]>
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; |
ebc2a1a6 | 604 | bool found_free; |
235b6217 | 605 | unsigned long tmp, max; |
ebc2a1a6 SL |
606 | |
607 | new_cluster: | |
608 | cluster = this_cpu_ptr(si->percpu_cluster); | |
609 | if (cluster_is_null(&cluster->index)) { | |
6b534915 HY |
610 | if (!cluster_list_empty(&si->free_clusters)) { |
611 | cluster->index = si->free_clusters.head; | |
ebc2a1a6 SL |
612 | cluster->next = cluster_next(&cluster->index) * |
613 | SWAPFILE_CLUSTER; | |
6b534915 | 614 | } else if (!cluster_list_empty(&si->discard_clusters)) { |
ebc2a1a6 SL |
615 | /* |
616 | * we don't have free cluster but have some clusters in | |
617 | * discarding, do discard now and reclaim them | |
618 | */ | |
619 | swap_do_scheduled_discard(si); | |
620 | *scan_base = *offset = si->cluster_next; | |
621 | goto new_cluster; | |
622 | } else | |
36005bae | 623 | return false; |
ebc2a1a6 SL |
624 | } |
625 | ||
626 | found_free = false; | |
627 | ||
628 | /* | |
629 | * Other CPUs can use our cluster if they can't find a free cluster, | |
630 | * check if there is still free entry in the cluster | |
631 | */ | |
632 | tmp = cluster->next; | |
235b6217 HY |
633 | max = min_t(unsigned long, si->max, |
634 | (cluster_next(&cluster->index) + 1) * SWAPFILE_CLUSTER); | |
635 | if (tmp >= max) { | |
636 | cluster_set_null(&cluster->index); | |
637 | goto new_cluster; | |
638 | } | |
639 | ci = lock_cluster(si, tmp); | |
640 | while (tmp < max) { | |
ebc2a1a6 SL |
641 | if (!si->swap_map[tmp]) { |
642 | found_free = true; | |
643 | break; | |
644 | } | |
645 | tmp++; | |
646 | } | |
235b6217 | 647 | unlock_cluster(ci); |
ebc2a1a6 SL |
648 | if (!found_free) { |
649 | cluster_set_null(&cluster->index); | |
650 | goto new_cluster; | |
651 | } | |
652 | cluster->next = tmp + 1; | |
653 | *offset = tmp; | |
654 | *scan_base = tmp; | |
36005bae | 655 | return found_free; |
2a8f9449 SL |
656 | } |
657 | ||
a2468cc9 AL |
658 | static void __del_from_avail_list(struct swap_info_struct *p) |
659 | { | |
660 | int nid; | |
661 | ||
662 | for_each_node(nid) | |
663 | plist_del(&p->avail_lists[nid], &swap_avail_heads[nid]); | |
664 | } | |
665 | ||
666 | static void del_from_avail_list(struct swap_info_struct *p) | |
667 | { | |
668 | spin_lock(&swap_avail_lock); | |
669 | __del_from_avail_list(p); | |
670 | spin_unlock(&swap_avail_lock); | |
671 | } | |
672 | ||
38d8b4e6 HY |
673 | static void swap_range_alloc(struct swap_info_struct *si, unsigned long offset, |
674 | unsigned int nr_entries) | |
675 | { | |
676 | unsigned int end = offset + nr_entries - 1; | |
677 | ||
678 | if (offset == si->lowest_bit) | |
679 | si->lowest_bit += nr_entries; | |
680 | if (end == si->highest_bit) | |
681 | si->highest_bit -= nr_entries; | |
682 | si->inuse_pages += nr_entries; | |
683 | if (si->inuse_pages == si->pages) { | |
684 | si->lowest_bit = si->max; | |
685 | si->highest_bit = 0; | |
a2468cc9 | 686 | del_from_avail_list(si); |
38d8b4e6 HY |
687 | } |
688 | } | |
689 | ||
a2468cc9 AL |
690 | static void add_to_avail_list(struct swap_info_struct *p) |
691 | { | |
692 | int nid; | |
693 | ||
694 | spin_lock(&swap_avail_lock); | |
695 | for_each_node(nid) { | |
696 | WARN_ON(!plist_node_empty(&p->avail_lists[nid])); | |
697 | plist_add(&p->avail_lists[nid], &swap_avail_heads[nid]); | |
698 | } | |
699 | spin_unlock(&swap_avail_lock); | |
700 | } | |
701 | ||
38d8b4e6 HY |
702 | static void swap_range_free(struct swap_info_struct *si, unsigned long offset, |
703 | unsigned int nr_entries) | |
704 | { | |
705 | unsigned long end = offset + nr_entries - 1; | |
706 | void (*swap_slot_free_notify)(struct block_device *, unsigned long); | |
707 | ||
708 | if (offset < si->lowest_bit) | |
709 | si->lowest_bit = offset; | |
710 | if (end > si->highest_bit) { | |
711 | bool was_full = !si->highest_bit; | |
712 | ||
713 | si->highest_bit = end; | |
a2468cc9 AL |
714 | if (was_full && (si->flags & SWP_WRITEOK)) |
715 | add_to_avail_list(si); | |
38d8b4e6 HY |
716 | } |
717 | atomic_long_add(nr_entries, &nr_swap_pages); | |
718 | si->inuse_pages -= nr_entries; | |
719 | if (si->flags & SWP_BLKDEV) | |
720 | swap_slot_free_notify = | |
721 | si->bdev->bd_disk->fops->swap_slot_free_notify; | |
722 | else | |
723 | swap_slot_free_notify = NULL; | |
724 | while (offset <= end) { | |
725 | frontswap_invalidate_page(si->type, offset); | |
726 | if (swap_slot_free_notify) | |
727 | swap_slot_free_notify(si->bdev, offset); | |
728 | offset++; | |
729 | } | |
730 | } | |
731 | ||
36005bae TC |
732 | static int scan_swap_map_slots(struct swap_info_struct *si, |
733 | unsigned char usage, int nr, | |
734 | swp_entry_t slots[]) | |
1da177e4 | 735 | { |
235b6217 | 736 | struct swap_cluster_info *ci; |
ebebbbe9 | 737 | unsigned long offset; |
c60aa176 | 738 | unsigned long scan_base; |
7992fde7 | 739 | unsigned long last_in_cluster = 0; |
048c27fd | 740 | int latency_ration = LATENCY_LIMIT; |
36005bae TC |
741 | int n_ret = 0; |
742 | ||
743 | if (nr > SWAP_BATCH) | |
744 | nr = SWAP_BATCH; | |
7dfad418 | 745 | |
886bb7e9 | 746 | /* |
7dfad418 HD |
747 | * We try to cluster swap pages by allocating them sequentially |
748 | * in swap. Once we've allocated SWAPFILE_CLUSTER pages this | |
749 | * way, however, we resort to first-free allocation, starting | |
750 | * a new cluster. This prevents us from scattering swap pages | |
751 | * all over the entire swap partition, so that we reduce | |
752 | * overall disk seek times between swap pages. -- sct | |
753 | * But we do now try to find an empty cluster. -Andrea | |
c60aa176 | 754 | * And we let swap pages go all over an SSD partition. Hugh |
7dfad418 HD |
755 | */ |
756 | ||
52b7efdb | 757 | si->flags += SWP_SCANNING; |
c60aa176 | 758 | scan_base = offset = si->cluster_next; |
ebebbbe9 | 759 | |
ebc2a1a6 SL |
760 | /* SSD algorithm */ |
761 | if (si->cluster_info) { | |
36005bae TC |
762 | if (scan_swap_map_try_ssd_cluster(si, &offset, &scan_base)) |
763 | goto checks; | |
764 | else | |
765 | goto scan; | |
ebc2a1a6 SL |
766 | } |
767 | ||
ebebbbe9 HD |
768 | if (unlikely(!si->cluster_nr--)) { |
769 | if (si->pages - si->inuse_pages < SWAPFILE_CLUSTER) { | |
770 | si->cluster_nr = SWAPFILE_CLUSTER - 1; | |
771 | goto checks; | |
772 | } | |
2a8f9449 | 773 | |
ec8acf20 | 774 | spin_unlock(&si->lock); |
7dfad418 | 775 | |
c60aa176 HD |
776 | /* |
777 | * If seek is expensive, start searching for new cluster from | |
778 | * start of partition, to minimize the span of allocated swap. | |
50088c44 CY |
779 | * If seek is cheap, that is the SWP_SOLIDSTATE si->cluster_info |
780 | * case, just handled by scan_swap_map_try_ssd_cluster() above. | |
c60aa176 | 781 | */ |
50088c44 | 782 | scan_base = offset = si->lowest_bit; |
7dfad418 HD |
783 | last_in_cluster = offset + SWAPFILE_CLUSTER - 1; |
784 | ||
785 | /* Locate the first empty (unaligned) cluster */ | |
786 | for (; last_in_cluster <= si->highest_bit; offset++) { | |
1da177e4 | 787 | if (si->swap_map[offset]) |
7dfad418 HD |
788 | last_in_cluster = offset + SWAPFILE_CLUSTER; |
789 | else if (offset == last_in_cluster) { | |
ec8acf20 | 790 | spin_lock(&si->lock); |
ebebbbe9 HD |
791 | offset -= SWAPFILE_CLUSTER - 1; |
792 | si->cluster_next = offset; | |
793 | si->cluster_nr = SWAPFILE_CLUSTER - 1; | |
c60aa176 HD |
794 | goto checks; |
795 | } | |
796 | if (unlikely(--latency_ration < 0)) { | |
797 | cond_resched(); | |
798 | latency_ration = LATENCY_LIMIT; | |
799 | } | |
800 | } | |
801 | ||
802 | offset = scan_base; | |
ec8acf20 | 803 | spin_lock(&si->lock); |
ebebbbe9 | 804 | si->cluster_nr = SWAPFILE_CLUSTER - 1; |
1da177e4 | 805 | } |
7dfad418 | 806 | |
ebebbbe9 | 807 | checks: |
ebc2a1a6 | 808 | if (si->cluster_info) { |
36005bae TC |
809 | while (scan_swap_map_ssd_cluster_conflict(si, offset)) { |
810 | /* take a break if we already got some slots */ | |
811 | if (n_ret) | |
812 | goto done; | |
813 | if (!scan_swap_map_try_ssd_cluster(si, &offset, | |
814 | &scan_base)) | |
815 | goto scan; | |
816 | } | |
ebc2a1a6 | 817 | } |
ebebbbe9 | 818 | if (!(si->flags & SWP_WRITEOK)) |
52b7efdb | 819 | goto no_page; |
7dfad418 HD |
820 | if (!si->highest_bit) |
821 | goto no_page; | |
ebebbbe9 | 822 | if (offset > si->highest_bit) |
c60aa176 | 823 | scan_base = offset = si->lowest_bit; |
c9e44410 | 824 | |
235b6217 | 825 | ci = lock_cluster(si, offset); |
b73d7fce HD |
826 | /* reuse swap entry of cache-only swap if not busy. */ |
827 | if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) { | |
c9e44410 | 828 | int swap_was_freed; |
235b6217 | 829 | unlock_cluster(ci); |
ec8acf20 | 830 | spin_unlock(&si->lock); |
bcd49e86 | 831 | swap_was_freed = __try_to_reclaim_swap(si, offset, TTRS_ANYWAY); |
ec8acf20 | 832 | spin_lock(&si->lock); |
c9e44410 KH |
833 | /* entry was freed successfully, try to use this again */ |
834 | if (swap_was_freed) | |
835 | goto checks; | |
836 | goto scan; /* check next one */ | |
837 | } | |
838 | ||
235b6217 HY |
839 | if (si->swap_map[offset]) { |
840 | unlock_cluster(ci); | |
36005bae TC |
841 | if (!n_ret) |
842 | goto scan; | |
843 | else | |
844 | goto done; | |
235b6217 | 845 | } |
2872bb2d HY |
846 | si->swap_map[offset] = usage; |
847 | inc_cluster_info_page(si, si->cluster_info, offset); | |
848 | unlock_cluster(ci); | |
ebebbbe9 | 849 | |
38d8b4e6 | 850 | swap_range_alloc(si, offset, 1); |
ebebbbe9 | 851 | si->cluster_next = offset + 1; |
36005bae TC |
852 | slots[n_ret++] = swp_entry(si->type, offset); |
853 | ||
854 | /* got enough slots or reach max slots? */ | |
855 | if ((n_ret == nr) || (offset >= si->highest_bit)) | |
856 | goto done; | |
857 | ||
858 | /* search for next available slot */ | |
859 | ||
860 | /* time to take a break? */ | |
861 | if (unlikely(--latency_ration < 0)) { | |
862 | if (n_ret) | |
863 | goto done; | |
864 | spin_unlock(&si->lock); | |
865 | cond_resched(); | |
866 | spin_lock(&si->lock); | |
867 | latency_ration = LATENCY_LIMIT; | |
868 | } | |
869 | ||
870 | /* try to get more slots in cluster */ | |
871 | if (si->cluster_info) { | |
872 | if (scan_swap_map_try_ssd_cluster(si, &offset, &scan_base)) | |
873 | goto checks; | |
874 | else | |
875 | goto done; | |
876 | } | |
36005bae TC |
877 | |
878 | /* non-ssd case, still more slots in cluster? */ | |
ca2c55a7 | 879 | if (si->cluster_nr && !si->swap_map[++offset]) { |
36005bae TC |
880 | --si->cluster_nr; |
881 | goto checks; | |
882 | } | |
7992fde7 | 883 | |
36005bae TC |
884 | done: |
885 | si->flags -= SWP_SCANNING; | |
886 | return n_ret; | |
7dfad418 | 887 | |
ebebbbe9 | 888 | scan: |
ec8acf20 | 889 | spin_unlock(&si->lock); |
7dfad418 | 890 | while (++offset <= si->highest_bit) { |
52b7efdb | 891 | if (!si->swap_map[offset]) { |
ec8acf20 | 892 | spin_lock(&si->lock); |
52b7efdb HD |
893 | goto checks; |
894 | } | |
c9e44410 | 895 | if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) { |
ec8acf20 | 896 | spin_lock(&si->lock); |
c9e44410 KH |
897 | goto checks; |
898 | } | |
048c27fd HD |
899 | if (unlikely(--latency_ration < 0)) { |
900 | cond_resched(); | |
901 | latency_ration = LATENCY_LIMIT; | |
902 | } | |
7dfad418 | 903 | } |
c60aa176 | 904 | offset = si->lowest_bit; |
a5998061 | 905 | while (offset < scan_base) { |
c60aa176 | 906 | if (!si->swap_map[offset]) { |
ec8acf20 | 907 | spin_lock(&si->lock); |
c60aa176 HD |
908 | goto checks; |
909 | } | |
c9e44410 | 910 | if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) { |
ec8acf20 | 911 | spin_lock(&si->lock); |
c9e44410 KH |
912 | goto checks; |
913 | } | |
c60aa176 HD |
914 | if (unlikely(--latency_ration < 0)) { |
915 | cond_resched(); | |
916 | latency_ration = LATENCY_LIMIT; | |
917 | } | |
a5998061 | 918 | offset++; |
c60aa176 | 919 | } |
ec8acf20 | 920 | spin_lock(&si->lock); |
7dfad418 HD |
921 | |
922 | no_page: | |
52b7efdb | 923 | si->flags -= SWP_SCANNING; |
36005bae | 924 | return n_ret; |
1da177e4 LT |
925 | } |
926 | ||
38d8b4e6 HY |
927 | static int swap_alloc_cluster(struct swap_info_struct *si, swp_entry_t *slot) |
928 | { | |
929 | unsigned long idx; | |
930 | struct swap_cluster_info *ci; | |
931 | unsigned long offset, i; | |
932 | unsigned char *map; | |
933 | ||
fe5266d5 HY |
934 | /* |
935 | * Should not even be attempting cluster allocations when huge | |
936 | * page swap is disabled. Warn and fail the allocation. | |
937 | */ | |
938 | if (!IS_ENABLED(CONFIG_THP_SWAP)) { | |
939 | VM_WARN_ON_ONCE(1); | |
940 | return 0; | |
941 | } | |
942 | ||
38d8b4e6 HY |
943 | if (cluster_list_empty(&si->free_clusters)) |
944 | return 0; | |
945 | ||
946 | idx = cluster_list_first(&si->free_clusters); | |
947 | offset = idx * SWAPFILE_CLUSTER; | |
948 | ci = lock_cluster(si, offset); | |
949 | alloc_cluster(si, idx); | |
e0709829 | 950 | cluster_set_count_flag(ci, SWAPFILE_CLUSTER, CLUSTER_FLAG_HUGE); |
38d8b4e6 HY |
951 | |
952 | map = si->swap_map + offset; | |
953 | for (i = 0; i < SWAPFILE_CLUSTER; i++) | |
954 | map[i] = SWAP_HAS_CACHE; | |
955 | unlock_cluster(ci); | |
956 | swap_range_alloc(si, offset, SWAPFILE_CLUSTER); | |
957 | *slot = swp_entry(si->type, offset); | |
958 | ||
959 | return 1; | |
960 | } | |
961 | ||
962 | static void swap_free_cluster(struct swap_info_struct *si, unsigned long idx) | |
963 | { | |
964 | unsigned long offset = idx * SWAPFILE_CLUSTER; | |
965 | struct swap_cluster_info *ci; | |
966 | ||
967 | ci = lock_cluster(si, offset); | |
979aafa5 | 968 | memset(si->swap_map + offset, 0, SWAPFILE_CLUSTER); |
38d8b4e6 HY |
969 | cluster_set_count_flag(ci, 0, 0); |
970 | free_cluster(si, idx); | |
971 | unlock_cluster(ci); | |
972 | swap_range_free(si, offset, SWAPFILE_CLUSTER); | |
973 | } | |
38d8b4e6 | 974 | |
36005bae TC |
975 | static unsigned long scan_swap_map(struct swap_info_struct *si, |
976 | unsigned char usage) | |
977 | { | |
978 | swp_entry_t entry; | |
979 | int n_ret; | |
980 | ||
981 | n_ret = scan_swap_map_slots(si, usage, 1, &entry); | |
982 | ||
983 | if (n_ret) | |
984 | return swp_offset(entry); | |
985 | else | |
986 | return 0; | |
987 | ||
988 | } | |
989 | ||
5d5e8f19 | 990 | int get_swap_pages(int n_goal, swp_entry_t swp_entries[], int entry_size) |
1da177e4 | 991 | { |
5d5e8f19 | 992 | unsigned long size = swap_entry_size(entry_size); |
adfab836 | 993 | struct swap_info_struct *si, *next; |
36005bae TC |
994 | long avail_pgs; |
995 | int n_ret = 0; | |
a2468cc9 | 996 | int node; |
1da177e4 | 997 | |
38d8b4e6 | 998 | /* Only single cluster request supported */ |
5d5e8f19 | 999 | WARN_ON_ONCE(n_goal > 1 && size == SWAPFILE_CLUSTER); |
38d8b4e6 | 1000 | |
5d5e8f19 | 1001 | avail_pgs = atomic_long_read(&nr_swap_pages) / size; |
36005bae | 1002 | if (avail_pgs <= 0) |
fb4f88dc | 1003 | goto noswap; |
36005bae TC |
1004 | |
1005 | if (n_goal > SWAP_BATCH) | |
1006 | n_goal = SWAP_BATCH; | |
1007 | ||
1008 | if (n_goal > avail_pgs) | |
1009 | n_goal = avail_pgs; | |
1010 | ||
5d5e8f19 | 1011 | atomic_long_sub(n_goal * size, &nr_swap_pages); |
fb4f88dc | 1012 | |
18ab4d4c DS |
1013 | spin_lock(&swap_avail_lock); |
1014 | ||
1015 | start_over: | |
a2468cc9 AL |
1016 | node = numa_node_id(); |
1017 | plist_for_each_entry_safe(si, next, &swap_avail_heads[node], avail_lists[node]) { | |
18ab4d4c | 1018 | /* requeue si to after same-priority siblings */ |
a2468cc9 | 1019 | plist_requeue(&si->avail_lists[node], &swap_avail_heads[node]); |
18ab4d4c | 1020 | spin_unlock(&swap_avail_lock); |
ec8acf20 | 1021 | spin_lock(&si->lock); |
adfab836 | 1022 | if (!si->highest_bit || !(si->flags & SWP_WRITEOK)) { |
