]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/mm/swapfile.c | |
3 | * | |
4 | * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds | |
5 | * Swap reorganised 29.12.95, Stephen Tweedie | |
6 | */ | |
7 | ||
1da177e4 | 8 | #include <linux/mm.h> |
6e84f315 | 9 | #include <linux/sched/mm.h> |
29930025 | 10 | #include <linux/sched/task.h> |
1da177e4 LT |
11 | #include <linux/hugetlb.h> |
12 | #include <linux/mman.h> | |
13 | #include <linux/slab.h> | |
14 | #include <linux/kernel_stat.h> | |
15 | #include <linux/swap.h> | |
16 | #include <linux/vmalloc.h> | |
17 | #include <linux/pagemap.h> | |
18 | #include <linux/namei.h> | |
072441e2 | 19 | #include <linux/shmem_fs.h> |
1da177e4 | 20 | #include <linux/blkdev.h> |
20137a49 | 21 | #include <linux/random.h> |
1da177e4 LT |
22 | #include <linux/writeback.h> |
23 | #include <linux/proc_fs.h> | |
24 | #include <linux/seq_file.h> | |
25 | #include <linux/init.h> | |
5ad64688 | 26 | #include <linux/ksm.h> |
1da177e4 LT |
27 | #include <linux/rmap.h> |
28 | #include <linux/security.h> | |
29 | #include <linux/backing-dev.h> | |
fc0abb14 | 30 | #include <linux/mutex.h> |
c59ede7b | 31 | #include <linux/capability.h> |
1da177e4 | 32 | #include <linux/syscalls.h> |
8a9f3ccd | 33 | #include <linux/memcontrol.h> |
66d7dd51 | 34 | #include <linux/poll.h> |
72788c38 | 35 | #include <linux/oom.h> |
38b5faf4 DM |
36 | #include <linux/frontswap.h> |
37 | #include <linux/swapfile.h> | |
f981c595 | 38 | #include <linux/export.h> |
67afa38e | 39 | #include <linux/swap_slots.h> |
155b5f88 | 40 | #include <linux/sort.h> |
1da177e4 LT |
41 | |
42 | #include <asm/pgtable.h> | |
43 | #include <asm/tlbflush.h> | |
44 | #include <linux/swapops.h> | |
5d1ea48b | 45 | #include <linux/swap_cgroup.h> |
1da177e4 | 46 | |
570a335b HD |
47 | static bool swap_count_continued(struct swap_info_struct *, pgoff_t, |
48 | unsigned char); | |
49 | static void free_swap_count_continuations(struct swap_info_struct *); | |
d4906e1a | 50 | static sector_t map_swap_entry(swp_entry_t, struct block_device**); |
570a335b | 51 | |
38b5faf4 | 52 | DEFINE_SPINLOCK(swap_lock); |
7c363b8c | 53 | static unsigned int nr_swapfiles; |
ec8acf20 | 54 | atomic_long_t nr_swap_pages; |
fb0fec50 CW |
55 | /* |
56 | * Some modules use swappable objects and may try to swap them out under | |
57 | * memory pressure (via the shrinker). Before doing so, they may wish to | |
58 | * check to see if any swap space is available. | |
59 | */ | |
60 | EXPORT_SYMBOL_GPL(nr_swap_pages); | |
ec8acf20 | 61 | /* protected with swap_lock. reading in vm_swap_full() doesn't need lock */ |
1da177e4 | 62 | long total_swap_pages; |
a2468cc9 | 63 | static int least_priority = -1; |
1da177e4 | 64 | |
1da177e4 LT |
65 | static const char Bad_file[] = "Bad swap file entry "; |
66 | static const char Unused_file[] = "Unused swap file entry "; | |
67 | static const char Bad_offset[] = "Bad swap offset entry "; | |
68 | static const char Unused_offset[] = "Unused swap offset entry "; | |
69 | ||
adfab836 DS |
70 | /* |
71 | * all active swap_info_structs | |
72 | * protected with swap_lock, and ordered by priority. | |
73 | */ | |
18ab4d4c DS |
74 | PLIST_HEAD(swap_active_head); |
75 | ||
76 | /* | |
77 | * all available (active, not full) swap_info_structs | |
78 | * protected with swap_avail_lock, ordered by priority. | |
79 | * This is used by get_swap_page() instead of swap_active_head | |
80 | * because swap_active_head includes all swap_info_structs, | |
81 | * but get_swap_page() doesn't need to look at full ones. | |
82 | * This uses its own lock instead of swap_lock because when a | |
83 | * swap_info_struct changes between not-full/full, it needs to | |
84 | * add/remove itself to/from this list, but the swap_info_struct->lock | |
85 | * is held and the locking order requires swap_lock to be taken | |
86 | * before any swap_info_struct->lock. | |
87 | */ | |
bfc6b1ca | 88 | static struct plist_head *swap_avail_heads; |
18ab4d4c | 89 | static DEFINE_SPINLOCK(swap_avail_lock); |
1da177e4 | 90 | |
38b5faf4 | 91 | struct swap_info_struct *swap_info[MAX_SWAPFILES]; |
1da177e4 | 92 | |
fc0abb14 | 93 | static DEFINE_MUTEX(swapon_mutex); |
1da177e4 | 94 | |
66d7dd51 KS |
95 | static DECLARE_WAIT_QUEUE_HEAD(proc_poll_wait); |
96 | /* Activity counter to indicate that a swapon or swapoff has occurred */ | |
97 | static atomic_t proc_poll_event = ATOMIC_INIT(0); | |
98 | ||
81a0298b HY |
99 | atomic_t nr_rotate_swap = ATOMIC_INIT(0); |
100 | ||
8d69aaee | 101 | static inline unsigned char swap_count(unsigned char ent) |
355cfa73 | 102 | { |
955c97f0 | 103 | return ent & ~SWAP_HAS_CACHE; /* may include COUNT_CONTINUED flag */ |
355cfa73 KH |
104 | } |
105 | ||
efa90a98 | 106 | /* returns 1 if swap entry is freed */ |
c9e44410 KH |
107 | static int |
108 | __try_to_reclaim_swap(struct swap_info_struct *si, unsigned long offset) | |
109 | { | |
efa90a98 | 110 | swp_entry_t entry = swp_entry(si->type, offset); |
c9e44410 KH |
111 | struct page *page; |
112 | int ret = 0; | |
113 | ||
f6ab1f7f | 114 | page = find_get_page(swap_address_space(entry), swp_offset(entry)); |
c9e44410 KH |
115 | if (!page) |
116 | return 0; | |
117 | /* | |
118 | * This function is called from scan_swap_map() and it's called | |
119 | * by vmscan.c at reclaiming pages. So, we hold a lock on a page, here. | |
120 | * We have to use trylock for avoiding deadlock. This is a special | |
121 | * case and you should use try_to_free_swap() with explicit lock_page() | |
122 | * in usual operations. | |
123 | */ | |
124 | if (trylock_page(page)) { | |
125 | ret = try_to_free_swap(page); | |
126 | unlock_page(page); | |
127 | } | |
09cbfeaf | 128 | put_page(page); |
c9e44410 KH |
129 | return ret; |
130 | } | |
355cfa73 | 131 | |
6a6ba831 HD |
132 | /* |
133 | * swapon tell device that all the old swap contents can be discarded, | |
134 | * to allow the swap device to optimize its wear-levelling. | |
135 | */ | |
136 | static int discard_swap(struct swap_info_struct *si) | |
137 | { | |
138 | struct swap_extent *se; | |
9625a5f2 HD |
139 | sector_t start_block; |
140 | sector_t nr_blocks; | |
6a6ba831 HD |
141 | int err = 0; |
142 | ||
9625a5f2 HD |
143 | /* Do not discard the swap header page! */ |
144 | se = &si->first_swap_extent; | |
145 | start_block = (se->start_block + 1) << (PAGE_SHIFT - 9); | |
146 | nr_blocks = ((sector_t)se->nr_pages - 1) << (PAGE_SHIFT - 9); | |
147 | if (nr_blocks) { | |
148 | err = blkdev_issue_discard(si->bdev, start_block, | |
dd3932ed | 149 | nr_blocks, GFP_KERNEL, 0); |
9625a5f2 HD |
150 | if (err) |
151 | return err; | |
152 | cond_resched(); | |
153 | } | |
6a6ba831 | 154 | |
9625a5f2 HD |
155 | list_for_each_entry(se, &si->first_swap_extent.list, list) { |
156 | start_block = se->start_block << (PAGE_SHIFT - 9); | |
157 | nr_blocks = (sector_t)se->nr_pages << (PAGE_SHIFT - 9); | |
6a6ba831 HD |
158 | |
159 | err = blkdev_issue_discard(si->bdev, start_block, | |
dd3932ed | 160 | nr_blocks, GFP_KERNEL, 0); |
6a6ba831 HD |
161 | if (err) |
162 | break; | |
163 | ||
164 | cond_resched(); | |
165 | } | |
166 | return err; /* That will often be -EOPNOTSUPP */ | |
167 | } | |
168 | ||
7992fde7 HD |
169 | /* |
170 | * swap allocation tell device that a cluster of swap can now be discarded, | |
171 | * to allow the swap device to optimize its wear-levelling. | |
172 | */ | |
173 | static void discard_swap_cluster(struct swap_info_struct *si, | |
174 | pgoff_t start_page, pgoff_t nr_pages) | |
175 | { | |
176 | struct swap_extent *se = si->curr_swap_extent; | |
177 | int found_extent = 0; | |
178 | ||
179 | while (nr_pages) { | |
7992fde7 HD |
180 | if (se->start_page <= start_page && |
181 | start_page < se->start_page + se->nr_pages) { | |
182 | pgoff_t offset = start_page - se->start_page; | |
183 | sector_t start_block = se->start_block + offset; | |
858a2990 | 184 | sector_t nr_blocks = se->nr_pages - offset; |
7992fde7 HD |
185 | |
186 | if (nr_blocks > nr_pages) | |
187 | nr_blocks = nr_pages; | |
188 | start_page += nr_blocks; | |
189 | nr_pages -= nr_blocks; | |
190 | ||
191 | if (!found_extent++) | |
192 | si->curr_swap_extent = se; | |
193 | ||
194 | start_block <<= PAGE_SHIFT - 9; | |
195 | nr_blocks <<= PAGE_SHIFT - 9; | |
196 | if (blkdev_issue_discard(si->bdev, start_block, | |
dd3932ed | 197 | nr_blocks, GFP_NOIO, 0)) |
7992fde7 HD |
198 | break; |
199 | } | |
200 | ||
a8ae4991 | 201 | se = list_next_entry(se, list); |
7992fde7 HD |
202 | } |
203 | } | |
204 | ||
38d8b4e6 HY |
205 | #ifdef CONFIG_THP_SWAP |
206 | #define SWAPFILE_CLUSTER HPAGE_PMD_NR | |
207 | #else | |
048c27fd | 208 | #define SWAPFILE_CLUSTER 256 |
38d8b4e6 | 209 | #endif |
048c27fd HD |
210 | #define LATENCY_LIMIT 256 |
211 | ||
2a8f9449 SL |
212 | static inline void cluster_set_flag(struct swap_cluster_info *info, |
213 | unsigned int flag) | |
214 | { | |
215 | info->flags = flag; | |
216 | } | |
217 | ||
218 | static inline unsigned int cluster_count(struct swap_cluster_info *info) | |
219 | { | |
220 | return info->data; | |
221 | } | |
222 | ||
223 | static inline void cluster_set_count(struct swap_cluster_info *info, | |
224 | unsigned int c) | |
225 | { | |
226 | info->data = c; | |
227 | } | |
228 | ||
229 | static inline void cluster_set_count_flag(struct swap_cluster_info *info, | |
230 | unsigned int c, unsigned int f) | |
231 | { | |
232 | info->flags = f; | |
233 | info->data = c; | |
234 | } | |
235 | ||
236 | static inline unsigned int cluster_next(struct swap_cluster_info *info) | |
237 | { | |
238 | return info->data; | |
239 | } | |
240 | ||
241 | static inline void cluster_set_next(struct swap_cluster_info *info, | |
242 | unsigned int n) | |
243 | { | |
244 | info->data = n; | |
245 | } | |
246 | ||
247 | static inline void cluster_set_next_flag(struct swap_cluster_info *info, | |
248 | unsigned int n, unsigned int f) | |
249 | { | |
250 | info->flags = f; | |
251 | info->data = n; | |
252 | } | |
253 | ||
254 | static inline bool cluster_is_free(struct swap_cluster_info *info) | |
255 | { | |
256 | return info->flags & CLUSTER_FLAG_FREE; | |
257 | } | |
258 | ||
259 | static inline bool cluster_is_null(struct swap_cluster_info *info) | |
260 | { | |
261 | return info->flags & CLUSTER_FLAG_NEXT_NULL; | |
262 | } | |
263 | ||
264 | static inline void cluster_set_null(struct swap_cluster_info *info) | |
265 | { | |
266 | info->flags = CLUSTER_FLAG_NEXT_NULL; | |
267 | info->data = 0; | |
268 | } | |
269 | ||
e0709829 HY |
270 | static inline bool cluster_is_huge(struct swap_cluster_info *info) |
271 | { | |
272 | return info->flags & CLUSTER_FLAG_HUGE; | |
273 | } | |
274 | ||
275 | static inline void cluster_clear_huge(struct swap_cluster_info *info) | |
276 | { | |
277 | info->flags &= ~CLUSTER_FLAG_HUGE; | |
278 | } | |
279 | ||
235b6217 HY |
280 | static inline struct swap_cluster_info *lock_cluster(struct swap_info_struct *si, |
281 | unsigned long offset) | |
282 | { | |
283 | struct swap_cluster_info *ci; | |
284 | ||
285 | ci = si->cluster_info; | |
286 | if (ci) { | |
287 | ci += offset / SWAPFILE_CLUSTER; | |
288 | spin_lock(&ci->lock); | |
289 | } | |
290 | return ci; | |
291 | } | |
292 | ||
293 | static inline void unlock_cluster(struct swap_cluster_info *ci) | |
294 | { | |
295 | if (ci) | |
296 | spin_unlock(&ci->lock); | |
297 | } | |
298 | ||
59d98bf3 HY |
299 | /* |
300 | * Determine the locking method in use for this device. Return | |
301 | * swap_cluster_info if SSD-style cluster-based locking is in place. | |
302 | */ | |
235b6217 | 303 | static inline struct swap_cluster_info *lock_cluster_or_swap_info( |
59d98bf3 | 304 | struct swap_info_struct *si, unsigned long offset) |
235b6217 HY |
305 | { |
306 | struct swap_cluster_info *ci; | |
307 | ||
59d98bf3 | 308 | /* Try to use fine-grained SSD-style locking if available: */ |
235b6217 | 309 | ci = lock_cluster(si, offset); |
59d98bf3 | 310 | /* Otherwise, fall back to traditional, coarse locking: */ |
235b6217 HY |
311 | if (!ci) |
312 | spin_lock(&si->lock); | |
313 | ||
314 | return ci; | |
315 | } | |
316 | ||
317 | static inline void unlock_cluster_or_swap_info(struct swap_info_struct *si, | |
318 | struct swap_cluster_info *ci) | |
319 | { | |
320 | if (ci) | |
321 | unlock_cluster(ci); | |
322 | else | |
323 | spin_unlock(&si->lock); | |
324 | } | |
325 | ||
6b534915 HY |
326 | static inline bool cluster_list_empty(struct swap_cluster_list *list) |
327 | { | |
328 | return cluster_is_null(&list->head); | |
329 | } | |
330 | ||
331 | static inline unsigned int cluster_list_first(struct swap_cluster_list *list) | |
332 | { | |
333 | return cluster_next(&list->head); | |
334 | } | |
335 | ||
336 | static void cluster_list_init(struct swap_cluster_list *list) | |
337 | { | |
338 | cluster_set_null(&list->head); | |
339 | cluster_set_null(&list->tail); | |
340 | } | |
341 | ||
342 | static void cluster_list_add_tail(struct swap_cluster_list *list, | |
343 | struct swap_cluster_info *ci, | |
344 | unsigned int idx) | |
345 | { | |
346 | if (cluster_list_empty(list)) { | |
347 | cluster_set_next_flag(&list->head, idx, 0); | |
348 | cluster_set_next_flag(&list->tail, idx, 0); | |
349 | } else { | |
235b6217 | 350 | struct swap_cluster_info *ci_tail; |
6b534915 HY |
351 | unsigned int tail = cluster_next(&list->tail); |
352 | ||
235b6217 HY |
353 | /* |
354 | * Nested cluster lock, but both cluster locks are | |
355 | * only acquired when we held swap_info_struct->lock | |
356 | */ | |
357 | ci_tail = ci + tail; | |
358 | spin_lock_nested(&ci_tail->lock, SINGLE_DEPTH_NESTING); | |
359 | cluster_set_next(ci_tail, idx); | |
0ef017d1 | 360 | spin_unlock(&ci_tail->lock); |
6b534915 HY |
361 | cluster_set_next_flag(&list->tail, idx, 0); |
362 | } | |
363 | } | |
364 | ||
365 | static unsigned int cluster_list_del_first(struct swap_cluster_list *list, | |
366 | struct swap_cluster_info *ci) | |
367 | { | |
368 | unsigned int idx; | |
369 | ||
370 | idx = cluster_next(&list->head); | |
371 | if (cluster_next(&list->tail) == idx) { | |
372 | cluster_set_null(&list->head); | |
373 | cluster_set_null(&list->tail); | |
374 | } else | |
375 | cluster_set_next_flag(&list->head, | |
376 | cluster_next(&ci[idx]), 0); | |
377 | ||
378 | return idx; | |
379 | } | |
380 | ||
815c2c54 SL |
381 | /* Add a cluster to discard list and schedule it to do discard */ |
382 | static void swap_cluster_schedule_discard(struct swap_info_struct *si, | |
383 | unsigned int idx) | |
384 | { | |
385 | /* | |
386 | * If scan_swap_map() can't find a free cluster, it will check | |
387 | * si->swap_map directly. To make sure the discarding cluster isn't | |
388 | * taken by scan_swap_map(), mark the swap entries bad (occupied). It | |
389 | * will be cleared after discard | |
390 | */ | |
391 | memset(si->swap_map + idx * SWAPFILE_CLUSTER, | |
392 | SWAP_MAP_BAD, SWAPFILE_CLUSTER); | |
393 | ||
6b534915 | 394 | cluster_list_add_tail(&si->discard_clusters, si->cluster_info, idx); |
815c2c54 SL |
395 | |
396 | schedule_work(&si->discard_work); | |
397 | } | |
398 | ||
38d8b4e6 HY |
399 | static void __free_cluster(struct swap_info_struct *si, unsigned long idx) |
400 | { | |
401 | struct swap_cluster_info *ci = si->cluster_info; | |
402 | ||
403 | cluster_set_flag(ci + idx, CLUSTER_FLAG_FREE); | |
404 | cluster_list_add_tail(&si->free_clusters, ci, idx); | |
405 | } | |
406 | ||
815c2c54 SL |
407 | /* |
408 | * Doing discard actually. After a cluster discard is finished, the cluster | |
409 | * will be added to free cluster list. caller should hold si->lock. | |
410 | */ | |
411 | static void swap_do_scheduled_discard(struct swap_info_struct *si) | |
412 | { | |
235b6217 | 413 | struct swap_cluster_info *info, *ci; |
815c2c54 SL |
414 | unsigned int idx; |
415 | ||
416 | info = si->cluster_info; | |
417 | ||
6b534915 HY |
418 | while (!cluster_list_empty(&si->discard_clusters)) { |
419 | idx = cluster_list_del_first(&si->discard_clusters, info); | |
815c2c54 SL |
420 | spin_unlock(&si->lock); |
421 | ||
422 | discard_swap_cluster(si, idx * SWAPFILE_CLUSTER, | |
423 | SWAPFILE_CLUSTER); | |
424 | ||
425 | spin_lock(&si->lock); | |
235b6217 | 426 | ci = lock_cluster(si, idx * SWAPFILE_CLUSTER); |
38d8b4e6 | 427 | __free_cluster(si, idx); |
815c2c54 SL |
428 | memset(si->swap_map + idx * SWAPFILE_CLUSTER, |
429 | 0, SWAPFILE_CLUSTER); | |
235b6217 | 430 | unlock_cluster(ci); |
815c2c54 SL |
431 | } |
432 | } | |
433 | ||
434 | static void swap_discard_work(struct work_struct *work) | |
435 | { | |
436 | struct swap_info_struct *si; | |
437 | ||
438 | si = container_of(work, struct swap_info_struct, discard_work); | |
439 | ||
440 | spin_lock(&si->lock); | |
441 | swap_do_scheduled_discard(si); | |
442 | spin_unlock(&si->lock); | |
443 | } | |
444 | ||
38d8b4e6 HY |
445 | static void alloc_cluster(struct swap_info_struct *si, unsigned long idx) |
446 | { | |
447 | struct swap_cluster_info *ci = si->cluster_info; | |
448 | ||
449 | VM_BUG_ON(cluster_list_first(&si->free_clusters) != idx); | |
450 | cluster_list_del_first(&si->free_clusters, ci); | |
451 | cluster_set_count_flag(ci + idx, 0, 0); | |
452 | } | |
453 | ||
454 | static void free_cluster(struct swap_info_struct *si, unsigned long idx) | |
455 | { | |
456 | struct swap_cluster_info *ci = si->cluster_info + idx; | |
457 | ||
458 | VM_BUG_ON(cluster_count(ci) != 0); | |
459 | /* | |
460 | * If the swap is discardable, prepare discard the cluster | |
461 | * instead of free it immediately. The cluster will be freed | |
462 | * after discard. | |
463 | */ | |
464 | if ((si->flags & (SWP_WRITEOK | SWP_PAGE_DISCARD)) == | |
465 | (SWP_WRITEOK | SWP_PAGE_DISCARD)) { | |
466 | swap_cluster_schedule_discard(si, idx); | |
467 | return; | |
468 | } | |
469 | ||
470 | __free_cluster(si, idx); | |
471 | } | |
472 | ||
2a8f9449 SL |
473 | /* |
474 | * The cluster corresponding to page_nr will be used. The cluster will be | |
475 | * removed from free cluster list and its usage counter will be increased. | |
476 | */ | |
