]>
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 | 8 | #include <linux/mm.h> |
6e84f315 | 9 | #include <linux/sched/mm.h> |
29930025 | 10 | #include <linux/sched/task.h> |
1da177e4 LT |
11 | #include <linux/hugetlb.h> |
12 | #include <linux/mman.h> | |
13 | #include <linux/slab.h> | |
14 | #include <linux/kernel_stat.h> | |
15 | #include <linux/swap.h> | |
16 | #include <linux/vmalloc.h> | |
17 | #include <linux/pagemap.h> | |
18 | #include <linux/namei.h> | |
072441e2 | 19 | #include <linux/shmem_fs.h> |
1da177e4 | 20 | #include <linux/blkdev.h> |
20137a49 | 21 | #include <linux/random.h> |
1da177e4 LT |
22 | #include <linux/writeback.h> |
23 | #include <linux/proc_fs.h> | |
24 | #include <linux/seq_file.h> | |
25 | #include <linux/init.h> | |
5ad64688 | 26 | #include <linux/ksm.h> |
1da177e4 LT |
27 | #include <linux/rmap.h> |
28 | #include <linux/security.h> | |
29 | #include <linux/backing-dev.h> | |
fc0abb14 | 30 | #include <linux/mutex.h> |
c59ede7b | 31 | #include <linux/capability.h> |
1da177e4 | 32 | #include <linux/syscalls.h> |
8a9f3ccd | 33 | #include <linux/memcontrol.h> |
66d7dd51 | 34 | #include <linux/poll.h> |
72788c38 | 35 | #include <linux/oom.h> |
38b5faf4 DM |
36 | #include <linux/frontswap.h> |
37 | #include <linux/swapfile.h> | |
f981c595 | 38 | #include <linux/export.h> |
67afa38e | 39 | #include <linux/swap_slots.h> |
155b5f88 | 40 | #include <linux/sort.h> |
1da177e4 LT |
41 | |
42 | #include <asm/pgtable.h> | |
43 | #include <asm/tlbflush.h> | |
44 | #include <linux/swapops.h> | |
5d1ea48b | 45 | #include <linux/swap_cgroup.h> |
1da177e4 | 46 | |
570a335b HD |
47 | static bool swap_count_continued(struct swap_info_struct *, pgoff_t, |
48 | unsigned char); | |
49 | static void free_swap_count_continuations(struct swap_info_struct *); | |
d4906e1a | 50 | static sector_t map_swap_entry(swp_entry_t, struct block_device**); |
570a335b | 51 | |
38b5faf4 | 52 | DEFINE_SPINLOCK(swap_lock); |
7c363b8c | 53 | static unsigned int nr_swapfiles; |
ec8acf20 | 54 | atomic_long_t nr_swap_pages; |
fb0fec50 CW |
55 | /* |
56 | * Some modules use swappable objects and may try to swap them out under | |
57 | * memory pressure (via the shrinker). Before doing so, they may wish to | |
58 | * check to see if any swap space is available. | |
59 | */ | |
60 | EXPORT_SYMBOL_GPL(nr_swap_pages); | |
ec8acf20 | 61 | /* protected with swap_lock. reading in vm_swap_full() doesn't need lock */ |
1da177e4 | 62 | long total_swap_pages; |
a2468cc9 | 63 | static int least_priority = -1; |
1da177e4 | 64 | |
1da177e4 LT |
65 | static const char Bad_file[] = "Bad swap file entry "; |
66 | static const char Unused_file[] = "Unused swap file entry "; | |
67 | static const char Bad_offset[] = "Bad swap offset entry "; | |
68 | static const char Unused_offset[] = "Unused swap offset entry "; | |
69 | ||
adfab836 DS |
70 | /* |
71 | * all active swap_info_structs | |
72 | * protected with swap_lock, and ordered by priority. | |
73 | */ | |
18ab4d4c DS |
74 | PLIST_HEAD(swap_active_head); |
75 | ||
76 | /* | |
77 | * all available (active, not full) swap_info_structs | |
78 | * protected with swap_avail_lock, ordered by priority. | |
79 | * This is used by get_swap_page() instead of swap_active_head | |
80 | * because swap_active_head includes all swap_info_structs, | |
81 | * but get_swap_page() doesn't need to look at full ones. | |
82 | * This uses its own lock instead of swap_lock because when a | |
83 | * swap_info_struct changes between not-full/full, it needs to | |
84 | * add/remove itself to/from this list, but the swap_info_struct->lock | |
85 | * is held and the locking order requires swap_lock to be taken | |
86 | * before any swap_info_struct->lock. | |
87 | */ | |
a2468cc9 | 88 | struct plist_head *swap_avail_heads; |
18ab4d4c | 89 | static DEFINE_SPINLOCK(swap_avail_lock); |
1da177e4 | 90 | |
38b5faf4 | 91 | struct swap_info_struct *swap_info[MAX_SWAPFILES]; |
1da177e4 | 92 | |
fc0abb14 | 93 | static DEFINE_MUTEX(swapon_mutex); |
1da177e4 | 94 | |
66d7dd51 KS |
95 | static DECLARE_WAIT_QUEUE_HEAD(proc_poll_wait); |
96 | /* Activity counter to indicate that a swapon or swapoff has occurred */ | |
97 | static atomic_t proc_poll_event = ATOMIC_INIT(0); | |
98 | ||
81a0298b HY |
99 | atomic_t nr_rotate_swap = ATOMIC_INIT(0); |
100 | ||
27a34b2d DJ |
101 | static struct swap_info_struct *swap_type_to_swap_info(int type) |
102 | { | |
103 | if (type >= READ_ONCE(nr_swapfiles)) | |
104 | return NULL; | |
105 | ||
106 | smp_rmb(); /* Pairs with smp_wmb in alloc_swap_info. */ | |
107 | return READ_ONCE(swap_info[type]); | |
108 | } | |
109 | ||
8d69aaee | 110 | static inline unsigned char swap_count(unsigned char ent) |
355cfa73 | 111 | { |
570a335b | 112 | return ent & ~SWAP_HAS_CACHE; /* may include SWAP_HAS_CONT flag */ |
355cfa73 KH |
113 | } |
114 | ||
efa90a98 | 115 | /* returns 1 if swap entry is freed */ |
c9e44410 KH |
116 | static int |
117 | __try_to_reclaim_swap(struct swap_info_struct *si, unsigned long offset) | |
118 | { | |
efa90a98 | 119 | swp_entry_t entry = swp_entry(si->type, offset); |
c9e44410 KH |
120 | struct page *page; |
121 | int ret = 0; | |
122 | ||
f6ab1f7f | 123 | page = find_get_page(swap_address_space(entry), swp_offset(entry)); |
c9e44410 KH |
124 | if (!page) |
125 | return 0; | |
126 | /* | |
127 | * This function is called from scan_swap_map() and it's called | |
128 | * by vmscan.c at reclaiming pages. So, we hold a lock on a page, here. | |
129 | * We have to use trylock for avoiding deadlock. This is a special | |
130 | * case and you should use try_to_free_swap() with explicit lock_page() | |
131 | * in usual operations. | |
132 | */ | |
133 | if (trylock_page(page)) { | |
134 | ret = try_to_free_swap(page); | |
135 | unlock_page(page); | |
136 | } | |
09cbfeaf | 137 | put_page(page); |
c9e44410 KH |
138 | return ret; |
139 | } | |
355cfa73 | 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, | |
dd3932ed | 158 | nr_blocks, GFP_KERNEL, 0); |
9625a5f2 HD |
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, | |
dd3932ed | 169 | nr_blocks, GFP_KERNEL, 0); |
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) { | |
7992fde7 HD |
189 | if (se->start_page <= start_page && |
190 | start_page < se->start_page + se->nr_pages) { | |
191 | pgoff_t offset = start_page - se->start_page; | |
192 | sector_t start_block = se->start_block + offset; | |
858a2990 | 193 | sector_t nr_blocks = se->nr_pages - offset; |
7992fde7 HD |
194 | |
195 | if (nr_blocks > nr_pages) | |
196 | nr_blocks = nr_pages; | |
197 | start_page += nr_blocks; | |
198 | nr_pages -= nr_blocks; | |
199 | ||
200 | if (!found_extent++) | |
201 | si->curr_swap_extent = se; | |
202 | ||
203 | start_block <<= PAGE_SHIFT - 9; | |
204 | nr_blocks <<= PAGE_SHIFT - 9; | |
205 | if (blkdev_issue_discard(si->bdev, start_block, | |
dd3932ed | 206 | nr_blocks, GFP_NOIO, 0)) |
7992fde7 HD |
207 | break; |
208 | } | |
209 | ||
a8ae4991 | 210 | se = list_next_entry(se, list); |
7992fde7 HD |
211 | } |
212 | } | |
213 | ||
38d8b4e6 HY |
214 | #ifdef CONFIG_THP_SWAP |
215 | #define SWAPFILE_CLUSTER HPAGE_PMD_NR | |
216 | #else | |
048c27fd | 217 | #define SWAPFILE_CLUSTER 256 |
38d8b4e6 | 218 | #endif |
048c27fd HD |
219 | #define LATENCY_LIMIT 256 |
220 | ||
2a8f9449 SL |
221 | static inline void cluster_set_flag(struct swap_cluster_info *info, |
222 | unsigned int flag) | |
223 | { | |
224 | info->flags = flag; | |
225 | } | |
226 | ||
227 | static inline unsigned int cluster_count(struct swap_cluster_info *info) | |
228 | { | |
229 | return info->data; | |
230 | } | |
231 | ||
232 | static inline void cluster_set_count(struct swap_cluster_info *info, | |
233 | unsigned int c) | |
234 | { | |
235 | info->data = c; | |
236 | } | |
237 | ||
238 | static inline void cluster_set_count_flag(struct swap_cluster_info *info, | |
239 | unsigned int c, unsigned int f) | |
240 | { | |
241 | info->flags = f; | |
242 | info->data = c; | |
243 | } | |
244 | ||
245 | static inline unsigned int cluster_next(struct swap_cluster_info *info) | |
246 | { | |
247 | return info->data; | |
248 | } | |
249 | ||
250 | static inline void cluster_set_next(struct swap_cluster_info *info, | |
251 | unsigned int n) | |
252 | { | |
253 | info->data = n; | |
254 | } | |
255 | ||
256 | static inline void cluster_set_next_flag(struct swap_cluster_info *info, | |
257 | unsigned int n, unsigned int f) | |
258 | { | |
259 | info->flags = f; | |
260 | info->data = n; | |
261 | } | |
262 | ||
263 | static inline bool cluster_is_free(struct swap_cluster_info *info) | |
264 | { | |
265 | return info->flags & CLUSTER_FLAG_FREE; | |
266 | } | |
267 | ||
268 | static inline bool cluster_is_null(struct swap_cluster_info *info) | |
269 | { | |
270 | return info->flags & CLUSTER_FLAG_NEXT_NULL; | |
271 | } | |
272 | ||
273 | static inline void cluster_set_null(struct swap_cluster_info *info) | |
274 | { | |
275 | info->flags = CLUSTER_FLAG_NEXT_NULL; | |
276 | info->data = 0; | |
277 | } | |
278 | ||
e0709829 HY |
279 | static inline bool cluster_is_huge(struct swap_cluster_info *info) |
280 | { | |
281 | return info->flags & CLUSTER_FLAG_HUGE; | |
282 | } | |
283 | ||
284 | static inline void cluster_clear_huge(struct swap_cluster_info *info) | |
285 | { | |
286 | info->flags &= ~CLUSTER_FLAG_HUGE; | |
287 | } | |
288 | ||
235b6217 HY |
289 | static inline struct swap_cluster_info *lock_cluster(struct swap_info_struct *si, |
290 | unsigned long offset) | |
291 | { | |
292 | struct swap_cluster_info *ci; | |
293 | ||
294 | ci = si->cluster_info; | |
295 | if (ci) { | |
296 | ci += offset / SWAPFILE_CLUSTER; | |
297 | spin_lock(&ci->lock); | |
298 | } | |
299 | return ci; | |
300 | } | |
301 | ||
302 | static inline void unlock_cluster(struct swap_cluster_info *ci) | |
303 | { | |
304 | if (ci) | |
305 | spin_unlock(&ci->lock); | |
306 | } | |
307 | ||
308 | static inline struct swap_cluster_info *lock_cluster_or_swap_info( | |
309 | struct swap_info_struct *si, | |
310 | unsigned long offset) | |
311 | { | |
312 | struct swap_cluster_info *ci; | |
313 | ||
314 | ci = lock_cluster(si, offset); | |
315 | if (!ci) | |
316 | spin_lock(&si->lock); | |
317 | ||
318 | return ci; | |
319 | } | |
320 | ||
321 | static inline void unlock_cluster_or_swap_info(struct swap_info_struct *si, | |
322 | struct swap_cluster_info *ci) | |
323 | { | |
324 | if (ci) | |
325 | unlock_cluster(ci); | |
326 | else | |
327 | spin_unlock(&si->lock); | |
328 | } | |
329 | ||
6b534915 HY |
330 | static inline bool cluster_list_empty(struct swap_cluster_list *list) |
331 | { | |
332 | return cluster_is_null(&list->head); | |
333 | } | |
334 | ||
335 | static inline unsigned int cluster_list_first(struct swap_cluster_list *list) | |
336 | { | |
337 | return cluster_next(&list->head); | |
338 | } | |
339 | ||
340 | static void cluster_list_init(struct swap_cluster_list *list) | |
341 | { | |
342 | cluster_set_null(&list->head); | |
343 | cluster_set_null(&list->tail); | |
344 | } | |
345 | ||
346 | static void cluster_list_add_tail(struct swap_cluster_list *list, | |
347 | struct swap_cluster_info *ci, | |
348 | unsigned int idx) | |
349 | { | |
350 | if (cluster_list_empty(list)) { | |
351 | cluster_set_next_flag(&list->head, idx, 0); | |
352 | cluster_set_next_flag(&list->tail, idx, 0); | |
353 | } else { | |
235b6217 | 354 | struct swap_cluster_info *ci_tail; |
6b534915 HY |
355 | unsigned int tail = cluster_next(&list->tail); |
356 | ||
235b6217 HY |
357 | /* |
358 | * Nested cluster lock, but both cluster locks are | |
359 | * only acquired when we held swap_info_struct->lock | |
360 | */ | |
361 | ci_tail = ci + tail; | |
362 | spin_lock_nested(&ci_tail->lock, SINGLE_DEPTH_NESTING); | |
363 | cluster_set_next(ci_tail, idx); | |
0ef017d1 | 364 | spin_unlock(&ci_tail->lock); |
6b534915 HY |
365 | cluster_set_next_flag(&list->tail, idx, 0); |
366 | } | |
367 | } | |
368 | ||
369 | static unsigned int cluster_list_del_first(struct swap_cluster_list *list, | |
370 | struct swap_cluster_info *ci) | |
371 | { | |
372 | unsigned int idx; | |
373 | ||
374 | idx = cluster_next(&list->head); | |
375 | if (cluster_next(&list->tail) == idx) { | |
376 | cluster_set_null(&list->head); | |
377 | cluster_set_null(&list->tail); | |
378 | } else | |
379 | cluster_set_next_flag(&list->head, | |
380 | cluster_next(&ci[idx]), 0); | |
381 | ||
382 | return idx; | |
383 | } | |
384 | ||
815c2c54 SL |
385 | /* Add a cluster to discard list and schedule it to do discard */ |
386 | static void swap_cluster_schedule_discard(struct swap_info_struct *si, | |
387 | unsigned int idx) | |
388 | { | |
389 | /* | |
390 | * If scan_swap_map() can't find a free cluster, it will check | |
391 | * si->swap_map directly. To make sure the discarding cluster isn't | |
392 | * taken by scan_swap_map(), mark the swap entries bad (occupied). It | |
393 | * will be cleared after discard | |
394 | */ | |
395 | memset(si->swap_map + idx * SWAPFILE_CLUSTER, | |
396 | SWAP_MAP_BAD, SWAPFILE_CLUSTER); | |
397 | ||
6b534915 | 398 | cluster_list_add_tail(&si->discard_clusters, si->cluster_info, idx); |
815c2c54 SL |
399 | |
400 | schedule_work(&si->discard_work); | |
401 | } | |
402 | ||
38d8b4e6 HY |
403 | static void __free_cluster(struct swap_info_struct *si, unsigned long idx) |
404 | { | |
405 | struct swap_cluster_info *ci = si->cluster_info; | |
406 | ||
407 | cluster_set_flag(ci + idx, CLUSTER_FLAG_FREE); | |
408 | cluster_list_add_tail(&si->free_clusters, ci, idx); | |
409 | } | |
410 | ||
815c2c54 SL |
411 | /* |
412 | * Doing discard actually. After a cluster discard is finished, the cluster | |
413 | * will be added to free cluster list. caller should hold si->lock. | |
414 | */ | |
415 | static void swap_do_scheduled_discard(struct swap_info_struct *si) | |
416 | { | |
235b6217 | 417 | struct swap_cluster_info *info, *ci; |
815c2c54 SL |
418 | unsigned int idx; |
419 | ||
420 | info = si->cluster_info; | |
421 | ||
6b534915 HY |
422 | while (!cluster_list_empty(&si->discard_clusters)) { |
423 | idx = cluster_list_del_first(&si->discard_clusters, info); | |
815c2c54 SL |
424 | spin_unlock(&si->lock); |
425 | ||
426 | discard_swap_cluster(si, idx * SWAPFILE_CLUSTER, | |
427 | SWAPFILE_CLUSTER); | |
428 | ||
429 | spin_lock(&si->lock); | |
235b6217 | 430 | ci = lock_cluster(si, idx * SWAPFILE_CLUSTER); |
38d8b4e6 | 431 | __free_cluster(si, idx); |
815c2c54 SL |
432 | memset(si->swap_map + idx * SWAPFILE_CLUSTER, |
433 | 0, SWAPFILE_CLUSTER); | |
235b6217 | 434 | unlock_cluster(ci); |
815c2c54 SL |
435 | } |
436 | } | |
437 | ||
438 | static void swap_discard_work(struct work_struct *work) | |
439 | { | |
440 | struct swap_info_struct *si; | |
441 | ||
442 | si = container_of(work, struct swap_info_struct, discard_work); | |
443 | ||
444 | spin_lock(&si->lock); | |
445 | swap_do_scheduled_discard(si); | |
446 | spin_unlock(&si->lock); | |
447 | } | |
448 | ||
38d8b4e6 HY |
449 | static void alloc_cluster(struct swap_info_struct *si, unsigned long idx) |
450 | { | |
451 | struct swap_cluster_info *ci = si->cluster_info; | |
452 | ||
453 | VM_BUG_ON(cluster_list_first(&si->free_clusters) != idx); | |
454 | cluster_list_del_first(&si->free_clusters, ci); | |
455 | cluster_set_count_flag(ci + idx, 0, 0); | |
456 | } | |
457 | ||
458 | static void free_cluster(struct swap_info_struct *si, unsigned long idx) | |
459 | { | |
460 | struct swap_cluster_info *ci = si->cluster_info + idx; | |
461 | ||
462 | VM_BUG_ON(cluster_count(ci) != 0); | |
463 | /* | |
464 | * If the swap is discardable, prepare discard the cluster | |
465 | * instead of free it immediately. The cluster will be freed | |
466 | * after discard. | |
467 | */ | |
468 | if ((si->flags & (SWP_WRITEOK | SWP_PAGE_DISCARD)) == | |
469 | (SWP_WRITEOK | SWP_PAGE_DISCARD)) { | |
470 | swap_cluster_schedule_discard(si, idx); | |
471 | return; | |
472 | } | |
473 | ||
474 | __free_cluster(si, idx); | |
475 | } | |
476 | ||
2a8f9449 SL |
477 | /* |
478 | * The cluster corresponding to page_nr will be used. The cluster will be | |
479 | * removed from free cluster list and its usage counter will be increased. | |
480 | */ | |
481 | static void inc_cluster_info_page(struct swap_info_struct *p, | |
482 | struct swap_cluster_info *cluster_info, unsigned long page_nr) | |
483 | { | |
484 | unsigned long idx = page_nr / SWAPFILE_CLUSTER; | |
485 | ||
486 | if (!cluster_info) | |
487 | return; | |
38d8b4e6 HY |
488 | if (cluster_is_free(&cluster_info[idx])) |
489 | alloc_cluster(p, idx); | |
2a8f9449 SL |
490 | |
491 | VM_BUG_ON(cluster_count(&cluster_info[idx]) >= SWAPFILE_CLUSTER); | |
492 | cluster_set_count(&cluster_info[idx], | |
493 | cluster_count(&cluster_info[idx]) + 1); | |
494 | } | |
495 | ||
496 | /* | |
497 | * The cluster corresponding to page_nr decreases one usage. If the usage | |
498 | * counter becomes 0, which means no page in the cluster is in using, we can | |
499 | * optionally discard the cluster and add it to free cluster list. | |
500 | */ | |
501 | static void dec_cluster_info_page(struct swap_info_struct *p, | |
502 | struct swap_cluster_info *cluster_info, unsigned long page_nr) | |
503 | { | |
504 | unsigned long idx = page_nr / SWAPFILE_CLUSTER; | |
505 | ||
506 | if (!cluster_info) | |
507 | return; | |
508 | ||
509 | VM_BUG_ON(cluster_count(&cluster_info[idx]) == 0); | |
510 | cluster_set_count(&cluster_info[idx], | |
511 | cluster_count(&cluster_info[idx]) - 1); | |
512 | ||
38d8b4e6 HY |
513 | if (cluster_count(&cluster_info[idx]) == 0) |
514 | free_cluster(p, idx); | |
2a8f9449 SL |
515 | } |
516 | ||
517 | /* | |
518 | * It's possible scan_swap_map() uses a free cluster in the middle of free | |
519 | * cluster list. Avoiding such abuse to avoid list corruption. | |
520 | */ | |
ebc2a1a6 SL |
521 | static bool |
522 | scan_swap_map_ssd_cluster_conflict(struct swap_info_struct *si, | |
2a8f9449 SL |
523 | unsigned long offset) |
524 | { | |
ebc2a1a6 SL |
525 | struct percpu_cluster *percpu_cluster; |
526 | bool conflict; | |
527 | ||
2a8f9449 | 528 | offset /= SWAPFILE_CLUSTER; |