18ab4d4c | 1023 | spin_lock(&swap_avail_lock); |
a2468cc9 | 1024 | if (plist_node_empty(&si->avail_lists[node])) { |
18ab4d4c DS |
1025 | spin_unlock(&si->lock); |
1026 | goto nextsi; | |
1027 | } | |
1028 | WARN(!si->highest_bit, | |
1029 | "swap_info %d in list but !highest_bit\n", | |
1030 | si->type); | |
1031 | WARN(!(si->flags & SWP_WRITEOK), | |
1032 | "swap_info %d in list but !SWP_WRITEOK\n", | |
1033 | si->type); | |
a2468cc9 | 1034 | __del_from_avail_list(si); |
ec8acf20 | 1035 | spin_unlock(&si->lock); |
18ab4d4c | 1036 | goto nextsi; |
ec8acf20 | 1037 | } |
5d5e8f19 | 1038 | if (size == SWAPFILE_CLUSTER) { |
bc4ae27d | 1039 | if (!(si->flags & SWP_FS)) |
f0eea189 HY |
1040 | n_ret = swap_alloc_cluster(si, swp_entries); |
1041 | } else | |
38d8b4e6 HY |
1042 | n_ret = scan_swap_map_slots(si, SWAP_HAS_CACHE, |
1043 | n_goal, swp_entries); | |
ec8acf20 | 1044 | spin_unlock(&si->lock); |
5d5e8f19 | 1045 | if (n_ret || size == SWAPFILE_CLUSTER) |
36005bae | 1046 | goto check_out; |
18ab4d4c | 1047 | pr_debug("scan_swap_map of si %d failed to find offset\n", |
36005bae TC |
1048 | si->type); |
1049 | ||
18ab4d4c DS |
1050 | spin_lock(&swap_avail_lock); |
1051 | nextsi: | |
adfab836 DS |
1052 | /* |
1053 | * if we got here, it's likely that si was almost full before, | |
1054 | * and since scan_swap_map() can drop the si->lock, multiple | |
1055 | * callers probably all tried to get a page from the same si | |
18ab4d4c DS |
1056 | * and it filled up before we could get one; or, the si filled |
1057 | * up between us dropping swap_avail_lock and taking si->lock. | |
1058 | * Since we dropped the swap_avail_lock, the swap_avail_head | |
1059 | * list may have been modified; so if next is still in the | |
36005bae TC |
1060 | * swap_avail_head list then try it, otherwise start over |
1061 | * if we have not gotten any slots. | |
adfab836 | 1062 | */ |
a2468cc9 | 1063 | if (plist_node_empty(&next->avail_lists[node])) |
18ab4d4c | 1064 | goto start_over; |
1da177e4 | 1065 | } |
fb4f88dc | 1066 | |
18ab4d4c DS |
1067 | spin_unlock(&swap_avail_lock); |
1068 | ||
36005bae TC |
1069 | check_out: |
1070 | if (n_ret < n_goal) | |
5d5e8f19 | 1071 | atomic_long_add((long)(n_goal - n_ret) * size, |
38d8b4e6 | 1072 | &nr_swap_pages); |
fb4f88dc | 1073 | noswap: |
36005bae TC |
1074 | return n_ret; |
1075 | } | |
1076 | ||
2de1a7e4 | 1077 | /* The only caller of this function is now suspend routine */ |
910321ea HD |
1078 | swp_entry_t get_swap_page_of_type(int type) |
1079 | { | |
c10d38cc | 1080 | struct swap_info_struct *si = swap_type_to_swap_info(type); |
910321ea HD |
1081 | pgoff_t offset; |
1082 | ||
c10d38cc DJ |
1083 | if (!si) |
1084 | goto fail; | |
1085 | ||
ec8acf20 | 1086 | spin_lock(&si->lock); |
c10d38cc | 1087 | if (si->flags & SWP_WRITEOK) { |
ec8acf20 | 1088 | atomic_long_dec(&nr_swap_pages); |
910321ea HD |
1089 | /* This is called for allocating swap entry, not cache */ |
1090 | offset = scan_swap_map(si, 1); | |
1091 | if (offset) { | |
ec8acf20 | 1092 | spin_unlock(&si->lock); |
910321ea HD |
1093 | return swp_entry(type, offset); |
1094 | } | |
ec8acf20 | 1095 | atomic_long_inc(&nr_swap_pages); |
910321ea | 1096 | } |
ec8acf20 | 1097 | spin_unlock(&si->lock); |
c10d38cc | 1098 | fail: |
910321ea HD |
1099 | return (swp_entry_t) {0}; |
1100 | } | |
1101 | ||
e8c26ab6 | 1102 | static struct swap_info_struct *__swap_info_get(swp_entry_t entry) |
1da177e4 | 1103 | { |
73c34b6a | 1104 | struct swap_info_struct *p; |
eb085574 | 1105 | unsigned long offset; |
1da177e4 LT |
1106 | |
1107 | if (!entry.val) | |
1108 | goto out; | |
eb085574 | 1109 | p = swp_swap_info(entry); |
c10d38cc | 1110 | if (!p) |
1da177e4 | 1111 | goto bad_nofile; |
1da177e4 LT |
1112 | if (!(p->flags & SWP_USED)) |
1113 | goto bad_device; | |
1114 | offset = swp_offset(entry); | |
1115 | if (offset >= p->max) | |
1116 | goto bad_offset; | |
1da177e4 LT |
1117 | return p; |
1118 | ||
1da177e4 | 1119 | bad_offset: |
6a991fc7 | 1120 | pr_err("swap_info_get: %s%08lx\n", Bad_offset, entry.val); |
1da177e4 LT |
1121 | goto out; |
1122 | bad_device: | |
6a991fc7 | 1123 | pr_err("swap_info_get: %s%08lx\n", Unused_file, entry.val); |
1da177e4 LT |
1124 | goto out; |
1125 | bad_nofile: | |
6a991fc7 | 1126 | pr_err("swap_info_get: %s%08lx\n", Bad_file, entry.val); |
1da177e4 LT |
1127 | out: |
1128 | return NULL; | |
886bb7e9 | 1129 | } |
1da177e4 | 1130 | |
e8c26ab6 TC |
1131 | static struct swap_info_struct *_swap_info_get(swp_entry_t entry) |
1132 | { | |
1133 | struct swap_info_struct *p; | |
1134 | ||
1135 | p = __swap_info_get(entry); | |
1136 | if (!p) | |
1137 | goto out; | |
1138 | if (!p->swap_map[swp_offset(entry)]) | |
1139 | goto bad_free; | |
1140 | return p; | |
1141 | ||
1142 | bad_free: | |
1143 | pr_err("swap_info_get: %s%08lx\n", Unused_offset, entry.val); | |
1144 | goto out; | |
1145 | out: | |
1146 | return NULL; | |
1147 | } | |
1148 | ||
235b6217 HY |
1149 | static struct swap_info_struct *swap_info_get(swp_entry_t entry) |
1150 | { | |
1151 | struct swap_info_struct *p; | |
1152 | ||
1153 | p = _swap_info_get(entry); | |
1154 | if (p) | |
1155 | spin_lock(&p->lock); | |
1156 | return p; | |
1157 | } | |
1158 | ||
7c00bafe TC |
1159 | static struct swap_info_struct *swap_info_get_cont(swp_entry_t entry, |
1160 | struct swap_info_struct *q) | |
1161 | { | |
1162 | struct swap_info_struct *p; | |
1163 | ||
1164 | p = _swap_info_get(entry); | |
1165 | ||
1166 | if (p != q) { | |
1167 | if (q != NULL) | |
1168 | spin_unlock(&q->lock); | |
1169 | if (p != NULL) | |
1170 | spin_lock(&p->lock); | |
1171 | } | |
1172 | return p; | |
1173 | } | |
1174 | ||
b32d5f32 HY |
1175 | static unsigned char __swap_entry_free_locked(struct swap_info_struct *p, |
1176 | unsigned long offset, | |
1177 | unsigned char usage) | |
1da177e4 | 1178 | { |
8d69aaee HD |
1179 | unsigned char count; |
1180 | unsigned char has_cache; | |
235b6217 | 1181 | |
253d553b | 1182 | count = p->swap_map[offset]; |
235b6217 | 1183 | |
253d553b HD |
1184 | has_cache = count & SWAP_HAS_CACHE; |
1185 | count &= ~SWAP_HAS_CACHE; | |
355cfa73 | 1186 | |
253d553b | 1187 | if (usage == SWAP_HAS_CACHE) { |
355cfa73 | 1188 | VM_BUG_ON(!has_cache); |
253d553b | 1189 | has_cache = 0; |
aaa46865 HD |
1190 | } else if (count == SWAP_MAP_SHMEM) { |
1191 | /* | |
1192 | * Or we could insist on shmem.c using a special | |
1193 | * swap_shmem_free() and free_shmem_swap_and_cache()... | |
1194 | */ | |
1195 | count = 0; | |
570a335b HD |
1196 | } else if ((count & ~COUNT_CONTINUED) <= SWAP_MAP_MAX) { |
1197 | if (count == COUNT_CONTINUED) { | |
1198 | if (swap_count_continued(p, offset, count)) | |
1199 | count = SWAP_MAP_MAX | COUNT_CONTINUED; | |
1200 | else | |
1201 | count = SWAP_MAP_MAX; | |
1202 | } else | |
1203 | count--; | |
1204 | } | |
253d553b | 1205 | |
253d553b | 1206 | usage = count | has_cache; |
7c00bafe TC |
1207 | p->swap_map[offset] = usage ? : SWAP_HAS_CACHE; |
1208 | ||
b32d5f32 HY |
1209 | return usage; |
1210 | } | |
1211 | ||
eb085574 HY |
1212 | /* |
1213 | * Check whether swap entry is valid in the swap device. If so, | |
1214 | * return pointer to swap_info_struct, and keep the swap entry valid | |
1215 | * via preventing the swap device from being swapoff, until | |
1216 | * put_swap_device() is called. Otherwise return NULL. | |
1217 | * | |
1218 | * The entirety of the RCU read critical section must come before the | |
1219 | * return from or after the call to synchronize_rcu() in | |
1220 | * enable_swap_info() or swapoff(). So if "si->flags & SWP_VALID" is | |
1221 | * true, the si->map, si->cluster_info, etc. must be valid in the | |
1222 | * critical section. | |
1223 | * | |
1224 | * Notice that swapoff or swapoff+swapon can still happen before the | |
1225 | * rcu_read_lock() in get_swap_device() or after the rcu_read_unlock() | |
1226 | * in put_swap_device() if there isn't any other way to prevent | |
1227 | * swapoff, such as page lock, page table lock, etc. The caller must | |
1228 | * be prepared for that. For example, the following situation is | |
1229 | * possible. | |
1230 | * | |
1231 | * CPU1 CPU2 | |
1232 | * do_swap_page() | |
1233 | * ... swapoff+swapon | |
1234 | * __read_swap_cache_async() | |
1235 | * swapcache_prepare() | |
1236 | * __swap_duplicate() | |
1237 | * // check swap_map | |
1238 | * // verify PTE not changed | |
1239 | * | |
1240 | * In __swap_duplicate(), the swap_map need to be checked before | |
1241 | * changing partly because the specified swap entry may be for another | |
1242 | * swap device which has been swapoff. And in do_swap_page(), after | |
1243 | * the page is read from the swap device, the PTE is verified not | |
1244 | * changed with the page table locked to check whether the swap device | |
1245 | * has been swapoff or swapoff+swapon. | |
1246 | */ | |
1247 | struct swap_info_struct *get_swap_device(swp_entry_t entry) | |
1248 | { | |
1249 | struct swap_info_struct *si; | |
1250 | unsigned long offset; | |
1251 | ||
1252 | if (!entry.val) | |
1253 | goto out; | |
1254 | si = swp_swap_info(entry); | |
1255 | if (!si) | |
1256 | goto bad_nofile; | |
1257 | ||
1258 | rcu_read_lock(); | |
1259 | if (!(si->flags & SWP_VALID)) | |
1260 | goto unlock_out; | |
1261 | offset = swp_offset(entry); | |
1262 | if (offset >= si->max) | |
1263 | goto unlock_out; | |
1264 | ||
1265 | return si; | |
1266 | bad_nofile: | |
1267 | pr_err("%s: %s%08lx\n", __func__, Bad_file, entry.val); | |
1268 | out: | |
1269 | return NULL; | |
1270 | unlock_out: | |
1271 | rcu_read_unlock(); | |
1272 | return NULL; | |
1273 | } | |
1274 | ||
b32d5f32 HY |
1275 | static unsigned char __swap_entry_free(struct swap_info_struct *p, |
1276 | swp_entry_t entry, unsigned char usage) | |
1277 | { | |
1278 | struct swap_cluster_info *ci; | |
1279 | unsigned long offset = swp_offset(entry); | |
1280 | ||
1281 | ci = lock_cluster_or_swap_info(p, offset); | |
1282 | usage = __swap_entry_free_locked(p, offset, usage); | |
7c00bafe | 1283 | unlock_cluster_or_swap_info(p, ci); |
10e364da HY |
1284 | if (!usage) |
1285 | free_swap_slot(entry); | |
7c00bafe TC |
1286 | |
1287 | return usage; | |
1288 | } | |
355cfa73 | 1289 | |
7c00bafe TC |
1290 | static void swap_entry_free(struct swap_info_struct *p, swp_entry_t entry) |
1291 | { | |
1292 | struct swap_cluster_info *ci; | |
1293 | unsigned long offset = swp_offset(entry); | |
1294 | unsigned char count; | |
1295 | ||
1296 | ci = lock_cluster(p, offset); | |
1297 | count = p->swap_map[offset]; | |
1298 | VM_BUG_ON(count != SWAP_HAS_CACHE); | |
1299 | p->swap_map[offset] = 0; | |
1300 | dec_cluster_info_page(p, p->cluster_info, offset); | |
235b6217 HY |
1301 | unlock_cluster(ci); |
1302 | ||
38d8b4e6 HY |
1303 | mem_cgroup_uncharge_swap(entry, 1); |
1304 | swap_range_free(p, offset, 1); | |
1da177e4 LT |
1305 | } |
1306 | ||
1307 | /* | |
2de1a7e4 | 1308 | * Caller has made sure that the swap device corresponding to entry |
1da177e4 LT |
1309 | * is still around or has not been recycled. |
1310 | */ | |
1311 | void swap_free(swp_entry_t entry) | |
1312 | { | |
73c34b6a | 1313 | struct swap_info_struct *p; |
1da177e4 | 1314 | |
235b6217 | 1315 | p = _swap_info_get(entry); |
10e364da HY |
1316 | if (p) |
1317 | __swap_entry_free(p, entry, 1); | |
1da177e4 LT |
1318 | } |
1319 | ||
cb4b86ba KH |
1320 | /* |
1321 | * Called after dropping swapcache to decrease refcnt to swap entries. | |
1322 | */ | |
a448f2d0 | 1323 | void put_swap_page(struct page *page, swp_entry_t entry) |
38d8b4e6 HY |
1324 | { |
1325 | unsigned long offset = swp_offset(entry); | |
1326 | unsigned long idx = offset / SWAPFILE_CLUSTER; | |
1327 | struct swap_cluster_info *ci; | |
1328 | struct swap_info_struct *si; | |
1329 | unsigned char *map; | |
a3aea839 HY |
1330 | unsigned int i, free_entries = 0; |
1331 | unsigned char val; | |
a448f2d0 | 1332 | int size = swap_entry_size(hpage_nr_pages(page)); |
fe5266d5 | 1333 | |
a3aea839 | 1334 | si = _swap_info_get(entry); |
38d8b4e6 HY |
1335 | if (!si) |
1336 | return; | |
1337 | ||
c2343d27 | 1338 | ci = lock_cluster_or_swap_info(si, offset); |
a448f2d0 | 1339 | if (size == SWAPFILE_CLUSTER) { |
a448f2d0 HY |
1340 | VM_BUG_ON(!cluster_is_huge(ci)); |
1341 | map = si->swap_map + offset; | |
1342 | for (i = 0; i < SWAPFILE_CLUSTER; i++) { | |
1343 | val = map[i]; | |
1344 | VM_BUG_ON(!(val & SWAP_HAS_CACHE)); | |
1345 | if (val == SWAP_HAS_CACHE) | |
1346 | free_entries++; | |
1347 | } | |
a448f2d0 | 1348 | cluster_clear_huge(ci); |
a448f2d0 | 1349 | if (free_entries == SWAPFILE_CLUSTER) { |
c2343d27 | 1350 | unlock_cluster_or_swap_info(si, ci); |
a448f2d0 | 1351 | spin_lock(&si->lock); |
a448f2d0 HY |
1352 | mem_cgroup_uncharge_swap(entry, SWAPFILE_CLUSTER); |
1353 | swap_free_cluster(si, idx); | |
1354 | spin_unlock(&si->lock); | |
1355 | return; | |
1356 | } | |
1357 | } | |
c2343d27 HY |
1358 | for (i = 0; i < size; i++, entry.val++) { |
1359 | if (!__swap_entry_free_locked(si, offset + i, SWAP_HAS_CACHE)) { | |
1360 | unlock_cluster_or_swap_info(si, ci); | |
1361 | free_swap_slot(entry); | |
1362 | if (i == size - 1) | |
1363 | return; | |
1364 | lock_cluster_or_swap_info(si, offset); | |
a3aea839 HY |
1365 | } |
1366 | } | |
c2343d27 | 1367 | unlock_cluster_or_swap_info(si, ci); |
38d8b4e6 | 1368 | } |
59807685 | 1369 | |
fe5266d5 | 1370 | #ifdef CONFIG_THP_SWAP |
59807685 HY |
1371 | int split_swap_cluster(swp_entry_t entry) |
1372 | { | |
1373 | struct swap_info_struct *si; | |
1374 | struct swap_cluster_info *ci; | |
1375 | unsigned long offset = swp_offset(entry); | |
1376 | ||
1377 | si = _swap_info_get(entry); | |
1378 | if (!si) | |
1379 | return -EBUSY; | |
1380 | ci = lock_cluster(si, offset); | |
1381 | cluster_clear_huge(ci); | |
1382 | unlock_cluster(ci); | |
1383 | return 0; | |
1384 | } | |
fe5266d5 | 1385 | #endif |
38d8b4e6 | 1386 | |
155b5f88 HY |
1387 | static int swp_entry_cmp(const void *ent1, const void *ent2) |
1388 | { | |
1389 | const swp_entry_t *e1 = ent1, *e2 = ent2; | |
1390 | ||
1391 | return (int)swp_type(*e1) - (int)swp_type(*e2); | |
1392 | } | |
1393 | ||
7c00bafe TC |
1394 | void swapcache_free_entries(swp_entry_t *entries, int n) |
1395 | { | |
1396 | struct swap_info_struct *p, *prev; | |
1397 | int i; | |
1398 | ||
1399 | if (n <= 0) | |
1400 | return; | |
1401 | ||
1402 | prev = NULL; | |
1403 | p = NULL; | |
155b5f88 HY |
1404 | |
1405 | /* | |
1406 | * Sort swap entries by swap device, so each lock is only taken once. | |
1407 | * nr_swapfiles isn't absolutely correct, but the overhead of sort() is | |
1408 | * so low that it isn't necessary to optimize further. | |
1409 | */ | |
1410 | if (nr_swapfiles > 1) | |
1411 | sort(entries, n, sizeof(entries[0]), swp_entry_cmp, NULL); | |
7c00bafe TC |
1412 | for (i = 0; i < n; ++i) { |
1413 | p = swap_info_get_cont(entries[i], prev); | |
1414 | if (p) | |
1415 | swap_entry_free(p, entries[i]); | |
7c00bafe TC |
1416 | prev = p; |
1417 | } | |
235b6217 | 1418 | if (p) |
7c00bafe | 1419 | spin_unlock(&p->lock); |
cb4b86ba KH |
1420 | } |
1421 | ||
1da177e4 | 1422 | /* |
c475a8ab | 1423 | * How many references to page are currently swapped out? |
570a335b HD |
1424 | * This does not give an exact answer when swap count is continued, |
1425 | * but does include the high COUNT_CONTINUED flag to allow for that. | |
1da177e4 | 1426 | */ |
bde05d1c | 1427 | int page_swapcount(struct page *page) |
1da177e4 | 1428 | { |
c475a8ab HD |
1429 | int count = 0; |
1430 | struct swap_info_struct *p; | |
235b6217 | 1431 | struct swap_cluster_info *ci; |
