]>
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 LT |
42 | |
43 | #include <asm/pgtable.h> | |
44 | #include <asm/tlbflush.h> | |
45 | #include <linux/swapops.h> | |
5d1ea48b | 46 | #include <linux/swap_cgroup.h> |
1da177e4 | 47 | |
570a335b HD |
48 | static bool swap_count_continued(struct swap_info_struct *, pgoff_t, |
49 | unsigned char); | |
50 | static void free_swap_count_continuations(struct swap_info_struct *); | |
d4906e1a | 51 | static sector_t map_swap_entry(swp_entry_t, struct block_device**); |
570a335b | 52 | |
38b5faf4 | 53 | DEFINE_SPINLOCK(swap_lock); |
7c363b8c | 54 | static unsigned int nr_swapfiles; |
ec8acf20 | 55 | atomic_long_t nr_swap_pages; |
fb0fec50 CW |
56 | /* |
57 | * Some modules use swappable objects and may try to swap them out under | |
58 | * memory pressure (via the shrinker). Before doing so, they may wish to | |
59 | * check to see if any swap space is available. | |
60 | */ | |
61 | EXPORT_SYMBOL_GPL(nr_swap_pages); | |
ec8acf20 | 62 | /* protected with swap_lock. reading in vm_swap_full() doesn't need lock */ |
1da177e4 | 63 | long total_swap_pages; |
a2468cc9 | 64 | static int least_priority = -1; |
1da177e4 | 65 | |
1da177e4 LT |
66 | static const char Bad_file[] = "Bad swap file entry "; |
67 | static const char Unused_file[] = "Unused swap file entry "; | |
68 | static const char Bad_offset[] = "Bad swap offset entry "; | |
69 | static const char Unused_offset[] = "Unused swap offset entry "; | |
70 | ||
adfab836 DS |
71 | /* |
72 | * all active swap_info_structs | |
73 | * protected with swap_lock, and ordered by priority. | |
74 | */ | |
18ab4d4c DS |
75 | PLIST_HEAD(swap_active_head); |
76 | ||
77 | /* | |
78 | * all available (active, not full) swap_info_structs | |
79 | * protected with swap_avail_lock, ordered by priority. | |
80 | * This is used by get_swap_page() instead of swap_active_head | |
81 | * because swap_active_head includes all swap_info_structs, | |
82 | * but get_swap_page() doesn't need to look at full ones. | |
83 | * This uses its own lock instead of swap_lock because when a | |
84 | * swap_info_struct changes between not-full/full, it needs to | |
85 | * add/remove itself to/from this list, but the swap_info_struct->lock | |
86 | * is held and the locking order requires swap_lock to be taken | |
87 | * before any swap_info_struct->lock. | |
88 | */ | |
bfc6b1ca | 89 | static struct plist_head *swap_avail_heads; |
18ab4d4c | 90 | static DEFINE_SPINLOCK(swap_avail_lock); |
1da177e4 | 91 | |
38b5faf4 | 92 | struct swap_info_struct *swap_info[MAX_SWAPFILES]; |
1da177e4 | 93 | |
fc0abb14 | 94 | static DEFINE_MUTEX(swapon_mutex); |
1da177e4 | 95 | |
66d7dd51 KS |
96 | static DECLARE_WAIT_QUEUE_HEAD(proc_poll_wait); |
97 | /* Activity counter to indicate that a swapon or swapoff has occurred */ | |
98 | static atomic_t proc_poll_event = ATOMIC_INIT(0); | |
99 | ||
81a0298b HY |
100 | atomic_t nr_rotate_swap = ATOMIC_INIT(0); |
101 | ||
c10d38cc DJ |
102 | static struct swap_info_struct *swap_type_to_swap_info(int type) |
103 | { | |
104 | if (type >= READ_ONCE(nr_swapfiles)) | |
105 | return NULL; | |
106 | ||
107 | smp_rmb(); /* Pairs with smp_wmb in alloc_swap_info. */ | |
108 | return READ_ONCE(swap_info[type]); | |
109 | } | |
110 | ||
8d69aaee | 111 | static inline unsigned char swap_count(unsigned char ent) |
355cfa73 | 112 | { |
955c97f0 | 113 | return ent & ~SWAP_HAS_CACHE; /* may include COUNT_CONTINUED flag */ |
355cfa73 KH |
114 | } |
115 | ||
bcd49e86 HY |
116 | /* Reclaim the swap entry anyway if possible */ |
117 | #define TTRS_ANYWAY 0x1 | |
118 | /* | |
119 | * Reclaim the swap entry if there are no more mappings of the | |
120 | * corresponding page | |
121 | */ | |
122 | #define TTRS_UNMAPPED 0x2 | |
123 | /* Reclaim the swap entry if swap is getting full*/ | |
124 | #define TTRS_FULL 0x4 | |
125 | ||
efa90a98 | 126 | /* returns 1 if swap entry is freed */ |
bcd49e86 HY |
127 | static int __try_to_reclaim_swap(struct swap_info_struct *si, |
128 | unsigned long offset, unsigned long flags) | |
c9e44410 | 129 | { |
efa90a98 | 130 | swp_entry_t entry = swp_entry(si->type, offset); |
c9e44410 KH |
131 | struct page *page; |
132 | int ret = 0; | |
133 | ||
bcd49e86 | 134 | page = find_get_page(swap_address_space(entry), offset); |
c9e44410 KH |
135 | if (!page) |
136 | return 0; | |
137 | /* | |
bcd49e86 HY |
138 | * When this function is called from scan_swap_map_slots() and it's |
139 | * called by vmscan.c at reclaiming pages. So, we hold a lock on a page, | |
140 | * here. We have to use trylock for avoiding deadlock. This is a special | |
c9e44410 KH |
141 | * case and you should use try_to_free_swap() with explicit lock_page() |
142 | * in usual operations. | |
143 | */ | |
144 | if (trylock_page(page)) { | |
bcd49e86 HY |
145 | if ((flags & TTRS_ANYWAY) || |
146 | ((flags & TTRS_UNMAPPED) && !page_mapped(page)) || | |
147 | ((flags & TTRS_FULL) && mem_cgroup_swap_full(page))) | |
148 | ret = try_to_free_swap(page); | |
c9e44410 KH |
149 | unlock_page(page); |
150 | } | |
09cbfeaf | 151 | put_page(page); |
c9e44410 KH |
152 | return ret; |
153 | } | |
355cfa73 | 154 | |
4efaceb1 AL |
155 | static inline struct swap_extent *first_se(struct swap_info_struct *sis) |
156 | { | |
157 | struct rb_node *rb = rb_first(&sis->swap_extent_root); | |
158 | return rb_entry(rb, struct swap_extent, rb_node); | |
159 | } | |
160 | ||
161 | static inline struct swap_extent *next_se(struct swap_extent *se) | |
162 | { | |
163 | struct rb_node *rb = rb_next(&se->rb_node); | |
164 | return rb ? rb_entry(rb, struct swap_extent, rb_node) : NULL; | |
165 | } | |
166 | ||
6a6ba831 HD |
167 | /* |
168 | * swapon tell device that all the old swap contents can be discarded, | |
169 | * to allow the swap device to optimize its wear-levelling. | |
170 | */ | |
171 | static int discard_swap(struct swap_info_struct *si) | |
172 | { | |
173 | struct swap_extent *se; | |
9625a5f2 HD |
174 | sector_t start_block; |
175 | sector_t nr_blocks; | |
6a6ba831 HD |
176 | int err = 0; |
177 | ||
9625a5f2 | 178 | /* Do not discard the swap header page! */ |
4efaceb1 | 179 | se = first_se(si); |
9625a5f2 HD |
180 | start_block = (se->start_block + 1) << (PAGE_SHIFT - 9); |
181 | nr_blocks = ((sector_t)se->nr_pages - 1) << (PAGE_SHIFT - 9); | |
182 | if (nr_blocks) { | |
183 | err = blkdev_issue_discard(si->bdev, start_block, | |
dd3932ed | 184 | nr_blocks, GFP_KERNEL, 0); |
9625a5f2 HD |
185 | if (err) |
186 | return err; | |
187 | cond_resched(); | |
188 | } | |
6a6ba831 | 189 | |
4efaceb1 | 190 | for (se = next_se(se); se; se = next_se(se)) { |
9625a5f2 HD |
191 | start_block = se->start_block << (PAGE_SHIFT - 9); |
192 | nr_blocks = (sector_t)se->nr_pages << (PAGE_SHIFT - 9); | |
6a6ba831 HD |
193 | |
194 | err = blkdev_issue_discard(si->bdev, start_block, | |
dd3932ed | 195 | nr_blocks, GFP_KERNEL, 0); |
6a6ba831 HD |
196 | if (err) |
197 | break; | |
198 | ||
199 | cond_resched(); | |
200 | } | |
201 | return err; /* That will often be -EOPNOTSUPP */ | |
202 | } | |
203 | ||
4efaceb1 AL |
204 | static struct swap_extent * |
205 | offset_to_swap_extent(struct swap_info_struct *sis, unsigned long offset) | |
206 | { | |
207 | struct swap_extent *se; | |
208 | struct rb_node *rb; | |
209 | ||
210 | rb = sis->swap_extent_root.rb_node; | |
211 | while (rb) { | |
212 | se = rb_entry(rb, struct swap_extent, rb_node); | |
213 | if (offset < se->start_page) | |
214 | rb = rb->rb_left; | |
215 | else if (offset >= se->start_page + se->nr_pages) | |
216 | rb = rb->rb_right; | |
217 | else | |
218 | return se; | |
219 | } | |
220 | /* It *must* be present */ | |
221 | BUG(); | |
222 | } | |
223 | ||
7992fde7 HD |
224 | /* |
225 | * swap allocation tell device that a cluster of swap can now be discarded, | |
226 | * to allow the swap device to optimize its wear-levelling. | |
227 | */ | |
228 | static void discard_swap_cluster(struct swap_info_struct *si, | |
229 | pgoff_t start_page, pgoff_t nr_pages) | |
230 | { | |
4efaceb1 | 231 | struct swap_extent *se = offset_to_swap_extent(si, start_page); |
7992fde7 HD |
232 | |
233 | while (nr_pages) { | |
4efaceb1 AL |
234 | pgoff_t offset = start_page - se->start_page; |
235 | sector_t start_block = se->start_block + offset; | |
236 | sector_t nr_blocks = se->nr_pages - offset; | |
237 | ||
238 | if (nr_blocks > nr_pages) | |
239 | nr_blocks = nr_pages; | |
240 | start_page += nr_blocks; | |
241 | nr_pages -= nr_blocks; | |
242 | ||
243 | start_block <<= PAGE_SHIFT - 9; | |
244 | nr_blocks <<= PAGE_SHIFT - 9; | |
245 | if (blkdev_issue_discard(si->bdev, start_block, | |
246 | nr_blocks, GFP_NOIO, 0)) | |
247 | break; | |
7992fde7 | 248 | |
4efaceb1 | 249 | se = next_se(se); |
7992fde7 HD |
250 | } |
251 | } | |
252 | ||
38d8b4e6 HY |
253 | #ifdef CONFIG_THP_SWAP |
254 | #define SWAPFILE_CLUSTER HPAGE_PMD_NR | |
a448f2d0 HY |
255 | |
256 | #define swap_entry_size(size) (size) | |
38d8b4e6 | 257 | #else |
048c27fd | 258 | #define SWAPFILE_CLUSTER 256 |
a448f2d0 HY |
259 | |
260 | /* | |
261 | * Define swap_entry_size() as constant to let compiler to optimize | |
262 | * out some code if !CONFIG_THP_SWAP | |
263 | */ | |
264 | #define swap_entry_size(size) 1 | |
38d8b4e6 | 265 | #endif |
048c27fd HD |
266 | #define LATENCY_LIMIT 256 |
267 | ||
2a8f9449 SL |
268 | static inline void cluster_set_flag(struct swap_cluster_info *info, |
269 | unsigned int flag) | |
270 | { | |
271 | info->flags = flag; | |
272 | } | |
273 | ||
274 | static inline unsigned int cluster_count(struct swap_cluster_info *info) | |
275 | { | |
276 | return info->data; | |
277 | } | |
278 | ||
279 | static inline void cluster_set_count(struct swap_cluster_info *info, | |
280 | unsigned int c) | |
281 | { | |
282 | info->data = c; | |
283 | } | |
284 | ||
285 | static inline void cluster_set_count_flag(struct swap_cluster_info *info, | |
286 | unsigned int c, unsigned int f) | |
287 | { | |
288 | info->flags = f; | |
289 | info->data = c; | |
290 | } | |
291 | ||
292 | static inline unsigned int cluster_next(struct swap_cluster_info *info) | |
293 | { | |
294 | return info->data; | |
295 | } | |
296 | ||
297 | static inline void cluster_set_next(struct swap_cluster_info *info, | |
298 | unsigned int n) | |
299 | { | |
300 | info->data = n; | |
301 | } | |
302 | ||
303 | static inline void cluster_set_next_flag(struct swap_cluster_info *info, | |
304 | unsigned int n, unsigned int f) | |
305 | { | |
306 | info->flags = f; | |
307 | info->data = n; | |
308 | } | |
309 | ||
310 | static inline bool cluster_is_free(struct swap_cluster_info *info) | |
311 | { | |
312 | return info->flags & CLUSTER_FLAG_FREE; | |
313 | } | |
314 | ||
315 | static inline bool cluster_is_null(struct swap_cluster_info *info) | |
316 | { | |
317 | return info->flags & CLUSTER_FLAG_NEXT_NULL; | |
318 | } | |
319 | ||
320 | static inline void cluster_set_null(struct swap_cluster_info *info) | |
321 | { | |
322 | info->flags = CLUSTER_FLAG_NEXT_NULL; | |
323 | info->data = 0; | |
324 | } | |
325 | ||
e0709829 HY |
326 | static inline bool cluster_is_huge(struct swap_cluster_info *info) |
327 | { | |
33ee011e HY |
328 | if (IS_ENABLED(CONFIG_THP_SWAP)) |
329 | return info->flags & CLUSTER_FLAG_HUGE; | |
330 | return false; | |
e0709829 HY |
331 | } |
332 | ||
333 | static inline void cluster_clear_huge(struct swap_cluster_info *info) | |
334 | { | |
335 | info->flags &= ~CLUSTER_FLAG_HUGE; | |
336 | } | |
337 | ||
235b6217 HY |
338 | static inline struct swap_cluster_info *lock_cluster(struct swap_info_struct *si, |
339 | unsigned long offset) | |
340 | { | |
341 | struct swap_cluster_info *ci; | |
342 | ||
343 | ci = si->cluster_info; | |
344 | if (ci) { | |
345 | ci += offset / SWAPFILE_CLUSTER; | |
346 | spin_lock(&ci->lock); | |
347 | } | |
348 | return ci; | |
349 | } | |
350 | ||
351 | static inline void unlock_cluster(struct swap_cluster_info *ci) | |
352 | { | |
353 | if (ci) | |
354 | spin_unlock(&ci->lock); | |
355 | } | |
356 | ||
59d98bf3 HY |
357 | /* |
358 | * Determine the locking method in use for this device. Return | |
359 | * swap_cluster_info if SSD-style cluster-based locking is in place. | |
360 | */ | |
235b6217 | 361 | static inline struct swap_cluster_info *lock_cluster_or_swap_info( |
59d98bf3 | 362 | struct swap_info_struct *si, unsigned long offset) |
235b6217 HY |
363 | { |
364 | struct swap_cluster_info *ci; | |
365 | ||
59d98bf3 | 366 | /* Try to use fine-grained SSD-style locking if available: */ |
235b6217 | 367 | ci = lock_cluster(si, offset); |
59d98bf3 | 368 | /* Otherwise, fall back to traditional, coarse locking: */ |
235b6217 HY |
369 | if (!ci) |
370 | spin_lock(&si->lock); | |
371 | ||
372 | return ci; | |
373 | } | |
374 | ||
375 | static inline void unlock_cluster_or_swap_info(struct swap_info_struct *si, | |
376 | struct swap_cluster_info *ci) | |
377 | { | |
378 | if (ci) | |
379 | unlock_cluster(ci); | |
380 | else | |
381 | spin_unlock(&si->lock); | |
382 | } | |
383 | ||
6b534915 HY |
384 | static inline bool cluster_list_empty(struct swap_cluster_list *list) |
385 | { | |
386 | return cluster_is_null(&list->head); | |
387 | } | |
388 | ||
389 | static inline unsigned int cluster_list_first(struct swap_cluster_list *list) | |
390 | { | |
391 | return cluster_next(&list->head); | |
392 | } | |
393 | ||
394 | static void cluster_list_init(struct swap_cluster_list *list) | |
395 | { | |
396 | cluster_set_null(&list->head); | |
397 | cluster_set_null(&list->tail); | |
398 | } | |
399 | ||
400 | static void cluster_list_add_tail(struct swap_cluster_list *list, | |
401 | struct swap_cluster_info *ci, | |
402 | unsigned int idx) | |
403 | { | |
404 | if (cluster_list_empty(list)) { | |
405 | cluster_set_next_flag(&list->head, idx, 0); | |
406 | cluster_set_next_flag(&list->tail, idx, 0); | |
407 | } else { | |
235b6217 | 408 | struct swap_cluster_info *ci_tail; |
6b534915 HY |
409 | unsigned int tail = cluster_next(&list->tail); |
410 | ||
235b6217 HY |
411 | /* |
412 | * Nested cluster lock, but both cluster locks are | |
413 | * only acquired when we held swap_info_struct->lock | |
414 | */ | |
415 | ci_tail = ci + tail; | |
416 | spin_lock_nested(&ci_tail->lock, SINGLE_DEPTH_NESTING); | |
417 | cluster_set_next(ci_tail, idx); | |
0ef017d1 | 418 | spin_unlock(&ci_tail->lock); |
6b534915 HY |
419 | cluster_set_next_flag(&list->tail, idx, 0); |
420 | } | |
421 | } | |
422 | ||
423 | static unsigned int cluster_list_del_first(struct swap_cluster_list *list, | |
424 | struct swap_cluster_info *ci) | |
425 | { | |
426 | unsigned int idx; | |
427 | ||
428 | idx = cluster_next(&list->head); | |
429 | if (cluster_next(&list->tail) == idx) { | |
430 | cluster_set_null(&list->head); | |
431 | cluster_set_null(&list->tail); | |
432 | } else | |
433 | cluster_set_next_flag(&list->head, | |
434 | cluster_next(&ci[idx]), 0); | |
435 | ||
436 | return idx; | |
437 | } | |
438 | ||
815c2c54 SL |
439 | /* Add a cluster to discard list and schedule it to do discard */ |
440 | static void swap_cluster_schedule_discard(struct swap_info_struct *si, | |
441 | unsigned int idx) | |
442 | { | |
443 | /* | |
444 | * If scan_swap_map() can't find a free cluster, it will check | |
445 | * si->swap_map directly. To make sure the discarding cluster isn't | |
446 | * taken by scan_swap_map(), mark the swap entries bad (occupied). It | |
447 | * will be cleared after discard | |
448 | */ | |
449 | memset(si->swap_map + idx * SWAPFILE_CLUSTER, | |
450 | SWAP_MAP_BAD, SWAPFILE_CLUSTER); | |
451 | ||
6b534915 | 452 | cluster_list_add_tail(&si->discard_clusters, si->cluster_info, idx); |
815c2c54 SL |
453 | |
454 | schedule_work(&si->discard_work); | |
455 | } | |
456 | ||
38d8b4e6 HY |
457 | static void __free_cluster(struct swap_info_struct *si, unsigned long idx) |
458 | { | |
459 | struct swap_cluster_info *ci = si->cluster_info; | |
460 | ||
461 | cluster_set_flag(ci + idx, CLUSTER_FLAG_FREE); | |
462 | cluster_list_add_tail(&si->free_clusters, ci, idx); | |
463 | } | |
464 | ||
815c2c54 SL |
465 | /* |
466 | * Doing discard actually. After a cluster discard is finished, the cluster | |
467 | * will be added to free cluster list. caller should hold si->lock. | |
468 | */ | |
469 | static void swap_do_scheduled_discard(struct swap_info_struct *si) | |
470 | { | |
235b6217 | 471 | struct swap_cluster_info *info, *ci; |
815c2c54 SL |
472 | unsigned int idx; |
473 | ||
474 | info = si->cluster_info; | |
475 | ||
6b534915 HY |
476 | while (!cluster_list_empty(&si->discard_clusters)) { |
477 | idx = cluster_list_del_first(&si->discard_clusters, info); | |
815c2c54 SL |
478 | spin_unlock(&si->lock); |
479 | ||
480 | discard_swap_cluster(si, idx * SWAPFILE_CLUSTER, | |
481 | SWAPFILE_CLUSTER); | |
482 | ||
483 | spin_lock(&si->lock); | |
235b6217 | 484 | ci = lock_cluster(si, idx * SWAPFILE_CLUSTER); |
38d8b4e6 | 485 | __free_cluster(si, idx); |
815c2c54 SL |
486 | memset(si->swap_map + idx * SWAPFILE_CLUSTER, |
487 | 0, SWAPFILE_CLUSTER); | |
235b6217 | 488 | unlock_cluster(ci); |
815c2c54 SL |
489 | } |
490 | } | |
491 | ||
492 | static void swap_discard_work(struct work_struct *work) | |
493 | { | |
494 | struct swap_info_struct *si; | |
495 | ||
496 | si = container_of(work, struct swap_info_struct, discard_work); | |
497 | ||
498 | spin_lock(&si->lock); | |
499 | swap_do_scheduled_discard(si); | |
500 | spin_unlock(&si->lock); | |
501 | } | |
502 | ||
38d8b4e6 HY |
503 | static void alloc_cluster(struct swap_info_struct *si, unsigned long idx) |
504 | { | |
505 | struct swap_cluster_info *ci = si->cluster_info; | |
506 | ||
507 | VM_BUG_ON(cluster_list_first(&si->free_clusters) != idx); | |
508 | cluster_list_del_first(&si->free_clusters, ci); | |
509 | cluster_set_count_flag(ci + idx, 0, 0); | |
510 | } | |
511 | ||
512 | static void free_cluster(struct swap_info_struct *si, unsigned long idx) | |
513 | { | |
514 | struct swap_cluster_info *ci = si->cluster_info + idx; | |
515 | ||
516 | VM_BUG_ON(cluster_count(ci) != 0); | |
517 | /* | |
518 | * If the swap is discardable, prepare discard the cluster | |
519 | * instead of free it immediately. The cluster will be freed | |
520 | * after discard. | |
521 | */ | |
522 | if ((si->flags & (SWP_WRITEOK | SWP_PAGE_DISCARD)) == | |
523 | (SWP_WRITEOK | SWP_PAGE_DISCARD)) { | |
524 | swap_cluster_schedule_discard(si, idx); | |
525 | return; | |
526 | } | |
527 | ||
528 | __free_cluster(si, idx); | |
529 | } | |
530 | ||
2a8f9449 SL |
531 | /* |
532 | * The cluster corresponding to page_nr will be used. The cluster will be | |
