]>
Commit | Line | Data |
---|---|---|
457c8996 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
1da177e4 LT |
2 | /* |
3 | * linux/mm/swapfile.c | |
4 | * | |
5 | * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds | |
6 | * Swap reorganised 29.12.95, Stephen Tweedie | |
7 | */ | |
8 | ||
1da177e4 | 9 | #include <linux/mm.h> |
6e84f315 | 10 | #include <linux/sched/mm.h> |
29930025 | 11 | #include <linux/sched/task.h> |
1da177e4 LT |
12 | #include <linux/hugetlb.h> |
13 | #include <linux/mman.h> | |
14 | #include <linux/slab.h> | |
15 | #include <linux/kernel_stat.h> | |
16 | #include <linux/swap.h> | |
17 | #include <linux/vmalloc.h> | |
18 | #include <linux/pagemap.h> | |
19 | #include <linux/namei.h> | |
072441e2 | 20 | #include <linux/shmem_fs.h> |
1da177e4 | 21 | #include <linux/blkdev.h> |
20137a49 | 22 | #include <linux/random.h> |
1da177e4 LT |
23 | #include <linux/writeback.h> |
24 | #include <linux/proc_fs.h> | |
25 | #include <linux/seq_file.h> | |
26 | #include <linux/init.h> | |
5ad64688 | 27 | #include <linux/ksm.h> |
1da177e4 LT |
28 | #include <linux/rmap.h> |
29 | #include <linux/security.h> | |
30 | #include <linux/backing-dev.h> | |
fc0abb14 | 31 | #include <linux/mutex.h> |
c59ede7b | 32 | #include <linux/capability.h> |
1da177e4 | 33 | #include <linux/syscalls.h> |
8a9f3ccd | 34 | #include <linux/memcontrol.h> |
66d7dd51 | 35 | #include <linux/poll.h> |
72788c38 | 36 | #include <linux/oom.h> |
38b5faf4 DM |
37 | #include <linux/frontswap.h> |
38 | #include <linux/swapfile.h> | |
f981c595 | 39 | #include <linux/export.h> |
67afa38e | 40 | #include <linux/swap_slots.h> |
155b5f88 | 41 | #include <linux/sort.h> |
1da177e4 LT |
42 | |
43 | #include <asm/pgtable.h> | |
44 | #include <asm/tlbflush.h> | |
45 | #include <linux/swapops.h> | |
5d1ea48b | 46 | #include <linux/swap_cgroup.h> |
1da177e4 | 47 | |
570a335b HD |
48 | static bool swap_count_continued(struct swap_info_struct *, pgoff_t, |
49 | unsigned char); | |
50 | static void free_swap_count_continuations(struct swap_info_struct *); | |
d4906e1a | 51 | static sector_t map_swap_entry(swp_entry_t, struct block_device**); |
570a335b | 52 | |
38b5faf4 | 53 | DEFINE_SPINLOCK(swap_lock); |
7c363b8c | 54 | static unsigned int nr_swapfiles; |
ec8acf20 | 55 | atomic_long_t nr_swap_pages; |
fb0fec50 CW |
56 | /* |
57 | * Some modules use swappable objects and may try to swap them out under | |
58 | * memory pressure (via the shrinker). Before doing so, they may wish to | |
59 | * check to see if any swap space is available. | |
60 | */ | |
61 | EXPORT_SYMBOL_GPL(nr_swap_pages); | |
ec8acf20 | 62 | /* protected with swap_lock. reading in vm_swap_full() doesn't need lock */ |
1da177e4 | 63 | long total_swap_pages; |
a2468cc9 | 64 | static int least_priority = -1; |
1da177e4 | 65 | |
1da177e4 LT |
66 | static const char Bad_file[] = "Bad swap file entry "; |
67 | static const char Unused_file[] = "Unused swap file entry "; | |
68 | static const char Bad_offset[] = "Bad swap offset entry "; | |
69 | static const char Unused_offset[] = "Unused swap offset entry "; | |
70 | ||
adfab836 DS |
71 | /* |
72 | * all active swap_info_structs | |
73 | * protected with swap_lock, and ordered by priority. | |
74 | */ | |
18ab4d4c DS |
75 | PLIST_HEAD(swap_active_head); |
76 | ||
77 | /* | |
78 | * all available (active, not full) swap_info_structs | |
79 | * protected with swap_avail_lock, ordered by priority. | |
80 | * This is used by get_swap_page() instead of swap_active_head | |
81 | * because swap_active_head includes all swap_info_structs, | |
82 | * but get_swap_page() doesn't need to look at full ones. | |
83 | * This uses its own lock instead of swap_lock because when a | |
84 | * swap_info_struct changes between not-full/full, it needs to | |
85 | * add/remove itself to/from this list, but the swap_info_struct->lock | |
86 | * is held and the locking order requires swap_lock to be taken | |
87 | * before any swap_info_struct->lock. | |
88 | */ | |
bfc6b1ca | 89 | static struct plist_head *swap_avail_heads; |
18ab4d4c | 90 | static DEFINE_SPINLOCK(swap_avail_lock); |
1da177e4 | 91 | |
38b5faf4 | 92 | struct swap_info_struct *swap_info[MAX_SWAPFILES]; |
1da177e4 | 93 | |
fc0abb14 | 94 | static DEFINE_MUTEX(swapon_mutex); |
1da177e4 | 95 | |
66d7dd51 KS |
96 | static DECLARE_WAIT_QUEUE_HEAD(proc_poll_wait); |
97 | /* Activity counter to indicate that a swapon or swapoff has occurred */ | |
98 | static atomic_t proc_poll_event = ATOMIC_INIT(0); | |
99 | ||
81a0298b HY |
100 | atomic_t nr_rotate_swap = ATOMIC_INIT(0); |
101 | ||
c10d38cc DJ |
102 | static struct swap_info_struct *swap_type_to_swap_info(int type) |
103 | { | |
104 | if (type >= READ_ONCE(nr_swapfiles)) | |
105 | return NULL; | |
106 | ||
107 | smp_rmb(); /* Pairs with smp_wmb in alloc_swap_info. */ | |
108 | return READ_ONCE(swap_info[type]); | |
109 | } | |
110 | ||
8d69aaee | 111 | static inline unsigned char swap_count(unsigned char ent) |
355cfa73 | 112 | { |
955c97f0 | 113 | return ent & ~SWAP_HAS_CACHE; /* may include COUNT_CONTINUED flag */ |
355cfa73 KH |
114 | } |
115 | ||
bcd49e86 HY |
116 | /* Reclaim the swap entry anyway if possible */ |
117 | #define TTRS_ANYWAY 0x1 | |
118 | /* | |
119 | * Reclaim the swap entry if there are no more mappings of the | |
120 | * corresponding page | |
121 | */ | |
122 | #define TTRS_UNMAPPED 0x2 | |
123 | /* Reclaim the swap entry if swap is getting full*/ | |
124 | #define TTRS_FULL 0x4 | |
125 | ||
efa90a98 | 126 | /* returns 1 if swap entry is freed */ |
bcd49e86 HY |
127 | static int __try_to_reclaim_swap(struct swap_info_struct *si, |
128 | unsigned long offset, unsigned long flags) | |
c9e44410 | 129 | { |
efa90a98 | 130 | swp_entry_t entry = swp_entry(si->type, offset); |
c9e44410 KH |
131 | struct page *page; |
132 | int ret = 0; | |
133 | ||
bcd49e86 | 134 | page = find_get_page(swap_address_space(entry), offset); |
c9e44410 KH |
135 | if (!page) |
136 | return 0; | |
137 | /* | |
bcd49e86 HY |
138 | * When this function is called from scan_swap_map_slots() and it's |
139 | * called by vmscan.c at reclaiming pages. So, we hold a lock on a page, | |
140 | * here. We have to use trylock for avoiding deadlock. This is a special | |
c9e44410 KH |
141 | * case and you should use try_to_free_swap() with explicit lock_page() |
142 | * in usual operations. | |
143 | */ | |
144 | if (trylock_page(page)) { | |
bcd49e86 HY |
145 | if ((flags & TTRS_ANYWAY) || |
146 | ((flags & TTRS_UNMAPPED) && !page_mapped(page)) || | |
147 | ((flags & TTRS_FULL) && mem_cgroup_swap_full(page))) | |
148 | ret = try_to_free_swap(page); | |
c9e44410 KH |
149 | unlock_page(page); |
150 | } | |
09cbfeaf | 151 | put_page(page); |
c9e44410 KH |
152 | return ret; |
153 | } | |
355cfa73 | 154 | |
4efaceb1 AL |
155 | static inline struct swap_extent *first_se(struct swap_info_struct *sis) |
156 | { | |
157 | struct rb_node *rb = rb_first(&sis->swap_extent_root); | |
158 | return rb_entry(rb, struct swap_extent, rb_node); | |
159 | } | |
160 | ||
161 | static inline struct swap_extent *next_se(struct swap_extent *se) | |
162 | { | |
163 | struct rb_node *rb = rb_next(&se->rb_node); | |
164 | return rb ? rb_entry(rb, struct swap_extent, rb_node) : NULL; | |
165 | } | |
166 | ||
6a6ba831 HD |
167 | /* |
168 | * swapon tell device that all the old swap contents can be discarded, | |
169 | * to allow the swap device to optimize its wear-levelling. | |
170 | */ | |
171 | static int discard_swap(struct swap_info_struct *si) | |
172 | { | |
173 | struct swap_extent *se; | |
9625a5f2 HD |
174 | sector_t start_block; |
175 | sector_t nr_blocks; | |
6a6ba831 HD |
176 | int err = 0; |
177 | ||
9625a5f2 | 178 | /* Do not discard the swap header page! */ |
4efaceb1 | 179 | se = first_se(si); |
9625a5f2 HD |
180 | start_block = (se->start_block + 1) << (PAGE_SHIFT - 9); |
181 | nr_blocks = ((sector_t)se->nr_pages - 1) << (PAGE_SHIFT - 9); | |
182 | if (nr_blocks) { | |
183 | err = blkdev_issue_discard(si->bdev, start_block, | |
dd3932ed | 184 | nr_blocks, GFP_KERNEL, 0); |
9625a5f2 HD |
185 | if (err) |
186 | return err; | |
187 | cond_resched(); | |
188 | } | |
6a6ba831 | 189 | |
4efaceb1 | 190 | for (se = next_se(se); se; se = next_se(se)) { |
9625a5f2 HD |
191 | start_block = se->start_block << (PAGE_SHIFT - 9); |
192 | nr_blocks = (sector_t)se->nr_pages << (PAGE_SHIFT - 9); | |
6a6ba831 HD |
193 | |
194 | err = blkdev_issue_discard(si->bdev, start_block, | |
dd3932ed | 195 | nr_blocks, GFP_KERNEL, 0); |
6a6ba831 HD |
196 | if (err) |
197 | break; | |
198 | ||
199 | cond_resched(); | |
200 | } | |
201 | return err; /* That will often be -EOPNOTSUPP */ | |
202 | } | |
203 | ||
4efaceb1 AL |
204 | static struct swap_extent * |
205 | offset_to_swap_extent(struct swap_info_struct *sis, unsigned long offset) | |
206 | { | |
207 | struct swap_extent *se; | |
208 | struct rb_node *rb; | |
209 | ||
210 | rb = sis->swap_extent_root.rb_node; | |
211 | while (rb) { | |
212 | se = rb_entry(rb, struct swap_extent, rb_node); | |
213 | if (offset < se->start_page) | |
214 | rb = rb->rb_left; | |
215 | else if (offset >= se->start_page + se->nr_pages) | |
216 | rb = rb->rb_right; | |
217 | else | |
218 | return se; | |
219 | } | |
220 | /* It *must* be present */ | |
221 | BUG(); | |
222 | } | |
223 | ||
7992fde7 HD |
224 | /* |
225 | * swap allocation tell device that a cluster of swap can now be discarded, | |
226 | * to allow the swap device to optimize its wear-levelling. | |
227 | */ | |
228 | static void discard_swap_cluster(struct swap_info_struct *si, | |
229 | pgoff_t start_page, pgoff_t nr_pages) | |
230 | { | |
4efaceb1 | 231 | struct swap_extent *se = offset_to_swap_extent(si, start_page); |
7992fde7 HD |
232 | |
233 | while (nr_pages) { | |
4efaceb1 AL |
234 | pgoff_t offset = start_page - se->start_page; |
235 | sector_t start_block = se->start_block + offset; | |
236 | sector_t nr_blocks = se->nr_pages - offset; | |
237 | ||
238 | if (nr_blocks > nr_pages) | |
239 | nr_blocks = nr_pages; | |
240 | start_page += nr_blocks; | |
241 | nr_pages -= nr_blocks; | |
242 | ||
243 | start_block <<= PAGE_SHIFT - 9; | |
244 | nr_blocks <<= PAGE_SHIFT - 9; | |
245 | if (blkdev_issue_discard(si->bdev, start_block, | |
246 | nr_blocks, GFP_NOIO, 0)) | |
247 | break; | |
7992fde7 | 248 | |
4efaceb1 | 249 | se = next_se(se); |
7992fde7 HD |
250 | } |
251 | } | |
252 | ||
38d8b4e6 HY |
253 | #ifdef CONFIG_THP_SWAP |
254 | #define SWAPFILE_CLUSTER HPAGE_PMD_NR | |
a448f2d0 HY |
255 | |
256 | #define swap_entry_size(size) (size) | |
38d8b4e6 | 257 | #else |
048c27fd | 258 | #define SWAPFILE_CLUSTER 256 |
a448f2d0 HY |
259 | |
260 | /* | |
261 | * Define swap_entry_size() as constant to let compiler to optimize | |
262 | * out some code if !CONFIG_THP_SWAP | |
263 | */ | |
264 | #define swap_entry_size(size) 1 | |
38d8b4e6 | 265 | #endif |
048c27fd HD |
266 | #define LATENCY_LIMIT 256 |
267 | ||
2a8f9449 SL |
268 | static inline void cluster_set_flag(struct swap_cluster_info *info, |
269 | unsigned int flag) | |
270 | { | |
271 | info->flags = flag; | |
272 | } | |
273 | ||
274 | static inline unsigned int cluster_count(struct swap_cluster_info *info) | |
275 | { | |
276 | return info->data; | |
277 | } | |
278 | ||
279 | static inline void cluster_set_count(struct swap_cluster_info *info, | |
280 | unsigned int c) | |
281 | { | |
282 | info->data = c; | |
283 | } | |
284 | ||
285 | static inline void cluster_set_count_flag(struct swap_cluster_info *info, | |
286 | unsigned int c, unsigned int f) | |
287 | { | |
288 | info->flags = f; | |
289 | info->data = c; | |
290 | } | |
291 | ||
292 | static inline unsigned int cluster_next(struct swap_cluster_info *info) | |
293 | { | |
294 | return info->data; | |
295 | } | |
296 | ||
297 | static inline void cluster_set_next(struct swap_cluster_info *info, | |
298 | unsigned int n) | |
299 | { | |
300 | info->data = n; | |
301 | } | |
302 | ||
303 | static inline void cluster_set_next_flag(struct swap_cluster_info *info, | |
304 | unsigned int n, unsigned int f) | |
305 | { | |
306 | info->flags = f; | |
307 | info->data = n; | |
308 | } | |
309 | ||
310 | static inline bool cluster_is_free(struct swap_cluster_info *info) | |
311 | { | |
312 | return info->flags & CLUSTER_FLAG_FREE; | |
313 | } | |
314 | ||
315 | static inline bool cluster_is_null(struct swap_cluster_info *info) | |
316 | { | |
317 | return info->flags & CLUSTER_FLAG_NEXT_NULL; | |
318 | } | |
319 | ||
320 | static inline void cluster_set_null(struct swap_cluster_info *info) | |
321 | { | |
322 | info->flags = CLUSTER_FLAG_NEXT_NULL; | |
323 | info->data = 0; | |
324 | } | |
325 | ||
e0709829 HY |
326 | static inline bool cluster_is_huge(struct swap_cluster_info *info) |
327 | { | |
33ee011e HY |
328 | if (IS_ENABLED(CONFIG_THP_SWAP)) |
329 | return info->flags & CLUSTER_FLAG_HUGE; | |
330 | return false; | |
e0709829 HY |
331 | } |
332 | ||
333 | static inline void cluster_clear_huge(struct swap_cluster_info *info) | |
334 | { | |
335 | info->flags &= ~CLUSTER_FLAG_HUGE; | |
336 | } | |
337 | ||
235b6217 HY |
338 | static inline struct swap_cluster_info *lock_cluster(struct swap_info_struct *si, |
339 | unsigned long offset) | |
340 | { | |
341 | struct swap_cluster_info *ci; | |
342 | ||
343 | ci = si->cluster_info; | |
344 | if (ci) { | |
345 | ci += offset / SWAPFILE_CLUSTER; | |
346 | spin_lock(&ci->lock); | |
347 | } | |
348 | return ci; | |
349 | } | |
350 | ||
351 | static inline void unlock_cluster(struct swap_cluster_info *ci) | |
352 | { | |
353 | if (ci) | |
354 | spin_unlock(&ci->lock); | |
355 | } | |
356 | ||
59d98bf3 HY |
357 | /* |
358 | * Determine the locking method in use for this device. Return | |
359 | * swap_cluster_info if SSD-style cluster-based locking is in place. | |
360 | */ | |
235b6217 | 361 | static inline struct swap_cluster_info *lock_cluster_or_swap_info( |
59d98bf3 | 362 | struct swap_info_struct *si, unsigned long offset) |
235b6217 HY |
363 | { |
364 | struct swap_cluster_info *ci; | |
365 | ||
59d98bf3 | 366 | /* Try to use fine-grained SSD-style locking if available: */ |
235b6217 | 367 | ci = lock_cluster(si, offset); |
59d98bf3 | 368 | /* Otherwise, fall back to traditional, coarse locking: */ |
235b6217 HY |
369 | if (!ci) |
370 | spin_lock(&si->lock); | |
371 | ||
372 | return ci; | |
373 | } | |
374 | ||
375 | static inline void unlock_cluster_or_swap_info(struct swap_info_struct *si, | |
376 | struct swap_cluster_info *ci) | |
377 | { | |
378 | if (ci) | |
379 | unlock_cluster(ci); | |
380 | else | |
381 | spin_unlock(&si->lock); | |
382 | } | |
383 | ||
6b534915 HY |
384 | static inline bool cluster_list_empty(struct swap_cluster_list *list) |
385 | { | |
386 | return cluster_is_null(&list->head); | |
387 | } | |
388 | ||
389 | static inline unsigned int cluster_list_first(struct swap_cluster_list *list) | |
390 | { | |
391 | return cluster_next(&list->head); | |
392 | } | |
393 | ||
394 | static void cluster_list_init(struct swap_cluster_list *list) | |
395 | { | |
396 | cluster_set_null(&list->head); | |
397 | cluster_set_null(&list->tail); | |
398 | } | |
399 | ||
400 | static void cluster_list_add_tail(struct swap_cluster_list *list, | |
401 | struct swap_cluster_info *ci, | |
402 | unsigned int idx) | |
403 | { | |
404 | if (cluster_list_empty(list)) { | |
405 | cluster_set_next_flag(&list->head, idx, 0); | |
406 | cluster_set_next_flag(&list->tail, idx, 0); | |
407 | } else { | |
235b6217 | 408 | struct swap_cluster_info *ci_tail; |
6b534915 HY |
409 | unsigned int tail = cluster_next(&list->tail); |
410 | ||
235b6217 HY |
411 | /* |
412 | * Nested cluster lock, but both cluster locks are | |
413 | * only acquired when we held swap_info_struct->lock | |
414 | */ | |
415 | ci_tail = ci + tail; | |
416 | spin_lock_nested(&ci_tail->lock, SINGLE_DEPTH_NESTING); | |
417 | cluster_set_next(ci_tail, idx); | |
0ef017d1 | 418 | spin_unlock(&ci_tail->lock); |
6b534915 HY |
419 | cluster_set_next_flag(&list->tail, idx, 0); |
420 | } | |
421 | } | |
422 | ||
423 | static unsigned int cluster_list_del_first(struct swap_cluster_list *list, | |
424 | struct swap_cluster_info *ci) | |
425 | { | |
426 | unsigned int idx; | |
427 | ||
428 | idx = cluster_next(&list->head); | |
429 | if (cluster_next(&list->tail) == idx) { | |
430 | cluster_set_null(&list->head); | |
431 | cluster_set_null(&list->tail); | |
432 | } else | |
433 | cluster_set_next_flag(&list->head, | |
434 | cluster_next(&ci[idx]), 0); | |
435 | ||
436 | return idx; | |
437 | } | |
438 | ||
815c2c54 SL |
439 | /* Add a cluster to discard list and schedule it to do discard */ |
440 | static void swap_cluster_schedule_discard(struct swap_info_struct *si, | |
441 | unsigned int idx) | |
442 | { | |
443 | /* | |
444 | * If scan_swap_map() can't find a free cluster, it will check | |
445 | * si->swap_map directly. To make sure the discarding cluster isn't | |
446 | * taken by scan_swap_map(), mark the swap entries bad (occupied). It | |
447 | * will be cleared after discard | |
448 | */ | |
449 | memset(si->swap_map + idx * SWAPFILE_CLUSTER, | |
450 | SWAP_MAP_BAD, SWAPFILE_CLUSTER); | |
451 | ||
6b534915 | 452 | cluster_list_add_tail(&si->discard_clusters, si->cluster_info, idx); |
815c2c54 SL |
453 | |
454 | schedule_work(&si->discard_work); | |
455 | } | |
456 | ||
38d8b4e6 HY |
457 | static void __free_cluster(struct swap_info_struct *si, unsigned long idx) |
458 | { | |
459 | struct swap_cluster_info *ci = si->cluster_info; | |
460 | ||
461 | cluster_set_flag(ci + idx, CLUSTER_FLAG_FREE); | |
462 | cluster_list_add_tail(&si->free_clusters, ci, idx); | |
463 | } | |
464 | ||
815c2c54 SL |
465 | /* |
466 | * Doing discard actually. After a cluster discard is finished, the cluster | |
467 | * will be added to free cluster list. caller should hold si->lock. | |
468 | */ | |
469 | static void swap_do_scheduled_discard(struct swap_info_struct *si) | |
470 | { | |
235b6217 | 471 | struct swap_cluster_info *info, *ci; |
815c2c54 SL |
472 | unsigned int idx; |
473 | ||
474 | info = si->cluster_info; | |
475 | ||
6b534915 HY |
476 | while (!cluster_list_empty(&si->discard_clusters)) { |
477 | idx = cluster_list_del_first(&si->discard_clusters, info); | |
815c2c54 SL |
478 | spin_unlock(&si->lock); |
479 | ||
480 | discard_swap_cluster(si, idx * SWAPFILE_CLUSTER, | |
481 | SWAPFILE_CLUSTER); | |
482 | ||
483 | spin_lock(&si->lock); | |
