Commit | Line | Data |
---|---|---|
457c8996 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
1da177e4 LT |
2 | /* |
3 | * linux/mm/swapfile.c | |
4 | * | |
5 | * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds | |
6 | * Swap reorganised 29.12.95, Stephen Tweedie | |
7 | */ | |
8 | ||
c97ab271 | 9 | #include <linux/blkdev.h> |
1da177e4 | 10 | #include <linux/mm.h> |
6e84f315 | 11 | #include <linux/sched/mm.h> |
29930025 | 12 | #include <linux/sched/task.h> |
1da177e4 LT |
13 | #include <linux/hugetlb.h> |
14 | #include <linux/mman.h> | |
15 | #include <linux/slab.h> | |
16 | #include <linux/kernel_stat.h> | |
17 | #include <linux/swap.h> | |
18 | #include <linux/vmalloc.h> | |
19 | #include <linux/pagemap.h> | |
20 | #include <linux/namei.h> | |
072441e2 | 21 | #include <linux/shmem_fs.h> |
e41d12f5 | 22 | #include <linux/blk-cgroup.h> |
20137a49 | 23 | #include <linux/random.h> |
1da177e4 LT |
24 | #include <linux/writeback.h> |
25 | #include <linux/proc_fs.h> | |
26 | #include <linux/seq_file.h> | |
27 | #include <linux/init.h> | |
5ad64688 | 28 | #include <linux/ksm.h> |
1da177e4 LT |
29 | #include <linux/rmap.h> |
30 | #include <linux/security.h> | |
31 | #include <linux/backing-dev.h> | |
fc0abb14 | 32 | #include <linux/mutex.h> |
c59ede7b | 33 | #include <linux/capability.h> |
1da177e4 | 34 | #include <linux/syscalls.h> |
8a9f3ccd | 35 | #include <linux/memcontrol.h> |
66d7dd51 | 36 | #include <linux/poll.h> |
72788c38 | 37 | #include <linux/oom.h> |
38b5faf4 DM |
38 | #include <linux/frontswap.h> |
39 | #include <linux/swapfile.h> | |
f981c595 | 40 | #include <linux/export.h> |
67afa38e | 41 | #include <linux/swap_slots.h> |
155b5f88 | 42 | #include <linux/sort.h> |
63d8620e | 43 | #include <linux/completion.h> |
1da177e4 | 44 | |
1da177e4 LT |
45 | #include <asm/tlbflush.h> |
46 | #include <linux/swapops.h> | |
5d1ea48b | 47 | #include <linux/swap_cgroup.h> |
014bb1de | 48 | #include "swap.h" |
1da177e4 | 49 | |
570a335b HD |
50 | static bool swap_count_continued(struct swap_info_struct *, pgoff_t, |
51 | unsigned char); | |
52 | static void free_swap_count_continuations(struct swap_info_struct *); | |
53 | ||
633423a0 | 54 | static DEFINE_SPINLOCK(swap_lock); |
7c363b8c | 55 | static unsigned int nr_swapfiles; |
ec8acf20 | 56 | atomic_long_t nr_swap_pages; |
fb0fec50 CW |
57 | /* |
58 | * Some modules use swappable objects and may try to swap them out under | |
59 | * memory pressure (via the shrinker). Before doing so, they may wish to | |
60 | * check to see if any swap space is available. | |
61 | */ | |
62 | EXPORT_SYMBOL_GPL(nr_swap_pages); | |
ec8acf20 | 63 | /* protected with swap_lock. reading in vm_swap_full() doesn't need lock */ |
1da177e4 | 64 | long total_swap_pages; |
a2468cc9 | 65 | static int least_priority = -1; |
be45a490 | 66 | unsigned long swapfile_maximum_size; |
5154e607 PX |
67 | #ifdef CONFIG_MIGRATION |
68 | bool swap_migration_ad_supported; | |
69 | #endif /* CONFIG_MIGRATION */ | |
1da177e4 | 70 | |
1da177e4 LT |
71 | static const char Bad_file[] = "Bad swap file entry "; |
72 | static const char Unused_file[] = "Unused swap file entry "; | |
73 | static const char Bad_offset[] = "Bad swap offset entry "; | |
74 | static const char Unused_offset[] = "Unused swap offset entry "; | |
75 | ||
adfab836 DS |
76 | /* |
77 | * all active swap_info_structs | |
78 | * protected with swap_lock, and ordered by priority. | |
79 | */ | |
633423a0 | 80 | static PLIST_HEAD(swap_active_head); |
18ab4d4c DS |
81 | |
82 | /* | |
83 | * all available (active, not full) swap_info_structs | |
84 | * protected with swap_avail_lock, ordered by priority. | |
e2e3fdc7 | 85 | * This is used by folio_alloc_swap() instead of swap_active_head |
18ab4d4c | 86 | * because swap_active_head includes all swap_info_structs, |
e2e3fdc7 | 87 | * but folio_alloc_swap() doesn't need to look at full ones. |
18ab4d4c DS |
88 | * This uses its own lock instead of swap_lock because when a |
89 | * swap_info_struct changes between not-full/full, it needs to | |
90 | * add/remove itself to/from this list, but the swap_info_struct->lock | |
91 | * is held and the locking order requires swap_lock to be taken | |
92 | * before any swap_info_struct->lock. | |
93 | */ | |
bfc6b1ca | 94 | static struct plist_head *swap_avail_heads; |
18ab4d4c | 95 | static DEFINE_SPINLOCK(swap_avail_lock); |
1da177e4 | 96 | |
38b5faf4 | 97 | struct swap_info_struct *swap_info[MAX_SWAPFILES]; |
1da177e4 | 98 | |
fc0abb14 | 99 | static DEFINE_MUTEX(swapon_mutex); |
1da177e4 | 100 | |
66d7dd51 KS |
101 | static DECLARE_WAIT_QUEUE_HEAD(proc_poll_wait); |
102 | /* Activity counter to indicate that a swapon or swapoff has occurred */ | |
103 | static atomic_t proc_poll_event = ATOMIC_INIT(0); | |
104 | ||
81a0298b HY |
105 | atomic_t nr_rotate_swap = ATOMIC_INIT(0); |
106 | ||
c10d38cc DJ |
107 | static struct swap_info_struct *swap_type_to_swap_info(int type) |
108 | { | |
a4b45114 | 109 | if (type >= MAX_SWAPFILES) |
c10d38cc DJ |
110 | return NULL; |
111 | ||
a4b45114 | 112 | return READ_ONCE(swap_info[type]); /* rcu_dereference() */ |
c10d38cc DJ |
113 | } |
114 | ||
8d69aaee | 115 | static inline unsigned char swap_count(unsigned char ent) |
355cfa73 | 116 | { |
955c97f0 | 117 | return ent & ~SWAP_HAS_CACHE; /* may include COUNT_CONTINUED flag */ |
355cfa73 KH |
118 | } |
119 | ||
bcd49e86 HY |
120 | /* Reclaim the swap entry anyway if possible */ |
121 | #define TTRS_ANYWAY 0x1 | |
122 | /* | |
123 | * Reclaim the swap entry if there are no more mappings of the | |
124 | * corresponding page | |
125 | */ | |
126 | #define TTRS_UNMAPPED 0x2 | |
127 | /* Reclaim the swap entry if swap is getting full*/ | |
128 | #define TTRS_FULL 0x4 | |
129 | ||
efa90a98 | 130 | /* returns 1 if swap entry is freed */ |
bcd49e86 HY |
131 | static int __try_to_reclaim_swap(struct swap_info_struct *si, |
132 | unsigned long offset, unsigned long flags) | |
c9e44410 | 133 | { |
efa90a98 | 134 | swp_entry_t entry = swp_entry(si->type, offset); |
c9e44410 KH |
135 | struct page *page; |
136 | int ret = 0; | |
137 | ||
bcd49e86 | 138 | page = find_get_page(swap_address_space(entry), offset); |
c9e44410 KH |
139 | if (!page) |
140 | return 0; | |
141 | /* | |
bcd49e86 HY |
142 | * When this function is called from scan_swap_map_slots() and it's |
143 | * called by vmscan.c at reclaiming pages. So, we hold a lock on a page, | |
144 | * here. We have to use trylock for avoiding deadlock. This is a special | |
c9e44410 KH |
145 | * case and you should use try_to_free_swap() with explicit lock_page() |
146 | * in usual operations. | |
147 | */ | |
148 | if (trylock_page(page)) { | |
bcd49e86 HY |
149 | if ((flags & TTRS_ANYWAY) || |
150 | ((flags & TTRS_UNMAPPED) && !page_mapped(page)) || | |
151 | ((flags & TTRS_FULL) && mem_cgroup_swap_full(page))) | |
152 | ret = try_to_free_swap(page); | |
c9e44410 KH |
153 | unlock_page(page); |
154 | } | |
09cbfeaf | 155 | put_page(page); |
c9e44410 KH |
156 | return ret; |
157 | } | |
355cfa73 | 158 | |
4efaceb1 AL |
159 | static inline struct swap_extent *first_se(struct swap_info_struct *sis) |
160 | { | |
161 | struct rb_node *rb = rb_first(&sis->swap_extent_root); | |
162 | return rb_entry(rb, struct swap_extent, rb_node); | |
163 | } | |
164 | ||
165 | static inline struct swap_extent *next_se(struct swap_extent *se) | |
166 | { | |
167 | struct rb_node *rb = rb_next(&se->rb_node); | |
168 | return rb ? rb_entry(rb, struct swap_extent, rb_node) : NULL; | |
169 | } | |
170 | ||
6a6ba831 HD |
171 | /* |
172 | * swapon tell device that all the old swap contents can be discarded, | |
173 | * to allow the swap device to optimize its wear-levelling. | |
174 | */ | |
175 | static int discard_swap(struct swap_info_struct *si) | |
176 | { | |
177 | struct swap_extent *se; | |
9625a5f2 HD |
178 | sector_t start_block; |
179 | sector_t nr_blocks; | |
6a6ba831 HD |
180 | int err = 0; |
181 | ||
9625a5f2 | 182 | /* Do not discard the swap header page! */ |
4efaceb1 | 183 | se = first_se(si); |
9625a5f2 HD |
184 | start_block = (se->start_block + 1) << (PAGE_SHIFT - 9); |
185 | nr_blocks = ((sector_t)se->nr_pages - 1) << (PAGE_SHIFT - 9); | |
186 | if (nr_blocks) { | |
187 | err = blkdev_issue_discard(si->bdev, start_block, | |
44abff2c | 188 | nr_blocks, GFP_KERNEL); |
9625a5f2 HD |
189 | if (err) |
190 | return err; | |
191 | cond_resched(); | |
192 | } | |
6a6ba831 | 193 | |
4efaceb1 | 194 | for (se = next_se(se); se; se = next_se(se)) { |
9625a5f2 HD |
195 | start_block = se->start_block << (PAGE_SHIFT - 9); |
196 | nr_blocks = (sector_t)se->nr_pages << (PAGE_SHIFT - 9); | |
6a6ba831 HD |
197 | |
198 | err = blkdev_issue_discard(si->bdev, start_block, | |
44abff2c | 199 | nr_blocks, GFP_KERNEL); |
6a6ba831 HD |
200 | if (err) |
201 | break; | |
202 | ||
203 | cond_resched(); | |
204 | } | |
205 | return err; /* That will often be -EOPNOTSUPP */ | |
206 | } | |
207 | ||
4efaceb1 AL |
208 | static struct swap_extent * |
209 | offset_to_swap_extent(struct swap_info_struct *sis, unsigned long offset) | |
210 | { | |
211 | struct swap_extent *se; | |
212 | struct rb_node *rb; | |
213 | ||
214 | rb = sis->swap_extent_root.rb_node; | |
215 | while (rb) { | |
216 | se = rb_entry(rb, struct swap_extent, rb_node); | |
217 | if (offset < se->start_page) | |
218 | rb = rb->rb_left; | |
219 | else if (offset >= se->start_page + se->nr_pages) | |
220 | rb = rb->rb_right; | |
221 | else | |
222 | return se; | |
223 | } | |
224 | /* It *must* be present */ | |
225 | BUG(); | |
226 | } | |
227 | ||
caf6912f JA |
228 | sector_t swap_page_sector(struct page *page) |
229 | { | |
230 | struct swap_info_struct *sis = page_swap_info(page); | |
231 | struct swap_extent *se; | |
232 | sector_t sector; | |
233 | pgoff_t offset; | |
234 | ||
235 | offset = __page_file_index(page); | |
236 | se = offset_to_swap_extent(sis, offset); | |
237 | sector = se->start_block + (offset - se->start_page); | |
238 | return sector << (PAGE_SHIFT - 9); | |
239 | } | |
240 | ||
7992fde7 HD |
241 | /* |
242 | * swap allocation tell device that a cluster of swap can now be discarded, | |
243 | * to allow the swap device to optimize its wear-levelling. | |
244 | */ | |
245 | static void discard_swap_cluster(struct swap_info_struct *si, | |
246 | pgoff_t start_page, pgoff_t nr_pages) | |
247 | { | |
4efaceb1 | 248 | struct swap_extent *se = offset_to_swap_extent(si, start_page); |
7992fde7 HD |
249 | |
250 | while (nr_pages) { | |
4efaceb1 AL |
251 | pgoff_t offset = start_page - se->start_page; |
252 | sector_t start_block = se->start_block + offset; | |
253 | sector_t nr_blocks = se->nr_pages - offset; | |
254 | ||
255 | if (nr_blocks > nr_pages) | |
256 | nr_blocks = nr_pages; | |
257 | start_page += nr_blocks; | |
258 | nr_pages -= nr_blocks; | |
259 | ||
260 | start_block <<= PAGE_SHIFT - 9; | |
261 | nr_blocks <<= PAGE_SHIFT - 9; | |
262 | if (blkdev_issue_discard(si->bdev, start_block, | |
44abff2c | 263 | nr_blocks, GFP_NOIO)) |
4efaceb1 | 264 | break; |
7992fde7 | 265 | |
4efaceb1 | 266 | se = next_se(se); |
7992fde7 HD |
267 | } |
268 | } | |
269 | ||
38d8b4e6 HY |
270 | #ifdef CONFIG_THP_SWAP |
271 | #define SWAPFILE_CLUSTER HPAGE_PMD_NR | |
a448f2d0 HY |
272 | |
273 | #define swap_entry_size(size) (size) | |
38d8b4e6 | 274 | #else |
048c27fd | 275 | #define SWAPFILE_CLUSTER 256 |
a448f2d0 HY |
276 | |
277 | /* | |
278 | * Define swap_entry_size() as constant to let compiler to optimize | |
279 | * out some code if !CONFIG_THP_SWAP | |
280 | */ | |
281 | #define swap_entry_size(size) 1 | |
38d8b4e6 | 282 | #endif |
048c27fd HD |
283 | #define LATENCY_LIMIT 256 |
284 | ||
2a8f9449 SL |
285 | static inline void cluster_set_flag(struct swap_cluster_info *info, |
286 | unsigned int flag) | |
287 | { | |
288 | info->flags = flag; | |
289 | } | |
290 | ||
291 | static inline unsigned int cluster_count(struct swap_cluster_info *info) | |
292 | { | |
293 | return info->data; | |
294 | } | |
295 | ||
296 | static inline void cluster_set_count(struct swap_cluster_info *info, | |
297 | unsigned int c) | |
298 | { | |
299 | info->data = c; | |
300 | } | |
301 | ||
302 | static inline void cluster_set_count_flag(struct swap_cluster_info *info, | |
303 | unsigned int c, unsigned int f) | |
304 | { | |
305 | info->flags = f; | |
306 | info->data = c; | |
307 | } | |
308 | ||
309 | static inline unsigned int cluster_next(struct swap_cluster_info *info) | |
310 | { | |
311 | return info->data; | |
312 | } | |
313 | ||
314 | static inline void cluster_set_next(struct swap_cluster_info *info, | |
315 | unsigned int n) | |
316 | { | |
317 | info->data = n; | |
318 | } | |
319 | ||
320 | static inline void cluster_set_next_flag(struct swap_cluster_info *info, | |
321 | unsigned int n, unsigned int f) | |
322 | { | |
323 | info->flags = f; | |
324 | info->data = n; | |
325 | } | |
326 | ||
327 | static inline bool cluster_is_free(struct swap_cluster_info *info) | |
328 | { | |
329 | return info->flags & CLUSTER_FLAG_FREE; | |
330 | } | |
331 | ||
332 | static inline bool cluster_is_null(struct swap_cluster_info *info) | |
333 | { | |
334 | return info->flags & CLUSTER_FLAG_NEXT_NULL; | |
335 | } | |
336 | ||
337 | static inline void cluster_set_null(struct swap_cluster_info *info) | |
338 | { | |
339 | info->flags = CLUSTER_FLAG_NEXT_NULL; | |
340 | info->data = 0; | |
341 | } | |
342 | ||
e0709829 HY |
343 | static inline bool cluster_is_huge(struct swap_cluster_info *info) |
344 | { | |
33ee011e HY |
345 | if (IS_ENABLED(CONFIG_THP_SWAP)) |
346 | return info->flags & CLUSTER_FLAG_HUGE; | |
347 | return false; | |
e0709829 HY |
348 | } |
349 | ||
350 | static inline void cluster_clear_huge(struct swap_cluster_info *info) | |
351 | { | |
352 | info->flags &= ~CLUSTER_FLAG_HUGE; | |
353 | } | |
354 | ||
235b6217 HY |
355 | static inline struct swap_cluster_info *lock_cluster(struct swap_info_struct *si, |
356 | unsigned long offset) | |
357 | { | |
358 | struct swap_cluster_info *ci; | |
359 | ||
360 | ci = si->cluster_info; | |
361 | if (ci) { | |
362 | ci += offset / SWAPFILE_CLUSTER; | |
363 | spin_lock(&ci->lock); | |
364 | } | |
365 | return ci; | |
366 | } | |
367 | ||
368 | static inline void unlock_cluster(struct swap_cluster_info *ci) | |
369 | { | |
370 | if (ci) | |
371 | spin_unlock(&ci->lock); | |
372 | } | |
373 | ||
59d98bf3 HY |
374 | /* |
375 | * Determine the locking method in use for this device. Return | |
376 | * swap_cluster_info if SSD-style cluster-based locking is in place. | |
377 | */ | |
235b6217 | 378 | static inline struct swap_cluster_info *lock_cluster_or_swap_info( |
59d98bf3 | 379 | struct swap_info_struct *si, unsigned long offset) |
235b6217 HY |
380 | { |
381 | struct swap_cluster_info *ci; | |
382 | ||
59d98bf3 | 383 | /* Try to use fine-grained SSD-style locking if available: */ |
235b6217 | 384 | ci = lock_cluster(si, offset); |
59d98bf3 | 385 | /* Otherwise, fall back to traditional, coarse locking: */ |
235b6217 HY |
386 | if (!ci) |
387 | spin_lock(&si->lock); | |
388 | ||
389 | return ci; | |
390 | } | |
391 | ||
392 | static inline void unlock_cluster_or_swap_info(struct swap_info_struct *si, | |
393 | struct swap_cluster_info *ci) | |
394 | { | |
395 | if (ci) | |
396 | unlock_cluster(ci); | |
397 | else | |
398 | spin_unlock(&si->lock); | |
399 | } | |
400 | ||
6b534915 HY |
401 | static inline bool cluster_list_empty(struct swap_cluster_list *list) |
402 | { | |
403 | return cluster_is_null(&list->head); | |
404 | } | |
405 | ||
406 | static inline unsigned int cluster_list_first(struct swap_cluster_list *list) | |
407 | { | |
408 | return cluster_next(&list->head); | |
409 | } | |
410 | ||
411 | static void cluster_list_init(struct swap_cluster_list *list) | |
412 | { | |
413 | cluster_set_null(&list->head); | |
414 | cluster_set_null(&list->tail); | |
415 | } | |
416 | ||
417 | static void cluster_list_add_tail(struct swap_cluster_list *list, | |
418 | struct swap_cluster_info *ci, | |
419 | unsigned int idx) | |
420 | { | |
421 | if (cluster_list_empty(list)) { | |
422 | cluster_set_next_flag(&list->head, idx, 0); | |
423 | cluster_set_next_flag(&list->tail, idx, 0); | |
424 | } else { | |
235b6217 | 425 | struct swap_cluster_info *ci_tail; |
6b534915 HY |
426 | unsigned int tail = cluster_next(&list->tail); |
427 | ||
235b6217 HY |
428 | /* |
429 | * Nested cluster lock, but both cluster locks are | |
430 | * only acquired when we held swap_info_struct->lock | |
431 | */ | |
432 | ci_tail = ci + tail; | |
433 | spin_lock_nested(&ci_tail->lock, SINGLE_DEPTH_NESTING); | |
434 | cluster_set_next(ci_tail, idx); | |
0ef017d1 | 435 | spin_unlock(&ci_tail->lock); |
6b534915 HY |
436 | cluster_set_next_flag(&list->tail, idx, 0); |
437 | } | |
438 | } | |
439 | ||
440 | static unsigned int cluster_list_del_first(struct swap_cluster_list *list, | |
441 | struct swap_cluster_info *ci) | |
442 | { | |
443 | unsigned int idx; | |
444 | ||
445 | idx = cluster_next(&list->head); | |
446 | if (cluster_next(&list->tail) == idx) { | |
447 | cluster_set_null(&list->head); | |
448 | cluster_set_null(&list->tail); | |
449 | } else | |
450 | cluster_set_next_flag(&list->head, | |
451 | cluster_next(&ci[idx]), 0); | |
452 | ||
453 | return idx; | |
454 | } | |
455 | ||
815c2c54 SL |
456 | /* Add a cluster to discard list and schedule it to do discard */ |
457 | static void swap_cluster_schedule_discard(struct swap_info_struct *si, | |
458 | unsigned int idx) | |
459 | { | |
460 | /* | |
bb243f7d | 461 | * If scan_swap_map_slots() can't find a free cluster, it will check |
815c2c54 | 462 | * si->swap_map directly. To make sure the discarding cluster isn't |
bb243f7d ML |
463 | * taken by scan_swap_map_slots(), mark the swap entries bad (occupied). |
464 | * It will be cleared after discard | |
815c2c54 SL |
465 | */ |
466 | memset(si->swap_map + idx * SWAPFILE_CLUSTER, | |
467 | SWAP_MAP_BAD, SWAPFILE_CLUSTER); | |
468 | ||
6b534915 | 469 | cluster_list_add_tail(&si->discard_clusters, si->cluster_info, idx); |
815c2c54 SL |
470 | |
471 | schedule_work(&si->discard_work); | |
