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> |
8b7787a5 | 45 | #include <linux/plist.h> |
1da177e4 | 46 | |
1da177e4 LT |
47 | #include <asm/tlbflush.h> |
48 | #include <linux/swapops.h> | |
5d1ea48b | 49 | #include <linux/swap_cgroup.h> |
00cde042 | 50 | #include "internal.h" |
014bb1de | 51 | #include "swap.h" |
1da177e4 | 52 | |
570a335b HD |
53 | static bool swap_count_continued(struct swap_info_struct *, pgoff_t, |
54 | unsigned char); | |
55 | static void free_swap_count_continuations(struct swap_info_struct *); | |
56 | ||
633423a0 | 57 | static DEFINE_SPINLOCK(swap_lock); |
7c363b8c | 58 | static unsigned int nr_swapfiles; |
ec8acf20 | 59 | atomic_long_t nr_swap_pages; |
fb0fec50 CW |
60 | /* |
61 | * Some modules use swappable objects and may try to swap them out under | |
62 | * memory pressure (via the shrinker). Before doing so, they may wish to | |
63 | * check to see if any swap space is available. | |
64 | */ | |
65 | EXPORT_SYMBOL_GPL(nr_swap_pages); | |
ec8acf20 | 66 | /* protected with swap_lock. reading in vm_swap_full() doesn't need lock */ |
1da177e4 | 67 | long total_swap_pages; |
a2468cc9 | 68 | static int least_priority = -1; |
be45a490 | 69 | unsigned long swapfile_maximum_size; |
5154e607 PX |
70 | #ifdef CONFIG_MIGRATION |
71 | bool swap_migration_ad_supported; | |
72 | #endif /* CONFIG_MIGRATION */ | |
1da177e4 | 73 | |
1da177e4 LT |
74 | static const char Bad_file[] = "Bad swap file entry "; |
75 | static const char Unused_file[] = "Unused swap file entry "; | |
76 | static const char Bad_offset[] = "Bad swap offset entry "; | |
77 | static const char Unused_offset[] = "Unused swap offset entry "; | |
78 | ||
adfab836 DS |
79 | /* |
80 | * all active swap_info_structs | |
81 | * protected with swap_lock, and ordered by priority. | |
82 | */ | |
633423a0 | 83 | static PLIST_HEAD(swap_active_head); |
18ab4d4c DS |
84 | |
85 | /* | |
86 | * all available (active, not full) swap_info_structs | |
87 | * protected with swap_avail_lock, ordered by priority. | |
e2e3fdc7 | 88 | * This is used by folio_alloc_swap() instead of swap_active_head |
18ab4d4c | 89 | * because swap_active_head includes all swap_info_structs, |
e2e3fdc7 | 90 | * but folio_alloc_swap() doesn't need to look at full ones. |
18ab4d4c DS |
91 | * This uses its own lock instead of swap_lock because when a |
92 | * swap_info_struct changes between not-full/full, it needs to | |
93 | * add/remove itself to/from this list, but the swap_info_struct->lock | |
94 | * is held and the locking order requires swap_lock to be taken | |
95 | * before any swap_info_struct->lock. | |
96 | */ | |
bfc6b1ca | 97 | static struct plist_head *swap_avail_heads; |
18ab4d4c | 98 | static DEFINE_SPINLOCK(swap_avail_lock); |
1da177e4 | 99 | |
42c06a0e | 100 | static struct swap_info_struct *swap_info[MAX_SWAPFILES]; |
1da177e4 | 101 | |
fc0abb14 | 102 | static DEFINE_MUTEX(swapon_mutex); |
1da177e4 | 103 | |
66d7dd51 KS |
104 | static DECLARE_WAIT_QUEUE_HEAD(proc_poll_wait); |
105 | /* Activity counter to indicate that a swapon or swapoff has occurred */ | |
106 | static atomic_t proc_poll_event = ATOMIC_INIT(0); | |
107 | ||
81a0298b HY |
108 | atomic_t nr_rotate_swap = ATOMIC_INIT(0); |
109 | ||
c10d38cc DJ |
110 | static struct swap_info_struct *swap_type_to_swap_info(int type) |
111 | { | |
a4b45114 | 112 | if (type >= MAX_SWAPFILES) |
c10d38cc DJ |
113 | return NULL; |
114 | ||
a4b45114 | 115 | return READ_ONCE(swap_info[type]); /* rcu_dereference() */ |
c10d38cc DJ |
116 | } |
117 | ||
8d69aaee | 118 | static inline unsigned char swap_count(unsigned char ent) |
355cfa73 | 119 | { |
955c97f0 | 120 | return ent & ~SWAP_HAS_CACHE; /* may include COUNT_CONTINUED flag */ |
355cfa73 KH |
121 | } |
122 | ||
bcd49e86 HY |
123 | /* Reclaim the swap entry anyway if possible */ |
124 | #define TTRS_ANYWAY 0x1 | |
125 | /* | |
126 | * Reclaim the swap entry if there are no more mappings of the | |
127 | * corresponding page | |
128 | */ | |
129 | #define TTRS_UNMAPPED 0x2 | |
130 | /* Reclaim the swap entry if swap is getting full*/ | |
131 | #define TTRS_FULL 0x4 | |
132 | ||
a62fb92a RR |
133 | /* |
134 | * returns number of pages in the folio that backs the swap entry. If positive, | |
135 | * the folio was reclaimed. If negative, the folio was not reclaimed. If 0, no | |
136 | * folio was associated with the swap entry. | |
137 | */ | |
bcd49e86 HY |
138 | static int __try_to_reclaim_swap(struct swap_info_struct *si, |
139 | unsigned long offset, unsigned long flags) | |
c9e44410 | 140 | { |
efa90a98 | 141 | swp_entry_t entry = swp_entry(si->type, offset); |
2c3f6194 | 142 | struct folio *folio; |
c9e44410 KH |
143 | int ret = 0; |
144 | ||
2c3f6194 | 145 | folio = filemap_get_folio(swap_address_space(entry), offset); |
66dabbb6 | 146 | if (IS_ERR(folio)) |
c9e44410 KH |
147 | return 0; |
148 | /* | |
bcd49e86 | 149 | * When this function is called from scan_swap_map_slots() and it's |
2c3f6194 | 150 | * called by vmscan.c at reclaiming folios. So we hold a folio lock |
bcd49e86 | 151 | * here. We have to use trylock for avoiding deadlock. This is a special |
2c3f6194 | 152 | * case and you should use folio_free_swap() with explicit folio_lock() |
c9e44410 KH |
153 | * in usual operations. |
154 | */ | |
2c3f6194 | 155 | if (folio_trylock(folio)) { |
bcd49e86 | 156 | if ((flags & TTRS_ANYWAY) || |
2c3f6194 | 157 | ((flags & TTRS_UNMAPPED) && !folio_mapped(folio)) || |
9202d527 | 158 | ((flags & TTRS_FULL) && mem_cgroup_swap_full(folio))) |
2c3f6194 MWO |
159 | ret = folio_free_swap(folio); |
160 | folio_unlock(folio); | |
c9e44410 | 161 | } |
a62fb92a | 162 | ret = ret ? folio_nr_pages(folio) : -folio_nr_pages(folio); |
2c3f6194 | 163 | folio_put(folio); |
c9e44410 KH |
164 | return ret; |
165 | } | |
355cfa73 | 166 | |
4efaceb1 AL |
167 | static inline struct swap_extent *first_se(struct swap_info_struct *sis) |
168 | { | |
169 | struct rb_node *rb = rb_first(&sis->swap_extent_root); | |
170 | return rb_entry(rb, struct swap_extent, rb_node); | |
171 | } | |
172 | ||
173 | static inline struct swap_extent *next_se(struct swap_extent *se) | |
174 | { | |
175 | struct rb_node *rb = rb_next(&se->rb_node); | |
176 | return rb ? rb_entry(rb, struct swap_extent, rb_node) : NULL; | |
177 | } | |
178 | ||
6a6ba831 HD |
179 | /* |
180 | * swapon tell device that all the old swap contents can be discarded, | |
181 | * to allow the swap device to optimize its wear-levelling. | |
182 | */ | |
183 | static int discard_swap(struct swap_info_struct *si) | |
184 | { | |
185 | struct swap_extent *se; | |
9625a5f2 HD |
186 | sector_t start_block; |
187 | sector_t nr_blocks; | |
6a6ba831 HD |
188 | int err = 0; |
189 | ||
9625a5f2 | 190 | /* Do not discard the swap header page! */ |
4efaceb1 | 191 | se = first_se(si); |
9625a5f2 HD |
192 | start_block = (se->start_block + 1) << (PAGE_SHIFT - 9); |
193 | nr_blocks = ((sector_t)se->nr_pages - 1) << (PAGE_SHIFT - 9); | |
194 | if (nr_blocks) { | |
195 | err = blkdev_issue_discard(si->bdev, start_block, | |
44abff2c | 196 | nr_blocks, GFP_KERNEL); |
9625a5f2 HD |
197 | if (err) |
198 | return err; | |
199 | cond_resched(); | |
200 | } | |
6a6ba831 | 201 | |
4efaceb1 | 202 | for (se = next_se(se); se; se = next_se(se)) { |
9625a5f2 HD |
203 | start_block = se->start_block << (PAGE_SHIFT - 9); |
204 | nr_blocks = (sector_t)se->nr_pages << (PAGE_SHIFT - 9); | |
6a6ba831 HD |
205 | |
206 | err = blkdev_issue_discard(si->bdev, start_block, | |
44abff2c | 207 | nr_blocks, GFP_KERNEL); |
6a6ba831 HD |
208 | if (err) |
209 | break; | |
210 | ||
211 | cond_resched(); | |
212 | } | |
213 | return err; /* That will often be -EOPNOTSUPP */ | |
214 | } | |
215 | ||
4efaceb1 AL |
216 | static struct swap_extent * |
217 | offset_to_swap_extent(struct swap_info_struct *sis, unsigned long offset) | |
218 | { | |
219 | struct swap_extent *se; | |
220 | struct rb_node *rb; | |
221 | ||
222 | rb = sis->swap_extent_root.rb_node; | |
223 | while (rb) { | |
224 | se = rb_entry(rb, struct swap_extent, rb_node); | |
225 | if (offset < se->start_page) | |
226 | rb = rb->rb_left; | |
227 | else if (offset >= se->start_page + se->nr_pages) | |
228 | rb = rb->rb_right; | |
229 | else | |
230 | return se; | |
231 | } | |
232 | /* It *must* be present */ | |
233 | BUG(); | |
234 | } | |
235 | ||
3a61e6f6 | 236 | sector_t swap_folio_sector(struct folio *folio) |
caf6912f | 237 | { |
3a61e6f6 | 238 | struct swap_info_struct *sis = swp_swap_info(folio->swap); |
caf6912f JA |
239 | struct swap_extent *se; |
240 | sector_t sector; | |
241 | pgoff_t offset; | |
242 | ||
3a61e6f6 | 243 | offset = swp_offset(folio->swap); |
caf6912f JA |
244 | se = offset_to_swap_extent(sis, offset); |
245 | sector = se->start_block + (offset - se->start_page); | |
246 | return sector << (PAGE_SHIFT - 9); | |
247 | } | |
248 | ||
7992fde7 HD |
249 | /* |
250 | * swap allocation tell device that a cluster of swap can now be discarded, | |
251 | * to allow the swap device to optimize its wear-levelling. | |
252 | */ | |
253 | static void discard_swap_cluster(struct swap_info_struct *si, | |
254 | pgoff_t start_page, pgoff_t nr_pages) | |
255 | { | |
4efaceb1 | 256 | struct swap_extent *se = offset_to_swap_extent(si, start_page); |
7992fde7 HD |
257 | |
258 | while (nr_pages) { | |
4efaceb1 AL |
259 | pgoff_t offset = start_page - se->start_page; |
260 | sector_t start_block = se->start_block + offset; | |
261 | sector_t nr_blocks = se->nr_pages - offset; | |
262 | ||
263 | if (nr_blocks > nr_pages) | |
264 | nr_blocks = nr_pages; | |
265 | start_page += nr_blocks; | |
266 | nr_pages -= nr_blocks; | |
267 | ||
268 | start_block <<= PAGE_SHIFT - 9; | |
269 | nr_blocks <<= PAGE_SHIFT - 9; | |
270 | if (blkdev_issue_discard(si->bdev, start_block, | |
44abff2c | 271 | nr_blocks, GFP_NOIO)) |
4efaceb1 | 272 | break; |
7992fde7 | 273 | |
4efaceb1 | 274 | se = next_se(se); |
7992fde7 HD |
275 | } |
276 | } | |
277 | ||
38d8b4e6 HY |
278 | #ifdef CONFIG_THP_SWAP |
279 | #define SWAPFILE_CLUSTER HPAGE_PMD_NR | |
a448f2d0 | 280 | |
9faaa0f8 | 281 | #define swap_entry_order(order) (order) |
38d8b4e6 | 282 | #else |
048c27fd | 283 | #define SWAPFILE_CLUSTER 256 |
a448f2d0 HY |
284 | |
285 | /* | |
9faaa0f8 | 286 | * Define swap_entry_order() as constant to let compiler to optimize |
a448f2d0 HY |
287 | * out some code if !CONFIG_THP_SWAP |
288 | */ | |
9faaa0f8 | 289 | #define swap_entry_order(order) 0 |
38d8b4e6 | 290 | #endif |
048c27fd HD |
291 | #define LATENCY_LIMIT 256 |
292 | ||
2a8f9449 SL |
293 | static inline void cluster_set_flag(struct swap_cluster_info *info, |
294 | unsigned int flag) | |
295 | { | |
296 | info->flags = flag; | |
297 | } | |
298 | ||
299 | static inline unsigned int cluster_count(struct swap_cluster_info *info) | |
300 | { | |
301 | return info->data; | |
302 | } | |
303 | ||
304 | static inline void cluster_set_count(struct swap_cluster_info *info, | |
305 | unsigned int c) | |
306 | { | |
307 | info->data = c; | |
308 | } | |
309 | ||
310 | static inline void cluster_set_count_flag(struct swap_cluster_info *info, | |
311 | unsigned int c, unsigned int f) | |
312 | { | |
313 | info->flags = f; | |
314 | info->data = c; | |
315 | } | |
316 | ||
317 | static inline unsigned int cluster_next(struct swap_cluster_info *info) | |
318 | { | |
319 | return info->data; | |
320 | } | |
321 | ||
322 | static inline void cluster_set_next(struct swap_cluster_info *info, | |
323 | unsigned int n) | |
324 | { | |
325 | info->data = n; | |
326 | } | |
327 | ||
328 | static inline void cluster_set_next_flag(struct swap_cluster_info *info, | |
329 | unsigned int n, unsigned int f) | |
330 | { | |
331 | info->flags = f; | |
332 | info->data = n; | |
333 | } | |
334 | ||
335 | static inline bool cluster_is_free(struct swap_cluster_info *info) | |
336 | { | |
337 | return info->flags & CLUSTER_FLAG_FREE; | |
338 | } | |
339 | ||
340 | static inline bool cluster_is_null(struct swap_cluster_info *info) | |
341 | { | |
342 | return info->flags & CLUSTER_FLAG_NEXT_NULL; | |
343 | } | |
344 | ||
345 | static inline void cluster_set_null(struct swap_cluster_info *info) | |
346 | { | |
347 | info->flags = CLUSTER_FLAG_NEXT_NULL; | |
348 | info->data = 0; | |
349 | } | |
350 | ||
235b6217 HY |
351 | static inline struct swap_cluster_info *lock_cluster(struct swap_info_struct *si, |
352 | unsigned long offset) | |
353 | { | |
354 | struct swap_cluster_info *ci; | |
355 | ||
356 | ci = si->cluster_info; | |
357 | if (ci) { | |
358 | ci += offset / SWAPFILE_CLUSTER; | |
359 | spin_lock(&ci->lock); | |
360 | } | |
361 | return ci; | |
362 | } | |
363 | ||
364 | static inline void unlock_cluster(struct swap_cluster_info *ci) | |
365 | { | |
366 | if (ci) | |
367 | spin_unlock(&ci->lock); | |
368 | } | |
369 | ||
59d98bf3 HY |
370 | /* |
371 | * Determine the locking method in use for this device. Return | |
372 | * swap_cluster_info if SSD-style cluster-based locking is in place. | |
373 | */ | |
235b6217 | 374 | static inline struct swap_cluster_info *lock_cluster_or_swap_info( |
59d98bf3 | 375 | struct swap_info_struct *si, unsigned long offset) |
235b6217 HY |
376 | { |
377 | struct swap_cluster_info *ci; | |
378 | ||
59d98bf3 | 379 | /* Try to use fine-grained SSD-style locking if available: */ |
235b6217 | 380 | ci = lock_cluster(si, offset); |
59d98bf3 | 381 | /* Otherwise, fall back to traditional, coarse locking: */ |
235b6217 HY |
382 | if (!ci) |
383 | spin_lock(&si->lock); | |
384 | ||
385 | return ci; | |
386 | } | |
387 | ||
388 | static inline void unlock_cluster_or_swap_info(struct swap_info_struct *si, | |
389 | struct swap_cluster_info *ci) | |
390 | { | |
391 | if (ci) | |
392 | unlock_cluster(ci); | |
393 | else | |
394 | spin_unlock(&si->lock); | |
395 | } | |
396 | ||
6b534915 HY |
397 | static inline bool cluster_list_empty(struct swap_cluster_list *list) |
398 | { | |
399 | return cluster_is_null(&list->head); | |
400 | } | |
401 | ||
402 | static inline unsigned int cluster_list_first(struct swap_cluster_list *list) | |
403 | { | |
404 | return cluster_next(&list->head); | |
405 | } | |
406 | ||
407 | static void cluster_list_init(struct swap_cluster_list *list) | |
408 | { | |
409 | cluster_set_null(&list->head); | |
410 | cluster_set_null(&list->tail); | |
411 | } | |
412 | ||
413 | static void cluster_list_add_tail(struct swap_cluster_list *list, | |
414 | struct swap_cluster_info *ci, | |
415 | unsigned int idx) | |
416 | { | |
417 | if (cluster_list_empty(list)) { | |
418 | cluster_set_next_flag(&list->head, idx, 0); | |
419 | cluster_set_next_flag(&list->tail, idx, 0); | |
420 | } else { | |
235b6217 | 421 | struct swap_cluster_info *ci_tail; |
6b534915 HY |
422 | unsigned int tail = cluster_next(&list->tail); |
423 | ||
235b6217 HY |
424 | /* |
425 | * Nested cluster lock, but both cluster locks are | |
426 | * only acquired when we held swap_info_struct->lock | |
427 | */ | |
428 | ci_tail = ci + tail; | |
429 | spin_lock_nested(&ci_tail->lock, SINGLE_DEPTH_NESTING); | |
430 | cluster_set_next(ci_tail, idx); | |
0ef017d1 | 431 | spin_unlock(&ci_tail->lock); |
6b534915 HY |
432 | cluster_set_next_flag(&list->tail, idx, 0); |
433 | } | |
434 | } | |
435 | ||
436 | static unsigned int cluster_list_del_first(struct swap_cluster_list *list, | |
437 | struct swap_cluster_info *ci) | |
438 | { | |
439 | unsigned int idx; | |
440 | ||
441 | idx = cluster_next(&list->head); | |
442 | if (cluster_next(&list->tail) == idx) { | |
443 | cluster_set_null(&list->head); | |
444 | cluster_set_null(&list->tail); | |
445 | } else | |
446 | cluster_set_next_flag(&list->head, | |
447 | cluster_next(&ci[idx]), 0); | |
448 | ||
449 | return idx; | |
450 | } | |
451 | ||
815c2c54 SL |
452 | /* Add a cluster to discard list and schedule it to do discard */ |
453 | static void swap_cluster_schedule_discard(struct swap_info_struct *si, | |
454 | unsigned int idx) | |
455 | { | |
456 | /* | |
bb243f7d | 457 | * If scan_swap_map_slots() can't find a free cluster, it will check |
815c2c54 | 458 | * si->swap_map directly. To make sure the discarding cluster isn't |
bb243f7d ML |
459 | * taken by scan_swap_map_slots(), mark the swap entries bad (occupied). |
460 | * It will be cleared after discard | |
815c2c54 SL |
461 | */ |
462 | memset(si->swap_map + idx * SWAPFILE_CLUSTER, | |
463 | SWAP_MAP_BAD, SWAPFILE_CLUSTER); | |
464 | ||
6b534915 | 465 | cluster_list_add_tail(&si->discard_clusters, si->cluster_info, idx); |
815c2c54 SL |
466 | |
467 | schedule_work(&si->discard_work); | |
468 | } | |
469 | ||
38d8b4e6 HY |
470 | static void __free_cluster(struct swap_info_struct *si, unsigned long idx) |
471 | { | |
472 | struct swap_cluster_info *ci = si->cluster_info; | |
473 | ||
474 | cluster_set_flag(ci + idx, CLUSTER_FLAG_FREE); | |
475 | cluster_list_add_tail(&si->free_clusters, ci, idx); | |
476 | } | |
477 | ||
815c2c54 SL |
478 | /* |
479 | * Doing discard actually. After a cluster discard is finished, the cluster | |
480 | * will be added to free cluster list. caller should hold si->lock. | |
481 | */ | |
482 | static void swap_do_scheduled_discard(struct swap_info_struct *si) | |
483 | { | |
235b6217 | 484 | struct swap_cluster_info *info, *ci; |
815c2c54 SL |
485 | unsigned int idx; |
486 | ||
487 | info = si->cluster_info; | |
488 | ||
6b534915 HY |
489 | while (!cluster_list_empty(&si->discard_clusters)) { |
490 | idx = cluster_list_del_first(&si->discard_clusters, info); | |
815c2c54 SL |
491 | spin_unlock(&si->lock); |
492 | ||
493 | discard_swap_cluster(si, idx * SWAPFILE_CLUSTER, | |
494 | SWAPFILE_CLUSTER); | |
495 | ||
496 | spin_lock(&si->lock); | |
235b6217 | 497 | ci = lock_cluster(si, idx * SWAPFILE_CLUSTER); |
38d8b4e6 | 498 | __free_cluster(si, idx); |
815c2c54 SL |
499 | memset(si->swap_map + idx * SWAPFILE_CLUSTER, |
500 | 0, SWAPFILE_CLUSTER); | |
235b6217 | 501 | unlock_cluster(ci); |
815c2c54 SL |
502 | } |
503 | } | |
504 | ||
505 | static void swap_discard_work(struct work_struct *work) | |
506 | { | |
507 | struct swap_info_struct *si; | |
508 | ||
509 | si = container_of(work, struct swap_info_struct, discard_work); | |
510 | ||
511 | spin_lock(&si->lock); | |
512 | swap_do_scheduled_discard(si); | |
513 | spin_unlock(&si->lock); | |
514 | } | |
515 | ||
63d8620e ML |
516 | static void swap_users_ref_free(struct percpu_ref *ref) |
517 | { | |
518 | struct swap_info_struct *si; | |
