]>
Commit | Line | Data |
---|---|---|
457c8996 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
1da177e4 LT |
2 | /* |
3 | * linux/mm/swapfile.c | |
4 | * | |
5 | * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds | |
6 | * Swap reorganised 29.12.95, Stephen Tweedie | |
7 | */ | |
8 | ||
1da177e4 | 9 | #include <linux/mm.h> |
6e84f315 | 10 | #include <linux/sched/mm.h> |
29930025 | 11 | #include <linux/sched/task.h> |
1da177e4 LT |
12 | #include <linux/hugetlb.h> |
13 | #include <linux/mman.h> | |
14 | #include <linux/slab.h> | |
15 | #include <linux/kernel_stat.h> | |
16 | #include <linux/swap.h> | |
17 | #include <linux/vmalloc.h> | |
18 | #include <linux/pagemap.h> | |
19 | #include <linux/namei.h> | |
072441e2 | 20 | #include <linux/shmem_fs.h> |
1da177e4 | 21 | #include <linux/blkdev.h> |
20137a49 | 22 | #include <linux/random.h> |
1da177e4 LT |
23 | #include <linux/writeback.h> |
24 | #include <linux/proc_fs.h> | |
25 | #include <linux/seq_file.h> | |
26 | #include <linux/init.h> | |
5ad64688 | 27 | #include <linux/ksm.h> |
1da177e4 LT |
28 | #include <linux/rmap.h> |
29 | #include <linux/security.h> | |
30 | #include <linux/backing-dev.h> | |
fc0abb14 | 31 | #include <linux/mutex.h> |
c59ede7b | 32 | #include <linux/capability.h> |
1da177e4 | 33 | #include <linux/syscalls.h> |
8a9f3ccd | 34 | #include <linux/memcontrol.h> |
66d7dd51 | 35 | #include <linux/poll.h> |
72788c38 | 36 | #include <linux/oom.h> |
38b5faf4 DM |
37 | #include <linux/frontswap.h> |
38 | #include <linux/swapfile.h> | |
f981c595 | 39 | #include <linux/export.h> |
67afa38e | 40 | #include <linux/swap_slots.h> |
155b5f88 | 41 | #include <linux/sort.h> |
1da177e4 LT |
42 | |
43 | #include <asm/pgtable.h> | |
44 | #include <asm/tlbflush.h> | |
45 | #include <linux/swapops.h> | |
5d1ea48b | 46 | #include <linux/swap_cgroup.h> |
1da177e4 | 47 | |
570a335b HD |
48 | static bool swap_count_continued(struct swap_info_struct *, pgoff_t, |
49 | unsigned char); | |
50 | static void free_swap_count_continuations(struct swap_info_struct *); | |
d4906e1a | 51 | static sector_t map_swap_entry(swp_entry_t, struct block_device**); |
570a335b | 52 | |
38b5faf4 | 53 | DEFINE_SPINLOCK(swap_lock); |
7c363b8c | 54 | static unsigned int nr_swapfiles; |
ec8acf20 | 55 | atomic_long_t nr_swap_pages; |
fb0fec50 CW |
56 | /* |
57 | * Some modules use swappable objects and may try to swap them out under | |
58 | * memory pressure (via the shrinker). Before doing so, they may wish to | |
59 | * check to see if any swap space is available. | |
60 | */ | |
61 | EXPORT_SYMBOL_GPL(nr_swap_pages); | |
ec8acf20 | 62 | /* protected with swap_lock. reading in vm_swap_full() doesn't need lock */ |
1da177e4 | 63 | long total_swap_pages; |
a2468cc9 | 64 | static int least_priority = -1; |
1da177e4 | 65 | |
1da177e4 LT |
66 | static const char Bad_file[] = "Bad swap file entry "; |
67 | static const char Unused_file[] = "Unused swap file entry "; | |
68 | static const char Bad_offset[] = "Bad swap offset entry "; | |
69 | static const char Unused_offset[] = "Unused swap offset entry "; | |
70 | ||
adfab836 DS |
71 | /* |
72 | * all active swap_info_structs | |
73 | * protected with swap_lock, and ordered by priority. | |
74 | */ | |
18ab4d4c DS |
75 | PLIST_HEAD(swap_active_head); |
76 | ||
77 | /* | |
78 | * all available (active, not full) swap_info_structs | |
79 | * protected with swap_avail_lock, ordered by priority. | |
80 | * This is used by get_swap_page() instead of swap_active_head | |
81 | * because swap_active_head includes all swap_info_structs, | |
82 | * but get_swap_page() doesn't need to look at full ones. | |
83 | * This uses its own lock instead of swap_lock because when a | |
84 | * swap_info_struct changes between not-full/full, it needs to | |
85 | * add/remove itself to/from this list, but the swap_info_struct->lock | |
86 | * is held and the locking order requires swap_lock to be taken | |
87 | * before any swap_info_struct->lock. | |
88 | */ | |
bfc6b1ca | 89 | static struct plist_head *swap_avail_heads; |
18ab4d4c | 90 | static DEFINE_SPINLOCK(swap_avail_lock); |
1da177e4 | 91 | |
38b5faf4 | 92 | struct swap_info_struct *swap_info[MAX_SWAPFILES]; |
1da177e4 | 93 | |
fc0abb14 | 94 | static DEFINE_MUTEX(swapon_mutex); |
1da177e4 | 95 | |
66d7dd51 KS |
96 | static DECLARE_WAIT_QUEUE_HEAD(proc_poll_wait); |
97 | /* Activity counter to indicate that a swapon or swapoff has occurred */ | |
98 | static atomic_t proc_poll_event = ATOMIC_INIT(0); | |
99 | ||
81a0298b HY |
100 | atomic_t nr_rotate_swap = ATOMIC_INIT(0); |
101 | ||
c10d38cc DJ |
102 | static struct swap_info_struct *swap_type_to_swap_info(int type) |
103 | { | |
104 | if (type >= READ_ONCE(nr_swapfiles)) | |
105 | return NULL; | |
106 | ||
107 | smp_rmb(); /* Pairs with smp_wmb in alloc_swap_info. */ | |
108 | return READ_ONCE(swap_info[type]); | |
109 | } | |
110 | ||
8d69aaee | 111 | static inline unsigned char swap_count(unsigned char ent) |
355cfa73 | 112 | { |
955c97f0 | 113 | return ent & ~SWAP_HAS_CACHE; /* may include COUNT_CONTINUED flag */ |
355cfa73 KH |
114 | } |
115 | ||
bcd49e86 HY |
116 | /* Reclaim the swap entry anyway if possible */ |
117 | #define TTRS_ANYWAY 0x1 | |
118 | /* | |
119 | * Reclaim the swap entry if there are no more mappings of the | |
120 | * corresponding page | |
121 | */ | |
122 | #define TTRS_UNMAPPED 0x2 | |
123 | /* Reclaim the swap entry if swap is getting full*/ | |
124 | #define TTRS_FULL 0x4 | |
125 | ||
efa90a98 | 126 | /* returns 1 if swap entry is freed */ |
bcd49e86 HY |
127 | static int __try_to_reclaim_swap(struct swap_info_struct *si, |
128 | unsigned long offset, unsigned long flags) | |
c9e44410 | 129 | { |
efa90a98 | 130 | swp_entry_t entry = swp_entry(si->type, offset); |
c9e44410 KH |
131 | struct page *page; |
132 | int ret = 0; | |
133 | ||
bcd49e86 | 134 | page = find_get_page(swap_address_space(entry), offset); |
c9e44410 KH |
135 | if (!page) |
136 | return 0; | |
137 | /* | |
bcd49e86 HY |
138 | * When this function is called from scan_swap_map_slots() and it's |
139 | * called by vmscan.c at reclaiming pages. So, we hold a lock on a page, | |
140 | * here. We have to use trylock for avoiding deadlock. This is a special | |
c9e44410 KH |
141 | * case and you should use try_to_free_swap() with explicit lock_page() |
142 | * in usual operations. | |
143 | */ | |
144 | if (trylock_page(page)) { | |
bcd49e86 HY |
145 | if ((flags & TTRS_ANYWAY) || |
146 | ((flags & TTRS_UNMAPPED) && !page_mapped(page)) || | |
147 | ((flags & TTRS_FULL) && mem_cgroup_swap_full(page))) | |
148 | ret = try_to_free_swap(page); | |
c9e44410 KH |
149 | unlock_page(page); |
150 | } | |
09cbfeaf | 151 | put_page(page); |
c9e44410 KH |
152 | return ret; |
153 | } | |
355cfa73 | 154 | |
4efaceb1 AL |
155 | static inline struct swap_extent *first_se(struct swap_info_struct *sis) |
156 | { | |
157 | struct rb_node *rb = rb_first(&sis->swap_extent_root); | |
158 | return rb_entry(rb, struct swap_extent, rb_node); | |
159 | } | |
160 | ||
161 | static inline struct swap_extent *next_se(struct swap_extent *se) | |
162 | { | |
163 | struct rb_node *rb = rb_next(&se->rb_node); | |
164 | return rb ? rb_entry(rb, struct swap_extent, rb_node) : NULL; | |
165 | } | |
166 | ||
6a6ba831 HD |
167 | /* |
168 | * swapon tell device that all the old swap contents can be discarded, | |
169 | * to allow the swap device to optimize its wear-levelling. | |
170 | */ | |
171 | static int discard_swap(struct swap_info_struct *si) | |
172 | { | |
173 | struct swap_extent *se; | |
9625a5f2 HD |
174 | sector_t start_block; |
175 | sector_t nr_blocks; | |
6a6ba831 HD |
176 | int err = 0; |
177 | ||
9625a5f2 | 178 | /* Do not discard the swap header page! */ |
4efaceb1 | 179 | se = first_se(si); |
9625a5f2 HD |
180 | start_block = (se->start_block + 1) << (PAGE_SHIFT - 9); |
181 | nr_blocks = ((sector_t)se->nr_pages - 1) << (PAGE_SHIFT - 9); | |
182 | if (nr_blocks) { | |
183 | err = blkdev_issue_discard(si->bdev, start_block, | |
dd3932ed | 184 | nr_blocks, GFP_KERNEL, 0); |
9625a5f2 HD |
185 | if (err) |
186 | return err; | |
187 | cond_resched(); | |
188 | } | |
6a6ba831 | 189 | |
4efaceb1 | 190 | for (se = next_se(se); se; se = next_se(se)) { |
9625a5f2 HD |
191 | start_block = se->start_block << (PAGE_SHIFT - 9); |
192 | nr_blocks = (sector_t)se->nr_pages << (PAGE_SHIFT - 9); | |
6a6ba831 HD |
193 | |
194 | err = blkdev_issue_discard(si->bdev, start_block, | |
dd3932ed | 195 | nr_blocks, GFP_KERNEL, 0); |
6a6ba831 HD |
196 | if (err) |
197 | break; | |
198 | ||
199 | cond_resched(); | |
200 | } | |
201 | return err; /* That will often be -EOPNOTSUPP */ | |
202 | } | |
203 | ||
4efaceb1 AL |
204 | static struct swap_extent * |
205 | offset_to_swap_extent(struct swap_info_struct *sis, unsigned long offset) | |
206 | { | |
207 | struct swap_extent *se; | |
208 | struct rb_node *rb; | |
209 | ||
210 | rb = sis->swap_extent_root.rb_node; | |
211 | while (rb) { | |
212 | se = rb_entry(rb, struct swap_extent, rb_node); | |
213 | if (offset < se->start_page) | |
214 | rb = rb->rb_left; | |
215 | else if (offset >= se->start_page + se->nr_pages) | |
216 | rb = rb->rb_right; | |
217 | else | |
218 | return se; | |
219 | } | |
220 | /* It *must* be present */ | |
221 | BUG(); | |
222 | } | |
223 | ||
7992fde7 HD |
224 | /* |
225 | * swap allocation tell device that a cluster of swap can now be discarded, | |
226 | * to allow the swap device to optimize its wear-levelling. | |
227 | */ | |
228 | static void discard_swap_cluster(struct swap_info_struct *si, | |
229 | pgoff_t start_page, pgoff_t nr_pages) | |
230 | { | |
4efaceb1 | 231 | struct swap_extent *se = offset_to_swap_extent(si, start_page); |
7992fde7 HD |
232 | |
233 | while (nr_pages) { | |
4efaceb1 AL |
234 | pgoff_t offset = start_page - se->start_page; |
235 | sector_t start_block = se->start_block + offset; | |
236 | sector_t nr_blocks = se->nr_pages - offset; | |
237 | ||
238 | if (nr_blocks > nr_pages) | |
239 | nr_blocks = nr_pages; | |
240 | start_page += nr_blocks; | |
241 | nr_pages -= nr_blocks; | |
242 | ||
243 | start_block <<= PAGE_SHIFT - 9; | |
244 | nr_blocks <<= PAGE_SHIFT - 9; | |
245 | if (blkdev_issue_discard(si->bdev, start_block, | |
246 | nr_blocks, GFP_NOIO, 0)) | |
247 | break; | |
7992fde7 | 248 | |
4efaceb1 | 249 | se = next_se(se); |
7992fde7 HD |
250 | } |
251 | } | |
252 | ||
38d8b4e6 HY |
253 | #ifdef CONFIG_THP_SWAP |
254 | #define SWAPFILE_CLUSTER HPAGE_PMD_NR | |
a448f2d0 HY |
255 | |
256 | #define swap_entry_size(size) (size) | |
38d8b4e6 | 257 | #else |
048c27fd | 258 | #define SWAPFILE_CLUSTER 256 |
a448f2d0 HY |
259 | |
260 | /* | |
261 | * Define swap_entry_size() as constant to let compiler to optimize | |
262 | * out some code if !CONFIG_THP_SWAP | |
263 | */ | |
264 | #define swap_entry_size(size) 1 | |
38d8b4e6 | 265 | #endif |
048c27fd HD |
266 | #define LATENCY_LIMIT 256 |
267 | ||
2a8f9449 SL |
268 | static inline void cluster_set_flag(struct swap_cluster_info *info, |
269 | unsigned int flag) | |
270 | { | |
271 | info->flags = flag; | |
272 | } | |
273 | ||
274 | static inline unsigned int cluster_count(struct swap_cluster_info *info) | |
275 | { | |
276 | return info->data; | |
277 | } | |
278 | ||
279 | static inline void cluster_set_count(struct swap_cluster_info *info, | |
280 | unsigned int c) | |
281 | { | |
282 | info->data = c; | |
283 | } | |
284 | ||
285 | static inline void cluster_set_count_flag(struct swap_cluster_info *info, | |
286 | unsigned int c, unsigned int f) | |
287 | { | |
288 | info->flags = f; | |
289 | info->data = c; | |
290 | } | |
291 | ||
292 | static inline unsigned int cluster_next(struct swap_cluster_info *info) | |
293 | { | |
294 | return info->data; | |
295 | } | |
296 | ||
297 | static inline void cluster_set_next(struct swap_cluster_info *info, | |
298 | unsigned int n) | |
299 | { | |
300 | info->data = n; | |
301 | } | |
302 | ||
303 | static inline void cluster_set_next_flag(struct swap_cluster_info *info, | |
304 | unsigned int n, unsigned int f) | |
305 | { | |
306 | info->flags = f; | |
307 | info->data = n; | |
308 | } | |
309 | ||
310 | static inline bool cluster_is_free(struct swap_cluster_info *info) | |
311 | { | |
312 | return info->flags & CLUSTER_FLAG_FREE; | |
313 | } | |
314 | ||
315 | static inline bool cluster_is_null(struct swap_cluster_info *info) | |
316 | { | |
317 | return info->flags & CLUSTER_FLAG_NEXT_NULL; | |
318 | } | |
319 | ||
320 | static inline void cluster_set_null(struct swap_cluster_info *info) | |
321 | { | |
322 | info->flags = CLUSTER_FLAG_NEXT_NULL; | |
323 | info->data = 0; | |
324 | } | |
325 | ||
e0709829 HY |
326 | static inline bool cluster_is_huge(struct swap_cluster_info *info) |
327 | { | |
33ee011e HY |
328 | if (IS_ENABLED(CONFIG_THP_SWAP)) |
329 | return info->flags & CLUSTER_FLAG_HUGE; | |
330 | return false; | |
e0709829 HY |
331 | } |
332 | ||
333 | static inline void cluster_clear_huge(struct swap_cluster_info *info) | |
334 | { | |
335 | info->flags &= ~CLUSTER_FLAG_HUGE; | |
336 | } | |
337 | ||
235b6217 HY |
338 | static inline struct swap_cluster_info *lock_cluster(struct swap_info_struct *si, |
339 | unsigned long offset) | |
340 | { | |
341 | struct swap_cluster_info *ci; | |
342 | ||
343 | ci = si->cluster_info; | |
344 | if (ci) { | |
345 | ci += offset / SWAPFILE_CLUSTER; | |
346 | spin_lock(&ci->lock); | |
347 | } | |
348 | return ci; | |
349 | } | |
350 | ||
351 | static inline void unlock_cluster(struct swap_cluster_info *ci) | |
352 | { | |
353 | if (ci) | |
354 | spin_unlock(&ci->lock); | |
355 | } | |
356 | ||
59d98bf3 HY |
357 | /* |
358 | * Determine the locking method in use for this device. Return | |
359 | * swap_cluster_info if SSD-style cluster-based locking is in place. | |
360 | */ | |
235b6217 | 361 | static inline struct swap_cluster_info *lock_cluster_or_swap_info( |
59d98bf3 | 362 | struct swap_info_struct *si, unsigned long offset) |
235b6217 HY |
363 | { |
364 | struct swap_cluster_info *ci; | |
365 | ||
59d98bf3 | 366 | /* Try to use fine-grained SSD-style locking if available: */ |
235b6217 | 367 | ci = lock_cluster(si, offset); |
59d98bf3 | 368 | /* Otherwise, fall back to traditional, coarse locking: */ |
235b6217 HY |
369 | if (!ci) |
370 | spin_lock(&si->lock); | |
371 | ||
372 | return ci; | |
373 | } | |
374 | ||
375 | static inline void unlock_cluster_or_swap_info(struct swap_info_struct *si, | |
376 | struct swap_cluster_info *ci) | |
377 | { | |
378 | if (ci) | |
379 | unlock_cluster(ci); | |
380 | else | |
381 | spin_unlock(&si->lock); | |
382 | } | |
383 | ||
6b534915 HY |
384 | static inline bool cluster_list_empty(struct swap_cluster_list *list) |
385 | { | |
386 | return cluster_is_null(&list->head); | |
387 | } | |
388 | ||
389 | static inline unsigned int cluster_list_first(struct swap_cluster_list *list) | |
390 | { | |
391 | return cluster_next(&list->head); | |
392 | } | |
393 | ||
394 | static void cluster_list_init(struct swap_cluster_list *list) | |
395 | { | |
396 | cluster_set_null(&list->head); | |
397 | cluster_set_null(&list->tail); | |
398 | } | |
399 | ||
400 | static void cluster_list_add_tail(struct swap_cluster_list *list, | |
401 | struct swap_cluster_info *ci, | |
402 | unsigned int idx) | |
403 | { | |
404 | if (cluster_list_empty(list)) { | |
405 | cluster_set_next_flag(&list->head, idx, 0); | |
406 | cluster_set_next_flag(&list->tail, idx, 0); | |
407 | } else { | |
235b6217 | 408 | struct swap_cluster_info *ci_tail; |
6b534915 HY |
409 | unsigned int tail = cluster_next(&list->tail); |
410 | ||
235b6217 HY |
411 | /* |
412 | * Nested cluster lock, but both cluster locks are | |
413 | * only acquired when we held swap_info_struct->lock | |
414 | */ | |
415 | ci_tail = ci + tail; | |
416 | spin_lock_nested(&ci_tail->lock, SINGLE_DEPTH_NESTING); | |
417 | cluster_set_next(ci_tail, idx); | |
0ef017d1 | 418 | spin_unlock(&ci_tail->lock); |
6b534915 HY |
419 | cluster_set_next_flag(&list->tail, idx, 0); |
420 | } | |
421 | } | |
422 | ||
423 | static unsigned int cluster_list_del_first(struct swap_cluster_list *list, | |
424 | struct swap_cluster_info *ci) | |
425 | { | |
426 | unsigned int idx; | |
427 | ||
428 | idx = cluster_next(&list->head); | |
429 | if (cluster_next(&list->tail) == idx) { | |
430 | cluster_set_null(&list->head); | |
431 | cluster_set_null(&list->tail); | |
432 | } else | |
433 | cluster_set_next_flag(&list->head, | |
434 | cluster_next(&ci[idx]), 0); | |
435 | ||
436 | return idx; | |
437 | } | |
438 | ||
815c2c54 SL |
439 | /* Add a cluster to discard list and schedule it to do discard */ |
440 | static void swap_cluster_schedule_discard(struct swap_info_struct *si, | |
441 | unsigned int idx) | |
442 | { | |
443 | /* | |
444 | * If scan_swap_map() can't find a free cluster, it will check | |
445 | * si->swap_map directly. To make sure the discarding cluster isn't | |
446 | * taken by scan_swap_map(), mark the swap entries bad (occupied). It | |
447 | * will be cleared after discard | |
448 | */ | |
449 | memset(si->swap_map + idx * SWAPFILE_CLUSTER, | |
450 | SWAP_MAP_BAD, SWAPFILE_CLUSTER); | |
451 | ||
6b534915 | 452 | cluster_list_add_tail(&si->discard_clusters, si->cluster_info, idx); |
