Commit | Line | Data |
---|---|---|
457c8996 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
1da177e4 LT |
2 | /* |
3 | * linux/mm/swapfile.c | |
4 | * | |
5 | * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds | |
6 | * Swap reorganised 29.12.95, Stephen Tweedie | |
7 | */ | |
8 | ||
c97ab271 | 9 | #include <linux/blkdev.h> |
1da177e4 | 10 | #include <linux/mm.h> |
6e84f315 | 11 | #include <linux/sched/mm.h> |
29930025 | 12 | #include <linux/sched/task.h> |
1da177e4 LT |
13 | #include <linux/hugetlb.h> |
14 | #include <linux/mman.h> | |
15 | #include <linux/slab.h> | |
16 | #include <linux/kernel_stat.h> | |
17 | #include <linux/swap.h> | |
18 | #include <linux/vmalloc.h> | |
19 | #include <linux/pagemap.h> | |
20 | #include <linux/namei.h> | |
072441e2 | 21 | #include <linux/shmem_fs.h> |
e41d12f5 | 22 | #include <linux/blk-cgroup.h> |
20137a49 | 23 | #include <linux/random.h> |
1da177e4 LT |
24 | #include <linux/writeback.h> |
25 | #include <linux/proc_fs.h> | |
26 | #include <linux/seq_file.h> | |
27 | #include <linux/init.h> | |
5ad64688 | 28 | #include <linux/ksm.h> |
1da177e4 LT |
29 | #include <linux/rmap.h> |
30 | #include <linux/security.h> | |
31 | #include <linux/backing-dev.h> | |
fc0abb14 | 32 | #include <linux/mutex.h> |
c59ede7b | 33 | #include <linux/capability.h> |
1da177e4 | 34 | #include <linux/syscalls.h> |
8a9f3ccd | 35 | #include <linux/memcontrol.h> |
66d7dd51 | 36 | #include <linux/poll.h> |
72788c38 | 37 | #include <linux/oom.h> |
38b5faf4 DM |
38 | #include <linux/frontswap.h> |
39 | #include <linux/swapfile.h> | |
f981c595 | 40 | #include <linux/export.h> |
67afa38e | 41 | #include <linux/swap_slots.h> |
155b5f88 | 42 | #include <linux/sort.h> |
63d8620e | 43 | #include <linux/completion.h> |
1da177e4 | 44 | |
1da177e4 LT |
45 | #include <asm/tlbflush.h> |
46 | #include <linux/swapops.h> | |
5d1ea48b | 47 | #include <linux/swap_cgroup.h> |
014bb1de | 48 | #include "swap.h" |
1da177e4 | 49 | |
570a335b HD |
50 | static bool swap_count_continued(struct swap_info_struct *, pgoff_t, |
51 | unsigned char); | |
52 | static void free_swap_count_continuations(struct swap_info_struct *); | |
53 | ||
633423a0 | 54 | static DEFINE_SPINLOCK(swap_lock); |
7c363b8c | 55 | static unsigned int nr_swapfiles; |
ec8acf20 | 56 | atomic_long_t nr_swap_pages; |
fb0fec50 CW |
57 | /* |
58 | * Some modules use swappable objects and may try to swap them out under | |
59 | * memory pressure (via the shrinker). Before doing so, they may wish to | |
60 | * check to see if any swap space is available. | |
61 | */ | |
62 | EXPORT_SYMBOL_GPL(nr_swap_pages); | |
ec8acf20 | 63 | /* protected with swap_lock. reading in vm_swap_full() doesn't need lock */ |
1da177e4 | 64 | long total_swap_pages; |
a2468cc9 | 65 | static int least_priority = -1; |
1da177e4 | 66 | |
1da177e4 LT |
67 | static const char Bad_file[] = "Bad swap file entry "; |
68 | static const char Unused_file[] = "Unused swap file entry "; | |
69 | static const char Bad_offset[] = "Bad swap offset entry "; | |
70 | static const char Unused_offset[] = "Unused swap offset entry "; | |
71 | ||
adfab836 DS |
72 | /* |
73 | * all active swap_info_structs | |
74 | * protected with swap_lock, and ordered by priority. | |
75 | */ | |
633423a0 | 76 | static PLIST_HEAD(swap_active_head); |
18ab4d4c DS |
77 | |
78 | /* | |
79 | * all available (active, not full) swap_info_structs | |
80 | * protected with swap_avail_lock, ordered by priority. | |
e2e3fdc7 | 81 | * This is used by folio_alloc_swap() instead of swap_active_head |
18ab4d4c | 82 | * because swap_active_head includes all swap_info_structs, |
e2e3fdc7 | 83 | * but folio_alloc_swap() doesn't need to look at full ones. |
18ab4d4c DS |
84 | * This uses its own lock instead of swap_lock because when a |
85 | * swap_info_struct changes between not-full/full, it needs to | |
86 | * add/remove itself to/from this list, but the swap_info_struct->lock | |
87 | * is held and the locking order requires swap_lock to be taken | |
88 | * before any swap_info_struct->lock. | |
89 | */ | |
bfc6b1ca | 90 | static struct plist_head *swap_avail_heads; |
18ab4d4c | 91 | static DEFINE_SPINLOCK(swap_avail_lock); |
1da177e4 | 92 | |
38b5faf4 | 93 | struct swap_info_struct *swap_info[MAX_SWAPFILES]; |
1da177e4 | 94 | |
fc0abb14 | 95 | static DEFINE_MUTEX(swapon_mutex); |
1da177e4 | 96 | |
66d7dd51 KS |
97 | static DECLARE_WAIT_QUEUE_HEAD(proc_poll_wait); |
98 | /* Activity counter to indicate that a swapon or swapoff has occurred */ | |
99 | static atomic_t proc_poll_event = ATOMIC_INIT(0); | |
100 | ||
81a0298b HY |
101 | atomic_t nr_rotate_swap = ATOMIC_INIT(0); |
102 | ||
c10d38cc DJ |
103 | static struct swap_info_struct *swap_type_to_swap_info(int type) |
104 | { | |
a4b45114 | 105 | if (type >= MAX_SWAPFILES) |
c10d38cc DJ |
106 | return NULL; |
107 | ||
a4b45114 | 108 | return READ_ONCE(swap_info[type]); /* rcu_dereference() */ |
c10d38cc DJ |
109 | } |
110 | ||
8d69aaee | 111 | static inline unsigned char swap_count(unsigned char ent) |
355cfa73 | 112 | { |
955c97f0 | 113 | return ent & ~SWAP_HAS_CACHE; /* may include COUNT_CONTINUED flag */ |
355cfa73 KH |
114 | } |
115 | ||
bcd49e86 HY |
116 | /* Reclaim the swap entry anyway if possible */ |
117 | #define TTRS_ANYWAY 0x1 | |
118 | /* | |
119 | * Reclaim the swap entry if there are no more mappings of the | |
120 | * corresponding page | |
121 | */ | |
122 | #define TTRS_UNMAPPED 0x2 | |
123 | /* Reclaim the swap entry if swap is getting full*/ | |
124 | #define TTRS_FULL 0x4 | |
125 | ||
efa90a98 | 126 | /* returns 1 if swap entry is freed */ |
bcd49e86 HY |
127 | static int __try_to_reclaim_swap(struct swap_info_struct *si, |
128 | unsigned long offset, unsigned long flags) | |
c9e44410 | 129 | { |
efa90a98 | 130 | swp_entry_t entry = swp_entry(si->type, offset); |
c9e44410 KH |
131 | struct page *page; |
132 | int ret = 0; | |
133 | ||
bcd49e86 | 134 | page = find_get_page(swap_address_space(entry), offset); |
c9e44410 KH |
135 | if (!page) |
136 | return 0; | |
137 | /* | |
bcd49e86 HY |
138 | * When this function is called from scan_swap_map_slots() and it's |
139 | * called by vmscan.c at reclaiming pages. So, we hold a lock on a page, | |
140 | * here. We have to use trylock for avoiding deadlock. This is a special | |
c9e44410 KH |
141 | * case and you should use try_to_free_swap() with explicit lock_page() |
142 | * in usual operations. | |
143 | */ | |
144 | if (trylock_page(page)) { | |
bcd49e86 HY |
145 | if ((flags & TTRS_ANYWAY) || |
146 | ((flags & TTRS_UNMAPPED) && !page_mapped(page)) || | |
147 | ((flags & TTRS_FULL) && mem_cgroup_swap_full(page))) | |
148 | ret = try_to_free_swap(page); | |
c9e44410 KH |
149 | unlock_page(page); |
150 | } | |
09cbfeaf | 151 | put_page(page); |
c9e44410 KH |
152 | return ret; |
153 | } | |
355cfa73 | 154 | |
4efaceb1 AL |
155 | static inline struct swap_extent *first_se(struct swap_info_struct *sis) |
156 | { | |
157 | struct rb_node *rb = rb_first(&sis->swap_extent_root); | |
158 | return rb_entry(rb, struct swap_extent, rb_node); | |
159 | } | |
160 | ||
161 | static inline struct swap_extent *next_se(struct swap_extent *se) | |
162 | { | |
163 | struct rb_node *rb = rb_next(&se->rb_node); | |
164 | return rb ? rb_entry(rb, struct swap_extent, rb_node) : NULL; | |
165 | } | |
166 | ||
6a6ba831 HD |
167 | /* |
168 | * swapon tell device that all the old swap contents can be discarded, | |
169 | * to allow the swap device to optimize its wear-levelling. | |
170 | */ | |
171 | static int discard_swap(struct swap_info_struct *si) | |
172 | { | |
173 | struct swap_extent *se; | |
9625a5f2 HD |
174 | sector_t start_block; |
175 | sector_t nr_blocks; | |
6a6ba831 HD |
176 | int err = 0; |
177 | ||
9625a5f2 | 178 | /* Do not discard the swap header page! */ |
4efaceb1 | 179 | se = first_se(si); |
9625a5f2 HD |
180 | start_block = (se->start_block + 1) << (PAGE_SHIFT - 9); |
181 | nr_blocks = ((sector_t)se->nr_pages - 1) << (PAGE_SHIFT - 9); | |
182 | if (nr_blocks) { | |
183 | err = blkdev_issue_discard(si->bdev, start_block, | |
44abff2c | 184 | nr_blocks, GFP_KERNEL); |
9625a5f2 HD |
185 | if (err) |
186 | return err; | |
187 | cond_resched(); | |
188 | } | |
6a6ba831 | 189 | |
4efaceb1 | 190 | for (se = next_se(se); se; se = next_se(se)) { |
9625a5f2 HD |
191 | start_block = se->start_block << (PAGE_SHIFT - 9); |
192 | nr_blocks = (sector_t)se->nr_pages << (PAGE_SHIFT - 9); | |
6a6ba831 HD |
193 | |
194 | err = blkdev_issue_discard(si->bdev, start_block, | |
44abff2c | 195 | nr_blocks, GFP_KERNEL); |
6a6ba831 HD |
196 | if (err) |
197 | break; | |
198 | ||
199 | cond_resched(); | |
200 | } | |
201 | return err; /* That will often be -EOPNOTSUPP */ | |
202 | } | |
203 | ||
4efaceb1 AL |
204 | static struct swap_extent * |
205 | offset_to_swap_extent(struct swap_info_struct *sis, unsigned long offset) | |
206 | { | |
207 | struct swap_extent *se; | |
208 | struct rb_node *rb; | |
209 | ||
210 | rb = sis->swap_extent_root.rb_node; | |
211 | while (rb) { | |
212 | se = rb_entry(rb, struct swap_extent, rb_node); | |
213 | if (offset < se->start_page) | |
214 | rb = rb->rb_left; | |
215 | else if (offset >= se->start_page + se->nr_pages) | |
216 | rb = rb->rb_right; | |
217 | else | |
218 | return se; | |
219 | } | |
220 | /* It *must* be present */ | |
221 | BUG(); | |
222 | } | |
223 | ||
caf6912f JA |
224 | sector_t swap_page_sector(struct page *page) |
225 | { | |
226 | struct swap_info_struct *sis = page_swap_info(page); | |
227 | struct swap_extent *se; | |
228 | sector_t sector; | |
229 | pgoff_t offset; | |
230 | ||
231 | offset = __page_file_index(page); | |
232 | se = offset_to_swap_extent(sis, offset); | |
233 | sector = se->start_block + (offset - se->start_page); | |
234 | return sector << (PAGE_SHIFT - 9); | |
235 | } | |
236 | ||
7992fde7 HD |
237 | /* |
238 | * swap allocation tell device that a cluster of swap can now be discarded, | |
239 | * to allow the swap device to optimize its wear-levelling. | |
240 | */ | |
241 | static void discard_swap_cluster(struct swap_info_struct *si, | |
242 | pgoff_t start_page, pgoff_t nr_pages) | |
243 | { | |
4efaceb1 | 244 | struct swap_extent *se = offset_to_swap_extent(si, start_page); |
7992fde7 HD |
245 | |
246 | while (nr_pages) { | |
4efaceb1 AL |
247 | pgoff_t offset = start_page - se->start_page; |
248 | sector_t start_block = se->start_block + offset; | |
249 | sector_t nr_blocks = se->nr_pages - offset; | |
250 | ||
251 | if (nr_blocks > nr_pages) | |
252 | nr_blocks = nr_pages; | |
253 | start_page += nr_blocks; | |
254 | nr_pages -= nr_blocks; | |
255 | ||
256 | start_block <<= PAGE_SHIFT - 9; | |
257 | nr_blocks <<= PAGE_SHIFT - 9; | |
258 | if (blkdev_issue_discard(si->bdev, start_block, | |
44abff2c | 259 | nr_blocks, GFP_NOIO)) |
4efaceb1 | 260 | break; |
7992fde7 | 261 | |
4efaceb1 | 262 | se = next_se(se); |
7992fde7 HD |
263 | } |
264 | } | |
265 | ||
38d8b4e6 HY |
266 | #ifdef CONFIG_THP_SWAP |
267 | #define SWAPFILE_CLUSTER HPAGE_PMD_NR | |
a448f2d0 HY |
268 | |
269 | #define swap_entry_size(size) (size) | |
38d8b4e6 | 270 | #else |
048c27fd | 271 | #define SWAPFILE_CLUSTER 256 |
a448f2d0 HY |
272 | |
273 | /* | |
274 | * Define swap_entry_size() as constant to let compiler to optimize | |
275 | * out some code if !CONFIG_THP_SWAP | |
276 | */ | |
277 | #define swap_entry_size(size) 1 | |
38d8b4e6 | 278 | #endif |
048c27fd HD |
279 | #define LATENCY_LIMIT 256 |
280 | ||
2a8f9449 SL |
281 | static inline void cluster_set_flag(struct swap_cluster_info *info, |
282 | unsigned int flag) | |
283 | { | |
284 | info->flags = flag; | |
285 | } | |
286 | ||
287 | static inline unsigned int cluster_count(struct swap_cluster_info *info) | |
288 | { | |
289 | return info->data; | |
290 | } | |
291 | ||
292 | static inline void cluster_set_count(struct swap_cluster_info *info, | |
293 | unsigned int c) | |
294 | { | |
295 | info->data = c; | |
296 | } | |
297 | ||
298 | static inline void cluster_set_count_flag(struct swap_cluster_info *info, | |
299 | unsigned int c, unsigned int f) | |
300 | { | |
301 | info->flags = f; | |
302 | info->data = c; | |
303 | } | |
304 | ||
305 | static inline unsigned int cluster_next(struct swap_cluster_info *info) | |
306 | { | |
307 | return info->data; | |
308 | } | |
309 | ||
310 | static inline void cluster_set_next(struct swap_cluster_info *info, | |
311 | unsigned int n) | |
312 | { | |
313 | info->data = n; | |
314 | } | |
315 | ||
316 | static inline void cluster_set_next_flag(struct swap_cluster_info *info, | |
317 | unsigned int n, unsigned int f) | |
318 | { | |
319 | info->flags = f; | |
320 | info->data = n; | |
321 | } | |
322 | ||
323 | static inline bool cluster_is_free(struct swap_cluster_info *info) | |
324 | { | |
325 | return info->flags & CLUSTER_FLAG_FREE; | |
326 | } | |
327 | ||
328 | static inline bool cluster_is_null(struct swap_cluster_info *info) | |
329 | { | |
330 | return info->flags & CLUSTER_FLAG_NEXT_NULL; | |
331 | } | |
332 | ||
333 | static inline void cluster_set_null(struct swap_cluster_info *info) | |
334 | { | |
335 | info->flags = CLUSTER_FLAG_NEXT_NULL; | |
336 | info->data = 0; | |
337 | } | |
338 | ||
e0709829 HY |
339 | static inline bool cluster_is_huge(struct swap_cluster_info *info) |
340 | { | |
33ee011e HY |
341 | if (IS_ENABLED(CONFIG_THP_SWAP)) |
342 | return info->flags & CLUSTER_FLAG_HUGE; | |
343 | return false; | |
e0709829 HY |
344 | } |
345 | ||
346 | static inline void cluster_clear_huge(struct swap_cluster_info *info) | |
347 | { | |
348 | info->flags &= ~CLUSTER_FLAG_HUGE; | |
349 | } | |
350 | ||
235b6217 HY |
351 | static inline struct swap_cluster_info *lock_cluster(struct swap_info_struct *si, |
352 | unsigned long offset) | |
353 | { | |
354 | struct swap_cluster_info *ci; | |
355 | ||
356 | ci = si->cluster_info; | |
357 | if (ci) { | |
358 | ci += offset / SWAPFILE_CLUSTER; | |
359 | spin_lock(&ci->lock); | |
360 | } | |
361 | return ci; | |
362 | } | |
363 | ||
364 | static inline void unlock_cluster(struct swap_cluster_info *ci) | |
365 | { | |
366 | if (ci) | |
367 | spin_unlock(&ci->lock); | |
368 | } | |
369 | ||
59d98bf3 HY |
370 | /* |
371 | * Determine the locking method in use for this device. Return | |
372 | * swap_cluster_info if SSD-style cluster-based locking is in place. | |
373 | */ | |
235b6217 | 374 | static inline struct swap_cluster_info *lock_cluster_or_swap_info( |
59d98bf3 | 375 | struct swap_info_struct *si, unsigned long offset) |
235b6217 HY |
376 | { |
377 | struct swap_cluster_info *ci; | |
378 | ||
59d98bf3 | 379 | /* Try to use fine-grained SSD-style locking if available: */ |
235b6217 | 380 | ci = lock_cluster(si, offset); |
59d98bf3 | 381 | /* Otherwise, fall back to traditional, coarse locking: */ |
235b6217 HY |
382 | if (!ci) |
383 | spin_lock(&si->lock); | |
384 | ||
385 | return ci; | |
386 | } | |
387 | ||
388 | static inline void unlock_cluster_or_swap_info(struct swap_info_struct *si, | |
389 | struct swap_cluster_info *ci) | |
390 | { | |
391 | if (ci) | |
392 | unlock_cluster(ci); | |
393 | else | |
394 | spin_unlock(&si->lock); | |
395 | } | |
396 | ||
6b534915 HY |
397 | static inline bool cluster_list_empty(struct swap_cluster_list *list) |
398 | { | |
399 | return cluster_is_null(&list->head); | |
400 | } | |
401 | ||
402 | static inline unsigned int cluster_list_first(struct swap_cluster_list *list) | |
403 | { | |
404 | return cluster_next(&list->head); | |
405 | } | |
406 | ||
407 | static void cluster_list_init(struct swap_cluster_list *list) | |
408 | { | |
409 | cluster_set_null(&list->head); | |
410 | cluster_set_null(&list->tail); | |
411 | } | |
412 | ||
413 | static void cluster_list_add_tail(struct swap_cluster_list *list, | |
414 | struct swap_cluster_info *ci, | |
415 | unsigned int idx) | |
416 | { | |
417 | if (cluster_list_empty(list)) { | |
418 | cluster_set_next_flag(&list->head, idx, 0); | |
419 | cluster_set_next_flag(&list->tail, idx, 0); | |
420 | } else { | |
235b6217 | 421 | struct swap_cluster_info *ci_tail; |
6b534915 HY |
422 | unsigned int tail = cluster_next(&list->tail); |
423 | ||
235b6217 HY |
424 | /* |
425 | * Nested cluster lock, but both cluster locks are | |
426 | * only acquired when we held swap_info_struct->lock | |
427 | */ | |
428 | ci_tail = ci + tail; | |
429 | spin_lock_nested(&ci_tail->lock, SINGLE_DEPTH_NESTING); | |
430 | cluster_set_next(ci_tail, idx); | |
0ef017d1 | 431 | spin_unlock(&ci_tail->lock); |
6b534915 HY |
432 | cluster_set_next_flag(&list->tail, idx, 0); |
433 | } | |
434 | } | |
435 | ||
436 | static unsigned int cluster_list_del_first(struct swap_cluster_list *list, | |
437 | struct swap_cluster_info *ci) | |
438 | { | |
439 | unsigned int idx; | |
440 | ||
441 | idx = cluster_next(&list->head); | |
442 | if (cluster_next(&list->tail) == idx) { | |
443 | cluster_set_null(&list->head); | |
444 | cluster_set_null(&list->tail); | |
445 | } else | |
446 | cluster_set_next_flag(&list->head, | |
447 | cluster_next(&ci[idx]), 0); | |
448 | ||
449 | return idx; | |
450 | } | |
451 | ||
815c2c54 SL |
452 | /* Add a cluster to discard list and schedule it to do discard */ |
453 | static void swap_cluster_schedule_discard(struct swap_info_struct *si, | |
454 | unsigned int idx) | |
455 | { | |
456 | /* | |
bb243f7d | 457 | * If scan_swap_map_slots() can't find a free cluster, it will check |
815c2c54 | 458 | * si->swap_map directly. To make sure the discarding cluster isn't |
bb243f7d ML |
459 | * taken by scan_swap_map_slots(), mark the swap entries bad (occupied). |
460 | * It will be cleared after discard | |
815c2c54 SL |
461 | */ |
462 | memset(si->swap_map + idx * SWAPFILE_CLUSTER, | |
463 | SWAP_MAP_BAD, SWAPFILE_CLUSTER); | |
464 | ||
6b534915 | 465 | cluster_list_add_tail(&si->discard_clusters, si->cluster_info, idx); |
815c2c54 SL |
466 | |
467 | schedule_work(&si->discard_work); | |
468 | } | |
469 | ||
38d8b4e6 HY |
470 | static void __free_cluster(struct swap_info_struct *si, unsigned long idx) |
471 | { | |