519 | ||
520 | si = container_of(ref, struct swap_info_struct, users); | |
521 | complete(&si->comp); | |
522 | } | |
523 | ||
38d8b4e6 HY |
524 | static void alloc_cluster(struct swap_info_struct *si, unsigned long idx) |
525 | { | |
526 | struct swap_cluster_info *ci = si->cluster_info; | |
527 | ||
528 | VM_BUG_ON(cluster_list_first(&si->free_clusters) != idx); | |
529 | cluster_list_del_first(&si->free_clusters, ci); | |
530 | cluster_set_count_flag(ci + idx, 0, 0); | |
531 | } | |
532 | ||
533 | static void free_cluster(struct swap_info_struct *si, unsigned long idx) | |
534 | { | |
535 | struct swap_cluster_info *ci = si->cluster_info + idx; | |
536 | ||
537 | VM_BUG_ON(cluster_count(ci) != 0); | |
538 | /* | |
539 | * If the swap is discardable, prepare discard the cluster | |
540 | * instead of free it immediately. The cluster will be freed | |
541 | * after discard. | |
542 | */ | |
543 | if ((si->flags & (SWP_WRITEOK | SWP_PAGE_DISCARD)) == | |
544 | (SWP_WRITEOK | SWP_PAGE_DISCARD)) { | |
545 | swap_cluster_schedule_discard(si, idx); | |
546 | return; | |
547 | } | |
548 | ||
549 | __free_cluster(si, idx); | |
550 | } | |
551 | ||
2a8f9449 SL |
552 | /* |
553 | * The cluster corresponding to page_nr will be used. The cluster will be | |
845982eb RR |
554 | * removed from free cluster list and its usage counter will be increased by |
555 | * count. | |
2a8f9449 | 556 | */ |
845982eb RR |
557 | static void add_cluster_info_page(struct swap_info_struct *p, |
558 | struct swap_cluster_info *cluster_info, unsigned long page_nr, | |
559 | unsigned long count) | |
2a8f9449 SL |
560 | { |
561 | unsigned long idx = page_nr / SWAPFILE_CLUSTER; | |
562 | ||
563 | if (!cluster_info) | |
564 | return; | |
38d8b4e6 HY |
565 | if (cluster_is_free(&cluster_info[idx])) |
566 | alloc_cluster(p, idx); | |
2a8f9449 | 567 | |
845982eb | 568 | VM_BUG_ON(cluster_count(&cluster_info[idx]) + count > SWAPFILE_CLUSTER); |
2a8f9449 | 569 | cluster_set_count(&cluster_info[idx], |
845982eb RR |
570 | cluster_count(&cluster_info[idx]) + count); |
571 | } | |
572 | ||
573 | /* | |
574 | * The cluster corresponding to page_nr will be used. The cluster will be | |
575 | * removed from free cluster list and its usage counter will be increased by 1. | |
576 | */ | |
577 | static void inc_cluster_info_page(struct swap_info_struct *p, | |
578 | struct swap_cluster_info *cluster_info, unsigned long page_nr) | |
579 | { | |
580 | add_cluster_info_page(p, cluster_info, page_nr, 1); | |
2a8f9449 SL |
581 | } |
582 | ||
583 | /* | |
584 | * The cluster corresponding to page_nr decreases one usage. If the usage | |
585 | * counter becomes 0, which means no page in the cluster is in using, we can | |
586 | * optionally discard the cluster and add it to free cluster list. | |
587 | */ | |
588 | static void dec_cluster_info_page(struct swap_info_struct *p, | |
589 | struct swap_cluster_info *cluster_info, unsigned long page_nr) | |
590 | { | |
591 | unsigned long idx = page_nr / SWAPFILE_CLUSTER; | |
592 | ||
593 | if (!cluster_info) | |
594 | return; | |
595 | ||
596 | VM_BUG_ON(cluster_count(&cluster_info[idx]) == 0); | |
597 | cluster_set_count(&cluster_info[idx], | |
598 | cluster_count(&cluster_info[idx]) - 1); | |
599 | ||
38d8b4e6 HY |
600 | if (cluster_count(&cluster_info[idx]) == 0) |
601 | free_cluster(p, idx); | |
2a8f9449 SL |
602 | } |
603 | ||
604 | /* | |
bb243f7d | 605 | * It's possible scan_swap_map_slots() uses a free cluster in the middle of free |
2a8f9449 SL |
606 | * cluster list. Avoiding such abuse to avoid list corruption. |
607 | */ | |
ebc2a1a6 SL |
608 | static bool |
609 | scan_swap_map_ssd_cluster_conflict(struct swap_info_struct *si, | |
845982eb | 610 | unsigned long offset, int order) |
2a8f9449 | 611 | { |
ebc2a1a6 SL |
612 | struct percpu_cluster *percpu_cluster; |
613 | bool conflict; | |
614 | ||
2a8f9449 | 615 | offset /= SWAPFILE_CLUSTER; |
6b534915 HY |
616 | conflict = !cluster_list_empty(&si->free_clusters) && |
617 | offset != cluster_list_first(&si->free_clusters) && | |
2a8f9449 | 618 | cluster_is_free(&si->cluster_info[offset]); |
ebc2a1a6 SL |
619 | |
620 | if (!conflict) | |
621 | return false; | |
622 | ||
623 | percpu_cluster = this_cpu_ptr(si->percpu_cluster); | |
845982eb RR |
624 | percpu_cluster->next[order] = SWAP_NEXT_INVALID; |
625 | return true; | |
626 | } | |
627 | ||
628 | static inline bool swap_range_empty(char *swap_map, unsigned int start, | |
629 | unsigned int nr_pages) | |
630 | { | |
631 | unsigned int i; | |
632 | ||
633 | for (i = 0; i < nr_pages; i++) { | |
634 | if (swap_map[start + i]) | |
635 | return false; | |
636 | } | |
637 | ||
ebc2a1a6 SL |
638 | return true; |
639 | } | |
640 | ||
641 | /* | |
845982eb RR |
642 | * Try to get swap entries with specified order from current cpu's swap entry |
643 | * pool (a cluster). This might involve allocating a new cluster for current CPU | |
644 | * too. | |
ebc2a1a6 | 645 | */ |
36005bae | 646 | static bool scan_swap_map_try_ssd_cluster(struct swap_info_struct *si, |
845982eb | 647 | unsigned long *offset, unsigned long *scan_base, int order) |
ebc2a1a6 | 648 | { |
845982eb | 649 | unsigned int nr_pages = 1 << order; |
ebc2a1a6 | 650 | struct percpu_cluster *cluster; |
235b6217 | 651 | struct swap_cluster_info *ci; |
14c62da2 | 652 | unsigned int tmp, max; |
ebc2a1a6 SL |
653 | |
654 | new_cluster: | |
655 | cluster = this_cpu_ptr(si->percpu_cluster); | |
845982eb | 656 | tmp = cluster->next[order]; |
14c62da2 | 657 | if (tmp == SWAP_NEXT_INVALID) { |
6b534915 | 658 | if (!cluster_list_empty(&si->free_clusters)) { |
14c62da2 | 659 | tmp = cluster_next(&si->free_clusters.head) * |
ebc2a1a6 | 660 | SWAPFILE_CLUSTER; |
6b534915 | 661 | } else if (!cluster_list_empty(&si->discard_clusters)) { |
ebc2a1a6 SL |
662 | /* |
663 | * we don't have free cluster but have some clusters in | |
49070588 HY |
664 | * discarding, do discard now and reclaim them, then |
665 | * reread cluster_next_cpu since we dropped si->lock | |
ebc2a1a6 SL |
666 | */ |
667 | swap_do_scheduled_discard(si); | |
49070588 HY |
668 | *scan_base = this_cpu_read(*si->cluster_next_cpu); |
669 | *offset = *scan_base; | |
ebc2a1a6 SL |
670 | goto new_cluster; |
671 | } else | |
36005bae | 672 | return false; |
ebc2a1a6 SL |
673 | } |
674 | ||
ebc2a1a6 SL |
675 | /* |
676 | * Other CPUs can use our cluster if they can't find a free cluster, | |
845982eb RR |
677 | * check if there is still free entry in the cluster, maintaining |
678 | * natural alignment. | |
ebc2a1a6 | 679 | */ |
14c62da2 | 680 | max = min_t(unsigned long, si->max, ALIGN(tmp + 1, SWAPFILE_CLUSTER)); |
7b9e2de1 WY |
681 | if (tmp < max) { |
682 | ci = lock_cluster(si, tmp); | |
683 | while (tmp < max) { | |
845982eb | 684 | if (swap_range_empty(si->swap_map, tmp, nr_pages)) |
7b9e2de1 | 685 | break; |
845982eb | 686 | tmp += nr_pages; |
7b9e2de1 WY |
687 | } |
688 | unlock_cluster(ci); | |
ebc2a1a6 | 689 | } |
0fd0e19e | 690 | if (tmp >= max) { |
845982eb | 691 | cluster->next[order] = SWAP_NEXT_INVALID; |
ebc2a1a6 SL |
692 | goto new_cluster; |
693 | } | |
ebc2a1a6 SL |
694 | *offset = tmp; |
695 | *scan_base = tmp; | |
845982eb RR |
696 | tmp += nr_pages; |
697 | cluster->next[order] = tmp < max ? tmp : SWAP_NEXT_INVALID; | |
fdff1deb | 698 | return true; |
2a8f9449 SL |
699 | } |
700 | ||
a2468cc9 AL |
701 | static void __del_from_avail_list(struct swap_info_struct *p) |
702 | { | |
703 | int nid; | |
704 | ||
6fe7d6b9 | 705 | assert_spin_locked(&p->lock); |
a2468cc9 AL |
706 | for_each_node(nid) |
707 | plist_del(&p->avail_lists[nid], &swap_avail_heads[nid]); | |
708 | } | |
709 | ||
710 | static void del_from_avail_list(struct swap_info_struct *p) | |
711 | { | |
712 | spin_lock(&swap_avail_lock); | |
713 | __del_from_avail_list(p); | |
714 | spin_unlock(&swap_avail_lock); | |
715 | } | |
716 | ||
38d8b4e6 HY |
717 | static void swap_range_alloc(struct swap_info_struct *si, unsigned long offset, |
718 | unsigned int nr_entries) | |
719 | { | |
720 | unsigned int end = offset + nr_entries - 1; | |
721 | ||
722 | if (offset == si->lowest_bit) | |
723 | si->lowest_bit += nr_entries; | |
724 | if (end == si->highest_bit) | |
a449bf58 | 725 | WRITE_ONCE(si->highest_bit, si->highest_bit - nr_entries); |
c8945306 | 726 | WRITE_ONCE(si->inuse_pages, si->inuse_pages + nr_entries); |
38d8b4e6 HY |
727 | if (si->inuse_pages == si->pages) { |
728 | si->lowest_bit = si->max; | |
729 | si->highest_bit = 0; | |
a2468cc9 | 730 | del_from_avail_list(si); |
38d8b4e6 HY |
731 | } |
732 | } | |
733 | ||
a2468cc9 AL |
734 | static void add_to_avail_list(struct swap_info_struct *p) |
735 | { | |
736 | int nid; | |
737 | ||
738 | spin_lock(&swap_avail_lock); | |
67490031 | 739 | for_each_node(nid) |
a2468cc9 | 740 | plist_add(&p->avail_lists[nid], &swap_avail_heads[nid]); |
a2468cc9 AL |
741 | spin_unlock(&swap_avail_lock); |
742 | } | |
743 | ||
38d8b4e6 HY |
744 | static void swap_range_free(struct swap_info_struct *si, unsigned long offset, |
745 | unsigned int nr_entries) | |
746 | { | |
3852f676 | 747 | unsigned long begin = offset; |
38d8b4e6 HY |
748 | unsigned long end = offset + nr_entries - 1; |
749 | void (*swap_slot_free_notify)(struct block_device *, unsigned long); | |
750 | ||
751 | if (offset < si->lowest_bit) | |
752 | si->lowest_bit = offset; | |
753 | if (end > si->highest_bit) { | |
754 | bool was_full = !si->highest_bit; | |
755 | ||
a449bf58 | 756 | WRITE_ONCE(si->highest_bit, end); |
a2468cc9 AL |
757 | if (was_full && (si->flags & SWP_WRITEOK)) |
758 | add_to_avail_list(si); | |
38d8b4e6 | 759 | } |
38d8b4e6 HY |
760 | if (si->flags & SWP_BLKDEV) |
761 | swap_slot_free_notify = | |
762 | si->bdev->bd_disk->fops->swap_slot_free_notify; | |
763 | else | |
764 | swap_slot_free_notify = NULL; | |
765 | while (offset <= end) { | |
8a84802e | 766 | arch_swap_invalidate_page(si->type, offset); |
38d8b4e6 HY |
767 | if (swap_slot_free_notify) |
768 | swap_slot_free_notify(si->bdev, offset); | |
769 | offset++; | |
770 | } | |
3852f676 | 771 | clear_shadow_from_swap_cache(si->type, begin, end); |
64cf264c YA |
772 | |
773 | /* | |
774 | * Make sure that try_to_unuse() observes si->inuse_pages reaching 0 | |
775 | * only after the above cleanups are done. | |
776 | */ | |
777 | smp_wmb(); | |
778 | atomic_long_add(nr_entries, &nr_swap_pages); | |
779 | WRITE_ONCE(si->inuse_pages, si->inuse_pages - nr_entries); | |
38d8b4e6 HY |
780 | } |
781 | ||
49070588 HY |
782 | static void set_cluster_next(struct swap_info_struct *si, unsigned long next) |
783 | { | |
784 | unsigned long prev; | |
785 | ||
786 | if (!(si->flags & SWP_SOLIDSTATE)) { | |
787 | si->cluster_next = next; | |
788 | return; | |
789 | } | |
790 | ||
791 | prev = this_cpu_read(*si->cluster_next_cpu); | |
792 | /* | |
793 | * Cross the swap address space size aligned trunk, choose | |
794 | * another trunk randomly to avoid lock contention on swap | |
795 | * address space if possible. | |
796 | */ | |
797 | if ((prev >> SWAP_ADDRESS_SPACE_SHIFT) != | |
798 | (next >> SWAP_ADDRESS_SPACE_SHIFT)) { | |
799 | /* No free swap slots available */ | |
800 | if (si->highest_bit <= si->lowest_bit) | |
801 | return; | |
e8a533cb | 802 | next = get_random_u32_inclusive(si->lowest_bit, si->highest_bit); |
49070588 HY |
803 | next = ALIGN_DOWN(next, SWAP_ADDRESS_SPACE_PAGES); |
804 | next = max_t(unsigned int, next, si->lowest_bit); | |
805 | } | |
806 | this_cpu_write(*si->cluster_next_cpu, next); | |
807 | } | |
808 | ||
4b9ae842 ML |
809 | static bool swap_offset_available_and_locked(struct swap_info_struct *si, |
810 | unsigned long offset) | |
811 | { | |
812 | if (data_race(!si->swap_map[offset])) { | |
813 | spin_lock(&si->lock); | |
814 | return true; | |
815 | } | |
816 | ||
817 | if (vm_swap_full() && READ_ONCE(si->swap_map[offset]) == SWAP_HAS_CACHE) { | |
818 | spin_lock(&si->lock); | |
819 | return true; | |
820 | } | |
821 | ||
822 | return false; | |
823 | } | |
824 | ||
36005bae TC |
825 | static int scan_swap_map_slots(struct swap_info_struct *si, |
826 | unsigned char usage, int nr, | |
845982eb | 827 | swp_entry_t slots[], int order) |
1da177e4 | 828 | { |
235b6217 | 829 | struct swap_cluster_info *ci; |
ebebbbe9 | 830 | unsigned long offset; |
c60aa176 | 831 | unsigned long scan_base; |
7992fde7 | 832 | unsigned long last_in_cluster = 0; |
048c27fd | 833 | int latency_ration = LATENCY_LIMIT; |
845982eb | 834 | unsigned int nr_pages = 1 << order; |
36005bae | 835 | int n_ret = 0; |
ed43af10 | 836 | bool scanned_many = false; |
36005bae | 837 | |
886bb7e9 | 838 | /* |
7dfad418 HD |
839 | * We try to cluster swap pages by allocating them sequentially |
840 | * in swap. Once we've allocated SWAPFILE_CLUSTER pages this | |
841 | * way, however, we resort to first-free allocation, starting | |
842 | * a new cluster. This prevents us from scattering swap pages | |
843 | * all over the entire swap partition, so that we reduce | |
844 | * overall disk seek times between swap pages. -- sct | |
845 | * But we do now try to find an empty cluster. -Andrea | |
c60aa176 | 846 | * And we let swap pages go all over an SSD partition. Hugh |
7dfad418 HD |
847 | */ |
848 | ||
845982eb RR |
849 | if (order > 0) { |
850 | /* | |
851 | * Should not even be attempting large allocations when huge | |
852 | * page swap is disabled. Warn and fail the allocation. | |
853 | */ | |
854 | if (!IS_ENABLED(CONFIG_THP_SWAP) || | |
855 | nr_pages > SWAPFILE_CLUSTER) { | |
856 | VM_WARN_ON_ONCE(1); | |
857 | return 0; | |
858 | } | |
859 | ||
860 | /* | |
861 | * Swapfile is not block device or not using clusters so unable | |
862 | * to allocate large entries. | |
863 | */ | |
864 | if (!(si->flags & SWP_BLKDEV) || !si->cluster_info) | |
865 | return 0; | |
866 | } | |
867 | ||
52b7efdb | 868 | si->flags += SWP_SCANNING; |
49070588 HY |
869 | /* |
870 | * Use percpu scan base for SSD to reduce lock contention on | |
871 | * cluster and swap cache. For HDD, sequential access is more | |
872 | * important. | |
873 | */ | |
874 | if (si->flags & SWP_SOLIDSTATE) | |
875 | scan_base = this_cpu_read(*si->cluster_next_cpu); | |
876 | else | |
877 | scan_base = si->cluster_next; | |
878 | offset = scan_base; | |
ebebbbe9 | 879 | |
ebc2a1a6 SL |
880 | /* SSD algorithm */ |
881 | if (si->cluster_info) { | |
845982eb RR |
882 | if (!scan_swap_map_try_ssd_cluster(si, &offset, &scan_base, order)) { |
883 | if (order > 0) | |
884 | goto no_page; | |
36005bae | 885 | goto scan; |
845982eb | 886 | } |
f4eaf51a | 887 | } else if (unlikely(!si->cluster_nr--)) { |
ebebbbe9 HD |
888 | if (si->pages - si->inuse_pages < SWAPFILE_CLUSTER) { |
889 | si->cluster_nr = SWAPFILE_CLUSTER - 1; | |
890 | goto checks; | |
891 | } | |
2a8f9449 | 892 | |
ec8acf20 | 893 | spin_unlock(&si->lock); |
7dfad418 | 894 | |
c60aa176 HD |
895 | /* |
896 | * If seek is expensive, start searching for new cluster from | |
897 | * start of partition, to minimize the span of allocated swap. | |
50088c44 CY |
898 | * If seek is cheap, that is the SWP_SOLIDSTATE si->cluster_info |
899 | * case, just handled by scan_swap_map_try_ssd_cluster() above. | |
c60aa176 | 900 | */ |
50088c44 | 901 | scan_base = offset = si->lowest_bit; |
7dfad418 HD |
902 | last_in_cluster = offset + SWAPFILE_CLUSTER - 1; |
903 | ||
904 | /* Locate the first empty (unaligned) cluster */ | |
905 | for (; last_in_cluster <= si->highest_bit; offset++) { | |
1da177e4 | 906 | if (si->swap_map[offset]) |
7dfad418 HD |
907 | last_in_cluster = offset + SWAPFILE_CLUSTER; |
908 | else if (offset == last_in_cluster) { | |
ec8acf20 | 909 | spin_lock(&si->lock); |
ebebbbe9 HD |
910 | offset -= SWAPFILE_CLUSTER - 1; |
911 | si->cluster_next = offset; | |
912 | si->cluster_nr = SWAPFILE_CLUSTER - 1; | |
c60aa176 HD |
913 | goto checks; |
914 | } | |
915 | if (unlikely(--latency_ration < 0)) { | |
916 | cond_resched(); | |
917 | latency_ration = LATENCY_LIMIT; | |
918 | } | |
919 | } | |
920 | ||
921 | offset = scan_base; | |
ec8acf20 | 922 | spin_lock(&si->lock); |
ebebbbe9 | 923 | si->cluster_nr = SWAPFILE_CLUSTER - 1; |
1da177e4 | 924 | } |
7dfad418 | 925 | |
ebebbbe9 | 926 | checks: |
ebc2a1a6 | 927 | if (si->cluster_info) { |
845982eb | 928 | while (scan_swap_map_ssd_cluster_conflict(si, offset, order)) { |
36005bae TC |
929 | /* take a break if we already got some slots */ |
930 | if (n_ret) | |
931 | goto done; | |
932 | if (!scan_swap_map_try_ssd_cluster(si, &offset, | |
845982eb RR |
933 | &scan_base, order)) { |
934 | if (order > 0) | |
935 | goto no_page; | |
36005bae | 936 | goto scan; |
845982eb | 937 | } |
36005bae | 938 | } |
ebc2a1a6 | 939 | } |
ebebbbe9 | 940 | if (!(si->flags & SWP_WRITEOK)) |
52b7efdb | 941 | goto no_page; |
7dfad418 HD |
942 | if (!si->highest_bit) |
943 | goto no_page; | |
ebebbbe9 | 944 | if (offset > si->highest_bit) |
c60aa176 | 945 | scan_base = offset = si->lowest_bit; |
c9e44410 | 946 | |
235b6217 | 947 | ci = lock_cluster(si, offset); |
b73d7fce HD |
948 | /* reuse swap entry of cache-only swap if not busy. */ |
949 | if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) { | |
c9e44410 | 950 | int swap_was_freed; |
235b6217 | 951 | unlock_cluster(ci); |
ec8acf20 | 952 | spin_unlock(&si->lock); |
bcd49e86 | 953 | swap_was_freed = __try_to_reclaim_swap(si, offset, TTRS_ANYWAY); |
ec8acf20 | 954 | spin_lock(&si->lock); |
c9e44410 | 955 | /* entry was freed successfully, try to use this again */ |
a62fb92a | 956 | if (swap_was_freed > 0) |
c9e44410 KH |
957 | goto checks; |
958 | goto scan; /* check next one */ | |
959 | } | |
960 | ||
235b6217 HY |
961 | if (si->swap_map[offset]) { |
962 | unlock_cluster(ci); | |
36005bae TC |
963 | if (!n_ret) |
964 | goto scan; | |
965 | else | |
966 | goto done; | |
235b6217 | 967 | } |
845982eb RR |
968 | memset(si->swap_map + offset, usage, nr_pages); |
969 | add_cluster_info_page(si, si->cluster_info, offset, nr_pages); | |
2872bb2d | 970 | unlock_cluster(ci); |
ebebbbe9 | 971 | |
845982eb | 972 | swap_range_alloc(si, offset, nr_pages); |
36005bae TC |
973 | slots[n_ret++] = swp_entry(si->type, offset); |
974 | ||
975 | /* got enough slots or reach max slots? */ | |
976 | if ((n_ret == nr) || (offset >= si->highest_bit)) | |
977 | goto done; | |
978 | ||
979 | /* search for next available slot */ | |
980 | ||
981 | /* time to take a break? */ | |
982 | if (unlikely(--latency_ration < 0)) { | |