472 | } | |
473 | ||
38d8b4e6 HY |
474 | static void __free_cluster(struct swap_info_struct *si, unsigned long idx) |
475 | { | |
476 | struct swap_cluster_info *ci = si->cluster_info; | |
477 | ||
478 | cluster_set_flag(ci + idx, CLUSTER_FLAG_FREE); | |
479 | cluster_list_add_tail(&si->free_clusters, ci, idx); | |
480 | } | |
481 | ||
815c2c54 SL |
482 | /* |
483 | * Doing discard actually. After a cluster discard is finished, the cluster | |
484 | * will be added to free cluster list. caller should hold si->lock. | |
485 | */ | |
486 | static void swap_do_scheduled_discard(struct swap_info_struct *si) | |
487 | { | |
235b6217 | 488 | struct swap_cluster_info *info, *ci; |
815c2c54 SL |
489 | unsigned int idx; |
490 | ||
491 | info = si->cluster_info; | |
492 | ||
6b534915 HY |
493 | while (!cluster_list_empty(&si->discard_clusters)) { |
494 | idx = cluster_list_del_first(&si->discard_clusters, info); | |
815c2c54 SL |
495 | spin_unlock(&si->lock); |
496 | ||
497 | discard_swap_cluster(si, idx * SWAPFILE_CLUSTER, | |
498 | SWAPFILE_CLUSTER); | |
499 | ||
500 | spin_lock(&si->lock); | |
235b6217 | 501 | ci = lock_cluster(si, idx * SWAPFILE_CLUSTER); |
38d8b4e6 | 502 | __free_cluster(si, idx); |
815c2c54 SL |
503 | memset(si->swap_map + idx * SWAPFILE_CLUSTER, |
504 | 0, SWAPFILE_CLUSTER); | |
235b6217 | 505 | unlock_cluster(ci); |
815c2c54 SL |
506 | } |
507 | } | |
508 | ||
509 | static void swap_discard_work(struct work_struct *work) | |
510 | { | |
511 | struct swap_info_struct *si; | |
512 | ||
513 | si = container_of(work, struct swap_info_struct, discard_work); | |
514 | ||
515 | spin_lock(&si->lock); | |
516 | swap_do_scheduled_discard(si); | |
517 | spin_unlock(&si->lock); | |
518 | } | |
519 | ||
63d8620e ML |
520 | static void swap_users_ref_free(struct percpu_ref *ref) |
521 | { | |
522 | struct swap_info_struct *si; | |
523 | ||
524 | si = container_of(ref, struct swap_info_struct, users); | |
525 | complete(&si->comp); | |
526 | } | |
527 | ||
38d8b4e6 HY |
528 | static void alloc_cluster(struct swap_info_struct *si, unsigned long idx) |
529 | { | |
530 | struct swap_cluster_info *ci = si->cluster_info; | |
531 | ||
532 | VM_BUG_ON(cluster_list_first(&si->free_clusters) != idx); | |
533 | cluster_list_del_first(&si->free_clusters, ci); | |
534 | cluster_set_count_flag(ci + idx, 0, 0); | |
535 | } | |
536 | ||
537 | static void free_cluster(struct swap_info_struct *si, unsigned long idx) | |
538 | { | |
539 | struct swap_cluster_info *ci = si->cluster_info + idx; | |
540 | ||
541 | VM_BUG_ON(cluster_count(ci) != 0); | |
542 | /* | |
543 | * If the swap is discardable, prepare discard the cluster | |
544 | * instead of free it immediately. The cluster will be freed | |
545 | * after discard. | |
546 | */ | |
547 | if ((si->flags & (SWP_WRITEOK | SWP_PAGE_DISCARD)) == | |
548 | (SWP_WRITEOK | SWP_PAGE_DISCARD)) { | |
549 | swap_cluster_schedule_discard(si, idx); | |
550 | return; | |
551 | } | |
552 | ||
553 | __free_cluster(si, idx); | |
554 | } | |
555 | ||
2a8f9449 SL |
556 | /* |
557 | * The cluster corresponding to page_nr will be used. The cluster will be | |
558 | * removed from free cluster list and its usage counter will be increased. | |
559 | */ | |
560 | static void inc_cluster_info_page(struct swap_info_struct *p, | |
561 | struct swap_cluster_info *cluster_info, unsigned long page_nr) | |
562 | { | |
563 | unsigned long idx = page_nr / SWAPFILE_CLUSTER; | |
564 | ||
565 | if (!cluster_info) | |
566 | return; | |
38d8b4e6 HY |
567 | if (cluster_is_free(&cluster_info[idx])) |
568 | alloc_cluster(p, idx); | |
2a8f9449 SL |
569 | |
570 | VM_BUG_ON(cluster_count(&cluster_info[idx]) >= SWAPFILE_CLUSTER); | |
571 | cluster_set_count(&cluster_info[idx], | |
572 | cluster_count(&cluster_info[idx]) + 1); | |
573 | } | |
574 | ||
575 | /* | |
576 | * The cluster corresponding to page_nr decreases one usage. If the usage | |
577 | * counter becomes 0, which means no page in the cluster is in using, we can | |
578 | * optionally discard the cluster and add it to free cluster list. | |
579 | */ | |
580 | static void dec_cluster_info_page(struct swap_info_struct *p, | |
581 | struct swap_cluster_info *cluster_info, unsigned long page_nr) | |
582 | { | |
583 | unsigned long idx = page_nr / SWAPFILE_CLUSTER; | |
584 | ||
585 | if (!cluster_info) | |
586 | return; | |
587 | ||
588 | VM_BUG_ON(cluster_count(&cluster_info[idx]) == 0); | |
589 | cluster_set_count(&cluster_info[idx], | |
590 | cluster_count(&cluster_info[idx]) - 1); | |
591 | ||
38d8b4e6 HY |
592 | if (cluster_count(&cluster_info[idx]) == 0) |
593 | free_cluster(p, idx); | |
2a8f9449 SL |
594 | } |
595 | ||
596 | /* | |
bb243f7d | 597 | * It's possible scan_swap_map_slots() uses a free cluster in the middle of free |
2a8f9449 SL |
598 | * cluster list. Avoiding such abuse to avoid list corruption. |
599 | */ | |
ebc2a1a6 SL |
600 | static bool |
601 | scan_swap_map_ssd_cluster_conflict(struct swap_info_struct *si, | |
2a8f9449 SL |
602 | unsigned long offset) |
603 | { | |
ebc2a1a6 SL |
604 | struct percpu_cluster *percpu_cluster; |
605 | bool conflict; | |
606 | ||
2a8f9449 | 607 | offset /= SWAPFILE_CLUSTER; |
6b534915 HY |
608 | conflict = !cluster_list_empty(&si->free_clusters) && |
609 | offset != cluster_list_first(&si->free_clusters) && | |
2a8f9449 | 610 | cluster_is_free(&si->cluster_info[offset]); |
ebc2a1a6 SL |
611 | |
612 | if (!conflict) | |
613 | return false; | |
614 | ||
615 | percpu_cluster = this_cpu_ptr(si->percpu_cluster); | |
616 | cluster_set_null(&percpu_cluster->index); | |
617 | return true; | |
618 | } | |
619 | ||
620 | /* | |
621 | * Try to get a swap entry from current cpu's swap entry pool (a cluster). This | |
622 | * might involve allocating a new cluster for current CPU too. | |
623 | */ | |
36005bae | 624 | static bool scan_swap_map_try_ssd_cluster(struct swap_info_struct *si, |
ebc2a1a6 SL |
625 | unsigned long *offset, unsigned long *scan_base) |
626 | { | |
627 | struct percpu_cluster *cluster; | |
235b6217 | 628 | struct swap_cluster_info *ci; |
235b6217 | 629 | unsigned long tmp, max; |
ebc2a1a6 SL |
630 | |
631 | new_cluster: | |
632 | cluster = this_cpu_ptr(si->percpu_cluster); | |
633 | if (cluster_is_null(&cluster->index)) { | |
6b534915 HY |
634 | if (!cluster_list_empty(&si->free_clusters)) { |
635 | cluster->index = si->free_clusters.head; | |
ebc2a1a6 SL |
636 | cluster->next = cluster_next(&cluster->index) * |
637 | SWAPFILE_CLUSTER; | |
6b534915 | 638 | } else if (!cluster_list_empty(&si->discard_clusters)) { |
ebc2a1a6 SL |
639 | /* |
640 | * we don't have free cluster but have some clusters in | |
49070588 HY |
641 | * discarding, do discard now and reclaim them, then |
642 | * reread cluster_next_cpu since we dropped si->lock | |
ebc2a1a6 SL |
643 | */ |
644 | swap_do_scheduled_discard(si); | |
49070588 HY |
645 | *scan_base = this_cpu_read(*si->cluster_next_cpu); |
646 | *offset = *scan_base; | |
ebc2a1a6 SL |
647 | goto new_cluster; |
648 | } else | |
36005bae | 649 | return false; |
ebc2a1a6 SL |
650 | } |
651 | ||
ebc2a1a6 SL |
652 | /* |
653 | * Other CPUs can use our cluster if they can't find a free cluster, | |
654 | * check if there is still free entry in the cluster | |
655 | */ | |
656 | tmp = cluster->next; | |
235b6217 HY |
657 | max = min_t(unsigned long, si->max, |
658 | (cluster_next(&cluster->index) + 1) * SWAPFILE_CLUSTER); | |
7b9e2de1 WY |
659 | if (tmp < max) { |
660 | ci = lock_cluster(si, tmp); | |
661 | while (tmp < max) { | |
662 | if (!si->swap_map[tmp]) | |
663 | break; | |
664 | tmp++; | |
665 | } | |
666 | unlock_cluster(ci); | |
ebc2a1a6 | 667 | } |
0fd0e19e | 668 | if (tmp >= max) { |
ebc2a1a6 SL |
669 | cluster_set_null(&cluster->index); |
670 | goto new_cluster; | |
671 | } | |
672 | cluster->next = tmp + 1; | |
673 | *offset = tmp; | |
674 | *scan_base = tmp; | |
fdff1deb | 675 | return true; |
2a8f9449 SL |
676 | } |
677 | ||
a2468cc9 AL |
678 | static void __del_from_avail_list(struct swap_info_struct *p) |
679 | { | |
680 | int nid; | |
681 | ||
682 | for_each_node(nid) | |
683 | plist_del(&p->avail_lists[nid], &swap_avail_heads[nid]); | |
684 | } | |
685 | ||
686 | static void del_from_avail_list(struct swap_info_struct *p) | |
687 | { | |
688 | spin_lock(&swap_avail_lock); | |
689 | __del_from_avail_list(p); | |
690 | spin_unlock(&swap_avail_lock); | |
691 | } | |
692 | ||
38d8b4e6 HY |
693 | static void swap_range_alloc(struct swap_info_struct *si, unsigned long offset, |
694 | unsigned int nr_entries) | |
695 | { | |
696 | unsigned int end = offset + nr_entries - 1; | |
697 | ||
698 | if (offset == si->lowest_bit) | |
699 | si->lowest_bit += nr_entries; | |
700 | if (end == si->highest_bit) | |
a449bf58 | 701 | WRITE_ONCE(si->highest_bit, si->highest_bit - nr_entries); |
c8945306 | 702 | WRITE_ONCE(si->inuse_pages, si->inuse_pages + nr_entries); |
38d8b4e6 HY |
703 | if (si->inuse_pages == si->pages) { |
704 | si->lowest_bit = si->max; | |
705 | si->highest_bit = 0; | |
a2468cc9 | 706 | del_from_avail_list(si); |
38d8b4e6 HY |
707 | } |
708 | } | |
709 | ||
a2468cc9 AL |
710 | static void add_to_avail_list(struct swap_info_struct *p) |
711 | { | |
712 | int nid; | |
713 | ||
714 | spin_lock(&swap_avail_lock); | |
715 | for_each_node(nid) { | |
716 | WARN_ON(!plist_node_empty(&p->avail_lists[nid])); | |
717 | plist_add(&p->avail_lists[nid], &swap_avail_heads[nid]); | |
718 | } | |
719 | spin_unlock(&swap_avail_lock); | |
720 | } | |
721 | ||
38d8b4e6 HY |
722 | static void swap_range_free(struct swap_info_struct *si, unsigned long offset, |
723 | unsigned int nr_entries) | |
724 | { | |
3852f676 | 725 | unsigned long begin = offset; |
38d8b4e6 HY |
726 | unsigned long end = offset + nr_entries - 1; |
727 | void (*swap_slot_free_notify)(struct block_device *, unsigned long); | |
728 | ||
729 | if (offset < si->lowest_bit) | |
730 | si->lowest_bit = offset; | |
731 | if (end > si->highest_bit) { | |
732 | bool was_full = !si->highest_bit; | |
733 | ||
a449bf58 | 734 | WRITE_ONCE(si->highest_bit, end); |
a2468cc9 AL |
735 | if (was_full && (si->flags & SWP_WRITEOK)) |
736 | add_to_avail_list(si); | |
38d8b4e6 HY |
737 | } |
738 | atomic_long_add(nr_entries, &nr_swap_pages); | |
c8945306 | 739 | WRITE_ONCE(si->inuse_pages, si->inuse_pages - nr_entries); |
38d8b4e6 HY |
740 | if (si->flags & SWP_BLKDEV) |
741 | swap_slot_free_notify = | |
742 | si->bdev->bd_disk->fops->swap_slot_free_notify; | |
743 | else | |
744 | swap_slot_free_notify = NULL; | |
745 | while (offset <= end) { | |
8a84802e | 746 | arch_swap_invalidate_page(si->type, offset); |
38d8b4e6 HY |
747 | frontswap_invalidate_page(si->type, offset); |
748 | if (swap_slot_free_notify) | |
749 | swap_slot_free_notify(si->bdev, offset); | |
750 | offset++; | |
751 | } | |
3852f676 | 752 | clear_shadow_from_swap_cache(si->type, begin, end); |
38d8b4e6 HY |
753 | } |
754 | ||
49070588 HY |
755 | static void set_cluster_next(struct swap_info_struct *si, unsigned long next) |
756 | { | |
757 | unsigned long prev; | |
758 | ||
759 | if (!(si->flags & SWP_SOLIDSTATE)) { | |
760 | si->cluster_next = next; | |
761 | return; | |
762 | } | |
763 | ||
764 | prev = this_cpu_read(*si->cluster_next_cpu); | |
765 | /* | |
766 | * Cross the swap address space size aligned trunk, choose | |
767 | * another trunk randomly to avoid lock contention on swap | |
768 | * address space if possible. | |
769 | */ | |
770 | if ((prev >> SWAP_ADDRESS_SPACE_SHIFT) != | |
771 | (next >> SWAP_ADDRESS_SPACE_SHIFT)) { | |
772 | /* No free swap slots available */ | |
773 | if (si->highest_bit <= si->lowest_bit) | |
774 | return; | |
775 | next = si->lowest_bit + | |
776 | prandom_u32_max(si->highest_bit - si->lowest_bit + 1); | |
777 | next = ALIGN_DOWN(next, SWAP_ADDRESS_SPACE_PAGES); | |
778 | next = max_t(unsigned int, next, si->lowest_bit); | |
779 | } | |
780 | this_cpu_write(*si->cluster_next_cpu, next); | |
781 | } | |
782 | ||
4b9ae842 ML |
783 | static bool swap_offset_available_and_locked(struct swap_info_struct *si, |
784 | unsigned long offset) | |
785 | { | |
786 | if (data_race(!si->swap_map[offset])) { | |
787 | spin_lock(&si->lock); | |
788 | return true; | |
789 | } | |
790 | ||
791 | if (vm_swap_full() && READ_ONCE(si->swap_map[offset]) == SWAP_HAS_CACHE) { | |
792 | spin_lock(&si->lock); | |
793 | return true; | |
794 | } | |
795 | ||
796 | return false; | |
797 | } | |
798 | ||
36005bae TC |
799 | static int scan_swap_map_slots(struct swap_info_struct *si, |
800 | unsigned char usage, int nr, | |
801 | swp_entry_t slots[]) | |
1da177e4 | 802 | { |
235b6217 | 803 | struct swap_cluster_info *ci; |
ebebbbe9 | 804 | unsigned long offset; |
c60aa176 | 805 | unsigned long scan_base; |
7992fde7 | 806 | unsigned long last_in_cluster = 0; |
048c27fd | 807 | int latency_ration = LATENCY_LIMIT; |
36005bae | 808 | int n_ret = 0; |
ed43af10 | 809 | bool scanned_many = false; |
36005bae | 810 | |
886bb7e9 | 811 | /* |
7dfad418 HD |
812 | * We try to cluster swap pages by allocating them sequentially |
813 | * in swap. Once we've allocated SWAPFILE_CLUSTER pages this | |
814 | * way, however, we resort to first-free allocation, starting | |
815 | * a new cluster. This prevents us from scattering swap pages | |
816 | * all over the entire swap partition, so that we reduce | |
817 | * overall disk seek times between swap pages. -- sct | |
818 | * But we do now try to find an empty cluster. -Andrea | |
c60aa176 | 819 | * And we let swap pages go all over an SSD partition. Hugh |
7dfad418 HD |
820 | */ |
821 | ||
52b7efdb | 822 | si->flags += SWP_SCANNING; |
49070588 HY |
823 | /* |
824 | * Use percpu scan base for SSD to reduce lock contention on | |
825 | * cluster and swap cache. For HDD, sequential access is more | |
826 | * important. | |
827 | */ | |
828 | if (si->flags & SWP_SOLIDSTATE) | |
829 | scan_base = this_cpu_read(*si->cluster_next_cpu); | |
830 | else | |
831 | scan_base = si->cluster_next; | |
832 | offset = scan_base; | |
ebebbbe9 | 833 | |
ebc2a1a6 SL |
834 | /* SSD algorithm */ |
835 | if (si->cluster_info) { | |
bd2d18da | 836 | if (!scan_swap_map_try_ssd_cluster(si, &offset, &scan_base)) |
36005bae | 837 | goto scan; |
f4eaf51a | 838 | } else if (unlikely(!si->cluster_nr--)) { |
ebebbbe9 HD |
839 | if (si->pages - si->inuse_pages < SWAPFILE_CLUSTER) { |
840 | si->cluster_nr = SWAPFILE_CLUSTER - 1; | |
841 | goto checks; | |
842 | } | |
2a8f9449 | 843 | |
ec8acf20 | 844 | spin_unlock(&si->lock); |
7dfad418 | 845 | |
c60aa176 HD |
846 | /* |
847 | * If seek is expensive, start searching for new cluster from | |
848 | * start of partition, to minimize the span of allocated swap. | |
50088c44 CY |
849 | * If seek is cheap, that is the SWP_SOLIDSTATE si->cluster_info |
850 | * case, just handled by scan_swap_map_try_ssd_cluster() above. | |
c60aa176 | 851 | */ |
50088c44 | 852 | scan_base = offset = si->lowest_bit; |
7dfad418 HD |
853 | last_in_cluster = offset + SWAPFILE_CLUSTER - 1; |
854 | ||
855 | /* Locate the first empty (unaligned) cluster */ | |
856 | for (; last_in_cluster <= si->highest_bit; offset++) { | |
1da177e4 | 857 | if (si->swap_map[offset]) |
7dfad418 HD |
858 | last_in_cluster = offset + SWAPFILE_CLUSTER; |
859 | else if (offset == last_in_cluster) { | |
ec8acf20 | 860 | spin_lock(&si->lock); |
ebebbbe9 HD |
861 | offset -= SWAPFILE_CLUSTER - 1; |
862 | si->cluster_next = offset; | |
863 | si->cluster_nr = SWAPFILE_CLUSTER - 1; | |
c60aa176 HD |
864 | goto checks; |
865 | } | |
866 | if (unlikely(--latency_ration < 0)) { | |
867 | cond_resched(); | |
868 | latency_ration = LATENCY_LIMIT; | |
869 | } | |
870 | } | |
871 | ||
872 | offset = scan_base; | |
ec8acf20 | 873 | spin_lock(&si->lock); |
ebebbbe9 | 874 | si->cluster_nr = SWAPFILE_CLUSTER - 1; |
1da177e4 | 875 | } |
7dfad418 | 876 | |
ebebbbe9 | 877 | checks: |
ebc2a1a6 | 878 | if (si->cluster_info) { |
36005bae TC |
879 | while (scan_swap_map_ssd_cluster_conflict(si, offset)) { |
880 | /* take a break if we already got some slots */ | |
881 | if (n_ret) | |
882 | goto done; | |
883 | if (!scan_swap_map_try_ssd_cluster(si, &offset, | |
884 | &scan_base)) | |
885 | goto scan; | |
886 | } | |
ebc2a1a6 | 887 | } |
ebebbbe9 | 888 | if (!(si->flags & SWP_WRITEOK)) |
52b7efdb | 889 | goto no_page; |
7dfad418 HD |
890 | if (!si->highest_bit) |
891 | goto no_page; | |
ebebbbe9 | 892 | if (offset > si->highest_bit) |
c60aa176 | 893 | scan_base = offset = si->lowest_bit; |
c9e44410 | 894 | |
235b6217 | 895 | ci = lock_cluster(si, offset); |
b73d7fce HD |
896 | /* reuse swap entry of cache-only swap if not busy. */ |
897 | if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) { | |
c9e44410 | 898 | int swap_was_freed; |
235b6217 | 899 | unlock_cluster(ci); |
ec8acf20 | 900 | spin_unlock(&si->lock); |
bcd49e86 | 901 | swap_was_freed = __try_to_reclaim_swap(si, offset, TTRS_ANYWAY); |
ec8acf20 | 902 | spin_lock(&si->lock); |
c9e44410 KH |
903 | /* entry was freed successfully, try to use this again */ |
904 | if (swap_was_freed) | |
905 | goto checks; | |
906 | goto scan; /* check next one */ | |
907 | } | |
908 | ||
235b6217 HY |
909 | if (si->swap_map[offset]) { |
910 | unlock_cluster(ci); | |
36005bae TC |
911 | if (!n_ret) |
912 | goto scan; | |
913 | else | |
914 | goto done; | |
235b6217 | 915 | } |
a449bf58 | 916 | WRITE_ONCE(si->swap_map[offset], usage); |
2872bb2d HY |
917 | inc_cluster_info_page(si, si->cluster_info, offset); |
918 | unlock_cluster(ci); | |
ebebbbe9 | 919 | |
38d8b4e6 | 920 | swap_range_alloc(si, offset, 1); |
36005bae TC |
921 | slots[n_ret++] = swp_entry(si->type, offset); |
922 | ||
923 | /* got enough slots or reach max slots? */ | |
924 | if ((n_ret == nr) || (offset >= si->highest_bit)) | |
925 | goto done; | |
926 | ||
927 | /* search for next available slot */ | |
928 | ||
929 | /* time to take a break? */ | |
930 | if (unlikely(--latency_ration < 0)) { | |
931 | if (n_ret) | |
932 | goto done; | |
933 | spin_unlock(&si->lock); | |
934 | cond_resched(); | |
935 | spin_lock(&si->lock); | |
