]>
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> |
1da177e4 LT |
10 | #include <linux/hugetlb.h> |
11 | #include <linux/mman.h> | |
12 | #include <linux/slab.h> | |
13 | #include <linux/kernel_stat.h> | |
14 | #include <linux/swap.h> | |
15 | #include <linux/vmalloc.h> | |
16 | #include <linux/pagemap.h> | |
17 | #include <linux/namei.h> | |
072441e2 | 18 | #include <linux/shmem_fs.h> |
1da177e4 | 19 | #include <linux/blkdev.h> |
20137a49 | 20 | #include <linux/random.h> |
1da177e4 LT |
21 | #include <linux/writeback.h> |
22 | #include <linux/proc_fs.h> | |
23 | #include <linux/seq_file.h> | |
24 | #include <linux/init.h> | |
5ad64688 | 25 | #include <linux/ksm.h> |
1da177e4 LT |
26 | #include <linux/rmap.h> |
27 | #include <linux/security.h> | |
28 | #include <linux/backing-dev.h> | |
fc0abb14 | 29 | #include <linux/mutex.h> |
c59ede7b | 30 | #include <linux/capability.h> |
1da177e4 | 31 | #include <linux/syscalls.h> |
8a9f3ccd | 32 | #include <linux/memcontrol.h> |
66d7dd51 | 33 | #include <linux/poll.h> |
72788c38 | 34 | #include <linux/oom.h> |
38b5faf4 DM |
35 | #include <linux/frontswap.h> |
36 | #include <linux/swapfile.h> | |
f981c595 | 37 | #include <linux/export.h> |
67afa38e | 38 | #include <linux/swap_slots.h> |
1da177e4 LT |
39 | |
40 | #include <asm/pgtable.h> | |
41 | #include <asm/tlbflush.h> | |
42 | #include <linux/swapops.h> | |
5d1ea48b | 43 | #include <linux/swap_cgroup.h> |
1da177e4 | 44 | |
570a335b HD |
45 | static bool swap_count_continued(struct swap_info_struct *, pgoff_t, |
46 | unsigned char); | |
47 | static void free_swap_count_continuations(struct swap_info_struct *); | |
d4906e1a | 48 | static sector_t map_swap_entry(swp_entry_t, struct block_device**); |
570a335b | 49 | |
38b5faf4 | 50 | DEFINE_SPINLOCK(swap_lock); |
7c363b8c | 51 | static unsigned int nr_swapfiles; |
ec8acf20 | 52 | atomic_long_t nr_swap_pages; |
fb0fec50 CW |
53 | /* |
54 | * Some modules use swappable objects and may try to swap them out under | |
55 | * memory pressure (via the shrinker). Before doing so, they may wish to | |
56 | * check to see if any swap space is available. | |
57 | */ | |
58 | EXPORT_SYMBOL_GPL(nr_swap_pages); | |
ec8acf20 | 59 | /* protected with swap_lock. reading in vm_swap_full() doesn't need lock */ |
1da177e4 | 60 | long total_swap_pages; |
78ecba08 | 61 | static int least_priority; |
1da177e4 | 62 | |
1da177e4 LT |
63 | static const char Bad_file[] = "Bad swap file entry "; |
64 | static const char Unused_file[] = "Unused swap file entry "; | |
65 | static const char Bad_offset[] = "Bad swap offset entry "; | |
66 | static const char Unused_offset[] = "Unused swap offset entry "; | |
67 | ||
adfab836 DS |
68 | /* |
69 | * all active swap_info_structs | |
70 | * protected with swap_lock, and ordered by priority. | |
71 | */ | |
18ab4d4c DS |
72 | PLIST_HEAD(swap_active_head); |
73 | ||
74 | /* | |
75 | * all available (active, not full) swap_info_structs | |
76 | * protected with swap_avail_lock, ordered by priority. | |
77 | * This is used by get_swap_page() instead of swap_active_head | |
78 | * because swap_active_head includes all swap_info_structs, | |
79 | * but get_swap_page() doesn't need to look at full ones. | |
80 | * This uses its own lock instead of swap_lock because when a | |
81 | * swap_info_struct changes between not-full/full, it needs to | |
82 | * add/remove itself to/from this list, but the swap_info_struct->lock | |
83 | * is held and the locking order requires swap_lock to be taken | |
84 | * before any swap_info_struct->lock. | |
85 | */ | |
86 | static PLIST_HEAD(swap_avail_head); | |
87 | static DEFINE_SPINLOCK(swap_avail_lock); | |
1da177e4 | 88 | |
38b5faf4 | 89 | struct swap_info_struct *swap_info[MAX_SWAPFILES]; |
1da177e4 | 90 | |
fc0abb14 | 91 | static DEFINE_MUTEX(swapon_mutex); |
1da177e4 | 92 | |
66d7dd51 KS |
93 | static DECLARE_WAIT_QUEUE_HEAD(proc_poll_wait); |
94 | /* Activity counter to indicate that a swapon or swapoff has occurred */ | |
95 | static atomic_t proc_poll_event = ATOMIC_INIT(0); | |
96 | ||
8d69aaee | 97 | static inline unsigned char swap_count(unsigned char ent) |
355cfa73 | 98 | { |
570a335b | 99 | return ent & ~SWAP_HAS_CACHE; /* may include SWAP_HAS_CONT flag */ |
355cfa73 KH |
100 | } |
101 | ||
efa90a98 | 102 | /* returns 1 if swap entry is freed */ |
c9e44410 KH |
103 | static int |
104 | __try_to_reclaim_swap(struct swap_info_struct *si, unsigned long offset) | |
105 | { | |
efa90a98 | 106 | swp_entry_t entry = swp_entry(si->type, offset); |
c9e44410 KH |
107 | struct page *page; |
108 | int ret = 0; | |
109 | ||
f6ab1f7f | 110 | page = find_get_page(swap_address_space(entry), swp_offset(entry)); |
c9e44410 KH |
111 | if (!page) |
112 | return 0; | |
113 | /* | |
114 | * This function is called from scan_swap_map() and it's called | |
115 | * by vmscan.c at reclaiming pages. So, we hold a lock on a page, here. | |
116 | * We have to use trylock for avoiding deadlock. This is a special | |
117 | * case and you should use try_to_free_swap() with explicit lock_page() | |
118 | * in usual operations. | |
119 | */ | |
120 | if (trylock_page(page)) { | |
121 | ret = try_to_free_swap(page); | |
122 | unlock_page(page); | |
123 | } | |
09cbfeaf | 124 | put_page(page); |
c9e44410 KH |
125 | return ret; |
126 | } | |
355cfa73 | 127 | |
6a6ba831 HD |
128 | /* |
129 | * swapon tell device that all the old swap contents can be discarded, | |
130 | * to allow the swap device to optimize its wear-levelling. | |
131 | */ | |
132 | static int discard_swap(struct swap_info_struct *si) | |
133 | { | |
134 | struct swap_extent *se; | |
9625a5f2 HD |
135 | sector_t start_block; |
136 | sector_t nr_blocks; | |
6a6ba831 HD |
137 | int err = 0; |
138 | ||
9625a5f2 HD |
139 | /* Do not discard the swap header page! */ |
140 | se = &si->first_swap_extent; | |
141 | start_block = (se->start_block + 1) << (PAGE_SHIFT - 9); | |
142 | nr_blocks = ((sector_t)se->nr_pages - 1) << (PAGE_SHIFT - 9); | |
143 | if (nr_blocks) { | |
144 | err = blkdev_issue_discard(si->bdev, start_block, | |
dd3932ed | 145 | nr_blocks, GFP_KERNEL, 0); |
9625a5f2 HD |
146 | if (err) |
147 | return err; | |
148 | cond_resched(); | |
149 | } | |
6a6ba831 | 150 | |
9625a5f2 HD |
151 | list_for_each_entry(se, &si->first_swap_extent.list, list) { |
152 | start_block = se->start_block << (PAGE_SHIFT - 9); | |
153 | nr_blocks = (sector_t)se->nr_pages << (PAGE_SHIFT - 9); | |
6a6ba831 HD |
154 | |
155 | err = blkdev_issue_discard(si->bdev, start_block, | |
dd3932ed | 156 | nr_blocks, GFP_KERNEL, 0); |
6a6ba831 HD |
157 | if (err) |
158 | break; | |
159 | ||
160 | cond_resched(); | |
161 | } | |
162 | return err; /* That will often be -EOPNOTSUPP */ | |
163 | } | |
164 | ||
7992fde7 HD |
165 | /* |
166 | * swap allocation tell device that a cluster of swap can now be discarded, | |
167 | * to allow the swap device to optimize its wear-levelling. | |
168 | */ | |
169 | static void discard_swap_cluster(struct swap_info_struct *si, | |
170 | pgoff_t start_page, pgoff_t nr_pages) | |
171 | { | |
172 | struct swap_extent *se = si->curr_swap_extent; | |
173 | int found_extent = 0; | |
174 | ||
175 | while (nr_pages) { | |
7992fde7 HD |
176 | if (se->start_page <= start_page && |
177 | start_page < se->start_page + se->nr_pages) { | |
178 | pgoff_t offset = start_page - se->start_page; | |
179 | sector_t start_block = se->start_block + offset; | |
858a2990 | 180 | sector_t nr_blocks = se->nr_pages - offset; |
7992fde7 HD |
181 | |
182 | if (nr_blocks > nr_pages) | |
183 | nr_blocks = nr_pages; | |
184 | start_page += nr_blocks; | |
185 | nr_pages -= nr_blocks; | |
186 | ||
187 | if (!found_extent++) | |
188 | si->curr_swap_extent = se; | |
189 | ||
190 | start_block <<= PAGE_SHIFT - 9; | |
191 | nr_blocks <<= PAGE_SHIFT - 9; | |
192 | if (blkdev_issue_discard(si->bdev, start_block, | |
dd3932ed | 193 | nr_blocks, GFP_NOIO, 0)) |
7992fde7 HD |
194 | break; |
195 | } | |
196 | ||
a8ae4991 | 197 | se = list_next_entry(se, list); |
7992fde7 HD |
198 | } |
199 | } | |
200 | ||
048c27fd HD |
201 | #define SWAPFILE_CLUSTER 256 |
202 | #define LATENCY_LIMIT 256 | |
203 | ||
2a8f9449 SL |
204 | static inline void cluster_set_flag(struct swap_cluster_info *info, |
205 | unsigned int flag) | |
206 | { | |
207 | info->flags = flag; | |
208 | } | |
209 | ||
210 | static inline unsigned int cluster_count(struct swap_cluster_info *info) | |
211 | { | |
212 | return info->data; | |
213 | } | |
214 | ||
215 | static inline void cluster_set_count(struct swap_cluster_info *info, | |
216 | unsigned int c) | |
217 | { | |
218 | info->data = c; | |
219 | } | |
220 | ||
221 | static inline void cluster_set_count_flag(struct swap_cluster_info *info, | |
222 | unsigned int c, unsigned int f) | |
223 | { | |
224 | info->flags = f; | |
225 | info->data = c; | |
226 | } | |
227 | ||
228 | static inline unsigned int cluster_next(struct swap_cluster_info *info) | |
229 | { | |
230 | return info->data; | |
231 | } | |
232 | ||
233 | static inline void cluster_set_next(struct swap_cluster_info *info, | |
234 | unsigned int n) | |
235 | { | |
236 | info->data = n; | |
237 | } | |
238 | ||
239 | static inline void cluster_set_next_flag(struct swap_cluster_info *info, | |
240 | unsigned int n, unsigned int f) | |
241 | { | |
242 | info->flags = f; | |
243 | info->data = n; | |
244 | } | |
245 | ||
246 | static inline bool cluster_is_free(struct swap_cluster_info *info) | |
247 | { | |
248 | return info->flags & CLUSTER_FLAG_FREE; | |
249 | } | |
250 | ||
251 | static inline bool cluster_is_null(struct swap_cluster_info *info) | |
252 | { | |
253 | return info->flags & CLUSTER_FLAG_NEXT_NULL; | |
254 | } | |
255 | ||
256 | static inline void cluster_set_null(struct swap_cluster_info *info) | |
257 | { | |
258 | info->flags = CLUSTER_FLAG_NEXT_NULL; | |
259 | info->data = 0; | |
260 | } | |
261 | ||
235b6217 HY |
262 | static inline struct swap_cluster_info *lock_cluster(struct swap_info_struct *si, |
263 | unsigned long offset) | |
264 | { | |
265 | struct swap_cluster_info *ci; | |
266 | ||
267 | ci = si->cluster_info; | |
268 | if (ci) { | |
269 | ci += offset / SWAPFILE_CLUSTER; | |
270 | spin_lock(&ci->lock); | |
271 | } | |
272 | return ci; | |
273 | } | |
274 | ||
275 | static inline void unlock_cluster(struct swap_cluster_info *ci) | |
276 | { | |
277 | if (ci) | |
278 | spin_unlock(&ci->lock); | |
279 | } | |
280 | ||
281 | static inline struct swap_cluster_info *lock_cluster_or_swap_info( | |
282 | struct swap_info_struct *si, | |
283 | unsigned long offset) | |
284 | { | |
285 | struct swap_cluster_info *ci; | |
286 | ||
287 | ci = lock_cluster(si, offset); | |
288 | if (!ci) | |
289 | spin_lock(&si->lock); | |
290 | ||
291 | return ci; | |
292 | } | |
293 | ||
294 | static inline void unlock_cluster_or_swap_info(struct swap_info_struct *si, | |
295 | struct swap_cluster_info *ci) | |
296 | { | |
297 | if (ci) | |
298 | unlock_cluster(ci); | |
299 | else | |
300 | spin_unlock(&si->lock); | |
301 | } | |
302 | ||
6b534915 HY |
303 | static inline bool cluster_list_empty(struct swap_cluster_list *list) |
304 | { | |
305 | return cluster_is_null(&list->head); | |
306 | } | |
307 | ||
308 | static inline unsigned int cluster_list_first(struct swap_cluster_list *list) | |
309 | { | |
310 | return cluster_next(&list->head); | |
311 | } | |
312 | ||
313 | static void cluster_list_init(struct swap_cluster_list *list) | |
314 | { | |
315 | cluster_set_null(&list->head); | |
316 | cluster_set_null(&list->tail); | |
317 | } | |
318 | ||
319 | static void cluster_list_add_tail(struct swap_cluster_list *list, | |
320 | struct swap_cluster_info *ci, | |
321 | unsigned int idx) | |
322 | { | |
323 | if (cluster_list_empty(list)) { | |
324 | cluster_set_next_flag(&list->head, idx, 0); | |
325 | cluster_set_next_flag(&list->tail, idx, 0); | |
326 | } else { | |
235b6217 | 327 | struct swap_cluster_info *ci_tail; |
6b534915 HY |
328 | unsigned int tail = cluster_next(&list->tail); |
329 | ||
235b6217 HY |
330 | /* |
331 | * Nested cluster lock, but both cluster locks are | |
332 | * only acquired when we held swap_info_struct->lock | |
333 | */ | |
334 | ci_tail = ci + tail; | |
335 | spin_lock_nested(&ci_tail->lock, SINGLE_DEPTH_NESTING); | |
336 | cluster_set_next(ci_tail, idx); | |
337 | unlock_cluster(ci_tail); | |
6b534915 HY |
338 | cluster_set_next_flag(&list->tail, idx, 0); |
339 | } | |
340 | } | |
341 | ||
342 | static unsigned int cluster_list_del_first(struct swap_cluster_list *list, | |
343 | struct swap_cluster_info *ci) | |
344 | { | |
345 | unsigned int idx; | |
346 | ||
347 | idx = cluster_next(&list->head); | |
348 | if (cluster_next(&list->tail) == idx) { | |
349 | cluster_set_null(&list->head); | |
350 | cluster_set_null(&list->tail); | |
351 | } else | |
352 | cluster_set_next_flag(&list->head, | |
353 | cluster_next(&ci[idx]), 0); | |
354 | ||
355 | return idx; | |
356 | } | |
357 | ||
815c2c54 SL |
358 | /* Add a cluster to discard list and schedule it to do discard */ |
359 | static void swap_cluster_schedule_discard(struct swap_info_struct *si, | |
360 | unsigned int idx) | |
361 | { | |
362 | /* | |
363 | * If scan_swap_map() can't find a free cluster, it will check | |
364 | * si->swap_map directly. To make sure the discarding cluster isn't | |
365 | * taken by scan_swap_map(), mark the swap entries bad (occupied). It | |
366 | * will be cleared after discard | |
367 | */ | |
368 | memset(si->swap_map + idx * SWAPFILE_CLUSTER, | |
369 | SWAP_MAP_BAD, SWAPFILE_CLUSTER); | |
370 | ||
6b534915 | 371 | cluster_list_add_tail(&si->discard_clusters, si->cluster_info, idx); |
815c2c54 SL |
372 | |
373 | schedule_work(&si->discard_work); | |
374 | } | |
375 | ||
376 | /* | |
377 | * Doing discard actually. After a cluster discard is finished, the cluster | |
378 | * will be added to free cluster list. caller should hold si->lock. | |
379 | */ | |
380 | static void swap_do_scheduled_discard(struct swap_info_struct *si) | |
381 | { | |
235b6217 | 382 | struct swap_cluster_info *info, *ci; |
815c2c54 SL |
383 | unsigned int idx; |
384 | ||
385 | info = si->cluster_info; | |
386 | ||
6b534915 HY |
387 | while (!cluster_list_empty(&si->discard_clusters)) { |
388 | idx = cluster_list_del_first(&si->discard_clusters, info); | |
815c2c54 SL |
389 | spin_unlock(&si->lock); |
390 | ||
391 | discard_swap_cluster(si, idx * SWAPFILE_CLUSTER, | |
392 | SWAPFILE_CLUSTER); | |
393 | ||
394 | spin_lock(&si->lock); | |
235b6217 HY |
395 | ci = lock_cluster(si, idx * SWAPFILE_CLUSTER); |
396 | cluster_set_flag(ci, CLUSTER_FLAG_FREE); | |
397 | unlock_cluster(ci); | |
6b534915 | 398 | cluster_list_add_tail(&si->free_clusters, info, idx); |
235b6217 | 399 | ci = lock_cluster(si, idx * SWAPFILE_CLUSTER); |
815c2c54 SL |
400 | memset(si->swap_map + idx * SWAPFILE_CLUSTER, |
401 | 0, SWAPFILE_CLUSTER); | |
235b6217 | 402 | unlock_cluster(ci); |
815c2c54 SL |
403 | } |
404 | } | |
405 | ||
406 | static void swap_discard_work(struct work_struct *work) | |
407 | { | |
408 | struct swap_info_struct *si; | |
409 | ||
410 | si = container_of(work, struct swap_info_struct, discard_work); | |
411 | ||
412 | spin_lock(&si->lock); | |
413 | swap_do_scheduled_discard(si); | |
414 | spin_unlock(&si->lock); | |
415 | } | |
416 | ||
2a8f9449 SL |
417 | /* |
418 | * The cluster corresponding to page_nr will be used. The cluster will be | |
419 | * removed from free cluster list and its usage counter will be increased. | |
420 | */ | |
421 | static void inc_cluster_info_page(struct swap_info_struct *p, | |
422 | struct swap_cluster_info *cluster_info, unsigned long page_nr) | |
423 | { | |
424 | unsigned long idx = page_nr / SWAPFILE_CLUSTER; | |
425 | ||
426 | if (!cluster_info) | |
427 | return; | |
428 | if (cluster_is_free(&cluster_info[idx])) { | |
6b534915 HY |
429 | VM_BUG_ON(cluster_list_first(&p->free_clusters) != idx); |
430 | cluster_list_del_first(&p->free_clusters, cluster_info); | |
2a8f9449 SL |
431 | cluster_set_count_flag(&cluster_info[idx], 0, 0); |