472 | struct swap_cluster_info *ci = si->cluster_info; | |
473 | ||
474 | cluster_set_flag(ci + idx, CLUSTER_FLAG_FREE); | |
475 | cluster_list_add_tail(&si->free_clusters, ci, idx); | |
476 | } | |
477 | ||
815c2c54 SL |
478 | /* |
479 | * Doing discard actually. After a cluster discard is finished, the cluster | |
480 | * will be added to free cluster list. caller should hold si->lock. | |
481 | */ | |
482 | static void swap_do_scheduled_discard(struct swap_info_struct *si) | |
483 | { | |
235b6217 | 484 | struct swap_cluster_info *info, *ci; |
815c2c54 SL |
485 | unsigned int idx; |
486 | ||
487 | info = si->cluster_info; | |
488 | ||
6b534915 HY |
489 | while (!cluster_list_empty(&si->discard_clusters)) { |
490 | idx = cluster_list_del_first(&si->discard_clusters, info); | |
815c2c54 SL |
491 | spin_unlock(&si->lock); |
492 | ||
493 | discard_swap_cluster(si, idx * SWAPFILE_CLUSTER, | |
494 | SWAPFILE_CLUSTER); | |
495 | ||
496 | spin_lock(&si->lock); | |
235b6217 | 497 | ci = lock_cluster(si, idx * SWAPFILE_CLUSTER); |
38d8b4e6 | 498 | __free_cluster(si, idx); |
815c2c54 SL |
499 | memset(si->swap_map + idx * SWAPFILE_CLUSTER, |
500 | 0, SWAPFILE_CLUSTER); | |
235b6217 | 501 | unlock_cluster(ci); |
815c2c54 SL |
502 | } |
503 | } | |
504 | ||
505 | static void swap_discard_work(struct work_struct *work) | |
506 | { | |
507 | struct swap_info_struct *si; | |
508 | ||
509 | si = container_of(work, struct swap_info_struct, discard_work); | |
510 | ||
511 | spin_lock(&si->lock); | |
512 | swap_do_scheduled_discard(si); | |
513 | spin_unlock(&si->lock); | |
514 | } | |
515 | ||
63d8620e ML |
516 | static void swap_users_ref_free(struct percpu_ref *ref) |
517 | { | |
518 | struct swap_info_struct *si; | |
519 | ||
520 | si = container_of(ref, struct swap_info_struct, users); | |
521 | complete(&si->comp); | |
522 | } | |
523 | ||
38d8b4e6 HY |
524 | static void alloc_cluster(struct swap_info_struct *si, unsigned long idx) |
525 | { | |
526 | struct swap_cluster_info *ci = si->cluster_info; | |
527 | ||
528 | VM_BUG_ON(cluster_list_first(&si->free_clusters) != idx); | |
529 | cluster_list_del_first(&si->free_clusters, ci); | |
530 | cluster_set_count_flag(ci + idx, 0, 0); | |
531 | } | |
532 | ||
533 | static void free_cluster(struct swap_info_struct *si, unsigned long idx) | |
534 | { | |
535 | struct swap_cluster_info *ci = si->cluster_info + idx; | |
536 | ||
537 | VM_BUG_ON(cluster_count(ci) != 0); | |
538 | /* | |
539 | * If the swap is discardable, prepare discard the cluster | |
540 | * instead of free it immediately. The cluster will be freed | |
541 | * after discard. | |
542 | */ | |
543 | if ((si->flags & (SWP_WRITEOK | SWP_PAGE_DISCARD)) == | |
544 | (SWP_WRITEOK | SWP_PAGE_DISCARD)) { | |
545 | swap_cluster_schedule_discard(si, idx); | |
546 | return; | |
547 | } | |
548 | ||
549 | __free_cluster(si, idx); | |
550 | } | |
551 | ||
2a8f9449 SL |
552 | /* |
553 | * The cluster corresponding to page_nr will be used. The cluster will be | |
554 | * removed from free cluster list and its usage counter will be increased. | |
555 | */ | |
556 | static void inc_cluster_info_page(struct swap_info_struct *p, | |
557 | struct swap_cluster_info *cluster_info, unsigned long page_nr) | |
558 | { | |
559 | unsigned long idx = page_nr / SWAPFILE_CLUSTER; | |
560 | ||
561 | if (!cluster_info) | |
562 | return; | |
38d8b4e6 HY |
563 | if (cluster_is_free(&cluster_info[idx])) |
564 | alloc_cluster(p, idx); | |
2a8f9449 SL |
565 | |
566 | VM_BUG_ON(cluster_count(&cluster_info[idx]) >= SWAPFILE_CLUSTER); | |
567 | cluster_set_count(&cluster_info[idx], | |
568 | cluster_count(&cluster_info[idx]) + 1); | |
569 | } | |
570 | ||
571 | /* | |
572 | * The cluster corresponding to page_nr decreases one usage. If the usage | |
573 | * counter becomes 0, which means no page in the cluster is in using, we can | |
574 | * optionally discard the cluster and add it to free cluster list. | |
575 | */ | |
576 | static void dec_cluster_info_page(struct swap_info_struct *p, | |
577 | struct swap_cluster_info *cluster_info, unsigned long page_nr) | |
578 | { | |
579 | unsigned long idx = page_nr / SWAPFILE_CLUSTER; | |
580 | ||
581 | if (!cluster_info) | |
582 | return; | |
583 | ||
584 | VM_BUG_ON(cluster_count(&cluster_info[idx]) == 0); | |
585 | cluster_set_count(&cluster_info[idx], | |
586 | cluster_count(&cluster_info[idx]) - 1); | |
587 | ||
38d8b4e6 HY |
588 | if (cluster_count(&cluster_info[idx]) == 0) |
589 | free_cluster(p, idx); | |
2a8f9449 SL |
590 | } |
591 | ||
592 | /* | |
bb243f7d | 593 | * It's possible scan_swap_map_slots() uses a free cluster in the middle of free |
2a8f9449 SL |
594 | * cluster list. Avoiding such abuse to avoid list corruption. |
595 | */ | |
ebc2a1a6 SL |
596 | static bool |
597 | scan_swap_map_ssd_cluster_conflict(struct swap_info_struct *si, | |
2a8f9449 SL |
598 | unsigned long offset) |
599 | { | |
ebc2a1a6 SL |
600 | struct percpu_cluster *percpu_cluster; |
601 | bool conflict; | |
602 | ||
2a8f9449 | 603 | offset /= SWAPFILE_CLUSTER; |
6b534915 HY |
604 | conflict = !cluster_list_empty(&si->free_clusters) && |
605 | offset != cluster_list_first(&si->free_clusters) && | |
2a8f9449 | 606 | cluster_is_free(&si->cluster_info[offset]); |
ebc2a1a6 SL |
607 | |
608 | if (!conflict) | |
609 | return false; | |
610 | ||
611 | percpu_cluster = this_cpu_ptr(si->percpu_cluster); | |
612 | cluster_set_null(&percpu_cluster->index); | |
613 | return true; | |
614 | } | |
615 | ||
616 | /* | |
617 | * Try to get a swap entry from current cpu's swap entry pool (a cluster). This | |
618 | * might involve allocating a new cluster for current CPU too. | |
619 | */ | |
36005bae | 620 | static bool scan_swap_map_try_ssd_cluster(struct swap_info_struct *si, |
ebc2a1a6 SL |
621 | unsigned long *offset, unsigned long *scan_base) |
622 | { | |
623 | struct percpu_cluster *cluster; | |
235b6217 | 624 | struct swap_cluster_info *ci; |
235b6217 | 625 | unsigned long tmp, max; |
ebc2a1a6 SL |
626 | |
627 | new_cluster: | |
628 | cluster = this_cpu_ptr(si->percpu_cluster); | |
629 | if (cluster_is_null(&cluster->index)) { | |
6b534915 HY |
630 | if (!cluster_list_empty(&si->free_clusters)) { |
631 | cluster->index = si->free_clusters.head; | |
ebc2a1a6 SL |
632 | cluster->next = cluster_next(&cluster->index) * |
633 | SWAPFILE_CLUSTER; | |
6b534915 | 634 | } else if (!cluster_list_empty(&si->discard_clusters)) { |
ebc2a1a6 SL |
635 | /* |
636 | * we don't have free cluster but have some clusters in | |
49070588 HY |
637 | * discarding, do discard now and reclaim them, then |
638 | * reread cluster_next_cpu since we dropped si->lock | |
ebc2a1a6 SL |
639 | */ |
640 | swap_do_scheduled_discard(si); | |
49070588 HY |
641 | *scan_base = this_cpu_read(*si->cluster_next_cpu); |
642 | *offset = *scan_base; | |
ebc2a1a6 SL |
643 | goto new_cluster; |
644 | } else | |
36005bae | 645 | return false; |
ebc2a1a6 SL |
646 | } |
647 | ||
ebc2a1a6 SL |
648 | /* |
649 | * Other CPUs can use our cluster if they can't find a free cluster, | |
650 | * check if there is still free entry in the cluster | |
651 | */ | |
652 | tmp = cluster->next; | |
235b6217 HY |
653 | max = min_t(unsigned long, si->max, |
654 | (cluster_next(&cluster->index) + 1) * SWAPFILE_CLUSTER); | |
7b9e2de1 WY |
655 | if (tmp < max) { |
656 | ci = lock_cluster(si, tmp); | |
657 | while (tmp < max) { | |
658 | if (!si->swap_map[tmp]) | |
659 | break; | |
660 | tmp++; | |
661 | } | |
662 | unlock_cluster(ci); | |
ebc2a1a6 | 663 | } |
0fd0e19e | 664 | if (tmp >= max) { |
ebc2a1a6 SL |
665 | cluster_set_null(&cluster->index); |
666 | goto new_cluster; | |
667 | } | |
668 | cluster->next = tmp + 1; | |
669 | *offset = tmp; | |
670 | *scan_base = tmp; | |
fdff1deb | 671 | return true; |
2a8f9449 SL |
672 | } |
673 | ||
a2468cc9 AL |
674 | static void __del_from_avail_list(struct swap_info_struct *p) |
675 | { | |
676 | int nid; | |
677 | ||
678 | for_each_node(nid) | |
679 | plist_del(&p->avail_lists[nid], &swap_avail_heads[nid]); | |
680 | } | |
681 | ||
682 | static void del_from_avail_list(struct swap_info_struct *p) | |
683 | { | |
684 | spin_lock(&swap_avail_lock); | |
685 | __del_from_avail_list(p); | |
686 | spin_unlock(&swap_avail_lock); | |
687 | } | |
688 | ||
38d8b4e6 HY |
689 | static void swap_range_alloc(struct swap_info_struct *si, unsigned long offset, |
690 | unsigned int nr_entries) | |
691 | { | |
692 | unsigned int end = offset + nr_entries - 1; | |
693 | ||
694 | if (offset == si->lowest_bit) | |
695 | si->lowest_bit += nr_entries; | |
696 | if (end == si->highest_bit) | |
a449bf58 | 697 | WRITE_ONCE(si->highest_bit, si->highest_bit - nr_entries); |
c8945306 | 698 | WRITE_ONCE(si->inuse_pages, si->inuse_pages + nr_entries); |
38d8b4e6 HY |
699 | if (si->inuse_pages == si->pages) { |
700 | si->lowest_bit = si->max; | |
701 | si->highest_bit = 0; | |
a2468cc9 | 702 | del_from_avail_list(si); |
38d8b4e6 HY |
703 | } |
704 | } | |
705 | ||
a2468cc9 AL |
706 | static void add_to_avail_list(struct swap_info_struct *p) |
707 | { | |
708 | int nid; | |
709 | ||
710 | spin_lock(&swap_avail_lock); | |
711 | for_each_node(nid) { | |
712 | WARN_ON(!plist_node_empty(&p->avail_lists[nid])); | |
713 | plist_add(&p->avail_lists[nid], &swap_avail_heads[nid]); | |
714 | } | |
715 | spin_unlock(&swap_avail_lock); | |
716 | } | |
717 | ||
38d8b4e6 HY |
718 | static void swap_range_free(struct swap_info_struct *si, unsigned long offset, |
719 | unsigned int nr_entries) | |
720 | { | |
3852f676 | 721 | unsigned long begin = offset; |
38d8b4e6 HY |
722 | unsigned long end = offset + nr_entries - 1; |
723 | void (*swap_slot_free_notify)(struct block_device *, unsigned long); | |
724 | ||
725 | if (offset < si->lowest_bit) | |
726 | si->lowest_bit = offset; | |
727 | if (end > si->highest_bit) { | |
728 | bool was_full = !si->highest_bit; | |
729 | ||
a449bf58 | 730 | WRITE_ONCE(si->highest_bit, end); |
a2468cc9 AL |
731 | if (was_full && (si->flags & SWP_WRITEOK)) |
732 | add_to_avail_list(si); | |
38d8b4e6 HY |
733 | } |
734 | atomic_long_add(nr_entries, &nr_swap_pages); | |
c8945306 | 735 | WRITE_ONCE(si->inuse_pages, si->inuse_pages - nr_entries); |
38d8b4e6 HY |
736 | if (si->flags & SWP_BLKDEV) |
737 | swap_slot_free_notify = | |
738 | si->bdev->bd_disk->fops->swap_slot_free_notify; | |
739 | else | |
740 | swap_slot_free_notify = NULL; | |
741 | while (offset <= end) { | |
8a84802e | 742 | arch_swap_invalidate_page(si->type, offset); |
38d8b4e6 HY |
743 | frontswap_invalidate_page(si->type, offset); |
744 | if (swap_slot_free_notify) | |
745 | swap_slot_free_notify(si->bdev, offset); | |
746 | offset++; | |
747 | } | |
3852f676 | 748 | clear_shadow_from_swap_cache(si->type, begin, end); |
38d8b4e6 HY |
749 | } |
750 | ||
49070588 HY |
751 | static void set_cluster_next(struct swap_info_struct *si, unsigned long next) |
752 | { | |
753 | unsigned long prev; | |
754 | ||
755 | if (!(si->flags & SWP_SOLIDSTATE)) { | |
756 | si->cluster_next = next; | |
757 | return; | |
758 | } | |
759 | ||
760 | prev = this_cpu_read(*si->cluster_next_cpu); | |
761 | /* | |
762 | * Cross the swap address space size aligned trunk, choose | |
763 | * another trunk randomly to avoid lock contention on swap | |
764 | * address space if possible. | |
765 | */ | |
766 | if ((prev >> SWAP_ADDRESS_SPACE_SHIFT) != | |
767 | (next >> SWAP_ADDRESS_SPACE_SHIFT)) { | |
768 | /* No free swap slots available */ | |
769 | if (si->highest_bit <= si->lowest_bit) | |
770 | return; | |
771 | next = si->lowest_bit + | |
772 | prandom_u32_max(si->highest_bit - si->lowest_bit + 1); | |
773 | next = ALIGN_DOWN(next, SWAP_ADDRESS_SPACE_PAGES); | |
774 | next = max_t(unsigned int, next, si->lowest_bit); | |
775 | } | |
776 | this_cpu_write(*si->cluster_next_cpu, next); | |
777 | } | |
778 | ||
4b9ae842 ML |
779 | static bool swap_offset_available_and_locked(struct swap_info_struct *si, |
780 | unsigned long offset) | |
781 | { | |
782 | if (data_race(!si->swap_map[offset])) { | |
783 | spin_lock(&si->lock); | |
784 | return true; | |
785 | } | |
786 | ||
787 | if (vm_swap_full() && READ_ONCE(si->swap_map[offset]) == SWAP_HAS_CACHE) { | |
788 | spin_lock(&si->lock); | |
789 | return true; | |
790 | } | |
791 | ||
792 | return false; | |
793 | } | |
794 | ||
36005bae TC |
795 | static int scan_swap_map_slots(struct swap_info_struct *si, |
796 | unsigned char usage, int nr, | |
797 | swp_entry_t slots[]) | |
1da177e4 | 798 | { |
235b6217 | 799 | struct swap_cluster_info *ci; |
ebebbbe9 | 800 | unsigned long offset; |
c60aa176 | 801 | unsigned long scan_base; |
7992fde7 | 802 | unsigned long last_in_cluster = 0; |
048c27fd | 803 | int latency_ration = LATENCY_LIMIT; |
36005bae | 804 | int n_ret = 0; |
ed43af10 | 805 | bool scanned_many = false; |
36005bae | 806 | |
886bb7e9 | 807 | /* |
7dfad418 HD |
808 | * We try to cluster swap pages by allocating them sequentially |
809 | * in swap. Once we've allocated SWAPFILE_CLUSTER pages this | |
810 | * way, however, we resort to first-free allocation, starting | |
811 | * a new cluster. This prevents us from scattering swap pages | |
812 | * all over the entire swap partition, so that we reduce | |
813 | * overall disk seek times between swap pages. -- sct | |
814 | * But we do now try to find an empty cluster. -Andrea | |
c60aa176 | 815 | * And we let swap pages go all over an SSD partition. Hugh |
7dfad418 HD |
816 | */ |
817 | ||
52b7efdb | 818 | si->flags += SWP_SCANNING; |
49070588 HY |
819 | /* |
820 | * Use percpu scan base for SSD to reduce lock contention on | |
821 | * cluster and swap cache. For HDD, sequential access is more | |
822 | * important. | |
823 | */ | |
824 | if (si->flags & SWP_SOLIDSTATE) | |
825 | scan_base = this_cpu_read(*si->cluster_next_cpu); | |
826 | else | |
827 | scan_base = si->cluster_next; | |
828 | offset = scan_base; | |
ebebbbe9 | 829 | |
ebc2a1a6 SL |
830 | /* SSD algorithm */ |
831 | if (si->cluster_info) { | |
bd2d18da | 832 | if (!scan_swap_map_try_ssd_cluster(si, &offset, &scan_base)) |
36005bae | 833 | goto scan; |
f4eaf51a | 834 | } else if (unlikely(!si->cluster_nr--)) { |
ebebbbe9 HD |
835 | if (si->pages - si->inuse_pages < SWAPFILE_CLUSTER) { |
836 | si->cluster_nr = SWAPFILE_CLUSTER - 1; | |
837 | goto checks; | |
838 | } | |
2a8f9449 | 839 | |
ec8acf20 | 840 | spin_unlock(&si->lock); |
7dfad418 | 841 | |
c60aa176 HD |
842 | /* |
843 | * If seek is expensive, start searching for new cluster from | |
844 | * start of partition, to minimize the span of allocated swap. | |
50088c44 CY |
845 | * If seek is cheap, that is the SWP_SOLIDSTATE si->cluster_info |
846 | * case, just handled by scan_swap_map_try_ssd_cluster() above. | |
c60aa176 | 847 | */ |
50088c44 | 848 | scan_base = offset = si->lowest_bit; |
7dfad418 HD |
849 | last_in_cluster = offset + SWAPFILE_CLUSTER - 1; |
850 | ||
851 | /* Locate the first empty (unaligned) cluster */ | |
852 | for (; last_in_cluster <= si->highest_bit; offset++) { | |
1da177e4 | 853 | if (si->swap_map[offset]) |
7dfad418 HD |
854 | last_in_cluster = offset + SWAPFILE_CLUSTER; |
855 | else if (offset == last_in_cluster) { | |
ec8acf20 | 856 | spin_lock(&si->lock); |
ebebbbe9 HD |
857 | offset -= SWAPFILE_CLUSTER - 1; |
858 | si->cluster_next = offset; | |
859 | si->cluster_nr = SWAPFILE_CLUSTER - 1; | |
c60aa176 HD |
860 | goto checks; |
861 | } | |
862 | if (unlikely(--latency_ration < 0)) { | |
863 | cond_resched(); | |
864 | latency_ration = LATENCY_LIMIT; | |
865 | } | |
866 | } | |
867 | ||
868 | offset = scan_base; | |
ec8acf20 | 869 | spin_lock(&si->lock); |
ebebbbe9 | 870 | si->cluster_nr = SWAPFILE_CLUSTER - 1; |
1da177e4 | 871 | } |
7dfad418 | 872 | |
ebebbbe9 | 873 | checks: |
ebc2a1a6 | 874 | if (si->cluster_info) { |
36005bae TC |
875 | while (scan_swap_map_ssd_cluster_conflict(si, offset)) { |
876 | /* take a break if we already got some slots */ | |
877 | if (n_ret) | |
878 | goto done; | |
879 | if (!scan_swap_map_try_ssd_cluster(si, &offset, | |
880 | &scan_base)) | |
881 | goto scan; | |
882 | } | |
ebc2a1a6 | 883 | } |
ebebbbe9 | 884 | if (!(si->flags & SWP_WRITEOK)) |
52b7efdb | 885 | goto no_page; |
7dfad418 HD |
886 | if (!si->highest_bit) |
887 | goto no_page; | |
ebebbbe9 | 888 | if (offset > si->highest_bit) |
c60aa176 | 889 | scan_base = offset = si->lowest_bit; |
c9e44410 | 890 | |
235b6217 | 891 | ci = lock_cluster(si, offset); |
b73d7fce HD |
892 | /* reuse swap entry of cache-only swap if not busy. */ |
893 | if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) { | |
c9e44410 | 894 | int swap_was_freed; |
235b6217 | 895 | unlock_cluster(ci); |
ec8acf20 | 896 | spin_unlock(&si->lock); |
bcd49e86 | 897 | swap_was_freed = __try_to_reclaim_swap(si, offset, TTRS_ANYWAY); |
ec8acf20 | 898 | spin_lock(&si->lock); |
c9e44410 KH |
899 | /* entry was freed successfully, try to use this again */ |
900 | if (swap_was_freed) | |
901 | goto checks; | |
902 | goto scan; /* check next one */ | |
903 | } | |
904 | ||
235b6217 HY |
905 | if (si->swap_map[offset]) { |
906 | unlock_cluster(ci); | |
36005bae TC |
907 | if (!n_ret) |
908 | goto scan; | |
909 | else | |
910 | goto done; | |
235b6217 | 911 | } |
a449bf58 | 912 | WRITE_ONCE(si->swap_map[offset], usage); |
2872bb2d HY |
913 | inc_cluster_info_page(si, si->cluster_info, offset); |
914 | unlock_cluster(ci); | |
ebebbbe9 | 915 | |
38d8b4e6 | 916 | swap_range_alloc(si, offset, 1); |
36005bae TC |
917 | slots[n_ret++] = swp_entry(si->type, offset); |
918 | ||
919 | /* got enough slots or reach max slots? */ | |
920 | if ((n_ret == nr) || (offset >= si->highest_bit)) | |
921 | goto done; | |
922 | ||
923 | /* search for next available slot */ | |
924 | ||
925 | /* time to take a break? */ | |
926 | if (unlikely(--latency_ration < 0)) { | |
927 | if (n_ret) | |
928 | goto done; | |
929 | spin_unlock(&si->lock); | |
930 | cond_resched(); | |
931 | spin_lock(&si->lock); | |
932 | latency_ration = LATENCY_LIMIT; | |
933 | } | |
934 | ||
935 | /* try to get more slots in cluster */ | |
936 | if (si->cluster_info) { | |
