]>
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 | ||
1da177e4 | 9 | #include <linux/mm.h> |
6e84f315 | 10 | #include <linux/sched/mm.h> |
29930025 | 11 | #include <linux/sched/task.h> |
1da177e4 LT |
12 | #include <linux/hugetlb.h> |
13 | #include <linux/mman.h> | |
14 | #include <linux/slab.h> | |
15 | #include <linux/kernel_stat.h> | |
16 | #include <linux/swap.h> | |
17 | #include <linux/vmalloc.h> | |
18 | #include <linux/pagemap.h> | |
19 | #include <linux/namei.h> | |
072441e2 | 20 | #include <linux/shmem_fs.h> |
1da177e4 | 21 | #include <linux/blkdev.h> |
20137a49 | 22 | #include <linux/random.h> |
1da177e4 LT |
23 | #include <linux/writeback.h> |
24 | #include <linux/proc_fs.h> | |
25 | #include <linux/seq_file.h> | |
26 | #include <linux/init.h> | |
5ad64688 | 27 | #include <linux/ksm.h> |
1da177e4 LT |
28 | #include <linux/rmap.h> |
29 | #include <linux/security.h> | |
30 | #include <linux/backing-dev.h> | |
fc0abb14 | 31 | #include <linux/mutex.h> |
c59ede7b | 32 | #include <linux/capability.h> |
1da177e4 | 33 | #include <linux/syscalls.h> |
8a9f3ccd | 34 | #include <linux/memcontrol.h> |
66d7dd51 | 35 | #include <linux/poll.h> |
72788c38 | 36 | #include <linux/oom.h> |
38b5faf4 DM |
37 | #include <linux/frontswap.h> |
38 | #include <linux/swapfile.h> | |
f981c595 | 39 | #include <linux/export.h> |
67afa38e | 40 | #include <linux/swap_slots.h> |
155b5f88 | 41 | #include <linux/sort.h> |
63d8620e | 42 | #include <linux/completion.h> |
1da177e4 | 43 | |
1da177e4 LT |
44 | #include <asm/tlbflush.h> |
45 | #include <linux/swapops.h> | |
5d1ea48b | 46 | #include <linux/swap_cgroup.h> |
1da177e4 | 47 | |
570a335b HD |
48 | static bool swap_count_continued(struct swap_info_struct *, pgoff_t, |
49 | unsigned char); | |
50 | static void free_swap_count_continuations(struct swap_info_struct *); | |
51 | ||
38b5faf4 | 52 | DEFINE_SPINLOCK(swap_lock); |
7c363b8c | 53 | static unsigned int nr_swapfiles; |
ec8acf20 | 54 | atomic_long_t nr_swap_pages; |
fb0fec50 CW |
55 | /* |
56 | * Some modules use swappable objects and may try to swap them out under | |
57 | * memory pressure (via the shrinker). Before doing so, they may wish to | |
58 | * check to see if any swap space is available. | |
59 | */ | |
60 | EXPORT_SYMBOL_GPL(nr_swap_pages); | |
ec8acf20 | 61 | /* protected with swap_lock. reading in vm_swap_full() doesn't need lock */ |
1da177e4 | 62 | long total_swap_pages; |
a2468cc9 | 63 | static int least_priority = -1; |
1da177e4 | 64 | |
1da177e4 LT |
65 | static const char Bad_file[] = "Bad swap file entry "; |
66 | static const char Unused_file[] = "Unused swap file entry "; | |
67 | static const char Bad_offset[] = "Bad swap offset entry "; | |
68 | static const char Unused_offset[] = "Unused swap offset entry "; | |
69 | ||
adfab836 DS |
70 | /* |
71 | * all active swap_info_structs | |
72 | * protected with swap_lock, and ordered by priority. | |
73 | */ | |
18ab4d4c DS |
74 | PLIST_HEAD(swap_active_head); |
75 | ||
76 | /* | |
77 | * all available (active, not full) swap_info_structs | |
78 | * protected with swap_avail_lock, ordered by priority. | |
79 | * This is used by get_swap_page() instead of swap_active_head | |
80 | * because swap_active_head includes all swap_info_structs, | |
81 | * but get_swap_page() doesn't need to look at full ones. | |
82 | * This uses its own lock instead of swap_lock because when a | |
83 | * swap_info_struct changes between not-full/full, it needs to | |
84 | * add/remove itself to/from this list, but the swap_info_struct->lock | |
85 | * is held and the locking order requires swap_lock to be taken | |
86 | * before any swap_info_struct->lock. | |
87 | */ | |
bfc6b1ca | 88 | static struct plist_head *swap_avail_heads; |
18ab4d4c | 89 | static DEFINE_SPINLOCK(swap_avail_lock); |
1da177e4 | 90 | |
38b5faf4 | 91 | struct swap_info_struct *swap_info[MAX_SWAPFILES]; |
1da177e4 | 92 | |
fc0abb14 | 93 | static DEFINE_MUTEX(swapon_mutex); |
1da177e4 | 94 | |
66d7dd51 KS |
95 | static DECLARE_WAIT_QUEUE_HEAD(proc_poll_wait); |
96 | /* Activity counter to indicate that a swapon or swapoff has occurred */ | |
97 | static atomic_t proc_poll_event = ATOMIC_INIT(0); | |
98 | ||
81a0298b HY |
99 | atomic_t nr_rotate_swap = ATOMIC_INIT(0); |
100 | ||
c10d38cc DJ |
101 | static struct swap_info_struct *swap_type_to_swap_info(int type) |
102 | { | |
103 | if (type >= READ_ONCE(nr_swapfiles)) | |
104 | return NULL; | |
105 | ||
106 | smp_rmb(); /* Pairs with smp_wmb in alloc_swap_info. */ | |
107 | return READ_ONCE(swap_info[type]); | |
108 | } | |
109 | ||
8d69aaee | 110 | static inline unsigned char swap_count(unsigned char ent) |
355cfa73 | 111 | { |
955c97f0 | 112 | return ent & ~SWAP_HAS_CACHE; /* may include COUNT_CONTINUED flag */ |
355cfa73 KH |
113 | } |
114 | ||
bcd49e86 HY |
115 | /* Reclaim the swap entry anyway if possible */ |
116 | #define TTRS_ANYWAY 0x1 | |
117 | /* | |
118 | * Reclaim the swap entry if there are no more mappings of the | |
119 | * corresponding page | |
120 | */ | |
121 | #define TTRS_UNMAPPED 0x2 | |
122 | /* Reclaim the swap entry if swap is getting full*/ | |
123 | #define TTRS_FULL 0x4 | |
124 | ||
efa90a98 | 125 | /* returns 1 if swap entry is freed */ |
bcd49e86 HY |
126 | static int __try_to_reclaim_swap(struct swap_info_struct *si, |
127 | unsigned long offset, unsigned long flags) | |
c9e44410 | 128 | { |
efa90a98 | 129 | swp_entry_t entry = swp_entry(si->type, offset); |
c9e44410 KH |
130 | struct page *page; |
131 | int ret = 0; | |
132 | ||
bcd49e86 | 133 | page = find_get_page(swap_address_space(entry), offset); |
c9e44410 KH |
134 | if (!page) |
135 | return 0; | |
136 | /* | |
bcd49e86 HY |
137 | * When this function is called from scan_swap_map_slots() and it's |
138 | * called by vmscan.c at reclaiming pages. So, we hold a lock on a page, | |
139 | * here. We have to use trylock for avoiding deadlock. This is a special | |
c9e44410 KH |
140 | * case and you should use try_to_free_swap() with explicit lock_page() |
141 | * in usual operations. | |
142 | */ | |
143 | if (trylock_page(page)) { | |
bcd49e86 HY |
144 | if ((flags & TTRS_ANYWAY) || |
145 | ((flags & TTRS_UNMAPPED) && !page_mapped(page)) || | |
146 | ((flags & TTRS_FULL) && mem_cgroup_swap_full(page))) | |
147 | ret = try_to_free_swap(page); | |
c9e44410 KH |
148 | unlock_page(page); |
149 | } | |
09cbfeaf | 150 | put_page(page); |
c9e44410 KH |
151 | return ret; |
152 | } | |
355cfa73 | 153 | |
4efaceb1 AL |
154 | static inline struct swap_extent *first_se(struct swap_info_struct *sis) |
155 | { | |
156 | struct rb_node *rb = rb_first(&sis->swap_extent_root); | |
157 | return rb_entry(rb, struct swap_extent, rb_node); | |
158 | } | |
159 | ||
160 | static inline struct swap_extent *next_se(struct swap_extent *se) | |
161 | { | |
162 | struct rb_node *rb = rb_next(&se->rb_node); | |
163 | return rb ? rb_entry(rb, struct swap_extent, rb_node) : NULL; | |
164 | } | |
165 | ||
6a6ba831 HD |
166 | /* |
167 | * swapon tell device that all the old swap contents can be discarded, | |
168 | * to allow the swap device to optimize its wear-levelling. | |
169 | */ | |
170 | static int discard_swap(struct swap_info_struct *si) | |
171 | { | |
172 | struct swap_extent *se; | |
9625a5f2 HD |
173 | sector_t start_block; |
174 | sector_t nr_blocks; | |
6a6ba831 HD |
175 | int err = 0; |
176 | ||
9625a5f2 | 177 | /* Do not discard the swap header page! */ |
4efaceb1 | 178 | se = first_se(si); |
9625a5f2 HD |
179 | start_block = (se->start_block + 1) << (PAGE_SHIFT - 9); |
180 | nr_blocks = ((sector_t)se->nr_pages - 1) << (PAGE_SHIFT - 9); | |
181 | if (nr_blocks) { | |
182 | err = blkdev_issue_discard(si->bdev, start_block, | |
dd3932ed | 183 | nr_blocks, GFP_KERNEL, 0); |
9625a5f2 HD |
184 | if (err) |
185 | return err; | |
186 | cond_resched(); | |
187 | } | |
6a6ba831 | 188 | |
4efaceb1 | 189 | for (se = next_se(se); se; se = next_se(se)) { |
9625a5f2 HD |
190 | start_block = se->start_block << (PAGE_SHIFT - 9); |
191 | nr_blocks = (sector_t)se->nr_pages << (PAGE_SHIFT - 9); | |
6a6ba831 HD |
192 | |
193 | err = blkdev_issue_discard(si->bdev, start_block, | |
dd3932ed | 194 | nr_blocks, GFP_KERNEL, 0); |
6a6ba831 HD |
195 | if (err) |
196 | break; | |
197 | ||
198 | cond_resched(); | |
199 | } | |
200 | return err; /* That will often be -EOPNOTSUPP */ | |
201 | } | |
202 | ||
4efaceb1 AL |
203 | static struct swap_extent * |
204 | offset_to_swap_extent(struct swap_info_struct *sis, unsigned long offset) | |
205 | { | |
206 | struct swap_extent *se; | |
207 | struct rb_node *rb; | |
208 | ||
209 | rb = sis->swap_extent_root.rb_node; | |
210 | while (rb) { | |
211 | se = rb_entry(rb, struct swap_extent, rb_node); | |
212 | if (offset < se->start_page) | |
213 | rb = rb->rb_left; | |
214 | else if (offset >= se->start_page + se->nr_pages) | |
215 | rb = rb->rb_right; | |
216 | else | |
217 | return se; | |
218 | } | |
219 | /* It *must* be present */ | |
220 | BUG(); | |
221 | } | |
222 | ||
caf6912f JA |
223 | sector_t swap_page_sector(struct page *page) |
224 | { | |
225 | struct swap_info_struct *sis = page_swap_info(page); | |
226 | struct swap_extent *se; | |
227 | sector_t sector; | |
228 | pgoff_t offset; | |
229 | ||
230 | offset = __page_file_index(page); | |
231 | se = offset_to_swap_extent(sis, offset); | |
232 | sector = se->start_block + (offset - se->start_page); | |
233 | return sector << (PAGE_SHIFT - 9); | |
234 | } | |
235 | ||
7992fde7 HD |
236 | /* |
237 | * swap allocation tell device that a cluster of swap can now be discarded, | |
238 | * to allow the swap device to optimize its wear-levelling. | |
239 | */ | |
240 | static void discard_swap_cluster(struct swap_info_struct *si, | |
241 | pgoff_t start_page, pgoff_t nr_pages) | |
242 | { | |
4efaceb1 | 243 | struct swap_extent *se = offset_to_swap_extent(si, start_page); |
7992fde7 HD |
244 | |
245 | while (nr_pages) { | |
4efaceb1 AL |
246 | pgoff_t offset = start_page - se->start_page; |
247 | sector_t start_block = se->start_block + offset; | |
248 | sector_t nr_blocks = se->nr_pages - offset; | |
249 | ||
250 | if (nr_blocks > nr_pages) | |
251 | nr_blocks = nr_pages; | |
252 | start_page += nr_blocks; | |
253 | nr_pages -= nr_blocks; | |
254 | ||
255 | start_block <<= PAGE_SHIFT - 9; | |
256 | nr_blocks <<= PAGE_SHIFT - 9; | |
257 | if (blkdev_issue_discard(si->bdev, start_block, | |
258 | nr_blocks, GFP_NOIO, 0)) | |
259 | break; | |
7992fde7 | 260 | |
4efaceb1 | 261 | se = next_se(se); |
7992fde7 HD |
262 | } |
263 | } | |
264 | ||
38d8b4e6 HY |
265 | #ifdef CONFIG_THP_SWAP |
266 | #define SWAPFILE_CLUSTER HPAGE_PMD_NR | |
a448f2d0 HY |
267 | |
268 | #define swap_entry_size(size) (size) | |
38d8b4e6 | 269 | #else |
048c27fd | 270 | #define SWAPFILE_CLUSTER 256 |
a448f2d0 HY |
271 | |
272 | /* | |
273 | * Define swap_entry_size() as constant to let compiler to optimize | |
274 | * out some code if !CONFIG_THP_SWAP | |
275 | */ | |
276 | #define swap_entry_size(size) 1 | |
38d8b4e6 | 277 | #endif |
048c27fd HD |
278 | #define LATENCY_LIMIT 256 |
279 | ||
2a8f9449 SL |
280 | static inline void cluster_set_flag(struct swap_cluster_info *info, |
281 | unsigned int flag) | |
282 | { | |
283 | info->flags = flag; | |
284 | } | |
285 | ||
286 | static inline unsigned int cluster_count(struct swap_cluster_info *info) | |
287 | { | |
288 | return info->data; | |
289 | } | |
290 | ||
291 | static inline void cluster_set_count(struct swap_cluster_info *info, | |
292 | unsigned int c) | |
293 | { | |
294 | info->data = c; | |
295 | } | |
296 | ||
297 | static inline void cluster_set_count_flag(struct swap_cluster_info *info, | |
298 | unsigned int c, unsigned int f) | |
299 | { | |
300 | info->flags = f; | |
301 | info->data = c; | |
302 | } | |
303 | ||
304 | static inline unsigned int cluster_next(struct swap_cluster_info *info) | |
305 | { | |
306 | return info->data; | |
307 | } | |
308 | ||
309 | static inline void cluster_set_next(struct swap_cluster_info *info, | |
310 | unsigned int n) | |
311 | { | |
312 | info->data = n; | |
313 | } | |
314 | ||
315 | static inline void cluster_set_next_flag(struct swap_cluster_info *info, | |
316 | unsigned int n, unsigned int f) | |
317 | { | |
318 | info->flags = f; | |
319 | info->data = n; | |
320 | } | |
321 | ||
322 | static inline bool cluster_is_free(struct swap_cluster_info *info) | |
323 | { | |
324 | return info->flags & CLUSTER_FLAG_FREE; | |
325 | } | |
326 | ||
327 | static inline bool cluster_is_null(struct swap_cluster_info *info) | |
328 | { | |
329 | return info->flags & CLUSTER_FLAG_NEXT_NULL; | |
330 | } | |
331 | ||
332 | static inline void cluster_set_null(struct swap_cluster_info *info) | |
333 | { | |
334 | info->flags = CLUSTER_FLAG_NEXT_NULL; | |
335 | info->data = 0; | |
336 | } | |
337 | ||
e0709829 HY |
338 | static inline bool cluster_is_huge(struct swap_cluster_info *info) |
339 | { | |
33ee011e HY |
340 | if (IS_ENABLED(CONFIG_THP_SWAP)) |
341 | return info->flags & CLUSTER_FLAG_HUGE; | |
342 | return false; | |
e0709829 HY |
343 | } |
344 | ||
345 | static inline void cluster_clear_huge(struct swap_cluster_info *info) | |
346 | { | |
347 | info->flags &= ~CLUSTER_FLAG_HUGE; | |
348 | } | |
349 | ||
235b6217 HY |
350 | static inline struct swap_cluster_info *lock_cluster(struct swap_info_struct *si, |
351 | unsigned long offset) | |
352 | { | |
353 | struct swap_cluster_info *ci; | |
354 | ||
355 | ci = si->cluster_info; | |
356 | if (ci) { | |
357 | ci += offset / SWAPFILE_CLUSTER; | |
358 | spin_lock(&ci->lock); | |
359 | } | |
360 | return ci; | |
361 | } | |
362 | ||
363 | static inline void unlock_cluster(struct swap_cluster_info *ci) | |
364 | { | |
365 | if (ci) | |
366 | spin_unlock(&ci->lock); | |
367 | } | |
368 | ||
59d98bf3 HY |
369 | /* |
370 | * Determine the locking method in use for this device. Return | |
371 | * swap_cluster_info if SSD-style cluster-based locking is in place. | |
372 | */ | |
235b6217 | 373 | static inline struct swap_cluster_info *lock_cluster_or_swap_info( |
59d98bf3 | 374 | struct swap_info_struct *si, unsigned long offset) |
235b6217 HY |
375 | { |
376 | struct swap_cluster_info *ci; | |
377 | ||
59d98bf3 | 378 | /* Try to use fine-grained SSD-style locking if available: */ |
235b6217 | 379 | ci = lock_cluster(si, offset); |
59d98bf3 | 380 | /* Otherwise, fall back to traditional, coarse locking: */ |
235b6217 HY |
381 | if (!ci) |
382 | spin_lock(&si->lock); | |
383 | ||
384 | return ci; | |
385 | } | |
386 | ||
387 | static inline void unlock_cluster_or_swap_info(struct swap_info_struct *si, | |
388 | struct swap_cluster_info *ci) | |
389 | { | |
390 | if (ci) | |
391 | unlock_cluster(ci); | |
392 | else | |
393 | spin_unlock(&si->lock); | |
394 | } | |
395 | ||
6b534915 HY |
396 | static inline bool cluster_list_empty(struct swap_cluster_list *list) |
397 | { | |
398 | return cluster_is_null(&list->head); | |
399 | } | |
400 | ||
401 | static inline unsigned int cluster_list_first(struct swap_cluster_list *list) | |
402 | { | |
403 | return cluster_next(&list->head); | |
404 | } | |
405 | ||
406 | static void cluster_list_init(struct swap_cluster_list *list) | |
407 | { | |
408 | cluster_set_null(&list->head); | |
409 | cluster_set_null(&list->tail); | |
410 | } | |
411 | ||
412 | static void cluster_list_add_tail(struct swap_cluster_list *list, | |
413 | struct swap_cluster_info *ci, | |
414 | unsigned int idx) | |
415 | { | |
416 | if (cluster_list_empty(list)) { | |
417 | cluster_set_next_flag(&list->head, idx, 0); | |
418 | cluster_set_next_flag(&list->tail, idx, 0); | |
419 | } else { | |
235b6217 | 420 | struct swap_cluster_info *ci_tail; |
6b534915 HY |
421 | unsigned int tail = cluster_next(&list->tail); |
422 | ||
235b6217 HY |
423 | /* |
424 | * Nested cluster lock, but both cluster locks are | |
425 | * only acquired when we held swap_info_struct->lock | |
426 | */ | |
427 | ci_tail = ci + tail; | |
428 | spin_lock_nested(&ci_tail->lock, SINGLE_DEPTH_NESTING); | |
429 | cluster_set_next(ci_tail, idx); | |
0ef017d1 | 430 | spin_unlock(&ci_tail->lock); |
6b534915 HY |
431 | cluster_set_next_flag(&list->tail, idx, 0); |
432 | } | |
433 | } | |
434 | ||
435 | static unsigned int cluster_list_del_first(struct swap_cluster_list *list, | |
436 | struct swap_cluster_info *ci) | |
437 | { | |
438 | unsigned int idx; | |
439 | ||
440 | idx = cluster_next(&list->head); | |
441 | if (cluster_next(&list->tail) == idx) { | |
442 | cluster_set_null(&list->head); | |
443 | cluster_set_null(&list->tail); | |
444 | } else | |
445 | cluster_set_next_flag(&list->head, | |
446 | cluster_next(&ci[idx]), 0); | |
447 | ||
448 | return idx; | |
449 | } | |
450 | ||
815c2c54 SL |
451 | /* Add a cluster to discard list and schedule it to do discard */ |
452 | static void swap_cluster_schedule_discard(struct swap_info_struct *si, | |
453 | unsigned int idx) | |
454 | { | |
455 | /* | |
456 | * If scan_swap_map() can't find a free cluster, it will check | |
457 | * si->swap_map directly. To make sure the discarding cluster isn't | |
458 | * taken by scan_swap_map(), mark the swap entries bad (occupied). It | |
459 | * will be cleared after discard | |
460 | */ | |
461 | memset(si->swap_map + idx * SWAPFILE_CLUSTER, | |
462 | SWAP_MAP_BAD, SWAPFILE_CLUSTER); | |
463 | ||
6b534915 | 464 | cluster_list_add_tail(&si->discard_clusters, si->cluster_info, idx); |
815c2c54 SL |
465 | |
466 | schedule_work(&si->discard_work); | |
467 | } | |
468 | ||
38d8b4e6 HY |
469 | static void __free_cluster(struct swap_info_struct *si, unsigned long idx) |
470 | { | |
471 | struct swap_cluster_info *ci = si->cluster_info; | |
472 | ||
473 | cluster_set_flag(ci + idx, CLUSTER_FLAG_FREE); | |
474 | cluster_list_add_tail(&si->free_clusters, ci, idx); | |
475 | } | |
476 | ||
815c2c54 SL |
477 | /* |
478 | * Doing discard actually. After a cluster discard is finished, the cluster | |
479 | * will be added to free cluster list. caller should hold si->lock. | |
480 | */ | |
481 | static void swap_do_scheduled_discard(struct swap_info_struct *si) | |
482 | { | |
235b6217 | 483 | struct swap_cluster_info *info, *ci; |
815c2c54 SL |
484 | unsigned int idx; |
485 | ||
486 | info = si->cluster_info; | |
487 | ||
6b534915 HY |
488 | while (!cluster_list_empty(&si->discard_clusters)) { |
489 | idx = cluster_list_del_first(&si->discard_clusters, info); | |
815c2c54 SL |
490 | spin_unlock(&si->lock); |
491 | ||
492 | discard_swap_cluster(si, idx * SWAPFILE_CLUSTER, | |
493 | SWAPFILE_CLUSTER); | |
494 | ||
495 | spin_lock(&si->lock); | |
235b6217 | 496 | ci = lock_cluster(si, idx * SWAPFILE_CLUSTER); |
38d8b4e6 | 497 | __free_cluster(si, idx); |
815c2c54 SL |
498 | memset(si->swap_map + idx * SWAPFILE_CLUSTER, |
499 | 0, SWAPFILE_CLUSTER); | |
