Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/mm/swapfile.c | |
3 | * | |
4 | * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds | |
5 | * Swap reorganised 29.12.95, Stephen Tweedie | |
6 | */ | |
7 | ||
1da177e4 | 8 | #include <linux/mm.h> |
6e84f315 | 9 | #include <linux/sched/mm.h> |
29930025 | 10 | #include <linux/sched/task.h> |
1da177e4 LT |
11 | #include <linux/hugetlb.h> |
12 | #include <linux/mman.h> | |
13 | #include <linux/slab.h> | |
14 | #include <linux/kernel_stat.h> | |
15 | #include <linux/swap.h> | |
16 | #include <linux/vmalloc.h> | |
17 | #include <linux/pagemap.h> | |
18 | #include <linux/namei.h> | |
072441e2 | 19 | #include <linux/shmem_fs.h> |
1da177e4 | 20 | #include <linux/blkdev.h> |
20137a49 | 21 | #include <linux/random.h> |
1da177e4 LT |
22 | #include <linux/writeback.h> |
23 | #include <linux/proc_fs.h> | |
24 | #include <linux/seq_file.h> | |
25 | #include <linux/init.h> | |
5ad64688 | 26 | #include <linux/ksm.h> |
1da177e4 LT |
27 | #include <linux/rmap.h> |
28 | #include <linux/security.h> | |
29 | #include <linux/backing-dev.h> | |
fc0abb14 | 30 | #include <linux/mutex.h> |
c59ede7b | 31 | #include <linux/capability.h> |
1da177e4 | 32 | #include <linux/syscalls.h> |
8a9f3ccd | 33 | #include <linux/memcontrol.h> |
66d7dd51 | 34 | #include <linux/poll.h> |
72788c38 | 35 | #include <linux/oom.h> |
38b5faf4 DM |
36 | #include <linux/frontswap.h> |
37 | #include <linux/swapfile.h> | |
f981c595 | 38 | #include <linux/export.h> |
67afa38e | 39 | #include <linux/swap_slots.h> |
1da177e4 LT |
40 | |
41 | #include <asm/pgtable.h> | |
42 | #include <asm/tlbflush.h> | |
43 | #include <linux/swapops.h> | |
5d1ea48b | 44 | #include <linux/swap_cgroup.h> |
1da177e4 | 45 | |
570a335b HD |
46 | static bool swap_count_continued(struct swap_info_struct *, pgoff_t, |
47 | unsigned char); | |
48 | static void free_swap_count_continuations(struct swap_info_struct *); | |
d4906e1a | 49 | static sector_t map_swap_entry(swp_entry_t, struct block_device**); |
570a335b | 50 | |
38b5faf4 | 51 | DEFINE_SPINLOCK(swap_lock); |
7c363b8c | 52 | static unsigned int nr_swapfiles; |
ec8acf20 | 53 | atomic_long_t nr_swap_pages; |
fb0fec50 CW |
54 | /* |
55 | * Some modules use swappable objects and may try to swap them out under | |
56 | * memory pressure (via the shrinker). Before doing so, they may wish to | |
57 | * check to see if any swap space is available. | |
58 | */ | |
59 | EXPORT_SYMBOL_GPL(nr_swap_pages); | |
ec8acf20 | 60 | /* protected with swap_lock. reading in vm_swap_full() doesn't need lock */ |
1da177e4 | 61 | long total_swap_pages; |
78ecba08 | 62 | static int least_priority; |
1da177e4 | 63 | |
1da177e4 LT |
64 | static const char Bad_file[] = "Bad swap file entry "; |
65 | static const char Unused_file[] = "Unused swap file entry "; | |
66 | static const char Bad_offset[] = "Bad swap offset entry "; | |
67 | static const char Unused_offset[] = "Unused swap offset entry "; | |
68 | ||
adfab836 DS |
69 | /* |
70 | * all active swap_info_structs | |
71 | * protected with swap_lock, and ordered by priority. | |
72 | */ | |
18ab4d4c DS |
73 | PLIST_HEAD(swap_active_head); |
74 | ||
75 | /* | |
76 | * all available (active, not full) swap_info_structs | |
77 | * protected with swap_avail_lock, ordered by priority. | |
78 | * This is used by get_swap_page() instead of swap_active_head | |
79 | * because swap_active_head includes all swap_info_structs, | |
80 | * but get_swap_page() doesn't need to look at full ones. | |
81 | * This uses its own lock instead of swap_lock because when a | |
82 | * swap_info_struct changes between not-full/full, it needs to | |
83 | * add/remove itself to/from this list, but the swap_info_struct->lock | |
84 | * is held and the locking order requires swap_lock to be taken | |
85 | * before any swap_info_struct->lock. | |
86 | */ | |
87 | static PLIST_HEAD(swap_avail_head); | |
88 | static DEFINE_SPINLOCK(swap_avail_lock); | |
1da177e4 | 89 | |
38b5faf4 | 90 | struct swap_info_struct *swap_info[MAX_SWAPFILES]; |
1da177e4 | 91 | |
fc0abb14 | 92 | static DEFINE_MUTEX(swapon_mutex); |
1da177e4 | 93 | |
66d7dd51 KS |
94 | static DECLARE_WAIT_QUEUE_HEAD(proc_poll_wait); |
95 | /* Activity counter to indicate that a swapon or swapoff has occurred */ | |
96 | static atomic_t proc_poll_event = ATOMIC_INIT(0); | |
97 | ||
8d69aaee | 98 | static inline unsigned char swap_count(unsigned char ent) |
355cfa73 | 99 | { |
570a335b | 100 | return ent & ~SWAP_HAS_CACHE; /* may include SWAP_HAS_CONT flag */ |
355cfa73 KH |
101 | } |
102 | ||
efa90a98 | 103 | /* returns 1 if swap entry is freed */ |
c9e44410 KH |
104 | static int |
105 | __try_to_reclaim_swap(struct swap_info_struct *si, unsigned long offset) | |
106 | { | |
efa90a98 | 107 | swp_entry_t entry = swp_entry(si->type, offset); |
c9e44410 KH |
108 | struct page *page; |
109 | int ret = 0; | |
110 | ||
f6ab1f7f | 111 | page = find_get_page(swap_address_space(entry), swp_offset(entry)); |
c9e44410 KH |
112 | if (!page) |
113 | return 0; | |
114 | /* | |
115 | * This function is called from scan_swap_map() and it's called | |
116 | * by vmscan.c at reclaiming pages. So, we hold a lock on a page, here. | |
117 | * We have to use trylock for avoiding deadlock. This is a special | |
118 | * case and you should use try_to_free_swap() with explicit lock_page() | |
119 | * in usual operations. | |
120 | */ | |
121 | if (trylock_page(page)) { | |
122 | ret = try_to_free_swap(page); | |
123 | unlock_page(page); | |
124 | } | |
09cbfeaf | 125 | put_page(page); |
c9e44410 KH |
126 | return ret; |
127 | } | |
355cfa73 | 128 | |
6a6ba831 HD |
129 | /* |
130 | * swapon tell device that all the old swap contents can be discarded, | |
131 | * to allow the swap device to optimize its wear-levelling. | |
132 | */ | |
133 | static int discard_swap(struct swap_info_struct *si) | |
134 | { | |
135 | struct swap_extent *se; | |
9625a5f2 HD |
136 | sector_t start_block; |
137 | sector_t nr_blocks; | |
6a6ba831 HD |
138 | int err = 0; |
139 | ||
9625a5f2 HD |
140 | /* Do not discard the swap header page! */ |
141 | se = &si->first_swap_extent; | |
142 | start_block = (se->start_block + 1) << (PAGE_SHIFT - 9); | |
143 | nr_blocks = ((sector_t)se->nr_pages - 1) << (PAGE_SHIFT - 9); | |
144 | if (nr_blocks) { | |
145 | err = blkdev_issue_discard(si->bdev, start_block, | |
dd3932ed | 146 | nr_blocks, GFP_KERNEL, 0); |
9625a5f2 HD |
147 | if (err) |
148 | return err; | |
149 | cond_resched(); | |
150 | } | |
6a6ba831 | 151 | |
9625a5f2 HD |
152 | list_for_each_entry(se, &si->first_swap_extent.list, list) { |
153 | start_block = se->start_block << (PAGE_SHIFT - 9); | |
154 | nr_blocks = (sector_t)se->nr_pages << (PAGE_SHIFT - 9); | |
6a6ba831 HD |
155 | |
156 | err = blkdev_issue_discard(si->bdev, start_block, | |
dd3932ed | 157 | nr_blocks, GFP_KERNEL, 0); |
6a6ba831 HD |
158 | if (err) |
159 | break; | |
160 | ||
161 | cond_resched(); | |
162 | } | |
163 | return err; /* That will often be -EOPNOTSUPP */ | |
164 | } | |
165 | ||
7992fde7 HD |
166 | /* |
167 | * swap allocation tell device that a cluster of swap can now be discarded, | |
168 | * to allow the swap device to optimize its wear-levelling. | |
169 | */ | |
170 | static void discard_swap_cluster(struct swap_info_struct *si, | |
171 | pgoff_t start_page, pgoff_t nr_pages) | |
172 | { | |
173 | struct swap_extent *se = si->curr_swap_extent; | |
174 | int found_extent = 0; | |
175 | ||
176 | while (nr_pages) { | |
7992fde7 HD |
177 | if (se->start_page <= start_page && |
178 | start_page < se->start_page + se->nr_pages) { | |
179 | pgoff_t offset = start_page - se->start_page; | |
180 | sector_t start_block = se->start_block + offset; | |
858a2990 | 181 | sector_t nr_blocks = se->nr_pages - offset; |
7992fde7 HD |
182 | |
183 | if (nr_blocks > nr_pages) | |
184 | nr_blocks = nr_pages; | |
185 | start_page += nr_blocks; | |
186 | nr_pages -= nr_blocks; | |
187 | ||
188 | if (!found_extent++) | |
189 | si->curr_swap_extent = se; | |
190 | ||
191 | start_block <<= PAGE_SHIFT - 9; | |
192 | nr_blocks <<= PAGE_SHIFT - 9; | |
193 | if (blkdev_issue_discard(si->bdev, start_block, | |
dd3932ed | 194 | nr_blocks, GFP_NOIO, 0)) |
7992fde7 HD |
195 | break; |
196 | } | |
197 | ||
a8ae4991 | 198 | se = list_next_entry(se, list); |
7992fde7 HD |
199 | } |
200 | } | |
201 | ||
048c27fd HD |
202 | #define SWAPFILE_CLUSTER 256 |
203 | #define LATENCY_LIMIT 256 | |
204 | ||
2a8f9449 SL |
205 | static inline void cluster_set_flag(struct swap_cluster_info *info, |
206 | unsigned int flag) | |
207 | { | |
208 | info->flags = flag; | |
209 | } | |
210 | ||
211 | static inline unsigned int cluster_count(struct swap_cluster_info *info) | |
212 | { | |
213 | return info->data; | |
214 | } | |
215 | ||
216 | static inline void cluster_set_count(struct swap_cluster_info *info, | |
217 | unsigned int c) | |
218 | { | |
219 | info->data = c; | |
220 | } | |
221 | ||
222 | static inline void cluster_set_count_flag(struct swap_cluster_info *info, | |
223 | unsigned int c, unsigned int f) | |
224 | { | |
225 | info->flags = f; | |
226 | info->data = c; | |
227 | } | |
228 | ||
229 | static inline unsigned int cluster_next(struct swap_cluster_info *info) | |
230 | { | |
231 | return info->data; | |
232 | } | |
233 | ||
234 | static inline void cluster_set_next(struct swap_cluster_info *info, | |
235 | unsigned int n) | |
236 | { | |
237 | info->data = n; | |
238 | } | |
239 | ||
240 | static inline void cluster_set_next_flag(struct swap_cluster_info *info, | |
241 | unsigned int n, unsigned int f) | |
242 | { | |
243 | info->flags = f; | |
244 | info->data = n; | |
245 | } | |
246 | ||
247 | static inline bool cluster_is_free(struct swap_cluster_info *info) | |
248 | { | |
249 | return info->flags & CLUSTER_FLAG_FREE; | |
250 | } | |
251 | ||
252 | static inline bool cluster_is_null(struct swap_cluster_info *info) | |
253 | { | |
254 | return info->flags & CLUSTER_FLAG_NEXT_NULL; | |
255 | } | |
256 | ||
257 | static inline void cluster_set_null(struct swap_cluster_info *info) | |
258 | { | |
259 | info->flags = CLUSTER_FLAG_NEXT_NULL; | |
260 | info->data = 0; | |
261 | } | |
262 | ||
235b6217 HY |
263 | static inline struct swap_cluster_info *lock_cluster(struct swap_info_struct *si, |
264 | unsigned long offset) | |
265 | { | |
266 | struct swap_cluster_info *ci; | |
267 | ||
268 | ci = si->cluster_info; | |
269 | if (ci) { | |
270 | ci += offset / SWAPFILE_CLUSTER; | |
271 | spin_lock(&ci->lock); | |
272 | } | |
273 | return ci; | |
274 | } | |
275 | ||
276 | static inline void unlock_cluster(struct swap_cluster_info *ci) | |
277 | { | |
278 | if (ci) | |
279 | spin_unlock(&ci->lock); | |
280 | } | |
281 | ||
282 | static inline struct swap_cluster_info *lock_cluster_or_swap_info( | |
283 | struct swap_info_struct *si, | |
284 | unsigned long offset) | |
285 | { | |
286 | struct swap_cluster_info *ci; | |
287 | ||
288 | ci = lock_cluster(si, offset); | |
289 | if (!ci) | |
290 | spin_lock(&si->lock); | |
291 | ||
292 | return ci; | |
293 | } | |
294 | ||
295 | static inline void unlock_cluster_or_swap_info(struct swap_info_struct *si, | |
296 | struct swap_cluster_info *ci) | |
297 | { | |
298 | if (ci) | |
299 | unlock_cluster(ci); | |
300 | else | |
301 | spin_unlock(&si->lock); | |
302 | } | |
303 | ||
6b534915 HY |
304 | static inline bool cluster_list_empty(struct swap_cluster_list *list) |
305 | { | |
306 | return cluster_is_null(&list->head); | |
307 | } | |
308 | ||
309 | static inline unsigned int cluster_list_first(struct swap_cluster_list *list) | |
310 | { | |
311 | return cluster_next(&list->head); | |
312 | } | |
313 | ||
314 | static void cluster_list_init(struct swap_cluster_list *list) | |
315 | { | |
316 | cluster_set_null(&list->head); | |
317 | cluster_set_null(&list->tail); | |
318 | } | |
319 | ||
320 | static void cluster_list_add_tail(struct swap_cluster_list *list, | |
321 | struct swap_cluster_info *ci, | |
322 | unsigned int idx) | |
323 | { | |
324 | if (cluster_list_empty(list)) { | |
325 | cluster_set_next_flag(&list->head, idx, 0); | |
326 | cluster_set_next_flag(&list->tail, idx, 0); | |
327 | } else { | |
235b6217 | 328 | struct swap_cluster_info *ci_tail; |
6b534915 HY |
329 | unsigned int tail = cluster_next(&list->tail); |
330 | ||
235b6217 HY |
331 | /* |
332 | * Nested cluster lock, but both cluster locks are | |
333 | * only acquired when we held swap_info_struct->lock | |
334 | */ | |
335 | ci_tail = ci + tail; | |
336 | spin_lock_nested(&ci_tail->lock, SINGLE_DEPTH_NESTING); | |
337 | cluster_set_next(ci_tail, idx); | |
338 | unlock_cluster(ci_tail); | |
6b534915 HY |
339 | cluster_set_next_flag(&list->tail, idx, 0); |
340 | } | |
341 | } | |
342 | ||
343 | static unsigned int cluster_list_del_first(struct swap_cluster_list *list, | |
344 | struct swap_cluster_info *ci) | |
345 | { | |
346 | unsigned int idx; | |
347 | ||
348 | idx = cluster_next(&list->head); | |
349 | if (cluster_next(&list->tail) == idx) { | |
350 | cluster_set_null(&list->head); | |
351 | cluster_set_null(&list->tail); | |
352 | } else | |
353 | cluster_set_next_flag(&list->head, | |
354 | cluster_next(&ci[idx]), 0); | |
355 | ||
356 | return idx; | |
357 | } | |
358 | ||
815c2c54 SL |
359 | /* Add a cluster to discard list and schedule it to do discard */ |
360 | static void swap_cluster_schedule_discard(struct swap_info_struct *si, | |
361 | unsigned int idx) | |
362 | { | |
363 | /* | |
364 | * If scan_swap_map() can't find a free cluster, it will check | |
365 | * si->swap_map directly. To make sure the discarding cluster isn't | |
366 | * taken by scan_swap_map(), mark the swap entries bad (occupied). It | |
367 | * will be cleared after discard | |
368 | */ | |
369 | memset(si->swap_map + idx * SWAPFILE_CLUSTER, | |
370 | SWAP_MAP_BAD, SWAPFILE_CLUSTER); | |
371 | ||
6b534915 | 372 | cluster_list_add_tail(&si->discard_clusters, si->cluster_info, idx); |
815c2c54 SL |
373 | |
374 | schedule_work(&si->discard_work); | |
375 | } | |
376 | ||
377 | /* | |
378 | * Doing discard actually. After a cluster discard is finished, the cluster | |
379 | * will be added to free cluster list. caller should hold si->lock. | |
380 | */ | |
381 | static void swap_do_scheduled_discard(struct swap_info_struct *si) | |
382 | { | |
235b6217 | 383 | struct swap_cluster_info *info, *ci; |
815c2c54 SL |
384 | unsigned int idx; |
385 | ||
386 | info = si->cluster_info; | |
387 | ||
6b534915 HY |
388 | while (!cluster_list_empty(&si->discard_clusters)) { |
389 | idx = cluster_list_del_first(&si->discard_clusters, info); | |
815c2c54 SL |
390 | spin_unlock(&si->lock); |
391 | ||
392 | discard_swap_cluster(si, idx * SWAPFILE_CLUSTER, | |
393 | SWAPFILE_CLUSTER); | |
394 | ||
395 | spin_lock(&si->lock); | |
235b6217 HY |
396 | ci = lock_cluster(si, idx * SWAPFILE_CLUSTER); |
397 | cluster_set_flag(ci, CLUSTER_FLAG_FREE); | |
398 | unlock_cluster(ci); | |
6b534915 | 399 | cluster_list_add_tail(&si->free_clusters, info, idx); |
235b6217 | 400 | ci = lock_cluster(si, idx * SWAPFILE_CLUSTER); |
815c2c54 SL |
401 | memset(si->swap_map + idx * SWAPFILE_CLUSTER, |
402 | 0, SWAPFILE_CLUSTER); | |
235b6217 | 403 | unlock_cluster(ci); |
815c2c54 SL |
404 | } |
405 | } | |
406 | ||
407 | static void swap_discard_work(struct work_struct *work) | |
408 | { | |
409 | struct swap_info_struct *si; | |
410 | ||
411 | si = container_of(work, struct swap_info_struct, discard_work); | |
412 | ||
413 | spin_lock(&si->lock); | |
414 | swap_do_scheduled_discard(si); | |
415 | spin_unlock(&si->lock); | |
416 | } | |
417 | ||
2a8f9449 SL |
418 | /* |
419 | * The cluster corresponding to page_nr will be used. The cluster will be | |
420 | * removed from free cluster list and its usage counter will be increased. | |
421 | */ | |
422 | static void inc_cluster_info_page(struct swap_info_struct *p, | |
423 | struct swap_cluster_info *cluster_info, unsigned long page_nr) | |
424 | { | |
425 | unsigned long idx = page_nr / SWAPFILE_CLUSTER; | |
426 | ||
427 | if (!cluster_info) | |
428 | return; | |
429 | if (cluster_is_free(&cluster_info[idx])) { | |
6b534915 HY |
430 | VM_BUG_ON(cluster_list_first(&p->free_clusters) != idx); |
