blob: b1cd120607230b0770c35e0823d6389782ae734a [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/mm/swapfile.c
3 *
4 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
5 * Swap reorganised 29.12.95, Stephen Tweedie
6 */
7
Linus Torvalds1da177e2005-04-16 15:20:36 -07008#include <linux/mm.h>
9#include <linux/hugetlb.h>
10#include <linux/mman.h>
11#include <linux/slab.h>
12#include <linux/kernel_stat.h>
13#include <linux/swap.h>
14#include <linux/vmalloc.h>
15#include <linux/pagemap.h>
16#include <linux/namei.h>
Hugh Dickins072441e2011-06-27 16:18:02 -070017#include <linux/shmem_fs.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070018#include <linux/blkdev.h>
Hugh Dickins20137a42009-01-06 14:39:54 -080019#include <linux/random.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070020#include <linux/writeback.h>
21#include <linux/proc_fs.h>
22#include <linux/seq_file.h>
23#include <linux/init.h>
Hugh Dickins5ad64682009-12-14 17:59:24 -080024#include <linux/ksm.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070025#include <linux/rmap.h>
26#include <linux/security.h>
27#include <linux/backing-dev.h>
Ingo Molnarfc0abb12006-01-18 17:42:33 -080028#include <linux/mutex.h>
Randy.Dunlapc59ede72006-01-11 12:17:46 -080029#include <linux/capability.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070030#include <linux/syscalls.h>
Balbir Singh8a9f3cc2008-02-07 00:13:53 -080031#include <linux/memcontrol.h>
Kay Sievers66d7dd52010-10-26 14:22:06 -070032#include <linux/poll.h>
David Rientjes72788c32011-05-24 17:11:40 -070033#include <linux/oom.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070034
35#include <asm/pgtable.h>
36#include <asm/tlbflush.h>
37#include <linux/swapops.h>
KAMEZAWA Hiroyuki27a7faa2009-01-07 18:07:58 -080038#include <linux/page_cgroup.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070039
Hugh Dickins570a335b2009-12-14 17:58:46 -080040static bool swap_count_continued(struct swap_info_struct *, pgoff_t,
41 unsigned char);
42static void free_swap_count_continuations(struct swap_info_struct *);
Lee Schermerhornd4906e12009-12-14 17:58:49 -080043static sector_t map_swap_entry(swp_entry_t, struct block_device**);
Hugh Dickins570a335b2009-12-14 17:58:46 -080044
Adrian Bunk7c363b82008-07-25 19:46:24 -070045static DEFINE_SPINLOCK(swap_lock);
46static unsigned int nr_swapfiles;
Hugh Dickinsb9627162009-01-06 14:39:41 -080047long nr_swap_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -070048long total_swap_pages;
Hugh Dickins78ecba02008-07-23 21:28:23 -070049static int least_priority;
Linus Torvalds1da177e2005-04-16 15:20:36 -070050
Linus Torvalds1da177e2005-04-16 15:20:36 -070051static const char Bad_file[] = "Bad swap file entry ";
52static const char Unused_file[] = "Unused swap file entry ";
53static const char Bad_offset[] = "Bad swap offset entry ";
54static const char Unused_offset[] = "Unused swap offset entry ";
55
Adrian Bunk7c363b82008-07-25 19:46:24 -070056static struct swap_list_t swap_list = {-1, -1};
Linus Torvalds1da177e2005-04-16 15:20:36 -070057
Hugh Dickinsefa90a92009-12-14 17:58:41 -080058static struct swap_info_struct *swap_info[MAX_SWAPFILES];
Linus Torvalds1da177e2005-04-16 15:20:36 -070059
Ingo Molnarfc0abb12006-01-18 17:42:33 -080060static DEFINE_MUTEX(swapon_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -070061
Kay Sievers66d7dd52010-10-26 14:22:06 -070062static DECLARE_WAIT_QUEUE_HEAD(proc_poll_wait);
63/* Activity counter to indicate that a swapon or swapoff has occurred */
64static atomic_t proc_poll_event = ATOMIC_INIT(0);
65
Hugh Dickins8d69aae2009-12-14 17:58:45 -080066static inline unsigned char swap_count(unsigned char ent)
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -070067{
Hugh Dickins570a335b2009-12-14 17:58:46 -080068 return ent & ~SWAP_HAS_CACHE; /* may include SWAP_HAS_CONT flag */
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -070069}
70
Hugh Dickinsefa90a92009-12-14 17:58:41 -080071/* returns 1 if swap entry is freed */
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -070072static int
73__try_to_reclaim_swap(struct swap_info_struct *si, unsigned long offset)
74{
Hugh Dickinsefa90a92009-12-14 17:58:41 -080075 swp_entry_t entry = swp_entry(si->type, offset);
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -070076 struct page *page;
77 int ret = 0;
78
79 page = find_get_page(&swapper_space, entry.val);
80 if (!page)
81 return 0;
82 /*
83 * This function is called from scan_swap_map() and it's called
84 * by vmscan.c at reclaiming pages. So, we hold a lock on a page, here.
85 * We have to use trylock for avoiding deadlock. This is a special
86 * case and you should use try_to_free_swap() with explicit lock_page()
87 * in usual operations.
88 */
89 if (trylock_page(page)) {
90 ret = try_to_free_swap(page);
91 unlock_page(page);
92 }
93 page_cache_release(page);
94 return ret;
95}
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -070096
Linus Torvalds1da177e2005-04-16 15:20:36 -070097/*
Hugh Dickins6a6ba832009-01-06 14:39:51 -080098 * swapon tell device that all the old swap contents can be discarded,
99 * to allow the swap device to optimize its wear-levelling.
100 */
101static int discard_swap(struct swap_info_struct *si)
102{
103 struct swap_extent *se;
Hugh Dickins9625a5f2009-12-14 17:58:42 -0800104 sector_t start_block;
105 sector_t nr_blocks;
Hugh Dickins6a6ba832009-01-06 14:39:51 -0800106 int err = 0;
107
Hugh Dickins9625a5f2009-12-14 17:58:42 -0800108 /* Do not discard the swap header page! */
109 se = &si->first_swap_extent;
110 start_block = (se->start_block + 1) << (PAGE_SHIFT - 9);
111 nr_blocks = ((sector_t)se->nr_pages - 1) << (PAGE_SHIFT - 9);
112 if (nr_blocks) {
113 err = blkdev_issue_discard(si->bdev, start_block,
Christoph Hellwigdd3932e2010-09-16 20:51:46 +0200114 nr_blocks, GFP_KERNEL, 0);
Hugh Dickins9625a5f2009-12-14 17:58:42 -0800115 if (err)
116 return err;
117 cond_resched();
118 }
Hugh Dickins6a6ba832009-01-06 14:39:51 -0800119
Hugh Dickins9625a5f2009-12-14 17:58:42 -0800120 list_for_each_entry(se, &si->first_swap_extent.list, list) {
121 start_block = se->start_block << (PAGE_SHIFT - 9);
122 nr_blocks = (sector_t)se->nr_pages << (PAGE_SHIFT - 9);
Hugh Dickins6a6ba832009-01-06 14:39:51 -0800123
124 err = blkdev_issue_discard(si->bdev, start_block,
Christoph Hellwigdd3932e2010-09-16 20:51:46 +0200125 nr_blocks, GFP_KERNEL, 0);
Hugh Dickins6a6ba832009-01-06 14:39:51 -0800126 if (err)
127 break;
128
129 cond_resched();
130 }
131 return err; /* That will often be -EOPNOTSUPP */
132}
133
Hugh Dickins7992fde2009-01-06 14:39:53 -0800134/*
135 * swap allocation tell device that a cluster of swap can now be discarded,
136 * to allow the swap device to optimize its wear-levelling.
137 */
138static void discard_swap_cluster(struct swap_info_struct *si,
139 pgoff_t start_page, pgoff_t nr_pages)
140{
141 struct swap_extent *se = si->curr_swap_extent;
142 int found_extent = 0;
143
144 while (nr_pages) {
145 struct list_head *lh;
146
147 if (se->start_page <= start_page &&
148 start_page < se->start_page + se->nr_pages) {
149 pgoff_t offset = start_page - se->start_page;
150 sector_t start_block = se->start_block + offset;
Hugh Dickins858a29902009-01-06 14:39:56 -0800151 sector_t nr_blocks = se->nr_pages - offset;
Hugh Dickins7992fde2009-01-06 14:39:53 -0800152
153 if (nr_blocks > nr_pages)
154 nr_blocks = nr_pages;
155 start_page += nr_blocks;
156 nr_pages -= nr_blocks;
157
158 if (!found_extent++)
159 si->curr_swap_extent = se;
160
161 start_block <<= PAGE_SHIFT - 9;
162 nr_blocks <<= PAGE_SHIFT - 9;
163 if (blkdev_issue_discard(si->bdev, start_block,
Christoph Hellwigdd3932e2010-09-16 20:51:46 +0200164 nr_blocks, GFP_NOIO, 0))
Hugh Dickins7992fde2009-01-06 14:39:53 -0800165 break;
166 }
167
168 lh = se->list.next;
Hugh Dickins7992fde2009-01-06 14:39:53 -0800169 se = list_entry(lh, struct swap_extent, list);
170 }
171}
172
173static int wait_for_discard(void *word)
174{
175 schedule();
176 return 0;
177}
178
Hugh Dickins048c27f2005-09-03 15:54:40 -0700179#define SWAPFILE_CLUSTER 256
180#define LATENCY_LIMIT 256
181
Cesar Eduardo Barros24b8ff72011-03-22 16:33:38 -0700182static unsigned long scan_swap_map(struct swap_info_struct *si,
183 unsigned char usage)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700184{
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800185 unsigned long offset;
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800186 unsigned long scan_base;
Hugh Dickins7992fde2009-01-06 14:39:53 -0800187 unsigned long last_in_cluster = 0;
Hugh Dickins048c27f2005-09-03 15:54:40 -0700188 int latency_ration = LATENCY_LIMIT;
Hugh Dickins7992fde2009-01-06 14:39:53 -0800189 int found_free_cluster = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700190
Hugh Dickins886bb7e2009-01-06 14:39:48 -0800191 /*
Hugh Dickins7dfad412005-09-03 15:54:38 -0700192 * We try to cluster swap pages by allocating them sequentially
193 * in swap. Once we've allocated SWAPFILE_CLUSTER pages this
194 * way, however, we resort to first-free allocation, starting
195 * a new cluster. This prevents us from scattering swap pages
196 * all over the entire swap partition, so that we reduce
197 * overall disk seek times between swap pages. -- sct
198 * But we do now try to find an empty cluster. -Andrea
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800199 * And we let swap pages go all over an SSD partition. Hugh
Hugh Dickins7dfad412005-09-03 15:54:38 -0700200 */
201
Hugh Dickins52b7efdb2005-09-03 15:54:39 -0700202 si->flags += SWP_SCANNING;
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800203 scan_base = offset = si->cluster_next;
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800204
205 if (unlikely(!si->cluster_nr--)) {
206 if (si->pages - si->inuse_pages < SWAPFILE_CLUSTER) {
207 si->cluster_nr = SWAPFILE_CLUSTER - 1;
208 goto checks;
209 }
Hugh Dickins7992fde2009-01-06 14:39:53 -0800210 if (si->flags & SWP_DISCARDABLE) {
211 /*
212 * Start range check on racing allocations, in case
213 * they overlap the cluster we eventually decide on
214 * (we scan without swap_lock to allow preemption).
215 * It's hardly conceivable that cluster_nr could be
216 * wrapped during our scan, but don't depend on it.
217 */
218 if (si->lowest_alloc)
219 goto checks;
220 si->lowest_alloc = si->max;
221 si->highest_alloc = 0;
222 }
Hugh Dickins5d337b92005-09-03 15:54:41 -0700223 spin_unlock(&swap_lock);
Hugh Dickins7dfad412005-09-03 15:54:38 -0700224
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800225 /*
226 * If seek is expensive, start searching for new cluster from
227 * start of partition, to minimize the span of allocated swap.
228 * But if seek is cheap, search from our current position, so
229 * that swap is allocated from all over the partition: if the
230 * Flash Translation Layer only remaps within limited zones,
231 * we don't want to wear out the first zone too quickly.
232 */
233 if (!(si->flags & SWP_SOLIDSTATE))
234 scan_base = offset = si->lowest_bit;
Hugh Dickins7dfad412005-09-03 15:54:38 -0700235 last_in_cluster = offset + SWAPFILE_CLUSTER - 1;
236
237 /* Locate the first empty (unaligned) cluster */
238 for (; last_in_cluster <= si->highest_bit; offset++) {
239 if (si->swap_map[offset])
240 last_in_cluster = offset + SWAPFILE_CLUSTER;
241 else if (offset == last_in_cluster) {
Hugh Dickins5d337b92005-09-03 15:54:41 -0700242 spin_lock(&swap_lock);
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800243 offset -= SWAPFILE_CLUSTER - 1;
244 si->cluster_next = offset;
245 si->cluster_nr = SWAPFILE_CLUSTER - 1;
Hugh Dickins7992fde2009-01-06 14:39:53 -0800246 found_free_cluster = 1;
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800247 goto checks;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700248 }
Hugh Dickins048c27f2005-09-03 15:54:40 -0700249 if (unlikely(--latency_ration < 0)) {
250 cond_resched();
251 latency_ration = LATENCY_LIMIT;
252 }
Hugh Dickins7dfad412005-09-03 15:54:38 -0700253 }
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800254
255 offset = si->lowest_bit;
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800256 last_in_cluster = offset + SWAPFILE_CLUSTER - 1;
257
258 /* Locate the first empty (unaligned) cluster */
259 for (; last_in_cluster < scan_base; offset++) {
260 if (si->swap_map[offset])
261 last_in_cluster = offset + SWAPFILE_CLUSTER;
262 else if (offset == last_in_cluster) {
263 spin_lock(&swap_lock);
264 offset -= SWAPFILE_CLUSTER - 1;
265 si->cluster_next = offset;
266 si->cluster_nr = SWAPFILE_CLUSTER - 1;
267 found_free_cluster = 1;
268 goto checks;
269 }
270 if (unlikely(--latency_ration < 0)) {
271 cond_resched();
272 latency_ration = LATENCY_LIMIT;
273 }
274 }
275
276 offset = scan_base;
Hugh Dickins5d337b92005-09-03 15:54:41 -0700277 spin_lock(&swap_lock);
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800278 si->cluster_nr = SWAPFILE_CLUSTER - 1;
Hugh Dickins7992fde2009-01-06 14:39:53 -0800279 si->lowest_alloc = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700280 }
Hugh Dickins7dfad412005-09-03 15:54:38 -0700281
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800282checks:
283 if (!(si->flags & SWP_WRITEOK))
Hugh Dickins52b7efdb2005-09-03 15:54:39 -0700284 goto no_page;
Hugh Dickins7dfad412005-09-03 15:54:38 -0700285 if (!si->highest_bit)
286 goto no_page;
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800287 if (offset > si->highest_bit)
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800288 scan_base = offset = si->lowest_bit;
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700289
Hugh Dickinsb73d7fc2010-09-09 16:38:09 -0700290 /* reuse swap entry of cache-only swap if not busy. */
291 if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) {
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700292 int swap_was_freed;
293 spin_unlock(&swap_lock);
294 swap_was_freed = __try_to_reclaim_swap(si, offset);
295 spin_lock(&swap_lock);
296 /* entry was freed successfully, try to use this again */
297 if (swap_was_freed)
298 goto checks;
299 goto scan; /* check next one */
300 }
301
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800302 if (si->swap_map[offset])
303 goto scan;
Hugh Dickins7dfad412005-09-03 15:54:38 -0700304
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800305 if (offset == si->lowest_bit)
306 si->lowest_bit++;
307 if (offset == si->highest_bit)
308 si->highest_bit--;
309 si->inuse_pages++;
310 if (si->inuse_pages == si->pages) {
311 si->lowest_bit = si->max;
312 si->highest_bit = 0;
313 }
Hugh Dickins253d5532009-12-14 17:58:44 -0800314 si->swap_map[offset] = usage;
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800315 si->cluster_next = offset + 1;
316 si->flags -= SWP_SCANNING;
Hugh Dickins7992fde2009-01-06 14:39:53 -0800317
318 if (si->lowest_alloc) {
319 /*
320 * Only set when SWP_DISCARDABLE, and there's a scan
321 * for a free cluster in progress or just completed.