936 | latency_ration = LATENCY_LIMIT; | |
937 | } | |
938 | ||
939 | /* try to get more slots in cluster */ | |
940 | if (si->cluster_info) { | |
941 | if (scan_swap_map_try_ssd_cluster(si, &offset, &scan_base)) | |
942 | goto checks; | |
f4eaf51a WY |
943 | } else if (si->cluster_nr && !si->swap_map[++offset]) { |
944 | /* non-ssd case, still more slots in cluster? */ | |
36005bae TC |
945 | --si->cluster_nr; |
946 | goto checks; | |
947 | } | |
7992fde7 | 948 | |
ed43af10 HY |
949 | /* |
950 | * Even if there's no free clusters available (fragmented), | |
951 | * try to scan a little more quickly with lock held unless we | |
952 | * have scanned too many slots already. | |
953 | */ | |
954 | if (!scanned_many) { | |
955 | unsigned long scan_limit; | |
956 | ||
957 | if (offset < scan_base) | |
958 | scan_limit = scan_base; | |
959 | else | |
960 | scan_limit = si->highest_bit; | |
961 | for (; offset <= scan_limit && --latency_ration > 0; | |
962 | offset++) { | |
963 | if (!si->swap_map[offset]) | |
964 | goto checks; | |
965 | } | |
966 | } | |
967 | ||
36005bae | 968 | done: |
49070588 | 969 | set_cluster_next(si, offset + 1); |
36005bae TC |
970 | si->flags -= SWP_SCANNING; |
971 | return n_ret; | |
7dfad418 | 972 | |
ebebbbe9 | 973 | scan: |
ec8acf20 | 974 | spin_unlock(&si->lock); |
a449bf58 | 975 | while (++offset <= READ_ONCE(si->highest_bit)) { |
4b9ae842 | 976 | if (swap_offset_available_and_locked(si, offset)) |
c9e44410 | 977 | goto checks; |
048c27fd HD |
978 | if (unlikely(--latency_ration < 0)) { |
979 | cond_resched(); | |
980 | latency_ration = LATENCY_LIMIT; | |
ed43af10 | 981 | scanned_many = true; |
048c27fd | 982 | } |
7dfad418 | 983 | } |
c60aa176 | 984 | offset = si->lowest_bit; |
a5998061 | 985 | while (offset < scan_base) { |
4b9ae842 | 986 | if (swap_offset_available_and_locked(si, offset)) |
c60aa176 | 987 | goto checks; |
c60aa176 HD |
988 | if (unlikely(--latency_ration < 0)) { |
989 | cond_resched(); | |
990 | latency_ration = LATENCY_LIMIT; | |
ed43af10 | 991 | scanned_many = true; |
c60aa176 | 992 | } |
a5998061 | 993 | offset++; |
c60aa176 | 994 | } |
ec8acf20 | 995 | spin_lock(&si->lock); |
7dfad418 HD |
996 | |
997 | no_page: | |
52b7efdb | 998 | si->flags -= SWP_SCANNING; |
36005bae | 999 | return n_ret; |
1da177e4 LT |
1000 | } |
1001 | ||
38d8b4e6 HY |
1002 | static int swap_alloc_cluster(struct swap_info_struct *si, swp_entry_t *slot) |
1003 | { | |
1004 | unsigned long idx; | |
1005 | struct swap_cluster_info *ci; | |
661c7566 | 1006 | unsigned long offset; |
38d8b4e6 | 1007 | |
fe5266d5 HY |
1008 | /* |
1009 | * Should not even be attempting cluster allocations when huge | |
1010 | * page swap is disabled. Warn and fail the allocation. | |
1011 | */ | |
1012 | if (!IS_ENABLED(CONFIG_THP_SWAP)) { | |
1013 | VM_WARN_ON_ONCE(1); | |
1014 | return 0; | |
1015 | } | |
1016 | ||
38d8b4e6 HY |
1017 | if (cluster_list_empty(&si->free_clusters)) |
1018 | return 0; | |
1019 | ||
1020 | idx = cluster_list_first(&si->free_clusters); | |
1021 | offset = idx * SWAPFILE_CLUSTER; | |
1022 | ci = lock_cluster(si, offset); | |
1023 | alloc_cluster(si, idx); | |
e0709829 | 1024 | cluster_set_count_flag(ci, SWAPFILE_CLUSTER, CLUSTER_FLAG_HUGE); |
38d8b4e6 | 1025 | |
661c7566 | 1026 | memset(si->swap_map + offset, SWAP_HAS_CACHE, SWAPFILE_CLUSTER); |
38d8b4e6 HY |
1027 | unlock_cluster(ci); |
1028 | swap_range_alloc(si, offset, SWAPFILE_CLUSTER); | |
1029 | *slot = swp_entry(si->type, offset); | |
1030 | ||
1031 | return 1; | |
1032 | } | |
1033 | ||
1034 | static void swap_free_cluster(struct swap_info_struct *si, unsigned long idx) | |
1035 | { | |
1036 | unsigned long offset = idx * SWAPFILE_CLUSTER; | |
1037 | struct swap_cluster_info *ci; | |
1038 | ||
1039 | ci = lock_cluster(si, offset); | |
979aafa5 | 1040 | memset(si->swap_map + offset, 0, SWAPFILE_CLUSTER); |
38d8b4e6 HY |
1041 | cluster_set_count_flag(ci, 0, 0); |
1042 | free_cluster(si, idx); | |
1043 | unlock_cluster(ci); | |
1044 | swap_range_free(si, offset, SWAPFILE_CLUSTER); | |
1045 | } | |
38d8b4e6 | 1046 | |
5d5e8f19 | 1047 | int get_swap_pages(int n_goal, swp_entry_t swp_entries[], int entry_size) |
1da177e4 | 1048 | { |
5d5e8f19 | 1049 | unsigned long size = swap_entry_size(entry_size); |
adfab836 | 1050 | struct swap_info_struct *si, *next; |
36005bae TC |
1051 | long avail_pgs; |
1052 | int n_ret = 0; | |
a2468cc9 | 1053 | int node; |
1da177e4 | 1054 | |
38d8b4e6 | 1055 | /* Only single cluster request supported */ |
5d5e8f19 | 1056 | WARN_ON_ONCE(n_goal > 1 && size == SWAPFILE_CLUSTER); |
38d8b4e6 | 1057 | |
b50da6e9 ZH |
1058 | spin_lock(&swap_avail_lock); |
1059 | ||
5d5e8f19 | 1060 | avail_pgs = atomic_long_read(&nr_swap_pages) / size; |
b50da6e9 ZH |
1061 | if (avail_pgs <= 0) { |
1062 | spin_unlock(&swap_avail_lock); | |
fb4f88dc | 1063 | goto noswap; |
b50da6e9 | 1064 | } |
36005bae | 1065 | |
08d3090f | 1066 | n_goal = min3((long)n_goal, (long)SWAP_BATCH, avail_pgs); |
36005bae | 1067 | |
5d5e8f19 | 1068 | atomic_long_sub(n_goal * size, &nr_swap_pages); |
fb4f88dc | 1069 | |
18ab4d4c | 1070 | start_over: |
a2468cc9 AL |
1071 | node = numa_node_id(); |
1072 | plist_for_each_entry_safe(si, next, &swap_avail_heads[node], avail_lists[node]) { | |
18ab4d4c | 1073 | /* requeue si to after same-priority siblings */ |
a2468cc9 | 1074 | plist_requeue(&si->avail_lists[node], &swap_avail_heads[node]); |
18ab4d4c | 1075 | spin_unlock(&swap_avail_lock); |
ec8acf20 | 1076 | spin_lock(&si->lock); |
adfab836 | 1077 | if (!si->highest_bit || !(si->flags & SWP_WRITEOK)) { |
18ab4d4c | 1078 | spin_lock(&swap_avail_lock); |
a2468cc9 | 1079 | if (plist_node_empty(&si->avail_lists[node])) { |
18ab4d4c DS |
1080 | spin_unlock(&si->lock); |
1081 | goto nextsi; | |
1082 | } | |
1083 | WARN(!si->highest_bit, | |
1084 | "swap_info %d in list but !highest_bit\n", | |
1085 | si->type); | |
1086 | WARN(!(si->flags & SWP_WRITEOK), | |
1087 | "swap_info %d in list but !SWP_WRITEOK\n", | |
1088 | si->type); | |
a2468cc9 | 1089 | __del_from_avail_list(si); |
ec8acf20 | 1090 | spin_unlock(&si->lock); |
18ab4d4c | 1091 | goto nextsi; |
ec8acf20 | 1092 | } |
5d5e8f19 | 1093 | if (size == SWAPFILE_CLUSTER) { |
41663430 | 1094 | if (si->flags & SWP_BLKDEV) |
f0eea189 HY |
1095 | n_ret = swap_alloc_cluster(si, swp_entries); |
1096 | } else | |
38d8b4e6 HY |
1097 | n_ret = scan_swap_map_slots(si, SWAP_HAS_CACHE, |
1098 | n_goal, swp_entries); | |
ec8acf20 | 1099 | spin_unlock(&si->lock); |
5d5e8f19 | 1100 | if (n_ret || size == SWAPFILE_CLUSTER) |
36005bae | 1101 | goto check_out; |
18ab4d4c | 1102 | pr_debug("scan_swap_map of si %d failed to find offset\n", |
36005bae TC |
1103 | si->type); |
1104 | ||
18ab4d4c DS |
1105 | spin_lock(&swap_avail_lock); |
1106 | nextsi: | |
adfab836 DS |
1107 | /* |
1108 | * if we got here, it's likely that si was almost full before, | |
bb243f7d ML |
1109 | * and since scan_swap_map_slots() can drop the si->lock, |
1110 | * multiple callers probably all tried to get a page from the | |
1111 | * same si and it filled up before we could get one; or, the si | |
1112 | * filled up between us dropping swap_avail_lock and taking | |
1113 | * si->lock. Since we dropped the swap_avail_lock, the | |
1114 | * swap_avail_head list may have been modified; so if next is | |
1115 | * still in the swap_avail_head list then try it, otherwise | |
1116 | * start over if we have not gotten any slots. | |
adfab836 | 1117 | */ |
a2468cc9 | 1118 | if (plist_node_empty(&next->avail_lists[node])) |
18ab4d4c | 1119 | goto start_over; |
1da177e4 | 1120 | } |
fb4f88dc | 1121 | |
18ab4d4c DS |
1122 | spin_unlock(&swap_avail_lock); |
1123 | ||
36005bae TC |
1124 | check_out: |
1125 | if (n_ret < n_goal) | |
5d5e8f19 | 1126 | atomic_long_add((long)(n_goal - n_ret) * size, |
38d8b4e6 | 1127 | &nr_swap_pages); |
fb4f88dc | 1128 | noswap: |
36005bae TC |
1129 | return n_ret; |
1130 | } | |
1131 | ||
afba72b1 | 1132 | static struct swap_info_struct *_swap_info_get(swp_entry_t entry) |
1da177e4 | 1133 | { |
73c34b6a | 1134 | struct swap_info_struct *p; |
eb085574 | 1135 | unsigned long offset; |
1da177e4 LT |
1136 | |
1137 | if (!entry.val) | |
1138 | goto out; | |
eb085574 | 1139 | p = swp_swap_info(entry); |
c10d38cc | 1140 | if (!p) |
1da177e4 | 1141 | goto bad_nofile; |
a449bf58 | 1142 | if (data_race(!(p->flags & SWP_USED))) |
1da177e4 LT |
1143 | goto bad_device; |
1144 | offset = swp_offset(entry); | |
1145 | if (offset >= p->max) | |
1146 | goto bad_offset; | |
afba72b1 ML |
1147 | if (data_race(!p->swap_map[swp_offset(entry)])) |
1148 | goto bad_free; | |
1da177e4 LT |
1149 | return p; |
1150 | ||
afba72b1 ML |
1151 | bad_free: |
1152 | pr_err("%s: %s%08lx\n", __func__, Unused_offset, entry.val); | |
1153 | goto out; | |
1da177e4 | 1154 | bad_offset: |
cf532faa | 1155 | pr_err("%s: %s%08lx\n", __func__, Bad_offset, entry.val); |
1da177e4 LT |
1156 | goto out; |
1157 | bad_device: | |
cf532faa | 1158 | pr_err("%s: %s%08lx\n", __func__, Unused_file, entry.val); |
1da177e4 LT |
1159 | goto out; |
1160 | bad_nofile: | |
cf532faa | 1161 | pr_err("%s: %s%08lx\n", __func__, Bad_file, entry.val); |
1da177e4 LT |
1162 | out: |
1163 | return NULL; | |
886bb7e9 | 1164 | } |
1da177e4 | 1165 | |
7c00bafe TC |
1166 | static struct swap_info_struct *swap_info_get_cont(swp_entry_t entry, |
1167 | struct swap_info_struct *q) | |
1168 | { | |
1169 | struct swap_info_struct *p; | |
1170 | ||
1171 | p = _swap_info_get(entry); | |
1172 | ||
1173 | if (p != q) { | |
1174 | if (q != NULL) | |
1175 | spin_unlock(&q->lock); | |
1176 | if (p != NULL) | |
1177 | spin_lock(&p->lock); | |
1178 | } | |
1179 | return p; | |
1180 | } | |
1181 | ||
b32d5f32 HY |
1182 | static unsigned char __swap_entry_free_locked(struct swap_info_struct *p, |
1183 | unsigned long offset, | |
1184 | unsigned char usage) | |
1da177e4 | 1185 | { |
8d69aaee HD |
1186 | unsigned char count; |
1187 | unsigned char has_cache; | |
235b6217 | 1188 | |
253d553b | 1189 | count = p->swap_map[offset]; |
235b6217 | 1190 | |
253d553b HD |
1191 | has_cache = count & SWAP_HAS_CACHE; |
1192 | count &= ~SWAP_HAS_CACHE; | |
355cfa73 | 1193 | |
253d553b | 1194 | if (usage == SWAP_HAS_CACHE) { |
355cfa73 | 1195 | VM_BUG_ON(!has_cache); |
253d553b | 1196 | has_cache = 0; |
aaa46865 HD |
1197 | } else if (count == SWAP_MAP_SHMEM) { |
1198 | /* | |
1199 | * Or we could insist on shmem.c using a special | |
1200 | * swap_shmem_free() and free_shmem_swap_and_cache()... | |
1201 | */ | |
1202 | count = 0; | |
570a335b HD |
1203 | } else if ((count & ~COUNT_CONTINUED) <= SWAP_MAP_MAX) { |
1204 | if (count == COUNT_CONTINUED) { | |
1205 | if (swap_count_continued(p, offset, count)) | |
1206 | count = SWAP_MAP_MAX | COUNT_CONTINUED; | |
1207 | else | |
1208 | count = SWAP_MAP_MAX; | |
1209 | } else | |
1210 | count--; | |
1211 | } | |
253d553b | 1212 | |
253d553b | 1213 | usage = count | has_cache; |
a449bf58 QC |
1214 | if (usage) |
1215 | WRITE_ONCE(p->swap_map[offset], usage); | |
1216 | else | |
1217 | WRITE_ONCE(p->swap_map[offset], SWAP_HAS_CACHE); | |
7c00bafe | 1218 | |
b32d5f32 HY |
1219 | return usage; |
1220 | } | |
1221 | ||
eb085574 HY |
1222 | /* |
1223 | * Check whether swap entry is valid in the swap device. If so, | |
1224 | * return pointer to swap_info_struct, and keep the swap entry valid | |
1225 | * via preventing the swap device from being swapoff, until | |
1226 | * put_swap_device() is called. Otherwise return NULL. | |
1227 | * | |
eb085574 | 1228 | * Notice that swapoff or swapoff+swapon can still happen before the |
63d8620e ML |
1229 | * percpu_ref_tryget_live() in get_swap_device() or after the |
1230 | * percpu_ref_put() in put_swap_device() if there isn't any other way | |
1231 | * to prevent swapoff, such as page lock, page table lock, etc. The | |
1232 | * caller must be prepared for that. For example, the following | |
1233 | * situation is possible. | |
eb085574 HY |
1234 | * |
1235 | * CPU1 CPU2 | |
1236 | * do_swap_page() | |
1237 | * ... swapoff+swapon | |
1238 | * __read_swap_cache_async() | |
1239 | * swapcache_prepare() | |
1240 | * __swap_duplicate() | |
1241 | * // check swap_map | |
1242 | * // verify PTE not changed | |
1243 | * | |
1244 | * In __swap_duplicate(), the swap_map need to be checked before | |
1245 | * changing partly because the specified swap entry may be for another | |
1246 | * swap device which has been swapoff. And in do_swap_page(), after | |
1247 | * the page is read from the swap device, the PTE is verified not | |
1248 | * changed with the page table locked to check whether the swap device | |
1249 | * has been swapoff or swapoff+swapon. | |
1250 | */ | |
1251 | struct swap_info_struct *get_swap_device(swp_entry_t entry) | |
1252 | { | |
1253 | struct swap_info_struct *si; | |
1254 | unsigned long offset; | |
1255 | ||
1256 | if (!entry.val) | |
1257 | goto out; | |
1258 | si = swp_swap_info(entry); | |
1259 | if (!si) | |
1260 | goto bad_nofile; | |
63d8620e ML |
1261 | if (!percpu_ref_tryget_live(&si->users)) |
1262 | goto out; | |
1263 | /* | |
1264 | * Guarantee the si->users are checked before accessing other | |
1265 | * fields of swap_info_struct. | |
1266 | * | |
1267 | * Paired with the spin_unlock() after setup_swap_info() in | |
1268 | * enable_swap_info(). | |
1269 | */ | |
1270 | smp_rmb(); | |
eb085574 HY |
1271 | offset = swp_offset(entry); |
1272 | if (offset >= si->max) | |
63d8620e | 1273 | goto put_out; |
eb085574 HY |
1274 | |
1275 | return si; | |
1276 | bad_nofile: | |
1277 | pr_err("%s: %s%08lx\n", __func__, Bad_file, entry.val); | |
1278 | out: | |
1279 | return NULL; | |
63d8620e | 1280 | put_out: |
23b230ba | 1281 | pr_err("%s: %s%08lx\n", __func__, Bad_offset, entry.val); |
63d8620e | 1282 | percpu_ref_put(&si->users); |
eb085574 HY |
1283 | return NULL; |
1284 | } | |
1285 | ||
b32d5f32 | 1286 | static unsigned char __swap_entry_free(struct swap_info_struct *p, |
33e16272 | 1287 | swp_entry_t entry) |
b32d5f32 HY |
1288 | { |
1289 | struct swap_cluster_info *ci; | |
1290 | unsigned long offset = swp_offset(entry); | |
33e16272 | 1291 | unsigned char usage; |
b32d5f32 HY |
1292 | |
1293 | ci = lock_cluster_or_swap_info(p, offset); | |
33e16272 | 1294 | usage = __swap_entry_free_locked(p, offset, 1); |
7c00bafe | 1295 | unlock_cluster_or_swap_info(p, ci); |
10e364da HY |
1296 | if (!usage) |
1297 | free_swap_slot(entry); | |
7c00bafe TC |
1298 | |
1299 | return usage; | |
1300 | } | |
355cfa73 | 1301 | |
7c00bafe TC |
1302 | static void swap_entry_free(struct swap_info_struct *p, swp_entry_t entry) |
1303 | { | |
1304 | struct swap_cluster_info *ci; | |
1305 | unsigned long offset = swp_offset(entry); | |
1306 | unsigned char count; | |
1307 | ||
1308 | ci = lock_cluster(p, offset); | |
1309 | count = p->swap_map[offset]; | |
1310 | VM_BUG_ON(count != SWAP_HAS_CACHE); | |
1311 | p->swap_map[offset] = 0; | |
1312 | dec_cluster_info_page(p, p->cluster_info, offset); | |
235b6217 HY |
1313 | unlock_cluster(ci); |
1314 | ||
38d8b4e6 HY |
1315 | mem_cgroup_uncharge_swap(entry, 1); |
1316 | swap_range_free(p, offset, 1); | |
1da177e4 LT |
1317 | } |
1318 | ||
1319 | /* | |
2de1a7e4 | 1320 | * Caller has made sure that the swap device corresponding to entry |
1da177e4 LT |
1321 | * is still around or has not been recycled. |
1322 | */ | |
1323 | void swap_free(swp_entry_t entry) | |
1324 | { | |
73c34b6a | 1325 | struct swap_info_struct *p; |
1da177e4 | 1326 | |
235b6217 | 1327 | p = _swap_info_get(entry); |
10e364da | 1328 | if (p) |
33e16272 | 1329 | __swap_entry_free(p, entry); |
1da177e4 LT |
1330 | } |
1331 | ||
cb4b86ba KH |
1332 | /* |
1333 | * Called after dropping swapcache to decrease refcnt to swap entries. | |
1334 | */ | |
a448f2d0 | 1335 | void put_swap_page(struct page *page, swp_entry_t entry) |
38d8b4e6 HY |
1336 | { |
1337 | unsigned long offset = swp_offset(entry); | |
1338 | unsigned long idx = offset / SWAPFILE_CLUSTER; | |
1339 | struct swap_cluster_info *ci; | |
1340 | struct swap_info_struct *si; | |
1341 | unsigned char *map; | |
a3aea839 HY |
1342 | unsigned int i, free_entries = 0; |
1343 | unsigned char val; | |
6c357848 | 1344 | int size = swap_entry_size(thp_nr_pages(page)); |
fe5266d5 | 1345 | |
a3aea839 | 1346 | si = _swap_info_get(entry); |
38d8b4e6 HY |
1347 | if (!si) |
1348 | return; | |
1349 | ||
c2343d27 | 1350 | ci = lock_cluster_or_swap_info(si, offset); |
a448f2d0 | 1351 | if (size == SWAPFILE_CLUSTER) { |
a448f2d0 HY |
1352 | VM_BUG_ON(!cluster_is_huge(ci)); |
1353 | map = si->swap_map + offset; | |
1354 | for (i = 0; i < SWAPFILE_CLUSTER; i++) { | |
1355 | val = map[i]; | |
1356 | VM_BUG_ON(!(val & SWAP_HAS_CACHE)); | |
1357 | if (val == SWAP_HAS_CACHE) | |
1358 | free_entries++; | |
1359 | } | |
a448f2d0 | 1360 | cluster_clear_huge(ci); |
a448f2d0 | 1361 | if (free_entries == SWAPFILE_CLUSTER) { |
c2343d27 | 1362 | unlock_cluster_or_swap_info(si, ci); |
a448f2d0 | 1363 | spin_lock(&si->lock); |
a448f2d0 HY |
1364 | mem_cgroup_uncharge_swap(entry, SWAPFILE_CLUSTER); |
1365 | swap_free_cluster(si, idx); | |
1366 | spin_unlock(&si->lock); | |
1367 | return; | |
1368 | } | |
1369 | } | |
c2343d27 HY |
1370 | for (i = 0; i < size; i++, entry.val++) { |
1371 | if (!__swap_entry_free_locked(si, offset + i, SWAP_HAS_CACHE)) { | |
1372 | unlock_cluster_or_swap_info(si, ci); | |
1373 | free_swap_slot(entry); | |
1374 | if (i == size - 1) | |
1375 | return; | |
1376 | lock_cluster_or_swap_info(si, offset); | |
a3aea839 HY |
1377 | } |
1378 | } | |
c2343d27 | 1379 | unlock_cluster_or_swap_info(si, ci); |
38d8b4e6 | 1380 | } |
59807685 | 1381 | |
fe5266d5 | 1382 | #ifdef CONFIG_THP_SWAP |
59807685 HY |
1383 | int split_swap_cluster(swp_entry_t entry) |
1384 | { | |
1385 | struct swap_info_struct *si; | |
1386 | struct swap_cluster_info *ci; | |
1387 | unsigned long offset = swp_offset(entry); | |
1388 | ||
1389 | si = _swap_info_get(entry); | |
1390 | if (!si) | |
1391 | return -EBUSY; | |
1392 | ci = lock_cluster(si, offset); | |
1393 | cluster_clear_huge(ci); | |
1394 | unlock_cluster(ci); | |
1395 | return 0; | |