6b534915 HY |
529 | conflict = !cluster_list_empty(&si->free_clusters) && |
530 | offset != cluster_list_first(&si->free_clusters) && | |
2a8f9449 | 531 | cluster_is_free(&si->cluster_info[offset]); |
ebc2a1a6 SL |
532 | |
533 | if (!conflict) | |
534 | return false; | |
535 | ||
536 | percpu_cluster = this_cpu_ptr(si->percpu_cluster); | |
537 | cluster_set_null(&percpu_cluster->index); | |
538 | return true; | |
539 | } | |
540 | ||
541 | /* | |
542 | * Try to get a swap entry from current cpu's swap entry pool (a cluster). This | |
543 | * might involve allocating a new cluster for current CPU too. | |
544 | */ | |
36005bae | 545 | static bool scan_swap_map_try_ssd_cluster(struct swap_info_struct *si, |
ebc2a1a6 SL |
546 | unsigned long *offset, unsigned long *scan_base) |
547 | { | |
548 | struct percpu_cluster *cluster; | |
235b6217 | 549 | struct swap_cluster_info *ci; |
ebc2a1a6 | 550 | bool found_free; |
235b6217 | 551 | unsigned long tmp, max; |
ebc2a1a6 SL |
552 | |
553 | new_cluster: | |
554 | cluster = this_cpu_ptr(si->percpu_cluster); | |
555 | if (cluster_is_null(&cluster->index)) { | |
6b534915 HY |
556 | if (!cluster_list_empty(&si->free_clusters)) { |
557 | cluster->index = si->free_clusters.head; | |
ebc2a1a6 SL |
558 | cluster->next = cluster_next(&cluster->index) * |
559 | SWAPFILE_CLUSTER; | |
6b534915 | 560 | } else if (!cluster_list_empty(&si->discard_clusters)) { |
ebc2a1a6 SL |
561 | /* |
562 | * we don't have free cluster but have some clusters in | |
563 | * discarding, do discard now and reclaim them | |
564 | */ | |
565 | swap_do_scheduled_discard(si); | |
566 | *scan_base = *offset = si->cluster_next; | |
567 | goto new_cluster; | |
568 | } else | |
36005bae | 569 | return false; |
ebc2a1a6 SL |
570 | } |
571 | ||
572 | found_free = false; | |
573 | ||
574 | /* | |
575 | * Other CPUs can use our cluster if they can't find a free cluster, | |
576 | * check if there is still free entry in the cluster | |
577 | */ | |
578 | tmp = cluster->next; | |
235b6217 HY |
579 | max = min_t(unsigned long, si->max, |
580 | (cluster_next(&cluster->index) + 1) * SWAPFILE_CLUSTER); | |
581 | if (tmp >= max) { | |
582 | cluster_set_null(&cluster->index); | |
583 | goto new_cluster; | |
584 | } | |
585 | ci = lock_cluster(si, tmp); | |
586 | while (tmp < max) { | |
ebc2a1a6 SL |
587 | if (!si->swap_map[tmp]) { |
588 | found_free = true; | |
589 | break; | |
590 | } | |
591 | tmp++; | |
592 | } | |
235b6217 | 593 | unlock_cluster(ci); |
ebc2a1a6 SL |
594 | if (!found_free) { |
595 | cluster_set_null(&cluster->index); | |
596 | goto new_cluster; | |
597 | } | |
598 | cluster->next = tmp + 1; | |
599 | *offset = tmp; | |
600 | *scan_base = tmp; | |
36005bae | 601 | return found_free; |
2a8f9449 SL |
602 | } |
603 | ||
a2468cc9 AL |
604 | static void __del_from_avail_list(struct swap_info_struct *p) |
605 | { | |
606 | int nid; | |
607 | ||
608 | for_each_node(nid) | |
609 | plist_del(&p->avail_lists[nid], &swap_avail_heads[nid]); | |
610 | } | |
611 | ||
612 | static void del_from_avail_list(struct swap_info_struct *p) | |
613 | { | |
614 | spin_lock(&swap_avail_lock); | |
615 | __del_from_avail_list(p); | |
616 | spin_unlock(&swap_avail_lock); | |
617 | } | |
618 | ||
38d8b4e6 HY |
619 | static void swap_range_alloc(struct swap_info_struct *si, unsigned long offset, |
620 | unsigned int nr_entries) | |
621 | { | |
622 | unsigned int end = offset + nr_entries - 1; | |
623 | ||
624 | if (offset == si->lowest_bit) | |
625 | si->lowest_bit += nr_entries; | |
626 | if (end == si->highest_bit) | |
627 | si->highest_bit -= nr_entries; | |
628 | si->inuse_pages += nr_entries; | |
629 | if (si->inuse_pages == si->pages) { | |
630 | si->lowest_bit = si->max; | |
631 | si->highest_bit = 0; | |
a2468cc9 | 632 | del_from_avail_list(si); |
38d8b4e6 HY |
633 | } |
634 | } | |
635 | ||
a2468cc9 AL |
636 | static void add_to_avail_list(struct swap_info_struct *p) |
637 | { | |
638 | int nid; | |
639 | ||
640 | spin_lock(&swap_avail_lock); | |
641 | for_each_node(nid) { | |
642 | WARN_ON(!plist_node_empty(&p->avail_lists[nid])); | |
643 | plist_add(&p->avail_lists[nid], &swap_avail_heads[nid]); | |
644 | } | |
645 | spin_unlock(&swap_avail_lock); | |
646 | } | |
647 | ||
38d8b4e6 HY |
648 | static void swap_range_free(struct swap_info_struct *si, unsigned long offset, |
649 | unsigned int nr_entries) | |
650 | { | |
651 | unsigned long end = offset + nr_entries - 1; | |
652 | void (*swap_slot_free_notify)(struct block_device *, unsigned long); | |
653 | ||
654 | if (offset < si->lowest_bit) | |
655 | si->lowest_bit = offset; | |
656 | if (end > si->highest_bit) { | |
657 | bool was_full = !si->highest_bit; | |
658 | ||
659 | si->highest_bit = end; | |
a2468cc9 AL |
660 | if (was_full && (si->flags & SWP_WRITEOK)) |
661 | add_to_avail_list(si); | |
38d8b4e6 HY |
662 | } |
663 | atomic_long_add(nr_entries, &nr_swap_pages); | |
664 | si->inuse_pages -= nr_entries; | |
665 | if (si->flags & SWP_BLKDEV) | |
666 | swap_slot_free_notify = | |
667 | si->bdev->bd_disk->fops->swap_slot_free_notify; | |
668 | else | |
669 | swap_slot_free_notify = NULL; | |
670 | while (offset <= end) { | |
671 | frontswap_invalidate_page(si->type, offset); | |
672 | if (swap_slot_free_notify) | |
673 | swap_slot_free_notify(si->bdev, offset); | |
674 | offset++; | |
675 | } | |
676 | } | |
677 | ||
36005bae TC |
678 | static int scan_swap_map_slots(struct swap_info_struct *si, |
679 | unsigned char usage, int nr, | |
680 | swp_entry_t slots[]) | |
1da177e4 | 681 | { |
235b6217 | 682 | struct swap_cluster_info *ci; |
ebebbbe9 | 683 | unsigned long offset; |
c60aa176 | 684 | unsigned long scan_base; |
7992fde7 | 685 | unsigned long last_in_cluster = 0; |
048c27fd | 686 | int latency_ration = LATENCY_LIMIT; |
36005bae TC |
687 | int n_ret = 0; |
688 | ||
689 | if (nr > SWAP_BATCH) | |
690 | nr = SWAP_BATCH; | |
7dfad418 | 691 | |
886bb7e9 | 692 | /* |
7dfad418 HD |
693 | * We try to cluster swap pages by allocating them sequentially |
694 | * in swap. Once we've allocated SWAPFILE_CLUSTER pages this | |
695 | * way, however, we resort to first-free allocation, starting | |
696 | * a new cluster. This prevents us from scattering swap pages | |
697 | * all over the entire swap partition, so that we reduce | |
698 | * overall disk seek times between swap pages. -- sct | |
699 | * But we do now try to find an empty cluster. -Andrea | |
c60aa176 | 700 | * And we let swap pages go all over an SSD partition. Hugh |
7dfad418 HD |
701 | */ |
702 | ||
52b7efdb | 703 | si->flags += SWP_SCANNING; |
c60aa176 | 704 | scan_base = offset = si->cluster_next; |
ebebbbe9 | 705 | |
ebc2a1a6 SL |
706 | /* SSD algorithm */ |
707 | if (si->cluster_info) { | |
36005bae TC |
708 | if (scan_swap_map_try_ssd_cluster(si, &offset, &scan_base)) |
709 | goto checks; | |
710 | else | |
711 | goto scan; | |
ebc2a1a6 SL |
712 | } |
713 | ||
ebebbbe9 HD |
714 | if (unlikely(!si->cluster_nr--)) { |
715 | if (si->pages - si->inuse_pages < SWAPFILE_CLUSTER) { | |
716 | si->cluster_nr = SWAPFILE_CLUSTER - 1; | |
717 | goto checks; | |
718 | } | |
2a8f9449 | 719 | |
ec8acf20 | 720 | spin_unlock(&si->lock); |
7dfad418 | 721 | |
c60aa176 HD |
722 | /* |
723 | * If seek is expensive, start searching for new cluster from | |
724 | * start of partition, to minimize the span of allocated swap. | |
50088c44 CY |
725 | * If seek is cheap, that is the SWP_SOLIDSTATE si->cluster_info |
726 | * case, just handled by scan_swap_map_try_ssd_cluster() above. | |
c60aa176 | 727 | */ |
50088c44 | 728 | scan_base = offset = si->lowest_bit; |
7dfad418 HD |
729 | last_in_cluster = offset + SWAPFILE_CLUSTER - 1; |
730 | ||
731 | /* Locate the first empty (unaligned) cluster */ | |
732 | for (; last_in_cluster <= si->highest_bit; offset++) { | |
1da177e4 | 733 | if (si->swap_map[offset]) |
7dfad418 HD |
734 | last_in_cluster = offset + SWAPFILE_CLUSTER; |
735 | else if (offset == last_in_cluster) { | |
ec8acf20 | 736 | spin_lock(&si->lock); |
ebebbbe9 HD |
737 | offset -= SWAPFILE_CLUSTER - 1; |
738 | si->cluster_next = offset; | |
739 | si->cluster_nr = SWAPFILE_CLUSTER - 1; | |
c60aa176 HD |
740 | goto checks; |
741 | } | |
742 | if (unlikely(--latency_ration < 0)) { | |
743 | cond_resched(); | |
744 | latency_ration = LATENCY_LIMIT; | |
745 | } | |
746 | } | |
747 | ||
748 | offset = scan_base; | |
ec8acf20 | 749 | spin_lock(&si->lock); |
ebebbbe9 | 750 | si->cluster_nr = SWAPFILE_CLUSTER - 1; |
1da177e4 | 751 | } |
7dfad418 | 752 | |
ebebbbe9 | 753 | checks: |
ebc2a1a6 | 754 | if (si->cluster_info) { |
36005bae TC |
755 | while (scan_swap_map_ssd_cluster_conflict(si, offset)) { |
756 | /* take a break if we already got some slots */ | |
757 | if (n_ret) | |
758 | goto done; | |
759 | if (!scan_swap_map_try_ssd_cluster(si, &offset, | |
760 | &scan_base)) | |
761 | goto scan; | |
762 | } | |
ebc2a1a6 | 763 | } |
ebebbbe9 | 764 | if (!(si->flags & SWP_WRITEOK)) |
52b7efdb | 765 | goto no_page; |
7dfad418 HD |
766 | if (!si->highest_bit) |
767 | goto no_page; | |
ebebbbe9 | 768 | if (offset > si->highest_bit) |
c60aa176 | 769 | scan_base = offset = si->lowest_bit; |
c9e44410 | 770 | |
235b6217 | 771 | ci = lock_cluster(si, offset); |
b73d7fce HD |
772 | /* reuse swap entry of cache-only swap if not busy. */ |
773 | if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) { | |
c9e44410 | 774 | int swap_was_freed; |
235b6217 | 775 | unlock_cluster(ci); |
ec8acf20 | 776 | spin_unlock(&si->lock); |
c9e44410 | 777 | swap_was_freed = __try_to_reclaim_swap(si, offset); |
ec8acf20 | 778 | spin_lock(&si->lock); |
c9e44410 KH |
779 | /* entry was freed successfully, try to use this again */ |
780 | if (swap_was_freed) | |
781 | goto checks; | |
782 | goto scan; /* check next one */ | |
783 | } | |
784 | ||
235b6217 HY |
785 | if (si->swap_map[offset]) { |
786 | unlock_cluster(ci); | |
36005bae TC |
787 | if (!n_ret) |
788 | goto scan; | |
789 | else | |
790 | goto done; | |
235b6217 | 791 | } |
2872bb2d HY |
792 | si->swap_map[offset] = usage; |
793 | inc_cluster_info_page(si, si->cluster_info, offset); | |
794 | unlock_cluster(ci); | |
ebebbbe9 | 795 | |
38d8b4e6 | 796 | swap_range_alloc(si, offset, 1); |
ebebbbe9 | 797 | si->cluster_next = offset + 1; |
36005bae TC |
798 | slots[n_ret++] = swp_entry(si->type, offset); |
799 | ||
800 | /* got enough slots or reach max slots? */ | |
801 | if ((n_ret == nr) || (offset >= si->highest_bit)) | |
802 | goto done; | |
803 | ||
804 | /* search for next available slot */ | |
805 | ||
806 | /* time to take a break? */ | |
807 | if (unlikely(--latency_ration < 0)) { | |
808 | if (n_ret) | |
809 | goto done; | |
810 | spin_unlock(&si->lock); | |
811 | cond_resched(); | |
812 | spin_lock(&si->lock); | |
813 | latency_ration = LATENCY_LIMIT; | |
814 | } | |
815 | ||
816 | /* try to get more slots in cluster */ | |
817 | if (si->cluster_info) { | |
818 | if (scan_swap_map_try_ssd_cluster(si, &offset, &scan_base)) | |
819 | goto checks; | |
820 | else | |
821 | goto done; | |
822 | } | |
823 | /* non-ssd case */ | |
824 | ++offset; | |
825 | ||
826 | /* non-ssd case, still more slots in cluster? */ | |
827 | if (si->cluster_nr && !si->swap_map[offset]) { | |
828 | --si->cluster_nr; | |
829 | goto checks; | |
830 | } | |
7992fde7 | 831 | |
36005bae TC |
832 | done: |
833 | si->flags -= SWP_SCANNING; | |
834 | return n_ret; | |
7dfad418 | 835 | |
ebebbbe9 | 836 | scan: |
ec8acf20 | 837 | spin_unlock(&si->lock); |
7dfad418 | 838 | while (++offset <= si->highest_bit) { |
52b7efdb | 839 | if (!si->swap_map[offset]) { |
ec8acf20 | 840 | spin_lock(&si->lock); |
52b7efdb HD |
841 | goto checks; |
842 | } | |
c9e44410 | 843 | if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) { |
ec8acf20 | 844 | spin_lock(&si->lock); |
c9e44410 KH |
845 | goto checks; |
846 | } | |
048c27fd HD |
847 | if (unlikely(--latency_ration < 0)) { |
848 | cond_resched(); | |
849 | latency_ration = LATENCY_LIMIT; | |
850 | } | |
7dfad418 | 851 | } |
c60aa176 | 852 | offset = si->lowest_bit; |
a5998061 | 853 | while (offset < scan_base) { |
c60aa176 | 854 | if (!si->swap_map[offset]) { |
ec8acf20 | 855 | spin_lock(&si->lock); |
c60aa176 HD |
856 | goto checks; |
857 | } | |
c9e44410 | 858 | if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) { |
ec8acf20 | 859 | spin_lock(&si->lock); |
c9e44410 KH |
860 | goto checks; |
861 | } | |
c60aa176 HD |
862 | if (unlikely(--latency_ration < 0)) { |
863 | cond_resched(); | |
864 | latency_ration = LATENCY_LIMIT; | |
865 | } | |
a5998061 | 866 | offset++; |
c60aa176 | 867 | } |
ec8acf20 | 868 | spin_lock(&si->lock); |
7dfad418 HD |
869 | |
870 | no_page: | |
52b7efdb | 871 | si->flags -= SWP_SCANNING; |
36005bae | 872 | return n_ret; |
1da177e4 LT |
873 | } |
874 | ||
38d8b4e6 HY |
875 | #ifdef CONFIG_THP_SWAP |
876 | static int swap_alloc_cluster(struct swap_info_struct *si, swp_entry_t *slot) | |
877 | { | |
878 | unsigned long idx; | |
879 | struct swap_cluster_info *ci; | |
880 | unsigned long offset, i; | |
881 | unsigned char *map; | |
882 | ||
883 | if (cluster_list_empty(&si->free_clusters)) | |
884 | return 0; | |
885 | ||
886 | idx = cluster_list_first(&si->free_clusters); | |
887 | offset = idx * SWAPFILE_CLUSTER; | |
888 | ci = lock_cluster(si, offset); | |
889 | alloc_cluster(si, idx); | |
e0709829 | 890 | cluster_set_count_flag(ci, SWAPFILE_CLUSTER, CLUSTER_FLAG_HUGE); |
38d8b4e6 HY |
891 | |
892 | map = si->swap_map + offset; | |
893 | for (i = 0; i < SWAPFILE_CLUSTER; i++) | |
894 | map[i] = SWAP_HAS_CACHE; | |
895 | unlock_cluster(ci); | |
896 | swap_range_alloc(si, offset, SWAPFILE_CLUSTER); | |
897 | *slot = swp_entry(si->type, offset); | |
898 | ||
899 | return 1; | |
900 | } | |
901 | ||
902 | static void swap_free_cluster(struct swap_info_struct *si, unsigned long idx) | |
903 | { | |
904 | unsigned long offset = idx * SWAPFILE_CLUSTER; | |
905 | struct swap_cluster_info *ci; | |
906 | ||
907 | ci = lock_cluster(si, offset); | |
908 | cluster_set_count_flag(ci, 0, 0); | |
909 | free_cluster(si, idx); | |
910 | unlock_cluster(ci); | |
911 | swap_range_free(si, offset, SWAPFILE_CLUSTER); | |
912 | } | |
913 | #else | |
914 | static int swap_alloc_cluster(struct swap_info_struct *si, swp_entry_t *slot) | |
915 | { | |
916 | VM_WARN_ON_ONCE(1); | |
917 | return 0; | |
918 | } | |
919 | #endif /* CONFIG_THP_SWAP */ | |
920 | ||
36005bae TC |
921 | static unsigned long scan_swap_map(struct swap_info_struct *si, |
922 | unsigned char usage) | |
923 | { | |
924 | swp_entry_t entry; | |
925 | int n_ret; | |
926 | ||
927 | n_ret = scan_swap_map_slots(si, usage, 1, &entry); | |
928 | ||
929 | if (n_ret) | |
930 | return swp_offset(entry); | |
931 | else | |
932 | return 0; | |
933 | ||
934 | } | |
935 | ||
38d8b4e6 | 936 | int get_swap_pages(int n_goal, bool cluster, swp_entry_t swp_entries[]) |
1da177e4 | 937 | { |
38d8b4e6 | 938 | unsigned long nr_pages = cluster ? SWAPFILE_CLUSTER : 1; |
adfab836 | 939 | struct swap_info_struct *si, *next; |
36005bae TC |
940 | long avail_pgs; |
941 | int n_ret = 0; | |
a2468cc9 | 942 | int node; |
1da177e4 | 943 | |
38d8b4e6 HY |
944 | /* Only single cluster request supported */ |
945 | WARN_ON_ONCE(n_goal > 1 && cluster); | |
946 | ||
947 | avail_pgs = atomic_long_read(&nr_swap_pages) / nr_pages; | |
36005bae | 948 | if (avail_pgs <= 0) |
fb4f88dc | 949 | goto noswap; |
36005bae TC |
950 | |
951 | if (n_goal > SWAP_BATCH) | |
952 | n_goal = SWAP_BATCH; | |
953 | ||
954 | if (n_goal > avail_pgs) | |
955 | n_goal = avail_pgs; | |
956 | ||
38d8b4e6 | 957 | atomic_long_sub(n_goal * nr_pages, &nr_swap_pages); |
fb4f88dc | 958 | |
18ab4d4c DS |
959 | spin_lock(&swap_avail_lock); |
960 | ||
961 | start_over: | |
a2468cc9 AL |
962 | node = numa_node_id(); |
963 | plist_for_each_entry_safe(si, next, &swap_avail_heads[node], avail_lists[node]) { | |
18ab4d4c | 964 | /* requeue si to after same-priority siblings */ |
a2468cc9 | 965 | plist_requeue(&si->avail_lists[node], &swap_avail_heads[node]); |
18ab4d4c | 966 | spin_unlock(&swap_avail_lock); |
ec8acf20 | 967 | spin_lock(&si->lock); |
adfab836 | 968 | if (!si->highest_bit || !(si->flags & SWP_WRITEOK)) { |
18ab4d4c | 969 | spin_lock(&swap_avail_lock); |
a2468cc9 | 970 | if (plist_node_empty(&si->avail_lists[node])) { |
18ab4d4c DS |
971 | spin_unlock(&si->lock); |
972 | goto nextsi; | |
973 | } | |
974 | WARN(!si->highest_bit, | |
975 | "swap_info %d in list but !highest_bit\n", | |
976 | si->type); | |
977 | WARN(!(si->flags & SWP_WRITEOK), | |
978 | "swap_info %d in list but !SWP_WRITEOK\n", | |
979 | si->type); | |
a2468cc9 | 980 | __del_from_avail_list(si); |
ec8acf20 | 981 | spin_unlock(&si->lock); |
18ab4d4c | 982 | goto nextsi; |
ec8acf20 | 983 | } |
f0eea189 HY |
984 | if (cluster) { |
985 | if (!(si->flags & SWP_FILE)) | |
986 | n_ret = swap_alloc_cluster(si, swp_entries); | |
987 | } else | |
38d8b4e6 HY |
988 | n_ret = scan_swap_map_slots(si, SWAP_HAS_CACHE, |
989 | n_goal, swp_entries); | |
ec8acf20 | 990 | spin_unlock(&si->lock); |
38d8b4e6 | 991 | if (n_ret || cluster) |
36005bae | 992 | goto check_out; |
18ab4d4c | 993 | pr_debug("scan_swap_map of si %d failed to find offset\n", |
36005bae TC |
994 | si->type); |
995 | ||
18ab4d4c DS |
996 | spin_lock(&swap_avail_lock); |
997 | nextsi: | |
adfab836 DS |
998 | /* |
999 | * if we got here, it's likely that si was almost full before, | |
1000 | * and since scan_swap_map() can drop the si->lock, multiple | |
1001 | * callers probably all tried to get a page from the same si | |
18ab4d4c DS |
1002 | * and it filled up before we could get one; or, the si filled |
1003 | * up between us dropping swap_avail_lock and taking si->lock. | |
1004 | * Since we dropped the swap_avail_lock, the swap_avail_head | |
1005 | * list may have been modified; so if next is still in the | |
36005bae TC |
1006 | * swap_avail_head list then try it, otherwise start over |
1007 | * if we have not gotten any slots. | |
adfab836 | 1008 | */ |
a2468cc9 | 1009 | if (plist_node_empty(&next->avail_lists[node])) |
18ab4d4c | 1010 | goto start_over; |
