blob: fafc26d1b1dc885d2541eda3bdc5705a4fe56012 [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 *
Mel Gormanf90ac392012-01-10 15:07:15 -0800670 * Hibration suspends storage while it is writing the image
671 * to disk so check that here.
Hugh Dickinsb73d7fc2010-09-09 16:38:09 -0700672 */
Mel Gormanf90ac392012-01-10 15:07:15 -0800673 if (pm_suspended_storage())
Hugh Dickinsb73d7fc2010-09-09 16:38:09 -0700674 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{
Johannes Weiner72835c82012-01-12 17:18:32 -0800850 struct mem_cgroup *memcg;
Hugh Dickins044d66c2008-02-07 00:14:04 -0800851 spinlock_t *ptl;
852 pte_t *pte;
853 int ret = 1;
854
Johannes Weiner72835c82012-01-12 17:18:32 -0800855 if (mem_cgroup_try_charge_swapin(vma->vm_mm, page,
856 GFP_KERNEL, &memcg)) {
Hugh Dickins044d66c2008-02-07 00:14:04 -0800857 ret = -ENOMEM;
KAMEZAWA Hiroyuki85d9fc82009-01-29 14:25:13 -0800858 goto out_nolock;
859 }
Hugh Dickins044d66c2008-02-07 00:14:04 -0800860
861 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
862 if (unlikely(!pte_same(*pte, swp_entry_to_pte(entry)))) {
863 if (ret > 0)
Johannes Weiner72835c82012-01-12 17:18:32 -0800864 mem_cgroup_cancel_charge_swapin(memcg);
Hugh Dickins044d66c2008-02-07 00:14:04 -0800865 ret = 0;
866 goto out;
867 }
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800868
KAMEZAWA Hiroyukib084d432010-03-05 13:41:42 -0800869 dec_mm_counter(vma->vm_mm, MM_SWAPENTS);
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -0800870 inc_mm_counter(vma->vm_mm, MM_ANONPAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700871 get_page(page);
872 set_pte_at(vma->vm_mm, addr, pte,
873 pte_mkold(mk_pte(page, vma->vm_page_prot)));
874 page_add_anon_rmap(page, vma, addr);
Johannes Weiner72835c82012-01-12 17:18:32 -0800875 mem_cgroup_commit_charge_swapin(page, memcg);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700876 swap_free(entry);
877 /*
878 * Move the page to the active list so it is not
879 * immediately swapped out again after swapon.
880 */
881 activate_page(page);
Hugh Dickins044d66c2008-02-07 00:14:04 -0800882out:
883 pte_unmap_unlock(pte, ptl);
KAMEZAWA Hiroyuki85d9fc82009-01-29 14:25:13 -0800884out_nolock:
Hugh Dickins044d66c2008-02-07 00:14:04 -0800885 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700886}
887
888static int unuse_pte_range(struct vm_area_struct *vma, pmd_t *pmd,
889 unsigned long addr, unsigned long end,
890 swp_entry_t entry, struct page *page)
891{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700892 pte_t swp_pte = swp_entry_to_pte(entry);
Hugh Dickins705e87c2005-10-29 18:16:27 -0700893 pte_t *pte;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800894 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700895
Hugh Dickins044d66c2008-02-07 00:14:04 -0800896 /*
897 * We don't actually need pte lock while scanning for swp_pte: since
898 * we hold page lock and mmap_sem, swp_pte cannot be inserted into the
899 * page table while we're scanning; though it could get zapped, and on
900 * some architectures (e.g. x86_32 with PAE) we might catch a glimpse
901 * of unmatched parts which look like swp_pte, so unuse_pte must
902 * recheck under pte lock. Scanning without pte lock lets it be
903 * preemptible whenever CONFIG_PREEMPT but not CONFIG_HIGHPTE.
904 */
905 pte = pte_offset_map(pmd, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700906 do {
907 /*
908 * swapoff spends a _lot_ of time in this loop!
909 * Test inline before going to call unuse_pte.
910 */
911 if (unlikely(pte_same(*pte, swp_pte))) {
Hugh Dickins044d66c2008-02-07 00:14:04 -0800912 pte_unmap(pte);
913 ret = unuse_pte(vma, pmd, addr, entry, page);
914 if (ret)
915 goto out;
916 pte = pte_offset_map(pmd, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700917 }
918 } while (pte++, addr += PAGE_SIZE, addr != end);
Hugh Dickins044d66c2008-02-07 00:14:04 -0800919 pte_unmap(pte - 1);
920out:
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800921 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700922}
923
924static inline int unuse_pmd_range(struct vm_area_struct *vma, pud_t *pud,
925 unsigned long addr, unsigned long end,
926 swp_entry_t entry, struct page *page)
927{
928 pmd_t *pmd;
929 unsigned long next;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800930 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700931
932 pmd = pmd_offset(pud, addr);
933 do {
934 next = pmd_addr_end(addr, end);
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -0700935 if (pmd_none_or_trans_huge_or_clear_bad(pmd))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700936 continue;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800937 ret = unuse_pte_range(vma, pmd, addr, next, entry, page);
938 if (ret)
939 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700940 } while (pmd++, addr = next, addr != end);
941 return 0;
942}
943
944static inline int unuse_pud_range(struct vm_area_struct *vma, pgd_t *pgd,
945 unsigned long addr, unsigned long end,
946 swp_entry_t entry, struct page *page)
947{
948 pud_t *pud;
949 unsigned long next;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800950 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700951
952 pud = pud_offset(pgd, addr);
953 do {
954 next = pud_addr_end(addr, end);
955 if (pud_none_or_clear_bad(pud))
956 continue;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800957 ret = unuse_pmd_range(vma, pud, addr, next, entry, page);
958 if (ret)
959 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700960 } while (pud++, addr = next, addr != end);
961 return 0;
962}
963
964static int unuse_vma(struct vm_area_struct *vma,
965 swp_entry_t entry, struct page *page)
966{
967 pgd_t *pgd;
968 unsigned long addr, end, next;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800969 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700970
Hugh Dickins3ca7b3c2009-12-14 17:58:57 -0800971 if (page_anon_vma(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700972 addr = page_address_in_vma(page, vma);
973 if (addr == -EFAULT)
974 return 0;
975 else
976 end = addr + PAGE_SIZE;
977 } else {
978 addr = vma->vm_start;
979 end = vma->vm_end;
980 }
981
982 pgd = pgd_offset(vma->vm_mm, addr);
983 do {
984 next = pgd_addr_end(addr, end);
985 if (pgd_none_or_clear_bad(pgd))
986 continue;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800987 ret = unuse_pud_range(vma, pgd, addr, next, entry, page);
988 if (ret)
989 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700990 } while (pgd++, addr = next, addr != end);
991 return 0;
992}
993
994static int unuse_mm(struct mm_struct *mm,
995 swp_entry_t entry, struct page *page)
996{
997 struct vm_area_struct *vma;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800998 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700999
1000 if (!down_read_trylock(&mm->mmap_sem)) {
1001 /*
Fernando Luis Vazquez Cao7d034312008-07-29 22:33:41 -07001002 * Activate page so shrink_inactive_list is unlikely to unmap
1003 * its ptes while lock is dropped, so swapoff can make progress.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001004 */
Hugh Dickinsc475a8a2005-06-21 17:15:12 -07001005 activate_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001006 unlock_page(page);
1007 down_read(&mm->mmap_sem);
1008 lock_page(page);
1009 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001010 for (vma = mm->mmap; vma; vma = vma->vm_next) {
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001011 if (vma->anon_vma && (ret = unuse_vma(vma, entry, page)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001012 break;
1013 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001014 up_read(&mm->mmap_sem);
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001015 return (ret < 0)? ret: 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001016}
1017
1018/*
1019 * Scan swap_map from current position to next entry still in use.
1020 * Recycle to start on reaching the end, returning 0 when empty.
1021 */
Hugh Dickins6eb396d2005-09-03 15:54:35 -07001022static unsigned int find_next_to_unuse(struct swap_info_struct *si,
1023 unsigned int prev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001024{
Hugh Dickins6eb396d2005-09-03 15:54:35 -07001025 unsigned int max = si->max;
1026 unsigned int i = prev;
Hugh Dickins8d69aae2009-12-14 17:58:45 -08001027 unsigned char count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001028
1029 /*
Hugh Dickins5d337b92005-09-03 15:54:41 -07001030 * No need for swap_lock here: we're just looking
Linus Torvalds1da177e2005-04-16 15:20:36 -07001031 * for whether an entry is in use, not modifying it; false
1032 * hits are okay, and sys_swapoff() has already prevented new
Hugh Dickins5d337b92005-09-03 15:54:41 -07001033 * allocations from this area (while holding swap_lock).
Linus Torvalds1da177e2005-04-16 15:20:36 -07001034 */
1035 for (;;) {
1036 if (++i >= max) {
1037 if (!prev) {
1038 i = 0;
1039 break;
1040 }
1041 /*
1042 * No entries in use at top of swap_map,
1043 * loop back to start and recheck there.
1044 */
1045 max = prev + 1;
1046 prev = 0;
1047 i = 1;
1048 }
1049 count = si->swap_map[i];
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07001050 if (count && swap_count(count) != SWAP_MAP_BAD)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001051 break;
1052 }
1053 return i;
1054}
1055
1056/*
1057 * We completely avoid races by reading each swap page in advance,
1058 * and then search for the process using it. All the necessary
1059 * page table adjustments can then be made atomically.