235b6217 | 484 | ci = lock_cluster(si, idx * SWAPFILE_CLUSTER); |
38d8b4e6 | 485 | __free_cluster(si, idx); |
815c2c54 SL |
486 | memset(si->swap_map + idx * SWAPFILE_CLUSTER, |
487 | 0, SWAPFILE_CLUSTER); | |
235b6217 | 488 | unlock_cluster(ci); |
815c2c54 SL |
489 | } |
490 | } | |
491 | ||
492 | static void swap_discard_work(struct work_struct *work) | |
493 | { | |
494 | struct swap_info_struct *si; | |
495 | ||
496 | si = container_of(work, struct swap_info_struct, discard_work); | |
497 | ||
498 | spin_lock(&si->lock); | |
499 | swap_do_scheduled_discard(si); | |
500 | spin_unlock(&si->lock); | |
501 | } | |
502 | ||
38d8b4e6 HY |
503 | static void alloc_cluster(struct swap_info_struct *si, unsigned long idx) |
504 | { | |
505 | struct swap_cluster_info *ci = si->cluster_info; | |
506 | ||
507 | VM_BUG_ON(cluster_list_first(&si->free_clusters) != idx); | |
508 | cluster_list_del_first(&si->free_clusters, ci); | |
509 | cluster_set_count_flag(ci + idx, 0, 0); | |
510 | } | |
511 | ||
512 | static void free_cluster(struct swap_info_struct *si, unsigned long idx) | |
513 | { | |
514 | struct swap_cluster_info *ci = si->cluster_info + idx; | |
515 | ||
516 | VM_BUG_ON(cluster_count(ci) != 0); | |
517 | /* | |
518 | * If the swap is discardable, prepare discard the cluster | |
519 | * instead of free it immediately. The cluster will be freed | |
520 | * after discard. | |
521 | */ | |
522 | if ((si->flags & (SWP_WRITEOK | SWP_PAGE_DISCARD)) == | |
523 | (SWP_WRITEOK | SWP_PAGE_DISCARD)) { | |
524 | swap_cluster_schedule_discard(si, idx); | |
525 | return; | |
526 | } | |
527 | ||
528 | __free_cluster(si, idx); | |
529 | } | |
530 | ||
2a8f9449 SL |
531 | /* |
532 | * The cluster corresponding to page_nr will be used. The cluster will be | |
533 | * removed from free cluster list and its usage counter will be increased. | |
534 | */ | |
535 | static void inc_cluster_info_page(struct swap_info_struct *p, | |
536 | struct swap_cluster_info *cluster_info, unsigned long page_nr) | |
537 | { | |
538 | unsigned long idx = page_nr / SWAPFILE_CLUSTER; | |
539 | ||
540 | if (!cluster_info) | |
541 | return; | |
38d8b4e6 HY |
542 | if (cluster_is_free(&cluster_info[idx])) |
543 | alloc_cluster(p, idx); | |
2a8f9449 SL |
544 | |
545 | VM_BUG_ON(cluster_count(&cluster_info[idx]) >= SWAPFILE_CLUSTER); | |
546 | cluster_set_count(&cluster_info[idx], | |
547 | cluster_count(&cluster_info[idx]) + 1); | |
548 | } | |
549 | ||
550 | /* | |
551 | * The cluster corresponding to page_nr decreases one usage. If the usage | |
552 | * counter becomes 0, which means no page in the cluster is in using, we can | |
553 | * optionally discard the cluster and add it to free cluster list. | |
554 | */ | |
555 | static void dec_cluster_info_page(struct swap_info_struct *p, | |
556 | struct swap_cluster_info *cluster_info, unsigned long page_nr) | |
557 | { | |
558 | unsigned long idx = page_nr / SWAPFILE_CLUSTER; | |
559 | ||
560 | if (!cluster_info) | |
561 | return; | |
562 | ||
563 | VM_BUG_ON(cluster_count(&cluster_info[idx]) == 0); | |
564 | cluster_set_count(&cluster_info[idx], | |
565 | cluster_count(&cluster_info[idx]) - 1); | |
566 | ||
38d8b4e6 HY |
567 | if (cluster_count(&cluster_info[idx]) == 0) |
568 | free_cluster(p, idx); | |
2a8f9449 SL |
569 | } |
570 | ||
571 | /* | |
572 | * It's possible scan_swap_map() uses a free cluster in the middle of free | |
573 | * cluster list. Avoiding such abuse to avoid list corruption. | |
574 | */ | |
ebc2a1a6 SL |
575 | static bool |
576 | scan_swap_map_ssd_cluster_conflict(struct swap_info_struct *si, | |
2a8f9449 SL |
577 | unsigned long offset) |
578 | { | |
ebc2a1a6 SL |
579 | struct percpu_cluster *percpu_cluster; |
580 | bool conflict; | |
581 | ||
2a8f9449 | 582 | offset /= SWAPFILE_CLUSTER; |
6b534915 HY |
583 | conflict = !cluster_list_empty(&si->free_clusters) && |
584 | offset != cluster_list_first(&si->free_clusters) && | |
2a8f9449 | 585 | cluster_is_free(&si->cluster_info[offset]); |
ebc2a1a6 SL |
586 | |
587 | if (!conflict) | |
588 | return false; | |
589 | ||
590 | percpu_cluster = this_cpu_ptr(si->percpu_cluster); | |
591 | cluster_set_null(&percpu_cluster->index); | |
592 | return true; | |
593 | } | |
594 | ||
595 | /* | |
596 | * Try to get a swap entry from current cpu's swap entry pool (a cluster). This | |
597 | * might involve allocating a new cluster for current CPU too. | |
598 | */ | |
36005bae | 599 | static bool scan_swap_map_try_ssd_cluster(struct swap_info_struct *si, |
ebc2a1a6 SL |
600 | unsigned long *offset, unsigned long *scan_base) |
601 | { | |
602 | struct percpu_cluster *cluster; | |
235b6217 | 603 | struct swap_cluster_info *ci; |
ebc2a1a6 | 604 | bool found_free; |
235b6217 | 605 | unsigned long tmp, max; |
ebc2a1a6 SL |
606 | |
607 | new_cluster: | |
608 | cluster = this_cpu_ptr(si->percpu_cluster); | |
609 | if (cluster_is_null(&cluster->index)) { | |
6b534915 HY |
610 | if (!cluster_list_empty(&si->free_clusters)) { |
611 | cluster->index = si->free_clusters.head; | |
ebc2a1a6 SL |
612 | cluster->next = cluster_next(&cluster->index) * |
613 | SWAPFILE_CLUSTER; | |
6b534915 | 614 | } else if (!cluster_list_empty(&si->discard_clusters)) { |
ebc2a1a6 SL |
615 | /* |
616 | * we don't have free cluster but have some clusters in | |
617 | * discarding, do discard now and reclaim them | |
618 | */ | |
619 | swap_do_scheduled_discard(si); | |
620 | *scan_base = *offset = si->cluster_next; | |
621 | goto new_cluster; | |
622 | } else | |
36005bae | 623 | return false; |
ebc2a1a6 SL |
624 | } |
625 | ||
626 | found_free = false; | |
627 | ||
628 | /* | |
629 | * Other CPUs can use our cluster if they can't find a free cluster, | |
630 | * check if there is still free entry in the cluster | |
631 | */ | |
632 | tmp = cluster->next; | |
235b6217 HY |
633 | max = min_t(unsigned long, si->max, |
634 | (cluster_next(&cluster->index) + 1) * SWAPFILE_CLUSTER); | |
635 | if (tmp >= max) { | |
636 | cluster_set_null(&cluster->index); | |
637 | goto new_cluster; | |
638 | } | |
639 | ci = lock_cluster(si, tmp); | |
640 | while (tmp < max) { | |
ebc2a1a6 SL |
641 | if (!si->swap_map[tmp]) { |
642 | found_free = true; | |
643 | break; | |
644 | } | |
645 | tmp++; | |
646 | } | |
235b6217 | 647 | unlock_cluster(ci); |
ebc2a1a6 SL |
648 | if (!found_free) { |
649 | cluster_set_null(&cluster->index); | |
650 | goto new_cluster; | |
651 | } | |
652 | cluster->next = tmp + 1; | |
653 | *offset = tmp; | |
654 | *scan_base = tmp; | |
36005bae | 655 | return found_free; |
2a8f9449 SL |
656 | } |
657 | ||
a2468cc9 AL |
658 | static void __del_from_avail_list(struct swap_info_struct *p) |
659 | { | |
660 | int nid; | |
661 | ||
662 | for_each_node(nid) | |
663 | plist_del(&p->avail_lists[nid], &swap_avail_heads[nid]); | |
664 | } | |
665 | ||
666 | static void del_from_avail_list(struct swap_info_struct *p) | |
667 | { | |
668 | spin_lock(&swap_avail_lock); | |
669 | __del_from_avail_list(p); | |
670 | spin_unlock(&swap_avail_lock); | |
671 | } | |
672 | ||
38d8b4e6 HY |
673 | static void swap_range_alloc(struct swap_info_struct *si, unsigned long offset, |
674 | unsigned int nr_entries) | |
675 | { | |
676 | unsigned int end = offset + nr_entries - 1; | |
677 | ||
678 | if (offset == si->lowest_bit) | |
679 | si->lowest_bit += nr_entries; | |
680 | if (end == si->highest_bit) | |
681 | si->highest_bit -= nr_entries; | |
682 | si->inuse_pages += nr_entries; | |
683 | if (si->inuse_pages == si->pages) { | |
684 | si->lowest_bit = si->max; | |
685 | si->highest_bit = 0; | |
a2468cc9 | 686 | del_from_avail_list(si); |
38d8b4e6 HY |
687 | } |
688 | } | |
689 | ||
a2468cc9 AL |
690 | static void add_to_avail_list(struct swap_info_struct *p) |
691 | { | |
692 | int nid; | |
693 | ||
694 | spin_lock(&swap_avail_lock); | |
695 | for_each_node(nid) { | |
696 | WARN_ON(!plist_node_empty(&p->avail_lists[nid])); | |
697 | plist_add(&p->avail_lists[nid], &swap_avail_heads[nid]); | |
698 | } | |
699 | spin_unlock(&swap_avail_lock); | |
700 | } | |
701 | ||
38d8b4e6 HY |
702 | static void swap_range_free(struct swap_info_struct *si, unsigned long offset, |
703 | unsigned int nr_entries) | |
704 | { | |
705 | unsigned long end = offset + nr_entries - 1; | |
706 | void (*swap_slot_free_notify)(struct block_device *, unsigned long); | |
707 | ||
708 | if (offset < si->lowest_bit) | |
709 | si->lowest_bit = offset; | |
710 | if (end > si->highest_bit) { | |
711 | bool was_full = !si->highest_bit; | |
712 | ||
713 | si->highest_bit = end; | |
a2468cc9 AL |
714 | if (was_full && (si->flags & SWP_WRITEOK)) |
715 | add_to_avail_list(si); | |
38d8b4e6 HY |
716 | } |
717 | atomic_long_add(nr_entries, &nr_swap_pages); | |
718 | si->inuse_pages -= nr_entries; | |
719 | if (si->flags & SWP_BLKDEV) | |
720 | swap_slot_free_notify = | |
721 | si->bdev->bd_disk->fops->swap_slot_free_notify; | |
722 | else | |
723 | swap_slot_free_notify = NULL; | |
724 | while (offset <= end) { | |
725 | frontswap_invalidate_page(si->type, offset); | |
726 | if (swap_slot_free_notify) | |
727 | swap_slot_free_notify(si->bdev, offset); | |
728 | offset++; | |
729 | } | |
730 | } | |
731 | ||
36005bae TC |
732 | static int scan_swap_map_slots(struct swap_info_struct *si, |
733 | unsigned char usage, int nr, | |
734 | swp_entry_t slots[]) | |
1da177e4 | 735 | { |
235b6217 | 736 | struct swap_cluster_info *ci; |
ebebbbe9 | 737 | unsigned long offset; |
c60aa176 | 738 | unsigned long scan_base; |
7992fde7 | 739 | unsigned long last_in_cluster = 0; |
048c27fd | 740 | int latency_ration = LATENCY_LIMIT; |
36005bae TC |
741 | int n_ret = 0; |
742 | ||
743 | if (nr > SWAP_BATCH) | |
744 | nr = SWAP_BATCH; | |
7dfad418 | 745 | |
886bb7e9 | 746 | /* |
7dfad418 HD |
747 | * We try to cluster swap pages by allocating them sequentially |
748 | * in swap. Once we've allocated SWAPFILE_CLUSTER pages this | |
749 | * way, however, we resort to first-free allocation, starting | |
750 | * a new cluster. This prevents us from scattering swap pages | |
751 | * all over the entire swap partition, so that we reduce | |
752 | * overall disk seek times between swap pages. -- sct | |
753 | * But we do now try to find an empty cluster. -Andrea | |
c60aa176 | 754 | * And we let swap pages go all over an SSD partition. Hugh |
7dfad418 HD |
755 | */ |
756 | ||
52b7efdb | 757 | si->flags += SWP_SCANNING; |
c60aa176 | 758 | scan_base = offset = si->cluster_next; |
ebebbbe9 | 759 | |
ebc2a1a6 SL |
760 | /* SSD algorithm */ |
761 | if (si->cluster_info) { | |
bd2d18da | 762 | if (!scan_swap_map_try_ssd_cluster(si, &offset, &scan_base)) |
36005bae | 763 | goto scan; |
f4eaf51a | 764 | } else if (unlikely(!si->cluster_nr--)) { |
ebebbbe9 HD |
765 | if (si->pages - si->inuse_pages < SWAPFILE_CLUSTER) { |
766 | si->cluster_nr = SWAPFILE_CLUSTER - 1; | |
767 | goto checks; | |
768 | } | |
2a8f9449 | 769 | |
ec8acf20 | 770 | spin_unlock(&si->lock); |
7dfad418 | 771 | |
c60aa176 HD |
772 | /* |
773 | * If seek is expensive, start searching for new cluster from | |
774 | * start of partition, to minimize the span of allocated swap. | |
50088c44 CY |
775 | * If seek is cheap, that is the SWP_SOLIDSTATE si->cluster_info |
776 | * case, just handled by scan_swap_map_try_ssd_cluster() above. | |
c60aa176 | 777 | */ |
50088c44 | 778 | scan_base = offset = si->lowest_bit; |
7dfad418 HD |
779 | last_in_cluster = offset + SWAPFILE_CLUSTER - 1; |
780 | ||
781 | /* Locate the first empty (unaligned) cluster */ | |
782 | for (; last_in_cluster <= si->highest_bit; offset++) { | |
1da177e4 | 783 | if (si->swap_map[offset]) |
7dfad418 HD |
784 | last_in_cluster = offset + SWAPFILE_CLUSTER; |
785 | else if (offset == last_in_cluster) { | |
ec8acf20 | 786 | spin_lock(&si->lock); |
ebebbbe9 HD |
787 | offset -= SWAPFILE_CLUSTER - 1; |
788 | si->cluster_next = offset; | |
789 | si->cluster_nr = SWAPFILE_CLUSTER - 1; | |
c60aa176 HD |
790 | goto checks; |
791 | } | |
792 | if (unlikely(--latency_ration < 0)) { | |
793 | cond_resched(); | |
794 | latency_ration = LATENCY_LIMIT; | |
795 | } | |
796 | } | |
797 | ||
798 | offset = scan_base; | |
ec8acf20 | 799 | spin_lock(&si->lock); |
ebebbbe9 | 800 | si->cluster_nr = SWAPFILE_CLUSTER - 1; |
1da177e4 | 801 | } |
7dfad418 | 802 | |
ebebbbe9 | 803 | checks: |
ebc2a1a6 | 804 | if (si->cluster_info) { |
36005bae TC |
805 | while (scan_swap_map_ssd_cluster_conflict(si, offset)) { |
806 | /* take a break if we already got some slots */ | |
807 | if (n_ret) | |
808 | goto done; | |
809 | if (!scan_swap_map_try_ssd_cluster(si, &offset, | |
810 | &scan_base)) | |
811 | goto scan; | |
812 | } | |
ebc2a1a6 | 813 | } |
ebebbbe9 | 814 | if (!(si->flags & SWP_WRITEOK)) |
52b7efdb | 815 | goto no_page; |
7dfad418 HD |
816 | if (!si->highest_bit) |
817 | goto no_page; | |
ebebbbe9 | 818 | if (offset > si->highest_bit) |
c60aa176 | 819 | scan_base = offset = si->lowest_bit; |
c9e44410 | 820 | |
235b6217 | 821 | ci = lock_cluster(si, offset); |
b73d7fce HD |
822 | /* reuse swap entry of cache-only swap if not busy. */ |
823 | if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) { | |
c9e44410 | 824 | int swap_was_freed; |
235b6217 | 825 | unlock_cluster(ci); |
ec8acf20 | 826 | spin_unlock(&si->lock); |
bcd49e86 | 827 | swap_was_freed = __try_to_reclaim_swap(si, offset, TTRS_ANYWAY); |
ec8acf20 | 828 | spin_lock(&si->lock); |
c9e44410 KH |
829 | /* entry was freed successfully, try to use this again */ |
830 | if (swap_was_freed) | |
831 | goto checks; | |
832 | goto scan; /* check next one */ | |
833 | } | |
834 | ||
235b6217 HY |
835 | if (si->swap_map[offset]) { |
836 | unlock_cluster(ci); | |
36005bae TC |
837 | if (!n_ret) |
838 | goto scan; | |
839 | else | |
840 | goto done; | |
235b6217 | 841 | } |
2872bb2d HY |
842 | si->swap_map[offset] = usage; |
843 | inc_cluster_info_page(si, si->cluster_info, offset); | |
844 | unlock_cluster(ci); | |
ebebbbe9 | 845 | |
38d8b4e6 | 846 | swap_range_alloc(si, offset, 1); |
ebebbbe9 | 847 | si->cluster_next = offset + 1; |
36005bae TC |
848 | slots[n_ret++] = swp_entry(si->type, offset); |
849 | ||
850 | /* got enough slots or reach max slots? */ | |
851 | if ((n_ret == nr) || (offset >= si->highest_bit)) | |
852 | goto done; | |
853 | ||
854 | /* search for next available slot */ | |
855 | ||
856 | /* time to take a break? */ | |
857 | if (unlikely(--latency_ration < 0)) { | |
858 | if (n_ret) | |
859 | goto done; | |
860 | spin_unlock(&si->lock); | |
861 | cond_resched(); | |
862 | spin_lock(&si->lock); | |
863 | latency_ration = LATENCY_LIMIT; | |
864 | } | |
865 | ||
866 | /* try to get more slots in cluster */ | |
867 | if (si->cluster_info) { | |
868 | if (scan_swap_map_try_ssd_cluster(si, &offset, &scan_base)) | |
869 | goto checks; | |
f4eaf51a WY |
870 | } else if (si->cluster_nr && !si->swap_map[++offset]) { |
871 | /* non-ssd case, still more slots in cluster? */ | |
36005bae TC |
872 | --si->cluster_nr; |
873 | goto checks; | |
874 | } | |
7992fde7 | 875 | |
36005bae TC |
876 | done: |
877 | si->flags -= SWP_SCANNING; | |
878 | return n_ret; | |
7dfad418 | 879 | |
ebebbbe9 | 880 | scan: |
ec8acf20 | 881 | spin_unlock(&si->lock); |
7dfad418 | 882 | while (++offset <= si->highest_bit) { |
52b7efdb | 883 | if (!si->swap_map[offset]) { |
ec8acf20 | 884 | spin_lock(&si->lock); |
52b7efdb HD |
885 | goto checks; |
886 | } | |
c9e44410 | 887 | if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) { |
ec8acf20 | 888 | spin_lock(&si->lock); |
c9e44410 KH |
889 | goto checks; |
890 | } | |
048c27fd HD |
891 | if (unlikely(--latency_ration < 0)) { |
892 | cond_resched(); | |
893 | latency_ration = LATENCY_LIMIT; | |
894 | } | |
7dfad418 | 895 | } |
c60aa176 | 896 | offset = si->lowest_bit; |
a5998061 | 897 | while (offset < scan_base) { |
c60aa176 | 898 | if (!si->swap_map[offset]) { |
ec8acf20 | 899 | spin_lock(&si->lock); |
c60aa176 HD |
900 | goto checks; |
901 | } | |
c9e44410 | 902 | if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) { |
ec8acf20 | 903 | spin_lock(&si->lock); |
c9e44410 KH |
904 | goto checks; |
905 | } | |
c60aa176 HD |
906 | if (unlikely(--latency_ration < 0)) { |
907 | cond_resched(); | |
908 | latency_ration = LATENCY_LIMIT; | |
909 | } | |
a5998061 | 910 | offset++; |
c60aa176 | 911 | } |
ec8acf20 | 912 | spin_lock(&si->lock); |
7dfad418 HD |
913 | |
914 | no_page: | |
52b7efdb | 915 | si->flags -= SWP_SCANNING; |
36005bae | 916 | return n_ret; |
1da177e4 LT |
917 | } |
918 | ||
38d8b4e6 HY |
919 | static int swap_alloc_cluster(struct swap_info_struct *si, swp_entry_t *slot) |
920 | { | |
921 | unsigned long idx; | |
922 | struct swap_cluster_info *ci; | |
923 | unsigned long offset, i; | |
924 | unsigned char *map; | |
925 | ||
fe5266d5 HY |
926 | /* |
927 | * Should not even be attempting cluster allocations when huge | |
928 | * page swap is disabled. Warn and fail the allocation. | |
929 | */ | |
930 | if (!IS_ENABLED(CONFIG_THP_SWAP)) { | |
931 | VM_WARN_ON_ONCE(1); | |
932 | return 0; | |
933 | } | |
934 | ||
38d8b4e6 HY |
935 | if (cluster_list_empty(&si->free_clusters)) |
936 | return 0; | |
937 | ||
938 | idx = cluster_list_first(&si->free_clusters); | |
939 | offset = idx * SWAPFILE_CLUSTER; | |
940 | ci = lock_cluster(si, offset); | |
941 | alloc_cluster(si, idx); | |
e0709829 | 942 | cluster_set_count_flag(ci, SWAPFILE_CLUSTER, CLUSTER_FLAG_HUGE); |
38d8b4e6 HY |
943 | |
944 | map = si->swap_map + offset; | |
945 | for (i = 0; i < SWAPFILE_CLUSTER; i++) | |
946 | map[i] = SWAP_HAS_CACHE; | |
947 | unlock_cluster(ci); | |
948 | swap_range_alloc(si, offset, SWAPFILE_CLUSTER); | |
949 | *slot = swp_entry(si->type, offset); | |
950 | ||
951 | return 1; | |
952 | } | |
953 | ||
954 | static void swap_free_cluster(struct swap_info_struct *si, unsigned long idx) | |
955 | { | |
956 | unsigned long offset = idx * SWAPFILE_CLUSTER; | |
957 | struct swap_cluster_info *ci; | |
958 | ||
959 | ci = lock_cluster(si, offset); | |
979aafa5 | 960 | memset(si->swap_map + offset, 0, SWAPFILE_CLUSTER); |
38d8b4e6 HY |
961 | cluster_set_count_flag(ci, 0, 0); |
962 | free_cluster(si, idx); | |
963 | unlock_cluster(ci); | |
964 | swap_range_free(si, offset, SWAPFILE_CLUSTER); | |
965 | } | |
38d8b4e6 | 966 | |
36005bae TC |
967 | static unsigned long scan_swap_map(struct swap_info_struct *si, |
968 | unsigned char usage) | |
969 | { | |
970 | swp_entry_t entry; | |
971 | int n_ret; | |
972 | ||
