Commit | Line | Data |
---|---|---|
457c8996 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
1da177e4 LT |
2 | /* |
3 | * linux/mm/swapfile.c | |
4 | * | |
5 | * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds | |
6 | * Swap reorganised 29.12.95, Stephen Tweedie | |
7 | */ | |
8 | ||
c97ab271 | 9 | #include <linux/blkdev.h> |
1da177e4 | 10 | #include <linux/mm.h> |
6e84f315 | 11 | #include <linux/sched/mm.h> |
29930025 | 12 | #include <linux/sched/task.h> |
1da177e4 LT |
13 | #include <linux/hugetlb.h> |
14 | #include <linux/mman.h> | |
15 | #include <linux/slab.h> | |
16 | #include <linux/kernel_stat.h> | |
17 | #include <linux/swap.h> | |
18 | #include <linux/vmalloc.h> | |
19 | #include <linux/pagemap.h> | |
20 | #include <linux/namei.h> | |
072441e2 | 21 | #include <linux/shmem_fs.h> |
e41d12f5 | 22 | #include <linux/blk-cgroup.h> |
20137a49 | 23 | #include <linux/random.h> |
1da177e4 LT |
24 | #include <linux/writeback.h> |
25 | #include <linux/proc_fs.h> | |
26 | #include <linux/seq_file.h> | |
27 | #include <linux/init.h> | |
5ad64688 | 28 | #include <linux/ksm.h> |
1da177e4 LT |
29 | #include <linux/rmap.h> |
30 | #include <linux/security.h> | |
31 | #include <linux/backing-dev.h> | |
fc0abb14 | 32 | #include <linux/mutex.h> |
c59ede7b | 33 | #include <linux/capability.h> |
1da177e4 | 34 | #include <linux/syscalls.h> |
8a9f3ccd | 35 | #include <linux/memcontrol.h> |
66d7dd51 | 36 | #include <linux/poll.h> |
72788c38 | 37 | #include <linux/oom.h> |
38b5faf4 | 38 | #include <linux/swapfile.h> |
f981c595 | 39 | #include <linux/export.h> |
155b5f88 | 40 | #include <linux/sort.h> |
63d8620e | 41 | #include <linux/completion.h> |
07f44ac3 | 42 | #include <linux/suspend.h> |
42c06a0e | 43 | #include <linux/zswap.h> |
8b7787a5 | 44 | #include <linux/plist.h> |
1da177e4 | 45 | |
1da177e4 LT |
46 | #include <asm/tlbflush.h> |
47 | #include <linux/swapops.h> | |
5d1ea48b | 48 | #include <linux/swap_cgroup.h> |
00cde042 | 49 | #include "internal.h" |
014bb1de | 50 | #include "swap.h" |
1da177e4 | 51 | |
570a335b HD |
52 | static bool swap_count_continued(struct swap_info_struct *, pgoff_t, |
53 | unsigned char); | |
54 | static void free_swap_count_continuations(struct swap_info_struct *); | |
64944ef6 KS |
55 | static void swap_entries_free(struct swap_info_struct *si, |
56 | struct swap_cluster_info *ci, | |
57 | swp_entry_t entry, unsigned int nr_pages); | |
27701521 | 58 | static void swap_range_alloc(struct swap_info_struct *si, |
5f843a9a | 59 | unsigned int nr_entries); |
862590ac | 60 | static bool folio_swapcache_freeable(struct folio *folio); |
0b310d9c KS |
61 | static struct swap_cluster_info *lock_cluster(struct swap_info_struct *si, |
62 | unsigned long offset); | |
4f79384a | 63 | static inline void unlock_cluster(struct swap_cluster_info *ci); |
570a335b | 64 | |
633423a0 | 65 | static DEFINE_SPINLOCK(swap_lock); |
7c363b8c | 66 | static unsigned int nr_swapfiles; |
ec8acf20 | 67 | atomic_long_t nr_swap_pages; |
fb0fec50 CW |
68 | /* |
69 | * Some modules use swappable objects and may try to swap them out under | |
70 | * memory pressure (via the shrinker). Before doing so, they may wish to | |
71 | * check to see if any swap space is available. | |
72 | */ | |
73 | EXPORT_SYMBOL_GPL(nr_swap_pages); | |
ec8acf20 | 74 | /* protected with swap_lock. reading in vm_swap_full() doesn't need lock */ |
1da177e4 | 75 | long total_swap_pages; |
a2468cc9 | 76 | static int least_priority = -1; |
be45a490 | 77 | unsigned long swapfile_maximum_size; |
5154e607 PX |
78 | #ifdef CONFIG_MIGRATION |
79 | bool swap_migration_ad_supported; | |
80 | #endif /* CONFIG_MIGRATION */ | |
1da177e4 | 81 | |
1da177e4 LT |
82 | static const char Bad_file[] = "Bad swap file entry "; |
83 | static const char Unused_file[] = "Unused swap file entry "; | |
84 | static const char Bad_offset[] = "Bad swap offset entry "; | |
85 | static const char Unused_offset[] = "Unused swap offset entry "; | |
86 | ||
adfab836 DS |
87 | /* |
88 | * all active swap_info_structs | |
89 | * protected with swap_lock, and ordered by priority. | |
90 | */ | |
633423a0 | 91 | static PLIST_HEAD(swap_active_head); |
18ab4d4c DS |
92 | |
93 | /* | |
94 | * all available (active, not full) swap_info_structs | |
95 | * protected with swap_avail_lock, ordered by priority. | |
e2e3fdc7 | 96 | * This is used by folio_alloc_swap() instead of swap_active_head |
18ab4d4c | 97 | * because swap_active_head includes all swap_info_structs, |
e2e3fdc7 | 98 | * but folio_alloc_swap() doesn't need to look at full ones. |
18ab4d4c DS |
99 | * This uses its own lock instead of swap_lock because when a |
100 | * swap_info_struct changes between not-full/full, it needs to | |
101 | * add/remove itself to/from this list, but the swap_info_struct->lock | |
102 | * is held and the locking order requires swap_lock to be taken | |
103 | * before any swap_info_struct->lock. | |
104 | */ | |
bfc6b1ca | 105 | static struct plist_head *swap_avail_heads; |
18ab4d4c | 106 | static DEFINE_SPINLOCK(swap_avail_lock); |
1da177e4 | 107 | |
42c06a0e | 108 | static struct swap_info_struct *swap_info[MAX_SWAPFILES]; |
1da177e4 | 109 | |
fc0abb14 | 110 | static DEFINE_MUTEX(swapon_mutex); |
1da177e4 | 111 | |
66d7dd51 KS |
112 | static DECLARE_WAIT_QUEUE_HEAD(proc_poll_wait); |
113 | /* Activity counter to indicate that a swapon or swapoff has occurred */ | |
114 | static atomic_t proc_poll_event = ATOMIC_INIT(0); | |
115 | ||
81a0298b HY |
116 | atomic_t nr_rotate_swap = ATOMIC_INIT(0); |
117 | ||
1b7e9002 KS |
118 | struct percpu_swap_cluster { |
119 | struct swap_info_struct *si[SWAP_NR_ORDERS]; | |
120 | unsigned long offset[SWAP_NR_ORDERS]; | |
121 | local_lock_t lock; | |
122 | }; | |
123 | ||
124 | static DEFINE_PER_CPU(struct percpu_swap_cluster, percpu_swap_cluster) = { | |
125 | .si = { NULL }, | |
126 | .offset = { SWAP_ENTRY_INVALID }, | |
127 | .lock = INIT_LOCAL_LOCK(), | |
128 | }; | |
129 | ||
c10d38cc DJ |
130 | static struct swap_info_struct *swap_type_to_swap_info(int type) |
131 | { | |
a4b45114 | 132 | if (type >= MAX_SWAPFILES) |
c10d38cc DJ |
133 | return NULL; |
134 | ||
a4b45114 | 135 | return READ_ONCE(swap_info[type]); /* rcu_dereference() */ |
c10d38cc DJ |
136 | } |
137 | ||
8d69aaee | 138 | static inline unsigned char swap_count(unsigned char ent) |
355cfa73 | 139 | { |
955c97f0 | 140 | return ent & ~SWAP_HAS_CACHE; /* may include COUNT_CONTINUED flag */ |
355cfa73 KH |
141 | } |
142 | ||
b228386c KS |
143 | /* |
144 | * Use the second highest bit of inuse_pages counter as the indicator | |
145 | * if one swap device is on the available plist, so the atomic can | |
146 | * still be updated arithmetically while having special data embedded. | |
147 | * | |
148 | * inuse_pages counter is the only thing indicating if a device should | |
149 | * be on avail_lists or not (except swapon / swapoff). By embedding the | |
150 | * off-list bit in the atomic counter, updates no longer need any lock | |
151 | * to check the list status. | |
152 | * | |
153 | * This bit will be set if the device is not on the plist and not | |
154 | * usable, will be cleared if the device is on the plist. | |
155 | */ | |
156 | #define SWAP_USAGE_OFFLIST_BIT (1UL << (BITS_PER_TYPE(atomic_t) - 2)) | |
157 | #define SWAP_USAGE_COUNTER_MASK (~SWAP_USAGE_OFFLIST_BIT) | |
158 | static long swap_usage_in_pages(struct swap_info_struct *si) | |
159 | { | |
160 | return atomic_long_read(&si->inuse_pages) & SWAP_USAGE_COUNTER_MASK; | |
161 | } | |
162 | ||
bcd49e86 HY |
163 | /* Reclaim the swap entry anyway if possible */ |
164 | #define TTRS_ANYWAY 0x1 | |
165 | /* | |
166 | * Reclaim the swap entry if there are no more mappings of the | |
167 | * corresponding page | |
168 | */ | |
169 | #define TTRS_UNMAPPED 0x2 | |
862590ac | 170 | /* Reclaim the swap entry if swap is getting full */ |
bcd49e86 | 171 | #define TTRS_FULL 0x4 |
862590ac | 172 | |
b4735d94 | 173 | static bool swap_only_has_cache(struct swap_info_struct *si, |
862590ac KS |
174 | unsigned long offset, int nr_pages) |
175 | { | |
176 | unsigned char *map = si->swap_map + offset; | |
177 | unsigned char *map_end = map + nr_pages; | |
178 | ||
179 | do { | |
180 | VM_BUG_ON(!(*map & SWAP_HAS_CACHE)); | |
181 | if (*map != SWAP_HAS_CACHE) | |
182 | return false; | |
183 | } while (++map < map_end); | |
184 | ||
185 | return true; | |
186 | } | |
bcd49e86 | 187 | |
bea67dcc BS |
188 | static bool swap_is_last_map(struct swap_info_struct *si, |
189 | unsigned long offset, int nr_pages, bool *has_cache) | |
190 | { | |
191 | unsigned char *map = si->swap_map + offset; | |
192 | unsigned char *map_end = map + nr_pages; | |
193 | unsigned char count = *map; | |
194 | ||
f2252acf | 195 | if (swap_count(count) != 1 && swap_count(count) != SWAP_MAP_SHMEM) |
bea67dcc BS |
196 | return false; |
197 | ||
198 | while (++map < map_end) { | |
199 | if (*map != count) | |
200 | return false; | |
201 | } | |
202 | ||
203 | *has_cache = !!(count & SWAP_HAS_CACHE); | |
204 | return true; | |
205 | } | |
206 | ||
a62fb92a RR |
207 | /* |
208 | * returns number of pages in the folio that backs the swap entry. If positive, | |
209 | * the folio was reclaimed. If negative, the folio was not reclaimed. If 0, no | |
210 | * folio was associated with the swap entry. | |
211 | */ | |
bcd49e86 HY |
212 | static int __try_to_reclaim_swap(struct swap_info_struct *si, |
213 | unsigned long offset, unsigned long flags) | |
c9e44410 | 214 | { |
efa90a98 | 215 | swp_entry_t entry = swp_entry(si->type, offset); |
862590ac KS |
216 | struct address_space *address_space = swap_address_space(entry); |
217 | struct swap_cluster_info *ci; | |
2c3f6194 | 218 | struct folio *folio; |
862590ac KS |
219 | int ret, nr_pages; |
220 | bool need_reclaim; | |
c9e44410 | 221 | |
fae85955 | 222 | again: |
862590ac | 223 | folio = filemap_get_folio(address_space, swap_cache_index(entry)); |
66dabbb6 | 224 | if (IS_ERR(folio)) |
c9e44410 | 225 | return 0; |
862590ac | 226 | |
862590ac KS |
227 | nr_pages = folio_nr_pages(folio); |
228 | ret = -nr_pages; | |
229 | ||
c9e44410 | 230 | /* |
bcd49e86 | 231 | * When this function is called from scan_swap_map_slots() and it's |
2c3f6194 | 232 | * called by vmscan.c at reclaiming folios. So we hold a folio lock |
bcd49e86 | 233 | * here. We have to use trylock for avoiding deadlock. This is a special |
2c3f6194 | 234 | * case and you should use folio_free_swap() with explicit folio_lock() |
c9e44410 KH |
235 | * in usual operations. |
236 | */ | |
862590ac KS |
237 | if (!folio_trylock(folio)) |
238 | goto out; | |
239 | ||
fae85955 KS |
240 | /* |
241 | * Offset could point to the middle of a large folio, or folio | |
242 | * may no longer point to the expected offset before it's locked. | |
243 | */ | |
818f916e | 244 | entry = folio->swap; |
fae85955 KS |
245 | if (offset < swp_offset(entry) || offset >= swp_offset(entry) + nr_pages) { |
246 | folio_unlock(folio); | |
247 | folio_put(folio); | |
248 | goto again; | |
249 | } | |
818f916e JP |
250 | offset = swp_offset(entry); |
251 | ||
862590ac KS |
252 | need_reclaim = ((flags & TTRS_ANYWAY) || |
253 | ((flags & TTRS_UNMAPPED) && !folio_mapped(folio)) || | |
254 | ((flags & TTRS_FULL) && mem_cgroup_swap_full(folio))); | |
255 | if (!need_reclaim || !folio_swapcache_freeable(folio)) | |
256 | goto out_unlock; | |
257 | ||
258 | /* | |
259 | * It's safe to delete the folio from swap cache only if the folio's | |
260 | * swap_map is HAS_CACHE only, which means the slots have no page table | |
261 | * reference or pending writeback, and can't be allocated to others. | |
262 | */ | |
0b310d9c | 263 | ci = lock_cluster(si, offset); |
b4735d94 | 264 | need_reclaim = swap_only_has_cache(si, offset, nr_pages); |
0b310d9c | 265 | unlock_cluster(ci); |
862590ac KS |
266 | if (!need_reclaim) |
267 | goto out_unlock; | |
268 | ||
3123fb0a | 269 | delete_from_swap_cache(folio); |
862590ac | 270 | folio_set_dirty(folio); |
862590ac KS |
271 | ret = nr_pages; |
272 | out_unlock: | |
273 | folio_unlock(folio); | |
274 | out: | |
2c3f6194 | 275 | folio_put(folio); |
c9e44410 KH |
276 | return ret; |
277 | } | |
355cfa73 | 278 | |
4efaceb1 AL |
279 | static inline struct swap_extent *first_se(struct swap_info_struct *sis) |
280 | { | |
281 | struct rb_node *rb = rb_first(&sis->swap_extent_root); | |
282 | return rb_entry(rb, struct swap_extent, rb_node); | |
283 | } | |
284 | ||
285 | static inline struct swap_extent *next_se(struct swap_extent *se) | |
286 | { | |
287 | struct rb_node *rb = rb_next(&se->rb_node); | |
288 | return rb ? rb_entry(rb, struct swap_extent, rb_node) : NULL; | |
289 | } | |
290 | ||
6a6ba831 HD |
291 | /* |
292 | * swapon tell device that all the old swap contents can be discarded, | |
293 | * to allow the swap device to optimize its wear-levelling. | |
294 | */ | |
295 | static int discard_swap(struct swap_info_struct *si) | |
296 | { | |
297 | struct swap_extent *se; | |
9625a5f2 HD |
298 | sector_t start_block; |
299 | sector_t nr_blocks; | |
6a6ba831 HD |
300 | int err = 0; |
301 | ||
9625a5f2 | 302 | /* Do not discard the swap header page! */ |
4efaceb1 | 303 | se = first_se(si); |
9625a5f2 HD |
304 | start_block = (se->start_block + 1) << (PAGE_SHIFT - 9); |
305 | nr_blocks = ((sector_t)se->nr_pages - 1) << (PAGE_SHIFT - 9); | |
306 | if (nr_blocks) { | |
307 | err = blkdev_issue_discard(si->bdev, start_block, | |
44abff2c | 308 | nr_blocks, GFP_KERNEL); |
9625a5f2 HD |
309 | if (err) |
310 | return err; | |
311 | cond_resched(); | |
312 | } | |
6a6ba831 | 313 | |
4efaceb1 | 314 | for (se = next_se(se); se; se = next_se(se)) { |
9625a5f2 HD |
315 | start_block = se->start_block << (PAGE_SHIFT - 9); |
316 | nr_blocks = (sector_t)se->nr_pages << (PAGE_SHIFT - 9); | |
6a6ba831 HD |
317 | |
318 | err = blkdev_issue_discard(si->bdev, start_block, | |
44abff2c | 319 | nr_blocks, GFP_KERNEL); |
6a6ba831 HD |
320 | if (err) |
321 | break; | |
322 | ||
323 | cond_resched(); | |
324 | } | |
325 | return err; /* That will often be -EOPNOTSUPP */ | |
326 | } | |
327 | ||
4efaceb1 AL |
328 | static struct swap_extent * |
329 | offset_to_swap_extent(struct swap_info_struct *sis, unsigned long offset) | |
330 | { | |
331 | struct swap_extent *se; | |
332 | struct rb_node *rb; | |
333 | ||
334 | rb = sis->swap_extent_root.rb_node; | |
335 | while (rb) { | |
336 | se = rb_entry(rb, struct swap_extent, rb_node); | |
337 | if (offset < se->start_page) | |
338 | rb = rb->rb_left; | |
339 | else if (offset >= se->start_page + se->nr_pages) | |
340 | rb = rb->rb_right; | |
341 | else | |
342 | return se; | |
343 | } | |
344 | /* It *must* be present */ | |
345 | BUG(); | |
346 | } | |
347 | ||
3a61e6f6 | 348 | sector_t swap_folio_sector(struct folio *folio) |
caf6912f | 349 | { |
3a61e6f6 | 350 | struct swap_info_struct *sis = swp_swap_info(folio->swap); |
caf6912f JA |
351 | struct swap_extent *se; |
352 | sector_t sector; | |
353 | pgoff_t offset; | |
354 | ||
3a61e6f6 | 355 | offset = swp_offset(folio->swap); |
caf6912f JA |
356 | se = offset_to_swap_extent(sis, offset); |
357 | sector = se->start_block + (offset - se->start_page); | |
358 | return sector << (PAGE_SHIFT - 9); | |
359 | } | |
360 | ||
7992fde7 HD |
361 | /* |
362 | * swap allocation tell device that a cluster of swap can now be discarded, | |
363 | * to allow the swap device to optimize its wear-levelling. | |
364 | */ | |
365 | static void discard_swap_cluster(struct swap_info_struct *si, | |
366 | pgoff_t start_page, pgoff_t nr_pages) | |
367 | { | |
4efaceb1 | 368 | struct swap_extent *se = offset_to_swap_extent(si, start_page); |
7992fde7 HD |
369 | |
370 | while (nr_pages) { | |
4efaceb1 AL |
371 | pgoff_t offset = start_page - se->start_page; |
372 | sector_t start_block = se->start_block + offset; | |
373 | sector_t nr_blocks = se->nr_pages - offset; | |
374 | ||
375 | if (nr_blocks > nr_pages) | |
376 | nr_blocks = nr_pages; | |
377 | start_page += nr_blocks; | |
378 | nr_pages -= nr_blocks; | |
379 | ||
380 | start_block <<= PAGE_SHIFT - 9; | |
381 | nr_blocks <<= PAGE_SHIFT - 9; | |
382 | if (blkdev_issue_discard(si->bdev, start_block, | |
44abff2c | 383 | nr_blocks, GFP_NOIO)) |
4efaceb1 | 384 | break; |
7992fde7 | 385 | |
4efaceb1 | 386 | se = next_se(se); |
7992fde7 HD |
387 | } |
388 | } | |
389 | ||
38d8b4e6 HY |
390 | #ifdef CONFIG_THP_SWAP |
391 | #define SWAPFILE_CLUSTER HPAGE_PMD_NR | |
a448f2d0 | 392 | |
9faaa0f8 | 393 | #define swap_entry_order(order) (order) |
38d8b4e6 | 394 | #else |
048c27fd | 395 | #define SWAPFILE_CLUSTER 256 |
a448f2d0 HY |
396 | |
397 | /* | |
9faaa0f8 | 398 | * Define swap_entry_order() as constant to let compiler to optimize |
a448f2d0 HY |
399 | * out some code if !CONFIG_THP_SWAP |
400 | */ | |
9faaa0f8 | 401 | #define swap_entry_order(order) 0 |
38d8b4e6 | 402 | #endif |
048c27fd HD |
403 | #define LATENCY_LIMIT 256 |
404 | ||
3b644773 KS |
405 | static inline bool cluster_is_empty(struct swap_cluster_info *info) |
406 | { | |
407 | return info->count == 0; | |
408 | } | |
409 | ||
410 | static inline bool cluster_is_discard(struct swap_cluster_info *info) | |
411 | { | |
412 | return info->flags == CLUSTER_FLAG_DISCARD; | |
413 | } | |
414 | ||
415 | static inline bool cluster_is_usable(struct swap_cluster_info *ci, int order) | |
2a8f9449 | 416 | { |
3b644773 KS |
417 | if (unlikely(ci->flags > CLUSTER_FLAG_USABLE)) |
418 | return false; | |
419 | if (!order) | |
420 | return true; | |
421 | return cluster_is_empty(ci) || order == ci->order; | |
2a8f9449 SL |
422 | } |
423 | ||
73ed0baa CL |
424 | static inline unsigned int cluster_index(struct swap_info_struct *si, |
425 | struct swap_cluster_info *ci) | |
2a8f9449 | 426 | { |
73ed0baa | 427 | return ci - si->cluster_info; |
2a8f9449 SL |
428 | } |
429 | ||
3f641cf9 KS |
430 | static inline struct swap_cluster_info *offset_to_cluster(struct swap_info_struct *si, |
431 | unsigned long offset) | |
432 | { | |
433 | return &si->cluster_info[offset / SWAPFILE_CLUSTER]; | |
434 | } | |
435 | ||
5f843a9a CL |
436 | static inline unsigned int cluster_offset(struct swap_info_struct *si, |
437 | struct swap_cluster_info *ci) | |
438 | { | |
439 | return cluster_index(si, ci) * SWAPFILE_CLUSTER; | |
440 | } | |
441 | ||
235b6217 HY |
442 | static inline struct swap_cluster_info *lock_cluster(struct swap_info_struct *si, |
443 | unsigned long offset) | |
444 | { | |
445 | struct swap_cluster_info *ci; | |
446 | ||
3f641cf9 | 447 | ci = offset_to_cluster(si, offset); |
0b310d9c | 448 | spin_lock(&ci->lock); |
235b6217 HY |
449 | |
450 | return ci; | |
451 | } | |
452 | ||
0b310d9c | 453 | static inline void unlock_cluster(struct swap_cluster_info *ci) |
235b6217 | 454 | { |
0b310d9c | 455 | spin_unlock(&ci->lock); |
235b6217 HY |
456 | } |
457 | ||
3494d184 KS |
458 | static void move_cluster(struct swap_info_struct *si, |
459 | struct swap_cluster_info *ci, struct list_head *list, | |
460 | enum swap_cluster_flags new_flags) | |
461 | { | |
462 | VM_WARN_ON(ci->flags == new_flags); | |
463 | ||
464 | BUILD_BUG_ON(1 << sizeof(ci->flags) * BITS_PER_BYTE < CLUSTER_FLAG_MAX); | |
3b644773 | 465 | lockdep_assert_held(&ci->lock); |
3494d184 | 466 | |
3b644773 KS |
467 | spin_lock(&si->lock); |
468 | if (ci->flags == CLUSTER_FLAG_NONE) | |
3494d184 | 469 | list_add_tail(&ci->list, list); |
3b644773 | 470 | else |
3494d184 | 471 | list_move_tail(&ci->list, list); |
3b644773 KS |
472 | spin_unlock(&si->lock); |
473 | ||
474 | if (ci->flags == CLUSTER_FLAG_FRAG) | |
475 | atomic_long_dec(&si->frag_cluster_nr[ci->order]); | |
476 | else if (new_flags == CLUSTER_FLAG_FRAG) | |
477 | atomic_long_inc(&si->frag_cluster_nr[ci->order]); | |
3494d184 | 478 | ci->flags = new_flags; |
3494d184 KS |
479 | } |
480 | ||
815c2c54 SL |
481 | /* Add a cluster to discard list and schedule it to do discard */ |