322 */
323 if (found_free_cluster) {
324 /*
325 * To optimize wear-levelling, discard the
326 * old data of the cluster, taking care not to
327 * discard any of its pages that have already
328 * been allocated by racing tasks (offset has
329 * already stepped over any at the beginning).
330 */
331 if (offset < si->highest_alloc &&
332 si->lowest_alloc <= last_in_cluster)
333 last_in_cluster = si->lowest_alloc - 1;
334 si->flags |= SWP_DISCARDING;
335 spin_unlock(&swap_lock);
336
337 if (offset < last_in_cluster)
338 discard_swap_cluster(si, offset,
339 last_in_cluster - offset + 1);
340
341 spin_lock(&swap_lock);
342 si->lowest_alloc = 0;
343 si->flags &= ~SWP_DISCARDING;
344
345 smp_mb(); /* wake_up_bit advises this */
346 wake_up_bit(&si->flags, ilog2(SWP_DISCARDING));
347
348 } else if (si->flags & SWP_DISCARDING) {
349 /*
350 * Delay using pages allocated by racing tasks
351 * until the whole discard has been issued. We
352 * could defer that delay until swap_writepage,
353 * but it's easier to keep this self-contained.
354 */
355 spin_unlock(&swap_lock);
356 wait_on_bit(&si->flags, ilog2(SWP_DISCARDING),
357 wait_for_discard, TASK_UNINTERRUPTIBLE);
358 spin_lock(&swap_lock);
359 } else {
360 /*
361 * Note pages allocated by racing tasks while
362 * scan for a free cluster is in progress, so
363 * that its final discard can exclude them.
364 */
365 if (offset < si->lowest_alloc)
366 si->lowest_alloc = offset;
367 if (offset > si->highest_alloc)
368 si->highest_alloc = offset;
369 }
370 }
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800371 return offset;
372
373scan:
Hugh Dickins5d337b92005-09-03 15:54:41 -0700374 spin_unlock(&swap_lock);
Hugh Dickins7dfad412005-09-03 15:54:38 -0700375 while (++offset <= si->highest_bit) {
Hugh Dickins52b7efdb2005-09-03 15:54:39 -0700376 if (!si->swap_map[offset]) {
Hugh Dickins5d337b92005-09-03 15:54:41 -0700377 spin_lock(&swap_lock);
Hugh Dickins52b7efdb2005-09-03 15:54:39 -0700378 goto checks;
379 }
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700380 if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) {
381 spin_lock(&swap_lock);
382 goto checks;
383 }
Hugh Dickins048c27f2005-09-03 15:54:40 -0700384 if (unlikely(--latency_ration < 0)) {
385 cond_resched();
386 latency_ration = LATENCY_LIMIT;
387 }
Hugh Dickins7dfad412005-09-03 15:54:38 -0700388 }
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800389 offset = si->lowest_bit;
390 while (++offset < scan_base) {
391 if (!si->swap_map[offset]) {
392 spin_lock(&swap_lock);
393 goto checks;
394 }
KAMEZAWA Hiroyukic9e44412009-06-16 15:32:54 -0700395 if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) {
396 spin_lock(&swap_lock);
397 goto checks;
398 }
Hugh Dickinsc60aa172009-01-06 14:39:55 -0800399 if (unlikely(--latency_ration < 0)) {
400 cond_resched();
401 latency_ration = LATENCY_LIMIT;
402 }
403 }
Hugh Dickins5d337b92005-09-03 15:54:41 -0700404 spin_lock(&swap_lock);
Hugh Dickins7dfad412005-09-03 15:54:38 -0700405
406no_page:
Hugh Dickins52b7efdb2005-09-03 15:54:39 -0700407 si->flags -= SWP_SCANNING;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700408 return 0;
409}
410
411swp_entry_t get_swap_page(void)
412{
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700413 struct swap_info_struct *si;
414 pgoff_t offset;
415 int type, next;
416 int wrapped = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700417
Hugh Dickins5d337b92005-09-03 15:54:41 -0700418 spin_lock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700419 if (nr_swap_pages <= 0)
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700420 goto noswap;
421 nr_swap_pages--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700422
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700423 for (type = swap_list.next; type >= 0 && wrapped < 2; type = next) {
Hugh Dickinsefa90a92009-12-14 17:58:41 -0800424 si = swap_info[type];
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700425 next = si->next;
426 if (next < 0 ||
Hugh Dickinsefa90a92009-12-14 17:58:41 -0800427 (!wrapped && si->prio != swap_info[next]->prio)) {
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700428 next = swap_list.head;
429 wrapped++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700430 }
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700431
432 if (!si->highest_bit)
433 continue;
434 if (!(si->flags & SWP_WRITEOK))
435 continue;
436
437 swap_list.next = next;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700438 /* This is called for allocating swap entry for cache */
Hugh Dickins253d5532009-12-14 17:58:44 -0800439 offset = scan_swap_map(si, SWAP_HAS_CACHE);
Hugh Dickins5d337b92005-09-03 15:54:41 -0700440 if (offset) {
441 spin_unlock(&swap_lock);
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700442 return swp_entry(type, offset);
Hugh Dickins5d337b92005-09-03 15:54:41 -0700443 }
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700444 next = swap_list.next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700445 }
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700446
447 nr_swap_pages++;
448noswap:
Hugh Dickins5d337b92005-09-03 15:54:41 -0700449 spin_unlock(&swap_lock);
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700450 return (swp_entry_t) {0};
Linus Torvalds1da177e2005-04-16 15:20:36 -0700451}
452
Hugh Dickins910321e2010-09-09 16:38:07 -0700453/* The only caller of this function is now susupend routine */
454swp_entry_t get_swap_page_of_type(int type)
455{
456 struct swap_info_struct *si;
457 pgoff_t offset;
458
459 spin_lock(&swap_lock);
460 si = swap_info[type];
461 if (si && (si->flags & SWP_WRITEOK)) {
462 nr_swap_pages--;
463 /* This is called for allocating swap entry, not cache */
464 offset = scan_swap_map(si, 1);
465 if (offset) {
466 spin_unlock(&swap_lock);
467 return swp_entry(type, offset);
468 }
469 nr_swap_pages++;
470 }
471 spin_unlock(&swap_lock);
472 return (swp_entry_t) {0};
473}
474
Hugh Dickins73c34b62009-12-14 17:58:43 -0800475static struct swap_info_struct *swap_info_get(swp_entry_t entry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700476{
Hugh Dickins73c34b62009-12-14 17:58:43 -0800477 struct swap_info_struct *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700478 unsigned long offset, type;
479
480 if (!entry.val)
481 goto out;
482 type = swp_type(entry);
483 if (type >= nr_swapfiles)
484 goto bad_nofile;
Hugh Dickinsefa90a92009-12-14 17:58:41 -0800485 p = swap_info[type];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700486 if (!(p->flags & SWP_USED))
487 goto bad_device;
488 offset = swp_offset(entry);
489 if (offset >= p->max)
490 goto bad_offset;
491 if (!p->swap_map[offset])
492 goto bad_free;
Hugh Dickins5d337b92005-09-03 15:54:41 -0700493 spin_lock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700494 return p;
495
496bad_free:
497 printk(KERN_ERR "swap_free: %s%08lx\n", Unused_offset, entry.val);
498 goto out;
499bad_offset:
500 printk(KERN_ERR "swap_free: %s%08lx\n", Bad_offset, entry.val);
501 goto out;
502bad_device:
503 printk(KERN_ERR "swap_free: %s%08lx\n", Unused_file, entry.val);
504 goto out;
505bad_nofile:
506 printk(KERN_ERR "swap_free: %s%08lx\n", Bad_file, entry.val);
507out:
508 return NULL;
Hugh Dickins886bb7e2009-01-06 14:39:48 -0800509}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700510
Hugh Dickins8d69aae2009-12-14 17:58:45 -0800511static unsigned char swap_entry_free(struct swap_info_struct *p,
512 swp_entry_t entry, unsigned char usage)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700513{
Hugh Dickins253d5532009-12-14 17:58:44 -0800514 unsigned long offset = swp_offset(entry);
Hugh Dickins8d69aae2009-12-14 17:58:45 -0800515 unsigned char count;
516 unsigned char has_cache;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700517
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700518 count = p->swap_map[offset];
Hugh Dickins253d5532009-12-14 17:58:44 -0800519 has_cache = count & SWAP_HAS_CACHE;
520 count &= ~SWAP_HAS_CACHE;
521
522 if (usage == SWAP_HAS_CACHE) {
523 VM_BUG_ON(!has_cache);
524 has_cache = 0;
Hugh Dickinsaaa46862009-12-14 17:58:47 -0800525 } else if (count == SWAP_MAP_SHMEM) {
526 /*
527 * Or we could insist on shmem.c using a special
528 * swap_shmem_free() and free_shmem_swap_and_cache()...
529 */
530 count = 0;
Hugh Dickins570a335b2009-12-14 17:58:46 -0800531 } else if ((count & ~COUNT_CONTINUED) <= SWAP_MAP_MAX) {
532 if (count == COUNT_CONTINUED) {
533 if (swap_count_continued(p, offset, count))
534 count = SWAP_MAP_MAX | COUNT_CONTINUED;
535 else
536 count = SWAP_MAP_MAX;
537 } else
538 count--;
539 }
Hugh Dickins253d5532009-12-14 17:58:44 -0800540
541 if (!count)
542 mem_cgroup_uncharge_swap(entry);
543
544 usage = count | has_cache;
545 p->swap_map[offset] = usage;
546
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700547 /* free if no reference */
Hugh Dickins253d5532009-12-14 17:58:44 -0800548 if (!usage) {
Nitin Guptab3a27d02010-05-17 11:02:43 +0530549 struct gendisk *disk = p->bdev->bd_disk;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700550 if (offset < p->lowest_bit)
551 p->lowest_bit = offset;
552 if (offset > p->highest_bit)
553 p->highest_bit = offset;
Hugh Dickinsefa90a92009-12-14 17:58:41 -0800554 if (swap_list.next >= 0 &&
555 p->prio > swap_info[swap_list.next]->prio)
556 swap_list.next = p->type;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700557 nr_swap_pages++;
558 p->inuse_pages--;
Nitin Guptab3a27d02010-05-17 11:02:43 +0530559 if ((p->flags & SWP_BLKDEV) &&
560 disk->fops->swap_slot_free_notify)
561 disk->fops->swap_slot_free_notify(p->bdev, offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700562 }
Hugh Dickins253d5532009-12-14 17:58:44 -0800563
564 return usage;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700565}
566
567/*
568 * Caller has made sure that the swapdevice corresponding to entry
569 * is still around or has not been recycled.
570 */
571void swap_free(swp_entry_t entry)
572{
Hugh Dickins73c34b62009-12-14 17:58:43 -0800573 struct swap_info_struct *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700574
575 p = swap_info_get(entry);
576 if (p) {
Hugh Dickins253d5532009-12-14 17:58:44 -0800577 swap_entry_free(p, entry, 1);
Hugh Dickins5d337b92005-09-03 15:54:41 -0700578 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700579 }
580}
581
582/*
KAMEZAWA Hiroyukicb4b86b2009-06-16 15:32:52 -0700583 * Called after dropping swapcache to decrease refcnt to swap entries.
584 */
585void swapcache_free(swp_entry_t entry, struct page *page)
586{
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700587 struct swap_info_struct *p;
Hugh Dickins8d69aae2009-12-14 17:58:45 -0800588 unsigned char count;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700589
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700590 p = swap_info_get(entry);
591 if (p) {
Hugh Dickins253d5532009-12-14 17:58:44 -0800592 count = swap_entry_free(p, entry, SWAP_HAS_CACHE);
593 if (page)
594 mem_cgroup_uncharge_swapcache(page, entry, count != 0);
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700595 spin_unlock(&swap_lock);
596 }
KAMEZAWA Hiroyukicb4b86b2009-06-16 15:32:52 -0700597}
598
599/*
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700600 * How many references to page are currently swapped out?
