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> |
1da177e4 | 42 | |
1da177e4 LT |
43 | #include <asm/tlbflush.h> |
44 | #include <linux/swapops.h> | |
5d1ea48b | 45 | #include <linux/swap_cgroup.h> |
1da177e4 | 46 | |
570a335b HD |
47 | static bool swap_count_continued(struct swap_info_struct *, pgoff_t, |
48 | unsigned char); | |
49 | static void free_swap_count_continuations(struct swap_info_struct *); | |
d4906e1a | 50 | static sector_t map_swap_entry(swp_entry_t, struct block_device**); |
570a335b | 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 | ||
7992fde7 HD |
223 | /* |
224 | * swap allocation tell device that a cluster of swap can now be discarded, | |
225 | * to allow the swap device to optimize its wear-levelling. | |
226 | */ | |
227 | static void discard_swap_cluster(struct swap_info_struct *si, | |
228 | pgoff_t start_page, pgoff_t nr_pages) | |
229 | { | |
4efaceb1 | 230 | struct swap_extent *se = offset_to_swap_extent(si, start_page); |
7992fde7 HD |
231 | |
232 | while (nr_pages) { | |
4efaceb1 AL |
233 | pgoff_t offset = start_page - se->start_page; |
234 | sector_t start_block = se->start_block + offset; | |
235 | sector_t nr_blocks = se->nr_pages - offset; | |
236 | ||
237 | if (nr_blocks > nr_pages) | |
238 | nr_blocks = nr_pages; | |
239 | start_page += nr_blocks; | |
240 | nr_pages -= nr_blocks; | |
241 | ||
242 | start_block <<= PAGE_SHIFT - 9; | |
243 | nr_blocks <<= PAGE_SHIFT - 9; | |
244 | if (blkdev_issue_discard(si->bdev, start_block, | |
245 | nr_blocks, GFP_NOIO, 0)) | |
246 | break; | |
7992fde7 | 247 | |
4efaceb1 | 248 | se = next_se(se); |
7992fde7 HD |
249 | } |
250 | } | |
251 | ||
38d8b4e6 HY |
252 | #ifdef CONFIG_THP_SWAP |
253 | #define SWAPFILE_CLUSTER HPAGE_PMD_NR | |
a448f2d0 HY |
254 | |
255 | #define swap_entry_size(size) (size) | |
38d8b4e6 | 256 | #else |
048c27fd | 257 | #define SWAPFILE_CLUSTER 256 |
a448f2d0 HY |
258 | |
259 | /* | |
260 | * Define swap_entry_size() as constant to let compiler to optimize | |
261 | * out some code if !CONFIG_THP_SWAP | |
262 | */ | |
263 | #define swap_entry_size(size) 1 | |
38d8b4e6 | 264 | #endif |
048c27fd HD |
265 | #define LATENCY_LIMIT 256 |
266 | ||
2a8f9449 SL |
267 | static inline void cluster_set_flag(struct swap_cluster_info *info, |
268 | unsigned int flag) | |
269 | { | |
270 | info->flags = flag; | |
271 | } | |
272 | ||
273 | static inline unsigned int cluster_count(struct swap_cluster_info *info) | |
274 | { | |
275 | return info->data; | |
276 | } | |
277 | ||
278 | static inline void cluster_set_count(struct swap_cluster_info *info, | |
279 | unsigned int c) | |
280 | { | |
281 | info->data = c; | |
282 | } | |
283 | ||
284 | static inline void cluster_set_count_flag(struct swap_cluster_info *info, | |
285 | unsigned int c, unsigned int f) | |
286 | { | |
287 | info->flags = f; | |
288 | info->data = c; | |
289 | } | |
290 | ||
291 | static inline unsigned int cluster_next(struct swap_cluster_info *info) | |
292 | { | |
293 | return info->data; | |
294 | } | |
295 | ||
296 | static inline void cluster_set_next(struct swap_cluster_info *info, | |
297 | unsigned int n) | |
298 | { | |
299 | info->data = n; | |
300 | } | |
301 | ||
302 | static inline void cluster_set_next_flag(struct swap_cluster_info *info, | |
303 | unsigned int n, unsigned int f) | |
304 | { | |
305 | info->flags = f; | |
306 | info->data = n; | |
307 | } | |
308 | ||
309 | static inline bool cluster_is_free(struct swap_cluster_info *info) | |
310 | { | |
311 | return info->flags & CLUSTER_FLAG_FREE; | |
312 | } | |
313 | ||
314 | static inline bool cluster_is_null(struct swap_cluster_info *info) | |
315 | { | |
316 | return info->flags & CLUSTER_FLAG_NEXT_NULL; | |
317 | } | |
318 | ||
319 | static inline void cluster_set_null(struct swap_cluster_info *info) | |
320 | { | |
321 | info->flags = CLUSTER_FLAG_NEXT_NULL; | |
322 | info->data = 0; | |
323 | } | |
324 | ||
e0709829 HY |
325 | static inline bool cluster_is_huge(struct swap_cluster_info *info) |
326 | { | |
33ee011e HY |
327 | if (IS_ENABLED(CONFIG_THP_SWAP)) |
328 | return info->flags & CLUSTER_FLAG_HUGE; | |
329 | return false; | |
e0709829 HY |
330 | } |
331 | ||
332 | static inline void cluster_clear_huge(struct swap_cluster_info *info) | |
333 | { | |
334 | info->flags &= ~CLUSTER_FLAG_HUGE; | |
335 | } | |
336 | ||
235b6217 HY |
337 | static inline struct swap_cluster_info *lock_cluster(struct swap_info_struct *si, |
338 | unsigned long offset) | |
339 | { | |
340 | struct swap_cluster_info *ci; | |
341 | ||
342 | ci = si->cluster_info; | |
343 | if (ci) { | |
344 | ci += offset / SWAPFILE_CLUSTER; | |
345 | spin_lock(&ci->lock); | |
346 | } | |
347 | return ci; | |
348 | } | |
349 | ||
350 | static inline void unlock_cluster(struct swap_cluster_info *ci) | |
351 | { | |
352 | if (ci) | |
353 | spin_unlock(&ci->lock); | |
354 | } | |
355 | ||
59d98bf3 HY |
356 | /* |
357 | * Determine the locking method in use for this device. Return | |
358 | * swap_cluster_info if SSD-style cluster-based locking is in place. | |
359 | */ | |
235b6217 | 360 | static inline struct swap_cluster_info *lock_cluster_or_swap_info( |
59d98bf3 | 361 | struct swap_info_struct *si, unsigned long offset) |
235b6217 HY |
362 | { |
363 | struct swap_cluster_info *ci; | |
364 | ||
59d98bf3 | 365 | /* Try to use fine-grained SSD-style locking if available: */ |
235b6217 | 366 | ci = lock_cluster(si, offset); |
59d98bf3 | 367 | /* Otherwise, fall back to traditional, coarse locking: */ |
235b6217 HY |
368 | if (!ci) |
369 | spin_lock(&si->lock); | |
370 | ||
371 | return ci; | |
372 | } | |
373 | ||
374 | static inline void unlock_cluster_or_swap_info(struct swap_info_struct *si, | |
375 | struct swap_cluster_info *ci) | |
376 | { | |
377 | if (ci) | |
378 | unlock_cluster(ci); | |
379 | else | |
380 | spin_unlock(&si->lock); | |
381 | } | |
382 | ||
6b534915 HY |
383 | static inline bool cluster_list_empty(struct swap_cluster_list *list) |
384 | { | |
385 | return cluster_is_null(&list->head); | |
386 | } | |
387 | ||
388 | static inline unsigned int cluster_list_first(struct swap_cluster_list *list) | |
389 | { | |
390 | return cluster_next(&list->head); | |
391 | } | |
392 | ||
393 | static void cluster_list_init(struct swap_cluster_list *list) | |
394 | { | |
395 | cluster_set_null(&list->head); | |
396 | cluster_set_null(&list->tail); | |
397 | } | |
398 | ||
399 | static void cluster_list_add_tail(struct swap_cluster_list *list, | |
400 | struct swap_cluster_info *ci, | |
401 | unsigned int idx) | |
402 | { | |
403 | if (cluster_list_empty(list)) { | |
404 | cluster_set_next_flag(&list->head, idx, 0); | |
405 | cluster_set_next_flag(&list->tail, idx, 0); | |
406 | } else { | |
235b6217 | 407 | struct swap_cluster_info *ci_tail; |
6b534915 HY |
408 | unsigned int tail = cluster_next(&list->tail); |
409 | ||
235b6217 HY |
410 | /* |
411 | * Nested cluster lock, but both cluster locks are | |
412 | * only acquired when we held swap_info_struct->lock | |
413 | */ | |
414 | ci_tail = ci + tail; | |
415 | spin_lock_nested(&ci_tail->lock, SINGLE_DEPTH_NESTING); | |
416 | cluster_set_next(ci_tail, idx); | |
0ef017d1 | 417 | spin_unlock(&ci_tail->lock); |
6b534915 HY |
418 | cluster_set_next_flag(&list->tail, idx, 0); |
419 | } | |
420 | } | |
421 | ||
422 | static unsigned int cluster_list_del_first(struct swap_cluster_list *list, | |
423 | struct swap_cluster_info *ci) | |
424 | { | |
425 | unsigned int idx; | |
426 | ||
427 | idx = cluster_next(&list->head); | |
428 | if (cluster_next(&list->tail) == idx) { | |
429 | cluster_set_null(&list->head); | |
430 | cluster_set_null(&list->tail); | |
431 | } else | |
432 | cluster_set_next_flag(&list->head, | |
433 | cluster_next(&ci[idx]), 0); | |
434 | ||
435 | return idx; | |
436 | } | |
437 | ||
815c2c54 SL |
438 | /* Add a cluster to discard list and schedule it to do discard */ |
439 | static void swap_cluster_schedule_discard(struct swap_info_struct *si, | |
440 | unsigned int idx) | |
441 | { | |
442 | /* | |
443 | * If scan_swap_map() can't find a free cluster, it will check | |
444 | * si->swap_map directly. To make sure the discarding cluster isn't | |
445 | * taken by scan_swap_map(), mark the swap entries bad (occupied). It | |
446 | * will be cleared after discard | |
447 | */ | |
448 | memset(si->swap_map + idx * SWAPFILE_CLUSTER, | |
449 | SWAP_MAP_BAD, SWAPFILE_CLUSTER); | |
450 | ||
6b534915 | 451 | cluster_list_add_tail(&si->discard_clusters, si->cluster_info, idx); |
815c2c54 SL |
452 | |
453 | schedule_work(&si->discard_work); | |
454 | } | |
455 | ||
38d8b4e6 HY |
456 | static void __free_cluster(struct swap_info_struct *si, unsigned long idx) |
457 | { | |
458 | struct swap_cluster_info *ci = si->cluster_info; | |
459 | ||
460 | cluster_set_flag(ci + idx, CLUSTER_FLAG_FREE); | |
461 | cluster_list_add_tail(&si->free_clusters, ci, idx); | |
462 | } | |
463 | ||
815c2c54 SL |
464 | /* |
465 | * Doing discard actually. After a cluster discard is finished, the cluster | |
466 | * will be added to free cluster list. caller should hold si->lock. | |
467 | */ | |
468 | static void swap_do_scheduled_discard(struct swap_info_struct *si) | |
469 | { | |
235b6217 | 470 | struct swap_cluster_info *info, *ci; |
815c2c54 SL |
471 | unsigned int idx; |
472 | ||
473 | info = si->cluster_info; | |
474 | ||
6b534915 HY |
475 | while (!cluster_list_empty(&si->discard_clusters)) { |
476 | idx = cluster_list_del_first(&si->discard_clusters, info); | |
815c2c54 SL |
477 | spin_unlock(&si->lock); |
478 | ||
479 | discard_swap_cluster(si, idx * SWAPFILE_CLUSTER, | |
480 | SWAPFILE_CLUSTER); | |
481 | ||
482 | spin_lock(&si->lock); | |
235b6217 | 483 | ci = lock_cluster(si, idx * SWAPFILE_CLUSTER); |
38d8b4e6 | 484 | __free_cluster(si, idx); |
815c2c54 SL |
485 | memset(si->swap_map + idx * SWAPFILE_CLUSTER, |
486 | 0, SWAPFILE_CLUSTER); | |
235b6217 | 487 | unlock_cluster(ci); |
815c2c54 SL |
488 | } |
489 | } | |
490 | ||
491 | static void swap_discard_work(struct work_struct *work) | |
492 | { | |
493 | struct swap_info_struct *si; | |
494 | ||
495 | si = container_of(work, struct swap_info_struct, discard_work); | |
496 | ||
497 | spin_lock(&si->lock); | |
498 | swap_do_scheduled_discard(si); | |
499 | spin_unlock(&si->lock); | |
500 | } | |
501 | ||
38d8b4e6 HY |
502 | static void alloc_cluster(struct swap_info_struct *si, unsigned long idx) |
503 | { | |
504 | struct swap_cluster_info *ci = si->cluster_info; | |
505 | ||
506 | VM_BUG_ON(cluster_list_first(&si->free_clusters) != idx); | |
507 | cluster_list_del_first(&si->free_clusters, ci); | |
508 | cluster_set_count_flag(ci + idx, 0, 0); | |
509 | } | |
510 | ||
511 | static void free_cluster(struct swap_info_struct *si, unsigned long idx) | |
512 | { | |
513 | struct swap_cluster_info *ci = si->cluster_info + idx; | |
514 | ||
515 | VM_BUG_ON(cluster_count(ci) != 0); | |
516 | /* | |
517 | * If the swap is discardable, prepare discard the cluster | |
518 | * instead of free it immediately. The cluster will be freed | |
519 | * after discard. | |
520 | */ | |
521 | if ((si->flags & (SWP_WRITEOK | SWP_PAGE_DISCARD)) == | |
522 | (SWP_WRITEOK | SWP_PAGE_DISCARD)) { | |
523 | swap_cluster_schedule_discard(si, idx); | |
524 | return; | |
525 | } | |
526 | ||
527 | __free_cluster(si, idx); | |
528 | } | |
529 | ||
2a8f9449 SL |
530 | /* |
531 | * The cluster corresponding to page_nr will be used. The cluster will be | |
532 | * removed from free cluster list and its usage counter will be increased. | |
533 | */ | |
534 | static void inc_cluster_info_page(struct swap_info_struct *p, | |
535 | struct swap_cluster_info *cluster_info, unsigned long page_nr) | |
536 | { | |
537 | unsigned long idx = page_nr / SWAPFILE_CLUSTER; | |
538 | ||
539 | if (!cluster_info) | |
540 | return; | |
38d8b4e6 HY |
541 | if (cluster_is_free(&cluster_info[idx])) |
542 | alloc_cluster(p, idx); | |
2a8f9449 SL |
543 | |
544 | VM_BUG_ON(cluster_count(&cluster_info[idx]) >= SWAPFILE_CLUSTER); | |
545 | cluster_set_count(&cluster_info[idx], | |
546 | cluster_count(&cluster_info[idx]) + 1); | |
547 | } | |
548 | ||
549 | /* | |
550 | * The cluster corresponding to page_nr decreases one usage. If the usage | |
551 | * counter becomes 0, which means no page in the cluster is in using, we can | |
552 | * optionally discard the cluster and add it to free cluster list. | |
553 | */ | |
554 | static void dec_cluster_info_page(struct swap_info_struct *p, | |
555 | struct swap_cluster_info *cluster_info, unsigned long page_nr) | |
556 | { | |
557 | unsigned long idx = page_nr / SWAPFILE_CLUSTER; | |
558 | ||
559 | if (!cluster_info) | |
560 | return; | |
561 | ||
562 | VM_BUG_ON(cluster_count(&cluster_info[idx]) == 0); | |
563 | cluster_set_count(&cluster_info[idx], | |
564 | cluster_count(&cluster_info[idx]) - 1); | |
565 | ||
38d8b4e6 HY |
566 | if (cluster_count(&cluster_info[idx]) == 0) |
567 | free_cluster(p, idx); | |
2a8f9449 SL |
568 | } |
569 | ||
570 | /* | |
571 | * It's possible scan_swap_map() uses a free cluster in the middle of free | |
572 | * cluster list. Avoiding such abuse to avoid list corruption. | |
573 | */ | |
ebc2a1a6 SL |
574 | static bool |
575 | scan_swap_map_ssd_cluster_conflict(struct swap_info_struct *si, | |
2a8f9449 SL |
576 | unsigned long offset) |
577 | { | |
ebc2a1a6 SL |
578 | struct percpu_cluster *percpu_cluster; |
579 | bool conflict; | |
580 | ||
2a8f9449 | 581 | offset /= SWAPFILE_CLUSTER; |
6b534915 HY |
582 | conflict = !cluster_list_empty(&si->free_clusters) && |
583 | offset != cluster_list_first(&si->free_clusters) && | |
2a8f9449 | 584 | cluster_is_free(&si->cluster_info[offset]); |
ebc2a1a6 SL |
585 | |
586 | if (!conflict) | |
587 | return false; | |
588 | ||
589 | percpu_cluster = this_cpu_ptr(si->percpu_cluster); | |
590 | cluster_set_null(&percpu_cluster->index); | |
591 | return true; | |
592 | } | |
593 | ||
594 | /* | |
595 | * Try to get a swap entry from current cpu's swap entry pool (a cluster). This | |
596 | * might involve allocating a new cluster for current CPU too. | |
597 | */ | |
36005bae | 598 | static bool scan_swap_map_try_ssd_cluster(struct swap_info_struct *si, |
ebc2a1a6 SL |
599 | unsigned long *offset, unsigned long *scan_base) |
600 | { | |
601 | struct percpu_cluster *cluster; | |
235b6217 | 602 | struct swap_cluster_info *ci; |
235b6217 | 603 | unsigned long tmp, max; |
ebc2a1a6 SL |
604 | |
605 | new_cluster: | |
606 | cluster = this_cpu_ptr(si->percpu_cluster); | |
607 | if (cluster_is_null(&cluster->index)) { | |
6b534915 HY |
608 | if (!cluster_list_empty(&si->free_clusters)) { |
609 | cluster->index = si->free_clusters.head; | |
ebc2a1a6 SL |
610 | cluster->next = cluster_next(&cluster->index) * |
611 | SWAPFILE_CLUSTER; | |
6b534915 | 612 | } else if (!cluster_list_empty(&si->discard_clusters)) { |
ebc2a1a6 SL |
613 | /* |
614 | * we don't have free cluster but have some clusters in | |
49070588 HY |
615 | * discarding, do discard now and reclaim them, then |
616 | * reread cluster_next_cpu since we dropped si->lock | |
ebc2a1a6 SL |
617 | */ |
618 | swap_do_scheduled_discard(si); | |
49070588 HY |
619 | *scan_base = this_cpu_read(*si->cluster_next_cpu); |
620 | *offset = *scan_base; | |
ebc2a1a6 SL |
621 | goto new_cluster; |
622 | } else | |
36005bae | 623 | return false; |
ebc2a1a6 SL |
624 | } |
625 | ||
ebc2a1a6 SL |
626 | /* |
627 | * Other CPUs can use our cluster if they can't find a free cluster, | |
628 | * check if there is still free entry in the cluster | |
629 | */ | |
630 | tmp = cluster->next; | |
235b6217 HY |
631 | max = min_t(unsigned long, si->max, |
632 | (cluster_next(&cluster->index) + 1) * SWAPFILE_CLUSTER); | |
7b9e2de1 WY |
633 | if (tmp < max) { |
634 | ci = lock_cluster(si, tmp); | |
635 | while (tmp < max) { | |
636 | if (!si->swap_map[tmp]) | |
637 | break; | |
638 | tmp++; | |
639 | } | |
640 | unlock_cluster(ci); | |
ebc2a1a6 | 641 | } |
0fd0e19e | 642 | if (tmp >= max) { |
ebc2a1a6 SL |
643 | cluster_set_null(&cluster->index); |
644 | goto new_cluster; | |
645 | } | |
646 | cluster->next = tmp + 1; | |
647 | *offset = tmp; | |
648 | *scan_base = tmp; | |
fdff1deb | 649 | return true; |
2a8f9449 SL |
650 | } |
651 | ||
a2468cc9 AL |
652 | static void __del_from_avail_list(struct swap_info_struct *p) |
653 | { | |
654 | int nid; | |
655 | ||
656 | for_each_node(nid) | |
657 | plist_del(&p->avail_lists[nid], &swap_avail_heads[nid]); | |
658 | } | |
659 | ||
660 | static void del_from_avail_list(struct swap_info_struct *p) | |
661 | { | |
662 | spin_lock(&swap_avail_lock); | |
663 | __del_from_avail_list(p); | |
664 | spin_unlock(&swap_avail_lock); | |
665 | } | |
666 | ||
38d8b4e6 HY |
667 | static void swap_range_alloc(struct swap_info_struct *si, unsigned long offset, |
668 | unsigned int nr_entries) | |
669 | { | |
670 | unsigned int end = offset + nr_entries - 1; | |
671 | ||
672 | if (offset == si->lowest_bit) | |
673 | si->lowest_bit += nr_entries; | |
674 | if (end == si->highest_bit) | |
a449bf58 | 675 | WRITE_ONCE(si->highest_bit, si->highest_bit - nr_entries); |
38d8b4e6 HY |
676 | si->inuse_pages += nr_entries; |
677 | if (si->inuse_pages == si->pages) { | |
678 | si->lowest_bit = si->max; | |
679 | si->highest_bit = 0; | |
a2468cc9 | 680 | del_from_avail_list(si); |
38d8b4e6 HY |
681 | } |
682 | } | |
683 | ||
a2468cc9 AL |
684 | static void add_to_avail_list(struct swap_info_struct *p) |
685 | { | |
686 | int nid; | |
687 | ||
688 | spin_lock(&swap_avail_lock); | |
689 | for_each_node(nid) { | |
690 | WARN_ON(!plist_node_empty(&p->avail_lists[nid])); | |
691 | plist_add(&p->avail_lists[nid], &swap_avail_heads[nid]); | |
692 | } | |
693 | spin_unlock(&swap_avail_lock); | |
