]>
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; |
78ecba08 | 63 | static int least_priority; |
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 | */ | |
88 | static PLIST_HEAD(swap_avail_head); | |
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 | ||
8d69aaee | 99 | static inline unsigned char swap_count(unsigned char ent) |
355cfa73 | 100 | { |
570a335b | 101 | return ent & ~SWAP_HAS_CACHE; /* may include SWAP_HAS_CONT flag */ |
355cfa73 KH |
102 | } |
103 | ||
efa90a98 | 104 | /* returns 1 if swap entry is freed */ |
c9e44410 KH |
105 | static int |
106 | __try_to_reclaim_swap(struct swap_info_struct *si, unsigned long offset) | |
107 | { | |
efa90a98 | 108 | swp_entry_t entry = swp_entry(si->type, offset); |
c9e44410 KH |
109 | struct page *page; |
110 | int ret = 0; | |
111 | ||
f6ab1f7f | 112 | page = find_get_page(swap_address_space(entry), swp_offset(entry)); |
c9e44410 KH |
113 | if (!page) |
114 | return 0; | |
115 | /* | |
116 | * This function is called from scan_swap_map() and it's called | |
117 | * by vmscan.c at reclaiming pages. So, we hold a lock on a page, here. | |
118 | * We have to use trylock for avoiding deadlock. This is a special | |
119 | * case and you should use try_to_free_swap() with explicit lock_page() | |
120 | * in usual operations. | |
121 | */ | |
122 | if (trylock_page(page)) { | |
123 | ret = try_to_free_swap(page); | |
124 | unlock_page(page); | |
125 | } | |
09cbfeaf | 126 | put_page(page); |
c9e44410 KH |
127 | return ret; |
128 | } | |
355cfa73 | 129 | |
6a6ba831 HD |
130 | /* |
131 | * swapon tell device that all the old swap contents can be discarded, | |
132 | * to allow the swap device to optimize its wear-levelling. | |
133 | */ | |
134 | static int discard_swap(struct swap_info_struct *si) | |
135 | { | |
136 | struct swap_extent *se; | |
9625a5f2 HD |
137 | sector_t start_block; |
138 | sector_t nr_blocks; | |
6a6ba831 HD |
139 | int err = 0; |
140 | ||
9625a5f2 HD |
141 | /* Do not discard the swap header page! */ |
142 | se = &si->first_swap_extent; | |
143 | start_block = (se->start_block + 1) << (PAGE_SHIFT - 9); | |
144 | nr_blocks = ((sector_t)se->nr_pages - 1) << (PAGE_SHIFT - 9); | |
145 | if (nr_blocks) { | |
146 | err = blkdev_issue_discard(si->bdev, start_block, | |
dd3932ed | 147 | nr_blocks, GFP_KERNEL, 0); |
9625a5f2 HD |
148 | if (err) |
149 | return err; | |
150 | cond_resched(); | |
151 | } | |
6a6ba831 | 152 | |
9625a5f2 HD |
153 | list_for_each_entry(se, &si->first_swap_extent.list, list) { |
154 | start_block = se->start_block << (PAGE_SHIFT - 9); | |
155 | nr_blocks = (sector_t)se->nr_pages << (PAGE_SHIFT - 9); | |
6a6ba831 HD |
156 | |
157 | err = blkdev_issue_discard(si->bdev, start_block, | |
dd3932ed | 158 | nr_blocks, GFP_KERNEL, 0); |
6a6ba831 HD |
159 | if (err) |
160 | break; | |
161 | ||
162 | cond_resched(); | |
163 | } | |
164 | return err; /* That will often be -EOPNOTSUPP */ | |
165 | } | |
166 | ||
7992fde7 HD |
167 | /* |
168 | * swap allocation tell device that a cluster of swap can now be discarded, | |
169 | * to allow the swap device to optimize its wear-levelling. | |
170 | */ | |
171 | static void discard_swap_cluster(struct swap_info_struct *si, | |
172 | pgoff_t start_page, pgoff_t nr_pages) | |
173 | { | |
174 | struct swap_extent *se = si->curr_swap_extent; | |
175 | int found_extent = 0; | |
176 | ||
177 | while (nr_pages) { | |
7992fde7 HD |
178 | if (se->start_page <= start_page && |
179 | start_page < se->start_page + se->nr_pages) { | |
180 | pgoff_t offset = start_page - se->start_page; | |
181 | sector_t start_block = se->start_block + offset; | |
858a2990 | 182 | sector_t nr_blocks = se->nr_pages - offset; |
7992fde7 HD |
183 | |
184 | if (nr_blocks > nr_pages) | |
185 | nr_blocks = nr_pages; | |
186 | start_page += nr_blocks; | |
187 | nr_pages -= nr_blocks; | |
188 | ||
189 | if (!found_extent++) | |
190 | si->curr_swap_extent = se; | |
191 | ||
192 | start_block <<= PAGE_SHIFT - 9; | |
193 | nr_blocks <<= PAGE_SHIFT - 9; | |
194 | if (blkdev_issue_discard(si->bdev, start_block, | |
dd3932ed | 195 | nr_blocks, GFP_NOIO, 0)) |
7992fde7 HD |
196 | break; |
197 | } | |
198 | ||
a8ae4991 | 199 | se = list_next_entry(se, list); |
7992fde7 HD |
200 | } |
201 | } | |
202 | ||
38d8b4e6 HY |
203 | #ifdef CONFIG_THP_SWAP |
204 | #define SWAPFILE_CLUSTER HPAGE_PMD_NR | |
205 | #else | |
048c27fd | 206 | #define SWAPFILE_CLUSTER 256 |
38d8b4e6 | 207 | #endif |
048c27fd HD |
208 | #define LATENCY_LIMIT 256 |
209 | ||
2a8f9449 SL |
210 | static inline void cluster_set_flag(struct swap_cluster_info *info, |
211 | unsigned int flag) | |
212 | { | |
213 | info->flags = flag; | |
214 | } | |
215 | ||
216 | static inline unsigned int cluster_count(struct swap_cluster_info *info) | |
217 | { | |
218 | return info->data; | |
219 | } | |
220 | ||
221 | static inline void cluster_set_count(struct swap_cluster_info *info, | |
222 | unsigned int c) | |
223 | { | |
224 | info->data = c; | |
225 | } | |
226 | ||
227 | static inline void cluster_set_count_flag(struct swap_cluster_info *info, | |
228 | unsigned int c, unsigned int f) | |
229 | { | |
230 | info->flags = f; | |
231 | info->data = c; | |
232 | } | |
233 | ||
234 | static inline unsigned int cluster_next(struct swap_cluster_info *info) | |
235 | { | |
236 | return info->data; | |
237 | } | |
238 | ||
239 | static inline void cluster_set_next(struct swap_cluster_info *info, | |
240 | unsigned int n) | |
241 | { | |
242 | info->data = n; | |
243 | } | |
244 | ||
245 | static inline void cluster_set_next_flag(struct swap_cluster_info *info, | |
246 | unsigned int n, unsigned int f) | |
247 | { | |
248 | info->flags = f; | |
249 | info->data = n; | |
250 | } | |
251 | ||
252 | static inline bool cluster_is_free(struct swap_cluster_info *info) | |
253 | { | |
254 | return info->flags & CLUSTER_FLAG_FREE; | |
255 | } | |
256 | ||
257 | static inline bool cluster_is_null(struct swap_cluster_info *info) | |
258 | { | |
259 | return info->flags & CLUSTER_FLAG_NEXT_NULL; | |
260 | } | |
261 | ||
262 | static inline void cluster_set_null(struct swap_cluster_info *info) | |
263 | { | |
264 | info->flags = CLUSTER_FLAG_NEXT_NULL; | |
265 | info->data = 0; | |
266 | } | |
267 | ||
e0709829 HY |
268 | static inline bool cluster_is_huge(struct swap_cluster_info *info) |
269 | { | |
270 | return info->flags & CLUSTER_FLAG_HUGE; | |
271 | } | |
272 | ||
273 | static inline void cluster_clear_huge(struct swap_cluster_info *info) | |
274 | { | |
275 | info->flags &= ~CLUSTER_FLAG_HUGE; | |
276 | } | |
277 | ||
235b6217 HY |
278 | static inline struct swap_cluster_info *lock_cluster(struct swap_info_struct *si, |
279 | unsigned long offset) | |
280 | { | |
281 | struct swap_cluster_info *ci; | |
282 | ||
283 | ci = si->cluster_info; | |
284 | if (ci) { | |
285 | ci += offset / SWAPFILE_CLUSTER; | |
286 | spin_lock(&ci->lock); | |
287 | } | |
288 | return ci; | |
289 | } | |
290 | ||
291 | static inline void unlock_cluster(struct swap_cluster_info *ci) | |
292 | { | |
293 | if (ci) | |
294 | spin_unlock(&ci->lock); | |
295 | } | |
296 | ||
297 | static inline struct swap_cluster_info *lock_cluster_or_swap_info( | |
298 | struct swap_info_struct *si, | |
299 | unsigned long offset) | |
300 | { | |
301 | struct swap_cluster_info *ci; | |
302 | ||
303 | ci = lock_cluster(si, offset); | |
304 | if (!ci) | |
305 | spin_lock(&si->lock); | |
306 | ||
307 | return ci; | |
308 | } | |
309 | ||
310 | static inline void unlock_cluster_or_swap_info(struct swap_info_struct *si, | |
311 | struct swap_cluster_info *ci) | |
312 | { | |
313 | if (ci) | |
314 | unlock_cluster(ci); | |
315 | else | |
316 | spin_unlock(&si->lock); | |
317 | } | |
318 | ||
6b534915 HY |
319 | static inline bool cluster_list_empty(struct swap_cluster_list *list) |
320 | { | |
321 | return cluster_is_null(&list->head); | |
322 | } | |
323 | ||
324 | static inline unsigned int cluster_list_first(struct swap_cluster_list *list) | |
325 | { | |
326 | return cluster_next(&list->head); | |
327 | } | |
328 | ||
329 | static void cluster_list_init(struct swap_cluster_list *list) | |
330 | { | |
331 | cluster_set_null(&list->head); | |
332 | cluster_set_null(&list->tail); | |
333 | } | |
334 | ||
335 | static void cluster_list_add_tail(struct swap_cluster_list *list, | |
336 | struct swap_cluster_info *ci, | |
337 | unsigned int idx) | |
338 | { | |
339 | if (cluster_list_empty(list)) { | |
340 | cluster_set_next_flag(&list->head, idx, 0); | |
341 | cluster_set_next_flag(&list->tail, idx, 0); | |
342 | } else { | |
235b6217 | 343 | struct swap_cluster_info *ci_tail; |
6b534915 HY |
344 | unsigned int tail = cluster_next(&list->tail); |
345 | ||
235b6217 HY |
346 | /* |
347 | * Nested cluster lock, but both cluster locks are | |
348 | * only acquired when we held swap_info_struct->lock | |
349 | */ | |
350 | ci_tail = ci + tail; | |
351 | spin_lock_nested(&ci_tail->lock, SINGLE_DEPTH_NESTING); | |
352 | cluster_set_next(ci_tail, idx); | |
0ef017d1 | 353 | spin_unlock(&ci_tail->lock); |
6b534915 HY |
354 | cluster_set_next_flag(&list->tail, idx, 0); |
355 | } | |
356 | } | |
357 | ||
358 | static unsigned int cluster_list_del_first(struct swap_cluster_list *list, | |
359 | struct swap_cluster_info *ci) | |
360 | { | |
361 | unsigned int idx; | |
362 | ||
363 | idx = cluster_next(&list->head); | |
364 | if (cluster_next(&list->tail) == idx) { | |
365 | cluster_set_null(&list->head); | |
366 | cluster_set_null(&list->tail); | |
367 | } else | |
368 | cluster_set_next_flag(&list->head, | |
369 | cluster_next(&ci[idx]), 0); | |
370 | ||
371 | return idx; | |
372 | } | |
373 | ||
815c2c54 SL |
374 | /* Add a cluster to discard list and schedule it to do discard */ |
375 | static void swap_cluster_schedule_discard(struct swap_info_struct *si, | |
376 | unsigned int idx) | |
377 | { | |
378 | /* | |
379 | * If scan_swap_map() can't find a free cluster, it will check | |
380 | * si->swap_map directly. To make sure the discarding cluster isn't | |
381 | * taken by scan_swap_map(), mark the swap entries bad (occupied). It | |
382 | * will be cleared after discard | |
383 | */ | |
384 | memset(si->swap_map + idx * SWAPFILE_CLUSTER, | |
385 | SWAP_MAP_BAD, SWAPFILE_CLUSTER); | |
386 | ||
6b534915 | 387 | cluster_list_add_tail(&si->discard_clusters, si->cluster_info, idx); |
815c2c54 SL |
388 | |
389 | schedule_work(&si->discard_work); | |
390 | } | |
391 | ||
38d8b4e6 HY |
392 | static void __free_cluster(struct swap_info_struct *si, unsigned long idx) |
393 | { | |
394 | struct swap_cluster_info *ci = si->cluster_info; | |
395 | ||
396 | cluster_set_flag(ci + idx, CLUSTER_FLAG_FREE); | |
397 | cluster_list_add_tail(&si->free_clusters, ci, idx); | |
398 | } | |
399 | ||
815c2c54 SL |
400 | /* |
401 | * Doing discard actually. After a cluster discard is finished, the cluster | |
402 | * will be added to free cluster list. caller should hold si->lock. | |
403 | */ | |
404 | static void swap_do_scheduled_discard(struct swap_info_struct *si) | |
405 | { | |
235b6217 | 406 | struct swap_cluster_info *info, *ci; |
815c2c54 SL |
407 | unsigned int idx; |
408 | ||
409 | info = si->cluster_info; | |
410 | ||
6b534915 HY |
411 | while (!cluster_list_empty(&si->discard_clusters)) { |
412 | idx = cluster_list_del_first(&si->discard_clusters, info); | |
815c2c54 SL |
413 | spin_unlock(&si->lock); |
414 | ||
415 | discard_swap_cluster(si, idx * SWAPFILE_CLUSTER, | |
416 | SWAPFILE_CLUSTER); | |
417 | ||
418 | spin_lock(&si->lock); | |
235b6217 | 419 | ci = lock_cluster(si, idx * SWAPFILE_CLUSTER); |
38d8b4e6 | 420 | __free_cluster(si, idx); |
815c2c54 SL |
421 | memset(si->swap_map + idx * SWAPFILE_CLUSTER, |
422 | 0, SWAPFILE_CLUSTER); | |
235b6217 | 423 | unlock_cluster(ci); |
815c2c54 SL |
424 | } |
425 | } | |
426 | ||
427 | static void swap_discard_work(struct work_struct *work) | |
428 | { | |
429 | struct swap_info_struct *si; | |
430 | ||
431 | si = container_of(work, struct swap_info_struct, discard_work); | |
432 | ||
433 | spin_lock(&si->lock); | |
434 | swap_do_scheduled_discard(si); | |
435 | spin_unlock(&si->lock); | |
436 | } | |
437 | ||
38d8b4e6 HY |
438 | static void alloc_cluster(struct swap_info_struct *si, unsigned long idx) |
439 | { | |
440 | struct swap_cluster_info *ci = si->cluster_info; | |
441 | ||
442 | VM_BUG_ON(cluster_list_first(&si->free_clusters) != idx); | |
443 | cluster_list_del_first(&si->free_clusters, ci); | |
444 | cluster_set_count_flag(ci + idx, 0, 0); | |
445 | } | |
446 | ||
447 | static void free_cluster(struct swap_info_struct *si, unsigned long idx) | |
448 | { | |
449 | struct swap_cluster_info *ci = si->cluster_info + idx; | |
450 | ||
451 | VM_BUG_ON(cluster_count(ci) != 0); | |
452 | /* | |
453 | * If the swap is discardable, prepare discard the cluster | |
454 | * instead of free it immediately. The cluster will be freed | |
455 | * after discard. | |
456 | */ | |
457 | if ((si->flags & (SWP_WRITEOK | SWP_PAGE_DISCARD)) == | |
458 | (SWP_WRITEOK | SWP_PAGE_DISCARD)) { | |
459 | swap_cluster_schedule_discard(si, idx); | |
460 | return; | |
461 | } | |
462 | ||
463 | __free_cluster(si, idx); | |
464 | } | |
465 | ||
2a8f9449 SL |
466 | /* |
467 | * The cluster corresponding to page_nr will be used. The cluster will be | |
468 | * removed from free cluster list and its usage counter will be increased. | |
469 | */ | |
470 | static void inc_cluster_info_page(struct swap_info_struct *p, | |
471 | struct swap_cluster_info *cluster_info, unsigned long page_nr) | |
472 | { | |
473 | unsigned long idx = page_nr / SWAPFILE_CLUSTER; | |
474 | ||
475 | if (!cluster_info) | |
476 | return; | |
38d8b4e6 HY |
477 | if (cluster_is_free(&cluster_info[idx])) |
478 | alloc_cluster(p, idx); | |
2a8f9449 SL |
479 | |
480 | VM_BUG_ON(cluster_count(&cluster_info[idx]) >= SWAPFILE_CLUSTER); | |
481 | cluster_set_count(&cluster_info[idx], | |
482 | cluster_count(&cluster_info[idx]) + 1); | |
483 | } | |
484 | ||
485 | /* | |
486 | * The cluster corresponding to page_nr decreases one usage. If the usage | |
487 | * counter becomes 0, which means no page in the cluster is in using, we can | |
488 | * optionally discard the cluster and add it to free cluster list. | |
489 | */ | |
490 | static void dec_cluster_info_page(struct swap_info_struct *p, | |
491 | struct swap_cluster_info *cluster_info, unsigned long page_nr) | |
492 | { | |
493 | unsigned long idx = page_nr / SWAPFILE_CLUSTER; | |
494 | ||
495 | if (!cluster_info) | |
496 | return; | |
497 | ||
498 | VM_BUG_ON(cluster_count(&cluster_info[idx]) == 0); | |
499 | cluster_set_count(&cluster_info[idx], | |
500 | cluster_count(&cluster_info[idx]) - 1); | |
501 | ||
38d8b4e6 HY |
502 | if (cluster_count(&cluster_info[idx]) == 0) |
503 | free_cluster(p, idx); | |
2a8f9449 SL |
504 | } |
505 | ||
506 | /* | |
507 | * It's possible scan_swap_map() uses a free cluster in the middle of free | |
508 | * cluster list. Avoiding such abuse to avoid list corruption. | |
509 | */ | |
ebc2a1a6 SL |
510 | static bool |
511 | scan_swap_map_ssd_cluster_conflict(struct swap_info_struct *si, | |
2a8f9449 SL |
512 | unsigned long offset) |
513 | { | |
ebc2a1a6 SL |
514 | struct percpu_cluster *percpu_cluster; |
515 | bool conflict; | |
516 | ||
2a8f9449 | 517 | offset /= SWAPFILE_CLUSTER; |
6b534915 HY |
518 | conflict = !cluster_list_empty(&si->free_clusters) && |
519 | offset != cluster_list_first(&si->free_clusters) && | |
2a8f9449 | 520 | cluster_is_free(&si->cluster_info[offset]); |
ebc2a1a6 SL |
521 | |
522 | if (!conflict) | |
523 | return false; | |
524 | ||
525 | percpu_cluster = this_cpu_ptr(si->percpu_cluster); | |
526 | cluster_set_null(&percpu_cluster->index); | |
527 | return true; | |
528 | } | |
529 | ||
530 | /* | |
531 | * Try to get a swap entry from current cpu's swap entry pool (a cluster). This | |
532 | * might involve allocating a new cluster for current CPU too. | |
533 | */ | |
36005bae | 534 | static bool scan_swap_map_try_ssd_cluster(struct swap_info_struct *si, |
ebc2a1a6 SL |
535 | unsigned long *offset, unsigned long *scan_base) |
536 | { | |
537 | struct percpu_cluster *cluster; | |
235b6217 | 538 | struct swap_cluster_info *ci; |
ebc2a1a6 | 539 | bool found_free; |
235b6217 | 540 | unsigned long tmp, max; |
ebc2a1a6 SL |
541 | |
542 | new_cluster: | |
543 | cluster = this_cpu_ptr(si->percpu_cluster); | |
544 | if (cluster_is_null(&cluster->index)) { | |
6b534915 HY |
545 | if (!cluster_list_empty(&si->free_clusters)) { |
546 | cluster->index = si->free_clusters.head; | |
ebc2a1a6 SL |
547 | cluster->next = cluster_next(&cluster->index) * |
548 | SWAPFILE_CLUSTER; | |
6b534915 | 549 | } else if (!cluster_list_empty(&si->discard_clusters)) { |
ebc2a1a6 SL |
550 | /* |