235b6217 | 500 | unlock_cluster(ci); |
815c2c54 SL |
501 | } |
502 | } | |
503 | ||
504 | static void swap_discard_work(struct work_struct *work) | |
505 | { | |
506 | struct swap_info_struct *si; | |
507 | ||
508 | si = container_of(work, struct swap_info_struct, discard_work); | |
509 | ||
510 | spin_lock(&si->lock); | |
511 | swap_do_scheduled_discard(si); | |
512 | spin_unlock(&si->lock); | |
513 | } | |
514 | ||
63d8620e ML |
515 | static void swap_users_ref_free(struct percpu_ref *ref) |
516 | { | |
517 | struct swap_info_struct *si; | |
518 | ||
519 | si = container_of(ref, struct swap_info_struct, users); | |
520 | complete(&si->comp); | |
521 | } | |
522 | ||
38d8b4e6 HY |
523 | static void alloc_cluster(struct swap_info_struct *si, unsigned long idx) |
524 | { | |
525 | struct swap_cluster_info *ci = si->cluster_info; | |
526 | ||
527 | VM_BUG_ON(cluster_list_first(&si->free_clusters) != idx); | |
528 | cluster_list_del_first(&si->free_clusters, ci); | |
529 | cluster_set_count_flag(ci + idx, 0, 0); | |
530 | } | |
531 | ||
532 | static void free_cluster(struct swap_info_struct *si, unsigned long idx) | |
533 | { | |
534 | struct swap_cluster_info *ci = si->cluster_info + idx; | |
535 | ||
536 | VM_BUG_ON(cluster_count(ci) != 0); | |
537 | /* | |
538 | * If the swap is discardable, prepare discard the cluster | |
539 | * instead of free it immediately. The cluster will be freed | |
540 | * after discard. | |
541 | */ | |
542 | if ((si->flags & (SWP_WRITEOK | SWP_PAGE_DISCARD)) == | |
543 | (SWP_WRITEOK | SWP_PAGE_DISCARD)) { | |
544 | swap_cluster_schedule_discard(si, idx); | |
545 | return; | |
546 | } | |
547 | ||
548 | __free_cluster(si, idx); | |
549 | } | |
550 | ||
2a8f9449 SL |
551 | /* |
552 | * The cluster corresponding to page_nr will be used. The cluster will be | |
553 | * removed from free cluster list and its usage counter will be increased. | |
554 | */ | |
555 | static void inc_cluster_info_page(struct swap_info_struct *p, | |
556 | struct swap_cluster_info *cluster_info, unsigned long page_nr) | |
557 | { | |
558 | unsigned long idx = page_nr / SWAPFILE_CLUSTER; | |
559 | ||
560 | if (!cluster_info) | |
561 | return; | |
38d8b4e6 HY |
562 | if (cluster_is_free(&cluster_info[idx])) |
563 | alloc_cluster(p, idx); | |
2a8f9449 SL |
564 | |
565 | VM_BUG_ON(cluster_count(&cluster_info[idx]) >= SWAPFILE_CLUSTER); | |
566 | cluster_set_count(&cluster_info[idx], | |
567 | cluster_count(&cluster_info[idx]) + 1); | |
568 | } | |
569 | ||
570 | /* | |
571 | * The cluster corresponding to page_nr decreases one usage. If the usage | |
572 | * counter becomes 0, which means no page in the cluster is in using, we can | |
573 | * optionally discard the cluster and add it to free cluster list. | |
574 | */ | |
575 | static void dec_cluster_info_page(struct swap_info_struct *p, | |
576 | struct swap_cluster_info *cluster_info, unsigned long page_nr) | |
577 | { | |
578 | unsigned long idx = page_nr / SWAPFILE_CLUSTER; | |
579 | ||
580 | if (!cluster_info) | |
581 | return; | |
582 | ||
583 | VM_BUG_ON(cluster_count(&cluster_info[idx]) == 0); | |
584 | cluster_set_count(&cluster_info[idx], | |
585 | cluster_count(&cluster_info[idx]) - 1); | |
586 | ||
38d8b4e6 HY |
587 | if (cluster_count(&cluster_info[idx]) == 0) |
588 | free_cluster(p, idx); | |
2a8f9449 SL |
589 | } |
590 | ||
591 | /* | |
592 | * It's possible scan_swap_map() uses a free cluster in the middle of free | |
593 | * cluster list. Avoiding such abuse to avoid list corruption. | |
594 | */ | |
ebc2a1a6 SL |
595 | static bool |
596 | scan_swap_map_ssd_cluster_conflict(struct swap_info_struct *si, | |
2a8f9449 SL |
597 | unsigned long offset) |
598 | { | |
ebc2a1a6 SL |
599 | struct percpu_cluster *percpu_cluster; |
600 | bool conflict; | |
601 | ||
2a8f9449 | 602 | offset /= SWAPFILE_CLUSTER; |
6b534915 HY |
603 | conflict = !cluster_list_empty(&si->free_clusters) && |
604 | offset != cluster_list_first(&si->free_clusters) && | |
2a8f9449 | 605 | cluster_is_free(&si->cluster_info[offset]); |
ebc2a1a6 SL |
606 | |
607 | if (!conflict) | |
608 | return false; | |
609 | ||
610 | percpu_cluster = this_cpu_ptr(si->percpu_cluster); | |
611 | cluster_set_null(&percpu_cluster->index); | |
612 | return true; | |
613 | } | |
614 | ||
615 | /* | |
616 | * Try to get a swap entry from current cpu's swap entry pool (a cluster). This | |
617 | * might involve allocating a new cluster for current CPU too. | |
618 | */ | |
36005bae | 619 | static bool scan_swap_map_try_ssd_cluster(struct swap_info_struct *si, |
ebc2a1a6 SL |
620 | unsigned long *offset, unsigned long *scan_base) |
621 | { | |
622 | struct percpu_cluster *cluster; | |
235b6217 | 623 | struct swap_cluster_info *ci; |
235b6217 | 624 | unsigned long tmp, max; |
ebc2a1a6 SL |
625 | |
626 | new_cluster: | |
627 | cluster = this_cpu_ptr(si->percpu_cluster); | |
628 | if (cluster_is_null(&cluster->index)) { | |
6b534915 HY |
629 | if (!cluster_list_empty(&si->free_clusters)) { |
630 | cluster->index = si->free_clusters.head; | |
ebc2a1a6 SL |
631 | cluster->next = cluster_next(&cluster->index) * |
632 | SWAPFILE_CLUSTER; | |
6b534915 | 633 | } else if (!cluster_list_empty(&si->discard_clusters)) { |
ebc2a1a6 SL |
634 | /* |
635 | * we don't have free cluster but have some clusters in | |
49070588 HY |
636 | * discarding, do discard now and reclaim them, then |
637 | * reread cluster_next_cpu since we dropped si->lock | |
ebc2a1a6 SL |
638 | */ |
639 | swap_do_scheduled_discard(si); | |
49070588 HY |
640 | *scan_base = this_cpu_read(*si->cluster_next_cpu); |
641 | *offset = *scan_base; | |
ebc2a1a6 SL |
642 | goto new_cluster; |
643 | } else | |
36005bae | 644 | return false; |
ebc2a1a6 SL |
645 | } |
646 | ||
ebc2a1a6 SL |
647 | /* |
648 | * Other CPUs can use our cluster if they can't find a free cluster, | |
649 | * check if there is still free entry in the cluster | |
650 | */ | |
651 | tmp = cluster->next; | |
235b6217 HY |
652 | max = min_t(unsigned long, si->max, |
653 | (cluster_next(&cluster->index) + 1) * SWAPFILE_CLUSTER); | |
7b9e2de1 WY |
654 | if (tmp < max) { |
655 | ci = lock_cluster(si, tmp); | |
656 | while (tmp < max) { | |
657 | if (!si->swap_map[tmp]) | |
658 | break; | |
659 | tmp++; | |
660 | } | |
661 | unlock_cluster(ci); | |
ebc2a1a6 | 662 | } |
0fd0e19e | 663 | if (tmp >= max) { |
ebc2a1a6 SL |
664 | cluster_set_null(&cluster->index); |
665 | goto new_cluster; | |
666 | } | |
667 | cluster->next = tmp + 1; | |
668 | *offset = tmp; | |
669 | *scan_base = tmp; | |
fdff1deb | 670 | return true; |
2a8f9449 SL |
671 | } |
672 | ||
a2468cc9 AL |
673 | static void __del_from_avail_list(struct swap_info_struct *p) |
674 | { | |
675 | int nid; | |
676 | ||
677 | for_each_node(nid) | |
678 | plist_del(&p->avail_lists[nid], &swap_avail_heads[nid]); | |
679 | } | |
680 | ||
681 | static void del_from_avail_list(struct swap_info_struct *p) | |
682 | { | |
683 | spin_lock(&swap_avail_lock); | |
684 | __del_from_avail_list(p); | |
685 | spin_unlock(&swap_avail_lock); | |
686 | } | |
687 | ||
38d8b4e6 HY |
688 | static void swap_range_alloc(struct swap_info_struct *si, unsigned long offset, |
689 | unsigned int nr_entries) | |
690 | { | |
691 | unsigned int end = offset + nr_entries - 1; | |
692 | ||
693 | if (offset == si->lowest_bit) | |
694 | si->lowest_bit += nr_entries; | |
695 | if (end == si->highest_bit) | |
a449bf58 | 696 | WRITE_ONCE(si->highest_bit, si->highest_bit - nr_entries); |
38d8b4e6 HY |
697 | si->inuse_pages += nr_entries; |
698 | if (si->inuse_pages == si->pages) { | |
699 | si->lowest_bit = si->max; | |
700 | si->highest_bit = 0; | |
a2468cc9 | 701 | del_from_avail_list(si); |
38d8b4e6 HY |
702 | } |
703 | } | |
704 | ||
a2468cc9 AL |
705 | static void add_to_avail_list(struct swap_info_struct *p) |
706 | { | |
707 | int nid; | |
708 | ||
709 | spin_lock(&swap_avail_lock); | |
710 | for_each_node(nid) { | |
711 | WARN_ON(!plist_node_empty(&p->avail_lists[nid])); | |
712 | plist_add(&p->avail_lists[nid], &swap_avail_heads[nid]); | |
713 | } | |
714 | spin_unlock(&swap_avail_lock); | |
715 | } | |
716 | ||
38d8b4e6 HY |
717 | static void swap_range_free(struct swap_info_struct *si, unsigned long offset, |
718 | unsigned int nr_entries) | |
719 | { | |
3852f676 | 720 | unsigned long begin = offset; |
38d8b4e6 HY |
721 | unsigned long end = offset + nr_entries - 1; |
722 | void (*swap_slot_free_notify)(struct block_device *, unsigned long); | |
723 | ||
724 | if (offset < si->lowest_bit) | |
725 | si->lowest_bit = offset; | |
726 | if (end > si->highest_bit) { | |
727 | bool was_full = !si->highest_bit; | |
728 | ||
a449bf58 | 729 | WRITE_ONCE(si->highest_bit, end); |
a2468cc9 AL |
730 | if (was_full && (si->flags & SWP_WRITEOK)) |
731 | add_to_avail_list(si); | |
38d8b4e6 HY |
732 | } |
733 | atomic_long_add(nr_entries, &nr_swap_pages); | |
734 | si->inuse_pages -= nr_entries; | |
735 | if (si->flags & SWP_BLKDEV) | |
736 | swap_slot_free_notify = | |
737 | si->bdev->bd_disk->fops->swap_slot_free_notify; | |
738 | else | |
739 | swap_slot_free_notify = NULL; | |
740 | while (offset <= end) { | |
8a84802e | 741 | arch_swap_invalidate_page(si->type, offset); |
38d8b4e6 HY |
742 | frontswap_invalidate_page(si->type, offset); |
743 | if (swap_slot_free_notify) | |
744 | swap_slot_free_notify(si->bdev, offset); | |
745 | offset++; | |
746 | } | |
3852f676 | 747 | clear_shadow_from_swap_cache(si->type, begin, end); |
38d8b4e6 HY |
748 | } |
749 | ||
49070588 HY |
750 | static void set_cluster_next(struct swap_info_struct *si, unsigned long next) |
751 | { | |
752 | unsigned long prev; | |
753 | ||
754 | if (!(si->flags & SWP_SOLIDSTATE)) { | |
755 | si->cluster_next = next; | |
756 | return; | |
757 | } | |
758 | ||
759 | prev = this_cpu_read(*si->cluster_next_cpu); | |
760 | /* | |
761 | * Cross the swap address space size aligned trunk, choose | |
762 | * another trunk randomly to avoid lock contention on swap | |
763 | * address space if possible. | |
764 | */ | |
765 | if ((prev >> SWAP_ADDRESS_SPACE_SHIFT) != | |
766 | (next >> SWAP_ADDRESS_SPACE_SHIFT)) { | |
767 | /* No free swap slots available */ | |
768 | if (si->highest_bit <= si->lowest_bit) | |
769 | return; | |
770 | next = si->lowest_bit + | |
771 | prandom_u32_max(si->highest_bit - si->lowest_bit + 1); | |
772 | next = ALIGN_DOWN(next, SWAP_ADDRESS_SPACE_PAGES); | |
773 | next = max_t(unsigned int, next, si->lowest_bit); | |
774 | } | |
775 | this_cpu_write(*si->cluster_next_cpu, next); | |
776 | } | |
777 | ||
36005bae TC |
778 | static int scan_swap_map_slots(struct swap_info_struct *si, |
779 | unsigned char usage, int nr, | |
780 | swp_entry_t slots[]) | |
1da177e4 | 781 | { |
235b6217 | 782 | struct swap_cluster_info *ci; |
ebebbbe9 | 783 | unsigned long offset; |
c60aa176 | 784 | unsigned long scan_base; |
7992fde7 | 785 | unsigned long last_in_cluster = 0; |
048c27fd | 786 | int latency_ration = LATENCY_LIMIT; |
36005bae | 787 | int n_ret = 0; |
ed43af10 | 788 | bool scanned_many = false; |
36005bae | 789 | |
886bb7e9 | 790 | /* |
7dfad418 HD |
791 | * We try to cluster swap pages by allocating them sequentially |
792 | * in swap. Once we've allocated SWAPFILE_CLUSTER pages this | |
793 | * way, however, we resort to first-free allocation, starting | |
794 | * a new cluster. This prevents us from scattering swap pages | |
795 | * all over the entire swap partition, so that we reduce | |
796 | * overall disk seek times between swap pages. -- sct | |
797 | * But we do now try to find an empty cluster. -Andrea | |
c60aa176 | 798 | * And we let swap pages go all over an SSD partition. Hugh |
7dfad418 HD |
799 | */ |
800 | ||
52b7efdb | 801 | si->flags += SWP_SCANNING; |
49070588 HY |
802 | /* |
803 | * Use percpu scan base for SSD to reduce lock contention on | |
804 | * cluster and swap cache. For HDD, sequential access is more | |
805 | * important. | |
806 | */ | |
807 | if (si->flags & SWP_SOLIDSTATE) | |
808 | scan_base = this_cpu_read(*si->cluster_next_cpu); | |
809 | else | |
810 | scan_base = si->cluster_next; | |
811 | offset = scan_base; | |
ebebbbe9 | 812 | |
ebc2a1a6 SL |
813 | /* SSD algorithm */ |
814 | if (si->cluster_info) { | |
bd2d18da | 815 | if (!scan_swap_map_try_ssd_cluster(si, &offset, &scan_base)) |
36005bae | 816 | goto scan; |
f4eaf51a | 817 | } else if (unlikely(!si->cluster_nr--)) { |
ebebbbe9 HD |
818 | if (si->pages - si->inuse_pages < SWAPFILE_CLUSTER) { |
819 | si->cluster_nr = SWAPFILE_CLUSTER - 1; | |
820 | goto checks; | |
821 | } | |
2a8f9449 | 822 | |
ec8acf20 | 823 | spin_unlock(&si->lock); |
7dfad418 | 824 | |
c60aa176 HD |
825 | /* |
826 | * If seek is expensive, start searching for new cluster from | |
827 | * start of partition, to minimize the span of allocated swap. | |
50088c44 CY |
828 | * If seek is cheap, that is the SWP_SOLIDSTATE si->cluster_info |
829 | * case, just handled by scan_swap_map_try_ssd_cluster() above. | |
c60aa176 | 830 | */ |
50088c44 | 831 | scan_base = offset = si->lowest_bit; |
7dfad418 HD |
832 | last_in_cluster = offset + SWAPFILE_CLUSTER - 1; |
833 | ||
834 | /* Locate the first empty (unaligned) cluster */ | |
835 | for (; last_in_cluster <= si->highest_bit; offset++) { | |
1da177e4 | 836 | if (si->swap_map[offset]) |
7dfad418 HD |
837 | last_in_cluster = offset + SWAPFILE_CLUSTER; |
838 | else if (offset == last_in_cluster) { | |
ec8acf20 | 839 | spin_lock(&si->lock); |
ebebbbe9 HD |
840 | offset -= SWAPFILE_CLUSTER - 1; |
841 | si->cluster_next = offset; | |
842 | si->cluster_nr = SWAPFILE_CLUSTER - 1; | |
c60aa176 HD |
843 | goto checks; |
844 | } | |
845 | if (unlikely(--latency_ration < 0)) { | |
846 | cond_resched(); | |
847 | latency_ration = LATENCY_LIMIT; | |
848 | } | |
849 | } | |
850 | ||
851 | offset = scan_base; | |
ec8acf20 | 852 | spin_lock(&si->lock); |
ebebbbe9 | 853 | si->cluster_nr = SWAPFILE_CLUSTER - 1; |
1da177e4 | 854 | } |
7dfad418 | 855 | |
ebebbbe9 | 856 | checks: |
ebc2a1a6 | 857 | if (si->cluster_info) { |
36005bae TC |
858 | while (scan_swap_map_ssd_cluster_conflict(si, offset)) { |
859 | /* take a break if we already got some slots */ | |
860 | if (n_ret) | |
861 | goto done; | |
862 | if (!scan_swap_map_try_ssd_cluster(si, &offset, | |
863 | &scan_base)) | |
864 | goto scan; | |
865 | } | |
ebc2a1a6 | 866 | } |
ebebbbe9 | 867 | if (!(si->flags & SWP_WRITEOK)) |
52b7efdb | 868 | goto no_page; |
7dfad418 HD |
869 | if (!si->highest_bit) |
870 | goto no_page; | |
ebebbbe9 | 871 | if (offset > si->highest_bit) |
c60aa176 | 872 | scan_base = offset = si->lowest_bit; |
c9e44410 | 873 | |
235b6217 | 874 | ci = lock_cluster(si, offset); |
b73d7fce HD |
875 | /* reuse swap entry of cache-only swap if not busy. */ |
876 | if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) { | |
c9e44410 | 877 | int swap_was_freed; |
235b6217 | 878 | unlock_cluster(ci); |
ec8acf20 | 879 | spin_unlock(&si->lock); |
bcd49e86 | 880 | swap_was_freed = __try_to_reclaim_swap(si, offset, TTRS_ANYWAY); |
ec8acf20 | 881 | spin_lock(&si->lock); |
c9e44410 KH |
882 | /* entry was freed successfully, try to use this again */ |
883 | if (swap_was_freed) | |
884 | goto checks; | |
885 | goto scan; /* check next one */ | |
886 | } | |
887 | ||
235b6217 HY |
888 | if (si->swap_map[offset]) { |
889 | unlock_cluster(ci); | |
36005bae TC |
890 | if (!n_ret) |
891 | goto scan; | |
892 | else | |
893 | goto done; | |
235b6217 | 894 | } |
a449bf58 | 895 | WRITE_ONCE(si->swap_map[offset], usage); |
2872bb2d HY |
896 | inc_cluster_info_page(si, si->cluster_info, offset); |
897 | unlock_cluster(ci); | |
ebebbbe9 | 898 | |
38d8b4e6 | 899 | swap_range_alloc(si, offset, 1); |
36005bae TC |
900 | slots[n_ret++] = swp_entry(si->type, offset); |
901 | ||
902 | /* got enough slots or reach max slots? */ | |
903 | if ((n_ret == nr) || (offset >= si->highest_bit)) | |
904 | goto done; | |
905 | ||
906 | /* search for next available slot */ | |
907 | ||
908 | /* time to take a break? */ | |
909 | if (unlikely(--latency_ration < 0)) { | |
910 | if (n_ret) | |
911 | goto done; | |
912 | spin_unlock(&si->lock); | |
913 | cond_resched(); | |
914 | spin_lock(&si->lock); | |
915 | latency_ration = LATENCY_LIMIT; | |
916 | } | |
917 | ||
918 | /* try to get more slots in cluster */ | |
919 | if (si->cluster_info) { | |
920 | if (scan_swap_map_try_ssd_cluster(si, &offset, &scan_base)) | |
921 | goto checks; | |
f4eaf51a WY |
922 | } else if (si->cluster_nr && !si->swap_map[++offset]) { |
923 | /* non-ssd case, still more slots in cluster? */ | |
36005bae TC |
924 | --si->cluster_nr; |
925 | goto checks; | |
926 | } | |
7992fde7 | 927 | |
ed43af10 HY |
928 | /* |
929 | * Even if there's no free clusters available (fragmented), | |
930 | * try to scan a little more quickly with lock held unless we | |
931 | * have scanned too many slots already. | |
932 | */ | |
933 | if (!scanned_many) { | |
934 | unsigned long scan_limit; | |
935 | ||
936 | if (offset < scan_base) | |
937 | scan_limit = scan_base; | |
938 | else | |
939 | scan_limit = si->highest_bit; | |
940 | for (; offset <= scan_limit && --latency_ration > 0; | |
941 | offset++) { | |
942 | if (!si->swap_map[offset]) | |
943 | goto checks; | |
944 | } | |
945 | } | |
946 | ||
36005bae | 947 | done: |
49070588 | 948 | set_cluster_next(si, offset + 1); |
36005bae TC |
949 | si->flags -= SWP_SCANNING; |
950 | return n_ret; | |
7dfad418 | 951 | |
ebebbbe9 | 952 | scan: |
ec8acf20 | 953 | spin_unlock(&si->lock); |
a449bf58 QC |
954 | while (++offset <= READ_ONCE(si->highest_bit)) { |
955 | if (data_race(!si->swap_map[offset])) { | |
ec8acf20 | 956 | spin_lock(&si->lock); |
52b7efdb HD |
957 | goto checks; |
958 | } | |
a449bf58 QC |
959 | if (vm_swap_full() && |
960 | READ_ONCE(si->swap_map[offset]) == SWAP_HAS_CACHE) { | |
ec8acf20 | 961 | spin_lock(&si->lock); |
c9e44410 KH |
962 | goto checks; |
963 | } | |
048c27fd HD |
964 | if (unlikely(--latency_ration < 0)) { |
965 | cond_resched(); | |
966 | latency_ration = LATENCY_LIMIT; | |
ed43af10 | 967 | scanned_many = true; |
048c27fd | 968 | } |
7dfad418 | 969 | } |
c60aa176 | 970 | offset = si->lowest_bit; |
a5998061 | 971 | while (offset < scan_base) { |
a449bf58 | 972 | if (data_race(!si->swap_map[offset])) { |
ec8acf20 | 973 | spin_lock(&si->lock); |
c60aa176 HD |
974 | goto checks; |
975 | } | |
a449bf58 QC |
976 | if (vm_swap_full() && |
977 | READ_ONCE(si->swap_map[offset]) == SWAP_HAS_CACHE) { | |
ec8acf20 | 978 | spin_lock(&si->lock); |
c9e44410 KH |
979 | goto checks; |
980 | } | |
c60aa176 HD |
981 | if (unlikely(--latency_ration < 0)) { |
982 | cond_resched(); | |
983 | latency_ration = LATENCY_LIMIT; | |
ed43af10 | 984 | scanned_many = true; |
c60aa176 | 985 | } |
a5998061 | 986 | offset++; |
c60aa176 | 987 | } |
ec8acf20 | 988 | spin_lock(&si->lock); |