815c2c54 SL |
453 | |
454 | schedule_work(&si->discard_work); | |
455 | } | |
456 | ||
38d8b4e6 HY |
457 | static void __free_cluster(struct swap_info_struct *si, unsigned long idx) |
458 | { | |
459 | struct swap_cluster_info *ci = si->cluster_info; | |
460 | ||
461 | cluster_set_flag(ci + idx, CLUSTER_FLAG_FREE); | |
462 | cluster_list_add_tail(&si->free_clusters, ci, idx); | |
463 | } | |
464 | ||
815c2c54 SL |
465 | /* |
466 | * Doing discard actually. After a cluster discard is finished, the cluster | |
467 | * will be added to free cluster list. caller should hold si->lock. | |
468 | */ | |
469 | static void swap_do_scheduled_discard(struct swap_info_struct *si) | |
470 | { | |
235b6217 | 471 | struct swap_cluster_info *info, *ci; |
815c2c54 SL |
472 | unsigned int idx; |
473 | ||
474 | info = si->cluster_info; | |
475 | ||
6b534915 HY |
476 | while (!cluster_list_empty(&si->discard_clusters)) { |
477 | idx = cluster_list_del_first(&si->discard_clusters, info); | |
815c2c54 SL |
478 | spin_unlock(&si->lock); |
479 | ||
480 | discard_swap_cluster(si, idx * SWAPFILE_CLUSTER, | |
481 | SWAPFILE_CLUSTER); | |
482 | ||
483 | spin_lock(&si->lock); | |
235b6217 | 484 | ci = lock_cluster(si, idx * SWAPFILE_CLUSTER); |
38d8b4e6 | 485 | __free_cluster(si, idx); |
815c2c54 SL |
486 | memset(si->swap_map + idx * SWAPFILE_CLUSTER, |
487 | 0, SWAPFILE_CLUSTER); | |
235b6217 | 488 | unlock_cluster(ci); |
815c2c54 SL |
489 | } |
490 | } | |
491 | ||
492 | static void swap_discard_work(struct work_struct *work) | |
493 | { | |
494 | struct swap_info_struct *si; | |
495 | ||
496 | si = container_of(work, struct swap_info_struct, discard_work); | |
497 | ||
498 | spin_lock(&si->lock); | |
499 | swap_do_scheduled_discard(si); | |
500 | spin_unlock(&si->lock); | |
501 | } | |
502 | ||
38d8b4e6 HY |
503 | static void alloc_cluster(struct swap_info_struct *si, unsigned long idx) |
504 | { | |
505 | struct swap_cluster_info *ci = si->cluster_info; | |
506 | ||
507 | VM_BUG_ON(cluster_list_first(&si->free_clusters) != idx); | |
508 | cluster_list_del_first(&si->free_clusters, ci); | |
509 | cluster_set_count_flag(ci + idx, 0, 0); | |
510 | } | |
511 | ||
512 | static void free_cluster(struct swap_info_struct *si, unsigned long idx) | |
513 | { | |
514 | struct swap_cluster_info *ci = si->cluster_info + idx; | |
515 | ||
516 | VM_BUG_ON(cluster_count(ci) != 0); | |
517 | /* | |
518 | * If the swap is discardable, prepare discard the cluster | |
519 | * instead of free it immediately. The cluster will be freed | |
520 | * after discard. | |
521 | */ | |
522 | if ((si->flags & (SWP_WRITEOK | SWP_PAGE_DISCARD)) == | |
523 | (SWP_WRITEOK | SWP_PAGE_DISCARD)) { | |
524 | swap_cluster_schedule_discard(si, idx); | |
525 | return; | |
526 | } | |
527 | ||
528 | __free_cluster(si, idx); | |
529 | } | |
530 | ||
2a8f9449 SL |
531 | /* |
532 | * The cluster corresponding to page_nr will be used. The cluster will be | |
533 | * removed from free cluster list and its usage counter will be increased. | |
534 | */ | |
535 | static void inc_cluster_info_page(struct swap_info_struct *p, | |
536 | struct swap_cluster_info *cluster_info, unsigned long page_nr) | |
537 | { | |
538 | unsigned long idx = page_nr / SWAPFILE_CLUSTER; | |
539 | ||
540 | if (!cluster_info) | |
541 | return; | |
38d8b4e6 HY |
542 | if (cluster_is_free(&cluster_info[idx])) |
543 | alloc_cluster(p, idx); | |
2a8f9449 SL |
544 | |
545 | VM_BUG_ON(cluster_count(&cluster_info[idx]) >= SWAPFILE_CLUSTER); | |
546 | cluster_set_count(&cluster_info[idx], | |
547 | cluster_count(&cluster_info[idx]) + 1); | |
548 | } | |
549 | ||
550 | /* | |
551 | * The cluster corresponding to page_nr decreases one usage. If the usage | |
552 | * counter becomes 0, which means no page in the cluster is in using, we can | |
553 | * optionally discard the cluster and add it to free cluster list. | |
554 | */ | |
555 | static void dec_cluster_info_page(struct swap_info_struct *p, | |
556 | struct swap_cluster_info *cluster_info, unsigned long page_nr) | |
557 | { | |
558 | unsigned long idx = page_nr / SWAPFILE_CLUSTER; | |
559 | ||
560 | if (!cluster_info) | |
561 | return; | |
562 | ||
563 | VM_BUG_ON(cluster_count(&cluster_info[idx]) == 0); | |
564 | cluster_set_count(&cluster_info[idx], | |
565 | cluster_count(&cluster_info[idx]) - 1); | |
566 | ||
38d8b4e6 HY |
567 | if (cluster_count(&cluster_info[idx]) == 0) |
568 | free_cluster(p, idx); | |
2a8f9449 SL |
569 | } |
570 | ||
571 | /* | |
572 | * It's possible scan_swap_map() uses a free cluster in the middle of free | |
573 | * cluster list. Avoiding such abuse to avoid list corruption. | |
574 | */ | |
ebc2a1a6 SL |
575 | static bool |
576 | scan_swap_map_ssd_cluster_conflict(struct swap_info_struct *si, | |
2a8f9449 SL |
577 | unsigned long offset) |
578 | { | |
ebc2a1a6 SL |
579 | struct percpu_cluster *percpu_cluster; |
580 | bool conflict; | |
581 | ||
2a8f9449 | 582 | offset /= SWAPFILE_CLUSTER; |
6b534915 HY |
583 | conflict = !cluster_list_empty(&si->free_clusters) && |
584 | offset != cluster_list_first(&si->free_clusters) && | |
2a8f9449 | 585 | cluster_is_free(&si->cluster_info[offset]); |
ebc2a1a6 SL |
586 | |
587 | if (!conflict) | |
588 | return false; | |
589 | ||
590 | percpu_cluster = this_cpu_ptr(si->percpu_cluster); | |
591 | cluster_set_null(&percpu_cluster->index); | |
592 | return true; | |
593 | } | |
594 | ||
595 | /* | |
596 | * Try to get a swap entry from current cpu's swap entry pool (a cluster). This | |
597 | * might involve allocating a new cluster for current CPU too. | |
598 | */ | |
36005bae | 599 | static bool scan_swap_map_try_ssd_cluster(struct swap_info_struct *si, |
ebc2a1a6 SL |
600 | unsigned long *offset, unsigned long *scan_base) |
601 | { | |
602 | struct percpu_cluster *cluster; | |
235b6217 | 603 | struct swap_cluster_info *ci; |
235b6217 | 604 | unsigned long tmp, max; |
ebc2a1a6 SL |
605 | |
606 | new_cluster: | |
607 | cluster = this_cpu_ptr(si->percpu_cluster); | |
608 | if (cluster_is_null(&cluster->index)) { | |
6b534915 HY |
609 | if (!cluster_list_empty(&si->free_clusters)) { |
610 | cluster->index = si->free_clusters.head; | |
ebc2a1a6 SL |
611 | cluster->next = cluster_next(&cluster->index) * |
612 | SWAPFILE_CLUSTER; | |
6b534915 | 613 | } else if (!cluster_list_empty(&si->discard_clusters)) { |
ebc2a1a6 SL |
614 | /* |
615 | * we don't have free cluster but have some clusters in | |
616 | * discarding, do discard now and reclaim them | |
617 | */ | |
618 | swap_do_scheduled_discard(si); | |
619 | *scan_base = *offset = si->cluster_next; | |
620 | goto new_cluster; | |
621 | } else | |
36005bae | 622 | return false; |
ebc2a1a6 SL |
623 | } |
624 | ||
ebc2a1a6 SL |
625 | /* |
626 | * Other CPUs can use our cluster if they can't find a free cluster, | |
627 | * check if there is still free entry in the cluster | |
628 | */ | |
629 | tmp = cluster->next; | |
235b6217 HY |
630 | max = min_t(unsigned long, si->max, |
631 | (cluster_next(&cluster->index) + 1) * SWAPFILE_CLUSTER); | |
7b9e2de1 WY |
632 | if (tmp < max) { |
633 | ci = lock_cluster(si, tmp); | |
634 | while (tmp < max) { | |
635 | if (!si->swap_map[tmp]) | |
636 | break; | |
637 | tmp++; | |
638 | } | |
639 | unlock_cluster(ci); | |
ebc2a1a6 | 640 | } |
0fd0e19e | 641 | if (tmp >= max) { |
ebc2a1a6 SL |
642 | cluster_set_null(&cluster->index); |
643 | goto new_cluster; | |
644 | } | |
645 | cluster->next = tmp + 1; | |
646 | *offset = tmp; | |
647 | *scan_base = tmp; | |
fdff1deb | 648 | return true; |
2a8f9449 SL |
649 | } |
650 | ||
a2468cc9 AL |
651 | static void __del_from_avail_list(struct swap_info_struct *p) |
652 | { | |
653 | int nid; | |
654 | ||
655 | for_each_node(nid) | |
656 | plist_del(&p->avail_lists[nid], &swap_avail_heads[nid]); | |
657 | } | |
658 | ||
659 | static void del_from_avail_list(struct swap_info_struct *p) | |
660 | { | |
661 | spin_lock(&swap_avail_lock); | |
662 | __del_from_avail_list(p); | |
663 | spin_unlock(&swap_avail_lock); | |
664 | } | |
665 | ||
38d8b4e6 HY |
666 | static void swap_range_alloc(struct swap_info_struct *si, unsigned long offset, |
667 | unsigned int nr_entries) | |
668 | { | |
669 | unsigned int end = offset + nr_entries - 1; | |
670 | ||
671 | if (offset == si->lowest_bit) | |
672 | si->lowest_bit += nr_entries; | |
673 | if (end == si->highest_bit) | |
674 | si->highest_bit -= nr_entries; | |
675 | si->inuse_pages += nr_entries; | |
676 | if (si->inuse_pages == si->pages) { | |
677 | si->lowest_bit = si->max; | |
678 | si->highest_bit = 0; | |
a2468cc9 | 679 | del_from_avail_list(si); |
38d8b4e6 HY |
680 | } |
681 | } | |
682 | ||
a2468cc9 AL |
683 | static void add_to_avail_list(struct swap_info_struct *p) |
684 | { | |
685 | int nid; | |
686 | ||
687 | spin_lock(&swap_avail_lock); | |
688 | for_each_node(nid) { | |
689 | WARN_ON(!plist_node_empty(&p->avail_lists[nid])); | |
690 | plist_add(&p->avail_lists[nid], &swap_avail_heads[nid]); | |
691 | } | |
692 | spin_unlock(&swap_avail_lock); | |
693 | } | |
694 | ||
38d8b4e6 HY |
695 | static void swap_range_free(struct swap_info_struct *si, unsigned long offset, |
696 | unsigned int nr_entries) | |
697 | { | |
698 | unsigned long end = offset + nr_entries - 1; | |
699 | void (*swap_slot_free_notify)(struct block_device *, unsigned long); | |
700 | ||
701 | if (offset < si->lowest_bit) | |
702 | si->lowest_bit = offset; | |
703 | if (end > si->highest_bit) { | |
704 | bool was_full = !si->highest_bit; | |
705 | ||
706 | si->highest_bit = end; | |
a2468cc9 AL |
707 | if (was_full && (si->flags & SWP_WRITEOK)) |
708 | add_to_avail_list(si); | |
38d8b4e6 HY |
709 | } |
710 | atomic_long_add(nr_entries, &nr_swap_pages); | |
711 | si->inuse_pages -= nr_entries; | |
712 | if (si->flags & SWP_BLKDEV) | |
713 | swap_slot_free_notify = | |
714 | si->bdev->bd_disk->fops->swap_slot_free_notify; | |
715 | else | |
716 | swap_slot_free_notify = NULL; | |
717 | while (offset <= end) { | |
718 | frontswap_invalidate_page(si->type, offset); | |
719 | if (swap_slot_free_notify) | |
720 | swap_slot_free_notify(si->bdev, offset); | |
721 | offset++; | |
722 | } | |
723 | } | |
724 | ||
36005bae TC |
725 | static int scan_swap_map_slots(struct swap_info_struct *si, |
726 | unsigned char usage, int nr, | |
727 | swp_entry_t slots[]) | |
1da177e4 | 728 | { |
235b6217 | 729 | struct swap_cluster_info *ci; |
ebebbbe9 | 730 | unsigned long offset; |
c60aa176 | 731 | unsigned long scan_base; |
7992fde7 | 732 | unsigned long last_in_cluster = 0; |
048c27fd | 733 | int latency_ration = LATENCY_LIMIT; |
36005bae | 734 | int n_ret = 0; |
ed43af10 | 735 | bool scanned_many = false; |
36005bae | 736 | |
886bb7e9 | 737 | /* |
7dfad418 HD |
738 | * We try to cluster swap pages by allocating them sequentially |
739 | * in swap. Once we've allocated SWAPFILE_CLUSTER pages this | |
740 | * way, however, we resort to first-free allocation, starting | |
741 | * a new cluster. This prevents us from scattering swap pages | |
742 | * all over the entire swap partition, so that we reduce | |
743 | * overall disk seek times between swap pages. -- sct | |
744 | * But we do now try to find an empty cluster. -Andrea | |
c60aa176 | 745 | * And we let swap pages go all over an SSD partition. Hugh |
7dfad418 HD |
746 | */ |
747 | ||
52b7efdb | 748 | si->flags += SWP_SCANNING; |
c60aa176 | 749 | scan_base = offset = si->cluster_next; |
ebebbbe9 | 750 | |
ebc2a1a6 SL |
751 | /* SSD algorithm */ |
752 | if (si->cluster_info) { | |
bd2d18da | 753 | if (!scan_swap_map_try_ssd_cluster(si, &offset, &scan_base)) |
36005bae | 754 | goto scan; |
f4eaf51a | 755 | } else if (unlikely(!si->cluster_nr--)) { |
ebebbbe9 HD |
756 | if (si->pages - si->inuse_pages < SWAPFILE_CLUSTER) { |
757 | si->cluster_nr = SWAPFILE_CLUSTER - 1; | |
758 | goto checks; | |
759 | } | |
2a8f9449 | 760 | |
ec8acf20 | 761 | spin_unlock(&si->lock); |
7dfad418 | 762 | |
c60aa176 HD |
763 | /* |
764 | * If seek is expensive, start searching for new cluster from | |
765 | * start of partition, to minimize the span of allocated swap. | |
50088c44 CY |
766 | * If seek is cheap, that is the SWP_SOLIDSTATE si->cluster_info |
767 | * case, just handled by scan_swap_map_try_ssd_cluster() above. | |
c60aa176 | 768 | */ |
50088c44 | 769 | scan_base = offset = si->lowest_bit; |
7dfad418 HD |
770 | last_in_cluster = offset + SWAPFILE_CLUSTER - 1; |
771 | ||
772 | /* Locate the first empty (unaligned) cluster */ | |
773 | for (; last_in_cluster <= si->highest_bit; offset++) { | |
1da177e4 | 774 | if (si->swap_map[offset]) |
7dfad418 HD |
775 | last_in_cluster = offset + SWAPFILE_CLUSTER; |
776 | else if (offset == last_in_cluster) { | |
ec8acf20 | 777 | spin_lock(&si->lock); |
ebebbbe9 HD |
778 | offset -= SWAPFILE_CLUSTER - 1; |
779 | si->cluster_next = offset; | |
780 | si->cluster_nr = SWAPFILE_CLUSTER - 1; | |
c60aa176 HD |
781 | goto checks; |
782 | } | |
783 | if (unlikely(--latency_ration < 0)) { | |
784 | cond_resched(); | |
785 | latency_ration = LATENCY_LIMIT; | |
786 | } | |
787 | } | |
788 | ||
789 | offset = scan_base; | |
ec8acf20 | 790 | spin_lock(&si->lock); |
ebebbbe9 | 791 | si->cluster_nr = SWAPFILE_CLUSTER - 1; |
1da177e4 | 792 | } |
7dfad418 | 793 | |
ebebbbe9 | 794 | checks: |
ebc2a1a6 | 795 | if (si->cluster_info) { |
36005bae TC |
796 | while (scan_swap_map_ssd_cluster_conflict(si, offset)) { |
797 | /* take a break if we already got some slots */ | |
798 | if (n_ret) | |
799 | goto done; | |
800 | if (!scan_swap_map_try_ssd_cluster(si, &offset, | |
801 | &scan_base)) | |
802 | goto scan; | |
803 | } | |
ebc2a1a6 | 804 | } |
ebebbbe9 | 805 | if (!(si->flags & SWP_WRITEOK)) |
52b7efdb | 806 | goto no_page; |
7dfad418 HD |
807 | if (!si->highest_bit) |
808 | goto no_page; | |
ebebbbe9 | 809 | if (offset > si->highest_bit) |
c60aa176 | 810 | scan_base = offset = si->lowest_bit; |
c9e44410 | 811 | |
235b6217 | 812 | ci = lock_cluster(si, offset); |
b73d7fce HD |
813 | /* reuse swap entry of cache-only swap if not busy. */ |
814 | if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) { | |
c9e44410 | 815 | int swap_was_freed; |
235b6217 | 816 | unlock_cluster(ci); |
ec8acf20 | 817 | spin_unlock(&si->lock); |
bcd49e86 | 818 | swap_was_freed = __try_to_reclaim_swap(si, offset, TTRS_ANYWAY); |
ec8acf20 | 819 | spin_lock(&si->lock); |
c9e44410 KH |
820 | /* entry was freed successfully, try to use this again */ |
821 | if (swap_was_freed) | |
822 | goto checks; | |
823 | goto scan; /* check next one */ | |
824 | } | |
825 | ||
235b6217 HY |
826 | if (si->swap_map[offset]) { |
827 | unlock_cluster(ci); | |
36005bae TC |
828 | if (!n_ret) |
829 | goto scan; | |
830 | else | |
831 | goto done; | |
235b6217 | 832 | } |
2872bb2d HY |
833 | si->swap_map[offset] = usage; |
834 | inc_cluster_info_page(si, si->cluster_info, offset); | |
835 | unlock_cluster(ci); | |
ebebbbe9 | 836 | |
38d8b4e6 | 837 | swap_range_alloc(si, offset, 1); |
ebebbbe9 | 838 | si->cluster_next = offset + 1; |
36005bae TC |
839 | slots[n_ret++] = swp_entry(si->type, offset); |
840 | ||
841 | /* got enough slots or reach max slots? */ | |
842 | if ((n_ret == nr) || (offset >= si->highest_bit)) | |
843 | goto done; | |
844 | ||
845 | /* search for next available slot */ | |
846 | ||
847 | /* time to take a break? */ | |
848 | if (unlikely(--latency_ration < 0)) { | |
849 | if (n_ret) | |
850 | goto done; | |
851 | spin_unlock(&si->lock); | |
852 | cond_resched(); | |
853 | spin_lock(&si->lock); | |
854 | latency_ration = LATENCY_LIMIT; | |
855 | } | |
856 | ||
857 | /* try to get more slots in cluster */ | |
858 | if (si->cluster_info) { | |
859 | if (scan_swap_map_try_ssd_cluster(si, &offset, &scan_base)) | |
860 | goto checks; | |
f4eaf51a WY |
861 | } else if (si->cluster_nr && !si->swap_map[++offset]) { |
862 | /* non-ssd case, still more slots in cluster? */ | |
36005bae TC |
863 | --si->cluster_nr; |
864 | goto checks; | |
865 | } | |
7992fde7 | 866 | |
ed43af10 HY |
867 | /* |
868 | * Even if there's no free clusters available (fragmented), | |
869 | * try to scan a little more quickly with lock held unless we | |
870 | * have scanned too many slots already. | |
871 | */ | |
872 | if (!scanned_many) { | |
873 | unsigned long scan_limit; | |
874 | ||
875 | if (offset < scan_base) | |
876 | scan_limit = scan_base; | |
877 | else | |
878 | scan_limit = si->highest_bit; | |
879 | for (; offset <= scan_limit && --latency_ration > 0; | |
880 | offset++) { | |
881 | if (!si->swap_map[offset]) | |
882 | goto checks; | |
883 | } | |
884 | } | |
885 | ||
36005bae TC |
886 | done: |
887 | si->flags -= SWP_SCANNING; | |
888 | return n_ret; | |
7dfad418 | 889 | |
ebebbbe9 | 890 | scan: |
ec8acf20 | 891 | spin_unlock(&si->lock); |
7dfad418 | 892 | while (++offset <= si->highest_bit) { |
52b7efdb | 893 | if (!si->swap_map[offset]) { |
ec8acf20 | 894 | spin_lock(&si->lock); |
52b7efdb HD |
895 | goto checks; |
896 | } | |
c9e44410 | 897 | if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) { |
ec8acf20 | 898 | spin_lock(&si->lock); |
c9e44410 KH |
899 | goto checks; |
900 | } | |
048c27fd HD |
901 | if (unlikely(--latency_ration < 0)) { |
902 | cond_resched(); | |