694 | } | |
695 | ||
38d8b4e6 HY |
696 | static void swap_range_free(struct swap_info_struct *si, unsigned long offset, |
697 | unsigned int nr_entries) | |
698 | { | |
3852f676 | 699 | unsigned long begin = offset; |
38d8b4e6 HY |
700 | unsigned long end = offset + nr_entries - 1; |
701 | void (*swap_slot_free_notify)(struct block_device *, unsigned long); | |
702 | ||
703 | if (offset < si->lowest_bit) | |
704 | si->lowest_bit = offset; | |
705 | if (end > si->highest_bit) { | |
706 | bool was_full = !si->highest_bit; | |
707 | ||
a449bf58 | 708 | WRITE_ONCE(si->highest_bit, end); |
a2468cc9 AL |
709 | if (was_full && (si->flags & SWP_WRITEOK)) |
710 | add_to_avail_list(si); | |
38d8b4e6 HY |
711 | } |
712 | atomic_long_add(nr_entries, &nr_swap_pages); | |
713 | si->inuse_pages -= nr_entries; | |
714 | if (si->flags & SWP_BLKDEV) | |
715 | swap_slot_free_notify = | |
716 | si->bdev->bd_disk->fops->swap_slot_free_notify; | |
717 | else | |
718 | swap_slot_free_notify = NULL; | |
719 | while (offset <= end) { | |
8a84802e | 720 | arch_swap_invalidate_page(si->type, offset); |
38d8b4e6 HY |
721 | frontswap_invalidate_page(si->type, offset); |
722 | if (swap_slot_free_notify) | |
723 | swap_slot_free_notify(si->bdev, offset); | |
724 | offset++; | |
725 | } | |
3852f676 | 726 | clear_shadow_from_swap_cache(si->type, begin, end); |
38d8b4e6 HY |
727 | } |
728 | ||
49070588 HY |
729 | static void set_cluster_next(struct swap_info_struct *si, unsigned long next) |
730 | { | |
731 | unsigned long prev; | |
732 | ||
733 | if (!(si->flags & SWP_SOLIDSTATE)) { | |
734 | si->cluster_next = next; | |
735 | return; | |
736 | } | |
737 | ||
738 | prev = this_cpu_read(*si->cluster_next_cpu); | |
739 | /* | |
740 | * Cross the swap address space size aligned trunk, choose | |
741 | * another trunk randomly to avoid lock contention on swap | |
742 | * address space if possible. | |
743 | */ | |
744 | if ((prev >> SWAP_ADDRESS_SPACE_SHIFT) != | |
745 | (next >> SWAP_ADDRESS_SPACE_SHIFT)) { | |
746 | /* No free swap slots available */ | |
747 | if (si->highest_bit <= si->lowest_bit) | |
748 | return; | |
749 | next = si->lowest_bit + | |
750 | prandom_u32_max(si->highest_bit - si->lowest_bit + 1); | |
751 | next = ALIGN_DOWN(next, SWAP_ADDRESS_SPACE_PAGES); | |
752 | next = max_t(unsigned int, next, si->lowest_bit); | |
753 | } | |
754 | this_cpu_write(*si->cluster_next_cpu, next); | |
755 | } | |
756 | ||
36005bae TC |
757 | static int scan_swap_map_slots(struct swap_info_struct *si, |
758 | unsigned char usage, int nr, | |
759 | swp_entry_t slots[]) | |
1da177e4 | 760 | { |
235b6217 | 761 | struct swap_cluster_info *ci; |
ebebbbe9 | 762 | unsigned long offset; |
c60aa176 | 763 | unsigned long scan_base; |
7992fde7 | 764 | unsigned long last_in_cluster = 0; |
048c27fd | 765 | int latency_ration = LATENCY_LIMIT; |
36005bae | 766 | int n_ret = 0; |
ed43af10 | 767 | bool scanned_many = false; |
36005bae | 768 | |
886bb7e9 | 769 | /* |
7dfad418 HD |
770 | * We try to cluster swap pages by allocating them sequentially |
771 | * in swap. Once we've allocated SWAPFILE_CLUSTER pages this | |
772 | * way, however, we resort to first-free allocation, starting | |
773 | * a new cluster. This prevents us from scattering swap pages | |
774 | * all over the entire swap partition, so that we reduce | |
775 | * overall disk seek times between swap pages. -- sct | |
776 | * But we do now try to find an empty cluster. -Andrea | |
c60aa176 | 777 | * And we let swap pages go all over an SSD partition. Hugh |
7dfad418 HD |
778 | */ |
779 | ||
52b7efdb | 780 | si->flags += SWP_SCANNING; |
49070588 HY |
781 | /* |
782 | * Use percpu scan base for SSD to reduce lock contention on | |
783 | * cluster and swap cache. For HDD, sequential access is more | |
784 | * important. | |
785 | */ | |
786 | if (si->flags & SWP_SOLIDSTATE) | |
787 | scan_base = this_cpu_read(*si->cluster_next_cpu); | |
788 | else | |
789 | scan_base = si->cluster_next; | |
790 | offset = scan_base; | |
ebebbbe9 | 791 | |
ebc2a1a6 SL |
792 | /* SSD algorithm */ |
793 | if (si->cluster_info) { | |
bd2d18da | 794 | if (!scan_swap_map_try_ssd_cluster(si, &offset, &scan_base)) |
36005bae | 795 | goto scan; |
f4eaf51a | 796 | } else if (unlikely(!si->cluster_nr--)) { |
ebebbbe9 HD |
797 | if (si->pages - si->inuse_pages < SWAPFILE_CLUSTER) { |
798 | si->cluster_nr = SWAPFILE_CLUSTER - 1; | |
799 | goto checks; | |
800 | } | |
2a8f9449 | 801 | |
ec8acf20 | 802 | spin_unlock(&si->lock); |
7dfad418 | 803 | |
c60aa176 HD |
804 | /* |
805 | * If seek is expensive, start searching for new cluster from | |
806 | * start of partition, to minimize the span of allocated swap. | |
50088c44 CY |
807 | * If seek is cheap, that is the SWP_SOLIDSTATE si->cluster_info |
808 | * case, just handled by scan_swap_map_try_ssd_cluster() above. | |
c60aa176 | 809 | */ |
50088c44 | 810 | scan_base = offset = si->lowest_bit; |
7dfad418 HD |
811 | last_in_cluster = offset + SWAPFILE_CLUSTER - 1; |
812 | ||
813 | /* Locate the first empty (unaligned) cluster */ | |
814 | for (; last_in_cluster <= si->highest_bit; offset++) { | |
1da177e4 | 815 | if (si->swap_map[offset]) |
7dfad418 HD |
816 | last_in_cluster = offset + SWAPFILE_CLUSTER; |
817 | else if (offset == last_in_cluster) { | |
ec8acf20 | 818 | spin_lock(&si->lock); |
ebebbbe9 HD |
819 | offset -= SWAPFILE_CLUSTER - 1; |
820 | si->cluster_next = offset; | |
821 | si->cluster_nr = SWAPFILE_CLUSTER - 1; | |
c60aa176 HD |
822 | goto checks; |
823 | } | |
824 | if (unlikely(--latency_ration < 0)) { | |
825 | cond_resched(); | |
826 | latency_ration = LATENCY_LIMIT; | |
827 | } | |
828 | } | |
829 | ||
830 | offset = scan_base; | |
ec8acf20 | 831 | spin_lock(&si->lock); |
ebebbbe9 | 832 | si->cluster_nr = SWAPFILE_CLUSTER - 1; |
1da177e4 | 833 | } |
7dfad418 | 834 | |
ebebbbe9 | 835 | checks: |
ebc2a1a6 | 836 | if (si->cluster_info) { |
36005bae TC |
837 | while (scan_swap_map_ssd_cluster_conflict(si, offset)) { |
838 | /* take a break if we already got some slots */ | |
839 | if (n_ret) | |
840 | goto done; | |
841 | if (!scan_swap_map_try_ssd_cluster(si, &offset, | |
842 | &scan_base)) | |
843 | goto scan; | |
844 | } | |
ebc2a1a6 | 845 | } |
ebebbbe9 | 846 | if (!(si->flags & SWP_WRITEOK)) |
52b7efdb | 847 | goto no_page; |
7dfad418 HD |
848 | if (!si->highest_bit) |
849 | goto no_page; | |
ebebbbe9 | 850 | if (offset > si->highest_bit) |
c60aa176 | 851 | scan_base = offset = si->lowest_bit; |
c9e44410 | 852 | |
235b6217 | 853 | ci = lock_cluster(si, offset); |
b73d7fce HD |
854 | /* reuse swap entry of cache-only swap if not busy. */ |
855 | if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) { | |
c9e44410 | 856 | int swap_was_freed; |
235b6217 | 857 | unlock_cluster(ci); |
ec8acf20 | 858 | spin_unlock(&si->lock); |
bcd49e86 | 859 | swap_was_freed = __try_to_reclaim_swap(si, offset, TTRS_ANYWAY); |
ec8acf20 | 860 | spin_lock(&si->lock); |
c9e44410 KH |
861 | /* entry was freed successfully, try to use this again */ |
862 | if (swap_was_freed) | |
863 | goto checks; | |
864 | goto scan; /* check next one */ | |
865 | } | |
866 | ||
235b6217 HY |
867 | if (si->swap_map[offset]) { |
868 | unlock_cluster(ci); | |
36005bae TC |
869 | if (!n_ret) |
870 | goto scan; | |
871 | else | |
872 | goto done; | |
235b6217 | 873 | } |
a449bf58 | 874 | WRITE_ONCE(si->swap_map[offset], usage); |
2872bb2d HY |
875 | inc_cluster_info_page(si, si->cluster_info, offset); |
876 | unlock_cluster(ci); | |
ebebbbe9 | 877 | |
38d8b4e6 | 878 | swap_range_alloc(si, offset, 1); |
36005bae TC |
879 | slots[n_ret++] = swp_entry(si->type, offset); |
880 | ||
881 | /* got enough slots or reach max slots? */ | |
882 | if ((n_ret == nr) || (offset >= si->highest_bit)) | |
883 | goto done; | |
884 | ||
885 | /* search for next available slot */ | |
886 | ||
887 | /* time to take a break? */ | |
888 | if (unlikely(--latency_ration < 0)) { | |
889 | if (n_ret) | |
890 | goto done; | |
891 | spin_unlock(&si->lock); | |
892 | cond_resched(); | |
893 | spin_lock(&si->lock); | |
894 | latency_ration = LATENCY_LIMIT; | |
895 | } | |
896 | ||
897 | /* try to get more slots in cluster */ | |
898 | if (si->cluster_info) { | |
899 | if (scan_swap_map_try_ssd_cluster(si, &offset, &scan_base)) | |
900 | goto checks; | |
f4eaf51a WY |
901 | } else if (si->cluster_nr && !si->swap_map[++offset]) { |
902 | /* non-ssd case, still more slots in cluster? */ | |
36005bae TC |
903 | --si->cluster_nr; |
904 | goto checks; | |
905 | } | |
7992fde7 | 906 | |
ed43af10 HY |
907 | /* |
908 | * Even if there's no free clusters available (fragmented), | |
909 | * try to scan a little more quickly with lock held unless we | |
910 | * have scanned too many slots already. | |
911 | */ | |
912 | if (!scanned_many) { | |
913 | unsigned long scan_limit; | |
914 | ||
915 | if (offset < scan_base) | |
916 | scan_limit = scan_base; | |
917 | else | |
918 | scan_limit = si->highest_bit; | |
919 | for (; offset <= scan_limit && --latency_ration > 0; | |
920 | offset++) { | |
921 | if (!si->swap_map[offset]) | |
922 | goto checks; | |
923 | } | |
924 | } | |
925 | ||
36005bae | 926 | done: |
49070588 | 927 | set_cluster_next(si, offset + 1); |
36005bae TC |
928 | si->flags -= SWP_SCANNING; |
929 | return n_ret; | |
7dfad418 | 930 | |
ebebbbe9 | 931 | scan: |
ec8acf20 | 932 | spin_unlock(&si->lock); |
a449bf58 QC |
933 | while (++offset <= READ_ONCE(si->highest_bit)) { |
934 | if (data_race(!si->swap_map[offset])) { | |
ec8acf20 | 935 | spin_lock(&si->lock); |
52b7efdb HD |
936 | goto checks; |
937 | } | |
a449bf58 QC |
938 | if (vm_swap_full() && |
939 | READ_ONCE(si->swap_map[offset]) == SWAP_HAS_CACHE) { | |
ec8acf20 | 940 | spin_lock(&si->lock); |
c9e44410 KH |
941 | goto checks; |
942 | } | |
048c27fd HD |
943 | if (unlikely(--latency_ration < 0)) { |
944 | cond_resched(); | |
945 | latency_ration = LATENCY_LIMIT; | |
ed43af10 | 946 | scanned_many = true; |
048c27fd | 947 | } |
7dfad418 | 948 | } |
c60aa176 | 949 | offset = si->lowest_bit; |
a5998061 | 950 | while (offset < scan_base) { |
a449bf58 | 951 | if (data_race(!si->swap_map[offset])) { |
ec8acf20 | 952 | spin_lock(&si->lock); |
c60aa176 HD |
953 | goto checks; |
954 | } | |
a449bf58 QC |
955 | if (vm_swap_full() && |
956 | READ_ONCE(si->swap_map[offset]) == SWAP_HAS_CACHE) { | |
ec8acf20 | 957 | spin_lock(&si->lock); |
c9e44410 KH |
958 | goto checks; |
959 | } | |
c60aa176 HD |
960 | if (unlikely(--latency_ration < 0)) { |
961 | cond_resched(); | |
962 | latency_ration = LATENCY_LIMIT; | |
ed43af10 | 963 | scanned_many = true; |
c60aa176 | 964 | } |
a5998061 | 965 | offset++; |
c60aa176 | 966 | } |
ec8acf20 | 967 | spin_lock(&si->lock); |
7dfad418 HD |
968 | |
969 | no_page: | |
52b7efdb | 970 | si->flags -= SWP_SCANNING; |
36005bae | 971 | return n_ret; |
1da177e4 LT |
972 | } |
973 | ||
38d8b4e6 HY |
974 | static int swap_alloc_cluster(struct swap_info_struct *si, swp_entry_t *slot) |
975 | { | |
976 | unsigned long idx; | |
977 | struct swap_cluster_info *ci; | |
661c7566 | 978 | unsigned long offset; |
38d8b4e6 | 979 | |
fe5266d5 HY |
980 | /* |
981 | * Should not even be attempting cluster allocations when huge | |
982 | * page swap is disabled. Warn and fail the allocation. | |
983 | */ | |
984 | if (!IS_ENABLED(CONFIG_THP_SWAP)) { | |
985 | VM_WARN_ON_ONCE(1); | |
986 | return 0; | |
987 | } | |
988 | ||
38d8b4e6 HY |
989 | if (cluster_list_empty(&si->free_clusters)) |
990 | return 0; | |
991 | ||
992 | idx = cluster_list_first(&si->free_clusters); | |
993 | offset = idx * SWAPFILE_CLUSTER; | |
994 | ci = lock_cluster(si, offset); | |
995 | alloc_cluster(si, idx); | |
e0709829 | 996 | cluster_set_count_flag(ci, SWAPFILE_CLUSTER, CLUSTER_FLAG_HUGE); |
38d8b4e6 | 997 | |
661c7566 | 998 | memset(si->swap_map + offset, SWAP_HAS_CACHE, SWAPFILE_CLUSTER); |
38d8b4e6 HY |
999 | unlock_cluster(ci); |
1000 | swap_range_alloc(si, offset, SWAPFILE_CLUSTER); | |
1001 | *slot = swp_entry(si->type, offset); | |
1002 | ||
1003 | return 1; | |
1004 | } | |
1005 | ||
1006 | static void swap_free_cluster(struct swap_info_struct *si, unsigned long idx) | |
1007 | { | |
1008 | unsigned long offset = idx * SWAPFILE_CLUSTER; | |
1009 | struct swap_cluster_info *ci; | |
1010 | ||
1011 | ci = lock_cluster(si, offset); | |
979aafa5 | 1012 | memset(si->swap_map + offset, 0, SWAPFILE_CLUSTER); |
38d8b4e6 HY |
1013 | cluster_set_count_flag(ci, 0, 0); |
1014 | free_cluster(si, idx); | |
1015 | unlock_cluster(ci); | |
1016 | swap_range_free(si, offset, SWAPFILE_CLUSTER); | |
1017 | } | |
38d8b4e6 | 1018 | |
36005bae TC |
1019 | static unsigned long scan_swap_map(struct swap_info_struct *si, |
1020 | unsigned char usage) | |
1021 | { | |
1022 | swp_entry_t entry; | |
1023 | int n_ret; | |
1024 | ||
1025 | n_ret = scan_swap_map_slots(si, usage, 1, &entry); | |
1026 | ||
1027 | if (n_ret) | |
1028 | return swp_offset(entry); | |
1029 | else | |
1030 | return 0; | |
1031 | ||
1032 | } | |
1033 | ||
5d5e8f19 | 1034 | int get_swap_pages(int n_goal, swp_entry_t swp_entries[], int entry_size) |
1da177e4 | 1035 | { |
5d5e8f19 | 1036 | unsigned long size = swap_entry_size(entry_size); |
adfab836 | 1037 | struct swap_info_struct *si, *next; |
36005bae TC |
1038 | long avail_pgs; |
1039 | int n_ret = 0; | |
a2468cc9 | 1040 | int node; |
1da177e4 | 1041 | |
38d8b4e6 | 1042 | /* Only single cluster request supported */ |
5d5e8f19 | 1043 | WARN_ON_ONCE(n_goal > 1 && size == SWAPFILE_CLUSTER); |
38d8b4e6 | 1044 | |
b50da6e9 ZH |
1045 | spin_lock(&swap_avail_lock); |
1046 | ||
5d5e8f19 | 1047 | avail_pgs = atomic_long_read(&nr_swap_pages) / size; |
b50da6e9 ZH |
1048 | if (avail_pgs <= 0) { |
1049 | spin_unlock(&swap_avail_lock); | |
fb4f88dc | 1050 | goto noswap; |
b50da6e9 | 1051 | } |
36005bae | 1052 | |
08d3090f | 1053 | n_goal = min3((long)n_goal, (long)SWAP_BATCH, avail_pgs); |
36005bae | 1054 | |
5d5e8f19 | 1055 | atomic_long_sub(n_goal * size, &nr_swap_pages); |
fb4f88dc | 1056 | |
18ab4d4c | 1057 | start_over: |
a2468cc9 AL |
1058 | node = numa_node_id(); |
1059 | plist_for_each_entry_safe(si, next, &swap_avail_heads[node], avail_lists[node]) { | |
18ab4d4c | 1060 | /* requeue si to after same-priority siblings */ |
a2468cc9 | 1061 | plist_requeue(&si->avail_lists[node], &swap_avail_heads[node]); |
18ab4d4c | 1062 | spin_unlock(&swap_avail_lock); |
ec8acf20 | 1063 | spin_lock(&si->lock); |
adfab836 | 1064 | if (!si->highest_bit || !(si->flags & SWP_WRITEOK)) { |
18ab4d4c | 1065 | spin_lock(&swap_avail_lock); |
a2468cc9 | 1066 | if (plist_node_empty(&si->avail_lists[node])) { |
18ab4d4c DS |
1067 | spin_unlock(&si->lock); |
1068 | goto nextsi; | |
1069 | } | |
1070 | WARN(!si->highest_bit, | |
1071 | "swap_info %d in list but !highest_bit\n", | |
1072 | si->type); | |
1073 | WARN(!(si->flags & SWP_WRITEOK), | |
1074 | "swap_info %d in list but !SWP_WRITEOK\n", | |
1075 | si->type); | |
a2468cc9 | 1076 | __del_from_avail_list(si); |
ec8acf20 | 1077 | spin_unlock(&si->lock); |
18ab4d4c | 1078 | goto nextsi; |
ec8acf20 | 1079 | } |
5d5e8f19 | 1080 | if (size == SWAPFILE_CLUSTER) { |
41663430 | 1081 | if (si->flags & SWP_BLKDEV) |
f0eea189 HY |
1082 | n_ret = swap_alloc_cluster(si, swp_entries); |
1083 | } else | |
38d8b4e6 HY |
1084 | n_ret = scan_swap_map_slots(si, SWAP_HAS_CACHE, |
1085 | n_goal, swp_entries); | |
ec8acf20 | 1086 | spin_unlock(&si->lock); |
5d5e8f19 | 1087 | if (n_ret || size == SWAPFILE_CLUSTER) |
36005bae | 1088 | goto check_out; |
18ab4d4c | 1089 | pr_debug("scan_swap_map of si %d failed to find offset\n", |
36005bae TC |
1090 | si->type); |
1091 | ||
18ab4d4c DS |
1092 | spin_lock(&swap_avail_lock); |
1093 | nextsi: | |
adfab836 DS |
1094 | /* |
1095 | * if we got here, it's likely that si was almost full before, | |
1096 | * and since scan_swap_map() can drop the si->lock, multiple | |
1097 | * callers probably all tried to get a page from the same si | |
18ab4d4c DS |
1098 | * and it filled up before we could get one; or, the si filled |
1099 | * up between us dropping swap_avail_lock and taking si->lock. | |
1100 | * Since we dropped the swap_avail_lock, the swap_avail_head | |
1101 | * list may have been modified; so if next is still in the | |
36005bae TC |
1102 | * swap_avail_head list then try it, otherwise start over |
1103 | * if we have not gotten any slots. | |
adfab836 | 1104 | */ |
a2468cc9 | 1105 | if (plist_node_empty(&next->avail_lists[node])) |
18ab4d4c | 1106 | goto start_over; |
1da177e4 | 1107 | } |
fb4f88dc | 1108 | |
18ab4d4c DS |
1109 | spin_unlock(&swap_avail_lock); |
1110 | ||
36005bae TC |
1111 | check_out: |
1112 | if (n_ret < n_goal) | |
5d5e8f19 | 1113 | atomic_long_add((long)(n_goal - n_ret) * size, |
38d8b4e6 | 1114 | &nr_swap_pages); |
fb4f88dc | 1115 | noswap: |
36005bae TC |
1116 | return n_ret; |
1117 | } | |
1118 | ||
2de1a7e4 | 1119 | /* The only caller of this function is now suspend routine */ |
910321ea HD |
1120 | swp_entry_t get_swap_page_of_type(int type) |
1121 | { | |
c10d38cc | 1122 | struct swap_info_struct *si = swap_type_to_swap_info(type); |
910321ea HD |
1123 | pgoff_t offset; |
1124 | ||
c10d38cc DJ |
1125 | if (!si) |
1126 | goto fail; | |
1127 | ||
ec8acf20 | 1128 | spin_lock(&si->lock); |
c10d38cc | 1129 | if (si->flags & SWP_WRITEOK) { |
910321ea HD |
1130 | /* This is called for allocating swap entry, not cache */ |
1131 | offset = scan_swap_map(si, 1); | |
1132 | if (offset) { | |
b50da6e9 | 1133 | atomic_long_dec(&nr_swap_pages); |
ec8acf20 | 1134 | spin_unlock(&si->lock); |
910321ea HD |
1135 | return swp_entry(type, offset); |
1136 | } | |
910321ea | 1137 | } |
ec8acf20 | 1138 | spin_unlock(&si->lock); |
c10d38cc | 1139 | fail: |
910321ea HD |