477 | static void inc_cluster_info_page(struct swap_info_struct *p, | |
478 | struct swap_cluster_info *cluster_info, unsigned long page_nr) | |
479 | { | |
480 | unsigned long idx = page_nr / SWAPFILE_CLUSTER; | |
481 | ||
482 | if (!cluster_info) | |
483 | return; | |
38d8b4e6 HY |
484 | if (cluster_is_free(&cluster_info[idx])) |
485 | alloc_cluster(p, idx); | |
2a8f9449 SL |
486 | |
487 | VM_BUG_ON(cluster_count(&cluster_info[idx]) >= SWAPFILE_CLUSTER); | |
488 | cluster_set_count(&cluster_info[idx], | |
489 | cluster_count(&cluster_info[idx]) + 1); | |
490 | } | |
491 | ||
492 | /* | |
493 | * The cluster corresponding to page_nr decreases one usage. If the usage | |
494 | * counter becomes 0, which means no page in the cluster is in using, we can | |
495 | * optionally discard the cluster and add it to free cluster list. | |
496 | */ | |
497 | static void dec_cluster_info_page(struct swap_info_struct *p, | |
498 | struct swap_cluster_info *cluster_info, unsigned long page_nr) | |
499 | { | |
500 | unsigned long idx = page_nr / SWAPFILE_CLUSTER; | |
501 | ||
502 | if (!cluster_info) | |
503 | return; | |
504 | ||
505 | VM_BUG_ON(cluster_count(&cluster_info[idx]) == 0); | |
506 | cluster_set_count(&cluster_info[idx], | |
507 | cluster_count(&cluster_info[idx]) - 1); | |
508 | ||
38d8b4e6 HY |
509 | if (cluster_count(&cluster_info[idx]) == 0) |
510 | free_cluster(p, idx); | |
2a8f9449 SL |
511 | } |
512 | ||
513 | /* | |
514 | * It's possible scan_swap_map() uses a free cluster in the middle of free | |
515 | * cluster list. Avoiding such abuse to avoid list corruption. | |
516 | */ | |
ebc2a1a6 SL |
517 | static bool |
518 | scan_swap_map_ssd_cluster_conflict(struct swap_info_struct *si, | |
2a8f9449 SL |
519 | unsigned long offset) |
520 | { | |
ebc2a1a6 SL |
521 | struct percpu_cluster *percpu_cluster; |
522 | bool conflict; | |
523 | ||
2a8f9449 | 524 | offset /= SWAPFILE_CLUSTER; |
6b534915 HY |
525 | conflict = !cluster_list_empty(&si->free_clusters) && |
526 | offset != cluster_list_first(&si->free_clusters) && | |
2a8f9449 | 527 | cluster_is_free(&si->cluster_info[offset]); |
ebc2a1a6 SL |
528 | |
529 | if (!conflict) | |
530 | return false; | |
531 | ||
532 | percpu_cluster = this_cpu_ptr(si->percpu_cluster); | |
533 | cluster_set_null(&percpu_cluster->index); | |
534 | return true; | |
535 | } | |
536 | ||
537 | /* | |
538 | * Try to get a swap entry from current cpu's swap entry pool (a cluster). This | |
539 | * might involve allocating a new cluster for current CPU too. | |
540 | */ | |
36005bae | 541 | static bool scan_swap_map_try_ssd_cluster(struct swap_info_struct *si, |
ebc2a1a6 SL |
542 | unsigned long *offset, unsigned long *scan_base) |
543 | { | |
544 | struct percpu_cluster *cluster; | |
235b6217 | 545 | struct swap_cluster_info *ci; |
ebc2a1a6 | 546 | bool found_free; |
235b6217 | 547 | unsigned long tmp, max; |
ebc2a1a6 SL |
548 | |
549 | new_cluster: | |
550 | cluster = this_cpu_ptr(si->percpu_cluster); | |
551 | if (cluster_is_null(&cluster->index)) { | |
6b534915 HY |
552 | if (!cluster_list_empty(&si->free_clusters)) { |
553 | cluster->index = si->free_clusters.head; | |
ebc2a1a6 SL |
554 | cluster->next = cluster_next(&cluster->index) * |
555 | SWAPFILE_CLUSTER; | |
6b534915 | 556 | } else if (!cluster_list_empty(&si->discard_clusters)) { |
ebc2a1a6 SL |
557 | /* |
558 | * we don't have free cluster but have some clusters in | |
559 | * discarding, do discard now and reclaim them | |
560 | */ | |
561 | swap_do_scheduled_discard(si); | |
562 | *scan_base = *offset = si->cluster_next; | |
563 | goto new_cluster; | |
564 | } else | |
36005bae | 565 | return false; |
ebc2a1a6 SL |
566 | } |
567 | ||
568 | found_free = false; | |
569 | ||
570 | /* | |
571 | * Other CPUs can use our cluster if they can't find a free cluster, | |
572 | * check if there is still free entry in the cluster | |
573 | */ | |
574 | tmp = cluster->next; | |
235b6217 HY |
575 | max = min_t(unsigned long, si->max, |
576 | (cluster_next(&cluster->index) + 1) * SWAPFILE_CLUSTER); | |
577 | if (tmp >= max) { | |
578 | cluster_set_null(&cluster->index); | |
579 | goto new_cluster; | |
580 | } | |
581 | ci = lock_cluster(si, tmp); | |
582 | while (tmp < max) { | |
ebc2a1a6 SL |
583 | if (!si->swap_map[tmp]) { |
584 | found_free = true; | |
585 | break; | |
586 | } | |
587 | tmp++; | |
588 | } | |
235b6217 | 589 | unlock_cluster(ci); |
ebc2a1a6 SL |
590 | if (!found_free) { |
591 | cluster_set_null(&cluster->index); | |
592 | goto new_cluster; | |
593 | } | |
594 | cluster->next = tmp + 1; | |
595 | *offset = tmp; | |
596 | *scan_base = tmp; | |
36005bae | 597 | return found_free; |
2a8f9449 SL |
598 | } |
599 | ||
a2468cc9 AL |
600 | static void __del_from_avail_list(struct swap_info_struct *p) |
601 | { | |
602 | int nid; | |
603 | ||
604 | for_each_node(nid) | |
605 | plist_del(&p->avail_lists[nid], &swap_avail_heads[nid]); | |
606 | } | |
607 | ||
608 | static void del_from_avail_list(struct swap_info_struct *p) | |
609 | { | |
610 | spin_lock(&swap_avail_lock); | |
611 | __del_from_avail_list(p); | |
612 | spin_unlock(&swap_avail_lock); | |
613 | } | |
614 | ||
38d8b4e6 HY |
615 | static void swap_range_alloc(struct swap_info_struct *si, unsigned long offset, |
616 | unsigned int nr_entries) | |
617 | { | |
618 | unsigned int end = offset + nr_entries - 1; | |
619 | ||
620 | if (offset == si->lowest_bit) | |
621 | si->lowest_bit += nr_entries; | |
622 | if (end == si->highest_bit) | |
623 | si->highest_bit -= nr_entries; | |
624 | si->inuse_pages += nr_entries; | |
625 | if (si->inuse_pages == si->pages) { | |
626 | si->lowest_bit = si->max; | |
627 | si->highest_bit = 0; | |
a2468cc9 | 628 | del_from_avail_list(si); |
38d8b4e6 HY |
629 | } |
630 | } | |
631 | ||
a2468cc9 AL |
632 | static void add_to_avail_list(struct swap_info_struct *p) |
633 | { | |
634 | int nid; | |
635 | ||
636 | spin_lock(&swap_avail_lock); | |
637 | for_each_node(nid) { | |
638 | WARN_ON(!plist_node_empty(&p->avail_lists[nid])); | |
639 | plist_add(&p->avail_lists[nid], &swap_avail_heads[nid]); | |
640 | } | |
641 | spin_unlock(&swap_avail_lock); | |
642 | } | |
643 | ||
38d8b4e6 HY |
644 | static void swap_range_free(struct swap_info_struct *si, unsigned long offset, |
645 | unsigned int nr_entries) | |
646 | { | |
647 | unsigned long end = offset + nr_entries - 1; | |
648 | void (*swap_slot_free_notify)(struct block_device *, unsigned long); | |
649 | ||
650 | if (offset < si->lowest_bit) | |
651 | si->lowest_bit = offset; | |
652 | if (end > si->highest_bit) { | |
653 | bool was_full = !si->highest_bit; | |
654 | ||
655 | si->highest_bit = end; | |
a2468cc9 AL |
656 | if (was_full && (si->flags & SWP_WRITEOK)) |
657 | add_to_avail_list(si); | |
38d8b4e6 HY |
658 | } |
659 | atomic_long_add(nr_entries, &nr_swap_pages); | |
660 | si->inuse_pages -= nr_entries; | |
661 | if (si->flags & SWP_BLKDEV) | |
662 | swap_slot_free_notify = | |
663 | si->bdev->bd_disk->fops->swap_slot_free_notify; | |
664 | else | |
665 | swap_slot_free_notify = NULL; | |
666 | while (offset <= end) { | |
667 | frontswap_invalidate_page(si->type, offset); | |
668 | if (swap_slot_free_notify) | |
669 | swap_slot_free_notify(si->bdev, offset); | |
670 | offset++; | |
671 | } | |
672 | } | |
673 | ||
36005bae TC |
674 | static int scan_swap_map_slots(struct swap_info_struct *si, |
675 | unsigned char usage, int nr, | |
676 | swp_entry_t slots[]) | |
1da177e4 | 677 | { |
235b6217 | 678 | struct swap_cluster_info *ci; |
ebebbbe9 | 679 | unsigned long offset; |
c60aa176 | 680 | unsigned long scan_base; |
7992fde7 | 681 | unsigned long last_in_cluster = 0; |
048c27fd | 682 | int latency_ration = LATENCY_LIMIT; |
36005bae TC |
683 | int n_ret = 0; |
684 | ||
685 | if (nr > SWAP_BATCH) | |
686 | nr = SWAP_BATCH; | |
7dfad418 | 687 | |
886bb7e9 | 688 | /* |
7dfad418 HD |
689 | * We try to cluster swap pages by allocating them sequentially |
690 | * in swap. Once we've allocated SWAPFILE_CLUSTER pages this | |
691 | * way, however, we resort to first-free allocation, starting | |
692 | * a new cluster. This prevents us from scattering swap pages | |
693 | * all over the entire swap partition, so that we reduce | |
694 | * overall disk seek times between swap pages. -- sct | |
695 | * But we do now try to find an empty cluster. -Andrea | |
c60aa176 | 696 | * And we let swap pages go all over an SSD partition. Hugh |
7dfad418 HD |
697 | */ |
698 | ||
52b7efdb | 699 | si->flags += SWP_SCANNING; |
c60aa176 | 700 | scan_base = offset = si->cluster_next; |
ebebbbe9 | 701 | |
ebc2a1a6 SL |
702 | /* SSD algorithm */ |
703 | if (si->cluster_info) { | |
36005bae TC |
704 | if (scan_swap_map_try_ssd_cluster(si, &offset, &scan_base)) |
705 | goto checks; | |
706 | else | |
707 | goto scan; | |
ebc2a1a6 SL |
708 | } |
709 | ||
ebebbbe9 HD |
710 | if (unlikely(!si->cluster_nr--)) { |
711 | if (si->pages - si->inuse_pages < SWAPFILE_CLUSTER) { | |
712 | si->cluster_nr = SWAPFILE_CLUSTER - 1; | |
713 | goto checks; | |
714 | } | |
2a8f9449 | 715 | |
ec8acf20 | 716 | spin_unlock(&si->lock); |
7dfad418 | 717 | |
c60aa176 HD |
718 | /* |
719 | * If seek is expensive, start searching for new cluster from | |
720 | * start of partition, to minimize the span of allocated swap. | |
50088c44 CY |
721 | * If seek is cheap, that is the SWP_SOLIDSTATE si->cluster_info |
722 | * case, just handled by scan_swap_map_try_ssd_cluster() above. | |
c60aa176 | 723 | */ |
50088c44 | 724 | scan_base = offset = si->lowest_bit; |
7dfad418 HD |
725 | last_in_cluster = offset + SWAPFILE_CLUSTER - 1; |
726 | ||
727 | /* Locate the first empty (unaligned) cluster */ | |
728 | for (; last_in_cluster <= si->highest_bit; offset++) { | |
1da177e4 | 729 | if (si->swap_map[offset]) |
7dfad418 HD |
730 | last_in_cluster = offset + SWAPFILE_CLUSTER; |
731 | else if (offset == last_in_cluster) { | |
ec8acf20 | 732 | spin_lock(&si->lock); |
ebebbbe9 HD |
733 | offset -= SWAPFILE_CLUSTER - 1; |
734 | si->cluster_next = offset; | |
735 | si->cluster_nr = SWAPFILE_CLUSTER - 1; | |
c60aa176 HD |
736 | goto checks; |
737 | } | |
738 | if (unlikely(--latency_ration < 0)) { | |
739 | cond_resched(); | |
740 | latency_ration = LATENCY_LIMIT; | |
741 | } | |
742 | } | |
743 | ||
744 | offset = scan_base; | |
ec8acf20 | 745 | spin_lock(&si->lock); |
ebebbbe9 | 746 | si->cluster_nr = SWAPFILE_CLUSTER - 1; |
1da177e4 | 747 | } |
7dfad418 | 748 | |
ebebbbe9 | 749 | checks: |
ebc2a1a6 | 750 | if (si->cluster_info) { |
36005bae TC |
751 | while (scan_swap_map_ssd_cluster_conflict(si, offset)) { |
752 | /* take a break if we already got some slots */ | |
753 | if (n_ret) | |
754 | goto done; | |
755 | if (!scan_swap_map_try_ssd_cluster(si, &offset, | |
756 | &scan_base)) | |
757 | goto scan; | |
758 | } | |
ebc2a1a6 | 759 | } |
ebebbbe9 | 760 | if (!(si->flags & SWP_WRITEOK)) |
52b7efdb | 761 | goto no_page; |
7dfad418 HD |
762 | if (!si->highest_bit) |
763 | goto no_page; | |
ebebbbe9 | 764 | if (offset > si->highest_bit) |
c60aa176 | 765 | scan_base = offset = si->lowest_bit; |
c9e44410 | 766 | |
235b6217 | 767 | ci = lock_cluster(si, offset); |
b73d7fce HD |
768 | /* reuse swap entry of cache-only swap if not busy. */ |
769 | if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) { | |
c9e44410 | 770 | int swap_was_freed; |
235b6217 | 771 | unlock_cluster(ci); |
ec8acf20 | 772 | spin_unlock(&si->lock); |
c9e44410 | 773 | swap_was_freed = __try_to_reclaim_swap(si, offset); |
ec8acf20 | 774 | spin_lock(&si->lock); |
c9e44410 KH |
775 | /* entry was freed successfully, try to use this again */ |
776 | if (swap_was_freed) | |
777 | goto checks; | |
778 | goto scan; /* check next one */ | |
779 | } | |
780 | ||
235b6217 HY |
781 | if (si->swap_map[offset]) { |
782 | unlock_cluster(ci); | |
36005bae TC |
783 | if (!n_ret) |
784 | goto scan; | |
785 | else | |
786 | goto done; | |
235b6217 | 787 | } |
2872bb2d HY |
788 | si->swap_map[offset] = usage; |
789 | inc_cluster_info_page(si, si->cluster_info, offset); | |
790 | unlock_cluster(ci); | |
ebebbbe9 | 791 | |
38d8b4e6 | 792 | swap_range_alloc(si, offset, 1); |
ebebbbe9 | 793 | si->cluster_next = offset + 1; |
36005bae TC |
794 | slots[n_ret++] = swp_entry(si->type, offset); |
795 | ||
796 | /* got enough slots or reach max slots? */ | |
797 | if ((n_ret == nr) || (offset >= si->highest_bit)) | |
798 | goto done; | |
799 | ||
800 | /* search for next available slot */ | |
801 | ||
802 | /* time to take a break? */ | |
803 | if (unlikely(--latency_ration < 0)) { | |
804 | if (n_ret) | |
805 | goto done; | |
806 | spin_unlock(&si->lock); | |
807 | cond_resched(); | |
808 | spin_lock(&si->lock); | |
809 | latency_ration = LATENCY_LIMIT; | |
810 | } | |
811 | ||
812 | /* try to get more slots in cluster */ | |
813 | if (si->cluster_info) { | |
814 | if (scan_swap_map_try_ssd_cluster(si, &offset, &scan_base)) | |
815 | goto checks; | |
816 | else | |
817 | goto done; | |
818 | } | |
819 | /* non-ssd case */ | |
820 | ++offset; | |
821 | ||
822 | /* non-ssd case, still more slots in cluster? */ | |
823 | if (si->cluster_nr && !si->swap_map[offset]) { | |
824 | --si->cluster_nr; | |
825 | goto checks; | |
826 | } | |
7992fde7 | 827 | |
36005bae TC |
828 | done: |
829 | si->flags -= SWP_SCANNING; | |
830 | return n_ret; | |
7dfad418 | 831 | |
ebebbbe9 | 832 | scan: |
ec8acf20 | 833 | spin_unlock(&si->lock); |
7dfad418 | 834 | while (++offset <= si->highest_bit) { |
52b7efdb | 835 | if (!si->swap_map[offset]) { |
ec8acf20 | 836 | spin_lock(&si->lock); |
52b7efdb HD |
837 | goto checks; |
838 | } | |
c9e44410 | 839 | if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) { |
ec8acf20 | 840 | spin_lock(&si->lock); |
c9e44410 KH |
841 | goto checks; |
842 | } | |
048c27fd HD |
843 | if (unlikely(--latency_ration < 0)) { |
844 | cond_resched(); | |
845 | latency_ration = LATENCY_LIMIT; | |
846 | } | |
7dfad418 | 847 | } |
c60aa176 | 848 | offset = si->lowest_bit; |
a5998061 | 849 | while (offset < scan_base) { |
c60aa176 | 850 | if (!si->swap_map[offset]) { |
ec8acf20 | 851 | spin_lock(&si->lock); |
c60aa176 HD |
852 | goto checks; |
853 | } | |
c9e44410 | 854 | if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) { |
ec8acf20 | 855 | spin_lock(&si->lock); |
c9e44410 KH |
856 | goto checks; |
857 | } | |
c60aa176 HD |
858 | if (unlikely(--latency_ration < 0)) { |
859 | cond_resched(); | |
860 | latency_ration = LATENCY_LIMIT; | |
861 | } | |
a5998061 | 862 | offset++; |
c60aa176 | 863 | } |
ec8acf20 | 864 | spin_lock(&si->lock); |
7dfad418 HD |
865 | |
866 | no_page: | |
52b7efdb | 867 | si->flags -= SWP_SCANNING; |
36005bae | 868 | return n_ret; |
1da177e4 LT |
869 | } |
870 | ||
38d8b4e6 HY |
871 | static int swap_alloc_cluster(struct swap_info_struct *si, swp_entry_t *slot) |
872 | { | |
873 | unsigned long idx; | |
874 | struct swap_cluster_info *ci; | |
875 | unsigned long offset, i; | |
876 | unsigned char *map; | |
877 | ||
fe5266d5 HY |
878 | /* |
879 | * Should not even be attempting cluster allocations when huge | |
880 | * page swap is disabled. Warn and fail the allocation. | |
881 | */ | |
882 | if (!IS_ENABLED(CONFIG_THP_SWAP)) { | |
883 | VM_WARN_ON_ONCE(1); | |
884 | return 0; | |
885 | } | |
886 | ||
38d8b4e6 HY |
887 | if (cluster_list_empty(&si->free_clusters)) |
888 | return 0; | |
889 | ||
890 | idx = cluster_list_first(&si->free_clusters); | |
891 | offset = idx * SWAPFILE_CLUSTER; | |
892 | ci = lock_cluster(si, offset); | |
893 | alloc_cluster(si, idx); | |
e0709829 | 894 | cluster_set_count_flag(ci, SWAPFILE_CLUSTER, CLUSTER_FLAG_HUGE); |
38d8b4e6 HY |
895 | |
896 | map = si->swap_map + offset; | |
897 | for (i = 0; i < SWAPFILE_CLUSTER; i++) | |
898 | map[i] = SWAP_HAS_CACHE; | |
899 | unlock_cluster(ci); | |
900 | swap_range_alloc(si, offset, SWAPFILE_CLUSTER); | |
901 | *slot = swp_entry(si->type, offset); | |
902 | ||
903 | return 1; | |
904 | } | |
905 | ||
906 | static void swap_free_cluster(struct swap_info_struct *si, unsigned long idx) | |
907 | { | |
908 | unsigned long offset = idx * SWAPFILE_CLUSTER; | |
909 | struct swap_cluster_info *ci; | |
910 | ||
911 | ci = lock_cluster(si, offset); | |
912 | cluster_set_count_flag(ci, 0, 0); | |
913 | free_cluster(si, idx); | |
914 | unlock_cluster(ci); | |
915 | swap_range_free(si, offset, SWAPFILE_CLUSTER); | |
916 | } | |
38d8b4e6 | 917 | |
36005bae TC |
918 | static unsigned long scan_swap_map(struct swap_info_struct *si, |
919 | unsigned char usage) | |
920 | { | |
921 | swp_entry_t entry; | |
922 | int n_ret; | |
923 | ||
924 | n_ret = scan_swap_map_slots(si, usage, 1, &entry); | |
925 | ||
926 | if (n_ret) | |
927 | return swp_offset(entry); | |
928 | else | |
929 | return 0; | |
930 | ||
931 | } | |
932 | ||
38d8b4e6 | 933 | int get_swap_pages(int n_goal, bool cluster, swp_entry_t swp_entries[]) |
1da177e4 | 934 | { |
38d8b4e6 | 935 | unsigned long nr_pages = cluster ? SWAPFILE_CLUSTER : 1; |
adfab836 | 936 | struct swap_info_struct *si, *next; |
36005bae TC |
937 | long avail_pgs; |
938 | int n_ret = 0; | |
a2468cc9 | 939 | int node; |
1da177e4 | 940 | |
38d8b4e6 HY |
941 | /* Only single cluster request supported */ |
942 | WARN_ON_ONCE(n_goal > 1 && cluster); | |
943 | ||
944 | avail_pgs = atomic_long_read(&nr_swap_pages) / nr_pages; | |
36005bae | 945 | if (avail_pgs <= 0) |
fb4f88dc | 946 | goto noswap; |
36005bae TC |
947 | |
948 | if (n_goal > SWAP_BATCH) | |
949 | n_goal = SWAP_BATCH; | |
950 | ||
951 | if (n_goal > avail_pgs) | |
952 | n_goal = avail_pgs; | |
953 | ||
38d8b4e6 | 954 | atomic_long_sub(n_goal * nr_pages, &nr_swap_pages); |
fb4f88dc | 955 | |
18ab4d4c DS |
956 | spin_lock(&swap_avail_lock); |
957 | ||
958 | start_over: | |
a2468cc9 AL |
959 | node = numa_node_id(); |
960 | plist_for_each_entry_safe(si, next, &swap_avail_heads[node], avail_lists[node]) { | |
18ab4d4c | 961 | /* requeue si to after same-priority siblings */ |
a2468cc9 | 962 | plist_requeue(&si->avail_lists[node], &swap_avail_heads[node]); |
18ab4d4c | 963 | spin_unlock(&swap_avail_lock); |
ec8acf20 | 964 | spin_lock(&si->lock); |