983 | if (n_ret) | |
984 | goto done; | |
985 | spin_unlock(&si->lock); | |
986 | cond_resched(); | |
987 | spin_lock(&si->lock); | |
988 | latency_ration = LATENCY_LIMIT; | |
989 | } | |
990 | ||
991 | /* try to get more slots in cluster */ | |
992 | if (si->cluster_info) { | |
845982eb | 993 | if (scan_swap_map_try_ssd_cluster(si, &offset, &scan_base, order)) |
36005bae | 994 | goto checks; |
845982eb RR |
995 | if (order > 0) |
996 | goto done; | |
f4eaf51a WY |
997 | } else if (si->cluster_nr && !si->swap_map[++offset]) { |
998 | /* non-ssd case, still more slots in cluster? */ | |
36005bae TC |
999 | --si->cluster_nr; |
1000 | goto checks; | |
1001 | } | |
7992fde7 | 1002 | |
ed43af10 HY |
1003 | /* |
1004 | * Even if there's no free clusters available (fragmented), | |
1005 | * try to scan a little more quickly with lock held unless we | |
1006 | * have scanned too many slots already. | |
1007 | */ | |
1008 | if (!scanned_many) { | |
1009 | unsigned long scan_limit; | |
1010 | ||
1011 | if (offset < scan_base) | |
1012 | scan_limit = scan_base; | |
1013 | else | |
1014 | scan_limit = si->highest_bit; | |
1015 | for (; offset <= scan_limit && --latency_ration > 0; | |
1016 | offset++) { | |
1017 | if (!si->swap_map[offset]) | |
1018 | goto checks; | |
1019 | } | |
1020 | } | |
1021 | ||
36005bae | 1022 | done: |
845982eb RR |
1023 | if (order == 0) |
1024 | set_cluster_next(si, offset + 1); | |
36005bae TC |
1025 | si->flags -= SWP_SCANNING; |
1026 | return n_ret; | |
7dfad418 | 1027 | |
ebebbbe9 | 1028 | scan: |
845982eb | 1029 | VM_WARN_ON(order > 0); |
ec8acf20 | 1030 | spin_unlock(&si->lock); |
a449bf58 | 1031 | while (++offset <= READ_ONCE(si->highest_bit)) { |
048c27fd HD |
1032 | if (unlikely(--latency_ration < 0)) { |
1033 | cond_resched(); | |
1034 | latency_ration = LATENCY_LIMIT; | |
ed43af10 | 1035 | scanned_many = true; |
048c27fd | 1036 | } |
de1ccfb6 CW |
1037 | if (swap_offset_available_and_locked(si, offset)) |
1038 | goto checks; | |
7dfad418 | 1039 | } |
c60aa176 | 1040 | offset = si->lowest_bit; |
a5998061 | 1041 | while (offset < scan_base) { |
c60aa176 HD |
1042 | if (unlikely(--latency_ration < 0)) { |
1043 | cond_resched(); | |
1044 | latency_ration = LATENCY_LIMIT; | |
ed43af10 | 1045 | scanned_many = true; |
c60aa176 | 1046 | } |
de1ccfb6 CW |
1047 | if (swap_offset_available_and_locked(si, offset)) |
1048 | goto checks; | |
a5998061 | 1049 | offset++; |
c60aa176 | 1050 | } |
ec8acf20 | 1051 | spin_lock(&si->lock); |
7dfad418 HD |
1052 | |
1053 | no_page: | |
52b7efdb | 1054 | si->flags -= SWP_SCANNING; |
36005bae | 1055 | return n_ret; |
1da177e4 LT |
1056 | } |
1057 | ||
38d8b4e6 HY |
1058 | static void swap_free_cluster(struct swap_info_struct *si, unsigned long idx) |
1059 | { | |
1060 | unsigned long offset = idx * SWAPFILE_CLUSTER; | |
1061 | struct swap_cluster_info *ci; | |
1062 | ||
1063 | ci = lock_cluster(si, offset); | |
979aafa5 | 1064 | memset(si->swap_map + offset, 0, SWAPFILE_CLUSTER); |
38d8b4e6 HY |
1065 | cluster_set_count_flag(ci, 0, 0); |
1066 | free_cluster(si, idx); | |
1067 | unlock_cluster(ci); | |
1068 | swap_range_free(si, offset, SWAPFILE_CLUSTER); | |
1069 | } | |
38d8b4e6 | 1070 | |
9faaa0f8 | 1071 | int get_swap_pages(int n_goal, swp_entry_t swp_entries[], int entry_order) |
1da177e4 | 1072 | { |
9faaa0f8 RR |
1073 | int order = swap_entry_order(entry_order); |
1074 | unsigned long size = 1 << order; | |
adfab836 | 1075 | struct swap_info_struct *si, *next; |
36005bae TC |
1076 | long avail_pgs; |
1077 | int n_ret = 0; | |
a2468cc9 | 1078 | int node; |
1da177e4 | 1079 | |
b50da6e9 ZH |
1080 | spin_lock(&swap_avail_lock); |
1081 | ||
5d5e8f19 | 1082 | avail_pgs = atomic_long_read(&nr_swap_pages) / size; |
b50da6e9 ZH |
1083 | if (avail_pgs <= 0) { |
1084 | spin_unlock(&swap_avail_lock); | |
fb4f88dc | 1085 | goto noswap; |
b50da6e9 | 1086 | } |
36005bae | 1087 | |
08d3090f | 1088 | n_goal = min3((long)n_goal, (long)SWAP_BATCH, avail_pgs); |
36005bae | 1089 | |
5d5e8f19 | 1090 | atomic_long_sub(n_goal * size, &nr_swap_pages); |
fb4f88dc | 1091 | |
18ab4d4c | 1092 | start_over: |
a2468cc9 AL |
1093 | node = numa_node_id(); |
1094 | plist_for_each_entry_safe(si, next, &swap_avail_heads[node], avail_lists[node]) { | |
18ab4d4c | 1095 | /* requeue si to after same-priority siblings */ |
a2468cc9 | 1096 | plist_requeue(&si->avail_lists[node], &swap_avail_heads[node]); |
18ab4d4c | 1097 | spin_unlock(&swap_avail_lock); |
ec8acf20 | 1098 | spin_lock(&si->lock); |
adfab836 | 1099 | if (!si->highest_bit || !(si->flags & SWP_WRITEOK)) { |
18ab4d4c | 1100 | spin_lock(&swap_avail_lock); |
a2468cc9 | 1101 | if (plist_node_empty(&si->avail_lists[node])) { |
18ab4d4c DS |
1102 | spin_unlock(&si->lock); |
1103 | goto nextsi; | |
1104 | } | |
1105 | WARN(!si->highest_bit, | |
1106 | "swap_info %d in list but !highest_bit\n", | |
1107 | si->type); | |
1108 | WARN(!(si->flags & SWP_WRITEOK), | |
1109 | "swap_info %d in list but !SWP_WRITEOK\n", | |
1110 | si->type); | |
a2468cc9 | 1111 | __del_from_avail_list(si); |
ec8acf20 | 1112 | spin_unlock(&si->lock); |
18ab4d4c | 1113 | goto nextsi; |
ec8acf20 | 1114 | } |
845982eb RR |
1115 | n_ret = scan_swap_map_slots(si, SWAP_HAS_CACHE, |
1116 | n_goal, swp_entries, order); | |
ec8acf20 | 1117 | spin_unlock(&si->lock); |
845982eb | 1118 | if (n_ret || size > 1) |
36005bae | 1119 | goto check_out; |
7717fc1a | 1120 | cond_resched(); |
36005bae | 1121 | |
18ab4d4c DS |
1122 | spin_lock(&swap_avail_lock); |
1123 | nextsi: | |
adfab836 DS |
1124 | /* |
1125 | * if we got here, it's likely that si was almost full before, | |
bb243f7d ML |
1126 | * and since scan_swap_map_slots() can drop the si->lock, |
1127 | * multiple callers probably all tried to get a page from the | |
1128 | * same si and it filled up before we could get one; or, the si | |
1129 | * filled up between us dropping swap_avail_lock and taking | |
1130 | * si->lock. Since we dropped the swap_avail_lock, the | |
1131 | * swap_avail_head list may have been modified; so if next is | |
1132 | * still in the swap_avail_head list then try it, otherwise | |
1133 | * start over if we have not gotten any slots. | |
adfab836 | 1134 | */ |
a2468cc9 | 1135 | if (plist_node_empty(&next->avail_lists[node])) |
18ab4d4c | 1136 | goto start_over; |
1da177e4 | 1137 | } |
fb4f88dc | 1138 | |
18ab4d4c DS |
1139 | spin_unlock(&swap_avail_lock); |
1140 | ||
36005bae TC |
1141 | check_out: |
1142 | if (n_ret < n_goal) | |
5d5e8f19 | 1143 | atomic_long_add((long)(n_goal - n_ret) * size, |
38d8b4e6 | 1144 | &nr_swap_pages); |
fb4f88dc | 1145 | noswap: |
36005bae TC |
1146 | return n_ret; |
1147 | } | |
1148 | ||
afba72b1 | 1149 | static struct swap_info_struct *_swap_info_get(swp_entry_t entry) |
1da177e4 | 1150 | { |
73c34b6a | 1151 | struct swap_info_struct *p; |
eb085574 | 1152 | unsigned long offset; |
1da177e4 LT |
1153 | |
1154 | if (!entry.val) | |
1155 | goto out; | |
eb085574 | 1156 | p = swp_swap_info(entry); |
c10d38cc | 1157 | if (!p) |
1da177e4 | 1158 | goto bad_nofile; |
a449bf58 | 1159 | if (data_race(!(p->flags & SWP_USED))) |
1da177e4 LT |
1160 | goto bad_device; |
1161 | offset = swp_offset(entry); | |
1162 | if (offset >= p->max) | |
1163 | goto bad_offset; | |
afba72b1 ML |
1164 | if (data_race(!p->swap_map[swp_offset(entry)])) |
1165 | goto bad_free; | |
1da177e4 LT |
1166 | return p; |
1167 | ||
afba72b1 ML |
1168 | bad_free: |
1169 | pr_err("%s: %s%08lx\n", __func__, Unused_offset, entry.val); | |
1170 | goto out; | |
1da177e4 | 1171 | bad_offset: |
cf532faa | 1172 | pr_err("%s: %s%08lx\n", __func__, Bad_offset, entry.val); |
1da177e4 LT |
1173 | goto out; |
1174 | bad_device: | |
cf532faa | 1175 | pr_err("%s: %s%08lx\n", __func__, Unused_file, entry.val); |
1da177e4 LT |
1176 | goto out; |
1177 | bad_nofile: | |
cf532faa | 1178 | pr_err("%s: %s%08lx\n", __func__, Bad_file, entry.val); |
1da177e4 LT |
1179 | out: |
1180 | return NULL; | |
886bb7e9 | 1181 | } |
1da177e4 | 1182 | |
7c00bafe TC |
1183 | static struct swap_info_struct *swap_info_get_cont(swp_entry_t entry, |
1184 | struct swap_info_struct *q) | |
1185 | { | |
1186 | struct swap_info_struct *p; | |
1187 | ||
1188 | p = _swap_info_get(entry); | |
1189 | ||
1190 | if (p != q) { | |
1191 | if (q != NULL) | |
1192 | spin_unlock(&q->lock); | |
1193 | if (p != NULL) | |
1194 | spin_lock(&p->lock); | |
1195 | } | |
1196 | return p; | |
1197 | } | |
1198 | ||
b32d5f32 HY |
1199 | static unsigned char __swap_entry_free_locked(struct swap_info_struct *p, |
1200 | unsigned long offset, | |
1201 | unsigned char usage) | |
1da177e4 | 1202 | { |
8d69aaee HD |
1203 | unsigned char count; |
1204 | unsigned char has_cache; | |
235b6217 | 1205 | |
253d553b | 1206 | count = p->swap_map[offset]; |
235b6217 | 1207 | |
253d553b HD |
1208 | has_cache = count & SWAP_HAS_CACHE; |
1209 | count &= ~SWAP_HAS_CACHE; | |
355cfa73 | 1210 | |
253d553b | 1211 | if (usage == SWAP_HAS_CACHE) { |
355cfa73 | 1212 | VM_BUG_ON(!has_cache); |
253d553b | 1213 | has_cache = 0; |
aaa46865 HD |
1214 | } else if (count == SWAP_MAP_SHMEM) { |
1215 | /* | |
1216 | * Or we could insist on shmem.c using a special | |
1217 | * swap_shmem_free() and free_shmem_swap_and_cache()... | |
1218 | */ | |
1219 | count = 0; | |
570a335b HD |
1220 | } else if ((count & ~COUNT_CONTINUED) <= SWAP_MAP_MAX) { |
1221 | if (count == COUNT_CONTINUED) { | |
1222 | if (swap_count_continued(p, offset, count)) | |
1223 | count = SWAP_MAP_MAX | COUNT_CONTINUED; | |
1224 | else | |
1225 | count = SWAP_MAP_MAX; | |
1226 | } else | |
1227 | count--; | |
1228 | } | |
253d553b | 1229 | |
253d553b | 1230 | usage = count | has_cache; |
a449bf58 QC |
1231 | if (usage) |
1232 | WRITE_ONCE(p->swap_map[offset], usage); | |
1233 | else | |
1234 | WRITE_ONCE(p->swap_map[offset], SWAP_HAS_CACHE); | |
7c00bafe | 1235 | |
b32d5f32 HY |
1236 | return usage; |
1237 | } | |
1238 | ||
eb085574 | 1239 | /* |
a95722a0 | 1240 | * When we get a swap entry, if there aren't some other ways to |
d4a34d7f HY |
1241 | * prevent swapoff, such as the folio in swap cache is locked, RCU |
1242 | * reader side is locked, etc., the swap entry may become invalid | |
1243 | * because of swapoff. Then, we need to enclose all swap related | |
1244 | * functions with get_swap_device() and put_swap_device(), unless the | |
1245 | * swap functions call get/put_swap_device() by themselves. | |
a95722a0 | 1246 | * |
d4a34d7f HY |
1247 | * RCU reader side lock (including any spinlock) is sufficient to |
1248 | * prevent swapoff, because synchronize_rcu() is called in swapoff() | |
1249 | * before freeing data structures. | |
82b1c07a | 1250 | * |
eb085574 HY |
1251 | * Check whether swap entry is valid in the swap device. If so, |
1252 | * return pointer to swap_info_struct, and keep the swap entry valid | |
1253 | * via preventing the swap device from being swapoff, until | |
1254 | * put_swap_device() is called. Otherwise return NULL. | |
1255 | * | |
eb085574 | 1256 | * Notice that swapoff or swapoff+swapon can still happen before the |
63d8620e ML |
1257 | * percpu_ref_tryget_live() in get_swap_device() or after the |
1258 | * percpu_ref_put() in put_swap_device() if there isn't any other way | |
a95722a0 HY |
1259 | * to prevent swapoff. The caller must be prepared for that. For |
1260 | * example, the following situation is possible. | |
eb085574 HY |
1261 | * |
1262 | * CPU1 CPU2 | |
1263 | * do_swap_page() | |
1264 | * ... swapoff+swapon | |
1265 | * __read_swap_cache_async() | |
1266 | * swapcache_prepare() | |
1267 | * __swap_duplicate() | |
1268 | * // check swap_map | |
1269 | * // verify PTE not changed | |
1270 | * | |
1271 | * In __swap_duplicate(), the swap_map need to be checked before | |
1272 | * changing partly because the specified swap entry may be for another | |
1273 | * swap device which has been swapoff. And in do_swap_page(), after | |
1274 | * the page is read from the swap device, the PTE is verified not | |
1275 | * changed with the page table locked to check whether the swap device | |
1276 | * has been swapoff or swapoff+swapon. | |
1277 | */ | |
1278 | struct swap_info_struct *get_swap_device(swp_entry_t entry) | |
1279 | { | |
1280 | struct swap_info_struct *si; | |
1281 | unsigned long offset; | |
1282 | ||
1283 | if (!entry.val) | |
1284 | goto out; | |
1285 | si = swp_swap_info(entry); | |
1286 | if (!si) | |
1287 | goto bad_nofile; | |
63d8620e ML |
1288 | if (!percpu_ref_tryget_live(&si->users)) |
1289 | goto out; | |
1290 | /* | |
1291 | * Guarantee the si->users are checked before accessing other | |
1292 | * fields of swap_info_struct. | |
1293 | * | |
1294 | * Paired with the spin_unlock() after setup_swap_info() in | |
1295 | * enable_swap_info(). | |
1296 | */ | |
1297 | smp_rmb(); | |
eb085574 HY |
1298 | offset = swp_offset(entry); |
1299 | if (offset >= si->max) | |
63d8620e | 1300 | goto put_out; |
eb085574 HY |
1301 | |
1302 | return si; | |
1303 | bad_nofile: | |
1304 | pr_err("%s: %s%08lx\n", __func__, Bad_file, entry.val); | |
1305 | out: | |
1306 | return NULL; | |
63d8620e | 1307 | put_out: |
23b230ba | 1308 | pr_err("%s: %s%08lx\n", __func__, Bad_offset, entry.val); |
63d8620e | 1309 | percpu_ref_put(&si->users); |
eb085574 HY |
1310 | return NULL; |
1311 | } | |
1312 | ||
b32d5f32 | 1313 | static unsigned char __swap_entry_free(struct swap_info_struct *p, |
33e16272 | 1314 | swp_entry_t entry) |
b32d5f32 HY |
1315 | { |
1316 | struct swap_cluster_info *ci; | |
1317 | unsigned long offset = swp_offset(entry); | |
33e16272 | 1318 | unsigned char usage; |
b32d5f32 HY |
1319 | |
1320 | ci = lock_cluster_or_swap_info(p, offset); | |
33e16272 | 1321 | usage = __swap_entry_free_locked(p, offset, 1); |
7c00bafe | 1322 | unlock_cluster_or_swap_info(p, ci); |
10e364da HY |
1323 | if (!usage) |
1324 | free_swap_slot(entry); | |
7c00bafe TC |
1325 | |
1326 | return usage; | |
1327 | } | |
355cfa73 | 1328 | |
7c00bafe TC |
1329 | static void swap_entry_free(struct swap_info_struct *p, swp_entry_t entry) |
1330 | { | |
1331 | struct swap_cluster_info *ci; | |
1332 | unsigned long offset = swp_offset(entry); | |
1333 | unsigned char count; | |
1334 | ||
1335 | ci = lock_cluster(p, offset); | |
1336 | count = p->swap_map[offset]; | |
1337 | VM_BUG_ON(count != SWAP_HAS_CACHE); | |
1338 | p->swap_map[offset] = 0; | |
1339 | dec_cluster_info_page(p, p->cluster_info, offset); | |
235b6217 HY |
1340 | unlock_cluster(ci); |
1341 | ||
38d8b4e6 HY |
1342 | mem_cgroup_uncharge_swap(entry, 1); |
1343 | swap_range_free(p, offset, 1); | |
1da177e4 LT |
1344 | } |
1345 | ||
1346 | /* | |
2de1a7e4 | 1347 | * Caller has made sure that the swap device corresponding to entry |
1da177e4 LT |
1348 | * is still around or has not been recycled. |
1349 | */ | |
1350 | void swap_free(swp_entry_t entry) | |
1351 | { | |
73c34b6a | 1352 | struct swap_info_struct *p; |
1da177e4 | 1353 | |
235b6217 | 1354 | p = _swap_info_get(entry); |
10e364da | 1355 | if (p) |
33e16272 | 1356 | __swap_entry_free(p, entry); |
1da177e4 LT |
1357 | } |
1358 | ||
cb4b86ba KH |
1359 | /* |
1360 | * Called after dropping swapcache to decrease refcnt to swap entries. | |
1361 | */ | |
4081f744 | 1362 | void put_swap_folio(struct folio *folio, swp_entry_t entry) |
38d8b4e6 HY |
1363 | { |
1364 | unsigned long offset = swp_offset(entry); | |
1365 | unsigned long idx = offset / SWAPFILE_CLUSTER; | |
1366 | struct swap_cluster_info *ci; | |
1367 | struct swap_info_struct *si; | |
1368 | unsigned char *map; | |
a3aea839 HY |
1369 | unsigned int i, free_entries = 0; |
1370 | unsigned char val; | |
9faaa0f8 | 1371 | int size = 1 << swap_entry_order(folio_order(folio)); |
fe5266d5 | 1372 | |
a3aea839 | 1373 | si = _swap_info_get(entry); |
38d8b4e6 HY |
1374 | if (!si) |
1375 | return; | |
1376 | ||
c2343d27 | 1377 | ci = lock_cluster_or_swap_info(si, offset); |
a448f2d0 | 1378 | if (size == SWAPFILE_CLUSTER) { |
a448f2d0 HY |
1379 | map = si->swap_map + offset; |
1380 | for (i = 0; i < SWAPFILE_CLUSTER; i++) { | |
1381 | val = map[i]; | |
1382 | VM_BUG_ON(!(val & SWAP_HAS_CACHE)); | |
1383 | if (val == SWAP_HAS_CACHE) | |
1384 | free_entries++; | |
1385 | } | |
a448f2d0 | 1386 | if (free_entries == SWAPFILE_CLUSTER) { |
c2343d27 | 1387 | unlock_cluster_or_swap_info(si, ci); |
a448f2d0 | 1388 | spin_lock(&si->lock); |
a448f2d0 HY |
1389 | mem_cgroup_uncharge_swap(entry, SWAPFILE_CLUSTER); |
1390 | swap_free_cluster(si, idx); | |
1391 | spin_unlock(&si->lock); | |
1392 | return; | |
1393 | } | |
1394 | } | |
c2343d27 HY |
1395 | for (i = 0; i < size; i++, entry.val++) { |
1396 | if (!__swap_entry_free_locked(si, offset + i, SWAP_HAS_CACHE)) { | |
1397 | unlock_cluster_or_swap_info(si, ci); | |
1398 | free_swap_slot(entry); | |
1399 | if (i == size - 1) | |
1400 | return; | |
1401 | lock_cluster_or_swap_info(si, offset); | |
a3aea839 HY |
1402 | } |
1403 | } | |
c2343d27 | 1404 | unlock_cluster_or_swap_info(si, ci); |
38d8b4e6 | 1405 | } |
59807685 | 1406 | |
155b5f88 HY |
1407 | static int swp_entry_cmp(const void *ent1, const void *ent2) |
1408 | { | |
1409 | const swp_entry_t *e1 = ent1, *e2 = ent2; | |
1410 | ||
1411 | return (int)swp_type(*e1) - (int)swp_type(*e2); | |
1412 | } | |
1413 | ||
7c00bafe TC |
1414 | void swapcache_free_entries(swp_entry_t *entries, int n) |
1415 | { | |
1416 | struct swap_info_struct *p, *prev; | |
1417 | int i; | |
1418 | ||
1419 | if (n <= 0) | |
1420 | return; | |
1421 | ||
1422 | prev = NULL; | |
1423 | p = NULL; | |
155b5f88 HY |
1424 | |
1425 | /* | |
1426 | * Sort swap entries by swap device, so each lock is only taken once. | |
1427 | * nr_swapfiles isn't absolutely correct, but the overhead of sort() is | |
1428 | * so low that it isn't necessary to optimize further. | |
1429 | */ | |
1430 | if (nr_swapfiles > 1) | |
1431 | sort(entries, n, sizeof(entries[0]), swp_entry_cmp, NULL); | |
7c00bafe TC |
1432 | for (i = 0; i < n; ++i) { |
1433 | p = swap_info_get_cont(entries[i], prev); | |
1434 | if (p) | |
1435 | swap_entry_free(p, entries[i]); | |
7c00bafe TC |
1436 | prev = p; |
1437 | } | |
235b6217 | 1438 | if (p) |
7c00bafe | 1439 | spin_unlock(&p->lock); |
cb4b86ba KH |
1440 | } |
1441 | ||
eb085574 | 1442 | int __swap_count(swp_entry_t entry) |
aa8d22a1 | 1443 | { |
f9f956b5 | 1444 | struct swap_info_struct *si = swp_swap_info(entry); |
aa8d22a1 MK |
1445 | pgoff_t offset = swp_offset(entry); |
1446 | ||