973 | n_ret = scan_swap_map_slots(si, usage, 1, &entry); | |
974 | ||
975 | if (n_ret) | |
976 | return swp_offset(entry); | |
977 | else | |
978 | return 0; | |
979 | ||
980 | } | |
981 | ||
5d5e8f19 | 982 | int get_swap_pages(int n_goal, swp_entry_t swp_entries[], int entry_size) |
1da177e4 | 983 | { |
5d5e8f19 | 984 | unsigned long size = swap_entry_size(entry_size); |
adfab836 | 985 | struct swap_info_struct *si, *next; |
36005bae TC |
986 | long avail_pgs; |
987 | int n_ret = 0; | |
a2468cc9 | 988 | int node; |
1da177e4 | 989 | |
38d8b4e6 | 990 | /* Only single cluster request supported */ |
5d5e8f19 | 991 | WARN_ON_ONCE(n_goal > 1 && size == SWAPFILE_CLUSTER); |
38d8b4e6 | 992 | |
5d5e8f19 | 993 | avail_pgs = atomic_long_read(&nr_swap_pages) / size; |
36005bae | 994 | if (avail_pgs <= 0) |
fb4f88dc | 995 | goto noswap; |
36005bae | 996 | |
08d3090f | 997 | n_goal = min3((long)n_goal, (long)SWAP_BATCH, avail_pgs); |
36005bae | 998 | |
5d5e8f19 | 999 | atomic_long_sub(n_goal * size, &nr_swap_pages); |
fb4f88dc | 1000 | |
18ab4d4c DS |
1001 | spin_lock(&swap_avail_lock); |
1002 | ||
1003 | start_over: | |
a2468cc9 AL |
1004 | node = numa_node_id(); |
1005 | plist_for_each_entry_safe(si, next, &swap_avail_heads[node], avail_lists[node]) { | |
18ab4d4c | 1006 | /* requeue si to after same-priority siblings */ |
a2468cc9 | 1007 | plist_requeue(&si->avail_lists[node], &swap_avail_heads[node]); |
18ab4d4c | 1008 | spin_unlock(&swap_avail_lock); |
ec8acf20 | 1009 | spin_lock(&si->lock); |
adfab836 | 1010 | if (!si->highest_bit || !(si->flags & SWP_WRITEOK)) { |
18ab4d4c | 1011 | spin_lock(&swap_avail_lock); |
a2468cc9 | 1012 | if (plist_node_empty(&si->avail_lists[node])) { |
18ab4d4c DS |
1013 | spin_unlock(&si->lock); |
1014 | goto nextsi; | |
1015 | } | |
1016 | WARN(!si->highest_bit, | |
1017 | "swap_info %d in list but !highest_bit\n", | |
1018 | si->type); | |
1019 | WARN(!(si->flags & SWP_WRITEOK), | |
1020 | "swap_info %d in list but !SWP_WRITEOK\n", | |
1021 | si->type); | |
a2468cc9 | 1022 | __del_from_avail_list(si); |
ec8acf20 | 1023 | spin_unlock(&si->lock); |
18ab4d4c | 1024 | goto nextsi; |
ec8acf20 | 1025 | } |
5d5e8f19 | 1026 | if (size == SWAPFILE_CLUSTER) { |
bc4ae27d | 1027 | if (!(si->flags & SWP_FS)) |
f0eea189 HY |
1028 | n_ret = swap_alloc_cluster(si, swp_entries); |
1029 | } else | |
38d8b4e6 HY |
1030 | n_ret = scan_swap_map_slots(si, SWAP_HAS_CACHE, |
1031 | n_goal, swp_entries); | |
ec8acf20 | 1032 | spin_unlock(&si->lock); |
5d5e8f19 | 1033 | if (n_ret || size == SWAPFILE_CLUSTER) |
36005bae | 1034 | goto check_out; |
18ab4d4c | 1035 | pr_debug("scan_swap_map of si %d failed to find offset\n", |
36005bae TC |
1036 | si->type); |
1037 | ||
18ab4d4c DS |
1038 | spin_lock(&swap_avail_lock); |
1039 | nextsi: | |
adfab836 DS |
1040 | /* |
1041 | * if we got here, it's likely that si was almost full before, | |
1042 | * and since scan_swap_map() can drop the si->lock, multiple | |
1043 | * callers probably all tried to get a page from the same si | |
18ab4d4c DS |
1044 | * and it filled up before we could get one; or, the si filled |
1045 | * up between us dropping swap_avail_lock and taking si->lock. | |
1046 | * Since we dropped the swap_avail_lock, the swap_avail_head | |
1047 | * list may have been modified; so if next is still in the | |
36005bae TC |
1048 | * swap_avail_head list then try it, otherwise start over |
1049 | * if we have not gotten any slots. | |
adfab836 | 1050 | */ |
a2468cc9 | 1051 | if (plist_node_empty(&next->avail_lists[node])) |
18ab4d4c | 1052 | goto start_over; |
1da177e4 | 1053 | } |
fb4f88dc | 1054 | |
18ab4d4c DS |
1055 | spin_unlock(&swap_avail_lock); |
1056 | ||
36005bae TC |
1057 | check_out: |
1058 | if (n_ret < n_goal) | |
5d5e8f19 | 1059 | atomic_long_add((long)(n_goal - n_ret) * size, |
38d8b4e6 | 1060 | &nr_swap_pages); |
fb4f88dc | 1061 | noswap: |
36005bae TC |
1062 | return n_ret; |
1063 | } | |
1064 | ||
2de1a7e4 | 1065 | /* The only caller of this function is now suspend routine */ |
910321ea HD |
1066 | swp_entry_t get_swap_page_of_type(int type) |
1067 | { | |
c10d38cc | 1068 | struct swap_info_struct *si = swap_type_to_swap_info(type); |
910321ea HD |
1069 | pgoff_t offset; |
1070 | ||
c10d38cc DJ |
1071 | if (!si) |
1072 | goto fail; | |
1073 | ||
ec8acf20 | 1074 | spin_lock(&si->lock); |
c10d38cc | 1075 | if (si->flags & SWP_WRITEOK) { |
ec8acf20 | 1076 | atomic_long_dec(&nr_swap_pages); |
910321ea HD |
1077 | /* This is called for allocating swap entry, not cache */ |
1078 | offset = scan_swap_map(si, 1); | |
1079 | if (offset) { | |
ec8acf20 | 1080 | spin_unlock(&si->lock); |
910321ea HD |
1081 | return swp_entry(type, offset); |
1082 | } | |
ec8acf20 | 1083 | atomic_long_inc(&nr_swap_pages); |
910321ea | 1084 | } |
ec8acf20 | 1085 | spin_unlock(&si->lock); |
c10d38cc | 1086 | fail: |
910321ea HD |
1087 | return (swp_entry_t) {0}; |
1088 | } | |
1089 | ||
e8c26ab6 | 1090 | static struct swap_info_struct *__swap_info_get(swp_entry_t entry) |
1da177e4 | 1091 | { |
73c34b6a | 1092 | struct swap_info_struct *p; |
eb085574 | 1093 | unsigned long offset; |
1da177e4 LT |
1094 | |
1095 | if (!entry.val) | |
1096 | goto out; | |
eb085574 | 1097 | p = swp_swap_info(entry); |
c10d38cc | 1098 | if (!p) |
1da177e4 | 1099 | goto bad_nofile; |
1da177e4 LT |
1100 | if (!(p->flags & SWP_USED)) |
1101 | goto bad_device; | |
1102 | offset = swp_offset(entry); | |
1103 | if (offset >= p->max) | |
1104 | goto bad_offset; | |
1da177e4 LT |
1105 | return p; |
1106 | ||
1da177e4 | 1107 | bad_offset: |
6a991fc7 | 1108 | pr_err("swap_info_get: %s%08lx\n", Bad_offset, entry.val); |
1da177e4 LT |
1109 | goto out; |
1110 | bad_device: | |
6a991fc7 | 1111 | pr_err("swap_info_get: %s%08lx\n", Unused_file, entry.val); |
1da177e4 LT |
1112 | goto out; |
1113 | bad_nofile: | |
6a991fc7 | 1114 | pr_err("swap_info_get: %s%08lx\n", Bad_file, entry.val); |
1da177e4 LT |
1115 | out: |
1116 | return NULL; | |
886bb7e9 | 1117 | } |
1da177e4 | 1118 | |
e8c26ab6 TC |
1119 | static struct swap_info_struct *_swap_info_get(swp_entry_t entry) |
1120 | { | |
1121 | struct swap_info_struct *p; | |
1122 | ||
1123 | p = __swap_info_get(entry); | |
1124 | if (!p) | |
1125 | goto out; | |
1126 | if (!p->swap_map[swp_offset(entry)]) | |
1127 | goto bad_free; | |
1128 | return p; | |
1129 | ||
1130 | bad_free: | |
1131 | pr_err("swap_info_get: %s%08lx\n", Unused_offset, entry.val); | |
1132 | goto out; | |
1133 | out: | |
1134 | return NULL; | |
1135 | } | |
1136 | ||
235b6217 HY |
1137 | static struct swap_info_struct *swap_info_get(swp_entry_t entry) |
1138 | { | |
1139 | struct swap_info_struct *p; | |
1140 | ||
1141 | p = _swap_info_get(entry); | |
1142 | if (p) | |
1143 | spin_lock(&p->lock); | |
1144 | return p; | |
1145 | } | |
1146 | ||
7c00bafe TC |
1147 | static struct swap_info_struct *swap_info_get_cont(swp_entry_t entry, |
1148 | struct swap_info_struct *q) | |
1149 | { | |
1150 | struct swap_info_struct *p; | |
1151 | ||
1152 | p = _swap_info_get(entry); | |
1153 | ||
1154 | if (p != q) { | |
1155 | if (q != NULL) | |
1156 | spin_unlock(&q->lock); | |
1157 | if (p != NULL) | |
1158 | spin_lock(&p->lock); | |
1159 | } | |
1160 | return p; | |
1161 | } | |
1162 | ||
b32d5f32 HY |
1163 | static unsigned char __swap_entry_free_locked(struct swap_info_struct *p, |
1164 | unsigned long offset, | |
1165 | unsigned char usage) | |
1da177e4 | 1166 | { |
8d69aaee HD |
1167 | unsigned char count; |
1168 | unsigned char has_cache; | |
235b6217 | 1169 | |
253d553b | 1170 | count = p->swap_map[offset]; |
235b6217 | 1171 | |
253d553b HD |
1172 | has_cache = count & SWAP_HAS_CACHE; |
1173 | count &= ~SWAP_HAS_CACHE; | |
355cfa73 | 1174 | |
253d553b | 1175 | if (usage == SWAP_HAS_CACHE) { |
355cfa73 | 1176 | VM_BUG_ON(!has_cache); |
253d553b | 1177 | has_cache = 0; |
aaa46865 HD |
1178 | } else if (count == SWAP_MAP_SHMEM) { |
1179 | /* | |
1180 | * Or we could insist on shmem.c using a special | |
1181 | * swap_shmem_free() and free_shmem_swap_and_cache()... | |
1182 | */ | |
1183 | count = 0; | |
570a335b HD |
1184 | } else if ((count & ~COUNT_CONTINUED) <= SWAP_MAP_MAX) { |
1185 | if (count == COUNT_CONTINUED) { | |
1186 | if (swap_count_continued(p, offset, count)) | |
1187 | count = SWAP_MAP_MAX | COUNT_CONTINUED; | |
1188 | else | |
1189 | count = SWAP_MAP_MAX; | |
1190 | } else | |
1191 | count--; | |
1192 | } | |
253d553b | 1193 | |
253d553b | 1194 | usage = count | has_cache; |
7c00bafe TC |
1195 | p->swap_map[offset] = usage ? : SWAP_HAS_CACHE; |
1196 | ||
b32d5f32 HY |
1197 | return usage; |
1198 | } | |
1199 | ||
eb085574 HY |
1200 | /* |
1201 | * Check whether swap entry is valid in the swap device. If so, | |
1202 | * return pointer to swap_info_struct, and keep the swap entry valid | |
1203 | * via preventing the swap device from being swapoff, until | |
1204 | * put_swap_device() is called. Otherwise return NULL. | |
1205 | * | |
1206 | * The entirety of the RCU read critical section must come before the | |
1207 | * return from or after the call to synchronize_rcu() in | |
1208 | * enable_swap_info() or swapoff(). So if "si->flags & SWP_VALID" is | |
1209 | * true, the si->map, si->cluster_info, etc. must be valid in the | |
1210 | * critical section. | |
1211 | * | |
1212 | * Notice that swapoff or swapoff+swapon can still happen before the | |
1213 | * rcu_read_lock() in get_swap_device() or after the rcu_read_unlock() | |
1214 | * in put_swap_device() if there isn't any other way to prevent | |
1215 | * swapoff, such as page lock, page table lock, etc. The caller must | |
1216 | * be prepared for that. For example, the following situation is | |
1217 | * possible. | |
1218 | * | |
1219 | * CPU1 CPU2 | |
1220 | * do_swap_page() | |
1221 | * ... swapoff+swapon | |
1222 | * __read_swap_cache_async() | |
1223 | * swapcache_prepare() | |
1224 | * __swap_duplicate() | |
1225 | * // check swap_map | |
1226 | * // verify PTE not changed | |
1227 | * | |
1228 | * In __swap_duplicate(), the swap_map need to be checked before | |
1229 | * changing partly because the specified swap entry may be for another | |
1230 | * swap device which has been swapoff. And in do_swap_page(), after | |
1231 | * the page is read from the swap device, the PTE is verified not | |
1232 | * changed with the page table locked to check whether the swap device | |
1233 | * has been swapoff or swapoff+swapon. | |
1234 | */ | |
1235 | struct swap_info_struct *get_swap_device(swp_entry_t entry) | |
1236 | { | |
1237 | struct swap_info_struct *si; | |
1238 | unsigned long offset; | |
1239 | ||
1240 | if (!entry.val) | |
1241 | goto out; | |
1242 | si = swp_swap_info(entry); | |
1243 | if (!si) | |
1244 | goto bad_nofile; | |
1245 | ||
1246 | rcu_read_lock(); | |
1247 | if (!(si->flags & SWP_VALID)) | |
1248 | goto unlock_out; | |
1249 | offset = swp_offset(entry); | |
1250 | if (offset >= si->max) | |
1251 | goto unlock_out; | |
1252 | ||
1253 | return si; | |
1254 | bad_nofile: | |
1255 | pr_err("%s: %s%08lx\n", __func__, Bad_file, entry.val); | |
1256 | out: | |
1257 | return NULL; | |
1258 | unlock_out: | |
1259 | rcu_read_unlock(); | |
1260 | return NULL; | |
1261 | } | |
1262 | ||
b32d5f32 HY |
1263 | static unsigned char __swap_entry_free(struct swap_info_struct *p, |
1264 | swp_entry_t entry, unsigned char usage) | |
1265 | { | |
1266 | struct swap_cluster_info *ci; | |
1267 | unsigned long offset = swp_offset(entry); | |
1268 | ||
1269 | ci = lock_cluster_or_swap_info(p, offset); | |
1270 | usage = __swap_entry_free_locked(p, offset, usage); | |
7c00bafe | 1271 | unlock_cluster_or_swap_info(p, ci); |
10e364da HY |
1272 | if (!usage) |
1273 | free_swap_slot(entry); | |
7c00bafe TC |
1274 | |
1275 | return usage; | |
1276 | } | |
355cfa73 | 1277 | |
7c00bafe TC |
1278 | static void swap_entry_free(struct swap_info_struct *p, swp_entry_t entry) |
1279 | { | |
1280 | struct swap_cluster_info *ci; | |
1281 | unsigned long offset = swp_offset(entry); | |
1282 | unsigned char count; | |
1283 | ||
1284 | ci = lock_cluster(p, offset); | |
1285 | count = p->swap_map[offset]; | |
1286 | VM_BUG_ON(count != SWAP_HAS_CACHE); | |
1287 | p->swap_map[offset] = 0; | |
1288 | dec_cluster_info_page(p, p->cluster_info, offset); | |
235b6217 HY |
1289 | unlock_cluster(ci); |
1290 | ||
38d8b4e6 HY |
1291 | mem_cgroup_uncharge_swap(entry, 1); |
1292 | swap_range_free(p, offset, 1); | |
1da177e4 LT |
1293 | } |
1294 | ||
1295 | /* | |
2de1a7e4 | 1296 | * Caller has made sure that the swap device corresponding to entry |
1da177e4 LT |
1297 | * is still around or has not been recycled. |
1298 | */ | |
1299 | void swap_free(swp_entry_t entry) | |
1300 | { | |
73c34b6a | 1301 | struct swap_info_struct *p; |
1da177e4 | 1302 | |
235b6217 | 1303 | p = _swap_info_get(entry); |
10e364da HY |
1304 | if (p) |
1305 | __swap_entry_free(p, entry, 1); | |
1da177e4 LT |
1306 | } |
1307 | ||
cb4b86ba KH |
1308 | /* |
1309 | * Called after dropping swapcache to decrease refcnt to swap entries. | |
1310 | */ | |
a448f2d0 | 1311 | void put_swap_page(struct page *page, swp_entry_t entry) |
38d8b4e6 HY |
1312 | { |
1313 | unsigned long offset = swp_offset(entry); | |
1314 | unsigned long idx = offset / SWAPFILE_CLUSTER; | |
1315 | struct swap_cluster_info *ci; | |
1316 | struct swap_info_struct *si; | |
1317 | unsigned char *map; | |
a3aea839 HY |
1318 | unsigned int i, free_entries = 0; |
1319 | unsigned char val; | |
a448f2d0 | 1320 | int size = swap_entry_size(hpage_nr_pages(page)); |
fe5266d5 | 1321 | |
a3aea839 | 1322 | si = _swap_info_get(entry); |
38d8b4e6 HY |
1323 | if (!si) |
1324 | return; | |
1325 | ||
c2343d27 | 1326 | ci = lock_cluster_or_swap_info(si, offset); |
a448f2d0 | 1327 | if (size == SWAPFILE_CLUSTER) { |
a448f2d0 HY |
1328 | VM_BUG_ON(!cluster_is_huge(ci)); |
1329 | map = si->swap_map + offset; | |
1330 | for (i = 0; i < SWAPFILE_CLUSTER; i++) { | |
1331 | val = map[i]; | |
1332 | VM_BUG_ON(!(val & SWAP_HAS_CACHE)); | |
1333 | if (val == SWAP_HAS_CACHE) | |
1334 | free_entries++; | |
1335 | } | |
a448f2d0 | 1336 | cluster_clear_huge(ci); |
a448f2d0 | 1337 | if (free_entries == SWAPFILE_CLUSTER) { |
c2343d27 | 1338 | unlock_cluster_or_swap_info(si, ci); |
a448f2d0 | 1339 | spin_lock(&si->lock); |
a448f2d0 HY |
1340 | mem_cgroup_uncharge_swap(entry, SWAPFILE_CLUSTER); |
1341 | swap_free_cluster(si, idx); | |
1342 | spin_unlock(&si->lock); | |
1343 | return; | |
1344 | } | |
1345 | } | |
c2343d27 HY |
1346 | for (i = 0; i < size; i++, entry.val++) { |
1347 | if (!__swap_entry_free_locked(si, offset + i, SWAP_HAS_CACHE)) { | |
1348 | unlock_cluster_or_swap_info(si, ci); | |
1349 | free_swap_slot(entry); | |
1350 | if (i == size - 1) | |
1351 | return; | |
1352 | lock_cluster_or_swap_info(si, offset); | |
a3aea839 HY |
1353 | } |
1354 | } | |
c2343d27 | 1355 | unlock_cluster_or_swap_info(si, ci); |
38d8b4e6 | 1356 | } |
59807685 | 1357 | |
fe5266d5 | 1358 | #ifdef CONFIG_THP_SWAP |
59807685 HY |
1359 | int split_swap_cluster(swp_entry_t entry) |
1360 | { | |
1361 | struct swap_info_struct *si; | |
1362 | struct swap_cluster_info *ci; | |
1363 | unsigned long offset = swp_offset(entry); | |
1364 | ||
1365 | si = _swap_info_get(entry); | |
1366 | if (!si) | |
1367 | return -EBUSY; | |
1368 | ci = lock_cluster(si, offset); | |
1369 | cluster_clear_huge(ci); | |
1370 | unlock_cluster(ci); | |
1371 | return 0; | |
1372 | } | |
fe5266d5 | 1373 | #endif |
38d8b4e6 | 1374 | |
155b5f88 HY |
1375 | static int swp_entry_cmp(const void *ent1, const void *ent2) |
1376 | { | |
1377 | const swp_entry_t *e1 = ent1, *e2 = ent2; | |
1378 | ||
1379 | return (int)swp_type(*e1) - (int)swp_type(*e2); | |
1380 | } | |
1381 | ||
7c00bafe TC |
1382 | void swapcache_free_entries(swp_entry_t *entries, int n) |
1383 | { | |
1384 | struct swap_info_struct *p, *prev; | |
1385 | int i; | |
1386 | ||
1387 | if (n <= 0) | |
1388 | return; | |
1389 | ||
1390 | prev = NULL; | |
1391 | p = NULL; | |
155b5f88 HY |
1392 | |
1393 | /* | |
1394 | * Sort swap entries by swap device, so each lock is only taken once. | |
1395 | * nr_swapfiles isn't absolutely correct, but the overhead of sort() is | |
1396 | * so low that it isn't necessary to optimize further. | |
1397 | */ | |
1398 | if (nr_swapfiles > 1) | |
1399 | sort(entries, n, sizeof(entries[0]), swp_entry_cmp, NULL); | |
7c00bafe TC |
1400 | for (i = 0; i < n; ++i) { |
1401 | p = swap_info_get_cont(entries[i], prev); | |
1402 | if (p) | |
1403 | swap_entry_free(p, entries[i]); | |
7c00bafe TC |
1404 | prev = p; |
1405 | } | |
235b6217 | 1406 | if (p) |
7c00bafe | 1407 | spin_unlock(&p->lock); |
cb4b86ba KH |
1408 | } |
1409 | ||
1da177e4 | 1410 | /* |
c475a8ab | 1411 | * How many references to page are currently swapped out? |
570a335b HD |
1412 | * This does not give an exact answer when swap count is continued, |
1413 | * but does include the high COUNT_CONTINUED flag to allow for that. | |
1da177e4 | 1414 | */ |
bde05d1c | 1415 | int page_swapcount(struct page *page) |
1da177e4 | 1416 | { |
c475a8ab HD |
1417 | int count = 0; |
1418 | struct swap_info_struct *p; | |
235b6217 | 1419 | struct swap_cluster_info *ci; |
1da177e4 | 1420 | swp_entry_t entry; |
235b6217 | 1421 | unsigned long offset; |
1da177e4 | 1422 | |
4c21e2f2 | 1423 | entry.val = page_private(page); |
235b6217 | 1424 | p = _swap_info_get(entry); |
1da177e4 | 1425 | if (p) { |
235b6217 HY |
1426 | offset = swp_offset(entry); |
1427 | ci = lock_cluster_or_swap_info(p, offset); | |
1428 | count = swap_count(p->swap_map[offset]); | |
1429 | unlock_cluster_or_swap_info(p, ci); | |
1da177e4 | 1430 | } |
c475a8ab | 1431 | return count; |
1da177e4 LT |
1432 | } |
1433 | ||
eb085574 | 1434 | int __swap_count(swp_entry_t entry) |
aa8d22a1 | 1435 | { |
eb085574 | 1436 | struct swap_info_struct *si; |
aa8d22a1 | 1437 | pgoff_t offset = swp_offset(entry); |
eb085574 | 1438 | int count = 0; |
aa8d22a1 | 1439 | |
eb085574 HY |
1440 | si = get_swap_device(entry); |
1441 | if (si) { | |
1442 | count = swap_count(si->swap_map[offset]); | |
1443 | put_swap_device(si); | |
1444 | } | |
1445 | return count; | |
aa8d22a1 MK |
1446 | } |
1447 | ||
322b8afe HY |
1448 | static int swap_swapcount(struct swap_info_struct *si, swp_entry_t entry) |
1449 | { | |
1450 | int count = 0; | |
1451 | pgoff_t offset = swp_offset(entry); | |
1452 | struct swap_cluster_info *ci; | |
1453 | ||
1454 | ci = lock_cluster_or_swap_info(si, offset); | |