Hugh Dickins570a335b2009-12-14 17:58:46 -0800601 * This does not give an exact answer when swap count is continued,
602 * but does include the high COUNT_CONTINUED flag to allow for that.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700603 */
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700604static inline int page_swapcount(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700605{
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700606 int count = 0;
607 struct swap_info_struct *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608 swp_entry_t entry;
609
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700610 entry.val = page_private(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700611 p = swap_info_get(entry);
612 if (p) {
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700613 count = swap_count(p->swap_map[swp_offset(entry)]);
Hugh Dickins5d337b92005-09-03 15:54:41 -0700614 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700615 }
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700616 return count;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700617}
618
619/*
Hugh Dickins7b1fe592009-01-06 14:39:34 -0800620 * We can write to an anon page without COW if there are no other references
621 * to it. And as a side-effect, free up its swap: because the old content
622 * on disk will never be read, and seeking back there to write new content
623 * later would only waste time away from clustering.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624 */
Hugh Dickins7b1fe592009-01-06 14:39:34 -0800625int reuse_swap_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700626{
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700627 int count;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700628
Hugh Dickins51726b12009-01-06 14:39:25 -0800629 VM_BUG_ON(!PageLocked(page));
Hugh Dickins5ad64682009-12-14 17:59:24 -0800630 if (unlikely(PageKsm(page)))
631 return 0;
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700632 count = page_mapcount(page);
Hugh Dickins7b1fe592009-01-06 14:39:34 -0800633 if (count <= 1 && PageSwapCache(page)) {
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700634 count += page_swapcount(page);
Hugh Dickins7b1fe592009-01-06 14:39:34 -0800635 if (count == 1 && !PageWriteback(page)) {
636 delete_from_swap_cache(page);
637 SetPageDirty(page);
638 }
639 }
Hugh Dickins5ad64682009-12-14 17:59:24 -0800640 return count <= 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700641}
642
643/*
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -0800644 * If swap is getting full, or if there are no more mappings of this page,
645 * then try_to_free_swap is called to free its swap space.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700646 */
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -0800647int try_to_free_swap(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700648{
Hugh Dickins51726b12009-01-06 14:39:25 -0800649 VM_BUG_ON(!PageLocked(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700650
651 if (!PageSwapCache(page))
652 return 0;
653 if (PageWriteback(page))
654 return 0;
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -0800655 if (page_swapcount(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700656 return 0;
657
Hugh Dickinsb73d7fc2010-09-09 16:38:09 -0700658 /*
659 * Once hibernation has begun to create its image of memory,
660 * there's a danger that one of the calls to try_to_free_swap()
661 * - most probably a call from __try_to_reclaim_swap() while
662 * hibernation is allocating its own swap pages for the image,
663 * but conceivably even a call from memory reclaim - will free
664 * the swap from a page which has already been recorded in the
665 * image as a clean swapcache page, and then reuse its swap for
666 * another page of the image. On waking from hibernation, the
667 * original page might be freed under memory pressure, then
668 * later read back in from swap, now with the wrong data.
669 *
670 * Hibernation clears bits from gfp_allowed_mask to prevent
671 * memory reclaim from writing to disk, so check that here.
672 */
673 if (!(gfp_allowed_mask & __GFP_IO))
674 return 0;
675
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -0800676 delete_from_swap_cache(page);
677 SetPageDirty(page);
678 return 1;
Rik van Riel68a223942008-10-18 20:26:23 -0700679}
680
681/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700682 * Free the swap entry like above, but also try to
683 * free the page cache entry if it is the last user.
684 */
Hugh Dickins2509ef22009-01-06 14:40:10 -0800685int free_swap_and_cache(swp_entry_t entry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700686{
Hugh Dickins2509ef22009-01-06 14:40:10 -0800687 struct swap_info_struct *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700688 struct page *page = NULL;
689
Andi Kleena7420aa2009-09-16 11:50:05 +0200690 if (non_swap_entry(entry))
Hugh Dickins2509ef22009-01-06 14:40:10 -0800691 return 1;
Christoph Lameter06972122006-06-23 02:03:35 -0700692
Linus Torvalds1da177e2005-04-16 15:20:36 -0700693 p = swap_info_get(entry);
694 if (p) {
Hugh Dickins253d5532009-12-14 17:58:44 -0800695 if (swap_entry_free(p, entry, 1) == SWAP_HAS_CACHE) {
Nick Piggin93fac702006-03-31 02:29:56 -0800696 page = find_get_page(&swapper_space, entry.val);
Nick Piggin8413ac92008-10-18 20:26:59 -0700697 if (page && !trylock_page(page)) {
Nick Piggin93fac702006-03-31 02:29:56 -0800698 page_cache_release(page);
699 page = NULL;
700 }
701 }
Hugh Dickins5d337b92005-09-03 15:54:41 -0700702 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700703 }
704 if (page) {
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -0800705 /*
706 * Not mapped elsewhere, or swap space full? Free it!
707 * Also recheck PageSwapCache now page is locked (above).
708 */
Nick Piggin93fac702006-03-31 02:29:56 -0800709 if (PageSwapCache(page) && !PageWriteback(page) &&
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -0800710 (!page_mapped(page) || vm_swap_full())) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700711 delete_from_swap_cache(page);
712 SetPageDirty(page);
713 }
714 unlock_page(page);
715 page_cache_release(page);
716 }
Hugh Dickins2509ef22009-01-06 14:40:10 -0800717 return p != NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700718}
719
Daisuke Nishimura02491442010-03-10 15:22:17 -0800720#ifdef CONFIG_CGROUP_MEM_RES_CTLR
721/**
722 * mem_cgroup_count_swap_user - count the user of a swap entry
723 * @ent: the swap entry to be checked
724 * @pagep: the pointer for the swap cache page of the entry to be stored
725 *
726 * Returns the number of the user of the swap entry. The number is valid only
727 * for swaps of anonymous pages.
728 * If the entry is found on swap cache, the page is stored to pagep with
729 * refcount of it being incremented.
730 */
731int mem_cgroup_count_swap_user(swp_entry_t ent, struct page **pagep)
732{
733 struct page *page;
734 struct swap_info_struct *p;
735 int count = 0;
736
737 page = find_get_page(&swapper_space, ent.val);
738 if (page)
739 count += page_mapcount(page);
740 p = swap_info_get(ent);
741 if (p) {
742 count += swap_count(p->swap_map[swp_offset(ent)]);
743 spin_unlock(&swap_lock);
744 }
745
746 *pagep = page;
747 return count;
748}
749#endif
750
Rafael J. Wysockib0cb1a12007-07-29 23:24:36 +0200751#ifdef CONFIG_HIBERNATION
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800752/*
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800753 * Find the swap type that corresponds to given device (if any).
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800754 *
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800755 * @offset - number of the PAGE_SIZE-sized block of the device, starting
756 * from 0, in which the swap header is expected to be located.
757 *
758 * This is needed for the suspend to disk (aka swsusp).
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800759 */
Rafael J. Wysocki7bf23682007-01-05 16:36:28 -0800760int swap_type_of(dev_t device, sector_t offset, struct block_device **bdev_p)
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800761{
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800762 struct block_device *bdev = NULL;
Hugh Dickinsefa90a92009-12-14 17:58:41 -0800763 int type;
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800764
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800765 if (device)
766 bdev = bdget(device);
767
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800768 spin_lock(&swap_lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -0800769 for (type = 0; type < nr_swapfiles; type++) {
770 struct swap_info_struct *sis = swap_info[type];
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800771
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800772 if (!(sis->flags & SWP_WRITEOK))
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800773 continue;
Rafael J. Wysockib6b5bce2006-08-27 01:23:25 -0700774
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800775 if (!bdev) {
Rafael J. Wysocki7bf23682007-01-05 16:36:28 -0800776 if (bdev_p)
Alan Jenkinsdddac6a2009-07-29 21:07:55 +0200777 *bdev_p = bdgrab(sis->bdev);
Rafael J. Wysocki7bf23682007-01-05 16:36:28 -0800778
Rafael J. Wysocki6e1819d2006-03-23 03:00:03 -0800779 spin_unlock(&swap_lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -0800780 return type;
Rafael J. Wysocki6e1819d2006-03-23 03:00:03 -0800781 }
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800782 if (bdev == sis->bdev) {
Hugh Dickins9625a5f2009-12-14 17:58:42 -0800783 struct swap_extent *se = &sis->first_swap_extent;
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800784
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800785 if (se->start_block == offset) {
Rafael J. Wysocki7bf23682007-01-05 16:36:28 -0800786 if (bdev_p)
Alan Jenkinsdddac6a2009-07-29 21:07:55 +0200787 *bdev_p = bdgrab(sis->bdev);
Rafael J. Wysocki7bf23682007-01-05 16:36:28 -0800788
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800789 spin_unlock(&swap_lock);
790 bdput(bdev);
Hugh Dickinsefa90a92009-12-14 17:58:41 -0800791 return type;
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800792 }
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800793 }
794 }
795 spin_unlock(&swap_lock);
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800796 if (bdev)
797 bdput(bdev);
798
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800799 return -ENODEV;
800}
801
802/*
Hugh Dickins73c34b62009-12-14 17:58:43 -0800803 * Get the (PAGE_SIZE) block corresponding to given offset on the swapdev
804 * corresponding to given index in swap_info (swap type).
805 */
806sector_t swapdev_block(int type, pgoff_t offset)
807{
808 struct block_device *bdev;
809
810 if ((unsigned int)type >= nr_swapfiles)
811 return 0;
812 if (!(swap_info[type]->flags & SWP_WRITEOK))
813 return 0;
Lee Schermerhornd4906e12009-12-14 17:58:49 -0800814 return map_swap_entry(swp_entry(type, offset), &bdev);
Hugh Dickins73c34b62009-12-14 17:58:43 -0800815}
816
817/*
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800818 * Return either the total number of swap pages of given type, or the number
819 * of free pages of that type (depending on @free)
820 *
821 * This is needed for software suspend
822 */
823unsigned int count_swap_pages(int type, int free)
824{
825 unsigned int n = 0;
826
Hugh Dickinsefa90a92009-12-14 17:58:41 -0800827 spin_lock(&swap_lock);
828 if ((unsigned int)type < nr_swapfiles) {
829 struct swap_info_struct *sis = swap_info[type];
830
831 if (sis->flags & SWP_WRITEOK) {
832 n = sis->pages;
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800833 if (free)
Hugh Dickinsefa90a92009-12-14 17:58:41 -0800834 n -= sis->inuse_pages;
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800835 }
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800836 }
Hugh Dickinsefa90a92009-12-14 17:58:41 -0800837 spin_unlock(&swap_lock);
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800838 return n;
839}
Hugh Dickins73c34b62009-12-14 17:58:43 -0800840#endif /* CONFIG_HIBERNATION */
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800841
Linus Torvalds1da177e2005-04-16 15:20:36 -0700842/*
Hugh Dickins72866f62005-10-29 18:15:55 -0700843 * No need to decide whether this PTE shares the swap entry with others,
844 * just let do_wp_page work it out if a write is requested later - to
845 * force COW, vm_page_prot omits write permission from any private vma.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700846 */
Hugh Dickins044d66c2008-02-07 00:14:04 -0800847static int unuse_pte(struct vm_area_struct *vma, pmd_t *pmd,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700848 unsigned long addr, swp_entry_t entry, struct page *page)
849{
KAMEZAWA Hiroyuki56039ef2011-03-23 16:42:19 -0700850 struct mem_cgroup *ptr;
Hugh Dickins044d66c2008-02-07 00:14:04 -0800851 spinlock_t *ptl;
852 pte_t *pte;
853 int ret = 1;
854
KAMEZAWA Hiroyuki85d9fc82009-01-29 14:25:13 -0800855 if (mem_cgroup_try_charge_swapin(vma->vm_mm, page, GFP_KERNEL, &ptr)) {
Hugh Dickins044d66c2008-02-07 00:14:04 -0800856 ret = -ENOMEM;
KAMEZAWA Hiroyuki85d9fc82009-01-29 14:25:13 -0800857 goto out_nolock;
858 }
Hugh Dickins044d66c2008-02-07 00:14:04 -0800859
860 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
861 if (unlikely(!pte_same(*pte, swp_entry_to_pte(entry)))) {
862 if (ret > 0)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -0800863 mem_cgroup_cancel_charge_swapin(ptr);
Hugh Dickins044d66c2008-02-07 00:14:04 -0800864 ret = 0;
865 goto out;
866 }
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800867
KAMEZAWA Hiroyukib084d432010-03-05 13:41:42 -0800868 dec_mm_counter(vma->vm_mm, MM_SWAPENTS);
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -0800869 inc_mm_counter(vma->vm_mm, MM_ANONPAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700870 get_page(page);
871 set_pte_at(vma->vm_mm, addr, pte,
872 pte_mkold(mk_pte(page, vma->vm_page_prot)));
873 page_add_anon_rmap(page, vma, addr);
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -0800874 mem_cgroup_commit_charge_swapin(page, ptr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700875 swap_free(entry);
876 /*
877 * Move the page to the active list so it is not
878 * immediately swapped out again after swapon.
879 */
880 activate_page(page);
Hugh Dickins044d66c2008-02-07 00:14:04 -0800881out:
882 pte_unmap_unlock(pte, ptl);
KAMEZAWA Hiroyuki85d9fc82009-01-29 14:25:13 -0800883out_nolock:
Hugh Dickins044d66c2008-02-07 00:14:04 -0800884 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700885}
886
887static int unuse_pte_range(struct vm_area_struct *vma, pmd_t *pmd,
888 unsigned long addr, unsigned long end,
889 swp_entry_t entry, struct page *page)
890{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700891 pte_t swp_pte = swp_entry_to_pte(entry);
Hugh Dickins705e87c2005-10-29 18:16:27 -0700892 pte_t *pte;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800893 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700894
Hugh Dickins044d66c2008-02-07 00:14:04 -0800895 /*
896 * We don't actually need pte lock while scanning for swp_pte: since
897 * we hold page lock and mmap_sem, swp_pte cannot be inserted into the
898 * page table while we're scanning; though it could get zapped, and on
899 * some architectures (e.g. x86_32 with PAE) we might catch a glimpse
900 * of unmatched parts which look like swp_pte, so unuse_pte must
901 * recheck under pte lock. Scanning without pte lock lets it be
902 * preemptible whenever CONFIG_PREEMPT but not CONFIG_HIGHPTE.
903 */
904 pte = pte_offset_map(pmd, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700905 do {
906 /*
907 * swapoff spends a _lot_ of time in this loop!
908 * Test inline before going to call unuse_pte.