f9f956b5 | 1447 | return swap_count(si->swap_map[offset]); |
aa8d22a1 MK |
1448 | } |
1449 | ||
14d01ee9 MWO |
1450 | /* |
1451 | * How many references to @entry are currently swapped out? | |
1452 | * This does not give an exact answer when swap count is continued, | |
1453 | * but does include the high COUNT_CONTINUED flag to allow for that. | |
1454 | */ | |
3ecdeb0f | 1455 | int swap_swapcount(struct swap_info_struct *si, swp_entry_t entry) |
322b8afe | 1456 | { |
322b8afe HY |
1457 | pgoff_t offset = swp_offset(entry); |
1458 | struct swap_cluster_info *ci; | |
14d01ee9 | 1459 | int count; |
322b8afe HY |
1460 | |
1461 | ci = lock_cluster_or_swap_info(si, offset); | |
1462 | count = swap_count(si->swap_map[offset]); | |
1463 | unlock_cluster_or_swap_info(si, ci); | |
1464 | return count; | |
1465 | } | |
1466 | ||
8334b962 MK |
1467 | /* |
1468 | * How many references to @entry are currently swapped out? | |
1469 | * This considers COUNT_CONTINUED so it returns exact answer. | |
1470 | */ | |
1471 | int swp_swapcount(swp_entry_t entry) | |
1472 | { | |
1473 | int count, tmp_count, n; | |
1474 | struct swap_info_struct *p; | |
235b6217 | 1475 | struct swap_cluster_info *ci; |
8334b962 MK |
1476 | struct page *page; |
1477 | pgoff_t offset; | |
1478 | unsigned char *map; | |
1479 | ||
235b6217 | 1480 | p = _swap_info_get(entry); |
8334b962 MK |
1481 | if (!p) |
1482 | return 0; | |
1483 | ||
235b6217 HY |
1484 | offset = swp_offset(entry); |
1485 | ||
1486 | ci = lock_cluster_or_swap_info(p, offset); | |
1487 | ||
1488 | count = swap_count(p->swap_map[offset]); | |
8334b962 MK |
1489 | if (!(count & COUNT_CONTINUED)) |
1490 | goto out; | |
1491 | ||
1492 | count &= ~COUNT_CONTINUED; | |
1493 | n = SWAP_MAP_MAX + 1; | |
1494 | ||
8334b962 MK |
1495 | page = vmalloc_to_page(p->swap_map + offset); |
1496 | offset &= ~PAGE_MASK; | |
1497 | VM_BUG_ON(page_private(page) != SWP_CONTINUED); | |
1498 | ||
1499 | do { | |
a8ae4991 | 1500 | page = list_next_entry(page, lru); |
829c3151 | 1501 | map = kmap_local_page(page); |
8334b962 | 1502 | tmp_count = map[offset]; |
829c3151 | 1503 | kunmap_local(map); |
8334b962 MK |
1504 | |
1505 | count += (tmp_count & ~COUNT_CONTINUED) * n; | |
1506 | n *= (SWAP_CONT_MAX + 1); | |
1507 | } while (tmp_count & COUNT_CONTINUED); | |
1508 | out: | |
235b6217 | 1509 | unlock_cluster_or_swap_info(p, ci); |
8334b962 MK |
1510 | return count; |
1511 | } | |
1512 | ||
e0709829 | 1513 | static bool swap_page_trans_huge_swapped(struct swap_info_struct *si, |
d7d0d389 | 1514 | swp_entry_t entry, int order) |
e0709829 HY |
1515 | { |
1516 | struct swap_cluster_info *ci; | |
1517 | unsigned char *map = si->swap_map; | |
d7d0d389 | 1518 | unsigned int nr_pages = 1 << order; |
e0709829 | 1519 | unsigned long roffset = swp_offset(entry); |
d7d0d389 | 1520 | unsigned long offset = round_down(roffset, nr_pages); |
e0709829 HY |
1521 | int i; |
1522 | bool ret = false; | |
1523 | ||
1524 | ci = lock_cluster_or_swap_info(si, offset); | |
d7d0d389 | 1525 | if (!ci || nr_pages == 1) { |
afa4711e | 1526 | if (swap_count(map[roffset])) |
e0709829 HY |
1527 | ret = true; |
1528 | goto unlock_out; | |
1529 | } | |
d7d0d389 | 1530 | for (i = 0; i < nr_pages; i++) { |
afa4711e | 1531 | if (swap_count(map[offset + i])) { |
e0709829 HY |
1532 | ret = true; |
1533 | break; | |
1534 | } | |
1535 | } | |
1536 | unlock_out: | |
1537 | unlock_cluster_or_swap_info(si, ci); | |
1538 | return ret; | |
1539 | } | |
1540 | ||
2397f780 | 1541 | static bool folio_swapped(struct folio *folio) |
e0709829 | 1542 | { |
3d2c9087 | 1543 | swp_entry_t entry = folio->swap; |
14d01ee9 MWO |
1544 | struct swap_info_struct *si = _swap_info_get(entry); |
1545 | ||
1546 | if (!si) | |
1547 | return false; | |
e0709829 | 1548 | |
2397f780 | 1549 | if (!IS_ENABLED(CONFIG_THP_SWAP) || likely(!folio_test_large(folio))) |
14d01ee9 | 1550 | return swap_swapcount(si, entry) != 0; |
e0709829 | 1551 | |
d7d0d389 | 1552 | return swap_page_trans_huge_swapped(si, entry, folio_order(folio)); |
e0709829 | 1553 | } |
ba3c4ce6 | 1554 | |
bdb0ed54 MWO |
1555 | /** |
1556 | * folio_free_swap() - Free the swap space used for this folio. | |
1557 | * @folio: The folio to remove. | |
1558 | * | |
1559 | * If swap is getting full, or if there are no more mappings of this folio, | |
1560 | * then call folio_free_swap to free its swap space. | |
1561 | * | |
1562 | * Return: true if we were able to release the swap space. | |
1da177e4 | 1563 | */ |
bdb0ed54 | 1564 | bool folio_free_swap(struct folio *folio) |
1da177e4 | 1565 | { |
2397f780 | 1566 | VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio); |
1da177e4 | 1567 | |
2397f780 | 1568 | if (!folio_test_swapcache(folio)) |
bdb0ed54 | 1569 | return false; |
2397f780 | 1570 | if (folio_test_writeback(folio)) |
bdb0ed54 | 1571 | return false; |
2397f780 | 1572 | if (folio_swapped(folio)) |
bdb0ed54 | 1573 | return false; |
1da177e4 | 1574 | |
b73d7fce HD |
1575 | /* |
1576 | * Once hibernation has begun to create its image of memory, | |
bdb0ed54 | 1577 | * there's a danger that one of the calls to folio_free_swap() |
b73d7fce HD |
1578 | * - most probably a call from __try_to_reclaim_swap() while |
1579 | * hibernation is allocating its own swap pages for the image, | |
1580 | * but conceivably even a call from memory reclaim - will free | |
bdb0ed54 MWO |
1581 | * the swap from a folio which has already been recorded in the |
1582 | * image as a clean swapcache folio, and then reuse its swap for | |
b73d7fce | 1583 | * another page of the image. On waking from hibernation, the |
bdb0ed54 | 1584 | * original folio might be freed under memory pressure, then |
b73d7fce HD |
1585 | * later read back in from swap, now with the wrong data. |
1586 | * | |
2de1a7e4 | 1587 | * Hibernation suspends storage while it is writing the image |
f90ac398 | 1588 | * to disk so check that here. |
b73d7fce | 1589 | */ |
f90ac398 | 1590 | if (pm_suspended_storage()) |
bdb0ed54 | 1591 | return false; |
b73d7fce | 1592 | |
75fa68a5 | 1593 | delete_from_swap_cache(folio); |
2397f780 | 1594 | folio_set_dirty(folio); |
bdb0ed54 | 1595 | return true; |
68a22394 RR |
1596 | } |
1597 | ||
a62fb92a RR |
1598 | /** |
1599 | * free_swap_and_cache_nr() - Release reference on range of swap entries and | |
1600 | * reclaim their cache if no more references remain. | |
1601 | * @entry: First entry of range. | |
1602 | * @nr: Number of entries in range. | |
1603 | * | |
1604 | * For each swap entry in the contiguous range, release a reference. If any swap | |
1605 | * entries become free, try to reclaim their underlying folios, if present. The | |
1606 | * offset range is defined by [entry.offset, entry.offset + nr). | |
1da177e4 | 1607 | */ |
a62fb92a | 1608 | void free_swap_and_cache_nr(swp_entry_t entry, int nr) |
1da177e4 | 1609 | { |
a62fb92a RR |
1610 | const unsigned long start_offset = swp_offset(entry); |
1611 | const unsigned long end_offset = start_offset + nr; | |
1612 | unsigned int type = swp_type(entry); | |
1613 | struct swap_info_struct *si; | |
1614 | bool any_only_cache = false; | |
1615 | unsigned long offset; | |
7c00bafe | 1616 | unsigned char count; |
1da177e4 | 1617 | |
a7420aa5 | 1618 | if (non_swap_entry(entry)) |
a62fb92a RR |
1619 | return; |
1620 | ||
1621 | si = get_swap_device(entry); | |
1622 | if (!si) | |
1623 | return; | |
1624 | ||
1625 | if (WARN_ON(end_offset > si->max)) | |
1626 | goto out; | |
0697212a | 1627 | |
a62fb92a RR |
1628 | /* |
1629 | * First free all entries in the range. | |
1630 | */ | |
1631 | for (offset = start_offset; offset < end_offset; offset++) { | |
1632 | if (data_race(si->swap_map[offset])) { | |
1633 | count = __swap_entry_free(si, swp_entry(type, offset)); | |
1634 | if (count == SWAP_HAS_CACHE) | |
1635 | any_only_cache = true; | |
1636 | } else { | |
1637 | WARN_ON_ONCE(1); | |
82b1c07a | 1638 | } |
a62fb92a RR |
1639 | } |
1640 | ||
1641 | /* | |
1642 | * Short-circuit the below loop if none of the entries had their | |
1643 | * reference drop to zero. | |
1644 | */ | |
1645 | if (!any_only_cache) | |
1646 | goto out; | |
82b1c07a | 1647 | |
a62fb92a RR |
1648 | /* |
1649 | * Now go back over the range trying to reclaim the swap cache. This is | |
1650 | * more efficient for large folios because we will only try to reclaim | |
1651 | * the swap once per folio in the common case. If we do | |
1652 | * __swap_entry_free() and __try_to_reclaim_swap() in the same loop, the | |
1653 | * latter will get a reference and lock the folio for every individual | |
1654 | * page but will only succeed once the swap slot for every subpage is | |
1655 | * zero. | |
1656 | */ | |
1657 | for (offset = start_offset; offset < end_offset; offset += nr) { | |
1658 | nr = 1; | |
1659 | if (READ_ONCE(si->swap_map[offset]) == SWAP_HAS_CACHE) { | |
1660 | /* | |
1661 | * Folios are always naturally aligned in swap so | |
1662 | * advance forward to the next boundary. Zero means no | |
1663 | * folio was found for the swap entry, so advance by 1 | |
1664 | * in this case. Negative value means folio was found | |
1665 | * but could not be reclaimed. Here we can still advance | |
1666 | * to the next boundary. | |
1667 | */ | |
1668 | nr = __try_to_reclaim_swap(si, offset, | |
bcd49e86 | 1669 | TTRS_UNMAPPED | TTRS_FULL); |
a62fb92a RR |
1670 | if (nr == 0) |
1671 | nr = 1; | |
1672 | else if (nr < 0) | |
1673 | nr = -nr; | |
1674 | nr = ALIGN(offset + 1, nr) - offset; | |
1675 | } | |
1da177e4 | 1676 | } |
a62fb92a RR |
1677 | |
1678 | out: | |
1679 | put_swap_device(si); | |
1da177e4 LT |
1680 | } |
1681 | ||
b0cb1a19 | 1682 | #ifdef CONFIG_HIBERNATION |
bb243f7d ML |
1683 | |
1684 | swp_entry_t get_swap_page_of_type(int type) | |
1685 | { | |
1686 | struct swap_info_struct *si = swap_type_to_swap_info(type); | |
1687 | swp_entry_t entry = {0}; | |
1688 | ||
1689 | if (!si) | |
1690 | goto fail; | |
1691 | ||
1692 | /* This is called for allocating swap entry, not cache */ | |
1693 | spin_lock(&si->lock); | |
845982eb | 1694 | if ((si->flags & SWP_WRITEOK) && scan_swap_map_slots(si, 1, 1, &entry, 0)) |
bb243f7d ML |
1695 | atomic_long_dec(&nr_swap_pages); |
1696 | spin_unlock(&si->lock); | |
1697 | fail: | |
1698 | return entry; | |
1699 | } | |
1700 | ||
f577eb30 | 1701 | /* |
915bae9e | 1702 | * Find the swap type that corresponds to given device (if any). |
f577eb30 | 1703 | * |
915bae9e RW |
1704 | * @offset - number of the PAGE_SIZE-sized block of the device, starting |
1705 | * from 0, in which the swap header is expected to be located. | |
1706 | * | |
1707 | * This is needed for the suspend to disk (aka swsusp). | |
f577eb30 | 1708 | */ |
21bd9005 | 1709 | int swap_type_of(dev_t device, sector_t offset) |
f577eb30 | 1710 | { |
efa90a98 | 1711 | int type; |
f577eb30 | 1712 | |
21bd9005 CH |
1713 | if (!device) |
1714 | return -1; | |
915bae9e | 1715 | |
f577eb30 | 1716 | spin_lock(&swap_lock); |
efa90a98 HD |
1717 | for (type = 0; type < nr_swapfiles; type++) { |
1718 | struct swap_info_struct *sis = swap_info[type]; | |
f577eb30 | 1719 | |
915bae9e | 1720 | if (!(sis->flags & SWP_WRITEOK)) |
f577eb30 | 1721 | continue; |
b6b5bce3 | 1722 | |
21bd9005 | 1723 | if (device == sis->bdev->bd_dev) { |
4efaceb1 | 1724 | struct swap_extent *se = first_se(sis); |
915bae9e | 1725 | |
915bae9e RW |
1726 | if (se->start_block == offset) { |
1727 | spin_unlock(&swap_lock); | |
efa90a98 | 1728 | return type; |
915bae9e | 1729 | } |
f577eb30 RW |
1730 | } |
1731 | } | |
1732 | spin_unlock(&swap_lock); | |
21bd9005 CH |
1733 | return -ENODEV; |
1734 | } | |
915bae9e | 1735 | |
21bd9005 CH |
1736 | int find_first_swap(dev_t *device) |
1737 | { | |
1738 | int type; | |
915bae9e | 1739 | |
21bd9005 CH |
1740 | spin_lock(&swap_lock); |
1741 | for (type = 0; type < nr_swapfiles; type++) { | |
1742 | struct swap_info_struct *sis = swap_info[type]; | |
1743 | ||
1744 | if (!(sis->flags & SWP_WRITEOK)) | |
1745 | continue; | |
1746 | *device = sis->bdev->bd_dev; | |
1747 | spin_unlock(&swap_lock); | |
1748 | return type; | |
1749 | } | |
1750 | spin_unlock(&swap_lock); | |
f577eb30 RW |
1751 | return -ENODEV; |
1752 | } | |
1753 | ||
73c34b6a HD |
1754 | /* |
1755 | * Get the (PAGE_SIZE) block corresponding to given offset on the swapdev | |
1756 | * corresponding to given index in swap_info (swap type). | |
1757 | */ | |
1758 | sector_t swapdev_block(int type, pgoff_t offset) | |
1759 | { | |
c10d38cc | 1760 | struct swap_info_struct *si = swap_type_to_swap_info(type); |
f885056a | 1761 | struct swap_extent *se; |
73c34b6a | 1762 | |
c10d38cc | 1763 | if (!si || !(si->flags & SWP_WRITEOK)) |
73c34b6a | 1764 | return 0; |
f885056a CH |
1765 | se = offset_to_swap_extent(si, offset); |
1766 | return se->start_block + (offset - se->start_page); | |
73c34b6a HD |
1767 | } |
1768 | ||
f577eb30 RW |
1769 | /* |
1770 | * Return either the total number of swap pages of given type, or the number | |
1771 | * of free pages of that type (depending on @free) | |
1772 | * | |
1773 | * This is needed for software suspend | |
1774 | */ | |
1775 | unsigned int count_swap_pages(int type, int free) | |
1776 | { | |
1777 | unsigned int n = 0; | |
1778 | ||
efa90a98 HD |
1779 | spin_lock(&swap_lock); |
1780 | if ((unsigned int)type < nr_swapfiles) { | |
1781 | struct swap_info_struct *sis = swap_info[type]; | |
1782 | ||
ec8acf20 | 1783 | spin_lock(&sis->lock); |
efa90a98 HD |
1784 | if (sis->flags & SWP_WRITEOK) { |
1785 | n = sis->pages; | |
f577eb30 | 1786 | if (free) |
efa90a98 | 1787 | n -= sis->inuse_pages; |
f577eb30 | 1788 | } |
ec8acf20 | 1789 | spin_unlock(&sis->lock); |
f577eb30 | 1790 | } |
efa90a98 | 1791 | spin_unlock(&swap_lock); |
f577eb30 RW |
1792 | return n; |
1793 | } | |
73c34b6a | 1794 | #endif /* CONFIG_HIBERNATION */ |
f577eb30 | 1795 | |
9f8bdb3f | 1796 | static inline int pte_same_as_swp(pte_t pte, pte_t swp_pte) |
179ef71c | 1797 | { |
099dd687 | 1798 | return pte_same(pte_swp_clear_flags(pte), swp_pte); |
179ef71c CG |
1799 | } |
1800 | ||
1da177e4 | 1801 | /* |
72866f6f HD |
1802 | * No need to decide whether this PTE shares the swap entry with others, |
1803 | * just let do_wp_page work it out if a write is requested later - to | |
1804 | * force COW, vm_page_prot omits write permission from any private vma. | |
1da177e4 | 1805 | */ |
044d66c1 | 1806 | static int unuse_pte(struct vm_area_struct *vma, pmd_t *pmd, |
f102cd8b | 1807 | unsigned long addr, swp_entry_t entry, struct folio *folio) |
1da177e4 | 1808 | { |
f00f4843 MWO |
1809 | struct page *page; |
1810 | struct folio *swapcache; | |
044d66c1 | 1811 | spinlock_t *ptl; |
c33c7948 | 1812 | pte_t *pte, new_pte, old_pte; |
f00f4843 | 1813 | bool hwpoisoned = false; |
044d66c1 HD |
1814 | int ret = 1; |
1815 | ||
f00f4843 | 1816 | swapcache = folio; |
96db66d9 MWO |
1817 | folio = ksm_might_need_to_copy(folio, vma, addr); |
1818 | if (unlikely(!folio)) | |
9e16b7fb | 1819 | return -ENOMEM; |
f00f4843 MWO |
1820 | else if (unlikely(folio == ERR_PTR(-EHWPOISON))) { |
1821 | hwpoisoned = true; | |
1822 | folio = swapcache; | |
1823 | } | |
1824 | ||
1825 | page = folio_file_page(folio, swp_offset(entry)); | |
1826 | if (PageHWPoison(page)) | |
f985fc32 | 1827 | hwpoisoned = true; |
9e16b7fb | 1828 | |
044d66c1 | 1829 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
c33c7948 RR |
1830 | if (unlikely(!pte || !pte_same_as_swp(ptep_get(pte), |
1831 | swp_entry_to_pte(entry)))) { | |
044d66c1 HD |
1832 | ret = 0; |
1833 | goto out; | |
1834 | } | |
8a9f3ccd | 1835 | |
c33c7948 RR |
1836 | old_pte = ptep_get(pte); |
1837 | ||
f00f4843 | 1838 | if (unlikely(hwpoisoned || !folio_test_uptodate(folio))) { |
6b970599 | 1839 | swp_entry_t swp_entry; |
9f186f9e ML |
1840 | |
1841 | dec_mm_counter(vma->vm_mm, MM_SWAPENTS); | |
f985fc32 | 1842 | if (hwpoisoned) { |
f00f4843 | 1843 | swp_entry = make_hwpoison_entry(page); |
6b970599 | 1844 | } else { |
af19487f | 1845 | swp_entry = make_poisoned_swp_entry(); |
6b970599 KW |
1846 | } |
1847 | new_pte = swp_entry_to_pte(swp_entry); | |
9f186f9e | 1848 | ret = 0; |
6b970599 | 1849 | goto setpte; |
9f186f9e ML |
1850 | } |
1851 | ||
b53e24c4 PC |
1852 | /* |
1853 | * Some architectures may have to restore extra metadata to the page | |
1854 | * when reading from swap. This metadata may be indexed by swap entry | |
1855 | * so this must be called before swap_free(). | |
1856 | */ | |
f238b8c3 | 1857 | arch_swap_restore(folio_swap(entry, folio), folio); |
b53e24c4 | 1858 | |
b084d435 | 1859 | dec_mm_counter(vma->vm_mm, MM_SWAPENTS); |
d559db08 | 1860 | inc_mm_counter(vma->vm_mm, MM_ANONPAGES); |
f00f4843 MWO |
1861 | folio_get(folio); |
1862 | if (folio == swapcache) { | |
1493a191 DH |
1863 | rmap_t rmap_flags = RMAP_NONE; |
1864 | ||
1865 | /* | |
f00f4843 MWO |
1866 | * See do_swap_page(): writeback would be problematic. |
1867 | * However, we do a folio_wait_writeback() just before this | |
1868 | * call and have the folio locked. | |
1493a191 | 1869 | */ |
f00f4843 | 1870 | VM_BUG_ON_FOLIO(folio_test_writeback(folio), folio); |
c33c7948 | 1871 | if (pte_swp_exclusive(old_pte)) |
1493a191 DH |
1872 | rmap_flags |= RMAP_EXCLUSIVE; |
1873 | ||
da7dc0af | 1874 | folio_add_anon_rmap_pte(folio, page, vma, addr, rmap_flags); |
00501b53 | 1875 | } else { /* ksm created a completely new copy */ |
f00f4843 MWO |
1876 | folio_add_new_anon_rmap(folio, vma, addr); |
1877 | folio_add_lru_vma(folio, vma); | |
00501b53 | 1878 | } |
14a762dd | 1879 | new_pte = pte_mkold(mk_pte(page, vma->vm_page_prot)); |
c33c7948 | 1880 | if (pte_swp_soft_dirty(old_pte)) |
14a762dd | 1881 | new_pte = pte_mksoft_dirty(new_pte); |
c33c7948 | 1882 | if (pte_swp_uffd_wp(old_pte)) |
14a762dd | 1883 | new_pte = pte_mkuffd_wp(new_pte); |
6b970599 | 1884 | setpte: |
14a762dd | 1885 | set_pte_at(vma->vm_mm, addr, pte, new_pte); |
1da177e4 | 1886 | swap_free(entry); |
044d66c1 | 1887 | out: |
d850fa72 HD |
1888 | if (pte) |
1889 | pte_unmap_unlock(pte, ptl); | |
f00f4843 MWO |
1890 | if (folio != swapcache) { |
1891 | folio_unlock(folio); | |
1892 | folio_put(folio); | |
9e16b7fb | 1893 | } |
044d66c1 | 1894 | return ret; |
1da177e4 LT |
1895 | } |
1896 | ||
1897 | static int unuse_pte_range(struct vm_area_struct *vma, pmd_t *pmd, | |
b56a2d8a | 1898 | unsigned long addr, unsigned long end, |
10a9c496 | 1899 | unsigned int type) |
1da177e4 | 1900 | { |
d850fa72 | 1901 | pte_t *pte = NULL; |