432 | } | |
433 | ||
434 | VM_BUG_ON(cluster_count(&cluster_info[idx]) >= SWAPFILE_CLUSTER); | |
435 | cluster_set_count(&cluster_info[idx], | |
436 | cluster_count(&cluster_info[idx]) + 1); | |
437 | } | |
438 | ||
439 | /* | |
440 | * The cluster corresponding to page_nr decreases one usage. If the usage | |
441 | * counter becomes 0, which means no page in the cluster is in using, we can | |
442 | * optionally discard the cluster and add it to free cluster list. | |
443 | */ | |
444 | static void dec_cluster_info_page(struct swap_info_struct *p, | |
445 | struct swap_cluster_info *cluster_info, unsigned long page_nr) | |
446 | { | |
447 | unsigned long idx = page_nr / SWAPFILE_CLUSTER; | |
448 | ||
449 | if (!cluster_info) | |
450 | return; | |
451 | ||
452 | VM_BUG_ON(cluster_count(&cluster_info[idx]) == 0); | |
453 | cluster_set_count(&cluster_info[idx], | |
454 | cluster_count(&cluster_info[idx]) - 1); | |
455 | ||
456 | if (cluster_count(&cluster_info[idx]) == 0) { | |
815c2c54 SL |
457 | /* |
458 | * If the swap is discardable, prepare discard the cluster | |
459 | * instead of free it immediately. The cluster will be freed | |
460 | * after discard. | |
461 | */ | |
edfe23da SL |
462 | if ((p->flags & (SWP_WRITEOK | SWP_PAGE_DISCARD)) == |
463 | (SWP_WRITEOK | SWP_PAGE_DISCARD)) { | |
815c2c54 SL |
464 | swap_cluster_schedule_discard(p, idx); |
465 | return; | |
466 | } | |
467 | ||
2a8f9449 | 468 | cluster_set_flag(&cluster_info[idx], CLUSTER_FLAG_FREE); |
6b534915 | 469 | cluster_list_add_tail(&p->free_clusters, cluster_info, idx); |
2a8f9449 SL |
470 | } |
471 | } | |
472 | ||
473 | /* | |
474 | * It's possible scan_swap_map() uses a free cluster in the middle of free | |
475 | * cluster list. Avoiding such abuse to avoid list corruption. | |
476 | */ | |
ebc2a1a6 SL |
477 | static bool |
478 | scan_swap_map_ssd_cluster_conflict(struct swap_info_struct *si, | |
2a8f9449 SL |
479 | unsigned long offset) |
480 | { | |
ebc2a1a6 SL |
481 | struct percpu_cluster *percpu_cluster; |
482 | bool conflict; | |
483 | ||
2a8f9449 | 484 | offset /= SWAPFILE_CLUSTER; |
6b534915 HY |
485 | conflict = !cluster_list_empty(&si->free_clusters) && |
486 | offset != cluster_list_first(&si->free_clusters) && | |
2a8f9449 | 487 | cluster_is_free(&si->cluster_info[offset]); |
ebc2a1a6 SL |
488 | |
489 | if (!conflict) | |
490 | return false; | |
491 | ||
492 | percpu_cluster = this_cpu_ptr(si->percpu_cluster); | |
493 | cluster_set_null(&percpu_cluster->index); | |
494 | return true; | |
495 | } | |
496 | ||
497 | /* | |
498 | * Try to get a swap entry from current cpu's swap entry pool (a cluster). This | |
499 | * might involve allocating a new cluster for current CPU too. | |
500 | */ | |
36005bae | 501 | static bool scan_swap_map_try_ssd_cluster(struct swap_info_struct *si, |
ebc2a1a6 SL |
502 | unsigned long *offset, unsigned long *scan_base) |
503 | { | |
504 | struct percpu_cluster *cluster; | |
235b6217 | 505 | struct swap_cluster_info *ci; |
ebc2a1a6 | 506 | bool found_free; |
235b6217 | 507 | unsigned long tmp, max; |
ebc2a1a6 SL |
508 | |
509 | new_cluster: | |
510 | cluster = this_cpu_ptr(si->percpu_cluster); | |
511 | if (cluster_is_null(&cluster->index)) { | |
6b534915 HY |
512 | if (!cluster_list_empty(&si->free_clusters)) { |
513 | cluster->index = si->free_clusters.head; | |
ebc2a1a6 SL |
514 | cluster->next = cluster_next(&cluster->index) * |
515 | SWAPFILE_CLUSTER; | |
6b534915 | 516 | } else if (!cluster_list_empty(&si->discard_clusters)) { |
ebc2a1a6 SL |
517 | /* |
518 | * we don't have free cluster but have some clusters in | |
519 | * discarding, do discard now and reclaim them | |
520 | */ | |
521 | swap_do_scheduled_discard(si); | |
522 | *scan_base = *offset = si->cluster_next; | |
523 | goto new_cluster; | |
524 | } else | |
36005bae | 525 | return false; |
ebc2a1a6 SL |
526 | } |
527 | ||
528 | found_free = false; | |
529 | ||
530 | /* | |
531 | * Other CPUs can use our cluster if they can't find a free cluster, | |
532 | * check if there is still free entry in the cluster | |
533 | */ | |
534 | tmp = cluster->next; | |
235b6217 HY |
535 | max = min_t(unsigned long, si->max, |
536 | (cluster_next(&cluster->index) + 1) * SWAPFILE_CLUSTER); | |
537 | if (tmp >= max) { | |
538 | cluster_set_null(&cluster->index); | |
539 | goto new_cluster; | |
540 | } | |
541 | ci = lock_cluster(si, tmp); | |
542 | while (tmp < max) { | |
ebc2a1a6 SL |
543 | if (!si->swap_map[tmp]) { |
544 | found_free = true; | |
545 | break; | |
546 | } | |
547 | tmp++; | |
548 | } | |
235b6217 | 549 | unlock_cluster(ci); |
ebc2a1a6 SL |
550 | if (!found_free) { |
551 | cluster_set_null(&cluster->index); | |
552 | goto new_cluster; | |
553 | } | |
554 | cluster->next = tmp + 1; | |
555 | *offset = tmp; | |
556 | *scan_base = tmp; | |
36005bae | 557 | return found_free; |
2a8f9449 SL |
558 | } |
559 | ||
36005bae TC |
560 | static int scan_swap_map_slots(struct swap_info_struct *si, |
561 | unsigned char usage, int nr, | |
562 | swp_entry_t slots[]) | |
1da177e4 | 563 | { |
235b6217 | 564 | struct swap_cluster_info *ci; |
ebebbbe9 | 565 | unsigned long offset; |
c60aa176 | 566 | unsigned long scan_base; |
7992fde7 | 567 | unsigned long last_in_cluster = 0; |
048c27fd | 568 | int latency_ration = LATENCY_LIMIT; |
36005bae TC |
569 | int n_ret = 0; |
570 | ||
571 | if (nr > SWAP_BATCH) | |
572 | nr = SWAP_BATCH; | |
7dfad418 | 573 | |
886bb7e9 | 574 | /* |
7dfad418 HD |
575 | * We try to cluster swap pages by allocating them sequentially |
576 | * in swap. Once we've allocated SWAPFILE_CLUSTER pages this | |
577 | * way, however, we resort to first-free allocation, starting | |
578 | * a new cluster. This prevents us from scattering swap pages | |
579 | * all over the entire swap partition, so that we reduce | |
580 | * overall disk seek times between swap pages. -- sct | |
581 | * But we do now try to find an empty cluster. -Andrea | |
c60aa176 | 582 | * And we let swap pages go all over an SSD partition. Hugh |
7dfad418 HD |
583 | */ |
584 | ||
52b7efdb | 585 | si->flags += SWP_SCANNING; |
c60aa176 | 586 | scan_base = offset = si->cluster_next; |
ebebbbe9 | 587 | |
ebc2a1a6 SL |
588 | /* SSD algorithm */ |
589 | if (si->cluster_info) { | |
36005bae TC |
590 | if (scan_swap_map_try_ssd_cluster(si, &offset, &scan_base)) |
591 | goto checks; | |
592 | else | |
593 | goto scan; | |
ebc2a1a6 SL |
594 | } |
595 | ||
ebebbbe9 HD |
596 | if (unlikely(!si->cluster_nr--)) { |
597 | if (si->pages - si->inuse_pages < SWAPFILE_CLUSTER) { | |
598 | si->cluster_nr = SWAPFILE_CLUSTER - 1; | |
599 | goto checks; | |
600 | } | |
2a8f9449 | 601 | |
ec8acf20 | 602 | spin_unlock(&si->lock); |
7dfad418 | 603 | |
c60aa176 HD |
604 | /* |
605 | * If seek is expensive, start searching for new cluster from | |
606 | * start of partition, to minimize the span of allocated swap. | |
50088c44 CY |
607 | * If seek is cheap, that is the SWP_SOLIDSTATE si->cluster_info |
608 | * case, just handled by scan_swap_map_try_ssd_cluster() above. | |
c60aa176 | 609 | */ |
50088c44 | 610 | scan_base = offset = si->lowest_bit; |
7dfad418 HD |
611 | last_in_cluster = offset + SWAPFILE_CLUSTER - 1; |
612 | ||
613 | /* Locate the first empty (unaligned) cluster */ | |
614 | for (; last_in_cluster <= si->highest_bit; offset++) { | |
1da177e4 | 615 | if (si->swap_map[offset]) |
7dfad418 HD |
616 | last_in_cluster = offset + SWAPFILE_CLUSTER; |
617 | else if (offset == last_in_cluster) { | |
ec8acf20 | 618 | spin_lock(&si->lock); |
ebebbbe9 HD |
619 | offset -= SWAPFILE_CLUSTER - 1; |
620 | si->cluster_next = offset; | |
621 | si->cluster_nr = SWAPFILE_CLUSTER - 1; | |
c60aa176 HD |
622 | goto checks; |
623 | } | |
624 | if (unlikely(--latency_ration < 0)) { | |
625 | cond_resched(); | |
626 | latency_ration = LATENCY_LIMIT; | |
627 | } | |
628 | } | |
629 | ||
630 | offset = scan_base; | |
ec8acf20 | 631 | spin_lock(&si->lock); |
ebebbbe9 | 632 | si->cluster_nr = SWAPFILE_CLUSTER - 1; |
1da177e4 | 633 | } |
7dfad418 | 634 | |
ebebbbe9 | 635 | checks: |
ebc2a1a6 | 636 | if (si->cluster_info) { |
36005bae TC |
637 | while (scan_swap_map_ssd_cluster_conflict(si, offset)) { |
638 | /* take a break if we already got some slots */ | |
639 | if (n_ret) | |
640 | goto done; | |
641 | if (!scan_swap_map_try_ssd_cluster(si, &offset, | |
642 | &scan_base)) | |
643 | goto scan; | |
644 | } | |
ebc2a1a6 | 645 | } |
ebebbbe9 | 646 | if (!(si->flags & SWP_WRITEOK)) |
52b7efdb | 647 | goto no_page; |
7dfad418 HD |
648 | if (!si->highest_bit) |
649 | goto no_page; | |
ebebbbe9 | 650 | if (offset > si->highest_bit) |
c60aa176 | 651 | scan_base = offset = si->lowest_bit; |
c9e44410 | 652 | |
235b6217 | 653 | ci = lock_cluster(si, offset); |
b73d7fce HD |
654 | /* reuse swap entry of cache-only swap if not busy. */ |
655 | if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) { | |
c9e44410 | 656 | int swap_was_freed; |
235b6217 | 657 | unlock_cluster(ci); |
ec8acf20 | 658 | spin_unlock(&si->lock); |
c9e44410 | 659 | swap_was_freed = __try_to_reclaim_swap(si, offset); |
ec8acf20 | 660 | spin_lock(&si->lock); |
c9e44410 KH |
661 | /* entry was freed successfully, try to use this again */ |
662 | if (swap_was_freed) | |
663 | goto checks; | |
664 | goto scan; /* check next one */ | |
665 | } | |
666 | ||
235b6217 HY |
667 | if (si->swap_map[offset]) { |
668 | unlock_cluster(ci); | |
36005bae TC |
669 | if (!n_ret) |
670 | goto scan; | |
671 | else | |
672 | goto done; | |
235b6217 | 673 | } |
ebebbbe9 HD |
674 | |
675 | if (offset == si->lowest_bit) | |
676 | si->lowest_bit++; | |
677 | if (offset == si->highest_bit) | |
678 | si->highest_bit--; | |
679 | si->inuse_pages++; | |
680 | if (si->inuse_pages == si->pages) { | |
681 | si->lowest_bit = si->max; | |
682 | si->highest_bit = 0; | |
18ab4d4c DS |
683 | spin_lock(&swap_avail_lock); |
684 | plist_del(&si->avail_list, &swap_avail_head); | |
685 | spin_unlock(&swap_avail_lock); | |
1da177e4 | 686 | } |
253d553b | 687 | si->swap_map[offset] = usage; |
2a8f9449 | 688 | inc_cluster_info_page(si, si->cluster_info, offset); |
235b6217 | 689 | unlock_cluster(ci); |
ebebbbe9 | 690 | si->cluster_next = offset + 1; |
36005bae TC |
691 | slots[n_ret++] = swp_entry(si->type, offset); |
692 | ||
693 | /* got enough slots or reach max slots? */ | |
694 | if ((n_ret == nr) || (offset >= si->highest_bit)) | |
695 | goto done; | |
696 | ||
697 | /* search for next available slot */ | |
698 | ||
699 | /* time to take a break? */ | |
700 | if (unlikely(--latency_ration < 0)) { | |
701 | if (n_ret) | |
702 | goto done; | |
703 | spin_unlock(&si->lock); | |
704 | cond_resched(); | |
705 | spin_lock(&si->lock); | |
706 | latency_ration = LATENCY_LIMIT; | |
707 | } | |
708 | ||
709 | /* try to get more slots in cluster */ | |
710 | if (si->cluster_info) { | |
711 | if (scan_swap_map_try_ssd_cluster(si, &offset, &scan_base)) | |
712 | goto checks; | |
713 | else | |
714 | goto done; | |
715 | } | |
716 | /* non-ssd case */ | |
717 | ++offset; | |
718 | ||
719 | /* non-ssd case, still more slots in cluster? */ | |
720 | if (si->cluster_nr && !si->swap_map[offset]) { | |
721 | --si->cluster_nr; | |
722 | goto checks; | |
723 | } | |
7992fde7 | 724 | |
36005bae TC |
725 | done: |
726 | si->flags -= SWP_SCANNING; | |
727 | return n_ret; | |
7dfad418 | 728 | |
ebebbbe9 | 729 | scan: |
ec8acf20 | 730 | spin_unlock(&si->lock); |
7dfad418 | 731 | while (++offset <= si->highest_bit) { |
52b7efdb | 732 | if (!si->swap_map[offset]) { |
ec8acf20 | 733 | spin_lock(&si->lock); |
52b7efdb HD |
734 | goto checks; |
735 | } | |
c9e44410 | 736 | if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) { |
ec8acf20 | 737 | spin_lock(&si->lock); |
c9e44410 KH |
738 | goto checks; |
739 | } | |
048c27fd HD |
740 | if (unlikely(--latency_ration < 0)) { |
741 | cond_resched(); | |
742 | latency_ration = LATENCY_LIMIT; | |
743 | } | |
7dfad418 | 744 | } |
c60aa176 | 745 | offset = si->lowest_bit; |
a5998061 | 746 | while (offset < scan_base) { |
c60aa176 | 747 | if (!si->swap_map[offset]) { |
ec8acf20 | 748 | spin_lock(&si->lock); |
c60aa176 HD |
749 | goto checks; |
750 | } | |
c9e44410 | 751 | if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) { |
ec8acf20 | 752 | spin_lock(&si->lock); |
c9e44410 KH |
753 | goto checks; |
754 | } | |
c60aa176 HD |
755 | if (unlikely(--latency_ration < 0)) { |
756 | cond_resched(); | |
757 | latency_ration = LATENCY_LIMIT; | |
758 | } | |
a5998061 | 759 | offset++; |
c60aa176 | 760 | } |
ec8acf20 | 761 | spin_lock(&si->lock); |
7dfad418 HD |
762 | |
763 | no_page: | |
52b7efdb | 764 | si->flags -= SWP_SCANNING; |
36005bae | 765 | return n_ret; |
1da177e4 LT |
766 | } |
767 | ||
36005bae TC |
768 | static unsigned long scan_swap_map(struct swap_info_struct *si, |
769 | unsigned char usage) | |
770 | { | |
771 | swp_entry_t entry; | |
772 | int n_ret; | |
773 | ||
774 | n_ret = scan_swap_map_slots(si, usage, 1, &entry); | |
775 | ||
776 | if (n_ret) | |
777 | return swp_offset(entry); | |
778 | else | |
779 | return 0; | |
780 | ||
781 | } | |
782 | ||
783 | int get_swap_pages(int n_goal, swp_entry_t swp_entries[]) | |
1da177e4 | 784 | { |
adfab836 | 785 | struct swap_info_struct *si, *next; |
36005bae TC |
786 | long avail_pgs; |
787 | int n_ret = 0; | |
1da177e4 | 788 | |
36005bae TC |
789 | avail_pgs = atomic_long_read(&nr_swap_pages); |
790 | if (avail_pgs <= 0) | |
fb4f88dc | 791 | goto noswap; |
36005bae TC |
792 | |
793 | if (n_goal > SWAP_BATCH) | |
794 | n_goal = SWAP_BATCH; | |
795 | ||
796 | if (n_goal > avail_pgs) | |
797 | n_goal = avail_pgs; | |
798 | ||
799 | atomic_long_sub(n_goal, &nr_swap_pages); | |
fb4f88dc | 800 | |
18ab4d4c DS |
801 | spin_lock(&swap_avail_lock); |
802 | ||
803 | start_over: | |
804 | plist_for_each_entry_safe(si, next, &swap_avail_head, avail_list) { | |
805 | /* requeue si to after same-priority siblings */ | |
806 | plist_requeue(&si->avail_list, &swap_avail_head); | |
807 | spin_unlock(&swap_avail_lock); | |
ec8acf20 | 808 | spin_lock(&si->lock); |
adfab836 | 809 | if (!si->highest_bit || !(si->flags & SWP_WRITEOK)) { |
18ab4d4c DS |
810 | spin_lock(&swap_avail_lock); |
811 | if (plist_node_empty(&si->avail_list)) { | |
812 | spin_unlock(&si->lock); | |
813 | goto nextsi; | |
814 | } | |
815 | WARN(!si->highest_bit, | |
816 | "swap_info %d in list but !highest_bit\n", | |
817 | si->type); | |
818 | WARN(!(si->flags & SWP_WRITEOK), | |
819 | "swap_info %d in list but !SWP_WRITEOK\n", | |
820 | si->type); | |
821 | plist_del(&si->avail_list, &swap_avail_head); | |
ec8acf20 | 822 | spin_unlock(&si->lock); |
18ab4d4c | 823 | goto nextsi; |
ec8acf20 | 824 | } |
36005bae TC |
825 | n_ret = scan_swap_map_slots(si, SWAP_HAS_CACHE, |
826 | n_goal, swp_entries); | |
ec8acf20 | 827 | spin_unlock(&si->lock); |
36005bae TC |
828 | if (n_ret) |
829 | goto check_out; | |
18ab4d4c | 830 | pr_debug("scan_swap_map of si %d failed to find offset\n", |
36005bae TC |
831 | si->type); |
832 | ||
18ab4d4c DS |
833 | spin_lock(&swap_avail_lock); |
834 | nextsi: | |
adfab836 DS |
835 | /* |
836 | * if we got here, it's likely that si was almost full before, | |
837 | * and since scan_swap_map() can drop the si->lock, multiple | |
838 | * callers probably all tried to get a page from the same si | |
18ab4d4c DS |
839 | * and it filled up before we could get one; or, the si filled |
840 | * up between us dropping swap_avail_lock and taking si->lock. | |
841 | * Since we dropped the swap_avail_lock, the swap_avail_head | |
842 | * list may have been modified; so if next is still in the | |
36005bae TC |
843 | * swap_avail_head list then try it, otherwise start over |
844 | * if we have not gotten any slots. | |
adfab836 | 845 | */ |
18ab4d4c DS |
846 | if (plist_node_empty(&next->avail_list)) |
847 | goto start_over; | |
1da177e4 | 848 | } |
fb4f88dc | 849 | |
18ab4d4c DS |
850 | spin_unlock(&swap_avail_lock); |
851 | ||
36005bae TC |
852 | check_out: |
853 | if (n_ret < n_goal) | |
854 | atomic_long_add((long) (n_goal-n_ret), &nr_swap_pages); | |