431 | cluster_list_del_first(&p->free_clusters, cluster_info); | |
2a8f9449 SL |
432 | cluster_set_count_flag(&cluster_info[idx], 0, 0); |
433 | } | |
434 | ||
435 | VM_BUG_ON(cluster_count(&cluster_info[idx]) >= SWAPFILE_CLUSTER); | |
436 | cluster_set_count(&cluster_info[idx], | |
437 | cluster_count(&cluster_info[idx]) + 1); | |
438 | } | |
439 | ||
440 | /* | |
441 | * The cluster corresponding to page_nr decreases one usage. If the usage | |
442 | * counter becomes 0, which means no page in the cluster is in using, we can | |
443 | * optionally discard the cluster and add it to free cluster list. | |
444 | */ | |
445 | static void dec_cluster_info_page(struct swap_info_struct *p, | |
446 | struct swap_cluster_info *cluster_info, unsigned long page_nr) | |
447 | { | |
448 | unsigned long idx = page_nr / SWAPFILE_CLUSTER; | |
449 | ||
450 | if (!cluster_info) | |
451 | return; | |
452 | ||
453 | VM_BUG_ON(cluster_count(&cluster_info[idx]) == 0); | |
454 | cluster_set_count(&cluster_info[idx], | |
455 | cluster_count(&cluster_info[idx]) - 1); | |
456 | ||
457 | if (cluster_count(&cluster_info[idx]) == 0) { | |
815c2c54 SL |
458 | /* |
459 | * If the swap is discardable, prepare discard the cluster | |
460 | * instead of free it immediately. The cluster will be freed | |
461 | * after discard. | |
462 | */ | |
edfe23da SL |
463 | if ((p->flags & (SWP_WRITEOK | SWP_PAGE_DISCARD)) == |
464 | (SWP_WRITEOK | SWP_PAGE_DISCARD)) { | |
815c2c54 SL |
465 | swap_cluster_schedule_discard(p, idx); |
466 | return; | |
467 | } | |
468 | ||
2a8f9449 | 469 | cluster_set_flag(&cluster_info[idx], CLUSTER_FLAG_FREE); |
6b534915 | 470 | cluster_list_add_tail(&p->free_clusters, cluster_info, idx); |
2a8f9449 SL |
471 | } |
472 | } | |
473 | ||
474 | /* | |
475 | * It's possible scan_swap_map() uses a free cluster in the middle of free | |
476 | * cluster list. Avoiding such abuse to avoid list corruption. | |
477 | */ | |
ebc2a1a6 SL |
478 | static bool |
479 | scan_swap_map_ssd_cluster_conflict(struct swap_info_struct *si, | |
2a8f9449 SL |
480 | unsigned long offset) |
481 | { | |
ebc2a1a6 SL |
482 | struct percpu_cluster *percpu_cluster; |
483 | bool conflict; | |
484 | ||
2a8f9449 | 485 | offset /= SWAPFILE_CLUSTER; |
6b534915 HY |
486 | conflict = !cluster_list_empty(&si->free_clusters) && |
487 | offset != cluster_list_first(&si->free_clusters) && | |
2a8f9449 | 488 | cluster_is_free(&si->cluster_info[offset]); |
ebc2a1a6 SL |
489 | |
490 | if (!conflict) | |
491 | return false; | |
492 | ||
493 | percpu_cluster = this_cpu_ptr(si->percpu_cluster); | |
494 | cluster_set_null(&percpu_cluster->index); | |
495 | return true; | |
496 | } | |
497 | ||
498 | /* | |
499 | * Try to get a swap entry from current cpu's swap entry pool (a cluster). This | |
500 | * might involve allocating a new cluster for current CPU too. | |
501 | */ | |
36005bae | 502 | static bool scan_swap_map_try_ssd_cluster(struct swap_info_struct *si, |
ebc2a1a6 SL |
503 | unsigned long *offset, unsigned long *scan_base) |
504 | { | |
505 | struct percpu_cluster *cluster; | |
235b6217 | 506 | struct swap_cluster_info *ci; |
ebc2a1a6 | 507 | bool found_free; |
235b6217 | 508 | unsigned long tmp, max; |
ebc2a1a6 SL |
509 | |
510 | new_cluster: | |
511 | cluster = this_cpu_ptr(si->percpu_cluster); | |
512 | if (cluster_is_null(&cluster->index)) { | |
6b534915 HY |
513 | if (!cluster_list_empty(&si->free_clusters)) { |
514 | cluster->index = si->free_clusters.head; | |
ebc2a1a6 SL |
515 | cluster->next = cluster_next(&cluster->index) * |
516 | SWAPFILE_CLUSTER; | |
6b534915 | 517 | } else if (!cluster_list_empty(&si->discard_clusters)) { |
ebc2a1a6 SL |
518 | /* |
519 | * we don't have free cluster but have some clusters in | |
520 | * discarding, do discard now and reclaim them | |
521 | */ | |
522 | swap_do_scheduled_discard(si); | |
523 | *scan_base = *offset = si->cluster_next; | |
524 | goto new_cluster; | |
525 | } else | |
36005bae | 526 | return false; |
ebc2a1a6 SL |
527 | } |
528 | ||
529 | found_free = false; | |
530 | ||
531 | /* | |
532 | * Other CPUs can use our cluster if they can't find a free cluster, | |
533 | * check if there is still free entry in the cluster | |
534 | */ | |
535 | tmp = cluster->next; | |
235b6217 HY |
536 | max = min_t(unsigned long, si->max, |
537 | (cluster_next(&cluster->index) + 1) * SWAPFILE_CLUSTER); | |
538 | if (tmp >= max) { | |
539 | cluster_set_null(&cluster->index); | |
540 | goto new_cluster; | |
541 | } | |
542 | ci = lock_cluster(si, tmp); | |
543 | while (tmp < max) { | |
ebc2a1a6 SL |
544 | if (!si->swap_map[tmp]) { |
545 | found_free = true; | |
546 | break; | |
547 | } | |
548 | tmp++; | |
549 | } | |
235b6217 | 550 | unlock_cluster(ci); |
ebc2a1a6 SL |
551 | if (!found_free) { |
552 | cluster_set_null(&cluster->index); | |
553 | goto new_cluster; | |
554 | } | |
555 | cluster->next = tmp + 1; | |
556 | *offset = tmp; | |
557 | *scan_base = tmp; | |
36005bae | 558 | return found_free; |
2a8f9449 SL |
559 | } |
560 | ||
36005bae TC |
561 | static int scan_swap_map_slots(struct swap_info_struct *si, |
562 | unsigned char usage, int nr, | |
563 | swp_entry_t slots[]) | |
1da177e4 | 564 | { |
235b6217 | 565 | struct swap_cluster_info *ci; |
ebebbbe9 | 566 | unsigned long offset; |
c60aa176 | 567 | unsigned long scan_base; |
7992fde7 | 568 | unsigned long last_in_cluster = 0; |
048c27fd | 569 | int latency_ration = LATENCY_LIMIT; |
36005bae TC |
570 | int n_ret = 0; |
571 | ||
572 | if (nr > SWAP_BATCH) | |
573 | nr = SWAP_BATCH; | |
7dfad418 | 574 | |
886bb7e9 | 575 | /* |
7dfad418 HD |
576 | * We try to cluster swap pages by allocating them sequentially |
577 | * in swap. Once we've allocated SWAPFILE_CLUSTER pages this | |
578 | * way, however, we resort to first-free allocation, starting | |
579 | * a new cluster. This prevents us from scattering swap pages | |
580 | * all over the entire swap partition, so that we reduce | |
581 | * overall disk seek times between swap pages. -- sct | |
582 | * But we do now try to find an empty cluster. -Andrea | |
c60aa176 | 583 | * And we let swap pages go all over an SSD partition. Hugh |
7dfad418 HD |
584 | */ |
585 | ||
52b7efdb | 586 | si->flags += SWP_SCANNING; |
c60aa176 | 587 | scan_base = offset = si->cluster_next; |
ebebbbe9 | 588 | |
ebc2a1a6 SL |
589 | /* SSD algorithm */ |
590 | if (si->cluster_info) { | |
36005bae TC |
591 | if (scan_swap_map_try_ssd_cluster(si, &offset, &scan_base)) |
592 | goto checks; | |
593 | else | |
594 | goto scan; | |
ebc2a1a6 SL |
595 | } |
596 | ||
ebebbbe9 HD |
597 | if (unlikely(!si->cluster_nr--)) { |
598 | if (si->pages - si->inuse_pages < SWAPFILE_CLUSTER) { | |
599 | si->cluster_nr = SWAPFILE_CLUSTER - 1; | |
600 | goto checks; | |
601 | } | |
2a8f9449 | 602 | |
ec8acf20 | 603 | spin_unlock(&si->lock); |
7dfad418 | 604 | |
c60aa176 HD |
605 | /* |
606 | * If seek is expensive, start searching for new cluster from | |
607 | * start of partition, to minimize the span of allocated swap. | |
50088c44 CY |
608 | * If seek is cheap, that is the SWP_SOLIDSTATE si->cluster_info |
609 | * case, just handled by scan_swap_map_try_ssd_cluster() above. | |
c60aa176 | 610 | */ |
50088c44 | 611 | scan_base = offset = si->lowest_bit; |
7dfad418 HD |
612 | last_in_cluster = offset + SWAPFILE_CLUSTER - 1; |
613 | ||
614 | /* Locate the first empty (unaligned) cluster */ | |
615 | for (; last_in_cluster <= si->highest_bit; offset++) { | |
1da177e4 | 616 | if (si->swap_map[offset]) |
7dfad418 HD |
617 | last_in_cluster = offset + SWAPFILE_CLUSTER; |
618 | else if (offset == last_in_cluster) { | |
ec8acf20 | 619 | spin_lock(&si->lock); |
ebebbbe9 HD |
620 | offset -= SWAPFILE_CLUSTER - 1; |
621 | si->cluster_next = offset; | |
622 | si->cluster_nr = SWAPFILE_CLUSTER - 1; | |
c60aa176 HD |
623 | goto checks; |
624 | } | |
625 | if (unlikely(--latency_ration < 0)) { | |
626 | cond_resched(); | |
627 | latency_ration = LATENCY_LIMIT; | |
628 | } | |
629 | } | |
630 | ||
631 | offset = scan_base; | |
ec8acf20 | 632 | spin_lock(&si->lock); |
ebebbbe9 | 633 | si->cluster_nr = SWAPFILE_CLUSTER - 1; |
1da177e4 | 634 | } |
7dfad418 | 635 | |
ebebbbe9 | 636 | checks: |
ebc2a1a6 | 637 | if (si->cluster_info) { |
36005bae TC |
638 | while (scan_swap_map_ssd_cluster_conflict(si, offset)) { |
639 | /* take a break if we already got some slots */ | |
640 | if (n_ret) | |
641 | goto done; | |
642 | if (!scan_swap_map_try_ssd_cluster(si, &offset, | |
643 | &scan_base)) | |
644 | goto scan; | |
645 | } | |
ebc2a1a6 | 646 | } |
ebebbbe9 | 647 | if (!(si->flags & SWP_WRITEOK)) |
52b7efdb | 648 | goto no_page; |
7dfad418 HD |
649 | if (!si->highest_bit) |
650 | goto no_page; | |
ebebbbe9 | 651 | if (offset > si->highest_bit) |
c60aa176 | 652 | scan_base = offset = si->lowest_bit; |
c9e44410 | 653 | |
235b6217 | 654 | ci = lock_cluster(si, offset); |
b73d7fce HD |
655 | /* reuse swap entry of cache-only swap if not busy. */ |
656 | if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) { | |
c9e44410 | 657 | int swap_was_freed; |
235b6217 | 658 | unlock_cluster(ci); |
ec8acf20 | 659 | spin_unlock(&si->lock); |
c9e44410 | 660 | swap_was_freed = __try_to_reclaim_swap(si, offset); |
ec8acf20 | 661 | spin_lock(&si->lock); |
c9e44410 KH |
662 | /* entry was freed successfully, try to use this again */ |
663 | if (swap_was_freed) | |
664 | goto checks; | |
665 | goto scan; /* check next one */ | |
666 | } | |
667 | ||
235b6217 HY |
668 | if (si->swap_map[offset]) { |
669 | unlock_cluster(ci); | |
36005bae TC |
670 | if (!n_ret) |
671 | goto scan; | |
672 | else | |
673 | goto done; | |
235b6217 | 674 | } |
ebebbbe9 HD |
675 | |
676 | if (offset == si->lowest_bit) | |
677 | si->lowest_bit++; | |
678 | if (offset == si->highest_bit) | |
679 | si->highest_bit--; | |
680 | si->inuse_pages++; | |
681 | if (si->inuse_pages == si->pages) { | |
682 | si->lowest_bit = si->max; | |
683 | si->highest_bit = 0; | |
18ab4d4c DS |
684 | spin_lock(&swap_avail_lock); |
685 | plist_del(&si->avail_list, &swap_avail_head); | |
686 | spin_unlock(&swap_avail_lock); | |
1da177e4 | 687 | } |
253d553b | 688 | si->swap_map[offset] = usage; |
2a8f9449 | 689 | inc_cluster_info_page(si, si->cluster_info, offset); |
235b6217 | 690 | unlock_cluster(ci); |
ebebbbe9 | 691 | si->cluster_next = offset + 1; |
36005bae TC |
692 | slots[n_ret++] = swp_entry(si->type, offset); |
693 | ||
694 | /* got enough slots or reach max slots? */ | |
695 | if ((n_ret == nr) || (offset >= si->highest_bit)) | |
696 | goto done; | |
697 | ||
698 | /* search for next available slot */ | |
699 | ||
700 | /* time to take a break? */ | |
701 | if (unlikely(--latency_ration < 0)) { | |
702 | if (n_ret) | |
703 | goto done; | |
704 | spin_unlock(&si->lock); | |
705 | cond_resched(); | |
706 | spin_lock(&si->lock); | |
707 | latency_ration = LATENCY_LIMIT; | |
708 | } | |
709 | ||
710 | /* try to get more slots in cluster */ | |
711 | if (si->cluster_info) { | |
712 | if (scan_swap_map_try_ssd_cluster(si, &offset, &scan_base)) | |
713 | goto checks; | |
714 | else | |
715 | goto done; | |
716 | } | |
717 | /* non-ssd case */ | |
718 | ++offset; | |
719 | ||
720 | /* non-ssd case, still more slots in cluster? */ | |
721 | if (si->cluster_nr && !si->swap_map[offset]) { | |
722 | --si->cluster_nr; | |
723 | goto checks; | |
724 | } | |
7992fde7 | 725 | |
36005bae TC |
726 | done: |
727 | si->flags -= SWP_SCANNING; | |
728 | return n_ret; | |
7dfad418 | 729 | |
ebebbbe9 | 730 | scan: |
ec8acf20 | 731 | spin_unlock(&si->lock); |
7dfad418 | 732 | while (++offset <= si->highest_bit) { |
52b7efdb | 733 | if (!si->swap_map[offset]) { |
ec8acf20 | 734 | spin_lock(&si->lock); |
52b7efdb HD |
735 | goto checks; |
736 | } | |
c9e44410 | 737 | if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) { |
ec8acf20 | 738 | spin_lock(&si->lock); |
c9e44410 KH |
739 | goto checks; |
740 | } | |
048c27fd HD |
741 | if (unlikely(--latency_ration < 0)) { |
742 | cond_resched(); | |
743 | latency_ration = LATENCY_LIMIT; | |
744 | } | |
7dfad418 | 745 | } |
c60aa176 | 746 | offset = si->lowest_bit; |
a5998061 | 747 | while (offset < scan_base) { |
c60aa176 | 748 | if (!si->swap_map[offset]) { |
ec8acf20 | 749 | spin_lock(&si->lock); |
c60aa176 HD |
750 | goto checks; |
751 | } | |
c9e44410 | 752 | if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) { |
ec8acf20 | 753 | spin_lock(&si->lock); |
c9e44410 KH |
754 | goto checks; |
755 | } | |
c60aa176 HD |
756 | if (unlikely(--latency_ration < 0)) { |
757 | cond_resched(); | |
758 | latency_ration = LATENCY_LIMIT; | |
759 | } | |
a5998061 | 760 | offset++; |
c60aa176 | 761 | } |
ec8acf20 | 762 | spin_lock(&si->lock); |
7dfad418 HD |
763 | |
764 | no_page: | |
52b7efdb | 765 | si->flags -= SWP_SCANNING; |
36005bae | 766 | return n_ret; |
1da177e4 LT |
767 | } |
768 | ||
36005bae TC |
769 | static unsigned long scan_swap_map(struct swap_info_struct *si, |
770 | unsigned char usage) | |
771 | { | |
772 | swp_entry_t entry; | |
773 | int n_ret; | |
774 | ||
775 | n_ret = scan_swap_map_slots(si, usage, 1, &entry); | |
776 | ||
777 | if (n_ret) | |
778 | return swp_offset(entry); | |
779 | else | |
780 | return 0; | |
781 | ||
782 | } | |
783 | ||
784 | int get_swap_pages(int n_goal, swp_entry_t swp_entries[]) | |
1da177e4 | 785 | { |
adfab836 | 786 | struct swap_info_struct *si, *next; |
36005bae TC |
787 | long avail_pgs; |
788 | int n_ret = 0; | |
1da177e4 | 789 | |
36005bae TC |
790 | avail_pgs = atomic_long_read(&nr_swap_pages); |
791 | if (avail_pgs <= 0) | |
fb4f88dc | 792 | goto noswap; |
36005bae TC |
793 | |
794 | if (n_goal > SWAP_BATCH) | |
795 | n_goal = SWAP_BATCH; | |
796 | ||
797 | if (n_goal > avail_pgs) | |
798 | n_goal = avail_pgs; | |
799 | ||
800 | atomic_long_sub(n_goal, &nr_swap_pages); | |
fb4f88dc | 801 | |
18ab4d4c DS |
802 | spin_lock(&swap_avail_lock); |
803 | ||
804 | start_over: | |
805 | plist_for_each_entry_safe(si, next, &swap_avail_head, avail_list) { | |
806 | /* requeue si to after same-priority siblings */ | |
807 | plist_requeue(&si->avail_list, &swap_avail_head); | |
808 | spin_unlock(&swap_avail_lock); | |
ec8acf20 | 809 | spin_lock(&si->lock); |
adfab836 | 810 | if (!si->highest_bit || !(si->flags & SWP_WRITEOK)) { |
18ab4d4c DS |
811 | spin_lock(&swap_avail_lock); |
812 | if (plist_node_empty(&si->avail_list)) { | |
813 | spin_unlock(&si->lock); | |
814 | goto nextsi; | |
815 | } | |
816 | WARN(!si->highest_bit, | |
817 | "swap_info %d in list but !highest_bit\n", | |
818 | si->type); | |
819 | WARN(!(si->flags & SWP_WRITEOK), | |
820 | "swap_info %d in list but !SWP_WRITEOK\n", | |
821 | si->type); | |
822 | plist_del(&si->avail_list, &swap_avail_head); | |
ec8acf20 | 823 | spin_unlock(&si->lock); |
18ab4d4c | 824 | goto nextsi; |
ec8acf20 | 825 | } |
36005bae TC |
826 | n_ret = scan_swap_map_slots(si, SWAP_HAS_CACHE, |
827 | n_goal, swp_entries); | |
ec8acf20 | 828 | spin_unlock(&si->lock); |
36005bae TC |
829 | if (n_ret) |
830 | goto check_out; | |
18ab4d4c | 831 | pr_debug("scan_swap_map of si %d failed to find offset\n", |
36005bae TC |
832 | si->type); |
833 | ||
18ab4d4c DS |
834 | spin_lock(&swap_avail_lock); |
835 | nextsi: | |
adfab836 DS |
836 | /* |
837 | * if we got here, it's likely that si was almost full before, | |
838 | * and since scan_swap_map() can drop the si->lock, multiple | |
839 | * callers probably all tried to get a page from the same si | |
18ab4d4c DS |
840 | * and it filled up before we could get one; or, the si filled |
841 | * up between us dropping swap_avail_lock and taking si->lock. | |
842 | * Since we dropped the swap_avail_lock, the swap_avail_head | |
843 | * list may have been modified; so if next is still in the | |
36005bae TC |
844 | * swap_avail_head list then try it, otherwise start over |
845 | * if we have not gotten any slots. | |
adfab836 | 846 | */ |
18ab4d4c DS |
847 | if (plist_node_empty(&next->avail_list)) |
848 | goto start_over; | |
1da177e4 | 849 | } |
fb4f88dc | 850 | |
18ab4d4c DS |
851 | spin_unlock(&swap_avail_lock); |
852 | ||
36005bae TC |
853 | check_out: |
854 | if (n_ret < n_goal) | |
855 | atomic_long_add((long) (n_goal-n_ret), &nr_swap_pages); | |
