]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
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 | ||
1da177e4 LT |
8 | #include <linux/mm.h> |
9 | #include <linux/hugetlb.h> | |
10 | #include <linux/mman.h> | |
11 | #include <linux/slab.h> | |
12 | #include <linux/kernel_stat.h> | |
13 | #include <linux/swap.h> | |
14 | #include <linux/vmalloc.h> | |
15 | #include <linux/pagemap.h> | |
16 | #include <linux/namei.h> | |
17 | #include <linux/shm.h> | |
18 | #include <linux/blkdev.h> | |
20137a49 | 19 | #include <linux/random.h> |
1da177e4 LT |
20 | #include <linux/writeback.h> |
21 | #include <linux/proc_fs.h> | |
22 | #include <linux/seq_file.h> | |
23 | #include <linux/init.h> | |
24 | #include <linux/module.h> | |
25 | #include <linux/rmap.h> | |
26 | #include <linux/security.h> | |
27 | #include <linux/backing-dev.h> | |
fc0abb14 | 28 | #include <linux/mutex.h> |
c59ede7b | 29 | #include <linux/capability.h> |
1da177e4 | 30 | #include <linux/syscalls.h> |
8a9f3ccd | 31 | #include <linux/memcontrol.h> |
1da177e4 LT |
32 | |
33 | #include <asm/pgtable.h> | |
34 | #include <asm/tlbflush.h> | |
35 | #include <linux/swapops.h> | |
27a7faa0 | 36 | #include <linux/page_cgroup.h> |
1da177e4 | 37 | |
7c363b8c AB |
38 | static DEFINE_SPINLOCK(swap_lock); |
39 | static unsigned int nr_swapfiles; | |
b962716b | 40 | long nr_swap_pages; |
1da177e4 LT |
41 | long total_swap_pages; |
42 | static int swap_overflow; | |
78ecba08 | 43 | static int least_priority; |
1da177e4 | 44 | |
1da177e4 LT |
45 | static const char Bad_file[] = "Bad swap file entry "; |
46 | static const char Unused_file[] = "Unused swap file entry "; | |
47 | static const char Bad_offset[] = "Bad swap offset entry "; | |
48 | static const char Unused_offset[] = "Unused swap offset entry "; | |
49 | ||
7c363b8c | 50 | static struct swap_list_t swap_list = {-1, -1}; |
1da177e4 | 51 | |
efa90a98 | 52 | static struct swap_info_struct *swap_info[MAX_SWAPFILES]; |
1da177e4 | 53 | |
fc0abb14 | 54 | static DEFINE_MUTEX(swapon_mutex); |
1da177e4 | 55 | |
355cfa73 KH |
56 | /* For reference count accounting in swap_map */ |
57 | /* enum for swap_map[] handling. internal use only */ | |
58 | enum { | |
59 | SWAP_MAP = 0, /* ops for reference from swap users */ | |
60 | SWAP_CACHE, /* ops for reference from swap cache */ | |
61 | }; | |
62 | ||
63 | static inline int swap_count(unsigned short ent) | |
64 | { | |
65 | return ent & SWAP_COUNT_MASK; | |
66 | } | |
67 | ||
68 | static inline bool swap_has_cache(unsigned short ent) | |
69 | { | |
70 | return !!(ent & SWAP_HAS_CACHE); | |
71 | } | |
72 | ||
73 | static inline unsigned short encode_swapmap(int count, bool has_cache) | |
74 | { | |
75 | unsigned short ret = count; | |
76 | ||
77 | if (has_cache) | |
78 | return SWAP_HAS_CACHE | ret; | |
79 | return ret; | |
80 | } | |
81 | ||
efa90a98 | 82 | /* returns 1 if swap entry is freed */ |
c9e44410 KH |
83 | static int |
84 | __try_to_reclaim_swap(struct swap_info_struct *si, unsigned long offset) | |
85 | { | |
efa90a98 | 86 | swp_entry_t entry = swp_entry(si->type, offset); |
c9e44410 KH |
87 | struct page *page; |
88 | int ret = 0; | |
89 | ||
90 | page = find_get_page(&swapper_space, entry.val); | |
91 | if (!page) | |
92 | return 0; | |
93 | /* | |
94 | * This function is called from scan_swap_map() and it's called | |
95 | * by vmscan.c at reclaiming pages. So, we hold a lock on a page, here. | |
96 | * We have to use trylock for avoiding deadlock. This is a special | |
97 | * case and you should use try_to_free_swap() with explicit lock_page() | |
98 | * in usual operations. | |
99 | */ | |
100 | if (trylock_page(page)) { | |
101 | ret = try_to_free_swap(page); | |
102 | unlock_page(page); | |
103 | } | |
104 | page_cache_release(page); | |
105 | return ret; | |
106 | } | |
355cfa73 | 107 | |
1da177e4 LT |
108 | /* |
109 | * We need this because the bdev->unplug_fn can sleep and we cannot | |
5d337b91 | 110 | * hold swap_lock while calling the unplug_fn. And swap_lock |
fc0abb14 | 111 | * cannot be turned into a mutex. |
1da177e4 LT |
112 | */ |
113 | static DECLARE_RWSEM(swap_unplug_sem); | |
114 | ||
1da177e4 LT |
115 | void swap_unplug_io_fn(struct backing_dev_info *unused_bdi, struct page *page) |
116 | { | |
117 | swp_entry_t entry; | |
118 | ||
119 | down_read(&swap_unplug_sem); | |
4c21e2f2 | 120 | entry.val = page_private(page); |
1da177e4 | 121 | if (PageSwapCache(page)) { |
efa90a98 | 122 | struct block_device *bdev = swap_info[swp_type(entry)]->bdev; |
1da177e4 LT |
123 | struct backing_dev_info *bdi; |
124 | ||
125 | /* | |
126 | * If the page is removed from swapcache from under us (with a | |
127 | * racy try_to_unuse/swapoff) we need an additional reference | |
4c21e2f2 HD |
128 | * count to avoid reading garbage from page_private(page) above. |
129 | * If the WARN_ON triggers during a swapoff it maybe the race | |
1da177e4 LT |
130 | * condition and it's harmless. However if it triggers without |
131 | * swapoff it signals a problem. | |
132 | */ | |
133 | WARN_ON(page_count(page) <= 1); | |
134 | ||
135 | bdi = bdev->bd_inode->i_mapping->backing_dev_info; | |
ba32311e | 136 | blk_run_backing_dev(bdi, page); |
1da177e4 LT |
137 | } |
138 | up_read(&swap_unplug_sem); | |
139 | } | |
140 | ||
6a6ba831 HD |
141 | /* |
142 | * swapon tell device that all the old swap contents can be discarded, | |
143 | * to allow the swap device to optimize its wear-levelling. | |
144 | */ | |
145 | static int discard_swap(struct swap_info_struct *si) | |
146 | { | |
147 | struct swap_extent *se; | |
9625a5f2 HD |
148 | sector_t start_block; |
149 | sector_t nr_blocks; | |
6a6ba831 HD |
150 | int err = 0; |
151 | ||
9625a5f2 HD |
152 | /* Do not discard the swap header page! */ |
153 | se = &si->first_swap_extent; | |
154 | start_block = (se->start_block + 1) << (PAGE_SHIFT - 9); | |
155 | nr_blocks = ((sector_t)se->nr_pages - 1) << (PAGE_SHIFT - 9); | |
156 | if (nr_blocks) { | |
157 | err = blkdev_issue_discard(si->bdev, start_block, | |
158 | nr_blocks, GFP_KERNEL, DISCARD_FL_BARRIER); | |
159 | if (err) | |
160 | return err; | |
161 | cond_resched(); | |
162 | } | |
6a6ba831 | 163 | |
9625a5f2 HD |
164 | list_for_each_entry(se, &si->first_swap_extent.list, list) { |
165 | start_block = se->start_block << (PAGE_SHIFT - 9); | |
166 | nr_blocks = (sector_t)se->nr_pages << (PAGE_SHIFT - 9); | |
6a6ba831 HD |
167 | |
168 | err = blkdev_issue_discard(si->bdev, start_block, | |
9625a5f2 | 169 | nr_blocks, GFP_KERNEL, DISCARD_FL_BARRIER); |
6a6ba831 HD |
170 | if (err) |
171 | break; | |
172 | ||
173 | cond_resched(); | |
174 | } | |
175 | return err; /* That will often be -EOPNOTSUPP */ | |
176 | } | |
177 | ||
7992fde7 HD |
178 | /* |
179 | * swap allocation tell device that a cluster of swap can now be discarded, | |
180 | * to allow the swap device to optimize its wear-levelling. | |
181 | */ | |
182 | static void discard_swap_cluster(struct swap_info_struct *si, | |
183 | pgoff_t start_page, pgoff_t nr_pages) | |
184 | { | |
185 | struct swap_extent *se = si->curr_swap_extent; | |
186 | int found_extent = 0; | |
187 | ||
188 | while (nr_pages) { | |
189 | struct list_head *lh; | |
190 | ||
191 | if (se->start_page <= start_page && | |
192 | start_page < se->start_page + se->nr_pages) { | |
193 | pgoff_t offset = start_page - se->start_page; | |
194 | sector_t start_block = se->start_block + offset; | |
858a2990 | 195 | sector_t nr_blocks = se->nr_pages - offset; |
7992fde7 HD |
196 | |
197 | if (nr_blocks > nr_pages) | |
198 | nr_blocks = nr_pages; | |
199 | start_page += nr_blocks; | |
200 | nr_pages -= nr_blocks; | |
201 | ||
202 | if (!found_extent++) | |
203 | si->curr_swap_extent = se; | |
204 | ||
205 | start_block <<= PAGE_SHIFT - 9; | |
206 | nr_blocks <<= PAGE_SHIFT - 9; | |
207 | if (blkdev_issue_discard(si->bdev, start_block, | |
9625a5f2 | 208 | nr_blocks, GFP_NOIO, DISCARD_FL_BARRIER)) |
7992fde7 HD |
209 | break; |
210 | } | |
211 | ||
212 | lh = se->list.next; | |
7992fde7 HD |
213 | se = list_entry(lh, struct swap_extent, list); |
214 | } | |
215 | } | |
216 | ||
217 | static int wait_for_discard(void *word) | |
218 | { | |
219 | schedule(); | |
220 | return 0; | |
221 | } | |
222 | ||
048c27fd HD |
223 | #define SWAPFILE_CLUSTER 256 |
224 | #define LATENCY_LIMIT 256 | |
225 | ||
355cfa73 KH |
226 | static inline unsigned long scan_swap_map(struct swap_info_struct *si, |
227 | int cache) | |
1da177e4 | 228 | { |
ebebbbe9 | 229 | unsigned long offset; |
c60aa176 | 230 | unsigned long scan_base; |
7992fde7 | 231 | unsigned long last_in_cluster = 0; |
048c27fd | 232 | int latency_ration = LATENCY_LIMIT; |
7992fde7 | 233 | int found_free_cluster = 0; |
7dfad418 | 234 | |
886bb7e9 | 235 | /* |
7dfad418 HD |
236 | * We try to cluster swap pages by allocating them sequentially |
237 | * in swap. Once we've allocated SWAPFILE_CLUSTER pages this | |
238 | * way, however, we resort to first-free allocation, starting | |
239 | * a new cluster. This prevents us from scattering swap pages | |
240 | * all over the entire swap partition, so that we reduce | |
241 | * overall disk seek times between swap pages. -- sct | |
242 | * But we do now try to find an empty cluster. -Andrea | |
c60aa176 | 243 | * And we let swap pages go all over an SSD partition. Hugh |
7dfad418 HD |
244 | */ |
245 | ||
52b7efdb | 246 | si->flags += SWP_SCANNING; |
c60aa176 | 247 | scan_base = offset = si->cluster_next; |
ebebbbe9 HD |
248 | |
249 | if (unlikely(!si->cluster_nr--)) { | |
250 | if (si->pages - si->inuse_pages < SWAPFILE_CLUSTER) { | |
251 | si->cluster_nr = SWAPFILE_CLUSTER - 1; | |
252 | goto checks; | |
253 | } | |
7992fde7 HD |
254 | if (si->flags & SWP_DISCARDABLE) { |
255 | /* | |
256 | * Start range check on racing allocations, in case | |
257 | * they overlap the cluster we eventually decide on | |
258 | * (we scan without swap_lock to allow preemption). | |
259 | * It's hardly conceivable that cluster_nr could be | |
260 | * wrapped during our scan, but don't depend on it. | |
261 | */ | |
262 | if (si->lowest_alloc) | |
263 | goto checks; | |
264 | si->lowest_alloc = si->max; | |
265 | si->highest_alloc = 0; | |
266 | } | |
5d337b91 | 267 | spin_unlock(&swap_lock); |
7dfad418 | 268 | |
c60aa176 HD |
269 | /* |
270 | * If seek is expensive, start searching for new cluster from | |
271 | * start of partition, to minimize the span of allocated swap. | |
272 | * But if seek is cheap, search from our current position, so | |
273 | * that swap is allocated from all over the partition: if the | |