937 | if (scan_swap_map_try_ssd_cluster(si, &offset, &scan_base)) | |
938 | goto checks; | |
f4eaf51a WY |
939 | } else if (si->cluster_nr && !si->swap_map[++offset]) { |
940 | /* non-ssd case, still more slots in cluster? */ | |
36005bae TC |
941 | --si->cluster_nr; |
942 | goto checks; | |
943 | } | |
7992fde7 | 944 | |
ed43af10 HY |
945 | /* |
946 | * Even if there's no free clusters available (fragmented), | |
947 | * try to scan a little more quickly with lock held unless we | |
948 | * have scanned too many slots already. | |
949 | */ | |
950 | if (!scanned_many) { | |
951 | unsigned long scan_limit; | |
952 | ||
953 | if (offset < scan_base) | |
954 | scan_limit = scan_base; | |
955 | else | |
956 | scan_limit = si->highest_bit; | |
957 | for (; offset <= scan_limit && --latency_ration > 0; | |
958 | offset++) { | |
959 | if (!si->swap_map[offset]) | |
960 | goto checks; | |
961 | } | |
962 | } | |
963 | ||
36005bae | 964 | done: |
49070588 | 965 | set_cluster_next(si, offset + 1); |
36005bae TC |
966 | si->flags -= SWP_SCANNING; |
967 | return n_ret; | |
7dfad418 | 968 | |
ebebbbe9 | 969 | scan: |
ec8acf20 | 970 | spin_unlock(&si->lock); |
a449bf58 | 971 | while (++offset <= READ_ONCE(si->highest_bit)) { |
4b9ae842 | 972 | if (swap_offset_available_and_locked(si, offset)) |
c9e44410 | 973 | goto checks; |
048c27fd HD |
974 | if (unlikely(--latency_ration < 0)) { |
975 | cond_resched(); | |
976 | latency_ration = LATENCY_LIMIT; | |
ed43af10 | 977 | scanned_many = true; |
048c27fd | 978 | } |
7dfad418 | 979 | } |
c60aa176 | 980 | offset = si->lowest_bit; |
a5998061 | 981 | while (offset < scan_base) { |
4b9ae842 | 982 | if (swap_offset_available_and_locked(si, offset)) |
c60aa176 | 983 | goto checks; |
c60aa176 HD |
984 | if (unlikely(--latency_ration < 0)) { |
985 | cond_resched(); | |
986 | latency_ration = LATENCY_LIMIT; | |
ed43af10 | 987 | scanned_many = true; |
c60aa176 | 988 | } |
a5998061 | 989 | offset++; |
c60aa176 | 990 | } |
ec8acf20 | 991 | spin_lock(&si->lock); |
7dfad418 HD |
992 | |
993 | no_page: | |
52b7efdb | 994 | si->flags -= SWP_SCANNING; |
36005bae | 995 | return n_ret; |
1da177e4 LT |
996 | } |
997 | ||
38d8b4e6 HY |
998 | static int swap_alloc_cluster(struct swap_info_struct *si, swp_entry_t *slot) |
999 | { | |
1000 | unsigned long idx; | |
1001 | struct swap_cluster_info *ci; | |
661c7566 | 1002 | unsigned long offset; |
38d8b4e6 | 1003 | |
fe5266d5 HY |
1004 | /* |
1005 | * Should not even be attempting cluster allocations when huge | |
1006 | * page swap is disabled. Warn and fail the allocation. | |
1007 | */ | |
1008 | if (!IS_ENABLED(CONFIG_THP_SWAP)) { | |
1009 | VM_WARN_ON_ONCE(1); | |
1010 | return 0; | |
1011 | } | |
1012 | ||
38d8b4e6 HY |
1013 | if (cluster_list_empty(&si->free_clusters)) |
1014 | return 0; | |
1015 | ||
1016 | idx = cluster_list_first(&si->free_clusters); | |
1017 | offset = idx * SWAPFILE_CLUSTER; | |
1018 | ci = lock_cluster(si, offset); | |
1019 | alloc_cluster(si, idx); | |
e0709829 | 1020 | cluster_set_count_flag(ci, SWAPFILE_CLUSTER, CLUSTER_FLAG_HUGE); |
38d8b4e6 | 1021 | |
661c7566 | 1022 | memset(si->swap_map + offset, SWAP_HAS_CACHE, SWAPFILE_CLUSTER); |
38d8b4e6 HY |
1023 | unlock_cluster(ci); |
1024 | swap_range_alloc(si, offset, SWAPFILE_CLUSTER); | |
1025 | *slot = swp_entry(si->type, offset); | |
1026 | ||
1027 | return 1; | |
1028 | } | |
1029 | ||
1030 | static void swap_free_cluster(struct swap_info_struct *si, unsigned long idx) | |
1031 | { | |
1032 | unsigned long offset = idx * SWAPFILE_CLUSTER; | |
1033 | struct swap_cluster_info *ci; | |
1034 | ||
1035 | ci = lock_cluster(si, offset); | |
979aafa5 | 1036 | memset(si->swap_map + offset, 0, SWAPFILE_CLUSTER); |
38d8b4e6 HY |
1037 | cluster_set_count_flag(ci, 0, 0); |
1038 | free_cluster(si, idx); | |
1039 | unlock_cluster(ci); | |
1040 | swap_range_free(si, offset, SWAPFILE_CLUSTER); | |
1041 | } | |
38d8b4e6 | 1042 | |
5d5e8f19 | 1043 | int get_swap_pages(int n_goal, swp_entry_t swp_entries[], int entry_size) |
1da177e4 | 1044 | { |
5d5e8f19 | 1045 | unsigned long size = swap_entry_size(entry_size); |
adfab836 | 1046 | struct swap_info_struct *si, *next; |
36005bae TC |
1047 | long avail_pgs; |
1048 | int n_ret = 0; | |
a2468cc9 | 1049 | int node; |
1da177e4 | 1050 | |
38d8b4e6 | 1051 | /* Only single cluster request supported */ |
5d5e8f19 | 1052 | WARN_ON_ONCE(n_goal > 1 && size == SWAPFILE_CLUSTER); |
38d8b4e6 | 1053 | |
b50da6e9 ZH |
1054 | spin_lock(&swap_avail_lock); |
1055 | ||
5d5e8f19 | 1056 | avail_pgs = atomic_long_read(&nr_swap_pages) / size; |
b50da6e9 ZH |
1057 | if (avail_pgs <= 0) { |
1058 | spin_unlock(&swap_avail_lock); | |
fb4f88dc | 1059 | goto noswap; |
b50da6e9 | 1060 | } |
36005bae | 1061 | |
08d3090f | 1062 | n_goal = min3((long)n_goal, (long)SWAP_BATCH, avail_pgs); |
36005bae | 1063 | |
5d5e8f19 | 1064 | atomic_long_sub(n_goal * size, &nr_swap_pages); |
fb4f88dc | 1065 | |
18ab4d4c | 1066 | start_over: |
a2468cc9 AL |
1067 | node = numa_node_id(); |
1068 | plist_for_each_entry_safe(si, next, &swap_avail_heads[node], avail_lists[node]) { | |
18ab4d4c | 1069 | /* requeue si to after same-priority siblings */ |
a2468cc9 | 1070 | plist_requeue(&si->avail_lists[node], &swap_avail_heads[node]); |
18ab4d4c | 1071 | spin_unlock(&swap_avail_lock); |
ec8acf20 | 1072 | spin_lock(&si->lock); |
adfab836 | 1073 | if (!si->highest_bit || !(si->flags & SWP_WRITEOK)) { |
18ab4d4c | 1074 | spin_lock(&swap_avail_lock); |
a2468cc9 | 1075 | if (plist_node_empty(&si->avail_lists[node])) { |
18ab4d4c DS |
1076 | spin_unlock(&si->lock); |
1077 | goto nextsi; | |
1078 | } | |
1079 | WARN(!si->highest_bit, | |
1080 | "swap_info %d in list but !highest_bit\n", | |
1081 | si->type); | |
1082 | WARN(!(si->flags & SWP_WRITEOK), | |
1083 | "swap_info %d in list but !SWP_WRITEOK\n", | |
1084 | si->type); | |
a2468cc9 | 1085 | __del_from_avail_list(si); |
ec8acf20 | 1086 | spin_unlock(&si->lock); |
18ab4d4c | 1087 | goto nextsi; |
ec8acf20 | 1088 | } |
5d5e8f19 | 1089 | if (size == SWAPFILE_CLUSTER) { |
41663430 | 1090 | if (si->flags & SWP_BLKDEV) |
f0eea189 HY |
1091 | n_ret = swap_alloc_cluster(si, swp_entries); |
1092 | } else | |
38d8b4e6 HY |
1093 | n_ret = scan_swap_map_slots(si, SWAP_HAS_CACHE, |
1094 | n_goal, swp_entries); | |
ec8acf20 | 1095 | spin_unlock(&si->lock); |
5d5e8f19 | 1096 | if (n_ret || size == SWAPFILE_CLUSTER) |
36005bae | 1097 | goto check_out; |
18ab4d4c | 1098 | pr_debug("scan_swap_map of si %d failed to find offset\n", |
36005bae TC |
1099 | si->type); |
1100 | ||
18ab4d4c DS |
1101 | spin_lock(&swap_avail_lock); |
1102 | nextsi: | |
adfab836 DS |
1103 | /* |
1104 | * if we got here, it's likely that si was almost full before, | |
bb243f7d ML |
1105 | * and since scan_swap_map_slots() can drop the si->lock, |
1106 | * multiple callers probably all tried to get a page from the | |
1107 | * same si and it filled up before we could get one; or, the si | |
1108 | * filled up between us dropping swap_avail_lock and taking | |
1109 | * si->lock. Since we dropped the swap_avail_lock, the | |
1110 | * swap_avail_head list may have been modified; so if next is | |
1111 | * still in the swap_avail_head list then try it, otherwise | |
1112 | * start over if we have not gotten any slots. | |
adfab836 | 1113 | */ |
a2468cc9 | 1114 | if (plist_node_empty(&next->avail_lists[node])) |
18ab4d4c | 1115 | goto start_over; |
1da177e4 | 1116 | } |
fb4f88dc | 1117 | |
18ab4d4c DS |
1118 | spin_unlock(&swap_avail_lock); |
1119 | ||
36005bae TC |
1120 | check_out: |
1121 | if (n_ret < n_goal) | |
5d5e8f19 | 1122 | atomic_long_add((long)(n_goal - n_ret) * size, |
38d8b4e6 | 1123 | &nr_swap_pages); |
fb4f88dc | 1124 | noswap: |
36005bae TC |
1125 | return n_ret; |
1126 | } | |
1127 | ||
afba72b1 | 1128 | static struct swap_info_struct *_swap_info_get(swp_entry_t entry) |
1da177e4 | 1129 | { |
73c34b6a | 1130 | struct swap_info_struct *p; |
eb085574 | 1131 | unsigned long offset; |
1da177e4 LT |
1132 | |
1133 | if (!entry.val) | |
1134 | goto out; | |
eb085574 | 1135 | p = swp_swap_info(entry); |
c10d38cc | 1136 | if (!p) |
1da177e4 | 1137 | goto bad_nofile; |
a449bf58 | 1138 | if (data_race(!(p->flags & SWP_USED))) |
1da177e4 LT |
1139 | goto bad_device; |
1140 | offset = swp_offset(entry); | |
1141 | if (offset >= p->max) | |
1142 | goto bad_offset; | |
afba72b1 ML |
1143 | if (data_race(!p->swap_map[swp_offset(entry)])) |
1144 | goto bad_free; | |
1da177e4 LT |
1145 | return p; |
1146 | ||
afba72b1 ML |
1147 | bad_free: |
1148 | pr_err("%s: %s%08lx\n", __func__, Unused_offset, entry.val); | |
1149 | goto out; | |
1da177e4 | 1150 | bad_offset: |
cf532faa | 1151 | pr_err("%s: %s%08lx\n", __func__, Bad_offset, entry.val); |
1da177e4 LT |
1152 | goto out; |
1153 | bad_device: | |
cf532faa | 1154 | pr_err("%s: %s%08lx\n", __func__, Unused_file, entry.val); |
1da177e4 LT |
1155 | goto out; |
1156 | bad_nofile: | |
cf532faa | 1157 | pr_err("%s: %s%08lx\n", __func__, Bad_file, entry.val); |
1da177e4 LT |
1158 | out: |
1159 | return NULL; | |
886bb7e9 | 1160 | } |
1da177e4 | 1161 | |
7c00bafe TC |
1162 | static struct swap_info_struct *swap_info_get_cont(swp_entry_t entry, |
1163 | struct swap_info_struct *q) | |
1164 | { | |
1165 | struct swap_info_struct *p; | |
1166 | ||
1167 | p = _swap_info_get(entry); | |
1168 | ||
1169 | if (p != q) { | |
1170 | if (q != NULL) | |
1171 | spin_unlock(&q->lock); | |
1172 | if (p != NULL) | |
1173 | spin_lock(&p->lock); | |
1174 | } | |
1175 | return p; | |
1176 | } | |
1177 | ||
b32d5f32 HY |
1178 | static unsigned char __swap_entry_free_locked(struct swap_info_struct *p, |
1179 | unsigned long offset, | |
1180 | unsigned char usage) | |
1da177e4 | 1181 | { |
8d69aaee HD |
1182 | unsigned char count; |
1183 | unsigned char has_cache; | |
235b6217 | 1184 | |
253d553b | 1185 | count = p->swap_map[offset]; |
235b6217 | 1186 | |
253d553b HD |
1187 | has_cache = count & SWAP_HAS_CACHE; |
1188 | count &= ~SWAP_HAS_CACHE; | |
355cfa73 | 1189 | |
253d553b | 1190 | if (usage == SWAP_HAS_CACHE) { |
355cfa73 | 1191 | VM_BUG_ON(!has_cache); |
253d553b | 1192 | has_cache = 0; |
aaa46865 HD |
1193 | } else if (count == SWAP_MAP_SHMEM) { |
1194 | /* | |
1195 | * Or we could insist on shmem.c using a special | |
1196 | * swap_shmem_free() and free_shmem_swap_and_cache()... | |
1197 | */ | |
1198 | count = 0; | |
570a335b HD |
1199 | } else if ((count & ~COUNT_CONTINUED) <= SWAP_MAP_MAX) { |
1200 | if (count == COUNT_CONTINUED) { | |
1201 | if (swap_count_continued(p, offset, count)) | |
1202 | count = SWAP_MAP_MAX | COUNT_CONTINUED; | |
1203 | else | |
1204 | count = SWAP_MAP_MAX; | |
1205 | } else | |
1206 | count--; | |
1207 | } | |
253d553b | 1208 | |
253d553b | 1209 | usage = count | has_cache; |
a449bf58 QC |
1210 | if (usage) |
1211 | WRITE_ONCE(p->swap_map[offset], usage); | |
1212 | else | |
1213 | WRITE_ONCE(p->swap_map[offset], SWAP_HAS_CACHE); | |
7c00bafe | 1214 | |
b32d5f32 HY |
1215 | return usage; |
1216 | } | |
1217 | ||
eb085574 HY |
1218 | /* |
1219 | * Check whether swap entry is valid in the swap device. If so, | |
1220 | * return pointer to swap_info_struct, and keep the swap entry valid | |
1221 | * via preventing the swap device from being swapoff, until | |
1222 | * put_swap_device() is called. Otherwise return NULL. | |
1223 | * | |
eb085574 | 1224 | * Notice that swapoff or swapoff+swapon can still happen before the |
63d8620e ML |
1225 | * percpu_ref_tryget_live() in get_swap_device() or after the |
1226 | * percpu_ref_put() in put_swap_device() if there isn't any other way | |
1227 | * to prevent swapoff, such as page lock, page table lock, etc. The | |
1228 | * caller must be prepared for that. For example, the following | |
1229 | * situation is possible. | |
eb085574 HY |
1230 | * |
1231 | * CPU1 CPU2 | |
1232 | * do_swap_page() | |
1233 | * ... swapoff+swapon | |
1234 | * __read_swap_cache_async() | |
1235 | * swapcache_prepare() | |
1236 | * __swap_duplicate() | |
1237 | * // check swap_map | |
1238 | * // verify PTE not changed | |
1239 | * | |
1240 | * In __swap_duplicate(), the swap_map need to be checked before | |
1241 | * changing partly because the specified swap entry may be for another | |
1242 | * swap device which has been swapoff. And in do_swap_page(), after | |
1243 | * the page is read from the swap device, the PTE is verified not | |
1244 | * changed with the page table locked to check whether the swap device | |
1245 | * has been swapoff or swapoff+swapon. | |
1246 | */ | |
1247 | struct swap_info_struct *get_swap_device(swp_entry_t entry) | |
1248 | { | |
1249 | struct swap_info_struct *si; | |
1250 | unsigned long offset; | |
1251 | ||
1252 | if (!entry.val) | |
1253 | goto out; | |
1254 | si = swp_swap_info(entry); | |
1255 | if (!si) | |
1256 | goto bad_nofile; | |
63d8620e ML |
1257 | if (!percpu_ref_tryget_live(&si->users)) |
1258 | goto out; | |
1259 | /* | |
1260 | * Guarantee the si->users are checked before accessing other | |
1261 | * fields of swap_info_struct. | |
1262 | * | |
1263 | * Paired with the spin_unlock() after setup_swap_info() in | |
1264 | * enable_swap_info(). | |
1265 | */ | |
1266 | smp_rmb(); | |
eb085574 HY |
1267 | offset = swp_offset(entry); |
1268 | if (offset >= si->max) | |
63d8620e | 1269 | goto put_out; |
eb085574 HY |
1270 | |
1271 | return si; | |
1272 | bad_nofile: | |
1273 | pr_err("%s: %s%08lx\n", __func__, Bad_file, entry.val); | |
1274 | out: | |
1275 | return NULL; | |
63d8620e | 1276 | put_out: |
23b230ba | 1277 | pr_err("%s: %s%08lx\n", __func__, Bad_offset, entry.val); |
63d8620e | 1278 | percpu_ref_put(&si->users); |
eb085574 HY |
1279 | return NULL; |
1280 | } | |
1281 | ||
b32d5f32 | 1282 | static unsigned char __swap_entry_free(struct swap_info_struct *p, |
33e16272 | 1283 | swp_entry_t entry) |
b32d5f32 HY |
1284 | { |
1285 | struct swap_cluster_info *ci; | |
1286 | unsigned long offset = swp_offset(entry); | |
33e16272 | 1287 | unsigned char usage; |
b32d5f32 HY |
1288 | |
1289 | ci = lock_cluster_or_swap_info(p, offset); | |
33e16272 | 1290 | usage = __swap_entry_free_locked(p, offset, 1); |
7c00bafe | 1291 | unlock_cluster_or_swap_info(p, ci); |
10e364da HY |
1292 | if (!usage) |
1293 | free_swap_slot(entry); | |
7c00bafe TC |
1294 | |
1295 | return usage; | |
1296 | } | |
355cfa73 | 1297 | |
7c00bafe TC |
1298 | static void swap_entry_free(struct swap_info_struct *p, swp_entry_t entry) |
1299 | { | |
1300 | struct swap_cluster_info *ci; | |
1301 | unsigned long offset = swp_offset(entry); | |
1302 | unsigned char count; | |
1303 | ||
1304 | ci = lock_cluster(p, offset); | |
1305 | count = p->swap_map[offset]; | |
1306 | VM_BUG_ON(count != SWAP_HAS_CACHE); | |
1307 | p->swap_map[offset] = 0; | |
1308 | dec_cluster_info_page(p, p->cluster_info, offset); | |
235b6217 HY |
1309 | unlock_cluster(ci); |
1310 | ||
38d8b4e6 HY |
1311 | mem_cgroup_uncharge_swap(entry, 1); |
1312 | swap_range_free(p, offset, 1); | |
1da177e4 LT |
1313 | } |
1314 | ||
1315 | /* | |
2de1a7e4 | 1316 | * Caller has made sure that the swap device corresponding to entry |
1da177e4 LT |
1317 | * is still around or has not been recycled. |
1318 | */ | |
1319 | void swap_free(swp_entry_t entry) | |
1320 | { | |
73c34b6a | 1321 | struct swap_info_struct *p; |
1da177e4 | 1322 | |
235b6217 | 1323 | p = _swap_info_get(entry); |
10e364da | 1324 | if (p) |
33e16272 | 1325 | __swap_entry_free(p, entry); |
1da177e4 LT |
1326 | } |
1327 | ||
cb4b86ba KH |
1328 | /* |
1329 | * Called after dropping swapcache to decrease refcnt to swap entries. | |
1330 | */ | |
a448f2d0 | 1331 | void put_swap_page(struct page *page, swp_entry_t entry) |
38d8b4e6 HY |
1332 | { |
1333 | unsigned long offset = swp_offset(entry); | |
1334 | unsigned long idx = offset / SWAPFILE_CLUSTER; | |
1335 | struct swap_cluster_info *ci; | |
1336 | struct swap_info_struct *si; | |
1337 | unsigned char *map; | |
a3aea839 HY |
1338 | unsigned int i, free_entries = 0; |
1339 | unsigned char val; | |
6c357848 | 1340 | int size = swap_entry_size(thp_nr_pages(page)); |
fe5266d5 | 1341 | |
a3aea839 | 1342 | si = _swap_info_get(entry); |
38d8b4e6 HY |
1343 | if (!si) |
1344 | return; | |
1345 | ||
c2343d27 | 1346 | ci = lock_cluster_or_swap_info(si, offset); |
a448f2d0 | 1347 | if (size == SWAPFILE_CLUSTER) { |
a448f2d0 HY |
1348 | VM_BUG_ON(!cluster_is_huge(ci)); |
1349 | map = si->swap_map + offset; | |
1350 | for (i = 0; i < SWAPFILE_CLUSTER; i++) { | |
1351 | val = map[i]; | |
1352 | VM_BUG_ON(!(val & SWAP_HAS_CACHE)); | |
1353 | if (val == SWAP_HAS_CACHE) | |
1354 | free_entries++; | |
1355 | } | |
a448f2d0 | 1356 | cluster_clear_huge(ci); |
a448f2d0 | 1357 | if (free_entries == SWAPFILE_CLUSTER) { |
c2343d27 | 1358 | unlock_cluster_or_swap_info(si, ci); |
a448f2d0 | 1359 | spin_lock(&si->lock); |
a448f2d0 HY |
1360 | mem_cgroup_uncharge_swap(entry, SWAPFILE_CLUSTER); |
1361 | swap_free_cluster(si, idx); | |
1362 | spin_unlock(&si->lock); | |
1363 | return; | |
1364 | } | |
1365 | } | |
c2343d27 HY |
1366 | for (i = 0; i < size; i++, entry.val++) { |
1367 | if (!__swap_entry_free_locked(si, offset + i, SWAP_HAS_CACHE)) { | |
1368 | unlock_cluster_or_swap_info(si, ci); | |
1369 | free_swap_slot(entry); | |
1370 | if (i == size - 1) | |
1371 | return; | |
1372 | lock_cluster_or_swap_info(si, offset); | |
a3aea839 HY |
1373 | } |
1374 | } | |
c2343d27 | 1375 | unlock_cluster_or_swap_info(si, ci); |
38d8b4e6 | 1376 | } |
59807685 | 1377 | |
fe5266d5 | 1378 | #ifdef CONFIG_THP_SWAP |
59807685 HY |
1379 | int split_swap_cluster(swp_entry_t entry) |
1380 | { | |
1381 | struct swap_info_struct *si; | |
1382 | struct swap_cluster_info *ci; | |
1383 | unsigned long offset = swp_offset(entry); | |
1384 | ||
1385 | si = _swap_info_get(entry); | |
1386 | if (!si) | |
1387 | return -EBUSY; | |
1388 | ci = lock_cluster(si, offset); | |
1389 | cluster_clear_huge(ci); | |
1390 | unlock_cluster(ci); | |
1391 | return 0; | |
1392 | } | |
fe5266d5 | 1393 | #endif |
38d8b4e6 | 1394 | |
155b5f88 HY |
1395 | static int swp_entry_cmp(const void *ent1, const void *ent2) |
1396 | { | |