1da177e4 | 1432 | swp_entry_t entry; |
235b6217 | 1433 | unsigned long offset; |
1da177e4 | 1434 | |
4c21e2f2 | 1435 | entry.val = page_private(page); |
235b6217 | 1436 | p = _swap_info_get(entry); |
1da177e4 | 1437 | if (p) { |
235b6217 HY |
1438 | offset = swp_offset(entry); |
1439 | ci = lock_cluster_or_swap_info(p, offset); | |
1440 | count = swap_count(p->swap_map[offset]); | |
1441 | unlock_cluster_or_swap_info(p, ci); | |
1da177e4 | 1442 | } |
c475a8ab | 1443 | return count; |
1da177e4 LT |
1444 | } |
1445 | ||
eb085574 | 1446 | int __swap_count(swp_entry_t entry) |
aa8d22a1 | 1447 | { |
eb085574 | 1448 | struct swap_info_struct *si; |
aa8d22a1 | 1449 | pgoff_t offset = swp_offset(entry); |
eb085574 | 1450 | int count = 0; |
aa8d22a1 | 1451 | |
eb085574 HY |
1452 | si = get_swap_device(entry); |
1453 | if (si) { | |
1454 | count = swap_count(si->swap_map[offset]); | |
1455 | put_swap_device(si); | |
1456 | } | |
1457 | return count; | |
aa8d22a1 MK |
1458 | } |
1459 | ||
322b8afe HY |
1460 | static int swap_swapcount(struct swap_info_struct *si, swp_entry_t entry) |
1461 | { | |
1462 | int count = 0; | |
1463 | pgoff_t offset = swp_offset(entry); | |
1464 | struct swap_cluster_info *ci; | |
1465 | ||
1466 | ci = lock_cluster_or_swap_info(si, offset); | |
1467 | count = swap_count(si->swap_map[offset]); | |
1468 | unlock_cluster_or_swap_info(si, ci); | |
1469 | return count; | |
1470 | } | |
1471 | ||
e8c26ab6 TC |
1472 | /* |
1473 | * How many references to @entry are currently swapped out? | |
1474 | * This does not give an exact answer when swap count is continued, | |
1475 | * but does include the high COUNT_CONTINUED flag to allow for that. | |
1476 | */ | |
1477 | int __swp_swapcount(swp_entry_t entry) | |
1478 | { | |
1479 | int count = 0; | |
e8c26ab6 | 1480 | struct swap_info_struct *si; |
e8c26ab6 | 1481 | |
eb085574 HY |
1482 | si = get_swap_device(entry); |
1483 | if (si) { | |
322b8afe | 1484 | count = swap_swapcount(si, entry); |
eb085574 HY |
1485 | put_swap_device(si); |
1486 | } | |
e8c26ab6 TC |
1487 | return count; |
1488 | } | |
1489 | ||
8334b962 MK |
1490 | /* |
1491 | * How many references to @entry are currently swapped out? | |
1492 | * This considers COUNT_CONTINUED so it returns exact answer. | |
1493 | */ | |
1494 | int swp_swapcount(swp_entry_t entry) | |
1495 | { | |
1496 | int count, tmp_count, n; | |
1497 | struct swap_info_struct *p; | |
235b6217 | 1498 | struct swap_cluster_info *ci; |
8334b962 MK |
1499 | struct page *page; |
1500 | pgoff_t offset; | |
1501 | unsigned char *map; | |
1502 | ||
235b6217 | 1503 | p = _swap_info_get(entry); |
8334b962 MK |
1504 | if (!p) |
1505 | return 0; | |
1506 | ||
235b6217 HY |
1507 | offset = swp_offset(entry); |
1508 | ||
1509 | ci = lock_cluster_or_swap_info(p, offset); | |
1510 | ||
1511 | count = swap_count(p->swap_map[offset]); | |
8334b962 MK |
1512 | if (!(count & COUNT_CONTINUED)) |
1513 | goto out; | |
1514 | ||
1515 | count &= ~COUNT_CONTINUED; | |
1516 | n = SWAP_MAP_MAX + 1; | |
1517 | ||
8334b962 MK |
1518 | page = vmalloc_to_page(p->swap_map + offset); |
1519 | offset &= ~PAGE_MASK; | |
1520 | VM_BUG_ON(page_private(page) != SWP_CONTINUED); | |
1521 | ||
1522 | do { | |
a8ae4991 | 1523 | page = list_next_entry(page, lru); |
8334b962 MK |
1524 | map = kmap_atomic(page); |
1525 | tmp_count = map[offset]; | |
1526 | kunmap_atomic(map); | |
1527 | ||
1528 | count += (tmp_count & ~COUNT_CONTINUED) * n; | |
1529 | n *= (SWAP_CONT_MAX + 1); | |
1530 | } while (tmp_count & COUNT_CONTINUED); | |
1531 | out: | |
235b6217 | 1532 | unlock_cluster_or_swap_info(p, ci); |
8334b962 MK |
1533 | return count; |
1534 | } | |
1535 | ||
e0709829 HY |
1536 | static bool swap_page_trans_huge_swapped(struct swap_info_struct *si, |
1537 | swp_entry_t entry) | |
1538 | { | |
1539 | struct swap_cluster_info *ci; | |
1540 | unsigned char *map = si->swap_map; | |
1541 | unsigned long roffset = swp_offset(entry); | |
1542 | unsigned long offset = round_down(roffset, SWAPFILE_CLUSTER); | |
1543 | int i; | |
1544 | bool ret = false; | |
1545 | ||
1546 | ci = lock_cluster_or_swap_info(si, offset); | |
1547 | if (!ci || !cluster_is_huge(ci)) { | |
afa4711e | 1548 | if (swap_count(map[roffset])) |
e0709829 HY |
1549 | ret = true; |
1550 | goto unlock_out; | |
1551 | } | |
1552 | for (i = 0; i < SWAPFILE_CLUSTER; i++) { | |
afa4711e | 1553 | if (swap_count(map[offset + i])) { |
e0709829 HY |
1554 | ret = true; |
1555 | break; | |
1556 | } | |
1557 | } | |
1558 | unlock_out: | |
1559 | unlock_cluster_or_swap_info(si, ci); | |
1560 | return ret; | |
1561 | } | |
1562 | ||
1563 | static bool page_swapped(struct page *page) | |
1564 | { | |
1565 | swp_entry_t entry; | |
1566 | struct swap_info_struct *si; | |
1567 | ||
fe5266d5 | 1568 | if (!IS_ENABLED(CONFIG_THP_SWAP) || likely(!PageTransCompound(page))) |
e0709829 HY |
1569 | return page_swapcount(page) != 0; |
1570 | ||
1571 | page = compound_head(page); | |
1572 | entry.val = page_private(page); | |
1573 | si = _swap_info_get(entry); | |
1574 | if (si) | |
1575 | return swap_page_trans_huge_swapped(si, entry); | |
1576 | return false; | |
1577 | } | |
ba3c4ce6 HY |
1578 | |
1579 | static int page_trans_huge_map_swapcount(struct page *page, int *total_mapcount, | |
1580 | int *total_swapcount) | |
1581 | { | |
1582 | int i, map_swapcount, _total_mapcount, _total_swapcount; | |
1583 | unsigned long offset = 0; | |
1584 | struct swap_info_struct *si; | |
1585 | struct swap_cluster_info *ci = NULL; | |
1586 | unsigned char *map = NULL; | |
1587 | int mapcount, swapcount = 0; | |
1588 | ||
1589 | /* hugetlbfs shouldn't call it */ | |
1590 | VM_BUG_ON_PAGE(PageHuge(page), page); | |
1591 | ||
fe5266d5 HY |
1592 | if (!IS_ENABLED(CONFIG_THP_SWAP) || likely(!PageTransCompound(page))) { |
1593 | mapcount = page_trans_huge_mapcount(page, total_mapcount); | |
ba3c4ce6 HY |
1594 | if (PageSwapCache(page)) |
1595 | swapcount = page_swapcount(page); | |
1596 | if (total_swapcount) | |
1597 | *total_swapcount = swapcount; | |
1598 | return mapcount + swapcount; | |
1599 | } | |
1600 | ||
1601 | page = compound_head(page); | |
1602 | ||
1603 | _total_mapcount = _total_swapcount = map_swapcount = 0; | |
1604 | if (PageSwapCache(page)) { | |
1605 | swp_entry_t entry; | |
1606 | ||
1607 | entry.val = page_private(page); | |
1608 | si = _swap_info_get(entry); | |
1609 | if (si) { | |
1610 | map = si->swap_map; | |
1611 | offset = swp_offset(entry); | |
1612 | } | |
1613 | } | |
1614 | if (map) | |
1615 | ci = lock_cluster(si, offset); | |
1616 | for (i = 0; i < HPAGE_PMD_NR; i++) { | |
1617 | mapcount = atomic_read(&page[i]._mapcount) + 1; | |
1618 | _total_mapcount += mapcount; | |
1619 | if (map) { | |
1620 | swapcount = swap_count(map[offset + i]); | |
1621 | _total_swapcount += swapcount; | |
1622 | } | |
1623 | map_swapcount = max(map_swapcount, mapcount + swapcount); | |
1624 | } | |
1625 | unlock_cluster(ci); | |
1626 | if (PageDoubleMap(page)) { | |
1627 | map_swapcount -= 1; | |
1628 | _total_mapcount -= HPAGE_PMD_NR; | |
1629 | } | |
1630 | mapcount = compound_mapcount(page); | |
1631 | map_swapcount += mapcount; | |
1632 | _total_mapcount += mapcount; | |
1633 | if (total_mapcount) | |
1634 | *total_mapcount = _total_mapcount; | |
1635 | if (total_swapcount) | |
1636 | *total_swapcount = _total_swapcount; | |
1637 | ||
1638 | return map_swapcount; | |
1639 | } | |
e0709829 | 1640 | |
1da177e4 | 1641 | /* |
7b1fe597 HD |
1642 | * We can write to an anon page without COW if there are no other references |
1643 | * to it. And as a side-effect, free up its swap: because the old content | |
1644 | * on disk will never be read, and seeking back there to write new content | |
1645 | * later would only waste time away from clustering. | |
6d0a07ed | 1646 | * |
ba3c4ce6 | 1647 | * NOTE: total_map_swapcount should not be relied upon by the caller if |
6d0a07ed AA |
1648 | * reuse_swap_page() returns false, but it may be always overwritten |
1649 | * (see the other implementation for CONFIG_SWAP=n). | |
1da177e4 | 1650 | */ |
ba3c4ce6 | 1651 | bool reuse_swap_page(struct page *page, int *total_map_swapcount) |
1da177e4 | 1652 | { |
ba3c4ce6 | 1653 | int count, total_mapcount, total_swapcount; |
c475a8ab | 1654 | |
309381fe | 1655 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
5ad64688 | 1656 | if (unlikely(PageKsm(page))) |
6d0a07ed | 1657 | return false; |
ba3c4ce6 HY |
1658 | count = page_trans_huge_map_swapcount(page, &total_mapcount, |
1659 | &total_swapcount); | |
1660 | if (total_map_swapcount) | |
1661 | *total_map_swapcount = total_mapcount + total_swapcount; | |
1662 | if (count == 1 && PageSwapCache(page) && | |
1663 | (likely(!PageTransCompound(page)) || | |
1664 | /* The remaining swap count will be freed soon */ | |
1665 | total_swapcount == page_swapcount(page))) { | |
f0571429 | 1666 | if (!PageWriteback(page)) { |
ba3c4ce6 | 1667 | page = compound_head(page); |
7b1fe597 HD |
1668 | delete_from_swap_cache(page); |
1669 | SetPageDirty(page); | |
f0571429 MK |
1670 | } else { |
1671 | swp_entry_t entry; | |
1672 | struct swap_info_struct *p; | |
1673 | ||
1674 | entry.val = page_private(page); | |
1675 | p = swap_info_get(entry); | |
1676 | if (p->flags & SWP_STABLE_WRITES) { | |
1677 | spin_unlock(&p->lock); | |
1678 | return false; | |
1679 | } | |
1680 | spin_unlock(&p->lock); | |
7b1fe597 HD |
1681 | } |
1682 | } | |
ba3c4ce6 | 1683 | |
5ad64688 | 1684 | return count <= 1; |
1da177e4 LT |
1685 | } |
1686 | ||
1687 | /* | |
a2c43eed HD |
1688 | * If swap is getting full, or if there are no more mappings of this page, |
1689 | * then try_to_free_swap is called to free its swap space. | |
1da177e4 | 1690 | */ |
a2c43eed | 1691 | int try_to_free_swap(struct page *page) |
1da177e4 | 1692 | { |
309381fe | 1693 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
1da177e4 LT |
1694 | |
1695 | if (!PageSwapCache(page)) | |
1696 | return 0; | |
1697 | if (PageWriteback(page)) | |
1698 | return 0; | |
e0709829 | 1699 | if (page_swapped(page)) |
1da177e4 LT |
1700 | return 0; |
1701 | ||
b73d7fce HD |
1702 | /* |
1703 | * Once hibernation has begun to create its image of memory, | |
1704 | * there's a danger that one of the calls to try_to_free_swap() | |
1705 | * - most probably a call from __try_to_reclaim_swap() while | |
1706 | * hibernation is allocating its own swap pages for the image, | |
1707 | * but conceivably even a call from memory reclaim - will free | |
1708 | * the swap from a page which has already been recorded in the | |
1709 | * image as a clean swapcache page, and then reuse its swap for | |
1710 | * another page of the image. On waking from hibernation, the | |
1711 | * original page might be freed under memory pressure, then | |
1712 | * later read back in from swap, now with the wrong data. | |
1713 | * | |
2de1a7e4 | 1714 | * Hibernation suspends storage while it is writing the image |
f90ac398 | 1715 | * to disk so check that here. |
b73d7fce | 1716 | */ |
f90ac398 | 1717 | if (pm_suspended_storage()) |
b73d7fce HD |
1718 | return 0; |
1719 | ||
e0709829 | 1720 | page = compound_head(page); |
a2c43eed HD |
1721 | delete_from_swap_cache(page); |
1722 | SetPageDirty(page); | |
1723 | return 1; | |
68a22394 RR |
1724 | } |
1725 | ||
1da177e4 LT |
1726 | /* |
1727 | * Free the swap entry like above, but also try to | |
1728 | * free the page cache entry if it is the last user. | |
1729 | */ | |
2509ef26 | 1730 | int free_swap_and_cache(swp_entry_t entry) |
1da177e4 | 1731 | { |
2509ef26 | 1732 | struct swap_info_struct *p; |
7c00bafe | 1733 | unsigned char count; |
1da177e4 | 1734 | |
a7420aa5 | 1735 | if (non_swap_entry(entry)) |
2509ef26 | 1736 | return 1; |
0697212a | 1737 | |
7c00bafe | 1738 | p = _swap_info_get(entry); |
1da177e4 | 1739 | if (p) { |
7c00bafe | 1740 | count = __swap_entry_free(p, entry, 1); |
e0709829 | 1741 | if (count == SWAP_HAS_CACHE && |
bcd49e86 HY |
1742 | !swap_page_trans_huge_swapped(p, entry)) |
1743 | __try_to_reclaim_swap(p, swp_offset(entry), | |
1744 | TTRS_UNMAPPED | TTRS_FULL); | |
1da177e4 | 1745 | } |
2509ef26 | 1746 | return p != NULL; |
1da177e4 LT |
1747 | } |
1748 | ||
b0cb1a19 | 1749 | #ifdef CONFIG_HIBERNATION |
f577eb30 | 1750 | /* |
915bae9e | 1751 | * Find the swap type that corresponds to given device (if any). |
f577eb30 | 1752 | * |
915bae9e RW |
1753 | * @offset - number of the PAGE_SIZE-sized block of the device, starting |
1754 | * from 0, in which the swap header is expected to be located. | |
1755 | * | |
1756 | * This is needed for the suspend to disk (aka swsusp). | |
f577eb30 | 1757 | */ |
7bf23687 | 1758 | int swap_type_of(dev_t device, sector_t offset, struct block_device **bdev_p) |
f577eb30 | 1759 | { |
915bae9e | 1760 | struct block_device *bdev = NULL; |
efa90a98 | 1761 | int type; |
f577eb30 | 1762 | |
915bae9e RW |
1763 | if (device) |
1764 | bdev = bdget(device); | |
1765 | ||
f577eb30 | 1766 | spin_lock(&swap_lock); |
efa90a98 HD |
1767 | for (type = 0; type < nr_swapfiles; type++) { |
1768 | struct swap_info_struct *sis = swap_info[type]; | |
f577eb30 | 1769 | |
915bae9e | 1770 | if (!(sis->flags & SWP_WRITEOK)) |
f577eb30 | 1771 | continue; |
b6b5bce3 | 1772 | |
915bae9e | 1773 | if (!bdev) { |
7bf23687 | 1774 | if (bdev_p) |
dddac6a7 | 1775 | *bdev_p = bdgrab(sis->bdev); |
7bf23687 | 1776 | |
6e1819d6 | 1777 | spin_unlock(&swap_lock); |
efa90a98 | 1778 | return type; |
6e1819d6 | 1779 | } |
915bae9e | 1780 | if (bdev == sis->bdev) { |
4efaceb1 | 1781 | struct swap_extent *se = first_se(sis); |
915bae9e | 1782 | |
915bae9e | 1783 | if (se->start_block == offset) { |
7bf23687 | 1784 | if (bdev_p) |
dddac6a7 | 1785 | *bdev_p = bdgrab(sis->bdev); |
7bf23687 | 1786 | |
915bae9e RW |
1787 | spin_unlock(&swap_lock); |
1788 | bdput(bdev); | |
efa90a98 | 1789 | return type; |
915bae9e | 1790 | } |
f577eb30 RW |
1791 | } |
1792 | } | |
1793 | spin_unlock(&swap_lock); | |
915bae9e RW |
1794 | if (bdev) |
1795 | bdput(bdev); | |
1796 | ||
f577eb30 RW |
1797 | return -ENODEV; |
1798 | } | |
1799 | ||
73c34b6a HD |
1800 | /* |
1801 | * Get the (PAGE_SIZE) block corresponding to given offset on the swapdev | |
1802 | * corresponding to given index in swap_info (swap type). | |
1803 | */ | |
1804 | sector_t swapdev_block(int type, pgoff_t offset) | |
1805 | { | |
1806 | struct block_device *bdev; | |
c10d38cc | 1807 | struct swap_info_struct *si = swap_type_to_swap_info(type); |
73c34b6a | 1808 | |
c10d38cc | 1809 | if (!si || !(si->flags & SWP_WRITEOK)) |
73c34b6a | 1810 | return 0; |
d4906e1a | 1811 | return map_swap_entry(swp_entry(type, offset), &bdev); |
73c34b6a HD |
1812 | } |
1813 | ||
f577eb30 RW |
1814 | /* |
1815 | * Return either the total number of swap pages of given type, or the number | |
1816 | * of free pages of that type (depending on @free) | |
1817 | * | |
1818 | * This is needed for software suspend | |
1819 | */ | |
1820 | unsigned int count_swap_pages(int type, int free) | |
1821 | { | |
1822 | unsigned int n = 0; | |
1823 | ||
efa90a98 HD |
1824 | spin_lock(&swap_lock); |
1825 | if ((unsigned int)type < nr_swapfiles) { | |
1826 | struct swap_info_struct *sis = swap_info[type]; | |
1827 | ||
ec8acf20 | 1828 | spin_lock(&sis->lock); |
efa90a98 HD |
1829 | if (sis->flags & SWP_WRITEOK) { |
1830 | n = sis->pages; | |
f577eb30 | 1831 | if (free) |
efa90a98 | 1832 | n -= sis->inuse_pages; |
f577eb30 | 1833 | } |
ec8acf20 | 1834 | spin_unlock(&sis->lock); |
f577eb30 | 1835 | } |
efa90a98 | 1836 | spin_unlock(&swap_lock); |
f577eb30 RW |
1837 | return n; |
1838 | } | |
