]>
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 | ||
38d8b4e6 HY |
202 | #ifdef CONFIG_THP_SWAP |
203 | #define SWAPFILE_CLUSTER HPAGE_PMD_NR | |
204 | #else | |
048c27fd | 205 | #define SWAPFILE_CLUSTER 256 |
38d8b4e6 | 206 | #endif |
048c27fd HD |
207 | #define LATENCY_LIMIT 256 |
208 | ||
2a8f9449 SL |
209 | static inline void cluster_set_flag(struct swap_cluster_info *info, |
210 | unsigned int flag) | |
211 | { | |
212 | info->flags = flag; | |
213 | } | |
214 | ||
215 | static inline unsigned int cluster_count(struct swap_cluster_info *info) | |
216 | { | |
217 | return info->data; | |
218 | } | |
219 | ||
220 | static inline void cluster_set_count(struct swap_cluster_info *info, | |
221 | unsigned int c) | |
222 | { | |
223 | info->data = c; | |
224 | } | |
225 | ||
226 | static inline void cluster_set_count_flag(struct swap_cluster_info *info, | |
227 | unsigned int c, unsigned int f) | |
228 | { | |
229 | info->flags = f; | |
230 | info->data = c; | |
231 | } | |
232 | ||
233 | static inline unsigned int cluster_next(struct swap_cluster_info *info) | |
234 | { | |
235 | return info->data; | |
236 | } | |
237 | ||
238 | static inline void cluster_set_next(struct swap_cluster_info *info, | |
239 | unsigned int n) | |
240 | { | |
241 | info->data = n; | |
242 | } | |
243 | ||
244 | static inline void cluster_set_next_flag(struct swap_cluster_info *info, | |
245 | unsigned int n, unsigned int f) | |
246 | { | |
247 | info->flags = f; | |
248 | info->data = n; | |
249 | } | |
250 | ||
251 | static inline bool cluster_is_free(struct swap_cluster_info *info) | |
252 | { | |
253 | return info->flags & CLUSTER_FLAG_FREE; | |
254 | } | |
255 | ||
256 | static inline bool cluster_is_null(struct swap_cluster_info *info) | |
257 | { | |
258 | return info->flags & CLUSTER_FLAG_NEXT_NULL; | |
259 | } | |
260 | ||
261 | static inline void cluster_set_null(struct swap_cluster_info *info) | |
262 | { | |
263 | info->flags = CLUSTER_FLAG_NEXT_NULL; | |
264 | info->data = 0; | |
265 | } | |
266 | ||
235b6217 HY |
267 | static inline struct swap_cluster_info *lock_cluster(struct swap_info_struct *si, |
268 | unsigned long offset) | |
269 | { | |
270 | struct swap_cluster_info *ci; | |
271 | ||
272 | ci = si->cluster_info; | |
273 | if (ci) { | |
274 | ci += offset / SWAPFILE_CLUSTER; | |
275 | spin_lock(&ci->lock); | |
276 | } | |
277 | return ci; | |
278 | } | |
279 | ||
280 | static inline void unlock_cluster(struct swap_cluster_info *ci) | |
281 | { | |
282 | if (ci) | |
283 | spin_unlock(&ci->lock); | |
284 | } | |
285 | ||
286 | static inline struct swap_cluster_info *lock_cluster_or_swap_info( | |
287 | struct swap_info_struct *si, | |
288 | unsigned long offset) | |
289 | { | |
290 | struct swap_cluster_info *ci; | |
291 | ||
292 | ci = lock_cluster(si, offset); | |
293 | if (!ci) | |
294 | spin_lock(&si->lock); | |
295 | ||
296 | return ci; | |
297 | } | |
298 | ||
299 | static inline void unlock_cluster_or_swap_info(struct swap_info_struct *si, | |
300 | struct swap_cluster_info *ci) | |
301 | { | |
302 | if (ci) | |
303 | unlock_cluster(ci); | |
304 | else | |
305 | spin_unlock(&si->lock); | |
306 | } | |
307 | ||
6b534915 HY |
308 | static inline bool cluster_list_empty(struct swap_cluster_list *list) |
309 | { | |
310 | return cluster_is_null(&list->head); | |
311 | } | |
312 | ||
313 | static inline unsigned int cluster_list_first(struct swap_cluster_list *list) | |
314 | { | |
315 | return cluster_next(&list->head); | |
316 | } | |
317 | ||
318 | static void cluster_list_init(struct swap_cluster_list *list) | |
319 | { | |
320 | cluster_set_null(&list->head); | |
321 | cluster_set_null(&list->tail); | |
322 | } | |
323 | ||
324 | static void cluster_list_add_tail(struct swap_cluster_list *list, | |
325 | struct swap_cluster_info *ci, | |
326 | unsigned int idx) | |
327 | { | |
328 | if (cluster_list_empty(list)) { | |
329 | cluster_set_next_flag(&list->head, idx, 0); | |
330 | cluster_set_next_flag(&list->tail, idx, 0); | |
331 | } else { | |
235b6217 | 332 | struct swap_cluster_info *ci_tail; |
6b534915 HY |
333 | unsigned int tail = cluster_next(&list->tail); |
334 | ||
235b6217 HY |
335 | /* |
336 | * Nested cluster lock, but both cluster locks are | |
337 | * only acquired when we held swap_info_struct->lock | |
338 | */ | |
339 | ci_tail = ci + tail; | |
340 | spin_lock_nested(&ci_tail->lock, SINGLE_DEPTH_NESTING); | |
341 | cluster_set_next(ci_tail, idx); | |
0ef017d1 | 342 | spin_unlock(&ci_tail->lock); |
6b534915 HY |
343 | cluster_set_next_flag(&list->tail, idx, 0); |
344 | } | |
345 | } | |
346 | ||
347 | static unsigned int cluster_list_del_first(struct swap_cluster_list *list, | |
348 | struct swap_cluster_info *ci) | |
349 | { | |
350 | unsigned int idx; | |
351 | ||
352 | idx = cluster_next(&list->head); | |
353 | if (cluster_next(&list->tail) == idx) { | |
354 | cluster_set_null(&list->head); | |
355 | cluster_set_null(&list->tail); | |
356 | } else | |
357 | cluster_set_next_flag(&list->head, | |
358 | cluster_next(&ci[idx]), 0); | |
359 | ||
360 | return idx; | |
361 | } | |
362 | ||
815c2c54 SL |
363 | /* Add a cluster to discard list and schedule it to do discard */ |
364 | static void swap_cluster_schedule_discard(struct swap_info_struct *si, | |
365 | unsigned int idx) | |