551 | * we don't have free cluster but have some clusters in | |
552 | * discarding, do discard now and reclaim them | |
553 | */ | |
554 | swap_do_scheduled_discard(si); | |
555 | *scan_base = *offset = si->cluster_next; | |
556 | goto new_cluster; | |
557 | } else | |
36005bae | 558 | return false; |
ebc2a1a6 SL |
559 | } |
560 | ||
561 | found_free = false; | |
562 | ||
563 | /* | |
564 | * Other CPUs can use our cluster if they can't find a free cluster, | |
565 | * check if there is still free entry in the cluster | |
566 | */ | |
567 | tmp = cluster->next; | |
235b6217 HY |
568 | max = min_t(unsigned long, si->max, |
569 | (cluster_next(&cluster->index) + 1) * SWAPFILE_CLUSTER); | |
570 | if (tmp >= max) { | |
571 | cluster_set_null(&cluster->index); | |
572 | goto new_cluster; | |
573 | } | |
574 | ci = lock_cluster(si, tmp); | |
575 | while (tmp < max) { | |
ebc2a1a6 SL |
576 | if (!si->swap_map[tmp]) { |
577 | found_free = true; | |
578 | break; | |
579 | } | |
580 | tmp++; | |
581 | } | |
235b6217 | 582 | unlock_cluster(ci); |
ebc2a1a6 SL |
583 | if (!found_free) { |
584 | cluster_set_null(&cluster->index); | |
585 | goto new_cluster; | |
586 | } | |
587 | cluster->next = tmp + 1; | |
588 | *offset = tmp; | |
589 | *scan_base = tmp; | |
36005bae | 590 | return found_free; |
2a8f9449 SL |
591 | } |
592 | ||
38d8b4e6 HY |
593 | static void swap_range_alloc(struct swap_info_struct *si, unsigned long offset, |
594 | unsigned int nr_entries) | |
595 | { | |
596 | unsigned int end = offset + nr_entries - 1; | |
597 | ||
598 | if (offset == si->lowest_bit) | |
599 | si->lowest_bit += nr_entries; | |
600 | if (end == si->highest_bit) | |
601 | si->highest_bit -= nr_entries; | |
602 | si->inuse_pages += nr_entries; | |
603 | if (si->inuse_pages == si->pages) { | |
604 | si->lowest_bit = si->max; | |
605 | si->highest_bit = 0; | |
606 | spin_lock(&swap_avail_lock); | |
607 | plist_del(&si->avail_list, &swap_avail_head); | |
608 | spin_unlock(&swap_avail_lock); | |
609 | } | |
610 | } | |
611 | ||
612 | static void swap_range_free(struct swap_info_struct *si, unsigned long offset, | |
613 | unsigned int nr_entries) | |
614 | { | |
615 | unsigned long end = offset + nr_entries - 1; | |
616 | void (*swap_slot_free_notify)(struct block_device *, unsigned long); | |
617 | ||
618 | if (offset < si->lowest_bit) | |
619 | si->lowest_bit = offset; | |
620 | if (end > si->highest_bit) { | |
621 | bool was_full = !si->highest_bit; | |
622 | ||
623 | si->highest_bit = end; | |
624 | if (was_full && (si->flags & SWP_WRITEOK)) { | |
625 | spin_lock(&swap_avail_lock); | |
626 | WARN_ON(!plist_node_empty(&si->avail_list)); | |
627 | if (plist_node_empty(&si->avail_list)) | |
628 | plist_add(&si->avail_list, &swap_avail_head); | |
629 | spin_unlock(&swap_avail_lock); | |
630 | } | |
631 | } | |
632 | atomic_long_add(nr_entries, &nr_swap_pages); | |
633 | si->inuse_pages -= nr_entries; | |
634 | if (si->flags & SWP_BLKDEV) | |
635 | swap_slot_free_notify = | |
636 | si->bdev->bd_disk->fops->swap_slot_free_notify; | |
637 | else | |
638 | swap_slot_free_notify = NULL; | |
639 | while (offset <= end) { | |
640 | frontswap_invalidate_page(si->type, offset); | |
641 | if (swap_slot_free_notify) | |
642 | swap_slot_free_notify(si->bdev, offset); | |
643 | offset++; | |
644 | } | |
645 | } | |
646 | ||
36005bae TC |
647 | static int scan_swap_map_slots(struct swap_info_struct *si, |
648 | unsigned char usage, int nr, | |
649 | swp_entry_t slots[]) | |
1da177e4 | 650 | { |
235b6217 | 651 | struct swap_cluster_info *ci; |
ebebbbe9 | 652 | unsigned long offset; |
c60aa176 | 653 | unsigned long scan_base; |
7992fde7 | 654 | unsigned long last_in_cluster = 0; |
048c27fd | 655 | int latency_ration = LATENCY_LIMIT; |
36005bae TC |
656 | int n_ret = 0; |
657 | ||
658 | if (nr > SWAP_BATCH) | |
659 | nr = SWAP_BATCH; | |
7dfad418 | 660 | |
886bb7e9 | 661 | /* |
7dfad418 HD |
662 | * We try to cluster swap pages by allocating them sequentially |
663 | * in swap. Once we've allocated SWAPFILE_CLUSTER pages this | |
664 | * way, however, we resort to first-free allocation, starting | |
665 | * a new cluster. This prevents us from scattering swap pages | |
666 | * all over the entire swap partition, so that we reduce | |
667 | * overall disk seek times between swap pages. -- sct | |
668 | * But we do now try to find an empty cluster. -Andrea | |
c60aa176 | 669 | * And we let swap pages go all over an SSD partition. Hugh |
7dfad418 HD |
670 | */ |
671 | ||
52b7efdb | 672 | si->flags += SWP_SCANNING; |
c60aa176 | 673 | scan_base = offset = si->cluster_next; |
ebebbbe9 | 674 | |
ebc2a1a6 SL |
675 | /* SSD algorithm */ |
676 | if (si->cluster_info) { | |
36005bae TC |
677 | if (scan_swap_map_try_ssd_cluster(si, &offset, &scan_base)) |
678 | goto checks; | |
679 | else | |
680 | goto scan; | |
ebc2a1a6 SL |
681 | } |
682 | ||
ebebbbe9 HD |
683 | if (unlikely(!si->cluster_nr--)) { |
684 | if (si->pages - si->inuse_pages < SWAPFILE_CLUSTER) { | |
685 | si->cluster_nr = SWAPFILE_CLUSTER - 1; | |
686 | goto checks; | |
687 | } | |
2a8f9449 | 688 | |
ec8acf20 | 689 | spin_unlock(&si->lock); |
7dfad418 | 690 | |
c60aa176 HD |
691 | /* |
692 | * If seek is expensive, start searching for new cluster from | |
693 | * start of partition, to minimize the span of allocated swap. | |
50088c44 CY |
694 | * If seek is cheap, that is the SWP_SOLIDSTATE si->cluster_info |
695 | * case, just handled by scan_swap_map_try_ssd_cluster() above. | |
c60aa176 | 696 | */ |
50088c44 | 697 | scan_base = offset = si->lowest_bit; |
7dfad418 HD |
698 | last_in_cluster = offset + SWAPFILE_CLUSTER - 1; |
699 | ||
700 | /* Locate the first empty (unaligned) cluster */ | |
701 | for (; last_in_cluster <= si->highest_bit; offset++) { | |
1da177e4 | 702 | if (si->swap_map[offset]) |
7dfad418 HD |
703 | last_in_cluster = offset + SWAPFILE_CLUSTER; |
704 | else if (offset == last_in_cluster) { | |
ec8acf20 | 705 | spin_lock(&si->lock); |
ebebbbe9 HD |
706 | offset -= SWAPFILE_CLUSTER - 1; |
707 | si->cluster_next = offset; | |
708 | si->cluster_nr = SWAPFILE_CLUSTER - 1; | |
c60aa176 HD |
709 | goto checks; |
710 | } | |
711 | if (unlikely(--latency_ration < 0)) { | |
712 | cond_resched(); | |
713 | latency_ration = LATENCY_LIMIT; | |
714 | } | |
715 | } | |
716 | ||
717 | offset = scan_base; | |
ec8acf20 | 718 | spin_lock(&si->lock); |
ebebbbe9 | 719 | si->cluster_nr = SWAPFILE_CLUSTER - 1; |
1da177e4 | 720 | } |
7dfad418 | 721 | |
ebebbbe9 | 722 | checks: |
ebc2a1a6 | 723 | if (si->cluster_info) { |
36005bae TC |
724 | while (scan_swap_map_ssd_cluster_conflict(si, offset)) { |
725 | /* take a break if we already got some slots */ | |
726 | if (n_ret) | |
727 | goto done; | |
728 | if (!scan_swap_map_try_ssd_cluster(si, &offset, | |
729 | &scan_base)) | |
730 | goto scan; | |
731 | } | |
ebc2a1a6 | 732 | } |
ebebbbe9 | 733 | if (!(si->flags & SWP_WRITEOK)) |
52b7efdb | 734 | goto no_page; |
7dfad418 HD |
735 | if (!si->highest_bit) |
736 | goto no_page; | |
ebebbbe9 | 737 | if (offset > si->highest_bit) |
c60aa176 | 738 | scan_base = offset = si->lowest_bit; |
c9e44410 | 739 | |
235b6217 | 740 | ci = lock_cluster(si, offset); |
b73d7fce HD |
741 | /* reuse swap entry of cache-only swap if not busy. */ |
742 | if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) { | |
c9e44410 | 743 | int swap_was_freed; |
235b6217 | 744 | unlock_cluster(ci); |
ec8acf20 | 745 | spin_unlock(&si->lock); |
c9e44410 | 746 | swap_was_freed = __try_to_reclaim_swap(si, offset); |
ec8acf20 | 747 | spin_lock(&si->lock); |
c9e44410 KH |
748 | /* entry was freed successfully, try to use this again */ |
749 | if (swap_was_freed) | |
750 | goto checks; | |
751 | goto scan; /* check next one */ | |
752 | } | |
753 | ||
235b6217 HY |
754 | if (si->swap_map[offset]) { |
755 | unlock_cluster(ci); | |
36005bae TC |
756 | if (!n_ret) |
757 | goto scan; | |
758 | else | |
759 | goto done; | |
235b6217 | 760 | } |
2872bb2d HY |
761 | si->swap_map[offset] = usage; |
762 | inc_cluster_info_page(si, si->cluster_info, offset); | |
763 | unlock_cluster(ci); | |
ebebbbe9 | 764 | |
38d8b4e6 | 765 | swap_range_alloc(si, offset, 1); |
ebebbbe9 | 766 | si->cluster_next = offset + 1; |
36005bae TC |
767 | slots[n_ret++] = swp_entry(si->type, offset); |
768 | ||
769 | /* got enough slots or reach max slots? */ | |
770 | if ((n_ret == nr) || (offset >= si->highest_bit)) | |
771 | goto done; | |
772 | ||
773 | /* search for next available slot */ | |
774 | ||
775 | /* time to take a break? */ | |
776 | if (unlikely(--latency_ration < 0)) { | |
777 | if (n_ret) | |
778 | goto done; | |
779 | spin_unlock(&si->lock); | |
780 | cond_resched(); | |
781 | spin_lock(&si->lock); | |
782 | latency_ration = LATENCY_LIMIT; | |
783 | } | |
784 | ||
785 | /* try to get more slots in cluster */ | |
786 | if (si->cluster_info) { | |
787 | if (scan_swap_map_try_ssd_cluster(si, &offset, &scan_base)) | |
788 | goto checks; | |
789 | else | |
790 | goto done; | |
791 | } | |
792 | /* non-ssd case */ | |
793 | ++offset; | |
794 | ||
795 | /* non-ssd case, still more slots in cluster? */ | |
796 | if (si->cluster_nr && !si->swap_map[offset]) { | |
797 | --si->cluster_nr; | |
798 | goto checks; | |
799 | } | |
7992fde7 | 800 | |
36005bae TC |
801 | done: |
802 | si->flags -= SWP_SCANNING; | |
803 | return n_ret; | |
7dfad418 | 804 | |
ebebbbe9 | 805 | scan: |
ec8acf20 | 806 | spin_unlock(&si->lock); |
7dfad418 | 807 | while (++offset <= si->highest_bit) { |
52b7efdb | 808 | if (!si->swap_map[offset]) { |
ec8acf20 | 809 | spin_lock(&si->lock); |
52b7efdb HD |
810 | goto checks; |
811 | } | |
c9e44410 | 812 | if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) { |
ec8acf20 | 813 | spin_lock(&si->lock); |
c9e44410 KH |
814 | goto checks; |
815 | } | |
048c27fd HD |
816 | if (unlikely(--latency_ration < 0)) { |
817 | cond_resched(); | |
818 | latency_ration = LATENCY_LIMIT; | |
819 | } | |
7dfad418 | 820 | } |
c60aa176 | 821 | offset = si->lowest_bit; |
a5998061 | 822 | while (offset < scan_base) { |
c60aa176 | 823 | if (!si->swap_map[offset]) { |
ec8acf20 | 824 | spin_lock(&si->lock); |
c60aa176 HD |
825 | goto checks; |
826 | } | |
c9e44410 | 827 | if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) { |
ec8acf20 | 828 | spin_lock(&si->lock); |
c9e44410 KH |
829 | goto checks; |
830 | } | |
c60aa176 HD |
831 | if (unlikely(--latency_ration < 0)) { |
832 | cond_resched(); | |
833 | latency_ration = LATENCY_LIMIT; | |
834 | } | |
a5998061 | 835 | offset++; |
c60aa176 | 836 | } |
ec8acf20 | 837 | spin_lock(&si->lock); |
7dfad418 HD |
838 | |
839 | no_page: | |
52b7efdb | 840 | si->flags -= SWP_SCANNING; |
36005bae | 841 | return n_ret; |
1da177e4 LT |
842 | } |
843 | ||
38d8b4e6 HY |
844 | #ifdef CONFIG_THP_SWAP |
845 | static int swap_alloc_cluster(struct swap_info_struct *si, swp_entry_t *slot) | |
846 | { | |
847 | unsigned long idx; | |
848 | struct swap_cluster_info *ci; | |
849 | unsigned long offset, i; | |
850 | unsigned char *map; | |
851 | ||
852 | if (cluster_list_empty(&si->free_clusters)) | |
853 | return 0; | |
854 | ||
855 | idx = cluster_list_first(&si->free_clusters); | |
856 | offset = idx * SWAPFILE_CLUSTER; | |
857 | ci = lock_cluster(si, offset); | |
858 | alloc_cluster(si, idx); | |
e0709829 | 859 | cluster_set_count_flag(ci, SWAPFILE_CLUSTER, CLUSTER_FLAG_HUGE); |
38d8b4e6 HY |
860 | |
861 | map = si->swap_map + offset; | |
862 | for (i = 0; i < SWAPFILE_CLUSTER; i++) | |
863 | map[i] = SWAP_HAS_CACHE; | |
864 | unlock_cluster(ci); | |
865 | swap_range_alloc(si, offset, SWAPFILE_CLUSTER); | |
866 | *slot = swp_entry(si->type, offset); | |
867 | ||
868 | return 1; | |
869 | } | |
870 | ||
871 | static void swap_free_cluster(struct swap_info_struct *si, unsigned long idx) | |
872 | { | |
873 | unsigned long offset = idx * SWAPFILE_CLUSTER; | |
874 | struct swap_cluster_info *ci; | |
875 | ||
876 | ci = lock_cluster(si, offset); | |
877 | cluster_set_count_flag(ci, 0, 0); | |
878 | free_cluster(si, idx); | |
879 | unlock_cluster(ci); | |
880 | swap_range_free(si, offset, SWAPFILE_CLUSTER); | |
881 | } | |
882 | #else | |
883 | static int swap_alloc_cluster(struct swap_info_struct *si, swp_entry_t *slot) | |
884 | { | |
885 | VM_WARN_ON_ONCE(1); | |
886 | return 0; | |
887 | } | |
888 | #endif /* CONFIG_THP_SWAP */ | |
889 | ||
36005bae TC |
890 | static unsigned long scan_swap_map(struct swap_info_struct *si, |
891 | unsigned char usage) | |
892 | { | |
893 | swp_entry_t entry; | |
894 | int n_ret; | |
895 | ||
896 | n_ret = scan_swap_map_slots(si, usage, 1, &entry); | |
897 | ||
898 | if (n_ret) | |
899 | return swp_offset(entry); | |
900 | else | |
901 | return 0; | |
902 | ||
903 | } | |
904 | ||
38d8b4e6 | 905 | int get_swap_pages(int n_goal, bool cluster, swp_entry_t swp_entries[]) |
1da177e4 | 906 | { |
38d8b4e6 | 907 | unsigned long nr_pages = cluster ? SWAPFILE_CLUSTER : 1; |
adfab836 | 908 | struct swap_info_struct *si, *next; |
36005bae TC |
909 | long avail_pgs; |
910 | int n_ret = 0; | |
1da177e4 | 911 | |
38d8b4e6 HY |
912 | /* Only single cluster request supported */ |
913 | WARN_ON_ONCE(n_goal > 1 && cluster); | |
914 | ||
915 | avail_pgs = atomic_long_read(&nr_swap_pages) / nr_pages; | |
36005bae | 916 | if (avail_pgs <= 0) |
fb4f88dc | 917 | goto noswap; |
36005bae TC |
918 | |
919 | if (n_goal > SWAP_BATCH) | |
920 | n_goal = SWAP_BATCH; | |
921 | ||
922 | if (n_goal > avail_pgs) | |
923 | n_goal = avail_pgs; | |
924 | ||
38d8b4e6 | 925 | atomic_long_sub(n_goal * nr_pages, &nr_swap_pages); |
fb4f88dc | 926 | |
18ab4d4c DS |
927 | spin_lock(&swap_avail_lock); |
928 | ||
929 | start_over: | |
930 | plist_for_each_entry_safe(si, next, &swap_avail_head, avail_list) { | |
931 | /* requeue si to after same-priority siblings */ | |
932 | plist_requeue(&si->avail_list, &swap_avail_head); | |
933 | spin_unlock(&swap_avail_lock); | |
ec8acf20 | 934 | spin_lock(&si->lock); |
adfab836 | 935 | if (!si->highest_bit || !(si->flags & SWP_WRITEOK)) { |
18ab4d4c DS |
936 | spin_lock(&swap_avail_lock); |
937 | if (plist_node_empty(&si->avail_list)) { | |
938 | spin_unlock(&si->lock); | |
939 | goto nextsi; | |
940 | } | |
941 | WARN(!si->highest_bit, | |
942 | "swap_info %d in list but !highest_bit\n", | |
943 | si->type); | |
944 | WARN(!(si->flags & SWP_WRITEOK), | |
945 | "swap_info %d in list but !SWP_WRITEOK\n", | |
946 | si->type); | |
947 | plist_del(&si->avail_list, &swap_avail_head); | |
ec8acf20 | 948 | spin_unlock(&si->lock); |
18ab4d4c | 949 | goto nextsi; |
ec8acf20 | 950 | } |
38d8b4e6 HY |
951 | if (cluster) |
952 | n_ret = swap_alloc_cluster(si, swp_entries); | |
953 | else | |
954 | n_ret = scan_swap_map_slots(si, SWAP_HAS_CACHE, | |
955 | n_goal, swp_entries); | |
ec8acf20 | 956 | spin_unlock(&si->lock); |
38d8b4e6 | 957 | if (n_ret || cluster) |
36005bae | 958 | goto check_out; |
18ab4d4c | 959 | pr_debug("scan_swap_map of si %d failed to find offset\n", |
36005bae TC |
960 | si->type); |
961 | ||
18ab4d4c DS |
962 | spin_lock(&swap_avail_lock); |
963 | nextsi: | |
adfab836 DS |
964 | /* |
965 | * if we got here, it's likely that si was almost full before, | |
966 | * and since scan_swap_map() can drop the si->lock, multiple | |
967 | * callers probably all tried to get a page from the same si | |
18ab4d4c DS |
968 | * and it filled up before we could get one; or, the si filled |
969 | * up between us dropping swap_avail_lock and taking si->lock. | |
970 | * Since we dropped the swap_avail_lock, the swap_avail_head | |
971 | * list may have been modified; so if next is still in the | |
36005bae TC |
972 | * swap_avail_head list then try it, otherwise start over |
973 | * if we have not gotten any slots. | |
adfab836 | 974 | */ |
18ab4d4c DS |
975 | if (plist_node_empty(&next->avail_list)) |
976 | goto start_over; | |
1da177e4 | 977 | } |
fb4f88dc | 978 | |
18ab4d4c DS |
979 | spin_unlock(&swap_avail_lock); |
980 | ||
36005bae TC |
981 | check_out: |
982 | if (n_ret < n_goal) | |
38d8b4e6 HY |
983 | atomic_long_add((long)(n_goal - n_ret) * nr_pages, |
984 | &nr_swap_pages); | |
fb4f88dc | 985 | noswap: |
36005bae TC |
986 | return n_ret; |
987 | } | |
988 | ||
2de1a7e4 | 989 | /* The only caller of this function is now suspend routine */ |
910321ea HD |
990 | swp_entry_t get_swap_page_of_type(int type) |
991 | { | |
992 | struct swap_info_struct *si; | |
993 | pgoff_t offset; | |
994 | ||
910321ea | 995 | si = swap_info[type]; |
ec8acf20 | 996 | spin_lock(&si->lock); |
910321ea | 997 | if (si && (si->flags & SWP_WRITEOK)) { |
ec8acf20 | 998 | atomic_long_dec(&nr_swap_pages); |
910321ea HD |
999 | /* This is called for allocating swap entry, not cache */ |
1000 | offset = scan_swap_map(si, 1); | |
1001 | if (offset) { | |
ec8acf20 | 1002 | spin_unlock(&si->lock); |