274 | * Flash Translation Layer only remaps within limited zones, | |
275 | * we don't want to wear out the first zone too quickly. | |
276 | */ | |
277 | if (!(si->flags & SWP_SOLIDSTATE)) | |
278 | scan_base = offset = si->lowest_bit; | |
7dfad418 HD |
279 | last_in_cluster = offset + SWAPFILE_CLUSTER - 1; |
280 | ||
281 | /* Locate the first empty (unaligned) cluster */ | |
282 | for (; last_in_cluster <= si->highest_bit; offset++) { | |
1da177e4 | 283 | if (si->swap_map[offset]) |
7dfad418 HD |
284 | last_in_cluster = offset + SWAPFILE_CLUSTER; |
285 | else if (offset == last_in_cluster) { | |
5d337b91 | 286 | spin_lock(&swap_lock); |
ebebbbe9 HD |
287 | offset -= SWAPFILE_CLUSTER - 1; |
288 | si->cluster_next = offset; | |
289 | si->cluster_nr = SWAPFILE_CLUSTER - 1; | |
7992fde7 | 290 | found_free_cluster = 1; |
ebebbbe9 | 291 | goto checks; |
1da177e4 | 292 | } |
048c27fd HD |
293 | if (unlikely(--latency_ration < 0)) { |
294 | cond_resched(); | |
295 | latency_ration = LATENCY_LIMIT; | |
296 | } | |
7dfad418 | 297 | } |
ebebbbe9 HD |
298 | |
299 | offset = si->lowest_bit; | |
c60aa176 HD |
300 | last_in_cluster = offset + SWAPFILE_CLUSTER - 1; |
301 | ||
302 | /* Locate the first empty (unaligned) cluster */ | |
303 | for (; last_in_cluster < scan_base; offset++) { | |
304 | if (si->swap_map[offset]) | |
305 | last_in_cluster = offset + SWAPFILE_CLUSTER; | |
306 | else if (offset == last_in_cluster) { | |
307 | spin_lock(&swap_lock); | |
308 | offset -= SWAPFILE_CLUSTER - 1; | |
309 | si->cluster_next = offset; | |
310 | si->cluster_nr = SWAPFILE_CLUSTER - 1; | |
311 | found_free_cluster = 1; | |
312 | goto checks; | |
313 | } | |
314 | if (unlikely(--latency_ration < 0)) { | |
315 | cond_resched(); | |
316 | latency_ration = LATENCY_LIMIT; | |
317 | } | |
318 | } | |
319 | ||
320 | offset = scan_base; | |
5d337b91 | 321 | spin_lock(&swap_lock); |
ebebbbe9 | 322 | si->cluster_nr = SWAPFILE_CLUSTER - 1; |
7992fde7 | 323 | si->lowest_alloc = 0; |
1da177e4 | 324 | } |
7dfad418 | 325 | |
ebebbbe9 HD |
326 | checks: |
327 | if (!(si->flags & SWP_WRITEOK)) | |
52b7efdb | 328 | goto no_page; |
7dfad418 HD |
329 | if (!si->highest_bit) |
330 | goto no_page; | |
ebebbbe9 | 331 | if (offset > si->highest_bit) |
c60aa176 | 332 | scan_base = offset = si->lowest_bit; |
c9e44410 KH |
333 | |
334 | /* reuse swap entry of cache-only swap if not busy. */ | |
335 | if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) { | |
336 | int swap_was_freed; | |
337 | spin_unlock(&swap_lock); | |
338 | swap_was_freed = __try_to_reclaim_swap(si, offset); | |
339 | spin_lock(&swap_lock); | |
340 | /* entry was freed successfully, try to use this again */ | |
341 | if (swap_was_freed) | |
342 | goto checks; | |
343 | goto scan; /* check next one */ | |
344 | } | |
345 | ||
ebebbbe9 HD |
346 | if (si->swap_map[offset]) |
347 | goto scan; | |
348 | ||
349 | if (offset == si->lowest_bit) | |
350 | si->lowest_bit++; | |
351 | if (offset == si->highest_bit) | |
352 | si->highest_bit--; | |
353 | si->inuse_pages++; | |
354 | if (si->inuse_pages == si->pages) { | |
355 | si->lowest_bit = si->max; | |
356 | si->highest_bit = 0; | |
1da177e4 | 357 | } |
355cfa73 KH |
358 | if (cache == SWAP_CACHE) /* at usual swap-out via vmscan.c */ |
359 | si->swap_map[offset] = encode_swapmap(0, true); | |
360 | else /* at suspend */ | |
361 | si->swap_map[offset] = encode_swapmap(1, false); | |
ebebbbe9 HD |
362 | si->cluster_next = offset + 1; |
363 | si->flags -= SWP_SCANNING; | |
7992fde7 HD |
364 | |
365 | if (si->lowest_alloc) { | |
366 | /* | |
367 | * Only set when SWP_DISCARDABLE, and there's a scan | |
368 | * for a free cluster in progress or just completed. | |
369 | */ | |
370 | if (found_free_cluster) { | |
371 | /* | |
372 | * To optimize wear-levelling, discard the | |
373 | * old data of the cluster, taking care not to | |
374 | * discard any of its pages that have already | |
375 | * been allocated by racing tasks (offset has | |
376 | * already stepped over any at the beginning). | |
377 | */ | |
378 | if (offset < si->highest_alloc && | |
379 | si->lowest_alloc <= last_in_cluster) | |
380 | last_in_cluster = si->lowest_alloc - 1; | |
381 | si->flags |= SWP_DISCARDING; | |
382 | spin_unlock(&swap_lock); | |
383 | ||
384 | if (offset < last_in_cluster) | |
385 | discard_swap_cluster(si, offset, | |
386 | last_in_cluster - offset + 1); | |
387 | ||
388 | spin_lock(&swap_lock); | |
389 | si->lowest_alloc = 0; | |
390 | si->flags &= ~SWP_DISCARDING; | |
391 | ||
392 | smp_mb(); /* wake_up_bit advises this */ | |
393 | wake_up_bit(&si->flags, ilog2(SWP_DISCARDING)); | |
394 | ||
395 | } else if (si->flags & SWP_DISCARDING) { | |
396 | /* | |
397 | * Delay using pages allocated by racing tasks | |
398 | * until the whole discard has been issued. We | |
399 | * could defer that delay until swap_writepage, | |
400 | * but it's easier to keep this self-contained. | |
401 | */ | |
402 | spin_unlock(&swap_lock); | |
403 | wait_on_bit(&si->flags, ilog2(SWP_DISCARDING), | |
404 | wait_for_discard, TASK_UNINTERRUPTIBLE); | |
405 | spin_lock(&swap_lock); | |
406 | } else { | |
407 | /* | |
408 | * Note pages allocated by racing tasks while | |
409 | * scan for a free cluster is in progress, so | |
410 | * that its final discard can exclude them. | |
411 | */ | |
412 | if (offset < si->lowest_alloc) | |
413 | si->lowest_alloc = offset; | |
414 | if (offset > si->highest_alloc) | |
415 | si->highest_alloc = offset; | |
416 | } | |
417 | } | |
ebebbbe9 | 418 | return offset; |
7dfad418 | 419 | |
ebebbbe9 | 420 | scan: |
5d337b91 | 421 | spin_unlock(&swap_lock); |
7dfad418 | 422 | while (++offset <= si->highest_bit) { |
52b7efdb | 423 | if (!si->swap_map[offset]) { |
5d337b91 | 424 | spin_lock(&swap_lock); |
52b7efdb HD |
425 | goto checks; |
426 | } | |
c9e44410 KH |
427 | if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) { |
428 | spin_lock(&swap_lock); | |
429 | goto checks; | |
430 | } | |
048c27fd HD |
431 | if (unlikely(--latency_ration < 0)) { |
432 | cond_resched(); | |
433 | latency_ration = LATENCY_LIMIT; | |
434 | } | |
7dfad418 | 435 | } |
c60aa176 HD |
436 | offset = si->lowest_bit; |
437 | while (++offset < scan_base) { | |
438 | if (!si->swap_map[offset]) { | |
439 | spin_lock(&swap_lock); | |
440 | goto checks; | |
441 | } | |
c9e44410 KH |
442 | if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) { |
443 | spin_lock(&swap_lock); | |
444 | goto checks; | |
445 | } | |
c60aa176 HD |
446 | if (unlikely(--latency_ration < 0)) { |
447 | cond_resched(); | |
448 | latency_ration = LATENCY_LIMIT; | |
449 | } | |
450 | } | |
5d337b91 | 451 | spin_lock(&swap_lock); |
7dfad418 HD |
452 | |
453 | no_page: | |
52b7efdb | 454 | si->flags -= SWP_SCANNING; |
1da177e4 LT |
455 | return 0; |
456 | } | |
457 | ||
458 | swp_entry_t get_swap_page(void) | |
459 | { | |
fb4f88dc HD |
460 | struct swap_info_struct *si; |
461 | pgoff_t offset; | |
462 | int type, next; | |
463 | int wrapped = 0; | |
1da177e4 | 464 | |
5d337b91 | 465 | spin_lock(&swap_lock); |
1da177e4 | 466 | if (nr_swap_pages <= 0) |
fb4f88dc HD |
467 | goto noswap; |
468 | nr_swap_pages--; | |
469 | ||
470 | for (type = swap_list.next; type >= 0 && wrapped < 2; type = next) { | |
efa90a98 | 471 | si = swap_info[type]; |
fb4f88dc HD |
472 | next = si->next; |
473 | if (next < 0 || | |
efa90a98 | 474 | (!wrapped && si->prio != swap_info[next]->prio)) { |
fb4f88dc HD |
475 | next = swap_list.head; |
476 | wrapped++; | |
1da177e4 | 477 | } |
fb4f88dc HD |
478 | |
479 | if (!si->highest_bit) | |
480 | continue; | |
481 | if (!(si->flags & SWP_WRITEOK)) | |
482 | continue; | |
483 | ||
484 | swap_list.next = next; | |
355cfa73 KH |
485 | /* This is called for allocating swap entry for cache */ |
486 | offset = scan_swap_map(si, SWAP_CACHE); | |
5d337b91 HD |
487 | if (offset) { |
488 | spin_unlock(&swap_lock); | |
fb4f88dc | 489 | return swp_entry(type, offset); |
5d337b91 | 490 | } |
fb4f88dc | 491 | next = swap_list.next; |
1da177e4 | 492 | } |
fb4f88dc HD |
493 | |
494 | nr_swap_pages++; | |
495 | noswap: | |
5d337b91 | 496 | spin_unlock(&swap_lock); |
fb4f88dc | 497 | return (swp_entry_t) {0}; |
1da177e4 LT |
498 | } |
499 | ||
355cfa73 | 500 | /* The only caller of this function is now susupend routine */ |
3a291a20 RW |
501 | swp_entry_t get_swap_page_of_type(int type) |
502 | { | |
503 | struct swap_info_struct *si; | |
504 | pgoff_t offset; | |
505 | ||
506 | spin_lock(&swap_lock); | |
efa90a98 HD |
507 | si = swap_info[type]; |
508 | if (si && (si->flags & SWP_WRITEOK)) { | |
3a291a20 | 509 | nr_swap_pages--; |
355cfa73 KH |
510 | /* This is called for allocating swap entry, not cache */ |
511 | offset = scan_swap_map(si, SWAP_MAP); | |
3a291a20 RW |
512 | if (offset) { |
513 | spin_unlock(&swap_lock); | |
514 | return swp_entry(type, offset); | |
515 | } | |
516 | nr_swap_pages++; | |
517 | } | |
518 | spin_unlock(&swap_lock); | |
519 | return (swp_entry_t) {0}; | |
520 | } | |
521 | ||
73c34b6a | 522 | static struct swap_info_struct *swap_info_get(swp_entry_t entry) |
1da177e4 | 523 | { |
73c34b6a | 524 | struct swap_info_struct *p; |
1da177e4 LT |
525 | unsigned long offset, type; |
526 | ||
527 | if (!entry.val) | |
528 | goto out; | |
529 | type = swp_type(entry); | |
530 | if (type >= nr_swapfiles) | |
531 | goto bad_nofile; | |
efa90a98 | 532 | p = swap_info[type]; |
1da177e4 LT |
533 | if (!(p->flags & SWP_USED)) |
534 | goto bad_device; | |
535 | offset = swp_offset(entry); | |
536 | if (offset >= p->max) | |
537 | goto bad_offset; | |
538 | if (!p->swap_map[offset]) | |
539 | goto bad_free; | |
5d337b91 | 540 | spin_lock(&swap_lock); |
1da177e4 LT |
541 | return p; |
542 | ||
543 | bad_free: | |
544 | printk(KERN_ERR "swap_free: %s%08lx\n", Unused_offset, entry.val); | |
545 | goto out; | |
546 | bad_offset: | |
547 | printk(KERN_ERR "swap_free: %s%08lx\n", Bad_offset, entry.val); | |
548 | goto out; | |
549 | bad_device: | |
550 | printk(KERN_ERR "swap_free: %s%08lx\n", Unused_file, entry.val); | |
551 | goto out; | |
552 | bad_nofile: | |
553 | printk(KERN_ERR "swap_free: %s%08lx\n", Bad_file, entry.val); | |
554 | out: | |
555 | return NULL; | |
886bb7e9 | 556 | } |
1da177e4 | 557 | |
355cfa73 KH |
558 | static int swap_entry_free(struct swap_info_struct *p, |
559 | swp_entry_t ent, int cache) | |
1da177e4 | 560 | { |
8c7c6e34 | 561 | unsigned long offset = swp_offset(ent); |
355cfa73 KH |
562 | int count = swap_count(p->swap_map[offset]); |
563 | bool has_cache; | |
564 | ||
565 | has_cache = swap_has_cache(p->swap_map[offset]); | |
566 | ||