366 | { | |
367 | /* | |
368 | * If scan_swap_map() can't find a free cluster, it will check | |
369 | * si->swap_map directly. To make sure the discarding cluster isn't | |
370 | * taken by scan_swap_map(), mark the swap entries bad (occupied). It | |
371 | * will be cleared after discard | |
372 | */ | |
373 | memset(si->swap_map + idx * SWAPFILE_CLUSTER, | |
374 | SWAP_MAP_BAD, SWAPFILE_CLUSTER); | |
375 | ||
6b534915 | 376 | cluster_list_add_tail(&si->discard_clusters, si->cluster_info, idx); |
815c2c54 SL |
377 | |
378 | schedule_work(&si->discard_work); | |
379 | } | |
380 | ||
38d8b4e6 HY |
381 | static void __free_cluster(struct swap_info_struct *si, unsigned long idx) |
382 | { | |
383 | struct swap_cluster_info *ci = si->cluster_info; | |
384 | ||
385 | cluster_set_flag(ci + idx, CLUSTER_FLAG_FREE); | |
386 | cluster_list_add_tail(&si->free_clusters, ci, idx); | |
387 | } | |
388 | ||
815c2c54 SL |
389 | /* |
390 | * Doing discard actually. After a cluster discard is finished, the cluster | |
391 | * will be added to free cluster list. caller should hold si->lock. | |
392 | */ | |
393 | static void swap_do_scheduled_discard(struct swap_info_struct *si) | |
394 | { | |
235b6217 | 395 | struct swap_cluster_info *info, *ci; |
815c2c54 SL |
396 | unsigned int idx; |
397 | ||
398 | info = si->cluster_info; | |
399 | ||
6b534915 HY |
400 | while (!cluster_list_empty(&si->discard_clusters)) { |
401 | idx = cluster_list_del_first(&si->discard_clusters, info); | |
815c2c54 SL |
402 | spin_unlock(&si->lock); |
403 | ||
404 | discard_swap_cluster(si, idx * SWAPFILE_CLUSTER, | |
405 | SWAPFILE_CLUSTER); | |
406 | ||
407 | spin_lock(&si->lock); | |
235b6217 | 408 | ci = lock_cluster(si, idx * SWAPFILE_CLUSTER); |
38d8b4e6 | 409 | __free_cluster(si, idx); |
815c2c54 SL |
410 | memset(si->swap_map + idx * SWAPFILE_CLUSTER, |
411 | 0, SWAPFILE_CLUSTER); | |
235b6217 | 412 | unlock_cluster(ci); |
815c2c54 SL |
413 | } |
414 | } | |
415 | ||
416 | static void swap_discard_work(struct work_struct *work) | |
417 | { | |
418 | struct swap_info_struct *si; | |
419 | ||
420 | si = container_of(work, struct swap_info_struct, discard_work); | |
421 | ||
422 | spin_lock(&si->lock); | |
423 | swap_do_scheduled_discard(si); | |
424 | spin_unlock(&si->lock); | |
425 | } | |
426 | ||
38d8b4e6 HY |
427 | static void alloc_cluster(struct swap_info_struct *si, unsigned long idx) |
428 | { | |
429 | struct swap_cluster_info *ci = si->cluster_info; | |
430 | ||
431 | VM_BUG_ON(cluster_list_first(&si->free_clusters) != idx); | |
432 | cluster_list_del_first(&si->free_clusters, ci); | |
433 | cluster_set_count_flag(ci + idx, 0, 0); | |
434 | } | |
435 | ||
436 | static void free_cluster(struct swap_info_struct *si, unsigned long idx) | |
437 | { | |
438 | struct swap_cluster_info *ci = si->cluster_info + idx; | |
439 | ||
440 | VM_BUG_ON(cluster_count(ci) != 0); | |
441 | /* | |
442 | * If the swap is discardable, prepare discard the cluster | |
443 | * instead of free it immediately. The cluster will be freed | |
444 | * after discard. | |
445 | */ | |
446 | if ((si->flags & (SWP_WRITEOK | SWP_PAGE_DISCARD)) == | |
447 | (SWP_WRITEOK | SWP_PAGE_DISCARD)) { | |
448 | swap_cluster_schedule_discard(si, idx); | |
449 | return; | |
450 | } | |
451 | ||
452 | __free_cluster(si, idx); | |
453 | } | |
454 | ||
2a8f9449 SL |
455 | /* |
456 | * The cluster corresponding to page_nr will be used. The cluster will be | |
457 | * removed from free cluster list and its usage counter will be increased. | |
458 | */ | |
459 | static void inc_cluster_info_page(struct swap_info_struct *p, | |
460 | struct swap_cluster_info *cluster_info, unsigned long page_nr) | |
461 | { | |
462 | unsigned long idx = page_nr / SWAPFILE_CLUSTER; | |
463 | ||
464 | if (!cluster_info) | |
465 | return; | |
38d8b4e6 HY |
466 | if (cluster_is_free(&cluster_info[idx])) |
467 | alloc_cluster(p, idx); | |
2a8f9449 SL |
468 | |
469 | VM_BUG_ON(cluster_count(&cluster_info[idx]) >= SWAPFILE_CLUSTER); | |
470 | cluster_set_count(&cluster_info[idx], | |
471 | cluster_count(&cluster_info[idx]) + 1); | |
472 | } | |
473 | ||
474 | /* | |
475 | * The cluster corresponding to page_nr decreases one usage. If the usage | |
476 | * counter becomes 0, which means no page in the cluster is in using, we can | |
477 | * optionally discard the cluster and add it to free cluster list. | |
478 | */ | |
479 | static void dec_cluster_info_page(struct swap_info_struct *p, | |
480 | struct swap_cluster_info *cluster_info, unsigned long page_nr) | |
481 | { | |
482 | unsigned long idx = page_nr / SWAPFILE_CLUSTER; | |
483 | ||
484 | if (!cluster_info) | |
485 | return; | |
486 | ||
487 | VM_BUG_ON(cluster_count(&cluster_info[idx]) == 0); | |
488 | cluster_set_count(&cluster_info[idx], | |
489 | cluster_count(&cluster_info[idx]) - 1); | |
490 | ||
38d8b4e6 HY |
491 | if (cluster_count(&cluster_info[idx]) == 0) |
492 | free_cluster(p, idx); | |
2a8f9449 SL |
493 | } |
494 | ||
495 | /* | |
496 | * It's possible scan_swap_map() uses a free cluster in the middle of free | |
497 | * cluster list. Avoiding such abuse to avoid list corruption. | |
498 | */ | |
ebc2a1a6 SL |
499 | static bool |
500 | scan_swap_map_ssd_cluster_conflict(struct swap_info_struct *si, | |
2a8f9449 SL |
501 | unsigned long offset) |