b56a2d8a | 1902 | struct swap_info_struct *si; |
1da177e4 | 1903 | |
b56a2d8a | 1904 | si = swap_info[type]; |
1da177e4 | 1905 | do { |
f102cd8b MWO |
1906 | struct folio *folio; |
1907 | unsigned long offset; | |
3f79b187 | 1908 | unsigned char swp_count; |
d850fa72 HD |
1909 | swp_entry_t entry; |
1910 | int ret; | |
c33c7948 | 1911 | pte_t ptent; |
d850fa72 HD |
1912 | |
1913 | if (!pte++) { | |
1914 | pte = pte_offset_map(pmd, addr); | |
1915 | if (!pte) | |
1916 | break; | |
1917 | } | |
f102cd8b | 1918 | |
c33c7948 | 1919 | ptent = ptep_get_lockless(pte); |
f102cd8b | 1920 | |
c33c7948 | 1921 | if (!is_swap_pte(ptent)) |
b56a2d8a VRP |
1922 | continue; |
1923 | ||
c33c7948 | 1924 | entry = pte_to_swp_entry(ptent); |
b56a2d8a VRP |
1925 | if (swp_type(entry) != type) |
1926 | continue; | |
1927 | ||
1928 | offset = swp_offset(entry); | |
b56a2d8a | 1929 | pte_unmap(pte); |
d850fa72 HD |
1930 | pte = NULL; |
1931 | ||
f102cd8b MWO |
1932 | folio = swap_cache_get_folio(entry, vma, addr); |
1933 | if (!folio) { | |
1934 | struct page *page; | |
8c63ca5b WD |
1935 | struct vm_fault vmf = { |
1936 | .vma = vma, | |
1937 | .address = addr, | |
824ddc60 | 1938 | .real_address = addr, |
8c63ca5b WD |
1939 | .pmd = pmd, |
1940 | }; | |
1941 | ||
ebc5951e AR |
1942 | page = swapin_readahead(entry, GFP_HIGHUSER_MOVABLE, |
1943 | &vmf); | |
f102cd8b MWO |
1944 | if (page) |
1945 | folio = page_folio(page); | |
ebc5951e | 1946 | } |
f102cd8b | 1947 | if (!folio) { |
3f79b187 KS |
1948 | swp_count = READ_ONCE(si->swap_map[offset]); |
1949 | if (swp_count == 0 || swp_count == SWAP_MAP_BAD) | |
d850fa72 | 1950 | continue; |
b56a2d8a VRP |
1951 | return -ENOMEM; |
1952 | } | |
1953 | ||
f102cd8b MWO |
1954 | folio_lock(folio); |
1955 | folio_wait_writeback(folio); | |
1956 | ret = unuse_pte(vma, pmd, addr, entry, folio); | |
b56a2d8a | 1957 | if (ret < 0) { |
f102cd8b MWO |
1958 | folio_unlock(folio); |
1959 | folio_put(folio); | |
d850fa72 | 1960 | return ret; |
b56a2d8a VRP |
1961 | } |
1962 | ||
f102cd8b MWO |
1963 | folio_free_swap(folio); |
1964 | folio_unlock(folio); | |
1965 | folio_put(folio); | |
d850fa72 | 1966 | } while (addr += PAGE_SIZE, addr != end); |
b56a2d8a | 1967 | |
d850fa72 HD |
1968 | if (pte) |
1969 | pte_unmap(pte); | |
1970 | return 0; | |
1da177e4 LT |
1971 | } |
1972 | ||
1973 | static inline int unuse_pmd_range(struct vm_area_struct *vma, pud_t *pud, | |
1974 | unsigned long addr, unsigned long end, | |
10a9c496 | 1975 | unsigned int type) |
1da177e4 LT |
1976 | { |
1977 | pmd_t *pmd; | |
1978 | unsigned long next; | |
8a9f3ccd | 1979 | int ret; |
1da177e4 LT |
1980 | |
1981 | pmd = pmd_offset(pud, addr); | |
1982 | do { | |
dc644a07 | 1983 | cond_resched(); |
1da177e4 | 1984 | next = pmd_addr_end(addr, end); |
10a9c496 | 1985 | ret = unuse_pte_range(vma, pmd, addr, next, type); |
8a9f3ccd BS |
1986 | if (ret) |
1987 | return ret; | |
1da177e4 LT |
1988 | } while (pmd++, addr = next, addr != end); |
1989 | return 0; | |
1990 | } | |
1991 | ||
c2febafc | 1992 | static inline int unuse_pud_range(struct vm_area_struct *vma, p4d_t *p4d, |
1da177e4 | 1993 | unsigned long addr, unsigned long end, |
10a9c496 | 1994 | unsigned int type) |
1da177e4 LT |
1995 | { |
1996 | pud_t *pud; | |
1997 | unsigned long next; | |
8a9f3ccd | 1998 | int ret; |
1da177e4 | 1999 | |
c2febafc | 2000 | pud = pud_offset(p4d, addr); |
1da177e4 LT |
2001 | do { |
2002 | next = pud_addr_end(addr, end); | |
2003 | if (pud_none_or_clear_bad(pud)) | |
2004 | continue; | |
10a9c496 | 2005 | ret = unuse_pmd_range(vma, pud, addr, next, type); |
8a9f3ccd BS |
2006 | if (ret) |
2007 | return ret; | |
1da177e4 LT |
2008 | } while (pud++, addr = next, addr != end); |
2009 | return 0; | |
2010 | } | |
2011 | ||
c2febafc KS |
2012 | static inline int unuse_p4d_range(struct vm_area_struct *vma, pgd_t *pgd, |
2013 | unsigned long addr, unsigned long end, | |
10a9c496 | 2014 | unsigned int type) |
c2febafc KS |
2015 | { |
2016 | p4d_t *p4d; | |
2017 | unsigned long next; | |
2018 | int ret; | |
2019 | ||
2020 | p4d = p4d_offset(pgd, addr); | |
2021 | do { | |
2022 | next = p4d_addr_end(addr, end); | |
2023 | if (p4d_none_or_clear_bad(p4d)) | |
2024 | continue; | |
10a9c496 | 2025 | ret = unuse_pud_range(vma, p4d, addr, next, type); |
c2febafc KS |
2026 | if (ret) |
2027 | return ret; | |
2028 | } while (p4d++, addr = next, addr != end); | |
2029 | return 0; | |
2030 | } | |
2031 | ||
10a9c496 | 2032 | static int unuse_vma(struct vm_area_struct *vma, unsigned int type) |
1da177e4 LT |
2033 | { |
2034 | pgd_t *pgd; | |
2035 | unsigned long addr, end, next; | |
8a9f3ccd | 2036 | int ret; |
1da177e4 | 2037 | |
b56a2d8a VRP |
2038 | addr = vma->vm_start; |
2039 | end = vma->vm_end; | |
1da177e4 LT |
2040 | |
2041 | pgd = pgd_offset(vma->vm_mm, addr); | |
2042 | do { | |
2043 | next = pgd_addr_end(addr, end); | |
2044 | if (pgd_none_or_clear_bad(pgd)) | |
2045 | continue; | |
10a9c496 | 2046 | ret = unuse_p4d_range(vma, pgd, addr, next, type); |
8a9f3ccd BS |
2047 | if (ret) |
2048 | return ret; | |
1da177e4 LT |
2049 | } while (pgd++, addr = next, addr != end); |
2050 | return 0; | |
2051 | } | |
2052 | ||
10a9c496 | 2053 | static int unuse_mm(struct mm_struct *mm, unsigned int type) |
1da177e4 LT |
2054 | { |
2055 | struct vm_area_struct *vma; | |
8a9f3ccd | 2056 | int ret = 0; |
208c09db | 2057 | VMA_ITERATOR(vmi, mm, 0); |
1da177e4 | 2058 | |
d8ed45c5 | 2059 | mmap_read_lock(mm); |
208c09db | 2060 | for_each_vma(vmi, vma) { |
b56a2d8a | 2061 | if (vma->anon_vma) { |
10a9c496 | 2062 | ret = unuse_vma(vma, type); |
b56a2d8a VRP |
2063 | if (ret) |
2064 | break; | |
2065 | } | |
208c09db | 2066 | |
dc644a07 | 2067 | cond_resched(); |
1da177e4 | 2068 | } |
d8ed45c5 | 2069 | mmap_read_unlock(mm); |
b56a2d8a | 2070 | return ret; |
1da177e4 LT |
2071 | } |
2072 | ||
2073 | /* | |
3c3115ad ML |
2074 | * Scan swap_map from current position to next entry still in use. |
2075 | * Return 0 if there are no inuse entries after prev till end of | |
2076 | * the map. | |
1da177e4 | 2077 | */ |
6eb396dc | 2078 | static unsigned int find_next_to_unuse(struct swap_info_struct *si, |
10a9c496 | 2079 | unsigned int prev) |
1da177e4 | 2080 | { |
b56a2d8a | 2081 | unsigned int i; |
8d69aaee | 2082 | unsigned char count; |
1da177e4 LT |
2083 | |
2084 | /* | |
5d337b91 | 2085 | * No need for swap_lock here: we're just looking |
1da177e4 LT |
2086 | * for whether an entry is in use, not modifying it; false |
2087 | * hits are okay, and sys_swapoff() has already prevented new | |
5d337b91 | 2088 | * allocations from this area (while holding swap_lock). |
1da177e4 | 2089 | */ |
b56a2d8a | 2090 | for (i = prev + 1; i < si->max; i++) { |
4db0c3c2 | 2091 | count = READ_ONCE(si->swap_map[i]); |
355cfa73 | 2092 | if (count && swap_count(count) != SWAP_MAP_BAD) |
10a9c496 | 2093 | break; |
dc644a07 HD |
2094 | if ((i % LATENCY_LIMIT) == 0) |
2095 | cond_resched(); | |
1da177e4 | 2096 | } |
b56a2d8a VRP |
2097 | |
2098 | if (i == si->max) | |
2099 | i = 0; | |
2100 | ||
1da177e4 LT |
2101 | return i; |
2102 | } | |
2103 | ||
10a9c496 | 2104 | static int try_to_unuse(unsigned int type) |
1da177e4 | 2105 | { |
b56a2d8a VRP |
2106 | struct mm_struct *prev_mm; |
2107 | struct mm_struct *mm; | |
2108 | struct list_head *p; | |
2109 | int retval = 0; | |
efa90a98 | 2110 | struct swap_info_struct *si = swap_info[type]; |
000085b9 | 2111 | struct folio *folio; |
1da177e4 | 2112 | swp_entry_t entry; |
b56a2d8a | 2113 | unsigned int i; |
1da177e4 | 2114 | |
21820948 | 2115 | if (!READ_ONCE(si->inuse_pages)) |
64cf264c | 2116 | goto success; |
1da177e4 | 2117 | |
b56a2d8a | 2118 | retry: |
10a9c496 | 2119 | retval = shmem_unuse(type); |
b56a2d8a | 2120 | if (retval) |
10a9c496 | 2121 | return retval; |
b56a2d8a VRP |
2122 | |
2123 | prev_mm = &init_mm; | |
2124 | mmget(prev_mm); | |
2125 | ||
2126 | spin_lock(&mmlist_lock); | |
2127 | p = &init_mm.mmlist; | |
21820948 | 2128 | while (READ_ONCE(si->inuse_pages) && |
64165b1a HD |
2129 | !signal_pending(current) && |
2130 | (p = p->next) != &init_mm.mmlist) { | |
1da177e4 | 2131 | |
b56a2d8a VRP |
2132 | mm = list_entry(p, struct mm_struct, mmlist); |
2133 | if (!mmget_not_zero(mm)) | |
2134 | continue; | |
2135 | spin_unlock(&mmlist_lock); | |
2136 | mmput(prev_mm); | |
2137 | prev_mm = mm; | |
10a9c496 | 2138 | retval = unuse_mm(mm, type); |
b56a2d8a VRP |
2139 | if (retval) { |
2140 | mmput(prev_mm); | |
10a9c496 | 2141 | return retval; |
1da177e4 LT |
2142 | } |
2143 | ||
2144 | /* | |
b56a2d8a VRP |
2145 | * Make sure that we aren't completely killing |
2146 | * interactive performance. | |
1da177e4 | 2147 | */ |
b56a2d8a VRP |
2148 | cond_resched(); |
2149 | spin_lock(&mmlist_lock); | |
2150 | } | |
2151 | spin_unlock(&mmlist_lock); | |
1da177e4 | 2152 | |
b56a2d8a | 2153 | mmput(prev_mm); |
1da177e4 | 2154 | |
b56a2d8a | 2155 | i = 0; |
21820948 | 2156 | while (READ_ONCE(si->inuse_pages) && |
64165b1a | 2157 | !signal_pending(current) && |
10a9c496 | 2158 | (i = find_next_to_unuse(si, i)) != 0) { |
1da177e4 | 2159 | |
b56a2d8a | 2160 | entry = swp_entry(type, i); |
000085b9 | 2161 | folio = filemap_get_folio(swap_address_space(entry), i); |
66dabbb6 | 2162 | if (IS_ERR(folio)) |
b56a2d8a | 2163 | continue; |
68bdc8d6 HD |
2164 | |
2165 | /* | |
000085b9 MWO |
2166 | * It is conceivable that a racing task removed this folio from |
2167 | * swap cache just before we acquired the page lock. The folio | |
b56a2d8a | 2168 | * might even be back in swap cache on another swap area. But |
000085b9 | 2169 | * that is okay, folio_free_swap() only removes stale folios. |
1da177e4 | 2170 | */ |
000085b9 MWO |
2171 | folio_lock(folio); |
2172 | folio_wait_writeback(folio); | |
2173 | folio_free_swap(folio); | |
2174 | folio_unlock(folio); | |
2175 | folio_put(folio); | |
1da177e4 LT |
2176 | } |
2177 | ||
b56a2d8a VRP |
2178 | /* |
2179 | * Lets check again to see if there are still swap entries in the map. | |
2180 | * If yes, we would need to do retry the unuse logic again. | |
2181 | * Under global memory pressure, swap entries can be reinserted back | |
2182 | * into process space after the mmlist loop above passes over them. | |
dd862deb | 2183 | * |
e2e3fdc7 MWO |
2184 | * Limit the number of retries? No: when mmget_not_zero() |
2185 | * above fails, that mm is likely to be freeing swap from | |
2186 | * exit_mmap(), which proceeds at its own independent pace; | |
2187 | * and even shmem_writepage() could have been preempted after | |
2188 | * folio_alloc_swap(), temporarily hiding that swap. It's easy | |
2189 | * and robust (though cpu-intensive) just to keep retrying. | |
b56a2d8a | 2190 | */ |
21820948 | 2191 | if (READ_ONCE(si->inuse_pages)) { |
64165b1a HD |
2192 | if (!signal_pending(current)) |
2193 | goto retry; | |
10a9c496 | 2194 | return -EINTR; |
64165b1a | 2195 | } |
10a9c496 | 2196 | |
64cf264c YA |
2197 | success: |
2198 | /* | |
2199 | * Make sure that further cleanups after try_to_unuse() returns happen | |
2200 | * after swap_range_free() reduces si->inuse_pages to 0. | |
2201 | */ | |
2202 | smp_mb(); | |
10a9c496 | 2203 | return 0; |
1da177e4 LT |
2204 | } |
2205 | ||
2206 | /* | |
5d337b91 HD |
2207 | * After a successful try_to_unuse, if no swap is now in use, we know |
2208 | * we can empty the mmlist. swap_lock must be held on entry and exit. | |
2209 | * Note that mmlist_lock nests inside swap_lock, and an mm must be | |
1da177e4 LT |
2210 | * added to the mmlist just after page_duplicate - before would be racy. |
2211 | */ | |
2212 | static void drain_mmlist(void) | |
2213 | { | |
2214 | struct list_head *p, *next; | |
efa90a98 | 2215 | unsigned int type; |
1da177e4 | 2216 | |
efa90a98 HD |
2217 | for (type = 0; type < nr_swapfiles; type++) |
2218 | if (swap_info[type]->inuse_pages) | |
1da177e4 LT |
2219 | return; |
2220 | spin_lock(&mmlist_lock); | |
2221 | list_for_each_safe(p, next, &init_mm.mmlist) | |
2222 | list_del_init(p); | |
2223 | spin_unlock(&mmlist_lock); | |
2224 | } | |
2225 | ||
1da177e4 LT |
2226 | /* |
2227 | * Free all of a swapdev's extent information | |
2228 | */ | |
2229 | static void destroy_swap_extents(struct swap_info_struct *sis) | |
2230 | { | |
4efaceb1 AL |
2231 | while (!RB_EMPTY_ROOT(&sis->swap_extent_root)) { |
2232 | struct rb_node *rb = sis->swap_extent_root.rb_node; | |
2233 | struct swap_extent *se = rb_entry(rb, struct swap_extent, rb_node); | |
1da177e4 | 2234 | |
4efaceb1 | 2235 | rb_erase(rb, &sis->swap_extent_root); |
1da177e4 LT |
2236 | kfree(se); |
2237 | } | |
62c230bc | 2238 | |
bc4ae27d | 2239 | if (sis->flags & SWP_ACTIVATED) { |
62c230bc MG |
2240 | struct file *swap_file = sis->swap_file; |
2241 | struct address_space *mapping = swap_file->f_mapping; | |
2242 | ||
bc4ae27d OS |
2243 | sis->flags &= ~SWP_ACTIVATED; |
2244 | if (mapping->a_ops->swap_deactivate) | |
2245 | mapping->a_ops->swap_deactivate(swap_file); | |
62c230bc | 2246 | } |
1da177e4 LT |
2247 | } |
2248 | ||
2249 | /* | |
2250 | * Add a block range (and the corresponding page range) into this swapdev's | |
4efaceb1 | 2251 | * extent tree. |
1da177e4 | 2252 | * |
11d31886 | 2253 | * This function rather assumes that it is called in ascending page order. |
1da177e4 | 2254 | */ |
a509bc1a | 2255 | int |
1da177e4 LT |
2256 | add_swap_extent(struct swap_info_struct *sis, unsigned long start_page, |
2257 | unsigned long nr_pages, sector_t start_block) | |
2258 | { | |
4efaceb1 | 2259 | struct rb_node **link = &sis->swap_extent_root.rb_node, *parent = NULL; |
1da177e4 LT |
2260 | struct swap_extent *se; |
2261 | struct swap_extent *new_se; | |
4efaceb1 AL |
2262 | |
2263 | /* | |
2264 | * place the new node at the right most since the | |
2265 | * function is called in ascending page order. | |
2266 | */ | |
2267 | while (*link) { | |
2268 | parent = *link; | |
2269 | link = &parent->rb_right; | |
2270 | } | |
2271 | ||
2272 | if (parent) { | |
2273 | se = rb_entry(parent, struct swap_extent, rb_node); | |
11d31886 HD |
2274 | BUG_ON(se->start_page + se->nr_pages != start_page); |
2275 | if (se->start_block + se->nr_pages == start_block) { | |
1da177e4 LT |
2276 | /* Merge it */ |
2277 | se->nr_pages += nr_pages; | |
2278 | return 0; | |
2279 | } | |
1da177e4 LT |
2280 | } |
2281 | ||
4efaceb1 | 2282 | /* No merge, insert a new extent. */ |
1da177e4 LT |
2283 | new_se = kmalloc(sizeof(*se), GFP_KERNEL); |
2284 | if (new_se == NULL) | |
2285 | return -ENOMEM; | |
2286 | new_se->start_page = start_page; | |
2287 | new_se->nr_pages = nr_pages; | |
2288 | new_se->start_block = start_block; | |
2289 | ||
4efaceb1 AL |
2290 | rb_link_node(&new_se->rb_node, parent, link); |
2291 | rb_insert_color(&new_se->rb_node, &sis->swap_extent_root); | |
53092a74 | 2292 | return 1; |
1da177e4 | 2293 | } |
aa8aa8a3 | 2294 | EXPORT_SYMBOL_GPL(add_swap_extent); |
1da177e4 LT |
2295 | |
2296 | /* | |
2297 | * A `swap extent' is a simple thing which maps a contiguous range of pages | |
ff351f4b | 2298 | * onto a contiguous range of disk blocks. A rbtree of swap extents is |
c9bdf768 | 2299 | * built at swapon time and is then used at swap_writepage/swap_read_folio |
1da177e4 LT |
2300 | * time for locating where on disk a page belongs. |
2301 | * | |
2302 | * If the swapfile is an S_ISBLK block device, a single extent is installed. | |
2303 | * This is done so that the main operating code can treat S_ISBLK and S_ISREG | |
2304 | * swap files identically. | |
2305 | * | |
2306 | * Whether the swapdev is an S_ISREG file or an S_ISBLK blockdev, the swap | |
ff351f4b | 2307 | * extent rbtree operates in PAGE_SIZE disk blocks. Both S_ISREG and S_ISBLK |
1da177e4 LT |
2308 | * swapfiles are handled *identically* after swapon time. |
2309 | * | |
2310 | * For S_ISREG swapfiles, setup_swap_extents() will walk all the file's blocks | |
ff351f4b ML |
2311 | * and will parse them into a rbtree, in PAGE_SIZE chunks. If some stray |
2312 | * blocks are found which do not fall within the PAGE_SIZE alignment | |
1da177e4 LT |
2313 | * requirements, they are simply tossed out - we will never use those blocks |
2314 | * for swapping. | |
2315 | * | |
1638045c DW |
2316 | * For all swap devices we set S_SWAPFILE across the life of the swapon. This |
2317 | * prevents users from writing to the swap device, which will corrupt memory. | |
1da177e4 LT |
2318 | * |
2319 | * The amount of disk space which a single swap extent represents varies. | |
2320 | * Typically it is in the 1-4 megabyte range. So we can have hundreds of | |
ff351f4b | 2321 | * extents in the rbtree. - akpm. |
1da177e4 | 2322 | */ |
53092a74 | 2323 | static int setup_swap_extents(struct swap_info_struct *sis, sector_t *span) |
1da177e4 | 2324 | { |
62c230bc MG |
2325 | struct file *swap_file = sis->swap_file; |
2326 | struct address_space *mapping = swap_file->f_mapping; | |
2327 | struct inode *inode = mapping->host; | |
1da177e4 LT |
2328 | int ret; |
2329 | ||
1da177e4 LT |
2330 | if (S_ISBLK(inode->i_mode)) { |
2331 | ret = add_swap_extent(sis, 0, sis->max, 0); | |
53092a74 | 2332 | *span = sis->pages; |
a509bc1a | 2333 | return ret; |
1da177e4 LT |
2334 | } |
2335 | ||
62c230bc | 2336 | if (mapping->a_ops->swap_activate) { |
a509bc1a | 2337 | ret = mapping->a_ops->swap_activate(sis, swap_file, span); |
4b60c0ff N |
2338 | if (ret < 0) |
2339 | return ret; | |
2340 | sis->flags |= SWP_ACTIVATED; | |
e1209d3a N |
2341 | if ((sis->flags & SWP_FS_OPS) && |
2342 | sio_pool_init() != 0) { | |
2343 | destroy_swap_extents(sis); | |
2344 | return -ENOMEM; | |
62c230bc | 2345 | } |
a509bc1a | 2346 | return ret; |
62c230bc MG |
2347 | } |
2348 | ||
a509bc1a | 2349 | return generic_swapfile_activate(sis, swap_file, span); |
1da177e4 LT |
2350 | } |
2351 | ||
a2468cc9 AL |
2352 | static int swap_node(struct swap_info_struct *p) |
2353 | { | |
2354 | struct block_device *bdev; | |
2355 | ||
2356 | if (p->bdev) | |
2357 | bdev = p->bdev; | |
2358 | else | |
2359 | bdev = p->swap_file->f_inode->i_sb->s_bdev; | |
2360 | ||
2361 | return bdev ? bdev->bd_disk->node_id : NUMA_NO_NODE; | |
2362 | } | |
2363 | ||
eb085574 HY |
2364 | static void setup_swap_info(struct swap_info_struct *p, int prio, |
2365 | unsigned char *swap_map, | |
2366 | struct swap_cluster_info *cluster_info) | |
40531542 | 2367 | { |
a2468cc9 AL |
2368 | int i; |
2369 | ||
40531542 CEB |
2370 | if (prio >= 0) |
2371 | p->prio = prio; | |