adfab836 | 965 | if (!si->highest_bit || !(si->flags & SWP_WRITEOK)) { |
18ab4d4c | 966 | spin_lock(&swap_avail_lock); |
a2468cc9 | 967 | if (plist_node_empty(&si->avail_lists[node])) { |
18ab4d4c DS |
968 | spin_unlock(&si->lock); |
969 | goto nextsi; | |
970 | } | |
971 | WARN(!si->highest_bit, | |
972 | "swap_info %d in list but !highest_bit\n", | |
973 | si->type); | |
974 | WARN(!(si->flags & SWP_WRITEOK), | |
975 | "swap_info %d in list but !SWP_WRITEOK\n", | |
976 | si->type); | |
a2468cc9 | 977 | __del_from_avail_list(si); |
ec8acf20 | 978 | spin_unlock(&si->lock); |
18ab4d4c | 979 | goto nextsi; |
ec8acf20 | 980 | } |
f0eea189 HY |
981 | if (cluster) { |
982 | if (!(si->flags & SWP_FILE)) | |
983 | n_ret = swap_alloc_cluster(si, swp_entries); | |
984 | } else | |
38d8b4e6 HY |
985 | n_ret = scan_swap_map_slots(si, SWAP_HAS_CACHE, |
986 | n_goal, swp_entries); | |
ec8acf20 | 987 | spin_unlock(&si->lock); |
38d8b4e6 | 988 | if (n_ret || cluster) |
36005bae | 989 | goto check_out; |
18ab4d4c | 990 | pr_debug("scan_swap_map of si %d failed to find offset\n", |
36005bae TC |
991 | si->type); |
992 | ||
18ab4d4c DS |
993 | spin_lock(&swap_avail_lock); |
994 | nextsi: | |
adfab836 DS |
995 | /* |
996 | * if we got here, it's likely that si was almost full before, | |
997 | * and since scan_swap_map() can drop the si->lock, multiple | |
998 | * callers probably all tried to get a page from the same si | |
18ab4d4c DS |
999 | * and it filled up before we could get one; or, the si filled |
1000 | * up between us dropping swap_avail_lock and taking si->lock. | |
1001 | * Since we dropped the swap_avail_lock, the swap_avail_head | |
1002 | * list may have been modified; so if next is still in the | |
36005bae TC |
1003 | * swap_avail_head list then try it, otherwise start over |
1004 | * if we have not gotten any slots. | |
adfab836 | 1005 | */ |
a2468cc9 | 1006 | if (plist_node_empty(&next->avail_lists[node])) |
18ab4d4c | 1007 | goto start_over; |
1da177e4 | 1008 | } |
fb4f88dc | 1009 | |
18ab4d4c DS |
1010 | spin_unlock(&swap_avail_lock); |
1011 | ||
36005bae TC |
1012 | check_out: |
1013 | if (n_ret < n_goal) | |
38d8b4e6 HY |
1014 | atomic_long_add((long)(n_goal - n_ret) * nr_pages, |
1015 | &nr_swap_pages); | |
fb4f88dc | 1016 | noswap: |
36005bae TC |
1017 | return n_ret; |
1018 | } | |
1019 | ||
2de1a7e4 | 1020 | /* The only caller of this function is now suspend routine */ |
910321ea HD |
1021 | swp_entry_t get_swap_page_of_type(int type) |
1022 | { | |
1023 | struct swap_info_struct *si; | |
1024 | pgoff_t offset; | |
1025 | ||
910321ea | 1026 | si = swap_info[type]; |
ec8acf20 | 1027 | spin_lock(&si->lock); |
910321ea | 1028 | if (si && (si->flags & SWP_WRITEOK)) { |
ec8acf20 | 1029 | atomic_long_dec(&nr_swap_pages); |
910321ea HD |
1030 | /* This is called for allocating swap entry, not cache */ |
1031 | offset = scan_swap_map(si, 1); | |
1032 | if (offset) { | |
ec8acf20 | 1033 | spin_unlock(&si->lock); |
910321ea HD |
1034 | return swp_entry(type, offset); |
1035 | } | |
ec8acf20 | 1036 | atomic_long_inc(&nr_swap_pages); |
910321ea | 1037 | } |
ec8acf20 | 1038 | spin_unlock(&si->lock); |
910321ea HD |
1039 | return (swp_entry_t) {0}; |
1040 | } | |
1041 | ||
e8c26ab6 | 1042 | static struct swap_info_struct *__swap_info_get(swp_entry_t entry) |
1da177e4 | 1043 | { |
73c34b6a | 1044 | struct swap_info_struct *p; |
1da177e4 LT |
1045 | unsigned long offset, type; |
1046 | ||
1047 | if (!entry.val) | |
1048 | goto out; | |
1049 | type = swp_type(entry); | |
1050 | if (type >= nr_swapfiles) | |
1051 | goto bad_nofile; | |
efa90a98 | 1052 | p = swap_info[type]; |
1da177e4 LT |
1053 | if (!(p->flags & SWP_USED)) |
1054 | goto bad_device; | |
1055 | offset = swp_offset(entry); | |
1056 | if (offset >= p->max) | |
1057 | goto bad_offset; | |
1da177e4 LT |
1058 | return p; |
1059 | ||
1da177e4 | 1060 | bad_offset: |
6a991fc7 | 1061 | pr_err("swap_info_get: %s%08lx\n", Bad_offset, entry.val); |
1da177e4 LT |
1062 | goto out; |
1063 | bad_device: | |
6a991fc7 | 1064 | pr_err("swap_info_get: %s%08lx\n", Unused_file, entry.val); |
1da177e4 LT |
1065 | goto out; |
1066 | bad_nofile: | |
6a991fc7 | 1067 | pr_err("swap_info_get: %s%08lx\n", Bad_file, entry.val); |
1da177e4 LT |
1068 | out: |
1069 | return NULL; | |
886bb7e9 | 1070 | } |
1da177e4 | 1071 | |
e8c26ab6 TC |
1072 | static struct swap_info_struct *_swap_info_get(swp_entry_t entry) |
1073 | { | |
1074 | struct swap_info_struct *p; | |
1075 | ||
1076 | p = __swap_info_get(entry); | |
1077 | if (!p) | |
1078 | goto out; | |
1079 | if (!p->swap_map[swp_offset(entry)]) | |
1080 | goto bad_free; | |
1081 | return p; | |
1082 | ||
1083 | bad_free: | |
1084 | pr_err("swap_info_get: %s%08lx\n", Unused_offset, entry.val); | |
1085 | goto out; | |
1086 | out: | |
1087 | return NULL; | |
1088 | } | |
1089 | ||
235b6217 HY |
1090 | static struct swap_info_struct *swap_info_get(swp_entry_t entry) |
1091 | { | |
1092 | struct swap_info_struct *p; | |
1093 | ||
1094 | p = _swap_info_get(entry); | |
1095 | if (p) | |
1096 | spin_lock(&p->lock); | |
1097 | return p; | |
1098 | } | |
1099 | ||
7c00bafe TC |
1100 | static struct swap_info_struct *swap_info_get_cont(swp_entry_t entry, |
1101 | struct swap_info_struct *q) | |
1102 | { | |
1103 | struct swap_info_struct *p; | |
1104 | ||
1105 | p = _swap_info_get(entry); | |
1106 | ||
1107 | if (p != q) { | |
1108 | if (q != NULL) | |
1109 | spin_unlock(&q->lock); | |
1110 | if (p != NULL) | |
1111 | spin_lock(&p->lock); | |
1112 | } | |
1113 | return p; | |
1114 | } | |
1115 | ||
1116 | static unsigned char __swap_entry_free(struct swap_info_struct *p, | |
1117 | swp_entry_t entry, unsigned char usage) | |
1da177e4 | 1118 | { |
235b6217 | 1119 | struct swap_cluster_info *ci; |
253d553b | 1120 | unsigned long offset = swp_offset(entry); |
8d69aaee HD |
1121 | unsigned char count; |
1122 | unsigned char has_cache; | |
235b6217 | 1123 | |
7c00bafe | 1124 | ci = lock_cluster_or_swap_info(p, offset); |
355cfa73 | 1125 | |
253d553b | 1126 | count = p->swap_map[offset]; |
235b6217 | 1127 | |
253d553b HD |
1128 | has_cache = count & SWAP_HAS_CACHE; |
1129 | count &= ~SWAP_HAS_CACHE; | |
355cfa73 | 1130 | |
253d553b | 1131 | if (usage == SWAP_HAS_CACHE) { |
355cfa73 | 1132 | VM_BUG_ON(!has_cache); |
253d553b | 1133 | has_cache = 0; |
aaa46865 HD |
1134 | } else if (count == SWAP_MAP_SHMEM) { |
1135 | /* | |
1136 | * Or we could insist on shmem.c using a special | |
1137 | * swap_shmem_free() and free_shmem_swap_and_cache()... | |
1138 | */ | |
1139 | count = 0; | |
570a335b HD |
1140 | } else if ((count & ~COUNT_CONTINUED) <= SWAP_MAP_MAX) { |
1141 | if (count == COUNT_CONTINUED) { | |
1142 | if (swap_count_continued(p, offset, count)) | |
1143 | count = SWAP_MAP_MAX | COUNT_CONTINUED; | |
1144 | else | |
1145 | count = SWAP_MAP_MAX; | |
1146 | } else | |
1147 | count--; | |
1148 | } | |
253d553b | 1149 | |
253d553b | 1150 | usage = count | has_cache; |
7c00bafe TC |
1151 | p->swap_map[offset] = usage ? : SWAP_HAS_CACHE; |
1152 | ||
1153 | unlock_cluster_or_swap_info(p, ci); | |
1154 | ||
1155 | return usage; | |
1156 | } | |
355cfa73 | 1157 | |
7c00bafe TC |
1158 | static void swap_entry_free(struct swap_info_struct *p, swp_entry_t entry) |
1159 | { | |
1160 | struct swap_cluster_info *ci; | |
1161 | unsigned long offset = swp_offset(entry); | |
1162 | unsigned char count; | |
1163 | ||
1164 | ci = lock_cluster(p, offset); | |
1165 | count = p->swap_map[offset]; | |
1166 | VM_BUG_ON(count != SWAP_HAS_CACHE); | |
1167 | p->swap_map[offset] = 0; | |
1168 | dec_cluster_info_page(p, p->cluster_info, offset); | |
235b6217 HY |
1169 | unlock_cluster(ci); |
1170 | ||
38d8b4e6 HY |
1171 | mem_cgroup_uncharge_swap(entry, 1); |
1172 | swap_range_free(p, offset, 1); | |
1da177e4 LT |
1173 | } |
1174 | ||
1175 | /* | |
2de1a7e4 | 1176 | * Caller has made sure that the swap device corresponding to entry |
1da177e4 LT |
1177 | * is still around or has not been recycled. |
1178 | */ | |
1179 | void swap_free(swp_entry_t entry) | |
1180 | { | |
73c34b6a | 1181 | struct swap_info_struct *p; |
1da177e4 | 1182 | |
235b6217 | 1183 | p = _swap_info_get(entry); |
7c00bafe TC |
1184 | if (p) { |
1185 | if (!__swap_entry_free(p, entry, 1)) | |
67afa38e | 1186 | free_swap_slot(entry); |
7c00bafe | 1187 | } |
1da177e4 LT |
1188 | } |
1189 | ||
cb4b86ba KH |
1190 | /* |
1191 | * Called after dropping swapcache to decrease refcnt to swap entries. | |
1192 | */ | |
75f6d6d2 | 1193 | static void swapcache_free(swp_entry_t entry) |
cb4b86ba | 1194 | { |
355cfa73 KH |
1195 | struct swap_info_struct *p; |
1196 | ||
235b6217 | 1197 | p = _swap_info_get(entry); |
7c00bafe TC |
1198 | if (p) { |
1199 | if (!__swap_entry_free(p, entry, SWAP_HAS_CACHE)) | |
67afa38e | 1200 | free_swap_slot(entry); |
7c00bafe TC |
1201 | } |
1202 | } | |
1203 | ||
75f6d6d2 | 1204 | static void swapcache_free_cluster(swp_entry_t entry) |
38d8b4e6 HY |
1205 | { |
1206 | unsigned long offset = swp_offset(entry); | |
1207 | unsigned long idx = offset / SWAPFILE_CLUSTER; | |
1208 | struct swap_cluster_info *ci; | |
1209 | struct swap_info_struct *si; | |
1210 | unsigned char *map; | |
a3aea839 HY |
1211 | unsigned int i, free_entries = 0; |
1212 | unsigned char val; | |
38d8b4e6 | 1213 | |
fe5266d5 HY |
1214 | if (!IS_ENABLED(CONFIG_THP_SWAP)) |
1215 | return; | |
1216 | ||
a3aea839 | 1217 | si = _swap_info_get(entry); |
38d8b4e6 HY |
1218 | if (!si) |
1219 | return; | |
1220 | ||
1221 | ci = lock_cluster(si, offset); | |
e0709829 | 1222 | VM_BUG_ON(!cluster_is_huge(ci)); |
38d8b4e6 HY |
1223 | map = si->swap_map + offset; |
1224 | for (i = 0; i < SWAPFILE_CLUSTER; i++) { | |
a3aea839 HY |
1225 | val = map[i]; |
1226 | VM_BUG_ON(!(val & SWAP_HAS_CACHE)); | |
1227 | if (val == SWAP_HAS_CACHE) | |
1228 | free_entries++; | |
1229 | } | |
1230 | if (!free_entries) { | |
1231 | for (i = 0; i < SWAPFILE_CLUSTER; i++) | |
1232 | map[i] &= ~SWAP_HAS_CACHE; | |
38d8b4e6 | 1233 | } |
e0709829 | 1234 | cluster_clear_huge(ci); |
38d8b4e6 | 1235 | unlock_cluster(ci); |
a3aea839 HY |
1236 | if (free_entries == SWAPFILE_CLUSTER) { |
1237 | spin_lock(&si->lock); | |
1238 | ci = lock_cluster(si, offset); | |
1239 | memset(map, 0, SWAPFILE_CLUSTER); | |
1240 | unlock_cluster(ci); | |
1241 | mem_cgroup_uncharge_swap(entry, SWAPFILE_CLUSTER); | |
1242 | swap_free_cluster(si, idx); | |
1243 | spin_unlock(&si->lock); | |
1244 | } else if (free_entries) { | |
1245 | for (i = 0; i < SWAPFILE_CLUSTER; i++, entry.val++) { | |
1246 | if (!__swap_entry_free(si, entry, SWAP_HAS_CACHE)) | |
1247 | free_swap_slot(entry); | |
1248 | } | |
1249 | } | |
38d8b4e6 | 1250 | } |
59807685 | 1251 | |
fe5266d5 | 1252 | #ifdef CONFIG_THP_SWAP |
59807685 HY |
1253 | int split_swap_cluster(swp_entry_t entry) |
1254 | { | |
1255 | struct swap_info_struct *si; | |
1256 | struct swap_cluster_info *ci; | |
1257 | unsigned long offset = swp_offset(entry); | |
1258 | ||
1259 | si = _swap_info_get(entry); | |
1260 | if (!si) | |
1261 | return -EBUSY; | |
1262 | ci = lock_cluster(si, offset); | |
1263 | cluster_clear_huge(ci); | |
1264 | unlock_cluster(ci); | |
1265 | return 0; | |
1266 | } | |
fe5266d5 | 1267 | #endif |
38d8b4e6 | 1268 | |
75f6d6d2 MK |
1269 | void put_swap_page(struct page *page, swp_entry_t entry) |
1270 | { | |
1271 | if (!PageTransHuge(page)) | |
1272 | swapcache_free(entry); | |
1273 | else | |
1274 | swapcache_free_cluster(entry); | |
1275 | } | |
1276 | ||
155b5f88 HY |
1277 | static int swp_entry_cmp(const void *ent1, const void *ent2) |
1278 | { | |
1279 | const swp_entry_t *e1 = ent1, *e2 = ent2; | |
1280 | ||
1281 | return (int)swp_type(*e1) - (int)swp_type(*e2); | |
1282 | } | |
1283 | ||
7c00bafe TC |
1284 | void swapcache_free_entries(swp_entry_t *entries, int n) |
1285 | { | |
1286 | struct swap_info_struct *p, *prev; | |
1287 | int i; | |
1288 | ||
1289 | if (n <= 0) | |
1290 | return; | |
1291 | ||
1292 | prev = NULL; | |
1293 | p = NULL; | |
155b5f88 HY |
1294 | |
1295 | /* | |
1296 | * Sort swap entries by swap device, so each lock is only taken once. | |
1297 | * nr_swapfiles isn't absolutely correct, but the overhead of sort() is | |
1298 | * so low that it isn't necessary to optimize further. | |
1299 | */ | |
1300 | if (nr_swapfiles > 1) | |
1301 | sort(entries, n, sizeof(entries[0]), swp_entry_cmp, NULL); | |
7c00bafe TC |
1302 | for (i = 0; i < n; ++i) { |
1303 | p = swap_info_get_cont(entries[i], prev); | |
1304 | if (p) | |
1305 | swap_entry_free(p, entries[i]); | |
7c00bafe TC |
1306 | prev = p; |
1307 | } | |
235b6217 | 1308 | if (p) |
7c00bafe | 1309 | spin_unlock(&p->lock); |
cb4b86ba KH |
1310 | } |
1311 | ||
1da177e4 | 1312 | /* |
c475a8ab | 1313 | * How many references to page are currently swapped out? |
570a335b HD |
1314 | * This does not give an exact answer when swap count is continued, |
1315 | * but does include the high COUNT_CONTINUED flag to allow for that. | |
1da177e4 | 1316 | */ |
bde05d1c | 1317 | int page_swapcount(struct page *page) |
1da177e4 | 1318 | { |
c475a8ab HD |
1319 | int count = 0; |
1320 | struct swap_info_struct *p; | |
235b6217 | 1321 | struct swap_cluster_info *ci; |
1da177e4 | 1322 | swp_entry_t entry; |
235b6217 | 1323 | unsigned long offset; |
1da177e4 | 1324 | |
4c21e2f2 | 1325 | entry.val = page_private(page); |
235b6217 | 1326 | p = _swap_info_get(entry); |
1da177e4 | 1327 | if (p) { |
235b6217 HY |
1328 | offset = swp_offset(entry); |
1329 | ci = lock_cluster_or_swap_info(p, offset); | |
1330 | count = swap_count(p->swap_map[offset]); | |
1331 | unlock_cluster_or_swap_info(p, ci); | |
1da177e4 | 1332 | } |
c475a8ab | 1333 | return count; |
1da177e4 LT |
1334 | } |
1335 | ||
aa8d22a1 MK |
1336 | int __swap_count(struct swap_info_struct *si, swp_entry_t entry) |
1337 | { | |
1338 | pgoff_t offset = swp_offset(entry); | |
1339 | ||
1340 | return swap_count(si->swap_map[offset]); | |
1341 | } | |
1342 | ||
322b8afe HY |
1343 | static int swap_swapcount(struct swap_info_struct *si, swp_entry_t entry) |
1344 | { | |
1345 | int count = 0; | |
1346 | pgoff_t offset = swp_offset(entry); | |
1347 | struct swap_cluster_info *ci; | |
1348 | ||
1349 | ci = lock_cluster_or_swap_info(si, offset); | |
1350 | count = swap_count(si->swap_map[offset]); | |
1351 | unlock_cluster_or_swap_info(si, ci); | |
1352 | return count; | |
1353 | } | |
1354 | ||
e8c26ab6 TC |
1355 | /* |
1356 | * How many references to @entry are currently swapped out? | |
1357 | * This does not give an exact answer when swap count is continued, | |
1358 | * but does include the high COUNT_CONTINUED flag to allow for that. | |
1359 | */ | |
1360 | int __swp_swapcount(swp_entry_t entry) | |
1361 | { | |
1362 | int count = 0; | |
e8c26ab6 | 1363 | struct swap_info_struct *si; |
e8c26ab6 TC |
1364 | |
1365 | si = __swap_info_get(entry); | |
322b8afe HY |
1366 | if (si) |
1367 | count = swap_swapcount(si, entry); | |
e8c26ab6 TC |
1368 | return count; |
1369 | } | |
1370 | ||
8334b962 MK |
1371 | /* |
1372 | * How many references to @entry are currently swapped out? | |
1373 | * This considers COUNT_CONTINUED so it returns exact answer. | |
1374 | */ | |
1375 | int swp_swapcount(swp_entry_t entry) | |
1376 | { | |
1377 | int count, tmp_count, n; | |
1378 | struct swap_info_struct *p; | |
235b6217 | 1379 | struct swap_cluster_info *ci; |
8334b962 MK |
1380 | struct page *page; |
1381 | pgoff_t offset; | |
1382 | unsigned char *map; | |
1383 | ||
235b6217 | 1384 | p = _swap_info_get(entry); |
8334b962 MK |
1385 | if (!p) |
1386 | return 0; | |
1387 | ||
235b6217 HY |
1388 | offset = swp_offset(entry); |
1389 | ||
1390 | ci = lock_cluster_or_swap_info(p, offset); | |
1391 | ||
1392 | count = swap_count(p->swap_map[offset]); | |
8334b962 MK |
1393 | if (!(count & COUNT_CONTINUED)) |
1394 | goto out; | |
1395 | ||
1396 | count &= ~COUNT_CONTINUED; | |
1397 | n = SWAP_MAP_MAX + 1; | |
1398 | ||
8334b962 MK |
1399 | page = vmalloc_to_page(p->swap_map + offset); |
1400 | offset &= ~PAGE_MASK; | |
1401 | VM_BUG_ON(page_private(page) != SWP_CONTINUED); | |
1402 | ||
1403 | do { | |
a8ae4991 | 1404 | page = list_next_entry(page, lru); |
8334b962 MK |
1405 | map = kmap_atomic(page); |
1406 | tmp_count = map[offset]; | |
1407 | kunmap_atomic(map); | |
1408 | ||
1409 | count += (tmp_count & ~COUNT_CONTINUED) * n; | |
1410 | n *= (SWAP_CONT_MAX + 1); | |
1411 | } while (tmp_count & COUNT_CONTINUED); | |
1412 | out: | |
235b6217 | 1413 | unlock_cluster_or_swap_info(p, ci); |
8334b962 MK |
1414 | return count; |
1415 | } | |
1416 | ||
e0709829 HY |
1417 | static bool swap_page_trans_huge_swapped(struct swap_info_struct *si, |
1418 | swp_entry_t entry) | |
1419 | { | |
1420 | struct swap_cluster_info *ci; | |
1421 | unsigned char *map = si->swap_map; | |
1422 | unsigned long roffset = swp_offset(entry); | |
1423 | unsigned long offset = round_down(roffset, SWAPFILE_CLUSTER); | |
1424 | int i; | |
1425 | bool ret = false; | |
1426 | ||
fe5266d5 HY |
1427 | if (!IS_ENABLED(CONFIG_THP_SWAP)) |
1428 | return swap_swapcount(si, entry) != 0; | |
1429 | ||
e0709829 HY |
1430 | ci = lock_cluster_or_swap_info(si, offset); |
1431 | if (!ci || !cluster_is_huge(ci)) { | |
1432 | if (map[roffset] != SWAP_HAS_CACHE) | |
1433 | ret = true; | |
1434 | goto unlock_out; | |
1435 | } | |
1436 | for (i = 0; i < SWAPFILE_CLUSTER; i++) { | |
1437 | if (map[offset + i] != SWAP_HAS_CACHE) { | |
1438 | ret = true; | |
1439 | break; | |
1440 | } | |
1441 | } | |
1442 | unlock_out: | |
1443 | unlock_cluster_or_swap_info(si, ci); | |
1444 | return ret; | |
1445 | } | |
1446 | ||
1447 | static bool page_swapped(struct page *page) | |
1448 | { | |
1449 | swp_entry_t entry; | |
1450 | struct swap_info_struct *si; | |
1451 | ||
fe5266d5 | 1452 | if (!IS_ENABLED(CONFIG_THP_SWAP) || likely(!PageTransCompound(page))) |
e0709829 HY |
1453 | return page_swapcount(page) != 0; |
1454 | ||
1455 | page = compound_head(page); | |
1456 | entry.val = page_private(page); | |
1457 | si = _swap_info_get(entry); | |
1458 | if (si) | |