502 | { | |
ebc2a1a6 SL |
503 | struct percpu_cluster *percpu_cluster; |
504 | bool conflict; | |
505 | ||
2a8f9449 | 506 | offset /= SWAPFILE_CLUSTER; |
6b534915 HY |
507 | conflict = !cluster_list_empty(&si->free_clusters) && |
508 | offset != cluster_list_first(&si->free_clusters) && | |
2a8f9449 | 509 | cluster_is_free(&si->cluster_info[offset]); |
ebc2a1a6 SL |
510 | |
511 | if (!conflict) | |
512 | return false; | |
513 | ||
514 | percpu_cluster = this_cpu_ptr(si->percpu_cluster); | |
515 | cluster_set_null(&percpu_cluster->index); | |
516 | return true; | |
517 | } | |
518 | ||
519 | /* | |
520 | * Try to get a swap entry from current cpu's swap entry pool (a cluster). This | |
521 | * might involve allocating a new cluster for current CPU too. | |
522 | */ | |
36005bae | 523 | static bool scan_swap_map_try_ssd_cluster(struct swap_info_struct *si, |
ebc2a1a6 SL |
524 | unsigned long *offset, unsigned long *scan_base) |
525 | { | |
526 | struct percpu_cluster *cluster; | |
235b6217 | 527 | struct swap_cluster_info *ci; |
ebc2a1a6 | 528 | bool found_free; |
235b6217 | 529 | unsigned long tmp, max; |
ebc2a1a6 SL |
530 | |
531 | new_cluster: | |
532 | cluster = this_cpu_ptr(si->percpu_cluster); | |
533 | if (cluster_is_null(&cluster->index)) { | |
6b534915 HY |
534 | if (!cluster_list_empty(&si->free_clusters)) { |
535 | cluster->index = si->free_clusters.head; | |
ebc2a1a6 SL |
536 | cluster->next = cluster_next(&cluster->index) * |
537 | SWAPFILE_CLUSTER; | |
6b534915 | 538 | } else if (!cluster_list_empty(&si->discard_clusters)) { |
ebc2a1a6 SL |
539 | /* |
540 | * we don't have free cluster but have some clusters in | |
541 | * discarding, do discard now and reclaim them | |
542 | */ | |
543 | swap_do_scheduled_discard(si); | |
544 | *scan_base = *offset = si->cluster_next; | |
545 | goto new_cluster; | |
546 | } else | |
36005bae | 547 | return false; |
ebc2a1a6 SL |
548 | } |
549 | ||
550 | found_free = false; | |
551 | ||
552 | /* | |
553 | * Other CPUs can use our cluster if they can't find a free cluster, | |
554 | * check if there is still free entry in the cluster | |
555 | */ | |
556 | tmp = cluster->next; | |
235b6217 HY |
557 | max = min_t(unsigned long, si->max, |
558 | (cluster_next(&cluster->index) + 1) * SWAPFILE_CLUSTER); | |
559 | if (tmp >= max) { | |
560 | cluster_set_null(&cluster->index); | |
561 | goto new_cluster; | |
562 | } | |
563 | ci = lock_cluster(si, tmp); | |
564 | while (tmp < max) { | |
ebc2a1a6 SL |
565 | if (!si->swap_map[tmp]) { |
566 | found_free = true; | |
567 | break; | |
568 | } | |
569 | tmp++; | |
570 | } | |
235b6217 | 571 | unlock_cluster(ci); |
ebc2a1a6 SL |
572 | if (!found_free) { |
573 | cluster_set_null(&cluster->index); | |
574 | goto new_cluster; | |
575 | } | |
576 | cluster->next = tmp + 1; | |
577 | *offset = tmp; | |
578 | *scan_base = tmp; | |
36005bae | 579 | return found_free; |
2a8f9449 SL |
580 | } |
581 | ||
38d8b4e6 HY |
582 | static void swap_range_alloc(struct swap_info_struct *si, unsigned long offset, |
583 | unsigned int nr_entries) | |
584 | { | |
585 | unsigned int end = offset + nr_entries - 1; | |
586 | ||
587 | if (offset == si->lowest_bit) | |
588 | si->lowest_bit += nr_entries; | |
589 | if (end == si->highest_bit) | |
590 | si->highest_bit -= nr_entries; | |
591 | si->inuse_pages += nr_entries; | |
592 | if (si->inuse_pages == si->pages) { | |
593 | si->lowest_bit = si->max; | |
594 | si->highest_bit = 0; | |
595 | spin_lock(&swap_avail_lock); | |
596 | plist_del(&si->avail_list, &swap_avail_head); | |
597 | spin_unlock(&swap_avail_lock); | |
598 | } | |
599 | } | |
600 | ||
601 | static void swap_range_free(struct swap_info_struct *si, unsigned long offset, | |
602 | unsigned int nr_entries) | |
603 | { | |
604 | unsigned long end = offset + nr_entries - 1; | |
605 | void (*swap_slot_free_notify)(struct block_device *, unsigned long); | |
606 | ||
607 | if (offset < si->lowest_bit) | |
608 | si->lowest_bit = offset; | |
609 | if (end > si->highest_bit) { | |
610 | bool was_full = !si->highest_bit; | |
611 | ||
612 | si->highest_bit = end; | |
613 | if (was_full && (si->flags & SWP_WRITEOK)) { | |
614 | spin_lock(&swap_avail_lock); | |
615 | WARN_ON(!plist_node_empty(&si->avail_list)); | |
616 | if (plist_node_empty(&si->avail_list)) | |
617 | plist_add(&si->avail_list, &swap_avail_head); | |
618 | spin_unlock(&swap_avail_lock); | |
619 | } | |
620 | } | |
621 | atomic_long_add(nr_entries, &nr_swap_pages); | |
622 | si->inuse_pages -= nr_entries; | |
623 | if (si->flags & SWP_BLKDEV) | |
624 | swap_slot_free_notify = | |
625 | si->bdev->bd_disk->fops->swap_slot_free_notify; | |
626 | else | |
627 | swap_slot_free_notify = NULL; | |
628 | while (offset <= end) { | |
629 | frontswap_invalidate_page(si->type, offset); | |
630 | if (swap_slot_free_notify) | |
631 | swap_slot_free_notify(si->bdev, offset); | |
632 | offset++; | |
633 | } | |
634 | } | |
635 | ||
36005bae TC |
636 | static int scan_swap_map_slots(struct swap_info_struct *si, |
637 | unsigned char usage, int nr, | |
638 | swp_entry_t slots[]) | |
1da177e4 | 639 | { |
235b6217 | 640 | struct swap_cluster_info *ci; |
ebebbbe9 | 641 | unsigned long offset; |
c60aa176 | 642 | unsigned long scan_base; |
7992fde7 | 643 | unsigned long last_in_cluster = 0; |