fb4f88dc | 855 | noswap: |
36005bae TC |
856 | return n_ret; |
857 | } | |
858 | ||
2de1a7e4 | 859 | /* The only caller of this function is now suspend routine */ |
910321ea HD |
860 | swp_entry_t get_swap_page_of_type(int type) |
861 | { | |
862 | struct swap_info_struct *si; | |
863 | pgoff_t offset; | |
864 | ||
910321ea | 865 | si = swap_info[type]; |
ec8acf20 | 866 | spin_lock(&si->lock); |
910321ea | 867 | if (si && (si->flags & SWP_WRITEOK)) { |
ec8acf20 | 868 | atomic_long_dec(&nr_swap_pages); |
910321ea HD |
869 | /* This is called for allocating swap entry, not cache */ |
870 | offset = scan_swap_map(si, 1); | |
871 | if (offset) { | |
ec8acf20 | 872 | spin_unlock(&si->lock); |
910321ea HD |
873 | return swp_entry(type, offset); |
874 | } | |
ec8acf20 | 875 | atomic_long_inc(&nr_swap_pages); |
910321ea | 876 | } |
ec8acf20 | 877 | spin_unlock(&si->lock); |
910321ea HD |
878 | return (swp_entry_t) {0}; |
879 | } | |
880 | ||
e8c26ab6 | 881 | static struct swap_info_struct *__swap_info_get(swp_entry_t entry) |
1da177e4 | 882 | { |
73c34b6a | 883 | struct swap_info_struct *p; |
1da177e4 LT |
884 | unsigned long offset, type; |
885 | ||
886 | if (!entry.val) | |
887 | goto out; | |
888 | type = swp_type(entry); | |
889 | if (type >= nr_swapfiles) | |
890 | goto bad_nofile; | |
efa90a98 | 891 | p = swap_info[type]; |
1da177e4 LT |
892 | if (!(p->flags & SWP_USED)) |
893 | goto bad_device; | |
894 | offset = swp_offset(entry); | |
895 | if (offset >= p->max) | |
896 | goto bad_offset; | |
1da177e4 LT |
897 | return p; |
898 | ||
1da177e4 | 899 | bad_offset: |
6a991fc7 | 900 | pr_err("swap_info_get: %s%08lx\n", Bad_offset, entry.val); |
1da177e4 LT |
901 | goto out; |
902 | bad_device: | |
6a991fc7 | 903 | pr_err("swap_info_get: %s%08lx\n", Unused_file, entry.val); |
1da177e4 LT |
904 | goto out; |
905 | bad_nofile: | |
6a991fc7 | 906 | pr_err("swap_info_get: %s%08lx\n", Bad_file, entry.val); |
1da177e4 LT |
907 | out: |
908 | return NULL; | |
886bb7e9 | 909 | } |
1da177e4 | 910 | |
e8c26ab6 TC |
911 | static struct swap_info_struct *_swap_info_get(swp_entry_t entry) |
912 | { | |
913 | struct swap_info_struct *p; | |
914 | ||
915 | p = __swap_info_get(entry); | |
916 | if (!p) | |
917 | goto out; | |
918 | if (!p->swap_map[swp_offset(entry)]) | |
919 | goto bad_free; | |
920 | return p; | |
921 | ||
922 | bad_free: | |
923 | pr_err("swap_info_get: %s%08lx\n", Unused_offset, entry.val); | |
924 | goto out; | |
925 | out: | |
926 | return NULL; | |
927 | } | |
928 | ||
235b6217 HY |
929 | static struct swap_info_struct *swap_info_get(swp_entry_t entry) |
930 | { | |
931 | struct swap_info_struct *p; | |
932 | ||
933 | p = _swap_info_get(entry); | |
934 | if (p) | |
935 | spin_lock(&p->lock); | |
936 | return p; | |
937 | } | |
938 | ||
7c00bafe TC |
939 | static struct swap_info_struct *swap_info_get_cont(swp_entry_t entry, |
940 | struct swap_info_struct *q) | |
941 | { | |
942 | struct swap_info_struct *p; | |
943 | ||
944 | p = _swap_info_get(entry); | |
945 | ||
946 | if (p != q) { | |
947 | if (q != NULL) | |
948 | spin_unlock(&q->lock); | |
949 | if (p != NULL) | |
950 | spin_lock(&p->lock); | |
951 | } | |
952 | return p; | |
953 | } | |
954 | ||
955 | static unsigned char __swap_entry_free(struct swap_info_struct *p, | |
956 | swp_entry_t entry, unsigned char usage) | |
1da177e4 | 957 | { |
235b6217 | 958 | struct swap_cluster_info *ci; |
253d553b | 959 | unsigned long offset = swp_offset(entry); |
8d69aaee HD |
960 | unsigned char count; |
961 | unsigned char has_cache; | |
235b6217 | 962 | |
7c00bafe | 963 | ci = lock_cluster_or_swap_info(p, offset); |
355cfa73 | 964 | |
253d553b | 965 | count = p->swap_map[offset]; |
235b6217 | 966 | |
253d553b HD |
967 | has_cache = count & SWAP_HAS_CACHE; |
968 | count &= ~SWAP_HAS_CACHE; | |
355cfa73 | 969 | |
253d553b | 970 | if (usage == SWAP_HAS_CACHE) { |
355cfa73 | 971 | VM_BUG_ON(!has_cache); |
253d553b | 972 | has_cache = 0; |
aaa46865 HD |
973 | } else if (count == SWAP_MAP_SHMEM) { |
974 | /* | |
975 | * Or we could insist on shmem.c using a special | |
976 | * swap_shmem_free() and free_shmem_swap_and_cache()... | |
977 | */ | |
978 | count = 0; | |
570a335b HD |
979 | } else if ((count & ~COUNT_CONTINUED) <= SWAP_MAP_MAX) { |
980 | if (count == COUNT_CONTINUED) { | |
981 | if (swap_count_continued(p, offset, count)) | |
982 | count = SWAP_MAP_MAX | COUNT_CONTINUED; | |
983 | else | |
984 | count = SWAP_MAP_MAX; | |
985 | } else | |
986 | count--; | |
987 | } | |
253d553b | 988 | |
253d553b | 989 | usage = count | has_cache; |
7c00bafe TC |
990 | p->swap_map[offset] = usage ? : SWAP_HAS_CACHE; |
991 | ||
992 | unlock_cluster_or_swap_info(p, ci); | |
993 | ||
994 | return usage; | |
995 | } | |
355cfa73 | 996 | |
7c00bafe TC |
997 | static void swap_entry_free(struct swap_info_struct *p, swp_entry_t entry) |
998 | { | |
999 | struct swap_cluster_info *ci; | |
1000 | unsigned long offset = swp_offset(entry); | |
1001 | unsigned char count; | |
1002 | ||
1003 | ci = lock_cluster(p, offset); | |
1004 | count = p->swap_map[offset]; | |
1005 | VM_BUG_ON(count != SWAP_HAS_CACHE); | |
1006 | p->swap_map[offset] = 0; | |
1007 | dec_cluster_info_page(p, p->cluster_info, offset); | |
235b6217 HY |
1008 | unlock_cluster(ci); |
1009 | ||
7c00bafe TC |
1010 | mem_cgroup_uncharge_swap(entry); |
1011 | if (offset < p->lowest_bit) | |
1012 | p->lowest_bit = offset; | |
1013 | if (offset > p->highest_bit) { | |
1014 | bool was_full = !p->highest_bit; | |
1015 | ||
1016 | p->highest_bit = offset; | |
1017 | if (was_full && (p->flags & SWP_WRITEOK)) { | |
1018 | spin_lock(&swap_avail_lock); | |
1019 | WARN_ON(!plist_node_empty(&p->avail_list)); | |
1020 | if (plist_node_empty(&p->avail_list)) | |
1021 | plist_add(&p->avail_list, | |
1022 | &swap_avail_head); | |
1023 | spin_unlock(&swap_avail_lock); | |
73744923 | 1024 | } |
1da177e4 | 1025 | } |
7c00bafe TC |
1026 | atomic_long_inc(&nr_swap_pages); |
1027 | p->inuse_pages--; | |
1028 | frontswap_invalidate_page(p->type, offset); | |
1029 | if (p->flags & SWP_BLKDEV) { | |
1030 | struct gendisk *disk = p->bdev->bd_disk; | |
1031 | ||
1032 | if (disk->fops->swap_slot_free_notify) | |
1033 | disk->fops->swap_slot_free_notify(p->bdev, | |
1034 | offset); | |
1035 | } | |
1da177e4 LT |
1036 | } |
1037 | ||
1038 | /* | |
2de1a7e4 | 1039 | * Caller has made sure that the swap device corresponding to entry |
1da177e4 LT |
1040 | * is still around or has not been recycled. |
1041 | */ | |
1042 | void swap_free(swp_entry_t entry) | |
1043 | { | |
73c34b6a | 1044 | struct swap_info_struct *p; |
1da177e4 | 1045 | |
235b6217 | 1046 | p = _swap_info_get(entry); |
7c00bafe TC |
1047 | if (p) { |
1048 | if (!__swap_entry_free(p, entry, 1)) | |
67afa38e | 1049 | free_swap_slot(entry); |
7c00bafe | 1050 | } |
1da177e4 LT |
1051 | } |
1052 | ||
cb4b86ba KH |
1053 | /* |
1054 | * Called after dropping swapcache to decrease refcnt to swap entries. | |
1055 | */ | |
0a31bc97 | 1056 | void swapcache_free(swp_entry_t entry) |
cb4b86ba | 1057 | { |
355cfa73 KH |
1058 | struct swap_info_struct *p; |
1059 | ||
235b6217 | 1060 | p = _swap_info_get(entry); |
7c00bafe TC |
1061 | if (p) { |
1062 | if (!__swap_entry_free(p, entry, SWAP_HAS_CACHE)) | |
67afa38e | 1063 | free_swap_slot(entry); |
7c00bafe TC |
1064 | } |
1065 | } | |
1066 | ||
1067 | void swapcache_free_entries(swp_entry_t *entries, int n) | |
1068 | { | |
1069 | struct swap_info_struct *p, *prev; | |
1070 | int i; | |
1071 | ||
1072 | if (n <= 0) | |
1073 | return; | |
1074 | ||
1075 | prev = NULL; | |
1076 | p = NULL; | |
1077 | for (i = 0; i < n; ++i) { | |
1078 | p = swap_info_get_cont(entries[i], prev); | |
1079 | if (p) | |
1080 | swap_entry_free(p, entries[i]); | |
1081 | else | |
1082 | break; | |
1083 | prev = p; | |
1084 | } | |
235b6217 | 1085 | if (p) |
7c00bafe | 1086 | spin_unlock(&p->lock); |
cb4b86ba KH |
1087 | } |
1088 | ||
1da177e4 | 1089 | /* |
c475a8ab | 1090 | * How many references to page are currently swapped out? |
570a335b HD |
1091 | * This does not give an exact answer when swap count is continued, |
1092 | * but does include the high COUNT_CONTINUED flag to allow for that. | |
1da177e4 | 1093 | */ |
bde05d1c | 1094 | int page_swapcount(struct page *page) |
1da177e4 | 1095 | { |
c475a8ab HD |
1096 | int count = 0; |
1097 | struct swap_info_struct *p; | |
235b6217 | 1098 | struct swap_cluster_info *ci; |
1da177e4 | 1099 | swp_entry_t entry; |
235b6217 | 1100 | unsigned long offset; |
1da177e4 | 1101 | |
4c21e2f2 | 1102 | entry.val = page_private(page); |
235b6217 | 1103 | p = _swap_info_get(entry); |
1da177e4 | 1104 | if (p) { |
235b6217 HY |
1105 | offset = swp_offset(entry); |
1106 | ci = lock_cluster_or_swap_info(p, offset); | |
1107 | count = swap_count(p->swap_map[offset]); | |
1108 | unlock_cluster_or_swap_info(p, ci); | |
1da177e4 | 1109 | } |
c475a8ab | 1110 | return count; |
1da177e4 LT |
1111 | } |
1112 | ||
e8c26ab6 TC |
1113 | /* |
1114 | * How many references to @entry are currently swapped out? | |
1115 | * This does not give an exact answer when swap count is continued, | |
1116 | * but does include the high COUNT_CONTINUED flag to allow for that. | |
1117 | */ | |
1118 | int __swp_swapcount(swp_entry_t entry) | |
1119 | { | |
1120 | int count = 0; | |
1121 | pgoff_t offset; | |
1122 | struct swap_info_struct *si; | |
1123 | struct swap_cluster_info *ci; | |
1124 | ||
1125 | si = __swap_info_get(entry); | |
1126 | if (si) { | |
1127 | offset = swp_offset(entry); | |
1128 | ci = lock_cluster_or_swap_info(si, offset); | |
1129 | count = swap_count(si->swap_map[offset]); | |
1130 | unlock_cluster_or_swap_info(si, ci); | |
1131 | } | |
1132 | return count; | |
1133 | } | |
1134 | ||
8334b962 MK |
1135 | /* |
1136 | * How many references to @entry are currently swapped out? | |
1137 | * This considers COUNT_CONTINUED so it returns exact answer. | |
1138 | */ | |
1139 | int swp_swapcount(swp_entry_t entry) | |
1140 | { | |
1141 | int count, tmp_count, n; | |
1142 | struct swap_info_struct *p; | |
235b6217 | 1143 | struct swap_cluster_info *ci; |
8334b962 MK |
1144 | struct page *page; |
1145 | pgoff_t offset; | |
1146 | unsigned char *map; | |
1147 | ||
235b6217 | 1148 | p = _swap_info_get(entry); |
8334b962 MK |
1149 | if (!p) |
1150 | return 0; | |
1151 | ||
235b6217 HY |
1152 | offset = swp_offset(entry); |
1153 | ||
1154 | ci = lock_cluster_or_swap_info(p, offset); | |
1155 | ||
1156 | count = swap_count(p->swap_map[offset]); | |
8334b962 MK |
1157 | if (!(count & COUNT_CONTINUED)) |
1158 | goto out; | |
1159 | ||
1160 | count &= ~COUNT_CONTINUED; | |
1161 | n = SWAP_MAP_MAX + 1; | |
1162 | ||
8334b962 MK |
1163 | page = vmalloc_to_page(p->swap_map + offset); |
1164 | offset &= ~PAGE_MASK; | |
1165 | VM_BUG_ON(page_private(page) != SWP_CONTINUED); | |
1166 | ||
1167 | do { | |
a8ae4991 | 1168 | page = list_next_entry(page, lru); |
8334b962 MK |
1169 | map = kmap_atomic(page); |
1170 | tmp_count = map[offset]; | |
1171 | kunmap_atomic(map); | |
1172 | ||
1173 | count += (tmp_count & ~COUNT_CONTINUED) * n; | |
1174 | n *= (SWAP_CONT_MAX + 1); | |
1175 | } while (tmp_count & COUNT_CONTINUED); | |
1176 | out: | |
235b6217 | 1177 | unlock_cluster_or_swap_info(p, ci); |
8334b962 MK |
1178 | return count; |
1179 | } | |
1180 | ||
1da177e4 | 1181 | /* |
7b1fe597 HD |
1182 | * We can write to an anon page without COW if there are no other references |
1183 | * to it. And as a side-effect, free up its swap: because the old content | |
1184 | * on disk will never be read, and seeking back there to write new content | |
1185 | * later would only waste time away from clustering. | |
6d0a07ed AA |
1186 | * |
1187 | * NOTE: total_mapcount should not be relied upon by the caller if | |
1188 | * reuse_swap_page() returns false, but it may be always overwritten | |
1189 | * (see the other implementation for CONFIG_SWAP=n). | |
1da177e4 | 1190 | */ |
6d0a07ed | 1191 | bool reuse_swap_page(struct page *page, int *total_mapcount) |
1da177e4 | 1192 | { |
c475a8ab HD |
1193 | int count; |
1194 | ||
309381fe | 1195 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
5ad64688 | 1196 | if (unlikely(PageKsm(page))) |
6d0a07ed AA |
1197 | return false; |
1198 | count = page_trans_huge_mapcount(page, total_mapcount); | |
7b1fe597 | 1199 | if (count <= 1 && PageSwapCache(page)) { |
c475a8ab | 1200 | count += page_swapcount(page); |
f0571429 MK |
1201 | if (count != 1) |
1202 | goto out; | |
1203 | if (!PageWriteback(page)) { | |
7b1fe597 HD |
1204 | delete_from_swap_cache(page); |
1205 | SetPageDirty(page); | |
f0571429 MK |
1206 | } else { |
1207 | swp_entry_t entry; | |
1208 | struct swap_info_struct *p; | |
1209 | ||
1210 | entry.val = page_private(page); | |
1211 | p = swap_info_get(entry); | |
1212 | if (p->flags & SWP_STABLE_WRITES) { | |
1213 | spin_unlock(&p->lock); | |
1214 | return false; | |
1215 | } | |
1216 | spin_unlock(&p->lock); | |
7b1fe597 HD |
1217 | } |
1218 | } | |
f0571429 | 1219 | out: |
5ad64688 | 1220 | return count <= 1; |
1da177e4 LT |
1221 | } |
1222 | ||
1223 | /* | |
a2c43eed HD |
1224 | * If swap is getting full, or if there are no more mappings of this page, |
1225 | * then try_to_free_swap is called to free its swap space. | |
1da177e4 | 1226 | */ |
a2c43eed | 1227 | int try_to_free_swap(struct page *page) |
1da177e4 | 1228 | { |
309381fe | 1229 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
1da177e4 LT |
1230 | |
1231 | if (!PageSwapCache(page)) | |
1232 | return 0; | |
1233 | if (PageWriteback(page)) | |
1234 | return 0; | |
a2c43eed | 1235 | if (page_swapcount(page)) |
1da177e4 LT |
1236 | return 0; |
1237 | ||
b73d7fce HD |
1238 | /* |
1239 | * Once hibernation has begun to create its image of memory, | |
1240 | * there's a danger that one of the calls to try_to_free_swap() | |
1241 | * - most probably a call from __try_to_reclaim_swap() while | |
1242 | * hibernation is allocating its own swap pages for the image, | |
1243 | * but conceivably even a call from memory reclaim - will free | |
1244 | * the swap from a page which has already been recorded in the | |
1245 | * image as a clean swapcache page, and then reuse its swap for | |
1246 | * another page of the image. On waking from hibernation, the | |
1247 | * original page might be freed under memory pressure, then | |
1248 | * later read back in from swap, now with the wrong data. | |
1249 | * | |
2de1a7e4 | 1250 | * Hibernation suspends storage while it is writing the image |
f90ac398 | 1251 | * to disk so check that here. |
b73d7fce | 1252 | */ |
f90ac398 | 1253 | if (pm_suspended_storage()) |
b73d7fce HD |
1254 | return 0; |
1255 | ||
a2c43eed HD |
1256 | delete_from_swap_cache(page); |
1257 | SetPageDirty(page); | |
1258 | return 1; | |
68a22394 RR |
1259 | } |
1260 | ||
1da177e4 LT |
1261 | /* |
1262 | * Free the swap entry like above, but also try to | |
1263 | * free the page cache entry if it is the last user. | |
1264 | */ | |
2509ef26 | 1265 | int free_swap_and_cache(swp_entry_t entry) |
1da177e4 | 1266 | { |
2509ef26 | 1267 | struct swap_info_struct *p; |
1da177e4 | 1268 | struct page *page = NULL; |
7c00bafe | 1269 | unsigned char count; |
1da177e4 | 1270 | |
a7420aa5 | 1271 | if (non_swap_entry(entry)) |
2509ef26 | 1272 | return 1; |
0697212a | 1273 | |
7c00bafe | 1274 | p = _swap_info_get(entry); |
1da177e4 | 1275 | if (p) { |
7c00bafe TC |
1276 | count = __swap_entry_free(p, entry, 1); |
1277 | if (count == SWAP_HAS_CACHE) { | |
33806f06 | 1278 | page = find_get_page(swap_address_space(entry), |
f6ab1f7f | 1279 | swp_offset(entry)); |
8413ac9d | 1280 | if (page && !trylock_page(page)) { |
09cbfeaf | 1281 | put_page(page); |
93fac704 NP |
1282 | page = NULL; |
1283 | } | |
7c00bafe | 1284 | } else if (!count) |
67afa38e | 1285 | free_swap_slot(entry); |
1da177e4 LT |
1286 | } |
1287 | if (page) { | |
a2c43eed HD |
1288 | /* |
1289 | * Not mapped elsewhere, or swap space full? Free it! | |
1290 | * Also recheck PageSwapCache now page is locked (above). | |
1291 | */ | |