1da177e4 | 1011 | } |
fb4f88dc | 1012 | |
18ab4d4c DS |
1013 | spin_unlock(&swap_avail_lock); |
1014 | ||
36005bae TC |
1015 | check_out: |
1016 | if (n_ret < n_goal) | |
38d8b4e6 HY |
1017 | atomic_long_add((long)(n_goal - n_ret) * nr_pages, |
1018 | &nr_swap_pages); | |
fb4f88dc | 1019 | noswap: |
36005bae TC |
1020 | return n_ret; |
1021 | } | |
1022 | ||
2de1a7e4 | 1023 | /* The only caller of this function is now suspend routine */ |
910321ea HD |
1024 | swp_entry_t get_swap_page_of_type(int type) |
1025 | { | |
27a34b2d | 1026 | struct swap_info_struct *si = swap_type_to_swap_info(type); |
910321ea HD |
1027 | pgoff_t offset; |
1028 | ||
27a34b2d DJ |
1029 | if (!si) |
1030 | goto fail; | |
1031 | ||
ec8acf20 | 1032 | spin_lock(&si->lock); |
27a34b2d | 1033 | if (si->flags & SWP_WRITEOK) { |
ec8acf20 | 1034 | atomic_long_dec(&nr_swap_pages); |
910321ea HD |
1035 | /* This is called for allocating swap entry, not cache */ |
1036 | offset = scan_swap_map(si, 1); | |
1037 | if (offset) { | |
ec8acf20 | 1038 | spin_unlock(&si->lock); |
910321ea HD |
1039 | return swp_entry(type, offset); |
1040 | } | |
ec8acf20 | 1041 | atomic_long_inc(&nr_swap_pages); |
910321ea | 1042 | } |
ec8acf20 | 1043 | spin_unlock(&si->lock); |
27a34b2d | 1044 | fail: |
910321ea HD |
1045 | return (swp_entry_t) {0}; |
1046 | } | |
1047 | ||
e8c26ab6 | 1048 | static struct swap_info_struct *__swap_info_get(swp_entry_t entry) |
1da177e4 | 1049 | { |
73c34b6a | 1050 | struct swap_info_struct *p; |
1da177e4 LT |
1051 | unsigned long offset, type; |
1052 | ||
1053 | if (!entry.val) | |
1054 | goto out; | |
1055 | type = swp_type(entry); | |
27a34b2d DJ |
1056 | p = swap_type_to_swap_info(type); |
1057 | if (!p) | |
1da177e4 | 1058 | goto bad_nofile; |
1da177e4 LT |
1059 | if (!(p->flags & SWP_USED)) |
1060 | goto bad_device; | |
1061 | offset = swp_offset(entry); | |
1062 | if (offset >= p->max) | |
1063 | goto bad_offset; | |
1da177e4 LT |
1064 | return p; |
1065 | ||
1da177e4 | 1066 | bad_offset: |
6a991fc7 | 1067 | pr_err("swap_info_get: %s%08lx\n", Bad_offset, entry.val); |
1da177e4 LT |
1068 | goto out; |
1069 | bad_device: | |
6a991fc7 | 1070 | pr_err("swap_info_get: %s%08lx\n", Unused_file, entry.val); |
1da177e4 LT |
1071 | goto out; |
1072 | bad_nofile: | |
6a991fc7 | 1073 | pr_err("swap_info_get: %s%08lx\n", Bad_file, entry.val); |
1da177e4 LT |
1074 | out: |
1075 | return NULL; | |
886bb7e9 | 1076 | } |
1da177e4 | 1077 | |
e8c26ab6 TC |
1078 | static struct swap_info_struct *_swap_info_get(swp_entry_t entry) |
1079 | { | |
1080 | struct swap_info_struct *p; | |
1081 | ||
1082 | p = __swap_info_get(entry); | |
1083 | if (!p) | |
1084 | goto out; | |
1085 | if (!p->swap_map[swp_offset(entry)]) | |
1086 | goto bad_free; | |
1087 | return p; | |
1088 | ||
1089 | bad_free: | |
1090 | pr_err("swap_info_get: %s%08lx\n", Unused_offset, entry.val); | |
1091 | goto out; | |
1092 | out: | |
1093 | return NULL; | |
1094 | } | |
1095 | ||
235b6217 HY |
1096 | static struct swap_info_struct *swap_info_get(swp_entry_t entry) |
1097 | { | |
1098 | struct swap_info_struct *p; | |
1099 | ||
1100 | p = _swap_info_get(entry); | |
1101 | if (p) | |
1102 | spin_lock(&p->lock); | |
1103 | return p; | |
1104 | } | |
1105 | ||
7c00bafe TC |
1106 | static struct swap_info_struct *swap_info_get_cont(swp_entry_t entry, |
1107 | struct swap_info_struct *q) | |
1108 | { | |
1109 | struct swap_info_struct *p; | |
1110 | ||
1111 | p = _swap_info_get(entry); | |
1112 | ||
1113 | if (p != q) { | |
1114 | if (q != NULL) | |
1115 | spin_unlock(&q->lock); | |
1116 | if (p != NULL) | |
1117 | spin_lock(&p->lock); | |
1118 | } | |
1119 | return p; | |
1120 | } | |
1121 | ||
1122 | static unsigned char __swap_entry_free(struct swap_info_struct *p, | |
1123 | swp_entry_t entry, unsigned char usage) | |
1da177e4 | 1124 | { |
235b6217 | 1125 | struct swap_cluster_info *ci; |
253d553b | 1126 | unsigned long offset = swp_offset(entry); |
8d69aaee HD |
1127 | unsigned char count; |
1128 | unsigned char has_cache; | |
235b6217 | 1129 | |
7c00bafe | 1130 | ci = lock_cluster_or_swap_info(p, offset); |
355cfa73 | 1131 | |
253d553b | 1132 | count = p->swap_map[offset]; |
235b6217 | 1133 | |
253d553b HD |
1134 | has_cache = count & SWAP_HAS_CACHE; |
1135 | count &= ~SWAP_HAS_CACHE; | |
355cfa73 | 1136 | |
253d553b | 1137 | if (usage == SWAP_HAS_CACHE) { |
355cfa73 | 1138 | VM_BUG_ON(!has_cache); |
253d553b | 1139 | has_cache = 0; |
aaa46865 HD |
1140 | } else if (count == SWAP_MAP_SHMEM) { |
1141 | /* | |
1142 | * Or we could insist on shmem.c using a special | |
1143 | * swap_shmem_free() and free_shmem_swap_and_cache()... | |
1144 | */ | |
1145 | count = 0; | |
570a335b HD |
1146 | } else if ((count & ~COUNT_CONTINUED) <= SWAP_MAP_MAX) { |
1147 | if (count == COUNT_CONTINUED) { | |
1148 | if (swap_count_continued(p, offset, count)) | |
1149 | count = SWAP_MAP_MAX | COUNT_CONTINUED; | |
1150 | else | |
1151 | count = SWAP_MAP_MAX; | |
1152 | } else | |
1153 | count--; | |
1154 | } | |
253d553b | 1155 | |
253d553b | 1156 | usage = count | has_cache; |
7c00bafe TC |
1157 | p->swap_map[offset] = usage ? : SWAP_HAS_CACHE; |
1158 | ||
1159 | unlock_cluster_or_swap_info(p, ci); | |
1160 | ||
1161 | return usage; | |
1162 | } | |
355cfa73 | 1163 | |
7c00bafe TC |
1164 | static void swap_entry_free(struct swap_info_struct *p, swp_entry_t entry) |
1165 | { | |
1166 | struct swap_cluster_info *ci; | |
1167 | unsigned long offset = swp_offset(entry); | |
1168 | unsigned char count; | |
1169 | ||
1170 | ci = lock_cluster(p, offset); | |
1171 | count = p->swap_map[offset]; | |
1172 | VM_BUG_ON(count != SWAP_HAS_CACHE); | |
1173 | p->swap_map[offset] = 0; | |
1174 | dec_cluster_info_page(p, p->cluster_info, offset); | |
235b6217 HY |
1175 | unlock_cluster(ci); |
1176 | ||
38d8b4e6 HY |
1177 | mem_cgroup_uncharge_swap(entry, 1); |
1178 | swap_range_free(p, offset, 1); | |
1da177e4 LT |
1179 | } |
1180 | ||
1181 | /* | |
2de1a7e4 | 1182 | * Caller has made sure that the swap device corresponding to entry |
1da177e4 LT |
1183 | * is still around or has not been recycled. |
1184 | */ | |
1185 | void swap_free(swp_entry_t entry) | |
1186 | { | |
73c34b6a | 1187 | struct swap_info_struct *p; |
1da177e4 | 1188 | |
235b6217 | 1189 | p = _swap_info_get(entry); |
7c00bafe TC |
1190 | if (p) { |
1191 | if (!__swap_entry_free(p, entry, 1)) | |
67afa38e | 1192 | free_swap_slot(entry); |
7c00bafe | 1193 | } |
1da177e4 LT |
1194 | } |
1195 | ||
cb4b86ba KH |
1196 | /* |
1197 | * Called after dropping swapcache to decrease refcnt to swap entries. | |
1198 | */ | |
75f6d6d2 | 1199 | static void swapcache_free(swp_entry_t entry) |
cb4b86ba | 1200 | { |
355cfa73 KH |
1201 | struct swap_info_struct *p; |
1202 | ||
235b6217 | 1203 | p = _swap_info_get(entry); |
7c00bafe TC |
1204 | if (p) { |
1205 | if (!__swap_entry_free(p, entry, SWAP_HAS_CACHE)) | |
67afa38e | 1206 | free_swap_slot(entry); |
7c00bafe TC |
1207 | } |
1208 | } | |
1209 | ||
38d8b4e6 | 1210 | #ifdef CONFIG_THP_SWAP |
75f6d6d2 | 1211 | static void swapcache_free_cluster(swp_entry_t entry) |
38d8b4e6 HY |
1212 | { |
1213 | unsigned long offset = swp_offset(entry); | |
1214 | unsigned long idx = offset / SWAPFILE_CLUSTER; | |
1215 | struct swap_cluster_info *ci; | |
1216 | struct swap_info_struct *si; | |
1217 | unsigned char *map; | |
a3aea839 HY |
1218 | unsigned int i, free_entries = 0; |
1219 | unsigned char val; | |
38d8b4e6 | 1220 | |
a3aea839 | 1221 | si = _swap_info_get(entry); |
38d8b4e6 HY |
1222 | if (!si) |
1223 | return; | |
1224 | ||
1225 | ci = lock_cluster(si, offset); | |
e0709829 | 1226 | VM_BUG_ON(!cluster_is_huge(ci)); |
38d8b4e6 HY |
1227 | map = si->swap_map + offset; |
1228 | for (i = 0; i < SWAPFILE_CLUSTER; i++) { | |
a3aea839 HY |
1229 | val = map[i]; |
1230 | VM_BUG_ON(!(val & SWAP_HAS_CACHE)); | |
1231 | if (val == SWAP_HAS_CACHE) | |
1232 | free_entries++; | |
1233 | } | |
1234 | if (!free_entries) { | |
1235 | for (i = 0; i < SWAPFILE_CLUSTER; i++) | |
1236 | map[i] &= ~SWAP_HAS_CACHE; | |
38d8b4e6 | 1237 | } |
e0709829 | 1238 | cluster_clear_huge(ci); |
38d8b4e6 | 1239 | unlock_cluster(ci); |
a3aea839 HY |
1240 | if (free_entries == SWAPFILE_CLUSTER) { |
1241 | spin_lock(&si->lock); | |
1242 | ci = lock_cluster(si, offset); | |
1243 | memset(map, 0, SWAPFILE_CLUSTER); | |
1244 | unlock_cluster(ci); | |
1245 | mem_cgroup_uncharge_swap(entry, SWAPFILE_CLUSTER); | |
1246 | swap_free_cluster(si, idx); | |
1247 | spin_unlock(&si->lock); | |
1248 | } else if (free_entries) { | |
1249 | for (i = 0; i < SWAPFILE_CLUSTER; i++, entry.val++) { | |
1250 | if (!__swap_entry_free(si, entry, SWAP_HAS_CACHE)) | |
1251 | free_swap_slot(entry); | |
1252 | } | |
1253 | } | |
38d8b4e6 | 1254 | } |
59807685 HY |
1255 | |
1256 | int split_swap_cluster(swp_entry_t entry) | |
1257 | { | |
1258 | struct swap_info_struct *si; | |
1259 | struct swap_cluster_info *ci; | |
1260 | unsigned long offset = swp_offset(entry); | |
1261 | ||
1262 | si = _swap_info_get(entry); | |
1263 | if (!si) | |
1264 | return -EBUSY; | |
1265 | ci = lock_cluster(si, offset); | |
1266 | cluster_clear_huge(ci); | |
1267 | unlock_cluster(ci); | |
1268 | return 0; | |
1269 | } | |
75f6d6d2 MK |
1270 | #else |
1271 | static inline void swapcache_free_cluster(swp_entry_t entry) | |
1272 | { | |
1273 | } | |
38d8b4e6 HY |
1274 | #endif /* CONFIG_THP_SWAP */ |
1275 | ||
75f6d6d2 MK |
1276 | void put_swap_page(struct page *page, swp_entry_t entry) |
1277 | { | |
1278 | if (!PageTransHuge(page)) | |
1279 | swapcache_free(entry); | |
1280 | else | |
1281 | swapcache_free_cluster(entry); | |
1282 | } | |
1283 | ||
155b5f88 HY |
1284 | static int swp_entry_cmp(const void *ent1, const void *ent2) |
1285 | { | |
1286 | const swp_entry_t *e1 = ent1, *e2 = ent2; | |
1287 | ||
1288 | return (int)swp_type(*e1) - (int)swp_type(*e2); | |
1289 | } | |
1290 | ||
7c00bafe TC |
1291 | void swapcache_free_entries(swp_entry_t *entries, int n) |
1292 | { | |
1293 | struct swap_info_struct *p, *prev; | |
1294 | int i; | |
1295 | ||
1296 | if (n <= 0) | |
1297 | return; | |
1298 | ||
1299 | prev = NULL; | |
1300 | p = NULL; | |
155b5f88 HY |
1301 | |
1302 | /* | |
1303 | * Sort swap entries by swap device, so each lock is only taken once. | |
1304 | * nr_swapfiles isn't absolutely correct, but the overhead of sort() is | |
1305 | * so low that it isn't necessary to optimize further. | |
1306 | */ | |
1307 | if (nr_swapfiles > 1) | |
1308 | sort(entries, n, sizeof(entries[0]), swp_entry_cmp, NULL); | |
7c00bafe TC |
1309 | for (i = 0; i < n; ++i) { |
1310 | p = swap_info_get_cont(entries[i], prev); | |
1311 | if (p) | |
1312 | swap_entry_free(p, entries[i]); | |
7c00bafe TC |
1313 | prev = p; |
1314 | } | |
235b6217 | 1315 | if (p) |
7c00bafe | 1316 | spin_unlock(&p->lock); |
cb4b86ba KH |
1317 | } |
1318 | ||
1da177e4 | 1319 | /* |
c475a8ab | 1320 | * How many references to page are currently swapped out? |
570a335b HD |
1321 | * This does not give an exact answer when swap count is continued, |
1322 | * but does include the high COUNT_CONTINUED flag to allow for that. | |
1da177e4 | 1323 | */ |
bde05d1c | 1324 | int page_swapcount(struct page *page) |
1da177e4 | 1325 | { |
c475a8ab HD |
1326 | int count = 0; |
1327 | struct swap_info_struct *p; | |
235b6217 | 1328 | struct swap_cluster_info *ci; |
1da177e4 | 1329 | swp_entry_t entry; |
235b6217 | 1330 | unsigned long offset; |
1da177e4 | 1331 | |
4c21e2f2 | 1332 | entry.val = page_private(page); |
235b6217 | 1333 | p = _swap_info_get(entry); |
1da177e4 | 1334 | if (p) { |
235b6217 HY |
1335 | offset = swp_offset(entry); |
1336 | ci = lock_cluster_or_swap_info(p, offset); | |
1337 | count = swap_count(p->swap_map[offset]); | |
1338 | unlock_cluster_or_swap_info(p, ci); | |
1da177e4 | 1339 | } |
c475a8ab | 1340 | return count; |
1da177e4 LT |
1341 | } |
1342 | ||
aa8d22a1 MK |
1343 | int __swap_count(struct swap_info_struct *si, swp_entry_t entry) |
1344 | { | |
1345 | pgoff_t offset = swp_offset(entry); | |
1346 | ||
1347 | return swap_count(si->swap_map[offset]); | |
1348 | } | |
1349 | ||
322b8afe HY |
1350 | static int swap_swapcount(struct swap_info_struct *si, swp_entry_t entry) |
1351 | { | |
1352 | int count = 0; | |
1353 | pgoff_t offset = swp_offset(entry); | |
1354 | struct swap_cluster_info *ci; | |
1355 | ||
1356 | ci = lock_cluster_or_swap_info(si, offset); | |
1357 | count = swap_count(si->swap_map[offset]); | |
1358 | unlock_cluster_or_swap_info(si, ci); | |
1359 | return count; | |
1360 | } | |
1361 | ||
e8c26ab6 TC |
1362 | /* |
1363 | * How many references to @entry are currently swapped out? | |
1364 | * This does not give an exact answer when swap count is continued, | |
1365 | * but does include the high COUNT_CONTINUED flag to allow for that. | |
1366 | */ | |
1367 | int __swp_swapcount(swp_entry_t entry) | |
1368 | { | |
1369 | int count = 0; | |
e8c26ab6 | 1370 | struct swap_info_struct *si; |
e8c26ab6 TC |
1371 | |
1372 | si = __swap_info_get(entry); | |
322b8afe HY |
1373 | if (si) |
1374 | count = swap_swapcount(si, entry); | |
e8c26ab6 TC |
1375 | return count; |
1376 | } | |
1377 | ||
8334b962 MK |
1378 | /* |
1379 | * How many references to @entry are currently swapped out? | |
1380 | * This considers COUNT_CONTINUED so it returns exact answer. | |
1381 | */ | |
1382 | int swp_swapcount(swp_entry_t entry) | |
1383 | { | |
1384 | int count, tmp_count, n; | |
1385 | struct swap_info_struct *p; | |
235b6217 | 1386 | struct swap_cluster_info *ci; |
8334b962 MK |
1387 | struct page *page; |
1388 | pgoff_t offset; | |
1389 | unsigned char *map; | |
1390 | ||
235b6217 | 1391 | p = _swap_info_get(entry); |
8334b962 MK |
1392 | if (!p) |
1393 | return 0; | |
1394 | ||
235b6217 HY |
1395 | offset = swp_offset(entry); |
1396 | ||
1397 | ci = lock_cluster_or_swap_info(p, offset); | |
1398 | ||
1399 | count = swap_count(p->swap_map[offset]); | |
8334b962 MK |
1400 | if (!(count & COUNT_CONTINUED)) |
1401 | goto out; | |
1402 | ||
1403 | count &= ~COUNT_CONTINUED; | |
1404 | n = SWAP_MAP_MAX + 1; | |
1405 | ||
8334b962 MK |
1406 | page = vmalloc_to_page(p->swap_map + offset); |
1407 | offset &= ~PAGE_MASK; | |
1408 | VM_BUG_ON(page_private(page) != SWP_CONTINUED); | |
1409 | ||
1410 | do { | |
a8ae4991 | 1411 | page = list_next_entry(page, lru); |
8334b962 MK |
1412 | map = kmap_atomic(page); |
1413 | tmp_count = map[offset]; | |
1414 | kunmap_atomic(map); | |
1415 | ||
1416 | count += (tmp_count & ~COUNT_CONTINUED) * n; | |
1417 | n *= (SWAP_CONT_MAX + 1); | |
1418 | } while (tmp_count & COUNT_CONTINUED); | |
1419 | out: | |
235b6217 | 1420 | unlock_cluster_or_swap_info(p, ci); |
8334b962 MK |
1421 | return count; |
1422 | } | |
1423 | ||
e0709829 HY |
1424 | #ifdef CONFIG_THP_SWAP |
1425 | static bool swap_page_trans_huge_swapped(struct swap_info_struct *si, | |
1426 | swp_entry_t entry) | |
1427 | { | |
1428 | struct swap_cluster_info *ci; | |
1429 | unsigned char *map = si->swap_map; | |
1430 | unsigned long roffset = swp_offset(entry); | |
1431 | unsigned long offset = round_down(roffset, SWAPFILE_CLUSTER); | |
1432 | int i; | |
1433 | bool ret = false; | |
1434 | ||
1435 | ci = lock_cluster_or_swap_info(si, offset); | |
1436 | if (!ci || !cluster_is_huge(ci)) { | |
1437 | if (map[roffset] != SWAP_HAS_CACHE) | |
1438 | ret = true; | |
1439 | goto unlock_out; | |
1440 | } | |
1441 | for (i = 0; i < SWAPFILE_CLUSTER; i++) { | |
1442 | if (map[offset + i] != SWAP_HAS_CACHE) { | |
1443 | ret = true; | |
1444 | break; | |
1445 | } | |
1446 | } | |
1447 | unlock_out: | |
1448 | unlock_cluster_or_swap_info(si, ci); | |
1449 | return ret; | |
1450 | } | |
1451 | ||
1452 | static bool page_swapped(struct page *page) | |
1453 | { | |
1454 | swp_entry_t entry; | |
1455 | struct swap_info_struct *si; | |
1456 | ||
1457 | if (likely(!PageTransCompound(page))) | |
1458 | return page_swapcount(page) != 0; | |
1459 | ||
1460 | page = compound_head(page); | |
1461 | entry.val = page_private(page); | |
1462 | si = _swap_info_get(entry); | |
1463 | if (si) | |
1464 | return swap_page_trans_huge_swapped(si, entry); | |
1465 | return false; | |
1466 | } | |
ba3c4ce6 HY |
1467 | |
1468 | static int page_trans_huge_map_swapcount(struct page *page, int *total_mapcount, | |
1469 | int *total_swapcount) | |
1470 | { | |
1471 | int i, map_swapcount, _total_mapcount, _total_swapcount; | |
1472 | unsigned long offset = 0; | |
1473 | struct swap_info_struct *si; | |
1474 | struct swap_cluster_info *ci = NULL; | |
1475 | unsigned char *map = NULL; | |
1476 | int mapcount, swapcount = 0; | |
1477 | ||
1478 | /* hugetlbfs shouldn't call it */ | |
1479 | VM_BUG_ON_PAGE(PageHuge(page), page); | |
1480 | ||
1481 | if (likely(!PageTransCompound(page))) { | |
1482 | mapcount = atomic_read(&page->_mapcount) + 1; | |
1483 | if (total_mapcount) | |
1484 | *total_mapcount = mapcount; | |
1485 | if (PageSwapCache(page)) | |
1486 | swapcount = page_swapcount(page); | |
1487 | if (total_swapcount) | |
1488 | *total_swapcount = swapcount; | |
1489 | return mapcount + swapcount; | |
1490 | } | |
1491 | ||
1492 | page = compound_head(page); | |
1493 | ||
1494 | _total_mapcount = _total_swapcount = map_swapcount = 0; | |
1495 | if (PageSwapCache(page)) { | |
1496 | swp_entry_t entry; | |
1497 | ||
1498 | entry.val = page_private(page); | |
1499 | si = _swap_info_get(entry); | |
1500 | if (si) { | |
1501 | map = si->swap_map; | |
1502 | offset = swp_offset(entry); | |
1503 | } | |
1504 | } | |
1505 | if (map) | |
1506 | ci = lock_cluster(si, offset); | |
1507 | for (i = 0; i < HPAGE_PMD_NR; i++) { | |
1508 | mapcount = atomic_read(&page[i]._mapcount) + 1; | |
1509 | _total_mapcount += mapcount; | |
1510 | if (map) { | |
1511 | swapcount = swap_count(map[offset + i]); | |
1512 | _total_swapcount += swapcount; | |
1513 | } | |
1514 | map_swapcount = max(map_swapcount, mapcount + swapcount); | |
1515 | } | |
1516 | unlock_cluster(ci); | |
1517 | if (PageDoubleMap(page)) { | |
1518 | map_swapcount -= 1; | |
1519 | _total_mapcount -= HPAGE_PMD_NR; | |
1520 | } | |