909 */
910 if (unlikely(pte_same(*pte, swp_pte))) {
Hugh Dickins044d66c2008-02-07 00:14:04 -0800911 pte_unmap(pte);
912 ret = unuse_pte(vma, pmd, addr, entry, page);
913 if (ret)
914 goto out;
915 pte = pte_offset_map(pmd, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700916 }
917 } while (pte++, addr += PAGE_SIZE, addr != end);
Hugh Dickins044d66c2008-02-07 00:14:04 -0800918 pte_unmap(pte - 1);
919out:
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800920 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700921}
922
923static inline int unuse_pmd_range(struct vm_area_struct *vma, pud_t *pud,
924 unsigned long addr, unsigned long end,
925 swp_entry_t entry, struct page *page)
926{
927 pmd_t *pmd;
928 unsigned long next;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800929 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700930
931 pmd = pmd_offset(pud, addr);
932 do {
933 next = pmd_addr_end(addr, end);
Andrea Arcangeli3f04f622011-01-13 15:46:47 -0800934 if (unlikely(pmd_trans_huge(*pmd)))
935 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700936 if (pmd_none_or_clear_bad(pmd))
937 continue;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800938 ret = unuse_pte_range(vma, pmd, addr, next, entry, page);
939 if (ret)
940 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700941 } while (pmd++, addr = next, addr != end);
942 return 0;
943}
944
945static inline int unuse_pud_range(struct vm_area_struct *vma, pgd_t *pgd,
946 unsigned long addr, unsigned long end,
947 swp_entry_t entry, struct page *page)
948{
949 pud_t *pud;
950 unsigned long next;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800951 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700952
953 pud = pud_offset(pgd, addr);
954 do {
955 next = pud_addr_end(addr, end);
956 if (pud_none_or_clear_bad(pud))
957 continue;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800958 ret = unuse_pmd_range(vma, pud, addr, next, entry, page);
959 if (ret)
960 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700961 } while (pud++, addr = next, addr != end);
962 return 0;
963}
964
965static int unuse_vma(struct vm_area_struct *vma,
966 swp_entry_t entry, struct page *page)
967{
968 pgd_t *pgd;
969 unsigned long addr, end, next;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800970 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700971
Hugh Dickins3ca7b3c2009-12-14 17:58:57 -0800972 if (page_anon_vma(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700973 addr = page_address_in_vma(page, vma);
974 if (addr == -EFAULT)
975 return 0;
976 else
977 end = addr + PAGE_SIZE;
978 } else {
979 addr = vma->vm_start;
980 end = vma->vm_end;
981 }
982
983 pgd = pgd_offset(vma->vm_mm, addr);
984 do {
985 next = pgd_addr_end(addr, end);
986 if (pgd_none_or_clear_bad(pgd))
987 continue;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800988 ret = unuse_pud_range(vma, pgd, addr, next, entry, page);
989 if (ret)
990 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700991 } while (pgd++, addr = next, addr != end);
992 return 0;
993}
994
995static int unuse_mm(struct mm_struct *mm,
996 swp_entry_t entry, struct page *page)
997{
998 struct vm_area_struct *vma;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800999 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001000
1001 if (!down_read_trylock(&mm->mmap_sem)) {
1002 /*
Fernando Luis Vazquez Cao7d034312008-07-29 22:33:41 -07001003 * Activate page so shrink_inactive_list is unlikely to unmap
1004 * its ptes while lock is dropped, so swapoff can make progress.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001005 */
Hugh Dickinsc475a8a2005-06-21 17:15:12 -07001006 activate_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001007 unlock_page(page);
1008 down_read(&mm->mmap_sem);
1009 lock_page(page);
1010 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001011 for (vma = mm->mmap; vma; vma = vma->vm_next) {
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001012 if (vma->anon_vma && (ret = unuse_vma(vma, entry, page)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001013 break;
1014 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001015 up_read(&mm->mmap_sem);
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001016 return (ret < 0)? ret: 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001017}
1018
1019/*
1020 * Scan swap_map from current position to next entry still in use.
1021 * Recycle to start on reaching the end, returning 0 when empty.
1022 */
Hugh Dickins6eb396d2005-09-03 15:54:35 -07001023static unsigned int find_next_to_unuse(struct swap_info_struct *si,
1024 unsigned int prev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001025{
Hugh Dickins6eb396d2005-09-03 15:54:35 -07001026 unsigned int max = si->max;
1027 unsigned int i = prev;
Hugh Dickins8d69aae2009-12-14 17:58:45 -08001028 unsigned char count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001029
1030 /*
Hugh Dickins5d337b92005-09-03 15:54:41 -07001031 * No need for swap_lock here: we're just looking
Linus Torvalds1da177e2005-04-16 15:20:36 -07001032 * for whether an entry is in use, not modifying it; false
1033 * hits are okay, and sys_swapoff() has already prevented new
Hugh Dickins5d337b92005-09-03 15:54:41 -07001034 * allocations from this area (while holding swap_lock).
Linus Torvalds1da177e2005-04-16 15:20:36 -07001035 */
1036 for (;;) {
1037 if (++i >= max) {
1038 if (!prev) {
1039 i = 0;
1040 break;
1041 }
1042 /*
1043 * No entries in use at top of swap_map,
1044 * loop back to start and recheck there.
1045 */
1046 max = prev + 1;
1047 prev = 0;
1048 i = 1;
1049 }
1050 count = si->swap_map[i];
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07001051 if (count && swap_count(count) != SWAP_MAP_BAD)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001052 break;
1053 }
1054 return i;
1055}
1056
1057/*
1058 * We completely avoid races by reading each swap page in advance,
1059 * and then search for the process using it. All the necessary
1060 * page table adjustments can then be made atomically.
1061 */
1062static int try_to_unuse(unsigned int type)
1063{
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001064 struct swap_info_struct *si = swap_info[type];
Linus Torvalds1da177e2005-04-16 15:20:36 -07001065 struct mm_struct *start_mm;
Hugh Dickins8d69aae2009-12-14 17:58:45 -08001066 unsigned char *swap_map;
1067 unsigned char swcount;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001068 struct page *page;
1069 swp_entry_t entry;
Hugh Dickins6eb396d2005-09-03 15:54:35 -07001070 unsigned int i = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001071 int retval = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001072
1073 /*
1074 * When searching mms for an entry, a good strategy is to
1075 * start at the first mm we freed the previous entry from
1076 * (though actually we don't notice whether we or coincidence
1077 * freed the entry). Initialize this start_mm with a hold.
1078 *
1079 * A simpler strategy would be to start at the last mm we
1080 * freed the previous entry from; but that would take less
1081 * advantage of mmlist ordering, which clusters forked mms
1082 * together, child after parent. If we race with dup_mmap(), we
1083 * prefer to resolve parent before child, lest we miss entries
1084 * duplicated after we scanned child: using last mm would invert
Hugh Dickins570a335b2009-12-14 17:58:46 -08001085 * that.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001086 */
1087 start_mm = &init_mm;
1088 atomic_inc(&init_mm.mm_users);
1089
1090 /*
1091 * Keep on scanning until all entries have gone. Usually,
1092 * one pass through swap_map is enough, but not necessarily:
1093 * there are races when an instance of an entry might be missed.
1094 */
1095 while ((i = find_next_to_unuse(si, i)) != 0) {
1096 if (signal_pending(current)) {
1097 retval = -EINTR;
1098 break;
1099 }
1100
Hugh Dickins886bb7e2009-01-06 14:39:48 -08001101 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001102 * Get a page for the entry, using the existing swap
1103 * cache page if there is one. Otherwise, get a clean
Hugh Dickins886bb7e2009-01-06 14:39:48 -08001104 * page and read the swap into it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001105 */
1106 swap_map = &si->swap_map[i];
1107 entry = swp_entry(type, i);
Hugh Dickins02098fe2008-02-04 22:28:42 -08001108 page = read_swap_cache_async(entry,
1109 GFP_HIGHUSER_MOVABLE, NULL, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001110 if (!page) {
1111 /*
1112 * Either swap_duplicate() failed because entry
1113 * has been freed independently, and will not be
1114 * reused since sys_swapoff() already disabled
1115 * allocation from here, or alloc_page() failed.
1116 */
1117 if (!*swap_map)
1118 continue;
1119 retval = -ENOMEM;
1120 break;
1121 }
1122
1123 /*
1124 * Don't hold on to start_mm if it looks like exiting.
1125 */
1126 if (atomic_read(&start_mm->mm_users) == 1) {
1127 mmput(start_mm);
1128 start_mm = &init_mm;
1129 atomic_inc(&init_mm.mm_users);
1130 }
1131
1132 /*
1133 * Wait for and lock page. When do_swap_page races with
1134 * try_to_unuse, do_swap_page can handle the fault much
1135 * faster than try_to_unuse can locate the entry. This
1136 * apparently redundant "wait_on_page_locked" lets try_to_unuse
1137 * defer to do_swap_page in such a case - in some tests,
1138 * do_swap_page and try_to_unuse repeatedly compete.
1139 */
1140 wait_on_page_locked(page);
1141 wait_on_page_writeback(page);
1142 lock_page(page);
1143 wait_on_page_writeback(page);
1144
1145 /*
1146 * Remove all references to entry.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001147 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001148 swcount = *swap_map;
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001149 if (swap_count(swcount) == SWAP_MAP_SHMEM) {
1150 retval = shmem_unuse(entry, page);
1151 /* page has already been unlocked and released */
1152 if (retval < 0)
1153 break;
1154 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001155 }
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001156 if (swap_count(swcount) && start_mm != &init_mm)
1157 retval = unuse_mm(start_mm, entry, page);
1158
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07001159 if (swap_count(*swap_map)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001160 int set_start_mm = (*swap_map >= swcount);
1161 struct list_head *p = &start_mm->mmlist;
1162 struct mm_struct *new_start_mm = start_mm;
1163 struct mm_struct *prev_mm = start_mm;
1164 struct mm_struct *mm;
1165
1166 atomic_inc(&new_start_mm->mm_users);
1167 atomic_inc(&prev_mm->mm_users);
1168 spin_lock(&mmlist_lock);
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001169 while (swap_count(*swap_map) && !retval &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001170 (p = p->next) != &start_mm->mmlist) {
1171 mm = list_entry(p, struct mm_struct, mmlist);
Hugh Dickins70af7c52006-06-23 02:03:44 -07001172 if (!atomic_inc_not_zero(&mm->mm_users))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001173 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001174 spin_unlock(&mmlist_lock);
1175 mmput(prev_mm);
1176 prev_mm = mm;
1177
1178 cond_resched();
1179
1180 swcount = *swap_map;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07001181 if (!swap_count(swcount)) /* any usage ? */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001182 ;
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001183 else if (mm == &init_mm)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001184 set_start_mm = 1;
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001185 else
Linus Torvalds1da177e2005-04-16 15:20:36 -07001186 retval = unuse_mm(mm, entry, page);
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07001187
Bo Liu32c5fc12009-11-02 16:50:33 +00001188 if (set_start_mm && *swap_map < swcount) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001189 mmput(new_start_mm);
1190 atomic_inc(&mm->mm_users);
1191 new_start_mm = mm;
1192 set_start_mm = 0;
1193 }
1194 spin_lock(&mmlist_lock);
1195 }
1196 spin_unlock(&mmlist_lock);
1197 mmput(prev_mm);
1198 mmput(start_mm);
1199 start_mm = new_start_mm;
1200 }
1201 if (retval) {
1202 unlock_page(page);
1203 page_cache_release(page);
1204 break;
1205 }
1206
1207 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001208 * If a reference remains (rare), we would like to leave
1209 * the page in the swap cache; but try_to_unmap could
1210 * then re-duplicate the entry once we drop page lock,
1211 * so we might loop indefinitely; also, that page could
1212 * not be swapped out to other storage meanwhile. So:
1213 * delete from cache even if there's another reference,
1214 * after ensuring that the data has been saved to disk -
1215 * since if the reference remains (rarer), it will be
1216 * read from disk into another page. Splitting into two
1217 * pages would be incorrect if swap supported "shared
1218 * private" pages, but they are handled by tmpfs files.
Hugh Dickins5ad64682009-12-14 17:59:24 -08001219 *
1220 * Given how unuse_vma() targets one particular offset
1221 * in an anon_vma, once the anon_vma has been determined,
1222 * this splitting happens to be just what is needed to
1223 * handle where KSM pages have been swapped out: re-reading
1224 * is unnecessarily slow, but we can fix that later on.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001225 */
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07001226 if (swap_count(*swap_map) &&
1227 PageDirty(page) && PageSwapCache(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001228 struct writeback_control wbc = {
1229 .sync_mode = WB_SYNC_NONE,
1230 };
1231
1232 swap_writepage(page, &wbc);
1233 lock_page(page);
1234 wait_on_page_writeback(page);
1235 }
Hugh Dickins68bdc8d2009-01-06 14:39:37 -08001236
1237 /*
1238 * It is conceivable that a racing task removed this page from
1239 * swap cache just before we acquired the page lock at the top,
1240 * or while we dropped it in unuse_mm(). The page might even
1241 * be back in swap cache on another swap area: that we must not
1242 * delete, since it may not have been written out to swap yet.
1243 */
1244 if (PageSwapCache(page) &&
1245 likely(page_private(page) == entry.val))
Hugh Dickins2e0e26c2008-02-04 22:28:53 -08001246 delete_from_swap_cache(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001247
1248 /*
1249 * So we could skip searching mms once swap count went
1250 * to 1, we did not mark any present ptes as dirty: must
Anderson Briglia2706a1b2007-07-15 23:38:09 -07001251 * mark page dirty so shrink_page_list will preserve it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001252 */
1253 SetPageDirty(page);
1254 unlock_page(page);
1255 page_cache_release(page);
1256
1257 /*
1258 * Make sure that we aren't completely killing
1259 * interactive performance.
1260 */
1261 cond_resched();
1262 }
1263
1264 mmput(start_mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001265 return retval;
1266}
1267
1268/*
Hugh Dickins5d337b92005-09-03 15:54:41 -07001269 * After a successful try_to_unuse, if no swap is now in use, we know
1270 * we can empty the mmlist. swap_lock must be held on entry and exit.
1271 * Note that mmlist_lock nests inside swap_lock, and an mm must be
Linus Torvalds1da177e2005-04-16 15:20:36 -07001272 * added to the mmlist just after page_duplicate - before would be racy.
1273 */
1274static void drain_mmlist(void)
1275{
1276 struct list_head *p, *next;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001277 unsigned int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001278
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001279 for (type = 0; type < nr_swapfiles; type++)
1280 if (swap_info[type]->inuse_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001281 return;
1282 spin_lock(&mmlist_lock);
1283 list_for_each_safe(p, next, &init_mm.mmlist)
1284 list_del_init(p);
1285 spin_unlock(&mmlist_lock);
1286}
1287
1288/*
1289 * Use this swapdev's extent info to locate the (PAGE_SIZE) block which
Lee Schermerhornd4906e12009-12-14 17:58:49 -08001290 * corresponds to page offset for the specified swap entry.
1291 * Note that the type of this function is sector_t, but it returns page offset
1292 * into the bdev, not sector offset.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001293 */
Lee Schermerhornd4906e12009-12-14 17:58:49 -08001294static sector_t map_swap_entry(swp_entry_t entry, struct block_device **bdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001295{
Hugh Dickinsf29ad6a2009-12-14 17:58:40 -08001296 struct swap_info_struct *sis;
1297 struct swap_extent *start_se;
1298 struct swap_extent *se;
1299 pgoff_t offset;
1300
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001301 sis = swap_info[swp_type(entry)];
Hugh Dickinsf29ad6a2009-12-14 17:58:40 -08001302 *bdev = sis->bdev;
1303
1304 offset = swp_offset(entry);
1305 start_se = sis->curr_swap_extent;
1306 se = start_se;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001307
1308 for ( ; ; ) {
1309 struct list_head *lh;
1310
1311 if (se->start_page <= offset &&
1312 offset < (se->start_page + se->nr_pages)) {
1313 return se->start_block + (offset - se->start_page);
1314 }
Hugh Dickins11d31882005-09-03 15:54:34 -07001315 lh = se->list.next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001316 se = list_entry(lh, struct swap_extent, list);
1317 sis->curr_swap_extent = se;
1318 BUG_ON(se == start_se); /* It *must* be present */
1319 }
1320}
1321
1322/*
Lee Schermerhornd4906e12009-12-14 17:58:49 -08001323 * Returns the page offset into bdev for the specified page's swap entry.