1455 | count = swap_count(si->swap_map[offset]); | |
1456 | unlock_cluster_or_swap_info(si, ci); | |
1457 | return count; | |
1458 | } | |
1459 | ||
e8c26ab6 TC |
1460 | /* |
1461 | * How many references to @entry are currently swapped out? | |
1462 | * This does not give an exact answer when swap count is continued, | |
1463 | * but does include the high COUNT_CONTINUED flag to allow for that. | |
1464 | */ | |
1465 | int __swp_swapcount(swp_entry_t entry) | |
1466 | { | |
1467 | int count = 0; | |
e8c26ab6 | 1468 | struct swap_info_struct *si; |
e8c26ab6 | 1469 | |
eb085574 HY |
1470 | si = get_swap_device(entry); |
1471 | if (si) { | |
322b8afe | 1472 | count = swap_swapcount(si, entry); |
eb085574 HY |
1473 | put_swap_device(si); |
1474 | } | |
e8c26ab6 TC |
1475 | return count; |
1476 | } | |
1477 | ||
8334b962 MK |
1478 | /* |
1479 | * How many references to @entry are currently swapped out? | |
1480 | * This considers COUNT_CONTINUED so it returns exact answer. | |
1481 | */ | |
1482 | int swp_swapcount(swp_entry_t entry) | |
1483 | { | |
1484 | int count, tmp_count, n; | |
1485 | struct swap_info_struct *p; | |
235b6217 | 1486 | struct swap_cluster_info *ci; |
8334b962 MK |
1487 | struct page *page; |
1488 | pgoff_t offset; | |
1489 | unsigned char *map; | |
1490 | ||
235b6217 | 1491 | p = _swap_info_get(entry); |
8334b962 MK |
1492 | if (!p) |
1493 | return 0; | |
1494 | ||
235b6217 HY |
1495 | offset = swp_offset(entry); |
1496 | ||
1497 | ci = lock_cluster_or_swap_info(p, offset); | |
1498 | ||
1499 | count = swap_count(p->swap_map[offset]); | |
8334b962 MK |
1500 | if (!(count & COUNT_CONTINUED)) |
1501 | goto out; | |
1502 | ||
1503 | count &= ~COUNT_CONTINUED; | |
1504 | n = SWAP_MAP_MAX + 1; | |
1505 | ||
8334b962 MK |
1506 | page = vmalloc_to_page(p->swap_map + offset); |
1507 | offset &= ~PAGE_MASK; | |
1508 | VM_BUG_ON(page_private(page) != SWP_CONTINUED); | |
1509 | ||
1510 | do { | |
a8ae4991 | 1511 | page = list_next_entry(page, lru); |
8334b962 MK |
1512 | map = kmap_atomic(page); |
1513 | tmp_count = map[offset]; | |
1514 | kunmap_atomic(map); | |
1515 | ||
1516 | count += (tmp_count & ~COUNT_CONTINUED) * n; | |
1517 | n *= (SWAP_CONT_MAX + 1); | |
1518 | } while (tmp_count & COUNT_CONTINUED); | |
1519 | out: | |
235b6217 | 1520 | unlock_cluster_or_swap_info(p, ci); |
8334b962 MK |
1521 | return count; |
1522 | } | |
1523 | ||
e0709829 HY |
1524 | static bool swap_page_trans_huge_swapped(struct swap_info_struct *si, |
1525 | swp_entry_t entry) | |
1526 | { | |
1527 | struct swap_cluster_info *ci; | |
1528 | unsigned char *map = si->swap_map; | |
1529 | unsigned long roffset = swp_offset(entry); | |
1530 | unsigned long offset = round_down(roffset, SWAPFILE_CLUSTER); | |
1531 | int i; | |
1532 | bool ret = false; | |
1533 | ||
1534 | ci = lock_cluster_or_swap_info(si, offset); | |
1535 | if (!ci || !cluster_is_huge(ci)) { | |
afa4711e | 1536 | if (swap_count(map[roffset])) |
e0709829 HY |
1537 | ret = true; |
1538 | goto unlock_out; | |
1539 | } | |
1540 | for (i = 0; i < SWAPFILE_CLUSTER; i++) { | |
afa4711e | 1541 | if (swap_count(map[offset + i])) { |
e0709829 HY |
1542 | ret = true; |
1543 | break; | |
1544 | } | |
1545 | } | |
1546 | unlock_out: | |
1547 | unlock_cluster_or_swap_info(si, ci); | |
1548 | return ret; | |
1549 | } | |
1550 | ||
1551 | static bool page_swapped(struct page *page) | |
1552 | { | |
1553 | swp_entry_t entry; | |
1554 | struct swap_info_struct *si; | |
1555 | ||
fe5266d5 | 1556 | if (!IS_ENABLED(CONFIG_THP_SWAP) || likely(!PageTransCompound(page))) |
e0709829 HY |
1557 | return page_swapcount(page) != 0; |
1558 | ||
1559 | page = compound_head(page); | |
1560 | entry.val = page_private(page); | |
1561 | si = _swap_info_get(entry); | |
1562 | if (si) | |
1563 | return swap_page_trans_huge_swapped(si, entry); | |
1564 | return false; | |
1565 | } | |
ba3c4ce6 HY |
1566 | |
1567 | static int page_trans_huge_map_swapcount(struct page *page, int *total_mapcount, | |
1568 | int *total_swapcount) | |
1569 | { | |
1570 | int i, map_swapcount, _total_mapcount, _total_swapcount; | |
1571 | unsigned long offset = 0; | |
1572 | struct swap_info_struct *si; | |
1573 | struct swap_cluster_info *ci = NULL; | |
1574 | unsigned char *map = NULL; | |
1575 | int mapcount, swapcount = 0; | |
1576 | ||
1577 | /* hugetlbfs shouldn't call it */ | |
1578 | VM_BUG_ON_PAGE(PageHuge(page), page); | |
1579 | ||
fe5266d5 HY |
1580 | if (!IS_ENABLED(CONFIG_THP_SWAP) || likely(!PageTransCompound(page))) { |
1581 | mapcount = page_trans_huge_mapcount(page, total_mapcount); | |
ba3c4ce6 HY |
1582 | if (PageSwapCache(page)) |
1583 | swapcount = page_swapcount(page); | |
1584 | if (total_swapcount) | |
1585 | *total_swapcount = swapcount; | |
1586 | return mapcount + swapcount; | |
1587 | } | |
1588 | ||
1589 | page = compound_head(page); | |
1590 | ||
1591 | _total_mapcount = _total_swapcount = map_swapcount = 0; | |
1592 | if (PageSwapCache(page)) { | |
1593 | swp_entry_t entry; | |
1594 | ||
1595 | entry.val = page_private(page); | |
1596 | si = _swap_info_get(entry); | |
1597 | if (si) { | |
1598 | map = si->swap_map; | |
1599 | offset = swp_offset(entry); | |
1600 | } | |
1601 | } | |
1602 | if (map) | |
1603 | ci = lock_cluster(si, offset); | |
1604 | for (i = 0; i < HPAGE_PMD_NR; i++) { | |
1605 | mapcount = atomic_read(&page[i]._mapcount) + 1; | |
1606 | _total_mapcount += mapcount; | |
1607 | if (map) { | |
1608 | swapcount = swap_count(map[offset + i]); | |
1609 | _total_swapcount += swapcount; | |
1610 | } | |
1611 | map_swapcount = max(map_swapcount, mapcount + swapcount); | |
1612 | } | |
1613 | unlock_cluster(ci); | |
1614 | if (PageDoubleMap(page)) { | |
1615 | map_swapcount -= 1; | |
1616 | _total_mapcount -= HPAGE_PMD_NR; | |
1617 | } | |
1618 | mapcount = compound_mapcount(page); | |
1619 | map_swapcount += mapcount; | |
1620 | _total_mapcount += mapcount; | |
1621 | if (total_mapcount) | |
1622 | *total_mapcount = _total_mapcount; | |
1623 | if (total_swapcount) | |
1624 | *total_swapcount = _total_swapcount; | |
1625 | ||
1626 | return map_swapcount; | |
1627 | } | |
e0709829 | 1628 | |
1da177e4 | 1629 | /* |
7b1fe597 HD |
1630 | * We can write to an anon page without COW if there are no other references |
1631 | * to it. And as a side-effect, free up its swap: because the old content | |
1632 | * on disk will never be read, and seeking back there to write new content | |
1633 | * later would only waste time away from clustering. | |
6d0a07ed | 1634 | * |
ba3c4ce6 | 1635 | * NOTE: total_map_swapcount should not be relied upon by the caller if |
6d0a07ed AA |
1636 | * reuse_swap_page() returns false, but it may be always overwritten |
1637 | * (see the other implementation for CONFIG_SWAP=n). | |
1da177e4 | 1638 | */ |
ba3c4ce6 | 1639 | bool reuse_swap_page(struct page *page, int *total_map_swapcount) |
1da177e4 | 1640 | { |
ba3c4ce6 | 1641 | int count, total_mapcount, total_swapcount; |
c475a8ab | 1642 | |
309381fe | 1643 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
5ad64688 | 1644 | if (unlikely(PageKsm(page))) |
6d0a07ed | 1645 | return false; |
ba3c4ce6 HY |
1646 | count = page_trans_huge_map_swapcount(page, &total_mapcount, |
1647 | &total_swapcount); | |
1648 | if (total_map_swapcount) | |
1649 | *total_map_swapcount = total_mapcount + total_swapcount; | |
1650 | if (count == 1 && PageSwapCache(page) && | |
1651 | (likely(!PageTransCompound(page)) || | |
1652 | /* The remaining swap count will be freed soon */ | |
1653 | total_swapcount == page_swapcount(page))) { | |
f0571429 | 1654 | if (!PageWriteback(page)) { |
ba3c4ce6 | 1655 | page = compound_head(page); |
7b1fe597 HD |
1656 | delete_from_swap_cache(page); |
1657 | SetPageDirty(page); | |
f0571429 MK |
1658 | } else { |
1659 | swp_entry_t entry; | |
1660 | struct swap_info_struct *p; | |
1661 | ||
1662 | entry.val = page_private(page); | |
1663 | p = swap_info_get(entry); | |
1664 | if (p->flags & SWP_STABLE_WRITES) { | |
1665 | spin_unlock(&p->lock); | |
1666 | return false; | |
1667 | } | |
1668 | spin_unlock(&p->lock); | |
7b1fe597 HD |
1669 | } |
1670 | } | |
ba3c4ce6 | 1671 | |
5ad64688 | 1672 | return count <= 1; |
1da177e4 LT |
1673 | } |
1674 | ||
1675 | /* | |
a2c43eed HD |
1676 | * If swap is getting full, or if there are no more mappings of this page, |
1677 | * then try_to_free_swap is called to free its swap space. | |
1da177e4 | 1678 | */ |
a2c43eed | 1679 | int try_to_free_swap(struct page *page) |
1da177e4 | 1680 | { |
309381fe | 1681 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
1da177e4 LT |
1682 | |
1683 | if (!PageSwapCache(page)) | |
1684 | return 0; | |
1685 | if (PageWriteback(page)) | |
1686 | return 0; | |
e0709829 | 1687 | if (page_swapped(page)) |
1da177e4 LT |
1688 | return 0; |
1689 | ||
b73d7fce HD |
1690 | /* |
1691 | * Once hibernation has begun to create its image of memory, | |
1692 | * there's a danger that one of the calls to try_to_free_swap() | |
1693 | * - most probably a call from __try_to_reclaim_swap() while | |
1694 | * hibernation is allocating its own swap pages for the image, | |
1695 | * but conceivably even a call from memory reclaim - will free | |
1696 | * the swap from a page which has already been recorded in the | |
1697 | * image as a clean swapcache page, and then reuse its swap for | |
1698 | * another page of the image. On waking from hibernation, the | |
1699 | * original page might be freed under memory pressure, then | |
1700 | * later read back in from swap, now with the wrong data. | |
1701 | * | |
2de1a7e4 | 1702 | * Hibernation suspends storage while it is writing the image |
f90ac398 | 1703 | * to disk so check that here. |
b73d7fce | 1704 | */ |
f90ac398 | 1705 | if (pm_suspended_storage()) |
b73d7fce HD |
1706 | return 0; |
1707 | ||
e0709829 | 1708 | page = compound_head(page); |
a2c43eed HD |
1709 | delete_from_swap_cache(page); |
1710 | SetPageDirty(page); | |
1711 | return 1; | |
68a22394 RR |
1712 | } |
1713 | ||
1da177e4 LT |
1714 | /* |
1715 | * Free the swap entry like above, but also try to | |
1716 | * free the page cache entry if it is the last user. | |
1717 | */ | |
2509ef26 | 1718 | int free_swap_and_cache(swp_entry_t entry) |
1da177e4 | 1719 | { |
2509ef26 | 1720 | struct swap_info_struct *p; |
7c00bafe | 1721 | unsigned char count; |
1da177e4 | 1722 | |
a7420aa5 | 1723 | if (non_swap_entry(entry)) |
2509ef26 | 1724 | return 1; |
0697212a | 1725 | |
7c00bafe | 1726 | p = _swap_info_get(entry); |
1da177e4 | 1727 | if (p) { |
7c00bafe | 1728 | count = __swap_entry_free(p, entry, 1); |
e0709829 | 1729 | if (count == SWAP_HAS_CACHE && |
bcd49e86 HY |
1730 | !swap_page_trans_huge_swapped(p, entry)) |
1731 | __try_to_reclaim_swap(p, swp_offset(entry), | |
1732 | TTRS_UNMAPPED | TTRS_FULL); | |
1da177e4 | 1733 | } |
2509ef26 | 1734 | return p != NULL; |
1da177e4 LT |
1735 | } |
1736 | ||
b0cb1a19 | 1737 | #ifdef CONFIG_HIBERNATION |
f577eb30 | 1738 | /* |
915bae9e | 1739 | * Find the swap type that corresponds to given device (if any). |
f577eb30 | 1740 | * |
915bae9e RW |
1741 | * @offset - number of the PAGE_SIZE-sized block of the device, starting |
1742 | * from 0, in which the swap header is expected to be located. | |
1743 | * | |
1744 | * This is needed for the suspend to disk (aka swsusp). | |
f577eb30 | 1745 | */ |
7bf23687 | 1746 | int swap_type_of(dev_t device, sector_t offset, struct block_device **bdev_p) |
f577eb30 | 1747 | { |
915bae9e | 1748 | struct block_device *bdev = NULL; |
efa90a98 | 1749 | int type; |
f577eb30 | 1750 | |
915bae9e RW |
1751 | if (device) |
1752 | bdev = bdget(device); | |
1753 | ||
f577eb30 | 1754 | spin_lock(&swap_lock); |
efa90a98 HD |
1755 | for (type = 0; type < nr_swapfiles; type++) { |
1756 | struct swap_info_struct *sis = swap_info[type]; | |
f577eb30 | 1757 | |
915bae9e | 1758 | if (!(sis->flags & SWP_WRITEOK)) |
f577eb30 | 1759 | continue; |
b6b5bce3 | 1760 | |
915bae9e | 1761 | if (!bdev) { |
7bf23687 | 1762 | if (bdev_p) |
dddac6a7 | 1763 | *bdev_p = bdgrab(sis->bdev); |
7bf23687 | 1764 | |
6e1819d6 | 1765 | spin_unlock(&swap_lock); |
efa90a98 | 1766 | return type; |
6e1819d6 | 1767 | } |
915bae9e | 1768 | if (bdev == sis->bdev) { |
4efaceb1 | 1769 | struct swap_extent *se = first_se(sis); |
915bae9e | 1770 | |
915bae9e | 1771 | if (se->start_block == offset) { |
7bf23687 | 1772 | if (bdev_p) |
dddac6a7 | 1773 | *bdev_p = bdgrab(sis->bdev); |
7bf23687 | 1774 | |
915bae9e RW |
1775 | spin_unlock(&swap_lock); |
1776 | bdput(bdev); | |
efa90a98 | 1777 | return type; |
915bae9e | 1778 | } |
f577eb30 RW |
1779 | } |
1780 | } | |
1781 | spin_unlock(&swap_lock); | |
915bae9e RW |
1782 | if (bdev) |
1783 | bdput(bdev); | |
1784 | ||
f577eb30 RW |
1785 | return -ENODEV; |
1786 | } | |
1787 | ||
73c34b6a HD |
1788 | /* |
1789 | * Get the (PAGE_SIZE) block corresponding to given offset on the swapdev | |
1790 | * corresponding to given index in swap_info (swap type). | |
1791 | */ | |
1792 | sector_t swapdev_block(int type, pgoff_t offset) | |
1793 | { | |
1794 | struct block_device *bdev; | |
c10d38cc | 1795 | struct swap_info_struct *si = swap_type_to_swap_info(type); |
73c34b6a | 1796 | |
c10d38cc | 1797 | if (!si || !(si->flags & SWP_WRITEOK)) |
73c34b6a | 1798 | return 0; |
d4906e1a | 1799 | return map_swap_entry(swp_entry(type, offset), &bdev); |
73c34b6a HD |
1800 | } |
1801 | ||
f577eb30 RW |
1802 | /* |
1803 | * Return either the total number of swap pages of given type, or the number | |
1804 | * of free pages of that type (depending on @free) | |
1805 | * | |
1806 | * This is needed for software suspend | |
1807 | */ | |
1808 | unsigned int count_swap_pages(int type, int free) | |
1809 | { | |
1810 | unsigned int n = 0; | |
1811 | ||
efa90a98 HD |
1812 | spin_lock(&swap_lock); |
1813 | if ((unsigned int)type < nr_swapfiles) { | |
1814 | struct swap_info_struct *sis = swap_info[type]; | |
1815 | ||
ec8acf20 | 1816 | spin_lock(&sis->lock); |
efa90a98 HD |
1817 | if (sis->flags & SWP_WRITEOK) { |
1818 | n = sis->pages; | |
f577eb30 | 1819 | if (free) |
efa90a98 | 1820 | n -= sis->inuse_pages; |
f577eb30 | 1821 | } |
ec8acf20 | 1822 | spin_unlock(&sis->lock); |
f577eb30 | 1823 | } |
efa90a98 | 1824 | spin_unlock(&swap_lock); |
f577eb30 RW |
1825 | return n; |
1826 | } | |
73c34b6a | 1827 | #endif /* CONFIG_HIBERNATION */ |
f577eb30 | 1828 | |
9f8bdb3f | 1829 | static inline int pte_same_as_swp(pte_t pte, pte_t swp_pte) |
179ef71c | 1830 | { |
9f8bdb3f | 1831 | return pte_same(pte_swp_clear_soft_dirty(pte), swp_pte); |
179ef71c CG |
1832 | } |
1833 | ||
1da177e4 | 1834 | /* |
72866f6f HD |
1835 | * No need to decide whether this PTE shares the swap entry with others, |
1836 | * just let do_wp_page work it out if a write is requested later - to | |
1837 | * force COW, vm_page_prot omits write permission from any private vma. | |
1da177e4 | 1838 | */ |
044d66c1 | 1839 | static int unuse_pte(struct vm_area_struct *vma, pmd_t *pmd, |
1da177e4 LT |
1840 | unsigned long addr, swp_entry_t entry, struct page *page) |
1841 | { | |
9e16b7fb | 1842 | struct page *swapcache; |
72835c86 | 1843 | struct mem_cgroup *memcg; |
044d66c1 HD |
1844 | spinlock_t *ptl; |
1845 | pte_t *pte; | |
1846 | int ret = 1; | |
1847 | ||
9e16b7fb HD |
1848 | swapcache = page; |
1849 | page = ksm_might_need_to_copy(page, vma, addr); | |
1850 | if (unlikely(!page)) | |
1851 | return -ENOMEM; | |
1852 | ||
f627c2f5 KS |
1853 | if (mem_cgroup_try_charge(page, vma->vm_mm, GFP_KERNEL, |
1854 | &memcg, false)) { | |
044d66c1 | 1855 | ret = -ENOMEM; |
85d9fc89 KH |
1856 | goto out_nolock; |
1857 | } | |
044d66c1 HD |
1858 | |
1859 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); | |
9f8bdb3f | 1860 | if (unlikely(!pte_same_as_swp(*pte, swp_entry_to_pte(entry)))) { |
f627c2f5 | 1861 | mem_cgroup_cancel_charge(page, memcg, false); |
044d66c1 HD |
1862 | ret = 0; |
1863 | goto out; | |
1864 | } | |
8a9f3ccd | 1865 | |
b084d435 | 1866 | dec_mm_counter(vma->vm_mm, MM_SWAPENTS); |
d559db08 | 1867 | inc_mm_counter(vma->vm_mm, MM_ANONPAGES); |
1da177e4 LT |
1868 | get_page(page); |
1869 | set_pte_at(vma->vm_mm, addr, pte, | |
1870 | pte_mkold(mk_pte(page, vma->vm_page_prot))); | |
00501b53 | 1871 | if (page == swapcache) { |
d281ee61 | 1872 | page_add_anon_rmap(page, vma, addr, false); |
f627c2f5 | 1873 | mem_cgroup_commit_charge(page, memcg, true, false); |
00501b53 | 1874 | } else { /* ksm created a completely new copy */ |
d281ee61 | 1875 | page_add_new_anon_rmap(page, vma, addr, false); |
f627c2f5 | 1876 | mem_cgroup_commit_charge(page, memcg, false, false); |
00501b53 JW |
1877 | lru_cache_add_active_or_unevictable(page, vma); |
1878 | } | |
1da177e4 LT |
1879 | swap_free(entry); |
1880 | /* | |
1881 | * Move the page to the active list so it is not | |
1882 | * immediately swapped out again after swapon. | |
1883 | */ | |
1884 | activate_page(page); | |
044d66c1 HD |
1885 | out: |
1886 | pte_unmap_unlock(pte, ptl); | |
85d9fc89 | 1887 | out_nolock: |
9e16b7fb HD |
1888 | if (page != swapcache) { |
1889 | unlock_page(page); | |
1890 | put_page(page); | |
1891 | } | |
044d66c1 | 1892 | return ret; |
1da177e4 LT |
1893 | } |
1894 | ||
1895 | static int unuse_pte_range(struct vm_area_struct *vma, pmd_t *pmd, | |
b56a2d8a VRP |
1896 | unsigned long addr, unsigned long end, |
1897 | unsigned int type, bool frontswap, | |
1898 | unsigned long *fs_pages_to_unuse) | |
1da177e4 | 1899 | { |
b56a2d8a VRP |
1900 | struct page *page; |
1901 | swp_entry_t entry; | |
705e87c0 | 1902 | pte_t *pte; |
b56a2d8a VRP |
1903 | struct swap_info_struct *si; |
1904 | unsigned long offset; | |
8a9f3ccd | 1905 | int ret = 0; |
b56a2d8a | 1906 | volatile unsigned char *swap_map; |
1da177e4 | 1907 | |
b56a2d8a | 1908 | si = swap_info[type]; |
044d66c1 | 1909 | pte = pte_offset_map(pmd, addr); |
1da177e4 | 1910 | do { |
b56a2d8a VRP |
1911 | struct vm_fault vmf; |
1912 | ||
1913 | if (!is_swap_pte(*pte)) | |
1914 | continue; | |
1915 | ||
1916 | entry = pte_to_swp_entry(*pte); | |
1917 | if (swp_type(entry) != type) | |
1918 | continue; | |
1919 | ||
1920 | offset = swp_offset(entry); | |
1921 | if (frontswap && !frontswap_test(si, offset)) | |
1922 | continue; | |
1923 | ||
1924 | pte_unmap(pte); | |
1925 | swap_map = &si->swap_map[offset]; | |
ebc5951e AR |
1926 | page = lookup_swap_cache(entry, vma, addr); |
1927 | if (!page) { | |
1928 | vmf.vma = vma; | |