1521 | mapcount = compound_mapcount(page); | |
1522 | map_swapcount += mapcount; | |
1523 | _total_mapcount += mapcount; | |
1524 | if (total_mapcount) | |
1525 | *total_mapcount = _total_mapcount; | |
1526 | if (total_swapcount) | |
1527 | *total_swapcount = _total_swapcount; | |
1528 | ||
1529 | return map_swapcount; | |
1530 | } | |
e0709829 HY |
1531 | #else |
1532 | #define swap_page_trans_huge_swapped(si, entry) swap_swapcount(si, entry) | |
1533 | #define page_swapped(page) (page_swapcount(page) != 0) | |
ba3c4ce6 HY |
1534 | |
1535 | static int page_trans_huge_map_swapcount(struct page *page, int *total_mapcount, | |
1536 | int *total_swapcount) | |
1537 | { | |
1538 | int mapcount, swapcount = 0; | |
1539 | ||
1540 | /* hugetlbfs shouldn't call it */ | |
1541 | VM_BUG_ON_PAGE(PageHuge(page), page); | |
1542 | ||
1543 | mapcount = page_trans_huge_mapcount(page, total_mapcount); | |
1544 | if (PageSwapCache(page)) | |
1545 | swapcount = page_swapcount(page); | |
1546 | if (total_swapcount) | |
1547 | *total_swapcount = swapcount; | |
1548 | return mapcount + swapcount; | |
1549 | } | |
e0709829 HY |
1550 | #endif |
1551 | ||
1da177e4 | 1552 | /* |
7b1fe597 HD |
1553 | * We can write to an anon page without COW if there are no other references |
1554 | * to it. And as a side-effect, free up its swap: because the old content | |
1555 | * on disk will never be read, and seeking back there to write new content | |
1556 | * later would only waste time away from clustering. | |
6d0a07ed | 1557 | * |
ba3c4ce6 | 1558 | * NOTE: total_map_swapcount should not be relied upon by the caller if |
6d0a07ed AA |
1559 | * reuse_swap_page() returns false, but it may be always overwritten |
1560 | * (see the other implementation for CONFIG_SWAP=n). | |
1da177e4 | 1561 | */ |
ba3c4ce6 | 1562 | bool reuse_swap_page(struct page *page, int *total_map_swapcount) |
1da177e4 | 1563 | { |
ba3c4ce6 | 1564 | int count, total_mapcount, total_swapcount; |
c475a8ab | 1565 | |
309381fe | 1566 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
5ad64688 | 1567 | if (unlikely(PageKsm(page))) |
6d0a07ed | 1568 | return false; |
ba3c4ce6 HY |
1569 | count = page_trans_huge_map_swapcount(page, &total_mapcount, |
1570 | &total_swapcount); | |
1571 | if (total_map_swapcount) | |
1572 | *total_map_swapcount = total_mapcount + total_swapcount; | |
1573 | if (count == 1 && PageSwapCache(page) && | |
1574 | (likely(!PageTransCompound(page)) || | |
1575 | /* The remaining swap count will be freed soon */ | |
1576 | total_swapcount == page_swapcount(page))) { | |
f0571429 | 1577 | if (!PageWriteback(page)) { |
ba3c4ce6 | 1578 | page = compound_head(page); |
7b1fe597 HD |
1579 | delete_from_swap_cache(page); |
1580 | SetPageDirty(page); | |
f0571429 MK |
1581 | } else { |
1582 | swp_entry_t entry; | |
1583 | struct swap_info_struct *p; | |
1584 | ||
1585 | entry.val = page_private(page); | |
1586 | p = swap_info_get(entry); | |
1587 | if (p->flags & SWP_STABLE_WRITES) { | |
1588 | spin_unlock(&p->lock); | |
1589 | return false; | |
1590 | } | |
1591 | spin_unlock(&p->lock); | |
7b1fe597 HD |
1592 | } |
1593 | } | |
ba3c4ce6 | 1594 | |
5ad64688 | 1595 | return count <= 1; |
1da177e4 LT |
1596 | } |
1597 | ||
1598 | /* | |
a2c43eed HD |
1599 | * If swap is getting full, or if there are no more mappings of this page, |
1600 | * then try_to_free_swap is called to free its swap space. | |
1da177e4 | 1601 | */ |
a2c43eed | 1602 | int try_to_free_swap(struct page *page) |
1da177e4 | 1603 | { |
309381fe | 1604 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
1da177e4 LT |
1605 | |
1606 | if (!PageSwapCache(page)) | |
1607 | return 0; | |
1608 | if (PageWriteback(page)) | |
1609 | return 0; | |
e0709829 | 1610 | if (page_swapped(page)) |
1da177e4 LT |
1611 | return 0; |
1612 | ||
b73d7fce HD |
1613 | /* |
1614 | * Once hibernation has begun to create its image of memory, | |
1615 | * there's a danger that one of the calls to try_to_free_swap() | |
1616 | * - most probably a call from __try_to_reclaim_swap() while | |
1617 | * hibernation is allocating its own swap pages for the image, | |
1618 | * but conceivably even a call from memory reclaim - will free | |
1619 | * the swap from a page which has already been recorded in the | |
1620 | * image as a clean swapcache page, and then reuse its swap for | |
1621 | * another page of the image. On waking from hibernation, the | |
1622 | * original page might be freed under memory pressure, then | |
1623 | * later read back in from swap, now with the wrong data. | |
1624 | * | |
2de1a7e4 | 1625 | * Hibernation suspends storage while it is writing the image |
f90ac398 | 1626 | * to disk so check that here. |
b73d7fce | 1627 | */ |
f90ac398 | 1628 | if (pm_suspended_storage()) |
b73d7fce HD |
1629 | return 0; |
1630 | ||
e0709829 | 1631 | page = compound_head(page); |
a2c43eed HD |
1632 | delete_from_swap_cache(page); |
1633 | SetPageDirty(page); | |
1634 | return 1; | |
68a22394 RR |
1635 | } |
1636 | ||
1da177e4 LT |
1637 | /* |
1638 | * Free the swap entry like above, but also try to | |
1639 | * free the page cache entry if it is the last user. | |
1640 | */ | |
2509ef26 | 1641 | int free_swap_and_cache(swp_entry_t entry) |
1da177e4 | 1642 | { |
2509ef26 | 1643 | struct swap_info_struct *p; |
1da177e4 | 1644 | struct page *page = NULL; |
7c00bafe | 1645 | unsigned char count; |
1da177e4 | 1646 | |
a7420aa5 | 1647 | if (non_swap_entry(entry)) |
2509ef26 | 1648 | return 1; |
0697212a | 1649 | |
7c00bafe | 1650 | p = _swap_info_get(entry); |
1da177e4 | 1651 | if (p) { |
7c00bafe | 1652 | count = __swap_entry_free(p, entry, 1); |
e0709829 HY |
1653 | if (count == SWAP_HAS_CACHE && |
1654 | !swap_page_trans_huge_swapped(p, entry)) { | |
33806f06 | 1655 | page = find_get_page(swap_address_space(entry), |
f6ab1f7f | 1656 | swp_offset(entry)); |
8413ac9d | 1657 | if (page && !trylock_page(page)) { |
09cbfeaf | 1658 | put_page(page); |
93fac704 NP |
1659 | page = NULL; |
1660 | } | |
7c00bafe | 1661 | } else if (!count) |
67afa38e | 1662 | free_swap_slot(entry); |
1da177e4 LT |
1663 | } |
1664 | if (page) { | |
a2c43eed HD |
1665 | /* |
1666 | * Not mapped elsewhere, or swap space full? Free it! | |
1667 | * Also recheck PageSwapCache now page is locked (above). | |
1668 | */ | |
93fac704 | 1669 | if (PageSwapCache(page) && !PageWriteback(page) && |
322b8afe | 1670 | (!page_mapped(page) || mem_cgroup_swap_full(page)) && |
e0709829 HY |
1671 | !swap_page_trans_huge_swapped(p, entry)) { |
1672 | page = compound_head(page); | |
1da177e4 LT |
1673 | delete_from_swap_cache(page); |
1674 | SetPageDirty(page); | |
1675 | } | |
1676 | unlock_page(page); | |
09cbfeaf | 1677 | put_page(page); |
1da177e4 | 1678 | } |
2509ef26 | 1679 | return p != NULL; |
1da177e4 LT |
1680 | } |
1681 | ||
b0cb1a19 | 1682 | #ifdef CONFIG_HIBERNATION |
f577eb30 | 1683 | /* |
915bae9e | 1684 | * Find the swap type that corresponds to given device (if any). |
f577eb30 | 1685 | * |
915bae9e RW |
1686 | * @offset - number of the PAGE_SIZE-sized block of the device, starting |
1687 | * from 0, in which the swap header is expected to be located. | |
1688 | * | |
1689 | * This is needed for the suspend to disk (aka swsusp). | |
f577eb30 | 1690 | */ |
7bf23687 | 1691 | int swap_type_of(dev_t device, sector_t offset, struct block_device **bdev_p) |
f577eb30 | 1692 | { |
915bae9e | 1693 | struct block_device *bdev = NULL; |
efa90a98 | 1694 | int type; |
f577eb30 | 1695 | |
915bae9e RW |
1696 | if (device) |
1697 | bdev = bdget(device); | |
1698 | ||
f577eb30 | 1699 | spin_lock(&swap_lock); |
efa90a98 HD |
1700 | for (type = 0; type < nr_swapfiles; type++) { |
1701 | struct swap_info_struct *sis = swap_info[type]; | |
f577eb30 | 1702 | |
915bae9e | 1703 | if (!(sis->flags & SWP_WRITEOK)) |
f577eb30 | 1704 | continue; |
b6b5bce3 | 1705 | |
915bae9e | 1706 | if (!bdev) { |
7bf23687 | 1707 | if (bdev_p) |
dddac6a7 | 1708 | *bdev_p = bdgrab(sis->bdev); |
7bf23687 | 1709 | |
6e1819d6 | 1710 | spin_unlock(&swap_lock); |
efa90a98 | 1711 | return type; |
6e1819d6 | 1712 | } |
915bae9e | 1713 | if (bdev == sis->bdev) { |
9625a5f2 | 1714 | struct swap_extent *se = &sis->first_swap_extent; |
915bae9e | 1715 | |
915bae9e | 1716 | if (se->start_block == offset) { |
7bf23687 | 1717 | if (bdev_p) |
dddac6a7 | 1718 | *bdev_p = bdgrab(sis->bdev); |
7bf23687 | 1719 | |
915bae9e RW |
1720 | spin_unlock(&swap_lock); |
1721 | bdput(bdev); | |
efa90a98 | 1722 | return type; |
915bae9e | 1723 | } |
f577eb30 RW |
1724 | } |
1725 | } | |
1726 | spin_unlock(&swap_lock); | |
915bae9e RW |
1727 | if (bdev) |
1728 | bdput(bdev); | |
1729 | ||
f577eb30 RW |
1730 | return -ENODEV; |
1731 | } | |
1732 | ||
73c34b6a HD |
1733 | /* |
1734 | * Get the (PAGE_SIZE) block corresponding to given offset on the swapdev | |
1735 | * corresponding to given index in swap_info (swap type). | |
1736 | */ | |
1737 | sector_t swapdev_block(int type, pgoff_t offset) | |
1738 | { | |
1739 | struct block_device *bdev; | |
27a34b2d | 1740 | struct swap_info_struct *si = swap_type_to_swap_info(type); |
73c34b6a | 1741 | |
27a34b2d | 1742 | if (!si || !(si->flags & SWP_WRITEOK)) |
73c34b6a | 1743 | return 0; |
d4906e1a | 1744 | return map_swap_entry(swp_entry(type, offset), &bdev); |
73c34b6a HD |
1745 | } |
1746 | ||
f577eb30 RW |
1747 | /* |
1748 | * Return either the total number of swap pages of given type, or the number | |
1749 | * of free pages of that type (depending on @free) | |
1750 | * | |
1751 | * This is needed for software suspend | |
1752 | */ | |
1753 | unsigned int count_swap_pages(int type, int free) | |
1754 | { | |
1755 | unsigned int n = 0; | |
1756 | ||
efa90a98 HD |
1757 | spin_lock(&swap_lock); |
1758 | if ((unsigned int)type < nr_swapfiles) { | |
1759 | struct swap_info_struct *sis = swap_info[type]; | |
1760 | ||
ec8acf20 | 1761 | spin_lock(&sis->lock); |
efa90a98 HD |
1762 | if (sis->flags & SWP_WRITEOK) { |
1763 | n = sis->pages; | |
f577eb30 | 1764 | if (free) |
efa90a98 | 1765 | n -= sis->inuse_pages; |
f577eb30 | 1766 | } |
ec8acf20 | 1767 | spin_unlock(&sis->lock); |
f577eb30 | 1768 | } |
efa90a98 | 1769 | spin_unlock(&swap_lock); |
f577eb30 RW |
1770 | return n; |
1771 | } | |
73c34b6a | 1772 | #endif /* CONFIG_HIBERNATION */ |
f577eb30 | 1773 | |
9f8bdb3f | 1774 | static inline int pte_same_as_swp(pte_t pte, pte_t swp_pte) |
179ef71c | 1775 | { |
9f8bdb3f | 1776 | return pte_same(pte_swp_clear_soft_dirty(pte), swp_pte); |
179ef71c CG |
1777 | } |
1778 | ||
1da177e4 | 1779 | /* |
72866f6f HD |
1780 | * No need to decide whether this PTE shares the swap entry with others, |
1781 | * just let do_wp_page work it out if a write is requested later - to | |
1782 | * force COW, vm_page_prot omits write permission from any private vma. | |
1da177e4 | 1783 | */ |
044d66c1 | 1784 | static int unuse_pte(struct vm_area_struct *vma, pmd_t *pmd, |
1da177e4 LT |
1785 | unsigned long addr, swp_entry_t entry, struct page *page) |
1786 | { | |
9e16b7fb | 1787 | struct page *swapcache; |
72835c86 | 1788 | struct mem_cgroup *memcg; |
044d66c1 HD |
1789 | spinlock_t *ptl; |
1790 | pte_t *pte; | |
1791 | int ret = 1; | |
1792 | ||
9e16b7fb HD |
1793 | swapcache = page; |
1794 | page = ksm_might_need_to_copy(page, vma, addr); | |
1795 | if (unlikely(!page)) | |
1796 | return -ENOMEM; | |
1797 | ||
f627c2f5 KS |
1798 | if (mem_cgroup_try_charge(page, vma->vm_mm, GFP_KERNEL, |
1799 | &memcg, false)) { | |
044d66c1 | 1800 | ret = -ENOMEM; |
85d9fc89 KH |
1801 | goto out_nolock; |
1802 | } | |
044d66c1 HD |
1803 | |
1804 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); | |
9f8bdb3f | 1805 | if (unlikely(!pte_same_as_swp(*pte, swp_entry_to_pte(entry)))) { |
f627c2f5 | 1806 | mem_cgroup_cancel_charge(page, memcg, false); |
044d66c1 HD |
1807 | ret = 0; |
1808 | goto out; | |
1809 | } | |
8a9f3ccd | 1810 | |
b084d435 | 1811 | dec_mm_counter(vma->vm_mm, MM_SWAPENTS); |
d559db08 | 1812 | inc_mm_counter(vma->vm_mm, MM_ANONPAGES); |
1da177e4 LT |
1813 | get_page(page); |
1814 | set_pte_at(vma->vm_mm, addr, pte, | |
1815 | pte_mkold(mk_pte(page, vma->vm_page_prot))); | |
00501b53 | 1816 | if (page == swapcache) { |
d281ee61 | 1817 | page_add_anon_rmap(page, vma, addr, false); |
f627c2f5 | 1818 | mem_cgroup_commit_charge(page, memcg, true, false); |
00501b53 | 1819 | } else { /* ksm created a completely new copy */ |
d281ee61 | 1820 | page_add_new_anon_rmap(page, vma, addr, false); |
f627c2f5 | 1821 | mem_cgroup_commit_charge(page, memcg, false, false); |
00501b53 JW |
1822 | lru_cache_add_active_or_unevictable(page, vma); |
1823 | } | |
1da177e4 LT |
1824 | swap_free(entry); |
1825 | /* | |
1826 | * Move the page to the active list so it is not | |
1827 | * immediately swapped out again after swapon. | |
1828 | */ | |
1829 | activate_page(page); | |
044d66c1 HD |
1830 | out: |
1831 | pte_unmap_unlock(pte, ptl); | |
85d9fc89 | 1832 | out_nolock: |
9e16b7fb HD |
1833 | if (page != swapcache) { |
1834 | unlock_page(page); | |
1835 | put_page(page); | |
1836 | } | |
044d66c1 | 1837 | return ret; |
1da177e4 LT |
1838 | } |
1839 | ||
1840 | static int unuse_pte_range(struct vm_area_struct *vma, pmd_t *pmd, | |
1841 | unsigned long addr, unsigned long end, | |
1842 | swp_entry_t entry, struct page *page) | |
1843 | { | |
1da177e4 | 1844 | pte_t swp_pte = swp_entry_to_pte(entry); |
705e87c0 | 1845 | pte_t *pte; |
8a9f3ccd | 1846 | int ret = 0; |
1da177e4 | 1847 | |
044d66c1 HD |
1848 | /* |
1849 | * We don't actually need pte lock while scanning for swp_pte: since | |
1850 | * we hold page lock and mmap_sem, swp_pte cannot be inserted into the | |
1851 | * page table while we're scanning; though it could get zapped, and on | |
1852 | * some architectures (e.g. x86_32 with PAE) we might catch a glimpse | |
1853 | * of unmatched parts which look like swp_pte, so unuse_pte must | |
1854 | * recheck under pte lock. Scanning without pte lock lets it be | |
2de1a7e4 | 1855 | * preemptable whenever CONFIG_PREEMPT but not CONFIG_HIGHPTE. |
044d66c1 HD |
1856 | */ |
1857 | pte = pte_offset_map(pmd, addr); | |
1da177e4 LT |
1858 | do { |
1859 | /* | |
1860 | * swapoff spends a _lot_ of time in this loop! | |
1861 | * Test inline before going to call unuse_pte. | |
1862 | */ | |
9f8bdb3f | 1863 | if (unlikely(pte_same_as_swp(*pte, swp_pte))) { |
044d66c1 HD |
1864 | pte_unmap(pte); |
1865 | ret = unuse_pte(vma, pmd, addr, entry, page); | |
1866 | if (ret) | |
1867 | goto out; | |
1868 | pte = pte_offset_map(pmd, addr); | |
1da177e4 LT |
1869 | } |
1870 | } while (pte++, addr += PAGE_SIZE, addr != end); | |
044d66c1 HD |
1871 | pte_unmap(pte - 1); |
1872 | out: | |
8a9f3ccd | 1873 | return ret; |
1da177e4 LT |
1874 | } |
1875 | ||
1876 | static inline int unuse_pmd_range(struct vm_area_struct *vma, pud_t *pud, | |
1877 | unsigned long addr, unsigned long end, | |
1878 | swp_entry_t entry, struct page *page) | |
1879 | { | |
1880 | pmd_t *pmd; | |
1881 | unsigned long next; | |
8a9f3ccd | 1882 | int ret; |
1da177e4 LT |
1883 | |
1884 | pmd = pmd_offset(pud, addr); | |
1885 | do { | |
dc644a07 | 1886 | cond_resched(); |
1da177e4 | 1887 | next = pmd_addr_end(addr, end); |
1a5a9906 | 1888 | if (pmd_none_or_trans_huge_or_clear_bad(pmd)) |
1da177e4 | 1889 | continue; |
8a9f3ccd BS |
1890 | ret = unuse_pte_range(vma, pmd, addr, next, entry, page); |
1891 | if (ret) | |
1892 | return ret; | |
1da177e4 LT |
1893 | } while (pmd++, addr = next, addr != end); |
1894 | return 0; | |
1895 | } | |
1896 | ||
c2febafc | 1897 | static inline int unuse_pud_range(struct vm_area_struct *vma, p4d_t *p4d, |
1da177e4 LT |
1898 | unsigned long addr, unsigned long end, |
1899 | swp_entry_t entry, struct page *page) | |
1900 | { | |
1901 | pud_t *pud; | |
1902 | unsigned long next; | |
8a9f3ccd | 1903 | int ret; |
1da177e4 | 1904 | |
c2febafc | 1905 | pud = pud_offset(p4d, addr); |
1da177e4 LT |
1906 | do { |
1907 | next = pud_addr_end(addr, end); | |
1908 | if (pud_none_or_clear_bad(pud)) | |
1909 | continue; | |
8a9f3ccd BS |
1910 | ret = unuse_pmd_range(vma, pud, addr, next, entry, page); |
1911 | if (ret) | |
1912 | return ret; | |
1da177e4 LT |
1913 | } while (pud++, addr = next, addr != end); |
1914 | return 0; | |
1915 | } | |
1916 | ||
c2febafc KS |
1917 | static inline int unuse_p4d_range(struct vm_area_struct *vma, pgd_t *pgd, |
1918 | unsigned long addr, unsigned long end, | |
1919 | swp_entry_t entry, struct page *page) | |
1920 | { | |
1921 | p4d_t *p4d; | |
1922 | unsigned long next; | |
1923 | int ret; | |
1924 | ||
1925 | p4d = p4d_offset(pgd, addr); | |
1926 | do { | |
1927 | next = p4d_addr_end(addr, end); | |
1928 | if (p4d_none_or_clear_bad(p4d)) | |
1929 | continue; | |
1930 | ret = unuse_pud_range(vma, p4d, addr, next, entry, page); | |
1931 | if (ret) | |
1932 | return ret; | |
1933 | } while (p4d++, addr = next, addr != end); | |
1934 | return 0; | |
1935 | } | |
1936 | ||
1da177e4 LT |
1937 | static int unuse_vma(struct vm_area_struct *vma, |
1938 | swp_entry_t entry, struct page *page) | |
1939 | { | |
1940 | pgd_t *pgd; | |
1941 | unsigned long addr, end, next; | |
8a9f3ccd | 1942 | int ret; |
1da177e4 | 1943 | |
3ca7b3c5 | 1944 | if (page_anon_vma(page)) { |
1da177e4 LT |
1945 | addr = page_address_in_vma(page, vma); |
1946 | if (addr == -EFAULT) | |
1947 | return 0; | |
1948 | else | |
1949 | end = addr + PAGE_SIZE; | |
1950 | } else { | |
1951 | addr = vma->vm_start; | |
1952 | end = vma->vm_end; | |
1953 | } | |
1954 | ||
1955 | pgd = pgd_offset(vma->vm_mm, addr); | |
1956 | do { | |
1957 | next = pgd_addr_end(addr, end); | |
1958 | if (pgd_none_or_clear_bad(pgd)) | |
1959 | continue; | |
c2febafc | 1960 | ret = unuse_p4d_range(vma, pgd, addr, next, entry, page); |
8a9f3ccd BS |
1961 | if (ret) |
1962 | return ret; | |
1da177e4 LT |
1963 | } while (pgd++, addr = next, addr != end); |
1964 | return 0; | |
1965 | } | |
1966 | ||
1967 | static int unuse_mm(struct mm_struct *mm, | |
1968 | swp_entry_t entry, struct page *page) | |
1969 | { | |
1970 | struct vm_area_struct *vma; | |
8a9f3ccd | 1971 | int ret = 0; |
1da177e4 LT |