1324 */
1325sector_t map_swap_page(struct page *page, struct block_device **bdev)
1326{
1327 swp_entry_t entry;
1328 entry.val = page_private(page);
1329 return map_swap_entry(entry, bdev);
1330}
1331
1332/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001333 * Free all of a swapdev's extent information
1334 */
1335static void destroy_swap_extents(struct swap_info_struct *sis)
1336{
Hugh Dickins9625a5f2009-12-14 17:58:42 -08001337 while (!list_empty(&sis->first_swap_extent.list)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001338 struct swap_extent *se;
1339
Hugh Dickins9625a5f2009-12-14 17:58:42 -08001340 se = list_entry(sis->first_swap_extent.list.next,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001341 struct swap_extent, list);
1342 list_del(&se->list);
1343 kfree(se);
1344 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001345}
1346
1347/*
1348 * Add a block range (and the corresponding page range) into this swapdev's
Hugh Dickins11d31882005-09-03 15:54:34 -07001349 * extent list. The extent list is kept sorted in page order.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001350 *
Hugh Dickins11d31882005-09-03 15:54:34 -07001351 * This function rather assumes that it is called in ascending page order.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001352 */
1353static int
1354add_swap_extent(struct swap_info_struct *sis, unsigned long start_page,
1355 unsigned long nr_pages, sector_t start_block)
1356{
1357 struct swap_extent *se;
1358 struct swap_extent *new_se;
1359 struct list_head *lh;
1360
Hugh Dickins9625a5f2009-12-14 17:58:42 -08001361 if (start_page == 0) {
1362 se = &sis->first_swap_extent;
1363 sis->curr_swap_extent = se;
1364 se->start_page = 0;
1365 se->nr_pages = nr_pages;
1366 se->start_block = start_block;
1367 return 1;
1368 } else {
1369 lh = sis->first_swap_extent.list.prev; /* Highest extent */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001370 se = list_entry(lh, struct swap_extent, list);
Hugh Dickins11d31882005-09-03 15:54:34 -07001371 BUG_ON(se->start_page + se->nr_pages != start_page);
1372 if (se->start_block + se->nr_pages == start_block) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001373 /* Merge it */
1374 se->nr_pages += nr_pages;
1375 return 0;
1376 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001377 }
1378
1379 /*
1380 * No merge. Insert a new extent, preserving ordering.
1381 */
1382 new_se = kmalloc(sizeof(*se), GFP_KERNEL);
1383 if (new_se == NULL)
1384 return -ENOMEM;
1385 new_se->start_page = start_page;
1386 new_se->nr_pages = nr_pages;
1387 new_se->start_block = start_block;
1388
Hugh Dickins9625a5f2009-12-14 17:58:42 -08001389 list_add_tail(&new_se->list, &sis->first_swap_extent.list);
Hugh Dickins53092a72005-09-03 15:54:34 -07001390 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001391}
1392
1393/*
1394 * A `swap extent' is a simple thing which maps a contiguous range of pages
1395 * onto a contiguous range of disk blocks. An ordered list of swap extents
1396 * is built at swapon time and is then used at swap_writepage/swap_readpage
1397 * time for locating where on disk a page belongs.
1398 *
1399 * If the swapfile is an S_ISBLK block device, a single extent is installed.
1400 * This is done so that the main operating code can treat S_ISBLK and S_ISREG
1401 * swap files identically.
1402 *
1403 * Whether the swapdev is an S_ISREG file or an S_ISBLK blockdev, the swap
1404 * extent list operates in PAGE_SIZE disk blocks. Both S_ISREG and S_ISBLK
1405 * swapfiles are handled *identically* after swapon time.
1406 *
1407 * For S_ISREG swapfiles, setup_swap_extents() will walk all the file's blocks
1408 * and will parse them into an ordered extent list, in PAGE_SIZE chunks. If
1409 * some stray blocks are found which do not fall within the PAGE_SIZE alignment
1410 * requirements, they are simply tossed out - we will never use those blocks
1411 * for swapping.
1412 *
Hugh Dickinsb0d9bcd2005-09-03 15:54:31 -07001413 * For S_ISREG swapfiles we set S_SWAPFILE across the life of the swapon. This
Linus Torvalds1da177e2005-04-16 15:20:36 -07001414 * prevents root from shooting her foot off by ftruncating an in-use swapfile,
1415 * which will scribble on the fs.
1416 *
1417 * The amount of disk space which a single swap extent represents varies.
1418 * Typically it is in the 1-4 megabyte range. So we can have hundreds of
1419 * extents in the list. To avoid much list walking, we cache the previous
1420 * search location in `curr_swap_extent', and start new searches from there.
1421 * This is extremely effective. The average number of iterations in
1422 * map_swap_page() has been measured at about 0.3 per page. - akpm.
1423 */
Hugh Dickins53092a72005-09-03 15:54:34 -07001424static int setup_swap_extents(struct swap_info_struct *sis, sector_t *span)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001425{
1426 struct inode *inode;
1427 unsigned blocks_per_page;
1428 unsigned long page_no;
1429 unsigned blkbits;
1430 sector_t probe_block;
1431 sector_t last_block;
Hugh Dickins53092a72005-09-03 15:54:34 -07001432 sector_t lowest_block = -1;
1433 sector_t highest_block = 0;
1434 int nr_extents = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001435 int ret;
1436
1437 inode = sis->swap_file->f_mapping->host;
1438 if (S_ISBLK(inode->i_mode)) {
1439 ret = add_swap_extent(sis, 0, sis->max, 0);
Hugh Dickins53092a72005-09-03 15:54:34 -07001440 *span = sis->pages;
Hugh Dickins9625a5f2009-12-14 17:58:42 -08001441 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001442 }
1443
1444 blkbits = inode->i_blkbits;
1445 blocks_per_page = PAGE_SIZE >> blkbits;
1446
1447 /*
1448 * Map all the blocks into the extent list. This code doesn't try
1449 * to be very smart.
1450 */
1451 probe_block = 0;
1452 page_no = 0;
1453 last_block = i_size_read(inode) >> blkbits;
1454 while ((probe_block + blocks_per_page) <= last_block &&
1455 page_no < sis->max) {
1456 unsigned block_in_page;
1457 sector_t first_block;
1458
1459 first_block = bmap(inode, probe_block);
1460 if (first_block == 0)
1461 goto bad_bmap;
1462
1463 /*
1464 * It must be PAGE_SIZE aligned on-disk
1465 */
1466 if (first_block & (blocks_per_page - 1)) {
1467 probe_block++;
1468 goto reprobe;
1469 }
1470
1471 for (block_in_page = 1; block_in_page < blocks_per_page;
1472 block_in_page++) {
1473 sector_t block;
1474
1475 block = bmap(inode, probe_block + block_in_page);
1476 if (block == 0)
1477 goto bad_bmap;
1478 if (block != first_block + block_in_page) {
1479 /* Discontiguity */
1480 probe_block++;
1481 goto reprobe;
1482 }
1483 }
1484
Hugh Dickins53092a72005-09-03 15:54:34 -07001485 first_block >>= (PAGE_SHIFT - blkbits);
1486 if (page_no) { /* exclude the header page */
1487 if (first_block < lowest_block)
1488 lowest_block = first_block;
1489 if (first_block > highest_block)
1490 highest_block = first_block;
1491 }
1492
Linus Torvalds1da177e2005-04-16 15:20:36 -07001493 /*
1494 * We found a PAGE_SIZE-length, PAGE_SIZE-aligned run of blocks
1495 */
Hugh Dickins53092a72005-09-03 15:54:34 -07001496 ret = add_swap_extent(sis, page_no, 1, first_block);
1497 if (ret < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001498 goto out;
Hugh Dickins53092a72005-09-03 15:54:34 -07001499 nr_extents += ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001500 page_no++;
1501 probe_block += blocks_per_page;
1502reprobe:
1503 continue;
1504 }
Hugh Dickins53092a72005-09-03 15:54:34 -07001505 ret = nr_extents;
1506 *span = 1 + highest_block - lowest_block;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001507 if (page_no == 0)
Hugh Dickinse2244ec2005-09-03 15:54:32 -07001508 page_no = 1; /* force Empty message */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001509 sis->max = page_no;
Hugh Dickinse2244ec2005-09-03 15:54:32 -07001510 sis->pages = page_no - 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001511 sis->highest_bit = page_no - 1;
Hugh Dickins9625a5f2009-12-14 17:58:42 -08001512out:
1513 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001514bad_bmap:
1515 printk(KERN_ERR "swapon: swapfile has holes\n");
1516 ret = -EINVAL;
Hugh Dickins9625a5f2009-12-14 17:58:42 -08001517 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001518}
1519
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07001520static void enable_swap_info(struct swap_info_struct *p, int prio,
1521 unsigned char *swap_map)
1522{
1523 int i, prev;
1524
1525 spin_lock(&swap_lock);
1526 if (prio >= 0)
1527 p->prio = prio;
1528 else
1529 p->prio = --least_priority;
1530 p->swap_map = swap_map;
1531 p->flags |= SWP_WRITEOK;
1532 nr_swap_pages += p->pages;
1533 total_swap_pages += p->pages;
1534
1535 /* insert swap space into swap_list: */
1536 prev = -1;
1537 for (i = swap_list.head; i >= 0; i = swap_info[i]->next) {
1538 if (p->prio >= swap_info[i]->prio)
1539 break;
1540 prev = i;
1541 }
1542 p->next = i;
1543 if (prev < 0)
1544 swap_list.head = swap_list.next = p->type;
1545 else
1546 swap_info[prev]->next = p->type;
1547 spin_unlock(&swap_lock);
1548}
1549
Heiko Carstensc4ea37c2009-01-14 14:14:28 +01001550SYSCALL_DEFINE1(swapoff, const char __user *, specialfile)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001551{
Hugh Dickins73c34b62009-12-14 17:58:43 -08001552 struct swap_info_struct *p = NULL;
Hugh Dickins8d69aae2009-12-14 17:58:45 -08001553 unsigned char *swap_map;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001554 struct file *swap_file, *victim;
1555 struct address_space *mapping;
1556 struct inode *inode;
Hugh Dickins73c34b62009-12-14 17:58:43 -08001557 char *pathname;
David Rientjes72788c32011-05-24 17:11:40 -07001558 int oom_score_adj;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001559 int i, type, prev;
1560 int err;
Hugh Dickins886bb7e2009-01-06 14:39:48 -08001561
Linus Torvalds1da177e2005-04-16 15:20:36 -07001562 if (!capable(CAP_SYS_ADMIN))
1563 return -EPERM;
1564
1565 pathname = getname(specialfile);
1566 err = PTR_ERR(pathname);
1567 if (IS_ERR(pathname))
1568 goto out;
1569
1570 victim = filp_open(pathname, O_RDWR|O_LARGEFILE, 0);
1571 putname(pathname);
1572 err = PTR_ERR(victim);
1573 if (IS_ERR(victim))
1574 goto out;
1575
1576 mapping = victim->f_mapping;
1577 prev = -1;
Hugh Dickins5d337b92005-09-03 15:54:41 -07001578 spin_lock(&swap_lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001579 for (type = swap_list.head; type >= 0; type = swap_info[type]->next) {
1580 p = swap_info[type];
Hugh Dickins22c6f8fd2009-01-06 14:39:48 -08001581 if (p->flags & SWP_WRITEOK) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001582 if (p->swap_file->f_mapping == mapping)
1583 break;
1584 }
1585 prev = type;
1586 }
1587 if (type < 0) {
1588 err = -EINVAL;
Hugh Dickins5d337b92005-09-03 15:54:41 -07001589 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001590 goto out_dput;
1591 }
1592 if (!security_vm_enough_memory(p->pages))
1593 vm_unacct_memory(p->pages);
1594 else {
1595 err = -ENOMEM;
Hugh Dickins5d337b92005-09-03 15:54:41 -07001596 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001597 goto out_dput;
1598 }
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001599 if (prev < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001600 swap_list.head = p->next;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001601 else
1602 swap_info[prev]->next = p->next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001603 if (type == swap_list.next) {
1604 /* just pick something that's safe... */
1605 swap_list.next = swap_list.head;
1606 }
Hugh Dickins78ecba02008-07-23 21:28:23 -07001607 if (p->prio < 0) {
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001608 for (i = p->next; i >= 0; i = swap_info[i]->next)
1609 swap_info[i]->prio = p->prio--;
Hugh Dickins78ecba02008-07-23 21:28:23 -07001610 least_priority++;
1611 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001612 nr_swap_pages -= p->pages;
1613 total_swap_pages -= p->pages;
1614 p->flags &= ~SWP_WRITEOK;
Hugh Dickins5d337b92005-09-03 15:54:41 -07001615 spin_unlock(&swap_lock);
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -07001616
David Rientjes72788c32011-05-24 17:11:40 -07001617 oom_score_adj = test_set_oom_score_adj(OOM_SCORE_ADJ_MAX);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001618 err = try_to_unuse(type);
David Rientjes43362a42011-10-31 17:07:18 -07001619 compare_swap_oom_score_adj(OOM_SCORE_ADJ_MAX, oom_score_adj);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001620
Linus Torvalds1da177e2005-04-16 15:20:36 -07001621 if (err) {
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07001622 /*
1623 * reading p->prio and p->swap_map outside the lock is
1624 * safe here because only sys_swapon and sys_swapoff
1625 * change them, and there can be no other sys_swapon or
1626 * sys_swapoff for this swap_info_struct at this point.