93fac704 | 1292 | if (PageSwapCache(page) && !PageWriteback(page) && |
5ccc5aba | 1293 | (!page_mapped(page) || mem_cgroup_swap_full(page))) { |
1da177e4 LT |
1294 | delete_from_swap_cache(page); |
1295 | SetPageDirty(page); | |
1296 | } | |
1297 | unlock_page(page); | |
09cbfeaf | 1298 | put_page(page); |
1da177e4 | 1299 | } |
2509ef26 | 1300 | return p != NULL; |
1da177e4 LT |
1301 | } |
1302 | ||
b0cb1a19 | 1303 | #ifdef CONFIG_HIBERNATION |
f577eb30 | 1304 | /* |
915bae9e | 1305 | * Find the swap type that corresponds to given device (if any). |
f577eb30 | 1306 | * |
915bae9e RW |
1307 | * @offset - number of the PAGE_SIZE-sized block of the device, starting |
1308 | * from 0, in which the swap header is expected to be located. | |
1309 | * | |
1310 | * This is needed for the suspend to disk (aka swsusp). | |
f577eb30 | 1311 | */ |
7bf23687 | 1312 | int swap_type_of(dev_t device, sector_t offset, struct block_device **bdev_p) |
f577eb30 | 1313 | { |
915bae9e | 1314 | struct block_device *bdev = NULL; |
efa90a98 | 1315 | int type; |
f577eb30 | 1316 | |
915bae9e RW |
1317 | if (device) |
1318 | bdev = bdget(device); | |
1319 | ||
f577eb30 | 1320 | spin_lock(&swap_lock); |
efa90a98 HD |
1321 | for (type = 0; type < nr_swapfiles; type++) { |
1322 | struct swap_info_struct *sis = swap_info[type]; | |
f577eb30 | 1323 | |
915bae9e | 1324 | if (!(sis->flags & SWP_WRITEOK)) |
f577eb30 | 1325 | continue; |
b6b5bce3 | 1326 | |
915bae9e | 1327 | if (!bdev) { |
7bf23687 | 1328 | if (bdev_p) |
dddac6a7 | 1329 | *bdev_p = bdgrab(sis->bdev); |
7bf23687 | 1330 | |
6e1819d6 | 1331 | spin_unlock(&swap_lock); |
efa90a98 | 1332 | return type; |
6e1819d6 | 1333 | } |
915bae9e | 1334 | if (bdev == sis->bdev) { |
9625a5f2 | 1335 | struct swap_extent *se = &sis->first_swap_extent; |
915bae9e | 1336 | |
915bae9e | 1337 | if (se->start_block == offset) { |
7bf23687 | 1338 | if (bdev_p) |
dddac6a7 | 1339 | *bdev_p = bdgrab(sis->bdev); |
7bf23687 | 1340 | |
915bae9e RW |
1341 | spin_unlock(&swap_lock); |
1342 | bdput(bdev); | |
efa90a98 | 1343 | return type; |
915bae9e | 1344 | } |
f577eb30 RW |
1345 | } |
1346 | } | |
1347 | spin_unlock(&swap_lock); | |
915bae9e RW |
1348 | if (bdev) |
1349 | bdput(bdev); | |
1350 | ||
f577eb30 RW |
1351 | return -ENODEV; |
1352 | } | |
1353 | ||
73c34b6a HD |
1354 | /* |
1355 | * Get the (PAGE_SIZE) block corresponding to given offset on the swapdev | |
1356 | * corresponding to given index in swap_info (swap type). | |
1357 | */ | |
1358 | sector_t swapdev_block(int type, pgoff_t offset) | |
1359 | { | |
1360 | struct block_device *bdev; | |
1361 | ||
1362 | if ((unsigned int)type >= nr_swapfiles) | |
1363 | return 0; | |
1364 | if (!(swap_info[type]->flags & SWP_WRITEOK)) | |
1365 | return 0; | |
d4906e1a | 1366 | return map_swap_entry(swp_entry(type, offset), &bdev); |
73c34b6a HD |
1367 | } |
1368 | ||
f577eb30 RW |
1369 | /* |
1370 | * Return either the total number of swap pages of given type, or the number | |
1371 | * of free pages of that type (depending on @free) | |
1372 | * | |
1373 | * This is needed for software suspend | |
1374 | */ | |
1375 | unsigned int count_swap_pages(int type, int free) | |
1376 | { | |
1377 | unsigned int n = 0; | |
1378 | ||
efa90a98 HD |
1379 | spin_lock(&swap_lock); |
1380 | if ((unsigned int)type < nr_swapfiles) { | |
1381 | struct swap_info_struct *sis = swap_info[type]; | |
1382 | ||
ec8acf20 | 1383 | spin_lock(&sis->lock); |
efa90a98 HD |
1384 | if (sis->flags & SWP_WRITEOK) { |
1385 | n = sis->pages; | |
f577eb30 | 1386 | if (free) |
efa90a98 | 1387 | n -= sis->inuse_pages; |
f577eb30 | 1388 | } |
ec8acf20 | 1389 | spin_unlock(&sis->lock); |
f577eb30 | 1390 | } |
efa90a98 | 1391 | spin_unlock(&swap_lock); |
f577eb30 RW |
1392 | return n; |
1393 | } | |
73c34b6a | 1394 | #endif /* CONFIG_HIBERNATION */ |
f577eb30 | 1395 | |
9f8bdb3f | 1396 | static inline int pte_same_as_swp(pte_t pte, pte_t swp_pte) |
179ef71c | 1397 | { |
9f8bdb3f | 1398 | return pte_same(pte_swp_clear_soft_dirty(pte), swp_pte); |
179ef71c CG |
1399 | } |
1400 | ||
1da177e4 | 1401 | /* |
72866f6f HD |
1402 | * No need to decide whether this PTE shares the swap entry with others, |
1403 | * just let do_wp_page work it out if a write is requested later - to | |
1404 | * force COW, vm_page_prot omits write permission from any private vma. | |
1da177e4 | 1405 | */ |
044d66c1 | 1406 | static int unuse_pte(struct vm_area_struct *vma, pmd_t *pmd, |
1da177e4 LT |
1407 | unsigned long addr, swp_entry_t entry, struct page *page) |
1408 | { | |
9e16b7fb | 1409 | struct page *swapcache; |
72835c86 | 1410 | struct mem_cgroup *memcg; |
044d66c1 HD |
1411 | spinlock_t *ptl; |
1412 | pte_t *pte; | |
1413 | int ret = 1; | |
1414 | ||
9e16b7fb HD |
1415 | swapcache = page; |
1416 | page = ksm_might_need_to_copy(page, vma, addr); | |
1417 | if (unlikely(!page)) | |
1418 | return -ENOMEM; | |
1419 | ||
f627c2f5 KS |
1420 | if (mem_cgroup_try_charge(page, vma->vm_mm, GFP_KERNEL, |
1421 | &memcg, false)) { | |
044d66c1 | 1422 | ret = -ENOMEM; |
85d9fc89 KH |
1423 | goto out_nolock; |
1424 | } | |
044d66c1 HD |
1425 | |
1426 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); | |
9f8bdb3f | 1427 | if (unlikely(!pte_same_as_swp(*pte, swp_entry_to_pte(entry)))) { |
f627c2f5 | 1428 | mem_cgroup_cancel_charge(page, memcg, false); |
044d66c1 HD |
1429 | ret = 0; |
1430 | goto out; | |
1431 | } | |
8a9f3ccd | 1432 | |
b084d435 | 1433 | dec_mm_counter(vma->vm_mm, MM_SWAPENTS); |
d559db08 | 1434 | inc_mm_counter(vma->vm_mm, MM_ANONPAGES); |
1da177e4 LT |
1435 | get_page(page); |
1436 | set_pte_at(vma->vm_mm, addr, pte, | |
1437 | pte_mkold(mk_pte(page, vma->vm_page_prot))); | |
00501b53 | 1438 | if (page == swapcache) { |
d281ee61 | 1439 | page_add_anon_rmap(page, vma, addr, false); |
f627c2f5 | 1440 | mem_cgroup_commit_charge(page, memcg, true, false); |
00501b53 | 1441 | } else { /* ksm created a completely new copy */ |
d281ee61 | 1442 | page_add_new_anon_rmap(page, vma, addr, false); |
f627c2f5 | 1443 | mem_cgroup_commit_charge(page, memcg, false, false); |
00501b53 JW |
1444 | lru_cache_add_active_or_unevictable(page, vma); |
1445 | } | |
1da177e4 LT |
1446 | swap_free(entry); |
1447 | /* | |
1448 | * Move the page to the active list so it is not | |
1449 | * immediately swapped out again after swapon. | |
1450 | */ | |
1451 | activate_page(page); | |
044d66c1 HD |
1452 | out: |
1453 | pte_unmap_unlock(pte, ptl); | |
85d9fc89 | 1454 | out_nolock: |
9e16b7fb HD |
1455 | if (page != swapcache) { |
1456 | unlock_page(page); | |
1457 | put_page(page); | |
1458 | } | |
044d66c1 | 1459 | return ret; |
1da177e4 LT |
1460 | } |
1461 | ||
1462 | static int unuse_pte_range(struct vm_area_struct *vma, pmd_t *pmd, | |
1463 | unsigned long addr, unsigned long end, | |
1464 | swp_entry_t entry, struct page *page) | |
1465 | { | |
1da177e4 | 1466 | pte_t swp_pte = swp_entry_to_pte(entry); |
705e87c0 | 1467 | pte_t *pte; |
8a9f3ccd | 1468 | int ret = 0; |
1da177e4 | 1469 | |
044d66c1 HD |
1470 | /* |
1471 | * We don't actually need pte lock while scanning for swp_pte: since | |
1472 | * we hold page lock and mmap_sem, swp_pte cannot be inserted into the | |
1473 | * page table while we're scanning; though it could get zapped, and on | |
1474 | * some architectures (e.g. x86_32 with PAE) we might catch a glimpse | |
1475 | * of unmatched parts which look like swp_pte, so unuse_pte must | |
1476 | * recheck under pte lock. Scanning without pte lock lets it be | |
2de1a7e4 | 1477 | * preemptable whenever CONFIG_PREEMPT but not CONFIG_HIGHPTE. |
044d66c1 HD |
1478 | */ |
1479 | pte = pte_offset_map(pmd, addr); | |
1da177e4 LT |
1480 | do { |
1481 | /* | |
1482 | * swapoff spends a _lot_ of time in this loop! | |
1483 | * Test inline before going to call unuse_pte. | |
1484 | */ | |
9f8bdb3f | 1485 | if (unlikely(pte_same_as_swp(*pte, swp_pte))) { |
044d66c1 HD |
1486 | pte_unmap(pte); |
1487 | ret = unuse_pte(vma, pmd, addr, entry, page); | |
1488 | if (ret) | |
1489 | goto out; | |
1490 | pte = pte_offset_map(pmd, addr); | |
1da177e4 LT |
1491 | } |
1492 | } while (pte++, addr += PAGE_SIZE, addr != end); | |
044d66c1 HD |
1493 | pte_unmap(pte - 1); |
1494 | out: | |
8a9f3ccd | 1495 | return ret; |
1da177e4 LT |
1496 | } |
1497 | ||
1498 | static inline int unuse_pmd_range(struct vm_area_struct *vma, pud_t *pud, | |
1499 | unsigned long addr, unsigned long end, | |
1500 | swp_entry_t entry, struct page *page) | |
1501 | { | |
1502 | pmd_t *pmd; | |
1503 | unsigned long next; | |
8a9f3ccd | 1504 | int ret; |
1da177e4 LT |
1505 | |
1506 | pmd = pmd_offset(pud, addr); | |
1507 | do { | |
dc644a07 | 1508 | cond_resched(); |
1da177e4 | 1509 | next = pmd_addr_end(addr, end); |
1a5a9906 | 1510 | if (pmd_none_or_trans_huge_or_clear_bad(pmd)) |
1da177e4 | 1511 | continue; |
8a9f3ccd BS |
1512 | ret = unuse_pte_range(vma, pmd, addr, next, entry, page); |
1513 | if (ret) | |
1514 | return ret; | |
1da177e4 LT |
1515 | } while (pmd++, addr = next, addr != end); |
1516 | return 0; | |
1517 | } | |
1518 | ||
1519 | static inline int unuse_pud_range(struct vm_area_struct *vma, pgd_t *pgd, | |
1520 | unsigned long addr, unsigned long end, | |
1521 | swp_entry_t entry, struct page *page) | |
1522 | { | |
1523 | pud_t *pud; | |
1524 | unsigned long next; | |
8a9f3ccd | 1525 | int ret; |
1da177e4 LT |
1526 | |
1527 | pud = pud_offset(pgd, addr); | |
1528 | do { | |
1529 | next = pud_addr_end(addr, end); | |
1530 | if (pud_none_or_clear_bad(pud)) | |
1531 | continue; | |
8a9f3ccd BS |
1532 | ret = unuse_pmd_range(vma, pud, addr, next, entry, page); |
1533 | if (ret) | |
1534 | return ret; | |
1da177e4 LT |
1535 | } while (pud++, addr = next, addr != end); |
1536 | return 0; | |
1537 | } | |
1538 | ||
1539 | static int unuse_vma(struct vm_area_struct *vma, | |
1540 | swp_entry_t entry, struct page *page) | |
1541 | { | |
1542 | pgd_t *pgd; | |
1543 | unsigned long addr, end, next; | |
8a9f3ccd | 1544 | int ret; |
1da177e4 | 1545 | |
3ca7b3c5 | 1546 | if (page_anon_vma(page)) { |
1da177e4 LT |
1547 | addr = page_address_in_vma(page, vma); |
1548 | if (addr == -EFAULT) | |
1549 | return 0; | |
1550 | else | |
1551 | end = addr + PAGE_SIZE; | |
1552 | } else { | |
1553 | addr = vma->vm_start; | |
1554 | end = vma->vm_end; | |
1555 | } | |
1556 | ||
1557 | pgd = pgd_offset(vma->vm_mm, addr); | |
1558 | do { | |
1559 | next = pgd_addr_end(addr, end); | |
1560 | if (pgd_none_or_clear_bad(pgd)) | |
1561 | continue; | |
8a9f3ccd BS |
1562 | ret = unuse_pud_range(vma, pgd, addr, next, entry, page); |
1563 | if (ret) | |
1564 | return ret; | |
1da177e4 LT |
1565 | } while (pgd++, addr = next, addr != end); |
1566 | return 0; | |
1567 | } | |
1568 | ||
1569 | static int unuse_mm(struct mm_struct *mm, | |
1570 | swp_entry_t entry, struct page *page) | |
1571 | { | |
1572 | struct vm_area_struct *vma; | |
8a9f3ccd | 1573 | int ret = 0; |
1da177e4 LT |
1574 | |
1575 | if (!down_read_trylock(&mm->mmap_sem)) { | |
1576 | /* | |
7d03431c FLVC |
1577 | * Activate page so shrink_inactive_list is unlikely to unmap |
1578 | * its ptes while lock is dropped, so swapoff can make progress. | |
1da177e4 | 1579 | */ |
c475a8ab | 1580 | activate_page(page); |
1da177e4 LT |
1581 | unlock_page(page); |
1582 | down_read(&mm->mmap_sem); | |
1583 | lock_page(page); | |
1584 | } | |
1da177e4 | 1585 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
8a9f3ccd | 1586 | if (vma->anon_vma && (ret = unuse_vma(vma, entry, page))) |
1da177e4 | 1587 | break; |
dc644a07 | 1588 | cond_resched(); |
1da177e4 | 1589 | } |
1da177e4 | 1590 | up_read(&mm->mmap_sem); |
8a9f3ccd | 1591 | return (ret < 0)? ret: 0; |
1da177e4 LT |
1592 | } |
1593 | ||
1594 | /* | |
38b5faf4 DM |
1595 | * Scan swap_map (or frontswap_map if frontswap parameter is true) |
1596 | * from current position to next entry still in use. | |
1da177e4 LT |
1597 | * Recycle to start on reaching the end, returning 0 when empty. |
1598 | */ | |
6eb396dc | 1599 | static unsigned int find_next_to_unuse(struct swap_info_struct *si, |
38b5faf4 | 1600 | unsigned int prev, bool frontswap) |
1da177e4 | 1601 | { |
6eb396dc HD |
1602 | unsigned int max = si->max; |
1603 | unsigned int i = prev; | |
8d69aaee | 1604 | unsigned char count; |
1da177e4 LT |
1605 | |
1606 | /* | |
5d337b91 | 1607 | * No need for swap_lock here: we're just looking |
1da177e4 LT |
1608 | * for whether an entry is in use, not modifying it; false |
1609 | * hits are okay, and sys_swapoff() has already prevented new | |
5d337b91 | 1610 | * allocations from this area (while holding swap_lock). |
1da177e4 LT |
1611 | */ |
1612 | for (;;) { | |
1613 | if (++i >= max) { | |
1614 | if (!prev) { | |
1615 | i = 0; | |
1616 | break; | |
1617 | } | |
1618 | /* | |
1619 | * No entries in use at top of swap_map, | |
1620 | * loop back to start and recheck there. | |
1621 | */ | |
1622 | max = prev + 1; | |
1623 | prev = 0; | |
1624 | i = 1; | |
1625 | } | |
4db0c3c2 | 1626 | count = READ_ONCE(si->swap_map[i]); |
355cfa73 | 1627 | if (count && swap_count(count) != SWAP_MAP_BAD) |
dc644a07 HD |
1628 | if (!frontswap || frontswap_test(si, i)) |
1629 | break; | |
1630 | if ((i % LATENCY_LIMIT) == 0) | |
1631 | cond_resched(); | |
1da177e4 LT |
1632 | } |
1633 | return i; | |
1634 | } | |
1635 | ||
1636 | /* | |
1637 | * We completely avoid races by reading each swap page in advance, | |
1638 | * and then search for the process using it. All the necessary | |
1639 | * page table adjustments can then be made atomically. | |
38b5faf4 DM |
1640 | * |
1641 | * if the boolean frontswap is true, only unuse pages_to_unuse pages; | |
1642 | * pages_to_unuse==0 means all pages; ignored if frontswap is false | |
1da177e4 | 1643 | */ |
38b5faf4 DM |
1644 | int try_to_unuse(unsigned int type, bool frontswap, |
1645 | unsigned long pages_to_unuse) | |
1da177e4 | 1646 | { |
efa90a98 | 1647 | struct swap_info_struct *si = swap_info[type]; |
1da177e4 | 1648 | struct mm_struct *start_mm; |
edfe23da SL |
1649 | volatile unsigned char *swap_map; /* swap_map is accessed without |
1650 | * locking. Mark it as volatile | |
1651 | * to prevent compiler doing | |
1652 | * something odd. | |
1653 | */ | |
8d69aaee | 1654 | unsigned char swcount; |
1da177e4 LT |
1655 | struct page *page; |
1656 | swp_entry_t entry; | |
6eb396dc | 1657 | unsigned int i = 0; |
1da177e4 | 1658 | int retval = 0; |
1da177e4 LT |
1659 | |
1660 | /* | |
1661 | * When searching mms for an entry, a good strategy is to | |
1662 | * start at the first mm we freed the previous entry from | |
1663 | * (though actually we don't notice whether we or coincidence | |
1664 | * freed the entry). Initialize this start_mm with a hold. | |
1665 | * | |
1666 | * A simpler strategy would be to start at the last mm we | |
1667 | * freed the previous entry from; but that would take less | |
1668 | * advantage of mmlist ordering, which clusters forked mms | |
1669 | * together, child after parent. If we race with dup_mmap(), we | |
1670 | * prefer to resolve parent before child, lest we miss entries | |
1671 | * duplicated after we scanned child: using last mm would invert | |
570a335b | 1672 | * that. |
1da177e4 LT |
1673 | */ |
1674 | start_mm = &init_mm; | |
3fce371b | 1675 | mmget(&init_mm); |
1da177e4 LT |
1676 | |
1677 | /* | |
1678 | * Keep on scanning until all entries have gone. Usually, | |
1679 | * one pass through swap_map is enough, but not necessarily: | |
1680 | * there are races when an instance of an entry might be missed. | |
1681 | */ | |
38b5faf4 | 1682 | while ((i = find_next_to_unuse(si, i, frontswap)) != 0) { |
1da177e4 LT |
1683 | if (signal_pending(current)) { |
1684 | retval = -EINTR; | |
1685 | break; | |
1686 | } | |
1687 | ||
886bb7e9 | 1688 | /* |
1da177e4 LT |
1689 | * Get a page for the entry, using the existing swap |
1690 | * cache page if there is one. Otherwise, get a clean | |
886bb7e9 | 1691 | * page and read the swap into it. |
1da177e4 LT |
1692 | */ |
1693 | swap_map = &si->swap_map[i]; | |
1694 | entry = swp_entry(type, i); | |
02098fea HD |