910321ea HD |
1003 | return swp_entry(type, offset); |
1004 | } | |
ec8acf20 | 1005 | atomic_long_inc(&nr_swap_pages); |
910321ea | 1006 | } |
ec8acf20 | 1007 | spin_unlock(&si->lock); |
910321ea HD |
1008 | return (swp_entry_t) {0}; |
1009 | } | |
1010 | ||
e8c26ab6 | 1011 | static struct swap_info_struct *__swap_info_get(swp_entry_t entry) |
1da177e4 | 1012 | { |
73c34b6a | 1013 | struct swap_info_struct *p; |
1da177e4 LT |
1014 | unsigned long offset, type; |
1015 | ||
1016 | if (!entry.val) | |
1017 | goto out; | |
1018 | type = swp_type(entry); | |
1019 | if (type >= nr_swapfiles) | |
1020 | goto bad_nofile; | |
efa90a98 | 1021 | p = swap_info[type]; |
1da177e4 LT |
1022 | if (!(p->flags & SWP_USED)) |
1023 | goto bad_device; | |
1024 | offset = swp_offset(entry); | |
1025 | if (offset >= p->max) | |
1026 | goto bad_offset; | |
1da177e4 LT |
1027 | return p; |
1028 | ||
1da177e4 | 1029 | bad_offset: |
6a991fc7 | 1030 | pr_err("swap_info_get: %s%08lx\n", Bad_offset, entry.val); |
1da177e4 LT |
1031 | goto out; |
1032 | bad_device: | |
6a991fc7 | 1033 | pr_err("swap_info_get: %s%08lx\n", Unused_file, entry.val); |
1da177e4 LT |
1034 | goto out; |
1035 | bad_nofile: | |
6a991fc7 | 1036 | pr_err("swap_info_get: %s%08lx\n", Bad_file, entry.val); |
1da177e4 LT |
1037 | out: |
1038 | return NULL; | |
886bb7e9 | 1039 | } |
1da177e4 | 1040 | |
e8c26ab6 TC |
1041 | static struct swap_info_struct *_swap_info_get(swp_entry_t entry) |
1042 | { | |
1043 | struct swap_info_struct *p; | |
1044 | ||
1045 | p = __swap_info_get(entry); | |
1046 | if (!p) | |
1047 | goto out; | |
1048 | if (!p->swap_map[swp_offset(entry)]) | |
1049 | goto bad_free; | |
1050 | return p; | |
1051 | ||
1052 | bad_free: | |
1053 | pr_err("swap_info_get: %s%08lx\n", Unused_offset, entry.val); | |
1054 | goto out; | |
1055 | out: | |
1056 | return NULL; | |
1057 | } | |
1058 | ||
235b6217 HY |
1059 | static struct swap_info_struct *swap_info_get(swp_entry_t entry) |
1060 | { | |
1061 | struct swap_info_struct *p; | |
1062 | ||
1063 | p = _swap_info_get(entry); | |
1064 | if (p) | |
1065 | spin_lock(&p->lock); | |
1066 | return p; | |
1067 | } | |
1068 | ||
7c00bafe TC |
1069 | static struct swap_info_struct *swap_info_get_cont(swp_entry_t entry, |
1070 | struct swap_info_struct *q) | |
1071 | { | |
1072 | struct swap_info_struct *p; | |
1073 | ||
1074 | p = _swap_info_get(entry); | |
1075 | ||
1076 | if (p != q) { | |
1077 | if (q != NULL) | |
1078 | spin_unlock(&q->lock); | |
1079 | if (p != NULL) | |
1080 | spin_lock(&p->lock); | |
1081 | } | |
1082 | return p; | |
1083 | } | |
1084 | ||
1085 | static unsigned char __swap_entry_free(struct swap_info_struct *p, | |
1086 | swp_entry_t entry, unsigned char usage) | |
1da177e4 | 1087 | { |
235b6217 | 1088 | struct swap_cluster_info *ci; |
253d553b | 1089 | unsigned long offset = swp_offset(entry); |
8d69aaee HD |
1090 | unsigned char count; |
1091 | unsigned char has_cache; | |
235b6217 | 1092 | |
7c00bafe | 1093 | ci = lock_cluster_or_swap_info(p, offset); |
355cfa73 | 1094 | |
253d553b | 1095 | count = p->swap_map[offset]; |
235b6217 | 1096 | |
253d553b HD |
1097 | has_cache = count & SWAP_HAS_CACHE; |
1098 | count &= ~SWAP_HAS_CACHE; | |
355cfa73 | 1099 | |
253d553b | 1100 | if (usage == SWAP_HAS_CACHE) { |
355cfa73 | 1101 | VM_BUG_ON(!has_cache); |
253d553b | 1102 | has_cache = 0; |
aaa46865 HD |
1103 | } else if (count == SWAP_MAP_SHMEM) { |
1104 | /* | |
1105 | * Or we could insist on shmem.c using a special | |
1106 | * swap_shmem_free() and free_shmem_swap_and_cache()... | |
1107 | */ | |
1108 | count = 0; | |
570a335b HD |
1109 | } else if ((count & ~COUNT_CONTINUED) <= SWAP_MAP_MAX) { |
1110 | if (count == COUNT_CONTINUED) { | |
1111 | if (swap_count_continued(p, offset, count)) | |
1112 | count = SWAP_MAP_MAX | COUNT_CONTINUED; | |
1113 | else | |
1114 | count = SWAP_MAP_MAX; | |
1115 | } else | |
1116 | count--; | |
1117 | } | |
253d553b | 1118 | |
253d553b | 1119 | usage = count | has_cache; |
7c00bafe TC |
1120 | p->swap_map[offset] = usage ? : SWAP_HAS_CACHE; |
1121 | ||
1122 | unlock_cluster_or_swap_info(p, ci); | |
1123 | ||
1124 | return usage; | |
1125 | } | |
355cfa73 | 1126 | |
7c00bafe TC |
1127 | static void swap_entry_free(struct swap_info_struct *p, swp_entry_t entry) |
1128 | { | |
1129 | struct swap_cluster_info *ci; | |
1130 | unsigned long offset = swp_offset(entry); | |
1131 | unsigned char count; | |
1132 | ||
1133 | ci = lock_cluster(p, offset); | |
1134 | count = p->swap_map[offset]; | |
1135 | VM_BUG_ON(count != SWAP_HAS_CACHE); | |
1136 | p->swap_map[offset] = 0; | |
1137 | dec_cluster_info_page(p, p->cluster_info, offset); | |
235b6217 HY |
1138 | unlock_cluster(ci); |
1139 | ||
38d8b4e6 HY |
1140 | mem_cgroup_uncharge_swap(entry, 1); |
1141 | swap_range_free(p, offset, 1); | |
1da177e4 LT |
1142 | } |
1143 | ||
1144 | /* | |
2de1a7e4 | 1145 | * Caller has made sure that the swap device corresponding to entry |
1da177e4 LT |
1146 | * is still around or has not been recycled. |
1147 | */ | |
1148 | void swap_free(swp_entry_t entry) | |
1149 | { | |
73c34b6a | 1150 | struct swap_info_struct *p; |
1da177e4 | 1151 | |
235b6217 | 1152 | p = _swap_info_get(entry); |
7c00bafe TC |
1153 | if (p) { |
1154 | if (!__swap_entry_free(p, entry, 1)) | |
67afa38e | 1155 | free_swap_slot(entry); |
7c00bafe | 1156 | } |
1da177e4 LT |
1157 | } |
1158 | ||
cb4b86ba KH |
1159 | /* |
1160 | * Called after dropping swapcache to decrease refcnt to swap entries. | |
1161 | */ | |
75f6d6d2 | 1162 | static void swapcache_free(swp_entry_t entry) |
cb4b86ba | 1163 | { |
355cfa73 KH |
1164 | struct swap_info_struct *p; |
1165 | ||
235b6217 | 1166 | p = _swap_info_get(entry); |
7c00bafe TC |
1167 | if (p) { |
1168 | if (!__swap_entry_free(p, entry, SWAP_HAS_CACHE)) | |
67afa38e | 1169 | free_swap_slot(entry); |
7c00bafe TC |
1170 | } |
1171 | } | |
1172 | ||
38d8b4e6 | 1173 | #ifdef CONFIG_THP_SWAP |
75f6d6d2 | 1174 | static void swapcache_free_cluster(swp_entry_t entry) |
38d8b4e6 HY |
1175 | { |
1176 | unsigned long offset = swp_offset(entry); | |
1177 | unsigned long idx = offset / SWAPFILE_CLUSTER; | |
1178 | struct swap_cluster_info *ci; | |
1179 | struct swap_info_struct *si; | |
1180 | unsigned char *map; | |
a3aea839 HY |
1181 | unsigned int i, free_entries = 0; |
1182 | unsigned char val; | |
38d8b4e6 | 1183 | |
a3aea839 | 1184 | si = _swap_info_get(entry); |
38d8b4e6 HY |
1185 | if (!si) |
1186 | return; | |
1187 | ||
1188 | ci = lock_cluster(si, offset); | |
e0709829 | 1189 | VM_BUG_ON(!cluster_is_huge(ci)); |
38d8b4e6 HY |
1190 | map = si->swap_map + offset; |
1191 | for (i = 0; i < SWAPFILE_CLUSTER; i++) { | |
a3aea839 HY |
1192 | val = map[i]; |
1193 | VM_BUG_ON(!(val & SWAP_HAS_CACHE)); | |
1194 | if (val == SWAP_HAS_CACHE) | |
1195 | free_entries++; | |
1196 | } | |
1197 | if (!free_entries) { | |
1198 | for (i = 0; i < SWAPFILE_CLUSTER; i++) | |
1199 | map[i] &= ~SWAP_HAS_CACHE; | |
38d8b4e6 | 1200 | } |
e0709829 | 1201 | cluster_clear_huge(ci); |
38d8b4e6 | 1202 | unlock_cluster(ci); |
a3aea839 HY |
1203 | if (free_entries == SWAPFILE_CLUSTER) { |
1204 | spin_lock(&si->lock); | |
1205 | ci = lock_cluster(si, offset); | |
1206 | memset(map, 0, SWAPFILE_CLUSTER); | |
1207 | unlock_cluster(ci); | |
1208 | mem_cgroup_uncharge_swap(entry, SWAPFILE_CLUSTER); | |
1209 | swap_free_cluster(si, idx); | |
1210 | spin_unlock(&si->lock); | |
1211 | } else if (free_entries) { | |
1212 | for (i = 0; i < SWAPFILE_CLUSTER; i++, entry.val++) { | |
1213 | if (!__swap_entry_free(si, entry, SWAP_HAS_CACHE)) | |
1214 | free_swap_slot(entry); | |
1215 | } | |
1216 | } | |
38d8b4e6 | 1217 | } |
75f6d6d2 MK |
1218 | #else |
1219 | static inline void swapcache_free_cluster(swp_entry_t entry) | |
1220 | { | |
1221 | } | |
38d8b4e6 HY |
1222 | #endif /* CONFIG_THP_SWAP */ |
1223 | ||
75f6d6d2 MK |
1224 | void put_swap_page(struct page *page, swp_entry_t entry) |
1225 | { | |
1226 | if (!PageTransHuge(page)) | |
1227 | swapcache_free(entry); | |
1228 | else | |
1229 | swapcache_free_cluster(entry); | |
1230 | } | |
1231 | ||
155b5f88 HY |
1232 | static int swp_entry_cmp(const void *ent1, const void *ent2) |
1233 | { | |
1234 | const swp_entry_t *e1 = ent1, *e2 = ent2; | |
1235 | ||
1236 | return (int)swp_type(*e1) - (int)swp_type(*e2); | |
1237 | } | |
1238 | ||
7c00bafe TC |
1239 | void swapcache_free_entries(swp_entry_t *entries, int n) |
1240 | { | |
1241 | struct swap_info_struct *p, *prev; | |
1242 | int i; | |
1243 | ||
1244 | if (n <= 0) | |
1245 | return; | |
1246 | ||
1247 | prev = NULL; | |
1248 | p = NULL; | |
155b5f88 HY |
1249 | |
1250 | /* | |
1251 | * Sort swap entries by swap device, so each lock is only taken once. | |
1252 | * nr_swapfiles isn't absolutely correct, but the overhead of sort() is | |
1253 | * so low that it isn't necessary to optimize further. | |
1254 | */ | |
1255 | if (nr_swapfiles > 1) | |
1256 | sort(entries, n, sizeof(entries[0]), swp_entry_cmp, NULL); | |
7c00bafe TC |
1257 | for (i = 0; i < n; ++i) { |
1258 | p = swap_info_get_cont(entries[i], prev); | |
1259 | if (p) | |
1260 | swap_entry_free(p, entries[i]); | |
7c00bafe TC |
1261 | prev = p; |
1262 | } | |
235b6217 | 1263 | if (p) |
7c00bafe | 1264 | spin_unlock(&p->lock); |
cb4b86ba KH |
1265 | } |
1266 | ||
1da177e4 | 1267 | /* |
c475a8ab | 1268 | * How many references to page are currently swapped out? |
570a335b HD |
1269 | * This does not give an exact answer when swap count is continued, |
1270 | * but does include the high COUNT_CONTINUED flag to allow for that. | |
1da177e4 | 1271 | */ |
bde05d1c | 1272 | int page_swapcount(struct page *page) |
1da177e4 | 1273 | { |
c475a8ab HD |
1274 | int count = 0; |
1275 | struct swap_info_struct *p; | |
235b6217 | 1276 | struct swap_cluster_info *ci; |
1da177e4 | 1277 | swp_entry_t entry; |
235b6217 | 1278 | unsigned long offset; |
1da177e4 | 1279 | |
4c21e2f2 | 1280 | entry.val = page_private(page); |
235b6217 | 1281 | p = _swap_info_get(entry); |
1da177e4 | 1282 | if (p) { |
235b6217 HY |
1283 | offset = swp_offset(entry); |
1284 | ci = lock_cluster_or_swap_info(p, offset); | |
1285 | count = swap_count(p->swap_map[offset]); | |
1286 | unlock_cluster_or_swap_info(p, ci); | |
1da177e4 | 1287 | } |
c475a8ab | 1288 | return count; |
1da177e4 LT |
1289 | } |
1290 | ||
322b8afe HY |
1291 | static int swap_swapcount(struct swap_info_struct *si, swp_entry_t entry) |
1292 | { | |
1293 | int count = 0; | |
1294 | pgoff_t offset = swp_offset(entry); | |
1295 | struct swap_cluster_info *ci; | |
1296 | ||
1297 | ci = lock_cluster_or_swap_info(si, offset); | |
1298 | count = swap_count(si->swap_map[offset]); | |
1299 | unlock_cluster_or_swap_info(si, ci); | |
1300 | return count; | |
1301 | } | |
1302 | ||
e8c26ab6 TC |
1303 | /* |
1304 | * How many references to @entry are currently swapped out? | |
1305 | * This does not give an exact answer when swap count is continued, | |
1306 | * but does include the high COUNT_CONTINUED flag to allow for that. | |
1307 | */ | |
1308 | int __swp_swapcount(swp_entry_t entry) | |
1309 | { | |
1310 | int count = 0; | |
e8c26ab6 | 1311 | struct swap_info_struct *si; |
e8c26ab6 TC |
1312 | |
1313 | si = __swap_info_get(entry); | |
322b8afe HY |
1314 | if (si) |
1315 | count = swap_swapcount(si, entry); | |
e8c26ab6 TC |
1316 | return count; |
1317 | } | |
1318 | ||
8334b962 MK |
1319 | /* |
1320 | * How many references to @entry are currently swapped out? | |
1321 | * This considers COUNT_CONTINUED so it returns exact answer. | |
1322 | */ | |
1323 | int swp_swapcount(swp_entry_t entry) | |
1324 | { | |
1325 | int count, tmp_count, n; | |
1326 | struct swap_info_struct *p; | |
235b6217 | 1327 | struct swap_cluster_info *ci; |
8334b962 MK |
1328 | struct page *page; |
1329 | pgoff_t offset; | |
1330 | unsigned char *map; | |
1331 | ||
235b6217 | 1332 | p = _swap_info_get(entry); |
8334b962 MK |
1333 | if (!p) |
1334 | return 0; | |
1335 | ||
235b6217 HY |
1336 | offset = swp_offset(entry); |
1337 | ||
1338 | ci = lock_cluster_or_swap_info(p, offset); | |
1339 | ||
1340 | count = swap_count(p->swap_map[offset]); | |
8334b962 MK |
1341 | if (!(count & COUNT_CONTINUED)) |
1342 | goto out; | |
1343 | ||
1344 | count &= ~COUNT_CONTINUED; | |
1345 | n = SWAP_MAP_MAX + 1; | |
1346 | ||
8334b962 MK |
1347 | page = vmalloc_to_page(p->swap_map + offset); |
1348 | offset &= ~PAGE_MASK; | |
1349 | VM_BUG_ON(page_private(page) != SWP_CONTINUED); | |
1350 | ||
1351 | do { | |
a8ae4991 | 1352 | page = list_next_entry(page, lru); |
8334b962 MK |
1353 | map = kmap_atomic(page); |
1354 | tmp_count = map[offset]; | |
1355 | kunmap_atomic(map); | |
1356 | ||
1357 | count += (tmp_count & ~COUNT_CONTINUED) * n; | |
1358 | n *= (SWAP_CONT_MAX + 1); | |
1359 | } while (tmp_count & COUNT_CONTINUED); | |
1360 | out: | |
235b6217 | 1361 | unlock_cluster_or_swap_info(p, ci); |
8334b962 MK |
1362 | return count; |
1363 | } | |
1364 | ||
e0709829 HY |
1365 | #ifdef CONFIG_THP_SWAP |
1366 | static bool swap_page_trans_huge_swapped(struct swap_info_struct *si, | |
1367 | swp_entry_t entry) | |
1368 | { | |
1369 | struct swap_cluster_info *ci; | |
1370 | unsigned char *map = si->swap_map; | |
1371 | unsigned long roffset = swp_offset(entry); | |
1372 | unsigned long offset = round_down(roffset, SWAPFILE_CLUSTER); | |
1373 | int i; | |
1374 | bool ret = false; | |
1375 | ||
1376 | ci = lock_cluster_or_swap_info(si, offset); | |
1377 | if (!ci || !cluster_is_huge(ci)) { | |
1378 | if (map[roffset] != SWAP_HAS_CACHE) | |
1379 | ret = true; | |
1380 | goto unlock_out; | |
1381 | } | |
1382 | for (i = 0; i < SWAPFILE_CLUSTER; i++) { | |
1383 | if (map[offset + i] != SWAP_HAS_CACHE) { | |
1384 | ret = true; | |
1385 | break; | |
1386 | } | |
1387 | } | |
1388 | unlock_out: | |
1389 | unlock_cluster_or_swap_info(si, ci); | |
1390 | return ret; | |
1391 | } | |
1392 | ||
1393 | static bool page_swapped(struct page *page) | |
1394 | { | |
1395 | swp_entry_t entry; | |
1396 | struct swap_info_struct *si; | |
1397 | ||
1398 | if (likely(!PageTransCompound(page))) | |
1399 | return page_swapcount(page) != 0; | |
1400 | ||
1401 | page = compound_head(page); | |
1402 | entry.val = page_private(page); | |
1403 | si = _swap_info_get(entry); | |
1404 | if (si) | |
1405 | return swap_page_trans_huge_swapped(si, entry); | |
1406 | return false; | |
1407 | } | |
1408 | #else | |
1409 | #define swap_page_trans_huge_swapped(si, entry) swap_swapcount(si, entry) | |
1410 | #define page_swapped(page) (page_swapcount(page) != 0) | |
1411 | #endif | |
1412 | ||
1da177e4 | 1413 | /* |
7b1fe597 HD |
1414 | * We can write to an anon page without COW if there are no other references |
1415 | * to it. And as a side-effect, free up its swap: because the old content | |
1416 | * on disk will never be read, and seeking back there to write new content | |
1417 | * later would only waste time away from clustering. | |
6d0a07ed AA |
1418 | * |
1419 | * NOTE: total_mapcount should not be relied upon by the caller if | |
1420 | * reuse_swap_page() returns false, but it may be always overwritten | |
1421 | * (see the other implementation for CONFIG_SWAP=n). | |
1da177e4 | 1422 | */ |
6d0a07ed | 1423 | bool reuse_swap_page(struct page *page, int *total_mapcount) |
1da177e4 | 1424 | { |
c475a8ab HD |
1425 | int count; |
1426 | ||
309381fe | 1427 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
5ad64688 | 1428 | if (unlikely(PageKsm(page))) |
6d0a07ed AA |
1429 | return false; |
1430 | count = page_trans_huge_mapcount(page, total_mapcount); | |
7b1fe597 | 1431 | if (count <= 1 && PageSwapCache(page)) { |
c475a8ab | 1432 | count += page_swapcount(page); |
f0571429 MK |
1433 | if (count != 1) |
1434 | goto out; | |
1435 | if (!PageWriteback(page)) { | |
7b1fe597 HD |
1436 | delete_from_swap_cache(page); |
1437 | SetPageDirty(page); | |
f0571429 MK |
1438 | } else { |
1439 | swp_entry_t entry; | |
1440 | struct swap_info_struct *p; | |
1441 | ||
1442 | entry.val = page_private(page); | |
1443 | p = swap_info_get(entry); | |
1444 | if (p->flags & SWP_STABLE_WRITES) { | |
1445 | spin_unlock(&p->lock); | |
1446 | return false; | |
1447 | } | |
1448 | spin_unlock(&p->lock); | |
7b1fe597 HD |
1449 | } |
1450 | } | |
f0571429 | 1451 | out: |
5ad64688 | 1452 | return count <= 1; |
1da177e4 LT |
1453 | } |
1454 | ||
1455 | /* | |
a2c43eed HD |
1456 | * If swap is getting full, or if there are no more mappings of this page, |
1457 | * then try_to_free_swap is called to free its swap space. | |
1da177e4 | 1458 | */ |
a2c43eed | 1459 | int try_to_free_swap(struct page *page) |
1da177e4 | 1460 | { |
309381fe | 1461 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
1da177e4 LT |
1462 | |
1463 | if (!PageSwapCache(page)) | |
1464 | return 0; | |