482 | static void swap_cluster_schedule_discard(struct swap_info_struct *si, | |
73ed0baa | 483 | struct swap_cluster_info *ci) |
815c2c54 | 484 | { |
3494d184 KS |
485 | VM_BUG_ON(ci->flags == CLUSTER_FLAG_FREE); |
486 | move_cluster(si, ci, &si->discard_clusters, CLUSTER_FLAG_DISCARD); | |
815c2c54 SL |
487 | schedule_work(&si->discard_work); |
488 | } | |
489 | ||
73ed0baa | 490 | static void __free_cluster(struct swap_info_struct *si, struct swap_cluster_info *ci) |
38d8b4e6 | 491 | { |
5f843a9a | 492 | lockdep_assert_held(&ci->lock); |
3494d184 | 493 | move_cluster(si, ci, &si->free_clusters, CLUSTER_FLAG_FREE); |
5f843a9a | 494 | ci->order = 0; |
38d8b4e6 HY |
495 | } |
496 | ||
3b644773 KS |
497 | /* |
498 | * Isolate and lock the first cluster that is not contented on a list, | |
499 | * clean its flag before taken off-list. Cluster flag must be in sync | |
500 | * with list status, so cluster updaters can always know the cluster | |
501 | * list status without touching si lock. | |
502 | * | |
503 | * Note it's possible that all clusters on a list are contented so | |
504 | * this returns NULL for an non-empty list. | |
505 | */ | |
506 | static struct swap_cluster_info *isolate_lock_cluster( | |
507 | struct swap_info_struct *si, struct list_head *list) | |
508 | { | |
509 | struct swap_cluster_info *ci, *ret = NULL; | |
510 | ||
511 | spin_lock(&si->lock); | |
512 | ||
513 | if (unlikely(!(si->flags & SWP_WRITEOK))) | |
514 | goto out; | |
515 | ||
516 | list_for_each_entry(ci, list, list) { | |
517 | if (!spin_trylock(&ci->lock)) | |
518 | continue; | |
519 | ||
520 | /* We may only isolate and clear flags of following lists */ | |
521 | VM_BUG_ON(!ci->flags); | |
522 | VM_BUG_ON(ci->flags > CLUSTER_FLAG_USABLE && | |
523 | ci->flags != CLUSTER_FLAG_FULL); | |
524 | ||
525 | list_del(&ci->list); | |
526 | ci->flags = CLUSTER_FLAG_NONE; | |
527 | ret = ci; | |
528 | break; | |
529 | } | |
530 | out: | |
531 | spin_unlock(&si->lock); | |
532 | ||
533 | return ret; | |
534 | } | |
535 | ||
815c2c54 SL |
536 | /* |
537 | * Doing discard actually. After a cluster discard is finished, the cluster | |
3b644773 KS |
538 | * will be added to free cluster list. Discard cluster is a bit special as |
539 | * they don't participate in allocation or reclaim, so clusters marked as | |
540 | * CLUSTER_FLAG_DISCARD must remain off-list or on discard list. | |
541 | */ | |
542 | static bool swap_do_scheduled_discard(struct swap_info_struct *si) | |
815c2c54 | 543 | { |
73ed0baa | 544 | struct swap_cluster_info *ci; |
3b644773 | 545 | bool ret = false; |
815c2c54 SL |
546 | unsigned int idx; |
547 | ||
3b644773 | 548 | spin_lock(&si->lock); |
73ed0baa CL |
549 | while (!list_empty(&si->discard_clusters)) { |
550 | ci = list_first_entry(&si->discard_clusters, struct swap_cluster_info, list); | |
3b644773 KS |
551 | /* |
552 | * Delete the cluster from list to prepare for discard, but keep | |
1b7e9002 | 553 | * the CLUSTER_FLAG_DISCARD flag, percpu_swap_cluster could be |
3b644773 KS |
554 | * pointing to it, or ran into by relocate_cluster. |
555 | */ | |
73ed0baa CL |
556 | list_del(&ci->list); |
557 | idx = cluster_index(si, ci); | |
815c2c54 | 558 | spin_unlock(&si->lock); |
815c2c54 SL |
559 | discard_swap_cluster(si, idx * SWAPFILE_CLUSTER, |
560 | SWAPFILE_CLUSTER); | |
561 | ||
73ed0baa | 562 | spin_lock(&ci->lock); |
3b644773 KS |
563 | /* |
564 | * Discard is done, clear its flags as it's off-list, then | |
565 | * return the cluster to allocation list. | |
566 | */ | |
567 | ci->flags = CLUSTER_FLAG_NONE; | |
3b644773 | 568 | __free_cluster(si, ci); |
73ed0baa | 569 | spin_unlock(&ci->lock); |
3b644773 KS |
570 | ret = true; |
571 | spin_lock(&si->lock); | |
815c2c54 | 572 | } |
3b644773 KS |
573 | spin_unlock(&si->lock); |
574 | return ret; | |
815c2c54 SL |
575 | } |
576 | ||
577 | static void swap_discard_work(struct work_struct *work) | |
578 | { | |
579 | struct swap_info_struct *si; | |
580 | ||
581 | si = container_of(work, struct swap_info_struct, discard_work); | |
582 | ||
815c2c54 | 583 | swap_do_scheduled_discard(si); |
815c2c54 SL |
584 | } |
585 | ||
63d8620e ML |
586 | static void swap_users_ref_free(struct percpu_ref *ref) |
587 | { | |
588 | struct swap_info_struct *si; | |
589 | ||
590 | si = container_of(ref, struct swap_info_struct, users); | |
591 | complete(&si->comp); | |
592 | } | |
593 | ||
3b644773 KS |
594 | /* |
595 | * Must be called after freeing if ci->count == 0, moves the cluster to free | |
596 | * or discard list. | |
597 | */ | |
73ed0baa | 598 | static void free_cluster(struct swap_info_struct *si, struct swap_cluster_info *ci) |
38d8b4e6 | 599 | { |
73ed0baa | 600 | VM_BUG_ON(ci->count != 0); |
3b644773 | 601 | VM_BUG_ON(ci->flags == CLUSTER_FLAG_FREE); |
5f843a9a | 602 | lockdep_assert_held(&ci->lock); |
661383c6 | 603 | |
38d8b4e6 HY |
604 | /* |
605 | * If the swap is discardable, prepare discard the cluster | |
606 | * instead of free it immediately. The cluster will be freed | |
607 | * after discard. | |
608 | */ | |
609 | if ((si->flags & (SWP_WRITEOK | SWP_PAGE_DISCARD)) == | |
610 | (SWP_WRITEOK | SWP_PAGE_DISCARD)) { | |
73ed0baa | 611 | swap_cluster_schedule_discard(si, ci); |
38d8b4e6 HY |
612 | return; |
613 | } | |
614 | ||
73ed0baa | 615 | __free_cluster(si, ci); |
38d8b4e6 HY |
616 | } |
617 | ||
3b644773 KS |
618 | /* |
619 | * Must be called after freeing if ci->count != 0, moves the cluster to | |
620 | * nonfull list. | |
621 | */ | |
622 | static void partial_free_cluster(struct swap_info_struct *si, | |
623 | struct swap_cluster_info *ci) | |
624 | { | |
625 | VM_BUG_ON(!ci->count || ci->count == SWAPFILE_CLUSTER); | |
626 | lockdep_assert_held(&ci->lock); | |
627 | ||
628 | if (ci->flags != CLUSTER_FLAG_NONFULL) | |
629 | move_cluster(si, ci, &si->nonfull_clusters[ci->order], | |
630 | CLUSTER_FLAG_NONFULL); | |
631 | } | |
632 | ||
633 | /* | |
634 | * Must be called after allocation, moves the cluster to full or frag list. | |
635 | * Note: allocation doesn't acquire si lock, and may drop the ci lock for | |
636 | * reclaim, so the cluster could be any where when called. | |
637 | */ | |
638 | static void relocate_cluster(struct swap_info_struct *si, | |
639 | struct swap_cluster_info *ci) | |
640 | { | |
641 | lockdep_assert_held(&ci->lock); | |
642 | ||
643 | /* Discard cluster must remain off-list or on discard list */ | |
644 | if (cluster_is_discard(ci)) | |
645 | return; | |
646 | ||
647 | if (!ci->count) { | |
57d910cf KS |
648 | if (ci->flags != CLUSTER_FLAG_FREE) |
649 | free_cluster(si, ci); | |
3b644773 KS |
650 | } else if (ci->count != SWAPFILE_CLUSTER) { |
651 | if (ci->flags != CLUSTER_FLAG_FRAG) | |
652 | move_cluster(si, ci, &si->frag_clusters[ci->order], | |
653 | CLUSTER_FLAG_FRAG); | |
654 | } else { | |
655 | if (ci->flags != CLUSTER_FLAG_FULL) | |
656 | move_cluster(si, ci, &si->full_clusters, | |
657 | CLUSTER_FLAG_FULL); | |
658 | } | |
659 | } | |
660 | ||
2a8f9449 | 661 | /* |
3b2561b5 KS |
662 | * The cluster corresponding to page_nr will be used. The cluster will not be |
663 | * added to free cluster list and its usage counter will be increased by 1. | |
664 | * Only used for initialization. | |
2a8f9449 | 665 | */ |
b85508d7 | 666 | static void inc_cluster_info_page(struct swap_info_struct *si, |
3b2561b5 | 667 | struct swap_cluster_info *cluster_info, unsigned long page_nr) |
2a8f9449 SL |
668 | { |
669 | unsigned long idx = page_nr / SWAPFILE_CLUSTER; | |
3b2561b5 | 670 | struct swap_cluster_info *ci; |
2a8f9449 | 671 | |
3b2561b5 KS |
672 | ci = cluster_info + idx; |
673 | ci->count++; | |
845982eb | 674 | |
3b2561b5 KS |
675 | VM_BUG_ON(ci->count > SWAPFILE_CLUSTER); |
676 | VM_BUG_ON(ci->flags); | |
2a8f9449 SL |
677 | } |
678 | ||
661383c6 KS |
679 | static bool cluster_reclaim_range(struct swap_info_struct *si, |
680 | struct swap_cluster_info *ci, | |
681 | unsigned long start, unsigned long end) | |
2a8f9449 | 682 | { |
661383c6 | 683 | unsigned char *map = si->swap_map; |
d563ced6 KS |
684 | unsigned long offset = start; |
685 | int nr_reclaim; | |
661383c6 KS |
686 | |
687 | spin_unlock(&ci->lock); | |
d563ced6 | 688 | do { |
661383c6 KS |
689 | switch (READ_ONCE(map[offset])) { |
690 | case 0: | |
d563ced6 KS |
691 | offset++; |
692 | break; | |
661383c6 | 693 | case SWAP_HAS_CACHE: |
3123fb0a | 694 | nr_reclaim = __try_to_reclaim_swap(si, offset, TTRS_ANYWAY); |
d563ced6 KS |
695 | if (nr_reclaim > 0) |
696 | offset += nr_reclaim; | |
697 | else | |
698 | goto out; | |
699 | break; | |
661383c6 KS |
700 | default: |
701 | goto out; | |
702 | } | |
d563ced6 | 703 | } while (offset < end); |
661383c6 | 704 | out: |
661383c6 | 705 | spin_lock(&ci->lock); |
661383c6 KS |
706 | /* |
707 | * Recheck the range no matter reclaim succeeded or not, the slot | |
708 | * could have been be freed while we are not holding the lock. | |
709 | */ | |
710 | for (offset = start; offset < end; offset++) | |
711 | if (READ_ONCE(map[offset])) | |
5f843a9a | 712 | return false; |
ebc2a1a6 | 713 | |
845982eb RR |
714 | return true; |
715 | } | |
716 | ||
661383c6 KS |
717 | static bool cluster_scan_range(struct swap_info_struct *si, |
718 | struct swap_cluster_info *ci, | |
3b644773 KS |
719 | unsigned long start, unsigned int nr_pages, |
720 | bool *need_reclaim) | |
661383c6 KS |
721 | { |
722 | unsigned long offset, end = start + nr_pages; | |
723 | unsigned char *map = si->swap_map; | |
5f843a9a | 724 | |
9b9cba72 BH |
725 | if (cluster_is_empty(ci)) |
726 | return true; | |
727 | ||
661383c6 KS |
728 | for (offset = start; offset < end; offset++) { |
729 | switch (READ_ONCE(map[offset])) { | |
730 | case 0: | |
731 | continue; | |
732 | case SWAP_HAS_CACHE: | |
733 | if (!vm_swap_full()) | |
734 | return false; | |
3b644773 | 735 | *need_reclaim = true; |
661383c6 KS |
736 | continue; |
737 | default: | |
738 | return false; | |
739 | } | |
740 | } | |
741 | ||
661383c6 KS |
742 | return true; |
743 | } | |
744 | ||
0ec8bc9e | 745 | static bool cluster_alloc_range(struct swap_info_struct *si, struct swap_cluster_info *ci, |
661383c6 KS |
746 | unsigned int start, unsigned char usage, |
747 | unsigned int order) | |
845982eb | 748 | { |
5f843a9a | 749 | unsigned int nr_pages = 1 << order; |
845982eb | 750 | |
9a0ddeb7 KS |
751 | lockdep_assert_held(&ci->lock); |
752 | ||
0ec8bc9e KS |
753 | if (!(si->flags & SWP_WRITEOK)) |
754 | return false; | |
755 | ||
3b644773 KS |
756 | /* |
757 | * The first allocation in a cluster makes the | |
758 | * cluster exclusive to this order | |
759 | */ | |
760 | if (cluster_is_empty(ci)) | |
5f843a9a | 761 | ci->order = order; |
845982eb | 762 | |
5f843a9a | 763 | memset(si->swap_map + start, usage, nr_pages); |
27701521 | 764 | swap_range_alloc(si, nr_pages); |
5f843a9a CL |
765 | ci->count += nr_pages; |
766 | ||
0ec8bc9e | 767 | return true; |
5f843a9a CL |
768 | } |
769 | ||
e3ae2dec KS |
770 | /* Try use a new cluster for current CPU and allocate from it. */ |
771 | static unsigned int alloc_swap_scan_cluster(struct swap_info_struct *si, | |
772 | struct swap_cluster_info *ci, | |
773 | unsigned long offset, | |
774 | unsigned int order, | |
5f843a9a CL |
775 | unsigned char usage) |
776 | { | |
e3ae2dec KS |
777 | unsigned int next = SWAP_ENTRY_INVALID, found = SWAP_ENTRY_INVALID; |
778 | unsigned long start = ALIGN_DOWN(offset, SWAPFILE_CLUSTER); | |
5f843a9a CL |
779 | unsigned long end = min(start + SWAPFILE_CLUSTER, si->max); |
780 | unsigned int nr_pages = 1 << order; | |
3b644773 | 781 | bool need_reclaim, ret; |
5f843a9a | 782 | |
3b644773 | 783 | lockdep_assert_held(&ci->lock); |
5f843a9a | 784 | |
e3ae2dec | 785 | if (end < nr_pages || ci->count + nr_pages > SWAPFILE_CLUSTER) |
3b644773 | 786 | goto out; |
5f843a9a | 787 | |
3b644773 KS |
788 | for (end -= nr_pages; offset <= end; offset += nr_pages) { |
789 | need_reclaim = false; | |
790 | if (!cluster_scan_range(si, ci, offset, nr_pages, &need_reclaim)) | |
791 | continue; | |
792 | if (need_reclaim) { | |
498c48c6 | 793 | ret = cluster_reclaim_range(si, ci, offset, offset + nr_pages); |
3b644773 KS |
794 | /* |
795 | * Reclaim drops ci->lock and cluster could be used | |
796 | * by another order. Not checking flag as off-list | |
797 | * cluster has no flag set, and change of list | |
798 | * won't cause fragmentation. | |
799 | */ | |
e3ae2dec | 800 | if (!cluster_is_usable(ci, order)) |
3b644773 | 801 | goto out; |
3b644773 KS |
802 | if (cluster_is_empty(ci)) |
803 | offset = start; | |
804 | /* Reclaim failed but cluster is usable, try next */ | |
805 | if (!ret) | |
806 | continue; | |
807 | } | |
e3ae2dec KS |
808 | if (!cluster_alloc_range(si, ci, offset, usage, order)) |
809 | break; | |
810 | found = offset; | |
5f843a9a | 811 | offset += nr_pages; |
e3ae2dec KS |
812 | if (ci->count < SWAPFILE_CLUSTER && offset <= end) |
813 | next = offset; | |
3b644773 | 814 | break; |
5f843a9a | 815 | } |
3b644773 KS |
816 | out: |
817 | relocate_cluster(si, ci); | |
5f843a9a | 818 | unlock_cluster(ci); |
1b7e9002 KS |
819 | if (si->flags & SWP_SOLIDSTATE) { |
820 | this_cpu_write(percpu_swap_cluster.offset[order], next); | |
821 | this_cpu_write(percpu_swap_cluster.si[order], si); | |
822 | } else { | |
bae8a4ef | 823 | si->global_cluster->next[order] = next; |
1b7e9002 | 824 | } |
e3ae2dec | 825 | return found; |
ebc2a1a6 SL |
826 | } |
827 | ||
5168a68e | 828 | static void swap_reclaim_full_clusters(struct swap_info_struct *si, bool force) |
2cacbdfd KS |
829 | { |
830 | long to_scan = 1; | |
831 | unsigned long offset, end; | |
832 | struct swap_cluster_info *ci; | |
833 | unsigned char *map = si->swap_map; | |
5168a68e | 834 | int nr_reclaim; |
2cacbdfd | 835 | |
5168a68e | 836 | if (force) |
b228386c | 837 | to_scan = swap_usage_in_pages(si) / SWAPFILE_CLUSTER; |
2cacbdfd | 838 | |
3b644773 | 839 | while ((ci = isolate_lock_cluster(si, &si->full_clusters))) { |
2cacbdfd KS |
840 | offset = cluster_offset(si, ci); |
841 | end = min(si->max, offset + SWAPFILE_CLUSTER); | |
842 | to_scan--; | |
843 | ||
844 | while (offset < end) { | |
845 | if (READ_ONCE(map[offset]) == SWAP_HAS_CACHE) { | |
3b644773 | 846 | spin_unlock(&ci->lock); |
2cacbdfd | 847 | nr_reclaim = __try_to_reclaim_swap(si, offset, |
3123fb0a | 848 | TTRS_ANYWAY); |
3b644773 | 849 | spin_lock(&ci->lock); |
5168a68e KS |
850 | if (nr_reclaim) { |
851 | offset += abs(nr_reclaim); | |
2cacbdfd KS |
852 | continue; |
853 | } | |
854 | } | |
855 | offset++; | |
856 | } | |
5168a68e | 857 | |
51f271c1 KS |
858 | /* in case no swap cache is reclaimed */ |
859 | if (ci->flags == CLUSTER_FLAG_NONE) | |
860 | relocate_cluster(si, ci); | |
861 | ||
3b644773 | 862 | unlock_cluster(ci); |
5168a68e | 863 | if (to_scan <= 0) |
2cacbdfd KS |
864 | break; |
865 | } | |
866 | } | |
867 | ||
5168a68e KS |
868 | static void swap_reclaim_work(struct work_struct *work) |
869 | { | |
870 | struct swap_info_struct *si; | |
871 | ||
872 | si = container_of(work, struct swap_info_struct, reclaim_work); | |
873 | ||
5168a68e | 874 | swap_reclaim_full_clusters(si, true); |
5168a68e KS |
875 | } |
876 | ||
ebc2a1a6 | 877 | /* |
1b7e9002 KS |
878 | * Try to allocate swap entries with specified order and try set a new |
879 | * cluster for current CPU too. | |
ebc2a1a6 | 880 | */ |
5f843a9a CL |
881 | static unsigned long cluster_alloc_swap_entry(struct swap_info_struct *si, int order, |
882 | unsigned char usage) | |
ebc2a1a6 | 883 | { |
661383c6 | 884 | struct swap_cluster_info *ci; |
1b7e9002 | 885 | unsigned int offset = SWAP_ENTRY_INVALID, found = SWAP_ENTRY_INVALID; |
ebc2a1a6 | 886 | |
0ff67f99 KS |
887 | /* |
888 | * Swapfile is not block device so unable | |
889 | * to allocate large entries. | |
890 | */ | |
891 | if (order && !(si->flags & SWP_BLKDEV)) | |
892 | return 0; | |
893 | ||
894 | if (!(si->flags & SWP_SOLIDSTATE)) { | |
bae8a4ef KS |
895 | /* Serialize HDD SWAP allocation for each device. */ |
896 | spin_lock(&si->global_cluster_lock); | |
897 | offset = si->global_cluster->next[order]; | |
0ff67f99 KS |
898 | if (offset == SWAP_ENTRY_INVALID) |
899 | goto new_cluster; | |
bae8a4ef | 900 | |
3b644773 KS |
901 | ci = lock_cluster(si, offset); |
902 | /* Cluster could have been used by another order */ | |
903 | if (cluster_is_usable(ci, order)) { | |
904 | if (cluster_is_empty(ci)) | |
905 | offset = cluster_offset(si, ci); | |
e3ae2dec KS |
906 | found = alloc_swap_scan_cluster(si, ci, offset, |
907 | order, usage); | |
3b644773 KS |
908 | } else { |
909 | unlock_cluster(ci); | |
910 | } | |
5f843a9a CL |
911 | if (found) |
912 | goto done; | |
ebc2a1a6 SL |
913 | } |
914 | ||
3b644773 KS |
915 | new_cluster: |
916 | ci = isolate_lock_cluster(si, &si->free_clusters); | |
917 | if (ci) { | |
e3ae2dec KS |
918 | found = alloc_swap_scan_cluster(si, ci, cluster_offset(si, ci), |
919 | order, usage); | |
3b644773 KS |
920 | if (found) |
921 | goto done; | |
5f843a9a CL |
922 | } |
923 | ||
5168a68e KS |
924 | /* Try reclaim from full clusters if free clusters list is drained */ |
925 | if (vm_swap_full()) | |
926 | swap_reclaim_full_clusters(si, false); | |
927 | ||
5f843a9a | 928 | if (order < PMD_ORDER) { |
3b644773 | 929 | unsigned int frags = 0, frags_existing; |
661383c6 | 930 | |
3b644773 | 931 | while ((ci = isolate_lock_cluster(si, &si->nonfull_clusters[order]))) { |
e3ae2dec KS |
932 | found = alloc_swap_scan_cluster(si, ci, cluster_offset(si, ci), |
933 | order, usage); | |
5f843a9a | 934 | if (found) |
d563ced6 | 935 | goto done; |
3b644773 KS |
936 | /* Clusters failed to allocate are moved to frag_clusters */ |
937 | frags++; | |
7b9e2de1 | 938 | } |
477cb7ba | 939 | |
3b644773 KS |
940 | frags_existing = atomic_long_read(&si->frag_cluster_nr[order]); |
941 | while (frags < frags_existing && | |
942 | (ci = isolate_lock_cluster(si, &si->frag_clusters[order]))) { | |
943 | atomic_long_dec(&si->frag_cluster_nr[order]); | |
661383c6 | 944 | /* |
3b644773 KS |
945 | * Rotate the frag list to iterate, they were all |
946 | * failing high order allocation or moved here due to | |
947 | * per-CPU usage, but they could contain newly released | |
948 | * reclaimable (eg. lazy-freed swap cache) slots. | |
661383c6 | 949 | */ |
e3ae2dec KS |
950 | found = alloc_swap_scan_cluster(si, ci, cluster_offset(si, ci), |
951 | order, usage); | |
d563ced6 KS |
952 | if (found) |
953 | goto done; | |
3b644773 | 954 | frags++; |
477cb7ba | 955 | } |
ebc2a1a6 | 956 | } |
5f843a9a | 957 | |
3b644773 KS |
958 | /* |
959 | * We don't have free cluster but have some clusters in | |
960 | * discarding, do discard now and reclaim them, then | |
961 | * reread cluster_next_cpu since we dropped si->lock | |
962 | */ | |
963 | if ((si->flags & SWP_PAGE_DISCARD) && swap_do_scheduled_discard(si)) | |
ebc2a1a6 | 964 | goto new_cluster; |
5f843a9a CL |
965 | |
966 | if (order) | |
967 | goto done; | |
968 | ||
477cb7ba | 969 | /* Order 0 stealing from higher order */ |
5f843a9a | 970 | for (int o = 1; o < SWAP_NR_ORDERS; o++) { |
661383c6 KS |
971 | /* |
972 | * Clusters here have at least one usable slots and can't fail order 0 | |
973 | * allocation, but reclaim may drop si->lock and race with another user. | |
974 | */ | |
3b644773 KS |
975 | while ((ci = isolate_lock_cluster(si, &si->frag_clusters[o]))) { |
976 | atomic_long_dec(&si->frag_cluster_nr[o]); | |
e3ae2dec KS |
977 | found = alloc_swap_scan_cluster(si, ci, cluster_offset(si, ci), |
978 | 0, usage); | |
661383c6 KS |
979 | if (found) |
980 | goto done; | |
477cb7ba KS |
981 | } |
982 | ||
3b644773 | 983 | while ((ci = isolate_lock_cluster(si, &si->nonfull_clusters[o]))) { |
e3ae2dec KS |
984 | found = alloc_swap_scan_cluster(si, ci, cluster_offset(si, ci), |
985 | 0, usage); | |
661383c6 KS |
986 | if (found) |