1929 | vmf.address = addr; | |
1930 | vmf.pmd = pmd; | |
1931 | page = swapin_readahead(entry, GFP_HIGHUSER_MOVABLE, | |
1932 | &vmf); | |
1933 | } | |
b56a2d8a VRP |
1934 | if (!page) { |
1935 | if (*swap_map == 0 || *swap_map == SWAP_MAP_BAD) | |
1936 | goto try_next; | |
1937 | return -ENOMEM; | |
1938 | } | |
1939 | ||
1940 | lock_page(page); | |
1941 | wait_on_page_writeback(page); | |
1942 | ret = unuse_pte(vma, pmd, addr, entry, page); | |
1943 | if (ret < 0) { | |
1944 | unlock_page(page); | |
1945 | put_page(page); | |
1946 | goto out; | |
1947 | } | |
1948 | ||
1949 | try_to_free_swap(page); | |
1950 | unlock_page(page); | |
1951 | put_page(page); | |
1952 | ||
1953 | if (*fs_pages_to_unuse && !--(*fs_pages_to_unuse)) { | |
1954 | ret = FRONTSWAP_PAGES_UNUSED; | |
1955 | goto out; | |
1da177e4 | 1956 | } |
b56a2d8a VRP |
1957 | try_next: |
1958 | pte = pte_offset_map(pmd, addr); | |
1da177e4 | 1959 | } while (pte++, addr += PAGE_SIZE, addr != end); |
044d66c1 | 1960 | pte_unmap(pte - 1); |
b56a2d8a VRP |
1961 | |
1962 | ret = 0; | |
044d66c1 | 1963 | out: |
8a9f3ccd | 1964 | return ret; |
1da177e4 LT |
1965 | } |
1966 | ||
1967 | static inline int unuse_pmd_range(struct vm_area_struct *vma, pud_t *pud, | |
1968 | unsigned long addr, unsigned long end, | |
b56a2d8a VRP |
1969 | unsigned int type, bool frontswap, |
1970 | unsigned long *fs_pages_to_unuse) | |
1da177e4 LT |
1971 | { |
1972 | pmd_t *pmd; | |
1973 | unsigned long next; | |
8a9f3ccd | 1974 | int ret; |
1da177e4 LT |
1975 | |
1976 | pmd = pmd_offset(pud, addr); | |
1977 | do { | |
dc644a07 | 1978 | cond_resched(); |
1da177e4 | 1979 | next = pmd_addr_end(addr, end); |
1a5a9906 | 1980 | if (pmd_none_or_trans_huge_or_clear_bad(pmd)) |
1da177e4 | 1981 | continue; |
b56a2d8a VRP |
1982 | ret = unuse_pte_range(vma, pmd, addr, next, type, |
1983 | frontswap, fs_pages_to_unuse); | |
8a9f3ccd BS |
1984 | if (ret) |
1985 | return ret; | |
1da177e4 LT |
1986 | } while (pmd++, addr = next, addr != end); |
1987 | return 0; | |
1988 | } | |
1989 | ||
c2febafc | 1990 | static inline int unuse_pud_range(struct vm_area_struct *vma, p4d_t *p4d, |
1da177e4 | 1991 | unsigned long addr, unsigned long end, |
b56a2d8a VRP |
1992 | unsigned int type, bool frontswap, |
1993 | unsigned long *fs_pages_to_unuse) | |
1da177e4 LT |
1994 | { |
1995 | pud_t *pud; | |
1996 | unsigned long next; | |
8a9f3ccd | 1997 | int ret; |
1da177e4 | 1998 | |
c2febafc | 1999 | pud = pud_offset(p4d, addr); |
1da177e4 LT |
2000 | do { |
2001 | next = pud_addr_end(addr, end); | |
2002 | if (pud_none_or_clear_bad(pud)) | |
2003 | continue; | |
b56a2d8a VRP |
2004 | ret = unuse_pmd_range(vma, pud, addr, next, type, |
2005 | frontswap, fs_pages_to_unuse); | |
8a9f3ccd BS |
2006 | if (ret) |
2007 | return ret; | |
1da177e4 LT |
2008 | } while (pud++, addr = next, addr != end); |
2009 | return 0; | |
2010 | } | |
2011 | ||
c2febafc KS |
2012 | static inline int unuse_p4d_range(struct vm_area_struct *vma, pgd_t *pgd, |
2013 | unsigned long addr, unsigned long end, | |
b56a2d8a VRP |
2014 | unsigned int type, bool frontswap, |
2015 | unsigned long *fs_pages_to_unuse) | |
c2febafc KS |
2016 | { |
2017 | p4d_t *p4d; | |
2018 | unsigned long next; | |
2019 | int ret; | |
2020 | ||
2021 | p4d = p4d_offset(pgd, addr); | |
2022 | do { | |
2023 | next = p4d_addr_end(addr, end); | |
2024 | if (p4d_none_or_clear_bad(p4d)) | |
2025 | continue; | |
b56a2d8a VRP |
2026 | ret = unuse_pud_range(vma, p4d, addr, next, type, |
2027 | frontswap, fs_pages_to_unuse); | |
c2febafc KS |
2028 | if (ret) |
2029 | return ret; | |
2030 | } while (p4d++, addr = next, addr != end); | |
2031 | return 0; | |
2032 | } | |
2033 | ||
b56a2d8a VRP |
2034 | static int unuse_vma(struct vm_area_struct *vma, unsigned int type, |
2035 | bool frontswap, unsigned long *fs_pages_to_unuse) | |
1da177e4 LT |
2036 | { |
2037 | pgd_t *pgd; | |
2038 | unsigned long addr, end, next; | |
8a9f3ccd | 2039 | int ret; |
1da177e4 | 2040 | |
b56a2d8a VRP |
2041 | addr = vma->vm_start; |
2042 | end = vma->vm_end; | |
1da177e4 LT |
2043 | |
2044 | pgd = pgd_offset(vma->vm_mm, addr); | |
2045 | do { | |
2046 | next = pgd_addr_end(addr, end); | |
2047 | if (pgd_none_or_clear_bad(pgd)) | |
2048 | continue; | |
b56a2d8a VRP |
2049 | ret = unuse_p4d_range(vma, pgd, addr, next, type, |
2050 | frontswap, fs_pages_to_unuse); | |
8a9f3ccd BS |
2051 | if (ret) |
2052 | return ret; | |
1da177e4 LT |
2053 | } while (pgd++, addr = next, addr != end); |
2054 | return 0; | |
2055 | } | |
2056 | ||
b56a2d8a VRP |
2057 | static int unuse_mm(struct mm_struct *mm, unsigned int type, |
2058 | bool frontswap, unsigned long *fs_pages_to_unuse) | |
1da177e4 LT |
2059 | { |
2060 | struct vm_area_struct *vma; | |
8a9f3ccd | 2061 | int ret = 0; |
1da177e4 | 2062 | |
b56a2d8a | 2063 | down_read(&mm->mmap_sem); |
1da177e4 | 2064 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
b56a2d8a VRP |
2065 | if (vma->anon_vma) { |
2066 | ret = unuse_vma(vma, type, frontswap, | |
2067 | fs_pages_to_unuse); | |
2068 | if (ret) | |
2069 | break; | |
2070 | } | |
dc644a07 | 2071 | cond_resched(); |
1da177e4 | 2072 | } |
1da177e4 | 2073 | up_read(&mm->mmap_sem); |
b56a2d8a | 2074 | return ret; |
1da177e4 LT |
2075 | } |
2076 | ||
2077 | /* | |
38b5faf4 | 2078 | * Scan swap_map (or frontswap_map if frontswap parameter is true) |
b56a2d8a VRP |
2079 | * from current position to next entry still in use. Return 0 |
2080 | * if there are no inuse entries after prev till end of the map. | |
1da177e4 | 2081 | */ |
6eb396dc | 2082 | static unsigned int find_next_to_unuse(struct swap_info_struct *si, |
38b5faf4 | 2083 | unsigned int prev, bool frontswap) |
1da177e4 | 2084 | { |
b56a2d8a | 2085 | unsigned int i; |
8d69aaee | 2086 | unsigned char count; |
1da177e4 LT |
2087 | |
2088 | /* | |
5d337b91 | 2089 | * No need for swap_lock here: we're just looking |
1da177e4 LT |
2090 | * for whether an entry is in use, not modifying it; false |
2091 | * hits are okay, and sys_swapoff() has already prevented new | |
5d337b91 | 2092 | * allocations from this area (while holding swap_lock). |
1da177e4 | 2093 | */ |
b56a2d8a | 2094 | for (i = prev + 1; i < si->max; i++) { |
4db0c3c2 | 2095 | count = READ_ONCE(si->swap_map[i]); |
355cfa73 | 2096 | if (count && swap_count(count) != SWAP_MAP_BAD) |
dc644a07 HD |
2097 | if (!frontswap || frontswap_test(si, i)) |
2098 | break; | |
2099 | if ((i % LATENCY_LIMIT) == 0) | |
2100 | cond_resched(); | |
1da177e4 | 2101 | } |
b56a2d8a VRP |
2102 | |
2103 | if (i == si->max) | |
2104 | i = 0; | |
2105 | ||
1da177e4 LT |
2106 | return i; |
2107 | } | |
2108 | ||
2109 | /* | |
b56a2d8a | 2110 | * If the boolean frontswap is true, only unuse pages_to_unuse pages; |
38b5faf4 | 2111 | * pages_to_unuse==0 means all pages; ignored if frontswap is false |
1da177e4 | 2112 | */ |
38b5faf4 DM |
2113 | int try_to_unuse(unsigned int type, bool frontswap, |
2114 | unsigned long pages_to_unuse) | |
1da177e4 | 2115 | { |
b56a2d8a VRP |
2116 | struct mm_struct *prev_mm; |
2117 | struct mm_struct *mm; | |
2118 | struct list_head *p; | |
2119 | int retval = 0; | |
efa90a98 | 2120 | struct swap_info_struct *si = swap_info[type]; |
1da177e4 LT |
2121 | struct page *page; |
2122 | swp_entry_t entry; | |
b56a2d8a | 2123 | unsigned int i; |
1da177e4 | 2124 | |
21820948 | 2125 | if (!READ_ONCE(si->inuse_pages)) |
b56a2d8a | 2126 | return 0; |
1da177e4 | 2127 | |
b56a2d8a VRP |
2128 | if (!frontswap) |
2129 | pages_to_unuse = 0; | |
2130 | ||
2131 | retry: | |
2132 | retval = shmem_unuse(type, frontswap, &pages_to_unuse); | |
2133 | if (retval) | |
2134 | goto out; | |
2135 | ||
2136 | prev_mm = &init_mm; | |
2137 | mmget(prev_mm); | |
2138 | ||
2139 | spin_lock(&mmlist_lock); | |
2140 | p = &init_mm.mmlist; | |
21820948 | 2141 | while (READ_ONCE(si->inuse_pages) && |
64165b1a HD |
2142 | !signal_pending(current) && |
2143 | (p = p->next) != &init_mm.mmlist) { | |
1da177e4 | 2144 | |
b56a2d8a VRP |
2145 | mm = list_entry(p, struct mm_struct, mmlist); |
2146 | if (!mmget_not_zero(mm)) | |
2147 | continue; | |
2148 | spin_unlock(&mmlist_lock); | |
2149 | mmput(prev_mm); | |
2150 | prev_mm = mm; | |
2151 | retval = unuse_mm(mm, type, frontswap, &pages_to_unuse); | |
1da177e4 | 2152 | |
b56a2d8a VRP |
2153 | if (retval) { |
2154 | mmput(prev_mm); | |
2155 | goto out; | |
1da177e4 LT |
2156 | } |
2157 | ||
2158 | /* | |
b56a2d8a VRP |
2159 | * Make sure that we aren't completely killing |
2160 | * interactive performance. | |
1da177e4 | 2161 | */ |
b56a2d8a VRP |
2162 | cond_resched(); |
2163 | spin_lock(&mmlist_lock); | |
2164 | } | |
2165 | spin_unlock(&mmlist_lock); | |
1da177e4 | 2166 | |
b56a2d8a | 2167 | mmput(prev_mm); |
1da177e4 | 2168 | |
b56a2d8a | 2169 | i = 0; |
21820948 | 2170 | while (READ_ONCE(si->inuse_pages) && |
64165b1a HD |
2171 | !signal_pending(current) && |
2172 | (i = find_next_to_unuse(si, i, frontswap)) != 0) { | |
1da177e4 | 2173 | |
b56a2d8a VRP |
2174 | entry = swp_entry(type, i); |
2175 | page = find_get_page(swap_address_space(entry), i); | |
2176 | if (!page) | |
2177 | continue; | |
68bdc8d6 HD |
2178 | |
2179 | /* | |
2180 | * It is conceivable that a racing task removed this page from | |
b56a2d8a VRP |
2181 | * swap cache just before we acquired the page lock. The page |
2182 | * might even be back in swap cache on another swap area. But | |
2183 | * that is okay, try_to_free_swap() only removes stale pages. | |
1da177e4 | 2184 | */ |
b56a2d8a VRP |
2185 | lock_page(page); |
2186 | wait_on_page_writeback(page); | |
2187 | try_to_free_swap(page); | |
1da177e4 | 2188 | unlock_page(page); |
09cbfeaf | 2189 | put_page(page); |
1da177e4 LT |
2190 | |
2191 | /* | |
b56a2d8a VRP |
2192 | * For frontswap, we just need to unuse pages_to_unuse, if |
2193 | * it was specified. Need not check frontswap again here as | |
2194 | * we already zeroed out pages_to_unuse if not frontswap. | |
1da177e4 | 2195 | */ |
b56a2d8a VRP |
2196 | if (pages_to_unuse && --pages_to_unuse == 0) |
2197 | goto out; | |
1da177e4 LT |
2198 | } |
2199 | ||
b56a2d8a VRP |
2200 | /* |
2201 | * Lets check again to see if there are still swap entries in the map. | |
2202 | * If yes, we would need to do retry the unuse logic again. | |
2203 | * Under global memory pressure, swap entries can be reinserted back | |
2204 | * into process space after the mmlist loop above passes over them. | |
dd862deb | 2205 | * |
af53d3e9 HD |
2206 | * Limit the number of retries? No: when mmget_not_zero() above fails, |
2207 | * that mm is likely to be freeing swap from exit_mmap(), which proceeds | |
2208 | * at its own independent pace; and even shmem_writepage() could have | |
2209 | * been preempted after get_swap_page(), temporarily hiding that swap. | |
2210 | * It's easy and robust (though cpu-intensive) just to keep retrying. | |
b56a2d8a | 2211 | */ |
21820948 | 2212 | if (READ_ONCE(si->inuse_pages)) { |
64165b1a HD |
2213 | if (!signal_pending(current)) |
2214 | goto retry; | |
2215 | retval = -EINTR; | |
2216 | } | |
b56a2d8a VRP |
2217 | out: |
2218 | return (retval == FRONTSWAP_PAGES_UNUSED) ? 0 : retval; | |
1da177e4 LT |
2219 | } |
2220 | ||
2221 | /* | |
5d337b91 HD |
2222 | * After a successful try_to_unuse, if no swap is now in use, we know |
2223 | * we can empty the mmlist. swap_lock must be held on entry and exit. | |
2224 | * Note that mmlist_lock nests inside swap_lock, and an mm must be | |
1da177e4 LT |
2225 | * added to the mmlist just after page_duplicate - before would be racy. |
2226 | */ | |
2227 | static void drain_mmlist(void) | |
2228 | { | |
2229 | struct list_head *p, *next; | |
efa90a98 | 2230 | unsigned int type; |
1da177e4 | 2231 | |
efa90a98 HD |
2232 | for (type = 0; type < nr_swapfiles; type++) |
2233 | if (swap_info[type]->inuse_pages) | |
1da177e4 LT |
2234 | return; |
2235 | spin_lock(&mmlist_lock); | |
2236 | list_for_each_safe(p, next, &init_mm.mmlist) | |
2237 | list_del_init(p); | |
2238 | spin_unlock(&mmlist_lock); | |
2239 | } | |
2240 | ||
2241 | /* | |
2242 | * Use this swapdev's extent info to locate the (PAGE_SIZE) block which | |
d4906e1a LS |
2243 | * corresponds to page offset for the specified swap entry. |
2244 | * Note that the type of this function is sector_t, but it returns page offset | |
2245 | * into the bdev, not sector offset. | |
1da177e4 | 2246 | */ |
d4906e1a | 2247 | static sector_t map_swap_entry(swp_entry_t entry, struct block_device **bdev) |
1da177e4 | 2248 | { |
f29ad6a9 | 2249 | struct swap_info_struct *sis; |
f29ad6a9 HD |
2250 | struct swap_extent *se; |
2251 | pgoff_t offset; | |
2252 | ||
c10d38cc | 2253 | sis = swp_swap_info(entry); |
f29ad6a9 HD |
2254 | *bdev = sis->bdev; |
2255 | ||
2256 | offset = swp_offset(entry); | |
4efaceb1 AL |
2257 | se = offset_to_swap_extent(sis, offset); |
2258 | return se->start_block + (offset - se->start_page); | |
1da177e4 LT |
2259 | } |
2260 | ||
d4906e1a LS |
2261 | /* |
2262 | * Returns the page offset into bdev for the specified page's swap entry. | |
2263 | */ | |
2264 | sector_t map_swap_page(struct page *page, struct block_device **bdev) | |
2265 | { | |
2266 | swp_entry_t entry; | |
2267 | entry.val = page_private(page); | |
2268 | return map_swap_entry(entry, bdev); | |
2269 | } | |
2270 | ||
1da177e4 LT |
2271 | /* |
2272 | * Free all of a swapdev's extent information | |
2273 | */ | |
2274 | static void destroy_swap_extents(struct swap_info_struct *sis) | |
2275 | { | |
4efaceb1 AL |
2276 | while (!RB_EMPTY_ROOT(&sis->swap_extent_root)) { |
2277 | struct rb_node *rb = sis->swap_extent_root.rb_node; | |
2278 | struct swap_extent *se = rb_entry(rb, struct swap_extent, rb_node); | |
1da177e4 | 2279 | |
4efaceb1 | 2280 | rb_erase(rb, &sis->swap_extent_root); |
1da177e4 LT |
2281 | kfree(se); |
2282 | } | |
62c230bc | 2283 | |
bc4ae27d | 2284 | if (sis->flags & SWP_ACTIVATED) { |
62c230bc MG |
2285 | struct file *swap_file = sis->swap_file; |
2286 | struct address_space *mapping = swap_file->f_mapping; | |
2287 | ||
bc4ae27d OS |
2288 | sis->flags &= ~SWP_ACTIVATED; |
2289 | if (mapping->a_ops->swap_deactivate) | |
2290 | mapping->a_ops->swap_deactivate(swap_file); | |
62c230bc | 2291 | } |
1da177e4 LT |
2292 | } |
2293 | ||
2294 | /* | |
2295 | * Add a block range (and the corresponding page range) into this swapdev's | |
4efaceb1 | 2296 | * extent tree. |
1da177e4 | 2297 | * |
11d31886 | 2298 | * This function rather assumes that it is called in ascending page order. |
1da177e4 | 2299 | */ |
a509bc1a | 2300 | int |
1da177e4 LT |
2301 | add_swap_extent(struct swap_info_struct *sis, unsigned long start_page, |
2302 | unsigned long nr_pages, sector_t start_block) | |
2303 | { | |
4efaceb1 | 2304 | struct rb_node **link = &sis->swap_extent_root.rb_node, *parent = NULL; |
1da177e4 LT |
2305 | struct swap_extent *se; |
2306 | struct swap_extent *new_se; | |
4efaceb1 AL |
2307 | |
2308 | /* | |
2309 | * place the new node at the right most since the | |
2310 | * function is called in ascending page order. | |
2311 | */ | |
2312 | while (*link) { | |
2313 | parent = *link; | |
2314 | link = &parent->rb_right; | |
2315 | } | |
2316 | ||
2317 | if (parent) { | |
2318 | se = rb_entry(parent, struct swap_extent, rb_node); | |
11d31886 HD |
2319 | BUG_ON(se->start_page + se->nr_pages != start_page); |
2320 | if (se->start_block + se->nr_pages == start_block) { | |
1da177e4 LT |
2321 | /* Merge it */ |
2322 | se->nr_pages += nr_pages; | |
2323 | return 0; | |
2324 | } | |
1da177e4 LT |
2325 | } |
2326 | ||
4efaceb1 | 2327 | /* No merge, insert a new extent. */ |
1da177e4 LT |
2328 | new_se = kmalloc(sizeof(*se), GFP_KERNEL); |
2329 | if (new_se == NULL) | |
2330 | return -ENOMEM; | |
2331 | new_se->start_page = start_page; | |
2332 | new_se->nr_pages = nr_pages; | |
2333 | new_se->start_block = start_block; | |
2334 | ||
4efaceb1 AL |
2335 | rb_link_node(&new_se->rb_node, parent, link); |
2336 | rb_insert_color(&new_se->rb_node, &sis->swap_extent_root); | |
53092a74 | 2337 | return 1; |
1da177e4 | 2338 | } |
aa8aa8a3 | 2339 | EXPORT_SYMBOL_GPL(add_swap_extent); |
1da177e4 LT |
2340 | |
2341 | /* | |
2342 | * A `swap extent' is a simple thing which maps a contiguous range of pages | |
2343 | * onto a contiguous range of disk blocks. An ordered list of swap extents | |
2344 | * is built at swapon time and is then used at swap_writepage/swap_readpage | |
2345 | * time for locating where on disk a page belongs. | |
2346 | * | |
2347 | * If the swapfile is an S_ISBLK block device, a single extent is installed. | |
2348 | * This is done so that the main operating code can treat S_ISBLK and S_ISREG | |
2349 | * swap files identically. | |
2350 | * | |
2351 | * Whether the swapdev is an S_ISREG file or an S_ISBLK blockdev, the swap | |
2352 | * extent list operates in PAGE_SIZE disk blocks. Both S_ISREG and S_ISBLK | |
2353 | * swapfiles are handled *identically* after swapon time. | |
2354 | * | |
2355 | * For S_ISREG swapfiles, setup_swap_extents() will walk all the file's blocks | |
2356 | * and will parse them into an ordered extent list, in PAGE_SIZE chunks. If | |
2357 | * some stray blocks are found which do not fall within the PAGE_SIZE alignment | |
2358 | * requirements, they are simply tossed out - we will never use those blocks | |
2359 | * for swapping. | |
2360 | * | |
1638045c DW |
2361 | * For all swap devices we set S_SWAPFILE across the life of the swapon. This |
2362 | * prevents users from writing to the swap device, which will corrupt memory. | |
1da177e4 LT |
2363 | * |
2364 | * The amount of disk space which a single swap extent represents varies. | |
2365 | * Typically it is in the 1-4 megabyte range. So we can have hundreds of | |
2366 | * extents in the list. To avoid much list walking, we cache the previous | |
2367 | * search location in `curr_swap_extent', and start new searches from there. | |
2368 | * This is extremely effective. The average number of iterations in | |
2369 | * map_swap_page() has been measured at about 0.3 per page. - akpm. | |
2370 | */ | |
53092a74 | 2371 | static int setup_swap_extents(struct swap_info_struct *sis, sector_t *span) |
1da177e4 | 2372 | { |
62c230bc MG |
2373 | struct file *swap_file = sis->swap_file; |
2374 | struct address_space *mapping = swap_file->f_mapping; | |
2375 | struct inode *inode = mapping->host; | |
1da177e4 LT |
2376 | int ret; |
2377 | ||
1da177e4 LT |
2378 | if (S_ISBLK(inode->i_mode)) { |
