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 | 38 | #include <linux/swapfile.h> |
f981c595 | 39 | #include <linux/export.h> |
67afa38e | 40 | #include <linux/swap_slots.h> |
155b5f88 | 41 | #include <linux/sort.h> |
63d8620e | 42 | #include <linux/completion.h> |
07f44ac3 | 43 | #include <linux/suspend.h> |
42c06a0e | 44 | #include <linux/zswap.h> |
1da177e4 | 45 | |
1da177e4 LT |
46 | #include <asm/tlbflush.h> |
47 | #include <linux/swapops.h> | |
5d1ea48b | 48 | #include <linux/swap_cgroup.h> |
00cde042 | 49 | #include "internal.h" |
014bb1de | 50 | #include "swap.h" |
1da177e4 | 51 | |
570a335b HD |
52 | static bool swap_count_continued(struct swap_info_struct *, pgoff_t, |
53 | unsigned char); | |
54 | static void free_swap_count_continuations(struct swap_info_struct *); | |
55 | ||
633423a0 | 56 | static DEFINE_SPINLOCK(swap_lock); |
7c363b8c | 57 | static unsigned int nr_swapfiles; |
ec8acf20 | 58 | atomic_long_t nr_swap_pages; |
fb0fec50 CW |
59 | /* |
60 | * Some modules use swappable objects and may try to swap them out under | |
61 | * memory pressure (via the shrinker). Before doing so, they may wish to | |
62 | * check to see if any swap space is available. | |
63 | */ | |
64 | EXPORT_SYMBOL_GPL(nr_swap_pages); | |
ec8acf20 | 65 | /* protected with swap_lock. reading in vm_swap_full() doesn't need lock */ |
1da177e4 | 66 | long total_swap_pages; |
a2468cc9 | 67 | static int least_priority = -1; |
be45a490 | 68 | unsigned long swapfile_maximum_size; |
5154e607 PX |
69 | #ifdef CONFIG_MIGRATION |
70 | bool swap_migration_ad_supported; | |
71 | #endif /* CONFIG_MIGRATION */ | |
1da177e4 | 72 | |
1da177e4 LT |
73 | static const char Bad_file[] = "Bad swap file entry "; |
74 | static const char Unused_file[] = "Unused swap file entry "; | |
75 | static const char Bad_offset[] = "Bad swap offset entry "; | |
76 | static const char Unused_offset[] = "Unused swap offset entry "; | |
77 | ||
adfab836 DS |
78 | /* |
79 | * all active swap_info_structs | |
80 | * protected with swap_lock, and ordered by priority. | |
81 | */ | |
633423a0 | 82 | static PLIST_HEAD(swap_active_head); |
18ab4d4c DS |
83 | |
84 | /* | |
85 | * all available (active, not full) swap_info_structs | |
86 | * protected with swap_avail_lock, ordered by priority. | |
e2e3fdc7 | 87 | * This is used by folio_alloc_swap() instead of swap_active_head |
18ab4d4c | 88 | * because swap_active_head includes all swap_info_structs, |
e2e3fdc7 | 89 | * but folio_alloc_swap() doesn't need to look at full ones. |
18ab4d4c DS |
90 | * This uses its own lock instead of swap_lock because when a |
91 | * swap_info_struct changes between not-full/full, it needs to | |
92 | * add/remove itself to/from this list, but the swap_info_struct->lock | |
93 | * is held and the locking order requires swap_lock to be taken | |
94 | * before any swap_info_struct->lock. | |
95 | */ | |
bfc6b1ca | 96 | static struct plist_head *swap_avail_heads; |
18ab4d4c | 97 | static DEFINE_SPINLOCK(swap_avail_lock); |
1da177e4 | 98 | |
42c06a0e | 99 | static struct swap_info_struct *swap_info[MAX_SWAPFILES]; |
1da177e4 | 100 | |
fc0abb14 | 101 | static DEFINE_MUTEX(swapon_mutex); |
1da177e4 | 102 | |
66d7dd51 KS |
103 | static DECLARE_WAIT_QUEUE_HEAD(proc_poll_wait); |
104 | /* Activity counter to indicate that a swapon or swapoff has occurred */ | |
105 | static atomic_t proc_poll_event = ATOMIC_INIT(0); | |
106 | ||
81a0298b HY |
107 | atomic_t nr_rotate_swap = ATOMIC_INIT(0); |
108 | ||
c10d38cc DJ |
109 | static struct swap_info_struct *swap_type_to_swap_info(int type) |
110 | { | |
a4b45114 | 111 | if (type >= MAX_SWAPFILES) |
c10d38cc DJ |
112 | return NULL; |
113 | ||
a4b45114 | 114 | return READ_ONCE(swap_info[type]); /* rcu_dereference() */ |
c10d38cc DJ |
115 | } |
116 | ||
8d69aaee | 117 | static inline unsigned char swap_count(unsigned char ent) |
355cfa73 | 118 | { |
955c97f0 | 119 | return ent & ~SWAP_HAS_CACHE; /* may include COUNT_CONTINUED flag */ |
355cfa73 KH |
120 | } |
121 | ||
bcd49e86 HY |
122 | /* Reclaim the swap entry anyway if possible */ |
123 | #define TTRS_ANYWAY 0x1 | |
124 | /* | |
125 | * Reclaim the swap entry if there are no more mappings of the | |
126 | * corresponding page | |
127 | */ | |
128 | #define TTRS_UNMAPPED 0x2 | |
129 | /* Reclaim the swap entry if swap is getting full*/ | |
130 | #define TTRS_FULL 0x4 | |
131 | ||
efa90a98 | 132 | /* returns 1 if swap entry is freed */ |
bcd49e86 HY |
133 | static int __try_to_reclaim_swap(struct swap_info_struct *si, |
134 | unsigned long offset, unsigned long flags) | |
c9e44410 | 135 | { |
efa90a98 | 136 | swp_entry_t entry = swp_entry(si->type, offset); |
2c3f6194 | 137 | struct folio *folio; |
c9e44410 KH |
138 | int ret = 0; |
139 | ||
2c3f6194 | 140 | folio = filemap_get_folio(swap_address_space(entry), offset); |
66dabbb6 | 141 | if (IS_ERR(folio)) |
c9e44410 KH |
142 | return 0; |
143 | /* | |
bcd49e86 | 144 | * When this function is called from scan_swap_map_slots() and it's |
2c3f6194 | 145 | * called by vmscan.c at reclaiming folios. So we hold a folio lock |
bcd49e86 | 146 | * here. We have to use trylock for avoiding deadlock. This is a special |
2c3f6194 | 147 | * case and you should use folio_free_swap() with explicit folio_lock() |
c9e44410 KH |
148 | * in usual operations. |
149 | */ | |
2c3f6194 | 150 | if (folio_trylock(folio)) { |
bcd49e86 | 151 | if ((flags & TTRS_ANYWAY) || |
2c3f6194 | 152 | ((flags & TTRS_UNMAPPED) && !folio_mapped(folio)) || |
9202d527 | 153 | ((flags & TTRS_FULL) && mem_cgroup_swap_full(folio))) |
2c3f6194 MWO |
154 | ret = folio_free_swap(folio); |
155 | folio_unlock(folio); | |
c9e44410 | 156 | } |
2c3f6194 | 157 | folio_put(folio); |
c9e44410 KH |
158 | return ret; |
159 | } | |
355cfa73 | 160 | |
4efaceb1 AL |
161 | static inline struct swap_extent *first_se(struct swap_info_struct *sis) |
162 | { | |
163 | struct rb_node *rb = rb_first(&sis->swap_extent_root); | |
164 | return rb_entry(rb, struct swap_extent, rb_node); | |
165 | } | |
166 | ||
167 | static inline struct swap_extent *next_se(struct swap_extent *se) | |
168 | { | |
169 | struct rb_node *rb = rb_next(&se->rb_node); | |
170 | return rb ? rb_entry(rb, struct swap_extent, rb_node) : NULL; | |
171 | } | |
172 | ||
6a6ba831 HD |
173 | /* |
174 | * swapon tell device that all the old swap contents can be discarded, | |
175 | * to allow the swap device to optimize its wear-levelling. | |
176 | */ | |
177 | static int discard_swap(struct swap_info_struct *si) | |
178 | { | |
179 | struct swap_extent *se; | |
9625a5f2 HD |
180 | sector_t start_block; |
181 | sector_t nr_blocks; | |
6a6ba831 HD |
182 | int err = 0; |
183 | ||
9625a5f2 | 184 | /* Do not discard the swap header page! */ |
4efaceb1 | 185 | se = first_se(si); |
9625a5f2 HD |
186 | start_block = (se->start_block + 1) << (PAGE_SHIFT - 9); |
187 | nr_blocks = ((sector_t)se->nr_pages - 1) << (PAGE_SHIFT - 9); | |
188 | if (nr_blocks) { | |
189 | err = blkdev_issue_discard(si->bdev, start_block, | |
44abff2c | 190 | nr_blocks, GFP_KERNEL); |
9625a5f2 HD |
191 | if (err) |
192 | return err; | |
193 | cond_resched(); | |
194 | } | |
6a6ba831 | 195 | |
4efaceb1 | 196 | for (se = next_se(se); se; se = next_se(se)) { |
9625a5f2 HD |
197 | start_block = se->start_block << (PAGE_SHIFT - 9); |
198 | nr_blocks = (sector_t)se->nr_pages << (PAGE_SHIFT - 9); | |
6a6ba831 HD |
199 | |
200 | err = blkdev_issue_discard(si->bdev, start_block, | |
44abff2c | 201 | nr_blocks, GFP_KERNEL); |
6a6ba831 HD |
202 | if (err) |
203 | break; | |
204 | ||
205 | cond_resched(); | |
206 | } | |
207 | return err; /* That will often be -EOPNOTSUPP */ | |
208 | } | |
209 | ||
4efaceb1 AL |
210 | static struct swap_extent * |
211 | offset_to_swap_extent(struct swap_info_struct *sis, unsigned long offset) | |
212 | { | |
213 | struct swap_extent *se; | |
214 | struct rb_node *rb; | |
215 | ||
216 | rb = sis->swap_extent_root.rb_node; | |
217 | while (rb) { | |
218 | se = rb_entry(rb, struct swap_extent, rb_node); | |
219 | if (offset < se->start_page) | |
220 | rb = rb->rb_left; | |
221 | else if (offset >= se->start_page + se->nr_pages) | |
222 | rb = rb->rb_right; | |
223 | else | |
224 | return se; | |
225 | } | |
226 | /* It *must* be present */ | |
227 | BUG(); | |
228 | } | |
229 | ||
caf6912f JA |
230 | sector_t swap_page_sector(struct page *page) |
231 | { | |
232 | struct swap_info_struct *sis = page_swap_info(page); | |
233 | struct swap_extent *se; | |
234 | sector_t sector; | |
235 | pgoff_t offset; | |
236 | ||
237 | offset = __page_file_index(page); | |
238 | se = offset_to_swap_extent(sis, offset); | |
239 | sector = se->start_block + (offset - se->start_page); | |
240 | return sector << (PAGE_SHIFT - 9); | |
241 | } | |
242 | ||
7992fde7 HD |
243 | /* |
244 | * swap allocation tell device that a cluster of swap can now be discarded, | |
245 | * to allow the swap device to optimize its wear-levelling. | |
246 | */ | |
247 | static void discard_swap_cluster(struct swap_info_struct *si, | |
248 | pgoff_t start_page, pgoff_t nr_pages) | |
249 | { | |
4efaceb1 | 250 | struct swap_extent *se = offset_to_swap_extent(si, start_page); |
7992fde7 HD |
251 | |
252 | while (nr_pages) { | |
4efaceb1 AL |
253 | pgoff_t offset = start_page - se->start_page; |
254 | sector_t start_block = se->start_block + offset; | |
255 | sector_t nr_blocks = se->nr_pages - offset; | |
256 | ||
257 | if (nr_blocks > nr_pages) | |
258 | nr_blocks = nr_pages; | |
259 | start_page += nr_blocks; | |
260 | nr_pages -= nr_blocks; | |
261 | ||
262 | start_block <<= PAGE_SHIFT - 9; | |
263 | nr_blocks <<= PAGE_SHIFT - 9; | |
264 | if (blkdev_issue_discard(si->bdev, start_block, | |
44abff2c | 265 | nr_blocks, GFP_NOIO)) |
4efaceb1 | 266 | break; |
7992fde7 | 267 | |
4efaceb1 | 268 | se = next_se(se); |
7992fde7 HD |
269 | } |
270 | } | |
271 | ||
38d8b4e6 HY |
272 | #ifdef CONFIG_THP_SWAP |
273 | #define SWAPFILE_CLUSTER HPAGE_PMD_NR | |
a448f2d0 HY |
274 | |
275 | #define swap_entry_size(size) (size) | |
38d8b4e6 | 276 | #else |
048c27fd | 277 | #define SWAPFILE_CLUSTER 256 |
a448f2d0 HY |
278 | |
279 | /* | |
280 | * Define swap_entry_size() as constant to let compiler to optimize | |
281 | * out some code if !CONFIG_THP_SWAP | |
282 | */ | |
283 | #define swap_entry_size(size) 1 | |
38d8b4e6 | 284 | #endif |
048c27fd HD |
285 | #define LATENCY_LIMIT 256 |
286 | ||
2a8f9449 SL |
287 | static inline void cluster_set_flag(struct swap_cluster_info *info, |
288 | unsigned int flag) | |
289 | { | |
290 | info->flags = flag; | |
291 | } | |
292 | ||
293 | static inline unsigned int cluster_count(struct swap_cluster_info *info) | |
294 | { | |
295 | return info->data; | |
296 | } | |
297 | ||
298 | static inline void cluster_set_count(struct swap_cluster_info *info, | |
299 | unsigned int c) | |
300 | { | |
301 | info->data = c; | |
302 | } | |
303 | ||
304 | static inline void cluster_set_count_flag(struct swap_cluster_info *info, | |
305 | unsigned int c, unsigned int f) | |
306 | { | |
307 | info->flags = f; | |
308 | info->data = c; | |
309 | } | |
310 | ||
311 | static inline unsigned int cluster_next(struct swap_cluster_info *info) | |
312 | { | |
313 | return info->data; | |
314 | } | |
315 | ||
316 | static inline void cluster_set_next(struct swap_cluster_info *info, | |
317 | unsigned int n) | |
318 | { | |
319 | info->data = n; | |
320 | } | |
321 | ||
322 | static inline void cluster_set_next_flag(struct swap_cluster_info *info, | |
323 | unsigned int n, unsigned int f) | |
324 | { | |
325 | info->flags = f; | |
326 | info->data = n; | |
327 | } | |
328 | ||
329 | static inline bool cluster_is_free(struct swap_cluster_info *info) | |
330 | { | |
331 | return info->flags & CLUSTER_FLAG_FREE; | |
332 | } | |
333 | ||
334 | static inline bool cluster_is_null(struct swap_cluster_info *info) | |
335 | { | |
336 | return info->flags & CLUSTER_FLAG_NEXT_NULL; | |
337 | } | |
338 | ||
339 | static inline void cluster_set_null(struct swap_cluster_info *info) | |
340 | { | |
341 | info->flags = CLUSTER_FLAG_NEXT_NULL; | |
342 | info->data = 0; | |
343 | } | |
344 | ||
e0709829 HY |
345 | static inline bool cluster_is_huge(struct swap_cluster_info *info) |
346 | { | |
33ee011e HY |
347 | if (IS_ENABLED(CONFIG_THP_SWAP)) |
348 | return info->flags & CLUSTER_FLAG_HUGE; | |
349 | return false; | |
e0709829 HY |
350 | } |
351 | ||
352 | static inline void cluster_clear_huge(struct swap_cluster_info *info) | |
353 | { | |
354 | info->flags &= ~CLUSTER_FLAG_HUGE; | |
355 | } | |
356 | ||
235b6217 HY |
357 | static inline struct swap_cluster_info *lock_cluster(struct swap_info_struct *si, |
358 | unsigned long offset) | |
359 | { | |
360 | struct swap_cluster_info *ci; | |
361 | ||
362 | ci = si->cluster_info; | |
363 | if (ci) { | |
364 | ci += offset / SWAPFILE_CLUSTER; | |
365 | spin_lock(&ci->lock); | |
366 | } | |
367 | return ci; | |
368 | } | |
369 | ||
370 | static inline void unlock_cluster(struct swap_cluster_info *ci) | |
371 | { | |
372 | if (ci) | |
373 | spin_unlock(&ci->lock); | |
374 | } | |
375 | ||
59d98bf3 HY |
376 | /* |
377 | * Determine the locking method in use for this device. Return | |
378 | * swap_cluster_info if SSD-style cluster-based locking is in place. | |
379 | */ | |
235b6217 | 380 | static inline struct swap_cluster_info *lock_cluster_or_swap_info( |
59d98bf3 | 381 | struct swap_info_struct *si, unsigned long offset) |
235b6217 HY |
382 | { |
383 | struct swap_cluster_info *ci; | |
384 | ||
59d98bf3 | 385 | /* Try to use fine-grained SSD-style locking if available: */ |
235b6217 | 386 | ci = lock_cluster(si, offset); |
59d98bf3 | 387 | /* Otherwise, fall back to traditional, coarse locking: */ |
235b6217 HY |
388 | if (!ci) |
389 | spin_lock(&si->lock); | |
390 | ||
391 | return ci; | |
392 | } | |
393 | ||
394 | static inline void unlock_cluster_or_swap_info(struct swap_info_struct *si, | |
395 | struct swap_cluster_info *ci) | |
396 | { | |
397 | if (ci) | |
398 | unlock_cluster(ci); | |
399 | else | |
400 | spin_unlock(&si->lock); | |
401 | } | |
402 | ||
6b534915 HY |
403 | static inline bool cluster_list_empty(struct swap_cluster_list *list) |
404 | { | |
405 | return cluster_is_null(&list->head); | |
406 | } | |
407 | ||
408 | static inline unsigned int cluster_list_first(struct swap_cluster_list *list) | |
409 | { | |
410 | return cluster_next(&list->head); | |
411 | } | |
412 | ||
413 | static void cluster_list_init(struct swap_cluster_list *list) | |
414 | { | |
415 | cluster_set_null(&list->head); | |
416 | cluster_set_null(&list->tail); | |
417 | } | |
418 | ||
419 | static void cluster_list_add_tail(struct swap_cluster_list *list, | |
420 | struct swap_cluster_info *ci, | |
421 | unsigned int idx) | |
422 | { | |
423 | if (cluster_list_empty(list)) { | |
424 | cluster_set_next_flag(&list->head, idx, 0); | |
425 | cluster_set_next_flag(&list->tail, idx, 0); | |
426 | } else { | |
235b6217 | 427 | struct swap_cluster_info *ci_tail; |
6b534915 HY |
428 | unsigned int tail = cluster_next(&list->tail); |
429 | ||
235b6217 HY |
430 | /* |
431 | * Nested cluster lock, but both cluster locks are | |
432 | * only acquired when we held swap_info_struct->lock | |
433 | */ | |
434 | ci_tail = ci + tail; | |
435 | spin_lock_nested(&ci_tail->lock, SINGLE_DEPTH_NESTING); | |
436 | cluster_set_next(ci_tail, idx); | |
0ef017d1 | 437 | spin_unlock(&ci_tail->lock); |
6b534915 HY |
438 | cluster_set_next_flag(&list->tail, idx, 0); |
439 | } | |
440 | } | |
441 | ||
442 | static unsigned int cluster_list_del_first(struct swap_cluster_list *list, | |
443 | struct swap_cluster_info *ci) | |
444 | { | |
445 | unsigned int idx; | |
446 | ||
447 | idx = cluster_next(&list->head); | |
448 | if (cluster_next(&list->tail) == idx) { | |
449 | cluster_set_null(&list->head); | |
450 | cluster_set_null(&list->tail); | |
451 | } else | |
452 | cluster_set_next_flag(&list->head, | |
453 | cluster_next(&ci[idx]), 0); | |
454 | ||
455 | return idx; | |
456 | } | |
457 | ||
815c2c54 SL |
458 | /* Add a cluster to discard list and schedule it to do discard */ |
459 | static void swap_cluster_schedule_discard(struct swap_info_struct *si, | |
460 | unsigned int idx) | |
461 | { | |
462 | /* | |
bb243f7d | 463 | * If scan_swap_map_slots() can't find a free cluster, it will check |
815c2c54 | 464 | * si->swap_map directly. To make sure the discarding cluster isn't |
bb243f7d ML |
465 | * taken by scan_swap_map_slots(), mark the swap entries bad (occupied). |
466 | * It will be cleared after discard | |
815c2c54 SL |
467 | */ |
468 | memset(si->swap_map + idx * SWAPFILE_CLUSTER, | |
469 | SWAP_MAP_BAD, SWAPFILE_CLUSTER); | |
470 | ||
6b534915 | 471 | cluster_list_add_tail(&si->discard_clusters, si->cluster_info, idx); |
815c2c54 SL |
472 | |
473 | schedule_work(&si->discard_work); | |
474 | } | |
475 | ||
38d8b4e6 HY |
476 | static void __free_cluster(struct swap_info_struct *si, unsigned long idx) |
477 | { | |
478 | struct swap_cluster_info *ci = si->cluster_info; | |
479 | ||
480 | cluster_set_flag(ci + idx, CLUSTER_FLAG_FREE); | |
481 | cluster_list_add_tail(&si->free_clusters, ci, idx); | |
482 | } | |
483 | ||
815c2c54 SL |
484 | /* |
485 | * Doing discard actually. After a cluster discard is finished, the cluster | |
486 | * will be added to free cluster list. caller should hold si->lock. | |
487 | */ | |
488 | static void swap_do_scheduled_discard(struct swap_info_struct *si) | |
489 | { | |
235b6217 | 490 | struct swap_cluster_info *info, *ci; |
815c2c54 SL |
491 | unsigned int idx; |
492 | ||
493 | info = si->cluster_info; | |
494 | ||
6b534915 HY |
495 | while (!cluster_list_empty(&si->discard_clusters)) { |