73c34b6a | 1839 | #endif /* CONFIG_HIBERNATION */ |
f577eb30 | 1840 | |
9f8bdb3f | 1841 | static inline int pte_same_as_swp(pte_t pte, pte_t swp_pte) |
179ef71c | 1842 | { |
9f8bdb3f | 1843 | return pte_same(pte_swp_clear_soft_dirty(pte), swp_pte); |
179ef71c CG |
1844 | } |
1845 | ||
1da177e4 | 1846 | /* |
72866f6f HD |
1847 | * No need to decide whether this PTE shares the swap entry with others, |
1848 | * just let do_wp_page work it out if a write is requested later - to | |
1849 | * force COW, vm_page_prot omits write permission from any private vma. | |
1da177e4 | 1850 | */ |
044d66c1 | 1851 | static int unuse_pte(struct vm_area_struct *vma, pmd_t *pmd, |
1da177e4 LT |
1852 | unsigned long addr, swp_entry_t entry, struct page *page) |
1853 | { | |
9e16b7fb | 1854 | struct page *swapcache; |
72835c86 | 1855 | struct mem_cgroup *memcg; |
044d66c1 HD |
1856 | spinlock_t *ptl; |
1857 | pte_t *pte; | |
1858 | int ret = 1; | |
1859 | ||
9e16b7fb HD |
1860 | swapcache = page; |
1861 | page = ksm_might_need_to_copy(page, vma, addr); | |
1862 | if (unlikely(!page)) | |
1863 | return -ENOMEM; | |
1864 | ||
f627c2f5 KS |
1865 | if (mem_cgroup_try_charge(page, vma->vm_mm, GFP_KERNEL, |
1866 | &memcg, false)) { | |
044d66c1 | 1867 | ret = -ENOMEM; |
85d9fc89 KH |
1868 | goto out_nolock; |
1869 | } | |
044d66c1 HD |
1870 | |
1871 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); | |
9f8bdb3f | 1872 | if (unlikely(!pte_same_as_swp(*pte, swp_entry_to_pte(entry)))) { |
f627c2f5 | 1873 | mem_cgroup_cancel_charge(page, memcg, false); |
044d66c1 HD |
1874 | ret = 0; |
1875 | goto out; | |
1876 | } | |
8a9f3ccd | 1877 | |
b084d435 | 1878 | dec_mm_counter(vma->vm_mm, MM_SWAPENTS); |
d559db08 | 1879 | inc_mm_counter(vma->vm_mm, MM_ANONPAGES); |
1da177e4 LT |
1880 | get_page(page); |
1881 | set_pte_at(vma->vm_mm, addr, pte, | |
1882 | pte_mkold(mk_pte(page, vma->vm_page_prot))); | |
00501b53 | 1883 | if (page == swapcache) { |
d281ee61 | 1884 | page_add_anon_rmap(page, vma, addr, false); |
f627c2f5 | 1885 | mem_cgroup_commit_charge(page, memcg, true, false); |
00501b53 | 1886 | } else { /* ksm created a completely new copy */ |
d281ee61 | 1887 | page_add_new_anon_rmap(page, vma, addr, false); |
f627c2f5 | 1888 | mem_cgroup_commit_charge(page, memcg, false, false); |
00501b53 JW |
1889 | lru_cache_add_active_or_unevictable(page, vma); |
1890 | } | |
1da177e4 LT |
1891 | swap_free(entry); |
1892 | /* | |
1893 | * Move the page to the active list so it is not | |
1894 | * immediately swapped out again after swapon. | |
1895 | */ | |
1896 | activate_page(page); | |
044d66c1 HD |
1897 | out: |
1898 | pte_unmap_unlock(pte, ptl); | |
85d9fc89 | 1899 | out_nolock: |
9e16b7fb HD |
1900 | if (page != swapcache) { |
1901 | unlock_page(page); | |
1902 | put_page(page); | |
1903 | } | |
044d66c1 | 1904 | return ret; |
1da177e4 LT |
1905 | } |
1906 | ||
1907 | static int unuse_pte_range(struct vm_area_struct *vma, pmd_t *pmd, | |
b56a2d8a VRP |
1908 | unsigned long addr, unsigned long end, |
1909 | unsigned int type, bool frontswap, | |
1910 | unsigned long *fs_pages_to_unuse) | |
1da177e4 | 1911 | { |
b56a2d8a VRP |
1912 | struct page *page; |
1913 | swp_entry_t entry; | |
705e87c0 | 1914 | pte_t *pte; |
b56a2d8a VRP |
1915 | struct swap_info_struct *si; |
1916 | unsigned long offset; | |
8a9f3ccd | 1917 | int ret = 0; |
b56a2d8a | 1918 | volatile unsigned char *swap_map; |
1da177e4 | 1919 | |
b56a2d8a | 1920 | si = swap_info[type]; |
044d66c1 | 1921 | pte = pte_offset_map(pmd, addr); |
1da177e4 | 1922 | do { |
b56a2d8a VRP |
1923 | struct vm_fault vmf; |
1924 | ||
1925 | if (!is_swap_pte(*pte)) | |
1926 | continue; | |
1927 | ||
1928 | entry = pte_to_swp_entry(*pte); | |
1929 | if (swp_type(entry) != type) | |
1930 | continue; | |
1931 | ||
1932 | offset = swp_offset(entry); | |
1933 | if (frontswap && !frontswap_test(si, offset)) | |
1934 | continue; | |
1935 | ||
1936 | pte_unmap(pte); | |
1937 | swap_map = &si->swap_map[offset]; | |
ebc5951e AR |
1938 | page = lookup_swap_cache(entry, vma, addr); |
1939 | if (!page) { | |
1940 | vmf.vma = vma; | |
1941 | vmf.address = addr; | |
1942 | vmf.pmd = pmd; | |
1943 | page = swapin_readahead(entry, GFP_HIGHUSER_MOVABLE, | |
1944 | &vmf); | |
1945 | } | |
b56a2d8a VRP |
1946 | if (!page) { |
1947 | if (*swap_map == 0 || *swap_map == SWAP_MAP_BAD) | |
1948 | goto try_next; | |
1949 | return -ENOMEM; | |
1950 | } | |
1951 | ||
1952 | lock_page(page); | |
1953 | wait_on_page_writeback(page); | |
1954 | ret = unuse_pte(vma, pmd, addr, entry, page); | |
1955 | if (ret < 0) { | |
1956 | unlock_page(page); | |
1957 | put_page(page); | |
1958 | goto out; | |
1959 | } | |
1960 | ||
1961 | try_to_free_swap(page); | |
1962 | unlock_page(page); | |
1963 | put_page(page); | |
1964 | ||
1965 | if (*fs_pages_to_unuse && !--(*fs_pages_to_unuse)) { | |
1966 | ret = FRONTSWAP_PAGES_UNUSED; | |
1967 | goto out; | |
1da177e4 | 1968 | } |
b56a2d8a VRP |
1969 | try_next: |
1970 | pte = pte_offset_map(pmd, addr); | |
1da177e4 | 1971 | } while (pte++, addr += PAGE_SIZE, addr != end); |
044d66c1 | 1972 | pte_unmap(pte - 1); |
b56a2d8a VRP |
1973 | |
1974 | ret = 0; | |
044d66c1 | 1975 | out: |
8a9f3ccd | 1976 | return ret; |
1da177e4 LT |
1977 | } |
1978 | ||
1979 | static inline int unuse_pmd_range(struct vm_area_struct *vma, pud_t *pud, | |
1980 | unsigned long addr, unsigned long end, | |
b56a2d8a VRP |
1981 | unsigned int type, bool frontswap, |
1982 | unsigned long *fs_pages_to_unuse) | |
1da177e4 LT |
1983 | { |
1984 | pmd_t *pmd; | |
1985 | unsigned long next; | |
8a9f3ccd | 1986 | int ret; |
1da177e4 LT |
1987 | |
1988 | pmd = pmd_offset(pud, addr); | |
1989 | do { | |
dc644a07 | 1990 | cond_resched(); |
1da177e4 | 1991 | next = pmd_addr_end(addr, end); |
1a5a9906 | 1992 | if (pmd_none_or_trans_huge_or_clear_bad(pmd)) |
1da177e4 | 1993 | continue; |
b56a2d8a VRP |
1994 | ret = unuse_pte_range(vma, pmd, addr, next, type, |
1995 | frontswap, fs_pages_to_unuse); | |
8a9f3ccd BS |
1996 | if (ret) |
1997 | return ret; | |
1da177e4 LT |
1998 | } while (pmd++, addr = next, addr != end); |
1999 | return 0; | |
2000 | } | |
2001 | ||
c2febafc | 2002 | static inline int unuse_pud_range(struct vm_area_struct *vma, p4d_t *p4d, |
1da177e4 | 2003 | unsigned long addr, unsigned long end, |
b56a2d8a VRP |
2004 | unsigned int type, bool frontswap, |
2005 | unsigned long *fs_pages_to_unuse) | |
1da177e4 LT |
2006 | { |
2007 | pud_t *pud; | |
2008 | unsigned long next; | |
8a9f3ccd | 2009 | int ret; |
1da177e4 | 2010 | |
c2febafc | 2011 | pud = pud_offset(p4d, addr); |
1da177e4 LT |
2012 | do { |
2013 | next = pud_addr_end(addr, end); | |
2014 | if (pud_none_or_clear_bad(pud)) | |
2015 | continue; | |
b56a2d8a VRP |
2016 | ret = unuse_pmd_range(vma, pud, addr, next, type, |
2017 | frontswap, fs_pages_to_unuse); | |
8a9f3ccd BS |
2018 | if (ret) |
2019 | return ret; | |
1da177e4 LT |
2020 | } while (pud++, addr = next, addr != end); |
2021 | return 0; | |
2022 | } | |
2023 | ||
c2febafc KS |
2024 | static inline int unuse_p4d_range(struct vm_area_struct *vma, pgd_t *pgd, |
2025 | unsigned long addr, unsigned long end, | |
b56a2d8a VRP |
2026 | unsigned int type, bool frontswap, |
2027 | unsigned long *fs_pages_to_unuse) | |
c2febafc KS |
2028 | { |
2029 | p4d_t *p4d; | |
2030 | unsigned long next; | |
2031 | int ret; | |
2032 | ||
2033 | p4d = p4d_offset(pgd, addr); | |
2034 | do { | |
2035 | next = p4d_addr_end(addr, end); | |
2036 | if (p4d_none_or_clear_bad(p4d)) | |
2037 | continue; | |
b56a2d8a VRP |
2038 | ret = unuse_pud_range(vma, p4d, addr, next, type, |
2039 | frontswap, fs_pages_to_unuse); | |
c2febafc KS |
2040 | if (ret) |
2041 | return ret; | |
2042 | } while (p4d++, addr = next, addr != end); | |
2043 | return 0; | |
2044 | } | |
2045 | ||
b56a2d8a VRP |
2046 | static int unuse_vma(struct vm_area_struct *vma, unsigned int type, |
2047 | bool frontswap, unsigned long *fs_pages_to_unuse) | |
1da177e4 LT |
2048 | { |
2049 | pgd_t *pgd; | |
2050 | unsigned long addr, end, next; | |
8a9f3ccd | 2051 | int ret; |
1da177e4 | 2052 | |
b56a2d8a VRP |
2053 | addr = vma->vm_start; |
2054 | end = vma->vm_end; | |
1da177e4 LT |
2055 | |
2056 | pgd = pgd_offset(vma->vm_mm, addr); | |
2057 | do { | |
2058 | next = pgd_addr_end(addr, end); | |
2059 | if (pgd_none_or_clear_bad(pgd)) | |
2060 | continue; | |
b56a2d8a VRP |
2061 | ret = unuse_p4d_range(vma, pgd, addr, next, type, |
2062 | frontswap, fs_pages_to_unuse); | |
8a9f3ccd BS |
2063 | if (ret) |
2064 | return ret; | |
1da177e4 LT |
2065 | } while (pgd++, addr = next, addr != end); |
2066 | return 0; | |
2067 | } | |
2068 | ||
b56a2d8a VRP |
2069 | static int unuse_mm(struct mm_struct *mm, unsigned int type, |
2070 | bool frontswap, unsigned long *fs_pages_to_unuse) | |
1da177e4 LT |
2071 | { |
2072 | struct vm_area_struct *vma; | |
8a9f3ccd | 2073 | int ret = 0; |
1da177e4 | 2074 | |
b56a2d8a | 2075 | down_read(&mm->mmap_sem); |
1da177e4 | 2076 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
b56a2d8a VRP |
2077 | if (vma->anon_vma) { |
2078 | ret = unuse_vma(vma, type, frontswap, | |
2079 | fs_pages_to_unuse); | |
2080 | if (ret) | |
2081 | break; | |
2082 | } | |
dc644a07 | 2083 | cond_resched(); |
1da177e4 | 2084 | } |
1da177e4 | 2085 | up_read(&mm->mmap_sem); |
b56a2d8a | 2086 | return ret; |
1da177e4 LT |
2087 | } |
2088 | ||
2089 | /* | |
38b5faf4 | 2090 | * Scan swap_map (or frontswap_map if frontswap parameter is true) |
b56a2d8a VRP |
2091 | * from current position to next entry still in use. Return 0 |
2092 | * if there are no inuse entries after prev till end of the map. | |
1da177e4 | 2093 | */ |
6eb396dc | 2094 | static unsigned int find_next_to_unuse(struct swap_info_struct *si, |
38b5faf4 | 2095 | unsigned int prev, bool frontswap) |
1da177e4 | 2096 | { |
b56a2d8a | 2097 | unsigned int i; |
8d69aaee | 2098 | unsigned char count; |
1da177e4 LT |
2099 | |
2100 | /* | |
5d337b91 | 2101 | * No need for swap_lock here: we're just looking |
1da177e4 LT |
2102 | * for whether an entry is in use, not modifying it; false |
2103 | * hits are okay, and sys_swapoff() has already prevented new | |
5d337b91 | 2104 | * allocations from this area (while holding swap_lock). |
1da177e4 | 2105 | */ |
b56a2d8a | 2106 | for (i = prev + 1; i < si->max; i++) { |
4db0c3c2 | 2107 | count = READ_ONCE(si->swap_map[i]); |
355cfa73 | 2108 | if (count && swap_count(count) != SWAP_MAP_BAD) |
dc644a07 HD |
2109 | if (!frontswap || frontswap_test(si, i)) |
2110 | break; | |
2111 | if ((i % LATENCY_LIMIT) == 0) | |
2112 | cond_resched(); | |
1da177e4 | 2113 | } |
b56a2d8a VRP |
2114 | |
2115 | if (i == si->max) | |
2116 | i = 0; | |
2117 | ||
1da177e4 LT |
2118 | return i; |
2119 | } | |
2120 | ||
2121 | /* | |
b56a2d8a | 2122 | * If the boolean frontswap is true, only unuse pages_to_unuse pages; |
38b5faf4 | 2123 | * pages_to_unuse==0 means all pages; ignored if frontswap is false |
1da177e4 | 2124 | */ |
38b5faf4 DM |
2125 | int try_to_unuse(unsigned int type, bool frontswap, |
2126 | unsigned long pages_to_unuse) | |
1da177e4 | 2127 | { |
b56a2d8a VRP |
2128 | struct mm_struct *prev_mm; |
2129 | struct mm_struct *mm; | |
2130 | struct list_head *p; | |
2131 | int retval = 0; | |
efa90a98 | 2132 | struct swap_info_struct *si = swap_info[type]; |
1da177e4 LT |
2133 | struct page *page; |
2134 | swp_entry_t entry; | |
b56a2d8a | 2135 | unsigned int i; |
1da177e4 | 2136 | |
21820948 | 2137 | if (!READ_ONCE(si->inuse_pages)) |
b56a2d8a | 2138 | return 0; |
1da177e4 | 2139 | |
b56a2d8a VRP |
2140 | if (!frontswap) |
2141 | pages_to_unuse = 0; | |
2142 | ||
2143 | retry: | |
2144 | retval = shmem_unuse(type, frontswap, &pages_to_unuse); | |
2145 | if (retval) | |
2146 | goto out; | |
2147 | ||
2148 | prev_mm = &init_mm; | |
2149 | mmget(prev_mm); | |
2150 | ||
2151 | spin_lock(&mmlist_lock); | |
2152 | p = &init_mm.mmlist; | |
21820948 | 2153 | while (READ_ONCE(si->inuse_pages) && |
64165b1a HD |
2154 | !signal_pending(current) && |
2155 | (p = p->next) != &init_mm.mmlist) { | |
1da177e4 | 2156 | |
b56a2d8a VRP |
2157 | mm = list_entry(p, struct mm_struct, mmlist); |
2158 | if (!mmget_not_zero(mm)) | |
2159 | continue; | |
2160 | spin_unlock(&mmlist_lock); | |
2161 | mmput(prev_mm); | |
2162 | prev_mm = mm; | |
2163 | retval = unuse_mm(mm, type, frontswap, &pages_to_unuse); | |
1da177e4 | 2164 | |
b56a2d8a VRP |
2165 | if (retval) { |
2166 | mmput(prev_mm); | |
2167 | goto out; | |
1da177e4 LT |
2168 | } |
2169 | ||
2170 | /* | |
b56a2d8a VRP |
2171 | * Make sure that we aren't completely killing |
2172 | * interactive performance. | |
1da177e4 | 2173 | */ |
b56a2d8a VRP |
2174 | cond_resched(); |
2175 | spin_lock(&mmlist_lock); | |
2176 | } | |
2177 | spin_unlock(&mmlist_lock); | |
1da177e4 | 2178 | |
b56a2d8a | 2179 | mmput(prev_mm); |
1da177e4 | 2180 | |
b56a2d8a | 2181 | i = 0; |
21820948 | 2182 | while (READ_ONCE(si->inuse_pages) && |
64165b1a HD |
2183 | !signal_pending(current) && |
2184 | (i = find_next_to_unuse(si, i, frontswap)) != 0) { | |
1da177e4 | 2185 | |
b56a2d8a VRP |
2186 | entry = swp_entry(type, i); |
2187 | page = find_get_page(swap_address_space(entry), i); | |
2188 | if (!page) | |
2189 | continue; | |
68bdc8d6 HD |
2190 | |
2191 | /* | |
2192 | * It is conceivable that a racing task removed this page from | |
b56a2d8a VRP |
2193 | * swap cache just before we acquired the page lock. The page |
2194 | * might even be back in swap cache on another swap area. But | |
2195 | * that is okay, try_to_free_swap() only removes stale pages. | |
1da177e4 | 2196 | */ |
b56a2d8a VRP |
2197 | lock_page(page); |
2198 | wait_on_page_writeback(page); | |
2199 | try_to_free_swap(page); | |
1da177e4 | 2200 | unlock_page(page); |
09cbfeaf | 2201 | put_page(page); |
1da177e4 LT |
2202 | |
2203 | /* | |
b56a2d8a VRP |
2204 | * For frontswap, we just need to unuse pages_to_unuse, if |
2205 | * it was specified. Need not check frontswap again here as | |
2206 | * we already zeroed out pages_to_unuse if not frontswap. | |
1da177e4 | 2207 | */ |
b56a2d8a VRP |
2208 | if (pages_to_unuse && --pages_to_unuse == 0) |
2209 | goto out; | |
1da177e4 LT |
2210 | } |
2211 | ||
b56a2d8a VRP |
2212 | /* |
2213 | * Lets check again to see if there are still swap entries in the map. | |
2214 | * If yes, we would need to do retry the unuse logic again. | |
2215 | * Under global memory pressure, swap entries can be reinserted back | |
2216 | * into process space after the mmlist loop above passes over them. | |
dd862deb | 2217 | * |
af53d3e9 HD |
2218 | * Limit the number of retries? No: when mmget_not_zero() above fails, |
2219 | * that mm is likely to be freeing swap from exit_mmap(), which proceeds | |
2220 | * at its own independent pace; and even shmem_writepage() could have | |
2221 | * been preempted after get_swap_page(), temporarily hiding that swap. | |
2222 | * It's easy and robust (though cpu-intensive) just to keep retrying. | |
b56a2d8a | 2223 | */ |
21820948 | 2224 | if (READ_ONCE(si->inuse_pages)) { |
64165b1a HD |
2225 | if (!signal_pending(current)) |
2226 | goto retry; | |
2227 | retval = -EINTR; | |
2228 | } | |
b56a2d8a VRP |
2229 | out: |
2230 | return (retval == FRONTSWAP_PAGES_UNUSED) ? 0 : retval; | |
1da177e4 LT |
2231 | } |
2232 | ||
2233 | /* | |
5d337b91 HD |
2234 | * After a successful try_to_unuse, if no swap is now in use, we know |
2235 | * we can empty the mmlist. swap_lock must be held on entry and exit. | |
2236 | * Note that mmlist_lock nests inside swap_lock, and an mm must be | |
1da177e4 LT |