048c27fd | 644 | int latency_ration = LATENCY_LIMIT; |
36005bae TC |
645 | int n_ret = 0; |
646 | ||
647 | if (nr > SWAP_BATCH) | |
648 | nr = SWAP_BATCH; | |
7dfad418 | 649 | |
886bb7e9 | 650 | /* |
7dfad418 HD |
651 | * We try to cluster swap pages by allocating them sequentially |
652 | * in swap. Once we've allocated SWAPFILE_CLUSTER pages this | |
653 | * way, however, we resort to first-free allocation, starting | |
654 | * a new cluster. This prevents us from scattering swap pages | |
655 | * all over the entire swap partition, so that we reduce | |
656 | * overall disk seek times between swap pages. -- sct | |
657 | * But we do now try to find an empty cluster. -Andrea | |
c60aa176 | 658 | * And we let swap pages go all over an SSD partition. Hugh |
7dfad418 HD |
659 | */ |
660 | ||
52b7efdb | 661 | si->flags += SWP_SCANNING; |
c60aa176 | 662 | scan_base = offset = si->cluster_next; |
ebebbbe9 | 663 | |
ebc2a1a6 SL |
664 | /* SSD algorithm */ |
665 | if (si->cluster_info) { | |
36005bae TC |
666 | if (scan_swap_map_try_ssd_cluster(si, &offset, &scan_base)) |
667 | goto checks; | |
668 | else | |
669 | goto scan; | |
ebc2a1a6 SL |
670 | } |
671 | ||
ebebbbe9 HD |
672 | if (unlikely(!si->cluster_nr--)) { |
673 | if (si->pages - si->inuse_pages < SWAPFILE_CLUSTER) { | |
674 | si->cluster_nr = SWAPFILE_CLUSTER - 1; | |
675 | goto checks; | |
676 | } | |
2a8f9449 | 677 | |
ec8acf20 | 678 | spin_unlock(&si->lock); |
7dfad418 | 679 | |
c60aa176 HD |
680 | /* |
681 | * If seek is expensive, start searching for new cluster from | |
682 | * start of partition, to minimize the span of allocated swap. | |
50088c44 CY |
683 | * If seek is cheap, that is the SWP_SOLIDSTATE si->cluster_info |
684 | * case, just handled by scan_swap_map_try_ssd_cluster() above. | |
c60aa176 | 685 | */ |
50088c44 | 686 | scan_base = offset = si->lowest_bit; |
7dfad418 HD |
687 | last_in_cluster = offset + SWAPFILE_CLUSTER - 1; |
688 | ||
689 | /* Locate the first empty (unaligned) cluster */ | |
690 | for (; last_in_cluster <= si->highest_bit; offset++) { | |
1da177e4 | 691 | if (si->swap_map[offset]) |
7dfad418 HD |
692 | last_in_cluster = offset + SWAPFILE_CLUSTER; |
693 | else if (offset == last_in_cluster) { | |
ec8acf20 | 694 | spin_lock(&si->lock); |
ebebbbe9 HD |
695 | offset -= SWAPFILE_CLUSTER - 1; |
696 | si->cluster_next = offset; | |
697 | si->cluster_nr = SWAPFILE_CLUSTER - 1; | |
c60aa176 HD |
698 | goto checks; |
699 | } | |
700 | if (unlikely(--latency_ration < 0)) { | |
701 | cond_resched(); | |
702 | latency_ration = LATENCY_LIMIT; | |
703 | } | |
704 | } | |
705 | ||
706 | offset = scan_base; | |
ec8acf20 | 707 | spin_lock(&si->lock); |
ebebbbe9 | 708 | si->cluster_nr = SWAPFILE_CLUSTER - 1; |
1da177e4 | 709 | } |
7dfad418 | 710 | |
ebebbbe9 | 711 | checks: |
ebc2a1a6 | 712 | if (si->cluster_info) { |
36005bae TC |
713 | while (scan_swap_map_ssd_cluster_conflict(si, offset)) { |
714 | /* take a break if we already got some slots */ | |
715 | if (n_ret) | |
716 | goto done; | |
717 | if (!scan_swap_map_try_ssd_cluster(si, &offset, | |
718 | &scan_base)) | |
719 | goto scan; | |
720 | } | |
ebc2a1a6 | 721 | } |
ebebbbe9 | 722 | if (!(si->flags & SWP_WRITEOK)) |
52b7efdb | 723 | goto no_page; |
7dfad418 HD |
724 | if (!si->highest_bit) |
725 | goto no_page; | |
ebebbbe9 | 726 | if (offset > si->highest_bit) |
c60aa176 | 727 | scan_base = offset = si->lowest_bit; |
c9e44410 | 728 | |
235b6217 | 729 | ci = lock_cluster(si, offset); |
b73d7fce HD |
730 | /* reuse swap entry of cache-only swap if not busy. */ |
731 | if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) { | |
c9e44410 | 732 | int swap_was_freed; |
235b6217 | 733 | unlock_cluster(ci); |
ec8acf20 | 734 | spin_unlock(&si->lock); |
c9e44410 | 735 | swap_was_freed = __try_to_reclaim_swap(si, offset); |
ec8acf20 | 736 | spin_lock(&si->lock); |
c9e44410 KH |
737 | /* entry was freed successfully, try to use this again */ |
738 | if (swap_was_freed) | |
739 | goto checks; | |
740 | goto scan; /* check next one */ | |
741 | } | |
742 | ||
235b6217 HY |
743 | if (si->swap_map[offset]) { |
744 | unlock_cluster(ci); | |
36005bae TC |
745 | if (!n_ret) |
746 | goto scan; | |
747 | else | |
748 | goto done; | |
235b6217 | 749 | } |
2872bb2d HY |
750 | si->swap_map[offset] = usage; |
751 | inc_cluster_info_page(si, si->cluster_info, offset); | |
752 | unlock_cluster(ci); | |
ebebbbe9 | 753 | |
38d8b4e6 | 754 | swap_range_alloc(si, offset, 1); |
ebebbbe9 | 755 | si->cluster_next = offset + 1; |
36005bae TC |
756 | slots[n_ret++] = swp_entry(si->type, offset); |
757 | ||
758 | /* got enough slots or reach max slots? */ | |
759 | if ((n_ret == nr) || (offset >= si->highest_bit)) | |
760 | goto done; | |
761 | ||
762 | /* search for next available slot */ | |
763 | ||
764 | /* time to take a break? */ | |
765 | if (unlikely(--latency_ration < 0)) { | |
766 | if (n_ret) | |
767 | goto done; | |
768 | spin_unlock(&si->lock); | |
769 | cond_resched(); | |
770 | spin_lock(&si->lock); | |
771 | latency_ration = LATENCY_LIMIT; | |
772 | } | |
773 | ||
774 | /* try to get more slots in cluster */ | |
775 | if (si->cluster_info) { | |
776 | if (scan_swap_map_try_ssd_cluster(si, &offset, &scan_base)) | |
777 | goto checks; | |
778 | else | |
779 | goto done; | |
780 | } | |