1627 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001628 /* re-insert swap space back into swap_list */
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07001629 enable_swap_info(p, p->prio, p->swap_map);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001630 goto out_dput;
1631 }
Hugh Dickins52b7efdb2005-09-03 15:54:39 -07001632
Hugh Dickins4cd3bb12005-09-03 15:54:33 -07001633 destroy_swap_extents(p);
Hugh Dickins570a335b2009-12-14 17:58:46 -08001634 if (p->flags & SWP_CONTINUED)
1635 free_swap_count_continuations(p);
1636
Ingo Molnarfc0abb12006-01-18 17:42:33 -08001637 mutex_lock(&swapon_mutex);
Hugh Dickins5d337b92005-09-03 15:54:41 -07001638 spin_lock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001639 drain_mmlist();
Hugh Dickins5d337b92005-09-03 15:54:41 -07001640
1641 /* wait for anyone still in scan_swap_map */
1642 p->highest_bit = 0; /* cuts scans short */
1643 while (p->flags >= SWP_SCANNING) {
1644 spin_unlock(&swap_lock);
Nishanth Aravamudan13e4b572005-09-10 00:27:25 -07001645 schedule_timeout_uninterruptible(1);
Hugh Dickins5d337b92005-09-03 15:54:41 -07001646 spin_lock(&swap_lock);
1647 }
1648
Linus Torvalds1da177e2005-04-16 15:20:36 -07001649 swap_file = p->swap_file;
1650 p->swap_file = NULL;
1651 p->max = 0;
1652 swap_map = p->swap_map;
1653 p->swap_map = NULL;
1654 p->flags = 0;
Hugh Dickins5d337b92005-09-03 15:54:41 -07001655 spin_unlock(&swap_lock);
Ingo Molnarfc0abb12006-01-18 17:42:33 -08001656 mutex_unlock(&swapon_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001657 vfree(swap_map);
KAMEZAWA Hiroyuki27a7faa2009-01-07 18:07:58 -08001658 /* Destroy swap account informatin */
1659 swap_cgroup_swapoff(type);
1660
Linus Torvalds1da177e2005-04-16 15:20:36 -07001661 inode = mapping->host;
1662 if (S_ISBLK(inode->i_mode)) {
1663 struct block_device *bdev = I_BDEV(inode);
1664 set_blocksize(bdev, p->old_block_size);
Tejun Heoe525fd82010-11-13 11:55:17 +01001665 blkdev_put(bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001666 } else {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001667 mutex_lock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001668 inode->i_flags &= ~S_SWAPFILE;
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001669 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001670 }
1671 filp_close(swap_file, NULL);
1672 err = 0;
Kay Sievers66d7dd52010-10-26 14:22:06 -07001673 atomic_inc(&proc_poll_event);
1674 wake_up_interruptible(&proc_poll_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001675
1676out_dput:
1677 filp_close(victim, NULL);
1678out:
1679 return err;
1680}
1681
1682#ifdef CONFIG_PROC_FS
Kay Sievers66d7dd52010-10-26 14:22:06 -07001683static unsigned swaps_poll(struct file *file, poll_table *wait)
1684{
Kay Sieversf1514632011-07-12 20:48:39 +02001685 struct seq_file *seq = file->private_data;
Kay Sievers66d7dd52010-10-26 14:22:06 -07001686
1687 poll_wait(file, &proc_poll_wait, wait);
1688
Kay Sieversf1514632011-07-12 20:48:39 +02001689 if (seq->poll_event != atomic_read(&proc_poll_event)) {
1690 seq->poll_event = atomic_read(&proc_poll_event);
Kay Sievers66d7dd52010-10-26 14:22:06 -07001691 return POLLIN | POLLRDNORM | POLLERR | POLLPRI;
1692 }
1693
1694 return POLLIN | POLLRDNORM;
1695}
1696
Linus Torvalds1da177e2005-04-16 15:20:36 -07001697/* iterator */
1698static void *swap_start(struct seq_file *swap, loff_t *pos)
1699{
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001700 struct swap_info_struct *si;
1701 int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001702 loff_t l = *pos;
1703
Ingo Molnarfc0abb12006-01-18 17:42:33 -08001704 mutex_lock(&swapon_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001705
Suleiman Souhlal881e4aa2006-12-06 20:32:28 -08001706 if (!l)
1707 return SEQ_START_TOKEN;
1708
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001709 for (type = 0; type < nr_swapfiles; type++) {
1710 smp_rmb(); /* read nr_swapfiles before swap_info[type] */
1711 si = swap_info[type];
1712 if (!(si->flags & SWP_USED) || !si->swap_map)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001713 continue;
Suleiman Souhlal881e4aa2006-12-06 20:32:28 -08001714 if (!--l)
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001715 return si;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001716 }
1717
1718 return NULL;
1719}
1720
1721static void *swap_next(struct seq_file *swap, void *v, loff_t *pos)
1722{
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001723 struct swap_info_struct *si = v;
1724 int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001725
Suleiman Souhlal881e4aa2006-12-06 20:32:28 -08001726 if (v == SEQ_START_TOKEN)
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001727 type = 0;
1728 else
1729 type = si->type + 1;
Suleiman Souhlal881e4aa2006-12-06 20:32:28 -08001730
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001731 for (; type < nr_swapfiles; type++) {
1732 smp_rmb(); /* read nr_swapfiles before swap_info[type] */
1733 si = swap_info[type];
1734 if (!(si->flags & SWP_USED) || !si->swap_map)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001735 continue;
1736 ++*pos;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001737 return si;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001738 }
1739
1740 return NULL;
1741}
1742
1743static void swap_stop(struct seq_file *swap, void *v)
1744{
Ingo Molnarfc0abb12006-01-18 17:42:33 -08001745 mutex_unlock(&swapon_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001746}
1747
1748static int swap_show(struct seq_file *swap, void *v)
1749{
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001750 struct swap_info_struct *si = v;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001751 struct file *file;
1752 int len;
1753
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001754 if (si == SEQ_START_TOKEN) {
Suleiman Souhlal881e4aa2006-12-06 20:32:28 -08001755 seq_puts(swap,"Filename\t\t\t\tType\t\tSize\tUsed\tPriority\n");
1756 return 0;
1757 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001758
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001759 file = si->swap_file;
Jan Blunckc32c2f62008-02-14 19:38:43 -08001760 len = seq_path(swap, &file->f_path, " \t\n\\");
Hugh Dickins6eb396d2005-09-03 15:54:35 -07001761 seq_printf(swap, "%*s%s\t%u\t%u\t%d\n",
Hugh Dickins886bb7e2009-01-06 14:39:48 -08001762 len < 40 ? 40 - len : 1, " ",
1763 S_ISBLK(file->f_path.dentry->d_inode->i_mode) ?
Linus Torvalds1da177e2005-04-16 15:20:36 -07001764 "partition" : "file\t",
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001765 si->pages << (PAGE_SHIFT - 10),
1766 si->inuse_pages << (PAGE_SHIFT - 10),
1767 si->prio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001768 return 0;
1769}
1770
Helge Deller15ad7cd2006-12-06 20:40:36 -08001771static const struct seq_operations swaps_op = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001772 .start = swap_start,
1773 .next = swap_next,
1774 .stop = swap_stop,
1775 .show = swap_show
1776};
1777
1778static int swaps_open(struct inode *inode, struct file *file)
1779{
Kay Sieversf1514632011-07-12 20:48:39 +02001780 struct seq_file *seq;
Kay Sievers66d7dd52010-10-26 14:22:06 -07001781 int ret;
1782
Kay Sievers66d7dd52010-10-26 14:22:06 -07001783 ret = seq_open(file, &swaps_op);
Kay Sieversf1514632011-07-12 20:48:39 +02001784 if (ret)
Kay Sievers66d7dd52010-10-26 14:22:06 -07001785 return ret;
Kay Sievers66d7dd52010-10-26 14:22:06 -07001786
Kay Sieversf1514632011-07-12 20:48:39 +02001787 seq = file->private_data;
1788 seq->poll_event = atomic_read(&proc_poll_event);
1789 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001790}
1791
Helge Deller15ad7cd2006-12-06 20:40:36 -08001792static const struct file_operations proc_swaps_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001793 .open = swaps_open,
1794 .read = seq_read,
1795 .llseek = seq_lseek,
1796 .release = seq_release,
Kay Sievers66d7dd52010-10-26 14:22:06 -07001797 .poll = swaps_poll,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001798};
1799
1800static int __init procswaps_init(void)
1801{
Denis V. Lunev3d71f862008-04-29 01:02:13 -07001802 proc_create("swaps", 0, NULL, &proc_swaps_operations);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001803 return 0;
1804}
1805__initcall(procswaps_init);
1806#endif /* CONFIG_PROC_FS */
1807
Jan Beulich17963162008-12-16 11:35:24 +00001808#ifdef MAX_SWAPFILES_CHECK
1809static int __init max_swapfiles_check(void)
1810{
1811 MAX_SWAPFILES_CHECK();
1812 return 0;
1813}
1814late_initcall(max_swapfiles_check);
1815#endif
1816
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07001817static struct swap_info_struct *alloc_swap_info(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001818{
Hugh Dickins73c34b62009-12-14 17:58:43 -08001819 struct swap_info_struct *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001820 unsigned int type;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001821
1822 p = kzalloc(sizeof(*p), GFP_KERNEL);
1823 if (!p)
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07001824 return ERR_PTR(-ENOMEM);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001825
Hugh Dickins5d337b92005-09-03 15:54:41 -07001826 spin_lock(&swap_lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001827 for (type = 0; type < nr_swapfiles; type++) {
1828 if (!(swap_info[type]->flags & SWP_USED))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001829 break;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001830 }
Christoph Lameter06972122006-06-23 02:03:35 -07001831 if (type >= MAX_SWAPFILES) {
Hugh Dickins5d337b92005-09-03 15:54:41 -07001832 spin_unlock(&swap_lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001833 kfree(p);
Cesar Eduardo Barros730c0582011-03-22 16:33:19 -07001834 return ERR_PTR(-EPERM);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001835 }
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001836 if (type >= nr_swapfiles) {
1837 p->type = type;
1838 swap_info[type] = p;
1839 /*
1840 * Write swap_info[type] before nr_swapfiles, in case a
1841 * racing procfs swap_start() or swap_next() is reading them.
1842 * (We never shrink nr_swapfiles, we never free this entry.)
1843 */
1844 smp_wmb();
1845 nr_swapfiles++;
1846 } else {
1847 kfree(p);
1848 p = swap_info[type];
1849 /*
1850 * Do not memset this entry: a racing procfs swap_next()
1851 * would be relying on p->type to remain valid.
1852 */
1853 }
Hugh Dickins9625a5f2009-12-14 17:58:42 -08001854 INIT_LIST_HEAD(&p->first_swap_extent.list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001855 p->flags = SWP_USED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001856 p->next = -1;
Hugh Dickins5d337b92005-09-03 15:54:41 -07001857 spin_unlock(&swap_lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001858
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07001859 return p;
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07001860}
1861
Cesar Eduardo Barros4d0e1e12011-03-22 16:33:26 -07001862static int claim_swapfile(struct swap_info_struct *p, struct inode *inode)
1863{
1864 int error;
1865
1866 if (S_ISBLK(inode->i_mode)) {
1867 p->bdev = bdgrab(I_BDEV(inode));
1868 error = blkdev_get(p->bdev,
1869 FMODE_READ | FMODE_WRITE | FMODE_EXCL,
1870 sys_swapon);
1871 if (error < 0) {
1872 p->bdev = NULL;
Cesar Eduardo Barros87ade722011-03-22 16:33:27 -07001873 return -EINVAL;
Cesar Eduardo Barros4d0e1e12011-03-22 16:33:26 -07001874 }
1875 p->old_block_size = block_size(p->bdev);
1876 error = set_blocksize(p->bdev, PAGE_SIZE);
1877 if (error < 0)
Cesar Eduardo Barros87ade722011-03-22 16:33:27 -07001878 return error;
Cesar Eduardo Barros4d0e1e12011-03-22 16:33:26 -07001879 p->flags |= SWP_BLKDEV;
1880 } else if (S_ISREG(inode->i_mode)) {
1881 p->bdev = inode->i_sb->s_bdev;
1882 mutex_lock(&inode->i_mutex);
Cesar Eduardo Barros87ade722011-03-22 16:33:27 -07001883 if (IS_SWAPFILE(inode))
1884 return -EBUSY;
1885 } else
1886 return -EINVAL;
Cesar Eduardo Barros4d0e1e12011-03-22 16:33:26 -07001887
1888 return 0;
Cesar Eduardo Barros4d0e1e12011-03-22 16:33:26 -07001889}
1890
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07001891static unsigned long read_swap_header(struct swap_info_struct *p,
1892 union swap_header *swap_header,
1893 struct inode *inode)
1894{
1895 int i;
1896 unsigned long maxpages;
1897 unsigned long swapfilepages;
1898
1899 if (memcmp("SWAPSPACE2", swap_header->magic.magic, 10)) {
1900 printk(KERN_ERR "Unable to find swap-space signature\n");
Cesar Eduardo Barros38719022011-03-22 16:33:30 -07001901 return 0;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07001902 }
1903
1904 /* swap partition endianess hack... */
1905 if (swab32(swap_header->info.version) == 1) {
1906 swab32s(&swap_header->info.version);
1907 swab32s(&swap_header->info.last_page);
1908 swab32s(&swap_header->info.nr_badpages);
1909 for (i = 0; i < swap_header->info.nr_badpages; i++)
1910 swab32s(&swap_header->info.badpages[i]);
1911 }
1912 /* Check the swap header's sub-version */
1913 if (swap_header->info.version != 1) {
1914 printk(KERN_WARNING
1915 "Unable to handle swap header version %d\n",
1916 swap_header->info.version);
Cesar Eduardo Barros38719022011-03-22 16:33:30 -07001917 return 0;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07001918 }
1919
1920 p->lowest_bit = 1;
1921 p->cluster_next = 1;
1922 p->cluster_nr = 0;
1923
1924 /*
1925 * Find out how many pages are allowed for a single swap
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001926 * device. There are three limiting factors: 1) the number
1927 * of bits for the swap offset in the swp_entry_t type, and
1928 * 2) the number of bits in the swap pte as defined by the
1929 * the different architectures, and 3) the number of free bits
1930 * in an exceptional radix_tree entry. In order to find the
1931 * largest possible bit mask, a swap entry with swap type 0
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07001932 * and swap offset ~0UL is created, encoded to a swap pte,
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001933 * decoded to a swp_entry_t again, and finally the swap
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07001934 * offset is extracted. This will mask all the bits from
1935 * the initial ~0UL mask that can't be encoded in either
1936 * the swp_entry_t or the architecture definition of a
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001937 * swap pte. Then the same is done for a radix_tree entry.