fb4f88dc | 856 | noswap: |
36005bae TC |
857 | return n_ret; |
858 | } | |
859 | ||
2de1a7e4 | 860 | /* The only caller of this function is now suspend routine */ |
910321ea HD |
861 | swp_entry_t get_swap_page_of_type(int type) |
862 | { | |
863 | struct swap_info_struct *si; | |
864 | pgoff_t offset; | |
865 | ||
910321ea | 866 | si = swap_info[type]; |
ec8acf20 | 867 | spin_lock(&si->lock); |
910321ea | 868 | if (si && (si->flags & SWP_WRITEOK)) { |
ec8acf20 | 869 | atomic_long_dec(&nr_swap_pages); |
910321ea HD |
870 | /* This is called for allocating swap entry, not cache */ |
871 | offset = scan_swap_map(si, 1); | |
872 | if (offset) { | |
ec8acf20 | 873 | spin_unlock(&si->lock); |
910321ea HD |
874 | return swp_entry(type, offset); |
875 | } | |
ec8acf20 | 876 | atomic_long_inc(&nr_swap_pages); |
910321ea | 877 | } |
ec8acf20 | 878 | spin_unlock(&si->lock); |
910321ea HD |
879 | return (swp_entry_t) {0}; |
880 | } | |
881 | ||
e8c26ab6 | 882 | static struct swap_info_struct *__swap_info_get(swp_entry_t entry) |
1da177e4 | 883 | { |
73c34b6a | 884 | struct swap_info_struct *p; |
1da177e4 LT |
885 | unsigned long offset, type; |
886 | ||
887 | if (!entry.val) | |
888 | goto out; | |
889 | type = swp_type(entry); | |
890 | if (type >= nr_swapfiles) | |
891 | goto bad_nofile; | |
efa90a98 | 892 | p = swap_info[type]; |
1da177e4 LT |
893 | if (!(p->flags & SWP_USED)) |
894 | goto bad_device; | |
895 | offset = swp_offset(entry); | |
896 | if (offset >= p->max) | |
897 | goto bad_offset; | |
1da177e4 LT |
898 | return p; |
899 | ||
1da177e4 | 900 | bad_offset: |
6a991fc7 | 901 | pr_err("swap_info_get: %s%08lx\n", Bad_offset, entry.val); |
1da177e4 LT |
902 | goto out; |
903 | bad_device: | |
6a991fc7 | 904 | pr_err("swap_info_get: %s%08lx\n", Unused_file, entry.val); |
1da177e4 LT |
905 | goto out; |
906 | bad_nofile: | |
6a991fc7 | 907 | pr_err("swap_info_get: %s%08lx\n", Bad_file, entry.val); |
1da177e4 LT |
908 | out: |
909 | return NULL; | |
886bb7e9 | 910 | } |
1da177e4 | 911 | |
e8c26ab6 TC |
912 | static struct swap_info_struct *_swap_info_get(swp_entry_t entry) |
913 | { | |
914 | struct swap_info_struct *p; | |
915 | ||
916 | p = __swap_info_get(entry); | |
917 | if (!p) | |
918 | goto out; | |
919 | if (!p->swap_map[swp_offset(entry)]) | |
920 | goto bad_free; | |
921 | return p; | |
922 | ||
923 | bad_free: | |
924 | pr_err("swap_info_get: %s%08lx\n", Unused_offset, entry.val); | |
925 | goto out; | |
926 | out: | |
927 | return NULL; | |
928 | } | |
929 | ||
235b6217 HY |
930 | static struct swap_info_struct *swap_info_get(swp_entry_t entry) |
931 | { | |
932 | struct swap_info_struct *p; | |
933 | ||
934 | p = _swap_info_get(entry); | |
935 | if (p) | |
936 | spin_lock(&p->lock); | |
937 | return p; | |
938 | } | |
939 | ||
7c00bafe TC |
940 | static struct swap_info_struct *swap_info_get_cont(swp_entry_t entry, |
941 | struct swap_info_struct *q) | |
942 | { | |
943 | struct swap_info_struct *p; | |
944 | ||
945 | p = _swap_info_get(entry); | |
946 | ||
947 | if (p != q) { | |
948 | if (q != NULL) | |
949 | spin_unlock(&q->lock); | |
950 | if (p != NULL) | |
951 | spin_lock(&p->lock); | |
952 | } | |
953 | return p; | |
954 | } | |
955 | ||
956 | static unsigned char __swap_entry_free(struct swap_info_struct *p, | |
957 | swp_entry_t entry, unsigned char usage) | |
1da177e4 | 958 | { |
235b6217 | 959 | struct swap_cluster_info *ci; |
253d553b | 960 | unsigned long offset = swp_offset(entry); |
8d69aaee HD |
961 | unsigned char count; |
962 | unsigned char has_cache; | |
235b6217 | 963 | |
7c00bafe | 964 | ci = lock_cluster_or_swap_info(p, offset); |
355cfa73 | 965 | |
253d553b | 966 | count = p->swap_map[offset]; |
235b6217 | 967 | |
253d553b HD |
968 | has_cache = count & SWAP_HAS_CACHE; |
969 | count &= ~SWAP_HAS_CACHE; | |
355cfa73 | 970 | |
253d553b | 971 | if (usage == SWAP_HAS_CACHE) { |
355cfa73 | 972 | VM_BUG_ON(!has_cache); |
253d553b | 973 | has_cache = 0; |
aaa46865 HD |
974 | } else if (count == SWAP_MAP_SHMEM) { |
975 | /* | |
976 | * Or we could insist on shmem.c using a special | |
977 | * swap_shmem_free() and free_shmem_swap_and_cache()... | |
978 | */ | |
979 | count = 0; | |
570a335b HD |
980 | } else if ((count & ~COUNT_CONTINUED) <= SWAP_MAP_MAX) { |
981 | if (count == COUNT_CONTINUED) { | |
982 | if (swap_count_continued(p, offset, count)) | |
983 | count = SWAP_MAP_MAX | COUNT_CONTINUED; | |
984 | else | |
985 | count = SWAP_MAP_MAX; | |
986 | } else | |
987 | count--; | |
988 | } | |
253d553b | 989 | |
253d553b | 990 | usage = count | has_cache; |
7c00bafe TC |
991 | p->swap_map[offset] = usage ? : SWAP_HAS_CACHE; |
992 | ||
993 | unlock_cluster_or_swap_info(p, ci); | |
994 | ||
995 | return usage; | |
996 | } | |
355cfa73 | 997 | |
7c00bafe TC |
998 | static void swap_entry_free(struct swap_info_struct *p, swp_entry_t entry) |
999 | { | |
1000 | struct swap_cluster_info *ci; | |
1001 | unsigned long offset = swp_offset(entry); | |
1002 | unsigned char count; | |
1003 | ||
1004 | ci = lock_cluster(p, offset); | |
1005 | count = p->swap_map[offset]; | |
1006 | VM_BUG_ON(count != SWAP_HAS_CACHE); | |
1007 | p->swap_map[offset] = 0; | |
1008 | dec_cluster_info_page(p, p->cluster_info, offset); | |
235b6217 HY |
1009 | unlock_cluster(ci); |
1010 | ||
7c00bafe TC |
1011 | mem_cgroup_uncharge_swap(entry); |
1012 | if (offset < p->lowest_bit) | |
1013 | p->lowest_bit = offset; | |
1014 | if (offset > p->highest_bit) { | |
1015 | bool was_full = !p->highest_bit; | |
1016 | ||
1017 | p->highest_bit = offset; | |
1018 | if (was_full && (p->flags & SWP_WRITEOK)) { | |
1019 | spin_lock(&swap_avail_lock); | |
1020 | WARN_ON(!plist_node_empty(&p->avail_list)); | |
1021 | if (plist_node_empty(&p->avail_list)) | |
1022 | plist_add(&p->avail_list, | |
1023 | &swap_avail_head); | |
1024 | spin_unlock(&swap_avail_lock); | |
73744923 | 1025 | } |
1da177e4 | 1026 | } |
7c00bafe TC |
1027 | atomic_long_inc(&nr_swap_pages); |
1028 | p->inuse_pages--; | |
1029 | frontswap_invalidate_page(p->type, offset); | |
1030 | if (p->flags & SWP_BLKDEV) { | |
1031 | struct gendisk *disk = p->bdev->bd_disk; | |
1032 | ||
1033 | if (disk->fops->swap_slot_free_notify) | |
1034 | disk->fops->swap_slot_free_notify(p->bdev, | |
1035 | offset); | |
1036 | } | |
1da177e4 LT |
1037 | } |
1038 | ||
1039 | /* | |
2de1a7e4 | 1040 | * Caller has made sure that the swap device corresponding to entry |
1da177e4 LT |
1041 | * is still around or has not been recycled. |
1042 | */ | |
1043 | void swap_free(swp_entry_t entry) | |
1044 | { | |
73c34b6a | 1045 | struct swap_info_struct *p; |
1da177e4 | 1046 | |
235b6217 | 1047 | p = _swap_info_get(entry); |
7c00bafe TC |
1048 | if (p) { |
1049 | if (!__swap_entry_free(p, entry, 1)) | |
67afa38e | 1050 | free_swap_slot(entry); |
7c00bafe | 1051 | } |
1da177e4 LT |
1052 | } |
1053 | ||
cb4b86ba KH |
1054 | /* |
1055 | * Called after dropping swapcache to decrease refcnt to swap entries. | |
1056 | */ | |
0a31bc97 | 1057 | void swapcache_free(swp_entry_t entry) |
cb4b86ba | 1058 | { |
355cfa73 KH |
1059 | struct swap_info_struct *p; |
1060 | ||
235b6217 | 1061 | p = _swap_info_get(entry); |
7c00bafe TC |
1062 | if (p) { |
1063 | if (!__swap_entry_free(p, entry, SWAP_HAS_CACHE)) | |
67afa38e | 1064 | free_swap_slot(entry); |
7c00bafe TC |
1065 | } |
1066 | } | |
1067 | ||
1068 | void swapcache_free_entries(swp_entry_t *entries, int n) | |
1069 | { | |
1070 | struct swap_info_struct *p, *prev; | |
1071 | int i; | |
1072 | ||
1073 | if (n <= 0) | |
1074 | return; | |
1075 | ||
1076 | prev = NULL; | |
1077 | p = NULL; | |
1078 | for (i = 0; i < n; ++i) { | |
1079 | p = swap_info_get_cont(entries[i], prev); | |
1080 | if (p) | |
1081 | swap_entry_free(p, entries[i]); | |
1082 | else | |
1083 | break; | |
1084 | prev = p; | |
1085 | } | |
235b6217 | 1086 | if (p) |
7c00bafe | 1087 | spin_unlock(&p->lock); |
cb4b86ba KH |
1088 | } |
1089 | ||
1da177e4 | 1090 | /* |
c475a8ab | 1091 | * How many references to page are currently swapped out? |
570a335b HD |
1092 | * This does not give an exact answer when swap count is continued, |
1093 | * but does include the high COUNT_CONTINUED flag to allow for that. | |
1da177e4 | 1094 | */ |
bde05d1c | 1095 | int page_swapcount(struct page *page) |
1da177e4 | 1096 | { |
c475a8ab HD |
1097 | int count = 0; |
1098 | struct swap_info_struct *p; | |
235b6217 | 1099 | struct swap_cluster_info *ci; |
1da177e4 | 1100 | swp_entry_t entry; |
235b6217 | 1101 | unsigned long offset; |
1da177e4 | 1102 | |
4c21e2f2 | 1103 | entry.val = page_private(page); |
235b6217 | 1104 | p = _swap_info_get(entry); |
1da177e4 | 1105 | if (p) { |
235b6217 HY |
1106 | offset = swp_offset(entry); |
1107 | ci = lock_cluster_or_swap_info(p, offset); | |
1108 | count = swap_count(p->swap_map[offset]); | |
1109 | unlock_cluster_or_swap_info(p, ci); | |
1da177e4 | 1110 | } |
c475a8ab | 1111 | return count; |
1da177e4 LT |
1112 | } |
1113 | ||
e8c26ab6 TC |
1114 | /* |
1115 | * How many references to @entry are currently swapped out? | |
1116 | * This does not give an exact answer when swap count is continued, | |
1117 | * but does include the high COUNT_CONTINUED flag to allow for that. | |
1118 | */ | |
1119 | int __swp_swapcount(swp_entry_t entry) | |
1120 | { | |
1121 | int count = 0; | |
1122 | pgoff_t offset; | |
1123 | struct swap_info_struct *si; | |
1124 | struct swap_cluster_info *ci; | |
1125 | ||
1126 | si = __swap_info_get(entry); | |
1127 | if (si) { | |
1128 | offset = swp_offset(entry); | |
1129 | ci = lock_cluster_or_swap_info(si, offset); | |
1130 | count = swap_count(si->swap_map[offset]); | |
1131 | unlock_cluster_or_swap_info(si, ci); | |
1132 | } | |
1133 | return count; | |
1134 | } | |
1135 | ||
8334b962 MK |
1136 | /* |
1137 | * How many references to @entry are currently swapped out? | |
1138 | * This considers COUNT_CONTINUED so it returns exact answer. | |
1139 | */ | |
1140 | int swp_swapcount(swp_entry_t entry) | |
1141 | { | |
1142 | int count, tmp_count, n; | |
1143 | struct swap_info_struct *p; | |
235b6217 | 1144 | struct swap_cluster_info *ci; |
8334b962 MK |
1145 | struct page *page; |
1146 | pgoff_t offset; | |
1147 | unsigned char *map; | |
1148 | ||
235b6217 | 1149 | p = _swap_info_get(entry); |
8334b962 MK |
1150 | if (!p) |
1151 | return 0; | |
1152 | ||
235b6217 HY |
1153 | offset = swp_offset(entry); |
1154 | ||
1155 | ci = lock_cluster_or_swap_info(p, offset); | |
1156 | ||
1157 | count = swap_count(p->swap_map[offset]); | |
8334b962 MK |
1158 | if (!(count & COUNT_CONTINUED)) |
1159 | goto out; | |
1160 | ||
1161 | count &= ~COUNT_CONTINUED; | |
1162 | n = SWAP_MAP_MAX + 1; | |
1163 | ||
8334b962 MK |
1164 | page = vmalloc_to_page(p->swap_map + offset); |
1165 | offset &= ~PAGE_MASK; | |
1166 | VM_BUG_ON(page_private(page) != SWP_CONTINUED); | |
1167 | ||
1168 | do { | |
a8ae4991 | 1169 | page = list_next_entry(page, lru); |
8334b962 MK |
1170 | map = kmap_atomic(page); |
1171 | tmp_count = map[offset]; | |
1172 | kunmap_atomic(map); | |
1173 | ||
1174 | count += (tmp_count & ~COUNT_CONTINUED) * n; | |
1175 | n *= (SWAP_CONT_MAX + 1); | |
1176 | } while (tmp_count & COUNT_CONTINUED); | |
1177 | out: | |
235b6217 | 1178 | unlock_cluster_or_swap_info(p, ci); |
8334b962 MK |
1179 | return count; |
1180 | } | |
1181 | ||
1da177e4 | 1182 | /* |
7b1fe597 HD |
1183 | * We can write to an anon page without COW if there are no other references |
1184 | * to it. And as a side-effect, free up its swap: because the old content | |
1185 | * on disk will never be read, and seeking back there to write new content | |
1186 | * later would only waste time away from clustering. | |
6d0a07ed AA |
1187 | * |
1188 | * NOTE: total_mapcount should not be relied upon by the caller if | |
1189 | * reuse_swap_page() returns false, but it may be always overwritten | |
1190 | * (see the other implementation for CONFIG_SWAP=n). | |
1da177e4 | 1191 | */ |
6d0a07ed | 1192 | bool reuse_swap_page(struct page *page, int *total_mapcount) |
1da177e4 | 1193 | { |
c475a8ab HD |
1194 | int count; |
1195 | ||
309381fe | 1196 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
5ad64688 | 1197 | if (unlikely(PageKsm(page))) |
6d0a07ed AA |
1198 | return false; |
1199 | count = page_trans_huge_mapcount(page, total_mapcount); | |
7b1fe597 | 1200 | if (count <= 1 && PageSwapCache(page)) { |
c475a8ab | 1201 | count += page_swapcount(page); |
f0571429 MK |
1202 | if (count != 1) |
1203 | goto out; | |
1204 | if (!PageWriteback(page)) { | |
7b1fe597 HD |
1205 | delete_from_swap_cache(page); |
1206 | SetPageDirty(page); | |
f0571429 MK |
1207 | } else { |
1208 | swp_entry_t entry; | |
1209 | struct swap_info_struct *p; | |
1210 | ||
1211 | entry.val = page_private(page); | |
1212 | p = swap_info_get(entry); | |
1213 | if (p->flags & SWP_STABLE_WRITES) { | |
1214 | spin_unlock(&p->lock); | |
1215 | return false; | |
1216 | } | |
1217 | spin_unlock(&p->lock); | |
7b1fe597 HD |
1218 | } |
1219 | } | |
f0571429 | 1220 | out: |
5ad64688 | 1221 | return count <= 1; |
1da177e4 LT |
1222 | } |
1223 | ||
1224 | /* | |
a2c43eed HD |
1225 | * If swap is getting full, or if there are no more mappings of this page, |
1226 | * then try_to_free_swap is called to free its swap space. | |
1da177e4 | 1227 | */ |
a2c43eed | 1228 | int try_to_free_swap(struct page *page) |
1da177e4 | 1229 | { |
309381fe | 1230 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
1da177e4 LT |
1231 | |
1232 | if (!PageSwapCache(page)) | |
1233 | return 0; | |
1234 | if (PageWriteback(page)) | |
1235 | return 0; | |
a2c43eed | 1236 | if (page_swapcount(page)) |
1da177e4 LT |
1237 | return 0; |
1238 | ||
b73d7fce HD |
1239 | /* |
1240 | * Once hibernation has begun to create its image of memory, | |
1241 | * there's a danger that one of the calls to try_to_free_swap() | |
1242 | * - most probably a call from __try_to_reclaim_swap() while | |
1243 | * hibernation is allocating its own swap pages for the image, | |
1244 | * but conceivably even a call from memory reclaim - will free | |
1245 | * the swap from a page which has already been recorded in the | |
1246 | * image as a clean swapcache page, and then reuse its swap for | |
1247 | * another page of the image. On waking from hibernation, the | |
1248 | * original page might be freed under memory pressure, then | |
1249 | * later read back in from swap, now with the wrong data. | |
1250 | * | |
2de1a7e4 | 1251 | * Hibernation suspends storage while it is writing the image |
f90ac398 | 1252 | * to disk so check that here. |
b73d7fce | 1253 | */ |
f90ac398 | 1254 | if (pm_suspended_storage()) |
b73d7fce HD |
1255 | return 0; |
1256 | ||
a2c43eed HD |
1257 | delete_from_swap_cache(page); |
1258 | SetPageDirty(page); | |
1259 | return 1; | |
68a22394 RR |
1260 | } |
1261 | ||
1da177e4 LT |
1262 | /* |
1263 | * Free the swap entry like above, but also try to | |
1264 | * free the page cache entry if it is the last user. | |
1265 | */ | |
2509ef26 | 1266 | int free_swap_and_cache(swp_entry_t entry) |
1da177e4 | 1267 | { |
2509ef26 | 1268 | struct swap_info_struct *p; |
1da177e4 | 1269 | struct page *page = NULL; |
7c00bafe | 1270 | unsigned char count; |
1da177e4 | 1271 | |
a7420aa5 | 1272 | if (non_swap_entry(entry)) |
2509ef26 | 1273 | return 1; |
0697212a | 1274 | |
7c00bafe | 1275 | p = _swap_info_get(entry); |
1da177e4 | 1276 | if (p) { |
7c00bafe TC |
1277 | count = __swap_entry_free(p, entry, 1); |
1278 | if (count == SWAP_HAS_CACHE) { | |
33806f06 | 1279 | page = find_get_page(swap_address_space(entry), |
f6ab1f7f | 1280 | swp_offset(entry)); |
8413ac9d | 1281 | if (page && !trylock_page(page)) { |
09cbfeaf | 1282 | put_page(page); |
93fac704 NP |
1283 | page = NULL; |
1284 | } | |
7c00bafe | 1285 | } else if (!count) |
67afa38e | 1286 | free_swap_slot(entry); |
1da177e4 LT |
1287 | } |
1288 | if (page) { | |
a2c43eed HD |
1289 | /* |
1290 | * Not mapped elsewhere, or swap space full? Free it! | |
1291 | * Also recheck PageSwapCache now page is locked (above). | |
1292 | */ | |
93fac704 | 1293 | if (PageSwapCache(page) && !PageWriteback(page) && |
5ccc5aba | 1294 | (!page_mapped(page) || mem_cgroup_swap_full(page))) { |