781 | /* non-ssd case */ | |
782 | ++offset; | |
783 | ||
784 | /* non-ssd case, still more slots in cluster? */ | |
785 | if (si->cluster_nr && !si->swap_map[offset]) { | |
786 | --si->cluster_nr; | |
787 | goto checks; | |
788 | } | |
7992fde7 | 789 | |
36005bae TC |
790 | done: |
791 | si->flags -= SWP_SCANNING; | |
792 | return n_ret; | |
7dfad418 | 793 | |
ebebbbe9 | 794 | scan: |
ec8acf20 | 795 | spin_unlock(&si->lock); |
7dfad418 | 796 | while (++offset <= si->highest_bit) { |
52b7efdb | 797 | if (!si->swap_map[offset]) { |
ec8acf20 | 798 | spin_lock(&si->lock); |
52b7efdb HD |
799 | goto checks; |
800 | } | |
c9e44410 | 801 | if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) { |
ec8acf20 | 802 | spin_lock(&si->lock); |
c9e44410 KH |
803 | goto checks; |
804 | } | |
048c27fd HD |
805 | if (unlikely(--latency_ration < 0)) { |
806 | cond_resched(); | |
807 | latency_ration = LATENCY_LIMIT; | |
808 | } | |
7dfad418 | 809 | } |
c60aa176 | 810 | offset = si->lowest_bit; |
a5998061 | 811 | while (offset < scan_base) { |
c60aa176 | 812 | if (!si->swap_map[offset]) { |
ec8acf20 | 813 | spin_lock(&si->lock); |
c60aa176 HD |
814 | goto checks; |
815 | } | |
c9e44410 | 816 | if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) { |
ec8acf20 | 817 | spin_lock(&si->lock); |
c9e44410 KH |
818 | goto checks; |
819 | } | |
c60aa176 HD |
820 | if (unlikely(--latency_ration < 0)) { |
821 | cond_resched(); | |
822 | latency_ration = LATENCY_LIMIT; | |
823 | } | |
a5998061 | 824 | offset++; |
c60aa176 | 825 | } |
ec8acf20 | 826 | spin_lock(&si->lock); |
7dfad418 HD |
827 | |
828 | no_page: | |
52b7efdb | 829 | si->flags -= SWP_SCANNING; |
36005bae | 830 | return n_ret; |
1da177e4 LT |
831 | } |
832 | ||
38d8b4e6 HY |
833 | #ifdef CONFIG_THP_SWAP |
834 | static int swap_alloc_cluster(struct swap_info_struct *si, swp_entry_t *slot) | |
835 | { | |
836 | unsigned long idx; | |
837 | struct swap_cluster_info *ci; | |
838 | unsigned long offset, i; | |
839 | unsigned char *map; | |
840 | ||
841 | if (cluster_list_empty(&si->free_clusters)) | |
842 | return 0; | |
843 | ||
844 | idx = cluster_list_first(&si->free_clusters); | |
845 | offset = idx * SWAPFILE_CLUSTER; | |
846 | ci = lock_cluster(si, offset); | |
847 | alloc_cluster(si, idx); | |
848 | cluster_set_count_flag(ci, SWAPFILE_CLUSTER, 0); | |
849 | ||
850 | map = si->swap_map + offset; | |
851 | for (i = 0; i < SWAPFILE_CLUSTER; i++) | |
852 | map[i] = SWAP_HAS_CACHE; | |
853 | unlock_cluster(ci); | |
854 | swap_range_alloc(si, offset, SWAPFILE_CLUSTER); | |
855 | *slot = swp_entry(si->type, offset); | |
856 | ||
857 | return 1; | |
858 | } | |
859 | ||
860 | static void swap_free_cluster(struct swap_info_struct *si, unsigned long idx) | |
861 | { | |
862 | unsigned long offset = idx * SWAPFILE_CLUSTER; | |
863 | struct swap_cluster_info *ci; | |
864 | ||
865 | ci = lock_cluster(si, offset); | |
866 | cluster_set_count_flag(ci, 0, 0); | |
867 | free_cluster(si, idx); | |
868 | unlock_cluster(ci); | |
869 | swap_range_free(si, offset, SWAPFILE_CLUSTER); | |
870 | } | |
871 | #else | |
872 | static int swap_alloc_cluster(struct swap_info_struct *si, swp_entry_t *slot) | |
873 | { | |
874 | VM_WARN_ON_ONCE(1); | |
875 | return 0; | |
876 | } | |
877 | #endif /* CONFIG_THP_SWAP */ | |
878 | ||
36005bae TC |
879 | static unsigned long scan_swap_map(struct swap_info_struct *si, |
880 | unsigned char usage) | |
881 | { | |
882 | swp_entry_t entry; | |
883 | int n_ret; | |
884 | ||
885 | n_ret = scan_swap_map_slots(si, usage, 1, &entry); | |
886 | ||
887 | if (n_ret) | |
888 | return swp_offset(entry); | |
889 | else | |
890 | return 0; | |
891 | ||
892 | } | |
893 | ||
38d8b4e6 | 894 | int get_swap_pages(int n_goal, bool cluster, swp_entry_t swp_entries[]) |
1da177e4 | 895 | { |
38d8b4e6 | 896 | unsigned long nr_pages = cluster ? SWAPFILE_CLUSTER : 1; |
adfab836 | 897 | struct swap_info_struct *si, *next; |
36005bae TC |
898 | long avail_pgs; |
899 | int n_ret = 0; | |
1da177e4 | 900 | |
38d8b4e6 HY |
901 | /* Only single cluster request supported */ |
902 | WARN_ON_ONCE(n_goal > 1 && cluster); | |
903 | ||
904 | avail_pgs = atomic_long_read(&nr_swap_pages) / nr_pages; | |
36005bae | 905 | if (avail_pgs <= 0) |
fb4f88dc | 906 | goto noswap; |
36005bae TC |
907 | |
908 | if (n_goal > SWAP_BATCH) | |
909 | n_goal = SWAP_BATCH; | |
910 | ||
911 | if (n_goal > avail_pgs) | |
912 | n_goal = avail_pgs; | |
913 | ||
38d8b4e6 | 914 | atomic_long_sub(n_goal * nr_pages, &nr_swap_pages); |
fb4f88dc | 915 | |
18ab4d4c DS |
916 | spin_lock(&swap_avail_lock); |
917 | ||
918 | start_over: | |
919 | plist_for_each_entry_safe(si, next, &swap_avail_head, avail_list) { | |
920 | /* requeue si to after same-priority siblings */ | |
921 | plist_requeue(&si->avail_list, &swap_avail_head); | |
922 | spin_unlock(&swap_avail_lock); | |
ec8acf20 | 923 | spin_lock(&si->lock); |
adfab836 | 924 | if (!si->highest_bit || !(si->flags & SWP_WRITEOK)) { |
18ab4d4c DS |
925 | spin_lock(&swap_avail_lock); |
926 | if (plist_node_empty(&si->avail_list)) { | |
927 | spin_unlock(&si->lock); | |
928 | goto nextsi; | |
929 | } | |
930 | WARN(!si->highest_bit, | |
931 | "swap_info %d in list but !highest_bit\n", | |
932 | si->type); | |
933 | WARN(!(si->flags & SWP_WRITEOK), | |