1459 | return swap_page_trans_huge_swapped(si, entry); | |
1460 | return false; | |
1461 | } | |
ba3c4ce6 HY |
1462 | |
1463 | static int page_trans_huge_map_swapcount(struct page *page, int *total_mapcount, | |
1464 | int *total_swapcount) | |
1465 | { | |
1466 | int i, map_swapcount, _total_mapcount, _total_swapcount; | |
1467 | unsigned long offset = 0; | |
1468 | struct swap_info_struct *si; | |
1469 | struct swap_cluster_info *ci = NULL; | |
1470 | unsigned char *map = NULL; | |
1471 | int mapcount, swapcount = 0; | |
1472 | ||
1473 | /* hugetlbfs shouldn't call it */ | |
1474 | VM_BUG_ON_PAGE(PageHuge(page), page); | |
1475 | ||
fe5266d5 HY |
1476 | if (!IS_ENABLED(CONFIG_THP_SWAP) || likely(!PageTransCompound(page))) { |
1477 | mapcount = page_trans_huge_mapcount(page, total_mapcount); | |
ba3c4ce6 HY |
1478 | if (PageSwapCache(page)) |
1479 | swapcount = page_swapcount(page); | |
1480 | if (total_swapcount) | |
1481 | *total_swapcount = swapcount; | |
1482 | return mapcount + swapcount; | |
1483 | } | |
1484 | ||
1485 | page = compound_head(page); | |
1486 | ||
1487 | _total_mapcount = _total_swapcount = map_swapcount = 0; | |
1488 | if (PageSwapCache(page)) { | |
1489 | swp_entry_t entry; | |
1490 | ||
1491 | entry.val = page_private(page); | |
1492 | si = _swap_info_get(entry); | |
1493 | if (si) { | |
1494 | map = si->swap_map; | |
1495 | offset = swp_offset(entry); | |
1496 | } | |
1497 | } | |
1498 | if (map) | |
1499 | ci = lock_cluster(si, offset); | |
1500 | for (i = 0; i < HPAGE_PMD_NR; i++) { | |
1501 | mapcount = atomic_read(&page[i]._mapcount) + 1; | |
1502 | _total_mapcount += mapcount; | |
1503 | if (map) { | |
1504 | swapcount = swap_count(map[offset + i]); | |
1505 | _total_swapcount += swapcount; | |
1506 | } | |
1507 | map_swapcount = max(map_swapcount, mapcount + swapcount); | |
1508 | } | |
1509 | unlock_cluster(ci); | |
1510 | if (PageDoubleMap(page)) { | |
1511 | map_swapcount -= 1; | |
1512 | _total_mapcount -= HPAGE_PMD_NR; | |
1513 | } | |
1514 | mapcount = compound_mapcount(page); | |
1515 | map_swapcount += mapcount; | |
1516 | _total_mapcount += mapcount; | |
1517 | if (total_mapcount) | |
1518 | *total_mapcount = _total_mapcount; | |
1519 | if (total_swapcount) | |
1520 | *total_swapcount = _total_swapcount; | |
1521 | ||
1522 | return map_swapcount; | |
1523 | } | |
e0709829 | 1524 | |
1da177e4 | 1525 | /* |
7b1fe597 HD |
1526 | * We can write to an anon page without COW if there are no other references |
1527 | * to it. And as a side-effect, free up its swap: because the old content | |
1528 | * on disk will never be read, and seeking back there to write new content | |
1529 | * later would only waste time away from clustering. | |
6d0a07ed | 1530 | * |
ba3c4ce6 | 1531 | * NOTE: total_map_swapcount should not be relied upon by the caller if |
6d0a07ed AA |
1532 | * reuse_swap_page() returns false, but it may be always overwritten |
1533 | * (see the other implementation for CONFIG_SWAP=n). | |
1da177e4 | 1534 | */ |
ba3c4ce6 | 1535 | bool reuse_swap_page(struct page *page, int *total_map_swapcount) |
1da177e4 | 1536 | { |
ba3c4ce6 | 1537 | int count, total_mapcount, total_swapcount; |
c475a8ab | 1538 | |
309381fe | 1539 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
5ad64688 | 1540 | if (unlikely(PageKsm(page))) |
6d0a07ed | 1541 | return false; |
ba3c4ce6 HY |
1542 | count = page_trans_huge_map_swapcount(page, &total_mapcount, |
1543 | &total_swapcount); | |
1544 | if (total_map_swapcount) | |
1545 | *total_map_swapcount = total_mapcount + total_swapcount; | |
1546 | if (count == 1 && PageSwapCache(page) && | |
1547 | (likely(!PageTransCompound(page)) || | |
1548 | /* The remaining swap count will be freed soon */ | |
1549 | total_swapcount == page_swapcount(page))) { | |
f0571429 | 1550 | if (!PageWriteback(page)) { |
ba3c4ce6 | 1551 | page = compound_head(page); |
7b1fe597 HD |
1552 | delete_from_swap_cache(page); |
1553 | SetPageDirty(page); | |
f0571429 MK |
1554 | } else { |
1555 | swp_entry_t entry; | |
1556 | struct swap_info_struct *p; | |
1557 | ||
1558 | entry.val = page_private(page); | |
1559 | p = swap_info_get(entry); | |
1560 | if (p->flags & SWP_STABLE_WRITES) { | |
1561 | spin_unlock(&p->lock); | |
1562 | return false; | |
1563 | } | |
1564 | spin_unlock(&p->lock); | |
7b1fe597 HD |
1565 | } |
1566 | } | |
ba3c4ce6 | 1567 | |
5ad64688 | 1568 | return count <= 1; |
1da177e4 LT |
1569 | } |
1570 | ||
1571 | /* | |
a2c43eed HD |
1572 | * If swap is getting full, or if there are no more mappings of this page, |
1573 | * then try_to_free_swap is called to free its swap space. | |
1da177e4 | 1574 | */ |
a2c43eed | 1575 | int try_to_free_swap(struct page *page) |
1da177e4 | 1576 | { |
309381fe | 1577 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
1da177e4 LT |
1578 | |
1579 | if (!PageSwapCache(page)) | |
1580 | return 0; | |
1581 | if (PageWriteback(page)) | |
1582 | return 0; | |
e0709829 | 1583 | if (page_swapped(page)) |
1da177e4 LT |
1584 | return 0; |
1585 | ||
b73d7fce HD |
1586 | /* |
1587 | * Once hibernation has begun to create its image of memory, | |
1588 | * there's a danger that one of the calls to try_to_free_swap() | |
1589 | * - most probably a call from __try_to_reclaim_swap() while | |
1590 | * hibernation is allocating its own swap pages for the image, | |
1591 | * but conceivably even a call from memory reclaim - will free | |
1592 | * the swap from a page which has already been recorded in the | |
1593 | * image as a clean swapcache page, and then reuse its swap for | |
1594 | * another page of the image. On waking from hibernation, the | |
1595 | * original page might be freed under memory pressure, then | |
1596 | * later read back in from swap, now with the wrong data. | |
1597 | * | |
2de1a7e4 | 1598 | * Hibernation suspends storage while it is writing the image |
f90ac398 | 1599 | * to disk so check that here. |
b73d7fce | 1600 | */ |
f90ac398 | 1601 | if (pm_suspended_storage()) |
b73d7fce HD |
1602 | return 0; |
1603 | ||
e0709829 | 1604 | page = compound_head(page); |
a2c43eed HD |
1605 | delete_from_swap_cache(page); |
1606 | SetPageDirty(page); | |
1607 | return 1; | |
68a22394 RR |
1608 | } |
1609 | ||
1da177e4 LT |
1610 | /* |
1611 | * Free the swap entry like above, but also try to | |
1612 | * free the page cache entry if it is the last user. | |
1613 | */ | |
2509ef26 | 1614 | int free_swap_and_cache(swp_entry_t entry) |
1da177e4 | 1615 | { |
2509ef26 | 1616 | struct swap_info_struct *p; |
1da177e4 | 1617 | struct page *page = NULL; |
7c00bafe | 1618 | unsigned char count; |
1da177e4 | 1619 | |
a7420aa5 | 1620 | if (non_swap_entry(entry)) |
2509ef26 | 1621 | return 1; |
0697212a | 1622 | |
7c00bafe | 1623 | p = _swap_info_get(entry); |
1da177e4 | 1624 | if (p) { |
7c00bafe | 1625 | count = __swap_entry_free(p, entry, 1); |
e0709829 HY |
1626 | if (count == SWAP_HAS_CACHE && |
1627 | !swap_page_trans_huge_swapped(p, entry)) { | |
33806f06 | 1628 | page = find_get_page(swap_address_space(entry), |
f6ab1f7f | 1629 | swp_offset(entry)); |
8413ac9d | 1630 | if (page && !trylock_page(page)) { |
09cbfeaf | 1631 | put_page(page); |
93fac704 NP |
1632 | page = NULL; |
1633 | } | |
7c00bafe | 1634 | } else if (!count) |
67afa38e | 1635 | free_swap_slot(entry); |
1da177e4 LT |
1636 | } |
1637 | if (page) { | |
a2c43eed HD |
1638 | /* |
1639 | * Not mapped elsewhere, or swap space full? Free it! | |
1640 | * Also recheck PageSwapCache now page is locked (above). | |
1641 | */ | |
93fac704 | 1642 | if (PageSwapCache(page) && !PageWriteback(page) && |
322b8afe | 1643 | (!page_mapped(page) || mem_cgroup_swap_full(page)) && |
e0709829 HY |
1644 | !swap_page_trans_huge_swapped(p, entry)) { |
1645 | page = compound_head(page); | |
1da177e4 LT |
1646 | delete_from_swap_cache(page); |
1647 | SetPageDirty(page); | |
1648 | } | |
1649 | unlock_page(page); | |
09cbfeaf | 1650 | put_page(page); |
1da177e4 | 1651 | } |
2509ef26 | 1652 | return p != NULL; |
1da177e4 LT |
1653 | } |
1654 | ||
b0cb1a19 | 1655 | #ifdef CONFIG_HIBERNATION |
f577eb30 | 1656 | /* |
915bae9e | 1657 | * Find the swap type that corresponds to given device (if any). |
f577eb30 | 1658 | * |
915bae9e RW |
1659 | * @offset - number of the PAGE_SIZE-sized block of the device, starting |
1660 | * from 0, in which the swap header is expected to be located. | |
1661 | * | |
1662 | * This is needed for the suspend to disk (aka swsusp). | |
f577eb30 | 1663 | */ |
7bf23687 | 1664 | int swap_type_of(dev_t device, sector_t offset, struct block_device **bdev_p) |
f577eb30 | 1665 | { |
915bae9e | 1666 | struct block_device *bdev = NULL; |
efa90a98 | 1667 | int type; |
f577eb30 | 1668 | |
915bae9e RW |
1669 | if (device) |
1670 | bdev = bdget(device); | |
1671 | ||
f577eb30 | 1672 | spin_lock(&swap_lock); |
efa90a98 HD |
1673 | for (type = 0; type < nr_swapfiles; type++) { |
1674 | struct swap_info_struct *sis = swap_info[type]; | |
f577eb30 | 1675 | |
915bae9e | 1676 | if (!(sis->flags & SWP_WRITEOK)) |
f577eb30 | 1677 | continue; |
b6b5bce3 | 1678 | |
915bae9e | 1679 | if (!bdev) { |
7bf23687 | 1680 | if (bdev_p) |
dddac6a7 | 1681 | *bdev_p = bdgrab(sis->bdev); |
7bf23687 | 1682 | |
6e1819d6 | 1683 | spin_unlock(&swap_lock); |
efa90a98 | 1684 | return type; |
6e1819d6 | 1685 | } |
915bae9e | 1686 | if (bdev == sis->bdev) { |
9625a5f2 | 1687 | struct swap_extent *se = &sis->first_swap_extent; |
915bae9e | 1688 | |
915bae9e | 1689 | if (se->start_block == offset) { |
7bf23687 | 1690 | if (bdev_p) |
dddac6a7 | 1691 | *bdev_p = bdgrab(sis->bdev); |
7bf23687 | 1692 | |
915bae9e RW |
1693 | spin_unlock(&swap_lock); |
1694 | bdput(bdev); | |
efa90a98 | 1695 | return type; |
915bae9e | 1696 | } |
f577eb30 RW |
1697 | } |
1698 | } | |
1699 | spin_unlock(&swap_lock); | |
915bae9e RW |
1700 | if (bdev) |
1701 | bdput(bdev); | |
1702 | ||
f577eb30 RW |
1703 | return -ENODEV; |
1704 | } | |
1705 | ||
73c34b6a HD |
1706 | /* |
1707 | * Get the (PAGE_SIZE) block corresponding to given offset on the swapdev | |
1708 | * corresponding to given index in swap_info (swap type). | |
1709 | */ | |
1710 | sector_t swapdev_block(int type, pgoff_t offset) | |
1711 | { | |
1712 | struct block_device *bdev; | |
1713 | ||
1714 | if ((unsigned int)type >= nr_swapfiles) | |
1715 | return 0; | |
1716 | if (!(swap_info[type]->flags & SWP_WRITEOK)) | |
1717 | return 0; | |
d4906e1a | 1718 | return map_swap_entry(swp_entry(type, offset), &bdev); |
73c34b6a HD |
1719 | } |
1720 | ||
f577eb30 RW |
1721 | /* |
1722 | * Return either the total number of swap pages of given type, or the number | |
1723 | * of free pages of that type (depending on @free) | |
1724 | * | |
1725 | * This is needed for software suspend | |
1726 | */ | |
1727 | unsigned int count_swap_pages(int type, int free) | |
1728 | { | |
1729 | unsigned int n = 0; | |
1730 | ||
efa90a98 HD |
1731 | spin_lock(&swap_lock); |
1732 | if ((unsigned int)type < nr_swapfiles) { | |
1733 | struct swap_info_struct *sis = swap_info[type]; | |
1734 | ||
ec8acf20 | 1735 | spin_lock(&sis->lock); |
efa90a98 HD |
1736 | if (sis->flags & SWP_WRITEOK) { |
1737 | n = sis->pages; | |
f577eb30 | 1738 | if (free) |
efa90a98 | 1739 | n -= sis->inuse_pages; |
f577eb30 | 1740 | } |
ec8acf20 | 1741 | spin_unlock(&sis->lock); |
f577eb30 | 1742 | } |
efa90a98 | 1743 | spin_unlock(&swap_lock); |
f577eb30 RW |
1744 | return n; |
1745 | } | |
73c34b6a | 1746 | #endif /* CONFIG_HIBERNATION */ |
f577eb30 | 1747 | |
9f8bdb3f | 1748 | static inline int pte_same_as_swp(pte_t pte, pte_t swp_pte) |
179ef71c | 1749 | { |
9f8bdb3f | 1750 | return pte_same(pte_swp_clear_soft_dirty(pte), swp_pte); |
179ef71c CG |
1751 | } |
1752 | ||
1da177e4 | 1753 | /* |
72866f6f HD |
1754 | * No need to decide whether this PTE shares the swap entry with others, |
1755 | * just let do_wp_page work it out if a write is requested later - to | |
1756 | * force COW, vm_page_prot omits write permission from any private vma. | |
1da177e4 | 1757 | */ |
044d66c1 | 1758 | static int unuse_pte(struct vm_area_struct *vma, pmd_t *pmd, |
1da177e4 LT |
1759 | unsigned long addr, swp_entry_t entry, struct page *page) |
1760 | { | |
9e16b7fb | 1761 | struct page *swapcache; |
72835c86 | 1762 | struct mem_cgroup *memcg; |
044d66c1 HD |
1763 | spinlock_t *ptl; |
1764 | pte_t *pte; | |
1765 | int ret = 1; | |
1766 | ||
9e16b7fb HD |
1767 | swapcache = page; |
1768 | page = ksm_might_need_to_copy(page, vma, addr); | |
1769 | if (unlikely(!page)) | |
1770 | return -ENOMEM; | |
1771 | ||
f627c2f5 KS |
1772 | if (mem_cgroup_try_charge(page, vma->vm_mm, GFP_KERNEL, |
1773 | &memcg, false)) { | |
044d66c1 | 1774 | ret = -ENOMEM; |
85d9fc89 KH |
1775 | goto out_nolock; |
1776 | } | |
044d66c1 HD |
1777 | |
1778 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); | |
9f8bdb3f | 1779 | if (unlikely(!pte_same_as_swp(*pte, swp_entry_to_pte(entry)))) { |
f627c2f5 | 1780 | mem_cgroup_cancel_charge(page, memcg, false); |
044d66c1 HD |
1781 | ret = 0; |
1782 | goto out; | |
1783 | } | |
8a9f3ccd | 1784 | |
b084d435 | 1785 | dec_mm_counter(vma->vm_mm, MM_SWAPENTS); |
d559db08 | 1786 | inc_mm_counter(vma->vm_mm, MM_ANONPAGES); |
1da177e4 LT |
1787 | get_page(page); |
1788 | set_pte_at(vma->vm_mm, addr, pte, | |
1789 | pte_mkold(mk_pte(page, vma->vm_page_prot))); | |
00501b53 | 1790 | if (page == swapcache) { |
d281ee61 | 1791 | page_add_anon_rmap(page, vma, addr, false); |
f627c2f5 | 1792 | mem_cgroup_commit_charge(page, memcg, true, false); |
00501b53 | 1793 | } else { /* ksm created a completely new copy */ |
d281ee61 | 1794 | page_add_new_anon_rmap(page, vma, addr, false); |
f627c2f5 | 1795 | mem_cgroup_commit_charge(page, memcg, false, false); |
00501b53 JW |
1796 | lru_cache_add_active_or_unevictable(page, vma); |
1797 | } | |
1da177e4 LT |
1798 | swap_free(entry); |
1799 | /* | |
1800 | * Move the page to the active list so it is not | |
1801 | * immediately swapped out again after swapon. | |
1802 | */ | |
1803 | activate_page(page); | |
044d66c1 HD |
1804 | out: |
1805 | pte_unmap_unlock(pte, ptl); | |
85d9fc89 | 1806 | out_nolock: |
9e16b7fb HD |
1807 | if (page != swapcache) { |
1808 | unlock_page(page); | |
1809 | put_page(page); | |
1810 | } | |
044d66c1 | 1811 | return ret; |
1da177e4 LT |
1812 | } |
1813 | ||
1814 | static int unuse_pte_range(struct vm_area_struct *vma, pmd_t *pmd, | |
1815 | unsigned long addr, unsigned long end, | |
1816 | swp_entry_t entry, struct page *page) | |
1817 | { | |
1da177e4 | 1818 | pte_t swp_pte = swp_entry_to_pte(entry); |
705e87c0 | 1819 | pte_t *pte; |
8a9f3ccd | 1820 | int ret = 0; |
1da177e4 | 1821 | |
044d66c1 HD |
1822 | /* |
1823 | * We don't actually need pte lock while scanning for swp_pte: since | |
1824 | * we hold page lock and mmap_sem, swp_pte cannot be inserted into the | |
1825 | * page table while we're scanning; though it could get zapped, and on | |
1826 | * some architectures (e.g. x86_32 with PAE) we might catch a glimpse | |
1827 | * of unmatched parts which look like swp_pte, so unuse_pte must | |
1828 | * recheck under pte lock. Scanning without pte lock lets it be | |
2de1a7e4 | 1829 | * preemptable whenever CONFIG_PREEMPT but not CONFIG_HIGHPTE. |
044d66c1 HD |
1830 | */ |
1831 | pte = pte_offset_map(pmd, addr); | |
1da177e4 LT |
1832 | do { |
1833 | /* | |
1834 | * swapoff spends a _lot_ of time in this loop! | |
1835 | * Test inline before going to call unuse_pte. | |
1836 | */ | |
9f8bdb3f | 1837 | if (unlikely(pte_same_as_swp(*pte, swp_pte))) { |
044d66c1 HD |
1838 | pte_unmap(pte); |
1839 | ret = unuse_pte(vma, pmd, addr, entry, page); | |
1840 | if (ret) | |
1841 | goto out; | |
1842 | pte = pte_offset_map(pmd, addr); | |
1da177e4 LT |
1843 | } |
1844 | } while (pte++, addr += PAGE_SIZE, addr != end); | |
044d66c1 HD |
1845 | pte_unmap(pte - 1); |
1846 | out: | |
8a9f3ccd | 1847 | return ret; |
1da177e4 LT |
1848 | } |
1849 | ||
1850 | static inline int unuse_pmd_range(struct vm_area_struct *vma, pud_t *pud, | |
1851 | unsigned long addr, unsigned long end, | |
1852 | swp_entry_t entry, struct page *page) | |
1853 | { | |
1854 | pmd_t *pmd; | |
1855 | unsigned long next; | |
8a9f3ccd | 1856 | int ret; |
1da177e4 LT |
1857 | |
1858 | pmd = pmd_offset(pud, addr); | |
1859 | do { | |
dc644a07 | 1860 | cond_resched(); |
1da177e4 | 1861 | next = pmd_addr_end(addr, end); |
1a5a9906 | 1862 | if (pmd_none_or_trans_huge_or_clear_bad(pmd)) |
1da177e4 | 1863 | continue; |
8a9f3ccd BS |
1864 | ret = unuse_pte_range(vma, pmd, addr, next, entry, page); |
1865 | if (ret) | |
1866 | return ret; | |
1da177e4 LT |
1867 | } while (pmd++, addr = next, addr != end); |
1868 | return 0; | |
1869 | } | |
1870 | ||
c2febafc | 1871 | static inline int unuse_pud_range(struct vm_area_struct *vma, p4d_t *p4d, |
1da177e4 LT |
1872 | unsigned long addr, unsigned long end, |
1873 | swp_entry_t entry, struct page *page) | |
1874 | { | |
1875 | pud_t *pud; | |
1876 | unsigned long next; | |
8a9f3ccd | 1877 | int ret; |
1da177e4 | 1878 | |
c2febafc | 1879 | pud = pud_offset(p4d, addr); |
1da177e4 LT |
1880 | do { |
1881 | next = pud_addr_end(addr, end); | |
1882 | if (pud_none_or_clear_bad(pud)) | |
1883 | continue; | |
8a9f3ccd BS |
1884 | ret = unuse_pmd_range(vma, pud, addr, next, entry, page); |
1885 | if (ret) | |
1886 | return ret; | |
1da177e4 LT |
1887 | } while (pud++, addr = next, addr != end); |
1888 | return 0; | |
1889 | } | |
1890 | ||
c2febafc KS |
1891 | static inline int unuse_p4d_range(struct vm_area_struct *vma, pgd_t *pgd, |
1892 | unsigned long addr, unsigned long end, | |
1893 | swp_entry_t entry, struct page *page) | |
1894 | { | |
1895 | p4d_t *p4d; | |
1896 | unsigned long next; | |
1897 | int ret; | |
1898 | ||
1899 | p4d = p4d_offset(pgd, addr); | |
1900 | do { | |
1901 | next = p4d_addr_end(addr, end); | |
1902 | if (p4d_none_or_clear_bad(p4d)) | |
1903 | continue; | |
1904 | ret = unuse_pud_range(vma, p4d, addr, next, entry, page); | |
1905 | if (ret) | |
1906 | return ret; | |
1907 | } while (p4d++, addr = next, addr != end); | |
1908 | return 0; | |