1da177e4 LT |
1295 | delete_from_swap_cache(page); |
1296 | SetPageDirty(page); | |
1297 | } | |
1298 | unlock_page(page); | |
09cbfeaf | 1299 | put_page(page); |
1da177e4 | 1300 | } |
2509ef26 | 1301 | return p != NULL; |
1da177e4 LT |
1302 | } |
1303 | ||
b0cb1a19 | 1304 | #ifdef CONFIG_HIBERNATION |
f577eb30 | 1305 | /* |
915bae9e | 1306 | * Find the swap type that corresponds to given device (if any). |
f577eb30 | 1307 | * |
915bae9e RW |
1308 | * @offset - number of the PAGE_SIZE-sized block of the device, starting |
1309 | * from 0, in which the swap header is expected to be located. | |
1310 | * | |
1311 | * This is needed for the suspend to disk (aka swsusp). | |
f577eb30 | 1312 | */ |
7bf23687 | 1313 | int swap_type_of(dev_t device, sector_t offset, struct block_device **bdev_p) |
f577eb30 | 1314 | { |
915bae9e | 1315 | struct block_device *bdev = NULL; |
efa90a98 | 1316 | int type; |
f577eb30 | 1317 | |
915bae9e RW |
1318 | if (device) |
1319 | bdev = bdget(device); | |
1320 | ||
f577eb30 | 1321 | spin_lock(&swap_lock); |
efa90a98 HD |
1322 | for (type = 0; type < nr_swapfiles; type++) { |
1323 | struct swap_info_struct *sis = swap_info[type]; | |
f577eb30 | 1324 | |
915bae9e | 1325 | if (!(sis->flags & SWP_WRITEOK)) |
f577eb30 | 1326 | continue; |
b6b5bce3 | 1327 | |
915bae9e | 1328 | if (!bdev) { |
7bf23687 | 1329 | if (bdev_p) |
dddac6a7 | 1330 | *bdev_p = bdgrab(sis->bdev); |
7bf23687 | 1331 | |
6e1819d6 | 1332 | spin_unlock(&swap_lock); |
efa90a98 | 1333 | return type; |
6e1819d6 | 1334 | } |
915bae9e | 1335 | if (bdev == sis->bdev) { |
9625a5f2 | 1336 | struct swap_extent *se = &sis->first_swap_extent; |
915bae9e | 1337 | |
915bae9e | 1338 | if (se->start_block == offset) { |
7bf23687 | 1339 | if (bdev_p) |
dddac6a7 | 1340 | *bdev_p = bdgrab(sis->bdev); |
7bf23687 | 1341 | |
915bae9e RW |
1342 | spin_unlock(&swap_lock); |
1343 | bdput(bdev); | |
efa90a98 | 1344 | return type; |
915bae9e | 1345 | } |
f577eb30 RW |
1346 | } |
1347 | } | |
1348 | spin_unlock(&swap_lock); | |
915bae9e RW |
1349 | if (bdev) |
1350 | bdput(bdev); | |
1351 | ||
f577eb30 RW |
1352 | return -ENODEV; |
1353 | } | |
1354 | ||
73c34b6a HD |
1355 | /* |
1356 | * Get the (PAGE_SIZE) block corresponding to given offset on the swapdev | |
1357 | * corresponding to given index in swap_info (swap type). | |
1358 | */ | |
1359 | sector_t swapdev_block(int type, pgoff_t offset) | |
1360 | { | |
1361 | struct block_device *bdev; | |
1362 | ||
1363 | if ((unsigned int)type >= nr_swapfiles) | |
1364 | return 0; | |
1365 | if (!(swap_info[type]->flags & SWP_WRITEOK)) | |
1366 | return 0; | |
d4906e1a | 1367 | return map_swap_entry(swp_entry(type, offset), &bdev); |
73c34b6a HD |
1368 | } |
1369 | ||
f577eb30 RW |
1370 | /* |
1371 | * Return either the total number of swap pages of given type, or the number | |
1372 | * of free pages of that type (depending on @free) | |
1373 | * | |
1374 | * This is needed for software suspend | |
1375 | */ | |
1376 | unsigned int count_swap_pages(int type, int free) | |
1377 | { | |
1378 | unsigned int n = 0; | |
1379 | ||
efa90a98 HD |
1380 | spin_lock(&swap_lock); |
1381 | if ((unsigned int)type < nr_swapfiles) { | |
1382 | struct swap_info_struct *sis = swap_info[type]; | |
1383 | ||
ec8acf20 | 1384 | spin_lock(&sis->lock); |
efa90a98 HD |
1385 | if (sis->flags & SWP_WRITEOK) { |
1386 | n = sis->pages; | |
f577eb30 | 1387 | if (free) |
efa90a98 | 1388 | n -= sis->inuse_pages; |
f577eb30 | 1389 | } |
ec8acf20 | 1390 | spin_unlock(&sis->lock); |
f577eb30 | 1391 | } |
efa90a98 | 1392 | spin_unlock(&swap_lock); |
f577eb30 RW |
1393 | return n; |
1394 | } | |
73c34b6a | 1395 | #endif /* CONFIG_HIBERNATION */ |
f577eb30 | 1396 | |
9f8bdb3f | 1397 | static inline int pte_same_as_swp(pte_t pte, pte_t swp_pte) |
179ef71c | 1398 | { |
9f8bdb3f | 1399 | return pte_same(pte_swp_clear_soft_dirty(pte), swp_pte); |
179ef71c CG |
1400 | } |
1401 | ||
1da177e4 | 1402 | /* |
72866f6f HD |
1403 | * No need to decide whether this PTE shares the swap entry with others, |
1404 | * just let do_wp_page work it out if a write is requested later - to | |
1405 | * force COW, vm_page_prot omits write permission from any private vma. | |
1da177e4 | 1406 | */ |
044d66c1 | 1407 | static int unuse_pte(struct vm_area_struct *vma, pmd_t *pmd, |
1da177e4 LT |
1408 | unsigned long addr, swp_entry_t entry, struct page *page) |
1409 | { | |
9e16b7fb | 1410 | struct page *swapcache; |
72835c86 | 1411 | struct mem_cgroup *memcg; |
044d66c1 HD |
1412 | spinlock_t *ptl; |
1413 | pte_t *pte; | |
1414 | int ret = 1; | |
1415 | ||
9e16b7fb HD |
1416 | swapcache = page; |
1417 | page = ksm_might_need_to_copy(page, vma, addr); | |
1418 | if (unlikely(!page)) | |
1419 | return -ENOMEM; | |
1420 | ||
f627c2f5 KS |
1421 | if (mem_cgroup_try_charge(page, vma->vm_mm, GFP_KERNEL, |
1422 | &memcg, false)) { | |
044d66c1 | 1423 | ret = -ENOMEM; |
85d9fc89 KH |
1424 | goto out_nolock; |
1425 | } | |
044d66c1 HD |
1426 | |
1427 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); | |
9f8bdb3f | 1428 | if (unlikely(!pte_same_as_swp(*pte, swp_entry_to_pte(entry)))) { |
f627c2f5 | 1429 | mem_cgroup_cancel_charge(page, memcg, false); |
044d66c1 HD |
1430 | ret = 0; |
1431 | goto out; | |
1432 | } | |
8a9f3ccd | 1433 | |
b084d435 | 1434 | dec_mm_counter(vma->vm_mm, MM_SWAPENTS); |
d559db08 | 1435 | inc_mm_counter(vma->vm_mm, MM_ANONPAGES); |
1da177e4 LT |
1436 | get_page(page); |
1437 | set_pte_at(vma->vm_mm, addr, pte, | |
1438 | pte_mkold(mk_pte(page, vma->vm_page_prot))); | |
00501b53 | 1439 | if (page == swapcache) { |
d281ee61 | 1440 | page_add_anon_rmap(page, vma, addr, false); |
f627c2f5 | 1441 | mem_cgroup_commit_charge(page, memcg, true, false); |
00501b53 | 1442 | } else { /* ksm created a completely new copy */ |
d281ee61 | 1443 | page_add_new_anon_rmap(page, vma, addr, false); |
f627c2f5 | 1444 | mem_cgroup_commit_charge(page, memcg, false, false); |
00501b53 JW |
1445 | lru_cache_add_active_or_unevictable(page, vma); |
1446 | } | |
1da177e4 LT |
1447 | swap_free(entry); |
1448 | /* | |
1449 | * Move the page to the active list so it is not | |
1450 | * immediately swapped out again after swapon. | |
1451 | */ | |
1452 | activate_page(page); | |
044d66c1 HD |
1453 | out: |
1454 | pte_unmap_unlock(pte, ptl); | |
85d9fc89 | 1455 | out_nolock: |
9e16b7fb HD |
1456 | if (page != swapcache) { |
1457 | unlock_page(page); | |
1458 | put_page(page); | |
1459 | } | |
044d66c1 | 1460 | return ret; |
1da177e4 LT |
1461 | } |
1462 | ||
1463 | static int unuse_pte_range(struct vm_area_struct *vma, pmd_t *pmd, | |
1464 | unsigned long addr, unsigned long end, | |
1465 | swp_entry_t entry, struct page *page) | |
1466 | { | |
1da177e4 | 1467 | pte_t swp_pte = swp_entry_to_pte(entry); |
705e87c0 | 1468 | pte_t *pte; |
8a9f3ccd | 1469 | int ret = 0; |
1da177e4 | 1470 | |
044d66c1 HD |
1471 | /* |
1472 | * We don't actually need pte lock while scanning for swp_pte: since | |
1473 | * we hold page lock and mmap_sem, swp_pte cannot be inserted into the | |
1474 | * page table while we're scanning; though it could get zapped, and on | |
1475 | * some architectures (e.g. x86_32 with PAE) we might catch a glimpse | |
1476 | * of unmatched parts which look like swp_pte, so unuse_pte must | |
1477 | * recheck under pte lock. Scanning without pte lock lets it be | |
2de1a7e4 | 1478 | * preemptable whenever CONFIG_PREEMPT but not CONFIG_HIGHPTE. |
044d66c1 HD |
1479 | */ |
1480 | pte = pte_offset_map(pmd, addr); | |
1da177e4 LT |
1481 | do { |
1482 | /* | |
1483 | * swapoff spends a _lot_ of time in this loop! | |
1484 | * Test inline before going to call unuse_pte. | |
1485 | */ | |
9f8bdb3f | 1486 | if (unlikely(pte_same_as_swp(*pte, swp_pte))) { |
044d66c1 HD |
1487 | pte_unmap(pte); |
1488 | ret = unuse_pte(vma, pmd, addr, entry, page); | |
1489 | if (ret) | |
1490 | goto out; | |
1491 | pte = pte_offset_map(pmd, addr); | |
1da177e4 LT |
1492 | } |
1493 | } while (pte++, addr += PAGE_SIZE, addr != end); | |
044d66c1 HD |
1494 | pte_unmap(pte - 1); |
1495 | out: | |
8a9f3ccd | 1496 | return ret; |
1da177e4 LT |
1497 | } |
1498 | ||
1499 | static inline int unuse_pmd_range(struct vm_area_struct *vma, pud_t *pud, | |
1500 | unsigned long addr, unsigned long end, | |
1501 | swp_entry_t entry, struct page *page) | |
1502 | { | |
1503 | pmd_t *pmd; | |
1504 | unsigned long next; | |
8a9f3ccd | 1505 | int ret; |
1da177e4 LT |
1506 | |
1507 | pmd = pmd_offset(pud, addr); | |
1508 | do { | |
dc644a07 | 1509 | cond_resched(); |
1da177e4 | 1510 | next = pmd_addr_end(addr, end); |
1a5a9906 | 1511 | if (pmd_none_or_trans_huge_or_clear_bad(pmd)) |
1da177e4 | 1512 | continue; |
8a9f3ccd BS |
1513 | ret = unuse_pte_range(vma, pmd, addr, next, entry, page); |
1514 | if (ret) | |
1515 | return ret; | |
1da177e4 LT |
1516 | } while (pmd++, addr = next, addr != end); |
1517 | return 0; | |
1518 | } | |
1519 | ||
c2febafc | 1520 | static inline int unuse_pud_range(struct vm_area_struct *vma, p4d_t *p4d, |
1da177e4 LT |
1521 | unsigned long addr, unsigned long end, |
1522 | swp_entry_t entry, struct page *page) | |
1523 | { | |
1524 | pud_t *pud; | |
1525 | unsigned long next; | |
8a9f3ccd | 1526 | int ret; |
1da177e4 | 1527 | |
c2febafc | 1528 | pud = pud_offset(p4d, addr); |
1da177e4 LT |
1529 | do { |
1530 | next = pud_addr_end(addr, end); | |
1531 | if (pud_none_or_clear_bad(pud)) | |
1532 | continue; | |
8a9f3ccd BS |
1533 | ret = unuse_pmd_range(vma, pud, addr, next, entry, page); |
1534 | if (ret) | |
1535 | return ret; | |
1da177e4 LT |
1536 | } while (pud++, addr = next, addr != end); |
1537 | return 0; | |
1538 | } | |
1539 | ||
c2febafc KS |
1540 | static inline int unuse_p4d_range(struct vm_area_struct *vma, pgd_t *pgd, |
1541 | unsigned long addr, unsigned long end, | |
1542 | swp_entry_t entry, struct page *page) | |
1543 | { | |
1544 | p4d_t *p4d; | |
1545 | unsigned long next; | |
1546 | int ret; | |
1547 | ||
1548 | p4d = p4d_offset(pgd, addr); | |
1549 | do { | |
1550 | next = p4d_addr_end(addr, end); | |
1551 | if (p4d_none_or_clear_bad(p4d)) | |
1552 | continue; | |
1553 | ret = unuse_pud_range(vma, p4d, addr, next, entry, page); | |
1554 | if (ret) | |
1555 | return ret; | |
1556 | } while (p4d++, addr = next, addr != end); | |
1557 | return 0; | |
1558 | } | |
1559 | ||
1da177e4 LT |
1560 | static int unuse_vma(struct vm_area_struct *vma, |
1561 | swp_entry_t entry, struct page *page) | |
1562 | { | |
1563 | pgd_t *pgd; | |
1564 | unsigned long addr, end, next; | |
8a9f3ccd | 1565 | int ret; |
1da177e4 | 1566 | |
3ca7b3c5 | 1567 | if (page_anon_vma(page)) { |
1da177e4 LT |
1568 | addr = page_address_in_vma(page, vma); |
1569 | if (addr == -EFAULT) | |
1570 | return 0; | |
1571 | else | |
1572 | end = addr + PAGE_SIZE; | |
1573 | } else { | |
1574 | addr = vma->vm_start; | |
1575 | end = vma->vm_end; | |
1576 | } | |
1577 | ||
1578 | pgd = pgd_offset(vma->vm_mm, addr); | |
1579 | do { | |
1580 | next = pgd_addr_end(addr, end); | |
1581 | if (pgd_none_or_clear_bad(pgd)) | |
1582 | continue; | |
c2febafc | 1583 | ret = unuse_p4d_range(vma, pgd, addr, next, entry, page); |
8a9f3ccd BS |
1584 | if (ret) |
1585 | return ret; | |
1da177e4 LT |
1586 | } while (pgd++, addr = next, addr != end); |
1587 | return 0; | |
1588 | } | |
1589 | ||
1590 | static int unuse_mm(struct mm_struct *mm, | |
1591 | swp_entry_t entry, struct page *page) | |
1592 | { | |
1593 | struct vm_area_struct *vma; | |
8a9f3ccd | 1594 | int ret = 0; |
1da177e4 LT |
1595 | |
1596 | if (!down_read_trylock(&mm->mmap_sem)) { | |
1597 | /* | |
7d03431c FLVC |
1598 | * Activate page so shrink_inactive_list is unlikely to unmap |
1599 | * its ptes while lock is dropped, so swapoff can make progress. | |
1da177e4 | 1600 | */ |
c475a8ab | 1601 | activate_page(page); |
1da177e4 LT |
1602 | unlock_page(page); |
1603 | down_read(&mm->mmap_sem); | |
1604 | lock_page(page); | |
1605 | } | |
1da177e4 | 1606 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
8a9f3ccd | 1607 | if (vma->anon_vma && (ret = unuse_vma(vma, entry, page))) |
1da177e4 | 1608 | break; |
dc644a07 | 1609 | cond_resched(); |
1da177e4 | 1610 | } |
1da177e4 | 1611 | up_read(&mm->mmap_sem); |
8a9f3ccd | 1612 | return (ret < 0)? ret: 0; |
1da177e4 LT |
1613 | } |
1614 | ||
1615 | /* | |
38b5faf4 DM |
1616 | * Scan swap_map (or frontswap_map if frontswap parameter is true) |
1617 | * from current position to next entry still in use. | |
1da177e4 LT |
1618 | * Recycle to start on reaching the end, returning 0 when empty. |
1619 | */ | |
6eb396dc | 1620 | static unsigned int find_next_to_unuse(struct swap_info_struct *si, |
38b5faf4 | 1621 | unsigned int prev, bool frontswap) |
1da177e4 | 1622 | { |
6eb396dc HD |
1623 | unsigned int max = si->max; |
1624 | unsigned int i = prev; | |
8d69aaee | 1625 | unsigned char count; |
1da177e4 LT |
1626 | |
1627 | /* | |
5d337b91 | 1628 | * No need for swap_lock here: we're just looking |
1da177e4 LT |
1629 | * for whether an entry is in use, not modifying it; false |
1630 | * hits are okay, and sys_swapoff() has already prevented new | |
5d337b91 | 1631 | * allocations from this area (while holding swap_lock). |
1da177e4 LT |
1632 | */ |
1633 | for (;;) { | |
1634 | if (++i >= max) { | |
1635 | if (!prev) { | |
1636 | i = 0; | |
1637 | break; | |
1638 | } | |
1639 | /* | |
1640 | * No entries in use at top of swap_map, | |
1641 | * loop back to start and recheck there. | |
1642 | */ | |
1643 | max = prev + 1; | |
1644 | prev = 0; | |
1645 | i = 1; | |
1646 | } | |
4db0c3c2 | 1647 | count = READ_ONCE(si->swap_map[i]); |
355cfa73 | 1648 | if (count && swap_count(count) != SWAP_MAP_BAD) |
dc644a07 HD |
1649 | if (!frontswap || frontswap_test(si, i)) |
1650 | break; | |
1651 | if ((i % LATENCY_LIMIT) == 0) | |
1652 | cond_resched(); | |
1da177e4 LT |
1653 | } |
1654 | return i; | |
1655 | } | |
1656 | ||
1657 | /* | |
1658 | * We completely avoid races by reading each swap page in advance, | |
1659 | * and then search for the process using it. All the necessary | |
1660 | * page table adjustments can then be made atomically. | |
38b5faf4 DM |
1661 | * |
1662 | * if the boolean frontswap is true, only unuse pages_to_unuse pages; | |
1663 | * pages_to_unuse==0 means all pages; ignored if frontswap is false | |
1da177e4 | 1664 | */ |
38b5faf4 DM |
1665 | int try_to_unuse(unsigned int type, bool frontswap, |
1666 | unsigned long pages_to_unuse) | |
1da177e4 | 1667 | { |
efa90a98 | 1668 | struct swap_info_struct *si = swap_info[type]; |
1da177e4 | 1669 | struct mm_struct *start_mm; |
edfe23da SL |
1670 | volatile unsigned char *swap_map; /* swap_map is accessed without |
1671 | * locking. Mark it as volatile | |
1672 | * to prevent compiler doing | |
1673 | * something odd. | |
1674 | */ | |
8d69aaee | 1675 | unsigned char swcount; |
1da177e4 LT |
1676 | struct page *page; |
1677 | swp_entry_t entry; | |
6eb396dc | 1678 | unsigned int i = 0; |
1da177e4 | 1679 | int retval = 0; |
1da177e4 LT |
1680 | |
1681 | /* | |
1682 | * When searching mms for an entry, a good strategy is to | |
1683 | * start at the first mm we freed the previous entry from | |
1684 | * (though actually we don't notice whether we or coincidence | |
1685 | * freed the entry). Initialize this start_mm with a hold. | |
1686 | * | |
1687 | * A simpler strategy would be to start at the last mm we | |
1688 | * freed the previous entry from; but that would take less | |
1689 | * advantage of mmlist ordering, which clusters forked mms | |
1690 | * together, child after parent. If we race with dup_mmap(), we | |
1691 | * prefer to resolve parent before child, lest we miss entries | |
1692 | * duplicated after we scanned child: using last mm would invert | |
570a335b | 1693 | * that. |
1da177e4 LT |
1694 | */ |
1695 | start_mm = &init_mm; | |
3fce371b | 1696 | mmget(&init_mm); |
1da177e4 LT |
1697 | |
1698 | /* | |
1699 | * Keep on scanning until all entries have gone. Usually, | |
1700 | * one pass through swap_map is enough, but not necessarily: | |
1701 | * there are races when an instance of an entry might be missed. | |
1702 | */ | |