1972 | |
1973 | if (!down_read_trylock(&mm->mmap_sem)) { | |
1974 | /* | |
7d03431c FLVC |
1975 | * Activate page so shrink_inactive_list is unlikely to unmap |
1976 | * its ptes while lock is dropped, so swapoff can make progress. | |
1da177e4 | 1977 | */ |
c475a8ab | 1978 | activate_page(page); |
1da177e4 LT |
1979 | unlock_page(page); |
1980 | down_read(&mm->mmap_sem); | |
1981 | lock_page(page); | |
1982 | } | |
1da177e4 | 1983 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
8a9f3ccd | 1984 | if (vma->anon_vma && (ret = unuse_vma(vma, entry, page))) |
1da177e4 | 1985 | break; |
dc644a07 | 1986 | cond_resched(); |
1da177e4 | 1987 | } |
1da177e4 | 1988 | up_read(&mm->mmap_sem); |
8a9f3ccd | 1989 | return (ret < 0)? ret: 0; |
1da177e4 LT |
1990 | } |
1991 | ||
1992 | /* | |
38b5faf4 DM |
1993 | * Scan swap_map (or frontswap_map if frontswap parameter is true) |
1994 | * from current position to next entry still in use. | |
1da177e4 LT |
1995 | * Recycle to start on reaching the end, returning 0 when empty. |
1996 | */ | |
6eb396dc | 1997 | static unsigned int find_next_to_unuse(struct swap_info_struct *si, |
38b5faf4 | 1998 | unsigned int prev, bool frontswap) |
1da177e4 | 1999 | { |
6eb396dc HD |
2000 | unsigned int max = si->max; |
2001 | unsigned int i = prev; | |
8d69aaee | 2002 | unsigned char count; |
1da177e4 LT |
2003 | |
2004 | /* | |
5d337b91 | 2005 | * No need for swap_lock here: we're just looking |
1da177e4 LT |
2006 | * for whether an entry is in use, not modifying it; false |
2007 | * hits are okay, and sys_swapoff() has already prevented new | |
5d337b91 | 2008 | * allocations from this area (while holding swap_lock). |
1da177e4 LT |
2009 | */ |
2010 | for (;;) { | |
2011 | if (++i >= max) { | |
2012 | if (!prev) { | |
2013 | i = 0; | |
2014 | break; | |
2015 | } | |
2016 | /* | |
2017 | * No entries in use at top of swap_map, | |
2018 | * loop back to start and recheck there. | |
2019 | */ | |
2020 | max = prev + 1; | |
2021 | prev = 0; | |
2022 | i = 1; | |
2023 | } | |
4db0c3c2 | 2024 | count = READ_ONCE(si->swap_map[i]); |
355cfa73 | 2025 | if (count && swap_count(count) != SWAP_MAP_BAD) |
dc644a07 HD |
2026 | if (!frontswap || frontswap_test(si, i)) |
2027 | break; | |
2028 | if ((i % LATENCY_LIMIT) == 0) | |
2029 | cond_resched(); | |
1da177e4 LT |
2030 | } |
2031 | return i; | |
2032 | } | |
2033 | ||
2034 | /* | |
2035 | * We completely avoid races by reading each swap page in advance, | |
2036 | * and then search for the process using it. All the necessary | |
2037 | * page table adjustments can then be made atomically. | |
38b5faf4 DM |
2038 | * |
2039 | * if the boolean frontswap is true, only unuse pages_to_unuse pages; | |
2040 | * pages_to_unuse==0 means all pages; ignored if frontswap is false | |
1da177e4 | 2041 | */ |
38b5faf4 DM |
2042 | int try_to_unuse(unsigned int type, bool frontswap, |
2043 | unsigned long pages_to_unuse) | |
1da177e4 | 2044 | { |
efa90a98 | 2045 | struct swap_info_struct *si = swap_info[type]; |
1da177e4 | 2046 | struct mm_struct *start_mm; |
edfe23da SL |
2047 | volatile unsigned char *swap_map; /* swap_map is accessed without |
2048 | * locking. Mark it as volatile | |
2049 | * to prevent compiler doing | |
2050 | * something odd. | |
2051 | */ | |
8d69aaee | 2052 | unsigned char swcount; |
1da177e4 LT |
2053 | struct page *page; |
2054 | swp_entry_t entry; | |
6eb396dc | 2055 | unsigned int i = 0; |
1da177e4 | 2056 | int retval = 0; |
1da177e4 LT |
2057 | |
2058 | /* | |
2059 | * When searching mms for an entry, a good strategy is to | |
2060 | * start at the first mm we freed the previous entry from | |
2061 | * (though actually we don't notice whether we or coincidence | |
2062 | * freed the entry). Initialize this start_mm with a hold. | |
2063 | * | |
2064 | * A simpler strategy would be to start at the last mm we | |
2065 | * freed the previous entry from; but that would take less | |
2066 | * advantage of mmlist ordering, which clusters forked mms | |
2067 | * together, child after parent. If we race with dup_mmap(), we | |
2068 | * prefer to resolve parent before child, lest we miss entries | |
2069 | * duplicated after we scanned child: using last mm would invert | |
570a335b | 2070 | * that. |
1da177e4 LT |
2071 | */ |
2072 | start_mm = &init_mm; | |
3fce371b | 2073 | mmget(&init_mm); |
1da177e4 LT |
2074 | |
2075 | /* | |
2076 | * Keep on scanning until all entries have gone. Usually, | |
2077 | * one pass through swap_map is enough, but not necessarily: | |
2078 | * there are races when an instance of an entry might be missed. | |
2079 | */ | |
38b5faf4 | 2080 | while ((i = find_next_to_unuse(si, i, frontswap)) != 0) { |
1da177e4 LT |
2081 | if (signal_pending(current)) { |
2082 | retval = -EINTR; | |
2083 | break; | |
2084 | } | |
2085 | ||
886bb7e9 | 2086 | /* |
1da177e4 LT |
2087 | * Get a page for the entry, using the existing swap |
2088 | * cache page if there is one. Otherwise, get a clean | |
886bb7e9 | 2089 | * page and read the swap into it. |
1da177e4 LT |
2090 | */ |
2091 | swap_map = &si->swap_map[i]; | |
2092 | entry = swp_entry(type, i); | |
02098fea | 2093 | page = read_swap_cache_async(entry, |
23955622 | 2094 | GFP_HIGHUSER_MOVABLE, NULL, 0, false); |
1da177e4 LT |
2095 | if (!page) { |
2096 | /* | |
2097 | * Either swap_duplicate() failed because entry | |
2098 | * has been freed independently, and will not be | |
2099 | * reused since sys_swapoff() already disabled | |
2100 | * allocation from here, or alloc_page() failed. | |
2101 | */ | |
edfe23da SL |
2102 | swcount = *swap_map; |
2103 | /* | |
2104 | * We don't hold lock here, so the swap entry could be | |
2105 | * SWAP_MAP_BAD (when the cluster is discarding). | |
2106 | * Instead of fail out, We can just skip the swap | |
2107 | * entry because swapoff will wait for discarding | |
2108 | * finish anyway. | |
2109 | */ | |
2110 | if (!swcount || swcount == SWAP_MAP_BAD) | |
1da177e4 LT |
2111 | continue; |
2112 | retval = -ENOMEM; | |
2113 | break; | |
2114 | } | |
2115 | ||
2116 | /* | |
2117 | * Don't hold on to start_mm if it looks like exiting. | |
2118 | */ | |
2119 | if (atomic_read(&start_mm->mm_users) == 1) { | |
2120 | mmput(start_mm); | |
2121 | start_mm = &init_mm; | |
3fce371b | 2122 | mmget(&init_mm); |
1da177e4 LT |
2123 | } |
2124 | ||
2125 | /* | |
2126 | * Wait for and lock page. When do_swap_page races with | |
2127 | * try_to_unuse, do_swap_page can handle the fault much | |
2128 | * faster than try_to_unuse can locate the entry. This | |
2129 | * apparently redundant "wait_on_page_locked" lets try_to_unuse | |
2130 | * defer to do_swap_page in such a case - in some tests, | |
2131 | * do_swap_page and try_to_unuse repeatedly compete. | |
2132 | */ | |
2133 | wait_on_page_locked(page); | |
2134 | wait_on_page_writeback(page); | |
2135 | lock_page(page); | |
2136 | wait_on_page_writeback(page); | |
2137 | ||
2138 | /* | |
2139 | * Remove all references to entry. | |
1da177e4 | 2140 | */ |
1da177e4 | 2141 | swcount = *swap_map; |
aaa46865 HD |
2142 | if (swap_count(swcount) == SWAP_MAP_SHMEM) { |
2143 | retval = shmem_unuse(entry, page); | |
2144 | /* page has already been unlocked and released */ | |
2145 | if (retval < 0) | |
2146 | break; | |
2147 | continue; | |
1da177e4 | 2148 | } |
aaa46865 HD |
2149 | if (swap_count(swcount) && start_mm != &init_mm) |
2150 | retval = unuse_mm(start_mm, entry, page); | |
2151 | ||
355cfa73 | 2152 | if (swap_count(*swap_map)) { |
1da177e4 LT |
2153 | int set_start_mm = (*swap_map >= swcount); |
2154 | struct list_head *p = &start_mm->mmlist; | |
2155 | struct mm_struct *new_start_mm = start_mm; | |
2156 | struct mm_struct *prev_mm = start_mm; | |
2157 | struct mm_struct *mm; | |
2158 | ||
3fce371b VN |
2159 | mmget(new_start_mm); |
2160 | mmget(prev_mm); | |
1da177e4 | 2161 | spin_lock(&mmlist_lock); |
aaa46865 | 2162 | while (swap_count(*swap_map) && !retval && |
1da177e4 LT |
2163 | (p = p->next) != &start_mm->mmlist) { |
2164 | mm = list_entry(p, struct mm_struct, mmlist); | |
388f7934 | 2165 | if (!mmget_not_zero(mm)) |
1da177e4 | 2166 | continue; |
1da177e4 LT |
2167 | spin_unlock(&mmlist_lock); |
2168 | mmput(prev_mm); | |
2169 | prev_mm = mm; | |
2170 | ||
2171 | cond_resched(); | |
2172 | ||
2173 | swcount = *swap_map; | |
355cfa73 | 2174 | if (!swap_count(swcount)) /* any usage ? */ |
1da177e4 | 2175 | ; |
aaa46865 | 2176 | else if (mm == &init_mm) |
1da177e4 | 2177 | set_start_mm = 1; |
aaa46865 | 2178 | else |
1da177e4 | 2179 | retval = unuse_mm(mm, entry, page); |
355cfa73 | 2180 | |
32c5fc10 | 2181 | if (set_start_mm && *swap_map < swcount) { |
1da177e4 | 2182 | mmput(new_start_mm); |
3fce371b | 2183 | mmget(mm); |
1da177e4 LT |
2184 | new_start_mm = mm; |
2185 | set_start_mm = 0; | |
2186 | } | |
2187 | spin_lock(&mmlist_lock); | |
2188 | } | |
2189 | spin_unlock(&mmlist_lock); | |
2190 | mmput(prev_mm); | |
2191 | mmput(start_mm); | |
2192 | start_mm = new_start_mm; | |
2193 | } | |
2194 | if (retval) { | |
2195 | unlock_page(page); | |
09cbfeaf | 2196 | put_page(page); |
1da177e4 LT |
2197 | break; |
2198 | } | |
2199 | ||
1da177e4 LT |
2200 | /* |
2201 | * If a reference remains (rare), we would like to leave | |
2202 | * the page in the swap cache; but try_to_unmap could | |
2203 | * then re-duplicate the entry once we drop page lock, | |
2204 | * so we might loop indefinitely; also, that page could | |
2205 | * not be swapped out to other storage meanwhile. So: | |
2206 | * delete from cache even if there's another reference, | |
2207 | * after ensuring that the data has been saved to disk - | |
2208 | * since if the reference remains (rarer), it will be | |
2209 | * read from disk into another page. Splitting into two | |
2210 | * pages would be incorrect if swap supported "shared | |
2211 | * private" pages, but they are handled by tmpfs files. | |
5ad64688 HD |
2212 | * |
2213 | * Given how unuse_vma() targets one particular offset | |
2214 | * in an anon_vma, once the anon_vma has been determined, | |
2215 | * this splitting happens to be just what is needed to | |
2216 | * handle where KSM pages have been swapped out: re-reading | |
2217 | * is unnecessarily slow, but we can fix that later on. | |
1da177e4 | 2218 | */ |
355cfa73 KH |
2219 | if (swap_count(*swap_map) && |
2220 | PageDirty(page) && PageSwapCache(page)) { | |
1da177e4 LT |
2221 | struct writeback_control wbc = { |
2222 | .sync_mode = WB_SYNC_NONE, | |
2223 | }; | |
2224 | ||
e0709829 | 2225 | swap_writepage(compound_head(page), &wbc); |
1da177e4 LT |
2226 | lock_page(page); |
2227 | wait_on_page_writeback(page); | |
2228 | } | |
68bdc8d6 HD |
2229 | |
2230 | /* | |
2231 | * It is conceivable that a racing task removed this page from | |
2232 | * swap cache just before we acquired the page lock at the top, | |
2233 | * or while we dropped it in unuse_mm(). The page might even | |
2234 | * be back in swap cache on another swap area: that we must not | |
2235 | * delete, since it may not have been written out to swap yet. | |
2236 | */ | |
2237 | if (PageSwapCache(page) && | |
e0709829 | 2238 | likely(page_private(page) == entry.val) && |
e25b4242 HY |
2239 | (!PageTransCompound(page) || |
2240 | !swap_page_trans_huge_swapped(si, entry))) | |
e0709829 | 2241 | delete_from_swap_cache(compound_head(page)); |
1da177e4 LT |
2242 | |
2243 | /* | |
2244 | * So we could skip searching mms once swap count went | |
2245 | * to 1, we did not mark any present ptes as dirty: must | |
2706a1b8 | 2246 | * mark page dirty so shrink_page_list will preserve it. |
1da177e4 LT |
2247 | */ |
2248 | SetPageDirty(page); | |
2249 | unlock_page(page); | |
09cbfeaf | 2250 | put_page(page); |
1da177e4 LT |
2251 | |
2252 | /* | |
2253 | * Make sure that we aren't completely killing | |
2254 | * interactive performance. | |
2255 | */ | |
2256 | cond_resched(); | |
38b5faf4 DM |
2257 | if (frontswap && pages_to_unuse > 0) { |
2258 | if (!--pages_to_unuse) | |
2259 | break; | |
2260 | } | |
1da177e4 LT |
2261 | } |
2262 | ||
2263 | mmput(start_mm); | |
1da177e4 LT |
2264 | return retval; |
2265 | } | |
2266 | ||
2267 | /* | |
5d337b91 HD |
2268 | * After a successful try_to_unuse, if no swap is now in use, we know |
2269 | * we can empty the mmlist. swap_lock must be held on entry and exit. | |
2270 | * Note that mmlist_lock nests inside swap_lock, and an mm must be | |
1da177e4 LT |
2271 | * added to the mmlist just after page_duplicate - before would be racy. |
2272 | */ | |
2273 | static void drain_mmlist(void) | |
2274 | { | |
2275 | struct list_head *p, *next; | |
efa90a98 | 2276 | unsigned int type; |
1da177e4 | 2277 | |
efa90a98 HD |
2278 | for (type = 0; type < nr_swapfiles; type++) |
2279 | if (swap_info[type]->inuse_pages) | |
1da177e4 LT |
2280 | return; |
2281 | spin_lock(&mmlist_lock); | |
2282 | list_for_each_safe(p, next, &init_mm.mmlist) | |
2283 | list_del_init(p); | |
2284 | spin_unlock(&mmlist_lock); | |
2285 | } | |
2286 | ||
2287 | /* | |
2288 | * Use this swapdev's extent info to locate the (PAGE_SIZE) block which | |
d4906e1a LS |
2289 | * corresponds to page offset for the specified swap entry. |
2290 | * Note that the type of this function is sector_t, but it returns page offset | |
2291 | * into the bdev, not sector offset. | |
1da177e4 | 2292 | */ |
d4906e1a | 2293 | static sector_t map_swap_entry(swp_entry_t entry, struct block_device **bdev) |
1da177e4 | 2294 | { |
f29ad6a9 HD |
2295 | struct swap_info_struct *sis; |
2296 | struct swap_extent *start_se; | |
2297 | struct swap_extent *se; | |
2298 | pgoff_t offset; | |
2299 | ||
27a34b2d | 2300 | sis = swp_swap_info(entry); |
f29ad6a9 HD |
2301 | *bdev = sis->bdev; |
2302 | ||
2303 | offset = swp_offset(entry); | |
2304 | start_se = sis->curr_swap_extent; | |
2305 | se = start_se; | |
1da177e4 LT |
2306 | |
2307 | for ( ; ; ) { | |
1da177e4 LT |
2308 | if (se->start_page <= offset && |
2309 | offset < (se->start_page + se->nr_pages)) { | |
2310 | return se->start_block + (offset - se->start_page); | |
2311 | } | |
a8ae4991 | 2312 | se = list_next_entry(se, list); |
1da177e4 LT |
2313 | sis->curr_swap_extent = se; |
2314 | BUG_ON(se == start_se); /* It *must* be present */ | |
2315 | } | |
2316 | } | |
2317 | ||
d4906e1a LS |
2318 | /* |
2319 | * Returns the page offset into bdev for the specified page's swap entry. | |
2320 | */ | |
2321 | sector_t map_swap_page(struct page *page, struct block_device **bdev) | |
2322 | { | |
2323 | swp_entry_t entry; | |
2324 | entry.val = page_private(page); | |
2325 | return map_swap_entry(entry, bdev); | |
2326 | } | |
2327 | ||
1da177e4 LT |
2328 | /* |
2329 | * Free all of a swapdev's extent information | |
2330 | */ | |
2331 | static void destroy_swap_extents(struct swap_info_struct *sis) | |
2332 | { | |
9625a5f2 | 2333 | while (!list_empty(&sis->first_swap_extent.list)) { |
1da177e4 LT |
2334 | struct swap_extent *se; |
2335 | ||
a8ae4991 | 2336 | se = list_first_entry(&sis->first_swap_extent.list, |
1da177e4 LT |
2337 | struct swap_extent, list); |
2338 | list_del(&se->list); | |
2339 | kfree(se); | |
2340 | } | |
62c230bc MG |
2341 | |
2342 | if (sis->flags & SWP_FILE) { | |
2343 | struct file *swap_file = sis->swap_file; | |
2344 | struct address_space *mapping = swap_file->f_mapping; | |
2345 | ||
2346 | sis->flags &= ~SWP_FILE; | |
2347 | mapping->a_ops->swap_deactivate(swap_file); | |
2348 | } | |
1da177e4 LT |
2349 | } |
2350 | ||
2351 | /* | |
2352 | * Add a block range (and the corresponding page range) into this swapdev's | |
11d31886 | 2353 | * extent list. The extent list is kept sorted in page order. |
1da177e4 | 2354 | * |
11d31886 | 2355 | * This function rather assumes that it is called in ascending page order. |
1da177e4 | 2356 | */ |
a509bc1a | 2357 | int |
1da177e4 LT |
2358 | add_swap_extent(struct swap_info_struct *sis, unsigned long start_page, |
2359 | unsigned long nr_pages, sector_t start_block) | |
2360 | { | |
2361 | struct swap_extent *se; | |
2362 | struct swap_extent *new_se; | |
2363 | struct list_head *lh; | |
2364 | ||
9625a5f2 HD |
2365 | if (start_page == 0) { |
2366 | se = &sis->first_swap_extent; | |
2367 | sis->curr_swap_extent = se; | |
2368 | se->start_page = 0; | |
2369 | se->nr_pages = nr_pages; | |
2370 | se->start_block = start_block; | |
2371 | return 1; | |
2372 | } else { | |
2373 | lh = sis->first_swap_extent.list.prev; /* Highest extent */ | |
1da177e4 | 2374 | se = list_entry(lh, struct swap_extent, list); |
11d31886 HD |
2375 | BUG_ON(se->start_page + se->nr_pages != start_page); |
2376 | if (se->start_block + se->nr_pages == start_block) { | |
1da177e4 LT |
2377 | /* Merge it */ |
2378 | se->nr_pages += nr_pages; | |
2379 | return 0; | |
2380 | } | |
1da177e4 LT |
2381 | } |
2382 | ||
2383 | /* | |
2384 | * No merge. Insert a new extent, preserving ordering. | |
2385 | */ | |
2386 | new_se = kmalloc(sizeof(*se), GFP_KERNEL); | |
2387 | if (new_se == NULL) | |
2388 | return -ENOMEM; | |
2389 | new_se->start_page = start_page; | |
2390 | new_se->nr_pages = nr_pages; | |
2391 | new_se->start_block = start_block; | |
2392 | ||
9625a5f2 | 2393 | list_add_tail(&new_se->list, &sis->first_swap_extent.list); |
53092a74 | 2394 | return 1; |
1da177e4 LT |
2395 | } |
2396 | ||
2397 | /* | |
2398 | * A `swap extent' is a simple thing which maps a contiguous range of pages | |
2399 | * onto a contiguous range of disk blocks. An ordered list of swap extents | |
2400 | * is built at swapon time and is then used at swap_writepage/swap_readpage | |
2401 | * time for locating where on disk a page belongs. | |
2402 | * | |
2403 | * If the swapfile is an S_ISBLK block device, a single extent is installed. | |
2404 | * This is done so that the main operating code can treat S_ISBLK and S_ISREG | |
2405 | * swap files identically. | |
2406 | * | |
2407 | * Whether the swapdev is an S_ISREG file or an S_ISBLK blockdev, the swap | |
2408 | * extent list operates in PAGE_SIZE disk blocks. Both S_ISREG and S_ISBLK | |
2409 | * swapfiles are handled *identically* after swapon time. | |
2410 | * | |
