Commit | Line | Data |
---|---|---|
457c8996 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
1da177e4 LT |
2 | /* |
3 | * linux/mm/swapfile.c | |
4 | * | |
5 | * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds | |
6 | * Swap reorganised 29.12.95, Stephen Tweedie | |
7 | */ | |
8 | ||
c97ab271 | 9 | #include <linux/blkdev.h> |
1da177e4 | 10 | #include <linux/mm.h> |
6e84f315 | 11 | #include <linux/sched/mm.h> |
29930025 | 12 | #include <linux/sched/task.h> |
1da177e4 LT |
13 | #include <linux/hugetlb.h> |
14 | #include <linux/mman.h> | |
15 | #include <linux/slab.h> | |
16 | #include <linux/kernel_stat.h> | |
17 | #include <linux/swap.h> | |
18 | #include <linux/vmalloc.h> | |
19 | #include <linux/pagemap.h> | |
20 | #include <linux/namei.h> | |
072441e2 | 21 | #include <linux/shmem_fs.h> |
e41d12f5 | 22 | #include <linux/blk-cgroup.h> |
20137a49 | 23 | #include <linux/random.h> |
1da177e4 LT |
24 | #include <linux/writeback.h> |
25 | #include <linux/proc_fs.h> | |
26 | #include <linux/seq_file.h> | |
27 | #include <linux/init.h> | |
5ad64688 | 28 | #include <linux/ksm.h> |
1da177e4 LT |
29 | #include <linux/rmap.h> |
30 | #include <linux/security.h> | |
31 | #include <linux/backing-dev.h> | |
fc0abb14 | 32 | #include <linux/mutex.h> |
c59ede7b | 33 | #include <linux/capability.h> |
1da177e4 | 34 | #include <linux/syscalls.h> |
8a9f3ccd | 35 | #include <linux/memcontrol.h> |
66d7dd51 | 36 | #include <linux/poll.h> |
72788c38 | 37 | #include <linux/oom.h> |
38b5faf4 DM |
38 | #include <linux/frontswap.h> |
39 | #include <linux/swapfile.h> | |
f981c595 | 40 | #include <linux/export.h> |
67afa38e | 41 | #include <linux/swap_slots.h> |
155b5f88 | 42 | #include <linux/sort.h> |
63d8620e | 43 | #include <linux/completion.h> |
07f44ac3 | 44 | #include <linux/suspend.h> |
1da177e4 | 45 | |
1da177e4 LT |
46 | #include <asm/tlbflush.h> |
47 | #include <linux/swapops.h> | |
5d1ea48b | 48 | #include <linux/swap_cgroup.h> |
014bb1de | 49 | #include "swap.h" |
1da177e4 | 50 | |
570a335b HD |
51 | static bool swap_count_continued(struct swap_info_struct *, pgoff_t, |
52 | unsigned char); | |
53 | static void free_swap_count_continuations(struct swap_info_struct *); | |
54 | ||
633423a0 | 55 | static DEFINE_SPINLOCK(swap_lock); |
7c363b8c | 56 | static unsigned int nr_swapfiles; |
ec8acf20 | 57 | atomic_long_t nr_swap_pages; |
fb0fec50 CW |
58 | /* |
59 | * Some modules use swappable objects and may try to swap them out under | |
60 | * memory pressure (via the shrinker). Before doing so, they may wish to | |
61 | * check to see if any swap space is available. | |
62 | */ | |
63 | EXPORT_SYMBOL_GPL(nr_swap_pages); | |
ec8acf20 | 64 | /* protected with swap_lock. reading in vm_swap_full() doesn't need lock */ |
1da177e4 | 65 | long total_swap_pages; |
a2468cc9 | 66 | static int least_priority = -1; |
be45a490 | 67 | unsigned long swapfile_maximum_size; |
5154e607 PX |
68 | #ifdef CONFIG_MIGRATION |
69 | bool swap_migration_ad_supported; | |
70 | #endif /* CONFIG_MIGRATION */ | |
1da177e4 | 71 | |
1da177e4 LT |
72 | static const char Bad_file[] = "Bad swap file entry "; |
73 | static const char Unused_file[] = "Unused swap file entry "; | |
74 | static const char Bad_offset[] = "Bad swap offset entry "; | |
75 | static const char Unused_offset[] = "Unused swap offset entry "; | |
76 | ||
adfab836 DS |
77 | /* |
78 | * all active swap_info_structs | |
79 | * protected with swap_lock, and ordered by priority. | |
80 | */ | |
633423a0 | 81 | static PLIST_HEAD(swap_active_head); |
18ab4d4c DS |
82 | |
83 | /* | |
84 | * all available (active, not full) swap_info_structs | |
85 | * protected with swap_avail_lock, ordered by priority. | |
e2e3fdc7 | 86 | * This is used by folio_alloc_swap() instead of swap_active_head |
18ab4d4c | 87 | * because swap_active_head includes all swap_info_structs, |
e2e3fdc7 | 88 | * but folio_alloc_swap() doesn't need to look at full ones. |
18ab4d4c DS |
89 | * This uses its own lock instead of swap_lock because when a |
90 | * swap_info_struct changes between not-full/full, it needs to | |
91 | * add/remove itself to/from this list, but the swap_info_struct->lock | |
92 | * is held and the locking order requires swap_lock to be taken | |
93 | * before any swap_info_struct->lock. | |
94 | */ | |
bfc6b1ca | 95 | static struct plist_head *swap_avail_heads; |
18ab4d4c | 96 | static DEFINE_SPINLOCK(swap_avail_lock); |
1da177e4 | 97 | |
38b5faf4 | 98 | struct swap_info_struct *swap_info[MAX_SWAPFILES]; |
1da177e4 | 99 | |
fc0abb14 | 100 | static DEFINE_MUTEX(swapon_mutex); |
1da177e4 | 101 | |
66d7dd51 KS |
102 | static DECLARE_WAIT_QUEUE_HEAD(proc_poll_wait); |
103 | /* Activity counter to indicate that a swapon or swapoff has occurred */ | |
104 | static atomic_t proc_poll_event = ATOMIC_INIT(0); | |
105 | ||
81a0298b HY |
106 | atomic_t nr_rotate_swap = ATOMIC_INIT(0); |
107 | ||
c10d38cc DJ |
108 | static struct swap_info_struct *swap_type_to_swap_info(int type) |
109 | { | |
a4b45114 | 110 | if (type >= MAX_SWAPFILES) |
c10d38cc DJ |
111 | return NULL; |
112 | ||
a4b45114 | 113 | return READ_ONCE(swap_info[type]); /* rcu_dereference() */ |
c10d38cc DJ |
114 | } |
115 | ||
8d69aaee | 116 | static inline unsigned char swap_count(unsigned char ent) |
355cfa73 | 117 | { |
955c97f0 | 118 | return ent & ~SWAP_HAS_CACHE; /* may include COUNT_CONTINUED flag */ |
355cfa73 KH |
119 | } |
120 | ||
bcd49e86 HY |
121 | /* Reclaim the swap entry anyway if possible */ |
122 | #define TTRS_ANYWAY 0x1 | |
123 | /* | |
124 | * Reclaim the swap entry if there are no more mappings of the | |
125 | * corresponding page | |
126 | */ | |
127 | #define TTRS_UNMAPPED 0x2 | |
128 | /* Reclaim the swap entry if swap is getting full*/ | |
129 | #define TTRS_FULL 0x4 | |
130 | ||
efa90a98 | 131 | /* returns 1 if swap entry is freed */ |
bcd49e86 HY |
132 | static int __try_to_reclaim_swap(struct swap_info_struct *si, |
133 | unsigned long offset, unsigned long flags) | |
c9e44410 | 134 | { |
efa90a98 | 135 | swp_entry_t entry = swp_entry(si->type, offset); |
2c3f6194 | 136 | struct folio *folio; |
c9e44410 KH |
137 | int ret = 0; |
138 | ||
2c3f6194 | 139 | folio = filemap_get_folio(swap_address_space(entry), offset); |
66dabbb6 | 140 | if (IS_ERR(folio)) |
c9e44410 KH |
141 | return 0; |
142 | /* | |
bcd49e86 | 143 | * When this function is called from scan_swap_map_slots() and it's |
2c3f6194 | 144 | * called by vmscan.c at reclaiming folios. So we hold a folio lock |
bcd49e86 | 145 | * here. We have to use trylock for avoiding deadlock. This is a special |
2c3f6194 | 146 | * case and you should use folio_free_swap() with explicit folio_lock() |
c9e44410 KH |
147 | * in usual operations. |
148 | */ | |
2c3f6194 | 149 | if (folio_trylock(folio)) { |
bcd49e86 | 150 | if ((flags & TTRS_ANYWAY) || |
2c3f6194 | 151 | ((flags & TTRS_UNMAPPED) && !folio_mapped(folio)) || |
9202d527 | 152 | ((flags & TTRS_FULL) && mem_cgroup_swap_full(folio))) |
2c3f6194 MWO |
153 | ret = folio_free_swap(folio); |
154 | folio_unlock(folio); | |
c9e44410 | 155 | } |
2c3f6194 | 156 | folio_put(folio); |
c9e44410 KH |
157 | return ret; |
158 | } | |
355cfa73 | 159 | |
4efaceb1 AL |
160 | static inline struct swap_extent *first_se(struct swap_info_struct *sis) |
161 | { | |
162 | struct rb_node *rb = rb_first(&sis->swap_extent_root); | |
163 | return rb_entry(rb, struct swap_extent, rb_node); | |
164 | } | |
165 | ||
166 | static inline struct swap_extent *next_se(struct swap_extent *se) | |
167 | { | |
168 | struct rb_node *rb = rb_next(&se->rb_node); | |
169 | return rb ? rb_entry(rb, struct swap_extent, rb_node) : NULL; | |
170 | } | |
171 | ||
6a6ba831 HD |
172 | /* |
173 | * swapon tell device that all the old swap contents can be discarded, | |
174 | * to allow the swap device to optimize its wear-levelling. | |
175 | */ | |
176 | static int discard_swap(struct swap_info_struct *si) | |
177 | { | |
178 | struct swap_extent *se; | |
9625a5f2 HD |
179 | sector_t start_block; |
180 | sector_t nr_blocks; | |
6a6ba831 HD |
181 | int err = 0; |
182 | ||
9625a5f2 | 183 | /* Do not discard the swap header page! */ |
4efaceb1 | 184 | se = first_se(si); |
9625a5f2 HD |
185 | start_block = (se->start_block + 1) << (PAGE_SHIFT - 9); |
186 | nr_blocks = ((sector_t)se->nr_pages - 1) << (PAGE_SHIFT - 9); | |
187 | if (nr_blocks) { | |
188 | err = blkdev_issue_discard(si->bdev, start_block, | |
44abff2c | 189 | nr_blocks, GFP_KERNEL); |
9625a5f2 HD |
190 | if (err) |
191 | return err; | |
192 | cond_resched(); | |
193 | } | |
6a6ba831 | 194 | |
4efaceb1 | 195 | for (se = next_se(se); se; se = next_se(se)) { |
9625a5f2 HD |
196 | start_block = se->start_block << (PAGE_SHIFT - 9); |
197 | nr_blocks = (sector_t)se->nr_pages << (PAGE_SHIFT - 9); | |
6a6ba831 HD |
198 | |
199 | err = blkdev_issue_discard(si->bdev, start_block, | |
44abff2c | 200 | nr_blocks, GFP_KERNEL); |
6a6ba831 HD |
201 | if (err) |
202 | break; | |
203 | ||
204 | cond_resched(); | |
205 | } | |
206 | return err; /* That will often be -EOPNOTSUPP */ | |
207 | } | |
208 | ||
4efaceb1 AL |
209 | static struct swap_extent * |
210 | offset_to_swap_extent(struct swap_info_struct *sis, unsigned long offset) | |
211 | { | |
212 | struct swap_extent *se; | |
213 | struct rb_node *rb; | |
214 | ||
215 | rb = sis->swap_extent_root.rb_node; | |
216 | while (rb) { | |
217 | se = rb_entry(rb, struct swap_extent, rb_node); | |
218 | if (offset < se->start_page) | |
219 | rb = rb->rb_left; | |
220 | else if (offset >= se->start_page + se->nr_pages) | |
221 | rb = rb->rb_right; | |
222 | else | |
223 | return se; | |
224 | } | |
225 | /* It *must* be present */ | |
226 | BUG(); | |
227 | } | |
228 | ||
caf6912f JA |
229 | sector_t swap_page_sector(struct page *page) |
230 | { | |
231 | struct swap_info_struct *sis = page_swap_info(page); | |
232 | struct swap_extent *se; | |
233 | sector_t sector; | |
234 | pgoff_t offset; | |
235 | ||
236 | offset = __page_file_index(page); | |
237 | se = offset_to_swap_extent(sis, offset); | |
238 | sector = se->start_block + (offset - se->start_page); | |
239 | return sector << (PAGE_SHIFT - 9); | |
240 | } | |
241 | ||
7992fde7 HD |
242 | /* |
243 | * swap allocation tell device that a cluster of swap can now be discarded, | |
244 | * to allow the swap device to optimize its wear-levelling. | |
245 | */ | |
246 | static void discard_swap_cluster(struct swap_info_struct *si, | |
247 | pgoff_t start_page, pgoff_t nr_pages) | |
248 | { | |
4efaceb1 | 249 | struct swap_extent *se = offset_to_swap_extent(si, start_page); |
7992fde7 HD |
250 | |
251 | while (nr_pages) { | |
4efaceb1 AL |
252 | pgoff_t offset = start_page - se->start_page; |
253 | sector_t start_block = se->start_block + offset; | |
254 | sector_t nr_blocks = se->nr_pages - offset; | |
255 | ||
256 | if (nr_blocks > nr_pages) | |
257 | nr_blocks = nr_pages; | |
258 | start_page += nr_blocks; | |
259 | nr_pages -= nr_blocks; | |
260 | ||
261 | start_block <<= PAGE_SHIFT - 9; | |
262 | nr_blocks <<= PAGE_SHIFT - 9; | |
263 | if (blkdev_issue_discard(si->bdev, start_block, | |
44abff2c | 264 | nr_blocks, GFP_NOIO)) |
4efaceb1 | 265 | break; |
7992fde7 | 266 | |
4efaceb1 | 267 | se = next_se(se); |
7992fde7 HD |
268 | } |
269 | } | |
270 | ||
38d8b4e6 HY |
271 | #ifdef CONFIG_THP_SWAP |
272 | #define SWAPFILE_CLUSTER HPAGE_PMD_NR | |
a448f2d0 HY |
273 | |
274 | #define swap_entry_size(size) (size) | |
38d8b4e6 | 275 | #else |
048c27fd | 276 | #define SWAPFILE_CLUSTER 256 |
a448f2d0 HY |
277 | |
278 | /* | |
279 | * Define swap_entry_size() as constant to let compiler to optimize | |
280 | * out some code if !CONFIG_THP_SWAP | |
281 | */ | |
282 | #define swap_entry_size(size) 1 | |
38d8b4e6 | 283 | #endif |
048c27fd HD |
284 | #define LATENCY_LIMIT 256 |
285 | ||
2a8f9449 SL |
286 | static inline void cluster_set_flag(struct swap_cluster_info *info, |
287 | unsigned int flag) | |
288 | { | |
289 | info->flags = flag; | |
290 | } | |
291 | ||
292 | static inline unsigned int cluster_count(struct swap_cluster_info *info) | |
293 | { | |
294 | return info->data; | |
295 | } | |
296 | ||
297 | static inline void cluster_set_count(struct swap_cluster_info *info, | |
298 | unsigned int c) | |
299 | { | |
300 | info->data = c; | |
301 | } | |
302 | ||
303 | static inline void cluster_set_count_flag(struct swap_cluster_info *info, | |
304 | unsigned int c, unsigned int f) | |
305 | { | |
306 | info->flags = f; | |
307 | info->data = c; | |
308 | } | |
309 | ||
310 | static inline unsigned int cluster_next(struct swap_cluster_info *info) | |
311 | { | |
312 | return info->data; | |
313 | } | |
314 | ||
315 | static inline void cluster_set_next(struct swap_cluster_info *info, | |
316 | unsigned int n) | |
317 | { | |
318 | info->data = n; | |
319 | } | |
320 | ||
321 | static inline void cluster_set_next_flag(struct swap_cluster_info *info, | |
322 | unsigned int n, unsigned int f) | |
323 | { | |
324 | info->flags = f; | |
325 | info->data = n; | |
326 | } | |
327 | ||
328 | static inline bool cluster_is_free(struct swap_cluster_info *info) | |
329 | { | |
330 | return info->flags & CLUSTER_FLAG_FREE; | |
331 | } | |
332 | ||
333 | static inline bool cluster_is_null(struct swap_cluster_info *info) | |
334 | { | |
335 | return info->flags & CLUSTER_FLAG_NEXT_NULL; | |
336 | } | |
337 | ||
338 | static inline void cluster_set_null(struct swap_cluster_info *info) | |
339 | { | |
340 | info->flags = CLUSTER_FLAG_NEXT_NULL; | |
341 | info->data = 0; | |
342 | } | |
343 | ||
e0709829 HY |
344 | static inline bool cluster_is_huge(struct swap_cluster_info *info) |
345 | { | |
33ee011e HY |
346 | if (IS_ENABLED(CONFIG_THP_SWAP)) |
347 | return info->flags & CLUSTER_FLAG_HUGE; | |
348 | return false; | |
e0709829 HY |
349 | } |
350 | ||
351 | static inline void cluster_clear_huge(struct swap_cluster_info *info) | |
352 | { | |
353 | info->flags &= ~CLUSTER_FLAG_HUGE; | |
354 | } | |
355 | ||
235b6217 HY |
356 | static inline struct swap_cluster_info *lock_cluster(struct swap_info_struct *si, |
357 | unsigned long offset) | |
358 | { | |
359 | struct swap_cluster_info *ci; | |
360 | ||
361 | ci = si->cluster_info; | |
362 | if (ci) { | |
363 | ci += offset / SWAPFILE_CLUSTER; | |
364 | spin_lock(&ci->lock); | |
365 | } | |
366 | return ci; | |
367 | } | |
368 | ||
369 | static inline void unlock_cluster(struct swap_cluster_info *ci) | |
370 | { | |
371 | if (ci) | |
372 | spin_unlock(&ci->lock); | |
373 | } | |
374 | ||
59d98bf3 HY |
375 | /* |
376 | * Determine the locking method in use for this device. Return | |
377 | * swap_cluster_info if SSD-style cluster-based locking is in place. | |
378 | */ | |
235b6217 | 379 | static inline struct swap_cluster_info *lock_cluster_or_swap_info( |
59d98bf3 | 380 | struct swap_info_struct *si, unsigned long offset) |
235b6217 HY |
381 | { |
382 | struct swap_cluster_info *ci; | |
383 | ||
59d98bf3 | 384 | /* Try to use fine-grained SSD-style locking if available: */ |
235b6217 | 385 | ci = lock_cluster(si, offset); |
59d98bf3 | 386 | /* Otherwise, fall back to traditional, coarse locking: */ |
235b6217 HY |
387 | if (!ci) |
388 | spin_lock(&si->lock); | |
389 | ||
390 | return ci; | |
391 | } | |
392 | ||
393 | static inline void unlock_cluster_or_swap_info(struct swap_info_struct *si, | |
394 | struct swap_cluster_info *ci) | |
395 | { | |
396 | if (ci) | |
397 | unlock_cluster(ci); | |
398 | else | |
399 | spin_unlock(&si->lock); | |
400 | } | |
401 | ||
6b534915 HY |
402 | static inline bool cluster_list_empty(struct swap_cluster_list *list) |
403 | { | |
404 | return cluster_is_null(&list->head); | |
405 | } | |
406 | ||
407 | static inline unsigned int cluster_list_first(struct swap_cluster_list *list) | |
408 | { | |
409 | return cluster_next(&list->head); | |
410 | } | |
411 | ||
412 | static void cluster_list_init(struct swap_cluster_list *list) | |
413 | { | |
414 | cluster_set_null(&list->head); | |
415 | cluster_set_null(&list->tail); | |
416 | } | |
417 | ||
418 | static void cluster_list_add_tail(struct swap_cluster_list *list, | |
419 | struct swap_cluster_info *ci, | |
420 | unsigned int idx) | |
421 | { | |
422 | if (cluster_list_empty(list)) { | |
423 | cluster_set_next_flag(&list->head, idx, 0); | |
424 | cluster_set_next_flag(&list->tail, idx, 0); | |
425 | } else { | |
235b6217 | 426 | struct swap_cluster_info *ci_tail; |
6b534915 HY |
427 | unsigned int tail = cluster_next(&list->tail); |
428 | ||
235b6217 HY |
429 | /* |
430 | * Nested cluster lock, but both cluster locks are | |
431 | * only acquired when we held swap_info_struct->lock | |
432 | */ | |
433 | ci_tail = ci + tail; | |
434 | spin_lock_nested(&ci_tail->lock, SINGLE_DEPTH_NESTING); | |
435 | cluster_set_next(ci_tail, idx); | |
0ef017d1 | 436 | spin_unlock(&ci_tail->lock); |
6b534915 HY |
437 | cluster_set_next_flag(&list->tail, idx, 0); |
438 | } | |
439 | } | |
440 | ||
441 | static unsigned int cluster_list_del_first(struct swap_cluster_list *list, | |
442 | struct swap_cluster_info *ci) | |
443 | { | |
444 | unsigned int idx; | |
445 | ||
446 | idx = cluster_next(&list->head); | |
447 | if (cluster_next(&list->tail) == idx) { | |
448 | cluster_set_null(&list->head); | |
449 | cluster_set_null(&list->tail); | |
450 | } else | |
451 | cluster_set_next_flag(&list->head, | |
452 | cluster_next(&ci[idx]), 0); | |
453 | ||
454 | return idx; | |
455 | } | |
456 | ||
815c2c54 SL |
457 | /* Add a cluster to discard list and schedule it to do discard */ |
458 | static void swap_cluster_schedule_discard(struct swap_info_struct *si, | |
459 | unsigned int idx) | |
460 | { | |
461 | /* | |
bb243f7d | 462 | * If scan_swap_map_slots() can't find a free cluster, it will check |
815c2c54 | 463 | * si->swap_map directly. To make sure the discarding cluster isn't |
bb243f7d ML |
464 | * taken by scan_swap_map_slots(), mark the swap entries bad (occupied). |
465 | * It will be cleared after discard | |
815c2c54 SL |
466 | */ |
467 | memset(si->swap_map + idx * SWAPFILE_CLUSTER, | |
468 | SWAP_MAP_BAD, SWAPFILE_CLUSTER); | |
469 | ||
6b534915 | 470 | cluster_list_add_tail(&si->discard_clusters, si->cluster_info, idx); |