496 | idx = cluster_list_del_first(&si->discard_clusters, info); | |
815c2c54 SL |
497 | spin_unlock(&si->lock); |
498 | ||
499 | discard_swap_cluster(si, idx * SWAPFILE_CLUSTER, | |
500 | SWAPFILE_CLUSTER); | |
501 | ||
502 | spin_lock(&si->lock); | |
235b6217 | 503 | ci = lock_cluster(si, idx * SWAPFILE_CLUSTER); |
38d8b4e6 | 504 | __free_cluster(si, idx); |
815c2c54 SL |
505 | memset(si->swap_map + idx * SWAPFILE_CLUSTER, |
506 | 0, SWAPFILE_CLUSTER); | |
235b6217 | 507 | unlock_cluster(ci); |
815c2c54 SL |
508 | } |
509 | } | |
510 | ||
511 | static void swap_discard_work(struct work_struct *work) | |
512 | { | |
513 | struct swap_info_struct *si; | |
514 | ||
515 | si = container_of(work, struct swap_info_struct, discard_work); | |
516 | ||
517 | spin_lock(&si->lock); | |
518 | swap_do_scheduled_discard(si); | |
519 | spin_unlock(&si->lock); | |
520 | } | |
521 | ||
63d8620e ML |
522 | static void swap_users_ref_free(struct percpu_ref *ref) |
523 | { | |
524 | struct swap_info_struct *si; | |
525 | ||
526 | si = container_of(ref, struct swap_info_struct, users); | |
527 | complete(&si->comp); | |
528 | } | |
529 | ||
38d8b4e6 HY |
530 | static void alloc_cluster(struct swap_info_struct *si, unsigned long idx) |
531 | { | |
532 | struct swap_cluster_info *ci = si->cluster_info; | |
533 | ||
534 | VM_BUG_ON(cluster_list_first(&si->free_clusters) != idx); | |
535 | cluster_list_del_first(&si->free_clusters, ci); | |
536 | cluster_set_count_flag(ci + idx, 0, 0); | |
537 | } | |
538 | ||
539 | static void free_cluster(struct swap_info_struct *si, unsigned long idx) | |
540 | { | |
541 | struct swap_cluster_info *ci = si->cluster_info + idx; | |
542 | ||
543 | VM_BUG_ON(cluster_count(ci) != 0); | |
544 | /* | |
545 | * If the swap is discardable, prepare discard the cluster | |
546 | * instead of free it immediately. The cluster will be freed | |
547 | * after discard. | |
548 | */ | |
549 | if ((si->flags & (SWP_WRITEOK | SWP_PAGE_DISCARD)) == | |
550 | (SWP_WRITEOK | SWP_PAGE_DISCARD)) { | |
551 | swap_cluster_schedule_discard(si, idx); | |
552 | return; | |
553 | } | |
554 | ||
555 | __free_cluster(si, idx); | |
556 | } | |
557 | ||
2a8f9449 SL |
558 | /* |
559 | * The cluster corresponding to page_nr will be used. The cluster will be | |
560 | * removed from free cluster list and its usage counter will be increased. | |
561 | */ | |
562 | static void inc_cluster_info_page(struct swap_info_struct *p, | |
563 | struct swap_cluster_info *cluster_info, unsigned long page_nr) | |
564 | { | |
565 | unsigned long idx = page_nr / SWAPFILE_CLUSTER; | |
566 | ||
567 | if (!cluster_info) | |
568 | return; | |
38d8b4e6 HY |
569 | if (cluster_is_free(&cluster_info[idx])) |
570 | alloc_cluster(p, idx); | |
2a8f9449 SL |
571 | |
572 | VM_BUG_ON(cluster_count(&cluster_info[idx]) >= SWAPFILE_CLUSTER); | |
573 | cluster_set_count(&cluster_info[idx], | |
574 | cluster_count(&cluster_info[idx]) + 1); | |
575 | } | |
576 | ||
577 | /* | |
578 | * The cluster corresponding to page_nr decreases one usage. If the usage | |
579 | * counter becomes 0, which means no page in the cluster is in using, we can | |
580 | * optionally discard the cluster and add it to free cluster list. | |
581 | */ | |
582 | static void dec_cluster_info_page(struct swap_info_struct *p, | |
583 | struct swap_cluster_info *cluster_info, unsigned long page_nr) | |
584 | { | |
585 | unsigned long idx = page_nr / SWAPFILE_CLUSTER; | |
586 | ||
587 | if (!cluster_info) | |
588 | return; | |
589 | ||
590 | VM_BUG_ON(cluster_count(&cluster_info[idx]) == 0); | |
591 | cluster_set_count(&cluster_info[idx], | |
592 | cluster_count(&cluster_info[idx]) - 1); | |
593 | ||
38d8b4e6 HY |
594 | if (cluster_count(&cluster_info[idx]) == 0) |
595 | free_cluster(p, idx); | |
2a8f9449 SL |
596 | } |
597 | ||
598 | /* | |
bb243f7d | 599 | * It's possible scan_swap_map_slots() uses a free cluster in the middle of free |
2a8f9449 SL |
600 | * cluster list. Avoiding such abuse to avoid list corruption. |
601 | */ | |
ebc2a1a6 SL |
602 | static bool |
603 | scan_swap_map_ssd_cluster_conflict(struct swap_info_struct *si, | |
2a8f9449 SL |
604 | unsigned long offset) |
605 | { | |
ebc2a1a6 SL |
606 | struct percpu_cluster *percpu_cluster; |
607 | bool conflict; | |
608 | ||
2a8f9449 | 609 | offset /= SWAPFILE_CLUSTER; |
6b534915 HY |
610 | conflict = !cluster_list_empty(&si->free_clusters) && |
611 | offset != cluster_list_first(&si->free_clusters) && | |
2a8f9449 | 612 | cluster_is_free(&si->cluster_info[offset]); |
ebc2a1a6 SL |
613 | |
614 | if (!conflict) | |
615 | return false; | |
616 | ||
617 | percpu_cluster = this_cpu_ptr(si->percpu_cluster); | |
618 | cluster_set_null(&percpu_cluster->index); | |
619 | return true; | |
620 | } | |
621 | ||
622 | /* | |
623 | * Try to get a swap entry from current cpu's swap entry pool (a cluster). This | |
624 | * might involve allocating a new cluster for current CPU too. | |
625 | */ | |
36005bae | 626 | static bool scan_swap_map_try_ssd_cluster(struct swap_info_struct *si, |
ebc2a1a6 SL |
627 | unsigned long *offset, unsigned long *scan_base) |
628 | { | |
629 | struct percpu_cluster *cluster; | |
235b6217 | 630 | struct swap_cluster_info *ci; |
235b6217 | 631 | unsigned long tmp, max; |
ebc2a1a6 SL |
632 | |
633 | new_cluster: | |
634 | cluster = this_cpu_ptr(si->percpu_cluster); | |
635 | if (cluster_is_null(&cluster->index)) { | |
6b534915 HY |
636 | if (!cluster_list_empty(&si->free_clusters)) { |
637 | cluster->index = si->free_clusters.head; | |
ebc2a1a6 SL |
638 | cluster->next = cluster_next(&cluster->index) * |
639 | SWAPFILE_CLUSTER; | |
6b534915 | 640 | } else if (!cluster_list_empty(&si->discard_clusters)) { |
ebc2a1a6 SL |
641 | /* |
642 | * we don't have free cluster but have some clusters in | |
49070588 HY |
643 | * discarding, do discard now and reclaim them, then |
644 | * reread cluster_next_cpu since we dropped si->lock | |
ebc2a1a6 SL |
645 | */ |
646 | swap_do_scheduled_discard(si); | |
49070588 HY |
647 | *scan_base = this_cpu_read(*si->cluster_next_cpu); |
648 | *offset = *scan_base; | |
ebc2a1a6 SL |
649 | goto new_cluster; |
650 | } else | |
36005bae | 651 | return false; |
ebc2a1a6 SL |
652 | } |
653 | ||
ebc2a1a6 SL |
654 | /* |
655 | * Other CPUs can use our cluster if they can't find a free cluster, | |
656 | * check if there is still free entry in the cluster | |
657 | */ | |
658 | tmp = cluster->next; | |
235b6217 HY |
659 | max = min_t(unsigned long, si->max, |
660 | (cluster_next(&cluster->index) + 1) * SWAPFILE_CLUSTER); | |
7b9e2de1 WY |
661 | if (tmp < max) { |
662 | ci = lock_cluster(si, tmp); | |
663 | while (tmp < max) { | |
664 | if (!si->swap_map[tmp]) | |
665 | break; | |
666 | tmp++; | |
667 | } | |
668 | unlock_cluster(ci); | |
ebc2a1a6 | 669 | } |
0fd0e19e | 670 | if (tmp >= max) { |
ebc2a1a6 SL |
671 | cluster_set_null(&cluster->index); |
672 | goto new_cluster; | |
673 | } | |
674 | cluster->next = tmp + 1; | |
675 | *offset = tmp; | |
676 | *scan_base = tmp; | |
fdff1deb | 677 | return true; |
2a8f9449 SL |
678 | } |
679 | ||
a2468cc9 AL |
680 | static void __del_from_avail_list(struct swap_info_struct *p) |
681 | { | |
682 | int nid; | |
683 | ||
6fe7d6b9 | 684 | assert_spin_locked(&p->lock); |
a2468cc9 AL |
685 | for_each_node(nid) |
686 | plist_del(&p->avail_lists[nid], &swap_avail_heads[nid]); | |
687 | } | |
688 | ||
689 | static void del_from_avail_list(struct swap_info_struct *p) | |
690 | { | |
691 | spin_lock(&swap_avail_lock); | |
692 | __del_from_avail_list(p); | |
693 | spin_unlock(&swap_avail_lock); | |
694 | } | |
695 | ||
38d8b4e6 HY |
696 | static void swap_range_alloc(struct swap_info_struct *si, unsigned long offset, |
697 | unsigned int nr_entries) | |
698 | { | |
699 | unsigned int end = offset + nr_entries - 1; | |
700 | ||
701 | if (offset == si->lowest_bit) | |
702 | si->lowest_bit += nr_entries; | |
703 | if (end == si->highest_bit) | |
a449bf58 | 704 | WRITE_ONCE(si->highest_bit, si->highest_bit - nr_entries); |
c8945306 | 705 | WRITE_ONCE(si->inuse_pages, si->inuse_pages + nr_entries); |
38d8b4e6 HY |
706 | if (si->inuse_pages == si->pages) { |
707 | si->lowest_bit = si->max; | |
708 | si->highest_bit = 0; | |
a2468cc9 | 709 | del_from_avail_list(si); |
38d8b4e6 HY |
710 | } |
711 | } | |
712 | ||
a2468cc9 AL |
713 | static void add_to_avail_list(struct swap_info_struct *p) |
714 | { | |
715 | int nid; | |
716 | ||
717 | spin_lock(&swap_avail_lock); | |
67490031 | 718 | for_each_node(nid) |
a2468cc9 | 719 | plist_add(&p->avail_lists[nid], &swap_avail_heads[nid]); |
a2468cc9 AL |
720 | spin_unlock(&swap_avail_lock); |
721 | } | |
722 | ||
38d8b4e6 HY |
723 | static void swap_range_free(struct swap_info_struct *si, unsigned long offset, |
724 | unsigned int nr_entries) | |
725 | { | |
3852f676 | 726 | unsigned long begin = offset; |
38d8b4e6 HY |
727 | unsigned long end = offset + nr_entries - 1; |
728 | void (*swap_slot_free_notify)(struct block_device *, unsigned long); | |
729 | ||
730 | if (offset < si->lowest_bit) | |
731 | si->lowest_bit = offset; | |
732 | if (end > si->highest_bit) { | |
733 | bool was_full = !si->highest_bit; | |
734 | ||
a449bf58 | 735 | WRITE_ONCE(si->highest_bit, end); |
a2468cc9 AL |
736 | if (was_full && (si->flags & SWP_WRITEOK)) |
737 | add_to_avail_list(si); | |
38d8b4e6 HY |
738 | } |
739 | atomic_long_add(nr_entries, &nr_swap_pages); | |
c8945306 | 740 | WRITE_ONCE(si->inuse_pages, si->inuse_pages - nr_entries); |
38d8b4e6 HY |
741 | if (si->flags & SWP_BLKDEV) |
742 | swap_slot_free_notify = | |
743 | si->bdev->bd_disk->fops->swap_slot_free_notify; | |
744 | else | |
745 | swap_slot_free_notify = NULL; | |
746 | while (offset <= end) { | |
8a84802e | 747 | arch_swap_invalidate_page(si->type, offset); |
42c06a0e | 748 | zswap_invalidate(si->type, offset); |
38d8b4e6 HY |
749 | if (swap_slot_free_notify) |
750 | swap_slot_free_notify(si->bdev, offset); | |
751 | offset++; | |
752 | } | |
3852f676 | 753 | clear_shadow_from_swap_cache(si->type, begin, end); |
38d8b4e6 HY |
754 | } |
755 | ||
49070588 HY |
756 | static void set_cluster_next(struct swap_info_struct *si, unsigned long next) |
757 | { | |
758 | unsigned long prev; | |
759 | ||
760 | if (!(si->flags & SWP_SOLIDSTATE)) { | |
761 | si->cluster_next = next; | |
762 | return; | |
763 | } | |
764 | ||
765 | prev = this_cpu_read(*si->cluster_next_cpu); | |
766 | /* | |
767 | * Cross the swap address space size aligned trunk, choose | |
768 | * another trunk randomly to avoid lock contention on swap | |
769 | * address space if possible. | |
770 | */ | |
771 | if ((prev >> SWAP_ADDRESS_SPACE_SHIFT) != | |
772 | (next >> SWAP_ADDRESS_SPACE_SHIFT)) { | |
773 | /* No free swap slots available */ | |
774 | if (si->highest_bit <= si->lowest_bit) | |
775 | return; | |
e8a533cb | 776 | next = get_random_u32_inclusive(si->lowest_bit, si->highest_bit); |
49070588 HY |
777 | next = ALIGN_DOWN(next, SWAP_ADDRESS_SPACE_PAGES); |
778 | next = max_t(unsigned int, next, si->lowest_bit); | |
779 | } | |
780 | this_cpu_write(*si->cluster_next_cpu, next); | |
781 | } | |
782 | ||
4b9ae842 ML |
783 | static bool swap_offset_available_and_locked(struct swap_info_struct *si, |
784 | unsigned long offset) | |
785 | { | |
786 | if (data_race(!si->swap_map[offset])) { | |
787 | spin_lock(&si->lock); | |
788 | return true; | |
789 | } | |
790 | ||
791 | if (vm_swap_full() && READ_ONCE(si->swap_map[offset]) == SWAP_HAS_CACHE) { | |
792 | spin_lock(&si->lock); | |
793 | return true; | |
794 | } | |
795 | ||
796 | return false; | |
797 | } | |
798 | ||
36005bae TC |
799 | static int scan_swap_map_slots(struct swap_info_struct *si, |
800 | unsigned char usage, int nr, | |
801 | swp_entry_t slots[]) | |
1da177e4 | 802 | { |
235b6217 | 803 | struct swap_cluster_info *ci; |
ebebbbe9 | 804 | unsigned long offset; |
c60aa176 | 805 | unsigned long scan_base; |
7992fde7 | 806 | unsigned long last_in_cluster = 0; |
048c27fd | 807 | int latency_ration = LATENCY_LIMIT; |
36005bae | 808 | int n_ret = 0; |
ed43af10 | 809 | bool scanned_many = false; |
36005bae | 810 | |
886bb7e9 | 811 | /* |
7dfad418 HD |
812 | * We try to cluster swap pages by allocating them sequentially |
813 | * in swap. Once we've allocated SWAPFILE_CLUSTER pages this | |
814 | * way, however, we resort to first-free allocation, starting | |
815 | * a new cluster. This prevents us from scattering swap pages | |
816 | * all over the entire swap partition, so that we reduce | |
817 | * overall disk seek times between swap pages. -- sct | |
818 | * But we do now try to find an empty cluster. -Andrea | |
c60aa176 | 819 | * And we let swap pages go all over an SSD partition. Hugh |
7dfad418 HD |
820 | */ |
821 | ||
52b7efdb | 822 | si->flags += SWP_SCANNING; |
49070588 HY |
823 | /* |
824 | * Use percpu scan base for SSD to reduce lock contention on | |
825 | * cluster and swap cache. For HDD, sequential access is more | |
826 | * important. | |
827 | */ | |
828 | if (si->flags & SWP_SOLIDSTATE) | |
829 | scan_base = this_cpu_read(*si->cluster_next_cpu); | |
830 | else | |
831 | scan_base = si->cluster_next; | |
832 | offset = scan_base; | |
ebebbbe9 | 833 | |
ebc2a1a6 SL |
834 | /* SSD algorithm */ |
835 | if (si->cluster_info) { | |
bd2d18da | 836 | if (!scan_swap_map_try_ssd_cluster(si, &offset, &scan_base)) |
36005bae | 837 | goto scan; |
f4eaf51a | 838 | } else if (unlikely(!si->cluster_nr--)) { |
ebebbbe9 HD |
839 | if (si->pages - si->inuse_pages < SWAPFILE_CLUSTER) { |
840 | si->cluster_nr = SWAPFILE_CLUSTER - 1; | |
841 | goto checks; | |
842 | } | |
2a8f9449 | 843 | |
ec8acf20 | 844 | spin_unlock(&si->lock); |
7dfad418 | 845 | |
c60aa176 HD |
846 | /* |
847 | * If seek is expensive, start searching for new cluster from | |
848 | * start of partition, to minimize the span of allocated swap. | |
50088c44 CY |
849 | * If seek is cheap, that is the SWP_SOLIDSTATE si->cluster_info |
850 | * case, just handled by scan_swap_map_try_ssd_cluster() above. | |
c60aa176 | 851 | */ |
50088c44 | 852 | scan_base = offset = si->lowest_bit; |
7dfad418 HD |
853 | last_in_cluster = offset + SWAPFILE_CLUSTER - 1; |
854 | ||
855 | /* Locate the first empty (unaligned) cluster */ | |
856 | for (; last_in_cluster <= si->highest_bit; offset++) { | |
1da177e4 | 857 | if (si->swap_map[offset]) |
7dfad418 HD |
858 | last_in_cluster = offset + SWAPFILE_CLUSTER; |
859 | else if (offset == last_in_cluster) { | |
ec8acf20 | 860 | spin_lock(&si->lock); |
ebebbbe9 HD |
861 | offset -= SWAPFILE_CLUSTER - 1; |
862 | si->cluster_next = offset; | |
863 | si->cluster_nr = SWAPFILE_CLUSTER - 1; | |
c60aa176 HD |
864 | goto checks; |
865 | } | |
866 | if (unlikely(--latency_ration < 0)) { | |
867 | cond_resched(); | |
868 | latency_ration = LATENCY_LIMIT; | |
869 | } | |
870 | } | |
871 | ||
872 | offset = scan_base; | |
ec8acf20 | 873 | spin_lock(&si->lock); |
ebebbbe9 | 874 | si->cluster_nr = SWAPFILE_CLUSTER - 1; |
1da177e4 | 875 | } |
7dfad418 | 876 | |
ebebbbe9 | 877 | checks: |
ebc2a1a6 | 878 | if (si->cluster_info) { |
36005bae TC |
879 | while (scan_swap_map_ssd_cluster_conflict(si, offset)) { |
880 | /* take a break if we already got some slots */ | |
881 | if (n_ret) | |
882 | goto done; | |
883 | if (!scan_swap_map_try_ssd_cluster(si, &offset, | |
884 | &scan_base)) | |
885 | goto scan; | |
886 | } | |
ebc2a1a6 | 887 | } |
ebebbbe9 | 888 | if (!(si->flags & SWP_WRITEOK)) |
52b7efdb | 889 | goto no_page; |
7dfad418 HD |
890 | if (!si->highest_bit) |
891 | goto no_page; | |
ebebbbe9 | 892 | if (offset > si->highest_bit) |
c60aa176 | 893 | scan_base = offset = si->lowest_bit; |
c9e44410 | 894 | |
235b6217 | 895 | ci = lock_cluster(si, offset); |
b73d7fce HD |
896 | /* reuse swap entry of cache-only swap if not busy. */ |
897 | if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) { | |
c9e44410 | 898 | int swap_was_freed; |
235b6217 | 899 | unlock_cluster(ci); |
ec8acf20 | 900 | spin_unlock(&si->lock); |
bcd49e86 | 901 | swap_was_freed = __try_to_reclaim_swap(si, offset, TTRS_ANYWAY); |
ec8acf20 | 902 | spin_lock(&si->lock); |
c9e44410 KH |
903 | /* entry was freed successfully, try to use this again */ |
904 | if (swap_was_freed) | |
905 | goto checks; | |
906 | goto scan; /* check next one */ | |
907 | } | |
908 | ||
235b6217 HY |
909 | if (si->swap_map[offset]) { |
910 | unlock_cluster(ci); | |
36005bae TC |
911 | if (!n_ret) |
912 | goto scan; | |
913 | else | |
914 | goto done; | |
235b6217 | 915 | } |
a449bf58 | 916 | WRITE_ONCE(si->swap_map[offset], usage); |
2872bb2d HY |
917 | inc_cluster_info_page(si, si->cluster_info, offset); |
918 | unlock_cluster(ci); | |
ebebbbe9 | 919 | |
38d8b4e6 | 920 | swap_range_alloc(si, offset, 1); |
36005bae TC |
921 | slots[n_ret++] = swp_entry(si->type, offset); |
922 | ||
923 | /* got enough slots or reach max slots? */ | |
924 | if ((n_ret == nr) || (offset >= si->highest_bit)) | |
925 | goto done; | |
926 | ||
927 | /* search for next available slot */ | |
928 | ||
929 | /* time to take a break? */ | |
930 | if (unlikely(--latency_ration < 0)) { | |
931 | if (n_ret) | |
932 | goto done; | |
933 | spin_unlock(&si->lock); | |
934 | cond_resched(); | |
935 | spin_lock(&si->lock); | |
936 | latency_ration = LATENCY_LIMIT; | |
937 | } | |
938 | ||
939 | /* try to get more slots in cluster */ | |
940 | if (si->cluster_info) { | |
941 | if (scan_swap_map_try_ssd_cluster(si, &offset, &scan_base)) | |
942 | goto checks; | |
f4eaf51a WY |
943 | } else if (si->cluster_nr && !si->swap_map[++offset]) { |
944 | /* non-ssd case, still more slots in cluster? */ | |
36005bae TC |
945 | --si->cluster_nr; |
946 | goto checks; | |
947 | } | |
7992fde7 | 948 | |
ed43af10 HY |
949 | /* |
