blob: 457b10baef59414f591fb0bfab2b54619c5209b0 [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 Dickinsbde05d12012-05-29 15:06:38 -0700604int 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
Rafael J. Wysockib0cb1a12007-07-29 23:24:36 +0200720#ifdef CONFIG_HIBERNATION
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800721/*
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800722 * Find the swap type that corresponds to given device (if any).
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800723 *
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800724 * @offset - number of the PAGE_SIZE-sized block of the device, starting
725 * from 0, in which the swap header is expected to be located.
726 *
727 * This is needed for the suspend to disk (aka swsusp).
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800728 */
Rafael J. Wysocki7bf23682007-01-05 16:36:28 -0800729int swap_type_of(dev_t device, sector_t offset, struct block_device **bdev_p)
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800730{
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800731 struct block_device *bdev = NULL;
Hugh Dickinsefa90a92009-12-14 17:58:41 -0800732 int type;
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800733
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800734 if (device)
735 bdev = bdget(device);
736
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800737 spin_lock(&swap_lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -0800738 for (type = 0; type < nr_swapfiles; type++) {
739 struct swap_info_struct *sis = swap_info[type];
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800740
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800741 if (!(sis->flags & SWP_WRITEOK))
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800742 continue;
Rafael J. Wysockib6b5bce2006-08-27 01:23:25 -0700743
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800744 if (!bdev) {
Rafael J. Wysocki7bf23682007-01-05 16:36:28 -0800745 if (bdev_p)
Alan Jenkinsdddac6a2009-07-29 21:07:55 +0200746 *bdev_p = bdgrab(sis->bdev);
Rafael J. Wysocki7bf23682007-01-05 16:36:28 -0800747
Rafael J. Wysocki6e1819d2006-03-23 03:00:03 -0800748 spin_unlock(&swap_lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -0800749 return type;
Rafael J. Wysocki6e1819d2006-03-23 03:00:03 -0800750 }
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800751 if (bdev == sis->bdev) {
Hugh Dickins9625a5f2009-12-14 17:58:42 -0800752 struct swap_extent *se = &sis->first_swap_extent;
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800753
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800754 if (se->start_block == offset) {
Rafael J. Wysocki7bf23682007-01-05 16:36:28 -0800755 if (bdev_p)
Alan Jenkinsdddac6a2009-07-29 21:07:55 +0200756 *bdev_p = bdgrab(sis->bdev);
Rafael J. Wysocki7bf23682007-01-05 16:36:28 -0800757
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800758 spin_unlock(&swap_lock);
759 bdput(bdev);
Hugh Dickinsefa90a92009-12-14 17:58:41 -0800760 return type;
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800761 }
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800762 }
763 }
764 spin_unlock(&swap_lock);
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800765 if (bdev)
766 bdput(bdev);
767
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800768 return -ENODEV;
769}
770
771/*
Hugh Dickins73c34b62009-12-14 17:58:43 -0800772 * Get the (PAGE_SIZE) block corresponding to given offset on the swapdev
773 * corresponding to given index in swap_info (swap type).
774 */
775sector_t swapdev_block(int type, pgoff_t offset)
776{
777 struct block_device *bdev;
778
779 if ((unsigned int)type >= nr_swapfiles)
780 return 0;
781 if (!(swap_info[type]->flags & SWP_WRITEOK))
782 return 0;
Lee Schermerhornd4906e12009-12-14 17:58:49 -0800783 return map_swap_entry(swp_entry(type, offset), &bdev);
Hugh Dickins73c34b62009-12-14 17:58:43 -0800784}
785
786/*
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800787 * Return either the total number of swap pages of given type, or the number
788 * of free pages of that type (depending on @free)
789 *
790 * This is needed for software suspend
791 */
792unsigned int count_swap_pages(int type, int free)
793{
794 unsigned int n = 0;
795
Hugh Dickinsefa90a92009-12-14 17:58:41 -0800796 spin_lock(&swap_lock);
797 if ((unsigned int)type < nr_swapfiles) {
798 struct swap_info_struct *sis = swap_info[type];
799
800 if (sis->flags & SWP_WRITEOK) {
801 n = sis->pages;
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800802 if (free)
Hugh Dickinsefa90a92009-12-14 17:58:41 -0800803 n -= sis->inuse_pages;
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800804 }
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800805 }
Hugh Dickinsefa90a92009-12-14 17:58:41 -0800806 spin_unlock(&swap_lock);
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800807 return n;
808}
Hugh Dickins73c34b62009-12-14 17:58:43 -0800809#endif /* CONFIG_HIBERNATION */
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800810
Linus Torvalds1da177e2005-04-16 15:20:36 -0700811/*
Hugh Dickins72866f62005-10-29 18:15:55 -0700812 * No need to decide whether this PTE shares the swap entry with others,
813 * just let do_wp_page work it out if a write is requested later - to
814 * force COW, vm_page_prot omits write permission from any private vma.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700815 */
Hugh Dickins044d66c2008-02-07 00:14:04 -0800816static int unuse_pte(struct vm_area_struct *vma, pmd_t *pmd,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700817 unsigned long addr, swp_entry_t entry, struct page *page)
818{
Johannes Weiner72835c82012-01-12 17:18:32 -0800819 struct mem_cgroup *memcg;
Hugh Dickins044d66c2008-02-07 00:14:04 -0800820 spinlock_t *ptl;
821 pte_t *pte;
822 int ret = 1;
823
Johannes Weiner72835c82012-01-12 17:18:32 -0800824 if (mem_cgroup_try_charge_swapin(vma->vm_mm, page,
825 GFP_KERNEL, &memcg)) {
Hugh Dickins044d66c2008-02-07 00:14:04 -0800826 ret = -ENOMEM;
KAMEZAWA Hiroyuki85d9fc82009-01-29 14:25:13 -0800827 goto out_nolock;
828 }
Hugh Dickins044d66c2008-02-07 00:14:04 -0800829
830 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
831 if (unlikely(!pte_same(*pte, swp_entry_to_pte(entry)))) {
832 if (ret > 0)
Johannes Weiner72835c82012-01-12 17:18:32 -0800833 mem_cgroup_cancel_charge_swapin(memcg);
Hugh Dickins044d66c2008-02-07 00:14:04 -0800834 ret = 0;
835 goto out;
836 }
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800837
KAMEZAWA Hiroyukib084d432010-03-05 13:41:42 -0800838 dec_mm_counter(vma->vm_mm, MM_SWAPENTS);
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -0800839 inc_mm_counter(vma->vm_mm, MM_ANONPAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700840 get_page(page);
841 set_pte_at(vma->vm_mm, addr, pte,
842 pte_mkold(mk_pte(page, vma->vm_page_prot)));
843 page_add_anon_rmap(page, vma, addr);
Johannes Weiner72835c82012-01-12 17:18:32 -0800844 mem_cgroup_commit_charge_swapin(page, memcg);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700845 swap_free(entry);
846 /*
847 * Move the page to the active list so it is not
848 * immediately swapped out again after swapon.
849 */
850 activate_page(page);
Hugh Dickins044d66c2008-02-07 00:14:04 -0800851out:
852 pte_unmap_unlock(pte, ptl);
KAMEZAWA Hiroyuki85d9fc82009-01-29 14:25:13 -0800853out_nolock:
Hugh Dickins044d66c2008-02-07 00:14:04 -0800854 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700855}
856
857static int unuse_pte_range(struct vm_area_struct *vma, pmd_t *pmd,
858 unsigned long addr, unsigned long end,
859 swp_entry_t entry, struct page *page)
860{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700861 pte_t swp_pte = swp_entry_to_pte(entry);
Hugh Dickins705e87c2005-10-29 18:16:27 -0700862 pte_t *pte;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800863 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700864
Hugh Dickins044d66c2008-02-07 00:14:04 -0800865 /*
866 * We don't actually need pte lock while scanning for swp_pte: since
867 * we hold page lock and mmap_sem, swp_pte cannot be inserted into the
868 * page table while we're scanning; though it could get zapped, and on
869 * some architectures (e.g. x86_32 with PAE) we might catch a glimpse
870 * of unmatched parts which look like swp_pte, so unuse_pte must
871 * recheck under pte lock. Scanning without pte lock lets it be
872 * preemptible whenever CONFIG_PREEMPT but not CONFIG_HIGHPTE.
873 */
874 pte = pte_offset_map(pmd, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700875 do {
876 /*
877 * swapoff spends a _lot_ of time in this loop!
878 * Test inline before going to call unuse_pte.
879 */
880 if (unlikely(pte_same(*pte, swp_pte))) {
Hugh Dickins044d66c2008-02-07 00:14:04 -0800881 pte_unmap(pte);
882 ret = unuse_pte(vma, pmd, addr, entry, page);
883 if (ret)
884 goto out;
885 pte = pte_offset_map(pmd, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700886 }
887 } while (pte++, addr += PAGE_SIZE, addr != end);
Hugh Dickins044d66c2008-02-07 00:14:04 -0800888 pte_unmap(pte - 1);
889out:
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800890 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700891}
892
893static inline int unuse_pmd_range(struct vm_area_struct *vma, pud_t *pud,
894 unsigned long addr, unsigned long end,
895 swp_entry_t entry, struct page *page)
896{
897 pmd_t *pmd;
898 unsigned long next;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800899 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700900
901 pmd = pmd_offset(pud, addr);
902 do {
903 next = pmd_addr_end(addr, end);
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -0700904 if (pmd_none_or_trans_huge_or_clear_bad(pmd))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700905 continue;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800906 ret = unuse_pte_range(vma, pmd, addr, next, entry, page);
907 if (ret)
908 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700909 } while (pmd++, addr = next, addr != end);
910 return 0;
911}
912
913static inline int unuse_pud_range(struct vm_area_struct *vma, pgd_t *pgd,
914 unsigned long addr, unsigned long end,
915 swp_entry_t entry, struct page *page)
916{
917 pud_t *pud;
918 unsigned long next;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800919 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700920
921 pud = pud_offset(pgd, addr);
922 do {
923 next = pud_addr_end(addr, end);
924 if (pud_none_or_clear_bad(pud))
925 continue;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800926 ret = unuse_pmd_range(vma, pud, addr, next, entry, page);
927 if (ret)
928 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700929 } while (pud++, addr = next, addr != end);
930 return 0;
931}
932
933static int unuse_vma(struct vm_area_struct *vma,
934 swp_entry_t entry, struct page *page)
935{
936 pgd_t *pgd;
937 unsigned long addr, end, next;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800938 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700939
Hugh Dickins3ca7b3c2009-12-14 17:58:57 -0800940 if (page_anon_vma(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700941 addr = page_address_in_vma(page, vma);
942 if (addr == -EFAULT)
943 return 0;
944 else
945 end = addr + PAGE_SIZE;
946 } else {
947 addr = vma->vm_start;
948 end = vma->vm_end;
949 }
950
951 pgd = pgd_offset(vma->vm_mm, addr);
952 do {
953 next = pgd_addr_end(addr, end);
954 if (pgd_none_or_clear_bad(pgd))
955 continue;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800956 ret = unuse_pud_range(vma, pgd, addr, next, entry, page);
957 if (ret)
958 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700959 } while (pgd++, addr = next, addr != end);
960 return 0;
961}
962
963static int unuse_mm(struct mm_struct *mm,
964 swp_entry_t entry, struct page *page)
965{
966 struct vm_area_struct *vma;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800967 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700968
969 if (!down_read_trylock(&mm->mmap_sem)) {
970 /*
Fernando Luis Vazquez Cao7d034312008-07-29 22:33:41 -0700971 * Activate page so shrink_inactive_list is unlikely to unmap
972 * its ptes while lock is dropped, so swapoff can make progress.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700973 */
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700974 activate_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700975 unlock_page(page);
976 down_read(&mm->mmap_sem);
977 lock_page(page);
978 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700979 for (vma = mm->mmap; vma; vma = vma->vm_next) {
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800980 if (vma->anon_vma && (ret = unuse_vma(vma, entry, page)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700981 break;
982 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700983 up_read(&mm->mmap_sem);
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800984 return (ret < 0)? ret: 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700985}
986
987/*
988 * Scan swap_map from current position to next entry still in use.