7dfad418 HD |
989 | |
990 | no_page: | |
52b7efdb | 991 | si->flags -= SWP_SCANNING; |
36005bae | 992 | return n_ret; |
1da177e4 LT |
993 | } |
994 | ||
38d8b4e6 HY |
995 | static int swap_alloc_cluster(struct swap_info_struct *si, swp_entry_t *slot) |
996 | { | |
997 | unsigned long idx; | |
998 | struct swap_cluster_info *ci; | |
661c7566 | 999 | unsigned long offset; |
38d8b4e6 | 1000 | |
fe5266d5 HY |
1001 | /* |
1002 | * Should not even be attempting cluster allocations when huge | |
1003 | * page swap is disabled. Warn and fail the allocation. | |
1004 | */ | |
1005 | if (!IS_ENABLED(CONFIG_THP_SWAP)) { | |
1006 | VM_WARN_ON_ONCE(1); | |
1007 | return 0; | |
1008 | } | |
1009 | ||
38d8b4e6 HY |
1010 | if (cluster_list_empty(&si->free_clusters)) |
1011 | return 0; | |
1012 | ||
1013 | idx = cluster_list_first(&si->free_clusters); | |
1014 | offset = idx * SWAPFILE_CLUSTER; | |
1015 | ci = lock_cluster(si, offset); | |
1016 | alloc_cluster(si, idx); | |
e0709829 | 1017 | cluster_set_count_flag(ci, SWAPFILE_CLUSTER, CLUSTER_FLAG_HUGE); |
38d8b4e6 | 1018 | |
661c7566 | 1019 | memset(si->swap_map + offset, SWAP_HAS_CACHE, SWAPFILE_CLUSTER); |
38d8b4e6 HY |
1020 | unlock_cluster(ci); |
1021 | swap_range_alloc(si, offset, SWAPFILE_CLUSTER); | |
1022 | *slot = swp_entry(si->type, offset); | |
1023 | ||
1024 | return 1; | |
1025 | } | |
1026 | ||
1027 | static void swap_free_cluster(struct swap_info_struct *si, unsigned long idx) | |
1028 | { | |
1029 | unsigned long offset = idx * SWAPFILE_CLUSTER; | |
1030 | struct swap_cluster_info *ci; | |
1031 | ||
1032 | ci = lock_cluster(si, offset); | |
979aafa5 | 1033 | memset(si->swap_map + offset, 0, SWAPFILE_CLUSTER); |
38d8b4e6 HY |
1034 | cluster_set_count_flag(ci, 0, 0); |
1035 | free_cluster(si, idx); | |
1036 | unlock_cluster(ci); | |
1037 | swap_range_free(si, offset, SWAPFILE_CLUSTER); | |
1038 | } | |
38d8b4e6 | 1039 | |
36005bae TC |
1040 | static unsigned long scan_swap_map(struct swap_info_struct *si, |
1041 | unsigned char usage) | |
1042 | { | |
1043 | swp_entry_t entry; | |
1044 | int n_ret; | |
1045 | ||
1046 | n_ret = scan_swap_map_slots(si, usage, 1, &entry); | |
1047 | ||
1048 | if (n_ret) | |
1049 | return swp_offset(entry); | |
1050 | else | |
1051 | return 0; | |
1052 | ||
1053 | } | |
1054 | ||
5d5e8f19 | 1055 | int get_swap_pages(int n_goal, swp_entry_t swp_entries[], int entry_size) |
1da177e4 | 1056 | { |
5d5e8f19 | 1057 | unsigned long size = swap_entry_size(entry_size); |
adfab836 | 1058 | struct swap_info_struct *si, *next; |
36005bae TC |
1059 | long avail_pgs; |
1060 | int n_ret = 0; | |
a2468cc9 | 1061 | int node; |
1da177e4 | 1062 | |
38d8b4e6 | 1063 | /* Only single cluster request supported */ |
5d5e8f19 | 1064 | WARN_ON_ONCE(n_goal > 1 && size == SWAPFILE_CLUSTER); |
38d8b4e6 | 1065 | |
b50da6e9 ZH |
1066 | spin_lock(&swap_avail_lock); |
1067 | ||
5d5e8f19 | 1068 | avail_pgs = atomic_long_read(&nr_swap_pages) / size; |
b50da6e9 ZH |
1069 | if (avail_pgs <= 0) { |
1070 | spin_unlock(&swap_avail_lock); | |
fb4f88dc | 1071 | goto noswap; |
b50da6e9 | 1072 | } |
36005bae | 1073 | |
08d3090f | 1074 | n_goal = min3((long)n_goal, (long)SWAP_BATCH, avail_pgs); |
36005bae | 1075 | |
5d5e8f19 | 1076 | atomic_long_sub(n_goal * size, &nr_swap_pages); |
fb4f88dc | 1077 | |
18ab4d4c | 1078 | start_over: |
a2468cc9 AL |
1079 | node = numa_node_id(); |
1080 | plist_for_each_entry_safe(si, next, &swap_avail_heads[node], avail_lists[node]) { | |
18ab4d4c | 1081 | /* requeue si to after same-priority siblings */ |
a2468cc9 | 1082 | plist_requeue(&si->avail_lists[node], &swap_avail_heads[node]); |
18ab4d4c | 1083 | spin_unlock(&swap_avail_lock); |
ec8acf20 | 1084 | spin_lock(&si->lock); |
adfab836 | 1085 | if (!si->highest_bit || !(si->flags & SWP_WRITEOK)) { |
18ab4d4c | 1086 | spin_lock(&swap_avail_lock); |
a2468cc9 | 1087 | if (plist_node_empty(&si->avail_lists[node])) { |
18ab4d4c DS |
1088 | spin_unlock(&si->lock); |
1089 | goto nextsi; | |
1090 | } | |
1091 | WARN(!si->highest_bit, | |
1092 | "swap_info %d in list but !highest_bit\n", | |
1093 | si->type); | |
1094 | WARN(!(si->flags & SWP_WRITEOK), | |
1095 | "swap_info %d in list but !SWP_WRITEOK\n", | |
1096 | si->type); | |
a2468cc9 | 1097 | __del_from_avail_list(si); |
ec8acf20 | 1098 | spin_unlock(&si->lock); |
18ab4d4c | 1099 | goto nextsi; |
ec8acf20 | 1100 | } |
5d5e8f19 | 1101 | if (size == SWAPFILE_CLUSTER) { |
41663430 | 1102 | if (si->flags & SWP_BLKDEV) |
f0eea189 HY |
1103 | n_ret = swap_alloc_cluster(si, swp_entries); |
1104 | } else | |
38d8b4e6 HY |
1105 | n_ret = scan_swap_map_slots(si, SWAP_HAS_CACHE, |
1106 | n_goal, swp_entries); | |
ec8acf20 | 1107 | spin_unlock(&si->lock); |
5d5e8f19 | 1108 | if (n_ret || size == SWAPFILE_CLUSTER) |
36005bae | 1109 | goto check_out; |
18ab4d4c | 1110 | pr_debug("scan_swap_map of si %d failed to find offset\n", |
36005bae TC |
1111 | si->type); |
1112 | ||
18ab4d4c DS |
1113 | spin_lock(&swap_avail_lock); |
1114 | nextsi: | |
adfab836 DS |
1115 | /* |
1116 | * if we got here, it's likely that si was almost full before, | |
1117 | * and since scan_swap_map() can drop the si->lock, multiple | |
1118 | * callers probably all tried to get a page from the same si | |
18ab4d4c DS |
1119 | * and it filled up before we could get one; or, the si filled |
1120 | * up between us dropping swap_avail_lock and taking si->lock. | |
1121 | * Since we dropped the swap_avail_lock, the swap_avail_head | |
1122 | * list may have been modified; so if next is still in the | |
36005bae TC |
1123 | * swap_avail_head list then try it, otherwise start over |
1124 | * if we have not gotten any slots. | |
adfab836 | 1125 | */ |
a2468cc9 | 1126 | if (plist_node_empty(&next->avail_lists[node])) |
18ab4d4c | 1127 | goto start_over; |
1da177e4 | 1128 | } |
fb4f88dc | 1129 | |
18ab4d4c DS |
1130 | spin_unlock(&swap_avail_lock); |
1131 | ||
36005bae TC |
1132 | check_out: |
1133 | if (n_ret < n_goal) | |
5d5e8f19 | 1134 | atomic_long_add((long)(n_goal - n_ret) * size, |
38d8b4e6 | 1135 | &nr_swap_pages); |
fb4f88dc | 1136 | noswap: |
36005bae TC |
1137 | return n_ret; |
1138 | } | |
1139 | ||
2de1a7e4 | 1140 | /* The only caller of this function is now suspend routine */ |
910321ea HD |
1141 | swp_entry_t get_swap_page_of_type(int type) |
1142 | { | |
c10d38cc | 1143 | struct swap_info_struct *si = swap_type_to_swap_info(type); |
910321ea HD |
1144 | pgoff_t offset; |
1145 | ||
c10d38cc DJ |
1146 | if (!si) |
1147 | goto fail; | |
1148 | ||
ec8acf20 | 1149 | spin_lock(&si->lock); |
c10d38cc | 1150 | if (si->flags & SWP_WRITEOK) { |
910321ea HD |
1151 | /* This is called for allocating swap entry, not cache */ |
1152 | offset = scan_swap_map(si, 1); | |
1153 | if (offset) { | |
b50da6e9 | 1154 | atomic_long_dec(&nr_swap_pages); |
ec8acf20 | 1155 | spin_unlock(&si->lock); |
910321ea HD |
1156 | return swp_entry(type, offset); |
1157 | } | |
910321ea | 1158 | } |
ec8acf20 | 1159 | spin_unlock(&si->lock); |
c10d38cc | 1160 | fail: |
910321ea HD |
1161 | return (swp_entry_t) {0}; |
1162 | } | |
1163 | ||
e8c26ab6 | 1164 | static struct swap_info_struct *__swap_info_get(swp_entry_t entry) |
1da177e4 | 1165 | { |
73c34b6a | 1166 | struct swap_info_struct *p; |
eb085574 | 1167 | unsigned long offset; |
1da177e4 LT |
1168 | |
1169 | if (!entry.val) | |
1170 | goto out; | |
eb085574 | 1171 | p = swp_swap_info(entry); |
c10d38cc | 1172 | if (!p) |
1da177e4 | 1173 | goto bad_nofile; |
a449bf58 | 1174 | if (data_race(!(p->flags & SWP_USED))) |
1da177e4 LT |
1175 | goto bad_device; |
1176 | offset = swp_offset(entry); | |
1177 | if (offset >= p->max) | |
1178 | goto bad_offset; | |
1da177e4 LT |
1179 | return p; |
1180 | ||
1da177e4 | 1181 | bad_offset: |
cf532faa | 1182 | pr_err("%s: %s%08lx\n", __func__, Bad_offset, entry.val); |
1da177e4 LT |
1183 | goto out; |
1184 | bad_device: | |
cf532faa | 1185 | pr_err("%s: %s%08lx\n", __func__, Unused_file, entry.val); |
1da177e4 LT |
1186 | goto out; |
1187 | bad_nofile: | |
cf532faa | 1188 | pr_err("%s: %s%08lx\n", __func__, Bad_file, entry.val); |
1da177e4 LT |
1189 | out: |
1190 | return NULL; | |
886bb7e9 | 1191 | } |
1da177e4 | 1192 | |
e8c26ab6 TC |
1193 | static struct swap_info_struct *_swap_info_get(swp_entry_t entry) |
1194 | { | |
1195 | struct swap_info_struct *p; | |
1196 | ||
1197 | p = __swap_info_get(entry); | |
1198 | if (!p) | |
1199 | goto out; | |
a449bf58 | 1200 | if (data_race(!p->swap_map[swp_offset(entry)])) |
e8c26ab6 TC |
1201 | goto bad_free; |
1202 | return p; | |
1203 | ||
1204 | bad_free: | |
cf532faa | 1205 | pr_err("%s: %s%08lx\n", __func__, Unused_offset, entry.val); |
e8c26ab6 TC |
1206 | out: |
1207 | return NULL; | |
1208 | } | |
1209 | ||
235b6217 HY |
1210 | static struct swap_info_struct *swap_info_get(swp_entry_t entry) |
1211 | { | |
1212 | struct swap_info_struct *p; | |
1213 | ||
1214 | p = _swap_info_get(entry); | |
1215 | if (p) | |
1216 | spin_lock(&p->lock); | |
1217 | return p; | |
1218 | } | |
1219 | ||
7c00bafe TC |
1220 | static struct swap_info_struct *swap_info_get_cont(swp_entry_t entry, |
1221 | struct swap_info_struct *q) | |
1222 | { | |
1223 | struct swap_info_struct *p; | |
1224 | ||
1225 | p = _swap_info_get(entry); | |
1226 | ||
1227 | if (p != q) { | |
1228 | if (q != NULL) | |
1229 | spin_unlock(&q->lock); | |
1230 | if (p != NULL) | |
1231 | spin_lock(&p->lock); | |
1232 | } | |
1233 | return p; | |
1234 | } | |
1235 | ||
b32d5f32 HY |
1236 | static unsigned char __swap_entry_free_locked(struct swap_info_struct *p, |
1237 | unsigned long offset, | |
1238 | unsigned char usage) | |
1da177e4 | 1239 | { |
8d69aaee HD |
1240 | unsigned char count; |
1241 | unsigned char has_cache; | |
235b6217 | 1242 | |
253d553b | 1243 | count = p->swap_map[offset]; |
235b6217 | 1244 | |
253d553b HD |
1245 | has_cache = count & SWAP_HAS_CACHE; |
1246 | count &= ~SWAP_HAS_CACHE; | |
355cfa73 | 1247 | |
253d553b | 1248 | if (usage == SWAP_HAS_CACHE) { |
355cfa73 | 1249 | VM_BUG_ON(!has_cache); |
253d553b | 1250 | has_cache = 0; |
aaa46865 HD |
1251 | } else if (count == SWAP_MAP_SHMEM) { |
1252 | /* | |
1253 | * Or we could insist on shmem.c using a special | |
1254 | * swap_shmem_free() and free_shmem_swap_and_cache()... | |
1255 | */ | |
1256 | count = 0; | |
570a335b HD |
1257 | } else if ((count & ~COUNT_CONTINUED) <= SWAP_MAP_MAX) { |
1258 | if (count == COUNT_CONTINUED) { | |
1259 | if (swap_count_continued(p, offset, count)) | |
1260 | count = SWAP_MAP_MAX | COUNT_CONTINUED; | |
1261 | else | |
1262 | count = SWAP_MAP_MAX; | |
1263 | } else | |
1264 | count--; | |
1265 | } | |
253d553b | 1266 | |
253d553b | 1267 | usage = count | has_cache; |
a449bf58 QC |
1268 | if (usage) |
1269 | WRITE_ONCE(p->swap_map[offset], usage); | |
1270 | else | |
1271 | WRITE_ONCE(p->swap_map[offset], SWAP_HAS_CACHE); | |
7c00bafe | 1272 | |
b32d5f32 HY |
1273 | return usage; |
1274 | } | |
1275 | ||
eb085574 HY |
1276 | /* |
1277 | * Check whether swap entry is valid in the swap device. If so, | |
1278 | * return pointer to swap_info_struct, and keep the swap entry valid | |
1279 | * via preventing the swap device from being swapoff, until | |
1280 | * put_swap_device() is called. Otherwise return NULL. | |
1281 | * | |
eb085574 | 1282 | * Notice that swapoff or swapoff+swapon can still happen before the |
63d8620e ML |
1283 | * percpu_ref_tryget_live() in get_swap_device() or after the |
1284 | * percpu_ref_put() in put_swap_device() if there isn't any other way | |
1285 | * to prevent swapoff, such as page lock, page table lock, etc. The | |
1286 | * caller must be prepared for that. For example, the following | |
1287 | * situation is possible. | |
eb085574 HY |
1288 | * |
1289 | * CPU1 CPU2 | |
1290 | * do_swap_page() | |
1291 | * ... swapoff+swapon | |
1292 | * __read_swap_cache_async() | |
1293 | * swapcache_prepare() | |
1294 | * __swap_duplicate() | |
1295 | * // check swap_map | |
1296 | * // verify PTE not changed | |
1297 | * | |
1298 | * In __swap_duplicate(), the swap_map need to be checked before | |
1299 | * changing partly because the specified swap entry may be for another | |
1300 | * swap device which has been swapoff. And in do_swap_page(), after | |
1301 | * the page is read from the swap device, the PTE is verified not | |
1302 | * changed with the page table locked to check whether the swap device | |
1303 | * has been swapoff or swapoff+swapon. | |
1304 | */ | |
1305 | struct swap_info_struct *get_swap_device(swp_entry_t entry) | |
1306 | { | |
1307 | struct swap_info_struct *si; | |
1308 | unsigned long offset; | |
1309 | ||
1310 | if (!entry.val) | |
1311 | goto out; | |
1312 | si = swp_swap_info(entry); | |
1313 | if (!si) | |
1314 | goto bad_nofile; | |
63d8620e ML |
1315 | if (!percpu_ref_tryget_live(&si->users)) |
1316 | goto out; | |
1317 | /* | |
1318 | * Guarantee the si->users are checked before accessing other | |
1319 | * fields of swap_info_struct. | |
1320 | * | |
1321 | * Paired with the spin_unlock() after setup_swap_info() in | |
1322 | * enable_swap_info(). | |
1323 | */ | |
1324 | smp_rmb(); | |
eb085574 HY |
1325 | offset = swp_offset(entry); |
1326 | if (offset >= si->max) | |
63d8620e | 1327 | goto put_out; |
eb085574 HY |
1328 | |
1329 | return si; | |
1330 | bad_nofile: | |
1331 | pr_err("%s: %s%08lx\n", __func__, Bad_file, entry.val); | |
1332 | out: | |
1333 | return NULL; | |
63d8620e ML |
1334 | put_out: |
1335 | percpu_ref_put(&si->users); | |
eb085574 HY |
1336 | return NULL; |
1337 | } | |
1338 | ||
b32d5f32 | 1339 | static unsigned char __swap_entry_free(struct swap_info_struct *p, |
33e16272 | 1340 | swp_entry_t entry) |
b32d5f32 HY |
1341 | { |
1342 | struct swap_cluster_info *ci; | |
1343 | unsigned long offset = swp_offset(entry); | |
33e16272 | 1344 | unsigned char usage; |
b32d5f32 HY |
1345 | |
1346 | ci = lock_cluster_or_swap_info(p, offset); | |
33e16272 | 1347 | usage = __swap_entry_free_locked(p, offset, 1); |
7c00bafe | 1348 | unlock_cluster_or_swap_info(p, ci); |
10e364da HY |
1349 | if (!usage) |
1350 | free_swap_slot(entry); | |
7c00bafe TC |
1351 | |
1352 | return usage; | |
1353 | } | |
355cfa73 | 1354 | |
7c00bafe TC |
1355 | static void swap_entry_free(struct swap_info_struct *p, swp_entry_t entry) |
1356 | { | |
1357 | struct swap_cluster_info *ci; | |
1358 | unsigned long offset = swp_offset(entry); | |
1359 | unsigned char count; | |
1360 | ||
1361 | ci = lock_cluster(p, offset); | |
1362 | count = p->swap_map[offset]; | |
1363 | VM_BUG_ON(count != SWAP_HAS_CACHE); | |
1364 | p->swap_map[offset] = 0; | |
1365 | dec_cluster_info_page(p, p->cluster_info, offset); | |
235b6217 HY |
1366 | unlock_cluster(ci); |
1367 | ||
38d8b4e6 HY |
1368 | mem_cgroup_uncharge_swap(entry, 1); |
1369 | swap_range_free(p, offset, 1); | |
1da177e4 LT |
1370 | } |
1371 | ||
1372 | /* | |
2de1a7e4 | 1373 | * Caller has made sure that the swap device corresponding to entry |
1da177e4 LT |
1374 | * is still around or has not been recycled. |
1375 | */ | |
1376 | void swap_free(swp_entry_t entry) | |
1377 | { | |
73c34b6a | 1378 | struct swap_info_struct *p; |
1da177e4 | 1379 | |
235b6217 | 1380 | p = _swap_info_get(entry); |
10e364da | 1381 | if (p) |
33e16272 | 1382 | __swap_entry_free(p, entry); |
1da177e4 LT |
1383 | } |
1384 | ||
cb4b86ba KH |
1385 | /* |
1386 | * Called after dropping swapcache to decrease refcnt to swap entries. | |
1387 | */ | |
a448f2d0 | 1388 | void put_swap_page(struct page *page, swp_entry_t entry) |
38d8b4e6 HY |
1389 | { |
1390 | unsigned long offset = swp_offset(entry); | |
1391 | unsigned long idx = offset / SWAPFILE_CLUSTER; | |
1392 | struct swap_cluster_info *ci; | |
1393 | struct swap_info_struct *si; | |
1394 | unsigned char *map; | |
a3aea839 HY |
1395 | unsigned int i, free_entries = 0; |
1396 | unsigned char val; | |
6c357848 | 1397 | int size = swap_entry_size(thp_nr_pages(page)); |
fe5266d5 | 1398 | |
a3aea839 | 1399 | si = _swap_info_get(entry); |
38d8b4e6 HY |
1400 | if (!si) |
1401 | return; | |
1402 | ||
c2343d27 | 1403 | ci = lock_cluster_or_swap_info(si, offset); |
a448f2d0 | 1404 | if (size == SWAPFILE_CLUSTER) { |
a448f2d0 HY |
1405 | VM_BUG_ON(!cluster_is_huge(ci)); |
1406 | map = si->swap_map + offset; | |
1407 | for (i = 0; i < SWAPFILE_CLUSTER; i++) { | |
1408 | val = map[i]; | |
1409 | VM_BUG_ON(!(val & SWAP_HAS_CACHE)); | |
1410 | if (val == SWAP_HAS_CACHE) | |
1411 | free_entries++; | |
1412 | } | |
a448f2d0 | 1413 | cluster_clear_huge(ci); |
a448f2d0 | 1414 | if (free_entries == SWAPFILE_CLUSTER) { |
c2343d27 | 1415 | unlock_cluster_or_swap_info(si, ci); |
a448f2d0 | 1416 | spin_lock(&si->lock); |
a448f2d0 HY |
1417 | mem_cgroup_uncharge_swap(entry, SWAPFILE_CLUSTER); |
1418 | swap_free_cluster(si, idx); | |
1419 | spin_unlock(&si->lock); | |
1420 | return; | |
1421 | } | |
1422 | } | |
c2343d27 HY |
1423 | for (i = 0; i < size; i++, entry.val++) { |
1424 | if (!__swap_entry_free_locked(si, offset + i, SWAP_HAS_CACHE)) { | |
1425 | unlock_cluster_or_swap_info(si, ci); | |
1426 | free_swap_slot(entry); | |
1427 | if (i == size - 1) | |
1428 | return; | |
1429 | lock_cluster_or_swap_info(si, offset); | |
a3aea839 HY |
1430 | } |
1431 | } | |
c2343d27 | 1432 | unlock_cluster_or_swap_info(si, ci); |
38d8b4e6 | 1433 | } |
59807685 | 1434 | |
fe5266d5 | 1435 | #ifdef CONFIG_THP_SWAP |
59807685 HY |
1436 | int split_swap_cluster(swp_entry_t entry) |
1437 | { | |
1438 | struct swap_info_struct *si; | |
1439 | struct swap_cluster_info *ci; | |
1440 | unsigned long offset = swp_offset(entry); | |
1441 | ||
1442 | si = _swap_info_get(entry); | |
1443 | if (!si) | |
1444 | return -EBUSY; | |
1445 | ci = lock_cluster(si, offset); | |
1446 | cluster_clear_huge(ci); | |
1447 | unlock_cluster(ci); | |
1448 | return 0; | |
1449 | } | |
fe5266d5 | 1450 | #endif |
38d8b4e6 | 1451 | |
155b5f88 HY |
1452 | static int swp_entry_cmp(const void *ent1, const void *ent2) |
1453 | { | |
1454 | const swp_entry_t *e1 = ent1, *e2 = ent2; | |
1455 | ||
1456 | return (int)swp_type(*e1) - (int)swp_type(*e2); | |
1457 | } | |
1458 | ||
7c00bafe TC |
1459 | void swapcache_free_entries(swp_entry_t *entries, int n) |
1460 | { | |
1461 | struct swap_info_struct *p, *prev; | |
1462 | int i; | |
1463 | ||
1464 | if (n <= 0) | |
1465 | return; | |
1466 | ||
1467 | prev = NULL; | |
1468 | p = NULL; | |
155b5f88 HY |
1469 | |
1470 | /* | |
1471 | * Sort swap entries by swap device, so each lock is only taken once. | |
1472 | * nr_swapfiles isn't absolutely correct, but the overhead of sort() is | |
1473 | * so low that it isn't necessary to optimize further. | |
1474 | */ | |
1475 | if (nr_swapfiles > 1) | |
1476 | sort(entries, n, sizeof(entries[0]), swp_entry_cmp, NULL); | |
7c00bafe TC |
1477 | for (i = 0; i < n; ++i) { |