2379 | ret = add_swap_extent(sis, 0, sis->max, 0); | |
53092a74 | 2380 | *span = sis->pages; |
a509bc1a | 2381 | return ret; |
1da177e4 LT |
2382 | } |
2383 | ||
62c230bc | 2384 | if (mapping->a_ops->swap_activate) { |
a509bc1a | 2385 | ret = mapping->a_ops->swap_activate(sis, swap_file, span); |
bc4ae27d OS |
2386 | if (ret >= 0) |
2387 | sis->flags |= SWP_ACTIVATED; | |
62c230bc | 2388 | if (!ret) { |
bc4ae27d | 2389 | sis->flags |= SWP_FS; |
62c230bc MG |
2390 | ret = add_swap_extent(sis, 0, sis->max, 0); |
2391 | *span = sis->pages; | |
2392 | } | |
a509bc1a | 2393 | return ret; |
62c230bc MG |
2394 | } |
2395 | ||
a509bc1a | 2396 | return generic_swapfile_activate(sis, swap_file, span); |
1da177e4 LT |
2397 | } |
2398 | ||
a2468cc9 AL |
2399 | static int swap_node(struct swap_info_struct *p) |
2400 | { | |
2401 | struct block_device *bdev; | |
2402 | ||
2403 | if (p->bdev) | |
2404 | bdev = p->bdev; | |
2405 | else | |
2406 | bdev = p->swap_file->f_inode->i_sb->s_bdev; | |
2407 | ||
2408 | return bdev ? bdev->bd_disk->node_id : NUMA_NO_NODE; | |
2409 | } | |
2410 | ||
eb085574 HY |
2411 | static void setup_swap_info(struct swap_info_struct *p, int prio, |
2412 | unsigned char *swap_map, | |
2413 | struct swap_cluster_info *cluster_info) | |
40531542 | 2414 | { |
a2468cc9 AL |
2415 | int i; |
2416 | ||
40531542 CEB |
2417 | if (prio >= 0) |
2418 | p->prio = prio; | |
2419 | else | |
2420 | p->prio = --least_priority; | |
18ab4d4c DS |
2421 | /* |
2422 | * the plist prio is negated because plist ordering is | |
2423 | * low-to-high, while swap ordering is high-to-low | |
2424 | */ | |
2425 | p->list.prio = -p->prio; | |
a2468cc9 AL |
2426 | for_each_node(i) { |
2427 | if (p->prio >= 0) | |
2428 | p->avail_lists[i].prio = -p->prio; | |
2429 | else { | |
2430 | if (swap_node(p) == i) | |
2431 | p->avail_lists[i].prio = 1; | |
2432 | else | |
2433 | p->avail_lists[i].prio = -p->prio; | |
2434 | } | |
2435 | } | |
40531542 | 2436 | p->swap_map = swap_map; |
2a8f9449 | 2437 | p->cluster_info = cluster_info; |
eb085574 HY |
2438 | } |
2439 | ||
2440 | static void _enable_swap_info(struct swap_info_struct *p) | |
2441 | { | |
2442 | p->flags |= SWP_WRITEOK | SWP_VALID; | |
ec8acf20 | 2443 | atomic_long_add(p->pages, &nr_swap_pages); |
40531542 CEB |
2444 | total_swap_pages += p->pages; |
2445 | ||
adfab836 | 2446 | assert_spin_locked(&swap_lock); |
adfab836 | 2447 | /* |
18ab4d4c DS |
2448 | * both lists are plists, and thus priority ordered. |
2449 | * swap_active_head needs to be priority ordered for swapoff(), | |
2450 | * which on removal of any swap_info_struct with an auto-assigned | |
2451 | * (i.e. negative) priority increments the auto-assigned priority | |
2452 | * of any lower-priority swap_info_structs. | |
2453 | * swap_avail_head needs to be priority ordered for get_swap_page(), | |
2454 | * which allocates swap pages from the highest available priority | |
2455 | * swap_info_struct. | |
adfab836 | 2456 | */ |
18ab4d4c | 2457 | plist_add(&p->list, &swap_active_head); |
a2468cc9 | 2458 | add_to_avail_list(p); |
cf0cac0a CEB |
2459 | } |
2460 | ||
2461 | static void enable_swap_info(struct swap_info_struct *p, int prio, | |
2462 | unsigned char *swap_map, | |
2a8f9449 | 2463 | struct swap_cluster_info *cluster_info, |
cf0cac0a CEB |
2464 | unsigned long *frontswap_map) |
2465 | { | |
4f89849d | 2466 | frontswap_init(p->type, frontswap_map); |
cf0cac0a | 2467 | spin_lock(&swap_lock); |
ec8acf20 | 2468 | spin_lock(&p->lock); |
eb085574 HY |
2469 | setup_swap_info(p, prio, swap_map, cluster_info); |
2470 | spin_unlock(&p->lock); | |
2471 | spin_unlock(&swap_lock); | |
2472 | /* | |
2473 | * Guarantee swap_map, cluster_info, etc. fields are valid | |
2474 | * between get/put_swap_device() if SWP_VALID bit is set | |
2475 | */ | |
2476 | synchronize_rcu(); | |
2477 | spin_lock(&swap_lock); | |
2478 | spin_lock(&p->lock); | |
2479 | _enable_swap_info(p); | |
ec8acf20 | 2480 | spin_unlock(&p->lock); |
cf0cac0a CEB |
2481 | spin_unlock(&swap_lock); |
2482 | } | |
2483 | ||
2484 | static void reinsert_swap_info(struct swap_info_struct *p) | |
2485 | { | |
2486 | spin_lock(&swap_lock); | |
ec8acf20 | 2487 | spin_lock(&p->lock); |
eb085574 HY |
2488 | setup_swap_info(p, p->prio, p->swap_map, p->cluster_info); |
2489 | _enable_swap_info(p); | |
ec8acf20 | 2490 | spin_unlock(&p->lock); |
40531542 CEB |
2491 | spin_unlock(&swap_lock); |
2492 | } | |
2493 | ||
67afa38e TC |
2494 | bool has_usable_swap(void) |
2495 | { | |
2496 | bool ret = true; | |
2497 | ||
2498 | spin_lock(&swap_lock); | |
2499 | if (plist_head_empty(&swap_active_head)) | |
2500 | ret = false; | |
2501 | spin_unlock(&swap_lock); | |
2502 | return ret; | |
2503 | } | |
2504 | ||
c4ea37c2 | 2505 | SYSCALL_DEFINE1(swapoff, const char __user *, specialfile) |
1da177e4 | 2506 | { |
73c34b6a | 2507 | struct swap_info_struct *p = NULL; |
8d69aaee | 2508 | unsigned char *swap_map; |
2a8f9449 | 2509 | struct swap_cluster_info *cluster_info; |
4f89849d | 2510 | unsigned long *frontswap_map; |
1da177e4 LT |
2511 | struct file *swap_file, *victim; |
2512 | struct address_space *mapping; | |
2513 | struct inode *inode; | |
91a27b2a | 2514 | struct filename *pathname; |
adfab836 | 2515 | int err, found = 0; |
5b808a23 | 2516 | unsigned int old_block_size; |
886bb7e9 | 2517 | |
1da177e4 LT |
2518 | if (!capable(CAP_SYS_ADMIN)) |
2519 | return -EPERM; | |
2520 | ||
191c5424 AV |
2521 | BUG_ON(!current->mm); |
2522 | ||
1da177e4 | 2523 | pathname = getname(specialfile); |
1da177e4 | 2524 | if (IS_ERR(pathname)) |
f58b59c1 | 2525 | return PTR_ERR(pathname); |
1da177e4 | 2526 | |
669abf4e | 2527 | victim = file_open_name(pathname, O_RDWR|O_LARGEFILE, 0); |
1da177e4 LT |
2528 | err = PTR_ERR(victim); |
2529 | if (IS_ERR(victim)) | |
2530 | goto out; | |
2531 | ||
2532 | mapping = victim->f_mapping; | |
5d337b91 | 2533 | spin_lock(&swap_lock); |
18ab4d4c | 2534 | plist_for_each_entry(p, &swap_active_head, list) { |
22c6f8fd | 2535 | if (p->flags & SWP_WRITEOK) { |
adfab836 DS |
2536 | if (p->swap_file->f_mapping == mapping) { |
2537 | found = 1; | |
1da177e4 | 2538 | break; |
adfab836 | 2539 | } |
1da177e4 | 2540 | } |
1da177e4 | 2541 | } |
adfab836 | 2542 | if (!found) { |
1da177e4 | 2543 | err = -EINVAL; |
5d337b91 | 2544 | spin_unlock(&swap_lock); |
1da177e4 LT |
2545 | goto out_dput; |
2546 | } | |
191c5424 | 2547 | if (!security_vm_enough_memory_mm(current->mm, p->pages)) |
1da177e4 LT |
2548 | vm_unacct_memory(p->pages); |
2549 | else { | |
2550 | err = -ENOMEM; | |
5d337b91 | 2551 | spin_unlock(&swap_lock); |
1da177e4 LT |
2552 | goto out_dput; |
2553 | } | |
a2468cc9 | 2554 | del_from_avail_list(p); |
ec8acf20 | 2555 | spin_lock(&p->lock); |
78ecba08 | 2556 | if (p->prio < 0) { |
adfab836 | 2557 | struct swap_info_struct *si = p; |
a2468cc9 | 2558 | int nid; |
adfab836 | 2559 | |
18ab4d4c | 2560 | plist_for_each_entry_continue(si, &swap_active_head, list) { |
adfab836 | 2561 | si->prio++; |
18ab4d4c | 2562 | si->list.prio--; |
a2468cc9 AL |
2563 | for_each_node(nid) { |
2564 | if (si->avail_lists[nid].prio != 1) | |
2565 | si->avail_lists[nid].prio--; | |
2566 | } | |
adfab836 | 2567 | } |
78ecba08 HD |
2568 | least_priority++; |
2569 | } | |
18ab4d4c | 2570 | plist_del(&p->list, &swap_active_head); |
ec8acf20 | 2571 | atomic_long_sub(p->pages, &nr_swap_pages); |
1da177e4 LT |
2572 | total_swap_pages -= p->pages; |
2573 | p->flags &= ~SWP_WRITEOK; | |
ec8acf20 | 2574 | spin_unlock(&p->lock); |
5d337b91 | 2575 | spin_unlock(&swap_lock); |
fb4f88dc | 2576 | |
039939a6 TC |
2577 | disable_swap_slots_cache_lock(); |
2578 | ||
e1e12d2f | 2579 | set_current_oom_origin(); |
adfab836 | 2580 | err = try_to_unuse(p->type, false, 0); /* force unuse all pages */ |
e1e12d2f | 2581 | clear_current_oom_origin(); |
1da177e4 | 2582 | |
1da177e4 LT |
2583 | if (err) { |
2584 | /* re-insert swap space back into swap_list */ | |
cf0cac0a | 2585 | reinsert_swap_info(p); |
039939a6 | 2586 | reenable_swap_slots_cache_unlock(); |
1da177e4 LT |
2587 | goto out_dput; |
2588 | } | |
52b7efdb | 2589 | |
039939a6 TC |
2590 | reenable_swap_slots_cache_unlock(); |
2591 | ||
eb085574 HY |
2592 | spin_lock(&swap_lock); |
2593 | spin_lock(&p->lock); | |
2594 | p->flags &= ~SWP_VALID; /* mark swap device as invalid */ | |
2595 | spin_unlock(&p->lock); | |
2596 | spin_unlock(&swap_lock); | |
2597 | /* | |
2598 | * wait for swap operations protected by get/put_swap_device() | |
2599 | * to complete | |
2600 | */ | |
2601 | synchronize_rcu(); | |
2602 | ||
815c2c54 SL |
2603 | flush_work(&p->discard_work); |
2604 | ||
5d337b91 | 2605 | destroy_swap_extents(p); |
570a335b HD |
2606 | if (p->flags & SWP_CONTINUED) |
2607 | free_swap_count_continuations(p); | |
2608 | ||
81a0298b HY |
2609 | if (!p->bdev || !blk_queue_nonrot(bdev_get_queue(p->bdev))) |
2610 | atomic_dec(&nr_rotate_swap); | |
2611 | ||
fc0abb14 | 2612 | mutex_lock(&swapon_mutex); |
5d337b91 | 2613 | spin_lock(&swap_lock); |
ec8acf20 | 2614 | spin_lock(&p->lock); |
5d337b91 HD |
2615 | drain_mmlist(); |
2616 | ||
52b7efdb | 2617 | /* wait for anyone still in scan_swap_map */ |
52b7efdb HD |
2618 | p->highest_bit = 0; /* cuts scans short */ |
2619 | while (p->flags >= SWP_SCANNING) { | |
ec8acf20 | 2620 | spin_unlock(&p->lock); |
5d337b91 | 2621 | spin_unlock(&swap_lock); |
13e4b57f | 2622 | schedule_timeout_uninterruptible(1); |
5d337b91 | 2623 | spin_lock(&swap_lock); |
ec8acf20 | 2624 | spin_lock(&p->lock); |
52b7efdb | 2625 | } |
52b7efdb | 2626 | |
1da177e4 | 2627 | swap_file = p->swap_file; |
5b808a23 | 2628 | old_block_size = p->old_block_size; |
1da177e4 LT |
2629 | p->swap_file = NULL; |
2630 | p->max = 0; | |
2631 | swap_map = p->swap_map; | |
2632 | p->swap_map = NULL; | |
2a8f9449 SL |
2633 | cluster_info = p->cluster_info; |
2634 | p->cluster_info = NULL; | |
4f89849d | 2635 | frontswap_map = frontswap_map_get(p); |
ec8acf20 | 2636 | spin_unlock(&p->lock); |
5d337b91 | 2637 | spin_unlock(&swap_lock); |
adfab836 | 2638 | frontswap_invalidate_area(p->type); |
58e97ba6 | 2639 | frontswap_map_set(p, NULL); |
fc0abb14 | 2640 | mutex_unlock(&swapon_mutex); |
ebc2a1a6 SL |
2641 | free_percpu(p->percpu_cluster); |
2642 | p->percpu_cluster = NULL; | |
1da177e4 | 2643 | vfree(swap_map); |
54f180d3 HY |
2644 | kvfree(cluster_info); |
2645 | kvfree(frontswap_map); | |
2de1a7e4 | 2646 | /* Destroy swap account information */ |
adfab836 | 2647 | swap_cgroup_swapoff(p->type); |
4b3ef9da | 2648 | exit_swap_address_space(p->type); |
27a7faa0 | 2649 | |
1da177e4 LT |
2650 | inode = mapping->host; |
2651 | if (S_ISBLK(inode->i_mode)) { | |
2652 | struct block_device *bdev = I_BDEV(inode); | |
1638045c | 2653 | |
5b808a23 | 2654 | set_blocksize(bdev, old_block_size); |
e525fd89 | 2655 | blkdev_put(bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL); |
1da177e4 | 2656 | } |
1638045c DW |
2657 | |
2658 | inode_lock(inode); | |
2659 | inode->i_flags &= ~S_SWAPFILE; | |
2660 | inode_unlock(inode); | |
1da177e4 | 2661 | filp_close(swap_file, NULL); |
f893ab41 WY |
2662 | |
2663 | /* | |
2664 | * Clear the SWP_USED flag after all resources are freed so that swapon | |
2665 | * can reuse this swap_info in alloc_swap_info() safely. It is ok to | |
2666 | * not hold p->lock after we cleared its SWP_WRITEOK. | |
2667 | */ | |
2668 | spin_lock(&swap_lock); | |
2669 | p->flags = 0; | |
2670 | spin_unlock(&swap_lock); | |
2671 | ||
1da177e4 | 2672 | err = 0; |
66d7dd51 KS |
2673 | atomic_inc(&proc_poll_event); |
2674 | wake_up_interruptible(&proc_poll_wait); | |
1da177e4 LT |
2675 | |
2676 | out_dput: | |
2677 | filp_close(victim, NULL); | |
2678 | out: | |
f58b59c1 | 2679 | putname(pathname); |
1da177e4 LT |
2680 | return err; |
2681 | } | |
2682 | ||
2683 | #ifdef CONFIG_PROC_FS | |
9dd95748 | 2684 | static __poll_t swaps_poll(struct file *file, poll_table *wait) |
66d7dd51 | 2685 | { |
f1514638 | 2686 | struct seq_file *seq = file->private_data; |
66d7dd51 KS |
2687 | |
2688 | poll_wait(file, &proc_poll_wait, wait); | |
2689 | ||
f1514638 KS |
2690 | if (seq->poll_event != atomic_read(&proc_poll_event)) { |
2691 | seq->poll_event = atomic_read(&proc_poll_event); | |
a9a08845 | 2692 | return EPOLLIN | EPOLLRDNORM | EPOLLERR | EPOLLPRI; |
66d7dd51 KS |
2693 | } |
2694 | ||
a9a08845 | 2695 | return EPOLLIN | EPOLLRDNORM; |
66d7dd51 KS |
2696 | } |
2697 | ||
1da177e4 LT |
2698 | /* iterator */ |
2699 | static void *swap_start(struct seq_file *swap, loff_t *pos) | |
2700 | { | |
efa90a98 HD |
2701 | struct swap_info_struct *si; |
2702 | int type; | |
1da177e4 LT |
2703 | loff_t l = *pos; |
2704 | ||
fc0abb14 | 2705 | mutex_lock(&swapon_mutex); |
1da177e4 | 2706 | |
881e4aab SS |
2707 | if (!l) |
2708 | return SEQ_START_TOKEN; | |
2709 | ||
c10d38cc | 2710 | for (type = 0; (si = swap_type_to_swap_info(type)); type++) { |
efa90a98 | 2711 | if (!(si->flags & SWP_USED) || !si->swap_map) |
1da177e4 | 2712 | continue; |
881e4aab | 2713 | if (!--l) |
efa90a98 | 2714 | return si; |
1da177e4 LT |
2715 | } |
2716 | ||
2717 | return NULL; | |
2718 | } | |
2719 | ||
2720 | static void *swap_next(struct seq_file *swap, void *v, loff_t *pos) | |
2721 | { | |
efa90a98 HD |
2722 | struct swap_info_struct *si = v; |
2723 | int type; | |
1da177e4 | 2724 | |
881e4aab | 2725 | if (v == SEQ_START_TOKEN) |
efa90a98 HD |
2726 | type = 0; |
2727 | else | |
2728 | type = si->type + 1; | |
881e4aab | 2729 | |
10c8d69f | 2730 | ++(*pos); |
c10d38cc | 2731 | for (; (si = swap_type_to_swap_info(type)); type++) { |
efa90a98 | 2732 | if (!(si->flags & SWP_USED) || !si->swap_map) |
1da177e4 | 2733 | continue; |
efa90a98 | 2734 | return si; |
1da177e4 LT |
2735 | } |
2736 | ||
2737 | return NULL; | |
2738 | } | |
2739 | ||
2740 | static void swap_stop(struct seq_file *swap, void *v) | |
2741 | { | |
fc0abb14 | 2742 | mutex_unlock(&swapon_mutex); |
1da177e4 LT |
2743 | } |
2744 | ||
2745 | static int swap_show(struct seq_file *swap, void *v) | |
2746 | { | |
efa90a98 | 2747 | struct swap_info_struct *si = v; |
1da177e4 LT |
2748 | struct file *file; |
2749 | int len; | |
2750 | ||
efa90a98 | 2751 | if (si == SEQ_START_TOKEN) { |
881e4aab SS |
2752 | seq_puts(swap,"Filename\t\t\t\tType\t\tSize\tUsed\tPriority\n"); |
2753 | return 0; | |
2754 | } | |
1da177e4 | 2755 | |
efa90a98 | 2756 | file = si->swap_file; |
2726d566 | 2757 | len = seq_file_path(swap, file, " \t\n\\"); |
6eb396dc | 2758 | seq_printf(swap, "%*s%s\t%u\t%u\t%d\n", |
886bb7e9 | 2759 | len < 40 ? 40 - len : 1, " ", |
496ad9aa | 2760 | S_ISBLK(file_inode(file)->i_mode) ? |
1da177e4 | 2761 | "partition" : "file\t", |
efa90a98 HD |
2762 | si->pages << (PAGE_SHIFT - 10), |
2763 | si->inuse_pages << (PAGE_SHIFT - 10), | |
2764 | si->prio); | |
1da177e4 LT |
2765 | return 0; |
2766 | } | |
2767 | ||
15ad7cdc | 2768 | static const struct seq_operations swaps_op = { |
1da177e4 LT |
2769 | .start = swap_start, |
2770 | .next = swap_next, | |
2771 | .stop = swap_stop, | |
2772 | .show = swap_show | |
2773 | }; | |
2774 | ||
2775 | static int swaps_open(struct inode *inode, struct file *file) | |
2776 | { | |
f1514638 | 2777 | struct seq_file *seq; |
66d7dd51 KS |
2778 | int ret; |
2779 | ||
66d7dd51 | 2780 | ret = seq_open(file, &swaps_op); |
f1514638 | 2781 | if (ret) |
66d7dd51 | 2782 | return ret; |
66d7dd51 | 2783 | |
f1514638 KS |
2784 | seq = file->private_data; |
2785 | seq->poll_event = atomic_read(&proc_poll_event); | |
2786 | return 0; | |
1da177e4 LT |
2787 | } |
2788 | ||
97a32539 | 2789 | static const struct proc_ops swaps_proc_ops = { |
d919b33d | 2790 | .proc_flags = PROC_ENTRY_PERMANENT, |
97a32539 AD |
2791 | .proc_open = swaps_open, |
2792 | .proc_read = seq_read, | |
2793 | .proc_lseek = seq_lseek, | |
2794 | .proc_release = seq_release, | |
2795 | .proc_poll = swaps_poll, | |
1da177e4 LT |
2796 | }; |
2797 | ||
2798 | static int __init procswaps_init(void) | |
2799 | { | |
97a32539 | 2800 | proc_create("swaps", 0, NULL, &swaps_proc_ops); |
1da177e4 LT |
2801 | return 0; |
2802 | } | |
2803 | __initcall(procswaps_init); | |
2804 | #endif /* CONFIG_PROC_FS */ | |
2805 | ||
1796316a JB |
2806 | #ifdef MAX_SWAPFILES_CHECK |
2807 | static int __init max_swapfiles_check(void) | |
2808 | { | |
2809 | MAX_SWAPFILES_CHECK(); | |
2810 | return 0; | |
2811 | } | |
2812 | late_initcall(max_swapfiles_check); | |
2813 | #endif | |
2814 | ||
53cbb243 | 2815 | static struct swap_info_struct *alloc_swap_info(void) |
1da177e4 | 2816 | { |
73c34b6a | 2817 | struct swap_info_struct *p; |
1da177e4 | 2818 | unsigned int type; |
a2468cc9 | 2819 | int i; |
efa90a98 | 2820 | |
96008744 | 2821 | p = kvzalloc(struct_size(p, avail_lists, nr_node_ids), GFP_KERNEL); |
efa90a98 | 2822 | if (!p) |
53cbb243 | 2823 | return ERR_PTR(-ENOMEM); |
efa90a98 | 2824 | |
5d337b91 | 2825 | spin_lock(&swap_lock); |
efa90a98 HD |
2826 | for (type = 0; type < nr_swapfiles; type++) { |
2827 | if (!(swap_info[type]->flags & SWP_USED)) | |
1da177e4 | 2828 | break; |
efa90a98 | 2829 | } |
0697212a | 2830 | if (type >= MAX_SWAPFILES) { |
5d337b91 | 2831 | spin_unlock(&swap_lock); |
873d7bcf | 2832 | kvfree(p); |
730c0581 | 2833 | return ERR_PTR(-EPERM); |
1da177e4 | 2834 | } |
efa90a98 HD |
2835 | if (type >= nr_swapfiles) { |
2836 | p->type = type; | |
c10d38cc | 2837 | WRITE_ONCE(swap_info[type], p); |
efa90a98 HD |
2838 | /* |
2839 | * Write swap_info[type] before nr_swapfiles, in case a | |
2840 | * racing procfs swap_start() or swap_next() is reading them. | |
2841 | * (We never shrink nr_swapfiles, we never free this entry.) | |
2842 | */ | |
2843 | smp_wmb(); | |
c10d38cc | 2844 | WRITE_ONCE(nr_swapfiles, nr_swapfiles + 1); |
efa90a98 | 2845 | } else { |
873d7bcf | 2846 | kvfree(p); |
efa90a98 HD |
2847 | p = swap_info[type]; |
2848 | /* | |
2849 | * Do not memset this entry: a racing procfs swap_next() | |
2850 | * would be relying on p->type to remain valid. | |
2851 | */ | |
2852 | } | |
4efaceb1 | 2853 | p->swap_extent_root = RB_ROOT; |
18ab4d4c | 2854 | plist_node_init(&p->list, 0); |
a2468cc9 AL |
2855 | for_each_node(i) |
2856 | plist_node_init(&p->avail_lists[i], 0); | |
1da177e4 | 2857 | p->flags = SWP_USED; |
5d337b91 | 2858 | spin_unlock(&swap_lock); |
ec8acf20 | 2859 | spin_lock_init(&p->lock); |
2628bd6f | 2860 | spin_lock_init(&p->cont_lock); |
efa90a98 | 2861 | |
53cbb243 | 2862 | return p; |
53cbb243 CEB |
2863 | } |
2864 | ||
4d0e1e10 CEB |