1465 | if (PageWriteback(page)) | |
1466 | return 0; | |
e0709829 | 1467 | if (page_swapped(page)) |
1da177e4 LT |
1468 | return 0; |
1469 | ||
b73d7fce HD |
1470 | /* |
1471 | * Once hibernation has begun to create its image of memory, | |
1472 | * there's a danger that one of the calls to try_to_free_swap() | |
1473 | * - most probably a call from __try_to_reclaim_swap() while | |
1474 | * hibernation is allocating its own swap pages for the image, | |
1475 | * but conceivably even a call from memory reclaim - will free | |
1476 | * the swap from a page which has already been recorded in the | |
1477 | * image as a clean swapcache page, and then reuse its swap for | |
1478 | * another page of the image. On waking from hibernation, the | |
1479 | * original page might be freed under memory pressure, then | |
1480 | * later read back in from swap, now with the wrong data. | |
1481 | * | |
2de1a7e4 | 1482 | * Hibernation suspends storage while it is writing the image |
f90ac398 | 1483 | * to disk so check that here. |
b73d7fce | 1484 | */ |
f90ac398 | 1485 | if (pm_suspended_storage()) |
b73d7fce HD |
1486 | return 0; |
1487 | ||
e0709829 | 1488 | page = compound_head(page); |
a2c43eed HD |
1489 | delete_from_swap_cache(page); |
1490 | SetPageDirty(page); | |
1491 | return 1; | |
68a22394 RR |
1492 | } |
1493 | ||
1da177e4 LT |
1494 | /* |
1495 | * Free the swap entry like above, but also try to | |
1496 | * free the page cache entry if it is the last user. | |
1497 | */ | |
2509ef26 | 1498 | int free_swap_and_cache(swp_entry_t entry) |
1da177e4 | 1499 | { |
2509ef26 | 1500 | struct swap_info_struct *p; |
1da177e4 | 1501 | struct page *page = NULL; |
7c00bafe | 1502 | unsigned char count; |
1da177e4 | 1503 | |
a7420aa5 | 1504 | if (non_swap_entry(entry)) |
2509ef26 | 1505 | return 1; |
0697212a | 1506 | |
7c00bafe | 1507 | p = _swap_info_get(entry); |
1da177e4 | 1508 | if (p) { |
7c00bafe | 1509 | count = __swap_entry_free(p, entry, 1); |
e0709829 HY |
1510 | if (count == SWAP_HAS_CACHE && |
1511 | !swap_page_trans_huge_swapped(p, entry)) { | |
33806f06 | 1512 | page = find_get_page(swap_address_space(entry), |
f6ab1f7f | 1513 | swp_offset(entry)); |
8413ac9d | 1514 | if (page && !trylock_page(page)) { |
09cbfeaf | 1515 | put_page(page); |
93fac704 NP |
1516 | page = NULL; |
1517 | } | |
7c00bafe | 1518 | } else if (!count) |
67afa38e | 1519 | free_swap_slot(entry); |
1da177e4 LT |
1520 | } |
1521 | if (page) { | |
a2c43eed HD |
1522 | /* |
1523 | * Not mapped elsewhere, or swap space full? Free it! | |
1524 | * Also recheck PageSwapCache now page is locked (above). | |
1525 | */ | |
93fac704 | 1526 | if (PageSwapCache(page) && !PageWriteback(page) && |
322b8afe | 1527 | (!page_mapped(page) || mem_cgroup_swap_full(page)) && |
e0709829 HY |
1528 | !swap_page_trans_huge_swapped(p, entry)) { |
1529 | page = compound_head(page); | |
1da177e4 LT |
1530 | delete_from_swap_cache(page); |
1531 | SetPageDirty(page); | |
1532 | } | |
1533 | unlock_page(page); | |
09cbfeaf | 1534 | put_page(page); |
1da177e4 | 1535 | } |
2509ef26 | 1536 | return p != NULL; |
1da177e4 LT |
1537 | } |
1538 | ||
b0cb1a19 | 1539 | #ifdef CONFIG_HIBERNATION |
f577eb30 | 1540 | /* |
915bae9e | 1541 | * Find the swap type that corresponds to given device (if any). |
f577eb30 | 1542 | * |
915bae9e RW |
1543 | * @offset - number of the PAGE_SIZE-sized block of the device, starting |
1544 | * from 0, in which the swap header is expected to be located. | |
1545 | * | |
1546 | * This is needed for the suspend to disk (aka swsusp). | |
f577eb30 | 1547 | */ |
7bf23687 | 1548 | int swap_type_of(dev_t device, sector_t offset, struct block_device **bdev_p) |
f577eb30 | 1549 | { |
915bae9e | 1550 | struct block_device *bdev = NULL; |
efa90a98 | 1551 | int type; |
f577eb30 | 1552 | |
915bae9e RW |
1553 | if (device) |
1554 | bdev = bdget(device); | |
1555 | ||
f577eb30 | 1556 | spin_lock(&swap_lock); |
efa90a98 HD |
1557 | for (type = 0; type < nr_swapfiles; type++) { |
1558 | struct swap_info_struct *sis = swap_info[type]; | |
f577eb30 | 1559 | |
915bae9e | 1560 | if (!(sis->flags & SWP_WRITEOK)) |
f577eb30 | 1561 | continue; |
b6b5bce3 | 1562 | |
915bae9e | 1563 | if (!bdev) { |
7bf23687 | 1564 | if (bdev_p) |
dddac6a7 | 1565 | *bdev_p = bdgrab(sis->bdev); |
7bf23687 | 1566 | |
6e1819d6 | 1567 | spin_unlock(&swap_lock); |
efa90a98 | 1568 | return type; |
6e1819d6 | 1569 | } |
915bae9e | 1570 | if (bdev == sis->bdev) { |
9625a5f2 | 1571 | struct swap_extent *se = &sis->first_swap_extent; |
915bae9e | 1572 | |
915bae9e | 1573 | if (se->start_block == offset) { |
7bf23687 | 1574 | if (bdev_p) |
dddac6a7 | 1575 | *bdev_p = bdgrab(sis->bdev); |
7bf23687 | 1576 | |
915bae9e RW |
1577 | spin_unlock(&swap_lock); |
1578 | bdput(bdev); | |
efa90a98 | 1579 | return type; |
915bae9e | 1580 | } |
f577eb30 RW |
1581 | } |
1582 | } | |
1583 | spin_unlock(&swap_lock); | |
915bae9e RW |
1584 | if (bdev) |
1585 | bdput(bdev); | |
1586 | ||
f577eb30 RW |
1587 | return -ENODEV; |
1588 | } | |
1589 | ||
73c34b6a HD |
1590 | /* |
1591 | * Get the (PAGE_SIZE) block corresponding to given offset on the swapdev | |
1592 | * corresponding to given index in swap_info (swap type). | |
1593 | */ | |
1594 | sector_t swapdev_block(int type, pgoff_t offset) | |
1595 | { | |
1596 | struct block_device *bdev; | |
1597 | ||
1598 | if ((unsigned int)type >= nr_swapfiles) | |
1599 | return 0; | |
1600 | if (!(swap_info[type]->flags & SWP_WRITEOK)) | |
1601 | return 0; | |
d4906e1a | 1602 | return map_swap_entry(swp_entry(type, offset), &bdev); |
73c34b6a HD |
1603 | } |
1604 | ||
f577eb30 RW |
1605 | /* |
1606 | * Return either the total number of swap pages of given type, or the number | |
1607 | * of free pages of that type (depending on @free) | |
1608 | * | |
1609 | * This is needed for software suspend | |
1610 | */ | |
1611 | unsigned int count_swap_pages(int type, int free) | |
1612 | { | |
1613 | unsigned int n = 0; | |
1614 | ||
efa90a98 HD |
1615 | spin_lock(&swap_lock); |
1616 | if ((unsigned int)type < nr_swapfiles) { | |
1617 | struct swap_info_struct *sis = swap_info[type]; | |
1618 | ||
ec8acf20 | 1619 | spin_lock(&sis->lock); |
efa90a98 HD |
1620 | if (sis->flags & SWP_WRITEOK) { |
1621 | n = sis->pages; | |
f577eb30 | 1622 | if (free) |
efa90a98 | 1623 | n -= sis->inuse_pages; |
f577eb30 | 1624 | } |
ec8acf20 | 1625 | spin_unlock(&sis->lock); |
f577eb30 | 1626 | } |
efa90a98 | 1627 | spin_unlock(&swap_lock); |
f577eb30 RW |
1628 | return n; |
1629 | } | |
73c34b6a | 1630 | #endif /* CONFIG_HIBERNATION */ |
f577eb30 | 1631 | |
9f8bdb3f | 1632 | static inline int pte_same_as_swp(pte_t pte, pte_t swp_pte) |
179ef71c | 1633 | { |
9f8bdb3f | 1634 | return pte_same(pte_swp_clear_soft_dirty(pte), swp_pte); |
179ef71c CG |
1635 | } |
1636 | ||
1da177e4 | 1637 | /* |
72866f6f HD |
1638 | * No need to decide whether this PTE shares the swap entry with others, |
1639 | * just let do_wp_page work it out if a write is requested later - to | |
1640 | * force COW, vm_page_prot omits write permission from any private vma. | |
1da177e4 | 1641 | */ |
044d66c1 | 1642 | static int unuse_pte(struct vm_area_struct *vma, pmd_t *pmd, |
1da177e4 LT |
1643 | unsigned long addr, swp_entry_t entry, struct page *page) |
1644 | { | |
9e16b7fb | 1645 | struct page *swapcache; |
72835c86 | 1646 | struct mem_cgroup *memcg; |
044d66c1 HD |
1647 | spinlock_t *ptl; |
1648 | pte_t *pte; | |
1649 | int ret = 1; | |
1650 | ||
9e16b7fb HD |
1651 | swapcache = page; |
1652 | page = ksm_might_need_to_copy(page, vma, addr); | |
1653 | if (unlikely(!page)) | |
1654 | return -ENOMEM; | |
1655 | ||
f627c2f5 KS |
1656 | if (mem_cgroup_try_charge(page, vma->vm_mm, GFP_KERNEL, |
1657 | &memcg, false)) { | |
044d66c1 | 1658 | ret = -ENOMEM; |
85d9fc89 KH |
1659 | goto out_nolock; |
1660 | } | |
044d66c1 HD |
1661 | |
1662 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); | |
9f8bdb3f | 1663 | if (unlikely(!pte_same_as_swp(*pte, swp_entry_to_pte(entry)))) { |
f627c2f5 | 1664 | mem_cgroup_cancel_charge(page, memcg, false); |
044d66c1 HD |
1665 | ret = 0; |
1666 | goto out; | |
1667 | } | |
8a9f3ccd | 1668 | |
b084d435 | 1669 | dec_mm_counter(vma->vm_mm, MM_SWAPENTS); |
d559db08 | 1670 | inc_mm_counter(vma->vm_mm, MM_ANONPAGES); |
1da177e4 LT |
1671 | get_page(page); |
1672 | set_pte_at(vma->vm_mm, addr, pte, | |
1673 | pte_mkold(mk_pte(page, vma->vm_page_prot))); | |
00501b53 | 1674 | if (page == swapcache) { |
d281ee61 | 1675 | page_add_anon_rmap(page, vma, addr, false); |
f627c2f5 | 1676 | mem_cgroup_commit_charge(page, memcg, true, false); |
00501b53 | 1677 | } else { /* ksm created a completely new copy */ |
d281ee61 | 1678 | page_add_new_anon_rmap(page, vma, addr, false); |
f627c2f5 | 1679 | mem_cgroup_commit_charge(page, memcg, false, false); |
00501b53 JW |
1680 | lru_cache_add_active_or_unevictable(page, vma); |
1681 | } | |
1da177e4 LT |
1682 | swap_free(entry); |
1683 | /* | |
1684 | * Move the page to the active list so it is not | |
1685 | * immediately swapped out again after swapon. | |
1686 | */ | |
1687 | activate_page(page); | |
044d66c1 HD |
1688 | out: |
1689 | pte_unmap_unlock(pte, ptl); | |
85d9fc89 | 1690 | out_nolock: |
9e16b7fb HD |
1691 | if (page != swapcache) { |
1692 | unlock_page(page); | |
1693 | put_page(page); | |
1694 | } | |
044d66c1 | 1695 | return ret; |
1da177e4 LT |
1696 | } |
1697 | ||
1698 | static int unuse_pte_range(struct vm_area_struct *vma, pmd_t *pmd, | |
1699 | unsigned long addr, unsigned long end, | |
1700 | swp_entry_t entry, struct page *page) | |
1701 | { | |
1da177e4 | 1702 | pte_t swp_pte = swp_entry_to_pte(entry); |
705e87c0 | 1703 | pte_t *pte; |
8a9f3ccd | 1704 | int ret = 0; |
1da177e4 | 1705 | |
044d66c1 HD |
1706 | /* |
1707 | * We don't actually need pte lock while scanning for swp_pte: since | |
1708 | * we hold page lock and mmap_sem, swp_pte cannot be inserted into the | |
1709 | * page table while we're scanning; though it could get zapped, and on | |
1710 | * some architectures (e.g. x86_32 with PAE) we might catch a glimpse | |
1711 | * of unmatched parts which look like swp_pte, so unuse_pte must | |
1712 | * recheck under pte lock. Scanning without pte lock lets it be | |
2de1a7e4 | 1713 | * preemptable whenever CONFIG_PREEMPT but not CONFIG_HIGHPTE. |
044d66c1 HD |
1714 | */ |
1715 | pte = pte_offset_map(pmd, addr); | |
1da177e4 LT |
1716 | do { |
1717 | /* | |
1718 | * swapoff spends a _lot_ of time in this loop! | |
1719 | * Test inline before going to call unuse_pte. | |
1720 | */ | |
9f8bdb3f | 1721 | if (unlikely(pte_same_as_swp(*pte, swp_pte))) { |
044d66c1 HD |
1722 | pte_unmap(pte); |
1723 | ret = unuse_pte(vma, pmd, addr, entry, page); | |
1724 | if (ret) | |
1725 | goto out; | |
1726 | pte = pte_offset_map(pmd, addr); | |
1da177e4 LT |
1727 | } |
1728 | } while (pte++, addr += PAGE_SIZE, addr != end); | |
044d66c1 HD |
1729 | pte_unmap(pte - 1); |
1730 | out: | |
8a9f3ccd | 1731 | return ret; |
1da177e4 LT |
1732 | } |
1733 | ||
1734 | static inline int unuse_pmd_range(struct vm_area_struct *vma, pud_t *pud, | |
1735 | unsigned long addr, unsigned long end, | |
1736 | swp_entry_t entry, struct page *page) | |
1737 | { | |
1738 | pmd_t *pmd; | |
1739 | unsigned long next; | |
8a9f3ccd | 1740 | int ret; |
1da177e4 LT |
1741 | |
1742 | pmd = pmd_offset(pud, addr); | |
1743 | do { | |
dc644a07 | 1744 | cond_resched(); |
1da177e4 | 1745 | next = pmd_addr_end(addr, end); |
1a5a9906 | 1746 | if (pmd_none_or_trans_huge_or_clear_bad(pmd)) |
1da177e4 | 1747 | continue; |
8a9f3ccd BS |
1748 | ret = unuse_pte_range(vma, pmd, addr, next, entry, page); |
1749 | if (ret) | |
1750 | return ret; | |
1da177e4 LT |
1751 | } while (pmd++, addr = next, addr != end); |
1752 | return 0; | |
1753 | } | |
1754 | ||
c2febafc | 1755 | static inline int unuse_pud_range(struct vm_area_struct *vma, p4d_t *p4d, |
1da177e4 LT |
1756 | unsigned long addr, unsigned long end, |
1757 | swp_entry_t entry, struct page *page) | |
1758 | { | |
1759 | pud_t *pud; | |
1760 | unsigned long next; | |
8a9f3ccd | 1761 | int ret; |
1da177e4 | 1762 | |
c2febafc | 1763 | pud = pud_offset(p4d, addr); |
1da177e4 LT |
1764 | do { |
1765 | next = pud_addr_end(addr, end); | |
1766 | if (pud_none_or_clear_bad(pud)) | |
1767 | continue; | |
8a9f3ccd BS |
1768 | ret = unuse_pmd_range(vma, pud, addr, next, entry, page); |
1769 | if (ret) | |
1770 | return ret; | |
1da177e4 LT |
1771 | } while (pud++, addr = next, addr != end); |
1772 | return 0; | |
1773 | } | |
1774 | ||
c2febafc KS |
1775 | static inline int unuse_p4d_range(struct vm_area_struct *vma, pgd_t *pgd, |
1776 | unsigned long addr, unsigned long end, | |
1777 | swp_entry_t entry, struct page *page) | |
1778 | { | |
1779 | p4d_t *p4d; | |
1780 | unsigned long next; | |
1781 | int ret; | |
1782 | ||
1783 | p4d = p4d_offset(pgd, addr); | |
1784 | do { | |
1785 | next = p4d_addr_end(addr, end); | |
1786 | if (p4d_none_or_clear_bad(p4d)) | |
1787 | continue; | |
1788 | ret = unuse_pud_range(vma, p4d, addr, next, entry, page); | |
1789 | if (ret) | |
1790 | return ret; | |
1791 | } while (p4d++, addr = next, addr != end); | |
1792 | return 0; | |
1793 | } | |
1794 | ||
1da177e4 LT |
1795 | static int unuse_vma(struct vm_area_struct *vma, |
1796 | swp_entry_t entry, struct page *page) | |
1797 | { | |
1798 | pgd_t *pgd; | |
1799 | unsigned long addr, end, next; | |
8a9f3ccd | 1800 | int ret; |
1da177e4 | 1801 | |
3ca7b3c5 | 1802 | if (page_anon_vma(page)) { |
1da177e4 LT |
1803 | addr = page_address_in_vma(page, vma); |
1804 | if (addr == -EFAULT) | |
1805 | return 0; | |
1806 | else | |
1807 | end = addr + PAGE_SIZE; | |
1808 | } else { | |
1809 | addr = vma->vm_start; | |
1810 | end = vma->vm_end; | |
1811 | } | |
1812 | ||
1813 | pgd = pgd_offset(vma->vm_mm, addr); | |
1814 | do { | |
1815 | next = pgd_addr_end(addr, end); | |
1816 | if (pgd_none_or_clear_bad(pgd)) | |
1817 | continue; | |
c2febafc | 1818 | ret = unuse_p4d_range(vma, pgd, addr, next, entry, page); |
8a9f3ccd BS |
1819 | if (ret) |
1820 | return ret; | |
1da177e4 LT |
1821 | } while (pgd++, addr = next, addr != end); |
1822 | return 0; | |
1823 | } | |
1824 | ||
1825 | static int unuse_mm(struct mm_struct *mm, | |
1826 | swp_entry_t entry, struct page *page) | |
1827 | { | |
1828 | struct vm_area_struct *vma; | |
8a9f3ccd | 1829 | int ret = 0; |
1da177e4 LT |
1830 | |
1831 | if (!down_read_trylock(&mm->mmap_sem)) { | |
1832 | /* | |
7d03431c FLVC |
1833 | * Activate page so shrink_inactive_list is unlikely to unmap |
1834 | * its ptes while lock is dropped, so swapoff can make progress. | |
1da177e4 | 1835 | */ |
c475a8ab | 1836 | activate_page(page); |
1da177e4 LT |
1837 | unlock_page(page); |
1838 | down_read(&mm->mmap_sem); | |
1839 | lock_page(page); | |
1840 | } | |
1da177e4 | 1841 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
8a9f3ccd | 1842 | if (vma->anon_vma && (ret = unuse_vma(vma, entry, page))) |
1da177e4 | 1843 | break; |
dc644a07 | 1844 | cond_resched(); |
1da177e4 | 1845 | } |
1da177e4 | 1846 | up_read(&mm->mmap_sem); |
8a9f3ccd | 1847 | return (ret < 0)? ret: 0; |
1da177e4 LT |
1848 | } |
1849 | ||
1850 | /* | |
38b5faf4 DM |
1851 | * Scan swap_map (or frontswap_map if frontswap parameter is true) |
1852 | * from current position to next entry still in use. | |
1da177e4 LT |
1853 | * Recycle to start on reaching the end, returning 0 when empty. |
1854 | */ | |
6eb396dc | 1855 | static unsigned int find_next_to_unuse(struct swap_info_struct *si, |
38b5faf4 | 1856 | unsigned int prev, bool frontswap) |
1da177e4 | 1857 | { |
6eb396dc HD |
1858 | unsigned int max = si->max; |
1859 | unsigned int i = prev; | |
8d69aaee | 1860 | unsigned char count; |
1da177e4 LT |
1861 | |
1862 | /* | |
5d337b91 | 1863 | * No need for swap_lock here: we're just looking |
1da177e4 LT |
1864 | * for whether an entry is in use, not modifying it; false |
1865 | * hits are okay, and sys_swapoff() has already prevented new | |
5d337b91 | 1866 | * allocations from this area (while holding swap_lock). |
1da177e4 LT |
1867 | */ |
1868 | for (;;) { | |
1869 | if (++i >= max) { | |
1870 | if (!prev) { | |
1871 | i = 0; | |
1872 | break; | |
1873 | } | |
1874 | /* | |
1875 | * No entries in use at top of swap_map, | |
1876 | * loop back to start and recheck there. | |
1877 | */ | |
1878 | max = prev + 1; | |
1879 | prev = 0; | |
1880 | i = 1; | |
1881 | } | |
4db0c3c2 | 1882 | count = READ_ONCE(si->swap_map[i]); |
355cfa73 | 1883 | if (count && swap_count(count) != SWAP_MAP_BAD) |
dc644a07 HD |
1884 | if (!frontswap || frontswap_test(si, i)) |
1885 | break; | |
1886 | if ((i % LATENCY_LIMIT) == 0) | |
1887 | cond_resched(); | |
1da177e4 LT |
1888 | } |
1889 | return i; | |
1890 | } | |
1891 | ||
1892 | /* | |