1478 | p = swap_info_get_cont(entries[i], prev); | |
1479 | if (p) | |
1480 | swap_entry_free(p, entries[i]); | |
7c00bafe TC |
1481 | prev = p; |
1482 | } | |
235b6217 | 1483 | if (p) |
7c00bafe | 1484 | spin_unlock(&p->lock); |
cb4b86ba KH |
1485 | } |
1486 | ||
1da177e4 | 1487 | /* |
c475a8ab | 1488 | * How many references to page are currently swapped out? |
570a335b HD |
1489 | * This does not give an exact answer when swap count is continued, |
1490 | * but does include the high COUNT_CONTINUED flag to allow for that. | |
1da177e4 | 1491 | */ |
bde05d1c | 1492 | int page_swapcount(struct page *page) |
1da177e4 | 1493 | { |
c475a8ab HD |
1494 | int count = 0; |
1495 | struct swap_info_struct *p; | |
235b6217 | 1496 | struct swap_cluster_info *ci; |
1da177e4 | 1497 | swp_entry_t entry; |
235b6217 | 1498 | unsigned long offset; |
1da177e4 | 1499 | |
4c21e2f2 | 1500 | entry.val = page_private(page); |
235b6217 | 1501 | p = _swap_info_get(entry); |
1da177e4 | 1502 | if (p) { |
235b6217 HY |
1503 | offset = swp_offset(entry); |
1504 | ci = lock_cluster_or_swap_info(p, offset); | |
1505 | count = swap_count(p->swap_map[offset]); | |
1506 | unlock_cluster_or_swap_info(p, ci); | |
1da177e4 | 1507 | } |
c475a8ab | 1508 | return count; |
1da177e4 LT |
1509 | } |
1510 | ||
eb085574 | 1511 | int __swap_count(swp_entry_t entry) |
aa8d22a1 | 1512 | { |
eb085574 | 1513 | struct swap_info_struct *si; |
aa8d22a1 | 1514 | pgoff_t offset = swp_offset(entry); |
eb085574 | 1515 | int count = 0; |
aa8d22a1 | 1516 | |
eb085574 HY |
1517 | si = get_swap_device(entry); |
1518 | if (si) { | |
1519 | count = swap_count(si->swap_map[offset]); | |
1520 | put_swap_device(si); | |
1521 | } | |
1522 | return count; | |
aa8d22a1 MK |
1523 | } |
1524 | ||
322b8afe HY |
1525 | static int swap_swapcount(struct swap_info_struct *si, swp_entry_t entry) |
1526 | { | |
1527 | int count = 0; | |
1528 | pgoff_t offset = swp_offset(entry); | |
1529 | struct swap_cluster_info *ci; | |
1530 | ||
1531 | ci = lock_cluster_or_swap_info(si, offset); | |
1532 | count = swap_count(si->swap_map[offset]); | |
1533 | unlock_cluster_or_swap_info(si, ci); | |
1534 | return count; | |
1535 | } | |
1536 | ||
e8c26ab6 TC |
1537 | /* |
1538 | * How many references to @entry are currently swapped out? | |
1539 | * This does not give an exact answer when swap count is continued, | |
1540 | * but does include the high COUNT_CONTINUED flag to allow for that. | |
1541 | */ | |
1542 | int __swp_swapcount(swp_entry_t entry) | |
1543 | { | |
1544 | int count = 0; | |
e8c26ab6 | 1545 | struct swap_info_struct *si; |
e8c26ab6 | 1546 | |
eb085574 HY |
1547 | si = get_swap_device(entry); |
1548 | if (si) { | |
322b8afe | 1549 | count = swap_swapcount(si, entry); |
eb085574 HY |
1550 | put_swap_device(si); |
1551 | } | |
e8c26ab6 TC |
1552 | return count; |
1553 | } | |
1554 | ||
8334b962 MK |
1555 | /* |
1556 | * How many references to @entry are currently swapped out? | |
1557 | * This considers COUNT_CONTINUED so it returns exact answer. | |
1558 | */ | |
1559 | int swp_swapcount(swp_entry_t entry) | |
1560 | { | |
1561 | int count, tmp_count, n; | |
1562 | struct swap_info_struct *p; | |
235b6217 | 1563 | struct swap_cluster_info *ci; |
8334b962 MK |
1564 | struct page *page; |
1565 | pgoff_t offset; | |
1566 | unsigned char *map; | |
1567 | ||
235b6217 | 1568 | p = _swap_info_get(entry); |
8334b962 MK |
1569 | if (!p) |
1570 | return 0; | |
1571 | ||
235b6217 HY |
1572 | offset = swp_offset(entry); |
1573 | ||
1574 | ci = lock_cluster_or_swap_info(p, offset); | |
1575 | ||
1576 | count = swap_count(p->swap_map[offset]); | |
8334b962 MK |
1577 | if (!(count & COUNT_CONTINUED)) |
1578 | goto out; | |
1579 | ||
1580 | count &= ~COUNT_CONTINUED; | |
1581 | n = SWAP_MAP_MAX + 1; | |
1582 | ||
8334b962 MK |
1583 | page = vmalloc_to_page(p->swap_map + offset); |
1584 | offset &= ~PAGE_MASK; | |
1585 | VM_BUG_ON(page_private(page) != SWP_CONTINUED); | |
1586 | ||
1587 | do { | |
a8ae4991 | 1588 | page = list_next_entry(page, lru); |
8334b962 MK |
1589 | map = kmap_atomic(page); |
1590 | tmp_count = map[offset]; | |
1591 | kunmap_atomic(map); | |
1592 | ||
1593 | count += (tmp_count & ~COUNT_CONTINUED) * n; | |
1594 | n *= (SWAP_CONT_MAX + 1); | |
1595 | } while (tmp_count & COUNT_CONTINUED); | |
1596 | out: | |
235b6217 | 1597 | unlock_cluster_or_swap_info(p, ci); |
8334b962 MK |
1598 | return count; |
1599 | } | |
1600 | ||
e0709829 HY |
1601 | static bool swap_page_trans_huge_swapped(struct swap_info_struct *si, |
1602 | swp_entry_t entry) | |
1603 | { | |
1604 | struct swap_cluster_info *ci; | |
1605 | unsigned char *map = si->swap_map; | |
1606 | unsigned long roffset = swp_offset(entry); | |
1607 | unsigned long offset = round_down(roffset, SWAPFILE_CLUSTER); | |
1608 | int i; | |
1609 | bool ret = false; | |
1610 | ||
1611 | ci = lock_cluster_or_swap_info(si, offset); | |
1612 | if (!ci || !cluster_is_huge(ci)) { | |
afa4711e | 1613 | if (swap_count(map[roffset])) |
e0709829 HY |
1614 | ret = true; |
1615 | goto unlock_out; | |
1616 | } | |
1617 | for (i = 0; i < SWAPFILE_CLUSTER; i++) { | |
afa4711e | 1618 | if (swap_count(map[offset + i])) { |
e0709829 HY |
1619 | ret = true; |
1620 | break; | |
1621 | } | |
1622 | } | |
1623 | unlock_out: | |
1624 | unlock_cluster_or_swap_info(si, ci); | |
1625 | return ret; | |
1626 | } | |
1627 | ||
1628 | static bool page_swapped(struct page *page) | |
1629 | { | |
1630 | swp_entry_t entry; | |
1631 | struct swap_info_struct *si; | |
1632 | ||
fe5266d5 | 1633 | if (!IS_ENABLED(CONFIG_THP_SWAP) || likely(!PageTransCompound(page))) |
e0709829 HY |
1634 | return page_swapcount(page) != 0; |
1635 | ||
1636 | page = compound_head(page); | |
1637 | entry.val = page_private(page); | |
1638 | si = _swap_info_get(entry); | |
1639 | if (si) | |
1640 | return swap_page_trans_huge_swapped(si, entry); | |
1641 | return false; | |
1642 | } | |
ba3c4ce6 HY |
1643 | |
1644 | static int page_trans_huge_map_swapcount(struct page *page, int *total_mapcount, | |
1645 | int *total_swapcount) | |
1646 | { | |
1647 | int i, map_swapcount, _total_mapcount, _total_swapcount; | |
1648 | unsigned long offset = 0; | |
1649 | struct swap_info_struct *si; | |
1650 | struct swap_cluster_info *ci = NULL; | |
1651 | unsigned char *map = NULL; | |
1652 | int mapcount, swapcount = 0; | |
1653 | ||
1654 | /* hugetlbfs shouldn't call it */ | |
1655 | VM_BUG_ON_PAGE(PageHuge(page), page); | |
1656 | ||
fe5266d5 HY |
1657 | if (!IS_ENABLED(CONFIG_THP_SWAP) || likely(!PageTransCompound(page))) { |
1658 | mapcount = page_trans_huge_mapcount(page, total_mapcount); | |
ba3c4ce6 HY |
1659 | if (PageSwapCache(page)) |
1660 | swapcount = page_swapcount(page); | |
1661 | if (total_swapcount) | |
1662 | *total_swapcount = swapcount; | |
1663 | return mapcount + swapcount; | |
1664 | } | |
1665 | ||
1666 | page = compound_head(page); | |
1667 | ||
1668 | _total_mapcount = _total_swapcount = map_swapcount = 0; | |
1669 | if (PageSwapCache(page)) { | |
1670 | swp_entry_t entry; | |
1671 | ||
1672 | entry.val = page_private(page); | |
1673 | si = _swap_info_get(entry); | |
1674 | if (si) { | |
1675 | map = si->swap_map; | |
1676 | offset = swp_offset(entry); | |
1677 | } | |
1678 | } | |
1679 | if (map) | |
1680 | ci = lock_cluster(si, offset); | |
1681 | for (i = 0; i < HPAGE_PMD_NR; i++) { | |
1682 | mapcount = atomic_read(&page[i]._mapcount) + 1; | |
1683 | _total_mapcount += mapcount; | |
1684 | if (map) { | |
1685 | swapcount = swap_count(map[offset + i]); | |
1686 | _total_swapcount += swapcount; | |
1687 | } | |
1688 | map_swapcount = max(map_swapcount, mapcount + swapcount); | |
1689 | } | |
1690 | unlock_cluster(ci); | |
1691 | if (PageDoubleMap(page)) { | |
1692 | map_swapcount -= 1; | |
1693 | _total_mapcount -= HPAGE_PMD_NR; | |
1694 | } | |
1695 | mapcount = compound_mapcount(page); | |
1696 | map_swapcount += mapcount; | |
1697 | _total_mapcount += mapcount; | |
1698 | if (total_mapcount) | |
1699 | *total_mapcount = _total_mapcount; | |
1700 | if (total_swapcount) | |
1701 | *total_swapcount = _total_swapcount; | |
1702 | ||
1703 | return map_swapcount; | |
1704 | } | |
e0709829 | 1705 | |
1da177e4 | 1706 | /* |
7b1fe597 HD |
1707 | * We can write to an anon page without COW if there are no other references |
1708 | * to it. And as a side-effect, free up its swap: because the old content | |
1709 | * on disk will never be read, and seeking back there to write new content | |
1710 | * later would only waste time away from clustering. | |
6d0a07ed | 1711 | * |
ba3c4ce6 | 1712 | * NOTE: total_map_swapcount should not be relied upon by the caller if |
6d0a07ed AA |
1713 | * reuse_swap_page() returns false, but it may be always overwritten |
1714 | * (see the other implementation for CONFIG_SWAP=n). | |
1da177e4 | 1715 | */ |
ba3c4ce6 | 1716 | bool reuse_swap_page(struct page *page, int *total_map_swapcount) |
1da177e4 | 1717 | { |
ba3c4ce6 | 1718 | int count, total_mapcount, total_swapcount; |
c475a8ab | 1719 | |
309381fe | 1720 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
5ad64688 | 1721 | if (unlikely(PageKsm(page))) |
6d0a07ed | 1722 | return false; |
ba3c4ce6 HY |
1723 | count = page_trans_huge_map_swapcount(page, &total_mapcount, |
1724 | &total_swapcount); | |
1725 | if (total_map_swapcount) | |
1726 | *total_map_swapcount = total_mapcount + total_swapcount; | |
1727 | if (count == 1 && PageSwapCache(page) && | |
1728 | (likely(!PageTransCompound(page)) || | |
1729 | /* The remaining swap count will be freed soon */ | |
1730 | total_swapcount == page_swapcount(page))) { | |
f0571429 | 1731 | if (!PageWriteback(page)) { |
ba3c4ce6 | 1732 | page = compound_head(page); |
7b1fe597 HD |
1733 | delete_from_swap_cache(page); |
1734 | SetPageDirty(page); | |
f0571429 MK |
1735 | } else { |
1736 | swp_entry_t entry; | |
1737 | struct swap_info_struct *p; | |
1738 | ||
1739 | entry.val = page_private(page); | |
1740 | p = swap_info_get(entry); | |
1741 | if (p->flags & SWP_STABLE_WRITES) { | |
1742 | spin_unlock(&p->lock); | |
1743 | return false; | |
1744 | } | |
1745 | spin_unlock(&p->lock); | |
7b1fe597 HD |
1746 | } |
1747 | } | |
ba3c4ce6 | 1748 | |
5ad64688 | 1749 | return count <= 1; |
1da177e4 LT |
1750 | } |
1751 | ||
1752 | /* | |
a2c43eed HD |
1753 | * If swap is getting full, or if there are no more mappings of this page, |
1754 | * then try_to_free_swap is called to free its swap space. | |
1da177e4 | 1755 | */ |
a2c43eed | 1756 | int try_to_free_swap(struct page *page) |
1da177e4 | 1757 | { |
309381fe | 1758 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
1da177e4 LT |
1759 | |
1760 | if (!PageSwapCache(page)) | |
1761 | return 0; | |
1762 | if (PageWriteback(page)) | |
1763 | return 0; | |
e0709829 | 1764 | if (page_swapped(page)) |
1da177e4 LT |
1765 | return 0; |
1766 | ||
b73d7fce HD |
1767 | /* |
1768 | * Once hibernation has begun to create its image of memory, | |
1769 | * there's a danger that one of the calls to try_to_free_swap() | |
1770 | * - most probably a call from __try_to_reclaim_swap() while | |
1771 | * hibernation is allocating its own swap pages for the image, | |
1772 | * but conceivably even a call from memory reclaim - will free | |
1773 | * the swap from a page which has already been recorded in the | |
1774 | * image as a clean swapcache page, and then reuse its swap for | |
1775 | * another page of the image. On waking from hibernation, the | |
1776 | * original page might be freed under memory pressure, then | |
1777 | * later read back in from swap, now with the wrong data. | |
1778 | * | |
2de1a7e4 | 1779 | * Hibernation suspends storage while it is writing the image |
f90ac398 | 1780 | * to disk so check that here. |
b73d7fce | 1781 | */ |
f90ac398 | 1782 | if (pm_suspended_storage()) |
b73d7fce HD |
1783 | return 0; |
1784 | ||
e0709829 | 1785 | page = compound_head(page); |
a2c43eed HD |
1786 | delete_from_swap_cache(page); |
1787 | SetPageDirty(page); | |
1788 | return 1; | |
68a22394 RR |
1789 | } |
1790 | ||
1da177e4 LT |
1791 | /* |
1792 | * Free the swap entry like above, but also try to | |
1793 | * free the page cache entry if it is the last user. | |
1794 | */ | |
2509ef26 | 1795 | int free_swap_and_cache(swp_entry_t entry) |
1da177e4 | 1796 | { |
2509ef26 | 1797 | struct swap_info_struct *p; |
7c00bafe | 1798 | unsigned char count; |
1da177e4 | 1799 | |
a7420aa5 | 1800 | if (non_swap_entry(entry)) |
2509ef26 | 1801 | return 1; |
0697212a | 1802 | |
7c00bafe | 1803 | p = _swap_info_get(entry); |
1da177e4 | 1804 | if (p) { |
33e16272 | 1805 | count = __swap_entry_free(p, entry); |
e0709829 | 1806 | if (count == SWAP_HAS_CACHE && |
bcd49e86 HY |
1807 | !swap_page_trans_huge_swapped(p, entry)) |
1808 | __try_to_reclaim_swap(p, swp_offset(entry), | |
1809 | TTRS_UNMAPPED | TTRS_FULL); | |
1da177e4 | 1810 | } |
2509ef26 | 1811 | return p != NULL; |
1da177e4 LT |
1812 | } |
1813 | ||
b0cb1a19 | 1814 | #ifdef CONFIG_HIBERNATION |
f577eb30 | 1815 | /* |
915bae9e | 1816 | * Find the swap type that corresponds to given device (if any). |
f577eb30 | 1817 | * |
915bae9e RW |
1818 | * @offset - number of the PAGE_SIZE-sized block of the device, starting |
1819 | * from 0, in which the swap header is expected to be located. | |
1820 | * | |
1821 | * This is needed for the suspend to disk (aka swsusp). | |
f577eb30 | 1822 | */ |
21bd9005 | 1823 | int swap_type_of(dev_t device, sector_t offset) |
f577eb30 | 1824 | { |
efa90a98 | 1825 | int type; |
f577eb30 | 1826 | |
21bd9005 CH |
1827 | if (!device) |
1828 | return -1; | |
915bae9e | 1829 | |
f577eb30 | 1830 | spin_lock(&swap_lock); |
efa90a98 HD |
1831 | for (type = 0; type < nr_swapfiles; type++) { |
1832 | struct swap_info_struct *sis = swap_info[type]; | |
f577eb30 | 1833 | |
915bae9e | 1834 | if (!(sis->flags & SWP_WRITEOK)) |
f577eb30 | 1835 | continue; |
b6b5bce3 | 1836 | |
21bd9005 | 1837 | if (device == sis->bdev->bd_dev) { |
4efaceb1 | 1838 | struct swap_extent *se = first_se(sis); |
915bae9e | 1839 | |
915bae9e RW |
1840 | if (se->start_block == offset) { |
1841 | spin_unlock(&swap_lock); | |
efa90a98 | 1842 | return type; |
915bae9e | 1843 | } |
f577eb30 RW |
1844 | } |
1845 | } | |
1846 | spin_unlock(&swap_lock); | |
21bd9005 CH |
1847 | return -ENODEV; |
1848 | } | |
915bae9e | 1849 | |
21bd9005 CH |
1850 | int find_first_swap(dev_t *device) |
1851 | { | |
1852 | int type; | |
915bae9e | 1853 | |
21bd9005 CH |
1854 | spin_lock(&swap_lock); |
1855 | for (type = 0; type < nr_swapfiles; type++) { | |
1856 | struct swap_info_struct *sis = swap_info[type]; | |
1857 | ||
1858 | if (!(sis->flags & SWP_WRITEOK)) | |
1859 | continue; | |
1860 | *device = sis->bdev->bd_dev; | |
1861 | spin_unlock(&swap_lock); | |
1862 | return type; | |
1863 | } | |
1864 | spin_unlock(&swap_lock); | |
f577eb30 RW |
1865 | return -ENODEV; |
1866 | } | |
1867 | ||
73c34b6a HD |
1868 | /* |
1869 | * Get the (PAGE_SIZE) block corresponding to given offset on the swapdev | |
1870 | * corresponding to given index in swap_info (swap type). | |
1871 | */ | |
1872 | sector_t swapdev_block(int type, pgoff_t offset) | |
1873 | { | |
c10d38cc | 1874 | struct swap_info_struct *si = swap_type_to_swap_info(type); |
f885056a | 1875 | struct swap_extent *se; |
73c34b6a | 1876 | |
c10d38cc | 1877 | if (!si || !(si->flags & SWP_WRITEOK)) |
73c34b6a | 1878 | return 0; |
f885056a CH |
1879 | se = offset_to_swap_extent(si, offset); |
1880 | return se->start_block + (offset - se->start_page); | |
73c34b6a HD |
1881 | } |
1882 | ||
f577eb30 RW |
1883 | /* |
1884 | * Return either the total number of swap pages of given type, or the number | |
1885 | * of free pages of that type (depending on @free) | |
1886 | * | |
1887 | * This is needed for software suspend | |
1888 | */ | |
1889 | unsigned int count_swap_pages(int type, int free) | |
1890 | { | |
1891 | unsigned int n = 0; | |
1892 | ||
efa90a98 HD |
1893 | spin_lock(&swap_lock); |
1894 | if ((unsigned int)type < nr_swapfiles) { | |
1895 | struct swap_info_struct *sis = swap_info[type]; | |
1896 | ||
ec8acf20 | 1897 | spin_lock(&sis->lock); |
efa90a98 HD |
1898 | if (sis->flags & SWP_WRITEOK) { |
1899 | n = sis->pages; | |
f577eb30 | 1900 | if (free) |
efa90a98 | 1901 | n -= sis->inuse_pages; |
f577eb30 | 1902 | } |
ec8acf20 | 1903 | spin_unlock(&sis->lock); |
f577eb30 | 1904 | } |
efa90a98 | 1905 | spin_unlock(&swap_lock); |
f577eb30 RW |
1906 | return n; |
1907 | } | |
73c34b6a | 1908 | #endif /* CONFIG_HIBERNATION */ |
f577eb30 | 1909 | |
9f8bdb3f | 1910 | static inline int pte_same_as_swp(pte_t pte, pte_t swp_pte) |
179ef71c | 1911 | { |
099dd687 | 1912 | return pte_same(pte_swp_clear_flags(pte), swp_pte); |
179ef71c CG |
1913 | } |
1914 | ||
1da177e4 | 1915 | /* |
72866f6f HD |
1916 | * No need to decide whether this PTE shares the swap entry with others, |
1917 | * just let do_wp_page work it out if a write is requested later - to | |
1918 | * force COW, vm_page_prot omits write permission from any private vma. | |
1da177e4 | 1919 | */ |
044d66c1 | 1920 | static int unuse_pte(struct vm_area_struct *vma, pmd_t *pmd, |
1da177e4 LT |
1921 | unsigned long addr, swp_entry_t entry, struct page *page) |
1922 | { | |
9e16b7fb | 1923 | struct page *swapcache; |
044d66c1 HD |
1924 | spinlock_t *ptl; |
1925 | pte_t *pte; | |
1926 | int ret = 1; | |
1927 | ||
9e16b7fb HD |
1928 | swapcache = page; |
1929 | page = ksm_might_need_to_copy(page, vma, addr); | |
1930 | if (unlikely(!page)) | |
1931 | return -ENOMEM; | |
1932 | ||
044d66c1 | 1933 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
9f8bdb3f | 1934 | if (unlikely(!pte_same_as_swp(*pte, swp_entry_to_pte(entry)))) { |
044d66c1 HD |
1935 | ret = 0; |
1936 | goto out; | |
1937 | } | |
8a9f3ccd | 1938 | |
b084d435 | 1939 | dec_mm_counter(vma->vm_mm, MM_SWAPENTS); |
d559db08 | 1940 | inc_mm_counter(vma->vm_mm, MM_ANONPAGES); |
1da177e4 LT |
1941 | get_page(page); |
1942 | set_pte_at(vma->vm_mm, addr, pte, | |
1943 | pte_mkold(mk_pte(page, vma->vm_page_prot))); | |
00501b53 | 1944 | if (page == swapcache) { |
be5d0a74 | 1945 | page_add_anon_rmap(page, vma, addr, false); |
00501b53 | 1946 | } else { /* ksm created a completely new copy */ |
be5d0a74 | 1947 | page_add_new_anon_rmap(page, vma, addr, false); |
b518154e | 1948 | lru_cache_add_inactive_or_unevictable(page, vma); |
00501b53 | 1949 | } |
1da177e4 | 1950 | swap_free(entry); |
044d66c1 HD |
1951 | out: |
1952 | pte_unmap_unlock(pte, ptl); | |
9e16b7fb HD |
1953 | if (page != swapcache) { |
1954 | unlock_page(page); | |
1955 | put_page(page); | |
1956 | } | |
044d66c1 | 1957 | return ret; |
1da177e4 LT |
1958 | } |
1959 | ||
1960 | static int unuse_pte_range(struct vm_area_struct *vma, pmd_t *pmd, | |
b56a2d8a VRP |
1961 | unsigned long addr, unsigned long end, |