1396 | } | |
fe5266d5 | 1397 | #endif |
38d8b4e6 | 1398 | |
155b5f88 HY |
1399 | static int swp_entry_cmp(const void *ent1, const void *ent2) |
1400 | { | |
1401 | const swp_entry_t *e1 = ent1, *e2 = ent2; | |
1402 | ||
1403 | return (int)swp_type(*e1) - (int)swp_type(*e2); | |
1404 | } | |
1405 | ||
7c00bafe TC |
1406 | void swapcache_free_entries(swp_entry_t *entries, int n) |
1407 | { | |
1408 | struct swap_info_struct *p, *prev; | |
1409 | int i; | |
1410 | ||
1411 | if (n <= 0) | |
1412 | return; | |
1413 | ||
1414 | prev = NULL; | |
1415 | p = NULL; | |
155b5f88 HY |
1416 | |
1417 | /* | |
1418 | * Sort swap entries by swap device, so each lock is only taken once. | |
1419 | * nr_swapfiles isn't absolutely correct, but the overhead of sort() is | |
1420 | * so low that it isn't necessary to optimize further. | |
1421 | */ | |
1422 | if (nr_swapfiles > 1) | |
1423 | sort(entries, n, sizeof(entries[0]), swp_entry_cmp, NULL); | |
7c00bafe TC |
1424 | for (i = 0; i < n; ++i) { |
1425 | p = swap_info_get_cont(entries[i], prev); | |
1426 | if (p) | |
1427 | swap_entry_free(p, entries[i]); | |
7c00bafe TC |
1428 | prev = p; |
1429 | } | |
235b6217 | 1430 | if (p) |
7c00bafe | 1431 | spin_unlock(&p->lock); |
cb4b86ba KH |
1432 | } |
1433 | ||
eb085574 | 1434 | int __swap_count(swp_entry_t entry) |
aa8d22a1 | 1435 | { |
eb085574 | 1436 | struct swap_info_struct *si; |
aa8d22a1 | 1437 | pgoff_t offset = swp_offset(entry); |
eb085574 | 1438 | int count = 0; |
aa8d22a1 | 1439 | |
eb085574 HY |
1440 | si = get_swap_device(entry); |
1441 | if (si) { | |
1442 | count = swap_count(si->swap_map[offset]); | |
1443 | put_swap_device(si); | |
1444 | } | |
1445 | return count; | |
aa8d22a1 MK |
1446 | } |
1447 | ||
14d01ee9 MWO |
1448 | /* |
1449 | * How many references to @entry are currently swapped out? | |
1450 | * This does not give an exact answer when swap count is continued, | |
1451 | * but does include the high COUNT_CONTINUED flag to allow for that. | |
1452 | */ | |
322b8afe HY |
1453 | static int swap_swapcount(struct swap_info_struct *si, swp_entry_t entry) |
1454 | { | |
322b8afe HY |
1455 | pgoff_t offset = swp_offset(entry); |
1456 | struct swap_cluster_info *ci; | |
14d01ee9 | 1457 | int count; |
322b8afe HY |
1458 | |
1459 | ci = lock_cluster_or_swap_info(si, offset); | |
1460 | count = swap_count(si->swap_map[offset]); | |
1461 | unlock_cluster_or_swap_info(si, ci); | |
1462 | return count; | |
1463 | } | |
1464 | ||
e8c26ab6 TC |
1465 | /* |
1466 | * How many references to @entry are currently swapped out? | |
1467 | * This does not give an exact answer when swap count is continued, | |
1468 | * but does include the high COUNT_CONTINUED flag to allow for that. | |
1469 | */ | |
1470 | int __swp_swapcount(swp_entry_t entry) | |
1471 | { | |
1472 | int count = 0; | |
e8c26ab6 | 1473 | struct swap_info_struct *si; |
e8c26ab6 | 1474 | |
eb085574 HY |
1475 | si = get_swap_device(entry); |
1476 | if (si) { | |
322b8afe | 1477 | count = swap_swapcount(si, entry); |
eb085574 HY |
1478 | put_swap_device(si); |
1479 | } | |
e8c26ab6 TC |
1480 | return count; |
1481 | } | |
1482 | ||
8334b962 MK |
1483 | /* |
1484 | * How many references to @entry are currently swapped out? | |
1485 | * This considers COUNT_CONTINUED so it returns exact answer. | |
1486 | */ | |
1487 | int swp_swapcount(swp_entry_t entry) | |
1488 | { | |
1489 | int count, tmp_count, n; | |
1490 | struct swap_info_struct *p; | |
235b6217 | 1491 | struct swap_cluster_info *ci; |
8334b962 MK |
1492 | struct page *page; |
1493 | pgoff_t offset; | |
1494 | unsigned char *map; | |
1495 | ||
235b6217 | 1496 | p = _swap_info_get(entry); |
8334b962 MK |
1497 | if (!p) |
1498 | return 0; | |
1499 | ||
235b6217 HY |
1500 | offset = swp_offset(entry); |
1501 | ||
1502 | ci = lock_cluster_or_swap_info(p, offset); | |
1503 | ||
1504 | count = swap_count(p->swap_map[offset]); | |
8334b962 MK |
1505 | if (!(count & COUNT_CONTINUED)) |
1506 | goto out; | |
1507 | ||
1508 | count &= ~COUNT_CONTINUED; | |
1509 | n = SWAP_MAP_MAX + 1; | |
1510 | ||
8334b962 MK |
1511 | page = vmalloc_to_page(p->swap_map + offset); |
1512 | offset &= ~PAGE_MASK; | |
1513 | VM_BUG_ON(page_private(page) != SWP_CONTINUED); | |
1514 | ||
1515 | do { | |
a8ae4991 | 1516 | page = list_next_entry(page, lru); |
8334b962 MK |
1517 | map = kmap_atomic(page); |
1518 | tmp_count = map[offset]; | |
1519 | kunmap_atomic(map); | |
1520 | ||
1521 | count += (tmp_count & ~COUNT_CONTINUED) * n; | |
1522 | n *= (SWAP_CONT_MAX + 1); | |
1523 | } while (tmp_count & COUNT_CONTINUED); | |
1524 | out: | |
235b6217 | 1525 | unlock_cluster_or_swap_info(p, ci); |
8334b962 MK |
1526 | return count; |
1527 | } | |
1528 | ||
e0709829 HY |
1529 | static bool swap_page_trans_huge_swapped(struct swap_info_struct *si, |
1530 | swp_entry_t entry) | |
1531 | { | |
1532 | struct swap_cluster_info *ci; | |
1533 | unsigned char *map = si->swap_map; | |
1534 | unsigned long roffset = swp_offset(entry); | |
1535 | unsigned long offset = round_down(roffset, SWAPFILE_CLUSTER); | |
1536 | int i; | |
1537 | bool ret = false; | |
1538 | ||
1539 | ci = lock_cluster_or_swap_info(si, offset); | |
1540 | if (!ci || !cluster_is_huge(ci)) { | |
afa4711e | 1541 | if (swap_count(map[roffset])) |
e0709829 HY |
1542 | ret = true; |
1543 | goto unlock_out; | |
1544 | } | |
1545 | for (i = 0; i < SWAPFILE_CLUSTER; i++) { | |
afa4711e | 1546 | if (swap_count(map[offset + i])) { |
e0709829 HY |
1547 | ret = true; |
1548 | break; | |
1549 | } | |
1550 | } | |
1551 | unlock_out: | |
1552 | unlock_cluster_or_swap_info(si, ci); | |
1553 | return ret; | |
1554 | } | |
1555 | ||
2397f780 | 1556 | static bool folio_swapped(struct folio *folio) |
e0709829 | 1557 | { |
14d01ee9 MWO |
1558 | swp_entry_t entry = folio_swap_entry(folio); |
1559 | struct swap_info_struct *si = _swap_info_get(entry); | |
1560 | ||
1561 | if (!si) | |
1562 | return false; | |
e0709829 | 1563 | |
2397f780 | 1564 | if (!IS_ENABLED(CONFIG_THP_SWAP) || likely(!folio_test_large(folio))) |
14d01ee9 | 1565 | return swap_swapcount(si, entry) != 0; |
e0709829 | 1566 | |
14d01ee9 | 1567 | return swap_page_trans_huge_swapped(si, entry); |
e0709829 | 1568 | } |
ba3c4ce6 | 1569 | |
bdb0ed54 MWO |
1570 | /** |
1571 | * folio_free_swap() - Free the swap space used for this folio. | |
1572 | * @folio: The folio to remove. | |
1573 | * | |
1574 | * If swap is getting full, or if there are no more mappings of this folio, | |
1575 | * then call folio_free_swap to free its swap space. | |
1576 | * | |
1577 | * Return: true if we were able to release the swap space. | |
1da177e4 | 1578 | */ |
bdb0ed54 | 1579 | bool folio_free_swap(struct folio *folio) |
1da177e4 | 1580 | { |
2397f780 | 1581 | VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio); |
1da177e4 | 1582 | |
2397f780 | 1583 | if (!folio_test_swapcache(folio)) |
bdb0ed54 | 1584 | return false; |
2397f780 | 1585 | if (folio_test_writeback(folio)) |
bdb0ed54 | 1586 | return false; |
2397f780 | 1587 | if (folio_swapped(folio)) |
bdb0ed54 | 1588 | return false; |
1da177e4 | 1589 | |
b73d7fce HD |
1590 | /* |
1591 | * Once hibernation has begun to create its image of memory, | |
bdb0ed54 | 1592 | * there's a danger that one of the calls to folio_free_swap() |
b73d7fce HD |
1593 | * - most probably a call from __try_to_reclaim_swap() while |
1594 | * hibernation is allocating its own swap pages for the image, | |
1595 | * but conceivably even a call from memory reclaim - will free | |
bdb0ed54 MWO |
1596 | * the swap from a folio which has already been recorded in the |
1597 | * image as a clean swapcache folio, and then reuse its swap for | |
b73d7fce | 1598 | * another page of the image. On waking from hibernation, the |
bdb0ed54 | 1599 | * original folio might be freed under memory pressure, then |
b73d7fce HD |
1600 | * later read back in from swap, now with the wrong data. |
1601 | * | |
2de1a7e4 | 1602 | * Hibernation suspends storage while it is writing the image |
f90ac398 | 1603 | * to disk so check that here. |
b73d7fce | 1604 | */ |
f90ac398 | 1605 | if (pm_suspended_storage()) |
bdb0ed54 | 1606 | return false; |
b73d7fce | 1607 | |
75fa68a5 | 1608 | delete_from_swap_cache(folio); |
2397f780 | 1609 | folio_set_dirty(folio); |
bdb0ed54 | 1610 | return true; |
68a22394 RR |
1611 | } |
1612 | ||
1da177e4 LT |
1613 | /* |
1614 | * Free the swap entry like above, but also try to | |
1615 | * free the page cache entry if it is the last user. | |
1616 | */ | |
2509ef26 | 1617 | int free_swap_and_cache(swp_entry_t entry) |
1da177e4 | 1618 | { |
2509ef26 | 1619 | struct swap_info_struct *p; |
7c00bafe | 1620 | unsigned char count; |
1da177e4 | 1621 | |
a7420aa5 | 1622 | if (non_swap_entry(entry)) |
2509ef26 | 1623 | return 1; |
0697212a | 1624 | |
7c00bafe | 1625 | p = _swap_info_get(entry); |
1da177e4 | 1626 | if (p) { |
33e16272 | 1627 | count = __swap_entry_free(p, entry); |
e0709829 | 1628 | if (count == SWAP_HAS_CACHE && |
bcd49e86 HY |
1629 | !swap_page_trans_huge_swapped(p, entry)) |
1630 | __try_to_reclaim_swap(p, swp_offset(entry), | |
1631 | TTRS_UNMAPPED | TTRS_FULL); | |
1da177e4 | 1632 | } |
2509ef26 | 1633 | return p != NULL; |
1da177e4 LT |
1634 | } |
1635 | ||
b0cb1a19 | 1636 | #ifdef CONFIG_HIBERNATION |
bb243f7d ML |
1637 | |
1638 | swp_entry_t get_swap_page_of_type(int type) | |
1639 | { | |
1640 | struct swap_info_struct *si = swap_type_to_swap_info(type); | |
1641 | swp_entry_t entry = {0}; | |
1642 | ||
1643 | if (!si) | |
1644 | goto fail; | |
1645 | ||
1646 | /* This is called for allocating swap entry, not cache */ | |
1647 | spin_lock(&si->lock); | |
1648 | if ((si->flags & SWP_WRITEOK) && scan_swap_map_slots(si, 1, 1, &entry)) | |
1649 | atomic_long_dec(&nr_swap_pages); | |
1650 | spin_unlock(&si->lock); | |
1651 | fail: | |
1652 | return entry; | |
1653 | } | |
1654 | ||
f577eb30 | 1655 | /* |
915bae9e | 1656 | * Find the swap type that corresponds to given device (if any). |
f577eb30 | 1657 | * |
915bae9e RW |
1658 | * @offset - number of the PAGE_SIZE-sized block of the device, starting |
1659 | * from 0, in which the swap header is expected to be located. | |
1660 | * | |
1661 | * This is needed for the suspend to disk (aka swsusp). | |
f577eb30 | 1662 | */ |
21bd9005 | 1663 | int swap_type_of(dev_t device, sector_t offset) |
f577eb30 | 1664 | { |
efa90a98 | 1665 | int type; |
f577eb30 | 1666 | |
21bd9005 CH |
1667 | if (!device) |
1668 | return -1; | |
915bae9e | 1669 | |
f577eb30 | 1670 | spin_lock(&swap_lock); |
efa90a98 HD |
1671 | for (type = 0; type < nr_swapfiles; type++) { |
1672 | struct swap_info_struct *sis = swap_info[type]; | |
f577eb30 | 1673 | |
915bae9e | 1674 | if (!(sis->flags & SWP_WRITEOK)) |
f577eb30 | 1675 | continue; |
b6b5bce3 | 1676 | |
21bd9005 | 1677 | if (device == sis->bdev->bd_dev) { |
4efaceb1 | 1678 | struct swap_extent *se = first_se(sis); |
915bae9e | 1679 | |
915bae9e RW |
1680 | if (se->start_block == offset) { |
1681 | spin_unlock(&swap_lock); | |
efa90a98 | 1682 | return type; |
915bae9e | 1683 | } |
f577eb30 RW |
1684 | } |
1685 | } | |
1686 | spin_unlock(&swap_lock); | |
21bd9005 CH |
1687 | return -ENODEV; |
1688 | } | |
915bae9e | 1689 | |
21bd9005 CH |
1690 | int find_first_swap(dev_t *device) |
1691 | { | |
1692 | int type; | |
915bae9e | 1693 | |
21bd9005 CH |
1694 | spin_lock(&swap_lock); |
1695 | for (type = 0; type < nr_swapfiles; type++) { | |
1696 | struct swap_info_struct *sis = swap_info[type]; | |
1697 | ||
1698 | if (!(sis->flags & SWP_WRITEOK)) | |
1699 | continue; | |
1700 | *device = sis->bdev->bd_dev; | |
1701 | spin_unlock(&swap_lock); | |
1702 | return type; | |
1703 | } | |
1704 | spin_unlock(&swap_lock); | |
f577eb30 RW |
1705 | return -ENODEV; |
1706 | } | |
1707 | ||
73c34b6a HD |
1708 | /* |
1709 | * Get the (PAGE_SIZE) block corresponding to given offset on the swapdev | |
1710 | * corresponding to given index in swap_info (swap type). | |
1711 | */ | |
1712 | sector_t swapdev_block(int type, pgoff_t offset) | |
1713 | { | |
c10d38cc | 1714 | struct swap_info_struct *si = swap_type_to_swap_info(type); |
f885056a | 1715 | struct swap_extent *se; |
73c34b6a | 1716 | |
c10d38cc | 1717 | if (!si || !(si->flags & SWP_WRITEOK)) |
73c34b6a | 1718 | return 0; |
f885056a CH |
1719 | se = offset_to_swap_extent(si, offset); |
1720 | return se->start_block + (offset - se->start_page); | |
73c34b6a HD |
1721 | } |
1722 | ||
f577eb30 RW |
1723 | /* |
1724 | * Return either the total number of swap pages of given type, or the number | |
1725 | * of free pages of that type (depending on @free) | |
1726 | * | |
1727 | * This is needed for software suspend | |
1728 | */ | |
1729 | unsigned int count_swap_pages(int type, int free) | |
1730 | { | |
1731 | unsigned int n = 0; | |
1732 | ||
efa90a98 HD |
1733 | spin_lock(&swap_lock); |
1734 | if ((unsigned int)type < nr_swapfiles) { | |
1735 | struct swap_info_struct *sis = swap_info[type]; | |
1736 | ||
ec8acf20 | 1737 | spin_lock(&sis->lock); |
efa90a98 HD |
1738 | if (sis->flags & SWP_WRITEOK) { |
1739 | n = sis->pages; | |
f577eb30 | 1740 | if (free) |
efa90a98 | 1741 | n -= sis->inuse_pages; |
f577eb30 | 1742 | } |
ec8acf20 | 1743 | spin_unlock(&sis->lock); |
f577eb30 | 1744 | } |
efa90a98 | 1745 | spin_unlock(&swap_lock); |
f577eb30 RW |
1746 | return n; |
1747 | } | |
73c34b6a | 1748 | #endif /* CONFIG_HIBERNATION */ |
f577eb30 | 1749 | |
9f8bdb3f | 1750 | static inline int pte_same_as_swp(pte_t pte, pte_t swp_pte) |
179ef71c | 1751 | { |
099dd687 | 1752 | return pte_same(pte_swp_clear_flags(pte), swp_pte); |
179ef71c CG |
1753 | } |
1754 | ||
1da177e4 | 1755 | /* |
72866f6f HD |
1756 | * No need to decide whether this PTE shares the swap entry with others, |
1757 | * just let do_wp_page work it out if a write is requested later - to | |
1758 | * force COW, vm_page_prot omits write permission from any private vma. | |
1da177e4 | 1759 | */ |
044d66c1 | 1760 | static int unuse_pte(struct vm_area_struct *vma, pmd_t *pmd, |
1da177e4 LT |
1761 | unsigned long addr, swp_entry_t entry, struct page *page) |
1762 | { | |
9e16b7fb | 1763 | struct page *swapcache; |
044d66c1 | 1764 | spinlock_t *ptl; |
14a762dd | 1765 | pte_t *pte, new_pte; |
044d66c1 HD |
1766 | int ret = 1; |
1767 | ||
9e16b7fb HD |
1768 | swapcache = page; |
1769 | page = ksm_might_need_to_copy(page, vma, addr); | |
1770 | if (unlikely(!page)) | |
1771 | return -ENOMEM; | |
1772 | ||
044d66c1 | 1773 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
9f8bdb3f | 1774 | if (unlikely(!pte_same_as_swp(*pte, swp_entry_to_pte(entry)))) { |
044d66c1 HD |
1775 | ret = 0; |
1776 | goto out; | |
1777 | } | |
8a9f3ccd | 1778 | |
9f186f9e ML |
1779 | if (unlikely(!PageUptodate(page))) { |
1780 | pte_t pteval; | |
1781 | ||
1782 | dec_mm_counter(vma->vm_mm, MM_SWAPENTS); | |
1783 | pteval = swp_entry_to_pte(make_swapin_error_entry(page)); | |
1784 | set_pte_at(vma->vm_mm, addr, pte, pteval); | |
1785 | swap_free(entry); | |
1786 | ret = 0; | |
1787 | goto out; | |
1788 | } | |
1789 | ||
78fbe906 DH |
1790 | /* See do_swap_page() */ |
1791 | BUG_ON(!PageAnon(page) && PageMappedToDisk(page)); | |
1792 | BUG_ON(PageAnon(page) && PageAnonExclusive(page)); | |
1793 | ||
b084d435 | 1794 | dec_mm_counter(vma->vm_mm, MM_SWAPENTS); |
d559db08 | 1795 | inc_mm_counter(vma->vm_mm, MM_ANONPAGES); |
1da177e4 | 1796 | get_page(page); |
00501b53 | 1797 | if (page == swapcache) { |
1493a191 DH |
1798 | rmap_t rmap_flags = RMAP_NONE; |
1799 | ||
1800 | /* | |
1801 | * See do_swap_page(): PageWriteback() would be problematic. | |
1802 | * However, we do a wait_on_page_writeback() just before this | |
1803 | * call and have the page locked. | |
1804 | */ | |
1805 | VM_BUG_ON_PAGE(PageWriteback(page), page); | |
1806 | if (pte_swp_exclusive(*pte)) | |
1807 | rmap_flags |= RMAP_EXCLUSIVE; | |
1808 | ||
1809 | page_add_anon_rmap(page, vma, addr, rmap_flags); | |
00501b53 | 1810 | } else { /* ksm created a completely new copy */ |
40f2bbf7 | 1811 | page_add_new_anon_rmap(page, vma, addr); |
b518154e | 1812 | lru_cache_add_inactive_or_unevictable(page, vma); |
00501b53 | 1813 | } |
14a762dd ML |
1814 | new_pte = pte_mkold(mk_pte(page, vma->vm_page_prot)); |
1815 | if (pte_swp_soft_dirty(*pte)) | |
1816 | new_pte = pte_mksoft_dirty(new_pte); | |
1817 | if (pte_swp_uffd_wp(*pte)) | |
1818 | new_pte = pte_mkuffd_wp(new_pte); | |
1819 | set_pte_at(vma->vm_mm, addr, pte, new_pte); | |
1da177e4 | 1820 | swap_free(entry); |
044d66c1 HD |
1821 | out: |
1822 | pte_unmap_unlock(pte, ptl); | |
9e16b7fb HD |
1823 | if (page != swapcache) { |
1824 | unlock_page(page); | |
1825 | put_page(page); | |
1826 | } | |
044d66c1 | 1827 | return ret; |
1da177e4 LT |
1828 | } |
1829 | ||
1830 | static int unuse_pte_range(struct vm_area_struct *vma, pmd_t *pmd, | |
b56a2d8a | 1831 | unsigned long addr, unsigned long end, |
10a9c496 | 1832 | unsigned int type) |
1da177e4 | 1833 | { |
b56a2d8a VRP |
1834 | struct page *page; |
1835 | swp_entry_t entry; | |
705e87c0 | 1836 | pte_t *pte; |
b56a2d8a VRP |
1837 | struct swap_info_struct *si; |
1838 | unsigned long offset; | |
8a9f3ccd | 1839 | int ret = 0; |
b56a2d8a | 1840 | volatile unsigned char *swap_map; |
1da177e4 | 1841 | |
b56a2d8a | 1842 | si = swap_info[type]; |
044d66c1 | 1843 | pte = pte_offset_map(pmd, addr); |
1da177e4 | 1844 | do { |
b56a2d8a VRP |
1845 | if (!is_swap_pte(*pte)) |
1846 | continue; | |
1847 | ||
1848 | entry = pte_to_swp_entry(*pte); | |
1849 | if (swp_type(entry) != type) | |
1850 | continue; | |
1851 | ||
1852 | offset = swp_offset(entry); | |
b56a2d8a VRP |
1853 | pte_unmap(pte); |
1854 | swap_map = &si->swap_map[offset]; | |
ebc5951e AR |
1855 | page = lookup_swap_cache(entry, vma, addr); |
1856 | if (!page) { | |
8c63ca5b WD |
1857 | struct vm_fault vmf = { |
1858 | .vma = vma, | |
1859 | .address = addr, | |
824ddc60 | 1860 | .real_address = addr, |
8c63ca5b WD |
1861 | .pmd = pmd, |
1862 | }; | |
1863 | ||
ebc5951e AR |
1864 | page = swapin_readahead(entry, GFP_HIGHUSER_MOVABLE, |
1865 | &vmf); | |
1866 | } | |
b56a2d8a VRP |