815c2c54 SL |
471 | |
472 | schedule_work(&si->discard_work); | |
473 | } | |
474 | ||
38d8b4e6 HY |
475 | static void __free_cluster(struct swap_info_struct *si, unsigned long idx) |
476 | { | |
477 | struct swap_cluster_info *ci = si->cluster_info; | |
478 | ||
479 | cluster_set_flag(ci + idx, CLUSTER_FLAG_FREE); | |
480 | cluster_list_add_tail(&si->free_clusters, ci, idx); | |
481 | } | |
482 | ||
815c2c54 SL |
483 | /* |
484 | * Doing discard actually. After a cluster discard is finished, the cluster | |
485 | * will be added to free cluster list. caller should hold si->lock. | |
486 | */ | |
487 | static void swap_do_scheduled_discard(struct swap_info_struct *si) | |
488 | { | |
235b6217 | 489 | struct swap_cluster_info *info, *ci; |
815c2c54 SL |
490 | unsigned int idx; |
491 | ||
492 | info = si->cluster_info; | |
493 | ||
6b534915 HY |
494 | while (!cluster_list_empty(&si->discard_clusters)) { |
495 | idx = cluster_list_del_first(&si->discard_clusters, info); | |
815c2c54 SL |
496 | spin_unlock(&si->lock); |
497 | ||
498 | discard_swap_cluster(si, idx * SWAPFILE_CLUSTER, | |
499 | SWAPFILE_CLUSTER); | |
500 | ||
501 | spin_lock(&si->lock); | |
235b6217 | 502 | ci = lock_cluster(si, idx * SWAPFILE_CLUSTER); |
38d8b4e6 | 503 | __free_cluster(si, idx); |
815c2c54 SL |
504 | memset(si->swap_map + idx * SWAPFILE_CLUSTER, |
505 | 0, SWAPFILE_CLUSTER); | |
235b6217 | 506 | unlock_cluster(ci); |
815c2c54 SL |
507 | } |
508 | } | |
509 | ||
510 | static void swap_discard_work(struct work_struct *work) | |
511 | { | |
512 | struct swap_info_struct *si; | |
513 | ||
514 | si = container_of(work, struct swap_info_struct, discard_work); | |
515 | ||
516 | spin_lock(&si->lock); | |
517 | swap_do_scheduled_discard(si); | |
518 | spin_unlock(&si->lock); | |
519 | } | |
520 | ||
63d8620e ML |
521 | static void swap_users_ref_free(struct percpu_ref *ref) |
522 | { | |
523 | struct swap_info_struct *si; | |
524 | ||
525 | si = container_of(ref, struct swap_info_struct, users); | |
526 | complete(&si->comp); | |
527 | } | |
528 | ||
38d8b4e6 HY |
529 | static void alloc_cluster(struct swap_info_struct *si, unsigned long idx) |
530 | { | |
531 | struct swap_cluster_info *ci = si->cluster_info; | |
532 | ||
533 | VM_BUG_ON(cluster_list_first(&si->free_clusters) != idx); | |
534 | cluster_list_del_first(&si->free_clusters, ci); | |
535 | cluster_set_count_flag(ci + idx, 0, 0); | |
536 | } | |
537 | ||
538 | static void free_cluster(struct swap_info_struct *si, unsigned long idx) | |
539 | { | |
540 | struct swap_cluster_info *ci = si->cluster_info + idx; | |
541 | ||
542 | VM_BUG_ON(cluster_count(ci) != 0); | |
543 | /* | |
544 | * If the swap is discardable, prepare discard the cluster | |
545 | * instead of free it immediately. The cluster will be freed | |
546 | * after discard. | |
547 | */ | |
548 | if ((si->flags & (SWP_WRITEOK | SWP_PAGE_DISCARD)) == | |
549 | (SWP_WRITEOK | SWP_PAGE_DISCARD)) { | |
550 | swap_cluster_schedule_discard(si, idx); | |
551 | return; | |
552 | } | |
553 | ||
554 | __free_cluster(si, idx); | |
555 | } | |
556 | ||
2a8f9449 SL |
557 | /* |
558 | * The cluster corresponding to page_nr will be used. The cluster will be | |
559 | * removed from free cluster list and its usage counter will be increased. | |
560 | */ | |
561 | static void inc_cluster_info_page(struct swap_info_struct *p, | |
562 | struct swap_cluster_info *cluster_info, unsigned long page_nr) | |
563 | { | |
564 | unsigned long idx = page_nr / SWAPFILE_CLUSTER; | |
565 | ||
566 | if (!cluster_info) | |
567 | return; | |
38d8b4e6 HY |
568 | if (cluster_is_free(&cluster_info[idx])) |
569 | alloc_cluster(p, idx); | |
2a8f9449 SL |
570 | |
571 | VM_BUG_ON(cluster_count(&cluster_info[idx]) >= SWAPFILE_CLUSTER); | |
572 | cluster_set_count(&cluster_info[idx], | |
573 | cluster_count(&cluster_info[idx]) + 1); | |
574 | } | |
575 | ||
576 | /* | |
577 | * The cluster corresponding to page_nr decreases one usage. If the usage | |
578 | * counter becomes 0, which means no page in the cluster is in using, we can | |
579 | * optionally discard the cluster and add it to free cluster list. | |
580 | */ | |
581 | static void dec_cluster_info_page(struct swap_info_struct *p, | |
582 | struct swap_cluster_info *cluster_info, unsigned long page_nr) | |
583 | { | |
584 | unsigned long idx = page_nr / SWAPFILE_CLUSTER; | |
585 | ||
586 | if (!cluster_info) | |
587 | return; | |
588 | ||
589 | VM_BUG_ON(cluster_count(&cluster_info[idx]) == 0); | |
590 | cluster_set_count(&cluster_info[idx], | |
591 | cluster_count(&cluster_info[idx]) - 1); | |
592 | ||
38d8b4e6 HY |
593 | if (cluster_count(&cluster_info[idx]) == 0) |
594 | free_cluster(p, idx); | |
2a8f9449 SL |
595 | } |
596 | ||
597 | /* | |
bb243f7d | 598 | * It's possible scan_swap_map_slots() uses a free cluster in the middle of free |
2a8f9449 SL |
599 | * cluster list. Avoiding such abuse to avoid list corruption. |
600 | */ | |
ebc2a1a6 SL |
601 | static bool |
602 | scan_swap_map_ssd_cluster_conflict(struct swap_info_struct *si, | |
2a8f9449 SL |
603 | unsigned long offset) |
604 | { | |
ebc2a1a6 SL |
605 | struct percpu_cluster *percpu_cluster; |
606 | bool conflict; | |
607 | ||
2a8f9449 | 608 | offset /= SWAPFILE_CLUSTER; |
6b534915 HY |
609 | conflict = !cluster_list_empty(&si->free_clusters) && |
610 | offset != cluster_list_first(&si->free_clusters) && | |
2a8f9449 | 611 | cluster_is_free(&si->cluster_info[offset]); |
ebc2a1a6 SL |
612 | |
613 | if (!conflict) | |
614 | return false; | |
615 | ||
616 | percpu_cluster = this_cpu_ptr(si->percpu_cluster); | |
617 | cluster_set_null(&percpu_cluster->index); | |
618 | return true; | |
619 | } | |
620 | ||
621 | /* | |
622 | * Try to get a swap entry from current cpu's swap entry pool (a cluster). This | |
623 | * might involve allocating a new cluster for current CPU too. | |
624 | */ | |
36005bae | 625 | static bool scan_swap_map_try_ssd_cluster(struct swap_info_struct *si, |
ebc2a1a6 SL |
626 | unsigned long *offset, unsigned long *scan_base) |
627 | { | |
628 | struct percpu_cluster *cluster; | |
235b6217 | 629 | struct swap_cluster_info *ci; |
235b6217 | 630 | unsigned long tmp, max; |
ebc2a1a6 SL |
631 | |
632 | new_cluster: | |
633 | cluster = this_cpu_ptr(si->percpu_cluster); | |
634 | if (cluster_is_null(&cluster->index)) { | |
6b534915 HY |
635 | if (!cluster_list_empty(&si->free_clusters)) { |
636 | cluster->index = si->free_clusters.head; | |
ebc2a1a6 SL |
637 | cluster->next = cluster_next(&cluster->index) * |
638 | SWAPFILE_CLUSTER; | |
6b534915 | 639 | } else if (!cluster_list_empty(&si->discard_clusters)) { |
ebc2a1a6 SL |
640 | /* |
641 | * we don't have free cluster but have some clusters in | |
49070588 HY |
642 | * discarding, do discard now and reclaim them, then |
643 | * reread cluster_next_cpu since we dropped si->lock | |
ebc2a1a6 SL |
644 | */ |
645 | swap_do_scheduled_discard(si); | |
49070588 HY |
646 | *scan_base = this_cpu_read(*si->cluster_next_cpu); |
647 | *offset = *scan_base; | |
ebc2a1a6 SL |
648 | goto new_cluster; |
649 | } else | |
36005bae | 650 | return false; |
ebc2a1a6 SL |
651 | } |
652 | ||
ebc2a1a6 SL |
653 | /* |
654 | * Other CPUs can use our cluster if they can't find a free cluster, | |
655 | * check if there is still free entry in the cluster | |
656 | */ | |
657 | tmp = cluster->next; | |
235b6217 HY |
658 | max = min_t(unsigned long, si->max, |
659 | (cluster_next(&cluster->index) + 1) * SWAPFILE_CLUSTER); | |
7b9e2de1 WY |
660 | if (tmp < max) { |
661 | ci = lock_cluster(si, tmp); | |
662 | while (tmp < max) { | |
663 | if (!si->swap_map[tmp]) | |
664 | break; | |
665 | tmp++; | |
666 | } | |
667 | unlock_cluster(ci); | |
ebc2a1a6 | 668 | } |
0fd0e19e | 669 | if (tmp >= max) { |
ebc2a1a6 SL |
670 | cluster_set_null(&cluster->index); |
671 | goto new_cluster; | |
672 | } | |
673 | cluster->next = tmp + 1; | |
674 | *offset = tmp; | |
675 | *scan_base = tmp; | |
fdff1deb | 676 | return true; |
2a8f9449 SL |
677 | } |
678 | ||
a2468cc9 AL |
679 | static void __del_from_avail_list(struct swap_info_struct *p) |
680 | { | |
681 | int nid; | |
682 | ||
6fe7d6b9 | 683 | assert_spin_locked(&p->lock); |
a2468cc9 AL |
684 | for_each_node(nid) |
685 | plist_del(&p->avail_lists[nid], &swap_avail_heads[nid]); | |
686 | } | |
687 | ||
688 | static void del_from_avail_list(struct swap_info_struct *p) | |
689 | { | |
690 | spin_lock(&swap_avail_lock); | |
691 | __del_from_avail_list(p); | |
692 | spin_unlock(&swap_avail_lock); | |
693 | } | |
694 | ||
38d8b4e6 HY |
695 | static void swap_range_alloc(struct swap_info_struct *si, unsigned long offset, |
696 | unsigned int nr_entries) | |
697 | { | |
698 | unsigned int end = offset + nr_entries - 1; | |
699 | ||
700 | if (offset == si->lowest_bit) | |
701 | si->lowest_bit += nr_entries; | |
702 | if (end == si->highest_bit) | |
a449bf58 | 703 | WRITE_ONCE(si->highest_bit, si->highest_bit - nr_entries); |
c8945306 | 704 | WRITE_ONCE(si->inuse_pages, si->inuse_pages + nr_entries); |
38d8b4e6 HY |
705 | if (si->inuse_pages == si->pages) { |
706 | si->lowest_bit = si->max; | |
707 | si->highest_bit = 0; | |
a2468cc9 | 708 | del_from_avail_list(si); |
38d8b4e6 HY |
709 | } |
710 | } | |
711 | ||
a2468cc9 AL |
712 | static void add_to_avail_list(struct swap_info_struct *p) |
713 | { | |
714 | int nid; | |
715 | ||
716 | spin_lock(&swap_avail_lock); | |
67490031 | 717 | for_each_node(nid) |
a2468cc9 | 718 | plist_add(&p->avail_lists[nid], &swap_avail_heads[nid]); |
a2468cc9 AL |
719 | spin_unlock(&swap_avail_lock); |
720 | } | |
721 | ||
38d8b4e6 HY |
722 | static void swap_range_free(struct swap_info_struct *si, unsigned long offset, |
723 | unsigned int nr_entries) | |
724 | { | |
3852f676 | 725 | unsigned long begin = offset; |
38d8b4e6 HY |
726 | unsigned long end = offset + nr_entries - 1; |
727 | void (*swap_slot_free_notify)(struct block_device *, unsigned long); | |
728 | ||
729 | if (offset < si->lowest_bit) | |
730 | si->lowest_bit = offset; | |
731 | if (end > si->highest_bit) { | |
732 | bool was_full = !si->highest_bit; | |
733 | ||
a449bf58 | 734 | WRITE_ONCE(si->highest_bit, end); |
a2468cc9 AL |
735 | if (was_full && (si->flags & SWP_WRITEOK)) |
736 | add_to_avail_list(si); | |
38d8b4e6 HY |
737 | } |
738 | atomic_long_add(nr_entries, &nr_swap_pages); | |
c8945306 | 739 | WRITE_ONCE(si->inuse_pages, si->inuse_pages - nr_entries); |
38d8b4e6 HY |
740 | if (si->flags & SWP_BLKDEV) |
741 | swap_slot_free_notify = | |
742 | si->bdev->bd_disk->fops->swap_slot_free_notify; | |
743 | else | |
744 | swap_slot_free_notify = NULL; | |
745 | while (offset <= end) { | |
8a84802e | 746 | arch_swap_invalidate_page(si->type, offset); |
38d8b4e6 HY |
747 | frontswap_invalidate_page(si->type, offset); |
748 | if (swap_slot_free_notify) | |
749 | swap_slot_free_notify(si->bdev, offset); | |
750 | offset++; | |
751 | } | |
3852f676 | 752 | clear_shadow_from_swap_cache(si->type, begin, end); |
38d8b4e6 HY |
753 | } |
754 | ||
49070588 HY |
755 | static void set_cluster_next(struct swap_info_struct *si, unsigned long next) |
756 | { | |
757 | unsigned long prev; | |
758 | ||
759 | if (!(si->flags & SWP_SOLIDSTATE)) { | |
760 | si->cluster_next = next; | |
761 | return; | |
762 | } | |
763 | ||
764 | prev = this_cpu_read(*si->cluster_next_cpu); | |
765 | /* | |
766 | * Cross the swap address space size aligned trunk, choose | |
767 | * another trunk randomly to avoid lock contention on swap | |
768 | * address space if possible. | |
769 | */ | |
770 | if ((prev >> SWAP_ADDRESS_SPACE_SHIFT) != | |
771 | (next >> SWAP_ADDRESS_SPACE_SHIFT)) { | |
772 | /* No free swap slots available */ | |
773 | if (si->highest_bit <= si->lowest_bit) | |
774 | return; | |
e8a533cb | 775 | next = get_random_u32_inclusive(si->lowest_bit, si->highest_bit); |
49070588 HY |
776 | next = ALIGN_DOWN(next, SWAP_ADDRESS_SPACE_PAGES); |
777 | next = max_t(unsigned int, next, si->lowest_bit); | |
778 | } | |
779 | this_cpu_write(*si->cluster_next_cpu, next); | |
780 | } | |
781 | ||
4b9ae842 ML |
782 | static bool swap_offset_available_and_locked(struct swap_info_struct *si, |
783 | unsigned long offset) | |
784 | { | |
785 | if (data_race(!si->swap_map[offset])) { | |
786 | spin_lock(&si->lock); | |
787 | return true; | |
788 | } | |
789 | ||
790 | if (vm_swap_full() && READ_ONCE(si->swap_map[offset]) == SWAP_HAS_CACHE) { | |
791 | spin_lock(&si->lock); | |
792 | return true; | |
793 | } | |
794 | ||
795 | return false; | |
796 | } | |
797 | ||
36005bae TC |
798 | static int scan_swap_map_slots(struct swap_info_struct *si, |
799 | unsigned char usage, int nr, | |
800 | swp_entry_t slots[]) | |
1da177e4 | 801 | { |
235b6217 | 802 | struct swap_cluster_info *ci; |
ebebbbe9 | 803 | unsigned long offset; |
c60aa176 | 804 | unsigned long scan_base; |
7992fde7 | 805 | unsigned long last_in_cluster = 0; |
048c27fd | 806 | int latency_ration = LATENCY_LIMIT; |
36005bae | 807 | int n_ret = 0; |
ed43af10 | 808 | bool scanned_many = false; |
36005bae | 809 | |
886bb7e9 | 810 | /* |
7dfad418 HD |
811 | * We try to cluster swap pages by allocating them sequentially |
812 | * in swap. Once we've allocated SWAPFILE_CLUSTER pages this | |
813 | * way, however, we resort to first-free allocation, starting | |
814 | * a new cluster. This prevents us from scattering swap pages | |
815 | * all over the entire swap partition, so that we reduce | |
816 | * overall disk seek times between swap pages. -- sct | |
817 | * But we do now try to find an empty cluster. -Andrea | |
c60aa176 | 818 | * And we let swap pages go all over an SSD partition. Hugh |
7dfad418 HD |
819 | */ |
820 | ||
52b7efdb | 821 | si->flags += SWP_SCANNING; |
49070588 HY |
822 | /* |
823 | * Use percpu scan base for SSD to reduce lock contention on | |
824 | * cluster and swap cache. For HDD, sequential access is more | |
825 | * important. | |
826 | */ | |
827 | if (si->flags & SWP_SOLIDSTATE) | |
828 | scan_base = this_cpu_read(*si->cluster_next_cpu); | |
829 | else | |
830 | scan_base = si->cluster_next; | |
831 | offset = scan_base; | |
ebebbbe9 | 832 | |
ebc2a1a6 SL |
833 | /* SSD algorithm */ |
834 | if (si->cluster_info) { | |
bd2d18da | 835 | if (!scan_swap_map_try_ssd_cluster(si, &offset, &scan_base)) |
36005bae | 836 | goto scan; |
f4eaf51a | 837 | } else if (unlikely(!si->cluster_nr--)) { |
ebebbbe9 HD |
838 | if (si->pages - si->inuse_pages < SWAPFILE_CLUSTER) { |
839 | si->cluster_nr = SWAPFILE_CLUSTER - 1; | |
840 | goto checks; | |
841 | } | |
2a8f9449 | 842 | |
ec8acf20 | 843 | spin_unlock(&si->lock); |
7dfad418 | 844 | |
c60aa176 HD |
845 | /* |
846 | * If seek is expensive, start searching for new cluster from | |
847 | * start of partition, to minimize the span of allocated swap. | |
50088c44 CY |
848 | * If seek is cheap, that is the SWP_SOLIDSTATE si->cluster_info |
849 | * case, just handled by scan_swap_map_try_ssd_cluster() above. | |
c60aa176 | 850 | */ |
50088c44 | 851 | scan_base = offset = si->lowest_bit; |
7dfad418 HD |
852 | last_in_cluster = offset + SWAPFILE_CLUSTER - 1; |
853 | ||
854 | /* Locate the first empty (unaligned) cluster */ | |
855 | for (; last_in_cluster <= si->highest_bit; offset++) { | |
1da177e4 | 856 | if (si->swap_map[offset]) |
7dfad418 HD |
857 | last_in_cluster = offset + SWAPFILE_CLUSTER; |
858 | else if (offset == last_in_cluster) { | |
ec8acf20 | 859 | spin_lock(&si->lock); |
ebebbbe9 HD |
860 | offset -= SWAPFILE_CLUSTER - 1; |
861 | si->cluster_next = offset; | |
862 | si->cluster_nr = SWAPFILE_CLUSTER - 1; | |
c60aa176 HD |
863 | goto checks; |
864 | } | |
865 | if (unlikely(--latency_ration < 0)) { | |
866 | cond_resched(); | |
867 | latency_ration = LATENCY_LIMIT; | |
868 | } | |
869 | } | |
870 | ||
871 | offset = scan_base; | |
ec8acf20 | 872 | spin_lock(&si->lock); |
ebebbbe9 | 873 | si->cluster_nr = SWAPFILE_CLUSTER - 1; |
1da177e4 | 874 | } |
7dfad418 | 875 | |
ebebbbe9 | 876 | checks: |
ebc2a1a6 | 877 | if (si->cluster_info) { |
36005bae TC |
878 | while (scan_swap_map_ssd_cluster_conflict(si, offset)) { |
879 | /* take a break if we already got some slots */ | |
880 | if (n_ret) | |
881 | goto done; | |
882 | if (!scan_swap_map_try_ssd_cluster(si, &offset, | |
883 | &scan_base)) | |
884 | goto scan; | |
885 | } | |
ebc2a1a6 | 886 | } |
ebebbbe9 | 887 | if (!(si->flags & SWP_WRITEOK)) |
52b7efdb | 888 | goto no_page; |
7dfad418 HD |
889 | if (!si->highest_bit) |
890 | goto no_page; | |
ebebbbe9 | 891 | if (offset > si->highest_bit) |
c60aa176 | 892 | scan_base = offset = si->lowest_bit; |
c9e44410 | 893 | |
235b6217 | 894 | ci = lock_cluster(si, offset); |
b73d7fce HD |
895 | /* reuse swap entry of cache-only swap if not busy. */ |
896 | if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) { | |
c9e44410 | 897 | int swap_was_freed; |
235b6217 | 898 | unlock_cluster(ci); |
ec8acf20 | 899 | spin_unlock(&si->lock); |
bcd49e86 | 900 | swap_was_freed = __try_to_reclaim_swap(si, offset, TTRS_ANYWAY); |
ec8acf20 | 901 | spin_lock(&si->lock); |
c9e44410 KH |
902 | /* entry was freed successfully, try to use this again */ |
903 | if (swap_was_freed) | |
904 | goto checks; | |
905 | goto scan; /* check next one */ | |
906 | } | |
907 | ||
235b6217 HY |
908 | if (si->swap_map[offset]) { |
909 | unlock_cluster(ci); | |
36005bae TC |
910 | if (!n_ret) |
911 | goto scan; | |
912 | else | |
913 | goto done; | |
235b6217 | 914 | } |
a449bf58 | 915 | WRITE_ONCE(si->swap_map[offset], usage); |
2872bb2d HY |
916 | inc_cluster_info_page(si, si->cluster_info, offset); |
917 | unlock_cluster(ci); | |
ebebbbe9 | 918 | |
38d8b4e6 | 919 | swap_range_alloc(si, offset, 1); |
36005bae TC |
920 | slots[n_ret++] = swp_entry(si->type, offset); |
921 | ||
922 | /* got enough slots or reach max slots? */ | |
923 | if ((n_ret == nr) || (offset >= si->highest_bit)) | |
924 | goto done; | |
925 | ||
926 | /* search for next available slot */ | |
927 | ||
928 | /* time to take a break? */ | |
929 | if (unlikely(--latency_ration < 0)) { | |
930 | if (n_ret) | |
931 | goto done; | |
932 | spin_unlock(&si->lock); | |
933 | cond_resched(); | |
934 | spin_lock(&si->lock); | |
935 | latency_ration = LATENCY_LIMIT; | |