2411 | * For S_ISREG swapfiles, setup_swap_extents() will walk all the file's blocks | |
2412 | * and will parse them into an ordered extent list, in PAGE_SIZE chunks. If | |
2413 | * some stray blocks are found which do not fall within the PAGE_SIZE alignment | |
2414 | * requirements, they are simply tossed out - we will never use those blocks | |
2415 | * for swapping. | |
2416 | * | |
b0d9bcd4 | 2417 | * For S_ISREG swapfiles we set S_SWAPFILE across the life of the swapon. This |
1da177e4 LT |
2418 | * prevents root from shooting her foot off by ftruncating an in-use swapfile, |
2419 | * which will scribble on the fs. | |
2420 | * | |
2421 | * The amount of disk space which a single swap extent represents varies. | |
2422 | * Typically it is in the 1-4 megabyte range. So we can have hundreds of | |
2423 | * extents in the list. To avoid much list walking, we cache the previous | |
2424 | * search location in `curr_swap_extent', and start new searches from there. | |
2425 | * This is extremely effective. The average number of iterations in | |
2426 | * map_swap_page() has been measured at about 0.3 per page. - akpm. | |
2427 | */ | |
53092a74 | 2428 | static int setup_swap_extents(struct swap_info_struct *sis, sector_t *span) |
1da177e4 | 2429 | { |
62c230bc MG |
2430 | struct file *swap_file = sis->swap_file; |
2431 | struct address_space *mapping = swap_file->f_mapping; | |
2432 | struct inode *inode = mapping->host; | |
1da177e4 LT |
2433 | int ret; |
2434 | ||
1da177e4 LT |
2435 | if (S_ISBLK(inode->i_mode)) { |
2436 | ret = add_swap_extent(sis, 0, sis->max, 0); | |
53092a74 | 2437 | *span = sis->pages; |
a509bc1a | 2438 | return ret; |
1da177e4 LT |
2439 | } |
2440 | ||
62c230bc | 2441 | if (mapping->a_ops->swap_activate) { |
a509bc1a | 2442 | ret = mapping->a_ops->swap_activate(sis, swap_file, span); |
62c230bc MG |
2443 | if (!ret) { |
2444 | sis->flags |= SWP_FILE; | |
2445 | ret = add_swap_extent(sis, 0, sis->max, 0); | |
2446 | *span = sis->pages; | |
2447 | } | |
a509bc1a | 2448 | return ret; |
62c230bc MG |
2449 | } |
2450 | ||
a509bc1a | 2451 | return generic_swapfile_activate(sis, swap_file, span); |
1da177e4 LT |
2452 | } |
2453 | ||
a2468cc9 AL |
2454 | static int swap_node(struct swap_info_struct *p) |
2455 | { | |
2456 | struct block_device *bdev; | |
2457 | ||
2458 | if (p->bdev) | |
2459 | bdev = p->bdev; | |
2460 | else | |
2461 | bdev = p->swap_file->f_inode->i_sb->s_bdev; | |
2462 | ||
2463 | return bdev ? bdev->bd_disk->node_id : NUMA_NO_NODE; | |
2464 | } | |
2465 | ||
cf0cac0a | 2466 | static void _enable_swap_info(struct swap_info_struct *p, int prio, |
2a8f9449 SL |
2467 | unsigned char *swap_map, |
2468 | struct swap_cluster_info *cluster_info) | |
40531542 | 2469 | { |
a2468cc9 AL |
2470 | int i; |
2471 | ||
40531542 CEB |
2472 | if (prio >= 0) |
2473 | p->prio = prio; | |
2474 | else | |
2475 | p->prio = --least_priority; | |
18ab4d4c DS |
2476 | /* |
2477 | * the plist prio is negated because plist ordering is | |
2478 | * low-to-high, while swap ordering is high-to-low | |
2479 | */ | |
2480 | p->list.prio = -p->prio; | |
a2468cc9 AL |
2481 | for_each_node(i) { |
2482 | if (p->prio >= 0) | |
2483 | p->avail_lists[i].prio = -p->prio; | |
2484 | else { | |
2485 | if (swap_node(p) == i) | |
2486 | p->avail_lists[i].prio = 1; | |
2487 | else | |
2488 | p->avail_lists[i].prio = -p->prio; | |
2489 | } | |
2490 | } | |
40531542 | 2491 | p->swap_map = swap_map; |
2a8f9449 | 2492 | p->cluster_info = cluster_info; |
40531542 | 2493 | p->flags |= SWP_WRITEOK; |
ec8acf20 | 2494 | atomic_long_add(p->pages, &nr_swap_pages); |
40531542 CEB |
2495 | total_swap_pages += p->pages; |
2496 | ||
adfab836 | 2497 | assert_spin_locked(&swap_lock); |
adfab836 | 2498 | /* |
18ab4d4c DS |
2499 | * both lists are plists, and thus priority ordered. |
2500 | * swap_active_head needs to be priority ordered for swapoff(), | |
2501 | * which on removal of any swap_info_struct with an auto-assigned | |
2502 | * (i.e. negative) priority increments the auto-assigned priority | |
2503 | * of any lower-priority swap_info_structs. | |
2504 | * swap_avail_head needs to be priority ordered for get_swap_page(), | |
2505 | * which allocates swap pages from the highest available priority | |
2506 | * swap_info_struct. | |
adfab836 | 2507 | */ |
18ab4d4c | 2508 | plist_add(&p->list, &swap_active_head); |
a2468cc9 | 2509 | add_to_avail_list(p); |
cf0cac0a CEB |
2510 | } |
2511 | ||
2512 | static void enable_swap_info(struct swap_info_struct *p, int prio, | |
2513 | unsigned char *swap_map, | |
2a8f9449 | 2514 | struct swap_cluster_info *cluster_info, |
cf0cac0a CEB |
2515 | unsigned long *frontswap_map) |
2516 | { | |
4f89849d | 2517 | frontswap_init(p->type, frontswap_map); |
cf0cac0a | 2518 | spin_lock(&swap_lock); |
ec8acf20 | 2519 | spin_lock(&p->lock); |
2a8f9449 | 2520 | _enable_swap_info(p, prio, swap_map, cluster_info); |
ec8acf20 | 2521 | spin_unlock(&p->lock); |
cf0cac0a CEB |
2522 | spin_unlock(&swap_lock); |
2523 | } | |
2524 | ||
2525 | static void reinsert_swap_info(struct swap_info_struct *p) | |
2526 | { | |
2527 | spin_lock(&swap_lock); | |
ec8acf20 | 2528 | spin_lock(&p->lock); |
2a8f9449 | 2529 | _enable_swap_info(p, p->prio, p->swap_map, p->cluster_info); |
ec8acf20 | 2530 | spin_unlock(&p->lock); |
40531542 CEB |
2531 | spin_unlock(&swap_lock); |
2532 | } | |
2533 | ||
67afa38e TC |
2534 | bool has_usable_swap(void) |
2535 | { | |
2536 | bool ret = true; | |
2537 | ||
2538 | spin_lock(&swap_lock); | |
2539 | if (plist_head_empty(&swap_active_head)) | |
2540 | ret = false; | |
2541 | spin_unlock(&swap_lock); | |
2542 | return ret; | |
2543 | } | |
2544 | ||
c4ea37c2 | 2545 | SYSCALL_DEFINE1(swapoff, const char __user *, specialfile) |
1da177e4 | 2546 | { |
73c34b6a | 2547 | struct swap_info_struct *p = NULL; |
8d69aaee | 2548 | unsigned char *swap_map; |
2a8f9449 | 2549 | struct swap_cluster_info *cluster_info; |
4f89849d | 2550 | unsigned long *frontswap_map; |
1da177e4 LT |
2551 | struct file *swap_file, *victim; |
2552 | struct address_space *mapping; | |
2553 | struct inode *inode; | |
91a27b2a | 2554 | struct filename *pathname; |
adfab836 | 2555 | int err, found = 0; |
5b808a23 | 2556 | unsigned int old_block_size; |
886bb7e9 | 2557 | |
1da177e4 LT |
2558 | if (!capable(CAP_SYS_ADMIN)) |
2559 | return -EPERM; | |
2560 | ||
191c5424 AV |
2561 | BUG_ON(!current->mm); |
2562 | ||
1da177e4 | 2563 | pathname = getname(specialfile); |
1da177e4 | 2564 | if (IS_ERR(pathname)) |
f58b59c1 | 2565 | return PTR_ERR(pathname); |
1da177e4 | 2566 | |
669abf4e | 2567 | victim = file_open_name(pathname, O_RDWR|O_LARGEFILE, 0); |
1da177e4 LT |
2568 | err = PTR_ERR(victim); |
2569 | if (IS_ERR(victim)) | |
2570 | goto out; | |
2571 | ||
2572 | mapping = victim->f_mapping; | |
5d337b91 | 2573 | spin_lock(&swap_lock); |
18ab4d4c | 2574 | plist_for_each_entry(p, &swap_active_head, list) { |
22c6f8fd | 2575 | if (p->flags & SWP_WRITEOK) { |
adfab836 DS |
2576 | if (p->swap_file->f_mapping == mapping) { |
2577 | found = 1; | |
1da177e4 | 2578 | break; |
adfab836 | 2579 | } |
1da177e4 | 2580 | } |
1da177e4 | 2581 | } |
adfab836 | 2582 | if (!found) { |
1da177e4 | 2583 | err = -EINVAL; |
5d337b91 | 2584 | spin_unlock(&swap_lock); |
1da177e4 LT |
2585 | goto out_dput; |
2586 | } | |
191c5424 | 2587 | if (!security_vm_enough_memory_mm(current->mm, p->pages)) |
1da177e4 LT |
2588 | vm_unacct_memory(p->pages); |
2589 | else { | |
2590 | err = -ENOMEM; | |
5d337b91 | 2591 | spin_unlock(&swap_lock); |
1da177e4 LT |
2592 | goto out_dput; |
2593 | } | |
a2468cc9 | 2594 | del_from_avail_list(p); |
ec8acf20 | 2595 | spin_lock(&p->lock); |
78ecba08 | 2596 | if (p->prio < 0) { |
adfab836 | 2597 | struct swap_info_struct *si = p; |
a2468cc9 | 2598 | int nid; |
adfab836 | 2599 | |
18ab4d4c | 2600 | plist_for_each_entry_continue(si, &swap_active_head, list) { |
adfab836 | 2601 | si->prio++; |
18ab4d4c | 2602 | si->list.prio--; |
a2468cc9 AL |
2603 | for_each_node(nid) { |
2604 | if (si->avail_lists[nid].prio != 1) | |
2605 | si->avail_lists[nid].prio--; | |
2606 | } | |
adfab836 | 2607 | } |
78ecba08 HD |
2608 | least_priority++; |
2609 | } | |
18ab4d4c | 2610 | plist_del(&p->list, &swap_active_head); |
ec8acf20 | 2611 | atomic_long_sub(p->pages, &nr_swap_pages); |
1da177e4 LT |
2612 | total_swap_pages -= p->pages; |
2613 | p->flags &= ~SWP_WRITEOK; | |
ec8acf20 | 2614 | spin_unlock(&p->lock); |
5d337b91 | 2615 | spin_unlock(&swap_lock); |
fb4f88dc | 2616 | |
039939a6 TC |
2617 | disable_swap_slots_cache_lock(); |
2618 | ||
e1e12d2f | 2619 | set_current_oom_origin(); |
adfab836 | 2620 | err = try_to_unuse(p->type, false, 0); /* force unuse all pages */ |
e1e12d2f | 2621 | clear_current_oom_origin(); |
1da177e4 | 2622 | |
1da177e4 LT |
2623 | if (err) { |
2624 | /* re-insert swap space back into swap_list */ | |
cf0cac0a | 2625 | reinsert_swap_info(p); |
039939a6 | 2626 | reenable_swap_slots_cache_unlock(); |
1da177e4 LT |
2627 | goto out_dput; |
2628 | } | |
52b7efdb | 2629 | |
039939a6 TC |
2630 | reenable_swap_slots_cache_unlock(); |
2631 | ||
815c2c54 SL |
2632 | flush_work(&p->discard_work); |
2633 | ||
5d337b91 | 2634 | destroy_swap_extents(p); |
570a335b HD |
2635 | if (p->flags & SWP_CONTINUED) |
2636 | free_swap_count_continuations(p); | |
2637 | ||
81a0298b HY |
2638 | if (!p->bdev || !blk_queue_nonrot(bdev_get_queue(p->bdev))) |
2639 | atomic_dec(&nr_rotate_swap); | |
2640 | ||
fc0abb14 | 2641 | mutex_lock(&swapon_mutex); |
5d337b91 | 2642 | spin_lock(&swap_lock); |
ec8acf20 | 2643 | spin_lock(&p->lock); |
5d337b91 HD |
2644 | drain_mmlist(); |
2645 | ||
52b7efdb | 2646 | /* wait for anyone still in scan_swap_map */ |
52b7efdb HD |
2647 | p->highest_bit = 0; /* cuts scans short */ |
2648 | while (p->flags >= SWP_SCANNING) { | |
ec8acf20 | 2649 | spin_unlock(&p->lock); |
5d337b91 | 2650 | spin_unlock(&swap_lock); |
13e4b57f | 2651 | schedule_timeout_uninterruptible(1); |
5d337b91 | 2652 | spin_lock(&swap_lock); |
ec8acf20 | 2653 | spin_lock(&p->lock); |
52b7efdb | 2654 | } |
52b7efdb | 2655 | |
1da177e4 | 2656 | swap_file = p->swap_file; |
5b808a23 | 2657 | old_block_size = p->old_block_size; |
1da177e4 LT |
2658 | p->swap_file = NULL; |
2659 | p->max = 0; | |
2660 | swap_map = p->swap_map; | |
2661 | p->swap_map = NULL; | |
2a8f9449 SL |
2662 | cluster_info = p->cluster_info; |
2663 | p->cluster_info = NULL; | |
4f89849d | 2664 | frontswap_map = frontswap_map_get(p); |
ec8acf20 | 2665 | spin_unlock(&p->lock); |
5d337b91 | 2666 | spin_unlock(&swap_lock); |
adfab836 | 2667 | frontswap_invalidate_area(p->type); |
58e97ba6 | 2668 | frontswap_map_set(p, NULL); |
fc0abb14 | 2669 | mutex_unlock(&swapon_mutex); |
ebc2a1a6 SL |
2670 | free_percpu(p->percpu_cluster); |
2671 | p->percpu_cluster = NULL; | |
1da177e4 | 2672 | vfree(swap_map); |
54f180d3 HY |
2673 | kvfree(cluster_info); |
2674 | kvfree(frontswap_map); | |
2de1a7e4 | 2675 | /* Destroy swap account information */ |
adfab836 | 2676 | swap_cgroup_swapoff(p->type); |
4b3ef9da | 2677 | exit_swap_address_space(p->type); |
27a7faa0 | 2678 | |
1da177e4 LT |
2679 | inode = mapping->host; |
2680 | if (S_ISBLK(inode->i_mode)) { | |
2681 | struct block_device *bdev = I_BDEV(inode); | |
5b808a23 | 2682 | set_blocksize(bdev, old_block_size); |
e525fd89 | 2683 | blkdev_put(bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL); |
1da177e4 | 2684 | } else { |
5955102c | 2685 | inode_lock(inode); |
1da177e4 | 2686 | inode->i_flags &= ~S_SWAPFILE; |
5955102c | 2687 | inode_unlock(inode); |
1da177e4 LT |
2688 | } |
2689 | filp_close(swap_file, NULL); | |
f893ab41 WY |
2690 | |
2691 | /* | |
2692 | * Clear the SWP_USED flag after all resources are freed so that swapon | |
2693 | * can reuse this swap_info in alloc_swap_info() safely. It is ok to | |
2694 | * not hold p->lock after we cleared its SWP_WRITEOK. | |
2695 | */ | |
2696 | spin_lock(&swap_lock); | |
2697 | p->flags = 0; | |
2698 | spin_unlock(&swap_lock); | |
2699 | ||
1da177e4 | 2700 | err = 0; |
66d7dd51 KS |
2701 | atomic_inc(&proc_poll_event); |
2702 | wake_up_interruptible(&proc_poll_wait); | |
1da177e4 LT |
2703 | |
2704 | out_dput: | |
2705 | filp_close(victim, NULL); | |
2706 | out: | |
f58b59c1 | 2707 | putname(pathname); |
1da177e4 LT |
2708 | return err; |
2709 | } | |
2710 | ||
2711 | #ifdef CONFIG_PROC_FS | |
66d7dd51 KS |
2712 | static unsigned swaps_poll(struct file *file, poll_table *wait) |
2713 | { | |
f1514638 | 2714 | struct seq_file *seq = file->private_data; |
66d7dd51 KS |
2715 | |
2716 | poll_wait(file, &proc_poll_wait, wait); | |
2717 | ||
f1514638 KS |
2718 | if (seq->poll_event != atomic_read(&proc_poll_event)) { |
2719 | seq->poll_event = atomic_read(&proc_poll_event); | |
66d7dd51 KS |
2720 | return POLLIN | POLLRDNORM | POLLERR | POLLPRI; |
2721 | } | |
2722 | ||
2723 | return POLLIN | POLLRDNORM; | |
2724 | } | |
2725 | ||
1da177e4 LT |
2726 | /* iterator */ |
2727 | static void *swap_start(struct seq_file *swap, loff_t *pos) | |
2728 | { | |
efa90a98 HD |
2729 | struct swap_info_struct *si; |
2730 | int type; | |
1da177e4 LT |
2731 | loff_t l = *pos; |
2732 | ||
fc0abb14 | 2733 | mutex_lock(&swapon_mutex); |
1da177e4 | 2734 | |
881e4aab SS |
2735 | if (!l) |
2736 | return SEQ_START_TOKEN; | |
2737 | ||
27a34b2d | 2738 | for (type = 0; (si = swap_type_to_swap_info(type)); type++) { |
efa90a98 | 2739 | if (!(si->flags & SWP_USED) || !si->swap_map) |
1da177e4 | 2740 | continue; |
881e4aab | 2741 | if (!--l) |
efa90a98 | 2742 | return si; |
1da177e4 LT |
2743 | } |
2744 | ||
2745 | return NULL; | |
2746 | } | |
2747 | ||
2748 | static void *swap_next(struct seq_file *swap, void *v, loff_t *pos) | |
2749 | { | |
efa90a98 HD |
2750 | struct swap_info_struct *si = v; |
2751 | int type; | |
1da177e4 | 2752 | |
881e4aab | 2753 | if (v == SEQ_START_TOKEN) |
efa90a98 HD |
2754 | type = 0; |
2755 | else | |
2756 | type = si->type + 1; | |
881e4aab | 2757 | |
27a34b2d | 2758 | for (; (si = swap_type_to_swap_info(type)); type++) { |
efa90a98 | 2759 | if (!(si->flags & SWP_USED) || !si->swap_map) |
1da177e4 LT |
2760 | continue; |
2761 | ++*pos; | |
efa90a98 | 2762 | return si; |
1da177e4 LT |
2763 | } |
2764 | ||
2765 | return NULL; | |
2766 | } | |
2767 | ||
2768 | static void swap_stop(struct seq_file *swap, void *v) | |
2769 | { | |
fc0abb14 | 2770 | mutex_unlock(&swapon_mutex); |
1da177e4 LT |
2771 | } |
2772 | ||
2773 | static int swap_show(struct seq_file *swap, void *v) | |
2774 | { | |
efa90a98 | 2775 | struct swap_info_struct *si = v; |
1da177e4 LT |
2776 | struct file *file; |
2777 | int len; | |
2778 | ||
efa90a98 | 2779 | if (si == SEQ_START_TOKEN) { |
881e4aab SS |
2780 | seq_puts(swap,"Filename\t\t\t\tType\t\tSize\tUsed\tPriority\n"); |
2781 | return 0; | |
2782 | } | |
1da177e4 | 2783 | |
efa90a98 | 2784 | file = si->swap_file; |
2726d566 | 2785 | len = seq_file_path(swap, file, " \t\n\\"); |
6eb396dc | 2786 | seq_printf(swap, "%*s%s\t%u\t%u\t%d\n", |
886bb7e9 | 2787 | len < 40 ? 40 - len : 1, " ", |
496ad9aa | 2788 | S_ISBLK(file_inode(file)->i_mode) ? |
1da177e4 | 2789 | "partition" : "file\t", |
efa90a98 HD |
2790 | si->pages << (PAGE_SHIFT - 10), |
2791 | si->inuse_pages << (PAGE_SHIFT - 10), | |
2792 | si->prio); | |
1da177e4 LT |
2793 | return 0; |
2794 | } | |
2795 | ||
15ad7cdc | 2796 | static const struct seq_operations swaps_op = { |
1da177e4 LT |
2797 | .start = swap_start, |
2798 | .next = swap_next, | |
2799 | .stop = swap_stop, | |
2800 | .show = swap_show | |
2801 | }; | |
2802 | ||
2803 | static int swaps_open(struct inode *inode, struct file *file) | |
2804 | { | |
f1514638 | 2805 | struct seq_file *seq; |
66d7dd51 KS |
2806 | int ret; |
2807 | ||
66d7dd51 | 2808 | ret = seq_open(file, &swaps_op); |
f1514638 | 2809 | if (ret) |
66d7dd51 | 2810 | return ret; |
66d7dd51 | 2811 | |
f1514638 KS |
2812 | seq = file->private_data; |
2813 | seq->poll_event = atomic_read(&proc_poll_event); | |
2814 | return 0; | |
1da177e4 LT |
2815 | } |
2816 | ||
15ad7cdc | 2817 | static const struct file_operations proc_swaps_operations = { |
1da177e4 LT |
2818 | .open = swaps_open, |
2819 | .read = seq_read, | |
2820 | .llseek = seq_lseek, | |
2821 | .release = seq_release, | |
66d7dd51 | 2822 | .poll = swaps_poll, |
1da177e4 LT |
2823 | }; |
2824 | ||
2825 | static int __init procswaps_init(void) | |
2826 | { | |
3d71f86f | 2827 | proc_create("swaps", 0, NULL, &proc_swaps_operations); |
1da177e4 LT |
2828 | return 0; |
2829 | } | |
2830 | __initcall(procswaps_init); | |
2831 | #endif /* CONFIG_PROC_FS */ | |
2832 | ||
1796316a JB |
2833 | #ifdef MAX_SWAPFILES_CHECK |
2834 | static int __init max_swapfiles_check(void) | |
2835 | { | |
2836 | MAX_SWAPFILES_CHECK(); | |
2837 | return 0; | |
2838 | } | |
2839 | late_initcall(max_swapfiles_check); | |
2840 | #endif | |
2841 | ||
53cbb243 | 2842 | static struct swap_info_struct *alloc_swap_info(void) |
1da177e4 | 2843 | { |
73c34b6a | 2844 | struct swap_info_struct *p; |
1da177e4 | 2845 | unsigned int type; |
a2468cc9 | 2846 | int i; |
92ee8bd3 | 2847 | int size = sizeof(*p) + nr_node_ids * sizeof(struct plist_node); |
efa90a98 | 2848 | |
92ee8bd3 | 2849 | p = kvzalloc(size, GFP_KERNEL); |
efa90a98 | 2850 | if (!p) |
53cbb243 | 2851 | return ERR_PTR(-ENOMEM); |
efa90a98 | 2852 | |
5d337b91 | 2853 | spin_lock(&swap_lock); |
efa90a98 HD |
2854 | for (type = 0; type < nr_swapfiles; type++) { |
2855 | if (!(swap_info[type]->flags & SWP_USED)) | |
1da177e4 | 2856 | break; |
efa90a98 | 2857 | } |
0697212a | 2858 | if (type >= MAX_SWAPFILES) { |
5d337b91 | 2859 | spin_unlock(&swap_lock); |
040e1b9f | 2860 | kvfree(p); |
730c0581 | 2861 | return ERR_PTR(-EPERM); |
1da177e4 | 2862 | } |
efa90a98 HD |
2863 | if (type >= nr_swapfiles) { |
2864 | p->type = type; | |
27a34b2d | 2865 | WRITE_ONCE(swap_info[type], p); |
efa90a98 HD |
2866 | /* |
2867 | * Write swap_info[type] before nr_swapfiles, in case a | |
2868 | * racing procfs swap_start() or swap_next() is reading them. | |
2869 | * (We never shrink nr_swapfiles, we never free this entry.) | |
2870 | */ | |
2871 | smp_wmb(); | |