2372 | else | |
2373 | p->prio = --least_priority; | |
18ab4d4c DS |
2374 | /* |
2375 | * the plist prio is negated because plist ordering is | |
2376 | * low-to-high, while swap ordering is high-to-low | |
2377 | */ | |
2378 | p->list.prio = -p->prio; | |
a2468cc9 AL |
2379 | for_each_node(i) { |
2380 | if (p->prio >= 0) | |
2381 | p->avail_lists[i].prio = -p->prio; | |
2382 | else { | |
2383 | if (swap_node(p) == i) | |
2384 | p->avail_lists[i].prio = 1; | |
2385 | else | |
2386 | p->avail_lists[i].prio = -p->prio; | |
2387 | } | |
2388 | } | |
40531542 | 2389 | p->swap_map = swap_map; |
2a8f9449 | 2390 | p->cluster_info = cluster_info; |
eb085574 HY |
2391 | } |
2392 | ||
2393 | static void _enable_swap_info(struct swap_info_struct *p) | |
2394 | { | |
63d8620e | 2395 | p->flags |= SWP_WRITEOK; |
ec8acf20 | 2396 | atomic_long_add(p->pages, &nr_swap_pages); |
40531542 CEB |
2397 | total_swap_pages += p->pages; |
2398 | ||
adfab836 | 2399 | assert_spin_locked(&swap_lock); |
adfab836 | 2400 | /* |
18ab4d4c DS |
2401 | * both lists are plists, and thus priority ordered. |
2402 | * swap_active_head needs to be priority ordered for swapoff(), | |
2403 | * which on removal of any swap_info_struct with an auto-assigned | |
2404 | * (i.e. negative) priority increments the auto-assigned priority | |
2405 | * of any lower-priority swap_info_structs. | |
e2e3fdc7 | 2406 | * swap_avail_head needs to be priority ordered for folio_alloc_swap(), |
18ab4d4c DS |
2407 | * which allocates swap pages from the highest available priority |
2408 | * swap_info_struct. | |
adfab836 | 2409 | */ |
18ab4d4c | 2410 | plist_add(&p->list, &swap_active_head); |
c70699e5 MW |
2411 | |
2412 | /* add to available list iff swap device is not full */ | |
2413 | if (p->highest_bit) | |
2414 | add_to_avail_list(p); | |
cf0cac0a CEB |
2415 | } |
2416 | ||
2417 | static void enable_swap_info(struct swap_info_struct *p, int prio, | |
2418 | unsigned char *swap_map, | |
42c06a0e | 2419 | struct swap_cluster_info *cluster_info) |
cf0cac0a CEB |
2420 | { |
2421 | spin_lock(&swap_lock); | |
ec8acf20 | 2422 | spin_lock(&p->lock); |
eb085574 HY |
2423 | setup_swap_info(p, prio, swap_map, cluster_info); |
2424 | spin_unlock(&p->lock); | |
2425 | spin_unlock(&swap_lock); | |
2426 | /* | |
63d8620e | 2427 | * Finished initializing swap device, now it's safe to reference it. |
eb085574 | 2428 | */ |
63d8620e | 2429 | percpu_ref_resurrect(&p->users); |
eb085574 HY |
2430 | spin_lock(&swap_lock); |
2431 | spin_lock(&p->lock); | |
2432 | _enable_swap_info(p); | |
ec8acf20 | 2433 | spin_unlock(&p->lock); |
cf0cac0a CEB |
2434 | spin_unlock(&swap_lock); |
2435 | } | |
2436 | ||
2437 | static void reinsert_swap_info(struct swap_info_struct *p) | |
2438 | { | |
2439 | spin_lock(&swap_lock); | |
ec8acf20 | 2440 | spin_lock(&p->lock); |
eb085574 HY |
2441 | setup_swap_info(p, p->prio, p->swap_map, p->cluster_info); |
2442 | _enable_swap_info(p); | |
ec8acf20 | 2443 | spin_unlock(&p->lock); |
40531542 CEB |
2444 | spin_unlock(&swap_lock); |
2445 | } | |
2446 | ||
67afa38e TC |
2447 | bool has_usable_swap(void) |
2448 | { | |
2449 | bool ret = true; | |
2450 | ||
2451 | spin_lock(&swap_lock); | |
2452 | if (plist_head_empty(&swap_active_head)) | |
2453 | ret = false; | |
2454 | spin_unlock(&swap_lock); | |
2455 | return ret; | |
2456 | } | |
2457 | ||
c4ea37c2 | 2458 | SYSCALL_DEFINE1(swapoff, const char __user *, specialfile) |
1da177e4 | 2459 | { |
73c34b6a | 2460 | struct swap_info_struct *p = NULL; |
8d69aaee | 2461 | unsigned char *swap_map; |
2a8f9449 | 2462 | struct swap_cluster_info *cluster_info; |
1da177e4 LT |
2463 | struct file *swap_file, *victim; |
2464 | struct address_space *mapping; | |
2465 | struct inode *inode; | |
91a27b2a | 2466 | struct filename *pathname; |
adfab836 | 2467 | int err, found = 0; |
5b808a23 | 2468 | unsigned int old_block_size; |
886bb7e9 | 2469 | |
1da177e4 LT |
2470 | if (!capable(CAP_SYS_ADMIN)) |
2471 | return -EPERM; | |
2472 | ||
191c5424 AV |
2473 | BUG_ON(!current->mm); |
2474 | ||
1da177e4 | 2475 | pathname = getname(specialfile); |
1da177e4 | 2476 | if (IS_ERR(pathname)) |
f58b59c1 | 2477 | return PTR_ERR(pathname); |
1da177e4 | 2478 | |
669abf4e | 2479 | victim = file_open_name(pathname, O_RDWR|O_LARGEFILE, 0); |
1da177e4 LT |
2480 | err = PTR_ERR(victim); |
2481 | if (IS_ERR(victim)) | |
2482 | goto out; | |
2483 | ||
2484 | mapping = victim->f_mapping; | |
5d337b91 | 2485 | spin_lock(&swap_lock); |
18ab4d4c | 2486 | plist_for_each_entry(p, &swap_active_head, list) { |
22c6f8fd | 2487 | if (p->flags & SWP_WRITEOK) { |
adfab836 DS |
2488 | if (p->swap_file->f_mapping == mapping) { |
2489 | found = 1; | |
1da177e4 | 2490 | break; |
adfab836 | 2491 | } |
1da177e4 | 2492 | } |
1da177e4 | 2493 | } |
adfab836 | 2494 | if (!found) { |
1da177e4 | 2495 | err = -EINVAL; |
5d337b91 | 2496 | spin_unlock(&swap_lock); |
1da177e4 LT |
2497 | goto out_dput; |
2498 | } | |
191c5424 | 2499 | if (!security_vm_enough_memory_mm(current->mm, p->pages)) |
1da177e4 LT |
2500 | vm_unacct_memory(p->pages); |
2501 | else { | |
2502 | err = -ENOMEM; | |
5d337b91 | 2503 | spin_unlock(&swap_lock); |
1da177e4 LT |
2504 | goto out_dput; |
2505 | } | |
ec8acf20 | 2506 | spin_lock(&p->lock); |
6fe7d6b9 | 2507 | del_from_avail_list(p); |
78ecba08 | 2508 | if (p->prio < 0) { |
adfab836 | 2509 | struct swap_info_struct *si = p; |
a2468cc9 | 2510 | int nid; |
adfab836 | 2511 | |
18ab4d4c | 2512 | plist_for_each_entry_continue(si, &swap_active_head, list) { |
adfab836 | 2513 | si->prio++; |
18ab4d4c | 2514 | si->list.prio--; |
a2468cc9 AL |
2515 | for_each_node(nid) { |
2516 | if (si->avail_lists[nid].prio != 1) | |
2517 | si->avail_lists[nid].prio--; | |
2518 | } | |
adfab836 | 2519 | } |
78ecba08 HD |
2520 | least_priority++; |
2521 | } | |
18ab4d4c | 2522 | plist_del(&p->list, &swap_active_head); |
ec8acf20 | 2523 | atomic_long_sub(p->pages, &nr_swap_pages); |
1da177e4 LT |
2524 | total_swap_pages -= p->pages; |
2525 | p->flags &= ~SWP_WRITEOK; | |
ec8acf20 | 2526 | spin_unlock(&p->lock); |
5d337b91 | 2527 | spin_unlock(&swap_lock); |
fb4f88dc | 2528 | |
039939a6 TC |
2529 | disable_swap_slots_cache_lock(); |
2530 | ||
e1e12d2f | 2531 | set_current_oom_origin(); |
10a9c496 | 2532 | err = try_to_unuse(p->type); |
e1e12d2f | 2533 | clear_current_oom_origin(); |
1da177e4 | 2534 | |
1da177e4 LT |
2535 | if (err) { |
2536 | /* re-insert swap space back into swap_list */ | |
cf0cac0a | 2537 | reinsert_swap_info(p); |
039939a6 | 2538 | reenable_swap_slots_cache_unlock(); |
1da177e4 LT |
2539 | goto out_dput; |
2540 | } | |
52b7efdb | 2541 | |
039939a6 TC |
2542 | reenable_swap_slots_cache_unlock(); |
2543 | ||
eb085574 | 2544 | /* |
63d8620e | 2545 | * Wait for swap operations protected by get/put_swap_device() |
d4a34d7f HY |
2546 | * to complete. Because of synchronize_rcu() here, all swap |
2547 | * operations protected by RCU reader side lock (including any | |
2548 | * spinlock) will be waited too. This makes it easy to | |
2549 | * prevent folio_test_swapcache() and the following swap cache | |
2550 | * operations from racing with swapoff. | |
eb085574 | 2551 | */ |
63d8620e | 2552 | percpu_ref_kill(&p->users); |
eb085574 | 2553 | synchronize_rcu(); |
63d8620e | 2554 | wait_for_completion(&p->comp); |
eb085574 | 2555 | |
815c2c54 SL |
2556 | flush_work(&p->discard_work); |
2557 | ||
5d337b91 | 2558 | destroy_swap_extents(p); |
570a335b HD |
2559 | if (p->flags & SWP_CONTINUED) |
2560 | free_swap_count_continuations(p); | |
2561 | ||
10f0d2a5 | 2562 | if (!p->bdev || !bdev_nonrot(p->bdev)) |
81a0298b HY |
2563 | atomic_dec(&nr_rotate_swap); |
2564 | ||
fc0abb14 | 2565 | mutex_lock(&swapon_mutex); |
5d337b91 | 2566 | spin_lock(&swap_lock); |
ec8acf20 | 2567 | spin_lock(&p->lock); |
5d337b91 HD |
2568 | drain_mmlist(); |
2569 | ||
bb243f7d | 2570 | /* wait for anyone still in scan_swap_map_slots */ |
52b7efdb HD |
2571 | p->highest_bit = 0; /* cuts scans short */ |
2572 | while (p->flags >= SWP_SCANNING) { | |
ec8acf20 | 2573 | spin_unlock(&p->lock); |
5d337b91 | 2574 | spin_unlock(&swap_lock); |
13e4b57f | 2575 | schedule_timeout_uninterruptible(1); |
5d337b91 | 2576 | spin_lock(&swap_lock); |
ec8acf20 | 2577 | spin_lock(&p->lock); |
52b7efdb | 2578 | } |
52b7efdb | 2579 | |
1da177e4 | 2580 | swap_file = p->swap_file; |
5b808a23 | 2581 | old_block_size = p->old_block_size; |
1da177e4 LT |
2582 | p->swap_file = NULL; |
2583 | p->max = 0; | |
2584 | swap_map = p->swap_map; | |
2585 | p->swap_map = NULL; | |
2a8f9449 SL |
2586 | cluster_info = p->cluster_info; |
2587 | p->cluster_info = NULL; | |
ec8acf20 | 2588 | spin_unlock(&p->lock); |
5d337b91 | 2589 | spin_unlock(&swap_lock); |
8a84802e | 2590 | arch_swap_invalidate_area(p->type); |
42c06a0e | 2591 | zswap_swapoff(p->type); |
fc0abb14 | 2592 | mutex_unlock(&swapon_mutex); |
ebc2a1a6 SL |
2593 | free_percpu(p->percpu_cluster); |
2594 | p->percpu_cluster = NULL; | |
49070588 HY |
2595 | free_percpu(p->cluster_next_cpu); |
2596 | p->cluster_next_cpu = NULL; | |
1da177e4 | 2597 | vfree(swap_map); |
54f180d3 | 2598 | kvfree(cluster_info); |
2de1a7e4 | 2599 | /* Destroy swap account information */ |
adfab836 | 2600 | swap_cgroup_swapoff(p->type); |
4b3ef9da | 2601 | exit_swap_address_space(p->type); |
27a7faa0 | 2602 | |
1da177e4 | 2603 | inode = mapping->host; |
16ca5dfd | 2604 | if (p->bdev_file) { |
4c6bca43 | 2605 | set_blocksize(p->bdev, old_block_size); |
16ca5dfd CB |
2606 | fput(p->bdev_file); |
2607 | p->bdev_file = NULL; | |
1da177e4 | 2608 | } |
1638045c DW |
2609 | |
2610 | inode_lock(inode); | |
2611 | inode->i_flags &= ~S_SWAPFILE; | |
2612 | inode_unlock(inode); | |
1da177e4 | 2613 | filp_close(swap_file, NULL); |
f893ab41 WY |
2614 | |
2615 | /* | |
2616 | * Clear the SWP_USED flag after all resources are freed so that swapon | |
2617 | * can reuse this swap_info in alloc_swap_info() safely. It is ok to | |
2618 | * not hold p->lock after we cleared its SWP_WRITEOK. | |
2619 | */ | |
2620 | spin_lock(&swap_lock); | |
2621 | p->flags = 0; | |
2622 | spin_unlock(&swap_lock); | |
2623 | ||
1da177e4 | 2624 | err = 0; |
66d7dd51 KS |
2625 | atomic_inc(&proc_poll_event); |
2626 | wake_up_interruptible(&proc_poll_wait); | |
1da177e4 LT |
2627 | |
2628 | out_dput: | |
2629 | filp_close(victim, NULL); | |
2630 | out: | |
f58b59c1 | 2631 | putname(pathname); |
1da177e4 LT |
2632 | return err; |
2633 | } | |
2634 | ||
2635 | #ifdef CONFIG_PROC_FS | |
9dd95748 | 2636 | static __poll_t swaps_poll(struct file *file, poll_table *wait) |
66d7dd51 | 2637 | { |
f1514638 | 2638 | struct seq_file *seq = file->private_data; |
66d7dd51 KS |
2639 | |
2640 | poll_wait(file, &proc_poll_wait, wait); | |
2641 | ||
f1514638 KS |
2642 | if (seq->poll_event != atomic_read(&proc_poll_event)) { |
2643 | seq->poll_event = atomic_read(&proc_poll_event); | |
a9a08845 | 2644 | return EPOLLIN | EPOLLRDNORM | EPOLLERR | EPOLLPRI; |
66d7dd51 KS |
2645 | } |
2646 | ||
a9a08845 | 2647 | return EPOLLIN | EPOLLRDNORM; |
66d7dd51 KS |
2648 | } |
2649 | ||
1da177e4 LT |
2650 | /* iterator */ |
2651 | static void *swap_start(struct seq_file *swap, loff_t *pos) | |
2652 | { | |
efa90a98 HD |
2653 | struct swap_info_struct *si; |
2654 | int type; | |
1da177e4 LT |
2655 | loff_t l = *pos; |
2656 | ||
fc0abb14 | 2657 | mutex_lock(&swapon_mutex); |
1da177e4 | 2658 | |
881e4aab SS |
2659 | if (!l) |
2660 | return SEQ_START_TOKEN; | |
2661 | ||
c10d38cc | 2662 | for (type = 0; (si = swap_type_to_swap_info(type)); type++) { |
efa90a98 | 2663 | if (!(si->flags & SWP_USED) || !si->swap_map) |
1da177e4 | 2664 | continue; |
881e4aab | 2665 | if (!--l) |
efa90a98 | 2666 | return si; |
1da177e4 LT |
2667 | } |
2668 | ||
2669 | return NULL; | |
2670 | } | |
2671 | ||
2672 | static void *swap_next(struct seq_file *swap, void *v, loff_t *pos) | |
2673 | { | |
efa90a98 HD |
2674 | struct swap_info_struct *si = v; |
2675 | int type; | |
1da177e4 | 2676 | |
881e4aab | 2677 | if (v == SEQ_START_TOKEN) |
efa90a98 HD |
2678 | type = 0; |
2679 | else | |
2680 | type = si->type + 1; | |
881e4aab | 2681 | |
10c8d69f | 2682 | ++(*pos); |
c10d38cc | 2683 | for (; (si = swap_type_to_swap_info(type)); type++) { |
efa90a98 | 2684 | if (!(si->flags & SWP_USED) || !si->swap_map) |
1da177e4 | 2685 | continue; |
efa90a98 | 2686 | return si; |
1da177e4 LT |
2687 | } |
2688 | ||
2689 | return NULL; | |
2690 | } | |
2691 | ||
2692 | static void swap_stop(struct seq_file *swap, void *v) | |
2693 | { | |
fc0abb14 | 2694 | mutex_unlock(&swapon_mutex); |
1da177e4 LT |
2695 | } |
2696 | ||
2697 | static int swap_show(struct seq_file *swap, void *v) | |
2698 | { | |
efa90a98 | 2699 | struct swap_info_struct *si = v; |
1da177e4 LT |
2700 | struct file *file; |
2701 | int len; | |
642929a2 | 2702 | unsigned long bytes, inuse; |
1da177e4 | 2703 | |
efa90a98 | 2704 | if (si == SEQ_START_TOKEN) { |
68d68ff6 | 2705 | seq_puts(swap, "Filename\t\t\t\tType\t\tSize\t\tUsed\t\tPriority\n"); |
881e4aab SS |
2706 | return 0; |
2707 | } | |
1da177e4 | 2708 | |
00cde042 Z |
2709 | bytes = K(si->pages); |
2710 | inuse = K(READ_ONCE(si->inuse_pages)); | |
6f793940 | 2711 | |
efa90a98 | 2712 | file = si->swap_file; |
2726d566 | 2713 | len = seq_file_path(swap, file, " \t\n\\"); |
642929a2 | 2714 | seq_printf(swap, "%*s%s\t%lu\t%s%lu\t%s%d\n", |
886bb7e9 | 2715 | len < 40 ? 40 - len : 1, " ", |
496ad9aa | 2716 | S_ISBLK(file_inode(file)->i_mode) ? |
1da177e4 | 2717 | "partition" : "file\t", |
6f793940 RD |
2718 | bytes, bytes < 10000000 ? "\t" : "", |
2719 | inuse, inuse < 10000000 ? "\t" : "", | |
efa90a98 | 2720 | si->prio); |
1da177e4 LT |
2721 | return 0; |
2722 | } | |
2723 | ||
15ad7cdc | 2724 | static const struct seq_operations swaps_op = { |
1da177e4 LT |
2725 | .start = swap_start, |
2726 | .next = swap_next, | |
2727 | .stop = swap_stop, | |
2728 | .show = swap_show | |
2729 | }; | |
2730 | ||
2731 | static int swaps_open(struct inode *inode, struct file *file) | |
2732 | { | |
f1514638 | 2733 | struct seq_file *seq; |
66d7dd51 KS |
2734 | int ret; |
2735 | ||
66d7dd51 | 2736 | ret = seq_open(file, &swaps_op); |
f1514638 | 2737 | if (ret) |
66d7dd51 | 2738 | return ret; |
66d7dd51 | 2739 | |
f1514638 KS |
2740 | seq = file->private_data; |
2741 | seq->poll_event = atomic_read(&proc_poll_event); | |
2742 | return 0; | |
1da177e4 LT |
2743 | } |
2744 | ||
97a32539 | 2745 | static const struct proc_ops swaps_proc_ops = { |
d919b33d | 2746 | .proc_flags = PROC_ENTRY_PERMANENT, |
97a32539 AD |
2747 | .proc_open = swaps_open, |
2748 | .proc_read = seq_read, | |
2749 | .proc_lseek = seq_lseek, | |
2750 | .proc_release = seq_release, | |
2751 | .proc_poll = swaps_poll, | |
1da177e4 LT |
2752 | }; |
2753 | ||
2754 | static int __init procswaps_init(void) | |
2755 | { | |
97a32539 | 2756 | proc_create("swaps", 0, NULL, &swaps_proc_ops); |
1da177e4 LT |
2757 | return 0; |
2758 | } | |
2759 | __initcall(procswaps_init); | |
2760 | #endif /* CONFIG_PROC_FS */ | |
2761 | ||
1796316a JB |
2762 | #ifdef MAX_SWAPFILES_CHECK |
2763 | static int __init max_swapfiles_check(void) | |
2764 | { | |
2765 | MAX_SWAPFILES_CHECK(); | |
2766 | return 0; | |
2767 | } | |
2768 | late_initcall(max_swapfiles_check); | |
2769 | #endif | |
2770 | ||
53cbb243 | 2771 | static struct swap_info_struct *alloc_swap_info(void) |
1da177e4 | 2772 | { |
73c34b6a | 2773 | struct swap_info_struct *p; |
b11a76b3 | 2774 | struct swap_info_struct *defer = NULL; |
1da177e4 | 2775 | unsigned int type; |
a2468cc9 | 2776 | int i; |
efa90a98 | 2777 | |
96008744 | 2778 | p = kvzalloc(struct_size(p, avail_lists, nr_node_ids), GFP_KERNEL); |
efa90a98 | 2779 | if (!p) |
53cbb243 | 2780 | return ERR_PTR(-ENOMEM); |
efa90a98 | 2781 | |
63d8620e ML |
2782 | if (percpu_ref_init(&p->users, swap_users_ref_free, |
2783 | PERCPU_REF_INIT_DEAD, GFP_KERNEL)) { | |
2784 | kvfree(p); | |
2785 | return ERR_PTR(-ENOMEM); | |
2786 | } | |
2787 | ||
5d337b91 | 2788 | spin_lock(&swap_lock); |
efa90a98 HD |
2789 | for (type = 0; type < nr_swapfiles; type++) { |
2790 | if (!(swap_info[type]->flags & SWP_USED)) | |
1da177e4 | 2791 | break; |
efa90a98 | 2792 | } |
0697212a | 2793 | if (type >= MAX_SWAPFILES) { |
5d337b91 | 2794 | spin_unlock(&swap_lock); |
63d8620e | 2795 | percpu_ref_exit(&p->users); |
873d7bcf | 2796 | kvfree(p); |
730c0581 | 2797 | return ERR_PTR(-EPERM); |
1da177e4 | 2798 | } |
efa90a98 HD |
2799 | if (type >= nr_swapfiles) { |
2800 | p->type = type; | |
efa90a98 | 2801 | /* |
a4b45114 HY |
2802 | * Publish the swap_info_struct after initializing it. |
2803 | * Note that kvzalloc() above zeroes all its fields. | |
efa90a98 | 2804 | */ |
a4b45114 HY |
2805 | smp_store_release(&swap_info[type], p); /* rcu_assign_pointer() */ |
2806 | nr_swapfiles++; | |
efa90a98 | 2807 | } else { |
b11a76b3 | 2808 | defer = p; |
efa90a98 HD |
2809 | p = swap_info[type]; |
2810 | /* | |
2811 | * Do not memset this entry: a racing procfs swap_next() | |
2812 | * would be relying on p->type to remain valid. | |
2813 | */ | |
2814 | } | |
4efaceb1 | 2815 | p->swap_extent_root = RB_ROOT; |
18ab4d4c | 2816 | plist_node_init(&p->list, 0); |
a2468cc9 AL |
2817 | for_each_node(i) |
2818 | plist_node_init(&p->avail_lists[i], 0); | |
1da177e4 | 2819 | p->flags = SWP_USED; |
5d337b91 | 2820 | spin_unlock(&swap_lock); |
63d8620e ML |
2821 | if (defer) { |
2822 | percpu_ref_exit(&defer->users); | |
2823 | kvfree(defer); | |
2824 | } | |
ec8acf20 | 2825 | spin_lock_init(&p->lock); |
2628bd6f | 2826 | spin_lock_init(&p->cont_lock); |
63d8620e | 2827 | init_completion(&p->comp); |
efa90a98 | 2828 | |
53cbb243 | 2829 | return p; |
53cbb243 CEB |
2830 | } |
2831 | ||
4d0e1e10 CEB |
2832 | static int claim_swapfile(struct swap_info_struct *p, struct inode *inode) |
2833 | { | |
2834 | int error; | |
2835 | ||
2836 | if (S_ISBLK(inode->i_mode)) { | |
16ca5dfd | 2837 | p->bdev_file = bdev_file_open_by_dev(inode->i_rdev, |
05bdb996 | 2838 | BLK_OPEN_READ | BLK_OPEN_WRITE, p, NULL); |
16ca5dfd CB |
2839 | if (IS_ERR(p->bdev_file)) { |
2840 | error = PTR_ERR(p->bdev_file); | |
2841 | p->bdev_file = NULL; | |
6f179af8 | 2842 | return error; |
4d0e1e10 | 2843 | } |
16ca5dfd | 2844 | p->bdev = file_bdev(p->bdev_file); |
4d0e1e10 CEB |
2845 | p->old_block_size = block_size(p->bdev); |
2846 | error = set_blocksize(p->bdev, PAGE_SIZE); | |
2847 | if (error < 0) | |
87ade72a | 2848 | return error; |
12d2966d NA |