2237 | * added to the mmlist just after page_duplicate - before would be racy. |
2238 | */ | |
2239 | static void drain_mmlist(void) | |
2240 | { | |
2241 | struct list_head *p, *next; | |
efa90a98 | 2242 | unsigned int type; |
1da177e4 | 2243 | |
efa90a98 HD |
2244 | for (type = 0; type < nr_swapfiles; type++) |
2245 | if (swap_info[type]->inuse_pages) | |
1da177e4 LT |
2246 | return; |
2247 | spin_lock(&mmlist_lock); | |
2248 | list_for_each_safe(p, next, &init_mm.mmlist) | |
2249 | list_del_init(p); | |
2250 | spin_unlock(&mmlist_lock); | |
2251 | } | |
2252 | ||
2253 | /* | |
2254 | * Use this swapdev's extent info to locate the (PAGE_SIZE) block which | |
d4906e1a LS |
2255 | * corresponds to page offset for the specified swap entry. |
2256 | * Note that the type of this function is sector_t, but it returns page offset | |
2257 | * into the bdev, not sector offset. | |
1da177e4 | 2258 | */ |
d4906e1a | 2259 | static sector_t map_swap_entry(swp_entry_t entry, struct block_device **bdev) |
1da177e4 | 2260 | { |
f29ad6a9 | 2261 | struct swap_info_struct *sis; |
f29ad6a9 HD |
2262 | struct swap_extent *se; |
2263 | pgoff_t offset; | |
2264 | ||
c10d38cc | 2265 | sis = swp_swap_info(entry); |
f29ad6a9 HD |
2266 | *bdev = sis->bdev; |
2267 | ||
2268 | offset = swp_offset(entry); | |
4efaceb1 AL |
2269 | se = offset_to_swap_extent(sis, offset); |
2270 | return se->start_block + (offset - se->start_page); | |
1da177e4 LT |
2271 | } |
2272 | ||
d4906e1a LS |
2273 | /* |
2274 | * Returns the page offset into bdev for the specified page's swap entry. | |
2275 | */ | |
2276 | sector_t map_swap_page(struct page *page, struct block_device **bdev) | |
2277 | { | |
2278 | swp_entry_t entry; | |
2279 | entry.val = page_private(page); | |
2280 | return map_swap_entry(entry, bdev); | |
2281 | } | |
2282 | ||
1da177e4 LT |
2283 | /* |
2284 | * Free all of a swapdev's extent information | |
2285 | */ | |
2286 | static void destroy_swap_extents(struct swap_info_struct *sis) | |
2287 | { | |
4efaceb1 AL |
2288 | while (!RB_EMPTY_ROOT(&sis->swap_extent_root)) { |
2289 | struct rb_node *rb = sis->swap_extent_root.rb_node; | |
2290 | struct swap_extent *se = rb_entry(rb, struct swap_extent, rb_node); | |
1da177e4 | 2291 | |
4efaceb1 | 2292 | rb_erase(rb, &sis->swap_extent_root); |
1da177e4 LT |
2293 | kfree(se); |
2294 | } | |
62c230bc | 2295 | |
bc4ae27d | 2296 | if (sis->flags & SWP_ACTIVATED) { |
62c230bc MG |
2297 | struct file *swap_file = sis->swap_file; |
2298 | struct address_space *mapping = swap_file->f_mapping; | |
2299 | ||
bc4ae27d OS |
2300 | sis->flags &= ~SWP_ACTIVATED; |
2301 | if (mapping->a_ops->swap_deactivate) | |
2302 | mapping->a_ops->swap_deactivate(swap_file); | |
62c230bc | 2303 | } |
1da177e4 LT |
2304 | } |
2305 | ||
2306 | /* | |
2307 | * Add a block range (and the corresponding page range) into this swapdev's | |
4efaceb1 | 2308 | * extent tree. |
1da177e4 | 2309 | * |
11d31886 | 2310 | * This function rather assumes that it is called in ascending page order. |
1da177e4 | 2311 | */ |
a509bc1a | 2312 | int |
1da177e4 LT |
2313 | add_swap_extent(struct swap_info_struct *sis, unsigned long start_page, |
2314 | unsigned long nr_pages, sector_t start_block) | |
2315 | { | |
4efaceb1 | 2316 | struct rb_node **link = &sis->swap_extent_root.rb_node, *parent = NULL; |
1da177e4 LT |
2317 | struct swap_extent *se; |
2318 | struct swap_extent *new_se; | |
4efaceb1 AL |
2319 | |
2320 | /* | |
2321 | * place the new node at the right most since the | |
2322 | * function is called in ascending page order. | |
2323 | */ | |
2324 | while (*link) { | |
2325 | parent = *link; | |
2326 | link = &parent->rb_right; | |
2327 | } | |
2328 | ||
2329 | if (parent) { | |
2330 | se = rb_entry(parent, struct swap_extent, rb_node); | |
11d31886 HD |
2331 | BUG_ON(se->start_page + se->nr_pages != start_page); |
2332 | if (se->start_block + se->nr_pages == start_block) { | |
1da177e4 LT |
2333 | /* Merge it */ |
2334 | se->nr_pages += nr_pages; | |
2335 | return 0; | |
2336 | } | |
1da177e4 LT |
2337 | } |
2338 | ||
4efaceb1 | 2339 | /* No merge, insert a new extent. */ |
1da177e4 LT |
2340 | new_se = kmalloc(sizeof(*se), GFP_KERNEL); |
2341 | if (new_se == NULL) | |
2342 | return -ENOMEM; | |
2343 | new_se->start_page = start_page; | |
2344 | new_se->nr_pages = nr_pages; | |
2345 | new_se->start_block = start_block; | |
2346 | ||
4efaceb1 AL |
2347 | rb_link_node(&new_se->rb_node, parent, link); |
2348 | rb_insert_color(&new_se->rb_node, &sis->swap_extent_root); | |
53092a74 | 2349 | return 1; |
1da177e4 | 2350 | } |
aa8aa8a3 | 2351 | EXPORT_SYMBOL_GPL(add_swap_extent); |
1da177e4 LT |
2352 | |
2353 | /* | |
2354 | * A `swap extent' is a simple thing which maps a contiguous range of pages | |
2355 | * onto a contiguous range of disk blocks. An ordered list of swap extents | |
2356 | * is built at swapon time and is then used at swap_writepage/swap_readpage | |
2357 | * time for locating where on disk a page belongs. | |
2358 | * | |
2359 | * If the swapfile is an S_ISBLK block device, a single extent is installed. | |
2360 | * This is done so that the main operating code can treat S_ISBLK and S_ISREG | |
2361 | * swap files identically. | |
2362 | * | |
2363 | * Whether the swapdev is an S_ISREG file or an S_ISBLK blockdev, the swap | |
2364 | * extent list operates in PAGE_SIZE disk blocks. Both S_ISREG and S_ISBLK | |
2365 | * swapfiles are handled *identically* after swapon time. | |
2366 | * | |
2367 | * For S_ISREG swapfiles, setup_swap_extents() will walk all the file's blocks | |
2368 | * and will parse them into an ordered extent list, in PAGE_SIZE chunks. If | |
2369 | * some stray blocks are found which do not fall within the PAGE_SIZE alignment | |
2370 | * requirements, they are simply tossed out - we will never use those blocks | |
2371 | * for swapping. | |
2372 | * | |
1638045c DW |
2373 | * For all swap devices we set S_SWAPFILE across the life of the swapon. This |
2374 | * prevents users from writing to the swap device, which will corrupt memory. | |
1da177e4 LT |
2375 | * |
2376 | * The amount of disk space which a single swap extent represents varies. | |
2377 | * Typically it is in the 1-4 megabyte range. So we can have hundreds of | |
2378 | * extents in the list. To avoid much list walking, we cache the previous | |
2379 | * search location in `curr_swap_extent', and start new searches from there. | |
2380 | * This is extremely effective. The average number of iterations in | |
2381 | * map_swap_page() has been measured at about 0.3 per page. - akpm. | |
2382 | */ | |
53092a74 | 2383 | static int setup_swap_extents(struct swap_info_struct *sis, sector_t *span) |
1da177e4 | 2384 | { |
62c230bc MG |
2385 | struct file *swap_file = sis->swap_file; |
2386 | struct address_space *mapping = swap_file->f_mapping; | |
2387 | struct inode *inode = mapping->host; | |
1da177e4 LT |
2388 | int ret; |
2389 | ||
1da177e4 LT |
2390 | if (S_ISBLK(inode->i_mode)) { |
2391 | ret = add_swap_extent(sis, 0, sis->max, 0); | |
53092a74 | 2392 | *span = sis->pages; |
a509bc1a | 2393 | return ret; |
1da177e4 LT |
2394 | } |
2395 | ||
62c230bc | 2396 | if (mapping->a_ops->swap_activate) { |
a509bc1a | 2397 | ret = mapping->a_ops->swap_activate(sis, swap_file, span); |
bc4ae27d OS |
2398 | if (ret >= 0) |
2399 | sis->flags |= SWP_ACTIVATED; | |
62c230bc | 2400 | if (!ret) { |
bc4ae27d | 2401 | sis->flags |= SWP_FS; |
62c230bc MG |
2402 | ret = add_swap_extent(sis, 0, sis->max, 0); |
2403 | *span = sis->pages; | |
2404 | } | |
a509bc1a | 2405 | return ret; |
62c230bc MG |
2406 | } |
2407 | ||
a509bc1a | 2408 | return generic_swapfile_activate(sis, swap_file, span); |
1da177e4 LT |
2409 | } |
2410 | ||
a2468cc9 AL |
2411 | static int swap_node(struct swap_info_struct *p) |
2412 | { | |
2413 | struct block_device *bdev; | |
2414 | ||
2415 | if (p->bdev) | |
2416 | bdev = p->bdev; | |
2417 | else | |
2418 | bdev = p->swap_file->f_inode->i_sb->s_bdev; | |
2419 | ||
2420 | return bdev ? bdev->bd_disk->node_id : NUMA_NO_NODE; | |
2421 | } | |
2422 | ||
eb085574 HY |
2423 | static void setup_swap_info(struct swap_info_struct *p, int prio, |
2424 | unsigned char *swap_map, | |
2425 | struct swap_cluster_info *cluster_info) | |
40531542 | 2426 | { |
a2468cc9 AL |
2427 | int i; |
2428 | ||
40531542 CEB |
2429 | if (prio >= 0) |
2430 | p->prio = prio; | |
2431 | else | |
2432 | p->prio = --least_priority; | |
18ab4d4c DS |
2433 | /* |
2434 | * the plist prio is negated because plist ordering is | |
2435 | * low-to-high, while swap ordering is high-to-low | |
2436 | */ | |
2437 | p->list.prio = -p->prio; | |
a2468cc9 AL |
2438 | for_each_node(i) { |
2439 | if (p->prio >= 0) | |
2440 | p->avail_lists[i].prio = -p->prio; | |
2441 | else { | |
2442 | if (swap_node(p) == i) | |
2443 | p->avail_lists[i].prio = 1; | |
2444 | else | |
2445 | p->avail_lists[i].prio = -p->prio; | |
2446 | } | |
2447 | } | |
40531542 | 2448 | p->swap_map = swap_map; |
2a8f9449 | 2449 | p->cluster_info = cluster_info; |
eb085574 HY |
2450 | } |
2451 | ||
2452 | static void _enable_swap_info(struct swap_info_struct *p) | |
2453 | { | |
2454 | p->flags |= SWP_WRITEOK | SWP_VALID; | |
ec8acf20 | 2455 | atomic_long_add(p->pages, &nr_swap_pages); |
40531542 CEB |
2456 | total_swap_pages += p->pages; |
2457 | ||
adfab836 | 2458 | assert_spin_locked(&swap_lock); |
adfab836 | 2459 | /* |
18ab4d4c DS |
2460 | * both lists are plists, and thus priority ordered. |
2461 | * swap_active_head needs to be priority ordered for swapoff(), | |
2462 | * which on removal of any swap_info_struct with an auto-assigned | |
2463 | * (i.e. negative) priority increments the auto-assigned priority | |
2464 | * of any lower-priority swap_info_structs. | |
2465 | * swap_avail_head needs to be priority ordered for get_swap_page(), | |
2466 | * which allocates swap pages from the highest available priority | |
2467 | * swap_info_struct. | |
adfab836 | 2468 | */ |
18ab4d4c | 2469 | plist_add(&p->list, &swap_active_head); |
a2468cc9 | 2470 | add_to_avail_list(p); |
cf0cac0a CEB |
2471 | } |
2472 | ||
2473 | static void enable_swap_info(struct swap_info_struct *p, int prio, | |
2474 | unsigned char *swap_map, | |
2a8f9449 | 2475 | struct swap_cluster_info *cluster_info, |
cf0cac0a CEB |
2476 | unsigned long *frontswap_map) |
2477 | { | |
4f89849d | 2478 | frontswap_init(p->type, frontswap_map); |
cf0cac0a | 2479 | spin_lock(&swap_lock); |
ec8acf20 | 2480 | spin_lock(&p->lock); |
eb085574 HY |
2481 | setup_swap_info(p, prio, swap_map, cluster_info); |
2482 | spin_unlock(&p->lock); | |
2483 | spin_unlock(&swap_lock); | |
2484 | /* | |
2485 | * Guarantee swap_map, cluster_info, etc. fields are valid | |
2486 | * between get/put_swap_device() if SWP_VALID bit is set | |
2487 | */ | |
2488 | synchronize_rcu(); | |
2489 | spin_lock(&swap_lock); | |
2490 | spin_lock(&p->lock); | |
2491 | _enable_swap_info(p); | |
ec8acf20 | 2492 | spin_unlock(&p->lock); |
cf0cac0a CEB |
2493 | spin_unlock(&swap_lock); |
2494 | } | |
2495 | ||
2496 | static void reinsert_swap_info(struct swap_info_struct *p) | |
2497 | { | |
2498 | spin_lock(&swap_lock); | |
ec8acf20 | 2499 | spin_lock(&p->lock); |
eb085574 HY |
2500 | setup_swap_info(p, p->prio, p->swap_map, p->cluster_info); |
2501 | _enable_swap_info(p); | |
ec8acf20 | 2502 | spin_unlock(&p->lock); |
40531542 CEB |
2503 | spin_unlock(&swap_lock); |
2504 | } | |
2505 | ||
67afa38e TC |
2506 | bool has_usable_swap(void) |
2507 | { | |
2508 | bool ret = true; | |
2509 | ||
2510 | spin_lock(&swap_lock); | |
2511 | if (plist_head_empty(&swap_active_head)) | |
2512 | ret = false; | |
2513 | spin_unlock(&swap_lock); | |
2514 | return ret; | |
2515 | } | |
2516 | ||
c4ea37c2 | 2517 | SYSCALL_DEFINE1(swapoff, const char __user *, specialfile) |
1da177e4 | 2518 | { |
73c34b6a | 2519 | struct swap_info_struct *p = NULL; |
8d69aaee | 2520 | unsigned char *swap_map; |
2a8f9449 | 2521 | struct swap_cluster_info *cluster_info; |
4f89849d | 2522 | unsigned long *frontswap_map; |
1da177e4 LT |
2523 | struct file *swap_file, *victim; |
2524 | struct address_space *mapping; | |
2525 | struct inode *inode; | |
91a27b2a | 2526 | struct filename *pathname; |
adfab836 | 2527 | int err, found = 0; |
5b808a23 | 2528 | unsigned int old_block_size; |
886bb7e9 | 2529 | |
1da177e4 LT |
2530 | if (!capable(CAP_SYS_ADMIN)) |
2531 | return -EPERM; | |
2532 | ||
191c5424 AV |
2533 | BUG_ON(!current->mm); |
2534 | ||
1da177e4 | 2535 | pathname = getname(specialfile); |
1da177e4 | 2536 | if (IS_ERR(pathname)) |
f58b59c1 | 2537 | return PTR_ERR(pathname); |
1da177e4 | 2538 | |
669abf4e | 2539 | victim = file_open_name(pathname, O_RDWR|O_LARGEFILE, 0); |
1da177e4 LT |
2540 | err = PTR_ERR(victim); |
2541 | if (IS_ERR(victim)) | |
2542 | goto out; | |
2543 | ||
2544 | mapping = victim->f_mapping; | |
5d337b91 | 2545 | spin_lock(&swap_lock); |
18ab4d4c | 2546 | plist_for_each_entry(p, &swap_active_head, list) { |
22c6f8fd | 2547 | if (p->flags & SWP_WRITEOK) { |
adfab836 DS |
2548 | if (p->swap_file->f_mapping == mapping) { |
2549 | found = 1; | |
1da177e4 | 2550 | break; |
adfab836 | 2551 | } |
1da177e4 | 2552 | } |
1da177e4 | 2553 | } |
adfab836 | 2554 | if (!found) { |
1da177e4 | 2555 | err = -EINVAL; |
5d337b91 | 2556 | spin_unlock(&swap_lock); |
1da177e4 LT |
2557 | goto out_dput; |
2558 | } | |
191c5424 | 2559 | if (!security_vm_enough_memory_mm(current->mm, p->pages)) |
1da177e4 LT |
2560 | vm_unacct_memory(p->pages); |
2561 | else { | |
2562 | err = -ENOMEM; | |
5d337b91 | 2563 | spin_unlock(&swap_lock); |
1da177e4 LT |
2564 | goto out_dput; |
2565 | } | |
a2468cc9 | 2566 | del_from_avail_list(p); |
ec8acf20 | 2567 | spin_lock(&p->lock); |
78ecba08 | 2568 | if (p->prio < 0) { |
adfab836 | 2569 | struct swap_info_struct *si = p; |
a2468cc9 | 2570 | int nid; |
adfab836 | 2571 | |
18ab4d4c | 2572 | plist_for_each_entry_continue(si, &swap_active_head, list) { |
adfab836 | 2573 | si->prio++; |
18ab4d4c | 2574 | si->list.prio--; |
a2468cc9 AL |
2575 | for_each_node(nid) { |
2576 | if (si->avail_lists[nid].prio != 1) | |
2577 | si->avail_lists[nid].prio--; | |
2578 | } | |
adfab836 | 2579 | } |
78ecba08 HD |
2580 | least_priority++; |
2581 | } | |
18ab4d4c | 2582 | plist_del(&p->list, &swap_active_head); |
ec8acf20 | 2583 | atomic_long_sub(p->pages, &nr_swap_pages); |
1da177e4 LT |
2584 | total_swap_pages -= p->pages; |
2585 | p->flags &= ~SWP_WRITEOK; | |
ec8acf20 | 2586 | spin_unlock(&p->lock); |
5d337b91 | 2587 | spin_unlock(&swap_lock); |
fb4f88dc | 2588 | |
039939a6 TC |
2589 | disable_swap_slots_cache_lock(); |
2590 | ||
e1e12d2f | 2591 | set_current_oom_origin(); |
adfab836 | 2592 | err = try_to_unuse(p->type, false, 0); /* force unuse all pages */ |
e1e12d2f | 2593 | clear_current_oom_origin(); |
1da177e4 | 2594 | |
1da177e4 LT |
2595 | if (err) { |
2596 | /* re-insert swap space back into swap_list */ | |
cf0cac0a | 2597 | reinsert_swap_info(p); |
039939a6 | 2598 | reenable_swap_slots_cache_unlock(); |
1da177e4 LT |
2599 | goto out_dput; |
2600 | } | |
52b7efdb | 2601 | |
039939a6 TC |
2602 | reenable_swap_slots_cache_unlock(); |
2603 | ||
eb085574 HY |
2604 | spin_lock(&swap_lock); |
2605 | spin_lock(&p->lock); | |
2606 | p->flags &= ~SWP_VALID; /* mark swap device as invalid */ | |
2607 | spin_unlock(&p->lock); | |
2608 | spin_unlock(&swap_lock); | |
2609 | /* | |
2610 | * wait for swap operations protected by get/put_swap_device() | |
2611 | * to complete | |
2612 | */ | |
2613 | synchronize_rcu(); | |
2614 | ||
815c2c54 SL |