1695 | page = read_swap_cache_async(entry, |
1696 | GFP_HIGHUSER_MOVABLE, NULL, 0); | |
1da177e4 LT |
1697 | if (!page) { |
1698 | /* | |
1699 | * Either swap_duplicate() failed because entry | |
1700 | * has been freed independently, and will not be | |
1701 | * reused since sys_swapoff() already disabled | |
1702 | * allocation from here, or alloc_page() failed. | |
1703 | */ | |
edfe23da SL |
1704 | swcount = *swap_map; |
1705 | /* | |
1706 | * We don't hold lock here, so the swap entry could be | |
1707 | * SWAP_MAP_BAD (when the cluster is discarding). | |
1708 | * Instead of fail out, We can just skip the swap | |
1709 | * entry because swapoff will wait for discarding | |
1710 | * finish anyway. | |
1711 | */ | |
1712 | if (!swcount || swcount == SWAP_MAP_BAD) | |
1da177e4 LT |
1713 | continue; |
1714 | retval = -ENOMEM; | |
1715 | break; | |
1716 | } | |
1717 | ||
1718 | /* | |
1719 | * Don't hold on to start_mm if it looks like exiting. | |
1720 | */ | |
1721 | if (atomic_read(&start_mm->mm_users) == 1) { | |
1722 | mmput(start_mm); | |
1723 | start_mm = &init_mm; | |
3fce371b | 1724 | mmget(&init_mm); |
1da177e4 LT |
1725 | } |
1726 | ||
1727 | /* | |
1728 | * Wait for and lock page. When do_swap_page races with | |
1729 | * try_to_unuse, do_swap_page can handle the fault much | |
1730 | * faster than try_to_unuse can locate the entry. This | |
1731 | * apparently redundant "wait_on_page_locked" lets try_to_unuse | |
1732 | * defer to do_swap_page in such a case - in some tests, | |
1733 | * do_swap_page and try_to_unuse repeatedly compete. | |
1734 | */ | |
1735 | wait_on_page_locked(page); | |
1736 | wait_on_page_writeback(page); | |
1737 | lock_page(page); | |
1738 | wait_on_page_writeback(page); | |
1739 | ||
1740 | /* | |
1741 | * Remove all references to entry. | |
1da177e4 | 1742 | */ |
1da177e4 | 1743 | swcount = *swap_map; |
aaa46865 HD |
1744 | if (swap_count(swcount) == SWAP_MAP_SHMEM) { |
1745 | retval = shmem_unuse(entry, page); | |
1746 | /* page has already been unlocked and released */ | |
1747 | if (retval < 0) | |
1748 | break; | |
1749 | continue; | |
1da177e4 | 1750 | } |
aaa46865 HD |
1751 | if (swap_count(swcount) && start_mm != &init_mm) |
1752 | retval = unuse_mm(start_mm, entry, page); | |
1753 | ||
355cfa73 | 1754 | if (swap_count(*swap_map)) { |
1da177e4 LT |
1755 | int set_start_mm = (*swap_map >= swcount); |
1756 | struct list_head *p = &start_mm->mmlist; | |
1757 | struct mm_struct *new_start_mm = start_mm; | |
1758 | struct mm_struct *prev_mm = start_mm; | |
1759 | struct mm_struct *mm; | |
1760 | ||
3fce371b VN |
1761 | mmget(new_start_mm); |
1762 | mmget(prev_mm); | |
1da177e4 | 1763 | spin_lock(&mmlist_lock); |
aaa46865 | 1764 | while (swap_count(*swap_map) && !retval && |
1da177e4 LT |
1765 | (p = p->next) != &start_mm->mmlist) { |
1766 | mm = list_entry(p, struct mm_struct, mmlist); | |
388f7934 | 1767 | if (!mmget_not_zero(mm)) |
1da177e4 | 1768 | continue; |
1da177e4 LT |
1769 | spin_unlock(&mmlist_lock); |
1770 | mmput(prev_mm); | |
1771 | prev_mm = mm; | |
1772 | ||
1773 | cond_resched(); | |
1774 | ||
1775 | swcount = *swap_map; | |
355cfa73 | 1776 | if (!swap_count(swcount)) /* any usage ? */ |
1da177e4 | 1777 | ; |
aaa46865 | 1778 | else if (mm == &init_mm) |
1da177e4 | 1779 | set_start_mm = 1; |
aaa46865 | 1780 | else |
1da177e4 | 1781 | retval = unuse_mm(mm, entry, page); |
355cfa73 | 1782 | |
32c5fc10 | 1783 | if (set_start_mm && *swap_map < swcount) { |
1da177e4 | 1784 | mmput(new_start_mm); |
3fce371b | 1785 | mmget(mm); |
1da177e4 LT |
1786 | new_start_mm = mm; |
1787 | set_start_mm = 0; | |
1788 | } | |
1789 | spin_lock(&mmlist_lock); | |
1790 | } | |
1791 | spin_unlock(&mmlist_lock); | |
1792 | mmput(prev_mm); | |
1793 | mmput(start_mm); | |
1794 | start_mm = new_start_mm; | |
1795 | } | |
1796 | if (retval) { | |
1797 | unlock_page(page); | |
09cbfeaf | 1798 | put_page(page); |
1da177e4 LT |
1799 | break; |
1800 | } | |
1801 | ||
1da177e4 LT |
1802 | /* |
1803 | * If a reference remains (rare), we would like to leave | |
1804 | * the page in the swap cache; but try_to_unmap could | |
1805 | * then re-duplicate the entry once we drop page lock, | |
1806 | * so we might loop indefinitely; also, that page could | |
1807 | * not be swapped out to other storage meanwhile. So: | |
1808 | * delete from cache even if there's another reference, | |
1809 | * after ensuring that the data has been saved to disk - | |
1810 | * since if the reference remains (rarer), it will be | |
1811 | * read from disk into another page. Splitting into two | |
1812 | * pages would be incorrect if swap supported "shared | |
1813 | * private" pages, but they are handled by tmpfs files. | |
5ad64688 HD |
1814 | * |
1815 | * Given how unuse_vma() targets one particular offset | |
1816 | * in an anon_vma, once the anon_vma has been determined, | |
1817 | * this splitting happens to be just what is needed to | |
1818 | * handle where KSM pages have been swapped out: re-reading | |
1819 | * is unnecessarily slow, but we can fix that later on. | |
1da177e4 | 1820 | */ |
355cfa73 KH |
1821 | if (swap_count(*swap_map) && |
1822 | PageDirty(page) && PageSwapCache(page)) { | |
1da177e4 LT |
1823 | struct writeback_control wbc = { |
1824 | .sync_mode = WB_SYNC_NONE, | |
1825 | }; | |
1826 | ||
1827 | swap_writepage(page, &wbc); | |
1828 | lock_page(page); | |
1829 | wait_on_page_writeback(page); | |
1830 | } | |
68bdc8d6 HD |
1831 | |
1832 | /* | |
1833 | * It is conceivable that a racing task removed this page from | |
1834 | * swap cache just before we acquired the page lock at the top, | |
1835 | * or while we dropped it in unuse_mm(). The page might even | |
1836 | * be back in swap cache on another swap area: that we must not | |
1837 | * delete, since it may not have been written out to swap yet. | |
1838 | */ | |
1839 | if (PageSwapCache(page) && | |
1840 | likely(page_private(page) == entry.val)) | |
2e0e26c7 | 1841 | delete_from_swap_cache(page); |
1da177e4 LT |
1842 | |
1843 | /* | |
1844 | * So we could skip searching mms once swap count went | |
1845 | * to 1, we did not mark any present ptes as dirty: must | |
2706a1b8 | 1846 | * mark page dirty so shrink_page_list will preserve it. |
1da177e4 LT |
1847 | */ |
1848 | SetPageDirty(page); | |
1849 | unlock_page(page); | |
09cbfeaf | 1850 | put_page(page); |
1da177e4 LT |
1851 | |
1852 | /* | |
1853 | * Make sure that we aren't completely killing | |
1854 | * interactive performance. | |
1855 | */ | |
1856 | cond_resched(); | |
38b5faf4 DM |
1857 | if (frontswap && pages_to_unuse > 0) { |
1858 | if (!--pages_to_unuse) | |
1859 | break; | |
1860 | } | |
1da177e4 LT |
1861 | } |
1862 | ||
1863 | mmput(start_mm); | |
1da177e4 LT |
1864 | return retval; |
1865 | } | |
1866 | ||
1867 | /* | |
5d337b91 HD |
1868 | * After a successful try_to_unuse, if no swap is now in use, we know |
1869 | * we can empty the mmlist. swap_lock must be held on entry and exit. | |
1870 | * Note that mmlist_lock nests inside swap_lock, and an mm must be | |
1da177e4 LT |
1871 | * added to the mmlist just after page_duplicate - before would be racy. |
1872 | */ | |
1873 | static void drain_mmlist(void) | |
1874 | { | |
1875 | struct list_head *p, *next; | |
efa90a98 | 1876 | unsigned int type; |
1da177e4 | 1877 | |
efa90a98 HD |
1878 | for (type = 0; type < nr_swapfiles; type++) |
1879 | if (swap_info[type]->inuse_pages) | |
1da177e4 LT |
1880 | return; |
1881 | spin_lock(&mmlist_lock); | |
1882 | list_for_each_safe(p, next, &init_mm.mmlist) | |
1883 | list_del_init(p); | |
1884 | spin_unlock(&mmlist_lock); | |
1885 | } | |
1886 | ||
1887 | /* | |
1888 | * Use this swapdev's extent info to locate the (PAGE_SIZE) block which | |
d4906e1a LS |
1889 | * corresponds to page offset for the specified swap entry. |
1890 | * Note that the type of this function is sector_t, but it returns page offset | |
1891 | * into the bdev, not sector offset. | |
1da177e4 | 1892 | */ |
d4906e1a | 1893 | static sector_t map_swap_entry(swp_entry_t entry, struct block_device **bdev) |
1da177e4 | 1894 | { |
f29ad6a9 HD |
1895 | struct swap_info_struct *sis; |
1896 | struct swap_extent *start_se; | |
1897 | struct swap_extent *se; | |
1898 | pgoff_t offset; | |
1899 | ||
efa90a98 | 1900 | sis = swap_info[swp_type(entry)]; |
f29ad6a9 HD |
1901 | *bdev = sis->bdev; |
1902 | ||
1903 | offset = swp_offset(entry); | |
1904 | start_se = sis->curr_swap_extent; | |
1905 | se = start_se; | |
1da177e4 LT |
1906 | |
1907 | for ( ; ; ) { | |
1da177e4 LT |
1908 | if (se->start_page <= offset && |
1909 | offset < (se->start_page + se->nr_pages)) { | |
1910 | return se->start_block + (offset - se->start_page); | |
1911 | } | |
a8ae4991 | 1912 | se = list_next_entry(se, list); |
1da177e4 LT |
1913 | sis->curr_swap_extent = se; |
1914 | BUG_ON(se == start_se); /* It *must* be present */ | |
1915 | } | |
1916 | } | |
1917 | ||
d4906e1a LS |
1918 | /* |
1919 | * Returns the page offset into bdev for the specified page's swap entry. | |
1920 | */ | |
1921 | sector_t map_swap_page(struct page *page, struct block_device **bdev) | |
1922 | { | |
1923 | swp_entry_t entry; | |
1924 | entry.val = page_private(page); | |
1925 | return map_swap_entry(entry, bdev); | |
1926 | } | |
1927 | ||
1da177e4 LT |
1928 | /* |
1929 | * Free all of a swapdev's extent information | |
1930 | */ | |
1931 | static void destroy_swap_extents(struct swap_info_struct *sis) | |
1932 | { | |
9625a5f2 | 1933 | while (!list_empty(&sis->first_swap_extent.list)) { |
1da177e4 LT |
1934 | struct swap_extent *se; |
1935 | ||
a8ae4991 | 1936 | se = list_first_entry(&sis->first_swap_extent.list, |
1da177e4 LT |
1937 | struct swap_extent, list); |
1938 | list_del(&se->list); | |
1939 | kfree(se); | |
1940 | } | |
62c230bc MG |
1941 | |
1942 | if (sis->flags & SWP_FILE) { | |
1943 | struct file *swap_file = sis->swap_file; | |
1944 | struct address_space *mapping = swap_file->f_mapping; | |
1945 | ||
1946 | sis->flags &= ~SWP_FILE; | |
1947 | mapping->a_ops->swap_deactivate(swap_file); | |
1948 | } | |
1da177e4 LT |
1949 | } |
1950 | ||
1951 | /* | |
1952 | * Add a block range (and the corresponding page range) into this swapdev's | |
11d31886 | 1953 | * extent list. The extent list is kept sorted in page order. |
1da177e4 | 1954 | * |
11d31886 | 1955 | * This function rather assumes that it is called in ascending page order. |
1da177e4 | 1956 | */ |
a509bc1a | 1957 | int |
1da177e4 LT |
1958 | add_swap_extent(struct swap_info_struct *sis, unsigned long start_page, |
1959 | unsigned long nr_pages, sector_t start_block) | |
1960 | { | |
1961 | struct swap_extent *se; | |
1962 | struct swap_extent *new_se; | |
1963 | struct list_head *lh; | |
1964 | ||
9625a5f2 HD |
1965 | if (start_page == 0) { |
1966 | se = &sis->first_swap_extent; | |
1967 | sis->curr_swap_extent = se; | |
1968 | se->start_page = 0; | |
1969 | se->nr_pages = nr_pages; | |
1970 | se->start_block = start_block; | |
1971 | return 1; | |
1972 | } else { | |
1973 | lh = sis->first_swap_extent.list.prev; /* Highest extent */ | |
1da177e4 | 1974 | se = list_entry(lh, struct swap_extent, list); |
11d31886 HD |
1975 | BUG_ON(se->start_page + se->nr_pages != start_page); |
1976 | if (se->start_block + se->nr_pages == start_block) { | |
1da177e4 LT |
1977 | /* Merge it */ |
1978 | se->nr_pages += nr_pages; | |
1979 | return 0; | |
1980 | } | |
1da177e4 LT |
1981 | } |
1982 | ||
1983 | /* | |
1984 | * No merge. Insert a new extent, preserving ordering. | |
1985 | */ | |
1986 | new_se = kmalloc(sizeof(*se), GFP_KERNEL); | |
1987 | if (new_se == NULL) | |
1988 | return -ENOMEM; | |
1989 | new_se->start_page = start_page; | |
1990 | new_se->nr_pages = nr_pages; | |
1991 | new_se->start_block = start_block; | |
1992 | ||
9625a5f2 | 1993 | list_add_tail(&new_se->list, &sis->first_swap_extent.list); |
53092a74 | 1994 | return 1; |
1da177e4 LT |
1995 | } |
1996 | ||
1997 | /* | |
1998 | * A `swap extent' is a simple thing which maps a contiguous range of pages | |
1999 | * onto a contiguous range of disk blocks. An ordered list of swap extents | |
2000 | * is built at swapon time and is then used at swap_writepage/swap_readpage | |
2001 | * time for locating where on disk a page belongs. | |
2002 | * | |
2003 | * If the swapfile is an S_ISBLK block device, a single extent is installed. | |
2004 | * This is done so that the main operating code can treat S_ISBLK and S_ISREG | |
2005 | * swap files identically. | |
2006 | * | |
2007 | * Whether the swapdev is an S_ISREG file or an S_ISBLK blockdev, the swap | |
2008 | * extent list operates in PAGE_SIZE disk blocks. Both S_ISREG and S_ISBLK | |
2009 | * swapfiles are handled *identically* after swapon time. | |
2010 | * | |
2011 | * For S_ISREG swapfiles, setup_swap_extents() will walk all the file's blocks | |
2012 | * and will parse them into an ordered extent list, in PAGE_SIZE chunks. If | |
2013 | * some stray blocks are found which do not fall within the PAGE_SIZE alignment | |
2014 | * requirements, they are simply tossed out - we will never use those blocks | |
2015 | * for swapping. | |
2016 | * | |
b0d9bcd4 | 2017 | * For S_ISREG swapfiles we set S_SWAPFILE across the life of the swapon. This |
1da177e4 LT |
2018 | * prevents root from shooting her foot off by ftruncating an in-use swapfile, |
2019 | * which will scribble on the fs. | |
2020 | * | |
2021 | * The amount of disk space which a single swap extent represents varies. | |
2022 | * Typically it is in the 1-4 megabyte range. So we can have hundreds of | |
2023 | * extents in the list. To avoid much list walking, we cache the previous | |
2024 | * search location in `curr_swap_extent', and start new searches from there. | |
2025 | * This is extremely effective. The average number of iterations in | |
2026 | * map_swap_page() has been measured at about 0.3 per page. - akpm. | |
2027 | */ | |
53092a74 | 2028 | static int setup_swap_extents(struct swap_info_struct *sis, sector_t *span) |
1da177e4 | 2029 | { |
62c230bc MG |
2030 | struct file *swap_file = sis->swap_file; |
2031 | struct address_space *mapping = swap_file->f_mapping; | |
2032 | struct inode *inode = mapping->host; | |
1da177e4 LT |
2033 | int ret; |
2034 | ||
1da177e4 LT |
2035 | if (S_ISBLK(inode->i_mode)) { |
2036 | ret = add_swap_extent(sis, 0, sis->max, 0); | |
53092a74 | 2037 | *span = sis->pages; |
a509bc1a | 2038 | return ret; |
1da177e4 LT |
2039 | } |
2040 | ||
62c230bc | 2041 | if (mapping->a_ops->swap_activate) { |
a509bc1a | 2042 | ret = mapping->a_ops->swap_activate(sis, swap_file, span); |
62c230bc MG |
2043 | if (!ret) { |
2044 | sis->flags |= SWP_FILE; | |
2045 | ret = add_swap_extent(sis, 0, sis->max, 0); | |
2046 | *span = sis->pages; | |
2047 | } | |
a509bc1a | 2048 | return ret; |
62c230bc MG |
2049 | } |
2050 | ||
a509bc1a | 2051 | return generic_swapfile_activate(sis, swap_file, span); |
1da177e4 LT |
2052 | } |
2053 | ||
cf0cac0a | 2054 | static void _enable_swap_info(struct swap_info_struct *p, int prio, |
2a8f9449 SL |
2055 | unsigned char *swap_map, |
2056 | struct swap_cluster_info *cluster_info) | |
40531542 | 2057 | { |
40531542 CEB |
2058 | if (prio >= 0) |
2059 | p->prio = prio; | |
2060 | else | |
2061 | p->prio = --least_priority; | |
18ab4d4c DS |
2062 | /* |
2063 | * the plist prio is negated because plist ordering is | |
2064 | * low-to-high, while swap ordering is high-to-low | |
2065 | */ | |
2066 | p->list.prio = -p->prio; | |
2067 | p->avail_list.prio = -p->prio; | |
40531542 | 2068 | p->swap_map = swap_map; |
2a8f9449 | 2069 | p->cluster_info = cluster_info; |
40531542 | 2070 | p->flags |= SWP_WRITEOK; |
ec8acf20 | 2071 | atomic_long_add(p->pages, &nr_swap_pages); |
40531542 CEB |
2072 | total_swap_pages += p->pages; |
2073 | ||
adfab836 | 2074 | assert_spin_locked(&swap_lock); |
adfab836 | 2075 | /* |
18ab4d4c DS |
2076 | * both lists are plists, and thus priority ordered. |
2077 | * swap_active_head needs to be priority ordered for swapoff(), | |
2078 | * which on removal of any swap_info_struct with an auto-assigned | |