1140 | return (swp_entry_t) {0}; |
1141 | } | |
1142 | ||
e8c26ab6 | 1143 | static struct swap_info_struct *__swap_info_get(swp_entry_t entry) |
1da177e4 | 1144 | { |
73c34b6a | 1145 | struct swap_info_struct *p; |
eb085574 | 1146 | unsigned long offset; |
1da177e4 LT |
1147 | |
1148 | if (!entry.val) | |
1149 | goto out; | |
eb085574 | 1150 | p = swp_swap_info(entry); |
c10d38cc | 1151 | if (!p) |
1da177e4 | 1152 | goto bad_nofile; |
a449bf58 | 1153 | if (data_race(!(p->flags & SWP_USED))) |
1da177e4 LT |
1154 | goto bad_device; |
1155 | offset = swp_offset(entry); | |
1156 | if (offset >= p->max) | |
1157 | goto bad_offset; | |
1da177e4 LT |
1158 | return p; |
1159 | ||
1da177e4 | 1160 | bad_offset: |
6a991fc7 | 1161 | pr_err("swap_info_get: %s%08lx\n", Bad_offset, entry.val); |
1da177e4 LT |
1162 | goto out; |
1163 | bad_device: | |
6a991fc7 | 1164 | pr_err("swap_info_get: %s%08lx\n", Unused_file, entry.val); |
1da177e4 LT |
1165 | goto out; |
1166 | bad_nofile: | |
6a991fc7 | 1167 | pr_err("swap_info_get: %s%08lx\n", Bad_file, entry.val); |
1da177e4 LT |
1168 | out: |
1169 | return NULL; | |
886bb7e9 | 1170 | } |
1da177e4 | 1171 | |
e8c26ab6 TC |
1172 | static struct swap_info_struct *_swap_info_get(swp_entry_t entry) |
1173 | { | |
1174 | struct swap_info_struct *p; | |
1175 | ||
1176 | p = __swap_info_get(entry); | |
1177 | if (!p) | |
1178 | goto out; | |
a449bf58 | 1179 | if (data_race(!p->swap_map[swp_offset(entry)])) |
e8c26ab6 TC |
1180 | goto bad_free; |
1181 | return p; | |
1182 | ||
1183 | bad_free: | |
1184 | pr_err("swap_info_get: %s%08lx\n", Unused_offset, entry.val); | |
e8c26ab6 TC |
1185 | out: |
1186 | return NULL; | |
1187 | } | |
1188 | ||
235b6217 HY |
1189 | static struct swap_info_struct *swap_info_get(swp_entry_t entry) |
1190 | { | |
1191 | struct swap_info_struct *p; | |
1192 | ||
1193 | p = _swap_info_get(entry); | |
1194 | if (p) | |
1195 | spin_lock(&p->lock); | |
1196 | return p; | |
1197 | } | |
1198 | ||
7c00bafe TC |
1199 | static struct swap_info_struct *swap_info_get_cont(swp_entry_t entry, |
1200 | struct swap_info_struct *q) | |
1201 | { | |
1202 | struct swap_info_struct *p; | |
1203 | ||
1204 | p = _swap_info_get(entry); | |
1205 | ||
1206 | if (p != q) { | |
1207 | if (q != NULL) | |
1208 | spin_unlock(&q->lock); | |
1209 | if (p != NULL) | |
1210 | spin_lock(&p->lock); | |
1211 | } | |
1212 | return p; | |
1213 | } | |
1214 | ||
b32d5f32 HY |
1215 | static unsigned char __swap_entry_free_locked(struct swap_info_struct *p, |
1216 | unsigned long offset, | |
1217 | unsigned char usage) | |
1da177e4 | 1218 | { |
8d69aaee HD |
1219 | unsigned char count; |
1220 | unsigned char has_cache; | |
235b6217 | 1221 | |
253d553b | 1222 | count = p->swap_map[offset]; |
235b6217 | 1223 | |
253d553b HD |
1224 | has_cache = count & SWAP_HAS_CACHE; |
1225 | count &= ~SWAP_HAS_CACHE; | |
355cfa73 | 1226 | |
253d553b | 1227 | if (usage == SWAP_HAS_CACHE) { |
355cfa73 | 1228 | VM_BUG_ON(!has_cache); |
253d553b | 1229 | has_cache = 0; |
aaa46865 HD |
1230 | } else if (count == SWAP_MAP_SHMEM) { |
1231 | /* | |
1232 | * Or we could insist on shmem.c using a special | |
1233 | * swap_shmem_free() and free_shmem_swap_and_cache()... | |
1234 | */ | |
1235 | count = 0; | |
570a335b HD |
1236 | } else if ((count & ~COUNT_CONTINUED) <= SWAP_MAP_MAX) { |
1237 | if (count == COUNT_CONTINUED) { | |
1238 | if (swap_count_continued(p, offset, count)) | |
1239 | count = SWAP_MAP_MAX | COUNT_CONTINUED; | |
1240 | else | |
1241 | count = SWAP_MAP_MAX; | |
1242 | } else | |
1243 | count--; | |
1244 | } | |
253d553b | 1245 | |
253d553b | 1246 | usage = count | has_cache; |
a449bf58 QC |
1247 | if (usage) |
1248 | WRITE_ONCE(p->swap_map[offset], usage); | |
1249 | else | |
1250 | WRITE_ONCE(p->swap_map[offset], SWAP_HAS_CACHE); | |
7c00bafe | 1251 | |
b32d5f32 HY |
1252 | return usage; |
1253 | } | |
1254 | ||
eb085574 HY |
1255 | /* |
1256 | * Check whether swap entry is valid in the swap device. If so, | |
1257 | * return pointer to swap_info_struct, and keep the swap entry valid | |
1258 | * via preventing the swap device from being swapoff, until | |
1259 | * put_swap_device() is called. Otherwise return NULL. | |
1260 | * | |
1261 | * The entirety of the RCU read critical section must come before the | |
1262 | * return from or after the call to synchronize_rcu() in | |
1263 | * enable_swap_info() or swapoff(). So if "si->flags & SWP_VALID" is | |
1264 | * true, the si->map, si->cluster_info, etc. must be valid in the | |
1265 | * critical section. | |
1266 | * | |
1267 | * Notice that swapoff or swapoff+swapon can still happen before the | |
1268 | * rcu_read_lock() in get_swap_device() or after the rcu_read_unlock() | |
1269 | * in put_swap_device() if there isn't any other way to prevent | |
1270 | * swapoff, such as page lock, page table lock, etc. The caller must | |
1271 | * be prepared for that. For example, the following situation is | |
1272 | * possible. | |
1273 | * | |
1274 | * CPU1 CPU2 | |
1275 | * do_swap_page() | |
1276 | * ... swapoff+swapon | |
1277 | * __read_swap_cache_async() | |
1278 | * swapcache_prepare() | |
1279 | * __swap_duplicate() | |
1280 | * // check swap_map | |
1281 | * // verify PTE not changed | |
1282 | * | |
1283 | * In __swap_duplicate(), the swap_map need to be checked before | |
1284 | * changing partly because the specified swap entry may be for another | |
1285 | * swap device which has been swapoff. And in do_swap_page(), after | |
1286 | * the page is read from the swap device, the PTE is verified not | |
1287 | * changed with the page table locked to check whether the swap device | |
1288 | * has been swapoff or swapoff+swapon. | |
1289 | */ | |
1290 | struct swap_info_struct *get_swap_device(swp_entry_t entry) | |
1291 | { | |
1292 | struct swap_info_struct *si; | |
1293 | unsigned long offset; | |
1294 | ||
1295 | if (!entry.val) | |
1296 | goto out; | |
1297 | si = swp_swap_info(entry); | |
1298 | if (!si) | |
1299 | goto bad_nofile; | |
1300 | ||
1301 | rcu_read_lock(); | |
a449bf58 | 1302 | if (data_race(!(si->flags & SWP_VALID))) |
eb085574 HY |
1303 | goto unlock_out; |
1304 | offset = swp_offset(entry); | |
1305 | if (offset >= si->max) | |
1306 | goto unlock_out; | |
1307 | ||
1308 | return si; | |
1309 | bad_nofile: | |
1310 | pr_err("%s: %s%08lx\n", __func__, Bad_file, entry.val); | |
1311 | out: | |
1312 | return NULL; | |
1313 | unlock_out: | |
1314 | rcu_read_unlock(); | |
1315 | return NULL; | |
1316 | } | |
1317 | ||
b32d5f32 | 1318 | static unsigned char __swap_entry_free(struct swap_info_struct *p, |
33e16272 | 1319 | swp_entry_t entry) |
b32d5f32 HY |
1320 | { |
1321 | struct swap_cluster_info *ci; | |
1322 | unsigned long offset = swp_offset(entry); | |
33e16272 | 1323 | unsigned char usage; |
b32d5f32 HY |
1324 | |
1325 | ci = lock_cluster_or_swap_info(p, offset); | |
33e16272 | 1326 | usage = __swap_entry_free_locked(p, offset, 1); |
7c00bafe | 1327 | unlock_cluster_or_swap_info(p, ci); |
10e364da HY |
1328 | if (!usage) |
1329 | free_swap_slot(entry); | |
7c00bafe TC |
1330 | |
1331 | return usage; | |
1332 | } | |
355cfa73 | 1333 | |
7c00bafe TC |
1334 | static void swap_entry_free(struct swap_info_struct *p, swp_entry_t entry) |
1335 | { | |
1336 | struct swap_cluster_info *ci; | |
1337 | unsigned long offset = swp_offset(entry); | |
1338 | unsigned char count; | |
1339 | ||
1340 | ci = lock_cluster(p, offset); | |
1341 | count = p->swap_map[offset]; | |
1342 | VM_BUG_ON(count != SWAP_HAS_CACHE); | |
1343 | p->swap_map[offset] = 0; | |
1344 | dec_cluster_info_page(p, p->cluster_info, offset); | |
235b6217 HY |
1345 | unlock_cluster(ci); |
1346 | ||
38d8b4e6 HY |
1347 | mem_cgroup_uncharge_swap(entry, 1); |
1348 | swap_range_free(p, offset, 1); | |
1da177e4 LT |
1349 | } |
1350 | ||
1351 | /* | |
2de1a7e4 | 1352 | * Caller has made sure that the swap device corresponding to entry |
1da177e4 LT |
1353 | * is still around or has not been recycled. |
1354 | */ | |
1355 | void swap_free(swp_entry_t entry) | |
1356 | { | |
73c34b6a | 1357 | struct swap_info_struct *p; |
1da177e4 | 1358 | |
235b6217 | 1359 | p = _swap_info_get(entry); |
10e364da | 1360 | if (p) |
33e16272 | 1361 | __swap_entry_free(p, entry); |
1da177e4 LT |
1362 | } |
1363 | ||
cb4b86ba KH |
1364 | /* |
1365 | * Called after dropping swapcache to decrease refcnt to swap entries. | |
1366 | */ | |
a448f2d0 | 1367 | void put_swap_page(struct page *page, swp_entry_t entry) |
38d8b4e6 HY |
1368 | { |
1369 | unsigned long offset = swp_offset(entry); | |
1370 | unsigned long idx = offset / SWAPFILE_CLUSTER; | |
1371 | struct swap_cluster_info *ci; | |
1372 | struct swap_info_struct *si; | |
1373 | unsigned char *map; | |
a3aea839 HY |
1374 | unsigned int i, free_entries = 0; |
1375 | unsigned char val; | |
6c357848 | 1376 | int size = swap_entry_size(thp_nr_pages(page)); |
fe5266d5 | 1377 | |
a3aea839 | 1378 | si = _swap_info_get(entry); |
38d8b4e6 HY |
1379 | if (!si) |
1380 | return; | |
1381 | ||
c2343d27 | 1382 | ci = lock_cluster_or_swap_info(si, offset); |
a448f2d0 | 1383 | if (size == SWAPFILE_CLUSTER) { |
a448f2d0 HY |
1384 | VM_BUG_ON(!cluster_is_huge(ci)); |
1385 | map = si->swap_map + offset; | |
1386 | for (i = 0; i < SWAPFILE_CLUSTER; i++) { | |
1387 | val = map[i]; | |
1388 | VM_BUG_ON(!(val & SWAP_HAS_CACHE)); | |
1389 | if (val == SWAP_HAS_CACHE) | |
1390 | free_entries++; | |
1391 | } | |
a448f2d0 | 1392 | cluster_clear_huge(ci); |
a448f2d0 | 1393 | if (free_entries == SWAPFILE_CLUSTER) { |
c2343d27 | 1394 | unlock_cluster_or_swap_info(si, ci); |
a448f2d0 | 1395 | spin_lock(&si->lock); |
a448f2d0 HY |
1396 | mem_cgroup_uncharge_swap(entry, SWAPFILE_CLUSTER); |
1397 | swap_free_cluster(si, idx); | |
1398 | spin_unlock(&si->lock); | |
1399 | return; | |
1400 | } | |
1401 | } | |
c2343d27 HY |
1402 | for (i = 0; i < size; i++, entry.val++) { |
1403 | if (!__swap_entry_free_locked(si, offset + i, SWAP_HAS_CACHE)) { | |
1404 | unlock_cluster_or_swap_info(si, ci); | |
1405 | free_swap_slot(entry); | |
1406 | if (i == size - 1) | |
1407 | return; | |
1408 | lock_cluster_or_swap_info(si, offset); | |
a3aea839 HY |
1409 | } |
1410 | } | |
c2343d27 | 1411 | unlock_cluster_or_swap_info(si, ci); |
38d8b4e6 | 1412 | } |
59807685 | 1413 | |
fe5266d5 | 1414 | #ifdef CONFIG_THP_SWAP |
59807685 HY |
1415 | int split_swap_cluster(swp_entry_t entry) |
1416 | { | |
1417 | struct swap_info_struct *si; | |
1418 | struct swap_cluster_info *ci; | |
1419 | unsigned long offset = swp_offset(entry); | |
1420 | ||
1421 | si = _swap_info_get(entry); | |
1422 | if (!si) | |
1423 | return -EBUSY; | |
1424 | ci = lock_cluster(si, offset); | |
1425 | cluster_clear_huge(ci); | |
1426 | unlock_cluster(ci); | |
1427 | return 0; | |
1428 | } | |
fe5266d5 | 1429 | #endif |
38d8b4e6 | 1430 | |
155b5f88 HY |
1431 | static int swp_entry_cmp(const void *ent1, const void *ent2) |
1432 | { | |
1433 | const swp_entry_t *e1 = ent1, *e2 = ent2; | |
1434 | ||
1435 | return (int)swp_type(*e1) - (int)swp_type(*e2); | |
1436 | } | |
1437 | ||
7c00bafe TC |
1438 | void swapcache_free_entries(swp_entry_t *entries, int n) |
1439 | { | |
1440 | struct swap_info_struct *p, *prev; | |
1441 | int i; | |
1442 | ||
1443 | if (n <= 0) | |
1444 | return; | |
1445 | ||
1446 | prev = NULL; | |
1447 | p = NULL; | |
155b5f88 HY |
1448 | |
1449 | /* | |
1450 | * Sort swap entries by swap device, so each lock is only taken once. | |
1451 | * nr_swapfiles isn't absolutely correct, but the overhead of sort() is | |
1452 | * so low that it isn't necessary to optimize further. | |
1453 | */ | |
1454 | if (nr_swapfiles > 1) | |
1455 | sort(entries, n, sizeof(entries[0]), swp_entry_cmp, NULL); | |
7c00bafe TC |
1456 | for (i = 0; i < n; ++i) { |
1457 | p = swap_info_get_cont(entries[i], prev); | |
1458 | if (p) | |
1459 | swap_entry_free(p, entries[i]); | |
7c00bafe TC |
1460 | prev = p; |
1461 | } | |
235b6217 | 1462 | if (p) |
7c00bafe | 1463 | spin_unlock(&p->lock); |
cb4b86ba KH |
1464 | } |
1465 | ||
1da177e4 | 1466 | /* |
c475a8ab | 1467 | * How many references to page are currently swapped out? |
570a335b HD |
1468 | * This does not give an exact answer when swap count is continued, |
1469 | * but does include the high COUNT_CONTINUED flag to allow for that. | |
1da177e4 | 1470 | */ |
bde05d1c | 1471 | int page_swapcount(struct page *page) |
1da177e4 | 1472 | { |
c475a8ab HD |
1473 | int count = 0; |
1474 | struct swap_info_struct *p; | |
235b6217 | 1475 | struct swap_cluster_info *ci; |
1da177e4 | 1476 | swp_entry_t entry; |
235b6217 | 1477 | unsigned long offset; |
1da177e4 | 1478 | |
4c21e2f2 | 1479 | entry.val = page_private(page); |
235b6217 | 1480 | p = _swap_info_get(entry); |
1da177e4 | 1481 | if (p) { |
235b6217 HY |
1482 | offset = swp_offset(entry); |
1483 | ci = lock_cluster_or_swap_info(p, offset); | |
1484 | count = swap_count(p->swap_map[offset]); | |
1485 | unlock_cluster_or_swap_info(p, ci); | |
1da177e4 | 1486 | } |
c475a8ab | 1487 | return count; |
1da177e4 LT |
1488 | } |
1489 | ||
eb085574 | 1490 | int __swap_count(swp_entry_t entry) |
aa8d22a1 | 1491 | { |
eb085574 | 1492 | struct swap_info_struct *si; |
aa8d22a1 | 1493 | pgoff_t offset = swp_offset(entry); |
eb085574 | 1494 | int count = 0; |
aa8d22a1 | 1495 | |
eb085574 HY |
1496 | si = get_swap_device(entry); |
1497 | if (si) { | |
1498 | count = swap_count(si->swap_map[offset]); | |
1499 | put_swap_device(si); | |
1500 | } | |
1501 | return count; | |
aa8d22a1 MK |
1502 | } |
1503 | ||
322b8afe HY |
1504 | static int swap_swapcount(struct swap_info_struct *si, swp_entry_t entry) |
1505 | { | |
1506 | int count = 0; | |
1507 | pgoff_t offset = swp_offset(entry); | |
1508 | struct swap_cluster_info *ci; | |
1509 | ||
1510 | ci = lock_cluster_or_swap_info(si, offset); | |
1511 | count = swap_count(si->swap_map[offset]); | |
1512 | unlock_cluster_or_swap_info(si, ci); | |
1513 | return count; | |
1514 | } | |
1515 | ||
e8c26ab6 TC |
1516 | /* |
1517 | * How many references to @entry are currently swapped out? | |
1518 | * This does not give an exact answer when swap count is continued, | |
1519 | * but does include the high COUNT_CONTINUED flag to allow for that. | |
1520 | */ | |
1521 | int __swp_swapcount(swp_entry_t entry) | |
1522 | { | |
1523 | int count = 0; | |
e8c26ab6 | 1524 | struct swap_info_struct *si; |
e8c26ab6 | 1525 | |
eb085574 HY |
1526 | si = get_swap_device(entry); |
1527 | if (si) { | |
322b8afe | 1528 | count = swap_swapcount(si, entry); |
eb085574 HY |
1529 | put_swap_device(si); |
1530 | } | |
e8c26ab6 TC |
1531 | return count; |
1532 | } | |
1533 | ||
8334b962 MK |
1534 | /* |
1535 | * How many references to @entry are currently swapped out? | |
1536 | * This considers COUNT_CONTINUED so it returns exact answer. | |
1537 | */ | |
1538 | int swp_swapcount(swp_entry_t entry) | |
1539 | { | |
1540 | int count, tmp_count, n; | |
1541 | struct swap_info_struct *p; | |
235b6217 | 1542 | struct swap_cluster_info *ci; |
8334b962 MK |
1543 | struct page *page; |
1544 | pgoff_t offset; | |
1545 | unsigned char *map; | |
1546 | ||
235b6217 | 1547 | p = _swap_info_get(entry); |
8334b962 MK |
1548 | if (!p) |
1549 | return 0; | |
1550 | ||
235b6217 HY |
1551 | offset = swp_offset(entry); |
1552 | ||
1553 | ci = lock_cluster_or_swap_info(p, offset); | |
1554 | ||
1555 | count = swap_count(p->swap_map[offset]); | |
8334b962 MK |
1556 | if (!(count & COUNT_CONTINUED)) |
1557 | goto out; | |
1558 | ||
1559 | count &= ~COUNT_CONTINUED; | |
1560 | n = SWAP_MAP_MAX + 1; | |
1561 | ||
8334b962 MK |
1562 | page = vmalloc_to_page(p->swap_map + offset); |
1563 | offset &= ~PAGE_MASK; | |
1564 | VM_BUG_ON(page_private(page) != SWP_CONTINUED); | |
1565 | ||
1566 | do { | |
a8ae4991 | 1567 | page = list_next_entry(page, lru); |
8334b962 MK |
1568 | map = kmap_atomic(page); |
1569 | tmp_count = map[offset]; | |
1570 | kunmap_atomic(map); | |
1571 | ||
1572 | count += (tmp_count & ~COUNT_CONTINUED) * n; | |
1573 | n *= (SWAP_CONT_MAX + 1); | |
1574 | } while (tmp_count & COUNT_CONTINUED); | |
1575 | out: | |
235b6217 | 1576 | unlock_cluster_or_swap_info(p, ci); |
8334b962 MK |
1577 | return count; |
1578 | } | |
1579 | ||
e0709829 HY |
1580 | static bool swap_page_trans_huge_swapped(struct swap_info_struct *si, |
1581 | swp_entry_t entry) | |
1582 | { | |
1583 | struct swap_cluster_info *ci; | |
1584 | unsigned char *map = si->swap_map; | |
1585 | unsigned long roffset = swp_offset(entry); | |
1586 | unsigned long offset = round_down(roffset, SWAPFILE_CLUSTER); | |
1587 | int i; | |
1588 | bool ret = false; | |
1589 | ||
1590 | ci = lock_cluster_or_swap_info(si, offset); | |
1591 | if (!ci || !cluster_is_huge(ci)) { | |
afa4711e | 1592 | if (swap_count(map[roffset])) |
e0709829 HY |
1593 | ret = true; |
1594 | goto unlock_out; | |
1595 | } | |
1596 | for (i = 0; i < SWAPFILE_CLUSTER; i++) { | |
afa4711e | 1597 | if (swap_count(map[offset + i])) { |
e0709829 HY |
1598 | ret = true; |
1599 | break; | |
1600 | } | |
1601 | } | |
1602 | unlock_out: | |
1603 | unlock_cluster_or_swap_info(si, ci); | |
1604 | return ret; | |
1605 | } | |
1606 | ||
1607 | static bool page_swapped(struct page *page) | |
1608 | { | |
1609 | swp_entry_t entry; | |
1610 | struct swap_info_struct *si; | |
1611 | ||
fe5266d5 | 1612 | if (!IS_ENABLED(CONFIG_THP_SWAP) || likely(!PageTransCompound(page))) |
e0709829 HY |
1613 | return page_swapcount(page) != 0; |
1614 | ||
1615 | page = compound_head(page); | |
1616 | entry.val = page_private(page); | |
1617 | si = _swap_info_get(entry); | |
1618 | if (si) | |
1619 | return swap_page_trans_huge_swapped(si, entry); | |
1620 | return false; | |
1621 | } | |
ba3c4ce6 HY |
1622 | |