533 | * removed from free cluster list and its usage counter will be increased. | |
534 | */ | |
535 | static void inc_cluster_info_page(struct swap_info_struct *p, | |
536 | struct swap_cluster_info *cluster_info, unsigned long page_nr) | |
537 | { | |
538 | unsigned long idx = page_nr / SWAPFILE_CLUSTER; | |
539 | ||
540 | if (!cluster_info) | |
541 | return; | |
38d8b4e6 HY |
542 | if (cluster_is_free(&cluster_info[idx])) |
543 | alloc_cluster(p, idx); | |
2a8f9449 SL |
544 | |
545 | VM_BUG_ON(cluster_count(&cluster_info[idx]) >= SWAPFILE_CLUSTER); | |
546 | cluster_set_count(&cluster_info[idx], | |
547 | cluster_count(&cluster_info[idx]) + 1); | |
548 | } | |
549 | ||
550 | /* | |
551 | * The cluster corresponding to page_nr decreases one usage. If the usage | |
552 | * counter becomes 0, which means no page in the cluster is in using, we can | |
553 | * optionally discard the cluster and add it to free cluster list. | |
554 | */ | |
555 | static void dec_cluster_info_page(struct swap_info_struct *p, | |
556 | struct swap_cluster_info *cluster_info, unsigned long page_nr) | |
557 | { | |
558 | unsigned long idx = page_nr / SWAPFILE_CLUSTER; | |
559 | ||
560 | if (!cluster_info) | |
561 | return; | |
562 | ||
563 | VM_BUG_ON(cluster_count(&cluster_info[idx]) == 0); | |
564 | cluster_set_count(&cluster_info[idx], | |
565 | cluster_count(&cluster_info[idx]) - 1); | |
566 | ||
38d8b4e6 HY |
567 | if (cluster_count(&cluster_info[idx]) == 0) |
568 | free_cluster(p, idx); | |
2a8f9449 SL |
569 | } |
570 | ||
571 | /* | |
572 | * It's possible scan_swap_map() uses a free cluster in the middle of free | |
573 | * cluster list. Avoiding such abuse to avoid list corruption. | |
574 | */ | |
ebc2a1a6 SL |
575 | static bool |
576 | scan_swap_map_ssd_cluster_conflict(struct swap_info_struct *si, | |
2a8f9449 SL |
577 | unsigned long offset) |
578 | { | |
ebc2a1a6 SL |
579 | struct percpu_cluster *percpu_cluster; |
580 | bool conflict; | |
581 | ||
2a8f9449 | 582 | offset /= SWAPFILE_CLUSTER; |
6b534915 HY |
583 | conflict = !cluster_list_empty(&si->free_clusters) && |
584 | offset != cluster_list_first(&si->free_clusters) && | |
2a8f9449 | 585 | cluster_is_free(&si->cluster_info[offset]); |
ebc2a1a6 SL |
586 | |
587 | if (!conflict) | |
588 | return false; | |
589 | ||
590 | percpu_cluster = this_cpu_ptr(si->percpu_cluster); | |
591 | cluster_set_null(&percpu_cluster->index); | |
592 | return true; | |
593 | } | |
594 | ||
595 | /* | |
596 | * Try to get a swap entry from current cpu's swap entry pool (a cluster). This | |
597 | * might involve allocating a new cluster for current CPU too. | |
598 | */ | |
36005bae | 599 | static bool scan_swap_map_try_ssd_cluster(struct swap_info_struct *si, |
ebc2a1a6 SL |
600 | unsigned long *offset, unsigned long *scan_base) |
601 | { | |
602 | struct percpu_cluster *cluster; | |
235b6217 | 603 | struct swap_cluster_info *ci; |
235b6217 | 604 | unsigned long tmp, max; |
ebc2a1a6 SL |
605 | |
606 | new_cluster: | |
607 | cluster = this_cpu_ptr(si->percpu_cluster); | |
608 | if (cluster_is_null(&cluster->index)) { | |
6b534915 HY |
609 | if (!cluster_list_empty(&si->free_clusters)) { |
610 | cluster->index = si->free_clusters.head; | |
ebc2a1a6 SL |
611 | cluster->next = cluster_next(&cluster->index) * |
612 | SWAPFILE_CLUSTER; | |
6b534915 | 613 | } else if (!cluster_list_empty(&si->discard_clusters)) { |
ebc2a1a6 SL |
614 | /* |
615 | * we don't have free cluster but have some clusters in | |
49070588 HY |
616 | * discarding, do discard now and reclaim them, then |
617 | * reread cluster_next_cpu since we dropped si->lock | |
ebc2a1a6 SL |
618 | */ |
619 | swap_do_scheduled_discard(si); | |
49070588 HY |
620 | *scan_base = this_cpu_read(*si->cluster_next_cpu); |
621 | *offset = *scan_base; | |
ebc2a1a6 SL |
622 | goto new_cluster; |
623 | } else | |
36005bae | 624 | return false; |
ebc2a1a6 SL |
625 | } |
626 | ||
ebc2a1a6 SL |
627 | /* |
628 | * Other CPUs can use our cluster if they can't find a free cluster, | |
629 | * check if there is still free entry in the cluster | |
630 | */ | |
631 | tmp = cluster->next; | |
235b6217 HY |
632 | max = min_t(unsigned long, si->max, |
633 | (cluster_next(&cluster->index) + 1) * SWAPFILE_CLUSTER); | |
7b9e2de1 WY |
634 | if (tmp < max) { |
635 | ci = lock_cluster(si, tmp); | |
636 | while (tmp < max) { | |
637 | if (!si->swap_map[tmp]) | |
638 | break; | |
639 | tmp++; | |
640 | } | |
641 | unlock_cluster(ci); | |
ebc2a1a6 | 642 | } |
0fd0e19e | 643 | if (tmp >= max) { |
ebc2a1a6 SL |
644 | cluster_set_null(&cluster->index); |
645 | goto new_cluster; | |
646 | } | |
647 | cluster->next = tmp + 1; | |
648 | *offset = tmp; | |
649 | *scan_base = tmp; | |
fdff1deb | 650 | return true; |
2a8f9449 SL |
651 | } |
652 | ||
a2468cc9 AL |
653 | static void __del_from_avail_list(struct swap_info_struct *p) |
654 | { | |
655 | int nid; | |
656 | ||
657 | for_each_node(nid) | |
658 | plist_del(&p->avail_lists[nid], &swap_avail_heads[nid]); | |
659 | } | |
660 | ||
661 | static void del_from_avail_list(struct swap_info_struct *p) | |
662 | { | |
663 | spin_lock(&swap_avail_lock); | |
664 | __del_from_avail_list(p); | |
665 | spin_unlock(&swap_avail_lock); | |
666 | } | |
667 | ||
38d8b4e6 HY |
668 | static void swap_range_alloc(struct swap_info_struct *si, unsigned long offset, |
669 | unsigned int nr_entries) | |
670 | { | |
671 | unsigned int end = offset + nr_entries - 1; | |
672 | ||
673 | if (offset == si->lowest_bit) | |
674 | si->lowest_bit += nr_entries; | |
675 | if (end == si->highest_bit) | |
676 | si->highest_bit -= nr_entries; | |
677 | si->inuse_pages += nr_entries; | |
678 | if (si->inuse_pages == si->pages) { | |
679 | si->lowest_bit = si->max; | |
680 | si->highest_bit = 0; | |
a2468cc9 | 681 | del_from_avail_list(si); |
38d8b4e6 HY |
682 | } |
683 | } | |
684 | ||
a2468cc9 AL |
685 | static void add_to_avail_list(struct swap_info_struct *p) |
686 | { | |
687 | int nid; | |
688 | ||
689 | spin_lock(&swap_avail_lock); | |
690 | for_each_node(nid) { | |
691 | WARN_ON(!plist_node_empty(&p->avail_lists[nid])); | |
692 | plist_add(&p->avail_lists[nid], &swap_avail_heads[nid]); | |
693 | } | |
694 | spin_unlock(&swap_avail_lock); | |
695 | } | |
696 | ||
38d8b4e6 HY |
697 | static void swap_range_free(struct swap_info_struct *si, unsigned long offset, |
698 | unsigned int nr_entries) | |
699 | { | |
700 | unsigned long end = offset + nr_entries - 1; | |
701 | void (*swap_slot_free_notify)(struct block_device *, unsigned long); | |
702 | ||
703 | if (offset < si->lowest_bit) | |
704 | si->lowest_bit = offset; | |
705 | if (end > si->highest_bit) { | |
706 | bool was_full = !si->highest_bit; | |
707 | ||
708 | si->highest_bit = end; | |
a2468cc9 AL |
709 | if (was_full && (si->flags & SWP_WRITEOK)) |
710 | add_to_avail_list(si); | |
38d8b4e6 HY |
711 | } |
712 | atomic_long_add(nr_entries, &nr_swap_pages); | |
713 | si->inuse_pages -= nr_entries; | |
714 | if (si->flags & SWP_BLKDEV) | |
715 | swap_slot_free_notify = | |
716 | si->bdev->bd_disk->fops->swap_slot_free_notify; | |
717 | else | |
718 | swap_slot_free_notify = NULL; | |
719 | while (offset <= end) { | |
720 | frontswap_invalidate_page(si->type, offset); | |
721 | if (swap_slot_free_notify) | |
722 | swap_slot_free_notify(si->bdev, offset); | |
723 | offset++; | |
724 | } | |
725 | } | |
726 | ||
49070588 HY |
727 | static void set_cluster_next(struct swap_info_struct *si, unsigned long next) |
728 | { | |
729 | unsigned long prev; | |
730 | ||
731 | if (!(si->flags & SWP_SOLIDSTATE)) { | |
732 | si->cluster_next = next; | |
733 | return; | |
734 | } | |
735 | ||
736 | prev = this_cpu_read(*si->cluster_next_cpu); | |
737 | /* | |
738 | * Cross the swap address space size aligned trunk, choose | |
739 | * another trunk randomly to avoid lock contention on swap | |
740 | * address space if possible. | |
741 | */ | |
742 | if ((prev >> SWAP_ADDRESS_SPACE_SHIFT) != | |
743 | (next >> SWAP_ADDRESS_SPACE_SHIFT)) { | |
744 | /* No free swap slots available */ | |
745 | if (si->highest_bit <= si->lowest_bit) | |
746 | return; | |
747 | next = si->lowest_bit + | |
748 | prandom_u32_max(si->highest_bit - si->lowest_bit + 1); | |
749 | next = ALIGN_DOWN(next, SWAP_ADDRESS_SPACE_PAGES); | |
750 | next = max_t(unsigned int, next, si->lowest_bit); | |
751 | } | |
752 | this_cpu_write(*si->cluster_next_cpu, next); | |
753 | } | |
754 | ||
36005bae TC |
755 | static int scan_swap_map_slots(struct swap_info_struct *si, |
756 | unsigned char usage, int nr, | |
757 | swp_entry_t slots[]) | |
1da177e4 | 758 | { |
235b6217 | 759 | struct swap_cluster_info *ci; |
ebebbbe9 | 760 | unsigned long offset; |
c60aa176 | 761 | unsigned long scan_base; |
7992fde7 | 762 | unsigned long last_in_cluster = 0; |
048c27fd | 763 | int latency_ration = LATENCY_LIMIT; |
36005bae | 764 | int n_ret = 0; |
ed43af10 | 765 | bool scanned_many = false; |
36005bae | 766 | |
886bb7e9 | 767 | /* |
7dfad418 HD |
768 | * We try to cluster swap pages by allocating them sequentially |
769 | * in swap. Once we've allocated SWAPFILE_CLUSTER pages this | |
770 | * way, however, we resort to first-free allocation, starting | |
771 | * a new cluster. This prevents us from scattering swap pages | |
772 | * all over the entire swap partition, so that we reduce | |
773 | * overall disk seek times between swap pages. -- sct | |
774 | * But we do now try to find an empty cluster. -Andrea | |
c60aa176 | 775 | * And we let swap pages go all over an SSD partition. Hugh |
7dfad418 HD |
776 | */ |
777 | ||
52b7efdb | 778 | si->flags += SWP_SCANNING; |
49070588 HY |
779 | /* |
780 | * Use percpu scan base for SSD to reduce lock contention on | |
781 | * cluster and swap cache. For HDD, sequential access is more | |
782 | * important. | |
783 | */ | |
784 | if (si->flags & SWP_SOLIDSTATE) | |
785 | scan_base = this_cpu_read(*si->cluster_next_cpu); | |
786 | else | |
787 | scan_base = si->cluster_next; | |
788 | offset = scan_base; | |
ebebbbe9 | 789 | |
ebc2a1a6 SL |
790 | /* SSD algorithm */ |
791 | if (si->cluster_info) { | |
bd2d18da | 792 | if (!scan_swap_map_try_ssd_cluster(si, &offset, &scan_base)) |
36005bae | 793 | goto scan; |
f4eaf51a | 794 | } else if (unlikely(!si->cluster_nr--)) { |
ebebbbe9 HD |
795 | if (si->pages - si->inuse_pages < SWAPFILE_CLUSTER) { |
796 | si->cluster_nr = SWAPFILE_CLUSTER - 1; | |
797 | goto checks; | |
798 | } | |
2a8f9449 | 799 | |
ec8acf20 | 800 | spin_unlock(&si->lock); |
7dfad418 | 801 | |
c60aa176 HD |
802 | /* |
803 | * If seek is expensive, start searching for new cluster from | |
804 | * start of partition, to minimize the span of allocated swap. | |
50088c44 CY |
805 | * If seek is cheap, that is the SWP_SOLIDSTATE si->cluster_info |
806 | * case, just handled by scan_swap_map_try_ssd_cluster() above. | |
c60aa176 | 807 | */ |
50088c44 | 808 | scan_base = offset = si->lowest_bit; |
7dfad418 HD |
809 | last_in_cluster = offset + SWAPFILE_CLUSTER - 1; |
810 | ||
811 | /* Locate the first empty (unaligned) cluster */ | |
812 | for (; last_in_cluster <= si->highest_bit; offset++) { | |
1da177e4 | 813 | if (si->swap_map[offset]) |
7dfad418 HD |
814 | last_in_cluster = offset + SWAPFILE_CLUSTER; |
815 | else if (offset == last_in_cluster) { | |
ec8acf20 | 816 | spin_lock(&si->lock); |
ebebbbe9 HD |
817 | offset -= SWAPFILE_CLUSTER - 1; |
818 | si->cluster_next = offset; | |
819 | si->cluster_nr = SWAPFILE_CLUSTER - 1; | |
c60aa176 HD |
820 | goto checks; |
821 | } | |
822 | if (unlikely(--latency_ration < 0)) { | |
823 | cond_resched(); | |
824 | latency_ration = LATENCY_LIMIT; | |
825 | } | |
826 | } | |
827 | ||
828 | offset = scan_base; | |
ec8acf20 | 829 | spin_lock(&si->lock); |
ebebbbe9 | 830 | si->cluster_nr = SWAPFILE_CLUSTER - 1; |
1da177e4 | 831 | } |
7dfad418 | 832 | |
ebebbbe9 | 833 | checks: |
ebc2a1a6 | 834 | if (si->cluster_info) { |
36005bae TC |
835 | while (scan_swap_map_ssd_cluster_conflict(si, offset)) { |
836 | /* take a break if we already got some slots */ | |
837 | if (n_ret) | |
838 | goto done; | |
839 | if (!scan_swap_map_try_ssd_cluster(si, &offset, | |
840 | &scan_base)) | |
841 | goto scan; | |
842 | } | |
ebc2a1a6 | 843 | } |
ebebbbe9 | 844 | if (!(si->flags & SWP_WRITEOK)) |
52b7efdb | 845 | goto no_page; |
7dfad418 HD |
846 | if (!si->highest_bit) |
847 | goto no_page; | |
ebebbbe9 | 848 | if (offset > si->highest_bit) |
c60aa176 | 849 | scan_base = offset = si->lowest_bit; |
c9e44410 | 850 | |
235b6217 | 851 | ci = lock_cluster(si, offset); |
b73d7fce HD |
852 | /* reuse swap entry of cache-only swap if not busy. */ |
853 | if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) { | |
c9e44410 | 854 | int swap_was_freed; |
235b6217 | 855 | unlock_cluster(ci); |
ec8acf20 | 856 | spin_unlock(&si->lock); |
bcd49e86 | 857 | swap_was_freed = __try_to_reclaim_swap(si, offset, TTRS_ANYWAY); |
ec8acf20 | 858 | spin_lock(&si->lock); |
c9e44410 KH |
859 | /* entry was freed successfully, try to use this again */ |
860 | if (swap_was_freed) | |
861 | goto checks; | |
862 | goto scan; /* check next one */ | |
863 | } | |
864 | ||
235b6217 HY |
865 | if (si->swap_map[offset]) { |
866 | unlock_cluster(ci); | |
36005bae TC |
867 | if (!n_ret) |
868 | goto scan; | |
869 | else | |
870 | goto done; | |
235b6217 | 871 | } |
2872bb2d HY |
872 | si->swap_map[offset] = usage; |
873 | inc_cluster_info_page(si, si->cluster_info, offset); | |
874 | unlock_cluster(ci); | |
ebebbbe9 | 875 | |
38d8b4e6 | 876 | swap_range_alloc(si, offset, 1); |
36005bae TC |
877 | slots[n_ret++] = swp_entry(si->type, offset); |
878 | ||
879 | /* got enough slots or reach max slots? */ | |
880 | if ((n_ret == nr) || (offset >= si->highest_bit)) | |
881 | goto done; | |
882 | ||
883 | /* search for next available slot */ | |
884 | ||
885 | /* time to take a break? */ | |
886 | if (unlikely(--latency_ration < 0)) { | |
887 | if (n_ret) | |
888 | goto done; | |
889 | spin_unlock(&si->lock); | |
890 | cond_resched(); | |
891 | spin_lock(&si->lock); | |
892 | latency_ration = LATENCY_LIMIT; | |
893 | } | |
894 | ||
895 | /* try to get more slots in cluster */ | |
896 | if (si->cluster_info) { | |
897 | if (scan_swap_map_try_ssd_cluster(si, &offset, &scan_base)) | |
898 | goto checks; | |
f4eaf51a WY |
899 | } else if (si->cluster_nr && !si->swap_map[++offset]) { |
900 | /* non-ssd case, still more slots in cluster? */ | |
36005bae TC |
901 | --si->cluster_nr; |
902 | goto checks; | |
903 | } | |
7992fde7 | 904 | |
ed43af10 HY |
905 | /* |
906 | * Even if there's no free clusters available (fragmented), | |
907 | * try to scan a little more quickly with lock held unless we | |
908 | * have scanned too many slots already. | |
909 | */ | |
910 | if (!scanned_many) { | |
911 | unsigned long scan_limit; | |
912 | ||
913 | if (offset < scan_base) | |
914 | scan_limit = scan_base; | |
915 | else | |
916 | scan_limit = si->highest_bit; | |
917 | for (; offset <= scan_limit && --latency_ration > 0; | |
918 | offset++) { | |
919 | if (!si->swap_map[offset]) | |
920 | goto checks; | |
921 | } | |
922 | } | |
923 | ||
36005bae | 924 | done: |
49070588 | 925 | set_cluster_next(si, offset + 1); |
36005bae TC |
926 | si->flags -= SWP_SCANNING; |
927 | return n_ret; | |
7dfad418 | 928 | |
ebebbbe9 | 929 | scan: |
ec8acf20 | 930 | spin_unlock(&si->lock); |
7dfad418 | 931 | while (++offset <= si->highest_bit) { |
52b7efdb | 932 | if (!si->swap_map[offset]) { |
ec8acf20 | 933 | spin_lock(&si->lock); |
52b7efdb HD |
934 | goto checks; |
935 | } | |
c9e44410 | 936 | if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) { |
ec8acf20 | 937 | spin_lock(&si->lock); |
c9e44410 KH |
938 | goto checks; |
939 | } | |
048c27fd HD |
940 | if (unlikely(--latency_ration < 0)) { |
941 | cond_resched(); | |
942 | latency_ration = LATENCY_LIMIT; | |
ed43af10 | 943 | scanned_many = true; |
048c27fd | 944 | } |
7dfad418 | 945 | } |
c60aa176 | 946 | offset = si->lowest_bit; |
a5998061 | 947 | while (offset < scan_base) { |
c60aa176 | 948 | if (!si->swap_map[offset]) { |
ec8acf20 | 949 | spin_lock(&si->lock); |
c60aa176 HD |
950 | goto checks; |
951 | } | |
c9e44410 | 952 | if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) { |
ec8acf20 | 953 | spin_lock(&si->lock); |
c9e44410 KH |
954 | goto checks; |
955 | } | |
c60aa176 HD |
956 | if (unlikely(--latency_ration < 0)) { |
957 | cond_resched(); | |
958 | latency_ration = LATENCY_LIMIT; | |
ed43af10 | 959 | scanned_many = true; |
c60aa176 | 960 | } |
a5998061 | 961 | offset++; |
c60aa176 | 962 | } |
ec8acf20 | 963 | spin_lock(&si->lock); |
7dfad418 HD |
964 | |
965 | no_page: | |
52b7efdb | 966 | si->flags -= SWP_SCANNING; |
36005bae | 967 | return n_ret; |
1da177e4 LT |
968 | } |
969 | ||
38d8b4e6 HY |
970 | static int swap_alloc_cluster(struct swap_info_struct *si, swp_entry_t *slot) |
971 | { | |
972 | unsigned long idx; | |
973 | struct swap_cluster_info *ci; | |
974 | unsigned long offset, i; | |
975 | unsigned char *map; | |
976 | ||
fe5266d5 HY |
977 | /* |
978 | * Should not even be attempting cluster allocations when huge | |
979 | * page swap is disabled. Warn and fail the allocation. | |
980 | */ | |
981 | if (!IS_ENABLED(CONFIG_THP_SWAP)) { | |
982 | VM_WARN_ON_ONCE(1); | |
983 | return 0; | |
984 | } | |
985 | ||
38d8b4e6 HY |
986 | if (cluster_list_empty(&si->free_clusters)) |
987 | return 0; | |
988 | ||
989 | idx = cluster_list_first(&si->free_clusters); | |
990 | offset = idx * SWAPFILE_CLUSTER; | |
991 | ci = lock_cluster(si, offset); | |
992 | alloc_cluster(si, idx); | |
e0709829 | 993 | cluster_set_count_flag(ci, SWAPFILE_CLUSTER, CLUSTER_FLAG_HUGE); |
38d8b4e6 HY |
994 | |
995 | map = si->swap_map + offset; | |
996 | for (i = 0; i < SWAPFILE_CLUSTER; i++) | |
997 | map[i] = SWAP_HAS_CACHE; | |
998 | unlock_cluster(ci); | |
999 | swap_range_alloc(si, offset, SWAPFILE_CLUSTER); | |
1000 | *slot = swp_entry(si->type, offset); | |
1001 | ||
1002 | return 1; | |
1003 | } | |
1004 | ||
1005 | static void swap_free_cluster(struct swap_info_struct *si, unsigned long idx) | |
1006 | { | |
1007 | unsigned long offset = idx * SWAPFILE_CLUSTER; | |
1008 | struct swap_cluster_info *ci; | |
1009 | ||
1010 | ci = lock_cluster(si, offset); | |
979aafa5 | 1011 | memset(si->swap_map + offset, 0, SWAPFILE_CLUSTER); |
38d8b4e6 HY |
1012 | cluster_set_count_flag(ci, 0, 0); |
1013 | free_cluster(si, idx); | |
1014 | unlock_cluster(ci); | |
1015 | swap_range_free(si, offset, SWAPFILE_CLUSTER); | |
1016 | } | |
38d8b4e6 | 1017 | |
36005bae TC |