987 | goto done; | |
5f843a9a CL |
988 | } |
989 | } | |
5f843a9a | 990 | done: |
1b7e9002 | 991 | if (!(si->flags & SWP_SOLIDSTATE)) |
bae8a4ef | 992 | spin_unlock(&si->global_cluster_lock); |
5f843a9a | 993 | return found; |
2a8f9449 SL |
994 | } |
995 | ||
b228386c KS |
996 | /* SWAP_USAGE_OFFLIST_BIT can only be set by this helper. */ |
997 | static void del_from_avail_list(struct swap_info_struct *si, bool swapoff) | |
a2468cc9 AL |
998 | { |
999 | int nid; | |
b228386c KS |
1000 | unsigned long pages; |
1001 | ||
1002 | spin_lock(&swap_avail_lock); | |
1003 | ||
1004 | if (swapoff) { | |
1005 | /* | |
1006 | * Forcefully remove it. Clear the SWP_WRITEOK flags for | |
1007 | * swapoff here so it's synchronized by both si->lock and | |
1008 | * swap_avail_lock, to ensure the result can be seen by | |
1009 | * add_to_avail_list. | |
1010 | */ | |
1011 | lockdep_assert_held(&si->lock); | |
1012 | si->flags &= ~SWP_WRITEOK; | |
1013 | atomic_long_or(SWAP_USAGE_OFFLIST_BIT, &si->inuse_pages); | |
1014 | } else { | |
1015 | /* | |
1016 | * If not called by swapoff, take it off-list only if it's | |
1017 | * full and SWAP_USAGE_OFFLIST_BIT is not set (strictly | |
1018 | * si->inuse_pages == pages), any concurrent slot freeing, | |
1019 | * or device already removed from plist by someone else | |
1020 | * will make this return false. | |
1021 | */ | |
1022 | pages = si->pages; | |
1023 | if (!atomic_long_try_cmpxchg(&si->inuse_pages, &pages, | |
1024 | pages | SWAP_USAGE_OFFLIST_BIT)) | |
1025 | goto skip; | |
1026 | } | |
a2468cc9 AL |
1027 | |
1028 | for_each_node(nid) | |
b85508d7 | 1029 | plist_del(&si->avail_lists[nid], &swap_avail_heads[nid]); |
b228386c KS |
1030 | |
1031 | skip: | |
1032 | spin_unlock(&swap_avail_lock); | |
a2468cc9 AL |
1033 | } |
1034 | ||
b228386c KS |
1035 | /* SWAP_USAGE_OFFLIST_BIT can only be cleared by this helper. */ |
1036 | static void add_to_avail_list(struct swap_info_struct *si, bool swapon) | |
a2468cc9 | 1037 | { |
b228386c KS |
1038 | int nid; |
1039 | long val; | |
1040 | unsigned long pages; | |
1041 | ||
a2468cc9 | 1042 | spin_lock(&swap_avail_lock); |
b228386c KS |
1043 | |
1044 | /* Corresponding to SWP_WRITEOK clearing in del_from_avail_list */ | |
1045 | if (swapon) { | |
1046 | lockdep_assert_held(&si->lock); | |
1047 | si->flags |= SWP_WRITEOK; | |
1048 | } else { | |
1049 | if (!(READ_ONCE(si->flags) & SWP_WRITEOK)) | |
1050 | goto skip; | |
1051 | } | |
1052 | ||
1053 | if (!(atomic_long_read(&si->inuse_pages) & SWAP_USAGE_OFFLIST_BIT)) | |
1054 | goto skip; | |
1055 | ||
1056 | val = atomic_long_fetch_and_relaxed(~SWAP_USAGE_OFFLIST_BIT, &si->inuse_pages); | |
1057 | ||
1058 | /* | |
1059 | * When device is full and device is on the plist, only one updater will | |
1060 | * see (inuse_pages == si->pages) and will call del_from_avail_list. If | |
1061 | * that updater happen to be here, just skip adding. | |
1062 | */ | |
1063 | pages = si->pages; | |
1064 | if (val == pages) { | |
1065 | /* Just like the cmpxchg in del_from_avail_list */ | |
1066 | if (atomic_long_try_cmpxchg(&si->inuse_pages, &pages, | |
1067 | pages | SWAP_USAGE_OFFLIST_BIT)) | |
1068 | goto skip; | |
1069 | } | |
1070 | ||
1071 | for_each_node(nid) | |
1072 | plist_add(&si->avail_lists[nid], &swap_avail_heads[nid]); | |
1073 | ||
1074 | skip: | |
a2468cc9 AL |
1075 | spin_unlock(&swap_avail_lock); |
1076 | } | |
1077 | ||
b228386c KS |
1078 | /* |
1079 | * swap_usage_add / swap_usage_sub of each slot are serialized by ci->lock | |
1080 | * within each cluster, so the total contribution to the global counter should | |
1081 | * always be positive and cannot exceed the total number of usable slots. | |
1082 | */ | |
1083 | static bool swap_usage_add(struct swap_info_struct *si, unsigned int nr_entries) | |
38d8b4e6 | 1084 | { |
b228386c | 1085 | long val = atomic_long_add_return_relaxed(nr_entries, &si->inuse_pages); |
5168a68e | 1086 | |
b228386c KS |
1087 | /* |
1088 | * If device is full, and SWAP_USAGE_OFFLIST_BIT is not set, | |
1089 | * remove it from the plist. | |
1090 | */ | |
1091 | if (unlikely(val == si->pages)) { | |
1092 | del_from_avail_list(si, false); | |
1093 | return true; | |
38d8b4e6 | 1094 | } |
b228386c KS |
1095 | |
1096 | return false; | |
38d8b4e6 HY |
1097 | } |
1098 | ||
b228386c | 1099 | static void swap_usage_sub(struct swap_info_struct *si, unsigned int nr_entries) |
a2468cc9 | 1100 | { |
b228386c | 1101 | long val = atomic_long_sub_return_relaxed(nr_entries, &si->inuse_pages); |
a2468cc9 | 1102 | |
b228386c KS |
1103 | /* |
1104 | * If device is not full, and SWAP_USAGE_OFFLIST_BIT is set, | |
2310f089 | 1105 | * add it to the plist. |
b228386c KS |
1106 | */ |
1107 | if (unlikely(val & SWAP_USAGE_OFFLIST_BIT)) | |
1108 | add_to_avail_list(si, false); | |
1109 | } | |
1110 | ||
1111 | static void swap_range_alloc(struct swap_info_struct *si, | |
1112 | unsigned int nr_entries) | |
1113 | { | |
1114 | if (swap_usage_add(si, nr_entries)) { | |
1115 | if (vm_swap_full()) | |
1116 | schedule_work(&si->reclaim_work); | |
1117 | } | |
a2468cc9 AL |
1118 | } |
1119 | ||
38d8b4e6 HY |
1120 | static void swap_range_free(struct swap_info_struct *si, unsigned long offset, |
1121 | unsigned int nr_entries) | |
1122 | { | |
3852f676 | 1123 | unsigned long begin = offset; |
38d8b4e6 HY |
1124 | unsigned long end = offset + nr_entries - 1; |
1125 | void (*swap_slot_free_notify)(struct block_device *, unsigned long); | |
0ca0c24e UA |
1126 | unsigned int i; |
1127 | ||
1128 | /* | |
1129 | * Use atomic clear_bit operations only on zeromap instead of non-atomic | |
1130 | * bitmap_clear to prevent adjacent bits corruption due to simultaneous writes. | |
1131 | */ | |
4f79384a | 1132 | for (i = 0; i < nr_entries; i++) { |
0ca0c24e | 1133 | clear_bit(offset + i, si->zeromap); |
4f79384a KS |
1134 | zswap_invalidate(swp_entry(si->type, offset + i)); |
1135 | } | |
38d8b4e6 | 1136 | |
38d8b4e6 HY |
1137 | if (si->flags & SWP_BLKDEV) |
1138 | swap_slot_free_notify = | |
1139 | si->bdev->bd_disk->fops->swap_slot_free_notify; | |
1140 | else | |
1141 | swap_slot_free_notify = NULL; | |
1142 | while (offset <= end) { | |
8a84802e | 1143 | arch_swap_invalidate_page(si->type, offset); |
38d8b4e6 HY |
1144 | if (swap_slot_free_notify) |
1145 | swap_slot_free_notify(si->bdev, offset); | |
1146 | offset++; | |
1147 | } | |
3852f676 | 1148 | clear_shadow_from_swap_cache(si->type, begin, end); |
64cf264c YA |
1149 | |
1150 | /* | |
1151 | * Make sure that try_to_unuse() observes si->inuse_pages reaching 0 | |
1152 | * only after the above cleanups are done. | |
1153 | */ | |
1154 | smp_wmb(); | |
1155 | atomic_long_add(nr_entries, &nr_swap_pages); | |
b228386c | 1156 | swap_usage_sub(si, nr_entries); |
38d8b4e6 HY |
1157 | } |
1158 | ||
9a0ddeb7 KS |
1159 | static bool get_swap_device_info(struct swap_info_struct *si) |
1160 | { | |
1161 | if (!percpu_ref_tryget_live(&si->users)) | |
1162 | return false; | |
1163 | /* | |
1164 | * Guarantee the si->users are checked before accessing other | |
1165 | * fields of swap_info_struct, and si->flags (SWP_WRITEOK) is | |
1166 | * up to dated. | |
1167 | * | |
1168 | * Paired with the spin_unlock() after setup_swap_info() in | |
1169 | * enable_swap_info(), and smp_wmb() in swapoff. | |
1170 | */ | |
1171 | smp_rmb(); | |
1172 | return true; | |
1173 | } | |
1174 | ||
1b7e9002 KS |
1175 | /* |
1176 | * Fast path try to get swap entries with specified order from current | |
1177 | * CPU's swap entry pool (a cluster). | |
1178 | */ | |
b487a2da KS |
1179 | static bool swap_alloc_fast(swp_entry_t *entry, |
1180 | int order) | |
1b7e9002 KS |
1181 | { |
1182 | struct swap_cluster_info *ci; | |
1183 | struct swap_info_struct *si; | |
0ff67f99 | 1184 | unsigned int offset, found = SWAP_ENTRY_INVALID; |
1b7e9002 KS |
1185 | |
1186 | /* | |
1187 | * Once allocated, swap_info_struct will never be completely freed, | |
1188 | * so checking it's liveness by get_swap_device_info is enough. | |
1189 | */ | |
1190 | si = this_cpu_read(percpu_swap_cluster.si[order]); | |
1191 | offset = this_cpu_read(percpu_swap_cluster.offset[order]); | |
1192 | if (!si || !offset || !get_swap_device_info(si)) | |
0ff67f99 | 1193 | return false; |
1b7e9002 | 1194 | |
0ff67f99 KS |
1195 | ci = lock_cluster(si, offset); |
1196 | if (cluster_is_usable(ci, order)) { | |
1b7e9002 KS |
1197 | if (cluster_is_empty(ci)) |
1198 | offset = cluster_offset(si, ci); | |
b487a2da | 1199 | found = alloc_swap_scan_cluster(si, ci, offset, order, SWAP_HAS_CACHE); |
0ff67f99 KS |
1200 | if (found) |
1201 | *entry = swp_entry(si->type, found); | |
1202 | } else { | |
1203 | unlock_cluster(ci); | |
1b7e9002 KS |
1204 | } |
1205 | ||
1206 | put_swap_device(si); | |
0ff67f99 | 1207 | return !!found; |
1b7e9002 KS |
1208 | } |
1209 | ||
b487a2da KS |
1210 | /* Rotate the device and switch to a new cluster */ |
1211 | static bool swap_alloc_slow(swp_entry_t *entry, | |
1212 | int order) | |
1da177e4 | 1213 | { |
a2468cc9 | 1214 | int node; |
b487a2da KS |
1215 | unsigned long offset; |
1216 | struct swap_info_struct *si, *next; | |
1da177e4 | 1217 | |
b487a2da | 1218 | node = numa_node_id(); |
b50da6e9 | 1219 | spin_lock(&swap_avail_lock); |
18ab4d4c | 1220 | start_over: |
a2468cc9 | 1221 | plist_for_each_entry_safe(si, next, &swap_avail_heads[node], avail_lists[node]) { |
b487a2da | 1222 | /* Rotate the device and switch to a new cluster */ |
a2468cc9 | 1223 | plist_requeue(&si->avail_lists[node], &swap_avail_heads[node]); |
18ab4d4c | 1224 | spin_unlock(&swap_avail_lock); |
9a0ddeb7 | 1225 | if (get_swap_device_info(si)) { |
0ff67f99 | 1226 | offset = cluster_alloc_swap_entry(si, order, SWAP_HAS_CACHE); |
9a0ddeb7 | 1227 | put_swap_device(si); |
0ff67f99 | 1228 | if (offset) { |
b487a2da KS |
1229 | *entry = swp_entry(si->type, offset); |
1230 | return true; | |
0ff67f99 KS |
1231 | } |
1232 | if (order) | |
b487a2da | 1233 | return false; |
9a0ddeb7 | 1234 | } |
36005bae | 1235 | |
18ab4d4c | 1236 | spin_lock(&swap_avail_lock); |
adfab836 DS |
1237 | /* |
1238 | * if we got here, it's likely that si was almost full before, | |
bb243f7d ML |
1239 | * and since scan_swap_map_slots() can drop the si->lock, |
1240 | * multiple callers probably all tried to get a page from the | |
1241 | * same si and it filled up before we could get one; or, the si | |
1242 | * filled up between us dropping swap_avail_lock and taking | |
1243 | * si->lock. Since we dropped the swap_avail_lock, the | |
1244 | * swap_avail_head list may have been modified; so if next is | |
1245 | * still in the swap_avail_head list then try it, otherwise | |
1246 | * start over if we have not gotten any slots. | |
adfab836 | 1247 | */ |
a2468cc9 | 1248 | if (plist_node_empty(&next->avail_lists[node])) |
18ab4d4c | 1249 | goto start_over; |
1da177e4 | 1250 | } |
18ab4d4c | 1251 | spin_unlock(&swap_avail_lock); |
b487a2da KS |
1252 | return false; |
1253 | } | |
1254 | ||
1255 | /** | |
1256 | * folio_alloc_swap - allocate swap space for a folio | |
1257 | * @folio: folio we want to move to swap | |
1258 | * @gfp: gfp mask for shadow nodes | |
1259 | * | |
1260 | * Allocate swap space for the folio and add the folio to the | |
1261 | * swap cache. | |
1262 | * | |
1263 | * Context: Caller needs to hold the folio lock. | |
1264 | * Return: Whether the folio was added to the swap cache. | |
1265 | */ | |
1266 | int folio_alloc_swap(struct folio *folio, gfp_t gfp) | |
1267 | { | |
1268 | unsigned int order = folio_order(folio); | |
1269 | unsigned int size = 1 << order; | |
1270 | swp_entry_t entry = {}; | |
1271 | ||
1272 | VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio); | |
1273 | VM_BUG_ON_FOLIO(!folio_test_uptodate(folio), folio); | |
1274 | ||
9a9794a8 KS |
1275 | if (order) { |
1276 | /* | |
1277 | * Reject large allocation when THP_SWAP is disabled, | |
1278 | * the caller should split the folio and try again. | |
1279 | */ | |
1280 | if (!IS_ENABLED(CONFIG_THP_SWAP)) | |
1281 | return -EAGAIN; | |
1282 | ||
1283 | /* | |
1284 | * Allocation size should never exceed cluster size | |
1285 | * (HPAGE_PMD_SIZE). | |
1286 | */ | |
1287 | if (size > SWAPFILE_CLUSTER) { | |
1288 | VM_WARN_ON_ONCE(1); | |
1289 | return -EINVAL; | |
1290 | } | |
b487a2da KS |
1291 | } |
1292 | ||
1293 | local_lock(&percpu_swap_cluster.lock); | |
1294 | if (!swap_alloc_fast(&entry, order)) | |
1295 | swap_alloc_slow(&entry, order); | |
1b7e9002 | 1296 | local_unlock(&percpu_swap_cluster.lock); |
b487a2da | 1297 | |
0ff67f99 | 1298 | /* Need to call this even if allocation failed, for MEMCG_SWAP_FAIL. */ |
b487a2da KS |
1299 | if (mem_cgroup_try_charge_swap(folio, entry)) |
1300 | goto out_free; | |
1301 | ||
1302 | if (!entry.val) | |
1303 | return -ENOMEM; | |
1304 | ||
1305 | /* | |
1306 | * XArray node allocations from PF_MEMALLOC contexts could | |
1307 | * completely exhaust the page allocator. __GFP_NOMEMALLOC | |
1308 | * stops emergency reserves from being allocated. | |
1309 | * | |
1310 | * TODO: this could cause a theoretical memory reclaim | |
1311 | * deadlock in the swap out path. | |
1312 | */ | |
1313 | if (add_to_swap_cache(folio, entry, gfp | __GFP_NOMEMALLOC, NULL)) | |
1314 | goto out_free; | |
1315 | ||
1316 | atomic_long_sub(size, &nr_swap_pages); | |
1317 | return 0; | |
1318 | ||
1319 | out_free: | |
1320 | put_swap_folio(folio, entry); | |
1321 | return -ENOMEM; | |
36005bae TC |
1322 | } |
1323 | ||
afba72b1 | 1324 | static struct swap_info_struct *_swap_info_get(swp_entry_t entry) |
1da177e4 | 1325 | { |
b85508d7 | 1326 | struct swap_info_struct *si; |
eb085574 | 1327 | unsigned long offset; |
1da177e4 LT |
1328 | |
1329 | if (!entry.val) | |
1330 | goto out; | |
b85508d7 BS |
1331 | si = swp_swap_info(entry); |
1332 | if (!si) | |
1da177e4 | 1333 | goto bad_nofile; |
b85508d7 | 1334 | if (data_race(!(si->flags & SWP_USED))) |
1da177e4 LT |
1335 | goto bad_device; |
1336 | offset = swp_offset(entry); | |
b85508d7 | 1337 | if (offset >= si->max) |
1da177e4 | 1338 | goto bad_offset; |
b85508d7 | 1339 | if (data_race(!si->swap_map[swp_offset(entry)])) |
afba72b1 | 1340 | goto bad_free; |
b85508d7 | 1341 | return si; |
1da177e4 | 1342 | |
afba72b1 ML |
1343 | bad_free: |
1344 | pr_err("%s: %s%08lx\n", __func__, Unused_offset, entry.val); | |
1345 | goto out; | |
1da177e4 | 1346 | bad_offset: |
cf532faa | 1347 | pr_err("%s: %s%08lx\n", __func__, Bad_offset, entry.val); |
1da177e4 LT |
1348 | goto out; |
1349 | bad_device: | |
cf532faa | 1350 | pr_err("%s: %s%08lx\n", __func__, Unused_file, entry.val); |
1da177e4 LT |
1351 | goto out; |
1352 | bad_nofile: | |
cf532faa | 1353 | pr_err("%s: %s%08lx\n", __func__, Bad_file, entry.val); |
1da177e4 LT |
1354 | out: |
1355 | return NULL; | |
886bb7e9 | 1356 | } |
1da177e4 | 1357 | |
9c1c38bc | 1358 | static unsigned char swap_entry_put_locked(struct swap_info_struct *si, |
835b8688 KS |
1359 | struct swap_cluster_info *ci, |
1360 | swp_entry_t entry, | |
9c1c38bc | 1361 | unsigned char usage) |
1da177e4 | 1362 | { |
835b8688 | 1363 | unsigned long offset = swp_offset(entry); |
8d69aaee HD |
1364 | unsigned char count; |
1365 | unsigned char has_cache; | |
235b6217 | 1366 | |
b85508d7 | 1367 | count = si->swap_map[offset]; |
235b6217 | 1368 | |
253d553b HD |
1369 | has_cache = count & SWAP_HAS_CACHE; |
1370 | count &= ~SWAP_HAS_CACHE; | |
355cfa73 | 1371 | |
253d553b | 1372 | if (usage == SWAP_HAS_CACHE) { |
355cfa73 | 1373 | VM_BUG_ON(!has_cache); |
253d553b | 1374 | has_cache = 0; |
aaa46865 HD |
1375 | } else if (count == SWAP_MAP_SHMEM) { |
1376 | /* | |
1377 | * Or we could insist on shmem.c using a special | |
1378 | * swap_shmem_free() and free_shmem_swap_and_cache()... | |
1379 | */ | |
1380 | count = 0; | |
570a335b HD |
1381 | } else if ((count & ~COUNT_CONTINUED) <= SWAP_MAP_MAX) { |
1382 | if (count == COUNT_CONTINUED) { | |
b85508d7 | 1383 | if (swap_count_continued(si, offset, count)) |
570a335b HD |
1384 | count = SWAP_MAP_MAX | COUNT_CONTINUED; |
1385 | else | |
1386 | count = SWAP_MAP_MAX; | |
1387 | } else | |
1388 | count--; | |
1389 | } | |
253d553b | 1390 | |
253d553b | 1391 | usage = count | has_cache; |
a449bf58 | 1392 | if (usage) |
b85508d7 | 1393 | WRITE_ONCE(si->swap_map[offset], usage); |
a449bf58 | 1394 | else |
835b8688 | 1395 | swap_entries_free(si, ci, entry, 1); |
7c00bafe | 1396 | |
b32d5f32 HY |
1397 | return usage; |
1398 | } | |
1399 | ||
eb085574 | 1400 | /* |
a95722a0 | 1401 | * When we get a swap entry, if there aren't some other ways to |
d4a34d7f HY |
1402 | * prevent swapoff, such as the folio in swap cache is locked, RCU |
1403 | * reader side is locked, etc., the swap entry may become invalid | |
1404 | * because of swapoff. Then, we need to enclose all swap related | |
1405 | * functions with get_swap_device() and put_swap_device(), unless the | |
1406 | * swap functions call get/put_swap_device() by themselves. | |
a95722a0 | 1407 | * |
d4a34d7f HY |
1408 | * RCU reader side lock (including any spinlock) is sufficient to |
1409 | * prevent swapoff, because synchronize_rcu() is called in swapoff() | |
1410 | * before freeing data structures. | |
82b1c07a | 1411 | * |
eb085574 HY |
1412 | * Check whether swap entry is valid in the swap device. If so, |
1413 | * return pointer to swap_info_struct, and keep the swap entry valid | |
1414 | * via preventing the swap device from being swapoff, until | |
1415 | * put_swap_device() is called. Otherwise return NULL. | |
1416 | * | |
eb085574 | 1417 | * Notice that swapoff or swapoff+swapon can still happen before the |
63d8620e ML |
1418 | * percpu_ref_tryget_live() in get_swap_device() or after the |
1419 | * percpu_ref_put() in put_swap_device() if there isn't any other way | |
a95722a0 HY |
1420 | * to prevent swapoff. The caller must be prepared for that. For |
1421 | * example, the following situation is possible. | |
eb085574 HY |
1422 | * |
1423 | * CPU1 CPU2 | |
1424 | * do_swap_page() | |
1425 | * ... swapoff+swapon | |
1426 | * __read_swap_cache_async() | |
1427 | * swapcache_prepare() | |
1428 | * __swap_duplicate() | |
1429 | * // check swap_map | |
1430 | * // verify PTE not changed | |
1431 | * | |
1432 | * In __swap_duplicate(), the swap_map need to be checked before | |
1433 | * changing partly because the specified swap entry may be for another | |
1434 | * swap device which has been swapoff. And in do_swap_page(), after | |
1435 | * the page is read from the swap device, the PTE is verified not | |
1436 | * changed with the page table locked to check whether the swap device | |
1437 | * has been swapoff or swapoff+swapon. | |
1438 | */ | |
1439 | struct swap_info_struct *get_swap_device(swp_entry_t entry) | |
1440 | { | |
1441 | struct swap_info_struct *si; | |
1442 | unsigned long offset; | |
1443 | ||
1444 | if (!entry.val) | |
1445 | goto out; | |
1446 | si = swp_swap_info(entry); | |
1447 | if (!si) | |
1448 | goto bad_nofile; | |
9a0ddeb7 | 1449 | if (!get_swap_device_info(si)) |
63d8620e | 1450 | goto out; |
eb085574 HY |
1451 | offset = swp_offset(entry); |
1452 | if (offset >= si->max) | |
63d8620e | 1453 | goto put_out; |
eb085574 HY |
1454 | |
1455 | return si; | |
1456 | bad_nofile: | |
1457 | pr_err("%s: %s%08lx\n", __func__, Bad_file, entry.val); | |
1458 | out: | |
1459 | return NULL; | |
63d8620e | 1460 | put_out: |
23b230ba | 1461 | pr_err("%s: %s%08lx\n", __func__, Bad_offset, entry.val); |
63d8620e | 1462 | percpu_ref_put(&si->users); |
eb085574 HY |
1463 | return NULL; |
1464 | } | |
1465 | ||
d4f8000b KS |
1466 | static void swap_entries_put_cache(struct swap_info_struct *si, |
1467 | swp_entry_t entry, int nr) | |
b32d5f32 | 1468 | { |
b32d5f32 | 1469 | unsigned long offset = swp_offset(entry); |
d4f8000b | 1470 | struct swap_cluster_info *ci; |
b32d5f32 | 1471 | |