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07001938 */
1939 maxpages = swp_offset(pte_to_swp_entry(
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001940 swp_entry_to_pte(swp_entry(0, ~0UL))));
1941 maxpages = swp_offset(radix_to_swp_entry(
1942 swp_to_radix_entry(swp_entry(0, maxpages)))) + 1;
1943
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07001944 if (maxpages > swap_header->info.last_page) {
1945 maxpages = swap_header->info.last_page + 1;
1946 /* p->max is an unsigned int: don't overflow it */
1947 if ((unsigned int)maxpages == 0)
1948 maxpages = UINT_MAX;
1949 }
1950 p->highest_bit = maxpages - 1;
1951
1952 if (!maxpages)
Cesar Eduardo Barros38719022011-03-22 16:33:30 -07001953 return 0;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07001954 swapfilepages = i_size_read(inode) >> PAGE_SHIFT;
1955 if (swapfilepages && maxpages > swapfilepages) {
1956 printk(KERN_WARNING
1957 "Swap area shorter than signature indicates\n");
Cesar Eduardo Barros38719022011-03-22 16:33:30 -07001958 return 0;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07001959 }
1960 if (swap_header->info.nr_badpages && S_ISREG(inode->i_mode))
Cesar Eduardo Barros38719022011-03-22 16:33:30 -07001961 return 0;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07001962 if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES)
Cesar Eduardo Barros38719022011-03-22 16:33:30 -07001963 return 0;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07001964
1965 return maxpages;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07001966}
1967
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07001968static int setup_swap_map_and_extents(struct swap_info_struct *p,
1969 union swap_header *swap_header,
1970 unsigned char *swap_map,
1971 unsigned long maxpages,
1972 sector_t *span)
1973{
1974 int i;
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07001975 unsigned int nr_good_pages;
1976 int nr_extents;
1977
1978 nr_good_pages = maxpages - 1; /* omit header page */
1979
1980 for (i = 0; i < swap_header->info.nr_badpages; i++) {
1981 unsigned int page_nr = swap_header->info.badpages[i];
Cesar Eduardo Barrosbdb8e3f2011-03-22 16:33:33 -07001982 if (page_nr == 0 || page_nr > swap_header->info.last_page)
1983 return -EINVAL;
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07001984 if (page_nr < maxpages) {
1985 swap_map[page_nr] = SWAP_MAP_BAD;
1986 nr_good_pages--;
1987 }
1988 }
1989
1990 if (nr_good_pages) {
1991 swap_map[0] = SWAP_MAP_BAD;
1992 p->max = maxpages;
1993 p->pages = nr_good_pages;
1994 nr_extents = setup_swap_extents(p, span);
Cesar Eduardo Barrosbdb8e3f2011-03-22 16:33:33 -07001995 if (nr_extents < 0)
1996 return nr_extents;
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07001997 nr_good_pages = p->pages;
1998 }
1999 if (!nr_good_pages) {
2000 printk(KERN_WARNING "Empty swap-file\n");
Cesar Eduardo Barrosbdb8e3f2011-03-22 16:33:33 -07002001 return -EINVAL;
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002002 }
2003
2004 return nr_extents;
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002005}
2006
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002007SYSCALL_DEFINE2(swapon, const char __user *, specialfile, int, swap_flags)
2008{
2009 struct swap_info_struct *p;
Cesar Eduardo Barros28b36bd2011-03-22 16:33:20 -07002010 char *name;
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002011 struct file *swap_file = NULL;
2012 struct address_space *mapping;
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07002013 int i;
2014 int prio;
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002015 int error;
2016 union swap_header *swap_header;
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002017 int nr_extents;
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002018 sector_t span;
2019 unsigned long maxpages;
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002020 unsigned char *swap_map = NULL;
2021 struct page *page = NULL;
2022 struct inode *inode = NULL;
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002023
2024 if (!capable(CAP_SYS_ADMIN))
2025 return -EPERM;
2026
2027 p = alloc_swap_info();
Cesar Eduardo Barros2542e5132011-03-22 16:33:18 -07002028 if (IS_ERR(p))
2029 return PTR_ERR(p);
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002030
Linus Torvalds1da177e2005-04-16 15:20:36 -07002031 name = getname(specialfile);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002032 if (IS_ERR(name)) {
Cesar Eduardo Barros7de7fb62011-03-22 16:33:22 -07002033 error = PTR_ERR(name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002034 name = NULL;
Cesar Eduardo Barrosbd690102011-03-22 16:33:25 -07002035 goto bad_swap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002036 }
2037 swap_file = filp_open(name, O_RDWR|O_LARGEFILE, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002038 if (IS_ERR(swap_file)) {
Cesar Eduardo Barros7de7fb62011-03-22 16:33:22 -07002039 error = PTR_ERR(swap_file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002040 swap_file = NULL;
Cesar Eduardo Barrosbd690102011-03-22 16:33:25 -07002041 goto bad_swap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002042 }
2043
2044 p->swap_file = swap_file;
2045 mapping = swap_file->f_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002046
Linus Torvalds1da177e2005-04-16 15:20:36 -07002047 for (i = 0; i < nr_swapfiles; i++) {
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002048 struct swap_info_struct *q = swap_info[i];
Linus Torvalds1da177e2005-04-16 15:20:36 -07002049
Cesar Eduardo Barrose8e6c2e2011-03-22 16:33:16 -07002050 if (q == p || !q->swap_file)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002051 continue;
Cesar Eduardo Barros7de7fb62011-03-22 16:33:22 -07002052 if (mapping == q->swap_file->f_mapping) {
2053 error = -EBUSY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002054 goto bad_swap;
Cesar Eduardo Barros7de7fb62011-03-22 16:33:22 -07002055 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002056 }
2057
Cesar Eduardo Barros21307812011-03-22 23:03:13 -03002058 inode = mapping->host;
2059 /* If S_ISREG(inode->i_mode) will do mutex_lock(&inode->i_mutex); */
Cesar Eduardo Barros4d0e1e12011-03-22 16:33:26 -07002060 error = claim_swapfile(p, inode);
2061 if (unlikely(error))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002062 goto bad_swap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002063
Linus Torvalds1da177e2005-04-16 15:20:36 -07002064 /*
2065 * Read the swap header.
2066 */
2067 if (!mapping->a_ops->readpage) {
2068 error = -EINVAL;
2069 goto bad_swap;
2070 }
Pekka Enberg090d2b12006-06-23 02:05:08 -07002071 page = read_mapping_page(mapping, 0, swap_file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002072 if (IS_ERR(page)) {
2073 error = PTR_ERR(page);
2074 goto bad_swap;
2075 }
Hugh Dickins81e33972009-01-06 14:39:49 -08002076 swap_header = kmap(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002077
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002078 maxpages = read_swap_header(p, swap_header, inode);
2079 if (unlikely(!maxpages)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002080 error = -EINVAL;
2081 goto bad_swap;
2082 }
Hugh Dickins886bb7e2009-01-06 14:39:48 -08002083
Hugh Dickins81e33972009-01-06 14:39:49 -08002084 /* OK, set up the swap map and apply the bad block list */
Cesar Eduardo Barros803d0c832011-03-22 16:33:14 -07002085 swap_map = vzalloc(maxpages);
Hugh Dickins81e33972009-01-06 14:39:49 -08002086 if (!swap_map) {
2087 error = -ENOMEM;
2088 goto bad_swap;
2089 }
2090
Cesar Eduardo Barros1421ef32011-03-22 16:33:31 -07002091 error = swap_cgroup_swapon(p->type, maxpages);
2092 if (error)
2093 goto bad_swap;
2094
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002095 nr_extents = setup_swap_map_and_extents(p, swap_header, swap_map,
2096 maxpages, &span);
2097 if (unlikely(nr_extents < 0)) {
2098 error = nr_extents;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002099 goto bad_swap;
2100 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002101
Suresh Jayaraman3bd0f0c2009-09-30 10:53:48 +02002102 if (p->bdev) {
2103 if (blk_queue_nonrot(bdev_get_queue(p->bdev))) {
2104 p->flags |= SWP_SOLIDSTATE;
2105 p->cluster_next = 1 + (random32() % p->highest_bit);
2106 }
Hugh Dickins33994462010-09-09 16:38:11 -07002107 if (discard_swap(p) == 0 && (swap_flags & SWAP_FLAG_DISCARD))
Suresh Jayaraman3bd0f0c2009-09-30 10:53:48 +02002108 p->flags |= SWP_DISCARDABLE;
Hugh Dickins20137a42009-01-06 14:39:54 -08002109 }
Hugh Dickins6a6ba832009-01-06 14:39:51 -08002110
Ingo Molnarfc0abb12006-01-18 17:42:33 -08002111 mutex_lock(&swapon_mutex);
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07002112 prio = -1;
Hugh Dickins78ecba02008-07-23 21:28:23 -07002113 if (swap_flags & SWAP_FLAG_PREFER)
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07002114 prio =
Hugh Dickins78ecba02008-07-23 21:28:23 -07002115 (swap_flags & SWAP_FLAG_PRIO_MASK) >> SWAP_FLAG_PRIO_SHIFT;
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07002116 enable_swap_info(p, prio, swap_map);
Cesar Eduardo Barrosc69dbfb2011-03-22 16:33:35 -07002117
2118 printk(KERN_INFO "Adding %uk swap on %s. "
2119 "Priority:%d extents:%d across:%lluk %s%s\n",
2120 p->pages<<(PAGE_SHIFT-10), name, p->prio,
2121 nr_extents, (unsigned long long)span<<(PAGE_SHIFT-10),
2122 (p->flags & SWP_SOLIDSTATE) ? "SS" : "",
2123 (p->flags & SWP_DISCARDABLE) ? "D" : "");
2124
Ingo Molnarfc0abb12006-01-18 17:42:33 -08002125 mutex_unlock(&swapon_mutex);
Kay Sievers66d7dd52010-10-26 14:22:06 -07002126 atomic_inc(&proc_poll_event);
2127 wake_up_interruptible(&proc_poll_wait);
2128
Cesar Eduardo Barros9b01c352011-03-22 16:33:24 -07002129 if (S_ISREG(inode->i_mode))
2130 inode->i_flags |= S_SWAPFILE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002131 error = 0;
2132 goto out;
2133bad_swap:
Cesar Eduardo Barrosbd690102011-03-22 16:33:25 -07002134 if (inode && S_ISBLK(inode->i_mode) && p->bdev) {
Cesar Eduardo Barrosf2090d22011-03-22 16:33:23 -07002135 set_blocksize(p->bdev, p->old_block_size);
2136 blkdev_put(p->bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002137 }
Hugh Dickins4cd3bb12005-09-03 15:54:33 -07002138 destroy_swap_extents(p);
Cesar Eduardo Barrose8e6c2e2011-03-22 16:33:16 -07002139 swap_cgroup_swapoff(p->type);
Hugh Dickins5d337b92005-09-03 15:54:41 -07002140 spin_lock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002141 p->swap_file = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002142 p->flags = 0;
Hugh Dickins5d337b92005-09-03 15:54:41 -07002143 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002144 vfree(swap_map);
Mel Gorman52c50562011-03-22 16:30:08 -07002145 if (swap_file) {
Cesar Eduardo Barros21307812011-03-22 23:03:13 -03002146 if (inode && S_ISREG(inode->i_mode)) {
Mel Gorman52c50562011-03-22 16:30:08 -07002147 mutex_unlock(&inode->i_mutex);
Cesar Eduardo Barros21307812011-03-22 23:03:13 -03002148 inode = NULL;
2149 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002150 filp_close(swap_file, NULL);
Mel Gorman52c50562011-03-22 16:30:08 -07002151 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002152out:
2153 if (page && !IS_ERR(page)) {
2154 kunmap(page);
2155 page_cache_release(page);
2156 }
2157 if (name)
2158 putname(name);
Cesar Eduardo Barros9b01c352011-03-22 16:33:24 -07002159 if (inode && S_ISREG(inode->i_mode))
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002160 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002161 return error;
2162}
2163
2164void si_swapinfo(struct sysinfo *val)
2165{
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002166 unsigned int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002167 unsigned long nr_to_be_unused = 0;
2168
Hugh Dickins5d337b92005-09-03 15:54:41 -07002169 spin_lock(&swap_lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002170 for (type = 0; type < nr_swapfiles; type++) {
2171 struct swap_info_struct *si = swap_info[type];
2172
2173 if ((si->flags & SWP_USED) && !(si->flags & SWP_WRITEOK))
2174 nr_to_be_unused += si->inuse_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002175 }
2176 val->freeswap = nr_swap_pages + nr_to_be_unused;
2177 val->totalswap = total_swap_pages + nr_to_be_unused;
Hugh Dickins5d337b92005-09-03 15:54:41 -07002178 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002179}
2180
2181/*
2182 * Verify that a swap entry is valid and increment its swap map count.
2183 *
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002184 * Returns error code in following case.
2185 * - success -> 0
2186 * - swp_entry is invalid -> EINVAL
2187 * - swp_entry is migration entry -> EINVAL
2188 * - swap-cache reference is requested but there is already one. -> EEXIST
2189 * - swap-cache reference is requested but the entry is not used. -> ENOENT
Hugh Dickins570a335b2009-12-14 17:58:46 -08002190 * - swap-mapped reference requested but needs continued swap count. -> ENOMEM
Linus Torvalds1da177e2005-04-16 15:20:36 -07002191 */
Hugh Dickins8d69aae2009-12-14 17:58:45 -08002192static int __swap_duplicate(swp_entry_t entry, unsigned char usage)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002193{
Hugh Dickins73c34b62009-12-14 17:58:43 -08002194 struct swap_info_struct *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002195 unsigned long offset, type;
Hugh Dickins8d69aae2009-12-14 17:58:45 -08002196 unsigned char count;
2197 unsigned char has_cache;
Hugh Dickins253d5532009-12-14 17:58:44 -08002198 int err = -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002199
Andi Kleena7420aa2009-09-16 11:50:05 +02002200 if (non_swap_entry(entry))
Hugh Dickins253d5532009-12-14 17:58:44 -08002201 goto out;
Christoph Lameter06972122006-06-23 02:03:35 -07002202
Linus Torvalds1da177e2005-04-16 15:20:36 -07002203 type = swp_type(entry);
2204 if (type >= nr_swapfiles)
2205 goto bad_file;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002206 p = swap_info[type];
Linus Torvalds1da177e2005-04-16 15:20:36 -07002207 offset = swp_offset(entry);
2208
Hugh Dickins5d337b92005-09-03 15:54:41 -07002209 spin_lock(&swap_lock);
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002210 if (unlikely(offset >= p->max))
2211 goto unlock_out;
2212
Hugh Dickins253d5532009-12-14 17:58:44 -08002213 count = p->swap_map[offset];
2214 has_cache = count & SWAP_HAS_CACHE;
2215 count &= ~SWAP_HAS_CACHE;
2216 err = 0;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002217
Hugh Dickins253d5532009-12-14 17:58:44 -08002218 if (usage == SWAP_HAS_CACHE) {
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002219
2220 /* set SWAP_HAS_CACHE if there is no cache and entry is used */
Hugh Dickins253d5532009-12-14 17:58:44 -08002221 if (!has_cache && count)
2222 has_cache = SWAP_HAS_CACHE;
2223 else if (has_cache) /* someone else added cache */
2224 err = -EEXIST;
2225 else /* no users remaining */
2226 err = -ENOENT;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002227
2228 } else if (count || has_cache) {
Hugh Dickins253d5532009-12-14 17:58:44 -08002229
Hugh Dickins570a335b2009-12-14 17:58:46 -08002230 if ((count & ~COUNT_CONTINUED) < SWAP_MAP_MAX)
2231 count += usage;
2232 else if ((count & ~COUNT_CONTINUED) > SWAP_MAP_MAX)
Hugh Dickins253d5532009-12-14 17:58:44 -08002233 err = -EINVAL;
Hugh Dickins570a335b2009-12-14 17:58:46 -08002234 else if (swap_count_continued(p, offset, count))
2235 count = COUNT_CONTINUED;
2236 else
2237 err = -ENOMEM;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002238 } else
Hugh Dickins253d5532009-12-14 17:58:44 -08002239 err = -ENOENT; /* unused swap entry */
2240
2241 p->swap_map[offset] = count | has_cache;
2242
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002243unlock_out:
Hugh Dickins5d337b92005-09-03 15:54:41 -07002244 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002245out:
Hugh Dickins253d5532009-12-14 17:58:44 -08002246 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002247
2248bad_file:
2249 printk(KERN_ERR "swap_dup: %s%08lx\n", Bad_file, entry.val);
2250 goto out;
2251}
Hugh Dickins253d5532009-12-14 17:58:44 -08002252
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002253/*
Hugh Dickinsaaa46862009-12-14 17:58:47 -08002254 * Help swapoff by noting that swap entry belongs to shmem/tmpfs
2255 * (in which case its reference count is never incremented).