1909 | } | |
1910 | ||
1da177e4 LT |
1911 | static int unuse_vma(struct vm_area_struct *vma, |
1912 | swp_entry_t entry, struct page *page) | |
1913 | { | |
1914 | pgd_t *pgd; | |
1915 | unsigned long addr, end, next; | |
8a9f3ccd | 1916 | int ret; |
1da177e4 | 1917 | |
3ca7b3c5 | 1918 | if (page_anon_vma(page)) { |
1da177e4 LT |
1919 | addr = page_address_in_vma(page, vma); |
1920 | if (addr == -EFAULT) | |
1921 | return 0; | |
1922 | else | |
1923 | end = addr + PAGE_SIZE; | |
1924 | } else { | |
1925 | addr = vma->vm_start; | |
1926 | end = vma->vm_end; | |
1927 | } | |
1928 | ||
1929 | pgd = pgd_offset(vma->vm_mm, addr); | |
1930 | do { | |
1931 | next = pgd_addr_end(addr, end); | |
1932 | if (pgd_none_or_clear_bad(pgd)) | |
1933 | continue; | |
c2febafc | 1934 | ret = unuse_p4d_range(vma, pgd, addr, next, entry, page); |
8a9f3ccd BS |
1935 | if (ret) |
1936 | return ret; | |
1da177e4 LT |
1937 | } while (pgd++, addr = next, addr != end); |
1938 | return 0; | |
1939 | } | |
1940 | ||
1941 | static int unuse_mm(struct mm_struct *mm, | |
1942 | swp_entry_t entry, struct page *page) | |
1943 | { | |
1944 | struct vm_area_struct *vma; | |
8a9f3ccd | 1945 | int ret = 0; |
1da177e4 LT |
1946 | |
1947 | if (!down_read_trylock(&mm->mmap_sem)) { | |
1948 | /* | |
7d03431c FLVC |
1949 | * Activate page so shrink_inactive_list is unlikely to unmap |
1950 | * its ptes while lock is dropped, so swapoff can make progress. | |
1da177e4 | 1951 | */ |
c475a8ab | 1952 | activate_page(page); |
1da177e4 LT |
1953 | unlock_page(page); |
1954 | down_read(&mm->mmap_sem); | |
1955 | lock_page(page); | |
1956 | } | |
1da177e4 | 1957 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
8a9f3ccd | 1958 | if (vma->anon_vma && (ret = unuse_vma(vma, entry, page))) |
1da177e4 | 1959 | break; |
dc644a07 | 1960 | cond_resched(); |
1da177e4 | 1961 | } |
1da177e4 | 1962 | up_read(&mm->mmap_sem); |
8a9f3ccd | 1963 | return (ret < 0)? ret: 0; |
1da177e4 LT |
1964 | } |
1965 | ||
1966 | /* | |
38b5faf4 DM |
1967 | * Scan swap_map (or frontswap_map if frontswap parameter is true) |
1968 | * from current position to next entry still in use. | |
1da177e4 LT |
1969 | * Recycle to start on reaching the end, returning 0 when empty. |
1970 | */ | |
6eb396dc | 1971 | static unsigned int find_next_to_unuse(struct swap_info_struct *si, |
38b5faf4 | 1972 | unsigned int prev, bool frontswap) |
1da177e4 | 1973 | { |
6eb396dc HD |
1974 | unsigned int max = si->max; |
1975 | unsigned int i = prev; | |
8d69aaee | 1976 | unsigned char count; |
1da177e4 LT |
1977 | |
1978 | /* | |
5d337b91 | 1979 | * No need for swap_lock here: we're just looking |
1da177e4 LT |
1980 | * for whether an entry is in use, not modifying it; false |
1981 | * hits are okay, and sys_swapoff() has already prevented new | |
5d337b91 | 1982 | * allocations from this area (while holding swap_lock). |
1da177e4 LT |
1983 | */ |
1984 | for (;;) { | |
1985 | if (++i >= max) { | |
1986 | if (!prev) { | |
1987 | i = 0; | |
1988 | break; | |
1989 | } | |
1990 | /* | |
1991 | * No entries in use at top of swap_map, | |
1992 | * loop back to start and recheck there. | |
1993 | */ | |
1994 | max = prev + 1; | |
1995 | prev = 0; | |
1996 | i = 1; | |
1997 | } | |
4db0c3c2 | 1998 | count = READ_ONCE(si->swap_map[i]); |
355cfa73 | 1999 | if (count && swap_count(count) != SWAP_MAP_BAD) |
dc644a07 HD |
2000 | if (!frontswap || frontswap_test(si, i)) |
2001 | break; | |
2002 | if ((i % LATENCY_LIMIT) == 0) | |
2003 | cond_resched(); | |
1da177e4 LT |
2004 | } |
2005 | return i; | |
2006 | } | |
2007 | ||
2008 | /* | |
2009 | * We completely avoid races by reading each swap page in advance, | |
2010 | * and then search for the process using it. All the necessary | |
2011 | * page table adjustments can then be made atomically. | |
38b5faf4 DM |
2012 | * |
2013 | * if the boolean frontswap is true, only unuse pages_to_unuse pages; | |
2014 | * pages_to_unuse==0 means all pages; ignored if frontswap is false | |
1da177e4 | 2015 | */ |
38b5faf4 DM |
2016 | int try_to_unuse(unsigned int type, bool frontswap, |
2017 | unsigned long pages_to_unuse) | |
1da177e4 | 2018 | { |
efa90a98 | 2019 | struct swap_info_struct *si = swap_info[type]; |
1da177e4 | 2020 | struct mm_struct *start_mm; |
edfe23da SL |
2021 | volatile unsigned char *swap_map; /* swap_map is accessed without |
2022 | * locking. Mark it as volatile | |
2023 | * to prevent compiler doing | |
2024 | * something odd. | |
2025 | */ | |
8d69aaee | 2026 | unsigned char swcount; |
1da177e4 LT |
2027 | struct page *page; |
2028 | swp_entry_t entry; | |
6eb396dc | 2029 | unsigned int i = 0; |
1da177e4 | 2030 | int retval = 0; |
1da177e4 LT |
2031 | |
2032 | /* | |
2033 | * When searching mms for an entry, a good strategy is to | |
2034 | * start at the first mm we freed the previous entry from | |
2035 | * (though actually we don't notice whether we or coincidence | |
2036 | * freed the entry). Initialize this start_mm with a hold. | |
2037 | * | |
2038 | * A simpler strategy would be to start at the last mm we | |
2039 | * freed the previous entry from; but that would take less | |
2040 | * advantage of mmlist ordering, which clusters forked mms | |
2041 | * together, child after parent. If we race with dup_mmap(), we | |
2042 | * prefer to resolve parent before child, lest we miss entries | |
2043 | * duplicated after we scanned child: using last mm would invert | |
570a335b | 2044 | * that. |
1da177e4 LT |
2045 | */ |
2046 | start_mm = &init_mm; | |
3fce371b | 2047 | mmget(&init_mm); |
1da177e4 LT |
2048 | |
2049 | /* | |
2050 | * Keep on scanning until all entries have gone. Usually, | |
2051 | * one pass through swap_map is enough, but not necessarily: | |
2052 | * there are races when an instance of an entry might be missed. | |
2053 | */ | |
38b5faf4 | 2054 | while ((i = find_next_to_unuse(si, i, frontswap)) != 0) { |
1da177e4 LT |
2055 | if (signal_pending(current)) { |
2056 | retval = -EINTR; | |
2057 | break; | |
2058 | } | |
2059 | ||
886bb7e9 | 2060 | /* |
1da177e4 LT |
2061 | * Get a page for the entry, using the existing swap |
2062 | * cache page if there is one. Otherwise, get a clean | |
886bb7e9 | 2063 | * page and read the swap into it. |
1da177e4 LT |
2064 | */ |
2065 | swap_map = &si->swap_map[i]; | |
2066 | entry = swp_entry(type, i); | |
02098fea | 2067 | page = read_swap_cache_async(entry, |
23955622 | 2068 | GFP_HIGHUSER_MOVABLE, NULL, 0, false); |
1da177e4 LT |
2069 | if (!page) { |
2070 | /* | |
2071 | * Either swap_duplicate() failed because entry | |
2072 | * has been freed independently, and will not be | |
2073 | * reused since sys_swapoff() already disabled | |
2074 | * allocation from here, or alloc_page() failed. | |
2075 | */ | |
edfe23da SL |
2076 | swcount = *swap_map; |
2077 | /* | |
2078 | * We don't hold lock here, so the swap entry could be | |
2079 | * SWAP_MAP_BAD (when the cluster is discarding). | |
2080 | * Instead of fail out, We can just skip the swap | |
2081 | * entry because swapoff will wait for discarding | |
2082 | * finish anyway. | |
2083 | */ | |
2084 | if (!swcount || swcount == SWAP_MAP_BAD) | |
1da177e4 LT |
2085 | continue; |
2086 | retval = -ENOMEM; | |
2087 | break; | |
2088 | } | |
2089 | ||
2090 | /* | |
2091 | * Don't hold on to start_mm if it looks like exiting. | |
2092 | */ | |
2093 | if (atomic_read(&start_mm->mm_users) == 1) { | |
2094 | mmput(start_mm); | |
2095 | start_mm = &init_mm; | |
3fce371b | 2096 | mmget(&init_mm); |
1da177e4 LT |
2097 | } |
2098 | ||
2099 | /* | |
2100 | * Wait for and lock page. When do_swap_page races with | |
2101 | * try_to_unuse, do_swap_page can handle the fault much | |
2102 | * faster than try_to_unuse can locate the entry. This | |
2103 | * apparently redundant "wait_on_page_locked" lets try_to_unuse | |
2104 | * defer to do_swap_page in such a case - in some tests, | |
2105 | * do_swap_page and try_to_unuse repeatedly compete. | |
2106 | */ | |
2107 | wait_on_page_locked(page); | |
2108 | wait_on_page_writeback(page); | |
2109 | lock_page(page); | |
2110 | wait_on_page_writeback(page); | |
2111 | ||
2112 | /* | |
2113 | * Remove all references to entry. | |
1da177e4 | 2114 | */ |
1da177e4 | 2115 | swcount = *swap_map; |
aaa46865 HD |
2116 | if (swap_count(swcount) == SWAP_MAP_SHMEM) { |
2117 | retval = shmem_unuse(entry, page); | |
2118 | /* page has already been unlocked and released */ | |
2119 | if (retval < 0) | |
2120 | break; | |
2121 | continue; | |
1da177e4 | 2122 | } |
aaa46865 HD |
2123 | if (swap_count(swcount) && start_mm != &init_mm) |
2124 | retval = unuse_mm(start_mm, entry, page); | |
2125 | ||
355cfa73 | 2126 | if (swap_count(*swap_map)) { |
1da177e4 LT |
2127 | int set_start_mm = (*swap_map >= swcount); |
2128 | struct list_head *p = &start_mm->mmlist; | |
2129 | struct mm_struct *new_start_mm = start_mm; | |
2130 | struct mm_struct *prev_mm = start_mm; | |
2131 | struct mm_struct *mm; | |
2132 | ||
3fce371b VN |
2133 | mmget(new_start_mm); |
2134 | mmget(prev_mm); | |
1da177e4 | 2135 | spin_lock(&mmlist_lock); |
aaa46865 | 2136 | while (swap_count(*swap_map) && !retval && |
1da177e4 LT |
2137 | (p = p->next) != &start_mm->mmlist) { |
2138 | mm = list_entry(p, struct mm_struct, mmlist); | |
388f7934 | 2139 | if (!mmget_not_zero(mm)) |
1da177e4 | 2140 | continue; |
1da177e4 LT |
2141 | spin_unlock(&mmlist_lock); |
2142 | mmput(prev_mm); | |
2143 | prev_mm = mm; | |
2144 | ||
2145 | cond_resched(); | |
2146 | ||
2147 | swcount = *swap_map; | |
355cfa73 | 2148 | if (!swap_count(swcount)) /* any usage ? */ |
1da177e4 | 2149 | ; |
aaa46865 | 2150 | else if (mm == &init_mm) |
1da177e4 | 2151 | set_start_mm = 1; |
aaa46865 | 2152 | else |
1da177e4 | 2153 | retval = unuse_mm(mm, entry, page); |
355cfa73 | 2154 | |
32c5fc10 | 2155 | if (set_start_mm && *swap_map < swcount) { |
1da177e4 | 2156 | mmput(new_start_mm); |
3fce371b | 2157 | mmget(mm); |
1da177e4 LT |
2158 | new_start_mm = mm; |
2159 | set_start_mm = 0; | |
2160 | } | |
2161 | spin_lock(&mmlist_lock); | |
2162 | } | |
2163 | spin_unlock(&mmlist_lock); | |
2164 | mmput(prev_mm); | |
2165 | mmput(start_mm); | |
2166 | start_mm = new_start_mm; | |
2167 | } | |
2168 | if (retval) { | |
2169 | unlock_page(page); | |
09cbfeaf | 2170 | put_page(page); |
1da177e4 LT |
2171 | break; |
2172 | } | |
2173 | ||
1da177e4 LT |
2174 | /* |
2175 | * If a reference remains (rare), we would like to leave | |
2176 | * the page in the swap cache; but try_to_unmap could | |
2177 | * then re-duplicate the entry once we drop page lock, | |
2178 | * so we might loop indefinitely; also, that page could | |
2179 | * not be swapped out to other storage meanwhile. So: | |
2180 | * delete from cache even if there's another reference, | |
2181 | * after ensuring that the data has been saved to disk - | |
2182 | * since if the reference remains (rarer), it will be | |
2183 | * read from disk into another page. Splitting into two | |
2184 | * pages would be incorrect if swap supported "shared | |
2185 | * private" pages, but they are handled by tmpfs files. | |
5ad64688 HD |
2186 | * |
2187 | * Given how unuse_vma() targets one particular offset | |
2188 | * in an anon_vma, once the anon_vma has been determined, | |
2189 | * this splitting happens to be just what is needed to | |
2190 | * handle where KSM pages have been swapped out: re-reading | |
2191 | * is unnecessarily slow, but we can fix that later on. | |
1da177e4 | 2192 | */ |
355cfa73 KH |
2193 | if (swap_count(*swap_map) && |
2194 | PageDirty(page) && PageSwapCache(page)) { | |
1da177e4 LT |
2195 | struct writeback_control wbc = { |
2196 | .sync_mode = WB_SYNC_NONE, | |
2197 | }; | |
2198 | ||
e0709829 | 2199 | swap_writepage(compound_head(page), &wbc); |
1da177e4 LT |
2200 | lock_page(page); |
2201 | wait_on_page_writeback(page); | |
2202 | } | |
68bdc8d6 HD |
2203 | |
2204 | /* | |
2205 | * It is conceivable that a racing task removed this page from | |
2206 | * swap cache just before we acquired the page lock at the top, | |
2207 | * or while we dropped it in unuse_mm(). The page might even | |
2208 | * be back in swap cache on another swap area: that we must not | |
2209 | * delete, since it may not have been written out to swap yet. | |
2210 | */ | |
2211 | if (PageSwapCache(page) && | |
e0709829 HY |
2212 | likely(page_private(page) == entry.val) && |
2213 | !page_swapped(page)) | |
2214 | delete_from_swap_cache(compound_head(page)); | |
1da177e4 LT |
2215 | |
2216 | /* | |
2217 | * So we could skip searching mms once swap count went | |
2218 | * to 1, we did not mark any present ptes as dirty: must | |
2706a1b8 | 2219 | * mark page dirty so shrink_page_list will preserve it. |
1da177e4 LT |
2220 | */ |
2221 | SetPageDirty(page); | |
2222 | unlock_page(page); | |
09cbfeaf | 2223 | put_page(page); |
1da177e4 LT |
2224 | |
2225 | /* | |
2226 | * Make sure that we aren't completely killing | |
2227 | * interactive performance. | |
2228 | */ | |
2229 | cond_resched(); | |
38b5faf4 DM |
2230 | if (frontswap && pages_to_unuse > 0) { |
2231 | if (!--pages_to_unuse) | |
2232 | break; | |
2233 | } | |
1da177e4 LT |
2234 | } |
2235 | ||
2236 | mmput(start_mm); | |
1da177e4 LT |
2237 | return retval; |
2238 | } | |
2239 | ||
2240 | /* | |
5d337b91 HD |
2241 | * After a successful try_to_unuse, if no swap is now in use, we know |
2242 | * we can empty the mmlist. swap_lock must be held on entry and exit. | |
2243 | * Note that mmlist_lock nests inside swap_lock, and an mm must be | |
1da177e4 LT |
2244 | * added to the mmlist just after page_duplicate - before would be racy. |
2245 | */ | |
2246 | static void drain_mmlist(void) | |
2247 | { | |
2248 | struct list_head *p, *next; | |
efa90a98 | 2249 | unsigned int type; |
1da177e4 | 2250 | |
efa90a98 HD |
2251 | for (type = 0; type < nr_swapfiles; type++) |
2252 | if (swap_info[type]->inuse_pages) | |
1da177e4 LT |
2253 | return; |
2254 | spin_lock(&mmlist_lock); | |
2255 | list_for_each_safe(p, next, &init_mm.mmlist) | |
2256 | list_del_init(p); | |
2257 | spin_unlock(&mmlist_lock); | |
2258 | } | |
2259 | ||
2260 | /* | |
2261 | * Use this swapdev's extent info to locate the (PAGE_SIZE) block which | |
d4906e1a LS |
2262 | * corresponds to page offset for the specified swap entry. |
2263 | * Note that the type of this function is sector_t, but it returns page offset | |
2264 | * into the bdev, not sector offset. | |
1da177e4 | 2265 | */ |
d4906e1a | 2266 | static sector_t map_swap_entry(swp_entry_t entry, struct block_device **bdev) |
1da177e4 | 2267 | { |
f29ad6a9 HD |
2268 | struct swap_info_struct *sis; |
2269 | struct swap_extent *start_se; | |
2270 | struct swap_extent *se; | |
2271 | pgoff_t offset; | |
2272 | ||
efa90a98 | 2273 | sis = swap_info[swp_type(entry)]; |
f29ad6a9 HD |
2274 | *bdev = sis->bdev; |
2275 | ||
2276 | offset = swp_offset(entry); | |
2277 | start_se = sis->curr_swap_extent; | |
2278 | se = start_se; | |
1da177e4 LT |
2279 | |
2280 | for ( ; ; ) { | |
1da177e4 LT |
2281 | if (se->start_page <= offset && |
2282 | offset < (se->start_page + se->nr_pages)) { | |
2283 | return se->start_block + (offset - se->start_page); | |
2284 | } | |
a8ae4991 | 2285 | se = list_next_entry(se, list); |
1da177e4 LT |
2286 | sis->curr_swap_extent = se; |
2287 | BUG_ON(se == start_se); /* It *must* be present */ | |
2288 | } | |
2289 | } | |
2290 | ||
d4906e1a LS |
2291 | /* |
2292 | * Returns the page offset into bdev for the specified page's swap entry. | |
2293 | */ | |
2294 | sector_t map_swap_page(struct page *page, struct block_device **bdev) | |
2295 | { | |
2296 | swp_entry_t entry; | |
2297 | entry.val = page_private(page); | |
2298 | return map_swap_entry(entry, bdev); | |
2299 | } | |
2300 | ||
1da177e4 LT |
2301 | /* |
2302 | * Free all of a swapdev's extent information | |
2303 | */ | |
2304 | static void destroy_swap_extents(struct swap_info_struct *sis) | |
2305 | { | |
9625a5f2 | 2306 | while (!list_empty(&sis->first_swap_extent.list)) { |
1da177e4 LT |
2307 | struct swap_extent *se; |
2308 | ||
a8ae4991 | 2309 | se = list_first_entry(&sis->first_swap_extent.list, |
1da177e4 LT |
2310 | struct swap_extent, list); |
2311 | list_del(&se->list); | |
2312 | kfree(se); | |
2313 | } | |
62c230bc MG |
2314 | |
2315 | if (sis->flags & SWP_FILE) { | |
2316 | struct file *swap_file = sis->swap_file; | |
2317 | struct address_space *mapping = swap_file->f_mapping; | |
2318 | ||
2319 | sis->flags &= ~SWP_FILE; | |
2320 | mapping->a_ops->swap_deactivate(swap_file); | |
2321 | } | |
1da177e4 LT |
2322 | } |
2323 | ||
2324 | /* | |
2325 | * Add a block range (and the corresponding page range) into this swapdev's | |
11d31886 | 2326 | * extent list. The extent list is kept sorted in page order. |
1da177e4 | 2327 | * |
11d31886 | 2328 | * This function rather assumes that it is called in ascending page order. |
1da177e4 | 2329 | */ |
a509bc1a | 2330 | int |
1da177e4 LT |
2331 | add_swap_extent(struct swap_info_struct *sis, unsigned long start_page, |
2332 | unsigned long nr_pages, sector_t start_block) | |
2333 | { | |
2334 | struct swap_extent *se; | |
2335 | struct swap_extent *new_se; | |
2336 | struct list_head *lh; | |
2337 | ||
9625a5f2 HD |
2338 | if (start_page == 0) { |
2339 | se = &sis->first_swap_extent; | |
2340 | sis->curr_swap_extent = se; | |
2341 | se->start_page = 0; | |
2342 | se->nr_pages = nr_pages; | |
2343 | se->start_block = start_block; | |
2344 | return 1; | |
2345 | } else { | |
2346 | lh = sis->first_swap_extent.list.prev; /* Highest extent */ | |
1da177e4 | 2347 | se = list_entry(lh, struct swap_extent, list); |
11d31886 HD |
2348 | BUG_ON(se->start_page + se->nr_pages != start_page); |
2349 | if (se->start_block + se->nr_pages == start_block) { | |
1da177e4 LT |
2350 | /* Merge it */ |
2351 | se->nr_pages += nr_pages; | |
2352 | return 0; | |
2353 | } | |
1da177e4 LT |
2354 | } |
2355 | ||
2356 | /* | |
2357 | * No merge. Insert a new extent, preserving ordering. | |
2358 | */ | |
2359 | new_se = kmalloc(sizeof(*se), GFP_KERNEL); | |
2360 | if (new_se == NULL) | |