2865 | static int claim_swapfile(struct swap_info_struct *p, struct inode *inode) |
2866 | { | |
2867 | int error; | |
2868 | ||
2869 | if (S_ISBLK(inode->i_mode)) { | |
2870 | p->bdev = bdgrab(I_BDEV(inode)); | |
2871 | error = blkdev_get(p->bdev, | |
6f179af8 | 2872 | FMODE_READ | FMODE_WRITE | FMODE_EXCL, p); |
4d0e1e10 CEB |
2873 | if (error < 0) { |
2874 | p->bdev = NULL; | |
6f179af8 | 2875 | return error; |
4d0e1e10 CEB |
2876 | } |
2877 | p->old_block_size = block_size(p->bdev); | |
2878 | error = set_blocksize(p->bdev, PAGE_SIZE); | |
2879 | if (error < 0) | |
87ade72a | 2880 | return error; |
12d2966d NA |
2881 | /* |
2882 | * Zoned block devices contain zones that have a sequential | |
2883 | * write only restriction. Hence zoned block devices are not | |
2884 | * suitable for swapping. Disallow them here. | |
2885 | */ | |
2886 | if (blk_queue_is_zoned(p->bdev->bd_queue)) | |
2887 | return -EINVAL; | |
4d0e1e10 CEB |
2888 | p->flags |= SWP_BLKDEV; |
2889 | } else if (S_ISREG(inode->i_mode)) { | |
2890 | p->bdev = inode->i_sb->s_bdev; | |
1638045c DW |
2891 | } |
2892 | ||
4d0e1e10 | 2893 | return 0; |
4d0e1e10 CEB |
2894 | } |
2895 | ||
377eeaa8 AK |
2896 | |
2897 | /* | |
2898 | * Find out how many pages are allowed for a single swap device. There | |
2899 | * are two limiting factors: | |
2900 | * 1) the number of bits for the swap offset in the swp_entry_t type, and | |
2901 | * 2) the number of bits in the swap pte, as defined by the different | |
2902 | * architectures. | |
2903 | * | |
2904 | * In order to find the largest possible bit mask, a swap entry with | |
2905 | * swap type 0 and swap offset ~0UL is created, encoded to a swap pte, | |
2906 | * decoded to a swp_entry_t again, and finally the swap offset is | |
2907 | * extracted. | |
2908 | * | |
2909 | * This will mask all the bits from the initial ~0UL mask that can't | |
2910 | * be encoded in either the swp_entry_t or the architecture definition | |
2911 | * of a swap pte. | |
2912 | */ | |
2913 | unsigned long generic_max_swapfile_size(void) | |
2914 | { | |
2915 | return swp_offset(pte_to_swp_entry( | |
2916 | swp_entry_to_pte(swp_entry(0, ~0UL)))) + 1; | |
2917 | } | |
2918 | ||
2919 | /* Can be overridden by an architecture for additional checks. */ | |
2920 | __weak unsigned long max_swapfile_size(void) | |
2921 | { | |
2922 | return generic_max_swapfile_size(); | |
2923 | } | |
2924 | ||
ca8bd38b CEB |
2925 | static unsigned long read_swap_header(struct swap_info_struct *p, |
2926 | union swap_header *swap_header, | |
2927 | struct inode *inode) | |
2928 | { | |
2929 | int i; | |
2930 | unsigned long maxpages; | |
2931 | unsigned long swapfilepages; | |
d6bbbd29 | 2932 | unsigned long last_page; |
ca8bd38b CEB |
2933 | |
2934 | if (memcmp("SWAPSPACE2", swap_header->magic.magic, 10)) { | |
465c47fd | 2935 | pr_err("Unable to find swap-space signature\n"); |
38719025 | 2936 | return 0; |
ca8bd38b CEB |
2937 | } |
2938 | ||
2939 | /* swap partition endianess hack... */ | |
2940 | if (swab32(swap_header->info.version) == 1) { | |
2941 | swab32s(&swap_header->info.version); | |
2942 | swab32s(&swap_header->info.last_page); | |
2943 | swab32s(&swap_header->info.nr_badpages); | |
dd111be6 JH |
2944 | if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES) |
2945 | return 0; | |
ca8bd38b CEB |
2946 | for (i = 0; i < swap_header->info.nr_badpages; i++) |
2947 | swab32s(&swap_header->info.badpages[i]); | |
2948 | } | |
2949 | /* Check the swap header's sub-version */ | |
2950 | if (swap_header->info.version != 1) { | |
465c47fd AM |
2951 | pr_warn("Unable to handle swap header version %d\n", |
2952 | swap_header->info.version); | |
38719025 | 2953 | return 0; |
ca8bd38b CEB |
2954 | } |
2955 | ||
2956 | p->lowest_bit = 1; | |
2957 | p->cluster_next = 1; | |
2958 | p->cluster_nr = 0; | |
2959 | ||
377eeaa8 | 2960 | maxpages = max_swapfile_size(); |
d6bbbd29 | 2961 | last_page = swap_header->info.last_page; |
a06ad633 TA |
2962 | if (!last_page) { |
2963 | pr_warn("Empty swap-file\n"); | |
2964 | return 0; | |
2965 | } | |
d6bbbd29 | 2966 | if (last_page > maxpages) { |
465c47fd | 2967 | pr_warn("Truncating oversized swap area, only using %luk out of %luk\n", |
d6bbbd29 RJ |
2968 | maxpages << (PAGE_SHIFT - 10), |
2969 | last_page << (PAGE_SHIFT - 10)); | |
2970 | } | |
2971 | if (maxpages > last_page) { | |
2972 | maxpages = last_page + 1; | |
ca8bd38b CEB |
2973 | /* p->max is an unsigned int: don't overflow it */ |
2974 | if ((unsigned int)maxpages == 0) | |
2975 | maxpages = UINT_MAX; | |
2976 | } | |
2977 | p->highest_bit = maxpages - 1; | |
2978 | ||
2979 | if (!maxpages) | |
38719025 | 2980 | return 0; |
ca8bd38b CEB |
2981 | swapfilepages = i_size_read(inode) >> PAGE_SHIFT; |
2982 | if (swapfilepages && maxpages > swapfilepages) { | |
465c47fd | 2983 | pr_warn("Swap area shorter than signature indicates\n"); |
38719025 | 2984 | return 0; |
ca8bd38b CEB |
2985 | } |
2986 | if (swap_header->info.nr_badpages && S_ISREG(inode->i_mode)) | |
38719025 | 2987 | return 0; |
ca8bd38b | 2988 | if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES) |
38719025 | 2989 | return 0; |
ca8bd38b CEB |
2990 | |
2991 | return maxpages; | |
ca8bd38b CEB |
2992 | } |
2993 | ||
4b3ef9da | 2994 | #define SWAP_CLUSTER_INFO_COLS \ |
235b6217 | 2995 | DIV_ROUND_UP(L1_CACHE_BYTES, sizeof(struct swap_cluster_info)) |
4b3ef9da HY |
2996 | #define SWAP_CLUSTER_SPACE_COLS \ |
2997 | DIV_ROUND_UP(SWAP_ADDRESS_SPACE_PAGES, SWAPFILE_CLUSTER) | |
2998 | #define SWAP_CLUSTER_COLS \ | |
2999 | max_t(unsigned int, SWAP_CLUSTER_INFO_COLS, SWAP_CLUSTER_SPACE_COLS) | |
235b6217 | 3000 | |
915d4d7b CEB |
3001 | static int setup_swap_map_and_extents(struct swap_info_struct *p, |
3002 | union swap_header *swap_header, | |
3003 | unsigned char *swap_map, | |
2a8f9449 | 3004 | struct swap_cluster_info *cluster_info, |
915d4d7b CEB |
3005 | unsigned long maxpages, |
3006 | sector_t *span) | |
3007 | { | |
235b6217 | 3008 | unsigned int j, k; |
915d4d7b CEB |
3009 | unsigned int nr_good_pages; |
3010 | int nr_extents; | |
2a8f9449 | 3011 | unsigned long nr_clusters = DIV_ROUND_UP(maxpages, SWAPFILE_CLUSTER); |
235b6217 HY |
3012 | unsigned long col = p->cluster_next / SWAPFILE_CLUSTER % SWAP_CLUSTER_COLS; |
3013 | unsigned long i, idx; | |
915d4d7b CEB |
3014 | |
3015 | nr_good_pages = maxpages - 1; /* omit header page */ | |
3016 | ||
6b534915 HY |
3017 | cluster_list_init(&p->free_clusters); |
3018 | cluster_list_init(&p->discard_clusters); | |
2a8f9449 | 3019 | |
915d4d7b CEB |
3020 | for (i = 0; i < swap_header->info.nr_badpages; i++) { |
3021 | unsigned int page_nr = swap_header->info.badpages[i]; | |
bdb8e3f6 CEB |
3022 | if (page_nr == 0 || page_nr > swap_header->info.last_page) |
3023 | return -EINVAL; | |
915d4d7b CEB |
3024 | if (page_nr < maxpages) { |
3025 | swap_map[page_nr] = SWAP_MAP_BAD; | |
3026 | nr_good_pages--; | |
2a8f9449 SL |
3027 | /* |
3028 | * Haven't marked the cluster free yet, no list | |
3029 | * operation involved | |
3030 | */ | |
3031 | inc_cluster_info_page(p, cluster_info, page_nr); | |
915d4d7b CEB |
3032 | } |
3033 | } | |
3034 | ||
2a8f9449 SL |
3035 | /* Haven't marked the cluster free yet, no list operation involved */ |
3036 | for (i = maxpages; i < round_up(maxpages, SWAPFILE_CLUSTER); i++) | |
3037 | inc_cluster_info_page(p, cluster_info, i); | |
3038 | ||
915d4d7b CEB |
3039 | if (nr_good_pages) { |
3040 | swap_map[0] = SWAP_MAP_BAD; | |
2a8f9449 SL |
3041 | /* |
3042 | * Not mark the cluster free yet, no list | |
3043 | * operation involved | |
3044 | */ | |
3045 | inc_cluster_info_page(p, cluster_info, 0); | |
915d4d7b CEB |
3046 | p->max = maxpages; |
3047 | p->pages = nr_good_pages; | |
3048 | nr_extents = setup_swap_extents(p, span); | |
bdb8e3f6 CEB |
3049 | if (nr_extents < 0) |
3050 | return nr_extents; | |
915d4d7b CEB |
3051 | nr_good_pages = p->pages; |
3052 | } | |
3053 | if (!nr_good_pages) { | |
465c47fd | 3054 | pr_warn("Empty swap-file\n"); |
bdb8e3f6 | 3055 | return -EINVAL; |
915d4d7b CEB |
3056 | } |
3057 | ||
2a8f9449 SL |
3058 | if (!cluster_info) |
3059 | return nr_extents; | |
3060 | ||
235b6217 | 3061 | |
4b3ef9da HY |
3062 | /* |
3063 | * Reduce false cache line sharing between cluster_info and | |
3064 | * sharing same address space. | |
3065 | */ | |
235b6217 HY |
3066 | for (k = 0; k < SWAP_CLUSTER_COLS; k++) { |
3067 | j = (k + col) % SWAP_CLUSTER_COLS; | |
3068 | for (i = 0; i < DIV_ROUND_UP(nr_clusters, SWAP_CLUSTER_COLS); i++) { | |
3069 | idx = i * SWAP_CLUSTER_COLS + j; | |
3070 | if (idx >= nr_clusters) | |
3071 | continue; | |
3072 | if (cluster_count(&cluster_info[idx])) | |
3073 | continue; | |
2a8f9449 | 3074 | cluster_set_flag(&cluster_info[idx], CLUSTER_FLAG_FREE); |
6b534915 HY |
3075 | cluster_list_add_tail(&p->free_clusters, cluster_info, |
3076 | idx); | |
2a8f9449 | 3077 | } |
2a8f9449 | 3078 | } |
915d4d7b | 3079 | return nr_extents; |
915d4d7b CEB |
3080 | } |
3081 | ||
dcf6b7dd RA |
3082 | /* |
3083 | * Helper to sys_swapon determining if a given swap | |
3084 | * backing device queue supports DISCARD operations. | |
3085 | */ | |
3086 | static bool swap_discardable(struct swap_info_struct *si) | |
3087 | { | |
3088 | struct request_queue *q = bdev_get_queue(si->bdev); | |
3089 | ||
3090 | if (!q || !blk_queue_discard(q)) | |
3091 | return false; | |
3092 | ||
3093 | return true; | |
3094 | } | |
3095 | ||
53cbb243 CEB |
3096 | SYSCALL_DEFINE2(swapon, const char __user *, specialfile, int, swap_flags) |
3097 | { | |
3098 | struct swap_info_struct *p; | |
91a27b2a | 3099 | struct filename *name; |
53cbb243 CEB |
3100 | struct file *swap_file = NULL; |
3101 | struct address_space *mapping; | |
40531542 | 3102 | int prio; |
53cbb243 CEB |
3103 | int error; |
3104 | union swap_header *swap_header; | |
915d4d7b | 3105 | int nr_extents; |
53cbb243 CEB |
3106 | sector_t span; |
3107 | unsigned long maxpages; | |
53cbb243 | 3108 | unsigned char *swap_map = NULL; |
2a8f9449 | 3109 | struct swap_cluster_info *cluster_info = NULL; |
38b5faf4 | 3110 | unsigned long *frontswap_map = NULL; |
53cbb243 CEB |
3111 | struct page *page = NULL; |
3112 | struct inode *inode = NULL; | |
7cbf3192 | 3113 | bool inced_nr_rotate_swap = false; |
53cbb243 | 3114 | |
d15cab97 HD |
3115 | if (swap_flags & ~SWAP_FLAGS_VALID) |
3116 | return -EINVAL; | |
3117 | ||
53cbb243 CEB |
3118 | if (!capable(CAP_SYS_ADMIN)) |
3119 | return -EPERM; | |
3120 | ||
a2468cc9 AL |
3121 | if (!swap_avail_heads) |
3122 | return -ENOMEM; | |
3123 | ||
53cbb243 | 3124 | p = alloc_swap_info(); |
2542e513 CEB |
3125 | if (IS_ERR(p)) |
3126 | return PTR_ERR(p); | |
53cbb243 | 3127 | |
815c2c54 SL |
3128 | INIT_WORK(&p->discard_work, swap_discard_work); |
3129 | ||
1da177e4 | 3130 | name = getname(specialfile); |
1da177e4 | 3131 | if (IS_ERR(name)) { |
7de7fb6b | 3132 | error = PTR_ERR(name); |
1da177e4 | 3133 | name = NULL; |
bd69010b | 3134 | goto bad_swap; |
1da177e4 | 3135 | } |
669abf4e | 3136 | swap_file = file_open_name(name, O_RDWR|O_LARGEFILE, 0); |
1da177e4 | 3137 | if (IS_ERR(swap_file)) { |
7de7fb6b | 3138 | error = PTR_ERR(swap_file); |
1da177e4 | 3139 | swap_file = NULL; |
bd69010b | 3140 | goto bad_swap; |
1da177e4 LT |
3141 | } |
3142 | ||
3143 | p->swap_file = swap_file; | |
3144 | mapping = swap_file->f_mapping; | |
2130781e | 3145 | inode = mapping->host; |
6f179af8 | 3146 | |
4d0e1e10 CEB |
3147 | error = claim_swapfile(p, inode); |
3148 | if (unlikely(error)) | |
1da177e4 | 3149 | goto bad_swap; |
1da177e4 | 3150 | |
d795a90e NA |
3151 | inode_lock(inode); |
3152 | if (IS_SWAPFILE(inode)) { | |
3153 | error = -EBUSY; | |
3154 | goto bad_swap_unlock_inode; | |
3155 | } | |
3156 | ||
1da177e4 LT |
3157 | /* |
3158 | * Read the swap header. | |
3159 | */ | |
3160 | if (!mapping->a_ops->readpage) { | |
3161 | error = -EINVAL; | |
d795a90e | 3162 | goto bad_swap_unlock_inode; |
1da177e4 | 3163 | } |
090d2b18 | 3164 | page = read_mapping_page(mapping, 0, swap_file); |
1da177e4 LT |
3165 | if (IS_ERR(page)) { |
3166 | error = PTR_ERR(page); | |
d795a90e | 3167 | goto bad_swap_unlock_inode; |
1da177e4 | 3168 | } |
81e33971 | 3169 | swap_header = kmap(page); |
1da177e4 | 3170 | |
ca8bd38b CEB |
3171 | maxpages = read_swap_header(p, swap_header, inode); |
3172 | if (unlikely(!maxpages)) { | |
1da177e4 | 3173 | error = -EINVAL; |
d795a90e | 3174 | goto bad_swap_unlock_inode; |
1da177e4 | 3175 | } |
886bb7e9 | 3176 | |
81e33971 | 3177 | /* OK, set up the swap map and apply the bad block list */ |
803d0c83 | 3178 | swap_map = vzalloc(maxpages); |
81e33971 HD |
3179 | if (!swap_map) { |
3180 | error = -ENOMEM; | |
d795a90e | 3181 | goto bad_swap_unlock_inode; |
81e33971 | 3182 | } |
f0571429 MK |
3183 | |
3184 | if (bdi_cap_stable_pages_required(inode_to_bdi(inode))) | |
3185 | p->flags |= SWP_STABLE_WRITES; | |
3186 | ||
539a6fea MK |
3187 | if (bdi_cap_synchronous_io(inode_to_bdi(inode))) |
3188 | p->flags |= SWP_SYNCHRONOUS_IO; | |
3189 | ||
2a8f9449 | 3190 | if (p->bdev && blk_queue_nonrot(bdev_get_queue(p->bdev))) { |
6f179af8 | 3191 | int cpu; |
235b6217 | 3192 | unsigned long ci, nr_cluster; |
6f179af8 | 3193 | |
2a8f9449 SL |
3194 | p->flags |= SWP_SOLIDSTATE; |
3195 | /* | |
3196 | * select a random position to start with to help wear leveling | |
3197 | * SSD | |
3198 | */ | |
3199 | p->cluster_next = 1 + (prandom_u32() % p->highest_bit); | |
235b6217 | 3200 | nr_cluster = DIV_ROUND_UP(maxpages, SWAPFILE_CLUSTER); |
2a8f9449 | 3201 | |
778e1cdd | 3202 | cluster_info = kvcalloc(nr_cluster, sizeof(*cluster_info), |
54f180d3 | 3203 | GFP_KERNEL); |
2a8f9449 SL |
3204 | if (!cluster_info) { |
3205 | error = -ENOMEM; | |
d795a90e | 3206 | goto bad_swap_unlock_inode; |
2a8f9449 | 3207 | } |
235b6217 HY |
3208 | |
3209 | for (ci = 0; ci < nr_cluster; ci++) | |
3210 | spin_lock_init(&((cluster_info + ci)->lock)); | |
3211 | ||
ebc2a1a6 SL |
3212 | p->percpu_cluster = alloc_percpu(struct percpu_cluster); |
3213 | if (!p->percpu_cluster) { | |
3214 | error = -ENOMEM; | |
d795a90e | 3215 | goto bad_swap_unlock_inode; |
ebc2a1a6 | 3216 | } |
6f179af8 | 3217 | for_each_possible_cpu(cpu) { |
ebc2a1a6 | 3218 | struct percpu_cluster *cluster; |
6f179af8 | 3219 | cluster = per_cpu_ptr(p->percpu_cluster, cpu); |
ebc2a1a6 SL |
3220 | cluster_set_null(&cluster->index); |
3221 | } | |
7cbf3192 | 3222 | } else { |
81a0298b | 3223 | atomic_inc(&nr_rotate_swap); |
7cbf3192 OS |
3224 | inced_nr_rotate_swap = true; |
3225 | } | |
1da177e4 | 3226 | |
1421ef3c CEB |
3227 | error = swap_cgroup_swapon(p->type, maxpages); |
3228 | if (error) | |
d795a90e | 3229 | goto bad_swap_unlock_inode; |
1421ef3c | 3230 | |
915d4d7b | 3231 | nr_extents = setup_swap_map_and_extents(p, swap_header, swap_map, |
2a8f9449 | 3232 | cluster_info, maxpages, &span); |
915d4d7b CEB |
3233 | if (unlikely(nr_extents < 0)) { |
3234 | error = nr_extents; | |
d795a90e | 3235 | goto bad_swap_unlock_inode; |
1da177e4 | 3236 | } |
38b5faf4 | 3237 | /* frontswap enabled? set up bit-per-page map for frontswap */ |
8ea1d2a1 | 3238 | if (IS_ENABLED(CONFIG_FRONTSWAP)) |
778e1cdd KC |
3239 | frontswap_map = kvcalloc(BITS_TO_LONGS(maxpages), |
3240 | sizeof(long), | |
54f180d3 | 3241 | GFP_KERNEL); |
1da177e4 | 3242 | |
2a8f9449 SL |
3243 | if (p->bdev &&(swap_flags & SWAP_FLAG_DISCARD) && swap_discardable(p)) { |
3244 | /* | |
3245 | * When discard is enabled for swap with no particular | |
3246 | * policy flagged, we set all swap discard flags here in | |
3247 | * order to sustain backward compatibility with older | |
3248 | * swapon(8) releases. | |
3249 | */ | |
3250 | p->flags |= (SWP_DISCARDABLE | SWP_AREA_DISCARD | | |
3251 | SWP_PAGE_DISCARD); | |
dcf6b7dd | 3252 | |
2a8f9449 SL |
3253 | /* |
3254 | * By flagging sys_swapon, a sysadmin can tell us to | |
3255 | * either do single-time area discards only, or to just | |
3256 | * perform discards for released swap page-clusters. | |
3257 | * Now it's time to adjust the p->flags accordingly. | |
3258 | */ | |
3259 | if (swap_flags & SWAP_FLAG_DISCARD_ONCE) | |
3260 | p->flags &= ~SWP_PAGE_DISCARD; | |
3261 | else if (swap_flags & SWAP_FLAG_DISCARD_PAGES) | |
3262 | p->flags &= ~SWP_AREA_DISCARD; | |
3263 | ||
3264 | /* issue a swapon-time discard if it's still required */ | |
3265 | if (p->flags & SWP_AREA_DISCARD) { | |
3266 | int err = discard_swap(p); | |
3267 | if (unlikely(err)) | |
3268 | pr_err("swapon: discard_swap(%p): %d\n", | |
3269 | p, err); | |
dcf6b7dd | 3270 | } |
20137a49 | 3271 | } |
6a6ba831 | 3272 | |
4b3ef9da HY |
3273 | error = init_swap_address_space(p->type, maxpages); |
3274 | if (error) | |
d795a90e | 3275 | goto bad_swap_unlock_inode; |
4b3ef9da | 3276 | |
dc617f29 DW |
3277 | /* |
3278 | * Flush any pending IO and dirty mappings before we start using this | |
3279 | * swap device. | |
3280 | */ | |
3281 | inode->i_flags |= S_SWAPFILE; | |
3282 | error = inode_drain_writes(inode); | |
3283 | if (error) { | |
3284 | inode->i_flags &= ~S_SWAPFILE; | |
d795a90e | 3285 | goto bad_swap_unlock_inode; |
dc617f29 DW |
3286 | } |
3287 | ||
fc0abb14 | 3288 | mutex_lock(&swapon_mutex); |
40531542 | 3289 | prio = -1; |
78ecba08 | 3290 | if (swap_flags & SWAP_FLAG_PREFER) |
40531542 | 3291 | prio = |
78ecba08 | 3292 | (swap_flags & SWAP_FLAG_PRIO_MASK) >> SWAP_FLAG_PRIO_SHIFT; |
2a8f9449 | 3293 | enable_swap_info(p, prio, swap_map, cluster_info, frontswap_map); |
c69dbfb8 | 3294 | |
756a025f | 3295 | pr_info("Adding %uk swap on %s. Priority:%d extents:%d across:%lluk %s%s%s%s%s\n", |
91a27b2a | 3296 | p->pages<<(PAGE_SHIFT-10), name->name, p->prio, |
c69dbfb8 CEB |
3297 | nr_extents, (unsigned long long)span<<(PAGE_SHIFT-10), |
3298 | (p->flags & SWP_SOLIDSTATE) ? "SS" : "", | |
38b5faf4 | 3299 | (p->flags & SWP_DISCARDABLE) ? "D" : "", |
dcf6b7dd RA |
3300 | (p->flags & SWP_AREA_DISCARD) ? "s" : "", |
3301 | (p->flags & SWP_PAGE_DISCARD) ? "c" : "", | |
38b5faf4 | 3302 | (frontswap_map) ? "FS" : ""); |
c69dbfb8 | 3303 | |
fc0abb14 | 3304 | mutex_unlock(&swapon_mutex); |
66d7dd51 KS |
3305 | atomic_inc(&proc_poll_event); |
3306 | wake_up_interruptible(&proc_poll_wait); | |
3307 | ||
1da177e4 LT |
3308 | error = 0; |
3309 | goto out; | |
d795a90e NA |
3310 | bad_swap_unlock_inode: |
3311 | inode_unlock(inode); | |
1da177e4 | 3312 | bad_swap: |
ebc2a1a6 SL |
3313 | free_percpu(p->percpu_cluster); |
3314 | p->percpu_cluster = NULL; | |
bd69010b | 3315 | if (inode && S_ISBLK(inode->i_mode) && p->bdev) { |