1962 | unsigned int type, bool frontswap, | |
1963 | unsigned long *fs_pages_to_unuse) | |
1da177e4 | 1964 | { |
b56a2d8a VRP |
1965 | struct page *page; |
1966 | swp_entry_t entry; | |
705e87c0 | 1967 | pte_t *pte; |
b56a2d8a VRP |
1968 | struct swap_info_struct *si; |
1969 | unsigned long offset; | |
8a9f3ccd | 1970 | int ret = 0; |
b56a2d8a | 1971 | volatile unsigned char *swap_map; |
1da177e4 | 1972 | |
b56a2d8a | 1973 | si = swap_info[type]; |
044d66c1 | 1974 | pte = pte_offset_map(pmd, addr); |
1da177e4 | 1975 | do { |
b56a2d8a VRP |
1976 | if (!is_swap_pte(*pte)) |
1977 | continue; | |
1978 | ||
1979 | entry = pte_to_swp_entry(*pte); | |
1980 | if (swp_type(entry) != type) | |
1981 | continue; | |
1982 | ||
1983 | offset = swp_offset(entry); | |
1984 | if (frontswap && !frontswap_test(si, offset)) | |
1985 | continue; | |
1986 | ||
1987 | pte_unmap(pte); | |
1988 | swap_map = &si->swap_map[offset]; | |
ebc5951e AR |
1989 | page = lookup_swap_cache(entry, vma, addr); |
1990 | if (!page) { | |
8c63ca5b WD |
1991 | struct vm_fault vmf = { |
1992 | .vma = vma, | |
1993 | .address = addr, | |
1994 | .pmd = pmd, | |
1995 | }; | |
1996 | ||
ebc5951e AR |
1997 | page = swapin_readahead(entry, GFP_HIGHUSER_MOVABLE, |
1998 | &vmf); | |
1999 | } | |
b56a2d8a VRP |
2000 | if (!page) { |
2001 | if (*swap_map == 0 || *swap_map == SWAP_MAP_BAD) | |
2002 | goto try_next; | |
2003 | return -ENOMEM; | |
2004 | } | |
2005 | ||
2006 | lock_page(page); | |
2007 | wait_on_page_writeback(page); | |
2008 | ret = unuse_pte(vma, pmd, addr, entry, page); | |
2009 | if (ret < 0) { | |
2010 | unlock_page(page); | |
2011 | put_page(page); | |
2012 | goto out; | |
2013 | } | |
2014 | ||
2015 | try_to_free_swap(page); | |
2016 | unlock_page(page); | |
2017 | put_page(page); | |
2018 | ||
2019 | if (*fs_pages_to_unuse && !--(*fs_pages_to_unuse)) { | |
2020 | ret = FRONTSWAP_PAGES_UNUSED; | |
2021 | goto out; | |
1da177e4 | 2022 | } |
b56a2d8a VRP |
2023 | try_next: |
2024 | pte = pte_offset_map(pmd, addr); | |
1da177e4 | 2025 | } while (pte++, addr += PAGE_SIZE, addr != end); |
044d66c1 | 2026 | pte_unmap(pte - 1); |
b56a2d8a VRP |
2027 | |
2028 | ret = 0; | |
044d66c1 | 2029 | out: |
8a9f3ccd | 2030 | return ret; |
1da177e4 LT |
2031 | } |
2032 | ||
2033 | static inline int unuse_pmd_range(struct vm_area_struct *vma, pud_t *pud, | |
2034 | unsigned long addr, unsigned long end, | |
b56a2d8a VRP |
2035 | unsigned int type, bool frontswap, |
2036 | unsigned long *fs_pages_to_unuse) | |
1da177e4 LT |
2037 | { |
2038 | pmd_t *pmd; | |
2039 | unsigned long next; | |
8a9f3ccd | 2040 | int ret; |
1da177e4 LT |
2041 | |
2042 | pmd = pmd_offset(pud, addr); | |
2043 | do { | |
dc644a07 | 2044 | cond_resched(); |
1da177e4 | 2045 | next = pmd_addr_end(addr, end); |
1a5a9906 | 2046 | if (pmd_none_or_trans_huge_or_clear_bad(pmd)) |
1da177e4 | 2047 | continue; |
b56a2d8a VRP |
2048 | ret = unuse_pte_range(vma, pmd, addr, next, type, |
2049 | frontswap, fs_pages_to_unuse); | |
8a9f3ccd BS |
2050 | if (ret) |
2051 | return ret; | |
1da177e4 LT |
2052 | } while (pmd++, addr = next, addr != end); |
2053 | return 0; | |
2054 | } | |
2055 | ||
c2febafc | 2056 | static inline int unuse_pud_range(struct vm_area_struct *vma, p4d_t *p4d, |
1da177e4 | 2057 | unsigned long addr, unsigned long end, |
b56a2d8a VRP |
2058 | unsigned int type, bool frontswap, |
2059 | unsigned long *fs_pages_to_unuse) | |
1da177e4 LT |
2060 | { |
2061 | pud_t *pud; | |
2062 | unsigned long next; | |
8a9f3ccd | 2063 | int ret; |
1da177e4 | 2064 | |
c2febafc | 2065 | pud = pud_offset(p4d, addr); |
1da177e4 LT |
2066 | do { |
2067 | next = pud_addr_end(addr, end); | |
2068 | if (pud_none_or_clear_bad(pud)) | |
2069 | continue; | |
b56a2d8a VRP |
2070 | ret = unuse_pmd_range(vma, pud, addr, next, type, |
2071 | frontswap, fs_pages_to_unuse); | |
8a9f3ccd BS |
2072 | if (ret) |
2073 | return ret; | |
1da177e4 LT |
2074 | } while (pud++, addr = next, addr != end); |
2075 | return 0; | |
2076 | } | |
2077 | ||
c2febafc KS |
2078 | static inline int unuse_p4d_range(struct vm_area_struct *vma, pgd_t *pgd, |
2079 | unsigned long addr, unsigned long end, | |
b56a2d8a VRP |
2080 | unsigned int type, bool frontswap, |
2081 | unsigned long *fs_pages_to_unuse) | |
c2febafc KS |
2082 | { |
2083 | p4d_t *p4d; | |
2084 | unsigned long next; | |
2085 | int ret; | |
2086 | ||
2087 | p4d = p4d_offset(pgd, addr); | |
2088 | do { | |
2089 | next = p4d_addr_end(addr, end); | |
2090 | if (p4d_none_or_clear_bad(p4d)) | |
2091 | continue; | |
b56a2d8a VRP |
2092 | ret = unuse_pud_range(vma, p4d, addr, next, type, |
2093 | frontswap, fs_pages_to_unuse); | |
c2febafc KS |
2094 | if (ret) |
2095 | return ret; | |
2096 | } while (p4d++, addr = next, addr != end); | |
2097 | return 0; | |
2098 | } | |
2099 | ||
b56a2d8a VRP |
2100 | static int unuse_vma(struct vm_area_struct *vma, unsigned int type, |
2101 | bool frontswap, unsigned long *fs_pages_to_unuse) | |
1da177e4 LT |
2102 | { |
2103 | pgd_t *pgd; | |
2104 | unsigned long addr, end, next; | |
8a9f3ccd | 2105 | int ret; |
1da177e4 | 2106 | |
b56a2d8a VRP |
2107 | addr = vma->vm_start; |
2108 | end = vma->vm_end; | |
1da177e4 LT |
2109 | |
2110 | pgd = pgd_offset(vma->vm_mm, addr); | |
2111 | do { | |
2112 | next = pgd_addr_end(addr, end); | |
2113 | if (pgd_none_or_clear_bad(pgd)) | |
2114 | continue; | |
b56a2d8a VRP |
2115 | ret = unuse_p4d_range(vma, pgd, addr, next, type, |
2116 | frontswap, fs_pages_to_unuse); | |
8a9f3ccd BS |
2117 | if (ret) |
2118 | return ret; | |
1da177e4 LT |
2119 | } while (pgd++, addr = next, addr != end); |
2120 | return 0; | |
2121 | } | |
2122 | ||
b56a2d8a VRP |
2123 | static int unuse_mm(struct mm_struct *mm, unsigned int type, |
2124 | bool frontswap, unsigned long *fs_pages_to_unuse) | |
1da177e4 LT |
2125 | { |
2126 | struct vm_area_struct *vma; | |
8a9f3ccd | 2127 | int ret = 0; |
1da177e4 | 2128 | |
d8ed45c5 | 2129 | mmap_read_lock(mm); |
1da177e4 | 2130 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
b56a2d8a VRP |
2131 | if (vma->anon_vma) { |
2132 | ret = unuse_vma(vma, type, frontswap, | |
2133 | fs_pages_to_unuse); | |
2134 | if (ret) | |
2135 | break; | |
2136 | } | |
dc644a07 | 2137 | cond_resched(); |
1da177e4 | 2138 | } |
d8ed45c5 | 2139 | mmap_read_unlock(mm); |
b56a2d8a | 2140 | return ret; |
1da177e4 LT |
2141 | } |
2142 | ||
2143 | /* | |
38b5faf4 | 2144 | * Scan swap_map (or frontswap_map if frontswap parameter is true) |
b56a2d8a VRP |
2145 | * from current position to next entry still in use. Return 0 |
2146 | * if there are no inuse entries after prev till end of the map. | |
1da177e4 | 2147 | */ |
6eb396dc | 2148 | static unsigned int find_next_to_unuse(struct swap_info_struct *si, |
38b5faf4 | 2149 | unsigned int prev, bool frontswap) |
1da177e4 | 2150 | { |
b56a2d8a | 2151 | unsigned int i; |
8d69aaee | 2152 | unsigned char count; |
1da177e4 LT |
2153 | |
2154 | /* | |
5d337b91 | 2155 | * No need for swap_lock here: we're just looking |
1da177e4 LT |
2156 | * for whether an entry is in use, not modifying it; false |
2157 | * hits are okay, and sys_swapoff() has already prevented new | |
5d337b91 | 2158 | * allocations from this area (while holding swap_lock). |
1da177e4 | 2159 | */ |
b56a2d8a | 2160 | for (i = prev + 1; i < si->max; i++) { |
4db0c3c2 | 2161 | count = READ_ONCE(si->swap_map[i]); |
355cfa73 | 2162 | if (count && swap_count(count) != SWAP_MAP_BAD) |
dc644a07 HD |
2163 | if (!frontswap || frontswap_test(si, i)) |
2164 | break; | |
2165 | if ((i % LATENCY_LIMIT) == 0) | |
2166 | cond_resched(); | |
1da177e4 | 2167 | } |
b56a2d8a VRP |
2168 | |
2169 | if (i == si->max) | |
2170 | i = 0; | |
2171 | ||
1da177e4 LT |
2172 | return i; |
2173 | } | |
2174 | ||
2175 | /* | |
b56a2d8a | 2176 | * If the boolean frontswap is true, only unuse pages_to_unuse pages; |
38b5faf4 | 2177 | * pages_to_unuse==0 means all pages; ignored if frontswap is false |
1da177e4 | 2178 | */ |
38b5faf4 DM |
2179 | int try_to_unuse(unsigned int type, bool frontswap, |
2180 | unsigned long pages_to_unuse) | |
1da177e4 | 2181 | { |
b56a2d8a VRP |
2182 | struct mm_struct *prev_mm; |
2183 | struct mm_struct *mm; | |
2184 | struct list_head *p; | |
2185 | int retval = 0; | |
efa90a98 | 2186 | struct swap_info_struct *si = swap_info[type]; |
1da177e4 LT |
2187 | struct page *page; |
2188 | swp_entry_t entry; | |
b56a2d8a | 2189 | unsigned int i; |
1da177e4 | 2190 | |
21820948 | 2191 | if (!READ_ONCE(si->inuse_pages)) |
b56a2d8a | 2192 | return 0; |
1da177e4 | 2193 | |
b56a2d8a VRP |
2194 | if (!frontswap) |
2195 | pages_to_unuse = 0; | |
2196 | ||
2197 | retry: | |
2198 | retval = shmem_unuse(type, frontswap, &pages_to_unuse); | |
2199 | if (retval) | |
2200 | goto out; | |
2201 | ||
2202 | prev_mm = &init_mm; | |
2203 | mmget(prev_mm); | |
2204 | ||
2205 | spin_lock(&mmlist_lock); | |
2206 | p = &init_mm.mmlist; | |
21820948 | 2207 | while (READ_ONCE(si->inuse_pages) && |
64165b1a HD |
2208 | !signal_pending(current) && |
2209 | (p = p->next) != &init_mm.mmlist) { | |
1da177e4 | 2210 | |
b56a2d8a VRP |
2211 | mm = list_entry(p, struct mm_struct, mmlist); |
2212 | if (!mmget_not_zero(mm)) | |
2213 | continue; | |
2214 | spin_unlock(&mmlist_lock); | |
2215 | mmput(prev_mm); | |
2216 | prev_mm = mm; | |
2217 | retval = unuse_mm(mm, type, frontswap, &pages_to_unuse); | |
1da177e4 | 2218 | |
b56a2d8a VRP |
2219 | if (retval) { |
2220 | mmput(prev_mm); | |
2221 | goto out; | |
1da177e4 LT |
2222 | } |
2223 | ||
2224 | /* | |
b56a2d8a VRP |
2225 | * Make sure that we aren't completely killing |
2226 | * interactive performance. | |
1da177e4 | 2227 | */ |
b56a2d8a VRP |
2228 | cond_resched(); |
2229 | spin_lock(&mmlist_lock); | |
2230 | } | |
2231 | spin_unlock(&mmlist_lock); | |
1da177e4 | 2232 | |
b56a2d8a | 2233 | mmput(prev_mm); |
1da177e4 | 2234 | |
b56a2d8a | 2235 | i = 0; |
21820948 | 2236 | while (READ_ONCE(si->inuse_pages) && |
64165b1a HD |
2237 | !signal_pending(current) && |
2238 | (i = find_next_to_unuse(si, i, frontswap)) != 0) { | |
1da177e4 | 2239 | |
b56a2d8a VRP |
2240 | entry = swp_entry(type, i); |
2241 | page = find_get_page(swap_address_space(entry), i); | |
2242 | if (!page) | |
2243 | continue; | |
68bdc8d6 HD |
2244 | |
2245 | /* | |
2246 | * It is conceivable that a racing task removed this page from | |
b56a2d8a VRP |
2247 | * swap cache just before we acquired the page lock. The page |
2248 | * might even be back in swap cache on another swap area. But | |
2249 | * that is okay, try_to_free_swap() only removes stale pages. | |
1da177e4 | 2250 | */ |
b56a2d8a VRP |
2251 | lock_page(page); |
2252 | wait_on_page_writeback(page); | |
2253 | try_to_free_swap(page); | |
1da177e4 | 2254 | unlock_page(page); |
09cbfeaf | 2255 | put_page(page); |
1da177e4 LT |
2256 | |
2257 | /* | |
b56a2d8a VRP |
2258 | * For frontswap, we just need to unuse pages_to_unuse, if |
2259 | * it was specified. Need not check frontswap again here as | |
2260 | * we already zeroed out pages_to_unuse if not frontswap. | |
1da177e4 | 2261 | */ |
b56a2d8a VRP |
2262 | if (pages_to_unuse && --pages_to_unuse == 0) |
2263 | goto out; | |
1da177e4 LT |
2264 | } |
2265 | ||
b56a2d8a VRP |
2266 | /* |
2267 | * Lets check again to see if there are still swap entries in the map. | |
2268 | * If yes, we would need to do retry the unuse logic again. | |
2269 | * Under global memory pressure, swap entries can be reinserted back | |
2270 | * into process space after the mmlist loop above passes over them. | |
dd862deb | 2271 | * |
af53d3e9 HD |
2272 | * Limit the number of retries? No: when mmget_not_zero() above fails, |
2273 | * that mm is likely to be freeing swap from exit_mmap(), which proceeds | |
2274 | * at its own independent pace; and even shmem_writepage() could have | |
2275 | * been preempted after get_swap_page(), temporarily hiding that swap. | |
2276 | * It's easy and robust (though cpu-intensive) just to keep retrying. | |
b56a2d8a | 2277 | */ |
21820948 | 2278 | if (READ_ONCE(si->inuse_pages)) { |
64165b1a HD |
2279 | if (!signal_pending(current)) |
2280 | goto retry; | |
2281 | retval = -EINTR; | |
2282 | } | |
b56a2d8a VRP |
2283 | out: |
2284 | return (retval == FRONTSWAP_PAGES_UNUSED) ? 0 : retval; | |
1da177e4 LT |
2285 | } |
2286 | ||
2287 | /* | |
5d337b91 HD |
2288 | * After a successful try_to_unuse, if no swap is now in use, we know |
2289 | * we can empty the mmlist. swap_lock must be held on entry and exit. | |
2290 | * Note that mmlist_lock nests inside swap_lock, and an mm must be | |
1da177e4 LT |
2291 | * added to the mmlist just after page_duplicate - before would be racy. |
2292 | */ | |
2293 | static void drain_mmlist(void) | |
2294 | { | |
2295 | struct list_head *p, *next; | |
efa90a98 | 2296 | unsigned int type; |
1da177e4 | 2297 | |
efa90a98 HD |
2298 | for (type = 0; type < nr_swapfiles; type++) |
2299 | if (swap_info[type]->inuse_pages) | |
1da177e4 LT |
2300 | return; |
2301 | spin_lock(&mmlist_lock); | |
2302 | list_for_each_safe(p, next, &init_mm.mmlist) | |
2303 | list_del_init(p); | |
2304 | spin_unlock(&mmlist_lock); | |
2305 | } | |
2306 | ||
1da177e4 LT |
2307 | /* |
2308 | * Free all of a swapdev's extent information | |
2309 | */ | |
2310 | static void destroy_swap_extents(struct swap_info_struct *sis) | |
2311 | { | |
4efaceb1 AL |
2312 | while (!RB_EMPTY_ROOT(&sis->swap_extent_root)) { |
2313 | struct rb_node *rb = sis->swap_extent_root.rb_node; | |
2314 | struct swap_extent *se = rb_entry(rb, struct swap_extent, rb_node); | |
1da177e4 | 2315 | |
4efaceb1 | 2316 | rb_erase(rb, &sis->swap_extent_root); |
1da177e4 LT |
2317 | kfree(se); |
2318 | } | |
62c230bc | 2319 | |
bc4ae27d | 2320 | if (sis->flags & SWP_ACTIVATED) { |
62c230bc MG |
2321 | struct file *swap_file = sis->swap_file; |
2322 | struct address_space *mapping = swap_file->f_mapping; | |
2323 | ||
bc4ae27d OS |
2324 | sis->flags &= ~SWP_ACTIVATED; |
2325 | if (mapping->a_ops->swap_deactivate) | |
2326 | mapping->a_ops->swap_deactivate(swap_file); | |
62c230bc | 2327 | } |
1da177e4 LT |
2328 | } |
2329 | ||
2330 | /* | |
2331 | * Add a block range (and the corresponding page range) into this swapdev's | |
4efaceb1 | 2332 | * extent tree. |
1da177e4 | 2333 | * |
11d31886 | 2334 | * This function rather assumes that it is called in ascending page order. |
1da177e4 | 2335 | */ |
a509bc1a | 2336 | int |
1da177e4 LT |
2337 | add_swap_extent(struct swap_info_struct *sis, unsigned long start_page, |
2338 | unsigned long nr_pages, sector_t start_block) | |
2339 | { | |
4efaceb1 | 2340 | struct rb_node **link = &sis->swap_extent_root.rb_node, *parent = NULL; |
1da177e4 LT |
2341 | struct swap_extent *se; |
2342 | struct swap_extent *new_se; | |
4efaceb1 AL |
2343 | |
2344 | /* | |
2345 | * place the new node at the right most since the | |
2346 | * function is called in ascending page order. | |
2347 | */ | |
2348 | while (*link) { | |
2349 | parent = *link; | |
2350 | link = &parent->rb_right; | |
2351 | } | |
2352 | ||
2353 | if (parent) { | |
2354 | se = rb_entry(parent, struct swap_extent, rb_node); | |
11d31886 HD |
2355 | BUG_ON(se->start_page + se->nr_pages != start_page); |
2356 | if (se->start_block + se->nr_pages == start_block) { | |
1da177e4 LT |
2357 | /* Merge it */ |
2358 | se->nr_pages += nr_pages; | |
2359 | return 0; | |
2360 | } | |
1da177e4 LT |
2361 | } |
2362 | ||
4efaceb1 | 2363 | /* No merge, insert a new extent. */ |
1da177e4 LT |
2364 | new_se = kmalloc(sizeof(*se), GFP_KERNEL); |
2365 | if (new_se == NULL) | |
2366 | return -ENOMEM; | |
2367 | new_se->start_page = start_page; | |
2368 | new_se->nr_pages = nr_pages; | |
2369 | new_se->start_block = start_block; | |
2370 | ||
4efaceb1 AL |
2371 | rb_link_node(&new_se->rb_node, parent, link); |
2372 | rb_insert_color(&new_se->rb_node, &sis->swap_extent_root); | |
53092a74 | 2373 | return 1; |
1da177e4 | 2374 | } |
aa8aa8a3 | 2375 | EXPORT_SYMBOL_GPL(add_swap_extent); |
1da177e4 LT |
2376 | |
2377 | /* | |
2378 | * A `swap extent' is a simple thing which maps a contiguous range of pages | |
2379 | * onto a contiguous range of disk blocks. An ordered list of swap extents | |
2380 | * is built at swapon time and is then used at swap_writepage/swap_readpage | |
2381 | * time for locating where on disk a page belongs. | |
2382 | * | |
2383 | * If the swapfile is an S_ISBLK block device, a single extent is installed. | |
2384 | * This is done so that the main operating code can treat S_ISBLK and S_ISREG | |
2385 | * swap files identically. | |
2386 | * | |
2387 | * Whether the swapdev is an S_ISREG file or an S_ISBLK blockdev, the swap | |
2388 | * extent list operates in PAGE_SIZE disk blocks. Both S_ISREG and S_ISBLK | |
2389 | * swapfiles are handled *identically* after swapon time. | |
2390 | * | |
2391 | * For S_ISREG swapfiles, setup_swap_extents() will walk all the file's blocks | |
2392 | * and will parse them into an ordered extent list, in PAGE_SIZE chunks. If | |
2393 | * some stray blocks are found which do not fall within the PAGE_SIZE alignment | |
2394 | * requirements, they are simply tossed out - we will never use those blocks | |
2395 | * for swapping. | |
2396 | * | |
1638045c DW |
2397 | * For all swap devices we set S_SWAPFILE across the life of the swapon. This |
2398 | * prevents users from writing to the swap device, which will corrupt memory. | |
1da177e4 LT |
2399 | * |
2400 | * The amount of disk space which a single swap extent represents varies. | |
2401 | * Typically it is in the 1-4 megabyte range. So we can have hundreds of | |
2402 | * extents in the list. To avoid much list walking, we cache the previous | |
2403 | * search location in `curr_swap_extent', and start new searches from there. | |
2404 | * This is extremely effective. The average number of iterations in | |
2405 | * map_swap_page() has been measured at about 0.3 per page. - akpm. | |
2406 | */ | |
53092a74 | 2407 | static int setup_swap_extents(struct swap_info_struct *sis, sector_t *span) |
1da177e4 | 2408 | { |
62c230bc MG |
2409 | struct file *swap_file = sis->swap_file; |
2410 | struct address_space *mapping = swap_file->f_mapping; | |
2411 | struct inode *inode = mapping->host; | |
1da177e4 LT |
2412 | int ret; |
2413 | ||
1da177e4 LT |
2414 | if (S_ISBLK(inode->i_mode)) { |
2415 | ret = add_swap_extent(sis, 0, sis->max, 0); | |
53092a74 | 2416 | *span = sis->pages; |
a509bc1a | 2417 | return ret; |
1da177e4 LT |
2418 | } |
2419 | ||
62c230bc | 2420 | if (mapping->a_ops->swap_activate) { |
a509bc1a | 2421 | ret = mapping->a_ops->swap_activate(sis, swap_file, span); |
bc4ae27d OS |
2422 | if (ret >= 0) |
2423 | sis->flags |= SWP_ACTIVATED; | |
62c230bc | 2424 | if (!ret) { |
32646315 | 2425 | sis->flags |= SWP_FS_OPS; |
62c230bc MG |