2361 | return -ENOMEM; | |
2362 | new_se->start_page = start_page; | |
2363 | new_se->nr_pages = nr_pages; | |
2364 | new_se->start_block = start_block; | |
2365 | ||
9625a5f2 | 2366 | list_add_tail(&new_se->list, &sis->first_swap_extent.list); |
53092a74 | 2367 | return 1; |
1da177e4 LT |
2368 | } |
2369 | ||
2370 | /* | |
2371 | * A `swap extent' is a simple thing which maps a contiguous range of pages | |
2372 | * onto a contiguous range of disk blocks. An ordered list of swap extents | |
2373 | * is built at swapon time and is then used at swap_writepage/swap_readpage | |
2374 | * time for locating where on disk a page belongs. | |
2375 | * | |
2376 | * If the swapfile is an S_ISBLK block device, a single extent is installed. | |
2377 | * This is done so that the main operating code can treat S_ISBLK and S_ISREG | |
2378 | * swap files identically. | |
2379 | * | |
2380 | * Whether the swapdev is an S_ISREG file or an S_ISBLK blockdev, the swap | |
2381 | * extent list operates in PAGE_SIZE disk blocks. Both S_ISREG and S_ISBLK | |
2382 | * swapfiles are handled *identically* after swapon time. | |
2383 | * | |
2384 | * For S_ISREG swapfiles, setup_swap_extents() will walk all the file's blocks | |
2385 | * and will parse them into an ordered extent list, in PAGE_SIZE chunks. If | |
2386 | * some stray blocks are found which do not fall within the PAGE_SIZE alignment | |
2387 | * requirements, they are simply tossed out - we will never use those blocks | |
2388 | * for swapping. | |
2389 | * | |
b0d9bcd4 | 2390 | * For S_ISREG swapfiles we set S_SWAPFILE across the life of the swapon. This |
1da177e4 LT |
2391 | * prevents root from shooting her foot off by ftruncating an in-use swapfile, |
2392 | * which will scribble on the fs. | |
2393 | * | |
2394 | * The amount of disk space which a single swap extent represents varies. | |
2395 | * Typically it is in the 1-4 megabyte range. So we can have hundreds of | |
2396 | * extents in the list. To avoid much list walking, we cache the previous | |
2397 | * search location in `curr_swap_extent', and start new searches from there. | |
2398 | * This is extremely effective. The average number of iterations in | |
2399 | * map_swap_page() has been measured at about 0.3 per page. - akpm. | |
2400 | */ | |
53092a74 | 2401 | static int setup_swap_extents(struct swap_info_struct *sis, sector_t *span) |
1da177e4 | 2402 | { |
62c230bc MG |
2403 | struct file *swap_file = sis->swap_file; |
2404 | struct address_space *mapping = swap_file->f_mapping; | |
2405 | struct inode *inode = mapping->host; | |
1da177e4 LT |
2406 | int ret; |
2407 | ||
1da177e4 LT |
2408 | if (S_ISBLK(inode->i_mode)) { |
2409 | ret = add_swap_extent(sis, 0, sis->max, 0); | |
53092a74 | 2410 | *span = sis->pages; |
a509bc1a | 2411 | return ret; |
1da177e4 LT |
2412 | } |
2413 | ||
62c230bc | 2414 | if (mapping->a_ops->swap_activate) { |
a509bc1a | 2415 | ret = mapping->a_ops->swap_activate(sis, swap_file, span); |
62c230bc MG |
2416 | if (!ret) { |
2417 | sis->flags |= SWP_FILE; | |
2418 | ret = add_swap_extent(sis, 0, sis->max, 0); | |
2419 | *span = sis->pages; | |
2420 | } | |
a509bc1a | 2421 | return ret; |
62c230bc MG |
2422 | } |
2423 | ||
a509bc1a | 2424 | return generic_swapfile_activate(sis, swap_file, span); |
1da177e4 LT |
2425 | } |
2426 | ||
a2468cc9 AL |
2427 | static int swap_node(struct swap_info_struct *p) |
2428 | { | |
2429 | struct block_device *bdev; | |
2430 | ||
2431 | if (p->bdev) | |
2432 | bdev = p->bdev; | |
2433 | else | |
2434 | bdev = p->swap_file->f_inode->i_sb->s_bdev; | |
2435 | ||
2436 | return bdev ? bdev->bd_disk->node_id : NUMA_NO_NODE; | |
2437 | } | |
2438 | ||
cf0cac0a | 2439 | static void _enable_swap_info(struct swap_info_struct *p, int prio, |
2a8f9449 SL |
2440 | unsigned char *swap_map, |
2441 | struct swap_cluster_info *cluster_info) | |
40531542 | 2442 | { |
a2468cc9 AL |
2443 | int i; |
2444 | ||
40531542 CEB |
2445 | if (prio >= 0) |
2446 | p->prio = prio; | |
2447 | else | |
2448 | p->prio = --least_priority; | |
18ab4d4c DS |
2449 | /* |
2450 | * the plist prio is negated because plist ordering is | |
2451 | * low-to-high, while swap ordering is high-to-low | |
2452 | */ | |
2453 | p->list.prio = -p->prio; | |
a2468cc9 AL |
2454 | for_each_node(i) { |
2455 | if (p->prio >= 0) | |
2456 | p->avail_lists[i].prio = -p->prio; | |
2457 | else { | |
2458 | if (swap_node(p) == i) | |
2459 | p->avail_lists[i].prio = 1; | |
2460 | else | |
2461 | p->avail_lists[i].prio = -p->prio; | |
2462 | } | |
2463 | } | |
40531542 | 2464 | p->swap_map = swap_map; |
2a8f9449 | 2465 | p->cluster_info = cluster_info; |
40531542 | 2466 | p->flags |= SWP_WRITEOK; |
ec8acf20 | 2467 | atomic_long_add(p->pages, &nr_swap_pages); |
40531542 CEB |
2468 | total_swap_pages += p->pages; |
2469 | ||
adfab836 | 2470 | assert_spin_locked(&swap_lock); |
adfab836 | 2471 | /* |
18ab4d4c DS |
2472 | * both lists are plists, and thus priority ordered. |
2473 | * swap_active_head needs to be priority ordered for swapoff(), | |
2474 | * which on removal of any swap_info_struct with an auto-assigned | |
2475 | * (i.e. negative) priority increments the auto-assigned priority | |
2476 | * of any lower-priority swap_info_structs. | |
2477 | * swap_avail_head needs to be priority ordered for get_swap_page(), | |
2478 | * which allocates swap pages from the highest available priority | |
2479 | * swap_info_struct. | |
adfab836 | 2480 | */ |
18ab4d4c | 2481 | plist_add(&p->list, &swap_active_head); |
a2468cc9 | 2482 | add_to_avail_list(p); |
cf0cac0a CEB |
2483 | } |
2484 | ||
2485 | static void enable_swap_info(struct swap_info_struct *p, int prio, | |
2486 | unsigned char *swap_map, | |
2a8f9449 | 2487 | struct swap_cluster_info *cluster_info, |
cf0cac0a CEB |
2488 | unsigned long *frontswap_map) |
2489 | { | |
4f89849d | 2490 | frontswap_init(p->type, frontswap_map); |
cf0cac0a | 2491 | spin_lock(&swap_lock); |
ec8acf20 | 2492 | spin_lock(&p->lock); |
2a8f9449 | 2493 | _enable_swap_info(p, prio, swap_map, cluster_info); |
ec8acf20 | 2494 | spin_unlock(&p->lock); |
cf0cac0a CEB |
2495 | spin_unlock(&swap_lock); |
2496 | } | |
2497 | ||
2498 | static void reinsert_swap_info(struct swap_info_struct *p) | |
2499 | { | |
2500 | spin_lock(&swap_lock); | |
ec8acf20 | 2501 | spin_lock(&p->lock); |
2a8f9449 | 2502 | _enable_swap_info(p, p->prio, p->swap_map, p->cluster_info); |
ec8acf20 | 2503 | spin_unlock(&p->lock); |
40531542 CEB |
2504 | spin_unlock(&swap_lock); |
2505 | } | |
2506 | ||
67afa38e TC |
2507 | bool has_usable_swap(void) |
2508 | { | |
2509 | bool ret = true; | |
2510 | ||
2511 | spin_lock(&swap_lock); | |
2512 | if (plist_head_empty(&swap_active_head)) | |
2513 | ret = false; | |
2514 | spin_unlock(&swap_lock); | |
2515 | return ret; | |
2516 | } | |
2517 | ||
c4ea37c2 | 2518 | SYSCALL_DEFINE1(swapoff, const char __user *, specialfile) |
1da177e4 | 2519 | { |
73c34b6a | 2520 | struct swap_info_struct *p = NULL; |
8d69aaee | 2521 | unsigned char *swap_map; |
2a8f9449 | 2522 | struct swap_cluster_info *cluster_info; |
4f89849d | 2523 | unsigned long *frontswap_map; |
1da177e4 LT |
2524 | struct file *swap_file, *victim; |
2525 | struct address_space *mapping; | |
2526 | struct inode *inode; | |
91a27b2a | 2527 | struct filename *pathname; |
adfab836 | 2528 | int err, found = 0; |
5b808a23 | 2529 | unsigned int old_block_size; |
886bb7e9 | 2530 | |
1da177e4 LT |
2531 | if (!capable(CAP_SYS_ADMIN)) |
2532 | return -EPERM; | |
2533 | ||
191c5424 AV |
2534 | BUG_ON(!current->mm); |
2535 | ||
1da177e4 | 2536 | pathname = getname(specialfile); |
1da177e4 | 2537 | if (IS_ERR(pathname)) |
f58b59c1 | 2538 | return PTR_ERR(pathname); |
1da177e4 | 2539 | |
669abf4e | 2540 | victim = file_open_name(pathname, O_RDWR|O_LARGEFILE, 0); |
1da177e4 LT |
2541 | err = PTR_ERR(victim); |
2542 | if (IS_ERR(victim)) | |
2543 | goto out; | |
2544 | ||
2545 | mapping = victim->f_mapping; | |
5d337b91 | 2546 | spin_lock(&swap_lock); |
18ab4d4c | 2547 | plist_for_each_entry(p, &swap_active_head, list) { |
22c6f8fd | 2548 | if (p->flags & SWP_WRITEOK) { |
adfab836 DS |
2549 | if (p->swap_file->f_mapping == mapping) { |
2550 | found = 1; | |
1da177e4 | 2551 | break; |
adfab836 | 2552 | } |
1da177e4 | 2553 | } |
1da177e4 | 2554 | } |
adfab836 | 2555 | if (!found) { |
1da177e4 | 2556 | err = -EINVAL; |
5d337b91 | 2557 | spin_unlock(&swap_lock); |
1da177e4 LT |
2558 | goto out_dput; |
2559 | } | |
191c5424 | 2560 | if (!security_vm_enough_memory_mm(current->mm, p->pages)) |
1da177e4 LT |
2561 | vm_unacct_memory(p->pages); |
2562 | else { | |
2563 | err = -ENOMEM; | |
5d337b91 | 2564 | spin_unlock(&swap_lock); |
1da177e4 LT |
2565 | goto out_dput; |
2566 | } | |
a2468cc9 | 2567 | del_from_avail_list(p); |
ec8acf20 | 2568 | spin_lock(&p->lock); |
78ecba08 | 2569 | if (p->prio < 0) { |
adfab836 | 2570 | struct swap_info_struct *si = p; |
a2468cc9 | 2571 | int nid; |
adfab836 | 2572 | |
18ab4d4c | 2573 | plist_for_each_entry_continue(si, &swap_active_head, list) { |
adfab836 | 2574 | si->prio++; |
18ab4d4c | 2575 | si->list.prio--; |
a2468cc9 AL |
2576 | for_each_node(nid) { |
2577 | if (si->avail_lists[nid].prio != 1) | |
2578 | si->avail_lists[nid].prio--; | |
2579 | } | |
adfab836 | 2580 | } |
78ecba08 HD |
2581 | least_priority++; |
2582 | } | |
18ab4d4c | 2583 | plist_del(&p->list, &swap_active_head); |
ec8acf20 | 2584 | atomic_long_sub(p->pages, &nr_swap_pages); |
1da177e4 LT |
2585 | total_swap_pages -= p->pages; |
2586 | p->flags &= ~SWP_WRITEOK; | |
ec8acf20 | 2587 | spin_unlock(&p->lock); |
5d337b91 | 2588 | spin_unlock(&swap_lock); |
fb4f88dc | 2589 | |
039939a6 TC |
2590 | disable_swap_slots_cache_lock(); |
2591 | ||
e1e12d2f | 2592 | set_current_oom_origin(); |
adfab836 | 2593 | err = try_to_unuse(p->type, false, 0); /* force unuse all pages */ |
e1e12d2f | 2594 | clear_current_oom_origin(); |
1da177e4 | 2595 | |
1da177e4 LT |
2596 | if (err) { |
2597 | /* re-insert swap space back into swap_list */ | |
cf0cac0a | 2598 | reinsert_swap_info(p); |
039939a6 | 2599 | reenable_swap_slots_cache_unlock(); |
1da177e4 LT |
2600 | goto out_dput; |
2601 | } | |
52b7efdb | 2602 | |
039939a6 TC |
2603 | reenable_swap_slots_cache_unlock(); |
2604 | ||
815c2c54 SL |
2605 | flush_work(&p->discard_work); |
2606 | ||
5d337b91 | 2607 | destroy_swap_extents(p); |
570a335b HD |
2608 | if (p->flags & SWP_CONTINUED) |
2609 | free_swap_count_continuations(p); | |
2610 | ||
81a0298b HY |
2611 | if (!p->bdev || !blk_queue_nonrot(bdev_get_queue(p->bdev))) |
2612 | atomic_dec(&nr_rotate_swap); | |
2613 | ||
fc0abb14 | 2614 | mutex_lock(&swapon_mutex); |
5d337b91 | 2615 | spin_lock(&swap_lock); |
ec8acf20 | 2616 | spin_lock(&p->lock); |
5d337b91 HD |
2617 | drain_mmlist(); |
2618 | ||
52b7efdb | 2619 | /* wait for anyone still in scan_swap_map */ |
52b7efdb HD |
2620 | p->highest_bit = 0; /* cuts scans short */ |
2621 | while (p->flags >= SWP_SCANNING) { | |
ec8acf20 | 2622 | spin_unlock(&p->lock); |
5d337b91 | 2623 | spin_unlock(&swap_lock); |
13e4b57f | 2624 | schedule_timeout_uninterruptible(1); |
5d337b91 | 2625 | spin_lock(&swap_lock); |
ec8acf20 | 2626 | spin_lock(&p->lock); |
52b7efdb | 2627 | } |
52b7efdb | 2628 | |
1da177e4 | 2629 | swap_file = p->swap_file; |
5b808a23 | 2630 | old_block_size = p->old_block_size; |
1da177e4 LT |
2631 | p->swap_file = NULL; |
2632 | p->max = 0; | |
2633 | swap_map = p->swap_map; | |
2634 | p->swap_map = NULL; | |
2a8f9449 SL |
2635 | cluster_info = p->cluster_info; |
2636 | p->cluster_info = NULL; | |
4f89849d | 2637 | frontswap_map = frontswap_map_get(p); |
ec8acf20 | 2638 | spin_unlock(&p->lock); |
5d337b91 | 2639 | spin_unlock(&swap_lock); |
adfab836 | 2640 | frontswap_invalidate_area(p->type); |
58e97ba6 | 2641 | frontswap_map_set(p, NULL); |
fc0abb14 | 2642 | mutex_unlock(&swapon_mutex); |
ebc2a1a6 SL |
2643 | free_percpu(p->percpu_cluster); |
2644 | p->percpu_cluster = NULL; | |
1da177e4 | 2645 | vfree(swap_map); |
54f180d3 HY |
2646 | kvfree(cluster_info); |
2647 | kvfree(frontswap_map); | |
2de1a7e4 | 2648 | /* Destroy swap account information */ |
adfab836 | 2649 | swap_cgroup_swapoff(p->type); |
4b3ef9da | 2650 | exit_swap_address_space(p->type); |
27a7faa0 | 2651 | |
1da177e4 LT |
2652 | inode = mapping->host; |
2653 | if (S_ISBLK(inode->i_mode)) { | |
2654 | struct block_device *bdev = I_BDEV(inode); | |
5b808a23 | 2655 | set_blocksize(bdev, old_block_size); |
e525fd89 | 2656 | blkdev_put(bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL); |
1da177e4 | 2657 | } else { |
5955102c | 2658 | inode_lock(inode); |
1da177e4 | 2659 | inode->i_flags &= ~S_SWAPFILE; |
5955102c | 2660 | inode_unlock(inode); |
1da177e4 LT |
2661 | } |
2662 | filp_close(swap_file, NULL); | |
f893ab41 WY |
2663 | |
2664 | /* | |
2665 | * Clear the SWP_USED flag after all resources are freed so that swapon | |
2666 | * can reuse this swap_info in alloc_swap_info() safely. It is ok to | |
2667 | * not hold p->lock after we cleared its SWP_WRITEOK. | |
2668 | */ | |
2669 | spin_lock(&swap_lock); | |
2670 | p->flags = 0; | |
2671 | spin_unlock(&swap_lock); | |
2672 | ||
1da177e4 | 2673 | err = 0; |
66d7dd51 KS |
2674 | atomic_inc(&proc_poll_event); |
2675 | wake_up_interruptible(&proc_poll_wait); | |
1da177e4 LT |
2676 | |
2677 | out_dput: | |
2678 | filp_close(victim, NULL); | |
2679 | out: | |
f58b59c1 | 2680 | putname(pathname); |
1da177e4 LT |
2681 | return err; |
2682 | } | |
2683 | ||
2684 | #ifdef CONFIG_PROC_FS | |
9dd95748 | 2685 | static __poll_t swaps_poll(struct file *file, poll_table *wait) |
66d7dd51 | 2686 | { |
f1514638 | 2687 | struct seq_file *seq = file->private_data; |
66d7dd51 KS |
2688 | |
2689 | poll_wait(file, &proc_poll_wait, wait); | |
2690 | ||
f1514638 KS |
2691 | if (seq->poll_event != atomic_read(&proc_poll_event)) { |
2692 | seq->poll_event = atomic_read(&proc_poll_event); | |
a9a08845 | 2693 | return EPOLLIN | EPOLLRDNORM | EPOLLERR | EPOLLPRI; |
66d7dd51 KS |
2694 | } |
2695 | ||
a9a08845 | 2696 | return EPOLLIN | EPOLLRDNORM; |
66d7dd51 KS |
2697 | } |
2698 | ||
1da177e4 LT |
2699 | /* iterator */ |
2700 | static void *swap_start(struct seq_file *swap, loff_t *pos) | |
2701 | { | |
efa90a98 HD |
2702 | struct swap_info_struct *si; |
2703 | int type; | |
1da177e4 LT |
2704 | loff_t l = *pos; |
2705 | ||
fc0abb14 | 2706 | mutex_lock(&swapon_mutex); |
1da177e4 | 2707 | |
881e4aab SS |
2708 | if (!l) |
2709 | return SEQ_START_TOKEN; | |
2710 | ||
efa90a98 HD |
2711 | for (type = 0; type < nr_swapfiles; type++) { |
2712 | smp_rmb(); /* read nr_swapfiles before swap_info[type] */ | |
2713 | si = swap_info[type]; | |
2714 | if (!(si->flags & SWP_USED) || !si->swap_map) | |
1da177e4 | 2715 | continue; |
881e4aab | 2716 | if (!--l) |
efa90a98 | 2717 | return si; |
1da177e4 LT |
2718 | } |
2719 | ||
2720 | return NULL; | |
2721 | } | |
2722 | ||
2723 | static void *swap_next(struct seq_file *swap, void *v, loff_t *pos) | |
2724 | { | |
efa90a98 HD |
2725 | struct swap_info_struct *si = v; |
2726 | int type; | |
1da177e4 | 2727 | |
881e4aab | 2728 | if (v == SEQ_START_TOKEN) |
efa90a98 HD |
2729 | type = 0; |
2730 | else | |
2731 | type = si->type + 1; | |
881e4aab | 2732 | |
efa90a98 HD |
2733 | for (; type < nr_swapfiles; type++) { |
2734 | smp_rmb(); /* read nr_swapfiles before swap_info[type] */ | |
2735 | si = swap_info[type]; | |
2736 | if (!(si->flags & SWP_USED) || !si->swap_map) | |
1da177e4 LT |
2737 | continue; |
2738 | ++*pos; | |
efa90a98 | 2739 | return si; |
1da177e4 LT |
2740 | } |
2741 | ||
2742 | return NULL; | |
2743 | } | |
2744 | ||
2745 | static void swap_stop(struct seq_file *swap, void *v) | |
2746 | { | |
fc0abb14 | 2747 | mutex_unlock(&swapon_mutex); |
1da177e4 LT |
2748 | } |
2749 | ||
2750 | static int swap_show(struct seq_file *swap, void *v) | |
2751 | { | |
efa90a98 | 2752 | struct swap_info_struct *si = v; |
1da177e4 LT |
2753 | struct file *file; |
2754 | int len; | |
2755 | ||
efa90a98 | 2756 | if (si == SEQ_START_TOKEN) { |
881e4aab SS |
2757 | seq_puts(swap,"Filename\t\t\t\tType\t\tSize\tUsed\tPriority\n"); |
2758 | return 0; | |
2759 | } | |
1da177e4 | 2760 | |
efa90a98 | 2761 | file = si->swap_file; |
2726d566 | 2762 | len = seq_file_path(swap, file, " \t\n\\"); |
6eb396dc | 2763 | seq_printf(swap, "%*s%s\t%u\t%u\t%d\n", |
886bb7e9 | 2764 | len < 40 ? 40 - len : 1, " ", |
496ad9aa | 2765 | S_ISBLK(file_inode(file)->i_mode) ? |
1da177e4 | 2766 | "partition" : "file\t", |
efa90a98 HD |
2767 | si->pages << (PAGE_SHIFT - 10), |
2768 | si->inuse_pages << (PAGE_SHIFT - 10), | |
2769 | si->prio); | |
1da177e4 LT |
2770 | return 0; |
2771 | } | |
2772 | ||
15ad7cdc | 2773 | static const struct seq_operations swaps_op = { |
1da177e4 LT |
2774 | .start = swap_start, |
2775 | .next = swap_next, | |
2776 | .stop = swap_stop, | |
2777 | .show = swap_show | |
2778 | }; | |
2779 | ||
2780 | static int swaps_open(struct inode *inode, struct file *file) | |
2781 | { | |
f1514638 | 2782 | struct seq_file *seq; |
66d7dd51 KS |
2783 | int ret; |
2784 | ||
66d7dd51 | 2785 | ret = seq_open(file, &swaps_op); |
f1514638 | 2786 | if (ret) |
66d7dd51 | 2787 | return ret; |
66d7dd51 | 2788 | |
f1514638 KS |
2789 | seq = file->private_data; |
2790 | seq->poll_event = atomic_read(&proc_poll_event); | |
2791 | return 0; | |
1da177e4 LT |
2792 | } |
2793 | ||
15ad7cdc | 2794 | static const struct file_operations proc_swaps_operations = { |
1da177e4 LT |
2795 | .open = swaps_open, |
2796 | .read = seq_read, | |
2797 | .llseek = seq_lseek, | |
2798 | .release = seq_release, | |
66d7dd51 | 2799 | .poll = swaps_poll, |
1da177e4 LT |
2800 | }; |
2801 | ||
2802 | static int __init procswaps_init(void) | |
2803 | { | |
3d71f86f | 2804 | proc_create("swaps", 0, NULL, &proc_swaps_operations); |
1da177e4 LT |
2805 | return 0; |
2806 | } | |
2807 | __initcall(procswaps_init); | |
2808 | #endif /* CONFIG_PROC_FS */ | |
2809 | ||
1796316a JB |
2810 | #ifdef MAX_SWAPFILES_CHECK |
2811 | static int __init max_swapfiles_check(void) | |
2812 | { | |
2813 | MAX_SWAPFILES_CHECK(); | |