1623 | static int page_trans_huge_map_swapcount(struct page *page, int *total_mapcount, | |
1624 | int *total_swapcount) | |
1625 | { | |
1626 | int i, map_swapcount, _total_mapcount, _total_swapcount; | |
1627 | unsigned long offset = 0; | |
1628 | struct swap_info_struct *si; | |
1629 | struct swap_cluster_info *ci = NULL; | |
1630 | unsigned char *map = NULL; | |
1631 | int mapcount, swapcount = 0; | |
1632 | ||
1633 | /* hugetlbfs shouldn't call it */ | |
1634 | VM_BUG_ON_PAGE(PageHuge(page), page); | |
1635 | ||
fe5266d5 HY |
1636 | if (!IS_ENABLED(CONFIG_THP_SWAP) || likely(!PageTransCompound(page))) { |
1637 | mapcount = page_trans_huge_mapcount(page, total_mapcount); | |
ba3c4ce6 HY |
1638 | if (PageSwapCache(page)) |
1639 | swapcount = page_swapcount(page); | |
1640 | if (total_swapcount) | |
1641 | *total_swapcount = swapcount; | |
1642 | return mapcount + swapcount; | |
1643 | } | |
1644 | ||
1645 | page = compound_head(page); | |
1646 | ||
1647 | _total_mapcount = _total_swapcount = map_swapcount = 0; | |
1648 | if (PageSwapCache(page)) { | |
1649 | swp_entry_t entry; | |
1650 | ||
1651 | entry.val = page_private(page); | |
1652 | si = _swap_info_get(entry); | |
1653 | if (si) { | |
1654 | map = si->swap_map; | |
1655 | offset = swp_offset(entry); | |
1656 | } | |
1657 | } | |
1658 | if (map) | |
1659 | ci = lock_cluster(si, offset); | |
1660 | for (i = 0; i < HPAGE_PMD_NR; i++) { | |
1661 | mapcount = atomic_read(&page[i]._mapcount) + 1; | |
1662 | _total_mapcount += mapcount; | |
1663 | if (map) { | |
1664 | swapcount = swap_count(map[offset + i]); | |
1665 | _total_swapcount += swapcount; | |
1666 | } | |
1667 | map_swapcount = max(map_swapcount, mapcount + swapcount); | |
1668 | } | |
1669 | unlock_cluster(ci); | |
1670 | if (PageDoubleMap(page)) { | |
1671 | map_swapcount -= 1; | |
1672 | _total_mapcount -= HPAGE_PMD_NR; | |
1673 | } | |
1674 | mapcount = compound_mapcount(page); | |
1675 | map_swapcount += mapcount; | |
1676 | _total_mapcount += mapcount; | |
1677 | if (total_mapcount) | |
1678 | *total_mapcount = _total_mapcount; | |
1679 | if (total_swapcount) | |
1680 | *total_swapcount = _total_swapcount; | |
1681 | ||
1682 | return map_swapcount; | |
1683 | } | |
e0709829 | 1684 | |
1da177e4 | 1685 | /* |
7b1fe597 HD |
1686 | * We can write to an anon page without COW if there are no other references |
1687 | * to it. And as a side-effect, free up its swap: because the old content | |
1688 | * on disk will never be read, and seeking back there to write new content | |
1689 | * later would only waste time away from clustering. | |
6d0a07ed | 1690 | * |
ba3c4ce6 | 1691 | * NOTE: total_map_swapcount should not be relied upon by the caller if |
6d0a07ed AA |
1692 | * reuse_swap_page() returns false, but it may be always overwritten |
1693 | * (see the other implementation for CONFIG_SWAP=n). | |
1da177e4 | 1694 | */ |
ba3c4ce6 | 1695 | bool reuse_swap_page(struct page *page, int *total_map_swapcount) |
1da177e4 | 1696 | { |
ba3c4ce6 | 1697 | int count, total_mapcount, total_swapcount; |
c475a8ab | 1698 | |
309381fe | 1699 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
5ad64688 | 1700 | if (unlikely(PageKsm(page))) |
6d0a07ed | 1701 | return false; |
ba3c4ce6 HY |
1702 | count = page_trans_huge_map_swapcount(page, &total_mapcount, |
1703 | &total_swapcount); | |
1704 | if (total_map_swapcount) | |
1705 | *total_map_swapcount = total_mapcount + total_swapcount; | |
1706 | if (count == 1 && PageSwapCache(page) && | |
1707 | (likely(!PageTransCompound(page)) || | |
1708 | /* The remaining swap count will be freed soon */ | |
1709 | total_swapcount == page_swapcount(page))) { | |
f0571429 | 1710 | if (!PageWriteback(page)) { |
ba3c4ce6 | 1711 | page = compound_head(page); |
7b1fe597 HD |
1712 | delete_from_swap_cache(page); |
1713 | SetPageDirty(page); | |
f0571429 MK |
1714 | } else { |
1715 | swp_entry_t entry; | |
1716 | struct swap_info_struct *p; | |
1717 | ||
1718 | entry.val = page_private(page); | |
1719 | p = swap_info_get(entry); | |
1720 | if (p->flags & SWP_STABLE_WRITES) { | |
1721 | spin_unlock(&p->lock); | |
1722 | return false; | |
1723 | } | |
1724 | spin_unlock(&p->lock); | |
7b1fe597 HD |
1725 | } |
1726 | } | |
ba3c4ce6 | 1727 | |
5ad64688 | 1728 | return count <= 1; |
1da177e4 LT |
1729 | } |
1730 | ||
1731 | /* | |
a2c43eed HD |
1732 | * If swap is getting full, or if there are no more mappings of this page, |
1733 | * then try_to_free_swap is called to free its swap space. | |
1da177e4 | 1734 | */ |
a2c43eed | 1735 | int try_to_free_swap(struct page *page) |
1da177e4 | 1736 | { |
309381fe | 1737 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
1da177e4 LT |
1738 | |
1739 | if (!PageSwapCache(page)) | |
1740 | return 0; | |
1741 | if (PageWriteback(page)) | |
1742 | return 0; | |
e0709829 | 1743 | if (page_swapped(page)) |
1da177e4 LT |
1744 | return 0; |
1745 | ||
b73d7fce HD |
1746 | /* |
1747 | * Once hibernation has begun to create its image of memory, | |
1748 | * there's a danger that one of the calls to try_to_free_swap() | |
1749 | * - most probably a call from __try_to_reclaim_swap() while | |
1750 | * hibernation is allocating its own swap pages for the image, | |
1751 | * but conceivably even a call from memory reclaim - will free | |
1752 | * the swap from a page which has already been recorded in the | |
1753 | * image as a clean swapcache page, and then reuse its swap for | |
1754 | * another page of the image. On waking from hibernation, the | |
1755 | * original page might be freed under memory pressure, then | |
1756 | * later read back in from swap, now with the wrong data. | |
1757 | * | |
2de1a7e4 | 1758 | * Hibernation suspends storage while it is writing the image |
f90ac398 | 1759 | * to disk so check that here. |
b73d7fce | 1760 | */ |
f90ac398 | 1761 | if (pm_suspended_storage()) |
b73d7fce HD |
1762 | return 0; |
1763 | ||
e0709829 | 1764 | page = compound_head(page); |
a2c43eed HD |
1765 | delete_from_swap_cache(page); |
1766 | SetPageDirty(page); | |
1767 | return 1; | |
68a22394 RR |
1768 | } |
1769 | ||
1da177e4 LT |
1770 | /* |
1771 | * Free the swap entry like above, but also try to | |
1772 | * free the page cache entry if it is the last user. | |
1773 | */ | |
2509ef26 | 1774 | int free_swap_and_cache(swp_entry_t entry) |
1da177e4 | 1775 | { |
2509ef26 | 1776 | struct swap_info_struct *p; |
7c00bafe | 1777 | unsigned char count; |
1da177e4 | 1778 | |
a7420aa5 | 1779 | if (non_swap_entry(entry)) |
2509ef26 | 1780 | return 1; |
0697212a | 1781 | |
7c00bafe | 1782 | p = _swap_info_get(entry); |
1da177e4 | 1783 | if (p) { |
33e16272 | 1784 | count = __swap_entry_free(p, entry); |
e0709829 | 1785 | if (count == SWAP_HAS_CACHE && |
bcd49e86 HY |
1786 | !swap_page_trans_huge_swapped(p, entry)) |
1787 | __try_to_reclaim_swap(p, swp_offset(entry), | |
1788 | TTRS_UNMAPPED | TTRS_FULL); | |
1da177e4 | 1789 | } |
2509ef26 | 1790 | return p != NULL; |
1da177e4 LT |
1791 | } |
1792 | ||
b0cb1a19 | 1793 | #ifdef CONFIG_HIBERNATION |
f577eb30 | 1794 | /* |
915bae9e | 1795 | * Find the swap type that corresponds to given device (if any). |
f577eb30 | 1796 | * |
915bae9e RW |
1797 | * @offset - number of the PAGE_SIZE-sized block of the device, starting |
1798 | * from 0, in which the swap header is expected to be located. | |
1799 | * | |
1800 | * This is needed for the suspend to disk (aka swsusp). | |
f577eb30 | 1801 | */ |
21bd9005 | 1802 | int swap_type_of(dev_t device, sector_t offset) |
f577eb30 | 1803 | { |
efa90a98 | 1804 | int type; |
f577eb30 | 1805 | |
21bd9005 CH |
1806 | if (!device) |
1807 | return -1; | |
915bae9e | 1808 | |
f577eb30 | 1809 | spin_lock(&swap_lock); |
efa90a98 HD |
1810 | for (type = 0; type < nr_swapfiles; type++) { |
1811 | struct swap_info_struct *sis = swap_info[type]; | |
f577eb30 | 1812 | |
915bae9e | 1813 | if (!(sis->flags & SWP_WRITEOK)) |
f577eb30 | 1814 | continue; |
b6b5bce3 | 1815 | |
21bd9005 | 1816 | if (device == sis->bdev->bd_dev) { |
4efaceb1 | 1817 | struct swap_extent *se = first_se(sis); |
915bae9e | 1818 | |
915bae9e RW |
1819 | if (se->start_block == offset) { |
1820 | spin_unlock(&swap_lock); | |
efa90a98 | 1821 | return type; |
915bae9e | 1822 | } |
f577eb30 RW |
1823 | } |
1824 | } | |
1825 | spin_unlock(&swap_lock); | |
21bd9005 CH |
1826 | return -ENODEV; |
1827 | } | |
915bae9e | 1828 | |
21bd9005 CH |
1829 | int find_first_swap(dev_t *device) |
1830 | { | |
1831 | int type; | |
915bae9e | 1832 | |
21bd9005 CH |
1833 | spin_lock(&swap_lock); |
1834 | for (type = 0; type < nr_swapfiles; type++) { | |
1835 | struct swap_info_struct *sis = swap_info[type]; | |
1836 | ||
1837 | if (!(sis->flags & SWP_WRITEOK)) | |
1838 | continue; | |
1839 | *device = sis->bdev->bd_dev; | |
1840 | spin_unlock(&swap_lock); | |
1841 | return type; | |
1842 | } | |
1843 | spin_unlock(&swap_lock); | |
f577eb30 RW |
1844 | return -ENODEV; |
1845 | } | |
1846 | ||
73c34b6a HD |
1847 | /* |
1848 | * Get the (PAGE_SIZE) block corresponding to given offset on the swapdev | |
1849 | * corresponding to given index in swap_info (swap type). | |
1850 | */ | |
1851 | sector_t swapdev_block(int type, pgoff_t offset) | |
1852 | { | |
1853 | struct block_device *bdev; | |
c10d38cc | 1854 | struct swap_info_struct *si = swap_type_to_swap_info(type); |
73c34b6a | 1855 | |
c10d38cc | 1856 | if (!si || !(si->flags & SWP_WRITEOK)) |
73c34b6a | 1857 | return 0; |
d4906e1a | 1858 | return map_swap_entry(swp_entry(type, offset), &bdev); |
73c34b6a HD |
1859 | } |
1860 | ||
f577eb30 RW |
1861 | /* |
1862 | * Return either the total number of swap pages of given type, or the number | |
1863 | * of free pages of that type (depending on @free) | |
1864 | * | |
1865 | * This is needed for software suspend | |
1866 | */ | |
1867 | unsigned int count_swap_pages(int type, int free) | |
1868 | { | |
1869 | unsigned int n = 0; | |
1870 | ||
efa90a98 HD |
1871 | spin_lock(&swap_lock); |
1872 | if ((unsigned int)type < nr_swapfiles) { | |
1873 | struct swap_info_struct *sis = swap_info[type]; | |
1874 | ||
ec8acf20 | 1875 | spin_lock(&sis->lock); |
efa90a98 HD |
1876 | if (sis->flags & SWP_WRITEOK) { |
1877 | n = sis->pages; | |
f577eb30 | 1878 | if (free) |
efa90a98 | 1879 | n -= sis->inuse_pages; |
f577eb30 | 1880 | } |
ec8acf20 | 1881 | spin_unlock(&sis->lock); |
f577eb30 | 1882 | } |
efa90a98 | 1883 | spin_unlock(&swap_lock); |
f577eb30 RW |
1884 | return n; |
1885 | } | |
73c34b6a | 1886 | #endif /* CONFIG_HIBERNATION */ |
f577eb30 | 1887 | |
9f8bdb3f | 1888 | static inline int pte_same_as_swp(pte_t pte, pte_t swp_pte) |
179ef71c | 1889 | { |
9f8bdb3f | 1890 | return pte_same(pte_swp_clear_soft_dirty(pte), swp_pte); |
179ef71c CG |
1891 | } |
1892 | ||
1da177e4 | 1893 | /* |
72866f6f HD |
1894 | * No need to decide whether this PTE shares the swap entry with others, |
1895 | * just let do_wp_page work it out if a write is requested later - to | |
1896 | * force COW, vm_page_prot omits write permission from any private vma. | |
1da177e4 | 1897 | */ |
044d66c1 | 1898 | static int unuse_pte(struct vm_area_struct *vma, pmd_t *pmd, |
1da177e4 LT |
1899 | unsigned long addr, swp_entry_t entry, struct page *page) |
1900 | { | |
9e16b7fb | 1901 | struct page *swapcache; |
044d66c1 HD |
1902 | spinlock_t *ptl; |
1903 | pte_t *pte; | |
1904 | int ret = 1; | |
1905 | ||
9e16b7fb HD |
1906 | swapcache = page; |
1907 | page = ksm_might_need_to_copy(page, vma, addr); | |
1908 | if (unlikely(!page)) | |
1909 | return -ENOMEM; | |
1910 | ||
044d66c1 | 1911 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
9f8bdb3f | 1912 | if (unlikely(!pte_same_as_swp(*pte, swp_entry_to_pte(entry)))) { |
044d66c1 HD |
1913 | ret = 0; |
1914 | goto out; | |
1915 | } | |
8a9f3ccd | 1916 | |
b084d435 | 1917 | dec_mm_counter(vma->vm_mm, MM_SWAPENTS); |
d559db08 | 1918 | inc_mm_counter(vma->vm_mm, MM_ANONPAGES); |
1da177e4 LT |
1919 | get_page(page); |
1920 | set_pte_at(vma->vm_mm, addr, pte, | |
1921 | pte_mkold(mk_pte(page, vma->vm_page_prot))); | |
00501b53 | 1922 | if (page == swapcache) { |
be5d0a74 | 1923 | page_add_anon_rmap(page, vma, addr, false); |
00501b53 | 1924 | } else { /* ksm created a completely new copy */ |
be5d0a74 | 1925 | page_add_new_anon_rmap(page, vma, addr, false); |
b518154e | 1926 | lru_cache_add_inactive_or_unevictable(page, vma); |
00501b53 | 1927 | } |
1da177e4 | 1928 | swap_free(entry); |
044d66c1 HD |
1929 | out: |
1930 | pte_unmap_unlock(pte, ptl); | |
9e16b7fb HD |
1931 | if (page != swapcache) { |
1932 | unlock_page(page); | |
1933 | put_page(page); | |
1934 | } | |
044d66c1 | 1935 | return ret; |
1da177e4 LT |
1936 | } |
1937 | ||
1938 | static int unuse_pte_range(struct vm_area_struct *vma, pmd_t *pmd, | |
b56a2d8a VRP |
1939 | unsigned long addr, unsigned long end, |
1940 | unsigned int type, bool frontswap, | |
1941 | unsigned long *fs_pages_to_unuse) | |
1da177e4 | 1942 | { |
b56a2d8a VRP |
1943 | struct page *page; |
1944 | swp_entry_t entry; | |
705e87c0 | 1945 | pte_t *pte; |
b56a2d8a VRP |
1946 | struct swap_info_struct *si; |
1947 | unsigned long offset; | |
8a9f3ccd | 1948 | int ret = 0; |
b56a2d8a | 1949 | volatile unsigned char *swap_map; |
1da177e4 | 1950 | |
b56a2d8a | 1951 | si = swap_info[type]; |
044d66c1 | 1952 | pte = pte_offset_map(pmd, addr); |
1da177e4 | 1953 | do { |
b56a2d8a VRP |
1954 | struct vm_fault vmf; |
1955 | ||
1956 | if (!is_swap_pte(*pte)) | |
1957 | continue; | |
1958 | ||
1959 | entry = pte_to_swp_entry(*pte); | |
1960 | if (swp_type(entry) != type) | |
1961 | continue; | |
1962 | ||
1963 | offset = swp_offset(entry); | |
1964 | if (frontswap && !frontswap_test(si, offset)) | |
1965 | continue; | |
1966 | ||
1967 | pte_unmap(pte); | |
1968 | swap_map = &si->swap_map[offset]; | |
ebc5951e AR |
1969 | page = lookup_swap_cache(entry, vma, addr); |
1970 | if (!page) { | |
1971 | vmf.vma = vma; | |
1972 | vmf.address = addr; | |
1973 | vmf.pmd = pmd; | |
1974 | page = swapin_readahead(entry, GFP_HIGHUSER_MOVABLE, | |
1975 | &vmf); | |
1976 | } | |
b56a2d8a VRP |
1977 | if (!page) { |
1978 | if (*swap_map == 0 || *swap_map == SWAP_MAP_BAD) | |
1979 | goto try_next; | |
1980 | return -ENOMEM; | |
1981 | } | |
1982 | ||
1983 | lock_page(page); | |
1984 | wait_on_page_writeback(page); | |
1985 | ret = unuse_pte(vma, pmd, addr, entry, page); | |
1986 | if (ret < 0) { | |
1987 | unlock_page(page); | |
1988 | put_page(page); | |
1989 | goto out; | |
1990 | } | |
1991 | ||
1992 | try_to_free_swap(page); | |
1993 | unlock_page(page); | |
1994 | put_page(page); | |
1995 | ||
1996 | if (*fs_pages_to_unuse && !--(*fs_pages_to_unuse)) { | |
1997 | ret = FRONTSWAP_PAGES_UNUSED; | |
1998 | goto out; | |
1da177e4 | 1999 | } |
b56a2d8a VRP |
2000 | try_next: |
2001 | pte = pte_offset_map(pmd, addr); | |
1da177e4 | 2002 | } while (pte++, addr += PAGE_SIZE, addr != end); |
044d66c1 | 2003 | pte_unmap(pte - 1); |
b56a2d8a VRP |
2004 | |
2005 | ret = 0; | |
044d66c1 | 2006 | out: |
8a9f3ccd | 2007 | return ret; |
1da177e4 LT |
2008 | } |
2009 | ||
2010 | static inline int unuse_pmd_range(struct vm_area_struct *vma, pud_t *pud, | |
2011 | unsigned long addr, unsigned long end, | |
b56a2d8a VRP |
2012 | unsigned int type, bool frontswap, |
2013 | unsigned long *fs_pages_to_unuse) | |
1da177e4 LT |
2014 | { |
2015 | pmd_t *pmd; | |
2016 | unsigned long next; | |
8a9f3ccd | 2017 | int ret; |
1da177e4 LT |
2018 | |
2019 | pmd = pmd_offset(pud, addr); | |
2020 | do { | |
dc644a07 | 2021 | cond_resched(); |
1da177e4 | 2022 | next = pmd_addr_end(addr, end); |
1a5a9906 | 2023 | if (pmd_none_or_trans_huge_or_clear_bad(pmd)) |
1da177e4 | 2024 | continue; |
b56a2d8a VRP |
2025 | ret = unuse_pte_range(vma, pmd, addr, next, type, |
2026 | frontswap, fs_pages_to_unuse); | |
8a9f3ccd BS |
2027 | if (ret) |
2028 | return ret; | |
1da177e4 LT |
2029 | } while (pmd++, addr = next, addr != end); |
2030 | return 0; | |
2031 | } | |
2032 | ||
c2febafc | 2033 | static inline int unuse_pud_range(struct vm_area_struct *vma, p4d_t *p4d, |
1da177e4 | 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 | pud_t *pud; | |
2039 | unsigned long next; | |
8a9f3ccd | 2040 | int ret; |
1da177e4 | 2041 | |
c2febafc | 2042 | pud = pud_offset(p4d, addr); |
1da177e4 LT |
2043 | do { |
2044 | next = pud_addr_end(addr, end); | |
2045 | if (pud_none_or_clear_bad(pud)) | |
2046 | continue; | |
b56a2d8a VRP |
2047 | ret = unuse_pmd_range(vma, pud, addr, next, type, |
2048 | frontswap, fs_pages_to_unuse); | |
8a9f3ccd BS |
2049 | if (ret) |
2050 | return ret; | |
1da177e4 LT |
2051 | } while (pud++, addr = next, addr != end); |
2052 | return 0; | |
2053 | } | |
2054 | ||
c2febafc KS |
2055 | static inline int unuse_p4d_range(struct vm_area_struct *vma, pgd_t *pgd, |
2056 | unsigned long addr, unsigned long end, | |
b56a2d8a VRP |
2057 | unsigned int type, bool frontswap, |
2058 | unsigned long *fs_pages_to_unuse) | |
c2febafc KS |
2059 | { |
2060 | p4d_t *p4d; | |
2061 | unsigned long next; | |
2062 | int ret; | |
2063 | ||
2064 | p4d = p4d_offset(pgd, addr); | |
2065 | do { | |
2066 | next = p4d_addr_end(addr, end); | |
2067 | if (p4d_none_or_clear_bad(p4d)) | |
2068 | continue; | |
b56a2d8a VRP |
2069 | ret = unuse_pud_range(vma, p4d, addr, next, type, |
2070 | frontswap, fs_pages_to_unuse); | |
c2febafc KS |
2071 | if (ret) |
2072 | return ret; | |
2073 | } while (p4d++, addr = next, addr != end); | |
2074 | return 0; | |
2075 | } | |
2076 | ||
b56a2d8a VRP |
2077 | static int unuse_vma(struct vm_area_struct *vma, unsigned int type, |
2078 | bool frontswap, unsigned long *fs_pages_to_unuse) | |
1da177e4 LT |
2079 | { |
2080 | pgd_t *pgd; | |
2081 | unsigned long addr, end, next; | |
8a9f3ccd | 2082 | int ret; |