1018 | static unsigned long scan_swap_map(struct swap_info_struct *si, |
1019 | unsigned char usage) | |
1020 | { | |
1021 | swp_entry_t entry; | |
1022 | int n_ret; | |
1023 | ||
1024 | n_ret = scan_swap_map_slots(si, usage, 1, &entry); | |
1025 | ||
1026 | if (n_ret) | |
1027 | return swp_offset(entry); | |
1028 | else | |
1029 | return 0; | |
1030 | ||
1031 | } | |
1032 | ||
5d5e8f19 | 1033 | int get_swap_pages(int n_goal, swp_entry_t swp_entries[], int entry_size) |
1da177e4 | 1034 | { |
5d5e8f19 | 1035 | unsigned long size = swap_entry_size(entry_size); |
adfab836 | 1036 | struct swap_info_struct *si, *next; |
36005bae TC |
1037 | long avail_pgs; |
1038 | int n_ret = 0; | |
a2468cc9 | 1039 | int node; |
1da177e4 | 1040 | |
38d8b4e6 | 1041 | /* Only single cluster request supported */ |
5d5e8f19 | 1042 | WARN_ON_ONCE(n_goal > 1 && size == SWAPFILE_CLUSTER); |
38d8b4e6 | 1043 | |
5d5e8f19 | 1044 | avail_pgs = atomic_long_read(&nr_swap_pages) / size; |
36005bae | 1045 | if (avail_pgs <= 0) |
fb4f88dc | 1046 | goto noswap; |
36005bae | 1047 | |
08d3090f | 1048 | n_goal = min3((long)n_goal, (long)SWAP_BATCH, avail_pgs); |
36005bae | 1049 | |
5d5e8f19 | 1050 | atomic_long_sub(n_goal * size, &nr_swap_pages); |
fb4f88dc | 1051 | |
18ab4d4c DS |
1052 | spin_lock(&swap_avail_lock); |
1053 | ||
1054 | start_over: | |
a2468cc9 AL |
1055 | node = numa_node_id(); |
1056 | plist_for_each_entry_safe(si, next, &swap_avail_heads[node], avail_lists[node]) { | |
18ab4d4c | 1057 | /* requeue si to after same-priority siblings */ |
a2468cc9 | 1058 | plist_requeue(&si->avail_lists[node], &swap_avail_heads[node]); |
18ab4d4c | 1059 | spin_unlock(&swap_avail_lock); |
ec8acf20 | 1060 | spin_lock(&si->lock); |
adfab836 | 1061 | if (!si->highest_bit || !(si->flags & SWP_WRITEOK)) { |
18ab4d4c | 1062 | spin_lock(&swap_avail_lock); |
a2468cc9 | 1063 | if (plist_node_empty(&si->avail_lists[node])) { |
18ab4d4c DS |
1064 | spin_unlock(&si->lock); |
1065 | goto nextsi; | |
1066 | } | |
1067 | WARN(!si->highest_bit, | |
1068 | "swap_info %d in list but !highest_bit\n", | |
1069 | si->type); | |
1070 | WARN(!(si->flags & SWP_WRITEOK), | |
1071 | "swap_info %d in list but !SWP_WRITEOK\n", | |
1072 | si->type); | |
a2468cc9 | 1073 | __del_from_avail_list(si); |
ec8acf20 | 1074 | spin_unlock(&si->lock); |
18ab4d4c | 1075 | goto nextsi; |
ec8acf20 | 1076 | } |
5d5e8f19 | 1077 | if (size == SWAPFILE_CLUSTER) { |
bc4ae27d | 1078 | if (!(si->flags & SWP_FS)) |
f0eea189 HY |
1079 | n_ret = swap_alloc_cluster(si, swp_entries); |
1080 | } else | |
38d8b4e6 HY |
1081 | n_ret = scan_swap_map_slots(si, SWAP_HAS_CACHE, |
1082 | n_goal, swp_entries); | |
ec8acf20 | 1083 | spin_unlock(&si->lock); |
5d5e8f19 | 1084 | if (n_ret || size == SWAPFILE_CLUSTER) |
36005bae | 1085 | goto check_out; |
18ab4d4c | 1086 | pr_debug("scan_swap_map of si %d failed to find offset\n", |
36005bae TC |
1087 | si->type); |
1088 | ||
18ab4d4c DS |
1089 | spin_lock(&swap_avail_lock); |
1090 | nextsi: | |
adfab836 DS |
1091 | /* |
1092 | * if we got here, it's likely that si was almost full before, | |
1093 | * and since scan_swap_map() can drop the si->lock, multiple | |
1094 | * callers probably all tried to get a page from the same si | |
18ab4d4c DS |
1095 | * and it filled up before we could get one; or, the si filled |
1096 | * up between us dropping swap_avail_lock and taking si->lock. | |
1097 | * Since we dropped the swap_avail_lock, the swap_avail_head | |
1098 | * list may have been modified; so if next is still in the | |
36005bae TC |
1099 | * swap_avail_head list then try it, otherwise start over |
1100 | * if we have not gotten any slots. | |
adfab836 | 1101 | */ |
a2468cc9 | 1102 | if (plist_node_empty(&next->avail_lists[node])) |
18ab4d4c | 1103 | goto start_over; |
1da177e4 | 1104 | } |
fb4f88dc | 1105 | |
18ab4d4c DS |
1106 | spin_unlock(&swap_avail_lock); |
1107 | ||
36005bae TC |
1108 | check_out: |
1109 | if (n_ret < n_goal) | |
5d5e8f19 | 1110 | atomic_long_add((long)(n_goal - n_ret) * size, |
38d8b4e6 | 1111 | &nr_swap_pages); |
fb4f88dc | 1112 | noswap: |
36005bae TC |
1113 | return n_ret; |
1114 | } | |
1115 | ||
2de1a7e4 | 1116 | /* The only caller of this function is now suspend routine */ |
910321ea HD |
1117 | swp_entry_t get_swap_page_of_type(int type) |
1118 | { | |
c10d38cc | 1119 | struct swap_info_struct *si = swap_type_to_swap_info(type); |
910321ea HD |
1120 | pgoff_t offset; |
1121 | ||
c10d38cc DJ |
1122 | if (!si) |
1123 | goto fail; | |
1124 | ||
ec8acf20 | 1125 | spin_lock(&si->lock); |
c10d38cc | 1126 | if (si->flags & SWP_WRITEOK) { |
ec8acf20 | 1127 | atomic_long_dec(&nr_swap_pages); |
910321ea HD |
1128 | /* This is called for allocating swap entry, not cache */ |
1129 | offset = scan_swap_map(si, 1); | |
1130 | if (offset) { | |
ec8acf20 | 1131 | spin_unlock(&si->lock); |
910321ea HD |
1132 | return swp_entry(type, offset); |
1133 | } | |
ec8acf20 | 1134 | atomic_long_inc(&nr_swap_pages); |
910321ea | 1135 | } |
ec8acf20 | 1136 | spin_unlock(&si->lock); |
c10d38cc | 1137 | fail: |
910321ea HD |
1138 | return (swp_entry_t) {0}; |
1139 | } | |
1140 | ||
e8c26ab6 | 1141 | static struct swap_info_struct *__swap_info_get(swp_entry_t entry) |
1da177e4 | 1142 | { |
73c34b6a | 1143 | struct swap_info_struct *p; |
eb085574 | 1144 | unsigned long offset; |
1da177e4 LT |
1145 | |
1146 | if (!entry.val) | |
1147 | goto out; | |
eb085574 | 1148 | p = swp_swap_info(entry); |
c10d38cc | 1149 | if (!p) |
1da177e4 | 1150 | goto bad_nofile; |
1da177e4 LT |
1151 | if (!(p->flags & SWP_USED)) |
1152 | goto bad_device; | |
1153 | offset = swp_offset(entry); | |
1154 | if (offset >= p->max) | |
1155 | goto bad_offset; | |
1da177e4 LT |
1156 | return p; |
1157 | ||
1da177e4 | 1158 | bad_offset: |
6a991fc7 | 1159 | pr_err("swap_info_get: %s%08lx\n", Bad_offset, entry.val); |
1da177e4 LT |
1160 | goto out; |
1161 | bad_device: | |
6a991fc7 | 1162 | pr_err("swap_info_get: %s%08lx\n", Unused_file, entry.val); |
1da177e4 LT |
1163 | goto out; |
1164 | bad_nofile: | |
6a991fc7 | 1165 | pr_err("swap_info_get: %s%08lx\n", Bad_file, entry.val); |
1da177e4 LT |
1166 | out: |
1167 | return NULL; | |
886bb7e9 | 1168 | } |
1da177e4 | 1169 | |
e8c26ab6 TC |
1170 | static struct swap_info_struct *_swap_info_get(swp_entry_t entry) |
1171 | { | |
1172 | struct swap_info_struct *p; | |
1173 | ||
1174 | p = __swap_info_get(entry); | |
1175 | if (!p) | |
1176 | goto out; | |
1177 | if (!p->swap_map[swp_offset(entry)]) | |
1178 | goto bad_free; | |
1179 | return p; | |
1180 | ||
1181 | bad_free: | |
1182 | pr_err("swap_info_get: %s%08lx\n", Unused_offset, entry.val); | |
1183 | goto out; | |
1184 | out: | |
1185 | return NULL; | |
1186 | } | |
1187 | ||
235b6217 HY |
1188 | static struct swap_info_struct *swap_info_get(swp_entry_t entry) |
1189 | { | |
1190 | struct swap_info_struct *p; | |
1191 | ||
1192 | p = _swap_info_get(entry); | |
1193 | if (p) | |
1194 | spin_lock(&p->lock); | |
1195 | return p; | |
1196 | } | |
1197 | ||
7c00bafe TC |
1198 | static struct swap_info_struct *swap_info_get_cont(swp_entry_t entry, |
1199 | struct swap_info_struct *q) | |
1200 | { | |
1201 | struct swap_info_struct *p; | |
1202 | ||
1203 | p = _swap_info_get(entry); | |
1204 | ||
1205 | if (p != q) { | |
1206 | if (q != NULL) | |
1207 | spin_unlock(&q->lock); | |
1208 | if (p != NULL) | |
1209 | spin_lock(&p->lock); | |
1210 | } | |
1211 | return p; | |
1212 | } | |
1213 | ||
b32d5f32 HY |
1214 | static unsigned char __swap_entry_free_locked(struct swap_info_struct *p, |
1215 | unsigned long offset, | |
1216 | unsigned char usage) | |
1da177e4 | 1217 | { |
8d69aaee HD |
1218 | unsigned char count; |
1219 | unsigned char has_cache; | |
235b6217 | 1220 | |
253d553b | 1221 | count = p->swap_map[offset]; |
235b6217 | 1222 | |
253d553b HD |
1223 | has_cache = count & SWAP_HAS_CACHE; |
1224 | count &= ~SWAP_HAS_CACHE; | |
355cfa73 | 1225 | |
253d553b | 1226 | if (usage == SWAP_HAS_CACHE) { |
355cfa73 | 1227 | VM_BUG_ON(!has_cache); |
253d553b | 1228 | has_cache = 0; |
aaa46865 HD |
1229 | } else if (count == SWAP_MAP_SHMEM) { |
1230 | /* | |
1231 | * Or we could insist on shmem.c using a special | |
1232 | * swap_shmem_free() and free_shmem_swap_and_cache()... | |
1233 | */ | |
1234 | count = 0; | |
570a335b HD |
1235 | } else if ((count & ~COUNT_CONTINUED) <= SWAP_MAP_MAX) { |
1236 | if (count == COUNT_CONTINUED) { | |
1237 | if (swap_count_continued(p, offset, count)) | |
1238 | count = SWAP_MAP_MAX | COUNT_CONTINUED; | |
1239 | else | |
1240 | count = SWAP_MAP_MAX; | |
1241 | } else | |
1242 | count--; | |
1243 | } | |
253d553b | 1244 | |
253d553b | 1245 | usage = count | has_cache; |
7c00bafe TC |
1246 | p->swap_map[offset] = usage ? : SWAP_HAS_CACHE; |
1247 | ||
b32d5f32 HY |
1248 | return usage; |
1249 | } | |
1250 | ||
eb085574 HY |
1251 | /* |
1252 | * Check whether swap entry is valid in the swap device. If so, | |
1253 | * return pointer to swap_info_struct, and keep the swap entry valid | |
1254 | * via preventing the swap device from being swapoff, until | |
1255 | * put_swap_device() is called. Otherwise return NULL. | |
1256 | * | |
1257 | * The entirety of the RCU read critical section must come before the | |
1258 | * return from or after the call to synchronize_rcu() in | |
1259 | * enable_swap_info() or swapoff(). So if "si->flags & SWP_VALID" is | |
1260 | * true, the si->map, si->cluster_info, etc. must be valid in the | |
1261 | * critical section. | |
1262 | * | |
1263 | * Notice that swapoff or swapoff+swapon can still happen before the | |
1264 | * rcu_read_lock() in get_swap_device() or after the rcu_read_unlock() | |
1265 | * in put_swap_device() if there isn't any other way to prevent | |
1266 | * swapoff, such as page lock, page table lock, etc. The caller must | |
1267 | * be prepared for that. For example, the following situation is | |
1268 | * possible. | |
1269 | * | |
1270 | * CPU1 CPU2 | |
1271 | * do_swap_page() | |
1272 | * ... swapoff+swapon | |
1273 | * __read_swap_cache_async() | |
1274 | * swapcache_prepare() | |
1275 | * __swap_duplicate() | |
1276 | * // check swap_map | |
1277 | * // verify PTE not changed | |
1278 | * | |
1279 | * In __swap_duplicate(), the swap_map need to be checked before | |
1280 | * changing partly because the specified swap entry may be for another | |
1281 | * swap device which has been swapoff. And in do_swap_page(), after | |
1282 | * the page is read from the swap device, the PTE is verified not | |
1283 | * changed with the page table locked to check whether the swap device | |
1284 | * has been swapoff or swapoff+swapon. | |
1285 | */ | |
1286 | struct swap_info_struct *get_swap_device(swp_entry_t entry) | |
1287 | { | |
1288 | struct swap_info_struct *si; | |
1289 | unsigned long offset; | |
1290 | ||
1291 | if (!entry.val) | |
1292 | goto out; | |
1293 | si = swp_swap_info(entry); | |
1294 | if (!si) | |
1295 | goto bad_nofile; | |
1296 | ||
1297 | rcu_read_lock(); | |
1298 | if (!(si->flags & SWP_VALID)) | |
1299 | goto unlock_out; | |
1300 | offset = swp_offset(entry); | |
1301 | if (offset >= si->max) | |
1302 | goto unlock_out; | |
1303 | ||
1304 | return si; | |
1305 | bad_nofile: | |
1306 | pr_err("%s: %s%08lx\n", __func__, Bad_file, entry.val); | |
1307 | out: | |
1308 | return NULL; | |
1309 | unlock_out: | |
1310 | rcu_read_unlock(); | |
1311 | return NULL; | |
1312 | } | |
1313 | ||
b32d5f32 | 1314 | static unsigned char __swap_entry_free(struct swap_info_struct *p, |
33e16272 | 1315 | swp_entry_t entry) |
b32d5f32 HY |
1316 | { |
1317 | struct swap_cluster_info *ci; | |
1318 | unsigned long offset = swp_offset(entry); | |
33e16272 | 1319 | unsigned char usage; |
b32d5f32 HY |
1320 | |
1321 | ci = lock_cluster_or_swap_info(p, offset); | |
33e16272 | 1322 | usage = __swap_entry_free_locked(p, offset, 1); |
7c00bafe | 1323 | unlock_cluster_or_swap_info(p, ci); |
10e364da HY |
1324 | if (!usage) |
1325 | free_swap_slot(entry); | |
7c00bafe TC |
1326 | |
1327 | return usage; | |
1328 | } | |
355cfa73 | 1329 | |
7c00bafe TC |
1330 | static void swap_entry_free(struct swap_info_struct *p, swp_entry_t entry) |
1331 | { | |
1332 | struct swap_cluster_info *ci; | |
1333 | unsigned long offset = swp_offset(entry); | |
1334 | unsigned char count; | |
1335 | ||
1336 | ci = lock_cluster(p, offset); | |
1337 | count = p->swap_map[offset]; | |
1338 | VM_BUG_ON(count != SWAP_HAS_CACHE); | |
1339 | p->swap_map[offset] = 0; | |
1340 | dec_cluster_info_page(p, p->cluster_info, offset); | |
235b6217 HY |
1341 | unlock_cluster(ci); |
1342 | ||
38d8b4e6 HY |
1343 | mem_cgroup_uncharge_swap(entry, 1); |
1344 | swap_range_free(p, offset, 1); | |
1da177e4 LT |
1345 | } |
1346 | ||
1347 | /* | |
2de1a7e4 | 1348 | * Caller has made sure that the swap device corresponding to entry |
1da177e4 LT |
1349 | * is still around or has not been recycled. |
1350 | */ | |
1351 | void swap_free(swp_entry_t entry) | |
1352 | { | |
73c34b6a | 1353 | struct swap_info_struct *p; |
1da177e4 | 1354 | |
235b6217 | 1355 | p = _swap_info_get(entry); |
10e364da | 1356 | if (p) |
33e16272 | 1357 | __swap_entry_free(p, entry); |
1da177e4 LT |
1358 | } |
1359 | ||
cb4b86ba KH |
1360 | /* |
1361 | * Called after dropping swapcache to decrease refcnt to swap entries. | |
1362 | */ | |
a448f2d0 | 1363 | void put_swap_page(struct page *page, swp_entry_t entry) |
38d8b4e6 HY |
1364 | { |
1365 | unsigned long offset = swp_offset(entry); | |
1366 | unsigned long idx = offset / SWAPFILE_CLUSTER; | |
1367 | struct swap_cluster_info *ci; | |
1368 | struct swap_info_struct *si; | |
1369 | unsigned char *map; | |
a3aea839 HY |
1370 | unsigned int i, free_entries = 0; |
1371 | unsigned char val; | |
a448f2d0 | 1372 | int size = swap_entry_size(hpage_nr_pages(page)); |
fe5266d5 | 1373 | |
a3aea839 | 1374 | si = _swap_info_get(entry); |
38d8b4e6 HY |
1375 | if (!si) |
1376 | return; | |
1377 | ||
c2343d27 | 1378 | ci = lock_cluster_or_swap_info(si, offset); |
a448f2d0 | 1379 | if (size == SWAPFILE_CLUSTER) { |
a448f2d0 HY |
1380 | VM_BUG_ON(!cluster_is_huge(ci)); |
1381 | map = si->swap_map + offset; | |
1382 | for (i = 0; i < SWAPFILE_CLUSTER; i++) { | |
1383 | val = map[i]; | |
1384 | VM_BUG_ON(!(val & SWAP_HAS_CACHE)); | |
1385 | if (val == SWAP_HAS_CACHE) | |
1386 | free_entries++; | |
1387 | } | |
a448f2d0 | 1388 | cluster_clear_huge(ci); |
a448f2d0 | 1389 | if (free_entries == SWAPFILE_CLUSTER) { |
c2343d27 | 1390 | unlock_cluster_or_swap_info(si, ci); |
a448f2d0 | 1391 | spin_lock(&si->lock); |
a448f2d0 HY |
1392 | mem_cgroup_uncharge_swap(entry, SWAPFILE_CLUSTER); |
1393 | swap_free_cluster(si, idx); | |
1394 | spin_unlock(&si->lock); | |
1395 | return; | |
1396 | } | |
1397 | } | |
c2343d27 HY |
1398 | for (i = 0; i < size; i++, entry.val++) { |
1399 | if (!__swap_entry_free_locked(si, offset + i, SWAP_HAS_CACHE)) { | |
1400 | unlock_cluster_or_swap_info(si, ci); | |
1401 | free_swap_slot(entry); | |
1402 | if (i == size - 1) | |
1403 | return; | |
1404 | lock_cluster_or_swap_info(si, offset); | |
a3aea839 HY |
1405 | } |
1406 | } | |
c2343d27 | 1407 | unlock_cluster_or_swap_info(si, ci); |
38d8b4e6 | 1408 | } |
59807685 | 1409 | |
fe5266d5 | 1410 | #ifdef CONFIG_THP_SWAP |
59807685 HY |
1411 | int split_swap_cluster(swp_entry_t entry) |
1412 | { | |
1413 | struct swap_info_struct *si; | |
1414 | struct swap_cluster_info *ci; | |
1415 | unsigned long offset = swp_offset(entry); | |
1416 | ||
1417 | si = _swap_info_get(entry); | |
1418 | if (!si) | |
1419 | return -EBUSY; | |
1420 | ci = lock_cluster(si, offset); | |
1421 | cluster_clear_huge(ci); | |
1422 | unlock_cluster(ci); | |
1423 | return 0; | |
1424 | } | |
fe5266d5 | 1425 | #endif |
38d8b4e6 | 1426 | |
155b5f88 HY |
1427 | static int swp_entry_cmp(const void *ent1, const void *ent2) |
1428 | { | |
1429 | const swp_entry_t *e1 = ent1, *e2 = ent2; | |
1430 | ||
1431 | return (int)swp_type(*e1) - (int)swp_type(*e2); | |
1432 | } | |
1433 | ||
7c00bafe TC |
1434 | void swapcache_free_entries(swp_entry_t *entries, int n) |
1435 | { | |
1436 | struct swap_info_struct *p, *prev; | |
1437 | int i; | |
1438 | ||
1439 | if (n <= 0) | |
1440 | return; | |
1441 | ||
1442 | prev = NULL; | |
1443 | p = NULL; | |
155b5f88 HY |
1444 | |
1445 | /* | |
1446 | * Sort swap entries by swap device, so each lock is only taken once. | |
1447 | * nr_swapfiles isn't absolutely correct, but the overhead of sort() is | |
1448 | * so low that it isn't necessary to optimize further. | |
1449 | */ | |
1450 | if (nr_swapfiles > 1) | |
1451 | sort(entries, n, sizeof(entries[0]), swp_entry_cmp, NULL); | |
7c00bafe TC |
1452 | for (i = 0; i < n; ++i) { |
1453 | p = swap_info_get_cont(entries[i], prev); | |
1454 | if (p) | |
1455 | swap_entry_free(p, entries[i]); | |
7c00bafe TC |
1456 | prev = p; |
1457 | } | |
235b6217 | 1458 | if (p) |
7c00bafe | 1459 | spin_unlock(&p->lock); |
cb4b86ba KH |
1460 | } |
1461 | ||
1da177e4 | 1462 | /* |
c475a8ab | 1463 | * How many references to page are currently swapped out? |
570a335b HD |
1464 | * This does not give an exact answer when swap count is continued, |
1465 | * but does include the high COUNT_CONTINUED flag to allow for that. | |
1da177e4 | 1466 | */ |
bde05d1c | 1467 | int page_swapcount(struct page *page) |
1da177e4 | 1468 | { |
c475a8ab HD |
1469 | int count = 0; |
1470 | struct swap_info_struct *p; | |
235b6217 | 1471 | struct swap_cluster_info *ci; |
1da177e4 | 1472 | swp_entry_t entry; |
235b6217 | 1473 | unsigned long offset; |
1da177e4 | 1474 | |
4c21e2f2 | 1475 | entry.val = page_private(page); |
235b6217 | 1476 | p = _swap_info_get(entry); |
1da177e4 | 1477 | if (p) { |
235b6217 HY |
1478 | offset = swp_offset(entry); |
1479 | ci = lock_cluster_or_swap_info(p, offset); | |
1480 | count = swap_count(p->swap_map[offset]); | |
1481 | unlock_cluster_or_swap_info(p, ci); | |
1da177e4 | 1482 | } |
c475a8ab | 1483 | return count; |
1da177e4 LT |
1484 | } |
1485 | ||
eb085574 | 1486 | int __swap_count(swp_entry_t entry) |
aa8d22a1 | 1487 | { |
eb085574 | 1488 | struct swap_info_struct *si; |
aa8d22a1 | 1489 | pgoff_t offset = swp_offset(entry); |
eb085574 | 1490 | int count = 0; |
aa8d22a1 | 1491 | |
eb085574 HY |
1492 | si = get_swap_device(entry); |
1493 | if (si) { | |
1494 | count = swap_count(si->swap_map[offset]); | |
1495 | put_swap_device(si); | |
1496 | } | |
1497 | return count; | |
aa8d22a1 MK |
1498 | } |
1499 | ||
322b8afe HY |
1500 | static int swap_swapcount(struct swap_info_struct *si, swp_entry_t entry) |
1501 | { | |
1502 | int count = 0; | |