2615 | flush_work(&p->discard_work); |
2616 | ||
5d337b91 | 2617 | destroy_swap_extents(p); |
570a335b HD |
2618 | if (p->flags & SWP_CONTINUED) |
2619 | free_swap_count_continuations(p); | |
2620 | ||
81a0298b HY |
2621 | if (!p->bdev || !blk_queue_nonrot(bdev_get_queue(p->bdev))) |
2622 | atomic_dec(&nr_rotate_swap); | |
2623 | ||
fc0abb14 | 2624 | mutex_lock(&swapon_mutex); |
5d337b91 | 2625 | spin_lock(&swap_lock); |
ec8acf20 | 2626 | spin_lock(&p->lock); |
5d337b91 HD |
2627 | drain_mmlist(); |
2628 | ||
52b7efdb | 2629 | /* wait for anyone still in scan_swap_map */ |
52b7efdb HD |
2630 | p->highest_bit = 0; /* cuts scans short */ |
2631 | while (p->flags >= SWP_SCANNING) { | |
ec8acf20 | 2632 | spin_unlock(&p->lock); |
5d337b91 | 2633 | spin_unlock(&swap_lock); |
13e4b57f | 2634 | schedule_timeout_uninterruptible(1); |
5d337b91 | 2635 | spin_lock(&swap_lock); |
ec8acf20 | 2636 | spin_lock(&p->lock); |
52b7efdb | 2637 | } |
52b7efdb | 2638 | |
1da177e4 | 2639 | swap_file = p->swap_file; |
5b808a23 | 2640 | old_block_size = p->old_block_size; |
1da177e4 LT |
2641 | p->swap_file = NULL; |
2642 | p->max = 0; | |
2643 | swap_map = p->swap_map; | |
2644 | p->swap_map = NULL; | |
2a8f9449 SL |
2645 | cluster_info = p->cluster_info; |
2646 | p->cluster_info = NULL; | |
4f89849d | 2647 | frontswap_map = frontswap_map_get(p); |
ec8acf20 | 2648 | spin_unlock(&p->lock); |
5d337b91 | 2649 | spin_unlock(&swap_lock); |
adfab836 | 2650 | frontswap_invalidate_area(p->type); |
58e97ba6 | 2651 | frontswap_map_set(p, NULL); |
fc0abb14 | 2652 | mutex_unlock(&swapon_mutex); |
ebc2a1a6 SL |
2653 | free_percpu(p->percpu_cluster); |
2654 | p->percpu_cluster = NULL; | |
1da177e4 | 2655 | vfree(swap_map); |
54f180d3 HY |
2656 | kvfree(cluster_info); |
2657 | kvfree(frontswap_map); | |
2de1a7e4 | 2658 | /* Destroy swap account information */ |
adfab836 | 2659 | swap_cgroup_swapoff(p->type); |
4b3ef9da | 2660 | exit_swap_address_space(p->type); |
27a7faa0 | 2661 | |
1da177e4 LT |
2662 | inode = mapping->host; |
2663 | if (S_ISBLK(inode->i_mode)) { | |
2664 | struct block_device *bdev = I_BDEV(inode); | |
1638045c | 2665 | |
5b808a23 | 2666 | set_blocksize(bdev, old_block_size); |
e525fd89 | 2667 | blkdev_put(bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL); |
1da177e4 | 2668 | } |
1638045c DW |
2669 | |
2670 | inode_lock(inode); | |
2671 | inode->i_flags &= ~S_SWAPFILE; | |
2672 | inode_unlock(inode); | |
1da177e4 | 2673 | filp_close(swap_file, NULL); |
f893ab41 WY |
2674 | |
2675 | /* | |
2676 | * Clear the SWP_USED flag after all resources are freed so that swapon | |
2677 | * can reuse this swap_info in alloc_swap_info() safely. It is ok to | |
2678 | * not hold p->lock after we cleared its SWP_WRITEOK. | |
2679 | */ | |
2680 | spin_lock(&swap_lock); | |
2681 | p->flags = 0; | |
2682 | spin_unlock(&swap_lock); | |
2683 | ||
1da177e4 | 2684 | err = 0; |
66d7dd51 KS |
2685 | atomic_inc(&proc_poll_event); |
2686 | wake_up_interruptible(&proc_poll_wait); | |
1da177e4 LT |
2687 | |
2688 | out_dput: | |
2689 | filp_close(victim, NULL); | |
2690 | out: | |
f58b59c1 | 2691 | putname(pathname); |
1da177e4 LT |
2692 | return err; |
2693 | } | |
2694 | ||
2695 | #ifdef CONFIG_PROC_FS | |
9dd95748 | 2696 | static __poll_t swaps_poll(struct file *file, poll_table *wait) |
66d7dd51 | 2697 | { |
f1514638 | 2698 | struct seq_file *seq = file->private_data; |
66d7dd51 KS |
2699 | |
2700 | poll_wait(file, &proc_poll_wait, wait); | |
2701 | ||
f1514638 KS |
2702 | if (seq->poll_event != atomic_read(&proc_poll_event)) { |
2703 | seq->poll_event = atomic_read(&proc_poll_event); | |
a9a08845 | 2704 | return EPOLLIN | EPOLLRDNORM | EPOLLERR | EPOLLPRI; |
66d7dd51 KS |
2705 | } |
2706 | ||
a9a08845 | 2707 | return EPOLLIN | EPOLLRDNORM; |
66d7dd51 KS |
2708 | } |
2709 | ||
1da177e4 LT |
2710 | /* iterator */ |
2711 | static void *swap_start(struct seq_file *swap, loff_t *pos) | |
2712 | { | |
efa90a98 HD |
2713 | struct swap_info_struct *si; |
2714 | int type; | |
1da177e4 LT |
2715 | loff_t l = *pos; |
2716 | ||
fc0abb14 | 2717 | mutex_lock(&swapon_mutex); |
1da177e4 | 2718 | |
881e4aab SS |
2719 | if (!l) |
2720 | return SEQ_START_TOKEN; | |
2721 | ||
c10d38cc | 2722 | for (type = 0; (si = swap_type_to_swap_info(type)); type++) { |
efa90a98 | 2723 | if (!(si->flags & SWP_USED) || !si->swap_map) |
1da177e4 | 2724 | continue; |
881e4aab | 2725 | if (!--l) |
efa90a98 | 2726 | return si; |
1da177e4 LT |
2727 | } |
2728 | ||
2729 | return NULL; | |
2730 | } | |
2731 | ||
2732 | static void *swap_next(struct seq_file *swap, void *v, loff_t *pos) | |
2733 | { | |
efa90a98 HD |
2734 | struct swap_info_struct *si = v; |
2735 | int type; | |
1da177e4 | 2736 | |
881e4aab | 2737 | if (v == SEQ_START_TOKEN) |
efa90a98 HD |
2738 | type = 0; |
2739 | else | |
2740 | type = si->type + 1; | |
881e4aab | 2741 | |
10c8d69f | 2742 | ++(*pos); |
c10d38cc | 2743 | for (; (si = swap_type_to_swap_info(type)); type++) { |
efa90a98 | 2744 | if (!(si->flags & SWP_USED) || !si->swap_map) |
1da177e4 | 2745 | continue; |
efa90a98 | 2746 | return si; |
1da177e4 LT |
2747 | } |
2748 | ||
2749 | return NULL; | |
2750 | } | |
2751 | ||
2752 | static void swap_stop(struct seq_file *swap, void *v) | |
2753 | { | |
fc0abb14 | 2754 | mutex_unlock(&swapon_mutex); |
1da177e4 LT |
2755 | } |
2756 | ||
2757 | static int swap_show(struct seq_file *swap, void *v) | |
2758 | { | |
efa90a98 | 2759 | struct swap_info_struct *si = v; |
1da177e4 LT |
2760 | struct file *file; |
2761 | int len; | |
2762 | ||
efa90a98 | 2763 | if (si == SEQ_START_TOKEN) { |
881e4aab SS |
2764 | seq_puts(swap,"Filename\t\t\t\tType\t\tSize\tUsed\tPriority\n"); |
2765 | return 0; | |
2766 | } | |
1da177e4 | 2767 | |
efa90a98 | 2768 | file = si->swap_file; |
2726d566 | 2769 | len = seq_file_path(swap, file, " \t\n\\"); |
6eb396dc | 2770 | seq_printf(swap, "%*s%s\t%u\t%u\t%d\n", |
886bb7e9 | 2771 | len < 40 ? 40 - len : 1, " ", |
496ad9aa | 2772 | S_ISBLK(file_inode(file)->i_mode) ? |
1da177e4 | 2773 | "partition" : "file\t", |
efa90a98 HD |
2774 | si->pages << (PAGE_SHIFT - 10), |
2775 | si->inuse_pages << (PAGE_SHIFT - 10), | |
2776 | si->prio); | |
1da177e4 LT |
2777 | return 0; |
2778 | } | |
2779 | ||
15ad7cdc | 2780 | static const struct seq_operations swaps_op = { |
1da177e4 LT |
2781 | .start = swap_start, |
2782 | .next = swap_next, | |
2783 | .stop = swap_stop, | |
2784 | .show = swap_show | |
2785 | }; | |
2786 | ||
2787 | static int swaps_open(struct inode *inode, struct file *file) | |
2788 | { | |
f1514638 | 2789 | struct seq_file *seq; |
66d7dd51 KS |
2790 | int ret; |
2791 | ||
66d7dd51 | 2792 | ret = seq_open(file, &swaps_op); |
f1514638 | 2793 | if (ret) |
66d7dd51 | 2794 | return ret; |
66d7dd51 | 2795 | |
f1514638 KS |
2796 | seq = file->private_data; |
2797 | seq->poll_event = atomic_read(&proc_poll_event); | |
2798 | return 0; | |
1da177e4 LT |
2799 | } |
2800 | ||
97a32539 | 2801 | static const struct proc_ops swaps_proc_ops = { |
d919b33d | 2802 | .proc_flags = PROC_ENTRY_PERMANENT, |
97a32539 AD |
2803 | .proc_open = swaps_open, |
2804 | .proc_read = seq_read, | |
2805 | .proc_lseek = seq_lseek, | |
2806 | .proc_release = seq_release, | |
2807 | .proc_poll = swaps_poll, | |
1da177e4 LT |
2808 | }; |
2809 | ||
2810 | static int __init procswaps_init(void) | |
2811 | { | |
97a32539 | 2812 | proc_create("swaps", 0, NULL, &swaps_proc_ops); |
1da177e4 LT |
2813 | return 0; |
2814 | } | |
2815 | __initcall(procswaps_init); | |
2816 | #endif /* CONFIG_PROC_FS */ | |
2817 | ||
1796316a JB |
2818 | #ifdef MAX_SWAPFILES_CHECK |
2819 | static int __init max_swapfiles_check(void) | |
2820 | { | |
2821 | MAX_SWAPFILES_CHECK(); | |
2822 | return 0; | |
2823 | } | |
2824 | late_initcall(max_swapfiles_check); | |
2825 | #endif | |
2826 | ||
53cbb243 | 2827 | static struct swap_info_struct *alloc_swap_info(void) |
1da177e4 | 2828 | { |
73c34b6a | 2829 | struct swap_info_struct *p; |
1da177e4 | 2830 | unsigned int type; |
a2468cc9 | 2831 | int i; |
efa90a98 | 2832 | |
96008744 | 2833 | p = kvzalloc(struct_size(p, avail_lists, nr_node_ids), GFP_KERNEL); |
efa90a98 | 2834 | if (!p) |
53cbb243 | 2835 | return ERR_PTR(-ENOMEM); |
efa90a98 | 2836 | |
5d337b91 | 2837 | spin_lock(&swap_lock); |
efa90a98 HD |
2838 | for (type = 0; type < nr_swapfiles; type++) { |
2839 | if (!(swap_info[type]->flags & SWP_USED)) | |
1da177e4 | 2840 | break; |
efa90a98 | 2841 | } |
0697212a | 2842 | if (type >= MAX_SWAPFILES) { |
5d337b91 | 2843 | spin_unlock(&swap_lock); |
873d7bcf | 2844 | kvfree(p); |
730c0581 | 2845 | return ERR_PTR(-EPERM); |
1da177e4 | 2846 | } |
efa90a98 HD |
2847 | if (type >= nr_swapfiles) { |
2848 | p->type = type; | |
c10d38cc | 2849 | WRITE_ONCE(swap_info[type], p); |
efa90a98 HD |
2850 | /* |
2851 | * Write swap_info[type] before nr_swapfiles, in case a | |
2852 | * racing procfs swap_start() or swap_next() is reading them. | |
2853 | * (We never shrink nr_swapfiles, we never free this entry.) | |
2854 | */ | |
2855 | smp_wmb(); | |
c10d38cc | 2856 | WRITE_ONCE(nr_swapfiles, nr_swapfiles + 1); |
efa90a98 | 2857 | } else { |
873d7bcf | 2858 | kvfree(p); |
efa90a98 HD |
2859 | p = swap_info[type]; |
2860 | /* | |
2861 | * Do not memset this entry: a racing procfs swap_next() | |
2862 | * would be relying on p->type to remain valid. | |
2863 | */ | |
2864 | } | |
4efaceb1 | 2865 | p->swap_extent_root = RB_ROOT; |
18ab4d4c | 2866 | plist_node_init(&p->list, 0); |
a2468cc9 AL |
2867 | for_each_node(i) |
2868 | plist_node_init(&p->avail_lists[i], 0); | |
1da177e4 | 2869 | p->flags = SWP_USED; |
5d337b91 | 2870 | spin_unlock(&swap_lock); |
ec8acf20 | 2871 | spin_lock_init(&p->lock); |
2628bd6f | 2872 | spin_lock_init(&p->cont_lock); |
efa90a98 | 2873 | |
53cbb243 | 2874 | return p; |
53cbb243 CEB |
2875 | } |
2876 | ||
4d0e1e10 CEB |
2877 | static int claim_swapfile(struct swap_info_struct *p, struct inode *inode) |
2878 | { | |
2879 | int error; | |
2880 | ||
2881 | if (S_ISBLK(inode->i_mode)) { | |
2882 | p->bdev = bdgrab(I_BDEV(inode)); | |
2883 | error = blkdev_get(p->bdev, | |
6f179af8 | 2884 | FMODE_READ | FMODE_WRITE | FMODE_EXCL, p); |
4d0e1e10 CEB |
2885 | if (error < 0) { |
2886 | p->bdev = NULL; | |
6f179af8 | 2887 | return error; |
4d0e1e10 CEB |
2888 | } |
2889 | p->old_block_size = block_size(p->bdev); | |
2890 | error = set_blocksize(p->bdev, PAGE_SIZE); | |
2891 | if (error < 0) | |
87ade72a | 2892 | return error; |
12d2966d NA |
2893 | /* |
2894 | * Zoned block devices contain zones that have a sequential | |
2895 | * write only restriction. Hence zoned block devices are not | |
2896 | * suitable for swapping. Disallow them here. | |
2897 | */ | |
2898 | if (blk_queue_is_zoned(p->bdev->bd_queue)) | |
2899 | return -EINVAL; | |
4d0e1e10 CEB |
2900 | p->flags |= SWP_BLKDEV; |
2901 | } else if (S_ISREG(inode->i_mode)) { | |
2902 | p->bdev = inode->i_sb->s_bdev; | |
1638045c DW |
2903 | } |
2904 | ||
4d0e1e10 | 2905 | return 0; |
4d0e1e10 CEB |
2906 | } |
2907 | ||
377eeaa8 AK |
2908 | |
2909 | /* | |
2910 | * Find out how many pages are allowed for a single swap device. There | |
2911 | * are two limiting factors: | |
2912 | * 1) the number of bits for the swap offset in the swp_entry_t type, and | |
2913 | * 2) the number of bits in the swap pte, as defined by the different | |
2914 | * architectures. | |
2915 | * | |
2916 | * In order to find the largest possible bit mask, a swap entry with | |
2917 | * swap type 0 and swap offset ~0UL is created, encoded to a swap pte, | |
2918 | * decoded to a swp_entry_t again, and finally the swap offset is | |
2919 | * extracted. | |
2920 | * | |
2921 | * This will mask all the bits from the initial ~0UL mask that can't | |
2922 | * be encoded in either the swp_entry_t or the architecture definition | |
2923 | * of a swap pte. | |
2924 | */ | |
2925 | unsigned long generic_max_swapfile_size(void) | |
2926 | { | |
2927 | return swp_offset(pte_to_swp_entry( | |
2928 | swp_entry_to_pte(swp_entry(0, ~0UL)))) + 1; | |
2929 | } | |
2930 | ||
2931 | /* Can be overridden by an architecture for additional checks. */ | |
2932 | __weak unsigned long max_swapfile_size(void) | |
2933 | { | |
2934 | return generic_max_swapfile_size(); | |
2935 | } | |
2936 | ||
ca8bd38b CEB |
2937 | static unsigned long read_swap_header(struct swap_info_struct *p, |
2938 | union swap_header *swap_header, | |
2939 | struct inode *inode) | |
2940 | { | |
2941 | int i; | |
2942 | unsigned long maxpages; | |
2943 | unsigned long swapfilepages; | |
d6bbbd29 | 2944 | unsigned long last_page; |
ca8bd38b CEB |
2945 | |
2946 | if (memcmp("SWAPSPACE2", swap_header->magic.magic, 10)) { | |
465c47fd | 2947 | pr_err("Unable to find swap-space signature\n"); |
38719025 | 2948 | return 0; |
ca8bd38b CEB |
2949 | } |
2950 | ||
2951 | /* swap partition endianess hack... */ | |
2952 | if (swab32(swap_header->info.version) == 1) { | |
2953 | swab32s(&swap_header->info.version); | |
2954 | swab32s(&swap_header->info.last_page); | |
2955 | swab32s(&swap_header->info.nr_badpages); | |
dd111be6 JH |
2956 | if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES) |
2957 | return 0; | |
ca8bd38b CEB |
2958 | for (i = 0; i < swap_header->info.nr_badpages; i++) |
2959 | swab32s(&swap_header->info.badpages[i]); | |
2960 | } | |
2961 | /* Check the swap header's sub-version */ | |
2962 | if (swap_header->info.version != 1) { | |
465c47fd AM |
2963 | pr_warn("Unable to handle swap header version %d\n", |
2964 | swap_header->info.version); | |
38719025 | 2965 | return 0; |
ca8bd38b CEB |
2966 | } |
2967 | ||
2968 | p->lowest_bit = 1; | |
2969 | p->cluster_next = 1; | |
2970 | p->cluster_nr = 0; | |
2971 | ||
377eeaa8 | 2972 | maxpages = max_swapfile_size(); |
d6bbbd29 | 2973 | last_page = swap_header->info.last_page; |
a06ad633 TA |
2974 | if (!last_page) { |
2975 | pr_warn("Empty swap-file\n"); | |
2976 | return 0; | |
2977 | } | |
d6bbbd29 | 2978 | if (last_page > maxpages) { |
465c47fd | 2979 | pr_warn("Truncating oversized swap area, only using %luk out of %luk\n", |
d6bbbd29 RJ |
2980 | maxpages << (PAGE_SHIFT - 10), |
2981 | last_page << (PAGE_SHIFT - 10)); | |
2982 | } | |
2983 | if (maxpages > last_page) { | |
2984 | maxpages = last_page + 1; | |
ca8bd38b CEB |
2985 | /* p->max is an unsigned int: don't overflow it */ |
2986 | if ((unsigned int)maxpages == 0) | |
2987 | maxpages = UINT_MAX; | |
2988 | } | |
2989 | p->highest_bit = maxpages - 1; | |
2990 | ||
2991 | if (!maxpages) | |
38719025 | 2992 | return 0; |
ca8bd38b CEB |
2993 | swapfilepages = i_size_read(inode) >> PAGE_SHIFT; |
2994 | if (swapfilepages && maxpages > swapfilepages) { | |
465c47fd | 2995 | pr_warn("Swap area shorter than signature indicates\n"); |
38719025 | 2996 | return 0; |
ca8bd38b CEB |
2997 | } |
2998 | if (swap_header->info.nr_badpages && S_ISREG(inode->i_mode)) | |
38719025 | 2999 | return 0; |
ca8bd38b | 3000 | if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES) |
38719025 | 3001 | return 0; |
ca8bd38b CEB |
3002 | |
3003 | return maxpages; | |
ca8bd38b CEB |
3004 | } |
3005 | ||
4b3ef9da | 3006 | #define SWAP_CLUSTER_INFO_COLS \ |
235b6217 | 3007 | DIV_ROUND_UP(L1_CACHE_BYTES, sizeof(struct swap_cluster_info)) |
4b3ef9da HY |
3008 | #define SWAP_CLUSTER_SPACE_COLS \ |
3009 | DIV_ROUND_UP(SWAP_ADDRESS_SPACE_PAGES, SWAPFILE_CLUSTER) | |
3010 | #define SWAP_CLUSTER_COLS \ | |
3011 | max_t(unsigned int, SWAP_CLUSTER_INFO_COLS, SWAP_CLUSTER_SPACE_COLS) | |
235b6217 | 3012 | |
915d4d7b CEB |
3013 | static int setup_swap_map_and_extents(struct swap_info_struct *p, |