936 | } | |
937 | ||
938 | /* try to get more slots in cluster */ | |
939 | if (si->cluster_info) { | |
940 | if (scan_swap_map_try_ssd_cluster(si, &offset, &scan_base)) | |
941 | goto checks; | |
f4eaf51a WY |
942 | } else if (si->cluster_nr && !si->swap_map[++offset]) { |
943 | /* non-ssd case, still more slots in cluster? */ | |
36005bae TC |
944 | --si->cluster_nr; |
945 | goto checks; | |
946 | } | |
7992fde7 | 947 | |
ed43af10 HY |
948 | /* |
949 | * Even if there's no free clusters available (fragmented), | |
950 | * try to scan a little more quickly with lock held unless we | |
951 | * have scanned too many slots already. | |
952 | */ | |
953 | if (!scanned_many) { | |
954 | unsigned long scan_limit; | |
955 | ||
956 | if (offset < scan_base) | |
957 | scan_limit = scan_base; | |
958 | else | |
959 | scan_limit = si->highest_bit; | |
960 | for (; offset <= scan_limit && --latency_ration > 0; | |
961 | offset++) { | |
962 | if (!si->swap_map[offset]) | |
963 | goto checks; | |
964 | } | |
965 | } | |
966 | ||
36005bae | 967 | done: |
49070588 | 968 | set_cluster_next(si, offset + 1); |
36005bae TC |
969 | si->flags -= SWP_SCANNING; |
970 | return n_ret; | |
7dfad418 | 971 | |
ebebbbe9 | 972 | scan: |
ec8acf20 | 973 | spin_unlock(&si->lock); |
a449bf58 | 974 | while (++offset <= READ_ONCE(si->highest_bit)) { |
048c27fd HD |
975 | if (unlikely(--latency_ration < 0)) { |
976 | cond_resched(); | |
977 | latency_ration = LATENCY_LIMIT; | |
ed43af10 | 978 | scanned_many = true; |
048c27fd | 979 | } |
de1ccfb6 CW |
980 | if (swap_offset_available_and_locked(si, offset)) |
981 | goto checks; | |
7dfad418 | 982 | } |
c60aa176 | 983 | offset = si->lowest_bit; |
a5998061 | 984 | while (offset < scan_base) { |
c60aa176 HD |
985 | if (unlikely(--latency_ration < 0)) { |
986 | cond_resched(); | |
987 | latency_ration = LATENCY_LIMIT; | |
ed43af10 | 988 | scanned_many = true; |
c60aa176 | 989 | } |
de1ccfb6 CW |
990 | if (swap_offset_available_and_locked(si, offset)) |
991 | goto checks; | |
a5998061 | 992 | offset++; |
c60aa176 | 993 | } |
ec8acf20 | 994 | spin_lock(&si->lock); |
7dfad418 HD |
995 | |
996 | no_page: | |
52b7efdb | 997 | si->flags -= SWP_SCANNING; |
36005bae | 998 | return n_ret; |
1da177e4 LT |
999 | } |
1000 | ||
38d8b4e6 HY |
1001 | static int swap_alloc_cluster(struct swap_info_struct *si, swp_entry_t *slot) |
1002 | { | |
1003 | unsigned long idx; | |
1004 | struct swap_cluster_info *ci; | |
661c7566 | 1005 | unsigned long offset; |
38d8b4e6 | 1006 | |
fe5266d5 HY |
1007 | /* |
1008 | * Should not even be attempting cluster allocations when huge | |
1009 | * page swap is disabled. Warn and fail the allocation. | |
1010 | */ | |
1011 | if (!IS_ENABLED(CONFIG_THP_SWAP)) { | |
1012 | VM_WARN_ON_ONCE(1); | |
1013 | return 0; | |
1014 | } | |
1015 | ||
38d8b4e6 HY |
1016 | if (cluster_list_empty(&si->free_clusters)) |
1017 | return 0; | |
1018 | ||
1019 | idx = cluster_list_first(&si->free_clusters); | |
1020 | offset = idx * SWAPFILE_CLUSTER; | |
1021 | ci = lock_cluster(si, offset); | |
1022 | alloc_cluster(si, idx); | |
e0709829 | 1023 | cluster_set_count_flag(ci, SWAPFILE_CLUSTER, CLUSTER_FLAG_HUGE); |
38d8b4e6 | 1024 | |
661c7566 | 1025 | memset(si->swap_map + offset, SWAP_HAS_CACHE, SWAPFILE_CLUSTER); |
38d8b4e6 HY |
1026 | unlock_cluster(ci); |
1027 | swap_range_alloc(si, offset, SWAPFILE_CLUSTER); | |
1028 | *slot = swp_entry(si->type, offset); | |
1029 | ||
1030 | return 1; | |
1031 | } | |
1032 | ||
1033 | static void swap_free_cluster(struct swap_info_struct *si, unsigned long idx) | |
1034 | { | |
1035 | unsigned long offset = idx * SWAPFILE_CLUSTER; | |
1036 | struct swap_cluster_info *ci; | |
1037 | ||
1038 | ci = lock_cluster(si, offset); | |
979aafa5 | 1039 | memset(si->swap_map + offset, 0, SWAPFILE_CLUSTER); |
38d8b4e6 HY |
1040 | cluster_set_count_flag(ci, 0, 0); |
1041 | free_cluster(si, idx); | |
1042 | unlock_cluster(ci); | |
1043 | swap_range_free(si, offset, SWAPFILE_CLUSTER); | |
1044 | } | |
38d8b4e6 | 1045 | |
5d5e8f19 | 1046 | int get_swap_pages(int n_goal, swp_entry_t swp_entries[], int entry_size) |
1da177e4 | 1047 | { |
5d5e8f19 | 1048 | unsigned long size = swap_entry_size(entry_size); |
adfab836 | 1049 | struct swap_info_struct *si, *next; |
36005bae TC |
1050 | long avail_pgs; |
1051 | int n_ret = 0; | |
a2468cc9 | 1052 | int node; |
1da177e4 | 1053 | |
38d8b4e6 | 1054 | /* Only single cluster request supported */ |
5d5e8f19 | 1055 | WARN_ON_ONCE(n_goal > 1 && size == SWAPFILE_CLUSTER); |
38d8b4e6 | 1056 | |
b50da6e9 ZH |
1057 | spin_lock(&swap_avail_lock); |
1058 | ||
5d5e8f19 | 1059 | avail_pgs = atomic_long_read(&nr_swap_pages) / size; |
b50da6e9 ZH |
1060 | if (avail_pgs <= 0) { |
1061 | spin_unlock(&swap_avail_lock); | |
fb4f88dc | 1062 | goto noswap; |
b50da6e9 | 1063 | } |
36005bae | 1064 | |
08d3090f | 1065 | n_goal = min3((long)n_goal, (long)SWAP_BATCH, avail_pgs); |
36005bae | 1066 | |
5d5e8f19 | 1067 | atomic_long_sub(n_goal * size, &nr_swap_pages); |
fb4f88dc | 1068 | |
18ab4d4c | 1069 | start_over: |
a2468cc9 AL |
1070 | node = numa_node_id(); |
1071 | plist_for_each_entry_safe(si, next, &swap_avail_heads[node], avail_lists[node]) { | |
18ab4d4c | 1072 | /* requeue si to after same-priority siblings */ |
a2468cc9 | 1073 | plist_requeue(&si->avail_lists[node], &swap_avail_heads[node]); |
18ab4d4c | 1074 | spin_unlock(&swap_avail_lock); |
ec8acf20 | 1075 | spin_lock(&si->lock); |
adfab836 | 1076 | if (!si->highest_bit || !(si->flags & SWP_WRITEOK)) { |
18ab4d4c | 1077 | spin_lock(&swap_avail_lock); |
a2468cc9 | 1078 | if (plist_node_empty(&si->avail_lists[node])) { |
18ab4d4c DS |
1079 | spin_unlock(&si->lock); |
1080 | goto nextsi; | |
1081 | } | |
1082 | WARN(!si->highest_bit, | |
1083 | "swap_info %d in list but !highest_bit\n", | |
1084 | si->type); | |
1085 | WARN(!(si->flags & SWP_WRITEOK), | |
1086 | "swap_info %d in list but !SWP_WRITEOK\n", | |
1087 | si->type); | |
a2468cc9 | 1088 | __del_from_avail_list(si); |
ec8acf20 | 1089 | spin_unlock(&si->lock); |
18ab4d4c | 1090 | goto nextsi; |
ec8acf20 | 1091 | } |
5d5e8f19 | 1092 | if (size == SWAPFILE_CLUSTER) { |
41663430 | 1093 | if (si->flags & SWP_BLKDEV) |
f0eea189 HY |
1094 | n_ret = swap_alloc_cluster(si, swp_entries); |
1095 | } else | |
38d8b4e6 HY |
1096 | n_ret = scan_swap_map_slots(si, SWAP_HAS_CACHE, |
1097 | n_goal, swp_entries); | |
ec8acf20 | 1098 | spin_unlock(&si->lock); |
5d5e8f19 | 1099 | if (n_ret || size == SWAPFILE_CLUSTER) |
36005bae | 1100 | goto check_out; |
7717fc1a | 1101 | cond_resched(); |
36005bae | 1102 | |
18ab4d4c DS |
1103 | spin_lock(&swap_avail_lock); |
1104 | nextsi: | |
adfab836 DS |
1105 | /* |
1106 | * if we got here, it's likely that si was almost full before, | |
bb243f7d ML |
1107 | * and since scan_swap_map_slots() can drop the si->lock, |
1108 | * multiple callers probably all tried to get a page from the | |
1109 | * same si and it filled up before we could get one; or, the si | |
1110 | * filled up between us dropping swap_avail_lock and taking | |
1111 | * si->lock. Since we dropped the swap_avail_lock, the | |
1112 | * swap_avail_head list may have been modified; so if next is | |
1113 | * still in the swap_avail_head list then try it, otherwise | |
1114 | * start over if we have not gotten any slots. | |
adfab836 | 1115 | */ |
a2468cc9 | 1116 | if (plist_node_empty(&next->avail_lists[node])) |
18ab4d4c | 1117 | goto start_over; |
1da177e4 | 1118 | } |
fb4f88dc | 1119 | |
18ab4d4c DS |
1120 | spin_unlock(&swap_avail_lock); |
1121 | ||
36005bae TC |
1122 | check_out: |
1123 | if (n_ret < n_goal) | |
5d5e8f19 | 1124 | atomic_long_add((long)(n_goal - n_ret) * size, |
38d8b4e6 | 1125 | &nr_swap_pages); |
fb4f88dc | 1126 | noswap: |
36005bae TC |
1127 | return n_ret; |
1128 | } | |
1129 | ||
afba72b1 | 1130 | static struct swap_info_struct *_swap_info_get(swp_entry_t entry) |
1da177e4 | 1131 | { |
73c34b6a | 1132 | struct swap_info_struct *p; |
eb085574 | 1133 | unsigned long offset; |
1da177e4 LT |
1134 | |
1135 | if (!entry.val) | |
1136 | goto out; | |
eb085574 | 1137 | p = swp_swap_info(entry); |
c10d38cc | 1138 | if (!p) |
1da177e4 | 1139 | goto bad_nofile; |
a449bf58 | 1140 | if (data_race(!(p->flags & SWP_USED))) |
1da177e4 LT |
1141 | goto bad_device; |
1142 | offset = swp_offset(entry); | |
1143 | if (offset >= p->max) | |
1144 | goto bad_offset; | |
afba72b1 ML |
1145 | if (data_race(!p->swap_map[swp_offset(entry)])) |
1146 | goto bad_free; | |
1da177e4 LT |
1147 | return p; |
1148 | ||
afba72b1 ML |
1149 | bad_free: |
1150 | pr_err("%s: %s%08lx\n", __func__, Unused_offset, entry.val); | |
1151 | goto out; | |
1da177e4 | 1152 | bad_offset: |
cf532faa | 1153 | pr_err("%s: %s%08lx\n", __func__, Bad_offset, entry.val); |
1da177e4 LT |
1154 | goto out; |
1155 | bad_device: | |
cf532faa | 1156 | pr_err("%s: %s%08lx\n", __func__, Unused_file, entry.val); |
1da177e4 LT |
1157 | goto out; |
1158 | bad_nofile: | |
cf532faa | 1159 | pr_err("%s: %s%08lx\n", __func__, Bad_file, entry.val); |
1da177e4 LT |
1160 | out: |
1161 | return NULL; | |
886bb7e9 | 1162 | } |
1da177e4 | 1163 | |
7c00bafe TC |
1164 | static struct swap_info_struct *swap_info_get_cont(swp_entry_t entry, |
1165 | struct swap_info_struct *q) | |
1166 | { | |
1167 | struct swap_info_struct *p; | |
1168 | ||
1169 | p = _swap_info_get(entry); | |
1170 | ||
1171 | if (p != q) { | |
1172 | if (q != NULL) | |
1173 | spin_unlock(&q->lock); | |
1174 | if (p != NULL) | |
1175 | spin_lock(&p->lock); | |
1176 | } | |
1177 | return p; | |
1178 | } | |
1179 | ||
b32d5f32 HY |
1180 | static unsigned char __swap_entry_free_locked(struct swap_info_struct *p, |
1181 | unsigned long offset, | |
1182 | unsigned char usage) | |
1da177e4 | 1183 | { |
8d69aaee HD |
1184 | unsigned char count; |
1185 | unsigned char has_cache; | |
235b6217 | 1186 | |
253d553b | 1187 | count = p->swap_map[offset]; |
235b6217 | 1188 | |
253d553b HD |
1189 | has_cache = count & SWAP_HAS_CACHE; |
1190 | count &= ~SWAP_HAS_CACHE; | |
355cfa73 | 1191 | |
253d553b | 1192 | if (usage == SWAP_HAS_CACHE) { |
355cfa73 | 1193 | VM_BUG_ON(!has_cache); |
253d553b | 1194 | has_cache = 0; |
aaa46865 HD |
1195 | } else if (count == SWAP_MAP_SHMEM) { |
1196 | /* | |
1197 | * Or we could insist on shmem.c using a special | |
1198 | * swap_shmem_free() and free_shmem_swap_and_cache()... | |
1199 | */ | |
1200 | count = 0; | |
570a335b HD |
1201 | } else if ((count & ~COUNT_CONTINUED) <= SWAP_MAP_MAX) { |
1202 | if (count == COUNT_CONTINUED) { | |
1203 | if (swap_count_continued(p, offset, count)) | |
1204 | count = SWAP_MAP_MAX | COUNT_CONTINUED; | |
1205 | else | |
1206 | count = SWAP_MAP_MAX; | |
1207 | } else | |
1208 | count--; | |
1209 | } | |
253d553b | 1210 | |
253d553b | 1211 | usage = count | has_cache; |
a449bf58 QC |
1212 | if (usage) |
1213 | WRITE_ONCE(p->swap_map[offset], usage); | |
1214 | else | |
1215 | WRITE_ONCE(p->swap_map[offset], SWAP_HAS_CACHE); | |
7c00bafe | 1216 | |
b32d5f32 HY |
1217 | return usage; |
1218 | } | |
1219 | ||
eb085574 | 1220 | /* |
a95722a0 HY |
1221 | * When we get a swap entry, if there aren't some other ways to |
1222 | * prevent swapoff, such as the folio in swap cache is locked, page | |
1223 | * table lock is held, etc., the swap entry may become invalid because | |
1224 | * of swapoff. Then, we need to enclose all swap related functions | |
1225 | * with get_swap_device() and put_swap_device(), unless the swap | |
1226 | * functions call get/put_swap_device() by themselves. | |
1227 | * | |
eb085574 HY |
1228 | * Check whether swap entry is valid in the swap device. If so, |
1229 | * return pointer to swap_info_struct, and keep the swap entry valid | |
1230 | * via preventing the swap device from being swapoff, until | |
1231 | * put_swap_device() is called. Otherwise return NULL. | |
1232 | * | |
eb085574 | 1233 | * Notice that swapoff or swapoff+swapon can still happen before the |
63d8620e ML |
1234 | * percpu_ref_tryget_live() in get_swap_device() or after the |
1235 | * percpu_ref_put() in put_swap_device() if there isn't any other way | |
a95722a0 HY |
1236 | * to prevent swapoff. The caller must be prepared for that. For |
1237 | * example, the following situation is possible. | |
eb085574 HY |
1238 | * |
1239 | * CPU1 CPU2 | |
1240 | * do_swap_page() | |
1241 | * ... swapoff+swapon | |
1242 | * __read_swap_cache_async() | |
1243 | * swapcache_prepare() | |
1244 | * __swap_duplicate() | |
1245 | * // check swap_map | |
1246 | * // verify PTE not changed | |
1247 | * | |
1248 | * In __swap_duplicate(), the swap_map need to be checked before | |
1249 | * changing partly because the specified swap entry may be for another | |
1250 | * swap device which has been swapoff. And in do_swap_page(), after | |
1251 | * the page is read from the swap device, the PTE is verified not | |
1252 | * changed with the page table locked to check whether the swap device | |
1253 | * has been swapoff or swapoff+swapon. | |
1254 | */ | |
1255 | struct swap_info_struct *get_swap_device(swp_entry_t entry) | |
1256 | { | |
1257 | struct swap_info_struct *si; | |
1258 | unsigned long offset; | |
1259 | ||
1260 | if (!entry.val) | |
1261 | goto out; | |
1262 | si = swp_swap_info(entry); | |
1263 | if (!si) | |
1264 | goto bad_nofile; | |
63d8620e ML |
1265 | if (!percpu_ref_tryget_live(&si->users)) |
1266 | goto out; | |
1267 | /* | |
1268 | * Guarantee the si->users are checked before accessing other | |
1269 | * fields of swap_info_struct. | |
1270 | * | |
1271 | * Paired with the spin_unlock() after setup_swap_info() in | |
1272 | * enable_swap_info(). | |
1273 | */ | |
1274 | smp_rmb(); | |
eb085574 HY |
1275 | offset = swp_offset(entry); |
1276 | if (offset >= si->max) | |
63d8620e | 1277 | goto put_out; |
eb085574 HY |
1278 | |
1279 | return si; | |
1280 | bad_nofile: | |
1281 | pr_err("%s: %s%08lx\n", __func__, Bad_file, entry.val); | |
1282 | out: | |
1283 | return NULL; | |
63d8620e | 1284 | put_out: |
23b230ba | 1285 | pr_err("%s: %s%08lx\n", __func__, Bad_offset, entry.val); |
63d8620e | 1286 | percpu_ref_put(&si->users); |
eb085574 HY |
1287 | return NULL; |
1288 | } | |
1289 | ||
b32d5f32 | 1290 | static unsigned char __swap_entry_free(struct swap_info_struct *p, |
33e16272 | 1291 | swp_entry_t entry) |
b32d5f32 HY |
1292 | { |
1293 | struct swap_cluster_info *ci; | |
1294 | unsigned long offset = swp_offset(entry); | |
33e16272 | 1295 | unsigned char usage; |
b32d5f32 HY |
1296 | |
1297 | ci = lock_cluster_or_swap_info(p, offset); | |
33e16272 | 1298 | usage = __swap_entry_free_locked(p, offset, 1); |
7c00bafe | 1299 | unlock_cluster_or_swap_info(p, ci); |
10e364da HY |
1300 | if (!usage) |
1301 | free_swap_slot(entry); | |
7c00bafe TC |
1302 | |
1303 | return usage; | |
1304 | } | |
355cfa73 | 1305 | |
7c00bafe TC |
1306 | static void swap_entry_free(struct swap_info_struct *p, swp_entry_t entry) |
1307 | { | |
1308 | struct swap_cluster_info *ci; | |
1309 | unsigned long offset = swp_offset(entry); | |
1310 | unsigned char count; | |
1311 | ||
1312 | ci = lock_cluster(p, offset); | |
1313 | count = p->swap_map[offset]; | |
1314 | VM_BUG_ON(count != SWAP_HAS_CACHE); | |
1315 | p->swap_map[offset] = 0; | |
1316 | dec_cluster_info_page(p, p->cluster_info, offset); | |
235b6217 HY |
1317 | unlock_cluster(ci); |
1318 | ||
38d8b4e6 HY |
1319 | mem_cgroup_uncharge_swap(entry, 1); |
1320 | swap_range_free(p, offset, 1); | |
1da177e4 LT |
1321 | } |
1322 | ||
1323 | /* | |
2de1a7e4 | 1324 | * Caller has made sure that the swap device corresponding to entry |
1da177e4 LT |
1325 | * is still around or has not been recycled. |
1326 | */ | |
1327 | void swap_free(swp_entry_t entry) | |
1328 | { | |
73c34b6a | 1329 | struct swap_info_struct *p; |
1da177e4 | 1330 | |
235b6217 | 1331 | p = _swap_info_get(entry); |
10e364da | 1332 | if (p) |
33e16272 | 1333 | __swap_entry_free(p, entry); |
1da177e4 LT |
1334 | } |
1335 | ||
cb4b86ba KH |
1336 | /* |
1337 | * Called after dropping swapcache to decrease refcnt to swap entries. | |
1338 | */ | |
4081f744 | 1339 | void put_swap_folio(struct folio *folio, swp_entry_t entry) |
38d8b4e6 HY |
1340 | { |
1341 | unsigned long offset = swp_offset(entry); | |
1342 | unsigned long idx = offset / SWAPFILE_CLUSTER; | |
1343 | struct swap_cluster_info *ci; | |
1344 | struct swap_info_struct *si; | |
1345 | unsigned char *map; | |
a3aea839 HY |
1346 | unsigned int i, free_entries = 0; |
1347 | unsigned char val; | |
4081f744 | 1348 | int size = swap_entry_size(folio_nr_pages(folio)); |
fe5266d5 | 1349 | |
a3aea839 | 1350 | si = _swap_info_get(entry); |
38d8b4e6 HY |
1351 | if (!si) |
1352 | return; | |
1353 | ||
c2343d27 | 1354 | ci = lock_cluster_or_swap_info(si, offset); |
a448f2d0 | 1355 | if (size == SWAPFILE_CLUSTER) { |
a448f2d0 HY |
1356 | VM_BUG_ON(!cluster_is_huge(ci)); |
1357 | map = si->swap_map + offset; | |
1358 | for (i = 0; i < SWAPFILE_CLUSTER; i++) { | |
1359 | val = map[i]; | |
1360 | VM_BUG_ON(!(val & SWAP_HAS_CACHE)); | |
1361 | if (val == SWAP_HAS_CACHE) | |
1362 | free_entries++; | |
1363 | } | |
a448f2d0 | 1364 | cluster_clear_huge(ci); |
a448f2d0 | 1365 | if (free_entries == SWAPFILE_CLUSTER) { |
c2343d27 | 1366 | unlock_cluster_or_swap_info(si, ci); |
a448f2d0 | 1367 | spin_lock(&si->lock); |
a448f2d0 HY |
1368 | mem_cgroup_uncharge_swap(entry, SWAPFILE_CLUSTER); |
1369 | swap_free_cluster(si, idx); | |
1370 | spin_unlock(&si->lock); | |
1371 | return; | |
1372 | } | |
1373 | } | |
c2343d27 HY |
1374 | for (i = 0; i < size; i++, entry.val++) { |
1375 | if (!__swap_entry_free_locked(si, offset + i, SWAP_HAS_CACHE)) { | |
1376 | unlock_cluster_or_swap_info(si, ci); | |
1377 | free_swap_slot(entry); | |
1378 | if (i == size - 1) | |
1379 | return; | |
1380 | lock_cluster_or_swap_info(si, offset); | |
a3aea839 HY |
1381 | } |
1382 | } | |
c2343d27 | 1383 | unlock_cluster_or_swap_info(si, ci); |
38d8b4e6 | 1384 | } |
59807685 | 1385 | |
fe5266d5 | 1386 | #ifdef CONFIG_THP_SWAP |
59807685 HY |
1387 | int split_swap_cluster(swp_entry_t entry) |
1388 | { | |
1389 | struct swap_info_struct *si; | |
1390 | struct swap_cluster_info *ci; | |