2256 */
2257void swap_shmem_alloc(swp_entry_t entry)
2258{
2259 __swap_duplicate(entry, SWAP_MAP_SHMEM);
2260}
2261
2262/*
Hugh Dickins08259d52010-03-05 13:42:25 -08002263 * Increase reference count of swap entry by 1.
2264 * Returns 0 for success, or -ENOMEM if a swap_count_continuation is required
2265 * but could not be atomically allocated. Returns 0, just as if it succeeded,
2266 * if __swap_duplicate() fails for another reason (-EINVAL or -ENOENT), which
2267 * might occur if a page table entry has got corrupted.
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002268 */
Hugh Dickins570a335b2009-12-14 17:58:46 -08002269int swap_duplicate(swp_entry_t entry)
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002270{
Hugh Dickins570a335b2009-12-14 17:58:46 -08002271 int err = 0;
2272
2273 while (!err && __swap_duplicate(entry, 1) == -ENOMEM)
2274 err = add_swap_count_continuation(entry, GFP_ATOMIC);
2275 return err;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002276}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002277
KAMEZAWA Hiroyukicb4b86b2009-06-16 15:32:52 -07002278/*
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002279 * @entry: swap entry for which we allocate swap cache.
2280 *
Hugh Dickins73c34b62009-12-14 17:58:43 -08002281 * Called when allocating swap cache for existing swap entry,
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002282 * This can return error codes. Returns 0 at success.
2283 * -EBUSY means there is a swap cache.
2284 * Note: return code is different from swap_duplicate().
KAMEZAWA Hiroyukicb4b86b2009-06-16 15:32:52 -07002285 */
2286int swapcache_prepare(swp_entry_t entry)
2287{
Hugh Dickins253d5532009-12-14 17:58:44 -08002288 return __swap_duplicate(entry, SWAP_HAS_CACHE);
KAMEZAWA Hiroyukicb4b86b2009-06-16 15:32:52 -07002289}
2290
Linus Torvalds1da177e2005-04-16 15:20:36 -07002291/*
Hugh Dickins5d337b92005-09-03 15:54:41 -07002292 * swap_lock prevents swap_map being freed. Don't grab an extra
Linus Torvalds1da177e2005-04-16 15:20:36 -07002293 * reference on the swaphandle, it doesn't matter if it becomes unused.
2294 */
2295int valid_swaphandles(swp_entry_t entry, unsigned long *offset)
2296{
Hugh Dickins89528982008-02-04 22:28:45 -08002297 struct swap_info_struct *si;
Hugh Dickins3f9e7942006-09-29 02:01:26 -07002298 int our_page_cluster = page_cluster;
Hugh Dickins89528982008-02-04 22:28:45 -08002299 pgoff_t target, toff;
2300 pgoff_t base, end;
2301 int nr_pages = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002302
Hugh Dickins3f9e7942006-09-29 02:01:26 -07002303 if (!our_page_cluster) /* no readahead */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002304 return 0;
Hugh Dickins89528982008-02-04 22:28:45 -08002305
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002306 si = swap_info[swp_type(entry)];
Hugh Dickins89528982008-02-04 22:28:45 -08002307 target = swp_offset(entry);
2308 base = (target >> our_page_cluster) << our_page_cluster;
2309 end = base + (1 << our_page_cluster);
2310 if (!base) /* first page is swap header */
2311 base++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002312
Hugh Dickins5d337b92005-09-03 15:54:41 -07002313 spin_lock(&swap_lock);
Hugh Dickins89528982008-02-04 22:28:45 -08002314 if (end > si->max) /* don't go beyond end of map */
2315 end = si->max;
2316
2317 /* Count contiguous allocated slots above our target */
2318 for (toff = target; ++toff < end; nr_pages++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002319 /* Don't read in free or bad pages */
Hugh Dickins89528982008-02-04 22:28:45 -08002320 if (!si->swap_map[toff])
Linus Torvalds1da177e2005-04-16 15:20:36 -07002321 break;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002322 if (swap_count(si->swap_map[toff]) == SWAP_MAP_BAD)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002323 break;
Hugh Dickins89528982008-02-04 22:28:45 -08002324 }
2325 /* Count contiguous allocated slots below our target */
2326 for (toff = target; --toff >= base; nr_pages++) {
2327 /* Don't read in free or bad pages */
2328 if (!si->swap_map[toff])
2329 break;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002330 if (swap_count(si->swap_map[toff]) == SWAP_MAP_BAD)
Hugh Dickins89528982008-02-04 22:28:45 -08002331 break;
2332 }
Hugh Dickins5d337b92005-09-03 15:54:41 -07002333 spin_unlock(&swap_lock);
Hugh Dickins89528982008-02-04 22:28:45 -08002334
2335 /*
2336 * Indicate starting offset, and return number of pages to get:
2337 * if only 1, say 0, since there's then no readahead to be done.
2338 */
2339 *offset = ++toff;
2340 return nr_pages? ++nr_pages: 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002341}
Hugh Dickins570a335b2009-12-14 17:58:46 -08002342
2343/*
2344 * add_swap_count_continuation - called when a swap count is duplicated
2345 * beyond SWAP_MAP_MAX, it allocates a new page and links that to the entry's
2346 * page of the original vmalloc'ed swap_map, to hold the continuation count
2347 * (for that entry and for its neighbouring PAGE_SIZE swap entries). Called
2348 * again when count is duplicated beyond SWAP_MAP_MAX * SWAP_CONT_MAX, etc.
2349 *
2350 * These continuation pages are seldom referenced: the common paths all work
2351 * on the original swap_map, only referring to a continuation page when the
2352 * low "digit" of a count is incremented or decremented through SWAP_MAP_MAX.
2353 *
2354 * add_swap_count_continuation(, GFP_ATOMIC) can be called while holding
2355 * page table locks; if it fails, add_swap_count_continuation(, GFP_KERNEL)
2356 * can be called after dropping locks.
2357 */
2358int add_swap_count_continuation(swp_entry_t entry, gfp_t gfp_mask)
2359{
2360 struct swap_info_struct *si;
2361 struct page *head;
2362 struct page *page;
2363 struct page *list_page;
2364 pgoff_t offset;
2365 unsigned char count;
2366
2367 /*
2368 * When debugging, it's easier to use __GFP_ZERO here; but it's better
2369 * for latency not to zero a page while GFP_ATOMIC and holding locks.
2370 */
2371 page = alloc_page(gfp_mask | __GFP_HIGHMEM);
2372
2373 si = swap_info_get(entry);
2374 if (!si) {
2375 /*
2376 * An acceptable race has occurred since the failing
2377 * __swap_duplicate(): the swap entry has been freed,
2378 * perhaps even the whole swap_map cleared for swapoff.
2379 */
2380 goto outer;
2381 }
2382
2383 offset = swp_offset(entry);
2384 count = si->swap_map[offset] & ~SWAP_HAS_CACHE;
2385
2386 if ((count & ~COUNT_CONTINUED) != SWAP_MAP_MAX) {
2387 /*
2388 * The higher the swap count, the more likely it is that tasks
2389 * will race to add swap count continuation: we need to avoid
2390 * over-provisioning.
2391 */
2392 goto out;
2393 }
2394
2395 if (!page) {
2396 spin_unlock(&swap_lock);
2397 return -ENOMEM;
2398 }
2399
2400 /*
2401 * We are fortunate that although vmalloc_to_page uses pte_offset_map,
2402 * no architecture is using highmem pages for kernel pagetables: so it
2403 * will not corrupt the GFP_ATOMIC caller's atomic pagetable kmaps.
2404 */
2405 head = vmalloc_to_page(si->swap_map + offset);
2406 offset &= ~PAGE_MASK;
2407
2408 /*
2409 * Page allocation does not initialize the page's lru field,
2410 * but it does always reset its private field.
2411 */
2412 if (!page_private(head)) {
2413 BUG_ON(count & COUNT_CONTINUED);
2414 INIT_LIST_HEAD(&head->lru);
2415 set_page_private(head, SWP_CONTINUED);
2416 si->flags |= SWP_CONTINUED;
2417 }
2418
2419 list_for_each_entry(list_page, &head->lru, lru) {
2420 unsigned char *map;
2421
2422 /*
2423 * If the previous map said no continuation, but we've found
2424 * a continuation page, free our allocation and use this one.
2425 */
2426 if (!(count & COUNT_CONTINUED))
2427 goto out;
2428
2429 map = kmap_atomic(list_page, KM_USER0) + offset;
2430 count = *map;
2431 kunmap_atomic(map, KM_USER0);
2432
2433 /*
2434 * If this continuation count now has some space in it,
2435 * free our allocation and use this one.
2436 */
2437 if ((count & ~COUNT_CONTINUED) != SWAP_CONT_MAX)
2438 goto out;
2439 }
2440
2441 list_add_tail(&page->lru, &head->lru);
2442 page = NULL; /* now it's attached, don't free it */
2443out:
2444 spin_unlock(&swap_lock);
2445outer:
2446 if (page)
2447 __free_page(page);
2448 return 0;
2449}
2450
2451/*
2452 * swap_count_continued - when the original swap_map count is incremented
2453 * from SWAP_MAP_MAX, check if there is already a continuation page to carry
2454 * into, carry if so, or else fail until a new continuation page is allocated;
2455 * when the original swap_map count is decremented from 0 with continuation,
2456 * borrow from the continuation and report whether it still holds more.
2457 * Called while __swap_duplicate() or swap_entry_free() holds swap_lock.
2458 */
2459static bool swap_count_continued(struct swap_info_struct *si,
2460 pgoff_t offset, unsigned char count)
2461{
2462 struct page *head;
2463 struct page *page;
2464 unsigned char *map;
2465
2466 head = vmalloc_to_page(si->swap_map + offset);
2467 if (page_private(head) != SWP_CONTINUED) {
2468 BUG_ON(count & COUNT_CONTINUED);
2469 return false; /* need to add count continuation */
2470 }
2471
2472 offset &= ~PAGE_MASK;
2473 page = list_entry(head->lru.next, struct page, lru);
2474 map = kmap_atomic(page, KM_USER0) + offset;
2475
2476 if (count == SWAP_MAP_MAX) /* initial increment from swap_map */
2477 goto init_map; /* jump over SWAP_CONT_MAX checks */
2478
2479 if (count == (SWAP_MAP_MAX | COUNT_CONTINUED)) { /* incrementing */
2480 /*
2481 * Think of how you add 1 to 999
2482 */
2483 while (*map == (SWAP_CONT_MAX | COUNT_CONTINUED)) {
2484 kunmap_atomic(map, KM_USER0);
2485 page = list_entry(page->lru.next, struct page, lru);
2486 BUG_ON(page == head);
2487 map = kmap_atomic(page, KM_USER0) + offset;
2488 }
2489 if (*map == SWAP_CONT_MAX) {
2490 kunmap_atomic(map, KM_USER0);
2491 page = list_entry(page->lru.next, struct page, lru);
2492 if (page == head)
2493 return false; /* add count continuation */
2494 map = kmap_atomic(page, KM_USER0) + offset;
2495init_map: *map = 0; /* we didn't zero the page */
2496 }
2497 *map += 1;
2498 kunmap_atomic(map, KM_USER0);
2499 page = list_entry(page->lru.prev, struct page, lru);
2500 while (page != head) {
2501 map = kmap_atomic(page, KM_USER0) + offset;
2502 *map = COUNT_CONTINUED;
2503 kunmap_atomic(map, KM_USER0);
2504 page = list_entry(page->lru.prev, struct page, lru);
2505 }
2506 return true; /* incremented */
2507
2508 } else { /* decrementing */
2509 /*
2510 * Think of how you subtract 1 from 1000
2511 */
2512 BUG_ON(count != COUNT_CONTINUED);
2513 while (*map == COUNT_CONTINUED) {
2514 kunmap_atomic(map, KM_USER0);
2515 page = list_entry(page->lru.next, struct page, lru);
2516 BUG_ON(page == head);
2517 map = kmap_atomic(page, KM_USER0) + offset;
2518 }
2519 BUG_ON(*map == 0);
2520 *map -= 1;
2521 if (*map == 0)
2522 count = 0;
2523 kunmap_atomic(map, KM_USER0);
2524 page = list_entry(page->lru.prev, struct page, lru);
2525 while (page != head) {
2526 map = kmap_atomic(page, KM_USER0) + offset;
2527 *map = SWAP_CONT_MAX | count;
2528 count = COUNT_CONTINUED;
2529 kunmap_atomic(map, KM_USER0);
2530 page = list_entry(page->lru.prev, struct page, lru);
2531 }
2532 return count == COUNT_CONTINUED;
2533 }
2534}
2535
2536/*
2537 * free_swap_count_continuations - swapoff free all the continuation pages
2538 * appended to the swap_map, after swap_map is quiesced, before vfree'ing it.
2539 */
2540static void free_swap_count_continuations(struct swap_info_struct *si)
2541{
2542 pgoff_t offset;
2543
2544 for (offset = 0; offset < si->max; offset += PAGE_SIZE) {
2545 struct page *head;
2546 head = vmalloc_to_page(si->swap_map + offset);
2547 if (page_private(head)) {
2548 struct list_head *this, *next;
2549 list_for_each_safe(this, next, &head->lru) {
2550 struct page *page;
2551 page = list_entry(this, struct page, lru);
2552 list_del(this);
2553 __free_page(page);
2554 }
2555 }
2556 }
2557}