950 | * Even if there's no free clusters available (fragmented), | |
951 | * try to scan a little more quickly with lock held unless we | |
952 | * have scanned too many slots already. | |
953 | */ | |
954 | if (!scanned_many) { | |
955 | unsigned long scan_limit; | |
956 | ||
957 | if (offset < scan_base) | |
958 | scan_limit = scan_base; | |
959 | else | |
960 | scan_limit = si->highest_bit; | |
961 | for (; offset <= scan_limit && --latency_ration > 0; | |
962 | offset++) { | |
963 | if (!si->swap_map[offset]) | |
964 | goto checks; | |
965 | } | |
966 | } | |
967 | ||
36005bae | 968 | done: |
49070588 | 969 | set_cluster_next(si, offset + 1); |
36005bae TC |
970 | si->flags -= SWP_SCANNING; |
971 | return n_ret; | |
7dfad418 | 972 | |
ebebbbe9 | 973 | scan: |
ec8acf20 | 974 | spin_unlock(&si->lock); |
a449bf58 | 975 | while (++offset <= READ_ONCE(si->highest_bit)) { |
048c27fd HD |
976 | if (unlikely(--latency_ration < 0)) { |
977 | cond_resched(); | |
978 | latency_ration = LATENCY_LIMIT; | |
ed43af10 | 979 | scanned_many = true; |
048c27fd | 980 | } |
de1ccfb6 CW |
981 | if (swap_offset_available_and_locked(si, offset)) |
982 | goto checks; | |
7dfad418 | 983 | } |
c60aa176 | 984 | offset = si->lowest_bit; |
a5998061 | 985 | while (offset < scan_base) { |
c60aa176 HD |
986 | if (unlikely(--latency_ration < 0)) { |
987 | cond_resched(); | |
988 | latency_ration = LATENCY_LIMIT; | |
ed43af10 | 989 | scanned_many = true; |
c60aa176 | 990 | } |
de1ccfb6 CW |
991 | if (swap_offset_available_and_locked(si, offset)) |
992 | goto checks; | |
a5998061 | 993 | offset++; |
c60aa176 | 994 | } |
ec8acf20 | 995 | spin_lock(&si->lock); |
7dfad418 HD |
996 | |
997 | no_page: | |
52b7efdb | 998 | si->flags -= SWP_SCANNING; |
36005bae | 999 | return n_ret; |
1da177e4 LT |
1000 | } |
1001 | ||
38d8b4e6 HY |
1002 | static int swap_alloc_cluster(struct swap_info_struct *si, swp_entry_t *slot) |
1003 | { | |
1004 | unsigned long idx; | |
1005 | struct swap_cluster_info *ci; | |
661c7566 | 1006 | unsigned long offset; |
38d8b4e6 | 1007 | |
fe5266d5 HY |
1008 | /* |
1009 | * Should not even be attempting cluster allocations when huge | |
1010 | * page swap is disabled. Warn and fail the allocation. | |
1011 | */ | |
1012 | if (!IS_ENABLED(CONFIG_THP_SWAP)) { | |
1013 | VM_WARN_ON_ONCE(1); | |
1014 | return 0; | |
1015 | } | |
1016 | ||
38d8b4e6 HY |
1017 | if (cluster_list_empty(&si->free_clusters)) |
1018 | return 0; | |
1019 | ||
1020 | idx = cluster_list_first(&si->free_clusters); | |
1021 | offset = idx * SWAPFILE_CLUSTER; | |
1022 | ci = lock_cluster(si, offset); | |
1023 | alloc_cluster(si, idx); | |
e0709829 | 1024 | cluster_set_count_flag(ci, SWAPFILE_CLUSTER, CLUSTER_FLAG_HUGE); |
38d8b4e6 | 1025 | |
661c7566 | 1026 | memset(si->swap_map + offset, SWAP_HAS_CACHE, SWAPFILE_CLUSTER); |
38d8b4e6 HY |
1027 | unlock_cluster(ci); |
1028 | swap_range_alloc(si, offset, SWAPFILE_CLUSTER); | |
1029 | *slot = swp_entry(si->type, offset); | |
1030 | ||
1031 | return 1; | |
1032 | } | |
1033 | ||
1034 | static void swap_free_cluster(struct swap_info_struct *si, unsigned long idx) | |
1035 | { | |
1036 | unsigned long offset = idx * SWAPFILE_CLUSTER; | |
1037 | struct swap_cluster_info *ci; | |
1038 | ||
1039 | ci = lock_cluster(si, offset); | |
979aafa5 | 1040 | memset(si->swap_map + offset, 0, SWAPFILE_CLUSTER); |
38d8b4e6 HY |
1041 | cluster_set_count_flag(ci, 0, 0); |
1042 | free_cluster(si, idx); | |
1043 | unlock_cluster(ci); | |
1044 | swap_range_free(si, offset, SWAPFILE_CLUSTER); | |
1045 | } | |
38d8b4e6 | 1046 | |
5d5e8f19 | 1047 | int get_swap_pages(int n_goal, swp_entry_t swp_entries[], int entry_size) |
1da177e4 | 1048 | { |
5d5e8f19 | 1049 | unsigned long size = swap_entry_size(entry_size); |
adfab836 | 1050 | struct swap_info_struct *si, *next; |
36005bae TC |
1051 | long avail_pgs; |
1052 | int n_ret = 0; | |
a2468cc9 | 1053 | int node; |
1da177e4 | 1054 | |
38d8b4e6 | 1055 | /* Only single cluster request supported */ |
5d5e8f19 | 1056 | WARN_ON_ONCE(n_goal > 1 && size == SWAPFILE_CLUSTER); |
38d8b4e6 | 1057 | |
b50da6e9 ZH |
1058 | spin_lock(&swap_avail_lock); |
1059 | ||
5d5e8f19 | 1060 | avail_pgs = atomic_long_read(&nr_swap_pages) / size; |
b50da6e9 ZH |
1061 | if (avail_pgs <= 0) { |
1062 | spin_unlock(&swap_avail_lock); | |
fb4f88dc | 1063 | goto noswap; |
b50da6e9 | 1064 | } |
36005bae | 1065 | |
08d3090f | 1066 | n_goal = min3((long)n_goal, (long)SWAP_BATCH, avail_pgs); |
36005bae | 1067 | |
5d5e8f19 | 1068 | atomic_long_sub(n_goal * size, &nr_swap_pages); |
fb4f88dc | 1069 | |
18ab4d4c | 1070 | start_over: |
a2468cc9 AL |
1071 | node = numa_node_id(); |
1072 | plist_for_each_entry_safe(si, next, &swap_avail_heads[node], avail_lists[node]) { | |
18ab4d4c | 1073 | /* requeue si to after same-priority siblings */ |
a2468cc9 | 1074 | plist_requeue(&si->avail_lists[node], &swap_avail_heads[node]); |
18ab4d4c | 1075 | spin_unlock(&swap_avail_lock); |
ec8acf20 | 1076 | spin_lock(&si->lock); |
adfab836 | 1077 | if (!si->highest_bit || !(si->flags & SWP_WRITEOK)) { |
18ab4d4c | 1078 | spin_lock(&swap_avail_lock); |
a2468cc9 | 1079 | if (plist_node_empty(&si->avail_lists[node])) { |
18ab4d4c DS |
1080 | spin_unlock(&si->lock); |
1081 | goto nextsi; | |
1082 | } | |
1083 | WARN(!si->highest_bit, | |
1084 | "swap_info %d in list but !highest_bit\n", | |
1085 | si->type); | |
1086 | WARN(!(si->flags & SWP_WRITEOK), | |
1087 | "swap_info %d in list but !SWP_WRITEOK\n", | |
1088 | si->type); | |
a2468cc9 | 1089 | __del_from_avail_list(si); |
ec8acf20 | 1090 | spin_unlock(&si->lock); |
18ab4d4c | 1091 | goto nextsi; |
ec8acf20 | 1092 | } |
5d5e8f19 | 1093 | if (size == SWAPFILE_CLUSTER) { |
41663430 | 1094 | if (si->flags & SWP_BLKDEV) |
f0eea189 HY |
1095 | n_ret = swap_alloc_cluster(si, swp_entries); |
1096 | } else | |
38d8b4e6 HY |
1097 | n_ret = scan_swap_map_slots(si, SWAP_HAS_CACHE, |
1098 | n_goal, swp_entries); | |
ec8acf20 | 1099 | spin_unlock(&si->lock); |
5d5e8f19 | 1100 | if (n_ret || size == SWAPFILE_CLUSTER) |
36005bae | 1101 | goto check_out; |
7717fc1a | 1102 | cond_resched(); |
36005bae | 1103 | |
18ab4d4c DS |
1104 | spin_lock(&swap_avail_lock); |
1105 | nextsi: | |
adfab836 DS |
1106 | /* |
1107 | * if we got here, it's likely that si was almost full before, | |
bb243f7d ML |
1108 | * and since scan_swap_map_slots() can drop the si->lock, |
1109 | * multiple callers probably all tried to get a page from the | |
1110 | * same si and it filled up before we could get one; or, the si | |
1111 | * filled up between us dropping swap_avail_lock and taking | |
1112 | * si->lock. Since we dropped the swap_avail_lock, the | |
1113 | * swap_avail_head list may have been modified; so if next is | |
1114 | * still in the swap_avail_head list then try it, otherwise | |
1115 | * start over if we have not gotten any slots. | |
adfab836 | 1116 | */ |
a2468cc9 | 1117 | if (plist_node_empty(&next->avail_lists[node])) |
18ab4d4c | 1118 | goto start_over; |
1da177e4 | 1119 | } |
fb4f88dc | 1120 | |
18ab4d4c DS |
1121 | spin_unlock(&swap_avail_lock); |
1122 | ||
36005bae TC |
1123 | check_out: |
1124 | if (n_ret < n_goal) | |
5d5e8f19 | 1125 | atomic_long_add((long)(n_goal - n_ret) * size, |
38d8b4e6 | 1126 | &nr_swap_pages); |
fb4f88dc | 1127 | noswap: |
36005bae TC |
1128 | return n_ret; |
1129 | } | |
1130 | ||
afba72b1 | 1131 | static struct swap_info_struct *_swap_info_get(swp_entry_t entry) |
1da177e4 | 1132 | { |
73c34b6a | 1133 | struct swap_info_struct *p; |
eb085574 | 1134 | unsigned long offset; |
1da177e4 LT |
1135 | |
1136 | if (!entry.val) | |
1137 | goto out; | |
eb085574 | 1138 | p = swp_swap_info(entry); |
c10d38cc | 1139 | if (!p) |
1da177e4 | 1140 | goto bad_nofile; |
a449bf58 | 1141 | if (data_race(!(p->flags & SWP_USED))) |
1da177e4 LT |
1142 | goto bad_device; |
1143 | offset = swp_offset(entry); | |
1144 | if (offset >= p->max) | |
1145 | goto bad_offset; | |
afba72b1 ML |
1146 | if (data_race(!p->swap_map[swp_offset(entry)])) |
1147 | goto bad_free; | |
1da177e4 LT |
1148 | return p; |
1149 | ||
afba72b1 ML |
1150 | bad_free: |
1151 | pr_err("%s: %s%08lx\n", __func__, Unused_offset, entry.val); | |
1152 | goto out; | |
1da177e4 | 1153 | bad_offset: |
cf532faa | 1154 | pr_err("%s: %s%08lx\n", __func__, Bad_offset, entry.val); |
1da177e4 LT |
1155 | goto out; |
1156 | bad_device: | |
cf532faa | 1157 | pr_err("%s: %s%08lx\n", __func__, Unused_file, entry.val); |
1da177e4 LT |
1158 | goto out; |
1159 | bad_nofile: | |
cf532faa | 1160 | pr_err("%s: %s%08lx\n", __func__, Bad_file, entry.val); |
1da177e4 LT |
1161 | out: |
1162 | return NULL; | |
886bb7e9 | 1163 | } |
1da177e4 | 1164 | |
7c00bafe TC |
1165 | static struct swap_info_struct *swap_info_get_cont(swp_entry_t entry, |
1166 | struct swap_info_struct *q) | |
1167 | { | |
1168 | struct swap_info_struct *p; | |
1169 | ||
1170 | p = _swap_info_get(entry); | |
1171 | ||
1172 | if (p != q) { | |
1173 | if (q != NULL) | |
1174 | spin_unlock(&q->lock); | |
1175 | if (p != NULL) | |
1176 | spin_lock(&p->lock); | |
1177 | } | |
1178 | return p; | |
1179 | } | |
1180 | ||
b32d5f32 HY |
1181 | static unsigned char __swap_entry_free_locked(struct swap_info_struct *p, |
1182 | unsigned long offset, | |
1183 | unsigned char usage) | |
1da177e4 | 1184 | { |
8d69aaee HD |
1185 | unsigned char count; |
1186 | unsigned char has_cache; | |
235b6217 | 1187 | |
253d553b | 1188 | count = p->swap_map[offset]; |
235b6217 | 1189 | |
253d553b HD |
1190 | has_cache = count & SWAP_HAS_CACHE; |
1191 | count &= ~SWAP_HAS_CACHE; | |
355cfa73 | 1192 | |
253d553b | 1193 | if (usage == SWAP_HAS_CACHE) { |
355cfa73 | 1194 | VM_BUG_ON(!has_cache); |
253d553b | 1195 | has_cache = 0; |
aaa46865 HD |
1196 | } else if (count == SWAP_MAP_SHMEM) { |
1197 | /* | |
1198 | * Or we could insist on shmem.c using a special | |
1199 | * swap_shmem_free() and free_shmem_swap_and_cache()... | |
1200 | */ | |
1201 | count = 0; | |
570a335b HD |
1202 | } else if ((count & ~COUNT_CONTINUED) <= SWAP_MAP_MAX) { |
1203 | if (count == COUNT_CONTINUED) { | |
1204 | if (swap_count_continued(p, offset, count)) | |
1205 | count = SWAP_MAP_MAX | COUNT_CONTINUED; | |
1206 | else | |
1207 | count = SWAP_MAP_MAX; | |
1208 | } else | |
1209 | count--; | |
1210 | } | |
253d553b | 1211 | |
253d553b | 1212 | usage = count | has_cache; |
a449bf58 QC |
1213 | if (usage) |
1214 | WRITE_ONCE(p->swap_map[offset], usage); | |
1215 | else | |
1216 | WRITE_ONCE(p->swap_map[offset], SWAP_HAS_CACHE); | |
7c00bafe | 1217 | |
b32d5f32 HY |
1218 | return usage; |
1219 | } | |
1220 | ||
eb085574 | 1221 | /* |
a95722a0 HY |
1222 | * When we get a swap entry, if there aren't some other ways to |
1223 | * prevent swapoff, such as the folio in swap cache is locked, page | |
1224 | * table lock is held, etc., the swap entry may become invalid because | |
1225 | * of swapoff. Then, we need to enclose all swap related functions | |
1226 | * with get_swap_device() and put_swap_device(), unless the swap | |
1227 | * functions call get/put_swap_device() by themselves. | |
1228 | * | |
eb085574 HY |
1229 | * Check whether swap entry is valid in the swap device. If so, |
1230 | * return pointer to swap_info_struct, and keep the swap entry valid | |
1231 | * via preventing the swap device from being swapoff, until | |
1232 | * put_swap_device() is called. Otherwise return NULL. | |
1233 | * | |
eb085574 | 1234 | * Notice that swapoff or swapoff+swapon can still happen before the |
63d8620e ML |
1235 | * percpu_ref_tryget_live() in get_swap_device() or after the |
1236 | * percpu_ref_put() in put_swap_device() if there isn't any other way | |
a95722a0 HY |
1237 | * to prevent swapoff. The caller must be prepared for that. For |
1238 | * example, the following situation is possible. | |
eb085574 HY |
1239 | * |
1240 | * CPU1 CPU2 | |
1241 | * do_swap_page() | |
1242 | * ... swapoff+swapon | |
1243 | * __read_swap_cache_async() | |
1244 | * swapcache_prepare() | |
1245 | * __swap_duplicate() | |
1246 | * // check swap_map | |
1247 | * // verify PTE not changed | |
1248 | * | |
1249 | * In __swap_duplicate(), the swap_map need to be checked before | |
1250 | * changing partly because the specified swap entry may be for another | |
1251 | * swap device which has been swapoff. And in do_swap_page(), after | |
1252 | * the page is read from the swap device, the PTE is verified not | |
1253 | * changed with the page table locked to check whether the swap device | |
1254 | * has been swapoff or swapoff+swapon. | |
1255 | */ | |
1256 | struct swap_info_struct *get_swap_device(swp_entry_t entry) | |
1257 | { | |
1258 | struct swap_info_struct *si; | |
1259 | unsigned long offset; | |
1260 | ||
1261 | if (!entry.val) | |
1262 | goto out; | |
1263 | si = swp_swap_info(entry); | |
1264 | if (!si) | |
1265 | goto bad_nofile; | |
63d8620e ML |
1266 | if (!percpu_ref_tryget_live(&si->users)) |
1267 | goto out; | |
1268 | /* | |
1269 | * Guarantee the si->users are checked before accessing other | |
1270 | * fields of swap_info_struct. | |
1271 | * | |
1272 | * Paired with the spin_unlock() after setup_swap_info() in | |
1273 | * enable_swap_info(). | |
1274 | */ | |
1275 | smp_rmb(); | |
eb085574 HY |
1276 | offset = swp_offset(entry); |
1277 | if (offset >= si->max) | |
63d8620e | 1278 | goto put_out; |
eb085574 HY |
1279 | |
1280 | return si; | |
1281 | bad_nofile: | |
1282 | pr_err("%s: %s%08lx\n", __func__, Bad_file, entry.val); | |
1283 | out: | |
1284 | return NULL; | |
63d8620e | 1285 | put_out: |
23b230ba | 1286 | pr_err("%s: %s%08lx\n", __func__, Bad_offset, entry.val); |
63d8620e | 1287 | percpu_ref_put(&si->users); |
eb085574 HY |
1288 | return NULL; |
1289 | } | |
1290 | ||
b32d5f32 | 1291 | static unsigned char __swap_entry_free(struct swap_info_struct *p, |
33e16272 | 1292 | swp_entry_t entry) |
b32d5f32 HY |
1293 | { |
1294 | struct swap_cluster_info *ci; | |
1295 | unsigned long offset = swp_offset(entry); | |
33e16272 | 1296 | unsigned char usage; |
b32d5f32 HY |
1297 | |
1298 | ci = lock_cluster_or_swap_info(p, offset); | |
33e16272 | 1299 | usage = __swap_entry_free_locked(p, offset, 1); |
7c00bafe | 1300 | unlock_cluster_or_swap_info(p, ci); |
10e364da HY |
1301 | if (!usage) |
1302 | free_swap_slot(entry); | |
7c00bafe TC |
1303 | |
1304 | return usage; | |
1305 | } | |
355cfa73 | 1306 | |
7c00bafe TC |
1307 | static void swap_entry_free(struct swap_info_struct *p, swp_entry_t entry) |
1308 | { | |
1309 | struct swap_cluster_info *ci; | |
1310 | unsigned long offset = swp_offset(entry); | |
1311 | unsigned char count; | |
1312 | ||
1313 | ci = lock_cluster(p, offset); | |
1314 | count = p->swap_map[offset]; | |
1315 | VM_BUG_ON(count != SWAP_HAS_CACHE); | |
1316 | p->swap_map[offset] = 0; | |
1317 | dec_cluster_info_page(p, p->cluster_info, offset); | |
235b6217 HY |
1318 | unlock_cluster(ci); |
1319 | ||
38d8b4e6 HY |
1320 | mem_cgroup_uncharge_swap(entry, 1); |
1321 | swap_range_free(p, offset, 1); | |
1da177e4 LT |
1322 | } |
1323 | ||
1324 | /* | |
2de1a7e4 | 1325 | * Caller has made sure that the swap device corresponding to entry |
1da177e4 LT |
1326 | * is still around or has not been recycled. |
1327 | */ | |
1328 | void swap_free(swp_entry_t entry) | |
1329 | { | |
73c34b6a | 1330 | struct swap_info_struct *p; |
1da177e4 | 1331 | |
235b6217 | 1332 | p = _swap_info_get(entry); |
10e364da | 1333 | if (p) |
33e16272 | 1334 | __swap_entry_free(p, entry); |
1da177e4 LT |
1335 | } |
1336 | ||
cb4b86ba KH |
1337 | /* |
1338 | * Called after dropping swapcache to decrease refcnt to swap entries. | |
1339 | */ | |
4081f744 | 1340 | void put_swap_folio(struct folio *folio, swp_entry_t entry) |
38d8b4e6 HY |
1341 | { |
1342 | unsigned long offset = swp_offset(entry); | |
1343 | unsigned long idx = offset / SWAPFILE_CLUSTER; | |
1344 | struct swap_cluster_info *ci; | |
1345 | struct swap_info_struct *si; | |
1346 | unsigned char *map; | |
a3aea839 HY |
1347 | unsigned int i, free_entries = 0; |
1348 | unsigned char val; | |
4081f744 | 1349 | int size = swap_entry_size(folio_nr_pages(folio)); |
fe5266d5 | 1350 | |
a3aea839 | 1351 | si = _swap_info_get(entry); |
38d8b4e6 HY |
1352 | if (!si) |
1353 | return; | |
1354 | ||
c2343d27 | 1355 | ci = lock_cluster_or_swap_info(si, offset); |
a448f2d0 | 1356 | if (size == SWAPFILE_CLUSTER) { |
a448f2d0 HY |
1357 | VM_BUG_ON(!cluster_is_huge(ci)); |
1358 | map = si->swap_map + offset; | |
1359 | for (i = 0; i < SWAPFILE_CLUSTER; i++) { | |
1360 | val = map[i]; | |
1361 | VM_BUG_ON(!(val & SWAP_HAS_CACHE)); | |
1362 | if (val == SWAP_HAS_CACHE) | |
1363 | free_entries++; | |
1364 | } | |
a448f2d0 | 1365 | cluster_clear_huge(ci); |
a448f2d0 | 1366 | if (free_entries == SWAPFILE_CLUSTER) { |
c2343d27 | 1367 | unlock_cluster_or_swap_info(si, ci); |
a448f2d0 | 1368 | spin_lock(&si->lock); |
a448f2d0 HY |
1369 | mem_cgroup_uncharge_swap(entry, SWAPFILE_CLUSTER); |
1370 | swap_free_cluster(si, idx); | |
1371 | spin_unlock(&si->lock); | |
1372 | return; | |
1373 | } | |
1374 | } | |
c2343d27 HY |
1375 | for (i = 0; i < size; i++, entry.val++) { |
1376 | if (!__swap_entry_free_locked(si, offset + i, SWAP_HAS_CACHE)) { | |
1377 | unlock_cluster_or_swap_info(si, ci); | |
1378 | free_swap_slot(entry); | |
1379 | if (i == size - 1) | |
1380 | return; | |
1381 | lock_cluster_or_swap_info(si, offset); | |
a3aea839 HY |
1382 | } |
1383 | } | |
c2343d27 | 1384 | unlock_cluster_or_swap_info(si, ci); |
38d8b4e6 | 1385 | } |
59807685 | 1386 | |
fe5266d5 | 1387 | #ifdef CONFIG_THP_SWAP |
59807685 HY |
1388 | int split_swap_cluster(swp_entry_t entry) |
1389 | { | |
1390 | struct swap_info_struct *si; | |
1391 | struct swap_cluster_info *ci; | |
1392 | unsigned long offset = swp_offset(entry); | |