2079 | * (i.e. negative) priority increments the auto-assigned priority | |
2080 | * of any lower-priority swap_info_structs. | |
2081 | * swap_avail_head needs to be priority ordered for get_swap_page(), | |
2082 | * which allocates swap pages from the highest available priority | |
2083 | * swap_info_struct. | |
adfab836 | 2084 | */ |
18ab4d4c DS |
2085 | plist_add(&p->list, &swap_active_head); |
2086 | spin_lock(&swap_avail_lock); | |
2087 | plist_add(&p->avail_list, &swap_avail_head); | |
2088 | spin_unlock(&swap_avail_lock); | |
cf0cac0a CEB |
2089 | } |
2090 | ||
2091 | static void enable_swap_info(struct swap_info_struct *p, int prio, | |
2092 | unsigned char *swap_map, | |
2a8f9449 | 2093 | struct swap_cluster_info *cluster_info, |
cf0cac0a CEB |
2094 | unsigned long *frontswap_map) |
2095 | { | |
4f89849d | 2096 | frontswap_init(p->type, frontswap_map); |
cf0cac0a | 2097 | spin_lock(&swap_lock); |
ec8acf20 | 2098 | spin_lock(&p->lock); |
2a8f9449 | 2099 | _enable_swap_info(p, prio, swap_map, cluster_info); |
ec8acf20 | 2100 | spin_unlock(&p->lock); |
cf0cac0a CEB |
2101 | spin_unlock(&swap_lock); |
2102 | } | |
2103 | ||
2104 | static void reinsert_swap_info(struct swap_info_struct *p) | |
2105 | { | |
2106 | spin_lock(&swap_lock); | |
ec8acf20 | 2107 | spin_lock(&p->lock); |
2a8f9449 | 2108 | _enable_swap_info(p, p->prio, p->swap_map, p->cluster_info); |
ec8acf20 | 2109 | spin_unlock(&p->lock); |
40531542 CEB |
2110 | spin_unlock(&swap_lock); |
2111 | } | |
2112 | ||
67afa38e TC |
2113 | bool has_usable_swap(void) |
2114 | { | |
2115 | bool ret = true; | |
2116 | ||
2117 | spin_lock(&swap_lock); | |
2118 | if (plist_head_empty(&swap_active_head)) | |
2119 | ret = false; | |
2120 | spin_unlock(&swap_lock); | |
2121 | return ret; | |
2122 | } | |
2123 | ||
c4ea37c2 | 2124 | SYSCALL_DEFINE1(swapoff, const char __user *, specialfile) |
1da177e4 | 2125 | { |
73c34b6a | 2126 | struct swap_info_struct *p = NULL; |
8d69aaee | 2127 | unsigned char *swap_map; |
2a8f9449 | 2128 | struct swap_cluster_info *cluster_info; |
4f89849d | 2129 | unsigned long *frontswap_map; |
1da177e4 LT |
2130 | struct file *swap_file, *victim; |
2131 | struct address_space *mapping; | |
2132 | struct inode *inode; | |
91a27b2a | 2133 | struct filename *pathname; |
adfab836 | 2134 | int err, found = 0; |
5b808a23 | 2135 | unsigned int old_block_size; |
886bb7e9 | 2136 | |
1da177e4 LT |
2137 | if (!capable(CAP_SYS_ADMIN)) |
2138 | return -EPERM; | |
2139 | ||
191c5424 AV |
2140 | BUG_ON(!current->mm); |
2141 | ||
1da177e4 | 2142 | pathname = getname(specialfile); |
1da177e4 | 2143 | if (IS_ERR(pathname)) |
f58b59c1 | 2144 | return PTR_ERR(pathname); |
1da177e4 | 2145 | |
669abf4e | 2146 | victim = file_open_name(pathname, O_RDWR|O_LARGEFILE, 0); |
1da177e4 LT |
2147 | err = PTR_ERR(victim); |
2148 | if (IS_ERR(victim)) | |
2149 | goto out; | |
2150 | ||
2151 | mapping = victim->f_mapping; | |
5d337b91 | 2152 | spin_lock(&swap_lock); |
18ab4d4c | 2153 | plist_for_each_entry(p, &swap_active_head, list) { |
22c6f8fd | 2154 | if (p->flags & SWP_WRITEOK) { |
adfab836 DS |
2155 | if (p->swap_file->f_mapping == mapping) { |
2156 | found = 1; | |
1da177e4 | 2157 | break; |
adfab836 | 2158 | } |
1da177e4 | 2159 | } |
1da177e4 | 2160 | } |
adfab836 | 2161 | if (!found) { |
1da177e4 | 2162 | err = -EINVAL; |
5d337b91 | 2163 | spin_unlock(&swap_lock); |
1da177e4 LT |
2164 | goto out_dput; |
2165 | } | |
191c5424 | 2166 | if (!security_vm_enough_memory_mm(current->mm, p->pages)) |
1da177e4 LT |
2167 | vm_unacct_memory(p->pages); |
2168 | else { | |
2169 | err = -ENOMEM; | |
5d337b91 | 2170 | spin_unlock(&swap_lock); |
1da177e4 LT |
2171 | goto out_dput; |
2172 | } | |
18ab4d4c DS |
2173 | spin_lock(&swap_avail_lock); |
2174 | plist_del(&p->avail_list, &swap_avail_head); | |
2175 | spin_unlock(&swap_avail_lock); | |
ec8acf20 | 2176 | spin_lock(&p->lock); |
78ecba08 | 2177 | if (p->prio < 0) { |
adfab836 DS |
2178 | struct swap_info_struct *si = p; |
2179 | ||
18ab4d4c | 2180 | plist_for_each_entry_continue(si, &swap_active_head, list) { |
adfab836 | 2181 | si->prio++; |
18ab4d4c DS |
2182 | si->list.prio--; |
2183 | si->avail_list.prio--; | |
adfab836 | 2184 | } |
78ecba08 HD |
2185 | least_priority++; |
2186 | } | |
18ab4d4c | 2187 | plist_del(&p->list, &swap_active_head); |
ec8acf20 | 2188 | atomic_long_sub(p->pages, &nr_swap_pages); |
1da177e4 LT |
2189 | total_swap_pages -= p->pages; |
2190 | p->flags &= ~SWP_WRITEOK; | |
ec8acf20 | 2191 | spin_unlock(&p->lock); |
5d337b91 | 2192 | spin_unlock(&swap_lock); |
fb4f88dc | 2193 | |
039939a6 TC |
2194 | disable_swap_slots_cache_lock(); |
2195 | ||
e1e12d2f | 2196 | set_current_oom_origin(); |
adfab836 | 2197 | err = try_to_unuse(p->type, false, 0); /* force unuse all pages */ |
e1e12d2f | 2198 | clear_current_oom_origin(); |
1da177e4 | 2199 | |
1da177e4 LT |
2200 | if (err) { |
2201 | /* re-insert swap space back into swap_list */ | |
cf0cac0a | 2202 | reinsert_swap_info(p); |
039939a6 | 2203 | reenable_swap_slots_cache_unlock(); |
1da177e4 LT |
2204 | goto out_dput; |
2205 | } | |
52b7efdb | 2206 | |
039939a6 TC |
2207 | reenable_swap_slots_cache_unlock(); |
2208 | ||
815c2c54 SL |
2209 | flush_work(&p->discard_work); |
2210 | ||
5d337b91 | 2211 | destroy_swap_extents(p); |
570a335b HD |
2212 | if (p->flags & SWP_CONTINUED) |
2213 | free_swap_count_continuations(p); | |
2214 | ||
fc0abb14 | 2215 | mutex_lock(&swapon_mutex); |
5d337b91 | 2216 | spin_lock(&swap_lock); |
ec8acf20 | 2217 | spin_lock(&p->lock); |
5d337b91 HD |
2218 | drain_mmlist(); |
2219 | ||
52b7efdb | 2220 | /* wait for anyone still in scan_swap_map */ |
52b7efdb HD |
2221 | p->highest_bit = 0; /* cuts scans short */ |
2222 | while (p->flags >= SWP_SCANNING) { | |
ec8acf20 | 2223 | spin_unlock(&p->lock); |
5d337b91 | 2224 | spin_unlock(&swap_lock); |
13e4b57f | 2225 | schedule_timeout_uninterruptible(1); |
5d337b91 | 2226 | spin_lock(&swap_lock); |
ec8acf20 | 2227 | spin_lock(&p->lock); |
52b7efdb | 2228 | } |
52b7efdb | 2229 | |
1da177e4 | 2230 | swap_file = p->swap_file; |
5b808a23 | 2231 | old_block_size = p->old_block_size; |
1da177e4 LT |
2232 | p->swap_file = NULL; |
2233 | p->max = 0; | |
2234 | swap_map = p->swap_map; | |
2235 | p->swap_map = NULL; | |
2a8f9449 SL |
2236 | cluster_info = p->cluster_info; |
2237 | p->cluster_info = NULL; | |
4f89849d | 2238 | frontswap_map = frontswap_map_get(p); |
ec8acf20 | 2239 | spin_unlock(&p->lock); |
5d337b91 | 2240 | spin_unlock(&swap_lock); |
adfab836 | 2241 | frontswap_invalidate_area(p->type); |
58e97ba6 | 2242 | frontswap_map_set(p, NULL); |
fc0abb14 | 2243 | mutex_unlock(&swapon_mutex); |
ebc2a1a6 SL |
2244 | free_percpu(p->percpu_cluster); |
2245 | p->percpu_cluster = NULL; | |
1da177e4 | 2246 | vfree(swap_map); |
2a8f9449 | 2247 | vfree(cluster_info); |
4f89849d | 2248 | vfree(frontswap_map); |
2de1a7e4 | 2249 | /* Destroy swap account information */ |
adfab836 | 2250 | swap_cgroup_swapoff(p->type); |
4b3ef9da | 2251 | exit_swap_address_space(p->type); |
27a7faa0 | 2252 | |
1da177e4 LT |
2253 | inode = mapping->host; |
2254 | if (S_ISBLK(inode->i_mode)) { | |
2255 | struct block_device *bdev = I_BDEV(inode); | |
5b808a23 | 2256 | set_blocksize(bdev, old_block_size); |
e525fd89 | 2257 | blkdev_put(bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL); |
1da177e4 | 2258 | } else { |
5955102c | 2259 | inode_lock(inode); |
1da177e4 | 2260 | inode->i_flags &= ~S_SWAPFILE; |
5955102c | 2261 | inode_unlock(inode); |
1da177e4 LT |
2262 | } |
2263 | filp_close(swap_file, NULL); | |
f893ab41 WY |
2264 | |
2265 | /* | |
2266 | * Clear the SWP_USED flag after all resources are freed so that swapon | |
2267 | * can reuse this swap_info in alloc_swap_info() safely. It is ok to | |
2268 | * not hold p->lock after we cleared its SWP_WRITEOK. | |
2269 | */ | |
2270 | spin_lock(&swap_lock); | |
2271 | p->flags = 0; | |
2272 | spin_unlock(&swap_lock); | |
2273 | ||
1da177e4 | 2274 | err = 0; |
66d7dd51 KS |
2275 | atomic_inc(&proc_poll_event); |
2276 | wake_up_interruptible(&proc_poll_wait); | |
1da177e4 LT |
2277 | |
2278 | out_dput: | |
2279 | filp_close(victim, NULL); | |
2280 | out: | |
f58b59c1 | 2281 | putname(pathname); |
1da177e4 LT |
2282 | return err; |
2283 | } | |
2284 | ||
2285 | #ifdef CONFIG_PROC_FS | |
66d7dd51 KS |
2286 | static unsigned swaps_poll(struct file *file, poll_table *wait) |
2287 | { | |
f1514638 | 2288 | struct seq_file *seq = file->private_data; |
66d7dd51 KS |
2289 | |
2290 | poll_wait(file, &proc_poll_wait, wait); | |
2291 | ||
f1514638 KS |
2292 | if (seq->poll_event != atomic_read(&proc_poll_event)) { |
2293 | seq->poll_event = atomic_read(&proc_poll_event); | |
66d7dd51 KS |
2294 | return POLLIN | POLLRDNORM | POLLERR | POLLPRI; |
2295 | } | |
2296 | ||
2297 | return POLLIN | POLLRDNORM; | |
2298 | } | |
2299 | ||
1da177e4 LT |
2300 | /* iterator */ |
2301 | static void *swap_start(struct seq_file *swap, loff_t *pos) | |
2302 | { | |
efa90a98 HD |
2303 | struct swap_info_struct *si; |
2304 | int type; | |
1da177e4 LT |
2305 | loff_t l = *pos; |
2306 | ||
fc0abb14 | 2307 | mutex_lock(&swapon_mutex); |
1da177e4 | 2308 | |
881e4aab SS |
2309 | if (!l) |
2310 | return SEQ_START_TOKEN; | |
2311 | ||
efa90a98 HD |
2312 | for (type = 0; type < nr_swapfiles; type++) { |
2313 | smp_rmb(); /* read nr_swapfiles before swap_info[type] */ | |
2314 | si = swap_info[type]; | |
2315 | if (!(si->flags & SWP_USED) || !si->swap_map) | |
1da177e4 | 2316 | continue; |
881e4aab | 2317 | if (!--l) |
efa90a98 | 2318 | return si; |
1da177e4 LT |
2319 | } |
2320 | ||
2321 | return NULL; | |
2322 | } | |
2323 | ||
2324 | static void *swap_next(struct seq_file *swap, void *v, loff_t *pos) | |
2325 | { | |
efa90a98 HD |
2326 | struct swap_info_struct *si = v; |
2327 | int type; | |
1da177e4 | 2328 | |
881e4aab | 2329 | if (v == SEQ_START_TOKEN) |
efa90a98 HD |
2330 | type = 0; |
2331 | else | |
2332 | type = si->type + 1; | |
881e4aab | 2333 | |
efa90a98 HD |
2334 | for (; type < nr_swapfiles; type++) { |
2335 | smp_rmb(); /* read nr_swapfiles before swap_info[type] */ | |
2336 | si = swap_info[type]; | |
2337 | if (!(si->flags & SWP_USED) || !si->swap_map) | |
1da177e4 LT |
2338 | continue; |
2339 | ++*pos; | |
efa90a98 | 2340 | return si; |
1da177e4 LT |
2341 | } |
2342 | ||
2343 | return NULL; | |
2344 | } | |
2345 | ||
2346 | static void swap_stop(struct seq_file *swap, void *v) | |
2347 | { | |
fc0abb14 | 2348 | mutex_unlock(&swapon_mutex); |
1da177e4 LT |
2349 | } |
2350 | ||
2351 | static int swap_show(struct seq_file *swap, void *v) | |
2352 | { | |
efa90a98 | 2353 | struct swap_info_struct *si = v; |
1da177e4 LT |
2354 | struct file *file; |
2355 | int len; | |
2356 | ||
efa90a98 | 2357 | if (si == SEQ_START_TOKEN) { |
881e4aab SS |
2358 | seq_puts(swap,"Filename\t\t\t\tType\t\tSize\tUsed\tPriority\n"); |
2359 | return 0; | |
2360 | } | |
1da177e4 | 2361 | |
efa90a98 | 2362 | file = si->swap_file; |
2726d566 | 2363 | len = seq_file_path(swap, file, " \t\n\\"); |
6eb396dc | 2364 | seq_printf(swap, "%*s%s\t%u\t%u\t%d\n", |
886bb7e9 | 2365 | len < 40 ? 40 - len : 1, " ", |
496ad9aa | 2366 | S_ISBLK(file_inode(file)->i_mode) ? |
1da177e4 | 2367 | "partition" : "file\t", |
efa90a98 HD |
2368 | si->pages << (PAGE_SHIFT - 10), |
2369 | si->inuse_pages << (PAGE_SHIFT - 10), | |
2370 | si->prio); | |
1da177e4 LT |
2371 | return 0; |
2372 | } | |
2373 | ||
15ad7cdc | 2374 | static const struct seq_operations swaps_op = { |
1da177e4 LT |
2375 | .start = swap_start, |
2376 | .next = swap_next, | |
2377 | .stop = swap_stop, | |
2378 | .show = swap_show | |
2379 | }; | |
2380 | ||
2381 | static int swaps_open(struct inode *inode, struct file *file) | |
2382 | { | |
f1514638 | 2383 | struct seq_file *seq; |
66d7dd51 KS |
2384 | int ret; |
2385 | ||
66d7dd51 | 2386 | ret = seq_open(file, &swaps_op); |
f1514638 | 2387 | if (ret) |
66d7dd51 | 2388 | return ret; |
66d7dd51 | 2389 | |
f1514638 KS |
2390 | seq = file->private_data; |
2391 | seq->poll_event = atomic_read(&proc_poll_event); | |
2392 | return 0; | |
1da177e4 LT |
2393 | } |
2394 | ||
15ad7cdc | 2395 | static const struct file_operations proc_swaps_operations = { |
1da177e4 LT |
2396 | .open = swaps_open, |
2397 | .read = seq_read, | |
2398 | .llseek = seq_lseek, | |
2399 | .release = seq_release, | |
66d7dd51 | 2400 | .poll = swaps_poll, |
1da177e4 LT |
2401 | }; |
2402 | ||
2403 | static int __init procswaps_init(void) | |
2404 | { | |
3d71f86f | 2405 | proc_create("swaps", 0, NULL, &proc_swaps_operations); |
1da177e4 LT |
2406 | return 0; |
2407 | } | |
2408 | __initcall(procswaps_init); | |
2409 | #endif /* CONFIG_PROC_FS */ | |
2410 | ||
1796316a JB |
2411 | #ifdef MAX_SWAPFILES_CHECK |
2412 | static int __init max_swapfiles_check(void) | |
2413 | { | |
2414 | MAX_SWAPFILES_CHECK(); | |
2415 | return 0; | |
2416 | } | |
2417 | late_initcall(max_swapfiles_check); | |
2418 | #endif | |
2419 | ||
53cbb243 | 2420 | static struct swap_info_struct *alloc_swap_info(void) |
1da177e4 | 2421 | { |
73c34b6a | 2422 | struct swap_info_struct *p; |
1da177e4 | 2423 | unsigned int type; |
efa90a98 HD |
2424 | |
2425 | p = kzalloc(sizeof(*p), GFP_KERNEL); | |
2426 | if (!p) | |
53cbb243 | 2427 | return ERR_PTR(-ENOMEM); |
efa90a98 | 2428 | |
5d337b91 | 2429 | spin_lock(&swap_lock); |
efa90a98 HD |
2430 | for (type = 0; type < nr_swapfiles; type++) { |
2431 | if (!(swap_info[type]->flags & SWP_USED)) | |
1da177e4 | 2432 | break; |
efa90a98 | 2433 | } |
0697212a | 2434 | if (type >= MAX_SWAPFILES) { |
5d337b91 | 2435 | spin_unlock(&swap_lock); |
efa90a98 | 2436 | kfree(p); |
730c0581 | 2437 | return ERR_PTR(-EPERM); |
1da177e4 | 2438 | } |
efa90a98 HD |
2439 | if (type >= nr_swapfiles) { |
2440 | p->type = type; | |
2441 | swap_info[type] = p; | |
2442 | /* | |
2443 | * Write swap_info[type] before nr_swapfiles, in case a | |
2444 | * racing procfs swap_start() or swap_next() is reading them. | |
2445 | * (We never shrink nr_swapfiles, we never free this entry.) | |
2446 | */ | |
2447 | smp_wmb(); | |
2448 | nr_swapfiles++; | |
2449 | } else { | |
2450 | kfree(p); | |
2451 | p = swap_info[type]; | |
2452 | /* | |
2453 | * Do not memset this entry: a racing procfs swap_next() | |
2454 | * would be relying on p->type to remain valid. | |
2455 | */ | |
2456 | } | |
9625a5f2 | 2457 | INIT_LIST_HEAD(&p->first_swap_extent.list); |
18ab4d4c DS |
2458 | plist_node_init(&p->list, 0); |
2459 | plist_node_init(&p->avail_list, 0); | |
1da177e4 | 2460 | p->flags = SWP_USED; |
5d337b91 | 2461 | spin_unlock(&swap_lock); |
ec8acf20 | 2462 | spin_lock_init(&p->lock); |
efa90a98 | 2463 | |
53cbb243 | 2464 | return p; |
53cbb243 CEB |
2465 | } |
2466 | ||
4d0e1e10 CEB |
2467 | static int claim_swapfile(struct swap_info_struct *p, struct inode *inode) |
2468 | { | |
2469 | int error; | |
2470 | ||
2471 | if (S_ISBLK(inode->i_mode)) { | |
2472 | p->bdev = bdgrab(I_BDEV(inode)); | |
2473 | error = blkdev_get(p->bdev, | |
6f179af8 | 2474 | FMODE_READ | FMODE_WRITE | FMODE_EXCL, p); |
4d0e1e10 CEB |
2475 | if (error < 0) { |
2476 | p->bdev = NULL; | |
6f179af8 | 2477 | return error; |
4d0e1e10 CEB |
2478 | } |
2479 | p->old_block_size = block_size(p->bdev); | |
2480 | error = set_blocksize(p->bdev, PAGE_SIZE); | |
2481 | if (error < 0) | |
87ade72a | 2482 | return error; |
4d0e1e10 CEB |
2483 | p->flags |= SWP_BLKDEV; |
2484 | } else if (S_ISREG(inode->i_mode)) { | |
2485 | p->bdev = inode->i_sb->s_bdev; | |
5955102c | 2486 | inode_lock(inode); |
87ade72a CEB |
2487 | if (IS_SWAPFILE(inode)) |
2488 | return -EBUSY; | |
2489 | } else | |
2490 | return -EINVAL; | |
4d0e1e10 CEB |
2491 | |
2492 | return 0; | |
4d0e1e10 CEB |
2493 | } |
2494 | ||
ca8bd38b CEB |
2495 | static unsigned long read_swap_header(struct swap_info_struct *p, |
2496 | union swap_header *swap_header, | |
2497 | struct inode *inode) | |
2498 | { | |
2499 | int i; | |
2500 | unsigned long maxpages; | |
2501 | unsigned long swapfilepages; | |
d6bbbd29 | 2502 | unsigned long last_page; |
ca8bd38b CEB |
2503 | |