27a34b2d | 2872 | WRITE_ONCE(nr_swapfiles, nr_swapfiles + 1); |
efa90a98 | 2873 | } else { |
040e1b9f | 2874 | kvfree(p); |
efa90a98 HD |
2875 | p = swap_info[type]; |
2876 | /* | |
2877 | * Do not memset this entry: a racing procfs swap_next() | |
2878 | * would be relying on p->type to remain valid. | |
2879 | */ | |
2880 | } | |
9625a5f2 | 2881 | INIT_LIST_HEAD(&p->first_swap_extent.list); |
18ab4d4c | 2882 | plist_node_init(&p->list, 0); |
a2468cc9 AL |
2883 | for_each_node(i) |
2884 | plist_node_init(&p->avail_lists[i], 0); | |
1da177e4 | 2885 | p->flags = SWP_USED; |
5d337b91 | 2886 | spin_unlock(&swap_lock); |
ec8acf20 | 2887 | spin_lock_init(&p->lock); |
2628bd6f | 2888 | spin_lock_init(&p->cont_lock); |
efa90a98 | 2889 | |
53cbb243 | 2890 | return p; |
53cbb243 CEB |
2891 | } |
2892 | ||
4d0e1e10 CEB |
2893 | static int claim_swapfile(struct swap_info_struct *p, struct inode *inode) |
2894 | { | |
2895 | int error; | |
2896 | ||
2897 | if (S_ISBLK(inode->i_mode)) { | |
2898 | p->bdev = bdgrab(I_BDEV(inode)); | |
2899 | error = blkdev_get(p->bdev, | |
6f179af8 | 2900 | FMODE_READ | FMODE_WRITE | FMODE_EXCL, p); |
4d0e1e10 CEB |
2901 | if (error < 0) { |
2902 | p->bdev = NULL; | |
6f179af8 | 2903 | return error; |
4d0e1e10 CEB |
2904 | } |
2905 | p->old_block_size = block_size(p->bdev); | |
2906 | error = set_blocksize(p->bdev, PAGE_SIZE); | |
2907 | if (error < 0) | |
87ade72a | 2908 | return error; |
4d0e1e10 CEB |
2909 | p->flags |= SWP_BLKDEV; |
2910 | } else if (S_ISREG(inode->i_mode)) { | |
2911 | p->bdev = inode->i_sb->s_bdev; | |
5955102c | 2912 | inode_lock(inode); |
87ade72a CEB |
2913 | if (IS_SWAPFILE(inode)) |
2914 | return -EBUSY; | |
2915 | } else | |
2916 | return -EINVAL; | |
4d0e1e10 CEB |
2917 | |
2918 | return 0; | |
4d0e1e10 CEB |
2919 | } |
2920 | ||
a98c27e1 AK |
2921 | |
2922 | /* | |
2923 | * Find out how many pages are allowed for a single swap device. There | |
2924 | * are two limiting factors: | |
2925 | * 1) the number of bits for the swap offset in the swp_entry_t type, and | |
2926 | * 2) the number of bits in the swap pte, as defined by the different | |
2927 | * architectures. | |
2928 | * | |
2929 | * In order to find the largest possible bit mask, a swap entry with | |
2930 | * swap type 0 and swap offset ~0UL is created, encoded to a swap pte, | |
2931 | * decoded to a swp_entry_t again, and finally the swap offset is | |
2932 | * extracted. | |
2933 | * | |
2934 | * This will mask all the bits from the initial ~0UL mask that can't | |
2935 | * be encoded in either the swp_entry_t or the architecture definition | |
2936 | * of a swap pte. | |
2937 | */ | |
2938 | unsigned long generic_max_swapfile_size(void) | |
2939 | { | |
2940 | return swp_offset(pte_to_swp_entry( | |
2941 | swp_entry_to_pte(swp_entry(0, ~0UL)))) + 1; | |
2942 | } | |
2943 | ||
2944 | /* Can be overridden by an architecture for additional checks. */ | |
2945 | __weak unsigned long max_swapfile_size(void) | |
2946 | { | |
2947 | return generic_max_swapfile_size(); | |
2948 | } | |
2949 | ||
ca8bd38b CEB |
2950 | static unsigned long read_swap_header(struct swap_info_struct *p, |
2951 | union swap_header *swap_header, | |
2952 | struct inode *inode) | |
2953 | { | |
2954 | int i; | |
2955 | unsigned long maxpages; | |
2956 | unsigned long swapfilepages; | |
d6bbbd29 | 2957 | unsigned long last_page; |
ca8bd38b CEB |
2958 | |
2959 | if (memcmp("SWAPSPACE2", swap_header->magic.magic, 10)) { | |
465c47fd | 2960 | pr_err("Unable to find swap-space signature\n"); |
38719025 | 2961 | return 0; |
ca8bd38b CEB |
2962 | } |
2963 | ||
2964 | /* swap partition endianess hack... */ | |
2965 | if (swab32(swap_header->info.version) == 1) { | |
2966 | swab32s(&swap_header->info.version); | |
2967 | swab32s(&swap_header->info.last_page); | |
2968 | swab32s(&swap_header->info.nr_badpages); | |
dd111be6 JH |
2969 | if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES) |
2970 | return 0; | |
ca8bd38b CEB |
2971 | for (i = 0; i < swap_header->info.nr_badpages; i++) |
2972 | swab32s(&swap_header->info.badpages[i]); | |
2973 | } | |
2974 | /* Check the swap header's sub-version */ | |
2975 | if (swap_header->info.version != 1) { | |
465c47fd AM |
2976 | pr_warn("Unable to handle swap header version %d\n", |
2977 | swap_header->info.version); | |
38719025 | 2978 | return 0; |
ca8bd38b CEB |
2979 | } |
2980 | ||
2981 | p->lowest_bit = 1; | |
2982 | p->cluster_next = 1; | |
2983 | p->cluster_nr = 0; | |
2984 | ||
a98c27e1 | 2985 | maxpages = max_swapfile_size(); |
d6bbbd29 | 2986 | last_page = swap_header->info.last_page; |
e4efe5ef TA |
2987 | if (!last_page) { |
2988 | pr_warn("Empty swap-file\n"); | |
2989 | return 0; | |
2990 | } | |
d6bbbd29 | 2991 | if (last_page > maxpages) { |
465c47fd | 2992 | pr_warn("Truncating oversized swap area, only using %luk out of %luk\n", |
d6bbbd29 RJ |
2993 | maxpages << (PAGE_SHIFT - 10), |
2994 | last_page << (PAGE_SHIFT - 10)); | |
2995 | } | |
2996 | if (maxpages > last_page) { | |
2997 | maxpages = last_page + 1; | |
ca8bd38b CEB |
2998 | /* p->max is an unsigned int: don't overflow it */ |
2999 | if ((unsigned int)maxpages == 0) | |
3000 | maxpages = UINT_MAX; | |
3001 | } | |
3002 | p->highest_bit = maxpages - 1; | |
3003 | ||
3004 | if (!maxpages) | |
38719025 | 3005 | return 0; |
ca8bd38b CEB |
3006 | swapfilepages = i_size_read(inode) >> PAGE_SHIFT; |
3007 | if (swapfilepages && maxpages > swapfilepages) { | |
465c47fd | 3008 | pr_warn("Swap area shorter than signature indicates\n"); |
38719025 | 3009 | return 0; |
ca8bd38b CEB |
3010 | } |
3011 | if (swap_header->info.nr_badpages && S_ISREG(inode->i_mode)) | |
38719025 | 3012 | return 0; |
ca8bd38b | 3013 | if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES) |
38719025 | 3014 | return 0; |
ca8bd38b CEB |
3015 | |
3016 | return maxpages; | |
ca8bd38b CEB |
3017 | } |
3018 | ||
4b3ef9da | 3019 | #define SWAP_CLUSTER_INFO_COLS \ |
235b6217 | 3020 | DIV_ROUND_UP(L1_CACHE_BYTES, sizeof(struct swap_cluster_info)) |
4b3ef9da HY |
3021 | #define SWAP_CLUSTER_SPACE_COLS \ |
3022 | DIV_ROUND_UP(SWAP_ADDRESS_SPACE_PAGES, SWAPFILE_CLUSTER) | |
3023 | #define SWAP_CLUSTER_COLS \ | |
3024 | max_t(unsigned int, SWAP_CLUSTER_INFO_COLS, SWAP_CLUSTER_SPACE_COLS) | |
235b6217 | 3025 | |
915d4d7b CEB |
3026 | static int setup_swap_map_and_extents(struct swap_info_struct *p, |
3027 | union swap_header *swap_header, | |
3028 | unsigned char *swap_map, | |
2a8f9449 | 3029 | struct swap_cluster_info *cluster_info, |
915d4d7b CEB |
3030 | unsigned long maxpages, |
3031 | sector_t *span) | |
3032 | { | |
235b6217 | 3033 | unsigned int j, k; |
915d4d7b CEB |
3034 | unsigned int nr_good_pages; |
3035 | int nr_extents; | |
2a8f9449 | 3036 | unsigned long nr_clusters = DIV_ROUND_UP(maxpages, SWAPFILE_CLUSTER); |
235b6217 HY |
3037 | unsigned long col = p->cluster_next / SWAPFILE_CLUSTER % SWAP_CLUSTER_COLS; |
3038 | unsigned long i, idx; | |
915d4d7b CEB |
3039 | |
3040 | nr_good_pages = maxpages - 1; /* omit header page */ | |
3041 | ||
6b534915 HY |
3042 | cluster_list_init(&p->free_clusters); |
3043 | cluster_list_init(&p->discard_clusters); | |
2a8f9449 | 3044 | |
915d4d7b CEB |
3045 | for (i = 0; i < swap_header->info.nr_badpages; i++) { |
3046 | unsigned int page_nr = swap_header->info.badpages[i]; | |
bdb8e3f6 CEB |
3047 | if (page_nr == 0 || page_nr > swap_header->info.last_page) |
3048 | return -EINVAL; | |
915d4d7b CEB |
3049 | if (page_nr < maxpages) { |
3050 | swap_map[page_nr] = SWAP_MAP_BAD; | |
3051 | nr_good_pages--; | |
2a8f9449 SL |
3052 | /* |
3053 | * Haven't marked the cluster free yet, no list | |
3054 | * operation involved | |
3055 | */ | |
3056 | inc_cluster_info_page(p, cluster_info, page_nr); | |
915d4d7b CEB |
3057 | } |
3058 | } | |
3059 | ||
2a8f9449 SL |
3060 | /* Haven't marked the cluster free yet, no list operation involved */ |
3061 | for (i = maxpages; i < round_up(maxpages, SWAPFILE_CLUSTER); i++) | |
3062 | inc_cluster_info_page(p, cluster_info, i); | |
3063 | ||
915d4d7b CEB |
3064 | if (nr_good_pages) { |
3065 | swap_map[0] = SWAP_MAP_BAD; | |
2a8f9449 SL |
3066 | /* |
3067 | * Not mark the cluster free yet, no list | |
3068 | * operation involved | |
3069 | */ | |
3070 | inc_cluster_info_page(p, cluster_info, 0); | |
915d4d7b CEB |
3071 | p->max = maxpages; |
3072 | p->pages = nr_good_pages; | |
3073 | nr_extents = setup_swap_extents(p, span); | |
bdb8e3f6 CEB |
3074 | if (nr_extents < 0) |
3075 | return nr_extents; | |
915d4d7b CEB |
3076 | nr_good_pages = p->pages; |
3077 | } | |
3078 | if (!nr_good_pages) { | |
465c47fd | 3079 | pr_warn("Empty swap-file\n"); |
bdb8e3f6 | 3080 | return -EINVAL; |
915d4d7b CEB |
3081 | } |
3082 | ||
2a8f9449 SL |
3083 | if (!cluster_info) |
3084 | return nr_extents; | |
3085 | ||
235b6217 | 3086 | |
4b3ef9da HY |
3087 | /* |
3088 | * Reduce false cache line sharing between cluster_info and | |
3089 | * sharing same address space. | |
3090 | */ | |
235b6217 HY |
3091 | for (k = 0; k < SWAP_CLUSTER_COLS; k++) { |
3092 | j = (k + col) % SWAP_CLUSTER_COLS; | |
3093 | for (i = 0; i < DIV_ROUND_UP(nr_clusters, SWAP_CLUSTER_COLS); i++) { | |
3094 | idx = i * SWAP_CLUSTER_COLS + j; | |
3095 | if (idx >= nr_clusters) | |
3096 | continue; | |
3097 | if (cluster_count(&cluster_info[idx])) | |
3098 | continue; | |
2a8f9449 | 3099 | cluster_set_flag(&cluster_info[idx], CLUSTER_FLAG_FREE); |
6b534915 HY |
3100 | cluster_list_add_tail(&p->free_clusters, cluster_info, |
3101 | idx); | |
2a8f9449 | 3102 | } |
2a8f9449 | 3103 | } |
915d4d7b | 3104 | return nr_extents; |
915d4d7b CEB |
3105 | } |
3106 | ||
dcf6b7dd RA |
3107 | /* |
3108 | * Helper to sys_swapon determining if a given swap | |
3109 | * backing device queue supports DISCARD operations. | |
3110 | */ | |
3111 | static bool swap_discardable(struct swap_info_struct *si) | |
3112 | { | |
3113 | struct request_queue *q = bdev_get_queue(si->bdev); | |
3114 | ||
3115 | if (!q || !blk_queue_discard(q)) | |
3116 | return false; | |
3117 | ||
3118 | return true; | |
3119 | } | |
3120 | ||
53cbb243 CEB |
3121 | SYSCALL_DEFINE2(swapon, const char __user *, specialfile, int, swap_flags) |
3122 | { | |
3123 | struct swap_info_struct *p; | |
91a27b2a | 3124 | struct filename *name; |
53cbb243 CEB |
3125 | struct file *swap_file = NULL; |
3126 | struct address_space *mapping; | |
40531542 | 3127 | int prio; |
53cbb243 CEB |
3128 | int error; |
3129 | union swap_header *swap_header; | |
915d4d7b | 3130 | int nr_extents; |
53cbb243 CEB |
3131 | sector_t span; |
3132 | unsigned long maxpages; | |
53cbb243 | 3133 | unsigned char *swap_map = NULL; |
2a8f9449 | 3134 | struct swap_cluster_info *cluster_info = NULL; |
38b5faf4 | 3135 | unsigned long *frontswap_map = NULL; |
53cbb243 CEB |
3136 | struct page *page = NULL; |
3137 | struct inode *inode = NULL; | |
53cbb243 | 3138 | |
d15cab97 HD |
3139 | if (swap_flags & ~SWAP_FLAGS_VALID) |
3140 | return -EINVAL; | |
3141 | ||
53cbb243 CEB |
3142 | if (!capable(CAP_SYS_ADMIN)) |
3143 | return -EPERM; | |
3144 | ||
a2468cc9 AL |
3145 | if (!swap_avail_heads) |
3146 | return -ENOMEM; | |
3147 | ||
53cbb243 | 3148 | p = alloc_swap_info(); |
2542e513 CEB |
3149 | if (IS_ERR(p)) |
3150 | return PTR_ERR(p); | |
53cbb243 | 3151 | |
815c2c54 SL |
3152 | INIT_WORK(&p->discard_work, swap_discard_work); |
3153 | ||
1da177e4 | 3154 | name = getname(specialfile); |
1da177e4 | 3155 | if (IS_ERR(name)) { |
7de7fb6b | 3156 | error = PTR_ERR(name); |
1da177e4 | 3157 | name = NULL; |
bd69010b | 3158 | goto bad_swap; |
1da177e4 | 3159 | } |
669abf4e | 3160 | swap_file = file_open_name(name, O_RDWR|O_LARGEFILE, 0); |
1da177e4 | 3161 | if (IS_ERR(swap_file)) { |
7de7fb6b | 3162 | error = PTR_ERR(swap_file); |
1da177e4 | 3163 | swap_file = NULL; |
bd69010b | 3164 | goto bad_swap; |
1da177e4 LT |
3165 | } |
3166 | ||
3167 | p->swap_file = swap_file; | |
3168 | mapping = swap_file->f_mapping; | |
2130781e | 3169 | inode = mapping->host; |
6f179af8 | 3170 | |
5955102c | 3171 | /* If S_ISREG(inode->i_mode) will do inode_lock(inode); */ |
4d0e1e10 CEB |
3172 | error = claim_swapfile(p, inode); |
3173 | if (unlikely(error)) | |
1da177e4 | 3174 | goto bad_swap; |
1da177e4 | 3175 | |
1da177e4 LT |
3176 | /* |
3177 | * Read the swap header. | |
3178 | */ | |
3179 | if (!mapping->a_ops->readpage) { | |
3180 | error = -EINVAL; | |
3181 | goto bad_swap; | |
3182 | } | |
090d2b18 | 3183 | page = read_mapping_page(mapping, 0, swap_file); |
1da177e4 LT |
3184 | if (IS_ERR(page)) { |
3185 | error = PTR_ERR(page); | |
3186 | goto bad_swap; | |
3187 | } | |
81e33971 | 3188 | swap_header = kmap(page); |
1da177e4 | 3189 | |
ca8bd38b CEB |
3190 | maxpages = read_swap_header(p, swap_header, inode); |
3191 | if (unlikely(!maxpages)) { | |
1da177e4 LT |
3192 | error = -EINVAL; |
3193 | goto bad_swap; | |
3194 | } | |
886bb7e9 | 3195 | |
81e33971 | 3196 | /* OK, set up the swap map and apply the bad block list */ |
803d0c83 | 3197 | swap_map = vzalloc(maxpages); |
81e33971 HD |
3198 | if (!swap_map) { |
3199 | error = -ENOMEM; | |
3200 | goto bad_swap; | |
3201 | } | |
f0571429 MK |
3202 | |
3203 | if (bdi_cap_stable_pages_required(inode_to_bdi(inode))) | |
3204 | p->flags |= SWP_STABLE_WRITES; | |
3205 | ||
539a6fea MK |
3206 | if (bdi_cap_synchronous_io(inode_to_bdi(inode))) |
3207 | p->flags |= SWP_SYNCHRONOUS_IO; | |
3208 | ||
2a8f9449 | 3209 | if (p->bdev && blk_queue_nonrot(bdev_get_queue(p->bdev))) { |
6f179af8 | 3210 | int cpu; |
235b6217 | 3211 | unsigned long ci, nr_cluster; |
6f179af8 | 3212 | |
2a8f9449 SL |
3213 | p->flags |= SWP_SOLIDSTATE; |
3214 | /* | |
3215 | * select a random position to start with to help wear leveling | |
3216 | * SSD | |
3217 | */ | |
3218 | p->cluster_next = 1 + (prandom_u32() % p->highest_bit); | |
235b6217 | 3219 | nr_cluster = DIV_ROUND_UP(maxpages, SWAPFILE_CLUSTER); |
2a8f9449 | 3220 | |
54f180d3 HY |
3221 | cluster_info = kvzalloc(nr_cluster * sizeof(*cluster_info), |
3222 | GFP_KERNEL); | |
2a8f9449 SL |
3223 | if (!cluster_info) { |
3224 | error = -ENOMEM; | |
3225 | goto bad_swap; | |
3226 | } | |
235b6217 HY |
3227 | |
3228 | for (ci = 0; ci < nr_cluster; ci++) | |
3229 | spin_lock_init(&((cluster_info + ci)->lock)); | |
3230 | ||
ebc2a1a6 SL |
3231 | p->percpu_cluster = alloc_percpu(struct percpu_cluster); |
3232 | if (!p->percpu_cluster) { | |
3233 | error = -ENOMEM; | |
3234 | goto bad_swap; | |
3235 | } | |
6f179af8 | 3236 | for_each_possible_cpu(cpu) { |
ebc2a1a6 | 3237 | struct percpu_cluster *cluster; |
6f179af8 | 3238 | cluster = per_cpu_ptr(p->percpu_cluster, cpu); |
ebc2a1a6 SL |
3239 | cluster_set_null(&cluster->index); |
3240 | } | |
81a0298b HY |
3241 | } else |
3242 | atomic_inc(&nr_rotate_swap); | |
1da177e4 | 3243 | |
1421ef3c CEB |
3244 | error = swap_cgroup_swapon(p->type, maxpages); |
3245 | if (error) | |
3246 | goto bad_swap; | |
3247 | ||
915d4d7b | 3248 | nr_extents = setup_swap_map_and_extents(p, swap_header, swap_map, |
2a8f9449 | 3249 | cluster_info, maxpages, &span); |
915d4d7b CEB |
3250 | if (unlikely(nr_extents < 0)) { |
3251 | error = nr_extents; | |
1da177e4 LT |
3252 | goto bad_swap; |
3253 | } | |
38b5faf4 | 3254 | /* frontswap enabled? set up bit-per-page map for frontswap */ |
8ea1d2a1 | 3255 | if (IS_ENABLED(CONFIG_FRONTSWAP)) |
54f180d3 HY |
3256 | frontswap_map = kvzalloc(BITS_TO_LONGS(maxpages) * sizeof(long), |
3257 | GFP_KERNEL); | |
1da177e4 | 3258 | |
2a8f9449 SL |
3259 | if (p->bdev &&(swap_flags & SWAP_FLAG_DISCARD) && swap_discardable(p)) { |
3260 | /* | |
3261 | * When discard is enabled for swap with no particular | |
3262 | * policy flagged, we set all swap discard flags here in | |
3263 | * order to sustain backward compatibility with older | |
3264 | * swapon(8) releases. | |
3265 | */ | |
3266 | p->flags |= (SWP_DISCARDABLE | SWP_AREA_DISCARD | | |
3267 | SWP_PAGE_DISCARD); | |
dcf6b7dd | 3268 | |
2a8f9449 SL |
3269 | /* |
3270 | * By flagging sys_swapon, a sysadmin can tell us to | |
3271 | * either do single-time area discards only, or to just | |
3272 | * perform discards for released swap page-clusters. | |
3273 | * Now it's time to adjust the p->flags accordingly. | |
3274 | */ | |
3275 | if (swap_flags & SWAP_FLAG_DISCARD_ONCE) | |
3276 | p->flags &= ~SWP_PAGE_DISCARD; | |
3277 | else if (swap_flags & SWAP_FLAG_DISCARD_PAGES) | |
3278 | p->flags &= ~SWP_AREA_DISCARD; | |
3279 | ||
3280 | /* issue a swapon-time discard if it's still required */ | |
3281 | if (p->flags & SWP_AREA_DISCARD) { | |
3282 | int err = discard_swap(p); | |
3283 | if (unlikely(err)) | |
3284 | pr_err("swapon: discard_swap(%p): %d\n", | |
3285 | p, err); | |
dcf6b7dd | 3286 | } |
20137a49 | 3287 | } |
6a6ba831 | 3288 | |
4b3ef9da HY |
3289 | error = init_swap_address_space(p->type, maxpages); |
3290 | if (error) | |
3291 | goto bad_swap; | |
3292 | ||
fc0abb14 | 3293 | mutex_lock(&swapon_mutex); |
40531542 | 3294 | prio = -1; |
78ecba08 | 3295 | if (swap_flags & SWAP_FLAG_PREFER) |
40531542 | 3296 | prio = |
78ecba08 | 3297 | (swap_flags & SWAP_FLAG_PRIO_MASK) >> SWAP_FLAG_PRIO_SHIFT; |
2a8f9449 | 3298 | enable_swap_info(p, prio, swap_map, cluster_info, frontswap_map); |
c69dbfb8 | 3299 | |
756a025f | 3300 | pr_info("Adding %uk swap on %s. Priority:%d extents:%d across:%lluk %s%s%s%s%s\n", |
91a27b2a | 3301 | p->pages<<(PAGE_SHIFT-10), name->name, p->prio, |
c69dbfb8 CEB |
3302 | nr_extents, (unsigned long long)span<<(PAGE_SHIFT-10), |
3303 | (p->flags & SWP_SOLIDSTATE) ? "SS" : "", | |
38b5faf4 | 3304 | (p->flags & SWP_DISCARDABLE) ? "D" : "", |
dcf6b7dd RA |
3305 | (p->flags & SWP_AREA_DISCARD) ? "s" : "", |
3306 | (p->flags & SWP_PAGE_DISCARD) ? "c" : "", | |
38b5faf4 | 3307 | (frontswap_map) ? "FS" : ""); |
c69dbfb8 | 3308 | |
fc0abb14 | 3309 | mutex_unlock(&swapon_mutex); |
66d7dd51 KS |
3310 | atomic_inc(&proc_poll_event); |
3311 | wake_up_interruptible(&proc_poll_wait); | |
3312 | ||
9b01c350 CEB |
3313 | if (S_ISREG(inode->i_mode)) |
3314 | inode->i_flags |= S_SWAPFILE; | |
1da177e4 LT |
3315 | error = 0; |
3316 | goto out; | |
3317 | bad_swap: | |
ebc2a1a6 SL |