1503 | pgoff_t offset = swp_offset(entry); | |
1504 | struct swap_cluster_info *ci; | |
1505 | ||
1506 | ci = lock_cluster_or_swap_info(si, offset); | |
1507 | count = swap_count(si->swap_map[offset]); | |
1508 | unlock_cluster_or_swap_info(si, ci); | |
1509 | return count; | |
1510 | } | |
1511 | ||
e8c26ab6 TC |
1512 | /* |
1513 | * How many references to @entry are currently swapped out? | |
1514 | * This does not give an exact answer when swap count is continued, | |
1515 | * but does include the high COUNT_CONTINUED flag to allow for that. | |
1516 | */ | |
1517 | int __swp_swapcount(swp_entry_t entry) | |
1518 | { | |
1519 | int count = 0; | |
e8c26ab6 | 1520 | struct swap_info_struct *si; |
e8c26ab6 | 1521 | |
eb085574 HY |
1522 | si = get_swap_device(entry); |
1523 | if (si) { | |
322b8afe | 1524 | count = swap_swapcount(si, entry); |
eb085574 HY |
1525 | put_swap_device(si); |
1526 | } | |
e8c26ab6 TC |
1527 | return count; |
1528 | } | |
1529 | ||
8334b962 MK |
1530 | /* |
1531 | * How many references to @entry are currently swapped out? | |
1532 | * This considers COUNT_CONTINUED so it returns exact answer. | |
1533 | */ | |
1534 | int swp_swapcount(swp_entry_t entry) | |
1535 | { | |
1536 | int count, tmp_count, n; | |
1537 | struct swap_info_struct *p; | |
235b6217 | 1538 | struct swap_cluster_info *ci; |
8334b962 MK |
1539 | struct page *page; |
1540 | pgoff_t offset; | |
1541 | unsigned char *map; | |
1542 | ||
235b6217 | 1543 | p = _swap_info_get(entry); |
8334b962 MK |
1544 | if (!p) |
1545 | return 0; | |
1546 | ||
235b6217 HY |
1547 | offset = swp_offset(entry); |
1548 | ||
1549 | ci = lock_cluster_or_swap_info(p, offset); | |
1550 | ||
1551 | count = swap_count(p->swap_map[offset]); | |
8334b962 MK |
1552 | if (!(count & COUNT_CONTINUED)) |
1553 | goto out; | |
1554 | ||
1555 | count &= ~COUNT_CONTINUED; | |
1556 | n = SWAP_MAP_MAX + 1; | |
1557 | ||
8334b962 MK |
1558 | page = vmalloc_to_page(p->swap_map + offset); |
1559 | offset &= ~PAGE_MASK; | |
1560 | VM_BUG_ON(page_private(page) != SWP_CONTINUED); | |
1561 | ||
1562 | do { | |
a8ae4991 | 1563 | page = list_next_entry(page, lru); |
8334b962 MK |
1564 | map = kmap_atomic(page); |
1565 | tmp_count = map[offset]; | |
1566 | kunmap_atomic(map); | |
1567 | ||
1568 | count += (tmp_count & ~COUNT_CONTINUED) * n; | |
1569 | n *= (SWAP_CONT_MAX + 1); | |
1570 | } while (tmp_count & COUNT_CONTINUED); | |
1571 | out: | |
235b6217 | 1572 | unlock_cluster_or_swap_info(p, ci); |
8334b962 MK |
1573 | return count; |
1574 | } | |
1575 | ||
e0709829 HY |
1576 | static bool swap_page_trans_huge_swapped(struct swap_info_struct *si, |
1577 | swp_entry_t entry) | |
1578 | { | |
1579 | struct swap_cluster_info *ci; | |
1580 | unsigned char *map = si->swap_map; | |
1581 | unsigned long roffset = swp_offset(entry); | |
1582 | unsigned long offset = round_down(roffset, SWAPFILE_CLUSTER); | |
1583 | int i; | |
1584 | bool ret = false; | |
1585 | ||
1586 | ci = lock_cluster_or_swap_info(si, offset); | |
1587 | if (!ci || !cluster_is_huge(ci)) { | |
afa4711e | 1588 | if (swap_count(map[roffset])) |
e0709829 HY |
1589 | ret = true; |
1590 | goto unlock_out; | |
1591 | } | |
1592 | for (i = 0; i < SWAPFILE_CLUSTER; i++) { | |
afa4711e | 1593 | if (swap_count(map[offset + i])) { |
e0709829 HY |
1594 | ret = true; |
1595 | break; | |
1596 | } | |
1597 | } | |
1598 | unlock_out: | |
1599 | unlock_cluster_or_swap_info(si, ci); | |
1600 | return ret; | |
1601 | } | |
1602 | ||
1603 | static bool page_swapped(struct page *page) | |
1604 | { | |
1605 | swp_entry_t entry; | |
1606 | struct swap_info_struct *si; | |
1607 | ||
fe5266d5 | 1608 | if (!IS_ENABLED(CONFIG_THP_SWAP) || likely(!PageTransCompound(page))) |
e0709829 HY |
1609 | return page_swapcount(page) != 0; |
1610 | ||
1611 | page = compound_head(page); | |
1612 | entry.val = page_private(page); | |
1613 | si = _swap_info_get(entry); | |
1614 | if (si) | |
1615 | return swap_page_trans_huge_swapped(si, entry); | |
1616 | return false; | |
1617 | } | |
ba3c4ce6 HY |
1618 | |
1619 | static int page_trans_huge_map_swapcount(struct page *page, int *total_mapcount, | |
1620 | int *total_swapcount) | |
1621 | { | |
1622 | int i, map_swapcount, _total_mapcount, _total_swapcount; | |
1623 | unsigned long offset = 0; | |
1624 | struct swap_info_struct *si; | |
1625 | struct swap_cluster_info *ci = NULL; | |
1626 | unsigned char *map = NULL; | |
1627 | int mapcount, swapcount = 0; | |
1628 | ||
1629 | /* hugetlbfs shouldn't call it */ | |
1630 | VM_BUG_ON_PAGE(PageHuge(page), page); | |
1631 | ||
fe5266d5 HY |
1632 | if (!IS_ENABLED(CONFIG_THP_SWAP) || likely(!PageTransCompound(page))) { |
1633 | mapcount = page_trans_huge_mapcount(page, total_mapcount); | |
ba3c4ce6 HY |
1634 | if (PageSwapCache(page)) |
1635 | swapcount = page_swapcount(page); | |
1636 | if (total_swapcount) | |
1637 | *total_swapcount = swapcount; | |
1638 | return mapcount + swapcount; | |
1639 | } | |
1640 | ||
1641 | page = compound_head(page); | |
1642 | ||
1643 | _total_mapcount = _total_swapcount = map_swapcount = 0; | |
1644 | if (PageSwapCache(page)) { | |
1645 | swp_entry_t entry; | |
1646 | ||
1647 | entry.val = page_private(page); | |
1648 | si = _swap_info_get(entry); | |
1649 | if (si) { | |
1650 | map = si->swap_map; | |
1651 | offset = swp_offset(entry); | |
1652 | } | |
1653 | } | |
1654 | if (map) | |
1655 | ci = lock_cluster(si, offset); | |
1656 | for (i = 0; i < HPAGE_PMD_NR; i++) { | |
1657 | mapcount = atomic_read(&page[i]._mapcount) + 1; | |
1658 | _total_mapcount += mapcount; | |
1659 | if (map) { | |
1660 | swapcount = swap_count(map[offset + i]); | |
1661 | _total_swapcount += swapcount; | |
1662 | } | |
1663 | map_swapcount = max(map_swapcount, mapcount + swapcount); | |
1664 | } | |
1665 | unlock_cluster(ci); | |
1666 | if (PageDoubleMap(page)) { | |
1667 | map_swapcount -= 1; | |
1668 | _total_mapcount -= HPAGE_PMD_NR; | |
1669 | } | |
1670 | mapcount = compound_mapcount(page); | |
1671 | map_swapcount += mapcount; | |
1672 | _total_mapcount += mapcount; | |
1673 | if (total_mapcount) | |
1674 | *total_mapcount = _total_mapcount; | |
1675 | if (total_swapcount) | |
1676 | *total_swapcount = _total_swapcount; | |
1677 | ||
1678 | return map_swapcount; | |
1679 | } | |
e0709829 | 1680 | |
1da177e4 | 1681 | /* |
7b1fe597 HD |
1682 | * We can write to an anon page without COW if there are no other references |
1683 | * to it. And as a side-effect, free up its swap: because the old content | |
1684 | * on disk will never be read, and seeking back there to write new content | |
1685 | * later would only waste time away from clustering. | |
6d0a07ed | 1686 | * |
ba3c4ce6 | 1687 | * NOTE: total_map_swapcount should not be relied upon by the caller if |
6d0a07ed AA |
1688 | * reuse_swap_page() returns false, but it may be always overwritten |
1689 | * (see the other implementation for CONFIG_SWAP=n). | |
1da177e4 | 1690 | */ |
ba3c4ce6 | 1691 | bool reuse_swap_page(struct page *page, int *total_map_swapcount) |
1da177e4 | 1692 | { |
ba3c4ce6 | 1693 | int count, total_mapcount, total_swapcount; |
c475a8ab | 1694 | |
309381fe | 1695 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
5ad64688 | 1696 | if (unlikely(PageKsm(page))) |
6d0a07ed | 1697 | return false; |
ba3c4ce6 HY |
1698 | count = page_trans_huge_map_swapcount(page, &total_mapcount, |
1699 | &total_swapcount); | |
1700 | if (total_map_swapcount) | |
1701 | *total_map_swapcount = total_mapcount + total_swapcount; | |
1702 | if (count == 1 && PageSwapCache(page) && | |
1703 | (likely(!PageTransCompound(page)) || | |
1704 | /* The remaining swap count will be freed soon */ | |
1705 | total_swapcount == page_swapcount(page))) { | |
f0571429 | 1706 | if (!PageWriteback(page)) { |
ba3c4ce6 | 1707 | page = compound_head(page); |
7b1fe597 HD |
1708 | delete_from_swap_cache(page); |
1709 | SetPageDirty(page); | |
f0571429 MK |
1710 | } else { |
1711 | swp_entry_t entry; | |
1712 | struct swap_info_struct *p; | |
1713 | ||
1714 | entry.val = page_private(page); | |
1715 | p = swap_info_get(entry); | |
1716 | if (p->flags & SWP_STABLE_WRITES) { | |
1717 | spin_unlock(&p->lock); | |
1718 | return false; | |
1719 | } | |
1720 | spin_unlock(&p->lock); | |
7b1fe597 HD |
1721 | } |
1722 | } | |
ba3c4ce6 | 1723 | |
5ad64688 | 1724 | return count <= 1; |
1da177e4 LT |
1725 | } |
1726 | ||
1727 | /* | |
a2c43eed HD |
1728 | * If swap is getting full, or if there are no more mappings of this page, |
1729 | * then try_to_free_swap is called to free its swap space. | |
1da177e4 | 1730 | */ |
a2c43eed | 1731 | int try_to_free_swap(struct page *page) |
1da177e4 | 1732 | { |
309381fe | 1733 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
1da177e4 LT |
1734 | |
1735 | if (!PageSwapCache(page)) | |
1736 | return 0; | |
1737 | if (PageWriteback(page)) | |
1738 | return 0; | |
e0709829 | 1739 | if (page_swapped(page)) |
1da177e4 LT |
1740 | return 0; |
1741 | ||
b73d7fce HD |
1742 | /* |
1743 | * Once hibernation has begun to create its image of memory, | |
1744 | * there's a danger that one of the calls to try_to_free_swap() | |
1745 | * - most probably a call from __try_to_reclaim_swap() while | |
1746 | * hibernation is allocating its own swap pages for the image, | |
1747 | * but conceivably even a call from memory reclaim - will free | |
1748 | * the swap from a page which has already been recorded in the | |
1749 | * image as a clean swapcache page, and then reuse its swap for | |
1750 | * another page of the image. On waking from hibernation, the | |
1751 | * original page might be freed under memory pressure, then | |
1752 | * later read back in from swap, now with the wrong data. | |
1753 | * | |
2de1a7e4 | 1754 | * Hibernation suspends storage while it is writing the image |
f90ac398 | 1755 | * to disk so check that here. |
b73d7fce | 1756 | */ |
f90ac398 | 1757 | if (pm_suspended_storage()) |
b73d7fce HD |
1758 | return 0; |
1759 | ||
e0709829 | 1760 | page = compound_head(page); |
a2c43eed HD |
1761 | delete_from_swap_cache(page); |
1762 | SetPageDirty(page); | |
1763 | return 1; | |
68a22394 RR |
1764 | } |
1765 | ||
1da177e4 LT |
1766 | /* |
1767 | * Free the swap entry like above, but also try to | |
1768 | * free the page cache entry if it is the last user. | |
1769 | */ | |
2509ef26 | 1770 | int free_swap_and_cache(swp_entry_t entry) |
1da177e4 | 1771 | { |
2509ef26 | 1772 | struct swap_info_struct *p; |
7c00bafe | 1773 | unsigned char count; |
1da177e4 | 1774 | |
a7420aa5 | 1775 | if (non_swap_entry(entry)) |
2509ef26 | 1776 | return 1; |
0697212a | 1777 | |
7c00bafe | 1778 | p = _swap_info_get(entry); |
1da177e4 | 1779 | if (p) { |
33e16272 | 1780 | count = __swap_entry_free(p, entry); |
e0709829 | 1781 | if (count == SWAP_HAS_CACHE && |
bcd49e86 HY |
1782 | !swap_page_trans_huge_swapped(p, entry)) |
1783 | __try_to_reclaim_swap(p, swp_offset(entry), | |
1784 | TTRS_UNMAPPED | TTRS_FULL); | |
1da177e4 | 1785 | } |
2509ef26 | 1786 | return p != NULL; |
1da177e4 LT |
1787 | } |
1788 | ||
b0cb1a19 | 1789 | #ifdef CONFIG_HIBERNATION |
f577eb30 | 1790 | /* |
915bae9e | 1791 | * Find the swap type that corresponds to given device (if any). |
f577eb30 | 1792 | * |
915bae9e RW |
1793 | * @offset - number of the PAGE_SIZE-sized block of the device, starting |
1794 | * from 0, in which the swap header is expected to be located. | |
1795 | * | |
1796 | * This is needed for the suspend to disk (aka swsusp). | |
f577eb30 | 1797 | */ |
7bf23687 | 1798 | int swap_type_of(dev_t device, sector_t offset, struct block_device **bdev_p) |
f577eb30 | 1799 | { |
915bae9e | 1800 | struct block_device *bdev = NULL; |
efa90a98 | 1801 | int type; |
f577eb30 | 1802 | |
915bae9e RW |
1803 | if (device) |
1804 | bdev = bdget(device); | |
1805 | ||
f577eb30 | 1806 | spin_lock(&swap_lock); |
efa90a98 HD |
1807 | for (type = 0; type < nr_swapfiles; type++) { |
1808 | struct swap_info_struct *sis = swap_info[type]; | |
f577eb30 | 1809 | |
915bae9e | 1810 | if (!(sis->flags & SWP_WRITEOK)) |
f577eb30 | 1811 | continue; |
b6b5bce3 | 1812 | |
915bae9e | 1813 | if (!bdev) { |
7bf23687 | 1814 | if (bdev_p) |
dddac6a7 | 1815 | *bdev_p = bdgrab(sis->bdev); |
7bf23687 | 1816 | |
6e1819d6 | 1817 | spin_unlock(&swap_lock); |
efa90a98 | 1818 | return type; |
6e1819d6 | 1819 | } |
915bae9e | 1820 | if (bdev == sis->bdev) { |
4efaceb1 | 1821 | struct swap_extent *se = first_se(sis); |
915bae9e | 1822 | |
915bae9e | 1823 | if (se->start_block == offset) { |
7bf23687 | 1824 | if (bdev_p) |
dddac6a7 | 1825 | *bdev_p = bdgrab(sis->bdev); |
7bf23687 | 1826 | |
915bae9e RW |
1827 | spin_unlock(&swap_lock); |
1828 | bdput(bdev); | |
efa90a98 | 1829 | return type; |
915bae9e | 1830 | } |
f577eb30 RW |
1831 | } |
1832 | } | |
1833 | spin_unlock(&swap_lock); | |
915bae9e RW |
1834 | if (bdev) |
1835 | bdput(bdev); | |
1836 | ||
f577eb30 RW |
1837 | return -ENODEV; |
1838 | } | |
1839 | ||
73c34b6a HD |
1840 | /* |
1841 | * Get the (PAGE_SIZE) block corresponding to given offset on the swapdev | |
1842 | * corresponding to given index in swap_info (swap type). | |
1843 | */ | |
1844 | sector_t swapdev_block(int type, pgoff_t offset) | |
1845 | { | |
1846 | struct block_device *bdev; | |
c10d38cc | 1847 | struct swap_info_struct *si = swap_type_to_swap_info(type); |
73c34b6a | 1848 | |
c10d38cc | 1849 | if (!si || !(si->flags & SWP_WRITEOK)) |
73c34b6a | 1850 | return 0; |
d4906e1a | 1851 | return map_swap_entry(swp_entry(type, offset), &bdev); |
73c34b6a HD |
1852 | } |
1853 | ||
f577eb30 RW |
1854 | /* |
1855 | * Return either the total number of swap pages of given type, or the number | |
1856 | * of free pages of that type (depending on @free) | |
1857 | * | |
1858 | * This is needed for software suspend | |
1859 | */ | |
1860 | unsigned int count_swap_pages(int type, int free) | |
1861 | { | |
1862 | unsigned int n = 0; | |
1863 | ||
efa90a98 HD |
1864 | spin_lock(&swap_lock); |
1865 | if ((unsigned int)type < nr_swapfiles) { | |
1866 | struct swap_info_struct *sis = swap_info[type]; | |
1867 | ||
ec8acf20 | 1868 | spin_lock(&sis->lock); |
efa90a98 HD |
1869 | if (sis->flags & SWP_WRITEOK) { |
1870 | n = sis->pages; | |
f577eb30 | 1871 | if (free) |
efa90a98 | 1872 | n -= sis->inuse_pages; |
f577eb30 | 1873 | } |
ec8acf20 | 1874 | spin_unlock(&sis->lock); |
f577eb30 | 1875 | } |
efa90a98 | 1876 | spin_unlock(&swap_lock); |
f577eb30 RW |
1877 | return n; |
1878 | } | |
73c34b6a | 1879 | #endif /* CONFIG_HIBERNATION */ |
f577eb30 | 1880 | |
9f8bdb3f | 1881 | static inline int pte_same_as_swp(pte_t pte, pte_t swp_pte) |
179ef71c | 1882 | { |
9f8bdb3f | 1883 | return pte_same(pte_swp_clear_soft_dirty(pte), swp_pte); |
179ef71c CG |
1884 | } |
1885 | ||
1da177e4 | 1886 | /* |
72866f6f HD |
1887 | * No need to decide whether this PTE shares the swap entry with others, |
1888 | * just let do_wp_page work it out if a write is requested later - to | |
1889 | * force COW, vm_page_prot omits write permission from any private vma. | |
1da177e4 | 1890 | */ |
044d66c1 | 1891 | static int unuse_pte(struct vm_area_struct *vma, pmd_t *pmd, |
1da177e4 LT |
1892 | unsigned long addr, swp_entry_t entry, struct page *page) |
1893 | { | |
9e16b7fb | 1894 | struct page *swapcache; |
72835c86 | 1895 | struct mem_cgroup *memcg; |
044d66c1 HD |
1896 | spinlock_t *ptl; |
1897 | pte_t *pte; | |
1898 | int ret = 1; | |
1899 | ||
9e16b7fb HD |
1900 | swapcache = page; |
1901 | page = ksm_might_need_to_copy(page, vma, addr); | |
1902 | if (unlikely(!page)) | |
1903 | return -ENOMEM; | |
1904 | ||
f627c2f5 KS |
1905 | if (mem_cgroup_try_charge(page, vma->vm_mm, GFP_KERNEL, |
1906 | &memcg, false)) { | |
044d66c1 | 1907 | ret = -ENOMEM; |
85d9fc89 KH |
1908 | goto out_nolock; |
1909 | } | |
044d66c1 HD |
1910 | |
1911 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); | |
9f8bdb3f | 1912 | if (unlikely(!pte_same_as_swp(*pte, swp_entry_to_pte(entry)))) { |
f627c2f5 | 1913 | mem_cgroup_cancel_charge(page, memcg, false); |
044d66c1 HD |
1914 | ret = 0; |
1915 | goto out; | |
1916 | } | |
8a9f3ccd | 1917 | |
b084d435 | 1918 | dec_mm_counter(vma->vm_mm, MM_SWAPENTS); |
d559db08 | 1919 | inc_mm_counter(vma->vm_mm, MM_ANONPAGES); |
1da177e4 LT |
1920 | get_page(page); |
1921 | set_pte_at(vma->vm_mm, addr, pte, | |
1922 | pte_mkold(mk_pte(page, vma->vm_page_prot))); | |
00501b53 | 1923 | if (page == swapcache) { |
d281ee61 | 1924 | page_add_anon_rmap(page, vma, addr, false); |
f627c2f5 | 1925 | mem_cgroup_commit_charge(page, memcg, true, false); |
00501b53 | 1926 | } else { /* ksm created a completely new copy */ |
d281ee61 | 1927 | page_add_new_anon_rmap(page, vma, addr, false); |
f627c2f5 | 1928 | mem_cgroup_commit_charge(page, memcg, false, false); |
00501b53 JW |
1929 | lru_cache_add_active_or_unevictable(page, vma); |
1930 | } | |
1da177e4 LT |
1931 | swap_free(entry); |
1932 | /* | |
1933 | * Move the page to the active list so it is not | |
1934 | * immediately swapped out again after swapon. | |
1935 | */ | |
1936 | activate_page(page); | |
044d66c1 HD |
1937 | out: |
1938 | pte_unmap_unlock(pte, ptl); | |
85d9fc89 | 1939 | out_nolock: |
9e16b7fb HD |
1940 | if (page != swapcache) { |
1941 | unlock_page(page); | |
1942 | put_page(page); | |
1943 | } | |
044d66c1 | 1944 | return ret; |
1da177e4 LT |
1945 | } |
1946 | ||
1947 | static int unuse_pte_range(struct vm_area_struct *vma, pmd_t *pmd, | |
b56a2d8a VRP |
1948 | unsigned long addr, unsigned long end, |
1949 | unsigned int type, bool frontswap, | |
1950 | unsigned long *fs_pages_to_unuse) | |
1da177e4 | 1951 | { |
b56a2d8a VRP |
1952 | struct page *page; |
1953 | swp_entry_t entry; | |
705e87c0 | 1954 | pte_t *pte; |
b56a2d8a VRP |
1955 | struct swap_info_struct *si; |
1956 | unsigned long offset; | |
8a9f3ccd | 1957 | int ret = 0; |
b56a2d8a | 1958 | volatile unsigned char *swap_map; |
1da177e4 | 1959 | |
b56a2d8a | 1960 | si = swap_info[type]; |
044d66c1 | 1961 | pte = pte_offset_map(pmd, addr); |
1da177e4 | 1962 | do { |
b56a2d8a VRP |
1963 | struct vm_fault vmf; |
1964 | ||
1965 | if (!is_swap_pte(*pte)) | |
1966 | continue; | |
1967 | ||
1968 | entry = pte_to_swp_entry(*pte); | |
1969 | if (swp_type(entry) != type) | |
1970 | continue; | |
1971 | ||
1972 | offset = swp_offset(entry); | |
1973 | if (frontswap && !frontswap_test(si, offset)) | |
1974 | continue; | |
1975 | ||
1976 | pte_unmap(pte); | |
1977 | swap_map = &si->swap_map[offset]; | |
ebc5951e AR |
1978 | page = lookup_swap_cache(entry, vma, addr); |
1979 | if (!page) { | |