38b5faf4 | 1703 | while ((i = find_next_to_unuse(si, i, frontswap)) != 0) { |
1da177e4 LT |
1704 | if (signal_pending(current)) { |
1705 | retval = -EINTR; | |
1706 | break; | |
1707 | } | |
1708 | ||
886bb7e9 | 1709 | /* |
1da177e4 LT |
1710 | * Get a page for the entry, using the existing swap |
1711 | * cache page if there is one. Otherwise, get a clean | |
886bb7e9 | 1712 | * page and read the swap into it. |
1da177e4 LT |
1713 | */ |
1714 | swap_map = &si->swap_map[i]; | |
1715 | entry = swp_entry(type, i); | |
02098fea HD |
1716 | page = read_swap_cache_async(entry, |
1717 | GFP_HIGHUSER_MOVABLE, NULL, 0); | |
1da177e4 LT |
1718 | if (!page) { |
1719 | /* | |
1720 | * Either swap_duplicate() failed because entry | |
1721 | * has been freed independently, and will not be | |
1722 | * reused since sys_swapoff() already disabled | |
1723 | * allocation from here, or alloc_page() failed. | |
1724 | */ | |
edfe23da SL |
1725 | swcount = *swap_map; |
1726 | /* | |
1727 | * We don't hold lock here, so the swap entry could be | |
1728 | * SWAP_MAP_BAD (when the cluster is discarding). | |
1729 | * Instead of fail out, We can just skip the swap | |
1730 | * entry because swapoff will wait for discarding | |
1731 | * finish anyway. | |
1732 | */ | |
1733 | if (!swcount || swcount == SWAP_MAP_BAD) | |
1da177e4 LT |
1734 | continue; |
1735 | retval = -ENOMEM; | |
1736 | break; | |
1737 | } | |
1738 | ||
1739 | /* | |
1740 | * Don't hold on to start_mm if it looks like exiting. | |
1741 | */ | |
1742 | if (atomic_read(&start_mm->mm_users) == 1) { | |
1743 | mmput(start_mm); | |
1744 | start_mm = &init_mm; | |
3fce371b | 1745 | mmget(&init_mm); |
1da177e4 LT |
1746 | } |
1747 | ||
1748 | /* | |
1749 | * Wait for and lock page. When do_swap_page races with | |
1750 | * try_to_unuse, do_swap_page can handle the fault much | |
1751 | * faster than try_to_unuse can locate the entry. This | |
1752 | * apparently redundant "wait_on_page_locked" lets try_to_unuse | |
1753 | * defer to do_swap_page in such a case - in some tests, | |
1754 | * do_swap_page and try_to_unuse repeatedly compete. | |
1755 | */ | |
1756 | wait_on_page_locked(page); | |
1757 | wait_on_page_writeback(page); | |
1758 | lock_page(page); | |
1759 | wait_on_page_writeback(page); | |
1760 | ||
1761 | /* | |
1762 | * Remove all references to entry. | |
1da177e4 | 1763 | */ |
1da177e4 | 1764 | swcount = *swap_map; |
aaa46865 HD |
1765 | if (swap_count(swcount) == SWAP_MAP_SHMEM) { |
1766 | retval = shmem_unuse(entry, page); | |
1767 | /* page has already been unlocked and released */ | |
1768 | if (retval < 0) | |
1769 | break; | |
1770 | continue; | |
1da177e4 | 1771 | } |
aaa46865 HD |
1772 | if (swap_count(swcount) && start_mm != &init_mm) |
1773 | retval = unuse_mm(start_mm, entry, page); | |
1774 | ||
355cfa73 | 1775 | if (swap_count(*swap_map)) { |
1da177e4 LT |
1776 | int set_start_mm = (*swap_map >= swcount); |
1777 | struct list_head *p = &start_mm->mmlist; | |
1778 | struct mm_struct *new_start_mm = start_mm; | |
1779 | struct mm_struct *prev_mm = start_mm; | |
1780 | struct mm_struct *mm; | |
1781 | ||
3fce371b VN |
1782 | mmget(new_start_mm); |
1783 | mmget(prev_mm); | |
1da177e4 | 1784 | spin_lock(&mmlist_lock); |
aaa46865 | 1785 | while (swap_count(*swap_map) && !retval && |
1da177e4 LT |
1786 | (p = p->next) != &start_mm->mmlist) { |
1787 | mm = list_entry(p, struct mm_struct, mmlist); | |
388f7934 | 1788 | if (!mmget_not_zero(mm)) |
1da177e4 | 1789 | continue; |
1da177e4 LT |
1790 | spin_unlock(&mmlist_lock); |
1791 | mmput(prev_mm); | |
1792 | prev_mm = mm; | |
1793 | ||
1794 | cond_resched(); | |
1795 | ||
1796 | swcount = *swap_map; | |
355cfa73 | 1797 | if (!swap_count(swcount)) /* any usage ? */ |
1da177e4 | 1798 | ; |
aaa46865 | 1799 | else if (mm == &init_mm) |
1da177e4 | 1800 | set_start_mm = 1; |
aaa46865 | 1801 | else |
1da177e4 | 1802 | retval = unuse_mm(mm, entry, page); |
355cfa73 | 1803 | |
32c5fc10 | 1804 | if (set_start_mm && *swap_map < swcount) { |
1da177e4 | 1805 | mmput(new_start_mm); |
3fce371b | 1806 | mmget(mm); |
1da177e4 LT |
1807 | new_start_mm = mm; |
1808 | set_start_mm = 0; | |
1809 | } | |
1810 | spin_lock(&mmlist_lock); | |
1811 | } | |
1812 | spin_unlock(&mmlist_lock); | |
1813 | mmput(prev_mm); | |
1814 | mmput(start_mm); | |
1815 | start_mm = new_start_mm; | |
1816 | } | |
1817 | if (retval) { | |
1818 | unlock_page(page); | |
09cbfeaf | 1819 | put_page(page); |
1da177e4 LT |
1820 | break; |
1821 | } | |
1822 | ||
1da177e4 LT |
1823 | /* |
1824 | * If a reference remains (rare), we would like to leave | |
1825 | * the page in the swap cache; but try_to_unmap could | |
1826 | * then re-duplicate the entry once we drop page lock, | |
1827 | * so we might loop indefinitely; also, that page could | |
1828 | * not be swapped out to other storage meanwhile. So: | |
1829 | * delete from cache even if there's another reference, | |
1830 | * after ensuring that the data has been saved to disk - | |
1831 | * since if the reference remains (rarer), it will be | |
1832 | * read from disk into another page. Splitting into two | |
1833 | * pages would be incorrect if swap supported "shared | |
1834 | * private" pages, but they are handled by tmpfs files. | |
5ad64688 HD |
1835 | * |
1836 | * Given how unuse_vma() targets one particular offset | |
1837 | * in an anon_vma, once the anon_vma has been determined, | |
1838 | * this splitting happens to be just what is needed to | |
1839 | * handle where KSM pages have been swapped out: re-reading | |
1840 | * is unnecessarily slow, but we can fix that later on. | |
1da177e4 | 1841 | */ |
355cfa73 KH |
1842 | if (swap_count(*swap_map) && |
1843 | PageDirty(page) && PageSwapCache(page)) { | |
1da177e4 LT |
1844 | struct writeback_control wbc = { |
1845 | .sync_mode = WB_SYNC_NONE, | |
1846 | }; | |
1847 | ||
1848 | swap_writepage(page, &wbc); | |
1849 | lock_page(page); | |
1850 | wait_on_page_writeback(page); | |
1851 | } | |
68bdc8d6 HD |
1852 | |
1853 | /* | |
1854 | * It is conceivable that a racing task removed this page from | |
1855 | * swap cache just before we acquired the page lock at the top, | |
1856 | * or while we dropped it in unuse_mm(). The page might even | |
1857 | * be back in swap cache on another swap area: that we must not | |
1858 | * delete, since it may not have been written out to swap yet. | |
1859 | */ | |
1860 | if (PageSwapCache(page) && | |
1861 | likely(page_private(page) == entry.val)) | |
2e0e26c7 | 1862 | delete_from_swap_cache(page); |
1da177e4 LT |
1863 | |
1864 | /* | |
1865 | * So we could skip searching mms once swap count went | |
1866 | * to 1, we did not mark any present ptes as dirty: must | |
2706a1b8 | 1867 | * mark page dirty so shrink_page_list will preserve it. |
1da177e4 LT |
1868 | */ |
1869 | SetPageDirty(page); | |
1870 | unlock_page(page); | |
09cbfeaf | 1871 | put_page(page); |
1da177e4 LT |
1872 | |
1873 | /* | |
1874 | * Make sure that we aren't completely killing | |
1875 | * interactive performance. | |
1876 | */ | |
1877 | cond_resched(); | |
38b5faf4 DM |
1878 | if (frontswap && pages_to_unuse > 0) { |
1879 | if (!--pages_to_unuse) | |
1880 | break; | |
1881 | } | |
1da177e4 LT |
1882 | } |
1883 | ||
1884 | mmput(start_mm); | |
1da177e4 LT |
1885 | return retval; |
1886 | } | |
1887 | ||
1888 | /* | |
5d337b91 HD |
1889 | * After a successful try_to_unuse, if no swap is now in use, we know |
1890 | * we can empty the mmlist. swap_lock must be held on entry and exit. | |
1891 | * Note that mmlist_lock nests inside swap_lock, and an mm must be | |
1da177e4 LT |
1892 | * added to the mmlist just after page_duplicate - before would be racy. |
1893 | */ | |
1894 | static void drain_mmlist(void) | |
1895 | { | |
1896 | struct list_head *p, *next; | |
efa90a98 | 1897 | unsigned int type; |
1da177e4 | 1898 | |
efa90a98 HD |
1899 | for (type = 0; type < nr_swapfiles; type++) |
1900 | if (swap_info[type]->inuse_pages) | |
1da177e4 LT |
1901 | return; |
1902 | spin_lock(&mmlist_lock); | |
1903 | list_for_each_safe(p, next, &init_mm.mmlist) | |
1904 | list_del_init(p); | |
1905 | spin_unlock(&mmlist_lock); | |
1906 | } | |
1907 | ||
1908 | /* | |
1909 | * Use this swapdev's extent info to locate the (PAGE_SIZE) block which | |
d4906e1a LS |
1910 | * corresponds to page offset for the specified swap entry. |
1911 | * Note that the type of this function is sector_t, but it returns page offset | |
1912 | * into the bdev, not sector offset. | |
1da177e4 | 1913 | */ |
d4906e1a | 1914 | static sector_t map_swap_entry(swp_entry_t entry, struct block_device **bdev) |
1da177e4 | 1915 | { |
f29ad6a9 HD |
1916 | struct swap_info_struct *sis; |
1917 | struct swap_extent *start_se; | |
1918 | struct swap_extent *se; | |
1919 | pgoff_t offset; | |
1920 | ||
efa90a98 | 1921 | sis = swap_info[swp_type(entry)]; |
f29ad6a9 HD |
1922 | *bdev = sis->bdev; |
1923 | ||
1924 | offset = swp_offset(entry); | |
1925 | start_se = sis->curr_swap_extent; | |
1926 | se = start_se; | |
1da177e4 LT |
1927 | |
1928 | for ( ; ; ) { | |
1da177e4 LT |
1929 | if (se->start_page <= offset && |
1930 | offset < (se->start_page + se->nr_pages)) { | |
1931 | return se->start_block + (offset - se->start_page); | |
1932 | } | |
a8ae4991 | 1933 | se = list_next_entry(se, list); |
1da177e4 LT |
1934 | sis->curr_swap_extent = se; |
1935 | BUG_ON(se == start_se); /* It *must* be present */ | |
1936 | } | |
1937 | } | |
1938 | ||
d4906e1a LS |
1939 | /* |
1940 | * Returns the page offset into bdev for the specified page's swap entry. | |
1941 | */ | |
1942 | sector_t map_swap_page(struct page *page, struct block_device **bdev) | |
1943 | { | |
1944 | swp_entry_t entry; | |
1945 | entry.val = page_private(page); | |
1946 | return map_swap_entry(entry, bdev); | |
1947 | } | |
1948 | ||
1da177e4 LT |
1949 | /* |
1950 | * Free all of a swapdev's extent information | |
1951 | */ | |
1952 | static void destroy_swap_extents(struct swap_info_struct *sis) | |
1953 | { | |
9625a5f2 | 1954 | while (!list_empty(&sis->first_swap_extent.list)) { |
1da177e4 LT |
1955 | struct swap_extent *se; |
1956 | ||
a8ae4991 | 1957 | se = list_first_entry(&sis->first_swap_extent.list, |
1da177e4 LT |
1958 | struct swap_extent, list); |
1959 | list_del(&se->list); | |
1960 | kfree(se); | |
1961 | } | |
62c230bc MG |
1962 | |
1963 | if (sis->flags & SWP_FILE) { | |
1964 | struct file *swap_file = sis->swap_file; | |
1965 | struct address_space *mapping = swap_file->f_mapping; | |
1966 | ||
1967 | sis->flags &= ~SWP_FILE; | |
1968 | mapping->a_ops->swap_deactivate(swap_file); | |
1969 | } | |
1da177e4 LT |
1970 | } |
1971 | ||
1972 | /* | |
1973 | * Add a block range (and the corresponding page range) into this swapdev's | |
11d31886 | 1974 | * extent list. The extent list is kept sorted in page order. |
1da177e4 | 1975 | * |
11d31886 | 1976 | * This function rather assumes that it is called in ascending page order. |
1da177e4 | 1977 | */ |
a509bc1a | 1978 | int |
1da177e4 LT |
1979 | add_swap_extent(struct swap_info_struct *sis, unsigned long start_page, |
1980 | unsigned long nr_pages, sector_t start_block) | |
1981 | { | |
1982 | struct swap_extent *se; | |
1983 | struct swap_extent *new_se; | |
1984 | struct list_head *lh; | |
1985 | ||
9625a5f2 HD |
1986 | if (start_page == 0) { |
1987 | se = &sis->first_swap_extent; | |
1988 | sis->curr_swap_extent = se; | |
1989 | se->start_page = 0; | |
1990 | se->nr_pages = nr_pages; | |
1991 | se->start_block = start_block; | |
1992 | return 1; | |
1993 | } else { | |
1994 | lh = sis->first_swap_extent.list.prev; /* Highest extent */ | |
1da177e4 | 1995 | se = list_entry(lh, struct swap_extent, list); |
11d31886 HD |
1996 | BUG_ON(se->start_page + se->nr_pages != start_page); |
1997 | if (se->start_block + se->nr_pages == start_block) { | |
1da177e4 LT |
1998 | /* Merge it */ |
1999 | se->nr_pages += nr_pages; | |
2000 | return 0; | |
2001 | } | |
1da177e4 LT |
2002 | } |
2003 | ||
2004 | /* | |
2005 | * No merge. Insert a new extent, preserving ordering. | |
2006 | */ | |
2007 | new_se = kmalloc(sizeof(*se), GFP_KERNEL); | |
2008 | if (new_se == NULL) | |
2009 | return -ENOMEM; | |
2010 | new_se->start_page = start_page; | |
2011 | new_se->nr_pages = nr_pages; | |
2012 | new_se->start_block = start_block; | |
2013 | ||
9625a5f2 | 2014 | list_add_tail(&new_se->list, &sis->first_swap_extent.list); |
53092a74 | 2015 | return 1; |
1da177e4 LT |
2016 | } |
2017 | ||
2018 | /* | |
2019 | * A `swap extent' is a simple thing which maps a contiguous range of pages | |
2020 | * onto a contiguous range of disk blocks. An ordered list of swap extents | |
2021 | * is built at swapon time and is then used at swap_writepage/swap_readpage | |
2022 | * time for locating where on disk a page belongs. | |
2023 | * | |
2024 | * If the swapfile is an S_ISBLK block device, a single extent is installed. | |
2025 | * This is done so that the main operating code can treat S_ISBLK and S_ISREG | |
2026 | * swap files identically. | |
2027 | * | |
2028 | * Whether the swapdev is an S_ISREG file or an S_ISBLK blockdev, the swap | |
2029 | * extent list operates in PAGE_SIZE disk blocks. Both S_ISREG and S_ISBLK | |
2030 | * swapfiles are handled *identically* after swapon time. | |
2031 | * | |
2032 | * For S_ISREG swapfiles, setup_swap_extents() will walk all the file's blocks | |
2033 | * and will parse them into an ordered extent list, in PAGE_SIZE chunks. If | |
2034 | * some stray blocks are found which do not fall within the PAGE_SIZE alignment | |
2035 | * requirements, they are simply tossed out - we will never use those blocks | |
2036 | * for swapping. | |
2037 | * | |
b0d9bcd4 | 2038 | * For S_ISREG swapfiles we set S_SWAPFILE across the life of the swapon. This |
1da177e4 LT |
2039 | * prevents root from shooting her foot off by ftruncating an in-use swapfile, |
2040 | * which will scribble on the fs. | |
2041 | * | |
2042 | * The amount of disk space which a single swap extent represents varies. | |
2043 | * Typically it is in the 1-4 megabyte range. So we can have hundreds of | |
2044 | * extents in the list. To avoid much list walking, we cache the previous | |
2045 | * search location in `curr_swap_extent', and start new searches from there. | |
2046 | * This is extremely effective. The average number of iterations in | |
2047 | * map_swap_page() has been measured at about 0.3 per page. - akpm. | |
2048 | */ | |
53092a74 | 2049 | static int setup_swap_extents(struct swap_info_struct *sis, sector_t *span) |
1da177e4 | 2050 | { |
62c230bc MG |
2051 | struct file *swap_file = sis->swap_file; |
2052 | struct address_space *mapping = swap_file->f_mapping; | |
2053 | struct inode *inode = mapping->host; | |
1da177e4 LT |
2054 | int ret; |
2055 | ||
1da177e4 LT |
2056 | if (S_ISBLK(inode->i_mode)) { |
2057 | ret = add_swap_extent(sis, 0, sis->max, 0); | |
53092a74 | 2058 | *span = sis->pages; |
a509bc1a | 2059 | return ret; |
1da177e4 LT |
2060 | } |
2061 | ||
62c230bc | 2062 | if (mapping->a_ops->swap_activate) { |
a509bc1a | 2063 | ret = mapping->a_ops->swap_activate(sis, swap_file, span); |
62c230bc MG |
2064 | if (!ret) { |
2065 | sis->flags |= SWP_FILE; | |
2066 | ret = add_swap_extent(sis, 0, sis->max, 0); | |
2067 | *span = sis->pages; | |
2068 | } | |
a509bc1a | 2069 | return ret; |
62c230bc MG |
2070 | } |
2071 | ||
a509bc1a | 2072 | return generic_swapfile_activate(sis, swap_file, span); |
1da177e4 LT |
2073 | } |
2074 | ||
cf0cac0a | 2075 | static void _enable_swap_info(struct swap_info_struct *p, int prio, |
2a8f9449 SL |
2076 | unsigned char *swap_map, |
2077 | struct swap_cluster_info *cluster_info) | |
40531542 | 2078 | { |
40531542 CEB |
2079 | if (prio >= 0) |
2080 | p->prio = prio; | |
2081 | else | |
2082 | p->prio = --least_priority; | |
18ab4d4c DS |
2083 | /* |
2084 | * the plist prio is negated because plist ordering is | |
2085 | * low-to-high, while swap ordering is high-to-low | |
2086 | */ | |
2087 | p->list.prio = -p->prio; | |
2088 | p->avail_list.prio = -p->prio; | |
40531542 | 2089 | p->swap_map = swap_map; |
2a8f9449 | 2090 | p->cluster_info = cluster_info; |
40531542 | 2091 | p->flags |= SWP_WRITEOK; |