2814 | return 0; | |
2815 | } | |
2816 | late_initcall(max_swapfiles_check); | |
2817 | #endif | |
2818 | ||
53cbb243 | 2819 | static struct swap_info_struct *alloc_swap_info(void) |
1da177e4 | 2820 | { |
73c34b6a | 2821 | struct swap_info_struct *p; |
1da177e4 | 2822 | unsigned int type; |
a2468cc9 | 2823 | int i; |
efa90a98 HD |
2824 | |
2825 | p = kzalloc(sizeof(*p), GFP_KERNEL); | |
2826 | if (!p) | |
53cbb243 | 2827 | return ERR_PTR(-ENOMEM); |
efa90a98 | 2828 | |
5d337b91 | 2829 | spin_lock(&swap_lock); |
efa90a98 HD |
2830 | for (type = 0; type < nr_swapfiles; type++) { |
2831 | if (!(swap_info[type]->flags & SWP_USED)) | |
1da177e4 | 2832 | break; |
efa90a98 | 2833 | } |
0697212a | 2834 | if (type >= MAX_SWAPFILES) { |
5d337b91 | 2835 | spin_unlock(&swap_lock); |
efa90a98 | 2836 | kfree(p); |
730c0581 | 2837 | return ERR_PTR(-EPERM); |
1da177e4 | 2838 | } |
efa90a98 HD |
2839 | if (type >= nr_swapfiles) { |
2840 | p->type = type; | |
2841 | swap_info[type] = p; | |
2842 | /* | |
2843 | * Write swap_info[type] before nr_swapfiles, in case a | |
2844 | * racing procfs swap_start() or swap_next() is reading them. | |
2845 | * (We never shrink nr_swapfiles, we never free this entry.) | |
2846 | */ | |
2847 | smp_wmb(); | |
2848 | nr_swapfiles++; | |
2849 | } else { | |
2850 | kfree(p); | |
2851 | p = swap_info[type]; | |
2852 | /* | |
2853 | * Do not memset this entry: a racing procfs swap_next() | |
2854 | * would be relying on p->type to remain valid. | |
2855 | */ | |
2856 | } | |
9625a5f2 | 2857 | INIT_LIST_HEAD(&p->first_swap_extent.list); |
18ab4d4c | 2858 | plist_node_init(&p->list, 0); |
a2468cc9 AL |
2859 | for_each_node(i) |
2860 | plist_node_init(&p->avail_lists[i], 0); | |
1da177e4 | 2861 | p->flags = SWP_USED; |
5d337b91 | 2862 | spin_unlock(&swap_lock); |
ec8acf20 | 2863 | spin_lock_init(&p->lock); |
2628bd6f | 2864 | spin_lock_init(&p->cont_lock); |
efa90a98 | 2865 | |
53cbb243 | 2866 | return p; |
53cbb243 CEB |
2867 | } |
2868 | ||
4d0e1e10 CEB |
2869 | static int claim_swapfile(struct swap_info_struct *p, struct inode *inode) |
2870 | { | |
2871 | int error; | |
2872 | ||
2873 | if (S_ISBLK(inode->i_mode)) { | |
2874 | p->bdev = bdgrab(I_BDEV(inode)); | |
2875 | error = blkdev_get(p->bdev, | |
6f179af8 | 2876 | FMODE_READ | FMODE_WRITE | FMODE_EXCL, p); |
4d0e1e10 CEB |
2877 | if (error < 0) { |
2878 | p->bdev = NULL; | |
6f179af8 | 2879 | return error; |
4d0e1e10 CEB |
2880 | } |
2881 | p->old_block_size = block_size(p->bdev); | |
2882 | error = set_blocksize(p->bdev, PAGE_SIZE); | |
2883 | if (error < 0) | |
87ade72a | 2884 | return error; |
4d0e1e10 CEB |
2885 | p->flags |= SWP_BLKDEV; |
2886 | } else if (S_ISREG(inode->i_mode)) { | |
2887 | p->bdev = inode->i_sb->s_bdev; | |
5955102c | 2888 | inode_lock(inode); |
87ade72a CEB |
2889 | if (IS_SWAPFILE(inode)) |
2890 | return -EBUSY; | |
2891 | } else | |
2892 | return -EINVAL; | |
4d0e1e10 CEB |
2893 | |
2894 | return 0; | |
4d0e1e10 CEB |
2895 | } |
2896 | ||
377eeaa8 AK |
2897 | |
2898 | /* | |
2899 | * Find out how many pages are allowed for a single swap device. There | |
2900 | * are two limiting factors: | |
2901 | * 1) the number of bits for the swap offset in the swp_entry_t type, and | |
2902 | * 2) the number of bits in the swap pte, as defined by the different | |
2903 | * architectures. | |
2904 | * | |
2905 | * In order to find the largest possible bit mask, a swap entry with | |
2906 | * swap type 0 and swap offset ~0UL is created, encoded to a swap pte, | |
2907 | * decoded to a swp_entry_t again, and finally the swap offset is | |
2908 | * extracted. | |
2909 | * | |
2910 | * This will mask all the bits from the initial ~0UL mask that can't | |
2911 | * be encoded in either the swp_entry_t or the architecture definition | |
2912 | * of a swap pte. | |
2913 | */ | |
2914 | unsigned long generic_max_swapfile_size(void) | |
2915 | { | |
2916 | return swp_offset(pte_to_swp_entry( | |
2917 | swp_entry_to_pte(swp_entry(0, ~0UL)))) + 1; | |
2918 | } | |
2919 | ||
2920 | /* Can be overridden by an architecture for additional checks. */ | |
2921 | __weak unsigned long max_swapfile_size(void) | |
2922 | { | |
2923 | return generic_max_swapfile_size(); | |
2924 | } | |
2925 | ||
ca8bd38b CEB |
2926 | static unsigned long read_swap_header(struct swap_info_struct *p, |
2927 | union swap_header *swap_header, | |
2928 | struct inode *inode) | |
2929 | { | |
2930 | int i; | |
2931 | unsigned long maxpages; | |
2932 | unsigned long swapfilepages; | |
d6bbbd29 | 2933 | unsigned long last_page; |
ca8bd38b CEB |
2934 | |
2935 | if (memcmp("SWAPSPACE2", swap_header->magic.magic, 10)) { | |
465c47fd | 2936 | pr_err("Unable to find swap-space signature\n"); |
38719025 | 2937 | return 0; |
ca8bd38b CEB |
2938 | } |
2939 | ||
2940 | /* swap partition endianess hack... */ | |
2941 | if (swab32(swap_header->info.version) == 1) { | |
2942 | swab32s(&swap_header->info.version); | |
2943 | swab32s(&swap_header->info.last_page); | |
2944 | swab32s(&swap_header->info.nr_badpages); | |
dd111be6 JH |
2945 | if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES) |
2946 | return 0; | |
ca8bd38b CEB |
2947 | for (i = 0; i < swap_header->info.nr_badpages; i++) |
2948 | swab32s(&swap_header->info.badpages[i]); | |
2949 | } | |
2950 | /* Check the swap header's sub-version */ | |
2951 | if (swap_header->info.version != 1) { | |
465c47fd AM |
2952 | pr_warn("Unable to handle swap header version %d\n", |
2953 | swap_header->info.version); | |
38719025 | 2954 | return 0; |
ca8bd38b CEB |
2955 | } |
2956 | ||
2957 | p->lowest_bit = 1; | |
2958 | p->cluster_next = 1; | |
2959 | p->cluster_nr = 0; | |
2960 | ||
377eeaa8 | 2961 | maxpages = max_swapfile_size(); |
d6bbbd29 | 2962 | last_page = swap_header->info.last_page; |
a06ad633 TA |
2963 | if (!last_page) { |
2964 | pr_warn("Empty swap-file\n"); | |
2965 | return 0; | |
2966 | } | |
d6bbbd29 | 2967 | if (last_page > maxpages) { |
465c47fd | 2968 | pr_warn("Truncating oversized swap area, only using %luk out of %luk\n", |
d6bbbd29 RJ |
2969 | maxpages << (PAGE_SHIFT - 10), |
2970 | last_page << (PAGE_SHIFT - 10)); | |
2971 | } | |
2972 | if (maxpages > last_page) { | |
2973 | maxpages = last_page + 1; | |
ca8bd38b CEB |
2974 | /* p->max is an unsigned int: don't overflow it */ |
2975 | if ((unsigned int)maxpages == 0) | |
2976 | maxpages = UINT_MAX; | |
2977 | } | |
2978 | p->highest_bit = maxpages - 1; | |
2979 | ||
2980 | if (!maxpages) | |
38719025 | 2981 | return 0; |
ca8bd38b CEB |
2982 | swapfilepages = i_size_read(inode) >> PAGE_SHIFT; |
2983 | if (swapfilepages && maxpages > swapfilepages) { | |
465c47fd | 2984 | pr_warn("Swap area shorter than signature indicates\n"); |
38719025 | 2985 | return 0; |
ca8bd38b CEB |
2986 | } |
2987 | if (swap_header->info.nr_badpages && S_ISREG(inode->i_mode)) | |
38719025 | 2988 | return 0; |
ca8bd38b | 2989 | if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES) |
38719025 | 2990 | return 0; |
ca8bd38b CEB |
2991 | |
2992 | return maxpages; | |
ca8bd38b CEB |
2993 | } |
2994 | ||
4b3ef9da | 2995 | #define SWAP_CLUSTER_INFO_COLS \ |
235b6217 | 2996 | DIV_ROUND_UP(L1_CACHE_BYTES, sizeof(struct swap_cluster_info)) |
4b3ef9da HY |
2997 | #define SWAP_CLUSTER_SPACE_COLS \ |
2998 | DIV_ROUND_UP(SWAP_ADDRESS_SPACE_PAGES, SWAPFILE_CLUSTER) | |
2999 | #define SWAP_CLUSTER_COLS \ | |
3000 | max_t(unsigned int, SWAP_CLUSTER_INFO_COLS, SWAP_CLUSTER_SPACE_COLS) | |
235b6217 | 3001 | |
915d4d7b CEB |
3002 | static int setup_swap_map_and_extents(struct swap_info_struct *p, |
3003 | union swap_header *swap_header, | |
3004 | unsigned char *swap_map, | |
2a8f9449 | 3005 | struct swap_cluster_info *cluster_info, |
915d4d7b CEB |
3006 | unsigned long maxpages, |
3007 | sector_t *span) | |
3008 | { | |
235b6217 | 3009 | unsigned int j, k; |
915d4d7b CEB |
3010 | unsigned int nr_good_pages; |
3011 | int nr_extents; | |
2a8f9449 | 3012 | unsigned long nr_clusters = DIV_ROUND_UP(maxpages, SWAPFILE_CLUSTER); |
235b6217 HY |
3013 | unsigned long col = p->cluster_next / SWAPFILE_CLUSTER % SWAP_CLUSTER_COLS; |
3014 | unsigned long i, idx; | |
915d4d7b CEB |
3015 | |
3016 | nr_good_pages = maxpages - 1; /* omit header page */ | |
3017 | ||
6b534915 HY |
3018 | cluster_list_init(&p->free_clusters); |
3019 | cluster_list_init(&p->discard_clusters); | |
2a8f9449 | 3020 | |
915d4d7b CEB |
3021 | for (i = 0; i < swap_header->info.nr_badpages; i++) { |
3022 | unsigned int page_nr = swap_header->info.badpages[i]; | |
bdb8e3f6 CEB |
3023 | if (page_nr == 0 || page_nr > swap_header->info.last_page) |
3024 | return -EINVAL; | |
915d4d7b CEB |
3025 | if (page_nr < maxpages) { |
3026 | swap_map[page_nr] = SWAP_MAP_BAD; | |
3027 | nr_good_pages--; | |
2a8f9449 SL |
3028 | /* |
3029 | * Haven't marked the cluster free yet, no list | |
3030 | * operation involved | |
3031 | */ | |
3032 | inc_cluster_info_page(p, cluster_info, page_nr); | |
915d4d7b CEB |
3033 | } |
3034 | } | |
3035 | ||
2a8f9449 SL |
3036 | /* Haven't marked the cluster free yet, no list operation involved */ |
3037 | for (i = maxpages; i < round_up(maxpages, SWAPFILE_CLUSTER); i++) | |
3038 | inc_cluster_info_page(p, cluster_info, i); | |
3039 | ||
915d4d7b CEB |
3040 | if (nr_good_pages) { |
3041 | swap_map[0] = SWAP_MAP_BAD; | |
2a8f9449 SL |
3042 | /* |
3043 | * Not mark the cluster free yet, no list | |
3044 | * operation involved | |
3045 | */ | |
3046 | inc_cluster_info_page(p, cluster_info, 0); | |
915d4d7b CEB |
3047 | p->max = maxpages; |
3048 | p->pages = nr_good_pages; | |
3049 | nr_extents = setup_swap_extents(p, span); | |
bdb8e3f6 CEB |
3050 | if (nr_extents < 0) |
3051 | return nr_extents; | |
915d4d7b CEB |
3052 | nr_good_pages = p->pages; |
3053 | } | |
3054 | if (!nr_good_pages) { | |
465c47fd | 3055 | pr_warn("Empty swap-file\n"); |
bdb8e3f6 | 3056 | return -EINVAL; |
915d4d7b CEB |
3057 | } |
3058 | ||
2a8f9449 SL |
3059 | if (!cluster_info) |
3060 | return nr_extents; | |
3061 | ||
235b6217 | 3062 | |
4b3ef9da HY |
3063 | /* |
3064 | * Reduce false cache line sharing between cluster_info and | |
3065 | * sharing same address space. | |
3066 | */ | |
235b6217 HY |
3067 | for (k = 0; k < SWAP_CLUSTER_COLS; k++) { |
3068 | j = (k + col) % SWAP_CLUSTER_COLS; | |
3069 | for (i = 0; i < DIV_ROUND_UP(nr_clusters, SWAP_CLUSTER_COLS); i++) { | |
3070 | idx = i * SWAP_CLUSTER_COLS + j; | |
3071 | if (idx >= nr_clusters) | |
3072 | continue; | |
3073 | if (cluster_count(&cluster_info[idx])) | |
3074 | continue; | |
2a8f9449 | 3075 | cluster_set_flag(&cluster_info[idx], CLUSTER_FLAG_FREE); |
6b534915 HY |
3076 | cluster_list_add_tail(&p->free_clusters, cluster_info, |
3077 | idx); | |
2a8f9449 | 3078 | } |
2a8f9449 | 3079 | } |
915d4d7b | 3080 | return nr_extents; |
915d4d7b CEB |
3081 | } |
3082 | ||
dcf6b7dd RA |
3083 | /* |
3084 | * Helper to sys_swapon determining if a given swap | |
3085 | * backing device queue supports DISCARD operations. | |
3086 | */ | |
3087 | static bool swap_discardable(struct swap_info_struct *si) | |
3088 | { | |
3089 | struct request_queue *q = bdev_get_queue(si->bdev); | |
3090 | ||
3091 | if (!q || !blk_queue_discard(q)) | |
3092 | return false; | |
3093 | ||
3094 | return true; | |
3095 | } | |
3096 | ||
53cbb243 CEB |
3097 | SYSCALL_DEFINE2(swapon, const char __user *, specialfile, int, swap_flags) |
3098 | { | |
3099 | struct swap_info_struct *p; | |
91a27b2a | 3100 | struct filename *name; |
53cbb243 CEB |
3101 | struct file *swap_file = NULL; |
3102 | struct address_space *mapping; | |
40531542 | 3103 | int prio; |
53cbb243 CEB |
3104 | int error; |
3105 | union swap_header *swap_header; | |
915d4d7b | 3106 | int nr_extents; |
53cbb243 CEB |
3107 | sector_t span; |
3108 | unsigned long maxpages; | |
53cbb243 | 3109 | unsigned char *swap_map = NULL; |
2a8f9449 | 3110 | struct swap_cluster_info *cluster_info = NULL; |
38b5faf4 | 3111 | unsigned long *frontswap_map = NULL; |
53cbb243 CEB |
3112 | struct page *page = NULL; |
3113 | struct inode *inode = NULL; | |
7cbf3192 | 3114 | bool inced_nr_rotate_swap = false; |
53cbb243 | 3115 | |
d15cab97 HD |
3116 | if (swap_flags & ~SWAP_FLAGS_VALID) |
3117 | return -EINVAL; | |
3118 | ||
53cbb243 CEB |
3119 | if (!capable(CAP_SYS_ADMIN)) |
3120 | return -EPERM; | |
3121 | ||
a2468cc9 AL |
3122 | if (!swap_avail_heads) |
3123 | return -ENOMEM; | |
3124 | ||
53cbb243 | 3125 | p = alloc_swap_info(); |
2542e513 CEB |
3126 | if (IS_ERR(p)) |
3127 | return PTR_ERR(p); | |
53cbb243 | 3128 | |
815c2c54 SL |
3129 | INIT_WORK(&p->discard_work, swap_discard_work); |
3130 | ||
1da177e4 | 3131 | name = getname(specialfile); |
1da177e4 | 3132 | if (IS_ERR(name)) { |
7de7fb6b | 3133 | error = PTR_ERR(name); |
1da177e4 | 3134 | name = NULL; |
bd69010b | 3135 | goto bad_swap; |
1da177e4 | 3136 | } |
669abf4e | 3137 | swap_file = file_open_name(name, O_RDWR|O_LARGEFILE, 0); |
1da177e4 | 3138 | if (IS_ERR(swap_file)) { |
7de7fb6b | 3139 | error = PTR_ERR(swap_file); |
1da177e4 | 3140 | swap_file = NULL; |
bd69010b | 3141 | goto bad_swap; |
1da177e4 LT |
3142 | } |
3143 | ||
3144 | p->swap_file = swap_file; | |
3145 | mapping = swap_file->f_mapping; | |
2130781e | 3146 | inode = mapping->host; |
6f179af8 | 3147 | |
5955102c | 3148 | /* If S_ISREG(inode->i_mode) will do inode_lock(inode); */ |
4d0e1e10 CEB |
3149 | error = claim_swapfile(p, inode); |
3150 | if (unlikely(error)) | |
1da177e4 | 3151 | goto bad_swap; |
1da177e4 | 3152 | |
1da177e4 LT |
3153 | /* |
3154 | * Read the swap header. | |
3155 | */ | |
3156 | if (!mapping->a_ops->readpage) { | |
3157 | error = -EINVAL; | |
3158 | goto bad_swap; | |
3159 | } | |
090d2b18 | 3160 | page = read_mapping_page(mapping, 0, swap_file); |
1da177e4 LT |
3161 | if (IS_ERR(page)) { |
3162 | error = PTR_ERR(page); | |
3163 | goto bad_swap; | |
3164 | } | |
81e33971 | 3165 | swap_header = kmap(page); |
1da177e4 | 3166 | |
ca8bd38b CEB |
3167 | maxpages = read_swap_header(p, swap_header, inode); |
3168 | if (unlikely(!maxpages)) { | |
1da177e4 LT |
3169 | error = -EINVAL; |
3170 | goto bad_swap; | |
3171 | } | |
886bb7e9 | 3172 | |
81e33971 | 3173 | /* OK, set up the swap map and apply the bad block list */ |
803d0c83 | 3174 | swap_map = vzalloc(maxpages); |
81e33971 HD |
3175 | if (!swap_map) { |
3176 | error = -ENOMEM; | |
3177 | goto bad_swap; | |
3178 | } | |
f0571429 MK |
3179 | |
3180 | if (bdi_cap_stable_pages_required(inode_to_bdi(inode))) | |
3181 | p->flags |= SWP_STABLE_WRITES; | |
3182 | ||
539a6fea MK |
3183 | if (bdi_cap_synchronous_io(inode_to_bdi(inode))) |
3184 | p->flags |= SWP_SYNCHRONOUS_IO; | |
3185 | ||
2a8f9449 | 3186 | if (p->bdev && blk_queue_nonrot(bdev_get_queue(p->bdev))) { |
6f179af8 | 3187 | int cpu; |
235b6217 | 3188 | unsigned long ci, nr_cluster; |
6f179af8 | 3189 | |
2a8f9449 SL |
3190 | p->flags |= SWP_SOLIDSTATE; |
3191 | /* | |
3192 | * select a random position to start with to help wear leveling | |
3193 | * SSD | |
3194 | */ | |
3195 | p->cluster_next = 1 + (prandom_u32() % p->highest_bit); | |
235b6217 | 3196 | nr_cluster = DIV_ROUND_UP(maxpages, SWAPFILE_CLUSTER); |
2a8f9449 | 3197 | |
778e1cdd | 3198 | cluster_info = kvcalloc(nr_cluster, sizeof(*cluster_info), |
54f180d3 | 3199 | GFP_KERNEL); |
2a8f9449 SL |
3200 | if (!cluster_info) { |
3201 | error = -ENOMEM; | |
3202 | goto bad_swap; | |
3203 | } | |
235b6217 HY |
3204 | |
3205 | for (ci = 0; ci < nr_cluster; ci++) | |
3206 | spin_lock_init(&((cluster_info + ci)->lock)); | |
3207 | ||
ebc2a1a6 SL |
3208 | p->percpu_cluster = alloc_percpu(struct percpu_cluster); |
3209 | if (!p->percpu_cluster) { | |
3210 | error = -ENOMEM; | |
3211 | goto bad_swap; | |
3212 | } | |
6f179af8 | 3213 | for_each_possible_cpu(cpu) { |
ebc2a1a6 | 3214 | struct percpu_cluster *cluster; |
6f179af8 | 3215 | cluster = per_cpu_ptr(p->percpu_cluster, cpu); |
ebc2a1a6 SL |
3216 | cluster_set_null(&cluster->index); |
3217 | } | |
7cbf3192 | 3218 | } else { |
81a0298b | 3219 | atomic_inc(&nr_rotate_swap); |
7cbf3192 OS |
3220 | inced_nr_rotate_swap = true; |
3221 | } | |
1da177e4 | 3222 | |
1421ef3c CEB |
3223 | error = swap_cgroup_swapon(p->type, maxpages); |
3224 | if (error) | |
3225 | goto bad_swap; | |
3226 | ||
915d4d7b | 3227 | nr_extents = setup_swap_map_and_extents(p, swap_header, swap_map, |
2a8f9449 | 3228 | cluster_info, maxpages, &span); |
915d4d7b CEB |
3229 | if (unlikely(nr_extents < 0)) { |
3230 | error = nr_extents; | |
1da177e4 LT |
3231 | goto bad_swap; |
3232 | } | |
38b5faf4 | 3233 | /* frontswap enabled? set up bit-per-page map for frontswap */ |
8ea1d2a1 | 3234 | if (IS_ENABLED(CONFIG_FRONTSWAP)) |
778e1cdd KC |
3235 | frontswap_map = kvcalloc(BITS_TO_LONGS(maxpages), |
3236 | sizeof(long), | |
54f180d3 | 3237 | GFP_KERNEL); |
1da177e4 | 3238 | |
2a8f9449 SL |
3239 | if (p->bdev &&(swap_flags & SWAP_FLAG_DISCARD) && swap_discardable(p)) { |
3240 | /* | |
3241 | * When discard is enabled for swap with no particular | |
3242 | * policy flagged, we set all swap discard flags here in | |
3243 | * order to sustain backward compatibility with older | |
3244 | * swapon(8) releases. | |
3245 | */ | |
3246 | p->flags |= (SWP_DISCARDABLE | SWP_AREA_DISCARD | | |
3247 | SWP_PAGE_DISCARD); | |
dcf6b7dd | 3248 | |
2a8f9449 SL |
3249 | /* |
3250 | * By flagging sys_swapon, a sysadmin can tell us to | |
3251 | * either do single-time area discards only, or to just | |
3252 | * perform discards for released swap page-clusters. | |
3253 | * Now it's time to adjust the p->flags accordingly. | |
3254 | */ | |
3255 | if (swap_flags & SWAP_FLAG_DISCARD_ONCE) | |
3256 | p->flags &= ~SWP_PAGE_DISCARD; | |
3257 | else if (swap_flags & SWAP_FLAG_DISCARD_PAGES) | |
3258 | p->flags &= ~SWP_AREA_DISCARD; | |
3259 | ||
3260 | /* issue a swapon-time discard if it's still required */ | |
3261 | if (p->flags & SWP_AREA_DISCARD) { | |
3262 | int err = discard_swap(p); | |
3263 | if (unlikely(err)) | |
3264 | pr_err("swapon: discard_swap(%p): %d\n", | |
3265 | p, err); | |
dcf6b7dd | 3266 | } |
20137a49 | 3267 | } |
6a6ba831 | 3268 | |