1980 | vmf.vma = vma; | |
1981 | vmf.address = addr; | |
1982 | vmf.pmd = pmd; | |
1983 | page = swapin_readahead(entry, GFP_HIGHUSER_MOVABLE, | |
1984 | &vmf); | |
1985 | } | |
b56a2d8a VRP |
1986 | if (!page) { |
1987 | if (*swap_map == 0 || *swap_map == SWAP_MAP_BAD) | |
1988 | goto try_next; | |
1989 | return -ENOMEM; | |
1990 | } | |
1991 | ||
1992 | lock_page(page); | |
1993 | wait_on_page_writeback(page); | |
1994 | ret = unuse_pte(vma, pmd, addr, entry, page); | |
1995 | if (ret < 0) { | |
1996 | unlock_page(page); | |
1997 | put_page(page); | |
1998 | goto out; | |
1999 | } | |
2000 | ||
2001 | try_to_free_swap(page); | |
2002 | unlock_page(page); | |
2003 | put_page(page); | |
2004 | ||
2005 | if (*fs_pages_to_unuse && !--(*fs_pages_to_unuse)) { | |
2006 | ret = FRONTSWAP_PAGES_UNUSED; | |
2007 | goto out; | |
1da177e4 | 2008 | } |
b56a2d8a VRP |
2009 | try_next: |
2010 | pte = pte_offset_map(pmd, addr); | |
1da177e4 | 2011 | } while (pte++, addr += PAGE_SIZE, addr != end); |
044d66c1 | 2012 | pte_unmap(pte - 1); |
b56a2d8a VRP |
2013 | |
2014 | ret = 0; | |
044d66c1 | 2015 | out: |
8a9f3ccd | 2016 | return ret; |
1da177e4 LT |
2017 | } |
2018 | ||
2019 | static inline int unuse_pmd_range(struct vm_area_struct *vma, pud_t *pud, | |
2020 | unsigned long addr, unsigned long end, | |
b56a2d8a VRP |
2021 | unsigned int type, bool frontswap, |
2022 | unsigned long *fs_pages_to_unuse) | |
1da177e4 LT |
2023 | { |
2024 | pmd_t *pmd; | |
2025 | unsigned long next; | |
8a9f3ccd | 2026 | int ret; |
1da177e4 LT |
2027 | |
2028 | pmd = pmd_offset(pud, addr); | |
2029 | do { | |
dc644a07 | 2030 | cond_resched(); |
1da177e4 | 2031 | next = pmd_addr_end(addr, end); |
1a5a9906 | 2032 | if (pmd_none_or_trans_huge_or_clear_bad(pmd)) |
1da177e4 | 2033 | continue; |
b56a2d8a VRP |
2034 | ret = unuse_pte_range(vma, pmd, addr, next, type, |
2035 | frontswap, fs_pages_to_unuse); | |
8a9f3ccd BS |
2036 | if (ret) |
2037 | return ret; | |
1da177e4 LT |
2038 | } while (pmd++, addr = next, addr != end); |
2039 | return 0; | |
2040 | } | |
2041 | ||
c2febafc | 2042 | static inline int unuse_pud_range(struct vm_area_struct *vma, p4d_t *p4d, |
1da177e4 | 2043 | unsigned long addr, unsigned long end, |
b56a2d8a VRP |
2044 | unsigned int type, bool frontswap, |
2045 | unsigned long *fs_pages_to_unuse) | |
1da177e4 LT |
2046 | { |
2047 | pud_t *pud; | |
2048 | unsigned long next; | |
8a9f3ccd | 2049 | int ret; |
1da177e4 | 2050 | |
c2febafc | 2051 | pud = pud_offset(p4d, addr); |
1da177e4 LT |
2052 | do { |
2053 | next = pud_addr_end(addr, end); | |
2054 | if (pud_none_or_clear_bad(pud)) | |
2055 | continue; | |
b56a2d8a VRP |
2056 | ret = unuse_pmd_range(vma, pud, addr, next, type, |
2057 | frontswap, fs_pages_to_unuse); | |
8a9f3ccd BS |
2058 | if (ret) |
2059 | return ret; | |
1da177e4 LT |
2060 | } while (pud++, addr = next, addr != end); |
2061 | return 0; | |
2062 | } | |
2063 | ||
c2febafc KS |
2064 | static inline int unuse_p4d_range(struct vm_area_struct *vma, pgd_t *pgd, |
2065 | unsigned long addr, unsigned long end, | |
b56a2d8a VRP |
2066 | unsigned int type, bool frontswap, |
2067 | unsigned long *fs_pages_to_unuse) | |
c2febafc KS |
2068 | { |
2069 | p4d_t *p4d; | |
2070 | unsigned long next; | |
2071 | int ret; | |
2072 | ||
2073 | p4d = p4d_offset(pgd, addr); | |
2074 | do { | |
2075 | next = p4d_addr_end(addr, end); | |
2076 | if (p4d_none_or_clear_bad(p4d)) | |
2077 | continue; | |
b56a2d8a VRP |
2078 | ret = unuse_pud_range(vma, p4d, addr, next, type, |
2079 | frontswap, fs_pages_to_unuse); | |
c2febafc KS |
2080 | if (ret) |
2081 | return ret; | |
2082 | } while (p4d++, addr = next, addr != end); | |
2083 | return 0; | |
2084 | } | |
2085 | ||
b56a2d8a VRP |
2086 | static int unuse_vma(struct vm_area_struct *vma, unsigned int type, |
2087 | bool frontswap, unsigned long *fs_pages_to_unuse) | |
1da177e4 LT |
2088 | { |
2089 | pgd_t *pgd; | |
2090 | unsigned long addr, end, next; | |
8a9f3ccd | 2091 | int ret; |
1da177e4 | 2092 | |
b56a2d8a VRP |
2093 | addr = vma->vm_start; |
2094 | end = vma->vm_end; | |
1da177e4 LT |
2095 | |
2096 | pgd = pgd_offset(vma->vm_mm, addr); | |
2097 | do { | |
2098 | next = pgd_addr_end(addr, end); | |
2099 | if (pgd_none_or_clear_bad(pgd)) | |
2100 | continue; | |
b56a2d8a VRP |
2101 | ret = unuse_p4d_range(vma, pgd, addr, next, type, |
2102 | frontswap, fs_pages_to_unuse); | |
8a9f3ccd BS |
2103 | if (ret) |
2104 | return ret; | |
1da177e4 LT |
2105 | } while (pgd++, addr = next, addr != end); |
2106 | return 0; | |
2107 | } | |
2108 | ||
b56a2d8a VRP |
2109 | static int unuse_mm(struct mm_struct *mm, unsigned int type, |
2110 | bool frontswap, unsigned long *fs_pages_to_unuse) | |
1da177e4 LT |
2111 | { |
2112 | struct vm_area_struct *vma; | |
8a9f3ccd | 2113 | int ret = 0; |
1da177e4 | 2114 | |
b56a2d8a | 2115 | down_read(&mm->mmap_sem); |
1da177e4 | 2116 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
b56a2d8a VRP |
2117 | if (vma->anon_vma) { |
2118 | ret = unuse_vma(vma, type, frontswap, | |
2119 | fs_pages_to_unuse); | |
2120 | if (ret) | |
2121 | break; | |
2122 | } | |
dc644a07 | 2123 | cond_resched(); |
1da177e4 | 2124 | } |
1da177e4 | 2125 | up_read(&mm->mmap_sem); |
b56a2d8a | 2126 | return ret; |
1da177e4 LT |
2127 | } |
2128 | ||
2129 | /* | |
38b5faf4 | 2130 | * Scan swap_map (or frontswap_map if frontswap parameter is true) |
b56a2d8a VRP |
2131 | * from current position to next entry still in use. Return 0 |
2132 | * if there are no inuse entries after prev till end of the map. | |
1da177e4 | 2133 | */ |
6eb396dc | 2134 | static unsigned int find_next_to_unuse(struct swap_info_struct *si, |
38b5faf4 | 2135 | unsigned int prev, bool frontswap) |
1da177e4 | 2136 | { |
b56a2d8a | 2137 | unsigned int i; |
8d69aaee | 2138 | unsigned char count; |
1da177e4 LT |
2139 | |
2140 | /* | |
5d337b91 | 2141 | * No need for swap_lock here: we're just looking |
1da177e4 LT |
2142 | * for whether an entry is in use, not modifying it; false |
2143 | * hits are okay, and sys_swapoff() has already prevented new | |
5d337b91 | 2144 | * allocations from this area (while holding swap_lock). |
1da177e4 | 2145 | */ |
b56a2d8a | 2146 | for (i = prev + 1; i < si->max; i++) { |
4db0c3c2 | 2147 | count = READ_ONCE(si->swap_map[i]); |
355cfa73 | 2148 | if (count && swap_count(count) != SWAP_MAP_BAD) |
dc644a07 HD |
2149 | if (!frontswap || frontswap_test(si, i)) |
2150 | break; | |
2151 | if ((i % LATENCY_LIMIT) == 0) | |
2152 | cond_resched(); | |
1da177e4 | 2153 | } |
b56a2d8a VRP |
2154 | |
2155 | if (i == si->max) | |
2156 | i = 0; | |
2157 | ||
1da177e4 LT |
2158 | return i; |
2159 | } | |
2160 | ||
2161 | /* | |
b56a2d8a | 2162 | * If the boolean frontswap is true, only unuse pages_to_unuse pages; |
38b5faf4 | 2163 | * pages_to_unuse==0 means all pages; ignored if frontswap is false |
1da177e4 | 2164 | */ |
38b5faf4 DM |
2165 | int try_to_unuse(unsigned int type, bool frontswap, |
2166 | unsigned long pages_to_unuse) | |
1da177e4 | 2167 | { |
b56a2d8a VRP |
2168 | struct mm_struct *prev_mm; |
2169 | struct mm_struct *mm; | |
2170 | struct list_head *p; | |
2171 | int retval = 0; | |
efa90a98 | 2172 | struct swap_info_struct *si = swap_info[type]; |
1da177e4 LT |
2173 | struct page *page; |
2174 | swp_entry_t entry; | |
b56a2d8a | 2175 | unsigned int i; |
1da177e4 | 2176 | |
21820948 | 2177 | if (!READ_ONCE(si->inuse_pages)) |
b56a2d8a | 2178 | return 0; |
1da177e4 | 2179 | |
b56a2d8a VRP |
2180 | if (!frontswap) |
2181 | pages_to_unuse = 0; | |
2182 | ||
2183 | retry: | |
2184 | retval = shmem_unuse(type, frontswap, &pages_to_unuse); | |
2185 | if (retval) | |
2186 | goto out; | |
2187 | ||
2188 | prev_mm = &init_mm; | |
2189 | mmget(prev_mm); | |
2190 | ||
2191 | spin_lock(&mmlist_lock); | |
2192 | p = &init_mm.mmlist; | |
21820948 | 2193 | while (READ_ONCE(si->inuse_pages) && |
64165b1a HD |
2194 | !signal_pending(current) && |
2195 | (p = p->next) != &init_mm.mmlist) { | |
1da177e4 | 2196 | |
b56a2d8a VRP |
2197 | mm = list_entry(p, struct mm_struct, mmlist); |
2198 | if (!mmget_not_zero(mm)) | |
2199 | continue; | |
2200 | spin_unlock(&mmlist_lock); | |
2201 | mmput(prev_mm); | |
2202 | prev_mm = mm; | |
2203 | retval = unuse_mm(mm, type, frontswap, &pages_to_unuse); | |
1da177e4 | 2204 | |
b56a2d8a VRP |
2205 | if (retval) { |
2206 | mmput(prev_mm); | |
2207 | goto out; | |
1da177e4 LT |
2208 | } |
2209 | ||
2210 | /* | |
b56a2d8a VRP |
2211 | * Make sure that we aren't completely killing |
2212 | * interactive performance. | |
1da177e4 | 2213 | */ |
b56a2d8a VRP |
2214 | cond_resched(); |
2215 | spin_lock(&mmlist_lock); | |
2216 | } | |
2217 | spin_unlock(&mmlist_lock); | |
1da177e4 | 2218 | |
b56a2d8a | 2219 | mmput(prev_mm); |
1da177e4 | 2220 | |
b56a2d8a | 2221 | i = 0; |
21820948 | 2222 | while (READ_ONCE(si->inuse_pages) && |
64165b1a HD |
2223 | !signal_pending(current) && |
2224 | (i = find_next_to_unuse(si, i, frontswap)) != 0) { | |
1da177e4 | 2225 | |
b56a2d8a VRP |
2226 | entry = swp_entry(type, i); |
2227 | page = find_get_page(swap_address_space(entry), i); | |
2228 | if (!page) | |
2229 | continue; | |
68bdc8d6 HD |
2230 | |
2231 | /* | |
2232 | * It is conceivable that a racing task removed this page from | |
b56a2d8a VRP |
2233 | * swap cache just before we acquired the page lock. The page |
2234 | * might even be back in swap cache on another swap area. But | |
2235 | * that is okay, try_to_free_swap() only removes stale pages. | |
1da177e4 | 2236 | */ |
b56a2d8a VRP |
2237 | lock_page(page); |
2238 | wait_on_page_writeback(page); | |
2239 | try_to_free_swap(page); | |
1da177e4 | 2240 | unlock_page(page); |
09cbfeaf | 2241 | put_page(page); |
1da177e4 LT |
2242 | |
2243 | /* | |
b56a2d8a VRP |
2244 | * For frontswap, we just need to unuse pages_to_unuse, if |
2245 | * it was specified. Need not check frontswap again here as | |
2246 | * we already zeroed out pages_to_unuse if not frontswap. | |
1da177e4 | 2247 | */ |
b56a2d8a VRP |
2248 | if (pages_to_unuse && --pages_to_unuse == 0) |
2249 | goto out; | |
1da177e4 LT |
2250 | } |
2251 | ||
b56a2d8a VRP |
2252 | /* |
2253 | * Lets check again to see if there are still swap entries in the map. | |
2254 | * If yes, we would need to do retry the unuse logic again. | |
2255 | * Under global memory pressure, swap entries can be reinserted back | |
2256 | * into process space after the mmlist loop above passes over them. | |
dd862deb | 2257 | * |
af53d3e9 HD |
2258 | * Limit the number of retries? No: when mmget_not_zero() above fails, |
2259 | * that mm is likely to be freeing swap from exit_mmap(), which proceeds | |
2260 | * at its own independent pace; and even shmem_writepage() could have | |
2261 | * been preempted after get_swap_page(), temporarily hiding that swap. | |
2262 | * It's easy and robust (though cpu-intensive) just to keep retrying. | |
b56a2d8a | 2263 | */ |
21820948 | 2264 | if (READ_ONCE(si->inuse_pages)) { |
64165b1a HD |
2265 | if (!signal_pending(current)) |
2266 | goto retry; | |
2267 | retval = -EINTR; | |
2268 | } | |
b56a2d8a VRP |
2269 | out: |
2270 | return (retval == FRONTSWAP_PAGES_UNUSED) ? 0 : retval; | |
1da177e4 LT |
2271 | } |
2272 | ||
2273 | /* | |
5d337b91 HD |
2274 | * After a successful try_to_unuse, if no swap is now in use, we know |
2275 | * we can empty the mmlist. swap_lock must be held on entry and exit. | |
2276 | * Note that mmlist_lock nests inside swap_lock, and an mm must be | |
1da177e4 LT |
2277 | * added to the mmlist just after page_duplicate - before would be racy. |
2278 | */ | |
2279 | static void drain_mmlist(void) | |
2280 | { | |
2281 | struct list_head *p, *next; | |
efa90a98 | 2282 | unsigned int type; |
1da177e4 | 2283 | |
efa90a98 HD |
2284 | for (type = 0; type < nr_swapfiles; type++) |
2285 | if (swap_info[type]->inuse_pages) | |
1da177e4 LT |
2286 | return; |
2287 | spin_lock(&mmlist_lock); | |
2288 | list_for_each_safe(p, next, &init_mm.mmlist) | |
2289 | list_del_init(p); | |
2290 | spin_unlock(&mmlist_lock); | |
2291 | } | |
2292 | ||
2293 | /* | |
2294 | * Use this swapdev's extent info to locate the (PAGE_SIZE) block which | |
d4906e1a LS |
2295 | * corresponds to page offset for the specified swap entry. |
2296 | * Note that the type of this function is sector_t, but it returns page offset | |
2297 | * into the bdev, not sector offset. | |
1da177e4 | 2298 | */ |
d4906e1a | 2299 | static sector_t map_swap_entry(swp_entry_t entry, struct block_device **bdev) |
1da177e4 | 2300 | { |
f29ad6a9 | 2301 | struct swap_info_struct *sis; |
f29ad6a9 HD |
2302 | struct swap_extent *se; |
2303 | pgoff_t offset; | |
2304 | ||
c10d38cc | 2305 | sis = swp_swap_info(entry); |
f29ad6a9 HD |
2306 | *bdev = sis->bdev; |
2307 | ||
2308 | offset = swp_offset(entry); | |
4efaceb1 AL |
2309 | se = offset_to_swap_extent(sis, offset); |
2310 | return se->start_block + (offset - se->start_page); | |
1da177e4 LT |
2311 | } |
2312 | ||
d4906e1a LS |
2313 | /* |
2314 | * Returns the page offset into bdev for the specified page's swap entry. | |
2315 | */ | |
2316 | sector_t map_swap_page(struct page *page, struct block_device **bdev) | |
2317 | { | |
2318 | swp_entry_t entry; | |
2319 | entry.val = page_private(page); | |
2320 | return map_swap_entry(entry, bdev); | |
2321 | } | |
2322 | ||
1da177e4 LT |
2323 | /* |
2324 | * Free all of a swapdev's extent information | |
2325 | */ | |
2326 | static void destroy_swap_extents(struct swap_info_struct *sis) | |
2327 | { | |
4efaceb1 AL |
2328 | while (!RB_EMPTY_ROOT(&sis->swap_extent_root)) { |
2329 | struct rb_node *rb = sis->swap_extent_root.rb_node; | |
2330 | struct swap_extent *se = rb_entry(rb, struct swap_extent, rb_node); | |
1da177e4 | 2331 | |
4efaceb1 | 2332 | rb_erase(rb, &sis->swap_extent_root); |
1da177e4 LT |
2333 | kfree(se); |
2334 | } | |
62c230bc | 2335 | |
bc4ae27d | 2336 | if (sis->flags & SWP_ACTIVATED) { |
62c230bc MG |
2337 | struct file *swap_file = sis->swap_file; |
2338 | struct address_space *mapping = swap_file->f_mapping; | |
2339 | ||
bc4ae27d OS |
2340 | sis->flags &= ~SWP_ACTIVATED; |
2341 | if (mapping->a_ops->swap_deactivate) | |
2342 | mapping->a_ops->swap_deactivate(swap_file); | |
62c230bc | 2343 | } |
1da177e4 LT |
2344 | } |
2345 | ||
2346 | /* | |
2347 | * Add a block range (and the corresponding page range) into this swapdev's | |
4efaceb1 | 2348 | * extent tree. |
1da177e4 | 2349 | * |
11d31886 | 2350 | * This function rather assumes that it is called in ascending page order. |
1da177e4 | 2351 | */ |
a509bc1a | 2352 | int |
1da177e4 LT |
2353 | add_swap_extent(struct swap_info_struct *sis, unsigned long start_page, |
2354 | unsigned long nr_pages, sector_t start_block) | |
2355 | { | |
4efaceb1 | 2356 | struct rb_node **link = &sis->swap_extent_root.rb_node, *parent = NULL; |
1da177e4 LT |
2357 | struct swap_extent *se; |
2358 | struct swap_extent *new_se; | |
4efaceb1 AL |
2359 | |
2360 | /* | |
2361 | * place the new node at the right most since the | |
2362 | * function is called in ascending page order. | |
2363 | */ | |
2364 | while (*link) { | |
2365 | parent = *link; | |
2366 | link = &parent->rb_right; | |
2367 | } | |
2368 | ||
2369 | if (parent) { | |
2370 | se = rb_entry(parent, struct swap_extent, rb_node); | |
11d31886 HD |
2371 | BUG_ON(se->start_page + se->nr_pages != start_page); |
2372 | if (se->start_block + se->nr_pages == start_block) { | |
1da177e4 LT |
2373 | /* Merge it */ |
2374 | se->nr_pages += nr_pages; | |
2375 | return 0; | |
2376 | } | |
1da177e4 LT |
2377 | } |
2378 | ||
4efaceb1 | 2379 | /* No merge, insert a new extent. */ |
1da177e4 LT |
2380 | new_se = kmalloc(sizeof(*se), GFP_KERNEL); |
2381 | if (new_se == NULL) | |
2382 | return -ENOMEM; | |
2383 | new_se->start_page = start_page; | |
2384 | new_se->nr_pages = nr_pages; | |
2385 | new_se->start_block = start_block; | |
2386 | ||
4efaceb1 AL |
2387 | rb_link_node(&new_se->rb_node, parent, link); |
2388 | rb_insert_color(&new_se->rb_node, &sis->swap_extent_root); | |
53092a74 | 2389 | return 1; |
1da177e4 | 2390 | } |
aa8aa8a3 | 2391 | EXPORT_SYMBOL_GPL(add_swap_extent); |
1da177e4 LT |
2392 | |
2393 | /* | |
2394 | * A `swap extent' is a simple thing which maps a contiguous range of pages | |
2395 | * onto a contiguous range of disk blocks. An ordered list of swap extents | |
2396 | * is built at swapon time and is then used at swap_writepage/swap_readpage | |
2397 | * time for locating where on disk a page belongs. | |
2398 | * | |
2399 | * If the swapfile is an S_ISBLK block device, a single extent is installed. | |
2400 | * This is done so that the main operating code can treat S_ISBLK and S_ISREG | |
2401 | * swap files identically. | |
2402 | * | |
2403 | * Whether the swapdev is an S_ISREG file or an S_ISBLK blockdev, the swap | |
2404 | * extent list operates in PAGE_SIZE disk blocks. Both S_ISREG and S_ISBLK | |
2405 | * swapfiles are handled *identically* after swapon time. | |
2406 | * | |
2407 | * For S_ISREG swapfiles, setup_swap_extents() will walk all the file's blocks | |
2408 | * and will parse them into an ordered extent list, in PAGE_SIZE chunks. If | |
2409 | * some stray blocks are found which do not fall within the PAGE_SIZE alignment | |
2410 | * requirements, they are simply tossed out - we will never use those blocks | |
2411 | * for swapping. | |
2412 | * | |
1638045c DW |
2413 | * For all swap devices we set S_SWAPFILE across the life of the swapon. This |
2414 | * prevents users from writing to the swap device, which will corrupt memory. | |
1da177e4 LT |
2415 | * |
2416 | * The amount of disk space which a single swap extent represents varies. | |
2417 | * Typically it is in the 1-4 megabyte range. So we can have hundreds of | |
2418 | * extents in the list. To avoid much list walking, we cache the previous | |
2419 | * search location in `curr_swap_extent', and start new searches from there. | |
2420 | * This is extremely effective. The average number of iterations in | |
2421 | * map_swap_page() has been measured at about 0.3 per page. - akpm. | |
2422 | */ | |
53092a74 | 2423 | static int setup_swap_extents(struct swap_info_struct *sis, sector_t *span) |
1da177e4 | 2424 | { |
62c230bc MG |
2425 | struct file *swap_file = sis->swap_file; |
2426 | struct address_space *mapping = swap_file->f_mapping; | |
2427 | struct inode *inode = mapping->host; | |
1da177e4 LT |
2428 | int ret; |
2429 | ||
1da177e4 LT |
2430 | if (S_ISBLK(inode->i_mode)) { |
2431 | ret = add_swap_extent(sis, 0, sis->max, 0); | |
53092a74 | 2432 | *span = sis->pages; |