1393 | ||
1394 | si = _swap_info_get(entry); | |
1395 | if (!si) | |
1396 | return -EBUSY; | |
1397 | ci = lock_cluster(si, offset); | |
1398 | cluster_clear_huge(ci); | |
1399 | unlock_cluster(ci); | |
1400 | return 0; | |
1401 | } | |
fe5266d5 | 1402 | #endif |
38d8b4e6 | 1403 | |
155b5f88 HY |
1404 | static int swp_entry_cmp(const void *ent1, const void *ent2) |
1405 | { | |
1406 | const swp_entry_t *e1 = ent1, *e2 = ent2; | |
1407 | ||
1408 | return (int)swp_type(*e1) - (int)swp_type(*e2); | |
1409 | } | |
1410 | ||
7c00bafe TC |
1411 | void swapcache_free_entries(swp_entry_t *entries, int n) |
1412 | { | |
1413 | struct swap_info_struct *p, *prev; | |
1414 | int i; | |
1415 | ||
1416 | if (n <= 0) | |
1417 | return; | |
1418 | ||
1419 | prev = NULL; | |
1420 | p = NULL; | |
155b5f88 HY |
1421 | |
1422 | /* | |
1423 | * Sort swap entries by swap device, so each lock is only taken once. | |
1424 | * nr_swapfiles isn't absolutely correct, but the overhead of sort() is | |
1425 | * so low that it isn't necessary to optimize further. | |
1426 | */ | |
1427 | if (nr_swapfiles > 1) | |
1428 | sort(entries, n, sizeof(entries[0]), swp_entry_cmp, NULL); | |
7c00bafe TC |
1429 | for (i = 0; i < n; ++i) { |
1430 | p = swap_info_get_cont(entries[i], prev); | |
1431 | if (p) | |
1432 | swap_entry_free(p, entries[i]); | |
7c00bafe TC |
1433 | prev = p; |
1434 | } | |
235b6217 | 1435 | if (p) |
7c00bafe | 1436 | spin_unlock(&p->lock); |
cb4b86ba KH |
1437 | } |
1438 | ||
eb085574 | 1439 | int __swap_count(swp_entry_t entry) |
aa8d22a1 | 1440 | { |
f9f956b5 | 1441 | struct swap_info_struct *si = swp_swap_info(entry); |
aa8d22a1 MK |
1442 | pgoff_t offset = swp_offset(entry); |
1443 | ||
f9f956b5 | 1444 | return swap_count(si->swap_map[offset]); |
aa8d22a1 MK |
1445 | } |
1446 | ||
14d01ee9 MWO |
1447 | /* |
1448 | * How many references to @entry are currently swapped out? | |
1449 | * This does not give an exact answer when swap count is continued, | |
1450 | * but does include the high COUNT_CONTINUED flag to allow for that. | |
1451 | */ | |
3ecdeb0f | 1452 | int swap_swapcount(struct swap_info_struct *si, swp_entry_t entry) |
322b8afe | 1453 | { |
322b8afe HY |
1454 | pgoff_t offset = swp_offset(entry); |
1455 | struct swap_cluster_info *ci; | |
14d01ee9 | 1456 | int count; |
322b8afe HY |
1457 | |
1458 | ci = lock_cluster_or_swap_info(si, offset); | |
1459 | count = swap_count(si->swap_map[offset]); | |
1460 | unlock_cluster_or_swap_info(si, ci); | |
1461 | return count; | |
1462 | } | |
1463 | ||
8334b962 MK |
1464 | /* |
1465 | * How many references to @entry are currently swapped out? | |
1466 | * This considers COUNT_CONTINUED so it returns exact answer. | |
1467 | */ | |
1468 | int swp_swapcount(swp_entry_t entry) | |
1469 | { | |
1470 | int count, tmp_count, n; | |
1471 | struct swap_info_struct *p; | |
235b6217 | 1472 | struct swap_cluster_info *ci; |
8334b962 MK |
1473 | struct page *page; |
1474 | pgoff_t offset; | |
1475 | unsigned char *map; | |
1476 | ||
235b6217 | 1477 | p = _swap_info_get(entry); |
8334b962 MK |
1478 | if (!p) |
1479 | return 0; | |
1480 | ||
235b6217 HY |
1481 | offset = swp_offset(entry); |
1482 | ||
1483 | ci = lock_cluster_or_swap_info(p, offset); | |
1484 | ||
1485 | count = swap_count(p->swap_map[offset]); | |
8334b962 MK |
1486 | if (!(count & COUNT_CONTINUED)) |
1487 | goto out; | |
1488 | ||
1489 | count &= ~COUNT_CONTINUED; | |
1490 | n = SWAP_MAP_MAX + 1; | |
1491 | ||
8334b962 MK |
1492 | page = vmalloc_to_page(p->swap_map + offset); |
1493 | offset &= ~PAGE_MASK; | |
1494 | VM_BUG_ON(page_private(page) != SWP_CONTINUED); | |
1495 | ||
1496 | do { | |
a8ae4991 | 1497 | page = list_next_entry(page, lru); |
8334b962 MK |
1498 | map = kmap_atomic(page); |
1499 | tmp_count = map[offset]; | |
1500 | kunmap_atomic(map); | |
1501 | ||
1502 | count += (tmp_count & ~COUNT_CONTINUED) * n; | |
1503 | n *= (SWAP_CONT_MAX + 1); | |
1504 | } while (tmp_count & COUNT_CONTINUED); | |
1505 | out: | |
235b6217 | 1506 | unlock_cluster_or_swap_info(p, ci); |
8334b962 MK |
1507 | return count; |
1508 | } | |
1509 | ||
e0709829 HY |
1510 | static bool swap_page_trans_huge_swapped(struct swap_info_struct *si, |
1511 | swp_entry_t entry) | |
1512 | { | |
1513 | struct swap_cluster_info *ci; | |
1514 | unsigned char *map = si->swap_map; | |
1515 | unsigned long roffset = swp_offset(entry); | |
1516 | unsigned long offset = round_down(roffset, SWAPFILE_CLUSTER); | |
1517 | int i; | |
1518 | bool ret = false; | |
1519 | ||
1520 | ci = lock_cluster_or_swap_info(si, offset); | |
1521 | if (!ci || !cluster_is_huge(ci)) { | |
afa4711e | 1522 | if (swap_count(map[roffset])) |
e0709829 HY |
1523 | ret = true; |
1524 | goto unlock_out; | |
1525 | } | |
1526 | for (i = 0; i < SWAPFILE_CLUSTER; i++) { | |
afa4711e | 1527 | if (swap_count(map[offset + i])) { |
e0709829 HY |
1528 | ret = true; |
1529 | break; | |
1530 | } | |
1531 | } | |
1532 | unlock_out: | |
1533 | unlock_cluster_or_swap_info(si, ci); | |
1534 | return ret; | |
1535 | } | |
1536 | ||
2397f780 | 1537 | static bool folio_swapped(struct folio *folio) |
e0709829 | 1538 | { |
3d2c9087 | 1539 | swp_entry_t entry = folio->swap; |
14d01ee9 MWO |
1540 | struct swap_info_struct *si = _swap_info_get(entry); |
1541 | ||
1542 | if (!si) | |
1543 | return false; | |
e0709829 | 1544 | |
2397f780 | 1545 | if (!IS_ENABLED(CONFIG_THP_SWAP) || likely(!folio_test_large(folio))) |
14d01ee9 | 1546 | return swap_swapcount(si, entry) != 0; |
e0709829 | 1547 | |
14d01ee9 | 1548 | return swap_page_trans_huge_swapped(si, entry); |
e0709829 | 1549 | } |
ba3c4ce6 | 1550 | |
bdb0ed54 MWO |
1551 | /** |
1552 | * folio_free_swap() - Free the swap space used for this folio. | |
1553 | * @folio: The folio to remove. | |
1554 | * | |
1555 | * If swap is getting full, or if there are no more mappings of this folio, | |
1556 | * then call folio_free_swap to free its swap space. | |
1557 | * | |
1558 | * Return: true if we were able to release the swap space. | |
1da177e4 | 1559 | */ |
bdb0ed54 | 1560 | bool folio_free_swap(struct folio *folio) |
1da177e4 | 1561 | { |
2397f780 | 1562 | VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio); |
1da177e4 | 1563 | |
2397f780 | 1564 | if (!folio_test_swapcache(folio)) |
bdb0ed54 | 1565 | return false; |
2397f780 | 1566 | if (folio_test_writeback(folio)) |
bdb0ed54 | 1567 | return false; |
2397f780 | 1568 | if (folio_swapped(folio)) |
bdb0ed54 | 1569 | return false; |
1da177e4 | 1570 | |
b73d7fce HD |
1571 | /* |
1572 | * Once hibernation has begun to create its image of memory, | |
bdb0ed54 | 1573 | * there's a danger that one of the calls to folio_free_swap() |
b73d7fce HD |
1574 | * - most probably a call from __try_to_reclaim_swap() while |
1575 | * hibernation is allocating its own swap pages for the image, | |
1576 | * but conceivably even a call from memory reclaim - will free | |
bdb0ed54 MWO |
1577 | * the swap from a folio which has already been recorded in the |
1578 | * image as a clean swapcache folio, and then reuse its swap for | |
b73d7fce | 1579 | * another page of the image. On waking from hibernation, the |
bdb0ed54 | 1580 | * original folio might be freed under memory pressure, then |
b73d7fce HD |
1581 | * later read back in from swap, now with the wrong data. |
1582 | * | |
2de1a7e4 | 1583 | * Hibernation suspends storage while it is writing the image |
f90ac398 | 1584 | * to disk so check that here. |
b73d7fce | 1585 | */ |
f90ac398 | 1586 | if (pm_suspended_storage()) |
bdb0ed54 | 1587 | return false; |
b73d7fce | 1588 | |
75fa68a5 | 1589 | delete_from_swap_cache(folio); |
2397f780 | 1590 | folio_set_dirty(folio); |
bdb0ed54 | 1591 | return true; |
68a22394 RR |
1592 | } |
1593 | ||
1da177e4 LT |
1594 | /* |
1595 | * Free the swap entry like above, but also try to | |
1596 | * free the page cache entry if it is the last user. | |
1597 | */ | |
2509ef26 | 1598 | int free_swap_and_cache(swp_entry_t entry) |
1da177e4 | 1599 | { |
2509ef26 | 1600 | struct swap_info_struct *p; |
7c00bafe | 1601 | unsigned char count; |
1da177e4 | 1602 | |
a7420aa5 | 1603 | if (non_swap_entry(entry)) |
2509ef26 | 1604 | return 1; |
0697212a | 1605 | |
7c00bafe | 1606 | p = _swap_info_get(entry); |
1da177e4 | 1607 | if (p) { |
33e16272 | 1608 | count = __swap_entry_free(p, entry); |
e0709829 | 1609 | if (count == SWAP_HAS_CACHE && |
bcd49e86 HY |
1610 | !swap_page_trans_huge_swapped(p, entry)) |
1611 | __try_to_reclaim_swap(p, swp_offset(entry), | |
1612 | TTRS_UNMAPPED | TTRS_FULL); | |
1da177e4 | 1613 | } |
2509ef26 | 1614 | return p != NULL; |
1da177e4 LT |
1615 | } |
1616 | ||
b0cb1a19 | 1617 | #ifdef CONFIG_HIBERNATION |
bb243f7d ML |
1618 | |
1619 | swp_entry_t get_swap_page_of_type(int type) | |
1620 | { | |
1621 | struct swap_info_struct *si = swap_type_to_swap_info(type); | |
1622 | swp_entry_t entry = {0}; | |
1623 | ||
1624 | if (!si) | |
1625 | goto fail; | |
1626 | ||
1627 | /* This is called for allocating swap entry, not cache */ | |
1628 | spin_lock(&si->lock); | |
1629 | if ((si->flags & SWP_WRITEOK) && scan_swap_map_slots(si, 1, 1, &entry)) | |
1630 | atomic_long_dec(&nr_swap_pages); | |
1631 | spin_unlock(&si->lock); | |
1632 | fail: | |
1633 | return entry; | |
1634 | } | |
1635 | ||
f577eb30 | 1636 | /* |
915bae9e | 1637 | * Find the swap type that corresponds to given device (if any). |
f577eb30 | 1638 | * |
915bae9e RW |
1639 | * @offset - number of the PAGE_SIZE-sized block of the device, starting |
1640 | * from 0, in which the swap header is expected to be located. | |
1641 | * | |
1642 | * This is needed for the suspend to disk (aka swsusp). | |
f577eb30 | 1643 | */ |
21bd9005 | 1644 | int swap_type_of(dev_t device, sector_t offset) |
f577eb30 | 1645 | { |
efa90a98 | 1646 | int type; |
f577eb30 | 1647 | |
21bd9005 CH |
1648 | if (!device) |
1649 | return -1; | |
915bae9e | 1650 | |
f577eb30 | 1651 | spin_lock(&swap_lock); |
efa90a98 HD |
1652 | for (type = 0; type < nr_swapfiles; type++) { |
1653 | struct swap_info_struct *sis = swap_info[type]; | |
f577eb30 | 1654 | |
915bae9e | 1655 | if (!(sis->flags & SWP_WRITEOK)) |
f577eb30 | 1656 | continue; |
b6b5bce3 | 1657 | |
21bd9005 | 1658 | if (device == sis->bdev->bd_dev) { |
4efaceb1 | 1659 | struct swap_extent *se = first_se(sis); |
915bae9e | 1660 | |
915bae9e RW |
1661 | if (se->start_block == offset) { |
1662 | spin_unlock(&swap_lock); | |
efa90a98 | 1663 | return type; |
915bae9e | 1664 | } |
f577eb30 RW |
1665 | } |
1666 | } | |
1667 | spin_unlock(&swap_lock); | |
21bd9005 CH |
1668 | return -ENODEV; |
1669 | } | |
915bae9e | 1670 | |
21bd9005 CH |
1671 | int find_first_swap(dev_t *device) |
1672 | { | |
1673 | int type; | |
915bae9e | 1674 | |
21bd9005 CH |
1675 | spin_lock(&swap_lock); |
1676 | for (type = 0; type < nr_swapfiles; type++) { | |
1677 | struct swap_info_struct *sis = swap_info[type]; | |
1678 | ||
1679 | if (!(sis->flags & SWP_WRITEOK)) | |
1680 | continue; | |
1681 | *device = sis->bdev->bd_dev; | |
1682 | spin_unlock(&swap_lock); | |
1683 | return type; | |
1684 | } | |
1685 | spin_unlock(&swap_lock); | |
f577eb30 RW |
1686 | return -ENODEV; |
1687 | } | |
1688 | ||
73c34b6a HD |
1689 | /* |
1690 | * Get the (PAGE_SIZE) block corresponding to given offset on the swapdev | |
1691 | * corresponding to given index in swap_info (swap type). | |
1692 | */ | |
1693 | sector_t swapdev_block(int type, pgoff_t offset) | |
1694 | { | |
c10d38cc | 1695 | struct swap_info_struct *si = swap_type_to_swap_info(type); |
f885056a | 1696 | struct swap_extent *se; |
73c34b6a | 1697 | |
c10d38cc | 1698 | if (!si || !(si->flags & SWP_WRITEOK)) |
73c34b6a | 1699 | return 0; |
f885056a CH |
1700 | se = offset_to_swap_extent(si, offset); |
1701 | return se->start_block + (offset - se->start_page); | |
73c34b6a HD |
1702 | } |
1703 | ||
f577eb30 RW |
1704 | /* |
1705 | * Return either the total number of swap pages of given type, or the number | |
1706 | * of free pages of that type (depending on @free) | |
1707 | * | |
1708 | * This is needed for software suspend | |
1709 | */ | |
1710 | unsigned int count_swap_pages(int type, int free) | |
1711 | { | |
1712 | unsigned int n = 0; | |
1713 | ||
efa90a98 HD |
1714 | spin_lock(&swap_lock); |
1715 | if ((unsigned int)type < nr_swapfiles) { | |
1716 | struct swap_info_struct *sis = swap_info[type]; | |
1717 | ||
ec8acf20 | 1718 | spin_lock(&sis->lock); |
efa90a98 HD |
1719 | if (sis->flags & SWP_WRITEOK) { |
1720 | n = sis->pages; | |
f577eb30 | 1721 | if (free) |
efa90a98 | 1722 | n -= sis->inuse_pages; |
f577eb30 | 1723 | } |
ec8acf20 | 1724 | spin_unlock(&sis->lock); |
f577eb30 | 1725 | } |
efa90a98 | 1726 | spin_unlock(&swap_lock); |
f577eb30 RW |
1727 | return n; |
1728 | } | |
73c34b6a | 1729 | #endif /* CONFIG_HIBERNATION */ |
f577eb30 | 1730 | |
9f8bdb3f | 1731 | static inline int pte_same_as_swp(pte_t pte, pte_t swp_pte) |
179ef71c | 1732 | { |
099dd687 | 1733 | return pte_same(pte_swp_clear_flags(pte), swp_pte); |
179ef71c CG |
1734 | } |
1735 | ||
1da177e4 | 1736 | /* |
72866f6f HD |
1737 | * No need to decide whether this PTE shares the swap entry with others, |
1738 | * just let do_wp_page work it out if a write is requested later - to | |
1739 | * force COW, vm_page_prot omits write permission from any private vma. | |
1da177e4 | 1740 | */ |
044d66c1 | 1741 | static int unuse_pte(struct vm_area_struct *vma, pmd_t *pmd, |
f102cd8b | 1742 | unsigned long addr, swp_entry_t entry, struct folio *folio) |
1da177e4 | 1743 | { |
f102cd8b | 1744 | struct page *page = folio_file_page(folio, swp_offset(entry)); |
9e16b7fb | 1745 | struct page *swapcache; |
044d66c1 | 1746 | spinlock_t *ptl; |
c33c7948 | 1747 | pte_t *pte, new_pte, old_pte; |
f985fc32 | 1748 | bool hwpoisoned = PageHWPoison(page); |
044d66c1 HD |
1749 | int ret = 1; |
1750 | ||
9e16b7fb HD |
1751 | swapcache = page; |
1752 | page = ksm_might_need_to_copy(page, vma, addr); | |
1753 | if (unlikely(!page)) | |
1754 | return -ENOMEM; | |
6b970599 | 1755 | else if (unlikely(PTR_ERR(page) == -EHWPOISON)) |
f985fc32 | 1756 | hwpoisoned = true; |
9e16b7fb | 1757 | |
044d66c1 | 1758 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
c33c7948 RR |
1759 | if (unlikely(!pte || !pte_same_as_swp(ptep_get(pte), |
1760 | swp_entry_to_pte(entry)))) { | |
044d66c1 HD |
1761 | ret = 0; |
1762 | goto out; | |
1763 | } | |
8a9f3ccd | 1764 | |
c33c7948 RR |
1765 | old_pte = ptep_get(pte); |
1766 | ||
f985fc32 | 1767 | if (unlikely(hwpoisoned || !PageUptodate(page))) { |
6b970599 | 1768 | swp_entry_t swp_entry; |
9f186f9e ML |
1769 | |
1770 | dec_mm_counter(vma->vm_mm, MM_SWAPENTS); | |
f985fc32 | 1771 | if (hwpoisoned) { |
6b970599 KW |
1772 | swp_entry = make_hwpoison_entry(swapcache); |
1773 | page = swapcache; | |
1774 | } else { | |
af19487f | 1775 | swp_entry = make_poisoned_swp_entry(); |
6b970599 KW |
1776 | } |
1777 | new_pte = swp_entry_to_pte(swp_entry); | |
9f186f9e | 1778 | ret = 0; |
6b970599 | 1779 | goto setpte; |
9f186f9e ML |
1780 | } |
1781 | ||
b53e24c4 PC |
1782 | /* |
1783 | * Some architectures may have to restore extra metadata to the page | |
1784 | * when reading from swap. This metadata may be indexed by swap entry | |
1785 | * so this must be called before swap_free(). | |
1786 | */ | |
1787 | arch_swap_restore(entry, page_folio(page)); | |
1788 | ||
78fbe906 DH |
1789 | /* See do_swap_page() */ |
1790 | BUG_ON(!PageAnon(page) && PageMappedToDisk(page)); | |
1791 | BUG_ON(PageAnon(page) && PageAnonExclusive(page)); | |
1792 | ||
b084d435 | 1793 | dec_mm_counter(vma->vm_mm, MM_SWAPENTS); |
d559db08 | 1794 | inc_mm_counter(vma->vm_mm, MM_ANONPAGES); |
1da177e4 | 1795 | get_page(page); |
00501b53 | 1796 | if (page == swapcache) { |
1493a191 DH |
1797 | rmap_t rmap_flags = RMAP_NONE; |
1798 | ||
1799 | /* | |
1800 | * See do_swap_page(): PageWriteback() would be problematic. | |
1801 | * However, we do a wait_on_page_writeback() just before this | |
1802 | * call and have the page locked. | |
1803 | */ | |
1804 | VM_BUG_ON_PAGE(PageWriteback(page), page); | |
c33c7948 | 1805 | if (pte_swp_exclusive(old_pte)) |
1493a191 DH |
1806 | rmap_flags |= RMAP_EXCLUSIVE; |
1807 | ||
1808 | page_add_anon_rmap(page, vma, addr, rmap_flags); | |
00501b53 | 1809 | } else { /* ksm created a completely new copy */ |
40f2bbf7 | 1810 | page_add_new_anon_rmap(page, vma, addr); |
b518154e | 1811 | lru_cache_add_inactive_or_unevictable(page, vma); |
00501b53 | 1812 | } |
14a762dd | 1813 | new_pte = pte_mkold(mk_pte(page, vma->vm_page_prot)); |
c33c7948 | 1814 | if (pte_swp_soft_dirty(old_pte)) |
14a762dd | 1815 | new_pte = pte_mksoft_dirty(new_pte); |
c33c7948 | 1816 | if (pte_swp_uffd_wp(old_pte)) |
14a762dd | 1817 | new_pte = pte_mkuffd_wp(new_pte); |
6b970599 | 1818 | setpte: |
14a762dd | 1819 | set_pte_at(vma->vm_mm, addr, pte, new_pte); |
1da177e4 | 1820 | swap_free(entry); |
044d66c1 | 1821 | out: |
d850fa72 HD |
1822 | if (pte) |
1823 | pte_unmap_unlock(pte, ptl); | |
9e16b7fb HD |
1824 | if (page != swapcache) { |
1825 | unlock_page(page); | |
1826 | put_page(page); | |
1827 | } | |
044d66c1 | 1828 | return ret; |
1da177e4 LT |
1829 | } |
1830 | ||
1831 | static int unuse_pte_range(struct vm_area_struct *vma, pmd_t *pmd, | |
b56a2d8a | 1832 | unsigned long addr, unsigned long end, |
10a9c496 | 1833 | unsigned int type) |
1da177e4 | 1834 | { |
d850fa72 | 1835 | pte_t *pte = NULL; |
b56a2d8a | 1836 | struct swap_info_struct *si; |
1da177e4 | 1837 | |
b56a2d8a | 1838 | si = swap_info[type]; |
1da177e4 | 1839 | do { |
f102cd8b MWO |
1840 | struct folio *folio; |
1841 | unsigned long offset; | |
3f79b187 | 1842 | unsigned char swp_count; |
d850fa72 HD |
1843 | swp_entry_t entry; |