989 * Recycle to start on reaching the end, returning 0 when empty.
990 */
Hugh Dickins6eb396d2005-09-03 15:54:35 -0700991static unsigned int find_next_to_unuse(struct swap_info_struct *si,
992 unsigned int prev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700993{
Hugh Dickins6eb396d2005-09-03 15:54:35 -0700994 unsigned int max = si->max;
995 unsigned int i = prev;
Hugh Dickins8d69aae2009-12-14 17:58:45 -0800996 unsigned char count;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700997
998 /*
Hugh Dickins5d337b92005-09-03 15:54:41 -0700999 * No need for swap_lock here: we're just looking
Linus Torvalds1da177e2005-04-16 15:20:36 -07001000 * for whether an entry is in use, not modifying it; false
1001 * hits are okay, and sys_swapoff() has already prevented new
Hugh Dickins5d337b92005-09-03 15:54:41 -07001002 * allocations from this area (while holding swap_lock).
Linus Torvalds1da177e2005-04-16 15:20:36 -07001003 */
1004 for (;;) {
1005 if (++i >= max) {
1006 if (!prev) {
1007 i = 0;
1008 break;
1009 }
1010 /*
1011 * No entries in use at top of swap_map,
1012 * loop back to start and recheck there.
1013 */
1014 max = prev + 1;
1015 prev = 0;
1016 i = 1;
1017 }
1018 count = si->swap_map[i];
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07001019 if (count && swap_count(count) != SWAP_MAP_BAD)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001020 break;
1021 }
1022 return i;
1023}
1024
1025/*
1026 * We completely avoid races by reading each swap page in advance,
1027 * and then search for the process using it. All the necessary
1028 * page table adjustments can then be made atomically.
1029 */
1030static int try_to_unuse(unsigned int type)
1031{
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001032 struct swap_info_struct *si = swap_info[type];
Linus Torvalds1da177e2005-04-16 15:20:36 -07001033 struct mm_struct *start_mm;
Hugh Dickins8d69aae2009-12-14 17:58:45 -08001034 unsigned char *swap_map;
1035 unsigned char swcount;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001036 struct page *page;
1037 swp_entry_t entry;
Hugh Dickins6eb396d2005-09-03 15:54:35 -07001038 unsigned int i = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001039 int retval = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001040
1041 /*
1042 * When searching mms for an entry, a good strategy is to
1043 * start at the first mm we freed the previous entry from
1044 * (though actually we don't notice whether we or coincidence
1045 * freed the entry). Initialize this start_mm with a hold.
1046 *
1047 * A simpler strategy would be to start at the last mm we
1048 * freed the previous entry from; but that would take less
1049 * advantage of mmlist ordering, which clusters forked mms
1050 * together, child after parent. If we race with dup_mmap(), we
1051 * prefer to resolve parent before child, lest we miss entries
1052 * duplicated after we scanned child: using last mm would invert
Hugh Dickins570a335b2009-12-14 17:58:46 -08001053 * that.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001054 */
1055 start_mm = &init_mm;
1056 atomic_inc(&init_mm.mm_users);
1057
1058 /*
1059 * Keep on scanning until all entries have gone. Usually,
1060 * one pass through swap_map is enough, but not necessarily:
1061 * there are races when an instance of an entry might be missed.
1062 */
1063 while ((i = find_next_to_unuse(si, i)) != 0) {
1064 if (signal_pending(current)) {
1065 retval = -EINTR;
1066 break;
1067 }
1068
Hugh Dickins886bb7e2009-01-06 14:39:48 -08001069 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001070 * Get a page for the entry, using the existing swap
1071 * cache page if there is one. Otherwise, get a clean
Hugh Dickins886bb7e2009-01-06 14:39:48 -08001072 * page and read the swap into it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001073 */
1074 swap_map = &si->swap_map[i];
1075 entry = swp_entry(type, i);
Hugh Dickins02098fe2008-02-04 22:28:42 -08001076 page = read_swap_cache_async(entry,
1077 GFP_HIGHUSER_MOVABLE, NULL, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001078 if (!page) {
1079 /*
1080 * Either swap_duplicate() failed because entry
1081 * has been freed independently, and will not be
1082 * reused since sys_swapoff() already disabled
1083 * allocation from here, or alloc_page() failed.
1084 */
1085 if (!*swap_map)
1086 continue;
1087 retval = -ENOMEM;
1088 break;
1089 }
1090
1091 /*
1092 * Don't hold on to start_mm if it looks like exiting.
1093 */
1094 if (atomic_read(&start_mm->mm_users) == 1) {
1095 mmput(start_mm);
1096 start_mm = &init_mm;
1097 atomic_inc(&init_mm.mm_users);
1098 }
1099
1100 /*
1101 * Wait for and lock page. When do_swap_page races with
1102 * try_to_unuse, do_swap_page can handle the fault much
1103 * faster than try_to_unuse can locate the entry. This
1104 * apparently redundant "wait_on_page_locked" lets try_to_unuse
1105 * defer to do_swap_page in such a case - in some tests,
1106 * do_swap_page and try_to_unuse repeatedly compete.
1107 */
1108 wait_on_page_locked(page);
1109 wait_on_page_writeback(page);
1110 lock_page(page);
1111 wait_on_page_writeback(page);
1112
1113 /*
1114 * Remove all references to entry.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001115 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001116 swcount = *swap_map;
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001117 if (swap_count(swcount) == SWAP_MAP_SHMEM) {
1118 retval = shmem_unuse(entry, page);
1119 /* page has already been unlocked and released */
1120 if (retval < 0)
1121 break;
1122 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001123 }
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001124 if (swap_count(swcount) && start_mm != &init_mm)
1125 retval = unuse_mm(start_mm, entry, page);
1126
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07001127 if (swap_count(*swap_map)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001128 int set_start_mm = (*swap_map >= swcount);
1129 struct list_head *p = &start_mm->mmlist;
1130 struct mm_struct *new_start_mm = start_mm;
1131 struct mm_struct *prev_mm = start_mm;
1132 struct mm_struct *mm;
1133
1134 atomic_inc(&new_start_mm->mm_users);
1135 atomic_inc(&prev_mm->mm_users);
1136 spin_lock(&mmlist_lock);
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001137 while (swap_count(*swap_map) && !retval &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001138 (p = p->next) != &start_mm->mmlist) {
1139 mm = list_entry(p, struct mm_struct, mmlist);
Hugh Dickins70af7c52006-06-23 02:03:44 -07001140 if (!atomic_inc_not_zero(&mm->mm_users))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001141 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001142 spin_unlock(&mmlist_lock);
1143 mmput(prev_mm);
1144 prev_mm = mm;
1145
1146 cond_resched();
1147
1148 swcount = *swap_map;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07001149 if (!swap_count(swcount)) /* any usage ? */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001150 ;
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001151 else if (mm == &init_mm)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001152 set_start_mm = 1;
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001153 else
Linus Torvalds1da177e2005-04-16 15:20:36 -07001154 retval = unuse_mm(mm, entry, page);
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07001155
Bo Liu32c5fc12009-11-02 16:50:33 +00001156 if (set_start_mm && *swap_map < swcount) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001157 mmput(new_start_mm);
1158 atomic_inc(&mm->mm_users);
1159 new_start_mm = mm;
1160 set_start_mm = 0;
1161 }
1162 spin_lock(&mmlist_lock);
1163 }
1164 spin_unlock(&mmlist_lock);
1165 mmput(prev_mm);
1166 mmput(start_mm);
1167 start_mm = new_start_mm;
1168 }
1169 if (retval) {
1170 unlock_page(page);
1171 page_cache_release(page);
1172 break;
1173 }
1174
1175 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001176 * If a reference remains (rare), we would like to leave
1177 * the page in the swap cache; but try_to_unmap could
1178 * then re-duplicate the entry once we drop page lock,
1179 * so we might loop indefinitely; also, that page could
1180 * not be swapped out to other storage meanwhile. So:
1181 * delete from cache even if there's another reference,
1182 * after ensuring that the data has been saved to disk -
1183 * since if the reference remains (rarer), it will be
1184 * read from disk into another page. Splitting into two
1185 * pages would be incorrect if swap supported "shared
1186 * private" pages, but they are handled by tmpfs files.