567 | if (cache == SWAP_MAP) { /* dropping usage count of swap */ | |
568 | if (count < SWAP_MAP_MAX) { | |
569 | count--; | |
570 | p->swap_map[offset] = encode_swapmap(count, has_cache); | |
1da177e4 | 571 | } |
355cfa73 KH |
572 | } else { /* dropping swap cache flag */ |
573 | VM_BUG_ON(!has_cache); | |
574 | p->swap_map[offset] = encode_swapmap(count, false); | |
575 | ||
576 | } | |
577 | /* return code. */ | |
578 | count = p->swap_map[offset]; | |
579 | /* free if no reference */ | |
580 | if (!count) { | |
581 | if (offset < p->lowest_bit) | |
582 | p->lowest_bit = offset; | |
583 | if (offset > p->highest_bit) | |
584 | p->highest_bit = offset; | |
efa90a98 HD |
585 | if (swap_list.next >= 0 && |
586 | p->prio > swap_info[swap_list.next]->prio) | |
587 | swap_list.next = p->type; | |
355cfa73 KH |
588 | nr_swap_pages++; |
589 | p->inuse_pages--; | |
1da177e4 | 590 | } |
8a9478ca KH |
591 | if (!swap_count(count)) |
592 | mem_cgroup_uncharge_swap(ent); | |
1da177e4 LT |
593 | return count; |
594 | } | |
595 | ||
596 | /* | |
597 | * Caller has made sure that the swapdevice corresponding to entry | |
598 | * is still around or has not been recycled. | |
599 | */ | |
600 | void swap_free(swp_entry_t entry) | |
601 | { | |
73c34b6a | 602 | struct swap_info_struct *p; |
1da177e4 LT |
603 | |
604 | p = swap_info_get(entry); | |
605 | if (p) { | |
355cfa73 | 606 | swap_entry_free(p, entry, SWAP_MAP); |
5d337b91 | 607 | spin_unlock(&swap_lock); |
1da177e4 LT |
608 | } |
609 | } | |
610 | ||
cb4b86ba KH |
611 | /* |
612 | * Called after dropping swapcache to decrease refcnt to swap entries. | |
613 | */ | |
614 | void swapcache_free(swp_entry_t entry, struct page *page) | |
615 | { | |
355cfa73 | 616 | struct swap_info_struct *p; |
8a9478ca | 617 | int ret; |
355cfa73 | 618 | |
355cfa73 KH |
619 | p = swap_info_get(entry); |
620 | if (p) { | |
8a9478ca KH |
621 | ret = swap_entry_free(p, entry, SWAP_CACHE); |
622 | if (page) { | |
623 | bool swapout; | |
624 | if (ret) | |
625 | swapout = true; /* the end of swap out */ | |
626 | else | |
627 | swapout = false; /* no more swap users! */ | |
628 | mem_cgroup_uncharge_swapcache(page, entry, swapout); | |
629 | } | |
355cfa73 KH |
630 | spin_unlock(&swap_lock); |
631 | } | |
cb4b86ba KH |
632 | } |
633 | ||
1da177e4 | 634 | /* |
c475a8ab | 635 | * How many references to page are currently swapped out? |
1da177e4 | 636 | */ |
c475a8ab | 637 | static inline int page_swapcount(struct page *page) |
1da177e4 | 638 | { |
c475a8ab HD |
639 | int count = 0; |
640 | struct swap_info_struct *p; | |
1da177e4 LT |
641 | swp_entry_t entry; |
642 | ||
4c21e2f2 | 643 | entry.val = page_private(page); |
1da177e4 LT |
644 | p = swap_info_get(entry); |
645 | if (p) { | |
355cfa73 | 646 | count = swap_count(p->swap_map[swp_offset(entry)]); |
5d337b91 | 647 | spin_unlock(&swap_lock); |
1da177e4 | 648 | } |
c475a8ab | 649 | return count; |
1da177e4 LT |
650 | } |
651 | ||
652 | /* | |
7b1fe597 HD |
653 | * We can write to an anon page without COW if there are no other references |
654 | * to it. And as a side-effect, free up its swap: because the old content | |
655 | * on disk will never be read, and seeking back there to write new content | |
656 | * later would only waste time away from clustering. | |
1da177e4 | 657 | */ |
7b1fe597 | 658 | int reuse_swap_page(struct page *page) |
1da177e4 | 659 | { |
c475a8ab HD |
660 | int count; |
661 | ||
51726b12 | 662 | VM_BUG_ON(!PageLocked(page)); |
c475a8ab | 663 | count = page_mapcount(page); |
7b1fe597 | 664 | if (count <= 1 && PageSwapCache(page)) { |
c475a8ab | 665 | count += page_swapcount(page); |
7b1fe597 HD |
666 | if (count == 1 && !PageWriteback(page)) { |
667 | delete_from_swap_cache(page); | |
668 | SetPageDirty(page); | |
669 | } | |
670 | } | |
c475a8ab | 671 | return count == 1; |
1da177e4 LT |
672 | } |
673 | ||
674 | /* | |
a2c43eed HD |
675 | * If swap is getting full, or if there are no more mappings of this page, |
676 | * then try_to_free_swap is called to free its swap space. | |
1da177e4 | 677 | */ |
a2c43eed | 678 | int try_to_free_swap(struct page *page) |
1da177e4 | 679 | { |
51726b12 | 680 | VM_BUG_ON(!PageLocked(page)); |
1da177e4 LT |
681 | |
682 | if (!PageSwapCache(page)) | |
683 | return 0; | |
684 | if (PageWriteback(page)) | |
685 | return 0; | |
a2c43eed | 686 | if (page_swapcount(page)) |
1da177e4 LT |
687 | return 0; |
688 | ||
a2c43eed HD |
689 | delete_from_swap_cache(page); |
690 | SetPageDirty(page); | |
691 | return 1; | |
68a22394 RR |
692 | } |
693 | ||
1da177e4 LT |
694 | /* |
695 | * Free the swap entry like above, but also try to | |
696 | * free the page cache entry if it is the last user. | |
697 | */ | |
2509ef26 | 698 | int free_swap_and_cache(swp_entry_t entry) |
1da177e4 | 699 | { |
2509ef26 | 700 | struct swap_info_struct *p; |
1da177e4 LT |
701 | struct page *page = NULL; |
702 | ||
a7420aa5 | 703 | if (non_swap_entry(entry)) |
2509ef26 | 704 | return 1; |
0697212a | 705 | |
1da177e4 LT |
706 | p = swap_info_get(entry); |
707 | if (p) { | |
355cfa73 | 708 | if (swap_entry_free(p, entry, SWAP_MAP) == SWAP_HAS_CACHE) { |
93fac704 | 709 | page = find_get_page(&swapper_space, entry.val); |
8413ac9d | 710 | if (page && !trylock_page(page)) { |
93fac704 NP |
711 | page_cache_release(page); |
712 | page = NULL; | |
713 | } | |
714 | } | |
5d337b91 | 715 | spin_unlock(&swap_lock); |
1da177e4 LT |
716 | } |
717 | if (page) { | |
a2c43eed HD |
718 | /* |
719 | * Not mapped elsewhere, or swap space full? Free it! | |
720 | * Also recheck PageSwapCache now page is locked (above). | |
721 | */ | |
93fac704 | 722 | if (PageSwapCache(page) && !PageWriteback(page) && |
a2c43eed | 723 | (!page_mapped(page) || vm_swap_full())) { |
1da177e4 LT |
724 | delete_from_swap_cache(page); |
725 | SetPageDirty(page); | |
726 | } | |
727 | unlock_page(page); | |
728 | page_cache_release(page); | |
729 | } | |
2509ef26 | 730 | return p != NULL; |
1da177e4 LT |
731 | } |
732 | ||
b0cb1a19 | 733 | #ifdef CONFIG_HIBERNATION |
f577eb30 | 734 | /* |
915bae9e | 735 | * Find the swap type that corresponds to given device (if any). |
f577eb30 | 736 | * |
915bae9e RW |
737 | * @offset - number of the PAGE_SIZE-sized block of the device, starting |
738 | * from 0, in which the swap header is expected to be located. | |
739 | * | |
740 | * This is needed for the suspend to disk (aka swsusp). | |
f577eb30 | 741 | */ |
7bf23687 | 742 | int swap_type_of(dev_t device, sector_t offset, struct block_device **bdev_p) |
f577eb30 | 743 | { |
915bae9e | 744 | struct block_device *bdev = NULL; |
efa90a98 | 745 | int type; |
f577eb30 | 746 | |
915bae9e RW |
747 | if (device) |
748 | bdev = bdget(device); | |
749 | ||
f577eb30 | 750 | spin_lock(&swap_lock); |
efa90a98 HD |
751 | for (type = 0; type < nr_swapfiles; type++) { |
752 | struct swap_info_struct *sis = swap_info[type]; | |
f577eb30 | 753 | |
915bae9e | 754 | if (!(sis->flags & SWP_WRITEOK)) |
f577eb30 | 755 | continue; |
b6b5bce3 | 756 | |
915bae9e | 757 | if (!bdev) { |
7bf23687 | 758 | if (bdev_p) |
dddac6a7 | 759 | *bdev_p = bdgrab(sis->bdev); |
7bf23687 | 760 | |
6e1819d6 | 761 | spin_unlock(&swap_lock); |
efa90a98 | 762 | return type; |
6e1819d6 | 763 | } |
915bae9e | 764 | if (bdev == sis->bdev) { |
9625a5f2 | 765 | struct swap_extent *se = &sis->first_swap_extent; |
915bae9e | 766 | |
915bae9e | 767 | if (se->start_block == offset) { |
7bf23687 | 768 | if (bdev_p) |
dddac6a7 | 769 | *bdev_p = bdgrab(sis->bdev); |
7bf23687 | 770 | |
915bae9e RW |
771 | spin_unlock(&swap_lock); |
772 | bdput(bdev); | |
efa90a98 | 773 | return type; |
915bae9e | 774 | } |
f577eb30 RW |
775 | } |
776 | } | |
777 | spin_unlock(&swap_lock); | |
915bae9e RW |
778 | if (bdev) |
779 | bdput(bdev); | |
780 | ||
f577eb30 RW |
781 | return -ENODEV; |
782 | } | |
783 | ||
73c34b6a HD |
784 | /* |
785 | * Get the (PAGE_SIZE) block corresponding to given offset on the swapdev | |
786 | * corresponding to given index in swap_info (swap type). | |
787 | */ | |
788 | sector_t swapdev_block(int type, pgoff_t offset) | |
789 | { | |
790 | struct block_device *bdev; | |
791 | ||
792 | if ((unsigned int)type >= nr_swapfiles) | |
793 | return 0; | |
794 | if (!(swap_info[type]->flags & SWP_WRITEOK)) | |
795 | return 0; | |
796 | return map_swap_page(swp_entry(type, offset), &bdev); | |
797 | } | |
798 | ||
f577eb30 RW |
799 | /* |
800 | * Return either the total number of swap pages of given type, or the number | |
801 | * of free pages of that type (depending on @free) | |
802 | * | |
803 | * This is needed for software suspend | |
804 | */ | |
805 | unsigned int count_swap_pages(int type, int free) | |
806 | { | |
807 | unsigned int n = 0; | |
808 | ||
efa90a98 HD |
809 | spin_lock(&swap_lock); |
810 | if ((unsigned int)type < nr_swapfiles) { | |
811 | struct swap_info_struct *sis = swap_info[type]; | |
812 | ||
813 | if (sis->flags & SWP_WRITEOK) { | |
814 | n = sis->pages; | |
f577eb30 | 815 | if (free) |
efa90a98 | 816 | n -= sis->inuse_pages; |
f577eb30 | 817 | } |
f577eb30 | 818 | } |
efa90a98 | 819 | spin_unlock(&swap_lock); |
f577eb30 RW |
820 | return n; |
821 | } | |
73c34b6a | 822 | #endif /* CONFIG_HIBERNATION */ |
f577eb30 | 823 | |
1da177e4 | 824 | /* |
72866f6f HD |
825 | * No need to decide whether this PTE shares the swap entry with others, |
826 | * just let do_wp_page work it out if a write is requested later - to | |
827 | * force COW, vm_page_prot omits write permission from any private vma. | |
1da177e4 | 828 | */ |
044d66c1 | 829 | static int unuse_pte(struct vm_area_struct *vma, pmd_t *pmd, |
1da177e4 LT |
830 | unsigned long addr, swp_entry_t entry, struct page *page) |
831 | { | |
7a81b88c | 832 | struct mem_cgroup *ptr = NULL; |
044d66c1 HD |
833 | spinlock_t *ptl; |
834 | pte_t *pte; | |
835 | int ret = 1; | |
836 | ||
85d9fc89 | 837 | if (mem_cgroup_try_charge_swapin(vma->vm_mm, page, GFP_KERNEL, &ptr)) { |
044d66c1 | 838 | ret = -ENOMEM; |
85d9fc89 KH |
839 | goto out_nolock; |
840 | } | |
044d66c1 HD |
841 | |
842 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); | |
843 | if (unlikely(!pte_same(*pte, swp_entry_to_pte(entry)))) { | |
844 | if (ret > 0) | |
7a81b88c | 845 | mem_cgroup_cancel_charge_swapin(ptr); |
044d66c1 HD |
846 | ret = 0; |
847 | goto out; | |
848 | } | |
8a9f3ccd | 849 | |
4294621f | 850 | inc_mm_counter(vma->vm_mm, anon_rss); |
1da177e4 LT |
851 | get_page(page); |
852 | set_pte_at(vma->vm_mm, addr, pte, | |
853 | pte_mkold(mk_pte(page, vma->vm_page_prot))); | |
854 | page_add_anon_rmap(page, vma, addr); | |