903 | latency_ration = LATENCY_LIMIT; | |
ed43af10 | 904 | scanned_many = true; |
048c27fd | 905 | } |
7dfad418 | 906 | } |
c60aa176 | 907 | offset = si->lowest_bit; |
a5998061 | 908 | while (offset < scan_base) { |
c60aa176 | 909 | if (!si->swap_map[offset]) { |
ec8acf20 | 910 | spin_lock(&si->lock); |
c60aa176 HD |
911 | goto checks; |
912 | } | |
c9e44410 | 913 | if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) { |
ec8acf20 | 914 | spin_lock(&si->lock); |
c9e44410 KH |
915 | goto checks; |
916 | } | |
c60aa176 HD |
917 | if (unlikely(--latency_ration < 0)) { |
918 | cond_resched(); | |
919 | latency_ration = LATENCY_LIMIT; | |
ed43af10 | 920 | scanned_many = true; |
c60aa176 | 921 | } |
a5998061 | 922 | offset++; |
c60aa176 | 923 | } |
ec8acf20 | 924 | spin_lock(&si->lock); |
7dfad418 HD |
925 | |
926 | no_page: | |
52b7efdb | 927 | si->flags -= SWP_SCANNING; |
36005bae | 928 | return n_ret; |
1da177e4 LT |
929 | } |
930 | ||
38d8b4e6 HY |
931 | static int swap_alloc_cluster(struct swap_info_struct *si, swp_entry_t *slot) |
932 | { | |
933 | unsigned long idx; | |
934 | struct swap_cluster_info *ci; | |
935 | unsigned long offset, i; | |
936 | unsigned char *map; | |
937 | ||
fe5266d5 HY |
938 | /* |
939 | * Should not even be attempting cluster allocations when huge | |
940 | * page swap is disabled. Warn and fail the allocation. | |
941 | */ | |
942 | if (!IS_ENABLED(CONFIG_THP_SWAP)) { | |
943 | VM_WARN_ON_ONCE(1); | |
944 | return 0; | |
945 | } | |
946 | ||
38d8b4e6 HY |
947 | if (cluster_list_empty(&si->free_clusters)) |
948 | return 0; | |
949 | ||
950 | idx = cluster_list_first(&si->free_clusters); | |
951 | offset = idx * SWAPFILE_CLUSTER; | |
952 | ci = lock_cluster(si, offset); | |
953 | alloc_cluster(si, idx); | |
e0709829 | 954 | cluster_set_count_flag(ci, SWAPFILE_CLUSTER, CLUSTER_FLAG_HUGE); |
38d8b4e6 HY |
955 | |
956 | map = si->swap_map + offset; | |
957 | for (i = 0; i < SWAPFILE_CLUSTER; i++) | |
958 | map[i] = SWAP_HAS_CACHE; | |
959 | unlock_cluster(ci); | |
960 | swap_range_alloc(si, offset, SWAPFILE_CLUSTER); | |
961 | *slot = swp_entry(si->type, offset); | |
962 | ||
963 | return 1; | |
964 | } | |
965 | ||
966 | static void swap_free_cluster(struct swap_info_struct *si, unsigned long idx) | |
967 | { | |
968 | unsigned long offset = idx * SWAPFILE_CLUSTER; | |
969 | struct swap_cluster_info *ci; | |
970 | ||
971 | ci = lock_cluster(si, offset); | |
979aafa5 | 972 | memset(si->swap_map + offset, 0, SWAPFILE_CLUSTER); |
38d8b4e6 HY |
973 | cluster_set_count_flag(ci, 0, 0); |
974 | free_cluster(si, idx); | |
975 | unlock_cluster(ci); | |
976 | swap_range_free(si, offset, SWAPFILE_CLUSTER); | |
977 | } | |
38d8b4e6 | 978 | |
36005bae TC |
979 | static unsigned long scan_swap_map(struct swap_info_struct *si, |
980 | unsigned char usage) | |
981 | { | |
982 | swp_entry_t entry; | |
983 | int n_ret; | |
984 | ||
985 | n_ret = scan_swap_map_slots(si, usage, 1, &entry); | |
986 | ||
987 | if (n_ret) | |
988 | return swp_offset(entry); | |
989 | else | |
990 | return 0; | |
991 | ||
992 | } | |
993 | ||
5d5e8f19 | 994 | int get_swap_pages(int n_goal, swp_entry_t swp_entries[], int entry_size) |
1da177e4 | 995 | { |
5d5e8f19 | 996 | unsigned long size = swap_entry_size(entry_size); |
adfab836 | 997 | struct swap_info_struct *si, *next; |
36005bae TC |
998 | long avail_pgs; |
999 | int n_ret = 0; | |
a2468cc9 | 1000 | int node; |
1da177e4 | 1001 | |
38d8b4e6 | 1002 | /* Only single cluster request supported */ |
5d5e8f19 | 1003 | WARN_ON_ONCE(n_goal > 1 && size == SWAPFILE_CLUSTER); |
38d8b4e6 | 1004 | |
5d5e8f19 | 1005 | avail_pgs = atomic_long_read(&nr_swap_pages) / size; |
36005bae | 1006 | if (avail_pgs <= 0) |
fb4f88dc | 1007 | goto noswap; |
36005bae | 1008 | |
08d3090f | 1009 | n_goal = min3((long)n_goal, (long)SWAP_BATCH, avail_pgs); |
36005bae | 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 | 1275 | static unsigned char __swap_entry_free(struct swap_info_struct *p, |
33e16272 | 1276 | swp_entry_t entry) |
b32d5f32 HY |
1277 | { |
1278 | struct swap_cluster_info *ci; | |
1279 | unsigned long offset = swp_offset(entry); | |
33e16272 | 1280 | unsigned char usage; |
b32d5f32 HY |
1281 | |
1282 | ci = lock_cluster_or_swap_info(p, offset); | |
33e16272 | 1283 | usage = __swap_entry_free_locked(p, offset, 1); |
7c00bafe | 1284 | unlock_cluster_or_swap_info(p, ci); |
10e364da HY |
1285 | if (!usage) |
1286 | free_swap_slot(entry); | |
7c00bafe TC |
1287 | |
1288 | return usage; | |
1289 | } | |
355cfa73 | 1290 | |
7c00bafe TC |
1291 | static void swap_entry_free(struct swap_info_struct *p, swp_entry_t entry) |
1292 | { | |
1293 | struct swap_cluster_info *ci; | |
1294 | unsigned long offset = swp_offset(entry); | |
1295 | unsigned char count; | |
1296 | ||
1297 | ci = lock_cluster(p, offset); | |
1298 | count = p->swap_map[offset]; | |
1299 | VM_BUG_ON(count != SWAP_HAS_CACHE); | |
1300 | p->swap_map[offset] = 0; | |
1301 | dec_cluster_info_page(p, p->cluster_info, offset); | |
235b6217 HY |
1302 | unlock_cluster(ci); |
1303 | ||
38d8b4e6 HY |
1304 | mem_cgroup_uncharge_swap(entry, 1); |
1305 | swap_range_free(p, offset, 1); | |
1da177e4 LT |
1306 | } |
1307 | ||
1308 | /* | |
2de1a7e4 | 1309 | * Caller has made sure that the swap device corresponding to entry |
1da177e4 LT |
1310 | * is still around or has not been recycled. |
1311 | */ | |
1312 | void swap_free(swp_entry_t entry) | |
1313 | { | |
73c34b6a | 1314 | struct swap_info_struct *p; |
1da177e4 | 1315 | |
235b6217 | 1316 | p = _swap_info_get(entry); |
10e364da | 1317 | if (p) |
33e16272 | 1318 | __swap_entry_free(p, entry); |
1da177e4 LT |
1319 | } |
1320 | ||
cb4b86ba KH |
1321 | /* |
1322 | * Called after dropping swapcache to decrease refcnt to swap entries. | |
1323 | */ | |
a448f2d0 | 1324 | void put_swap_page(struct page *page, swp_entry_t entry) |
38d8b4e6 HY |
1325 | { |
1326 | unsigned long offset = swp_offset(entry); | |
1327 | unsigned long idx = offset / SWAPFILE_CLUSTER; | |
1328 | struct swap_cluster_info *ci; | |
1329 | struct swap_info_struct *si; | |
1330 | unsigned char *map; | |
a3aea839 HY |
1331 | unsigned int i, free_entries = 0; |
1332 | unsigned char val; | |
a448f2d0 | 1333 | int size = swap_entry_size(hpage_nr_pages(page)); |
fe5266d5 | 1334 | |
a3aea839 | 1335 | si = _swap_info_get(entry); |
38d8b4e6 HY |
1336 | if (!si) |
1337 | return; | |
1338 | ||
c2343d27 | 1339 | ci = lock_cluster_or_swap_info(si, offset); |
a448f2d0 | 1340 | if (size == SWAPFILE_CLUSTER) { |
a448f2d0 HY |
1341 | VM_BUG_ON(!cluster_is_huge(ci)); |
1342 | map = si->swap_map + offset; | |
1343 | for (i = 0; i < SWAPFILE_CLUSTER; i++) { | |
1344 | val = map[i]; | |
1345 | VM_BUG_ON(!(val & SWAP_HAS_CACHE)); | |
1346 | if (val == SWAP_HAS_CACHE) | |
1347 | free_entries++; | |
1348 | } | |
a448f2d0 | 1349 | cluster_clear_huge(ci); |
a448f2d0 | 1350 | if (free_entries == SWAPFILE_CLUSTER) { |
c2343d27 | 1351 | unlock_cluster_or_swap_info(si, ci); |
a448f2d0 | 1352 | spin_lock(&si->lock); |
a448f2d0 HY |
1353 | mem_cgroup_uncharge_swap(entry, SWAPFILE_CLUSTER); |
1354 | swap_free_cluster(si, idx); | |
1355 | spin_unlock(&si->lock); | |
1356 | return; | |
1357 | } | |
1358 | } | |
c2343d27 HY |
1359 | for (i = 0; i < size; i++, entry.val++) { |
1360 | if (!__swap_entry_free_locked(si, offset + i, SWAP_HAS_CACHE)) { | |
1361 | unlock_cluster_or_swap_info(si, ci); | |
1362 | free_swap_slot(entry); | |
1363 | if (i == size - 1) | |
1364 | return; | |
1365 | lock_cluster_or_swap_info(si, offset); | |
a3aea839 HY |
1366 | } |
1367 | } | |
c2343d27 | 1368 | unlock_cluster_or_swap_info(si, ci); |
38d8b4e6 | 1369 | } |
59807685 | 1370 | |
fe5266d5 | 1371 | #ifdef CONFIG_THP_SWAP |
59807685 HY |
1372 | int split_swap_cluster(swp_entry_t entry) |
1373 | { | |
1374 | struct swap_info_struct *si; | |
1375 | struct swap_cluster_info *ci; | |
1376 | unsigned long offset = swp_offset(entry); | |
1377 | ||
1378 | si = _swap_info_get(entry); | |
1379 | if (!si) | |
1380 | return -EBUSY; | |
1381 | ci = lock_cluster(si, offset); | |
1382 | cluster_clear_huge(ci); | |
1383 | unlock_cluster(ci); | |
1384 | return 0; | |
1385 | } | |
fe5266d5 | 1386 | #endif |
38d8b4e6 | 1387 | |
155b5f88 HY |
1388 | static int swp_entry_cmp(const void *ent1, const void *ent2) |
1389 | { | |
1390 | const swp_entry_t *e1 = ent1, *e2 = ent2; | |
1391 | ||
1392 | return (int)swp_type(*e1) - (int)swp_type(*e2); | |
1393 | } | |
1394 | ||
7c00bafe TC |
1395 | void swapcache_free_entries(swp_entry_t *entries, int n) |
1396 | { | |
1397 | struct swap_info_struct *p, *prev; | |
1398 | int i; | |
1399 | ||
1400 | if (n <= 0) | |
1401 | return; | |
1402 | ||
1403 | prev = NULL; | |
1404 | p = NULL; | |
155b5f88 HY |
1405 | |
1406 | /* | |
1407 | * Sort swap entries by swap device, so each lock is only taken once. | |
1408 | * nr_swapfiles isn't absolutely correct, but the overhead of sort() is | |
1409 | * so low that it isn't necessary to optimize further. | |
1410 | */ | |
1411 | if (nr_swapfiles > 1) | |
1412 | sort(entries, n, sizeof(entries[0]), swp_entry_cmp, NULL); | |
7c00bafe TC |
1413 | for (i = 0; i < n; ++i) { |
1414 | p = swap_info_get_cont(entries[i], prev); | |
1415 | if (p) | |
1416 | swap_entry_free(p, entries[i]); | |
7c00bafe TC |
1417 | prev = p; |
1418 | } | |
235b6217 | 1419 | if (p) |
7c00bafe | 1420 | spin_unlock(&p->lock); |
cb4b86ba KH |
1421 | } |
1422 | ||
1da177e4 | 1423 | /* |
c475a8ab | 1424 | * How many references to page are currently swapped out? |
570a335b HD |
1425 | * This does not give an exact answer when swap count is continued, |
1426 | * but does include the high COUNT_CONTINUED flag to allow for that. | |
1da177e4 | 1427 | */ |
bde05d1c | 1428 | int page_swapcount(struct page *page) |
1da177e4 | 1429 | { |
c475a8ab HD |
1430 | int count = 0; |
1431 | struct swap_info_struct *p; | |
235b6217 | 1432 | struct swap_cluster_info *ci; |
1da177e4 | 1433 | swp_entry_t entry; |
235b6217 | 1434 | unsigned long offset; |
1da177e4 | 1435 | |
4c21e2f2 | 1436 | entry.val = page_private(page); |
235b6217 | 1437 | p = _swap_info_get(entry); |
1da177e4 | 1438 | if (p) { |
235b6217 HY |
1439 | offset = swp_offset(entry); |
1440 | ci = lock_cluster_or_swap_info(p, offset); | |
1441 | count = swap_count(p->swap_map[offset]); | |
1442 | unlock_cluster_or_swap_info(p, ci); | |
1da177e4 | 1443 | } |
c475a8ab | 1444 | return count; |
1da177e4 LT |
1445 | } |
1446 | ||
eb085574 | 1447 | int __swap_count(swp_entry_t entry) |
aa8d22a1 | 1448 | { |
eb085574 | 1449 | struct swap_info_struct *si; |
aa8d22a1 | 1450 | pgoff_t offset = swp_offset(entry); |
eb085574 | 1451 | int count = 0; |
aa8d22a1 | 1452 | |
eb085574 HY |
1453 | si = get_swap_device(entry); |
1454 | if (si) { | |
1455 | count = swap_count(si->swap_map[offset]); | |
1456 | put_swap_device(si); | |
1457 | } | |
1458 | return count; | |
aa8d22a1 MK |
1459 | } |
1460 | ||
322b8afe HY |
1461 | static int swap_swapcount(struct swap_info_struct *si, swp_entry_t entry) |
1462 | { | |
1463 | int count = 0; | |
1464 | pgoff_t offset = swp_offset(entry); | |
1465 | struct swap_cluster_info *ci; | |
1466 | ||
1467 | ci = lock_cluster_or_swap_info(si, offset); | |
1468 | count = swap_count(si->swap_map[offset]); | |
1469 | unlock_cluster_or_swap_info(si, ci); | |
1470 | return count; | |
1471 | } | |
1472 | ||
e8c26ab6 TC |
1473 | /* |
1474 | * How many references to @entry are currently swapped out? | |
1475 | * This does not give an exact answer when swap count is continued, | |
1476 | * but does include the high COUNT_CONTINUED flag to allow for that. | |
1477 | */ | |
1478 | int __swp_swapcount(swp_entry_t entry) | |
1479 | { | |
1480 | int count = 0; | |
e8c26ab6 | 1481 | struct swap_info_struct *si; |
e8c26ab6 | 1482 | |
eb085574 HY |
1483 | si = get_swap_device(entry); |
1484 | if (si) { | |
322b8afe | 1485 | count = swap_swapcount(si, entry); |
eb085574 HY |
1486 | put_swap_device(si); |
1487 | } | |
e8c26ab6 TC |
1488 | return count; |
1489 | } | |
1490 | ||
8334b962 MK |
1491 | /* |
1492 | * How many references to @entry are currently swapped out? | |
1493 | * This considers COUNT_CONTINUED so it returns exact answer. | |
1494 | */ | |
1495 | int swp_swapcount(swp_entry_t entry) | |
1496 | { | |
1497 | int count, tmp_count, n; | |
1498 | struct swap_info_struct *p; | |
235b6217 | 1499 | struct swap_cluster_info *ci; |
8334b962 MK |
1500 | struct page *page; |
1501 | pgoff_t offset; | |
1502 | unsigned char *map; | |
1503 | ||
235b6217 | 1504 | p = _swap_info_get(entry); |
8334b962 MK |
1505 | if (!p) |
1506 | return 0; | |
1507 | ||
235b6217 HY |
1508 | offset = swp_offset(entry); |
1509 | ||
1510 | ci = lock_cluster_or_swap_info(p, offset); | |
1511 | ||
1512 | count = swap_count(p->swap_map[offset]); | |
8334b962 MK |
1513 | if (!(count & COUNT_CONTINUED)) |
1514 | goto out; | |
1515 | ||
1516 | count &= ~COUNT_CONTINUED; | |
1517 | n = SWAP_MAP_MAX + 1; | |
1518 | ||
8334b962 MK |
1519 | page = vmalloc_to_page(p->swap_map + offset); |
1520 | offset &= ~PAGE_MASK; | |
1521 | VM_BUG_ON(page_private(page) != SWP_CONTINUED); | |
1522 | ||
1523 | do { | |
a8ae4991 | 1524 | page = list_next_entry(page, lru); |
8334b962 MK |
1525 | map = kmap_atomic(page); |
1526 | tmp_count = map[offset]; | |
1527 | kunmap_atomic(map); | |
1528 | ||
1529 | count += (tmp_count & ~COUNT_CONTINUED) * n; | |
1530 | n *= (SWAP_CONT_MAX + 1); | |
1531 | } while (tmp_count & COUNT_CONTINUED); | |
1532 | out: | |
235b6217 | 1533 | unlock_cluster_or_swap_info(p, ci); |
8334b962 MK |
1534 | return count; |
1535 | } | |
1536 | ||
e0709829 HY |
1537 | static bool swap_page_trans_huge_swapped(struct swap_info_struct *si, |
1538 | swp_entry_t entry) | |
1539 | { | |
1540 | struct swap_cluster_info *ci; | |
1541 | unsigned char *map = si->swap_map; | |
1542 | unsigned long roffset = swp_offset(entry); | |
1543 | unsigned long offset = round_down(roffset, SWAPFILE_CLUSTER); | |
1544 | int i; | |
1545 | bool ret = false; | |
1546 | ||
1547 | ci = lock_cluster_or_swap_info(si, offset); | |
1548 | if (!ci || !cluster_is_huge(ci)) { | |
afa4711e | 1549 | if (swap_count(map[roffset])) |
e0709829 HY |
1550 | ret = true; |
1551 | goto unlock_out; | |
1552 | } | |
1553 | for (i = 0; i < SWAPFILE_CLUSTER; i++) { | |
afa4711e | 1554 | if (swap_count(map[offset + i])) { |
e0709829 HY |
1555 | ret = true; |
1556 | break; | |
1557 | } | |
1558 | } | |
1559 | unlock_out: | |
1560 | unlock_cluster_or_swap_info(si, ci); | |
1561 | return ret; | |
1562 | } | |
1563 | ||
1564 | static bool page_swapped(struct page *page) | |
1565 | { | |
1566 | swp_entry_t entry; | |
1567 | struct swap_info_struct *si; | |
1568 | ||
fe5266d5 | 1569 | if (!IS_ENABLED(CONFIG_THP_SWAP) || likely(!PageTransCompound(page))) |
e0709829 HY |
1570 | return page_swapcount(page) != 0; |
1571 | ||
1572 | page = compound_head(page); | |
1573 | entry.val = page_private(page); | |
1574 | si = _swap_info_get(entry); | |
1575 | if (si) | |
1576 | return swap_page_trans_huge_swapped(si, entry); | |
1577 | return false; | |
1578 | } | |
ba3c4ce6 HY |
1579 | |
1580 | static int page_trans_huge_map_swapcount(struct page *page, int *total_mapcount, | |
1581 | int *total_swapcount) | |
1582 | { | |
1583 | int i, map_swapcount, _total_mapcount, _total_swapcount; | |
1584 | unsigned long offset = 0; | |
1585 | struct swap_info_struct *si; | |
1586 | struct swap_cluster_info *ci = NULL; | |
1587 | unsigned char *map = NULL; | |
1588 | int mapcount, swapcount = 0; | |
1589 | ||
1590 | /* hugetlbfs shouldn't call it */ | |
1591 | VM_BUG_ON_PAGE(PageHuge(page), page); | |
1592 | ||
fe5266d5 HY |
1593 | if (!IS_ENABLED(CONFIG_THP_SWAP) || likely(!PageTransCompound(page))) { |
1594 | mapcount = page_trans_huge_mapcount(page, total_mapcount); | |
ba3c4ce6 HY |
1595 | if (PageSwapCache(page)) |
1596 | swapcount = page_swapcount(page); | |
1597 | if (total_swapcount) | |
1598 | *total_swapcount = swapcount; | |
1599 | return mapcount + swapcount; | |
1600 | } | |
1601 | ||
1602 | page = compound_head(page); | |
1603 | ||
1604 | _total_mapcount = _total_swapcount = map_swapcount = 0; | |
1605 | if (PageSwapCache(page)) { | |
1606 | swp_entry_t entry; | |
1607 | ||
1608 | entry.val = page_private(page); | |
1609 | si = _swap_info_get(entry); | |
1610 | if (si) { | |
1611 | map = si->swap_map; | |
1612 | offset = swp_offset(entry); | |
1613 | } | |
1614 | } | |
1615 | if (map) | |
1616 | ci = lock_cluster(si, offset); | |
1617 | for (i = 0; i < HPAGE_PMD_NR; i++) { | |
1618 | mapcount = atomic_read(&page[i]._mapcount) + 1; | |
1619 | _total_mapcount += mapcount; | |
1620 | if (map) { | |
1621 | swapcount = swap_count(map[offset + i]); | |
1622 | _total_swapcount += swapcount; | |
1623 | } | |
1624 | map_swapcount = max(map_swapcount, mapcount + swapcount); | |
1625 | } | |
1626 | unlock_cluster(ci); | |
1627 | if (PageDoubleMap(page)) { | |
1628 | map_swapcount -= 1; | |
1629 | _total_mapcount -= HPAGE_PMD_NR; | |
1630 | } | |
1631 | mapcount = compound_mapcount(page); | |
1632 | map_swapcount += mapcount; | |
1633 | _total_mapcount += mapcount; | |
1634 | if (total_mapcount) | |
1635 | *total_mapcount = _total_mapcount; | |
1636 | if (total_swapcount) | |