2504 | if (memcmp("SWAPSPACE2", swap_header->magic.magic, 10)) { | |
465c47fd | 2505 | pr_err("Unable to find swap-space signature\n"); |
38719025 | 2506 | return 0; |
ca8bd38b CEB |
2507 | } |
2508 | ||
2509 | /* swap partition endianess hack... */ | |
2510 | if (swab32(swap_header->info.version) == 1) { | |
2511 | swab32s(&swap_header->info.version); | |
2512 | swab32s(&swap_header->info.last_page); | |
2513 | swab32s(&swap_header->info.nr_badpages); | |
dd111be6 JH |
2514 | if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES) |
2515 | return 0; | |
ca8bd38b CEB |
2516 | for (i = 0; i < swap_header->info.nr_badpages; i++) |
2517 | swab32s(&swap_header->info.badpages[i]); | |
2518 | } | |
2519 | /* Check the swap header's sub-version */ | |
2520 | if (swap_header->info.version != 1) { | |
465c47fd AM |
2521 | pr_warn("Unable to handle swap header version %d\n", |
2522 | swap_header->info.version); | |
38719025 | 2523 | return 0; |
ca8bd38b CEB |
2524 | } |
2525 | ||
2526 | p->lowest_bit = 1; | |
2527 | p->cluster_next = 1; | |
2528 | p->cluster_nr = 0; | |
2529 | ||
2530 | /* | |
2531 | * Find out how many pages are allowed for a single swap | |
9b15b817 | 2532 | * device. There are two limiting factors: 1) the number |
a2c16d6c HD |
2533 | * of bits for the swap offset in the swp_entry_t type, and |
2534 | * 2) the number of bits in the swap pte as defined by the | |
9b15b817 | 2535 | * different architectures. In order to find the |
a2c16d6c | 2536 | * largest possible bit mask, a swap entry with swap type 0 |
ca8bd38b | 2537 | * and swap offset ~0UL is created, encoded to a swap pte, |
a2c16d6c | 2538 | * decoded to a swp_entry_t again, and finally the swap |
ca8bd38b CEB |
2539 | * offset is extracted. This will mask all the bits from |
2540 | * the initial ~0UL mask that can't be encoded in either | |
2541 | * the swp_entry_t or the architecture definition of a | |
9b15b817 | 2542 | * swap pte. |
ca8bd38b CEB |
2543 | */ |
2544 | maxpages = swp_offset(pte_to_swp_entry( | |
9b15b817 | 2545 | swp_entry_to_pte(swp_entry(0, ~0UL)))) + 1; |
d6bbbd29 RJ |
2546 | last_page = swap_header->info.last_page; |
2547 | if (last_page > maxpages) { | |
465c47fd | 2548 | pr_warn("Truncating oversized swap area, only using %luk out of %luk\n", |
d6bbbd29 RJ |
2549 | maxpages << (PAGE_SHIFT - 10), |
2550 | last_page << (PAGE_SHIFT - 10)); | |
2551 | } | |
2552 | if (maxpages > last_page) { | |
2553 | maxpages = last_page + 1; | |
ca8bd38b CEB |
2554 | /* p->max is an unsigned int: don't overflow it */ |
2555 | if ((unsigned int)maxpages == 0) | |
2556 | maxpages = UINT_MAX; | |
2557 | } | |
2558 | p->highest_bit = maxpages - 1; | |
2559 | ||
2560 | if (!maxpages) | |
38719025 | 2561 | return 0; |
ca8bd38b CEB |
2562 | swapfilepages = i_size_read(inode) >> PAGE_SHIFT; |
2563 | if (swapfilepages && maxpages > swapfilepages) { | |
465c47fd | 2564 | pr_warn("Swap area shorter than signature indicates\n"); |
38719025 | 2565 | return 0; |
ca8bd38b CEB |
2566 | } |
2567 | if (swap_header->info.nr_badpages && S_ISREG(inode->i_mode)) | |
38719025 | 2568 | return 0; |
ca8bd38b | 2569 | if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES) |
38719025 | 2570 | return 0; |
ca8bd38b CEB |
2571 | |
2572 | return maxpages; | |
ca8bd38b CEB |
2573 | } |
2574 | ||
4b3ef9da | 2575 | #define SWAP_CLUSTER_INFO_COLS \ |
235b6217 | 2576 | DIV_ROUND_UP(L1_CACHE_BYTES, sizeof(struct swap_cluster_info)) |
4b3ef9da HY |
2577 | #define SWAP_CLUSTER_SPACE_COLS \ |
2578 | DIV_ROUND_UP(SWAP_ADDRESS_SPACE_PAGES, SWAPFILE_CLUSTER) | |
2579 | #define SWAP_CLUSTER_COLS \ | |
2580 | max_t(unsigned int, SWAP_CLUSTER_INFO_COLS, SWAP_CLUSTER_SPACE_COLS) | |
235b6217 | 2581 | |
915d4d7b CEB |
2582 | static int setup_swap_map_and_extents(struct swap_info_struct *p, |
2583 | union swap_header *swap_header, | |
2584 | unsigned char *swap_map, | |
2a8f9449 | 2585 | struct swap_cluster_info *cluster_info, |
915d4d7b CEB |
2586 | unsigned long maxpages, |
2587 | sector_t *span) | |
2588 | { | |
235b6217 | 2589 | unsigned int j, k; |
915d4d7b CEB |
2590 | unsigned int nr_good_pages; |
2591 | int nr_extents; | |
2a8f9449 | 2592 | unsigned long nr_clusters = DIV_ROUND_UP(maxpages, SWAPFILE_CLUSTER); |
235b6217 HY |
2593 | unsigned long col = p->cluster_next / SWAPFILE_CLUSTER % SWAP_CLUSTER_COLS; |
2594 | unsigned long i, idx; | |
915d4d7b CEB |
2595 | |
2596 | nr_good_pages = maxpages - 1; /* omit header page */ | |
2597 | ||
6b534915 HY |
2598 | cluster_list_init(&p->free_clusters); |
2599 | cluster_list_init(&p->discard_clusters); | |
2a8f9449 | 2600 | |
915d4d7b CEB |
2601 | for (i = 0; i < swap_header->info.nr_badpages; i++) { |
2602 | unsigned int page_nr = swap_header->info.badpages[i]; | |
bdb8e3f6 CEB |
2603 | if (page_nr == 0 || page_nr > swap_header->info.last_page) |
2604 | return -EINVAL; | |
915d4d7b CEB |
2605 | if (page_nr < maxpages) { |
2606 | swap_map[page_nr] = SWAP_MAP_BAD; | |
2607 | nr_good_pages--; | |
2a8f9449 SL |
2608 | /* |
2609 | * Haven't marked the cluster free yet, no list | |
2610 | * operation involved | |
2611 | */ | |
2612 | inc_cluster_info_page(p, cluster_info, page_nr); | |
915d4d7b CEB |
2613 | } |
2614 | } | |
2615 | ||
2a8f9449 SL |
2616 | /* Haven't marked the cluster free yet, no list operation involved */ |
2617 | for (i = maxpages; i < round_up(maxpages, SWAPFILE_CLUSTER); i++) | |
2618 | inc_cluster_info_page(p, cluster_info, i); | |
2619 | ||
915d4d7b CEB |
2620 | if (nr_good_pages) { |
2621 | swap_map[0] = SWAP_MAP_BAD; | |
2a8f9449 SL |
2622 | /* |
2623 | * Not mark the cluster free yet, no list | |
2624 | * operation involved | |
2625 | */ | |
2626 | inc_cluster_info_page(p, cluster_info, 0); | |
915d4d7b CEB |
2627 | p->max = maxpages; |
2628 | p->pages = nr_good_pages; | |
2629 | nr_extents = setup_swap_extents(p, span); | |
bdb8e3f6 CEB |
2630 | if (nr_extents < 0) |
2631 | return nr_extents; | |
915d4d7b CEB |
2632 | nr_good_pages = p->pages; |
2633 | } | |
2634 | if (!nr_good_pages) { | |
465c47fd | 2635 | pr_warn("Empty swap-file\n"); |
bdb8e3f6 | 2636 | return -EINVAL; |
915d4d7b CEB |
2637 | } |
2638 | ||
2a8f9449 SL |
2639 | if (!cluster_info) |
2640 | return nr_extents; | |
2641 | ||
235b6217 | 2642 | |
4b3ef9da HY |
2643 | /* |
2644 | * Reduce false cache line sharing between cluster_info and | |
2645 | * sharing same address space. | |
2646 | */ | |
235b6217 HY |
2647 | for (k = 0; k < SWAP_CLUSTER_COLS; k++) { |
2648 | j = (k + col) % SWAP_CLUSTER_COLS; | |
2649 | for (i = 0; i < DIV_ROUND_UP(nr_clusters, SWAP_CLUSTER_COLS); i++) { | |
2650 | idx = i * SWAP_CLUSTER_COLS + j; | |
2651 | if (idx >= nr_clusters) | |
2652 | continue; | |
2653 | if (cluster_count(&cluster_info[idx])) | |
2654 | continue; | |
2a8f9449 | 2655 | cluster_set_flag(&cluster_info[idx], CLUSTER_FLAG_FREE); |
6b534915 HY |
2656 | cluster_list_add_tail(&p->free_clusters, cluster_info, |
2657 | idx); | |
2a8f9449 | 2658 | } |
2a8f9449 | 2659 | } |
915d4d7b | 2660 | return nr_extents; |
915d4d7b CEB |
2661 | } |
2662 | ||
dcf6b7dd RA |
2663 | /* |
2664 | * Helper to sys_swapon determining if a given swap | |
2665 | * backing device queue supports DISCARD operations. | |
2666 | */ | |
2667 | static bool swap_discardable(struct swap_info_struct *si) | |
2668 | { | |
2669 | struct request_queue *q = bdev_get_queue(si->bdev); | |
2670 | ||
2671 | if (!q || !blk_queue_discard(q)) | |
2672 | return false; | |
2673 | ||
2674 | return true; | |
2675 | } | |
2676 | ||
53cbb243 CEB |
2677 | SYSCALL_DEFINE2(swapon, const char __user *, specialfile, int, swap_flags) |
2678 | { | |
2679 | struct swap_info_struct *p; | |
91a27b2a | 2680 | struct filename *name; |
53cbb243 CEB |
2681 | struct file *swap_file = NULL; |
2682 | struct address_space *mapping; | |
40531542 | 2683 | int prio; |
53cbb243 CEB |
2684 | int error; |
2685 | union swap_header *swap_header; | |
915d4d7b | 2686 | int nr_extents; |
53cbb243 CEB |
2687 | sector_t span; |
2688 | unsigned long maxpages; | |
53cbb243 | 2689 | unsigned char *swap_map = NULL; |
2a8f9449 | 2690 | struct swap_cluster_info *cluster_info = NULL; |
38b5faf4 | 2691 | unsigned long *frontswap_map = NULL; |
53cbb243 CEB |
2692 | struct page *page = NULL; |
2693 | struct inode *inode = NULL; | |
53cbb243 | 2694 | |
d15cab97 HD |
2695 | if (swap_flags & ~SWAP_FLAGS_VALID) |
2696 | return -EINVAL; | |
2697 | ||
53cbb243 CEB |
2698 | if (!capable(CAP_SYS_ADMIN)) |
2699 | return -EPERM; | |
2700 | ||
2701 | p = alloc_swap_info(); | |
2542e513 CEB |
2702 | if (IS_ERR(p)) |
2703 | return PTR_ERR(p); | |
53cbb243 | 2704 | |
815c2c54 SL |
2705 | INIT_WORK(&p->discard_work, swap_discard_work); |
2706 | ||
1da177e4 | 2707 | name = getname(specialfile); |
1da177e4 | 2708 | if (IS_ERR(name)) { |
7de7fb6b | 2709 | error = PTR_ERR(name); |
1da177e4 | 2710 | name = NULL; |
bd69010b | 2711 | goto bad_swap; |
1da177e4 | 2712 | } |
669abf4e | 2713 | swap_file = file_open_name(name, O_RDWR|O_LARGEFILE, 0); |
1da177e4 | 2714 | if (IS_ERR(swap_file)) { |
7de7fb6b | 2715 | error = PTR_ERR(swap_file); |
1da177e4 | 2716 | swap_file = NULL; |
bd69010b | 2717 | goto bad_swap; |
1da177e4 LT |
2718 | } |
2719 | ||
2720 | p->swap_file = swap_file; | |
2721 | mapping = swap_file->f_mapping; | |
2130781e | 2722 | inode = mapping->host; |
6f179af8 | 2723 | |
5955102c | 2724 | /* If S_ISREG(inode->i_mode) will do inode_lock(inode); */ |
4d0e1e10 CEB |
2725 | error = claim_swapfile(p, inode); |
2726 | if (unlikely(error)) | |
1da177e4 | 2727 | goto bad_swap; |
1da177e4 | 2728 | |
1da177e4 LT |
2729 | /* |
2730 | * Read the swap header. | |
2731 | */ | |
2732 | if (!mapping->a_ops->readpage) { | |
2733 | error = -EINVAL; | |
2734 | goto bad_swap; | |
2735 | } | |
090d2b18 | 2736 | page = read_mapping_page(mapping, 0, swap_file); |
1da177e4 LT |
2737 | if (IS_ERR(page)) { |
2738 | error = PTR_ERR(page); | |
2739 | goto bad_swap; | |
2740 | } | |
81e33971 | 2741 | swap_header = kmap(page); |
1da177e4 | 2742 | |
ca8bd38b CEB |
2743 | maxpages = read_swap_header(p, swap_header, inode); |
2744 | if (unlikely(!maxpages)) { | |
1da177e4 LT |
2745 | error = -EINVAL; |
2746 | goto bad_swap; | |
2747 | } | |
886bb7e9 | 2748 | |
81e33971 | 2749 | /* OK, set up the swap map and apply the bad block list */ |
803d0c83 | 2750 | swap_map = vzalloc(maxpages); |
81e33971 HD |
2751 | if (!swap_map) { |
2752 | error = -ENOMEM; | |
2753 | goto bad_swap; | |
2754 | } | |
f0571429 MK |
2755 | |
2756 | if (bdi_cap_stable_pages_required(inode_to_bdi(inode))) | |
2757 | p->flags |= SWP_STABLE_WRITES; | |
2758 | ||
2a8f9449 | 2759 | if (p->bdev && blk_queue_nonrot(bdev_get_queue(p->bdev))) { |
6f179af8 | 2760 | int cpu; |
235b6217 | 2761 | unsigned long ci, nr_cluster; |
6f179af8 | 2762 | |
2a8f9449 SL |
2763 | p->flags |= SWP_SOLIDSTATE; |
2764 | /* | |
2765 | * select a random position to start with to help wear leveling | |
2766 | * SSD | |
2767 | */ | |
2768 | p->cluster_next = 1 + (prandom_u32() % p->highest_bit); | |
235b6217 | 2769 | nr_cluster = DIV_ROUND_UP(maxpages, SWAPFILE_CLUSTER); |
2a8f9449 | 2770 | |
235b6217 | 2771 | cluster_info = vzalloc(nr_cluster * sizeof(*cluster_info)); |
2a8f9449 SL |
2772 | if (!cluster_info) { |
2773 | error = -ENOMEM; | |
2774 | goto bad_swap; | |
2775 | } | |
235b6217 HY |
2776 | |
2777 | for (ci = 0; ci < nr_cluster; ci++) | |
2778 | spin_lock_init(&((cluster_info + ci)->lock)); | |
2779 | ||
ebc2a1a6 SL |
2780 | p->percpu_cluster = alloc_percpu(struct percpu_cluster); |
2781 | if (!p->percpu_cluster) { | |
2782 | error = -ENOMEM; | |
2783 | goto bad_swap; | |
2784 | } | |
6f179af8 | 2785 | for_each_possible_cpu(cpu) { |
ebc2a1a6 | 2786 | struct percpu_cluster *cluster; |
6f179af8 | 2787 | cluster = per_cpu_ptr(p->percpu_cluster, cpu); |
ebc2a1a6 SL |
2788 | cluster_set_null(&cluster->index); |
2789 | } | |
2a8f9449 | 2790 | } |
1da177e4 | 2791 | |
1421ef3c CEB |
2792 | error = swap_cgroup_swapon(p->type, maxpages); |
2793 | if (error) | |
2794 | goto bad_swap; | |
2795 | ||
915d4d7b | 2796 | nr_extents = setup_swap_map_and_extents(p, swap_header, swap_map, |
2a8f9449 | 2797 | cluster_info, maxpages, &span); |
915d4d7b CEB |
2798 | if (unlikely(nr_extents < 0)) { |
2799 | error = nr_extents; | |
1da177e4 LT |
2800 | goto bad_swap; |
2801 | } | |
38b5faf4 | 2802 | /* frontswap enabled? set up bit-per-page map for frontswap */ |
8ea1d2a1 | 2803 | if (IS_ENABLED(CONFIG_FRONTSWAP)) |
7b57976d | 2804 | frontswap_map = vzalloc(BITS_TO_LONGS(maxpages) * sizeof(long)); |
1da177e4 | 2805 | |
2a8f9449 SL |
2806 | if (p->bdev &&(swap_flags & SWAP_FLAG_DISCARD) && swap_discardable(p)) { |
2807 | /* | |
2808 | * When discard is enabled for swap with no particular | |
2809 | * policy flagged, we set all swap discard flags here in | |
2810 | * order to sustain backward compatibility with older | |
2811 | * swapon(8) releases. | |
2812 | */ | |
2813 | p->flags |= (SWP_DISCARDABLE | SWP_AREA_DISCARD | | |
2814 | SWP_PAGE_DISCARD); | |
dcf6b7dd | 2815 | |
2a8f9449 SL |
2816 | /* |
2817 | * By flagging sys_swapon, a sysadmin can tell us to | |
2818 | * either do single-time area discards only, or to just | |
2819 | * perform discards for released swap page-clusters. | |
2820 | * Now it's time to adjust the p->flags accordingly. | |
2821 | */ | |
2822 | if (swap_flags & SWAP_FLAG_DISCARD_ONCE) | |
2823 | p->flags &= ~SWP_PAGE_DISCARD; | |
2824 | else if (swap_flags & SWAP_FLAG_DISCARD_PAGES) | |
2825 | p->flags &= ~SWP_AREA_DISCARD; | |
2826 | ||
2827 | /* issue a swapon-time discard if it's still required */ | |
2828 | if (p->flags & SWP_AREA_DISCARD) { | |
2829 | int err = discard_swap(p); | |
2830 | if (unlikely(err)) | |
2831 | pr_err("swapon: discard_swap(%p): %d\n", | |
2832 | p, err); | |
dcf6b7dd | 2833 | } |
20137a49 | 2834 | } |
6a6ba831 | 2835 | |
4b3ef9da HY |
2836 | error = init_swap_address_space(p->type, maxpages); |
2837 | if (error) | |
2838 | goto bad_swap; | |
2839 | ||
fc0abb14 | 2840 | mutex_lock(&swapon_mutex); |
40531542 | 2841 | prio = -1; |
78ecba08 | 2842 | if (swap_flags & SWAP_FLAG_PREFER) |
40531542 | 2843 | prio = |
78ecba08 | 2844 | (swap_flags & SWAP_FLAG_PRIO_MASK) >> SWAP_FLAG_PRIO_SHIFT; |
2a8f9449 | 2845 | enable_swap_info(p, prio, swap_map, cluster_info, frontswap_map); |
c69dbfb8 | 2846 | |
756a025f | 2847 | pr_info("Adding %uk swap on %s. Priority:%d extents:%d across:%lluk %s%s%s%s%s\n", |
91a27b2a | 2848 | p->pages<<(PAGE_SHIFT-10), name->name, p->prio, |
c69dbfb8 CEB |
2849 | nr_extents, (unsigned long long)span<<(PAGE_SHIFT-10), |
2850 | (p->flags & SWP_SOLIDSTATE) ? "SS" : "", | |
38b5faf4 | 2851 | (p->flags & SWP_DISCARDABLE) ? "D" : "", |
dcf6b7dd RA |
2852 | (p->flags & SWP_AREA_DISCARD) ? "s" : "", |
2853 | (p->flags & SWP_PAGE_DISCARD) ? "c" : "", | |
38b5faf4 | 2854 | (frontswap_map) ? "FS" : ""); |
c69dbfb8 | 2855 | |
fc0abb14 | 2856 | mutex_unlock(&swapon_mutex); |
66d7dd51 KS |
2857 | atomic_inc(&proc_poll_event); |
2858 | wake_up_interruptible(&proc_poll_wait); | |
2859 | ||
9b01c350 CEB |
2860 | if (S_ISREG(inode->i_mode)) |
2861 | inode->i_flags |= S_SWAPFILE; | |
1da177e4 LT |
2862 | error = 0; |
2863 | goto out; | |
2864 | bad_swap: | |
ebc2a1a6 SL |
2865 | free_percpu(p->percpu_cluster); |
2866 | p->percpu_cluster = NULL; | |
bd69010b | 2867 | if (inode && S_ISBLK(inode->i_mode) && p->bdev) { |
f2090d2d CEB |
2868 | set_blocksize(p->bdev, p->old_block_size); |
2869 | blkdev_put(p->bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL); | |
1da177e4 | 2870 | } |
4cd3bb10 | 2871 | destroy_swap_extents(p); |
e8e6c2ec | 2872 | swap_cgroup_swapoff(p->type); |
5d337b91 | 2873 | spin_lock(&swap_lock); |
1da177e4 | 2874 | p->swap_file = NULL; |
1da177e4 | 2875 | p->flags = 0; |
5d337b91 | 2876 | spin_unlock(&swap_lock); |
1da177e4 | 2877 | vfree(swap_map); |
2a8f9449 | 2878 | vfree(cluster_info); |
52c50567 | 2879 | if (swap_file) { |
2130781e | 2880 | if (inode && S_ISREG(inode->i_mode)) { |
5955102c | 2881 | inode_unlock(inode); |
2130781e CEB |
2882 | inode = NULL; |
2883 | } | |
1da177e4 | 2884 | filp_close(swap_file, NULL); |