7a81b88c | 855 | mem_cgroup_commit_charge_swapin(page, ptr); |
1da177e4 LT |
856 | swap_free(entry); |
857 | /* | |
858 | * Move the page to the active list so it is not | |
859 | * immediately swapped out again after swapon. | |
860 | */ | |
861 | activate_page(page); | |
044d66c1 HD |
862 | out: |
863 | pte_unmap_unlock(pte, ptl); | |
85d9fc89 | 864 | out_nolock: |
044d66c1 | 865 | return ret; |
1da177e4 LT |
866 | } |
867 | ||
868 | static int unuse_pte_range(struct vm_area_struct *vma, pmd_t *pmd, | |
869 | unsigned long addr, unsigned long end, | |
870 | swp_entry_t entry, struct page *page) | |
871 | { | |
1da177e4 | 872 | pte_t swp_pte = swp_entry_to_pte(entry); |
705e87c0 | 873 | pte_t *pte; |
8a9f3ccd | 874 | int ret = 0; |
1da177e4 | 875 | |
044d66c1 HD |
876 | /* |
877 | * We don't actually need pte lock while scanning for swp_pte: since | |
878 | * we hold page lock and mmap_sem, swp_pte cannot be inserted into the | |
879 | * page table while we're scanning; though it could get zapped, and on | |
880 | * some architectures (e.g. x86_32 with PAE) we might catch a glimpse | |
881 | * of unmatched parts which look like swp_pte, so unuse_pte must | |
882 | * recheck under pte lock. Scanning without pte lock lets it be | |
883 | * preemptible whenever CONFIG_PREEMPT but not CONFIG_HIGHPTE. | |
884 | */ | |
885 | pte = pte_offset_map(pmd, addr); | |
1da177e4 LT |
886 | do { |
887 | /* | |
888 | * swapoff spends a _lot_ of time in this loop! | |
889 | * Test inline before going to call unuse_pte. | |
890 | */ | |
891 | if (unlikely(pte_same(*pte, swp_pte))) { | |
044d66c1 HD |
892 | pte_unmap(pte); |
893 | ret = unuse_pte(vma, pmd, addr, entry, page); | |
894 | if (ret) | |
895 | goto out; | |
896 | pte = pte_offset_map(pmd, addr); | |
1da177e4 LT |
897 | } |
898 | } while (pte++, addr += PAGE_SIZE, addr != end); | |
044d66c1 HD |
899 | pte_unmap(pte - 1); |
900 | out: | |
8a9f3ccd | 901 | return ret; |
1da177e4 LT |
902 | } |
903 | ||
904 | static inline int unuse_pmd_range(struct vm_area_struct *vma, pud_t *pud, | |
905 | unsigned long addr, unsigned long end, | |
906 | swp_entry_t entry, struct page *page) | |
907 | { | |
908 | pmd_t *pmd; | |
909 | unsigned long next; | |
8a9f3ccd | 910 | int ret; |
1da177e4 LT |
911 | |
912 | pmd = pmd_offset(pud, addr); | |
913 | do { | |
914 | next = pmd_addr_end(addr, end); | |
915 | if (pmd_none_or_clear_bad(pmd)) | |
916 | continue; | |
8a9f3ccd BS |
917 | ret = unuse_pte_range(vma, pmd, addr, next, entry, page); |
918 | if (ret) | |
919 | return ret; | |
1da177e4 LT |
920 | } while (pmd++, addr = next, addr != end); |
921 | return 0; | |
922 | } | |
923 | ||
924 | static inline int unuse_pud_range(struct vm_area_struct *vma, pgd_t *pgd, | |
925 | unsigned long addr, unsigned long end, | |
926 | swp_entry_t entry, struct page *page) | |
927 | { | |
928 | pud_t *pud; | |
929 | unsigned long next; | |
8a9f3ccd | 930 | int ret; |
1da177e4 LT |
931 | |
932 | pud = pud_offset(pgd, addr); | |
933 | do { | |
934 | next = pud_addr_end(addr, end); | |
935 | if (pud_none_or_clear_bad(pud)) | |
936 | continue; | |
8a9f3ccd BS |
937 | ret = unuse_pmd_range(vma, pud, addr, next, entry, page); |
938 | if (ret) | |
939 | return ret; | |
1da177e4 LT |
940 | } while (pud++, addr = next, addr != end); |
941 | return 0; | |
942 | } | |
943 | ||
944 | static int unuse_vma(struct vm_area_struct *vma, | |
945 | swp_entry_t entry, struct page *page) | |
946 | { | |
947 | pgd_t *pgd; | |
948 | unsigned long addr, end, next; | |
8a9f3ccd | 949 | int ret; |
1da177e4 LT |
950 | |
951 | if (page->mapping) { | |
952 | addr = page_address_in_vma(page, vma); | |
953 | if (addr == -EFAULT) | |
954 | return 0; | |
955 | else | |
956 | end = addr + PAGE_SIZE; | |
957 | } else { | |
958 | addr = vma->vm_start; | |
959 | end = vma->vm_end; | |
960 | } | |
961 | ||
962 | pgd = pgd_offset(vma->vm_mm, addr); | |
963 | do { | |
964 | next = pgd_addr_end(addr, end); | |
965 | if (pgd_none_or_clear_bad(pgd)) | |
966 | continue; | |
8a9f3ccd BS |
967 | ret = unuse_pud_range(vma, pgd, addr, next, entry, page); |
968 | if (ret) | |
969 | return ret; | |
1da177e4 LT |
970 | } while (pgd++, addr = next, addr != end); |
971 | return 0; | |
972 | } | |
973 | ||
974 | static int unuse_mm(struct mm_struct *mm, | |
975 | swp_entry_t entry, struct page *page) | |
976 | { | |
977 | struct vm_area_struct *vma; | |
8a9f3ccd | 978 | int ret = 0; |
1da177e4 LT |
979 | |
980 | if (!down_read_trylock(&mm->mmap_sem)) { | |
981 | /* | |
7d03431c FLVC |
982 | * Activate page so shrink_inactive_list is unlikely to unmap |
983 | * its ptes while lock is dropped, so swapoff can make progress. | |
1da177e4 | 984 | */ |
c475a8ab | 985 | activate_page(page); |
1da177e4 LT |
986 | unlock_page(page); |
987 | down_read(&mm->mmap_sem); | |
988 | lock_page(page); | |
989 | } | |
1da177e4 | 990 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
8a9f3ccd | 991 | if (vma->anon_vma && (ret = unuse_vma(vma, entry, page))) |
1da177e4 LT |
992 | break; |
993 | } | |
1da177e4 | 994 | up_read(&mm->mmap_sem); |
8a9f3ccd | 995 | return (ret < 0)? ret: 0; |
1da177e4 LT |
996 | } |
997 | ||
998 | /* | |
999 | * Scan swap_map from current position to next entry still in use. | |
1000 | * Recycle to start on reaching the end, returning 0 when empty. | |
1001 | */ | |
6eb396dc HD |
1002 | static unsigned int find_next_to_unuse(struct swap_info_struct *si, |
1003 | unsigned int prev) | |
1da177e4 | 1004 | { |
6eb396dc HD |
1005 | unsigned int max = si->max; |
1006 | unsigned int i = prev; | |
1da177e4 LT |
1007 | int count; |
1008 | ||
1009 | /* | |
5d337b91 | 1010 | * No need for swap_lock here: we're just looking |
1da177e4 LT |
1011 | * for whether an entry is in use, not modifying it; false |
1012 | * hits are okay, and sys_swapoff() has already prevented new | |
5d337b91 | 1013 | * allocations from this area (while holding swap_lock). |
1da177e4 LT |
1014 | */ |
1015 | for (;;) { | |
1016 | if (++i >= max) { | |
1017 | if (!prev) { | |
1018 | i = 0; | |
1019 | break; | |
1020 | } | |
1021 | /* | |
1022 | * No entries in use at top of swap_map, | |
1023 | * loop back to start and recheck there. | |
1024 | */ | |
1025 | max = prev + 1; | |
1026 | prev = 0; | |
1027 | i = 1; | |
1028 | } | |
1029 | count = si->swap_map[i]; | |
355cfa73 | 1030 | if (count && swap_count(count) != SWAP_MAP_BAD) |
1da177e4 LT |
1031 | break; |
1032 | } | |
1033 | return i; | |
1034 | } | |
1035 | ||
1036 | /* | |
1037 | * We completely avoid races by reading each swap page in advance, | |
1038 | * and then search for the process using it. All the necessary | |
1039 | * page table adjustments can then be made atomically. | |
1040 | */ | |
1041 | static int try_to_unuse(unsigned int type) | |
1042 | { | |
efa90a98 | 1043 | struct swap_info_struct *si = swap_info[type]; |
1da177e4 LT |
1044 | struct mm_struct *start_mm; |
1045 | unsigned short *swap_map; | |
1046 | unsigned short swcount; | |
1047 | struct page *page; | |
1048 | swp_entry_t entry; | |
6eb396dc | 1049 | unsigned int i = 0; |
1da177e4 LT |
1050 | int retval = 0; |
1051 | int reset_overflow = 0; | |
1052 | int shmem; | |
1053 | ||
1054 | /* | |
1055 | * When searching mms for an entry, a good strategy is to | |
1056 | * start at the first mm we freed the previous entry from | |
1057 | * (though actually we don't notice whether we or coincidence | |
1058 | * freed the entry). Initialize this start_mm with a hold. | |
1059 | * | |
1060 | * A simpler strategy would be to start at the last mm we | |
1061 | * freed the previous entry from; but that would take less | |
1062 | * advantage of mmlist ordering, which clusters forked mms | |
1063 | * together, child after parent. If we race with dup_mmap(), we | |
1064 | * prefer to resolve parent before child, lest we miss entries | |
1065 | * duplicated after we scanned child: using last mm would invert | |
1066 | * that. Though it's only a serious concern when an overflowed | |
1067 | * swap count is reset from SWAP_MAP_MAX, preventing a rescan. | |
1068 | */ | |
1069 | start_mm = &init_mm; | |
1070 | atomic_inc(&init_mm.mm_users); | |
1071 | ||
1072 | /* | |
1073 | * Keep on scanning until all entries have gone. Usually, | |
1074 | * one pass through swap_map is enough, but not necessarily: | |
1075 | * there are races when an instance of an entry might be missed. | |
1076 | */ | |
1077 | while ((i = find_next_to_unuse(si, i)) != 0) { | |
1078 | if (signal_pending(current)) { | |
1079 | retval = -EINTR; | |
1080 | break; | |
1081 | } | |
1082 | ||
886bb7e9 | 1083 | /* |
1da177e4 LT |
1084 | * Get a page for the entry, using the existing swap |
1085 | * cache page if there is one. Otherwise, get a clean | |
886bb7e9 | 1086 | * page and read the swap into it. |
1da177e4 LT |
1087 | */ |
1088 | swap_map = &si->swap_map[i]; | |
1089 | entry = swp_entry(type, i); | |
02098fea HD |
1090 | page = read_swap_cache_async(entry, |
1091 | GFP_HIGHUSER_MOVABLE, NULL, 0); | |
1da177e4 LT |
1092 | if (!page) { |
1093 | /* | |
1094 | * Either swap_duplicate() failed because entry | |
1095 | * has been freed independently, and will not be | |
1096 | * reused since sys_swapoff() already disabled | |
1097 | * allocation from here, or alloc_page() failed. | |
1098 | */ | |
1099 | if (!*swap_map) | |
1100 | continue; | |
1101 | retval = -ENOMEM; | |
1102 | break; | |
1103 | } | |
1104 | ||
1105 | /* | |
1106 | * Don't hold on to start_mm if it looks like exiting. | |
1107 | */ | |
1108 | if (atomic_read(&start_mm->mm_users) == 1) { | |
1109 | mmput(start_mm); | |
1110 | start_mm = &init_mm; | |
1111 | atomic_inc(&init_mm.mm_users); | |
1112 | } | |
1113 | ||
1114 | /* | |
1115 | * Wait for and lock page. When do_swap_page races with | |
1116 | * try_to_unuse, do_swap_page can handle the fault much | |
1117 | * faster than try_to_unuse can locate the entry. This | |
1118 | * apparently redundant "wait_on_page_locked" lets try_to_unuse | |
1119 | * defer to do_swap_page in such a case - in some tests, | |
1120 | * do_swap_page and try_to_unuse repeatedly compete. | |
1121 | */ | |
1122 | wait_on_page_locked(page); | |
1123 | wait_on_page_writeback(page); | |
1124 | lock_page(page); | |
1125 | wait_on_page_writeback(page); | |
1126 | ||
1127 | /* | |
1128 | * Remove all references to entry. | |
1129 | * Whenever we reach init_mm, there's no address space | |
1130 | * to search, but use it as a reminder to search shmem. | |
1131 | */ | |
1132 | shmem = 0; | |
1133 | swcount = *swap_map; | |
355cfa73 | 1134 | if (swap_count(swcount)) { |