1060 */
1061static int try_to_unuse(unsigned int type)
1062{
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001063 struct swap_info_struct *si = swap_info[type];
Linus Torvalds1da177e2005-04-16 15:20:36 -07001064 struct mm_struct *start_mm;
Hugh Dickins8d69aae2009-12-14 17:58:45 -08001065 unsigned char *swap_map;
1066 unsigned char swcount;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001067 struct page *page;
1068 swp_entry_t entry;
Hugh Dickins6eb396d2005-09-03 15:54:35 -07001069 unsigned int i = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001070 int retval = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001071
1072 /*
1073 * When searching mms for an entry, a good strategy is to
1074 * start at the first mm we freed the previous entry from
1075 * (though actually we don't notice whether we or coincidence
1076 * freed the entry). Initialize this start_mm with a hold.
1077 *
1078 * A simpler strategy would be to start at the last mm we
1079 * freed the previous entry from; but that would take less
1080 * advantage of mmlist ordering, which clusters forked mms
1081 * together, child after parent. If we race with dup_mmap(), we
1082 * prefer to resolve parent before child, lest we miss entries
1083 * duplicated after we scanned child: using last mm would invert
Hugh Dickins570a335b2009-12-14 17:58:46 -08001084 * that.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001085 */
1086 start_mm = &init_mm;
1087 atomic_inc(&init_mm.mm_users);
1088
1089 /*
1090 * Keep on scanning until all entries have gone. Usually,
1091 * one pass through swap_map is enough, but not necessarily:
1092 * there are races when an instance of an entry might be missed.
1093 */
1094 while ((i = find_next_to_unuse(si, i)) != 0) {
1095 if (signal_pending(current)) {
1096 retval = -EINTR;
1097 break;
1098 }
1099
Hugh Dickins886bb7e2009-01-06 14:39:48 -08001100 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001101 * Get a page for the entry, using the existing swap
1102 * cache page if there is one. Otherwise, get a clean
Hugh Dickins886bb7e2009-01-06 14:39:48 -08001103 * page and read the swap into it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001104 */
1105 swap_map = &si->swap_map[i];
1106 entry = swp_entry(type, i);
Hugh Dickins02098fe2008-02-04 22:28:42 -08001107 page = read_swap_cache_async(entry,
1108 GFP_HIGHUSER_MOVABLE, NULL, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001109 if (!page) {
1110 /*
1111 * Either swap_duplicate() failed because entry
1112 * has been freed independently, and will not be
1113 * reused since sys_swapoff() already disabled
1114 * allocation from here, or alloc_page() failed.
1115 */
1116 if (!*swap_map)
1117 continue;
1118 retval = -ENOMEM;
1119 break;
1120 }
1121
1122 /*
1123 * Don't hold on to start_mm if it looks like exiting.
1124 */
1125 if (atomic_read(&start_mm->mm_users) == 1) {
1126 mmput(start_mm);
1127 start_mm = &init_mm;
1128 atomic_inc(&init_mm.mm_users);
1129 }
1130
1131 /*
1132 * Wait for and lock page. When do_swap_page races with
1133 * try_to_unuse, do_swap_page can handle the fault much
1134 * faster than try_to_unuse can locate the entry. This
1135 * apparently redundant "wait_on_page_locked" lets try_to_unuse
1136 * defer to do_swap_page in such a case - in some tests,
1137 * do_swap_page and try_to_unuse repeatedly compete.
1138 */
1139 wait_on_page_locked(page);
1140 wait_on_page_writeback(page);
1141 lock_page(page);
1142 wait_on_page_writeback(page);
1143
1144 /*
1145 * Remove all references to entry.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001146 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001147 swcount = *swap_map;
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001148 if (swap_count(swcount) == SWAP_MAP_SHMEM) {
1149 retval = shmem_unuse(entry, page);
1150 /* page has already been unlocked and released */
1151 if (retval < 0)
1152 break;
1153 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001154 }
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001155 if (swap_count(swcount) && start_mm != &init_mm)
1156 retval = unuse_mm(start_mm, entry, page);
1157
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07001158 if (swap_count(*swap_map)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001159 int set_start_mm = (*swap_map >= swcount);
1160 struct list_head *p = &start_mm->mmlist;
1161 struct mm_struct *new_start_mm = start_mm;
1162 struct mm_struct *prev_mm = start_mm;
1163 struct mm_struct *mm;
1164
1165 atomic_inc(&new_start_mm->mm_users);
1166 atomic_inc(&prev_mm->mm_users);
1167 spin_lock(&mmlist_lock);
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001168 while (swap_count(*swap_map) && !retval &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001169 (p = p->next) != &start_mm->mmlist) {
1170 mm = list_entry(p, struct mm_struct, mmlist);
Hugh Dickins70af7c52006-06-23 02:03:44 -07001171 if (!atomic_inc_not_zero(&mm->mm_users))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001172 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001173 spin_unlock(&mmlist_lock);
1174 mmput(prev_mm);
1175 prev_mm = mm;
1176
1177 cond_resched();
1178
1179 swcount = *swap_map;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07001180 if (!swap_count(swcount)) /* any usage ? */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001181 ;
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001182 else if (mm == &init_mm)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001183 set_start_mm = 1;
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001184 else
Linus Torvalds1da177e2005-04-16 15:20:36 -07001185 retval = unuse_mm(mm, entry, page);
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07001186
Bo Liu32c5fc12009-11-02 16:50:33 +00001187 if (set_start_mm && *swap_map < swcount) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001188 mmput(new_start_mm);
1189 atomic_inc(&mm->mm_users);
1190 new_start_mm = mm;
1191 set_start_mm = 0;
1192 }
1193 spin_lock(&mmlist_lock);
1194 }
1195 spin_unlock(&mmlist_lock);
1196 mmput(prev_mm);
1197 mmput(start_mm);
1198 start_mm = new_start_mm;
1199 }
1200 if (retval) {
1201 unlock_page(page);
1202 page_cache_release(page);
1203 break;
1204 }
1205
1206 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001207 * If a reference remains (rare), we would like to leave
1208 * the page in the swap cache; but try_to_unmap could
1209 * then re-duplicate the entry once we drop page lock,
1210 * so we might loop indefinitely; also, that page could
1211 * not be swapped out to other storage meanwhile. So:
1212 * delete from cache even if there's another reference,
1213 * after ensuring that the data has been saved to disk -
1214 * since if the reference remains (rarer), it will be
1215 * read from disk into another page. Splitting into two
1216 * pages would be incorrect if swap supported "shared
1217 * private" pages, but they are handled by tmpfs files.
Hugh Dickins5ad64682009-12-14 17:59:24 -08001218 *
1219 * Given how unuse_vma() targets one particular offset
1220 * in an anon_vma, once the anon_vma has been determined,
1221 * this splitting happens to be just what is needed to
1222 * handle where KSM pages have been swapped out: re-reading
1223 * is unnecessarily slow, but we can fix that later on.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001224 */
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07001225 if (swap_count(*swap_map) &&
1226 PageDirty(page) && PageSwapCache(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001227 struct writeback_control wbc = {
1228 .sync_mode = WB_SYNC_NONE,
1229 };
1230
1231 swap_writepage(page, &wbc);
1232 lock_page(page);
1233 wait_on_page_writeback(page);
1234 }
Hugh Dickins68bdc8d2009-01-06 14:39:37 -08001235
1236 /*
1237 * It is conceivable that a racing task removed this page from
1238 * swap cache just before we acquired the page lock at the top,
1239 * or while we dropped it in unuse_mm(). The page might even
1240 * be back in swap cache on another swap area: that we must not
1241 * delete, since it may not have been written out to swap yet.
1242 */
1243 if (PageSwapCache(page) &&
1244 likely(page_private(page) == entry.val))
Hugh Dickins2e0e26c2008-02-04 22:28:53 -08001245 delete_from_swap_cache(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001246
1247 /*
1248 * So we could skip searching mms once swap count went
1249 * to 1, we did not mark any present ptes as dirty: must
Anderson Briglia2706a1b2007-07-15 23:38:09 -07001250 * mark page dirty so shrink_page_list will preserve it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001251 */
1252 SetPageDirty(page);
1253 unlock_page(page);
1254 page_cache_release(page);
1255
1256 /*
1257 * Make sure that we aren't completely killing
1258 * interactive performance.
1259 */
1260 cond_resched();
1261 }
1262
1263 mmput(start_mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001264 return retval;
1265}
1266
1267/*
Hugh Dickins5d337b92005-09-03 15:54:41 -07001268 * After a successful try_to_unuse, if no swap is now in use, we know
1269 * we can empty the mmlist. swap_lock must be held on entry and exit.
1270 * Note that mmlist_lock nests inside swap_lock, and an mm must be
Linus Torvalds1da177e2005-04-16 15:20:36 -07001271 * added to the mmlist just after page_duplicate - before would be racy.
1272 */
1273static void drain_mmlist(void)
1274{
1275 struct list_head *p, *next;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001276 unsigned int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001277
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001278 for (type = 0; type < nr_swapfiles; type++)
1279 if (swap_info[type]->inuse_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001280 return;
1281 spin_lock(&mmlist_lock);
1282 list_for_each_safe(p, next, &init_mm.mmlist)
1283 list_del_init(p);
1284 spin_unlock(&mmlist_lock);
1285}
1286
1287/*
1288 * Use this swapdev's extent info to locate the (PAGE_SIZE) block which
Lee Schermerhornd4906e12009-12-14 17:58:49 -08001289 * corresponds to page offset for the specified swap entry.