934 | "swap_info %d in list but !SWP_WRITEOK\n", | |
935 | si->type); | |
936 | plist_del(&si->avail_list, &swap_avail_head); | |
ec8acf20 | 937 | spin_unlock(&si->lock); |
18ab4d4c | 938 | goto nextsi; |
ec8acf20 | 939 | } |
38d8b4e6 HY |
940 | if (cluster) |
941 | n_ret = swap_alloc_cluster(si, swp_entries); | |
942 | else | |
943 | n_ret = scan_swap_map_slots(si, SWAP_HAS_CACHE, | |
944 | n_goal, swp_entries); | |
ec8acf20 | 945 | spin_unlock(&si->lock); |
38d8b4e6 | 946 | if (n_ret || cluster) |
36005bae | 947 | goto check_out; |
18ab4d4c | 948 | pr_debug("scan_swap_map of si %d failed to find offset\n", |
36005bae TC |
949 | si->type); |
950 | ||
18ab4d4c DS |
951 | spin_lock(&swap_avail_lock); |
952 | nextsi: | |
adfab836 DS |
953 | /* |
954 | * if we got here, it's likely that si was almost full before, | |
955 | * and since scan_swap_map() can drop the si->lock, multiple | |
956 | * callers probably all tried to get a page from the same si | |
18ab4d4c DS |
957 | * and it filled up before we could get one; or, the si filled |
958 | * up between us dropping swap_avail_lock and taking si->lock. | |
959 | * Since we dropped the swap_avail_lock, the swap_avail_head | |
960 | * list may have been modified; so if next is still in the | |
36005bae TC |
961 | * swap_avail_head list then try it, otherwise start over |
962 | * if we have not gotten any slots. | |
adfab836 | 963 | */ |
18ab4d4c DS |
964 | if (plist_node_empty(&next->avail_list)) |
965 | goto start_over; | |
1da177e4 | 966 | } |
fb4f88dc | 967 | |
18ab4d4c DS |
968 | spin_unlock(&swap_avail_lock); |
969 | ||
36005bae TC |
970 | check_out: |
971 | if (n_ret < n_goal) | |
38d8b4e6 HY |
972 | atomic_long_add((long)(n_goal - n_ret) * nr_pages, |
973 | &nr_swap_pages); | |
fb4f88dc | 974 | noswap: |
36005bae TC |
975 | return n_ret; |
976 | } | |
977 | ||
2de1a7e4 | 978 | /* The only caller of this function is now suspend routine */ |
910321ea HD |
979 | swp_entry_t get_swap_page_of_type(int type) |
980 | { | |
981 | struct swap_info_struct *si; | |
982 | pgoff_t offset; | |
983 | ||
910321ea | 984 | si = swap_info[type]; |
ec8acf20 | 985 | spin_lock(&si->lock); |
910321ea | 986 | if (si && (si->flags & SWP_WRITEOK)) { |
ec8acf20 | 987 | atomic_long_dec(&nr_swap_pages); |
910321ea HD |
988 | /* This is called for allocating swap entry, not cache */ |
989 | offset = scan_swap_map(si, 1); | |
990 | if (offset) { | |
ec8acf20 | 991 | spin_unlock(&si->lock); |
910321ea HD |
992 | return swp_entry(type, offset); |
993 | } | |
ec8acf20 | 994 | atomic_long_inc(&nr_swap_pages); |
910321ea | 995 | } |
ec8acf20 | 996 | spin_unlock(&si->lock); |
910321ea HD |
997 | return (swp_entry_t) {0}; |
998 | } | |
999 | ||
e8c26ab6 | 1000 | static struct swap_info_struct *__swap_info_get(swp_entry_t entry) |
1da177e4 | 1001 | { |
73c34b6a | 1002 | struct swap_info_struct *p; |
1da177e4 LT |
1003 | unsigned long offset, type; |
1004 | ||
1005 | if (!entry.val) | |
1006 | goto out; | |
1007 | type = swp_type(entry); | |
1008 | if (type >= nr_swapfiles) | |
1009 | goto bad_nofile; | |
efa90a98 | 1010 | p = swap_info[type]; |
1da177e4 LT |
1011 | if (!(p->flags & SWP_USED)) |
1012 | goto bad_device; | |
1013 | offset = swp_offset(entry); | |
1014 | if (offset >= p->max) | |
1015 | goto bad_offset; | |
1da177e4 LT |
1016 | return p; |
1017 | ||
1da177e4 | 1018 | bad_offset: |
6a991fc7 | 1019 | pr_err("swap_info_get: %s%08lx\n", Bad_offset, entry.val); |
1da177e4 LT |
1020 | goto out; |
1021 | bad_device: | |
6a991fc7 | 1022 | pr_err("swap_info_get: %s%08lx\n", Unused_file, entry.val); |
1da177e4 LT |
1023 | goto out; |
1024 | bad_nofile: | |
6a991fc7 | 1025 | pr_err("swap_info_get: %s%08lx\n", Bad_file, entry.val); |
1da177e4 LT |
1026 | out: |
1027 | return NULL; | |
886bb7e9 | 1028 | } |
1da177e4 | 1029 | |
e8c26ab6 TC |
1030 | static struct swap_info_struct *_swap_info_get(swp_entry_t entry) |
1031 | { | |
1032 | struct swap_info_struct *p; | |
1033 | ||
1034 | p = __swap_info_get(entry); | |
1035 | if (!p) | |
1036 | goto out; | |
1037 | if (!p->swap_map[swp_offset(entry)]) | |
1038 | goto bad_free; | |
1039 | return p; | |
1040 | ||
1041 | bad_free: | |
1042 | pr_err("swap_info_get: %s%08lx\n", Unused_offset, entry.val); | |
1043 | goto out; | |
1044 | out: | |
1045 | return NULL; | |
1046 | } | |
1047 | ||
235b6217 HY |
1048 | static struct swap_info_struct *swap_info_get(swp_entry_t entry) |
1049 | { | |
1050 | struct swap_info_struct *p; | |
1051 | ||
1052 | p = _swap_info_get(entry); | |
1053 | if (p) | |
1054 | spin_lock(&p->lock); | |
1055 | return p; | |
1056 | } | |
1057 | ||
7c00bafe TC |
1058 | static struct swap_info_struct *swap_info_get_cont(swp_entry_t entry, |
1059 | struct swap_info_struct *q) | |
1060 | { | |
1061 | struct swap_info_struct *p; | |
1062 | ||
1063 | p = _swap_info_get(entry); | |
1064 | ||
1065 | if (p != q) { | |
1066 | if (q != NULL) | |
1067 | spin_unlock(&q->lock); | |
1068 | if (p != NULL) | |
1069 | spin_lock(&p->lock); | |
1070 | } | |
1071 | return p; | |
1072 | } | |
1073 | ||
1074 | static unsigned char __swap_entry_free(struct swap_info_struct *p, | |
1075 | swp_entry_t entry, unsigned char usage) | |
1da177e4 | 1076 | { |
235b6217 | 1077 | struct swap_cluster_info *ci; |
253d553b | 1078 | unsigned long offset = swp_offset(entry); |
8d69aaee HD |