1da177e4 LT |
1135 | if (start_mm == &init_mm) |
1136 | shmem = shmem_unuse(entry, page); | |
1137 | else | |
1138 | retval = unuse_mm(start_mm, entry, page); | |
1139 | } | |
355cfa73 | 1140 | if (swap_count(*swap_map)) { |
1da177e4 LT |
1141 | int set_start_mm = (*swap_map >= swcount); |
1142 | struct list_head *p = &start_mm->mmlist; | |
1143 | struct mm_struct *new_start_mm = start_mm; | |
1144 | struct mm_struct *prev_mm = start_mm; | |
1145 | struct mm_struct *mm; | |
1146 | ||
1147 | atomic_inc(&new_start_mm->mm_users); | |
1148 | atomic_inc(&prev_mm->mm_users); | |
1149 | spin_lock(&mmlist_lock); | |
355cfa73 | 1150 | while (swap_count(*swap_map) && !retval && !shmem && |
1da177e4 LT |
1151 | (p = p->next) != &start_mm->mmlist) { |
1152 | mm = list_entry(p, struct mm_struct, mmlist); | |
70af7c5c | 1153 | if (!atomic_inc_not_zero(&mm->mm_users)) |
1da177e4 | 1154 | continue; |
1da177e4 LT |
1155 | spin_unlock(&mmlist_lock); |
1156 | mmput(prev_mm); | |
1157 | prev_mm = mm; | |
1158 | ||
1159 | cond_resched(); | |
1160 | ||
1161 | swcount = *swap_map; | |
355cfa73 | 1162 | if (!swap_count(swcount)) /* any usage ? */ |
1da177e4 LT |
1163 | ; |
1164 | else if (mm == &init_mm) { | |
1165 | set_start_mm = 1; | |
1166 | shmem = shmem_unuse(entry, page); | |
1167 | } else | |
1168 | retval = unuse_mm(mm, entry, page); | |
355cfa73 | 1169 | |
32c5fc10 | 1170 | if (set_start_mm && *swap_map < swcount) { |
1da177e4 LT |
1171 | mmput(new_start_mm); |
1172 | atomic_inc(&mm->mm_users); | |
1173 | new_start_mm = mm; | |
1174 | set_start_mm = 0; | |
1175 | } | |
1176 | spin_lock(&mmlist_lock); | |
1177 | } | |
1178 | spin_unlock(&mmlist_lock); | |
1179 | mmput(prev_mm); | |
1180 | mmput(start_mm); | |
1181 | start_mm = new_start_mm; | |
1182 | } | |
2e0e26c7 HD |
1183 | if (shmem) { |
1184 | /* page has already been unlocked and released */ | |
1185 | if (shmem > 0) | |
1186 | continue; | |
1187 | retval = shmem; | |
1188 | break; | |
1189 | } | |
1da177e4 LT |
1190 | if (retval) { |
1191 | unlock_page(page); | |
1192 | page_cache_release(page); | |
1193 | break; | |
1194 | } | |
1195 | ||
1196 | /* | |
355cfa73 KH |
1197 | * How could swap count reach 0x7ffe ? |
1198 | * There's no way to repeat a swap page within an mm | |
1199 | * (except in shmem, where it's the shared object which takes | |
1200 | * the reference count)? | |
1201 | * We believe SWAP_MAP_MAX cannot occur.(if occur, unsigned | |
1202 | * short is too small....) | |
1da177e4 LT |
1203 | * If that's wrong, then we should worry more about |
1204 | * exit_mmap() and do_munmap() cases described above: | |
1205 | * we might be resetting SWAP_MAP_MAX too early here. | |
1206 | * We know "Undead"s can happen, they're okay, so don't | |
1207 | * report them; but do report if we reset SWAP_MAP_MAX. | |
1208 | */ | |
355cfa73 KH |
1209 | /* We might release the lock_page() in unuse_mm(). */ |
1210 | if (!PageSwapCache(page) || page_private(page) != entry.val) | |
1211 | goto retry; | |
1212 | ||
1213 | if (swap_count(*swap_map) == SWAP_MAP_MAX) { | |
5d337b91 | 1214 | spin_lock(&swap_lock); |
355cfa73 | 1215 | *swap_map = encode_swapmap(0, true); |
5d337b91 | 1216 | spin_unlock(&swap_lock); |
1da177e4 LT |
1217 | reset_overflow = 1; |
1218 | } | |
1219 | ||
1220 | /* | |
1221 | * If a reference remains (rare), we would like to leave | |
1222 | * the page in the swap cache; but try_to_unmap could | |
1223 | * then re-duplicate the entry once we drop page lock, | |
1224 | * so we might loop indefinitely; also, that page could | |
1225 | * not be swapped out to other storage meanwhile. So: | |
1226 | * delete from cache even if there's another reference, | |
1227 | * after ensuring that the data has been saved to disk - | |
1228 | * since if the reference remains (rarer), it will be | |
1229 | * read from disk into another page. Splitting into two | |
1230 | * pages would be incorrect if swap supported "shared | |
1231 | * private" pages, but they are handled by tmpfs files. | |
1da177e4 | 1232 | */ |
355cfa73 KH |
1233 | if (swap_count(*swap_map) && |
1234 | PageDirty(page) && PageSwapCache(page)) { | |
1da177e4 LT |
1235 | struct writeback_control wbc = { |
1236 | .sync_mode = WB_SYNC_NONE, | |
1237 | }; | |
1238 | ||
1239 | swap_writepage(page, &wbc); | |
1240 | lock_page(page); | |
1241 | wait_on_page_writeback(page); | |
1242 | } | |
68bdc8d6 HD |
1243 | |
1244 | /* | |
1245 | * It is conceivable that a racing task removed this page from | |
1246 | * swap cache just before we acquired the page lock at the top, | |
1247 | * or while we dropped it in unuse_mm(). The page might even | |
1248 | * be back in swap cache on another swap area: that we must not | |
1249 | * delete, since it may not have been written out to swap yet. | |
1250 | */ | |
1251 | if (PageSwapCache(page) && | |
1252 | likely(page_private(page) == entry.val)) | |
2e0e26c7 | 1253 | delete_from_swap_cache(page); |
1da177e4 LT |
1254 | |
1255 | /* | |
1256 | * So we could skip searching mms once swap count went | |
1257 | * to 1, we did not mark any present ptes as dirty: must | |
2706a1b8 | 1258 | * mark page dirty so shrink_page_list will preserve it. |
1da177e4 LT |
1259 | */ |
1260 | SetPageDirty(page); | |
355cfa73 | 1261 | retry: |
1da177e4 LT |
1262 | unlock_page(page); |
1263 | page_cache_release(page); | |
1264 | ||
1265 | /* | |
1266 | * Make sure that we aren't completely killing | |
1267 | * interactive performance. | |
1268 | */ | |
1269 | cond_resched(); | |
1270 | } | |
1271 | ||
1272 | mmput(start_mm); | |
1273 | if (reset_overflow) { | |
1274 | printk(KERN_WARNING "swapoff: cleared swap entry overflow\n"); | |
1275 | swap_overflow = 0; | |
1276 | } | |
1277 | return retval; | |
1278 | } | |
1279 | ||
1280 | /* | |
5d337b91 HD |
1281 | * After a successful try_to_unuse, if no swap is now in use, we know |
1282 | * we can empty the mmlist. swap_lock must be held on entry and exit. | |
1283 | * Note that mmlist_lock nests inside swap_lock, and an mm must be | |
1da177e4 LT |
1284 | * added to the mmlist just after page_duplicate - before would be racy. |
1285 | */ | |
1286 | static void drain_mmlist(void) | |
1287 | { | |
1288 | struct list_head *p, *next; | |
efa90a98 | 1289 | unsigned int type; |
1da177e4 | 1290 | |
efa90a98 HD |
1291 | for (type = 0; type < nr_swapfiles; type++) |
1292 | if (swap_info[type]->inuse_pages) | |
1da177e4 LT |
1293 | return; |
1294 | spin_lock(&mmlist_lock); | |
1295 | list_for_each_safe(p, next, &init_mm.mmlist) | |
1296 | list_del_init(p); | |
1297 | spin_unlock(&mmlist_lock); | |
1298 | } | |
1299 | ||
1300 | /* | |
1301 | * Use this swapdev's extent info to locate the (PAGE_SIZE) block which | |
f29ad6a9 HD |
1302 | * corresponds to page offset `offset'. Note that the type of this function |
1303 | * is sector_t, but it returns page offset into the bdev, not sector offset. | |
1da177e4 | 1304 | */ |
f29ad6a9 | 1305 | sector_t map_swap_page(swp_entry_t entry, struct block_device **bdev) |
1da177e4 | 1306 | { |
f29ad6a9 HD |
1307 | struct swap_info_struct *sis; |
1308 | struct swap_extent *start_se; | |
1309 | struct swap_extent *se; | |
1310 | pgoff_t offset; | |
1311 | ||
efa90a98 | 1312 | sis = swap_info[swp_type(entry)]; |
f29ad6a9 HD |
1313 | *bdev = sis->bdev; |
1314 | ||
1315 | offset = swp_offset(entry); | |
1316 | start_se = sis->curr_swap_extent; | |
1317 | se = start_se; | |
1da177e4 LT |
1318 | |
1319 | for ( ; ; ) { | |
1320 | struct list_head *lh; | |
1321 | ||
1322 | if (se->start_page <= offset && | |
1323 | offset < (se->start_page + se->nr_pages)) { | |
1324 | return se->start_block + (offset - se->start_page); | |
1325 | } | |
11d31886 | 1326 | lh = se->list.next; |
1da177e4 LT |
1327 | se = list_entry(lh, struct swap_extent, list); |
1328 | sis->curr_swap_extent = se; | |
1329 | BUG_ON(se == start_se); /* It *must* be present */ | |
1330 | } | |
1331 | } | |
1332 | ||
1333 | /* | |
1334 | * Free all of a swapdev's extent information | |
1335 | */ | |
1336 | static void destroy_swap_extents(struct swap_info_struct *sis) | |
1337 | { | |
9625a5f2 | 1338 | while (!list_empty(&sis->first_swap_extent.list)) { |
1da177e4 LT |
1339 | struct swap_extent *se; |
1340 | ||
9625a5f2 | 1341 | se = list_entry(sis->first_swap_extent.list.next, |
1da177e4 LT |
1342 | struct swap_extent, list); |
1343 | list_del(&se->list); | |
1344 | kfree(se); | |
1345 | } | |
1da177e4 LT |
1346 | } |
1347 | ||
1348 | /* | |
1349 | * Add a block range (and the corresponding page range) into this swapdev's | |
11d31886 | 1350 | * extent list. The extent list is kept sorted in page order. |
1da177e4 | 1351 | * |
11d31886 | 1352 | * This function rather assumes that it is called in ascending page order. |
1da177e4 LT |
1353 | */ |
1354 | static int | |
1355 | add_swap_extent(struct swap_info_struct *sis, unsigned long start_page, | |
1356 | unsigned long nr_pages, sector_t start_block) | |
1357 | { | |
1358 | struct swap_extent *se; | |
1359 | struct swap_extent *new_se; | |
1360 | struct list_head *lh; | |
1361 | ||
9625a5f2 HD |
1362 | if (start_page == 0) { |
1363 | se = &sis->first_swap_extent; | |
1364 | sis->curr_swap_extent = se; | |
1365 | se->start_page = 0; | |
1366 | se->nr_pages = nr_pages; | |
1367 | se->start_block = start_block; | |
1368 | return 1; | |
1369 | } else { | |
1370 | lh = sis->first_swap_extent.list.prev; /* Highest extent */ | |
1da177e4 | 1371 | se = list_entry(lh, struct swap_extent, list); |
11d31886 HD |
1372 | BUG_ON(se->start_page + se->nr_pages != start_page); |
1373 | if (se->start_block + se->nr_pages == start_block) { | |
1da177e4 LT |
1374 | /* Merge it */ |
1375 | se->nr_pages += nr_pages; | |
1376 | return 0; | |
1377 | } | |
1da177e4 LT |
1378 | } |
1379 | ||
1380 | /* | |
1381 | * No merge. Insert a new extent, preserving ordering. | |
1382 | */ | |
1383 | new_se = kmalloc(sizeof(*se), GFP_KERNEL); | |
1384 | if (new_se == NULL) | |
1385 | return -ENOMEM; | |
1386 | new_se->start_page = start_page; | |
1387 | new_se->nr_pages = nr_pages; | |
1388 | new_se->start_block = start_block; | |
1389 | ||
9625a5f2 | 1390 | list_add_tail(&new_se->list, &sis->first_swap_extent.list); |
53092a74 | 1391 | return 1; |
1da177e4 LT |
1392 | } |
1393 | ||
1394 | /* | |
1395 | * A `swap extent' is a simple thing which maps a contiguous range of pages | |
1396 | * onto a contiguous range of disk blocks. An ordered list of swap extents | |
1397 | * is built at swapon time and is then used at swap_writepage/swap_readpage | |
1398 | * time for locating where on disk a page belongs. | |
1399 | * | |