1290 * Note that the type of this function is sector_t, but it returns page offset
1291 * into the bdev, not sector offset.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001292 */
Lee Schermerhornd4906e12009-12-14 17:58:49 -08001293static sector_t map_swap_entry(swp_entry_t entry, struct block_device **bdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001294{
Hugh Dickinsf29ad6a2009-12-14 17:58:40 -08001295 struct swap_info_struct *sis;
1296 struct swap_extent *start_se;
1297 struct swap_extent *se;
1298 pgoff_t offset;
1299
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001300 sis = swap_info[swp_type(entry)];
Hugh Dickinsf29ad6a2009-12-14 17:58:40 -08001301 *bdev = sis->bdev;
1302
1303 offset = swp_offset(entry);
1304 start_se = sis->curr_swap_extent;
1305 se = start_se;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001306
1307 for ( ; ; ) {
1308 struct list_head *lh;
1309
1310 if (se->start_page <= offset &&
1311 offset < (se->start_page + se->nr_pages)) {
1312 return se->start_block + (offset - se->start_page);
1313 }
Hugh Dickins11d31882005-09-03 15:54:34 -07001314 lh = se->list.next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001315 se = list_entry(lh, struct swap_extent, list);
1316 sis->curr_swap_extent = se;
1317 BUG_ON(se == start_se); /* It *must* be present */
1318 }
1319}
1320
1321/*
Lee Schermerhornd4906e12009-12-14 17:58:49 -08001322 * Returns the page offset into bdev for the specified page's swap entry.
1323 */
1324sector_t map_swap_page(struct page *page, struct block_device **bdev)
1325{
1326 swp_entry_t entry;
1327 entry.val = page_private(page);
1328 return map_swap_entry(entry, bdev);
1329}
1330
1331/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001332 * Free all of a swapdev's extent information
1333 */
1334static void destroy_swap_extents(struct swap_info_struct *sis)
1335{
Hugh Dickins9625a5f2009-12-14 17:58:42 -08001336 while (!list_empty(&sis->first_swap_extent.list)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001337 struct swap_extent *se;
1338
Hugh Dickins9625a5f2009-12-14 17:58:42 -08001339 se = list_entry(sis->first_swap_extent.list.next,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001340 struct swap_extent, list);
1341 list_del(&se->list);
1342 kfree(se);
1343 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001344}
1345
1346/*
1347 * Add a block range (and the corresponding page range) into this swapdev's
Hugh Dickins11d31882005-09-03 15:54:34 -07001348 * extent list. The extent list is kept sorted in page order.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001349 *
Hugh Dickins11d31882005-09-03 15:54:34 -07001350 * This function rather assumes that it is called in ascending page order.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001351 */
1352static int
1353add_swap_extent(struct swap_info_struct *sis, unsigned long start_page,
1354 unsigned long nr_pages, sector_t start_block)
1355{
1356 struct swap_extent *se;
1357 struct swap_extent *new_se;
1358 struct list_head *lh;
1359
Hugh Dickins9625a5f2009-12-14 17:58:42 -08001360 if (start_page == 0) {
1361 se = &sis->first_swap_extent;
1362 sis->curr_swap_extent = se;
1363 se->start_page = 0;
1364 se->nr_pages = nr_pages;
1365 se->start_block = start_block;
1366 return 1;
1367 } else {
1368 lh = sis->first_swap_extent.list.prev; /* Highest extent */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001369 se = list_entry(lh, struct swap_extent, list);
Hugh Dickins11d31882005-09-03 15:54:34 -07001370 BUG_ON(se->start_page + se->nr_pages != start_page);
1371 if (se->start_block + se->nr_pages == start_block) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001372 /* Merge it */
1373 se->nr_pages += nr_pages;
1374 return 0;
1375 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001376 }
1377
1378 /*
1379 * No merge. Insert a new extent, preserving ordering.
1380 */
1381 new_se = kmalloc(sizeof(*se), GFP_KERNEL);
1382 if (new_se == NULL)
1383 return -ENOMEM;
1384 new_se->start_page = start_page;
1385 new_se->nr_pages = nr_pages;
1386 new_se->start_block = start_block;
1387
Hugh Dickins9625a5f2009-12-14 17:58:42 -08001388 list_add_tail(&new_se->list, &sis->first_swap_extent.list);
Hugh Dickins53092a72005-09-03 15:54:34 -07001389 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001390}
1391
1392/*
1393 * A `swap extent' is a simple thing which maps a contiguous range of pages
1394 * onto a contiguous range of disk blocks. An ordered list of swap extents
1395 * is built at swapon time and is then used at swap_writepage/swap_readpage
1396 * time for locating where on disk a page belongs.
1397 *
1398 * If the swapfile is an S_ISBLK block device, a single extent is installed.
1399 * This is done so that the main operating code can treat S_ISBLK and S_ISREG
1400 * swap files identically.
1401 *
1402 * Whether the swapdev is an S_ISREG file or an S_ISBLK blockdev, the swap
1403 * extent list operates in PAGE_SIZE disk blocks. Both S_ISREG and S_ISBLK
1404 * swapfiles are handled *identically* after swapon time.
1405 *
1406 * For S_ISREG swapfiles, setup_swap_extents() will walk all the file's blocks
1407 * and will parse them into an ordered extent list, in PAGE_SIZE chunks. If
1408 * some stray blocks are found which do not fall within the PAGE_SIZE alignment
1409 * requirements, they are simply tossed out - we will never use those blocks
1410 * for swapping.
1411 *
Hugh Dickinsb0d9bcd2005-09-03 15:54:31 -07001412 * For S_ISREG swapfiles we set S_SWAPFILE across the life of the swapon. This
Linus Torvalds1da177e2005-04-16 15:20:36 -07001413 * prevents root from shooting her foot off by ftruncating an in-use swapfile,
1414 * which will scribble on the fs.
1415 *
1416 * The amount of disk space which a single swap extent represents varies.
1417 * Typically it is in the 1-4 megabyte range. So we can have hundreds of
1418 * extents in the list. To avoid much list walking, we cache the previous
1419 * search location in `curr_swap_extent', and start new searches from there.
1420 * This is extremely effective. The average number of iterations in
1421 * map_swap_page() has been measured at about 0.3 per page. - akpm.
1422 */
Hugh Dickins53092a72005-09-03 15:54:34 -07001423static int setup_swap_extents(struct swap_info_struct *sis, sector_t *span)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001424{
1425 struct inode *inode;
1426 unsigned blocks_per_page;
1427 unsigned long page_no;
1428 unsigned blkbits;
1429 sector_t probe_block;
1430 sector_t last_block;
Hugh Dickins53092a72005-09-03 15:54:34 -07001431 sector_t lowest_block = -1;
1432 sector_t highest_block = 0;
1433 int nr_extents = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001434 int ret;
1435
1436 inode = sis->swap_file->f_mapping->host;
1437 if (S_ISBLK(inode->i_mode)) {
1438 ret = add_swap_extent(sis, 0, sis->max, 0);
Hugh Dickins53092a72005-09-03 15:54:34 -07001439 *span = sis->pages;
Hugh Dickins9625a5f2009-12-14 17:58:42 -08001440 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001441 }
1442
1443 blkbits = inode->i_blkbits;
1444 blocks_per_page = PAGE_SIZE >> blkbits;
1445
1446 /*
1447 * Map all the blocks into the extent list. This code doesn't try
1448 * to be very smart.
1449 */
1450 probe_block = 0;
1451 page_no = 0;
1452 last_block = i_size_read(inode) >> blkbits;
1453 while ((probe_block + blocks_per_page) <= last_block &&
1454 page_no < sis->max) {
1455 unsigned block_in_page;
1456 sector_t first_block;
1457
1458 first_block = bmap(inode, probe_block);
1459 if (first_block == 0)
1460 goto bad_bmap;
1461
1462 /*
1463 * It must be PAGE_SIZE aligned on-disk
1464 */
1465 if (first_block & (blocks_per_page - 1)) {
1466 probe_block++;
1467 goto reprobe;
1468 }
1469
1470 for (block_in_page = 1; block_in_page < blocks_per_page;
1471 block_in_page++) {
1472 sector_t block;
1473
1474 block = bmap(inode, probe_block + block_in_page);
1475 if (block == 0)
1476 goto bad_bmap;
1477 if (block != first_block + block_in_page) {
1478 /* Discontiguity */
1479 probe_block++;
1480 goto reprobe;
1481 }
1482 }
1483
Hugh Dickins53092a72005-09-03 15:54:34 -07001484 first_block >>= (PAGE_SHIFT - blkbits);
1485 if (page_no) { /* exclude the header page */
1486 if (first_block < lowest_block)
1487 lowest_block = first_block;
1488 if (first_block > highest_block)
1489 highest_block = first_block;
1490 }
1491
Linus Torvalds1da177e2005-04-16 15:20:36 -07001492 /*
1493 * We found a PAGE_SIZE-length, PAGE_SIZE-aligned run of blocks
1494 */
Hugh Dickins53092a72005-09-03 15:54:34 -07001495 ret = add_swap_extent(sis, page_no, 1, first_block);
1496 if (ret < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001497 goto out;
Hugh Dickins53092a72005-09-03 15:54:34 -07001498 nr_extents += ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001499 page_no++;
1500 probe_block += blocks_per_page;
1501reprobe:
1502 continue;
1503 }
Hugh Dickins53092a72005-09-03 15:54:34 -07001504 ret = nr_extents;
1505 *span = 1 + highest_block - lowest_block;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001506 if (page_no == 0)
Hugh Dickinse2244ec2005-09-03 15:54:32 -07001507 page_no = 1; /* force Empty message */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001508 sis->max = page_no;
Hugh Dickinse2244ec2005-09-03 15:54:32 -07001509 sis->pages = page_no - 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001510 sis->highest_bit = page_no - 1;
Hugh Dickins9625a5f2009-12-14 17:58:42 -08001511out:
1512 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001513bad_bmap:
1514 printk(KERN_ERR "swapon: swapfile has holes\n");
1515 ret = -EINVAL;
Hugh Dickins9625a5f2009-12-14 17:58:42 -08001516 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001517}
1518
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07001519static void enable_swap_info(struct swap_info_struct *p, int prio,
1520 unsigned char *swap_map)
1521{
1522 int i, prev;
1523
1524 spin_lock(&swap_lock);
1525 if (prio >= 0)
1526 p->prio = prio;
1527 else
1528 p->prio = --least_priority;
1529 p->swap_map = swap_map;
1530 p->flags |= SWP_WRITEOK;
1531 nr_swap_pages += p->pages;
1532 total_swap_pages += p->pages;
1533
1534 /* insert swap space into swap_list: */
1535 prev = -1;
1536 for (i = swap_list.head; i >= 0; i = swap_info[i]->next) {
1537 if (p->prio >= swap_info[i]->prio)
1538 break;
1539 prev = i;
1540 }
1541 p->next = i;
1542 if (prev < 0)