1397 | const swp_entry_t *e1 = ent1, *e2 = ent2; | |
1398 | ||
1399 | return (int)swp_type(*e1) - (int)swp_type(*e2); | |
1400 | } | |
1401 | ||
7c00bafe TC |
1402 | void swapcache_free_entries(swp_entry_t *entries, int n) |
1403 | { | |
1404 | struct swap_info_struct *p, *prev; | |
1405 | int i; | |
1406 | ||
1407 | if (n <= 0) | |
1408 | return; | |
1409 | ||
1410 | prev = NULL; | |
1411 | p = NULL; | |
155b5f88 HY |
1412 | |
1413 | /* | |
1414 | * Sort swap entries by swap device, so each lock is only taken once. | |
1415 | * nr_swapfiles isn't absolutely correct, but the overhead of sort() is | |
1416 | * so low that it isn't necessary to optimize further. | |
1417 | */ | |
1418 | if (nr_swapfiles > 1) | |
1419 | sort(entries, n, sizeof(entries[0]), swp_entry_cmp, NULL); | |
7c00bafe TC |
1420 | for (i = 0; i < n; ++i) { |
1421 | p = swap_info_get_cont(entries[i], prev); | |
1422 | if (p) | |
1423 | swap_entry_free(p, entries[i]); | |
7c00bafe TC |
1424 | prev = p; |
1425 | } | |
235b6217 | 1426 | if (p) |
7c00bafe | 1427 | spin_unlock(&p->lock); |
cb4b86ba KH |
1428 | } |
1429 | ||
1da177e4 | 1430 | /* |
c475a8ab | 1431 | * How many references to page are currently swapped out? |
570a335b HD |
1432 | * This does not give an exact answer when swap count is continued, |
1433 | * but does include the high COUNT_CONTINUED flag to allow for that. | |
1da177e4 | 1434 | */ |
3db3264d | 1435 | static int page_swapcount(struct page *page) |
1da177e4 | 1436 | { |
c475a8ab HD |
1437 | int count = 0; |
1438 | struct swap_info_struct *p; | |
235b6217 | 1439 | struct swap_cluster_info *ci; |
1da177e4 | 1440 | swp_entry_t entry; |
235b6217 | 1441 | unsigned long offset; |
1da177e4 | 1442 | |
4c21e2f2 | 1443 | entry.val = page_private(page); |
235b6217 | 1444 | p = _swap_info_get(entry); |
1da177e4 | 1445 | if (p) { |
235b6217 HY |
1446 | offset = swp_offset(entry); |
1447 | ci = lock_cluster_or_swap_info(p, offset); | |
1448 | count = swap_count(p->swap_map[offset]); | |
1449 | unlock_cluster_or_swap_info(p, ci); | |
1da177e4 | 1450 | } |
c475a8ab | 1451 | return count; |
1da177e4 LT |
1452 | } |
1453 | ||
eb085574 | 1454 | int __swap_count(swp_entry_t entry) |
aa8d22a1 | 1455 | { |
eb085574 | 1456 | struct swap_info_struct *si; |
aa8d22a1 | 1457 | pgoff_t offset = swp_offset(entry); |
eb085574 | 1458 | int count = 0; |
aa8d22a1 | 1459 | |
eb085574 HY |
1460 | si = get_swap_device(entry); |
1461 | if (si) { | |
1462 | count = swap_count(si->swap_map[offset]); | |
1463 | put_swap_device(si); | |
1464 | } | |
1465 | return count; | |
aa8d22a1 MK |
1466 | } |
1467 | ||
322b8afe HY |
1468 | static int swap_swapcount(struct swap_info_struct *si, swp_entry_t entry) |
1469 | { | |
1470 | int count = 0; | |
1471 | pgoff_t offset = swp_offset(entry); | |
1472 | struct swap_cluster_info *ci; | |
1473 | ||
1474 | ci = lock_cluster_or_swap_info(si, offset); | |
1475 | count = swap_count(si->swap_map[offset]); | |
1476 | unlock_cluster_or_swap_info(si, ci); | |
1477 | return count; | |
1478 | } | |
1479 | ||
e8c26ab6 TC |
1480 | /* |
1481 | * How many references to @entry are currently swapped out? | |
1482 | * This does not give an exact answer when swap count is continued, | |
1483 | * but does include the high COUNT_CONTINUED flag to allow for that. | |
1484 | */ | |
1485 | int __swp_swapcount(swp_entry_t entry) | |
1486 | { | |
1487 | int count = 0; | |
e8c26ab6 | 1488 | struct swap_info_struct *si; |
e8c26ab6 | 1489 | |
eb085574 HY |
1490 | si = get_swap_device(entry); |
1491 | if (si) { | |
322b8afe | 1492 | count = swap_swapcount(si, entry); |
eb085574 HY |
1493 | put_swap_device(si); |
1494 | } | |
e8c26ab6 TC |
1495 | return count; |
1496 | } | |
1497 | ||
8334b962 MK |
1498 | /* |
1499 | * How many references to @entry are currently swapped out? | |
1500 | * This considers COUNT_CONTINUED so it returns exact answer. | |
1501 | */ | |
1502 | int swp_swapcount(swp_entry_t entry) | |
1503 | { | |
1504 | int count, tmp_count, n; | |
1505 | struct swap_info_struct *p; | |
235b6217 | 1506 | struct swap_cluster_info *ci; |
8334b962 MK |
1507 | struct page *page; |
1508 | pgoff_t offset; | |
1509 | unsigned char *map; | |
1510 | ||
235b6217 | 1511 | p = _swap_info_get(entry); |
8334b962 MK |
1512 | if (!p) |
1513 | return 0; | |
1514 | ||
235b6217 HY |
1515 | offset = swp_offset(entry); |
1516 | ||
1517 | ci = lock_cluster_or_swap_info(p, offset); | |
1518 | ||
1519 | count = swap_count(p->swap_map[offset]); | |
8334b962 MK |
1520 | if (!(count & COUNT_CONTINUED)) |
1521 | goto out; | |
1522 | ||
1523 | count &= ~COUNT_CONTINUED; | |
1524 | n = SWAP_MAP_MAX + 1; | |
1525 | ||
8334b962 MK |
1526 | page = vmalloc_to_page(p->swap_map + offset); |
1527 | offset &= ~PAGE_MASK; | |
1528 | VM_BUG_ON(page_private(page) != SWP_CONTINUED); | |
1529 | ||
1530 | do { | |
a8ae4991 | 1531 | page = list_next_entry(page, lru); |
8334b962 MK |
1532 | map = kmap_atomic(page); |
1533 | tmp_count = map[offset]; | |
1534 | kunmap_atomic(map); | |
1535 | ||
1536 | count += (tmp_count & ~COUNT_CONTINUED) * n; | |
1537 | n *= (SWAP_CONT_MAX + 1); | |
1538 | } while (tmp_count & COUNT_CONTINUED); | |
1539 | out: | |
235b6217 | 1540 | unlock_cluster_or_swap_info(p, ci); |
8334b962 MK |
1541 | return count; |
1542 | } | |
1543 | ||
e0709829 HY |
1544 | static bool swap_page_trans_huge_swapped(struct swap_info_struct *si, |
1545 | swp_entry_t entry) | |
1546 | { | |
1547 | struct swap_cluster_info *ci; | |
1548 | unsigned char *map = si->swap_map; | |
1549 | unsigned long roffset = swp_offset(entry); | |
1550 | unsigned long offset = round_down(roffset, SWAPFILE_CLUSTER); | |
1551 | int i; | |
1552 | bool ret = false; | |
1553 | ||
1554 | ci = lock_cluster_or_swap_info(si, offset); | |
1555 | if (!ci || !cluster_is_huge(ci)) { | |
afa4711e | 1556 | if (swap_count(map[roffset])) |
e0709829 HY |
1557 | ret = true; |
1558 | goto unlock_out; | |
1559 | } | |
1560 | for (i = 0; i < SWAPFILE_CLUSTER; i++) { | |
afa4711e | 1561 | if (swap_count(map[offset + i])) { |
e0709829 HY |
1562 | ret = true; |
1563 | break; | |
1564 | } | |
1565 | } | |
1566 | unlock_out: | |
1567 | unlock_cluster_or_swap_info(si, ci); | |
1568 | return ret; | |
1569 | } | |
1570 | ||
2397f780 | 1571 | static bool folio_swapped(struct folio *folio) |
e0709829 HY |
1572 | { |
1573 | swp_entry_t entry; | |
1574 | struct swap_info_struct *si; | |
1575 | ||
2397f780 MWO |
1576 | if (!IS_ENABLED(CONFIG_THP_SWAP) || likely(!folio_test_large(folio))) |
1577 | return page_swapcount(&folio->page) != 0; | |
e0709829 | 1578 | |
2397f780 | 1579 | entry = folio_swap_entry(folio); |
e0709829 HY |
1580 | si = _swap_info_get(entry); |
1581 | if (si) | |
1582 | return swap_page_trans_huge_swapped(si, entry); | |
1583 | return false; | |
1584 | } | |
ba3c4ce6 | 1585 | |
1da177e4 | 1586 | /* |
a2c43eed HD |
1587 | * If swap is getting full, or if there are no more mappings of this page, |
1588 | * then try_to_free_swap is called to free its swap space. | |
1da177e4 | 1589 | */ |
a2c43eed | 1590 | int try_to_free_swap(struct page *page) |
1da177e4 | 1591 | { |
2397f780 MWO |
1592 | struct folio *folio = page_folio(page); |
1593 | VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio); | |
1da177e4 | 1594 | |
2397f780 | 1595 | if (!folio_test_swapcache(folio)) |
1da177e4 | 1596 | return 0; |
2397f780 | 1597 | if (folio_test_writeback(folio)) |
1da177e4 | 1598 | return 0; |
2397f780 | 1599 | if (folio_swapped(folio)) |
1da177e4 LT |
1600 | return 0; |
1601 | ||
b73d7fce HD |
1602 | /* |
1603 | * Once hibernation has begun to create its image of memory, | |
1604 | * there's a danger that one of the calls to try_to_free_swap() | |
1605 | * - most probably a call from __try_to_reclaim_swap() while | |
1606 | * hibernation is allocating its own swap pages for the image, | |
1607 | * but conceivably even a call from memory reclaim - will free | |
1608 | * the swap from a page which has already been recorded in the | |
1609 | * image as a clean swapcache page, and then reuse its swap for | |
1610 | * another page of the image. On waking from hibernation, the | |
1611 | * original page might be freed under memory pressure, then | |
1612 | * later read back in from swap, now with the wrong data. | |
1613 | * | |
2de1a7e4 | 1614 | * Hibernation suspends storage while it is writing the image |
f90ac398 | 1615 | * to disk so check that here. |
b73d7fce | 1616 | */ |
f90ac398 | 1617 | if (pm_suspended_storage()) |
b73d7fce HD |
1618 | return 0; |
1619 | ||
75fa68a5 | 1620 | delete_from_swap_cache(folio); |
2397f780 | 1621 | folio_set_dirty(folio); |
a2c43eed | 1622 | return 1; |
68a22394 RR |
1623 | } |
1624 | ||
1da177e4 LT |
1625 | /* |
1626 | * Free the swap entry like above, but also try to | |
1627 | * free the page cache entry if it is the last user. | |
1628 | */ | |
2509ef26 | 1629 | int free_swap_and_cache(swp_entry_t entry) |
1da177e4 | 1630 | { |
2509ef26 | 1631 | struct swap_info_struct *p; |
7c00bafe | 1632 | unsigned char count; |
1da177e4 | 1633 | |
a7420aa5 | 1634 | if (non_swap_entry(entry)) |
2509ef26 | 1635 | return 1; |
0697212a | 1636 | |
7c00bafe | 1637 | p = _swap_info_get(entry); |
1da177e4 | 1638 | if (p) { |
33e16272 | 1639 | count = __swap_entry_free(p, entry); |
e0709829 | 1640 | if (count == SWAP_HAS_CACHE && |
bcd49e86 HY |
1641 | !swap_page_trans_huge_swapped(p, entry)) |
1642 | __try_to_reclaim_swap(p, swp_offset(entry), | |
1643 | TTRS_UNMAPPED | TTRS_FULL); | |
1da177e4 | 1644 | } |
2509ef26 | 1645 | return p != NULL; |
1da177e4 LT |
1646 | } |
1647 | ||
b0cb1a19 | 1648 | #ifdef CONFIG_HIBERNATION |
bb243f7d ML |
1649 | |
1650 | swp_entry_t get_swap_page_of_type(int type) | |
1651 | { | |
1652 | struct swap_info_struct *si = swap_type_to_swap_info(type); | |
1653 | swp_entry_t entry = {0}; | |
1654 | ||
1655 | if (!si) | |
1656 | goto fail; | |
1657 | ||
1658 | /* This is called for allocating swap entry, not cache */ | |
1659 | spin_lock(&si->lock); | |
1660 | if ((si->flags & SWP_WRITEOK) && scan_swap_map_slots(si, 1, 1, &entry)) | |
1661 | atomic_long_dec(&nr_swap_pages); | |
1662 | spin_unlock(&si->lock); | |
1663 | fail: | |
1664 | return entry; | |
1665 | } | |
1666 | ||
f577eb30 | 1667 | /* |
915bae9e | 1668 | * Find the swap type that corresponds to given device (if any). |
f577eb30 | 1669 | * |
915bae9e RW |
1670 | * @offset - number of the PAGE_SIZE-sized block of the device, starting |
1671 | * from 0, in which the swap header is expected to be located. | |
1672 | * | |
1673 | * This is needed for the suspend to disk (aka swsusp). | |
f577eb30 | 1674 | */ |
21bd9005 | 1675 | int swap_type_of(dev_t device, sector_t offset) |
f577eb30 | 1676 | { |
efa90a98 | 1677 | int type; |
f577eb30 | 1678 | |
21bd9005 CH |
1679 | if (!device) |
1680 | return -1; | |
915bae9e | 1681 | |
f577eb30 | 1682 | spin_lock(&swap_lock); |
efa90a98 HD |
1683 | for (type = 0; type < nr_swapfiles; type++) { |
1684 | struct swap_info_struct *sis = swap_info[type]; | |
f577eb30 | 1685 | |
915bae9e | 1686 | if (!(sis->flags & SWP_WRITEOK)) |
f577eb30 | 1687 | continue; |
b6b5bce3 | 1688 | |
21bd9005 | 1689 | if (device == sis->bdev->bd_dev) { |
4efaceb1 | 1690 | struct swap_extent *se = first_se(sis); |
915bae9e | 1691 | |
915bae9e RW |
1692 | if (se->start_block == offset) { |
1693 | spin_unlock(&swap_lock); | |
efa90a98 | 1694 | return type; |
915bae9e | 1695 | } |
f577eb30 RW |
1696 | } |
1697 | } | |
1698 | spin_unlock(&swap_lock); | |
21bd9005 CH |
1699 | return -ENODEV; |
1700 | } | |
915bae9e | 1701 | |
21bd9005 CH |
1702 | int find_first_swap(dev_t *device) |
1703 | { | |
1704 | int type; | |
915bae9e | 1705 | |
21bd9005 CH |
1706 | spin_lock(&swap_lock); |
1707 | for (type = 0; type < nr_swapfiles; type++) { | |
1708 | struct swap_info_struct *sis = swap_info[type]; | |
1709 | ||
1710 | if (!(sis->flags & SWP_WRITEOK)) | |
1711 | continue; | |
1712 | *device = sis->bdev->bd_dev; | |
1713 | spin_unlock(&swap_lock); | |
1714 | return type; | |
1715 | } | |
1716 | spin_unlock(&swap_lock); | |
f577eb30 RW |
1717 | return -ENODEV; |
1718 | } | |
1719 | ||
73c34b6a HD |
1720 | /* |
1721 | * Get the (PAGE_SIZE) block corresponding to given offset on the swapdev | |
1722 | * corresponding to given index in swap_info (swap type). | |
1723 | */ | |
1724 | sector_t swapdev_block(int type, pgoff_t offset) | |
1725 | { | |
c10d38cc | 1726 | struct swap_info_struct *si = swap_type_to_swap_info(type); |
f885056a | 1727 | struct swap_extent *se; |
73c34b6a | 1728 | |
c10d38cc | 1729 | if (!si || !(si->flags & SWP_WRITEOK)) |
73c34b6a | 1730 | return 0; |
f885056a CH |
1731 | se = offset_to_swap_extent(si, offset); |
1732 | return se->start_block + (offset - se->start_page); | |
73c34b6a HD |
1733 | } |
1734 | ||
f577eb30 RW |
1735 | /* |
1736 | * Return either the total number of swap pages of given type, or the number | |
1737 | * of free pages of that type (depending on @free) | |
1738 | * | |
1739 | * This is needed for software suspend | |
1740 | */ | |
1741 | unsigned int count_swap_pages(int type, int free) | |
1742 | { | |
1743 | unsigned int n = 0; | |
1744 | ||
efa90a98 HD |
1745 | spin_lock(&swap_lock); |
1746 | if ((unsigned int)type < nr_swapfiles) { | |
1747 | struct swap_info_struct *sis = swap_info[type]; | |
1748 | ||
ec8acf20 | 1749 | spin_lock(&sis->lock); |
efa90a98 HD |
1750 | if (sis->flags & SWP_WRITEOK) { |
1751 | n = sis->pages; | |
f577eb30 | 1752 | if (free) |
efa90a98 | 1753 | n -= sis->inuse_pages; |
f577eb30 | 1754 | } |
ec8acf20 | 1755 | spin_unlock(&sis->lock); |
f577eb30 | 1756 | } |
efa90a98 | 1757 | spin_unlock(&swap_lock); |
f577eb30 RW |
1758 | return n; |
1759 | } | |
73c34b6a | 1760 | #endif /* CONFIG_HIBERNATION */ |
f577eb30 | 1761 | |
9f8bdb3f | 1762 | static inline int pte_same_as_swp(pte_t pte, pte_t swp_pte) |
179ef71c | 1763 | { |
099dd687 | 1764 | return pte_same(pte_swp_clear_flags(pte), swp_pte); |
179ef71c CG |
1765 | } |
1766 | ||
1da177e4 | 1767 | /* |
72866f6f HD |
1768 | * No need to decide whether this PTE shares the swap entry with others, |
1769 | * just let do_wp_page work it out if a write is requested later - to | |
1770 | * force COW, vm_page_prot omits write permission from any private vma. | |
1da177e4 | 1771 | */ |
044d66c1 | 1772 | static int unuse_pte(struct vm_area_struct *vma, pmd_t *pmd, |
1da177e4 LT |
1773 | unsigned long addr, swp_entry_t entry, struct page *page) |
1774 | { | |
9e16b7fb | 1775 | struct page *swapcache; |
044d66c1 | 1776 | spinlock_t *ptl; |
14a762dd | 1777 | pte_t *pte, new_pte; |
044d66c1 HD |
1778 | int ret = 1; |
1779 | ||
9e16b7fb HD |
1780 | swapcache = page; |
1781 | page = ksm_might_need_to_copy(page, vma, addr); | |
1782 | if (unlikely(!page)) | |
1783 | return -ENOMEM; | |
1784 | ||
044d66c1 | 1785 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
9f8bdb3f | 1786 | if (unlikely(!pte_same_as_swp(*pte, swp_entry_to_pte(entry)))) { |
044d66c1 HD |
1787 | ret = 0; |
1788 | goto out; | |
1789 | } | |
8a9f3ccd | 1790 | |
9f186f9e ML |
1791 | if (unlikely(!PageUptodate(page))) { |
1792 | pte_t pteval; | |
1793 | ||
1794 | dec_mm_counter(vma->vm_mm, MM_SWAPENTS); | |
1795 | pteval = swp_entry_to_pte(make_swapin_error_entry(page)); | |
1796 | set_pte_at(vma->vm_mm, addr, pte, pteval); | |
1797 | swap_free(entry); | |
1798 | ret = 0; | |
1799 | goto out; | |
1800 | } | |
1801 | ||
78fbe906 DH |
1802 | /* See do_swap_page() */ |
1803 | BUG_ON(!PageAnon(page) && PageMappedToDisk(page)); | |
1804 | BUG_ON(PageAnon(page) && PageAnonExclusive(page)); | |
1805 | ||
b084d435 | 1806 | dec_mm_counter(vma->vm_mm, MM_SWAPENTS); |
d559db08 | 1807 | inc_mm_counter(vma->vm_mm, MM_ANONPAGES); |
1da177e4 | 1808 | get_page(page); |
00501b53 | 1809 | if (page == swapcache) { |
1493a191 DH |
1810 | rmap_t rmap_flags = RMAP_NONE; |
1811 | ||
1812 | /* | |
1813 | * See do_swap_page(): PageWriteback() would be problematic. | |
1814 | * However, we do a wait_on_page_writeback() just before this | |
1815 | * call and have the page locked. | |
1816 | */ | |
1817 | VM_BUG_ON_PAGE(PageWriteback(page), page); | |
1818 | if (pte_swp_exclusive(*pte)) | |
1819 | rmap_flags |= RMAP_EXCLUSIVE; | |
1820 | ||
1821 | page_add_anon_rmap(page, vma, addr, rmap_flags); | |
00501b53 | 1822 | } else { /* ksm created a completely new copy */ |
40f2bbf7 | 1823 | page_add_new_anon_rmap(page, vma, addr); |
b518154e | 1824 | lru_cache_add_inactive_or_unevictable(page, vma); |
00501b53 | 1825 | } |
14a762dd ML |
1826 | new_pte = pte_mkold(mk_pte(page, vma->vm_page_prot)); |
1827 | if (pte_swp_soft_dirty(*pte)) | |
1828 | new_pte = pte_mksoft_dirty(new_pte); | |
1829 | if (pte_swp_uffd_wp(*pte)) | |
1830 | new_pte = pte_mkuffd_wp(new_pte); | |
1831 | set_pte_at(vma->vm_mm, addr, pte, new_pte); | |
1da177e4 | 1832 | swap_free(entry); |
044d66c1 HD |
1833 | out: |
1834 | pte_unmap_unlock(pte, ptl); | |
9e16b7fb HD |
1835 | if (page != swapcache) { |
1836 | unlock_page(page); | |
1837 | put_page(page); | |
1838 | } | |
044d66c1 | 1839 | return ret; |
1da177e4 LT |
1840 | } |
1841 | ||
1842 | static int unuse_pte_range(struct vm_area_struct *vma, pmd_t *pmd, | |
b56a2d8a | 1843 | unsigned long addr, unsigned long end, |
10a9c496 | 1844 | unsigned int type) |
1da177e4 | 1845 | { |
b56a2d8a VRP |
1846 | struct page *page; |
1847 | swp_entry_t entry; | |
705e87c0 | 1848 | pte_t *pte; |
b56a2d8a VRP |
1849 | struct swap_info_struct *si; |
1850 | unsigned long offset; | |
8a9f3ccd | 1851 | int ret = 0; |
b56a2d8a | 1852 | volatile unsigned char *swap_map; |
1da177e4 | 1853 | |
b56a2d8a | 1854 | si = swap_info[type]; |
044d66c1 | 1855 | pte = pte_offset_map(pmd, addr); |
1da177e4 | 1856 | do { |
b56a2d8a VRP |
1857 | if (!is_swap_pte(*pte)) |
1858 | continue; | |
1859 | ||
1860 | entry = pte_to_swp_entry(*pte); | |
1861 | if (swp_type(entry) != type) | |
1862 | continue; | |
1863 | ||
1864 | offset = swp_offset(entry); | |
b56a2d8a VRP |
1865 | pte_unmap(pte); |
1866 | swap_map = &si->swap_map[offset]; | |
ebc5951e AR |
1867 | page = lookup_swap_cache(entry, vma, addr); |