1867 | if (!page) { |
1868 | if (*swap_map == 0 || *swap_map == SWAP_MAP_BAD) | |
1869 | goto try_next; | |
1870 | return -ENOMEM; | |
1871 | } | |
1872 | ||
1873 | lock_page(page); | |
1874 | wait_on_page_writeback(page); | |
1875 | ret = unuse_pte(vma, pmd, addr, entry, page); | |
1876 | if (ret < 0) { | |
1877 | unlock_page(page); | |
1878 | put_page(page); | |
1879 | goto out; | |
1880 | } | |
1881 | ||
1882 | try_to_free_swap(page); | |
1883 | unlock_page(page); | |
1884 | put_page(page); | |
b56a2d8a VRP |
1885 | try_next: |
1886 | pte = pte_offset_map(pmd, addr); | |
1da177e4 | 1887 | } while (pte++, addr += PAGE_SIZE, addr != end); |
044d66c1 | 1888 | pte_unmap(pte - 1); |
b56a2d8a VRP |
1889 | |
1890 | ret = 0; | |
044d66c1 | 1891 | out: |
8a9f3ccd | 1892 | return ret; |
1da177e4 LT |
1893 | } |
1894 | ||
1895 | static inline int unuse_pmd_range(struct vm_area_struct *vma, pud_t *pud, | |
1896 | unsigned long addr, unsigned long end, | |
10a9c496 | 1897 | unsigned int type) |
1da177e4 LT |
1898 | { |
1899 | pmd_t *pmd; | |
1900 | unsigned long next; | |
8a9f3ccd | 1901 | int ret; |
1da177e4 LT |
1902 | |
1903 | pmd = pmd_offset(pud, addr); | |
1904 | do { | |
dc644a07 | 1905 | cond_resched(); |
1da177e4 | 1906 | next = pmd_addr_end(addr, end); |
1a5a9906 | 1907 | if (pmd_none_or_trans_huge_or_clear_bad(pmd)) |
1da177e4 | 1908 | continue; |
10a9c496 | 1909 | ret = unuse_pte_range(vma, pmd, addr, next, type); |
8a9f3ccd BS |
1910 | if (ret) |
1911 | return ret; | |
1da177e4 LT |
1912 | } while (pmd++, addr = next, addr != end); |
1913 | return 0; | |
1914 | } | |
1915 | ||
c2febafc | 1916 | static inline int unuse_pud_range(struct vm_area_struct *vma, p4d_t *p4d, |
1da177e4 | 1917 | unsigned long addr, unsigned long end, |
10a9c496 | 1918 | unsigned int type) |
1da177e4 LT |
1919 | { |
1920 | pud_t *pud; | |
1921 | unsigned long next; | |
8a9f3ccd | 1922 | int ret; |
1da177e4 | 1923 | |
c2febafc | 1924 | pud = pud_offset(p4d, addr); |
1da177e4 LT |
1925 | do { |
1926 | next = pud_addr_end(addr, end); | |
1927 | if (pud_none_or_clear_bad(pud)) | |
1928 | continue; | |
10a9c496 | 1929 | ret = unuse_pmd_range(vma, pud, addr, next, type); |
8a9f3ccd BS |
1930 | if (ret) |
1931 | return ret; | |
1da177e4 LT |
1932 | } while (pud++, addr = next, addr != end); |
1933 | return 0; | |
1934 | } | |
1935 | ||
c2febafc KS |
1936 | static inline int unuse_p4d_range(struct vm_area_struct *vma, pgd_t *pgd, |
1937 | unsigned long addr, unsigned long end, | |
10a9c496 | 1938 | unsigned int type) |
c2febafc KS |
1939 | { |
1940 | p4d_t *p4d; | |
1941 | unsigned long next; | |
1942 | int ret; | |
1943 | ||
1944 | p4d = p4d_offset(pgd, addr); | |
1945 | do { | |
1946 | next = p4d_addr_end(addr, end); | |
1947 | if (p4d_none_or_clear_bad(p4d)) | |
1948 | continue; | |
10a9c496 | 1949 | ret = unuse_pud_range(vma, p4d, addr, next, type); |
c2febafc KS |
1950 | if (ret) |
1951 | return ret; | |
1952 | } while (p4d++, addr = next, addr != end); | |
1953 | return 0; | |
1954 | } | |
1955 | ||
10a9c496 | 1956 | static int unuse_vma(struct vm_area_struct *vma, unsigned int type) |
1da177e4 LT |
1957 | { |
1958 | pgd_t *pgd; | |
1959 | unsigned long addr, end, next; | |
8a9f3ccd | 1960 | int ret; |
1da177e4 | 1961 | |
b56a2d8a VRP |
1962 | addr = vma->vm_start; |
1963 | end = vma->vm_end; | |
1da177e4 LT |
1964 | |
1965 | pgd = pgd_offset(vma->vm_mm, addr); | |
1966 | do { | |
1967 | next = pgd_addr_end(addr, end); | |
1968 | if (pgd_none_or_clear_bad(pgd)) | |
1969 | continue; | |
10a9c496 | 1970 | ret = unuse_p4d_range(vma, pgd, addr, next, type); |
8a9f3ccd BS |
1971 | if (ret) |
1972 | return ret; | |
1da177e4 LT |
1973 | } while (pgd++, addr = next, addr != end); |
1974 | return 0; | |
1975 | } | |
1976 | ||
10a9c496 | 1977 | static int unuse_mm(struct mm_struct *mm, unsigned int type) |
1da177e4 LT |
1978 | { |
1979 | struct vm_area_struct *vma; | |
8a9f3ccd | 1980 | int ret = 0; |
208c09db | 1981 | VMA_ITERATOR(vmi, mm, 0); |
1da177e4 | 1982 | |
d8ed45c5 | 1983 | mmap_read_lock(mm); |
208c09db | 1984 | for_each_vma(vmi, vma) { |
b56a2d8a | 1985 | if (vma->anon_vma) { |
10a9c496 | 1986 | ret = unuse_vma(vma, type); |
b56a2d8a VRP |
1987 | if (ret) |
1988 | break; | |
1989 | } | |
208c09db | 1990 | |
dc644a07 | 1991 | cond_resched(); |
1da177e4 | 1992 | } |
d8ed45c5 | 1993 | mmap_read_unlock(mm); |
b56a2d8a | 1994 | return ret; |
1da177e4 LT |
1995 | } |
1996 | ||
1997 | /* | |
3c3115ad ML |
1998 | * Scan swap_map from current position to next entry still in use. |
1999 | * Return 0 if there are no inuse entries after prev till end of | |
2000 | * the map. | |
1da177e4 | 2001 | */ |
6eb396dc | 2002 | static unsigned int find_next_to_unuse(struct swap_info_struct *si, |
10a9c496 | 2003 | unsigned int prev) |
1da177e4 | 2004 | { |
b56a2d8a | 2005 | unsigned int i; |
8d69aaee | 2006 | unsigned char count; |
1da177e4 LT |
2007 | |
2008 | /* | |
5d337b91 | 2009 | * No need for swap_lock here: we're just looking |
1da177e4 LT |
2010 | * for whether an entry is in use, not modifying it; false |
2011 | * hits are okay, and sys_swapoff() has already prevented new | |
5d337b91 | 2012 | * allocations from this area (while holding swap_lock). |
1da177e4 | 2013 | */ |
b56a2d8a | 2014 | for (i = prev + 1; i < si->max; i++) { |
4db0c3c2 | 2015 | count = READ_ONCE(si->swap_map[i]); |
355cfa73 | 2016 | if (count && swap_count(count) != SWAP_MAP_BAD) |
10a9c496 | 2017 | break; |
dc644a07 HD |
2018 | if ((i % LATENCY_LIMIT) == 0) |
2019 | cond_resched(); | |
1da177e4 | 2020 | } |
b56a2d8a VRP |
2021 | |
2022 | if (i == si->max) | |
2023 | i = 0; | |
2024 | ||
1da177e4 LT |
2025 | return i; |
2026 | } | |
2027 | ||
10a9c496 | 2028 | static int try_to_unuse(unsigned int type) |
1da177e4 | 2029 | { |
b56a2d8a VRP |
2030 | struct mm_struct *prev_mm; |
2031 | struct mm_struct *mm; | |
2032 | struct list_head *p; | |
2033 | int retval = 0; | |
efa90a98 | 2034 | struct swap_info_struct *si = swap_info[type]; |
1da177e4 LT |
2035 | struct page *page; |
2036 | swp_entry_t entry; | |
b56a2d8a | 2037 | unsigned int i; |
1da177e4 | 2038 | |
21820948 | 2039 | if (!READ_ONCE(si->inuse_pages)) |
b56a2d8a | 2040 | return 0; |
1da177e4 | 2041 | |
b56a2d8a | 2042 | retry: |
10a9c496 | 2043 | retval = shmem_unuse(type); |
b56a2d8a | 2044 | if (retval) |
10a9c496 | 2045 | return retval; |
b56a2d8a VRP |
2046 | |
2047 | prev_mm = &init_mm; | |
2048 | mmget(prev_mm); | |
2049 | ||
2050 | spin_lock(&mmlist_lock); | |
2051 | p = &init_mm.mmlist; | |
21820948 | 2052 | while (READ_ONCE(si->inuse_pages) && |
64165b1a HD |
2053 | !signal_pending(current) && |
2054 | (p = p->next) != &init_mm.mmlist) { | |
1da177e4 | 2055 | |
b56a2d8a VRP |
2056 | mm = list_entry(p, struct mm_struct, mmlist); |
2057 | if (!mmget_not_zero(mm)) | |
2058 | continue; | |
2059 | spin_unlock(&mmlist_lock); | |
2060 | mmput(prev_mm); | |
2061 | prev_mm = mm; | |
10a9c496 | 2062 | retval = unuse_mm(mm, type); |
b56a2d8a VRP |
2063 | if (retval) { |
2064 | mmput(prev_mm); | |
10a9c496 | 2065 | return retval; |
1da177e4 LT |
2066 | } |
2067 | ||
2068 | /* | |
b56a2d8a VRP |
2069 | * Make sure that we aren't completely killing |
2070 | * interactive performance. | |
1da177e4 | 2071 | */ |
b56a2d8a VRP |
2072 | cond_resched(); |
2073 | spin_lock(&mmlist_lock); | |
2074 | } | |
2075 | spin_unlock(&mmlist_lock); | |
1da177e4 | 2076 | |
b56a2d8a | 2077 | mmput(prev_mm); |
1da177e4 | 2078 | |
b56a2d8a | 2079 | i = 0; |
21820948 | 2080 | while (READ_ONCE(si->inuse_pages) && |
64165b1a | 2081 | !signal_pending(current) && |
10a9c496 | 2082 | (i = find_next_to_unuse(si, i)) != 0) { |
1da177e4 | 2083 | |
b56a2d8a VRP |
2084 | entry = swp_entry(type, i); |
2085 | page = find_get_page(swap_address_space(entry), i); | |
2086 | if (!page) | |
2087 | continue; | |
68bdc8d6 HD |
2088 | |
2089 | /* | |
2090 | * It is conceivable that a racing task removed this page from | |
b56a2d8a VRP |
2091 | * swap cache just before we acquired the page lock. The page |
2092 | * might even be back in swap cache on another swap area. But | |
2093 | * that is okay, try_to_free_swap() only removes stale pages. | |
1da177e4 | 2094 | */ |
b56a2d8a VRP |
2095 | lock_page(page); |
2096 | wait_on_page_writeback(page); | |
2097 | try_to_free_swap(page); | |
1da177e4 | 2098 | unlock_page(page); |
09cbfeaf | 2099 | put_page(page); |
1da177e4 LT |
2100 | } |
2101 | ||
b56a2d8a VRP |
2102 | /* |
2103 | * Lets check again to see if there are still swap entries in the map. | |
2104 | * If yes, we would need to do retry the unuse logic again. | |
2105 | * Under global memory pressure, swap entries can be reinserted back | |
2106 | * into process space after the mmlist loop above passes over them. | |
dd862deb | 2107 | * |
e2e3fdc7 MWO |
2108 | * Limit the number of retries? No: when mmget_not_zero() |
2109 | * above fails, that mm is likely to be freeing swap from | |
2110 | * exit_mmap(), which proceeds at its own independent pace; | |
2111 | * and even shmem_writepage() could have been preempted after | |
2112 | * folio_alloc_swap(), temporarily hiding that swap. It's easy | |
2113 | * and robust (though cpu-intensive) just to keep retrying. | |
b56a2d8a | 2114 | */ |
21820948 | 2115 | if (READ_ONCE(si->inuse_pages)) { |
64165b1a HD |
2116 | if (!signal_pending(current)) |
2117 | goto retry; | |
10a9c496 | 2118 | return -EINTR; |
64165b1a | 2119 | } |
10a9c496 CH |
2120 | |
2121 | return 0; | |
1da177e4 LT |
2122 | } |
2123 | ||
2124 | /* | |
5d337b91 HD |
2125 | * After a successful try_to_unuse, if no swap is now in use, we know |
2126 | * we can empty the mmlist. swap_lock must be held on entry and exit. | |
2127 | * Note that mmlist_lock nests inside swap_lock, and an mm must be | |
1da177e4 LT |
2128 | * added to the mmlist just after page_duplicate - before would be racy. |
2129 | */ | |
2130 | static void drain_mmlist(void) | |
2131 | { | |
2132 | struct list_head *p, *next; | |
efa90a98 | 2133 | unsigned int type; |
1da177e4 | 2134 | |
efa90a98 HD |
2135 | for (type = 0; type < nr_swapfiles; type++) |
2136 | if (swap_info[type]->inuse_pages) | |
1da177e4 LT |
2137 | return; |
2138 | spin_lock(&mmlist_lock); | |
2139 | list_for_each_safe(p, next, &init_mm.mmlist) | |
2140 | list_del_init(p); | |
2141 | spin_unlock(&mmlist_lock); | |
2142 | } | |
2143 | ||
1da177e4 LT |
2144 | /* |
2145 | * Free all of a swapdev's extent information | |
2146 | */ | |
2147 | static void destroy_swap_extents(struct swap_info_struct *sis) | |
2148 | { | |
4efaceb1 AL |
2149 | while (!RB_EMPTY_ROOT(&sis->swap_extent_root)) { |
2150 | struct rb_node *rb = sis->swap_extent_root.rb_node; | |
2151 | struct swap_extent *se = rb_entry(rb, struct swap_extent, rb_node); | |
1da177e4 | 2152 | |
4efaceb1 | 2153 | rb_erase(rb, &sis->swap_extent_root); |
1da177e4 LT |
2154 | kfree(se); |
2155 | } | |
62c230bc | 2156 | |
bc4ae27d | 2157 | if (sis->flags & SWP_ACTIVATED) { |
62c230bc MG |
2158 | struct file *swap_file = sis->swap_file; |
2159 | struct address_space *mapping = swap_file->f_mapping; | |
2160 | ||
bc4ae27d OS |
2161 | sis->flags &= ~SWP_ACTIVATED; |
2162 | if (mapping->a_ops->swap_deactivate) | |
2163 | mapping->a_ops->swap_deactivate(swap_file); | |
62c230bc | 2164 | } |
1da177e4 LT |
2165 | } |
2166 | ||
2167 | /* | |
2168 | * Add a block range (and the corresponding page range) into this swapdev's | |
4efaceb1 | 2169 | * extent tree. |
1da177e4 | 2170 | * |
11d31886 | 2171 | * This function rather assumes that it is called in ascending page order. |
1da177e4 | 2172 | */ |
a509bc1a | 2173 | int |
1da177e4 LT |
2174 | add_swap_extent(struct swap_info_struct *sis, unsigned long start_page, |
2175 | unsigned long nr_pages, sector_t start_block) | |
2176 | { | |
4efaceb1 | 2177 | struct rb_node **link = &sis->swap_extent_root.rb_node, *parent = NULL; |
1da177e4 LT |
2178 | struct swap_extent *se; |
2179 | struct swap_extent *new_se; | |
4efaceb1 AL |
2180 | |
2181 | /* | |
2182 | * place the new node at the right most since the | |
2183 | * function is called in ascending page order. | |
2184 | */ | |
2185 | while (*link) { | |
2186 | parent = *link; | |
2187 | link = &parent->rb_right; | |
2188 | } | |
2189 | ||
2190 | if (parent) { | |
2191 | se = rb_entry(parent, struct swap_extent, rb_node); | |
11d31886 HD |
2192 | BUG_ON(se->start_page + se->nr_pages != start_page); |
2193 | if (se->start_block + se->nr_pages == start_block) { | |
1da177e4 LT |
2194 | /* Merge it */ |
2195 | se->nr_pages += nr_pages; | |
2196 | return 0; | |
2197 | } | |
1da177e4 LT |
2198 | } |
2199 | ||
4efaceb1 | 2200 | /* No merge, insert a new extent. */ |
1da177e4 LT |
2201 | new_se = kmalloc(sizeof(*se), GFP_KERNEL); |
2202 | if (new_se == NULL) | |
2203 | return -ENOMEM; | |
2204 | new_se->start_page = start_page; | |
2205 | new_se->nr_pages = nr_pages; | |
2206 | new_se->start_block = start_block; | |
2207 | ||
4efaceb1 AL |
2208 | rb_link_node(&new_se->rb_node, parent, link); |
2209 | rb_insert_color(&new_se->rb_node, &sis->swap_extent_root); | |
53092a74 | 2210 | return 1; |
1da177e4 | 2211 | } |
aa8aa8a3 | 2212 | EXPORT_SYMBOL_GPL(add_swap_extent); |
1da177e4 LT |
2213 | |
2214 | /* | |
2215 | * A `swap extent' is a simple thing which maps a contiguous range of pages | |
ff351f4b ML |
2216 | * onto a contiguous range of disk blocks. A rbtree of swap extents is |
2217 | * built at swapon time and is then used at swap_writepage/swap_readpage | |
1da177e4 LT |
2218 | * time for locating where on disk a page belongs. |
2219 | * | |
2220 | * If the swapfile is an S_ISBLK block device, a single extent is installed. | |
2221 | * This is done so that the main operating code can treat S_ISBLK and S_ISREG | |
2222 | * swap files identically. | |
2223 | * | |
2224 | * Whether the swapdev is an S_ISREG file or an S_ISBLK blockdev, the swap | |
ff351f4b | 2225 | * extent rbtree operates in PAGE_SIZE disk blocks. Both S_ISREG and S_ISBLK |
1da177e4 LT |
2226 | * swapfiles are handled *identically* after swapon time. |
2227 | * | |
2228 | * For S_ISREG swapfiles, setup_swap_extents() will walk all the file's blocks | |
ff351f4b ML |
2229 | * and will parse them into a rbtree, in PAGE_SIZE chunks. If some stray |
2230 | * blocks are found which do not fall within the PAGE_SIZE alignment | |
1da177e4 LT |
2231 | * requirements, they are simply tossed out - we will never use those blocks |
2232 | * for swapping. | |
2233 | * | |
1638045c DW |
2234 | * For all swap devices we set S_SWAPFILE across the life of the swapon. This |
2235 | * prevents users from writing to the swap device, which will corrupt memory. | |
1da177e4 LT |
2236 | * |
2237 | * The amount of disk space which a single swap extent represents varies. | |
2238 | * Typically it is in the 1-4 megabyte range. So we can have hundreds of | |
ff351f4b | 2239 | * extents in the rbtree. - akpm. |
1da177e4 | 2240 | */ |
53092a74 | 2241 | static int setup_swap_extents(struct swap_info_struct *sis, sector_t *span) |
1da177e4 | 2242 | { |
62c230bc MG |
2243 | struct file *swap_file = sis->swap_file; |
2244 | struct address_space *mapping = swap_file->f_mapping; | |
2245 | struct inode *inode = mapping->host; | |
1da177e4 LT |
2246 | int ret; |
2247 | ||
1da177e4 LT |
2248 | if (S_ISBLK(inode->i_mode)) { |
2249 | ret = add_swap_extent(sis, 0, sis->max, 0); | |
53092a74 | 2250 | *span = sis->pages; |
a509bc1a | 2251 | return ret; |
1da177e4 LT |
2252 | } |
2253 | ||
62c230bc | 2254 | if (mapping->a_ops->swap_activate) { |
a509bc1a | 2255 | ret = mapping->a_ops->swap_activate(sis, swap_file, span); |
4b60c0ff N |
2256 | if (ret < 0) |
2257 | return ret; | |
2258 | sis->flags |= SWP_ACTIVATED; | |
e1209d3a N |
2259 | if ((sis->flags & SWP_FS_OPS) && |
2260 | sio_pool_init() != 0) { | |
2261 | destroy_swap_extents(sis); | |
2262 | return -ENOMEM; | |
62c230bc | 2263 | } |
a509bc1a | 2264 | return ret; |
62c230bc MG |
2265 | } |
2266 | ||
a509bc1a | 2267 | return generic_swapfile_activate(sis, swap_file, span); |
1da177e4 LT |
2268 | } |
2269 | ||
a2468cc9 AL |
2270 | static int swap_node(struct swap_info_struct *p) |
2271 | { | |
2272 | struct block_device *bdev; | |
2273 | ||
2274 | if (p->bdev) | |
2275 | bdev = p->bdev; | |
2276 | else | |
2277 | bdev = p->swap_file->f_inode->i_sb->s_bdev; | |
2278 | ||
2279 | return bdev ? bdev->bd_disk->node_id : NUMA_NO_NODE; | |
2280 | } | |
2281 | ||
eb085574 HY |
2282 | static void setup_swap_info(struct swap_info_struct *p, int prio, |
2283 | unsigned char *swap_map, | |
2284 | struct swap_cluster_info *cluster_info) | |
40531542 | 2285 | { |
a2468cc9 AL |
2286 | int i; |
2287 | ||
40531542 CEB |
2288 | if (prio >= 0) |
2289 | p->prio = prio; | |
2290 | else | |
2291 | p->prio = --least_priority; | |
18ab4d4c DS |
2292 | /* |
2293 | * the plist prio is negated because plist ordering is | |
2294 | * low-to-high, while swap ordering is high-to-low | |
2295 | */ | |
2296 | p->list.prio = -p->prio; | |
a2468cc9 AL |
2297 | for_each_node(i) { |
2298 | if (p->prio >= 0) | |
2299 | p->avail_lists[i].prio = -p->prio; | |
2300 | else { | |
2301 | if (swap_node(p) == i) | |
2302 | p->avail_lists[i].prio = 1; | |
2303 | else | |
2304 | p->avail_lists[i].prio = -p->prio; | |
2305 | } | |
2306 | } | |
40531542 | 2307 | p->swap_map = swap_map; |
2a8f9449 | 2308 | p->cluster_info = cluster_info; |
eb085574 HY |
2309 | } |
2310 | ||
2311 | static void _enable_swap_info(struct swap_info_struct *p) | |
2312 | { | |
63d8620e | 2313 | p->flags |= SWP_WRITEOK; |
ec8acf20 | 2314 | atomic_long_add(p->pages, &nr_swap_pages); |
40531542 CEB |
2315 | total_swap_pages += p->pages; |
2316 | ||
adfab836 | 2317 | assert_spin_locked(&swap_lock); |
adfab836 | 2318 | /* |
18ab4d4c DS |
2319 | * both lists are plists, and thus priority ordered. |
2320 | * swap_active_head needs to be priority ordered for swapoff(), | |
2321 | * which on removal of any swap_info_struct with an auto-assigned | |
2322 | * (i.e. negative) priority increments the auto-assigned priority | |
2323 | * of any lower-priority swap_info_structs. | |