1543 swap_list.head = swap_list.next = p->type;
1544 else
1545 swap_info[prev]->next = p->type;
1546 spin_unlock(&swap_lock);
1547}
1548
Heiko Carstensc4ea37c2009-01-14 14:14:28 +01001549SYSCALL_DEFINE1(swapoff, const char __user *, specialfile)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001550{
Hugh Dickins73c34b62009-12-14 17:58:43 -08001551 struct swap_info_struct *p = NULL;
Hugh Dickins8d69aae2009-12-14 17:58:45 -08001552 unsigned char *swap_map;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001553 struct file *swap_file, *victim;
1554 struct address_space *mapping;
1555 struct inode *inode;
Hugh Dickins73c34b62009-12-14 17:58:43 -08001556 char *pathname;
David Rientjes72788c32011-05-24 17:11:40 -07001557 int oom_score_adj;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001558 int i, type, prev;
1559 int err;
Hugh Dickins886bb7e2009-01-06 14:39:48 -08001560
Linus Torvalds1da177e2005-04-16 15:20:36 -07001561 if (!capable(CAP_SYS_ADMIN))
1562 return -EPERM;
1563
Al Viro191c5422012-02-13 03:58:52 +00001564 BUG_ON(!current->mm);
1565
Linus Torvalds1da177e2005-04-16 15:20:36 -07001566 pathname = getname(specialfile);
1567 err = PTR_ERR(pathname);
1568 if (IS_ERR(pathname))
1569 goto out;
1570
1571 victim = filp_open(pathname, O_RDWR|O_LARGEFILE, 0);
1572 putname(pathname);
1573 err = PTR_ERR(victim);
1574 if (IS_ERR(victim))
1575 goto out;
1576
1577 mapping = victim->f_mapping;
1578 prev = -1;
Hugh Dickins5d337b92005-09-03 15:54:41 -07001579 spin_lock(&swap_lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001580 for (type = swap_list.head; type >= 0; type = swap_info[type]->next) {
1581 p = swap_info[type];
Hugh Dickins22c6f8f2009-01-06 14:39:48 -08001582 if (p->flags & SWP_WRITEOK) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001583 if (p->swap_file->f_mapping == mapping)
1584 break;
1585 }
1586 prev = type;
1587 }
1588 if (type < 0) {
1589 err = -EINVAL;
Hugh Dickins5d337b92005-09-03 15:54:41 -07001590 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001591 goto out_dput;
1592 }
Al Viro191c5422012-02-13 03:58:52 +00001593 if (!security_vm_enough_memory_mm(current->mm, p->pages))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001594 vm_unacct_memory(p->pages);
1595 else {
1596 err = -ENOMEM;
Hugh Dickins5d337b92005-09-03 15:54:41 -07001597 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001598 goto out_dput;
1599 }
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001600 if (prev < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001601 swap_list.head = p->next;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001602 else
1603 swap_info[prev]->next = p->next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001604 if (type == swap_list.next) {
1605 /* just pick something that's safe... */
1606 swap_list.next = swap_list.head;
1607 }
Hugh Dickins78ecba02008-07-23 21:28:23 -07001608 if (p->prio < 0) {
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001609 for (i = p->next; i >= 0; i = swap_info[i]->next)
1610 swap_info[i]->prio = p->prio--;
Hugh Dickins78ecba02008-07-23 21:28:23 -07001611 least_priority++;
1612 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001613 nr_swap_pages -= p->pages;
1614 total_swap_pages -= p->pages;
1615 p->flags &= ~SWP_WRITEOK;
Hugh Dickins5d337b92005-09-03 15:54:41 -07001616 spin_unlock(&swap_lock);
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -07001617
David Rientjes72788c32011-05-24 17:11:40 -07001618 oom_score_adj = test_set_oom_score_adj(OOM_SCORE_ADJ_MAX);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001619 err = try_to_unuse(type);
David Rientjes43362a42011-10-31 17:07:18 -07001620 compare_swap_oom_score_adj(OOM_SCORE_ADJ_MAX, oom_score_adj);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001621
Linus Torvalds1da177e2005-04-16 15:20:36 -07001622 if (err) {
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07001623 /*
1624 * reading p->prio and p->swap_map outside the lock is
1625 * safe here because only sys_swapon and sys_swapoff
1626 * change them, and there can be no other sys_swapon or
1627 * sys_swapoff for this swap_info_struct at this point.
1628 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001629 /* re-insert swap space back into swap_list */
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07001630 enable_swap_info(p, p->prio, p->swap_map);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001631 goto out_dput;
1632 }
Hugh Dickins52b7efdb2005-09-03 15:54:39 -07001633
Hugh Dickins4cd3bb12005-09-03 15:54:33 -07001634 destroy_swap_extents(p);
Hugh Dickins570a335b2009-12-14 17:58:46 -08001635 if (p->flags & SWP_CONTINUED)
1636 free_swap_count_continuations(p);
1637
Ingo Molnarfc0abb12006-01-18 17:42:33 -08001638 mutex_lock(&swapon_mutex);
Hugh Dickins5d337b92005-09-03 15:54:41 -07001639 spin_lock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001640 drain_mmlist();
Hugh Dickins5d337b92005-09-03 15:54:41 -07001641
1642 /* wait for anyone still in scan_swap_map */
1643 p->highest_bit = 0; /* cuts scans short */
1644 while (p->flags >= SWP_SCANNING) {
1645 spin_unlock(&swap_lock);
Nishanth Aravamudan13e4b572005-09-10 00:27:25 -07001646 schedule_timeout_uninterruptible(1);
Hugh Dickins5d337b92005-09-03 15:54:41 -07001647 spin_lock(&swap_lock);
1648 }
1649
Linus Torvalds1da177e2005-04-16 15:20:36 -07001650 swap_file = p->swap_file;
1651 p->swap_file = NULL;
1652 p->max = 0;
1653 swap_map = p->swap_map;
1654 p->swap_map = NULL;
1655 p->flags = 0;
Hugh Dickins5d337b92005-09-03 15:54:41 -07001656 spin_unlock(&swap_lock);
Ingo Molnarfc0abb12006-01-18 17:42:33 -08001657 mutex_unlock(&swapon_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001658 vfree(swap_map);
KAMEZAWA Hiroyuki27a7faa2009-01-07 18:07:58 -08001659 /* Destroy swap account informatin */
1660 swap_cgroup_swapoff(type);
1661
Linus Torvalds1da177e2005-04-16 15:20:36 -07001662 inode = mapping->host;
1663 if (S_ISBLK(inode->i_mode)) {
1664 struct block_device *bdev = I_BDEV(inode);
1665 set_blocksize(bdev, p->old_block_size);
Tejun Heoe525fd82010-11-13 11:55:17 +01001666 blkdev_put(bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001667 } else {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001668 mutex_lock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001669 inode->i_flags &= ~S_SWAPFILE;
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001670 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001671 }
1672 filp_close(swap_file, NULL);
1673 err = 0;
Kay Sievers66d7dd52010-10-26 14:22:06 -07001674 atomic_inc(&proc_poll_event);
1675 wake_up_interruptible(&proc_poll_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001676
1677out_dput:
1678 filp_close(victim, NULL);
1679out:
1680 return err;
1681}
1682
1683#ifdef CONFIG_PROC_FS
Kay Sievers66d7dd52010-10-26 14:22:06 -07001684static unsigned swaps_poll(struct file *file, poll_table *wait)
1685{
Kay Sieversf1514632011-07-12 20:48:39 +02001686 struct seq_file *seq = file->private_data;
Kay Sievers66d7dd52010-10-26 14:22:06 -07001687
1688 poll_wait(file, &proc_poll_wait, wait);
1689
Kay Sieversf1514632011-07-12 20:48:39 +02001690 if (seq->poll_event != atomic_read(&proc_poll_event)) {
1691 seq->poll_event = atomic_read(&proc_poll_event);
Kay Sievers66d7dd52010-10-26 14:22:06 -07001692 return POLLIN | POLLRDNORM | POLLERR | POLLPRI;
1693 }
1694
1695 return POLLIN | POLLRDNORM;
1696}
1697
Linus Torvalds1da177e2005-04-16 15:20:36 -07001698/* iterator */
1699static void *swap_start(struct seq_file *swap, loff_t *pos)
1700{
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001701 struct swap_info_struct *si;
1702 int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001703 loff_t l = *pos;
1704
Ingo Molnarfc0abb12006-01-18 17:42:33 -08001705 mutex_lock(&swapon_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001706
Suleiman Souhlal881e4aa2006-12-06 20:32:28 -08001707 if (!l)
1708 return SEQ_START_TOKEN;
1709
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001710 for (type = 0; type < nr_swapfiles; type++) {
1711 smp_rmb(); /* read nr_swapfiles before swap_info[type] */
1712 si = swap_info[type];
1713 if (!(si->flags & SWP_USED) || !si->swap_map)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001714 continue;
Suleiman Souhlal881e4aa2006-12-06 20:32:28 -08001715 if (!--l)
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001716 return si;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001717 }
1718
1719 return NULL;
1720}
1721
1722static void *swap_next(struct seq_file *swap, void *v, loff_t *pos)
1723{
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001724 struct swap_info_struct *si = v;
1725 int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001726
Suleiman Souhlal881e4aa2006-12-06 20:32:28 -08001727 if (v == SEQ_START_TOKEN)
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001728 type = 0;
1729 else
1730 type = si->type + 1;
Suleiman Souhlal881e4aa2006-12-06 20:32:28 -08001731
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001732 for (; type < nr_swapfiles; type++) {
1733 smp_rmb(); /* read nr_swapfiles before swap_info[type] */
1734 si = swap_info[type];
1735 if (!(si->flags & SWP_USED) || !si->swap_map)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001736 continue;
1737 ++*pos;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001738 return si;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001739 }
1740
1741 return NULL;
1742}
1743
1744static void swap_stop(struct seq_file *swap, void *v)
1745{
Ingo Molnarfc0abb12006-01-18 17:42:33 -08001746 mutex_unlock(&swapon_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001747}
1748
1749static int swap_show(struct seq_file *swap, void *v)
1750{
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001751 struct swap_info_struct *si = v;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001752 struct file *file;
1753 int len;
1754
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001755 if (si == SEQ_START_TOKEN) {
Suleiman Souhlal881e4aa2006-12-06 20:32:28 -08001756 seq_puts(swap,"Filename\t\t\t\tType\t\tSize\tUsed\tPriority\n");
1757 return 0;
1758 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001759
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001760 file = si->swap_file;
Jan Blunckc32c2f62008-02-14 19:38:43 -08001761 len = seq_path(swap, &file->f_path, " \t\n\\");
Hugh Dickins6eb396d2005-09-03 15:54:35 -07001762 seq_printf(swap, "%*s%s\t%u\t%u\t%d\n",