52c50567 | 2885 | } |
1da177e4 LT |
2886 | out: |
2887 | if (page && !IS_ERR(page)) { | |
2888 | kunmap(page); | |
09cbfeaf | 2889 | put_page(page); |
1da177e4 LT |
2890 | } |
2891 | if (name) | |
2892 | putname(name); | |
9b01c350 | 2893 | if (inode && S_ISREG(inode->i_mode)) |
5955102c | 2894 | inode_unlock(inode); |
039939a6 TC |
2895 | if (!error) |
2896 | enable_swap_slots_cache(); | |
1da177e4 LT |
2897 | return error; |
2898 | } | |
2899 | ||
2900 | void si_swapinfo(struct sysinfo *val) | |
2901 | { | |
efa90a98 | 2902 | unsigned int type; |
1da177e4 LT |
2903 | unsigned long nr_to_be_unused = 0; |
2904 | ||
5d337b91 | 2905 | spin_lock(&swap_lock); |
efa90a98 HD |
2906 | for (type = 0; type < nr_swapfiles; type++) { |
2907 | struct swap_info_struct *si = swap_info[type]; | |
2908 | ||
2909 | if ((si->flags & SWP_USED) && !(si->flags & SWP_WRITEOK)) | |
2910 | nr_to_be_unused += si->inuse_pages; | |
1da177e4 | 2911 | } |
ec8acf20 | 2912 | val->freeswap = atomic_long_read(&nr_swap_pages) + nr_to_be_unused; |
1da177e4 | 2913 | val->totalswap = total_swap_pages + nr_to_be_unused; |
5d337b91 | 2914 | spin_unlock(&swap_lock); |
1da177e4 LT |
2915 | } |
2916 | ||
2917 | /* | |
2918 | * Verify that a swap entry is valid and increment its swap map count. | |
2919 | * | |
355cfa73 KH |
2920 | * Returns error code in following case. |
2921 | * - success -> 0 | |
2922 | * - swp_entry is invalid -> EINVAL | |
2923 | * - swp_entry is migration entry -> EINVAL | |
2924 | * - swap-cache reference is requested but there is already one. -> EEXIST | |
2925 | * - swap-cache reference is requested but the entry is not used. -> ENOENT | |
570a335b | 2926 | * - swap-mapped reference requested but needs continued swap count. -> ENOMEM |
1da177e4 | 2927 | */ |
8d69aaee | 2928 | static int __swap_duplicate(swp_entry_t entry, unsigned char usage) |
1da177e4 | 2929 | { |
73c34b6a | 2930 | struct swap_info_struct *p; |
235b6217 | 2931 | struct swap_cluster_info *ci; |
1da177e4 | 2932 | unsigned long offset, type; |
8d69aaee HD |
2933 | unsigned char count; |
2934 | unsigned char has_cache; | |
253d553b | 2935 | int err = -EINVAL; |
1da177e4 | 2936 | |
a7420aa5 | 2937 | if (non_swap_entry(entry)) |
253d553b | 2938 | goto out; |
0697212a | 2939 | |
1da177e4 LT |
2940 | type = swp_type(entry); |
2941 | if (type >= nr_swapfiles) | |
2942 | goto bad_file; | |
efa90a98 | 2943 | p = swap_info[type]; |
1da177e4 | 2944 | offset = swp_offset(entry); |
355cfa73 | 2945 | if (unlikely(offset >= p->max)) |
235b6217 HY |
2946 | goto out; |
2947 | ||
2948 | ci = lock_cluster_or_swap_info(p, offset); | |
355cfa73 | 2949 | |
253d553b | 2950 | count = p->swap_map[offset]; |
edfe23da SL |
2951 | |
2952 | /* | |
2953 | * swapin_readahead() doesn't check if a swap entry is valid, so the | |
2954 | * swap entry could be SWAP_MAP_BAD. Check here with lock held. | |
2955 | */ | |
2956 | if (unlikely(swap_count(count) == SWAP_MAP_BAD)) { | |
2957 | err = -ENOENT; | |
2958 | goto unlock_out; | |
2959 | } | |
2960 | ||
253d553b HD |
2961 | has_cache = count & SWAP_HAS_CACHE; |
2962 | count &= ~SWAP_HAS_CACHE; | |
2963 | err = 0; | |
355cfa73 | 2964 | |
253d553b | 2965 | if (usage == SWAP_HAS_CACHE) { |
355cfa73 KH |
2966 | |
2967 | /* set SWAP_HAS_CACHE if there is no cache and entry is used */ | |
253d553b HD |
2968 | if (!has_cache && count) |
2969 | has_cache = SWAP_HAS_CACHE; | |
2970 | else if (has_cache) /* someone else added cache */ | |
2971 | err = -EEXIST; | |
2972 | else /* no users remaining */ | |
2973 | err = -ENOENT; | |
355cfa73 KH |
2974 | |
2975 | } else if (count || has_cache) { | |
253d553b | 2976 | |
570a335b HD |
2977 | if ((count & ~COUNT_CONTINUED) < SWAP_MAP_MAX) |
2978 | count += usage; | |
2979 | else if ((count & ~COUNT_CONTINUED) > SWAP_MAP_MAX) | |
253d553b | 2980 | err = -EINVAL; |
570a335b HD |
2981 | else if (swap_count_continued(p, offset, count)) |
2982 | count = COUNT_CONTINUED; | |
2983 | else | |
2984 | err = -ENOMEM; | |
355cfa73 | 2985 | } else |
253d553b HD |
2986 | err = -ENOENT; /* unused swap entry */ |
2987 | ||
2988 | p->swap_map[offset] = count | has_cache; | |
2989 | ||
355cfa73 | 2990 | unlock_out: |
235b6217 | 2991 | unlock_cluster_or_swap_info(p, ci); |
1da177e4 | 2992 | out: |
253d553b | 2993 | return err; |
1da177e4 LT |
2994 | |
2995 | bad_file: | |
465c47fd | 2996 | pr_err("swap_dup: %s%08lx\n", Bad_file, entry.val); |
1da177e4 LT |
2997 | goto out; |
2998 | } | |
253d553b | 2999 | |
aaa46865 HD |
3000 | /* |
3001 | * Help swapoff by noting that swap entry belongs to shmem/tmpfs | |
3002 | * (in which case its reference count is never incremented). | |
3003 | */ | |
3004 | void swap_shmem_alloc(swp_entry_t entry) | |
3005 | { | |
3006 | __swap_duplicate(entry, SWAP_MAP_SHMEM); | |
3007 | } | |
3008 | ||
355cfa73 | 3009 | /* |
08259d58 HD |
3010 | * Increase reference count of swap entry by 1. |
3011 | * Returns 0 for success, or -ENOMEM if a swap_count_continuation is required | |
3012 | * but could not be atomically allocated. Returns 0, just as if it succeeded, | |
3013 | * if __swap_duplicate() fails for another reason (-EINVAL or -ENOENT), which | |
3014 | * might occur if a page table entry has got corrupted. | |
355cfa73 | 3015 | */ |
570a335b | 3016 | int swap_duplicate(swp_entry_t entry) |
355cfa73 | 3017 | { |
570a335b HD |
3018 | int err = 0; |
3019 | ||
3020 | while (!err && __swap_duplicate(entry, 1) == -ENOMEM) | |
3021 | err = add_swap_count_continuation(entry, GFP_ATOMIC); | |
3022 | return err; | |
355cfa73 | 3023 | } |
1da177e4 | 3024 | |
cb4b86ba | 3025 | /* |
355cfa73 KH |
3026 | * @entry: swap entry for which we allocate swap cache. |
3027 | * | |
73c34b6a | 3028 | * Called when allocating swap cache for existing swap entry, |
355cfa73 KH |
3029 | * This can return error codes. Returns 0 at success. |
3030 | * -EBUSY means there is a swap cache. | |
3031 | * Note: return code is different from swap_duplicate(). | |
cb4b86ba KH |
3032 | */ |
3033 | int swapcache_prepare(swp_entry_t entry) | |
3034 | { | |
253d553b | 3035 | return __swap_duplicate(entry, SWAP_HAS_CACHE); |
cb4b86ba KH |
3036 | } |
3037 | ||
f981c595 MG |
3038 | struct swap_info_struct *page_swap_info(struct page *page) |
3039 | { | |
3040 | swp_entry_t swap = { .val = page_private(page) }; | |
f981c595 MG |
3041 | return swap_info[swp_type(swap)]; |
3042 | } | |
3043 | ||
3044 | /* | |
3045 | * out-of-line __page_file_ methods to avoid include hell. | |
3046 | */ | |
3047 | struct address_space *__page_file_mapping(struct page *page) | |
3048 | { | |
309381fe | 3049 | VM_BUG_ON_PAGE(!PageSwapCache(page), page); |
f981c595 MG |
3050 | return page_swap_info(page)->swap_file->f_mapping; |
3051 | } | |
3052 | EXPORT_SYMBOL_GPL(__page_file_mapping); | |
3053 | ||
3054 | pgoff_t __page_file_index(struct page *page) | |
3055 | { | |
3056 | swp_entry_t swap = { .val = page_private(page) }; | |
309381fe | 3057 | VM_BUG_ON_PAGE(!PageSwapCache(page), page); |
f981c595 MG |
3058 | return swp_offset(swap); |
3059 | } | |
3060 | EXPORT_SYMBOL_GPL(__page_file_index); | |
3061 | ||
570a335b HD |
3062 | /* |
3063 | * add_swap_count_continuation - called when a swap count is duplicated | |
3064 | * beyond SWAP_MAP_MAX, it allocates a new page and links that to the entry's | |
3065 | * page of the original vmalloc'ed swap_map, to hold the continuation count | |
3066 | * (for that entry and for its neighbouring PAGE_SIZE swap entries). Called | |
3067 | * again when count is duplicated beyond SWAP_MAP_MAX * SWAP_CONT_MAX, etc. | |
3068 | * | |
3069 | * These continuation pages are seldom referenced: the common paths all work | |
3070 | * on the original swap_map, only referring to a continuation page when the | |
3071 | * low "digit" of a count is incremented or decremented through SWAP_MAP_MAX. | |
3072 | * | |
3073 | * add_swap_count_continuation(, GFP_ATOMIC) can be called while holding | |
3074 | * page table locks; if it fails, add_swap_count_continuation(, GFP_KERNEL) | |
3075 | * can be called after dropping locks. | |
3076 | */ | |
3077 | int add_swap_count_continuation(swp_entry_t entry, gfp_t gfp_mask) | |
3078 | { | |
3079 | struct swap_info_struct *si; | |
235b6217 | 3080 | struct swap_cluster_info *ci; |
570a335b HD |
3081 | struct page *head; |
3082 | struct page *page; | |
3083 | struct page *list_page; | |
3084 | pgoff_t offset; | |
3085 | unsigned char count; | |
3086 | ||
3087 | /* | |
3088 | * When debugging, it's easier to use __GFP_ZERO here; but it's better | |
3089 | * for latency not to zero a page while GFP_ATOMIC and holding locks. | |
3090 | */ | |
3091 | page = alloc_page(gfp_mask | __GFP_HIGHMEM); | |
3092 | ||
3093 | si = swap_info_get(entry); | |
3094 | if (!si) { | |
3095 | /* | |
3096 | * An acceptable race has occurred since the failing | |
3097 | * __swap_duplicate(): the swap entry has been freed, | |
3098 | * perhaps even the whole swap_map cleared for swapoff. | |
3099 | */ | |
3100 | goto outer; | |
3101 | } | |
3102 | ||
3103 | offset = swp_offset(entry); | |
235b6217 HY |
3104 | |
3105 | ci = lock_cluster(si, offset); | |
3106 | ||
570a335b HD |
3107 | count = si->swap_map[offset] & ~SWAP_HAS_CACHE; |
3108 | ||
3109 | if ((count & ~COUNT_CONTINUED) != SWAP_MAP_MAX) { | |
3110 | /* | |
3111 | * The higher the swap count, the more likely it is that tasks | |
3112 | * will race to add swap count continuation: we need to avoid | |
3113 | * over-provisioning. | |
3114 | */ | |
3115 | goto out; | |
3116 | } | |
3117 | ||
3118 | if (!page) { | |
235b6217 | 3119 | unlock_cluster(ci); |
ec8acf20 | 3120 | spin_unlock(&si->lock); |
570a335b HD |
3121 | return -ENOMEM; |
3122 | } | |
3123 | ||
3124 | /* | |
3125 | * We are fortunate that although vmalloc_to_page uses pte_offset_map, | |
2de1a7e4 SJ |
3126 | * no architecture is using highmem pages for kernel page tables: so it |
3127 | * will not corrupt the GFP_ATOMIC caller's atomic page table kmaps. | |
570a335b HD |
3128 | */ |
3129 | head = vmalloc_to_page(si->swap_map + offset); | |
3130 | offset &= ~PAGE_MASK; | |
3131 | ||
3132 | /* | |
3133 | * Page allocation does not initialize the page's lru field, | |
3134 | * but it does always reset its private field. | |
3135 | */ | |
3136 | if (!page_private(head)) { | |
3137 | BUG_ON(count & COUNT_CONTINUED); | |
3138 | INIT_LIST_HEAD(&head->lru); | |
3139 | set_page_private(head, SWP_CONTINUED); | |
3140 | si->flags |= SWP_CONTINUED; | |
3141 | } | |
3142 | ||
3143 | list_for_each_entry(list_page, &head->lru, lru) { | |
3144 | unsigned char *map; | |
3145 | ||
3146 | /* | |
3147 | * If the previous map said no continuation, but we've found | |
3148 | * a continuation page, free our allocation and use this one. | |
3149 | */ | |
3150 | if (!(count & COUNT_CONTINUED)) | |
3151 | goto out; | |
3152 | ||
9b04c5fe | 3153 | map = kmap_atomic(list_page) + offset; |
570a335b | 3154 | count = *map; |
9b04c5fe | 3155 | kunmap_atomic(map); |
570a335b HD |
3156 | |
3157 | /* | |
3158 | * If this continuation count now has some space in it, | |
3159 | * free our allocation and use this one. | |
3160 | */ | |
3161 | if ((count & ~COUNT_CONTINUED) != SWAP_CONT_MAX) | |
3162 | goto out; | |
3163 | } | |
3164 | ||
3165 | list_add_tail(&page->lru, &head->lru); | |
3166 | page = NULL; /* now it's attached, don't free it */ | |
3167 | out: | |
235b6217 | 3168 | unlock_cluster(ci); |
ec8acf20 | 3169 | spin_unlock(&si->lock); |
570a335b HD |
3170 | outer: |
3171 | if (page) | |
3172 | __free_page(page); | |
3173 | return 0; | |
3174 | } | |
3175 | ||
3176 | /* | |
3177 | * swap_count_continued - when the original swap_map count is incremented | |
3178 | * from SWAP_MAP_MAX, check if there is already a continuation page to carry | |
3179 | * into, carry if so, or else fail until a new continuation page is allocated; | |
3180 | * when the original swap_map count is decremented from 0 with continuation, | |
3181 | * borrow from the continuation and report whether it still holds more. | |
235b6217 HY |
3182 | * Called while __swap_duplicate() or swap_entry_free() holds swap or cluster |
3183 | * lock. | |
570a335b HD |
3184 | */ |
3185 | static bool swap_count_continued(struct swap_info_struct *si, | |
3186 | pgoff_t offset, unsigned char count) | |
3187 | { | |
3188 | struct page *head; | |
3189 | struct page *page; | |
3190 | unsigned char *map; | |
3191 | ||
3192 | head = vmalloc_to_page(si->swap_map + offset); | |
3193 | if (page_private(head) != SWP_CONTINUED) { | |
3194 | BUG_ON(count & COUNT_CONTINUED); | |
3195 | return false; /* need to add count continuation */ | |
3196 | } | |
3197 | ||
3198 | offset &= ~PAGE_MASK; | |
3199 | page = list_entry(head->lru.next, struct page, lru); | |
9b04c5fe | 3200 | map = kmap_atomic(page) + offset; |
570a335b HD |
3201 | |
3202 | if (count == SWAP_MAP_MAX) /* initial increment from swap_map */ | |
3203 | goto init_map; /* jump over SWAP_CONT_MAX checks */ | |
3204 | ||
3205 | if (count == (SWAP_MAP_MAX | COUNT_CONTINUED)) { /* incrementing */ | |
3206 | /* | |
3207 | * Think of how you add 1 to 999 | |
3208 | */ | |
3209 | while (*map == (SWAP_CONT_MAX | COUNT_CONTINUED)) { | |
9b04c5fe | 3210 | kunmap_atomic(map); |
570a335b HD |
3211 | page = list_entry(page->lru.next, struct page, lru); |
3212 | BUG_ON(page == head); | |
9b04c5fe | 3213 | map = kmap_atomic(page) + offset; |
570a335b HD |
3214 | } |
3215 | if (*map == SWAP_CONT_MAX) { | |
9b04c5fe | 3216 | kunmap_atomic(map); |
570a335b HD |
3217 | page = list_entry(page->lru.next, struct page, lru); |
3218 | if (page == head) | |
3219 | return false; /* add count continuation */ | |
9b04c5fe | 3220 | map = kmap_atomic(page) + offset; |
570a335b HD |
3221 | init_map: *map = 0; /* we didn't zero the page */ |
3222 | } | |
3223 | *map += 1; | |
9b04c5fe | 3224 | kunmap_atomic(map); |
570a335b HD |
3225 | page = list_entry(page->lru.prev, struct page, lru); |
3226 | while (page != head) { | |
9b04c5fe | 3227 | map = kmap_atomic(page) + offset; |
570a335b | 3228 | *map = COUNT_CONTINUED; |
9b04c5fe | 3229 | kunmap_atomic(map); |
570a335b HD |
3230 | page = list_entry(page->lru.prev, struct page, lru); |
3231 | } | |
3232 | return true; /* incremented */ | |
3233 | ||
3234 | } else { /* decrementing */ | |
3235 | /* | |
3236 | * Think of how you subtract 1 from 1000 | |
3237 | */ | |
3238 | BUG_ON(count != COUNT_CONTINUED); | |
3239 | while (*map == COUNT_CONTINUED) { | |
9b04c5fe | 3240 | kunmap_atomic(map); |
570a335b HD |
3241 | page = list_entry(page->lru.next, struct page, lru); |
3242 | BUG_ON(page == head); | |
9b04c5fe | 3243 | map = kmap_atomic(page) + offset; |
570a335b HD |
3244 | } |
3245 | BUG_ON(*map == 0); | |
3246 | *map -= 1; | |
3247 | if (*map == 0) | |
3248 | count = 0; | |
9b04c5fe | 3249 | kunmap_atomic(map); |
570a335b HD |
3250 | page = list_entry(page->lru.prev, struct page, lru); |
3251 | while (page != head) { | |
9b04c5fe | 3252 | map = kmap_atomic(page) + offset; |
570a335b HD |
3253 | *map = SWAP_CONT_MAX | count; |
3254 | count = COUNT_CONTINUED; | |
9b04c5fe | 3255 | kunmap_atomic(map); |
570a335b HD |
3256 | page = list_entry(page->lru.prev, struct page, lru); |
3257 | } | |
3258 | return count == COUNT_CONTINUED; | |
3259 | } | |
3260 | } | |
3261 | ||
3262 | /* | |
3263 | * free_swap_count_continuations - swapoff free all the continuation pages | |
3264 | * appended to the swap_map, after swap_map is quiesced, before vfree'ing it. | |
3265 | */ | |
3266 | static void free_swap_count_continuations(struct swap_info_struct *si) | |
3267 | { | |
3268 | pgoff_t offset; | |
3269 | ||
3270 | for (offset = 0; offset < si->max; offset += PAGE_SIZE) { | |
3271 | struct page *head; | |
3272 | head = vmalloc_to_page(si->swap_map + offset); | |
3273 | if (page_private(head)) { | |
0d576d20 GT |
3274 | struct page *page, *next; |
3275 | ||
3276 | list_for_each_entry_safe(page, next, &head->lru, lru) { | |
3277 | list_del(&page->lru); | |
570a335b HD |
3278 | __free_page(page); |
3279 | } | |
3280 | } | |
3281 | } | |
3282 | } |