Hugh Dickins5ad64682009-12-14 17:59:24 -08001187 *
1188 * Given how unuse_vma() targets one particular offset
1189 * in an anon_vma, once the anon_vma has been determined,
1190 * this splitting happens to be just what is needed to
1191 * handle where KSM pages have been swapped out: re-reading
1192 * is unnecessarily slow, but we can fix that later on.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001193 */
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07001194 if (swap_count(*swap_map) &&
1195 PageDirty(page) && PageSwapCache(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001196 struct writeback_control wbc = {
1197 .sync_mode = WB_SYNC_NONE,
1198 };
1199
1200 swap_writepage(page, &wbc);
1201 lock_page(page);
1202 wait_on_page_writeback(page);
1203 }
Hugh Dickins68bdc8d2009-01-06 14:39:37 -08001204
1205 /*
1206 * It is conceivable that a racing task removed this page from
1207 * swap cache just before we acquired the page lock at the top,
1208 * or while we dropped it in unuse_mm(). The page might even
1209 * be back in swap cache on another swap area: that we must not
1210 * delete, since it may not have been written out to swap yet.
1211 */
1212 if (PageSwapCache(page) &&
1213 likely(page_private(page) == entry.val))
Hugh Dickins2e0e26c2008-02-04 22:28:53 -08001214 delete_from_swap_cache(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001215
1216 /*
1217 * So we could skip searching mms once swap count went
1218 * to 1, we did not mark any present ptes as dirty: must
Anderson Briglia2706a1b2007-07-15 23:38:09 -07001219 * mark page dirty so shrink_page_list will preserve it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001220 */
1221 SetPageDirty(page);
1222 unlock_page(page);
1223 page_cache_release(page);
1224
1225 /*
1226 * Make sure that we aren't completely killing
1227 * interactive performance.
1228 */
1229 cond_resched();
1230 }
1231
1232 mmput(start_mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001233 return retval;
1234}
1235
1236/*
Hugh Dickins5d337b92005-09-03 15:54:41 -07001237 * After a successful try_to_unuse, if no swap is now in use, we know
1238 * we can empty the mmlist. swap_lock must be held on entry and exit.
1239 * Note that mmlist_lock nests inside swap_lock, and an mm must be
Linus Torvalds1da177e2005-04-16 15:20:36 -07001240 * added to the mmlist just after page_duplicate - before would be racy.
1241 */
1242static void drain_mmlist(void)
1243{
1244 struct list_head *p, *next;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001245 unsigned int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001246
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001247 for (type = 0; type < nr_swapfiles; type++)
1248 if (swap_info[type]->inuse_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001249 return;
1250 spin_lock(&mmlist_lock);
1251 list_for_each_safe(p, next, &init_mm.mmlist)
1252 list_del_init(p);
1253 spin_unlock(&mmlist_lock);
1254}
1255
1256/*
1257 * Use this swapdev's extent info to locate the (PAGE_SIZE) block which
Lee Schermerhornd4906e12009-12-14 17:58:49 -08001258 * corresponds to page offset for the specified swap entry.
1259 * Note that the type of this function is sector_t, but it returns page offset
1260 * into the bdev, not sector offset.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001261 */
Lee Schermerhornd4906e12009-12-14 17:58:49 -08001262static sector_t map_swap_entry(swp_entry_t entry, struct block_device **bdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001263{
Hugh Dickinsf29ad6a2009-12-14 17:58:40 -08001264 struct swap_info_struct *sis;
1265 struct swap_extent *start_se;
1266 struct swap_extent *se;
1267 pgoff_t offset;
1268
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001269 sis = swap_info[swp_type(entry)];
Hugh Dickinsf29ad6a2009-12-14 17:58:40 -08001270 *bdev = sis->bdev;
1271
1272 offset = swp_offset(entry);
1273 start_se = sis->curr_swap_extent;
1274 se = start_se;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001275
1276 for ( ; ; ) {
1277 struct list_head *lh;
1278
1279 if (se->start_page <= offset &&
1280 offset < (se->start_page + se->nr_pages)) {
1281 return se->start_block + (offset - se->start_page);
1282 }
Hugh Dickins11d31882005-09-03 15:54:34 -07001283 lh = se->list.next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001284 se = list_entry(lh, struct swap_extent, list);
1285 sis->curr_swap_extent = se;
1286 BUG_ON(se == start_se); /* It *must* be present */
1287 }
1288}
1289
1290/*
Lee Schermerhornd4906e12009-12-14 17:58:49 -08001291 * Returns the page offset into bdev for the specified page's swap entry.
1292 */
1293sector_t map_swap_page(struct page *page, struct block_device **bdev)
1294{
1295 swp_entry_t entry;
1296 entry.val = page_private(page);
1297 return map_swap_entry(entry, bdev);
1298}
1299
1300/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001301 * Free all of a swapdev's extent information
1302 */
1303static void destroy_swap_extents(struct swap_info_struct *sis)
1304{
Hugh Dickins9625a5f2009-12-14 17:58:42 -08001305 while (!list_empty(&sis->first_swap_extent.list)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001306 struct swap_extent *se;
1307
Hugh Dickins9625a5f2009-12-14 17:58:42 -08001308 se = list_entry(sis->first_swap_extent.list.next,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001309 struct swap_extent, list);
1310 list_del(&se->list);
1311 kfree(se);
1312 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001313}
1314
1315/*
1316 * Add a block range (and the corresponding page range) into this swapdev's
Hugh Dickins11d31882005-09-03 15:54:34 -07001317 * extent list. The extent list is kept sorted in page order.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001318 *
Hugh Dickins11d31882005-09-03 15:54:34 -07001319 * This function rather assumes that it is called in ascending page order.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001320 */
1321static int
1322add_swap_extent(struct swap_info_struct *sis, unsigned long start_page,
1323 unsigned long nr_pages, sector_t start_block)
1324{
1325 struct swap_extent *se;
1326 struct swap_extent *new_se;
1327 struct list_head *lh;
1328
Hugh Dickins9625a5f2009-12-14 17:58:42 -08001329 if (start_page == 0) {
1330 se = &sis->first_swap_extent;
1331 sis->curr_swap_extent = se;
1332 se->start_page = 0;
1333 se->nr_pages = nr_pages;
1334 se->start_block = start_block;
1335 return 1;
1336 } else {
1337 lh = sis->first_swap_extent.list.prev; /* Highest extent */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001338 se = list_entry(lh, struct swap_extent, list);
Hugh Dickins11d31882005-09-03 15:54:34 -07001339 BUG_ON(se->start_page + se->nr_pages != start_page);
1340 if (se->start_block + se->nr_pages == start_block) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001341 /* Merge it */
1342 se->nr_pages += nr_pages;
1343 return 0;
1344 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001345 }
1346
1347 /*
1348 * No merge. Insert a new extent, preserving ordering.
1349 */
1350 new_se = kmalloc(sizeof(*se), GFP_KERNEL);
1351 if (new_se == NULL)
1352 return -ENOMEM;
1353 new_se->start_page = start_page;
1354 new_se->nr_pages = nr_pages;
1355 new_se->start_block = start_block;
1356
Hugh Dickins9625a5f2009-12-14 17:58:42 -08001357 list_add_tail(&new_se->list, &sis->first_swap_extent.list);
Hugh Dickins53092a72005-09-03 15:54:34 -07001358 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001359}
1360
1361/*
1362 * A `swap extent' is a simple thing which maps a contiguous range of pages
1363 * onto a contiguous range of disk blocks. An ordered list of swap extents
1364 * is built at swapon time and is then used at swap_writepage/swap_readpage
1365 * time for locating where on disk a page belongs.
1366 *
1367 * If the swapfile is an S_ISBLK block device, a single extent is installed.
1368 * This is done so that the main operating code can treat S_ISBLK and S_ISREG
1369 * swap files identically.
1370 *
1371 * Whether the swapdev is an S_ISREG file or an S_ISBLK blockdev, the swap
1372 * extent list operates in PAGE_SIZE disk blocks. Both S_ISREG and S_ISBLK
1373 * swapfiles are handled *identically* after swapon time.
1374 *
1375 * For S_ISREG swapfiles, setup_swap_extents() will walk all the file's blocks
1376 * and will parse them into an ordered extent list, in PAGE_SIZE chunks. If
1377 * some stray blocks are found which do not fall within the PAGE_SIZE alignment
1378 * requirements, they are simply tossed out - we will never use those blocks
1379 * for swapping.
1380 *
Hugh Dickinsb0d9bcd2005-09-03 15:54:31 -07001381 * For S_ISREG swapfiles we set S_SWAPFILE across the life of the swapon. This
Linus Torvalds1da177e2005-04-16 15:20:36 -07001382 * prevents root from shooting her foot off by ftruncating an in-use swapfile,
1383 * which will scribble on the fs.
1384 *
1385 * The amount of disk space which a single swap extent represents varies.
1386 * Typically it is in the 1-4 megabyte range. So we can have hundreds of
1387 * extents in the list. To avoid much list walking, we cache the previous
1388 * search location in `curr_swap_extent', and start new searches from there.
1389 * This is extremely effective. The average number of iterations in
1390 * map_swap_page() has been measured at about 0.3 per page. - akpm.
1391 */
Hugh Dickins53092a72005-09-03 15:54:34 -07001392static int setup_swap_extents(struct swap_info_struct *sis, sector_t *span)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001393{
1394 struct inode *inode;
1395 unsigned blocks_per_page;
1396 unsigned long page_no;
1397 unsigned blkbits;
1398 sector_t probe_block;
1399 sector_t last_block;
Hugh Dickins53092a72005-09-03 15:54:34 -07001400 sector_t lowest_block = -1;
1401 sector_t highest_block = 0;
1402 int nr_extents = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001403 int ret;
1404
1405 inode = sis->swap_file->f_mapping->host;
1406 if (S_ISBLK(inode->i_mode)) {
1407 ret = add_swap_extent(sis, 0, sis->max, 0);
Hugh Dickins53092a72005-09-03 15:54:34 -07001408 *span = sis->pages;
Hugh Dickins9625a5f2009-12-14 17:58:42 -08001409 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001410 }
1411
1412 blkbits = inode->i_blkbits;
1413 blocks_per_page = PAGE_SIZE >> blkbits;
1414
1415 /*
1416 * Map all the blocks into the extent list. This code doesn't try
1417 * to be very smart.