1079 | unsigned char count; |
1080 | unsigned char has_cache; | |
235b6217 | 1081 | |
7c00bafe | 1082 | ci = lock_cluster_or_swap_info(p, offset); |
355cfa73 | 1083 | |
253d553b | 1084 | count = p->swap_map[offset]; |
235b6217 | 1085 | |
253d553b HD |
1086 | has_cache = count & SWAP_HAS_CACHE; |
1087 | count &= ~SWAP_HAS_CACHE; | |
355cfa73 | 1088 | |
253d553b | 1089 | if (usage == SWAP_HAS_CACHE) { |
355cfa73 | 1090 | VM_BUG_ON(!has_cache); |
253d553b | 1091 | has_cache = 0; |
aaa46865 HD |
1092 | } else if (count == SWAP_MAP_SHMEM) { |
1093 | /* | |
1094 | * Or we could insist on shmem.c using a special | |
1095 | * swap_shmem_free() and free_shmem_swap_and_cache()... | |
1096 | */ | |
1097 | count = 0; | |
570a335b HD |
1098 | } else if ((count & ~COUNT_CONTINUED) <= SWAP_MAP_MAX) { |
1099 | if (count == COUNT_CONTINUED) { | |
1100 | if (swap_count_continued(p, offset, count)) | |
1101 | count = SWAP_MAP_MAX | COUNT_CONTINUED; | |
1102 | else | |
1103 | count = SWAP_MAP_MAX; | |
1104 | } else | |
1105 | count--; | |
1106 | } | |
253d553b | 1107 | |
253d553b | 1108 | usage = count | has_cache; |
7c00bafe TC |
1109 | p->swap_map[offset] = usage ? : SWAP_HAS_CACHE; |
1110 | ||
1111 | unlock_cluster_or_swap_info(p, ci); | |
1112 | ||
1113 | return usage; | |
1114 | } | |
355cfa73 | 1115 | |
7c00bafe TC |
1116 | static void swap_entry_free(struct swap_info_struct *p, swp_entry_t entry) |
1117 | { | |
1118 | struct swap_cluster_info *ci; | |
1119 | unsigned long offset = swp_offset(entry); | |
1120 | unsigned char count; | |
1121 | ||
1122 | ci = lock_cluster(p, offset); | |
1123 | count = p->swap_map[offset]; | |
1124 | VM_BUG_ON(count != SWAP_HAS_CACHE); | |
1125 | p->swap_map[offset] = 0; | |
1126 | dec_cluster_info_page(p, p->cluster_info, offset); | |
235b6217 HY |
1127 | unlock_cluster(ci); |
1128 | ||
38d8b4e6 HY |
1129 | mem_cgroup_uncharge_swap(entry, 1); |
1130 | swap_range_free(p, offset, 1); | |
1da177e4 LT |
1131 | } |
1132 | ||
1133 | /* | |
2de1a7e4 | 1134 | * Caller has made sure that the swap device corresponding to entry |
1da177e4 LT |
1135 | * is still around or has not been recycled. |
1136 | */ | |
1137 | void swap_free(swp_entry_t entry) | |
1138 | { | |
73c34b6a | 1139 | struct swap_info_struct *p; |
1da177e4 | 1140 | |
235b6217 | 1141 | p = _swap_info_get(entry); |
7c00bafe TC |
1142 | if (p) { |
1143 | if (!__swap_entry_free(p, entry, 1)) | |
67afa38e | 1144 | free_swap_slot(entry); |
7c00bafe | 1145 | } |
1da177e4 LT |
1146 | } |
1147 | ||
cb4b86ba KH |
1148 | /* |
1149 | * Called after dropping swapcache to decrease refcnt to swap entries. | |
1150 | */ | |
75f6d6d2 | 1151 | static void swapcache_free(swp_entry_t entry) |
cb4b86ba | 1152 | { |
355cfa73 KH |
1153 | struct swap_info_struct *p; |
1154 | ||
235b6217 | 1155 | p = _swap_info_get(entry); |
7c00bafe TC |
1156 | if (p) { |
1157 | if (!__swap_entry_free(p, entry, SWAP_HAS_CACHE)) | |
67afa38e | 1158 | free_swap_slot(entry); |
7c00bafe TC |
1159 | } |
1160 | } | |
1161 | ||
38d8b4e6 | 1162 | #ifdef CONFIG_THP_SWAP |
75f6d6d2 | 1163 | static void swapcache_free_cluster(swp_entry_t entry) |
38d8b4e6 HY |
1164 | { |
1165 | unsigned long offset = swp_offset(entry); | |
1166 | unsigned long idx = offset / SWAPFILE_CLUSTER; | |
1167 | struct swap_cluster_info *ci; | |
1168 | struct swap_info_struct *si; | |
1169 | unsigned char *map; | |
1170 | unsigned int i; | |
1171 | ||
1172 | si = swap_info_get(entry); | |
1173 | if (!si) | |
1174 | return; | |
1175 | ||
1176 | ci = lock_cluster(si, offset); | |
1177 | map = si->swap_map + offset; | |
1178 | for (i = 0; i < SWAPFILE_CLUSTER; i++) { | |
1179 | VM_BUG_ON(map[i] != SWAP_HAS_CACHE); | |
1180 | map[i] = 0; | |
1181 | } | |
1182 | unlock_cluster(ci); | |
1183 | mem_cgroup_uncharge_swap(entry, SWAPFILE_CLUSTER); | |
1184 | swap_free_cluster(si, idx); | |
1185 | spin_unlock(&si->lock); | |
1186 | } | |
75f6d6d2 MK |
1187 | #else |
1188 | static inline void swapcache_free_cluster(swp_entry_t entry) | |
1189 | { | |
1190 | } | |
38d8b4e6 HY |
1191 | #endif /* CONFIG_THP_SWAP */ |
1192 | ||
75f6d6d2 MK |
1193 | void put_swap_page(struct page *page, swp_entry_t entry) |
1194 | { | |
1195 | if (!PageTransHuge(page)) | |
1196 | swapcache_free(entry); | |
1197 | else | |
1198 | swapcache_free_cluster(entry); | |
1199 | } | |
1200 | ||
7c00bafe TC |
1201 | void swapcache_free_entries(swp_entry_t *entries, int n) |
1202 | { | |
1203 | struct swap_info_struct *p, *prev; | |
1204 | int i; | |
1205 | ||
1206 | if (n <= 0) | |
1207 | return; | |
1208 | ||
1209 | prev = NULL; | |
1210 | p = NULL; | |
1211 | for (i = 0; i < n; ++i) { | |
1212 | p = swap_info_get_cont(entries[i], prev); | |
1213 | if (p) | |
1214 | swap_entry_free(p, entries[i]); | |
7c00bafe TC |
1215 | prev = p; |
1216 | } | |
235b6217 | 1217 | if (p) |
7c00bafe | 1218 | spin_unlock(&p->lock); |
cb4b86ba KH |
1219 | } |
1220 | ||
1da177e4 | 1221 | /* |
c475a8ab | 1222 | * How many references to page are currently swapped out? |
570a335b HD |
1223 | * This does not give an exact answer when swap count is continued, |
1224 | * but does include the high COUNT_CONTINUED flag to allow for that. | |
1da177e4 | 1225 | */ |
bde05d1c | 1226 | int page_swapcount(struct page *page) |
1da177e4 | 1227 | { |
c475a8ab HD |
1228 | int count = 0; |