1da177e4 | 2083 | |
b56a2d8a VRP |
2084 | addr = vma->vm_start; |
2085 | end = vma->vm_end; | |
1da177e4 LT |
2086 | |
2087 | pgd = pgd_offset(vma->vm_mm, addr); | |
2088 | do { | |
2089 | next = pgd_addr_end(addr, end); | |
2090 | if (pgd_none_or_clear_bad(pgd)) | |
2091 | continue; | |
b56a2d8a VRP |
2092 | ret = unuse_p4d_range(vma, pgd, addr, next, type, |
2093 | frontswap, fs_pages_to_unuse); | |
8a9f3ccd BS |
2094 | if (ret) |
2095 | return ret; | |
1da177e4 LT |
2096 | } while (pgd++, addr = next, addr != end); |
2097 | return 0; | |
2098 | } | |
2099 | ||
b56a2d8a VRP |
2100 | static int unuse_mm(struct mm_struct *mm, unsigned int type, |
2101 | bool frontswap, unsigned long *fs_pages_to_unuse) | |
1da177e4 LT |
2102 | { |
2103 | struct vm_area_struct *vma; | |
8a9f3ccd | 2104 | int ret = 0; |
1da177e4 | 2105 | |
d8ed45c5 | 2106 | mmap_read_lock(mm); |
1da177e4 | 2107 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
b56a2d8a VRP |
2108 | if (vma->anon_vma) { |
2109 | ret = unuse_vma(vma, type, frontswap, | |
2110 | fs_pages_to_unuse); | |
2111 | if (ret) | |
2112 | break; | |
2113 | } | |
dc644a07 | 2114 | cond_resched(); |
1da177e4 | 2115 | } |
d8ed45c5 | 2116 | mmap_read_unlock(mm); |
b56a2d8a | 2117 | return ret; |
1da177e4 LT |
2118 | } |
2119 | ||
2120 | /* | |
38b5faf4 | 2121 | * Scan swap_map (or frontswap_map if frontswap parameter is true) |
b56a2d8a VRP |
2122 | * from current position to next entry still in use. Return 0 |
2123 | * if there are no inuse entries after prev till end of the map. | |
1da177e4 | 2124 | */ |
6eb396dc | 2125 | static unsigned int find_next_to_unuse(struct swap_info_struct *si, |
38b5faf4 | 2126 | unsigned int prev, bool frontswap) |
1da177e4 | 2127 | { |
b56a2d8a | 2128 | unsigned int i; |
8d69aaee | 2129 | unsigned char count; |
1da177e4 LT |
2130 | |
2131 | /* | |
5d337b91 | 2132 | * No need for swap_lock here: we're just looking |
1da177e4 LT |
2133 | * for whether an entry is in use, not modifying it; false |
2134 | * hits are okay, and sys_swapoff() has already prevented new | |
5d337b91 | 2135 | * allocations from this area (while holding swap_lock). |
1da177e4 | 2136 | */ |
b56a2d8a | 2137 | for (i = prev + 1; i < si->max; i++) { |
4db0c3c2 | 2138 | count = READ_ONCE(si->swap_map[i]); |
355cfa73 | 2139 | if (count && swap_count(count) != SWAP_MAP_BAD) |
dc644a07 HD |
2140 | if (!frontswap || frontswap_test(si, i)) |
2141 | break; | |
2142 | if ((i % LATENCY_LIMIT) == 0) | |
2143 | cond_resched(); | |
1da177e4 | 2144 | } |
b56a2d8a VRP |
2145 | |
2146 | if (i == si->max) | |
2147 | i = 0; | |
2148 | ||
1da177e4 LT |
2149 | return i; |
2150 | } | |
2151 | ||
2152 | /* | |
b56a2d8a | 2153 | * If the boolean frontswap is true, only unuse pages_to_unuse pages; |
38b5faf4 | 2154 | * pages_to_unuse==0 means all pages; ignored if frontswap is false |
1da177e4 | 2155 | */ |
38b5faf4 DM |
2156 | int try_to_unuse(unsigned int type, bool frontswap, |
2157 | unsigned long pages_to_unuse) | |
1da177e4 | 2158 | { |
b56a2d8a VRP |
2159 | struct mm_struct *prev_mm; |
2160 | struct mm_struct *mm; | |
2161 | struct list_head *p; | |
2162 | int retval = 0; | |
efa90a98 | 2163 | struct swap_info_struct *si = swap_info[type]; |
1da177e4 LT |
2164 | struct page *page; |
2165 | swp_entry_t entry; | |
b56a2d8a | 2166 | unsigned int i; |
1da177e4 | 2167 | |
21820948 | 2168 | if (!READ_ONCE(si->inuse_pages)) |
b56a2d8a | 2169 | return 0; |
1da177e4 | 2170 | |
b56a2d8a VRP |
2171 | if (!frontswap) |
2172 | pages_to_unuse = 0; | |
2173 | ||
2174 | retry: | |
2175 | retval = shmem_unuse(type, frontswap, &pages_to_unuse); | |
2176 | if (retval) | |
2177 | goto out; | |
2178 | ||
2179 | prev_mm = &init_mm; | |
2180 | mmget(prev_mm); | |
2181 | ||
2182 | spin_lock(&mmlist_lock); | |
2183 | p = &init_mm.mmlist; | |
21820948 | 2184 | while (READ_ONCE(si->inuse_pages) && |
64165b1a HD |
2185 | !signal_pending(current) && |
2186 | (p = p->next) != &init_mm.mmlist) { | |
1da177e4 | 2187 | |
b56a2d8a VRP |
2188 | mm = list_entry(p, struct mm_struct, mmlist); |
2189 | if (!mmget_not_zero(mm)) | |
2190 | continue; | |
2191 | spin_unlock(&mmlist_lock); | |
2192 | mmput(prev_mm); | |
2193 | prev_mm = mm; | |
2194 | retval = unuse_mm(mm, type, frontswap, &pages_to_unuse); | |
1da177e4 | 2195 | |
b56a2d8a VRP |
2196 | if (retval) { |
2197 | mmput(prev_mm); | |
2198 | goto out; | |
1da177e4 LT |
2199 | } |
2200 | ||
2201 | /* | |
b56a2d8a VRP |
2202 | * Make sure that we aren't completely killing |
2203 | * interactive performance. | |
1da177e4 | 2204 | */ |
b56a2d8a VRP |
2205 | cond_resched(); |
2206 | spin_lock(&mmlist_lock); | |
2207 | } | |
2208 | spin_unlock(&mmlist_lock); | |
1da177e4 | 2209 | |
b56a2d8a | 2210 | mmput(prev_mm); |
1da177e4 | 2211 | |
b56a2d8a | 2212 | i = 0; |
21820948 | 2213 | while (READ_ONCE(si->inuse_pages) && |
64165b1a HD |
2214 | !signal_pending(current) && |
2215 | (i = find_next_to_unuse(si, i, frontswap)) != 0) { | |
1da177e4 | 2216 | |
b56a2d8a VRP |
2217 | entry = swp_entry(type, i); |
2218 | page = find_get_page(swap_address_space(entry), i); | |
2219 | if (!page) | |
2220 | continue; | |
68bdc8d6 HD |
2221 | |
2222 | /* | |
2223 | * It is conceivable that a racing task removed this page from | |
b56a2d8a VRP |
2224 | * swap cache just before we acquired the page lock. The page |
2225 | * might even be back in swap cache on another swap area. But | |
2226 | * that is okay, try_to_free_swap() only removes stale pages. | |
1da177e4 | 2227 | */ |
b56a2d8a VRP |
2228 | lock_page(page); |
2229 | wait_on_page_writeback(page); | |
2230 | try_to_free_swap(page); | |
1da177e4 | 2231 | unlock_page(page); |
09cbfeaf | 2232 | put_page(page); |
1da177e4 LT |
2233 | |
2234 | /* | |
b56a2d8a VRP |
2235 | * For frontswap, we just need to unuse pages_to_unuse, if |
2236 | * it was specified. Need not check frontswap again here as | |
2237 | * we already zeroed out pages_to_unuse if not frontswap. | |
1da177e4 | 2238 | */ |
b56a2d8a VRP |
2239 | if (pages_to_unuse && --pages_to_unuse == 0) |
2240 | goto out; | |
1da177e4 LT |
2241 | } |
2242 | ||
b56a2d8a VRP |
2243 | /* |
2244 | * Lets check again to see if there are still swap entries in the map. | |
2245 | * If yes, we would need to do retry the unuse logic again. | |
2246 | * Under global memory pressure, swap entries can be reinserted back | |
2247 | * into process space after the mmlist loop above passes over them. | |
dd862deb | 2248 | * |
af53d3e9 HD |
2249 | * Limit the number of retries? No: when mmget_not_zero() above fails, |
2250 | * that mm is likely to be freeing swap from exit_mmap(), which proceeds | |
2251 | * at its own independent pace; and even shmem_writepage() could have | |
2252 | * been preempted after get_swap_page(), temporarily hiding that swap. | |
2253 | * It's easy and robust (though cpu-intensive) just to keep retrying. | |
b56a2d8a | 2254 | */ |
21820948 | 2255 | if (READ_ONCE(si->inuse_pages)) { |
64165b1a HD |
2256 | if (!signal_pending(current)) |
2257 | goto retry; | |
2258 | retval = -EINTR; | |
2259 | } | |
b56a2d8a VRP |
2260 | out: |
2261 | return (retval == FRONTSWAP_PAGES_UNUSED) ? 0 : retval; | |
1da177e4 LT |
2262 | } |
2263 | ||
2264 | /* | |
5d337b91 HD |
2265 | * After a successful try_to_unuse, if no swap is now in use, we know |
2266 | * we can empty the mmlist. swap_lock must be held on entry and exit. | |
2267 | * Note that mmlist_lock nests inside swap_lock, and an mm must be | |
1da177e4 LT |
2268 | * added to the mmlist just after page_duplicate - before would be racy. |
2269 | */ | |
2270 | static void drain_mmlist(void) | |
2271 | { | |
2272 | struct list_head *p, *next; | |
efa90a98 | 2273 | unsigned int type; |
1da177e4 | 2274 | |
efa90a98 HD |
2275 | for (type = 0; type < nr_swapfiles; type++) |
2276 | if (swap_info[type]->inuse_pages) | |
1da177e4 LT |
2277 | return; |
2278 | spin_lock(&mmlist_lock); | |
2279 | list_for_each_safe(p, next, &init_mm.mmlist) | |
2280 | list_del_init(p); | |
2281 | spin_unlock(&mmlist_lock); | |
2282 | } | |
2283 | ||
2284 | /* | |
2285 | * Use this swapdev's extent info to locate the (PAGE_SIZE) block which | |
d4906e1a LS |
2286 | * corresponds to page offset for the specified swap entry. |
2287 | * Note that the type of this function is sector_t, but it returns page offset | |
2288 | * into the bdev, not sector offset. | |
1da177e4 | 2289 | */ |
d4906e1a | 2290 | static sector_t map_swap_entry(swp_entry_t entry, struct block_device **bdev) |
1da177e4 | 2291 | { |
f29ad6a9 | 2292 | struct swap_info_struct *sis; |
f29ad6a9 HD |
2293 | struct swap_extent *se; |
2294 | pgoff_t offset; | |
2295 | ||
c10d38cc | 2296 | sis = swp_swap_info(entry); |
f29ad6a9 HD |
2297 | *bdev = sis->bdev; |
2298 | ||
2299 | offset = swp_offset(entry); | |
4efaceb1 AL |
2300 | se = offset_to_swap_extent(sis, offset); |
2301 | return se->start_block + (offset - se->start_page); | |
1da177e4 LT |
2302 | } |
2303 | ||
d4906e1a LS |
2304 | /* |
2305 | * Returns the page offset into bdev for the specified page's swap entry. | |
2306 | */ | |
2307 | sector_t map_swap_page(struct page *page, struct block_device **bdev) | |
2308 | { | |
2309 | swp_entry_t entry; | |
2310 | entry.val = page_private(page); | |
2311 | return map_swap_entry(entry, bdev); | |
2312 | } | |
2313 | ||
1da177e4 LT |
2314 | /* |
2315 | * Free all of a swapdev's extent information | |
2316 | */ | |
2317 | static void destroy_swap_extents(struct swap_info_struct *sis) | |
2318 | { | |
4efaceb1 AL |
2319 | while (!RB_EMPTY_ROOT(&sis->swap_extent_root)) { |
2320 | struct rb_node *rb = sis->swap_extent_root.rb_node; | |
2321 | struct swap_extent *se = rb_entry(rb, struct swap_extent, rb_node); | |
1da177e4 | 2322 | |
4efaceb1 | 2323 | rb_erase(rb, &sis->swap_extent_root); |
1da177e4 LT |
2324 | kfree(se); |
2325 | } | |
62c230bc | 2326 | |
bc4ae27d | 2327 | if (sis->flags & SWP_ACTIVATED) { |
62c230bc MG |
2328 | struct file *swap_file = sis->swap_file; |
2329 | struct address_space *mapping = swap_file->f_mapping; | |
2330 | ||
bc4ae27d OS |
2331 | sis->flags &= ~SWP_ACTIVATED; |
2332 | if (mapping->a_ops->swap_deactivate) | |
2333 | mapping->a_ops->swap_deactivate(swap_file); | |
62c230bc | 2334 | } |
1da177e4 LT |
2335 | } |
2336 | ||
2337 | /* | |
2338 | * Add a block range (and the corresponding page range) into this swapdev's | |
4efaceb1 | 2339 | * extent tree. |
1da177e4 | 2340 | * |
11d31886 | 2341 | * This function rather assumes that it is called in ascending page order. |
1da177e4 | 2342 | */ |
a509bc1a | 2343 | int |
1da177e4 LT |
2344 | add_swap_extent(struct swap_info_struct *sis, unsigned long start_page, |
2345 | unsigned long nr_pages, sector_t start_block) | |
2346 | { | |
4efaceb1 | 2347 | struct rb_node **link = &sis->swap_extent_root.rb_node, *parent = NULL; |
1da177e4 LT |
2348 | struct swap_extent *se; |
2349 | struct swap_extent *new_se; | |
4efaceb1 AL |
2350 | |
2351 | /* | |
2352 | * place the new node at the right most since the | |
2353 | * function is called in ascending page order. | |
2354 | */ | |
2355 | while (*link) { | |
2356 | parent = *link; | |
2357 | link = &parent->rb_right; | |
2358 | } | |
2359 | ||
2360 | if (parent) { | |
2361 | se = rb_entry(parent, struct swap_extent, rb_node); | |
11d31886 HD |
2362 | BUG_ON(se->start_page + se->nr_pages != start_page); |
2363 | if (se->start_block + se->nr_pages == start_block) { | |
1da177e4 LT |
2364 | /* Merge it */ |
2365 | se->nr_pages += nr_pages; | |
2366 | return 0; | |
2367 | } | |
1da177e4 LT |
2368 | } |
2369 | ||
4efaceb1 | 2370 | /* No merge, insert a new extent. */ |
1da177e4 LT |
2371 | new_se = kmalloc(sizeof(*se), GFP_KERNEL); |
2372 | if (new_se == NULL) | |
2373 | return -ENOMEM; | |
2374 | new_se->start_page = start_page; | |
2375 | new_se->nr_pages = nr_pages; | |
2376 | new_se->start_block = start_block; | |
2377 | ||
4efaceb1 AL |
2378 | rb_link_node(&new_se->rb_node, parent, link); |
2379 | rb_insert_color(&new_se->rb_node, &sis->swap_extent_root); | |
53092a74 | 2380 | return 1; |
1da177e4 | 2381 | } |
aa8aa8a3 | 2382 | EXPORT_SYMBOL_GPL(add_swap_extent); |
1da177e4 LT |
2383 | |
2384 | /* | |
2385 | * A `swap extent' is a simple thing which maps a contiguous range of pages | |
2386 | * onto a contiguous range of disk blocks. An ordered list of swap extents | |
2387 | * is built at swapon time and is then used at swap_writepage/swap_readpage | |
2388 | * time for locating where on disk a page belongs. | |
2389 | * | |
2390 | * If the swapfile is an S_ISBLK block device, a single extent is installed. | |
2391 | * This is done so that the main operating code can treat S_ISBLK and S_ISREG | |
2392 | * swap files identically. | |
2393 | * | |
2394 | * Whether the swapdev is an S_ISREG file or an S_ISBLK blockdev, the swap | |
2395 | * extent list operates in PAGE_SIZE disk blocks. Both S_ISREG and S_ISBLK | |
2396 | * swapfiles are handled *identically* after swapon time. | |
2397 | * | |
2398 | * For S_ISREG swapfiles, setup_swap_extents() will walk all the file's blocks | |
2399 | * and will parse them into an ordered extent list, in PAGE_SIZE chunks. If | |
2400 | * some stray blocks are found which do not fall within the PAGE_SIZE alignment | |
2401 | * requirements, they are simply tossed out - we will never use those blocks | |
2402 | * for swapping. | |
2403 | * | |
1638045c DW |
2404 | * For all swap devices we set S_SWAPFILE across the life of the swapon. This |
2405 | * prevents users from writing to the swap device, which will corrupt memory. | |
1da177e4 LT |
2406 | * |
2407 | * The amount of disk space which a single swap extent represents varies. | |
2408 | * Typically it is in the 1-4 megabyte range. So we can have hundreds of | |
2409 | * extents in the list. To avoid much list walking, we cache the previous | |
2410 | * search location in `curr_swap_extent', and start new searches from there. | |
2411 | * This is extremely effective. The average number of iterations in | |
2412 | * map_swap_page() has been measured at about 0.3 per page. - akpm. | |
2413 | */ | |
53092a74 | 2414 | static int setup_swap_extents(struct swap_info_struct *sis, sector_t *span) |
1da177e4 | 2415 | { |
62c230bc MG |
2416 | struct file *swap_file = sis->swap_file; |
2417 | struct address_space *mapping = swap_file->f_mapping; | |
2418 | struct inode *inode = mapping->host; | |
1da177e4 LT |
2419 | int ret; |
2420 | ||
1da177e4 LT |
2421 | if (S_ISBLK(inode->i_mode)) { |
2422 | ret = add_swap_extent(sis, 0, sis->max, 0); | |
53092a74 | 2423 | *span = sis->pages; |
a509bc1a | 2424 | return ret; |
1da177e4 LT |
2425 | } |
2426 | ||
62c230bc | 2427 | if (mapping->a_ops->swap_activate) { |
a509bc1a | 2428 | ret = mapping->a_ops->swap_activate(sis, swap_file, span); |
bc4ae27d OS |
2429 | if (ret >= 0) |
2430 | sis->flags |= SWP_ACTIVATED; | |
62c230bc | 2431 | if (!ret) { |
32646315 | 2432 | sis->flags |= SWP_FS_OPS; |
62c230bc MG |
2433 | ret = add_swap_extent(sis, 0, sis->max, 0); |
2434 | *span = sis->pages; | |
2435 | } | |
a509bc1a | 2436 | return ret; |
62c230bc MG |
2437 | } |
2438 | ||
a509bc1a | 2439 | return generic_swapfile_activate(sis, swap_file, span); |
1da177e4 LT |
2440 | } |
2441 | ||
a2468cc9 AL |
2442 | static int swap_node(struct swap_info_struct *p) |
2443 | { | |
2444 | struct block_device *bdev; | |
2445 | ||
2446 | if (p->bdev) | |
2447 | bdev = p->bdev; | |
2448 | else | |
2449 | bdev = p->swap_file->f_inode->i_sb->s_bdev; | |
2450 | ||
2451 | return bdev ? bdev->bd_disk->node_id : NUMA_NO_NODE; | |
2452 | } | |
2453 | ||
eb085574 HY |
2454 | static void setup_swap_info(struct swap_info_struct *p, int prio, |
2455 | unsigned char *swap_map, | |
2456 | struct swap_cluster_info *cluster_info) | |
40531542 | 2457 | { |
a2468cc9 AL |
2458 | int i; |
2459 | ||
40531542 CEB |
2460 | if (prio >= 0) |
2461 | p->prio = prio; | |
2462 | else | |
2463 | p->prio = --least_priority; | |
18ab4d4c DS |
2464 | /* |
2465 | * the plist prio is negated because plist ordering is | |
2466 | * low-to-high, while swap ordering is high-to-low | |
2467 | */ | |
2468 | p->list.prio = -p->prio; | |
a2468cc9 AL |
2469 | for_each_node(i) { |
2470 | if (p->prio >= 0) | |
2471 | p->avail_lists[i].prio = -p->prio; | |
2472 | else { | |
2473 | if (swap_node(p) == i) | |
2474 | p->avail_lists[i].prio = 1; | |
2475 | else | |
2476 | p->avail_lists[i].prio = -p->prio; | |
2477 | } | |
2478 | } | |
40531542 | 2479 | p->swap_map = swap_map; |
2a8f9449 | 2480 | p->cluster_info = cluster_info; |
eb085574 HY |
2481 | } |
2482 | ||
2483 | static void _enable_swap_info(struct swap_info_struct *p) | |
2484 | { | |
2485 | p->flags |= SWP_WRITEOK | SWP_VALID; | |
ec8acf20 | 2486 | atomic_long_add(p->pages, &nr_swap_pages); |
40531542 CEB |
2487 | total_swap_pages += p->pages; |
2488 | ||
adfab836 | 2489 | assert_spin_locked(&swap_lock); |
adfab836 | 2490 | /* |
18ab4d4c DS |
2491 | * both lists are plists, and thus priority ordered. |
2492 | * swap_active_head needs to be priority ordered for swapoff(), | |
2493 | * which on removal of any swap_info_struct with an auto-assigned | |
2494 | * (i.e. negative) priority increments the auto-assigned priority | |
2495 | * of any lower-priority swap_info_structs. | |
2496 | * swap_avail_head needs to be priority ordered for get_swap_page(), | |
2497 | * which allocates swap pages from the highest available priority | |
2498 | * swap_info_struct. | |
adfab836 | 2499 | */ |
18ab4d4c | 2500 | plist_add(&p->list, &swap_active_head); |
a2468cc9 | 2501 | add_to_avail_list(p); |
cf0cac0a CEB |
2502 | } |
2503 | ||
2504 | static void enable_swap_info(struct swap_info_struct *p, int prio, | |
2505 | unsigned char *swap_map, | |