1418 */
1419 probe_block = 0;
1420 page_no = 0;
1421 last_block = i_size_read(inode) >> blkbits;
1422 while ((probe_block + blocks_per_page) <= last_block &&
1423 page_no < sis->max) {
1424 unsigned block_in_page;
1425 sector_t first_block;
1426
1427 first_block = bmap(inode, probe_block);
1428 if (first_block == 0)
1429 goto bad_bmap;
1430
1431 /*
1432 * It must be PAGE_SIZE aligned on-disk
1433 */
1434 if (first_block & (blocks_per_page - 1)) {
1435 probe_block++;
1436 goto reprobe;
1437 }
1438
1439 for (block_in_page = 1; block_in_page < blocks_per_page;
1440 block_in_page++) {
1441 sector_t block;
1442
1443 block = bmap(inode, probe_block + block_in_page);
1444 if (block == 0)
1445 goto bad_bmap;
1446 if (block != first_block + block_in_page) {
1447 /* Discontiguity */
1448 probe_block++;
1449 goto reprobe;
1450 }
1451 }
1452
Hugh Dickins53092a72005-09-03 15:54:34 -07001453 first_block >>= (PAGE_SHIFT - blkbits);
1454 if (page_no) { /* exclude the header page */
1455 if (first_block < lowest_block)
1456 lowest_block = first_block;
1457 if (first_block > highest_block)
1458 highest_block = first_block;
1459 }
1460
Linus Torvalds1da177e2005-04-16 15:20:36 -07001461 /*
1462 * We found a PAGE_SIZE-length, PAGE_SIZE-aligned run of blocks
1463 */
Hugh Dickins53092a72005-09-03 15:54:34 -07001464 ret = add_swap_extent(sis, page_no, 1, first_block);
1465 if (ret < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001466 goto out;
Hugh Dickins53092a72005-09-03 15:54:34 -07001467 nr_extents += ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001468 page_no++;
1469 probe_block += blocks_per_page;
1470reprobe:
1471 continue;
1472 }
Hugh Dickins53092a72005-09-03 15:54:34 -07001473 ret = nr_extents;
1474 *span = 1 + highest_block - lowest_block;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001475 if (page_no == 0)
Hugh Dickinse2244ec2005-09-03 15:54:32 -07001476 page_no = 1; /* force Empty message */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001477 sis->max = page_no;
Hugh Dickinse2244ec2005-09-03 15:54:32 -07001478 sis->pages = page_no - 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001479 sis->highest_bit = page_no - 1;
Hugh Dickins9625a5f2009-12-14 17:58:42 -08001480out:
1481 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001482bad_bmap:
1483 printk(KERN_ERR "swapon: swapfile has holes\n");
1484 ret = -EINVAL;
Hugh Dickins9625a5f2009-12-14 17:58:42 -08001485 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001486}
1487
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07001488static void enable_swap_info(struct swap_info_struct *p, int prio,
1489 unsigned char *swap_map)
1490{
1491 int i, prev;
1492
1493 spin_lock(&swap_lock);
1494 if (prio >= 0)
1495 p->prio = prio;
1496 else
1497 p->prio = --least_priority;
1498 p->swap_map = swap_map;
1499 p->flags |= SWP_WRITEOK;
1500 nr_swap_pages += p->pages;
1501 total_swap_pages += p->pages;
1502
1503 /* insert swap space into swap_list: */
1504 prev = -1;
1505 for (i = swap_list.head; i >= 0; i = swap_info[i]->next) {
1506 if (p->prio >= swap_info[i]->prio)
1507 break;
1508 prev = i;
1509 }
1510 p->next = i;
1511 if (prev < 0)
1512 swap_list.head = swap_list.next = p->type;
1513 else
1514 swap_info[prev]->next = p->type;
1515 spin_unlock(&swap_lock);
1516}
1517
Heiko Carstensc4ea37c2009-01-14 14:14:28 +01001518SYSCALL_DEFINE1(swapoff, const char __user *, specialfile)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001519{
Hugh Dickins73c34b62009-12-14 17:58:43 -08001520 struct swap_info_struct *p = NULL;
Hugh Dickins8d69aae2009-12-14 17:58:45 -08001521 unsigned char *swap_map;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001522 struct file *swap_file, *victim;
1523 struct address_space *mapping;
1524 struct inode *inode;
Hugh Dickins73c34b62009-12-14 17:58:43 -08001525 char *pathname;
David Rientjes72788c32011-05-24 17:11:40 -07001526 int oom_score_adj;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001527 int i, type, prev;
1528 int err;
Hugh Dickins886bb7e2009-01-06 14:39:48 -08001529
Linus Torvalds1da177e2005-04-16 15:20:36 -07001530 if (!capable(CAP_SYS_ADMIN))
1531 return -EPERM;
1532
Al Viro191c5422012-02-13 03:58:52 +00001533 BUG_ON(!current->mm);
1534
Linus Torvalds1da177e2005-04-16 15:20:36 -07001535 pathname = getname(specialfile);
1536 err = PTR_ERR(pathname);
1537 if (IS_ERR(pathname))
1538 goto out;
1539
1540 victim = filp_open(pathname, O_RDWR|O_LARGEFILE, 0);
1541 putname(pathname);
1542 err = PTR_ERR(victim);
1543 if (IS_ERR(victim))
1544 goto out;
1545
1546 mapping = victim->f_mapping;
1547 prev = -1;
Hugh Dickins5d337b92005-09-03 15:54:41 -07001548 spin_lock(&swap_lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001549 for (type = swap_list.head; type >= 0; type = swap_info[type]->next) {
1550 p = swap_info[type];
Hugh Dickins22c6f8f2009-01-06 14:39:48 -08001551 if (p->flags & SWP_WRITEOK) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001552 if (p->swap_file->f_mapping == mapping)
1553 break;
1554 }
1555 prev = type;
1556 }
1557 if (type < 0) {
1558 err = -EINVAL;
Hugh Dickins5d337b92005-09-03 15:54:41 -07001559 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001560 goto out_dput;
1561 }
Al Viro191c5422012-02-13 03:58:52 +00001562 if (!security_vm_enough_memory_mm(current->mm, p->pages))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001563 vm_unacct_memory(p->pages);
1564 else {
1565 err = -ENOMEM;
Hugh Dickins5d337b92005-09-03 15:54:41 -07001566 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001567 goto out_dput;
1568 }
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001569 if (prev < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001570 swap_list.head = p->next;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001571 else
1572 swap_info[prev]->next = p->next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001573 if (type == swap_list.next) {
1574 /* just pick something that's safe... */
1575 swap_list.next = swap_list.head;
1576 }
Hugh Dickins78ecba02008-07-23 21:28:23 -07001577 if (p->prio < 0) {
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001578 for (i = p->next; i >= 0; i = swap_info[i]->next)
1579 swap_info[i]->prio = p->prio--;
Hugh Dickins78ecba02008-07-23 21:28:23 -07001580 least_priority++;
1581 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001582 nr_swap_pages -= p->pages;
1583 total_swap_pages -= p->pages;
1584 p->flags &= ~SWP_WRITEOK;
Hugh Dickins5d337b92005-09-03 15:54:41 -07001585 spin_unlock(&swap_lock);
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -07001586
David Rientjes72788c32011-05-24 17:11:40 -07001587 oom_score_adj = test_set_oom_score_adj(OOM_SCORE_ADJ_MAX);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001588 err = try_to_unuse(type);
David Rientjes43362a42011-10-31 17:07:18 -07001589 compare_swap_oom_score_adj(OOM_SCORE_ADJ_MAX, oom_score_adj);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001590
Linus Torvalds1da177e2005-04-16 15:20:36 -07001591 if (err) {
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07001592 /*
1593 * reading p->prio and p->swap_map outside the lock is
1594 * safe here because only sys_swapon and sys_swapoff
1595 * change them, and there can be no other sys_swapon or
1596 * sys_swapoff for this swap_info_struct at this point.