0b310d9c | 1472 | ci = lock_cluster(si, offset); |
d4f8000b KS |
1473 | if (swap_only_has_cache(si, offset, nr)) |
1474 | swap_entries_free(si, ci, entry, nr); | |
1475 | else { | |
1476 | for (int i = 0; i < nr; i++, entry.val++) | |
1477 | swap_entry_put_locked(si, ci, entry, SWAP_HAS_CACHE); | |
1478 | } | |
4f79384a | 1479 | unlock_cluster(ci); |
7c00bafe | 1480 | } |
355cfa73 | 1481 | |
4d71d906 KS |
1482 | static bool swap_entries_put_map(struct swap_info_struct *si, |
1483 | swp_entry_t entry, int nr) | |
bea67dcc BS |
1484 | { |
1485 | unsigned long offset = swp_offset(entry); | |
bea67dcc BS |
1486 | struct swap_cluster_info *ci; |
1487 | bool has_cache = false; | |
1488 | unsigned char count; | |
1489 | int i; | |
1490 | ||
f2252acf | 1491 | if (nr <= 1) |
bea67dcc | 1492 | goto fallback; |
f2252acf KS |
1493 | count = swap_count(data_race(si->swap_map[offset])); |
1494 | if (count != 1 && count != SWAP_MAP_SHMEM) | |
bea67dcc BS |
1495 | goto fallback; |
1496 | ||
0b310d9c | 1497 | ci = lock_cluster(si, offset); |
bea67dcc | 1498 | if (!swap_is_last_map(si, offset, nr, &has_cache)) { |
4d71d906 | 1499 | goto locked_fallback; |
bea67dcc | 1500 | } |
4f79384a | 1501 | if (!has_cache) |
64944ef6 | 1502 | swap_entries_free(si, ci, entry, nr); |
46e0ab2c KS |
1503 | else |
1504 | for (i = 0; i < nr; i++) | |
1505 | WRITE_ONCE(si->swap_map[offset + i], SWAP_HAS_CACHE); | |
0b310d9c | 1506 | unlock_cluster(ci); |
bea67dcc | 1507 | |
bea67dcc BS |
1508 | return has_cache; |
1509 | ||
1510 | fallback: | |
4d71d906 KS |
1511 | ci = lock_cluster(si, offset); |
1512 | locked_fallback: | |
1513 | for (i = 0; i < nr; i++, entry.val++) { | |
1514 | count = swap_entry_put_locked(si, ci, entry, 1); | |
1515 | if (count == SWAP_HAS_CACHE) | |
1516 | has_cache = true; | |
bea67dcc | 1517 | } |
4d71d906 | 1518 | unlock_cluster(ci); |
bea67dcc | 1519 | return has_cache; |
4d71d906 | 1520 | |
bea67dcc BS |
1521 | } |
1522 | ||
650975d2 | 1523 | /* |
ec9827cd KS |
1524 | * Only functions with "_nr" suffix are able to free entries spanning |
1525 | * cross multi clusters, so ensure the range is within a single cluster | |
1526 | * when freeing entries with functions without "_nr" suffix. | |
650975d2 | 1527 | */ |
4d71d906 KS |
1528 | static bool swap_entries_put_map_nr(struct swap_info_struct *si, |
1529 | swp_entry_t entry, int nr) | |
1530 | { | |
1531 | int cluster_nr, cluster_rest; | |
1532 | unsigned long offset = swp_offset(entry); | |
1533 | bool has_cache = false; | |
1534 | ||
1535 | cluster_rest = SWAPFILE_CLUSTER - offset % SWAPFILE_CLUSTER; | |
1536 | while (nr) { | |
1537 | cluster_nr = min(nr, cluster_rest); | |
1538 | has_cache |= swap_entries_put_map(si, entry, cluster_nr); | |
1539 | cluster_rest = SWAPFILE_CLUSTER; | |
1540 | nr -= cluster_nr; | |
1541 | entry.val += cluster_nr; | |
1542 | } | |
1543 | ||
bea67dcc BS |
1544 | return has_cache; |
1545 | } | |
1546 | ||
650975d2 | 1547 | /* |
64944ef6 KS |
1548 | * Check if it's the last ref of swap entry in the freeing path. |
1549 | * Qualified vlaue includes 1, SWAP_HAS_CACHE or SWAP_MAP_SHMEM. | |
650975d2 | 1550 | */ |
64944ef6 KS |
1551 | static inline bool __maybe_unused swap_is_last_ref(unsigned char count) |
1552 | { | |
1553 | return (count == SWAP_HAS_CACHE) || (count == 1) || | |
1554 | (count == SWAP_MAP_SHMEM); | |
1555 | } | |
1556 | ||
1557 | /* | |
1558 | * Drop the last ref of swap entries, caller have to ensure all entries | |
1559 | * belong to the same cgroup and cluster. | |
1560 | */ | |
1561 | static void swap_entries_free(struct swap_info_struct *si, | |
1562 | struct swap_cluster_info *ci, | |
1563 | swp_entry_t entry, unsigned int nr_pages) | |
7c00bafe | 1564 | { |
7c00bafe | 1565 | unsigned long offset = swp_offset(entry); |
b85508d7 | 1566 | unsigned char *map = si->swap_map + offset; |
650975d2 | 1567 | unsigned char *map_end = map + nr_pages; |
3f641cf9 KS |
1568 | |
1569 | /* It should never free entries across different clusters */ | |
1570 | VM_BUG_ON(ci != offset_to_cluster(si, offset + nr_pages - 1)); | |
3b644773 KS |
1571 | VM_BUG_ON(cluster_is_empty(ci)); |
1572 | VM_BUG_ON(ci->count < nr_pages); | |
1573 | ||
1574 | ci->count -= nr_pages; | |
650975d2 | 1575 | do { |
64944ef6 | 1576 | VM_BUG_ON(!swap_is_last_ref(*map)); |
650975d2 KS |
1577 | *map = 0; |
1578 | } while (++map < map_end); | |
235b6217 | 1579 | |
650975d2 | 1580 | mem_cgroup_uncharge_swap(entry, nr_pages); |
b85508d7 | 1581 | swap_range_free(si, offset, nr_pages); |
3b644773 KS |
1582 | |
1583 | if (!ci->count) | |
1584 | free_cluster(si, ci); | |
1585 | else | |
1586 | partial_free_cluster(si, ci); | |
1da177e4 LT |
1587 | } |
1588 | ||
54f7a49c BS |
1589 | /* |
1590 | * Caller has made sure that the swap device corresponding to entry | |
1591 | * is still around or has not been recycled. | |
1592 | */ | |
ebfba004 CH |
1593 | void swap_free_nr(swp_entry_t entry, int nr_pages) |
1594 | { | |
1595 | int nr; | |
1596 | struct swap_info_struct *sis; | |
1597 | unsigned long offset = swp_offset(entry); | |
1598 | ||
1599 | sis = _swap_info_get(entry); | |
1600 | if (!sis) | |
1601 | return; | |
1602 | ||
1603 | while (nr_pages) { | |
1604 | nr = min_t(int, nr_pages, SWAPFILE_CLUSTER - offset % SWAPFILE_CLUSTER); | |
ec9827cd | 1605 | swap_entries_put_map(sis, swp_entry(sis->type, offset), nr); |
ebfba004 CH |
1606 | offset += nr; |
1607 | nr_pages -= nr; | |
1608 | } | |
1609 | } | |
1610 | ||
cb4b86ba KH |
1611 | /* |
1612 | * Called after dropping swapcache to decrease refcnt to swap entries. | |
1613 | */ | |
4081f744 | 1614 | void put_swap_folio(struct folio *folio, swp_entry_t entry) |
38d8b4e6 | 1615 | { |
38d8b4e6 | 1616 | struct swap_info_struct *si; |
9faaa0f8 | 1617 | int size = 1 << swap_entry_order(folio_order(folio)); |
fe5266d5 | 1618 | |
a3aea839 | 1619 | si = _swap_info_get(entry); |
38d8b4e6 HY |
1620 | if (!si) |
1621 | return; | |
1622 | ||
d4f8000b | 1623 | swap_entries_put_cache(si, entry, size); |
38d8b4e6 | 1624 | } |
59807685 | 1625 | |
eb085574 | 1626 | int __swap_count(swp_entry_t entry) |
aa8d22a1 | 1627 | { |
f9f956b5 | 1628 | struct swap_info_struct *si = swp_swap_info(entry); |
aa8d22a1 MK |
1629 | pgoff_t offset = swp_offset(entry); |
1630 | ||
f9f956b5 | 1631 | return swap_count(si->swap_map[offset]); |
aa8d22a1 MK |
1632 | } |
1633 | ||
14d01ee9 MWO |
1634 | /* |
1635 | * How many references to @entry are currently swapped out? | |
1636 | * This does not give an exact answer when swap count is continued, | |
1637 | * but does include the high COUNT_CONTINUED flag to allow for that. | |
1638 | */ | |
c523aa89 | 1639 | bool swap_entry_swapped(struct swap_info_struct *si, swp_entry_t entry) |
322b8afe | 1640 | { |
322b8afe HY |
1641 | pgoff_t offset = swp_offset(entry); |
1642 | struct swap_cluster_info *ci; | |
14d01ee9 | 1643 | int count; |
322b8afe | 1644 | |
0b310d9c | 1645 | ci = lock_cluster(si, offset); |
322b8afe | 1646 | count = swap_count(si->swap_map[offset]); |
0b310d9c | 1647 | unlock_cluster(ci); |
c523aa89 | 1648 | return !!count; |
322b8afe HY |
1649 | } |
1650 | ||
8334b962 MK |
1651 | /* |
1652 | * How many references to @entry are currently swapped out? | |
1653 | * This considers COUNT_CONTINUED so it returns exact answer. | |
1654 | */ | |
1655 | int swp_swapcount(swp_entry_t entry) | |
1656 | { | |
1657 | int count, tmp_count, n; | |
b85508d7 | 1658 | struct swap_info_struct *si; |
235b6217 | 1659 | struct swap_cluster_info *ci; |
8334b962 MK |
1660 | struct page *page; |
1661 | pgoff_t offset; | |
1662 | unsigned char *map; | |
1663 | ||
b85508d7 BS |
1664 | si = _swap_info_get(entry); |
1665 | if (!si) | |
8334b962 MK |
1666 | return 0; |
1667 | ||
235b6217 HY |
1668 | offset = swp_offset(entry); |
1669 | ||
0b310d9c | 1670 | ci = lock_cluster(si, offset); |
235b6217 | 1671 | |
b85508d7 | 1672 | count = swap_count(si->swap_map[offset]); |
8334b962 MK |
1673 | if (!(count & COUNT_CONTINUED)) |
1674 | goto out; | |
1675 | ||
1676 | count &= ~COUNT_CONTINUED; | |
1677 | n = SWAP_MAP_MAX + 1; | |
1678 | ||
b85508d7 | 1679 | page = vmalloc_to_page(si->swap_map + offset); |
8334b962 MK |
1680 | offset &= ~PAGE_MASK; |
1681 | VM_BUG_ON(page_private(page) != SWP_CONTINUED); | |
1682 | ||
1683 | do { | |
a8ae4991 | 1684 | page = list_next_entry(page, lru); |
829c3151 | 1685 | map = kmap_local_page(page); |
8334b962 | 1686 | tmp_count = map[offset]; |
829c3151 | 1687 | kunmap_local(map); |
8334b962 MK |
1688 | |
1689 | count += (tmp_count & ~COUNT_CONTINUED) * n; | |
1690 | n *= (SWAP_CONT_MAX + 1); | |
1691 | } while (tmp_count & COUNT_CONTINUED); | |
1692 | out: | |
0b310d9c | 1693 | unlock_cluster(ci); |
8334b962 MK |
1694 | return count; |
1695 | } | |
1696 | ||
e0709829 | 1697 | static bool swap_page_trans_huge_swapped(struct swap_info_struct *si, |
d7d0d389 | 1698 | swp_entry_t entry, int order) |
e0709829 HY |
1699 | { |
1700 | struct swap_cluster_info *ci; | |
1701 | unsigned char *map = si->swap_map; | |
d7d0d389 | 1702 | unsigned int nr_pages = 1 << order; |
e0709829 | 1703 | unsigned long roffset = swp_offset(entry); |
d7d0d389 | 1704 | unsigned long offset = round_down(roffset, nr_pages); |
e0709829 HY |
1705 | int i; |
1706 | bool ret = false; | |
1707 | ||
0b310d9c KS |
1708 | ci = lock_cluster(si, offset); |
1709 | if (nr_pages == 1) { | |
afa4711e | 1710 | if (swap_count(map[roffset])) |
e0709829 HY |
1711 | ret = true; |
1712 | goto unlock_out; | |
1713 | } | |
d7d0d389 | 1714 | for (i = 0; i < nr_pages; i++) { |
afa4711e | 1715 | if (swap_count(map[offset + i])) { |
e0709829 HY |
1716 | ret = true; |
1717 | break; | |
1718 | } | |
1719 | } | |
1720 | unlock_out: | |
0b310d9c | 1721 | unlock_cluster(ci); |
e0709829 HY |
1722 | return ret; |
1723 | } | |
1724 | ||
2397f780 | 1725 | static bool folio_swapped(struct folio *folio) |
e0709829 | 1726 | { |
3d2c9087 | 1727 | swp_entry_t entry = folio->swap; |
14d01ee9 MWO |
1728 | struct swap_info_struct *si = _swap_info_get(entry); |
1729 | ||
1730 | if (!si) | |
1731 | return false; | |
e0709829 | 1732 | |
2397f780 | 1733 | if (!IS_ENABLED(CONFIG_THP_SWAP) || likely(!folio_test_large(folio))) |
c523aa89 | 1734 | return swap_entry_swapped(si, entry); |
e0709829 | 1735 | |
d7d0d389 | 1736 | return swap_page_trans_huge_swapped(si, entry, folio_order(folio)); |
e0709829 | 1737 | } |
ba3c4ce6 | 1738 | |
862590ac | 1739 | static bool folio_swapcache_freeable(struct folio *folio) |
1da177e4 | 1740 | { |
2397f780 | 1741 | VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio); |
1da177e4 | 1742 | |
2397f780 | 1743 | if (!folio_test_swapcache(folio)) |
bdb0ed54 | 1744 | return false; |
2397f780 | 1745 | if (folio_test_writeback(folio)) |
bdb0ed54 | 1746 | return false; |
1da177e4 | 1747 | |
b73d7fce HD |
1748 | /* |
1749 | * Once hibernation has begun to create its image of memory, | |
bdb0ed54 | 1750 | * there's a danger that one of the calls to folio_free_swap() |
b73d7fce HD |
1751 | * - most probably a call from __try_to_reclaim_swap() while |
1752 | * hibernation is allocating its own swap pages for the image, | |
1753 | * but conceivably even a call from memory reclaim - will free | |
bdb0ed54 MWO |
1754 | * the swap from a folio which has already been recorded in the |
1755 | * image as a clean swapcache folio, and then reuse its swap for | |
b73d7fce | 1756 | * another page of the image. On waking from hibernation, the |
bdb0ed54 | 1757 | * original folio might be freed under memory pressure, then |
b73d7fce HD |
1758 | * later read back in from swap, now with the wrong data. |
1759 | * | |
2de1a7e4 | 1760 | * Hibernation suspends storage while it is writing the image |
f90ac398 | 1761 | * to disk so check that here. |
b73d7fce | 1762 | */ |
f90ac398 | 1763 | if (pm_suspended_storage()) |
bdb0ed54 | 1764 | return false; |
b73d7fce | 1765 | |
862590ac KS |
1766 | return true; |
1767 | } | |
1768 | ||
1769 | /** | |
1770 | * folio_free_swap() - Free the swap space used for this folio. | |
1771 | * @folio: The folio to remove. | |
1772 | * | |
1773 | * If swap is getting full, or if there are no more mappings of this folio, | |
1774 | * then call folio_free_swap to free its swap space. | |
1775 | * | |
1776 | * Return: true if we were able to release the swap space. | |
1777 | */ | |
1778 | bool folio_free_swap(struct folio *folio) | |
1779 | { | |
1780 | if (!folio_swapcache_freeable(folio)) | |
1781 | return false; | |
1782 | if (folio_swapped(folio)) | |
1783 | return false; | |
1784 | ||
75fa68a5 | 1785 | delete_from_swap_cache(folio); |
2397f780 | 1786 | folio_set_dirty(folio); |
bdb0ed54 | 1787 | return true; |
68a22394 RR |
1788 | } |
1789 | ||
a62fb92a RR |
1790 | /** |
1791 | * free_swap_and_cache_nr() - Release reference on range of swap entries and | |
1792 | * reclaim their cache if no more references remain. | |
1793 | * @entry: First entry of range. | |
1794 | * @nr: Number of entries in range. | |
1795 | * | |
1796 | * For each swap entry in the contiguous range, release a reference. If any swap | |
1797 | * entries become free, try to reclaim their underlying folios, if present. The | |
1798 | * offset range is defined by [entry.offset, entry.offset + nr). | |
1da177e4 | 1799 | */ |
a62fb92a | 1800 | void free_swap_and_cache_nr(swp_entry_t entry, int nr) |
1da177e4 | 1801 | { |
a62fb92a RR |
1802 | const unsigned long start_offset = swp_offset(entry); |
1803 | const unsigned long end_offset = start_offset + nr; | |
a62fb92a RR |
1804 | struct swap_info_struct *si; |
1805 | bool any_only_cache = false; | |
1806 | unsigned long offset; | |
1da177e4 | 1807 | |
a62fb92a RR |
1808 | si = get_swap_device(entry); |
1809 | if (!si) | |
1810 | return; | |
1811 | ||
1812 | if (WARN_ON(end_offset > si->max)) | |
1813 | goto out; | |
0697212a | 1814 | |
a62fb92a RR |
1815 | /* |
1816 | * First free all entries in the range. | |
1817 | */ | |
4d71d906 | 1818 | any_only_cache = swap_entries_put_map_nr(si, entry, nr); |
a62fb92a RR |
1819 | |
1820 | /* | |
1821 | * Short-circuit the below loop if none of the entries had their | |
1822 | * reference drop to zero. | |
1823 | */ | |
1824 | if (!any_only_cache) | |
1825 | goto out; | |
82b1c07a | 1826 | |
a62fb92a | 1827 | /* |
4d71d906 | 1828 | * Now go back over the range trying to reclaim the swap cache. |
a62fb92a RR |
1829 | */ |
1830 | for (offset = start_offset; offset < end_offset; offset += nr) { | |
1831 | nr = 1; | |
1832 | if (READ_ONCE(si->swap_map[offset]) == SWAP_HAS_CACHE) { | |
1833 | /* | |
1834 | * Folios are always naturally aligned in swap so | |
1835 | * advance forward to the next boundary. Zero means no | |
1836 | * folio was found for the swap entry, so advance by 1 | |
1837 | * in this case. Negative value means folio was found | |
1838 | * but could not be reclaimed. Here we can still advance | |
1839 | * to the next boundary. | |
1840 | */ | |
1841 | nr = __try_to_reclaim_swap(si, offset, | |
862590ac | 1842 | TTRS_UNMAPPED | TTRS_FULL); |
a62fb92a RR |
1843 | if (nr == 0) |
1844 | nr = 1; | |
1845 | else if (nr < 0) | |
1846 | nr = -nr; | |
1847 | nr = ALIGN(offset + 1, nr) - offset; | |
1848 | } | |
1da177e4 | 1849 | } |
a62fb92a RR |
1850 | |
1851 | out: | |
1852 | put_swap_device(si); | |
1da177e4 LT |
1853 | } |
1854 | ||
b0cb1a19 | 1855 | #ifdef CONFIG_HIBERNATION |
bb243f7d ML |
1856 | |
1857 | swp_entry_t get_swap_page_of_type(int type) | |
1858 | { | |
1859 | struct swap_info_struct *si = swap_type_to_swap_info(type); | |
0ff67f99 | 1860 | unsigned long offset; |
bb243f7d ML |
1861 | swp_entry_t entry = {0}; |
1862 | ||
1863 | if (!si) | |
1864 | goto fail; | |
1865 | ||
1866 | /* This is called for allocating swap entry, not cache */ | |
9a0ddeb7 | 1867 | if (get_swap_device_info(si)) { |
0ff67f99 KS |
1868 | if (si->flags & SWP_WRITEOK) { |
1869 | offset = cluster_alloc_swap_entry(si, 0, 1); | |
1870 | if (offset) { | |
1871 | entry = swp_entry(si->type, offset); | |
1872 | atomic_long_dec(&nr_swap_pages); | |
1873 | } | |
1874 | } | |
9a0ddeb7 KS |
1875 | put_swap_device(si); |
1876 | } | |
bb243f7d ML |
1877 | fail: |
1878 | return entry; | |
1879 | } | |
1880 | ||
f577eb30 | 1881 | /* |
915bae9e | 1882 | * Find the swap type that corresponds to given device (if any). |
f577eb30 | 1883 | * |
915bae9e RW |
1884 | * @offset - number of the PAGE_SIZE-sized block of the device, starting |
1885 | * from 0, in which the swap header is expected to be located. | |
1886 | * | |
1887 | * This is needed for the suspend to disk (aka swsusp). | |
f577eb30 | 1888 | */ |
21bd9005 | 1889 | int swap_type_of(dev_t device, sector_t offset) |
f577eb30 | 1890 | { |
efa90a98 | 1891 | int type; |
f577eb30 | 1892 | |
21bd9005 CH |
1893 | if (!device) |
1894 | return -1; | |
915bae9e | 1895 | |
f577eb30 | 1896 | spin_lock(&swap_lock); |
efa90a98 HD |
1897 | for (type = 0; type < nr_swapfiles; type++) { |
1898 | struct swap_info_struct *sis = swap_info[type]; | |
f577eb30 | 1899 | |
915bae9e | 1900 | if (!(sis->flags & SWP_WRITEOK)) |
f577eb30 | 1901 | continue; |
b6b5bce3 | 1902 | |
21bd9005 | 1903 | if (device == sis->bdev->bd_dev) { |
4efaceb1 | 1904 | struct swap_extent *se = first_se(sis); |
915bae9e | 1905 | |
915bae9e RW |
1906 | if (se->start_block == offset) { |
1907 | spin_unlock(&swap_lock); | |
efa90a98 | 1908 | return type; |
915bae9e | 1909 | } |
f577eb30 RW |
1910 | } |
1911 | } | |
1912 | spin_unlock(&swap_lock); | |
21bd9005 CH |
1913 | return -ENODEV; |
1914 | } | |
915bae9e | 1915 | |
21bd9005 CH |
1916 | int find_first_swap(dev_t *device) |
1917 | { | |
1918 | int type; | |
915bae9e | 1919 | |
21bd9005 CH |
1920 | spin_lock(&swap_lock); |
1921 | for (type = 0; type < nr_swapfiles; type++) { | |
1922 | struct swap_info_struct *sis = swap_info[type]; | |
1923 | ||
1924 | if (!(sis->flags & SWP_WRITEOK)) | |
1925 | continue; | |
1926 | *device = sis->bdev->bd_dev; | |
1927 | spin_unlock(&swap_lock); | |
1928 | return type; | |
1929 | } | |
1930 | spin_unlock(&swap_lock); | |
f577eb30 RW |
1931 | return -ENODEV; |
1932 | } | |
1933 | ||
73c34b6a HD |
1934 | /* |
1935 | * Get the (PAGE_SIZE) block corresponding to given offset on the swapdev | |
1936 | * corresponding to given index in swap_info (swap type). | |
1937 | */ | |
1938 | sector_t swapdev_block(int type, pgoff_t offset) | |
1939 | { | |
c10d38cc | 1940 | struct swap_info_struct *si = swap_type_to_swap_info(type); |
f885056a | 1941 | struct swap_extent *se; |
73c34b6a | 1942 | |
c10d38cc | 1943 | if (!si || !(si->flags & SWP_WRITEOK)) |
73c34b6a | 1944 | return 0; |
f885056a CH |
1945 | se = offset_to_swap_extent(si, offset); |
1946 | return se->start_block + (offset - se->start_page); | |
73c34b6a HD |
1947 | } |
1948 | ||
f577eb30 RW |
1949 | /* |
1950 | * Return either the total number of swap pages of given type, or the number | |
1951 | * of free pages of that type (depending on @free) | |
1952 | * | |
1953 | * This is needed for software suspend | |
1954 | */ | |
1955 | unsigned int count_swap_pages(int type, int free) | |
1956 | { | |
1957 | unsigned int n = 0; | |
1958 | ||
efa90a98 HD |
1959 | spin_lock(&swap_lock); |
1960 | if ((unsigned int)type < nr_swapfiles) { | |
1961 | struct swap_info_struct *sis = swap_info[type]; | |
1962 | ||
ec8acf20 | 1963 | spin_lock(&sis->lock); |
efa90a98 HD |
1964 | if (sis->flags & SWP_WRITEOK) { |
1965 | n = sis->pages; | |
f577eb30 | 1966 | if (free) |
b228386c | 1967 | n -= swap_usage_in_pages(sis); |
f577eb30 | 1968 | } |
ec8acf20 | 1969 | spin_unlock(&sis->lock); |
f577eb30 | 1970 | } |
efa90a98 | 1971 | spin_unlock(&swap_lock); |
f577eb30 RW |
1972 | return n; |
1973 | } | |
73c34b6a | 1974 | #endif /* CONFIG_HIBERNATION */ |
f577eb30 | 1975 | |
9f8bdb3f | 1976 | static inline int pte_same_as_swp(pte_t pte, pte_t swp_pte) |
179ef71c | 1977 | { |
099dd687 | 1978 | return pte_same(pte_swp_clear_flags(pte), swp_pte); |