1893 | * We completely avoid races by reading each swap page in advance, | |
1894 | * and then search for the process using it. All the necessary | |
1895 | * page table adjustments can then be made atomically. | |
38b5faf4 DM |
1896 | * |
1897 | * if the boolean frontswap is true, only unuse pages_to_unuse pages; | |
1898 | * pages_to_unuse==0 means all pages; ignored if frontswap is false | |
1da177e4 | 1899 | */ |
38b5faf4 DM |
1900 | int try_to_unuse(unsigned int type, bool frontswap, |
1901 | unsigned long pages_to_unuse) | |
1da177e4 | 1902 | { |
efa90a98 | 1903 | struct swap_info_struct *si = swap_info[type]; |
1da177e4 | 1904 | struct mm_struct *start_mm; |
edfe23da SL |
1905 | volatile unsigned char *swap_map; /* swap_map is accessed without |
1906 | * locking. Mark it as volatile | |
1907 | * to prevent compiler doing | |
1908 | * something odd. | |
1909 | */ | |
8d69aaee | 1910 | unsigned char swcount; |
1da177e4 LT |
1911 | struct page *page; |
1912 | swp_entry_t entry; | |
6eb396dc | 1913 | unsigned int i = 0; |
1da177e4 | 1914 | int retval = 0; |
1da177e4 LT |
1915 | |
1916 | /* | |
1917 | * When searching mms for an entry, a good strategy is to | |
1918 | * start at the first mm we freed the previous entry from | |
1919 | * (though actually we don't notice whether we or coincidence | |
1920 | * freed the entry). Initialize this start_mm with a hold. | |
1921 | * | |
1922 | * A simpler strategy would be to start at the last mm we | |
1923 | * freed the previous entry from; but that would take less | |
1924 | * advantage of mmlist ordering, which clusters forked mms | |
1925 | * together, child after parent. If we race with dup_mmap(), we | |
1926 | * prefer to resolve parent before child, lest we miss entries | |
1927 | * duplicated after we scanned child: using last mm would invert | |
570a335b | 1928 | * that. |
1da177e4 LT |
1929 | */ |
1930 | start_mm = &init_mm; | |
3fce371b | 1931 | mmget(&init_mm); |
1da177e4 LT |
1932 | |
1933 | /* | |
1934 | * Keep on scanning until all entries have gone. Usually, | |
1935 | * one pass through swap_map is enough, but not necessarily: | |
1936 | * there are races when an instance of an entry might be missed. | |
1937 | */ | |
38b5faf4 | 1938 | while ((i = find_next_to_unuse(si, i, frontswap)) != 0) { |
1da177e4 LT |
1939 | if (signal_pending(current)) { |
1940 | retval = -EINTR; | |
1941 | break; | |
1942 | } | |
1943 | ||
886bb7e9 | 1944 | /* |
1da177e4 LT |
1945 | * Get a page for the entry, using the existing swap |
1946 | * cache page if there is one. Otherwise, get a clean | |
886bb7e9 | 1947 | * page and read the swap into it. |
1da177e4 LT |
1948 | */ |
1949 | swap_map = &si->swap_map[i]; | |
1950 | entry = swp_entry(type, i); | |
02098fea | 1951 | page = read_swap_cache_async(entry, |
23955622 | 1952 | GFP_HIGHUSER_MOVABLE, NULL, 0, false); |
1da177e4 LT |
1953 | if (!page) { |
1954 | /* | |
1955 | * Either swap_duplicate() failed because entry | |
1956 | * has been freed independently, and will not be | |
1957 | * reused since sys_swapoff() already disabled | |
1958 | * allocation from here, or alloc_page() failed. | |
1959 | */ | |
edfe23da SL |
1960 | swcount = *swap_map; |
1961 | /* | |
1962 | * We don't hold lock here, so the swap entry could be | |
1963 | * SWAP_MAP_BAD (when the cluster is discarding). | |
1964 | * Instead of fail out, We can just skip the swap | |
1965 | * entry because swapoff will wait for discarding | |
1966 | * finish anyway. | |
1967 | */ | |
1968 | if (!swcount || swcount == SWAP_MAP_BAD) | |
1da177e4 LT |
1969 | continue; |
1970 | retval = -ENOMEM; | |
1971 | break; | |
1972 | } | |
1973 | ||
1974 | /* | |
1975 | * Don't hold on to start_mm if it looks like exiting. | |
1976 | */ | |
1977 | if (atomic_read(&start_mm->mm_users) == 1) { | |
1978 | mmput(start_mm); | |
1979 | start_mm = &init_mm; | |
3fce371b | 1980 | mmget(&init_mm); |
1da177e4 LT |
1981 | } |
1982 | ||
1983 | /* | |
1984 | * Wait for and lock page. When do_swap_page races with | |
1985 | * try_to_unuse, do_swap_page can handle the fault much | |
1986 | * faster than try_to_unuse can locate the entry. This | |
1987 | * apparently redundant "wait_on_page_locked" lets try_to_unuse | |
1988 | * defer to do_swap_page in such a case - in some tests, | |
1989 | * do_swap_page and try_to_unuse repeatedly compete. | |
1990 | */ | |
1991 | wait_on_page_locked(page); | |
1992 | wait_on_page_writeback(page); | |
1993 | lock_page(page); | |
1994 | wait_on_page_writeback(page); | |
1995 | ||
1996 | /* | |
1997 | * Remove all references to entry. | |
1da177e4 | 1998 | */ |
1da177e4 | 1999 | swcount = *swap_map; |
aaa46865 HD |
2000 | if (swap_count(swcount) == SWAP_MAP_SHMEM) { |
2001 | retval = shmem_unuse(entry, page); | |
2002 | /* page has already been unlocked and released */ | |
2003 | if (retval < 0) | |
2004 | break; | |
2005 | continue; | |
1da177e4 | 2006 | } |
aaa46865 HD |
2007 | if (swap_count(swcount) && start_mm != &init_mm) |
2008 | retval = unuse_mm(start_mm, entry, page); | |
2009 | ||
355cfa73 | 2010 | if (swap_count(*swap_map)) { |
1da177e4 LT |
2011 | int set_start_mm = (*swap_map >= swcount); |
2012 | struct list_head *p = &start_mm->mmlist; | |
2013 | struct mm_struct *new_start_mm = start_mm; | |
2014 | struct mm_struct *prev_mm = start_mm; | |
2015 | struct mm_struct *mm; | |
2016 | ||
3fce371b VN |
2017 | mmget(new_start_mm); |
2018 | mmget(prev_mm); | |
1da177e4 | 2019 | spin_lock(&mmlist_lock); |
aaa46865 | 2020 | while (swap_count(*swap_map) && !retval && |
1da177e4 LT |
2021 | (p = p->next) != &start_mm->mmlist) { |
2022 | mm = list_entry(p, struct mm_struct, mmlist); | |
388f7934 | 2023 | if (!mmget_not_zero(mm)) |
1da177e4 | 2024 | continue; |
1da177e4 LT |
2025 | spin_unlock(&mmlist_lock); |
2026 | mmput(prev_mm); | |
2027 | prev_mm = mm; | |
2028 | ||
2029 | cond_resched(); | |
2030 | ||
2031 | swcount = *swap_map; | |
355cfa73 | 2032 | if (!swap_count(swcount)) /* any usage ? */ |
1da177e4 | 2033 | ; |
aaa46865 | 2034 | else if (mm == &init_mm) |
1da177e4 | 2035 | set_start_mm = 1; |
aaa46865 | 2036 | else |
1da177e4 | 2037 | retval = unuse_mm(mm, entry, page); |
355cfa73 | 2038 | |
32c5fc10 | 2039 | if (set_start_mm && *swap_map < swcount) { |
1da177e4 | 2040 | mmput(new_start_mm); |
3fce371b | 2041 | mmget(mm); |
1da177e4 LT |
2042 | new_start_mm = mm; |
2043 | set_start_mm = 0; | |
2044 | } | |
2045 | spin_lock(&mmlist_lock); | |
2046 | } | |
2047 | spin_unlock(&mmlist_lock); | |
2048 | mmput(prev_mm); | |
2049 | mmput(start_mm); | |
2050 | start_mm = new_start_mm; | |
2051 | } | |
2052 | if (retval) { | |
2053 | unlock_page(page); | |
09cbfeaf | 2054 | put_page(page); |
1da177e4 LT |
2055 | break; |
2056 | } | |
2057 | ||
1da177e4 LT |
2058 | /* |
2059 | * If a reference remains (rare), we would like to leave | |
2060 | * the page in the swap cache; but try_to_unmap could | |
2061 | * then re-duplicate the entry once we drop page lock, | |
2062 | * so we might loop indefinitely; also, that page could | |
2063 | * not be swapped out to other storage meanwhile. So: | |
2064 | * delete from cache even if there's another reference, | |
2065 | * after ensuring that the data has been saved to disk - | |
2066 | * since if the reference remains (rarer), it will be | |
2067 | * read from disk into another page. Splitting into two | |
2068 | * pages would be incorrect if swap supported "shared | |
2069 | * private" pages, but they are handled by tmpfs files. | |
5ad64688 HD |
2070 | * |
2071 | * Given how unuse_vma() targets one particular offset | |
2072 | * in an anon_vma, once the anon_vma has been determined, | |
2073 | * this splitting happens to be just what is needed to | |
2074 | * handle where KSM pages have been swapped out: re-reading | |
2075 | * is unnecessarily slow, but we can fix that later on. | |
1da177e4 | 2076 | */ |
355cfa73 KH |
2077 | if (swap_count(*swap_map) && |
2078 | PageDirty(page) && PageSwapCache(page)) { | |
1da177e4 LT |
2079 | struct writeback_control wbc = { |
2080 | .sync_mode = WB_SYNC_NONE, | |
2081 | }; | |
2082 | ||
e0709829 | 2083 | swap_writepage(compound_head(page), &wbc); |
1da177e4 LT |
2084 | lock_page(page); |
2085 | wait_on_page_writeback(page); | |
2086 | } | |
68bdc8d6 HD |
2087 | |
2088 | /* | |
2089 | * It is conceivable that a racing task removed this page from | |
2090 | * swap cache just before we acquired the page lock at the top, | |
2091 | * or while we dropped it in unuse_mm(). The page might even | |
2092 | * be back in swap cache on another swap area: that we must not | |
2093 | * delete, since it may not have been written out to swap yet. | |
2094 | */ | |
2095 | if (PageSwapCache(page) && | |
e0709829 HY |
2096 | likely(page_private(page) == entry.val) && |
2097 | !page_swapped(page)) | |
2098 | delete_from_swap_cache(compound_head(page)); | |
1da177e4 LT |
2099 | |
2100 | /* | |
2101 | * So we could skip searching mms once swap count went | |
2102 | * to 1, we did not mark any present ptes as dirty: must | |
2706a1b8 | 2103 | * mark page dirty so shrink_page_list will preserve it. |
1da177e4 LT |
2104 | */ |
2105 | SetPageDirty(page); | |
2106 | unlock_page(page); | |
09cbfeaf | 2107 | put_page(page); |
1da177e4 LT |
2108 | |
2109 | /* | |
2110 | * Make sure that we aren't completely killing | |
2111 | * interactive performance. | |
2112 | */ | |
2113 | cond_resched(); | |
38b5faf4 DM |
2114 | if (frontswap && pages_to_unuse > 0) { |
2115 | if (!--pages_to_unuse) | |
2116 | break; | |
2117 | } | |
1da177e4 LT |
2118 | } |
2119 | ||
2120 | mmput(start_mm); | |
1da177e4 LT |
2121 | return retval; |
2122 | } | |
2123 | ||
2124 | /* | |
5d337b91 HD |
2125 | * After a successful try_to_unuse, if no swap is now in use, we know |
2126 | * we can empty the mmlist. swap_lock must be held on entry and exit. | |
2127 | * Note that mmlist_lock nests inside swap_lock, and an mm must be | |
1da177e4 LT |
2128 | * added to the mmlist just after page_duplicate - before would be racy. |
2129 | */ | |
2130 | static void drain_mmlist(void) | |
2131 | { | |
2132 | struct list_head *p, *next; | |
efa90a98 | 2133 | unsigned int type; |
1da177e4 | 2134 | |
efa90a98 HD |
2135 | for (type = 0; type < nr_swapfiles; type++) |
2136 | if (swap_info[type]->inuse_pages) | |
1da177e4 LT |
2137 | return; |
2138 | spin_lock(&mmlist_lock); | |
2139 | list_for_each_safe(p, next, &init_mm.mmlist) | |
2140 | list_del_init(p); | |
2141 | spin_unlock(&mmlist_lock); | |
2142 | } | |
2143 | ||
2144 | /* | |
2145 | * Use this swapdev's extent info to locate the (PAGE_SIZE) block which | |
d4906e1a LS |
2146 | * corresponds to page offset for the specified swap entry. |
2147 | * Note that the type of this function is sector_t, but it returns page offset | |
2148 | * into the bdev, not sector offset. | |
1da177e4 | 2149 | */ |
d4906e1a | 2150 | static sector_t map_swap_entry(swp_entry_t entry, struct block_device **bdev) |
1da177e4 | 2151 | { |
f29ad6a9 HD |
2152 | struct swap_info_struct *sis; |
2153 | struct swap_extent *start_se; | |
2154 | struct swap_extent *se; | |
2155 | pgoff_t offset; | |
2156 | ||
efa90a98 | 2157 | sis = swap_info[swp_type(entry)]; |
f29ad6a9 HD |
2158 | *bdev = sis->bdev; |
2159 | ||
2160 | offset = swp_offset(entry); | |
2161 | start_se = sis->curr_swap_extent; | |
2162 | se = start_se; | |
1da177e4 LT |
2163 | |
2164 | for ( ; ; ) { | |
1da177e4 LT |
2165 | if (se->start_page <= offset && |
2166 | offset < (se->start_page + se->nr_pages)) { | |
2167 | return se->start_block + (offset - se->start_page); | |
2168 | } | |
a8ae4991 | 2169 | se = list_next_entry(se, list); |
1da177e4 LT |
2170 | sis->curr_swap_extent = se; |
2171 | BUG_ON(se == start_se); /* It *must* be present */ | |
2172 | } | |
2173 | } | |
2174 | ||
d4906e1a LS |
2175 | /* |
2176 | * Returns the page offset into bdev for the specified page's swap entry. | |
2177 | */ | |
2178 | sector_t map_swap_page(struct page *page, struct block_device **bdev) | |
2179 | { | |
2180 | swp_entry_t entry; | |
2181 | entry.val = page_private(page); | |
2182 | return map_swap_entry(entry, bdev); | |
2183 | } | |
2184 | ||
1da177e4 LT |
2185 | /* |
2186 | * Free all of a swapdev's extent information | |
2187 | */ | |
2188 | static void destroy_swap_extents(struct swap_info_struct *sis) | |
2189 | { | |
9625a5f2 | 2190 | while (!list_empty(&sis->first_swap_extent.list)) { |
1da177e4 LT |
2191 | struct swap_extent *se; |
2192 | ||
a8ae4991 | 2193 | se = list_first_entry(&sis->first_swap_extent.list, |
1da177e4 LT |
2194 | struct swap_extent, list); |
2195 | list_del(&se->list); | |
2196 | kfree(se); | |
2197 | } | |
62c230bc MG |
2198 | |
2199 | if (sis->flags & SWP_FILE) { | |
2200 | struct file *swap_file = sis->swap_file; | |
2201 | struct address_space *mapping = swap_file->f_mapping; | |
2202 | ||
2203 | sis->flags &= ~SWP_FILE; | |
2204 | mapping->a_ops->swap_deactivate(swap_file); | |
2205 | } | |
1da177e4 LT |
2206 | } |
2207 | ||
2208 | /* | |
2209 | * Add a block range (and the corresponding page range) into this swapdev's | |
11d31886 | 2210 | * extent list. The extent list is kept sorted in page order. |
1da177e4 | 2211 | * |
11d31886 | 2212 | * This function rather assumes that it is called in ascending page order. |
1da177e4 | 2213 | */ |
a509bc1a | 2214 | int |
1da177e4 LT |
2215 | add_swap_extent(struct swap_info_struct *sis, unsigned long start_page, |
2216 | unsigned long nr_pages, sector_t start_block) | |
2217 | { | |
2218 | struct swap_extent *se; | |
2219 | struct swap_extent *new_se; | |
2220 | struct list_head *lh; | |
2221 | ||
9625a5f2 HD |
2222 | if (start_page == 0) { |
2223 | se = &sis->first_swap_extent; | |
2224 | sis->curr_swap_extent = se; | |
2225 | se->start_page = 0; | |
2226 | se->nr_pages = nr_pages; | |
2227 | se->start_block = start_block; | |
2228 | return 1; | |
2229 | } else { | |
2230 | lh = sis->first_swap_extent.list.prev; /* Highest extent */ | |
1da177e4 | 2231 | se = list_entry(lh, struct swap_extent, list); |
11d31886 HD |
2232 | BUG_ON(se->start_page + se->nr_pages != start_page); |
2233 | if (se->start_block + se->nr_pages == start_block) { | |
1da177e4 LT |
2234 | /* Merge it */ |
2235 | se->nr_pages += nr_pages; | |
2236 | return 0; | |
2237 | } | |
1da177e4 LT |
2238 | } |
2239 | ||
2240 | /* | |
2241 | * No merge. Insert a new extent, preserving ordering. | |
2242 | */ | |
2243 | new_se = kmalloc(sizeof(*se), GFP_KERNEL); | |
2244 | if (new_se == NULL) | |
2245 | return -ENOMEM; | |
2246 | new_se->start_page = start_page; | |
2247 | new_se->nr_pages = nr_pages; | |
2248 | new_se->start_block = start_block; | |
2249 | ||
9625a5f2 | 2250 | list_add_tail(&new_se->list, &sis->first_swap_extent.list); |
53092a74 | 2251 | return 1; |
1da177e4 LT |
2252 | } |
2253 | ||
2254 | /* | |
2255 | * A `swap extent' is a simple thing which maps a contiguous range of pages | |
2256 | * onto a contiguous range of disk blocks. An ordered list of swap extents | |
2257 | * is built at swapon time and is then used at swap_writepage/swap_readpage | |
2258 | * time for locating where on disk a page belongs. | |
2259 | * | |
2260 | * If the swapfile is an S_ISBLK block device, a single extent is installed. | |
2261 | * This is done so that the main operating code can treat S_ISBLK and S_ISREG | |
2262 | * swap files identically. | |
2263 | * | |
2264 | * Whether the swapdev is an S_ISREG file or an S_ISBLK blockdev, the swap | |
2265 | * extent list operates in PAGE_SIZE disk blocks. Both S_ISREG and S_ISBLK | |
2266 | * swapfiles are handled *identically* after swapon time. | |
2267 | * | |
2268 | * For S_ISREG swapfiles, setup_swap_extents() will walk all the file's blocks | |
2269 | * and will parse them into an ordered extent list, in PAGE_SIZE chunks. If | |
2270 | * some stray blocks are found which do not fall within the PAGE_SIZE alignment | |
2271 | * requirements, they are simply tossed out - we will never use those blocks | |
2272 | * for swapping. | |
2273 | * | |
b0d9bcd4 | 2274 | * For S_ISREG swapfiles we set S_SWAPFILE across the life of the swapon. This |
1da177e4 LT |
2275 | * prevents root from shooting her foot off by ftruncating an in-use swapfile, |
2276 | * which will scribble on the fs. | |
2277 | * | |
2278 | * The amount of disk space which a single swap extent represents varies. | |
2279 | * Typically it is in the 1-4 megabyte range. So we can have hundreds of | |
2280 | * extents in the list. To avoid much list walking, we cache the previous | |
2281 | * search location in `curr_swap_extent', and start new searches from there. | |
2282 | * This is extremely effective. The average number of iterations in | |
2283 | * map_swap_page() has been measured at about 0.3 per page. - akpm. | |
2284 | */ | |