a509bc1a | 2433 | return ret; |
1da177e4 LT |
2434 | } |
2435 | ||
62c230bc | 2436 | if (mapping->a_ops->swap_activate) { |
a509bc1a | 2437 | ret = mapping->a_ops->swap_activate(sis, swap_file, span); |
bc4ae27d OS |
2438 | if (ret >= 0) |
2439 | sis->flags |= SWP_ACTIVATED; | |
62c230bc | 2440 | if (!ret) { |
bc4ae27d | 2441 | sis->flags |= SWP_FS; |
62c230bc MG |
2442 | ret = add_swap_extent(sis, 0, sis->max, 0); |
2443 | *span = sis->pages; | |
2444 | } | |
a509bc1a | 2445 | return ret; |
62c230bc MG |
2446 | } |
2447 | ||
a509bc1a | 2448 | return generic_swapfile_activate(sis, swap_file, span); |
1da177e4 LT |
2449 | } |
2450 | ||
a2468cc9 AL |
2451 | static int swap_node(struct swap_info_struct *p) |
2452 | { | |
2453 | struct block_device *bdev; | |
2454 | ||
2455 | if (p->bdev) | |
2456 | bdev = p->bdev; | |
2457 | else | |
2458 | bdev = p->swap_file->f_inode->i_sb->s_bdev; | |
2459 | ||
2460 | return bdev ? bdev->bd_disk->node_id : NUMA_NO_NODE; | |
2461 | } | |
2462 | ||
eb085574 HY |
2463 | static void setup_swap_info(struct swap_info_struct *p, int prio, |
2464 | unsigned char *swap_map, | |
2465 | struct swap_cluster_info *cluster_info) | |
40531542 | 2466 | { |
a2468cc9 AL |
2467 | int i; |
2468 | ||
40531542 CEB |
2469 | if (prio >= 0) |
2470 | p->prio = prio; | |
2471 | else | |
2472 | p->prio = --least_priority; | |
18ab4d4c DS |
2473 | /* |
2474 | * the plist prio is negated because plist ordering is | |
2475 | * low-to-high, while swap ordering is high-to-low | |
2476 | */ | |
2477 | p->list.prio = -p->prio; | |
a2468cc9 AL |
2478 | for_each_node(i) { |
2479 | if (p->prio >= 0) | |
2480 | p->avail_lists[i].prio = -p->prio; | |
2481 | else { | |
2482 | if (swap_node(p) == i) | |
2483 | p->avail_lists[i].prio = 1; | |
2484 | else | |
2485 | p->avail_lists[i].prio = -p->prio; | |
2486 | } | |
2487 | } | |
40531542 | 2488 | p->swap_map = swap_map; |
2a8f9449 | 2489 | p->cluster_info = cluster_info; |
eb085574 HY |
2490 | } |
2491 | ||
2492 | static void _enable_swap_info(struct swap_info_struct *p) | |
2493 | { | |
2494 | p->flags |= SWP_WRITEOK | SWP_VALID; | |
ec8acf20 | 2495 | atomic_long_add(p->pages, &nr_swap_pages); |
40531542 CEB |
2496 | total_swap_pages += p->pages; |
2497 | ||
adfab836 | 2498 | assert_spin_locked(&swap_lock); |
adfab836 | 2499 | /* |
18ab4d4c DS |
2500 | * both lists are plists, and thus priority ordered. |
2501 | * swap_active_head needs to be priority ordered for swapoff(), | |
2502 | * which on removal of any swap_info_struct with an auto-assigned | |
2503 | * (i.e. negative) priority increments the auto-assigned priority | |
2504 | * of any lower-priority swap_info_structs. | |
2505 | * swap_avail_head needs to be priority ordered for get_swap_page(), | |
2506 | * which allocates swap pages from the highest available priority | |
2507 | * swap_info_struct. | |
adfab836 | 2508 | */ |
18ab4d4c | 2509 | plist_add(&p->list, &swap_active_head); |
a2468cc9 | 2510 | add_to_avail_list(p); |
cf0cac0a CEB |
2511 | } |
2512 | ||
2513 | static void enable_swap_info(struct swap_info_struct *p, int prio, | |
2514 | unsigned char *swap_map, | |
2a8f9449 | 2515 | struct swap_cluster_info *cluster_info, |
cf0cac0a CEB |
2516 | unsigned long *frontswap_map) |
2517 | { | |
4f89849d | 2518 | frontswap_init(p->type, frontswap_map); |
cf0cac0a | 2519 | spin_lock(&swap_lock); |
ec8acf20 | 2520 | spin_lock(&p->lock); |
eb085574 HY |
2521 | setup_swap_info(p, prio, swap_map, cluster_info); |
2522 | spin_unlock(&p->lock); | |
2523 | spin_unlock(&swap_lock); | |
2524 | /* | |
2525 | * Guarantee swap_map, cluster_info, etc. fields are valid | |
2526 | * between get/put_swap_device() if SWP_VALID bit is set | |
2527 | */ | |
2528 | synchronize_rcu(); | |
2529 | spin_lock(&swap_lock); | |
2530 | spin_lock(&p->lock); | |
2531 | _enable_swap_info(p); | |
ec8acf20 | 2532 | spin_unlock(&p->lock); |
cf0cac0a CEB |
2533 | spin_unlock(&swap_lock); |
2534 | } | |
2535 | ||
2536 | static void reinsert_swap_info(struct swap_info_struct *p) | |
2537 | { | |
2538 | spin_lock(&swap_lock); | |
ec8acf20 | 2539 | spin_lock(&p->lock); |
eb085574 HY |
2540 | setup_swap_info(p, p->prio, p->swap_map, p->cluster_info); |
2541 | _enable_swap_info(p); | |
ec8acf20 | 2542 | spin_unlock(&p->lock); |
40531542 CEB |
2543 | spin_unlock(&swap_lock); |
2544 | } | |
2545 | ||
67afa38e TC |
2546 | bool has_usable_swap(void) |
2547 | { | |
2548 | bool ret = true; | |
2549 | ||
2550 | spin_lock(&swap_lock); | |
2551 | if (plist_head_empty(&swap_active_head)) | |
2552 | ret = false; | |
2553 | spin_unlock(&swap_lock); | |
2554 | return ret; | |
2555 | } | |
2556 | ||
c4ea37c2 | 2557 | SYSCALL_DEFINE1(swapoff, const char __user *, specialfile) |
1da177e4 | 2558 | { |
73c34b6a | 2559 | struct swap_info_struct *p = NULL; |
8d69aaee | 2560 | unsigned char *swap_map; |
2a8f9449 | 2561 | struct swap_cluster_info *cluster_info; |
4f89849d | 2562 | unsigned long *frontswap_map; |
1da177e4 LT |
2563 | struct file *swap_file, *victim; |
2564 | struct address_space *mapping; | |
2565 | struct inode *inode; | |
91a27b2a | 2566 | struct filename *pathname; |
adfab836 | 2567 | int err, found = 0; |
5b808a23 | 2568 | unsigned int old_block_size; |
886bb7e9 | 2569 | |
1da177e4 LT |
2570 | if (!capable(CAP_SYS_ADMIN)) |
2571 | return -EPERM; | |
2572 | ||
191c5424 AV |
2573 | BUG_ON(!current->mm); |
2574 | ||
1da177e4 | 2575 | pathname = getname(specialfile); |
1da177e4 | 2576 | if (IS_ERR(pathname)) |
f58b59c1 | 2577 | return PTR_ERR(pathname); |
1da177e4 | 2578 | |
669abf4e | 2579 | victim = file_open_name(pathname, O_RDWR|O_LARGEFILE, 0); |
1da177e4 LT |
2580 | err = PTR_ERR(victim); |
2581 | if (IS_ERR(victim)) | |
2582 | goto out; | |
2583 | ||
2584 | mapping = victim->f_mapping; | |
5d337b91 | 2585 | spin_lock(&swap_lock); |
18ab4d4c | 2586 | plist_for_each_entry(p, &swap_active_head, list) { |
22c6f8fd | 2587 | if (p->flags & SWP_WRITEOK) { |
adfab836 DS |
2588 | if (p->swap_file->f_mapping == mapping) { |
2589 | found = 1; | |
1da177e4 | 2590 | break; |
adfab836 | 2591 | } |
1da177e4 | 2592 | } |
1da177e4 | 2593 | } |
adfab836 | 2594 | if (!found) { |
1da177e4 | 2595 | err = -EINVAL; |
5d337b91 | 2596 | spin_unlock(&swap_lock); |
1da177e4 LT |
2597 | goto out_dput; |
2598 | } | |
191c5424 | 2599 | if (!security_vm_enough_memory_mm(current->mm, p->pages)) |
1da177e4 LT |
2600 | vm_unacct_memory(p->pages); |
2601 | else { | |
2602 | err = -ENOMEM; | |
5d337b91 | 2603 | spin_unlock(&swap_lock); |
1da177e4 LT |
2604 | goto out_dput; |
2605 | } | |
a2468cc9 | 2606 | del_from_avail_list(p); |
ec8acf20 | 2607 | spin_lock(&p->lock); |
78ecba08 | 2608 | if (p->prio < 0) { |
adfab836 | 2609 | struct swap_info_struct *si = p; |
a2468cc9 | 2610 | int nid; |
adfab836 | 2611 | |
18ab4d4c | 2612 | plist_for_each_entry_continue(si, &swap_active_head, list) { |
adfab836 | 2613 | si->prio++; |
18ab4d4c | 2614 | si->list.prio--; |
a2468cc9 AL |
2615 | for_each_node(nid) { |
2616 | if (si->avail_lists[nid].prio != 1) | |
2617 | si->avail_lists[nid].prio--; | |
2618 | } | |
adfab836 | 2619 | } |
78ecba08 HD |
2620 | least_priority++; |
2621 | } | |
18ab4d4c | 2622 | plist_del(&p->list, &swap_active_head); |
ec8acf20 | 2623 | atomic_long_sub(p->pages, &nr_swap_pages); |
1da177e4 LT |
2624 | total_swap_pages -= p->pages; |
2625 | p->flags &= ~SWP_WRITEOK; | |
ec8acf20 | 2626 | spin_unlock(&p->lock); |
5d337b91 | 2627 | spin_unlock(&swap_lock); |
fb4f88dc | 2628 | |
039939a6 TC |
2629 | disable_swap_slots_cache_lock(); |
2630 | ||
e1e12d2f | 2631 | set_current_oom_origin(); |
adfab836 | 2632 | err = try_to_unuse(p->type, false, 0); /* force unuse all pages */ |
e1e12d2f | 2633 | clear_current_oom_origin(); |
1da177e4 | 2634 | |
1da177e4 LT |
2635 | if (err) { |
2636 | /* re-insert swap space back into swap_list */ | |
cf0cac0a | 2637 | reinsert_swap_info(p); |
039939a6 | 2638 | reenable_swap_slots_cache_unlock(); |
1da177e4 LT |
2639 | goto out_dput; |
2640 | } | |
52b7efdb | 2641 | |
039939a6 TC |
2642 | reenable_swap_slots_cache_unlock(); |
2643 | ||
eb085574 HY |
2644 | spin_lock(&swap_lock); |
2645 | spin_lock(&p->lock); | |
2646 | p->flags &= ~SWP_VALID; /* mark swap device as invalid */ | |
2647 | spin_unlock(&p->lock); | |
2648 | spin_unlock(&swap_lock); | |
2649 | /* | |
2650 | * wait for swap operations protected by get/put_swap_device() | |
2651 | * to complete | |
2652 | */ | |
2653 | synchronize_rcu(); | |
2654 | ||
815c2c54 SL |
2655 | flush_work(&p->discard_work); |
2656 | ||
5d337b91 | 2657 | destroy_swap_extents(p); |
570a335b HD |
2658 | if (p->flags & SWP_CONTINUED) |
2659 | free_swap_count_continuations(p); | |
2660 | ||
81a0298b HY |
2661 | if (!p->bdev || !blk_queue_nonrot(bdev_get_queue(p->bdev))) |
2662 | atomic_dec(&nr_rotate_swap); | |
2663 | ||
fc0abb14 | 2664 | mutex_lock(&swapon_mutex); |
5d337b91 | 2665 | spin_lock(&swap_lock); |
ec8acf20 | 2666 | spin_lock(&p->lock); |
5d337b91 HD |
2667 | drain_mmlist(); |
2668 | ||
52b7efdb | 2669 | /* wait for anyone still in scan_swap_map */ |
52b7efdb HD |
2670 | p->highest_bit = 0; /* cuts scans short */ |
2671 | while (p->flags >= SWP_SCANNING) { | |
ec8acf20 | 2672 | spin_unlock(&p->lock); |
5d337b91 | 2673 | spin_unlock(&swap_lock); |
13e4b57f | 2674 | schedule_timeout_uninterruptible(1); |
5d337b91 | 2675 | spin_lock(&swap_lock); |
ec8acf20 | 2676 | spin_lock(&p->lock); |
52b7efdb | 2677 | } |
52b7efdb | 2678 | |
1da177e4 | 2679 | swap_file = p->swap_file; |
5b808a23 | 2680 | old_block_size = p->old_block_size; |
1da177e4 LT |
2681 | p->swap_file = NULL; |
2682 | p->max = 0; | |
2683 | swap_map = p->swap_map; | |
2684 | p->swap_map = NULL; | |
2a8f9449 SL |
2685 | cluster_info = p->cluster_info; |
2686 | p->cluster_info = NULL; | |
4f89849d | 2687 | frontswap_map = frontswap_map_get(p); |
ec8acf20 | 2688 | spin_unlock(&p->lock); |
5d337b91 | 2689 | spin_unlock(&swap_lock); |
adfab836 | 2690 | frontswap_invalidate_area(p->type); |
58e97ba6 | 2691 | frontswap_map_set(p, NULL); |
fc0abb14 | 2692 | mutex_unlock(&swapon_mutex); |
ebc2a1a6 SL |
2693 | free_percpu(p->percpu_cluster); |
2694 | p->percpu_cluster = NULL; | |
49070588 HY |
2695 | free_percpu(p->cluster_next_cpu); |
2696 | p->cluster_next_cpu = NULL; | |
1da177e4 | 2697 | vfree(swap_map); |
54f180d3 HY |
2698 | kvfree(cluster_info); |
2699 | kvfree(frontswap_map); | |
2de1a7e4 | 2700 | /* Destroy swap account information */ |
adfab836 | 2701 | swap_cgroup_swapoff(p->type); |
4b3ef9da | 2702 | exit_swap_address_space(p->type); |
27a7faa0 | 2703 | |
1da177e4 LT |
2704 | inode = mapping->host; |
2705 | if (S_ISBLK(inode->i_mode)) { | |
2706 | struct block_device *bdev = I_BDEV(inode); | |
1638045c | 2707 | |
5b808a23 | 2708 | set_blocksize(bdev, old_block_size); |
e525fd89 | 2709 | blkdev_put(bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL); |
1da177e4 | 2710 | } |
1638045c DW |
2711 | |
2712 | inode_lock(inode); | |
2713 | inode->i_flags &= ~S_SWAPFILE; | |
2714 | inode_unlock(inode); | |
1da177e4 | 2715 | filp_close(swap_file, NULL); |
f893ab41 WY |
2716 | |
2717 | /* | |
2718 | * Clear the SWP_USED flag after all resources are freed so that swapon | |
2719 | * can reuse this swap_info in alloc_swap_info() safely. It is ok to | |
2720 | * not hold p->lock after we cleared its SWP_WRITEOK. | |
2721 | */ | |
2722 | spin_lock(&swap_lock); | |
2723 | p->flags = 0; | |
2724 | spin_unlock(&swap_lock); | |
2725 | ||
1da177e4 | 2726 | err = 0; |
66d7dd51 KS |
2727 | atomic_inc(&proc_poll_event); |
2728 | wake_up_interruptible(&proc_poll_wait); | |
1da177e4 LT |
2729 | |
2730 | out_dput: | |
2731 | filp_close(victim, NULL); | |
2732 | out: | |
f58b59c1 | 2733 | putname(pathname); |
1da177e4 LT |
2734 | return err; |
2735 | } | |
2736 | ||
2737 | #ifdef CONFIG_PROC_FS | |
9dd95748 | 2738 | static __poll_t swaps_poll(struct file *file, poll_table *wait) |
66d7dd51 | 2739 | { |
f1514638 | 2740 | struct seq_file *seq = file->private_data; |
66d7dd51 KS |
2741 | |
2742 | poll_wait(file, &proc_poll_wait, wait); | |
2743 | ||
f1514638 KS |
2744 | if (seq->poll_event != atomic_read(&proc_poll_event)) { |
2745 | seq->poll_event = atomic_read(&proc_poll_event); | |
a9a08845 | 2746 | return EPOLLIN | EPOLLRDNORM | EPOLLERR | EPOLLPRI; |
66d7dd51 KS |
2747 | } |
2748 | ||
a9a08845 | 2749 | return EPOLLIN | EPOLLRDNORM; |
66d7dd51 KS |
2750 | } |
2751 | ||
1da177e4 LT |
2752 | /* iterator */ |
2753 | static void *swap_start(struct seq_file *swap, loff_t *pos) | |
2754 | { | |
efa90a98 HD |
2755 | struct swap_info_struct *si; |
2756 | int type; | |
1da177e4 LT |
2757 | loff_t l = *pos; |
2758 | ||
fc0abb14 | 2759 | mutex_lock(&swapon_mutex); |
1da177e4 | 2760 | |
881e4aab SS |
2761 | if (!l) |
2762 | return SEQ_START_TOKEN; | |
2763 | ||
c10d38cc | 2764 | for (type = 0; (si = swap_type_to_swap_info(type)); type++) { |
efa90a98 | 2765 | if (!(si->flags & SWP_USED) || !si->swap_map) |
1da177e4 | 2766 | continue; |
881e4aab | 2767 | if (!--l) |
efa90a98 | 2768 | return si; |
1da177e4 LT |
2769 | } |
2770 | ||
2771 | return NULL; | |
2772 | } | |
2773 | ||
2774 | static void *swap_next(struct seq_file *swap, void *v, loff_t *pos) | |
2775 | { | |
efa90a98 HD |
2776 | struct swap_info_struct *si = v; |
2777 | int type; | |
1da177e4 | 2778 | |
881e4aab | 2779 | if (v == SEQ_START_TOKEN) |
efa90a98 HD |
2780 | type = 0; |
2781 | else | |
2782 | type = si->type + 1; | |
881e4aab | 2783 | |
10c8d69f | 2784 | ++(*pos); |
c10d38cc | 2785 | for (; (si = swap_type_to_swap_info(type)); type++) { |
efa90a98 | 2786 | if (!(si->flags & SWP_USED) || !si->swap_map) |
1da177e4 | 2787 | continue; |
efa90a98 | 2788 | return si; |
1da177e4 LT |
2789 | } |
2790 | ||
2791 | return NULL; | |
2792 | } | |
2793 | ||
2794 | static void swap_stop(struct seq_file *swap, void *v) | |
2795 | { | |
fc0abb14 | 2796 | mutex_unlock(&swapon_mutex); |
1da177e4 LT |
2797 | } |
2798 | ||
2799 | static int swap_show(struct seq_file *swap, void *v) | |
2800 | { | |
efa90a98 | 2801 | struct swap_info_struct *si = v; |
1da177e4 LT |
2802 | struct file *file; |
2803 | int len; | |
2804 | ||
efa90a98 | 2805 | if (si == SEQ_START_TOKEN) { |
881e4aab SS |
2806 | seq_puts(swap,"Filename\t\t\t\tType\t\tSize\tUsed\tPriority\n"); |
2807 | return 0; | |
2808 | } | |
1da177e4 | 2809 | |
efa90a98 | 2810 | file = si->swap_file; |
2726d566 | 2811 | len = seq_file_path(swap, file, " \t\n\\"); |
6eb396dc | 2812 | seq_printf(swap, "%*s%s\t%u\t%u\t%d\n", |
886bb7e9 | 2813 | len < 40 ? 40 - len : 1, " ", |
496ad9aa | 2814 | S_ISBLK(file_inode(file)->i_mode) ? |
1da177e4 | 2815 | "partition" : "file\t", |
efa90a98 HD |
2816 | si->pages << (PAGE_SHIFT - 10), |
2817 | si->inuse_pages << (PAGE_SHIFT - 10), | |
2818 | si->prio); | |
1da177e4 LT |
2819 | return 0; |
2820 | } | |
2821 | ||
15ad7cdc | 2822 | static const struct seq_operations swaps_op = { |
1da177e4 LT |
2823 | .start = swap_start, |
2824 | .next = swap_next, | |
2825 | .stop = swap_stop, | |
2826 | .show = swap_show | |
2827 | }; | |
2828 | ||
2829 | static int swaps_open(struct inode *inode, struct file *file) | |
2830 | { | |
f1514638 | 2831 | struct seq_file *seq; |
66d7dd51 KS |
2832 | int ret; |
2833 | ||
66d7dd51 | 2834 | ret = seq_open(file, &swaps_op); |
f1514638 | 2835 | if (ret) |
66d7dd51 | 2836 | return ret; |
66d7dd51 | 2837 | |
f1514638 KS |
2838 | seq = file->private_data; |
2839 | seq->poll_event = atomic_read(&proc_poll_event); | |
2840 | return 0; | |
1da177e4 LT |
2841 | } |
2842 | ||
97a32539 | 2843 | static const struct proc_ops swaps_proc_ops = { |
d919b33d | 2844 | .proc_flags = PROC_ENTRY_PERMANENT, |
97a32539 AD |
2845 | .proc_open = swaps_open, |
2846 | .proc_read = seq_read, | |
2847 | .proc_lseek = seq_lseek, | |
2848 | .proc_release = seq_release, | |
2849 | .proc_poll = swaps_poll, | |
1da177e4 LT |
2850 | }; |
2851 | ||
2852 | static int __init procswaps_init(void) | |
2853 | { | |
97a32539 | 2854 | proc_create("swaps", 0, NULL, &swaps_proc_ops); |
1da177e4 LT |
2855 | return 0; |
2856 | } | |
2857 | __initcall(procswaps_init); | |
2858 | #endif /* CONFIG_PROC_FS */ | |
2859 | ||
1796316a JB |
2860 | #ifdef MAX_SWAPFILES_CHECK |
2861 | static int __init max_swapfiles_check(void) | |
2862 | { | |
2863 | MAX_SWAPFILES_CHECK(); | |
2864 | return 0; | |
2865 | } | |
2866 | late_initcall(max_swapfiles_check); | |
2867 | #endif | |
2868 | ||
53cbb243 | 2869 | static struct swap_info_struct *alloc_swap_info(void) |
1da177e4 | 2870 | { |
73c34b6a | 2871 | struct swap_info_struct *p; |
1da177e4 | 2872 | unsigned int type; |
a2468cc9 | 2873 | int i; |
efa90a98 | 2874 | |
96008744 | 2875 | p = kvzalloc(struct_size(p, avail_lists, nr_node_ids), GFP_KERNEL); |
efa90a98 | 2876 | if (!p) |
53cbb243 | 2877 | return ERR_PTR(-ENOMEM); |
efa90a98 | 2878 | |
5d337b91 | 2879 | spin_lock(&swap_lock); |
efa90a98 HD |
2880 | for (type = 0; type < nr_swapfiles; type++) { |
2881 | if (!(swap_info[type]->flags & SWP_USED)) | |
1da177e4 | 2882 | break; |
efa90a98 | 2883 | } |
0697212a | 2884 | if (type >= MAX_SWAPFILES) { |
5d337b91 | 2885 | spin_unlock(&swap_lock); |
873d7bcf | 2886 | kvfree(p); |
730c0581 | 2887 | return ERR_PTR(-EPERM); |
1da177e4 | 2888 | } |
efa90a98 HD |
2889 | if (type >= nr_swapfiles) { |
2890 | p->type = type; | |
c10d38cc | 2891 | WRITE_ONCE(swap_info[type], p); |
efa90a98 HD |
2892 | /* |
2893 | * Write swap_info[type] before nr_swapfiles, in case a | |
2894 | * racing procfs swap_start() or swap_next() is reading them. | |
2895 | * (We never shrink nr_swapfiles, we never free this entry.) | |
2896 | */ | |
2897 | smp_wmb(); | |
c10d38cc | 2898 | WRITE_ONCE(nr_swapfiles, nr_swapfiles + 1); |
efa90a98 | 2899 | } else { |
873d7bcf | 2900 | kvfree(p); |
efa90a98 HD |
2901 | p = swap_info[type]; |
2902 | /* | |
2903 | * Do not memset this entry: a racing procfs swap_next() | |
2904 | * would be relying on p->type to remain valid. | |
2905 | */ | |
2906 | } | |
4efaceb1 | 2907 | p->swap_extent_root = RB_ROOT; |
18ab4d4c | 2908 | plist_node_init(&p->list, 0); |
a2468cc9 AL |
2909 | for_each_node(i) |
2910 | plist_node_init(&p->avail_lists[i], 0); | |
1da177e4 | 2911 | p->flags = SWP_USED; |
5d337b91 | 2912 | spin_unlock(&swap_lock); |
ec8acf20 | 2913 | spin_lock_init(&p->lock); |
2628bd6f | 2914 | spin_lock_init(&p->cont_lock); |
efa90a98 | 2915 | |
53cbb243 | 2916 | return p; |
53cbb243 CEB |
2917 | } |
2918 | ||
4d0e1e10 CEB |
2919 | static int claim_swapfile(struct swap_info_struct *p, struct inode *inode) |
2920 | { | |
2921 | int error; | |
2922 | ||
2923 | if (S_ISBLK(inode->i_mode)) { | |