179ef71c CG |
1979 | } |
1980 | ||
1da177e4 | 1981 | /* |
72866f6f HD |
1982 | * No need to decide whether this PTE shares the swap entry with others, |
1983 | * just let do_wp_page work it out if a write is requested later - to | |
1984 | * force COW, vm_page_prot omits write permission from any private vma. | |
1da177e4 | 1985 | */ |
044d66c1 | 1986 | static int unuse_pte(struct vm_area_struct *vma, pmd_t *pmd, |
f102cd8b | 1987 | unsigned long addr, swp_entry_t entry, struct folio *folio) |
1da177e4 | 1988 | { |
f00f4843 MWO |
1989 | struct page *page; |
1990 | struct folio *swapcache; | |
044d66c1 | 1991 | spinlock_t *ptl; |
c33c7948 | 1992 | pte_t *pte, new_pte, old_pte; |
f00f4843 | 1993 | bool hwpoisoned = false; |
044d66c1 HD |
1994 | int ret = 1; |
1995 | ||
f00f4843 | 1996 | swapcache = folio; |
96db66d9 MWO |
1997 | folio = ksm_might_need_to_copy(folio, vma, addr); |
1998 | if (unlikely(!folio)) | |
9e16b7fb | 1999 | return -ENOMEM; |
f00f4843 MWO |
2000 | else if (unlikely(folio == ERR_PTR(-EHWPOISON))) { |
2001 | hwpoisoned = true; | |
2002 | folio = swapcache; | |
2003 | } | |
2004 | ||
2005 | page = folio_file_page(folio, swp_offset(entry)); | |
2006 | if (PageHWPoison(page)) | |
f985fc32 | 2007 | hwpoisoned = true; |
9e16b7fb | 2008 | |
044d66c1 | 2009 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
c33c7948 RR |
2010 | if (unlikely(!pte || !pte_same_as_swp(ptep_get(pte), |
2011 | swp_entry_to_pte(entry)))) { | |
044d66c1 HD |
2012 | ret = 0; |
2013 | goto out; | |
2014 | } | |
8a9f3ccd | 2015 | |
c33c7948 RR |
2016 | old_pte = ptep_get(pte); |
2017 | ||
f00f4843 | 2018 | if (unlikely(hwpoisoned || !folio_test_uptodate(folio))) { |
6b970599 | 2019 | swp_entry_t swp_entry; |
9f186f9e ML |
2020 | |
2021 | dec_mm_counter(vma->vm_mm, MM_SWAPENTS); | |
f985fc32 | 2022 | if (hwpoisoned) { |
f00f4843 | 2023 | swp_entry = make_hwpoison_entry(page); |
6b970599 | 2024 | } else { |
af19487f | 2025 | swp_entry = make_poisoned_swp_entry(); |
6b970599 KW |
2026 | } |
2027 | new_pte = swp_entry_to_pte(swp_entry); | |
9f186f9e | 2028 | ret = 0; |
6b970599 | 2029 | goto setpte; |
9f186f9e ML |
2030 | } |
2031 | ||
b53e24c4 PC |
2032 | /* |
2033 | * Some architectures may have to restore extra metadata to the page | |
2034 | * when reading from swap. This metadata may be indexed by swap entry | |
2035 | * so this must be called before swap_free(). | |
2036 | */ | |
f238b8c3 | 2037 | arch_swap_restore(folio_swap(entry, folio), folio); |
b53e24c4 | 2038 | |
b084d435 | 2039 | dec_mm_counter(vma->vm_mm, MM_SWAPENTS); |
d559db08 | 2040 | inc_mm_counter(vma->vm_mm, MM_ANONPAGES); |
f00f4843 MWO |
2041 | folio_get(folio); |
2042 | if (folio == swapcache) { | |
1493a191 DH |
2043 | rmap_t rmap_flags = RMAP_NONE; |
2044 | ||
2045 | /* | |
f00f4843 MWO |
2046 | * See do_swap_page(): writeback would be problematic. |
2047 | * However, we do a folio_wait_writeback() just before this | |
2048 | * call and have the folio locked. | |
1493a191 | 2049 | */ |
f00f4843 | 2050 | VM_BUG_ON_FOLIO(folio_test_writeback(folio), folio); |
c33c7948 | 2051 | if (pte_swp_exclusive(old_pte)) |
1493a191 | 2052 | rmap_flags |= RMAP_EXCLUSIVE; |
9ae2feac BS |
2053 | /* |
2054 | * We currently only expect small !anon folios, which are either | |
2055 | * fully exclusive or fully shared. If we ever get large folios | |
2056 | * here, we have to be careful. | |
2057 | */ | |
2058 | if (!folio_test_anon(folio)) { | |
2059 | VM_WARN_ON_ONCE(folio_test_large(folio)); | |
2060 | VM_WARN_ON_FOLIO(!folio_test_locked(folio), folio); | |
2061 | folio_add_new_anon_rmap(folio, vma, addr, rmap_flags); | |
2062 | } else { | |
2063 | folio_add_anon_rmap_pte(folio, page, vma, addr, rmap_flags); | |
2064 | } | |
00501b53 | 2065 | } else { /* ksm created a completely new copy */ |
15bde4ab | 2066 | folio_add_new_anon_rmap(folio, vma, addr, RMAP_EXCLUSIVE); |
f00f4843 | 2067 | folio_add_lru_vma(folio, vma); |
00501b53 | 2068 | } |
14a762dd | 2069 | new_pte = pte_mkold(mk_pte(page, vma->vm_page_prot)); |
c33c7948 | 2070 | if (pte_swp_soft_dirty(old_pte)) |
14a762dd | 2071 | new_pte = pte_mksoft_dirty(new_pte); |
c33c7948 | 2072 | if (pte_swp_uffd_wp(old_pte)) |
14a762dd | 2073 | new_pte = pte_mkuffd_wp(new_pte); |
6b970599 | 2074 | setpte: |
14a762dd | 2075 | set_pte_at(vma->vm_mm, addr, pte, new_pte); |
1da177e4 | 2076 | swap_free(entry); |
044d66c1 | 2077 | out: |
d850fa72 HD |
2078 | if (pte) |
2079 | pte_unmap_unlock(pte, ptl); | |
f00f4843 MWO |
2080 | if (folio != swapcache) { |
2081 | folio_unlock(folio); | |
2082 | folio_put(folio); | |
9e16b7fb | 2083 | } |
044d66c1 | 2084 | return ret; |
1da177e4 LT |
2085 | } |
2086 | ||
2087 | static int unuse_pte_range(struct vm_area_struct *vma, pmd_t *pmd, | |
b56a2d8a | 2088 | unsigned long addr, unsigned long end, |
10a9c496 | 2089 | unsigned int type) |
1da177e4 | 2090 | { |
d850fa72 | 2091 | pte_t *pte = NULL; |
b56a2d8a | 2092 | struct swap_info_struct *si; |
1da177e4 | 2093 | |
b56a2d8a | 2094 | si = swap_info[type]; |
1da177e4 | 2095 | do { |
f102cd8b MWO |
2096 | struct folio *folio; |
2097 | unsigned long offset; | |
3f79b187 | 2098 | unsigned char swp_count; |
d850fa72 HD |
2099 | swp_entry_t entry; |
2100 | int ret; | |
c33c7948 | 2101 | pte_t ptent; |
d850fa72 HD |
2102 | |
2103 | if (!pte++) { | |
2104 | pte = pte_offset_map(pmd, addr); | |
2105 | if (!pte) | |
2106 | break; | |
2107 | } | |
f102cd8b | 2108 | |
c33c7948 | 2109 | ptent = ptep_get_lockless(pte); |
f102cd8b | 2110 | |
c33c7948 | 2111 | if (!is_swap_pte(ptent)) |
b56a2d8a VRP |
2112 | continue; |
2113 | ||
c33c7948 | 2114 | entry = pte_to_swp_entry(ptent); |
b56a2d8a VRP |
2115 | if (swp_type(entry) != type) |
2116 | continue; | |
2117 | ||
2118 | offset = swp_offset(entry); | |
b56a2d8a | 2119 | pte_unmap(pte); |
d850fa72 HD |
2120 | pte = NULL; |
2121 | ||
f102cd8b MWO |
2122 | folio = swap_cache_get_folio(entry, vma, addr); |
2123 | if (!folio) { | |
8c63ca5b WD |
2124 | struct vm_fault vmf = { |
2125 | .vma = vma, | |
2126 | .address = addr, | |
824ddc60 | 2127 | .real_address = addr, |
8c63ca5b WD |
2128 | .pmd = pmd, |
2129 | }; | |
2130 | ||
94dc8bff | 2131 | folio = swapin_readahead(entry, GFP_HIGHUSER_MOVABLE, |
ebc5951e AR |
2132 | &vmf); |
2133 | } | |
f102cd8b | 2134 | if (!folio) { |
3f79b187 KS |
2135 | swp_count = READ_ONCE(si->swap_map[offset]); |
2136 | if (swp_count == 0 || swp_count == SWAP_MAP_BAD) | |
d850fa72 | 2137 | continue; |
b56a2d8a VRP |
2138 | return -ENOMEM; |
2139 | } | |
2140 | ||
f102cd8b MWO |
2141 | folio_lock(folio); |
2142 | folio_wait_writeback(folio); | |
2143 | ret = unuse_pte(vma, pmd, addr, entry, folio); | |
b56a2d8a | 2144 | if (ret < 0) { |
f102cd8b MWO |
2145 | folio_unlock(folio); |
2146 | folio_put(folio); | |
d850fa72 | 2147 | return ret; |
b56a2d8a VRP |
2148 | } |
2149 | ||
f102cd8b MWO |
2150 | folio_free_swap(folio); |
2151 | folio_unlock(folio); | |
2152 | folio_put(folio); | |
d850fa72 | 2153 | } while (addr += PAGE_SIZE, addr != end); |
b56a2d8a | 2154 | |
d850fa72 HD |
2155 | if (pte) |
2156 | pte_unmap(pte); | |
2157 | return 0; | |
1da177e4 LT |
2158 | } |
2159 | ||
2160 | static inline int unuse_pmd_range(struct vm_area_struct *vma, pud_t *pud, | |
2161 | unsigned long addr, unsigned long end, | |
10a9c496 | 2162 | unsigned int type) |
1da177e4 LT |
2163 | { |
2164 | pmd_t *pmd; | |
2165 | unsigned long next; | |
8a9f3ccd | 2166 | int ret; |
1da177e4 LT |
2167 | |
2168 | pmd = pmd_offset(pud, addr); | |
2169 | do { | |
dc644a07 | 2170 | cond_resched(); |
1da177e4 | 2171 | next = pmd_addr_end(addr, end); |
10a9c496 | 2172 | ret = unuse_pte_range(vma, pmd, addr, next, type); |
8a9f3ccd BS |
2173 | if (ret) |
2174 | return ret; | |
1da177e4 LT |
2175 | } while (pmd++, addr = next, addr != end); |
2176 | return 0; | |
2177 | } | |
2178 | ||
c2febafc | 2179 | static inline int unuse_pud_range(struct vm_area_struct *vma, p4d_t *p4d, |
1da177e4 | 2180 | unsigned long addr, unsigned long end, |
10a9c496 | 2181 | unsigned int type) |
1da177e4 LT |
2182 | { |
2183 | pud_t *pud; | |
2184 | unsigned long next; | |
8a9f3ccd | 2185 | int ret; |
1da177e4 | 2186 | |
c2febafc | 2187 | pud = pud_offset(p4d, addr); |
1da177e4 LT |
2188 | do { |
2189 | next = pud_addr_end(addr, end); | |
2190 | if (pud_none_or_clear_bad(pud)) | |
2191 | continue; | |
10a9c496 | 2192 | ret = unuse_pmd_range(vma, pud, addr, next, type); |
8a9f3ccd BS |
2193 | if (ret) |
2194 | return ret; | |
1da177e4 LT |
2195 | } while (pud++, addr = next, addr != end); |
2196 | return 0; | |
2197 | } | |
2198 | ||
c2febafc KS |
2199 | static inline int unuse_p4d_range(struct vm_area_struct *vma, pgd_t *pgd, |
2200 | unsigned long addr, unsigned long end, | |
10a9c496 | 2201 | unsigned int type) |
c2febafc KS |
2202 | { |
2203 | p4d_t *p4d; | |
2204 | unsigned long next; | |
2205 | int ret; | |
2206 | ||
2207 | p4d = p4d_offset(pgd, addr); | |
2208 | do { | |
2209 | next = p4d_addr_end(addr, end); | |
2210 | if (p4d_none_or_clear_bad(p4d)) | |
2211 | continue; | |
10a9c496 | 2212 | ret = unuse_pud_range(vma, p4d, addr, next, type); |
c2febafc KS |
2213 | if (ret) |
2214 | return ret; | |
2215 | } while (p4d++, addr = next, addr != end); | |
2216 | return 0; | |
2217 | } | |
2218 | ||
10a9c496 | 2219 | static int unuse_vma(struct vm_area_struct *vma, unsigned int type) |
1da177e4 LT |
2220 | { |
2221 | pgd_t *pgd; | |
2222 | unsigned long addr, end, next; | |
8a9f3ccd | 2223 | int ret; |
1da177e4 | 2224 | |
b56a2d8a VRP |
2225 | addr = vma->vm_start; |
2226 | end = vma->vm_end; | |
1da177e4 LT |
2227 | |
2228 | pgd = pgd_offset(vma->vm_mm, addr); | |
2229 | do { | |
2230 | next = pgd_addr_end(addr, end); | |
2231 | if (pgd_none_or_clear_bad(pgd)) | |
2232 | continue; | |
10a9c496 | 2233 | ret = unuse_p4d_range(vma, pgd, addr, next, type); |
8a9f3ccd BS |
2234 | if (ret) |
2235 | return ret; | |
1da177e4 LT |
2236 | } while (pgd++, addr = next, addr != end); |
2237 | return 0; | |
2238 | } | |
2239 | ||
10a9c496 | 2240 | static int unuse_mm(struct mm_struct *mm, unsigned int type) |
1da177e4 LT |
2241 | { |
2242 | struct vm_area_struct *vma; | |
8a9f3ccd | 2243 | int ret = 0; |
208c09db | 2244 | VMA_ITERATOR(vmi, mm, 0); |
1da177e4 | 2245 | |
d8ed45c5 | 2246 | mmap_read_lock(mm); |
208c09db | 2247 | for_each_vma(vmi, vma) { |
7528c4fb | 2248 | if (vma->anon_vma && !is_vm_hugetlb_page(vma)) { |
10a9c496 | 2249 | ret = unuse_vma(vma, type); |
b56a2d8a VRP |
2250 | if (ret) |
2251 | break; | |
2252 | } | |
208c09db | 2253 | |
dc644a07 | 2254 | cond_resched(); |
1da177e4 | 2255 | } |
d8ed45c5 | 2256 | mmap_read_unlock(mm); |
b56a2d8a | 2257 | return ret; |
1da177e4 LT |
2258 | } |
2259 | ||
2260 | /* | |
3c3115ad ML |
2261 | * Scan swap_map from current position to next entry still in use. |
2262 | * Return 0 if there are no inuse entries after prev till end of | |
2263 | * the map. | |
1da177e4 | 2264 | */ |
6eb396dc | 2265 | static unsigned int find_next_to_unuse(struct swap_info_struct *si, |
10a9c496 | 2266 | unsigned int prev) |
1da177e4 | 2267 | { |
b56a2d8a | 2268 | unsigned int i; |
8d69aaee | 2269 | unsigned char count; |
1da177e4 LT |
2270 | |
2271 | /* | |
5d337b91 | 2272 | * No need for swap_lock here: we're just looking |
1da177e4 LT |
2273 | * for whether an entry is in use, not modifying it; false |
2274 | * hits are okay, and sys_swapoff() has already prevented new | |
5d337b91 | 2275 | * allocations from this area (while holding swap_lock). |
1da177e4 | 2276 | */ |
b56a2d8a | 2277 | for (i = prev + 1; i < si->max; i++) { |
4db0c3c2 | 2278 | count = READ_ONCE(si->swap_map[i]); |
355cfa73 | 2279 | if (count && swap_count(count) != SWAP_MAP_BAD) |
10a9c496 | 2280 | break; |
dc644a07 HD |
2281 | if ((i % LATENCY_LIMIT) == 0) |
2282 | cond_resched(); | |
1da177e4 | 2283 | } |
b56a2d8a VRP |
2284 | |
2285 | if (i == si->max) | |
2286 | i = 0; | |
2287 | ||
1da177e4 LT |
2288 | return i; |
2289 | } | |
2290 | ||
10a9c496 | 2291 | static int try_to_unuse(unsigned int type) |
1da177e4 | 2292 | { |
b56a2d8a VRP |
2293 | struct mm_struct *prev_mm; |
2294 | struct mm_struct *mm; | |
2295 | struct list_head *p; | |
2296 | int retval = 0; | |
efa90a98 | 2297 | struct swap_info_struct *si = swap_info[type]; |
000085b9 | 2298 | struct folio *folio; |
1da177e4 | 2299 | swp_entry_t entry; |
b56a2d8a | 2300 | unsigned int i; |
1da177e4 | 2301 | |
b228386c | 2302 | if (!swap_usage_in_pages(si)) |
64cf264c | 2303 | goto success; |
1da177e4 | 2304 | |
b56a2d8a | 2305 | retry: |
10a9c496 | 2306 | retval = shmem_unuse(type); |
b56a2d8a | 2307 | if (retval) |
10a9c496 | 2308 | return retval; |
b56a2d8a VRP |
2309 | |
2310 | prev_mm = &init_mm; | |
2311 | mmget(prev_mm); | |
2312 | ||
2313 | spin_lock(&mmlist_lock); | |
2314 | p = &init_mm.mmlist; | |
b228386c | 2315 | while (swap_usage_in_pages(si) && |
64165b1a HD |
2316 | !signal_pending(current) && |
2317 | (p = p->next) != &init_mm.mmlist) { | |
1da177e4 | 2318 | |
b56a2d8a VRP |
2319 | mm = list_entry(p, struct mm_struct, mmlist); |
2320 | if (!mmget_not_zero(mm)) | |
2321 | continue; | |
2322 | spin_unlock(&mmlist_lock); | |
2323 | mmput(prev_mm); | |
2324 | prev_mm = mm; | |
10a9c496 | 2325 | retval = unuse_mm(mm, type); |
b56a2d8a VRP |
2326 | if (retval) { |
2327 | mmput(prev_mm); | |
10a9c496 | 2328 | return retval; |
1da177e4 LT |
2329 | } |
2330 | ||
2331 | /* | |
b56a2d8a VRP |
2332 | * Make sure that we aren't completely killing |
2333 | * interactive performance. | |
1da177e4 | 2334 | */ |
b56a2d8a VRP |
2335 | cond_resched(); |
2336 | spin_lock(&mmlist_lock); | |
2337 | } | |
2338 | spin_unlock(&mmlist_lock); | |
1da177e4 | 2339 | |
b56a2d8a | 2340 | mmput(prev_mm); |
1da177e4 | 2341 | |
b56a2d8a | 2342 | i = 0; |
b228386c | 2343 | while (swap_usage_in_pages(si) && |
64165b1a | 2344 | !signal_pending(current) && |
10a9c496 | 2345 | (i = find_next_to_unuse(si, i)) != 0) { |
1da177e4 | 2346 | |
b56a2d8a | 2347 | entry = swp_entry(type, i); |
7aad25b4 | 2348 | folio = filemap_get_folio(swap_address_space(entry), swap_cache_index(entry)); |
66dabbb6 | 2349 | if (IS_ERR(folio)) |
b56a2d8a | 2350 | continue; |
68bdc8d6 HD |
2351 | |
2352 | /* | |
000085b9 MWO |
2353 | * It is conceivable that a racing task removed this folio from |
2354 | * swap cache just before we acquired the page lock. The folio | |
b56a2d8a | 2355 | * might even be back in swap cache on another swap area. But |
000085b9 | 2356 | * that is okay, folio_free_swap() only removes stale folios. |
1da177e4 | 2357 | */ |
000085b9 MWO |
2358 | folio_lock(folio); |
2359 | folio_wait_writeback(folio); | |
2360 | folio_free_swap(folio); | |
2361 | folio_unlock(folio); | |
2362 | folio_put(folio); | |
1da177e4 LT |
2363 | } |
2364 | ||
b56a2d8a VRP |
2365 | /* |
2366 | * Lets check again to see if there are still swap entries in the map. | |
2367 | * If yes, we would need to do retry the unuse logic again. | |
2368 | * Under global memory pressure, swap entries can be reinserted back | |
2369 | * into process space after the mmlist loop above passes over them. | |
dd862deb | 2370 | * |
e2e3fdc7 MWO |
2371 | * Limit the number of retries? No: when mmget_not_zero() |
2372 | * above fails, that mm is likely to be freeing swap from | |
2373 | * exit_mmap(), which proceeds at its own independent pace; | |
84798514 | 2374 | * and even shmem_writeout() could have been preempted after |
e2e3fdc7 MWO |
2375 | * folio_alloc_swap(), temporarily hiding that swap. It's easy |
2376 | * and robust (though cpu-intensive) just to keep retrying. | |
b56a2d8a | 2377 | */ |
b228386c | 2378 | if (swap_usage_in_pages(si)) { |
64165b1a HD |
2379 | if (!signal_pending(current)) |
2380 | goto retry; | |
10a9c496 | 2381 | return -EINTR; |
64165b1a | 2382 | } |
10a9c496 | 2383 | |
64cf264c YA |
2384 | success: |
2385 | /* | |
2386 | * Make sure that further cleanups after try_to_unuse() returns happen | |
2387 | * after swap_range_free() reduces si->inuse_pages to 0. | |
2388 | */ | |
2389 | smp_mb(); | |
10a9c496 | 2390 | return 0; |
1da177e4 LT |
2391 | } |
2392 | ||
2393 | /* | |
5d337b91 HD |
2394 | * After a successful try_to_unuse, if no swap is now in use, we know |
2395 | * we can empty the mmlist. swap_lock must be held on entry and exit. | |
2396 | * Note that mmlist_lock nests inside swap_lock, and an mm must be | |
1da177e4 LT |
2397 | * added to the mmlist just after page_duplicate - before would be racy. |
2398 | */ | |
2399 | static void drain_mmlist(void) | |
2400 | { | |
2401 | struct list_head *p, *next; | |
efa90a98 | 2402 | unsigned int type; |
1da177e4 | 2403 | |
efa90a98 | 2404 | for (type = 0; type < nr_swapfiles; type++) |
b228386c | 2405 | if (swap_usage_in_pages(swap_info[type])) |
1da177e4 LT |
2406 | return; |
2407 | spin_lock(&mmlist_lock); | |
2408 | list_for_each_safe(p, next, &init_mm.mmlist) | |
2409 | list_del_init(p); | |
2410 | spin_unlock(&mmlist_lock); | |
2411 | } | |
2412 | ||
1da177e4 LT |
2413 | /* |
2414 | * Free all of a swapdev's extent information | |
2415 | */ | |
2416 | static void destroy_swap_extents(struct swap_info_struct *sis) | |
2417 | { | |
4efaceb1 AL |
2418 | while (!RB_EMPTY_ROOT(&sis->swap_extent_root)) { |
2419 | struct rb_node *rb = sis->swap_extent_root.rb_node; | |
2420 | struct swap_extent *se = rb_entry(rb, struct swap_extent, rb_node); | |
1da177e4 | 2421 | |
4efaceb1 | 2422 | rb_erase(rb, &sis->swap_extent_root); |
1da177e4 LT |
2423 | kfree(se); |
2424 | } | |
62c230bc | 2425 | |
bc4ae27d | 2426 | if (sis->flags & SWP_ACTIVATED) { |
62c230bc MG |
2427 | struct file *swap_file = sis->swap_file; |
2428 | struct address_space *mapping = swap_file->f_mapping; | |
2429 | ||
bc4ae27d OS |
2430 | sis->flags &= ~SWP_ACTIVATED; |
2431 | if (mapping->a_ops->swap_deactivate) | |
2432 | mapping->a_ops->swap_deactivate(swap_file); | |
62c230bc | 2433 | } |
1da177e4 LT |
2434 | } |
2435 | ||
2436 | /* | |
2437 | * Add a block range (and the corresponding page range) into this swapdev's | |
4efaceb1 | 2438 | * extent tree. |
1da177e4 | 2439 | * |
11d31886 | 2440 | * This function rather assumes that it is called in ascending page order. |
1da177e4 | 2441 | */ |
a509bc1a | 2442 | int |
1da177e4 LT |
2443 | add_swap_extent(struct swap_info_struct *sis, unsigned long start_page, |
2444 | unsigned long nr_pages, sector_t start_block) | |
2445 | { | |
4efaceb1 | 2446 | struct rb_node **link = &sis->swap_extent_root.rb_node, *parent = NULL; |
1da177e4 LT |
2447 | struct swap_extent *se; |
2448 | struct swap_extent *new_se; | |
4efaceb1 AL |
2449 | |
2450 | /* | |
2451 | * place the new node at the right most since the | |
2452 | * function is called in ascending page order. | |
2453 | */ | |
2454 | while (*link) { | |
2455 | parent = *link; | |
2456 | link = &parent->rb_right; | |
2457 | } | |
2458 | ||
2459 | if (parent) { | |
2460 | se = rb_entry(parent, struct swap_extent, rb_node); | |
11d31886 HD |
2461 | BUG_ON(se->start_page + se->nr_pages != start_page); |
2462 | if (se->start_block + se->nr_pages == start_block) { | |
1da177e4 LT |
2463 | /* Merge it */ |
2464 | se->nr_pages += nr_pages; | |
2465 | return 0; | |
2466 | } | |
1da177e4 LT |
2467 | } |
2468 | ||
4efaceb1 | 2469 | /* No merge, insert a new extent. */ |
1da177e4 LT |
2470 | new_se = kmalloc(sizeof(*se), GFP_KERNEL); |
2471 | if (new_se == NULL) | |
2472 | return -ENOMEM; | |
2473 | new_se->start_page = start_page; | |
2474 | new_se->nr_pages = nr_pages; | |
2475 | new_se->start_block = start_block; | |
2476 | ||
4efaceb1 AL |
2477 | rb_link_node(&new_se->rb_node, parent, link); |
2478 | rb_insert_color(&new_se->rb_node, &sis->swap_extent_root); | |
53092a74 | 2479 | return 1; |
1da177e4 | 2480 | } |
aa8aa8a3 | 2481 | EXPORT_SYMBOL_GPL(add_swap_extent); |
1da177e4 LT |
2482 | |
2483 | /* | |
2484 | * A `swap extent' is a simple thing which maps a contiguous range of pages | |