ec8acf20 | 2092 | atomic_long_add(p->pages, &nr_swap_pages); |
40531542 CEB |
2093 | total_swap_pages += p->pages; |
2094 | ||
adfab836 | 2095 | assert_spin_locked(&swap_lock); |
adfab836 | 2096 | /* |
18ab4d4c DS |
2097 | * both lists are plists, and thus priority ordered. |
2098 | * swap_active_head needs to be priority ordered for swapoff(), | |
2099 | * which on removal of any swap_info_struct with an auto-assigned | |
2100 | * (i.e. negative) priority increments the auto-assigned priority | |
2101 | * of any lower-priority swap_info_structs. | |
2102 | * swap_avail_head needs to be priority ordered for get_swap_page(), | |
2103 | * which allocates swap pages from the highest available priority | |
2104 | * swap_info_struct. | |
adfab836 | 2105 | */ |
18ab4d4c DS |
2106 | plist_add(&p->list, &swap_active_head); |
2107 | spin_lock(&swap_avail_lock); | |
2108 | plist_add(&p->avail_list, &swap_avail_head); | |
2109 | spin_unlock(&swap_avail_lock); | |
cf0cac0a CEB |
2110 | } |
2111 | ||
2112 | static void enable_swap_info(struct swap_info_struct *p, int prio, | |
2113 | unsigned char *swap_map, | |
2a8f9449 | 2114 | struct swap_cluster_info *cluster_info, |
cf0cac0a CEB |
2115 | unsigned long *frontswap_map) |
2116 | { | |
4f89849d | 2117 | frontswap_init(p->type, frontswap_map); |
cf0cac0a | 2118 | spin_lock(&swap_lock); |
ec8acf20 | 2119 | spin_lock(&p->lock); |
2a8f9449 | 2120 | _enable_swap_info(p, prio, swap_map, cluster_info); |
ec8acf20 | 2121 | spin_unlock(&p->lock); |
cf0cac0a CEB |
2122 | spin_unlock(&swap_lock); |
2123 | } | |
2124 | ||
2125 | static void reinsert_swap_info(struct swap_info_struct *p) | |
2126 | { | |
2127 | spin_lock(&swap_lock); | |
ec8acf20 | 2128 | spin_lock(&p->lock); |
2a8f9449 | 2129 | _enable_swap_info(p, p->prio, p->swap_map, p->cluster_info); |
ec8acf20 | 2130 | spin_unlock(&p->lock); |
40531542 CEB |
2131 | spin_unlock(&swap_lock); |
2132 | } | |
2133 | ||
67afa38e TC |
2134 | bool has_usable_swap(void) |
2135 | { | |
2136 | bool ret = true; | |
2137 | ||
2138 | spin_lock(&swap_lock); | |
2139 | if (plist_head_empty(&swap_active_head)) | |
2140 | ret = false; | |
2141 | spin_unlock(&swap_lock); | |
2142 | return ret; | |
2143 | } | |
2144 | ||
c4ea37c2 | 2145 | SYSCALL_DEFINE1(swapoff, const char __user *, specialfile) |
1da177e4 | 2146 | { |
73c34b6a | 2147 | struct swap_info_struct *p = NULL; |
8d69aaee | 2148 | unsigned char *swap_map; |
2a8f9449 | 2149 | struct swap_cluster_info *cluster_info; |
4f89849d | 2150 | unsigned long *frontswap_map; |
1da177e4 LT |
2151 | struct file *swap_file, *victim; |
2152 | struct address_space *mapping; | |
2153 | struct inode *inode; | |
91a27b2a | 2154 | struct filename *pathname; |
adfab836 | 2155 | int err, found = 0; |
5b808a23 | 2156 | unsigned int old_block_size; |
886bb7e9 | 2157 | |
1da177e4 LT |
2158 | if (!capable(CAP_SYS_ADMIN)) |
2159 | return -EPERM; | |
2160 | ||
191c5424 AV |
2161 | BUG_ON(!current->mm); |
2162 | ||
1da177e4 | 2163 | pathname = getname(specialfile); |
1da177e4 | 2164 | if (IS_ERR(pathname)) |
f58b59c1 | 2165 | return PTR_ERR(pathname); |
1da177e4 | 2166 | |
669abf4e | 2167 | victim = file_open_name(pathname, O_RDWR|O_LARGEFILE, 0); |
1da177e4 LT |
2168 | err = PTR_ERR(victim); |
2169 | if (IS_ERR(victim)) | |
2170 | goto out; | |
2171 | ||
2172 | mapping = victim->f_mapping; | |
5d337b91 | 2173 | spin_lock(&swap_lock); |
18ab4d4c | 2174 | plist_for_each_entry(p, &swap_active_head, list) { |
22c6f8fd | 2175 | if (p->flags & SWP_WRITEOK) { |
adfab836 DS |
2176 | if (p->swap_file->f_mapping == mapping) { |
2177 | found = 1; | |
1da177e4 | 2178 | break; |
adfab836 | 2179 | } |
1da177e4 | 2180 | } |
1da177e4 | 2181 | } |
adfab836 | 2182 | if (!found) { |
1da177e4 | 2183 | err = -EINVAL; |
5d337b91 | 2184 | spin_unlock(&swap_lock); |
1da177e4 LT |
2185 | goto out_dput; |
2186 | } | |
191c5424 | 2187 | if (!security_vm_enough_memory_mm(current->mm, p->pages)) |
1da177e4 LT |
2188 | vm_unacct_memory(p->pages); |
2189 | else { | |
2190 | err = -ENOMEM; | |
5d337b91 | 2191 | spin_unlock(&swap_lock); |
1da177e4 LT |
2192 | goto out_dput; |
2193 | } | |
18ab4d4c DS |
2194 | spin_lock(&swap_avail_lock); |
2195 | plist_del(&p->avail_list, &swap_avail_head); | |
2196 | spin_unlock(&swap_avail_lock); | |
ec8acf20 | 2197 | spin_lock(&p->lock); |
78ecba08 | 2198 | if (p->prio < 0) { |
adfab836 DS |
2199 | struct swap_info_struct *si = p; |
2200 | ||
18ab4d4c | 2201 | plist_for_each_entry_continue(si, &swap_active_head, list) { |
adfab836 | 2202 | si->prio++; |
18ab4d4c DS |
2203 | si->list.prio--; |
2204 | si->avail_list.prio--; | |
adfab836 | 2205 | } |
78ecba08 HD |
2206 | least_priority++; |
2207 | } | |
18ab4d4c | 2208 | plist_del(&p->list, &swap_active_head); |
ec8acf20 | 2209 | atomic_long_sub(p->pages, &nr_swap_pages); |
1da177e4 LT |
2210 | total_swap_pages -= p->pages; |
2211 | p->flags &= ~SWP_WRITEOK; | |
ec8acf20 | 2212 | spin_unlock(&p->lock); |
5d337b91 | 2213 | spin_unlock(&swap_lock); |
fb4f88dc | 2214 | |
039939a6 TC |
2215 | disable_swap_slots_cache_lock(); |
2216 | ||
e1e12d2f | 2217 | set_current_oom_origin(); |
adfab836 | 2218 | err = try_to_unuse(p->type, false, 0); /* force unuse all pages */ |
e1e12d2f | 2219 | clear_current_oom_origin(); |
1da177e4 | 2220 | |
1da177e4 LT |
2221 | if (err) { |
2222 | /* re-insert swap space back into swap_list */ | |
cf0cac0a | 2223 | reinsert_swap_info(p); |
039939a6 | 2224 | reenable_swap_slots_cache_unlock(); |
1da177e4 LT |
2225 | goto out_dput; |
2226 | } | |
52b7efdb | 2227 | |
039939a6 TC |
2228 | reenable_swap_slots_cache_unlock(); |
2229 | ||
815c2c54 SL |
2230 | flush_work(&p->discard_work); |
2231 | ||
5d337b91 | 2232 | destroy_swap_extents(p); |
570a335b HD |
2233 | if (p->flags & SWP_CONTINUED) |
2234 | free_swap_count_continuations(p); | |
2235 | ||
fc0abb14 | 2236 | mutex_lock(&swapon_mutex); |
5d337b91 | 2237 | spin_lock(&swap_lock); |
ec8acf20 | 2238 | spin_lock(&p->lock); |
5d337b91 HD |
2239 | drain_mmlist(); |
2240 | ||
52b7efdb | 2241 | /* wait for anyone still in scan_swap_map */ |
52b7efdb HD |
2242 | p->highest_bit = 0; /* cuts scans short */ |
2243 | while (p->flags >= SWP_SCANNING) { | |
ec8acf20 | 2244 | spin_unlock(&p->lock); |
5d337b91 | 2245 | spin_unlock(&swap_lock); |
13e4b57f | 2246 | schedule_timeout_uninterruptible(1); |
5d337b91 | 2247 | spin_lock(&swap_lock); |
ec8acf20 | 2248 | spin_lock(&p->lock); |
52b7efdb | 2249 | } |
52b7efdb | 2250 | |
1da177e4 | 2251 | swap_file = p->swap_file; |
5b808a23 | 2252 | old_block_size = p->old_block_size; |
1da177e4 LT |
2253 | p->swap_file = NULL; |
2254 | p->max = 0; | |
2255 | swap_map = p->swap_map; | |
2256 | p->swap_map = NULL; | |
2a8f9449 SL |
2257 | cluster_info = p->cluster_info; |
2258 | p->cluster_info = NULL; | |
4f89849d | 2259 | frontswap_map = frontswap_map_get(p); |
ec8acf20 | 2260 | spin_unlock(&p->lock); |
5d337b91 | 2261 | spin_unlock(&swap_lock); |
adfab836 | 2262 | frontswap_invalidate_area(p->type); |
58e97ba6 | 2263 | frontswap_map_set(p, NULL); |
fc0abb14 | 2264 | mutex_unlock(&swapon_mutex); |
ebc2a1a6 SL |
2265 | free_percpu(p->percpu_cluster); |
2266 | p->percpu_cluster = NULL; | |
1da177e4 | 2267 | vfree(swap_map); |
2a8f9449 | 2268 | vfree(cluster_info); |
4f89849d | 2269 | vfree(frontswap_map); |
2de1a7e4 | 2270 | /* Destroy swap account information */ |
adfab836 | 2271 | swap_cgroup_swapoff(p->type); |
4b3ef9da | 2272 | exit_swap_address_space(p->type); |
27a7faa0 | 2273 | |
1da177e4 LT |
2274 | inode = mapping->host; |
2275 | if (S_ISBLK(inode->i_mode)) { | |
2276 | struct block_device *bdev = I_BDEV(inode); | |
5b808a23 | 2277 | set_blocksize(bdev, old_block_size); |
e525fd89 | 2278 | blkdev_put(bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL); |
1da177e4 | 2279 | } else { |
5955102c | 2280 | inode_lock(inode); |
1da177e4 | 2281 | inode->i_flags &= ~S_SWAPFILE; |
5955102c | 2282 | inode_unlock(inode); |
1da177e4 LT |
2283 | } |
2284 | filp_close(swap_file, NULL); | |
f893ab41 WY |
2285 | |
2286 | /* | |
2287 | * Clear the SWP_USED flag after all resources are freed so that swapon | |
2288 | * can reuse this swap_info in alloc_swap_info() safely. It is ok to | |
2289 | * not hold p->lock after we cleared its SWP_WRITEOK. | |
2290 | */ | |
2291 | spin_lock(&swap_lock); | |
2292 | p->flags = 0; | |
2293 | spin_unlock(&swap_lock); | |
2294 | ||
1da177e4 | 2295 | err = 0; |
66d7dd51 KS |
2296 | atomic_inc(&proc_poll_event); |
2297 | wake_up_interruptible(&proc_poll_wait); | |
1da177e4 LT |
2298 | |
2299 | out_dput: | |
2300 | filp_close(victim, NULL); | |
2301 | out: | |
f58b59c1 | 2302 | putname(pathname); |
1da177e4 LT |
2303 | return err; |
2304 | } | |
2305 | ||
2306 | #ifdef CONFIG_PROC_FS | |
66d7dd51 KS |
2307 | static unsigned swaps_poll(struct file *file, poll_table *wait) |
2308 | { | |
f1514638 | 2309 | struct seq_file *seq = file->private_data; |
66d7dd51 KS |
2310 | |
2311 | poll_wait(file, &proc_poll_wait, wait); | |
2312 | ||
f1514638 KS |
2313 | if (seq->poll_event != atomic_read(&proc_poll_event)) { |
2314 | seq->poll_event = atomic_read(&proc_poll_event); | |
66d7dd51 KS |
2315 | return POLLIN | POLLRDNORM | POLLERR | POLLPRI; |
2316 | } | |
2317 | ||
2318 | return POLLIN | POLLRDNORM; | |
2319 | } | |
2320 | ||
1da177e4 LT |
2321 | /* iterator */ |
2322 | static void *swap_start(struct seq_file *swap, loff_t *pos) | |
2323 | { | |
efa90a98 HD |
2324 | struct swap_info_struct *si; |
2325 | int type; | |
1da177e4 LT |
2326 | loff_t l = *pos; |
2327 | ||
fc0abb14 | 2328 | mutex_lock(&swapon_mutex); |
1da177e4 | 2329 | |
881e4aab SS |
2330 | if (!l) |
2331 | return SEQ_START_TOKEN; | |
2332 | ||
efa90a98 HD |
2333 | for (type = 0; type < nr_swapfiles; type++) { |
2334 | smp_rmb(); /* read nr_swapfiles before swap_info[type] */ | |
2335 | si = swap_info[type]; | |
2336 | if (!(si->flags & SWP_USED) || !si->swap_map) | |
1da177e4 | 2337 | continue; |
881e4aab | 2338 | if (!--l) |
efa90a98 | 2339 | return si; |
1da177e4 LT |
2340 | } |
2341 | ||
2342 | return NULL; | |
2343 | } | |
2344 | ||
2345 | static void *swap_next(struct seq_file *swap, void *v, loff_t *pos) | |
2346 | { | |
efa90a98 HD |
2347 | struct swap_info_struct *si = v; |
2348 | int type; | |
1da177e4 | 2349 | |
881e4aab | 2350 | if (v == SEQ_START_TOKEN) |
efa90a98 HD |
2351 | type = 0; |
2352 | else | |
2353 | type = si->type + 1; | |
881e4aab | 2354 | |
efa90a98 HD |
2355 | for (; type < nr_swapfiles; type++) { |
2356 | smp_rmb(); /* read nr_swapfiles before swap_info[type] */ | |
2357 | si = swap_info[type]; | |
2358 | if (!(si->flags & SWP_USED) || !si->swap_map) | |
1da177e4 LT |
2359 | continue; |
2360 | ++*pos; | |
efa90a98 | 2361 | return si; |
1da177e4 LT |
2362 | } |
2363 | ||
2364 | return NULL; | |
2365 | } | |
2366 | ||
2367 | static void swap_stop(struct seq_file *swap, void *v) | |
2368 | { | |
fc0abb14 | 2369 | mutex_unlock(&swapon_mutex); |
1da177e4 LT |
2370 | } |
2371 | ||
2372 | static int swap_show(struct seq_file *swap, void *v) | |
2373 | { | |
efa90a98 | 2374 | struct swap_info_struct *si = v; |
1da177e4 LT |
2375 | struct file *file; |
2376 | int len; | |
2377 | ||
efa90a98 | 2378 | if (si == SEQ_START_TOKEN) { |
881e4aab SS |
2379 | seq_puts(swap,"Filename\t\t\t\tType\t\tSize\tUsed\tPriority\n"); |
2380 | return 0; | |
2381 | } | |
1da177e4 | 2382 | |
efa90a98 | 2383 | file = si->swap_file; |
2726d566 | 2384 | len = seq_file_path(swap, file, " \t\n\\"); |
6eb396dc | 2385 | seq_printf(swap, "%*s%s\t%u\t%u\t%d\n", |
886bb7e9 | 2386 | len < 40 ? 40 - len : 1, " ", |
496ad9aa | 2387 | S_ISBLK(file_inode(file)->i_mode) ? |
1da177e4 | 2388 | "partition" : "file\t", |
efa90a98 HD |
2389 | si->pages << (PAGE_SHIFT - 10), |
2390 | si->inuse_pages << (PAGE_SHIFT - 10), | |
2391 | si->prio); | |
1da177e4 LT |
2392 | return 0; |
2393 | } | |
2394 | ||
15ad7cdc | 2395 | static const struct seq_operations swaps_op = { |
1da177e4 LT |
2396 | .start = swap_start, |
2397 | .next = swap_next, | |
2398 | .stop = swap_stop, | |
2399 | .show = swap_show | |
2400 | }; | |
2401 | ||
2402 | static int swaps_open(struct inode *inode, struct file *file) | |
2403 | { | |
f1514638 | 2404 | struct seq_file *seq; |
66d7dd51 KS |
2405 | int ret; |
2406 | ||
66d7dd51 | 2407 | ret = seq_open(file, &swaps_op); |
f1514638 | 2408 | if (ret) |
66d7dd51 | 2409 | return ret; |
66d7dd51 | 2410 | |
f1514638 KS |
2411 | seq = file->private_data; |
2412 | seq->poll_event = atomic_read(&proc_poll_event); | |
2413 | return 0; | |
1da177e4 LT |
2414 | } |
2415 | ||
15ad7cdc | 2416 | static const struct file_operations proc_swaps_operations = { |
1da177e4 LT |
2417 | .open = swaps_open, |
2418 | .read = seq_read, | |
2419 | .llseek = seq_lseek, | |
2420 | .release = seq_release, | |
66d7dd51 | 2421 | .poll = swaps_poll, |
1da177e4 LT |
2422 | }; |
2423 | ||
2424 | static int __init procswaps_init(void) | |
2425 | { | |
3d71f86f | 2426 | proc_create("swaps", 0, NULL, &proc_swaps_operations); |
1da177e4 LT |
2427 | return 0; |
2428 | } | |
2429 | __initcall(procswaps_init); | |
2430 | #endif /* CONFIG_PROC_FS */ | |
2431 | ||
1796316a JB |
2432 | #ifdef MAX_SWAPFILES_CHECK |
2433 | static int __init max_swapfiles_check(void) | |
2434 | { | |
2435 | MAX_SWAPFILES_CHECK(); | |
2436 | return 0; | |
2437 | } | |
2438 | late_initcall(max_swapfiles_check); | |
2439 | #endif | |
2440 | ||
53cbb243 | 2441 | static struct swap_info_struct *alloc_swap_info(void) |
1da177e4 | 2442 | { |
73c34b6a | 2443 | struct swap_info_struct *p; |
1da177e4 | 2444 | unsigned int type; |
efa90a98 HD |
2445 | |
2446 | p = kzalloc(sizeof(*p), GFP_KERNEL); | |
2447 | if (!p) | |
53cbb243 | 2448 | return ERR_PTR(-ENOMEM); |
efa90a98 | 2449 | |
5d337b91 | 2450 | spin_lock(&swap_lock); |
efa90a98 HD |
2451 | for (type = 0; type < nr_swapfiles; type++) { |
2452 | if (!(swap_info[type]->flags & SWP_USED)) | |
1da177e4 | 2453 | break; |
efa90a98 | 2454 | } |
0697212a | 2455 | if (type >= MAX_SWAPFILES) { |
5d337b91 | 2456 | spin_unlock(&swap_lock); |
efa90a98 | 2457 | kfree(p); |
730c0581 | 2458 | return ERR_PTR(-EPERM); |
1da177e4 | 2459 | } |
efa90a98 HD |
2460 | if (type >= nr_swapfiles) { |
2461 | p->type = type; | |
2462 | swap_info[type] = p; | |
2463 | /* | |
2464 | * Write swap_info[type] before nr_swapfiles, in case a | |
2465 | * racing procfs swap_start() or swap_next() is reading them. | |
2466 | * (We never shrink nr_swapfiles, we never free this entry.) | |
2467 | */ | |
2468 | smp_wmb(); | |
2469 | nr_swapfiles++; | |
2470 | } else { | |
2471 | kfree(p); | |
2472 | p = swap_info[type]; | |
2473 | /* | |
2474 | * Do not memset this entry: a racing procfs swap_next() | |
2475 | * would be relying on p->type to remain valid. | |
2476 | */ | |
2477 | } | |
9625a5f2 | 2478 | INIT_LIST_HEAD(&p->first_swap_extent.list); |
18ab4d4c DS |
2479 | plist_node_init(&p->list, 0); |
2480 | plist_node_init(&p->avail_list, 0); | |
1da177e4 | 2481 | p->flags = SWP_USED; |
5d337b91 | 2482 | spin_unlock(&swap_lock); |
ec8acf20 | 2483 | spin_lock_init(&p->lock); |
efa90a98 | 2484 | |
53cbb243 | 2485 | return p; |
53cbb243 CEB |
2486 | } |
2487 | ||
4d0e1e10 CEB |
2488 | static int claim_swapfile(struct swap_info_struct *p, struct inode *inode) |
2489 | { | |
2490 | int error; | |
2491 | ||
2492 | if (S_ISBLK(inode->i_mode)) { | |
2493 | p->bdev = bdgrab(I_BDEV(inode)); | |
2494 | error = blkdev_get(p->bdev, | |
6f179af8 | 2495 | FMODE_READ | FMODE_WRITE | FMODE_EXCL, p); |
4d0e1e10 CEB |
2496 | if (error < 0) { |
2497 | p->bdev = NULL; | |
6f179af8 | 2498 | return error; |
4d0e1e10 CEB |
2499 | } |
2500 | p->old_block_size = block_size(p->bdev); | |
2501 | error = set_blocksize(p->bdev, PAGE_SIZE); | |
2502 | if (error < 0) | |
87ade72a | 2503 | return error; |
4d0e1e10 CEB |
2504 | p->flags |= SWP_BLKDEV; |
2505 | } else if (S_ISREG(inode->i_mode)) { | |
2506 | p->bdev = inode->i_sb->s_bdev; | |
5955102c | 2507 | inode_lock(inode); |
87ade72a CEB |
2508 | if (IS_SWAPFILE(inode)) |
2509 | return -EBUSY; | |
2510 | } else | |
2511 | return -EINVAL; | |
4d0e1e10 CEB |
2512 | |
2513 | return 0; | |
4d0e1e10 CEB |
2514 | } |
2515 | ||
ca8bd38b CEB |
2516 | static unsigned long read_swap_header(struct swap_info_struct *p, |