1597 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001598 /* re-insert swap space back into swap_list */
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07001599 enable_swap_info(p, p->prio, p->swap_map);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001600 goto out_dput;
1601 }
Hugh Dickins52b7efdb2005-09-03 15:54:39 -07001602
Hugh Dickins4cd3bb12005-09-03 15:54:33 -07001603 destroy_swap_extents(p);
Hugh Dickins570a335b2009-12-14 17:58:46 -08001604 if (p->flags & SWP_CONTINUED)
1605 free_swap_count_continuations(p);
1606
Ingo Molnarfc0abb12006-01-18 17:42:33 -08001607 mutex_lock(&swapon_mutex);
Hugh Dickins5d337b92005-09-03 15:54:41 -07001608 spin_lock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001609 drain_mmlist();
Hugh Dickins5d337b92005-09-03 15:54:41 -07001610
1611 /* wait for anyone still in scan_swap_map */
1612 p->highest_bit = 0; /* cuts scans short */
1613 while (p->flags >= SWP_SCANNING) {
1614 spin_unlock(&swap_lock);
Nishanth Aravamudan13e4b572005-09-10 00:27:25 -07001615 schedule_timeout_uninterruptible(1);
Hugh Dickins5d337b92005-09-03 15:54:41 -07001616 spin_lock(&swap_lock);
1617 }
1618
Linus Torvalds1da177e2005-04-16 15:20:36 -07001619 swap_file = p->swap_file;
1620 p->swap_file = NULL;
1621 p->max = 0;
1622 swap_map = p->swap_map;
1623 p->swap_map = NULL;
1624 p->flags = 0;
Hugh Dickins5d337b92005-09-03 15:54:41 -07001625 spin_unlock(&swap_lock);
Ingo Molnarfc0abb12006-01-18 17:42:33 -08001626 mutex_unlock(&swapon_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001627 vfree(swap_map);
KAMEZAWA Hiroyuki27a7faa2009-01-07 18:07:58 -08001628 /* Destroy swap account informatin */
1629 swap_cgroup_swapoff(type);
1630
Linus Torvalds1da177e2005-04-16 15:20:36 -07001631 inode = mapping->host;
1632 if (S_ISBLK(inode->i_mode)) {
1633 struct block_device *bdev = I_BDEV(inode);
1634 set_blocksize(bdev, p->old_block_size);
Tejun Heoe525fd82010-11-13 11:55:17 +01001635 blkdev_put(bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001636 } else {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001637 mutex_lock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001638 inode->i_flags &= ~S_SWAPFILE;
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001639 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001640 }
1641 filp_close(swap_file, NULL);
1642 err = 0;
Kay Sievers66d7dd52010-10-26 14:22:06 -07001643 atomic_inc(&proc_poll_event);
1644 wake_up_interruptible(&proc_poll_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001645
1646out_dput:
1647 filp_close(victim, NULL);
1648out:
1649 return err;
1650}
1651
1652#ifdef CONFIG_PROC_FS
Kay Sievers66d7dd52010-10-26 14:22:06 -07001653static unsigned swaps_poll(struct file *file, poll_table *wait)
1654{
Kay Sieversf1514632011-07-12 20:48:39 +02001655 struct seq_file *seq = file->private_data;
Kay Sievers66d7dd52010-10-26 14:22:06 -07001656
1657 poll_wait(file, &proc_poll_wait, wait);
1658
Kay Sieversf1514632011-07-12 20:48:39 +02001659 if (seq->poll_event != atomic_read(&proc_poll_event)) {
1660 seq->poll_event = atomic_read(&proc_poll_event);
Kay Sievers66d7dd52010-10-26 14:22:06 -07001661 return POLLIN | POLLRDNORM | POLLERR | POLLPRI;
1662 }
1663
1664 return POLLIN | POLLRDNORM;
1665}
1666
Linus Torvalds1da177e2005-04-16 15:20:36 -07001667/* iterator */
1668static void *swap_start(struct seq_file *swap, loff_t *pos)
1669{
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001670 struct swap_info_struct *si;
1671 int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001672 loff_t l = *pos;
1673
Ingo Molnarfc0abb12006-01-18 17:42:33 -08001674 mutex_lock(&swapon_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001675
Suleiman Souhlal881e4aa2006-12-06 20:32:28 -08001676 if (!l)
1677 return SEQ_START_TOKEN;
1678
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001679 for (type = 0; type < nr_swapfiles; type++) {
1680 smp_rmb(); /* read nr_swapfiles before swap_info[type] */
1681 si = swap_info[type];
1682 if (!(si->flags & SWP_USED) || !si->swap_map)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001683 continue;
Suleiman Souhlal881e4aa2006-12-06 20:32:28 -08001684 if (!--l)
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001685 return si;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001686 }
1687
1688 return NULL;
1689}
1690
1691static void *swap_next(struct seq_file *swap, void *v, loff_t *pos)
1692{
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001693 struct swap_info_struct *si = v;
1694 int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001695
Suleiman Souhlal881e4aa2006-12-06 20:32:28 -08001696 if (v == SEQ_START_TOKEN)
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001697 type = 0;
1698 else
1699 type = si->type + 1;
Suleiman Souhlal881e4aa2006-12-06 20:32:28 -08001700
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001701 for (; type < nr_swapfiles; type++) {
1702 smp_rmb(); /* read nr_swapfiles before swap_info[type] */
1703 si = swap_info[type];
1704 if (!(si->flags & SWP_USED) || !si->swap_map)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001705 continue;
1706 ++*pos;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001707 return si;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001708 }
1709
1710 return NULL;
1711}
1712
1713static void swap_stop(struct seq_file *swap, void *v)
1714{
Ingo Molnarfc0abb12006-01-18 17:42:33 -08001715 mutex_unlock(&swapon_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001716}
1717
1718static int swap_show(struct seq_file *swap, void *v)
1719{
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001720 struct swap_info_struct *si = v;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001721 struct file *file;
1722 int len;
1723
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001724 if (si == SEQ_START_TOKEN) {
Suleiman Souhlal881e4aa2006-12-06 20:32:28 -08001725 seq_puts(swap,"Filename\t\t\t\tType\t\tSize\tUsed\tPriority\n");
1726 return 0;
1727 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001728
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001729 file = si->swap_file;
Jan Blunckc32c2f62008-02-14 19:38:43 -08001730 len = seq_path(swap, &file->f_path, " \t\n\\");
Hugh Dickins6eb396d2005-09-03 15:54:35 -07001731 seq_printf(swap, "%*s%s\t%u\t%u\t%d\n",
Hugh Dickins886bb7e2009-01-06 14:39:48 -08001732 len < 40 ? 40 - len : 1, " ",
1733 S_ISBLK(file->f_path.dentry->d_inode->i_mode) ?
Linus Torvalds1da177e2005-04-16 15:20:36 -07001734 "partition" : "file\t",
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001735 si->pages << (PAGE_SHIFT - 10),
1736 si->inuse_pages << (PAGE_SHIFT - 10),
1737 si->prio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001738 return 0;
1739}
1740
Helge Deller15ad7cd2006-12-06 20:40:36 -08001741static const struct seq_operations swaps_op = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001742 .start = swap_start,
1743 .next = swap_next,
1744 .stop = swap_stop,
1745 .show = swap_show
1746};
1747
1748static int swaps_open(struct inode *inode, struct file *file)
1749{
Kay Sieversf1514632011-07-12 20:48:39 +02001750 struct seq_file *seq;
Kay Sievers66d7dd52010-10-26 14:22:06 -07001751 int ret;
1752
Kay Sievers66d7dd52010-10-26 14:22:06 -07001753 ret = seq_open(file, &swaps_op);
Kay Sieversf1514632011-07-12 20:48:39 +02001754 if (ret)
Kay Sievers66d7dd52010-10-26 14:22:06 -07001755 return ret;
Kay Sievers66d7dd52010-10-26 14:22:06 -07001756
Kay Sieversf1514632011-07-12 20:48:39 +02001757 seq = file->private_data;
1758 seq->poll_event = atomic_read(&proc_poll_event);
1759 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001760}
1761
Helge Deller15ad7cd2006-12-06 20:40:36 -08001762static const struct file_operations proc_swaps_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001763 .open = swaps_open,
1764 .read = seq_read,
1765 .llseek = seq_lseek,
1766 .release = seq_release,
Kay Sievers66d7dd52010-10-26 14:22:06 -07001767 .poll = swaps_poll,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001768};
1769
1770static int __init procswaps_init(void)
1771{
Denis V. Lunev3d71f862008-04-29 01:02:13 -07001772 proc_create("swaps", 0, NULL, &proc_swaps_operations);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001773 return 0;
1774}
1775__initcall(procswaps_init);
1776#endif /* CONFIG_PROC_FS */
1777
Jan Beulich17963162008-12-16 11:35:24 +00001778#ifdef MAX_SWAPFILES_CHECK
1779static int __init max_swapfiles_check(void)
1780{
1781 MAX_SWAPFILES_CHECK();
1782 return 0;
1783}
1784late_initcall(max_swapfiles_check);
1785#endif
1786
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07001787static struct swap_info_struct *alloc_swap_info(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001788{
Hugh Dickins73c34b62009-12-14 17:58:43 -08001789 struct swap_info_struct *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001790 unsigned int type;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001791
1792 p = kzalloc(sizeof(*p), GFP_KERNEL);
1793 if (!p)
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07001794 return ERR_PTR(-ENOMEM);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001795
Hugh Dickins5d337b92005-09-03 15:54:41 -07001796 spin_lock(&swap_lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001797 for (type = 0; type < nr_swapfiles; type++) {
1798 if (!(swap_info[type]->flags & SWP_USED))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001799 break;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001800 }
Christoph Lameter06972122006-06-23 02:03:35 -07001801 if (type >= MAX_SWAPFILES) {
Hugh Dickins5d337b92005-09-03 15:54:41 -07001802 spin_unlock(&swap_lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001803 kfree(p);
Cesar Eduardo Barros730c0582011-03-22 16:33:19 -07001804 return ERR_PTR(-EPERM);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001805 }
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001806 if (type >= nr_swapfiles) {
1807 p->type = type;
1808 swap_info[type] = p;
1809 /*
1810 * Write swap_info[type] before nr_swapfiles, in case a
1811 * racing procfs swap_start() or swap_next() is reading them.
1812 * (We never shrink nr_swapfiles, we never free this entry.)
1813 */
1814 smp_wmb();
1815 nr_swapfiles++;
1816 } else {
1817 kfree(p);
1818 p = swap_info[type];
1819 /*
1820 * Do not memset this entry: a racing procfs swap_next()
1821 * would be relying on p->type to remain valid.
1822 */
1823 }
Hugh Dickins9625a5f2009-12-14 17:58:42 -08001824 INIT_LIST_HEAD(&p->first_swap_extent.list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001825 p->flags = SWP_USED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001826 p->next = -1;
Hugh Dickins5d337b92005-09-03 15:54:41 -07001827 spin_unlock(&swap_lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08001828
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07001829 return p;
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07001830}
1831
Cesar Eduardo Barros4d0e1e12011-03-22 16:33:26 -07001832static int claim_swapfile(struct swap_info_struct *p, struct inode *inode)
1833{
1834 int error;
1835
1836 if (S_ISBLK(inode->i_mode)) {
1837 p->bdev = bdgrab(I_BDEV(inode));
1838 error = blkdev_get(p->bdev,
1839 FMODE_READ | FMODE_WRITE | FMODE_EXCL,
1840 sys_swapon);
1841 if (error < 0) {
1842 p->bdev = NULL;
Cesar Eduardo Barros87ade722011-03-22 16:33:27 -07001843 return -EINVAL;
Cesar Eduardo Barros4d0e1e12011-03-22 16:33:26 -07001844 }
1845 p->old_block_size = block_size(p->bdev);
1846 error = set_blocksize(p->bdev, PAGE_SIZE);
1847 if (error < 0)
Cesar Eduardo Barros87ade722011-03-22 16:33:27 -07001848 return error;
Cesar Eduardo Barros4d0e1e12011-03-22 16:33:26 -07001849 p->flags |= SWP_BLKDEV;
1850 } else if (S_ISREG(inode->i_mode)) {
1851 p->bdev = inode->i_sb->s_bdev;
1852 mutex_lock(&inode->i_mutex);
Cesar Eduardo Barros87ade722011-03-22 16:33:27 -07001853 if (IS_SWAPFILE(inode))
1854 return -EBUSY;
1855 } else
1856 return -EINVAL;
Cesar Eduardo Barros4d0e1e12011-03-22 16:33:26 -07001857
1858 return 0;
Cesar Eduardo Barros4d0e1e12011-03-22 16:33:26 -07001859}
1860
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07001861static unsigned long read_swap_header(struct swap_info_struct *p,
1862 union swap_header *swap_header,
1863 struct inode *inode)
1864{
1865 int i;
1866 unsigned long maxpages;
1867 unsigned long swapfilepages;
1868
1869 if (memcmp("SWAPSPACE2", swap_header->magic.magic, 10)) {
1870 printk(KERN_ERR "Unable to find swap-space signature\n");
Cesar Eduardo Barros38719022011-03-22 16:33:30 -07001871 return 0;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07001872 }
1873
1874 /* swap partition endianess hack... */
1875 if (swab32(swap_header->info.version) == 1) {
1876 swab32s(&swap_header->info.version);
1877 swab32s(&swap_header->info.last_page);
1878 swab32s(&swap_header->info.nr_badpages);
1879 for (i = 0; i < swap_header->info.nr_badpages; i++)
1880 swab32s(&swap_header->info.badpages[i]);
1881 }
1882 /* Check the swap header's sub-version */
1883 if (swap_header->info.version != 1) {
1884 printk(KERN_WARNING
1885 "Unable to handle swap header version %d\n",
1886 swap_header->info.version);
Cesar Eduardo Barros38719022011-03-22 16:33:30 -07001887 return 0;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07001888 }
1889
1890 p->lowest_bit = 1;
1891 p->cluster_next = 1;
1892 p->cluster_nr = 0;
1893
1894 /*
1895 * Find out how many pages are allowed for a single swap
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001896 * device. There are three limiting factors: 1) the number
1897 * of bits for the swap offset in the swp_entry_t type, and
1898 * 2) the number of bits in the swap pte as defined by the
1899 * the different architectures, and 3) the number of free bits
1900 * in an exceptional radix_tree entry. In order to find the
1901 * largest possible bit mask, a swap entry with swap type 0
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07001902 * and swap offset ~0UL is created, encoded to a swap pte,
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001903 * decoded to a swp_entry_t again, and finally the swap
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07001904 * offset is extracted. This will mask all the bits from
1905 * the initial ~0UL mask that can't be encoded in either
1906 * the swp_entry_t or the architecture definition of a
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001907 * swap pte. Then the same is done for a radix_tree entry.