2924 | p->bdev = bdgrab(I_BDEV(inode)); | |
2925 | error = blkdev_get(p->bdev, | |
6f179af8 | 2926 | FMODE_READ | FMODE_WRITE | FMODE_EXCL, p); |
4d0e1e10 CEB |
2927 | if (error < 0) { |
2928 | p->bdev = NULL; | |
6f179af8 | 2929 | return error; |
4d0e1e10 CEB |
2930 | } |
2931 | p->old_block_size = block_size(p->bdev); | |
2932 | error = set_blocksize(p->bdev, PAGE_SIZE); | |
2933 | if (error < 0) | |
87ade72a | 2934 | return error; |
12d2966d NA |
2935 | /* |
2936 | * Zoned block devices contain zones that have a sequential | |
2937 | * write only restriction. Hence zoned block devices are not | |
2938 | * suitable for swapping. Disallow them here. | |
2939 | */ | |
2940 | if (blk_queue_is_zoned(p->bdev->bd_queue)) | |
2941 | return -EINVAL; | |
4d0e1e10 CEB |
2942 | p->flags |= SWP_BLKDEV; |
2943 | } else if (S_ISREG(inode->i_mode)) { | |
2944 | p->bdev = inode->i_sb->s_bdev; | |
1638045c DW |
2945 | } |
2946 | ||
4d0e1e10 | 2947 | return 0; |
4d0e1e10 CEB |
2948 | } |
2949 | ||
377eeaa8 AK |
2950 | |
2951 | /* | |
2952 | * Find out how many pages are allowed for a single swap device. There | |
2953 | * are two limiting factors: | |
2954 | * 1) the number of bits for the swap offset in the swp_entry_t type, and | |
2955 | * 2) the number of bits in the swap pte, as defined by the different | |
2956 | * architectures. | |
2957 | * | |
2958 | * In order to find the largest possible bit mask, a swap entry with | |
2959 | * swap type 0 and swap offset ~0UL is created, encoded to a swap pte, | |
2960 | * decoded to a swp_entry_t again, and finally the swap offset is | |
2961 | * extracted. | |
2962 | * | |
2963 | * This will mask all the bits from the initial ~0UL mask that can't | |
2964 | * be encoded in either the swp_entry_t or the architecture definition | |
2965 | * of a swap pte. | |
2966 | */ | |
2967 | unsigned long generic_max_swapfile_size(void) | |
2968 | { | |
2969 | return swp_offset(pte_to_swp_entry( | |
2970 | swp_entry_to_pte(swp_entry(0, ~0UL)))) + 1; | |
2971 | } | |
2972 | ||
2973 | /* Can be overridden by an architecture for additional checks. */ | |
2974 | __weak unsigned long max_swapfile_size(void) | |
2975 | { | |
2976 | return generic_max_swapfile_size(); | |
2977 | } | |
2978 | ||
ca8bd38b CEB |
2979 | static unsigned long read_swap_header(struct swap_info_struct *p, |
2980 | union swap_header *swap_header, | |
2981 | struct inode *inode) | |
2982 | { | |
2983 | int i; | |
2984 | unsigned long maxpages; | |
2985 | unsigned long swapfilepages; | |
d6bbbd29 | 2986 | unsigned long last_page; |
ca8bd38b CEB |
2987 | |
2988 | if (memcmp("SWAPSPACE2", swap_header->magic.magic, 10)) { | |
465c47fd | 2989 | pr_err("Unable to find swap-space signature\n"); |
38719025 | 2990 | return 0; |
ca8bd38b CEB |
2991 | } |
2992 | ||
2993 | /* swap partition endianess hack... */ | |
2994 | if (swab32(swap_header->info.version) == 1) { | |
2995 | swab32s(&swap_header->info.version); | |
2996 | swab32s(&swap_header->info.last_page); | |
2997 | swab32s(&swap_header->info.nr_badpages); | |
dd111be6 JH |
2998 | if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES) |
2999 | return 0; | |
ca8bd38b CEB |
3000 | for (i = 0; i < swap_header->info.nr_badpages; i++) |
3001 | swab32s(&swap_header->info.badpages[i]); | |
3002 | } | |
3003 | /* Check the swap header's sub-version */ | |
3004 | if (swap_header->info.version != 1) { | |
465c47fd AM |
3005 | pr_warn("Unable to handle swap header version %d\n", |
3006 | swap_header->info.version); | |
38719025 | 3007 | return 0; |
ca8bd38b CEB |
3008 | } |
3009 | ||
3010 | p->lowest_bit = 1; | |
3011 | p->cluster_next = 1; | |
3012 | p->cluster_nr = 0; | |
3013 | ||
377eeaa8 | 3014 | maxpages = max_swapfile_size(); |
d6bbbd29 | 3015 | last_page = swap_header->info.last_page; |
a06ad633 TA |
3016 | if (!last_page) { |
3017 | pr_warn("Empty swap-file\n"); | |
3018 | return 0; | |
3019 | } | |
d6bbbd29 | 3020 | if (last_page > maxpages) { |
465c47fd | 3021 | pr_warn("Truncating oversized swap area, only using %luk out of %luk\n", |
d6bbbd29 RJ |
3022 | maxpages << (PAGE_SHIFT - 10), |
3023 | last_page << (PAGE_SHIFT - 10)); | |
3024 | } | |
3025 | if (maxpages > last_page) { | |
3026 | maxpages = last_page + 1; | |
ca8bd38b CEB |
3027 | /* p->max is an unsigned int: don't overflow it */ |
3028 | if ((unsigned int)maxpages == 0) | |
3029 | maxpages = UINT_MAX; | |
3030 | } | |
3031 | p->highest_bit = maxpages - 1; | |
3032 | ||
3033 | if (!maxpages) | |
38719025 | 3034 | return 0; |
ca8bd38b CEB |
3035 | swapfilepages = i_size_read(inode) >> PAGE_SHIFT; |
3036 | if (swapfilepages && maxpages > swapfilepages) { | |
465c47fd | 3037 | pr_warn("Swap area shorter than signature indicates\n"); |
38719025 | 3038 | return 0; |
ca8bd38b CEB |
3039 | } |
3040 | if (swap_header->info.nr_badpages && S_ISREG(inode->i_mode)) | |
38719025 | 3041 | return 0; |
ca8bd38b | 3042 | if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES) |
38719025 | 3043 | return 0; |
ca8bd38b CEB |
3044 | |
3045 | return maxpages; | |
ca8bd38b CEB |
3046 | } |
3047 | ||
4b3ef9da | 3048 | #define SWAP_CLUSTER_INFO_COLS \ |
235b6217 | 3049 | DIV_ROUND_UP(L1_CACHE_BYTES, sizeof(struct swap_cluster_info)) |
4b3ef9da HY |
3050 | #define SWAP_CLUSTER_SPACE_COLS \ |
3051 | DIV_ROUND_UP(SWAP_ADDRESS_SPACE_PAGES, SWAPFILE_CLUSTER) | |
3052 | #define SWAP_CLUSTER_COLS \ | |
3053 | max_t(unsigned int, SWAP_CLUSTER_INFO_COLS, SWAP_CLUSTER_SPACE_COLS) | |
235b6217 | 3054 | |
915d4d7b CEB |
3055 | static int setup_swap_map_and_extents(struct swap_info_struct *p, |
3056 | union swap_header *swap_header, | |
3057 | unsigned char *swap_map, | |
2a8f9449 | 3058 | struct swap_cluster_info *cluster_info, |
915d4d7b CEB |
3059 | unsigned long maxpages, |
3060 | sector_t *span) | |
3061 | { | |
235b6217 | 3062 | unsigned int j, k; |
915d4d7b CEB |
3063 | unsigned int nr_good_pages; |
3064 | int nr_extents; | |
2a8f9449 | 3065 | unsigned long nr_clusters = DIV_ROUND_UP(maxpages, SWAPFILE_CLUSTER); |
235b6217 HY |
3066 | unsigned long col = p->cluster_next / SWAPFILE_CLUSTER % SWAP_CLUSTER_COLS; |
3067 | unsigned long i, idx; | |
915d4d7b CEB |
3068 | |
3069 | nr_good_pages = maxpages - 1; /* omit header page */ | |
3070 | ||
6b534915 HY |
3071 | cluster_list_init(&p->free_clusters); |
3072 | cluster_list_init(&p->discard_clusters); | |
2a8f9449 | 3073 | |
915d4d7b CEB |
3074 | for (i = 0; i < swap_header->info.nr_badpages; i++) { |
3075 | unsigned int page_nr = swap_header->info.badpages[i]; | |
bdb8e3f6 CEB |
3076 | if (page_nr == 0 || page_nr > swap_header->info.last_page) |
3077 | return -EINVAL; | |
915d4d7b CEB |
3078 | if (page_nr < maxpages) { |
3079 | swap_map[page_nr] = SWAP_MAP_BAD; | |
3080 | nr_good_pages--; | |
2a8f9449 SL |
3081 | /* |
3082 | * Haven't marked the cluster free yet, no list | |
3083 | * operation involved | |
3084 | */ | |
3085 | inc_cluster_info_page(p, cluster_info, page_nr); | |
915d4d7b CEB |
3086 | } |
3087 | } | |
3088 | ||
2a8f9449 SL |
3089 | /* Haven't marked the cluster free yet, no list operation involved */ |
3090 | for (i = maxpages; i < round_up(maxpages, SWAPFILE_CLUSTER); i++) | |
3091 | inc_cluster_info_page(p, cluster_info, i); | |
3092 | ||
915d4d7b CEB |
3093 | if (nr_good_pages) { |
3094 | swap_map[0] = SWAP_MAP_BAD; | |
2a8f9449 SL |
3095 | /* |
3096 | * Not mark the cluster free yet, no list | |
3097 | * operation involved | |
3098 | */ | |
3099 | inc_cluster_info_page(p, cluster_info, 0); | |
915d4d7b CEB |
3100 | p->max = maxpages; |
3101 | p->pages = nr_good_pages; | |
3102 | nr_extents = setup_swap_extents(p, span); | |
bdb8e3f6 CEB |
3103 | if (nr_extents < 0) |
3104 | return nr_extents; | |
915d4d7b CEB |
3105 | nr_good_pages = p->pages; |
3106 | } | |
3107 | if (!nr_good_pages) { | |
465c47fd | 3108 | pr_warn("Empty swap-file\n"); |
bdb8e3f6 | 3109 | return -EINVAL; |
915d4d7b CEB |
3110 | } |
3111 | ||
2a8f9449 SL |
3112 | if (!cluster_info) |
3113 | return nr_extents; | |
3114 | ||
235b6217 | 3115 | |
4b3ef9da HY |
3116 | /* |
3117 | * Reduce false cache line sharing between cluster_info and | |
3118 | * sharing same address space. | |
3119 | */ | |
235b6217 HY |
3120 | for (k = 0; k < SWAP_CLUSTER_COLS; k++) { |
3121 | j = (k + col) % SWAP_CLUSTER_COLS; | |
3122 | for (i = 0; i < DIV_ROUND_UP(nr_clusters, SWAP_CLUSTER_COLS); i++) { | |
3123 | idx = i * SWAP_CLUSTER_COLS + j; | |
3124 | if (idx >= nr_clusters) | |
3125 | continue; | |
3126 | if (cluster_count(&cluster_info[idx])) | |
3127 | continue; | |
2a8f9449 | 3128 | cluster_set_flag(&cluster_info[idx], CLUSTER_FLAG_FREE); |
6b534915 HY |
3129 | cluster_list_add_tail(&p->free_clusters, cluster_info, |
3130 | idx); | |
2a8f9449 | 3131 | } |
2a8f9449 | 3132 | } |
915d4d7b | 3133 | return nr_extents; |
915d4d7b CEB |
3134 | } |
3135 | ||
dcf6b7dd RA |
3136 | /* |
3137 | * Helper to sys_swapon determining if a given swap | |
3138 | * backing device queue supports DISCARD operations. | |
3139 | */ | |
3140 | static bool swap_discardable(struct swap_info_struct *si) | |
3141 | { | |
3142 | struct request_queue *q = bdev_get_queue(si->bdev); | |
3143 | ||
3144 | if (!q || !blk_queue_discard(q)) | |
3145 | return false; | |
3146 | ||
3147 | return true; | |
3148 | } | |
3149 | ||
53cbb243 CEB |
3150 | SYSCALL_DEFINE2(swapon, const char __user *, specialfile, int, swap_flags) |
3151 | { | |
3152 | struct swap_info_struct *p; | |
91a27b2a | 3153 | struct filename *name; |
53cbb243 CEB |
3154 | struct file *swap_file = NULL; |
3155 | struct address_space *mapping; | |
40531542 | 3156 | int prio; |
53cbb243 CEB |
3157 | int error; |
3158 | union swap_header *swap_header; | |
915d4d7b | 3159 | int nr_extents; |
53cbb243 CEB |
3160 | sector_t span; |
3161 | unsigned long maxpages; | |
53cbb243 | 3162 | unsigned char *swap_map = NULL; |
2a8f9449 | 3163 | struct swap_cluster_info *cluster_info = NULL; |
38b5faf4 | 3164 | unsigned long *frontswap_map = NULL; |
53cbb243 CEB |
3165 | struct page *page = NULL; |
3166 | struct inode *inode = NULL; | |
7cbf3192 | 3167 | bool inced_nr_rotate_swap = false; |
53cbb243 | 3168 | |
d15cab97 HD |
3169 | if (swap_flags & ~SWAP_FLAGS_VALID) |
3170 | return -EINVAL; | |
3171 | ||
53cbb243 CEB |
3172 | if (!capable(CAP_SYS_ADMIN)) |
3173 | return -EPERM; | |
3174 | ||
a2468cc9 AL |
3175 | if (!swap_avail_heads) |
3176 | return -ENOMEM; | |
3177 | ||
53cbb243 | 3178 | p = alloc_swap_info(); |
2542e513 CEB |
3179 | if (IS_ERR(p)) |
3180 | return PTR_ERR(p); | |
53cbb243 | 3181 | |
815c2c54 SL |
3182 | INIT_WORK(&p->discard_work, swap_discard_work); |
3183 | ||
1da177e4 | 3184 | name = getname(specialfile); |
1da177e4 | 3185 | if (IS_ERR(name)) { |
7de7fb6b | 3186 | error = PTR_ERR(name); |
1da177e4 | 3187 | name = NULL; |
bd69010b | 3188 | goto bad_swap; |
1da177e4 | 3189 | } |
669abf4e | 3190 | swap_file = file_open_name(name, O_RDWR|O_LARGEFILE, 0); |
1da177e4 | 3191 | if (IS_ERR(swap_file)) { |
7de7fb6b | 3192 | error = PTR_ERR(swap_file); |
1da177e4 | 3193 | swap_file = NULL; |
bd69010b | 3194 | goto bad_swap; |
1da177e4 LT |
3195 | } |
3196 | ||
3197 | p->swap_file = swap_file; | |
3198 | mapping = swap_file->f_mapping; | |
2130781e | 3199 | inode = mapping->host; |
6f179af8 | 3200 | |
4d0e1e10 CEB |
3201 | error = claim_swapfile(p, inode); |
3202 | if (unlikely(error)) | |
1da177e4 | 3203 | goto bad_swap; |
1da177e4 | 3204 | |
d795a90e NA |
3205 | inode_lock(inode); |
3206 | if (IS_SWAPFILE(inode)) { | |
3207 | error = -EBUSY; | |
3208 | goto bad_swap_unlock_inode; | |
3209 | } | |
3210 | ||
1da177e4 LT |
3211 | /* |
3212 | * Read the swap header. | |
3213 | */ | |
3214 | if (!mapping->a_ops->readpage) { | |
3215 | error = -EINVAL; | |
d795a90e | 3216 | goto bad_swap_unlock_inode; |
1da177e4 | 3217 | } |
090d2b18 | 3218 | page = read_mapping_page(mapping, 0, swap_file); |
1da177e4 LT |
3219 | if (IS_ERR(page)) { |
3220 | error = PTR_ERR(page); | |
d795a90e | 3221 | goto bad_swap_unlock_inode; |
1da177e4 | 3222 | } |
81e33971 | 3223 | swap_header = kmap(page); |
1da177e4 | 3224 | |
ca8bd38b CEB |
3225 | maxpages = read_swap_header(p, swap_header, inode); |
3226 | if (unlikely(!maxpages)) { | |
1da177e4 | 3227 | error = -EINVAL; |
d795a90e | 3228 | goto bad_swap_unlock_inode; |
1da177e4 | 3229 | } |
886bb7e9 | 3230 | |
81e33971 | 3231 | /* OK, set up the swap map and apply the bad block list */ |
803d0c83 | 3232 | swap_map = vzalloc(maxpages); |
81e33971 HD |
3233 | if (!swap_map) { |
3234 | error = -ENOMEM; | |
d795a90e | 3235 | goto bad_swap_unlock_inode; |
81e33971 | 3236 | } |
f0571429 MK |
3237 | |
3238 | if (bdi_cap_stable_pages_required(inode_to_bdi(inode))) | |
3239 | p->flags |= SWP_STABLE_WRITES; | |
3240 | ||
539a6fea MK |
3241 | if (bdi_cap_synchronous_io(inode_to_bdi(inode))) |
3242 | p->flags |= SWP_SYNCHRONOUS_IO; | |
3243 | ||
2a8f9449 | 3244 | if (p->bdev && blk_queue_nonrot(bdev_get_queue(p->bdev))) { |
6f179af8 | 3245 | int cpu; |
235b6217 | 3246 | unsigned long ci, nr_cluster; |
6f179af8 | 3247 | |
2a8f9449 | 3248 | p->flags |= SWP_SOLIDSTATE; |
49070588 HY |
3249 | p->cluster_next_cpu = alloc_percpu(unsigned int); |
3250 | if (!p->cluster_next_cpu) { | |
3251 | error = -ENOMEM; | |
3252 | goto bad_swap_unlock_inode; | |
3253 | } | |
2a8f9449 SL |
3254 | /* |
3255 | * select a random position to start with to help wear leveling | |
3256 | * SSD | |
3257 | */ | |
49070588 HY |
3258 | for_each_possible_cpu(cpu) { |
3259 | per_cpu(*p->cluster_next_cpu, cpu) = | |
3260 | 1 + prandom_u32_max(p->highest_bit); | |
3261 | } | |
235b6217 | 3262 | nr_cluster = DIV_ROUND_UP(maxpages, SWAPFILE_CLUSTER); |
2a8f9449 | 3263 | |
778e1cdd | 3264 | cluster_info = kvcalloc(nr_cluster, sizeof(*cluster_info), |
54f180d3 | 3265 | GFP_KERNEL); |
2a8f9449 SL |
3266 | if (!cluster_info) { |
3267 | error = -ENOMEM; | |
d795a90e | 3268 | goto bad_swap_unlock_inode; |
2a8f9449 | 3269 | } |
235b6217 HY |
3270 | |
3271 | for (ci = 0; ci < nr_cluster; ci++) | |
3272 | spin_lock_init(&((cluster_info + ci)->lock)); | |
3273 | ||
ebc2a1a6 SL |
3274 | p->percpu_cluster = alloc_percpu(struct percpu_cluster); |
3275 | if (!p->percpu_cluster) { | |
3276 | error = -ENOMEM; | |
d795a90e | 3277 | goto bad_swap_unlock_inode; |
ebc2a1a6 | 3278 | } |
6f179af8 | 3279 | for_each_possible_cpu(cpu) { |
ebc2a1a6 | 3280 | struct percpu_cluster *cluster; |
6f179af8 | 3281 | cluster = per_cpu_ptr(p->percpu_cluster, cpu); |
ebc2a1a6 SL |
3282 | cluster_set_null(&cluster->index); |
3283 | } | |
7cbf3192 | 3284 | } else { |
81a0298b | 3285 | atomic_inc(&nr_rotate_swap); |
7cbf3192 OS |
3286 | inced_nr_rotate_swap = true; |
3287 | } | |
1da177e4 | 3288 | |
1421ef3c CEB |
3289 | error = swap_cgroup_swapon(p->type, maxpages); |
3290 | if (error) | |
d795a90e | 3291 | goto bad_swap_unlock_inode; |
1421ef3c | 3292 | |
915d4d7b | 3293 | nr_extents = setup_swap_map_and_extents(p, swap_header, swap_map, |
2a8f9449 | 3294 | cluster_info, maxpages, &span); |
915d4d7b CEB |
3295 | if (unlikely(nr_extents < 0)) { |
3296 | error = nr_extents; | |
d795a90e | 3297 | goto bad_swap_unlock_inode; |
1da177e4 | 3298 | } |
38b5faf4 | 3299 | /* frontswap enabled? set up bit-per-page map for frontswap */ |
8ea1d2a1 | 3300 | if (IS_ENABLED(CONFIG_FRONTSWAP)) |
778e1cdd KC |
3301 | frontswap_map = kvcalloc(BITS_TO_LONGS(maxpages), |
3302 | sizeof(long), | |
54f180d3 | 3303 | GFP_KERNEL); |
1da177e4 | 3304 | |
2a8f9449 SL |
3305 | if (p->bdev &&(swap_flags & SWAP_FLAG_DISCARD) && swap_discardable(p)) { |
3306 | /* | |
3307 | * When discard is enabled for swap with no particular | |
3308 | * policy flagged, we set all swap discard flags here in | |
3309 | * order to sustain backward compatibility with older | |
3310 | * swapon(8) releases. | |
3311 | */ | |
3312 | p->flags |= (SWP_DISCARDABLE | SWP_AREA_DISCARD | | |
3313 | SWP_PAGE_DISCARD); | |
dcf6b7dd | 3314 | |
2a8f9449 SL |
3315 | /* |
3316 | * By flagging sys_swapon, a sysadmin can tell us to | |
3317 | * either do single-time area discards only, or to just | |
3318 | * perform discards for released swap page-clusters. | |
3319 | * Now it's time to adjust the p->flags accordingly. | |
3320 | */ | |
3321 | if (swap_flags & SWAP_FLAG_DISCARD_ONCE) | |
3322 | p->flags &= ~SWP_PAGE_DISCARD; | |
3323 | else if (swap_flags & SWAP_FLAG_DISCARD_PAGES) | |
3324 | p->flags &= ~SWP_AREA_DISCARD; | |
3325 | ||
3326 | /* issue a swapon-time discard if it's still required */ | |
3327 | if (p->flags & SWP_AREA_DISCARD) { | |
3328 | int err = discard_swap(p); | |
3329 | if (unlikely(err)) | |
3330 | pr_err("swapon: discard_swap(%p): %d\n", | |
3331 | p, err); | |
dcf6b7dd | 3332 | } |
20137a49 | 3333 | } |
6a6ba831 | 3334 | |
4b3ef9da HY |
3335 | error = init_swap_address_space(p->type, maxpages); |
3336 | if (error) | |
d795a90e | 3337 | goto bad_swap_unlock_inode; |
4b3ef9da | 3338 | |
dc617f29 DW |
3339 | /* |
3340 | * Flush any pending IO and dirty mappings before we start using this | |
3341 | * swap device. | |
3342 | */ | |
3343 | inode->i_flags |= S_SWAPFILE; | |
3344 | error = inode_drain_writes(inode); | |
3345 | if (error) { | |
3346 | inode->i_flags &= ~S_SWAPFILE; | |
d795a90e | 3347 | goto bad_swap_unlock_inode; |
dc617f29 DW |
3348 | } |
3349 | ||
fc0abb14 | 3350 | mutex_lock(&swapon_mutex); |
40531542 | 3351 | prio = -1; |
78ecba08 | 3352 | if (swap_flags & SWAP_FLAG_PREFER) |
40531542 | 3353 | prio = |
78ecba08 | 3354 | (swap_flags & SWAP_FLAG_PRIO_MASK) >> SWAP_FLAG_PRIO_SHIFT; |
2a8f9449 | 3355 | enable_swap_info(p, prio, swap_map, cluster_info, frontswap_map); |
c69dbfb8 | 3356 | |
756a025f | 3357 | pr_info("Adding %uk swap on %s. Priority:%d extents:%d across:%lluk %s%s%s%s%s\n", |
91a27b2a | 3358 | p->pages<<(PAGE_SHIFT-10), name->name, p->prio, |
c69dbfb8 CEB |
3359 | nr_extents, (unsigned long long)span<<(PAGE_SHIFT-10), |
3360 | (p->flags & SWP_SOLIDSTATE) ? "SS" : "", | |
38b5faf4 | 3361 | (p->flags & SWP_DISCARDABLE) ? "D" : "", |
dcf6b7dd RA |
3362 | (p->flags & SWP_AREA_DISCARD) ? "s" : "", |
3363 | (p->flags & SWP_PAGE_DISCARD) ? "c" : "", | |
38b5faf4 | 3364 | (frontswap_map) ? "FS" : ""); |
c69dbfb8 | 3365 | |
fc0abb14 | 3366 | mutex_unlock(&swapon_mutex); |
66d7dd51 KS |
3367 | atomic_inc(&proc_poll_event); |
3368 | wake_up_interruptible(&proc_poll_wait); | |
3369 | ||
1da177e4 LT |
3370 | error = 0; |
3371 | goto out; | |
d795a90e NA |
3372 | bad_swap_unlock_inode: |
3373 | inode_unlock(inode); | |
1da177e4 | 3374 | bad_swap: |
ebc2a1a6 SL |
3375 | free_percpu(p->percpu_cluster); |
3376 | p->percpu_cluster = NULL; | |
49070588 HY |