1844 | int ret; | |
c33c7948 | 1845 | pte_t ptent; |
d850fa72 HD |
1846 | |
1847 | if (!pte++) { | |
1848 | pte = pte_offset_map(pmd, addr); | |
1849 | if (!pte) | |
1850 | break; | |
1851 | } | |
f102cd8b | 1852 | |
c33c7948 | 1853 | ptent = ptep_get_lockless(pte); |
f102cd8b | 1854 | |
c33c7948 | 1855 | if (!is_swap_pte(ptent)) |
b56a2d8a VRP |
1856 | continue; |
1857 | ||
c33c7948 | 1858 | entry = pte_to_swp_entry(ptent); |
b56a2d8a VRP |
1859 | if (swp_type(entry) != type) |
1860 | continue; | |
1861 | ||
1862 | offset = swp_offset(entry); | |
b56a2d8a | 1863 | pte_unmap(pte); |
d850fa72 HD |
1864 | pte = NULL; |
1865 | ||
f102cd8b MWO |
1866 | folio = swap_cache_get_folio(entry, vma, addr); |
1867 | if (!folio) { | |
1868 | struct page *page; | |
8c63ca5b WD |
1869 | struct vm_fault vmf = { |
1870 | .vma = vma, | |
1871 | .address = addr, | |
824ddc60 | 1872 | .real_address = addr, |
8c63ca5b WD |
1873 | .pmd = pmd, |
1874 | }; | |
1875 | ||
ebc5951e AR |
1876 | page = swapin_readahead(entry, GFP_HIGHUSER_MOVABLE, |
1877 | &vmf); | |
f102cd8b MWO |
1878 | if (page) |
1879 | folio = page_folio(page); | |
ebc5951e | 1880 | } |
f102cd8b | 1881 | if (!folio) { |
3f79b187 KS |
1882 | swp_count = READ_ONCE(si->swap_map[offset]); |
1883 | if (swp_count == 0 || swp_count == SWAP_MAP_BAD) | |
d850fa72 | 1884 | continue; |
b56a2d8a VRP |
1885 | return -ENOMEM; |
1886 | } | |
1887 | ||
f102cd8b MWO |
1888 | folio_lock(folio); |
1889 | folio_wait_writeback(folio); | |
1890 | ret = unuse_pte(vma, pmd, addr, entry, folio); | |
b56a2d8a | 1891 | if (ret < 0) { |
f102cd8b MWO |
1892 | folio_unlock(folio); |
1893 | folio_put(folio); | |
d850fa72 | 1894 | return ret; |
b56a2d8a VRP |
1895 | } |
1896 | ||
f102cd8b MWO |
1897 | folio_free_swap(folio); |
1898 | folio_unlock(folio); | |
1899 | folio_put(folio); | |
d850fa72 | 1900 | } while (addr += PAGE_SIZE, addr != end); |
b56a2d8a | 1901 | |
d850fa72 HD |
1902 | if (pte) |
1903 | pte_unmap(pte); | |
1904 | return 0; | |
1da177e4 LT |
1905 | } |
1906 | ||
1907 | static inline int unuse_pmd_range(struct vm_area_struct *vma, pud_t *pud, | |
1908 | unsigned long addr, unsigned long end, | |
10a9c496 | 1909 | unsigned int type) |
1da177e4 LT |
1910 | { |
1911 | pmd_t *pmd; | |
1912 | unsigned long next; | |
8a9f3ccd | 1913 | int ret; |
1da177e4 LT |
1914 | |
1915 | pmd = pmd_offset(pud, addr); | |
1916 | do { | |
dc644a07 | 1917 | cond_resched(); |
1da177e4 | 1918 | next = pmd_addr_end(addr, end); |
10a9c496 | 1919 | ret = unuse_pte_range(vma, pmd, addr, next, type); |
8a9f3ccd BS |
1920 | if (ret) |
1921 | return ret; | |
1da177e4 LT |
1922 | } while (pmd++, addr = next, addr != end); |
1923 | return 0; | |
1924 | } | |
1925 | ||
c2febafc | 1926 | static inline int unuse_pud_range(struct vm_area_struct *vma, p4d_t *p4d, |
1da177e4 | 1927 | unsigned long addr, unsigned long end, |
10a9c496 | 1928 | unsigned int type) |
1da177e4 LT |
1929 | { |
1930 | pud_t *pud; | |
1931 | unsigned long next; | |
8a9f3ccd | 1932 | int ret; |
1da177e4 | 1933 | |
c2febafc | 1934 | pud = pud_offset(p4d, addr); |
1da177e4 LT |
1935 | do { |
1936 | next = pud_addr_end(addr, end); | |
1937 | if (pud_none_or_clear_bad(pud)) | |
1938 | continue; | |
10a9c496 | 1939 | ret = unuse_pmd_range(vma, pud, addr, next, type); |
8a9f3ccd BS |
1940 | if (ret) |
1941 | return ret; | |
1da177e4 LT |
1942 | } while (pud++, addr = next, addr != end); |
1943 | return 0; | |
1944 | } | |
1945 | ||
c2febafc KS |
1946 | static inline int unuse_p4d_range(struct vm_area_struct *vma, pgd_t *pgd, |
1947 | unsigned long addr, unsigned long end, | |
10a9c496 | 1948 | unsigned int type) |
c2febafc KS |
1949 | { |
1950 | p4d_t *p4d; | |
1951 | unsigned long next; | |
1952 | int ret; | |
1953 | ||
1954 | p4d = p4d_offset(pgd, addr); | |
1955 | do { | |
1956 | next = p4d_addr_end(addr, end); | |
1957 | if (p4d_none_or_clear_bad(p4d)) | |
1958 | continue; | |
10a9c496 | 1959 | ret = unuse_pud_range(vma, p4d, addr, next, type); |
c2febafc KS |
1960 | if (ret) |
1961 | return ret; | |
1962 | } while (p4d++, addr = next, addr != end); | |
1963 | return 0; | |
1964 | } | |
1965 | ||
10a9c496 | 1966 | static int unuse_vma(struct vm_area_struct *vma, unsigned int type) |
1da177e4 LT |
1967 | { |
1968 | pgd_t *pgd; | |
1969 | unsigned long addr, end, next; | |
8a9f3ccd | 1970 | int ret; |
1da177e4 | 1971 | |
b56a2d8a VRP |
1972 | addr = vma->vm_start; |
1973 | end = vma->vm_end; | |
1da177e4 LT |
1974 | |
1975 | pgd = pgd_offset(vma->vm_mm, addr); | |
1976 | do { | |
1977 | next = pgd_addr_end(addr, end); | |
1978 | if (pgd_none_or_clear_bad(pgd)) | |
1979 | continue; | |
10a9c496 | 1980 | ret = unuse_p4d_range(vma, pgd, addr, next, type); |
8a9f3ccd BS |
1981 | if (ret) |
1982 | return ret; | |
1da177e4 LT |
1983 | } while (pgd++, addr = next, addr != end); |
1984 | return 0; | |
1985 | } | |
1986 | ||
10a9c496 | 1987 | static int unuse_mm(struct mm_struct *mm, unsigned int type) |
1da177e4 LT |
1988 | { |
1989 | struct vm_area_struct *vma; | |
8a9f3ccd | 1990 | int ret = 0; |
208c09db | 1991 | VMA_ITERATOR(vmi, mm, 0); |
1da177e4 | 1992 | |
d8ed45c5 | 1993 | mmap_read_lock(mm); |
208c09db | 1994 | for_each_vma(vmi, vma) { |
b56a2d8a | 1995 | if (vma->anon_vma) { |
10a9c496 | 1996 | ret = unuse_vma(vma, type); |
b56a2d8a VRP |
1997 | if (ret) |
1998 | break; | |
1999 | } | |
208c09db | 2000 | |
dc644a07 | 2001 | cond_resched(); |
1da177e4 | 2002 | } |
d8ed45c5 | 2003 | mmap_read_unlock(mm); |
b56a2d8a | 2004 | return ret; |
1da177e4 LT |
2005 | } |
2006 | ||
2007 | /* | |
3c3115ad ML |
2008 | * Scan swap_map from current position to next entry still in use. |
2009 | * Return 0 if there are no inuse entries after prev till end of | |
2010 | * the map. | |
1da177e4 | 2011 | */ |
6eb396dc | 2012 | static unsigned int find_next_to_unuse(struct swap_info_struct *si, |
10a9c496 | 2013 | unsigned int prev) |
1da177e4 | 2014 | { |
b56a2d8a | 2015 | unsigned int i; |
8d69aaee | 2016 | unsigned char count; |
1da177e4 LT |
2017 | |
2018 | /* | |
5d337b91 | 2019 | * No need for swap_lock here: we're just looking |
1da177e4 LT |
2020 | * for whether an entry is in use, not modifying it; false |
2021 | * hits are okay, and sys_swapoff() has already prevented new | |
5d337b91 | 2022 | * allocations from this area (while holding swap_lock). |
1da177e4 | 2023 | */ |
b56a2d8a | 2024 | for (i = prev + 1; i < si->max; i++) { |
4db0c3c2 | 2025 | count = READ_ONCE(si->swap_map[i]); |
355cfa73 | 2026 | if (count && swap_count(count) != SWAP_MAP_BAD) |
10a9c496 | 2027 | break; |
dc644a07 HD |
2028 | if ((i % LATENCY_LIMIT) == 0) |
2029 | cond_resched(); | |
1da177e4 | 2030 | } |
b56a2d8a VRP |
2031 | |
2032 | if (i == si->max) | |
2033 | i = 0; | |
2034 | ||
1da177e4 LT |
2035 | return i; |
2036 | } | |
2037 | ||
10a9c496 | 2038 | static int try_to_unuse(unsigned int type) |
1da177e4 | 2039 | { |
b56a2d8a VRP |
2040 | struct mm_struct *prev_mm; |
2041 | struct mm_struct *mm; | |
2042 | struct list_head *p; | |
2043 | int retval = 0; | |
efa90a98 | 2044 | struct swap_info_struct *si = swap_info[type]; |
000085b9 | 2045 | struct folio *folio; |
1da177e4 | 2046 | swp_entry_t entry; |
b56a2d8a | 2047 | unsigned int i; |
1da177e4 | 2048 | |
21820948 | 2049 | if (!READ_ONCE(si->inuse_pages)) |
b56a2d8a | 2050 | return 0; |
1da177e4 | 2051 | |
b56a2d8a | 2052 | retry: |
10a9c496 | 2053 | retval = shmem_unuse(type); |
b56a2d8a | 2054 | if (retval) |
10a9c496 | 2055 | return retval; |
b56a2d8a VRP |
2056 | |
2057 | prev_mm = &init_mm; | |
2058 | mmget(prev_mm); | |
2059 | ||
2060 | spin_lock(&mmlist_lock); | |
2061 | p = &init_mm.mmlist; | |
21820948 | 2062 | while (READ_ONCE(si->inuse_pages) && |
64165b1a HD |
2063 | !signal_pending(current) && |
2064 | (p = p->next) != &init_mm.mmlist) { | |
1da177e4 | 2065 | |
b56a2d8a VRP |
2066 | mm = list_entry(p, struct mm_struct, mmlist); |
2067 | if (!mmget_not_zero(mm)) | |
2068 | continue; | |
2069 | spin_unlock(&mmlist_lock); | |
2070 | mmput(prev_mm); | |
2071 | prev_mm = mm; | |
10a9c496 | 2072 | retval = unuse_mm(mm, type); |
b56a2d8a VRP |
2073 | if (retval) { |
2074 | mmput(prev_mm); | |
10a9c496 | 2075 | return retval; |
1da177e4 LT |
2076 | } |
2077 | ||
2078 | /* | |
b56a2d8a VRP |
2079 | * Make sure that we aren't completely killing |
2080 | * interactive performance. | |
1da177e4 | 2081 | */ |
b56a2d8a VRP |
2082 | cond_resched(); |
2083 | spin_lock(&mmlist_lock); | |
2084 | } | |
2085 | spin_unlock(&mmlist_lock); | |
1da177e4 | 2086 | |
b56a2d8a | 2087 | mmput(prev_mm); |
1da177e4 | 2088 | |
b56a2d8a | 2089 | i = 0; |
21820948 | 2090 | while (READ_ONCE(si->inuse_pages) && |
64165b1a | 2091 | !signal_pending(current) && |
10a9c496 | 2092 | (i = find_next_to_unuse(si, i)) != 0) { |
1da177e4 | 2093 | |
b56a2d8a | 2094 | entry = swp_entry(type, i); |
000085b9 | 2095 | folio = filemap_get_folio(swap_address_space(entry), i); |
66dabbb6 | 2096 | if (IS_ERR(folio)) |
b56a2d8a | 2097 | continue; |
68bdc8d6 HD |
2098 | |
2099 | /* | |
000085b9 MWO |
2100 | * It is conceivable that a racing task removed this folio from |
2101 | * swap cache just before we acquired the page lock. The folio | |
b56a2d8a | 2102 | * might even be back in swap cache on another swap area. But |
000085b9 | 2103 | * that is okay, folio_free_swap() only removes stale folios. |
1da177e4 | 2104 | */ |
000085b9 MWO |
2105 | folio_lock(folio); |
2106 | folio_wait_writeback(folio); | |
2107 | folio_free_swap(folio); | |
2108 | folio_unlock(folio); | |
2109 | folio_put(folio); | |
1da177e4 LT |
2110 | } |
2111 | ||
b56a2d8a VRP |
2112 | /* |
2113 | * Lets check again to see if there are still swap entries in the map. | |
2114 | * If yes, we would need to do retry the unuse logic again. | |
2115 | * Under global memory pressure, swap entries can be reinserted back | |
2116 | * into process space after the mmlist loop above passes over them. | |
dd862deb | 2117 | * |
e2e3fdc7 MWO |
2118 | * Limit the number of retries? No: when mmget_not_zero() |
2119 | * above fails, that mm is likely to be freeing swap from | |
2120 | * exit_mmap(), which proceeds at its own independent pace; | |
2121 | * and even shmem_writepage() could have been preempted after | |
2122 | * folio_alloc_swap(), temporarily hiding that swap. It's easy | |
2123 | * and robust (though cpu-intensive) just to keep retrying. | |
b56a2d8a | 2124 | */ |
21820948 | 2125 | if (READ_ONCE(si->inuse_pages)) { |
64165b1a HD |
2126 | if (!signal_pending(current)) |
2127 | goto retry; | |
10a9c496 | 2128 | return -EINTR; |
64165b1a | 2129 | } |
10a9c496 CH |
2130 | |
2131 | return 0; | |
1da177e4 LT |
2132 | } |
2133 | ||
2134 | /* | |
5d337b91 HD |
2135 | * After a successful try_to_unuse, if no swap is now in use, we know |
2136 | * we can empty the mmlist. swap_lock must be held on entry and exit. | |
2137 | * Note that mmlist_lock nests inside swap_lock, and an mm must be | |
1da177e4 LT |
2138 | * added to the mmlist just after page_duplicate - before would be racy. |
2139 | */ | |
2140 | static void drain_mmlist(void) | |
2141 | { | |
2142 | struct list_head *p, *next; | |
efa90a98 | 2143 | unsigned int type; |
1da177e4 | 2144 | |
efa90a98 HD |
2145 | for (type = 0; type < nr_swapfiles; type++) |
2146 | if (swap_info[type]->inuse_pages) | |
1da177e4 LT |
2147 | return; |
2148 | spin_lock(&mmlist_lock); | |
2149 | list_for_each_safe(p, next, &init_mm.mmlist) | |
2150 | list_del_init(p); | |
2151 | spin_unlock(&mmlist_lock); | |
2152 | } | |
2153 | ||
1da177e4 LT |
2154 | /* |
2155 | * Free all of a swapdev's extent information | |
2156 | */ | |
2157 | static void destroy_swap_extents(struct swap_info_struct *sis) | |
2158 | { | |
4efaceb1 AL |
2159 | while (!RB_EMPTY_ROOT(&sis->swap_extent_root)) { |
2160 | struct rb_node *rb = sis->swap_extent_root.rb_node; | |
2161 | struct swap_extent *se = rb_entry(rb, struct swap_extent, rb_node); | |
1da177e4 | 2162 | |
4efaceb1 | 2163 | rb_erase(rb, &sis->swap_extent_root); |
1da177e4 LT |
2164 | kfree(se); |
2165 | } | |
62c230bc | 2166 | |
bc4ae27d | 2167 | if (sis->flags & SWP_ACTIVATED) { |
62c230bc MG |
2168 | struct file *swap_file = sis->swap_file; |
2169 | struct address_space *mapping = swap_file->f_mapping; | |
2170 | ||
bc4ae27d OS |
2171 | sis->flags &= ~SWP_ACTIVATED; |
2172 | if (mapping->a_ops->swap_deactivate) | |
2173 | mapping->a_ops->swap_deactivate(swap_file); | |
62c230bc | 2174 | } |
1da177e4 LT |
2175 | } |
2176 | ||
2177 | /* | |
2178 | * Add a block range (and the corresponding page range) into this swapdev's | |
4efaceb1 | 2179 | * extent tree. |
1da177e4 | 2180 | * |
11d31886 | 2181 | * This function rather assumes that it is called in ascending page order. |
1da177e4 | 2182 | */ |
a509bc1a | 2183 | int |
1da177e4 LT |
2184 | add_swap_extent(struct swap_info_struct *sis, unsigned long start_page, |
2185 | unsigned long nr_pages, sector_t start_block) | |
2186 | { | |
4efaceb1 | 2187 | struct rb_node **link = &sis->swap_extent_root.rb_node, *parent = NULL; |
1da177e4 LT |
2188 | struct swap_extent *se; |
2189 | struct swap_extent *new_se; | |
4efaceb1 AL |
2190 | |
2191 | /* | |
2192 | * place the new node at the right most since the | |
2193 | * function is called in ascending page order. | |
2194 | */ | |
2195 | while (*link) { | |
2196 | parent = *link; | |
2197 | link = &parent->rb_right; | |
2198 | } | |
2199 | ||
2200 | if (parent) { | |
2201 | se = rb_entry(parent, struct swap_extent, rb_node); | |
11d31886 HD |
2202 | BUG_ON(se->start_page + se->nr_pages != start_page); |
2203 | if (se->start_block + se->nr_pages == start_block) { | |
1da177e4 LT |
2204 | /* Merge it */ |
2205 | se->nr_pages += nr_pages; | |
2206 | return 0; | |
2207 | } | |
1da177e4 LT |
2208 | } |
2209 | ||
4efaceb1 | 2210 | /* No merge, insert a new extent. */ |
1da177e4 LT |
2211 | new_se = kmalloc(sizeof(*se), GFP_KERNEL); |
2212 | if (new_se == NULL) | |
2213 | return -ENOMEM; | |
2214 | new_se->start_page = start_page; | |
2215 | new_se->nr_pages = nr_pages; | |
2216 | new_se->start_block = start_block; | |
2217 | ||
4efaceb1 AL |
2218 | rb_link_node(&new_se->rb_node, parent, link); |
2219 | rb_insert_color(&new_se->rb_node, &sis->swap_extent_root); | |
53092a74 | 2220 | return 1; |
1da177e4 | 2221 | } |
aa8aa8a3 | 2222 | EXPORT_SYMBOL_GPL(add_swap_extent); |
1da177e4 LT |
2223 | |
2224 | /* | |
2225 | * A `swap extent' is a simple thing which maps a contiguous range of pages | |
ff351f4b ML |
2226 | * onto a contiguous range of disk blocks. A rbtree of swap extents is |
2227 | * built at swapon time and is then used at swap_writepage/swap_readpage | |
1da177e4 LT |
2228 | * time for locating where on disk a page belongs. |
2229 | * | |
2230 | * If the swapfile is an S_ISBLK block device, a single extent is installed. | |
2231 | * This is done so that the main operating code can treat S_ISBLK and S_ISREG | |
2232 | * swap files identically. | |
2233 | * | |
2234 | * Whether the swapdev is an S_ISREG file or an S_ISBLK blockdev, the swap | |
ff351f4b | 2235 | * extent rbtree operates in PAGE_SIZE disk blocks. Both S_ISREG and S_ISBLK |
1da177e4 LT |
2236 | * swapfiles are handled *identically* after swapon time. |
2237 | * | |
2238 | * For S_ISREG swapfiles, setup_swap_extents() will walk all the file's blocks | |
ff351f4b ML |
2239 | * and will parse them into a rbtree, in PAGE_SIZE chunks. If some stray |
2240 | * blocks are found which do not fall within the PAGE_SIZE alignment | |
1da177e4 LT |
2241 | * requirements, they are simply tossed out - we will never use those blocks |
2242 | * for swapping. | |
2243 | * | |
1638045c DW |
2244 | * For all swap devices we set S_SWAPFILE across the life of the swapon. This |
2245 | * prevents users from writing to the swap device, which will corrupt memory. | |
1da177e4 LT |
2246 | * |
2247 | * The amount of disk space which a single swap extent represents varies. | |
2248 | * Typically it is in the 1-4 megabyte range. So we can have hundreds of | |
ff351f4b | 2249 | * extents in the rbtree. - akpm. |
1da177e4 | 2250 | */ |
53092a74 | 2251 | static int setup_swap_extents(struct swap_info_struct *sis, sector_t *span) |
1da177e4 | 2252 | { |
62c230bc MG |
2253 | struct file *swap_file = sis->swap_file; |
2254 | struct address_space *mapping = swap_file->f_mapping; | |
2255 | struct inode *inode = mapping->host; | |
1da177e4 LT |
2256 | int ret; |
2257 | ||
1da177e4 LT |
2258 | if (S_ISBLK(inode->i_mode)) { |
2259 | ret = add_swap_extent(sis, 0, sis->max, 0); | |
53092a74 | 2260 | *span = sis->pages; |
a509bc1a | 2261 | return ret; |
1da177e4 LT |
2262 | } |
2263 | ||
62c230bc | 2264 | if (mapping->a_ops->swap_activate) { |
a509bc1a | 2265 | ret = mapping->a_ops->swap_activate(sis, swap_file, span); |
4b60c0ff N |
2266 | if (ret < 0) |
2267 | return ret; | |
2268 | sis->flags |= SWP_ACTIVATED; | |
e1209d3a N |
2269 | if ((sis->flags & SWP_FS_OPS) && |
2270 | sio_pool_init() != 0) { | |
2271 | destroy_swap_extents(sis); | |
2272 | return -ENOMEM; | |
62c230bc | 2273 | } |
a509bc1a | 2274 | return ret; |
62c230bc MG |
2275 | } |
2276 | ||
a509bc1a | 2277 | return generic_swapfile_activate(sis, swap_file, span); |
1da177e4 LT |
2278 | } |
2279 | ||
a2468cc9 AL |
2280 | static int swap_node(struct swap_info_struct *p) |
2281 | { | |
2282 | struct block_device *bdev; | |
2283 | ||
2284 | if (p->bdev) | |
2285 | bdev = p->bdev; | |
2286 | else | |
2287 | bdev = p->swap_file->f_inode->i_sb->s_bdev; | |
2288 | ||
2289 | return bdev ? bdev->bd_disk->node_id : NUMA_NO_NODE; | |
2290 | } | |
2291 | ||
eb085574 HY |
2292 | static void setup_swap_info(struct swap_info_struct *p, int prio, |
2293 | unsigned char *swap_map, | |
2294 | struct swap_cluster_info *cluster_info) | |