1391 | unsigned long offset = swp_offset(entry); | |
1392 | ||
1393 | si = _swap_info_get(entry); | |
1394 | if (!si) | |
1395 | return -EBUSY; | |
1396 | ci = lock_cluster(si, offset); | |
1397 | cluster_clear_huge(ci); | |
1398 | unlock_cluster(ci); | |
1399 | return 0; | |
1400 | } | |
fe5266d5 | 1401 | #endif |
38d8b4e6 | 1402 | |
155b5f88 HY |
1403 | static int swp_entry_cmp(const void *ent1, const void *ent2) |
1404 | { | |
1405 | const swp_entry_t *e1 = ent1, *e2 = ent2; | |
1406 | ||
1407 | return (int)swp_type(*e1) - (int)swp_type(*e2); | |
1408 | } | |
1409 | ||
7c00bafe TC |
1410 | void swapcache_free_entries(swp_entry_t *entries, int n) |
1411 | { | |
1412 | struct swap_info_struct *p, *prev; | |
1413 | int i; | |
1414 | ||
1415 | if (n <= 0) | |
1416 | return; | |
1417 | ||
1418 | prev = NULL; | |
1419 | p = NULL; | |
155b5f88 HY |
1420 | |
1421 | /* | |
1422 | * Sort swap entries by swap device, so each lock is only taken once. | |
1423 | * nr_swapfiles isn't absolutely correct, but the overhead of sort() is | |
1424 | * so low that it isn't necessary to optimize further. | |
1425 | */ | |
1426 | if (nr_swapfiles > 1) | |
1427 | sort(entries, n, sizeof(entries[0]), swp_entry_cmp, NULL); | |
7c00bafe TC |
1428 | for (i = 0; i < n; ++i) { |
1429 | p = swap_info_get_cont(entries[i], prev); | |
1430 | if (p) | |
1431 | swap_entry_free(p, entries[i]); | |
7c00bafe TC |
1432 | prev = p; |
1433 | } | |
235b6217 | 1434 | if (p) |
7c00bafe | 1435 | spin_unlock(&p->lock); |
cb4b86ba KH |
1436 | } |
1437 | ||
eb085574 | 1438 | int __swap_count(swp_entry_t entry) |
aa8d22a1 | 1439 | { |
f9f956b5 | 1440 | struct swap_info_struct *si = swp_swap_info(entry); |
aa8d22a1 MK |
1441 | pgoff_t offset = swp_offset(entry); |
1442 | ||
f9f956b5 | 1443 | return swap_count(si->swap_map[offset]); |
aa8d22a1 MK |
1444 | } |
1445 | ||
14d01ee9 MWO |
1446 | /* |
1447 | * How many references to @entry are currently swapped out? | |
1448 | * This does not give an exact answer when swap count is continued, | |
1449 | * but does include the high COUNT_CONTINUED flag to allow for that. | |
1450 | */ | |
3ecdeb0f | 1451 | int swap_swapcount(struct swap_info_struct *si, swp_entry_t entry) |
322b8afe | 1452 | { |
322b8afe HY |
1453 | pgoff_t offset = swp_offset(entry); |
1454 | struct swap_cluster_info *ci; | |
14d01ee9 | 1455 | int count; |
322b8afe HY |
1456 | |
1457 | ci = lock_cluster_or_swap_info(si, offset); | |
1458 | count = swap_count(si->swap_map[offset]); | |
1459 | unlock_cluster_or_swap_info(si, ci); | |
1460 | return count; | |
1461 | } | |
1462 | ||
8334b962 MK |
1463 | /* |
1464 | * How many references to @entry are currently swapped out? | |
1465 | * This considers COUNT_CONTINUED so it returns exact answer. | |
1466 | */ | |
1467 | int swp_swapcount(swp_entry_t entry) | |
1468 | { | |
1469 | int count, tmp_count, n; | |
1470 | struct swap_info_struct *p; | |
235b6217 | 1471 | struct swap_cluster_info *ci; |
8334b962 MK |
1472 | struct page *page; |
1473 | pgoff_t offset; | |
1474 | unsigned char *map; | |
1475 | ||
235b6217 | 1476 | p = _swap_info_get(entry); |
8334b962 MK |
1477 | if (!p) |
1478 | return 0; | |
1479 | ||
235b6217 HY |
1480 | offset = swp_offset(entry); |
1481 | ||
1482 | ci = lock_cluster_or_swap_info(p, offset); | |
1483 | ||
1484 | count = swap_count(p->swap_map[offset]); | |
8334b962 MK |
1485 | if (!(count & COUNT_CONTINUED)) |
1486 | goto out; | |
1487 | ||
1488 | count &= ~COUNT_CONTINUED; | |
1489 | n = SWAP_MAP_MAX + 1; | |
1490 | ||
8334b962 MK |
1491 | page = vmalloc_to_page(p->swap_map + offset); |
1492 | offset &= ~PAGE_MASK; | |
1493 | VM_BUG_ON(page_private(page) != SWP_CONTINUED); | |
1494 | ||
1495 | do { | |
a8ae4991 | 1496 | page = list_next_entry(page, lru); |
8334b962 MK |
1497 | map = kmap_atomic(page); |
1498 | tmp_count = map[offset]; | |
1499 | kunmap_atomic(map); | |
1500 | ||
1501 | count += (tmp_count & ~COUNT_CONTINUED) * n; | |
1502 | n *= (SWAP_CONT_MAX + 1); | |
1503 | } while (tmp_count & COUNT_CONTINUED); | |
1504 | out: | |
235b6217 | 1505 | unlock_cluster_or_swap_info(p, ci); |
8334b962 MK |
1506 | return count; |
1507 | } | |
1508 | ||
e0709829 HY |
1509 | static bool swap_page_trans_huge_swapped(struct swap_info_struct *si, |
1510 | swp_entry_t entry) | |
1511 | { | |
1512 | struct swap_cluster_info *ci; | |
1513 | unsigned char *map = si->swap_map; | |
1514 | unsigned long roffset = swp_offset(entry); | |
1515 | unsigned long offset = round_down(roffset, SWAPFILE_CLUSTER); | |
1516 | int i; | |
1517 | bool ret = false; | |
1518 | ||
1519 | ci = lock_cluster_or_swap_info(si, offset); | |
1520 | if (!ci || !cluster_is_huge(ci)) { | |
afa4711e | 1521 | if (swap_count(map[roffset])) |
e0709829 HY |
1522 | ret = true; |
1523 | goto unlock_out; | |
1524 | } | |
1525 | for (i = 0; i < SWAPFILE_CLUSTER; i++) { | |
afa4711e | 1526 | if (swap_count(map[offset + i])) { |
e0709829 HY |
1527 | ret = true; |
1528 | break; | |
1529 | } | |
1530 | } | |
1531 | unlock_out: | |
1532 | unlock_cluster_or_swap_info(si, ci); | |
1533 | return ret; | |
1534 | } | |
1535 | ||
2397f780 | 1536 | static bool folio_swapped(struct folio *folio) |
e0709829 | 1537 | { |
14d01ee9 MWO |
1538 | swp_entry_t entry = folio_swap_entry(folio); |
1539 | struct swap_info_struct *si = _swap_info_get(entry); | |
1540 | ||
1541 | if (!si) | |
1542 | return false; | |
e0709829 | 1543 | |
2397f780 | 1544 | if (!IS_ENABLED(CONFIG_THP_SWAP) || likely(!folio_test_large(folio))) |
14d01ee9 | 1545 | return swap_swapcount(si, entry) != 0; |
e0709829 | 1546 | |
14d01ee9 | 1547 | return swap_page_trans_huge_swapped(si, entry); |
e0709829 | 1548 | } |
ba3c4ce6 | 1549 | |
bdb0ed54 MWO |
1550 | /** |
1551 | * folio_free_swap() - Free the swap space used for this folio. | |
1552 | * @folio: The folio to remove. | |
1553 | * | |
1554 | * If swap is getting full, or if there are no more mappings of this folio, | |
1555 | * then call folio_free_swap to free its swap space. | |
1556 | * | |
1557 | * Return: true if we were able to release the swap space. | |
1da177e4 | 1558 | */ |
bdb0ed54 | 1559 | bool folio_free_swap(struct folio *folio) |
1da177e4 | 1560 | { |
2397f780 | 1561 | VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio); |
1da177e4 | 1562 | |
2397f780 | 1563 | if (!folio_test_swapcache(folio)) |
bdb0ed54 | 1564 | return false; |
2397f780 | 1565 | if (folio_test_writeback(folio)) |
bdb0ed54 | 1566 | return false; |
2397f780 | 1567 | if (folio_swapped(folio)) |
bdb0ed54 | 1568 | return false; |
1da177e4 | 1569 | |
b73d7fce HD |
1570 | /* |
1571 | * Once hibernation has begun to create its image of memory, | |
bdb0ed54 | 1572 | * there's a danger that one of the calls to folio_free_swap() |
b73d7fce HD |
1573 | * - most probably a call from __try_to_reclaim_swap() while |
1574 | * hibernation is allocating its own swap pages for the image, | |
1575 | * but conceivably even a call from memory reclaim - will free | |
bdb0ed54 MWO |
1576 | * the swap from a folio which has already been recorded in the |
1577 | * image as a clean swapcache folio, and then reuse its swap for | |
b73d7fce | 1578 | * another page of the image. On waking from hibernation, the |
bdb0ed54 | 1579 | * original folio might be freed under memory pressure, then |
b73d7fce HD |
1580 | * later read back in from swap, now with the wrong data. |
1581 | * | |
2de1a7e4 | 1582 | * Hibernation suspends storage while it is writing the image |
f90ac398 | 1583 | * to disk so check that here. |
b73d7fce | 1584 | */ |
f90ac398 | 1585 | if (pm_suspended_storage()) |
bdb0ed54 | 1586 | return false; |
b73d7fce | 1587 | |
75fa68a5 | 1588 | delete_from_swap_cache(folio); |
2397f780 | 1589 | folio_set_dirty(folio); |
bdb0ed54 | 1590 | return true; |
68a22394 RR |
1591 | } |
1592 | ||
1da177e4 LT |
1593 | /* |
1594 | * Free the swap entry like above, but also try to | |
1595 | * free the page cache entry if it is the last user. | |
1596 | */ | |
2509ef26 | 1597 | int free_swap_and_cache(swp_entry_t entry) |
1da177e4 | 1598 | { |
2509ef26 | 1599 | struct swap_info_struct *p; |
7c00bafe | 1600 | unsigned char count; |
1da177e4 | 1601 | |
a7420aa5 | 1602 | if (non_swap_entry(entry)) |
2509ef26 | 1603 | return 1; |
0697212a | 1604 | |
7c00bafe | 1605 | p = _swap_info_get(entry); |
1da177e4 | 1606 | if (p) { |
33e16272 | 1607 | count = __swap_entry_free(p, entry); |
e0709829 | 1608 | if (count == SWAP_HAS_CACHE && |
bcd49e86 HY |
1609 | !swap_page_trans_huge_swapped(p, entry)) |
1610 | __try_to_reclaim_swap(p, swp_offset(entry), | |
1611 | TTRS_UNMAPPED | TTRS_FULL); | |
1da177e4 | 1612 | } |
2509ef26 | 1613 | return p != NULL; |
1da177e4 LT |
1614 | } |
1615 | ||
b0cb1a19 | 1616 | #ifdef CONFIG_HIBERNATION |
bb243f7d ML |
1617 | |
1618 | swp_entry_t get_swap_page_of_type(int type) | |
1619 | { | |
1620 | struct swap_info_struct *si = swap_type_to_swap_info(type); | |
1621 | swp_entry_t entry = {0}; | |
1622 | ||
1623 | if (!si) | |
1624 | goto fail; | |
1625 | ||
1626 | /* This is called for allocating swap entry, not cache */ | |
1627 | spin_lock(&si->lock); | |
1628 | if ((si->flags & SWP_WRITEOK) && scan_swap_map_slots(si, 1, 1, &entry)) | |
1629 | atomic_long_dec(&nr_swap_pages); | |
1630 | spin_unlock(&si->lock); | |
1631 | fail: | |
1632 | return entry; | |
1633 | } | |
1634 | ||
f577eb30 | 1635 | /* |
915bae9e | 1636 | * Find the swap type that corresponds to given device (if any). |
f577eb30 | 1637 | * |
915bae9e RW |
1638 | * @offset - number of the PAGE_SIZE-sized block of the device, starting |
1639 | * from 0, in which the swap header is expected to be located. | |
1640 | * | |
1641 | * This is needed for the suspend to disk (aka swsusp). | |
f577eb30 | 1642 | */ |
21bd9005 | 1643 | int swap_type_of(dev_t device, sector_t offset) |
f577eb30 | 1644 | { |
efa90a98 | 1645 | int type; |
f577eb30 | 1646 | |
21bd9005 CH |
1647 | if (!device) |
1648 | return -1; | |
915bae9e | 1649 | |
f577eb30 | 1650 | spin_lock(&swap_lock); |
efa90a98 HD |
1651 | for (type = 0; type < nr_swapfiles; type++) { |
1652 | struct swap_info_struct *sis = swap_info[type]; | |
f577eb30 | 1653 | |
915bae9e | 1654 | if (!(sis->flags & SWP_WRITEOK)) |
f577eb30 | 1655 | continue; |
b6b5bce3 | 1656 | |
21bd9005 | 1657 | if (device == sis->bdev->bd_dev) { |
4efaceb1 | 1658 | struct swap_extent *se = first_se(sis); |
915bae9e | 1659 | |
915bae9e RW |
1660 | if (se->start_block == offset) { |
1661 | spin_unlock(&swap_lock); | |
efa90a98 | 1662 | return type; |
915bae9e | 1663 | } |
f577eb30 RW |
1664 | } |
1665 | } | |
1666 | spin_unlock(&swap_lock); | |
21bd9005 CH |
1667 | return -ENODEV; |
1668 | } | |
915bae9e | 1669 | |
21bd9005 CH |
1670 | int find_first_swap(dev_t *device) |
1671 | { | |
1672 | int type; | |
915bae9e | 1673 | |
21bd9005 CH |
1674 | spin_lock(&swap_lock); |
1675 | for (type = 0; type < nr_swapfiles; type++) { | |
1676 | struct swap_info_struct *sis = swap_info[type]; | |
1677 | ||
1678 | if (!(sis->flags & SWP_WRITEOK)) | |
1679 | continue; | |
1680 | *device = sis->bdev->bd_dev; | |
1681 | spin_unlock(&swap_lock); | |
1682 | return type; | |
1683 | } | |
1684 | spin_unlock(&swap_lock); | |
f577eb30 RW |
1685 | return -ENODEV; |
1686 | } | |
1687 | ||
73c34b6a HD |
1688 | /* |
1689 | * Get the (PAGE_SIZE) block corresponding to given offset on the swapdev | |
1690 | * corresponding to given index in swap_info (swap type). | |
1691 | */ | |
1692 | sector_t swapdev_block(int type, pgoff_t offset) | |
1693 | { | |
c10d38cc | 1694 | struct swap_info_struct *si = swap_type_to_swap_info(type); |
f885056a | 1695 | struct swap_extent *se; |
73c34b6a | 1696 | |
c10d38cc | 1697 | if (!si || !(si->flags & SWP_WRITEOK)) |
73c34b6a | 1698 | return 0; |
f885056a CH |
1699 | se = offset_to_swap_extent(si, offset); |
1700 | return se->start_block + (offset - se->start_page); | |
73c34b6a HD |
1701 | } |
1702 | ||
f577eb30 RW |
1703 | /* |
1704 | * Return either the total number of swap pages of given type, or the number | |
1705 | * of free pages of that type (depending on @free) | |
1706 | * | |
1707 | * This is needed for software suspend | |
1708 | */ | |
1709 | unsigned int count_swap_pages(int type, int free) | |
1710 | { | |
1711 | unsigned int n = 0; | |
1712 | ||
efa90a98 HD |
1713 | spin_lock(&swap_lock); |
1714 | if ((unsigned int)type < nr_swapfiles) { | |
1715 | struct swap_info_struct *sis = swap_info[type]; | |
1716 | ||
ec8acf20 | 1717 | spin_lock(&sis->lock); |
efa90a98 HD |
1718 | if (sis->flags & SWP_WRITEOK) { |
1719 | n = sis->pages; | |
f577eb30 | 1720 | if (free) |
efa90a98 | 1721 | n -= sis->inuse_pages; |
f577eb30 | 1722 | } |
ec8acf20 | 1723 | spin_unlock(&sis->lock); |
f577eb30 | 1724 | } |
efa90a98 | 1725 | spin_unlock(&swap_lock); |
f577eb30 RW |
1726 | return n; |
1727 | } | |
73c34b6a | 1728 | #endif /* CONFIG_HIBERNATION */ |
f577eb30 | 1729 | |
9f8bdb3f | 1730 | static inline int pte_same_as_swp(pte_t pte, pte_t swp_pte) |
179ef71c | 1731 | { |
099dd687 | 1732 | return pte_same(pte_swp_clear_flags(pte), swp_pte); |
179ef71c CG |
1733 | } |
1734 | ||
1da177e4 | 1735 | /* |
72866f6f HD |
1736 | * No need to decide whether this PTE shares the swap entry with others, |
1737 | * just let do_wp_page work it out if a write is requested later - to | |
1738 | * force COW, vm_page_prot omits write permission from any private vma. | |
1da177e4 | 1739 | */ |
044d66c1 | 1740 | static int unuse_pte(struct vm_area_struct *vma, pmd_t *pmd, |
f102cd8b | 1741 | unsigned long addr, swp_entry_t entry, struct folio *folio) |
1da177e4 | 1742 | { |
f102cd8b | 1743 | struct page *page = folio_file_page(folio, swp_offset(entry)); |
9e16b7fb | 1744 | struct page *swapcache; |
044d66c1 | 1745 | spinlock_t *ptl; |
c33c7948 | 1746 | pte_t *pte, new_pte, old_pte; |
6b970599 | 1747 | bool hwposioned = false; |
044d66c1 HD |
1748 | int ret = 1; |
1749 | ||
9e16b7fb HD |
1750 | swapcache = page; |
1751 | page = ksm_might_need_to_copy(page, vma, addr); | |
1752 | if (unlikely(!page)) | |
1753 | return -ENOMEM; | |
6b970599 KW |
1754 | else if (unlikely(PTR_ERR(page) == -EHWPOISON)) |
1755 | hwposioned = true; | |
9e16b7fb | 1756 | |
044d66c1 | 1757 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
c33c7948 RR |
1758 | if (unlikely(!pte || !pte_same_as_swp(ptep_get(pte), |
1759 | swp_entry_to_pte(entry)))) { | |
044d66c1 HD |
1760 | ret = 0; |
1761 | goto out; | |
1762 | } | |
8a9f3ccd | 1763 | |
c33c7948 RR |
1764 | old_pte = ptep_get(pte); |
1765 | ||
6b970599 KW |
1766 | if (unlikely(hwposioned || !PageUptodate(page))) { |
1767 | swp_entry_t swp_entry; | |
9f186f9e ML |
1768 | |
1769 | dec_mm_counter(vma->vm_mm, MM_SWAPENTS); | |
6b970599 KW |
1770 | if (hwposioned) { |
1771 | swp_entry = make_hwpoison_entry(swapcache); | |
1772 | page = swapcache; | |
1773 | } else { | |
af19487f | 1774 | swp_entry = make_poisoned_swp_entry(); |
6b970599 KW |
1775 | } |
1776 | new_pte = swp_entry_to_pte(swp_entry); | |
9f186f9e | 1777 | ret = 0; |
6b970599 | 1778 | goto setpte; |
9f186f9e ML |
1779 | } |
1780 | ||
b53e24c4 PC |
1781 | /* |
1782 | * Some architectures may have to restore extra metadata to the page | |
1783 | * when reading from swap. This metadata may be indexed by swap entry | |
1784 | * so this must be called before swap_free(). | |
1785 | */ | |
1786 | arch_swap_restore(entry, page_folio(page)); | |
1787 | ||
78fbe906 DH |
1788 | /* See do_swap_page() */ |
1789 | BUG_ON(!PageAnon(page) && PageMappedToDisk(page)); | |
1790 | BUG_ON(PageAnon(page) && PageAnonExclusive(page)); | |
1791 | ||
b084d435 | 1792 | dec_mm_counter(vma->vm_mm, MM_SWAPENTS); |
d559db08 | 1793 | inc_mm_counter(vma->vm_mm, MM_ANONPAGES); |
1da177e4 | 1794 | get_page(page); |
00501b53 | 1795 | if (page == swapcache) { |
1493a191 DH |
1796 | rmap_t rmap_flags = RMAP_NONE; |
1797 | ||
1798 | /* | |
1799 | * See do_swap_page(): PageWriteback() would be problematic. | |
1800 | * However, we do a wait_on_page_writeback() just before this | |
1801 | * call and have the page locked. | |
1802 | */ | |
1803 | VM_BUG_ON_PAGE(PageWriteback(page), page); | |
c33c7948 | 1804 | if (pte_swp_exclusive(old_pte)) |
1493a191 DH |
1805 | rmap_flags |= RMAP_EXCLUSIVE; |
1806 | ||
1807 | page_add_anon_rmap(page, vma, addr, rmap_flags); | |
00501b53 | 1808 | } else { /* ksm created a completely new copy */ |
40f2bbf7 | 1809 | page_add_new_anon_rmap(page, vma, addr); |
b518154e | 1810 | lru_cache_add_inactive_or_unevictable(page, vma); |
00501b53 | 1811 | } |
14a762dd | 1812 | new_pte = pte_mkold(mk_pte(page, vma->vm_page_prot)); |
c33c7948 | 1813 | if (pte_swp_soft_dirty(old_pte)) |
14a762dd | 1814 | new_pte = pte_mksoft_dirty(new_pte); |
c33c7948 | 1815 | if (pte_swp_uffd_wp(old_pte)) |
14a762dd | 1816 | new_pte = pte_mkuffd_wp(new_pte); |
6b970599 | 1817 | setpte: |
14a762dd | 1818 | set_pte_at(vma->vm_mm, addr, pte, new_pte); |
1da177e4 | 1819 | swap_free(entry); |
044d66c1 | 1820 | out: |
d850fa72 HD |
1821 | if (pte) |
1822 | pte_unmap_unlock(pte, ptl); | |
9e16b7fb HD |
1823 | if (page != swapcache) { |
1824 | unlock_page(page); | |
1825 | put_page(page); | |
1826 | } | |
044d66c1 | 1827 | return ret; |
1da177e4 LT |
1828 | } |
1829 | ||
1830 | static int unuse_pte_range(struct vm_area_struct *vma, pmd_t *pmd, | |
b56a2d8a | 1831 | unsigned long addr, unsigned long end, |
10a9c496 | 1832 | unsigned int type) |
1da177e4 | 1833 | { |
d850fa72 | 1834 | pte_t *pte = NULL; |
b56a2d8a | 1835 | struct swap_info_struct *si; |
1da177e4 | 1836 | |
b56a2d8a | 1837 | si = swap_info[type]; |
1da177e4 | 1838 | do { |
f102cd8b MWO |
1839 | struct folio *folio; |
1840 | unsigned long offset; | |
3f79b187 | 1841 | unsigned char swp_count; |
d850fa72 HD |
1842 | swp_entry_t entry; |
1843 | int ret; | |
c33c7948 | 1844 | pte_t ptent; |
d850fa72 HD |
1845 | |
1846 | if (!pte++) { | |
1847 | pte = pte_offset_map(pmd, addr); | |
1848 | if (!pte) | |
1849 | break; | |
1850 | } | |
f102cd8b | 1851 | |
c33c7948 | 1852 | ptent = ptep_get_lockless(pte); |
f102cd8b | 1853 | |
c33c7948 | 1854 | if (!is_swap_pte(ptent)) |
b56a2d8a VRP |
1855 | continue; |
1856 | ||
c33c7948 | 1857 | entry = pte_to_swp_entry(ptent); |
b56a2d8a VRP |
1858 | if (swp_type(entry) != type) |