ff351f4b | 2485 | * onto a contiguous range of disk blocks. A rbtree of swap extents is |
c9bdf768 | 2486 | * built at swapon time and is then used at swap_writepage/swap_read_folio |
1da177e4 LT |
2487 | * time for locating where on disk a page belongs. |
2488 | * | |
2489 | * If the swapfile is an S_ISBLK block device, a single extent is installed. | |
2490 | * This is done so that the main operating code can treat S_ISBLK and S_ISREG | |
2491 | * swap files identically. | |
2492 | * | |
2493 | * Whether the swapdev is an S_ISREG file or an S_ISBLK blockdev, the swap | |
ff351f4b | 2494 | * extent rbtree operates in PAGE_SIZE disk blocks. Both S_ISREG and S_ISBLK |
1da177e4 LT |
2495 | * swapfiles are handled *identically* after swapon time. |
2496 | * | |
2497 | * For S_ISREG swapfiles, setup_swap_extents() will walk all the file's blocks | |
ff351f4b ML |
2498 | * and will parse them into a rbtree, in PAGE_SIZE chunks. If some stray |
2499 | * blocks are found which do not fall within the PAGE_SIZE alignment | |
1da177e4 LT |
2500 | * requirements, they are simply tossed out - we will never use those blocks |
2501 | * for swapping. | |
2502 | * | |
1638045c DW |
2503 | * For all swap devices we set S_SWAPFILE across the life of the swapon. This |
2504 | * prevents users from writing to the swap device, which will corrupt memory. | |
1da177e4 LT |
2505 | * |
2506 | * The amount of disk space which a single swap extent represents varies. | |
2507 | * Typically it is in the 1-4 megabyte range. So we can have hundreds of | |
ff351f4b | 2508 | * extents in the rbtree. - akpm. |
1da177e4 | 2509 | */ |
53092a74 | 2510 | static int setup_swap_extents(struct swap_info_struct *sis, sector_t *span) |
1da177e4 | 2511 | { |
62c230bc MG |
2512 | struct file *swap_file = sis->swap_file; |
2513 | struct address_space *mapping = swap_file->f_mapping; | |
2514 | struct inode *inode = mapping->host; | |
1da177e4 LT |
2515 | int ret; |
2516 | ||
1da177e4 LT |
2517 | if (S_ISBLK(inode->i_mode)) { |
2518 | ret = add_swap_extent(sis, 0, sis->max, 0); | |
53092a74 | 2519 | *span = sis->pages; |
a509bc1a | 2520 | return ret; |
1da177e4 LT |
2521 | } |
2522 | ||
62c230bc | 2523 | if (mapping->a_ops->swap_activate) { |
a509bc1a | 2524 | ret = mapping->a_ops->swap_activate(sis, swap_file, span); |
4b60c0ff N |
2525 | if (ret < 0) |
2526 | return ret; | |
2527 | sis->flags |= SWP_ACTIVATED; | |
e1209d3a N |
2528 | if ((sis->flags & SWP_FS_OPS) && |
2529 | sio_pool_init() != 0) { | |
2530 | destroy_swap_extents(sis); | |
2531 | return -ENOMEM; | |
62c230bc | 2532 | } |
a509bc1a | 2533 | return ret; |
62c230bc MG |
2534 | } |
2535 | ||
a509bc1a | 2536 | return generic_swapfile_activate(sis, swap_file, span); |
1da177e4 LT |
2537 | } |
2538 | ||
b85508d7 | 2539 | static int swap_node(struct swap_info_struct *si) |
a2468cc9 AL |
2540 | { |
2541 | struct block_device *bdev; | |
2542 | ||
b85508d7 BS |
2543 | if (si->bdev) |
2544 | bdev = si->bdev; | |
a2468cc9 | 2545 | else |
b85508d7 | 2546 | bdev = si->swap_file->f_inode->i_sb->s_bdev; |
a2468cc9 AL |
2547 | |
2548 | return bdev ? bdev->bd_disk->node_id : NUMA_NO_NODE; | |
2549 | } | |
2550 | ||
b85508d7 | 2551 | static void setup_swap_info(struct swap_info_struct *si, int prio, |
eb085574 | 2552 | unsigned char *swap_map, |
0ca0c24e UA |
2553 | struct swap_cluster_info *cluster_info, |
2554 | unsigned long *zeromap) | |
40531542 | 2555 | { |
a2468cc9 AL |
2556 | int i; |
2557 | ||
40531542 | 2558 | if (prio >= 0) |
b85508d7 | 2559 | si->prio = prio; |
40531542 | 2560 | else |
b85508d7 | 2561 | si->prio = --least_priority; |
18ab4d4c DS |
2562 | /* |
2563 | * the plist prio is negated because plist ordering is | |
2564 | * low-to-high, while swap ordering is high-to-low | |
2565 | */ | |
b85508d7 | 2566 | si->list.prio = -si->prio; |
a2468cc9 | 2567 | for_each_node(i) { |
b85508d7 BS |
2568 | if (si->prio >= 0) |
2569 | si->avail_lists[i].prio = -si->prio; | |
a2468cc9 | 2570 | else { |
b85508d7 BS |
2571 | if (swap_node(si) == i) |
2572 | si->avail_lists[i].prio = 1; | |
a2468cc9 | 2573 | else |
b85508d7 | 2574 | si->avail_lists[i].prio = -si->prio; |
a2468cc9 AL |
2575 | } |
2576 | } | |
b85508d7 BS |
2577 | si->swap_map = swap_map; |
2578 | si->cluster_info = cluster_info; | |
0ca0c24e | 2579 | si->zeromap = zeromap; |
eb085574 HY |
2580 | } |
2581 | ||
b85508d7 | 2582 | static void _enable_swap_info(struct swap_info_struct *si) |
eb085574 | 2583 | { |
b85508d7 BS |
2584 | atomic_long_add(si->pages, &nr_swap_pages); |
2585 | total_swap_pages += si->pages; | |
40531542 | 2586 | |
adfab836 | 2587 | assert_spin_locked(&swap_lock); |
adfab836 | 2588 | /* |
18ab4d4c DS |
2589 | * both lists are plists, and thus priority ordered. |
2590 | * swap_active_head needs to be priority ordered for swapoff(), | |
2591 | * which on removal of any swap_info_struct with an auto-assigned | |
2592 | * (i.e. negative) priority increments the auto-assigned priority | |
2593 | * of any lower-priority swap_info_structs. | |
e2e3fdc7 | 2594 | * swap_avail_head needs to be priority ordered for folio_alloc_swap(), |
18ab4d4c DS |
2595 | * which allocates swap pages from the highest available priority |
2596 | * swap_info_struct. | |
adfab836 | 2597 | */ |
b85508d7 | 2598 | plist_add(&si->list, &swap_active_head); |
c70699e5 | 2599 | |
b228386c KS |
2600 | /* Add back to available list */ |
2601 | add_to_avail_list(si, true); | |
cf0cac0a CEB |
2602 | } |
2603 | ||
b85508d7 | 2604 | static void enable_swap_info(struct swap_info_struct *si, int prio, |
cf0cac0a | 2605 | unsigned char *swap_map, |
0ca0c24e UA |
2606 | struct swap_cluster_info *cluster_info, |
2607 | unsigned long *zeromap) | |
cf0cac0a CEB |
2608 | { |
2609 | spin_lock(&swap_lock); | |
b85508d7 | 2610 | spin_lock(&si->lock); |
0ca0c24e | 2611 | setup_swap_info(si, prio, swap_map, cluster_info, zeromap); |
b85508d7 | 2612 | spin_unlock(&si->lock); |
eb085574 HY |
2613 | spin_unlock(&swap_lock); |
2614 | /* | |
63d8620e | 2615 | * Finished initializing swap device, now it's safe to reference it. |
eb085574 | 2616 | */ |
b85508d7 | 2617 | percpu_ref_resurrect(&si->users); |
eb085574 | 2618 | spin_lock(&swap_lock); |
b85508d7 BS |
2619 | spin_lock(&si->lock); |
2620 | _enable_swap_info(si); | |
2621 | spin_unlock(&si->lock); | |
cf0cac0a CEB |
2622 | spin_unlock(&swap_lock); |
2623 | } | |
2624 | ||
b85508d7 | 2625 | static void reinsert_swap_info(struct swap_info_struct *si) |
cf0cac0a CEB |
2626 | { |
2627 | spin_lock(&swap_lock); | |
b85508d7 | 2628 | spin_lock(&si->lock); |
0ca0c24e | 2629 | setup_swap_info(si, si->prio, si->swap_map, si->cluster_info, si->zeromap); |
b85508d7 BS |
2630 | _enable_swap_info(si); |
2631 | spin_unlock(&si->lock); | |
40531542 CEB |
2632 | spin_unlock(&swap_lock); |
2633 | } | |
2634 | ||
9a0ddeb7 KS |
2635 | /* |
2636 | * Called after clearing SWP_WRITEOK, ensures cluster_alloc_range | |
2637 | * see the updated flags, so there will be no more allocations. | |
2638 | */ | |
2639 | static void wait_for_allocation(struct swap_info_struct *si) | |
2640 | { | |
2641 | unsigned long offset; | |
2642 | unsigned long end = ALIGN(si->max, SWAPFILE_CLUSTER); | |
2643 | struct swap_cluster_info *ci; | |
2644 | ||
2645 | BUG_ON(si->flags & SWP_WRITEOK); | |
2646 | ||
2647 | for (offset = 0; offset < end; offset += SWAPFILE_CLUSTER) { | |
2648 | ci = lock_cluster(si, offset); | |
2649 | unlock_cluster(ci); | |
9a0ddeb7 KS |
2650 | } |
2651 | } | |
2652 | ||
1b7e9002 KS |
2653 | /* |
2654 | * Called after swap device's reference count is dead, so | |
2655 | * neither scan nor allocation will use it. | |
2656 | */ | |
2657 | static void flush_percpu_swap_cluster(struct swap_info_struct *si) | |
2658 | { | |
2659 | int cpu, i; | |
2660 | struct swap_info_struct **pcp_si; | |
2661 | ||
2662 | for_each_possible_cpu(cpu) { | |
2663 | pcp_si = per_cpu_ptr(percpu_swap_cluster.si, cpu); | |
2664 | /* | |
2665 | * Invalidate the percpu swap cluster cache, si->users | |
2666 | * is dead, so no new user will point to it, just flush | |
2667 | * any existing user. | |
2668 | */ | |
2669 | for (i = 0; i < SWAP_NR_ORDERS; i++) | |
2670 | cmpxchg(&pcp_si[i], si, NULL); | |
2671 | } | |
2672 | } | |
2673 | ||
2674 | ||
c4ea37c2 | 2675 | SYSCALL_DEFINE1(swapoff, const char __user *, specialfile) |
1da177e4 | 2676 | { |
73c34b6a | 2677 | struct swap_info_struct *p = NULL; |
8d69aaee | 2678 | unsigned char *swap_map; |
0ca0c24e | 2679 | unsigned long *zeromap; |
2a8f9449 | 2680 | struct swap_cluster_info *cluster_info; |
1da177e4 LT |
2681 | struct file *swap_file, *victim; |
2682 | struct address_space *mapping; | |
2683 | struct inode *inode; | |
91a27b2a | 2684 | struct filename *pathname; |
adfab836 | 2685 | int err, found = 0; |
886bb7e9 | 2686 | |
1da177e4 LT |
2687 | if (!capable(CAP_SYS_ADMIN)) |
2688 | return -EPERM; | |
2689 | ||
191c5424 AV |
2690 | BUG_ON(!current->mm); |
2691 | ||
1da177e4 | 2692 | pathname = getname(specialfile); |
1da177e4 | 2693 | if (IS_ERR(pathname)) |
f58b59c1 | 2694 | return PTR_ERR(pathname); |
1da177e4 | 2695 | |
669abf4e | 2696 | victim = file_open_name(pathname, O_RDWR|O_LARGEFILE, 0); |
1da177e4 LT |
2697 | err = PTR_ERR(victim); |
2698 | if (IS_ERR(victim)) | |
2699 | goto out; | |
2700 | ||
2701 | mapping = victim->f_mapping; | |
5d337b91 | 2702 | spin_lock(&swap_lock); |
18ab4d4c | 2703 | plist_for_each_entry(p, &swap_active_head, list) { |
22c6f8fd | 2704 | if (p->flags & SWP_WRITEOK) { |
adfab836 DS |
2705 | if (p->swap_file->f_mapping == mapping) { |
2706 | found = 1; | |
1da177e4 | 2707 | break; |
adfab836 | 2708 | } |
1da177e4 | 2709 | } |
1da177e4 | 2710 | } |
adfab836 | 2711 | if (!found) { |
1da177e4 | 2712 | err = -EINVAL; |
5d337b91 | 2713 | spin_unlock(&swap_lock); |
1da177e4 LT |
2714 | goto out_dput; |
2715 | } | |
191c5424 | 2716 | if (!security_vm_enough_memory_mm(current->mm, p->pages)) |
1da177e4 LT |
2717 | vm_unacct_memory(p->pages); |
2718 | else { | |
2719 | err = -ENOMEM; | |
5d337b91 | 2720 | spin_unlock(&swap_lock); |
1da177e4 LT |
2721 | goto out_dput; |
2722 | } | |
ec8acf20 | 2723 | spin_lock(&p->lock); |
b228386c | 2724 | del_from_avail_list(p, true); |
78ecba08 | 2725 | if (p->prio < 0) { |
adfab836 | 2726 | struct swap_info_struct *si = p; |
a2468cc9 | 2727 | int nid; |
adfab836 | 2728 | |
18ab4d4c | 2729 | plist_for_each_entry_continue(si, &swap_active_head, list) { |
adfab836 | 2730 | si->prio++; |
18ab4d4c | 2731 | si->list.prio--; |
a2468cc9 AL |
2732 | for_each_node(nid) { |
2733 | if (si->avail_lists[nid].prio != 1) | |
2734 | si->avail_lists[nid].prio--; | |
2735 | } | |
adfab836 | 2736 | } |
78ecba08 HD |
2737 | least_priority++; |
2738 | } | |
18ab4d4c | 2739 | plist_del(&p->list, &swap_active_head); |
ec8acf20 | 2740 | atomic_long_sub(p->pages, &nr_swap_pages); |
1da177e4 | 2741 | total_swap_pages -= p->pages; |
ec8acf20 | 2742 | spin_unlock(&p->lock); |
5d337b91 | 2743 | spin_unlock(&swap_lock); |
fb4f88dc | 2744 | |
9a0ddeb7 KS |
2745 | wait_for_allocation(p); |
2746 | ||
e1e12d2f | 2747 | set_current_oom_origin(); |
10a9c496 | 2748 | err = try_to_unuse(p->type); |
e1e12d2f | 2749 | clear_current_oom_origin(); |
1da177e4 | 2750 | |
1da177e4 LT |
2751 | if (err) { |
2752 | /* re-insert swap space back into swap_list */ | |
cf0cac0a | 2753 | reinsert_swap_info(p); |
1da177e4 LT |
2754 | goto out_dput; |
2755 | } | |
52b7efdb | 2756 | |
eb085574 | 2757 | /* |
63d8620e | 2758 | * Wait for swap operations protected by get/put_swap_device() |
d4a34d7f HY |
2759 | * to complete. Because of synchronize_rcu() here, all swap |
2760 | * operations protected by RCU reader side lock (including any | |
2761 | * spinlock) will be waited too. This makes it easy to | |
2762 | * prevent folio_test_swapcache() and the following swap cache | |
2763 | * operations from racing with swapoff. | |
eb085574 | 2764 | */ |
63d8620e | 2765 | percpu_ref_kill(&p->users); |
eb085574 | 2766 | synchronize_rcu(); |
63d8620e | 2767 | wait_for_completion(&p->comp); |
eb085574 | 2768 | |
815c2c54 | 2769 | flush_work(&p->discard_work); |
5168a68e | 2770 | flush_work(&p->reclaim_work); |
1b7e9002 | 2771 | flush_percpu_swap_cluster(p); |
815c2c54 | 2772 | |
5d337b91 | 2773 | destroy_swap_extents(p); |
570a335b HD |
2774 | if (p->flags & SWP_CONTINUED) |
2775 | free_swap_count_continuations(p); | |
2776 | ||
10f0d2a5 | 2777 | if (!p->bdev || !bdev_nonrot(p->bdev)) |
81a0298b HY |
2778 | atomic_dec(&nr_rotate_swap); |
2779 | ||
fc0abb14 | 2780 | mutex_lock(&swapon_mutex); |
5d337b91 | 2781 | spin_lock(&swap_lock); |
ec8acf20 | 2782 | spin_lock(&p->lock); |
5d337b91 HD |
2783 | drain_mmlist(); |
2784 | ||
1da177e4 LT |
2785 | swap_file = p->swap_file; |
2786 | p->swap_file = NULL; | |
2787 | p->max = 0; | |
2788 | swap_map = p->swap_map; | |
2789 | p->swap_map = NULL; | |
0ca0c24e UA |
2790 | zeromap = p->zeromap; |
2791 | p->zeromap = NULL; | |
2a8f9449 SL |
2792 | cluster_info = p->cluster_info; |
2793 | p->cluster_info = NULL; | |
ec8acf20 | 2794 | spin_unlock(&p->lock); |
5d337b91 | 2795 | spin_unlock(&swap_lock); |
8a84802e | 2796 | arch_swap_invalidate_area(p->type); |
42c06a0e | 2797 | zswap_swapoff(p->type); |
fc0abb14 | 2798 | mutex_unlock(&swapon_mutex); |
bae8a4ef KS |
2799 | kfree(p->global_cluster); |
2800 | p->global_cluster = NULL; | |
1da177e4 | 2801 | vfree(swap_map); |
0ca0c24e | 2802 | kvfree(zeromap); |
54f180d3 | 2803 | kvfree(cluster_info); |
2de1a7e4 | 2804 | /* Destroy swap account information */ |
adfab836 | 2805 | swap_cgroup_swapoff(p->type); |
4b3ef9da | 2806 | exit_swap_address_space(p->type); |
27a7faa0 | 2807 | |
1da177e4 | 2808 | inode = mapping->host; |
1638045c DW |
2809 | |
2810 | inode_lock(inode); | |
2811 | inode->i_flags &= ~S_SWAPFILE; | |
2812 | inode_unlock(inode); | |
1da177e4 | 2813 | filp_close(swap_file, NULL); |
f893ab41 WY |
2814 | |
2815 | /* | |
2816 | * Clear the SWP_USED flag after all resources are freed so that swapon | |
2817 | * can reuse this swap_info in alloc_swap_info() safely. It is ok to | |
2818 | * not hold p->lock after we cleared its SWP_WRITEOK. | |
2819 | */ | |
2820 | spin_lock(&swap_lock); | |
2821 | p->flags = 0; | |
2822 | spin_unlock(&swap_lock); | |
2823 | ||
1da177e4 | 2824 | err = 0; |
66d7dd51 KS |
2825 | atomic_inc(&proc_poll_event); |
2826 | wake_up_interruptible(&proc_poll_wait); | |
1da177e4 LT |
2827 | |
2828 | out_dput: | |
2829 | filp_close(victim, NULL); | |
2830 | out: | |
f58b59c1 | 2831 | putname(pathname); |
1da177e4 LT |
2832 | return err; |
2833 | } | |
2834 | ||
2835 | #ifdef CONFIG_PROC_FS | |
9dd95748 | 2836 | static __poll_t swaps_poll(struct file *file, poll_table *wait) |
66d7dd51 | 2837 | { |
f1514638 | 2838 | struct seq_file *seq = file->private_data; |
66d7dd51 KS |
2839 | |
2840 | poll_wait(file, &proc_poll_wait, wait); | |
2841 | ||
f1514638 KS |
2842 | if (seq->poll_event != atomic_read(&proc_poll_event)) { |
2843 | seq->poll_event = atomic_read(&proc_poll_event); | |
a9a08845 | 2844 | return EPOLLIN | EPOLLRDNORM | EPOLLERR | EPOLLPRI; |
66d7dd51 KS |
2845 | } |
2846 | ||
a9a08845 | 2847 | return EPOLLIN | EPOLLRDNORM; |
66d7dd51 KS |
2848 | } |
2849 | ||
1da177e4 LT |
2850 | /* iterator */ |
2851 | static void *swap_start(struct seq_file *swap, loff_t *pos) | |
2852 | { | |
efa90a98 HD |
2853 | struct swap_info_struct *si; |
2854 | int type; | |
1da177e4 LT |
2855 | loff_t l = *pos; |
2856 | ||
fc0abb14 | 2857 | mutex_lock(&swapon_mutex); |
1da177e4 | 2858 | |
881e4aab SS |
2859 | if (!l) |
2860 | return SEQ_START_TOKEN; | |
2861 | ||
c10d38cc | 2862 | for (type = 0; (si = swap_type_to_swap_info(type)); type++) { |
efa90a98 | 2863 | if (!(si->flags & SWP_USED) || !si->swap_map) |
1da177e4 | 2864 | continue; |
881e4aab | 2865 | if (!--l) |
efa90a98 | 2866 | return si; |
1da177e4 LT |
2867 | } |
2868 | ||
2869 | return NULL; | |
2870 | } | |
2871 | ||
2872 | static void *swap_next(struct seq_file *swap, void *v, loff_t *pos) | |
2873 | { | |
efa90a98 HD |
2874 | struct swap_info_struct *si = v; |
2875 | int type; | |
1da177e4 | 2876 | |
881e4aab | 2877 | if (v == SEQ_START_TOKEN) |
efa90a98 HD |
2878 | type = 0; |
2879 | else | |
2880 | type = si->type + 1; | |
881e4aab | 2881 | |
10c8d69f | 2882 | ++(*pos); |
c10d38cc | 2883 | for (; (si = swap_type_to_swap_info(type)); type++) { |
efa90a98 | 2884 | if (!(si->flags & SWP_USED) || !si->swap_map) |
1da177e4 | 2885 | continue; |
efa90a98 | 2886 | return si; |
1da177e4 LT |
2887 | } |
2888 | ||
2889 | return NULL; | |
2890 | } | |
2891 | ||
2892 | static void swap_stop(struct seq_file *swap, void *v) | |
2893 | { | |
fc0abb14 | 2894 | mutex_unlock(&swapon_mutex); |
1da177e4 LT |
2895 | } |
2896 | ||
2897 | static int swap_show(struct seq_file *swap, void *v) | |
2898 | { | |
efa90a98 | 2899 | struct swap_info_struct *si = v; |
1da177e4 LT |
2900 | struct file *file; |
2901 | int len; | |
642929a2 | 2902 | unsigned long bytes, inuse; |
1da177e4 | 2903 | |
efa90a98 | 2904 | if (si == SEQ_START_TOKEN) { |
68d68ff6 | 2905 | seq_puts(swap, "Filename\t\t\t\tType\t\tSize\t\tUsed\t\tPriority\n"); |
881e4aab SS |
2906 | return 0; |
2907 | } | |
1da177e4 | 2908 | |
00cde042 | 2909 | bytes = K(si->pages); |
b228386c | 2910 | inuse = K(swap_usage_in_pages(si)); |
6f793940 | 2911 | |
efa90a98 | 2912 | file = si->swap_file; |
2726d566 | 2913 | len = seq_file_path(swap, file, " \t\n\\"); |
642929a2 | 2914 | seq_printf(swap, "%*s%s\t%lu\t%s%lu\t%s%d\n", |
886bb7e9 | 2915 | len < 40 ? 40 - len : 1, " ", |
496ad9aa | 2916 | S_ISBLK(file_inode(file)->i_mode) ? |
1da177e4 | 2917 | "partition" : "file\t", |
6f793940 RD |
2918 | bytes, bytes < 10000000 ? "\t" : "", |
2919 | inuse, inuse < 10000000 ? "\t" : "", | |
efa90a98 | 2920 | si->prio); |
1da177e4 LT |
2921 | return 0; |
2922 | } | |
2923 | ||
15ad7cdc | 2924 | static const struct seq_operations swaps_op = { |
1da177e4 LT |
2925 | .start = swap_start, |
2926 | .next = swap_next, | |
2927 | .stop = swap_stop, | |
2928 | .show = swap_show | |
2929 | }; | |
2930 | ||
2931 | static int swaps_open(struct inode *inode, struct file *file) | |
2932 | { | |
f1514638 | 2933 | struct seq_file *seq; |
66d7dd51 KS |
2934 | int ret; |
2935 | ||
66d7dd51 | 2936 | ret = seq_open(file, &swaps_op); |
f1514638 | 2937 | if (ret) |
66d7dd51 | 2938 | return ret; |
66d7dd51 | 2939 | |
f1514638 KS |
2940 | seq = file->private_data; |
2941 | seq->poll_event = atomic_read(&proc_poll_event); | |
2942 | return 0; | |
1da177e4 LT |
2943 | } |
2944 | ||
97a32539 | 2945 | static const struct proc_ops swaps_proc_ops = { |
d919b33d | 2946 | .proc_flags = PROC_ENTRY_PERMANENT, |
97a32539 AD |
2947 | .proc_open = swaps_open, |
2948 | .proc_read = seq_read, | |
2949 | .proc_lseek = seq_lseek, | |
2950 | .proc_release = seq_release, | |
2951 | .proc_poll = swaps_poll, | |
1da177e4 LT |
2952 | }; |
2953 | ||
2954 | static int __init procswaps_init(void) | |
2955 | { | |
97a32539 | 2956 | proc_create("swaps", 0, NULL, &swaps_proc_ops); |
1da177e4 LT |
2957 | return 0; |
2958 | } | |
2959 | __initcall(procswaps_init); | |
2960 | #endif /* CONFIG_PROC_FS */ | |
2961 | ||
1796316a JB |
2962 | #ifdef MAX_SWAPFILES_CHECK |
2963 | static int __init max_swapfiles_check(void) | |
2964 | { | |
2965 | MAX_SWAPFILES_CHECK(); | |
2966 | return 0; | |
2967 | } | |
2968 | late_initcall(max_swapfiles_check); | |
2969 | #endif | |
2970 | ||
53cbb243 | 2971 | static struct swap_info_struct *alloc_swap_info(void) |
1da177e4 | 2972 | { |
73c34b6a | 2973 | struct swap_info_struct *p; |
b11a76b3 | 2974 | struct swap_info_struct *defer = NULL; |
1da177e4 | 2975 | unsigned int type; |
a2468cc9 | 2976 | int i; |
efa90a98 | 2977 | |
96008744 | 2978 | p = kvzalloc(struct_size(p, avail_lists, nr_node_ids), GFP_KERNEL); |
efa90a98 | 2979 | if (!p) |
53cbb243 | 2980 | return ERR_PTR(-ENOMEM); |