3318 | free_percpu(p->percpu_cluster); |
3319 | p->percpu_cluster = NULL; | |
bd69010b | 3320 | if (inode && S_ISBLK(inode->i_mode) && p->bdev) { |
f2090d2d CEB |
3321 | set_blocksize(p->bdev, p->old_block_size); |
3322 | blkdev_put(p->bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL); | |
1da177e4 | 3323 | } |
4cd3bb10 | 3324 | destroy_swap_extents(p); |
e8e6c2ec | 3325 | swap_cgroup_swapoff(p->type); |
5d337b91 | 3326 | spin_lock(&swap_lock); |
1da177e4 | 3327 | p->swap_file = NULL; |
1da177e4 | 3328 | p->flags = 0; |
5d337b91 | 3329 | spin_unlock(&swap_lock); |
1da177e4 | 3330 | vfree(swap_map); |
8606a1a9 | 3331 | kvfree(cluster_info); |
b6b1fd2a | 3332 | kvfree(frontswap_map); |
52c50567 | 3333 | if (swap_file) { |
2130781e | 3334 | if (inode && S_ISREG(inode->i_mode)) { |
5955102c | 3335 | inode_unlock(inode); |
2130781e CEB |
3336 | inode = NULL; |
3337 | } | |
1da177e4 | 3338 | filp_close(swap_file, NULL); |
52c50567 | 3339 | } |
1da177e4 LT |
3340 | out: |
3341 | if (page && !IS_ERR(page)) { | |
3342 | kunmap(page); | |
09cbfeaf | 3343 | put_page(page); |
1da177e4 LT |
3344 | } |
3345 | if (name) | |
3346 | putname(name); | |
9b01c350 | 3347 | if (inode && S_ISREG(inode->i_mode)) |
5955102c | 3348 | inode_unlock(inode); |
039939a6 TC |
3349 | if (!error) |
3350 | enable_swap_slots_cache(); | |
1da177e4 LT |
3351 | return error; |
3352 | } | |
3353 | ||
3354 | void si_swapinfo(struct sysinfo *val) | |
3355 | { | |
efa90a98 | 3356 | unsigned int type; |
1da177e4 LT |
3357 | unsigned long nr_to_be_unused = 0; |
3358 | ||
5d337b91 | 3359 | spin_lock(&swap_lock); |
efa90a98 HD |
3360 | for (type = 0; type < nr_swapfiles; type++) { |
3361 | struct swap_info_struct *si = swap_info[type]; | |
3362 | ||
3363 | if ((si->flags & SWP_USED) && !(si->flags & SWP_WRITEOK)) | |
3364 | nr_to_be_unused += si->inuse_pages; | |
1da177e4 | 3365 | } |
ec8acf20 | 3366 | val->freeswap = atomic_long_read(&nr_swap_pages) + nr_to_be_unused; |
1da177e4 | 3367 | val->totalswap = total_swap_pages + nr_to_be_unused; |
5d337b91 | 3368 | spin_unlock(&swap_lock); |
1da177e4 LT |
3369 | } |
3370 | ||
3371 | /* | |
3372 | * Verify that a swap entry is valid and increment its swap map count. | |
3373 | * | |
355cfa73 KH |
3374 | * Returns error code in following case. |
3375 | * - success -> 0 | |
3376 | * - swp_entry is invalid -> EINVAL | |
3377 | * - swp_entry is migration entry -> EINVAL | |
3378 | * - swap-cache reference is requested but there is already one. -> EEXIST | |
3379 | * - swap-cache reference is requested but the entry is not used. -> ENOENT | |
570a335b | 3380 | * - swap-mapped reference requested but needs continued swap count. -> ENOMEM |
1da177e4 | 3381 | */ |
8d69aaee | 3382 | static int __swap_duplicate(swp_entry_t entry, unsigned char usage) |
1da177e4 | 3383 | { |
73c34b6a | 3384 | struct swap_info_struct *p; |
235b6217 | 3385 | struct swap_cluster_info *ci; |
27a34b2d | 3386 | unsigned long offset; |
8d69aaee HD |
3387 | unsigned char count; |
3388 | unsigned char has_cache; | |
253d553b | 3389 | int err = -EINVAL; |
1da177e4 | 3390 | |
a7420aa5 | 3391 | if (non_swap_entry(entry)) |
253d553b | 3392 | goto out; |
0697212a | 3393 | |
27a34b2d DJ |
3394 | p = swp_swap_info(entry); |
3395 | if (!p) | |
1da177e4 | 3396 | goto bad_file; |
27a34b2d | 3397 | |
1da177e4 | 3398 | offset = swp_offset(entry); |
355cfa73 | 3399 | if (unlikely(offset >= p->max)) |
235b6217 HY |
3400 | goto out; |
3401 | ||
3402 | ci = lock_cluster_or_swap_info(p, offset); | |
355cfa73 | 3403 | |
253d553b | 3404 | count = p->swap_map[offset]; |
edfe23da SL |
3405 | |
3406 | /* | |
3407 | * swapin_readahead() doesn't check if a swap entry is valid, so the | |
3408 | * swap entry could be SWAP_MAP_BAD. Check here with lock held. | |
3409 | */ | |
3410 | if (unlikely(swap_count(count) == SWAP_MAP_BAD)) { | |
3411 | err = -ENOENT; | |
3412 | goto unlock_out; | |
3413 | } | |
3414 | ||
253d553b HD |
3415 | has_cache = count & SWAP_HAS_CACHE; |
3416 | count &= ~SWAP_HAS_CACHE; | |
3417 | err = 0; | |
355cfa73 | 3418 | |
253d553b | 3419 | if (usage == SWAP_HAS_CACHE) { |
355cfa73 KH |
3420 | |
3421 | /* set SWAP_HAS_CACHE if there is no cache and entry is used */ | |
253d553b HD |
3422 | if (!has_cache && count) |
3423 | has_cache = SWAP_HAS_CACHE; | |
3424 | else if (has_cache) /* someone else added cache */ | |
3425 | err = -EEXIST; | |
3426 | else /* no users remaining */ | |
3427 | err = -ENOENT; | |
355cfa73 KH |
3428 | |
3429 | } else if (count || has_cache) { | |
253d553b | 3430 | |
570a335b HD |
3431 | if ((count & ~COUNT_CONTINUED) < SWAP_MAP_MAX) |
3432 | count += usage; | |
3433 | else if ((count & ~COUNT_CONTINUED) > SWAP_MAP_MAX) | |
253d553b | 3434 | err = -EINVAL; |
570a335b HD |
3435 | else if (swap_count_continued(p, offset, count)) |
3436 | count = COUNT_CONTINUED; | |
3437 | else | |
3438 | err = -ENOMEM; | |
355cfa73 | 3439 | } else |
253d553b HD |
3440 | err = -ENOENT; /* unused swap entry */ |
3441 | ||
3442 | p->swap_map[offset] = count | has_cache; | |
3443 | ||
355cfa73 | 3444 | unlock_out: |
235b6217 | 3445 | unlock_cluster_or_swap_info(p, ci); |
1da177e4 | 3446 | out: |
253d553b | 3447 | return err; |
1da177e4 LT |
3448 | |
3449 | bad_file: | |
465c47fd | 3450 | pr_err("swap_dup: %s%08lx\n", Bad_file, entry.val); |
1da177e4 LT |
3451 | goto out; |
3452 | } | |
253d553b | 3453 | |
aaa46865 HD |
3454 | /* |
3455 | * Help swapoff by noting that swap entry belongs to shmem/tmpfs | |
3456 | * (in which case its reference count is never incremented). | |
3457 | */ | |
3458 | void swap_shmem_alloc(swp_entry_t entry) | |
3459 | { | |
3460 | __swap_duplicate(entry, SWAP_MAP_SHMEM); | |
3461 | } | |
3462 | ||
355cfa73 | 3463 | /* |
08259d58 HD |
3464 | * Increase reference count of swap entry by 1. |
3465 | * Returns 0 for success, or -ENOMEM if a swap_count_continuation is required | |
3466 | * but could not be atomically allocated. Returns 0, just as if it succeeded, | |
3467 | * if __swap_duplicate() fails for another reason (-EINVAL or -ENOENT), which | |
3468 | * might occur if a page table entry has got corrupted. | |
355cfa73 | 3469 | */ |
570a335b | 3470 | int swap_duplicate(swp_entry_t entry) |
355cfa73 | 3471 | { |
570a335b HD |
3472 | int err = 0; |
3473 | ||
3474 | while (!err && __swap_duplicate(entry, 1) == -ENOMEM) | |
3475 | err = add_swap_count_continuation(entry, GFP_ATOMIC); | |
3476 | return err; | |
355cfa73 | 3477 | } |
1da177e4 | 3478 | |
cb4b86ba | 3479 | /* |
355cfa73 KH |
3480 | * @entry: swap entry for which we allocate swap cache. |
3481 | * | |
73c34b6a | 3482 | * Called when allocating swap cache for existing swap entry, |
355cfa73 KH |
3483 | * This can return error codes. Returns 0 at success. |
3484 | * -EBUSY means there is a swap cache. | |
3485 | * Note: return code is different from swap_duplicate(). | |
cb4b86ba KH |
3486 | */ |
3487 | int swapcache_prepare(swp_entry_t entry) | |
3488 | { | |
253d553b | 3489 | return __swap_duplicate(entry, SWAP_HAS_CACHE); |
cb4b86ba KH |
3490 | } |
3491 | ||
0bcac06f MK |
3492 | struct swap_info_struct *swp_swap_info(swp_entry_t entry) |
3493 | { | |
27a34b2d | 3494 | return swap_type_to_swap_info(swp_type(entry)); |
0bcac06f MK |
3495 | } |
3496 | ||
f981c595 MG |
3497 | struct swap_info_struct *page_swap_info(struct page *page) |
3498 | { | |
0bcac06f MK |
3499 | swp_entry_t entry = { .val = page_private(page) }; |
3500 | return swp_swap_info(entry); | |
f981c595 MG |
3501 | } |
3502 | ||
3503 | /* | |
3504 | * out-of-line __page_file_ methods to avoid include hell. | |
3505 | */ | |
3506 | struct address_space *__page_file_mapping(struct page *page) | |
3507 | { | |
f981c595 MG |
3508 | return page_swap_info(page)->swap_file->f_mapping; |
3509 | } | |
3510 | EXPORT_SYMBOL_GPL(__page_file_mapping); | |
3511 | ||
3512 | pgoff_t __page_file_index(struct page *page) | |
3513 | { | |
3514 | swp_entry_t swap = { .val = page_private(page) }; | |
f981c595 MG |
3515 | return swp_offset(swap); |
3516 | } | |
3517 | EXPORT_SYMBOL_GPL(__page_file_index); | |
3518 | ||
570a335b HD |
3519 | /* |
3520 | * add_swap_count_continuation - called when a swap count is duplicated | |
3521 | * beyond SWAP_MAP_MAX, it allocates a new page and links that to the entry's | |
3522 | * page of the original vmalloc'ed swap_map, to hold the continuation count | |
3523 | * (for that entry and for its neighbouring PAGE_SIZE swap entries). Called | |
3524 | * again when count is duplicated beyond SWAP_MAP_MAX * SWAP_CONT_MAX, etc. | |
3525 | * | |
3526 | * These continuation pages are seldom referenced: the common paths all work | |
3527 | * on the original swap_map, only referring to a continuation page when the | |
3528 | * low "digit" of a count is incremented or decremented through SWAP_MAP_MAX. | |
3529 | * | |
3530 | * add_swap_count_continuation(, GFP_ATOMIC) can be called while holding | |
3531 | * page table locks; if it fails, add_swap_count_continuation(, GFP_KERNEL) | |
3532 | * can be called after dropping locks. | |
3533 | */ | |
3534 | int add_swap_count_continuation(swp_entry_t entry, gfp_t gfp_mask) | |
3535 | { | |
3536 | struct swap_info_struct *si; | |
235b6217 | 3537 | struct swap_cluster_info *ci; |
570a335b HD |
3538 | struct page *head; |
3539 | struct page *page; | |
3540 | struct page *list_page; | |
3541 | pgoff_t offset; | |
3542 | unsigned char count; | |
3543 | ||
3544 | /* | |
3545 | * When debugging, it's easier to use __GFP_ZERO here; but it's better | |
3546 | * for latency not to zero a page while GFP_ATOMIC and holding locks. | |
3547 | */ | |
3548 | page = alloc_page(gfp_mask | __GFP_HIGHMEM); | |
3549 | ||
3550 | si = swap_info_get(entry); | |
3551 | if (!si) { | |
3552 | /* | |
3553 | * An acceptable race has occurred since the failing | |
3554 | * __swap_duplicate(): the swap entry has been freed, | |
3555 | * perhaps even the whole swap_map cleared for swapoff. | |
3556 | */ | |
3557 | goto outer; | |
3558 | } | |
3559 | ||
3560 | offset = swp_offset(entry); | |
235b6217 HY |
3561 | |
3562 | ci = lock_cluster(si, offset); | |
3563 | ||
570a335b HD |
3564 | count = si->swap_map[offset] & ~SWAP_HAS_CACHE; |
3565 | ||
3566 | if ((count & ~COUNT_CONTINUED) != SWAP_MAP_MAX) { | |
3567 | /* | |
3568 | * The higher the swap count, the more likely it is that tasks | |
3569 | * will race to add swap count continuation: we need to avoid | |
3570 | * over-provisioning. | |
3571 | */ | |
3572 | goto out; | |
3573 | } | |
3574 | ||
3575 | if (!page) { | |
235b6217 | 3576 | unlock_cluster(ci); |
ec8acf20 | 3577 | spin_unlock(&si->lock); |
570a335b HD |
3578 | return -ENOMEM; |
3579 | } | |
3580 | ||
3581 | /* | |
3582 | * We are fortunate that although vmalloc_to_page uses pte_offset_map, | |
2de1a7e4 SJ |
3583 | * no architecture is using highmem pages for kernel page tables: so it |
3584 | * will not corrupt the GFP_ATOMIC caller's atomic page table kmaps. | |
570a335b HD |
3585 | */ |
3586 | head = vmalloc_to_page(si->swap_map + offset); | |
3587 | offset &= ~PAGE_MASK; | |
3588 | ||
2628bd6f | 3589 | spin_lock(&si->cont_lock); |
570a335b HD |
3590 | /* |
3591 | * Page allocation does not initialize the page's lru field, | |
3592 | * but it does always reset its private field. | |
3593 | */ | |
3594 | if (!page_private(head)) { | |
3595 | BUG_ON(count & COUNT_CONTINUED); | |
3596 | INIT_LIST_HEAD(&head->lru); | |
3597 | set_page_private(head, SWP_CONTINUED); | |
3598 | si->flags |= SWP_CONTINUED; | |
3599 | } | |
3600 | ||
3601 | list_for_each_entry(list_page, &head->lru, lru) { | |
3602 | unsigned char *map; | |
3603 | ||
3604 | /* | |
3605 | * If the previous map said no continuation, but we've found | |
3606 | * a continuation page, free our allocation and use this one. | |
3607 | */ | |
3608 | if (!(count & COUNT_CONTINUED)) | |
2628bd6f | 3609 | goto out_unlock_cont; |
570a335b | 3610 | |
9b04c5fe | 3611 | map = kmap_atomic(list_page) + offset; |
570a335b | 3612 | count = *map; |
9b04c5fe | 3613 | kunmap_atomic(map); |
570a335b HD |
3614 | |
3615 | /* | |
3616 | * If this continuation count now has some space in it, | |
3617 | * free our allocation and use this one. | |
3618 | */ | |
3619 | if ((count & ~COUNT_CONTINUED) != SWAP_CONT_MAX) | |
2628bd6f | 3620 | goto out_unlock_cont; |
570a335b HD |
3621 | } |
3622 | ||
3623 | list_add_tail(&page->lru, &head->lru); | |
3624 | page = NULL; /* now it's attached, don't free it */ | |
2628bd6f HY |
3625 | out_unlock_cont: |
3626 | spin_unlock(&si->cont_lock); | |
570a335b | 3627 | out: |
235b6217 | 3628 | unlock_cluster(ci); |
ec8acf20 | 3629 | spin_unlock(&si->lock); |
570a335b HD |
3630 | outer: |
3631 | if (page) | |
3632 | __free_page(page); | |
3633 | return 0; | |
3634 | } | |
3635 | ||
3636 | /* | |
3637 | * swap_count_continued - when the original swap_map count is incremented | |
3638 | * from SWAP_MAP_MAX, check if there is already a continuation page to carry | |
3639 | * into, carry if so, or else fail until a new continuation page is allocated; | |
3640 | * when the original swap_map count is decremented from 0 with continuation, | |
3641 | * borrow from the continuation and report whether it still holds more. | |
235b6217 HY |
3642 | * Called while __swap_duplicate() or swap_entry_free() holds swap or cluster |
3643 | * lock. | |
570a335b HD |
3644 | */ |
3645 | static bool swap_count_continued(struct swap_info_struct *si, | |
3646 | pgoff_t offset, unsigned char count) | |
3647 | { | |
3648 | struct page *head; | |
3649 | struct page *page; | |
3650 | unsigned char *map; | |
2628bd6f | 3651 | bool ret; |
570a335b HD |
3652 | |
3653 | head = vmalloc_to_page(si->swap_map + offset); | |
3654 | if (page_private(head) != SWP_CONTINUED) { | |
3655 | BUG_ON(count & COUNT_CONTINUED); | |
3656 | return false; /* need to add count continuation */ | |
3657 | } | |
3658 | ||
2628bd6f | 3659 | spin_lock(&si->cont_lock); |
570a335b HD |
3660 | offset &= ~PAGE_MASK; |
3661 | page = list_entry(head->lru.next, struct page, lru); | |
9b04c5fe | 3662 | map = kmap_atomic(page) + offset; |
570a335b HD |
3663 | |
3664 | if (count == SWAP_MAP_MAX) /* initial increment from swap_map */ | |
3665 | goto init_map; /* jump over SWAP_CONT_MAX checks */ | |
3666 | ||
3667 | if (count == (SWAP_MAP_MAX | COUNT_CONTINUED)) { /* incrementing */ | |
3668 | /* | |
3669 | * Think of how you add 1 to 999 | |
3670 | */ | |
3671 | while (*map == (SWAP_CONT_MAX | COUNT_CONTINUED)) { | |
9b04c5fe | 3672 | kunmap_atomic(map); |
570a335b HD |
3673 | page = list_entry(page->lru.next, struct page, lru); |
3674 | BUG_ON(page == head); | |
9b04c5fe | 3675 | map = kmap_atomic(page) + offset; |
570a335b HD |
3676 | } |
3677 | if (*map == SWAP_CONT_MAX) { | |
9b04c5fe | 3678 | kunmap_atomic(map); |
570a335b | 3679 | page = list_entry(page->lru.next, struct page, lru); |
2628bd6f HY |
3680 | if (page == head) { |
3681 | ret = false; /* add count continuation */ | |
3682 | goto out; | |
3683 | } | |
9b04c5fe | 3684 | map = kmap_atomic(page) + offset; |
570a335b HD |
3685 | init_map: *map = 0; /* we didn't zero the page */ |
3686 | } | |
3687 | *map += 1; | |
9b04c5fe | 3688 | kunmap_atomic(map); |
570a335b HD |
3689 | page = list_entry(page->lru.prev, struct page, lru); |
3690 | while (page != head) { | |
9b04c5fe | 3691 | map = kmap_atomic(page) + offset; |
570a335b | 3692 | *map = COUNT_CONTINUED; |
9b04c5fe | 3693 | kunmap_atomic(map); |
570a335b HD |
3694 | page = list_entry(page->lru.prev, struct page, lru); |
3695 | } | |
2628bd6f | 3696 | ret = true; /* incremented */ |
570a335b HD |
3697 | |
3698 | } else { /* decrementing */ | |
3699 | /* | |
3700 | * Think of how you subtract 1 from 1000 | |
3701 | */ | |
3702 | BUG_ON(count != COUNT_CONTINUED); | |
3703 | while (*map == COUNT_CONTINUED) { | |
9b04c5fe | 3704 | kunmap_atomic(map); |
570a335b HD |
3705 | page = list_entry(page->lru.next, struct page, lru); |
3706 | BUG_ON(page == head); | |
9b04c5fe | 3707 | map = kmap_atomic(page) + offset; |
570a335b HD |
3708 | } |
3709 | BUG_ON(*map == 0); | |
3710 | *map -= 1; | |
3711 | if (*map == 0) | |
3712 | count = 0; | |
9b04c5fe | 3713 | kunmap_atomic(map); |
570a335b HD |
3714 | page = list_entry(page->lru.prev, struct page, lru); |
3715 | while (page != head) { | |
9b04c5fe | 3716 | map = kmap_atomic(page) + offset; |
570a335b HD |
3717 | *map = SWAP_CONT_MAX | count; |
3718 | count = COUNT_CONTINUED; | |
9b04c5fe | 3719 | kunmap_atomic(map); |
570a335b HD |
3720 | page = list_entry(page->lru.prev, struct page, lru); |
3721 | } | |
2628bd6f | 3722 | ret = count == COUNT_CONTINUED; |
570a335b | 3723 | } |
2628bd6f HY |
3724 | out: |
3725 | spin_unlock(&si->cont_lock); | |
3726 | return ret; | |
570a335b HD |
3727 | } |
3728 | ||
3729 | /* | |
3730 | * free_swap_count_continuations - swapoff free all the continuation pages | |
3731 | * appended to the swap_map, after swap_map is quiesced, before vfree'ing it. | |
3732 | */ | |
3733 | static void free_swap_count_continuations(struct swap_info_struct *si) | |
3734 | { | |
3735 | pgoff_t offset; | |
3736 | ||
3737 | for (offset = 0; offset < si->max; offset += PAGE_SIZE) { | |
3738 | struct page *head; | |
3739 | head = vmalloc_to_page(si->swap_map + offset); | |
3740 | if (page_private(head)) { | |
0d576d20 GT |
3741 | struct page *page, *next; |
3742 | ||
3743 | list_for_each_entry_safe(page, next, &head->lru, lru) { | |
3744 | list_del(&page->lru); | |
570a335b HD |
3745 | __free_page(page); |
3746 | } | |
3747 | } | |
3748 | } | |
3749 | } | |
a2468cc9 AL |
3750 | |
3751 | static int __init swapfile_init(void) | |
3752 | { | |
3753 | int nid; | |
3754 | ||
3755 | swap_avail_heads = kmalloc_array(nr_node_ids, sizeof(struct plist_head), | |
3756 | GFP_KERNEL); | |
3757 | if (!swap_avail_heads) { | |
3758 | pr_emerg("Not enough memory for swap heads, swap is disabled\n"); | |
3759 | return -ENOMEM; | |
3760 | } | |
3761 | ||
3762 | for_each_node(nid) | |
3763 | plist_head_init(&swap_avail_heads[nid]); | |
3764 | ||
3765 | return 0; | |
3766 | } | |
3767 | subsys_initcall(swapfile_init); |