1637 | *total_swapcount = _total_swapcount; | |
1638 | ||
1639 | return map_swapcount; | |
1640 | } | |
e0709829 | 1641 | |
1da177e4 | 1642 | /* |
7b1fe597 HD |
1643 | * We can write to an anon page without COW if there are no other references |
1644 | * to it. And as a side-effect, free up its swap: because the old content | |
1645 | * on disk will never be read, and seeking back there to write new content | |
1646 | * later would only waste time away from clustering. | |
6d0a07ed | 1647 | * |
ba3c4ce6 | 1648 | * NOTE: total_map_swapcount should not be relied upon by the caller if |
6d0a07ed AA |
1649 | * reuse_swap_page() returns false, but it may be always overwritten |
1650 | * (see the other implementation for CONFIG_SWAP=n). | |
1da177e4 | 1651 | */ |
ba3c4ce6 | 1652 | bool reuse_swap_page(struct page *page, int *total_map_swapcount) |
1da177e4 | 1653 | { |
ba3c4ce6 | 1654 | int count, total_mapcount, total_swapcount; |
c475a8ab | 1655 | |
309381fe | 1656 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
5ad64688 | 1657 | if (unlikely(PageKsm(page))) |
6d0a07ed | 1658 | return false; |
ba3c4ce6 HY |
1659 | count = page_trans_huge_map_swapcount(page, &total_mapcount, |
1660 | &total_swapcount); | |
1661 | if (total_map_swapcount) | |
1662 | *total_map_swapcount = total_mapcount + total_swapcount; | |
1663 | if (count == 1 && PageSwapCache(page) && | |
1664 | (likely(!PageTransCompound(page)) || | |
1665 | /* The remaining swap count will be freed soon */ | |
1666 | total_swapcount == page_swapcount(page))) { | |
f0571429 | 1667 | if (!PageWriteback(page)) { |
ba3c4ce6 | 1668 | page = compound_head(page); |
7b1fe597 HD |
1669 | delete_from_swap_cache(page); |
1670 | SetPageDirty(page); | |
f0571429 MK |
1671 | } else { |
1672 | swp_entry_t entry; | |
1673 | struct swap_info_struct *p; | |
1674 | ||
1675 | entry.val = page_private(page); | |
1676 | p = swap_info_get(entry); | |
1677 | if (p->flags & SWP_STABLE_WRITES) { | |
1678 | spin_unlock(&p->lock); | |
1679 | return false; | |
1680 | } | |
1681 | spin_unlock(&p->lock); | |
7b1fe597 HD |
1682 | } |
1683 | } | |
ba3c4ce6 | 1684 | |
5ad64688 | 1685 | return count <= 1; |
1da177e4 LT |
1686 | } |
1687 | ||
1688 | /* | |
a2c43eed HD |
1689 | * If swap is getting full, or if there are no more mappings of this page, |
1690 | * then try_to_free_swap is called to free its swap space. | |
1da177e4 | 1691 | */ |
a2c43eed | 1692 | int try_to_free_swap(struct page *page) |
1da177e4 | 1693 | { |
309381fe | 1694 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
1da177e4 LT |
1695 | |
1696 | if (!PageSwapCache(page)) | |
1697 | return 0; | |
1698 | if (PageWriteback(page)) | |
1699 | return 0; | |
e0709829 | 1700 | if (page_swapped(page)) |
1da177e4 LT |
1701 | return 0; |
1702 | ||
b73d7fce HD |
1703 | /* |
1704 | * Once hibernation has begun to create its image of memory, | |
1705 | * there's a danger that one of the calls to try_to_free_swap() | |
1706 | * - most probably a call from __try_to_reclaim_swap() while | |
1707 | * hibernation is allocating its own swap pages for the image, | |
1708 | * but conceivably even a call from memory reclaim - will free | |
1709 | * the swap from a page which has already been recorded in the | |
1710 | * image as a clean swapcache page, and then reuse its swap for | |
1711 | * another page of the image. On waking from hibernation, the | |
1712 | * original page might be freed under memory pressure, then | |
1713 | * later read back in from swap, now with the wrong data. | |
1714 | * | |
2de1a7e4 | 1715 | * Hibernation suspends storage while it is writing the image |
f90ac398 | 1716 | * to disk so check that here. |
b73d7fce | 1717 | */ |
f90ac398 | 1718 | if (pm_suspended_storage()) |
b73d7fce HD |
1719 | return 0; |
1720 | ||
e0709829 | 1721 | page = compound_head(page); |
a2c43eed HD |
1722 | delete_from_swap_cache(page); |
1723 | SetPageDirty(page); | |
1724 | return 1; | |
68a22394 RR |
1725 | } |
1726 | ||
1da177e4 LT |
1727 | /* |
1728 | * Free the swap entry like above, but also try to | |
1729 | * free the page cache entry if it is the last user. | |
1730 | */ | |
2509ef26 | 1731 | int free_swap_and_cache(swp_entry_t entry) |
1da177e4 | 1732 | { |
2509ef26 | 1733 | struct swap_info_struct *p; |
7c00bafe | 1734 | unsigned char count; |
1da177e4 | 1735 | |
a7420aa5 | 1736 | if (non_swap_entry(entry)) |
2509ef26 | 1737 | return 1; |
0697212a | 1738 | |
7c00bafe | 1739 | p = _swap_info_get(entry); |
1da177e4 | 1740 | if (p) { |
33e16272 | 1741 | count = __swap_entry_free(p, entry); |
e0709829 | 1742 | if (count == SWAP_HAS_CACHE && |
bcd49e86 HY |
1743 | !swap_page_trans_huge_swapped(p, entry)) |
1744 | __try_to_reclaim_swap(p, swp_offset(entry), | |
1745 | TTRS_UNMAPPED | TTRS_FULL); | |
1da177e4 | 1746 | } |
2509ef26 | 1747 | return p != NULL; |
1da177e4 LT |
1748 | } |
1749 | ||
b0cb1a19 | 1750 | #ifdef CONFIG_HIBERNATION |
f577eb30 | 1751 | /* |
915bae9e | 1752 | * Find the swap type that corresponds to given device (if any). |
f577eb30 | 1753 | * |
915bae9e RW |
1754 | * @offset - number of the PAGE_SIZE-sized block of the device, starting |
1755 | * from 0, in which the swap header is expected to be located. | |
1756 | * | |
1757 | * This is needed for the suspend to disk (aka swsusp). | |
f577eb30 | 1758 | */ |
7bf23687 | 1759 | int swap_type_of(dev_t device, sector_t offset, struct block_device **bdev_p) |
f577eb30 | 1760 | { |
915bae9e | 1761 | struct block_device *bdev = NULL; |
efa90a98 | 1762 | int type; |
f577eb30 | 1763 | |
915bae9e RW |
1764 | if (device) |
1765 | bdev = bdget(device); | |
1766 | ||
f577eb30 | 1767 | spin_lock(&swap_lock); |
efa90a98 HD |
1768 | for (type = 0; type < nr_swapfiles; type++) { |
1769 | struct swap_info_struct *sis = swap_info[type]; | |
f577eb30 | 1770 | |
915bae9e | 1771 | if (!(sis->flags & SWP_WRITEOK)) |
f577eb30 | 1772 | continue; |
b6b5bce3 | 1773 | |
915bae9e | 1774 | if (!bdev) { |
7bf23687 | 1775 | if (bdev_p) |
dddac6a7 | 1776 | *bdev_p = bdgrab(sis->bdev); |
7bf23687 | 1777 | |
6e1819d6 | 1778 | spin_unlock(&swap_lock); |
efa90a98 | 1779 | return type; |
6e1819d6 | 1780 | } |
915bae9e | 1781 | if (bdev == sis->bdev) { |
4efaceb1 | 1782 | struct swap_extent *se = first_se(sis); |
915bae9e | 1783 | |
915bae9e | 1784 | if (se->start_block == offset) { |
7bf23687 | 1785 | if (bdev_p) |
dddac6a7 | 1786 | *bdev_p = bdgrab(sis->bdev); |
7bf23687 | 1787 | |
915bae9e RW |
1788 | spin_unlock(&swap_lock); |
1789 | bdput(bdev); | |
efa90a98 | 1790 | return type; |
915bae9e | 1791 | } |
f577eb30 RW |
1792 | } |
1793 | } | |
1794 | spin_unlock(&swap_lock); | |
915bae9e RW |
1795 | if (bdev) |
1796 | bdput(bdev); | |
1797 | ||
f577eb30 RW |
1798 | return -ENODEV; |
1799 | } | |
1800 | ||
73c34b6a HD |
1801 | /* |
1802 | * Get the (PAGE_SIZE) block corresponding to given offset on the swapdev | |
1803 | * corresponding to given index in swap_info (swap type). | |
1804 | */ | |
1805 | sector_t swapdev_block(int type, pgoff_t offset) | |
1806 | { | |
1807 | struct block_device *bdev; | |
c10d38cc | 1808 | struct swap_info_struct *si = swap_type_to_swap_info(type); |
73c34b6a | 1809 | |
c10d38cc | 1810 | if (!si || !(si->flags & SWP_WRITEOK)) |
73c34b6a | 1811 | return 0; |
d4906e1a | 1812 | return map_swap_entry(swp_entry(type, offset), &bdev); |
73c34b6a HD |
1813 | } |
1814 | ||
f577eb30 RW |
1815 | /* |
1816 | * Return either the total number of swap pages of given type, or the number | |
1817 | * of free pages of that type (depending on @free) | |
1818 | * | |
1819 | * This is needed for software suspend | |
1820 | */ | |
1821 | unsigned int count_swap_pages(int type, int free) | |
1822 | { | |
1823 | unsigned int n = 0; | |
1824 | ||
efa90a98 HD |
1825 | spin_lock(&swap_lock); |
1826 | if ((unsigned int)type < nr_swapfiles) { | |
1827 | struct swap_info_struct *sis = swap_info[type]; | |
1828 | ||
ec8acf20 | 1829 | spin_lock(&sis->lock); |
efa90a98 HD |
1830 | if (sis->flags & SWP_WRITEOK) { |
1831 | n = sis->pages; | |
f577eb30 | 1832 | if (free) |
efa90a98 | 1833 | n -= sis->inuse_pages; |
f577eb30 | 1834 | } |
ec8acf20 | 1835 | spin_unlock(&sis->lock); |
f577eb30 | 1836 | } |
efa90a98 | 1837 | spin_unlock(&swap_lock); |
f577eb30 RW |
1838 | return n; |
1839 | } | |
73c34b6a | 1840 | #endif /* CONFIG_HIBERNATION */ |
f577eb30 | 1841 | |
9f8bdb3f | 1842 | static inline int pte_same_as_swp(pte_t pte, pte_t swp_pte) |
179ef71c | 1843 | { |
9f8bdb3f | 1844 | return pte_same(pte_swp_clear_soft_dirty(pte), swp_pte); |
179ef71c CG |
1845 | } |
1846 | ||
1da177e4 | 1847 | /* |
72866f6f HD |
1848 | * No need to decide whether this PTE shares the swap entry with others, |
1849 | * just let do_wp_page work it out if a write is requested later - to | |
1850 | * force COW, vm_page_prot omits write permission from any private vma. | |
1da177e4 | 1851 | */ |
044d66c1 | 1852 | static int unuse_pte(struct vm_area_struct *vma, pmd_t *pmd, |
1da177e4 LT |
1853 | unsigned long addr, swp_entry_t entry, struct page *page) |
1854 | { | |
9e16b7fb | 1855 | struct page *swapcache; |
72835c86 | 1856 | struct mem_cgroup *memcg; |
044d66c1 HD |
1857 | spinlock_t *ptl; |
1858 | pte_t *pte; | |
1859 | int ret = 1; | |
1860 | ||
9e16b7fb HD |
1861 | swapcache = page; |
1862 | page = ksm_might_need_to_copy(page, vma, addr); | |
1863 | if (unlikely(!page)) | |
1864 | return -ENOMEM; | |
1865 | ||
f627c2f5 KS |
1866 | if (mem_cgroup_try_charge(page, vma->vm_mm, GFP_KERNEL, |
1867 | &memcg, false)) { | |
044d66c1 | 1868 | ret = -ENOMEM; |
85d9fc89 KH |
1869 | goto out_nolock; |
1870 | } | |
044d66c1 HD |
1871 | |
1872 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); | |
9f8bdb3f | 1873 | if (unlikely(!pte_same_as_swp(*pte, swp_entry_to_pte(entry)))) { |
f627c2f5 | 1874 | mem_cgroup_cancel_charge(page, memcg, false); |
044d66c1 HD |
1875 | ret = 0; |
1876 | goto out; | |
1877 | } | |
8a9f3ccd | 1878 | |
b084d435 | 1879 | dec_mm_counter(vma->vm_mm, MM_SWAPENTS); |
d559db08 | 1880 | inc_mm_counter(vma->vm_mm, MM_ANONPAGES); |
1da177e4 LT |
1881 | get_page(page); |
1882 | set_pte_at(vma->vm_mm, addr, pte, | |
1883 | pte_mkold(mk_pte(page, vma->vm_page_prot))); | |
00501b53 | 1884 | if (page == swapcache) { |
d281ee61 | 1885 | page_add_anon_rmap(page, vma, addr, false); |
f627c2f5 | 1886 | mem_cgroup_commit_charge(page, memcg, true, false); |
00501b53 | 1887 | } else { /* ksm created a completely new copy */ |
d281ee61 | 1888 | page_add_new_anon_rmap(page, vma, addr, false); |
f627c2f5 | 1889 | mem_cgroup_commit_charge(page, memcg, false, false); |
00501b53 JW |
1890 | lru_cache_add_active_or_unevictable(page, vma); |
1891 | } | |
1da177e4 LT |
1892 | swap_free(entry); |
1893 | /* | |
1894 | * Move the page to the active list so it is not | |
1895 | * immediately swapped out again after swapon. | |
1896 | */ | |
1897 | activate_page(page); | |
044d66c1 HD |
1898 | out: |
1899 | pte_unmap_unlock(pte, ptl); | |
85d9fc89 | 1900 | out_nolock: |
9e16b7fb HD |
1901 | if (page != swapcache) { |
1902 | unlock_page(page); | |
1903 | put_page(page); | |
1904 | } | |
044d66c1 | 1905 | return ret; |
1da177e4 LT |
1906 | } |
1907 | ||
1908 | static int unuse_pte_range(struct vm_area_struct *vma, pmd_t *pmd, | |
b56a2d8a VRP |
1909 | unsigned long addr, unsigned long end, |
1910 | unsigned int type, bool frontswap, | |
1911 | unsigned long *fs_pages_to_unuse) | |
1da177e4 | 1912 | { |
b56a2d8a VRP |
1913 | struct page *page; |
1914 | swp_entry_t entry; | |
705e87c0 | 1915 | pte_t *pte; |
b56a2d8a VRP |
1916 | struct swap_info_struct *si; |
1917 | unsigned long offset; | |
8a9f3ccd | 1918 | int ret = 0; |
b56a2d8a | 1919 | volatile unsigned char *swap_map; |
1da177e4 | 1920 | |
b56a2d8a | 1921 | si = swap_info[type]; |
044d66c1 | 1922 | pte = pte_offset_map(pmd, addr); |
1da177e4 | 1923 | do { |
b56a2d8a VRP |
1924 | struct vm_fault vmf; |
1925 | ||
1926 | if (!is_swap_pte(*pte)) | |
1927 | continue; | |
1928 | ||
1929 | entry = pte_to_swp_entry(*pte); | |
1930 | if (swp_type(entry) != type) | |
1931 | continue; | |
1932 | ||
1933 | offset = swp_offset(entry); | |
1934 | if (frontswap && !frontswap_test(si, offset)) | |
1935 | continue; | |
1936 | ||
1937 | pte_unmap(pte); | |
1938 | swap_map = &si->swap_map[offset]; | |
ebc5951e AR |
1939 | page = lookup_swap_cache(entry, vma, addr); |
1940 | if (!page) { | |
1941 | vmf.vma = vma; | |
1942 | vmf.address = addr; | |
1943 | vmf.pmd = pmd; | |
1944 | page = swapin_readahead(entry, GFP_HIGHUSER_MOVABLE, | |
1945 | &vmf); | |
1946 | } | |
b56a2d8a VRP |
1947 | if (!page) { |
1948 | if (*swap_map == 0 || *swap_map == SWAP_MAP_BAD) | |
1949 | goto try_next; | |
1950 | return -ENOMEM; | |
1951 | } | |
1952 | ||
1953 | lock_page(page); | |
1954 | wait_on_page_writeback(page); | |
1955 | ret = unuse_pte(vma, pmd, addr, entry, page); | |
1956 | if (ret < 0) { | |
1957 | unlock_page(page); | |
1958 | put_page(page); | |
1959 | goto out; | |
1960 | } | |
1961 | ||
1962 | try_to_free_swap(page); | |
1963 | unlock_page(page); | |
1964 | put_page(page); | |
1965 | ||
1966 | if (*fs_pages_to_unuse && !--(*fs_pages_to_unuse)) { | |
1967 | ret = FRONTSWAP_PAGES_UNUSED; | |
1968 | goto out; | |
1da177e4 | 1969 | } |
b56a2d8a VRP |
1970 | try_next: |
1971 | pte = pte_offset_map(pmd, addr); | |
1da177e4 | 1972 | } while (pte++, addr += PAGE_SIZE, addr != end); |
044d66c1 | 1973 | pte_unmap(pte - 1); |
b56a2d8a VRP |
1974 | |
1975 | ret = 0; | |
044d66c1 | 1976 | out: |
8a9f3ccd | 1977 | return ret; |
1da177e4 LT |
1978 | } |
1979 | ||
1980 | static inline int unuse_pmd_range(struct vm_area_struct *vma, pud_t *pud, | |
1981 | unsigned long addr, unsigned long end, | |
b56a2d8a VRP |
1982 | unsigned int type, bool frontswap, |
1983 | unsigned long *fs_pages_to_unuse) | |
1da177e4 LT |
1984 | { |
1985 | pmd_t *pmd; | |
1986 | unsigned long next; | |
8a9f3ccd | 1987 | int ret; |
1da177e4 LT |
1988 | |
1989 | pmd = pmd_offset(pud, addr); | |
1990 | do { | |
dc644a07 | 1991 | cond_resched(); |
1da177e4 | 1992 | next = pmd_addr_end(addr, end); |
1a5a9906 | 1993 | if (pmd_none_or_trans_huge_or_clear_bad(pmd)) |
1da177e4 | 1994 | continue; |
b56a2d8a VRP |
1995 | ret = unuse_pte_range(vma, pmd, addr, next, type, |
1996 | frontswap, fs_pages_to_unuse); | |
8a9f3ccd BS |
1997 | if (ret) |
1998 | return ret; | |
1da177e4 LT |
1999 | } while (pmd++, addr = next, addr != end); |
2000 | return 0; | |
2001 | } | |
2002 | ||
c2febafc | 2003 | static inline int unuse_pud_range(struct vm_area_struct *vma, p4d_t *p4d, |
1da177e4 | 2004 | unsigned long addr, unsigned long end, |
b56a2d8a VRP |
2005 | unsigned int type, bool frontswap, |
2006 | unsigned long *fs_pages_to_unuse) | |
1da177e4 LT |
2007 | { |
2008 | pud_t *pud; | |
2009 | unsigned long next; | |
8a9f3ccd | 2010 | int ret; |
1da177e4 | 2011 | |
c2febafc | 2012 | pud = pud_offset(p4d, addr); |
1da177e4 LT |
2013 | do { |
2014 | next = pud_addr_end(addr, end); | |
2015 | if (pud_none_or_clear_bad(pud)) | |
2016 | continue; | |
b56a2d8a VRP |
2017 | ret = unuse_pmd_range(vma, pud, addr, next, type, |
2018 | frontswap, fs_pages_to_unuse); | |
8a9f3ccd BS |
2019 | if (ret) |
2020 | return ret; | |
1da177e4 LT |
2021 | } while (pud++, addr = next, addr != end); |
2022 | return 0; | |
2023 | } | |
2024 | ||
c2febafc KS |
2025 | static inline int unuse_p4d_range(struct vm_area_struct *vma, pgd_t *pgd, |
2026 | unsigned long addr, unsigned long end, | |
b56a2d8a VRP |
2027 | unsigned int type, bool frontswap, |
2028 | unsigned long *fs_pages_to_unuse) | |
c2febafc KS |
2029 | { |
2030 | p4d_t *p4d; | |
2031 | unsigned long next; | |
2032 | int ret; | |
2033 | ||
2034 | p4d = p4d_offset(pgd, addr); | |
2035 | do { | |
2036 | next = p4d_addr_end(addr, end); | |
2037 | if (p4d_none_or_clear_bad(p4d)) | |
2038 | continue; | |
b56a2d8a VRP |
2039 | ret = unuse_pud_range(vma, p4d, addr, next, type, |
2040 | frontswap, fs_pages_to_unuse); | |
c2febafc KS |
2041 | if (ret) |
2042 | return ret; | |
2043 | } while (p4d++, addr = next, addr != end); | |
2044 | return 0; | |
2045 | } | |
2046 | ||
b56a2d8a VRP |
2047 | static int unuse_vma(struct vm_area_struct *vma, unsigned int type, |
2048 | bool frontswap, unsigned long *fs_pages_to_unuse) | |
1da177e4 LT |
2049 | { |
2050 | pgd_t *pgd; | |
2051 | unsigned long addr, end, next; | |
8a9f3ccd | 2052 | int ret; |
1da177e4 | 2053 | |
b56a2d8a VRP |
2054 | addr = vma->vm_start; |
2055 | end = vma->vm_end; | |
1da177e4 LT |
2056 | |
2057 | pgd = pgd_offset(vma->vm_mm, addr); | |
2058 | do { | |
2059 | next = pgd_addr_end(addr, end); | |
2060 | if (pgd_none_or_clear_bad(pgd)) | |
2061 | continue; | |
b56a2d8a VRP |
2062 | ret = unuse_p4d_range(vma, pgd, addr, next, type, |
2063 | frontswap, fs_pages_to_unuse); | |
8a9f3ccd BS |
2064 | if (ret) |
2065 | return ret; | |
1da177e4 LT |
2066 | } while (pgd++, addr = next, addr != end); |
2067 | return 0; | |
2068 | } | |
2069 | ||
b56a2d8a VRP |
2070 | static int unuse_mm(struct mm_struct *mm, unsigned int type, |
2071 | bool frontswap, unsigned long *fs_pages_to_unuse) | |
1da177e4 LT |
2072 | { |
2073 | struct vm_area_struct *vma; | |
8a9f3ccd | 2074 | int ret = 0; |
1da177e4 | 2075 | |
b56a2d8a | 2076 | down_read(&mm->mmap_sem); |