1868 | if (!page) { | |
8c63ca5b WD |
1869 | struct vm_fault vmf = { |
1870 | .vma = vma, | |
1871 | .address = addr, | |
824ddc60 | 1872 | .real_address = addr, |
8c63ca5b WD |
1873 | .pmd = pmd, |
1874 | }; | |
1875 | ||
ebc5951e AR |
1876 | page = swapin_readahead(entry, GFP_HIGHUSER_MOVABLE, |
1877 | &vmf); | |
1878 | } | |
b56a2d8a VRP |
1879 | if (!page) { |
1880 | if (*swap_map == 0 || *swap_map == SWAP_MAP_BAD) | |
1881 | goto try_next; | |
1882 | return -ENOMEM; | |
1883 | } | |
1884 | ||
1885 | lock_page(page); | |
1886 | wait_on_page_writeback(page); | |
1887 | ret = unuse_pte(vma, pmd, addr, entry, page); | |
1888 | if (ret < 0) { | |
1889 | unlock_page(page); | |
1890 | put_page(page); | |
1891 | goto out; | |
1892 | } | |
1893 | ||
1894 | try_to_free_swap(page); | |
1895 | unlock_page(page); | |
1896 | put_page(page); | |
b56a2d8a VRP |
1897 | try_next: |
1898 | pte = pte_offset_map(pmd, addr); | |
1da177e4 | 1899 | } while (pte++, addr += PAGE_SIZE, addr != end); |
044d66c1 | 1900 | pte_unmap(pte - 1); |
b56a2d8a VRP |
1901 | |
1902 | ret = 0; | |
044d66c1 | 1903 | out: |
8a9f3ccd | 1904 | return ret; |
1da177e4 LT |
1905 | } |
1906 | ||
1907 | static inline int unuse_pmd_range(struct vm_area_struct *vma, pud_t *pud, | |
1908 | unsigned long addr, unsigned long end, | |
10a9c496 | 1909 | unsigned int type) |
1da177e4 LT |
1910 | { |
1911 | pmd_t *pmd; | |
1912 | unsigned long next; | |
8a9f3ccd | 1913 | int ret; |
1da177e4 LT |
1914 | |
1915 | pmd = pmd_offset(pud, addr); | |
1916 | do { | |
dc644a07 | 1917 | cond_resched(); |
1da177e4 | 1918 | next = pmd_addr_end(addr, end); |
1a5a9906 | 1919 | if (pmd_none_or_trans_huge_or_clear_bad(pmd)) |
1da177e4 | 1920 | continue; |
10a9c496 | 1921 | ret = unuse_pte_range(vma, pmd, addr, next, type); |
8a9f3ccd BS |
1922 | if (ret) |
1923 | return ret; | |
1da177e4 LT |
1924 | } while (pmd++, addr = next, addr != end); |
1925 | return 0; | |
1926 | } | |
1927 | ||
c2febafc | 1928 | static inline int unuse_pud_range(struct vm_area_struct *vma, p4d_t *p4d, |
1da177e4 | 1929 | unsigned long addr, unsigned long end, |
10a9c496 | 1930 | unsigned int type) |
1da177e4 LT |
1931 | { |
1932 | pud_t *pud; | |
1933 | unsigned long next; | |
8a9f3ccd | 1934 | int ret; |
1da177e4 | 1935 | |
c2febafc | 1936 | pud = pud_offset(p4d, addr); |
1da177e4 LT |
1937 | do { |
1938 | next = pud_addr_end(addr, end); | |
1939 | if (pud_none_or_clear_bad(pud)) | |
1940 | continue; | |
10a9c496 | 1941 | ret = unuse_pmd_range(vma, pud, addr, next, type); |
8a9f3ccd BS |
1942 | if (ret) |
1943 | return ret; | |
1da177e4 LT |
1944 | } while (pud++, addr = next, addr != end); |
1945 | return 0; | |
1946 | } | |
1947 | ||
c2febafc KS |
1948 | static inline int unuse_p4d_range(struct vm_area_struct *vma, pgd_t *pgd, |
1949 | unsigned long addr, unsigned long end, | |
10a9c496 | 1950 | unsigned int type) |
c2febafc KS |
1951 | { |
1952 | p4d_t *p4d; | |
1953 | unsigned long next; | |
1954 | int ret; | |
1955 | ||
1956 | p4d = p4d_offset(pgd, addr); | |
1957 | do { | |
1958 | next = p4d_addr_end(addr, end); | |
1959 | if (p4d_none_or_clear_bad(p4d)) | |
1960 | continue; | |
10a9c496 | 1961 | ret = unuse_pud_range(vma, p4d, addr, next, type); |
c2febafc KS |
1962 | if (ret) |
1963 | return ret; | |
1964 | } while (p4d++, addr = next, addr != end); | |
1965 | return 0; | |
1966 | } | |
1967 | ||
10a9c496 | 1968 | static int unuse_vma(struct vm_area_struct *vma, unsigned int type) |
1da177e4 LT |
1969 | { |
1970 | pgd_t *pgd; | |
1971 | unsigned long addr, end, next; | |
8a9f3ccd | 1972 | int ret; |
1da177e4 | 1973 | |
b56a2d8a VRP |
1974 | addr = vma->vm_start; |
1975 | end = vma->vm_end; | |
1da177e4 LT |
1976 | |
1977 | pgd = pgd_offset(vma->vm_mm, addr); | |
1978 | do { | |
1979 | next = pgd_addr_end(addr, end); | |
1980 | if (pgd_none_or_clear_bad(pgd)) | |
1981 | continue; | |
10a9c496 | 1982 | ret = unuse_p4d_range(vma, pgd, addr, next, type); |
8a9f3ccd BS |
1983 | if (ret) |
1984 | return ret; | |
1da177e4 LT |
1985 | } while (pgd++, addr = next, addr != end); |
1986 | return 0; | |
1987 | } | |
1988 | ||
10a9c496 | 1989 | static int unuse_mm(struct mm_struct *mm, unsigned int type) |
1da177e4 LT |
1990 | { |
1991 | struct vm_area_struct *vma; | |
8a9f3ccd | 1992 | int ret = 0; |
1da177e4 | 1993 | |
d8ed45c5 | 1994 | mmap_read_lock(mm); |
1da177e4 | 1995 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
b56a2d8a | 1996 | if (vma->anon_vma) { |
10a9c496 | 1997 | ret = unuse_vma(vma, type); |
b56a2d8a VRP |
1998 | if (ret) |
1999 | break; | |
2000 | } | |
dc644a07 | 2001 | cond_resched(); |
1da177e4 | 2002 | } |
d8ed45c5 | 2003 | mmap_read_unlock(mm); |
b56a2d8a | 2004 | return ret; |
1da177e4 LT |
2005 | } |
2006 | ||
2007 | /* | |
3c3115ad ML |
2008 | * Scan swap_map from current position to next entry still in use. |
2009 | * Return 0 if there are no inuse entries after prev till end of | |
2010 | * the map. | |
1da177e4 | 2011 | */ |
6eb396dc | 2012 | static unsigned int find_next_to_unuse(struct swap_info_struct *si, |
10a9c496 | 2013 | unsigned int prev) |
1da177e4 | 2014 | { |
b56a2d8a | 2015 | unsigned int i; |
8d69aaee | 2016 | unsigned char count; |
1da177e4 LT |
2017 | |
2018 | /* | |
5d337b91 | 2019 | * No need for swap_lock here: we're just looking |
1da177e4 LT |
2020 | * for whether an entry is in use, not modifying it; false |
2021 | * hits are okay, and sys_swapoff() has already prevented new | |
5d337b91 | 2022 | * allocations from this area (while holding swap_lock). |
1da177e4 | 2023 | */ |
b56a2d8a | 2024 | for (i = prev + 1; i < si->max; i++) { |
4db0c3c2 | 2025 | count = READ_ONCE(si->swap_map[i]); |
355cfa73 | 2026 | if (count && swap_count(count) != SWAP_MAP_BAD) |
10a9c496 | 2027 | break; |
dc644a07 HD |
2028 | if ((i % LATENCY_LIMIT) == 0) |
2029 | cond_resched(); | |
1da177e4 | 2030 | } |
b56a2d8a VRP |
2031 | |
2032 | if (i == si->max) | |
2033 | i = 0; | |
2034 | ||
1da177e4 LT |
2035 | return i; |
2036 | } | |
2037 | ||
10a9c496 | 2038 | static int try_to_unuse(unsigned int type) |
1da177e4 | 2039 | { |
b56a2d8a VRP |
2040 | struct mm_struct *prev_mm; |
2041 | struct mm_struct *mm; | |
2042 | struct list_head *p; | |
2043 | int retval = 0; | |
efa90a98 | 2044 | struct swap_info_struct *si = swap_info[type]; |
1da177e4 LT |
2045 | struct page *page; |
2046 | swp_entry_t entry; | |
b56a2d8a | 2047 | unsigned int i; |
1da177e4 | 2048 | |
21820948 | 2049 | if (!READ_ONCE(si->inuse_pages)) |
b56a2d8a | 2050 | return 0; |
1da177e4 | 2051 | |
b56a2d8a | 2052 | retry: |
10a9c496 | 2053 | retval = shmem_unuse(type); |
b56a2d8a | 2054 | if (retval) |
10a9c496 | 2055 | return retval; |
b56a2d8a VRP |
2056 | |
2057 | prev_mm = &init_mm; | |
2058 | mmget(prev_mm); | |
2059 | ||
2060 | spin_lock(&mmlist_lock); | |
2061 | p = &init_mm.mmlist; | |
21820948 | 2062 | while (READ_ONCE(si->inuse_pages) && |
64165b1a HD |
2063 | !signal_pending(current) && |
2064 | (p = p->next) != &init_mm.mmlist) { | |
1da177e4 | 2065 | |
b56a2d8a VRP |
2066 | mm = list_entry(p, struct mm_struct, mmlist); |
2067 | if (!mmget_not_zero(mm)) | |
2068 | continue; | |
2069 | spin_unlock(&mmlist_lock); | |
2070 | mmput(prev_mm); | |
2071 | prev_mm = mm; | |
10a9c496 | 2072 | retval = unuse_mm(mm, type); |
b56a2d8a VRP |
2073 | if (retval) { |
2074 | mmput(prev_mm); | |
10a9c496 | 2075 | return retval; |
1da177e4 LT |
2076 | } |
2077 | ||
2078 | /* | |
b56a2d8a VRP |
2079 | * Make sure that we aren't completely killing |
2080 | * interactive performance. | |
1da177e4 | 2081 | */ |
b56a2d8a VRP |
2082 | cond_resched(); |
2083 | spin_lock(&mmlist_lock); | |
2084 | } | |
2085 | spin_unlock(&mmlist_lock); | |
1da177e4 | 2086 | |
b56a2d8a | 2087 | mmput(prev_mm); |
1da177e4 | 2088 | |
b56a2d8a | 2089 | i = 0; |
21820948 | 2090 | while (READ_ONCE(si->inuse_pages) && |
64165b1a | 2091 | !signal_pending(current) && |
10a9c496 | 2092 | (i = find_next_to_unuse(si, i)) != 0) { |
1da177e4 | 2093 | |
b56a2d8a VRP |
2094 | entry = swp_entry(type, i); |
2095 | page = find_get_page(swap_address_space(entry), i); | |
2096 | if (!page) | |
2097 | continue; | |
68bdc8d6 HD |
2098 | |
2099 | /* | |
2100 | * It is conceivable that a racing task removed this page from | |
b56a2d8a VRP |
2101 | * swap cache just before we acquired the page lock. The page |
2102 | * might even be back in swap cache on another swap area. But | |
2103 | * that is okay, try_to_free_swap() only removes stale pages. | |
1da177e4 | 2104 | */ |
b56a2d8a VRP |
2105 | lock_page(page); |
2106 | wait_on_page_writeback(page); | |
2107 | try_to_free_swap(page); | |
1da177e4 | 2108 | unlock_page(page); |
09cbfeaf | 2109 | put_page(page); |
1da177e4 LT |
2110 | } |
2111 | ||
b56a2d8a VRP |
2112 | /* |
2113 | * Lets check again to see if there are still swap entries in the map. | |
2114 | * If yes, we would need to do retry the unuse logic again. | |
2115 | * Under global memory pressure, swap entries can be reinserted back | |
2116 | * into process space after the mmlist loop above passes over them. | |
dd862deb | 2117 | * |
e2e3fdc7 MWO |
2118 | * Limit the number of retries? No: when mmget_not_zero() |
2119 | * above fails, that mm is likely to be freeing swap from | |
2120 | * exit_mmap(), which proceeds at its own independent pace; | |
2121 | * and even shmem_writepage() could have been preempted after | |
2122 | * folio_alloc_swap(), temporarily hiding that swap. It's easy | |
2123 | * and robust (though cpu-intensive) just to keep retrying. | |
b56a2d8a | 2124 | */ |
21820948 | 2125 | if (READ_ONCE(si->inuse_pages)) { |
64165b1a HD |
2126 | if (!signal_pending(current)) |
2127 | goto retry; | |
10a9c496 | 2128 | return -EINTR; |
64165b1a | 2129 | } |
10a9c496 CH |
2130 | |
2131 | return 0; | |
1da177e4 LT |
2132 | } |
2133 | ||
2134 | /* | |
5d337b91 HD |
2135 | * After a successful try_to_unuse, if no swap is now in use, we know |
2136 | * we can empty the mmlist. swap_lock must be held on entry and exit. | |
2137 | * Note that mmlist_lock nests inside swap_lock, and an mm must be | |
1da177e4 LT |
2138 | * added to the mmlist just after page_duplicate - before would be racy. |
2139 | */ | |
2140 | static void drain_mmlist(void) | |
2141 | { | |
2142 | struct list_head *p, *next; | |
efa90a98 | 2143 | unsigned int type; |
1da177e4 | 2144 | |
efa90a98 HD |
2145 | for (type = 0; type < nr_swapfiles; type++) |
2146 | if (swap_info[type]->inuse_pages) | |
1da177e4 LT |
2147 | return; |
2148 | spin_lock(&mmlist_lock); | |
2149 | list_for_each_safe(p, next, &init_mm.mmlist) | |
2150 | list_del_init(p); | |
2151 | spin_unlock(&mmlist_lock); | |
2152 | } | |
2153 | ||
1da177e4 LT |
2154 | /* |
2155 | * Free all of a swapdev's extent information | |
2156 | */ | |
2157 | static void destroy_swap_extents(struct swap_info_struct *sis) | |
2158 | { | |
4efaceb1 AL |
2159 | while (!RB_EMPTY_ROOT(&sis->swap_extent_root)) { |
2160 | struct rb_node *rb = sis->swap_extent_root.rb_node; | |
2161 | struct swap_extent *se = rb_entry(rb, struct swap_extent, rb_node); | |
1da177e4 | 2162 | |
4efaceb1 | 2163 | rb_erase(rb, &sis->swap_extent_root); |
1da177e4 LT |
2164 | kfree(se); |
2165 | } | |
62c230bc | 2166 | |
bc4ae27d | 2167 | if (sis->flags & SWP_ACTIVATED) { |
62c230bc MG |
2168 | struct file *swap_file = sis->swap_file; |
2169 | struct address_space *mapping = swap_file->f_mapping; | |
2170 | ||
bc4ae27d OS |
2171 | sis->flags &= ~SWP_ACTIVATED; |
2172 | if (mapping->a_ops->swap_deactivate) | |
2173 | mapping->a_ops->swap_deactivate(swap_file); | |
62c230bc | 2174 | } |
1da177e4 LT |
2175 | } |
2176 | ||
2177 | /* | |
2178 | * Add a block range (and the corresponding page range) into this swapdev's | |
4efaceb1 | 2179 | * extent tree. |
1da177e4 | 2180 | * |
11d31886 | 2181 | * This function rather assumes that it is called in ascending page order. |
1da177e4 | 2182 | */ |
a509bc1a | 2183 | int |
1da177e4 LT |
2184 | add_swap_extent(struct swap_info_struct *sis, unsigned long start_page, |
2185 | unsigned long nr_pages, sector_t start_block) | |
2186 | { | |
4efaceb1 | 2187 | struct rb_node **link = &sis->swap_extent_root.rb_node, *parent = NULL; |
1da177e4 LT |
2188 | struct swap_extent *se; |
2189 | struct swap_extent *new_se; | |
4efaceb1 AL |
2190 | |
2191 | /* | |
2192 | * place the new node at the right most since the | |
2193 | * function is called in ascending page order. | |
2194 | */ | |
2195 | while (*link) { | |
2196 | parent = *link; | |
2197 | link = &parent->rb_right; | |
2198 | } | |
2199 | ||
2200 | if (parent) { | |
2201 | se = rb_entry(parent, struct swap_extent, rb_node); | |
11d31886 HD |
2202 | BUG_ON(se->start_page + se->nr_pages != start_page); |
2203 | if (se->start_block + se->nr_pages == start_block) { | |
1da177e4 LT |
2204 | /* Merge it */ |
2205 | se->nr_pages += nr_pages; | |
2206 | return 0; | |
2207 | } | |
1da177e4 LT |
2208 | } |
2209 | ||
4efaceb1 | 2210 | /* No merge, insert a new extent. */ |
1da177e4 LT |
2211 | new_se = kmalloc(sizeof(*se), GFP_KERNEL); |
2212 | if (new_se == NULL) | |
2213 | return -ENOMEM; | |
2214 | new_se->start_page = start_page; | |
2215 | new_se->nr_pages = nr_pages; | |
2216 | new_se->start_block = start_block; | |
2217 | ||
4efaceb1 AL |
2218 | rb_link_node(&new_se->rb_node, parent, link); |
2219 | rb_insert_color(&new_se->rb_node, &sis->swap_extent_root); | |
53092a74 | 2220 | return 1; |
1da177e4 | 2221 | } |
aa8aa8a3 | 2222 | EXPORT_SYMBOL_GPL(add_swap_extent); |
1da177e4 LT |
2223 | |
2224 | /* | |
2225 | * A `swap extent' is a simple thing which maps a contiguous range of pages | |
ff351f4b ML |
2226 | * onto a contiguous range of disk blocks. A rbtree of swap extents is |
2227 | * built at swapon time and is then used at swap_writepage/swap_readpage | |
1da177e4 LT |
2228 | * time for locating where on disk a page belongs. |
2229 | * | |
2230 | * If the swapfile is an S_ISBLK block device, a single extent is installed. | |
2231 | * This is done so that the main operating code can treat S_ISBLK and S_ISREG | |
2232 | * swap files identically. | |
2233 | * | |
2234 | * Whether the swapdev is an S_ISREG file or an S_ISBLK blockdev, the swap | |
ff351f4b | 2235 | * extent rbtree operates in PAGE_SIZE disk blocks. Both S_ISREG and S_ISBLK |
1da177e4 LT |
2236 | * swapfiles are handled *identically* after swapon time. |
2237 | * | |
2238 | * For S_ISREG swapfiles, setup_swap_extents() will walk all the file's blocks | |
ff351f4b ML |
2239 | * and will parse them into a rbtree, in PAGE_SIZE chunks. If some stray |
2240 | * blocks are found which do not fall within the PAGE_SIZE alignment | |
1da177e4 LT |
2241 | * requirements, they are simply tossed out - we will never use those blocks |
2242 | * for swapping. | |
2243 | * | |
1638045c DW |
2244 | * For all swap devices we set S_SWAPFILE across the life of the swapon. This |
2245 | * prevents users from writing to the swap device, which will corrupt memory. | |
1da177e4 LT |
2246 | * |
2247 | * The amount of disk space which a single swap extent represents varies. | |
2248 | * Typically it is in the 1-4 megabyte range. So we can have hundreds of | |
ff351f4b | 2249 | * extents in the rbtree. - akpm. |
1da177e4 | 2250 | */ |
53092a74 | 2251 | static int setup_swap_extents(struct swap_info_struct *sis, sector_t *span) |
1da177e4 | 2252 | { |
62c230bc MG |
2253 | struct file *swap_file = sis->swap_file; |
2254 | struct address_space *mapping = swap_file->f_mapping; | |
2255 | struct inode *inode = mapping->host; | |
1da177e4 LT |
2256 | int ret; |
2257 | ||
1da177e4 LT |
2258 | if (S_ISBLK(inode->i_mode)) { |
2259 | ret = add_swap_extent(sis, 0, sis->max, 0); | |
53092a74 | 2260 | *span = sis->pages; |
a509bc1a | 2261 | return ret; |
1da177e4 LT |
2262 | } |
2263 | ||
62c230bc | 2264 | if (mapping->a_ops->swap_activate) { |
a509bc1a | 2265 | ret = mapping->a_ops->swap_activate(sis, swap_file, span); |
4b60c0ff N |
2266 | if (ret < 0) |
2267 | return ret; | |
2268 | sis->flags |= SWP_ACTIVATED; | |
e1209d3a N |
2269 | if ((sis->flags & SWP_FS_OPS) && |
2270 | sio_pool_init() != 0) { | |
2271 | destroy_swap_extents(sis); | |
2272 | return -ENOMEM; | |
62c230bc | 2273 | } |
a509bc1a | 2274 | return ret; |
62c230bc MG |
2275 | } |
2276 | ||
a509bc1a | 2277 | return generic_swapfile_activate(sis, swap_file, span); |
1da177e4 LT |
2278 | } |
2279 | ||
a2468cc9 AL |
2280 | static int swap_node(struct swap_info_struct *p) |
2281 | { | |
2282 | struct block_device *bdev; | |
2283 | ||
2284 | if (p->bdev) | |
2285 | bdev = p->bdev; | |
2286 | else | |
2287 | bdev = p->swap_file->f_inode->i_sb->s_bdev; | |
2288 | ||
2289 | return bdev ? bdev->bd_disk->node_id : NUMA_NO_NODE; | |
2290 | } | |
2291 | ||
eb085574 HY |
2292 | static void setup_swap_info(struct swap_info_struct *p, int prio, |
2293 | unsigned char *swap_map, | |
2294 | struct swap_cluster_info *cluster_info) | |
40531542 | 2295 | { |
a2468cc9 AL |
2296 | int i; |
2297 | ||
40531542 CEB |
2298 | if (prio >= 0) |
2299 | p->prio = prio; | |
2300 | else | |
2301 | p->prio = --least_priority; | |
18ab4d4c DS |
2302 | /* |
2303 | * the plist prio is negated because plist ordering is | |
2304 | * low-to-high, while swap ordering is high-to-low | |
2305 | */ | |
2306 | p->list.prio = -p->prio; | |
a2468cc9 AL |
2307 | for_each_node(i) { |
2308 | if (p->prio >= 0) | |
2309 | p->avail_lists[i].prio = -p->prio; | |
2310 | else { | |
2311 | if (swap_node(p) == i) | |
2312 | p->avail_lists[i].prio = 1; | |
2313 | else | |
2314 | p->avail_lists[i].prio = -p->prio; | |
2315 | } | |
2316 | } | |
40531542 | 2317 | p->swap_map = swap_map; |
2a8f9449 | 2318 | p->cluster_info = cluster_info; |
eb085574 HY |
2319 | } |
2320 | ||
2321 | static void _enable_swap_info(struct swap_info_struct *p) | |
2322 | { | |
63d8620e | 2323 | p->flags |= SWP_WRITEOK; |
ec8acf20 | 2324 | atomic_long_add(p->pages, &nr_swap_pages); |
40531542 CEB |
2325 | total_swap_pages += p->pages; |
2326 | ||
adfab836 | 2327 | assert_spin_locked(&swap_lock); |
adfab836 | 2328 | /* |
18ab4d4c DS |
2329 | * both lists are plists, and thus priority ordered. |