3014 | union swap_header *swap_header, | |
3015 | unsigned char *swap_map, | |
2a8f9449 | 3016 | struct swap_cluster_info *cluster_info, |
915d4d7b CEB |
3017 | unsigned long maxpages, |
3018 | sector_t *span) | |
3019 | { | |
235b6217 | 3020 | unsigned int j, k; |
915d4d7b CEB |
3021 | unsigned int nr_good_pages; |
3022 | int nr_extents; | |
2a8f9449 | 3023 | unsigned long nr_clusters = DIV_ROUND_UP(maxpages, SWAPFILE_CLUSTER); |
235b6217 HY |
3024 | unsigned long col = p->cluster_next / SWAPFILE_CLUSTER % SWAP_CLUSTER_COLS; |
3025 | unsigned long i, idx; | |
915d4d7b CEB |
3026 | |
3027 | nr_good_pages = maxpages - 1; /* omit header page */ | |
3028 | ||
6b534915 HY |
3029 | cluster_list_init(&p->free_clusters); |
3030 | cluster_list_init(&p->discard_clusters); | |
2a8f9449 | 3031 | |
915d4d7b CEB |
3032 | for (i = 0; i < swap_header->info.nr_badpages; i++) { |
3033 | unsigned int page_nr = swap_header->info.badpages[i]; | |
bdb8e3f6 CEB |
3034 | if (page_nr == 0 || page_nr > swap_header->info.last_page) |
3035 | return -EINVAL; | |
915d4d7b CEB |
3036 | if (page_nr < maxpages) { |
3037 | swap_map[page_nr] = SWAP_MAP_BAD; | |
3038 | nr_good_pages--; | |
2a8f9449 SL |
3039 | /* |
3040 | * Haven't marked the cluster free yet, no list | |
3041 | * operation involved | |
3042 | */ | |
3043 | inc_cluster_info_page(p, cluster_info, page_nr); | |
915d4d7b CEB |
3044 | } |
3045 | } | |
3046 | ||
2a8f9449 SL |
3047 | /* Haven't marked the cluster free yet, no list operation involved */ |
3048 | for (i = maxpages; i < round_up(maxpages, SWAPFILE_CLUSTER); i++) | |
3049 | inc_cluster_info_page(p, cluster_info, i); | |
3050 | ||
915d4d7b CEB |
3051 | if (nr_good_pages) { |
3052 | swap_map[0] = SWAP_MAP_BAD; | |
2a8f9449 SL |
3053 | /* |
3054 | * Not mark the cluster free yet, no list | |
3055 | * operation involved | |
3056 | */ | |
3057 | inc_cluster_info_page(p, cluster_info, 0); | |
915d4d7b CEB |
3058 | p->max = maxpages; |
3059 | p->pages = nr_good_pages; | |
3060 | nr_extents = setup_swap_extents(p, span); | |
bdb8e3f6 CEB |
3061 | if (nr_extents < 0) |
3062 | return nr_extents; | |
915d4d7b CEB |
3063 | nr_good_pages = p->pages; |
3064 | } | |
3065 | if (!nr_good_pages) { | |
465c47fd | 3066 | pr_warn("Empty swap-file\n"); |
bdb8e3f6 | 3067 | return -EINVAL; |
915d4d7b CEB |
3068 | } |
3069 | ||
2a8f9449 SL |
3070 | if (!cluster_info) |
3071 | return nr_extents; | |
3072 | ||
235b6217 | 3073 | |
4b3ef9da HY |
3074 | /* |
3075 | * Reduce false cache line sharing between cluster_info and | |
3076 | * sharing same address space. | |
3077 | */ | |
235b6217 HY |
3078 | for (k = 0; k < SWAP_CLUSTER_COLS; k++) { |
3079 | j = (k + col) % SWAP_CLUSTER_COLS; | |
3080 | for (i = 0; i < DIV_ROUND_UP(nr_clusters, SWAP_CLUSTER_COLS); i++) { | |
3081 | idx = i * SWAP_CLUSTER_COLS + j; | |
3082 | if (idx >= nr_clusters) | |
3083 | continue; | |
3084 | if (cluster_count(&cluster_info[idx])) | |
3085 | continue; | |
2a8f9449 | 3086 | cluster_set_flag(&cluster_info[idx], CLUSTER_FLAG_FREE); |
6b534915 HY |
3087 | cluster_list_add_tail(&p->free_clusters, cluster_info, |
3088 | idx); | |
2a8f9449 | 3089 | } |
2a8f9449 | 3090 | } |
915d4d7b | 3091 | return nr_extents; |
915d4d7b CEB |
3092 | } |
3093 | ||
dcf6b7dd RA |
3094 | /* |
3095 | * Helper to sys_swapon determining if a given swap | |
3096 | * backing device queue supports DISCARD operations. | |
3097 | */ | |
3098 | static bool swap_discardable(struct swap_info_struct *si) | |
3099 | { | |
3100 | struct request_queue *q = bdev_get_queue(si->bdev); | |
3101 | ||
3102 | if (!q || !blk_queue_discard(q)) | |
3103 | return false; | |
3104 | ||
3105 | return true; | |
3106 | } | |
3107 | ||
53cbb243 CEB |
3108 | SYSCALL_DEFINE2(swapon, const char __user *, specialfile, int, swap_flags) |
3109 | { | |
3110 | struct swap_info_struct *p; | |
91a27b2a | 3111 | struct filename *name; |
53cbb243 CEB |
3112 | struct file *swap_file = NULL; |
3113 | struct address_space *mapping; | |
40531542 | 3114 | int prio; |
53cbb243 CEB |
3115 | int error; |
3116 | union swap_header *swap_header; | |
915d4d7b | 3117 | int nr_extents; |
53cbb243 CEB |
3118 | sector_t span; |
3119 | unsigned long maxpages; | |
53cbb243 | 3120 | unsigned char *swap_map = NULL; |
2a8f9449 | 3121 | struct swap_cluster_info *cluster_info = NULL; |
38b5faf4 | 3122 | unsigned long *frontswap_map = NULL; |
53cbb243 CEB |
3123 | struct page *page = NULL; |
3124 | struct inode *inode = NULL; | |
7cbf3192 | 3125 | bool inced_nr_rotate_swap = false; |
53cbb243 | 3126 | |
d15cab97 HD |
3127 | if (swap_flags & ~SWAP_FLAGS_VALID) |
3128 | return -EINVAL; | |
3129 | ||
53cbb243 CEB |
3130 | if (!capable(CAP_SYS_ADMIN)) |
3131 | return -EPERM; | |
3132 | ||
a2468cc9 AL |
3133 | if (!swap_avail_heads) |
3134 | return -ENOMEM; | |
3135 | ||
53cbb243 | 3136 | p = alloc_swap_info(); |
2542e513 CEB |
3137 | if (IS_ERR(p)) |
3138 | return PTR_ERR(p); | |
53cbb243 | 3139 | |
815c2c54 SL |
3140 | INIT_WORK(&p->discard_work, swap_discard_work); |
3141 | ||
1da177e4 | 3142 | name = getname(specialfile); |
1da177e4 | 3143 | if (IS_ERR(name)) { |
7de7fb6b | 3144 | error = PTR_ERR(name); |
1da177e4 | 3145 | name = NULL; |
bd69010b | 3146 | goto bad_swap; |
1da177e4 | 3147 | } |
669abf4e | 3148 | swap_file = file_open_name(name, O_RDWR|O_LARGEFILE, 0); |
1da177e4 | 3149 | if (IS_ERR(swap_file)) { |
7de7fb6b | 3150 | error = PTR_ERR(swap_file); |
1da177e4 | 3151 | swap_file = NULL; |
bd69010b | 3152 | goto bad_swap; |
1da177e4 LT |
3153 | } |
3154 | ||
3155 | p->swap_file = swap_file; | |
3156 | mapping = swap_file->f_mapping; | |
2130781e | 3157 | inode = mapping->host; |
6f179af8 | 3158 | |
4d0e1e10 CEB |
3159 | error = claim_swapfile(p, inode); |
3160 | if (unlikely(error)) | |
1da177e4 | 3161 | goto bad_swap; |
1da177e4 | 3162 | |
d795a90e NA |
3163 | inode_lock(inode); |
3164 | if (IS_SWAPFILE(inode)) { | |
3165 | error = -EBUSY; | |
3166 | goto bad_swap_unlock_inode; | |
3167 | } | |
3168 | ||
1da177e4 LT |
3169 | /* |
3170 | * Read the swap header. | |
3171 | */ | |
3172 | if (!mapping->a_ops->readpage) { | |
3173 | error = -EINVAL; | |
d795a90e | 3174 | goto bad_swap_unlock_inode; |
1da177e4 | 3175 | } |
090d2b18 | 3176 | page = read_mapping_page(mapping, 0, swap_file); |
1da177e4 LT |
3177 | if (IS_ERR(page)) { |
3178 | error = PTR_ERR(page); | |
d795a90e | 3179 | goto bad_swap_unlock_inode; |
1da177e4 | 3180 | } |
81e33971 | 3181 | swap_header = kmap(page); |
1da177e4 | 3182 | |
ca8bd38b CEB |
3183 | maxpages = read_swap_header(p, swap_header, inode); |
3184 | if (unlikely(!maxpages)) { | |
1da177e4 | 3185 | error = -EINVAL; |
d795a90e | 3186 | goto bad_swap_unlock_inode; |
1da177e4 | 3187 | } |
886bb7e9 | 3188 | |
81e33971 | 3189 | /* OK, set up the swap map and apply the bad block list */ |
803d0c83 | 3190 | swap_map = vzalloc(maxpages); |
81e33971 HD |
3191 | if (!swap_map) { |
3192 | error = -ENOMEM; | |
d795a90e | 3193 | goto bad_swap_unlock_inode; |
81e33971 | 3194 | } |
f0571429 MK |
3195 | |
3196 | if (bdi_cap_stable_pages_required(inode_to_bdi(inode))) | |
3197 | p->flags |= SWP_STABLE_WRITES; | |
3198 | ||
539a6fea MK |
3199 | if (bdi_cap_synchronous_io(inode_to_bdi(inode))) |
3200 | p->flags |= SWP_SYNCHRONOUS_IO; | |
3201 | ||
2a8f9449 | 3202 | if (p->bdev && blk_queue_nonrot(bdev_get_queue(p->bdev))) { |
6f179af8 | 3203 | int cpu; |
235b6217 | 3204 | unsigned long ci, nr_cluster; |
6f179af8 | 3205 | |
2a8f9449 SL |
3206 | p->flags |= SWP_SOLIDSTATE; |
3207 | /* | |
3208 | * select a random position to start with to help wear leveling | |
3209 | * SSD | |
3210 | */ | |
3211 | p->cluster_next = 1 + (prandom_u32() % p->highest_bit); | |
235b6217 | 3212 | nr_cluster = DIV_ROUND_UP(maxpages, SWAPFILE_CLUSTER); |
2a8f9449 | 3213 | |
778e1cdd | 3214 | cluster_info = kvcalloc(nr_cluster, sizeof(*cluster_info), |
54f180d3 | 3215 | GFP_KERNEL); |
2a8f9449 SL |
3216 | if (!cluster_info) { |
3217 | error = -ENOMEM; | |
d795a90e | 3218 | goto bad_swap_unlock_inode; |
2a8f9449 | 3219 | } |
235b6217 HY |
3220 | |
3221 | for (ci = 0; ci < nr_cluster; ci++) | |
3222 | spin_lock_init(&((cluster_info + ci)->lock)); | |
3223 | ||
ebc2a1a6 SL |
3224 | p->percpu_cluster = alloc_percpu(struct percpu_cluster); |
3225 | if (!p->percpu_cluster) { | |
3226 | error = -ENOMEM; | |
d795a90e | 3227 | goto bad_swap_unlock_inode; |
ebc2a1a6 | 3228 | } |
6f179af8 | 3229 | for_each_possible_cpu(cpu) { |
ebc2a1a6 | 3230 | struct percpu_cluster *cluster; |
6f179af8 | 3231 | cluster = per_cpu_ptr(p->percpu_cluster, cpu); |
ebc2a1a6 SL |
3232 | cluster_set_null(&cluster->index); |
3233 | } | |
7cbf3192 | 3234 | } else { |
81a0298b | 3235 | atomic_inc(&nr_rotate_swap); |
7cbf3192 OS |
3236 | inced_nr_rotate_swap = true; |
3237 | } | |
1da177e4 | 3238 | |
1421ef3c CEB |
3239 | error = swap_cgroup_swapon(p->type, maxpages); |
3240 | if (error) | |
d795a90e | 3241 | goto bad_swap_unlock_inode; |
1421ef3c | 3242 | |
915d4d7b | 3243 | nr_extents = setup_swap_map_and_extents(p, swap_header, swap_map, |
2a8f9449 | 3244 | cluster_info, maxpages, &span); |
915d4d7b CEB |
3245 | if (unlikely(nr_extents < 0)) { |
3246 | error = nr_extents; | |
d795a90e | 3247 | goto bad_swap_unlock_inode; |
1da177e4 | 3248 | } |
38b5faf4 | 3249 | /* frontswap enabled? set up bit-per-page map for frontswap */ |
8ea1d2a1 | 3250 | if (IS_ENABLED(CONFIG_FRONTSWAP)) |
778e1cdd KC |
3251 | frontswap_map = kvcalloc(BITS_TO_LONGS(maxpages), |
3252 | sizeof(long), | |
54f180d3 | 3253 | GFP_KERNEL); |
1da177e4 | 3254 | |
2a8f9449 SL |
3255 | if (p->bdev &&(swap_flags & SWAP_FLAG_DISCARD) && swap_discardable(p)) { |
3256 | /* | |
3257 | * When discard is enabled for swap with no particular | |
3258 | * policy flagged, we set all swap discard flags here in | |
3259 | * order to sustain backward compatibility with older | |
3260 | * swapon(8) releases. | |
3261 | */ | |
3262 | p->flags |= (SWP_DISCARDABLE | SWP_AREA_DISCARD | | |
3263 | SWP_PAGE_DISCARD); | |
dcf6b7dd | 3264 | |
2a8f9449 SL |
3265 | /* |
3266 | * By flagging sys_swapon, a sysadmin can tell us to | |
3267 | * either do single-time area discards only, or to just | |
3268 | * perform discards for released swap page-clusters. | |
3269 | * Now it's time to adjust the p->flags accordingly. | |
3270 | */ | |
3271 | if (swap_flags & SWAP_FLAG_DISCARD_ONCE) | |
3272 | p->flags &= ~SWP_PAGE_DISCARD; | |
3273 | else if (swap_flags & SWAP_FLAG_DISCARD_PAGES) | |
3274 | p->flags &= ~SWP_AREA_DISCARD; | |
3275 | ||
3276 | /* issue a swapon-time discard if it's still required */ | |
3277 | if (p->flags & SWP_AREA_DISCARD) { | |
3278 | int err = discard_swap(p); | |
3279 | if (unlikely(err)) | |
3280 | pr_err("swapon: discard_swap(%p): %d\n", | |
3281 | p, err); | |
dcf6b7dd | 3282 | } |
20137a49 | 3283 | } |
6a6ba831 | 3284 | |
4b3ef9da HY |
3285 | error = init_swap_address_space(p->type, maxpages); |
3286 | if (error) | |
d795a90e | 3287 | goto bad_swap_unlock_inode; |
4b3ef9da | 3288 | |
dc617f29 DW |
3289 | /* |
3290 | * Flush any pending IO and dirty mappings before we start using this | |
3291 | * swap device. | |
3292 | */ | |
3293 | inode->i_flags |= S_SWAPFILE; | |
3294 | error = inode_drain_writes(inode); | |
3295 | if (error) { | |
3296 | inode->i_flags &= ~S_SWAPFILE; | |
d795a90e | 3297 | goto bad_swap_unlock_inode; |
dc617f29 DW |
3298 | } |
3299 | ||
fc0abb14 | 3300 | mutex_lock(&swapon_mutex); |
40531542 | 3301 | prio = -1; |
78ecba08 | 3302 | if (swap_flags & SWAP_FLAG_PREFER) |
40531542 | 3303 | prio = |
78ecba08 | 3304 | (swap_flags & SWAP_FLAG_PRIO_MASK) >> SWAP_FLAG_PRIO_SHIFT; |
2a8f9449 | 3305 | enable_swap_info(p, prio, swap_map, cluster_info, frontswap_map); |
c69dbfb8 | 3306 | |
756a025f | 3307 | pr_info("Adding %uk swap on %s. Priority:%d extents:%d across:%lluk %s%s%s%s%s\n", |
91a27b2a | 3308 | p->pages<<(PAGE_SHIFT-10), name->name, p->prio, |
c69dbfb8 CEB |
3309 | nr_extents, (unsigned long long)span<<(PAGE_SHIFT-10), |
3310 | (p->flags & SWP_SOLIDSTATE) ? "SS" : "", | |
38b5faf4 | 3311 | (p->flags & SWP_DISCARDABLE) ? "D" : "", |
dcf6b7dd RA |
3312 | (p->flags & SWP_AREA_DISCARD) ? "s" : "", |
3313 | (p->flags & SWP_PAGE_DISCARD) ? "c" : "", | |
38b5faf4 | 3314 | (frontswap_map) ? "FS" : ""); |
c69dbfb8 | 3315 | |
fc0abb14 | 3316 | mutex_unlock(&swapon_mutex); |
66d7dd51 KS |
3317 | atomic_inc(&proc_poll_event); |
3318 | wake_up_interruptible(&proc_poll_wait); | |
3319 | ||
1da177e4 LT |
3320 | error = 0; |
3321 | goto out; | |
d795a90e NA |
3322 | bad_swap_unlock_inode: |
3323 | inode_unlock(inode); | |
1da177e4 | 3324 | bad_swap: |
ebc2a1a6 SL |
3325 | free_percpu(p->percpu_cluster); |
3326 | p->percpu_cluster = NULL; | |
bd69010b | 3327 | if (inode && S_ISBLK(inode->i_mode) && p->bdev) { |
f2090d2d CEB |
3328 | set_blocksize(p->bdev, p->old_block_size); |
3329 | blkdev_put(p->bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL); | |
1da177e4 | 3330 | } |
d795a90e | 3331 | inode = NULL; |
4cd3bb10 | 3332 | destroy_swap_extents(p); |
e8e6c2ec | 3333 | swap_cgroup_swapoff(p->type); |
5d337b91 | 3334 | spin_lock(&swap_lock); |
1da177e4 | 3335 | p->swap_file = NULL; |
1da177e4 | 3336 | p->flags = 0; |
5d337b91 | 3337 | spin_unlock(&swap_lock); |
1da177e4 | 3338 | vfree(swap_map); |
8606a1a9 | 3339 | kvfree(cluster_info); |
b6b1fd2a | 3340 | kvfree(frontswap_map); |
7cbf3192 OS |
3341 | if (inced_nr_rotate_swap) |
3342 | atomic_dec(&nr_rotate_swap); | |
d795a90e | 3343 | if (swap_file) |
1da177e4 LT |
3344 | filp_close(swap_file, NULL); |
3345 | out: | |
3346 | if (page && !IS_ERR(page)) { | |
3347 | kunmap(page); | |
09cbfeaf | 3348 | put_page(page); |
1da177e4 LT |
3349 | } |
3350 | if (name) | |
3351 | putname(name); | |
1638045c | 3352 | if (inode) |
5955102c | 3353 | inode_unlock(inode); |
039939a6 TC |
3354 | if (!error) |
3355 | enable_swap_slots_cache(); | |
1da177e4 LT |
3356 | return error; |
3357 | } | |
3358 | ||
3359 | void si_swapinfo(struct sysinfo *val) | |
3360 | { | |
efa90a98 | 3361 | unsigned int type; |
1da177e4 LT |
3362 | unsigned long nr_to_be_unused = 0; |
3363 | ||
5d337b91 | 3364 | spin_lock(&swap_lock); |
efa90a98 HD |
3365 | for (type = 0; type < nr_swapfiles; type++) { |
3366 | struct swap_info_struct *si = swap_info[type]; | |
3367 | ||
3368 | if ((si->flags & SWP_USED) && !(si->flags & SWP_WRITEOK)) | |
3369 | nr_to_be_unused += si->inuse_pages; | |
1da177e4 | 3370 | } |
ec8acf20 | 3371 | val->freeswap = atomic_long_read(&nr_swap_pages) + nr_to_be_unused; |
1da177e4 | 3372 | val->totalswap = total_swap_pages + nr_to_be_unused; |
5d337b91 | 3373 | spin_unlock(&swap_lock); |
1da177e4 LT |
3374 | } |
3375 | ||
3376 | /* | |
3377 | * Verify that a swap entry is valid and increment its swap map count. | |
3378 | * | |
355cfa73 KH |
3379 | * Returns error code in following case. |
3380 | * - success -> 0 | |
3381 | * - swp_entry is invalid -> EINVAL | |
3382 | * - swp_entry is migration entry -> EINVAL | |
3383 | * - swap-cache reference is requested but there is already one. -> EEXIST | |
3384 | * - swap-cache reference is requested but the entry is not used. -> ENOENT | |
570a335b | 3385 | * - swap-mapped reference requested but needs continued swap count. -> ENOMEM |
1da177e4 | 3386 | */ |
8d69aaee | 3387 | static int __swap_duplicate(swp_entry_t entry, unsigned char usage) |
1da177e4 | 3388 | { |
73c34b6a | 3389 | struct swap_info_struct *p; |
235b6217 | 3390 | struct swap_cluster_info *ci; |
c10d38cc | 3391 | unsigned long offset; |
8d69aaee HD |
3392 | unsigned char count; |
3393 | unsigned char has_cache; | |
253d553b | 3394 | int err = -EINVAL; |