1da177e4 | 2077 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
b56a2d8a VRP |
2078 | if (vma->anon_vma) { |
2079 | ret = unuse_vma(vma, type, frontswap, | |
2080 | fs_pages_to_unuse); | |
2081 | if (ret) | |
2082 | break; | |
2083 | } | |
dc644a07 | 2084 | cond_resched(); |
1da177e4 | 2085 | } |
1da177e4 | 2086 | up_read(&mm->mmap_sem); |
b56a2d8a | 2087 | return ret; |
1da177e4 LT |
2088 | } |
2089 | ||
2090 | /* | |
38b5faf4 | 2091 | * Scan swap_map (or frontswap_map if frontswap parameter is true) |
b56a2d8a VRP |
2092 | * from current position to next entry still in use. Return 0 |
2093 | * if there are no inuse entries after prev till end of the map. | |
1da177e4 | 2094 | */ |
6eb396dc | 2095 | static unsigned int find_next_to_unuse(struct swap_info_struct *si, |
38b5faf4 | 2096 | unsigned int prev, bool frontswap) |
1da177e4 | 2097 | { |
b56a2d8a | 2098 | unsigned int i; |
8d69aaee | 2099 | unsigned char count; |
1da177e4 LT |
2100 | |
2101 | /* | |
5d337b91 | 2102 | * No need for swap_lock here: we're just looking |
1da177e4 LT |
2103 | * for whether an entry is in use, not modifying it; false |
2104 | * hits are okay, and sys_swapoff() has already prevented new | |
5d337b91 | 2105 | * allocations from this area (while holding swap_lock). |
1da177e4 | 2106 | */ |
b56a2d8a | 2107 | for (i = prev + 1; i < si->max; i++) { |
4db0c3c2 | 2108 | count = READ_ONCE(si->swap_map[i]); |
355cfa73 | 2109 | if (count && swap_count(count) != SWAP_MAP_BAD) |
dc644a07 HD |
2110 | if (!frontswap || frontswap_test(si, i)) |
2111 | break; | |
2112 | if ((i % LATENCY_LIMIT) == 0) | |
2113 | cond_resched(); | |
1da177e4 | 2114 | } |
b56a2d8a VRP |
2115 | |
2116 | if (i == si->max) | |
2117 | i = 0; | |
2118 | ||
1da177e4 LT |
2119 | return i; |
2120 | } | |
2121 | ||
2122 | /* | |
b56a2d8a | 2123 | * If the boolean frontswap is true, only unuse pages_to_unuse pages; |
38b5faf4 | 2124 | * pages_to_unuse==0 means all pages; ignored if frontswap is false |
1da177e4 | 2125 | */ |
38b5faf4 DM |
2126 | int try_to_unuse(unsigned int type, bool frontswap, |
2127 | unsigned long pages_to_unuse) | |
1da177e4 | 2128 | { |
b56a2d8a VRP |
2129 | struct mm_struct *prev_mm; |
2130 | struct mm_struct *mm; | |
2131 | struct list_head *p; | |
2132 | int retval = 0; | |
efa90a98 | 2133 | struct swap_info_struct *si = swap_info[type]; |
1da177e4 LT |
2134 | struct page *page; |
2135 | swp_entry_t entry; | |
b56a2d8a | 2136 | unsigned int i; |
1da177e4 | 2137 | |
21820948 | 2138 | if (!READ_ONCE(si->inuse_pages)) |
b56a2d8a | 2139 | return 0; |
1da177e4 | 2140 | |
b56a2d8a VRP |
2141 | if (!frontswap) |
2142 | pages_to_unuse = 0; | |
2143 | ||
2144 | retry: | |
2145 | retval = shmem_unuse(type, frontswap, &pages_to_unuse); | |
2146 | if (retval) | |
2147 | goto out; | |
2148 | ||
2149 | prev_mm = &init_mm; | |
2150 | mmget(prev_mm); | |
2151 | ||
2152 | spin_lock(&mmlist_lock); | |
2153 | p = &init_mm.mmlist; | |
21820948 | 2154 | while (READ_ONCE(si->inuse_pages) && |
64165b1a HD |
2155 | !signal_pending(current) && |
2156 | (p = p->next) != &init_mm.mmlist) { | |
1da177e4 | 2157 | |
b56a2d8a VRP |
2158 | mm = list_entry(p, struct mm_struct, mmlist); |
2159 | if (!mmget_not_zero(mm)) | |
2160 | continue; | |
2161 | spin_unlock(&mmlist_lock); | |
2162 | mmput(prev_mm); | |
2163 | prev_mm = mm; | |
2164 | retval = unuse_mm(mm, type, frontswap, &pages_to_unuse); | |
1da177e4 | 2165 | |
b56a2d8a VRP |
2166 | if (retval) { |
2167 | mmput(prev_mm); | |
2168 | goto out; | |
1da177e4 LT |
2169 | } |
2170 | ||
2171 | /* | |
b56a2d8a VRP |
2172 | * Make sure that we aren't completely killing |
2173 | * interactive performance. | |
1da177e4 | 2174 | */ |
b56a2d8a VRP |
2175 | cond_resched(); |
2176 | spin_lock(&mmlist_lock); | |
2177 | } | |
2178 | spin_unlock(&mmlist_lock); | |
1da177e4 | 2179 | |
b56a2d8a | 2180 | mmput(prev_mm); |
1da177e4 | 2181 | |
b56a2d8a | 2182 | i = 0; |
21820948 | 2183 | while (READ_ONCE(si->inuse_pages) && |
64165b1a HD |
2184 | !signal_pending(current) && |
2185 | (i = find_next_to_unuse(si, i, frontswap)) != 0) { | |
1da177e4 | 2186 | |
b56a2d8a VRP |
2187 | entry = swp_entry(type, i); |
2188 | page = find_get_page(swap_address_space(entry), i); | |
2189 | if (!page) | |
2190 | continue; | |
68bdc8d6 HD |
2191 | |
2192 | /* | |
2193 | * It is conceivable that a racing task removed this page from | |
b56a2d8a VRP |
2194 | * swap cache just before we acquired the page lock. The page |
2195 | * might even be back in swap cache on another swap area. But | |
2196 | * that is okay, try_to_free_swap() only removes stale pages. | |
1da177e4 | 2197 | */ |
b56a2d8a VRP |
2198 | lock_page(page); |
2199 | wait_on_page_writeback(page); | |
2200 | try_to_free_swap(page); | |
1da177e4 | 2201 | unlock_page(page); |
09cbfeaf | 2202 | put_page(page); |
1da177e4 LT |
2203 | |
2204 | /* | |
b56a2d8a VRP |
2205 | * For frontswap, we just need to unuse pages_to_unuse, if |
2206 | * it was specified. Need not check frontswap again here as | |
2207 | * we already zeroed out pages_to_unuse if not frontswap. | |
1da177e4 | 2208 | */ |
b56a2d8a VRP |
2209 | if (pages_to_unuse && --pages_to_unuse == 0) |
2210 | goto out; | |
1da177e4 LT |
2211 | } |
2212 | ||
b56a2d8a VRP |
2213 | /* |
2214 | * Lets check again to see if there are still swap entries in the map. | |
2215 | * If yes, we would need to do retry the unuse logic again. | |
2216 | * Under global memory pressure, swap entries can be reinserted back | |
2217 | * into process space after the mmlist loop above passes over them. | |
dd862deb | 2218 | * |
af53d3e9 HD |
2219 | * Limit the number of retries? No: when mmget_not_zero() above fails, |
2220 | * that mm is likely to be freeing swap from exit_mmap(), which proceeds | |
2221 | * at its own independent pace; and even shmem_writepage() could have | |
2222 | * been preempted after get_swap_page(), temporarily hiding that swap. | |
2223 | * It's easy and robust (though cpu-intensive) just to keep retrying. | |
b56a2d8a | 2224 | */ |
21820948 | 2225 | if (READ_ONCE(si->inuse_pages)) { |
64165b1a HD |
2226 | if (!signal_pending(current)) |
2227 | goto retry; | |
2228 | retval = -EINTR; | |
2229 | } | |
b56a2d8a VRP |
2230 | out: |
2231 | return (retval == FRONTSWAP_PAGES_UNUSED) ? 0 : retval; | |
1da177e4 LT |
2232 | } |
2233 | ||
2234 | /* | |
5d337b91 HD |
2235 | * After a successful try_to_unuse, if no swap is now in use, we know |
2236 | * we can empty the mmlist. swap_lock must be held on entry and exit. | |
2237 | * Note that mmlist_lock nests inside swap_lock, and an mm must be | |
1da177e4 LT |
2238 | * added to the mmlist just after page_duplicate - before would be racy. |
2239 | */ | |
2240 | static void drain_mmlist(void) | |
2241 | { | |
2242 | struct list_head *p, *next; | |
efa90a98 | 2243 | unsigned int type; |
1da177e4 | 2244 | |
efa90a98 HD |
2245 | for (type = 0; type < nr_swapfiles; type++) |
2246 | if (swap_info[type]->inuse_pages) | |
1da177e4 LT |
2247 | return; |
2248 | spin_lock(&mmlist_lock); | |
2249 | list_for_each_safe(p, next, &init_mm.mmlist) | |
2250 | list_del_init(p); | |
2251 | spin_unlock(&mmlist_lock); | |
2252 | } | |
2253 | ||
2254 | /* | |
2255 | * Use this swapdev's extent info to locate the (PAGE_SIZE) block which | |
d4906e1a LS |
2256 | * corresponds to page offset for the specified swap entry. |
2257 | * Note that the type of this function is sector_t, but it returns page offset | |
2258 | * into the bdev, not sector offset. | |
1da177e4 | 2259 | */ |
d4906e1a | 2260 | static sector_t map_swap_entry(swp_entry_t entry, struct block_device **bdev) |
1da177e4 | 2261 | { |
f29ad6a9 | 2262 | struct swap_info_struct *sis; |
f29ad6a9 HD |
2263 | struct swap_extent *se; |
2264 | pgoff_t offset; | |
2265 | ||
c10d38cc | 2266 | sis = swp_swap_info(entry); |
f29ad6a9 HD |
2267 | *bdev = sis->bdev; |
2268 | ||
2269 | offset = swp_offset(entry); | |
4efaceb1 AL |
2270 | se = offset_to_swap_extent(sis, offset); |
2271 | return se->start_block + (offset - se->start_page); | |
1da177e4 LT |
2272 | } |
2273 | ||
d4906e1a LS |
2274 | /* |
2275 | * Returns the page offset into bdev for the specified page's swap entry. | |
2276 | */ | |
2277 | sector_t map_swap_page(struct page *page, struct block_device **bdev) | |
2278 | { | |
2279 | swp_entry_t entry; | |
2280 | entry.val = page_private(page); | |
2281 | return map_swap_entry(entry, bdev); | |
2282 | } | |
2283 | ||
1da177e4 LT |
2284 | /* |
2285 | * Free all of a swapdev's extent information | |
2286 | */ | |
2287 | static void destroy_swap_extents(struct swap_info_struct *sis) | |
2288 | { | |
4efaceb1 AL |
2289 | while (!RB_EMPTY_ROOT(&sis->swap_extent_root)) { |
2290 | struct rb_node *rb = sis->swap_extent_root.rb_node; | |
2291 | struct swap_extent *se = rb_entry(rb, struct swap_extent, rb_node); | |
1da177e4 | 2292 | |
4efaceb1 | 2293 | rb_erase(rb, &sis->swap_extent_root); |
1da177e4 LT |
2294 | kfree(se); |
2295 | } | |
62c230bc | 2296 | |
bc4ae27d | 2297 | if (sis->flags & SWP_ACTIVATED) { |
62c230bc MG |
2298 | struct file *swap_file = sis->swap_file; |
2299 | struct address_space *mapping = swap_file->f_mapping; | |
2300 | ||
bc4ae27d OS |
2301 | sis->flags &= ~SWP_ACTIVATED; |
2302 | if (mapping->a_ops->swap_deactivate) | |
2303 | mapping->a_ops->swap_deactivate(swap_file); | |
62c230bc | 2304 | } |
1da177e4 LT |
2305 | } |
2306 | ||
2307 | /* | |
2308 | * Add a block range (and the corresponding page range) into this swapdev's | |
4efaceb1 | 2309 | * extent tree. |
1da177e4 | 2310 | * |
11d31886 | 2311 | * This function rather assumes that it is called in ascending page order. |
1da177e4 | 2312 | */ |
a509bc1a | 2313 | int |
1da177e4 LT |
2314 | add_swap_extent(struct swap_info_struct *sis, unsigned long start_page, |
2315 | unsigned long nr_pages, sector_t start_block) | |
2316 | { | |
4efaceb1 | 2317 | struct rb_node **link = &sis->swap_extent_root.rb_node, *parent = NULL; |
1da177e4 LT |
2318 | struct swap_extent *se; |
2319 | struct swap_extent *new_se; | |
4efaceb1 AL |
2320 | |
2321 | /* | |
2322 | * place the new node at the right most since the | |
2323 | * function is called in ascending page order. | |
2324 | */ | |
2325 | while (*link) { | |
2326 | parent = *link; | |
2327 | link = &parent->rb_right; | |
2328 | } | |
2329 | ||
2330 | if (parent) { | |
2331 | se = rb_entry(parent, struct swap_extent, rb_node); | |
11d31886 HD |
2332 | BUG_ON(se->start_page + se->nr_pages != start_page); |
2333 | if (se->start_block + se->nr_pages == start_block) { | |
1da177e4 LT |
2334 | /* Merge it */ |
2335 | se->nr_pages += nr_pages; | |
2336 | return 0; | |
2337 | } | |
1da177e4 LT |
2338 | } |
2339 | ||
4efaceb1 | 2340 | /* No merge, insert a new extent. */ |
1da177e4 LT |
2341 | new_se = kmalloc(sizeof(*se), GFP_KERNEL); |
2342 | if (new_se == NULL) | |
2343 | return -ENOMEM; | |
2344 | new_se->start_page = start_page; | |
2345 | new_se->nr_pages = nr_pages; | |
2346 | new_se->start_block = start_block; | |
2347 | ||
4efaceb1 AL |
2348 | rb_link_node(&new_se->rb_node, parent, link); |
2349 | rb_insert_color(&new_se->rb_node, &sis->swap_extent_root); | |
53092a74 | 2350 | return 1; |
1da177e4 | 2351 | } |
aa8aa8a3 | 2352 | EXPORT_SYMBOL_GPL(add_swap_extent); |
1da177e4 LT |
2353 | |
2354 | /* | |
2355 | * A `swap extent' is a simple thing which maps a contiguous range of pages | |
2356 | * onto a contiguous range of disk blocks. An ordered list of swap extents | |
2357 | * is built at swapon time and is then used at swap_writepage/swap_readpage | |
2358 | * time for locating where on disk a page belongs. | |
2359 | * | |
2360 | * If the swapfile is an S_ISBLK block device, a single extent is installed. | |
2361 | * This is done so that the main operating code can treat S_ISBLK and S_ISREG | |
2362 | * swap files identically. | |
2363 | * | |
2364 | * Whether the swapdev is an S_ISREG file or an S_ISBLK blockdev, the swap | |
2365 | * extent list operates in PAGE_SIZE disk blocks. Both S_ISREG and S_ISBLK | |
2366 | * swapfiles are handled *identically* after swapon time. | |
2367 | * | |
2368 | * For S_ISREG swapfiles, setup_swap_extents() will walk all the file's blocks | |
2369 | * and will parse them into an ordered extent list, in PAGE_SIZE chunks. If | |
2370 | * some stray blocks are found which do not fall within the PAGE_SIZE alignment | |
2371 | * requirements, they are simply tossed out - we will never use those blocks | |
2372 | * for swapping. | |
2373 | * | |
1638045c DW |
2374 | * For all swap devices we set S_SWAPFILE across the life of the swapon. This |
2375 | * prevents users from writing to the swap device, which will corrupt memory. | |
1da177e4 LT |
2376 | * |
2377 | * The amount of disk space which a single swap extent represents varies. | |
2378 | * Typically it is in the 1-4 megabyte range. So we can have hundreds of | |
2379 | * extents in the list. To avoid much list walking, we cache the previous | |
2380 | * search location in `curr_swap_extent', and start new searches from there. | |
2381 | * This is extremely effective. The average number of iterations in | |
2382 | * map_swap_page() has been measured at about 0.3 per page. - akpm. | |
2383 | */ | |
53092a74 | 2384 | static int setup_swap_extents(struct swap_info_struct *sis, sector_t *span) |
1da177e4 | 2385 | { |
62c230bc MG |
2386 | struct file *swap_file = sis->swap_file; |
2387 | struct address_space *mapping = swap_file->f_mapping; | |
2388 | struct inode *inode = mapping->host; | |
1da177e4 LT |
2389 | int ret; |
2390 | ||
1da177e4 LT |
2391 | if (S_ISBLK(inode->i_mode)) { |
2392 | ret = add_swap_extent(sis, 0, sis->max, 0); | |
53092a74 | 2393 | *span = sis->pages; |
a509bc1a | 2394 | return ret; |
1da177e4 LT |
2395 | } |
2396 | ||
62c230bc | 2397 | if (mapping->a_ops->swap_activate) { |
a509bc1a | 2398 | ret = mapping->a_ops->swap_activate(sis, swap_file, span); |
bc4ae27d OS |
2399 | if (ret >= 0) |
2400 | sis->flags |= SWP_ACTIVATED; | |
62c230bc | 2401 | if (!ret) { |
bc4ae27d | 2402 | sis->flags |= SWP_FS; |
62c230bc MG |
2403 | ret = add_swap_extent(sis, 0, sis->max, 0); |
2404 | *span = sis->pages; | |
2405 | } | |
a509bc1a | 2406 | return ret; |
62c230bc MG |
2407 | } |
2408 | ||
a509bc1a | 2409 | return generic_swapfile_activate(sis, swap_file, span); |
1da177e4 LT |
2410 | } |
2411 | ||
a2468cc9 AL |
2412 | static int swap_node(struct swap_info_struct *p) |
2413 | { | |
2414 | struct block_device *bdev; | |
2415 | ||
2416 | if (p->bdev) | |
2417 | bdev = p->bdev; | |
2418 | else | |
2419 | bdev = p->swap_file->f_inode->i_sb->s_bdev; | |
2420 | ||
2421 | return bdev ? bdev->bd_disk->node_id : NUMA_NO_NODE; | |
2422 | } | |
2423 | ||
eb085574 HY |
2424 | static void setup_swap_info(struct swap_info_struct *p, int prio, |
2425 | unsigned char *swap_map, | |
2426 | struct swap_cluster_info *cluster_info) | |
40531542 | 2427 | { |
a2468cc9 AL |
2428 | int i; |
2429 | ||
40531542 CEB |
2430 | if (prio >= 0) |
2431 | p->prio = prio; | |
2432 | else | |
2433 | p->prio = --least_priority; | |
18ab4d4c DS |
2434 | /* |
2435 | * the plist prio is negated because plist ordering is | |
2436 | * low-to-high, while swap ordering is high-to-low | |
2437 | */ | |
2438 | p->list.prio = -p->prio; | |
a2468cc9 AL |
2439 | for_each_node(i) { |
2440 | if (p->prio >= 0) | |
2441 | p->avail_lists[i].prio = -p->prio; | |
2442 | else { | |
2443 | if (swap_node(p) == i) | |
2444 | p->avail_lists[i].prio = 1; | |
2445 | else | |
2446 | p->avail_lists[i].prio = -p->prio; | |
2447 | } | |
2448 | } | |
40531542 | 2449 | p->swap_map = swap_map; |
2a8f9449 | 2450 | p->cluster_info = cluster_info; |
eb085574 HY |
2451 | } |
2452 | ||
2453 | static void _enable_swap_info(struct swap_info_struct *p) | |
2454 | { | |
2455 | p->flags |= SWP_WRITEOK | SWP_VALID; | |
ec8acf20 | 2456 | atomic_long_add(p->pages, &nr_swap_pages); |
40531542 CEB |
2457 | total_swap_pages += p->pages; |
2458 | ||
adfab836 | 2459 | assert_spin_locked(&swap_lock); |
adfab836 | 2460 | /* |
18ab4d4c DS |
2461 | * both lists are plists, and thus priority ordered. |
2462 | * swap_active_head needs to be priority ordered for swapoff(), | |
2463 | * which on removal of any swap_info_struct with an auto-assigned | |
2464 | * (i.e. negative) priority increments the auto-assigned priority | |
2465 | * of any lower-priority swap_info_structs. | |
2466 | * swap_avail_head needs to be priority ordered for get_swap_page(), | |
2467 | * which allocates swap pages from the highest available priority | |
2468 | * swap_info_struct. | |
adfab836 | 2469 | */ |
18ab4d4c | 2470 | plist_add(&p->list, &swap_active_head); |
a2468cc9 | 2471 | add_to_avail_list(p); |
cf0cac0a CEB |
2472 | } |
2473 | ||
2474 | static void enable_swap_info(struct swap_info_struct *p, int prio, | |
2475 | unsigned char *swap_map, | |
2a8f9449 | 2476 | struct swap_cluster_info *cluster_info, |
cf0cac0a CEB |
2477 | unsigned long *frontswap_map) |
2478 | { | |
4f89849d | 2479 | frontswap_init(p->type, frontswap_map); |
cf0cac0a | 2480 | spin_lock(&swap_lock); |
ec8acf20 | 2481 | spin_lock(&p->lock); |
eb085574 HY |
2482 | setup_swap_info(p, prio, swap_map, cluster_info); |
2483 | spin_unlock(&p->lock); | |