2849 | /* |
2850 | * Zoned block devices contain zones that have a sequential | |
2851 | * write only restriction. Hence zoned block devices are not | |
2852 | * suitable for swapping. Disallow them here. | |
2853 | */ | |
9964e674 | 2854 | if (bdev_is_zoned(p->bdev)) |
12d2966d | 2855 | return -EINVAL; |
4d0e1e10 CEB |
2856 | p->flags |= SWP_BLKDEV; |
2857 | } else if (S_ISREG(inode->i_mode)) { | |
2858 | p->bdev = inode->i_sb->s_bdev; | |
1638045c DW |
2859 | } |
2860 | ||
4d0e1e10 | 2861 | return 0; |
4d0e1e10 CEB |
2862 | } |
2863 | ||
377eeaa8 AK |
2864 | |
2865 | /* | |
2866 | * Find out how many pages are allowed for a single swap device. There | |
2867 | * are two limiting factors: | |
2868 | * 1) the number of bits for the swap offset in the swp_entry_t type, and | |
2869 | * 2) the number of bits in the swap pte, as defined by the different | |
2870 | * architectures. | |
2871 | * | |
2872 | * In order to find the largest possible bit mask, a swap entry with | |
2873 | * swap type 0 and swap offset ~0UL is created, encoded to a swap pte, | |
2874 | * decoded to a swp_entry_t again, and finally the swap offset is | |
2875 | * extracted. | |
2876 | * | |
2877 | * This will mask all the bits from the initial ~0UL mask that can't | |
2878 | * be encoded in either the swp_entry_t or the architecture definition | |
2879 | * of a swap pte. | |
2880 | */ | |
2881 | unsigned long generic_max_swapfile_size(void) | |
2882 | { | |
2883 | return swp_offset(pte_to_swp_entry( | |
2884 | swp_entry_to_pte(swp_entry(0, ~0UL)))) + 1; | |
2885 | } | |
2886 | ||
2887 | /* Can be overridden by an architecture for additional checks. */ | |
be45a490 | 2888 | __weak unsigned long arch_max_swapfile_size(void) |
377eeaa8 AK |
2889 | { |
2890 | return generic_max_swapfile_size(); | |
2891 | } | |
2892 | ||
ca8bd38b CEB |
2893 | static unsigned long read_swap_header(struct swap_info_struct *p, |
2894 | union swap_header *swap_header, | |
2895 | struct inode *inode) | |
2896 | { | |
2897 | int i; | |
2898 | unsigned long maxpages; | |
2899 | unsigned long swapfilepages; | |
d6bbbd29 | 2900 | unsigned long last_page; |
ca8bd38b CEB |
2901 | |
2902 | if (memcmp("SWAPSPACE2", swap_header->magic.magic, 10)) { | |
465c47fd | 2903 | pr_err("Unable to find swap-space signature\n"); |
38719025 | 2904 | return 0; |
ca8bd38b CEB |
2905 | } |
2906 | ||
041711ce | 2907 | /* swap partition endianness hack... */ |
ca8bd38b CEB |
2908 | if (swab32(swap_header->info.version) == 1) { |
2909 | swab32s(&swap_header->info.version); | |
2910 | swab32s(&swap_header->info.last_page); | |
2911 | swab32s(&swap_header->info.nr_badpages); | |
dd111be6 JH |
2912 | if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES) |
2913 | return 0; | |
ca8bd38b CEB |
2914 | for (i = 0; i < swap_header->info.nr_badpages; i++) |
2915 | swab32s(&swap_header->info.badpages[i]); | |
2916 | } | |
2917 | /* Check the swap header's sub-version */ | |
2918 | if (swap_header->info.version != 1) { | |
465c47fd AM |
2919 | pr_warn("Unable to handle swap header version %d\n", |
2920 | swap_header->info.version); | |
38719025 | 2921 | return 0; |
ca8bd38b CEB |
2922 | } |
2923 | ||
2924 | p->lowest_bit = 1; | |
2925 | p->cluster_next = 1; | |
2926 | p->cluster_nr = 0; | |
2927 | ||
be45a490 | 2928 | maxpages = swapfile_maximum_size; |
d6bbbd29 | 2929 | last_page = swap_header->info.last_page; |
a06ad633 TA |
2930 | if (!last_page) { |
2931 | pr_warn("Empty swap-file\n"); | |
2932 | return 0; | |
2933 | } | |
d6bbbd29 | 2934 | if (last_page > maxpages) { |
465c47fd | 2935 | pr_warn("Truncating oversized swap area, only using %luk out of %luk\n", |
00cde042 | 2936 | K(maxpages), K(last_page)); |
d6bbbd29 RJ |
2937 | } |
2938 | if (maxpages > last_page) { | |
2939 | maxpages = last_page + 1; | |
ca8bd38b CEB |
2940 | /* p->max is an unsigned int: don't overflow it */ |
2941 | if ((unsigned int)maxpages == 0) | |
2942 | maxpages = UINT_MAX; | |
2943 | } | |
2944 | p->highest_bit = maxpages - 1; | |
2945 | ||
2946 | if (!maxpages) | |
38719025 | 2947 | return 0; |
ca8bd38b CEB |
2948 | swapfilepages = i_size_read(inode) >> PAGE_SHIFT; |
2949 | if (swapfilepages && maxpages > swapfilepages) { | |
465c47fd | 2950 | pr_warn("Swap area shorter than signature indicates\n"); |
38719025 | 2951 | return 0; |
ca8bd38b CEB |
2952 | } |
2953 | if (swap_header->info.nr_badpages && S_ISREG(inode->i_mode)) | |
38719025 | 2954 | return 0; |
ca8bd38b | 2955 | if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES) |
38719025 | 2956 | return 0; |
ca8bd38b CEB |
2957 | |
2958 | return maxpages; | |
ca8bd38b CEB |
2959 | } |
2960 | ||
4b3ef9da | 2961 | #define SWAP_CLUSTER_INFO_COLS \ |
235b6217 | 2962 | DIV_ROUND_UP(L1_CACHE_BYTES, sizeof(struct swap_cluster_info)) |
4b3ef9da HY |
2963 | #define SWAP_CLUSTER_SPACE_COLS \ |
2964 | DIV_ROUND_UP(SWAP_ADDRESS_SPACE_PAGES, SWAPFILE_CLUSTER) | |
2965 | #define SWAP_CLUSTER_COLS \ | |
2966 | max_t(unsigned int, SWAP_CLUSTER_INFO_COLS, SWAP_CLUSTER_SPACE_COLS) | |
235b6217 | 2967 | |
915d4d7b CEB |
2968 | static int setup_swap_map_and_extents(struct swap_info_struct *p, |
2969 | union swap_header *swap_header, | |
2970 | unsigned char *swap_map, | |
2a8f9449 | 2971 | struct swap_cluster_info *cluster_info, |
915d4d7b CEB |
2972 | unsigned long maxpages, |
2973 | sector_t *span) | |
2974 | { | |
235b6217 | 2975 | unsigned int j, k; |
915d4d7b CEB |
2976 | unsigned int nr_good_pages; |
2977 | int nr_extents; | |
2a8f9449 | 2978 | unsigned long nr_clusters = DIV_ROUND_UP(maxpages, SWAPFILE_CLUSTER); |
235b6217 HY |
2979 | unsigned long col = p->cluster_next / SWAPFILE_CLUSTER % SWAP_CLUSTER_COLS; |
2980 | unsigned long i, idx; | |
915d4d7b CEB |
2981 | |
2982 | nr_good_pages = maxpages - 1; /* omit header page */ | |
2983 | ||
6b534915 HY |
2984 | cluster_list_init(&p->free_clusters); |
2985 | cluster_list_init(&p->discard_clusters); | |
2a8f9449 | 2986 | |
915d4d7b CEB |
2987 | for (i = 0; i < swap_header->info.nr_badpages; i++) { |
2988 | unsigned int page_nr = swap_header->info.badpages[i]; | |
bdb8e3f6 CEB |
2989 | if (page_nr == 0 || page_nr > swap_header->info.last_page) |
2990 | return -EINVAL; | |
915d4d7b CEB |
2991 | if (page_nr < maxpages) { |
2992 | swap_map[page_nr] = SWAP_MAP_BAD; | |
2993 | nr_good_pages--; | |
2a8f9449 SL |
2994 | /* |
2995 | * Haven't marked the cluster free yet, no list | |
2996 | * operation involved | |
2997 | */ | |
2998 | inc_cluster_info_page(p, cluster_info, page_nr); | |
915d4d7b CEB |
2999 | } |
3000 | } | |
3001 | ||
2a8f9449 SL |
3002 | /* Haven't marked the cluster free yet, no list operation involved */ |
3003 | for (i = maxpages; i < round_up(maxpages, SWAPFILE_CLUSTER); i++) | |
3004 | inc_cluster_info_page(p, cluster_info, i); | |
3005 | ||
915d4d7b CEB |
3006 | if (nr_good_pages) { |
3007 | swap_map[0] = SWAP_MAP_BAD; | |
2a8f9449 SL |
3008 | /* |
3009 | * Not mark the cluster free yet, no list | |
3010 | * operation involved | |
3011 | */ | |
3012 | inc_cluster_info_page(p, cluster_info, 0); | |
915d4d7b CEB |
3013 | p->max = maxpages; |
3014 | p->pages = nr_good_pages; | |
3015 | nr_extents = setup_swap_extents(p, span); | |
bdb8e3f6 CEB |
3016 | if (nr_extents < 0) |
3017 | return nr_extents; | |
915d4d7b CEB |
3018 | nr_good_pages = p->pages; |
3019 | } | |
3020 | if (!nr_good_pages) { | |
465c47fd | 3021 | pr_warn("Empty swap-file\n"); |
bdb8e3f6 | 3022 | return -EINVAL; |
915d4d7b CEB |
3023 | } |
3024 | ||
2a8f9449 SL |
3025 | if (!cluster_info) |
3026 | return nr_extents; | |
3027 | ||
235b6217 | 3028 | |
4b3ef9da HY |
3029 | /* |
3030 | * Reduce false cache line sharing between cluster_info and | |
3031 | * sharing same address space. | |
3032 | */ | |
235b6217 HY |
3033 | for (k = 0; k < SWAP_CLUSTER_COLS; k++) { |
3034 | j = (k + col) % SWAP_CLUSTER_COLS; | |
3035 | for (i = 0; i < DIV_ROUND_UP(nr_clusters, SWAP_CLUSTER_COLS); i++) { | |
3036 | idx = i * SWAP_CLUSTER_COLS + j; | |
3037 | if (idx >= nr_clusters) | |
3038 | continue; | |
3039 | if (cluster_count(&cluster_info[idx])) | |
3040 | continue; | |
2a8f9449 | 3041 | cluster_set_flag(&cluster_info[idx], CLUSTER_FLAG_FREE); |
6b534915 HY |
3042 | cluster_list_add_tail(&p->free_clusters, cluster_info, |
3043 | idx); | |
2a8f9449 | 3044 | } |
2a8f9449 | 3045 | } |
915d4d7b | 3046 | return nr_extents; |
915d4d7b CEB |
3047 | } |
3048 | ||
53cbb243 CEB |
3049 | SYSCALL_DEFINE2(swapon, const char __user *, specialfile, int, swap_flags) |
3050 | { | |
3051 | struct swap_info_struct *p; | |
91a27b2a | 3052 | struct filename *name; |
53cbb243 CEB |
3053 | struct file *swap_file = NULL; |
3054 | struct address_space *mapping; | |
51cc3a66 | 3055 | struct dentry *dentry; |
40531542 | 3056 | int prio; |
53cbb243 CEB |
3057 | int error; |
3058 | union swap_header *swap_header; | |
915d4d7b | 3059 | int nr_extents; |
53cbb243 CEB |
3060 | sector_t span; |
3061 | unsigned long maxpages; | |
53cbb243 | 3062 | unsigned char *swap_map = NULL; |
2a8f9449 | 3063 | struct swap_cluster_info *cluster_info = NULL; |
53cbb243 CEB |
3064 | struct page *page = NULL; |
3065 | struct inode *inode = NULL; | |
7cbf3192 | 3066 | bool inced_nr_rotate_swap = false; |
53cbb243 | 3067 | |
d15cab97 HD |
3068 | if (swap_flags & ~SWAP_FLAGS_VALID) |
3069 | return -EINVAL; | |
3070 | ||
53cbb243 CEB |
3071 | if (!capable(CAP_SYS_ADMIN)) |
3072 | return -EPERM; | |
3073 | ||
a2468cc9 AL |
3074 | if (!swap_avail_heads) |
3075 | return -ENOMEM; | |
3076 | ||
53cbb243 | 3077 | p = alloc_swap_info(); |
2542e513 CEB |
3078 | if (IS_ERR(p)) |
3079 | return PTR_ERR(p); | |
53cbb243 | 3080 | |
815c2c54 SL |
3081 | INIT_WORK(&p->discard_work, swap_discard_work); |
3082 | ||
1da177e4 | 3083 | name = getname(specialfile); |
1da177e4 | 3084 | if (IS_ERR(name)) { |
7de7fb6b | 3085 | error = PTR_ERR(name); |
1da177e4 | 3086 | name = NULL; |
bd69010b | 3087 | goto bad_swap; |
1da177e4 | 3088 | } |
669abf4e | 3089 | swap_file = file_open_name(name, O_RDWR|O_LARGEFILE, 0); |
1da177e4 | 3090 | if (IS_ERR(swap_file)) { |
7de7fb6b | 3091 | error = PTR_ERR(swap_file); |
1da177e4 | 3092 | swap_file = NULL; |
bd69010b | 3093 | goto bad_swap; |
1da177e4 LT |
3094 | } |
3095 | ||
3096 | p->swap_file = swap_file; | |
3097 | mapping = swap_file->f_mapping; | |
51cc3a66 | 3098 | dentry = swap_file->f_path.dentry; |
2130781e | 3099 | inode = mapping->host; |
6f179af8 | 3100 | |
4d0e1e10 CEB |
3101 | error = claim_swapfile(p, inode); |
3102 | if (unlikely(error)) | |
1da177e4 | 3103 | goto bad_swap; |
1da177e4 | 3104 | |
d795a90e | 3105 | inode_lock(inode); |
51cc3a66 HD |
3106 | if (d_unlinked(dentry) || cant_mount(dentry)) { |
3107 | error = -ENOENT; | |
3108 | goto bad_swap_unlock_inode; | |
3109 | } | |
d795a90e NA |
3110 | if (IS_SWAPFILE(inode)) { |
3111 | error = -EBUSY; | |
3112 | goto bad_swap_unlock_inode; | |
3113 | } | |
3114 | ||
1da177e4 LT |
3115 | /* |
3116 | * Read the swap header. | |
3117 | */ | |
7e0a1265 | 3118 | if (!mapping->a_ops->read_folio) { |
1da177e4 | 3119 | error = -EINVAL; |
d795a90e | 3120 | goto bad_swap_unlock_inode; |
1da177e4 | 3121 | } |
090d2b18 | 3122 | page = read_mapping_page(mapping, 0, swap_file); |
1da177e4 LT |
3123 | if (IS_ERR(page)) { |
3124 | error = PTR_ERR(page); | |
d795a90e | 3125 | goto bad_swap_unlock_inode; |
1da177e4 | 3126 | } |
81e33971 | 3127 | swap_header = kmap(page); |
1da177e4 | 3128 | |
ca8bd38b CEB |
3129 | maxpages = read_swap_header(p, swap_header, inode); |
3130 | if (unlikely(!maxpages)) { | |
1da177e4 | 3131 | error = -EINVAL; |
d795a90e | 3132 | goto bad_swap_unlock_inode; |
1da177e4 | 3133 | } |
886bb7e9 | 3134 | |
81e33971 | 3135 | /* OK, set up the swap map and apply the bad block list */ |
803d0c83 | 3136 | swap_map = vzalloc(maxpages); |
81e33971 HD |
3137 | if (!swap_map) { |
3138 | error = -ENOMEM; | |
d795a90e | 3139 | goto bad_swap_unlock_inode; |
81e33971 | 3140 | } |
f0571429 | 3141 | |
36d25489 | 3142 | if (p->bdev && bdev_stable_writes(p->bdev)) |
f0571429 MK |
3143 | p->flags |= SWP_STABLE_WRITES; |
3144 | ||
3222d8c2 | 3145 | if (p->bdev && bdev_synchronous(p->bdev)) |
539a6fea MK |
3146 | p->flags |= SWP_SYNCHRONOUS_IO; |
3147 | ||
10f0d2a5 | 3148 | if (p->bdev && bdev_nonrot(p->bdev)) { |
845982eb | 3149 | int cpu, i; |
235b6217 | 3150 | unsigned long ci, nr_cluster; |
6f179af8 | 3151 | |
2a8f9449 | 3152 | p->flags |= SWP_SOLIDSTATE; |
49070588 HY |
3153 | p->cluster_next_cpu = alloc_percpu(unsigned int); |
3154 | if (!p->cluster_next_cpu) { | |
3155 | error = -ENOMEM; | |
3156 | goto bad_swap_unlock_inode; | |
3157 | } | |
2a8f9449 SL |
3158 | /* |
3159 | * select a random position to start with to help wear leveling | |
3160 | * SSD | |
3161 | */ | |
49070588 HY |
3162 | for_each_possible_cpu(cpu) { |
3163 | per_cpu(*p->cluster_next_cpu, cpu) = | |
e8a533cb | 3164 | get_random_u32_inclusive(1, p->highest_bit); |
49070588 | 3165 | } |
235b6217 | 3166 | nr_cluster = DIV_ROUND_UP(maxpages, SWAPFILE_CLUSTER); |
2a8f9449 | 3167 | |
778e1cdd | 3168 | cluster_info = kvcalloc(nr_cluster, sizeof(*cluster_info), |
54f180d3 | 3169 | GFP_KERNEL); |
2a8f9449 SL |
3170 | if (!cluster_info) { |
3171 | error = -ENOMEM; | |
d795a90e | 3172 | goto bad_swap_unlock_inode; |
2a8f9449 | 3173 | } |
235b6217 HY |
3174 | |
3175 | for (ci = 0; ci < nr_cluster; ci++) | |
3176 | spin_lock_init(&((cluster_info + ci)->lock)); | |
3177 | ||
ebc2a1a6 SL |
3178 | p->percpu_cluster = alloc_percpu(struct percpu_cluster); |
3179 | if (!p->percpu_cluster) { | |
3180 | error = -ENOMEM; | |
d795a90e | 3181 | goto bad_swap_unlock_inode; |
ebc2a1a6 | 3182 | } |
6f179af8 | 3183 | for_each_possible_cpu(cpu) { |
ebc2a1a6 | 3184 | struct percpu_cluster *cluster; |
14c62da2 | 3185 | |
6f179af8 | 3186 | cluster = per_cpu_ptr(p->percpu_cluster, cpu); |
845982eb RR |
3187 | for (i = 0; i < SWAP_NR_ORDERS; i++) |
3188 | cluster->next[i] = SWAP_NEXT_INVALID; | |
ebc2a1a6 | 3189 | } |
7cbf3192 | 3190 | } else { |
81a0298b | 3191 | atomic_inc(&nr_rotate_swap); |
7cbf3192 OS |
3192 | inced_nr_rotate_swap = true; |
3193 | } | |
1da177e4 | 3194 | |
1421ef3c CEB |
3195 | error = swap_cgroup_swapon(p->type, maxpages); |
3196 | if (error) | |
d795a90e | 3197 | goto bad_swap_unlock_inode; |
1421ef3c | 3198 | |
915d4d7b | 3199 | nr_extents = setup_swap_map_and_extents(p, swap_header, swap_map, |
2a8f9449 | 3200 | cluster_info, maxpages, &span); |
915d4d7b CEB |
3201 | if (unlikely(nr_extents < 0)) { |
3202 | error = nr_extents; | |
d795a90e | 3203 | goto bad_swap_unlock_inode; |
1da177e4 | 3204 | } |
1da177e4 | 3205 | |
70200574 CH |
3206 | if ((swap_flags & SWAP_FLAG_DISCARD) && |
3207 | p->bdev && bdev_max_discard_sectors(p->bdev)) { | |
2a8f9449 SL |
3208 | /* |
3209 | * When discard is enabled for swap with no particular | |
3210 | * policy flagged, we set all swap discard flags here in | |
3211 | * order to sustain backward compatibility with older | |
3212 | * swapon(8) releases. | |
3213 | */ | |
3214 | p->flags |= (SWP_DISCARDABLE | SWP_AREA_DISCARD | | |
3215 | SWP_PAGE_DISCARD); | |
dcf6b7dd | 3216 | |
2a8f9449 SL |
3217 | /* |
3218 | * By flagging sys_swapon, a sysadmin can tell us to | |
3219 | * either do single-time area discards only, or to just | |
3220 | * perform discards for released swap page-clusters. | |
3221 | * Now it's time to adjust the p->flags accordingly. | |
3222 | */ | |
3223 | if (swap_flags & SWAP_FLAG_DISCARD_ONCE) | |
3224 | p->flags &= ~SWP_PAGE_DISCARD; | |
3225 | else if (swap_flags & SWAP_FLAG_DISCARD_PAGES) | |
3226 | p->flags &= ~SWP_AREA_DISCARD; | |
3227 | ||
3228 | /* issue a swapon-time discard if it's still required */ | |
3229 | if (p->flags & SWP_AREA_DISCARD) { | |
3230 | int err = discard_swap(p); | |
3231 | if (unlikely(err)) | |
3232 | pr_err("swapon: discard_swap(%p): %d\n", | |
3233 | p, err); | |
dcf6b7dd | 3234 | } |
20137a49 | 3235 | } |
6a6ba831 | 3236 | |
4b3ef9da HY |
3237 | error = init_swap_address_space(p->type, maxpages); |
3238 | if (error) | |
d795a90e | 3239 | goto bad_swap_unlock_inode; |
4b3ef9da | 3240 | |
44c7c734 | 3241 | error = zswap_swapon(p->type, maxpages); |
bb29fd77 CZ |
3242 | if (error) |
3243 | goto free_swap_address_space; | |
3244 | ||
dc617f29 DW |
3245 | /* |
3246 | * Flush any pending IO and dirty mappings before we start using this | |
3247 | * swap device. | |
3248 | */ | |
3249 | inode->i_flags |= S_SWAPFILE; | |
3250 | error = inode_drain_writes(inode); | |
3251 | if (error) { | |
3252 | inode->i_flags &= ~S_SWAPFILE; | |
bb29fd77 | 3253 | goto free_swap_zswap; |
dc617f29 DW |
3254 | } |
3255 | ||
fc0abb14 | 3256 | mutex_lock(&swapon_mutex); |
40531542 | 3257 | prio = -1; |
78ecba08 | 3258 | if (swap_flags & SWAP_FLAG_PREFER) |
40531542 | 3259 | prio = |
78ecba08 | 3260 | (swap_flags & SWAP_FLAG_PRIO_MASK) >> SWAP_FLAG_PRIO_SHIFT; |
42c06a0e | 3261 | enable_swap_info(p, prio, swap_map, cluster_info); |
c69dbfb8 | 3262 | |
42c06a0e | 3263 | pr_info("Adding %uk swap on %s. Priority:%d extents:%d across:%lluk %s%s%s%s\n", |
00cde042 Z |
3264 | K(p->pages), name->name, p->prio, nr_extents, |
3265 | K((unsigned long long)span), | |
c69dbfb8 | 3266 | (p->flags & SWP_SOLIDSTATE) ? "SS" : "", |
38b5faf4 | 3267 | (p->flags & SWP_DISCARDABLE) ? "D" : "", |
dcf6b7dd | 3268 | (p->flags & SWP_AREA_DISCARD) ? "s" : "", |
42c06a0e | 3269 | (p->flags & SWP_PAGE_DISCARD) ? "c" : ""); |
c69dbfb8 | 3270 | |
fc0abb14 | 3271 | mutex_unlock(&swapon_mutex); |
66d7dd51 KS |
3272 | atomic_inc(&proc_poll_event); |
3273 | wake_up_interruptible(&proc_poll_wait); | |
3274 | ||
1da177e4 LT |
3275 | error = 0; |
3276 | goto out; | |
bb29fd77 CZ |
3277 | free_swap_zswap: |
3278 | zswap_swapoff(p->type); | |
822bca52 ML |
3279 | free_swap_address_space: |
3280 | exit_swap_address_space(p->type); | |
d795a90e NA |
3281 | bad_swap_unlock_inode: |
3282 | inode_unlock(inode); | |
1da177e4 | 3283 | bad_swap: |
ebc2a1a6 SL |
3284 | free_percpu(p->percpu_cluster); |
3285 | p->percpu_cluster = NULL; | |
49070588 HY |
3286 | free_percpu(p->cluster_next_cpu); |
3287 | p->cluster_next_cpu = NULL; | |
16ca5dfd | 3288 | if (p->bdev_file) { |
f2090d2d | 3289 | set_blocksize(p->bdev, p->old_block_size); |
16ca5dfd CB |
3290 | fput(p->bdev_file); |
3291 | p->bdev_file = NULL; | |
1da177e4 | 3292 | } |
d795a90e | 3293 | inode = NULL; |
4cd3bb10 | 3294 | destroy_swap_extents(p); |
e8e6c2ec | 3295 | swap_cgroup_swapoff(p->type); |
5d337b91 | 3296 | spin_lock(&swap_lock); |