e2e3fdc7 | 2324 | * swap_avail_head needs to be priority ordered for folio_alloc_swap(), |
18ab4d4c DS |
2325 | * which allocates swap pages from the highest available priority |
2326 | * swap_info_struct. | |
adfab836 | 2327 | */ |
18ab4d4c | 2328 | plist_add(&p->list, &swap_active_head); |
a2468cc9 | 2329 | add_to_avail_list(p); |
cf0cac0a CEB |
2330 | } |
2331 | ||
2332 | static void enable_swap_info(struct swap_info_struct *p, int prio, | |
2333 | unsigned char *swap_map, | |
2a8f9449 | 2334 | struct swap_cluster_info *cluster_info, |
cf0cac0a CEB |
2335 | unsigned long *frontswap_map) |
2336 | { | |
1cf53c89 CH |
2337 | if (IS_ENABLED(CONFIG_FRONTSWAP)) |
2338 | frontswap_init(p->type, frontswap_map); | |
cf0cac0a | 2339 | spin_lock(&swap_lock); |
ec8acf20 | 2340 | spin_lock(&p->lock); |
eb085574 HY |
2341 | setup_swap_info(p, prio, swap_map, cluster_info); |
2342 | spin_unlock(&p->lock); | |
2343 | spin_unlock(&swap_lock); | |
2344 | /* | |
63d8620e | 2345 | * Finished initializing swap device, now it's safe to reference it. |
eb085574 | 2346 | */ |
63d8620e | 2347 | percpu_ref_resurrect(&p->users); |
eb085574 HY |
2348 | spin_lock(&swap_lock); |
2349 | spin_lock(&p->lock); | |
2350 | _enable_swap_info(p); | |
ec8acf20 | 2351 | spin_unlock(&p->lock); |
cf0cac0a CEB |
2352 | spin_unlock(&swap_lock); |
2353 | } | |
2354 | ||
2355 | static void reinsert_swap_info(struct swap_info_struct *p) | |
2356 | { | |
2357 | spin_lock(&swap_lock); | |
ec8acf20 | 2358 | spin_lock(&p->lock); |
eb085574 HY |
2359 | setup_swap_info(p, p->prio, p->swap_map, p->cluster_info); |
2360 | _enable_swap_info(p); | |
ec8acf20 | 2361 | spin_unlock(&p->lock); |
40531542 CEB |
2362 | spin_unlock(&swap_lock); |
2363 | } | |
2364 | ||
67afa38e TC |
2365 | bool has_usable_swap(void) |
2366 | { | |
2367 | bool ret = true; | |
2368 | ||
2369 | spin_lock(&swap_lock); | |
2370 | if (plist_head_empty(&swap_active_head)) | |
2371 | ret = false; | |
2372 | spin_unlock(&swap_lock); | |
2373 | return ret; | |
2374 | } | |
2375 | ||
c4ea37c2 | 2376 | SYSCALL_DEFINE1(swapoff, const char __user *, specialfile) |
1da177e4 | 2377 | { |
73c34b6a | 2378 | struct swap_info_struct *p = NULL; |
8d69aaee | 2379 | unsigned char *swap_map; |
2a8f9449 | 2380 | struct swap_cluster_info *cluster_info; |
4f89849d | 2381 | unsigned long *frontswap_map; |
1da177e4 LT |
2382 | struct file *swap_file, *victim; |
2383 | struct address_space *mapping; | |
2384 | struct inode *inode; | |
91a27b2a | 2385 | struct filename *pathname; |
adfab836 | 2386 | int err, found = 0; |
5b808a23 | 2387 | unsigned int old_block_size; |
886bb7e9 | 2388 | |
1da177e4 LT |
2389 | if (!capable(CAP_SYS_ADMIN)) |
2390 | return -EPERM; | |
2391 | ||
191c5424 AV |
2392 | BUG_ON(!current->mm); |
2393 | ||
1da177e4 | 2394 | pathname = getname(specialfile); |
1da177e4 | 2395 | if (IS_ERR(pathname)) |
f58b59c1 | 2396 | return PTR_ERR(pathname); |
1da177e4 | 2397 | |
669abf4e | 2398 | victim = file_open_name(pathname, O_RDWR|O_LARGEFILE, 0); |
1da177e4 LT |
2399 | err = PTR_ERR(victim); |
2400 | if (IS_ERR(victim)) | |
2401 | goto out; | |
2402 | ||
2403 | mapping = victim->f_mapping; | |
5d337b91 | 2404 | spin_lock(&swap_lock); |
18ab4d4c | 2405 | plist_for_each_entry(p, &swap_active_head, list) { |
22c6f8fd | 2406 | if (p->flags & SWP_WRITEOK) { |
adfab836 DS |
2407 | if (p->swap_file->f_mapping == mapping) { |
2408 | found = 1; | |
1da177e4 | 2409 | break; |
adfab836 | 2410 | } |
1da177e4 | 2411 | } |
1da177e4 | 2412 | } |
adfab836 | 2413 | if (!found) { |
1da177e4 | 2414 | err = -EINVAL; |
5d337b91 | 2415 | spin_unlock(&swap_lock); |
1da177e4 LT |
2416 | goto out_dput; |
2417 | } | |
191c5424 | 2418 | if (!security_vm_enough_memory_mm(current->mm, p->pages)) |
1da177e4 LT |
2419 | vm_unacct_memory(p->pages); |
2420 | else { | |
2421 | err = -ENOMEM; | |
5d337b91 | 2422 | spin_unlock(&swap_lock); |
1da177e4 LT |
2423 | goto out_dput; |
2424 | } | |
a2468cc9 | 2425 | del_from_avail_list(p); |
ec8acf20 | 2426 | spin_lock(&p->lock); |
78ecba08 | 2427 | if (p->prio < 0) { |
adfab836 | 2428 | struct swap_info_struct *si = p; |
a2468cc9 | 2429 | int nid; |
adfab836 | 2430 | |
18ab4d4c | 2431 | plist_for_each_entry_continue(si, &swap_active_head, list) { |
adfab836 | 2432 | si->prio++; |
18ab4d4c | 2433 | si->list.prio--; |
a2468cc9 AL |
2434 | for_each_node(nid) { |
2435 | if (si->avail_lists[nid].prio != 1) | |
2436 | si->avail_lists[nid].prio--; | |
2437 | } | |
adfab836 | 2438 | } |
78ecba08 HD |
2439 | least_priority++; |
2440 | } | |
18ab4d4c | 2441 | plist_del(&p->list, &swap_active_head); |
ec8acf20 | 2442 | atomic_long_sub(p->pages, &nr_swap_pages); |
1da177e4 LT |
2443 | total_swap_pages -= p->pages; |
2444 | p->flags &= ~SWP_WRITEOK; | |
ec8acf20 | 2445 | spin_unlock(&p->lock); |
5d337b91 | 2446 | spin_unlock(&swap_lock); |
fb4f88dc | 2447 | |
039939a6 TC |
2448 | disable_swap_slots_cache_lock(); |
2449 | ||
e1e12d2f | 2450 | set_current_oom_origin(); |
10a9c496 | 2451 | err = try_to_unuse(p->type); |
e1e12d2f | 2452 | clear_current_oom_origin(); |
1da177e4 | 2453 | |
1da177e4 LT |
2454 | if (err) { |
2455 | /* re-insert swap space back into swap_list */ | |
cf0cac0a | 2456 | reinsert_swap_info(p); |
039939a6 | 2457 | reenable_swap_slots_cache_unlock(); |
1da177e4 LT |
2458 | goto out_dput; |
2459 | } | |
52b7efdb | 2460 | |
039939a6 TC |
2461 | reenable_swap_slots_cache_unlock(); |
2462 | ||
eb085574 | 2463 | /* |
63d8620e ML |
2464 | * Wait for swap operations protected by get/put_swap_device() |
2465 | * to complete. | |
2466 | * | |
2467 | * We need synchronize_rcu() here to protect the accessing to | |
2468 | * the swap cache data structure. | |
eb085574 | 2469 | */ |
63d8620e | 2470 | percpu_ref_kill(&p->users); |
eb085574 | 2471 | synchronize_rcu(); |
63d8620e | 2472 | wait_for_completion(&p->comp); |
eb085574 | 2473 | |
815c2c54 SL |
2474 | flush_work(&p->discard_work); |
2475 | ||
5d337b91 | 2476 | destroy_swap_extents(p); |
570a335b HD |
2477 | if (p->flags & SWP_CONTINUED) |
2478 | free_swap_count_continuations(p); | |
2479 | ||
10f0d2a5 | 2480 | if (!p->bdev || !bdev_nonrot(p->bdev)) |
81a0298b HY |
2481 | atomic_dec(&nr_rotate_swap); |
2482 | ||
fc0abb14 | 2483 | mutex_lock(&swapon_mutex); |
5d337b91 | 2484 | spin_lock(&swap_lock); |
ec8acf20 | 2485 | spin_lock(&p->lock); |
5d337b91 HD |
2486 | drain_mmlist(); |
2487 | ||
bb243f7d | 2488 | /* wait for anyone still in scan_swap_map_slots */ |
52b7efdb HD |
2489 | p->highest_bit = 0; /* cuts scans short */ |
2490 | while (p->flags >= SWP_SCANNING) { | |
ec8acf20 | 2491 | spin_unlock(&p->lock); |
5d337b91 | 2492 | spin_unlock(&swap_lock); |
13e4b57f | 2493 | schedule_timeout_uninterruptible(1); |
5d337b91 | 2494 | spin_lock(&swap_lock); |
ec8acf20 | 2495 | spin_lock(&p->lock); |
52b7efdb | 2496 | } |
52b7efdb | 2497 | |
1da177e4 | 2498 | swap_file = p->swap_file; |
5b808a23 | 2499 | old_block_size = p->old_block_size; |
1da177e4 LT |
2500 | p->swap_file = NULL; |
2501 | p->max = 0; | |
2502 | swap_map = p->swap_map; | |
2503 | p->swap_map = NULL; | |
2a8f9449 SL |
2504 | cluster_info = p->cluster_info; |
2505 | p->cluster_info = NULL; | |
4f89849d | 2506 | frontswap_map = frontswap_map_get(p); |
ec8acf20 | 2507 | spin_unlock(&p->lock); |
5d337b91 | 2508 | spin_unlock(&swap_lock); |
8a84802e | 2509 | arch_swap_invalidate_area(p->type); |
adfab836 | 2510 | frontswap_invalidate_area(p->type); |
58e97ba6 | 2511 | frontswap_map_set(p, NULL); |
fc0abb14 | 2512 | mutex_unlock(&swapon_mutex); |
ebc2a1a6 SL |
2513 | free_percpu(p->percpu_cluster); |
2514 | p->percpu_cluster = NULL; | |
49070588 HY |
2515 | free_percpu(p->cluster_next_cpu); |
2516 | p->cluster_next_cpu = NULL; | |
1da177e4 | 2517 | vfree(swap_map); |
54f180d3 HY |
2518 | kvfree(cluster_info); |
2519 | kvfree(frontswap_map); | |
2de1a7e4 | 2520 | /* Destroy swap account information */ |
adfab836 | 2521 | swap_cgroup_swapoff(p->type); |
4b3ef9da | 2522 | exit_swap_address_space(p->type); |
27a7faa0 | 2523 | |
1da177e4 LT |
2524 | inode = mapping->host; |
2525 | if (S_ISBLK(inode->i_mode)) { | |
2526 | struct block_device *bdev = I_BDEV(inode); | |
1638045c | 2527 | |
5b808a23 | 2528 | set_blocksize(bdev, old_block_size); |
e525fd89 | 2529 | blkdev_put(bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL); |
1da177e4 | 2530 | } |
1638045c DW |
2531 | |
2532 | inode_lock(inode); | |
2533 | inode->i_flags &= ~S_SWAPFILE; | |
2534 | inode_unlock(inode); | |
1da177e4 | 2535 | filp_close(swap_file, NULL); |
f893ab41 WY |
2536 | |
2537 | /* | |
2538 | * Clear the SWP_USED flag after all resources are freed so that swapon | |
2539 | * can reuse this swap_info in alloc_swap_info() safely. It is ok to | |
2540 | * not hold p->lock after we cleared its SWP_WRITEOK. | |
2541 | */ | |
2542 | spin_lock(&swap_lock); | |
2543 | p->flags = 0; | |
2544 | spin_unlock(&swap_lock); | |
2545 | ||
1da177e4 | 2546 | err = 0; |
66d7dd51 KS |
2547 | atomic_inc(&proc_poll_event); |
2548 | wake_up_interruptible(&proc_poll_wait); | |
1da177e4 LT |
2549 | |
2550 | out_dput: | |
2551 | filp_close(victim, NULL); | |
2552 | out: | |
f58b59c1 | 2553 | putname(pathname); |
1da177e4 LT |
2554 | return err; |
2555 | } | |
2556 | ||
2557 | #ifdef CONFIG_PROC_FS | |
9dd95748 | 2558 | static __poll_t swaps_poll(struct file *file, poll_table *wait) |
66d7dd51 | 2559 | { |
f1514638 | 2560 | struct seq_file *seq = file->private_data; |
66d7dd51 KS |
2561 | |
2562 | poll_wait(file, &proc_poll_wait, wait); | |
2563 | ||
f1514638 KS |
2564 | if (seq->poll_event != atomic_read(&proc_poll_event)) { |
2565 | seq->poll_event = atomic_read(&proc_poll_event); | |
a9a08845 | 2566 | return EPOLLIN | EPOLLRDNORM | EPOLLERR | EPOLLPRI; |
66d7dd51 KS |
2567 | } |
2568 | ||
a9a08845 | 2569 | return EPOLLIN | EPOLLRDNORM; |
66d7dd51 KS |
2570 | } |
2571 | ||
1da177e4 LT |
2572 | /* iterator */ |
2573 | static void *swap_start(struct seq_file *swap, loff_t *pos) | |
2574 | { | |
efa90a98 HD |
2575 | struct swap_info_struct *si; |
2576 | int type; | |
1da177e4 LT |
2577 | loff_t l = *pos; |
2578 | ||
fc0abb14 | 2579 | mutex_lock(&swapon_mutex); |
1da177e4 | 2580 | |
881e4aab SS |
2581 | if (!l) |
2582 | return SEQ_START_TOKEN; | |
2583 | ||
c10d38cc | 2584 | for (type = 0; (si = swap_type_to_swap_info(type)); type++) { |
efa90a98 | 2585 | if (!(si->flags & SWP_USED) || !si->swap_map) |
1da177e4 | 2586 | continue; |
881e4aab | 2587 | if (!--l) |
efa90a98 | 2588 | return si; |
1da177e4 LT |
2589 | } |
2590 | ||
2591 | return NULL; | |
2592 | } | |
2593 | ||
2594 | static void *swap_next(struct seq_file *swap, void *v, loff_t *pos) | |
2595 | { | |
efa90a98 HD |
2596 | struct swap_info_struct *si = v; |
2597 | int type; | |
1da177e4 | 2598 | |
881e4aab | 2599 | if (v == SEQ_START_TOKEN) |
efa90a98 HD |
2600 | type = 0; |
2601 | else | |
2602 | type = si->type + 1; | |
881e4aab | 2603 | |
10c8d69f | 2604 | ++(*pos); |
c10d38cc | 2605 | for (; (si = swap_type_to_swap_info(type)); type++) { |
efa90a98 | 2606 | if (!(si->flags & SWP_USED) || !si->swap_map) |
1da177e4 | 2607 | continue; |
efa90a98 | 2608 | return si; |
1da177e4 LT |
2609 | } |
2610 | ||
2611 | return NULL; | |
2612 | } | |
2613 | ||
2614 | static void swap_stop(struct seq_file *swap, void *v) | |
2615 | { | |
fc0abb14 | 2616 | mutex_unlock(&swapon_mutex); |
1da177e4 LT |
2617 | } |
2618 | ||
2619 | static int swap_show(struct seq_file *swap, void *v) | |
2620 | { | |
efa90a98 | 2621 | struct swap_info_struct *si = v; |
1da177e4 LT |
2622 | struct file *file; |
2623 | int len; | |
642929a2 | 2624 | unsigned long bytes, inuse; |
1da177e4 | 2625 | |
efa90a98 | 2626 | if (si == SEQ_START_TOKEN) { |
68d68ff6 | 2627 | seq_puts(swap, "Filename\t\t\t\tType\t\tSize\t\tUsed\t\tPriority\n"); |
881e4aab SS |
2628 | return 0; |
2629 | } | |
1da177e4 | 2630 | |
6f793940 | 2631 | bytes = si->pages << (PAGE_SHIFT - 10); |
c8945306 | 2632 | inuse = READ_ONCE(si->inuse_pages) << (PAGE_SHIFT - 10); |
6f793940 | 2633 | |
efa90a98 | 2634 | file = si->swap_file; |
2726d566 | 2635 | len = seq_file_path(swap, file, " \t\n\\"); |
642929a2 | 2636 | seq_printf(swap, "%*s%s\t%lu\t%s%lu\t%s%d\n", |
886bb7e9 | 2637 | len < 40 ? 40 - len : 1, " ", |
496ad9aa | 2638 | S_ISBLK(file_inode(file)->i_mode) ? |
1da177e4 | 2639 | "partition" : "file\t", |
6f793940 RD |
2640 | bytes, bytes < 10000000 ? "\t" : "", |
2641 | inuse, inuse < 10000000 ? "\t" : "", | |
efa90a98 | 2642 | si->prio); |
1da177e4 LT |
2643 | return 0; |
2644 | } | |
2645 | ||
15ad7cdc | 2646 | static const struct seq_operations swaps_op = { |
1da177e4 LT |
2647 | .start = swap_start, |
2648 | .next = swap_next, | |
2649 | .stop = swap_stop, | |
2650 | .show = swap_show | |
2651 | }; | |
2652 | ||
2653 | static int swaps_open(struct inode *inode, struct file *file) | |
2654 | { | |
f1514638 | 2655 | struct seq_file *seq; |
66d7dd51 KS |
2656 | int ret; |
2657 | ||
66d7dd51 | 2658 | ret = seq_open(file, &swaps_op); |
f1514638 | 2659 | if (ret) |
66d7dd51 | 2660 | return ret; |
66d7dd51 | 2661 | |
f1514638 KS |
2662 | seq = file->private_data; |
2663 | seq->poll_event = atomic_read(&proc_poll_event); | |
2664 | return 0; | |
1da177e4 LT |
2665 | } |
2666 | ||
97a32539 | 2667 | static const struct proc_ops swaps_proc_ops = { |
d919b33d | 2668 | .proc_flags = PROC_ENTRY_PERMANENT, |
97a32539 AD |
2669 | .proc_open = swaps_open, |
2670 | .proc_read = seq_read, | |
2671 | .proc_lseek = seq_lseek, | |
2672 | .proc_release = seq_release, | |
2673 | .proc_poll = swaps_poll, | |
1da177e4 LT |
2674 | }; |
2675 | ||
2676 | static int __init procswaps_init(void) | |
2677 | { | |
97a32539 | 2678 | proc_create("swaps", 0, NULL, &swaps_proc_ops); |
1da177e4 LT |
2679 | return 0; |
2680 | } | |
2681 | __initcall(procswaps_init); | |
2682 | #endif /* CONFIG_PROC_FS */ | |
2683 | ||
1796316a JB |
2684 | #ifdef MAX_SWAPFILES_CHECK |
2685 | static int __init max_swapfiles_check(void) | |
2686 | { | |
2687 | MAX_SWAPFILES_CHECK(); | |
2688 | return 0; | |
2689 | } | |
2690 | late_initcall(max_swapfiles_check); | |
2691 | #endif | |
2692 | ||
53cbb243 | 2693 | static struct swap_info_struct *alloc_swap_info(void) |
1da177e4 | 2694 | { |
73c34b6a | 2695 | struct swap_info_struct *p; |
b11a76b3 | 2696 | struct swap_info_struct *defer = NULL; |
1da177e4 | 2697 | unsigned int type; |
a2468cc9 | 2698 | int i; |
efa90a98 | 2699 | |
96008744 | 2700 | p = kvzalloc(struct_size(p, avail_lists, nr_node_ids), GFP_KERNEL); |
efa90a98 | 2701 | if (!p) |
53cbb243 | 2702 | return ERR_PTR(-ENOMEM); |
efa90a98 | 2703 | |
63d8620e ML |
2704 | if (percpu_ref_init(&p->users, swap_users_ref_free, |
2705 | PERCPU_REF_INIT_DEAD, GFP_KERNEL)) { | |
2706 | kvfree(p); | |
2707 | return ERR_PTR(-ENOMEM); | |
2708 | } | |
2709 | ||
5d337b91 | 2710 | spin_lock(&swap_lock); |
efa90a98 HD |
2711 | for (type = 0; type < nr_swapfiles; type++) { |
2712 | if (!(swap_info[type]->flags & SWP_USED)) | |
1da177e4 | 2713 | break; |
efa90a98 | 2714 | } |
0697212a | 2715 | if (type >= MAX_SWAPFILES) { |
5d337b91 | 2716 | spin_unlock(&swap_lock); |
63d8620e | 2717 | percpu_ref_exit(&p->users); |
873d7bcf | 2718 | kvfree(p); |
730c0581 | 2719 | return ERR_PTR(-EPERM); |
1da177e4 | 2720 | } |
efa90a98 HD |
2721 | if (type >= nr_swapfiles) { |
2722 | p->type = type; | |
efa90a98 | 2723 | /* |
a4b45114 HY |
2724 | * Publish the swap_info_struct after initializing it. |
2725 | * Note that kvzalloc() above zeroes all its fields. | |
efa90a98 | 2726 | */ |
a4b45114 HY |
2727 | smp_store_release(&swap_info[type], p); /* rcu_assign_pointer() */ |
2728 | nr_swapfiles++; | |
efa90a98 | 2729 | } else { |
b11a76b3 | 2730 | defer = p; |
efa90a98 HD |
2731 | p = swap_info[type]; |
2732 | /* | |
2733 | * Do not memset this entry: a racing procfs swap_next() | |
2734 | * would be relying on p->type to remain valid. | |
2735 | */ | |
2736 | } | |
4efaceb1 | 2737 | p->swap_extent_root = RB_ROOT; |
18ab4d4c | 2738 | plist_node_init(&p->list, 0); |
a2468cc9 AL |
2739 | for_each_node(i) |
2740 | plist_node_init(&p->avail_lists[i], 0); | |
1da177e4 | 2741 | p->flags = SWP_USED; |
5d337b91 | 2742 | spin_unlock(&swap_lock); |
63d8620e ML |
2743 | if (defer) { |
2744 | percpu_ref_exit(&defer->users); | |
2745 | kvfree(defer); | |
2746 | } | |
ec8acf20 | 2747 | spin_lock_init(&p->lock); |
2628bd6f | 2748 | spin_lock_init(&p->cont_lock); |
63d8620e | 2749 | init_completion(&p->comp); |
efa90a98 | 2750 | |
53cbb243 | 2751 | return p; |
53cbb243 CEB |
2752 | } |
2753 | ||
4d0e1e10 CEB |
2754 | static int claim_swapfile(struct swap_info_struct *p, struct inode *inode) |
2755 | { | |
2756 | int error; | |
2757 | ||
2758 | if (S_ISBLK(inode->i_mode)) { | |
ef16e1d9 | 2759 | p->bdev = blkdev_get_by_dev(inode->i_rdev, |
6f179af8 | 2760 | FMODE_READ | FMODE_WRITE | FMODE_EXCL, p); |
ef16e1d9 CH |
2761 | if (IS_ERR(p->bdev)) { |
2762 | error = PTR_ERR(p->bdev); | |
4d0e1e10 | 2763 | p->bdev = NULL; |
6f179af8 | 2764 | return error; |
4d0e1e10 CEB |
2765 | } |
2766 | p->old_block_size = block_size(p->bdev); | |
2767 | error = set_blocksize(p->bdev, PAGE_SIZE); | |
2768 | if (error < 0) | |
87ade72a | 2769 | return error; |
12d2966d NA |
2770 | /* |
2771 | * Zoned block devices contain zones that have a sequential | |
2772 | * write only restriction. Hence zoned block devices are not | |
2773 | * suitable for swapping. Disallow them here. | |
2774 | */ | |
9964e674 | 2775 | if (bdev_is_zoned(p->bdev)) |
12d2966d | 2776 | return -EINVAL; |
4d0e1e10 CEB |
2777 | p->flags |= SWP_BLKDEV; |
2778 | } else if (S_ISREG(inode->i_mode)) { | |
2779 | p->bdev = inode->i_sb->s_bdev; | |
1638045c DW |
2780 | } |
2781 | ||
4d0e1e10 | 2782 | return 0; |
4d0e1e10 CEB |
2783 | } |
2784 | ||
377eeaa8 AK |
2785 | |
2786 | /* | |
2787 | * Find out how many pages are allowed for a single swap device. There | |
2788 | * are two limiting factors: | |
2789 | * 1) the number of bits for the swap offset in the swp_entry_t type, and | |
2790 | * 2) the number of bits in the swap pte, as defined by the different | |
2791 | * architectures. | |
2792 | * | |