1859 | continue; | |
1860 | ||
1861 | offset = swp_offset(entry); | |
b56a2d8a | 1862 | pte_unmap(pte); |
d850fa72 HD |
1863 | pte = NULL; |
1864 | ||
f102cd8b MWO |
1865 | folio = swap_cache_get_folio(entry, vma, addr); |
1866 | if (!folio) { | |
1867 | struct page *page; | |
8c63ca5b WD |
1868 | struct vm_fault vmf = { |
1869 | .vma = vma, | |
1870 | .address = addr, | |
824ddc60 | 1871 | .real_address = addr, |
8c63ca5b WD |
1872 | .pmd = pmd, |
1873 | }; | |
1874 | ||
ebc5951e AR |
1875 | page = swapin_readahead(entry, GFP_HIGHUSER_MOVABLE, |
1876 | &vmf); | |
f102cd8b MWO |
1877 | if (page) |
1878 | folio = page_folio(page); | |
ebc5951e | 1879 | } |
f102cd8b | 1880 | if (!folio) { |
3f79b187 KS |
1881 | swp_count = READ_ONCE(si->swap_map[offset]); |
1882 | if (swp_count == 0 || swp_count == SWAP_MAP_BAD) | |
d850fa72 | 1883 | continue; |
b56a2d8a VRP |
1884 | return -ENOMEM; |
1885 | } | |
1886 | ||
f102cd8b MWO |
1887 | folio_lock(folio); |
1888 | folio_wait_writeback(folio); | |
1889 | ret = unuse_pte(vma, pmd, addr, entry, folio); | |
b56a2d8a | 1890 | if (ret < 0) { |
f102cd8b MWO |
1891 | folio_unlock(folio); |
1892 | folio_put(folio); | |
d850fa72 | 1893 | return ret; |
b56a2d8a VRP |
1894 | } |
1895 | ||
f102cd8b MWO |
1896 | folio_free_swap(folio); |
1897 | folio_unlock(folio); | |
1898 | folio_put(folio); | |
d850fa72 | 1899 | } while (addr += PAGE_SIZE, addr != end); |
b56a2d8a | 1900 | |
d850fa72 HD |
1901 | if (pte) |
1902 | pte_unmap(pte); | |
1903 | return 0; | |
1da177e4 LT |
1904 | } |
1905 | ||
1906 | static inline int unuse_pmd_range(struct vm_area_struct *vma, pud_t *pud, | |
1907 | unsigned long addr, unsigned long end, | |
10a9c496 | 1908 | unsigned int type) |
1da177e4 LT |
1909 | { |
1910 | pmd_t *pmd; | |
1911 | unsigned long next; | |
8a9f3ccd | 1912 | int ret; |
1da177e4 LT |
1913 | |
1914 | pmd = pmd_offset(pud, addr); | |
1915 | do { | |
dc644a07 | 1916 | cond_resched(); |
1da177e4 | 1917 | next = pmd_addr_end(addr, end); |
10a9c496 | 1918 | ret = unuse_pte_range(vma, pmd, addr, next, type); |
8a9f3ccd BS |
1919 | if (ret) |
1920 | return ret; | |
1da177e4 LT |
1921 | } while (pmd++, addr = next, addr != end); |
1922 | return 0; | |
1923 | } | |
1924 | ||
c2febafc | 1925 | static inline int unuse_pud_range(struct vm_area_struct *vma, p4d_t *p4d, |
1da177e4 | 1926 | unsigned long addr, unsigned long end, |
10a9c496 | 1927 | unsigned int type) |
1da177e4 LT |
1928 | { |
1929 | pud_t *pud; | |
1930 | unsigned long next; | |
8a9f3ccd | 1931 | int ret; |
1da177e4 | 1932 | |
c2febafc | 1933 | pud = pud_offset(p4d, addr); |
1da177e4 LT |
1934 | do { |
1935 | next = pud_addr_end(addr, end); | |
1936 | if (pud_none_or_clear_bad(pud)) | |
1937 | continue; | |
10a9c496 | 1938 | ret = unuse_pmd_range(vma, pud, addr, next, type); |
8a9f3ccd BS |
1939 | if (ret) |
1940 | return ret; | |
1da177e4 LT |
1941 | } while (pud++, addr = next, addr != end); |
1942 | return 0; | |
1943 | } | |
1944 | ||
c2febafc KS |
1945 | static inline int unuse_p4d_range(struct vm_area_struct *vma, pgd_t *pgd, |
1946 | unsigned long addr, unsigned long end, | |
10a9c496 | 1947 | unsigned int type) |
c2febafc KS |
1948 | { |
1949 | p4d_t *p4d; | |
1950 | unsigned long next; | |
1951 | int ret; | |
1952 | ||
1953 | p4d = p4d_offset(pgd, addr); | |
1954 | do { | |
1955 | next = p4d_addr_end(addr, end); | |
1956 | if (p4d_none_or_clear_bad(p4d)) | |
1957 | continue; | |
10a9c496 | 1958 | ret = unuse_pud_range(vma, p4d, addr, next, type); |
c2febafc KS |
1959 | if (ret) |
1960 | return ret; | |
1961 | } while (p4d++, addr = next, addr != end); | |
1962 | return 0; | |
1963 | } | |
1964 | ||
10a9c496 | 1965 | static int unuse_vma(struct vm_area_struct *vma, unsigned int type) |
1da177e4 LT |
1966 | { |
1967 | pgd_t *pgd; | |
1968 | unsigned long addr, end, next; | |
8a9f3ccd | 1969 | int ret; |
1da177e4 | 1970 | |
b56a2d8a VRP |
1971 | addr = vma->vm_start; |
1972 | end = vma->vm_end; | |
1da177e4 LT |
1973 | |
1974 | pgd = pgd_offset(vma->vm_mm, addr); | |
1975 | do { | |
1976 | next = pgd_addr_end(addr, end); | |
1977 | if (pgd_none_or_clear_bad(pgd)) | |
1978 | continue; | |
10a9c496 | 1979 | ret = unuse_p4d_range(vma, pgd, addr, next, type); |
8a9f3ccd BS |
1980 | if (ret) |
1981 | return ret; | |
1da177e4 LT |
1982 | } while (pgd++, addr = next, addr != end); |
1983 | return 0; | |
1984 | } | |
1985 | ||
10a9c496 | 1986 | static int unuse_mm(struct mm_struct *mm, unsigned int type) |
1da177e4 LT |
1987 | { |
1988 | struct vm_area_struct *vma; | |
8a9f3ccd | 1989 | int ret = 0; |
208c09db | 1990 | VMA_ITERATOR(vmi, mm, 0); |
1da177e4 | 1991 | |
d8ed45c5 | 1992 | mmap_read_lock(mm); |
208c09db | 1993 | for_each_vma(vmi, vma) { |
b56a2d8a | 1994 | if (vma->anon_vma) { |
10a9c496 | 1995 | ret = unuse_vma(vma, type); |
b56a2d8a VRP |
1996 | if (ret) |
1997 | break; | |
1998 | } | |
208c09db | 1999 | |
dc644a07 | 2000 | cond_resched(); |
1da177e4 | 2001 | } |
d8ed45c5 | 2002 | mmap_read_unlock(mm); |
b56a2d8a | 2003 | return ret; |
1da177e4 LT |
2004 | } |
2005 | ||
2006 | /* | |
3c3115ad ML |
2007 | * Scan swap_map from current position to next entry still in use. |
2008 | * Return 0 if there are no inuse entries after prev till end of | |
2009 | * the map. | |
1da177e4 | 2010 | */ |
6eb396dc | 2011 | static unsigned int find_next_to_unuse(struct swap_info_struct *si, |
10a9c496 | 2012 | unsigned int prev) |
1da177e4 | 2013 | { |
b56a2d8a | 2014 | unsigned int i; |
8d69aaee | 2015 | unsigned char count; |
1da177e4 LT |
2016 | |
2017 | /* | |
5d337b91 | 2018 | * No need for swap_lock here: we're just looking |
1da177e4 LT |
2019 | * for whether an entry is in use, not modifying it; false |
2020 | * hits are okay, and sys_swapoff() has already prevented new | |
5d337b91 | 2021 | * allocations from this area (while holding swap_lock). |
1da177e4 | 2022 | */ |
b56a2d8a | 2023 | for (i = prev + 1; i < si->max; i++) { |
4db0c3c2 | 2024 | count = READ_ONCE(si->swap_map[i]); |
355cfa73 | 2025 | if (count && swap_count(count) != SWAP_MAP_BAD) |
10a9c496 | 2026 | break; |
dc644a07 HD |
2027 | if ((i % LATENCY_LIMIT) == 0) |
2028 | cond_resched(); | |
1da177e4 | 2029 | } |
b56a2d8a VRP |
2030 | |
2031 | if (i == si->max) | |
2032 | i = 0; | |
2033 | ||
1da177e4 LT |
2034 | return i; |
2035 | } | |
2036 | ||
10a9c496 | 2037 | static int try_to_unuse(unsigned int type) |
1da177e4 | 2038 | { |
b56a2d8a VRP |
2039 | struct mm_struct *prev_mm; |
2040 | struct mm_struct *mm; | |
2041 | struct list_head *p; | |
2042 | int retval = 0; | |
efa90a98 | 2043 | struct swap_info_struct *si = swap_info[type]; |
000085b9 | 2044 | struct folio *folio; |
1da177e4 | 2045 | swp_entry_t entry; |
b56a2d8a | 2046 | unsigned int i; |
1da177e4 | 2047 | |
21820948 | 2048 | if (!READ_ONCE(si->inuse_pages)) |
b56a2d8a | 2049 | return 0; |
1da177e4 | 2050 | |
b56a2d8a | 2051 | retry: |
10a9c496 | 2052 | retval = shmem_unuse(type); |
b56a2d8a | 2053 | if (retval) |
10a9c496 | 2054 | return retval; |
b56a2d8a VRP |
2055 | |
2056 | prev_mm = &init_mm; | |
2057 | mmget(prev_mm); | |
2058 | ||
2059 | spin_lock(&mmlist_lock); | |
2060 | p = &init_mm.mmlist; | |
21820948 | 2061 | while (READ_ONCE(si->inuse_pages) && |
64165b1a HD |
2062 | !signal_pending(current) && |
2063 | (p = p->next) != &init_mm.mmlist) { | |
1da177e4 | 2064 | |
b56a2d8a VRP |
2065 | mm = list_entry(p, struct mm_struct, mmlist); |
2066 | if (!mmget_not_zero(mm)) | |
2067 | continue; | |
2068 | spin_unlock(&mmlist_lock); | |
2069 | mmput(prev_mm); | |
2070 | prev_mm = mm; | |
10a9c496 | 2071 | retval = unuse_mm(mm, type); |
b56a2d8a VRP |
2072 | if (retval) { |
2073 | mmput(prev_mm); | |
10a9c496 | 2074 | return retval; |
1da177e4 LT |
2075 | } |
2076 | ||
2077 | /* | |
b56a2d8a VRP |
2078 | * Make sure that we aren't completely killing |
2079 | * interactive performance. | |
1da177e4 | 2080 | */ |
b56a2d8a VRP |
2081 | cond_resched(); |
2082 | spin_lock(&mmlist_lock); | |
2083 | } | |
2084 | spin_unlock(&mmlist_lock); | |
1da177e4 | 2085 | |
b56a2d8a | 2086 | mmput(prev_mm); |
1da177e4 | 2087 | |
b56a2d8a | 2088 | i = 0; |
21820948 | 2089 | while (READ_ONCE(si->inuse_pages) && |
64165b1a | 2090 | !signal_pending(current) && |
10a9c496 | 2091 | (i = find_next_to_unuse(si, i)) != 0) { |
1da177e4 | 2092 | |
b56a2d8a | 2093 | entry = swp_entry(type, i); |
000085b9 | 2094 | folio = filemap_get_folio(swap_address_space(entry), i); |
66dabbb6 | 2095 | if (IS_ERR(folio)) |
b56a2d8a | 2096 | continue; |
68bdc8d6 HD |
2097 | |
2098 | /* | |
000085b9 MWO |
2099 | * It is conceivable that a racing task removed this folio from |
2100 | * swap cache just before we acquired the page lock. The folio | |
b56a2d8a | 2101 | * might even be back in swap cache on another swap area. But |
000085b9 | 2102 | * that is okay, folio_free_swap() only removes stale folios. |
1da177e4 | 2103 | */ |
000085b9 MWO |
2104 | folio_lock(folio); |
2105 | folio_wait_writeback(folio); | |
2106 | folio_free_swap(folio); | |
2107 | folio_unlock(folio); | |
2108 | folio_put(folio); | |
1da177e4 LT |
2109 | } |
2110 | ||
b56a2d8a VRP |
2111 | /* |
2112 | * Lets check again to see if there are still swap entries in the map. | |
2113 | * If yes, we would need to do retry the unuse logic again. | |
2114 | * Under global memory pressure, swap entries can be reinserted back | |
2115 | * into process space after the mmlist loop above passes over them. | |
dd862deb | 2116 | * |
e2e3fdc7 MWO |
2117 | * Limit the number of retries? No: when mmget_not_zero() |
2118 | * above fails, that mm is likely to be freeing swap from | |
2119 | * exit_mmap(), which proceeds at its own independent pace; | |
2120 | * and even shmem_writepage() could have been preempted after | |
2121 | * folio_alloc_swap(), temporarily hiding that swap. It's easy | |
2122 | * and robust (though cpu-intensive) just to keep retrying. | |
b56a2d8a | 2123 | */ |
21820948 | 2124 | if (READ_ONCE(si->inuse_pages)) { |
64165b1a HD |
2125 | if (!signal_pending(current)) |
2126 | goto retry; | |
10a9c496 | 2127 | return -EINTR; |
64165b1a | 2128 | } |
10a9c496 CH |
2129 | |
2130 | return 0; | |
1da177e4 LT |
2131 | } |
2132 | ||
2133 | /* | |
5d337b91 HD |
2134 | * After a successful try_to_unuse, if no swap is now in use, we know |
2135 | * we can empty the mmlist. swap_lock must be held on entry and exit. | |
2136 | * Note that mmlist_lock nests inside swap_lock, and an mm must be | |
1da177e4 LT |
2137 | * added to the mmlist just after page_duplicate - before would be racy. |
2138 | */ | |
2139 | static void drain_mmlist(void) | |
2140 | { | |
2141 | struct list_head *p, *next; | |
efa90a98 | 2142 | unsigned int type; |
1da177e4 | 2143 | |
efa90a98 HD |
2144 | for (type = 0; type < nr_swapfiles; type++) |
2145 | if (swap_info[type]->inuse_pages) | |
1da177e4 LT |
2146 | return; |
2147 | spin_lock(&mmlist_lock); | |
2148 | list_for_each_safe(p, next, &init_mm.mmlist) | |
2149 | list_del_init(p); | |
2150 | spin_unlock(&mmlist_lock); | |
2151 | } | |
2152 | ||
1da177e4 LT |
2153 | /* |
2154 | * Free all of a swapdev's extent information | |
2155 | */ | |
2156 | static void destroy_swap_extents(struct swap_info_struct *sis) | |
2157 | { | |
4efaceb1 AL |
2158 | while (!RB_EMPTY_ROOT(&sis->swap_extent_root)) { |
2159 | struct rb_node *rb = sis->swap_extent_root.rb_node; | |
2160 | struct swap_extent *se = rb_entry(rb, struct swap_extent, rb_node); | |
1da177e4 | 2161 | |
4efaceb1 | 2162 | rb_erase(rb, &sis->swap_extent_root); |
1da177e4 LT |
2163 | kfree(se); |
2164 | } | |
62c230bc | 2165 | |
bc4ae27d | 2166 | if (sis->flags & SWP_ACTIVATED) { |
62c230bc MG |
2167 | struct file *swap_file = sis->swap_file; |
2168 | struct address_space *mapping = swap_file->f_mapping; | |
2169 | ||
bc4ae27d OS |
2170 | sis->flags &= ~SWP_ACTIVATED; |
2171 | if (mapping->a_ops->swap_deactivate) | |
2172 | mapping->a_ops->swap_deactivate(swap_file); | |
62c230bc | 2173 | } |
1da177e4 LT |
2174 | } |
2175 | ||
2176 | /* | |
2177 | * Add a block range (and the corresponding page range) into this swapdev's | |
4efaceb1 | 2178 | * extent tree. |
1da177e4 | 2179 | * |
11d31886 | 2180 | * This function rather assumes that it is called in ascending page order. |
1da177e4 | 2181 | */ |
a509bc1a | 2182 | int |
1da177e4 LT |
2183 | add_swap_extent(struct swap_info_struct *sis, unsigned long start_page, |
2184 | unsigned long nr_pages, sector_t start_block) | |
2185 | { | |
4efaceb1 | 2186 | struct rb_node **link = &sis->swap_extent_root.rb_node, *parent = NULL; |
1da177e4 LT |
2187 | struct swap_extent *se; |
2188 | struct swap_extent *new_se; | |
4efaceb1 AL |
2189 | |
2190 | /* | |
2191 | * place the new node at the right most since the | |
2192 | * function is called in ascending page order. | |
2193 | */ | |
2194 | while (*link) { | |
2195 | parent = *link; | |
2196 | link = &parent->rb_right; | |
2197 | } | |
2198 | ||
2199 | if (parent) { | |
2200 | se = rb_entry(parent, struct swap_extent, rb_node); | |
11d31886 HD |
2201 | BUG_ON(se->start_page + se->nr_pages != start_page); |
2202 | if (se->start_block + se->nr_pages == start_block) { | |
1da177e4 LT |
2203 | /* Merge it */ |
2204 | se->nr_pages += nr_pages; | |
2205 | return 0; | |
2206 | } | |
1da177e4 LT |
2207 | } |
2208 | ||
4efaceb1 | 2209 | /* No merge, insert a new extent. */ |
1da177e4 LT |
2210 | new_se = kmalloc(sizeof(*se), GFP_KERNEL); |
2211 | if (new_se == NULL) | |
2212 | return -ENOMEM; | |
2213 | new_se->start_page = start_page; | |
2214 | new_se->nr_pages = nr_pages; | |
2215 | new_se->start_block = start_block; | |
2216 | ||
4efaceb1 AL |
2217 | rb_link_node(&new_se->rb_node, parent, link); |
2218 | rb_insert_color(&new_se->rb_node, &sis->swap_extent_root); | |
53092a74 | 2219 | return 1; |
1da177e4 | 2220 | } |
aa8aa8a3 | 2221 | EXPORT_SYMBOL_GPL(add_swap_extent); |
1da177e4 LT |
2222 | |
2223 | /* | |
2224 | * A `swap extent' is a simple thing which maps a contiguous range of pages | |
ff351f4b ML |
2225 | * onto a contiguous range of disk blocks. A rbtree of swap extents is |
2226 | * built at swapon time and is then used at swap_writepage/swap_readpage | |
1da177e4 LT |
2227 | * time for locating where on disk a page belongs. |
2228 | * | |
2229 | * If the swapfile is an S_ISBLK block device, a single extent is installed. | |
2230 | * This is done so that the main operating code can treat S_ISBLK and S_ISREG | |
2231 | * swap files identically. | |
2232 | * | |
2233 | * Whether the swapdev is an S_ISREG file or an S_ISBLK blockdev, the swap | |
ff351f4b | 2234 | * extent rbtree operates in PAGE_SIZE disk blocks. Both S_ISREG and S_ISBLK |
1da177e4 LT |
2235 | * swapfiles are handled *identically* after swapon time. |
2236 | * | |
2237 | * For S_ISREG swapfiles, setup_swap_extents() will walk all the file's blocks | |
ff351f4b ML |
2238 | * and will parse them into a rbtree, in PAGE_SIZE chunks. If some stray |
2239 | * blocks are found which do not fall within the PAGE_SIZE alignment | |
1da177e4 LT |
2240 | * requirements, they are simply tossed out - we will never use those blocks |
2241 | * for swapping. | |
2242 | * | |
1638045c DW |
2243 | * For all swap devices we set S_SWAPFILE across the life of the swapon. This |
2244 | * prevents users from writing to the swap device, which will corrupt memory. | |
1da177e4 LT |
2245 | * |
2246 | * The amount of disk space which a single swap extent represents varies. | |
2247 | * Typically it is in the 1-4 megabyte range. So we can have hundreds of | |
ff351f4b | 2248 | * extents in the rbtree. - akpm. |
1da177e4 | 2249 | */ |
53092a74 | 2250 | static int setup_swap_extents(struct swap_info_struct *sis, sector_t *span) |
1da177e4 | 2251 | { |
62c230bc MG |
2252 | struct file *swap_file = sis->swap_file; |
2253 | struct address_space *mapping = swap_file->f_mapping; | |
2254 | struct inode *inode = mapping->host; | |
1da177e4 LT |
2255 | int ret; |
2256 | ||
1da177e4 LT |
2257 | if (S_ISBLK(inode->i_mode)) { |
2258 | ret = add_swap_extent(sis, 0, sis->max, 0); | |
53092a74 | 2259 | *span = sis->pages; |
a509bc1a | 2260 | return ret; |
1da177e4 LT |
2261 | } |
2262 | ||
62c230bc | 2263 | if (mapping->a_ops->swap_activate) { |
a509bc1a | 2264 | ret = mapping->a_ops->swap_activate(sis, swap_file, span); |
4b60c0ff N |
2265 | if (ret < 0) |
2266 | return ret; | |
2267 | sis->flags |= SWP_ACTIVATED; | |
e1209d3a N |
2268 | if ((sis->flags & SWP_FS_OPS) && |
2269 | sio_pool_init() != 0) { | |
2270 | destroy_swap_extents(sis); | |
2271 | return -ENOMEM; | |
62c230bc | 2272 | } |
a509bc1a | 2273 | return ret; |
62c230bc MG |
2274 | } |
2275 | ||
a509bc1a | 2276 | return generic_swapfile_activate(sis, swap_file, span); |
1da177e4 LT |
2277 | } |
2278 | ||
a2468cc9 AL |
2279 | static int swap_node(struct swap_info_struct *p) |
2280 | { | |
2281 | struct block_device *bdev; | |
2282 | ||
2283 | if (p->bdev) | |
2284 | bdev = p->bdev; | |
2285 | else | |
2286 | bdev = p->swap_file->f_inode->i_sb->s_bdev; | |
2287 | ||
2288 | return bdev ? bdev->bd_disk->node_id : NUMA_NO_NODE; | |
2289 | } | |
2290 | ||
eb085574 HY |
2291 | static void setup_swap_info(struct swap_info_struct *p, int prio, |
2292 | unsigned char *swap_map, | |
2293 | struct swap_cluster_info *cluster_info) | |
40531542 | 2294 | { |
a2468cc9 AL |
2295 | int i; |
2296 | ||
40531542 CEB |
2297 | if (prio >= 0) |
2298 | p->prio = prio; | |
2299 | else | |
2300 | p->prio = --least_priority; | |
18ab4d4c DS |
2301 | /* |
2302 | * the plist prio is negated because plist ordering is | |
2303 | * low-to-high, while swap ordering is high-to-low | |
2304 | */ | |
2305 | p->list.prio = -p->prio; | |
a2468cc9 AL |
2306 | for_each_node(i) { |
2307 | if (p->prio >= 0) | |
2308 | p->avail_lists[i].prio = -p->prio; | |
2309 | else { | |
2310 | if (swap_node(p) == i) | |
2311 | p->avail_lists[i].prio = 1; | |
2312 | else | |
2313 | p->avail_lists[i].prio = -p->prio; | |
2314 | } | |
2315 | } | |
40531542 | 2316 | p->swap_map = swap_map; |
2a8f9449 | 2317 | p->cluster_info = cluster_info; |
eb085574 HY |
2318 | } |
2319 | ||
2320 | static void _enable_swap_info(struct swap_info_struct *p) | |
2321 | { | |