2426 | ret = add_swap_extent(sis, 0, sis->max, 0); |
2427 | *span = sis->pages; | |
2428 | } | |
a509bc1a | 2429 | return ret; |
62c230bc MG |
2430 | } |
2431 | ||
a509bc1a | 2432 | return generic_swapfile_activate(sis, swap_file, span); |
1da177e4 LT |
2433 | } |
2434 | ||
a2468cc9 AL |
2435 | static int swap_node(struct swap_info_struct *p) |
2436 | { | |
2437 | struct block_device *bdev; | |
2438 | ||
2439 | if (p->bdev) | |
2440 | bdev = p->bdev; | |
2441 | else | |
2442 | bdev = p->swap_file->f_inode->i_sb->s_bdev; | |
2443 | ||
2444 | return bdev ? bdev->bd_disk->node_id : NUMA_NO_NODE; | |
2445 | } | |
2446 | ||
eb085574 HY |
2447 | static void setup_swap_info(struct swap_info_struct *p, int prio, |
2448 | unsigned char *swap_map, | |
2449 | struct swap_cluster_info *cluster_info) | |
40531542 | 2450 | { |
a2468cc9 AL |
2451 | int i; |
2452 | ||
40531542 CEB |
2453 | if (prio >= 0) |
2454 | p->prio = prio; | |
2455 | else | |
2456 | p->prio = --least_priority; | |
18ab4d4c DS |
2457 | /* |
2458 | * the plist prio is negated because plist ordering is | |
2459 | * low-to-high, while swap ordering is high-to-low | |
2460 | */ | |
2461 | p->list.prio = -p->prio; | |
a2468cc9 AL |
2462 | for_each_node(i) { |
2463 | if (p->prio >= 0) | |
2464 | p->avail_lists[i].prio = -p->prio; | |
2465 | else { | |
2466 | if (swap_node(p) == i) | |
2467 | p->avail_lists[i].prio = 1; | |
2468 | else | |
2469 | p->avail_lists[i].prio = -p->prio; | |
2470 | } | |
2471 | } | |
40531542 | 2472 | p->swap_map = swap_map; |
2a8f9449 | 2473 | p->cluster_info = cluster_info; |
eb085574 HY |
2474 | } |
2475 | ||
2476 | static void _enable_swap_info(struct swap_info_struct *p) | |
2477 | { | |
63d8620e | 2478 | p->flags |= SWP_WRITEOK; |
ec8acf20 | 2479 | atomic_long_add(p->pages, &nr_swap_pages); |
40531542 CEB |
2480 | total_swap_pages += p->pages; |
2481 | ||
adfab836 | 2482 | assert_spin_locked(&swap_lock); |
adfab836 | 2483 | /* |
18ab4d4c DS |
2484 | * both lists are plists, and thus priority ordered. |
2485 | * swap_active_head needs to be priority ordered for swapoff(), | |
2486 | * which on removal of any swap_info_struct with an auto-assigned | |
2487 | * (i.e. negative) priority increments the auto-assigned priority | |
2488 | * of any lower-priority swap_info_structs. | |
2489 | * swap_avail_head needs to be priority ordered for get_swap_page(), | |
2490 | * which allocates swap pages from the highest available priority | |
2491 | * swap_info_struct. | |
adfab836 | 2492 | */ |
18ab4d4c | 2493 | plist_add(&p->list, &swap_active_head); |
a2468cc9 | 2494 | add_to_avail_list(p); |
cf0cac0a CEB |
2495 | } |
2496 | ||
2497 | static void enable_swap_info(struct swap_info_struct *p, int prio, | |
2498 | unsigned char *swap_map, | |
2a8f9449 | 2499 | struct swap_cluster_info *cluster_info, |
cf0cac0a CEB |
2500 | unsigned long *frontswap_map) |
2501 | { | |
4f89849d | 2502 | frontswap_init(p->type, frontswap_map); |
cf0cac0a | 2503 | spin_lock(&swap_lock); |
ec8acf20 | 2504 | spin_lock(&p->lock); |
eb085574 HY |
2505 | setup_swap_info(p, prio, swap_map, cluster_info); |
2506 | spin_unlock(&p->lock); | |
2507 | spin_unlock(&swap_lock); | |
2508 | /* | |
63d8620e | 2509 | * Finished initializing swap device, now it's safe to reference it. |
eb085574 | 2510 | */ |
63d8620e | 2511 | percpu_ref_resurrect(&p->users); |
eb085574 HY |
2512 | spin_lock(&swap_lock); |
2513 | spin_lock(&p->lock); | |
2514 | _enable_swap_info(p); | |
ec8acf20 | 2515 | spin_unlock(&p->lock); |
cf0cac0a CEB |
2516 | spin_unlock(&swap_lock); |
2517 | } | |
2518 | ||
2519 | static void reinsert_swap_info(struct swap_info_struct *p) | |
2520 | { | |
2521 | spin_lock(&swap_lock); | |
ec8acf20 | 2522 | spin_lock(&p->lock); |
eb085574 HY |
2523 | setup_swap_info(p, p->prio, p->swap_map, p->cluster_info); |
2524 | _enable_swap_info(p); | |
ec8acf20 | 2525 | spin_unlock(&p->lock); |
40531542 CEB |
2526 | spin_unlock(&swap_lock); |
2527 | } | |
2528 | ||
67afa38e TC |
2529 | bool has_usable_swap(void) |
2530 | { | |
2531 | bool ret = true; | |
2532 | ||
2533 | spin_lock(&swap_lock); | |
2534 | if (plist_head_empty(&swap_active_head)) | |
2535 | ret = false; | |
2536 | spin_unlock(&swap_lock); | |
2537 | return ret; | |
2538 | } | |
2539 | ||
c4ea37c2 | 2540 | SYSCALL_DEFINE1(swapoff, const char __user *, specialfile) |
1da177e4 | 2541 | { |
73c34b6a | 2542 | struct swap_info_struct *p = NULL; |
8d69aaee | 2543 | unsigned char *swap_map; |
2a8f9449 | 2544 | struct swap_cluster_info *cluster_info; |
4f89849d | 2545 | unsigned long *frontswap_map; |
1da177e4 LT |
2546 | struct file *swap_file, *victim; |
2547 | struct address_space *mapping; | |
2548 | struct inode *inode; | |
91a27b2a | 2549 | struct filename *pathname; |
adfab836 | 2550 | int err, found = 0; |
5b808a23 | 2551 | unsigned int old_block_size; |
886bb7e9 | 2552 | |
1da177e4 LT |
2553 | if (!capable(CAP_SYS_ADMIN)) |
2554 | return -EPERM; | |
2555 | ||
191c5424 AV |
2556 | BUG_ON(!current->mm); |
2557 | ||
1da177e4 | 2558 | pathname = getname(specialfile); |
1da177e4 | 2559 | if (IS_ERR(pathname)) |
f58b59c1 | 2560 | return PTR_ERR(pathname); |
1da177e4 | 2561 | |
669abf4e | 2562 | victim = file_open_name(pathname, O_RDWR|O_LARGEFILE, 0); |
1da177e4 LT |
2563 | err = PTR_ERR(victim); |
2564 | if (IS_ERR(victim)) | |
2565 | goto out; | |
2566 | ||
2567 | mapping = victim->f_mapping; | |
5d337b91 | 2568 | spin_lock(&swap_lock); |
18ab4d4c | 2569 | plist_for_each_entry(p, &swap_active_head, list) { |
22c6f8fd | 2570 | if (p->flags & SWP_WRITEOK) { |
adfab836 DS |
2571 | if (p->swap_file->f_mapping == mapping) { |
2572 | found = 1; | |
1da177e4 | 2573 | break; |
adfab836 | 2574 | } |
1da177e4 | 2575 | } |
1da177e4 | 2576 | } |
adfab836 | 2577 | if (!found) { |
1da177e4 | 2578 | err = -EINVAL; |
5d337b91 | 2579 | spin_unlock(&swap_lock); |
1da177e4 LT |
2580 | goto out_dput; |
2581 | } | |
191c5424 | 2582 | if (!security_vm_enough_memory_mm(current->mm, p->pages)) |
1da177e4 LT |
2583 | vm_unacct_memory(p->pages); |
2584 | else { | |
2585 | err = -ENOMEM; | |
5d337b91 | 2586 | spin_unlock(&swap_lock); |
1da177e4 LT |
2587 | goto out_dput; |
2588 | } | |
a2468cc9 | 2589 | del_from_avail_list(p); |
ec8acf20 | 2590 | spin_lock(&p->lock); |
78ecba08 | 2591 | if (p->prio < 0) { |
adfab836 | 2592 | struct swap_info_struct *si = p; |
a2468cc9 | 2593 | int nid; |
adfab836 | 2594 | |
18ab4d4c | 2595 | plist_for_each_entry_continue(si, &swap_active_head, list) { |
adfab836 | 2596 | si->prio++; |
18ab4d4c | 2597 | si->list.prio--; |
a2468cc9 AL |
2598 | for_each_node(nid) { |
2599 | if (si->avail_lists[nid].prio != 1) | |
2600 | si->avail_lists[nid].prio--; | |
2601 | } | |
adfab836 | 2602 | } |
78ecba08 HD |
2603 | least_priority++; |
2604 | } | |
18ab4d4c | 2605 | plist_del(&p->list, &swap_active_head); |
ec8acf20 | 2606 | atomic_long_sub(p->pages, &nr_swap_pages); |
1da177e4 LT |
2607 | total_swap_pages -= p->pages; |
2608 | p->flags &= ~SWP_WRITEOK; | |
ec8acf20 | 2609 | spin_unlock(&p->lock); |
5d337b91 | 2610 | spin_unlock(&swap_lock); |
fb4f88dc | 2611 | |
039939a6 TC |
2612 | disable_swap_slots_cache_lock(); |
2613 | ||
e1e12d2f | 2614 | set_current_oom_origin(); |
adfab836 | 2615 | err = try_to_unuse(p->type, false, 0); /* force unuse all pages */ |
e1e12d2f | 2616 | clear_current_oom_origin(); |
1da177e4 | 2617 | |
1da177e4 LT |
2618 | if (err) { |
2619 | /* re-insert swap space back into swap_list */ | |
cf0cac0a | 2620 | reinsert_swap_info(p); |
039939a6 | 2621 | reenable_swap_slots_cache_unlock(); |
1da177e4 LT |
2622 | goto out_dput; |
2623 | } | |
52b7efdb | 2624 | |
039939a6 TC |
2625 | reenable_swap_slots_cache_unlock(); |
2626 | ||
eb085574 | 2627 | /* |
63d8620e ML |
2628 | * Wait for swap operations protected by get/put_swap_device() |
2629 | * to complete. | |
2630 | * | |
2631 | * We need synchronize_rcu() here to protect the accessing to | |
2632 | * the swap cache data structure. | |
eb085574 | 2633 | */ |
63d8620e | 2634 | percpu_ref_kill(&p->users); |
eb085574 | 2635 | synchronize_rcu(); |
63d8620e | 2636 | wait_for_completion(&p->comp); |
eb085574 | 2637 | |
815c2c54 SL |
2638 | flush_work(&p->discard_work); |
2639 | ||
5d337b91 | 2640 | destroy_swap_extents(p); |
570a335b HD |
2641 | if (p->flags & SWP_CONTINUED) |
2642 | free_swap_count_continuations(p); | |
2643 | ||
81a0298b HY |
2644 | if (!p->bdev || !blk_queue_nonrot(bdev_get_queue(p->bdev))) |
2645 | atomic_dec(&nr_rotate_swap); | |
2646 | ||
fc0abb14 | 2647 | mutex_lock(&swapon_mutex); |
5d337b91 | 2648 | spin_lock(&swap_lock); |
ec8acf20 | 2649 | spin_lock(&p->lock); |
5d337b91 HD |
2650 | drain_mmlist(); |
2651 | ||
52b7efdb | 2652 | /* wait for anyone still in scan_swap_map */ |
52b7efdb HD |
2653 | p->highest_bit = 0; /* cuts scans short */ |
2654 | while (p->flags >= SWP_SCANNING) { | |
ec8acf20 | 2655 | spin_unlock(&p->lock); |
5d337b91 | 2656 | spin_unlock(&swap_lock); |
13e4b57f | 2657 | schedule_timeout_uninterruptible(1); |
5d337b91 | 2658 | spin_lock(&swap_lock); |
ec8acf20 | 2659 | spin_lock(&p->lock); |
52b7efdb | 2660 | } |
52b7efdb | 2661 | |
1da177e4 | 2662 | swap_file = p->swap_file; |
5b808a23 | 2663 | old_block_size = p->old_block_size; |
1da177e4 LT |
2664 | p->swap_file = NULL; |
2665 | p->max = 0; | |
2666 | swap_map = p->swap_map; | |
2667 | p->swap_map = NULL; | |
2a8f9449 SL |
2668 | cluster_info = p->cluster_info; |
2669 | p->cluster_info = NULL; | |
4f89849d | 2670 | frontswap_map = frontswap_map_get(p); |
ec8acf20 | 2671 | spin_unlock(&p->lock); |
5d337b91 | 2672 | spin_unlock(&swap_lock); |
8a84802e | 2673 | arch_swap_invalidate_area(p->type); |
adfab836 | 2674 | frontswap_invalidate_area(p->type); |
58e97ba6 | 2675 | frontswap_map_set(p, NULL); |
fc0abb14 | 2676 | mutex_unlock(&swapon_mutex); |
ebc2a1a6 SL |
2677 | free_percpu(p->percpu_cluster); |
2678 | p->percpu_cluster = NULL; | |
49070588 HY |
2679 | free_percpu(p->cluster_next_cpu); |
2680 | p->cluster_next_cpu = NULL; | |
1da177e4 | 2681 | vfree(swap_map); |
54f180d3 HY |
2682 | kvfree(cluster_info); |
2683 | kvfree(frontswap_map); | |
2de1a7e4 | 2684 | /* Destroy swap account information */ |
adfab836 | 2685 | swap_cgroup_swapoff(p->type); |
4b3ef9da | 2686 | exit_swap_address_space(p->type); |
27a7faa0 | 2687 | |
1da177e4 LT |
2688 | inode = mapping->host; |
2689 | if (S_ISBLK(inode->i_mode)) { | |
2690 | struct block_device *bdev = I_BDEV(inode); | |
1638045c | 2691 | |
5b808a23 | 2692 | set_blocksize(bdev, old_block_size); |
e525fd89 | 2693 | blkdev_put(bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL); |
1da177e4 | 2694 | } |
1638045c DW |
2695 | |
2696 | inode_lock(inode); | |
2697 | inode->i_flags &= ~S_SWAPFILE; | |
2698 | inode_unlock(inode); | |
1da177e4 | 2699 | filp_close(swap_file, NULL); |
f893ab41 WY |
2700 | |
2701 | /* | |
2702 | * Clear the SWP_USED flag after all resources are freed so that swapon | |
2703 | * can reuse this swap_info in alloc_swap_info() safely. It is ok to | |
2704 | * not hold p->lock after we cleared its SWP_WRITEOK. | |
2705 | */ | |
2706 | spin_lock(&swap_lock); | |
2707 | p->flags = 0; | |
2708 | spin_unlock(&swap_lock); | |
2709 | ||
1da177e4 | 2710 | err = 0; |
66d7dd51 KS |
2711 | atomic_inc(&proc_poll_event); |
2712 | wake_up_interruptible(&proc_poll_wait); | |
1da177e4 LT |
2713 | |
2714 | out_dput: | |
2715 | filp_close(victim, NULL); | |
2716 | out: | |
f58b59c1 | 2717 | putname(pathname); |
1da177e4 LT |
2718 | return err; |
2719 | } | |
2720 | ||
2721 | #ifdef CONFIG_PROC_FS | |
9dd95748 | 2722 | static __poll_t swaps_poll(struct file *file, poll_table *wait) |
66d7dd51 | 2723 | { |
f1514638 | 2724 | struct seq_file *seq = file->private_data; |
66d7dd51 KS |
2725 | |
2726 | poll_wait(file, &proc_poll_wait, wait); | |
2727 | ||
f1514638 KS |
2728 | if (seq->poll_event != atomic_read(&proc_poll_event)) { |
2729 | seq->poll_event = atomic_read(&proc_poll_event); | |
a9a08845 | 2730 | return EPOLLIN | EPOLLRDNORM | EPOLLERR | EPOLLPRI; |
66d7dd51 KS |
2731 | } |
2732 | ||
a9a08845 | 2733 | return EPOLLIN | EPOLLRDNORM; |
66d7dd51 KS |
2734 | } |
2735 | ||
1da177e4 LT |
2736 | /* iterator */ |
2737 | static void *swap_start(struct seq_file *swap, loff_t *pos) | |
2738 | { | |
efa90a98 HD |
2739 | struct swap_info_struct *si; |
2740 | int type; | |
1da177e4 LT |
2741 | loff_t l = *pos; |
2742 | ||
fc0abb14 | 2743 | mutex_lock(&swapon_mutex); |
1da177e4 | 2744 | |
881e4aab SS |
2745 | if (!l) |
2746 | return SEQ_START_TOKEN; | |
2747 | ||
c10d38cc | 2748 | for (type = 0; (si = swap_type_to_swap_info(type)); type++) { |
efa90a98 | 2749 | if (!(si->flags & SWP_USED) || !si->swap_map) |
1da177e4 | 2750 | continue; |
881e4aab | 2751 | if (!--l) |
efa90a98 | 2752 | return si; |
1da177e4 LT |
2753 | } |
2754 | ||
2755 | return NULL; | |
2756 | } | |
2757 | ||
2758 | static void *swap_next(struct seq_file *swap, void *v, loff_t *pos) | |
2759 | { | |
efa90a98 HD |
2760 | struct swap_info_struct *si = v; |
2761 | int type; | |
1da177e4 | 2762 | |
881e4aab | 2763 | if (v == SEQ_START_TOKEN) |
efa90a98 HD |
2764 | type = 0; |
2765 | else | |
2766 | type = si->type + 1; | |
881e4aab | 2767 | |
10c8d69f | 2768 | ++(*pos); |
c10d38cc | 2769 | for (; (si = swap_type_to_swap_info(type)); type++) { |
efa90a98 | 2770 | if (!(si->flags & SWP_USED) || !si->swap_map) |
1da177e4 | 2771 | continue; |
efa90a98 | 2772 | return si; |
1da177e4 LT |
2773 | } |
2774 | ||
2775 | return NULL; | |
2776 | } | |
2777 | ||
2778 | static void swap_stop(struct seq_file *swap, void *v) | |
2779 | { | |
fc0abb14 | 2780 | mutex_unlock(&swapon_mutex); |
1da177e4 LT |
2781 | } |
2782 | ||
2783 | static int swap_show(struct seq_file *swap, void *v) | |
2784 | { | |
efa90a98 | 2785 | struct swap_info_struct *si = v; |
1da177e4 LT |
2786 | struct file *file; |
2787 | int len; | |
6f793940 | 2788 | unsigned int bytes, inuse; |
1da177e4 | 2789 | |
efa90a98 | 2790 | if (si == SEQ_START_TOKEN) { |
68d68ff6 | 2791 | seq_puts(swap, "Filename\t\t\t\tType\t\tSize\t\tUsed\t\tPriority\n"); |
881e4aab SS |
2792 | return 0; |
2793 | } | |
1da177e4 | 2794 | |
6f793940 RD |
2795 | bytes = si->pages << (PAGE_SHIFT - 10); |
2796 | inuse = si->inuse_pages << (PAGE_SHIFT - 10); | |
2797 | ||
efa90a98 | 2798 | file = si->swap_file; |
2726d566 | 2799 | len = seq_file_path(swap, file, " \t\n\\"); |
6f793940 | 2800 | seq_printf(swap, "%*s%s\t%u\t%s%u\t%s%d\n", |
886bb7e9 | 2801 | len < 40 ? 40 - len : 1, " ", |
496ad9aa | 2802 | S_ISBLK(file_inode(file)->i_mode) ? |
1da177e4 | 2803 | "partition" : "file\t", |
6f793940 RD |
2804 | bytes, bytes < 10000000 ? "\t" : "", |
2805 | inuse, inuse < 10000000 ? "\t" : "", | |
efa90a98 | 2806 | si->prio); |
1da177e4 LT |
2807 | return 0; |
2808 | } | |
2809 | ||
15ad7cdc | 2810 | static const struct seq_operations swaps_op = { |
1da177e4 LT |
2811 | .start = swap_start, |
2812 | .next = swap_next, | |
2813 | .stop = swap_stop, | |
2814 | .show = swap_show | |
2815 | }; | |
2816 | ||
2817 | static int swaps_open(struct inode *inode, struct file *file) | |
2818 | { | |
f1514638 | 2819 | struct seq_file *seq; |
66d7dd51 KS |
2820 | int ret; |
2821 | ||
66d7dd51 | 2822 | ret = seq_open(file, &swaps_op); |
f1514638 | 2823 | if (ret) |
66d7dd51 | 2824 | return ret; |
66d7dd51 | 2825 | |
f1514638 KS |
2826 | seq = file->private_data; |
2827 | seq->poll_event = atomic_read(&proc_poll_event); | |
2828 | return 0; | |
1da177e4 LT |
2829 | } |
2830 | ||
97a32539 | 2831 | static const struct proc_ops swaps_proc_ops = { |
d919b33d | 2832 | .proc_flags = PROC_ENTRY_PERMANENT, |
97a32539 AD |
2833 | .proc_open = swaps_open, |
2834 | .proc_read = seq_read, | |
2835 | .proc_lseek = seq_lseek, | |
2836 | .proc_release = seq_release, | |
2837 | .proc_poll = swaps_poll, | |
1da177e4 LT |
2838 | }; |
2839 | ||
2840 | static int __init procswaps_init(void) | |
2841 | { | |
97a32539 | 2842 | proc_create("swaps", 0, NULL, &swaps_proc_ops); |
1da177e4 LT |
2843 | return 0; |
2844 | } | |
2845 | __initcall(procswaps_init); | |
2846 | #endif /* CONFIG_PROC_FS */ | |
2847 | ||
1796316a JB |
2848 | #ifdef MAX_SWAPFILES_CHECK |
2849 | static int __init max_swapfiles_check(void) | |
2850 | { | |
2851 | MAX_SWAPFILES_CHECK(); | |
2852 | return 0; | |
2853 | } | |
2854 | late_initcall(max_swapfiles_check); | |
2855 | #endif | |
2856 | ||
53cbb243 | 2857 | static struct swap_info_struct *alloc_swap_info(void) |
1da177e4 | 2858 | { |
73c34b6a | 2859 | struct swap_info_struct *p; |
b11a76b3 | 2860 | struct swap_info_struct *defer = NULL; |
1da177e4 | 2861 | unsigned int type; |
a2468cc9 | 2862 | int i; |
efa90a98 | 2863 | |
96008744 | 2864 | p = kvzalloc(struct_size(p, avail_lists, nr_node_ids), GFP_KERNEL); |
efa90a98 | 2865 | if (!p) |
53cbb243 | 2866 | return ERR_PTR(-ENOMEM); |
efa90a98 | 2867 | |
63d8620e ML |
2868 | if (percpu_ref_init(&p->users, swap_users_ref_free, |
2869 | PERCPU_REF_INIT_DEAD, GFP_KERNEL)) { | |
2870 | kvfree(p); | |
2871 | return ERR_PTR(-ENOMEM); | |
2872 | } | |
2873 | ||
5d337b91 | 2874 | spin_lock(&swap_lock); |
efa90a98 HD |
2875 | for (type = 0; type < nr_swapfiles; type++) { |
2876 | if (!(swap_info[type]->flags & SWP_USED)) | |
1da177e4 | 2877 | break; |
efa90a98 | 2878 | } |
0697212a | 2879 | if (type >= MAX_SWAPFILES) { |
5d337b91 | 2880 | spin_unlock(&swap_lock); |
63d8620e | 2881 | percpu_ref_exit(&p->users); |
873d7bcf | 2882 | kvfree(p); |
730c0581 | 2883 | return ERR_PTR(-EPERM); |
1da177e4 | 2884 | } |
efa90a98 HD |
2885 | if (type >= nr_swapfiles) { |
2886 | p->type = type; | |
c10d38cc | 2887 | WRITE_ONCE(swap_info[type], p); |
efa90a98 HD |
2888 | /* |
2889 | * Write swap_info[type] before nr_swapfiles, in case a | |
2890 | * racing procfs swap_start() or swap_next() is reading them. | |
2891 | * (We never shrink nr_swapfiles, we never free this entry.) | |
2892 | */ | |
2893 | smp_wmb(); | |
c10d38cc | 2894 | WRITE_ONCE(nr_swapfiles, nr_swapfiles + 1); |
efa90a98 | 2895 | } else { |
b11a76b3 | 2896 | defer = p; |
efa90a98 HD |
2897 | p = swap_info[type]; |
2898 | /* | |
2899 | * Do not memset this entry: a racing procfs swap_next() | |
2900 | * would be relying on p->type to remain valid. | |
2901 | */ | |
2902 | } | |
4efaceb1 | 2903 | p->swap_extent_root = RB_ROOT; |
18ab4d4c | 2904 | plist_node_init(&p->list, 0); |
a2468cc9 AL |
2905 | for_each_node(i) |
2906 | plist_node_init(&p->avail_lists[i], 0); | |
1da177e4 | 2907 | p->flags = SWP_USED; |
5d337b91 | 2908 | spin_unlock(&swap_lock); |
63d8620e ML |
2909 | if (defer) { |
2910 | percpu_ref_exit(&defer->users); | |
2911 | kvfree(defer); | |
2912 | } | |
ec8acf20 | 2913 | spin_lock_init(&p->lock); |
2628bd6f | 2914 | spin_lock_init(&p->cont_lock); |
63d8620e | 2915 | init_completion(&p->comp); |
efa90a98 | 2916 | |
53cbb243 | 2917 | return p; |
53cbb243 CEB |
2918 | } |