efa90a98 | 2981 | |
63d8620e ML |
2982 | if (percpu_ref_init(&p->users, swap_users_ref_free, |
2983 | PERCPU_REF_INIT_DEAD, GFP_KERNEL)) { | |
2984 | kvfree(p); | |
2985 | return ERR_PTR(-ENOMEM); | |
2986 | } | |
2987 | ||
5d337b91 | 2988 | spin_lock(&swap_lock); |
efa90a98 HD |
2989 | for (type = 0; type < nr_swapfiles; type++) { |
2990 | if (!(swap_info[type]->flags & SWP_USED)) | |
1da177e4 | 2991 | break; |
efa90a98 | 2992 | } |
0697212a | 2993 | if (type >= MAX_SWAPFILES) { |
5d337b91 | 2994 | spin_unlock(&swap_lock); |
63d8620e | 2995 | percpu_ref_exit(&p->users); |
873d7bcf | 2996 | kvfree(p); |
730c0581 | 2997 | return ERR_PTR(-EPERM); |
1da177e4 | 2998 | } |
efa90a98 HD |
2999 | if (type >= nr_swapfiles) { |
3000 | p->type = type; | |
efa90a98 | 3001 | /* |
a4b45114 HY |
3002 | * Publish the swap_info_struct after initializing it. |
3003 | * Note that kvzalloc() above zeroes all its fields. | |
efa90a98 | 3004 | */ |
a4b45114 HY |
3005 | smp_store_release(&swap_info[type], p); /* rcu_assign_pointer() */ |
3006 | nr_swapfiles++; | |
efa90a98 | 3007 | } else { |
b11a76b3 | 3008 | defer = p; |
efa90a98 HD |
3009 | p = swap_info[type]; |
3010 | /* | |
3011 | * Do not memset this entry: a racing procfs swap_next() | |
3012 | * would be relying on p->type to remain valid. | |
3013 | */ | |
3014 | } | |
4efaceb1 | 3015 | p->swap_extent_root = RB_ROOT; |
18ab4d4c | 3016 | plist_node_init(&p->list, 0); |
a2468cc9 AL |
3017 | for_each_node(i) |
3018 | plist_node_init(&p->avail_lists[i], 0); | |
1da177e4 | 3019 | p->flags = SWP_USED; |
5d337b91 | 3020 | spin_unlock(&swap_lock); |
63d8620e ML |
3021 | if (defer) { |
3022 | percpu_ref_exit(&defer->users); | |
3023 | kvfree(defer); | |
3024 | } | |
ec8acf20 | 3025 | spin_lock_init(&p->lock); |
2628bd6f | 3026 | spin_lock_init(&p->cont_lock); |
b228386c | 3027 | atomic_long_set(&p->inuse_pages, SWAP_USAGE_OFFLIST_BIT); |
63d8620e | 3028 | init_completion(&p->comp); |
efa90a98 | 3029 | |
53cbb243 | 3030 | return p; |
53cbb243 CEB |
3031 | } |
3032 | ||
b85508d7 | 3033 | static int claim_swapfile(struct swap_info_struct *si, struct inode *inode) |
4d0e1e10 | 3034 | { |
4d0e1e10 | 3035 | if (S_ISBLK(inode->i_mode)) { |
b85508d7 | 3036 | si->bdev = I_BDEV(inode); |
12d2966d NA |
3037 | /* |
3038 | * Zoned block devices contain zones that have a sequential | |
3039 | * write only restriction. Hence zoned block devices are not | |
3040 | * suitable for swapping. Disallow them here. | |
3041 | */ | |
b85508d7 | 3042 | if (bdev_is_zoned(si->bdev)) |
12d2966d | 3043 | return -EINVAL; |
b85508d7 | 3044 | si->flags |= SWP_BLKDEV; |
4d0e1e10 | 3045 | } else if (S_ISREG(inode->i_mode)) { |
b85508d7 | 3046 | si->bdev = inode->i_sb->s_bdev; |
1638045c DW |
3047 | } |
3048 | ||
4d0e1e10 | 3049 | return 0; |
4d0e1e10 CEB |
3050 | } |
3051 | ||
377eeaa8 AK |
3052 | |
3053 | /* | |
3054 | * Find out how many pages are allowed for a single swap device. There | |
3055 | * are two limiting factors: | |
3056 | * 1) the number of bits for the swap offset in the swp_entry_t type, and | |
3057 | * 2) the number of bits in the swap pte, as defined by the different | |
3058 | * architectures. | |
3059 | * | |
3060 | * In order to find the largest possible bit mask, a swap entry with | |
3061 | * swap type 0 and swap offset ~0UL is created, encoded to a swap pte, | |
3062 | * decoded to a swp_entry_t again, and finally the swap offset is | |
3063 | * extracted. | |
3064 | * | |
3065 | * This will mask all the bits from the initial ~0UL mask that can't | |
3066 | * be encoded in either the swp_entry_t or the architecture definition | |
3067 | * of a swap pte. | |
3068 | */ | |
3069 | unsigned long generic_max_swapfile_size(void) | |
3070 | { | |
3071 | return swp_offset(pte_to_swp_entry( | |
3072 | swp_entry_to_pte(swp_entry(0, ~0UL)))) + 1; | |
3073 | } | |
3074 | ||
3075 | /* Can be overridden by an architecture for additional checks. */ | |
be45a490 | 3076 | __weak unsigned long arch_max_swapfile_size(void) |
377eeaa8 AK |
3077 | { |
3078 | return generic_max_swapfile_size(); | |
3079 | } | |
3080 | ||
b85508d7 | 3081 | static unsigned long read_swap_header(struct swap_info_struct *si, |
ca8bd38b CEB |
3082 | union swap_header *swap_header, |
3083 | struct inode *inode) | |
3084 | { | |
3085 | int i; | |
3086 | unsigned long maxpages; | |
3087 | unsigned long swapfilepages; | |
d6bbbd29 | 3088 | unsigned long last_page; |
ca8bd38b CEB |
3089 | |
3090 | if (memcmp("SWAPSPACE2", swap_header->magic.magic, 10)) { | |
465c47fd | 3091 | pr_err("Unable to find swap-space signature\n"); |
38719025 | 3092 | return 0; |
ca8bd38b CEB |
3093 | } |
3094 | ||
041711ce | 3095 | /* swap partition endianness hack... */ |
ca8bd38b CEB |
3096 | if (swab32(swap_header->info.version) == 1) { |
3097 | swab32s(&swap_header->info.version); | |
3098 | swab32s(&swap_header->info.last_page); | |
3099 | swab32s(&swap_header->info.nr_badpages); | |
dd111be6 JH |
3100 | if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES) |
3101 | return 0; | |
ca8bd38b CEB |
3102 | for (i = 0; i < swap_header->info.nr_badpages; i++) |
3103 | swab32s(&swap_header->info.badpages[i]); | |
3104 | } | |
3105 | /* Check the swap header's sub-version */ | |
3106 | if (swap_header->info.version != 1) { | |
465c47fd AM |
3107 | pr_warn("Unable to handle swap header version %d\n", |
3108 | swap_header->info.version); | |
38719025 | 3109 | return 0; |
ca8bd38b CEB |
3110 | } |
3111 | ||
be45a490 | 3112 | maxpages = swapfile_maximum_size; |
d6bbbd29 | 3113 | last_page = swap_header->info.last_page; |
a06ad633 TA |
3114 | if (!last_page) { |
3115 | pr_warn("Empty swap-file\n"); | |
3116 | return 0; | |
3117 | } | |
d6bbbd29 | 3118 | if (last_page > maxpages) { |
465c47fd | 3119 | pr_warn("Truncating oversized swap area, only using %luk out of %luk\n", |
00cde042 | 3120 | K(maxpages), K(last_page)); |
d6bbbd29 RJ |
3121 | } |
3122 | if (maxpages > last_page) { | |
3123 | maxpages = last_page + 1; | |
ca8bd38b CEB |
3124 | /* p->max is an unsigned int: don't overflow it */ |
3125 | if ((unsigned int)maxpages == 0) | |
3126 | maxpages = UINT_MAX; | |
3127 | } | |
ca8bd38b CEB |
3128 | |
3129 | if (!maxpages) | |
38719025 | 3130 | return 0; |
ca8bd38b CEB |
3131 | swapfilepages = i_size_read(inode) >> PAGE_SHIFT; |
3132 | if (swapfilepages && maxpages > swapfilepages) { | |
465c47fd | 3133 | pr_warn("Swap area shorter than signature indicates\n"); |
38719025 | 3134 | return 0; |
ca8bd38b CEB |
3135 | } |
3136 | if (swap_header->info.nr_badpages && S_ISREG(inode->i_mode)) | |
38719025 | 3137 | return 0; |
ca8bd38b | 3138 | if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES) |
38719025 | 3139 | return 0; |
ca8bd38b CEB |
3140 | |
3141 | return maxpages; | |
ca8bd38b CEB |
3142 | } |
3143 | ||
b85508d7 | 3144 | static int setup_swap_map_and_extents(struct swap_info_struct *si, |
915d4d7b CEB |
3145 | union swap_header *swap_header, |
3146 | unsigned char *swap_map, | |
3147 | unsigned long maxpages, | |
3148 | sector_t *span) | |
3149 | { | |
915d4d7b | 3150 | unsigned int nr_good_pages; |
b843786b | 3151 | unsigned long i; |
915d4d7b CEB |
3152 | int nr_extents; |
3153 | ||
3154 | nr_good_pages = maxpages - 1; /* omit header page */ | |
3155 | ||
3156 | for (i = 0; i < swap_header->info.nr_badpages; i++) { | |
3157 | unsigned int page_nr = swap_header->info.badpages[i]; | |
bdb8e3f6 CEB |
3158 | if (page_nr == 0 || page_nr > swap_header->info.last_page) |
3159 | return -EINVAL; | |
915d4d7b CEB |
3160 | if (page_nr < maxpages) { |
3161 | swap_map[page_nr] = SWAP_MAP_BAD; | |
3162 | nr_good_pages--; | |
3163 | } | |
3164 | } | |
3165 | ||
3166 | if (nr_good_pages) { | |
3167 | swap_map[0] = SWAP_MAP_BAD; | |
b85508d7 BS |
3168 | si->max = maxpages; |
3169 | si->pages = nr_good_pages; | |
3170 | nr_extents = setup_swap_extents(si, span); | |
bdb8e3f6 CEB |
3171 | if (nr_extents < 0) |
3172 | return nr_extents; | |
b85508d7 | 3173 | nr_good_pages = si->pages; |
915d4d7b CEB |
3174 | } |
3175 | if (!nr_good_pages) { | |
465c47fd | 3176 | pr_warn("Empty swap-file\n"); |
bdb8e3f6 | 3177 | return -EINVAL; |
915d4d7b CEB |
3178 | } |
3179 | ||
b843786b JW |
3180 | return nr_extents; |
3181 | } | |
3182 | ||
a46a6bc2 BH |
3183 | #define SWAP_CLUSTER_INFO_COLS \ |
3184 | DIV_ROUND_UP(L1_CACHE_BYTES, sizeof(struct swap_cluster_info)) | |
3185 | #define SWAP_CLUSTER_SPACE_COLS \ | |
3186 | DIV_ROUND_UP(SWAP_ADDRESS_SPACE_PAGES, SWAPFILE_CLUSTER) | |
3187 | #define SWAP_CLUSTER_COLS \ | |
3188 | max_t(unsigned int, SWAP_CLUSTER_INFO_COLS, SWAP_CLUSTER_SPACE_COLS) | |
3189 | ||
b843786b JW |
3190 | static struct swap_cluster_info *setup_clusters(struct swap_info_struct *si, |
3191 | union swap_header *swap_header, | |
3192 | unsigned long maxpages) | |
3193 | { | |
3194 | unsigned long nr_clusters = DIV_ROUND_UP(maxpages, SWAPFILE_CLUSTER); | |
b843786b | 3195 | struct swap_cluster_info *cluster_info; |
a46a6bc2 | 3196 | unsigned long i, j, idx; |
1b7e9002 | 3197 | int err = -ENOMEM; |
b843786b JW |
3198 | |
3199 | cluster_info = kvcalloc(nr_clusters, sizeof(*cluster_info), GFP_KERNEL); | |
2a8f9449 | 3200 | if (!cluster_info) |
b843786b JW |
3201 | goto err; |
3202 | ||
3203 | for (i = 0; i < nr_clusters; i++) | |
3204 | spin_lock_init(&cluster_info[i].lock); | |
3205 | ||
1b7e9002 | 3206 | if (!(si->flags & SWP_SOLIDSTATE)) { |
bae8a4ef KS |
3207 | si->global_cluster = kmalloc(sizeof(*si->global_cluster), |
3208 | GFP_KERNEL); | |
3209 | if (!si->global_cluster) | |
3210 | goto err_free; | |
b843786b | 3211 | for (i = 0; i < SWAP_NR_ORDERS; i++) |
bae8a4ef KS |
3212 | si->global_cluster->next[i] = SWAP_ENTRY_INVALID; |
3213 | spin_lock_init(&si->global_cluster_lock); | |
b843786b JW |
3214 | } |
3215 | ||
3216 | /* | |
3217 | * Mark unusable pages as unavailable. The clusters aren't | |
3218 | * marked free yet, so no list operations are involved yet. | |
3219 | * | |
3220 | * See setup_swap_map_and_extents(): header page, bad pages, | |
3221 | * and the EOF part of the last cluster. | |
3222 | */ | |
3223 | inc_cluster_info_page(si, cluster_info, 0); | |
3224 | for (i = 0; i < swap_header->info.nr_badpages; i++) | |
3225 | inc_cluster_info_page(si, cluster_info, | |
3226 | swap_header->info.badpages[i]); | |
3227 | for (i = maxpages; i < round_up(maxpages, SWAPFILE_CLUSTER); i++) | |
3228 | inc_cluster_info_page(si, cluster_info, i); | |
3229 | ||
3230 | INIT_LIST_HEAD(&si->free_clusters); | |
3231 | INIT_LIST_HEAD(&si->full_clusters); | |
3232 | INIT_LIST_HEAD(&si->discard_clusters); | |
3233 | ||
3234 | for (i = 0; i < SWAP_NR_ORDERS; i++) { | |
3235 | INIT_LIST_HEAD(&si->nonfull_clusters[i]); | |
3236 | INIT_LIST_HEAD(&si->frag_clusters[i]); | |
3b644773 | 3237 | atomic_long_set(&si->frag_cluster_nr[i], 0); |
b843786b | 3238 | } |
2a8f9449 | 3239 | |
4b3ef9da HY |
3240 | /* |
3241 | * Reduce false cache line sharing between cluster_info and | |
3242 | * sharing same address space. | |
3243 | */ | |
a46a6bc2 | 3244 | for (j = 0; j < SWAP_CLUSTER_COLS; j++) { |
235b6217 | 3245 | for (i = 0; i < DIV_ROUND_UP(nr_clusters, SWAP_CLUSTER_COLS); i++) { |
73ed0baa | 3246 | struct swap_cluster_info *ci; |
235b6217 | 3247 | idx = i * SWAP_CLUSTER_COLS + j; |
73ed0baa | 3248 | ci = cluster_info + idx; |
235b6217 HY |
3249 | if (idx >= nr_clusters) |
3250 | continue; | |
5f843a9a CL |
3251 | if (ci->count) { |
3252 | ci->flags = CLUSTER_FLAG_NONFULL; | |
b85508d7 | 3253 | list_add_tail(&ci->list, &si->nonfull_clusters[0]); |
235b6217 | 3254 | continue; |
5f843a9a | 3255 | } |
73ed0baa | 3256 | ci->flags = CLUSTER_FLAG_FREE; |
b85508d7 | 3257 | list_add_tail(&ci->list, &si->free_clusters); |
2a8f9449 | 3258 | } |
2a8f9449 | 3259 | } |
b843786b JW |
3260 | |
3261 | return cluster_info; | |
3262 | ||
3263 | err_free: | |
3264 | kvfree(cluster_info); | |
3265 | err: | |
3266 | return ERR_PTR(err); | |
915d4d7b CEB |
3267 | } |
3268 | ||
53cbb243 CEB |
3269 | SYSCALL_DEFINE2(swapon, const char __user *, specialfile, int, swap_flags) |
3270 | { | |
b85508d7 | 3271 | struct swap_info_struct *si; |
91a27b2a | 3272 | struct filename *name; |
53cbb243 CEB |
3273 | struct file *swap_file = NULL; |
3274 | struct address_space *mapping; | |
51cc3a66 | 3275 | struct dentry *dentry; |
40531542 | 3276 | int prio; |
53cbb243 CEB |
3277 | int error; |
3278 | union swap_header *swap_header; | |
915d4d7b | 3279 | int nr_extents; |
53cbb243 CEB |
3280 | sector_t span; |
3281 | unsigned long maxpages; | |
53cbb243 | 3282 | unsigned char *swap_map = NULL; |
0ca0c24e | 3283 | unsigned long *zeromap = NULL; |
2a8f9449 | 3284 | struct swap_cluster_info *cluster_info = NULL; |
97b76796 | 3285 | struct folio *folio = NULL; |
53cbb243 | 3286 | struct inode *inode = NULL; |
7cbf3192 | 3287 | bool inced_nr_rotate_swap = false; |
53cbb243 | 3288 | |
d15cab97 HD |
3289 | if (swap_flags & ~SWAP_FLAGS_VALID) |
3290 | return -EINVAL; | |
3291 | ||
53cbb243 CEB |
3292 | if (!capable(CAP_SYS_ADMIN)) |
3293 | return -EPERM; | |
3294 | ||
a2468cc9 AL |
3295 | if (!swap_avail_heads) |
3296 | return -ENOMEM; | |
3297 | ||
b85508d7 BS |
3298 | si = alloc_swap_info(); |
3299 | if (IS_ERR(si)) | |
3300 | return PTR_ERR(si); | |
53cbb243 | 3301 | |
b85508d7 | 3302 | INIT_WORK(&si->discard_work, swap_discard_work); |
5168a68e | 3303 | INIT_WORK(&si->reclaim_work, swap_reclaim_work); |
815c2c54 | 3304 | |
1da177e4 | 3305 | name = getname(specialfile); |
1da177e4 | 3306 | if (IS_ERR(name)) { |
7de7fb6b | 3307 | error = PTR_ERR(name); |
1da177e4 | 3308 | name = NULL; |
bd69010b | 3309 | goto bad_swap; |
1da177e4 | 3310 | } |
51d908b3 | 3311 | swap_file = file_open_name(name, O_RDWR | O_LARGEFILE | O_EXCL, 0); |
1da177e4 | 3312 | if (IS_ERR(swap_file)) { |
7de7fb6b | 3313 | error = PTR_ERR(swap_file); |
1da177e4 | 3314 | swap_file = NULL; |
bd69010b | 3315 | goto bad_swap; |
1da177e4 LT |
3316 | } |
3317 | ||
b85508d7 | 3318 | si->swap_file = swap_file; |
1da177e4 | 3319 | mapping = swap_file->f_mapping; |
51cc3a66 | 3320 | dentry = swap_file->f_path.dentry; |
2130781e | 3321 | inode = mapping->host; |
6f179af8 | 3322 | |
b85508d7 | 3323 | error = claim_swapfile(si, inode); |
4d0e1e10 | 3324 | if (unlikely(error)) |
1da177e4 | 3325 | goto bad_swap; |
1da177e4 | 3326 | |
d795a90e | 3327 | inode_lock(inode); |
51cc3a66 HD |
3328 | if (d_unlinked(dentry) || cant_mount(dentry)) { |
3329 | error = -ENOENT; | |
3330 | goto bad_swap_unlock_inode; | |
3331 | } | |
d795a90e NA |
3332 | if (IS_SWAPFILE(inode)) { |
3333 | error = -EBUSY; | |
3334 | goto bad_swap_unlock_inode; | |
3335 | } | |
3336 | ||
6ba0982c LC |
3337 | /* |
3338 | * The swap subsystem needs a major overhaul to support this. | |
3339 | * It doesn't work yet so just disable it for now. | |
3340 | */ | |
3341 | if (mapping_min_folio_order(mapping) > 0) { | |
3342 | error = -EINVAL; | |
3343 | goto bad_swap_unlock_inode; | |
3344 | } | |
3345 | ||
1da177e4 LT |
3346 | /* |
3347 | * Read the swap header. | |
3348 | */ | |
7e0a1265 | 3349 | if (!mapping->a_ops->read_folio) { |
1da177e4 | 3350 | error = -EINVAL; |
d795a90e | 3351 | goto bad_swap_unlock_inode; |
1da177e4 | 3352 | } |
97b76796 MWO |
3353 | folio = read_mapping_folio(mapping, 0, swap_file); |
3354 | if (IS_ERR(folio)) { | |
3355 | error = PTR_ERR(folio); | |
d795a90e | 3356 | goto bad_swap_unlock_inode; |
1da177e4 | 3357 | } |
97b76796 | 3358 | swap_header = kmap_local_folio(folio, 0); |
1da177e4 | 3359 | |
b85508d7 | 3360 | maxpages = read_swap_header(si, swap_header, inode); |
ca8bd38b | 3361 | if (unlikely(!maxpages)) { |
1da177e4 | 3362 | error = -EINVAL; |
d795a90e | 3363 | goto bad_swap_unlock_inode; |
1da177e4 | 3364 | } |
886bb7e9 | 3365 | |
81e33971 | 3366 | /* OK, set up the swap map and apply the bad block list */ |
803d0c83 | 3367 | swap_map = vzalloc(maxpages); |
81e33971 HD |
3368 | if (!swap_map) { |
3369 | error = -ENOMEM; | |
d795a90e | 3370 | goto bad_swap_unlock_inode; |
81e33971 | 3371 | } |
f0571429 | 3372 | |
b843786b JW |
3373 | error = swap_cgroup_swapon(si->type, maxpages); |
3374 | if (error) | |
3375 | goto bad_swap_unlock_inode; | |
3376 | ||
3377 | nr_extents = setup_swap_map_and_extents(si, swap_header, swap_map, | |
3378 | maxpages, &span); | |
3379 | if (unlikely(nr_extents < 0)) { | |
3380 | error = nr_extents; | |
3381 | goto bad_swap_unlock_inode; | |
3382 | } | |
3383 | ||
0ca0c24e UA |
3384 | /* |
3385 | * Use kvmalloc_array instead of bitmap_zalloc as the allocation order might | |
3386 | * be above MAX_PAGE_ORDER incase of a large swap file. | |
3387 | */ | |
3388 | zeromap = kvmalloc_array(BITS_TO_LONGS(maxpages), sizeof(long), | |
3389 | GFP_KERNEL | __GFP_ZERO); | |
3390 | if (!zeromap) { | |
3391 | error = -ENOMEM; | |
3392 | goto bad_swap_unlock_inode; | |
3393 | } | |
3394 | ||
b85508d7 BS |
3395 | if (si->bdev && bdev_stable_writes(si->bdev)) |
3396 | si->flags |= SWP_STABLE_WRITES; | |
f0571429 | 3397 | |
b85508d7 BS |
3398 | if (si->bdev && bdev_synchronous(si->bdev)) |
3399 | si->flags |= SWP_SYNCHRONOUS_IO; | |
539a6fea | 3400 | |
b85508d7 | 3401 | if (si->bdev && bdev_nonrot(si->bdev)) { |
b85508d7 | 3402 | si->flags |= SWP_SOLIDSTATE; |
7cbf3192 | 3403 | } else { |
81a0298b | 3404 | atomic_inc(&nr_rotate_swap); |
7cbf3192 OS |
3405 | inced_nr_rotate_swap = true; |
3406 | } | |
1da177e4 | 3407 | |
72774330 KS |
3408 | cluster_info = setup_clusters(si, swap_header, maxpages); |
3409 | if (IS_ERR(cluster_info)) { | |
3410 | error = PTR_ERR(cluster_info); | |
3411 | cluster_info = NULL; | |
3412 | goto bad_swap_unlock_inode; | |
3413 | } | |
3414 | ||
70200574 | 3415 | if ((swap_flags & SWAP_FLAG_DISCARD) && |
b85508d7 | 3416 | si->bdev && bdev_max_discard_sectors(si->bdev)) { |
2a8f9449 SL |
3417 | /* |
3418 | * When discard is enabled for swap with no particular | |
3419 | * policy flagged, we set all swap discard flags here in | |
3420 | * order to sustain backward compatibility with older | |
3421 | * swapon(8) releases. | |
3422 | */ | |
b85508d7 | 3423 | si->flags |= (SWP_DISCARDABLE | SWP_AREA_DISCARD | |
2a8f9449 | 3424 | SWP_PAGE_DISCARD); |
dcf6b7dd | 3425 | |
2a8f9449 SL |
3426 | /* |
3427 | * By flagging sys_swapon, a sysadmin can tell us to | |
3428 | * either do single-time area discards only, or to just | |
3429 | * perform discards for released swap page-clusters. | |
3430 | * Now it's time to adjust the p->flags accordingly. | |
3431 | */ | |
3432 | if (swap_flags & SWAP_FLAG_DISCARD_ONCE) | |
b85508d7 | 3433 | si->flags &= ~SWP_PAGE_DISCARD; |
2a8f9449 | 3434 | else if (swap_flags & SWAP_FLAG_DISCARD_PAGES) |
b85508d7 | 3435 | si->flags &= ~SWP_AREA_DISCARD; |
2a8f9449 SL |
3436 | |
3437 | /* issue a swapon-time discard if it's still required */ | |
b85508d7 BS |
3438 | if (si->flags & SWP_AREA_DISCARD) { |
3439 | int err = discard_swap(si); | |
2a8f9449 SL |
3440 | if (unlikely(err)) |
3441 | pr_err("swapon: discard_swap(%p): %d\n", | |
b85508d7 | 3442 | si, err); |
dcf6b7dd | 3443 | } |
20137a49 | 3444 | } |
6a6ba831 | 3445 | |
b85508d7 | 3446 | error = init_swap_address_space(si->type, maxpages); |
4b3ef9da | 3447 | if (error) |
d795a90e | 3448 | goto bad_swap_unlock_inode; |
4b3ef9da | 3449 | |
b85508d7 | 3450 | error = zswap_swapon(si->type, maxpages); |
bb29fd77 CZ |
3451 | if (error) |
3452 | goto free_swap_address_space; | |
3453 | ||
dc617f29 DW |
3454 | /* |
3455 | * Flush any pending IO and dirty mappings before we start using this | |
3456 | * swap device. | |
3457 | */ | |
3458 | inode->i_flags |= S_SWAPFILE; | |
3459 | error = inode_drain_writes(inode); | |
3460 | if (error) { | |
3461 | inode->i_flags &= ~S_SWAPFILE; | |
bb29fd77 | 3462 | goto free_swap_zswap; |
dc617f29 DW |
3463 | } |
3464 | ||
fc0abb14 | 3465 | mutex_lock(&swapon_mutex); |
40531542 | 3466 | prio = -1; |
78ecba08 | 3467 | if (swap_flags & SWAP_FLAG_PREFER) |