53092a74 | 2285 | static int setup_swap_extents(struct swap_info_struct *sis, sector_t *span) |
1da177e4 | 2286 | { |
62c230bc MG |
2287 | struct file *swap_file = sis->swap_file; |
2288 | struct address_space *mapping = swap_file->f_mapping; | |
2289 | struct inode *inode = mapping->host; | |
1da177e4 LT |
2290 | int ret; |
2291 | ||
1da177e4 LT |
2292 | if (S_ISBLK(inode->i_mode)) { |
2293 | ret = add_swap_extent(sis, 0, sis->max, 0); | |
53092a74 | 2294 | *span = sis->pages; |
a509bc1a | 2295 | return ret; |
1da177e4 LT |
2296 | } |
2297 | ||
62c230bc | 2298 | if (mapping->a_ops->swap_activate) { |
a509bc1a | 2299 | ret = mapping->a_ops->swap_activate(sis, swap_file, span); |
62c230bc MG |
2300 | if (!ret) { |
2301 | sis->flags |= SWP_FILE; | |
2302 | ret = add_swap_extent(sis, 0, sis->max, 0); | |
2303 | *span = sis->pages; | |
2304 | } | |
a509bc1a | 2305 | return ret; |
62c230bc MG |
2306 | } |
2307 | ||
a509bc1a | 2308 | return generic_swapfile_activate(sis, swap_file, span); |
1da177e4 LT |
2309 | } |
2310 | ||
cf0cac0a | 2311 | static void _enable_swap_info(struct swap_info_struct *p, int prio, |
2a8f9449 SL |
2312 | unsigned char *swap_map, |
2313 | struct swap_cluster_info *cluster_info) | |
40531542 | 2314 | { |
40531542 CEB |
2315 | if (prio >= 0) |
2316 | p->prio = prio; | |
2317 | else | |
2318 | p->prio = --least_priority; | |
18ab4d4c DS |
2319 | /* |
2320 | * the plist prio is negated because plist ordering is | |
2321 | * low-to-high, while swap ordering is high-to-low | |
2322 | */ | |
2323 | p->list.prio = -p->prio; | |
2324 | p->avail_list.prio = -p->prio; | |
40531542 | 2325 | p->swap_map = swap_map; |
2a8f9449 | 2326 | p->cluster_info = cluster_info; |
40531542 | 2327 | p->flags |= SWP_WRITEOK; |
ec8acf20 | 2328 | atomic_long_add(p->pages, &nr_swap_pages); |
40531542 CEB |
2329 | total_swap_pages += p->pages; |
2330 | ||
adfab836 | 2331 | assert_spin_locked(&swap_lock); |
adfab836 | 2332 | /* |
18ab4d4c DS |
2333 | * both lists are plists, and thus priority ordered. |
2334 | * swap_active_head needs to be priority ordered for swapoff(), | |
2335 | * which on removal of any swap_info_struct with an auto-assigned | |
2336 | * (i.e. negative) priority increments the auto-assigned priority | |
2337 | * of any lower-priority swap_info_structs. | |
2338 | * swap_avail_head needs to be priority ordered for get_swap_page(), | |
2339 | * which allocates swap pages from the highest available priority | |
2340 | * swap_info_struct. | |
adfab836 | 2341 | */ |
18ab4d4c DS |
2342 | plist_add(&p->list, &swap_active_head); |
2343 | spin_lock(&swap_avail_lock); | |
2344 | plist_add(&p->avail_list, &swap_avail_head); | |
2345 | spin_unlock(&swap_avail_lock); | |
cf0cac0a CEB |
2346 | } |
2347 | ||
2348 | static void enable_swap_info(struct swap_info_struct *p, int prio, | |
2349 | unsigned char *swap_map, | |
2a8f9449 | 2350 | struct swap_cluster_info *cluster_info, |
cf0cac0a CEB |
2351 | unsigned long *frontswap_map) |
2352 | { | |
4f89849d | 2353 | frontswap_init(p->type, frontswap_map); |
cf0cac0a | 2354 | spin_lock(&swap_lock); |
ec8acf20 | 2355 | spin_lock(&p->lock); |
2a8f9449 | 2356 | _enable_swap_info(p, prio, swap_map, cluster_info); |
ec8acf20 | 2357 | spin_unlock(&p->lock); |
cf0cac0a CEB |
2358 | spin_unlock(&swap_lock); |
2359 | } | |
2360 | ||
2361 | static void reinsert_swap_info(struct swap_info_struct *p) | |
2362 | { | |
2363 | spin_lock(&swap_lock); | |
ec8acf20 | 2364 | spin_lock(&p->lock); |
2a8f9449 | 2365 | _enable_swap_info(p, p->prio, p->swap_map, p->cluster_info); |
ec8acf20 | 2366 | spin_unlock(&p->lock); |
40531542 CEB |
2367 | spin_unlock(&swap_lock); |
2368 | } | |
2369 | ||
67afa38e TC |
2370 | bool has_usable_swap(void) |
2371 | { | |
2372 | bool ret = true; | |
2373 | ||
2374 | spin_lock(&swap_lock); | |
2375 | if (plist_head_empty(&swap_active_head)) | |
2376 | ret = false; | |
2377 | spin_unlock(&swap_lock); | |
2378 | return ret; | |
2379 | } | |
2380 | ||
c4ea37c2 | 2381 | SYSCALL_DEFINE1(swapoff, const char __user *, specialfile) |
1da177e4 | 2382 | { |
73c34b6a | 2383 | struct swap_info_struct *p = NULL; |
8d69aaee | 2384 | unsigned char *swap_map; |
2a8f9449 | 2385 | struct swap_cluster_info *cluster_info; |
4f89849d | 2386 | unsigned long *frontswap_map; |
1da177e4 LT |
2387 | struct file *swap_file, *victim; |
2388 | struct address_space *mapping; | |
2389 | struct inode *inode; | |
91a27b2a | 2390 | struct filename *pathname; |
adfab836 | 2391 | int err, found = 0; |
5b808a23 | 2392 | unsigned int old_block_size; |
886bb7e9 | 2393 | |
1da177e4 LT |
2394 | if (!capable(CAP_SYS_ADMIN)) |
2395 | return -EPERM; | |
2396 | ||
191c5424 AV |
2397 | BUG_ON(!current->mm); |
2398 | ||
1da177e4 | 2399 | pathname = getname(specialfile); |
1da177e4 | 2400 | if (IS_ERR(pathname)) |
f58b59c1 | 2401 | return PTR_ERR(pathname); |
1da177e4 | 2402 | |
669abf4e | 2403 | victim = file_open_name(pathname, O_RDWR|O_LARGEFILE, 0); |
1da177e4 LT |
2404 | err = PTR_ERR(victim); |
2405 | if (IS_ERR(victim)) | |
2406 | goto out; | |
2407 | ||
2408 | mapping = victim->f_mapping; | |
5d337b91 | 2409 | spin_lock(&swap_lock); |
18ab4d4c | 2410 | plist_for_each_entry(p, &swap_active_head, list) { |
22c6f8fd | 2411 | if (p->flags & SWP_WRITEOK) { |
adfab836 DS |
2412 | if (p->swap_file->f_mapping == mapping) { |
2413 | found = 1; | |
1da177e4 | 2414 | break; |
adfab836 | 2415 | } |
1da177e4 | 2416 | } |
1da177e4 | 2417 | } |
adfab836 | 2418 | if (!found) { |
1da177e4 | 2419 | err = -EINVAL; |
5d337b91 | 2420 | spin_unlock(&swap_lock); |
1da177e4 LT |
2421 | goto out_dput; |
2422 | } | |
191c5424 | 2423 | if (!security_vm_enough_memory_mm(current->mm, p->pages)) |
1da177e4 LT |
2424 | vm_unacct_memory(p->pages); |
2425 | else { | |
2426 | err = -ENOMEM; | |
5d337b91 | 2427 | spin_unlock(&swap_lock); |
1da177e4 LT |
2428 | goto out_dput; |
2429 | } | |
18ab4d4c DS |
2430 | spin_lock(&swap_avail_lock); |
2431 | plist_del(&p->avail_list, &swap_avail_head); | |
2432 | spin_unlock(&swap_avail_lock); | |
ec8acf20 | 2433 | spin_lock(&p->lock); |
78ecba08 | 2434 | if (p->prio < 0) { |
adfab836 DS |
2435 | struct swap_info_struct *si = p; |
2436 | ||
18ab4d4c | 2437 | plist_for_each_entry_continue(si, &swap_active_head, list) { |
adfab836 | 2438 | si->prio++; |
18ab4d4c DS |
2439 | si->list.prio--; |
2440 | si->avail_list.prio--; | |
adfab836 | 2441 | } |
78ecba08 HD |
2442 | least_priority++; |
2443 | } | |
18ab4d4c | 2444 | plist_del(&p->list, &swap_active_head); |
ec8acf20 | 2445 | atomic_long_sub(p->pages, &nr_swap_pages); |
1da177e4 LT |
2446 | total_swap_pages -= p->pages; |
2447 | p->flags &= ~SWP_WRITEOK; | |
ec8acf20 | 2448 | spin_unlock(&p->lock); |
5d337b91 | 2449 | spin_unlock(&swap_lock); |
fb4f88dc | 2450 | |
039939a6 TC |
2451 | disable_swap_slots_cache_lock(); |
2452 | ||
e1e12d2f | 2453 | set_current_oom_origin(); |
adfab836 | 2454 | err = try_to_unuse(p->type, false, 0); /* force unuse all pages */ |
e1e12d2f | 2455 | clear_current_oom_origin(); |
1da177e4 | 2456 | |
1da177e4 LT |
2457 | if (err) { |
2458 | /* re-insert swap space back into swap_list */ | |
cf0cac0a | 2459 | reinsert_swap_info(p); |
039939a6 | 2460 | reenable_swap_slots_cache_unlock(); |
1da177e4 LT |
2461 | goto out_dput; |
2462 | } | |
52b7efdb | 2463 | |
039939a6 TC |
2464 | reenable_swap_slots_cache_unlock(); |
2465 | ||
815c2c54 SL |
2466 | flush_work(&p->discard_work); |
2467 | ||
5d337b91 | 2468 | destroy_swap_extents(p); |
570a335b HD |
2469 | if (p->flags & SWP_CONTINUED) |
2470 | free_swap_count_continuations(p); | |
2471 | ||
fc0abb14 | 2472 | mutex_lock(&swapon_mutex); |
5d337b91 | 2473 | spin_lock(&swap_lock); |
ec8acf20 | 2474 | spin_lock(&p->lock); |
5d337b91 HD |
2475 | drain_mmlist(); |
2476 | ||
52b7efdb | 2477 | /* wait for anyone still in scan_swap_map */ |
52b7efdb HD |
2478 | p->highest_bit = 0; /* cuts scans short */ |
2479 | while (p->flags >= SWP_SCANNING) { | |
ec8acf20 | 2480 | spin_unlock(&p->lock); |
5d337b91 | 2481 | spin_unlock(&swap_lock); |
13e4b57f | 2482 | schedule_timeout_uninterruptible(1); |
5d337b91 | 2483 | spin_lock(&swap_lock); |
ec8acf20 | 2484 | spin_lock(&p->lock); |
52b7efdb | 2485 | } |
52b7efdb | 2486 | |
1da177e4 | 2487 | swap_file = p->swap_file; |
5b808a23 | 2488 | old_block_size = p->old_block_size; |
1da177e4 LT |
2489 | p->swap_file = NULL; |
2490 | p->max = 0; | |
2491 | swap_map = p->swap_map; | |
2492 | p->swap_map = NULL; | |
2a8f9449 SL |
2493 | cluster_info = p->cluster_info; |
2494 | p->cluster_info = NULL; | |
4f89849d | 2495 | frontswap_map = frontswap_map_get(p); |
ec8acf20 | 2496 | spin_unlock(&p->lock); |
5d337b91 | 2497 | spin_unlock(&swap_lock); |
adfab836 | 2498 | frontswap_invalidate_area(p->type); |
58e97ba6 | 2499 | frontswap_map_set(p, NULL); |
fc0abb14 | 2500 | mutex_unlock(&swapon_mutex); |
ebc2a1a6 SL |
2501 | free_percpu(p->percpu_cluster); |
2502 | p->percpu_cluster = NULL; | |
1da177e4 | 2503 | vfree(swap_map); |
54f180d3 HY |
2504 | kvfree(cluster_info); |
2505 | kvfree(frontswap_map); | |
2de1a7e4 | 2506 | /* Destroy swap account information */ |
adfab836 | 2507 | swap_cgroup_swapoff(p->type); |
4b3ef9da | 2508 | exit_swap_address_space(p->type); |
27a7faa0 | 2509 | |
1da177e4 LT |
2510 | inode = mapping->host; |
2511 | if (S_ISBLK(inode->i_mode)) { | |
2512 | struct block_device *bdev = I_BDEV(inode); | |
5b808a23 | 2513 | set_blocksize(bdev, old_block_size); |
e525fd89 | 2514 | blkdev_put(bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL); |
1da177e4 | 2515 | } else { |
5955102c | 2516 | inode_lock(inode); |
1da177e4 | 2517 | inode->i_flags &= ~S_SWAPFILE; |
5955102c | 2518 | inode_unlock(inode); |
1da177e4 LT |
2519 | } |
2520 | filp_close(swap_file, NULL); | |
f893ab41 WY |
2521 | |
2522 | /* | |
2523 | * Clear the SWP_USED flag after all resources are freed so that swapon | |
2524 | * can reuse this swap_info in alloc_swap_info() safely. It is ok to | |
2525 | * not hold p->lock after we cleared its SWP_WRITEOK. | |
2526 | */ | |
2527 | spin_lock(&swap_lock); | |
2528 | p->flags = 0; | |
2529 | spin_unlock(&swap_lock); | |
2530 | ||
1da177e4 | 2531 | err = 0; |
66d7dd51 KS |
2532 | atomic_inc(&proc_poll_event); |
2533 | wake_up_interruptible(&proc_poll_wait); | |
1da177e4 LT |
2534 | |
2535 | out_dput: | |
2536 | filp_close(victim, NULL); | |
2537 | out: | |
f58b59c1 | 2538 | putname(pathname); |
1da177e4 LT |
2539 | return err; |
2540 | } | |
2541 | ||
2542 | #ifdef CONFIG_PROC_FS | |
66d7dd51 KS |
2543 | static unsigned swaps_poll(struct file *file, poll_table *wait) |
2544 | { | |
f1514638 | 2545 | struct seq_file *seq = file->private_data; |
66d7dd51 KS |
2546 | |
2547 | poll_wait(file, &proc_poll_wait, wait); | |
2548 | ||
f1514638 KS |
2549 | if (seq->poll_event != atomic_read(&proc_poll_event)) { |
2550 | seq->poll_event = atomic_read(&proc_poll_event); | |
66d7dd51 KS |
2551 | return POLLIN | POLLRDNORM | POLLERR | POLLPRI; |
2552 | } | |
2553 | ||
2554 | return POLLIN | POLLRDNORM; | |
2555 | } | |
2556 | ||
1da177e4 LT |
2557 | /* iterator */ |
2558 | static void *swap_start(struct seq_file *swap, loff_t *pos) | |
2559 | { | |
efa90a98 HD |
2560 | struct swap_info_struct *si; |
2561 | int type; | |
1da177e4 LT |
2562 | loff_t l = *pos; |
2563 | ||
fc0abb14 | 2564 | mutex_lock(&swapon_mutex); |
1da177e4 | 2565 | |
881e4aab SS |
2566 | if (!l) |
2567 | return SEQ_START_TOKEN; | |
2568 | ||
efa90a98 HD |
2569 | for (type = 0; type < nr_swapfiles; type++) { |
2570 | smp_rmb(); /* read nr_swapfiles before swap_info[type] */ | |
2571 | si = swap_info[type]; | |
2572 | if (!(si->flags & SWP_USED) || !si->swap_map) | |
1da177e4 | 2573 | continue; |
881e4aab | 2574 | if (!--l) |
efa90a98 | 2575 | return si; |
1da177e4 LT |
2576 | } |
2577 | ||
2578 | return NULL; | |
2579 | } | |
2580 | ||
2581 | static void *swap_next(struct seq_file *swap, void *v, loff_t *pos) | |
2582 | { | |
efa90a98 HD |
2583 | struct swap_info_struct *si = v; |
2584 | int type; | |
1da177e4 | 2585 | |
881e4aab | 2586 | if (v == SEQ_START_TOKEN) |
efa90a98 HD |
2587 | type = 0; |
2588 | else | |
2589 | type = si->type + 1; | |
881e4aab | 2590 | |
efa90a98 HD |
2591 | for (; type < nr_swapfiles; type++) { |
2592 | smp_rmb(); /* read nr_swapfiles before swap_info[type] */ | |
2593 | si = swap_info[type]; | |
2594 | if (!(si->flags & SWP_USED) || !si->swap_map) | |
1da177e4 LT |
2595 | continue; |
2596 | ++*pos; | |
efa90a98 | 2597 | return si; |
1da177e4 LT |
2598 | } |
2599 | ||
2600 | return NULL; | |
2601 | } | |
2602 | ||
2603 | static void swap_stop(struct seq_file *swap, void *v) | |
2604 | { | |
fc0abb14 | 2605 | mutex_unlock(&swapon_mutex); |
1da177e4 LT |
2606 | } |
2607 | ||
2608 | static int swap_show(struct seq_file *swap, void *v) | |
2609 | { | |
efa90a98 | 2610 | struct swap_info_struct *si = v; |
1da177e4 LT |
2611 | struct file *file; |
2612 | int len; | |
2613 | ||
efa90a98 | 2614 | if (si == SEQ_START_TOKEN) { |
881e4aab SS |
2615 | seq_puts(swap,"Filename\t\t\t\tType\t\tSize\tUsed\tPriority\n"); |
2616 | return 0; | |
2617 | } | |
1da177e4 | 2618 | |
efa90a98 | 2619 | file = si->swap_file; |
2726d566 | 2620 | len = seq_file_path(swap, file, " \t\n\\"); |
6eb396dc | 2621 | seq_printf(swap, "%*s%s\t%u\t%u\t%d\n", |
886bb7e9 | 2622 | len < 40 ? 40 - len : 1, " ", |
496ad9aa | 2623 | S_ISBLK(file_inode(file)->i_mode) ? |
1da177e4 | 2624 | "partition" : "file\t", |
efa90a98 HD |
2625 | si->pages << (PAGE_SHIFT - 10), |
2626 | si->inuse_pages << (PAGE_SHIFT - 10), | |
2627 | si->prio); | |
1da177e4 LT |
2628 | return 0; |
2629 | } | |
2630 | ||
15ad7cdc | 2631 | static const struct seq_operations swaps_op = { |
1da177e4 LT |
2632 | .start = swap_start, |
2633 | .next = swap_next, | |
2634 | .stop = swap_stop, | |
2635 | .show = swap_show | |
2636 | }; | |
2637 | ||
2638 | static int swaps_open(struct inode *inode, struct file *file) | |
2639 | { | |
f1514638 | 2640 | struct seq_file *seq; |
66d7dd51 KS |
2641 | int ret; |
2642 | ||
66d7dd51 | 2643 | ret = seq_open(file, &swaps_op); |
f1514638 | 2644 | if (ret) |
66d7dd51 | 2645 | return ret; |
66d7dd51 | 2646 | |
f1514638 KS |
2647 | seq = file->private_data; |
2648 | seq->poll_event = atomic_read(&proc_poll_event); | |
2649 | return 0; | |
1da177e4 LT |
2650 | } |
2651 | ||
15ad7cdc | 2652 | static const struct file_operations proc_swaps_operations = { |
1da177e4 LT |
2653 | .open = swaps_open, |
2654 | .read = seq_read, | |
2655 | .llseek = seq_lseek, | |
2656 | .release = seq_release, | |
66d7dd51 | 2657 | .poll = swaps_poll, |
1da177e4 LT |
2658 | }; |
2659 | ||
2660 | static int __init procswaps_init(void) | |
2661 | { | |
3d71f86f | 2662 | proc_create("swaps", 0, NULL, &proc_swaps_operations); |
1da177e4 LT |
2663 | return 0; |
2664 | } | |
2665 | __initcall(procswaps_init); | |
2666 | #endif /* CONFIG_PROC_FS */ | |
2667 | ||
1796316a JB |
2668 | #ifdef MAX_SWAPFILES_CHECK |
2669 | static int __init max_swapfiles_check(void) | |
2670 | { | |
2671 | MAX_SWAPFILES_CHECK(); | |
2672 | return 0; | |
2673 | } | |
2674 | late_initcall(max_swapfiles_check); | |
2675 | #endif | |
2676 | ||
53cbb243 | 2677 | static struct swap_info_struct *alloc_swap_info(void) |
1da177e4 | 2678 | { |
73c34b6a | 2679 | struct swap_info_struct *p; |
1da177e4 | 2680 | unsigned int type; |
efa90a98 HD |
2681 | |
2682 | p = kzalloc(sizeof(*p), GFP_KERNEL); | |
2683 | if (!p) | |
53cbb243 | 2684 | return ERR_PTR(-ENOMEM); |
efa90a98 | 2685 | |
5d337b91 | 2686 | spin_lock(&swap_lock); |
efa90a98 HD |
2687 | for (type = 0; type < nr_swapfiles; type++) { |
2688 | if (!(swap_info[type]->flags & SWP_USED)) | |
1da177e4 | 2689 | break; |
efa90a98 | 2690 | } |
0697212a | 2691 | if (type >= MAX_SWAPFILES) { |
5d337b91 | 2692 | spin_unlock(&swap_lock); |
efa90a98 | 2693 | kfree(p); |
730c0581 | 2694 | return ERR_PTR(-EPERM); |
1da177e4 | 2695 | } |
efa90a98 HD |
2696 | if (type >= nr_swapfiles) { |
2697 | p->type = type; | |
2698 | swap_info[type] = p; | |
2699 | /* | |
2700 | * Write swap_info[type] before nr_swapfiles, in case a | |
2701 | * racing procfs swap_start() or swap_next() is reading them. | |
2702 | * (We never shrink nr_swapfiles, we never free this entry.) | |
2703 | */ | |
2704 | smp_wmb(); | |
2705 | nr_swapfiles++; | |
2706 | } else { | |
2707 | kfree(p); | |
2708 | p = swap_info[type]; | |
2709 | /* | |
2710 | * Do not memset this entry: a racing procfs swap_next() | |
2711 | * would be relying on p->type to remain valid. | |
2712 | */ | |
2713 | } | |
9625a5f2 | 2714 | INIT_LIST_HEAD(&p->first_swap_extent.list); |
18ab4d4c DS |
2715 | plist_node_init(&p->list, 0); |
2716 | plist_node_init(&p->avail_list, 0); | |
1da177e4 | 2717 | p->flags = SWP_USED; |
5d337b91 | 2718 | spin_unlock(&swap_lock); |
ec8acf20 | 2719 | spin_lock_init(&p->lock); |
efa90a98 | 2720 | |
53cbb243 | 2721 | return p; |