2793 | * In order to find the largest possible bit mask, a swap entry with | |
2794 | * swap type 0 and swap offset ~0UL is created, encoded to a swap pte, | |
2795 | * decoded to a swp_entry_t again, and finally the swap offset is | |
2796 | * extracted. | |
2797 | * | |
2798 | * This will mask all the bits from the initial ~0UL mask that can't | |
2799 | * be encoded in either the swp_entry_t or the architecture definition | |
2800 | * of a swap pte. | |
2801 | */ | |
2802 | unsigned long generic_max_swapfile_size(void) | |
2803 | { | |
2804 | return swp_offset(pte_to_swp_entry( | |
2805 | swp_entry_to_pte(swp_entry(0, ~0UL)))) + 1; | |
2806 | } | |
2807 | ||
2808 | /* Can be overridden by an architecture for additional checks. */ | |
be45a490 | 2809 | __weak unsigned long arch_max_swapfile_size(void) |
377eeaa8 AK |
2810 | { |
2811 | return generic_max_swapfile_size(); | |
2812 | } | |
2813 | ||
ca8bd38b CEB |
2814 | static unsigned long read_swap_header(struct swap_info_struct *p, |
2815 | union swap_header *swap_header, | |
2816 | struct inode *inode) | |
2817 | { | |
2818 | int i; | |
2819 | unsigned long maxpages; | |
2820 | unsigned long swapfilepages; | |
d6bbbd29 | 2821 | unsigned long last_page; |
ca8bd38b CEB |
2822 | |
2823 | if (memcmp("SWAPSPACE2", swap_header->magic.magic, 10)) { | |
465c47fd | 2824 | pr_err("Unable to find swap-space signature\n"); |
38719025 | 2825 | return 0; |
ca8bd38b CEB |
2826 | } |
2827 | ||
041711ce | 2828 | /* swap partition endianness hack... */ |
ca8bd38b CEB |
2829 | if (swab32(swap_header->info.version) == 1) { |
2830 | swab32s(&swap_header->info.version); | |
2831 | swab32s(&swap_header->info.last_page); | |
2832 | swab32s(&swap_header->info.nr_badpages); | |
dd111be6 JH |
2833 | if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES) |
2834 | return 0; | |
ca8bd38b CEB |
2835 | for (i = 0; i < swap_header->info.nr_badpages; i++) |
2836 | swab32s(&swap_header->info.badpages[i]); | |
2837 | } | |
2838 | /* Check the swap header's sub-version */ | |
2839 | if (swap_header->info.version != 1) { | |
465c47fd AM |
2840 | pr_warn("Unable to handle swap header version %d\n", |
2841 | swap_header->info.version); | |
38719025 | 2842 | return 0; |
ca8bd38b CEB |
2843 | } |
2844 | ||
2845 | p->lowest_bit = 1; | |
2846 | p->cluster_next = 1; | |
2847 | p->cluster_nr = 0; | |
2848 | ||
be45a490 | 2849 | maxpages = swapfile_maximum_size; |
d6bbbd29 | 2850 | last_page = swap_header->info.last_page; |
a06ad633 TA |
2851 | if (!last_page) { |
2852 | pr_warn("Empty swap-file\n"); | |
2853 | return 0; | |
2854 | } | |
d6bbbd29 | 2855 | if (last_page > maxpages) { |
465c47fd | 2856 | pr_warn("Truncating oversized swap area, only using %luk out of %luk\n", |
d6bbbd29 RJ |
2857 | maxpages << (PAGE_SHIFT - 10), |
2858 | last_page << (PAGE_SHIFT - 10)); | |
2859 | } | |
2860 | if (maxpages > last_page) { | |
2861 | maxpages = last_page + 1; | |
ca8bd38b CEB |
2862 | /* p->max is an unsigned int: don't overflow it */ |
2863 | if ((unsigned int)maxpages == 0) | |
2864 | maxpages = UINT_MAX; | |
2865 | } | |
2866 | p->highest_bit = maxpages - 1; | |
2867 | ||
2868 | if (!maxpages) | |
38719025 | 2869 | return 0; |
ca8bd38b CEB |
2870 | swapfilepages = i_size_read(inode) >> PAGE_SHIFT; |
2871 | if (swapfilepages && maxpages > swapfilepages) { | |
465c47fd | 2872 | pr_warn("Swap area shorter than signature indicates\n"); |
38719025 | 2873 | return 0; |
ca8bd38b CEB |
2874 | } |
2875 | if (swap_header->info.nr_badpages && S_ISREG(inode->i_mode)) | |
38719025 | 2876 | return 0; |
ca8bd38b | 2877 | if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES) |
38719025 | 2878 | return 0; |
ca8bd38b CEB |
2879 | |
2880 | return maxpages; | |
ca8bd38b CEB |
2881 | } |
2882 | ||
4b3ef9da | 2883 | #define SWAP_CLUSTER_INFO_COLS \ |
235b6217 | 2884 | DIV_ROUND_UP(L1_CACHE_BYTES, sizeof(struct swap_cluster_info)) |
4b3ef9da HY |
2885 | #define SWAP_CLUSTER_SPACE_COLS \ |
2886 | DIV_ROUND_UP(SWAP_ADDRESS_SPACE_PAGES, SWAPFILE_CLUSTER) | |
2887 | #define SWAP_CLUSTER_COLS \ | |
2888 | max_t(unsigned int, SWAP_CLUSTER_INFO_COLS, SWAP_CLUSTER_SPACE_COLS) | |
235b6217 | 2889 | |
915d4d7b CEB |
2890 | static int setup_swap_map_and_extents(struct swap_info_struct *p, |
2891 | union swap_header *swap_header, | |
2892 | unsigned char *swap_map, | |
2a8f9449 | 2893 | struct swap_cluster_info *cluster_info, |
915d4d7b CEB |
2894 | unsigned long maxpages, |
2895 | sector_t *span) | |
2896 | { | |
235b6217 | 2897 | unsigned int j, k; |
915d4d7b CEB |
2898 | unsigned int nr_good_pages; |
2899 | int nr_extents; | |
2a8f9449 | 2900 | unsigned long nr_clusters = DIV_ROUND_UP(maxpages, SWAPFILE_CLUSTER); |
235b6217 HY |
2901 | unsigned long col = p->cluster_next / SWAPFILE_CLUSTER % SWAP_CLUSTER_COLS; |
2902 | unsigned long i, idx; | |
915d4d7b CEB |
2903 | |
2904 | nr_good_pages = maxpages - 1; /* omit header page */ | |
2905 | ||
6b534915 HY |
2906 | cluster_list_init(&p->free_clusters); |
2907 | cluster_list_init(&p->discard_clusters); | |
2a8f9449 | 2908 | |
915d4d7b CEB |
2909 | for (i = 0; i < swap_header->info.nr_badpages; i++) { |
2910 | unsigned int page_nr = swap_header->info.badpages[i]; | |
bdb8e3f6 CEB |
2911 | if (page_nr == 0 || page_nr > swap_header->info.last_page) |
2912 | return -EINVAL; | |
915d4d7b CEB |
2913 | if (page_nr < maxpages) { |
2914 | swap_map[page_nr] = SWAP_MAP_BAD; | |
2915 | nr_good_pages--; | |
2a8f9449 SL |
2916 | /* |
2917 | * Haven't marked the cluster free yet, no list | |
2918 | * operation involved | |
2919 | */ | |
2920 | inc_cluster_info_page(p, cluster_info, page_nr); | |
915d4d7b CEB |
2921 | } |
2922 | } | |
2923 | ||
2a8f9449 SL |
2924 | /* Haven't marked the cluster free yet, no list operation involved */ |
2925 | for (i = maxpages; i < round_up(maxpages, SWAPFILE_CLUSTER); i++) | |
2926 | inc_cluster_info_page(p, cluster_info, i); | |
2927 | ||
915d4d7b CEB |
2928 | if (nr_good_pages) { |
2929 | swap_map[0] = SWAP_MAP_BAD; | |
2a8f9449 SL |
2930 | /* |
2931 | * Not mark the cluster free yet, no list | |
2932 | * operation involved | |
2933 | */ | |
2934 | inc_cluster_info_page(p, cluster_info, 0); | |
915d4d7b CEB |
2935 | p->max = maxpages; |
2936 | p->pages = nr_good_pages; | |
2937 | nr_extents = setup_swap_extents(p, span); | |
bdb8e3f6 CEB |
2938 | if (nr_extents < 0) |
2939 | return nr_extents; | |
915d4d7b CEB |
2940 | nr_good_pages = p->pages; |
2941 | } | |
2942 | if (!nr_good_pages) { | |
465c47fd | 2943 | pr_warn("Empty swap-file\n"); |
bdb8e3f6 | 2944 | return -EINVAL; |
915d4d7b CEB |
2945 | } |
2946 | ||
2a8f9449 SL |
2947 | if (!cluster_info) |
2948 | return nr_extents; | |
2949 | ||
235b6217 | 2950 | |
4b3ef9da HY |
2951 | /* |
2952 | * Reduce false cache line sharing between cluster_info and | |
2953 | * sharing same address space. | |
2954 | */ | |
235b6217 HY |
2955 | for (k = 0; k < SWAP_CLUSTER_COLS; k++) { |
2956 | j = (k + col) % SWAP_CLUSTER_COLS; | |
2957 | for (i = 0; i < DIV_ROUND_UP(nr_clusters, SWAP_CLUSTER_COLS); i++) { | |
2958 | idx = i * SWAP_CLUSTER_COLS + j; | |
2959 | if (idx >= nr_clusters) | |
2960 | continue; | |
2961 | if (cluster_count(&cluster_info[idx])) | |
2962 | continue; | |
2a8f9449 | 2963 | cluster_set_flag(&cluster_info[idx], CLUSTER_FLAG_FREE); |
6b534915 HY |
2964 | cluster_list_add_tail(&p->free_clusters, cluster_info, |
2965 | idx); | |
2a8f9449 | 2966 | } |
2a8f9449 | 2967 | } |
915d4d7b | 2968 | return nr_extents; |
915d4d7b CEB |
2969 | } |
2970 | ||
53cbb243 CEB |
2971 | SYSCALL_DEFINE2(swapon, const char __user *, specialfile, int, swap_flags) |
2972 | { | |
2973 | struct swap_info_struct *p; | |
91a27b2a | 2974 | struct filename *name; |
53cbb243 CEB |
2975 | struct file *swap_file = NULL; |
2976 | struct address_space *mapping; | |
51cc3a66 | 2977 | struct dentry *dentry; |
40531542 | 2978 | int prio; |
53cbb243 CEB |
2979 | int error; |
2980 | union swap_header *swap_header; | |
915d4d7b | 2981 | int nr_extents; |
53cbb243 CEB |
2982 | sector_t span; |
2983 | unsigned long maxpages; | |
53cbb243 | 2984 | unsigned char *swap_map = NULL; |
2a8f9449 | 2985 | struct swap_cluster_info *cluster_info = NULL; |
38b5faf4 | 2986 | unsigned long *frontswap_map = NULL; |
53cbb243 CEB |
2987 | struct page *page = NULL; |
2988 | struct inode *inode = NULL; | |
7cbf3192 | 2989 | bool inced_nr_rotate_swap = false; |
53cbb243 | 2990 | |
d15cab97 HD |
2991 | if (swap_flags & ~SWAP_FLAGS_VALID) |
2992 | return -EINVAL; | |
2993 | ||
53cbb243 CEB |
2994 | if (!capable(CAP_SYS_ADMIN)) |
2995 | return -EPERM; | |
2996 | ||
a2468cc9 AL |
2997 | if (!swap_avail_heads) |
2998 | return -ENOMEM; | |
2999 | ||
53cbb243 | 3000 | p = alloc_swap_info(); |
2542e513 CEB |
3001 | if (IS_ERR(p)) |
3002 | return PTR_ERR(p); | |
53cbb243 | 3003 | |
815c2c54 SL |
3004 | INIT_WORK(&p->discard_work, swap_discard_work); |
3005 | ||
1da177e4 | 3006 | name = getname(specialfile); |
1da177e4 | 3007 | if (IS_ERR(name)) { |
7de7fb6b | 3008 | error = PTR_ERR(name); |
1da177e4 | 3009 | name = NULL; |
bd69010b | 3010 | goto bad_swap; |
1da177e4 | 3011 | } |
669abf4e | 3012 | swap_file = file_open_name(name, O_RDWR|O_LARGEFILE, 0); |
1da177e4 | 3013 | if (IS_ERR(swap_file)) { |
7de7fb6b | 3014 | error = PTR_ERR(swap_file); |
1da177e4 | 3015 | swap_file = NULL; |
bd69010b | 3016 | goto bad_swap; |
1da177e4 LT |
3017 | } |
3018 | ||
3019 | p->swap_file = swap_file; | |
3020 | mapping = swap_file->f_mapping; | |
51cc3a66 | 3021 | dentry = swap_file->f_path.dentry; |
2130781e | 3022 | inode = mapping->host; |
6f179af8 | 3023 | |
4d0e1e10 CEB |
3024 | error = claim_swapfile(p, inode); |
3025 | if (unlikely(error)) | |
1da177e4 | 3026 | goto bad_swap; |
1da177e4 | 3027 | |
d795a90e | 3028 | inode_lock(inode); |
51cc3a66 HD |
3029 | if (d_unlinked(dentry) || cant_mount(dentry)) { |
3030 | error = -ENOENT; | |
3031 | goto bad_swap_unlock_inode; | |
3032 | } | |
d795a90e NA |
3033 | if (IS_SWAPFILE(inode)) { |
3034 | error = -EBUSY; | |
3035 | goto bad_swap_unlock_inode; | |
3036 | } | |
3037 | ||
1da177e4 LT |
3038 | /* |
3039 | * Read the swap header. | |
3040 | */ | |
7e0a1265 | 3041 | if (!mapping->a_ops->read_folio) { |
1da177e4 | 3042 | error = -EINVAL; |
d795a90e | 3043 | goto bad_swap_unlock_inode; |
1da177e4 | 3044 | } |
090d2b18 | 3045 | page = read_mapping_page(mapping, 0, swap_file); |
1da177e4 LT |
3046 | if (IS_ERR(page)) { |
3047 | error = PTR_ERR(page); | |
d795a90e | 3048 | goto bad_swap_unlock_inode; |
1da177e4 | 3049 | } |
81e33971 | 3050 | swap_header = kmap(page); |
1da177e4 | 3051 | |
ca8bd38b CEB |
3052 | maxpages = read_swap_header(p, swap_header, inode); |
3053 | if (unlikely(!maxpages)) { | |
1da177e4 | 3054 | error = -EINVAL; |
d795a90e | 3055 | goto bad_swap_unlock_inode; |
1da177e4 | 3056 | } |
886bb7e9 | 3057 | |
81e33971 | 3058 | /* OK, set up the swap map and apply the bad block list */ |
803d0c83 | 3059 | swap_map = vzalloc(maxpages); |
81e33971 HD |
3060 | if (!swap_map) { |
3061 | error = -ENOMEM; | |
d795a90e | 3062 | goto bad_swap_unlock_inode; |
81e33971 | 3063 | } |
f0571429 | 3064 | |
36d25489 | 3065 | if (p->bdev && bdev_stable_writes(p->bdev)) |
f0571429 MK |
3066 | p->flags |= SWP_STABLE_WRITES; |
3067 | ||
a8b456d0 | 3068 | if (p->bdev && p->bdev->bd_disk->fops->rw_page) |
539a6fea MK |
3069 | p->flags |= SWP_SYNCHRONOUS_IO; |
3070 | ||
10f0d2a5 | 3071 | if (p->bdev && bdev_nonrot(p->bdev)) { |
6f179af8 | 3072 | int cpu; |
235b6217 | 3073 | unsigned long ci, nr_cluster; |
6f179af8 | 3074 | |
2a8f9449 | 3075 | p->flags |= SWP_SOLIDSTATE; |
49070588 HY |
3076 | p->cluster_next_cpu = alloc_percpu(unsigned int); |
3077 | if (!p->cluster_next_cpu) { | |
3078 | error = -ENOMEM; | |
3079 | goto bad_swap_unlock_inode; | |
3080 | } | |
2a8f9449 SL |
3081 | /* |
3082 | * select a random position to start with to help wear leveling | |
3083 | * SSD | |
3084 | */ | |
49070588 HY |
3085 | for_each_possible_cpu(cpu) { |
3086 | per_cpu(*p->cluster_next_cpu, cpu) = | |
3087 | 1 + prandom_u32_max(p->highest_bit); | |
3088 | } | |
235b6217 | 3089 | nr_cluster = DIV_ROUND_UP(maxpages, SWAPFILE_CLUSTER); |
2a8f9449 | 3090 | |
778e1cdd | 3091 | cluster_info = kvcalloc(nr_cluster, sizeof(*cluster_info), |
54f180d3 | 3092 | GFP_KERNEL); |
2a8f9449 SL |
3093 | if (!cluster_info) { |
3094 | error = -ENOMEM; | |
d795a90e | 3095 | goto bad_swap_unlock_inode; |
2a8f9449 | 3096 | } |
235b6217 HY |
3097 | |
3098 | for (ci = 0; ci < nr_cluster; ci++) | |
3099 | spin_lock_init(&((cluster_info + ci)->lock)); | |
3100 | ||
ebc2a1a6 SL |
3101 | p->percpu_cluster = alloc_percpu(struct percpu_cluster); |
3102 | if (!p->percpu_cluster) { | |
3103 | error = -ENOMEM; | |
d795a90e | 3104 | goto bad_swap_unlock_inode; |
ebc2a1a6 | 3105 | } |
6f179af8 | 3106 | for_each_possible_cpu(cpu) { |
ebc2a1a6 | 3107 | struct percpu_cluster *cluster; |
6f179af8 | 3108 | cluster = per_cpu_ptr(p->percpu_cluster, cpu); |
ebc2a1a6 SL |
3109 | cluster_set_null(&cluster->index); |
3110 | } | |
7cbf3192 | 3111 | } else { |
81a0298b | 3112 | atomic_inc(&nr_rotate_swap); |
7cbf3192 OS |
3113 | inced_nr_rotate_swap = true; |
3114 | } | |
1da177e4 | 3115 | |
1421ef3c CEB |
3116 | error = swap_cgroup_swapon(p->type, maxpages); |
3117 | if (error) | |
d795a90e | 3118 | goto bad_swap_unlock_inode; |
1421ef3c | 3119 | |
915d4d7b | 3120 | nr_extents = setup_swap_map_and_extents(p, swap_header, swap_map, |
2a8f9449 | 3121 | cluster_info, maxpages, &span); |
915d4d7b CEB |
3122 | if (unlikely(nr_extents < 0)) { |
3123 | error = nr_extents; | |
d795a90e | 3124 | goto bad_swap_unlock_inode; |
1da177e4 | 3125 | } |
38b5faf4 | 3126 | /* frontswap enabled? set up bit-per-page map for frontswap */ |
8ea1d2a1 | 3127 | if (IS_ENABLED(CONFIG_FRONTSWAP)) |
778e1cdd KC |
3128 | frontswap_map = kvcalloc(BITS_TO_LONGS(maxpages), |
3129 | sizeof(long), | |
54f180d3 | 3130 | GFP_KERNEL); |
1da177e4 | 3131 | |
70200574 CH |
3132 | if ((swap_flags & SWAP_FLAG_DISCARD) && |
3133 | p->bdev && bdev_max_discard_sectors(p->bdev)) { | |
2a8f9449 SL |
3134 | /* |
3135 | * When discard is enabled for swap with no particular | |
3136 | * policy flagged, we set all swap discard flags here in | |
3137 | * order to sustain backward compatibility with older | |
3138 | * swapon(8) releases. | |
3139 | */ | |
3140 | p->flags |= (SWP_DISCARDABLE | SWP_AREA_DISCARD | | |
3141 | SWP_PAGE_DISCARD); | |
dcf6b7dd | 3142 | |
2a8f9449 SL |
3143 | /* |
3144 | * By flagging sys_swapon, a sysadmin can tell us to | |
3145 | * either do single-time area discards only, or to just | |
3146 | * perform discards for released swap page-clusters. | |
3147 | * Now it's time to adjust the p->flags accordingly. | |
3148 | */ | |
3149 | if (swap_flags & SWAP_FLAG_DISCARD_ONCE) | |
3150 | p->flags &= ~SWP_PAGE_DISCARD; | |
3151 | else if (swap_flags & SWAP_FLAG_DISCARD_PAGES) | |
3152 | p->flags &= ~SWP_AREA_DISCARD; | |
3153 | ||
3154 | /* issue a swapon-time discard if it's still required */ | |
3155 | if (p->flags & SWP_AREA_DISCARD) { | |
3156 | int err = discard_swap(p); | |
3157 | if (unlikely(err)) | |
3158 | pr_err("swapon: discard_swap(%p): %d\n", | |
3159 | p, err); | |
dcf6b7dd | 3160 | } |
20137a49 | 3161 | } |
6a6ba831 | 3162 | |
4b3ef9da HY |
3163 | error = init_swap_address_space(p->type, maxpages); |
3164 | if (error) | |
d795a90e | 3165 | goto bad_swap_unlock_inode; |
4b3ef9da | 3166 | |
dc617f29 DW |
3167 | /* |
3168 | * Flush any pending IO and dirty mappings before we start using this | |
3169 | * swap device. | |
3170 | */ | |
3171 | inode->i_flags |= S_SWAPFILE; | |
3172 | error = inode_drain_writes(inode); | |
3173 | if (error) { | |
3174 | inode->i_flags &= ~S_SWAPFILE; | |
822bca52 | 3175 | goto free_swap_address_space; |
dc617f29 DW |
3176 | } |
3177 | ||
fc0abb14 | 3178 | mutex_lock(&swapon_mutex); |
40531542 | 3179 | prio = -1; |
78ecba08 | 3180 | if (swap_flags & SWAP_FLAG_PREFER) |
40531542 | 3181 | prio = |
78ecba08 | 3182 | (swap_flags & SWAP_FLAG_PRIO_MASK) >> SWAP_FLAG_PRIO_SHIFT; |
2a8f9449 | 3183 | enable_swap_info(p, prio, swap_map, cluster_info, frontswap_map); |
c69dbfb8 | 3184 | |
756a025f | 3185 | pr_info("Adding %uk swap on %s. Priority:%d extents:%d across:%lluk %s%s%s%s%s\n", |
91a27b2a | 3186 | p->pages<<(PAGE_SHIFT-10), name->name, p->prio, |
c69dbfb8 CEB |
3187 | nr_extents, (unsigned long long)span<<(PAGE_SHIFT-10), |
3188 | (p->flags & SWP_SOLIDSTATE) ? "SS" : "", | |
38b5faf4 | 3189 | (p->flags & SWP_DISCARDABLE) ? "D" : "", |
dcf6b7dd RA |
3190 | (p->flags & SWP_AREA_DISCARD) ? "s" : "", |
3191 | (p->flags & SWP_PAGE_DISCARD) ? "c" : "", | |
38b5faf4 | 3192 | (frontswap_map) ? "FS" : ""); |
c69dbfb8 | 3193 | |
fc0abb14 | 3194 | mutex_unlock(&swapon_mutex); |
66d7dd51 KS |
3195 | atomic_inc(&proc_poll_event); |
3196 | wake_up_interruptible(&proc_poll_wait); | |
3197 | ||
1da177e4 LT |
3198 | error = 0; |
3199 | goto out; | |
822bca52 ML |
3200 | free_swap_address_space: |
3201 | exit_swap_address_space(p->type); | |
d795a90e NA |
3202 | bad_swap_unlock_inode: |
3203 | inode_unlock(inode); | |
1da177e4 | 3204 | bad_swap: |
ebc2a1a6 SL |
3205 | free_percpu(p->percpu_cluster); |
3206 | p->percpu_cluster = NULL; | |
49070588 HY |
3207 | free_percpu(p->cluster_next_cpu); |
3208 | p->cluster_next_cpu = NULL; | |
bd69010b | 3209 | if (inode && S_ISBLK(inode->i_mode) && p->bdev) { |
f2090d2d CEB |
3210 | set_blocksize(p->bdev, p->old_block_size); |
3211 | blkdev_put(p->bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL); | |
1da177e4 | 3212 | } |
d795a90e | 3213 | inode = NULL; |
4cd3bb10 | 3214 | destroy_swap_extents(p); |
e8e6c2ec | 3215 | swap_cgroup_swapoff(p->type); |
5d337b91 | 3216 | spin_lock(&swap_lock); |
1da177e4 | 3217 | p->swap_file = NULL; |
1da177e4 | 3218 | p->flags = 0; |
5d337b91 | 3219 | spin_unlock(&swap_lock); |
1da177e4 | 3220 | vfree(swap_map); |
8606a1a9 | 3221 | kvfree(cluster_info); |
b6b1fd2a | 3222 | kvfree(frontswap_map); |
7cbf3192 OS |
3223 | if (inced_nr_rotate_swap) |
3224 | atomic_dec(&nr_rotate_swap); | |
d795a90e | 3225 | if (swap_file) |
1da177e4 LT |
3226 | filp_close(swap_file, NULL); |
3227 | out: | |