Hugh Dickins886bb7e2009-01-06 14:39:48 -08001763 len < 40 ? 40 - len : 1, " ",
1764 S_ISBLK(file->f_path.dentry->d_inode->i_mode) ?
Linus Torvalds1da177e2005-04-16 15:20:36 -07001765 "partition" : "file\t",
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001766 si->pages << (PAGE_SHIFT - 10),
1767 si->inuse_pages << (PAGE_SHIFT - 10),
1768 si->prio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001769 return 0;
1770}
1771
Helge Deller15ad7cd2006-12-06 20:40:36 -08001772static const struct seq_operations swaps_op = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001773 .start = swap_start,
1774 .next = swap_next,
1775 .stop = swap_stop,
1776 .show = swap_show
1777};
1778
1779static int swaps_open(struct inode *inode, struct file *file)
1780{
Kay Sieversf1514632011-07-12 20:48:39 +02001781 struct seq_file *seq;
Kay Sievers66d7dd52010-10-26 14:22:06 -07001782 int ret;
1783
Kay Sievers66d7dd52010-10-26 14:22:06 -07001784 ret = seq_open(file, &swaps_op);
Kay Sieversf1514632011-07-12 20:48:39 +02001785 if (ret)
Kay Sievers66d7dd52010-10-26 14:22:06 -07001786 return ret;
Kay Sievers66d7dd52010-10-26 14:22:06 -07001787
Kay Sieversf1514632011-07-12 20:48:39 +02001788 seq = file->private_data;
1789 seq->poll_event = atomic_read(&proc_poll_event);
1790 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001791}
1792
Helge Deller15ad7cd2006-12-06 20:40:36 -08001793static const struct file_operations proc_swaps_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001794 .open = swaps_open,
1795 .read = seq_read,
1796 .llseek = seq_lseek,
1797 .release = seq_release,
Kay Sievers66d7dd52010-10-26 14:22:06 -07001798 .poll = swaps_poll,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001799};
1800
1801static int __init procswaps_init(void)
1802{
Denis V. Lunev3d71f862008-04-29 01:02:13 -07001803 proc_create("swaps", 0, NULL, &proc_swaps_operations);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001804 return 0;
1805}
1806__initcall(procswaps_init);
1807#endif /* CONFIG_PROC_FS */
1808
Jan Beulich17963162008-12-16 11:35:24 +00001809#ifdef MAX_SWAPFILES_CHECK
1810static int __init max_swapfiles_check(void)
1811{
1812 MAX_SWAPFILES_CHECK();
1813 return 0;
1814}
1815late_initcall(max_swapfiles_check);
1816#endif
1817
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07001818static struct swap_info_struct *alloc_swap_info(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001819{
Hugh Dickins73c34b62009-12-14 17:58:43 -08001820 struct swap_info_struct *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001821 unsigned int type;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001822
1823 p = kzalloc(sizeof(*p), GFP_KERNEL);
1824 if (!p)
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07001825 return ERR_PTR(-ENOMEM);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001826
Hugh Dickins5d337b92005-09-03 15:54:41 -07001827 spin_lock(&swap_lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001828 for (type = 0; type < nr_swapfiles; type++) {
1829 if (!(swap_info[type]->flags & SWP_USED))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001830 break;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001831 }
Christoph Lameter06972122006-06-23 02:03:35 -07001832 if (type >= MAX_SWAPFILES) {
Hugh Dickins5d337b92005-09-03 15:54:41 -07001833 spin_unlock(&swap_lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001834 kfree(p);
Cesar Eduardo Barros730c0582011-03-22 16:33:19 -07001835 return ERR_PTR(-EPERM);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001836 }
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001837 if (type >= nr_swapfiles) {
1838 p->type = type;
1839 swap_info[type] = p;
1840 /*
1841 * Write swap_info[type] before nr_swapfiles, in case a
1842 * racing procfs swap_start() or swap_next() is reading them.
1843 * (We never shrink nr_swapfiles, we never free this entry.)
1844 */
1845 smp_wmb();
1846 nr_swapfiles++;
1847 } else {
1848 kfree(p);
1849 p = swap_info[type];
1850 /*
1851 * Do not memset this entry: a racing procfs swap_next()
1852 * would be relying on p->type to remain valid.
1853 */
1854 }
Hugh Dickins9625a5f2009-12-14 17:58:42 -08001855 INIT_LIST_HEAD(&p->first_swap_extent.list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001856 p->flags = SWP_USED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001857 p->next = -1;
Hugh Dickins5d337b92005-09-03 15:54:41 -07001858 spin_unlock(&swap_lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001859
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07001860 return p;
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07001861}
1862
Cesar Eduardo Barros4d0e1e12011-03-22 16:33:26 -07001863static int claim_swapfile(struct swap_info_struct *p, struct inode *inode)
1864{
1865 int error;
1866
1867 if (S_ISBLK(inode->i_mode)) {
1868 p->bdev = bdgrab(I_BDEV(inode));
1869 error = blkdev_get(p->bdev,
1870 FMODE_READ | FMODE_WRITE | FMODE_EXCL,
1871 sys_swapon);
1872 if (error < 0) {
1873 p->bdev = NULL;
Cesar Eduardo Barros87ade722011-03-22 16:33:27 -07001874 return -EINVAL;
Cesar Eduardo Barros4d0e1e12011-03-22 16:33:26 -07001875 }
1876 p->old_block_size = block_size(p->bdev);
1877 error = set_blocksize(p->bdev, PAGE_SIZE);
1878 if (error < 0)
Cesar Eduardo Barros87ade722011-03-22 16:33:27 -07001879 return error;
Cesar Eduardo Barros4d0e1e12011-03-22 16:33:26 -07001880 p->flags |= SWP_BLKDEV;
1881 } else if (S_ISREG(inode->i_mode)) {
1882 p->bdev = inode->i_sb->s_bdev;
1883 mutex_lock(&inode->i_mutex);
Cesar Eduardo Barros87ade722011-03-22 16:33:27 -07001884 if (IS_SWAPFILE(inode))
1885 return -EBUSY;
1886 } else
1887 return -EINVAL;
Cesar Eduardo Barros4d0e1e12011-03-22 16:33:26 -07001888
1889 return 0;
Cesar Eduardo Barros4d0e1e12011-03-22 16:33:26 -07001890}
1891
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07001892static unsigned long read_swap_header(struct swap_info_struct *p,
1893 union swap_header *swap_header,
1894 struct inode *inode)
1895{
1896 int i;
1897 unsigned long maxpages;
1898 unsigned long swapfilepages;
1899
1900 if (memcmp("SWAPSPACE2", swap_header->magic.magic, 10)) {
1901 printk(KERN_ERR "Unable to find swap-space signature\n");
Cesar Eduardo Barros38719022011-03-22 16:33:30 -07001902 return 0;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07001903 }
1904
1905 /* swap partition endianess hack... */
1906 if (swab32(swap_header->info.version) == 1) {
1907 swab32s(&swap_header->info.version);
1908 swab32s(&swap_header->info.last_page);
1909 swab32s(&swap_header->info.nr_badpages);
1910 for (i = 0; i < swap_header->info.nr_badpages; i++)
1911 swab32s(&swap_header->info.badpages[i]);
1912 }
1913 /* Check the swap header's sub-version */
1914 if (swap_header->info.version != 1) {
1915 printk(KERN_WARNING
1916 "Unable to handle swap header version %d\n",
1917 swap_header->info.version);
Cesar Eduardo Barros38719022011-03-22 16:33:30 -07001918 return 0;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07001919 }
1920
1921 p->lowest_bit = 1;
1922 p->cluster_next = 1;
1923 p->cluster_nr = 0;
1924
1925 /*
1926 * Find out how many pages are allowed for a single swap
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001927 * device. There are three limiting factors: 1) the number
1928 * of bits for the swap offset in the swp_entry_t type, and
1929 * 2) the number of bits in the swap pte as defined by the
1930 * the different architectures, and 3) the number of free bits
1931 * in an exceptional radix_tree entry. In order to find the
1932 * largest possible bit mask, a swap entry with swap type 0
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07001933 * and swap offset ~0UL is created, encoded to a swap pte,
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001934 * decoded to a swp_entry_t again, and finally the swap
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07001935 * offset is extracted. This will mask all the bits from
1936 * the initial ~0UL mask that can't be encoded in either
1937 * the swp_entry_t or the architecture definition of a
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001938 * swap pte. Then the same is done for a radix_tree entry.