2517 | union swap_header *swap_header, | |
2518 | struct inode *inode) | |
2519 | { | |
2520 | int i; | |
2521 | unsigned long maxpages; | |
2522 | unsigned long swapfilepages; | |
d6bbbd29 | 2523 | unsigned long last_page; |
ca8bd38b CEB |
2524 | |
2525 | if (memcmp("SWAPSPACE2", swap_header->magic.magic, 10)) { | |
465c47fd | 2526 | pr_err("Unable to find swap-space signature\n"); |
38719025 | 2527 | return 0; |
ca8bd38b CEB |
2528 | } |
2529 | ||
2530 | /* swap partition endianess hack... */ | |
2531 | if (swab32(swap_header->info.version) == 1) { | |
2532 | swab32s(&swap_header->info.version); | |
2533 | swab32s(&swap_header->info.last_page); | |
2534 | swab32s(&swap_header->info.nr_badpages); | |
dd111be6 JH |
2535 | if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES) |
2536 | return 0; | |
ca8bd38b CEB |
2537 | for (i = 0; i < swap_header->info.nr_badpages; i++) |
2538 | swab32s(&swap_header->info.badpages[i]); | |
2539 | } | |
2540 | /* Check the swap header's sub-version */ | |
2541 | if (swap_header->info.version != 1) { | |
465c47fd AM |
2542 | pr_warn("Unable to handle swap header version %d\n", |
2543 | swap_header->info.version); | |
38719025 | 2544 | return 0; |
ca8bd38b CEB |
2545 | } |
2546 | ||
2547 | p->lowest_bit = 1; | |
2548 | p->cluster_next = 1; | |
2549 | p->cluster_nr = 0; | |
2550 | ||
2551 | /* | |
2552 | * Find out how many pages are allowed for a single swap | |
9b15b817 | 2553 | * device. There are two limiting factors: 1) the number |
a2c16d6c HD |
2554 | * of bits for the swap offset in the swp_entry_t type, and |
2555 | * 2) the number of bits in the swap pte as defined by the | |
9b15b817 | 2556 | * different architectures. In order to find the |
a2c16d6c | 2557 | * largest possible bit mask, a swap entry with swap type 0 |
ca8bd38b | 2558 | * and swap offset ~0UL is created, encoded to a swap pte, |
a2c16d6c | 2559 | * decoded to a swp_entry_t again, and finally the swap |
ca8bd38b CEB |
2560 | * offset is extracted. This will mask all the bits from |
2561 | * the initial ~0UL mask that can't be encoded in either | |
2562 | * the swp_entry_t or the architecture definition of a | |
9b15b817 | 2563 | * swap pte. |
ca8bd38b CEB |
2564 | */ |
2565 | maxpages = swp_offset(pte_to_swp_entry( | |
9b15b817 | 2566 | swp_entry_to_pte(swp_entry(0, ~0UL)))) + 1; |
d6bbbd29 RJ |
2567 | last_page = swap_header->info.last_page; |
2568 | if (last_page > maxpages) { | |
465c47fd | 2569 | pr_warn("Truncating oversized swap area, only using %luk out of %luk\n", |
d6bbbd29 RJ |
2570 | maxpages << (PAGE_SHIFT - 10), |
2571 | last_page << (PAGE_SHIFT - 10)); | |
2572 | } | |
2573 | if (maxpages > last_page) { | |
2574 | maxpages = last_page + 1; | |
ca8bd38b CEB |
2575 | /* p->max is an unsigned int: don't overflow it */ |
2576 | if ((unsigned int)maxpages == 0) | |
2577 | maxpages = UINT_MAX; | |
2578 | } | |
2579 | p->highest_bit = maxpages - 1; | |
2580 | ||
2581 | if (!maxpages) | |
38719025 | 2582 | return 0; |
ca8bd38b CEB |
2583 | swapfilepages = i_size_read(inode) >> PAGE_SHIFT; |
2584 | if (swapfilepages && maxpages > swapfilepages) { | |
465c47fd | 2585 | pr_warn("Swap area shorter than signature indicates\n"); |
38719025 | 2586 | return 0; |
ca8bd38b CEB |
2587 | } |
2588 | if (swap_header->info.nr_badpages && S_ISREG(inode->i_mode)) | |
38719025 | 2589 | return 0; |
ca8bd38b | 2590 | if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES) |
38719025 | 2591 | return 0; |
ca8bd38b CEB |
2592 | |
2593 | return maxpages; | |
ca8bd38b CEB |
2594 | } |
2595 | ||
4b3ef9da | 2596 | #define SWAP_CLUSTER_INFO_COLS \ |
235b6217 | 2597 | DIV_ROUND_UP(L1_CACHE_BYTES, sizeof(struct swap_cluster_info)) |
4b3ef9da HY |
2598 | #define SWAP_CLUSTER_SPACE_COLS \ |
2599 | DIV_ROUND_UP(SWAP_ADDRESS_SPACE_PAGES, SWAPFILE_CLUSTER) | |
2600 | #define SWAP_CLUSTER_COLS \ | |
2601 | max_t(unsigned int, SWAP_CLUSTER_INFO_COLS, SWAP_CLUSTER_SPACE_COLS) | |
235b6217 | 2602 | |
915d4d7b CEB |
2603 | static int setup_swap_map_and_extents(struct swap_info_struct *p, |
2604 | union swap_header *swap_header, | |
2605 | unsigned char *swap_map, | |
2a8f9449 | 2606 | struct swap_cluster_info *cluster_info, |
915d4d7b CEB |
2607 | unsigned long maxpages, |
2608 | sector_t *span) | |
2609 | { | |
235b6217 | 2610 | unsigned int j, k; |
915d4d7b CEB |
2611 | unsigned int nr_good_pages; |
2612 | int nr_extents; | |
2a8f9449 | 2613 | unsigned long nr_clusters = DIV_ROUND_UP(maxpages, SWAPFILE_CLUSTER); |
235b6217 HY |
2614 | unsigned long col = p->cluster_next / SWAPFILE_CLUSTER % SWAP_CLUSTER_COLS; |
2615 | unsigned long i, idx; | |
915d4d7b CEB |
2616 | |
2617 | nr_good_pages = maxpages - 1; /* omit header page */ | |
2618 | ||
6b534915 HY |
2619 | cluster_list_init(&p->free_clusters); |
2620 | cluster_list_init(&p->discard_clusters); | |
2a8f9449 | 2621 | |
915d4d7b CEB |
2622 | for (i = 0; i < swap_header->info.nr_badpages; i++) { |
2623 | unsigned int page_nr = swap_header->info.badpages[i]; | |
bdb8e3f6 CEB |
2624 | if (page_nr == 0 || page_nr > swap_header->info.last_page) |
2625 | return -EINVAL; | |
915d4d7b CEB |
2626 | if (page_nr < maxpages) { |
2627 | swap_map[page_nr] = SWAP_MAP_BAD; | |
2628 | nr_good_pages--; | |
2a8f9449 SL |
2629 | /* |
2630 | * Haven't marked the cluster free yet, no list | |
2631 | * operation involved | |
2632 | */ | |
2633 | inc_cluster_info_page(p, cluster_info, page_nr); | |
915d4d7b CEB |
2634 | } |
2635 | } | |
2636 | ||
2a8f9449 SL |
2637 | /* Haven't marked the cluster free yet, no list operation involved */ |
2638 | for (i = maxpages; i < round_up(maxpages, SWAPFILE_CLUSTER); i++) | |
2639 | inc_cluster_info_page(p, cluster_info, i); | |
2640 | ||
915d4d7b CEB |
2641 | if (nr_good_pages) { |
2642 | swap_map[0] = SWAP_MAP_BAD; | |
2a8f9449 SL |
2643 | /* |
2644 | * Not mark the cluster free yet, no list | |
2645 | * operation involved | |
2646 | */ | |
2647 | inc_cluster_info_page(p, cluster_info, 0); | |
915d4d7b CEB |
2648 | p->max = maxpages; |
2649 | p->pages = nr_good_pages; | |
2650 | nr_extents = setup_swap_extents(p, span); | |
bdb8e3f6 CEB |
2651 | if (nr_extents < 0) |
2652 | return nr_extents; | |
915d4d7b CEB |
2653 | nr_good_pages = p->pages; |
2654 | } | |
2655 | if (!nr_good_pages) { | |
465c47fd | 2656 | pr_warn("Empty swap-file\n"); |
bdb8e3f6 | 2657 | return -EINVAL; |
915d4d7b CEB |
2658 | } |
2659 | ||
2a8f9449 SL |
2660 | if (!cluster_info) |
2661 | return nr_extents; | |
2662 | ||
235b6217 | 2663 | |
4b3ef9da HY |
2664 | /* |
2665 | * Reduce false cache line sharing between cluster_info and | |
2666 | * sharing same address space. | |
2667 | */ | |
235b6217 HY |
2668 | for (k = 0; k < SWAP_CLUSTER_COLS; k++) { |
2669 | j = (k + col) % SWAP_CLUSTER_COLS; | |
2670 | for (i = 0; i < DIV_ROUND_UP(nr_clusters, SWAP_CLUSTER_COLS); i++) { | |
2671 | idx = i * SWAP_CLUSTER_COLS + j; | |
2672 | if (idx >= nr_clusters) | |
2673 | continue; | |
2674 | if (cluster_count(&cluster_info[idx])) | |
2675 | continue; | |
2a8f9449 | 2676 | cluster_set_flag(&cluster_info[idx], CLUSTER_FLAG_FREE); |
6b534915 HY |
2677 | cluster_list_add_tail(&p->free_clusters, cluster_info, |
2678 | idx); | |
2a8f9449 | 2679 | } |
2a8f9449 | 2680 | } |
915d4d7b | 2681 | return nr_extents; |
915d4d7b CEB |
2682 | } |
2683 | ||
dcf6b7dd RA |
2684 | /* |
2685 | * Helper to sys_swapon determining if a given swap | |
2686 | * backing device queue supports DISCARD operations. | |
2687 | */ | |
2688 | static bool swap_discardable(struct swap_info_struct *si) | |
2689 | { | |
2690 | struct request_queue *q = bdev_get_queue(si->bdev); | |
2691 | ||
2692 | if (!q || !blk_queue_discard(q)) | |
2693 | return false; | |
2694 | ||
2695 | return true; | |
2696 | } | |
2697 | ||
53cbb243 CEB |
2698 | SYSCALL_DEFINE2(swapon, const char __user *, specialfile, int, swap_flags) |
2699 | { | |
2700 | struct swap_info_struct *p; | |
91a27b2a | 2701 | struct filename *name; |
53cbb243 CEB |
2702 | struct file *swap_file = NULL; |
2703 | struct address_space *mapping; | |
40531542 | 2704 | int prio; |
53cbb243 CEB |
2705 | int error; |
2706 | union swap_header *swap_header; | |
915d4d7b | 2707 | int nr_extents; |
53cbb243 CEB |
2708 | sector_t span; |
2709 | unsigned long maxpages; | |
53cbb243 | 2710 | unsigned char *swap_map = NULL; |
2a8f9449 | 2711 | struct swap_cluster_info *cluster_info = NULL; |
38b5faf4 | 2712 | unsigned long *frontswap_map = NULL; |
53cbb243 CEB |
2713 | struct page *page = NULL; |
2714 | struct inode *inode = NULL; | |
53cbb243 | 2715 | |
d15cab97 HD |
2716 | if (swap_flags & ~SWAP_FLAGS_VALID) |
2717 | return -EINVAL; | |
2718 | ||
53cbb243 CEB |
2719 | if (!capable(CAP_SYS_ADMIN)) |
2720 | return -EPERM; | |
2721 | ||
2722 | p = alloc_swap_info(); | |
2542e513 CEB |
2723 | if (IS_ERR(p)) |
2724 | return PTR_ERR(p); | |
53cbb243 | 2725 | |
815c2c54 SL |
2726 | INIT_WORK(&p->discard_work, swap_discard_work); |
2727 | ||
1da177e4 | 2728 | name = getname(specialfile); |
1da177e4 | 2729 | if (IS_ERR(name)) { |
7de7fb6b | 2730 | error = PTR_ERR(name); |
1da177e4 | 2731 | name = NULL; |
bd69010b | 2732 | goto bad_swap; |
1da177e4 | 2733 | } |
669abf4e | 2734 | swap_file = file_open_name(name, O_RDWR|O_LARGEFILE, 0); |
1da177e4 | 2735 | if (IS_ERR(swap_file)) { |
7de7fb6b | 2736 | error = PTR_ERR(swap_file); |
1da177e4 | 2737 | swap_file = NULL; |
bd69010b | 2738 | goto bad_swap; |
1da177e4 LT |
2739 | } |
2740 | ||
2741 | p->swap_file = swap_file; | |
2742 | mapping = swap_file->f_mapping; | |
2130781e | 2743 | inode = mapping->host; |
6f179af8 | 2744 | |
5955102c | 2745 | /* If S_ISREG(inode->i_mode) will do inode_lock(inode); */ |
4d0e1e10 CEB |
2746 | error = claim_swapfile(p, inode); |
2747 | if (unlikely(error)) | |
1da177e4 | 2748 | goto bad_swap; |
1da177e4 | 2749 | |
1da177e4 LT |
2750 | /* |
2751 | * Read the swap header. | |
2752 | */ | |
2753 | if (!mapping->a_ops->readpage) { | |
2754 | error = -EINVAL; | |
2755 | goto bad_swap; | |
2756 | } | |
090d2b18 | 2757 | page = read_mapping_page(mapping, 0, swap_file); |
1da177e4 LT |
2758 | if (IS_ERR(page)) { |
2759 | error = PTR_ERR(page); | |
2760 | goto bad_swap; | |
2761 | } | |
81e33971 | 2762 | swap_header = kmap(page); |
1da177e4 | 2763 | |
ca8bd38b CEB |
2764 | maxpages = read_swap_header(p, swap_header, inode); |
2765 | if (unlikely(!maxpages)) { | |
1da177e4 LT |
2766 | error = -EINVAL; |
2767 | goto bad_swap; | |
2768 | } | |
886bb7e9 | 2769 | |
81e33971 | 2770 | /* OK, set up the swap map and apply the bad block list */ |
803d0c83 | 2771 | swap_map = vzalloc(maxpages); |
81e33971 HD |
2772 | if (!swap_map) { |
2773 | error = -ENOMEM; | |
2774 | goto bad_swap; | |
2775 | } | |
f0571429 MK |
2776 | |
2777 | if (bdi_cap_stable_pages_required(inode_to_bdi(inode))) | |
2778 | p->flags |= SWP_STABLE_WRITES; | |
2779 | ||
2a8f9449 | 2780 | if (p->bdev && blk_queue_nonrot(bdev_get_queue(p->bdev))) { |
6f179af8 | 2781 | int cpu; |
235b6217 | 2782 | unsigned long ci, nr_cluster; |
6f179af8 | 2783 | |
2a8f9449 SL |
2784 | p->flags |= SWP_SOLIDSTATE; |
2785 | /* | |
2786 | * select a random position to start with to help wear leveling | |
2787 | * SSD | |
2788 | */ | |
2789 | p->cluster_next = 1 + (prandom_u32() % p->highest_bit); | |
235b6217 | 2790 | nr_cluster = DIV_ROUND_UP(maxpages, SWAPFILE_CLUSTER); |
2a8f9449 | 2791 | |
235b6217 | 2792 | cluster_info = vzalloc(nr_cluster * sizeof(*cluster_info)); |
2a8f9449 SL |
2793 | if (!cluster_info) { |
2794 | error = -ENOMEM; | |
2795 | goto bad_swap; | |
2796 | } | |
235b6217 HY |
2797 | |
2798 | for (ci = 0; ci < nr_cluster; ci++) | |
2799 | spin_lock_init(&((cluster_info + ci)->lock)); | |
2800 | ||
ebc2a1a6 SL |
2801 | p->percpu_cluster = alloc_percpu(struct percpu_cluster); |
2802 | if (!p->percpu_cluster) { | |
2803 | error = -ENOMEM; | |
2804 | goto bad_swap; | |
2805 | } | |
6f179af8 | 2806 | for_each_possible_cpu(cpu) { |
ebc2a1a6 | 2807 | struct percpu_cluster *cluster; |
6f179af8 | 2808 | cluster = per_cpu_ptr(p->percpu_cluster, cpu); |
ebc2a1a6 SL |
2809 | cluster_set_null(&cluster->index); |
2810 | } | |
2a8f9449 | 2811 | } |
1da177e4 | 2812 | |
1421ef3c CEB |
2813 | error = swap_cgroup_swapon(p->type, maxpages); |
2814 | if (error) | |
2815 | goto bad_swap; | |
2816 | ||
915d4d7b | 2817 | nr_extents = setup_swap_map_and_extents(p, swap_header, swap_map, |
2a8f9449 | 2818 | cluster_info, maxpages, &span); |
915d4d7b CEB |
2819 | if (unlikely(nr_extents < 0)) { |
2820 | error = nr_extents; | |
1da177e4 LT |
2821 | goto bad_swap; |
2822 | } | |
38b5faf4 | 2823 | /* frontswap enabled? set up bit-per-page map for frontswap */ |
8ea1d2a1 | 2824 | if (IS_ENABLED(CONFIG_FRONTSWAP)) |
7b57976d | 2825 | frontswap_map = vzalloc(BITS_TO_LONGS(maxpages) * sizeof(long)); |
1da177e4 | 2826 | |
2a8f9449 SL |
2827 | if (p->bdev &&(swap_flags & SWAP_FLAG_DISCARD) && swap_discardable(p)) { |
2828 | /* | |
2829 | * When discard is enabled for swap with no particular | |
2830 | * policy flagged, we set all swap discard flags here in | |
2831 | * order to sustain backward compatibility with older | |
2832 | * swapon(8) releases. | |
2833 | */ | |
2834 | p->flags |= (SWP_DISCARDABLE | SWP_AREA_DISCARD | | |
2835 | SWP_PAGE_DISCARD); | |
dcf6b7dd | 2836 | |
2a8f9449 SL |
2837 | /* |
2838 | * By flagging sys_swapon, a sysadmin can tell us to | |
2839 | * either do single-time area discards only, or to just | |
2840 | * perform discards for released swap page-clusters. | |
2841 | * Now it's time to adjust the p->flags accordingly. | |
2842 | */ | |
2843 | if (swap_flags & SWAP_FLAG_DISCARD_ONCE) | |
2844 | p->flags &= ~SWP_PAGE_DISCARD; | |
2845 | else if (swap_flags & SWAP_FLAG_DISCARD_PAGES) | |
2846 | p->flags &= ~SWP_AREA_DISCARD; | |
2847 | ||
2848 | /* issue a swapon-time discard if it's still required */ | |
2849 | if (p->flags & SWP_AREA_DISCARD) { | |
2850 | int err = discard_swap(p); | |
2851 | if (unlikely(err)) | |
2852 | pr_err("swapon: discard_swap(%p): %d\n", | |
2853 | p, err); | |
dcf6b7dd | 2854 | } |
20137a49 | 2855 | } |
6a6ba831 | 2856 | |
4b3ef9da HY |
2857 | error = init_swap_address_space(p->type, maxpages); |
2858 | if (error) | |
2859 | goto bad_swap; | |
2860 | ||
fc0abb14 | 2861 | mutex_lock(&swapon_mutex); |
40531542 | 2862 | prio = -1; |
78ecba08 | 2863 | if (swap_flags & SWAP_FLAG_PREFER) |
40531542 | 2864 | prio = |
78ecba08 | 2865 | (swap_flags & SWAP_FLAG_PRIO_MASK) >> SWAP_FLAG_PRIO_SHIFT; |
2a8f9449 | 2866 | enable_swap_info(p, prio, swap_map, cluster_info, frontswap_map); |
c69dbfb8 | 2867 | |
756a025f | 2868 | pr_info("Adding %uk swap on %s. Priority:%d extents:%d across:%lluk %s%s%s%s%s\n", |
91a27b2a | 2869 | p->pages<<(PAGE_SHIFT-10), name->name, p->prio, |
c69dbfb8 CEB |
2870 | nr_extents, (unsigned long long)span<<(PAGE_SHIFT-10), |
2871 | (p->flags & SWP_SOLIDSTATE) ? "SS" : "", | |
38b5faf4 | 2872 | (p->flags & SWP_DISCARDABLE) ? "D" : "", |
dcf6b7dd RA |
2873 | (p->flags & SWP_AREA_DISCARD) ? "s" : "", |
2874 | (p->flags & SWP_PAGE_DISCARD) ? "c" : "", | |
38b5faf4 | 2875 | (frontswap_map) ? "FS" : ""); |
c69dbfb8 | 2876 | |
fc0abb14 | 2877 | mutex_unlock(&swapon_mutex); |
66d7dd51 KS |
2878 | atomic_inc(&proc_poll_event); |
2879 | wake_up_interruptible(&proc_poll_wait); | |
2880 | ||
9b01c350 CEB |
2881 | if (S_ISREG(inode->i_mode)) |
2882 | inode->i_flags |= S_SWAPFILE; | |
1da177e4 LT |
2883 | error = 0; |
2884 | goto out; | |
2885 | bad_swap: | |
ebc2a1a6 SL |
2886 | free_percpu(p->percpu_cluster); |
2887 | p->percpu_cluster = NULL; | |
bd69010b | 2888 | if (inode && S_ISBLK(inode->i_mode) && p->bdev) { |
f2090d2d CEB |
2889 | set_blocksize(p->bdev, p->old_block_size); |
2890 | blkdev_put(p->bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL); | |
1da177e4 | 2891 | } |
4cd3bb10 | 2892 | destroy_swap_extents(p); |
e8e6c2ec | 2893 | swap_cgroup_swapoff(p->type); |
5d337b91 | 2894 | spin_lock(&swap_lock); |
1da177e4 | 2895 | p->swap_file = NULL; |
1da177e4 | 2896 | p->flags = 0; |
5d337b91 | 2897 | spin_unlock(&swap_lock); |
1da177e4 | 2898 | vfree(swap_map); |
2a8f9449 | 2899 | vfree(cluster_info); |
52c50567 | 2900 | if (swap_file) { |
2130781e | 2901 | if (inode && S_ISREG(inode->i_mode)) { |