40531542 | 2295 | { |
a2468cc9 AL |
2296 | int i; |
2297 | ||
40531542 CEB |
2298 | if (prio >= 0) |
2299 | p->prio = prio; | |
2300 | else | |
2301 | p->prio = --least_priority; | |
18ab4d4c DS |
2302 | /* |
2303 | * the plist prio is negated because plist ordering is | |
2304 | * low-to-high, while swap ordering is high-to-low | |
2305 | */ | |
2306 | p->list.prio = -p->prio; | |
a2468cc9 AL |
2307 | for_each_node(i) { |
2308 | if (p->prio >= 0) | |
2309 | p->avail_lists[i].prio = -p->prio; | |
2310 | else { | |
2311 | if (swap_node(p) == i) | |
2312 | p->avail_lists[i].prio = 1; | |
2313 | else | |
2314 | p->avail_lists[i].prio = -p->prio; | |
2315 | } | |
2316 | } | |
40531542 | 2317 | p->swap_map = swap_map; |
2a8f9449 | 2318 | p->cluster_info = cluster_info; |
eb085574 HY |
2319 | } |
2320 | ||
2321 | static void _enable_swap_info(struct swap_info_struct *p) | |
2322 | { | |
63d8620e | 2323 | p->flags |= SWP_WRITEOK; |
ec8acf20 | 2324 | atomic_long_add(p->pages, &nr_swap_pages); |
40531542 CEB |
2325 | total_swap_pages += p->pages; |
2326 | ||
adfab836 | 2327 | assert_spin_locked(&swap_lock); |
adfab836 | 2328 | /* |
18ab4d4c DS |
2329 | * both lists are plists, and thus priority ordered. |
2330 | * swap_active_head needs to be priority ordered for swapoff(), | |
2331 | * which on removal of any swap_info_struct with an auto-assigned | |
2332 | * (i.e. negative) priority increments the auto-assigned priority | |
2333 | * of any lower-priority swap_info_structs. | |
e2e3fdc7 | 2334 | * swap_avail_head needs to be priority ordered for folio_alloc_swap(), |
18ab4d4c DS |
2335 | * which allocates swap pages from the highest available priority |
2336 | * swap_info_struct. | |
adfab836 | 2337 | */ |
18ab4d4c | 2338 | plist_add(&p->list, &swap_active_head); |
c70699e5 MW |
2339 | |
2340 | /* add to available list iff swap device is not full */ | |
2341 | if (p->highest_bit) | |
2342 | add_to_avail_list(p); | |
cf0cac0a CEB |
2343 | } |
2344 | ||
2345 | static void enable_swap_info(struct swap_info_struct *p, int prio, | |
2346 | unsigned char *swap_map, | |
42c06a0e | 2347 | struct swap_cluster_info *cluster_info) |
cf0cac0a | 2348 | { |
42c06a0e JW |
2349 | zswap_swapon(p->type); |
2350 | ||
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; |
1da177e4 LT |
2393 | struct file *swap_file, *victim; |
2394 | struct address_space *mapping; | |
2395 | struct inode *inode; | |
91a27b2a | 2396 | struct filename *pathname; |
adfab836 | 2397 | int err, found = 0; |
5b808a23 | 2398 | unsigned int old_block_size; |
886bb7e9 | 2399 | |
1da177e4 LT |
2400 | if (!capable(CAP_SYS_ADMIN)) |
2401 | return -EPERM; | |
2402 | ||
191c5424 AV |
2403 | BUG_ON(!current->mm); |
2404 | ||
1da177e4 | 2405 | pathname = getname(specialfile); |
1da177e4 | 2406 | if (IS_ERR(pathname)) |
f58b59c1 | 2407 | return PTR_ERR(pathname); |
1da177e4 | 2408 | |
669abf4e | 2409 | victim = file_open_name(pathname, O_RDWR|O_LARGEFILE, 0); |
1da177e4 LT |
2410 | err = PTR_ERR(victim); |
2411 | if (IS_ERR(victim)) | |
2412 | goto out; | |
2413 | ||
2414 | mapping = victim->f_mapping; | |
5d337b91 | 2415 | spin_lock(&swap_lock); |
18ab4d4c | 2416 | plist_for_each_entry(p, &swap_active_head, list) { |
22c6f8fd | 2417 | if (p->flags & SWP_WRITEOK) { |
adfab836 DS |
2418 | if (p->swap_file->f_mapping == mapping) { |
2419 | found = 1; | |
1da177e4 | 2420 | break; |
adfab836 | 2421 | } |
1da177e4 | 2422 | } |
1da177e4 | 2423 | } |
adfab836 | 2424 | if (!found) { |
1da177e4 | 2425 | err = -EINVAL; |
5d337b91 | 2426 | spin_unlock(&swap_lock); |
1da177e4 LT |
2427 | goto out_dput; |
2428 | } | |
191c5424 | 2429 | if (!security_vm_enough_memory_mm(current->mm, p->pages)) |
1da177e4 LT |
2430 | vm_unacct_memory(p->pages); |
2431 | else { | |
2432 | err = -ENOMEM; | |
5d337b91 | 2433 | spin_unlock(&swap_lock); |
1da177e4 LT |
2434 | goto out_dput; |
2435 | } | |
ec8acf20 | 2436 | spin_lock(&p->lock); |
6fe7d6b9 | 2437 | del_from_avail_list(p); |
78ecba08 | 2438 | if (p->prio < 0) { |
adfab836 | 2439 | struct swap_info_struct *si = p; |
a2468cc9 | 2440 | int nid; |
adfab836 | 2441 | |
18ab4d4c | 2442 | plist_for_each_entry_continue(si, &swap_active_head, list) { |
adfab836 | 2443 | si->prio++; |
18ab4d4c | 2444 | si->list.prio--; |
a2468cc9 AL |
2445 | for_each_node(nid) { |
2446 | if (si->avail_lists[nid].prio != 1) | |
2447 | si->avail_lists[nid].prio--; | |
2448 | } | |
adfab836 | 2449 | } |
78ecba08 HD |
2450 | least_priority++; |
2451 | } | |
18ab4d4c | 2452 | plist_del(&p->list, &swap_active_head); |
ec8acf20 | 2453 | atomic_long_sub(p->pages, &nr_swap_pages); |
1da177e4 LT |
2454 | total_swap_pages -= p->pages; |
2455 | p->flags &= ~SWP_WRITEOK; | |
ec8acf20 | 2456 | spin_unlock(&p->lock); |
5d337b91 | 2457 | spin_unlock(&swap_lock); |
fb4f88dc | 2458 | |
039939a6 TC |
2459 | disable_swap_slots_cache_lock(); |
2460 | ||
e1e12d2f | 2461 | set_current_oom_origin(); |
10a9c496 | 2462 | err = try_to_unuse(p->type); |
e1e12d2f | 2463 | clear_current_oom_origin(); |
1da177e4 | 2464 | |
1da177e4 LT |
2465 | if (err) { |
2466 | /* re-insert swap space back into swap_list */ | |
cf0cac0a | 2467 | reinsert_swap_info(p); |
039939a6 | 2468 | reenable_swap_slots_cache_unlock(); |
1da177e4 LT |
2469 | goto out_dput; |
2470 | } | |
52b7efdb | 2471 | |
039939a6 TC |
2472 | reenable_swap_slots_cache_unlock(); |
2473 | ||
eb085574 | 2474 | /* |
63d8620e ML |
2475 | * Wait for swap operations protected by get/put_swap_device() |
2476 | * to complete. | |
2477 | * | |
2478 | * We need synchronize_rcu() here to protect the accessing to | |
2479 | * the swap cache data structure. | |
eb085574 | 2480 | */ |
63d8620e | 2481 | percpu_ref_kill(&p->users); |
eb085574 | 2482 | synchronize_rcu(); |
63d8620e | 2483 | wait_for_completion(&p->comp); |
eb085574 | 2484 | |
815c2c54 SL |
2485 | flush_work(&p->discard_work); |
2486 | ||
5d337b91 | 2487 | destroy_swap_extents(p); |
570a335b HD |
2488 | if (p->flags & SWP_CONTINUED) |
2489 | free_swap_count_continuations(p); | |
2490 | ||
10f0d2a5 | 2491 | if (!p->bdev || !bdev_nonrot(p->bdev)) |
81a0298b HY |
2492 | atomic_dec(&nr_rotate_swap); |
2493 | ||
fc0abb14 | 2494 | mutex_lock(&swapon_mutex); |
5d337b91 | 2495 | spin_lock(&swap_lock); |
ec8acf20 | 2496 | spin_lock(&p->lock); |
5d337b91 HD |
2497 | drain_mmlist(); |
2498 | ||
bb243f7d | 2499 | /* wait for anyone still in scan_swap_map_slots */ |
52b7efdb HD |
2500 | p->highest_bit = 0; /* cuts scans short */ |
2501 | while (p->flags >= SWP_SCANNING) { | |
ec8acf20 | 2502 | spin_unlock(&p->lock); |
5d337b91 | 2503 | spin_unlock(&swap_lock); |
13e4b57f | 2504 | schedule_timeout_uninterruptible(1); |
5d337b91 | 2505 | spin_lock(&swap_lock); |
ec8acf20 | 2506 | spin_lock(&p->lock); |
52b7efdb | 2507 | } |
52b7efdb | 2508 | |
1da177e4 | 2509 | swap_file = p->swap_file; |
5b808a23 | 2510 | old_block_size = p->old_block_size; |
1da177e4 LT |
2511 | p->swap_file = NULL; |
2512 | p->max = 0; | |
2513 | swap_map = p->swap_map; | |
2514 | p->swap_map = NULL; | |
2a8f9449 SL |
2515 | cluster_info = p->cluster_info; |
2516 | p->cluster_info = NULL; | |
ec8acf20 | 2517 | spin_unlock(&p->lock); |
5d337b91 | 2518 | spin_unlock(&swap_lock); |
8a84802e | 2519 | arch_swap_invalidate_area(p->type); |
42c06a0e | 2520 | zswap_swapoff(p->type); |
fc0abb14 | 2521 | mutex_unlock(&swapon_mutex); |
ebc2a1a6 SL |
2522 | free_percpu(p->percpu_cluster); |
2523 | p->percpu_cluster = NULL; | |
49070588 HY |
2524 | free_percpu(p->cluster_next_cpu); |
2525 | p->cluster_next_cpu = NULL; | |
1da177e4 | 2526 | vfree(swap_map); |
54f180d3 | 2527 | kvfree(cluster_info); |
2de1a7e4 | 2528 | /* Destroy swap account information */ |
adfab836 | 2529 | swap_cgroup_swapoff(p->type); |
4b3ef9da | 2530 | exit_swap_address_space(p->type); |
27a7faa0 | 2531 | |
1da177e4 | 2532 | inode = mapping->host; |
4c6bca43 JK |
2533 | if (p->bdev_handle) { |
2534 | set_blocksize(p->bdev, old_block_size); | |
2535 | bdev_release(p->bdev_handle); | |
2536 | p->bdev_handle = NULL; | |
1da177e4 | 2537 | } |
1638045c DW |
2538 | |
2539 | inode_lock(inode); | |
2540 | inode->i_flags &= ~S_SWAPFILE; | |
2541 | inode_unlock(inode); | |
1da177e4 | 2542 | filp_close(swap_file, NULL); |
f893ab41 WY |
2543 | |
2544 | /* | |
2545 | * Clear the SWP_USED flag after all resources are freed so that swapon | |
2546 | * can reuse this swap_info in alloc_swap_info() safely. It is ok to | |
2547 | * not hold p->lock after we cleared its SWP_WRITEOK. | |
2548 | */ | |
2549 | spin_lock(&swap_lock); | |
2550 | p->flags = 0; | |
2551 | spin_unlock(&swap_lock); | |
2552 | ||
1da177e4 | 2553 | err = 0; |
66d7dd51 KS |
2554 | atomic_inc(&proc_poll_event); |
2555 | wake_up_interruptible(&proc_poll_wait); | |
1da177e4 LT |
2556 | |
2557 | out_dput: | |
2558 | filp_close(victim, NULL); | |
2559 | out: | |
f58b59c1 | 2560 | putname(pathname); |
1da177e4 LT |
2561 | return err; |
2562 | } | |
2563 | ||
2564 | #ifdef CONFIG_PROC_FS | |
9dd95748 | 2565 | static __poll_t swaps_poll(struct file *file, poll_table *wait) |
66d7dd51 | 2566 | { |
f1514638 | 2567 | struct seq_file *seq = file->private_data; |
66d7dd51 KS |
2568 | |
2569 | poll_wait(file, &proc_poll_wait, wait); | |
2570 | ||
f1514638 KS |
2571 | if (seq->poll_event != atomic_read(&proc_poll_event)) { |
2572 | seq->poll_event = atomic_read(&proc_poll_event); | |
a9a08845 | 2573 | return EPOLLIN | EPOLLRDNORM | EPOLLERR | EPOLLPRI; |
66d7dd51 KS |
2574 | } |
2575 | ||
a9a08845 | 2576 | return EPOLLIN | EPOLLRDNORM; |
66d7dd51 KS |
2577 | } |
2578 | ||
1da177e4 LT |
2579 | /* iterator */ |
2580 | static void *swap_start(struct seq_file *swap, loff_t *pos) | |
2581 | { | |
efa90a98 HD |
2582 | struct swap_info_struct *si; |
2583 | int type; | |
1da177e4 LT |
2584 | loff_t l = *pos; |
2585 | ||
fc0abb14 | 2586 | mutex_lock(&swapon_mutex); |
1da177e4 | 2587 | |
881e4aab SS |
2588 | if (!l) |
2589 | return SEQ_START_TOKEN; | |
2590 | ||
c10d38cc | 2591 | for (type = 0; (si = swap_type_to_swap_info(type)); type++) { |
efa90a98 | 2592 | if (!(si->flags & SWP_USED) || !si->swap_map) |
1da177e4 | 2593 | continue; |
881e4aab | 2594 | if (!--l) |
efa90a98 | 2595 | return si; |
1da177e4 LT |
2596 | } |
2597 | ||
2598 | return NULL; | |
2599 | } | |
2600 | ||
2601 | static void *swap_next(struct seq_file *swap, void *v, loff_t *pos) | |
2602 | { | |
efa90a98 HD |
2603 | struct swap_info_struct *si = v; |
2604 | int type; | |
1da177e4 | 2605 | |
881e4aab | 2606 | if (v == SEQ_START_TOKEN) |
efa90a98 HD |
2607 | type = 0; |
2608 | else | |
2609 | type = si->type + 1; | |
881e4aab | 2610 | |
10c8d69f | 2611 | ++(*pos); |
c10d38cc | 2612 | for (; (si = swap_type_to_swap_info(type)); type++) { |
efa90a98 | 2613 | if (!(si->flags & SWP_USED) || !si->swap_map) |
1da177e4 | 2614 | continue; |
efa90a98 | 2615 | return si; |
1da177e4 LT |
2616 | } |
2617 | ||
2618 | return NULL; | |
2619 | } | |
2620 | ||
2621 | static void swap_stop(struct seq_file *swap, void *v) | |
2622 | { | |
fc0abb14 | 2623 | mutex_unlock(&swapon_mutex); |
1da177e4 LT |
2624 | } |
2625 | ||
2626 | static int swap_show(struct seq_file *swap, void *v) | |
2627 | { | |
efa90a98 | 2628 | struct swap_info_struct *si = v; |
1da177e4 LT |
2629 | struct file *file; |
2630 | int len; | |
642929a2 | 2631 | unsigned long bytes, inuse; |
1da177e4 | 2632 | |
efa90a98 | 2633 | if (si == SEQ_START_TOKEN) { |
68d68ff6 | 2634 | seq_puts(swap, "Filename\t\t\t\tType\t\tSize\t\tUsed\t\tPriority\n"); |
881e4aab SS |
2635 | return 0; |
2636 | } | |
1da177e4 | 2637 | |
00cde042 Z |
2638 | bytes = K(si->pages); |
2639 | inuse = K(READ_ONCE(si->inuse_pages)); | |
6f793940 | 2640 | |
efa90a98 | 2641 | file = si->swap_file; |
2726d566 | 2642 | len = seq_file_path(swap, file, " \t\n\\"); |
642929a2 | 2643 | seq_printf(swap, "%*s%s\t%lu\t%s%lu\t%s%d\n", |
886bb7e9 | 2644 | len < 40 ? 40 - len : 1, " ", |
496ad9aa | 2645 | S_ISBLK(file_inode(file)->i_mode) ? |
1da177e4 | 2646 | "partition" : "file\t", |
6f793940 RD |
2647 | bytes, bytes < 10000000 ? "\t" : "", |
2648 | inuse, inuse < 10000000 ? "\t" : "", | |
efa90a98 | 2649 | si->prio); |
1da177e4 LT |
2650 | return 0; |
2651 | } | |
2652 | ||
15ad7cdc | 2653 | static const struct seq_operations swaps_op = { |
1da177e4 LT |
2654 | .start = swap_start, |
2655 | .next = swap_next, | |
2656 | .stop = swap_stop, | |
2657 | .show = swap_show | |
2658 | }; | |
2659 | ||
2660 | static int swaps_open(struct inode *inode, struct file *file) | |
2661 | { | |
f1514638 | 2662 | struct seq_file *seq; |
66d7dd51 KS |
2663 | int ret; |
2664 | ||
66d7dd51 | 2665 | ret = seq_open(file, &swaps_op); |
f1514638 | 2666 | if (ret) |
66d7dd51 | 2667 | return ret; |
66d7dd51 | 2668 | |
f1514638 KS |
2669 | seq = file->private_data; |
2670 | seq->poll_event = atomic_read(&proc_poll_event); | |
2671 | return 0; | |
1da177e4 LT |
2672 | } |
2673 | ||
97a32539 | 2674 | static const struct proc_ops swaps_proc_ops = { |
d919b33d | 2675 | .proc_flags = PROC_ENTRY_PERMANENT, |
97a32539 AD |
2676 | .proc_open = swaps_open, |
2677 | .proc_read = seq_read, | |
2678 | .proc_lseek = seq_lseek, | |
2679 | .proc_release = seq_release, | |
2680 | .proc_poll = swaps_poll, | |
1da177e4 LT |
2681 | }; |
2682 | ||
2683 | static int __init procswaps_init(void) | |
2684 | { | |
97a32539 | 2685 | proc_create("swaps", 0, NULL, &swaps_proc_ops); |
1da177e4 LT |
2686 | return 0; |
2687 | } | |
2688 | __initcall(procswaps_init); | |
2689 | #endif /* CONFIG_PROC_FS */ | |
2690 | ||
1796316a JB |
2691 | #ifdef MAX_SWAPFILES_CHECK |
2692 | static int __init max_swapfiles_check(void) | |
2693 | { | |
2694 | MAX_SWAPFILES_CHECK(); | |
2695 | return 0; | |
2696 | } | |
2697 | late_initcall(max_swapfiles_check); | |
2698 | #endif | |
2699 | ||
53cbb243 | 2700 | static struct swap_info_struct *alloc_swap_info(void) |
1da177e4 | 2701 | { |
73c34b6a | 2702 | struct swap_info_struct *p; |
b11a76b3 | 2703 | struct swap_info_struct *defer = NULL; |
1da177e4 | 2704 | unsigned int type; |
a2468cc9 | 2705 | int i; |
efa90a98 | 2706 | |
96008744 | 2707 | p = kvzalloc(struct_size(p, avail_lists, nr_node_ids), GFP_KERNEL); |
efa90a98 | 2708 | if (!p) |
53cbb243 | 2709 | return ERR_PTR(-ENOMEM); |
efa90a98 | 2710 | |
63d8620e ML |
2711 | if (percpu_ref_init(&p->users, swap_users_ref_free, |
2712 | PERCPU_REF_INIT_DEAD, GFP_KERNEL)) { | |
2713 | kvfree(p); | |
2714 | return ERR_PTR(-ENOMEM); | |
2715 | } | |
2716 | ||
5d337b91 | 2717 | spin_lock(&swap_lock); |
efa90a98 HD |
2718 | for (type = 0; type < nr_swapfiles; type++) { |
2719 | if (!(swap_info[type]->flags & SWP_USED)) | |
1da177e4 | 2720 | break; |
efa90a98 | 2721 | } |
0697212a | 2722 | if (type >= MAX_SWAPFILES) { |
5d337b91 | 2723 | spin_unlock(&swap_lock); |
63d8620e | 2724 | percpu_ref_exit(&p->users); |
873d7bcf | 2725 | kvfree(p); |
730c0581 | 2726 | return ERR_PTR(-EPERM); |
1da177e4 | 2727 | } |
efa90a98 HD |
2728 | if (type >= nr_swapfiles) { |
2729 | p->type = type; | |
efa90a98 | 2730 | /* |
a4b45114 HY |
2731 | * Publish the swap_info_struct after initializing it. |
2732 | * Note that kvzalloc() above zeroes all its fields. | |
efa90a98 | 2733 | */ |
a4b45114 HY |
2734 | smp_store_release(&swap_info[type], p); /* rcu_assign_pointer() */ |
2735 | nr_swapfiles++; | |
efa90a98 | 2736 | } else { |
b11a76b3 | 2737 | defer = p; |
efa90a98 HD |
2738 | p = swap_info[type]; |
2739 | /* | |
2740 | * Do not memset this entry: a racing procfs swap_next() | |
2741 | * would be relying on p->type to remain valid. | |
2742 | */ | |
2743 | } | |
4efaceb1 | 2744 | p->swap_extent_root = RB_ROOT; |
18ab4d4c | 2745 | plist_node_init(&p->list, 0); |
a2468cc9 AL |
2746 | for_each_node(i) |
2747 | plist_node_init(&p->avail_lists[i], 0); | |
1da177e4 | 2748 | p->flags = SWP_USED; |
5d337b91 | 2749 | spin_unlock(&swap_lock); |
63d8620e ML |
2750 | if (defer) { |
2751 | percpu_ref_exit(&defer->users); | |
2752 | kvfree(defer); | |
2753 | } | |
ec8acf20 | 2754 | spin_lock_init(&p->lock); |
2628bd6f | 2755 | spin_lock_init(&p->cont_lock); |
63d8620e | 2756 | init_completion(&p->comp); |
efa90a98 | 2757 | |
53cbb243 | 2758 | return p; |
53cbb243 CEB |
2759 | } |
2760 | ||
4d0e1e10 CEB |
2761 | static int claim_swapfile(struct swap_info_struct *p, struct inode *inode) |
2762 | { | |
2763 | int error; | |
2764 | ||
2765 | if (S_ISBLK(inode->i_mode)) { | |
4c6bca43 | 2766 | p->bdev_handle = bdev_open_by_dev(inode->i_rdev, |
05bdb996 | 2767 | BLK_OPEN_READ | BLK_OPEN_WRITE, p, NULL); |
4c6bca43 JK |
2768 | if (IS_ERR(p->bdev_handle)) { |
2769 | error = PTR_ERR(p->bdev_handle); | |
2770 | p->bdev_handle = NULL; | |
6f179af8 | 2771 | return error; |
4d0e1e10 | 2772 | } |
4c6bca43 | 2773 | p->bdev = p->bdev_handle->bdev; |
4d0e1e10 CEB |
2774 | p->old_block_size = block_size(p->bdev); |
2775 | error = set_blocksize(p->bdev, PAGE_SIZE); | |
2776 | if (error < 0) | |
87ade72a | 2777 | return error; |
12d2966d NA |
2778 | /* |
2779 | * Zoned block devices contain zones that have a sequential | |
2780 | * write only restriction. Hence zoned block devices are not | |
2781 | * suitable for swapping. Disallow them here. | |
2782 | */ | |
9964e674 | 2783 | if (bdev_is_zoned(p->bdev)) |
12d2966d | 2784 | return -EINVAL; |
4d0e1e10 CEB |
2785 | p->flags |= SWP_BLKDEV; |
2786 | } else if (S_ISREG(inode->i_mode)) { | |
2787 | p->bdev = inode->i_sb->s_bdev; | |
1638045c DW |
2788 | } |
2789 | ||
4d0e1e10 | 2790 | return 0; |
4d0e1e10 CEB |
2791 | } |
2792 | ||
377eeaa8 AK |
2793 | |
2794 | /* | |
2795 | * Find out how many pages are allowed for a single swap device. There | |
2796 | * are two limiting factors: | |
2797 | * 1) the number of bits for the swap offset in the swp_entry_t type, and | |