2330 | * swap_active_head needs to be priority ordered for swapoff(), | |
2331 | * which on removal of any swap_info_struct with an auto-assigned | |
2332 | * (i.e. negative) priority increments the auto-assigned priority | |
2333 | * of any lower-priority swap_info_structs. | |
e2e3fdc7 | 2334 | * swap_avail_head needs to be priority ordered for folio_alloc_swap(), |
18ab4d4c DS |
2335 | * which allocates swap pages from the highest available priority |
2336 | * swap_info_struct. | |
adfab836 | 2337 | */ |
18ab4d4c | 2338 | plist_add(&p->list, &swap_active_head); |
a2468cc9 | 2339 | add_to_avail_list(p); |
cf0cac0a CEB |
2340 | } |
2341 | ||
2342 | static void enable_swap_info(struct swap_info_struct *p, int prio, | |
2343 | unsigned char *swap_map, | |
2a8f9449 | 2344 | struct swap_cluster_info *cluster_info, |
cf0cac0a CEB |
2345 | unsigned long *frontswap_map) |
2346 | { | |
1cf53c89 CH |
2347 | if (IS_ENABLED(CONFIG_FRONTSWAP)) |
2348 | frontswap_init(p->type, frontswap_map); | |
cf0cac0a | 2349 | spin_lock(&swap_lock); |
ec8acf20 | 2350 | spin_lock(&p->lock); |
eb085574 HY |
2351 | setup_swap_info(p, prio, swap_map, cluster_info); |
2352 | spin_unlock(&p->lock); | |
2353 | spin_unlock(&swap_lock); | |
2354 | /* | |
63d8620e | 2355 | * Finished initializing swap device, now it's safe to reference it. |
eb085574 | 2356 | */ |
63d8620e | 2357 | percpu_ref_resurrect(&p->users); |
eb085574 HY |
2358 | spin_lock(&swap_lock); |
2359 | spin_lock(&p->lock); | |
2360 | _enable_swap_info(p); | |
ec8acf20 | 2361 | spin_unlock(&p->lock); |
cf0cac0a CEB |
2362 | spin_unlock(&swap_lock); |
2363 | } | |
2364 | ||
2365 | static void reinsert_swap_info(struct swap_info_struct *p) | |
2366 | { | |
2367 | spin_lock(&swap_lock); | |
ec8acf20 | 2368 | spin_lock(&p->lock); |
eb085574 HY |
2369 | setup_swap_info(p, p->prio, p->swap_map, p->cluster_info); |
2370 | _enable_swap_info(p); | |
ec8acf20 | 2371 | spin_unlock(&p->lock); |
40531542 CEB |
2372 | spin_unlock(&swap_lock); |
2373 | } | |
2374 | ||
67afa38e TC |
2375 | bool has_usable_swap(void) |
2376 | { | |
2377 | bool ret = true; | |
2378 | ||
2379 | spin_lock(&swap_lock); | |
2380 | if (plist_head_empty(&swap_active_head)) | |
2381 | ret = false; | |
2382 | spin_unlock(&swap_lock); | |
2383 | return ret; | |
2384 | } | |
2385 | ||
c4ea37c2 | 2386 | SYSCALL_DEFINE1(swapoff, const char __user *, specialfile) |
1da177e4 | 2387 | { |
73c34b6a | 2388 | struct swap_info_struct *p = NULL; |
8d69aaee | 2389 | unsigned char *swap_map; |
2a8f9449 | 2390 | struct swap_cluster_info *cluster_info; |
4f89849d | 2391 | unsigned long *frontswap_map; |
1da177e4 LT |
2392 | struct file *swap_file, *victim; |
2393 | struct address_space *mapping; | |
2394 | struct inode *inode; | |
91a27b2a | 2395 | struct filename *pathname; |
adfab836 | 2396 | int err, found = 0; |
5b808a23 | 2397 | unsigned int old_block_size; |
886bb7e9 | 2398 | |
1da177e4 LT |
2399 | if (!capable(CAP_SYS_ADMIN)) |
2400 | return -EPERM; | |
2401 | ||
191c5424 AV |
2402 | BUG_ON(!current->mm); |
2403 | ||
1da177e4 | 2404 | pathname = getname(specialfile); |
1da177e4 | 2405 | if (IS_ERR(pathname)) |
f58b59c1 | 2406 | return PTR_ERR(pathname); |
1da177e4 | 2407 | |
669abf4e | 2408 | victim = file_open_name(pathname, O_RDWR|O_LARGEFILE, 0); |
1da177e4 LT |
2409 | err = PTR_ERR(victim); |
2410 | if (IS_ERR(victim)) | |
2411 | goto out; | |
2412 | ||
2413 | mapping = victim->f_mapping; | |
5d337b91 | 2414 | spin_lock(&swap_lock); |
18ab4d4c | 2415 | plist_for_each_entry(p, &swap_active_head, list) { |
22c6f8fd | 2416 | if (p->flags & SWP_WRITEOK) { |
adfab836 DS |
2417 | if (p->swap_file->f_mapping == mapping) { |
2418 | found = 1; | |
1da177e4 | 2419 | break; |
adfab836 | 2420 | } |
1da177e4 | 2421 | } |
1da177e4 | 2422 | } |
adfab836 | 2423 | if (!found) { |
1da177e4 | 2424 | err = -EINVAL; |
5d337b91 | 2425 | spin_unlock(&swap_lock); |
1da177e4 LT |
2426 | goto out_dput; |
2427 | } | |
191c5424 | 2428 | if (!security_vm_enough_memory_mm(current->mm, p->pages)) |
1da177e4 LT |
2429 | vm_unacct_memory(p->pages); |
2430 | else { | |
2431 | err = -ENOMEM; | |
5d337b91 | 2432 | spin_unlock(&swap_lock); |
1da177e4 LT |
2433 | goto out_dput; |
2434 | } | |
a2468cc9 | 2435 | del_from_avail_list(p); |
ec8acf20 | 2436 | spin_lock(&p->lock); |
78ecba08 | 2437 | if (p->prio < 0) { |
adfab836 | 2438 | struct swap_info_struct *si = p; |
a2468cc9 | 2439 | int nid; |
adfab836 | 2440 | |
18ab4d4c | 2441 | plist_for_each_entry_continue(si, &swap_active_head, list) { |
adfab836 | 2442 | si->prio++; |
18ab4d4c | 2443 | si->list.prio--; |
a2468cc9 AL |
2444 | for_each_node(nid) { |
2445 | if (si->avail_lists[nid].prio != 1) | |
2446 | si->avail_lists[nid].prio--; | |
2447 | } | |
adfab836 | 2448 | } |
78ecba08 HD |
2449 | least_priority++; |
2450 | } | |
18ab4d4c | 2451 | plist_del(&p->list, &swap_active_head); |
ec8acf20 | 2452 | atomic_long_sub(p->pages, &nr_swap_pages); |
1da177e4 LT |
2453 | total_swap_pages -= p->pages; |
2454 | p->flags &= ~SWP_WRITEOK; | |
ec8acf20 | 2455 | spin_unlock(&p->lock); |
5d337b91 | 2456 | spin_unlock(&swap_lock); |
fb4f88dc | 2457 | |
039939a6 TC |
2458 | disable_swap_slots_cache_lock(); |
2459 | ||
e1e12d2f | 2460 | set_current_oom_origin(); |
10a9c496 | 2461 | err = try_to_unuse(p->type); |
e1e12d2f | 2462 | clear_current_oom_origin(); |
1da177e4 | 2463 | |
1da177e4 LT |
2464 | if (err) { |
2465 | /* re-insert swap space back into swap_list */ | |
cf0cac0a | 2466 | reinsert_swap_info(p); |
039939a6 | 2467 | reenable_swap_slots_cache_unlock(); |
1da177e4 LT |
2468 | goto out_dput; |
2469 | } | |
52b7efdb | 2470 | |
039939a6 TC |
2471 | reenable_swap_slots_cache_unlock(); |
2472 | ||
eb085574 | 2473 | /* |
63d8620e ML |
2474 | * Wait for swap operations protected by get/put_swap_device() |
2475 | * to complete. | |
2476 | * | |
2477 | * We need synchronize_rcu() here to protect the accessing to | |
2478 | * the swap cache data structure. | |
eb085574 | 2479 | */ |
63d8620e | 2480 | percpu_ref_kill(&p->users); |
eb085574 | 2481 | synchronize_rcu(); |
63d8620e | 2482 | wait_for_completion(&p->comp); |
eb085574 | 2483 | |
815c2c54 SL |
2484 | flush_work(&p->discard_work); |
2485 | ||
5d337b91 | 2486 | destroy_swap_extents(p); |
570a335b HD |
2487 | if (p->flags & SWP_CONTINUED) |
2488 | free_swap_count_continuations(p); | |
2489 | ||
10f0d2a5 | 2490 | if (!p->bdev || !bdev_nonrot(p->bdev)) |
81a0298b HY |
2491 | atomic_dec(&nr_rotate_swap); |
2492 | ||
fc0abb14 | 2493 | mutex_lock(&swapon_mutex); |
5d337b91 | 2494 | spin_lock(&swap_lock); |
ec8acf20 | 2495 | spin_lock(&p->lock); |
5d337b91 HD |
2496 | drain_mmlist(); |
2497 | ||
bb243f7d | 2498 | /* wait for anyone still in scan_swap_map_slots */ |
52b7efdb HD |
2499 | p->highest_bit = 0; /* cuts scans short */ |
2500 | while (p->flags >= SWP_SCANNING) { | |
ec8acf20 | 2501 | spin_unlock(&p->lock); |
5d337b91 | 2502 | spin_unlock(&swap_lock); |
13e4b57f | 2503 | schedule_timeout_uninterruptible(1); |
5d337b91 | 2504 | spin_lock(&swap_lock); |
ec8acf20 | 2505 | spin_lock(&p->lock); |
52b7efdb | 2506 | } |
52b7efdb | 2507 | |
1da177e4 | 2508 | swap_file = p->swap_file; |
5b808a23 | 2509 | old_block_size = p->old_block_size; |
1da177e4 LT |
2510 | p->swap_file = NULL; |
2511 | p->max = 0; | |
2512 | swap_map = p->swap_map; | |
2513 | p->swap_map = NULL; | |
2a8f9449 SL |
2514 | cluster_info = p->cluster_info; |
2515 | p->cluster_info = NULL; | |
4f89849d | 2516 | frontswap_map = frontswap_map_get(p); |
ec8acf20 | 2517 | spin_unlock(&p->lock); |
5d337b91 | 2518 | spin_unlock(&swap_lock); |
8a84802e | 2519 | arch_swap_invalidate_area(p->type); |
adfab836 | 2520 | frontswap_invalidate_area(p->type); |
58e97ba6 | 2521 | frontswap_map_set(p, NULL); |
fc0abb14 | 2522 | mutex_unlock(&swapon_mutex); |
ebc2a1a6 SL |
2523 | free_percpu(p->percpu_cluster); |
2524 | p->percpu_cluster = NULL; | |
49070588 HY |
2525 | free_percpu(p->cluster_next_cpu); |
2526 | p->cluster_next_cpu = NULL; | |
1da177e4 | 2527 | vfree(swap_map); |
54f180d3 HY |
2528 | kvfree(cluster_info); |
2529 | kvfree(frontswap_map); | |
2de1a7e4 | 2530 | /* Destroy swap account information */ |
adfab836 | 2531 | swap_cgroup_swapoff(p->type); |
4b3ef9da | 2532 | exit_swap_address_space(p->type); |
27a7faa0 | 2533 | |
1da177e4 LT |
2534 | inode = mapping->host; |
2535 | if (S_ISBLK(inode->i_mode)) { | |
2536 | struct block_device *bdev = I_BDEV(inode); | |
1638045c | 2537 | |
5b808a23 | 2538 | set_blocksize(bdev, old_block_size); |
e525fd89 | 2539 | blkdev_put(bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL); |
1da177e4 | 2540 | } |
1638045c DW |
2541 | |
2542 | inode_lock(inode); | |
2543 | inode->i_flags &= ~S_SWAPFILE; | |
2544 | inode_unlock(inode); | |
1da177e4 | 2545 | filp_close(swap_file, NULL); |
f893ab41 WY |
2546 | |
2547 | /* | |
2548 | * Clear the SWP_USED flag after all resources are freed so that swapon | |
2549 | * can reuse this swap_info in alloc_swap_info() safely. It is ok to | |
2550 | * not hold p->lock after we cleared its SWP_WRITEOK. | |
2551 | */ | |
2552 | spin_lock(&swap_lock); | |
2553 | p->flags = 0; | |
2554 | spin_unlock(&swap_lock); | |
2555 | ||
1da177e4 | 2556 | err = 0; |
66d7dd51 KS |
2557 | atomic_inc(&proc_poll_event); |
2558 | wake_up_interruptible(&proc_poll_wait); | |
1da177e4 LT |
2559 | |
2560 | out_dput: | |
2561 | filp_close(victim, NULL); | |
2562 | out: | |
f58b59c1 | 2563 | putname(pathname); |
1da177e4 LT |
2564 | return err; |
2565 | } | |
2566 | ||
2567 | #ifdef CONFIG_PROC_FS | |
9dd95748 | 2568 | static __poll_t swaps_poll(struct file *file, poll_table *wait) |
66d7dd51 | 2569 | { |
f1514638 | 2570 | struct seq_file *seq = file->private_data; |
66d7dd51 KS |
2571 | |
2572 | poll_wait(file, &proc_poll_wait, wait); | |
2573 | ||
f1514638 KS |
2574 | if (seq->poll_event != atomic_read(&proc_poll_event)) { |
2575 | seq->poll_event = atomic_read(&proc_poll_event); | |
a9a08845 | 2576 | return EPOLLIN | EPOLLRDNORM | EPOLLERR | EPOLLPRI; |
66d7dd51 KS |
2577 | } |
2578 | ||
a9a08845 | 2579 | return EPOLLIN | EPOLLRDNORM; |
66d7dd51 KS |
2580 | } |
2581 | ||
1da177e4 LT |
2582 | /* iterator */ |
2583 | static void *swap_start(struct seq_file *swap, loff_t *pos) | |
2584 | { | |
efa90a98 HD |
2585 | struct swap_info_struct *si; |
2586 | int type; | |
1da177e4 LT |
2587 | loff_t l = *pos; |
2588 | ||
fc0abb14 | 2589 | mutex_lock(&swapon_mutex); |
1da177e4 | 2590 | |
881e4aab SS |
2591 | if (!l) |
2592 | return SEQ_START_TOKEN; | |
2593 | ||
c10d38cc | 2594 | for (type = 0; (si = swap_type_to_swap_info(type)); type++) { |
efa90a98 | 2595 | if (!(si->flags & SWP_USED) || !si->swap_map) |
1da177e4 | 2596 | continue; |
881e4aab | 2597 | if (!--l) |
efa90a98 | 2598 | return si; |
1da177e4 LT |
2599 | } |
2600 | ||
2601 | return NULL; | |
2602 | } | |
2603 | ||
2604 | static void *swap_next(struct seq_file *swap, void *v, loff_t *pos) | |
2605 | { | |
efa90a98 HD |
2606 | struct swap_info_struct *si = v; |
2607 | int type; | |
1da177e4 | 2608 | |
881e4aab | 2609 | if (v == SEQ_START_TOKEN) |
efa90a98 HD |
2610 | type = 0; |
2611 | else | |
2612 | type = si->type + 1; | |
881e4aab | 2613 | |
10c8d69f | 2614 | ++(*pos); |
c10d38cc | 2615 | for (; (si = swap_type_to_swap_info(type)); type++) { |
efa90a98 | 2616 | if (!(si->flags & SWP_USED) || !si->swap_map) |
1da177e4 | 2617 | continue; |
efa90a98 | 2618 | return si; |
1da177e4 LT |
2619 | } |
2620 | ||
2621 | return NULL; | |
2622 | } | |
2623 | ||
2624 | static void swap_stop(struct seq_file *swap, void *v) | |
2625 | { | |
fc0abb14 | 2626 | mutex_unlock(&swapon_mutex); |
1da177e4 LT |
2627 | } |
2628 | ||
2629 | static int swap_show(struct seq_file *swap, void *v) | |
2630 | { | |
efa90a98 | 2631 | struct swap_info_struct *si = v; |
1da177e4 LT |
2632 | struct file *file; |
2633 | int len; | |
642929a2 | 2634 | unsigned long bytes, inuse; |
1da177e4 | 2635 | |
efa90a98 | 2636 | if (si == SEQ_START_TOKEN) { |
68d68ff6 | 2637 | seq_puts(swap, "Filename\t\t\t\tType\t\tSize\t\tUsed\t\tPriority\n"); |
881e4aab SS |
2638 | return 0; |
2639 | } | |
1da177e4 | 2640 | |
6f793940 | 2641 | bytes = si->pages << (PAGE_SHIFT - 10); |
c8945306 | 2642 | inuse = READ_ONCE(si->inuse_pages) << (PAGE_SHIFT - 10); |
6f793940 | 2643 | |
efa90a98 | 2644 | file = si->swap_file; |
2726d566 | 2645 | len = seq_file_path(swap, file, " \t\n\\"); |
642929a2 | 2646 | seq_printf(swap, "%*s%s\t%lu\t%s%lu\t%s%d\n", |
886bb7e9 | 2647 | len < 40 ? 40 - len : 1, " ", |
496ad9aa | 2648 | S_ISBLK(file_inode(file)->i_mode) ? |
1da177e4 | 2649 | "partition" : "file\t", |
6f793940 RD |
2650 | bytes, bytes < 10000000 ? "\t" : "", |
2651 | inuse, inuse < 10000000 ? "\t" : "", | |
efa90a98 | 2652 | si->prio); |
1da177e4 LT |
2653 | return 0; |
2654 | } | |
2655 | ||
15ad7cdc | 2656 | static const struct seq_operations swaps_op = { |
1da177e4 LT |
2657 | .start = swap_start, |
2658 | .next = swap_next, | |
2659 | .stop = swap_stop, | |
2660 | .show = swap_show | |
2661 | }; | |
2662 | ||
2663 | static int swaps_open(struct inode *inode, struct file *file) | |
2664 | { | |
f1514638 | 2665 | struct seq_file *seq; |
66d7dd51 KS |
2666 | int ret; |
2667 | ||
66d7dd51 | 2668 | ret = seq_open(file, &swaps_op); |
f1514638 | 2669 | if (ret) |
66d7dd51 | 2670 | return ret; |
66d7dd51 | 2671 | |
f1514638 KS |
2672 | seq = file->private_data; |
2673 | seq->poll_event = atomic_read(&proc_poll_event); | |
2674 | return 0; | |
1da177e4 LT |
2675 | } |
2676 | ||
97a32539 | 2677 | static const struct proc_ops swaps_proc_ops = { |
d919b33d | 2678 | .proc_flags = PROC_ENTRY_PERMANENT, |
97a32539 AD |
2679 | .proc_open = swaps_open, |
2680 | .proc_read = seq_read, | |
2681 | .proc_lseek = seq_lseek, | |
2682 | .proc_release = seq_release, | |
2683 | .proc_poll = swaps_poll, | |
1da177e4 LT |
2684 | }; |
2685 | ||
2686 | static int __init procswaps_init(void) | |
2687 | { | |
97a32539 | 2688 | proc_create("swaps", 0, NULL, &swaps_proc_ops); |
1da177e4 LT |
2689 | return 0; |
2690 | } | |
2691 | __initcall(procswaps_init); | |
2692 | #endif /* CONFIG_PROC_FS */ | |
2693 | ||
1796316a JB |
2694 | #ifdef MAX_SWAPFILES_CHECK |
2695 | static int __init max_swapfiles_check(void) | |
2696 | { | |
2697 | MAX_SWAPFILES_CHECK(); | |
2698 | return 0; | |
2699 | } | |
2700 | late_initcall(max_swapfiles_check); | |
2701 | #endif | |
2702 | ||
53cbb243 | 2703 | static struct swap_info_struct *alloc_swap_info(void) |
1da177e4 | 2704 | { |
73c34b6a | 2705 | struct swap_info_struct *p; |
b11a76b3 | 2706 | struct swap_info_struct *defer = NULL; |
1da177e4 | 2707 | unsigned int type; |
a2468cc9 | 2708 | int i; |
efa90a98 | 2709 | |
96008744 | 2710 | p = kvzalloc(struct_size(p, avail_lists, nr_node_ids), GFP_KERNEL); |
efa90a98 | 2711 | if (!p) |
53cbb243 | 2712 | return ERR_PTR(-ENOMEM); |
efa90a98 | 2713 | |
63d8620e ML |
2714 | if (percpu_ref_init(&p->users, swap_users_ref_free, |
2715 | PERCPU_REF_INIT_DEAD, GFP_KERNEL)) { | |
2716 | kvfree(p); | |
2717 | return ERR_PTR(-ENOMEM); | |
2718 | } | |
2719 | ||
5d337b91 | 2720 | spin_lock(&swap_lock); |
efa90a98 HD |
2721 | for (type = 0; type < nr_swapfiles; type++) { |
2722 | if (!(swap_info[type]->flags & SWP_USED)) | |
1da177e4 | 2723 | break; |
efa90a98 | 2724 | } |
0697212a | 2725 | if (type >= MAX_SWAPFILES) { |
5d337b91 | 2726 | spin_unlock(&swap_lock); |
63d8620e | 2727 | percpu_ref_exit(&p->users); |
873d7bcf | 2728 | kvfree(p); |
730c0581 | 2729 | return ERR_PTR(-EPERM); |
1da177e4 | 2730 | } |
efa90a98 HD |
2731 | if (type >= nr_swapfiles) { |
2732 | p->type = type; | |
efa90a98 | 2733 | /* |
a4b45114 HY |
2734 | * Publish the swap_info_struct after initializing it. |
2735 | * Note that kvzalloc() above zeroes all its fields. | |
efa90a98 | 2736 | */ |
a4b45114 HY |
2737 | smp_store_release(&swap_info[type], p); /* rcu_assign_pointer() */ |
2738 | nr_swapfiles++; | |
efa90a98 | 2739 | } else { |
b11a76b3 | 2740 | defer = p; |
efa90a98 HD |
2741 | p = swap_info[type]; |
2742 | /* | |
2743 | * Do not memset this entry: a racing procfs swap_next() | |
2744 | * would be relying on p->type to remain valid. | |
2745 | */ | |
2746 | } | |
4efaceb1 | 2747 | p->swap_extent_root = RB_ROOT; |
18ab4d4c | 2748 | plist_node_init(&p->list, 0); |
a2468cc9 AL |
2749 | for_each_node(i) |
2750 | plist_node_init(&p->avail_lists[i], 0); | |
1da177e4 | 2751 | p->flags = SWP_USED; |
5d337b91 | 2752 | spin_unlock(&swap_lock); |
63d8620e ML |
2753 | if (defer) { |
2754 | percpu_ref_exit(&defer->users); | |
2755 | kvfree(defer); | |
2756 | } | |
ec8acf20 | 2757 | spin_lock_init(&p->lock); |
2628bd6f | 2758 | spin_lock_init(&p->cont_lock); |
63d8620e | 2759 | init_completion(&p->comp); |
efa90a98 | 2760 | |
53cbb243 | 2761 | return p; |
53cbb243 CEB |
2762 | } |
2763 | ||
4d0e1e10 CEB |
2764 | static int claim_swapfile(struct swap_info_struct *p, struct inode *inode) |
2765 | { | |
2766 | int error; | |
2767 | ||
2768 | if (S_ISBLK(inode->i_mode)) { | |
ef16e1d9 | 2769 | p->bdev = blkdev_get_by_dev(inode->i_rdev, |
6f179af8 | 2770 | FMODE_READ | FMODE_WRITE | FMODE_EXCL, p); |
ef16e1d9 CH |
2771 | if (IS_ERR(p->bdev)) { |
2772 | error = PTR_ERR(p->bdev); | |
4d0e1e10 | 2773 | p->bdev = NULL; |
6f179af8 | 2774 | return error; |
4d0e1e10 CEB |
2775 | } |
2776 | p->old_block_size = block_size(p->bdev); | |
2777 | error = set_blocksize(p->bdev, PAGE_SIZE); | |
2778 | if (error < 0) | |
87ade72a | 2779 | return error; |
12d2966d NA |
2780 | /* |
2781 | * Zoned block devices contain zones that have a sequential | |
2782 | * write only restriction. Hence zoned block devices are not | |
2783 | * suitable for swapping. Disallow them here. | |
2784 | */ | |
9964e674 | 2785 | if (bdev_is_zoned(p->bdev)) |
12d2966d | 2786 | return -EINVAL; |
4d0e1e10 CEB |
2787 | p->flags |= SWP_BLKDEV; |
2788 | } else if (S_ISREG(inode->i_mode)) { | |
2789 | p->bdev = inode->i_sb->s_bdev; | |
1638045c DW |
2790 | } |
2791 | ||
4d0e1e10 | 2792 | return 0; |
4d0e1e10 CEB |
2793 | } |
2794 | ||
377eeaa8 AK |
2795 | |
2796 | /* | |