5955102c | 2902 | inode_unlock(inode); |
2130781e CEB |
2903 | inode = NULL; |
2904 | } | |
1da177e4 | 2905 | filp_close(swap_file, NULL); |
52c50567 | 2906 | } |
1da177e4 LT |
2907 | out: |
2908 | if (page && !IS_ERR(page)) { | |
2909 | kunmap(page); | |
09cbfeaf | 2910 | put_page(page); |
1da177e4 LT |
2911 | } |
2912 | if (name) | |
2913 | putname(name); | |
9b01c350 | 2914 | if (inode && S_ISREG(inode->i_mode)) |
5955102c | 2915 | inode_unlock(inode); |
039939a6 TC |
2916 | if (!error) |
2917 | enable_swap_slots_cache(); | |
1da177e4 LT |
2918 | return error; |
2919 | } | |
2920 | ||
2921 | void si_swapinfo(struct sysinfo *val) | |
2922 | { | |
efa90a98 | 2923 | unsigned int type; |
1da177e4 LT |
2924 | unsigned long nr_to_be_unused = 0; |
2925 | ||
5d337b91 | 2926 | spin_lock(&swap_lock); |
efa90a98 HD |
2927 | for (type = 0; type < nr_swapfiles; type++) { |
2928 | struct swap_info_struct *si = swap_info[type]; | |
2929 | ||
2930 | if ((si->flags & SWP_USED) && !(si->flags & SWP_WRITEOK)) | |
2931 | nr_to_be_unused += si->inuse_pages; | |
1da177e4 | 2932 | } |
ec8acf20 | 2933 | val->freeswap = atomic_long_read(&nr_swap_pages) + nr_to_be_unused; |
1da177e4 | 2934 | val->totalswap = total_swap_pages + nr_to_be_unused; |
5d337b91 | 2935 | spin_unlock(&swap_lock); |
1da177e4 LT |
2936 | } |
2937 | ||
2938 | /* | |
2939 | * Verify that a swap entry is valid and increment its swap map count. | |
2940 | * | |
355cfa73 KH |
2941 | * Returns error code in following case. |
2942 | * - success -> 0 | |
2943 | * - swp_entry is invalid -> EINVAL | |
2944 | * - swp_entry is migration entry -> EINVAL | |
2945 | * - swap-cache reference is requested but there is already one. -> EEXIST | |
2946 | * - swap-cache reference is requested but the entry is not used. -> ENOENT | |
570a335b | 2947 | * - swap-mapped reference requested but needs continued swap count. -> ENOMEM |
1da177e4 | 2948 | */ |
8d69aaee | 2949 | static int __swap_duplicate(swp_entry_t entry, unsigned char usage) |
1da177e4 | 2950 | { |
73c34b6a | 2951 | struct swap_info_struct *p; |
235b6217 | 2952 | struct swap_cluster_info *ci; |
1da177e4 | 2953 | unsigned long offset, type; |
8d69aaee HD |
2954 | unsigned char count; |
2955 | unsigned char has_cache; | |
253d553b | 2956 | int err = -EINVAL; |
1da177e4 | 2957 | |
a7420aa5 | 2958 | if (non_swap_entry(entry)) |
253d553b | 2959 | goto out; |
0697212a | 2960 | |
1da177e4 LT |
2961 | type = swp_type(entry); |
2962 | if (type >= nr_swapfiles) | |
2963 | goto bad_file; | |
efa90a98 | 2964 | p = swap_info[type]; |
1da177e4 | 2965 | offset = swp_offset(entry); |
355cfa73 | 2966 | if (unlikely(offset >= p->max)) |
235b6217 HY |
2967 | goto out; |
2968 | ||
2969 | ci = lock_cluster_or_swap_info(p, offset); | |
355cfa73 | 2970 | |
253d553b | 2971 | count = p->swap_map[offset]; |
edfe23da SL |
2972 | |
2973 | /* | |
2974 | * swapin_readahead() doesn't check if a swap entry is valid, so the | |
2975 | * swap entry could be SWAP_MAP_BAD. Check here with lock held. | |
2976 | */ | |
2977 | if (unlikely(swap_count(count) == SWAP_MAP_BAD)) { | |
2978 | err = -ENOENT; | |
2979 | goto unlock_out; | |
2980 | } | |
2981 | ||
253d553b HD |
2982 | has_cache = count & SWAP_HAS_CACHE; |
2983 | count &= ~SWAP_HAS_CACHE; | |
2984 | err = 0; | |
355cfa73 | 2985 | |
253d553b | 2986 | if (usage == SWAP_HAS_CACHE) { |
355cfa73 KH |
2987 | |
2988 | /* set SWAP_HAS_CACHE if there is no cache and entry is used */ | |
253d553b HD |
2989 | if (!has_cache && count) |
2990 | has_cache = SWAP_HAS_CACHE; | |
2991 | else if (has_cache) /* someone else added cache */ | |
2992 | err = -EEXIST; | |
2993 | else /* no users remaining */ | |
2994 | err = -ENOENT; | |
355cfa73 KH |
2995 | |
2996 | } else if (count || has_cache) { | |
253d553b | 2997 | |
570a335b HD |
2998 | if ((count & ~COUNT_CONTINUED) < SWAP_MAP_MAX) |
2999 | count += usage; | |
3000 | else if ((count & ~COUNT_CONTINUED) > SWAP_MAP_MAX) | |
253d553b | 3001 | err = -EINVAL; |
570a335b HD |
3002 | else if (swap_count_continued(p, offset, count)) |
3003 | count = COUNT_CONTINUED; | |
3004 | else | |
3005 | err = -ENOMEM; | |
355cfa73 | 3006 | } else |
253d553b HD |
3007 | err = -ENOENT; /* unused swap entry */ |
3008 | ||
3009 | p->swap_map[offset] = count | has_cache; | |
3010 | ||
355cfa73 | 3011 | unlock_out: |
235b6217 | 3012 | unlock_cluster_or_swap_info(p, ci); |
1da177e4 | 3013 | out: |
253d553b | 3014 | return err; |
1da177e4 LT |
3015 | |
3016 | bad_file: | |
465c47fd | 3017 | pr_err("swap_dup: %s%08lx\n", Bad_file, entry.val); |
1da177e4 LT |
3018 | goto out; |
3019 | } | |
253d553b | 3020 | |
aaa46865 HD |
3021 | /* |
3022 | * Help swapoff by noting that swap entry belongs to shmem/tmpfs | |
3023 | * (in which case its reference count is never incremented). | |
3024 | */ | |
3025 | void swap_shmem_alloc(swp_entry_t entry) | |
3026 | { | |
3027 | __swap_duplicate(entry, SWAP_MAP_SHMEM); | |
3028 | } | |
3029 | ||
355cfa73 | 3030 | /* |
08259d58 HD |
3031 | * Increase reference count of swap entry by 1. |
3032 | * Returns 0 for success, or -ENOMEM if a swap_count_continuation is required | |
3033 | * but could not be atomically allocated. Returns 0, just as if it succeeded, | |
3034 | * if __swap_duplicate() fails for another reason (-EINVAL or -ENOENT), which | |
3035 | * might occur if a page table entry has got corrupted. | |
355cfa73 | 3036 | */ |
570a335b | 3037 | int swap_duplicate(swp_entry_t entry) |
355cfa73 | 3038 | { |
570a335b HD |
3039 | int err = 0; |
3040 | ||
3041 | while (!err && __swap_duplicate(entry, 1) == -ENOMEM) | |
3042 | err = add_swap_count_continuation(entry, GFP_ATOMIC); | |
3043 | return err; | |
355cfa73 | 3044 | } |
1da177e4 | 3045 | |
cb4b86ba | 3046 | /* |
355cfa73 KH |
3047 | * @entry: swap entry for which we allocate swap cache. |
3048 | * | |
73c34b6a | 3049 | * Called when allocating swap cache for existing swap entry, |
355cfa73 KH |
3050 | * This can return error codes. Returns 0 at success. |
3051 | * -EBUSY means there is a swap cache. | |
3052 | * Note: return code is different from swap_duplicate(). | |
cb4b86ba KH |
3053 | */ |
3054 | int swapcache_prepare(swp_entry_t entry) | |
3055 | { | |
253d553b | 3056 | return __swap_duplicate(entry, SWAP_HAS_CACHE); |
cb4b86ba KH |
3057 | } |
3058 | ||
f981c595 MG |
3059 | struct swap_info_struct *page_swap_info(struct page *page) |
3060 | { | |
3061 | swp_entry_t swap = { .val = page_private(page) }; | |
f981c595 MG |
3062 | return swap_info[swp_type(swap)]; |
3063 | } | |
3064 | ||
3065 | /* | |
3066 | * out-of-line __page_file_ methods to avoid include hell. | |
3067 | */ | |
3068 | struct address_space *__page_file_mapping(struct page *page) | |
3069 | { | |
309381fe | 3070 | VM_BUG_ON_PAGE(!PageSwapCache(page), page); |
f981c595 MG |
3071 | return page_swap_info(page)->swap_file->f_mapping; |
3072 | } | |
3073 | EXPORT_SYMBOL_GPL(__page_file_mapping); | |
3074 | ||
3075 | pgoff_t __page_file_index(struct page *page) | |
3076 | { | |
3077 | swp_entry_t swap = { .val = page_private(page) }; | |
309381fe | 3078 | VM_BUG_ON_PAGE(!PageSwapCache(page), page); |
f981c595 MG |
3079 | return swp_offset(swap); |
3080 | } | |
3081 | EXPORT_SYMBOL_GPL(__page_file_index); | |
3082 | ||
570a335b HD |
3083 | /* |
3084 | * add_swap_count_continuation - called when a swap count is duplicated | |
3085 | * beyond SWAP_MAP_MAX, it allocates a new page and links that to the entry's | |
3086 | * page of the original vmalloc'ed swap_map, to hold the continuation count | |
3087 | * (for that entry and for its neighbouring PAGE_SIZE swap entries). Called | |
3088 | * again when count is duplicated beyond SWAP_MAP_MAX * SWAP_CONT_MAX, etc. | |
3089 | * | |
3090 | * These continuation pages are seldom referenced: the common paths all work | |
3091 | * on the original swap_map, only referring to a continuation page when the | |
3092 | * low "digit" of a count is incremented or decremented through SWAP_MAP_MAX. | |
3093 | * | |
3094 | * add_swap_count_continuation(, GFP_ATOMIC) can be called while holding | |
3095 | * page table locks; if it fails, add_swap_count_continuation(, GFP_KERNEL) | |
3096 | * can be called after dropping locks. | |
3097 | */ | |
3098 | int add_swap_count_continuation(swp_entry_t entry, gfp_t gfp_mask) | |
3099 | { | |
3100 | struct swap_info_struct *si; | |
235b6217 | 3101 | struct swap_cluster_info *ci; |
570a335b HD |
3102 | struct page *head; |
3103 | struct page *page; | |
3104 | struct page *list_page; | |
3105 | pgoff_t offset; | |
3106 | unsigned char count; | |
3107 | ||
3108 | /* | |
3109 | * When debugging, it's easier to use __GFP_ZERO here; but it's better | |
3110 | * for latency not to zero a page while GFP_ATOMIC and holding locks. | |
3111 | */ | |
3112 | page = alloc_page(gfp_mask | __GFP_HIGHMEM); | |
3113 | ||
3114 | si = swap_info_get(entry); | |
3115 | if (!si) { | |
3116 | /* | |
3117 | * An acceptable race has occurred since the failing | |
3118 | * __swap_duplicate(): the swap entry has been freed, | |
3119 | * perhaps even the whole swap_map cleared for swapoff. | |
3120 | */ | |
3121 | goto outer; | |
3122 | } | |
3123 | ||
3124 | offset = swp_offset(entry); | |
235b6217 HY |
3125 | |
3126 | ci = lock_cluster(si, offset); | |
3127 | ||
570a335b HD |
3128 | count = si->swap_map[offset] & ~SWAP_HAS_CACHE; |
3129 | ||
3130 | if ((count & ~COUNT_CONTINUED) != SWAP_MAP_MAX) { | |
3131 | /* | |
3132 | * The higher the swap count, the more likely it is that tasks | |
3133 | * will race to add swap count continuation: we need to avoid | |
3134 | * over-provisioning. | |
3135 | */ | |
3136 | goto out; | |
3137 | } | |
3138 | ||
3139 | if (!page) { | |
235b6217 | 3140 | unlock_cluster(ci); |
ec8acf20 | 3141 | spin_unlock(&si->lock); |
570a335b HD |
3142 | return -ENOMEM; |
3143 | } | |
3144 | ||
3145 | /* | |
3146 | * We are fortunate that although vmalloc_to_page uses pte_offset_map, | |
2de1a7e4 SJ |
3147 | * no architecture is using highmem pages for kernel page tables: so it |
3148 | * will not corrupt the GFP_ATOMIC caller's atomic page table kmaps. | |
570a335b HD |
3149 | */ |
3150 | head = vmalloc_to_page(si->swap_map + offset); | |
3151 | offset &= ~PAGE_MASK; | |
3152 | ||
3153 | /* | |
3154 | * Page allocation does not initialize the page's lru field, | |
3155 | * but it does always reset its private field. | |
3156 | */ | |
3157 | if (!page_private(head)) { | |
3158 | BUG_ON(count & COUNT_CONTINUED); | |
3159 | INIT_LIST_HEAD(&head->lru); | |
3160 | set_page_private(head, SWP_CONTINUED); | |
3161 | si->flags |= SWP_CONTINUED; | |
3162 | } | |
3163 | ||
3164 | list_for_each_entry(list_page, &head->lru, lru) { | |
3165 | unsigned char *map; | |
3166 | ||
3167 | /* | |
3168 | * If the previous map said no continuation, but we've found | |
3169 | * a continuation page, free our allocation and use this one. | |
3170 | */ | |
3171 | if (!(count & COUNT_CONTINUED)) | |
3172 | goto out; | |
3173 | ||
9b04c5fe | 3174 | map = kmap_atomic(list_page) + offset; |
570a335b | 3175 | count = *map; |
9b04c5fe | 3176 | kunmap_atomic(map); |
570a335b HD |
3177 | |
3178 | /* | |
3179 | * If this continuation count now has some space in it, | |
3180 | * free our allocation and use this one. | |
3181 | */ | |
3182 | if ((count & ~COUNT_CONTINUED) != SWAP_CONT_MAX) | |
3183 | goto out; | |
3184 | } | |
3185 | ||
3186 | list_add_tail(&page->lru, &head->lru); | |
3187 | page = NULL; /* now it's attached, don't free it */ | |
3188 | out: | |
235b6217 | 3189 | unlock_cluster(ci); |
ec8acf20 | 3190 | spin_unlock(&si->lock); |
570a335b HD |
3191 | outer: |
3192 | if (page) | |
3193 | __free_page(page); | |
3194 | return 0; | |
3195 | } | |
3196 | ||
3197 | /* | |
3198 | * swap_count_continued - when the original swap_map count is incremented | |
3199 | * from SWAP_MAP_MAX, check if there is already a continuation page to carry | |
3200 | * into, carry if so, or else fail until a new continuation page is allocated; | |
3201 | * when the original swap_map count is decremented from 0 with continuation, | |
3202 | * borrow from the continuation and report whether it still holds more. | |
235b6217 HY |
3203 | * Called while __swap_duplicate() or swap_entry_free() holds swap or cluster |
3204 | * lock. | |
570a335b HD |
3205 | */ |
3206 | static bool swap_count_continued(struct swap_info_struct *si, | |
3207 | pgoff_t offset, unsigned char count) | |
3208 | { | |
3209 | struct page *head; | |
3210 | struct page *page; | |
3211 | unsigned char *map; | |
3212 | ||
3213 | head = vmalloc_to_page(si->swap_map + offset); | |
3214 | if (page_private(head) != SWP_CONTINUED) { | |
3215 | BUG_ON(count & COUNT_CONTINUED); | |
3216 | return false; /* need to add count continuation */ | |
3217 | } | |
3218 | ||
3219 | offset &= ~PAGE_MASK; | |
3220 | page = list_entry(head->lru.next, struct page, lru); | |
9b04c5fe | 3221 | map = kmap_atomic(page) + offset; |
570a335b HD |
3222 | |
3223 | if (count == SWAP_MAP_MAX) /* initial increment from swap_map */ | |
3224 | goto init_map; /* jump over SWAP_CONT_MAX checks */ | |
3225 | ||
3226 | if (count == (SWAP_MAP_MAX | COUNT_CONTINUED)) { /* incrementing */ | |
3227 | /* | |
3228 | * Think of how you add 1 to 999 | |
3229 | */ | |
3230 | while (*map == (SWAP_CONT_MAX | COUNT_CONTINUED)) { | |
9b04c5fe | 3231 | kunmap_atomic(map); |
570a335b HD |
3232 | page = list_entry(page->lru.next, struct page, lru); |
3233 | BUG_ON(page == head); | |
9b04c5fe | 3234 | map = kmap_atomic(page) + offset; |
570a335b HD |
3235 | } |
3236 | if (*map == SWAP_CONT_MAX) { | |
9b04c5fe | 3237 | kunmap_atomic(map); |
570a335b HD |
3238 | page = list_entry(page->lru.next, struct page, lru); |
3239 | if (page == head) | |
3240 | return false; /* add count continuation */ | |
9b04c5fe | 3241 | map = kmap_atomic(page) + offset; |
570a335b HD |
3242 | init_map: *map = 0; /* we didn't zero the page */ |
3243 | } | |
3244 | *map += 1; | |
9b04c5fe | 3245 | kunmap_atomic(map); |
570a335b HD |
3246 | page = list_entry(page->lru.prev, struct page, lru); |
3247 | while (page != head) { | |
9b04c5fe | 3248 | map = kmap_atomic(page) + offset; |
570a335b | 3249 | *map = COUNT_CONTINUED; |
9b04c5fe | 3250 | kunmap_atomic(map); |
570a335b HD |
3251 | page = list_entry(page->lru.prev, struct page, lru); |
3252 | } | |
3253 | return true; /* incremented */ | |
3254 | ||
3255 | } else { /* decrementing */ | |
3256 | /* | |
3257 | * Think of how you subtract 1 from 1000 | |
3258 | */ | |
3259 | BUG_ON(count != COUNT_CONTINUED); | |
3260 | while (*map == COUNT_CONTINUED) { | |
9b04c5fe | 3261 | kunmap_atomic(map); |
570a335b HD |
3262 | page = list_entry(page->lru.next, struct page, lru); |
3263 | BUG_ON(page == head); | |
9b04c5fe | 3264 | map = kmap_atomic(page) + offset; |
570a335b HD |
3265 | } |
3266 | BUG_ON(*map == 0); | |
3267 | *map -= 1; | |
3268 | if (*map == 0) | |
3269 | count = 0; | |
9b04c5fe | 3270 | kunmap_atomic(map); |
570a335b HD |
3271 | page = list_entry(page->lru.prev, struct page, lru); |
3272 | while (page != head) { | |
9b04c5fe | 3273 | map = kmap_atomic(page) + offset; |
570a335b HD |
3274 | *map = SWAP_CONT_MAX | count; |
3275 | count = COUNT_CONTINUED; | |
9b04c5fe | 3276 | kunmap_atomic(map); |
570a335b HD |
3277 | page = list_entry(page->lru.prev, struct page, lru); |
3278 | } | |
3279 | return count == COUNT_CONTINUED; | |
3280 | } | |
3281 | } | |
3282 | ||
3283 | /* | |
3284 | * free_swap_count_continuations - swapoff free all the continuation pages | |
3285 | * appended to the swap_map, after swap_map is quiesced, before vfree'ing it. | |
3286 | */ | |
3287 | static void free_swap_count_continuations(struct swap_info_struct *si) | |
3288 | { | |
3289 | pgoff_t offset; | |
3290 | ||
3291 | for (offset = 0; offset < si->max; offset += PAGE_SIZE) { | |
3292 | struct page *head; | |
3293 | head = vmalloc_to_page(si->swap_map + offset); | |
3294 | if (page_private(head)) { | |
0d576d20 GT |
3295 | struct page *page, *next; |
3296 | ||
3297 | list_for_each_entry_safe(page, next, &head->lru, lru) { | |
3298 | list_del(&page->lru); | |
570a335b HD |
3299 | __free_page(page); |
3300 | } | |
3301 | } | |
3302 | } | |
3303 | } |