2919 | ||
4d0e1e10 CEB |
2920 | static int claim_swapfile(struct swap_info_struct *p, struct inode *inode) |
2921 | { | |
2922 | int error; | |
2923 | ||
2924 | if (S_ISBLK(inode->i_mode)) { | |
ef16e1d9 | 2925 | p->bdev = blkdev_get_by_dev(inode->i_rdev, |
6f179af8 | 2926 | FMODE_READ | FMODE_WRITE | FMODE_EXCL, p); |
ef16e1d9 CH |
2927 | if (IS_ERR(p->bdev)) { |
2928 | error = PTR_ERR(p->bdev); | |
4d0e1e10 | 2929 | p->bdev = NULL; |
6f179af8 | 2930 | return error; |
4d0e1e10 CEB |
2931 | } |
2932 | p->old_block_size = block_size(p->bdev); | |
2933 | error = set_blocksize(p->bdev, PAGE_SIZE); | |
2934 | if (error < 0) | |
87ade72a | 2935 | return error; |
12d2966d NA |
2936 | /* |
2937 | * Zoned block devices contain zones that have a sequential | |
2938 | * write only restriction. Hence zoned block devices are not | |
2939 | * suitable for swapping. Disallow them here. | |
2940 | */ | |
e556f6ba | 2941 | if (blk_queue_is_zoned(p->bdev->bd_disk->queue)) |
12d2966d | 2942 | return -EINVAL; |
4d0e1e10 CEB |
2943 | p->flags |= SWP_BLKDEV; |
2944 | } else if (S_ISREG(inode->i_mode)) { | |
2945 | p->bdev = inode->i_sb->s_bdev; | |
1638045c DW |
2946 | } |
2947 | ||
4d0e1e10 | 2948 | return 0; |
4d0e1e10 CEB |
2949 | } |
2950 | ||
377eeaa8 AK |
2951 | |
2952 | /* | |
2953 | * Find out how many pages are allowed for a single swap device. There | |
2954 | * are two limiting factors: | |
2955 | * 1) the number of bits for the swap offset in the swp_entry_t type, and | |
2956 | * 2) the number of bits in the swap pte, as defined by the different | |
2957 | * architectures. | |
2958 | * | |
2959 | * In order to find the largest possible bit mask, a swap entry with | |
2960 | * swap type 0 and swap offset ~0UL is created, encoded to a swap pte, | |
2961 | * decoded to a swp_entry_t again, and finally the swap offset is | |
2962 | * extracted. | |
2963 | * | |
2964 | * This will mask all the bits from the initial ~0UL mask that can't | |
2965 | * be encoded in either the swp_entry_t or the architecture definition | |
2966 | * of a swap pte. | |
2967 | */ | |
2968 | unsigned long generic_max_swapfile_size(void) | |
2969 | { | |
2970 | return swp_offset(pte_to_swp_entry( | |
2971 | swp_entry_to_pte(swp_entry(0, ~0UL)))) + 1; | |
2972 | } | |
2973 | ||
2974 | /* Can be overridden by an architecture for additional checks. */ | |
2975 | __weak unsigned long max_swapfile_size(void) | |
2976 | { | |
2977 | return generic_max_swapfile_size(); | |
2978 | } | |
2979 | ||
ca8bd38b CEB |
2980 | static unsigned long read_swap_header(struct swap_info_struct *p, |
2981 | union swap_header *swap_header, | |
2982 | struct inode *inode) | |
2983 | { | |
2984 | int i; | |
2985 | unsigned long maxpages; | |
2986 | unsigned long swapfilepages; | |
d6bbbd29 | 2987 | unsigned long last_page; |
ca8bd38b CEB |
2988 | |
2989 | if (memcmp("SWAPSPACE2", swap_header->magic.magic, 10)) { | |
465c47fd | 2990 | pr_err("Unable to find swap-space signature\n"); |
38719025 | 2991 | return 0; |
ca8bd38b CEB |
2992 | } |
2993 | ||
2994 | /* swap partition endianess hack... */ | |
2995 | if (swab32(swap_header->info.version) == 1) { | |
2996 | swab32s(&swap_header->info.version); | |
2997 | swab32s(&swap_header->info.last_page); | |
2998 | swab32s(&swap_header->info.nr_badpages); | |
dd111be6 JH |
2999 | if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES) |
3000 | return 0; | |
ca8bd38b CEB |
3001 | for (i = 0; i < swap_header->info.nr_badpages; i++) |
3002 | swab32s(&swap_header->info.badpages[i]); | |
3003 | } | |
3004 | /* Check the swap header's sub-version */ | |
3005 | if (swap_header->info.version != 1) { | |
465c47fd AM |
3006 | pr_warn("Unable to handle swap header version %d\n", |
3007 | swap_header->info.version); | |
38719025 | 3008 | return 0; |
ca8bd38b CEB |
3009 | } |
3010 | ||
3011 | p->lowest_bit = 1; | |
3012 | p->cluster_next = 1; | |
3013 | p->cluster_nr = 0; | |
3014 | ||
377eeaa8 | 3015 | maxpages = max_swapfile_size(); |
d6bbbd29 | 3016 | last_page = swap_header->info.last_page; |
a06ad633 TA |
3017 | if (!last_page) { |
3018 | pr_warn("Empty swap-file\n"); | |
3019 | return 0; | |
3020 | } | |
d6bbbd29 | 3021 | if (last_page > maxpages) { |
465c47fd | 3022 | pr_warn("Truncating oversized swap area, only using %luk out of %luk\n", |
d6bbbd29 RJ |
3023 | maxpages << (PAGE_SHIFT - 10), |
3024 | last_page << (PAGE_SHIFT - 10)); | |
3025 | } | |
3026 | if (maxpages > last_page) { | |
3027 | maxpages = last_page + 1; | |
ca8bd38b CEB |
3028 | /* p->max is an unsigned int: don't overflow it */ |
3029 | if ((unsigned int)maxpages == 0) | |
3030 | maxpages = UINT_MAX; | |
3031 | } | |
3032 | p->highest_bit = maxpages - 1; | |
3033 | ||
3034 | if (!maxpages) | |
38719025 | 3035 | return 0; |
ca8bd38b CEB |
3036 | swapfilepages = i_size_read(inode) >> PAGE_SHIFT; |
3037 | if (swapfilepages && maxpages > swapfilepages) { | |
465c47fd | 3038 | pr_warn("Swap area shorter than signature indicates\n"); |
38719025 | 3039 | return 0; |
ca8bd38b CEB |
3040 | } |
3041 | if (swap_header->info.nr_badpages && S_ISREG(inode->i_mode)) | |
38719025 | 3042 | return 0; |
ca8bd38b | 3043 | if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES) |
38719025 | 3044 | return 0; |
ca8bd38b CEB |
3045 | |
3046 | return maxpages; | |
ca8bd38b CEB |
3047 | } |
3048 | ||
4b3ef9da | 3049 | #define SWAP_CLUSTER_INFO_COLS \ |
235b6217 | 3050 | DIV_ROUND_UP(L1_CACHE_BYTES, sizeof(struct swap_cluster_info)) |
4b3ef9da HY |
3051 | #define SWAP_CLUSTER_SPACE_COLS \ |
3052 | DIV_ROUND_UP(SWAP_ADDRESS_SPACE_PAGES, SWAPFILE_CLUSTER) | |
3053 | #define SWAP_CLUSTER_COLS \ | |
3054 | max_t(unsigned int, SWAP_CLUSTER_INFO_COLS, SWAP_CLUSTER_SPACE_COLS) | |
235b6217 | 3055 | |
915d4d7b CEB |
3056 | static int setup_swap_map_and_extents(struct swap_info_struct *p, |
3057 | union swap_header *swap_header, | |
3058 | unsigned char *swap_map, | |
2a8f9449 | 3059 | struct swap_cluster_info *cluster_info, |
915d4d7b CEB |
3060 | unsigned long maxpages, |
3061 | sector_t *span) | |
3062 | { | |
235b6217 | 3063 | unsigned int j, k; |
915d4d7b CEB |
3064 | unsigned int nr_good_pages; |
3065 | int nr_extents; | |
2a8f9449 | 3066 | unsigned long nr_clusters = DIV_ROUND_UP(maxpages, SWAPFILE_CLUSTER); |
235b6217 HY |
3067 | unsigned long col = p->cluster_next / SWAPFILE_CLUSTER % SWAP_CLUSTER_COLS; |
3068 | unsigned long i, idx; | |
915d4d7b CEB |
3069 | |
3070 | nr_good_pages = maxpages - 1; /* omit header page */ | |
3071 | ||
6b534915 HY |
3072 | cluster_list_init(&p->free_clusters); |
3073 | cluster_list_init(&p->discard_clusters); | |
2a8f9449 | 3074 | |
915d4d7b CEB |
3075 | for (i = 0; i < swap_header->info.nr_badpages; i++) { |
3076 | unsigned int page_nr = swap_header->info.badpages[i]; | |
bdb8e3f6 CEB |
3077 | if (page_nr == 0 || page_nr > swap_header->info.last_page) |
3078 | return -EINVAL; | |
915d4d7b CEB |
3079 | if (page_nr < maxpages) { |
3080 | swap_map[page_nr] = SWAP_MAP_BAD; | |
3081 | nr_good_pages--; | |
2a8f9449 SL |
3082 | /* |
3083 | * Haven't marked the cluster free yet, no list | |
3084 | * operation involved | |
3085 | */ | |
3086 | inc_cluster_info_page(p, cluster_info, page_nr); | |
915d4d7b CEB |
3087 | } |
3088 | } | |
3089 | ||
2a8f9449 SL |
3090 | /* Haven't marked the cluster free yet, no list operation involved */ |
3091 | for (i = maxpages; i < round_up(maxpages, SWAPFILE_CLUSTER); i++) | |
3092 | inc_cluster_info_page(p, cluster_info, i); | |
3093 | ||
915d4d7b CEB |
3094 | if (nr_good_pages) { |
3095 | swap_map[0] = SWAP_MAP_BAD; | |
2a8f9449 SL |
3096 | /* |
3097 | * Not mark the cluster free yet, no list | |
3098 | * operation involved | |
3099 | */ | |
3100 | inc_cluster_info_page(p, cluster_info, 0); | |
915d4d7b CEB |
3101 | p->max = maxpages; |
3102 | p->pages = nr_good_pages; | |
3103 | nr_extents = setup_swap_extents(p, span); | |
bdb8e3f6 CEB |
3104 | if (nr_extents < 0) |
3105 | return nr_extents; | |
915d4d7b CEB |
3106 | nr_good_pages = p->pages; |
3107 | } | |
3108 | if (!nr_good_pages) { | |
465c47fd | 3109 | pr_warn("Empty swap-file\n"); |
bdb8e3f6 | 3110 | return -EINVAL; |
915d4d7b CEB |
3111 | } |
3112 | ||
2a8f9449 SL |
3113 | if (!cluster_info) |
3114 | return nr_extents; | |
3115 | ||
235b6217 | 3116 | |
4b3ef9da HY |
3117 | /* |
3118 | * Reduce false cache line sharing between cluster_info and | |
3119 | * sharing same address space. | |
3120 | */ | |
235b6217 HY |
3121 | for (k = 0; k < SWAP_CLUSTER_COLS; k++) { |
3122 | j = (k + col) % SWAP_CLUSTER_COLS; | |
3123 | for (i = 0; i < DIV_ROUND_UP(nr_clusters, SWAP_CLUSTER_COLS); i++) { | |
3124 | idx = i * SWAP_CLUSTER_COLS + j; | |
3125 | if (idx >= nr_clusters) | |
3126 | continue; | |
3127 | if (cluster_count(&cluster_info[idx])) | |
3128 | continue; | |
2a8f9449 | 3129 | cluster_set_flag(&cluster_info[idx], CLUSTER_FLAG_FREE); |
6b534915 HY |
3130 | cluster_list_add_tail(&p->free_clusters, cluster_info, |
3131 | idx); | |
2a8f9449 | 3132 | } |
2a8f9449 | 3133 | } |
915d4d7b | 3134 | return nr_extents; |
915d4d7b CEB |
3135 | } |
3136 | ||
dcf6b7dd RA |
3137 | /* |
3138 | * Helper to sys_swapon determining if a given swap | |
3139 | * backing device queue supports DISCARD operations. | |
3140 | */ | |
3141 | static bool swap_discardable(struct swap_info_struct *si) | |
3142 | { | |
3143 | struct request_queue *q = bdev_get_queue(si->bdev); | |
3144 | ||
3145 | if (!q || !blk_queue_discard(q)) | |
3146 | return false; | |
3147 | ||
3148 | return true; | |
3149 | } | |
3150 | ||
53cbb243 CEB |
3151 | SYSCALL_DEFINE2(swapon, const char __user *, specialfile, int, swap_flags) |
3152 | { | |
3153 | struct swap_info_struct *p; | |
91a27b2a | 3154 | struct filename *name; |
53cbb243 CEB |
3155 | struct file *swap_file = NULL; |
3156 | struct address_space *mapping; | |
40531542 | 3157 | int prio; |
53cbb243 CEB |
3158 | int error; |
3159 | union swap_header *swap_header; | |
915d4d7b | 3160 | int nr_extents; |
53cbb243 CEB |
3161 | sector_t span; |
3162 | unsigned long maxpages; | |
53cbb243 | 3163 | unsigned char *swap_map = NULL; |
2a8f9449 | 3164 | struct swap_cluster_info *cluster_info = NULL; |
38b5faf4 | 3165 | unsigned long *frontswap_map = NULL; |
53cbb243 CEB |
3166 | struct page *page = NULL; |
3167 | struct inode *inode = NULL; | |
7cbf3192 | 3168 | bool inced_nr_rotate_swap = false; |
53cbb243 | 3169 | |
d15cab97 HD |
3170 | if (swap_flags & ~SWAP_FLAGS_VALID) |
3171 | return -EINVAL; | |
3172 | ||
53cbb243 CEB |
3173 | if (!capable(CAP_SYS_ADMIN)) |
3174 | return -EPERM; | |
3175 | ||
a2468cc9 AL |
3176 | if (!swap_avail_heads) |
3177 | return -ENOMEM; | |
3178 | ||
53cbb243 | 3179 | p = alloc_swap_info(); |
2542e513 CEB |
3180 | if (IS_ERR(p)) |
3181 | return PTR_ERR(p); | |
53cbb243 | 3182 | |
815c2c54 SL |
3183 | INIT_WORK(&p->discard_work, swap_discard_work); |
3184 | ||
1da177e4 | 3185 | name = getname(specialfile); |
1da177e4 | 3186 | if (IS_ERR(name)) { |
7de7fb6b | 3187 | error = PTR_ERR(name); |
1da177e4 | 3188 | name = NULL; |
bd69010b | 3189 | goto bad_swap; |
1da177e4 | 3190 | } |
669abf4e | 3191 | swap_file = file_open_name(name, O_RDWR|O_LARGEFILE, 0); |
1da177e4 | 3192 | if (IS_ERR(swap_file)) { |
7de7fb6b | 3193 | error = PTR_ERR(swap_file); |
1da177e4 | 3194 | swap_file = NULL; |
bd69010b | 3195 | goto bad_swap; |
1da177e4 LT |
3196 | } |
3197 | ||
3198 | p->swap_file = swap_file; | |
3199 | mapping = swap_file->f_mapping; | |
2130781e | 3200 | inode = mapping->host; |
6f179af8 | 3201 | |
4d0e1e10 CEB |
3202 | error = claim_swapfile(p, inode); |
3203 | if (unlikely(error)) | |
1da177e4 | 3204 | goto bad_swap; |
1da177e4 | 3205 | |
d795a90e NA |
3206 | inode_lock(inode); |
3207 | if (IS_SWAPFILE(inode)) { | |
3208 | error = -EBUSY; | |
3209 | goto bad_swap_unlock_inode; | |
3210 | } | |
3211 | ||
1da177e4 LT |
3212 | /* |
3213 | * Read the swap header. | |
3214 | */ | |
3215 | if (!mapping->a_ops->readpage) { | |
3216 | error = -EINVAL; | |
d795a90e | 3217 | goto bad_swap_unlock_inode; |
1da177e4 | 3218 | } |
090d2b18 | 3219 | page = read_mapping_page(mapping, 0, swap_file); |
1da177e4 LT |
3220 | if (IS_ERR(page)) { |
3221 | error = PTR_ERR(page); | |
d795a90e | 3222 | goto bad_swap_unlock_inode; |
1da177e4 | 3223 | } |
81e33971 | 3224 | swap_header = kmap(page); |
1da177e4 | 3225 | |
ca8bd38b CEB |
3226 | maxpages = read_swap_header(p, swap_header, inode); |
3227 | if (unlikely(!maxpages)) { | |
1da177e4 | 3228 | error = -EINVAL; |
d795a90e | 3229 | goto bad_swap_unlock_inode; |
1da177e4 | 3230 | } |
886bb7e9 | 3231 | |
81e33971 | 3232 | /* OK, set up the swap map and apply the bad block list */ |
803d0c83 | 3233 | swap_map = vzalloc(maxpages); |
81e33971 HD |
3234 | if (!swap_map) { |
3235 | error = -ENOMEM; | |
d795a90e | 3236 | goto bad_swap_unlock_inode; |
81e33971 | 3237 | } |
f0571429 | 3238 | |
1cb039f3 | 3239 | if (p->bdev && blk_queue_stable_writes(p->bdev->bd_disk->queue)) |
f0571429 MK |
3240 | p->flags |= SWP_STABLE_WRITES; |
3241 | ||
a8b456d0 | 3242 | if (p->bdev && p->bdev->bd_disk->fops->rw_page) |
539a6fea MK |
3243 | p->flags |= SWP_SYNCHRONOUS_IO; |
3244 | ||
2a8f9449 | 3245 | if (p->bdev && blk_queue_nonrot(bdev_get_queue(p->bdev))) { |
6f179af8 | 3246 | int cpu; |
235b6217 | 3247 | unsigned long ci, nr_cluster; |
6f179af8 | 3248 | |
2a8f9449 | 3249 | p->flags |= SWP_SOLIDSTATE; |
49070588 HY |
3250 | p->cluster_next_cpu = alloc_percpu(unsigned int); |
3251 | if (!p->cluster_next_cpu) { | |
3252 | error = -ENOMEM; | |
3253 | goto bad_swap_unlock_inode; | |
3254 | } | |
2a8f9449 SL |
3255 | /* |
3256 | * select a random position to start with to help wear leveling | |
3257 | * SSD | |
3258 | */ | |
49070588 HY |
3259 | for_each_possible_cpu(cpu) { |
3260 | per_cpu(*p->cluster_next_cpu, cpu) = | |
3261 | 1 + prandom_u32_max(p->highest_bit); | |
3262 | } | |
235b6217 | 3263 | nr_cluster = DIV_ROUND_UP(maxpages, SWAPFILE_CLUSTER); |
2a8f9449 | 3264 | |
778e1cdd | 3265 | cluster_info = kvcalloc(nr_cluster, sizeof(*cluster_info), |
54f180d3 | 3266 | GFP_KERNEL); |
2a8f9449 SL |
3267 | if (!cluster_info) { |
3268 | error = -ENOMEM; | |
d795a90e | 3269 | goto bad_swap_unlock_inode; |
2a8f9449 | 3270 | } |
235b6217 HY |
3271 | |
3272 | for (ci = 0; ci < nr_cluster; ci++) | |
3273 | spin_lock_init(&((cluster_info + ci)->lock)); | |
3274 | ||
ebc2a1a6 SL |
3275 | p->percpu_cluster = alloc_percpu(struct percpu_cluster); |
3276 | if (!p->percpu_cluster) { | |
3277 | error = -ENOMEM; | |
d795a90e | 3278 | goto bad_swap_unlock_inode; |
ebc2a1a6 | 3279 | } |
6f179af8 | 3280 | for_each_possible_cpu(cpu) { |
ebc2a1a6 | 3281 | struct percpu_cluster *cluster; |
6f179af8 | 3282 | cluster = per_cpu_ptr(p->percpu_cluster, cpu); |
ebc2a1a6 SL |
3283 | cluster_set_null(&cluster->index); |
3284 | } | |
7cbf3192 | 3285 | } else { |
81a0298b | 3286 | atomic_inc(&nr_rotate_swap); |
7cbf3192 OS |
3287 | inced_nr_rotate_swap = true; |
3288 | } | |
1da177e4 | 3289 | |
1421ef3c CEB |
3290 | error = swap_cgroup_swapon(p->type, maxpages); |
3291 | if (error) | |
d795a90e | 3292 | goto bad_swap_unlock_inode; |
1421ef3c | 3293 | |
915d4d7b | 3294 | nr_extents = setup_swap_map_and_extents(p, swap_header, swap_map, |
2a8f9449 | 3295 | cluster_info, maxpages, &span); |
915d4d7b CEB |
3296 | if (unlikely(nr_extents < 0)) { |
3297 | error = nr_extents; | |
d795a90e | 3298 | goto bad_swap_unlock_inode; |
1da177e4 | 3299 | } |
38b5faf4 | 3300 | /* frontswap enabled? set up bit-per-page map for frontswap */ |
8ea1d2a1 | 3301 | if (IS_ENABLED(CONFIG_FRONTSWAP)) |
778e1cdd KC |
3302 | frontswap_map = kvcalloc(BITS_TO_LONGS(maxpages), |
3303 | sizeof(long), | |
54f180d3 | 3304 | GFP_KERNEL); |
1da177e4 | 3305 | |
68d68ff6 | 3306 | if (p->bdev && (swap_flags & SWAP_FLAG_DISCARD) && swap_discardable(p)) { |
2a8f9449 SL |
3307 | /* |
3308 | * When discard is enabled for swap with no particular | |
3309 | * policy flagged, we set all swap discard flags here in | |
3310 | * order to sustain backward compatibility with older | |
3311 | * swapon(8) releases. | |
3312 | */ | |
3313 | p->flags |= (SWP_DISCARDABLE | SWP_AREA_DISCARD | | |
3314 | SWP_PAGE_DISCARD); | |
dcf6b7dd | 3315 | |
2a8f9449 SL |
3316 | /* |
3317 | * By flagging sys_swapon, a sysadmin can tell us to | |
3318 | * either do single-time area discards only, or to just | |
3319 | * perform discards for released swap page-clusters. | |
3320 | * Now it's time to adjust the p->flags accordingly. | |
3321 | */ | |
3322 | if (swap_flags & SWAP_FLAG_DISCARD_ONCE) | |
3323 | p->flags &= ~SWP_PAGE_DISCARD; | |
3324 | else if (swap_flags & SWAP_FLAG_DISCARD_PAGES) | |
3325 | p->flags &= ~SWP_AREA_DISCARD; | |
3326 | ||
3327 | /* issue a swapon-time discard if it's still required */ | |
3328 | if (p->flags & SWP_AREA_DISCARD) { | |
3329 | int err = discard_swap(p); | |
3330 | if (unlikely(err)) | |
3331 | pr_err("swapon: discard_swap(%p): %d\n", | |
3332 | p, err); | |
dcf6b7dd | 3333 | } |
20137a49 | 3334 | } |
6a6ba831 | 3335 | |
4b3ef9da HY |
3336 | error = init_swap_address_space(p->type, maxpages); |
3337 | if (error) | |
d795a90e | 3338 | goto bad_swap_unlock_inode; |
4b3ef9da | 3339 | |
dc617f29 DW |
3340 | /* |
3341 | * Flush any pending IO and dirty mappings before we start using this | |
3342 | * swap device. | |
3343 | */ | |
3344 | inode->i_flags |= S_SWAPFILE; | |
3345 | error = inode_drain_writes(inode); | |
3346 | if (error) { | |
3347 | inode->i_flags &= ~S_SWAPFILE; | |
822bca52 | 3348 | goto free_swap_address_space; |
dc617f29 DW |
3349 | } |
3350 | ||
fc0abb14 | 3351 | mutex_lock(&swapon_mutex); |
40531542 | 3352 | prio = -1; |
78ecba08 | 3353 | if (swap_flags & SWAP_FLAG_PREFER) |
40531542 | 3354 | prio = |
78ecba08 | 3355 | (swap_flags & SWAP_FLAG_PRIO_MASK) >> SWAP_FLAG_PRIO_SHIFT; |
2a8f9449 | 3356 | enable_swap_info(p, prio, swap_map, cluster_info, frontswap_map); |
c69dbfb8 | 3357 | |
756a025f | 3358 | pr_info("Adding %uk swap on %s. Priority:%d extents:%d across:%lluk %s%s%s%s%s\n", |
91a27b2a | 3359 | p->pages<<(PAGE_SHIFT-10), name->name, p->prio, |
c69dbfb8 CEB |
3360 | nr_extents, (unsigned long long)span<<(PAGE_SHIFT-10), |
3361 | (p->flags & SWP_SOLIDSTATE) ? "SS" : "", | |
38b5faf4 | 3362 | (p->flags & SWP_DISCARDABLE) ? "D" : "", |
dcf6b7dd RA |
3363 | (p->flags & SWP_AREA_DISCARD) ? "s" : "", |
3364 | (p->flags & SWP_PAGE_DISCARD) ? "c" : "", | |
38b5faf4 | 3365 | (frontswap_map) ? "FS" : ""); |
c69dbfb8 | 3366 | |
fc0abb14 | 3367 | mutex_unlock(&swapon_mutex); |
66d7dd51 KS |
3368 | atomic_inc(&proc_poll_event); |
3369 | wake_up_interruptible(&proc_poll_wait); | |
3370 | ||
1da177e4 LT |
3371 | error = 0; |
3372 | goto out; | |
822bca52 ML |
3373 | free_swap_address_space: |
3374 | exit_swap_address_space(p->type); | |
d795a90e NA |
3375 | bad_swap_unlock_inode: |
3376 | inode_unlock(inode); | |