1da177e4 | 3395 | |
eb085574 | 3396 | p = get_swap_device(entry); |
c10d38cc | 3397 | if (!p) |
235b6217 HY |
3398 | goto out; |
3399 | ||
eb085574 | 3400 | offset = swp_offset(entry); |
235b6217 | 3401 | ci = lock_cluster_or_swap_info(p, offset); |
355cfa73 | 3402 | |
253d553b | 3403 | count = p->swap_map[offset]; |
edfe23da SL |
3404 | |
3405 | /* | |
3406 | * swapin_readahead() doesn't check if a swap entry is valid, so the | |
3407 | * swap entry could be SWAP_MAP_BAD. Check here with lock held. | |
3408 | */ | |
3409 | if (unlikely(swap_count(count) == SWAP_MAP_BAD)) { | |
3410 | err = -ENOENT; | |
3411 | goto unlock_out; | |
3412 | } | |
3413 | ||
253d553b HD |
3414 | has_cache = count & SWAP_HAS_CACHE; |
3415 | count &= ~SWAP_HAS_CACHE; | |
3416 | err = 0; | |
355cfa73 | 3417 | |
253d553b | 3418 | if (usage == SWAP_HAS_CACHE) { |
355cfa73 KH |
3419 | |
3420 | /* set SWAP_HAS_CACHE if there is no cache and entry is used */ | |
253d553b HD |
3421 | if (!has_cache && count) |
3422 | has_cache = SWAP_HAS_CACHE; | |
3423 | else if (has_cache) /* someone else added cache */ | |
3424 | err = -EEXIST; | |
3425 | else /* no users remaining */ | |
3426 | err = -ENOENT; | |
355cfa73 KH |
3427 | |
3428 | } else if (count || has_cache) { | |
253d553b | 3429 | |
570a335b HD |
3430 | if ((count & ~COUNT_CONTINUED) < SWAP_MAP_MAX) |
3431 | count += usage; | |
3432 | else if ((count & ~COUNT_CONTINUED) > SWAP_MAP_MAX) | |
253d553b | 3433 | err = -EINVAL; |
570a335b HD |
3434 | else if (swap_count_continued(p, offset, count)) |
3435 | count = COUNT_CONTINUED; | |
3436 | else | |
3437 | err = -ENOMEM; | |
355cfa73 | 3438 | } else |
253d553b HD |
3439 | err = -ENOENT; /* unused swap entry */ |
3440 | ||
3441 | p->swap_map[offset] = count | has_cache; | |
3442 | ||
355cfa73 | 3443 | unlock_out: |
235b6217 | 3444 | unlock_cluster_or_swap_info(p, ci); |
1da177e4 | 3445 | out: |
eb085574 HY |
3446 | if (p) |
3447 | put_swap_device(p); | |
253d553b | 3448 | return err; |
1da177e4 | 3449 | } |
253d553b | 3450 | |
aaa46865 HD |
3451 | /* |
3452 | * Help swapoff by noting that swap entry belongs to shmem/tmpfs | |
3453 | * (in which case its reference count is never incremented). | |
3454 | */ | |
3455 | void swap_shmem_alloc(swp_entry_t entry) | |
3456 | { | |
3457 | __swap_duplicate(entry, SWAP_MAP_SHMEM); | |
3458 | } | |
3459 | ||
355cfa73 | 3460 | /* |
08259d58 HD |
3461 | * Increase reference count of swap entry by 1. |
3462 | * Returns 0 for success, or -ENOMEM if a swap_count_continuation is required | |
3463 | * but could not be atomically allocated. Returns 0, just as if it succeeded, | |
3464 | * if __swap_duplicate() fails for another reason (-EINVAL or -ENOENT), which | |
3465 | * might occur if a page table entry has got corrupted. | |
355cfa73 | 3466 | */ |
570a335b | 3467 | int swap_duplicate(swp_entry_t entry) |
355cfa73 | 3468 | { |
570a335b HD |
3469 | int err = 0; |
3470 | ||
3471 | while (!err && __swap_duplicate(entry, 1) == -ENOMEM) | |
3472 | err = add_swap_count_continuation(entry, GFP_ATOMIC); | |
3473 | return err; | |
355cfa73 | 3474 | } |
1da177e4 | 3475 | |
cb4b86ba | 3476 | /* |
355cfa73 KH |
3477 | * @entry: swap entry for which we allocate swap cache. |
3478 | * | |
73c34b6a | 3479 | * Called when allocating swap cache for existing swap entry, |
355cfa73 | 3480 | * This can return error codes. Returns 0 at success. |
3eeba135 | 3481 | * -EEXIST means there is a swap cache. |
355cfa73 | 3482 | * Note: return code is different from swap_duplicate(). |
cb4b86ba KH |
3483 | */ |
3484 | int swapcache_prepare(swp_entry_t entry) | |
3485 | { | |
253d553b | 3486 | return __swap_duplicate(entry, SWAP_HAS_CACHE); |
cb4b86ba KH |
3487 | } |
3488 | ||
0bcac06f MK |
3489 | struct swap_info_struct *swp_swap_info(swp_entry_t entry) |
3490 | { | |
c10d38cc | 3491 | return swap_type_to_swap_info(swp_type(entry)); |
0bcac06f MK |
3492 | } |
3493 | ||
f981c595 MG |
3494 | struct swap_info_struct *page_swap_info(struct page *page) |
3495 | { | |
0bcac06f MK |
3496 | swp_entry_t entry = { .val = page_private(page) }; |
3497 | return swp_swap_info(entry); | |
f981c595 MG |
3498 | } |
3499 | ||
3500 | /* | |
3501 | * out-of-line __page_file_ methods to avoid include hell. | |
3502 | */ | |
3503 | struct address_space *__page_file_mapping(struct page *page) | |
3504 | { | |
f981c595 MG |
3505 | return page_swap_info(page)->swap_file->f_mapping; |
3506 | } | |
3507 | EXPORT_SYMBOL_GPL(__page_file_mapping); | |
3508 | ||
3509 | pgoff_t __page_file_index(struct page *page) | |
3510 | { | |
3511 | swp_entry_t swap = { .val = page_private(page) }; | |
f981c595 MG |
3512 | return swp_offset(swap); |
3513 | } | |
3514 | EXPORT_SYMBOL_GPL(__page_file_index); | |
3515 | ||
570a335b HD |
3516 | /* |
3517 | * add_swap_count_continuation - called when a swap count is duplicated | |
3518 | * beyond SWAP_MAP_MAX, it allocates a new page and links that to the entry's | |
3519 | * page of the original vmalloc'ed swap_map, to hold the continuation count | |
3520 | * (for that entry and for its neighbouring PAGE_SIZE swap entries). Called | |
3521 | * again when count is duplicated beyond SWAP_MAP_MAX * SWAP_CONT_MAX, etc. | |
3522 | * | |
3523 | * These continuation pages are seldom referenced: the common paths all work | |
3524 | * on the original swap_map, only referring to a continuation page when the | |
3525 | * low "digit" of a count is incremented or decremented through SWAP_MAP_MAX. | |
3526 | * | |
3527 | * add_swap_count_continuation(, GFP_ATOMIC) can be called while holding | |
3528 | * page table locks; if it fails, add_swap_count_continuation(, GFP_KERNEL) | |
3529 | * can be called after dropping locks. | |
3530 | */ | |
3531 | int add_swap_count_continuation(swp_entry_t entry, gfp_t gfp_mask) | |
3532 | { | |
3533 | struct swap_info_struct *si; | |
235b6217 | 3534 | struct swap_cluster_info *ci; |
570a335b HD |
3535 | struct page *head; |
3536 | struct page *page; | |
3537 | struct page *list_page; | |
3538 | pgoff_t offset; | |
3539 | unsigned char count; | |
eb085574 | 3540 | int ret = 0; |
570a335b HD |
3541 | |
3542 | /* | |
3543 | * When debugging, it's easier to use __GFP_ZERO here; but it's better | |
3544 | * for latency not to zero a page while GFP_ATOMIC and holding locks. | |
3545 | */ | |
3546 | page = alloc_page(gfp_mask | __GFP_HIGHMEM); | |
3547 | ||
eb085574 | 3548 | si = get_swap_device(entry); |
570a335b HD |
3549 | if (!si) { |
3550 | /* | |
3551 | * An acceptable race has occurred since the failing | |
eb085574 | 3552 | * __swap_duplicate(): the swap device may be swapoff |
570a335b HD |
3553 | */ |
3554 | goto outer; | |
3555 | } | |
eb085574 | 3556 | spin_lock(&si->lock); |
570a335b HD |
3557 | |
3558 | offset = swp_offset(entry); | |
235b6217 HY |
3559 | |
3560 | ci = lock_cluster(si, offset); | |
3561 | ||
570a335b HD |
3562 | count = si->swap_map[offset] & ~SWAP_HAS_CACHE; |
3563 | ||
3564 | if ((count & ~COUNT_CONTINUED) != SWAP_MAP_MAX) { | |
3565 | /* | |
3566 | * The higher the swap count, the more likely it is that tasks | |
3567 | * will race to add swap count continuation: we need to avoid | |
3568 | * over-provisioning. | |
3569 | */ | |
3570 | goto out; | |
3571 | } | |
3572 | ||
3573 | if (!page) { | |
eb085574 HY |
3574 | ret = -ENOMEM; |
3575 | goto out; | |
570a335b HD |
3576 | } |
3577 | ||
3578 | /* | |
3579 | * We are fortunate that although vmalloc_to_page uses pte_offset_map, | |
2de1a7e4 SJ |
3580 | * no architecture is using highmem pages for kernel page tables: so it |
3581 | * will not corrupt the GFP_ATOMIC caller's atomic page table kmaps. | |
570a335b HD |
3582 | */ |
3583 | head = vmalloc_to_page(si->swap_map + offset); | |
3584 | offset &= ~PAGE_MASK; | |
3585 | ||
2628bd6f | 3586 | spin_lock(&si->cont_lock); |
570a335b HD |
3587 | /* |
3588 | * Page allocation does not initialize the page's lru field, | |
3589 | * but it does always reset its private field. | |
3590 | */ | |
3591 | if (!page_private(head)) { | |
3592 | BUG_ON(count & COUNT_CONTINUED); | |
3593 | INIT_LIST_HEAD(&head->lru); | |
3594 | set_page_private(head, SWP_CONTINUED); | |
3595 | si->flags |= SWP_CONTINUED; | |
3596 | } | |
3597 | ||
3598 | list_for_each_entry(list_page, &head->lru, lru) { | |
3599 | unsigned char *map; | |
3600 | ||
3601 | /* | |
3602 | * If the previous map said no continuation, but we've found | |
3603 | * a continuation page, free our allocation and use this one. | |
3604 | */ | |
3605 | if (!(count & COUNT_CONTINUED)) | |
2628bd6f | 3606 | goto out_unlock_cont; |
570a335b | 3607 | |
9b04c5fe | 3608 | map = kmap_atomic(list_page) + offset; |
570a335b | 3609 | count = *map; |
9b04c5fe | 3610 | kunmap_atomic(map); |
570a335b HD |
3611 | |
3612 | /* | |
3613 | * If this continuation count now has some space in it, | |
3614 | * free our allocation and use this one. | |
3615 | */ | |
3616 | if ((count & ~COUNT_CONTINUED) != SWAP_CONT_MAX) | |
2628bd6f | 3617 | goto out_unlock_cont; |
570a335b HD |
3618 | } |
3619 | ||
3620 | list_add_tail(&page->lru, &head->lru); | |
3621 | page = NULL; /* now it's attached, don't free it */ | |
2628bd6f HY |
3622 | out_unlock_cont: |
3623 | spin_unlock(&si->cont_lock); | |
570a335b | 3624 | out: |
235b6217 | 3625 | unlock_cluster(ci); |
ec8acf20 | 3626 | spin_unlock(&si->lock); |
eb085574 | 3627 | put_swap_device(si); |
570a335b HD |
3628 | outer: |
3629 | if (page) | |
3630 | __free_page(page); | |
eb085574 | 3631 | return ret; |
570a335b HD |
3632 | } |
3633 | ||
3634 | /* | |
3635 | * swap_count_continued - when the original swap_map count is incremented | |
3636 | * from SWAP_MAP_MAX, check if there is already a continuation page to carry | |
3637 | * into, carry if so, or else fail until a new continuation page is allocated; | |
3638 | * when the original swap_map count is decremented from 0 with continuation, | |
3639 | * borrow from the continuation and report whether it still holds more. | |
235b6217 HY |
3640 | * Called while __swap_duplicate() or swap_entry_free() holds swap or cluster |
3641 | * lock. | |
570a335b HD |
3642 | */ |
3643 | static bool swap_count_continued(struct swap_info_struct *si, | |
3644 | pgoff_t offset, unsigned char count) | |
3645 | { | |
3646 | struct page *head; | |
3647 | struct page *page; | |
3648 | unsigned char *map; | |
2628bd6f | 3649 | bool ret; |
570a335b HD |
3650 | |
3651 | head = vmalloc_to_page(si->swap_map + offset); | |
3652 | if (page_private(head) != SWP_CONTINUED) { | |
3653 | BUG_ON(count & COUNT_CONTINUED); | |
3654 | return false; /* need to add count continuation */ | |
3655 | } | |
3656 | ||
2628bd6f | 3657 | spin_lock(&si->cont_lock); |
570a335b | 3658 | offset &= ~PAGE_MASK; |
213516ac | 3659 | page = list_next_entry(head, lru); |
9b04c5fe | 3660 | map = kmap_atomic(page) + offset; |
570a335b HD |
3661 | |
3662 | if (count == SWAP_MAP_MAX) /* initial increment from swap_map */ | |
3663 | goto init_map; /* jump over SWAP_CONT_MAX checks */ | |
3664 | ||
3665 | if (count == (SWAP_MAP_MAX | COUNT_CONTINUED)) { /* incrementing */ | |
3666 | /* | |
3667 | * Think of how you add 1 to 999 | |
3668 | */ | |
3669 | while (*map == (SWAP_CONT_MAX | COUNT_CONTINUED)) { | |
9b04c5fe | 3670 | kunmap_atomic(map); |
213516ac | 3671 | page = list_next_entry(page, lru); |
570a335b | 3672 | BUG_ON(page == head); |
9b04c5fe | 3673 | map = kmap_atomic(page) + offset; |
570a335b HD |
3674 | } |
3675 | if (*map == SWAP_CONT_MAX) { | |
9b04c5fe | 3676 | kunmap_atomic(map); |
213516ac | 3677 | page = list_next_entry(page, lru); |
2628bd6f HY |
3678 | if (page == head) { |
3679 | ret = false; /* add count continuation */ | |
3680 | goto out; | |
3681 | } | |
9b04c5fe | 3682 | map = kmap_atomic(page) + offset; |
570a335b HD |
3683 | init_map: *map = 0; /* we didn't zero the page */ |
3684 | } | |
3685 | *map += 1; | |
9b04c5fe | 3686 | kunmap_atomic(map); |
213516ac | 3687 | while ((page = list_prev_entry(page, lru)) != head) { |
9b04c5fe | 3688 | map = kmap_atomic(page) + offset; |
570a335b | 3689 | *map = COUNT_CONTINUED; |
9b04c5fe | 3690 | kunmap_atomic(map); |
570a335b | 3691 | } |
2628bd6f | 3692 | ret = true; /* incremented */ |
570a335b HD |
3693 | |
3694 | } else { /* decrementing */ | |
3695 | /* | |
3696 | * Think of how you subtract 1 from 1000 | |
3697 | */ | |
3698 | BUG_ON(count != COUNT_CONTINUED); | |
3699 | while (*map == COUNT_CONTINUED) { | |
9b04c5fe | 3700 | kunmap_atomic(map); |
213516ac | 3701 | page = list_next_entry(page, lru); |
570a335b | 3702 | BUG_ON(page == head); |
9b04c5fe | 3703 | map = kmap_atomic(page) + offset; |
570a335b HD |
3704 | } |
3705 | BUG_ON(*map == 0); | |
3706 | *map -= 1; | |
3707 | if (*map == 0) | |
3708 | count = 0; | |
9b04c5fe | 3709 | kunmap_atomic(map); |
213516ac | 3710 | while ((page = list_prev_entry(page, lru)) != head) { |
9b04c5fe | 3711 | map = kmap_atomic(page) + offset; |
570a335b HD |
3712 | *map = SWAP_CONT_MAX | count; |
3713 | count = COUNT_CONTINUED; | |
9b04c5fe | 3714 | kunmap_atomic(map); |
570a335b | 3715 | } |
2628bd6f | 3716 | ret = count == COUNT_CONTINUED; |
570a335b | 3717 | } |
2628bd6f HY |
3718 | out: |
3719 | spin_unlock(&si->cont_lock); | |
3720 | return ret; | |
570a335b HD |
3721 | } |
3722 | ||
3723 | /* | |
3724 | * free_swap_count_continuations - swapoff free all the continuation pages | |
3725 | * appended to the swap_map, after swap_map is quiesced, before vfree'ing it. | |
3726 | */ | |
3727 | static void free_swap_count_continuations(struct swap_info_struct *si) | |
3728 | { | |
3729 | pgoff_t offset; | |
3730 | ||
3731 | for (offset = 0; offset < si->max; offset += PAGE_SIZE) { | |
3732 | struct page *head; | |
3733 | head = vmalloc_to_page(si->swap_map + offset); | |
3734 | if (page_private(head)) { | |
0d576d20 GT |
3735 | struct page *page, *next; |
3736 | ||
3737 | list_for_each_entry_safe(page, next, &head->lru, lru) { | |
3738 | list_del(&page->lru); | |
570a335b HD |
3739 | __free_page(page); |
3740 | } | |
3741 | } | |
3742 | } | |
3743 | } | |
a2468cc9 | 3744 | |
2cf85583 TH |
3745 | #if defined(CONFIG_MEMCG) && defined(CONFIG_BLK_CGROUP) |
3746 | void mem_cgroup_throttle_swaprate(struct mem_cgroup *memcg, int node, | |
3747 | gfp_t gfp_mask) | |
3748 | { | |
3749 | struct swap_info_struct *si, *next; | |
3750 | if (!(gfp_mask & __GFP_IO) || !memcg) | |
3751 | return; | |
3752 | ||
3753 | if (!blk_cgroup_congested()) | |
3754 | return; | |
3755 | ||
3756 | /* | |
3757 | * We've already scheduled a throttle, avoid taking the global swap | |
3758 | * lock. | |
3759 | */ | |
3760 | if (current->throttle_queue) | |
3761 | return; | |
3762 | ||
3763 | spin_lock(&swap_avail_lock); | |
3764 | plist_for_each_entry_safe(si, next, &swap_avail_heads[node], | |
3765 | avail_lists[node]) { | |
3766 | if (si->bdev) { | |
3767 | blkcg_schedule_throttle(bdev_get_queue(si->bdev), | |
3768 | true); | |
3769 | break; | |
3770 | } | |
3771 | } | |
3772 | spin_unlock(&swap_avail_lock); | |
3773 | } | |
3774 | #endif | |
3775 | ||
a2468cc9 AL |
3776 | static int __init swapfile_init(void) |
3777 | { | |
3778 | int nid; | |
3779 | ||
3780 | swap_avail_heads = kmalloc_array(nr_node_ids, sizeof(struct plist_head), | |
3781 | GFP_KERNEL); | |
3782 | if (!swap_avail_heads) { | |
3783 | pr_emerg("Not enough memory for swap heads, swap is disabled\n"); | |
3784 | return -ENOMEM; | |
3785 | } | |
3786 | ||
3787 | for_each_node(nid) | |
3788 | plist_head_init(&swap_avail_heads[nid]); | |
3789 | ||
3790 | return 0; | |
3791 | } | |
3792 | subsys_initcall(swapfile_init); |