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07001939 */
1940 maxpages = swp_offset(pte_to_swp_entry(
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001941 swp_entry_to_pte(swp_entry(0, ~0UL))));
1942 maxpages = swp_offset(radix_to_swp_entry(
1943 swp_to_radix_entry(swp_entry(0, maxpages)))) + 1;
1944
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07001945 if (maxpages > swap_header->info.last_page) {
1946 maxpages = swap_header->info.last_page + 1;
1947 /* p->max is an unsigned int: don't overflow it */
1948 if ((unsigned int)maxpages == 0)
1949 maxpages = UINT_MAX;
1950 }
1951 p->highest_bit = maxpages - 1;
1952
1953 if (!maxpages)
Cesar Eduardo Barros38719022011-03-22 16:33:30 -07001954 return 0;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07001955 swapfilepages = i_size_read(inode) >> PAGE_SHIFT;
1956 if (swapfilepages && maxpages > swapfilepages) {
1957 printk(KERN_WARNING
1958 "Swap area shorter than signature indicates\n");
Cesar Eduardo Barros38719022011-03-22 16:33:30 -07001959 return 0;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07001960 }
1961 if (swap_header->info.nr_badpages && S_ISREG(inode->i_mode))
Cesar Eduardo Barros38719022011-03-22 16:33:30 -07001962 return 0;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07001963 if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES)
Cesar Eduardo Barros38719022011-03-22 16:33:30 -07001964 return 0;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07001965
1966 return maxpages;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07001967}
1968
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07001969static int setup_swap_map_and_extents(struct swap_info_struct *p,
1970 union swap_header *swap_header,
1971 unsigned char *swap_map,
1972 unsigned long maxpages,
1973 sector_t *span)
1974{
1975 int i;
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07001976 unsigned int nr_good_pages;
1977 int nr_extents;
1978
1979 nr_good_pages = maxpages - 1; /* omit header page */
1980
1981 for (i = 0; i < swap_header->info.nr_badpages; i++) {
1982 unsigned int page_nr = swap_header->info.badpages[i];
Cesar Eduardo Barrosbdb8e3f2011-03-22 16:33:33 -07001983 if (page_nr == 0 || page_nr > swap_header->info.last_page)
1984 return -EINVAL;
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07001985 if (page_nr < maxpages) {
1986 swap_map[page_nr] = SWAP_MAP_BAD;
1987 nr_good_pages--;
1988 }
1989 }
1990
1991 if (nr_good_pages) {
1992 swap_map[0] = SWAP_MAP_BAD;
1993 p->max = maxpages;
1994 p->pages = nr_good_pages;
1995 nr_extents = setup_swap_extents(p, span);
Cesar Eduardo Barrosbdb8e3f2011-03-22 16:33:33 -07001996 if (nr_extents < 0)
1997 return nr_extents;
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07001998 nr_good_pages = p->pages;
1999 }
2000 if (!nr_good_pages) {
2001 printk(KERN_WARNING "Empty swap-file\n");
Cesar Eduardo Barrosbdb8e3f2011-03-22 16:33:33 -07002002 return -EINVAL;
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002003 }
2004
2005 return nr_extents;
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002006}
2007
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002008SYSCALL_DEFINE2(swapon, const char __user *, specialfile, int, swap_flags)
2009{
2010 struct swap_info_struct *p;
Cesar Eduardo Barros28b36bd2011-03-22 16:33:20 -07002011 char *name;
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002012 struct file *swap_file = NULL;
2013 struct address_space *mapping;
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07002014 int i;
2015 int prio;
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002016 int error;
2017 union swap_header *swap_header;
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002018 int nr_extents;
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002019 sector_t span;
2020 unsigned long maxpages;
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002021 unsigned char *swap_map = NULL;
2022 struct page *page = NULL;
2023 struct inode *inode = NULL;
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002024
Hugh Dickinsd15cab92012-03-28 14:42:42 -07002025 if (swap_flags & ~SWAP_FLAGS_VALID)
2026 return -EINVAL;
2027
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002028 if (!capable(CAP_SYS_ADMIN))
2029 return -EPERM;
2030
2031 p = alloc_swap_info();
Cesar Eduardo Barros2542e5132011-03-22 16:33:18 -07002032 if (IS_ERR(p))
2033 return PTR_ERR(p);
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002034
Linus Torvalds1da177e2005-04-16 15:20:36 -07002035 name = getname(specialfile);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002036 if (IS_ERR(name)) {
Cesar Eduardo Barros7de7fb62011-03-22 16:33:22 -07002037 error = PTR_ERR(name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002038 name = NULL;
Cesar Eduardo Barrosbd690102011-03-22 16:33:25 -07002039 goto bad_swap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002040 }
2041 swap_file = filp_open(name, O_RDWR|O_LARGEFILE, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002042 if (IS_ERR(swap_file)) {
Cesar Eduardo Barros7de7fb62011-03-22 16:33:22 -07002043 error = PTR_ERR(swap_file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002044 swap_file = NULL;
Cesar Eduardo Barrosbd690102011-03-22 16:33:25 -07002045 goto bad_swap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002046 }
2047
2048 p->swap_file = swap_file;
2049 mapping = swap_file->f_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002050
Linus Torvalds1da177e2005-04-16 15:20:36 -07002051 for (i = 0; i < nr_swapfiles; i++) {
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002052 struct swap_info_struct *q = swap_info[i];
Linus Torvalds1da177e2005-04-16 15:20:36 -07002053
Cesar Eduardo Barrose8e6c2e2011-03-22 16:33:16 -07002054 if (q == p || !q->swap_file)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002055 continue;
Cesar Eduardo Barros7de7fb62011-03-22 16:33:22 -07002056 if (mapping == q->swap_file->f_mapping) {
2057 error = -EBUSY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002058 goto bad_swap;
Cesar Eduardo Barros7de7fb62011-03-22 16:33:22 -07002059 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002060 }
2061
Cesar Eduardo Barros21307812011-03-22 23:03:13 -03002062 inode = mapping->host;
2063 /* If S_ISREG(inode->i_mode) will do mutex_lock(&inode->i_mutex); */
Cesar Eduardo Barros4d0e1e12011-03-22 16:33:26 -07002064 error = claim_swapfile(p, inode);
2065 if (unlikely(error))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002066 goto bad_swap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002067
Linus Torvalds1da177e2005-04-16 15:20:36 -07002068 /*
2069 * Read the swap header.
2070 */
2071 if (!mapping->a_ops->readpage) {
2072 error = -EINVAL;
2073 goto bad_swap;
2074 }
Pekka Enberg090d2b12006-06-23 02:05:08 -07002075 page = read_mapping_page(mapping, 0, swap_file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002076 if (IS_ERR(page)) {
2077 error = PTR_ERR(page);
2078 goto bad_swap;
2079 }
Hugh Dickins81e33972009-01-06 14:39:49 -08002080 swap_header = kmap(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002081
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002082 maxpages = read_swap_header(p, swap_header, inode);
2083 if (unlikely(!maxpages)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002084 error = -EINVAL;
2085 goto bad_swap;
2086 }
Hugh Dickins886bb7e2009-01-06 14:39:48 -08002087
Hugh Dickins81e33972009-01-06 14:39:49 -08002088 /* OK, set up the swap map and apply the bad block list */
Cesar Eduardo Barros803d0c832011-03-22 16:33:14 -07002089 swap_map = vzalloc(maxpages);
Hugh Dickins81e33972009-01-06 14:39:49 -08002090 if (!swap_map) {
2091 error = -ENOMEM;
2092 goto bad_swap;
2093 }
2094
Cesar Eduardo Barros1421ef32011-03-22 16:33:31 -07002095 error = swap_cgroup_swapon(p->type, maxpages);
2096 if (error)
2097 goto bad_swap;
2098
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002099 nr_extents = setup_swap_map_and_extents(p, swap_header, swap_map,
2100 maxpages, &span);
2101 if (unlikely(nr_extents < 0)) {
2102 error = nr_extents;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002103 goto bad_swap;
2104 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002105
Suresh Jayaraman3bd0f0c2009-09-30 10:53:48 +02002106 if (p->bdev) {
2107 if (blk_queue_nonrot(bdev_get_queue(p->bdev))) {
2108 p->flags |= SWP_SOLIDSTATE;
2109 p->cluster_next = 1 + (random32() % p->highest_bit);
2110 }
Shaohua Li052b1982012-03-21 16:34:17 -07002111 if ((swap_flags & SWAP_FLAG_DISCARD) && discard_swap(p) == 0)
Suresh Jayaraman3bd0f0c2009-09-30 10:53:48 +02002112 p->flags |= SWP_DISCARDABLE;
Hugh Dickins20137a42009-01-06 14:39:54 -08002113 }
Hugh Dickins6a6ba832009-01-06 14:39:51 -08002114
Ingo Molnarfc0abb12006-01-18 17:42:33 -08002115 mutex_lock(&swapon_mutex);
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07002116 prio = -1;
Hugh Dickins78ecba02008-07-23 21:28:23 -07002117 if (swap_flags & SWAP_FLAG_PREFER)
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07002118 prio =
Hugh Dickins78ecba02008-07-23 21:28:23 -07002119 (swap_flags & SWAP_FLAG_PRIO_MASK) >> SWAP_FLAG_PRIO_SHIFT;
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07002120 enable_swap_info(p, prio, swap_map);
Cesar Eduardo Barrosc69dbfb2011-03-22 16:33:35 -07002121
2122 printk(KERN_INFO "Adding %uk swap on %s. "
2123 "Priority:%d extents:%d across:%lluk %s%s\n",