2484 | spin_unlock(&swap_lock); | |
2485 | /* | |
2486 | * Guarantee swap_map, cluster_info, etc. fields are valid | |
2487 | * between get/put_swap_device() if SWP_VALID bit is set | |
2488 | */ | |
2489 | synchronize_rcu(); | |
2490 | spin_lock(&swap_lock); | |
2491 | spin_lock(&p->lock); | |
2492 | _enable_swap_info(p); | |
ec8acf20 | 2493 | spin_unlock(&p->lock); |
cf0cac0a CEB |
2494 | spin_unlock(&swap_lock); |
2495 | } | |
2496 | ||
2497 | static void reinsert_swap_info(struct swap_info_struct *p) | |
2498 | { | |
2499 | spin_lock(&swap_lock); | |
ec8acf20 | 2500 | spin_lock(&p->lock); |
eb085574 HY |
2501 | setup_swap_info(p, p->prio, p->swap_map, p->cluster_info); |
2502 | _enable_swap_info(p); | |
ec8acf20 | 2503 | spin_unlock(&p->lock); |
40531542 CEB |
2504 | spin_unlock(&swap_lock); |
2505 | } | |
2506 | ||
67afa38e TC |
2507 | bool has_usable_swap(void) |
2508 | { | |
2509 | bool ret = true; | |
2510 | ||
2511 | spin_lock(&swap_lock); | |
2512 | if (plist_head_empty(&swap_active_head)) | |
2513 | ret = false; | |
2514 | spin_unlock(&swap_lock); | |
2515 | return ret; | |
2516 | } | |
2517 | ||
c4ea37c2 | 2518 | SYSCALL_DEFINE1(swapoff, const char __user *, specialfile) |
1da177e4 | 2519 | { |
73c34b6a | 2520 | struct swap_info_struct *p = NULL; |
8d69aaee | 2521 | unsigned char *swap_map; |
2a8f9449 | 2522 | struct swap_cluster_info *cluster_info; |
4f89849d | 2523 | unsigned long *frontswap_map; |
1da177e4 LT |
2524 | struct file *swap_file, *victim; |
2525 | struct address_space *mapping; | |
2526 | struct inode *inode; | |
91a27b2a | 2527 | struct filename *pathname; |
adfab836 | 2528 | int err, found = 0; |
5b808a23 | 2529 | unsigned int old_block_size; |
886bb7e9 | 2530 | |
1da177e4 LT |
2531 | if (!capable(CAP_SYS_ADMIN)) |
2532 | return -EPERM; | |
2533 | ||
191c5424 AV |
2534 | BUG_ON(!current->mm); |
2535 | ||
1da177e4 | 2536 | pathname = getname(specialfile); |
1da177e4 | 2537 | if (IS_ERR(pathname)) |
f58b59c1 | 2538 | return PTR_ERR(pathname); |
1da177e4 | 2539 | |
669abf4e | 2540 | victim = file_open_name(pathname, O_RDWR|O_LARGEFILE, 0); |
1da177e4 LT |
2541 | err = PTR_ERR(victim); |
2542 | if (IS_ERR(victim)) | |
2543 | goto out; | |
2544 | ||
2545 | mapping = victim->f_mapping; | |
5d337b91 | 2546 | spin_lock(&swap_lock); |
18ab4d4c | 2547 | plist_for_each_entry(p, &swap_active_head, list) { |
22c6f8fd | 2548 | if (p->flags & SWP_WRITEOK) { |
adfab836 DS |
2549 | if (p->swap_file->f_mapping == mapping) { |
2550 | found = 1; | |
1da177e4 | 2551 | break; |
adfab836 | 2552 | } |
1da177e4 | 2553 | } |
1da177e4 | 2554 | } |
adfab836 | 2555 | if (!found) { |
1da177e4 | 2556 | err = -EINVAL; |
5d337b91 | 2557 | spin_unlock(&swap_lock); |
1da177e4 LT |
2558 | goto out_dput; |
2559 | } | |
191c5424 | 2560 | if (!security_vm_enough_memory_mm(current->mm, p->pages)) |
1da177e4 LT |
2561 | vm_unacct_memory(p->pages); |
2562 | else { | |
2563 | err = -ENOMEM; | |
5d337b91 | 2564 | spin_unlock(&swap_lock); |
1da177e4 LT |
2565 | goto out_dput; |
2566 | } | |
a2468cc9 | 2567 | del_from_avail_list(p); |
ec8acf20 | 2568 | spin_lock(&p->lock); |
78ecba08 | 2569 | if (p->prio < 0) { |
adfab836 | 2570 | struct swap_info_struct *si = p; |
a2468cc9 | 2571 | int nid; |
adfab836 | 2572 | |
18ab4d4c | 2573 | plist_for_each_entry_continue(si, &swap_active_head, list) { |
adfab836 | 2574 | si->prio++; |
18ab4d4c | 2575 | si->list.prio--; |
a2468cc9 AL |
2576 | for_each_node(nid) { |
2577 | if (si->avail_lists[nid].prio != 1) | |
2578 | si->avail_lists[nid].prio--; | |
2579 | } | |
adfab836 | 2580 | } |
78ecba08 HD |
2581 | least_priority++; |
2582 | } | |
18ab4d4c | 2583 | plist_del(&p->list, &swap_active_head); |
ec8acf20 | 2584 | atomic_long_sub(p->pages, &nr_swap_pages); |
1da177e4 LT |
2585 | total_swap_pages -= p->pages; |
2586 | p->flags &= ~SWP_WRITEOK; | |
ec8acf20 | 2587 | spin_unlock(&p->lock); |
5d337b91 | 2588 | spin_unlock(&swap_lock); |
fb4f88dc | 2589 | |
039939a6 TC |
2590 | disable_swap_slots_cache_lock(); |
2591 | ||
e1e12d2f | 2592 | set_current_oom_origin(); |
adfab836 | 2593 | err = try_to_unuse(p->type, false, 0); /* force unuse all pages */ |
e1e12d2f | 2594 | clear_current_oom_origin(); |
1da177e4 | 2595 | |
1da177e4 LT |
2596 | if (err) { |
2597 | /* re-insert swap space back into swap_list */ | |
cf0cac0a | 2598 | reinsert_swap_info(p); |
039939a6 | 2599 | reenable_swap_slots_cache_unlock(); |
1da177e4 LT |
2600 | goto out_dput; |
2601 | } | |
52b7efdb | 2602 | |
039939a6 TC |
2603 | reenable_swap_slots_cache_unlock(); |
2604 | ||
eb085574 HY |
2605 | spin_lock(&swap_lock); |
2606 | spin_lock(&p->lock); | |
2607 | p->flags &= ~SWP_VALID; /* mark swap device as invalid */ | |
2608 | spin_unlock(&p->lock); | |
2609 | spin_unlock(&swap_lock); | |
2610 | /* | |
2611 | * wait for swap operations protected by get/put_swap_device() | |
2612 | * to complete | |
2613 | */ | |
2614 | synchronize_rcu(); | |
2615 | ||
815c2c54 SL |
2616 | flush_work(&p->discard_work); |
2617 | ||
5d337b91 | 2618 | destroy_swap_extents(p); |
570a335b HD |
2619 | if (p->flags & SWP_CONTINUED) |
2620 | free_swap_count_continuations(p); | |
2621 | ||
81a0298b HY |
2622 | if (!p->bdev || !blk_queue_nonrot(bdev_get_queue(p->bdev))) |
2623 | atomic_dec(&nr_rotate_swap); | |
2624 | ||
fc0abb14 | 2625 | mutex_lock(&swapon_mutex); |
5d337b91 | 2626 | spin_lock(&swap_lock); |
ec8acf20 | 2627 | spin_lock(&p->lock); |
5d337b91 HD |
2628 | drain_mmlist(); |
2629 | ||
52b7efdb | 2630 | /* wait for anyone still in scan_swap_map */ |
52b7efdb HD |
2631 | p->highest_bit = 0; /* cuts scans short */ |
2632 | while (p->flags >= SWP_SCANNING) { | |
ec8acf20 | 2633 | spin_unlock(&p->lock); |
5d337b91 | 2634 | spin_unlock(&swap_lock); |
13e4b57f | 2635 | schedule_timeout_uninterruptible(1); |
5d337b91 | 2636 | spin_lock(&swap_lock); |
ec8acf20 | 2637 | spin_lock(&p->lock); |
52b7efdb | 2638 | } |
52b7efdb | 2639 | |
1da177e4 | 2640 | swap_file = p->swap_file; |
5b808a23 | 2641 | old_block_size = p->old_block_size; |
1da177e4 LT |
2642 | p->swap_file = NULL; |
2643 | p->max = 0; | |
2644 | swap_map = p->swap_map; | |
2645 | p->swap_map = NULL; | |
2a8f9449 SL |
2646 | cluster_info = p->cluster_info; |
2647 | p->cluster_info = NULL; | |
4f89849d | 2648 | frontswap_map = frontswap_map_get(p); |
ec8acf20 | 2649 | spin_unlock(&p->lock); |
5d337b91 | 2650 | spin_unlock(&swap_lock); |
adfab836 | 2651 | frontswap_invalidate_area(p->type); |
58e97ba6 | 2652 | frontswap_map_set(p, NULL); |
fc0abb14 | 2653 | mutex_unlock(&swapon_mutex); |
ebc2a1a6 SL |
2654 | free_percpu(p->percpu_cluster); |
2655 | p->percpu_cluster = NULL; | |
1da177e4 | 2656 | vfree(swap_map); |
54f180d3 HY |
2657 | kvfree(cluster_info); |
2658 | kvfree(frontswap_map); | |
2de1a7e4 | 2659 | /* Destroy swap account information */ |
adfab836 | 2660 | swap_cgroup_swapoff(p->type); |
4b3ef9da | 2661 | exit_swap_address_space(p->type); |
27a7faa0 | 2662 | |
1da177e4 LT |
2663 | inode = mapping->host; |
2664 | if (S_ISBLK(inode->i_mode)) { | |
2665 | struct block_device *bdev = I_BDEV(inode); | |
1638045c | 2666 | |
5b808a23 | 2667 | set_blocksize(bdev, old_block_size); |
e525fd89 | 2668 | blkdev_put(bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL); |
1da177e4 | 2669 | } |
1638045c DW |
2670 | |
2671 | inode_lock(inode); | |
2672 | inode->i_flags &= ~S_SWAPFILE; | |
2673 | inode_unlock(inode); | |
1da177e4 | 2674 | filp_close(swap_file, NULL); |
f893ab41 WY |
2675 | |
2676 | /* | |
2677 | * Clear the SWP_USED flag after all resources are freed so that swapon | |
2678 | * can reuse this swap_info in alloc_swap_info() safely. It is ok to | |
2679 | * not hold p->lock after we cleared its SWP_WRITEOK. | |
2680 | */ | |
2681 | spin_lock(&swap_lock); | |
2682 | p->flags = 0; | |
2683 | spin_unlock(&swap_lock); | |
2684 | ||
1da177e4 | 2685 | err = 0; |
66d7dd51 KS |
2686 | atomic_inc(&proc_poll_event); |
2687 | wake_up_interruptible(&proc_poll_wait); | |
1da177e4 LT |
2688 | |
2689 | out_dput: | |
2690 | filp_close(victim, NULL); | |
2691 | out: | |
f58b59c1 | 2692 | putname(pathname); |
1da177e4 LT |
2693 | return err; |
2694 | } | |
2695 | ||
2696 | #ifdef CONFIG_PROC_FS | |
9dd95748 | 2697 | static __poll_t swaps_poll(struct file *file, poll_table *wait) |
66d7dd51 | 2698 | { |
f1514638 | 2699 | struct seq_file *seq = file->private_data; |
66d7dd51 KS |
2700 | |
2701 | poll_wait(file, &proc_poll_wait, wait); | |
2702 | ||
f1514638 KS |
2703 | if (seq->poll_event != atomic_read(&proc_poll_event)) { |
2704 | seq->poll_event = atomic_read(&proc_poll_event); | |
a9a08845 | 2705 | return EPOLLIN | EPOLLRDNORM | EPOLLERR | EPOLLPRI; |
66d7dd51 KS |
2706 | } |
2707 | ||
a9a08845 | 2708 | return EPOLLIN | EPOLLRDNORM; |
66d7dd51 KS |
2709 | } |
2710 | ||
1da177e4 LT |
2711 | /* iterator */ |
2712 | static void *swap_start(struct seq_file *swap, loff_t *pos) | |
2713 | { | |
efa90a98 HD |
2714 | struct swap_info_struct *si; |
2715 | int type; | |
1da177e4 LT |
2716 | loff_t l = *pos; |
2717 | ||
fc0abb14 | 2718 | mutex_lock(&swapon_mutex); |
1da177e4 | 2719 | |
881e4aab SS |
2720 | if (!l) |
2721 | return SEQ_START_TOKEN; | |
2722 | ||
c10d38cc | 2723 | for (type = 0; (si = swap_type_to_swap_info(type)); type++) { |
efa90a98 | 2724 | if (!(si->flags & SWP_USED) || !si->swap_map) |
1da177e4 | 2725 | continue; |
881e4aab | 2726 | if (!--l) |
efa90a98 | 2727 | return si; |
1da177e4 LT |
2728 | } |
2729 | ||
2730 | return NULL; | |
2731 | } | |
2732 | ||
2733 | static void *swap_next(struct seq_file *swap, void *v, loff_t *pos) | |
2734 | { | |
efa90a98 HD |
2735 | struct swap_info_struct *si = v; |
2736 | int type; | |
1da177e4 | 2737 | |
881e4aab | 2738 | if (v == SEQ_START_TOKEN) |
efa90a98 HD |
2739 | type = 0; |
2740 | else | |
2741 | type = si->type + 1; | |
881e4aab | 2742 | |
10c8d69f | 2743 | ++(*pos); |
c10d38cc | 2744 | for (; (si = swap_type_to_swap_info(type)); type++) { |
efa90a98 | 2745 | if (!(si->flags & SWP_USED) || !si->swap_map) |
1da177e4 | 2746 | continue; |
efa90a98 | 2747 | return si; |
1da177e4 LT |
2748 | } |
2749 | ||
2750 | return NULL; | |
2751 | } | |
2752 | ||
2753 | static void swap_stop(struct seq_file *swap, void *v) | |
2754 | { | |
fc0abb14 | 2755 | mutex_unlock(&swapon_mutex); |
1da177e4 LT |
2756 | } |
2757 | ||
2758 | static int swap_show(struct seq_file *swap, void *v) | |
2759 | { | |
efa90a98 | 2760 | struct swap_info_struct *si = v; |
1da177e4 LT |
2761 | struct file *file; |
2762 | int len; | |
2763 | ||
efa90a98 | 2764 | if (si == SEQ_START_TOKEN) { |
881e4aab SS |
2765 | seq_puts(swap,"Filename\t\t\t\tType\t\tSize\tUsed\tPriority\n"); |
2766 | return 0; | |
2767 | } | |
1da177e4 | 2768 | |
efa90a98 | 2769 | file = si->swap_file; |
2726d566 | 2770 | len = seq_file_path(swap, file, " \t\n\\"); |
6eb396dc | 2771 | seq_printf(swap, "%*s%s\t%u\t%u\t%d\n", |
886bb7e9 | 2772 | len < 40 ? 40 - len : 1, " ", |
496ad9aa | 2773 | S_ISBLK(file_inode(file)->i_mode) ? |
1da177e4 | 2774 | "partition" : "file\t", |
efa90a98 HD |
2775 | si->pages << (PAGE_SHIFT - 10), |
2776 | si->inuse_pages << (PAGE_SHIFT - 10), | |
2777 | si->prio); | |
1da177e4 LT |
2778 | return 0; |
2779 | } | |
2780 | ||
15ad7cdc | 2781 | static const struct seq_operations swaps_op = { |
1da177e4 LT |
2782 | .start = swap_start, |
2783 | .next = swap_next, | |
2784 | .stop = swap_stop, | |
2785 | .show = swap_show | |
2786 | }; | |
2787 | ||
2788 | static int swaps_open(struct inode *inode, struct file *file) | |
2789 | { | |
f1514638 | 2790 | struct seq_file *seq; |
66d7dd51 KS |
2791 | int ret; |
2792 | ||
66d7dd51 | 2793 | ret = seq_open(file, &swaps_op); |
f1514638 | 2794 | if (ret) |
66d7dd51 | 2795 | return ret; |
66d7dd51 | 2796 | |
f1514638 KS |
2797 | seq = file->private_data; |
2798 | seq->poll_event = atomic_read(&proc_poll_event); | |
2799 | return 0; | |
1da177e4 LT |
2800 | } |
2801 | ||
97a32539 | 2802 | static const struct proc_ops swaps_proc_ops = { |
d919b33d | 2803 | .proc_flags = PROC_ENTRY_PERMANENT, |
97a32539 AD |
2804 | .proc_open = swaps_open, |
2805 | .proc_read = seq_read, | |
2806 | .proc_lseek = seq_lseek, | |
2807 | .proc_release = seq_release, | |
2808 | .proc_poll = swaps_poll, | |
1da177e4 LT |
2809 | }; |
2810 | ||
2811 | static int __init procswaps_init(void) | |
2812 | { | |
97a32539 | 2813 | proc_create("swaps", 0, NULL, &swaps_proc_ops); |
1da177e4 LT |
2814 | return 0; |
2815 | } | |
2816 | __initcall(procswaps_init); | |
2817 | #endif /* CONFIG_PROC_FS */ | |
2818 | ||
1796316a JB |
2819 | #ifdef MAX_SWAPFILES_CHECK |
2820 | static int __init max_swapfiles_check(void) | |
2821 | { | |
2822 | MAX_SWAPFILES_CHECK(); | |
2823 | return 0; | |
2824 | } | |
2825 | late_initcall(max_swapfiles_check); | |
2826 | #endif | |
2827 | ||
53cbb243 | 2828 | static struct swap_info_struct *alloc_swap_info(void) |
1da177e4 | 2829 | { |
73c34b6a | 2830 | struct swap_info_struct *p; |
1da177e4 | 2831 | unsigned int type; |
a2468cc9 | 2832 | int i; |
efa90a98 | 2833 | |
96008744 | 2834 | p = kvzalloc(struct_size(p, avail_lists, nr_node_ids), GFP_KERNEL); |
efa90a98 | 2835 | if (!p) |
53cbb243 | 2836 | return ERR_PTR(-ENOMEM); |
efa90a98 | 2837 | |
5d337b91 | 2838 | spin_lock(&swap_lock); |
efa90a98 HD |
2839 | for (type = 0; type < nr_swapfiles; type++) { |
2840 | if (!(swap_info[type]->flags & SWP_USED)) | |
1da177e4 | 2841 | break; |
efa90a98 | 2842 | } |
0697212a | 2843 | if (type >= MAX_SWAPFILES) { |
5d337b91 | 2844 | spin_unlock(&swap_lock); |
873d7bcf | 2845 | kvfree(p); |
730c0581 | 2846 | return ERR_PTR(-EPERM); |
1da177e4 | 2847 | } |
efa90a98 HD |
2848 | if (type >= nr_swapfiles) { |
2849 | p->type = type; | |
c10d38cc | 2850 | WRITE_ONCE(swap_info[type], p); |
efa90a98 HD |
2851 | /* |
2852 | * Write swap_info[type] before nr_swapfiles, in case a | |
2853 | * racing procfs swap_start() or swap_next() is reading them. | |
2854 | * (We never shrink nr_swapfiles, we never free this entry.) | |
2855 | */ | |
2856 | smp_wmb(); | |
c10d38cc | 2857 | WRITE_ONCE(nr_swapfiles, nr_swapfiles + 1); |
efa90a98 | 2858 | } else { |
873d7bcf | 2859 | kvfree(p); |
efa90a98 HD |
2860 | p = swap_info[type]; |
2861 | /* | |
2862 | * Do not memset this entry: a racing procfs swap_next() | |
2863 | * would be relying on p->type to remain valid. | |
2864 | */ | |
2865 | } | |
4efaceb1 | 2866 | p->swap_extent_root = RB_ROOT; |
18ab4d4c | 2867 | plist_node_init(&p->list, 0); |
a2468cc9 AL |
2868 | for_each_node(i) |
2869 | plist_node_init(&p->avail_lists[i], 0); | |
1da177e4 | 2870 | p->flags = SWP_USED; |
5d337b91 | 2871 | spin_unlock(&swap_lock); |
ec8acf20 | 2872 | spin_lock_init(&p->lock); |
2628bd6f | 2873 | spin_lock_init(&p->cont_lock); |
efa90a98 | 2874 | |
53cbb243 | 2875 | return p; |
53cbb243 CEB |
2876 | } |
2877 | ||
4d0e1e10 CEB |
2878 | static int claim_swapfile(struct swap_info_struct *p, struct inode *inode) |
2879 | { | |
2880 | int error; | |
2881 | ||
2882 | if (S_ISBLK(inode->i_mode)) { | |
2883 | p->bdev = bdgrab(I_BDEV(inode)); | |
2884 | error = blkdev_get(p->bdev, | |
6f179af8 | 2885 | FMODE_READ | FMODE_WRITE | FMODE_EXCL, p); |
4d0e1e10 CEB |
2886 | if (error < 0) { |
2887 | p->bdev = NULL; | |
6f179af8 | 2888 | return error; |
4d0e1e10 CEB |
2889 | } |
2890 | p->old_block_size = block_size(p->bdev); | |
2891 | error = set_blocksize(p->bdev, PAGE_SIZE); | |
2892 | if (error < 0) | |
87ade72a | 2893 | return error; |
12d2966d NA |
2894 | /* |
2895 | * Zoned block devices contain zones that have a sequential | |
2896 | * write only restriction. Hence zoned block devices are not | |
2897 | * suitable for swapping. Disallow them here. | |
2898 | */ | |
2899 | if (blk_queue_is_zoned(p->bdev->bd_queue)) | |
2900 | return -EINVAL; | |
4d0e1e10 CEB |
2901 | p->flags |= SWP_BLKDEV; |
2902 | } else if (S_ISREG(inode->i_mode)) { | |
2903 | p->bdev = inode->i_sb->s_bdev; | |
1638045c DW |
2904 | } |
2905 | ||
4d0e1e10 | 2906 | return 0; |
4d0e1e10 CEB |
2907 | } |
2908 | ||
377eeaa8 AK |
2909 | |
2910 | /* | |
2911 | * Find out how many pages are allowed for a single swap device. There | |
2912 | * are two limiting factors: | |
2913 | * 1) the number of bits for the swap offset in the swp_entry_t type, and | |
2914 | * 2) the number of bits in the swap pte, as defined by the different | |
2915 | * architectures. | |
2916 | * | |
2917 | * In order to find the largest possible bit mask, a swap entry with | |
2918 | * swap type 0 and swap offset ~0UL is created, encoded to a swap pte, | |
2919 | * decoded to a swp_entry_t again, and finally the swap offset is | |
2920 | * extracted. | |
2921 | * | |
2922 | * This will mask all the bits from the initial ~0UL mask that can't | |
2923 | * be encoded in either the swp_entry_t or the architecture definition | |
2924 | * of a swap pte. | |
2925 | */ | |
2926 | unsigned long generic_max_swapfile_size(void) | |
2927 | { | |
2928 | return swp_offset(pte_to_swp_entry( | |
2929 | swp_entry_to_pte(swp_entry(0, ~0UL)))) + 1; | |
2930 | } | |
2931 | ||
2932 | /* Can be overridden by an architecture for additional checks. */ | |
2933 | __weak unsigned long max_swapfile_size(void) | |
2934 | { | |
2935 | return generic_max_swapfile_size(); | |
2936 | } | |
2937 | ||
ca8bd38b CEB |