4b3ef9da HY |
3269 | error = init_swap_address_space(p->type, maxpages); |
3270 | if (error) | |
3271 | goto bad_swap; | |
3272 | ||
fc0abb14 | 3273 | mutex_lock(&swapon_mutex); |
40531542 | 3274 | prio = -1; |
78ecba08 | 3275 | if (swap_flags & SWAP_FLAG_PREFER) |
40531542 | 3276 | prio = |
78ecba08 | 3277 | (swap_flags & SWAP_FLAG_PRIO_MASK) >> SWAP_FLAG_PRIO_SHIFT; |
2a8f9449 | 3278 | enable_swap_info(p, prio, swap_map, cluster_info, frontswap_map); |
c69dbfb8 | 3279 | |
756a025f | 3280 | pr_info("Adding %uk swap on %s. Priority:%d extents:%d across:%lluk %s%s%s%s%s\n", |
91a27b2a | 3281 | p->pages<<(PAGE_SHIFT-10), name->name, p->prio, |
c69dbfb8 CEB |
3282 | nr_extents, (unsigned long long)span<<(PAGE_SHIFT-10), |
3283 | (p->flags & SWP_SOLIDSTATE) ? "SS" : "", | |
38b5faf4 | 3284 | (p->flags & SWP_DISCARDABLE) ? "D" : "", |
dcf6b7dd RA |
3285 | (p->flags & SWP_AREA_DISCARD) ? "s" : "", |
3286 | (p->flags & SWP_PAGE_DISCARD) ? "c" : "", | |
38b5faf4 | 3287 | (frontswap_map) ? "FS" : ""); |
c69dbfb8 | 3288 | |
fc0abb14 | 3289 | mutex_unlock(&swapon_mutex); |
66d7dd51 KS |
3290 | atomic_inc(&proc_poll_event); |
3291 | wake_up_interruptible(&proc_poll_wait); | |
3292 | ||
9b01c350 CEB |
3293 | if (S_ISREG(inode->i_mode)) |
3294 | inode->i_flags |= S_SWAPFILE; | |
1da177e4 LT |
3295 | error = 0; |
3296 | goto out; | |
3297 | bad_swap: | |
ebc2a1a6 SL |
3298 | free_percpu(p->percpu_cluster); |
3299 | p->percpu_cluster = NULL; | |
bd69010b | 3300 | if (inode && S_ISBLK(inode->i_mode) && p->bdev) { |
f2090d2d CEB |
3301 | set_blocksize(p->bdev, p->old_block_size); |
3302 | blkdev_put(p->bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL); | |
1da177e4 | 3303 | } |
4cd3bb10 | 3304 | destroy_swap_extents(p); |
e8e6c2ec | 3305 | swap_cgroup_swapoff(p->type); |
5d337b91 | 3306 | spin_lock(&swap_lock); |
1da177e4 | 3307 | p->swap_file = NULL; |
1da177e4 | 3308 | p->flags = 0; |
5d337b91 | 3309 | spin_unlock(&swap_lock); |
1da177e4 | 3310 | vfree(swap_map); |
8606a1a9 | 3311 | kvfree(cluster_info); |
b6b1fd2a | 3312 | kvfree(frontswap_map); |
7cbf3192 OS |
3313 | if (inced_nr_rotate_swap) |
3314 | atomic_dec(&nr_rotate_swap); | |
52c50567 | 3315 | if (swap_file) { |
2130781e | 3316 | if (inode && S_ISREG(inode->i_mode)) { |
5955102c | 3317 | inode_unlock(inode); |
2130781e CEB |
3318 | inode = NULL; |
3319 | } | |
1da177e4 | 3320 | filp_close(swap_file, NULL); |
52c50567 | 3321 | } |
1da177e4 LT |
3322 | out: |
3323 | if (page && !IS_ERR(page)) { | |
3324 | kunmap(page); | |
09cbfeaf | 3325 | put_page(page); |
1da177e4 LT |
3326 | } |
3327 | if (name) | |
3328 | putname(name); | |
9b01c350 | 3329 | if (inode && S_ISREG(inode->i_mode)) |
5955102c | 3330 | inode_unlock(inode); |
039939a6 TC |
3331 | if (!error) |
3332 | enable_swap_slots_cache(); | |
1da177e4 LT |
3333 | return error; |
3334 | } | |
3335 | ||
3336 | void si_swapinfo(struct sysinfo *val) | |
3337 | { | |
efa90a98 | 3338 | unsigned int type; |
1da177e4 LT |
3339 | unsigned long nr_to_be_unused = 0; |
3340 | ||
5d337b91 | 3341 | spin_lock(&swap_lock); |
efa90a98 HD |
3342 | for (type = 0; type < nr_swapfiles; type++) { |
3343 | struct swap_info_struct *si = swap_info[type]; | |
3344 | ||
3345 | if ((si->flags & SWP_USED) && !(si->flags & SWP_WRITEOK)) | |
3346 | nr_to_be_unused += si->inuse_pages; | |
1da177e4 | 3347 | } |
ec8acf20 | 3348 | val->freeswap = atomic_long_read(&nr_swap_pages) + nr_to_be_unused; |
1da177e4 | 3349 | val->totalswap = total_swap_pages + nr_to_be_unused; |
5d337b91 | 3350 | spin_unlock(&swap_lock); |
1da177e4 LT |
3351 | } |
3352 | ||
3353 | /* | |
3354 | * Verify that a swap entry is valid and increment its swap map count. | |
3355 | * | |
355cfa73 KH |
3356 | * Returns error code in following case. |
3357 | * - success -> 0 | |
3358 | * - swp_entry is invalid -> EINVAL | |
3359 | * - swp_entry is migration entry -> EINVAL | |
3360 | * - swap-cache reference is requested but there is already one. -> EEXIST | |
3361 | * - swap-cache reference is requested but the entry is not used. -> ENOENT | |
570a335b | 3362 | * - swap-mapped reference requested but needs continued swap count. -> ENOMEM |
1da177e4 | 3363 | */ |
8d69aaee | 3364 | static int __swap_duplicate(swp_entry_t entry, unsigned char usage) |
1da177e4 | 3365 | { |
73c34b6a | 3366 | struct swap_info_struct *p; |
235b6217 | 3367 | struct swap_cluster_info *ci; |
1da177e4 | 3368 | unsigned long offset, type; |
8d69aaee HD |
3369 | unsigned char count; |
3370 | unsigned char has_cache; | |
253d553b | 3371 | int err = -EINVAL; |
1da177e4 | 3372 | |
a7420aa5 | 3373 | if (non_swap_entry(entry)) |
253d553b | 3374 | goto out; |
0697212a | 3375 | |
1da177e4 LT |
3376 | type = swp_type(entry); |
3377 | if (type >= nr_swapfiles) | |
3378 | goto bad_file; | |
efa90a98 | 3379 | p = swap_info[type]; |
1da177e4 | 3380 | offset = swp_offset(entry); |
355cfa73 | 3381 | if (unlikely(offset >= p->max)) |
235b6217 HY |
3382 | goto out; |
3383 | ||
3384 | ci = lock_cluster_or_swap_info(p, offset); | |
355cfa73 | 3385 | |
253d553b | 3386 | count = p->swap_map[offset]; |
edfe23da SL |
3387 | |
3388 | /* | |
3389 | * swapin_readahead() doesn't check if a swap entry is valid, so the | |
3390 | * swap entry could be SWAP_MAP_BAD. Check here with lock held. | |
3391 | */ | |
3392 | if (unlikely(swap_count(count) == SWAP_MAP_BAD)) { | |
3393 | err = -ENOENT; | |
3394 | goto unlock_out; | |
3395 | } | |
3396 | ||
253d553b HD |
3397 | has_cache = count & SWAP_HAS_CACHE; |
3398 | count &= ~SWAP_HAS_CACHE; | |
3399 | err = 0; | |
355cfa73 | 3400 | |
253d553b | 3401 | if (usage == SWAP_HAS_CACHE) { |
355cfa73 KH |
3402 | |
3403 | /* set SWAP_HAS_CACHE if there is no cache and entry is used */ | |
253d553b HD |
3404 | if (!has_cache && count) |
3405 | has_cache = SWAP_HAS_CACHE; | |
3406 | else if (has_cache) /* someone else added cache */ | |
3407 | err = -EEXIST; | |
3408 | else /* no users remaining */ | |
3409 | err = -ENOENT; | |
355cfa73 KH |
3410 | |
3411 | } else if (count || has_cache) { | |
253d553b | 3412 | |
570a335b HD |
3413 | if ((count & ~COUNT_CONTINUED) < SWAP_MAP_MAX) |
3414 | count += usage; | |
3415 | else if ((count & ~COUNT_CONTINUED) > SWAP_MAP_MAX) | |
253d553b | 3416 | err = -EINVAL; |
570a335b HD |
3417 | else if (swap_count_continued(p, offset, count)) |
3418 | count = COUNT_CONTINUED; | |
3419 | else | |
3420 | err = -ENOMEM; | |
355cfa73 | 3421 | } else |
253d553b HD |
3422 | err = -ENOENT; /* unused swap entry */ |
3423 | ||
3424 | p->swap_map[offset] = count | has_cache; | |
3425 | ||
355cfa73 | 3426 | unlock_out: |
235b6217 | 3427 | unlock_cluster_or_swap_info(p, ci); |
1da177e4 | 3428 | out: |
253d553b | 3429 | return err; |
1da177e4 LT |
3430 | |
3431 | bad_file: | |
465c47fd | 3432 | pr_err("swap_dup: %s%08lx\n", Bad_file, entry.val); |
1da177e4 LT |
3433 | goto out; |
3434 | } | |
253d553b | 3435 | |
aaa46865 HD |
3436 | /* |
3437 | * Help swapoff by noting that swap entry belongs to shmem/tmpfs | |
3438 | * (in which case its reference count is never incremented). | |
3439 | */ | |
3440 | void swap_shmem_alloc(swp_entry_t entry) | |
3441 | { | |
3442 | __swap_duplicate(entry, SWAP_MAP_SHMEM); | |
3443 | } | |
3444 | ||
355cfa73 | 3445 | /* |
08259d58 HD |
3446 | * Increase reference count of swap entry by 1. |
3447 | * Returns 0 for success, or -ENOMEM if a swap_count_continuation is required | |
3448 | * but could not be atomically allocated. Returns 0, just as if it succeeded, | |
3449 | * if __swap_duplicate() fails for another reason (-EINVAL or -ENOENT), which | |
3450 | * might occur if a page table entry has got corrupted. | |
355cfa73 | 3451 | */ |
570a335b | 3452 | int swap_duplicate(swp_entry_t entry) |
355cfa73 | 3453 | { |
570a335b HD |
3454 | int err = 0; |
3455 | ||
3456 | while (!err && __swap_duplicate(entry, 1) == -ENOMEM) | |
3457 | err = add_swap_count_continuation(entry, GFP_ATOMIC); | |
3458 | return err; | |
355cfa73 | 3459 | } |
1da177e4 | 3460 | |
cb4b86ba | 3461 | /* |
355cfa73 KH |
3462 | * @entry: swap entry for which we allocate swap cache. |
3463 | * | |
73c34b6a | 3464 | * Called when allocating swap cache for existing swap entry, |
355cfa73 KH |
3465 | * This can return error codes. Returns 0 at success. |
3466 | * -EBUSY means there is a swap cache. | |
3467 | * Note: return code is different from swap_duplicate(). | |
cb4b86ba KH |
3468 | */ |
3469 | int swapcache_prepare(swp_entry_t entry) | |
3470 | { | |
253d553b | 3471 | return __swap_duplicate(entry, SWAP_HAS_CACHE); |
cb4b86ba KH |
3472 | } |
3473 | ||
0bcac06f MK |
3474 | struct swap_info_struct *swp_swap_info(swp_entry_t entry) |
3475 | { | |
3476 | return swap_info[swp_type(entry)]; | |
3477 | } | |
3478 | ||
f981c595 MG |
3479 | struct swap_info_struct *page_swap_info(struct page *page) |
3480 | { | |
0bcac06f MK |
3481 | swp_entry_t entry = { .val = page_private(page) }; |
3482 | return swp_swap_info(entry); | |
f981c595 MG |
3483 | } |
3484 | ||
3485 | /* | |
3486 | * out-of-line __page_file_ methods to avoid include hell. | |
3487 | */ | |
3488 | struct address_space *__page_file_mapping(struct page *page) | |
3489 | { | |
f981c595 MG |
3490 | return page_swap_info(page)->swap_file->f_mapping; |
3491 | } | |
3492 | EXPORT_SYMBOL_GPL(__page_file_mapping); | |
3493 | ||
3494 | pgoff_t __page_file_index(struct page *page) | |
3495 | { | |
3496 | swp_entry_t swap = { .val = page_private(page) }; | |
f981c595 MG |
3497 | return swp_offset(swap); |
3498 | } | |
3499 | EXPORT_SYMBOL_GPL(__page_file_index); | |
3500 | ||
570a335b HD |
3501 | /* |
3502 | * add_swap_count_continuation - called when a swap count is duplicated | |
3503 | * beyond SWAP_MAP_MAX, it allocates a new page and links that to the entry's | |
3504 | * page of the original vmalloc'ed swap_map, to hold the continuation count | |
3505 | * (for that entry and for its neighbouring PAGE_SIZE swap entries). Called | |
3506 | * again when count is duplicated beyond SWAP_MAP_MAX * SWAP_CONT_MAX, etc. | |
3507 | * | |
3508 | * These continuation pages are seldom referenced: the common paths all work | |
3509 | * on the original swap_map, only referring to a continuation page when the | |
3510 | * low "digit" of a count is incremented or decremented through SWAP_MAP_MAX. | |
3511 | * | |
3512 | * add_swap_count_continuation(, GFP_ATOMIC) can be called while holding | |
3513 | * page table locks; if it fails, add_swap_count_continuation(, GFP_KERNEL) | |
3514 | * can be called after dropping locks. | |
3515 | */ | |
3516 | int add_swap_count_continuation(swp_entry_t entry, gfp_t gfp_mask) | |
3517 | { | |
3518 | struct swap_info_struct *si; | |
235b6217 | 3519 | struct swap_cluster_info *ci; |
570a335b HD |
3520 | struct page *head; |
3521 | struct page *page; | |
3522 | struct page *list_page; | |
3523 | pgoff_t offset; | |
3524 | unsigned char count; | |
3525 | ||
3526 | /* | |
3527 | * When debugging, it's easier to use __GFP_ZERO here; but it's better | |
3528 | * for latency not to zero a page while GFP_ATOMIC and holding locks. | |
3529 | */ | |
3530 | page = alloc_page(gfp_mask | __GFP_HIGHMEM); | |
3531 | ||
3532 | si = swap_info_get(entry); | |
3533 | if (!si) { | |
3534 | /* | |
3535 | * An acceptable race has occurred since the failing | |
3536 | * __swap_duplicate(): the swap entry has been freed, | |
3537 | * perhaps even the whole swap_map cleared for swapoff. | |
3538 | */ | |
3539 | goto outer; | |
3540 | } | |
3541 | ||
3542 | offset = swp_offset(entry); | |
235b6217 HY |
3543 | |
3544 | ci = lock_cluster(si, offset); | |
3545 | ||
570a335b HD |
3546 | count = si->swap_map[offset] & ~SWAP_HAS_CACHE; |
3547 | ||
3548 | if ((count & ~COUNT_CONTINUED) != SWAP_MAP_MAX) { | |
3549 | /* | |
3550 | * The higher the swap count, the more likely it is that tasks | |
3551 | * will race to add swap count continuation: we need to avoid | |
3552 | * over-provisioning. | |
3553 | */ | |
3554 | goto out; | |
3555 | } | |
3556 | ||
3557 | if (!page) { | |
235b6217 | 3558 | unlock_cluster(ci); |
ec8acf20 | 3559 | spin_unlock(&si->lock); |
570a335b HD |
3560 | return -ENOMEM; |
3561 | } | |
3562 | ||
3563 | /* | |
3564 | * We are fortunate that although vmalloc_to_page uses pte_offset_map, | |
2de1a7e4 SJ |
3565 | * no architecture is using highmem pages for kernel page tables: so it |
3566 | * will not corrupt the GFP_ATOMIC caller's atomic page table kmaps. | |
570a335b HD |
3567 | */ |
3568 | head = vmalloc_to_page(si->swap_map + offset); | |
3569 | offset &= ~PAGE_MASK; | |
3570 | ||
2628bd6f | 3571 | spin_lock(&si->cont_lock); |
570a335b HD |
3572 | /* |
3573 | * Page allocation does not initialize the page's lru field, | |
3574 | * but it does always reset its private field. | |
3575 | */ | |
3576 | if (!page_private(head)) { | |
3577 | BUG_ON(count & COUNT_CONTINUED); | |
3578 | INIT_LIST_HEAD(&head->lru); | |
3579 | set_page_private(head, SWP_CONTINUED); | |
3580 | si->flags |= SWP_CONTINUED; | |
3581 | } | |
3582 | ||
3583 | list_for_each_entry(list_page, &head->lru, lru) { | |
3584 | unsigned char *map; | |
3585 | ||
3586 | /* | |
3587 | * If the previous map said no continuation, but we've found | |
3588 | * a continuation page, free our allocation and use this one. | |
3589 | */ | |
3590 | if (!(count & COUNT_CONTINUED)) | |
2628bd6f | 3591 | goto out_unlock_cont; |
570a335b | 3592 | |
9b04c5fe | 3593 | map = kmap_atomic(list_page) + offset; |
570a335b | 3594 | count = *map; |
9b04c5fe | 3595 | kunmap_atomic(map); |
570a335b HD |
3596 | |
3597 | /* | |
3598 | * If this continuation count now has some space in it, | |
3599 | * free our allocation and use this one. | |
3600 | */ | |
3601 | if ((count & ~COUNT_CONTINUED) != SWAP_CONT_MAX) | |
2628bd6f | 3602 | goto out_unlock_cont; |
570a335b HD |
3603 | } |
3604 | ||
3605 | list_add_tail(&page->lru, &head->lru); | |
3606 | page = NULL; /* now it's attached, don't free it */ | |
2628bd6f HY |
3607 | out_unlock_cont: |
3608 | spin_unlock(&si->cont_lock); | |
570a335b | 3609 | out: |
235b6217 | 3610 | unlock_cluster(ci); |
ec8acf20 | 3611 | spin_unlock(&si->lock); |
570a335b HD |
3612 | outer: |
3613 | if (page) | |
3614 | __free_page(page); | |
3615 | return 0; | |
3616 | } | |
3617 | ||
3618 | /* | |
3619 | * swap_count_continued - when the original swap_map count is incremented | |
3620 | * from SWAP_MAP_MAX, check if there is already a continuation page to carry | |
3621 | * into, carry if so, or else fail until a new continuation page is allocated; | |
3622 | * when the original swap_map count is decremented from 0 with continuation, | |
3623 | * borrow from the continuation and report whether it still holds more. | |
235b6217 HY |
3624 | * Called while __swap_duplicate() or swap_entry_free() holds swap or cluster |
3625 | * lock. | |
570a335b HD |
3626 | */ |
3627 | static bool swap_count_continued(struct swap_info_struct *si, | |
3628 | pgoff_t offset, unsigned char count) | |
3629 | { | |
3630 | struct page *head; | |
3631 | struct page *page; | |
3632 | unsigned char *map; | |
2628bd6f | 3633 | bool ret; |
570a335b HD |
3634 | |
3635 | head = vmalloc_to_page(si->swap_map + offset); | |
3636 | if (page_private(head) != SWP_CONTINUED) { | |
3637 | BUG_ON(count & COUNT_CONTINUED); | |
3638 | return false; /* need to add count continuation */ | |
3639 | } | |
3640 | ||
2628bd6f | 3641 | spin_lock(&si->cont_lock); |
570a335b HD |
3642 | offset &= ~PAGE_MASK; |
3643 | page = list_entry(head->lru.next, struct page, lru); | |
9b04c5fe | 3644 | map = kmap_atomic(page) + offset; |
570a335b HD |
3645 | |
3646 | if (count == SWAP_MAP_MAX) /* initial increment from swap_map */ | |
3647 | goto init_map; /* jump over SWAP_CONT_MAX checks */ | |
3648 | ||
3649 | if (count == (SWAP_MAP_MAX | COUNT_CONTINUED)) { /* incrementing */ | |
3650 | /* | |
3651 | * Think of how you add 1 to 999 | |
3652 | */ | |
3653 | while (*map == (SWAP_CONT_MAX | COUNT_CONTINUED)) { | |
9b04c5fe | 3654 | kunmap_atomic(map); |
570a335b HD |
3655 | page = list_entry(page->lru.next, struct page, lru); |
3656 | BUG_ON(page == head); | |
9b04c5fe | 3657 | map = kmap_atomic(page) + offset; |
570a335b HD |
3658 | } |
3659 | if (*map == SWAP_CONT_MAX) { | |
9b04c5fe | 3660 | kunmap_atomic(map); |
570a335b | 3661 | page = list_entry(page->lru.next, struct page, lru); |
2628bd6f HY |
3662 | if (page == head) { |
3663 | ret = false; /* add count continuation */ | |
3664 | goto out; | |
3665 | } | |
9b04c5fe | 3666 | map = kmap_atomic(page) + offset; |
570a335b HD |
3667 | init_map: *map = 0; /* we didn't zero the page */ |
3668 | } | |
3669 | *map += 1; | |
9b04c5fe | 3670 | kunmap_atomic(map); |
570a335b HD |
3671 | page = list_entry(page->lru.prev, struct page, lru); |
3672 | while (page != head) { | |
9b04c5fe | 3673 | map = kmap_atomic(page) + offset; |
570a335b | 3674 | *map = COUNT_CONTINUED; |
9b04c5fe | 3675 | kunmap_atomic(map); |
570a335b HD |
3676 | page = list_entry(page->lru.prev, struct page, lru); |
3677 | } | |
2628bd6f | 3678 | ret = true; /* incremented */ |
570a335b HD |
3679 | |
3680 | } else { /* decrementing */ | |
3681 | /* | |
3682 | * Think of how you subtract 1 from 1000 | |
3683 | */ | |
3684 | BUG_ON(count != COUNT_CONTINUED); | |
3685 | while (*map == COUNT_CONTINUED) { | |
9b04c5fe | 3686 | kunmap_atomic(map); |
570a335b HD |
3687 | page = list_entry(page->lru.next, struct page, lru); |
3688 | BUG_ON(page == head); | |
9b04c5fe | 3689 | map = kmap_atomic(page) + offset; |
570a335b HD |
3690 | } |
3691 | BUG_ON(*map == 0); | |
3692 | *map -= 1; | |
3693 | if (*map == 0) | |
3694 | count = 0; | |
9b04c5fe | 3695 | kunmap_atomic(map); |
570a335b HD |
3696 | page = list_entry(page->lru.prev, struct page, lru); |
3697 | while (page != head) { | |
9b04c5fe | 3698 | map = kmap_atomic(page) + offset; |
570a335b HD |
3699 | *map = SWAP_CONT_MAX | count; |
3700 | count = COUNT_CONTINUED; | |
9b04c5fe | 3701 | kunmap_atomic(map); |
570a335b HD |
3702 | page = list_entry(page->lru.prev, struct page, lru); |
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); |