1400 | * If the swapfile is an S_ISBLK block device, a single extent is installed. | |
1401 | * This is done so that the main operating code can treat S_ISBLK and S_ISREG | |
1402 | * swap files identically. | |
1403 | * | |
1404 | * Whether the swapdev is an S_ISREG file or an S_ISBLK blockdev, the swap | |
1405 | * extent list operates in PAGE_SIZE disk blocks. Both S_ISREG and S_ISBLK | |
1406 | * swapfiles are handled *identically* after swapon time. | |
1407 | * | |
1408 | * For S_ISREG swapfiles, setup_swap_extents() will walk all the file's blocks | |
1409 | * and will parse them into an ordered extent list, in PAGE_SIZE chunks. If | |
1410 | * some stray blocks are found which do not fall within the PAGE_SIZE alignment | |
1411 | * requirements, they are simply tossed out - we will never use those blocks | |
1412 | * for swapping. | |
1413 | * | |
b0d9bcd4 | 1414 | * For S_ISREG swapfiles we set S_SWAPFILE across the life of the swapon. This |
1da177e4 LT |
1415 | * prevents root from shooting her foot off by ftruncating an in-use swapfile, |
1416 | * which will scribble on the fs. | |
1417 | * | |
1418 | * The amount of disk space which a single swap extent represents varies. | |
1419 | * Typically it is in the 1-4 megabyte range. So we can have hundreds of | |
1420 | * extents in the list. To avoid much list walking, we cache the previous | |
1421 | * search location in `curr_swap_extent', and start new searches from there. | |
1422 | * This is extremely effective. The average number of iterations in | |
1423 | * map_swap_page() has been measured at about 0.3 per page. - akpm. | |
1424 | */ | |
53092a74 | 1425 | static int setup_swap_extents(struct swap_info_struct *sis, sector_t *span) |
1da177e4 LT |
1426 | { |
1427 | struct inode *inode; | |
1428 | unsigned blocks_per_page; | |
1429 | unsigned long page_no; | |
1430 | unsigned blkbits; | |
1431 | sector_t probe_block; | |
1432 | sector_t last_block; | |
53092a74 HD |
1433 | sector_t lowest_block = -1; |
1434 | sector_t highest_block = 0; | |
1435 | int nr_extents = 0; | |
1da177e4 LT |
1436 | int ret; |
1437 | ||
1438 | inode = sis->swap_file->f_mapping->host; | |
1439 | if (S_ISBLK(inode->i_mode)) { | |
1440 | ret = add_swap_extent(sis, 0, sis->max, 0); | |
53092a74 | 1441 | *span = sis->pages; |
9625a5f2 | 1442 | goto out; |
1da177e4 LT |
1443 | } |
1444 | ||
1445 | blkbits = inode->i_blkbits; | |
1446 | blocks_per_page = PAGE_SIZE >> blkbits; | |
1447 | ||
1448 | /* | |
1449 | * Map all the blocks into the extent list. This code doesn't try | |
1450 | * to be very smart. | |
1451 | */ | |
1452 | probe_block = 0; | |
1453 | page_no = 0; | |
1454 | last_block = i_size_read(inode) >> blkbits; | |
1455 | while ((probe_block + blocks_per_page) <= last_block && | |
1456 | page_no < sis->max) { | |
1457 | unsigned block_in_page; | |
1458 | sector_t first_block; | |
1459 | ||
1460 | first_block = bmap(inode, probe_block); | |
1461 | if (first_block == 0) | |
1462 | goto bad_bmap; | |
1463 | ||
1464 | /* | |
1465 | * It must be PAGE_SIZE aligned on-disk | |
1466 | */ | |
1467 | if (first_block & (blocks_per_page - 1)) { | |
1468 | probe_block++; | |
1469 | goto reprobe; | |
1470 | } | |
1471 | ||
1472 | for (block_in_page = 1; block_in_page < blocks_per_page; | |
1473 | block_in_page++) { | |
1474 | sector_t block; | |
1475 | ||
1476 | block = bmap(inode, probe_block + block_in_page); | |
1477 | if (block == 0) | |
1478 | goto bad_bmap; | |
1479 | if (block != first_block + block_in_page) { | |
1480 | /* Discontiguity */ | |
1481 | probe_block++; | |
1482 | goto reprobe; | |
1483 | } | |
1484 | } | |
1485 | ||
53092a74 HD |
1486 | first_block >>= (PAGE_SHIFT - blkbits); |
1487 | if (page_no) { /* exclude the header page */ | |
1488 | if (first_block < lowest_block) | |
1489 | lowest_block = first_block; | |
1490 | if (first_block > highest_block) | |
1491 | highest_block = first_block; | |
1492 | } | |
1493 | ||
1da177e4 LT |
1494 | /* |
1495 | * We found a PAGE_SIZE-length, PAGE_SIZE-aligned run of blocks | |
1496 | */ | |
53092a74 HD |
1497 | ret = add_swap_extent(sis, page_no, 1, first_block); |
1498 | if (ret < 0) | |
1da177e4 | 1499 | goto out; |
53092a74 | 1500 | nr_extents += ret; |
1da177e4 LT |
1501 | page_no++; |
1502 | probe_block += blocks_per_page; | |
1503 | reprobe: | |
1504 | continue; | |
1505 | } | |
53092a74 HD |
1506 | ret = nr_extents; |
1507 | *span = 1 + highest_block - lowest_block; | |
1da177e4 | 1508 | if (page_no == 0) |
e2244ec2 | 1509 | page_no = 1; /* force Empty message */ |
1da177e4 | 1510 | sis->max = page_no; |
e2244ec2 | 1511 | sis->pages = page_no - 1; |
1da177e4 | 1512 | sis->highest_bit = page_no - 1; |
9625a5f2 HD |
1513 | out: |
1514 | return ret; | |
1da177e4 LT |
1515 | bad_bmap: |
1516 | printk(KERN_ERR "swapon: swapfile has holes\n"); | |
1517 | ret = -EINVAL; | |
9625a5f2 | 1518 | goto out; |
1da177e4 LT |
1519 | } |
1520 | ||
c4ea37c2 | 1521 | SYSCALL_DEFINE1(swapoff, const char __user *, specialfile) |
1da177e4 | 1522 | { |
73c34b6a | 1523 | struct swap_info_struct *p = NULL; |
1da177e4 LT |
1524 | unsigned short *swap_map; |
1525 | struct file *swap_file, *victim; | |
1526 | struct address_space *mapping; | |
1527 | struct inode *inode; | |
73c34b6a | 1528 | char *pathname; |
1da177e4 LT |
1529 | int i, type, prev; |
1530 | int err; | |
886bb7e9 | 1531 | |
1da177e4 LT |
1532 | if (!capable(CAP_SYS_ADMIN)) |
1533 | return -EPERM; | |
1534 | ||
1535 | 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; | |
5d337b91 | 1548 | spin_lock(&swap_lock); |
efa90a98 HD |
1549 | for (type = swap_list.head; type >= 0; type = swap_info[type]->next) { |
1550 | p = swap_info[type]; | |
22c6f8fd | 1551 | if (p->flags & SWP_WRITEOK) { |
1da177e4 LT |
1552 | if (p->swap_file->f_mapping == mapping) |
1553 | break; | |
1554 | } | |
1555 | prev = type; | |
1556 | } | |
1557 | if (type < 0) { | |
1558 | err = -EINVAL; | |
5d337b91 | 1559 | spin_unlock(&swap_lock); |
1da177e4 LT |
1560 | goto out_dput; |
1561 | } | |
1562 | if (!security_vm_enough_memory(p->pages)) | |
1563 | vm_unacct_memory(p->pages); | |
1564 | else { | |
1565 | err = -ENOMEM; | |
5d337b91 | 1566 | spin_unlock(&swap_lock); |
1da177e4 LT |
1567 | goto out_dput; |
1568 | } | |
efa90a98 | 1569 | if (prev < 0) |
1da177e4 | 1570 | swap_list.head = p->next; |
efa90a98 HD |
1571 | else |
1572 | swap_info[prev]->next = p->next; | |
1da177e4 LT |
1573 | if (type == swap_list.next) { |
1574 | /* just pick something that's safe... */ | |
1575 | swap_list.next = swap_list.head; | |
1576 | } | |
78ecba08 | 1577 | if (p->prio < 0) { |
efa90a98 HD |
1578 | for (i = p->next; i >= 0; i = swap_info[i]->next) |
1579 | swap_info[i]->prio = p->prio--; | |
78ecba08 HD |
1580 | least_priority++; |
1581 | } | |
1da177e4 LT |
1582 | nr_swap_pages -= p->pages; |
1583 | total_swap_pages -= p->pages; | |
1584 | p->flags &= ~SWP_WRITEOK; | |
5d337b91 | 1585 | spin_unlock(&swap_lock); |
fb4f88dc | 1586 | |
35451bee | 1587 | current->flags |= PF_OOM_ORIGIN; |
1da177e4 | 1588 | err = try_to_unuse(type); |
35451bee | 1589 | current->flags &= ~PF_OOM_ORIGIN; |
1da177e4 | 1590 | |
1da177e4 LT |
1591 | if (err) { |
1592 | /* re-insert swap space back into swap_list */ | |
5d337b91 | 1593 | spin_lock(&swap_lock); |
78ecba08 HD |
1594 | if (p->prio < 0) |
1595 | p->prio = --least_priority; | |
1596 | prev = -1; | |
efa90a98 HD |
1597 | for (i = swap_list.head; i >= 0; i = swap_info[i]->next) { |
1598 | if (p->prio >= swap_info[i]->prio) | |
1da177e4 | 1599 | break; |
78ecba08 HD |
1600 | prev = i; |
1601 | } | |
1da177e4 LT |
1602 | p->next = i; |
1603 | if (prev < 0) | |
efa90a98 | 1604 | swap_list.head = swap_list.next = type; |
1da177e4 | 1605 | else |
efa90a98 | 1606 | swap_info[prev]->next = type; |
1da177e4 LT |
1607 | nr_swap_pages += p->pages; |
1608 | total_swap_pages += p->pages; | |
1609 | p->flags |= SWP_WRITEOK; | |
5d337b91 | 1610 | spin_unlock(&swap_lock); |
1da177e4 LT |
1611 | goto out_dput; |
1612 | } | |
52b7efdb HD |
1613 | |
1614 | /* wait for any unplug function to finish */ | |
1615 | down_write(&swap_unplug_sem); | |
1616 | up_write(&swap_unplug_sem); | |
1617 | ||
5d337b91 | 1618 | destroy_swap_extents(p); |
fc0abb14 | 1619 | mutex_lock(&swapon_mutex); |
5d337b91 HD |
1620 | spin_lock(&swap_lock); |
1621 | drain_mmlist(); | |
1622 | ||
52b7efdb | 1623 | /* wait for anyone still in scan_swap_map */ |
52b7efdb HD |
1624 | p->highest_bit = 0; /* cuts scans short */ |
1625 | while (p->flags >= SWP_SCANNING) { | |
5d337b91 | 1626 | spin_unlock(&swap_lock); |
13e4b57f | 1627 | schedule_timeout_uninterruptible(1); |
5d337b91 | 1628 | spin_lock(&swap_lock); |
52b7efdb | 1629 | } |
52b7efdb | 1630 | |
1da177e4 LT |
1631 | swap_file = p->swap_file; |
1632 | p->swap_file = NULL; | |
1633 | p->max = 0; | |
1634 | swap_map = p->swap_map; | |
1635 | p->swap_map = NULL; | |
1636 | p->flags = 0; | |
5d337b91 | 1637 | spin_unlock(&swap_lock); |
fc0abb14 | 1638 | mutex_unlock(&swapon_mutex); |
1da177e4 | 1639 | vfree(swap_map); |
27a7faa0 KH |
1640 | /* Destroy swap account informatin */ |
1641 | swap_cgroup_swapoff(type); | |
1642 | ||
1da177e4 LT |
1643 | inode = mapping->host; |
1644 | if (S_ISBLK(inode->i_mode)) { | |
1645 | struct block_device *bdev = I_BDEV(inode); | |
1646 | set_blocksize(bdev, p->old_block_size); | |
1647 | bd_release(bdev); | |
1648 | } else { | |
1b1dcc1b | 1649 | mutex_lock(&inode->i_mutex); |
1da177e4 | 1650 | inode->i_flags &= ~S_SWAPFILE; |
1b1dcc1b | 1651 | mutex_unlock(&inode->i_mutex); |
1da177e4 LT |
1652 | } |
1653 | filp_close(swap_file, NULL); | |
1654 | err = 0; | |
1655 | ||
1656 | out_dput: | |
1657 | filp_close(victim, NULL); | |
1658 | out: | |
1659 | return err; | |
1660 | } | |
1661 | ||
1662 | #ifdef CONFIG_PROC_FS | |
1663 | /* iterator */ | |
1664 | static void *swap_start(struct seq_file *swap, loff_t *pos) | |
1665 | { | |
efa90a98 HD |
1666 | struct swap_info_struct *si; |
1667 | int type; | |
1da177e4 LT |
1668 | loff_t l = *pos; |
1669 | ||
fc0abb14 | 1670 | mutex_lock(&swapon_mutex); |
1da177e4 | 1671 | |
881e4aab SS |
1672 | if (!l) |
1673 | return SEQ_START_TOKEN; | |
1674 | ||
efa90a98 HD |
1675 | for (type = 0; type < nr_swapfiles; type++) { |
1676 | smp_rmb(); /* read nr_swapfiles before swap_info[type] */ | |
1677 | si = swap_info[type]; | |
1678 | if (!(si->flags & SWP_USED) || !si->swap_map) | |
1da177e4 | 1679 | continue; |
881e4aab | 1680 | if (!--l) |
efa90a98 | 1681 | return si; |
1da177e4 LT |
1682 | } |
1683 | ||
1684 | return NULL; | |
1685 | } | |
1686 | ||