2798 | * 2) the number of bits in the swap pte, as defined by the different | |
2799 | * architectures. | |
2800 | * | |
2801 | * In order to find the largest possible bit mask, a swap entry with | |
2802 | * swap type 0 and swap offset ~0UL is created, encoded to a swap pte, | |
2803 | * decoded to a swp_entry_t again, and finally the swap offset is | |
2804 | * extracted. | |
2805 | * | |
2806 | * This will mask all the bits from the initial ~0UL mask that can't | |
2807 | * be encoded in either the swp_entry_t or the architecture definition | |
2808 | * of a swap pte. | |
2809 | */ | |
2810 | unsigned long generic_max_swapfile_size(void) | |
2811 | { | |
2812 | return swp_offset(pte_to_swp_entry( | |
2813 | swp_entry_to_pte(swp_entry(0, ~0UL)))) + 1; | |
2814 | } | |
2815 | ||
2816 | /* Can be overridden by an architecture for additional checks. */ | |
be45a490 | 2817 | __weak unsigned long arch_max_swapfile_size(void) |
377eeaa8 AK |
2818 | { |
2819 | return generic_max_swapfile_size(); | |
2820 | } | |
2821 | ||
ca8bd38b CEB |
2822 | static unsigned long read_swap_header(struct swap_info_struct *p, |
2823 | union swap_header *swap_header, | |
2824 | struct inode *inode) | |
2825 | { | |
2826 | int i; | |
2827 | unsigned long maxpages; | |
2828 | unsigned long swapfilepages; | |
d6bbbd29 | 2829 | unsigned long last_page; |
ca8bd38b CEB |
2830 | |
2831 | if (memcmp("SWAPSPACE2", swap_header->magic.magic, 10)) { | |
465c47fd | 2832 | pr_err("Unable to find swap-space signature\n"); |
38719025 | 2833 | return 0; |
ca8bd38b CEB |
2834 | } |
2835 | ||
041711ce | 2836 | /* swap partition endianness hack... */ |
ca8bd38b CEB |
2837 | if (swab32(swap_header->info.version) == 1) { |
2838 | swab32s(&swap_header->info.version); | |
2839 | swab32s(&swap_header->info.last_page); | |
2840 | swab32s(&swap_header->info.nr_badpages); | |
dd111be6 JH |
2841 | if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES) |
2842 | return 0; | |
ca8bd38b CEB |
2843 | for (i = 0; i < swap_header->info.nr_badpages; i++) |
2844 | swab32s(&swap_header->info.badpages[i]); | |
2845 | } | |
2846 | /* Check the swap header's sub-version */ | |
2847 | if (swap_header->info.version != 1) { | |
465c47fd AM |
2848 | pr_warn("Unable to handle swap header version %d\n", |
2849 | swap_header->info.version); | |
38719025 | 2850 | return 0; |
ca8bd38b CEB |
2851 | } |
2852 | ||
2853 | p->lowest_bit = 1; | |
2854 | p->cluster_next = 1; | |
2855 | p->cluster_nr = 0; | |
2856 | ||
be45a490 | 2857 | maxpages = swapfile_maximum_size; |
d6bbbd29 | 2858 | last_page = swap_header->info.last_page; |
a06ad633 TA |
2859 | if (!last_page) { |
2860 | pr_warn("Empty swap-file\n"); | |
2861 | return 0; | |
2862 | } | |
d6bbbd29 | 2863 | if (last_page > maxpages) { |
465c47fd | 2864 | pr_warn("Truncating oversized swap area, only using %luk out of %luk\n", |
00cde042 | 2865 | K(maxpages), K(last_page)); |
d6bbbd29 RJ |
2866 | } |
2867 | if (maxpages > last_page) { | |
2868 | maxpages = last_page + 1; | |
ca8bd38b CEB |
2869 | /* p->max is an unsigned int: don't overflow it */ |
2870 | if ((unsigned int)maxpages == 0) | |
2871 | maxpages = UINT_MAX; | |
2872 | } | |
2873 | p->highest_bit = maxpages - 1; | |
2874 | ||
2875 | if (!maxpages) | |
38719025 | 2876 | return 0; |
ca8bd38b CEB |
2877 | swapfilepages = i_size_read(inode) >> PAGE_SHIFT; |
2878 | if (swapfilepages && maxpages > swapfilepages) { | |
465c47fd | 2879 | pr_warn("Swap area shorter than signature indicates\n"); |
38719025 | 2880 | return 0; |
ca8bd38b CEB |
2881 | } |
2882 | if (swap_header->info.nr_badpages && S_ISREG(inode->i_mode)) | |
38719025 | 2883 | return 0; |
ca8bd38b | 2884 | if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES) |
38719025 | 2885 | return 0; |
ca8bd38b CEB |
2886 | |
2887 | return maxpages; | |
ca8bd38b CEB |
2888 | } |
2889 | ||
4b3ef9da | 2890 | #define SWAP_CLUSTER_INFO_COLS \ |
235b6217 | 2891 | DIV_ROUND_UP(L1_CACHE_BYTES, sizeof(struct swap_cluster_info)) |
4b3ef9da HY |
2892 | #define SWAP_CLUSTER_SPACE_COLS \ |
2893 | DIV_ROUND_UP(SWAP_ADDRESS_SPACE_PAGES, SWAPFILE_CLUSTER) | |
2894 | #define SWAP_CLUSTER_COLS \ | |
2895 | max_t(unsigned int, SWAP_CLUSTER_INFO_COLS, SWAP_CLUSTER_SPACE_COLS) | |
235b6217 | 2896 | |
915d4d7b CEB |
2897 | static int setup_swap_map_and_extents(struct swap_info_struct *p, |
2898 | union swap_header *swap_header, | |
2899 | unsigned char *swap_map, | |
2a8f9449 | 2900 | struct swap_cluster_info *cluster_info, |
915d4d7b CEB |
2901 | unsigned long maxpages, |
2902 | sector_t *span) | |
2903 | { | |
235b6217 | 2904 | unsigned int j, k; |
915d4d7b CEB |
2905 | unsigned int nr_good_pages; |
2906 | int nr_extents; | |
2a8f9449 | 2907 | unsigned long nr_clusters = DIV_ROUND_UP(maxpages, SWAPFILE_CLUSTER); |
235b6217 HY |
2908 | unsigned long col = p->cluster_next / SWAPFILE_CLUSTER % SWAP_CLUSTER_COLS; |
2909 | unsigned long i, idx; | |
915d4d7b CEB |
2910 | |
2911 | nr_good_pages = maxpages - 1; /* omit header page */ | |
2912 | ||
6b534915 HY |
2913 | cluster_list_init(&p->free_clusters); |
2914 | cluster_list_init(&p->discard_clusters); | |
2a8f9449 | 2915 | |
915d4d7b CEB |
2916 | for (i = 0; i < swap_header->info.nr_badpages; i++) { |
2917 | unsigned int page_nr = swap_header->info.badpages[i]; | |
bdb8e3f6 CEB |
2918 | if (page_nr == 0 || page_nr > swap_header->info.last_page) |
2919 | return -EINVAL; | |
915d4d7b CEB |
2920 | if (page_nr < maxpages) { |
2921 | swap_map[page_nr] = SWAP_MAP_BAD; | |
2922 | nr_good_pages--; | |
2a8f9449 SL |
2923 | /* |
2924 | * Haven't marked the cluster free yet, no list | |
2925 | * operation involved | |
2926 | */ | |
2927 | inc_cluster_info_page(p, cluster_info, page_nr); | |
915d4d7b CEB |
2928 | } |
2929 | } | |
2930 | ||
2a8f9449 SL |
2931 | /* Haven't marked the cluster free yet, no list operation involved */ |
2932 | for (i = maxpages; i < round_up(maxpages, SWAPFILE_CLUSTER); i++) | |
2933 | inc_cluster_info_page(p, cluster_info, i); | |
2934 | ||
915d4d7b CEB |
2935 | if (nr_good_pages) { |
2936 | swap_map[0] = SWAP_MAP_BAD; | |
2a8f9449 SL |
2937 | /* |
2938 | * Not mark the cluster free yet, no list | |
2939 | * operation involved | |
2940 | */ | |
2941 | inc_cluster_info_page(p, cluster_info, 0); | |
915d4d7b CEB |
2942 | p->max = maxpages; |
2943 | p->pages = nr_good_pages; | |
2944 | nr_extents = setup_swap_extents(p, span); | |
bdb8e3f6 CEB |
2945 | if (nr_extents < 0) |
2946 | return nr_extents; | |
915d4d7b CEB |
2947 | nr_good_pages = p->pages; |
2948 | } | |
2949 | if (!nr_good_pages) { | |
465c47fd | 2950 | pr_warn("Empty swap-file\n"); |
bdb8e3f6 | 2951 | return -EINVAL; |
915d4d7b CEB |
2952 | } |
2953 | ||
2a8f9449 SL |
2954 | if (!cluster_info) |
2955 | return nr_extents; | |
2956 | ||
235b6217 | 2957 | |
4b3ef9da HY |
2958 | /* |
2959 | * Reduce false cache line sharing between cluster_info and | |
2960 | * sharing same address space. | |
2961 | */ | |
235b6217 HY |
2962 | for (k = 0; k < SWAP_CLUSTER_COLS; k++) { |
2963 | j = (k + col) % SWAP_CLUSTER_COLS; | |
2964 | for (i = 0; i < DIV_ROUND_UP(nr_clusters, SWAP_CLUSTER_COLS); i++) { | |
2965 | idx = i * SWAP_CLUSTER_COLS + j; | |
2966 | if (idx >= nr_clusters) | |
2967 | continue; | |
2968 | if (cluster_count(&cluster_info[idx])) | |
2969 | continue; | |
2a8f9449 | 2970 | cluster_set_flag(&cluster_info[idx], CLUSTER_FLAG_FREE); |
6b534915 HY |
2971 | cluster_list_add_tail(&p->free_clusters, cluster_info, |
2972 | idx); | |
2a8f9449 | 2973 | } |
2a8f9449 | 2974 | } |
915d4d7b | 2975 | return nr_extents; |
915d4d7b CEB |
2976 | } |
2977 | ||
53cbb243 CEB |
2978 | SYSCALL_DEFINE2(swapon, const char __user *, specialfile, int, swap_flags) |
2979 | { | |
2980 | struct swap_info_struct *p; | |
91a27b2a | 2981 | struct filename *name; |
53cbb243 CEB |
2982 | struct file *swap_file = NULL; |
2983 | struct address_space *mapping; | |
51cc3a66 | 2984 | struct dentry *dentry; |
40531542 | 2985 | int prio; |
53cbb243 CEB |
2986 | int error; |
2987 | union swap_header *swap_header; | |
915d4d7b | 2988 | int nr_extents; |
53cbb243 CEB |
2989 | sector_t span; |
2990 | unsigned long maxpages; | |
53cbb243 | 2991 | unsigned char *swap_map = NULL; |
2a8f9449 | 2992 | struct swap_cluster_info *cluster_info = NULL; |
53cbb243 CEB |
2993 | struct page *page = NULL; |
2994 | struct inode *inode = NULL; | |
7cbf3192 | 2995 | bool inced_nr_rotate_swap = false; |
53cbb243 | 2996 | |
d15cab97 HD |
2997 | if (swap_flags & ~SWAP_FLAGS_VALID) |
2998 | return -EINVAL; | |
2999 | ||
53cbb243 CEB |
3000 | if (!capable(CAP_SYS_ADMIN)) |
3001 | return -EPERM; | |
3002 | ||
a2468cc9 AL |
3003 | if (!swap_avail_heads) |
3004 | return -ENOMEM; | |
3005 | ||
53cbb243 | 3006 | p = alloc_swap_info(); |
2542e513 CEB |
3007 | if (IS_ERR(p)) |
3008 | return PTR_ERR(p); | |
53cbb243 | 3009 | |
815c2c54 SL |
3010 | INIT_WORK(&p->discard_work, swap_discard_work); |
3011 | ||
1da177e4 | 3012 | name = getname(specialfile); |
1da177e4 | 3013 | if (IS_ERR(name)) { |
7de7fb6b | 3014 | error = PTR_ERR(name); |
1da177e4 | 3015 | name = NULL; |
bd69010b | 3016 | goto bad_swap; |
1da177e4 | 3017 | } |
669abf4e | 3018 | swap_file = file_open_name(name, O_RDWR|O_LARGEFILE, 0); |
1da177e4 | 3019 | if (IS_ERR(swap_file)) { |
7de7fb6b | 3020 | error = PTR_ERR(swap_file); |
1da177e4 | 3021 | swap_file = NULL; |
bd69010b | 3022 | goto bad_swap; |
1da177e4 LT |
3023 | } |
3024 | ||
3025 | p->swap_file = swap_file; | |
3026 | mapping = swap_file->f_mapping; | |
51cc3a66 | 3027 | dentry = swap_file->f_path.dentry; |
2130781e | 3028 | inode = mapping->host; |
6f179af8 | 3029 | |
4d0e1e10 CEB |
3030 | error = claim_swapfile(p, inode); |
3031 | if (unlikely(error)) | |
1da177e4 | 3032 | goto bad_swap; |
1da177e4 | 3033 | |
d795a90e | 3034 | inode_lock(inode); |
51cc3a66 HD |
3035 | if (d_unlinked(dentry) || cant_mount(dentry)) { |
3036 | error = -ENOENT; | |
3037 | goto bad_swap_unlock_inode; | |
3038 | } | |
d795a90e NA |
3039 | if (IS_SWAPFILE(inode)) { |
3040 | error = -EBUSY; | |
3041 | goto bad_swap_unlock_inode; | |
3042 | } | |
3043 | ||
1da177e4 LT |
3044 | /* |
3045 | * Read the swap header. | |
3046 | */ | |
7e0a1265 | 3047 | if (!mapping->a_ops->read_folio) { |
1da177e4 | 3048 | error = -EINVAL; |
d795a90e | 3049 | goto bad_swap_unlock_inode; |
1da177e4 | 3050 | } |
090d2b18 | 3051 | page = read_mapping_page(mapping, 0, swap_file); |
1da177e4 LT |
3052 | if (IS_ERR(page)) { |
3053 | error = PTR_ERR(page); | |
d795a90e | 3054 | goto bad_swap_unlock_inode; |
1da177e4 | 3055 | } |
81e33971 | 3056 | swap_header = kmap(page); |
1da177e4 | 3057 | |
ca8bd38b CEB |
3058 | maxpages = read_swap_header(p, swap_header, inode); |
3059 | if (unlikely(!maxpages)) { | |
1da177e4 | 3060 | error = -EINVAL; |
d795a90e | 3061 | goto bad_swap_unlock_inode; |
1da177e4 | 3062 | } |
886bb7e9 | 3063 | |
81e33971 | 3064 | /* OK, set up the swap map and apply the bad block list */ |
803d0c83 | 3065 | swap_map = vzalloc(maxpages); |
81e33971 HD |
3066 | if (!swap_map) { |
3067 | error = -ENOMEM; | |
d795a90e | 3068 | goto bad_swap_unlock_inode; |
81e33971 | 3069 | } |
f0571429 | 3070 | |
36d25489 | 3071 | if (p->bdev && bdev_stable_writes(p->bdev)) |
f0571429 MK |
3072 | p->flags |= SWP_STABLE_WRITES; |
3073 | ||
3222d8c2 | 3074 | if (p->bdev && bdev_synchronous(p->bdev)) |
539a6fea MK |
3075 | p->flags |= SWP_SYNCHRONOUS_IO; |
3076 | ||
10f0d2a5 | 3077 | if (p->bdev && bdev_nonrot(p->bdev)) { |
6f179af8 | 3078 | int cpu; |
235b6217 | 3079 | unsigned long ci, nr_cluster; |
6f179af8 | 3080 | |
2a8f9449 | 3081 | p->flags |= SWP_SOLIDSTATE; |
49070588 HY |
3082 | p->cluster_next_cpu = alloc_percpu(unsigned int); |
3083 | if (!p->cluster_next_cpu) { | |
3084 | error = -ENOMEM; | |
3085 | goto bad_swap_unlock_inode; | |
3086 | } | |
2a8f9449 SL |
3087 | /* |
3088 | * select a random position to start with to help wear leveling | |
3089 | * SSD | |
3090 | */ | |
49070588 HY |
3091 | for_each_possible_cpu(cpu) { |
3092 | per_cpu(*p->cluster_next_cpu, cpu) = | |
e8a533cb | 3093 | get_random_u32_inclusive(1, p->highest_bit); |
49070588 | 3094 | } |
235b6217 | 3095 | nr_cluster = DIV_ROUND_UP(maxpages, SWAPFILE_CLUSTER); |
2a8f9449 | 3096 | |
778e1cdd | 3097 | cluster_info = kvcalloc(nr_cluster, sizeof(*cluster_info), |
54f180d3 | 3098 | GFP_KERNEL); |
2a8f9449 SL |
3099 | if (!cluster_info) { |
3100 | error = -ENOMEM; | |
d795a90e | 3101 | goto bad_swap_unlock_inode; |
2a8f9449 | 3102 | } |
235b6217 HY |
3103 | |
3104 | for (ci = 0; ci < nr_cluster; ci++) | |
3105 | spin_lock_init(&((cluster_info + ci)->lock)); | |
3106 | ||
ebc2a1a6 SL |
3107 | p->percpu_cluster = alloc_percpu(struct percpu_cluster); |
3108 | if (!p->percpu_cluster) { | |
3109 | error = -ENOMEM; | |
d795a90e | 3110 | goto bad_swap_unlock_inode; |
ebc2a1a6 | 3111 | } |
6f179af8 | 3112 | for_each_possible_cpu(cpu) { |
ebc2a1a6 | 3113 | struct percpu_cluster *cluster; |
6f179af8 | 3114 | cluster = per_cpu_ptr(p->percpu_cluster, cpu); |
ebc2a1a6 SL |
3115 | cluster_set_null(&cluster->index); |
3116 | } | |
7cbf3192 | 3117 | } else { |
81a0298b | 3118 | atomic_inc(&nr_rotate_swap); |
7cbf3192 OS |
3119 | inced_nr_rotate_swap = true; |
3120 | } | |
1da177e4 | 3121 | |
1421ef3c CEB |
3122 | error = swap_cgroup_swapon(p->type, maxpages); |
3123 | if (error) | |
d795a90e | 3124 | goto bad_swap_unlock_inode; |
1421ef3c | 3125 | |
915d4d7b | 3126 | nr_extents = setup_swap_map_and_extents(p, swap_header, swap_map, |
2a8f9449 | 3127 | cluster_info, maxpages, &span); |
915d4d7b CEB |
3128 | if (unlikely(nr_extents < 0)) { |
3129 | error = nr_extents; | |
d795a90e | 3130 | goto bad_swap_unlock_inode; |
1da177e4 | 3131 | } |
1da177e4 | 3132 | |
70200574 CH |
3133 | if ((swap_flags & SWAP_FLAG_DISCARD) && |
3134 | p->bdev && bdev_max_discard_sectors(p->bdev)) { | |
2a8f9449 SL |
3135 | /* |
3136 | * When discard is enabled for swap with no particular | |
3137 | * policy flagged, we set all swap discard flags here in | |
3138 | * order to sustain backward compatibility with older | |
3139 | * swapon(8) releases. | |
3140 | */ | |
3141 | p->flags |= (SWP_DISCARDABLE | SWP_AREA_DISCARD | | |
3142 | SWP_PAGE_DISCARD); | |
dcf6b7dd | 3143 | |
2a8f9449 SL |
3144 | /* |
3145 | * By flagging sys_swapon, a sysadmin can tell us to | |
3146 | * either do single-time area discards only, or to just | |
3147 | * perform discards for released swap page-clusters. | |
3148 | * Now it's time to adjust the p->flags accordingly. | |
3149 | */ | |
3150 | if (swap_flags & SWAP_FLAG_DISCARD_ONCE) | |
3151 | p->flags &= ~SWP_PAGE_DISCARD; | |
3152 | else if (swap_flags & SWAP_FLAG_DISCARD_PAGES) | |
3153 | p->flags &= ~SWP_AREA_DISCARD; | |
3154 | ||
3155 | /* issue a swapon-time discard if it's still required */ | |
3156 | if (p->flags & SWP_AREA_DISCARD) { | |
3157 | int err = discard_swap(p); | |
3158 | if (unlikely(err)) | |
3159 | pr_err("swapon: discard_swap(%p): %d\n", | |
3160 | p, err); | |
dcf6b7dd | 3161 | } |
20137a49 | 3162 | } |
6a6ba831 | 3163 | |
4b3ef9da HY |
3164 | error = init_swap_address_space(p->type, maxpages); |
3165 | if (error) | |
d795a90e | 3166 | goto bad_swap_unlock_inode; |
4b3ef9da | 3167 | |
dc617f29 DW |
3168 | /* |
3169 | * Flush any pending IO and dirty mappings before we start using this | |
3170 | * swap device. | |
3171 | */ | |
3172 | inode->i_flags |= S_SWAPFILE; | |
3173 | error = inode_drain_writes(inode); | |
3174 | if (error) { | |
3175 | inode->i_flags &= ~S_SWAPFILE; | |
822bca52 | 3176 | goto free_swap_address_space; |
dc617f29 DW |
3177 | } |
3178 | ||
fc0abb14 | 3179 | mutex_lock(&swapon_mutex); |
40531542 | 3180 | prio = -1; |
78ecba08 | 3181 | if (swap_flags & SWAP_FLAG_PREFER) |
40531542 | 3182 | prio = |
78ecba08 | 3183 | (swap_flags & SWAP_FLAG_PRIO_MASK) >> SWAP_FLAG_PRIO_SHIFT; |
42c06a0e | 3184 | enable_swap_info(p, prio, swap_map, cluster_info); |
c69dbfb8 | 3185 | |
42c06a0e | 3186 | pr_info("Adding %uk swap on %s. Priority:%d extents:%d across:%lluk %s%s%s%s\n", |
00cde042 Z |
3187 | K(p->pages), name->name, p->prio, nr_extents, |
3188 | K((unsigned long long)span), | |
c69dbfb8 | 3189 | (p->flags & SWP_SOLIDSTATE) ? "SS" : "", |
38b5faf4 | 3190 | (p->flags & SWP_DISCARDABLE) ? "D" : "", |
dcf6b7dd | 3191 | (p->flags & SWP_AREA_DISCARD) ? "s" : "", |
42c06a0e | 3192 | (p->flags & SWP_PAGE_DISCARD) ? "c" : ""); |
c69dbfb8 | 3193 | |
fc0abb14 | 3194 | mutex_unlock(&swapon_mutex); |
66d7dd51 KS |
3195 | atomic_inc(&proc_poll_event); |
3196 | wake_up_interruptible(&proc_poll_wait); | |
3197 | ||
1da177e4 LT |
3198 | error = 0; |
3199 | goto out; | |
822bca52 ML |
3200 | free_swap_address_space: |
3201 | exit_swap_address_space(p->type); | |
d795a90e NA |
3202 | bad_swap_unlock_inode: |
3203 | inode_unlock(inode); | |
1da177e4 | 3204 | bad_swap: |
ebc2a1a6 SL |
3205 | free_percpu(p->percpu_cluster); |
3206 | p->percpu_cluster = NULL; | |
49070588 HY |
3207 | free_percpu(p->cluster_next_cpu); |
3208 | p->cluster_next_cpu = NULL; | |
4c6bca43 | 3209 | if (p->bdev_handle) { |
f2090d2d | 3210 | set_blocksize(p->bdev, p->old_block_size); |
4c6bca43 JK |
3211 | bdev_release(p->bdev_handle); |
3212 | p->bdev_handle = NULL; | |
1da177e4 | 3213 | } |
d795a90e | 3214 | inode = NULL; |
4cd3bb10 | 3215 | destroy_swap_extents(p); |
e8e6c2ec | 3216 | swap_cgroup_swapoff(p->type); |
5d337b91 | 3217 | spin_lock(&swap_lock); |
1da177e4 | 3218 | p->swap_file = NULL; |
1da177e4 | 3219 | p->flags = 0; |
5d337b91 | 3220 | spin_unlock(&swap_lock); |
1da177e4 | 3221 | vfree(swap_map); |
8606a1a9 | 3222 | kvfree(cluster_info); |
7cbf3192 OS |
3223 | if (inced_nr_rotate_swap) |
3224 | atomic_dec(&nr_rotate_swap); | |
d795a90e | 3225 | if (swap_file) |
1da177e4 LT |
3226 | filp_close(swap_file, NULL); |
3227 | out: | |
