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