2124 p->pages<<(PAGE_SHIFT-10), name, p->prio,
2125 nr_extents, (unsigned long long)span<<(PAGE_SHIFT-10),
2126 (p->flags & SWP_SOLIDSTATE) ? "SS" : "",
2127 (p->flags & SWP_DISCARDABLE) ? "D" : "");
2128
Ingo Molnarfc0abb12006-01-18 17:42:33 -08002129 mutex_unlock(&swapon_mutex);
Kay Sievers66d7dd52010-10-26 14:22:06 -07002130 atomic_inc(&proc_poll_event);
2131 wake_up_interruptible(&proc_poll_wait);
2132
Cesar Eduardo Barros9b01c352011-03-22 16:33:24 -07002133 if (S_ISREG(inode->i_mode))
2134 inode->i_flags |= S_SWAPFILE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002135 error = 0;
2136 goto out;
2137bad_swap:
Cesar Eduardo Barrosbd690102011-03-22 16:33:25 -07002138 if (inode && S_ISBLK(inode->i_mode) && p->bdev) {
Cesar Eduardo Barrosf2090d22011-03-22 16:33:23 -07002139 set_blocksize(p->bdev, p->old_block_size);
2140 blkdev_put(p->bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002141 }
Hugh Dickins4cd3bb12005-09-03 15:54:33 -07002142 destroy_swap_extents(p);
Cesar Eduardo Barrose8e6c2e2011-03-22 16:33:16 -07002143 swap_cgroup_swapoff(p->type);
Hugh Dickins5d337b92005-09-03 15:54:41 -07002144 spin_lock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002145 p->swap_file = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002146 p->flags = 0;
Hugh Dickins5d337b92005-09-03 15:54:41 -07002147 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002148 vfree(swap_map);
Mel Gorman52c50562011-03-22 16:30:08 -07002149 if (swap_file) {
Cesar Eduardo Barros21307812011-03-22 23:03:13 -03002150 if (inode && S_ISREG(inode->i_mode)) {
Mel Gorman52c50562011-03-22 16:30:08 -07002151 mutex_unlock(&inode->i_mutex);
Cesar Eduardo Barros21307812011-03-22 23:03:13 -03002152 inode = NULL;
2153 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002154 filp_close(swap_file, NULL);
Mel Gorman52c50562011-03-22 16:30:08 -07002155 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002156out:
2157 if (page && !IS_ERR(page)) {
2158 kunmap(page);
2159 page_cache_release(page);
2160 }
2161 if (name)
2162 putname(name);
Cesar Eduardo Barros9b01c352011-03-22 16:33:24 -07002163 if (inode && S_ISREG(inode->i_mode))
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002164 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002165 return error;
2166}
2167
2168void si_swapinfo(struct sysinfo *val)
2169{
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002170 unsigned int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002171 unsigned long nr_to_be_unused = 0;
2172
Hugh Dickins5d337b92005-09-03 15:54:41 -07002173 spin_lock(&swap_lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002174 for (type = 0; type < nr_swapfiles; type++) {
2175 struct swap_info_struct *si = swap_info[type];
2176
2177 if ((si->flags & SWP_USED) && !(si->flags & SWP_WRITEOK))
2178 nr_to_be_unused += si->inuse_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002179 }
2180 val->freeswap = nr_swap_pages + nr_to_be_unused;
2181 val->totalswap = total_swap_pages + nr_to_be_unused;
Hugh Dickins5d337b92005-09-03 15:54:41 -07002182 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002183}
2184
2185/*
2186 * Verify that a swap entry is valid and increment its swap map count.
2187 *
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002188 * Returns error code in following case.
2189 * - success -> 0
2190 * - swp_entry is invalid -> EINVAL
2191 * - swp_entry is migration entry -> EINVAL
2192 * - swap-cache reference is requested but there is already one. -> EEXIST
2193 * - swap-cache reference is requested but the entry is not used. -> ENOENT
Hugh Dickins570a335b2009-12-14 17:58:46 -08002194 * - swap-mapped reference requested but needs continued swap count. -> ENOMEM
Linus Torvalds1da177e2005-04-16 15:20:36 -07002195 */
Hugh Dickins8d69aae2009-12-14 17:58:45 -08002196static int __swap_duplicate(swp_entry_t entry, unsigned char usage)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002197{
Hugh Dickins73c34b62009-12-14 17:58:43 -08002198 struct swap_info_struct *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002199 unsigned long offset, type;
Hugh Dickins8d69aae2009-12-14 17:58:45 -08002200 unsigned char count;
2201 unsigned char has_cache;
Hugh Dickins253d5532009-12-14 17:58:44 -08002202 int err = -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002203
Andi Kleena7420aa2009-09-16 11:50:05 +02002204 if (non_swap_entry(entry))
Hugh Dickins253d5532009-12-14 17:58:44 -08002205 goto out;
Christoph Lameter06972122006-06-23 02:03:35 -07002206
Linus Torvalds1da177e2005-04-16 15:20:36 -07002207 type = swp_type(entry);
2208 if (type >= nr_swapfiles)
2209 goto bad_file;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002210 p = swap_info[type];
Linus Torvalds1da177e2005-04-16 15:20:36 -07002211 offset = swp_offset(entry);
2212
Hugh Dickins5d337b92005-09-03 15:54:41 -07002213 spin_lock(&swap_lock);
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002214 if (unlikely(offset >= p->max))
2215 goto unlock_out;
2216
Hugh Dickins253d5532009-12-14 17:58:44 -08002217 count = p->swap_map[offset];
2218 has_cache = count & SWAP_HAS_CACHE;
2219 count &= ~SWAP_HAS_CACHE;
2220 err = 0;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002221
Hugh Dickins253d5532009-12-14 17:58:44 -08002222 if (usage == SWAP_HAS_CACHE) {
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002223
2224 /* set SWAP_HAS_CACHE if there is no cache and entry is used */
Hugh Dickins253d5532009-12-14 17:58:44 -08002225 if (!has_cache && count)
2226 has_cache = SWAP_HAS_CACHE;
2227 else if (has_cache) /* someone else added cache */
2228 err = -EEXIST;
2229 else /* no users remaining */
2230 err = -ENOENT;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002231
2232 } else if (count || has_cache) {
Hugh Dickins253d5532009-12-14 17:58:44 -08002233
Hugh Dickins570a335b2009-12-14 17:58:46 -08002234 if ((count & ~COUNT_CONTINUED) < SWAP_MAP_MAX)
2235 count += usage;
2236 else if ((count & ~COUNT_CONTINUED) > SWAP_MAP_MAX)
Hugh Dickins253d5532009-12-14 17:58:44 -08002237 err = -EINVAL;
Hugh Dickins570a335b2009-12-14 17:58:46 -08002238 else if (swap_count_continued(p, offset, count))
2239 count = COUNT_CONTINUED;
2240 else
2241 err = -ENOMEM;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002242 } else
Hugh Dickins253d5532009-12-14 17:58:44 -08002243 err = -ENOENT; /* unused swap entry */
2244
2245 p->swap_map[offset] = count | has_cache;
2246
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002247unlock_out:
Hugh Dickins5d337b92005-09-03 15:54:41 -07002248 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002249out:
Hugh Dickins253d5532009-12-14 17:58:44 -08002250 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002251
2252bad_file:
2253 printk(KERN_ERR "swap_dup: %s%08lx\n", Bad_file, entry.val);
2254 goto out;
2255}
Hugh Dickins253d5532009-12-14 17:58:44 -08002256
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002257/*
Hugh Dickinsaaa46862009-12-14 17:58:47 -08002258 * Help swapoff by noting that swap entry belongs to shmem/tmpfs
2259 * (in which case its reference count is never incremented).
2260 */
2261void swap_shmem_alloc(swp_entry_t entry)
2262{
2263 __swap_duplicate(entry, SWAP_MAP_SHMEM);
2264}
2265
2266/*
Hugh Dickins08259d52010-03-05 13:42:25 -08002267 * Increase reference count of swap entry by 1.
2268 * Returns 0 for success, or -ENOMEM if a swap_count_continuation is required
2269 * but could not be atomically allocated. Returns 0, just as if it succeeded,
2270 * if __swap_duplicate() fails for another reason (-EINVAL or -ENOENT), which
2271 * might occur if a page table entry has got corrupted.
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002272 */
Hugh Dickins570a335b2009-12-14 17:58:46 -08002273int swap_duplicate(swp_entry_t entry)
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002274{
Hugh Dickins570a335b2009-12-14 17:58:46 -08002275 int err = 0;
2276
2277 while (!err && __swap_duplicate(entry, 1) == -ENOMEM)
2278 err = add_swap_count_continuation(entry, GFP_ATOMIC);
2279 return err;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002280}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002281
KAMEZAWA Hiroyukicb4b86b2009-06-16 15:32:52 -07002282/*
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002283 * @entry: swap entry for which we allocate swap cache.
2284 *
Hugh Dickins73c34b62009-12-14 17:58:43 -08002285 * Called when allocating swap cache for existing swap entry,
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002286 * This can return error codes. Returns 0 at success.
2287 * -EBUSY means there is a swap cache.
2288 * Note: return code is different from swap_duplicate().
KAMEZAWA Hiroyukicb4b86b2009-06-16 15:32:52 -07002289 */
2290int swapcache_prepare(swp_entry_t entry)
2291{
Hugh Dickins253d5532009-12-14 17:58:44 -08002292 return __swap_duplicate(entry, SWAP_HAS_CACHE);
KAMEZAWA Hiroyukicb4b86b2009-06-16 15:32:52 -07002293}
2294
Linus Torvalds1da177e2005-04-16 15:20:36 -07002295/*
Hugh Dickins570a335b2009-12-14 17:58:46 -08002296 * add_swap_count_continuation - called when a swap count is duplicated
2297 * beyond SWAP_MAP_MAX, it allocates a new page and links that to the entry's
2298 * page of the original vmalloc'ed swap_map, to hold the continuation count
2299 * (for that entry and for its neighbouring PAGE_SIZE swap entries). Called
2300 * again when count is duplicated beyond SWAP_MAP_MAX * SWAP_CONT_MAX, etc.