63d8620e | 2322 | p->flags |= SWP_WRITEOK; |
ec8acf20 | 2323 | atomic_long_add(p->pages, &nr_swap_pages); |
40531542 CEB |
2324 | total_swap_pages += p->pages; |
2325 | ||
adfab836 | 2326 | assert_spin_locked(&swap_lock); |
adfab836 | 2327 | /* |
18ab4d4c DS |
2328 | * both lists are plists, and thus priority ordered. |
2329 | * swap_active_head needs to be priority ordered for swapoff(), | |
2330 | * which on removal of any swap_info_struct with an auto-assigned | |
2331 | * (i.e. negative) priority increments the auto-assigned priority | |
2332 | * of any lower-priority swap_info_structs. | |
e2e3fdc7 | 2333 | * swap_avail_head needs to be priority ordered for folio_alloc_swap(), |
18ab4d4c DS |
2334 | * which allocates swap pages from the highest available priority |
2335 | * swap_info_struct. | |
adfab836 | 2336 | */ |
18ab4d4c | 2337 | plist_add(&p->list, &swap_active_head); |
c70699e5 MW |
2338 | |
2339 | /* add to available list iff swap device is not full */ | |
2340 | if (p->highest_bit) | |
2341 | add_to_avail_list(p); | |
cf0cac0a CEB |
2342 | } |
2343 | ||
2344 | static void enable_swap_info(struct swap_info_struct *p, int prio, | |
2345 | unsigned char *swap_map, | |
2a8f9449 | 2346 | struct swap_cluster_info *cluster_info, |
cf0cac0a CEB |
2347 | unsigned long *frontswap_map) |
2348 | { | |
1cf53c89 CH |
2349 | if (IS_ENABLED(CONFIG_FRONTSWAP)) |
2350 | frontswap_init(p->type, frontswap_map); | |
cf0cac0a | 2351 | spin_lock(&swap_lock); |
ec8acf20 | 2352 | spin_lock(&p->lock); |
eb085574 HY |
2353 | setup_swap_info(p, prio, swap_map, cluster_info); |
2354 | spin_unlock(&p->lock); | |
2355 | spin_unlock(&swap_lock); | |
2356 | /* | |
63d8620e | 2357 | * Finished initializing swap device, now it's safe to reference it. |
eb085574 | 2358 | */ |
63d8620e | 2359 | percpu_ref_resurrect(&p->users); |
eb085574 HY |
2360 | spin_lock(&swap_lock); |
2361 | spin_lock(&p->lock); | |
2362 | _enable_swap_info(p); | |
ec8acf20 | 2363 | spin_unlock(&p->lock); |
cf0cac0a CEB |
2364 | spin_unlock(&swap_lock); |
2365 | } | |
2366 | ||
2367 | static void reinsert_swap_info(struct swap_info_struct *p) | |
2368 | { | |
2369 | spin_lock(&swap_lock); | |
ec8acf20 | 2370 | spin_lock(&p->lock); |
eb085574 HY |
2371 | setup_swap_info(p, p->prio, p->swap_map, p->cluster_info); |
2372 | _enable_swap_info(p); | |
ec8acf20 | 2373 | spin_unlock(&p->lock); |
40531542 CEB |
2374 | spin_unlock(&swap_lock); |
2375 | } | |
2376 | ||
67afa38e TC |
2377 | bool has_usable_swap(void) |
2378 | { | |
2379 | bool ret = true; | |
2380 | ||
2381 | spin_lock(&swap_lock); | |
2382 | if (plist_head_empty(&swap_active_head)) | |
2383 | ret = false; | |
2384 | spin_unlock(&swap_lock); | |
2385 | return ret; | |
2386 | } | |
2387 | ||
c4ea37c2 | 2388 | SYSCALL_DEFINE1(swapoff, const char __user *, specialfile) |
1da177e4 | 2389 | { |
73c34b6a | 2390 | struct swap_info_struct *p = NULL; |
8d69aaee | 2391 | unsigned char *swap_map; |
2a8f9449 | 2392 | struct swap_cluster_info *cluster_info; |
4f89849d | 2393 | unsigned long *frontswap_map; |
1da177e4 LT |
2394 | struct file *swap_file, *victim; |
2395 | struct address_space *mapping; | |
2396 | struct inode *inode; | |
91a27b2a | 2397 | struct filename *pathname; |
adfab836 | 2398 | int err, found = 0; |
5b808a23 | 2399 | unsigned int old_block_size; |
886bb7e9 | 2400 | |
1da177e4 LT |
2401 | if (!capable(CAP_SYS_ADMIN)) |
2402 | return -EPERM; | |
2403 | ||
191c5424 AV |
2404 | BUG_ON(!current->mm); |
2405 | ||
1da177e4 | 2406 | pathname = getname(specialfile); |
1da177e4 | 2407 | if (IS_ERR(pathname)) |
f58b59c1 | 2408 | return PTR_ERR(pathname); |
1da177e4 | 2409 | |
669abf4e | 2410 | victim = file_open_name(pathname, O_RDWR|O_LARGEFILE, 0); |
1da177e4 LT |
2411 | err = PTR_ERR(victim); |
2412 | if (IS_ERR(victim)) | |
2413 | goto out; | |
2414 | ||
2415 | mapping = victim->f_mapping; | |
5d337b91 | 2416 | spin_lock(&swap_lock); |
18ab4d4c | 2417 | plist_for_each_entry(p, &swap_active_head, list) { |
22c6f8fd | 2418 | if (p->flags & SWP_WRITEOK) { |
adfab836 DS |
2419 | if (p->swap_file->f_mapping == mapping) { |
2420 | found = 1; | |
1da177e4 | 2421 | break; |
adfab836 | 2422 | } |
1da177e4 | 2423 | } |
1da177e4 | 2424 | } |
adfab836 | 2425 | if (!found) { |
1da177e4 | 2426 | err = -EINVAL; |
5d337b91 | 2427 | spin_unlock(&swap_lock); |
1da177e4 LT |
2428 | goto out_dput; |
2429 | } | |
191c5424 | 2430 | if (!security_vm_enough_memory_mm(current->mm, p->pages)) |
1da177e4 LT |
2431 | vm_unacct_memory(p->pages); |
2432 | else { | |
2433 | err = -ENOMEM; | |
5d337b91 | 2434 | spin_unlock(&swap_lock); |
1da177e4 LT |
2435 | goto out_dput; |
2436 | } | |
ec8acf20 | 2437 | spin_lock(&p->lock); |
6fe7d6b9 | 2438 | del_from_avail_list(p); |
78ecba08 | 2439 | if (p->prio < 0) { |
adfab836 | 2440 | struct swap_info_struct *si = p; |
a2468cc9 | 2441 | int nid; |
adfab836 | 2442 | |
18ab4d4c | 2443 | plist_for_each_entry_continue(si, &swap_active_head, list) { |
adfab836 | 2444 | si->prio++; |
18ab4d4c | 2445 | si->list.prio--; |
a2468cc9 AL |
2446 | for_each_node(nid) { |
2447 | if (si->avail_lists[nid].prio != 1) | |
2448 | si->avail_lists[nid].prio--; | |
2449 | } | |
adfab836 | 2450 | } |
78ecba08 HD |
2451 | least_priority++; |
2452 | } | |
18ab4d4c | 2453 | plist_del(&p->list, &swap_active_head); |
ec8acf20 | 2454 | atomic_long_sub(p->pages, &nr_swap_pages); |
1da177e4 LT |
2455 | total_swap_pages -= p->pages; |
2456 | p->flags &= ~SWP_WRITEOK; | |
ec8acf20 | 2457 | spin_unlock(&p->lock); |
5d337b91 | 2458 | spin_unlock(&swap_lock); |
fb4f88dc | 2459 | |
039939a6 TC |
2460 | disable_swap_slots_cache_lock(); |
2461 | ||
e1e12d2f | 2462 | set_current_oom_origin(); |
10a9c496 | 2463 | err = try_to_unuse(p->type); |
e1e12d2f | 2464 | clear_current_oom_origin(); |
1da177e4 | 2465 | |
1da177e4 LT |
2466 | if (err) { |
2467 | /* re-insert swap space back into swap_list */ | |
cf0cac0a | 2468 | reinsert_swap_info(p); |
039939a6 | 2469 | reenable_swap_slots_cache_unlock(); |
1da177e4 LT |
2470 | goto out_dput; |
2471 | } | |
52b7efdb | 2472 | |
039939a6 TC |
2473 | reenable_swap_slots_cache_unlock(); |
2474 | ||
eb085574 | 2475 | /* |
63d8620e ML |
2476 | * Wait for swap operations protected by get/put_swap_device() |
2477 | * to complete. | |
2478 | * | |
2479 | * We need synchronize_rcu() here to protect the accessing to | |
2480 | * the swap cache data structure. | |
eb085574 | 2481 | */ |
63d8620e | 2482 | percpu_ref_kill(&p->users); |
eb085574 | 2483 | synchronize_rcu(); |
63d8620e | 2484 | wait_for_completion(&p->comp); |
eb085574 | 2485 | |
815c2c54 SL |
2486 | flush_work(&p->discard_work); |
2487 | ||
5d337b91 | 2488 | destroy_swap_extents(p); |
570a335b HD |
2489 | if (p->flags & SWP_CONTINUED) |
2490 | free_swap_count_continuations(p); | |
2491 | ||
10f0d2a5 | 2492 | if (!p->bdev || !bdev_nonrot(p->bdev)) |
81a0298b HY |
2493 | atomic_dec(&nr_rotate_swap); |
2494 | ||
fc0abb14 | 2495 | mutex_lock(&swapon_mutex); |
5d337b91 | 2496 | spin_lock(&swap_lock); |
ec8acf20 | 2497 | spin_lock(&p->lock); |
5d337b91 HD |
2498 | drain_mmlist(); |
2499 | ||
bb243f7d | 2500 | /* wait for anyone still in scan_swap_map_slots */ |
52b7efdb HD |
2501 | p->highest_bit = 0; /* cuts scans short */ |
2502 | while (p->flags >= SWP_SCANNING) { | |
ec8acf20 | 2503 | spin_unlock(&p->lock); |
5d337b91 | 2504 | spin_unlock(&swap_lock); |
13e4b57f | 2505 | schedule_timeout_uninterruptible(1); |
5d337b91 | 2506 | spin_lock(&swap_lock); |
ec8acf20 | 2507 | spin_lock(&p->lock); |
52b7efdb | 2508 | } |
52b7efdb | 2509 | |
1da177e4 | 2510 | swap_file = p->swap_file; |
5b808a23 | 2511 | old_block_size = p->old_block_size; |
1da177e4 LT |
2512 | p->swap_file = NULL; |
2513 | p->max = 0; | |
2514 | swap_map = p->swap_map; | |
2515 | p->swap_map = NULL; | |
2a8f9449 SL |
2516 | cluster_info = p->cluster_info; |
2517 | p->cluster_info = NULL; | |
4f89849d | 2518 | frontswap_map = frontswap_map_get(p); |
ec8acf20 | 2519 | spin_unlock(&p->lock); |
5d337b91 | 2520 | spin_unlock(&swap_lock); |
8a84802e | 2521 | arch_swap_invalidate_area(p->type); |
adfab836 | 2522 | frontswap_invalidate_area(p->type); |
58e97ba6 | 2523 | frontswap_map_set(p, NULL); |
fc0abb14 | 2524 | mutex_unlock(&swapon_mutex); |
ebc2a1a6 SL |
2525 | free_percpu(p->percpu_cluster); |
2526 | p->percpu_cluster = NULL; | |
49070588 HY |
2527 | free_percpu(p->cluster_next_cpu); |
2528 | p->cluster_next_cpu = NULL; | |
1da177e4 | 2529 | vfree(swap_map); |
54f180d3 HY |
2530 | kvfree(cluster_info); |
2531 | kvfree(frontswap_map); | |
2de1a7e4 | 2532 | /* Destroy swap account information */ |
adfab836 | 2533 | swap_cgroup_swapoff(p->type); |
4b3ef9da | 2534 | exit_swap_address_space(p->type); |
27a7faa0 | 2535 | |
1da177e4 LT |
2536 | inode = mapping->host; |
2537 | if (S_ISBLK(inode->i_mode)) { | |
2538 | struct block_device *bdev = I_BDEV(inode); | |
1638045c | 2539 | |
5b808a23 | 2540 | set_blocksize(bdev, old_block_size); |
2736e8ee | 2541 | blkdev_put(bdev, p); |
1da177e4 | 2542 | } |
1638045c DW |
2543 | |
2544 | inode_lock(inode); | |
2545 | inode->i_flags &= ~S_SWAPFILE; | |
2546 | inode_unlock(inode); | |
1da177e4 | 2547 | filp_close(swap_file, NULL); |
f893ab41 WY |
2548 | |
2549 | /* | |
2550 | * Clear the SWP_USED flag after all resources are freed so that swapon | |
2551 | * can reuse this swap_info in alloc_swap_info() safely. It is ok to | |
2552 | * not hold p->lock after we cleared its SWP_WRITEOK. | |
2553 | */ | |
2554 | spin_lock(&swap_lock); | |
2555 | p->flags = 0; | |
2556 | spin_unlock(&swap_lock); | |
2557 | ||
1da177e4 | 2558 | err = 0; |
66d7dd51 KS |
2559 | atomic_inc(&proc_poll_event); |
2560 | wake_up_interruptible(&proc_poll_wait); | |
1da177e4 LT |
2561 | |
2562 | out_dput: | |
2563 | filp_close(victim, NULL); | |
2564 | out: | |
f58b59c1 | 2565 | putname(pathname); |
1da177e4 LT |
2566 | return err; |
2567 | } | |
2568 | ||
2569 | #ifdef CONFIG_PROC_FS | |
9dd95748 | 2570 | static __poll_t swaps_poll(struct file *file, poll_table *wait) |
66d7dd51 | 2571 | { |
f1514638 | 2572 | struct seq_file *seq = file->private_data; |
66d7dd51 KS |
2573 | |
2574 | poll_wait(file, &proc_poll_wait, wait); | |
2575 | ||
f1514638 KS |
2576 | if (seq->poll_event != atomic_read(&proc_poll_event)) { |
2577 | seq->poll_event = atomic_read(&proc_poll_event); | |
a9a08845 | 2578 | return EPOLLIN | EPOLLRDNORM | EPOLLERR | EPOLLPRI; |
66d7dd51 KS |
2579 | } |
2580 | ||
a9a08845 | 2581 | return EPOLLIN | EPOLLRDNORM; |
66d7dd51 KS |
2582 | } |
2583 | ||
1da177e4 LT |
2584 | /* iterator */ |
2585 | static void *swap_start(struct seq_file *swap, loff_t *pos) | |
2586 | { | |
efa90a98 HD |
2587 | struct swap_info_struct *si; |
2588 | int type; | |
1da177e4 LT |
2589 | loff_t l = *pos; |
2590 | ||
fc0abb14 | 2591 | mutex_lock(&swapon_mutex); |
1da177e4 | 2592 | |
881e4aab SS |
2593 | if (!l) |
2594 | return SEQ_START_TOKEN; | |
2595 | ||
c10d38cc | 2596 | for (type = 0; (si = swap_type_to_swap_info(type)); type++) { |
efa90a98 | 2597 | if (!(si->flags & SWP_USED) || !si->swap_map) |
1da177e4 | 2598 | continue; |
881e4aab | 2599 | if (!--l) |
efa90a98 | 2600 | return si; |
1da177e4 LT |
2601 | } |
2602 | ||
2603 | return NULL; | |
2604 | } | |
2605 | ||
2606 | static void *swap_next(struct seq_file *swap, void *v, loff_t *pos) | |
2607 | { | |
efa90a98 HD |
2608 | struct swap_info_struct *si = v; |
2609 | int type; | |
1da177e4 | 2610 | |
881e4aab | 2611 | if (v == SEQ_START_TOKEN) |
efa90a98 HD |
2612 | type = 0; |
2613 | else | |
2614 | type = si->type + 1; | |
881e4aab | 2615 | |
10c8d69f | 2616 | ++(*pos); |
c10d38cc | 2617 | for (; (si = swap_type_to_swap_info(type)); type++) { |
efa90a98 | 2618 | if (!(si->flags & SWP_USED) || !si->swap_map) |
1da177e4 | 2619 | continue; |
efa90a98 | 2620 | return si; |
1da177e4 LT |
2621 | } |
2622 | ||
2623 | return NULL; | |
2624 | } | |
2625 | ||
2626 | static void swap_stop(struct seq_file *swap, void *v) | |
2627 | { | |
fc0abb14 | 2628 | mutex_unlock(&swapon_mutex); |
1da177e4 LT |
2629 | } |
2630 | ||
2631 | static int swap_show(struct seq_file *swap, void *v) | |
2632 | { | |
efa90a98 | 2633 | struct swap_info_struct *si = v; |
1da177e4 LT |
2634 | struct file *file; |
2635 | int len; | |
642929a2 | 2636 | unsigned long bytes, inuse; |
1da177e4 | 2637 | |
efa90a98 | 2638 | if (si == SEQ_START_TOKEN) { |
68d68ff6 | 2639 | seq_puts(swap, "Filename\t\t\t\tType\t\tSize\t\tUsed\t\tPriority\n"); |
881e4aab SS |
2640 | return 0; |
2641 | } | |
1da177e4 | 2642 | |
6f793940 | 2643 | bytes = si->pages << (PAGE_SHIFT - 10); |
c8945306 | 2644 | inuse = READ_ONCE(si->inuse_pages) << (PAGE_SHIFT - 10); |
6f793940 | 2645 | |
efa90a98 | 2646 | file = si->swap_file; |
2726d566 | 2647 | len = seq_file_path(swap, file, " \t\n\\"); |
642929a2 | 2648 | seq_printf(swap, "%*s%s\t%lu\t%s%lu\t%s%d\n", |
886bb7e9 | 2649 | len < 40 ? 40 - len : 1, " ", |
496ad9aa | 2650 | S_ISBLK(file_inode(file)->i_mode) ? |
1da177e4 | 2651 | "partition" : "file\t", |
6f793940 RD |
2652 | bytes, bytes < 10000000 ? "\t" : "", |
2653 | inuse, inuse < 10000000 ? "\t" : "", | |
efa90a98 | 2654 | si->prio); |
1da177e4 LT |
2655 | return 0; |
2656 | } | |
2657 | ||
15ad7cdc | 2658 | static const struct seq_operations swaps_op = { |
1da177e4 LT |
2659 | .start = swap_start, |
2660 | .next = swap_next, | |
2661 | .stop = swap_stop, | |
2662 | .show = swap_show | |
2663 | }; | |
2664 | ||
2665 | static int swaps_open(struct inode *inode, struct file *file) | |
2666 | { | |
f1514638 | 2667 | struct seq_file *seq; |
66d7dd51 KS |
2668 | int ret; |
2669 | ||
66d7dd51 | 2670 | ret = seq_open(file, &swaps_op); |
f1514638 | 2671 | if (ret) |
66d7dd51 | 2672 | return ret; |
66d7dd51 | 2673 | |
f1514638 KS |
2674 | seq = file->private_data; |
2675 | seq->poll_event = atomic_read(&proc_poll_event); | |
2676 | return 0; | |
1da177e4 LT |
2677 | } |
2678 | ||
97a32539 | 2679 | static const struct proc_ops swaps_proc_ops = { |
d919b33d | 2680 | .proc_flags = PROC_ENTRY_PERMANENT, |
97a32539 AD |
2681 | .proc_open = swaps_open, |
2682 | .proc_read = seq_read, | |
2683 | .proc_lseek = seq_lseek, | |
2684 | .proc_release = seq_release, | |
2685 | .proc_poll = swaps_poll, | |
1da177e4 LT |
2686 | }; |
2687 | ||
2688 | static int __init procswaps_init(void) | |
2689 | { | |
97a32539 | 2690 | proc_create("swaps", 0, NULL, &swaps_proc_ops); |
1da177e4 LT |
2691 | return 0; |
2692 | } | |
2693 | __initcall(procswaps_init); | |
2694 | #endif /* CONFIG_PROC_FS */ | |
2695 | ||
1796316a JB |
2696 | #ifdef MAX_SWAPFILES_CHECK |
2697 | static int __init max_swapfiles_check(void) | |
2698 | { | |
2699 | MAX_SWAPFILES_CHECK(); | |
2700 | return 0; | |
2701 | } | |
2702 | late_initcall(max_swapfiles_check); | |
2703 | #endif | |
2704 | ||
53cbb243 | 2705 | static struct swap_info_struct *alloc_swap_info(void) |
1da177e4 | 2706 | { |
73c34b6a | 2707 | struct swap_info_struct *p; |
b11a76b3 | 2708 | struct swap_info_struct *defer = NULL; |
1da177e4 | 2709 | unsigned int type; |
a2468cc9 | 2710 | int i; |
efa90a98 | 2711 | |
96008744 | 2712 | p = kvzalloc(struct_size(p, avail_lists, nr_node_ids), GFP_KERNEL); |
efa90a98 | 2713 | if (!p) |
53cbb243 | 2714 | return ERR_PTR(-ENOMEM); |
efa90a98 | 2715 | |
63d8620e ML |
2716 | if (percpu_ref_init(&p->users, swap_users_ref_free, |
2717 | PERCPU_REF_INIT_DEAD, GFP_KERNEL)) { | |
2718 | kvfree(p); | |
2719 | return ERR_PTR(-ENOMEM); | |
2720 | } | |
2721 | ||
5d337b91 | 2722 | spin_lock(&swap_lock); |
efa90a98 HD |
2723 | for (type = 0; type < nr_swapfiles; type++) { |
2724 | if (!(swap_info[type]->flags & SWP_USED)) | |
1da177e4 | 2725 | break; |
efa90a98 | 2726 | } |
0697212a | 2727 | if (type >= MAX_SWAPFILES) { |
5d337b91 | 2728 | spin_unlock(&swap_lock); |
63d8620e | 2729 | percpu_ref_exit(&p->users); |
873d7bcf | 2730 | kvfree(p); |
730c0581 | 2731 | return ERR_PTR(-EPERM); |
1da177e4 | 2732 | } |
efa90a98 HD |
2733 | if (type >= nr_swapfiles) { |
2734 | p->type = type; | |
efa90a98 | 2735 | /* |
a4b45114 HY |
2736 | * Publish the swap_info_struct after initializing it. |
2737 | * Note that kvzalloc() above zeroes all its fields. | |
efa90a98 | 2738 | */ |
a4b45114 HY |
2739 | smp_store_release(&swap_info[type], p); /* rcu_assign_pointer() */ |
2740 | nr_swapfiles++; | |
efa90a98 | 2741 | } else { |
b11a76b3 | 2742 | defer = p; |
efa90a98 HD |
2743 | p = swap_info[type]; |
2744 | /* | |
2745 | * Do not memset this entry: a racing procfs swap_next() | |
2746 | * would be relying on p->type to remain valid. | |
2747 | */ | |
2748 | } | |
4efaceb1 | 2749 | p->swap_extent_root = RB_ROOT; |
18ab4d4c | 2750 | plist_node_init(&p->list, 0); |
a2468cc9 AL |
2751 | for_each_node(i) |
2752 | plist_node_init(&p->avail_lists[i], 0); | |
1da177e4 | 2753 | p->flags = SWP_USED; |
5d337b91 | 2754 | spin_unlock(&swap_lock); |
63d8620e ML |
2755 | if (defer) { |
2756 | percpu_ref_exit(&defer->users); | |
2757 | kvfree(defer); | |
2758 | } | |
ec8acf20 | 2759 | spin_lock_init(&p->lock); |
2628bd6f | 2760 | spin_lock_init(&p->cont_lock); |
63d8620e | 2761 | init_completion(&p->comp); |
efa90a98 | 2762 | |
53cbb243 | 2763 | return p; |
53cbb243 CEB |
2764 | } |
2765 | ||
4d0e1e10 CEB |
2766 | static int claim_swapfile(struct swap_info_struct *p, struct inode *inode) |
2767 | { | |
2768 | int error; | |
2769 | ||
2770 | if (S_ISBLK(inode->i_mode)) { | |
ef16e1d9 | 2771 | p->bdev = blkdev_get_by_dev(inode->i_rdev, |
05bdb996 | 2772 | BLK_OPEN_READ | BLK_OPEN_WRITE, p, NULL); |
ef16e1d9 CH |
2773 | if (IS_ERR(p->bdev)) { |
2774 | error = PTR_ERR(p->bdev); | |
4d0e1e10 | 2775 | p->bdev = NULL; |
6f179af8 | 2776 | return error; |
4d0e1e10 CEB |
2777 | } |
2778 | p->old_block_size = block_size(p->bdev); | |
2779 | error = set_blocksize(p->bdev, PAGE_SIZE); | |
2780 | if (error < 0) | |
87ade72a | 2781 | return error; |
12d2966d NA |
2782 | /* |
2783 | * Zoned block devices contain zones that have a sequential | |
2784 | * write only restriction. Hence zoned block devices are not | |
2785 | * suitable for swapping. Disallow them here. | |
2786 | */ | |
9964e674 | 2787 | if (bdev_is_zoned(p->bdev)) |
12d2966d | 2788 | return -EINVAL; |
4d0e1e10 CEB |
2789 | p->flags |= SWP_BLKDEV; |