f2090d2d CEB |
3316 | set_blocksize(p->bdev, p->old_block_size); |
3317 | blkdev_put(p->bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL); | |
1da177e4 | 3318 | } |
d795a90e | 3319 | inode = NULL; |
4cd3bb10 | 3320 | destroy_swap_extents(p); |
e8e6c2ec | 3321 | swap_cgroup_swapoff(p->type); |
5d337b91 | 3322 | spin_lock(&swap_lock); |
1da177e4 | 3323 | p->swap_file = NULL; |
1da177e4 | 3324 | p->flags = 0; |
5d337b91 | 3325 | spin_unlock(&swap_lock); |
1da177e4 | 3326 | vfree(swap_map); |
8606a1a9 | 3327 | kvfree(cluster_info); |
b6b1fd2a | 3328 | kvfree(frontswap_map); |
7cbf3192 OS |
3329 | if (inced_nr_rotate_swap) |
3330 | atomic_dec(&nr_rotate_swap); | |
d795a90e | 3331 | if (swap_file) |
1da177e4 LT |
3332 | filp_close(swap_file, NULL); |
3333 | out: | |
3334 | if (page && !IS_ERR(page)) { | |
3335 | kunmap(page); | |
09cbfeaf | 3336 | put_page(page); |
1da177e4 LT |
3337 | } |
3338 | if (name) | |
3339 | putname(name); | |
1638045c | 3340 | if (inode) |
5955102c | 3341 | inode_unlock(inode); |
039939a6 TC |
3342 | if (!error) |
3343 | enable_swap_slots_cache(); | |
1da177e4 LT |
3344 | return error; |
3345 | } | |
3346 | ||
3347 | void si_swapinfo(struct sysinfo *val) | |
3348 | { | |
efa90a98 | 3349 | unsigned int type; |
1da177e4 LT |
3350 | unsigned long nr_to_be_unused = 0; |
3351 | ||
5d337b91 | 3352 | spin_lock(&swap_lock); |
efa90a98 HD |
3353 | for (type = 0; type < nr_swapfiles; type++) { |
3354 | struct swap_info_struct *si = swap_info[type]; | |
3355 | ||
3356 | if ((si->flags & SWP_USED) && !(si->flags & SWP_WRITEOK)) | |
3357 | nr_to_be_unused += si->inuse_pages; | |
1da177e4 | 3358 | } |
ec8acf20 | 3359 | val->freeswap = atomic_long_read(&nr_swap_pages) + nr_to_be_unused; |
1da177e4 | 3360 | val->totalswap = total_swap_pages + nr_to_be_unused; |
5d337b91 | 3361 | spin_unlock(&swap_lock); |
1da177e4 LT |
3362 | } |
3363 | ||
3364 | /* | |
3365 | * Verify that a swap entry is valid and increment its swap map count. | |
3366 | * | |
355cfa73 KH |
3367 | * Returns error code in following case. |
3368 | * - success -> 0 | |
3369 | * - swp_entry is invalid -> EINVAL | |
3370 | * - swp_entry is migration entry -> EINVAL | |
3371 | * - swap-cache reference is requested but there is already one. -> EEXIST | |
3372 | * - swap-cache reference is requested but the entry is not used. -> ENOENT | |
570a335b | 3373 | * - swap-mapped reference requested but needs continued swap count. -> ENOMEM |
1da177e4 | 3374 | */ |
8d69aaee | 3375 | static int __swap_duplicate(swp_entry_t entry, unsigned char usage) |
1da177e4 | 3376 | { |
73c34b6a | 3377 | struct swap_info_struct *p; |
235b6217 | 3378 | struct swap_cluster_info *ci; |
c10d38cc | 3379 | unsigned long offset; |
8d69aaee HD |
3380 | unsigned char count; |
3381 | unsigned char has_cache; | |
253d553b | 3382 | int err = -EINVAL; |
1da177e4 | 3383 | |
eb085574 | 3384 | p = get_swap_device(entry); |
c10d38cc | 3385 | if (!p) |
235b6217 HY |
3386 | goto out; |
3387 | ||
eb085574 | 3388 | offset = swp_offset(entry); |
235b6217 | 3389 | ci = lock_cluster_or_swap_info(p, offset); |
355cfa73 | 3390 | |
253d553b | 3391 | count = p->swap_map[offset]; |
edfe23da SL |
3392 | |
3393 | /* | |
3394 | * swapin_readahead() doesn't check if a swap entry is valid, so the | |
3395 | * swap entry could be SWAP_MAP_BAD. Check here with lock held. | |
3396 | */ | |
3397 | if (unlikely(swap_count(count) == SWAP_MAP_BAD)) { | |
3398 | err = -ENOENT; | |
3399 | goto unlock_out; | |
3400 | } | |
3401 | ||
253d553b HD |
3402 | has_cache = count & SWAP_HAS_CACHE; |
3403 | count &= ~SWAP_HAS_CACHE; | |
3404 | err = 0; | |
355cfa73 | 3405 | |
253d553b | 3406 | if (usage == SWAP_HAS_CACHE) { |
355cfa73 KH |
3407 | |
3408 | /* set SWAP_HAS_CACHE if there is no cache and entry is used */ | |
253d553b HD |
3409 | if (!has_cache && count) |
3410 | has_cache = SWAP_HAS_CACHE; | |
3411 | else if (has_cache) /* someone else added cache */ | |
3412 | err = -EEXIST; | |
3413 | else /* no users remaining */ | |
3414 | err = -ENOENT; | |
355cfa73 KH |
3415 | |
3416 | } else if (count || has_cache) { | |
253d553b | 3417 | |
570a335b HD |
3418 | if ((count & ~COUNT_CONTINUED) < SWAP_MAP_MAX) |
3419 | count += usage; | |
3420 | else if ((count & ~COUNT_CONTINUED) > SWAP_MAP_MAX) | |
253d553b | 3421 | err = -EINVAL; |
570a335b HD |
3422 | else if (swap_count_continued(p, offset, count)) |
3423 | count = COUNT_CONTINUED; | |
3424 | else | |
3425 | err = -ENOMEM; | |
355cfa73 | 3426 | } else |
253d553b HD |
3427 | err = -ENOENT; /* unused swap entry */ |
3428 | ||
3429 | p->swap_map[offset] = count | has_cache; | |
3430 | ||
355cfa73 | 3431 | unlock_out: |
235b6217 | 3432 | unlock_cluster_or_swap_info(p, ci); |
1da177e4 | 3433 | out: |
eb085574 HY |
3434 | if (p) |
3435 | put_swap_device(p); | |
253d553b | 3436 | return err; |
1da177e4 | 3437 | } |
253d553b | 3438 | |
aaa46865 HD |
3439 | /* |
3440 | * Help swapoff by noting that swap entry belongs to shmem/tmpfs | |
3441 | * (in which case its reference count is never incremented). | |
3442 | */ | |
3443 | void swap_shmem_alloc(swp_entry_t entry) | |
3444 | { | |
3445 | __swap_duplicate(entry, SWAP_MAP_SHMEM); | |
3446 | } | |
3447 | ||
355cfa73 | 3448 | /* |
08259d58 HD |
3449 | * Increase reference count of swap entry by 1. |
3450 | * Returns 0 for success, or -ENOMEM if a swap_count_continuation is required | |
3451 | * but could not be atomically allocated. Returns 0, just as if it succeeded, | |
3452 | * if __swap_duplicate() fails for another reason (-EINVAL or -ENOENT), which | |
3453 | * might occur if a page table entry has got corrupted. | |
355cfa73 | 3454 | */ |
570a335b | 3455 | int swap_duplicate(swp_entry_t entry) |
355cfa73 | 3456 | { |
570a335b HD |
3457 | int err = 0; |
3458 | ||
3459 | while (!err && __swap_duplicate(entry, 1) == -ENOMEM) | |
3460 | err = add_swap_count_continuation(entry, GFP_ATOMIC); | |
3461 | return err; | |
355cfa73 | 3462 | } |
1da177e4 | 3463 | |
cb4b86ba | 3464 | /* |
355cfa73 KH |
3465 | * @entry: swap entry for which we allocate swap cache. |
3466 | * | |
73c34b6a | 3467 | * Called when allocating swap cache for existing swap entry, |
355cfa73 | 3468 | * This can return error codes. Returns 0 at success. |
3eeba135 | 3469 | * -EEXIST means there is a swap cache. |
355cfa73 | 3470 | * Note: return code is different from swap_duplicate(). |
cb4b86ba KH |
3471 | */ |
3472 | int swapcache_prepare(swp_entry_t entry) | |
3473 | { | |
253d553b | 3474 | return __swap_duplicate(entry, SWAP_HAS_CACHE); |
cb4b86ba KH |
3475 | } |
3476 | ||
0bcac06f MK |
3477 | struct swap_info_struct *swp_swap_info(swp_entry_t entry) |
3478 | { | |
c10d38cc | 3479 | return swap_type_to_swap_info(swp_type(entry)); |
0bcac06f MK |
3480 | } |
3481 | ||
f981c595 MG |
3482 | struct swap_info_struct *page_swap_info(struct page *page) |
3483 | { | |
0bcac06f MK |
3484 | swp_entry_t entry = { .val = page_private(page) }; |
3485 | return swp_swap_info(entry); | |
f981c595 MG |
3486 | } |
3487 | ||
3488 | /* | |
3489 | * out-of-line __page_file_ methods to avoid include hell. | |
3490 | */ | |
3491 | struct address_space *__page_file_mapping(struct page *page) | |
3492 | { | |
f981c595 MG |
3493 | return page_swap_info(page)->swap_file->f_mapping; |
3494 | } | |
3495 | EXPORT_SYMBOL_GPL(__page_file_mapping); | |
3496 | ||
3497 | pgoff_t __page_file_index(struct page *page) | |
3498 | { | |
3499 | swp_entry_t swap = { .val = page_private(page) }; | |
f981c595 MG |
3500 | return swp_offset(swap); |
3501 | } | |
3502 | EXPORT_SYMBOL_GPL(__page_file_index); | |
3503 | ||
570a335b HD |
3504 | /* |
3505 | * add_swap_count_continuation - called when a swap count is duplicated | |
3506 | * beyond SWAP_MAP_MAX, it allocates a new page and links that to the entry's | |
3507 | * page of the original vmalloc'ed swap_map, to hold the continuation count | |
3508 | * (for that entry and for its neighbouring PAGE_SIZE swap entries). Called | |
3509 | * again when count is duplicated beyond SWAP_MAP_MAX * SWAP_CONT_MAX, etc. | |
3510 | * | |
3511 | * These continuation pages are seldom referenced: the common paths all work | |
3512 | * on the original swap_map, only referring to a continuation page when the | |
3513 | * low "digit" of a count is incremented or decremented through SWAP_MAP_MAX. | |
3514 | * | |
3515 | * add_swap_count_continuation(, GFP_ATOMIC) can be called while holding | |
3516 | * page table locks; if it fails, add_swap_count_continuation(, GFP_KERNEL) | |
3517 | * can be called after dropping locks. | |
3518 | */ | |
3519 | int add_swap_count_continuation(swp_entry_t entry, gfp_t gfp_mask) | |
3520 | { | |
3521 | struct swap_info_struct *si; | |
235b6217 | 3522 | struct swap_cluster_info *ci; |
570a335b HD |
3523 | struct page *head; |
3524 | struct page *page; | |
3525 | struct page *list_page; | |
3526 | pgoff_t offset; | |
3527 | unsigned char count; | |
eb085574 | 3528 | int ret = 0; |
570a335b HD |
3529 | |
3530 | /* | |
3531 | * When debugging, it's easier to use __GFP_ZERO here; but it's better | |
3532 | * for latency not to zero a page while GFP_ATOMIC and holding locks. | |
3533 | */ | |
3534 | page = alloc_page(gfp_mask | __GFP_HIGHMEM); | |
3535 | ||
eb085574 | 3536 | si = get_swap_device(entry); |
570a335b HD |
3537 | if (!si) { |
3538 | /* | |
3539 | * An acceptable race has occurred since the failing | |
eb085574 | 3540 | * __swap_duplicate(): the swap device may be swapoff |
570a335b HD |
3541 | */ |
3542 | goto outer; | |
3543 | } | |
eb085574 | 3544 | spin_lock(&si->lock); |
570a335b HD |
3545 | |
3546 | offset = swp_offset(entry); | |
235b6217 HY |
3547 | |
3548 | ci = lock_cluster(si, offset); | |
3549 | ||
570a335b HD |
3550 | count = si->swap_map[offset] & ~SWAP_HAS_CACHE; |
3551 | ||
3552 | if ((count & ~COUNT_CONTINUED) != SWAP_MAP_MAX) { | |
3553 | /* | |
3554 | * The higher the swap count, the more likely it is that tasks | |
3555 | * will race to add swap count continuation: we need to avoid | |
3556 | * over-provisioning. | |
3557 | */ | |
3558 | goto out; | |
3559 | } | |
3560 | ||
3561 | if (!page) { | |
eb085574 HY |
3562 | ret = -ENOMEM; |
3563 | goto out; | |
570a335b HD |
3564 | } |
3565 | ||
3566 | /* | |
3567 | * We are fortunate that although vmalloc_to_page uses pte_offset_map, | |
2de1a7e4 SJ |
3568 | * no architecture is using highmem pages for kernel page tables: so it |
3569 | * will not corrupt the GFP_ATOMIC caller's atomic page table kmaps. | |
570a335b HD |
3570 | */ |
3571 | head = vmalloc_to_page(si->swap_map + offset); | |
3572 | offset &= ~PAGE_MASK; | |
3573 | ||
2628bd6f | 3574 | spin_lock(&si->cont_lock); |
570a335b HD |
3575 | /* |
3576 | * Page allocation does not initialize the page's lru field, | |
3577 | * but it does always reset its private field. | |
3578 | */ | |
3579 | if (!page_private(head)) { | |
3580 | BUG_ON(count & COUNT_CONTINUED); | |
3581 | INIT_LIST_HEAD(&head->lru); | |
3582 | set_page_private(head, SWP_CONTINUED); | |
3583 | si->flags |= SWP_CONTINUED; | |
3584 | } | |
3585 | ||
3586 | list_for_each_entry(list_page, &head->lru, lru) { | |
3587 | unsigned char *map; | |
3588 | ||
3589 | /* | |
3590 | * If the previous map said no continuation, but we've found | |
3591 | * a continuation page, free our allocation and use this one. | |
3592 | */ | |
3593 | if (!(count & COUNT_CONTINUED)) | |
2628bd6f | 3594 | goto out_unlock_cont; |
570a335b | 3595 | |
9b04c5fe | 3596 | map = kmap_atomic(list_page) + offset; |
570a335b | 3597 | count = *map; |
9b04c5fe | 3598 | kunmap_atomic(map); |
570a335b HD |
3599 | |
3600 | /* | |
3601 | * If this continuation count now has some space in it, | |
3602 | * free our allocation and use this one. | |
3603 | */ | |
3604 | if ((count & ~COUNT_CONTINUED) != SWAP_CONT_MAX) | |
2628bd6f | 3605 | goto out_unlock_cont; |
570a335b HD |
3606 | } |
3607 | ||
3608 | list_add_tail(&page->lru, &head->lru); | |
3609 | page = NULL; /* now it's attached, don't free it */ | |
2628bd6f HY |
3610 | out_unlock_cont: |
3611 | spin_unlock(&si->cont_lock); | |
570a335b | 3612 | out: |
235b6217 | 3613 | unlock_cluster(ci); |
ec8acf20 | 3614 | spin_unlock(&si->lock); |
eb085574 | 3615 | put_swap_device(si); |
570a335b HD |
3616 | outer: |
3617 | if (page) | |
3618 | __free_page(page); | |
eb085574 | 3619 | return ret; |
570a335b HD |
3620 | } |
3621 | ||
3622 | /* | |
3623 | * swap_count_continued - when the original swap_map count is incremented | |
3624 | * from SWAP_MAP_MAX, check if there is already a continuation page to carry | |
3625 | * into, carry if so, or else fail until a new continuation page is allocated; | |
3626 | * when the original swap_map count is decremented from 0 with continuation, | |
3627 | * borrow from the continuation and report whether it still holds more. | |
235b6217 HY |
3628 | * Called while __swap_duplicate() or swap_entry_free() holds swap or cluster |
3629 | * lock. | |
570a335b HD |
3630 | */ |
3631 | static bool swap_count_continued(struct swap_info_struct *si, | |
3632 | pgoff_t offset, unsigned char count) | |
3633 | { | |
3634 | struct page *head; | |
3635 | struct page *page; | |
3636 | unsigned char *map; | |
2628bd6f | 3637 | bool ret; |
570a335b HD |
3638 | |
3639 | head = vmalloc_to_page(si->swap_map + offset); | |
3640 | if (page_private(head) != SWP_CONTINUED) { | |
3641 | BUG_ON(count & COUNT_CONTINUED); | |
3642 | return false; /* need to add count continuation */ | |
3643 | } | |
3644 | ||
2628bd6f | 3645 | spin_lock(&si->cont_lock); |
570a335b | 3646 | offset &= ~PAGE_MASK; |
213516ac | 3647 | page = list_next_entry(head, lru); |
9b04c5fe | 3648 | map = kmap_atomic(page) + offset; |
570a335b HD |
3649 | |
3650 | if (count == SWAP_MAP_MAX) /* initial increment from swap_map */ | |
3651 | goto init_map; /* jump over SWAP_CONT_MAX checks */ | |
3652 | ||
3653 | if (count == (SWAP_MAP_MAX | COUNT_CONTINUED)) { /* incrementing */ | |
3654 | /* | |
3655 | * Think of how you add 1 to 999 | |
3656 | */ | |
3657 | while (*map == (SWAP_CONT_MAX | COUNT_CONTINUED)) { | |
9b04c5fe | 3658 | kunmap_atomic(map); |
213516ac | 3659 | page = list_next_entry(page, lru); |
570a335b | 3660 | BUG_ON(page == head); |
9b04c5fe | 3661 | map = kmap_atomic(page) + offset; |
570a335b HD |
3662 | } |
3663 | if (*map == SWAP_CONT_MAX) { | |
9b04c5fe | 3664 | kunmap_atomic(map); |
213516ac | 3665 | page = list_next_entry(page, lru); |
2628bd6f HY |
3666 | if (page == head) { |
3667 | ret = false; /* add count continuation */ | |
3668 | goto out; | |
3669 | } | |
9b04c5fe | 3670 | map = kmap_atomic(page) + offset; |
570a335b HD |
3671 | init_map: *map = 0; /* we didn't zero the page */ |
3672 | } | |
3673 | *map += 1; | |
9b04c5fe | 3674 | kunmap_atomic(map); |
213516ac | 3675 | while ((page = list_prev_entry(page, lru)) != head) { |
9b04c5fe | 3676 | map = kmap_atomic(page) + offset; |
570a335b | 3677 | *map = COUNT_CONTINUED; |
9b04c5fe | 3678 | kunmap_atomic(map); |
570a335b | 3679 | } |
2628bd6f | 3680 | ret = true; /* incremented */ |
570a335b HD |
3681 | |
3682 | } else { /* decrementing */ | |
3683 | /* | |
3684 | * Think of how you subtract 1 from 1000 | |
3685 | */ | |
3686 | BUG_ON(count != COUNT_CONTINUED); | |
3687 | while (*map == COUNT_CONTINUED) { | |
9b04c5fe | 3688 | kunmap_atomic(map); |
213516ac | 3689 | page = list_next_entry(page, lru); |
570a335b | 3690 | BUG_ON(page == head); |
9b04c5fe | 3691 | map = kmap_atomic(page) + offset; |
570a335b HD |
3692 | } |
3693 | BUG_ON(*map == 0); | |
3694 | *map -= 1; | |
3695 | if (*map == 0) | |
3696 | count = 0; | |
9b04c5fe | 3697 | kunmap_atomic(map); |
213516ac | 3698 | while ((page = list_prev_entry(page, lru)) != head) { |
9b04c5fe | 3699 | map = kmap_atomic(page) + offset; |
570a335b HD |
3700 | *map = SWAP_CONT_MAX | count; |
3701 | count = COUNT_CONTINUED; | |
9b04c5fe | 3702 | kunmap_atomic(map); |
570a335b | 3703 | } |
2628bd6f | 3704 | ret = count == COUNT_CONTINUED; |
570a335b | 3705 | } |
2628bd6f HY |
3706 | out: |
3707 | spin_unlock(&si->cont_lock); | |
3708 | return ret; | |
570a335b HD |
3709 | } |
3710 | ||
3711 | /* | |
3712 | * free_swap_count_continuations - swapoff free all the continuation pages | |
3713 | * appended to the swap_map, after swap_map is quiesced, before vfree'ing it. | |
3714 | */ | |
3715 | static void free_swap_count_continuations(struct swap_info_struct *si) | |
3716 | { | |
3717 | pgoff_t offset; | |
3718 | ||
3719 | for (offset = 0; offset < si->max; offset += PAGE_SIZE) { | |
3720 | struct page *head; | |
3721 | head = vmalloc_to_page(si->swap_map + offset); | |
3722 | if (page_private(head)) { | |
0d576d20 GT |
3723 | struct page *page, *next; |
3724 | ||
3725 | list_for_each_entry_safe(page, next, &head->lru, lru) { | |
3726 | list_del(&page->lru); | |
570a335b HD |
3727 | __free_page(page); |
3728 | } | |
3729 | } | |
3730 | } | |
3731 | } | |
a2468cc9 | 3732 | |
2cf85583 TH |
3733 | #if defined(CONFIG_MEMCG) && defined(CONFIG_BLK_CGROUP) |
3734 | void mem_cgroup_throttle_swaprate(struct mem_cgroup *memcg, int node, | |
3735 | gfp_t gfp_mask) | |
3736 | { | |
3737 | struct swap_info_struct *si, *next; | |
3738 | if (!(gfp_mask & __GFP_IO) || !memcg) | |
3739 | return; | |
3740 | ||
3741 | if (!blk_cgroup_congested()) | |
3742 | return; | |
3743 | ||
3744 | /* | |
3745 | * We've already scheduled a throttle, avoid taking the global swap | |
3746 | * lock. | |
3747 | */ | |
3748 | if (current->throttle_queue) | |
3749 | return; | |
3750 | ||
3751 | spin_lock(&swap_avail_lock); | |
3752 | plist_for_each_entry_safe(si, next, &swap_avail_heads[node], | |
3753 | avail_lists[node]) { | |
3754 | if (si->bdev) { | |
3755 | blkcg_schedule_throttle(bdev_get_queue(si->bdev), | |
3756 | true); | |
3757 | break; | |
3758 | } | |
3759 | } | |
3760 | spin_unlock(&swap_avail_lock); | |
3761 | } | |
3762 | #endif | |
3763 | ||
a2468cc9 AL |
3764 | static int __init swapfile_init(void) |
3765 | { | |
3766 | int nid; | |
3767 | ||
3768 | swap_avail_heads = kmalloc_array(nr_node_ids, sizeof(struct plist_head), | |
3769 | GFP_KERNEL); | |
3770 | if (!swap_avail_heads) { | |
3771 | pr_emerg("Not enough memory for swap heads, swap is disabled\n"); | |
3772 | return -ENOMEM; | |
3773 | } | |
3774 | ||
3775 | for_each_node(nid) | |
3776 | plist_head_init(&swap_avail_heads[nid]); | |
3777 | ||
3778 | return 0; | |
3779 | } | |
3780 | subsys_initcall(swapfile_init); |