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07001908 */
1909 maxpages = swp_offset(pte_to_swp_entry(
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001910 swp_entry_to_pte(swp_entry(0, ~0UL))));
1911 maxpages = swp_offset(radix_to_swp_entry(
1912 swp_to_radix_entry(swp_entry(0, maxpages)))) + 1;
1913
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07001914 if (maxpages > swap_header->info.last_page) {
1915 maxpages = swap_header->info.last_page + 1;
1916 /* p->max is an unsigned int: don't overflow it */
1917 if ((unsigned int)maxpages == 0)
1918 maxpages = UINT_MAX;
1919 }
1920 p->highest_bit = maxpages - 1;
1921
1922 if (!maxpages)
Cesar Eduardo Barros38719022011-03-22 16:33:30 -07001923 return 0;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07001924 swapfilepages = i_size_read(inode) >> PAGE_SHIFT;
1925 if (swapfilepages && maxpages > swapfilepages) {
1926 printk(KERN_WARNING
1927 "Swap area shorter than signature indicates\n");
Cesar Eduardo Barros38719022011-03-22 16:33:30 -07001928 return 0;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07001929 }
1930 if (swap_header->info.nr_badpages && S_ISREG(inode->i_mode))
Cesar Eduardo Barros38719022011-03-22 16:33:30 -07001931 return 0;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07001932 if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES)
Cesar Eduardo Barros38719022011-03-22 16:33:30 -07001933 return 0;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07001934
1935 return maxpages;
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07001936}
1937
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07001938static int setup_swap_map_and_extents(struct swap_info_struct *p,
1939 union swap_header *swap_header,
1940 unsigned char *swap_map,
1941 unsigned long maxpages,
1942 sector_t *span)
1943{
1944 int i;
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07001945 unsigned int nr_good_pages;
1946 int nr_extents;
1947
1948 nr_good_pages = maxpages - 1; /* omit header page */
1949
1950 for (i = 0; i < swap_header->info.nr_badpages; i++) {
1951 unsigned int page_nr = swap_header->info.badpages[i];
Cesar Eduardo Barrosbdb8e3f2011-03-22 16:33:33 -07001952 if (page_nr == 0 || page_nr > swap_header->info.last_page)
1953 return -EINVAL;
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07001954 if (page_nr < maxpages) {
1955 swap_map[page_nr] = SWAP_MAP_BAD;
1956 nr_good_pages--;
1957 }
1958 }
1959
1960 if (nr_good_pages) {
1961 swap_map[0] = SWAP_MAP_BAD;
1962 p->max = maxpages;
1963 p->pages = nr_good_pages;
1964 nr_extents = setup_swap_extents(p, span);
Cesar Eduardo Barrosbdb8e3f2011-03-22 16:33:33 -07001965 if (nr_extents < 0)
1966 return nr_extents;
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07001967 nr_good_pages = p->pages;
1968 }
1969 if (!nr_good_pages) {
1970 printk(KERN_WARNING "Empty swap-file\n");
Cesar Eduardo Barrosbdb8e3f2011-03-22 16:33:33 -07001971 return -EINVAL;
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07001972 }
1973
1974 return nr_extents;
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07001975}
1976
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07001977SYSCALL_DEFINE2(swapon, const char __user *, specialfile, int, swap_flags)
1978{
1979 struct swap_info_struct *p;
Cesar Eduardo Barros28b36bd2011-03-22 16:33:20 -07001980 char *name;
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07001981 struct file *swap_file = NULL;
1982 struct address_space *mapping;
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07001983 int i;
1984 int prio;
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07001985 int error;
1986 union swap_header *swap_header;
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07001987 int nr_extents;
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07001988 sector_t span;
1989 unsigned long maxpages;
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07001990 unsigned char *swap_map = NULL;
1991 struct page *page = NULL;
1992 struct inode *inode = NULL;
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07001993
Hugh Dickinsd15cab92012-03-28 14:42:42 -07001994 if (swap_flags & ~SWAP_FLAGS_VALID)
1995 return -EINVAL;
1996
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07001997 if (!capable(CAP_SYS_ADMIN))
1998 return -EPERM;
1999
2000 p = alloc_swap_info();
Cesar Eduardo Barros2542e5132011-03-22 16:33:18 -07002001 if (IS_ERR(p))
2002 return PTR_ERR(p);
Cesar Eduardo Barros53cbb242011-03-22 16:33:17 -07002003
Linus Torvalds1da177e2005-04-16 15:20:36 -07002004 name = getname(specialfile);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002005 if (IS_ERR(name)) {
Cesar Eduardo Barros7de7fb62011-03-22 16:33:22 -07002006 error = PTR_ERR(name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002007 name = NULL;
Cesar Eduardo Barrosbd690102011-03-22 16:33:25 -07002008 goto bad_swap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002009 }
2010 swap_file = filp_open(name, O_RDWR|O_LARGEFILE, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002011 if (IS_ERR(swap_file)) {
Cesar Eduardo Barros7de7fb62011-03-22 16:33:22 -07002012 error = PTR_ERR(swap_file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002013 swap_file = NULL;
Cesar Eduardo Barrosbd690102011-03-22 16:33:25 -07002014 goto bad_swap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002015 }
2016
2017 p->swap_file = swap_file;
2018 mapping = swap_file->f_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002019
Linus Torvalds1da177e2005-04-16 15:20:36 -07002020 for (i = 0; i < nr_swapfiles; i++) {
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002021 struct swap_info_struct *q = swap_info[i];
Linus Torvalds1da177e2005-04-16 15:20:36 -07002022
Cesar Eduardo Barrose8e6c2e2011-03-22 16:33:16 -07002023 if (q == p || !q->swap_file)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002024 continue;
Cesar Eduardo Barros7de7fb62011-03-22 16:33:22 -07002025 if (mapping == q->swap_file->f_mapping) {
2026 error = -EBUSY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002027 goto bad_swap;
Cesar Eduardo Barros7de7fb62011-03-22 16:33:22 -07002028 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002029 }
2030
Cesar Eduardo Barros21307812011-03-22 23:03:13 -03002031 inode = mapping->host;
2032 /* If S_ISREG(inode->i_mode) will do mutex_lock(&inode->i_mutex); */
Cesar Eduardo Barros4d0e1e12011-03-22 16:33:26 -07002033 error = claim_swapfile(p, inode);
2034 if (unlikely(error))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002035 goto bad_swap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002036
Linus Torvalds1da177e2005-04-16 15:20:36 -07002037 /*
2038 * Read the swap header.