2790 | } else if (S_ISREG(inode->i_mode)) { | |
2791 | p->bdev = inode->i_sb->s_bdev; | |
1638045c DW |
2792 | } |
2793 | ||
4d0e1e10 | 2794 | return 0; |
4d0e1e10 CEB |
2795 | } |
2796 | ||
377eeaa8 AK |
2797 | |
2798 | /* | |
2799 | * Find out how many pages are allowed for a single swap device. There | |
2800 | * are two limiting factors: | |
2801 | * 1) the number of bits for the swap offset in the swp_entry_t type, and | |
2802 | * 2) the number of bits in the swap pte, as defined by the different | |
2803 | * architectures. | |
2804 | * | |
2805 | * In order to find the largest possible bit mask, a swap entry with | |
2806 | * swap type 0 and swap offset ~0UL is created, encoded to a swap pte, | |
2807 | * decoded to a swp_entry_t again, and finally the swap offset is | |
2808 | * extracted. | |
2809 | * | |
2810 | * This will mask all the bits from the initial ~0UL mask that can't | |
2811 | * be encoded in either the swp_entry_t or the architecture definition | |
2812 | * of a swap pte. | |
2813 | */ | |
2814 | unsigned long generic_max_swapfile_size(void) | |
2815 | { | |
2816 | return swp_offset(pte_to_swp_entry( | |
2817 | swp_entry_to_pte(swp_entry(0, ~0UL)))) + 1; | |
2818 | } | |
2819 | ||
2820 | /* Can be overridden by an architecture for additional checks. */ | |
be45a490 | 2821 | __weak unsigned long arch_max_swapfile_size(void) |
377eeaa8 AK |
2822 | { |
2823 | return generic_max_swapfile_size(); | |
2824 | } | |
2825 | ||
ca8bd38b CEB |
2826 | static unsigned long read_swap_header(struct swap_info_struct *p, |
2827 | union swap_header *swap_header, | |
2828 | struct inode *inode) | |
2829 | { | |
2830 | int i; | |
2831 | unsigned long maxpages; | |
2832 | unsigned long swapfilepages; | |
d6bbbd29 | 2833 | unsigned long last_page; |
ca8bd38b CEB |
2834 | |
2835 | if (memcmp("SWAPSPACE2", swap_header->magic.magic, 10)) { | |
465c47fd | 2836 | pr_err("Unable to find swap-space signature\n"); |
38719025 | 2837 | return 0; |
ca8bd38b CEB |
2838 | } |
2839 | ||
041711ce | 2840 | /* swap partition endianness hack... */ |
ca8bd38b CEB |
2841 | if (swab32(swap_header->info.version) == 1) { |
2842 | swab32s(&swap_header->info.version); | |
2843 | swab32s(&swap_header->info.last_page); | |
2844 | swab32s(&swap_header->info.nr_badpages); | |
dd111be6 JH |
2845 | if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES) |
2846 | return 0; | |
ca8bd38b CEB |
2847 | for (i = 0; i < swap_header->info.nr_badpages; i++) |
2848 | swab32s(&swap_header->info.badpages[i]); | |
2849 | } | |
2850 | /* Check the swap header's sub-version */ | |
2851 | if (swap_header->info.version != 1) { | |
465c47fd AM |
2852 | pr_warn("Unable to handle swap header version %d\n", |
2853 | swap_header->info.version); | |
38719025 | 2854 | return 0; |
ca8bd38b CEB |
2855 | } |
2856 | ||
2857 | p->lowest_bit = 1; | |
2858 | p->cluster_next = 1; | |
2859 | p->cluster_nr = 0; | |
2860 | ||
be45a490 | 2861 | maxpages = swapfile_maximum_size; |
d6bbbd29 | 2862 | last_page = swap_header->info.last_page; |
a06ad633 TA |
2863 | if (!last_page) { |
2864 | pr_warn("Empty swap-file\n"); | |
2865 | return 0; | |
2866 | } | |
d6bbbd29 | 2867 | if (last_page > maxpages) { |
465c47fd | 2868 | pr_warn("Truncating oversized swap area, only using %luk out of %luk\n", |
d6bbbd29 RJ |
2869 | maxpages << (PAGE_SHIFT - 10), |
2870 | last_page << (PAGE_SHIFT - 10)); | |
2871 | } | |
2872 | if (maxpages > last_page) { | |
2873 | maxpages = last_page + 1; | |
ca8bd38b CEB |
2874 | /* p->max is an unsigned int: don't overflow it */ |
2875 | if ((unsigned int)maxpages == 0) | |
2876 | maxpages = UINT_MAX; | |
2877 | } | |
2878 | p->highest_bit = maxpages - 1; | |
2879 | ||
2880 | if (!maxpages) | |
38719025 | 2881 | return 0; |
ca8bd38b CEB |
2882 | swapfilepages = i_size_read(inode) >> PAGE_SHIFT; |
2883 | if (swapfilepages && maxpages > swapfilepages) { | |
465c47fd | 2884 | pr_warn("Swap area shorter than signature indicates\n"); |
38719025 | 2885 | return 0; |
ca8bd38b CEB |
2886 | } |
2887 | if (swap_header->info.nr_badpages && S_ISREG(inode->i_mode)) | |
38719025 | 2888 | return 0; |
ca8bd38b | 2889 | if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES) |
38719025 | 2890 | return 0; |
ca8bd38b CEB |
2891 | |
2892 | return maxpages; | |
ca8bd38b CEB |
2893 | } |
2894 | ||
4b3ef9da | 2895 | #define SWAP_CLUSTER_INFO_COLS \ |
235b6217 | 2896 | DIV_ROUND_UP(L1_CACHE_BYTES, sizeof(struct swap_cluster_info)) |
4b3ef9da HY |
2897 | #define SWAP_CLUSTER_SPACE_COLS \ |
2898 | DIV_ROUND_UP(SWAP_ADDRESS_SPACE_PAGES, SWAPFILE_CLUSTER) | |
2899 | #define SWAP_CLUSTER_COLS \ | |
2900 | max_t(unsigned int, SWAP_CLUSTER_INFO_COLS, SWAP_CLUSTER_SPACE_COLS) | |
235b6217 | 2901 | |
915d4d7b CEB |
2902 | static int setup_swap_map_and_extents(struct swap_info_struct *p, |
2903 | union swap_header *swap_header, | |
2904 | unsigned char *swap_map, | |
2a8f9449 | 2905 | struct swap_cluster_info *cluster_info, |
915d4d7b CEB |
2906 | unsigned long maxpages, |
2907 | sector_t *span) | |
2908 | { | |
235b6217 | 2909 | unsigned int j, k; |
915d4d7b CEB |
2910 | unsigned int nr_good_pages; |
2911 | int nr_extents; | |
2a8f9449 | 2912 | unsigned long nr_clusters = DIV_ROUND_UP(maxpages, SWAPFILE_CLUSTER); |
235b6217 HY |
2913 | unsigned long col = p->cluster_next / SWAPFILE_CLUSTER % SWAP_CLUSTER_COLS; |
2914 | unsigned long i, idx; | |
915d4d7b CEB |
2915 | |
2916 | nr_good_pages = maxpages - 1; /* omit header page */ | |
2917 | ||
6b534915 HY |
2918 | cluster_list_init(&p->free_clusters); |
2919 | cluster_list_init(&p->discard_clusters); | |
2a8f9449 | 2920 | |
915d4d7b CEB |
2921 | for (i = 0; i < swap_header->info.nr_badpages; i++) { |
2922 | unsigned int page_nr = swap_header->info.badpages[i]; | |
bdb8e3f6 CEB |
2923 | if (page_nr == 0 || page_nr > swap_header->info.last_page) |
2924 | return -EINVAL; | |
915d4d7b CEB |
2925 | if (page_nr < maxpages) { |
2926 | swap_map[page_nr] = SWAP_MAP_BAD; | |
2927 | nr_good_pages--; | |
2a8f9449 SL |
2928 | /* |
2929 | * Haven't marked the cluster free yet, no list | |
2930 | * operation involved | |
2931 | */ | |
2932 | inc_cluster_info_page(p, cluster_info, page_nr); | |
915d4d7b CEB |
2933 | } |
2934 | } | |
2935 | ||
2a8f9449 SL |
2936 | /* Haven't marked the cluster free yet, no list operation involved */ |
2937 | for (i = maxpages; i < round_up(maxpages, SWAPFILE_CLUSTER); i++) | |
2938 | inc_cluster_info_page(p, cluster_info, i); | |
2939 | ||
915d4d7b CEB |
2940 | if (nr_good_pages) { |
2941 | swap_map[0] = SWAP_MAP_BAD; | |
2a8f9449 SL |
2942 | /* |
2943 | * Not mark the cluster free yet, no list | |
2944 | * operation involved | |
2945 | */ | |
2946 | inc_cluster_info_page(p, cluster_info, 0); | |
915d4d7b CEB |
2947 | p->max = maxpages; |
2948 | p->pages = nr_good_pages; | |
2949 | nr_extents = setup_swap_extents(p, span); | |
bdb8e3f6 CEB |
2950 | if (nr_extents < 0) |
2951 | return nr_extents; | |
915d4d7b CEB |
2952 | nr_good_pages = p->pages; |
2953 | } | |
2954 | if (!nr_good_pages) { | |
465c47fd | 2955 | pr_warn("Empty swap-file\n"); |
bdb8e3f6 | 2956 | return -EINVAL; |
915d4d7b CEB |
2957 | } |
2958 | ||
2a8f9449 SL |
2959 | if (!cluster_info) |
2960 | return nr_extents; | |
2961 | ||
235b6217 | 2962 | |
4b3ef9da HY |
2963 | /* |
2964 | * Reduce false cache line sharing between cluster_info and | |
2965 | * sharing same address space. | |
2966 | */ | |
235b6217 HY |
2967 | for (k = 0; k < SWAP_CLUSTER_COLS; k++) { |
2968 | j = (k + col) % SWAP_CLUSTER_COLS; | |
2969 | for (i = 0; i < DIV_ROUND_UP(nr_clusters, SWAP_CLUSTER_COLS); i++) { | |
2970 | idx = i * SWAP_CLUSTER_COLS + j; | |
2971 | if (idx >= nr_clusters) | |
2972 | continue; | |
2973 | if (cluster_count(&cluster_info[idx])) | |
2974 | continue; | |
2a8f9449 | 2975 | cluster_set_flag(&cluster_info[idx], CLUSTER_FLAG_FREE); |
6b534915 HY |
2976 | cluster_list_add_tail(&p->free_clusters, cluster_info, |
2977 | idx); | |
2a8f9449 | 2978 | } |
2a8f9449 | 2979 | } |
915d4d7b | 2980 | return nr_extents; |
915d4d7b CEB |
2981 | } |
2982 | ||
53cbb243 CEB |
2983 | SYSCALL_DEFINE2(swapon, const char __user *, specialfile, int, swap_flags) |
2984 | { | |
2985 | struct swap_info_struct *p; | |
91a27b2a | 2986 | struct filename *name; |
53cbb243 CEB |
2987 | struct file *swap_file = NULL; |
2988 | struct address_space *mapping; | |
51cc3a66 | 2989 | struct dentry *dentry; |
40531542 | 2990 | int prio; |
53cbb243 CEB |
2991 | int error; |
2992 | union swap_header *swap_header; | |
915d4d7b | 2993 | int nr_extents; |
53cbb243 CEB |
2994 | sector_t span; |
2995 | unsigned long maxpages; | |
53cbb243 | 2996 | unsigned char *swap_map = NULL; |
2a8f9449 | 2997 | struct swap_cluster_info *cluster_info = NULL; |
38b5faf4 | 2998 | unsigned long *frontswap_map = NULL; |
53cbb243 CEB |
2999 | struct page *page = NULL; |
3000 | struct inode *inode = NULL; | |
7cbf3192 | 3001 | bool inced_nr_rotate_swap = false; |
53cbb243 | 3002 | |
d15cab97 HD |
3003 | if (swap_flags & ~SWAP_FLAGS_VALID) |
3004 | return -EINVAL; | |
3005 | ||
53cbb243 CEB |
3006 | if (!capable(CAP_SYS_ADMIN)) |
3007 | return -EPERM; | |
3008 | ||
a2468cc9 AL |
3009 | if (!swap_avail_heads) |
3010 | return -ENOMEM; | |
3011 | ||
53cbb243 | 3012 | p = alloc_swap_info(); |
2542e513 CEB |
3013 | if (IS_ERR(p)) |
3014 | return PTR_ERR(p); | |
53cbb243 | 3015 | |
815c2c54 SL |
3016 | INIT_WORK(&p->discard_work, swap_discard_work); |
3017 | ||
1da177e4 | 3018 | name = getname(specialfile); |
1da177e4 | 3019 | if (IS_ERR(name)) { |
7de7fb6b | 3020 | error = PTR_ERR(name); |
1da177e4 | 3021 | name = NULL; |
bd69010b | 3022 | goto bad_swap; |
1da177e4 | 3023 | } |
669abf4e | 3024 | swap_file = file_open_name(name, O_RDWR|O_LARGEFILE, 0); |
1da177e4 | 3025 | if (IS_ERR(swap_file)) { |
7de7fb6b | 3026 | error = PTR_ERR(swap_file); |
1da177e4 | 3027 | swap_file = NULL; |
bd69010b | 3028 | goto bad_swap; |
1da177e4 LT |
3029 | } |
3030 | ||
3031 | p->swap_file = swap_file; | |
3032 | mapping = swap_file->f_mapping; | |
51cc3a66 | 3033 | dentry = swap_file->f_path.dentry; |
2130781e | 3034 | inode = mapping->host; |
6f179af8 | 3035 | |
4d0e1e10 CEB |
3036 | error = claim_swapfile(p, inode); |
3037 | if (unlikely(error)) | |
1da177e4 | 3038 | goto bad_swap; |
1da177e4 | 3039 | |
d795a90e | 3040 | inode_lock(inode); |
51cc3a66 HD |
3041 | if (d_unlinked(dentry) || cant_mount(dentry)) { |
3042 | error = -ENOENT; | |
3043 | goto bad_swap_unlock_inode; | |
3044 | } | |
d795a90e NA |
3045 | if (IS_SWAPFILE(inode)) { |
3046 | error = -EBUSY; | |
3047 | goto bad_swap_unlock_inode; | |
3048 | } | |
3049 | ||
1da177e4 LT |
3050 | /* |
3051 | * Read the swap header. | |
3052 | */ | |
7e0a1265 | 3053 | if (!mapping->a_ops->read_folio) { |
1da177e4 | 3054 | error = -EINVAL; |
d795a90e | 3055 | goto bad_swap_unlock_inode; |
1da177e4 | 3056 | } |
090d2b18 | 3057 | page = read_mapping_page(mapping, 0, swap_file); |
1da177e4 LT |
3058 | if (IS_ERR(page)) { |
3059 | error = PTR_ERR(page); | |
d795a90e | 3060 | goto bad_swap_unlock_inode; |
1da177e4 | 3061 | } |
81e33971 | 3062 | swap_header = kmap(page); |
1da177e4 | 3063 | |
ca8bd38b CEB |
3064 | maxpages = read_swap_header(p, swap_header, inode); |
3065 | if (unlikely(!maxpages)) { | |
1da177e4 | 3066 | error = -EINVAL; |
d795a90e | 3067 | goto bad_swap_unlock_inode; |
1da177e4 | 3068 | } |
886bb7e9 | 3069 | |
81e33971 | 3070 | /* OK, set up the swap map and apply the bad block list */ |
803d0c83 | 3071 | swap_map = vzalloc(maxpages); |
81e33971 HD |
3072 | if (!swap_map) { |
3073 | error = -ENOMEM; | |
d795a90e | 3074 | goto bad_swap_unlock_inode; |
81e33971 | 3075 | } |
f0571429 | 3076 | |
36d25489 | 3077 | if (p->bdev && bdev_stable_writes(p->bdev)) |
f0571429 MK |
3078 | p->flags |= SWP_STABLE_WRITES; |
3079 | ||
3222d8c2 | 3080 | if (p->bdev && bdev_synchronous(p->bdev)) |
539a6fea MK |
3081 | p->flags |= SWP_SYNCHRONOUS_IO; |
3082 | ||
10f0d2a5 | 3083 | if (p->bdev && bdev_nonrot(p->bdev)) { |
6f179af8 | 3084 | int cpu; |
235b6217 | 3085 | unsigned long ci, nr_cluster; |
6f179af8 | 3086 | |
2a8f9449 | 3087 | p->flags |= SWP_SOLIDSTATE; |
49070588 HY |
3088 | p->cluster_next_cpu = alloc_percpu(unsigned int); |
3089 | if (!p->cluster_next_cpu) { | |
3090 | error = -ENOMEM; | |
3091 | goto bad_swap_unlock_inode; | |
3092 | } | |
2a8f9449 SL |
3093 | /* |
3094 | * select a random position to start with to help wear leveling | |
3095 | * SSD | |
3096 | */ | |
49070588 HY |
3097 | for_each_possible_cpu(cpu) { |
3098 | per_cpu(*p->cluster_next_cpu, cpu) = | |
e8a533cb | 3099 | get_random_u32_inclusive(1, p->highest_bit); |
49070588 | 3100 | } |
235b6217 | 3101 | nr_cluster = DIV_ROUND_UP(maxpages, SWAPFILE_CLUSTER); |
2a8f9449 | 3102 | |
778e1cdd | 3103 | cluster_info = kvcalloc(nr_cluster, sizeof(*cluster_info), |
54f180d3 | 3104 | GFP_KERNEL); |
2a8f9449 SL |
3105 | if (!cluster_info) { |
3106 | error = -ENOMEM; | |
d795a90e | 3107 | goto bad_swap_unlock_inode; |
2a8f9449 | 3108 | } |
235b6217 HY |
3109 | |
3110 | for (ci = 0; ci < nr_cluster; ci++) | |
3111 | spin_lock_init(&((cluster_info + ci)->lock)); | |
3112 | ||
ebc2a1a6 SL |
3113 | p->percpu_cluster = alloc_percpu(struct percpu_cluster); |
3114 | if (!p->percpu_cluster) { | |
3115 | error = -ENOMEM; | |
d795a90e | 3116 | goto bad_swap_unlock_inode; |
ebc2a1a6 | 3117 | } |
6f179af8 | 3118 | for_each_possible_cpu(cpu) { |
ebc2a1a6 | 3119 | struct percpu_cluster *cluster; |
6f179af8 | 3120 | cluster = per_cpu_ptr(p->percpu_cluster, cpu); |
ebc2a1a6 SL |
3121 | cluster_set_null(&cluster->index); |
3122 | } | |
7cbf3192 | 3123 | } else { |
81a0298b | 3124 | atomic_inc(&nr_rotate_swap); |
7cbf3192 OS |
3125 | inced_nr_rotate_swap = true; |
3126 | } | |
1da177e4 | 3127 | |
1421ef3c CEB |
3128 | error = swap_cgroup_swapon(p->type, maxpages); |
3129 | if (error) | |
d795a90e | 3130 | goto bad_swap_unlock_inode; |
1421ef3c | 3131 | |
915d4d7b | 3132 | nr_extents = setup_swap_map_and_extents(p, swap_header, swap_map, |
2a8f9449 | 3133 | cluster_info, maxpages, &span); |
915d4d7b CEB |
3134 | if (unlikely(nr_extents < 0)) { |
3135 | error = nr_extents; | |
d795a90e | 3136 | goto bad_swap_unlock_inode; |
1da177e4 | 3137 | } |
38b5faf4 | 3138 | /* frontswap enabled? set up bit-per-page map for frontswap */ |
8ea1d2a1 | 3139 | if (IS_ENABLED(CONFIG_FRONTSWAP)) |
778e1cdd KC |
3140 | frontswap_map = kvcalloc(BITS_TO_LONGS(maxpages), |
3141 | sizeof(long), | |
54f180d3 | 3142 | GFP_KERNEL); |
1da177e4 | 3143 | |
70200574 CH |
3144 | if ((swap_flags & SWAP_FLAG_DISCARD) && |
3145 | p->bdev && bdev_max_discard_sectors(p->bdev)) { | |
2a8f9449 SL |
3146 | /* |
3147 | * When discard is enabled for swap with no particular | |
3148 | * policy flagged, we set all swap discard flags here in | |
3149 | * order to sustain backward compatibility with older | |
3150 | * swapon(8) releases. | |
3151 | */ | |
3152 | p->flags |= (SWP_DISCARDABLE | SWP_AREA_DISCARD | | |
3153 | SWP_PAGE_DISCARD); | |
dcf6b7dd | 3154 | |
2a8f9449 SL |
3155 | /* |
3156 | * By flagging sys_swapon, a sysadmin can tell us to | |
3157 | * either do single-time area discards only, or to just | |
3158 | * perform discards for released swap page-clusters. | |
3159 | * Now it's time to adjust the p->flags accordingly. | |
3160 | */ | |
3161 | if (swap_flags & SWAP_FLAG_DISCARD_ONCE) | |
3162 | p->flags &= ~SWP_PAGE_DISCARD; | |
3163 | else if (swap_flags & SWAP_FLAG_DISCARD_PAGES) | |
3164 | p->flags &= ~SWP_AREA_DISCARD; | |
3165 | ||
3166 | /* issue a swapon-time discard if it's still required */ | |
3167 | if (p->flags & SWP_AREA_DISCARD) { | |
3168 | int err = discard_swap(p); | |
3169 | if (unlikely(err)) | |
3170 | pr_err("swapon: discard_swap(%p): %d\n", | |
3171 | p, err); | |
dcf6b7dd | 3172 | } |
20137a49 | 3173 | } |
6a6ba831 | 3174 | |
4b3ef9da HY |
3175 | error = init_swap_address_space(p->type, maxpages); |
3176 | if (error) | |
d795a90e | 3177 | goto bad_swap_unlock_inode; |
4b3ef9da | 3178 | |
dc617f29 DW |
3179 | /* |
3180 | * Flush any pending IO and dirty mappings before we start using this | |
3181 | * swap device. | |
3182 | */ | |
3183 | inode->i_flags |= S_SWAPFILE; | |
3184 | error = inode_drain_writes(inode); | |
3185 | if (error) { | |
3186 | inode->i_flags &= ~S_SWAPFILE; | |
822bca52 | 3187 | goto free_swap_address_space; |
dc617f29 DW |
3188 | } |
3189 | ||
fc0abb14 | 3190 | mutex_lock(&swapon_mutex); |
40531542 | 3191 | prio = -1; |
78ecba08 | 3192 | if (swap_flags & SWAP_FLAG_PREFER) |
40531542 | 3193 | prio = |
78ecba08 | 3194 | (swap_flags & SWAP_FLAG_PRIO_MASK) >> SWAP_FLAG_PRIO_SHIFT; |
2a8f9449 | 3195 | enable_swap_info(p, prio, swap_map, cluster_info, frontswap_map); |
c69dbfb8 | 3196 | |
756a025f | 3197 | pr_info("Adding %uk swap on %s. Priority:%d extents:%d across:%lluk %s%s%s%s%s\n", |
91a27b2a | 3198 | p->pages<<(PAGE_SHIFT-10), name->name, p->prio, |
c69dbfb8 CEB |
3199 | nr_extents, (unsigned long long)span<<(PAGE_SHIFT-10), |
3200 | (p->flags & SWP_SOLIDSTATE) ? "SS" : "", | |
38b5faf4 | 3201 | (p->flags & SWP_DISCARDABLE) ? "D" : "", |
dcf6b7dd RA |
3202 | (p->flags & SWP_AREA_DISCARD) ? "s" : "", |
3203 | (p->flags & SWP_PAGE_DISCARD) ? "c" : "", | |
38b5faf4 | 3204 | (frontswap_map) ? "FS" : ""); |
c69dbfb8 | 3205 | |
fc0abb14 | 3206 | mutex_unlock(&swapon_mutex); |
66d7dd51 KS |
3207 | atomic_inc(&proc_poll_event); |
3208 | wake_up_interruptible(&proc_poll_wait); | |
3209 | ||
1da177e4 LT |
3210 | error = 0; |
3211 | goto out; | |
822bca52 ML |
3212 | free_swap_address_space: |
3213 | exit_swap_address_space(p->type); | |
d795a90e NA |
3214 | bad_swap_unlock_inode: |
3215 | inode_unlock(inode); | |
1da177e4 | 3216 | bad_swap: |
ebc2a1a6 SL |
3217 | free_percpu(p->percpu_cluster); |
3218 | p->percpu_cluster = NULL; | |
49070588 HY |
3219 | free_percpu(p->cluster_next_cpu); |
3220 | p->cluster_next_cpu = NULL; | |
bd69010b | 3221 | if (inode && S_ISBLK(inode->i_mode) && p->bdev) { |
f2090d2d | 3222 | set_blocksize(p->bdev, p->old_block_size); |
2736e8ee | 3223 | blkdev_put(p->bdev, p); |
1da177e4 | 3224 | } |
d795a90e | 3225 | inode = NULL; |
4cd3bb10 | 3226 | destroy_swap_extents(p); |
e8e6c2ec | 3227 | swap_cgroup_swapoff(p->type); |