53cbb243 CEB |
2722 | } |
2723 | ||
4d0e1e10 CEB |
2724 | static int claim_swapfile(struct swap_info_struct *p, struct inode *inode) |
2725 | { | |
2726 | int error; | |
2727 | ||
2728 | if (S_ISBLK(inode->i_mode)) { | |
2729 | p->bdev = bdgrab(I_BDEV(inode)); | |
2730 | error = blkdev_get(p->bdev, | |
6f179af8 | 2731 | FMODE_READ | FMODE_WRITE | FMODE_EXCL, p); |
4d0e1e10 CEB |
2732 | if (error < 0) { |
2733 | p->bdev = NULL; | |
6f179af8 | 2734 | return error; |
4d0e1e10 CEB |
2735 | } |
2736 | p->old_block_size = block_size(p->bdev); | |
2737 | error = set_blocksize(p->bdev, PAGE_SIZE); | |
2738 | if (error < 0) | |
87ade72a | 2739 | return error; |
4d0e1e10 CEB |
2740 | p->flags |= SWP_BLKDEV; |
2741 | } else if (S_ISREG(inode->i_mode)) { | |
2742 | p->bdev = inode->i_sb->s_bdev; | |
5955102c | 2743 | inode_lock(inode); |
87ade72a CEB |
2744 | if (IS_SWAPFILE(inode)) |
2745 | return -EBUSY; | |
2746 | } else | |
2747 | return -EINVAL; | |
4d0e1e10 CEB |
2748 | |
2749 | return 0; | |
4d0e1e10 CEB |
2750 | } |
2751 | ||
ca8bd38b CEB |
2752 | static unsigned long read_swap_header(struct swap_info_struct *p, |
2753 | union swap_header *swap_header, | |
2754 | struct inode *inode) | |
2755 | { | |
2756 | int i; | |
2757 | unsigned long maxpages; | |
2758 | unsigned long swapfilepages; | |
d6bbbd29 | 2759 | unsigned long last_page; |
ca8bd38b CEB |
2760 | |
2761 | if (memcmp("SWAPSPACE2", swap_header->magic.magic, 10)) { | |
465c47fd | 2762 | pr_err("Unable to find swap-space signature\n"); |
38719025 | 2763 | return 0; |
ca8bd38b CEB |
2764 | } |
2765 | ||
2766 | /* swap partition endianess hack... */ | |
2767 | if (swab32(swap_header->info.version) == 1) { | |
2768 | swab32s(&swap_header->info.version); | |
2769 | swab32s(&swap_header->info.last_page); | |
2770 | swab32s(&swap_header->info.nr_badpages); | |
dd111be6 JH |
2771 | if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES) |
2772 | return 0; | |
ca8bd38b CEB |
2773 | for (i = 0; i < swap_header->info.nr_badpages; i++) |
2774 | swab32s(&swap_header->info.badpages[i]); | |
2775 | } | |
2776 | /* Check the swap header's sub-version */ | |
2777 | if (swap_header->info.version != 1) { | |
465c47fd AM |
2778 | pr_warn("Unable to handle swap header version %d\n", |
2779 | swap_header->info.version); | |
38719025 | 2780 | return 0; |
ca8bd38b CEB |
2781 | } |
2782 | ||
2783 | p->lowest_bit = 1; | |
2784 | p->cluster_next = 1; | |
2785 | p->cluster_nr = 0; | |
2786 | ||
2787 | /* | |
2788 | * Find out how many pages are allowed for a single swap | |
9b15b817 | 2789 | * device. There are two limiting factors: 1) the number |
a2c16d6c HD |
2790 | * of bits for the swap offset in the swp_entry_t type, and |
2791 | * 2) the number of bits in the swap pte as defined by the | |
9b15b817 | 2792 | * different architectures. In order to find the |
a2c16d6c | 2793 | * largest possible bit mask, a swap entry with swap type 0 |
ca8bd38b | 2794 | * and swap offset ~0UL is created, encoded to a swap pte, |
a2c16d6c | 2795 | * decoded to a swp_entry_t again, and finally the swap |
ca8bd38b CEB |
2796 | * offset is extracted. This will mask all the bits from |
2797 | * the initial ~0UL mask that can't be encoded in either | |
2798 | * the swp_entry_t or the architecture definition of a | |
9b15b817 | 2799 | * swap pte. |
ca8bd38b CEB |
2800 | */ |
2801 | maxpages = swp_offset(pte_to_swp_entry( | |
9b15b817 | 2802 | swp_entry_to_pte(swp_entry(0, ~0UL)))) + 1; |
d6bbbd29 RJ |
2803 | last_page = swap_header->info.last_page; |
2804 | if (last_page > maxpages) { | |
465c47fd | 2805 | pr_warn("Truncating oversized swap area, only using %luk out of %luk\n", |
d6bbbd29 RJ |
2806 | maxpages << (PAGE_SHIFT - 10), |
2807 | last_page << (PAGE_SHIFT - 10)); | |
2808 | } | |
2809 | if (maxpages > last_page) { | |
2810 | maxpages = last_page + 1; | |
ca8bd38b CEB |
2811 | /* p->max is an unsigned int: don't overflow it */ |
2812 | if ((unsigned int)maxpages == 0) | |
2813 | maxpages = UINT_MAX; | |
2814 | } | |
2815 | p->highest_bit = maxpages - 1; | |
2816 | ||
2817 | if (!maxpages) | |
38719025 | 2818 | return 0; |
ca8bd38b CEB |
2819 | swapfilepages = i_size_read(inode) >> PAGE_SHIFT; |
2820 | if (swapfilepages && maxpages > swapfilepages) { | |
465c47fd | 2821 | pr_warn("Swap area shorter than signature indicates\n"); |
38719025 | 2822 | return 0; |
ca8bd38b CEB |
2823 | } |
2824 | if (swap_header->info.nr_badpages && S_ISREG(inode->i_mode)) | |
38719025 | 2825 | return 0; |
ca8bd38b | 2826 | if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES) |
38719025 | 2827 | return 0; |
ca8bd38b CEB |
2828 | |
2829 | return maxpages; | |
ca8bd38b CEB |
2830 | } |
2831 | ||
4b3ef9da | 2832 | #define SWAP_CLUSTER_INFO_COLS \ |
235b6217 | 2833 | DIV_ROUND_UP(L1_CACHE_BYTES, sizeof(struct swap_cluster_info)) |
4b3ef9da HY |
2834 | #define SWAP_CLUSTER_SPACE_COLS \ |
2835 | DIV_ROUND_UP(SWAP_ADDRESS_SPACE_PAGES, SWAPFILE_CLUSTER) | |
2836 | #define SWAP_CLUSTER_COLS \ | |
2837 | max_t(unsigned int, SWAP_CLUSTER_INFO_COLS, SWAP_CLUSTER_SPACE_COLS) | |
235b6217 | 2838 | |
915d4d7b CEB |
2839 | static int setup_swap_map_and_extents(struct swap_info_struct *p, |
2840 | union swap_header *swap_header, | |
2841 | unsigned char *swap_map, | |
2a8f9449 | 2842 | struct swap_cluster_info *cluster_info, |
915d4d7b CEB |
2843 | unsigned long maxpages, |
2844 | sector_t *span) | |
2845 | { | |
235b6217 | 2846 | unsigned int j, k; |
915d4d7b CEB |
2847 | unsigned int nr_good_pages; |
2848 | int nr_extents; | |
2a8f9449 | 2849 | unsigned long nr_clusters = DIV_ROUND_UP(maxpages, SWAPFILE_CLUSTER); |
235b6217 HY |
2850 | unsigned long col = p->cluster_next / SWAPFILE_CLUSTER % SWAP_CLUSTER_COLS; |
2851 | unsigned long i, idx; | |
915d4d7b CEB |
2852 | |
2853 | nr_good_pages = maxpages - 1; /* omit header page */ | |
2854 | ||
6b534915 HY |
2855 | cluster_list_init(&p->free_clusters); |
2856 | cluster_list_init(&p->discard_clusters); | |
2a8f9449 | 2857 | |
915d4d7b CEB |
2858 | for (i = 0; i < swap_header->info.nr_badpages; i++) { |
2859 | unsigned int page_nr = swap_header->info.badpages[i]; | |
bdb8e3f6 CEB |
2860 | if (page_nr == 0 || page_nr > swap_header->info.last_page) |
2861 | return -EINVAL; | |
915d4d7b CEB |
2862 | if (page_nr < maxpages) { |
2863 | swap_map[page_nr] = SWAP_MAP_BAD; | |
2864 | nr_good_pages--; | |
2a8f9449 SL |
2865 | /* |
2866 | * Haven't marked the cluster free yet, no list | |
2867 | * operation involved | |
2868 | */ | |
2869 | inc_cluster_info_page(p, cluster_info, page_nr); | |
915d4d7b CEB |
2870 | } |
2871 | } | |
2872 | ||
2a8f9449 SL |
2873 | /* Haven't marked the cluster free yet, no list operation involved */ |
2874 | for (i = maxpages; i < round_up(maxpages, SWAPFILE_CLUSTER); i++) | |
2875 | inc_cluster_info_page(p, cluster_info, i); | |
2876 | ||
915d4d7b CEB |
2877 | if (nr_good_pages) { |
2878 | swap_map[0] = SWAP_MAP_BAD; | |
2a8f9449 SL |
2879 | /* |
2880 | * Not mark the cluster free yet, no list | |
2881 | * operation involved | |
2882 | */ | |
2883 | inc_cluster_info_page(p, cluster_info, 0); | |
915d4d7b CEB |
2884 | p->max = maxpages; |
2885 | p->pages = nr_good_pages; | |
2886 | nr_extents = setup_swap_extents(p, span); | |
bdb8e3f6 CEB |
2887 | if (nr_extents < 0) |
2888 | return nr_extents; | |
915d4d7b CEB |
2889 | nr_good_pages = p->pages; |
2890 | } | |
2891 | if (!nr_good_pages) { | |
465c47fd | 2892 | pr_warn("Empty swap-file\n"); |
bdb8e3f6 | 2893 | return -EINVAL; |
915d4d7b CEB |
2894 | } |
2895 | ||
2a8f9449 SL |
2896 | if (!cluster_info) |
2897 | return nr_extents; | |
2898 | ||
235b6217 | 2899 | |
4b3ef9da HY |
2900 | /* |
2901 | * Reduce false cache line sharing between cluster_info and | |
2902 | * sharing same address space. | |
2903 | */ | |
235b6217 HY |
2904 | for (k = 0; k < SWAP_CLUSTER_COLS; k++) { |
2905 | j = (k + col) % SWAP_CLUSTER_COLS; | |
2906 | for (i = 0; i < DIV_ROUND_UP(nr_clusters, SWAP_CLUSTER_COLS); i++) { | |
2907 | idx = i * SWAP_CLUSTER_COLS + j; | |
2908 | if (idx >= nr_clusters) | |
2909 | continue; | |
2910 | if (cluster_count(&cluster_info[idx])) | |
2911 | continue; | |
2a8f9449 | 2912 | cluster_set_flag(&cluster_info[idx], CLUSTER_FLAG_FREE); |
6b534915 HY |
2913 | cluster_list_add_tail(&p->free_clusters, cluster_info, |
2914 | idx); | |
2a8f9449 | 2915 | } |
2a8f9449 | 2916 | } |
915d4d7b | 2917 | return nr_extents; |
915d4d7b CEB |
2918 | } |
2919 | ||
dcf6b7dd RA |
2920 | /* |
2921 | * Helper to sys_swapon determining if a given swap | |
2922 | * backing device queue supports DISCARD operations. | |
2923 | */ | |
2924 | static bool swap_discardable(struct swap_info_struct *si) | |
2925 | { | |
2926 | struct request_queue *q = bdev_get_queue(si->bdev); | |
2927 | ||
2928 | if (!q || !blk_queue_discard(q)) | |
2929 | return false; | |
2930 | ||
2931 | return true; | |
2932 | } | |
2933 | ||
53cbb243 CEB |
2934 | SYSCALL_DEFINE2(swapon, const char __user *, specialfile, int, swap_flags) |
2935 | { | |
2936 | struct swap_info_struct *p; | |
91a27b2a | 2937 | struct filename *name; |
53cbb243 CEB |
2938 | struct file *swap_file = NULL; |
2939 | struct address_space *mapping; | |
40531542 | 2940 | int prio; |
53cbb243 CEB |
2941 | int error; |
2942 | union swap_header *swap_header; | |
915d4d7b | 2943 | int nr_extents; |
53cbb243 CEB |
2944 | sector_t span; |
2945 | unsigned long maxpages; | |
53cbb243 | 2946 | unsigned char *swap_map = NULL; |
2a8f9449 | 2947 | struct swap_cluster_info *cluster_info = NULL; |
38b5faf4 | 2948 | unsigned long *frontswap_map = NULL; |
53cbb243 CEB |
2949 | struct page *page = NULL; |
2950 | struct inode *inode = NULL; | |
53cbb243 | 2951 | |
d15cab97 HD |
2952 | if (swap_flags & ~SWAP_FLAGS_VALID) |
2953 | return -EINVAL; | |
2954 | ||
53cbb243 CEB |
2955 | if (!capable(CAP_SYS_ADMIN)) |
2956 | return -EPERM; | |
2957 | ||
2958 | p = alloc_swap_info(); | |
2542e513 CEB |
2959 | if (IS_ERR(p)) |
2960 | return PTR_ERR(p); | |
53cbb243 | 2961 | |
815c2c54 SL |
2962 | INIT_WORK(&p->discard_work, swap_discard_work); |
2963 | ||
1da177e4 | 2964 | name = getname(specialfile); |
1da177e4 | 2965 | if (IS_ERR(name)) { |
7de7fb6b | 2966 | error = PTR_ERR(name); |
1da177e4 | 2967 | name = NULL; |
bd69010b | 2968 | goto bad_swap; |
1da177e4 | 2969 | } |
669abf4e | 2970 | swap_file = file_open_name(name, O_RDWR|O_LARGEFILE, 0); |
1da177e4 | 2971 | if (IS_ERR(swap_file)) { |
7de7fb6b | 2972 | error = PTR_ERR(swap_file); |
1da177e4 | 2973 | swap_file = NULL; |
bd69010b | 2974 | goto bad_swap; |
1da177e4 LT |
2975 | } |
2976 | ||
2977 | p->swap_file = swap_file; | |
2978 | mapping = swap_file->f_mapping; | |
2130781e | 2979 | inode = mapping->host; |
6f179af8 | 2980 | |
5955102c | 2981 | /* If S_ISREG(inode->i_mode) will do inode_lock(inode); */ |
4d0e1e10 CEB |
2982 | error = claim_swapfile(p, inode); |
2983 | if (unlikely(error)) | |
1da177e4 | 2984 | goto bad_swap; |
1da177e4 | 2985 | |
1da177e4 LT |
2986 | /* |
2987 | * Read the swap header. | |
2988 | */ | |
2989 | if (!mapping->a_ops->readpage) { | |
2990 | error = -EINVAL; | |
2991 | goto bad_swap; | |
2992 | } | |
090d2b18 | 2993 | page = read_mapping_page(mapping, 0, swap_file); |
1da177e4 LT |
2994 | if (IS_ERR(page)) { |
2995 | error = PTR_ERR(page); | |
2996 | goto bad_swap; | |
2997 | } | |
81e33971 | 2998 | swap_header = kmap(page); |
1da177e4 | 2999 | |
ca8bd38b CEB |
3000 | maxpages = read_swap_header(p, swap_header, inode); |
3001 | if (unlikely(!maxpages)) { | |
1da177e4 LT |
3002 | error = -EINVAL; |
3003 | goto bad_swap; | |
3004 | } | |
886bb7e9 | 3005 | |
81e33971 | 3006 | /* OK, set up the swap map and apply the bad block list */ |
803d0c83 | 3007 | swap_map = vzalloc(maxpages); |
81e33971 HD |
3008 | if (!swap_map) { |
3009 | error = -ENOMEM; | |
3010 | goto bad_swap; | |
3011 | } | |
f0571429 MK |
3012 | |
3013 | if (bdi_cap_stable_pages_required(inode_to_bdi(inode))) | |
3014 | p->flags |= SWP_STABLE_WRITES; | |
3015 | ||
2a8f9449 | 3016 | if (p->bdev && blk_queue_nonrot(bdev_get_queue(p->bdev))) { |
6f179af8 | 3017 | int cpu; |
235b6217 | 3018 | unsigned long ci, nr_cluster; |
6f179af8 | 3019 | |
2a8f9449 SL |
3020 | p->flags |= SWP_SOLIDSTATE; |
3021 | /* | |
3022 | * select a random position to start with to help wear leveling | |
3023 | * SSD | |
3024 | */ | |
3025 | p->cluster_next = 1 + (prandom_u32() % p->highest_bit); | |
235b6217 | 3026 | nr_cluster = DIV_ROUND_UP(maxpages, SWAPFILE_CLUSTER); |
2a8f9449 | 3027 | |
54f180d3 HY |
3028 | cluster_info = kvzalloc(nr_cluster * sizeof(*cluster_info), |
3029 | GFP_KERNEL); | |
2a8f9449 SL |
3030 | if (!cluster_info) { |
3031 | error = -ENOMEM; | |
3032 | goto bad_swap; | |
3033 | } | |
235b6217 HY |
3034 | |
3035 | for (ci = 0; ci < nr_cluster; ci++) | |
3036 | spin_lock_init(&((cluster_info + ci)->lock)); | |
3037 | ||
ebc2a1a6 SL |
3038 | p->percpu_cluster = alloc_percpu(struct percpu_cluster); |
3039 | if (!p->percpu_cluster) { | |
3040 | error = -ENOMEM; | |
3041 | goto bad_swap; | |
3042 | } | |
6f179af8 | 3043 | for_each_possible_cpu(cpu) { |
ebc2a1a6 | 3044 | struct percpu_cluster *cluster; |
6f179af8 | 3045 | cluster = per_cpu_ptr(p->percpu_cluster, cpu); |
ebc2a1a6 SL |
3046 | cluster_set_null(&cluster->index); |
3047 | } | |
2a8f9449 | 3048 | } |
1da177e4 | 3049 | |
1421ef3c CEB |
3050 | error = swap_cgroup_swapon(p->type, maxpages); |
3051 | if (error) | |
3052 | goto bad_swap; | |
3053 | ||
915d4d7b | 3054 | nr_extents = setup_swap_map_and_extents(p, swap_header, swap_map, |
2a8f9449 | 3055 | cluster_info, maxpages, &span); |
915d4d7b CEB |
3056 | if (unlikely(nr_extents < 0)) { |
3057 | error = nr_extents; | |
1da177e4 LT |
3058 | goto bad_swap; |
3059 | } | |
38b5faf4 | 3060 | /* frontswap enabled? set up bit-per-page map for frontswap */ |
8ea1d2a1 | 3061 | if (IS_ENABLED(CONFIG_FRONTSWAP)) |
54f180d3 HY |
3062 | frontswap_map = kvzalloc(BITS_TO_LONGS(maxpages) * sizeof(long), |
3063 | GFP_KERNEL); | |
1da177e4 | 3064 | |
2a8f9449 SL |
3065 | if (p->bdev &&(swap_flags & SWAP_FLAG_DISCARD) && swap_discardable(p)) { |
3066 | /* | |
3067 | * When discard is enabled for swap with no particular | |
3068 | * policy flagged, we set all swap discard flags here in | |
3069 | * order to sustain backward compatibility with older | |
3070 | * swapon(8) releases. | |
3071 | */ | |
3072 | p->flags |= (SWP_DISCARDABLE | SWP_AREA_DISCARD | | |
3073 | SWP_PAGE_DISCARD); | |
dcf6b7dd | 3074 | |
2a8f9449 SL |
3075 | /* |
3076 | * By flagging sys_swapon, a sysadmin can tell us to | |
3077 | * either do single-time area discards only, or to just | |
3078 | * perform discards for released swap page-clusters. | |
3079 | * Now it's time to adjust the p->flags accordingly. | |
3080 | */ | |
3081 | if (swap_flags & SWAP_FLAG_DISCARD_ONCE) | |
3082 | p->flags &= ~SWP_PAGE_DISCARD; | |
3083 | else if (swap_flags & SWAP_FLAG_DISCARD_PAGES) | |
3084 | p->flags &= ~SWP_AREA_DISCARD; | |
3085 | ||
3086 | /* issue a swapon-time discard if it's still required */ | |
3087 | if (p->flags & SWP_AREA_DISCARD) { | |
3088 | int err = discard_swap(p); | |
3089 | if (unlikely(err)) | |
3090 | pr_err("swapon: discard_swap(%p): %d\n", | |
3091 | p, err); | |
dcf6b7dd | 3092 | } |
20137a49 | 3093 | } |
6a6ba831 | 3094 | |
4b3ef9da HY |
3095 | error = init_swap_address_space(p->type, maxpages); |
3096 | if (error) | |
3097 | goto bad_swap; | |
3098 | ||
fc0abb14 | 3099 | mutex_lock(&swapon_mutex); |
40531542 | 3100 | prio = -1; |
78ecba08 | 3101 | if (swap_flags & SWAP_FLAG_PREFER) |
40531542 | 3102 | prio = |
78ecba08 | 3103 | (swap_flags & SWAP_FLAG_PRIO_MASK) >> SWAP_FLAG_PRIO_SHIFT; |
2a8f9449 | 3104 | enable_swap_info(p, prio, swap_map, cluster_info, frontswap_map); |
c69dbfb8 | 3105 | |
756a025f | 3106 | pr_info("Adding %uk swap on %s. Priority:%d extents:%d across:%lluk %s%s%s%s%s\n", |
91a27b2a | 3107 | p->pages<<(PAGE_SHIFT-10), name->name, p->prio, |
c69dbfb8 CEB |
3108 | nr_extents, (unsigned long long)span<<(PAGE_SHIFT-10), |
3109 | (p->flags & SWP_SOLIDSTATE) ? "SS" : "", | |
38b5faf4 | 3110 | (p->flags & SWP_DISCARDABLE) ? "D" : "", |
dcf6b7dd RA |
3111 | (p->flags & SWP_AREA_DISCARD) ? "s" : "", |
3112 | (p->flags & SWP_PAGE_DISCARD) ? "c" : "", | |
38b5faf4 | 3113 | (frontswap_map) ? "FS" : ""); |
c69dbfb8 | 3114 | |
fc0abb14 | 3115 | mutex_unlock(&swapon_mutex); |
66d7dd51 KS |
3116 | atomic_inc(&proc_poll_event); |
3117 | wake_up_interruptible(&proc_poll_wait); | |
3118 | ||
9b01c350 CEB |
3119 | if (S_ISREG(inode->i_mode)) |
3120 | inode->i_flags |= S_SWAPFILE; | |
1da177e4 LT |
3121 | error = 0; |
3122 | goto out; | |
3123 | bad_swap: | |
ebc2a1a6 SL |
3124 | free_percpu(p->percpu_cluster); |
3125 | p->percpu_cluster = NULL; | |
bd69010b | 3126 | if (inode && S_ISBLK(inode->i_mode) && p->bdev) { |