2039 */
2040 if (!mapping->a_ops->readpage) {
2041 error = -EINVAL;
2042 goto bad_swap;
2043 }
Pekka Enberg090d2b12006-06-23 02:05:08 -07002044 page = read_mapping_page(mapping, 0, swap_file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002045 if (IS_ERR(page)) {
2046 error = PTR_ERR(page);
2047 goto bad_swap;
2048 }
Hugh Dickins81e33972009-01-06 14:39:49 -08002049 swap_header = kmap(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002050
Cesar Eduardo Barrosca8bd382011-03-22 16:33:29 -07002051 maxpages = read_swap_header(p, swap_header, inode);
2052 if (unlikely(!maxpages)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002053 error = -EINVAL;
2054 goto bad_swap;
2055 }
Hugh Dickins886bb7e2009-01-06 14:39:48 -08002056
Hugh Dickins81e33972009-01-06 14:39:49 -08002057 /* OK, set up the swap map and apply the bad block list */
Cesar Eduardo Barros803d0c832011-03-22 16:33:14 -07002058 swap_map = vzalloc(maxpages);
Hugh Dickins81e33972009-01-06 14:39:49 -08002059 if (!swap_map) {
2060 error = -ENOMEM;
2061 goto bad_swap;
2062 }
2063
Cesar Eduardo Barros1421ef32011-03-22 16:33:31 -07002064 error = swap_cgroup_swapon(p->type, maxpages);
2065 if (error)
2066 goto bad_swap;
2067
Cesar Eduardo Barros915d4d72011-03-22 16:33:32 -07002068 nr_extents = setup_swap_map_and_extents(p, swap_header, swap_map,
2069 maxpages, &span);
2070 if (unlikely(nr_extents < 0)) {
2071 error = nr_extents;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002072 goto bad_swap;
2073 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002074
Suresh Jayaraman3bd0f0c2009-09-30 10:53:48 +02002075 if (p->bdev) {
2076 if (blk_queue_nonrot(bdev_get_queue(p->bdev))) {
2077 p->flags |= SWP_SOLIDSTATE;
2078 p->cluster_next = 1 + (random32() % p->highest_bit);
2079 }
Shaohua Li052b1982012-03-21 16:34:17 -07002080 if ((swap_flags & SWAP_FLAG_DISCARD) && discard_swap(p) == 0)
Suresh Jayaraman3bd0f0c2009-09-30 10:53:48 +02002081 p->flags |= SWP_DISCARDABLE;
Hugh Dickins20137a42009-01-06 14:39:54 -08002082 }
Hugh Dickins6a6ba832009-01-06 14:39:51 -08002083
Ingo Molnarfc0abb12006-01-18 17:42:33 -08002084 mutex_lock(&swapon_mutex);
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07002085 prio = -1;
Hugh Dickins78ecba02008-07-23 21:28:23 -07002086 if (swap_flags & SWAP_FLAG_PREFER)
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07002087 prio =
Hugh Dickins78ecba02008-07-23 21:28:23 -07002088 (swap_flags & SWAP_FLAG_PRIO_MASK) >> SWAP_FLAG_PRIO_SHIFT;
Cesar Eduardo Barros40531542011-03-22 16:33:37 -07002089 enable_swap_info(p, prio, swap_map);
Cesar Eduardo Barrosc69dbfb2011-03-22 16:33:35 -07002090
2091 printk(KERN_INFO "Adding %uk swap on %s. "
2092 "Priority:%d extents:%d across:%lluk %s%s\n",
2093 p->pages<<(PAGE_SHIFT-10), name, p->prio,
2094 nr_extents, (unsigned long long)span<<(PAGE_SHIFT-10),
2095 (p->flags & SWP_SOLIDSTATE) ? "SS" : "",
2096 (p->flags & SWP_DISCARDABLE) ? "D" : "");
2097
Ingo Molnarfc0abb12006-01-18 17:42:33 -08002098 mutex_unlock(&swapon_mutex);
Kay Sievers66d7dd52010-10-26 14:22:06 -07002099 atomic_inc(&proc_poll_event);
2100 wake_up_interruptible(&proc_poll_wait);
2101
Cesar Eduardo Barros9b01c352011-03-22 16:33:24 -07002102 if (S_ISREG(inode->i_mode))
2103 inode->i_flags |= S_SWAPFILE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002104 error = 0;
2105 goto out;
2106bad_swap:
Cesar Eduardo Barrosbd690102011-03-22 16:33:25 -07002107 if (inode && S_ISBLK(inode->i_mode) && p->bdev) {
Cesar Eduardo Barrosf2090d22011-03-22 16:33:23 -07002108 set_blocksize(p->bdev, p->old_block_size);
2109 blkdev_put(p->bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002110 }
Hugh Dickins4cd3bb12005-09-03 15:54:33 -07002111 destroy_swap_extents(p);
Cesar Eduardo Barrose8e6c2e2011-03-22 16:33:16 -07002112 swap_cgroup_swapoff(p->type);
Hugh Dickins5d337b92005-09-03 15:54:41 -07002113 spin_lock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002114 p->swap_file = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002115 p->flags = 0;
Hugh Dickins5d337b92005-09-03 15:54:41 -07002116 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002117 vfree(swap_map);
Mel Gorman52c50562011-03-22 16:30:08 -07002118 if (swap_file) {
Cesar Eduardo Barros21307812011-03-22 23:03:13 -03002119 if (inode && S_ISREG(inode->i_mode)) {
Mel Gorman52c50562011-03-22 16:30:08 -07002120 mutex_unlock(&inode->i_mutex);
Cesar Eduardo Barros21307812011-03-22 23:03:13 -03002121 inode = NULL;
2122 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002123 filp_close(swap_file, NULL);
Mel Gorman52c50562011-03-22 16:30:08 -07002124 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002125out:
2126 if (page && !IS_ERR(page)) {
2127 kunmap(page);
2128 page_cache_release(page);
2129 }
2130 if (name)
2131 putname(name);
Cesar Eduardo Barros9b01c352011-03-22 16:33:24 -07002132 if (inode && S_ISREG(inode->i_mode))
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002133 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002134 return error;
2135}
2136
2137void si_swapinfo(struct sysinfo *val)
2138{
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002139 unsigned int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002140 unsigned long nr_to_be_unused = 0;
2141
Hugh Dickins5d337b92005-09-03 15:54:41 -07002142 spin_lock(&swap_lock);
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002143 for (type = 0; type < nr_swapfiles; type++) {
2144 struct swap_info_struct *si = swap_info[type];
2145
2146 if ((si->flags & SWP_USED) && !(si->flags & SWP_WRITEOK))
2147 nr_to_be_unused += si->inuse_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002148 }
2149 val->freeswap = nr_swap_pages + nr_to_be_unused;
2150 val->totalswap = total_swap_pages + nr_to_be_unused;
Hugh Dickins5d337b92005-09-03 15:54:41 -07002151 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002152}
2153
2154/*
2155 * Verify that a swap entry is valid and increment its swap map count.
2156 *
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002157 * Returns error code in following case.
2158 * - success -> 0
2159 * - swp_entry is invalid -> EINVAL
2160 * - swp_entry is migration entry -> EINVAL
2161 * - swap-cache reference is requested but there is already one. -> EEXIST
2162 * - swap-cache reference is requested but the entry is not used. -> ENOENT
Hugh Dickins570a335b2009-12-14 17:58:46 -08002163 * - swap-mapped reference requested but needs continued swap count. -> ENOMEM
Linus Torvalds1da177e2005-04-16 15:20:36 -07002164 */
Hugh Dickins8d69aae2009-12-14 17:58:45 -08002165static int __swap_duplicate(swp_entry_t entry, unsigned char usage)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002166{
Hugh Dickins73c34b62009-12-14 17:58:43 -08002167 struct swap_info_struct *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002168 unsigned long offset, type;
Hugh Dickins8d69aae2009-12-14 17:58:45 -08002169 unsigned char count;
2170 unsigned char has_cache;
Hugh Dickins253d5532009-12-14 17:58:44 -08002171 int err = -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002172
Andi Kleena7420aa2009-09-16 11:50:05 +02002173 if (non_swap_entry(entry))
Hugh Dickins253d5532009-12-14 17:58:44 -08002174 goto out;
Christoph Lameter06972122006-06-23 02:03:35 -07002175
Linus Torvalds1da177e2005-04-16 15:20:36 -07002176 type = swp_type(entry);
2177 if (type >= nr_swapfiles)
2178 goto bad_file;
Hugh Dickinsefa90a92009-12-14 17:58:41 -08002179 p = swap_info[type];
Linus Torvalds1da177e2005-04-16 15:20:36 -07002180 offset = swp_offset(entry);
2181
Hugh Dickins5d337b92005-09-03 15:54:41 -07002182 spin_lock(&swap_lock);
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002183 if (unlikely(offset >= p->max))
2184 goto unlock_out;
2185
Hugh Dickins253d5532009-12-14 17:58:44 -08002186 count = p->swap_map[offset];
2187 has_cache = count & SWAP_HAS_CACHE;
2188 count &= ~SWAP_HAS_CACHE;
2189 err = 0;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002190
Hugh Dickins253d5532009-12-14 17:58:44 -08002191 if (usage == SWAP_HAS_CACHE) {
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002192
2193 /* set SWAP_HAS_CACHE if there is no cache and entry is used */
Hugh Dickins253d5532009-12-14 17:58:44 -08002194 if (!has_cache && count)
2195 has_cache = SWAP_HAS_CACHE;
2196 else if (has_cache) /* someone else added cache */
2197 err = -EEXIST;
2198 else /* no users remaining */
2199 err = -ENOENT;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002200
2201 } else if (count || has_cache) {
Hugh Dickins253d5532009-12-14 17:58:44 -08002202
Hugh Dickins570a335b2009-12-14 17:58:46 -08002203 if ((count & ~COUNT_CONTINUED) < SWAP_MAP_MAX)
2204 count += usage;
2205 else if ((count & ~COUNT_CONTINUED) > SWAP_MAP_MAX)
Hugh Dickins253d5532009-12-14 17:58:44 -08002206 err = -EINVAL;
Hugh Dickins570a335b2009-12-14 17:58:46 -08002207 else if (swap_count_continued(p, offset, count))
2208 count = COUNT_CONTINUED;
2209 else
2210 err = -ENOMEM;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002211 } else
Hugh Dickins253d5532009-12-14 17:58:44 -08002212 err = -ENOENT; /* unused swap entry */
2213
2214 p->swap_map[offset] = count | has_cache;
2215
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002216unlock_out:
Hugh Dickins5d337b92005-09-03 15:54:41 -07002217 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002218out:
Hugh Dickins253d5532009-12-14 17:58:44 -08002219 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002220
2221bad_file:
2222 printk(KERN_ERR "swap_dup: %s%08lx\n", Bad_file, entry.val);
2223 goto out;
2224}
Hugh Dickins253d5532009-12-14 17:58:44 -08002225
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002226/*
Hugh Dickinsaaa46862009-12-14 17:58:47 -08002227 * Help swapoff by noting that swap entry belongs to shmem/tmpfs
2228 * (in which case its reference count is never incremented).
2229 */
2230void swap_shmem_alloc(swp_entry_t entry)
2231{
2232 __swap_duplicate(entry, SWAP_MAP_SHMEM);
2233}
2234
2235/*
Hugh Dickins08259d52010-03-05 13:42:25 -08002236 * Increase reference count of swap entry by 1.
2237 * Returns 0 for success, or -ENOMEM if a swap_count_continuation is required
2238 * but could not be atomically allocated. Returns 0, just as if it succeeded,
2239 * if __swap_duplicate() fails for another reason (-EINVAL or -ENOENT), which
2240 * might occur if a page table entry has got corrupted.
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002241 */
Hugh Dickins570a335b2009-12-14 17:58:46 -08002242int swap_duplicate(swp_entry_t entry)
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002243{
Hugh Dickins570a335b2009-12-14 17:58:46 -08002244 int err = 0;
2245
2246 while (!err && __swap_duplicate(entry, 1) == -ENOMEM)
2247 err = add_swap_count_continuation(entry, GFP_ATOMIC);
2248 return err;
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002249}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002250
KAMEZAWA Hiroyukicb4b86b2009-06-16 15:32:52 -07002251/*
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002252 * @entry: swap entry for which we allocate swap cache.