0eb7d2c3 | 3468 | prio = swap_flags & SWAP_FLAG_PRIO_MASK; |
0ca0c24e | 3469 | enable_swap_info(si, prio, swap_map, cluster_info, zeromap); |
c69dbfb8 | 3470 | |
42c06a0e | 3471 | pr_info("Adding %uk swap on %s. Priority:%d extents:%d across:%lluk %s%s%s%s\n", |
b85508d7 | 3472 | K(si->pages), name->name, si->prio, nr_extents, |
00cde042 | 3473 | K((unsigned long long)span), |
b85508d7 BS |
3474 | (si->flags & SWP_SOLIDSTATE) ? "SS" : "", |
3475 | (si->flags & SWP_DISCARDABLE) ? "D" : "", | |
3476 | (si->flags & SWP_AREA_DISCARD) ? "s" : "", | |
3477 | (si->flags & SWP_PAGE_DISCARD) ? "c" : ""); | |
c69dbfb8 | 3478 | |
fc0abb14 | 3479 | mutex_unlock(&swapon_mutex); |
66d7dd51 KS |
3480 | atomic_inc(&proc_poll_event); |
3481 | wake_up_interruptible(&proc_poll_wait); | |
3482 | ||
1da177e4 LT |
3483 | error = 0; |
3484 | goto out; | |
bb29fd77 | 3485 | free_swap_zswap: |
b85508d7 | 3486 | zswap_swapoff(si->type); |
822bca52 | 3487 | free_swap_address_space: |
b85508d7 | 3488 | exit_swap_address_space(si->type); |
d795a90e NA |
3489 | bad_swap_unlock_inode: |
3490 | inode_unlock(inode); | |
1da177e4 | 3491 | bad_swap: |
bae8a4ef KS |
3492 | kfree(si->global_cluster); |
3493 | si->global_cluster = NULL; | |
d795a90e | 3494 | inode = NULL; |
b85508d7 BS |
3495 | destroy_swap_extents(si); |
3496 | swap_cgroup_swapoff(si->type); | |
5d337b91 | 3497 | spin_lock(&swap_lock); |
b85508d7 BS |
3498 | si->swap_file = NULL; |
3499 | si->flags = 0; | |
5d337b91 | 3500 | spin_unlock(&swap_lock); |
1da177e4 | 3501 | vfree(swap_map); |
0ca0c24e | 3502 | kvfree(zeromap); |
8606a1a9 | 3503 | kvfree(cluster_info); |
7cbf3192 OS |
3504 | if (inced_nr_rotate_swap) |
3505 | atomic_dec(&nr_rotate_swap); | |
d795a90e | 3506 | if (swap_file) |
1da177e4 LT |
3507 | filp_close(swap_file, NULL); |
3508 | out: | |
97b76796 MWO |
3509 | if (!IS_ERR_OR_NULL(folio)) |
3510 | folio_release_kmap(folio, swap_header); | |
1da177e4 LT |
3511 | if (name) |
3512 | putname(name); | |
1638045c | 3513 | if (inode) |
5955102c | 3514 | inode_unlock(inode); |
1da177e4 LT |
3515 | return error; |
3516 | } | |
3517 | ||
3518 | void si_swapinfo(struct sysinfo *val) | |
3519 | { | |
efa90a98 | 3520 | unsigned int type; |
1da177e4 LT |
3521 | unsigned long nr_to_be_unused = 0; |
3522 | ||
5d337b91 | 3523 | spin_lock(&swap_lock); |
efa90a98 HD |
3524 | for (type = 0; type < nr_swapfiles; type++) { |
3525 | struct swap_info_struct *si = swap_info[type]; | |
3526 | ||
3527 | if ((si->flags & SWP_USED) && !(si->flags & SWP_WRITEOK)) | |
b228386c | 3528 | nr_to_be_unused += swap_usage_in_pages(si); |
1da177e4 | 3529 | } |
ec8acf20 | 3530 | val->freeswap = atomic_long_read(&nr_swap_pages) + nr_to_be_unused; |
1da177e4 | 3531 | val->totalswap = total_swap_pages + nr_to_be_unused; |
5d337b91 | 3532 | spin_unlock(&swap_lock); |
1da177e4 LT |
3533 | } |
3534 | ||
3535 | /* | |
9f101bef | 3536 | * Verify that nr swap entries are valid and increment their swap map counts. |
1da177e4 | 3537 | * |
355cfa73 KH |
3538 | * Returns error code in following case. |
3539 | * - success -> 0 | |
3540 | * - swp_entry is invalid -> EINVAL | |
355cfa73 KH |
3541 | * - swap-cache reference is requested but there is already one. -> EEXIST |
3542 | * - swap-cache reference is requested but the entry is not used. -> ENOENT | |
570a335b | 3543 | * - swap-mapped reference requested but needs continued swap count. -> ENOMEM |
1da177e4 | 3544 | */ |
9f101bef | 3545 | static int __swap_duplicate(swp_entry_t entry, unsigned char usage, int nr) |
1da177e4 | 3546 | { |
b85508d7 | 3547 | struct swap_info_struct *si; |
235b6217 | 3548 | struct swap_cluster_info *ci; |
c10d38cc | 3549 | unsigned long offset; |
8d69aaee HD |
3550 | unsigned char count; |
3551 | unsigned char has_cache; | |
9f101bef | 3552 | int err, i; |
1da177e4 | 3553 | |
b85508d7 | 3554 | si = swp_swap_info(entry); |
c3e99839 | 3555 | if (WARN_ON_ONCE(!si)) { |
3556 | pr_err("%s%08lx\n", Bad_file, entry.val); | |
3557 | return -EINVAL; | |
3558 | } | |
235b6217 | 3559 | |
eb085574 | 3560 | offset = swp_offset(entry); |
9f101bef BS |
3561 | VM_WARN_ON(nr > SWAPFILE_CLUSTER - offset % SWAPFILE_CLUSTER); |
3562 | VM_WARN_ON(usage == 1 && nr > 1); | |
0b310d9c | 3563 | ci = lock_cluster(si, offset); |
355cfa73 | 3564 | |
253d553b | 3565 | err = 0; |
9f101bef | 3566 | for (i = 0; i < nr; i++) { |
b85508d7 | 3567 | count = si->swap_map[offset + i]; |
355cfa73 | 3568 | |
9f101bef BS |
3569 | /* |
3570 | * swapin_readahead() doesn't check if a swap entry is valid, so the | |
3571 | * swap entry could be SWAP_MAP_BAD. Check here with lock held. | |
3572 | */ | |
3573 | if (unlikely(swap_count(count) == SWAP_MAP_BAD)) { | |
3574 | err = -ENOENT; | |
3575 | goto unlock_out; | |
3576 | } | |
355cfa73 | 3577 | |
9f101bef BS |
3578 | has_cache = count & SWAP_HAS_CACHE; |
3579 | count &= ~SWAP_HAS_CACHE; | |
3580 | ||
3581 | if (!count && !has_cache) { | |
253d553b | 3582 | err = -ENOENT; |
9f101bef BS |
3583 | } else if (usage == SWAP_HAS_CACHE) { |
3584 | if (has_cache) | |
3585 | err = -EEXIST; | |
3586 | } else if ((count & ~COUNT_CONTINUED) > SWAP_MAP_MAX) { | |
3587 | err = -EINVAL; | |
3588 | } | |
3589 | ||
3590 | if (err) | |
3591 | goto unlock_out; | |
3592 | } | |
355cfa73 | 3593 | |
9f101bef | 3594 | for (i = 0; i < nr; i++) { |
b85508d7 | 3595 | count = si->swap_map[offset + i]; |
9f101bef BS |
3596 | has_cache = count & SWAP_HAS_CACHE; |
3597 | count &= ~SWAP_HAS_CACHE; | |
253d553b | 3598 | |
9f101bef BS |
3599 | if (usage == SWAP_HAS_CACHE) |
3600 | has_cache = SWAP_HAS_CACHE; | |
3601 | else if ((count & ~COUNT_CONTINUED) < SWAP_MAP_MAX) | |
570a335b | 3602 | count += usage; |
b85508d7 | 3603 | else if (swap_count_continued(si, offset + i, count)) |
570a335b | 3604 | count = COUNT_CONTINUED; |
9f101bef BS |
3605 | else { |
3606 | /* | |
3607 | * Don't need to rollback changes, because if | |
3608 | * usage == 1, there must be nr == 1. | |
3609 | */ | |
570a335b | 3610 | err = -ENOMEM; |
9f101bef BS |
3611 | goto unlock_out; |
3612 | } | |
253d553b | 3613 | |
b85508d7 | 3614 | WRITE_ONCE(si->swap_map[offset + i], count | has_cache); |
9f101bef | 3615 | } |
253d553b | 3616 | |
355cfa73 | 3617 | unlock_out: |
0b310d9c | 3618 | unlock_cluster(ci); |
253d553b | 3619 | return err; |
1da177e4 | 3620 | } |
253d553b | 3621 | |
aaa46865 HD |
3622 | /* |
3623 | * Help swapoff by noting that swap entry belongs to shmem/tmpfs | |
3624 | * (in which case its reference count is never incremented). | |
3625 | */ | |
65018076 | 3626 | void swap_shmem_alloc(swp_entry_t entry, int nr) |
aaa46865 | 3627 | { |
65018076 | 3628 | __swap_duplicate(entry, SWAP_MAP_SHMEM, nr); |
aaa46865 HD |
3629 | } |
3630 | ||
355cfa73 | 3631 | /* |
08259d58 HD |
3632 | * Increase reference count of swap entry by 1. |
3633 | * Returns 0 for success, or -ENOMEM if a swap_count_continuation is required | |
3634 | * but could not be atomically allocated. Returns 0, just as if it succeeded, | |
3635 | * if __swap_duplicate() fails for another reason (-EINVAL or -ENOENT), which | |
3636 | * might occur if a page table entry has got corrupted. | |
355cfa73 | 3637 | */ |
570a335b | 3638 | int swap_duplicate(swp_entry_t entry) |
355cfa73 | 3639 | { |
570a335b HD |
3640 | int err = 0; |
3641 | ||
9f101bef | 3642 | while (!err && __swap_duplicate(entry, 1, 1) == -ENOMEM) |
570a335b HD |
3643 | err = add_swap_count_continuation(entry, GFP_ATOMIC); |
3644 | return err; | |
355cfa73 | 3645 | } |
1da177e4 | 3646 | |
cb4b86ba | 3647 | /* |
9f101bef | 3648 | * @entry: first swap entry from which we allocate nr swap cache. |
355cfa73 | 3649 | * |
9f101bef | 3650 | * Called when allocating swap cache for existing swap entries, |
355cfa73 | 3651 | * This can return error codes. Returns 0 at success. |
3eeba135 | 3652 | * -EEXIST means there is a swap cache. |
355cfa73 | 3653 | * Note: return code is different from swap_duplicate(). |
cb4b86ba | 3654 | */ |
9f101bef | 3655 | int swapcache_prepare(swp_entry_t entry, int nr) |
cb4b86ba | 3656 | { |
9f101bef | 3657 | return __swap_duplicate(entry, SWAP_HAS_CACHE, nr); |
cb4b86ba KH |
3658 | } |
3659 | ||
ec9827cd KS |
3660 | /* |
3661 | * Caller should ensure entries belong to the same folio so | |
3662 | * the entries won't span cross cluster boundary. | |
3663 | */ | |
9f101bef | 3664 | void swapcache_clear(struct swap_info_struct *si, swp_entry_t entry, int nr) |
13ddaf26 | 3665 | { |
ec9827cd | 3666 | swap_entries_put_cache(si, entry, nr); |
13ddaf26 KS |
3667 | } |
3668 | ||
0bcac06f MK |
3669 | struct swap_info_struct *swp_swap_info(swp_entry_t entry) |
3670 | { | |
c10d38cc | 3671 | return swap_type_to_swap_info(swp_type(entry)); |
0bcac06f MK |
3672 | } |
3673 | ||
570a335b HD |
3674 | /* |
3675 | * add_swap_count_continuation - called when a swap count is duplicated | |
3676 | * beyond SWAP_MAP_MAX, it allocates a new page and links that to the entry's | |
3677 | * page of the original vmalloc'ed swap_map, to hold the continuation count | |
3678 | * (for that entry and for its neighbouring PAGE_SIZE swap entries). Called | |
3679 | * again when count is duplicated beyond SWAP_MAP_MAX * SWAP_CONT_MAX, etc. | |
3680 | * | |
3681 | * These continuation pages are seldom referenced: the common paths all work | |
3682 | * on the original swap_map, only referring to a continuation page when the | |
3683 | * low "digit" of a count is incremented or decremented through SWAP_MAP_MAX. | |
3684 | * | |
3685 | * add_swap_count_continuation(, GFP_ATOMIC) can be called while holding | |
3686 | * page table locks; if it fails, add_swap_count_continuation(, GFP_KERNEL) | |
3687 | * can be called after dropping locks. | |
3688 | */ | |
3689 | int add_swap_count_continuation(swp_entry_t entry, gfp_t gfp_mask) | |
3690 | { | |
3691 | struct swap_info_struct *si; | |
235b6217 | 3692 | struct swap_cluster_info *ci; |
570a335b HD |
3693 | struct page *head; |
3694 | struct page *page; | |
3695 | struct page *list_page; | |
3696 | pgoff_t offset; | |
3697 | unsigned char count; | |
eb085574 | 3698 | int ret = 0; |
570a335b HD |
3699 | |
3700 | /* | |
3701 | * When debugging, it's easier to use __GFP_ZERO here; but it's better | |
3702 | * for latency not to zero a page while GFP_ATOMIC and holding locks. | |
3703 | */ | |
3704 | page = alloc_page(gfp_mask | __GFP_HIGHMEM); | |
3705 | ||
eb085574 | 3706 | si = get_swap_device(entry); |
570a335b HD |
3707 | if (!si) { |
3708 | /* | |
3709 | * An acceptable race has occurred since the failing | |
eb085574 | 3710 | * __swap_duplicate(): the swap device may be swapoff |
570a335b HD |
3711 | */ |
3712 | goto outer; | |
3713 | } | |
3714 | ||
3715 | offset = swp_offset(entry); | |
235b6217 HY |
3716 | |
3717 | ci = lock_cluster(si, offset); | |
3718 | ||
d8aa24e0 | 3719 | count = swap_count(si->swap_map[offset]); |
570a335b HD |
3720 | |
3721 | if ((count & ~COUNT_CONTINUED) != SWAP_MAP_MAX) { | |
3722 | /* | |
3723 | * The higher the swap count, the more likely it is that tasks | |
3724 | * will race to add swap count continuation: we need to avoid | |
3725 | * over-provisioning. | |
3726 | */ | |
3727 | goto out; | |
3728 | } | |
3729 | ||
3730 | if (!page) { | |
eb085574 HY |
3731 | ret = -ENOMEM; |
3732 | goto out; | |
570a335b HD |
3733 | } |
3734 | ||
570a335b HD |
3735 | head = vmalloc_to_page(si->swap_map + offset); |
3736 | offset &= ~PAGE_MASK; | |
3737 | ||
2628bd6f | 3738 | spin_lock(&si->cont_lock); |
570a335b HD |
3739 | /* |
3740 | * Page allocation does not initialize the page's lru field, | |
3741 | * but it does always reset its private field. | |
3742 | */ | |
3743 | if (!page_private(head)) { | |
3744 | BUG_ON(count & COUNT_CONTINUED); | |
3745 | INIT_LIST_HEAD(&head->lru); | |
3746 | set_page_private(head, SWP_CONTINUED); | |
3747 | si->flags |= SWP_CONTINUED; | |
3748 | } | |
3749 | ||
3750 | list_for_each_entry(list_page, &head->lru, lru) { | |
3751 | unsigned char *map; | |
3752 | ||
3753 | /* | |
3754 | * If the previous map said no continuation, but we've found | |
3755 | * a continuation page, free our allocation and use this one. | |
3756 | */ | |
3757 | if (!(count & COUNT_CONTINUED)) | |
2628bd6f | 3758 | goto out_unlock_cont; |
570a335b | 3759 | |
829c3151 | 3760 | map = kmap_local_page(list_page) + offset; |
570a335b | 3761 | count = *map; |
829c3151 | 3762 | kunmap_local(map); |
570a335b HD |
3763 | |
3764 | /* | |
3765 | * If this continuation count now has some space in it, | |
3766 | * free our allocation and use this one. | |
3767 | */ | |
3768 | if ((count & ~COUNT_CONTINUED) != SWAP_CONT_MAX) | |
2628bd6f | 3769 | goto out_unlock_cont; |
570a335b HD |
3770 | } |
3771 | ||
3772 | list_add_tail(&page->lru, &head->lru); | |
3773 | page = NULL; /* now it's attached, don't free it */ | |
2628bd6f HY |
3774 | out_unlock_cont: |
3775 | spin_unlock(&si->cont_lock); | |
570a335b | 3776 | out: |
235b6217 | 3777 | unlock_cluster(ci); |
eb085574 | 3778 | put_swap_device(si); |
570a335b HD |
3779 | outer: |
3780 | if (page) | |
3781 | __free_page(page); | |
eb085574 | 3782 | return ret; |
570a335b HD |
3783 | } |
3784 | ||
3785 | /* | |
3786 | * swap_count_continued - when the original swap_map count is incremented | |
3787 | * from SWAP_MAP_MAX, check if there is already a continuation page to carry | |
3788 | * into, carry if so, or else fail until a new continuation page is allocated; | |
3789 | * when the original swap_map count is decremented from 0 with continuation, | |
3790 | * borrow from the continuation and report whether it still holds more. | |
9c1c38bc | 3791 | * Called while __swap_duplicate() or caller of swap_entry_put_locked() |
f80ddc14 | 3792 | * holds cluster lock. |
570a335b HD |
3793 | */ |
3794 | static bool swap_count_continued(struct swap_info_struct *si, | |
3795 | pgoff_t offset, unsigned char count) | |
3796 | { | |
3797 | struct page *head; | |
3798 | struct page *page; | |
3799 | unsigned char *map; | |
2628bd6f | 3800 | bool ret; |
570a335b HD |
3801 | |
3802 | head = vmalloc_to_page(si->swap_map + offset); | |
3803 | if (page_private(head) != SWP_CONTINUED) { | |
3804 | BUG_ON(count & COUNT_CONTINUED); | |
3805 | return false; /* need to add count continuation */ | |
3806 | } | |
3807 | ||
2628bd6f | 3808 | spin_lock(&si->cont_lock); |
570a335b | 3809 | offset &= ~PAGE_MASK; |
213516ac | 3810 | page = list_next_entry(head, lru); |
829c3151 | 3811 | map = kmap_local_page(page) + offset; |
570a335b HD |
3812 | |
3813 | if (count == SWAP_MAP_MAX) /* initial increment from swap_map */ | |
3814 | goto init_map; /* jump over SWAP_CONT_MAX checks */ | |
3815 | ||
3816 | if (count == (SWAP_MAP_MAX | COUNT_CONTINUED)) { /* incrementing */ | |
3817 | /* | |
3818 | * Think of how you add 1 to 999 | |
3819 | */ | |
3820 | while (*map == (SWAP_CONT_MAX | COUNT_CONTINUED)) { | |
829c3151 | 3821 | kunmap_local(map); |
213516ac | 3822 | page = list_next_entry(page, lru); |
570a335b | 3823 | BUG_ON(page == head); |
829c3151 | 3824 | map = kmap_local_page(page) + offset; |
570a335b HD |
3825 | } |
3826 | if (*map == SWAP_CONT_MAX) { | |
829c3151 | 3827 | kunmap_local(map); |
213516ac | 3828 | page = list_next_entry(page, lru); |
2628bd6f HY |
3829 | if (page == head) { |
3830 | ret = false; /* add count continuation */ | |
3831 | goto out; | |
3832 | } | |
829c3151 | 3833 | map = kmap_local_page(page) + offset; |
570a335b HD |
3834 | init_map: *map = 0; /* we didn't zero the page */ |
3835 | } | |
3836 | *map += 1; | |
829c3151 | 3837 | kunmap_local(map); |
213516ac | 3838 | while ((page = list_prev_entry(page, lru)) != head) { |
829c3151 | 3839 | map = kmap_local_page(page) + offset; |
570a335b | 3840 | *map = COUNT_CONTINUED; |
829c3151 | 3841 | kunmap_local(map); |
570a335b | 3842 | } |
2628bd6f | 3843 | ret = true; /* incremented */ |
570a335b HD |
3844 | |
3845 | } else { /* decrementing */ | |
3846 | /* | |
3847 | * Think of how you subtract 1 from 1000 | |
3848 | */ | |
3849 | BUG_ON(count != COUNT_CONTINUED); | |
3850 | while (*map == COUNT_CONTINUED) { | |
829c3151 | 3851 | kunmap_local(map); |
213516ac | 3852 | page = list_next_entry(page, lru); |
570a335b | 3853 | BUG_ON(page == head); |
829c3151 | 3854 | map = kmap_local_page(page) + offset; |
570a335b HD |
3855 | } |
3856 | BUG_ON(*map == 0); | |
3857 | *map -= 1; | |
3858 | if (*map == 0) | |
3859 | count = 0; | |
829c3151 | 3860 | kunmap_local(map); |
213516ac | 3861 | while ((page = list_prev_entry(page, lru)) != head) { |
829c3151 | 3862 | map = kmap_local_page(page) + offset; |
570a335b HD |
3863 | *map = SWAP_CONT_MAX | count; |
3864 | count = COUNT_CONTINUED; | |
829c3151 | 3865 | kunmap_local(map); |
570a335b | 3866 | } |
2628bd6f | 3867 | ret = count == COUNT_CONTINUED; |
570a335b | 3868 | } |
2628bd6f HY |
3869 | out: |
3870 | spin_unlock(&si->cont_lock); | |
3871 | return ret; | |
570a335b HD |
3872 | } |
3873 | ||
3874 | /* | |
3875 | * free_swap_count_continuations - swapoff free all the continuation pages | |
3876 | * appended to the swap_map, after swap_map is quiesced, before vfree'ing it. | |
3877 | */ | |
3878 | static void free_swap_count_continuations(struct swap_info_struct *si) | |
3879 | { | |
3880 | pgoff_t offset; | |
3881 | ||
3882 | for (offset = 0; offset < si->max; offset += PAGE_SIZE) { | |
3883 | struct page *head; | |
3884 | head = vmalloc_to_page(si->swap_map + offset); | |
3885 | if (page_private(head)) { | |
0d576d20 GT |
3886 | struct page *page, *next; |
3887 | ||
3888 | list_for_each_entry_safe(page, next, &head->lru, lru) { | |
3889 | list_del(&page->lru); | |
570a335b HD |
3890 | __free_page(page); |
3891 | } | |
3892 | } | |
3893 | } | |
3894 | } | |
a2468cc9 | 3895 | |
2cf85583 | 3896 | #if defined(CONFIG_MEMCG) && defined(CONFIG_BLK_CGROUP) |
0ff67f99 KS |
3897 | static bool __has_usable_swap(void) |
3898 | { | |
3899 | return !plist_head_empty(&swap_active_head); | |
3900 | } | |
3901 | ||
3e4fb13a | 3902 | void __folio_throttle_swaprate(struct folio *folio, gfp_t gfp) |
2cf85583 TH |
3903 | { |
3904 | struct swap_info_struct *si, *next; | |
3e4fb13a | 3905 | int nid = folio_nid(folio); |
6caa6a07 | 3906 | |
3e4fb13a | 3907 | if (!(gfp & __GFP_IO)) |
2cf85583 TH |
3908 | return; |
3909 | ||
80e75021 KW |
3910 | if (!__has_usable_swap()) |
3911 | return; | |
3912 | ||
2cf85583 TH |
3913 | if (!blk_cgroup_congested()) |
3914 | return; | |
3915 | ||
3916 | /* | |
3917 | * We've already scheduled a throttle, avoid taking the global swap | |
3918 | * lock. | |
3919 | */ | |
f05837ed | 3920 | if (current->throttle_disk) |
2cf85583 TH |
3921 | return; |
3922 | ||
3923 | spin_lock(&swap_avail_lock); | |
6caa6a07 JW |
3924 | plist_for_each_entry_safe(si, next, &swap_avail_heads[nid], |
3925 | avail_lists[nid]) { | |
2cf85583 | 3926 | if (si->bdev) { |
de185b56 | 3927 | blkcg_schedule_throttle(si->bdev->bd_disk, true); |
2cf85583 TH |
3928 | break; |
3929 | } | |
3930 | } | |
3931 | spin_unlock(&swap_avail_lock); | |
3932 | } | |
3933 | #endif | |
3934 | ||
a2468cc9 AL |
3935 | static int __init swapfile_init(void) |
3936 | { | |
3937 | int nid; | |
3938 | ||
3939 | swap_avail_heads = kmalloc_array(nr_node_ids, sizeof(struct plist_head), | |
3940 | GFP_KERNEL); | |
3941 | if (!swap_avail_heads) { | |
3942 | pr_emerg("Not enough memory for swap heads, swap is disabled\n"); | |
3943 | return -ENOMEM; | |
3944 | } | |
3945 | ||
3946 | for_each_node(nid) | |
3947 | plist_head_init(&swap_avail_heads[nid]); | |
3948 | ||
be45a490 PX |
3949 | swapfile_maximum_size = arch_max_swapfile_size(); |
3950 | ||
5154e607 PX |
3951 | #ifdef CONFIG_MIGRATION |
3952 | if (swapfile_maximum_size >= (1UL << SWP_MIG_TOTAL_BITS)) | |
3953 | swap_migration_ad_supported = true; | |
3954 | #endif /* CONFIG_MIGRATION */ | |
3955 | ||
a2468cc9 AL |
3956 | return 0; |
3957 | } | |
3958 | subsys_initcall(swapfile_init); |