2797 | * Find out how many pages are allowed for a single swap device. There | |
2798 | * are two limiting factors: | |
2799 | * 1) the number of bits for the swap offset in the swp_entry_t type, and | |
2800 | * 2) the number of bits in the swap pte, as defined by the different | |
2801 | * architectures. | |
2802 | * | |
2803 | * In order to find the largest possible bit mask, a swap entry with | |
2804 | * swap type 0 and swap offset ~0UL is created, encoded to a swap pte, | |
2805 | * decoded to a swp_entry_t again, and finally the swap offset is | |
2806 | * extracted. | |
2807 | * | |
2808 | * This will mask all the bits from the initial ~0UL mask that can't | |
2809 | * be encoded in either the swp_entry_t or the architecture definition | |
2810 | * of a swap pte. | |
2811 | */ | |
2812 | unsigned long generic_max_swapfile_size(void) | |
2813 | { | |
2814 | return swp_offset(pte_to_swp_entry( | |
2815 | swp_entry_to_pte(swp_entry(0, ~0UL)))) + 1; | |
2816 | } | |
2817 | ||
2818 | /* Can be overridden by an architecture for additional checks. */ | |
2819 | __weak unsigned long max_swapfile_size(void) | |
2820 | { | |
2821 | return generic_max_swapfile_size(); | |
2822 | } | |
2823 | ||
ca8bd38b CEB |
2824 | static unsigned long read_swap_header(struct swap_info_struct *p, |
2825 | union swap_header *swap_header, | |
2826 | struct inode *inode) | |
2827 | { | |
2828 | int i; | |
2829 | unsigned long maxpages; | |
2830 | unsigned long swapfilepages; | |
d6bbbd29 | 2831 | unsigned long last_page; |
ca8bd38b CEB |
2832 | |
2833 | if (memcmp("SWAPSPACE2", swap_header->magic.magic, 10)) { | |
465c47fd | 2834 | pr_err("Unable to find swap-space signature\n"); |
38719025 | 2835 | return 0; |
ca8bd38b CEB |
2836 | } |
2837 | ||
041711ce | 2838 | /* swap partition endianness hack... */ |
ca8bd38b CEB |
2839 | if (swab32(swap_header->info.version) == 1) { |
2840 | swab32s(&swap_header->info.version); | |
2841 | swab32s(&swap_header->info.last_page); | |
2842 | swab32s(&swap_header->info.nr_badpages); | |
dd111be6 JH |
2843 | if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES) |
2844 | return 0; | |
ca8bd38b CEB |
2845 | for (i = 0; i < swap_header->info.nr_badpages; i++) |
2846 | swab32s(&swap_header->info.badpages[i]); | |
2847 | } | |
2848 | /* Check the swap header's sub-version */ | |
2849 | if (swap_header->info.version != 1) { | |
465c47fd AM |
2850 | pr_warn("Unable to handle swap header version %d\n", |
2851 | swap_header->info.version); | |
38719025 | 2852 | return 0; |
ca8bd38b CEB |
2853 | } |
2854 | ||
2855 | p->lowest_bit = 1; | |
2856 | p->cluster_next = 1; | |
2857 | p->cluster_nr = 0; | |
2858 | ||
377eeaa8 | 2859 | maxpages = max_swapfile_size(); |
d6bbbd29 | 2860 | last_page = swap_header->info.last_page; |
a06ad633 TA |
2861 | if (!last_page) { |
2862 | pr_warn("Empty swap-file\n"); | |
2863 | return 0; | |
2864 | } | |
d6bbbd29 | 2865 | if (last_page > maxpages) { |
465c47fd | 2866 | pr_warn("Truncating oversized swap area, only using %luk out of %luk\n", |
d6bbbd29 RJ |
2867 | maxpages << (PAGE_SHIFT - 10), |
2868 | last_page << (PAGE_SHIFT - 10)); | |
2869 | } | |
2870 | if (maxpages > last_page) { | |
2871 | maxpages = last_page + 1; | |
ca8bd38b CEB |
2872 | /* p->max is an unsigned int: don't overflow it */ |
2873 | if ((unsigned int)maxpages == 0) | |
2874 | maxpages = UINT_MAX; | |
2875 | } | |
2876 | p->highest_bit = maxpages - 1; | |
2877 | ||
2878 | if (!maxpages) | |
38719025 | 2879 | return 0; |
ca8bd38b CEB |
2880 | swapfilepages = i_size_read(inode) >> PAGE_SHIFT; |
2881 | if (swapfilepages && maxpages > swapfilepages) { | |
465c47fd | 2882 | pr_warn("Swap area shorter than signature indicates\n"); |
38719025 | 2883 | return 0; |
ca8bd38b CEB |
2884 | } |
2885 | if (swap_header->info.nr_badpages && S_ISREG(inode->i_mode)) | |
38719025 | 2886 | return 0; |
ca8bd38b | 2887 | if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES) |
38719025 | 2888 | return 0; |
ca8bd38b CEB |
2889 | |
2890 | return maxpages; | |
ca8bd38b CEB |
2891 | } |
2892 | ||
4b3ef9da | 2893 | #define SWAP_CLUSTER_INFO_COLS \ |
235b6217 | 2894 | DIV_ROUND_UP(L1_CACHE_BYTES, sizeof(struct swap_cluster_info)) |
4b3ef9da HY |
2895 | #define SWAP_CLUSTER_SPACE_COLS \ |
2896 | DIV_ROUND_UP(SWAP_ADDRESS_SPACE_PAGES, SWAPFILE_CLUSTER) | |
2897 | #define SWAP_CLUSTER_COLS \ | |
2898 | max_t(unsigned int, SWAP_CLUSTER_INFO_COLS, SWAP_CLUSTER_SPACE_COLS) | |
235b6217 | 2899 | |
915d4d7b CEB |
2900 | static int setup_swap_map_and_extents(struct swap_info_struct *p, |
2901 | union swap_header *swap_header, | |
2902 | unsigned char *swap_map, | |
2a8f9449 | 2903 | struct swap_cluster_info *cluster_info, |
915d4d7b CEB |
2904 | unsigned long maxpages, |
2905 | sector_t *span) | |
2906 | { | |
235b6217 | 2907 | unsigned int j, k; |
915d4d7b CEB |
2908 | unsigned int nr_good_pages; |
2909 | int nr_extents; | |
2a8f9449 | 2910 | unsigned long nr_clusters = DIV_ROUND_UP(maxpages, SWAPFILE_CLUSTER); |
235b6217 HY |
2911 | unsigned long col = p->cluster_next / SWAPFILE_CLUSTER % SWAP_CLUSTER_COLS; |
2912 | unsigned long i, idx; | |
915d4d7b CEB |
2913 | |
2914 | nr_good_pages = maxpages - 1; /* omit header page */ | |
2915 | ||
6b534915 HY |
2916 | cluster_list_init(&p->free_clusters); |
2917 | cluster_list_init(&p->discard_clusters); | |
2a8f9449 | 2918 | |
915d4d7b CEB |
2919 | for (i = 0; i < swap_header->info.nr_badpages; i++) { |
2920 | unsigned int page_nr = swap_header->info.badpages[i]; | |
bdb8e3f6 CEB |
2921 | if (page_nr == 0 || page_nr > swap_header->info.last_page) |
2922 | return -EINVAL; | |
915d4d7b CEB |
2923 | if (page_nr < maxpages) { |
2924 | swap_map[page_nr] = SWAP_MAP_BAD; | |
2925 | nr_good_pages--; | |
2a8f9449 SL |
2926 | /* |
2927 | * Haven't marked the cluster free yet, no list | |
2928 | * operation involved | |
2929 | */ | |
2930 | inc_cluster_info_page(p, cluster_info, page_nr); | |
915d4d7b CEB |
2931 | } |
2932 | } | |
2933 | ||
2a8f9449 SL |
2934 | /* Haven't marked the cluster free yet, no list operation involved */ |
2935 | for (i = maxpages; i < round_up(maxpages, SWAPFILE_CLUSTER); i++) | |
2936 | inc_cluster_info_page(p, cluster_info, i); | |
2937 | ||
915d4d7b CEB |
2938 | if (nr_good_pages) { |
2939 | swap_map[0] = SWAP_MAP_BAD; | |
2a8f9449 SL |
2940 | /* |
2941 | * Not mark the cluster free yet, no list | |
2942 | * operation involved | |
2943 | */ | |
2944 | inc_cluster_info_page(p, cluster_info, 0); | |
915d4d7b CEB |
2945 | p->max = maxpages; |
2946 | p->pages = nr_good_pages; | |
2947 | nr_extents = setup_swap_extents(p, span); | |
bdb8e3f6 CEB |
2948 | if (nr_extents < 0) |
2949 | return nr_extents; | |
915d4d7b CEB |
2950 | nr_good_pages = p->pages; |
2951 | } | |
2952 | if (!nr_good_pages) { | |
465c47fd | 2953 | pr_warn("Empty swap-file\n"); |
bdb8e3f6 | 2954 | return -EINVAL; |
915d4d7b CEB |
2955 | } |
2956 | ||
2a8f9449 SL |
2957 | if (!cluster_info) |
2958 | return nr_extents; | |
2959 | ||
235b6217 | 2960 | |
4b3ef9da HY |
2961 | /* |
2962 | * Reduce false cache line sharing between cluster_info and | |
2963 | * sharing same address space. | |
2964 | */ | |
235b6217 HY |
2965 | for (k = 0; k < SWAP_CLUSTER_COLS; k++) { |
2966 | j = (k + col) % SWAP_CLUSTER_COLS; | |
2967 | for (i = 0; i < DIV_ROUND_UP(nr_clusters, SWAP_CLUSTER_COLS); i++) { | |
2968 | idx = i * SWAP_CLUSTER_COLS + j; | |
2969 | if (idx >= nr_clusters) | |
2970 | continue; | |
2971 | if (cluster_count(&cluster_info[idx])) | |
2972 | continue; | |
2a8f9449 | 2973 | cluster_set_flag(&cluster_info[idx], CLUSTER_FLAG_FREE); |
6b534915 HY |
2974 | cluster_list_add_tail(&p->free_clusters, cluster_info, |
2975 | idx); | |
2a8f9449 | 2976 | } |
2a8f9449 | 2977 | } |
915d4d7b | 2978 | return nr_extents; |
915d4d7b CEB |
2979 | } |
2980 | ||
53cbb243 CEB |
2981 | SYSCALL_DEFINE2(swapon, const char __user *, specialfile, int, swap_flags) |
2982 | { | |
2983 | struct swap_info_struct *p; | |
91a27b2a | 2984 | struct filename *name; |
53cbb243 CEB |
2985 | struct file *swap_file = NULL; |
2986 | struct address_space *mapping; | |
51cc3a66 | 2987 | struct dentry *dentry; |
40531542 | 2988 | int prio; |
53cbb243 CEB |
2989 | int error; |
2990 | union swap_header *swap_header; | |
915d4d7b | 2991 | int nr_extents; |
53cbb243 CEB |
2992 | sector_t span; |
2993 | unsigned long maxpages; | |
53cbb243 | 2994 | unsigned char *swap_map = NULL; |
2a8f9449 | 2995 | struct swap_cluster_info *cluster_info = NULL; |
38b5faf4 | 2996 | unsigned long *frontswap_map = NULL; |
53cbb243 CEB |
2997 | struct page *page = NULL; |
2998 | struct inode *inode = NULL; | |
7cbf3192 | 2999 | bool inced_nr_rotate_swap = false; |
53cbb243 | 3000 | |
d15cab97 HD |
3001 | if (swap_flags & ~SWAP_FLAGS_VALID) |
3002 | return -EINVAL; | |
3003 | ||
53cbb243 CEB |
3004 | if (!capable(CAP_SYS_ADMIN)) |
3005 | return -EPERM; | |
3006 | ||
a2468cc9 AL |
3007 | if (!swap_avail_heads) |
3008 | return -ENOMEM; | |
3009 | ||
53cbb243 | 3010 | p = alloc_swap_info(); |
2542e513 CEB |
3011 | if (IS_ERR(p)) |
3012 | return PTR_ERR(p); | |
53cbb243 | 3013 | |
815c2c54 SL |
3014 | INIT_WORK(&p->discard_work, swap_discard_work); |
3015 | ||
1da177e4 | 3016 | name = getname(specialfile); |
1da177e4 | 3017 | if (IS_ERR(name)) { |
7de7fb6b | 3018 | error = PTR_ERR(name); |
1da177e4 | 3019 | name = NULL; |
bd69010b | 3020 | goto bad_swap; |
1da177e4 | 3021 | } |
669abf4e | 3022 | swap_file = file_open_name(name, O_RDWR|O_LARGEFILE, 0); |
1da177e4 | 3023 | if (IS_ERR(swap_file)) { |
7de7fb6b | 3024 | error = PTR_ERR(swap_file); |
1da177e4 | 3025 | swap_file = NULL; |
bd69010b | 3026 | goto bad_swap; |
1da177e4 LT |
3027 | } |
3028 | ||
3029 | p->swap_file = swap_file; | |
3030 | mapping = swap_file->f_mapping; | |
51cc3a66 | 3031 | dentry = swap_file->f_path.dentry; |
2130781e | 3032 | inode = mapping->host; |
6f179af8 | 3033 | |
4d0e1e10 CEB |
3034 | error = claim_swapfile(p, inode); |
3035 | if (unlikely(error)) | |
1da177e4 | 3036 | goto bad_swap; |
1da177e4 | 3037 | |
d795a90e | 3038 | inode_lock(inode); |
51cc3a66 HD |
3039 | if (d_unlinked(dentry) || cant_mount(dentry)) { |
3040 | error = -ENOENT; | |
3041 | goto bad_swap_unlock_inode; | |
3042 | } | |
d795a90e NA |
3043 | if (IS_SWAPFILE(inode)) { |
3044 | error = -EBUSY; | |
3045 | goto bad_swap_unlock_inode; | |
3046 | } | |
3047 | ||
1da177e4 LT |
3048 | /* |
3049 | * Read the swap header. | |
3050 | */ | |
7e0a1265 | 3051 | if (!mapping->a_ops->read_folio) { |
1da177e4 | 3052 | error = -EINVAL; |
d795a90e | 3053 | goto bad_swap_unlock_inode; |
1da177e4 | 3054 | } |
090d2b18 | 3055 | page = read_mapping_page(mapping, 0, swap_file); |
1da177e4 LT |
3056 | if (IS_ERR(page)) { |
3057 | error = PTR_ERR(page); | |
d795a90e | 3058 | goto bad_swap_unlock_inode; |
1da177e4 | 3059 | } |
81e33971 | 3060 | swap_header = kmap(page); |
1da177e4 | 3061 | |
ca8bd38b CEB |
3062 | maxpages = read_swap_header(p, swap_header, inode); |
3063 | if (unlikely(!maxpages)) { | |
1da177e4 | 3064 | error = -EINVAL; |
d795a90e | 3065 | goto bad_swap_unlock_inode; |
1da177e4 | 3066 | } |
886bb7e9 | 3067 | |
81e33971 | 3068 | /* OK, set up the swap map and apply the bad block list */ |
803d0c83 | 3069 | swap_map = vzalloc(maxpages); |
81e33971 HD |
3070 | if (!swap_map) { |
3071 | error = -ENOMEM; | |
d795a90e | 3072 | goto bad_swap_unlock_inode; |
81e33971 | 3073 | } |
f0571429 | 3074 | |
36d25489 | 3075 | if (p->bdev && bdev_stable_writes(p->bdev)) |
f0571429 MK |
3076 | p->flags |= SWP_STABLE_WRITES; |
3077 | ||
a8b456d0 | 3078 | if (p->bdev && p->bdev->bd_disk->fops->rw_page) |
539a6fea MK |
3079 | p->flags |= SWP_SYNCHRONOUS_IO; |
3080 | ||
10f0d2a5 | 3081 | if (p->bdev && bdev_nonrot(p->bdev)) { |
6f179af8 | 3082 | int cpu; |
235b6217 | 3083 | unsigned long ci, nr_cluster; |
6f179af8 | 3084 | |
2a8f9449 | 3085 | p->flags |= SWP_SOLIDSTATE; |
49070588 HY |
3086 | p->cluster_next_cpu = alloc_percpu(unsigned int); |
3087 | if (!p->cluster_next_cpu) { | |
3088 | error = -ENOMEM; | |
3089 | goto bad_swap_unlock_inode; | |
3090 | } | |
2a8f9449 SL |
3091 | /* |
3092 | * select a random position to start with to help wear leveling | |
3093 | * SSD | |
3094 | */ | |
49070588 HY |
3095 | for_each_possible_cpu(cpu) { |
3096 | per_cpu(*p->cluster_next_cpu, cpu) = | |
3097 | 1 + prandom_u32_max(p->highest_bit); | |
3098 | } | |
235b6217 | 3099 | nr_cluster = DIV_ROUND_UP(maxpages, SWAPFILE_CLUSTER); |
2a8f9449 | 3100 | |
778e1cdd | 3101 | cluster_info = kvcalloc(nr_cluster, sizeof(*cluster_info), |
54f180d3 | 3102 | GFP_KERNEL); |
2a8f9449 SL |
3103 | if (!cluster_info) { |
3104 | error = -ENOMEM; | |
d795a90e | 3105 | goto bad_swap_unlock_inode; |
2a8f9449 | 3106 | } |
235b6217 HY |
3107 | |
3108 | for (ci = 0; ci < nr_cluster; ci++) | |
3109 | spin_lock_init(&((cluster_info + ci)->lock)); | |
3110 | ||
ebc2a1a6 SL |
3111 | p->percpu_cluster = alloc_percpu(struct percpu_cluster); |
3112 | if (!p->percpu_cluster) { | |
3113 | error = -ENOMEM; | |
d795a90e | 3114 | goto bad_swap_unlock_inode; |
ebc2a1a6 | 3115 | } |
6f179af8 | 3116 | for_each_possible_cpu(cpu) { |
ebc2a1a6 | 3117 | struct percpu_cluster *cluster; |
6f179af8 | 3118 | cluster = per_cpu_ptr(p->percpu_cluster, cpu); |
ebc2a1a6 SL |
3119 | cluster_set_null(&cluster->index); |
3120 | } | |
7cbf3192 | 3121 | } else { |
81a0298b | 3122 | atomic_inc(&nr_rotate_swap); |
7cbf3192 OS |
3123 | inced_nr_rotate_swap = true; |
3124 | } | |
1da177e4 | 3125 | |
1421ef3c CEB |
3126 | error = swap_cgroup_swapon(p->type, maxpages); |
3127 | if (error) | |
d795a90e | 3128 | goto bad_swap_unlock_inode; |
1421ef3c | 3129 | |
915d4d7b | 3130 | nr_extents = setup_swap_map_and_extents(p, swap_header, swap_map, |
2a8f9449 | 3131 | cluster_info, maxpages, &span); |
915d4d7b CEB |
3132 | if (unlikely(nr_extents < 0)) { |
3133 | error = nr_extents; | |
d795a90e | 3134 | goto bad_swap_unlock_inode; |
1da177e4 | 3135 | } |
38b5faf4 | 3136 | /* frontswap enabled? set up bit-per-page map for frontswap */ |
8ea1d2a1 | 3137 | if (IS_ENABLED(CONFIG_FRONTSWAP)) |
778e1cdd KC |
3138 | frontswap_map = kvcalloc(BITS_TO_LONGS(maxpages), |
3139 | sizeof(long), | |
54f180d3 | 3140 | GFP_KERNEL); |
1da177e4 | 3141 | |
70200574 CH |
3142 | if ((swap_flags & SWAP_FLAG_DISCARD) && |
3143 | p->bdev && bdev_max_discard_sectors(p->bdev)) { | |
2a8f9449 SL |
3144 | /* |
3145 | * When discard is enabled for swap with no particular | |
3146 | * policy flagged, we set all swap discard flags here in | |
3147 | * order to sustain backward compatibility with older | |
3148 | * swapon(8) releases. | |
3149 | */ | |
3150 | p->flags |= (SWP_DISCARDABLE | SWP_AREA_DISCARD | | |
3151 | SWP_PAGE_DISCARD); | |
dcf6b7dd | 3152 | |
2a8f9449 SL |
3153 | /* |
3154 | * By flagging sys_swapon, a sysadmin can tell us to | |
3155 | * either do single-time area discards only, or to just | |
3156 | * perform discards for released swap page-clusters. | |
3157 | * Now it's time to adjust the p->flags accordingly. | |
3158 | */ | |
3159 | if (swap_flags & SWAP_FLAG_DISCARD_ONCE) | |
3160 | p->flags &= ~SWP_PAGE_DISCARD; | |
3161 | else if (swap_flags & SWAP_FLAG_DISCARD_PAGES) | |
3162 | p->flags &= ~SWP_AREA_DISCARD; | |
3163 | ||
3164 | /* issue a swapon-time discard if it's still required */ | |
3165 | if (p->flags & SWP_AREA_DISCARD) { | |
3166 | int err = discard_swap(p); | |
3167 | if (unlikely(err)) | |
3168 | pr_err("swapon: discard_swap(%p): %d\n", | |
3169 | p, err); | |
dcf6b7dd | 3170 | } |
20137a49 | 3171 | } |
6a6ba831 | 3172 | |
4b3ef9da HY |
3173 | error = init_swap_address_space(p->type, maxpages); |
3174 | if (error) | |
d795a90e | 3175 | goto bad_swap_unlock_inode; |
4b3ef9da | 3176 | |
dc617f29 DW |
3177 | /* |
3178 | * Flush any pending IO and dirty mappings before we start using this | |
3179 | * swap device. | |
3180 | */ | |
3181 | inode->i_flags |= S_SWAPFILE; | |
3182 | error = inode_drain_writes(inode); | |
3183 | if (error) { | |
3184 | inode->i_flags &= ~S_SWAPFILE; | |
822bca52 | 3185 | goto free_swap_address_space; |
dc617f29 DW |
3186 | } |
3187 | ||
fc0abb14 | 3188 | mutex_lock(&swapon_mutex); |
40531542 | 3189 | prio = -1; |
78ecba08 | 3190 | if (swap_flags & SWAP_FLAG_PREFER) |
40531542 | 3191 | prio = |
78ecba08 | 3192 | (swap_flags & SWAP_FLAG_PRIO_MASK) >> SWAP_FLAG_PRIO_SHIFT; |
2a8f9449 | 3193 | enable_swap_info(p, prio, swap_map, cluster_info, frontswap_map); |
c69dbfb8 | 3194 | |
756a025f | 3195 | pr_info("Adding %uk swap on %s. Priority:%d extents:%d across:%lluk %s%s%s%s%s\n", |
91a27b2a | 3196 | p->pages<<(PAGE_SHIFT-10), name->name, p->prio, |
c69dbfb8 CEB |
3197 | nr_extents, (unsigned long long)span<<(PAGE_SHIFT-10), |
3198 | (p->flags & SWP_SOLIDSTATE) ? "SS" : "", | |
38b5faf4 | 3199 | (p->flags & SWP_DISCARDABLE) ? "D" : "", |
dcf6b7dd RA |
3200 | (p->flags & SWP_AREA_DISCARD) ? "s" : "", |
3201 | (p->flags & SWP_PAGE_DISCARD) ? "c" : "", | |
38b5faf4 | 3202 | (frontswap_map) ? "FS" : ""); |
c69dbfb8 | 3203 | |
fc0abb14 | 3204 | mutex_unlock(&swapon_mutex); |
66d7dd51 KS |
3205 | atomic_inc(&proc_poll_event); |
3206 | wake_up_interruptible(&proc_poll_wait); | |
3207 | ||
1da177e4 LT |
3208 | error = 0; |
3209 | goto out; | |
822bca52 ML |
3210 | free_swap_address_space: |
3211 | exit_swap_address_space(p->type); | |
d795a90e NA |
3212 | bad_swap_unlock_inode: |
3213 | inode_unlock(inode); | |
1da177e4 | 3214 | bad_swap: |
ebc2a1a6 SL |
3215 | free_percpu(p->percpu_cluster); |
3216 | p->percpu_cluster = NULL; | |
49070588 HY |
3217 | free_percpu(p->cluster_next_cpu); |
3218 | p->cluster_next_cpu = NULL; | |
bd69010b | 3219 | if (inode && S_ISBLK(inode->i_mode) && p->bdev) { |
f2090d2d CEB |
3220 | set_blocksize(p->bdev, p->old_block_size); |
3221 | blkdev_put(p->bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL); | |
1da177e4 | 3222 | } |
d795a90e | 3223 | inode = NULL; |
4cd3bb10 | 3224 | destroy_swap_extents(p); |
e8e6c2ec | 3225 | swap_cgroup_swapoff(p->type); |
5d337b91 | 3226 | spin_lock(&swap_lock); |
1da177e4 | 3227 | p->swap_file = NULL; |
1da177e4 | 3228 | p->flags = 0; |
5d337b91 | 3229 | spin_unlock(&swap_lock); |