2253 *
Hugh Dickins73c34b62009-12-14 17:58:43 -08002254 * Called when allocating swap cache for existing swap entry,
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -07002255 * This can return error codes. Returns 0 at success.
2256 * -EBUSY means there is a swap cache.
2257 * Note: return code is different from swap_duplicate().
KAMEZAWA Hiroyukicb4b86b2009-06-16 15:32:52 -07002258 */
2259int swapcache_prepare(swp_entry_t entry)
2260{
Hugh Dickins253d5532009-12-14 17:58:44 -08002261 return __swap_duplicate(entry, SWAP_HAS_CACHE);
KAMEZAWA Hiroyukicb4b86b2009-06-16 15:32:52 -07002262}
2263
Linus Torvalds1da177e2005-04-16 15:20:36 -07002264/*
Hugh Dickins570a335b2009-12-14 17:58:46 -08002265 * add_swap_count_continuation - called when a swap count is duplicated
2266 * beyond SWAP_MAP_MAX, it allocates a new page and links that to the entry's
2267 * page of the original vmalloc'ed swap_map, to hold the continuation count
2268 * (for that entry and for its neighbouring PAGE_SIZE swap entries). Called
2269 * again when count is duplicated beyond SWAP_MAP_MAX * SWAP_CONT_MAX, etc.
2270 *
2271 * These continuation pages are seldom referenced: the common paths all work
2272 * on the original swap_map, only referring to a continuation page when the
2273 * low "digit" of a count is incremented or decremented through SWAP_MAP_MAX.
2274 *
2275 * add_swap_count_continuation(, GFP_ATOMIC) can be called while holding
2276 * page table locks; if it fails, add_swap_count_continuation(, GFP_KERNEL)
2277 * can be called after dropping locks.
2278 */
2279int add_swap_count_continuation(swp_entry_t entry, gfp_t gfp_mask)
2280{
2281 struct swap_info_struct *si;
2282 struct page *head;
2283 struct page *page;
2284 struct page *list_page;
2285 pgoff_t offset;
2286 unsigned char count;
2287
2288 /*
2289 * When debugging, it's easier to use __GFP_ZERO here; but it's better
2290 * for latency not to zero a page while GFP_ATOMIC and holding locks.
2291 */
2292 page = alloc_page(gfp_mask | __GFP_HIGHMEM);
2293
2294 si = swap_info_get(entry);
2295 if (!si) {
2296 /*
2297 * An acceptable race has occurred since the failing
2298 * __swap_duplicate(): the swap entry has been freed,
2299 * perhaps even the whole swap_map cleared for swapoff.
2300 */
2301 goto outer;
2302 }
2303
2304 offset = swp_offset(entry);
2305 count = si->swap_map[offset] & ~SWAP_HAS_CACHE;
2306
2307 if ((count & ~COUNT_CONTINUED) != SWAP_MAP_MAX) {
2308 /*
2309 * The higher the swap count, the more likely it is that tasks
2310 * will race to add swap count continuation: we need to avoid
2311 * over-provisioning.
2312 */
2313 goto out;
2314 }
2315
2316 if (!page) {
2317 spin_unlock(&swap_lock);
2318 return -ENOMEM;
2319 }
2320
2321 /*
2322 * We are fortunate that although vmalloc_to_page uses pte_offset_map,
2323 * no architecture is using highmem pages for kernel pagetables: so it
2324 * will not corrupt the GFP_ATOMIC caller's atomic pagetable kmaps.
2325 */
2326 head = vmalloc_to_page(si->swap_map + offset);
2327 offset &= ~PAGE_MASK;
2328
2329 /*
2330 * Page allocation does not initialize the page's lru field,
2331 * but it does always reset its private field.
2332 */
2333 if (!page_private(head)) {
2334 BUG_ON(count & COUNT_CONTINUED);
2335 INIT_LIST_HEAD(&head->lru);
2336 set_page_private(head, SWP_CONTINUED);
2337 si->flags |= SWP_CONTINUED;
2338 }
2339
2340 list_for_each_entry(list_page, &head->lru, lru) {
2341 unsigned char *map;
2342
2343 /*
2344 * If the previous map said no continuation, but we've found
2345 * a continuation page, free our allocation and use this one.
2346 */
2347 if (!(count & COUNT_CONTINUED))
2348 goto out;
2349
Cong Wang9b04c5f2011-11-25 23:14:39 +08002350 map = kmap_atomic(list_page) + offset;
Hugh Dickins570a335b2009-12-14 17:58:46 -08002351 count = *map;
Cong Wang9b04c5f2011-11-25 23:14:39 +08002352 kunmap_atomic(map);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002353
2354 /*
2355 * If this continuation count now has some space in it,
2356 * free our allocation and use this one.
2357 */
2358 if ((count & ~COUNT_CONTINUED) != SWAP_CONT_MAX)
2359 goto out;
2360 }
2361
2362 list_add_tail(&page->lru, &head->lru);
2363 page = NULL; /* now it's attached, don't free it */
2364out:
2365 spin_unlock(&swap_lock);
2366outer:
2367 if (page)
2368 __free_page(page);
2369 return 0;
2370}
2371
2372/*
2373 * swap_count_continued - when the original swap_map count is incremented
2374 * from SWAP_MAP_MAX, check if there is already a continuation page to carry
2375 * into, carry if so, or else fail until a new continuation page is allocated;
2376 * when the original swap_map count is decremented from 0 with continuation,
2377 * borrow from the continuation and report whether it still holds more.
2378 * Called while __swap_duplicate() or swap_entry_free() holds swap_lock.
2379 */
2380static bool swap_count_continued(struct swap_info_struct *si,
2381 pgoff_t offset, unsigned char count)
2382{
2383 struct page *head;
2384 struct page *page;
2385 unsigned char *map;
2386
2387 head = vmalloc_to_page(si->swap_map + offset);
2388 if (page_private(head) != SWP_CONTINUED) {
2389 BUG_ON(count & COUNT_CONTINUED);
2390 return false; /* need to add count continuation */
2391 }
2392
2393 offset &= ~PAGE_MASK;
2394 page = list_entry(head->lru.next, struct page, lru);
Cong Wang9b04c5f2011-11-25 23:14:39 +08002395 map = kmap_atomic(page) + offset;
Hugh Dickins570a335b2009-12-14 17:58:46 -08002396
2397 if (count == SWAP_MAP_MAX) /* initial increment from swap_map */
2398 goto init_map; /* jump over SWAP_CONT_MAX checks */
2399
2400 if (count == (SWAP_MAP_MAX | COUNT_CONTINUED)) { /* incrementing */
2401 /*
2402 * Think of how you add 1 to 999
2403 */
2404 while (*map == (SWAP_CONT_MAX | COUNT_CONTINUED)) {
Cong Wang9b04c5f2011-11-25 23:14:39 +08002405 kunmap_atomic(map);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002406 page = list_entry(page->lru.next, struct page, lru);
2407 BUG_ON(page == head);
Cong Wang9b04c5f2011-11-25 23:14:39 +08002408 map = kmap_atomic(page) + offset;
Hugh Dickins570a335b2009-12-14 17:58:46 -08002409 }
2410 if (*map == SWAP_CONT_MAX) {
Cong Wang9b04c5f2011-11-25 23:14:39 +08002411 kunmap_atomic(map);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002412 page = list_entry(page->lru.next, struct page, lru);
2413 if (page == head)
2414 return false; /* add count continuation */
Cong Wang9b04c5f2011-11-25 23:14:39 +08002415 map = kmap_atomic(page) + offset;
Hugh Dickins570a335b2009-12-14 17:58:46 -08002416init_map: *map = 0; /* we didn't zero the page */
2417 }
2418 *map += 1;
Cong Wang9b04c5f2011-11-25 23:14:39 +08002419 kunmap_atomic(map);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002420 page = list_entry(page->lru.prev, struct page, lru);
2421 while (page != head) {
Cong Wang9b04c5f2011-11-25 23:14:39 +08002422 map = kmap_atomic(page) + offset;
Hugh Dickins570a335b2009-12-14 17:58:46 -08002423 *map = COUNT_CONTINUED;
Cong Wang9b04c5f2011-11-25 23:14:39 +08002424 kunmap_atomic(map);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002425 page = list_entry(page->lru.prev, struct page, lru);
2426 }
2427 return true; /* incremented */
2428
2429 } else { /* decrementing */
2430 /*
2431 * Think of how you subtract 1 from 1000
2432 */
2433 BUG_ON(count != COUNT_CONTINUED);
2434 while (*map == COUNT_CONTINUED) {
Cong Wang9b04c5f2011-11-25 23:14:39 +08002435 kunmap_atomic(map);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002436 page = list_entry(page->lru.next, struct page, lru);
2437 BUG_ON(page == head);
Cong Wang9b04c5f2011-11-25 23:14:39 +08002438 map = kmap_atomic(page) + offset;
Hugh Dickins570a335b2009-12-14 17:58:46 -08002439 }
2440 BUG_ON(*map == 0);
2441 *map -= 1;
2442 if (*map == 0)
2443 count = 0;
Cong Wang9b04c5f2011-11-25 23:14:39 +08002444 kunmap_atomic(map);
Hugh Dickins570a335b2009-12-14 17:58:46 -08002445 page = list_entry(page->lru.prev, struct page, lru);
2446 while (page != head) {
Cong Wang9b04c5f2011-11-25 23:14:39 +08002447 map = kmap_atomic(page) + offset;
Hugh Dickins570a335b2009-12-14 17:58:46 -08002448 *map = SWAP_CONT_MAX | count;
2449 count = COUNT_CONTINUED;
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 }
2453 return count == COUNT_CONTINUED;
2454 }
2455}
2456
2457/*
2458 * free_swap_count_continuations - swapoff free all the continuation pages
2459 * appended to the swap_map, after swap_map is quiesced, before vfree'ing it.
2460 */
2461static void free_swap_count_continuations(struct swap_info_struct *si)
2462{
2463 pgoff_t offset;
2464
2465 for (offset = 0; offset < si->max; offset += PAGE_SIZE) {
2466 struct page *head;
2467 head = vmalloc_to_page(si->swap_map + offset);
2468 if (page_private(head)) {
2469 struct list_head *this, *next;
2470 list_for_each_safe(this, next, &head->lru) {
2471 struct page *page;
2472 page = list_entry(this, struct page, lru);
2473 list_del(this);
2474 __free_page(page);
2475 }
2476 }
2477 }
2478}