2a8f9449 | 2506 | struct swap_cluster_info *cluster_info, |
cf0cac0a CEB |
2507 | unsigned long *frontswap_map) |
2508 | { | |
4f89849d | 2509 | frontswap_init(p->type, frontswap_map); |
cf0cac0a | 2510 | spin_lock(&swap_lock); |
ec8acf20 | 2511 | spin_lock(&p->lock); |
eb085574 HY |
2512 | setup_swap_info(p, prio, swap_map, cluster_info); |
2513 | spin_unlock(&p->lock); | |
2514 | spin_unlock(&swap_lock); | |
2515 | /* | |
2516 | * Guarantee swap_map, cluster_info, etc. fields are valid | |
2517 | * between get/put_swap_device() if SWP_VALID bit is set | |
2518 | */ | |
2519 | synchronize_rcu(); | |
2520 | spin_lock(&swap_lock); | |
2521 | spin_lock(&p->lock); | |
2522 | _enable_swap_info(p); | |
ec8acf20 | 2523 | spin_unlock(&p->lock); |
cf0cac0a CEB |
2524 | spin_unlock(&swap_lock); |
2525 | } | |
2526 | ||
2527 | static void reinsert_swap_info(struct swap_info_struct *p) | |
2528 | { | |
2529 | spin_lock(&swap_lock); | |
ec8acf20 | 2530 | spin_lock(&p->lock); |
eb085574 HY |
2531 | setup_swap_info(p, p->prio, p->swap_map, p->cluster_info); |
2532 | _enable_swap_info(p); | |
ec8acf20 | 2533 | spin_unlock(&p->lock); |
40531542 CEB |
2534 | spin_unlock(&swap_lock); |
2535 | } | |
2536 | ||
67afa38e TC |
2537 | bool has_usable_swap(void) |
2538 | { | |
2539 | bool ret = true; | |
2540 | ||
2541 | spin_lock(&swap_lock); | |
2542 | if (plist_head_empty(&swap_active_head)) | |
2543 | ret = false; | |
2544 | spin_unlock(&swap_lock); | |
2545 | return ret; | |
2546 | } | |
2547 | ||
c4ea37c2 | 2548 | SYSCALL_DEFINE1(swapoff, const char __user *, specialfile) |
1da177e4 | 2549 | { |
73c34b6a | 2550 | struct swap_info_struct *p = NULL; |
8d69aaee | 2551 | unsigned char *swap_map; |
2a8f9449 | 2552 | struct swap_cluster_info *cluster_info; |
4f89849d | 2553 | unsigned long *frontswap_map; |
1da177e4 LT |
2554 | struct file *swap_file, *victim; |
2555 | struct address_space *mapping; | |
2556 | struct inode *inode; | |
91a27b2a | 2557 | struct filename *pathname; |
adfab836 | 2558 | int err, found = 0; |
5b808a23 | 2559 | unsigned int old_block_size; |
886bb7e9 | 2560 | |
1da177e4 LT |
2561 | if (!capable(CAP_SYS_ADMIN)) |
2562 | return -EPERM; | |
2563 | ||
191c5424 AV |
2564 | BUG_ON(!current->mm); |
2565 | ||
1da177e4 | 2566 | pathname = getname(specialfile); |
1da177e4 | 2567 | if (IS_ERR(pathname)) |
f58b59c1 | 2568 | return PTR_ERR(pathname); |
1da177e4 | 2569 | |
669abf4e | 2570 | victim = file_open_name(pathname, O_RDWR|O_LARGEFILE, 0); |
1da177e4 LT |
2571 | err = PTR_ERR(victim); |
2572 | if (IS_ERR(victim)) | |
2573 | goto out; | |
2574 | ||
2575 | mapping = victim->f_mapping; | |
5d337b91 | 2576 | spin_lock(&swap_lock); |
18ab4d4c | 2577 | plist_for_each_entry(p, &swap_active_head, list) { |
22c6f8fd | 2578 | if (p->flags & SWP_WRITEOK) { |
adfab836 DS |
2579 | if (p->swap_file->f_mapping == mapping) { |
2580 | found = 1; | |
1da177e4 | 2581 | break; |
adfab836 | 2582 | } |
1da177e4 | 2583 | } |
1da177e4 | 2584 | } |
adfab836 | 2585 | if (!found) { |
1da177e4 | 2586 | err = -EINVAL; |
5d337b91 | 2587 | spin_unlock(&swap_lock); |
1da177e4 LT |
2588 | goto out_dput; |
2589 | } | |
191c5424 | 2590 | if (!security_vm_enough_memory_mm(current->mm, p->pages)) |
1da177e4 LT |
2591 | vm_unacct_memory(p->pages); |
2592 | else { | |
2593 | err = -ENOMEM; | |
5d337b91 | 2594 | spin_unlock(&swap_lock); |
1da177e4 LT |
2595 | goto out_dput; |
2596 | } | |
a2468cc9 | 2597 | del_from_avail_list(p); |
ec8acf20 | 2598 | spin_lock(&p->lock); |
78ecba08 | 2599 | if (p->prio < 0) { |
adfab836 | 2600 | struct swap_info_struct *si = p; |
a2468cc9 | 2601 | int nid; |
adfab836 | 2602 | |
18ab4d4c | 2603 | plist_for_each_entry_continue(si, &swap_active_head, list) { |
adfab836 | 2604 | si->prio++; |
18ab4d4c | 2605 | si->list.prio--; |
a2468cc9 AL |
2606 | for_each_node(nid) { |
2607 | if (si->avail_lists[nid].prio != 1) | |
2608 | si->avail_lists[nid].prio--; | |
2609 | } | |
adfab836 | 2610 | } |
78ecba08 HD |
2611 | least_priority++; |
2612 | } | |
18ab4d4c | 2613 | plist_del(&p->list, &swap_active_head); |
ec8acf20 | 2614 | atomic_long_sub(p->pages, &nr_swap_pages); |
1da177e4 LT |
2615 | total_swap_pages -= p->pages; |
2616 | p->flags &= ~SWP_WRITEOK; | |
ec8acf20 | 2617 | spin_unlock(&p->lock); |
5d337b91 | 2618 | spin_unlock(&swap_lock); |
fb4f88dc | 2619 | |
039939a6 TC |
2620 | disable_swap_slots_cache_lock(); |
2621 | ||
e1e12d2f | 2622 | set_current_oom_origin(); |
adfab836 | 2623 | err = try_to_unuse(p->type, false, 0); /* force unuse all pages */ |
e1e12d2f | 2624 | clear_current_oom_origin(); |
1da177e4 | 2625 | |
1da177e4 LT |
2626 | if (err) { |
2627 | /* re-insert swap space back into swap_list */ | |
cf0cac0a | 2628 | reinsert_swap_info(p); |
039939a6 | 2629 | reenable_swap_slots_cache_unlock(); |
1da177e4 LT |
2630 | goto out_dput; |
2631 | } | |
52b7efdb | 2632 | |
039939a6 TC |
2633 | reenable_swap_slots_cache_unlock(); |
2634 | ||
eb085574 HY |
2635 | spin_lock(&swap_lock); |
2636 | spin_lock(&p->lock); | |
2637 | p->flags &= ~SWP_VALID; /* mark swap device as invalid */ | |
2638 | spin_unlock(&p->lock); | |
2639 | spin_unlock(&swap_lock); | |
2640 | /* | |
2641 | * wait for swap operations protected by get/put_swap_device() | |
2642 | * to complete | |
2643 | */ | |
2644 | synchronize_rcu(); | |
2645 | ||
815c2c54 SL |
2646 | flush_work(&p->discard_work); |
2647 | ||
5d337b91 | 2648 | destroy_swap_extents(p); |
570a335b HD |
2649 | if (p->flags & SWP_CONTINUED) |
2650 | free_swap_count_continuations(p); | |
2651 | ||
81a0298b HY |
2652 | if (!p->bdev || !blk_queue_nonrot(bdev_get_queue(p->bdev))) |
2653 | atomic_dec(&nr_rotate_swap); | |
2654 | ||
fc0abb14 | 2655 | mutex_lock(&swapon_mutex); |
5d337b91 | 2656 | spin_lock(&swap_lock); |
ec8acf20 | 2657 | spin_lock(&p->lock); |
5d337b91 HD |
2658 | drain_mmlist(); |
2659 | ||
52b7efdb | 2660 | /* wait for anyone still in scan_swap_map */ |
52b7efdb HD |
2661 | p->highest_bit = 0; /* cuts scans short */ |
2662 | while (p->flags >= SWP_SCANNING) { | |
ec8acf20 | 2663 | spin_unlock(&p->lock); |
5d337b91 | 2664 | spin_unlock(&swap_lock); |
13e4b57f | 2665 | schedule_timeout_uninterruptible(1); |
5d337b91 | 2666 | spin_lock(&swap_lock); |
ec8acf20 | 2667 | spin_lock(&p->lock); |
52b7efdb | 2668 | } |
52b7efdb | 2669 | |
1da177e4 | 2670 | swap_file = p->swap_file; |
5b808a23 | 2671 | old_block_size = p->old_block_size; |
1da177e4 LT |
2672 | p->swap_file = NULL; |
2673 | p->max = 0; | |
2674 | swap_map = p->swap_map; | |
2675 | p->swap_map = NULL; | |
2a8f9449 SL |
2676 | cluster_info = p->cluster_info; |
2677 | p->cluster_info = NULL; | |
4f89849d | 2678 | frontswap_map = frontswap_map_get(p); |
ec8acf20 | 2679 | spin_unlock(&p->lock); |
5d337b91 | 2680 | spin_unlock(&swap_lock); |
8a84802e | 2681 | arch_swap_invalidate_area(p->type); |
adfab836 | 2682 | frontswap_invalidate_area(p->type); |
58e97ba6 | 2683 | frontswap_map_set(p, NULL); |
fc0abb14 | 2684 | mutex_unlock(&swapon_mutex); |
ebc2a1a6 SL |
2685 | free_percpu(p->percpu_cluster); |
2686 | p->percpu_cluster = NULL; | |
49070588 HY |
2687 | free_percpu(p->cluster_next_cpu); |
2688 | p->cluster_next_cpu = NULL; | |
1da177e4 | 2689 | vfree(swap_map); |
54f180d3 HY |
2690 | kvfree(cluster_info); |
2691 | kvfree(frontswap_map); | |
2de1a7e4 | 2692 | /* Destroy swap account information */ |
adfab836 | 2693 | swap_cgroup_swapoff(p->type); |
4b3ef9da | 2694 | exit_swap_address_space(p->type); |
27a7faa0 | 2695 | |
1da177e4 LT |
2696 | inode = mapping->host; |
2697 | if (S_ISBLK(inode->i_mode)) { | |
2698 | struct block_device *bdev = I_BDEV(inode); | |
1638045c | 2699 | |
5b808a23 | 2700 | set_blocksize(bdev, old_block_size); |
e525fd89 | 2701 | blkdev_put(bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL); |
1da177e4 | 2702 | } |
1638045c DW |
2703 | |
2704 | inode_lock(inode); | |
2705 | inode->i_flags &= ~S_SWAPFILE; | |
2706 | inode_unlock(inode); | |
1da177e4 | 2707 | filp_close(swap_file, NULL); |
f893ab41 WY |
2708 | |
2709 | /* | |
2710 | * Clear the SWP_USED flag after all resources are freed so that swapon | |
2711 | * can reuse this swap_info in alloc_swap_info() safely. It is ok to | |
2712 | * not hold p->lock after we cleared its SWP_WRITEOK. | |
2713 | */ | |
2714 | spin_lock(&swap_lock); | |
2715 | p->flags = 0; | |
2716 | spin_unlock(&swap_lock); | |
2717 | ||
1da177e4 | 2718 | err = 0; |
66d7dd51 KS |
2719 | atomic_inc(&proc_poll_event); |
2720 | wake_up_interruptible(&proc_poll_wait); | |
1da177e4 LT |
2721 | |
2722 | out_dput: | |
2723 | filp_close(victim, NULL); | |
2724 | out: | |
f58b59c1 | 2725 | putname(pathname); |
1da177e4 LT |
2726 | return err; |
2727 | } | |
2728 | ||
2729 | #ifdef CONFIG_PROC_FS | |
9dd95748 | 2730 | static __poll_t swaps_poll(struct file *file, poll_table *wait) |
66d7dd51 | 2731 | { |
f1514638 | 2732 | struct seq_file *seq = file->private_data; |
66d7dd51 KS |
2733 | |
2734 | poll_wait(file, &proc_poll_wait, wait); | |
2735 | ||
f1514638 KS |
2736 | if (seq->poll_event != atomic_read(&proc_poll_event)) { |
2737 | seq->poll_event = atomic_read(&proc_poll_event); | |
a9a08845 | 2738 | return EPOLLIN | EPOLLRDNORM | EPOLLERR | EPOLLPRI; |
66d7dd51 KS |
2739 | } |
2740 | ||
a9a08845 | 2741 | return EPOLLIN | EPOLLRDNORM; |
66d7dd51 KS |
2742 | } |
2743 | ||
1da177e4 LT |
2744 | /* iterator */ |
2745 | static void *swap_start(struct seq_file *swap, loff_t *pos) | |
2746 | { | |
efa90a98 HD |
2747 | struct swap_info_struct *si; |
2748 | int type; | |
1da177e4 LT |
2749 | loff_t l = *pos; |
2750 | ||
fc0abb14 | 2751 | mutex_lock(&swapon_mutex); |
1da177e4 | 2752 | |
881e4aab SS |
2753 | if (!l) |
2754 | return SEQ_START_TOKEN; | |
2755 | ||
c10d38cc | 2756 | for (type = 0; (si = swap_type_to_swap_info(type)); type++) { |
efa90a98 | 2757 | if (!(si->flags & SWP_USED) || !si->swap_map) |
1da177e4 | 2758 | continue; |
881e4aab | 2759 | if (!--l) |
efa90a98 | 2760 | return si; |
1da177e4 LT |
2761 | } |
2762 | ||
2763 | return NULL; | |
2764 | } | |
2765 | ||
2766 | static void *swap_next(struct seq_file *swap, void *v, loff_t *pos) | |
2767 | { | |
efa90a98 HD |
2768 | struct swap_info_struct *si = v; |
2769 | int type; | |
1da177e4 | 2770 | |
881e4aab | 2771 | if (v == SEQ_START_TOKEN) |
efa90a98 HD |
2772 | type = 0; |
2773 | else | |
2774 | type = si->type + 1; | |
881e4aab | 2775 | |
10c8d69f | 2776 | ++(*pos); |
c10d38cc | 2777 | for (; (si = swap_type_to_swap_info(type)); type++) { |
efa90a98 | 2778 | if (!(si->flags & SWP_USED) || !si->swap_map) |
1da177e4 | 2779 | continue; |
efa90a98 | 2780 | return si; |
1da177e4 LT |
2781 | } |
2782 | ||
2783 | return NULL; | |
2784 | } | |
2785 | ||
2786 | static void swap_stop(struct seq_file *swap, void *v) | |
2787 | { | |
fc0abb14 | 2788 | mutex_unlock(&swapon_mutex); |
1da177e4 LT |
2789 | } |
2790 | ||
2791 | static int swap_show(struct seq_file *swap, void *v) | |
2792 | { | |
efa90a98 | 2793 | struct swap_info_struct *si = v; |
1da177e4 LT |
2794 | struct file *file; |
2795 | int len; | |
6f793940 | 2796 | unsigned int bytes, inuse; |
1da177e4 | 2797 | |
efa90a98 | 2798 | if (si == SEQ_START_TOKEN) { |
6f793940 | 2799 | seq_puts(swap,"Filename\t\t\t\tType\t\tSize\t\tUsed\t\tPriority\n"); |
881e4aab SS |
2800 | return 0; |
2801 | } | |
1da177e4 | 2802 | |
6f793940 RD |
2803 | bytes = si->pages << (PAGE_SHIFT - 10); |
2804 | inuse = si->inuse_pages << (PAGE_SHIFT - 10); | |
2805 | ||
efa90a98 | 2806 | file = si->swap_file; |
2726d566 | 2807 | len = seq_file_path(swap, file, " \t\n\\"); |
6f793940 | 2808 | seq_printf(swap, "%*s%s\t%u\t%s%u\t%s%d\n", |
886bb7e9 | 2809 | len < 40 ? 40 - len : 1, " ", |
496ad9aa | 2810 | S_ISBLK(file_inode(file)->i_mode) ? |
1da177e4 | 2811 | "partition" : "file\t", |
6f793940 RD |
2812 | bytes, bytes < 10000000 ? "\t" : "", |
2813 | inuse, inuse < 10000000 ? "\t" : "", | |
efa90a98 | 2814 | si->prio); |
1da177e4 LT |
2815 | return 0; |
2816 | } | |
2817 | ||
15ad7cdc | 2818 | static const struct seq_operations swaps_op = { |
1da177e4 LT |
2819 | .start = swap_start, |
2820 | .next = swap_next, | |
2821 | .stop = swap_stop, | |
2822 | .show = swap_show | |
2823 | }; | |
2824 | ||
2825 | static int swaps_open(struct inode *inode, struct file *file) | |
2826 | { | |
f1514638 | 2827 | struct seq_file *seq; |
66d7dd51 KS |
2828 | int ret; |
2829 | ||
66d7dd51 | 2830 | ret = seq_open(file, &swaps_op); |
f1514638 | 2831 | if (ret) |
66d7dd51 | 2832 | return ret; |
66d7dd51 | 2833 | |
f1514638 KS |
2834 | seq = file->private_data; |
2835 | seq->poll_event = atomic_read(&proc_poll_event); | |
2836 | return 0; | |
1da177e4 LT |
2837 | } |
2838 | ||
97a32539 | 2839 | static const struct proc_ops swaps_proc_ops = { |
d919b33d | 2840 | .proc_flags = PROC_ENTRY_PERMANENT, |
97a32539 AD |
2841 | .proc_open = swaps_open, |
2842 | .proc_read = seq_read, | |
2843 | .proc_lseek = seq_lseek, | |
2844 | .proc_release = seq_release, | |
2845 | .proc_poll = swaps_poll, | |
1da177e4 LT |
2846 | }; |
2847 | ||
2848 | static int __init procswaps_init(void) | |
2849 | { | |
97a32539 | 2850 | proc_create("swaps", 0, NULL, &swaps_proc_ops); |
1da177e4 LT |
2851 | return 0; |
2852 | } | |
2853 | __initcall(procswaps_init); | |
2854 | #endif /* CONFIG_PROC_FS */ | |
2855 | ||
1796316a JB |
2856 | #ifdef MAX_SWAPFILES_CHECK |
2857 | static int __init max_swapfiles_check(void) | |
2858 | { | |
2859 | MAX_SWAPFILES_CHECK(); | |
2860 | return 0; | |
2861 | } | |
2862 | late_initcall(max_swapfiles_check); | |
2863 | #endif | |
2864 | ||
53cbb243 | 2865 | static struct swap_info_struct *alloc_swap_info(void) |
1da177e4 | 2866 | { |
73c34b6a | 2867 | struct swap_info_struct *p; |
b11a76b3 | 2868 | struct swap_info_struct *defer = NULL; |
1da177e4 | 2869 | unsigned int type; |
a2468cc9 | 2870 | int i; |
efa90a98 | 2871 | |
96008744 | 2872 | p = kvzalloc(struct_size(p, avail_lists, nr_node_ids), GFP_KERNEL); |
efa90a98 | 2873 | if (!p) |
53cbb243 | 2874 | return ERR_PTR(-ENOMEM); |
efa90a98 | 2875 | |
5d337b91 | 2876 | spin_lock(&swap_lock); |
efa90a98 HD |
2877 | for (type = 0; type < nr_swapfiles; type++) { |
2878 | if (!(swap_info[type]->flags & SWP_USED)) | |
1da177e4 | 2879 | break; |
efa90a98 | 2880 | } |
0697212a | 2881 | if (type >= MAX_SWAPFILES) { |
5d337b91 | 2882 | spin_unlock(&swap_lock); |
873d7bcf | 2883 | kvfree(p); |
730c0581 | 2884 | return ERR_PTR(-EPERM); |
1da177e4 | 2885 | } |
efa90a98 HD |
2886 | if (type >= nr_swapfiles) { |
2887 | p->type = type; | |
c10d38cc | 2888 | WRITE_ONCE(swap_info[type], p); |
efa90a98 HD |
2889 | /* |
2890 | * Write swap_info[type] before nr_swapfiles, in case a | |
2891 | * racing procfs swap_start() or swap_next() is reading them. | |
2892 | * (We never shrink nr_swapfiles, we never free this entry.) | |
2893 | */ | |
2894 | smp_wmb(); | |
c10d38cc | 2895 | WRITE_ONCE(nr_swapfiles, nr_swapfiles + 1); |
efa90a98 | 2896 | } else { |
b11a76b3 | 2897 | defer = p; |
efa90a98 HD |
2898 | p = swap_info[type]; |
2899 | /* | |
2900 | * Do not memset this entry: a racing procfs swap_next() | |
2901 | * would be relying on p->type to remain valid. | |
2902 | */ | |
2903 | } | |
4efaceb1 | 2904 | p->swap_extent_root = RB_ROOT; |
18ab4d4c | 2905 | plist_node_init(&p->list, 0); |
a2468cc9 AL |
2906 | for_each_node(i) |
2907 | plist_node_init(&p->avail_lists[i], 0); | |
1da177e4 | 2908 | p->flags = SWP_USED; |
5d337b91 | 2909 | spin_unlock(&swap_lock); |
b11a76b3 | 2910 | kvfree(defer); |
ec8acf20 | 2911 | spin_lock_init(&p->lock); |
2628bd6f | 2912 | spin_lock_init(&p->cont_lock); |
efa90a98 | 2913 | |
53cbb243 | 2914 | return p; |
53cbb243 CEB |
2915 | } |
2916 | ||
4d0e1e10 CEB |
2917 | static int claim_swapfile(struct swap_info_struct *p, struct inode *inode) |
2918 | { | |
2919 | int error; | |
2920 | ||
2921 | if (S_ISBLK(inode->i_mode)) { | |
ef16e1d9 | 2922 | p->bdev = blkdev_get_by_dev(inode->i_rdev, |
6f179af8 | 2923 | FMODE_READ | FMODE_WRITE | FMODE_EXCL, p); |
ef16e1d9 CH |
2924 | if (IS_ERR(p->bdev)) { |
2925 | error = PTR_ERR(p->bdev); | |
4d0e1e10 | 2926 | p->bdev = NULL; |
6f179af8 | 2927 | return error; |
4d0e1e10 CEB |
2928 | } |
2929 | p->old_block_size = block_size(p->bdev); | |
2930 | error = set_blocksize(p->bdev, PAGE_SIZE); | |
2931 | if (error < 0) | |
87ade72a | 2932 | return error; |
12d2966d NA |
2933 | /* |
2934 | * Zoned block devices contain zones that have a sequential | |
2935 | * write only restriction. Hence zoned block devices are not | |
2936 | * suitable for swapping. Disallow them here. | |
2937 | */ | |
e556f6ba | 2938 | if (blk_queue_is_zoned(p->bdev->bd_disk->queue)) |
12d2966d | 2939 | return -EINVAL; |
4d0e1e10 CEB |
2940 | p->flags |= SWP_BLKDEV; |
2941 | } else if (S_ISREG(inode->i_mode)) { | |
2942 | p->bdev = inode->i_sb->s_bdev; | |
1638045c DW |
2943 | } |
2944 | ||
4d0e1e10 | 2945 | return 0; |
4d0e1e10 CEB |
2946 | } |
2947 | ||
377eeaa8 AK |
2948 | |
2949 | /* | |
2950 | * Find out how many pages are allowed for a single swap device. There | |
2951 | * are two limiting factors: | |
2952 | * 1) the number of bits for the swap offset in the swp_entry_t type, and | |
2953 | * 2) the number of bits in the swap pte, as defined by the different | |
2954 | * architectures. | |
2955 | * | |
2956 | * In order to find the largest possible bit mask, a swap entry with | |
2957 | * swap type 0 and swap offset ~0UL is created, encoded to a swap pte, | |
2958 | * decoded to a swp_entry_t again, and finally the swap offset is | |
2959 | * extracted. | |
2960 | * | |
2961 | * This will mask all the bits from the initial ~0UL mask that can't | |
2962 | * be encoded in either the swp_entry_t or the architecture definition | |
2963 | * of a swap pte. | |
2964 | */ | |
2965 | unsigned long generic_max_swapfile_size(void) | |