3228 | if (page && !IS_ERR(page)) { | |
3229 | kunmap(page); | |
09cbfeaf | 3230 | put_page(page); |
1da177e4 LT |
3231 | } |
3232 | if (name) | |
3233 | putname(name); | |
1638045c | 3234 | if (inode) |
5955102c | 3235 | inode_unlock(inode); |
039939a6 TC |
3236 | if (!error) |
3237 | enable_swap_slots_cache(); | |
1da177e4 LT |
3238 | return error; |
3239 | } | |
3240 | ||
3241 | void si_swapinfo(struct sysinfo *val) | |
3242 | { | |
efa90a98 | 3243 | unsigned int type; |
1da177e4 LT |
3244 | unsigned long nr_to_be_unused = 0; |
3245 | ||
5d337b91 | 3246 | spin_lock(&swap_lock); |
efa90a98 HD |
3247 | for (type = 0; type < nr_swapfiles; type++) { |
3248 | struct swap_info_struct *si = swap_info[type]; | |
3249 | ||
3250 | if ((si->flags & SWP_USED) && !(si->flags & SWP_WRITEOK)) | |
c8945306 | 3251 | nr_to_be_unused += READ_ONCE(si->inuse_pages); |
1da177e4 | 3252 | } |
ec8acf20 | 3253 | val->freeswap = atomic_long_read(&nr_swap_pages) + nr_to_be_unused; |
1da177e4 | 3254 | val->totalswap = total_swap_pages + nr_to_be_unused; |
5d337b91 | 3255 | spin_unlock(&swap_lock); |
1da177e4 LT |
3256 | } |
3257 | ||
3258 | /* | |
3259 | * Verify that a swap entry is valid and increment its swap map count. | |
3260 | * | |
355cfa73 KH |
3261 | * Returns error code in following case. |
3262 | * - success -> 0 | |
3263 | * - swp_entry is invalid -> EINVAL | |
3264 | * - swp_entry is migration entry -> EINVAL | |
3265 | * - swap-cache reference is requested but there is already one. -> EEXIST | |
3266 | * - swap-cache reference is requested but the entry is not used. -> ENOENT | |
570a335b | 3267 | * - swap-mapped reference requested but needs continued swap count. -> ENOMEM |
1da177e4 | 3268 | */ |
8d69aaee | 3269 | static int __swap_duplicate(swp_entry_t entry, unsigned char usage) |
1da177e4 | 3270 | { |
73c34b6a | 3271 | struct swap_info_struct *p; |
235b6217 | 3272 | struct swap_cluster_info *ci; |
c10d38cc | 3273 | unsigned long offset; |
8d69aaee HD |
3274 | unsigned char count; |
3275 | unsigned char has_cache; | |
9d9a0334 | 3276 | int err; |
1da177e4 | 3277 | |
c07aee4f | 3278 | p = swp_swap_info(entry); |
235b6217 | 3279 | |
eb085574 | 3280 | offset = swp_offset(entry); |
235b6217 | 3281 | ci = lock_cluster_or_swap_info(p, offset); |
355cfa73 | 3282 | |
253d553b | 3283 | count = p->swap_map[offset]; |
edfe23da SL |
3284 | |
3285 | /* | |
3286 | * swapin_readahead() doesn't check if a swap entry is valid, so the | |
3287 | * swap entry could be SWAP_MAP_BAD. Check here with lock held. | |
3288 | */ | |
3289 | if (unlikely(swap_count(count) == SWAP_MAP_BAD)) { | |
3290 | err = -ENOENT; | |
3291 | goto unlock_out; | |
3292 | } | |
3293 | ||
253d553b HD |
3294 | has_cache = count & SWAP_HAS_CACHE; |
3295 | count &= ~SWAP_HAS_CACHE; | |
3296 | err = 0; | |
355cfa73 | 3297 | |
253d553b | 3298 | if (usage == SWAP_HAS_CACHE) { |
355cfa73 KH |
3299 | |
3300 | /* set SWAP_HAS_CACHE if there is no cache and entry is used */ | |
253d553b HD |
3301 | if (!has_cache && count) |
3302 | has_cache = SWAP_HAS_CACHE; | |
3303 | else if (has_cache) /* someone else added cache */ | |
3304 | err = -EEXIST; | |
3305 | else /* no users remaining */ | |
3306 | err = -ENOENT; | |
355cfa73 KH |
3307 | |
3308 | } else if (count || has_cache) { | |
253d553b | 3309 | |
570a335b HD |
3310 | if ((count & ~COUNT_CONTINUED) < SWAP_MAP_MAX) |
3311 | count += usage; | |
3312 | else if ((count & ~COUNT_CONTINUED) > SWAP_MAP_MAX) | |
253d553b | 3313 | err = -EINVAL; |
570a335b HD |
3314 | else if (swap_count_continued(p, offset, count)) |
3315 | count = COUNT_CONTINUED; | |
3316 | else | |
3317 | err = -ENOMEM; | |
355cfa73 | 3318 | } else |
253d553b HD |
3319 | err = -ENOENT; /* unused swap entry */ |
3320 | ||
a449bf58 | 3321 | WRITE_ONCE(p->swap_map[offset], count | has_cache); |
253d553b | 3322 | |
355cfa73 | 3323 | unlock_out: |
235b6217 | 3324 | unlock_cluster_or_swap_info(p, ci); |
253d553b | 3325 | return err; |
1da177e4 | 3326 | } |
253d553b | 3327 | |
aaa46865 HD |
3328 | /* |
3329 | * Help swapoff by noting that swap entry belongs to shmem/tmpfs | |
3330 | * (in which case its reference count is never incremented). | |
3331 | */ | |
3332 | void swap_shmem_alloc(swp_entry_t entry) | |
3333 | { | |
3334 | __swap_duplicate(entry, SWAP_MAP_SHMEM); | |
3335 | } | |
3336 | ||
355cfa73 | 3337 | /* |
08259d58 HD |
3338 | * Increase reference count of swap entry by 1. |
3339 | * Returns 0 for success, or -ENOMEM if a swap_count_continuation is required | |
3340 | * but could not be atomically allocated. Returns 0, just as if it succeeded, | |
3341 | * if __swap_duplicate() fails for another reason (-EINVAL or -ENOENT), which | |
3342 | * might occur if a page table entry has got corrupted. | |
355cfa73 | 3343 | */ |
570a335b | 3344 | int swap_duplicate(swp_entry_t entry) |
355cfa73 | 3345 | { |
570a335b HD |
3346 | int err = 0; |
3347 | ||
3348 | while (!err && __swap_duplicate(entry, 1) == -ENOMEM) | |
3349 | err = add_swap_count_continuation(entry, GFP_ATOMIC); | |
3350 | return err; | |
355cfa73 | 3351 | } |
1da177e4 | 3352 | |
cb4b86ba | 3353 | /* |
355cfa73 KH |
3354 | * @entry: swap entry for which we allocate swap cache. |
3355 | * | |
73c34b6a | 3356 | * Called when allocating swap cache for existing swap entry, |
355cfa73 | 3357 | * This can return error codes. Returns 0 at success. |
3eeba135 | 3358 | * -EEXIST means there is a swap cache. |
355cfa73 | 3359 | * Note: return code is different from swap_duplicate(). |
cb4b86ba KH |
3360 | */ |
3361 | int swapcache_prepare(swp_entry_t entry) | |
3362 | { | |
253d553b | 3363 | return __swap_duplicate(entry, SWAP_HAS_CACHE); |
cb4b86ba KH |
3364 | } |
3365 | ||
0bcac06f MK |
3366 | struct swap_info_struct *swp_swap_info(swp_entry_t entry) |
3367 | { | |
c10d38cc | 3368 | return swap_type_to_swap_info(swp_type(entry)); |
0bcac06f MK |
3369 | } |
3370 | ||
f981c595 MG |
3371 | struct swap_info_struct *page_swap_info(struct page *page) |
3372 | { | |
cfeed8ff | 3373 | swp_entry_t entry = page_swap_entry(page); |
0bcac06f | 3374 | return swp_swap_info(entry); |
f981c595 MG |
3375 | } |
3376 | ||
3377 | /* | |
2f52578f | 3378 | * out-of-line methods to avoid include hell. |
f981c595 | 3379 | */ |
2f52578f | 3380 | struct address_space *swapcache_mapping(struct folio *folio) |
f981c595 | 3381 | { |
2f52578f | 3382 | return page_swap_info(&folio->page)->swap_file->f_mapping; |
f981c595 | 3383 | } |
2f52578f | 3384 | EXPORT_SYMBOL_GPL(swapcache_mapping); |
f981c595 MG |
3385 | |
3386 | pgoff_t __page_file_index(struct page *page) | |
3387 | { | |
cfeed8ff | 3388 | swp_entry_t swap = page_swap_entry(page); |
f981c595 MG |
3389 | return swp_offset(swap); |
3390 | } | |
3391 | EXPORT_SYMBOL_GPL(__page_file_index); | |
3392 | ||
570a335b HD |
3393 | /* |
3394 | * add_swap_count_continuation - called when a swap count is duplicated | |
3395 | * beyond SWAP_MAP_MAX, it allocates a new page and links that to the entry's | |
3396 | * page of the original vmalloc'ed swap_map, to hold the continuation count | |
3397 | * (for that entry and for its neighbouring PAGE_SIZE swap entries). Called | |
3398 | * again when count is duplicated beyond SWAP_MAP_MAX * SWAP_CONT_MAX, etc. | |
3399 | * | |
3400 | * These continuation pages are seldom referenced: the common paths all work | |
3401 | * on the original swap_map, only referring to a continuation page when the | |
3402 | * low "digit" of a count is incremented or decremented through SWAP_MAP_MAX. | |
3403 | * | |
3404 | * add_swap_count_continuation(, GFP_ATOMIC) can be called while holding | |
3405 | * page table locks; if it fails, add_swap_count_continuation(, GFP_KERNEL) | |
3406 | * can be called after dropping locks. | |
3407 | */ | |
3408 | int add_swap_count_continuation(swp_entry_t entry, gfp_t gfp_mask) | |
3409 | { | |
3410 | struct swap_info_struct *si; | |
235b6217 | 3411 | struct swap_cluster_info *ci; |
570a335b HD |
3412 | struct page *head; |
3413 | struct page *page; | |
3414 | struct page *list_page; | |
3415 | pgoff_t offset; | |
3416 | unsigned char count; | |
eb085574 | 3417 | int ret = 0; |
570a335b HD |
3418 | |
3419 | /* | |
3420 | * When debugging, it's easier to use __GFP_ZERO here; but it's better | |
3421 | * for latency not to zero a page while GFP_ATOMIC and holding locks. | |
3422 | */ | |
3423 | page = alloc_page(gfp_mask | __GFP_HIGHMEM); | |
3424 | ||
eb085574 | 3425 | si = get_swap_device(entry); |
570a335b HD |
3426 | if (!si) { |
3427 | /* | |
3428 | * An acceptable race has occurred since the failing | |
eb085574 | 3429 | * __swap_duplicate(): the swap device may be swapoff |
570a335b HD |
3430 | */ |
3431 | goto outer; | |
3432 | } | |
eb085574 | 3433 | spin_lock(&si->lock); |
570a335b HD |
3434 | |
3435 | offset = swp_offset(entry); | |
235b6217 HY |
3436 | |
3437 | ci = lock_cluster(si, offset); | |
3438 | ||
d8aa24e0 | 3439 | count = swap_count(si->swap_map[offset]); |
570a335b HD |
3440 | |
3441 | if ((count & ~COUNT_CONTINUED) != SWAP_MAP_MAX) { | |
3442 | /* | |
3443 | * The higher the swap count, the more likely it is that tasks | |
3444 | * will race to add swap count continuation: we need to avoid | |
3445 | * over-provisioning. | |
3446 | */ | |
3447 | goto out; | |
3448 | } | |
3449 | ||
3450 | if (!page) { | |
eb085574 HY |
3451 | ret = -ENOMEM; |
3452 | goto out; | |
570a335b HD |
3453 | } |
3454 | ||
570a335b HD |
3455 | head = vmalloc_to_page(si->swap_map + offset); |
3456 | offset &= ~PAGE_MASK; | |
3457 | ||
2628bd6f | 3458 | spin_lock(&si->cont_lock); |
570a335b HD |
3459 | /* |
3460 | * Page allocation does not initialize the page's lru field, | |
3461 | * but it does always reset its private field. | |
3462 | */ | |
3463 | if (!page_private(head)) { | |
3464 | BUG_ON(count & COUNT_CONTINUED); | |
3465 | INIT_LIST_HEAD(&head->lru); | |
3466 | set_page_private(head, SWP_CONTINUED); | |
3467 | si->flags |= SWP_CONTINUED; | |
3468 | } | |
3469 | ||
3470 | list_for_each_entry(list_page, &head->lru, lru) { | |
3471 | unsigned char *map; | |
3472 | ||
3473 | /* | |
3474 | * If the previous map said no continuation, but we've found | |
3475 | * a continuation page, free our allocation and use this one. | |
3476 | */ | |
3477 | if (!(count & COUNT_CONTINUED)) | |
2628bd6f | 3478 | goto out_unlock_cont; |
570a335b | 3479 | |
9b04c5fe | 3480 | map = kmap_atomic(list_page) + offset; |
570a335b | 3481 | count = *map; |
9b04c5fe | 3482 | kunmap_atomic(map); |
570a335b HD |
3483 | |
3484 | /* | |
3485 | * If this continuation count now has some space in it, | |
3486 | * free our allocation and use this one. | |
3487 | */ | |
3488 | if ((count & ~COUNT_CONTINUED) != SWAP_CONT_MAX) | |
2628bd6f | 3489 | goto out_unlock_cont; |
570a335b HD |
3490 | } |
3491 | ||
3492 | list_add_tail(&page->lru, &head->lru); | |
3493 | page = NULL; /* now it's attached, don't free it */ | |
2628bd6f HY |
3494 | out_unlock_cont: |
3495 | spin_unlock(&si->cont_lock); | |
570a335b | 3496 | out: |
235b6217 | 3497 | unlock_cluster(ci); |
ec8acf20 | 3498 | spin_unlock(&si->lock); |
eb085574 | 3499 | put_swap_device(si); |
570a335b HD |
3500 | outer: |
3501 | if (page) | |
3502 | __free_page(page); | |
eb085574 | 3503 | return ret; |
570a335b HD |
3504 | } |
3505 | ||
3506 | /* | |
3507 | * swap_count_continued - when the original swap_map count is incremented | |
3508 | * from SWAP_MAP_MAX, check if there is already a continuation page to carry | |
3509 | * into, carry if so, or else fail until a new continuation page is allocated; | |
3510 | * when the original swap_map count is decremented from 0 with continuation, | |
3511 | * borrow from the continuation and report whether it still holds more. | |
235b6217 HY |
3512 | * Called while __swap_duplicate() or swap_entry_free() holds swap or cluster |
3513 | * lock. | |
570a335b HD |
3514 | */ |
3515 | static bool swap_count_continued(struct swap_info_struct *si, | |
3516 | pgoff_t offset, unsigned char count) | |
3517 | { | |
3518 | struct page *head; | |
3519 | struct page *page; | |
3520 | unsigned char *map; | |
2628bd6f | 3521 | bool ret; |
570a335b HD |
3522 | |
3523 | head = vmalloc_to_page(si->swap_map + offset); | |
3524 | if (page_private(head) != SWP_CONTINUED) { | |
3525 | BUG_ON(count & COUNT_CONTINUED); | |
3526 | return false; /* need to add count continuation */ | |
3527 | } | |
3528 | ||
2628bd6f | 3529 | spin_lock(&si->cont_lock); |
570a335b | 3530 | offset &= ~PAGE_MASK; |
213516ac | 3531 | page = list_next_entry(head, lru); |
9b04c5fe | 3532 | map = kmap_atomic(page) + offset; |
570a335b HD |
3533 | |
3534 | if (count == SWAP_MAP_MAX) /* initial increment from swap_map */ | |
3535 | goto init_map; /* jump over SWAP_CONT_MAX checks */ | |
3536 | ||
3537 | if (count == (SWAP_MAP_MAX | COUNT_CONTINUED)) { /* incrementing */ | |
3538 | /* | |
3539 | * Think of how you add 1 to 999 | |
3540 | */ | |
3541 | while (*map == (SWAP_CONT_MAX | COUNT_CONTINUED)) { | |
9b04c5fe | 3542 | kunmap_atomic(map); |
213516ac | 3543 | page = list_next_entry(page, lru); |
570a335b | 3544 | BUG_ON(page == head); |
9b04c5fe | 3545 | map = kmap_atomic(page) + offset; |
570a335b HD |
3546 | } |
3547 | if (*map == SWAP_CONT_MAX) { | |
9b04c5fe | 3548 | kunmap_atomic(map); |
213516ac | 3549 | page = list_next_entry(page, lru); |
2628bd6f HY |
3550 | if (page == head) { |
3551 | ret = false; /* add count continuation */ | |
3552 | goto out; | |
3553 | } | |
9b04c5fe | 3554 | map = kmap_atomic(page) + offset; |
570a335b HD |
3555 | init_map: *map = 0; /* we didn't zero the page */ |
3556 | } | |
3557 | *map += 1; | |
9b04c5fe | 3558 | kunmap_atomic(map); |
213516ac | 3559 | while ((page = list_prev_entry(page, lru)) != head) { |
9b04c5fe | 3560 | map = kmap_atomic(page) + offset; |
570a335b | 3561 | *map = COUNT_CONTINUED; |
9b04c5fe | 3562 | kunmap_atomic(map); |
570a335b | 3563 | } |
2628bd6f | 3564 | ret = true; /* incremented */ |
570a335b HD |
3565 | |
3566 | } else { /* decrementing */ | |
3567 | /* | |
3568 | * Think of how you subtract 1 from 1000 | |
3569 | */ | |
3570 | BUG_ON(count != COUNT_CONTINUED); | |
3571 | while (*map == COUNT_CONTINUED) { | |
9b04c5fe | 3572 | kunmap_atomic(map); |
213516ac | 3573 | page = list_next_entry(page, lru); |
570a335b | 3574 | BUG_ON(page == head); |
9b04c5fe | 3575 | map = kmap_atomic(page) + offset; |
570a335b HD |
3576 | } |
3577 | BUG_ON(*map == 0); | |
3578 | *map -= 1; | |
3579 | if (*map == 0) | |
3580 | count = 0; | |
9b04c5fe | 3581 | kunmap_atomic(map); |
213516ac | 3582 | while ((page = list_prev_entry(page, lru)) != head) { |
9b04c5fe | 3583 | map = kmap_atomic(page) + offset; |
570a335b HD |
3584 | *map = SWAP_CONT_MAX | count; |
3585 | count = COUNT_CONTINUED; | |
9b04c5fe | 3586 | kunmap_atomic(map); |
570a335b | 3587 | } |
2628bd6f | 3588 | ret = count == COUNT_CONTINUED; |
570a335b | 3589 | } |
2628bd6f HY |
3590 | out: |
3591 | spin_unlock(&si->cont_lock); | |
3592 | return ret; | |
570a335b HD |
3593 | } |
3594 | ||
3595 | /* | |
3596 | * free_swap_count_continuations - swapoff free all the continuation pages | |
3597 | * appended to the swap_map, after swap_map is quiesced, before vfree'ing it. | |
3598 | */ | |
3599 | static void free_swap_count_continuations(struct swap_info_struct *si) | |
3600 | { | |
3601 | pgoff_t offset; | |
3602 | ||
3603 | for (offset = 0; offset < si->max; offset += PAGE_SIZE) { | |
3604 | struct page *head; | |
3605 | head = vmalloc_to_page(si->swap_map + offset); | |
3606 | if (page_private(head)) { | |
0d576d20 GT |
3607 | struct page *page, *next; |
3608 | ||
3609 | list_for_each_entry_safe(page, next, &head->lru, lru) { | |
3610 | list_del(&page->lru); | |
570a335b HD |
3611 | __free_page(page); |
3612 | } | |
3613 | } | |
3614 | } | |
3615 | } | |
a2468cc9 | 3616 | |
2cf85583 | 3617 | #if defined(CONFIG_MEMCG) && defined(CONFIG_BLK_CGROUP) |
3e4fb13a | 3618 | void __folio_throttle_swaprate(struct folio *folio, gfp_t gfp) |
2cf85583 TH |
3619 | { |
3620 | struct swap_info_struct *si, *next; | |
3e4fb13a | 3621 | int nid = folio_nid(folio); |
6caa6a07 | 3622 | |
3e4fb13a | 3623 | if (!(gfp & __GFP_IO)) |
2cf85583 TH |
3624 | return; |
3625 | ||
3626 | if (!blk_cgroup_congested()) | |
3627 | return; | |
3628 | ||
3629 | /* | |
3630 | * We've already scheduled a throttle, avoid taking the global swap | |
3631 | * lock. | |
3632 | */ | |
f05837ed | 3633 | if (current->throttle_disk) |
2cf85583 TH |
3634 | return; |
3635 | ||
3636 | spin_lock(&swap_avail_lock); | |
6caa6a07 JW |
3637 | plist_for_each_entry_safe(si, next, &swap_avail_heads[nid], |
3638 | avail_lists[nid]) { | |
2cf85583 | 3639 | if (si->bdev) { |
de185b56 | 3640 | blkcg_schedule_throttle(si->bdev->bd_disk, true); |
2cf85583 TH |
3641 | break; |
3642 | } | |
3643 | } | |
3644 | spin_unlock(&swap_avail_lock); | |
3645 | } | |
3646 | #endif | |
3647 | ||
a2468cc9 AL |
3648 | static int __init swapfile_init(void) |
3649 | { | |
3650 | int nid; | |
3651 | ||
3652 | swap_avail_heads = kmalloc_array(nr_node_ids, sizeof(struct plist_head), | |
3653 | GFP_KERNEL); | |
3654 | if (!swap_avail_heads) { | |
3655 | pr_emerg("Not enough memory for swap heads, swap is disabled\n"); | |
3656 | return -ENOMEM; | |
3657 | } | |
3658 | ||
3659 | for_each_node(nid) | |
3660 | plist_head_init(&swap_avail_heads[nid]); | |
3661 | ||
be45a490 PX |
3662 | swapfile_maximum_size = arch_max_swapfile_size(); |
3663 | ||
5154e607 PX |
3664 | #ifdef CONFIG_MIGRATION |
3665 | if (swapfile_maximum_size >= (1UL << SWP_MIG_TOTAL_BITS)) | |
3666 | swap_migration_ad_supported = true; | |
3667 | #endif /* CONFIG_MIGRATION */ | |
3668 | ||
a2468cc9 AL |
3669 | return 0; |
3670 | } | |
3671 | subsys_initcall(swapfile_init); |