1da177e4 | 3297 | p->swap_file = NULL; |
1da177e4 | 3298 | p->flags = 0; |
5d337b91 | 3299 | spin_unlock(&swap_lock); |
1da177e4 | 3300 | vfree(swap_map); |
8606a1a9 | 3301 | kvfree(cluster_info); |
7cbf3192 OS |
3302 | if (inced_nr_rotate_swap) |
3303 | atomic_dec(&nr_rotate_swap); | |
d795a90e | 3304 | if (swap_file) |
1da177e4 LT |
3305 | filp_close(swap_file, NULL); |
3306 | out: | |
3307 | if (page && !IS_ERR(page)) { | |
3308 | kunmap(page); | |
09cbfeaf | 3309 | put_page(page); |
1da177e4 LT |
3310 | } |
3311 | if (name) | |
3312 | putname(name); | |
1638045c | 3313 | if (inode) |
5955102c | 3314 | inode_unlock(inode); |
039939a6 TC |
3315 | if (!error) |
3316 | enable_swap_slots_cache(); | |
1da177e4 LT |
3317 | return error; |
3318 | } | |
3319 | ||
3320 | void si_swapinfo(struct sysinfo *val) | |
3321 | { | |
efa90a98 | 3322 | unsigned int type; |
1da177e4 LT |
3323 | unsigned long nr_to_be_unused = 0; |
3324 | ||
5d337b91 | 3325 | spin_lock(&swap_lock); |
efa90a98 HD |
3326 | for (type = 0; type < nr_swapfiles; type++) { |
3327 | struct swap_info_struct *si = swap_info[type]; | |
3328 | ||
3329 | if ((si->flags & SWP_USED) && !(si->flags & SWP_WRITEOK)) | |
c8945306 | 3330 | nr_to_be_unused += READ_ONCE(si->inuse_pages); |
1da177e4 | 3331 | } |
ec8acf20 | 3332 | val->freeswap = atomic_long_read(&nr_swap_pages) + nr_to_be_unused; |
1da177e4 | 3333 | val->totalswap = total_swap_pages + nr_to_be_unused; |
5d337b91 | 3334 | spin_unlock(&swap_lock); |
1da177e4 LT |
3335 | } |
3336 | ||
3337 | /* | |
3338 | * Verify that a swap entry is valid and increment its swap map count. | |
3339 | * | |
355cfa73 KH |
3340 | * Returns error code in following case. |
3341 | * - success -> 0 | |
3342 | * - swp_entry is invalid -> EINVAL | |
3343 | * - swp_entry is migration entry -> EINVAL | |
3344 | * - swap-cache reference is requested but there is already one. -> EEXIST | |
3345 | * - swap-cache reference is requested but the entry is not used. -> ENOENT | |
570a335b | 3346 | * - swap-mapped reference requested but needs continued swap count. -> ENOMEM |
1da177e4 | 3347 | */ |
8d69aaee | 3348 | static int __swap_duplicate(swp_entry_t entry, unsigned char usage) |
1da177e4 | 3349 | { |
73c34b6a | 3350 | struct swap_info_struct *p; |
235b6217 | 3351 | struct swap_cluster_info *ci; |
c10d38cc | 3352 | unsigned long offset; |
8d69aaee HD |
3353 | unsigned char count; |
3354 | unsigned char has_cache; | |
9d9a0334 | 3355 | int err; |
1da177e4 | 3356 | |
c07aee4f | 3357 | p = swp_swap_info(entry); |
235b6217 | 3358 | |
eb085574 | 3359 | offset = swp_offset(entry); |
235b6217 | 3360 | ci = lock_cluster_or_swap_info(p, offset); |
355cfa73 | 3361 | |
253d553b | 3362 | count = p->swap_map[offset]; |
edfe23da SL |
3363 | |
3364 | /* | |
3365 | * swapin_readahead() doesn't check if a swap entry is valid, so the | |
3366 | * swap entry could be SWAP_MAP_BAD. Check here with lock held. | |
3367 | */ | |
3368 | if (unlikely(swap_count(count) == SWAP_MAP_BAD)) { | |
3369 | err = -ENOENT; | |
3370 | goto unlock_out; | |
3371 | } | |
3372 | ||
253d553b HD |
3373 | has_cache = count & SWAP_HAS_CACHE; |
3374 | count &= ~SWAP_HAS_CACHE; | |
3375 | err = 0; | |
355cfa73 | 3376 | |
253d553b | 3377 | if (usage == SWAP_HAS_CACHE) { |
355cfa73 KH |
3378 | |
3379 | /* set SWAP_HAS_CACHE if there is no cache and entry is used */ | |
253d553b HD |
3380 | if (!has_cache && count) |
3381 | has_cache = SWAP_HAS_CACHE; | |
3382 | else if (has_cache) /* someone else added cache */ | |
3383 | err = -EEXIST; | |
3384 | else /* no users remaining */ | |
3385 | err = -ENOENT; | |
355cfa73 KH |
3386 | |
3387 | } else if (count || has_cache) { | |
253d553b | 3388 | |
570a335b HD |
3389 | if ((count & ~COUNT_CONTINUED) < SWAP_MAP_MAX) |
3390 | count += usage; | |
3391 | else if ((count & ~COUNT_CONTINUED) > SWAP_MAP_MAX) | |
253d553b | 3392 | err = -EINVAL; |
570a335b HD |
3393 | else if (swap_count_continued(p, offset, count)) |
3394 | count = COUNT_CONTINUED; | |
3395 | else | |
3396 | err = -ENOMEM; | |
355cfa73 | 3397 | } else |
253d553b HD |
3398 | err = -ENOENT; /* unused swap entry */ |
3399 | ||
e26f0b93 BS |
3400 | if (!err) |
3401 | WRITE_ONCE(p->swap_map[offset], count | has_cache); | |
253d553b | 3402 | |
355cfa73 | 3403 | unlock_out: |
235b6217 | 3404 | unlock_cluster_or_swap_info(p, ci); |
253d553b | 3405 | return err; |
1da177e4 | 3406 | } |
253d553b | 3407 | |
aaa46865 HD |
3408 | /* |
3409 | * Help swapoff by noting that swap entry belongs to shmem/tmpfs | |
3410 | * (in which case its reference count is never incremented). | |
3411 | */ | |
3412 | void swap_shmem_alloc(swp_entry_t entry) | |
3413 | { | |
3414 | __swap_duplicate(entry, SWAP_MAP_SHMEM); | |
3415 | } | |
3416 | ||
355cfa73 | 3417 | /* |
08259d58 HD |
3418 | * Increase reference count of swap entry by 1. |
3419 | * Returns 0 for success, or -ENOMEM if a swap_count_continuation is required | |
3420 | * but could not be atomically allocated. Returns 0, just as if it succeeded, | |
3421 | * if __swap_duplicate() fails for another reason (-EINVAL or -ENOENT), which | |
3422 | * might occur if a page table entry has got corrupted. | |
355cfa73 | 3423 | */ |
570a335b | 3424 | int swap_duplicate(swp_entry_t entry) |
355cfa73 | 3425 | { |
570a335b HD |
3426 | int err = 0; |
3427 | ||
3428 | while (!err && __swap_duplicate(entry, 1) == -ENOMEM) | |
3429 | err = add_swap_count_continuation(entry, GFP_ATOMIC); | |
3430 | return err; | |
355cfa73 | 3431 | } |
1da177e4 | 3432 | |
cb4b86ba | 3433 | /* |
355cfa73 KH |
3434 | * @entry: swap entry for which we allocate swap cache. |
3435 | * | |
73c34b6a | 3436 | * Called when allocating swap cache for existing swap entry, |
355cfa73 | 3437 | * This can return error codes. Returns 0 at success. |
3eeba135 | 3438 | * -EEXIST means there is a swap cache. |
355cfa73 | 3439 | * Note: return code is different from swap_duplicate(). |
cb4b86ba KH |
3440 | */ |
3441 | int swapcache_prepare(swp_entry_t entry) | |
3442 | { | |
253d553b | 3443 | return __swap_duplicate(entry, SWAP_HAS_CACHE); |
cb4b86ba KH |
3444 | } |
3445 | ||
13ddaf26 KS |
3446 | void swapcache_clear(struct swap_info_struct *si, swp_entry_t entry) |
3447 | { | |
3448 | struct swap_cluster_info *ci; | |
3449 | unsigned long offset = swp_offset(entry); | |
3450 | unsigned char usage; | |
3451 | ||
3452 | ci = lock_cluster_or_swap_info(si, offset); | |
3453 | usage = __swap_entry_free_locked(si, offset, SWAP_HAS_CACHE); | |
3454 | unlock_cluster_or_swap_info(si, ci); | |
3455 | if (!usage) | |
3456 | free_swap_slot(entry); | |
3457 | } | |
3458 | ||
0bcac06f MK |
3459 | struct swap_info_struct *swp_swap_info(swp_entry_t entry) |
3460 | { | |
c10d38cc | 3461 | return swap_type_to_swap_info(swp_type(entry)); |
0bcac06f MK |
3462 | } |
3463 | ||
f981c595 | 3464 | /* |
2f52578f | 3465 | * out-of-line methods to avoid include hell. |
f981c595 | 3466 | */ |
2f52578f | 3467 | struct address_space *swapcache_mapping(struct folio *folio) |
f981c595 | 3468 | { |
69fe7d67 | 3469 | return swp_swap_info(folio->swap)->swap_file->f_mapping; |
f981c595 | 3470 | } |
2f52578f | 3471 | EXPORT_SYMBOL_GPL(swapcache_mapping); |
f981c595 MG |
3472 | |
3473 | pgoff_t __page_file_index(struct page *page) | |
3474 | { | |
cfeed8ff | 3475 | swp_entry_t swap = page_swap_entry(page); |
f981c595 MG |
3476 | return swp_offset(swap); |
3477 | } | |
3478 | EXPORT_SYMBOL_GPL(__page_file_index); | |
3479 | ||
570a335b HD |
3480 | /* |
3481 | * add_swap_count_continuation - called when a swap count is duplicated | |
3482 | * beyond SWAP_MAP_MAX, it allocates a new page and links that to the entry's | |
3483 | * page of the original vmalloc'ed swap_map, to hold the continuation count | |
3484 | * (for that entry and for its neighbouring PAGE_SIZE swap entries). Called | |
3485 | * again when count is duplicated beyond SWAP_MAP_MAX * SWAP_CONT_MAX, etc. | |
3486 | * | |
3487 | * These continuation pages are seldom referenced: the common paths all work | |
3488 | * on the original swap_map, only referring to a continuation page when the | |
3489 | * low "digit" of a count is incremented or decremented through SWAP_MAP_MAX. | |
3490 | * | |
3491 | * add_swap_count_continuation(, GFP_ATOMIC) can be called while holding | |
3492 | * page table locks; if it fails, add_swap_count_continuation(, GFP_KERNEL) | |
3493 | * can be called after dropping locks. | |
3494 | */ | |
3495 | int add_swap_count_continuation(swp_entry_t entry, gfp_t gfp_mask) | |
3496 | { | |
3497 | struct swap_info_struct *si; | |
235b6217 | 3498 | struct swap_cluster_info *ci; |
570a335b HD |
3499 | struct page *head; |
3500 | struct page *page; | |
3501 | struct page *list_page; | |
3502 | pgoff_t offset; | |
3503 | unsigned char count; | |
eb085574 | 3504 | int ret = 0; |
570a335b HD |
3505 | |
3506 | /* | |
3507 | * When debugging, it's easier to use __GFP_ZERO here; but it's better | |
3508 | * for latency not to zero a page while GFP_ATOMIC and holding locks. | |
3509 | */ | |
3510 | page = alloc_page(gfp_mask | __GFP_HIGHMEM); | |
3511 | ||
eb085574 | 3512 | si = get_swap_device(entry); |
570a335b HD |
3513 | if (!si) { |
3514 | /* | |
3515 | * An acceptable race has occurred since the failing | |
eb085574 | 3516 | * __swap_duplicate(): the swap device may be swapoff |
570a335b HD |
3517 | */ |
3518 | goto outer; | |
3519 | } | |
eb085574 | 3520 | spin_lock(&si->lock); |
570a335b HD |
3521 | |
3522 | offset = swp_offset(entry); | |
235b6217 HY |
3523 | |
3524 | ci = lock_cluster(si, offset); | |
3525 | ||
d8aa24e0 | 3526 | count = swap_count(si->swap_map[offset]); |
570a335b HD |
3527 | |
3528 | if ((count & ~COUNT_CONTINUED) != SWAP_MAP_MAX) { | |
3529 | /* | |
3530 | * The higher the swap count, the more likely it is that tasks | |
3531 | * will race to add swap count continuation: we need to avoid | |
3532 | * over-provisioning. | |
3533 | */ | |
3534 | goto out; | |
3535 | } | |
3536 | ||
3537 | if (!page) { | |
eb085574 HY |
3538 | ret = -ENOMEM; |
3539 | goto out; | |
570a335b HD |
3540 | } |
3541 | ||
570a335b HD |
3542 | head = vmalloc_to_page(si->swap_map + offset); |
3543 | offset &= ~PAGE_MASK; | |
3544 | ||
2628bd6f | 3545 | spin_lock(&si->cont_lock); |
570a335b HD |
3546 | /* |
3547 | * Page allocation does not initialize the page's lru field, | |
3548 | * but it does always reset its private field. | |
3549 | */ | |
3550 | if (!page_private(head)) { | |
3551 | BUG_ON(count & COUNT_CONTINUED); | |
3552 | INIT_LIST_HEAD(&head->lru); | |
3553 | set_page_private(head, SWP_CONTINUED); | |
3554 | si->flags |= SWP_CONTINUED; | |
3555 | } | |
3556 | ||
3557 | list_for_each_entry(list_page, &head->lru, lru) { | |
3558 | unsigned char *map; | |
3559 | ||
3560 | /* | |
3561 | * If the previous map said no continuation, but we've found | |
3562 | * a continuation page, free our allocation and use this one. | |
3563 | */ | |
3564 | if (!(count & COUNT_CONTINUED)) | |
2628bd6f | 3565 | goto out_unlock_cont; |
570a335b | 3566 | |
829c3151 | 3567 | map = kmap_local_page(list_page) + offset; |
570a335b | 3568 | count = *map; |
829c3151 | 3569 | kunmap_local(map); |
570a335b HD |
3570 | |
3571 | /* | |
3572 | * If this continuation count now has some space in it, | |
3573 | * free our allocation and use this one. | |
3574 | */ | |
3575 | if ((count & ~COUNT_CONTINUED) != SWAP_CONT_MAX) | |
2628bd6f | 3576 | goto out_unlock_cont; |
570a335b HD |
3577 | } |
3578 | ||
3579 | list_add_tail(&page->lru, &head->lru); | |
3580 | page = NULL; /* now it's attached, don't free it */ | |
2628bd6f HY |
3581 | out_unlock_cont: |
3582 | spin_unlock(&si->cont_lock); | |
570a335b | 3583 | out: |
235b6217 | 3584 | unlock_cluster(ci); |
ec8acf20 | 3585 | spin_unlock(&si->lock); |
eb085574 | 3586 | put_swap_device(si); |
570a335b HD |
3587 | outer: |
3588 | if (page) | |
3589 | __free_page(page); | |
eb085574 | 3590 | return ret; |
570a335b HD |
3591 | } |
3592 | ||
3593 | /* | |
3594 | * swap_count_continued - when the original swap_map count is incremented | |
3595 | * from SWAP_MAP_MAX, check if there is already a continuation page to carry | |
3596 | * into, carry if so, or else fail until a new continuation page is allocated; | |
3597 | * when the original swap_map count is decremented from 0 with continuation, | |
3598 | * borrow from the continuation and report whether it still holds more. | |
235b6217 HY |
3599 | * Called while __swap_duplicate() or swap_entry_free() holds swap or cluster |
3600 | * lock. | |
570a335b HD |
3601 | */ |
3602 | static bool swap_count_continued(struct swap_info_struct *si, | |
3603 | pgoff_t offset, unsigned char count) | |
3604 | { | |
3605 | struct page *head; | |
3606 | struct page *page; | |
3607 | unsigned char *map; | |
2628bd6f | 3608 | bool ret; |
570a335b HD |
3609 | |
3610 | head = vmalloc_to_page(si->swap_map + offset); | |
3611 | if (page_private(head) != SWP_CONTINUED) { | |
3612 | BUG_ON(count & COUNT_CONTINUED); | |
3613 | return false; /* need to add count continuation */ | |
3614 | } | |
3615 | ||
2628bd6f | 3616 | spin_lock(&si->cont_lock); |
570a335b | 3617 | offset &= ~PAGE_MASK; |
213516ac | 3618 | page = list_next_entry(head, lru); |
829c3151 | 3619 | map = kmap_local_page(page) + offset; |
570a335b HD |
3620 | |
3621 | if (count == SWAP_MAP_MAX) /* initial increment from swap_map */ | |
3622 | goto init_map; /* jump over SWAP_CONT_MAX checks */ | |
3623 | ||
3624 | if (count == (SWAP_MAP_MAX | COUNT_CONTINUED)) { /* incrementing */ | |
3625 | /* | |
3626 | * Think of how you add 1 to 999 | |
3627 | */ | |
3628 | while (*map == (SWAP_CONT_MAX | COUNT_CONTINUED)) { | |
829c3151 | 3629 | kunmap_local(map); |
213516ac | 3630 | page = list_next_entry(page, lru); |
570a335b | 3631 | BUG_ON(page == head); |
829c3151 | 3632 | map = kmap_local_page(page) + offset; |
570a335b HD |
3633 | } |
3634 | if (*map == SWAP_CONT_MAX) { | |
829c3151 | 3635 | kunmap_local(map); |
213516ac | 3636 | page = list_next_entry(page, lru); |
2628bd6f HY |
3637 | if (page == head) { |
3638 | ret = false; /* add count continuation */ | |
3639 | goto out; | |
3640 | } | |
829c3151 | 3641 | map = kmap_local_page(page) + offset; |
570a335b HD |
3642 | init_map: *map = 0; /* we didn't zero the page */ |
3643 | } | |
3644 | *map += 1; | |
829c3151 | 3645 | kunmap_local(map); |
213516ac | 3646 | while ((page = list_prev_entry(page, lru)) != head) { |
829c3151 | 3647 | map = kmap_local_page(page) + offset; |
570a335b | 3648 | *map = COUNT_CONTINUED; |
829c3151 | 3649 | kunmap_local(map); |
570a335b | 3650 | } |
2628bd6f | 3651 | ret = true; /* incremented */ |
570a335b HD |
3652 | |
3653 | } else { /* decrementing */ | |
3654 | /* | |
3655 | * Think of how you subtract 1 from 1000 | |
3656 | */ | |
3657 | BUG_ON(count != COUNT_CONTINUED); | |
3658 | while (*map == COUNT_CONTINUED) { | |
829c3151 | 3659 | kunmap_local(map); |
213516ac | 3660 | page = list_next_entry(page, lru); |
570a335b | 3661 | BUG_ON(page == head); |
829c3151 | 3662 | map = kmap_local_page(page) + offset; |
570a335b HD |
3663 | } |
3664 | BUG_ON(*map == 0); | |
3665 | *map -= 1; | |
3666 | if (*map == 0) | |
3667 | count = 0; | |
829c3151 | 3668 | kunmap_local(map); |
213516ac | 3669 | while ((page = list_prev_entry(page, lru)) != head) { |
829c3151 | 3670 | map = kmap_local_page(page) + offset; |
570a335b HD |
3671 | *map = SWAP_CONT_MAX | count; |
3672 | count = COUNT_CONTINUED; | |
829c3151 | 3673 | kunmap_local(map); |
570a335b | 3674 | } |
2628bd6f | 3675 | ret = count == COUNT_CONTINUED; |
570a335b | 3676 | } |
2628bd6f HY |
3677 | out: |
3678 | spin_unlock(&si->cont_lock); | |
3679 | return ret; | |
570a335b HD |
3680 | } |
3681 | ||
3682 | /* | |
3683 | * free_swap_count_continuations - swapoff free all the continuation pages | |
3684 | * appended to the swap_map, after swap_map is quiesced, before vfree'ing it. | |
3685 | */ | |
3686 | static void free_swap_count_continuations(struct swap_info_struct *si) | |
3687 | { | |
3688 | pgoff_t offset; | |
3689 | ||
3690 | for (offset = 0; offset < si->max; offset += PAGE_SIZE) { | |
3691 | struct page *head; | |
3692 | head = vmalloc_to_page(si->swap_map + offset); | |
3693 | if (page_private(head)) { | |
0d576d20 GT |
3694 | struct page *page, *next; |
3695 | ||
3696 | list_for_each_entry_safe(page, next, &head->lru, lru) { | |
3697 | list_del(&page->lru); | |
570a335b HD |
3698 | __free_page(page); |
3699 | } | |
3700 | } | |
3701 | } | |
3702 | } | |
a2468cc9 | 3703 | |
2cf85583 | 3704 | #if defined(CONFIG_MEMCG) && defined(CONFIG_BLK_CGROUP) |
3e4fb13a | 3705 | void __folio_throttle_swaprate(struct folio *folio, gfp_t gfp) |
2cf85583 TH |
3706 | { |
3707 | struct swap_info_struct *si, *next; | |
3e4fb13a | 3708 | int nid = folio_nid(folio); |
6caa6a07 | 3709 | |
3e4fb13a | 3710 | if (!(gfp & __GFP_IO)) |
2cf85583 TH |
3711 | return; |
3712 | ||
3713 | if (!blk_cgroup_congested()) | |
3714 | return; | |
3715 | ||
3716 | /* | |
3717 | * We've already scheduled a throttle, avoid taking the global swap | |
3718 | * lock. | |
3719 | */ | |
f05837ed | 3720 | if (current->throttle_disk) |
2cf85583 TH |
3721 | return; |
3722 | ||
3723 | spin_lock(&swap_avail_lock); | |
6caa6a07 JW |
3724 | plist_for_each_entry_safe(si, next, &swap_avail_heads[nid], |
3725 | avail_lists[nid]) { | |
2cf85583 | 3726 | if (si->bdev) { |
de185b56 | 3727 | blkcg_schedule_throttle(si->bdev->bd_disk, true); |
2cf85583 TH |
3728 | break; |
3729 | } | |
3730 | } | |
3731 | spin_unlock(&swap_avail_lock); | |
3732 | } | |
3733 | #endif | |
3734 | ||
a2468cc9 AL |
3735 | static int __init swapfile_init(void) |
3736 | { | |
3737 | int nid; | |
3738 | ||
3739 | swap_avail_heads = kmalloc_array(nr_node_ids, sizeof(struct plist_head), | |
3740 | GFP_KERNEL); | |
3741 | if (!swap_avail_heads) { | |
3742 | pr_emerg("Not enough memory for swap heads, swap is disabled\n"); | |
3743 | return -ENOMEM; | |
3744 | } | |
3745 | ||
3746 | for_each_node(nid) | |
3747 | plist_head_init(&swap_avail_heads[nid]); | |
3748 | ||
be45a490 PX |
3749 | swapfile_maximum_size = arch_max_swapfile_size(); |
3750 | ||
5154e607 PX |
3751 | #ifdef CONFIG_MIGRATION |
3752 | if (swapfile_maximum_size >= (1UL << SWP_MIG_TOTAL_BITS)) | |
3753 | swap_migration_ad_supported = true; | |
3754 | #endif /* CONFIG_MIGRATION */ | |
3755 | ||
a2468cc9 AL |
3756 | return 0; |
3757 | } | |
3758 | subsys_initcall(swapfile_init); |