2301 *
2302 * These continuation pages are seldom referenced: the common paths all work
2303 * on the original swap_map, only referring to a continuation page when the
2304 * low "digit" of a count is incremented or decremented through SWAP_MAP_MAX.
2305 *
2306 * add_swap_count_continuation(, GFP_ATOMIC) can be called while holding
2307 * page table locks; if it fails, add_swap_count_continuation(, GFP_KERNEL)
2308 * can be called after dropping locks.
2309 */
2310int add_swap_count_continuation(swp_entry_t entry, gfp_t gfp_mask)
2311{
2312 struct swap_info_struct *si;
2313 struct page *head;
2314 struct page *page;
2315 struct page *list_page;
2316 pgoff_t offset;
2317 unsigned char count;
2318
2319 /*
2320 * When debugging, it's easier to use __GFP_ZERO here; but it's better
2321 * for latency not to zero a page while GFP_ATOMIC and holding locks.
2322 */
2323 page = alloc_page(gfp_mask | __GFP_HIGHMEM);
2324
2325 si = swap_info_get(entry);
2326 if (!si) {
2327 /*
2328 * An acceptable race has occurred since the failing
2329 * __swap_duplicate(): the swap entry has been freed,
2330 * perhaps even the whole swap_map cleared for swapoff.
2331 */
2332 goto outer;
2333 }
2334
2335 offset = swp_offset(entry);
2336 count = si->swap_map[offset] & ~SWAP_HAS_CACHE;
2337
2338 if ((count & ~COUNT_CONTINUED) != SWAP_MAP_MAX) {
2339 /*
2340 * The higher the swap count, the more likely it is that tasks
2341 * will race to add swap count continuation: we need to avoid
2342 * over-provisioning.
2343 */
2344 goto out;
2345 }
2346
2347 if (!page) {
2348 spin_unlock(&swap_lock);
2349 return -ENOMEM;
2350 }
2351
2352 /*
2353 * We are fortunate that although vmalloc_to_page uses pte_offset_map,
2354 * no architecture is using highmem pages for kernel pagetables: so it
2355 * will not corrupt the GFP_ATOMIC caller's atomic pagetable kmaps.
2356 */
2357 head = vmalloc_to_page(si->swap_map + offset);
2358 offset &= ~PAGE_MASK;
2359
2360 /*
2361 * Page allocation does not initialize the page's lru field,
2362 * but it does always reset its private field.
2363 */
2364 if (!page_private(head)) {
2365 BUG_ON(count & COUNT_CONTINUED);
2366 INIT_LIST_HEAD(&head->lru);
2367 set_page_private(head, SWP_CONTINUED);
2368 si->flags |= SWP_CONTINUED;
2369 }
2370
2371 list_for_each_entry(list_page, &head->lru, lru) {
2372 unsigned char *map;
2373
2374 /*
2375 * If the previous map said no continuation, but we've found
2376 * a continuation page, free our allocation and use this one.
2377 */
2378 if (!(count & COUNT_CONTINUED))
2379 goto out;
2380
Cong Wang9b04c5f2011-11-25 23:14:39 +08002381 map = kmap_atomic(list_page) + offset;
Hugh Dickins570a335b2009-12-14 17:58:46 -08002382 count = *map;
Cong Wang9b04c5f2011-11-25 23:14:39 +08002383 kunmap_atomic(map);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002384
2385 /*
2386 * If this continuation count now has some space in it,
2387 * free our allocation and use this one.
2388 */
2389 if ((count & ~COUNT_CONTINUED) != SWAP_CONT_MAX)
2390 goto out;
2391 }
2392
2393 list_add_tail(&page->lru, &head->lru);
2394 page = NULL; /* now it's attached, don't free it */
2395out:
2396 spin_unlock(&swap_lock);
2397outer:
2398 if (page)
2399 __free_page(page);
2400 return 0;
2401}
2402
2403/*
2404 * swap_count_continued - when the original swap_map count is incremented
2405 * from SWAP_MAP_MAX, check if there is already a continuation page to carry
2406 * into, carry if so, or else fail until a new continuation page is allocated;
2407 * when the original swap_map count is decremented from 0 with continuation,
2408 * borrow from the continuation and report whether it still holds more.
2409 * Called while __swap_duplicate() or swap_entry_free() holds swap_lock.
2410 */
2411static bool swap_count_continued(struct swap_info_struct *si,
2412 pgoff_t offset, unsigned char count)
2413{
2414 struct page *head;
2415 struct page *page;
2416 unsigned char *map;
2417
2418 head = vmalloc_to_page(si->swap_map + offset);
2419 if (page_private(head) != SWP_CONTINUED) {
2420 BUG_ON(count & COUNT_CONTINUED);
2421 return false; /* need to add count continuation */
2422 }
2423
2424 offset &= ~PAGE_MASK;
2425 page = list_entry(head->lru.next, struct page, lru);
Cong Wang9b04c5f2011-11-25 23:14:39 +08002426 map = kmap_atomic(page) + offset;
Hugh Dickins570a335b2009-12-14 17:58:46 -08002427
2428 if (count == SWAP_MAP_MAX) /* initial increment from swap_map */
2429 goto init_map; /* jump over SWAP_CONT_MAX checks */
2430
2431 if (count == (SWAP_MAP_MAX | COUNT_CONTINUED)) { /* incrementing */
2432 /*
2433 * Think of how you add 1 to 999
2434 */
2435 while (*map == (SWAP_CONT_MAX | COUNT_CONTINUED)) {
Cong Wang9b04c5f2011-11-25 23:14:39 +08002436 kunmap_atomic(map);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002437 page = list_entry(page->lru.next, struct page, lru);
2438 BUG_ON(page == head);
Cong Wang9b04c5f2011-11-25 23:14:39 +08002439 map = kmap_atomic(page) + offset;
Hugh Dickins570a335b2009-12-14 17:58:46 -08002440 }
2441 if (*map == SWAP_CONT_MAX) {
Cong Wang9b04c5f2011-11-25 23:14:39 +08002442 kunmap_atomic(map);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002443 page = list_entry(page->lru.next, struct page, lru);
2444 if (page == head)
2445 return false; /* add count continuation */
Cong Wang9b04c5f2011-11-25 23:14:39 +08002446 map = kmap_atomic(page) + offset;
Hugh Dickins570a335b2009-12-14 17:58:46 -08002447init_map: *map = 0; /* we didn't zero the page */
2448 }
2449 *map += 1;
Cong Wang9b04c5f2011-11-25 23:14:39 +08002450 kunmap_atomic(map);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002451 page = list_entry(page->lru.prev, struct page, lru);
2452 while (page != head) {
Cong Wang9b04c5f2011-11-25 23:14:39 +08002453 map = kmap_atomic(page) + offset;
Hugh Dickins570a335b2009-12-14 17:58:46 -08002454 *map = COUNT_CONTINUED;
Cong Wang9b04c5f2011-11-25 23:14:39 +08002455 kunmap_atomic(map);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002456 page = list_entry(page->lru.prev, struct page, lru);
2457 }
2458 return true; /* incremented */
2459
2460 } else { /* decrementing */
2461 /*
2462 * Think of how you subtract 1 from 1000
2463 */
2464 BUG_ON(count != COUNT_CONTINUED);
2465 while (*map == COUNT_CONTINUED) {
Cong Wang9b04c5f2011-11-25 23:14:39 +08002466 kunmap_atomic(map);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002467 page = list_entry(page->lru.next, struct page, lru);
2468 BUG_ON(page == head);
Cong Wang9b04c5f2011-11-25 23:14:39 +08002469 map = kmap_atomic(page) + offset;
Hugh Dickins570a335b2009-12-14 17:58:46 -08002470 }
2471 BUG_ON(*map == 0);
2472 *map -= 1;
2473 if (*map == 0)
2474 count = 0;
Cong Wang9b04c5f2011-11-25 23:14:39 +08002475 kunmap_atomic(map);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002476 page = list_entry(page->lru.prev, struct page, lru);
2477 while (page != head) {
Cong Wang9b04c5f2011-11-25 23:14:39 +08002478 map = kmap_atomic(page) + offset;
Hugh Dickins570a335b2009-12-14 17:58:46 -08002479 *map = SWAP_CONT_MAX | count;
2480 count = COUNT_CONTINUED;
Cong Wang9b04c5f2011-11-25 23:14:39 +08002481 kunmap_atomic(map);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002482 page = list_entry(page->lru.prev, struct page, lru);
2483 }
2484 return count == COUNT_CONTINUED;
2485 }
2486}
2487
2488/*
2489 * free_swap_count_continuations - swapoff free all the continuation pages
2490 * appended to the swap_map, after swap_map is quiesced, before vfree'ing it.
2491 */
2492static void free_swap_count_continuations(struct swap_info_struct *si)
2493{
2494 pgoff_t offset;
2495
2496 for (offset = 0; offset < si->max; offset += PAGE_SIZE) {
2497 struct page *head;
2498 head = vmalloc_to_page(si->swap_map + offset);
2499 if (page_private(head)) {
2500 struct list_head *this, *next;
2501 list_for_each_safe(this, next, &head->lru) {
2502 struct page *page;
2503 page = list_entry(this, struct page, lru);
2504 list_del(this);
2505 __free_page(page);
2506 }
2507 }
2508 }
2509}