5d337b91 | 3228 | spin_lock(&swap_lock); |
1da177e4 | 3229 | p->swap_file = NULL; |
1da177e4 | 3230 | p->flags = 0; |
5d337b91 | 3231 | spin_unlock(&swap_lock); |
1da177e4 | 3232 | vfree(swap_map); |
8606a1a9 | 3233 | kvfree(cluster_info); |
b6b1fd2a | 3234 | kvfree(frontswap_map); |
7cbf3192 OS |
3235 | if (inced_nr_rotate_swap) |
3236 | atomic_dec(&nr_rotate_swap); | |
d795a90e | 3237 | if (swap_file) |
1da177e4 LT |
3238 | filp_close(swap_file, NULL); |
3239 | out: | |
3240 | if (page && !IS_ERR(page)) { | |
3241 | kunmap(page); | |
09cbfeaf | 3242 | put_page(page); |
1da177e4 LT |
3243 | } |
3244 | if (name) | |
3245 | putname(name); | |
1638045c | 3246 | if (inode) |
5955102c | 3247 | inode_unlock(inode); |
039939a6 TC |
3248 | if (!error) |
3249 | enable_swap_slots_cache(); | |
1da177e4 LT |
3250 | return error; |
3251 | } | |
3252 | ||
3253 | void si_swapinfo(struct sysinfo *val) | |
3254 | { | |
efa90a98 | 3255 | unsigned int type; |
1da177e4 LT |
3256 | unsigned long nr_to_be_unused = 0; |
3257 | ||
5d337b91 | 3258 | spin_lock(&swap_lock); |
efa90a98 HD |
3259 | for (type = 0; type < nr_swapfiles; type++) { |
3260 | struct swap_info_struct *si = swap_info[type]; | |
3261 | ||
3262 | if ((si->flags & SWP_USED) && !(si->flags & SWP_WRITEOK)) | |
c8945306 | 3263 | nr_to_be_unused += READ_ONCE(si->inuse_pages); |
1da177e4 | 3264 | } |
ec8acf20 | 3265 | val->freeswap = atomic_long_read(&nr_swap_pages) + nr_to_be_unused; |
1da177e4 | 3266 | val->totalswap = total_swap_pages + nr_to_be_unused; |
5d337b91 | 3267 | spin_unlock(&swap_lock); |
1da177e4 LT |
3268 | } |
3269 | ||
3270 | /* | |
3271 | * Verify that a swap entry is valid and increment its swap map count. | |
3272 | * | |
355cfa73 KH |
3273 | * Returns error code in following case. |
3274 | * - success -> 0 | |
3275 | * - swp_entry is invalid -> EINVAL | |
3276 | * - swp_entry is migration entry -> EINVAL | |
3277 | * - swap-cache reference is requested but there is already one. -> EEXIST | |
3278 | * - swap-cache reference is requested but the entry is not used. -> ENOENT | |
570a335b | 3279 | * - swap-mapped reference requested but needs continued swap count. -> ENOMEM |
1da177e4 | 3280 | */ |
8d69aaee | 3281 | static int __swap_duplicate(swp_entry_t entry, unsigned char usage) |
1da177e4 | 3282 | { |
73c34b6a | 3283 | struct swap_info_struct *p; |
235b6217 | 3284 | struct swap_cluster_info *ci; |
c10d38cc | 3285 | unsigned long offset; |
8d69aaee HD |
3286 | unsigned char count; |
3287 | unsigned char has_cache; | |
9d9a0334 | 3288 | int err; |
1da177e4 | 3289 | |
c07aee4f | 3290 | p = swp_swap_info(entry); |
235b6217 | 3291 | |
eb085574 | 3292 | offset = swp_offset(entry); |
235b6217 | 3293 | ci = lock_cluster_or_swap_info(p, offset); |
355cfa73 | 3294 | |
253d553b | 3295 | count = p->swap_map[offset]; |
edfe23da SL |
3296 | |
3297 | /* | |
3298 | * swapin_readahead() doesn't check if a swap entry is valid, so the | |
3299 | * swap entry could be SWAP_MAP_BAD. Check here with lock held. | |
3300 | */ | |
3301 | if (unlikely(swap_count(count) == SWAP_MAP_BAD)) { | |
3302 | err = -ENOENT; | |
3303 | goto unlock_out; | |
3304 | } | |
3305 | ||
253d553b HD |
3306 | has_cache = count & SWAP_HAS_CACHE; |
3307 | count &= ~SWAP_HAS_CACHE; | |
3308 | err = 0; | |
355cfa73 | 3309 | |
253d553b | 3310 | if (usage == SWAP_HAS_CACHE) { |
355cfa73 KH |
3311 | |
3312 | /* set SWAP_HAS_CACHE if there is no cache and entry is used */ | |
253d553b HD |
3313 | if (!has_cache && count) |
3314 | has_cache = SWAP_HAS_CACHE; | |
3315 | else if (has_cache) /* someone else added cache */ | |
3316 | err = -EEXIST; | |
3317 | else /* no users remaining */ | |
3318 | err = -ENOENT; | |
355cfa73 KH |
3319 | |
3320 | } else if (count || has_cache) { | |
253d553b | 3321 | |
570a335b HD |
3322 | if ((count & ~COUNT_CONTINUED) < SWAP_MAP_MAX) |
3323 | count += usage; | |
3324 | else if ((count & ~COUNT_CONTINUED) > SWAP_MAP_MAX) | |
253d553b | 3325 | err = -EINVAL; |
570a335b HD |
3326 | else if (swap_count_continued(p, offset, count)) |
3327 | count = COUNT_CONTINUED; | |
3328 | else | |
3329 | err = -ENOMEM; | |
355cfa73 | 3330 | } else |
253d553b HD |
3331 | err = -ENOENT; /* unused swap entry */ |
3332 | ||
a449bf58 | 3333 | WRITE_ONCE(p->swap_map[offset], count | has_cache); |
253d553b | 3334 | |
355cfa73 | 3335 | unlock_out: |
235b6217 | 3336 | unlock_cluster_or_swap_info(p, ci); |
253d553b | 3337 | return err; |
1da177e4 | 3338 | } |
253d553b | 3339 | |
aaa46865 HD |
3340 | /* |
3341 | * Help swapoff by noting that swap entry belongs to shmem/tmpfs | |
3342 | * (in which case its reference count is never incremented). | |
3343 | */ | |
3344 | void swap_shmem_alloc(swp_entry_t entry) | |
3345 | { | |
3346 | __swap_duplicate(entry, SWAP_MAP_SHMEM); | |
3347 | } | |
3348 | ||
355cfa73 | 3349 | /* |
08259d58 HD |
3350 | * Increase reference count of swap entry by 1. |
3351 | * Returns 0 for success, or -ENOMEM if a swap_count_continuation is required | |
3352 | * but could not be atomically allocated. Returns 0, just as if it succeeded, | |
3353 | * if __swap_duplicate() fails for another reason (-EINVAL or -ENOENT), which | |
3354 | * might occur if a page table entry has got corrupted. | |
355cfa73 | 3355 | */ |
570a335b | 3356 | int swap_duplicate(swp_entry_t entry) |
355cfa73 | 3357 | { |
570a335b HD |
3358 | int err = 0; |
3359 | ||
3360 | while (!err && __swap_duplicate(entry, 1) == -ENOMEM) | |
3361 | err = add_swap_count_continuation(entry, GFP_ATOMIC); | |
3362 | return err; | |
355cfa73 | 3363 | } |
1da177e4 | 3364 | |
cb4b86ba | 3365 | /* |
355cfa73 KH |
3366 | * @entry: swap entry for which we allocate swap cache. |
3367 | * | |
73c34b6a | 3368 | * Called when allocating swap cache for existing swap entry, |
355cfa73 | 3369 | * This can return error codes. Returns 0 at success. |
3eeba135 | 3370 | * -EEXIST means there is a swap cache. |
355cfa73 | 3371 | * Note: return code is different from swap_duplicate(). |
cb4b86ba KH |
3372 | */ |
3373 | int swapcache_prepare(swp_entry_t entry) | |
3374 | { | |
253d553b | 3375 | return __swap_duplicate(entry, SWAP_HAS_CACHE); |
cb4b86ba KH |
3376 | } |
3377 | ||
0bcac06f MK |
3378 | struct swap_info_struct *swp_swap_info(swp_entry_t entry) |
3379 | { | |
c10d38cc | 3380 | return swap_type_to_swap_info(swp_type(entry)); |
0bcac06f MK |
3381 | } |
3382 | ||
f981c595 MG |
3383 | struct swap_info_struct *page_swap_info(struct page *page) |
3384 | { | |
0bcac06f MK |
3385 | swp_entry_t entry = { .val = page_private(page) }; |
3386 | return swp_swap_info(entry); | |
f981c595 MG |
3387 | } |
3388 | ||
3389 | /* | |
2f52578f | 3390 | * out-of-line methods to avoid include hell. |
f981c595 | 3391 | */ |
2f52578f | 3392 | struct address_space *swapcache_mapping(struct folio *folio) |
f981c595 | 3393 | { |
2f52578f | 3394 | return page_swap_info(&folio->page)->swap_file->f_mapping; |
f981c595 | 3395 | } |
2f52578f | 3396 | EXPORT_SYMBOL_GPL(swapcache_mapping); |
f981c595 MG |
3397 | |
3398 | pgoff_t __page_file_index(struct page *page) | |
3399 | { | |
3400 | swp_entry_t swap = { .val = page_private(page) }; | |
f981c595 MG |
3401 | return swp_offset(swap); |
3402 | } | |
3403 | EXPORT_SYMBOL_GPL(__page_file_index); | |
3404 | ||
570a335b HD |
3405 | /* |
3406 | * add_swap_count_continuation - called when a swap count is duplicated | |
3407 | * beyond SWAP_MAP_MAX, it allocates a new page and links that to the entry's | |
3408 | * page of the original vmalloc'ed swap_map, to hold the continuation count | |
3409 | * (for that entry and for its neighbouring PAGE_SIZE swap entries). Called | |
3410 | * again when count is duplicated beyond SWAP_MAP_MAX * SWAP_CONT_MAX, etc. | |
3411 | * | |
3412 | * These continuation pages are seldom referenced: the common paths all work | |
3413 | * on the original swap_map, only referring to a continuation page when the | |
3414 | * low "digit" of a count is incremented or decremented through SWAP_MAP_MAX. | |
3415 | * | |
3416 | * add_swap_count_continuation(, GFP_ATOMIC) can be called while holding | |
3417 | * page table locks; if it fails, add_swap_count_continuation(, GFP_KERNEL) | |
3418 | * can be called after dropping locks. | |
3419 | */ | |
3420 | int add_swap_count_continuation(swp_entry_t entry, gfp_t gfp_mask) | |
3421 | { | |
3422 | struct swap_info_struct *si; | |
235b6217 | 3423 | struct swap_cluster_info *ci; |
570a335b HD |
3424 | struct page *head; |
3425 | struct page *page; | |
3426 | struct page *list_page; | |
3427 | pgoff_t offset; | |
3428 | unsigned char count; | |
eb085574 | 3429 | int ret = 0; |
570a335b HD |
3430 | |
3431 | /* | |
3432 | * When debugging, it's easier to use __GFP_ZERO here; but it's better | |
3433 | * for latency not to zero a page while GFP_ATOMIC and holding locks. | |
3434 | */ | |
3435 | page = alloc_page(gfp_mask | __GFP_HIGHMEM); | |
3436 | ||
eb085574 | 3437 | si = get_swap_device(entry); |
570a335b HD |
3438 | if (!si) { |
3439 | /* | |
3440 | * An acceptable race has occurred since the failing | |
eb085574 | 3441 | * __swap_duplicate(): the swap device may be swapoff |
570a335b HD |
3442 | */ |
3443 | goto outer; | |
3444 | } | |
eb085574 | 3445 | spin_lock(&si->lock); |
570a335b HD |
3446 | |
3447 | offset = swp_offset(entry); | |
235b6217 HY |
3448 | |
3449 | ci = lock_cluster(si, offset); | |
3450 | ||
d8aa24e0 | 3451 | count = swap_count(si->swap_map[offset]); |
570a335b HD |
3452 | |
3453 | if ((count & ~COUNT_CONTINUED) != SWAP_MAP_MAX) { | |
3454 | /* | |
3455 | * The higher the swap count, the more likely it is that tasks | |
3456 | * will race to add swap count continuation: we need to avoid | |
3457 | * over-provisioning. | |
3458 | */ | |
3459 | goto out; | |
3460 | } | |
3461 | ||
3462 | if (!page) { | |
eb085574 HY |
3463 | ret = -ENOMEM; |
3464 | goto out; | |
570a335b HD |
3465 | } |
3466 | ||
570a335b HD |
3467 | head = vmalloc_to_page(si->swap_map + offset); |
3468 | offset &= ~PAGE_MASK; | |
3469 | ||
2628bd6f | 3470 | spin_lock(&si->cont_lock); |
570a335b HD |
3471 | /* |
3472 | * Page allocation does not initialize the page's lru field, | |
3473 | * but it does always reset its private field. | |
3474 | */ | |
3475 | if (!page_private(head)) { | |
3476 | BUG_ON(count & COUNT_CONTINUED); | |
3477 | INIT_LIST_HEAD(&head->lru); | |
3478 | set_page_private(head, SWP_CONTINUED); | |
3479 | si->flags |= SWP_CONTINUED; | |
3480 | } | |
3481 | ||
3482 | list_for_each_entry(list_page, &head->lru, lru) { | |
3483 | unsigned char *map; | |
3484 | ||
3485 | /* | |
3486 | * If the previous map said no continuation, but we've found | |
3487 | * a continuation page, free our allocation and use this one. | |
3488 | */ | |
3489 | if (!(count & COUNT_CONTINUED)) | |
2628bd6f | 3490 | goto out_unlock_cont; |
570a335b | 3491 | |
9b04c5fe | 3492 | map = kmap_atomic(list_page) + offset; |
570a335b | 3493 | count = *map; |
9b04c5fe | 3494 | kunmap_atomic(map); |
570a335b HD |
3495 | |
3496 | /* | |
3497 | * If this continuation count now has some space in it, | |
3498 | * free our allocation and use this one. | |
3499 | */ | |
3500 | if ((count & ~COUNT_CONTINUED) != SWAP_CONT_MAX) | |
2628bd6f | 3501 | goto out_unlock_cont; |
570a335b HD |
3502 | } |
3503 | ||
3504 | list_add_tail(&page->lru, &head->lru); | |
3505 | page = NULL; /* now it's attached, don't free it */ | |
2628bd6f HY |
3506 | out_unlock_cont: |
3507 | spin_unlock(&si->cont_lock); | |
570a335b | 3508 | out: |
235b6217 | 3509 | unlock_cluster(ci); |
ec8acf20 | 3510 | spin_unlock(&si->lock); |
eb085574 | 3511 | put_swap_device(si); |
570a335b HD |
3512 | outer: |
3513 | if (page) | |
3514 | __free_page(page); | |
eb085574 | 3515 | return ret; |
570a335b HD |
3516 | } |
3517 | ||
3518 | /* | |
3519 | * swap_count_continued - when the original swap_map count is incremented | |
3520 | * from SWAP_MAP_MAX, check if there is already a continuation page to carry | |
3521 | * into, carry if so, or else fail until a new continuation page is allocated; | |
3522 | * when the original swap_map count is decremented from 0 with continuation, | |
3523 | * borrow from the continuation and report whether it still holds more. | |
235b6217 HY |
3524 | * Called while __swap_duplicate() or swap_entry_free() holds swap or cluster |
3525 | * lock. | |
570a335b HD |
3526 | */ |
3527 | static bool swap_count_continued(struct swap_info_struct *si, | |
3528 | pgoff_t offset, unsigned char count) | |
3529 | { | |
3530 | struct page *head; | |
3531 | struct page *page; | |
3532 | unsigned char *map; | |
2628bd6f | 3533 | bool ret; |
570a335b HD |
3534 | |
3535 | head = vmalloc_to_page(si->swap_map + offset); | |
3536 | if (page_private(head) != SWP_CONTINUED) { | |
3537 | BUG_ON(count & COUNT_CONTINUED); | |
3538 | return false; /* need to add count continuation */ | |
3539 | } | |
3540 | ||
2628bd6f | 3541 | spin_lock(&si->cont_lock); |
570a335b | 3542 | offset &= ~PAGE_MASK; |
213516ac | 3543 | page = list_next_entry(head, lru); |
9b04c5fe | 3544 | map = kmap_atomic(page) + offset; |
570a335b HD |
3545 | |
3546 | if (count == SWAP_MAP_MAX) /* initial increment from swap_map */ | |
3547 | goto init_map; /* jump over SWAP_CONT_MAX checks */ | |
3548 | ||
3549 | if (count == (SWAP_MAP_MAX | COUNT_CONTINUED)) { /* incrementing */ | |
3550 | /* | |
3551 | * Think of how you add 1 to 999 | |
3552 | */ | |
3553 | while (*map == (SWAP_CONT_MAX | COUNT_CONTINUED)) { | |
9b04c5fe | 3554 | kunmap_atomic(map); |
213516ac | 3555 | page = list_next_entry(page, lru); |
570a335b | 3556 | BUG_ON(page == head); |
9b04c5fe | 3557 | map = kmap_atomic(page) + offset; |
570a335b HD |
3558 | } |
3559 | if (*map == SWAP_CONT_MAX) { | |
9b04c5fe | 3560 | kunmap_atomic(map); |
213516ac | 3561 | page = list_next_entry(page, lru); |
2628bd6f HY |
3562 | if (page == head) { |
3563 | ret = false; /* add count continuation */ | |
3564 | goto out; | |
3565 | } | |
9b04c5fe | 3566 | map = kmap_atomic(page) + offset; |
570a335b HD |
3567 | init_map: *map = 0; /* we didn't zero the page */ |
3568 | } | |
3569 | *map += 1; | |
9b04c5fe | 3570 | kunmap_atomic(map); |
213516ac | 3571 | while ((page = list_prev_entry(page, lru)) != head) { |
9b04c5fe | 3572 | map = kmap_atomic(page) + offset; |
570a335b | 3573 | *map = COUNT_CONTINUED; |
9b04c5fe | 3574 | kunmap_atomic(map); |
570a335b | 3575 | } |
2628bd6f | 3576 | ret = true; /* incremented */ |
570a335b HD |
3577 | |
3578 | } else { /* decrementing */ | |
3579 | /* | |
3580 | * Think of how you subtract 1 from 1000 | |
3581 | */ | |
3582 | BUG_ON(count != COUNT_CONTINUED); | |
3583 | while (*map == COUNT_CONTINUED) { | |
9b04c5fe | 3584 | kunmap_atomic(map); |
213516ac | 3585 | page = list_next_entry(page, lru); |
570a335b | 3586 | BUG_ON(page == head); |
9b04c5fe | 3587 | map = kmap_atomic(page) + offset; |
570a335b HD |
3588 | } |
3589 | BUG_ON(*map == 0); | |
3590 | *map -= 1; | |
3591 | if (*map == 0) | |
3592 | count = 0; | |
9b04c5fe | 3593 | kunmap_atomic(map); |
213516ac | 3594 | while ((page = list_prev_entry(page, lru)) != head) { |
9b04c5fe | 3595 | map = kmap_atomic(page) + offset; |
570a335b HD |
3596 | *map = SWAP_CONT_MAX | count; |
3597 | count = COUNT_CONTINUED; | |
9b04c5fe | 3598 | kunmap_atomic(map); |
570a335b | 3599 | } |
2628bd6f | 3600 | ret = count == COUNT_CONTINUED; |
570a335b | 3601 | } |
2628bd6f HY |
3602 | out: |
3603 | spin_unlock(&si->cont_lock); | |
3604 | return ret; | |
570a335b HD |
3605 | } |
3606 | ||
3607 | /* | |
3608 | * free_swap_count_continuations - swapoff free all the continuation pages | |
3609 | * appended to the swap_map, after swap_map is quiesced, before vfree'ing it. | |
3610 | */ | |
3611 | static void free_swap_count_continuations(struct swap_info_struct *si) | |
3612 | { | |
3613 | pgoff_t offset; | |
3614 | ||
3615 | for (offset = 0; offset < si->max; offset += PAGE_SIZE) { | |
3616 | struct page *head; | |
3617 | head = vmalloc_to_page(si->swap_map + offset); | |
3618 | if (page_private(head)) { | |
0d576d20 GT |
3619 | struct page *page, *next; |
3620 | ||
3621 | list_for_each_entry_safe(page, next, &head->lru, lru) { | |
3622 | list_del(&page->lru); | |
570a335b HD |
3623 | __free_page(page); |
3624 | } | |
3625 | } | |
3626 | } | |
3627 | } | |
a2468cc9 | 3628 | |
2cf85583 | 3629 | #if defined(CONFIG_MEMCG) && defined(CONFIG_BLK_CGROUP) |
3e4fb13a | 3630 | void __folio_throttle_swaprate(struct folio *folio, gfp_t gfp) |
2cf85583 TH |
3631 | { |
3632 | struct swap_info_struct *si, *next; | |
3e4fb13a | 3633 | int nid = folio_nid(folio); |
6caa6a07 | 3634 | |
3e4fb13a | 3635 | if (!(gfp & __GFP_IO)) |
2cf85583 TH |
3636 | return; |
3637 | ||
3638 | if (!blk_cgroup_congested()) | |
3639 | return; | |
3640 | ||
3641 | /* | |
3642 | * We've already scheduled a throttle, avoid taking the global swap | |
3643 | * lock. | |
3644 | */ | |
f05837ed | 3645 | if (current->throttle_disk) |
2cf85583 TH |
3646 | return; |
3647 | ||
3648 | spin_lock(&swap_avail_lock); | |
6caa6a07 JW |
3649 | plist_for_each_entry_safe(si, next, &swap_avail_heads[nid], |
3650 | avail_lists[nid]) { | |
2cf85583 | 3651 | if (si->bdev) { |
de185b56 | 3652 | blkcg_schedule_throttle(si->bdev->bd_disk, true); |
2cf85583 TH |
3653 | break; |
3654 | } | |
3655 | } | |
3656 | spin_unlock(&swap_avail_lock); | |
3657 | } | |
3658 | #endif | |
3659 | ||
a2468cc9 AL |
3660 | static int __init swapfile_init(void) |
3661 | { | |
3662 | int nid; | |
3663 | ||
3664 | swap_avail_heads = kmalloc_array(nr_node_ids, sizeof(struct plist_head), | |
3665 | GFP_KERNEL); | |
3666 | if (!swap_avail_heads) { | |
3667 | pr_emerg("Not enough memory for swap heads, swap is disabled\n"); | |
3668 | return -ENOMEM; | |
3669 | } | |
3670 | ||
3671 | for_each_node(nid) | |
3672 | plist_head_init(&swap_avail_heads[nid]); | |
3673 | ||
be45a490 PX |
3674 | swapfile_maximum_size = arch_max_swapfile_size(); |
3675 | ||
5154e607 PX |
3676 | #ifdef CONFIG_MIGRATION |
3677 | if (swapfile_maximum_size >= (1UL << SWP_MIG_TOTAL_BITS)) | |
3678 | swap_migration_ad_supported = true; | |
3679 | #endif /* CONFIG_MIGRATION */ | |
3680 | ||
a2468cc9 AL |
3681 | return 0; |
3682 | } | |
3683 | subsys_initcall(swapfile_init); |