1687 | static void *swap_next(struct seq_file *swap, void *v, loff_t *pos) | |
1688 | { | |
efa90a98 HD |
1689 | struct swap_info_struct *si = v; |
1690 | int type; | |
1da177e4 | 1691 | |
881e4aab | 1692 | if (v == SEQ_START_TOKEN) |
efa90a98 HD |
1693 | type = 0; |
1694 | else | |
1695 | type = si->type + 1; | |
881e4aab | 1696 | |
efa90a98 HD |
1697 | for (; type < nr_swapfiles; type++) { |
1698 | smp_rmb(); /* read nr_swapfiles before swap_info[type] */ | |
1699 | si = swap_info[type]; | |
1700 | if (!(si->flags & SWP_USED) || !si->swap_map) | |
1da177e4 LT |
1701 | continue; |
1702 | ++*pos; | |
efa90a98 | 1703 | return si; |
1da177e4 LT |
1704 | } |
1705 | ||
1706 | return NULL; | |
1707 | } | |
1708 | ||
1709 | static void swap_stop(struct seq_file *swap, void *v) | |
1710 | { | |
fc0abb14 | 1711 | mutex_unlock(&swapon_mutex); |
1da177e4 LT |
1712 | } |
1713 | ||
1714 | static int swap_show(struct seq_file *swap, void *v) | |
1715 | { | |
efa90a98 | 1716 | struct swap_info_struct *si = v; |
1da177e4 LT |
1717 | struct file *file; |
1718 | int len; | |
1719 | ||
efa90a98 | 1720 | if (si == SEQ_START_TOKEN) { |
881e4aab SS |
1721 | seq_puts(swap,"Filename\t\t\t\tType\t\tSize\tUsed\tPriority\n"); |
1722 | return 0; | |
1723 | } | |
1da177e4 | 1724 | |
efa90a98 | 1725 | file = si->swap_file; |
c32c2f63 | 1726 | len = seq_path(swap, &file->f_path, " \t\n\\"); |
6eb396dc | 1727 | seq_printf(swap, "%*s%s\t%u\t%u\t%d\n", |
886bb7e9 HD |
1728 | len < 40 ? 40 - len : 1, " ", |
1729 | S_ISBLK(file->f_path.dentry->d_inode->i_mode) ? | |
1da177e4 | 1730 | "partition" : "file\t", |
efa90a98 HD |
1731 | si->pages << (PAGE_SHIFT - 10), |
1732 | si->inuse_pages << (PAGE_SHIFT - 10), | |
1733 | si->prio); | |
1da177e4 LT |
1734 | return 0; |
1735 | } | |
1736 | ||
15ad7cdc | 1737 | static const struct seq_operations swaps_op = { |
1da177e4 LT |
1738 | .start = swap_start, |
1739 | .next = swap_next, | |
1740 | .stop = swap_stop, | |
1741 | .show = swap_show | |
1742 | }; | |
1743 | ||
1744 | static int swaps_open(struct inode *inode, struct file *file) | |
1745 | { | |
1746 | return seq_open(file, &swaps_op); | |
1747 | } | |
1748 | ||
15ad7cdc | 1749 | static const struct file_operations proc_swaps_operations = { |
1da177e4 LT |
1750 | .open = swaps_open, |
1751 | .read = seq_read, | |
1752 | .llseek = seq_lseek, | |
1753 | .release = seq_release, | |
1754 | }; | |
1755 | ||
1756 | static int __init procswaps_init(void) | |
1757 | { | |
3d71f86f | 1758 | proc_create("swaps", 0, NULL, &proc_swaps_operations); |
1da177e4 LT |
1759 | return 0; |
1760 | } | |
1761 | __initcall(procswaps_init); | |
1762 | #endif /* CONFIG_PROC_FS */ | |
1763 | ||
1796316a JB |
1764 | #ifdef MAX_SWAPFILES_CHECK |
1765 | static int __init max_swapfiles_check(void) | |
1766 | { | |
1767 | MAX_SWAPFILES_CHECK(); | |
1768 | return 0; | |
1769 | } | |
1770 | late_initcall(max_swapfiles_check); | |
1771 | #endif | |
1772 | ||
1da177e4 LT |
1773 | /* |
1774 | * Written 01/25/92 by Simmule Turner, heavily changed by Linus. | |
1775 | * | |
1776 | * The swapon system call | |
1777 | */ | |
c4ea37c2 | 1778 | SYSCALL_DEFINE2(swapon, const char __user *, specialfile, int, swap_flags) |
1da177e4 | 1779 | { |
73c34b6a | 1780 | struct swap_info_struct *p; |
1da177e4 LT |
1781 | char *name = NULL; |
1782 | struct block_device *bdev = NULL; | |
1783 | struct file *swap_file = NULL; | |
1784 | struct address_space *mapping; | |
1785 | unsigned int type; | |
1786 | int i, prev; | |
1787 | int error; | |
1da177e4 | 1788 | union swap_header *swap_header = NULL; |
6eb396dc HD |
1789 | unsigned int nr_good_pages = 0; |
1790 | int nr_extents = 0; | |
53092a74 | 1791 | sector_t span; |
1da177e4 | 1792 | unsigned long maxpages = 1; |
73fd8748 | 1793 | unsigned long swapfilepages; |
78ecba08 | 1794 | unsigned short *swap_map = NULL; |
1da177e4 LT |
1795 | struct page *page = NULL; |
1796 | struct inode *inode = NULL; | |
1797 | int did_down = 0; | |
1798 | ||
1799 | if (!capable(CAP_SYS_ADMIN)) | |
1800 | return -EPERM; | |
efa90a98 HD |
1801 | |
1802 | p = kzalloc(sizeof(*p), GFP_KERNEL); | |
1803 | if (!p) | |
1804 | return -ENOMEM; | |
1805 | ||
5d337b91 | 1806 | spin_lock(&swap_lock); |
efa90a98 HD |
1807 | for (type = 0; type < nr_swapfiles; type++) { |
1808 | if (!(swap_info[type]->flags & SWP_USED)) | |
1da177e4 | 1809 | break; |
efa90a98 | 1810 | } |
1da177e4 | 1811 | error = -EPERM; |
0697212a | 1812 | if (type >= MAX_SWAPFILES) { |
5d337b91 | 1813 | spin_unlock(&swap_lock); |
efa90a98 | 1814 | kfree(p); |
1da177e4 LT |
1815 | goto out; |
1816 | } | |
efa90a98 HD |
1817 | if (type >= nr_swapfiles) { |
1818 | p->type = type; | |
1819 | swap_info[type] = p; | |
1820 | /* | |
1821 | * Write swap_info[type] before nr_swapfiles, in case a | |
1822 | * racing procfs swap_start() or swap_next() is reading them. | |
1823 | * (We never shrink nr_swapfiles, we never free this entry.) | |
1824 | */ | |
1825 | smp_wmb(); | |
1826 | nr_swapfiles++; | |
1827 | } else { | |
1828 | kfree(p); | |
1829 | p = swap_info[type]; | |
1830 | /* | |
1831 | * Do not memset this entry: a racing procfs swap_next() | |
1832 | * would be relying on p->type to remain valid. | |
1833 | */ | |
1834 | } | |
9625a5f2 | 1835 | INIT_LIST_HEAD(&p->first_swap_extent.list); |
1da177e4 | 1836 | p->flags = SWP_USED; |
1da177e4 | 1837 | p->next = -1; |
5d337b91 | 1838 | spin_unlock(&swap_lock); |
efa90a98 | 1839 | |
1da177e4 LT |
1840 | name = getname(specialfile); |
1841 | error = PTR_ERR(name); | |
1842 | if (IS_ERR(name)) { | |
1843 | name = NULL; | |
1844 | goto bad_swap_2; | |
1845 | } | |
1846 | swap_file = filp_open(name, O_RDWR|O_LARGEFILE, 0); | |
1847 | error = PTR_ERR(swap_file); | |
1848 | if (IS_ERR(swap_file)) { | |
1849 | swap_file = NULL; | |
1850 | goto bad_swap_2; | |
1851 | } | |
1852 | ||
1853 | p->swap_file = swap_file; | |
1854 | mapping = swap_file->f_mapping; | |
1855 | inode = mapping->host; | |
1856 | ||
1857 | error = -EBUSY; | |
1858 | for (i = 0; i < nr_swapfiles; i++) { | |
efa90a98 | 1859 | struct swap_info_struct *q = swap_info[i]; |
1da177e4 LT |
1860 | |
1861 | if (i == type || !q->swap_file) | |
1862 | continue; | |
1863 | if (mapping == q->swap_file->f_mapping) | |
1864 | goto bad_swap; | |
1865 | } | |
1866 | ||
1867 | error = -EINVAL; | |
1868 | if (S_ISBLK(inode->i_mode)) { | |
1869 | bdev = I_BDEV(inode); | |
1870 | error = bd_claim(bdev, sys_swapon); | |
1871 | if (error < 0) { | |
1872 | bdev = NULL; | |
f7b3a435 | 1873 | error = -EINVAL; |
1da177e4 LT |
1874 | goto bad_swap; |
1875 | } | |
1876 | p->old_block_size = block_size(bdev); | |
1877 | error = set_blocksize(bdev, PAGE_SIZE); | |
1878 | if (error < 0) | |
1879 | goto bad_swap; | |
1880 | p->bdev = bdev; | |
1881 | } else if (S_ISREG(inode->i_mode)) { | |
1882 | p->bdev = inode->i_sb->s_bdev; | |
1b1dcc1b | 1883 | mutex_lock(&inode->i_mutex); |
1da177e4 LT |
1884 | did_down = 1; |
1885 | if (IS_SWAPFILE(inode)) { | |
1886 | error = -EBUSY; | |
1887 | goto bad_swap; | |
1888 | } | |
1889 | } else { | |
1890 | goto bad_swap; | |
1891 | } | |
1892 | ||
73fd8748 | 1893 | swapfilepages = i_size_read(inode) >> PAGE_SHIFT; |
1da177e4 LT |
1894 | |
1895 | /* | |
1896 | * Read the swap header. | |
1897 | */ | |
1898 | if (!mapping->a_ops->readpage) { | |
1899 | error = -EINVAL; | |
1900 | goto bad_swap; | |
1901 | } | |
090d2b18 | 1902 | page = read_mapping_page(mapping, 0, swap_file); |
1da177e4 LT |
1903 | if (IS_ERR(page)) { |
1904 | error = PTR_ERR(page); | |
1905 | goto bad_swap; | |
1906 | } | |
81e33971 | 1907 | swap_header = kmap(page); |
1da177e4 | 1908 | |
81e33971 | 1909 | if (memcmp("SWAPSPACE2", swap_header->magic.magic, 10)) { |
e97a3111 | 1910 | printk(KERN_ERR "Unable to find swap-space signature\n"); |
1da177e4 LT |
1911 | error = -EINVAL; |
1912 | goto bad_swap; | |
1913 | } | |
886bb7e9 | 1914 | |
81e33971 HD |
1915 | /* swap partition endianess hack... */ |
1916 | if (swab32(swap_header->info.version) == 1) { | |
1917 | swab32s(&swap_header->info.version); | |
1918 | swab32s(&swap_header->info.last_page); | |
1919 | swab32s(&swap_header->info.nr_badpages); | |
1920 | for (i = 0; i < swap_header->info.nr_badpages; i++) | |
1921 | swab32s(&swap_header->info.badpages[i]); | |
1922 | } | |
1923 | /* Check the swap header's sub-version */ | |
1924 | if (swap_header->info.version != 1) { | |
1925 | printk(KERN_WARNING | |
1926 | "Unable to handle swap header version %d\n", | |
1927 | swap_header->info.version); | |
1da177e4 LT |
1928 | error = -EINVAL; |
1929 | goto bad_swap; | |
81e33971 | 1930 | } |
1da177e4 | 1931 | |
81e33971 HD |
1932 | p->lowest_bit = 1; |
1933 | p->cluster_next = 1; | |
efa90a98 | 1934 | p->cluster_nr = 0; |
52b7efdb | 1935 | |
81e33971 HD |
1936 | /* |
1937 | * Find out how many pages are allowed for a single swap | |
1938 | * device. There are two limiting factors: 1) the number of | |
1939 | * bits for the swap offset in the swp_entry_t type and | |
1940 | * 2) the number of bits in the a swap pte as defined by | |
1941 | * the different architectures. In order to find the | |
1942 | * largest possible bit mask a swap entry with swap type 0 | |
1943 | * and swap offset ~0UL is created, encoded to a swap pte, | |
1944 | * decoded to a swp_entry_t again and finally the swap | |
1945 | * offset is extracted. This will mask all the bits from | |
1946 | * the initial ~0UL mask that can't be encoded in either | |
1947 | * the swp_entry_t or the architecture definition of a | |
1948 | * swap pte. | |
1949 | */ | |
1950 | maxpages = swp_offset(pte_to_swp_entry( | |
1951 | swp_entry_to_pte(swp_entry(0, ~0UL)))) - 1; | |
1952 | if (maxpages > swap_header->info.last_page) | |
1953 | maxpages = swap_header->info.last_page; | |
1954 | p->highest_bit = maxpages - 1; | |
1da177e4 | 1955 | |
81e33971 HD |
1956 | error = -EINVAL; |
1957 | if (!maxpages) | |
1958 | goto bad_swap; | |
1959 | if (swapfilepages && maxpages > swapfilepages) { | |
1960 | printk(KERN_WARNING | |
1961 | "Swap area shorter than signature indicates\n"); | |
1962 | goto bad_swap; | |
1963 | } | |
1964 | if (swap_header->info.nr_badpages && S_ISREG(inode->i_mode)) | |
1965 | goto bad_swap; | |
1966 | if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES) | |
1967 | goto bad_swap; | |
cd105df4 | 1968 | |
81e33971 HD |
1969 | /* OK, set up the swap map and apply the bad block list */ |
1970 | swap_map = vmalloc(maxpages * sizeof(short)); | |
1971 | if (!swap_map) { | |
1972 | error = -ENOMEM; | |
1973 | goto bad_swap; | |
1974 | } | |
1da177e4 | 1975 | |
81e33971 HD |