3228 | if (page && !IS_ERR(page)) { | |
3229 | kunmap(page); | |
09cbfeaf | 3230 | put_page(page); |
1da177e4 LT |
3231 | } |
3232 | if (name) | |
3233 | putname(name); | |
1638045c | 3234 | if (inode) |
5955102c | 3235 | inode_unlock(inode); |
039939a6 TC |
3236 | if (!error) |
3237 | enable_swap_slots_cache(); | |
1da177e4 LT |
3238 | return error; |
3239 | } | |
3240 | ||
3241 | void si_swapinfo(struct sysinfo *val) | |
3242 | { | |
efa90a98 | 3243 | unsigned int type; |
1da177e4 LT |
3244 | unsigned long nr_to_be_unused = 0; |
3245 | ||
5d337b91 | 3246 | spin_lock(&swap_lock); |
efa90a98 HD |
3247 | for (type = 0; type < nr_swapfiles; type++) { |
3248 | struct swap_info_struct *si = swap_info[type]; | |
3249 | ||
3250 | if ((si->flags & SWP_USED) && !(si->flags & SWP_WRITEOK)) | |
c8945306 | 3251 | nr_to_be_unused += READ_ONCE(si->inuse_pages); |
1da177e4 | 3252 | } |
ec8acf20 | 3253 | val->freeswap = atomic_long_read(&nr_swap_pages) + nr_to_be_unused; |
1da177e4 | 3254 | val->totalswap = total_swap_pages + nr_to_be_unused; |
5d337b91 | 3255 | spin_unlock(&swap_lock); |
1da177e4 LT |
3256 | } |
3257 | ||
3258 | /* | |
3259 | * Verify that a swap entry is valid and increment its swap map count. | |
3260 | * | |
355cfa73 KH |
3261 | * Returns error code in following case. |
3262 | * - success -> 0 | |
3263 | * - swp_entry is invalid -> EINVAL | |
3264 | * - swp_entry is migration entry -> EINVAL | |
3265 | * - swap-cache reference is requested but there is already one. -> EEXIST | |
3266 | * - swap-cache reference is requested but the entry is not used. -> ENOENT | |
570a335b | 3267 | * - swap-mapped reference requested but needs continued swap count. -> ENOMEM |
1da177e4 | 3268 | */ |
8d69aaee | 3269 | static int __swap_duplicate(swp_entry_t entry, unsigned char usage) |
1da177e4 | 3270 | { |
73c34b6a | 3271 | struct swap_info_struct *p; |
235b6217 | 3272 | struct swap_cluster_info *ci; |
c10d38cc | 3273 | unsigned long offset; |
8d69aaee HD |
3274 | unsigned char count; |
3275 | unsigned char has_cache; | |
9d9a0334 | 3276 | int err; |
1da177e4 | 3277 | |
eb085574 | 3278 | p = get_swap_device(entry); |
c10d38cc | 3279 | if (!p) |
9d9a0334 | 3280 | return -EINVAL; |
235b6217 | 3281 | |
eb085574 | 3282 | offset = swp_offset(entry); |
235b6217 | 3283 | ci = lock_cluster_or_swap_info(p, offset); |
355cfa73 | 3284 | |
253d553b | 3285 | count = p->swap_map[offset]; |
edfe23da SL |
3286 | |
3287 | /* | |
3288 | * swapin_readahead() doesn't check if a swap entry is valid, so the | |
3289 | * swap entry could be SWAP_MAP_BAD. Check here with lock held. | |
3290 | */ | |
3291 | if (unlikely(swap_count(count) == SWAP_MAP_BAD)) { | |
3292 | err = -ENOENT; | |
3293 | goto unlock_out; | |
3294 | } | |
3295 | ||
253d553b HD |
3296 | has_cache = count & SWAP_HAS_CACHE; |
3297 | count &= ~SWAP_HAS_CACHE; | |
3298 | err = 0; | |
355cfa73 | 3299 | |
253d553b | 3300 | if (usage == SWAP_HAS_CACHE) { |
355cfa73 KH |
3301 | |
3302 | /* set SWAP_HAS_CACHE if there is no cache and entry is used */ | |
253d553b HD |
3303 | if (!has_cache && count) |
3304 | has_cache = SWAP_HAS_CACHE; | |
3305 | else if (has_cache) /* someone else added cache */ | |
3306 | err = -EEXIST; | |
3307 | else /* no users remaining */ | |
3308 | err = -ENOENT; | |
355cfa73 KH |
3309 | |
3310 | } else if (count || has_cache) { | |
253d553b | 3311 | |
570a335b HD |
3312 | if ((count & ~COUNT_CONTINUED) < SWAP_MAP_MAX) |
3313 | count += usage; | |
3314 | else if ((count & ~COUNT_CONTINUED) > SWAP_MAP_MAX) | |
253d553b | 3315 | err = -EINVAL; |
570a335b HD |
3316 | else if (swap_count_continued(p, offset, count)) |
3317 | count = COUNT_CONTINUED; | |
3318 | else | |
3319 | err = -ENOMEM; | |
355cfa73 | 3320 | } else |
253d553b HD |
3321 | err = -ENOENT; /* unused swap entry */ |
3322 | ||
a449bf58 | 3323 | WRITE_ONCE(p->swap_map[offset], count | has_cache); |
253d553b | 3324 | |
355cfa73 | 3325 | unlock_out: |
235b6217 | 3326 | unlock_cluster_or_swap_info(p, ci); |
dab8dfff | 3327 | put_swap_device(p); |
253d553b | 3328 | return err; |
1da177e4 | 3329 | } |
253d553b | 3330 | |
aaa46865 HD |
3331 | /* |
3332 | * Help swapoff by noting that swap entry belongs to shmem/tmpfs | |
3333 | * (in which case its reference count is never incremented). | |
3334 | */ | |
3335 | void swap_shmem_alloc(swp_entry_t entry) | |
3336 | { | |
3337 | __swap_duplicate(entry, SWAP_MAP_SHMEM); | |
3338 | } | |
3339 | ||
355cfa73 | 3340 | /* |
08259d58 HD |
3341 | * Increase reference count of swap entry by 1. |
3342 | * Returns 0 for success, or -ENOMEM if a swap_count_continuation is required | |
3343 | * but could not be atomically allocated. Returns 0, just as if it succeeded, | |
3344 | * if __swap_duplicate() fails for another reason (-EINVAL or -ENOENT), which | |
3345 | * might occur if a page table entry has got corrupted. | |
355cfa73 | 3346 | */ |
570a335b | 3347 | int swap_duplicate(swp_entry_t entry) |
355cfa73 | 3348 | { |
570a335b HD |
3349 | int err = 0; |
3350 | ||
3351 | while (!err && __swap_duplicate(entry, 1) == -ENOMEM) | |
3352 | err = add_swap_count_continuation(entry, GFP_ATOMIC); | |
3353 | return err; | |
355cfa73 | 3354 | } |
1da177e4 | 3355 | |
cb4b86ba | 3356 | /* |
355cfa73 KH |
3357 | * @entry: swap entry for which we allocate swap cache. |
3358 | * | |
73c34b6a | 3359 | * Called when allocating swap cache for existing swap entry, |
355cfa73 | 3360 | * This can return error codes. Returns 0 at success. |
3eeba135 | 3361 | * -EEXIST means there is a swap cache. |
355cfa73 | 3362 | * Note: return code is different from swap_duplicate(). |
cb4b86ba KH |
3363 | */ |
3364 | int swapcache_prepare(swp_entry_t entry) | |
3365 | { | |
253d553b | 3366 | return __swap_duplicate(entry, SWAP_HAS_CACHE); |
cb4b86ba KH |
3367 | } |
3368 | ||
0bcac06f MK |
3369 | struct swap_info_struct *swp_swap_info(swp_entry_t entry) |
3370 | { | |
c10d38cc | 3371 | return swap_type_to_swap_info(swp_type(entry)); |
0bcac06f MK |
3372 | } |
3373 | ||
f981c595 MG |
3374 | struct swap_info_struct *page_swap_info(struct page *page) |
3375 | { | |
0bcac06f MK |
3376 | swp_entry_t entry = { .val = page_private(page) }; |
3377 | return swp_swap_info(entry); | |
f981c595 MG |
3378 | } |
3379 | ||
3380 | /* | |
2f52578f | 3381 | * out-of-line methods to avoid include hell. |
f981c595 | 3382 | */ |
2f52578f | 3383 | struct address_space *swapcache_mapping(struct folio *folio) |
f981c595 | 3384 | { |
2f52578f | 3385 | return page_swap_info(&folio->page)->swap_file->f_mapping; |
f981c595 | 3386 | } |
2f52578f | 3387 | EXPORT_SYMBOL_GPL(swapcache_mapping); |
f981c595 MG |
3388 | |
3389 | pgoff_t __page_file_index(struct page *page) | |
3390 | { | |
3391 | swp_entry_t swap = { .val = page_private(page) }; | |
f981c595 MG |
3392 | return swp_offset(swap); |
3393 | } | |
3394 | EXPORT_SYMBOL_GPL(__page_file_index); | |
3395 | ||
570a335b HD |
3396 | /* |
3397 | * add_swap_count_continuation - called when a swap count is duplicated | |
3398 | * beyond SWAP_MAP_MAX, it allocates a new page and links that to the entry's | |
3399 | * page of the original vmalloc'ed swap_map, to hold the continuation count | |
3400 | * (for that entry and for its neighbouring PAGE_SIZE swap entries). Called | |
3401 | * again when count is duplicated beyond SWAP_MAP_MAX * SWAP_CONT_MAX, etc. | |
3402 | * | |
3403 | * These continuation pages are seldom referenced: the common paths all work | |
3404 | * on the original swap_map, only referring to a continuation page when the | |
3405 | * low "digit" of a count is incremented or decremented through SWAP_MAP_MAX. | |
3406 | * | |
3407 | * add_swap_count_continuation(, GFP_ATOMIC) can be called while holding | |
3408 | * page table locks; if it fails, add_swap_count_continuation(, GFP_KERNEL) | |
3409 | * can be called after dropping locks. | |
3410 | */ | |
3411 | int add_swap_count_continuation(swp_entry_t entry, gfp_t gfp_mask) | |
3412 | { | |
3413 | struct swap_info_struct *si; | |
235b6217 | 3414 | struct swap_cluster_info *ci; |
570a335b HD |
3415 | struct page *head; |
3416 | struct page *page; | |
3417 | struct page *list_page; | |
3418 | pgoff_t offset; | |
3419 | unsigned char count; | |
eb085574 | 3420 | int ret = 0; |
570a335b HD |
3421 | |
3422 | /* | |
3423 | * When debugging, it's easier to use __GFP_ZERO here; but it's better | |
3424 | * for latency not to zero a page while GFP_ATOMIC and holding locks. | |
3425 | */ | |
3426 | page = alloc_page(gfp_mask | __GFP_HIGHMEM); | |
3427 | ||
eb085574 | 3428 | si = get_swap_device(entry); |
570a335b HD |
3429 | if (!si) { |
3430 | /* | |
3431 | * An acceptable race has occurred since the failing | |
eb085574 | 3432 | * __swap_duplicate(): the swap device may be swapoff |
570a335b HD |
3433 | */ |
3434 | goto outer; | |
3435 | } | |
eb085574 | 3436 | spin_lock(&si->lock); |
570a335b HD |
3437 | |
3438 | offset = swp_offset(entry); | |
235b6217 HY |
3439 | |
3440 | ci = lock_cluster(si, offset); | |
3441 | ||
d8aa24e0 | 3442 | count = swap_count(si->swap_map[offset]); |
570a335b HD |
3443 | |
3444 | if ((count & ~COUNT_CONTINUED) != SWAP_MAP_MAX) { | |
3445 | /* | |
3446 | * The higher the swap count, the more likely it is that tasks | |
3447 | * will race to add swap count continuation: we need to avoid | |
3448 | * over-provisioning. | |
3449 | */ | |
3450 | goto out; | |
3451 | } | |
3452 | ||
3453 | if (!page) { | |
eb085574 HY |
3454 | ret = -ENOMEM; |
3455 | goto out; | |
570a335b HD |
3456 | } |
3457 | ||
3458 | /* | |
3459 | * We are fortunate that although vmalloc_to_page uses pte_offset_map, | |
2de1a7e4 SJ |
3460 | * no architecture is using highmem pages for kernel page tables: so it |
3461 | * will not corrupt the GFP_ATOMIC caller's atomic page table kmaps. | |
570a335b HD |
3462 | */ |
3463 | head = vmalloc_to_page(si->swap_map + offset); | |
3464 | offset &= ~PAGE_MASK; | |
3465 | ||
2628bd6f | 3466 | spin_lock(&si->cont_lock); |
570a335b HD |
3467 | /* |
3468 | * Page allocation does not initialize the page's lru field, | |
3469 | * but it does always reset its private field. | |
3470 | */ | |
3471 | if (!page_private(head)) { | |
3472 | BUG_ON(count & COUNT_CONTINUED); | |
3473 | INIT_LIST_HEAD(&head->lru); | |
3474 | set_page_private(head, SWP_CONTINUED); | |
3475 | si->flags |= SWP_CONTINUED; | |
3476 | } | |
3477 | ||
3478 | list_for_each_entry(list_page, &head->lru, lru) { | |
3479 | unsigned char *map; | |
3480 | ||
3481 | /* | |
3482 | * If the previous map said no continuation, but we've found | |
3483 | * a continuation page, free our allocation and use this one. | |
3484 | */ | |
3485 | if (!(count & COUNT_CONTINUED)) | |
2628bd6f | 3486 | goto out_unlock_cont; |
570a335b | 3487 | |
9b04c5fe | 3488 | map = kmap_atomic(list_page) + offset; |
570a335b | 3489 | count = *map; |
9b04c5fe | 3490 | kunmap_atomic(map); |
570a335b HD |
3491 | |
3492 | /* | |
3493 | * If this continuation count now has some space in it, | |
3494 | * free our allocation and use this one. | |
3495 | */ | |
3496 | if ((count & ~COUNT_CONTINUED) != SWAP_CONT_MAX) | |
2628bd6f | 3497 | goto out_unlock_cont; |
570a335b HD |
3498 | } |
3499 | ||
3500 | list_add_tail(&page->lru, &head->lru); | |
3501 | page = NULL; /* now it's attached, don't free it */ | |
2628bd6f HY |
3502 | out_unlock_cont: |
3503 | spin_unlock(&si->cont_lock); | |
570a335b | 3504 | out: |
235b6217 | 3505 | unlock_cluster(ci); |
ec8acf20 | 3506 | spin_unlock(&si->lock); |
eb085574 | 3507 | put_swap_device(si); |
570a335b HD |
3508 | outer: |
3509 | if (page) | |
3510 | __free_page(page); | |
eb085574 | 3511 | return ret; |
570a335b HD |
3512 | } |
3513 | ||
3514 | /* | |
3515 | * swap_count_continued - when the original swap_map count is incremented | |
3516 | * from SWAP_MAP_MAX, check if there is already a continuation page to carry | |
3517 | * into, carry if so, or else fail until a new continuation page is allocated; | |
3518 | * when the original swap_map count is decremented from 0 with continuation, | |
3519 | * borrow from the continuation and report whether it still holds more. | |
235b6217 HY |
3520 | * Called while __swap_duplicate() or swap_entry_free() holds swap or cluster |
3521 | * lock. | |
570a335b HD |
3522 | */ |
3523 | static bool swap_count_continued(struct swap_info_struct *si, | |
3524 | pgoff_t offset, unsigned char count) | |
3525 | { | |
3526 | struct page *head; | |
3527 | struct page *page; | |
3528 | unsigned char *map; | |
2628bd6f | 3529 | bool ret; |
570a335b HD |
3530 | |
3531 | head = vmalloc_to_page(si->swap_map + offset); | |
3532 | if (page_private(head) != SWP_CONTINUED) { | |
3533 | BUG_ON(count & COUNT_CONTINUED); | |
3534 | return false; /* need to add count continuation */ | |
3535 | } | |
3536 | ||
2628bd6f | 3537 | spin_lock(&si->cont_lock); |
570a335b | 3538 | offset &= ~PAGE_MASK; |
213516ac | 3539 | page = list_next_entry(head, lru); |
9b04c5fe | 3540 | map = kmap_atomic(page) + offset; |
570a335b HD |
3541 | |
3542 | if (count == SWAP_MAP_MAX) /* initial increment from swap_map */ | |
3543 | goto init_map; /* jump over SWAP_CONT_MAX checks */ | |
3544 | ||
3545 | if (count == (SWAP_MAP_MAX | COUNT_CONTINUED)) { /* incrementing */ | |
3546 | /* | |
3547 | * Think of how you add 1 to 999 | |
3548 | */ | |
3549 | while (*map == (SWAP_CONT_MAX | COUNT_CONTINUED)) { | |
9b04c5fe | 3550 | kunmap_atomic(map); |
213516ac | 3551 | page = list_next_entry(page, lru); |
570a335b | 3552 | BUG_ON(page == head); |
9b04c5fe | 3553 | map = kmap_atomic(page) + offset; |
570a335b HD |
3554 | } |
3555 | if (*map == SWAP_CONT_MAX) { | |
9b04c5fe | 3556 | kunmap_atomic(map); |
213516ac | 3557 | page = list_next_entry(page, lru); |
2628bd6f HY |
3558 | if (page == head) { |
3559 | ret = false; /* add count continuation */ | |
3560 | goto out; | |
3561 | } | |
9b04c5fe | 3562 | map = kmap_atomic(page) + offset; |
570a335b HD |
3563 | init_map: *map = 0; /* we didn't zero the page */ |
3564 | } | |
3565 | *map += 1; | |
9b04c5fe | 3566 | kunmap_atomic(map); |
213516ac | 3567 | while ((page = list_prev_entry(page, lru)) != head) { |
9b04c5fe | 3568 | map = kmap_atomic(page) + offset; |
570a335b | 3569 | *map = COUNT_CONTINUED; |
9b04c5fe | 3570 | kunmap_atomic(map); |
570a335b | 3571 | } |
2628bd6f | 3572 | ret = true; /* incremented */ |
570a335b HD |
3573 | |
3574 | } else { /* decrementing */ | |
3575 | /* | |
3576 | * Think of how you subtract 1 from 1000 | |
3577 | */ | |
3578 | BUG_ON(count != COUNT_CONTINUED); | |
3579 | while (*map == COUNT_CONTINUED) { | |
9b04c5fe | 3580 | kunmap_atomic(map); |
213516ac | 3581 | page = list_next_entry(page, lru); |
570a335b | 3582 | BUG_ON(page == head); |
9b04c5fe | 3583 | map = kmap_atomic(page) + offset; |
570a335b HD |
3584 | } |
3585 | BUG_ON(*map == 0); | |
3586 | *map -= 1; | |
3587 | if (*map == 0) | |
3588 | count = 0; | |
9b04c5fe | 3589 | kunmap_atomic(map); |
213516ac | 3590 | while ((page = list_prev_entry(page, lru)) != head) { |
9b04c5fe | 3591 | map = kmap_atomic(page) + offset; |
570a335b HD |
3592 | *map = SWAP_CONT_MAX | count; |
3593 | count = COUNT_CONTINUED; | |
9b04c5fe | 3594 | kunmap_atomic(map); |
570a335b | 3595 | } |
2628bd6f | 3596 | ret = count == COUNT_CONTINUED; |
570a335b | 3597 | } |
2628bd6f HY |
3598 | out: |
3599 | spin_unlock(&si->cont_lock); | |
3600 | return ret; | |
570a335b HD |
3601 | } |
3602 | ||
3603 | /* | |
3604 | * free_swap_count_continuations - swapoff free all the continuation pages | |
3605 | * appended to the swap_map, after swap_map is quiesced, before vfree'ing it. | |
3606 | */ | |
3607 | static void free_swap_count_continuations(struct swap_info_struct *si) | |
3608 | { | |
3609 | pgoff_t offset; | |
3610 | ||
3611 | for (offset = 0; offset < si->max; offset += PAGE_SIZE) { | |
3612 | struct page *head; | |
3613 | head = vmalloc_to_page(si->swap_map + offset); | |
3614 | if (page_private(head)) { | |
0d576d20 GT |
3615 | struct page *page, *next; |
3616 | ||
3617 | list_for_each_entry_safe(page, next, &head->lru, lru) { | |
3618 | list_del(&page->lru); | |
570a335b HD |
3619 | __free_page(page); |
3620 | } | |
3621 | } | |
3622 | } | |
3623 | } | |
a2468cc9 | 3624 | |
2cf85583 | 3625 | #if defined(CONFIG_MEMCG) && defined(CONFIG_BLK_CGROUP) |
01c4b28c | 3626 | void __cgroup_throttle_swaprate(struct page *page, gfp_t gfp_mask) |
2cf85583 TH |
3627 | { |
3628 | struct swap_info_struct *si, *next; | |
6caa6a07 JW |
3629 | int nid = page_to_nid(page); |
3630 | ||
3631 | if (!(gfp_mask & __GFP_IO)) | |
2cf85583 TH |
3632 | return; |
3633 | ||
3634 | if (!blk_cgroup_congested()) | |
3635 | return; | |
3636 | ||
3637 | /* | |
3638 | * We've already scheduled a throttle, avoid taking the global swap | |
3639 | * lock. | |
3640 | */ | |
3641 | if (current->throttle_queue) | |
3642 | return; | |
3643 | ||
3644 | spin_lock(&swap_avail_lock); | |
6caa6a07 JW |
3645 | plist_for_each_entry_safe(si, next, &swap_avail_heads[nid], |
3646 | avail_lists[nid]) { | |
2cf85583 | 3647 | if (si->bdev) { |
6caa6a07 | 3648 | blkcg_schedule_throttle(bdev_get_queue(si->bdev), true); |
2cf85583 TH |
3649 | break; |
3650 | } | |
3651 | } | |
3652 | spin_unlock(&swap_avail_lock); | |
3653 | } | |
3654 | #endif | |
3655 | ||
a2468cc9 AL |
3656 | static int __init swapfile_init(void) |
3657 | { | |
3658 | int nid; | |
3659 | ||
3660 | swap_avail_heads = kmalloc_array(nr_node_ids, sizeof(struct plist_head), | |
3661 | GFP_KERNEL); | |
3662 | if (!swap_avail_heads) { | |
3663 | pr_emerg("Not enough memory for swap heads, swap is disabled\n"); | |
3664 | return -ENOMEM; | |
3665 | } | |
3666 | ||
3667 | for_each_node(nid) | |
3668 | plist_head_init(&swap_avail_heads[nid]); | |
3669 | ||
be45a490 PX |
3670 | swapfile_maximum_size = arch_max_swapfile_size(); |
3671 | ||
5154e607 PX |
3672 | #ifdef CONFIG_MIGRATION |
3673 | if (swapfile_maximum_size >= (1UL << SWP_MIG_TOTAL_BITS)) | |
3674 | swap_migration_ad_supported = true; | |
3675 | #endif /* CONFIG_MIGRATION */ | |
3676 | ||
a2468cc9 AL |
3677 | return 0; |
3678 | } | |
3679 | subsys_initcall(swapfile_init); |