1da177e4 | 3230 | vfree(swap_map); |
8606a1a9 | 3231 | kvfree(cluster_info); |
b6b1fd2a | 3232 | kvfree(frontswap_map); |
7cbf3192 OS |
3233 | if (inced_nr_rotate_swap) |
3234 | atomic_dec(&nr_rotate_swap); | |
d795a90e | 3235 | if (swap_file) |
1da177e4 LT |
3236 | filp_close(swap_file, NULL); |
3237 | out: | |
3238 | if (page && !IS_ERR(page)) { | |
3239 | kunmap(page); | |
09cbfeaf | 3240 | put_page(page); |
1da177e4 LT |
3241 | } |
3242 | if (name) | |
3243 | putname(name); | |
1638045c | 3244 | if (inode) |
5955102c | 3245 | inode_unlock(inode); |
039939a6 TC |
3246 | if (!error) |
3247 | enable_swap_slots_cache(); | |
1da177e4 LT |
3248 | return error; |
3249 | } | |
3250 | ||
3251 | void si_swapinfo(struct sysinfo *val) | |
3252 | { | |
efa90a98 | 3253 | unsigned int type; |
1da177e4 LT |
3254 | unsigned long nr_to_be_unused = 0; |
3255 | ||
5d337b91 | 3256 | spin_lock(&swap_lock); |
efa90a98 HD |
3257 | for (type = 0; type < nr_swapfiles; type++) { |
3258 | struct swap_info_struct *si = swap_info[type]; | |
3259 | ||
3260 | if ((si->flags & SWP_USED) && !(si->flags & SWP_WRITEOK)) | |
c8945306 | 3261 | nr_to_be_unused += READ_ONCE(si->inuse_pages); |
1da177e4 | 3262 | } |
ec8acf20 | 3263 | val->freeswap = atomic_long_read(&nr_swap_pages) + nr_to_be_unused; |
1da177e4 | 3264 | val->totalswap = total_swap_pages + nr_to_be_unused; |
5d337b91 | 3265 | spin_unlock(&swap_lock); |
1da177e4 LT |
3266 | } |
3267 | ||
3268 | /* | |
3269 | * Verify that a swap entry is valid and increment its swap map count. | |
3270 | * | |
355cfa73 KH |
3271 | * Returns error code in following case. |
3272 | * - success -> 0 | |
3273 | * - swp_entry is invalid -> EINVAL | |
3274 | * - swp_entry is migration entry -> EINVAL | |
3275 | * - swap-cache reference is requested but there is already one. -> EEXIST | |
3276 | * - swap-cache reference is requested but the entry is not used. -> ENOENT | |
570a335b | 3277 | * - swap-mapped reference requested but needs continued swap count. -> ENOMEM |
1da177e4 | 3278 | */ |
8d69aaee | 3279 | static int __swap_duplicate(swp_entry_t entry, unsigned char usage) |
1da177e4 | 3280 | { |
73c34b6a | 3281 | struct swap_info_struct *p; |
235b6217 | 3282 | struct swap_cluster_info *ci; |
c10d38cc | 3283 | unsigned long offset; |
8d69aaee HD |
3284 | unsigned char count; |
3285 | unsigned char has_cache; | |
9d9a0334 | 3286 | int err; |
1da177e4 | 3287 | |
eb085574 | 3288 | p = get_swap_device(entry); |
c10d38cc | 3289 | if (!p) |
9d9a0334 | 3290 | return -EINVAL; |
235b6217 | 3291 | |
eb085574 | 3292 | offset = swp_offset(entry); |
235b6217 | 3293 | ci = lock_cluster_or_swap_info(p, offset); |
355cfa73 | 3294 | |
253d553b | 3295 | count = p->swap_map[offset]; |
edfe23da SL |
3296 | |
3297 | /* | |
3298 | * swapin_readahead() doesn't check if a swap entry is valid, so the | |
3299 | * swap entry could be SWAP_MAP_BAD. Check here with lock held. | |
3300 | */ | |
3301 | if (unlikely(swap_count(count) == SWAP_MAP_BAD)) { | |
3302 | err = -ENOENT; | |
3303 | goto unlock_out; | |
3304 | } | |
3305 | ||
253d553b HD |
3306 | has_cache = count & SWAP_HAS_CACHE; |
3307 | count &= ~SWAP_HAS_CACHE; | |
3308 | err = 0; | |
355cfa73 | 3309 | |
253d553b | 3310 | if (usage == SWAP_HAS_CACHE) { |
355cfa73 KH |
3311 | |
3312 | /* set SWAP_HAS_CACHE if there is no cache and entry is used */ | |
253d553b HD |
3313 | if (!has_cache && count) |
3314 | has_cache = SWAP_HAS_CACHE; | |
3315 | else if (has_cache) /* someone else added cache */ | |
3316 | err = -EEXIST; | |
3317 | else /* no users remaining */ | |
3318 | err = -ENOENT; | |
355cfa73 KH |
3319 | |
3320 | } else if (count || has_cache) { | |
253d553b | 3321 | |
570a335b HD |
3322 | if ((count & ~COUNT_CONTINUED) < SWAP_MAP_MAX) |
3323 | count += usage; | |
3324 | else if ((count & ~COUNT_CONTINUED) > SWAP_MAP_MAX) | |
253d553b | 3325 | err = -EINVAL; |
570a335b HD |
3326 | else if (swap_count_continued(p, offset, count)) |
3327 | count = COUNT_CONTINUED; | |
3328 | else | |
3329 | err = -ENOMEM; | |
355cfa73 | 3330 | } else |
253d553b HD |
3331 | err = -ENOENT; /* unused swap entry */ |
3332 | ||
a449bf58 | 3333 | WRITE_ONCE(p->swap_map[offset], count | has_cache); |
253d553b | 3334 | |
355cfa73 | 3335 | unlock_out: |
235b6217 | 3336 | unlock_cluster_or_swap_info(p, ci); |
dab8dfff | 3337 | put_swap_device(p); |
253d553b | 3338 | return err; |
1da177e4 | 3339 | } |
253d553b | 3340 | |
aaa46865 HD |
3341 | /* |
3342 | * Help swapoff by noting that swap entry belongs to shmem/tmpfs | |
3343 | * (in which case its reference count is never incremented). | |
3344 | */ | |
3345 | void swap_shmem_alloc(swp_entry_t entry) | |
3346 | { | |
3347 | __swap_duplicate(entry, SWAP_MAP_SHMEM); | |
3348 | } | |
3349 | ||
355cfa73 | 3350 | /* |
08259d58 HD |
3351 | * Increase reference count of swap entry by 1. |
3352 | * Returns 0 for success, or -ENOMEM if a swap_count_continuation is required | |
3353 | * but could not be atomically allocated. Returns 0, just as if it succeeded, | |
3354 | * if __swap_duplicate() fails for another reason (-EINVAL or -ENOENT), which | |
3355 | * might occur if a page table entry has got corrupted. | |
355cfa73 | 3356 | */ |
570a335b | 3357 | int swap_duplicate(swp_entry_t entry) |
355cfa73 | 3358 | { |
570a335b HD |
3359 | int err = 0; |
3360 | ||
3361 | while (!err && __swap_duplicate(entry, 1) == -ENOMEM) | |
3362 | err = add_swap_count_continuation(entry, GFP_ATOMIC); | |
3363 | return err; | |
355cfa73 | 3364 | } |
1da177e4 | 3365 | |
cb4b86ba | 3366 | /* |
355cfa73 KH |
3367 | * @entry: swap entry for which we allocate swap cache. |
3368 | * | |
73c34b6a | 3369 | * Called when allocating swap cache for existing swap entry, |
355cfa73 | 3370 | * This can return error codes. Returns 0 at success. |
3eeba135 | 3371 | * -EEXIST means there is a swap cache. |
355cfa73 | 3372 | * Note: return code is different from swap_duplicate(). |
cb4b86ba KH |
3373 | */ |
3374 | int swapcache_prepare(swp_entry_t entry) | |
3375 | { | |
253d553b | 3376 | return __swap_duplicate(entry, SWAP_HAS_CACHE); |
cb4b86ba KH |
3377 | } |
3378 | ||
0bcac06f MK |
3379 | struct swap_info_struct *swp_swap_info(swp_entry_t entry) |
3380 | { | |
c10d38cc | 3381 | return swap_type_to_swap_info(swp_type(entry)); |
0bcac06f MK |
3382 | } |
3383 | ||
f981c595 MG |
3384 | struct swap_info_struct *page_swap_info(struct page *page) |
3385 | { | |
0bcac06f MK |
3386 | swp_entry_t entry = { .val = page_private(page) }; |
3387 | return swp_swap_info(entry); | |
f981c595 MG |
3388 | } |
3389 | ||
3390 | /* | |
2f52578f | 3391 | * out-of-line methods to avoid include hell. |
f981c595 | 3392 | */ |
2f52578f | 3393 | struct address_space *swapcache_mapping(struct folio *folio) |
f981c595 | 3394 | { |
2f52578f | 3395 | return page_swap_info(&folio->page)->swap_file->f_mapping; |
f981c595 | 3396 | } |
2f52578f | 3397 | EXPORT_SYMBOL_GPL(swapcache_mapping); |
f981c595 MG |
3398 | |
3399 | pgoff_t __page_file_index(struct page *page) | |
3400 | { | |
3401 | swp_entry_t swap = { .val = page_private(page) }; | |
f981c595 MG |
3402 | return swp_offset(swap); |
3403 | } | |
3404 | EXPORT_SYMBOL_GPL(__page_file_index); | |
3405 | ||
570a335b HD |
3406 | /* |
3407 | * add_swap_count_continuation - called when a swap count is duplicated | |
3408 | * beyond SWAP_MAP_MAX, it allocates a new page and links that to the entry's | |
3409 | * page of the original vmalloc'ed swap_map, to hold the continuation count | |
3410 | * (for that entry and for its neighbouring PAGE_SIZE swap entries). Called | |
3411 | * again when count is duplicated beyond SWAP_MAP_MAX * SWAP_CONT_MAX, etc. | |
3412 | * | |
3413 | * These continuation pages are seldom referenced: the common paths all work | |
3414 | * on the original swap_map, only referring to a continuation page when the | |
3415 | * low "digit" of a count is incremented or decremented through SWAP_MAP_MAX. | |
3416 | * | |
3417 | * add_swap_count_continuation(, GFP_ATOMIC) can be called while holding | |
3418 | * page table locks; if it fails, add_swap_count_continuation(, GFP_KERNEL) | |
3419 | * can be called after dropping locks. | |
3420 | */ | |
3421 | int add_swap_count_continuation(swp_entry_t entry, gfp_t gfp_mask) | |
3422 | { | |
3423 | struct swap_info_struct *si; | |
235b6217 | 3424 | struct swap_cluster_info *ci; |
570a335b HD |
3425 | struct page *head; |
3426 | struct page *page; | |
3427 | struct page *list_page; | |
3428 | pgoff_t offset; | |
3429 | unsigned char count; | |
eb085574 | 3430 | int ret = 0; |
570a335b HD |
3431 | |
3432 | /* | |
3433 | * When debugging, it's easier to use __GFP_ZERO here; but it's better | |
3434 | * for latency not to zero a page while GFP_ATOMIC and holding locks. | |
3435 | */ | |
3436 | page = alloc_page(gfp_mask | __GFP_HIGHMEM); | |
3437 | ||
eb085574 | 3438 | si = get_swap_device(entry); |
570a335b HD |
3439 | if (!si) { |
3440 | /* | |
3441 | * An acceptable race has occurred since the failing | |
eb085574 | 3442 | * __swap_duplicate(): the swap device may be swapoff |
570a335b HD |
3443 | */ |
3444 | goto outer; | |
3445 | } | |
eb085574 | 3446 | spin_lock(&si->lock); |
570a335b HD |
3447 | |
3448 | offset = swp_offset(entry); | |
235b6217 HY |
3449 | |
3450 | ci = lock_cluster(si, offset); | |
3451 | ||
d8aa24e0 | 3452 | count = swap_count(si->swap_map[offset]); |
570a335b HD |
3453 | |
3454 | if ((count & ~COUNT_CONTINUED) != SWAP_MAP_MAX) { | |
3455 | /* | |
3456 | * The higher the swap count, the more likely it is that tasks | |
3457 | * will race to add swap count continuation: we need to avoid | |
3458 | * over-provisioning. | |
3459 | */ | |
3460 | goto out; | |
3461 | } | |
3462 | ||
3463 | if (!page) { | |
eb085574 HY |
3464 | ret = -ENOMEM; |
3465 | goto out; | |
570a335b HD |
3466 | } |
3467 | ||
3468 | /* | |
3469 | * We are fortunate that although vmalloc_to_page uses pte_offset_map, | |
2de1a7e4 SJ |
3470 | * no architecture is using highmem pages for kernel page tables: so it |
3471 | * will not corrupt the GFP_ATOMIC caller's atomic page table kmaps. | |
570a335b HD |
3472 | */ |
3473 | head = vmalloc_to_page(si->swap_map + offset); | |
3474 | offset &= ~PAGE_MASK; | |
3475 | ||
2628bd6f | 3476 | spin_lock(&si->cont_lock); |
570a335b HD |
3477 | /* |
3478 | * Page allocation does not initialize the page's lru field, | |
3479 | * but it does always reset its private field. | |
3480 | */ | |
3481 | if (!page_private(head)) { | |
3482 | BUG_ON(count & COUNT_CONTINUED); | |
3483 | INIT_LIST_HEAD(&head->lru); | |
3484 | set_page_private(head, SWP_CONTINUED); | |
3485 | si->flags |= SWP_CONTINUED; | |
3486 | } | |
3487 | ||
3488 | list_for_each_entry(list_page, &head->lru, lru) { | |
3489 | unsigned char *map; | |
3490 | ||
3491 | /* | |
3492 | * If the previous map said no continuation, but we've found | |
3493 | * a continuation page, free our allocation and use this one. | |
3494 | */ | |
3495 | if (!(count & COUNT_CONTINUED)) | |
2628bd6f | 3496 | goto out_unlock_cont; |
570a335b | 3497 | |
9b04c5fe | 3498 | map = kmap_atomic(list_page) + offset; |
570a335b | 3499 | count = *map; |
9b04c5fe | 3500 | kunmap_atomic(map); |
570a335b HD |
3501 | |
3502 | /* | |
3503 | * If this continuation count now has some space in it, | |
3504 | * free our allocation and use this one. | |
3505 | */ | |
3506 | if ((count & ~COUNT_CONTINUED) != SWAP_CONT_MAX) | |
2628bd6f | 3507 | goto out_unlock_cont; |
570a335b HD |
3508 | } |
3509 | ||
3510 | list_add_tail(&page->lru, &head->lru); | |
3511 | page = NULL; /* now it's attached, don't free it */ | |
2628bd6f HY |
3512 | out_unlock_cont: |
3513 | spin_unlock(&si->cont_lock); | |
570a335b | 3514 | out: |
235b6217 | 3515 | unlock_cluster(ci); |
ec8acf20 | 3516 | spin_unlock(&si->lock); |
eb085574 | 3517 | put_swap_device(si); |
570a335b HD |
3518 | outer: |
3519 | if (page) | |
3520 | __free_page(page); | |
eb085574 | 3521 | return ret; |
570a335b HD |
3522 | } |
3523 | ||
3524 | /* | |
3525 | * swap_count_continued - when the original swap_map count is incremented | |
3526 | * from SWAP_MAP_MAX, check if there is already a continuation page to carry | |
3527 | * into, carry if so, or else fail until a new continuation page is allocated; | |
3528 | * when the original swap_map count is decremented from 0 with continuation, | |
3529 | * borrow from the continuation and report whether it still holds more. | |
235b6217 HY |
3530 | * Called while __swap_duplicate() or swap_entry_free() holds swap or cluster |
3531 | * lock. | |
570a335b HD |
3532 | */ |
3533 | static bool swap_count_continued(struct swap_info_struct *si, | |
3534 | pgoff_t offset, unsigned char count) | |
3535 | { | |
3536 | struct page *head; | |
3537 | struct page *page; | |
3538 | unsigned char *map; | |
2628bd6f | 3539 | bool ret; |
570a335b HD |
3540 | |
3541 | head = vmalloc_to_page(si->swap_map + offset); | |
3542 | if (page_private(head) != SWP_CONTINUED) { | |
3543 | BUG_ON(count & COUNT_CONTINUED); | |
3544 | return false; /* need to add count continuation */ | |
3545 | } | |
3546 | ||
2628bd6f | 3547 | spin_lock(&si->cont_lock); |
570a335b | 3548 | offset &= ~PAGE_MASK; |
213516ac | 3549 | page = list_next_entry(head, lru); |
9b04c5fe | 3550 | map = kmap_atomic(page) + offset; |
570a335b HD |
3551 | |
3552 | if (count == SWAP_MAP_MAX) /* initial increment from swap_map */ | |
3553 | goto init_map; /* jump over SWAP_CONT_MAX checks */ | |
3554 | ||
3555 | if (count == (SWAP_MAP_MAX | COUNT_CONTINUED)) { /* incrementing */ | |
3556 | /* | |
3557 | * Think of how you add 1 to 999 | |
3558 | */ | |
3559 | while (*map == (SWAP_CONT_MAX | COUNT_CONTINUED)) { | |
9b04c5fe | 3560 | kunmap_atomic(map); |
213516ac | 3561 | page = list_next_entry(page, lru); |
570a335b | 3562 | BUG_ON(page == head); |
9b04c5fe | 3563 | map = kmap_atomic(page) + offset; |
570a335b HD |
3564 | } |
3565 | if (*map == SWAP_CONT_MAX) { | |
9b04c5fe | 3566 | kunmap_atomic(map); |
213516ac | 3567 | page = list_next_entry(page, lru); |
2628bd6f HY |
3568 | if (page == head) { |
3569 | ret = false; /* add count continuation */ | |
3570 | goto out; | |
3571 | } | |
9b04c5fe | 3572 | map = kmap_atomic(page) + offset; |
570a335b HD |
3573 | init_map: *map = 0; /* we didn't zero the page */ |
3574 | } | |
3575 | *map += 1; | |
9b04c5fe | 3576 | kunmap_atomic(map); |
213516ac | 3577 | while ((page = list_prev_entry(page, lru)) != head) { |
9b04c5fe | 3578 | map = kmap_atomic(page) + offset; |
570a335b | 3579 | *map = COUNT_CONTINUED; |
9b04c5fe | 3580 | kunmap_atomic(map); |
570a335b | 3581 | } |
2628bd6f | 3582 | ret = true; /* incremented */ |
570a335b HD |
3583 | |
3584 | } else { /* decrementing */ | |
3585 | /* | |
3586 | * Think of how you subtract 1 from 1000 | |
3587 | */ | |
3588 | BUG_ON(count != COUNT_CONTINUED); | |
3589 | while (*map == COUNT_CONTINUED) { | |
9b04c5fe | 3590 | kunmap_atomic(map); |
213516ac | 3591 | page = list_next_entry(page, lru); |
570a335b | 3592 | BUG_ON(page == head); |
9b04c5fe | 3593 | map = kmap_atomic(page) + offset; |
570a335b HD |
3594 | } |
3595 | BUG_ON(*map == 0); | |
3596 | *map -= 1; | |
3597 | if (*map == 0) | |
3598 | count = 0; | |
9b04c5fe | 3599 | kunmap_atomic(map); |
213516ac | 3600 | while ((page = list_prev_entry(page, lru)) != head) { |
9b04c5fe | 3601 | map = kmap_atomic(page) + offset; |
570a335b HD |
3602 | *map = SWAP_CONT_MAX | count; |
3603 | count = COUNT_CONTINUED; | |
9b04c5fe | 3604 | kunmap_atomic(map); |
570a335b | 3605 | } |
2628bd6f | 3606 | ret = count == COUNT_CONTINUED; |
570a335b | 3607 | } |
2628bd6f HY |
3608 | out: |
3609 | spin_unlock(&si->cont_lock); | |
3610 | return ret; | |
570a335b HD |
3611 | } |
3612 | ||
3613 | /* | |
3614 | * free_swap_count_continuations - swapoff free all the continuation pages | |
3615 | * appended to the swap_map, after swap_map is quiesced, before vfree'ing it. | |
3616 | */ | |
3617 | static void free_swap_count_continuations(struct swap_info_struct *si) | |
3618 | { | |
3619 | pgoff_t offset; | |
3620 | ||
3621 | for (offset = 0; offset < si->max; offset += PAGE_SIZE) { | |
3622 | struct page *head; | |
3623 | head = vmalloc_to_page(si->swap_map + offset); | |
3624 | if (page_private(head)) { | |
0d576d20 GT |
3625 | struct page *page, *next; |
3626 | ||
3627 | list_for_each_entry_safe(page, next, &head->lru, lru) { | |
3628 | list_del(&page->lru); | |
570a335b HD |
3629 | __free_page(page); |
3630 | } | |
3631 | } | |
3632 | } | |
3633 | } | |
a2468cc9 | 3634 | |
2cf85583 | 3635 | #if defined(CONFIG_MEMCG) && defined(CONFIG_BLK_CGROUP) |
01c4b28c | 3636 | void __cgroup_throttle_swaprate(struct page *page, gfp_t gfp_mask) |
2cf85583 TH |
3637 | { |
3638 | struct swap_info_struct *si, *next; | |
6caa6a07 JW |
3639 | int nid = page_to_nid(page); |
3640 | ||
3641 | if (!(gfp_mask & __GFP_IO)) | |
2cf85583 TH |
3642 | return; |
3643 | ||
3644 | if (!blk_cgroup_congested()) | |
3645 | return; | |
3646 | ||
3647 | /* | |
3648 | * We've already scheduled a throttle, avoid taking the global swap | |
3649 | * lock. | |
3650 | */ | |
3651 | if (current->throttle_queue) | |
3652 | return; | |
3653 | ||
3654 | spin_lock(&swap_avail_lock); | |
6caa6a07 JW |
3655 | plist_for_each_entry_safe(si, next, &swap_avail_heads[nid], |
3656 | avail_lists[nid]) { | |
2cf85583 | 3657 | if (si->bdev) { |
de185b56 | 3658 | blkcg_schedule_throttle(si->bdev->bd_disk, true); |
2cf85583 TH |
3659 | break; |
3660 | } | |
3661 | } | |
3662 | spin_unlock(&swap_avail_lock); | |
3663 | } | |
3664 | #endif | |
3665 | ||
a2468cc9 AL |
3666 | static int __init swapfile_init(void) |
3667 | { | |
3668 | int nid; | |
3669 | ||
3670 | swap_avail_heads = kmalloc_array(nr_node_ids, sizeof(struct plist_head), | |
3671 | GFP_KERNEL); | |
3672 | if (!swap_avail_heads) { | |
3673 | pr_emerg("Not enough memory for swap heads, swap is disabled\n"); | |
3674 | return -ENOMEM; | |
3675 | } | |
3676 | ||
3677 | for_each_node(nid) | |
3678 | plist_head_init(&swap_avail_heads[nid]); | |
3679 | ||
3680 | return 0; | |
3681 | } | |
3682 | subsys_initcall(swapfile_init); |