1976 | memset(swap_map, 0, maxpages * sizeof(short)); |
1977 | for (i = 0; i < swap_header->info.nr_badpages; i++) { | |
1978 | int page_nr = swap_header->info.badpages[i]; | |
1979 | if (page_nr <= 0 || page_nr >= swap_header->info.last_page) { | |
1980 | error = -EINVAL; | |
1da177e4 | 1981 | goto bad_swap; |
81e33971 HD |
1982 | } |
1983 | swap_map[page_nr] = SWAP_MAP_BAD; | |
1da177e4 | 1984 | } |
27a7faa0 KH |
1985 | |
1986 | error = swap_cgroup_swapon(type, maxpages); | |
1987 | if (error) | |
1988 | goto bad_swap; | |
1989 | ||
81e33971 HD |
1990 | nr_good_pages = swap_header->info.last_page - |
1991 | swap_header->info.nr_badpages - | |
1992 | 1 /* header page */; | |
e2244ec2 | 1993 | |
e2244ec2 | 1994 | if (nr_good_pages) { |
78ecba08 | 1995 | swap_map[0] = SWAP_MAP_BAD; |
e2244ec2 HD |
1996 | p->max = maxpages; |
1997 | p->pages = nr_good_pages; | |
53092a74 HD |
1998 | nr_extents = setup_swap_extents(p, &span); |
1999 | if (nr_extents < 0) { | |
2000 | error = nr_extents; | |
e2244ec2 | 2001 | goto bad_swap; |
53092a74 | 2002 | } |
e2244ec2 HD |
2003 | nr_good_pages = p->pages; |
2004 | } | |
1da177e4 LT |
2005 | if (!nr_good_pages) { |
2006 | printk(KERN_WARNING "Empty swap-file\n"); | |
2007 | error = -EINVAL; | |
2008 | goto bad_swap; | |
2009 | } | |
1da177e4 | 2010 | |
3bd0f0c7 SJ |
2011 | if (p->bdev) { |
2012 | if (blk_queue_nonrot(bdev_get_queue(p->bdev))) { | |
2013 | p->flags |= SWP_SOLIDSTATE; | |
2014 | p->cluster_next = 1 + (random32() % p->highest_bit); | |
2015 | } | |
2016 | if (discard_swap(p) == 0) | |
2017 | p->flags |= SWP_DISCARDABLE; | |
20137a49 | 2018 | } |
6a6ba831 | 2019 | |
fc0abb14 | 2020 | mutex_lock(&swapon_mutex); |
5d337b91 | 2021 | spin_lock(&swap_lock); |
78ecba08 HD |
2022 | if (swap_flags & SWAP_FLAG_PREFER) |
2023 | p->prio = | |
2024 | (swap_flags & SWAP_FLAG_PRIO_MASK) >> SWAP_FLAG_PRIO_SHIFT; | |
2025 | else | |
2026 | p->prio = --least_priority; | |
2027 | p->swap_map = swap_map; | |
22c6f8fd | 2028 | p->flags |= SWP_WRITEOK; |
1da177e4 LT |
2029 | nr_swap_pages += nr_good_pages; |
2030 | total_swap_pages += nr_good_pages; | |
53092a74 | 2031 | |
6eb396dc | 2032 | printk(KERN_INFO "Adding %uk swap on %s. " |
20137a49 | 2033 | "Priority:%d extents:%d across:%lluk %s%s\n", |
53092a74 | 2034 | nr_good_pages<<(PAGE_SHIFT-10), name, p->prio, |
6a6ba831 | 2035 | nr_extents, (unsigned long long)span<<(PAGE_SHIFT-10), |
20137a49 HD |
2036 | (p->flags & SWP_SOLIDSTATE) ? "SS" : "", |
2037 | (p->flags & SWP_DISCARDABLE) ? "D" : ""); | |
1da177e4 LT |
2038 | |
2039 | /* insert swap space into swap_list: */ | |
2040 | prev = -1; | |
efa90a98 HD |
2041 | for (i = swap_list.head; i >= 0; i = swap_info[i]->next) { |
2042 | if (p->prio >= swap_info[i]->prio) | |
1da177e4 | 2043 | break; |
1da177e4 LT |
2044 | prev = i; |
2045 | } | |
2046 | p->next = i; | |
efa90a98 HD |
2047 | if (prev < 0) |
2048 | swap_list.head = swap_list.next = type; | |
2049 | else | |
2050 | swap_info[prev]->next = type; | |
5d337b91 | 2051 | spin_unlock(&swap_lock); |
fc0abb14 | 2052 | mutex_unlock(&swapon_mutex); |
1da177e4 LT |
2053 | error = 0; |
2054 | goto out; | |
2055 | bad_swap: | |
2056 | if (bdev) { | |
2057 | set_blocksize(bdev, p->old_block_size); | |
2058 | bd_release(bdev); | |
2059 | } | |
4cd3bb10 | 2060 | destroy_swap_extents(p); |
27a7faa0 | 2061 | swap_cgroup_swapoff(type); |
1da177e4 | 2062 | bad_swap_2: |
5d337b91 | 2063 | spin_lock(&swap_lock); |
1da177e4 | 2064 | p->swap_file = NULL; |
1da177e4 | 2065 | p->flags = 0; |
5d337b91 | 2066 | spin_unlock(&swap_lock); |
1da177e4 LT |
2067 | vfree(swap_map); |
2068 | if (swap_file) | |
2069 | filp_close(swap_file, NULL); | |
2070 | out: | |
2071 | if (page && !IS_ERR(page)) { | |
2072 | kunmap(page); | |
2073 | page_cache_release(page); | |
2074 | } | |
2075 | if (name) | |
2076 | putname(name); | |
2077 | if (did_down) { | |
2078 | if (!error) | |
2079 | inode->i_flags |= S_SWAPFILE; | |
1b1dcc1b | 2080 | mutex_unlock(&inode->i_mutex); |
1da177e4 LT |
2081 | } |
2082 | return error; | |
2083 | } | |
2084 | ||
2085 | void si_swapinfo(struct sysinfo *val) | |
2086 | { | |
efa90a98 | 2087 | unsigned int type; |
1da177e4 LT |
2088 | unsigned long nr_to_be_unused = 0; |
2089 | ||
5d337b91 | 2090 | spin_lock(&swap_lock); |
efa90a98 HD |
2091 | for (type = 0; type < nr_swapfiles; type++) { |
2092 | struct swap_info_struct *si = swap_info[type]; | |
2093 | ||
2094 | if ((si->flags & SWP_USED) && !(si->flags & SWP_WRITEOK)) | |
2095 | nr_to_be_unused += si->inuse_pages; | |
1da177e4 LT |
2096 | } |
2097 | val->freeswap = nr_swap_pages + nr_to_be_unused; | |
2098 | val->totalswap = total_swap_pages + nr_to_be_unused; | |
5d337b91 | 2099 | spin_unlock(&swap_lock); |
1da177e4 LT |
2100 | } |
2101 | ||
2102 | /* | |
2103 | * Verify that a swap entry is valid and increment its swap map count. | |
2104 | * | |
2105 | * Note: if swap_map[] reaches SWAP_MAP_MAX the entries are treated as | |
2106 | * "permanent", but will be reclaimed by the next swapoff. | |
355cfa73 KH |
2107 | * Returns error code in following case. |
2108 | * - success -> 0 | |
2109 | * - swp_entry is invalid -> EINVAL | |
2110 | * - swp_entry is migration entry -> EINVAL | |
2111 | * - swap-cache reference is requested but there is already one. -> EEXIST | |
2112 | * - swap-cache reference is requested but the entry is not used. -> ENOENT | |
1da177e4 | 2113 | */ |
355cfa73 | 2114 | static int __swap_duplicate(swp_entry_t entry, bool cache) |
1da177e4 | 2115 | { |
73c34b6a | 2116 | struct swap_info_struct *p; |
1da177e4 | 2117 | unsigned long offset, type; |
355cfa73 KH |
2118 | int result = -EINVAL; |
2119 | int count; | |
2120 | bool has_cache; | |
1da177e4 | 2121 | |
a7420aa5 | 2122 | if (non_swap_entry(entry)) |
355cfa73 | 2123 | return -EINVAL; |
0697212a | 2124 | |
1da177e4 LT |
2125 | type = swp_type(entry); |
2126 | if (type >= nr_swapfiles) | |
2127 | goto bad_file; | |
efa90a98 | 2128 | p = swap_info[type]; |
1da177e4 LT |
2129 | offset = swp_offset(entry); |
2130 | ||
5d337b91 | 2131 | spin_lock(&swap_lock); |
355cfa73 KH |
2132 | |
2133 | if (unlikely(offset >= p->max)) | |
2134 | goto unlock_out; | |
2135 | ||
2136 | count = swap_count(p->swap_map[offset]); | |
2137 | has_cache = swap_has_cache(p->swap_map[offset]); | |
2138 | ||
2139 | if (cache == SWAP_CACHE) { /* called for swapcache/swapin-readahead */ | |
2140 | ||
2141 | /* set SWAP_HAS_CACHE if there is no cache and entry is used */ | |
2142 | if (!has_cache && count) { | |
2143 | p->swap_map[offset] = encode_swapmap(count, true); | |
2144 | result = 0; | |
2145 | } else if (has_cache) /* someone added cache */ | |
2146 | result = -EEXIST; | |
2147 | else if (!count) /* no users */ | |
2148 | result = -ENOENT; | |
2149 | ||
2150 | } else if (count || has_cache) { | |
2151 | if (count < SWAP_MAP_MAX - 1) { | |
2152 | p->swap_map[offset] = encode_swapmap(count + 1, | |
2153 | has_cache); | |
2154 | result = 0; | |
2155 | } else if (count <= SWAP_MAP_MAX) { | |
1da177e4 | 2156 | if (swap_overflow++ < 5) |
355cfa73 KH |
2157 | printk(KERN_WARNING |
2158 | "swap_dup: swap entry overflow\n"); | |
2159 | p->swap_map[offset] = encode_swapmap(SWAP_MAP_MAX, | |
2160 | has_cache); | |
2161 | result = 0; | |
1da177e4 | 2162 | } |
355cfa73 KH |
2163 | } else |
2164 | result = -ENOENT; /* unused swap entry */ | |
2165 | unlock_out: | |
5d337b91 | 2166 | spin_unlock(&swap_lock); |
1da177e4 LT |
2167 | out: |
2168 | return result; | |
2169 | ||
2170 | bad_file: | |
2171 | printk(KERN_ERR "swap_dup: %s%08lx\n", Bad_file, entry.val); | |
2172 | goto out; | |
2173 | } | |
355cfa73 KH |
2174 | /* |
2175 | * increase reference count of swap entry by 1. | |
2176 | */ | |
2177 | void swap_duplicate(swp_entry_t entry) | |
2178 | { | |
2179 | __swap_duplicate(entry, SWAP_MAP); | |
2180 | } | |
1da177e4 | 2181 | |
cb4b86ba | 2182 | /* |
355cfa73 KH |
2183 | * @entry: swap entry for which we allocate swap cache. |
2184 | * | |
73c34b6a | 2185 | * Called when allocating swap cache for existing swap entry, |
355cfa73 KH |
2186 | * This can return error codes. Returns 0 at success. |
2187 | * -EBUSY means there is a swap cache. | |
2188 | * Note: return code is different from swap_duplicate(). | |
cb4b86ba KH |
2189 | */ |
2190 | int swapcache_prepare(swp_entry_t entry) | |
2191 | { | |
355cfa73 | 2192 | return __swap_duplicate(entry, SWAP_CACHE); |
cb4b86ba KH |
2193 | } |
2194 | ||
1da177e4 | 2195 | /* |
5d337b91 | 2196 | * swap_lock prevents swap_map being freed. Don't grab an extra |
1da177e4 LT |
2197 | * reference on the swaphandle, it doesn't matter if it becomes unused. |
2198 | */ | |
2199 | int valid_swaphandles(swp_entry_t entry, unsigned long *offset) | |
2200 | { | |
8952898b | 2201 | struct swap_info_struct *si; |
3f9e7949 | 2202 | int our_page_cluster = page_cluster; |
8952898b HD |
2203 | pgoff_t target, toff; |
2204 | pgoff_t base, end; | |
2205 | int nr_pages = 0; | |
1da177e4 | 2206 | |
3f9e7949 | 2207 | if (!our_page_cluster) /* no readahead */ |
1da177e4 | 2208 | return 0; |
8952898b | 2209 | |
efa90a98 | 2210 | si = swap_info[swp_type(entry)]; |
8952898b HD |
2211 | target = swp_offset(entry); |
2212 | base = (target >> our_page_cluster) << our_page_cluster; | |
2213 | end = base + (1 << our_page_cluster); | |
2214 | if (!base) /* first page is swap header */ | |
2215 | base++; | |
1da177e4 | 2216 | |
5d337b91 | 2217 | spin_lock(&swap_lock); |
8952898b HD |
2218 | if (end > si->max) /* don't go beyond end of map */ |
2219 | end = si->max; | |
2220 | ||
2221 | /* Count contiguous allocated slots above our target */ | |
2222 | for (toff = target; ++toff < end; nr_pages++) { | |
2223 | /* Don't read in free or bad pages */ | |
2224 | if (!si->swap_map[toff]) | |
2225 | break; | |
355cfa73 | 2226 | if (swap_count(si->swap_map[toff]) == SWAP_MAP_BAD) |
1da177e4 | 2227 | break; |
8952898b HD |
2228 | } |
2229 | /* Count contiguous allocated slots below our target */ | |
2230 | for (toff = target; --toff >= base; nr_pages++) { | |
1da177e4 | 2231 | /* Don't read in free or bad pages */ |
8952898b | 2232 | if (!si->swap_map[toff]) |
1da177e4 | 2233 | break; |
355cfa73 | 2234 | if (swap_count(si->swap_map[toff]) == SWAP_MAP_BAD) |
1da177e4 | 2235 | break; |
8952898b | 2236 | } |
5d337b91 | 2237 | spin_unlock(&swap_lock); |
8952898b HD |
2238 | |
2239 | /* | |
2240 | * Indicate starting offset, and return number of pages to get: | |
2241 | * if only 1, say 0, since there's then no readahead to be done. | |
2242 | */ | |
2243 | *offset = ++toff; | |
2244 | return nr_pages? ++nr_pages: 0; | |
1da177e4 | 2245 | } |