2966 | { | |
2967 | return swp_offset(pte_to_swp_entry( | |
2968 | swp_entry_to_pte(swp_entry(0, ~0UL)))) + 1; | |
2969 | } | |
2970 | ||
2971 | /* Can be overridden by an architecture for additional checks. */ | |
2972 | __weak unsigned long max_swapfile_size(void) | |
2973 | { | |
2974 | return generic_max_swapfile_size(); | |
2975 | } | |
2976 | ||
ca8bd38b CEB |
2977 | static unsigned long read_swap_header(struct swap_info_struct *p, |
2978 | union swap_header *swap_header, | |
2979 | struct inode *inode) | |
2980 | { | |
2981 | int i; | |
2982 | unsigned long maxpages; | |
2983 | unsigned long swapfilepages; | |
d6bbbd29 | 2984 | unsigned long last_page; |
ca8bd38b CEB |
2985 | |
2986 | if (memcmp("SWAPSPACE2", swap_header->magic.magic, 10)) { | |
465c47fd | 2987 | pr_err("Unable to find swap-space signature\n"); |
38719025 | 2988 | return 0; |
ca8bd38b CEB |
2989 | } |
2990 | ||
2991 | /* swap partition endianess hack... */ | |
2992 | if (swab32(swap_header->info.version) == 1) { | |
2993 | swab32s(&swap_header->info.version); | |
2994 | swab32s(&swap_header->info.last_page); | |
2995 | swab32s(&swap_header->info.nr_badpages); | |
dd111be6 JH |
2996 | if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES) |
2997 | return 0; | |
ca8bd38b CEB |
2998 | for (i = 0; i < swap_header->info.nr_badpages; i++) |
2999 | swab32s(&swap_header->info.badpages[i]); | |
3000 | } | |
3001 | /* Check the swap header's sub-version */ | |
3002 | if (swap_header->info.version != 1) { | |
465c47fd AM |
3003 | pr_warn("Unable to handle swap header version %d\n", |
3004 | swap_header->info.version); | |
38719025 | 3005 | return 0; |
ca8bd38b CEB |
3006 | } |
3007 | ||
3008 | p->lowest_bit = 1; | |
3009 | p->cluster_next = 1; | |
3010 | p->cluster_nr = 0; | |
3011 | ||
377eeaa8 | 3012 | maxpages = max_swapfile_size(); |
d6bbbd29 | 3013 | last_page = swap_header->info.last_page; |
a06ad633 TA |
3014 | if (!last_page) { |
3015 | pr_warn("Empty swap-file\n"); | |
3016 | return 0; | |
3017 | } | |
d6bbbd29 | 3018 | if (last_page > maxpages) { |
465c47fd | 3019 | pr_warn("Truncating oversized swap area, only using %luk out of %luk\n", |
d6bbbd29 RJ |
3020 | maxpages << (PAGE_SHIFT - 10), |
3021 | last_page << (PAGE_SHIFT - 10)); | |
3022 | } | |
3023 | if (maxpages > last_page) { | |
3024 | maxpages = last_page + 1; | |
ca8bd38b CEB |
3025 | /* p->max is an unsigned int: don't overflow it */ |
3026 | if ((unsigned int)maxpages == 0) | |
3027 | maxpages = UINT_MAX; | |
3028 | } | |
3029 | p->highest_bit = maxpages - 1; | |
3030 | ||
3031 | if (!maxpages) | |
38719025 | 3032 | return 0; |
ca8bd38b CEB |
3033 | swapfilepages = i_size_read(inode) >> PAGE_SHIFT; |
3034 | if (swapfilepages && maxpages > swapfilepages) { | |
465c47fd | 3035 | pr_warn("Swap area shorter than signature indicates\n"); |
38719025 | 3036 | return 0; |
ca8bd38b CEB |
3037 | } |
3038 | if (swap_header->info.nr_badpages && S_ISREG(inode->i_mode)) | |
38719025 | 3039 | return 0; |
ca8bd38b | 3040 | if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES) |
38719025 | 3041 | return 0; |
ca8bd38b CEB |
3042 | |
3043 | return maxpages; | |
ca8bd38b CEB |
3044 | } |
3045 | ||
4b3ef9da | 3046 | #define SWAP_CLUSTER_INFO_COLS \ |
235b6217 | 3047 | DIV_ROUND_UP(L1_CACHE_BYTES, sizeof(struct swap_cluster_info)) |
4b3ef9da HY |
3048 | #define SWAP_CLUSTER_SPACE_COLS \ |
3049 | DIV_ROUND_UP(SWAP_ADDRESS_SPACE_PAGES, SWAPFILE_CLUSTER) | |
3050 | #define SWAP_CLUSTER_COLS \ | |
3051 | max_t(unsigned int, SWAP_CLUSTER_INFO_COLS, SWAP_CLUSTER_SPACE_COLS) | |
235b6217 | 3052 | |
915d4d7b CEB |
3053 | static int setup_swap_map_and_extents(struct swap_info_struct *p, |
3054 | union swap_header *swap_header, | |
3055 | unsigned char *swap_map, | |
2a8f9449 | 3056 | struct swap_cluster_info *cluster_info, |
915d4d7b CEB |
3057 | unsigned long maxpages, |
3058 | sector_t *span) | |
3059 | { | |
235b6217 | 3060 | unsigned int j, k; |
915d4d7b CEB |
3061 | unsigned int nr_good_pages; |
3062 | int nr_extents; | |
2a8f9449 | 3063 | unsigned long nr_clusters = DIV_ROUND_UP(maxpages, SWAPFILE_CLUSTER); |
235b6217 HY |
3064 | unsigned long col = p->cluster_next / SWAPFILE_CLUSTER % SWAP_CLUSTER_COLS; |
3065 | unsigned long i, idx; | |
915d4d7b CEB |
3066 | |
3067 | nr_good_pages = maxpages - 1; /* omit header page */ | |
3068 | ||
6b534915 HY |
3069 | cluster_list_init(&p->free_clusters); |
3070 | cluster_list_init(&p->discard_clusters); | |
2a8f9449 | 3071 | |
915d4d7b CEB |
3072 | for (i = 0; i < swap_header->info.nr_badpages; i++) { |
3073 | unsigned int page_nr = swap_header->info.badpages[i]; | |
bdb8e3f6 CEB |
3074 | if (page_nr == 0 || page_nr > swap_header->info.last_page) |
3075 | return -EINVAL; | |
915d4d7b CEB |
3076 | if (page_nr < maxpages) { |
3077 | swap_map[page_nr] = SWAP_MAP_BAD; | |
3078 | nr_good_pages--; | |
2a8f9449 SL |
3079 | /* |
3080 | * Haven't marked the cluster free yet, no list | |
3081 | * operation involved | |
3082 | */ | |
3083 | inc_cluster_info_page(p, cluster_info, page_nr); | |
915d4d7b CEB |
3084 | } |
3085 | } | |
3086 | ||
2a8f9449 SL |
3087 | /* Haven't marked the cluster free yet, no list operation involved */ |
3088 | for (i = maxpages; i < round_up(maxpages, SWAPFILE_CLUSTER); i++) | |
3089 | inc_cluster_info_page(p, cluster_info, i); | |
3090 | ||
915d4d7b CEB |
3091 | if (nr_good_pages) { |
3092 | swap_map[0] = SWAP_MAP_BAD; | |
2a8f9449 SL |
3093 | /* |
3094 | * Not mark the cluster free yet, no list | |
3095 | * operation involved | |
3096 | */ | |
3097 | inc_cluster_info_page(p, cluster_info, 0); | |
915d4d7b CEB |
3098 | p->max = maxpages; |
3099 | p->pages = nr_good_pages; | |
3100 | nr_extents = setup_swap_extents(p, span); | |
bdb8e3f6 CEB |
3101 | if (nr_extents < 0) |
3102 | return nr_extents; | |
915d4d7b CEB |
3103 | nr_good_pages = p->pages; |
3104 | } | |
3105 | if (!nr_good_pages) { | |
465c47fd | 3106 | pr_warn("Empty swap-file\n"); |
bdb8e3f6 | 3107 | return -EINVAL; |
915d4d7b CEB |
3108 | } |
3109 | ||
2a8f9449 SL |
3110 | if (!cluster_info) |
3111 | return nr_extents; | |
3112 | ||
235b6217 | 3113 | |
4b3ef9da HY |
3114 | /* |
3115 | * Reduce false cache line sharing between cluster_info and | |
3116 | * sharing same address space. | |
3117 | */ | |
235b6217 HY |
3118 | for (k = 0; k < SWAP_CLUSTER_COLS; k++) { |
3119 | j = (k + col) % SWAP_CLUSTER_COLS; | |
3120 | for (i = 0; i < DIV_ROUND_UP(nr_clusters, SWAP_CLUSTER_COLS); i++) { | |
3121 | idx = i * SWAP_CLUSTER_COLS + j; | |
3122 | if (idx >= nr_clusters) | |
3123 | continue; | |
3124 | if (cluster_count(&cluster_info[idx])) | |
3125 | continue; | |
2a8f9449 | 3126 | cluster_set_flag(&cluster_info[idx], CLUSTER_FLAG_FREE); |
6b534915 HY |
3127 | cluster_list_add_tail(&p->free_clusters, cluster_info, |
3128 | idx); | |
2a8f9449 | 3129 | } |
2a8f9449 | 3130 | } |
915d4d7b | 3131 | return nr_extents; |
915d4d7b CEB |
3132 | } |
3133 | ||
dcf6b7dd RA |
3134 | /* |
3135 | * Helper to sys_swapon determining if a given swap | |
3136 | * backing device queue supports DISCARD operations. | |
3137 | */ | |
3138 | static bool swap_discardable(struct swap_info_struct *si) | |
3139 | { | |
3140 | struct request_queue *q = bdev_get_queue(si->bdev); | |
3141 | ||
3142 | if (!q || !blk_queue_discard(q)) | |
3143 | return false; | |
3144 | ||
3145 | return true; | |
3146 | } | |
3147 | ||
53cbb243 CEB |
3148 | SYSCALL_DEFINE2(swapon, const char __user *, specialfile, int, swap_flags) |
3149 | { | |
3150 | struct swap_info_struct *p; | |
91a27b2a | 3151 | struct filename *name; |
53cbb243 CEB |
3152 | struct file *swap_file = NULL; |
3153 | struct address_space *mapping; | |
40531542 | 3154 | int prio; |
53cbb243 CEB |
3155 | int error; |
3156 | union swap_header *swap_header; | |
915d4d7b | 3157 | int nr_extents; |
53cbb243 CEB |
3158 | sector_t span; |
3159 | unsigned long maxpages; | |
53cbb243 | 3160 | unsigned char *swap_map = NULL; |
2a8f9449 | 3161 | struct swap_cluster_info *cluster_info = NULL; |
38b5faf4 | 3162 | unsigned long *frontswap_map = NULL; |
53cbb243 CEB |
3163 | struct page *page = NULL; |
3164 | struct inode *inode = NULL; | |
7cbf3192 | 3165 | bool inced_nr_rotate_swap = false; |
53cbb243 | 3166 | |
d15cab97 HD |
3167 | if (swap_flags & ~SWAP_FLAGS_VALID) |
3168 | return -EINVAL; | |
3169 | ||
53cbb243 CEB |
3170 | if (!capable(CAP_SYS_ADMIN)) |
3171 | return -EPERM; | |
3172 | ||
a2468cc9 AL |
3173 | if (!swap_avail_heads) |
3174 | return -ENOMEM; | |
3175 | ||
53cbb243 | 3176 | p = alloc_swap_info(); |
2542e513 CEB |
3177 | if (IS_ERR(p)) |
3178 | return PTR_ERR(p); | |
53cbb243 | 3179 | |
815c2c54 SL |
3180 | INIT_WORK(&p->discard_work, swap_discard_work); |
3181 | ||
1da177e4 | 3182 | name = getname(specialfile); |
1da177e4 | 3183 | if (IS_ERR(name)) { |
7de7fb6b | 3184 | error = PTR_ERR(name); |
1da177e4 | 3185 | name = NULL; |
bd69010b | 3186 | goto bad_swap; |
1da177e4 | 3187 | } |
669abf4e | 3188 | swap_file = file_open_name(name, O_RDWR|O_LARGEFILE, 0); |
1da177e4 | 3189 | if (IS_ERR(swap_file)) { |
7de7fb6b | 3190 | error = PTR_ERR(swap_file); |
1da177e4 | 3191 | swap_file = NULL; |
bd69010b | 3192 | goto bad_swap; |
1da177e4 LT |
3193 | } |
3194 | ||
3195 | p->swap_file = swap_file; | |
3196 | mapping = swap_file->f_mapping; | |
2130781e | 3197 | inode = mapping->host; |
6f179af8 | 3198 | |
4d0e1e10 CEB |
3199 | error = claim_swapfile(p, inode); |
3200 | if (unlikely(error)) | |
1da177e4 | 3201 | goto bad_swap; |
1da177e4 | 3202 | |
d795a90e NA |
3203 | inode_lock(inode); |
3204 | if (IS_SWAPFILE(inode)) { | |
3205 | error = -EBUSY; | |
3206 | goto bad_swap_unlock_inode; | |
3207 | } | |
3208 | ||
1da177e4 LT |
3209 | /* |
3210 | * Read the swap header. | |
3211 | */ | |
3212 | if (!mapping->a_ops->readpage) { | |
3213 | error = -EINVAL; | |
d795a90e | 3214 | goto bad_swap_unlock_inode; |
1da177e4 | 3215 | } |
090d2b18 | 3216 | page = read_mapping_page(mapping, 0, swap_file); |
1da177e4 LT |
3217 | if (IS_ERR(page)) { |
3218 | error = PTR_ERR(page); | |
d795a90e | 3219 | goto bad_swap_unlock_inode; |
1da177e4 | 3220 | } |
81e33971 | 3221 | swap_header = kmap(page); |
1da177e4 | 3222 | |
ca8bd38b CEB |
3223 | maxpages = read_swap_header(p, swap_header, inode); |
3224 | if (unlikely(!maxpages)) { | |
1da177e4 | 3225 | error = -EINVAL; |
d795a90e | 3226 | goto bad_swap_unlock_inode; |
1da177e4 | 3227 | } |
886bb7e9 | 3228 | |
81e33971 | 3229 | /* OK, set up the swap map and apply the bad block list */ |
803d0c83 | 3230 | swap_map = vzalloc(maxpages); |
81e33971 HD |
3231 | if (!swap_map) { |
3232 | error = -ENOMEM; | |
d795a90e | 3233 | goto bad_swap_unlock_inode; |
81e33971 | 3234 | } |
f0571429 | 3235 | |
1cb039f3 | 3236 | if (p->bdev && blk_queue_stable_writes(p->bdev->bd_disk->queue)) |
f0571429 MK |
3237 | p->flags |= SWP_STABLE_WRITES; |
3238 | ||
a8b456d0 | 3239 | if (p->bdev && p->bdev->bd_disk->fops->rw_page) |
539a6fea MK |
3240 | p->flags |= SWP_SYNCHRONOUS_IO; |
3241 | ||
2a8f9449 | 3242 | if (p->bdev && blk_queue_nonrot(bdev_get_queue(p->bdev))) { |
6f179af8 | 3243 | int cpu; |
235b6217 | 3244 | unsigned long ci, nr_cluster; |
6f179af8 | 3245 | |
2a8f9449 | 3246 | p->flags |= SWP_SOLIDSTATE; |
49070588 HY |
3247 | p->cluster_next_cpu = alloc_percpu(unsigned int); |
3248 | if (!p->cluster_next_cpu) { | |
3249 | error = -ENOMEM; | |
3250 | goto bad_swap_unlock_inode; | |
3251 | } | |
2a8f9449 SL |
3252 | /* |
3253 | * select a random position to start with to help wear leveling | |
3254 | * SSD | |
3255 | */ | |
49070588 HY |
3256 | for_each_possible_cpu(cpu) { |
3257 | per_cpu(*p->cluster_next_cpu, cpu) = | |
3258 | 1 + prandom_u32_max(p->highest_bit); | |
3259 | } | |
235b6217 | 3260 | nr_cluster = DIV_ROUND_UP(maxpages, SWAPFILE_CLUSTER); |
2a8f9449 | 3261 | |
778e1cdd | 3262 | cluster_info = kvcalloc(nr_cluster, sizeof(*cluster_info), |
54f180d3 | 3263 | GFP_KERNEL); |
2a8f9449 SL |
3264 | if (!cluster_info) { |
3265 | error = -ENOMEM; | |
d795a90e | 3266 | goto bad_swap_unlock_inode; |
2a8f9449 | 3267 | } |
235b6217 HY |
3268 | |
3269 | for (ci = 0; ci < nr_cluster; ci++) | |
3270 | spin_lock_init(&((cluster_info + ci)->lock)); | |
3271 | ||
ebc2a1a6 SL |
3272 | p->percpu_cluster = alloc_percpu(struct percpu_cluster); |
3273 | if (!p->percpu_cluster) { | |
3274 | error = -ENOMEM; | |
d795a90e | 3275 | goto bad_swap_unlock_inode; |
ebc2a1a6 | 3276 | } |
6f179af8 | 3277 | for_each_possible_cpu(cpu) { |
ebc2a1a6 | 3278 | struct percpu_cluster *cluster; |
6f179af8 | 3279 | cluster = per_cpu_ptr(p->percpu_cluster, cpu); |
ebc2a1a6 SL |
3280 | cluster_set_null(&cluster->index); |
3281 | } | |
7cbf3192 | 3282 | } else { |
81a0298b | 3283 | atomic_inc(&nr_rotate_swap); |
7cbf3192 OS |
3284 | inced_nr_rotate_swap = true; |
3285 | } | |
1da177e4 | 3286 | |
1421ef3c CEB |
3287 | error = swap_cgroup_swapon(p->type, maxpages); |
3288 | if (error) | |
d795a90e | 3289 | goto bad_swap_unlock_inode; |
1421ef3c | 3290 | |
915d4d7b | 3291 | nr_extents = setup_swap_map_and_extents(p, swap_header, swap_map, |
2a8f9449 | 3292 | cluster_info, maxpages, &span); |
915d4d7b CEB |
3293 | if (unlikely(nr_extents < 0)) { |
3294 | error = nr_extents; | |
d795a90e | 3295 | goto bad_swap_unlock_inode; |
1da177e4 | 3296 | } |
38b5faf4 | 3297 | /* frontswap enabled? set up bit-per-page map for frontswap */ |
8ea1d2a1 | 3298 | if (IS_ENABLED(CONFIG_FRONTSWAP)) |
778e1cdd KC |
3299 | frontswap_map = kvcalloc(BITS_TO_LONGS(maxpages), |
3300 | sizeof(long), | |
54f180d3 | 3301 | GFP_KERNEL); |
1da177e4 | 3302 | |
2a8f9449 SL |
3303 | if (p->bdev &&(swap_flags & SWAP_FLAG_DISCARD) && swap_discardable(p)) { |
3304 | /* | |
3305 | * When discard is enabled for swap with no particular | |
3306 | * policy flagged, we set all swap discard flags here in | |
3307 | * order to sustain backward compatibility with older | |
3308 | * swapon(8) releases. | |
3309 | */ | |
3310 | p->flags |= (SWP_DISCARDABLE | SWP_AREA_DISCARD | | |
3311 | SWP_PAGE_DISCARD); | |
dcf6b7dd | 3312 | |
2a8f9449 SL |
3313 | /* |
3314 | * By flagging sys_swapon, a sysadmin can tell us to | |
3315 | * either do single-time area discards only, or to just | |
3316 | * perform discards for released swap page-clusters. | |
3317 | * Now it's time to adjust the p->flags accordingly. | |
3318 | */ | |
3319 | if (swap_flags & SWAP_FLAG_DISCARD_ONCE) | |
3320 | p->flags &= ~SWP_PAGE_DISCARD; | |
3321 | else if (swap_flags & SWAP_FLAG_DISCARD_PAGES) | |
3322 | p->flags &= ~SWP_AREA_DISCARD; | |
3323 | ||
3324 | /* issue a swapon-time discard if it's still required */ | |
3325 | if (p->flags & SWP_AREA_DISCARD) { | |
3326 | int err = discard_swap(p); | |
3327 | if (unlikely(err)) | |
3328 | pr_err("swapon: discard_swap(%p): %d\n", | |
3329 | p, err); | |
dcf6b7dd | 3330 | } |
20137a49 | 3331 | } |
6a6ba831 | 3332 | |
4b3ef9da HY |
3333 | error = init_swap_address_space(p->type, maxpages); |
3334 | if (error) | |
d795a90e | 3335 | goto bad_swap_unlock_inode; |
4b3ef9da | 3336 | |
dc617f29 DW |
3337 | /* |
3338 | * Flush any pending IO and dirty mappings before we start using this | |
3339 | * swap device. | |
3340 | */ | |
3341 | inode->i_flags |= S_SWAPFILE; | |
3342 | error = inode_drain_writes(inode); | |
3343 | if (error) { | |
3344 | inode->i_flags &= ~S_SWAPFILE; | |
822bca52 | 3345 | goto free_swap_address_space; |
dc617f29 DW |
3346 | } |
3347 | ||
fc0abb14 | 3348 | mutex_lock(&swapon_mutex); |
40531542 | 3349 | prio = -1; |
78ecba08 | 3350 | if (swap_flags & SWAP_FLAG_PREFER) |
40531542 | 3351 | prio = |
78ecba08 | 3352 | (swap_flags & SWAP_FLAG_PRIO_MASK) >> SWAP_FLAG_PRIO_SHIFT; |
2a8f9449 | 3353 | enable_swap_info(p, prio, swap_map, cluster_info, frontswap_map); |
c69dbfb8 | 3354 | |
756a025f | 3355 | pr_info("Adding %uk swap on %s. Priority:%d extents:%d across:%lluk %s%s%s%s%s\n", |
91a27b2a | 3356 | p->pages<<(PAGE_SHIFT-10), name->name, p->prio, |
c69dbfb8 CEB |
3357 | nr_extents, (unsigned long long)span<<(PAGE_SHIFT-10), |
3358 | (p->flags & SWP_SOLIDSTATE) ? "SS" : "", | |
38b5faf4 | 3359 | (p->flags & SWP_DISCARDABLE) ? "D" : "", |
dcf6b7dd RA |
3360 | (p->flags & SWP_AREA_DISCARD) ? "s" : "", |
3361 | (p->flags & SWP_PAGE_DISCARD) ? "c" : "", | |
38b5faf4 | 3362 | (frontswap_map) ? "FS" : ""); |
c69dbfb8 | 3363 | |
fc0abb14 | 3364 | mutex_unlock(&swapon_mutex); |
66d7dd51 KS |
3365 | atomic_inc(&proc_poll_event); |
3366 | wake_up_interruptible(&proc_poll_wait); | |
3367 | ||
1da177e4 LT |
3368 | error = 0; |
3369 | goto out; | |
822bca52 ML |
3370 | free_swap_address_space: |
3371 | exit_swap_address_space(p->type); | |
d795a90e NA |
3372 | bad_swap_unlock_inode: |
3373 | inode_unlock(inode); | |
1da177e4 | 3374 | bad_swap: |
ebc2a1a6 SL |
3375 | free_percpu(p->percpu_cluster); |
3376 | p->percpu_cluster = NULL; | |
49070588 HY |
3377 | free_percpu(p->cluster_next_cpu); |
3378 | p->cluster_next_cpu = NULL; | |
bd69010b | 3379 | if (inode && S_ISBLK(inode->i_mode) && p->bdev) { |
f2090d2d CEB |
3380 | set_blocksize(p->bdev, p->old_block_size); |
3381 | blkdev_put(p->bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL); | |
1da177e4 | 3382 | } |
d795a90e | 3383 | inode = NULL; |
4cd3bb10 | 3384 | destroy_swap_extents(p); |
e8e6c2ec | 3385 | swap_cgroup_swapoff(p->type); |
5d337b91 | 3386 | spin_lock(&swap_lock); |
1da177e4 | 3387 | p->swap_file = NULL; |
1da177e4 | 3388 | p->flags = 0; |
5d337b91 | 3389 | spin_unlock(&swap_lock); |
1da177e4 | 3390 | vfree(swap_map); |
8606a1a9 | 3391 | kvfree(cluster_info); |
b6b1fd2a | 3392 | kvfree(frontswap_map); |
7cbf3192 OS |
3393 | if (inced_nr_rotate_swap) |
3394 | atomic_dec(&nr_rotate_swap); | |
d795a90e | 3395 | if (swap_file) |
1da177e4 LT |
3396 | filp_close(swap_file, NULL); |
3397 | out: | |