2938 | static unsigned long read_swap_header(struct swap_info_struct *p, |
2939 | union swap_header *swap_header, | |
2940 | struct inode *inode) | |
2941 | { | |
2942 | int i; | |
2943 | unsigned long maxpages; | |
2944 | unsigned long swapfilepages; | |
d6bbbd29 | 2945 | unsigned long last_page; |
ca8bd38b CEB |
2946 | |
2947 | if (memcmp("SWAPSPACE2", swap_header->magic.magic, 10)) { | |
465c47fd | 2948 | pr_err("Unable to find swap-space signature\n"); |
38719025 | 2949 | return 0; |
ca8bd38b CEB |
2950 | } |
2951 | ||
2952 | /* swap partition endianess hack... */ | |
2953 | if (swab32(swap_header->info.version) == 1) { | |
2954 | swab32s(&swap_header->info.version); | |
2955 | swab32s(&swap_header->info.last_page); | |
2956 | swab32s(&swap_header->info.nr_badpages); | |
dd111be6 JH |
2957 | if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES) |
2958 | return 0; | |
ca8bd38b CEB |
2959 | for (i = 0; i < swap_header->info.nr_badpages; i++) |
2960 | swab32s(&swap_header->info.badpages[i]); | |
2961 | } | |
2962 | /* Check the swap header's sub-version */ | |
2963 | if (swap_header->info.version != 1) { | |
465c47fd AM |
2964 | pr_warn("Unable to handle swap header version %d\n", |
2965 | swap_header->info.version); | |
38719025 | 2966 | return 0; |
ca8bd38b CEB |
2967 | } |
2968 | ||
2969 | p->lowest_bit = 1; | |
2970 | p->cluster_next = 1; | |
2971 | p->cluster_nr = 0; | |
2972 | ||
377eeaa8 | 2973 | maxpages = max_swapfile_size(); |
d6bbbd29 | 2974 | last_page = swap_header->info.last_page; |
a06ad633 TA |
2975 | if (!last_page) { |
2976 | pr_warn("Empty swap-file\n"); | |
2977 | return 0; | |
2978 | } | |
d6bbbd29 | 2979 | if (last_page > maxpages) { |
465c47fd | 2980 | pr_warn("Truncating oversized swap area, only using %luk out of %luk\n", |
d6bbbd29 RJ |
2981 | maxpages << (PAGE_SHIFT - 10), |
2982 | last_page << (PAGE_SHIFT - 10)); | |
2983 | } | |
2984 | if (maxpages > last_page) { | |
2985 | maxpages = last_page + 1; | |
ca8bd38b CEB |
2986 | /* p->max is an unsigned int: don't overflow it */ |
2987 | if ((unsigned int)maxpages == 0) | |
2988 | maxpages = UINT_MAX; | |
2989 | } | |
2990 | p->highest_bit = maxpages - 1; | |
2991 | ||
2992 | if (!maxpages) | |
38719025 | 2993 | return 0; |
ca8bd38b CEB |
2994 | swapfilepages = i_size_read(inode) >> PAGE_SHIFT; |
2995 | if (swapfilepages && maxpages > swapfilepages) { | |
465c47fd | 2996 | pr_warn("Swap area shorter than signature indicates\n"); |
38719025 | 2997 | return 0; |
ca8bd38b CEB |
2998 | } |
2999 | if (swap_header->info.nr_badpages && S_ISREG(inode->i_mode)) | |
38719025 | 3000 | return 0; |
ca8bd38b | 3001 | if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES) |
38719025 | 3002 | return 0; |
ca8bd38b CEB |
3003 | |
3004 | return maxpages; | |
ca8bd38b CEB |
3005 | } |
3006 | ||
4b3ef9da | 3007 | #define SWAP_CLUSTER_INFO_COLS \ |
235b6217 | 3008 | DIV_ROUND_UP(L1_CACHE_BYTES, sizeof(struct swap_cluster_info)) |
4b3ef9da HY |
3009 | #define SWAP_CLUSTER_SPACE_COLS \ |
3010 | DIV_ROUND_UP(SWAP_ADDRESS_SPACE_PAGES, SWAPFILE_CLUSTER) | |
3011 | #define SWAP_CLUSTER_COLS \ | |
3012 | max_t(unsigned int, SWAP_CLUSTER_INFO_COLS, SWAP_CLUSTER_SPACE_COLS) | |
235b6217 | 3013 | |
915d4d7b CEB |
3014 | static int setup_swap_map_and_extents(struct swap_info_struct *p, |
3015 | union swap_header *swap_header, | |
3016 | unsigned char *swap_map, | |
2a8f9449 | 3017 | struct swap_cluster_info *cluster_info, |
915d4d7b CEB |
3018 | unsigned long maxpages, |
3019 | sector_t *span) | |
3020 | { | |
235b6217 | 3021 | unsigned int j, k; |
915d4d7b CEB |
3022 | unsigned int nr_good_pages; |
3023 | int nr_extents; | |
2a8f9449 | 3024 | unsigned long nr_clusters = DIV_ROUND_UP(maxpages, SWAPFILE_CLUSTER); |
235b6217 HY |
3025 | unsigned long col = p->cluster_next / SWAPFILE_CLUSTER % SWAP_CLUSTER_COLS; |
3026 | unsigned long i, idx; | |
915d4d7b CEB |
3027 | |
3028 | nr_good_pages = maxpages - 1; /* omit header page */ | |
3029 | ||
6b534915 HY |
3030 | cluster_list_init(&p->free_clusters); |
3031 | cluster_list_init(&p->discard_clusters); | |
2a8f9449 | 3032 | |
915d4d7b CEB |
3033 | for (i = 0; i < swap_header->info.nr_badpages; i++) { |
3034 | unsigned int page_nr = swap_header->info.badpages[i]; | |
bdb8e3f6 CEB |
3035 | if (page_nr == 0 || page_nr > swap_header->info.last_page) |
3036 | return -EINVAL; | |
915d4d7b CEB |
3037 | if (page_nr < maxpages) { |
3038 | swap_map[page_nr] = SWAP_MAP_BAD; | |
3039 | nr_good_pages--; | |
2a8f9449 SL |
3040 | /* |
3041 | * Haven't marked the cluster free yet, no list | |
3042 | * operation involved | |
3043 | */ | |
3044 | inc_cluster_info_page(p, cluster_info, page_nr); | |
915d4d7b CEB |
3045 | } |
3046 | } | |
3047 | ||
2a8f9449 SL |
3048 | /* Haven't marked the cluster free yet, no list operation involved */ |
3049 | for (i = maxpages; i < round_up(maxpages, SWAPFILE_CLUSTER); i++) | |
3050 | inc_cluster_info_page(p, cluster_info, i); | |
3051 | ||
915d4d7b CEB |
3052 | if (nr_good_pages) { |
3053 | swap_map[0] = SWAP_MAP_BAD; | |
2a8f9449 SL |
3054 | /* |
3055 | * Not mark the cluster free yet, no list | |
3056 | * operation involved | |
3057 | */ | |
3058 | inc_cluster_info_page(p, cluster_info, 0); | |
915d4d7b CEB |
3059 | p->max = maxpages; |
3060 | p->pages = nr_good_pages; | |
3061 | nr_extents = setup_swap_extents(p, span); | |
bdb8e3f6 CEB |
3062 | if (nr_extents < 0) |
3063 | return nr_extents; | |
915d4d7b CEB |
3064 | nr_good_pages = p->pages; |
3065 | } | |
3066 | if (!nr_good_pages) { | |
465c47fd | 3067 | pr_warn("Empty swap-file\n"); |
bdb8e3f6 | 3068 | return -EINVAL; |
915d4d7b CEB |
3069 | } |
3070 | ||
2a8f9449 SL |
3071 | if (!cluster_info) |
3072 | return nr_extents; | |
3073 | ||
235b6217 | 3074 | |
4b3ef9da HY |
3075 | /* |
3076 | * Reduce false cache line sharing between cluster_info and | |
3077 | * sharing same address space. | |
3078 | */ | |
235b6217 HY |
3079 | for (k = 0; k < SWAP_CLUSTER_COLS; k++) { |
3080 | j = (k + col) % SWAP_CLUSTER_COLS; | |
3081 | for (i = 0; i < DIV_ROUND_UP(nr_clusters, SWAP_CLUSTER_COLS); i++) { | |
3082 | idx = i * SWAP_CLUSTER_COLS + j; | |
3083 | if (idx >= nr_clusters) | |
3084 | continue; | |
3085 | if (cluster_count(&cluster_info[idx])) | |
3086 | continue; | |
2a8f9449 | 3087 | cluster_set_flag(&cluster_info[idx], CLUSTER_FLAG_FREE); |
6b534915 HY |
3088 | cluster_list_add_tail(&p->free_clusters, cluster_info, |
3089 | idx); | |
2a8f9449 | 3090 | } |
2a8f9449 | 3091 | } |
915d4d7b | 3092 | return nr_extents; |
915d4d7b CEB |
3093 | } |
3094 | ||
dcf6b7dd RA |
3095 | /* |
3096 | * Helper to sys_swapon determining if a given swap | |
3097 | * backing device queue supports DISCARD operations. | |
3098 | */ | |
3099 | static bool swap_discardable(struct swap_info_struct *si) | |
3100 | { | |
3101 | struct request_queue *q = bdev_get_queue(si->bdev); | |
3102 | ||
3103 | if (!q || !blk_queue_discard(q)) | |
3104 | return false; | |
3105 | ||
3106 | return true; | |
3107 | } | |
3108 | ||
53cbb243 CEB |
3109 | SYSCALL_DEFINE2(swapon, const char __user *, specialfile, int, swap_flags) |
3110 | { | |
3111 | struct swap_info_struct *p; | |
91a27b2a | 3112 | struct filename *name; |
53cbb243 CEB |
3113 | struct file *swap_file = NULL; |
3114 | struct address_space *mapping; | |
40531542 | 3115 | int prio; |
53cbb243 CEB |
3116 | int error; |
3117 | union swap_header *swap_header; | |
915d4d7b | 3118 | int nr_extents; |
53cbb243 CEB |
3119 | sector_t span; |
3120 | unsigned long maxpages; | |
53cbb243 | 3121 | unsigned char *swap_map = NULL; |
2a8f9449 | 3122 | struct swap_cluster_info *cluster_info = NULL; |
38b5faf4 | 3123 | unsigned long *frontswap_map = NULL; |
53cbb243 CEB |
3124 | struct page *page = NULL; |
3125 | struct inode *inode = NULL; | |
7cbf3192 | 3126 | bool inced_nr_rotate_swap = false; |
53cbb243 | 3127 | |
d15cab97 HD |
3128 | if (swap_flags & ~SWAP_FLAGS_VALID) |
3129 | return -EINVAL; | |
3130 | ||
53cbb243 CEB |
3131 | if (!capable(CAP_SYS_ADMIN)) |
3132 | return -EPERM; | |
3133 | ||
a2468cc9 AL |
3134 | if (!swap_avail_heads) |
3135 | return -ENOMEM; | |
3136 | ||
53cbb243 | 3137 | p = alloc_swap_info(); |
2542e513 CEB |
3138 | if (IS_ERR(p)) |
3139 | return PTR_ERR(p); | |
53cbb243 | 3140 | |
815c2c54 SL |
3141 | INIT_WORK(&p->discard_work, swap_discard_work); |
3142 | ||
1da177e4 | 3143 | name = getname(specialfile); |
1da177e4 | 3144 | if (IS_ERR(name)) { |
7de7fb6b | 3145 | error = PTR_ERR(name); |
1da177e4 | 3146 | name = NULL; |
bd69010b | 3147 | goto bad_swap; |
1da177e4 | 3148 | } |
669abf4e | 3149 | swap_file = file_open_name(name, O_RDWR|O_LARGEFILE, 0); |
1da177e4 | 3150 | if (IS_ERR(swap_file)) { |
7de7fb6b | 3151 | error = PTR_ERR(swap_file); |
1da177e4 | 3152 | swap_file = NULL; |
bd69010b | 3153 | goto bad_swap; |
1da177e4 LT |
3154 | } |
3155 | ||
3156 | p->swap_file = swap_file; | |
3157 | mapping = swap_file->f_mapping; | |
2130781e | 3158 | inode = mapping->host; |
6f179af8 | 3159 | |
4d0e1e10 CEB |
3160 | error = claim_swapfile(p, inode); |
3161 | if (unlikely(error)) | |
1da177e4 | 3162 | goto bad_swap; |
1da177e4 | 3163 | |
d795a90e NA |
3164 | inode_lock(inode); |
3165 | if (IS_SWAPFILE(inode)) { | |
3166 | error = -EBUSY; | |
3167 | goto bad_swap_unlock_inode; | |
3168 | } | |
3169 | ||
1da177e4 LT |
3170 | /* |
3171 | * Read the swap header. | |
3172 | */ | |
3173 | if (!mapping->a_ops->readpage) { | |
3174 | error = -EINVAL; | |
d795a90e | 3175 | goto bad_swap_unlock_inode; |
1da177e4 | 3176 | } |
090d2b18 | 3177 | page = read_mapping_page(mapping, 0, swap_file); |
1da177e4 LT |
3178 | if (IS_ERR(page)) { |
3179 | error = PTR_ERR(page); | |
d795a90e | 3180 | goto bad_swap_unlock_inode; |
1da177e4 | 3181 | } |
81e33971 | 3182 | swap_header = kmap(page); |
1da177e4 | 3183 | |
ca8bd38b CEB |
3184 | maxpages = read_swap_header(p, swap_header, inode); |
3185 | if (unlikely(!maxpages)) { | |
1da177e4 | 3186 | error = -EINVAL; |
d795a90e | 3187 | goto bad_swap_unlock_inode; |
1da177e4 | 3188 | } |
886bb7e9 | 3189 | |
81e33971 | 3190 | /* OK, set up the swap map and apply the bad block list */ |
803d0c83 | 3191 | swap_map = vzalloc(maxpages); |
81e33971 HD |
3192 | if (!swap_map) { |
3193 | error = -ENOMEM; | |
d795a90e | 3194 | goto bad_swap_unlock_inode; |
81e33971 | 3195 | } |
f0571429 MK |
3196 | |
3197 | if (bdi_cap_stable_pages_required(inode_to_bdi(inode))) | |
3198 | p->flags |= SWP_STABLE_WRITES; | |
3199 | ||
539a6fea MK |
3200 | if (bdi_cap_synchronous_io(inode_to_bdi(inode))) |
3201 | p->flags |= SWP_SYNCHRONOUS_IO; | |
3202 | ||
2a8f9449 | 3203 | if (p->bdev && blk_queue_nonrot(bdev_get_queue(p->bdev))) { |
6f179af8 | 3204 | int cpu; |
235b6217 | 3205 | unsigned long ci, nr_cluster; |
6f179af8 | 3206 | |
2a8f9449 SL |
3207 | p->flags |= SWP_SOLIDSTATE; |
3208 | /* | |
3209 | * select a random position to start with to help wear leveling | |
3210 | * SSD | |
3211 | */ | |
09fe06ce | 3212 | p->cluster_next = 1 + prandom_u32_max(p->highest_bit); |
235b6217 | 3213 | nr_cluster = DIV_ROUND_UP(maxpages, SWAPFILE_CLUSTER); |
2a8f9449 | 3214 | |
778e1cdd | 3215 | cluster_info = kvcalloc(nr_cluster, sizeof(*cluster_info), |
54f180d3 | 3216 | GFP_KERNEL); |
2a8f9449 SL |
3217 | if (!cluster_info) { |
3218 | error = -ENOMEM; | |
d795a90e | 3219 | goto bad_swap_unlock_inode; |
2a8f9449 | 3220 | } |
235b6217 HY |
3221 | |
3222 | for (ci = 0; ci < nr_cluster; ci++) | |
3223 | spin_lock_init(&((cluster_info + ci)->lock)); | |
3224 | ||
ebc2a1a6 SL |
3225 | p->percpu_cluster = alloc_percpu(struct percpu_cluster); |
3226 | if (!p->percpu_cluster) { | |
3227 | error = -ENOMEM; | |
d795a90e | 3228 | goto bad_swap_unlock_inode; |
ebc2a1a6 | 3229 | } |
6f179af8 | 3230 | for_each_possible_cpu(cpu) { |
ebc2a1a6 | 3231 | struct percpu_cluster *cluster; |
6f179af8 | 3232 | cluster = per_cpu_ptr(p->percpu_cluster, cpu); |
ebc2a1a6 SL |
3233 | cluster_set_null(&cluster->index); |
3234 | } | |
7cbf3192 | 3235 | } else { |
81a0298b | 3236 | atomic_inc(&nr_rotate_swap); |
7cbf3192 OS |
3237 | inced_nr_rotate_swap = true; |
3238 | } | |
1da177e4 | 3239 | |
1421ef3c CEB |
3240 | error = swap_cgroup_swapon(p->type, maxpages); |
3241 | if (error) | |
d795a90e | 3242 | goto bad_swap_unlock_inode; |
1421ef3c | 3243 | |
915d4d7b | 3244 | nr_extents = setup_swap_map_and_extents(p, swap_header, swap_map, |
2a8f9449 | 3245 | cluster_info, maxpages, &span); |
915d4d7b CEB |
3246 | if (unlikely(nr_extents < 0)) { |
3247 | error = nr_extents; | |
d795a90e | 3248 | goto bad_swap_unlock_inode; |
1da177e4 | 3249 | } |
38b5faf4 | 3250 | /* frontswap enabled? set up bit-per-page map for frontswap */ |
8ea1d2a1 | 3251 | if (IS_ENABLED(CONFIG_FRONTSWAP)) |
778e1cdd KC |
3252 | frontswap_map = kvcalloc(BITS_TO_LONGS(maxpages), |
3253 | sizeof(long), | |
54f180d3 | 3254 | GFP_KERNEL); |
1da177e4 | 3255 | |
2a8f9449 SL |
3256 | if (p->bdev &&(swap_flags & SWAP_FLAG_DISCARD) && swap_discardable(p)) { |
3257 | /* | |
3258 | * When discard is enabled for swap with no particular | |
3259 | * policy flagged, we set all swap discard flags here in | |
3260 | * order to sustain backward compatibility with older | |
3261 | * swapon(8) releases. | |
3262 | */ | |
3263 | p->flags |= (SWP_DISCARDABLE | SWP_AREA_DISCARD | | |
3264 | SWP_PAGE_DISCARD); | |
dcf6b7dd | 3265 | |
2a8f9449 SL |
3266 | /* |
3267 | * By flagging sys_swapon, a sysadmin can tell us to | |
3268 | * either do single-time area discards only, or to just | |
3269 | * perform discards for released swap page-clusters. | |
3270 | * Now it's time to adjust the p->flags accordingly. | |
3271 | */ | |
3272 | if (swap_flags & SWAP_FLAG_DISCARD_ONCE) | |
3273 | p->flags &= ~SWP_PAGE_DISCARD; | |
3274 | else if (swap_flags & SWAP_FLAG_DISCARD_PAGES) | |
3275 | p->flags &= ~SWP_AREA_DISCARD; | |
3276 | ||
3277 | /* issue a swapon-time discard if it's still required */ | |
3278 | if (p->flags & SWP_AREA_DISCARD) { | |
3279 | int err = discard_swap(p); | |
3280 | if (unlikely(err)) | |
3281 | pr_err("swapon: discard_swap(%p): %d\n", | |
3282 | p, err); | |
dcf6b7dd | 3283 | } |
20137a49 | 3284 | } |
6a6ba831 | 3285 | |
4b3ef9da HY |
3286 | error = init_swap_address_space(p->type, maxpages); |
3287 | if (error) | |
d795a90e | 3288 | goto bad_swap_unlock_inode; |
4b3ef9da | 3289 | |
dc617f29 DW |
3290 | /* |
3291 | * Flush any pending IO and dirty mappings before we start using this | |
3292 | * swap device. | |
3293 | */ | |
3294 | inode->i_flags |= S_SWAPFILE; | |
3295 | error = inode_drain_writes(inode); | |
3296 | if (error) { | |
3297 | inode->i_flags &= ~S_SWAPFILE; | |
d795a90e | 3298 | goto bad_swap_unlock_inode; |
dc617f29 DW |
3299 | } |
3300 | ||
fc0abb14 | 3301 | mutex_lock(&swapon_mutex); |
40531542 | 3302 | prio = -1; |
78ecba08 | 3303 | if (swap_flags & SWAP_FLAG_PREFER) |
40531542 | 3304 | prio = |
78ecba08 | 3305 | (swap_flags & SWAP_FLAG_PRIO_MASK) >> SWAP_FLAG_PRIO_SHIFT; |
2a8f9449 | 3306 | enable_swap_info(p, prio, swap_map, cluster_info, frontswap_map); |
c69dbfb8 | 3307 | |
756a025f | 3308 | pr_info("Adding %uk swap on %s. Priority:%d extents:%d across:%lluk %s%s%s%s%s\n", |
91a27b2a | 3309 | p->pages<<(PAGE_SHIFT-10), name->name, p->prio, |
c69dbfb8 CEB |
3310 | nr_extents, (unsigned long long)span<<(PAGE_SHIFT-10), |
3311 | (p->flags & SWP_SOLIDSTATE) ? "SS" : "", | |
38b5faf4 | 3312 | (p->flags & SWP_DISCARDABLE) ? "D" : "", |
dcf6b7dd RA |
3313 | (p->flags & SWP_AREA_DISCARD) ? "s" : "", |
3314 | (p->flags & SWP_PAGE_DISCARD) ? "c" : "", | |
38b5faf4 | 3315 | (frontswap_map) ? "FS" : ""); |
c69dbfb8 | 3316 | |
fc0abb14 | 3317 | mutex_unlock(&swapon_mutex); |
66d7dd51 KS |
3318 | atomic_inc(&proc_poll_event); |
3319 | wake_up_interruptible(&proc_poll_wait); | |
3320 | ||
1da177e4 LT |
3321 | error = 0; |
3322 | goto out; | |
d795a90e NA |
3323 | bad_swap_unlock_inode: |
3324 | inode_unlock(inode); | |
1da177e4 | 3325 | bad_swap: |
ebc2a1a6 SL |
3326 | free_percpu(p->percpu_cluster); |
3327 | p->percpu_cluster = NULL; | |
bd69010b | 3328 | if (inode && S_ISBLK(inode->i_mode) && p->bdev) { |
f2090d2d CEB |
3329 | set_blocksize(p->bdev, p->old_block_size); |
3330 | blkdev_put(p->bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL); | |
1da177e4 | 3331 | } |
d795a90e | 3332 | inode = NULL; |
4cd3bb10 | 3333 | destroy_swap_extents(p); |
e8e6c2ec | 3334 | swap_cgroup_swapoff(p->type); |
5d337b91 | 3335 | spin_lock(&swap_lock); |
1da177e4 | 3336 | p->swap_file = NULL; |
1da177e4 | 3337 | p->flags = 0; |
5d337b91 | 3338 | spin_unlock(&swap_lock); |
1da177e4 | 3339 | vfree(swap_map); |
8606a1a9 | 3340 | kvfree(cluster_info); |
b6b1fd2a | 3341 | kvfree(frontswap_map); |
7cbf3192 OS |
3342 | if (inced_nr_rotate_swap) |
3343 | atomic_dec(&nr_rotate_swap); | |
d795a90e | 3344 | if (swap_file) |
1da177e4 LT |
3345 | filp_close(swap_file, NULL); |
3346 | out: | |
3347 | if (page && !IS_ERR(page)) { | |
3348 | kunmap(page); | |
09cbfeaf | 3349 | put_page(page); |
1da177e4 LT |
3350 | } |
3351 | if (name) | |
3352 | putname(name); | |
1638045c | 3353 | if (inode) |
5955102c | 3354 | inode_unlock(inode); |
039939a6 TC |
3355 | if (!error) |
3356 | enable_swap_slots_cache(); | |
1da177e4 LT |
3357 | return error; |
3358 | } | |
3359 | ||
3360 | void si_swapinfo(struct sysinfo *val) | |