f2090d2d CEB |
3127 | set_blocksize(p->bdev, p->old_block_size); |
3128 | blkdev_put(p->bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL); | |
1da177e4 | 3129 | } |
4cd3bb10 | 3130 | destroy_swap_extents(p); |
e8e6c2ec | 3131 | swap_cgroup_swapoff(p->type); |
5d337b91 | 3132 | spin_lock(&swap_lock); |
1da177e4 | 3133 | p->swap_file = NULL; |
1da177e4 | 3134 | p->flags = 0; |
5d337b91 | 3135 | spin_unlock(&swap_lock); |
1da177e4 | 3136 | vfree(swap_map); |
2a8f9449 | 3137 | vfree(cluster_info); |
52c50567 | 3138 | if (swap_file) { |
2130781e | 3139 | if (inode && S_ISREG(inode->i_mode)) { |
5955102c | 3140 | inode_unlock(inode); |
2130781e CEB |
3141 | inode = NULL; |
3142 | } | |
1da177e4 | 3143 | filp_close(swap_file, NULL); |
52c50567 | 3144 | } |
1da177e4 LT |
3145 | out: |
3146 | if (page && !IS_ERR(page)) { | |
3147 | kunmap(page); | |
09cbfeaf | 3148 | put_page(page); |
1da177e4 LT |
3149 | } |
3150 | if (name) | |
3151 | putname(name); | |
9b01c350 | 3152 | if (inode && S_ISREG(inode->i_mode)) |
5955102c | 3153 | inode_unlock(inode); |
039939a6 TC |
3154 | if (!error) |
3155 | enable_swap_slots_cache(); | |
1da177e4 LT |
3156 | return error; |
3157 | } | |
3158 | ||
3159 | void si_swapinfo(struct sysinfo *val) | |
3160 | { | |
efa90a98 | 3161 | unsigned int type; |
1da177e4 LT |
3162 | unsigned long nr_to_be_unused = 0; |
3163 | ||
5d337b91 | 3164 | spin_lock(&swap_lock); |
efa90a98 HD |
3165 | for (type = 0; type < nr_swapfiles; type++) { |
3166 | struct swap_info_struct *si = swap_info[type]; | |
3167 | ||
3168 | if ((si->flags & SWP_USED) && !(si->flags & SWP_WRITEOK)) | |
3169 | nr_to_be_unused += si->inuse_pages; | |
1da177e4 | 3170 | } |
ec8acf20 | 3171 | val->freeswap = atomic_long_read(&nr_swap_pages) + nr_to_be_unused; |
1da177e4 | 3172 | val->totalswap = total_swap_pages + nr_to_be_unused; |
5d337b91 | 3173 | spin_unlock(&swap_lock); |
1da177e4 LT |
3174 | } |
3175 | ||
3176 | /* | |
3177 | * Verify that a swap entry is valid and increment its swap map count. | |
3178 | * | |
355cfa73 KH |
3179 | * Returns error code in following case. |
3180 | * - success -> 0 | |
3181 | * - swp_entry is invalid -> EINVAL | |
3182 | * - swp_entry is migration entry -> EINVAL | |
3183 | * - swap-cache reference is requested but there is already one. -> EEXIST | |
3184 | * - swap-cache reference is requested but the entry is not used. -> ENOENT | |
570a335b | 3185 | * - swap-mapped reference requested but needs continued swap count. -> ENOMEM |
1da177e4 | 3186 | */ |
8d69aaee | 3187 | static int __swap_duplicate(swp_entry_t entry, unsigned char usage) |
1da177e4 | 3188 | { |
73c34b6a | 3189 | struct swap_info_struct *p; |
235b6217 | 3190 | struct swap_cluster_info *ci; |
1da177e4 | 3191 | unsigned long offset, type; |
8d69aaee HD |
3192 | unsigned char count; |
3193 | unsigned char has_cache; | |
253d553b | 3194 | int err = -EINVAL; |
1da177e4 | 3195 | |
a7420aa5 | 3196 | if (non_swap_entry(entry)) |
253d553b | 3197 | goto out; |
0697212a | 3198 | |
1da177e4 LT |
3199 | type = swp_type(entry); |
3200 | if (type >= nr_swapfiles) | |
3201 | goto bad_file; | |
efa90a98 | 3202 | p = swap_info[type]; |
1da177e4 | 3203 | offset = swp_offset(entry); |
355cfa73 | 3204 | if (unlikely(offset >= p->max)) |
235b6217 HY |
3205 | goto out; |
3206 | ||
3207 | ci = lock_cluster_or_swap_info(p, offset); | |
355cfa73 | 3208 | |
253d553b | 3209 | count = p->swap_map[offset]; |
edfe23da SL |
3210 | |
3211 | /* | |
3212 | * swapin_readahead() doesn't check if a swap entry is valid, so the | |
3213 | * swap entry could be SWAP_MAP_BAD. Check here with lock held. | |
3214 | */ | |
3215 | if (unlikely(swap_count(count) == SWAP_MAP_BAD)) { | |
3216 | err = -ENOENT; | |
3217 | goto unlock_out; | |
3218 | } | |
3219 | ||
253d553b HD |
3220 | has_cache = count & SWAP_HAS_CACHE; |
3221 | count &= ~SWAP_HAS_CACHE; | |
3222 | err = 0; | |
355cfa73 | 3223 | |
253d553b | 3224 | if (usage == SWAP_HAS_CACHE) { |
355cfa73 KH |
3225 | |
3226 | /* set SWAP_HAS_CACHE if there is no cache and entry is used */ | |
253d553b HD |
3227 | if (!has_cache && count) |
3228 | has_cache = SWAP_HAS_CACHE; | |
3229 | else if (has_cache) /* someone else added cache */ | |
3230 | err = -EEXIST; | |
3231 | else /* no users remaining */ | |
3232 | err = -ENOENT; | |
355cfa73 KH |
3233 | |
3234 | } else if (count || has_cache) { | |
253d553b | 3235 | |
570a335b HD |
3236 | if ((count & ~COUNT_CONTINUED) < SWAP_MAP_MAX) |
3237 | count += usage; | |
3238 | else if ((count & ~COUNT_CONTINUED) > SWAP_MAP_MAX) | |
253d553b | 3239 | err = -EINVAL; |
570a335b HD |
3240 | else if (swap_count_continued(p, offset, count)) |
3241 | count = COUNT_CONTINUED; | |
3242 | else | |
3243 | err = -ENOMEM; | |
355cfa73 | 3244 | } else |
253d553b HD |
3245 | err = -ENOENT; /* unused swap entry */ |
3246 | ||
3247 | p->swap_map[offset] = count | has_cache; | |
3248 | ||
355cfa73 | 3249 | unlock_out: |
235b6217 | 3250 | unlock_cluster_or_swap_info(p, ci); |
1da177e4 | 3251 | out: |
253d553b | 3252 | return err; |
1da177e4 LT |
3253 | |
3254 | bad_file: | |
465c47fd | 3255 | pr_err("swap_dup: %s%08lx\n", Bad_file, entry.val); |
1da177e4 LT |
3256 | goto out; |
3257 | } | |
253d553b | 3258 | |
aaa46865 HD |
3259 | /* |
3260 | * Help swapoff by noting that swap entry belongs to shmem/tmpfs | |
3261 | * (in which case its reference count is never incremented). | |
3262 | */ | |
3263 | void swap_shmem_alloc(swp_entry_t entry) | |
3264 | { | |
3265 | __swap_duplicate(entry, SWAP_MAP_SHMEM); | |
3266 | } | |
3267 | ||
355cfa73 | 3268 | /* |
08259d58 HD |
3269 | * Increase reference count of swap entry by 1. |
3270 | * Returns 0 for success, or -ENOMEM if a swap_count_continuation is required | |
3271 | * but could not be atomically allocated. Returns 0, just as if it succeeded, | |
3272 | * if __swap_duplicate() fails for another reason (-EINVAL or -ENOENT), which | |
3273 | * might occur if a page table entry has got corrupted. | |
355cfa73 | 3274 | */ |
570a335b | 3275 | int swap_duplicate(swp_entry_t entry) |
355cfa73 | 3276 | { |
570a335b HD |
3277 | int err = 0; |
3278 | ||
3279 | while (!err && __swap_duplicate(entry, 1) == -ENOMEM) | |
3280 | err = add_swap_count_continuation(entry, GFP_ATOMIC); | |
3281 | return err; | |
355cfa73 | 3282 | } |
1da177e4 | 3283 | |
cb4b86ba | 3284 | /* |
355cfa73 KH |
3285 | * @entry: swap entry for which we allocate swap cache. |
3286 | * | |
73c34b6a | 3287 | * Called when allocating swap cache for existing swap entry, |
355cfa73 KH |
3288 | * This can return error codes. Returns 0 at success. |
3289 | * -EBUSY means there is a swap cache. | |
3290 | * Note: return code is different from swap_duplicate(). | |
cb4b86ba KH |
3291 | */ |
3292 | int swapcache_prepare(swp_entry_t entry) | |
3293 | { | |
253d553b | 3294 | return __swap_duplicate(entry, SWAP_HAS_CACHE); |
cb4b86ba KH |
3295 | } |
3296 | ||
f981c595 MG |
3297 | struct swap_info_struct *page_swap_info(struct page *page) |
3298 | { | |
3299 | swp_entry_t swap = { .val = page_private(page) }; | |
f981c595 MG |
3300 | return swap_info[swp_type(swap)]; |
3301 | } | |
3302 | ||
3303 | /* | |
3304 | * out-of-line __page_file_ methods to avoid include hell. | |
3305 | */ | |
3306 | struct address_space *__page_file_mapping(struct page *page) | |
3307 | { | |
309381fe | 3308 | VM_BUG_ON_PAGE(!PageSwapCache(page), page); |
f981c595 MG |
3309 | return page_swap_info(page)->swap_file->f_mapping; |
3310 | } | |
3311 | EXPORT_SYMBOL_GPL(__page_file_mapping); | |
3312 | ||
3313 | pgoff_t __page_file_index(struct page *page) | |
3314 | { | |
3315 | swp_entry_t swap = { .val = page_private(page) }; | |
309381fe | 3316 | VM_BUG_ON_PAGE(!PageSwapCache(page), page); |
f981c595 MG |
3317 | return swp_offset(swap); |
3318 | } | |
3319 | EXPORT_SYMBOL_GPL(__page_file_index); | |
3320 | ||
570a335b HD |
3321 | /* |
3322 | * add_swap_count_continuation - called when a swap count is duplicated | |
3323 | * beyond SWAP_MAP_MAX, it allocates a new page and links that to the entry's | |
3324 | * page of the original vmalloc'ed swap_map, to hold the continuation count | |
3325 | * (for that entry and for its neighbouring PAGE_SIZE swap entries). Called | |
3326 | * again when count is duplicated beyond SWAP_MAP_MAX * SWAP_CONT_MAX, etc. | |
3327 | * | |
3328 | * These continuation pages are seldom referenced: the common paths all work | |
3329 | * on the original swap_map, only referring to a continuation page when the | |
3330 | * low "digit" of a count is incremented or decremented through SWAP_MAP_MAX. | |
3331 | * | |
3332 | * add_swap_count_continuation(, GFP_ATOMIC) can be called while holding | |
3333 | * page table locks; if it fails, add_swap_count_continuation(, GFP_KERNEL) | |
3334 | * can be called after dropping locks. | |
3335 | */ | |
3336 | int add_swap_count_continuation(swp_entry_t entry, gfp_t gfp_mask) | |
3337 | { | |
3338 | struct swap_info_struct *si; | |
235b6217 | 3339 | struct swap_cluster_info *ci; |
570a335b HD |
3340 | struct page *head; |
3341 | struct page *page; | |
3342 | struct page *list_page; | |
3343 | pgoff_t offset; | |
3344 | unsigned char count; | |
3345 | ||
3346 | /* | |
3347 | * When debugging, it's easier to use __GFP_ZERO here; but it's better | |
3348 | * for latency not to zero a page while GFP_ATOMIC and holding locks. | |
3349 | */ | |
3350 | page = alloc_page(gfp_mask | __GFP_HIGHMEM); | |
3351 | ||
3352 | si = swap_info_get(entry); | |
3353 | if (!si) { | |
3354 | /* | |
3355 | * An acceptable race has occurred since the failing | |
3356 | * __swap_duplicate(): the swap entry has been freed, | |
3357 | * perhaps even the whole swap_map cleared for swapoff. | |
3358 | */ | |
3359 | goto outer; | |
3360 | } | |
3361 | ||
3362 | offset = swp_offset(entry); | |
235b6217 HY |
3363 | |
3364 | ci = lock_cluster(si, offset); | |
3365 | ||
570a335b HD |
3366 | count = si->swap_map[offset] & ~SWAP_HAS_CACHE; |
3367 | ||
3368 | if ((count & ~COUNT_CONTINUED) != SWAP_MAP_MAX) { | |
3369 | /* | |
3370 | * The higher the swap count, the more likely it is that tasks | |
3371 | * will race to add swap count continuation: we need to avoid | |
3372 | * over-provisioning. | |
3373 | */ | |
3374 | goto out; | |
3375 | } | |
3376 | ||
3377 | if (!page) { | |
235b6217 | 3378 | unlock_cluster(ci); |
ec8acf20 | 3379 | spin_unlock(&si->lock); |
570a335b HD |
3380 | return -ENOMEM; |
3381 | } | |
3382 | ||
3383 | /* | |
3384 | * We are fortunate that although vmalloc_to_page uses pte_offset_map, | |
2de1a7e4 SJ |
3385 | * no architecture is using highmem pages for kernel page tables: so it |
3386 | * will not corrupt the GFP_ATOMIC caller's atomic page table kmaps. | |
570a335b HD |
3387 | */ |
3388 | head = vmalloc_to_page(si->swap_map + offset); | |
3389 | offset &= ~PAGE_MASK; | |
3390 | ||
3391 | /* | |
3392 | * Page allocation does not initialize the page's lru field, | |
3393 | * but it does always reset its private field. | |
3394 | */ | |
3395 | if (!page_private(head)) { | |
3396 | BUG_ON(count & COUNT_CONTINUED); | |
3397 | INIT_LIST_HEAD(&head->lru); | |
3398 | set_page_private(head, SWP_CONTINUED); | |
3399 | si->flags |= SWP_CONTINUED; | |
3400 | } | |
3401 | ||
3402 | list_for_each_entry(list_page, &head->lru, lru) { | |
3403 | unsigned char *map; | |
3404 | ||
3405 | /* | |
3406 | * If the previous map said no continuation, but we've found | |
3407 | * a continuation page, free our allocation and use this one. | |
3408 | */ | |
3409 | if (!(count & COUNT_CONTINUED)) | |
3410 | goto out; | |
3411 | ||
9b04c5fe | 3412 | map = kmap_atomic(list_page) + offset; |
570a335b | 3413 | count = *map; |
9b04c5fe | 3414 | kunmap_atomic(map); |
570a335b HD |
3415 | |
3416 | /* | |
3417 | * If this continuation count now has some space in it, | |
3418 | * free our allocation and use this one. | |
3419 | */ | |
3420 | if ((count & ~COUNT_CONTINUED) != SWAP_CONT_MAX) | |
3421 | goto out; | |
3422 | } | |
3423 | ||
3424 | list_add_tail(&page->lru, &head->lru); | |
3425 | page = NULL; /* now it's attached, don't free it */ | |
3426 | out: | |
235b6217 | 3427 | unlock_cluster(ci); |
ec8acf20 | 3428 | spin_unlock(&si->lock); |
570a335b HD |
3429 | outer: |
3430 | if (page) | |
3431 | __free_page(page); | |
3432 | return 0; | |
3433 | } | |
3434 | ||
3435 | /* | |
3436 | * swap_count_continued - when the original swap_map count is incremented | |
3437 | * from SWAP_MAP_MAX, check if there is already a continuation page to carry | |
3438 | * into, carry if so, or else fail until a new continuation page is allocated; | |
3439 | * when the original swap_map count is decremented from 0 with continuation, | |
3440 | * borrow from the continuation and report whether it still holds more. | |
235b6217 HY |
3441 | * Called while __swap_duplicate() or swap_entry_free() holds swap or cluster |
3442 | * lock. | |
570a335b HD |
3443 | */ |
3444 | static bool swap_count_continued(struct swap_info_struct *si, | |
3445 | pgoff_t offset, unsigned char count) | |
3446 | { | |
3447 | struct page *head; | |
3448 | struct page *page; | |
3449 | unsigned char *map; | |
3450 | ||
3451 | head = vmalloc_to_page(si->swap_map + offset); | |
3452 | if (page_private(head) != SWP_CONTINUED) { | |
3453 | BUG_ON(count & COUNT_CONTINUED); | |
3454 | return false; /* need to add count continuation */ | |
3455 | } | |
3456 | ||
3457 | offset &= ~PAGE_MASK; | |
3458 | page = list_entry(head->lru.next, struct page, lru); | |
9b04c5fe | 3459 | map = kmap_atomic(page) + offset; |
570a335b HD |
3460 | |
3461 | if (count == SWAP_MAP_MAX) /* initial increment from swap_map */ | |
3462 | goto init_map; /* jump over SWAP_CONT_MAX checks */ | |
3463 | ||
3464 | if (count == (SWAP_MAP_MAX | COUNT_CONTINUED)) { /* incrementing */ | |
3465 | /* | |
3466 | * Think of how you add 1 to 999 | |
3467 | */ | |
3468 | while (*map == (SWAP_CONT_MAX | COUNT_CONTINUED)) { | |
9b04c5fe | 3469 | kunmap_atomic(map); |
570a335b HD |
3470 | page = list_entry(page->lru.next, struct page, lru); |
3471 | BUG_ON(page == head); | |
9b04c5fe | 3472 | map = kmap_atomic(page) + offset; |
570a335b HD |
3473 | } |
3474 | if (*map == SWAP_CONT_MAX) { | |
9b04c5fe | 3475 | kunmap_atomic(map); |
570a335b HD |
3476 | page = list_entry(page->lru.next, struct page, lru); |
3477 | if (page == head) | |
3478 | return false; /* add count continuation */ | |
9b04c5fe | 3479 | map = kmap_atomic(page) + offset; |
570a335b HD |
3480 | init_map: *map = 0; /* we didn't zero the page */ |
3481 | } | |
3482 | *map += 1; | |
9b04c5fe | 3483 | kunmap_atomic(map); |
570a335b HD |
3484 | page = list_entry(page->lru.prev, struct page, lru); |
3485 | while (page != head) { | |
9b04c5fe | 3486 | map = kmap_atomic(page) + offset; |
570a335b | 3487 | *map = COUNT_CONTINUED; |
9b04c5fe | 3488 | kunmap_atomic(map); |
570a335b HD |
3489 | page = list_entry(page->lru.prev, struct page, lru); |
3490 | } | |
3491 | return true; /* incremented */ | |
3492 | ||
3493 | } else { /* decrementing */ | |
3494 | /* | |
3495 | * Think of how you subtract 1 from 1000 | |
3496 | */ | |
3497 | BUG_ON(count != COUNT_CONTINUED); | |
3498 | while (*map == COUNT_CONTINUED) { | |
9b04c5fe | 3499 | kunmap_atomic(map); |
570a335b HD |
3500 | page = list_entry(page->lru.next, struct page, lru); |
3501 | BUG_ON(page == head); | |
9b04c5fe | 3502 | map = kmap_atomic(page) + offset; |
570a335b HD |
3503 | } |
3504 | BUG_ON(*map == 0); | |
3505 | *map -= 1; | |
3506 | if (*map == 0) | |
3507 | count = 0; | |
9b04c5fe | 3508 | kunmap_atomic(map); |
570a335b HD |
3509 | page = list_entry(page->lru.prev, struct page, lru); |
3510 | while (page != head) { | |
9b04c5fe | 3511 | map = kmap_atomic(page) + offset; |
570a335b HD |
3512 | *map = SWAP_CONT_MAX | count; |
3513 | count = COUNT_CONTINUED; | |
9b04c5fe | 3514 | kunmap_atomic(map); |
570a335b HD |
3515 | page = list_entry(page->lru.prev, struct page, lru); |
3516 | } | |
3517 | return count == COUNT_CONTINUED; | |
3518 | } | |
3519 | } | |
3520 | ||
3521 | /* | |
3522 | * free_swap_count_continuations - swapoff free all the continuation pages | |
3523 | * appended to the swap_map, after swap_map is quiesced, before vfree'ing it. | |
3524 | */ | |
3525 | static void free_swap_count_continuations(struct swap_info_struct *si) | |
3526 | { | |
3527 | pgoff_t offset; | |
3528 | ||
3529 | for (offset = 0; offset < si->max; offset += PAGE_SIZE) { | |
3530 | struct page *head; | |
3531 | head = vmalloc_to_page(si->swap_map + offset); | |
3532 | if (page_private(head)) { | |
0d576d20 GT |
3533 | struct page *page, *next; |
3534 | ||
3535 | list_for_each_entry_safe(page, next, &head->lru, lru) { | |
3536 | list_del(&page->lru); | |
570a335b HD |
3537 | __free_page(page); |
3538 | } | |
3539 | } | |
3540 | } | |
3541 | } |