3377 | free_percpu(p->cluster_next_cpu); |
3378 | p->cluster_next_cpu = NULL; | |
bd69010b | 3379 | if (inode && S_ISBLK(inode->i_mode) && p->bdev) { |
f2090d2d CEB |
3380 | set_blocksize(p->bdev, p->old_block_size); |
3381 | blkdev_put(p->bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL); | |
1da177e4 | 3382 | } |
d795a90e | 3383 | inode = NULL; |
4cd3bb10 | 3384 | destroy_swap_extents(p); |
e8e6c2ec | 3385 | swap_cgroup_swapoff(p->type); |
5d337b91 | 3386 | spin_lock(&swap_lock); |
1da177e4 | 3387 | p->swap_file = NULL; |
1da177e4 | 3388 | p->flags = 0; |
5d337b91 | 3389 | spin_unlock(&swap_lock); |
1da177e4 | 3390 | vfree(swap_map); |
8606a1a9 | 3391 | kvfree(cluster_info); |
b6b1fd2a | 3392 | kvfree(frontswap_map); |
7cbf3192 OS |
3393 | if (inced_nr_rotate_swap) |
3394 | atomic_dec(&nr_rotate_swap); | |
d795a90e | 3395 | if (swap_file) |
1da177e4 LT |
3396 | filp_close(swap_file, NULL); |
3397 | out: | |
3398 | if (page && !IS_ERR(page)) { | |
3399 | kunmap(page); | |
09cbfeaf | 3400 | put_page(page); |
1da177e4 LT |
3401 | } |
3402 | if (name) | |
3403 | putname(name); | |
1638045c | 3404 | if (inode) |
5955102c | 3405 | inode_unlock(inode); |
039939a6 TC |
3406 | if (!error) |
3407 | enable_swap_slots_cache(); | |
1da177e4 LT |
3408 | return error; |
3409 | } | |
3410 | ||
3411 | void si_swapinfo(struct sysinfo *val) | |
3412 | { | |
efa90a98 | 3413 | unsigned int type; |
1da177e4 LT |
3414 | unsigned long nr_to_be_unused = 0; |
3415 | ||
5d337b91 | 3416 | spin_lock(&swap_lock); |
efa90a98 HD |
3417 | for (type = 0; type < nr_swapfiles; type++) { |
3418 | struct swap_info_struct *si = swap_info[type]; | |
3419 | ||
3420 | if ((si->flags & SWP_USED) && !(si->flags & SWP_WRITEOK)) | |
3421 | nr_to_be_unused += si->inuse_pages; | |
1da177e4 | 3422 | } |
ec8acf20 | 3423 | val->freeswap = atomic_long_read(&nr_swap_pages) + nr_to_be_unused; |
1da177e4 | 3424 | val->totalswap = total_swap_pages + nr_to_be_unused; |
5d337b91 | 3425 | spin_unlock(&swap_lock); |
1da177e4 LT |
3426 | } |
3427 | ||
3428 | /* | |
3429 | * Verify that a swap entry is valid and increment its swap map count. | |
3430 | * | |
355cfa73 KH |
3431 | * Returns error code in following case. |
3432 | * - success -> 0 | |
3433 | * - swp_entry is invalid -> EINVAL | |
3434 | * - swp_entry is migration entry -> EINVAL | |
3435 | * - swap-cache reference is requested but there is already one. -> EEXIST | |
3436 | * - swap-cache reference is requested but the entry is not used. -> ENOENT | |
570a335b | 3437 | * - swap-mapped reference requested but needs continued swap count. -> ENOMEM |
1da177e4 | 3438 | */ |
8d69aaee | 3439 | static int __swap_duplicate(swp_entry_t entry, unsigned char usage) |
1da177e4 | 3440 | { |
73c34b6a | 3441 | struct swap_info_struct *p; |
235b6217 | 3442 | struct swap_cluster_info *ci; |
c10d38cc | 3443 | unsigned long offset; |
8d69aaee HD |
3444 | unsigned char count; |
3445 | unsigned char has_cache; | |
253d553b | 3446 | int err = -EINVAL; |
1da177e4 | 3447 | |
eb085574 | 3448 | p = get_swap_device(entry); |
c10d38cc | 3449 | if (!p) |
235b6217 HY |
3450 | goto out; |
3451 | ||
eb085574 | 3452 | offset = swp_offset(entry); |
235b6217 | 3453 | ci = lock_cluster_or_swap_info(p, offset); |
355cfa73 | 3454 | |
253d553b | 3455 | count = p->swap_map[offset]; |
edfe23da SL |
3456 | |
3457 | /* | |
3458 | * swapin_readahead() doesn't check if a swap entry is valid, so the | |
3459 | * swap entry could be SWAP_MAP_BAD. Check here with lock held. | |
3460 | */ | |
3461 | if (unlikely(swap_count(count) == SWAP_MAP_BAD)) { | |
3462 | err = -ENOENT; | |
3463 | goto unlock_out; | |
3464 | } | |
3465 | ||
253d553b HD |
3466 | has_cache = count & SWAP_HAS_CACHE; |
3467 | count &= ~SWAP_HAS_CACHE; | |
3468 | err = 0; | |
355cfa73 | 3469 | |
253d553b | 3470 | if (usage == SWAP_HAS_CACHE) { |
355cfa73 KH |
3471 | |
3472 | /* set SWAP_HAS_CACHE if there is no cache and entry is used */ | |
253d553b HD |
3473 | if (!has_cache && count) |
3474 | has_cache = SWAP_HAS_CACHE; | |
3475 | else if (has_cache) /* someone else added cache */ | |
3476 | err = -EEXIST; | |
3477 | else /* no users remaining */ | |
3478 | err = -ENOENT; | |
355cfa73 KH |
3479 | |
3480 | } else if (count || has_cache) { | |
253d553b | 3481 | |
570a335b HD |
3482 | if ((count & ~COUNT_CONTINUED) < SWAP_MAP_MAX) |
3483 | count += usage; | |
3484 | else if ((count & ~COUNT_CONTINUED) > SWAP_MAP_MAX) | |
253d553b | 3485 | err = -EINVAL; |
570a335b HD |
3486 | else if (swap_count_continued(p, offset, count)) |
3487 | count = COUNT_CONTINUED; | |
3488 | else | |
3489 | err = -ENOMEM; | |
355cfa73 | 3490 | } else |
253d553b HD |
3491 | err = -ENOENT; /* unused swap entry */ |
3492 | ||
3493 | p->swap_map[offset] = count | has_cache; | |
3494 | ||
355cfa73 | 3495 | unlock_out: |
235b6217 | 3496 | unlock_cluster_or_swap_info(p, ci); |
1da177e4 | 3497 | out: |
eb085574 HY |
3498 | if (p) |
3499 | put_swap_device(p); | |
253d553b | 3500 | return err; |
1da177e4 | 3501 | } |
253d553b | 3502 | |
aaa46865 HD |
3503 | /* |
3504 | * Help swapoff by noting that swap entry belongs to shmem/tmpfs | |
3505 | * (in which case its reference count is never incremented). | |
3506 | */ | |
3507 | void swap_shmem_alloc(swp_entry_t entry) | |
3508 | { | |
3509 | __swap_duplicate(entry, SWAP_MAP_SHMEM); | |
3510 | } | |
3511 | ||
355cfa73 | 3512 | /* |
08259d58 HD |
3513 | * Increase reference count of swap entry by 1. |
3514 | * Returns 0 for success, or -ENOMEM if a swap_count_continuation is required | |
3515 | * but could not be atomically allocated. Returns 0, just as if it succeeded, | |
3516 | * if __swap_duplicate() fails for another reason (-EINVAL or -ENOENT), which | |
3517 | * might occur if a page table entry has got corrupted. | |
355cfa73 | 3518 | */ |
570a335b | 3519 | int swap_duplicate(swp_entry_t entry) |
355cfa73 | 3520 | { |
570a335b HD |
3521 | int err = 0; |
3522 | ||
3523 | while (!err && __swap_duplicate(entry, 1) == -ENOMEM) | |
3524 | err = add_swap_count_continuation(entry, GFP_ATOMIC); | |
3525 | return err; | |
355cfa73 | 3526 | } |
1da177e4 | 3527 | |
cb4b86ba | 3528 | /* |
355cfa73 KH |
3529 | * @entry: swap entry for which we allocate swap cache. |
3530 | * | |
73c34b6a | 3531 | * Called when allocating swap cache for existing swap entry, |
355cfa73 | 3532 | * This can return error codes. Returns 0 at success. |
3eeba135 | 3533 | * -EEXIST means there is a swap cache. |
355cfa73 | 3534 | * Note: return code is different from swap_duplicate(). |
cb4b86ba KH |
3535 | */ |
3536 | int swapcache_prepare(swp_entry_t entry) | |
3537 | { | |
253d553b | 3538 | return __swap_duplicate(entry, SWAP_HAS_CACHE); |
cb4b86ba KH |
3539 | } |
3540 | ||
0bcac06f MK |
3541 | struct swap_info_struct *swp_swap_info(swp_entry_t entry) |
3542 | { | |
c10d38cc | 3543 | return swap_type_to_swap_info(swp_type(entry)); |
0bcac06f MK |
3544 | } |
3545 | ||
f981c595 MG |
3546 | struct swap_info_struct *page_swap_info(struct page *page) |
3547 | { | |
0bcac06f MK |
3548 | swp_entry_t entry = { .val = page_private(page) }; |
3549 | return swp_swap_info(entry); | |
f981c595 MG |
3550 | } |
3551 | ||
3552 | /* | |
3553 | * out-of-line __page_file_ methods to avoid include hell. | |
3554 | */ | |
3555 | struct address_space *__page_file_mapping(struct page *page) | |
3556 | { | |
f981c595 MG |
3557 | return page_swap_info(page)->swap_file->f_mapping; |
3558 | } | |
3559 | EXPORT_SYMBOL_GPL(__page_file_mapping); | |
3560 | ||
3561 | pgoff_t __page_file_index(struct page *page) | |
3562 | { | |
3563 | swp_entry_t swap = { .val = page_private(page) }; | |
f981c595 MG |
3564 | return swp_offset(swap); |
3565 | } | |
3566 | EXPORT_SYMBOL_GPL(__page_file_index); | |
3567 | ||
570a335b HD |
3568 | /* |
3569 | * add_swap_count_continuation - called when a swap count is duplicated | |
3570 | * beyond SWAP_MAP_MAX, it allocates a new page and links that to the entry's | |
3571 | * page of the original vmalloc'ed swap_map, to hold the continuation count | |
3572 | * (for that entry and for its neighbouring PAGE_SIZE swap entries). Called | |
3573 | * again when count is duplicated beyond SWAP_MAP_MAX * SWAP_CONT_MAX, etc. | |
3574 | * | |
3575 | * These continuation pages are seldom referenced: the common paths all work | |
3576 | * on the original swap_map, only referring to a continuation page when the | |
3577 | * low "digit" of a count is incremented or decremented through SWAP_MAP_MAX. | |
3578 | * | |
3579 | * add_swap_count_continuation(, GFP_ATOMIC) can be called while holding | |
3580 | * page table locks; if it fails, add_swap_count_continuation(, GFP_KERNEL) | |
3581 | * can be called after dropping locks. | |
3582 | */ | |
3583 | int add_swap_count_continuation(swp_entry_t entry, gfp_t gfp_mask) | |
3584 | { | |
3585 | struct swap_info_struct *si; | |
235b6217 | 3586 | struct swap_cluster_info *ci; |
570a335b HD |
3587 | struct page *head; |
3588 | struct page *page; | |
3589 | struct page *list_page; | |
3590 | pgoff_t offset; | |
3591 | unsigned char count; | |
eb085574 | 3592 | int ret = 0; |
570a335b HD |
3593 | |
3594 | /* | |
3595 | * When debugging, it's easier to use __GFP_ZERO here; but it's better | |
3596 | * for latency not to zero a page while GFP_ATOMIC and holding locks. | |
3597 | */ | |
3598 | page = alloc_page(gfp_mask | __GFP_HIGHMEM); | |
3599 | ||
eb085574 | 3600 | si = get_swap_device(entry); |
570a335b HD |
3601 | if (!si) { |
3602 | /* | |
3603 | * An acceptable race has occurred since the failing | |
eb085574 | 3604 | * __swap_duplicate(): the swap device may be swapoff |
570a335b HD |
3605 | */ |
3606 | goto outer; | |
3607 | } | |
eb085574 | 3608 | spin_lock(&si->lock); |
570a335b HD |
3609 | |
3610 | offset = swp_offset(entry); | |
235b6217 HY |
3611 | |
3612 | ci = lock_cluster(si, offset); | |
3613 | ||
570a335b HD |
3614 | count = si->swap_map[offset] & ~SWAP_HAS_CACHE; |
3615 | ||
3616 | if ((count & ~COUNT_CONTINUED) != SWAP_MAP_MAX) { | |
3617 | /* | |
3618 | * The higher the swap count, the more likely it is that tasks | |
3619 | * will race to add swap count continuation: we need to avoid | |
3620 | * over-provisioning. | |
3621 | */ | |
3622 | goto out; | |
3623 | } | |
3624 | ||
3625 | if (!page) { | |
eb085574 HY |
3626 | ret = -ENOMEM; |
3627 | goto out; | |
570a335b HD |
3628 | } |
3629 | ||
3630 | /* | |
3631 | * We are fortunate that although vmalloc_to_page uses pte_offset_map, | |
2de1a7e4 SJ |
3632 | * no architecture is using highmem pages for kernel page tables: so it |
3633 | * will not corrupt the GFP_ATOMIC caller's atomic page table kmaps. | |
570a335b HD |
3634 | */ |
3635 | head = vmalloc_to_page(si->swap_map + offset); | |
3636 | offset &= ~PAGE_MASK; | |
3637 | ||
2628bd6f | 3638 | spin_lock(&si->cont_lock); |
570a335b HD |
3639 | /* |
3640 | * Page allocation does not initialize the page's lru field, | |
3641 | * but it does always reset its private field. | |
3642 | */ | |
3643 | if (!page_private(head)) { | |
3644 | BUG_ON(count & COUNT_CONTINUED); | |
3645 | INIT_LIST_HEAD(&head->lru); | |
3646 | set_page_private(head, SWP_CONTINUED); | |
3647 | si->flags |= SWP_CONTINUED; | |
3648 | } | |
3649 | ||
3650 | list_for_each_entry(list_page, &head->lru, lru) { | |
3651 | unsigned char *map; | |
3652 | ||
3653 | /* | |
3654 | * If the previous map said no continuation, but we've found | |
3655 | * a continuation page, free our allocation and use this one. | |
3656 | */ | |
3657 | if (!(count & COUNT_CONTINUED)) | |
2628bd6f | 3658 | goto out_unlock_cont; |
570a335b | 3659 | |
9b04c5fe | 3660 | map = kmap_atomic(list_page) + offset; |
570a335b | 3661 | count = *map; |
9b04c5fe | 3662 | kunmap_atomic(map); |
570a335b HD |
3663 | |
3664 | /* | |
3665 | * If this continuation count now has some space in it, | |
3666 | * free our allocation and use this one. | |
3667 | */ | |
3668 | if ((count & ~COUNT_CONTINUED) != SWAP_CONT_MAX) | |
2628bd6f | 3669 | goto out_unlock_cont; |
570a335b HD |
3670 | } |
3671 | ||
3672 | list_add_tail(&page->lru, &head->lru); | |
3673 | page = NULL; /* now it's attached, don't free it */ | |
2628bd6f HY |
3674 | out_unlock_cont: |
3675 | spin_unlock(&si->cont_lock); | |
570a335b | 3676 | out: |
235b6217 | 3677 | unlock_cluster(ci); |
ec8acf20 | 3678 | spin_unlock(&si->lock); |
eb085574 | 3679 | put_swap_device(si); |
570a335b HD |
3680 | outer: |
3681 | if (page) | |
3682 | __free_page(page); | |
eb085574 | 3683 | return ret; |
570a335b HD |
3684 | } |
3685 | ||
3686 | /* | |
3687 | * swap_count_continued - when the original swap_map count is incremented | |
3688 | * from SWAP_MAP_MAX, check if there is already a continuation page to carry | |
3689 | * into, carry if so, or else fail until a new continuation page is allocated; | |
3690 | * when the original swap_map count is decremented from 0 with continuation, | |
3691 | * borrow from the continuation and report whether it still holds more. | |
235b6217 HY |
3692 | * Called while __swap_duplicate() or swap_entry_free() holds swap or cluster |
3693 | * lock. | |
570a335b HD |
3694 | */ |
3695 | static bool swap_count_continued(struct swap_info_struct *si, | |
3696 | pgoff_t offset, unsigned char count) | |
3697 | { | |
3698 | struct page *head; | |
3699 | struct page *page; | |
3700 | unsigned char *map; | |
2628bd6f | 3701 | bool ret; |
570a335b HD |
3702 | |
3703 | head = vmalloc_to_page(si->swap_map + offset); | |
3704 | if (page_private(head) != SWP_CONTINUED) { | |
3705 | BUG_ON(count & COUNT_CONTINUED); | |
3706 | return false; /* need to add count continuation */ | |
3707 | } | |
3708 | ||
2628bd6f | 3709 | spin_lock(&si->cont_lock); |
570a335b | 3710 | offset &= ~PAGE_MASK; |
213516ac | 3711 | page = list_next_entry(head, lru); |
9b04c5fe | 3712 | map = kmap_atomic(page) + offset; |
570a335b HD |
3713 | |
3714 | if (count == SWAP_MAP_MAX) /* initial increment from swap_map */ | |
3715 | goto init_map; /* jump over SWAP_CONT_MAX checks */ | |
3716 | ||
3717 | if (count == (SWAP_MAP_MAX | COUNT_CONTINUED)) { /* incrementing */ | |
3718 | /* | |
3719 | * Think of how you add 1 to 999 | |
3720 | */ | |
3721 | while (*map == (SWAP_CONT_MAX | COUNT_CONTINUED)) { | |
9b04c5fe | 3722 | kunmap_atomic(map); |
213516ac | 3723 | page = list_next_entry(page, lru); |
570a335b | 3724 | BUG_ON(page == head); |
9b04c5fe | 3725 | map = kmap_atomic(page) + offset; |
570a335b HD |
3726 | } |
3727 | if (*map == SWAP_CONT_MAX) { | |
9b04c5fe | 3728 | kunmap_atomic(map); |
213516ac | 3729 | page = list_next_entry(page, lru); |
2628bd6f HY |
3730 | if (page == head) { |
3731 | ret = false; /* add count continuation */ | |
3732 | goto out; | |
3733 | } | |
9b04c5fe | 3734 | map = kmap_atomic(page) + offset; |
570a335b HD |
3735 | init_map: *map = 0; /* we didn't zero the page */ |
3736 | } | |
3737 | *map += 1; | |
9b04c5fe | 3738 | kunmap_atomic(map); |
213516ac | 3739 | while ((page = list_prev_entry(page, lru)) != head) { |
9b04c5fe | 3740 | map = kmap_atomic(page) + offset; |
570a335b | 3741 | *map = COUNT_CONTINUED; |
9b04c5fe | 3742 | kunmap_atomic(map); |
570a335b | 3743 | } |
2628bd6f | 3744 | ret = true; /* incremented */ |
570a335b HD |
3745 | |
3746 | } else { /* decrementing */ | |
3747 | /* | |
3748 | * Think of how you subtract 1 from 1000 | |
3749 | */ | |
3750 | BUG_ON(count != COUNT_CONTINUED); | |
3751 | while (*map == COUNT_CONTINUED) { | |
9b04c5fe | 3752 | kunmap_atomic(map); |
213516ac | 3753 | page = list_next_entry(page, lru); |
570a335b | 3754 | BUG_ON(page == head); |
9b04c5fe | 3755 | map = kmap_atomic(page) + offset; |
570a335b HD |
3756 | } |
3757 | BUG_ON(*map == 0); | |
3758 | *map -= 1; | |
3759 | if (*map == 0) | |
3760 | count = 0; | |
9b04c5fe | 3761 | kunmap_atomic(map); |
213516ac | 3762 | while ((page = list_prev_entry(page, lru)) != head) { |
9b04c5fe | 3763 | map = kmap_atomic(page) + offset; |
570a335b HD |
3764 | *map = SWAP_CONT_MAX | count; |
3765 | count = COUNT_CONTINUED; | |
9b04c5fe | 3766 | kunmap_atomic(map); |
570a335b | 3767 | } |
2628bd6f | 3768 | ret = count == COUNT_CONTINUED; |
570a335b | 3769 | } |
2628bd6f HY |
3770 | out: |
3771 | spin_unlock(&si->cont_lock); | |
3772 | return ret; | |
570a335b HD |
3773 | } |
3774 | ||
3775 | /* | |
3776 | * free_swap_count_continuations - swapoff free all the continuation pages | |
3777 | * appended to the swap_map, after swap_map is quiesced, before vfree'ing it. | |
3778 | */ | |
3779 | static void free_swap_count_continuations(struct swap_info_struct *si) | |
3780 | { | |
3781 | pgoff_t offset; | |
3782 | ||
3783 | for (offset = 0; offset < si->max; offset += PAGE_SIZE) { | |
3784 | struct page *head; | |
3785 | head = vmalloc_to_page(si->swap_map + offset); | |
3786 | if (page_private(head)) { | |
0d576d20 GT |
3787 | struct page *page, *next; |
3788 | ||
3789 | list_for_each_entry_safe(page, next, &head->lru, lru) { | |
3790 | list_del(&page->lru); | |
570a335b HD |
3791 | __free_page(page); |
3792 | } | |
3793 | } | |
3794 | } | |
3795 | } | |
a2468cc9 | 3796 | |
2cf85583 TH |
3797 | #if defined(CONFIG_MEMCG) && defined(CONFIG_BLK_CGROUP) |
3798 | void mem_cgroup_throttle_swaprate(struct mem_cgroup *memcg, int node, | |
3799 | gfp_t gfp_mask) | |
3800 | { | |
3801 | struct swap_info_struct *si, *next; | |
3802 | if (!(gfp_mask & __GFP_IO) || !memcg) | |
3803 | return; | |
3804 | ||
3805 | if (!blk_cgroup_congested()) | |
3806 | return; | |
3807 | ||
3808 | /* | |
3809 | * We've already scheduled a throttle, avoid taking the global swap | |
3810 | * lock. | |
3811 | */ | |
3812 | if (current->throttle_queue) | |
3813 | return; | |
3814 | ||
3815 | spin_lock(&swap_avail_lock); | |
3816 | plist_for_each_entry_safe(si, next, &swap_avail_heads[node], | |
3817 | avail_lists[node]) { | |
3818 | if (si->bdev) { | |
3819 | blkcg_schedule_throttle(bdev_get_queue(si->bdev), | |
3820 | true); | |
3821 | break; | |
3822 | } | |
3823 | } | |
3824 | spin_unlock(&swap_avail_lock); | |
3825 | } | |
3826 | #endif | |
3827 | ||
a2468cc9 AL |
3828 | static int __init swapfile_init(void) |
3829 | { | |
3830 | int nid; | |
3831 | ||
3832 | swap_avail_heads = kmalloc_array(nr_node_ids, sizeof(struct plist_head), | |
3833 | GFP_KERNEL); | |
3834 | if (!swap_avail_heads) { | |
3835 | pr_emerg("Not enough memory for swap heads, swap is disabled\n"); | |
3836 | return -ENOMEM; | |
3837 | } | |
3838 | ||
3839 | for_each_node(nid) | |
3840 | plist_head_init(&swap_avail_heads[nid]); | |
3841 | ||
3842 | return 0; | |
3843 | } | |
3844 | subsys_initcall(swapfile_init); |