3398 | if (page && !IS_ERR(page)) { | |
3399 | kunmap(page); | |
09cbfeaf | 3400 | put_page(page); |
1da177e4 LT |
3401 | } |
3402 | if (name) | |
3403 | putname(name); | |
1638045c | 3404 | if (inode) |
5955102c | 3405 | inode_unlock(inode); |
039939a6 TC |
3406 | if (!error) |
3407 | enable_swap_slots_cache(); | |
1da177e4 LT |
3408 | return error; |
3409 | } | |
3410 | ||
3411 | void si_swapinfo(struct sysinfo *val) | |
3412 | { | |
efa90a98 | 3413 | unsigned int type; |
1da177e4 LT |
3414 | unsigned long nr_to_be_unused = 0; |
3415 | ||
5d337b91 | 3416 | spin_lock(&swap_lock); |
efa90a98 HD |
3417 | for (type = 0; type < nr_swapfiles; type++) { |
3418 | struct swap_info_struct *si = swap_info[type]; | |
3419 | ||
3420 | if ((si->flags & SWP_USED) && !(si->flags & SWP_WRITEOK)) | |
3421 | nr_to_be_unused += si->inuse_pages; | |
1da177e4 | 3422 | } |
ec8acf20 | 3423 | val->freeswap = atomic_long_read(&nr_swap_pages) + nr_to_be_unused; |
1da177e4 | 3424 | val->totalswap = total_swap_pages + nr_to_be_unused; |
5d337b91 | 3425 | spin_unlock(&swap_lock); |
1da177e4 LT |
3426 | } |
3427 | ||
3428 | /* | |
3429 | * Verify that a swap entry is valid and increment its swap map count. | |
3430 | * | |
355cfa73 KH |
3431 | * Returns error code in following case. |
3432 | * - success -> 0 | |
3433 | * - swp_entry is invalid -> EINVAL | |
3434 | * - swp_entry is migration entry -> EINVAL | |
3435 | * - swap-cache reference is requested but there is already one. -> EEXIST | |
3436 | * - swap-cache reference is requested but the entry is not used. -> ENOENT | |
570a335b | 3437 | * - swap-mapped reference requested but needs continued swap count. -> ENOMEM |
1da177e4 | 3438 | */ |
8d69aaee | 3439 | static int __swap_duplicate(swp_entry_t entry, unsigned char usage) |
1da177e4 | 3440 | { |
73c34b6a | 3441 | struct swap_info_struct *p; |
235b6217 | 3442 | struct swap_cluster_info *ci; |
c10d38cc | 3443 | unsigned long offset; |
8d69aaee HD |
3444 | unsigned char count; |
3445 | unsigned char has_cache; | |
9d9a0334 | 3446 | int err; |
1da177e4 | 3447 | |
eb085574 | 3448 | p = get_swap_device(entry); |
c10d38cc | 3449 | if (!p) |
9d9a0334 | 3450 | return -EINVAL; |
235b6217 | 3451 | |
eb085574 | 3452 | offset = swp_offset(entry); |
235b6217 | 3453 | ci = lock_cluster_or_swap_info(p, offset); |
355cfa73 | 3454 | |
253d553b | 3455 | count = p->swap_map[offset]; |
edfe23da SL |
3456 | |
3457 | /* | |
3458 | * swapin_readahead() doesn't check if a swap entry is valid, so the | |
3459 | * swap entry could be SWAP_MAP_BAD. Check here with lock held. | |
3460 | */ | |
3461 | if (unlikely(swap_count(count) == SWAP_MAP_BAD)) { | |
3462 | err = -ENOENT; | |
3463 | goto unlock_out; | |
3464 | } | |
3465 | ||
253d553b HD |
3466 | has_cache = count & SWAP_HAS_CACHE; |
3467 | count &= ~SWAP_HAS_CACHE; | |
3468 | err = 0; | |
355cfa73 | 3469 | |
253d553b | 3470 | if (usage == SWAP_HAS_CACHE) { |
355cfa73 KH |
3471 | |
3472 | /* set SWAP_HAS_CACHE if there is no cache and entry is used */ | |
253d553b HD |
3473 | if (!has_cache && count) |
3474 | has_cache = SWAP_HAS_CACHE; | |
3475 | else if (has_cache) /* someone else added cache */ | |
3476 | err = -EEXIST; | |
3477 | else /* no users remaining */ | |
3478 | err = -ENOENT; | |
355cfa73 KH |
3479 | |
3480 | } else if (count || has_cache) { | |
253d553b | 3481 | |
570a335b HD |
3482 | if ((count & ~COUNT_CONTINUED) < SWAP_MAP_MAX) |
3483 | count += usage; | |
3484 | else if ((count & ~COUNT_CONTINUED) > SWAP_MAP_MAX) | |
253d553b | 3485 | err = -EINVAL; |
570a335b HD |
3486 | else if (swap_count_continued(p, offset, count)) |
3487 | count = COUNT_CONTINUED; | |
3488 | else | |
3489 | err = -ENOMEM; | |
355cfa73 | 3490 | } else |
253d553b HD |
3491 | err = -ENOENT; /* unused swap entry */ |
3492 | ||
a449bf58 | 3493 | WRITE_ONCE(p->swap_map[offset], count | has_cache); |
253d553b | 3494 | |
355cfa73 | 3495 | unlock_out: |
235b6217 | 3496 | unlock_cluster_or_swap_info(p, ci); |
eb085574 HY |
3497 | if (p) |
3498 | put_swap_device(p); | |
253d553b | 3499 | return err; |
1da177e4 | 3500 | } |
253d553b | 3501 | |
aaa46865 HD |
3502 | /* |
3503 | * Help swapoff by noting that swap entry belongs to shmem/tmpfs | |
3504 | * (in which case its reference count is never incremented). | |
3505 | */ | |
3506 | void swap_shmem_alloc(swp_entry_t entry) | |
3507 | { | |
3508 | __swap_duplicate(entry, SWAP_MAP_SHMEM); | |
3509 | } | |
3510 | ||
355cfa73 | 3511 | /* |
08259d58 HD |
3512 | * Increase reference count of swap entry by 1. |
3513 | * Returns 0 for success, or -ENOMEM if a swap_count_continuation is required | |
3514 | * but could not be atomically allocated. Returns 0, just as if it succeeded, | |
3515 | * if __swap_duplicate() fails for another reason (-EINVAL or -ENOENT), which | |
3516 | * might occur if a page table entry has got corrupted. | |
355cfa73 | 3517 | */ |
570a335b | 3518 | int swap_duplicate(swp_entry_t entry) |
355cfa73 | 3519 | { |
570a335b HD |
3520 | int err = 0; |
3521 | ||
3522 | while (!err && __swap_duplicate(entry, 1) == -ENOMEM) | |
3523 | err = add_swap_count_continuation(entry, GFP_ATOMIC); | |
3524 | return err; | |
355cfa73 | 3525 | } |
1da177e4 | 3526 | |
cb4b86ba | 3527 | /* |
355cfa73 KH |
3528 | * @entry: swap entry for which we allocate swap cache. |
3529 | * | |
73c34b6a | 3530 | * Called when allocating swap cache for existing swap entry, |
355cfa73 | 3531 | * This can return error codes. Returns 0 at success. |
3eeba135 | 3532 | * -EEXIST means there is a swap cache. |
355cfa73 | 3533 | * Note: return code is different from swap_duplicate(). |
cb4b86ba KH |
3534 | */ |
3535 | int swapcache_prepare(swp_entry_t entry) | |
3536 | { | |
253d553b | 3537 | return __swap_duplicate(entry, SWAP_HAS_CACHE); |
cb4b86ba KH |
3538 | } |
3539 | ||
0bcac06f MK |
3540 | struct swap_info_struct *swp_swap_info(swp_entry_t entry) |
3541 | { | |
c10d38cc | 3542 | return swap_type_to_swap_info(swp_type(entry)); |
0bcac06f MK |
3543 | } |
3544 | ||
f981c595 MG |
3545 | struct swap_info_struct *page_swap_info(struct page *page) |
3546 | { | |
0bcac06f MK |
3547 | swp_entry_t entry = { .val = page_private(page) }; |
3548 | return swp_swap_info(entry); | |
f981c595 MG |
3549 | } |
3550 | ||
3551 | /* | |
3552 | * out-of-line __page_file_ methods to avoid include hell. | |
3553 | */ | |
3554 | struct address_space *__page_file_mapping(struct page *page) | |
3555 | { | |
f981c595 MG |
3556 | return page_swap_info(page)->swap_file->f_mapping; |
3557 | } | |
3558 | EXPORT_SYMBOL_GPL(__page_file_mapping); | |
3559 | ||
3560 | pgoff_t __page_file_index(struct page *page) | |
3561 | { | |
3562 | swp_entry_t swap = { .val = page_private(page) }; | |
f981c595 MG |
3563 | return swp_offset(swap); |
3564 | } | |
3565 | EXPORT_SYMBOL_GPL(__page_file_index); | |
3566 | ||
570a335b HD |
3567 | /* |
3568 | * add_swap_count_continuation - called when a swap count is duplicated | |
3569 | * beyond SWAP_MAP_MAX, it allocates a new page and links that to the entry's | |
3570 | * page of the original vmalloc'ed swap_map, to hold the continuation count | |
3571 | * (for that entry and for its neighbouring PAGE_SIZE swap entries). Called | |
3572 | * again when count is duplicated beyond SWAP_MAP_MAX * SWAP_CONT_MAX, etc. | |
3573 | * | |
3574 | * These continuation pages are seldom referenced: the common paths all work | |
3575 | * on the original swap_map, only referring to a continuation page when the | |
3576 | * low "digit" of a count is incremented or decremented through SWAP_MAP_MAX. | |
3577 | * | |
3578 | * add_swap_count_continuation(, GFP_ATOMIC) can be called while holding | |
3579 | * page table locks; if it fails, add_swap_count_continuation(, GFP_KERNEL) | |
3580 | * can be called after dropping locks. | |
3581 | */ | |
3582 | int add_swap_count_continuation(swp_entry_t entry, gfp_t gfp_mask) | |
3583 | { | |
3584 | struct swap_info_struct *si; | |
235b6217 | 3585 | struct swap_cluster_info *ci; |
570a335b HD |
3586 | struct page *head; |
3587 | struct page *page; | |
3588 | struct page *list_page; | |
3589 | pgoff_t offset; | |
3590 | unsigned char count; | |
eb085574 | 3591 | int ret = 0; |
570a335b HD |
3592 | |
3593 | /* | |
3594 | * When debugging, it's easier to use __GFP_ZERO here; but it's better | |
3595 | * for latency not to zero a page while GFP_ATOMIC and holding locks. | |
3596 | */ | |
3597 | page = alloc_page(gfp_mask | __GFP_HIGHMEM); | |
3598 | ||
eb085574 | 3599 | si = get_swap_device(entry); |
570a335b HD |
3600 | if (!si) { |
3601 | /* | |
3602 | * An acceptable race has occurred since the failing | |
eb085574 | 3603 | * __swap_duplicate(): the swap device may be swapoff |
570a335b HD |
3604 | */ |
3605 | goto outer; | |
3606 | } | |
eb085574 | 3607 | spin_lock(&si->lock); |
570a335b HD |
3608 | |
3609 | offset = swp_offset(entry); | |
235b6217 HY |
3610 | |
3611 | ci = lock_cluster(si, offset); | |
3612 | ||
d8aa24e0 | 3613 | count = swap_count(si->swap_map[offset]); |
570a335b HD |
3614 | |
3615 | if ((count & ~COUNT_CONTINUED) != SWAP_MAP_MAX) { | |
3616 | /* | |
3617 | * The higher the swap count, the more likely it is that tasks | |
3618 | * will race to add swap count continuation: we need to avoid | |
3619 | * over-provisioning. | |
3620 | */ | |
3621 | goto out; | |
3622 | } | |
3623 | ||
3624 | if (!page) { | |
eb085574 HY |
3625 | ret = -ENOMEM; |
3626 | goto out; | |
570a335b HD |
3627 | } |
3628 | ||
3629 | /* | |
3630 | * We are fortunate that although vmalloc_to_page uses pte_offset_map, | |
2de1a7e4 SJ |
3631 | * no architecture is using highmem pages for kernel page tables: so it |
3632 | * will not corrupt the GFP_ATOMIC caller's atomic page table kmaps. | |
570a335b HD |
3633 | */ |
3634 | head = vmalloc_to_page(si->swap_map + offset); | |
3635 | offset &= ~PAGE_MASK; | |
3636 | ||
2628bd6f | 3637 | spin_lock(&si->cont_lock); |
570a335b HD |
3638 | /* |
3639 | * Page allocation does not initialize the page's lru field, | |
3640 | * but it does always reset its private field. | |
3641 | */ | |
3642 | if (!page_private(head)) { | |
3643 | BUG_ON(count & COUNT_CONTINUED); | |
3644 | INIT_LIST_HEAD(&head->lru); | |
3645 | set_page_private(head, SWP_CONTINUED); | |
3646 | si->flags |= SWP_CONTINUED; | |
3647 | } | |
3648 | ||
3649 | list_for_each_entry(list_page, &head->lru, lru) { | |
3650 | unsigned char *map; | |
3651 | ||
3652 | /* | |
3653 | * If the previous map said no continuation, but we've found | |
3654 | * a continuation page, free our allocation and use this one. | |
3655 | */ | |
3656 | if (!(count & COUNT_CONTINUED)) | |
2628bd6f | 3657 | goto out_unlock_cont; |
570a335b | 3658 | |
9b04c5fe | 3659 | map = kmap_atomic(list_page) + offset; |
570a335b | 3660 | count = *map; |
9b04c5fe | 3661 | kunmap_atomic(map); |
570a335b HD |
3662 | |
3663 | /* | |
3664 | * If this continuation count now has some space in it, | |
3665 | * free our allocation and use this one. | |
3666 | */ | |
3667 | if ((count & ~COUNT_CONTINUED) != SWAP_CONT_MAX) | |
2628bd6f | 3668 | goto out_unlock_cont; |
570a335b HD |
3669 | } |
3670 | ||
3671 | list_add_tail(&page->lru, &head->lru); | |
3672 | page = NULL; /* now it's attached, don't free it */ | |
2628bd6f HY |
3673 | out_unlock_cont: |
3674 | spin_unlock(&si->cont_lock); | |
570a335b | 3675 | out: |
235b6217 | 3676 | unlock_cluster(ci); |
ec8acf20 | 3677 | spin_unlock(&si->lock); |
eb085574 | 3678 | put_swap_device(si); |
570a335b HD |
3679 | outer: |
3680 | if (page) | |
3681 | __free_page(page); | |
eb085574 | 3682 | return ret; |
570a335b HD |
3683 | } |
3684 | ||
3685 | /* | |
3686 | * swap_count_continued - when the original swap_map count is incremented | |
3687 | * from SWAP_MAP_MAX, check if there is already a continuation page to carry | |
3688 | * into, carry if so, or else fail until a new continuation page is allocated; | |
3689 | * when the original swap_map count is decremented from 0 with continuation, | |
3690 | * borrow from the continuation and report whether it still holds more. | |
235b6217 HY |
3691 | * Called while __swap_duplicate() or swap_entry_free() holds swap or cluster |
3692 | * lock. | |
570a335b HD |
3693 | */ |
3694 | static bool swap_count_continued(struct swap_info_struct *si, | |
3695 | pgoff_t offset, unsigned char count) | |
3696 | { | |
3697 | struct page *head; | |
3698 | struct page *page; | |
3699 | unsigned char *map; | |
2628bd6f | 3700 | bool ret; |
570a335b HD |
3701 | |
3702 | head = vmalloc_to_page(si->swap_map + offset); | |
3703 | if (page_private(head) != SWP_CONTINUED) { | |
3704 | BUG_ON(count & COUNT_CONTINUED); | |
3705 | return false; /* need to add count continuation */ | |
3706 | } | |
3707 | ||
2628bd6f | 3708 | spin_lock(&si->cont_lock); |
570a335b | 3709 | offset &= ~PAGE_MASK; |
213516ac | 3710 | page = list_next_entry(head, lru); |
9b04c5fe | 3711 | map = kmap_atomic(page) + offset; |
570a335b HD |
3712 | |
3713 | if (count == SWAP_MAP_MAX) /* initial increment from swap_map */ | |
3714 | goto init_map; /* jump over SWAP_CONT_MAX checks */ | |
3715 | ||
3716 | if (count == (SWAP_MAP_MAX | COUNT_CONTINUED)) { /* incrementing */ | |
3717 | /* | |
3718 | * Think of how you add 1 to 999 | |
3719 | */ | |
3720 | while (*map == (SWAP_CONT_MAX | COUNT_CONTINUED)) { | |
9b04c5fe | 3721 | kunmap_atomic(map); |
213516ac | 3722 | page = list_next_entry(page, lru); |
570a335b | 3723 | BUG_ON(page == head); |
9b04c5fe | 3724 | map = kmap_atomic(page) + offset; |
570a335b HD |
3725 | } |
3726 | if (*map == SWAP_CONT_MAX) { | |
9b04c5fe | 3727 | kunmap_atomic(map); |
213516ac | 3728 | page = list_next_entry(page, lru); |
2628bd6f HY |
3729 | if (page == head) { |
3730 | ret = false; /* add count continuation */ | |
3731 | goto out; | |
3732 | } | |
9b04c5fe | 3733 | map = kmap_atomic(page) + offset; |
570a335b HD |
3734 | init_map: *map = 0; /* we didn't zero the page */ |
3735 | } | |
3736 | *map += 1; | |
9b04c5fe | 3737 | kunmap_atomic(map); |
213516ac | 3738 | while ((page = list_prev_entry(page, lru)) != head) { |
9b04c5fe | 3739 | map = kmap_atomic(page) + offset; |
570a335b | 3740 | *map = COUNT_CONTINUED; |
9b04c5fe | 3741 | kunmap_atomic(map); |
570a335b | 3742 | } |
2628bd6f | 3743 | ret = true; /* incremented */ |
570a335b HD |
3744 | |
3745 | } else { /* decrementing */ | |
3746 | /* | |
3747 | * Think of how you subtract 1 from 1000 | |
3748 | */ | |
3749 | BUG_ON(count != COUNT_CONTINUED); | |
3750 | while (*map == COUNT_CONTINUED) { | |
9b04c5fe | 3751 | kunmap_atomic(map); |
213516ac | 3752 | page = list_next_entry(page, lru); |
570a335b | 3753 | BUG_ON(page == head); |
9b04c5fe | 3754 | map = kmap_atomic(page) + offset; |
570a335b HD |
3755 | } |
3756 | BUG_ON(*map == 0); | |
3757 | *map -= 1; | |
3758 | if (*map == 0) | |
3759 | count = 0; | |
9b04c5fe | 3760 | kunmap_atomic(map); |
213516ac | 3761 | while ((page = list_prev_entry(page, lru)) != head) { |
9b04c5fe | 3762 | map = kmap_atomic(page) + offset; |
570a335b HD |
3763 | *map = SWAP_CONT_MAX | count; |
3764 | count = COUNT_CONTINUED; | |
9b04c5fe | 3765 | kunmap_atomic(map); |
570a335b | 3766 | } |
2628bd6f | 3767 | ret = count == COUNT_CONTINUED; |
570a335b | 3768 | } |
2628bd6f HY |
3769 | out: |
3770 | spin_unlock(&si->cont_lock); | |
3771 | return ret; | |
570a335b HD |
3772 | } |
3773 | ||
3774 | /* | |
3775 | * free_swap_count_continuations - swapoff free all the continuation pages | |
3776 | * appended to the swap_map, after swap_map is quiesced, before vfree'ing it. | |
3777 | */ | |
3778 | static void free_swap_count_continuations(struct swap_info_struct *si) | |
3779 | { | |
3780 | pgoff_t offset; | |
3781 | ||
3782 | for (offset = 0; offset < si->max; offset += PAGE_SIZE) { | |
3783 | struct page *head; | |
3784 | head = vmalloc_to_page(si->swap_map + offset); | |
3785 | if (page_private(head)) { | |
0d576d20 GT |
3786 | struct page *page, *next; |
3787 | ||
3788 | list_for_each_entry_safe(page, next, &head->lru, lru) { | |
3789 | list_del(&page->lru); | |
570a335b HD |
3790 | __free_page(page); |
3791 | } | |
3792 | } | |
3793 | } | |
3794 | } | |
a2468cc9 | 3795 | |
2cf85583 | 3796 | #if defined(CONFIG_MEMCG) && defined(CONFIG_BLK_CGROUP) |
6caa6a07 | 3797 | void cgroup_throttle_swaprate(struct page *page, gfp_t gfp_mask) |
2cf85583 TH |
3798 | { |
3799 | struct swap_info_struct *si, *next; | |
6caa6a07 JW |
3800 | int nid = page_to_nid(page); |
3801 | ||
3802 | if (!(gfp_mask & __GFP_IO)) | |
2cf85583 TH |
3803 | return; |
3804 | ||
3805 | if (!blk_cgroup_congested()) | |
3806 | return; | |
3807 | ||
3808 | /* | |
3809 | * We've already scheduled a throttle, avoid taking the global swap | |
3810 | * lock. | |
3811 | */ | |
3812 | if (current->throttle_queue) | |
3813 | return; | |
3814 | ||
3815 | spin_lock(&swap_avail_lock); | |
6caa6a07 JW |
3816 | plist_for_each_entry_safe(si, next, &swap_avail_heads[nid], |
3817 | avail_lists[nid]) { | |
2cf85583 | 3818 | if (si->bdev) { |
6caa6a07 | 3819 | blkcg_schedule_throttle(bdev_get_queue(si->bdev), true); |
2cf85583 TH |
3820 | break; |
3821 | } | |
3822 | } | |
3823 | spin_unlock(&swap_avail_lock); | |
3824 | } | |
3825 | #endif | |
3826 | ||
a2468cc9 AL |
3827 | static int __init swapfile_init(void) |
3828 | { | |
3829 | int nid; | |
3830 | ||
3831 | swap_avail_heads = kmalloc_array(nr_node_ids, sizeof(struct plist_head), | |
3832 | GFP_KERNEL); | |
3833 | if (!swap_avail_heads) { | |
3834 | pr_emerg("Not enough memory for swap heads, swap is disabled\n"); | |
3835 | return -ENOMEM; | |
3836 | } | |
3837 | ||
3838 | for_each_node(nid) | |
3839 | plist_head_init(&swap_avail_heads[nid]); | |
3840 | ||
3841 | return 0; | |
3842 | } | |
3843 | subsys_initcall(swapfile_init); |