1da177e4 | 3377 | bad_swap: |
ebc2a1a6 SL |
3378 | free_percpu(p->percpu_cluster); |
3379 | p->percpu_cluster = NULL; | |
49070588 HY |
3380 | free_percpu(p->cluster_next_cpu); |
3381 | p->cluster_next_cpu = NULL; | |
bd69010b | 3382 | if (inode && S_ISBLK(inode->i_mode) && p->bdev) { |
f2090d2d CEB |
3383 | set_blocksize(p->bdev, p->old_block_size); |
3384 | blkdev_put(p->bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL); | |
1da177e4 | 3385 | } |
d795a90e | 3386 | inode = NULL; |
4cd3bb10 | 3387 | destroy_swap_extents(p); |
e8e6c2ec | 3388 | swap_cgroup_swapoff(p->type); |
5d337b91 | 3389 | spin_lock(&swap_lock); |
1da177e4 | 3390 | p->swap_file = NULL; |
1da177e4 | 3391 | p->flags = 0; |
5d337b91 | 3392 | spin_unlock(&swap_lock); |
1da177e4 | 3393 | vfree(swap_map); |
8606a1a9 | 3394 | kvfree(cluster_info); |
b6b1fd2a | 3395 | kvfree(frontswap_map); |
7cbf3192 OS |
3396 | if (inced_nr_rotate_swap) |
3397 | atomic_dec(&nr_rotate_swap); | |
d795a90e | 3398 | if (swap_file) |
1da177e4 LT |
3399 | filp_close(swap_file, NULL); |
3400 | out: | |
3401 | if (page && !IS_ERR(page)) { | |
3402 | kunmap(page); | |
09cbfeaf | 3403 | put_page(page); |
1da177e4 LT |
3404 | } |
3405 | if (name) | |
3406 | putname(name); | |
1638045c | 3407 | if (inode) |
5955102c | 3408 | inode_unlock(inode); |
039939a6 TC |
3409 | if (!error) |
3410 | enable_swap_slots_cache(); | |
1da177e4 LT |
3411 | return error; |
3412 | } | |
3413 | ||
3414 | void si_swapinfo(struct sysinfo *val) | |
3415 | { | |
efa90a98 | 3416 | unsigned int type; |
1da177e4 LT |
3417 | unsigned long nr_to_be_unused = 0; |
3418 | ||
5d337b91 | 3419 | spin_lock(&swap_lock); |
efa90a98 HD |
3420 | for (type = 0; type < nr_swapfiles; type++) { |
3421 | struct swap_info_struct *si = swap_info[type]; | |
3422 | ||
3423 | if ((si->flags & SWP_USED) && !(si->flags & SWP_WRITEOK)) | |
3424 | nr_to_be_unused += si->inuse_pages; | |
1da177e4 | 3425 | } |
ec8acf20 | 3426 | val->freeswap = atomic_long_read(&nr_swap_pages) + nr_to_be_unused; |
1da177e4 | 3427 | val->totalswap = total_swap_pages + nr_to_be_unused; |
5d337b91 | 3428 | spin_unlock(&swap_lock); |
1da177e4 LT |
3429 | } |
3430 | ||
3431 | /* | |
3432 | * Verify that a swap entry is valid and increment its swap map count. | |
3433 | * | |
355cfa73 KH |
3434 | * Returns error code in following case. |
3435 | * - success -> 0 | |
3436 | * - swp_entry is invalid -> EINVAL | |
3437 | * - swp_entry is migration entry -> EINVAL | |
3438 | * - swap-cache reference is requested but there is already one. -> EEXIST | |
3439 | * - swap-cache reference is requested but the entry is not used. -> ENOENT | |
570a335b | 3440 | * - swap-mapped reference requested but needs continued swap count. -> ENOMEM |
1da177e4 | 3441 | */ |
8d69aaee | 3442 | static int __swap_duplicate(swp_entry_t entry, unsigned char usage) |
1da177e4 | 3443 | { |
73c34b6a | 3444 | struct swap_info_struct *p; |
235b6217 | 3445 | struct swap_cluster_info *ci; |
c10d38cc | 3446 | unsigned long offset; |
8d69aaee HD |
3447 | unsigned char count; |
3448 | unsigned char has_cache; | |
9d9a0334 | 3449 | int err; |
1da177e4 | 3450 | |
eb085574 | 3451 | p = get_swap_device(entry); |
c10d38cc | 3452 | if (!p) |
9d9a0334 | 3453 | return -EINVAL; |
235b6217 | 3454 | |
eb085574 | 3455 | offset = swp_offset(entry); |
235b6217 | 3456 | ci = lock_cluster_or_swap_info(p, offset); |
355cfa73 | 3457 | |
253d553b | 3458 | count = p->swap_map[offset]; |
edfe23da SL |
3459 | |
3460 | /* | |
3461 | * swapin_readahead() doesn't check if a swap entry is valid, so the | |
3462 | * swap entry could be SWAP_MAP_BAD. Check here with lock held. | |
3463 | */ | |
3464 | if (unlikely(swap_count(count) == SWAP_MAP_BAD)) { | |
3465 | err = -ENOENT; | |
3466 | goto unlock_out; | |
3467 | } | |
3468 | ||
253d553b HD |
3469 | has_cache = count & SWAP_HAS_CACHE; |
3470 | count &= ~SWAP_HAS_CACHE; | |
3471 | err = 0; | |
355cfa73 | 3472 | |
253d553b | 3473 | if (usage == SWAP_HAS_CACHE) { |
355cfa73 KH |
3474 | |
3475 | /* set SWAP_HAS_CACHE if there is no cache and entry is used */ | |
253d553b HD |
3476 | if (!has_cache && count) |
3477 | has_cache = SWAP_HAS_CACHE; | |
3478 | else if (has_cache) /* someone else added cache */ | |
3479 | err = -EEXIST; | |
3480 | else /* no users remaining */ | |
3481 | err = -ENOENT; | |
355cfa73 KH |
3482 | |
3483 | } else if (count || has_cache) { | |
253d553b | 3484 | |
570a335b HD |
3485 | if ((count & ~COUNT_CONTINUED) < SWAP_MAP_MAX) |
3486 | count += usage; | |
3487 | else if ((count & ~COUNT_CONTINUED) > SWAP_MAP_MAX) | |
253d553b | 3488 | err = -EINVAL; |
570a335b HD |
3489 | else if (swap_count_continued(p, offset, count)) |
3490 | count = COUNT_CONTINUED; | |
3491 | else | |
3492 | err = -ENOMEM; | |
355cfa73 | 3493 | } else |
253d553b HD |
3494 | err = -ENOENT; /* unused swap entry */ |
3495 | ||
a449bf58 | 3496 | WRITE_ONCE(p->swap_map[offset], count | has_cache); |
253d553b | 3497 | |
355cfa73 | 3498 | unlock_out: |
235b6217 | 3499 | unlock_cluster_or_swap_info(p, ci); |
eb085574 HY |
3500 | if (p) |
3501 | put_swap_device(p); | |
253d553b | 3502 | return err; |
1da177e4 | 3503 | } |
253d553b | 3504 | |
aaa46865 HD |
3505 | /* |
3506 | * Help swapoff by noting that swap entry belongs to shmem/tmpfs | |
3507 | * (in which case its reference count is never incremented). | |
3508 | */ | |
3509 | void swap_shmem_alloc(swp_entry_t entry) | |
3510 | { | |
3511 | __swap_duplicate(entry, SWAP_MAP_SHMEM); | |
3512 | } | |
3513 | ||
355cfa73 | 3514 | /* |
08259d58 HD |
3515 | * Increase reference count of swap entry by 1. |
3516 | * Returns 0 for success, or -ENOMEM if a swap_count_continuation is required | |
3517 | * but could not be atomically allocated. Returns 0, just as if it succeeded, | |
3518 | * if __swap_duplicate() fails for another reason (-EINVAL or -ENOENT), which | |
3519 | * might occur if a page table entry has got corrupted. | |
355cfa73 | 3520 | */ |
570a335b | 3521 | int swap_duplicate(swp_entry_t entry) |
355cfa73 | 3522 | { |
570a335b HD |
3523 | int err = 0; |
3524 | ||
3525 | while (!err && __swap_duplicate(entry, 1) == -ENOMEM) | |
3526 | err = add_swap_count_continuation(entry, GFP_ATOMIC); | |
3527 | return err; | |
355cfa73 | 3528 | } |
1da177e4 | 3529 | |
cb4b86ba | 3530 | /* |
355cfa73 KH |
3531 | * @entry: swap entry for which we allocate swap cache. |
3532 | * | |
73c34b6a | 3533 | * Called when allocating swap cache for existing swap entry, |
355cfa73 | 3534 | * This can return error codes. Returns 0 at success. |
3eeba135 | 3535 | * -EEXIST means there is a swap cache. |
355cfa73 | 3536 | * Note: return code is different from swap_duplicate(). |
cb4b86ba KH |
3537 | */ |
3538 | int swapcache_prepare(swp_entry_t entry) | |
3539 | { | |
253d553b | 3540 | return __swap_duplicate(entry, SWAP_HAS_CACHE); |
cb4b86ba KH |
3541 | } |
3542 | ||
0bcac06f MK |
3543 | struct swap_info_struct *swp_swap_info(swp_entry_t entry) |
3544 | { | |
c10d38cc | 3545 | return swap_type_to_swap_info(swp_type(entry)); |
0bcac06f MK |
3546 | } |
3547 | ||
f981c595 MG |
3548 | struct swap_info_struct *page_swap_info(struct page *page) |
3549 | { | |
0bcac06f MK |
3550 | swp_entry_t entry = { .val = page_private(page) }; |
3551 | return swp_swap_info(entry); | |
f981c595 MG |
3552 | } |
3553 | ||
3554 | /* | |
3555 | * out-of-line __page_file_ methods to avoid include hell. | |
3556 | */ | |
3557 | struct address_space *__page_file_mapping(struct page *page) | |
3558 | { | |
f981c595 MG |
3559 | return page_swap_info(page)->swap_file->f_mapping; |
3560 | } | |
3561 | EXPORT_SYMBOL_GPL(__page_file_mapping); | |
3562 | ||
3563 | pgoff_t __page_file_index(struct page *page) | |
3564 | { | |
3565 | swp_entry_t swap = { .val = page_private(page) }; | |
f981c595 MG |
3566 | return swp_offset(swap); |
3567 | } | |
3568 | EXPORT_SYMBOL_GPL(__page_file_index); | |
3569 | ||
570a335b HD |
3570 | /* |
3571 | * add_swap_count_continuation - called when a swap count is duplicated | |
3572 | * beyond SWAP_MAP_MAX, it allocates a new page and links that to the entry's | |
3573 | * page of the original vmalloc'ed swap_map, to hold the continuation count | |
3574 | * (for that entry and for its neighbouring PAGE_SIZE swap entries). Called | |
3575 | * again when count is duplicated beyond SWAP_MAP_MAX * SWAP_CONT_MAX, etc. | |
3576 | * | |
3577 | * These continuation pages are seldom referenced: the common paths all work | |
3578 | * on the original swap_map, only referring to a continuation page when the | |
3579 | * low "digit" of a count is incremented or decremented through SWAP_MAP_MAX. | |
3580 | * | |
3581 | * add_swap_count_continuation(, GFP_ATOMIC) can be called while holding | |
3582 | * page table locks; if it fails, add_swap_count_continuation(, GFP_KERNEL) | |
3583 | * can be called after dropping locks. | |
3584 | */ | |
3585 | int add_swap_count_continuation(swp_entry_t entry, gfp_t gfp_mask) | |
3586 | { | |
3587 | struct swap_info_struct *si; | |
235b6217 | 3588 | struct swap_cluster_info *ci; |
570a335b HD |
3589 | struct page *head; |
3590 | struct page *page; | |
3591 | struct page *list_page; | |
3592 | pgoff_t offset; | |
3593 | unsigned char count; | |
eb085574 | 3594 | int ret = 0; |
570a335b HD |
3595 | |
3596 | /* | |
3597 | * When debugging, it's easier to use __GFP_ZERO here; but it's better | |
3598 | * for latency not to zero a page while GFP_ATOMIC and holding locks. | |
3599 | */ | |
3600 | page = alloc_page(gfp_mask | __GFP_HIGHMEM); | |
3601 | ||
eb085574 | 3602 | si = get_swap_device(entry); |
570a335b HD |
3603 | if (!si) { |
3604 | /* | |
3605 | * An acceptable race has occurred since the failing | |
eb085574 | 3606 | * __swap_duplicate(): the swap device may be swapoff |
570a335b HD |
3607 | */ |
3608 | goto outer; | |
3609 | } | |
eb085574 | 3610 | spin_lock(&si->lock); |
570a335b HD |
3611 | |
3612 | offset = swp_offset(entry); | |
235b6217 HY |
3613 | |
3614 | ci = lock_cluster(si, offset); | |
3615 | ||
d8aa24e0 | 3616 | count = swap_count(si->swap_map[offset]); |
570a335b HD |
3617 | |
3618 | if ((count & ~COUNT_CONTINUED) != SWAP_MAP_MAX) { | |
3619 | /* | |
3620 | * The higher the swap count, the more likely it is that tasks | |
3621 | * will race to add swap count continuation: we need to avoid | |
3622 | * over-provisioning. | |
3623 | */ | |
3624 | goto out; | |
3625 | } | |
3626 | ||
3627 | if (!page) { | |
eb085574 HY |
3628 | ret = -ENOMEM; |
3629 | goto out; | |
570a335b HD |
3630 | } |
3631 | ||
3632 | /* | |
3633 | * We are fortunate that although vmalloc_to_page uses pte_offset_map, | |
2de1a7e4 SJ |
3634 | * no architecture is using highmem pages for kernel page tables: so it |
3635 | * will not corrupt the GFP_ATOMIC caller's atomic page table kmaps. | |
570a335b HD |
3636 | */ |
3637 | head = vmalloc_to_page(si->swap_map + offset); | |
3638 | offset &= ~PAGE_MASK; | |
3639 | ||
2628bd6f | 3640 | spin_lock(&si->cont_lock); |
570a335b HD |
3641 | /* |
3642 | * Page allocation does not initialize the page's lru field, | |
3643 | * but it does always reset its private field. | |
3644 | */ | |
3645 | if (!page_private(head)) { | |
3646 | BUG_ON(count & COUNT_CONTINUED); | |
3647 | INIT_LIST_HEAD(&head->lru); | |
3648 | set_page_private(head, SWP_CONTINUED); | |
3649 | si->flags |= SWP_CONTINUED; | |
3650 | } | |
3651 | ||
3652 | list_for_each_entry(list_page, &head->lru, lru) { | |
3653 | unsigned char *map; | |
3654 | ||
3655 | /* | |
3656 | * If the previous map said no continuation, but we've found | |
3657 | * a continuation page, free our allocation and use this one. | |
3658 | */ | |
3659 | if (!(count & COUNT_CONTINUED)) | |
2628bd6f | 3660 | goto out_unlock_cont; |
570a335b | 3661 | |
9b04c5fe | 3662 | map = kmap_atomic(list_page) + offset; |
570a335b | 3663 | count = *map; |
9b04c5fe | 3664 | kunmap_atomic(map); |
570a335b HD |
3665 | |
3666 | /* | |
3667 | * If this continuation count now has some space in it, | |
3668 | * free our allocation and use this one. | |
3669 | */ | |
3670 | if ((count & ~COUNT_CONTINUED) != SWAP_CONT_MAX) | |
2628bd6f | 3671 | goto out_unlock_cont; |
570a335b HD |
3672 | } |
3673 | ||
3674 | list_add_tail(&page->lru, &head->lru); | |
3675 | page = NULL; /* now it's attached, don't free it */ | |
2628bd6f HY |
3676 | out_unlock_cont: |
3677 | spin_unlock(&si->cont_lock); | |
570a335b | 3678 | out: |
235b6217 | 3679 | unlock_cluster(ci); |
ec8acf20 | 3680 | spin_unlock(&si->lock); |
eb085574 | 3681 | put_swap_device(si); |
570a335b HD |
3682 | outer: |
3683 | if (page) | |
3684 | __free_page(page); | |
eb085574 | 3685 | return ret; |
570a335b HD |
3686 | } |
3687 | ||
3688 | /* | |
3689 | * swap_count_continued - when the original swap_map count is incremented | |
3690 | * from SWAP_MAP_MAX, check if there is already a continuation page to carry | |
3691 | * into, carry if so, or else fail until a new continuation page is allocated; | |
3692 | * when the original swap_map count is decremented from 0 with continuation, | |
3693 | * borrow from the continuation and report whether it still holds more. | |
235b6217 HY |
3694 | * Called while __swap_duplicate() or swap_entry_free() holds swap or cluster |
3695 | * lock. | |
570a335b HD |
3696 | */ |
3697 | static bool swap_count_continued(struct swap_info_struct *si, | |
3698 | pgoff_t offset, unsigned char count) | |
3699 | { | |
3700 | struct page *head; | |
3701 | struct page *page; | |
3702 | unsigned char *map; | |
2628bd6f | 3703 | bool ret; |
570a335b HD |
3704 | |
3705 | head = vmalloc_to_page(si->swap_map + offset); | |
3706 | if (page_private(head) != SWP_CONTINUED) { | |
3707 | BUG_ON(count & COUNT_CONTINUED); | |
3708 | return false; /* need to add count continuation */ | |
3709 | } | |
3710 | ||
2628bd6f | 3711 | spin_lock(&si->cont_lock); |
570a335b | 3712 | offset &= ~PAGE_MASK; |
213516ac | 3713 | page = list_next_entry(head, lru); |
9b04c5fe | 3714 | map = kmap_atomic(page) + offset; |
570a335b HD |
3715 | |
3716 | if (count == SWAP_MAP_MAX) /* initial increment from swap_map */ | |
3717 | goto init_map; /* jump over SWAP_CONT_MAX checks */ | |
3718 | ||
3719 | if (count == (SWAP_MAP_MAX | COUNT_CONTINUED)) { /* incrementing */ | |
3720 | /* | |
3721 | * Think of how you add 1 to 999 | |
3722 | */ | |
3723 | while (*map == (SWAP_CONT_MAX | COUNT_CONTINUED)) { | |
9b04c5fe | 3724 | kunmap_atomic(map); |
213516ac | 3725 | page = list_next_entry(page, lru); |
570a335b | 3726 | BUG_ON(page == head); |
9b04c5fe | 3727 | map = kmap_atomic(page) + offset; |
570a335b HD |
3728 | } |
3729 | if (*map == SWAP_CONT_MAX) { | |
9b04c5fe | 3730 | kunmap_atomic(map); |
213516ac | 3731 | page = list_next_entry(page, lru); |
2628bd6f HY |
3732 | if (page == head) { |
3733 | ret = false; /* add count continuation */ | |
3734 | goto out; | |
3735 | } | |
9b04c5fe | 3736 | map = kmap_atomic(page) + offset; |
570a335b HD |
3737 | init_map: *map = 0; /* we didn't zero the page */ |
3738 | } | |
3739 | *map += 1; | |
9b04c5fe | 3740 | kunmap_atomic(map); |
213516ac | 3741 | while ((page = list_prev_entry(page, lru)) != head) { |
9b04c5fe | 3742 | map = kmap_atomic(page) + offset; |
570a335b | 3743 | *map = COUNT_CONTINUED; |
9b04c5fe | 3744 | kunmap_atomic(map); |
570a335b | 3745 | } |
2628bd6f | 3746 | ret = true; /* incremented */ |
570a335b HD |
3747 | |
3748 | } else { /* decrementing */ | |
3749 | /* | |
3750 | * Think of how you subtract 1 from 1000 | |
3751 | */ | |
3752 | BUG_ON(count != COUNT_CONTINUED); | |
3753 | while (*map == COUNT_CONTINUED) { | |
9b04c5fe | 3754 | kunmap_atomic(map); |
213516ac | 3755 | page = list_next_entry(page, lru); |
570a335b | 3756 | BUG_ON(page == head); |
9b04c5fe | 3757 | map = kmap_atomic(page) + offset; |
570a335b HD |
3758 | } |
3759 | BUG_ON(*map == 0); | |
3760 | *map -= 1; | |
3761 | if (*map == 0) | |
3762 | count = 0; | |
9b04c5fe | 3763 | kunmap_atomic(map); |
213516ac | 3764 | while ((page = list_prev_entry(page, lru)) != head) { |
9b04c5fe | 3765 | map = kmap_atomic(page) + offset; |
570a335b HD |
3766 | *map = SWAP_CONT_MAX | count; |
3767 | count = COUNT_CONTINUED; | |
9b04c5fe | 3768 | kunmap_atomic(map); |
570a335b | 3769 | } |
2628bd6f | 3770 | ret = count == COUNT_CONTINUED; |
570a335b | 3771 | } |
2628bd6f HY |
3772 | out: |
3773 | spin_unlock(&si->cont_lock); | |
3774 | return ret; | |
570a335b HD |
3775 | } |
3776 | ||
3777 | /* | |
3778 | * free_swap_count_continuations - swapoff free all the continuation pages | |
3779 | * appended to the swap_map, after swap_map is quiesced, before vfree'ing it. | |
3780 | */ | |
3781 | static void free_swap_count_continuations(struct swap_info_struct *si) | |
3782 | { | |
3783 | pgoff_t offset; | |
3784 | ||
3785 | for (offset = 0; offset < si->max; offset += PAGE_SIZE) { | |
3786 | struct page *head; | |
3787 | head = vmalloc_to_page(si->swap_map + offset); | |
3788 | if (page_private(head)) { | |
0d576d20 GT |
3789 | struct page *page, *next; |
3790 | ||
3791 | list_for_each_entry_safe(page, next, &head->lru, lru) { | |
3792 | list_del(&page->lru); | |
570a335b HD |
3793 | __free_page(page); |
3794 | } | |
3795 | } | |
3796 | } | |
3797 | } | |
a2468cc9 | 3798 | |
2cf85583 | 3799 | #if defined(CONFIG_MEMCG) && defined(CONFIG_BLK_CGROUP) |
6caa6a07 | 3800 | void cgroup_throttle_swaprate(struct page *page, gfp_t gfp_mask) |
2cf85583 TH |
3801 | { |
3802 | struct swap_info_struct *si, *next; | |
6caa6a07 JW |
3803 | int nid = page_to_nid(page); |
3804 | ||
3805 | if (!(gfp_mask & __GFP_IO)) | |
2cf85583 TH |
3806 | return; |
3807 | ||
3808 | if (!blk_cgroup_congested()) | |
3809 | return; | |
3810 | ||
3811 | /* | |
3812 | * We've already scheduled a throttle, avoid taking the global swap | |
3813 | * lock. | |
3814 | */ | |
3815 | if (current->throttle_queue) | |
3816 | return; | |
3817 | ||
3818 | spin_lock(&swap_avail_lock); | |
6caa6a07 JW |
3819 | plist_for_each_entry_safe(si, next, &swap_avail_heads[nid], |
3820 | avail_lists[nid]) { | |
2cf85583 | 3821 | if (si->bdev) { |
6caa6a07 | 3822 | blkcg_schedule_throttle(bdev_get_queue(si->bdev), true); |
2cf85583 TH |
3823 | break; |
3824 | } | |
3825 | } | |
3826 | spin_unlock(&swap_avail_lock); | |
3827 | } | |
3828 | #endif | |
3829 | ||
a2468cc9 AL |
3830 | static int __init swapfile_init(void) |
3831 | { | |
3832 | int nid; | |
3833 | ||
3834 | swap_avail_heads = kmalloc_array(nr_node_ids, sizeof(struct plist_head), | |
3835 | GFP_KERNEL); | |
3836 | if (!swap_avail_heads) { | |
3837 | pr_emerg("Not enough memory for swap heads, swap is disabled\n"); | |
3838 | return -ENOMEM; | |
3839 | } | |
3840 | ||
3841 | for_each_node(nid) | |
3842 | plist_head_init(&swap_avail_heads[nid]); | |
3843 | ||
3844 | return 0; | |
3845 | } | |
3846 | subsys_initcall(swapfile_init); |