3361 | { | |
efa90a98 | 3362 | unsigned int type; |
1da177e4 LT |
3363 | unsigned long nr_to_be_unused = 0; |
3364 | ||
5d337b91 | 3365 | spin_lock(&swap_lock); |
efa90a98 HD |
3366 | for (type = 0; type < nr_swapfiles; type++) { |
3367 | struct swap_info_struct *si = swap_info[type]; | |
3368 | ||
3369 | if ((si->flags & SWP_USED) && !(si->flags & SWP_WRITEOK)) | |
3370 | nr_to_be_unused += si->inuse_pages; | |
1da177e4 | 3371 | } |
ec8acf20 | 3372 | val->freeswap = atomic_long_read(&nr_swap_pages) + nr_to_be_unused; |
1da177e4 | 3373 | val->totalswap = total_swap_pages + nr_to_be_unused; |
5d337b91 | 3374 | spin_unlock(&swap_lock); |
1da177e4 LT |
3375 | } |
3376 | ||
3377 | /* | |
3378 | * Verify that a swap entry is valid and increment its swap map count. | |
3379 | * | |
355cfa73 KH |
3380 | * Returns error code in following case. |
3381 | * - success -> 0 | |
3382 | * - swp_entry is invalid -> EINVAL | |
3383 | * - swp_entry is migration entry -> EINVAL | |
3384 | * - swap-cache reference is requested but there is already one. -> EEXIST | |
3385 | * - swap-cache reference is requested but the entry is not used. -> ENOENT | |
570a335b | 3386 | * - swap-mapped reference requested but needs continued swap count. -> ENOMEM |
1da177e4 | 3387 | */ |
8d69aaee | 3388 | static int __swap_duplicate(swp_entry_t entry, unsigned char usage) |
1da177e4 | 3389 | { |
73c34b6a | 3390 | struct swap_info_struct *p; |
235b6217 | 3391 | struct swap_cluster_info *ci; |
c10d38cc | 3392 | unsigned long offset; |
8d69aaee HD |
3393 | unsigned char count; |
3394 | unsigned char has_cache; | |
253d553b | 3395 | int err = -EINVAL; |
1da177e4 | 3396 | |
eb085574 | 3397 | p = get_swap_device(entry); |
c10d38cc | 3398 | if (!p) |
235b6217 HY |
3399 | goto out; |
3400 | ||
eb085574 | 3401 | offset = swp_offset(entry); |
235b6217 | 3402 | ci = lock_cluster_or_swap_info(p, offset); |
355cfa73 | 3403 | |
253d553b | 3404 | count = p->swap_map[offset]; |
edfe23da SL |
3405 | |
3406 | /* | |
3407 | * swapin_readahead() doesn't check if a swap entry is valid, so the | |
3408 | * swap entry could be SWAP_MAP_BAD. Check here with lock held. | |
3409 | */ | |
3410 | if (unlikely(swap_count(count) == SWAP_MAP_BAD)) { | |
3411 | err = -ENOENT; | |
3412 | goto unlock_out; | |
3413 | } | |
3414 | ||
253d553b HD |
3415 | has_cache = count & SWAP_HAS_CACHE; |
3416 | count &= ~SWAP_HAS_CACHE; | |
3417 | err = 0; | |
355cfa73 | 3418 | |
253d553b | 3419 | if (usage == SWAP_HAS_CACHE) { |
355cfa73 KH |
3420 | |
3421 | /* set SWAP_HAS_CACHE if there is no cache and entry is used */ | |
253d553b HD |
3422 | if (!has_cache && count) |
3423 | has_cache = SWAP_HAS_CACHE; | |
3424 | else if (has_cache) /* someone else added cache */ | |
3425 | err = -EEXIST; | |
3426 | else /* no users remaining */ | |
3427 | err = -ENOENT; | |
355cfa73 KH |
3428 | |
3429 | } else if (count || has_cache) { | |
253d553b | 3430 | |
570a335b HD |
3431 | if ((count & ~COUNT_CONTINUED) < SWAP_MAP_MAX) |
3432 | count += usage; | |
3433 | else if ((count & ~COUNT_CONTINUED) > SWAP_MAP_MAX) | |
253d553b | 3434 | err = -EINVAL; |
570a335b HD |
3435 | else if (swap_count_continued(p, offset, count)) |
3436 | count = COUNT_CONTINUED; | |
3437 | else | |
3438 | err = -ENOMEM; | |
355cfa73 | 3439 | } else |
253d553b HD |
3440 | err = -ENOENT; /* unused swap entry */ |
3441 | ||
3442 | p->swap_map[offset] = count | has_cache; | |
3443 | ||
355cfa73 | 3444 | unlock_out: |
235b6217 | 3445 | unlock_cluster_or_swap_info(p, ci); |
1da177e4 | 3446 | out: |
eb085574 HY |
3447 | if (p) |
3448 | put_swap_device(p); | |
253d553b | 3449 | return err; |
1da177e4 | 3450 | } |
253d553b | 3451 | |
aaa46865 HD |
3452 | /* |
3453 | * Help swapoff by noting that swap entry belongs to shmem/tmpfs | |
3454 | * (in which case its reference count is never incremented). | |
3455 | */ | |
3456 | void swap_shmem_alloc(swp_entry_t entry) | |
3457 | { | |
3458 | __swap_duplicate(entry, SWAP_MAP_SHMEM); | |
3459 | } | |
3460 | ||
355cfa73 | 3461 | /* |
08259d58 HD |
3462 | * Increase reference count of swap entry by 1. |
3463 | * Returns 0 for success, or -ENOMEM if a swap_count_continuation is required | |
3464 | * but could not be atomically allocated. Returns 0, just as if it succeeded, | |
3465 | * if __swap_duplicate() fails for another reason (-EINVAL or -ENOENT), which | |
3466 | * might occur if a page table entry has got corrupted. | |
355cfa73 | 3467 | */ |
570a335b | 3468 | int swap_duplicate(swp_entry_t entry) |
355cfa73 | 3469 | { |
570a335b HD |
3470 | int err = 0; |
3471 | ||
3472 | while (!err && __swap_duplicate(entry, 1) == -ENOMEM) | |
3473 | err = add_swap_count_continuation(entry, GFP_ATOMIC); | |
3474 | return err; | |
355cfa73 | 3475 | } |
1da177e4 | 3476 | |
cb4b86ba | 3477 | /* |
355cfa73 KH |
3478 | * @entry: swap entry for which we allocate swap cache. |
3479 | * | |
73c34b6a | 3480 | * Called when allocating swap cache for existing swap entry, |
355cfa73 | 3481 | * This can return error codes. Returns 0 at success. |
3eeba135 | 3482 | * -EEXIST means there is a swap cache. |
355cfa73 | 3483 | * Note: return code is different from swap_duplicate(). |
cb4b86ba KH |
3484 | */ |
3485 | int swapcache_prepare(swp_entry_t entry) | |
3486 | { | |
253d553b | 3487 | return __swap_duplicate(entry, SWAP_HAS_CACHE); |
cb4b86ba KH |
3488 | } |
3489 | ||
0bcac06f MK |
3490 | struct swap_info_struct *swp_swap_info(swp_entry_t entry) |
3491 | { | |
c10d38cc | 3492 | return swap_type_to_swap_info(swp_type(entry)); |
0bcac06f MK |
3493 | } |
3494 | ||
f981c595 MG |
3495 | struct swap_info_struct *page_swap_info(struct page *page) |
3496 | { | |
0bcac06f MK |
3497 | swp_entry_t entry = { .val = page_private(page) }; |
3498 | return swp_swap_info(entry); | |
f981c595 MG |
3499 | } |
3500 | ||
3501 | /* | |
3502 | * out-of-line __page_file_ methods to avoid include hell. | |
3503 | */ | |
3504 | struct address_space *__page_file_mapping(struct page *page) | |
3505 | { | |
f981c595 MG |
3506 | return page_swap_info(page)->swap_file->f_mapping; |
3507 | } | |
3508 | EXPORT_SYMBOL_GPL(__page_file_mapping); | |
3509 | ||
3510 | pgoff_t __page_file_index(struct page *page) | |
3511 | { | |
3512 | swp_entry_t swap = { .val = page_private(page) }; | |
f981c595 MG |
3513 | return swp_offset(swap); |
3514 | } | |
3515 | EXPORT_SYMBOL_GPL(__page_file_index); | |
3516 | ||
570a335b HD |
3517 | /* |
3518 | * add_swap_count_continuation - called when a swap count is duplicated | |
3519 | * beyond SWAP_MAP_MAX, it allocates a new page and links that to the entry's | |
3520 | * page of the original vmalloc'ed swap_map, to hold the continuation count | |
3521 | * (for that entry and for its neighbouring PAGE_SIZE swap entries). Called | |
3522 | * again when count is duplicated beyond SWAP_MAP_MAX * SWAP_CONT_MAX, etc. | |
3523 | * | |
3524 | * These continuation pages are seldom referenced: the common paths all work | |
3525 | * on the original swap_map, only referring to a continuation page when the | |
3526 | * low "digit" of a count is incremented or decremented through SWAP_MAP_MAX. | |
3527 | * | |
3528 | * add_swap_count_continuation(, GFP_ATOMIC) can be called while holding | |
3529 | * page table locks; if it fails, add_swap_count_continuation(, GFP_KERNEL) | |
3530 | * can be called after dropping locks. | |
3531 | */ | |
3532 | int add_swap_count_continuation(swp_entry_t entry, gfp_t gfp_mask) | |
3533 | { | |
3534 | struct swap_info_struct *si; | |
235b6217 | 3535 | struct swap_cluster_info *ci; |
570a335b HD |
3536 | struct page *head; |
3537 | struct page *page; | |
3538 | struct page *list_page; | |
3539 | pgoff_t offset; | |
3540 | unsigned char count; | |
eb085574 | 3541 | int ret = 0; |
570a335b HD |
3542 | |
3543 | /* | |
3544 | * When debugging, it's easier to use __GFP_ZERO here; but it's better | |
3545 | * for latency not to zero a page while GFP_ATOMIC and holding locks. | |
3546 | */ | |
3547 | page = alloc_page(gfp_mask | __GFP_HIGHMEM); | |
3548 | ||
eb085574 | 3549 | si = get_swap_device(entry); |
570a335b HD |
3550 | if (!si) { |
3551 | /* | |
3552 | * An acceptable race has occurred since the failing | |
eb085574 | 3553 | * __swap_duplicate(): the swap device may be swapoff |
570a335b HD |
3554 | */ |
3555 | goto outer; | |
3556 | } | |
eb085574 | 3557 | spin_lock(&si->lock); |
570a335b HD |
3558 | |
3559 | offset = swp_offset(entry); | |
235b6217 HY |
3560 | |
3561 | ci = lock_cluster(si, offset); | |
3562 | ||
570a335b HD |
3563 | count = si->swap_map[offset] & ~SWAP_HAS_CACHE; |
3564 | ||
3565 | if ((count & ~COUNT_CONTINUED) != SWAP_MAP_MAX) { | |
3566 | /* | |
3567 | * The higher the swap count, the more likely it is that tasks | |
3568 | * will race to add swap count continuation: we need to avoid | |
3569 | * over-provisioning. | |
3570 | */ | |
3571 | goto out; | |
3572 | } | |
3573 | ||
3574 | if (!page) { | |
eb085574 HY |
3575 | ret = -ENOMEM; |
3576 | goto out; | |
570a335b HD |
3577 | } |
3578 | ||
3579 | /* | |
3580 | * We are fortunate that although vmalloc_to_page uses pte_offset_map, | |
2de1a7e4 SJ |
3581 | * no architecture is using highmem pages for kernel page tables: so it |
3582 | * will not corrupt the GFP_ATOMIC caller's atomic page table kmaps. | |
570a335b HD |
3583 | */ |
3584 | head = vmalloc_to_page(si->swap_map + offset); | |
3585 | offset &= ~PAGE_MASK; | |
3586 | ||
2628bd6f | 3587 | spin_lock(&si->cont_lock); |
570a335b HD |
3588 | /* |
3589 | * Page allocation does not initialize the page's lru field, | |
3590 | * but it does always reset its private field. | |
3591 | */ | |
3592 | if (!page_private(head)) { | |
3593 | BUG_ON(count & COUNT_CONTINUED); | |
3594 | INIT_LIST_HEAD(&head->lru); | |
3595 | set_page_private(head, SWP_CONTINUED); | |
3596 | si->flags |= SWP_CONTINUED; | |
3597 | } | |
3598 | ||
3599 | list_for_each_entry(list_page, &head->lru, lru) { | |
3600 | unsigned char *map; | |
3601 | ||
3602 | /* | |
3603 | * If the previous map said no continuation, but we've found | |
3604 | * a continuation page, free our allocation and use this one. | |
3605 | */ | |
3606 | if (!(count & COUNT_CONTINUED)) | |
2628bd6f | 3607 | goto out_unlock_cont; |
570a335b | 3608 | |
9b04c5fe | 3609 | map = kmap_atomic(list_page) + offset; |
570a335b | 3610 | count = *map; |
9b04c5fe | 3611 | kunmap_atomic(map); |
570a335b HD |
3612 | |
3613 | /* | |
3614 | * If this continuation count now has some space in it, | |
3615 | * free our allocation and use this one. | |
3616 | */ | |
3617 | if ((count & ~COUNT_CONTINUED) != SWAP_CONT_MAX) | |
2628bd6f | 3618 | goto out_unlock_cont; |
570a335b HD |
3619 | } |
3620 | ||
3621 | list_add_tail(&page->lru, &head->lru); | |
3622 | page = NULL; /* now it's attached, don't free it */ | |
2628bd6f HY |
3623 | out_unlock_cont: |
3624 | spin_unlock(&si->cont_lock); | |
570a335b | 3625 | out: |
235b6217 | 3626 | unlock_cluster(ci); |
ec8acf20 | 3627 | spin_unlock(&si->lock); |
eb085574 | 3628 | put_swap_device(si); |
570a335b HD |
3629 | outer: |
3630 | if (page) | |
3631 | __free_page(page); | |
eb085574 | 3632 | return ret; |
570a335b HD |
3633 | } |
3634 | ||
3635 | /* | |
3636 | * swap_count_continued - when the original swap_map count is incremented | |
3637 | * from SWAP_MAP_MAX, check if there is already a continuation page to carry | |
3638 | * into, carry if so, or else fail until a new continuation page is allocated; | |
3639 | * when the original swap_map count is decremented from 0 with continuation, | |
3640 | * borrow from the continuation and report whether it still holds more. | |
235b6217 HY |
3641 | * Called while __swap_duplicate() or swap_entry_free() holds swap or cluster |
3642 | * lock. | |
570a335b HD |
3643 | */ |
3644 | static bool swap_count_continued(struct swap_info_struct *si, | |
3645 | pgoff_t offset, unsigned char count) | |
3646 | { | |
3647 | struct page *head; | |
3648 | struct page *page; | |
3649 | unsigned char *map; | |
2628bd6f | 3650 | bool ret; |
570a335b HD |
3651 | |
3652 | head = vmalloc_to_page(si->swap_map + offset); | |
3653 | if (page_private(head) != SWP_CONTINUED) { | |
3654 | BUG_ON(count & COUNT_CONTINUED); | |
3655 | return false; /* need to add count continuation */ | |
3656 | } | |
3657 | ||
2628bd6f | 3658 | spin_lock(&si->cont_lock); |
570a335b | 3659 | offset &= ~PAGE_MASK; |
213516ac | 3660 | page = list_next_entry(head, lru); |
9b04c5fe | 3661 | map = kmap_atomic(page) + offset; |
570a335b HD |
3662 | |
3663 | if (count == SWAP_MAP_MAX) /* initial increment from swap_map */ | |
3664 | goto init_map; /* jump over SWAP_CONT_MAX checks */ | |
3665 | ||
3666 | if (count == (SWAP_MAP_MAX | COUNT_CONTINUED)) { /* incrementing */ | |
3667 | /* | |
3668 | * Think of how you add 1 to 999 | |
3669 | */ | |
3670 | while (*map == (SWAP_CONT_MAX | COUNT_CONTINUED)) { | |
9b04c5fe | 3671 | kunmap_atomic(map); |
213516ac | 3672 | page = list_next_entry(page, lru); |
570a335b | 3673 | BUG_ON(page == head); |
9b04c5fe | 3674 | map = kmap_atomic(page) + offset; |
570a335b HD |
3675 | } |
3676 | if (*map == SWAP_CONT_MAX) { | |
9b04c5fe | 3677 | kunmap_atomic(map); |
213516ac | 3678 | page = list_next_entry(page, lru); |
2628bd6f HY |
3679 | if (page == head) { |
3680 | ret = false; /* add count continuation */ | |
3681 | goto out; | |
3682 | } | |
9b04c5fe | 3683 | map = kmap_atomic(page) + offset; |
570a335b HD |
3684 | init_map: *map = 0; /* we didn't zero the page */ |
3685 | } | |
3686 | *map += 1; | |
9b04c5fe | 3687 | kunmap_atomic(map); |
213516ac | 3688 | while ((page = list_prev_entry(page, lru)) != head) { |
9b04c5fe | 3689 | map = kmap_atomic(page) + offset; |
570a335b | 3690 | *map = COUNT_CONTINUED; |
9b04c5fe | 3691 | kunmap_atomic(map); |
570a335b | 3692 | } |
2628bd6f | 3693 | ret = true; /* incremented */ |
570a335b HD |
3694 | |
3695 | } else { /* decrementing */ | |
3696 | /* | |
3697 | * Think of how you subtract 1 from 1000 | |
3698 | */ | |
3699 | BUG_ON(count != COUNT_CONTINUED); | |
3700 | while (*map == COUNT_CONTINUED) { | |
9b04c5fe | 3701 | kunmap_atomic(map); |
213516ac | 3702 | page = list_next_entry(page, lru); |
570a335b | 3703 | BUG_ON(page == head); |
9b04c5fe | 3704 | map = kmap_atomic(page) + offset; |
570a335b HD |
3705 | } |
3706 | BUG_ON(*map == 0); | |
3707 | *map -= 1; | |
3708 | if (*map == 0) | |
3709 | count = 0; | |
9b04c5fe | 3710 | kunmap_atomic(map); |
213516ac | 3711 | while ((page = list_prev_entry(page, lru)) != head) { |
9b04c5fe | 3712 | map = kmap_atomic(page) + offset; |
570a335b HD |
3713 | *map = SWAP_CONT_MAX | count; |
3714 | count = COUNT_CONTINUED; | |
9b04c5fe | 3715 | kunmap_atomic(map); |
570a335b | 3716 | } |
2628bd6f | 3717 | ret = count == COUNT_CONTINUED; |
570a335b | 3718 | } |
2628bd6f HY |
3719 | out: |
3720 | spin_unlock(&si->cont_lock); | |
3721 | return ret; | |
570a335b HD |
3722 | } |
3723 | ||
3724 | /* | |
3725 | * free_swap_count_continuations - swapoff free all the continuation pages | |
3726 | * appended to the swap_map, after swap_map is quiesced, before vfree'ing it. | |
3727 | */ | |
3728 | static void free_swap_count_continuations(struct swap_info_struct *si) | |
3729 | { | |
3730 | pgoff_t offset; | |
3731 | ||
3732 | for (offset = 0; offset < si->max; offset += PAGE_SIZE) { | |
3733 | struct page *head; | |
3734 | head = vmalloc_to_page(si->swap_map + offset); | |
3735 | if (page_private(head)) { | |
0d576d20 GT |
3736 | struct page *page, *next; |
3737 | ||
3738 | list_for_each_entry_safe(page, next, &head->lru, lru) { | |
3739 | list_del(&page->lru); | |
570a335b HD |
3740 | __free_page(page); |
3741 | } | |
3742 | } | |
3743 | } | |
3744 | } | |
a2468cc9 | 3745 | |
2cf85583 TH |
3746 | #if defined(CONFIG_MEMCG) && defined(CONFIG_BLK_CGROUP) |
3747 | void mem_cgroup_throttle_swaprate(struct mem_cgroup *memcg, int node, | |
3748 | gfp_t gfp_mask) | |
3749 | { | |
3750 | struct swap_info_struct *si, *next; | |
3751 | if (!(gfp_mask & __GFP_IO) || !memcg) | |
3752 | return; | |
3753 | ||
3754 | if (!blk_cgroup_congested()) | |
3755 | return; | |
3756 | ||
3757 | /* | |
3758 | * We've already scheduled a throttle, avoid taking the global swap | |
3759 | * lock. | |
3760 | */ | |
3761 | if (current->throttle_queue) | |
3762 | return; | |
3763 | ||
3764 | spin_lock(&swap_avail_lock); | |
3765 | plist_for_each_entry_safe(si, next, &swap_avail_heads[node], | |
3766 | avail_lists[node]) { | |
3767 | if (si->bdev) { | |
3768 | blkcg_schedule_throttle(bdev_get_queue(si->bdev), | |
3769 | true); | |
3770 | break; | |
3771 | } | |
3772 | } | |
3773 | spin_unlock(&swap_avail_lock); | |
3774 | } | |
3775 | #endif | |
3776 | ||
a2468cc9 AL |
3777 | static int __init swapfile_init(void) |
3778 | { | |
3779 | int nid; | |
3780 | ||
3781 | swap_avail_heads = kmalloc_array(nr_node_ids, sizeof(struct plist_head), | |
3782 | GFP_KERNEL); | |
3783 | if (!swap_avail_heads) { | |
3784 | pr_emerg("Not enough memory for swap heads, swap is disabled\n"); | |
3785 | return -ENOMEM; | |
3786 | } | |
3787 | ||
3788 | for_each_node(nid) | |
3789 | plist_head_init(&swap_avail_heads[nid]); | |
3790 | ||
3791 | return 0; | |
3792 | } | |
3793 | subsys_initcall(swapfile_init); |