1229 | struct swap_info_struct *p; | |
235b6217 | 1230 | struct swap_cluster_info *ci; |
1da177e4 | 1231 | swp_entry_t entry; |
235b6217 | 1232 | unsigned long offset; |
1da177e4 | 1233 | |
4c21e2f2 | 1234 | entry.val = page_private(page); |
235b6217 | 1235 | p = _swap_info_get(entry); |
1da177e4 | 1236 | if (p) { |
235b6217 HY |
1237 | offset = swp_offset(entry); |
1238 | ci = lock_cluster_or_swap_info(p, offset); | |
1239 | count = swap_count(p->swap_map[offset]); | |
1240 | unlock_cluster_or_swap_info(p, ci); | |
1da177e4 | 1241 | } |
c475a8ab | 1242 | return count; |
1da177e4 LT |
1243 | } |
1244 | ||
322b8afe HY |
1245 | static int swap_swapcount(struct swap_info_struct *si, swp_entry_t entry) |
1246 | { | |
1247 | int count = 0; | |
1248 | pgoff_t offset = swp_offset(entry); | |
1249 | struct swap_cluster_info *ci; | |
1250 | ||
1251 | ci = lock_cluster_or_swap_info(si, offset); | |
1252 | count = swap_count(si->swap_map[offset]); | |
1253 | unlock_cluster_or_swap_info(si, ci); | |
1254 | return count; | |
1255 | } | |
1256 | ||
e8c26ab6 TC |
1257 | /* |
1258 | * How many references to @entry are currently swapped out? | |
1259 | * This does not give an exact answer when swap count is continued, | |
1260 | * but does include the high COUNT_CONTINUED flag to allow for that. | |
1261 | */ | |
1262 | int __swp_swapcount(swp_entry_t entry) | |
1263 | { | |
1264 | int count = 0; | |
e8c26ab6 | 1265 | struct swap_info_struct *si; |
e8c26ab6 TC |
1266 | |
1267 | si = __swap_info_get(entry); | |
322b8afe HY |
1268 | if (si) |
1269 | count = swap_swapcount(si, entry); | |
e8c26ab6 TC |
1270 | return count; |
1271 | } | |
1272 | ||
8334b962 MK |
1273 | /* |
1274 | * How many references to @entry are currently swapped out? | |
1275 | * This considers COUNT_CONTINUED so it returns exact answer. | |
1276 | */ | |
1277 | int swp_swapcount(swp_entry_t entry) | |
1278 | { | |
1279 | int count, tmp_count, n; | |
1280 | struct swap_info_struct *p; | |
235b6217 | 1281 | struct swap_cluster_info *ci; |
8334b962 MK |
1282 | struct page *page; |
1283 | pgoff_t offset; | |
1284 | unsigned char *map; | |
1285 | ||
235b6217 | 1286 | p = _swap_info_get(entry); |
8334b962 MK |
1287 | if (!p) |
1288 | return 0; | |
1289 | ||
235b6217 HY |
1290 | offset = swp_offset(entry); |
1291 | ||
1292 | ci = lock_cluster_or_swap_info(p, offset); | |
1293 | ||
1294 | count = swap_count(p->swap_map[offset]); | |
8334b962 MK |
1295 | if (!(count & COUNT_CONTINUED)) |
1296 | goto out; | |
1297 | ||
1298 | count &= ~COUNT_CONTINUED; | |
1299 | n = SWAP_MAP_MAX + 1; | |
1300 | ||
8334b962 MK |
1301 | page = vmalloc_to_page(p->swap_map + offset); |
1302 | offset &= ~PAGE_MASK; | |
1303 | VM_BUG_ON(page_private(page) != SWP_CONTINUED); | |
1304 | ||
1305 | do { | |
a8ae4991 | 1306 | page = list_next_entry(page, lru); |
8334b962 MK |
1307 | map = kmap_atomic(page); |
1308 | tmp_count = map[offset]; | |
1309 | kunmap_atomic(map); | |
1310 | ||
1311 | count += (tmp_count & ~COUNT_CONTINUED) * n; | |
1312 | n *= (SWAP_CONT_MAX + 1); | |
1313 | } while (tmp_count & COUNT_CONTINUED); | |
1314 | out: | |
235b6217 | 1315 | unlock_cluster_or_swap_info(p, ci); |
8334b962 MK |
1316 | return count; |
1317 | } | |
1318 | ||
1da177e4 | 1319 | /* |
7b1fe597 HD |
1320 | * We can write to an anon page without COW if there are no other references |
1321 | * to it. And as a side-effect, free up its swap: because the old content | |
1322 | * on disk will never be read, and seeking back there to write new content | |
1323 | * later would only waste time away from clustering. | |
6d0a07ed AA |
1324 | * |
1325 | * NOTE: total_mapcount should not be relied upon by the caller if | |
1326 | * reuse_swap_page() returns false, but it may be always overwritten | |
1327 | * (see the other implementation for CONFIG_SWAP=n). | |
1da177e4 | 1328 | */ |
6d0a07ed | 1329 | bool reuse_swap_page(struct page *page, int *total_mapcount) |
1da177e4 | 1330 | { |
c475a8ab HD |
1331 | int count; |
1332 | ||
309381fe | 1333 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
5ad64688 | 1334 | if (unlikely(PageKsm(page))) |
6d0a07ed AA |
1335 | return false; |
1336 | count = page_trans_huge_mapcount(page, total_mapcount); | |
7b1fe597 | 1337 | if (count <= 1 && PageSwapCache(page)) { |
c475a8ab | 1338 | count += page_swapcount(page); |
f0571429 MK |
1339 | if (count != 1) |
1340 | goto out; | |
1341 | if (!PageWriteback(page)) { | |
7b1fe597 HD |
1342 | delete_from_swap_cache(page); |
1343 | SetPageDirty(page); | |
f0571429 MK |
1344 | } else { |
1345 | swp_entry_t entry; | |
1346 | struct swap_info_struct *p; | |
1347 | ||
1348 | entry.val = page_private(page); | |
1349 | p = swap_info_get(entry); | |
1350 | if (p->flags & SWP_STABLE_WRITES) { | |
1351 | spin_unlock(&p->lock); | |
1352 | return false; | |
1353 | } | |
1354 | spin_unlock(&p->lock); | |
7b1fe597 HD |
1355 | } |
1356 | } | |
f0571429 | 1357 | out: |
5ad64688 | 1358 | return count <= 1; |
1da177e4 LT |
1359 | } |
1360 | ||
1361 | /* | |
a2c43eed HD |
1362 | * If swap is getting full, or if there are no more mappings of this page, |
1363 | * then try_to_free_swap is called to free its swap space. | |
1da177e4 | 1364 | */ |
a2c43eed | 1365 | int try_to_free_swap(struct page *page) |
1da177e4 | 1366 | { |
309381fe | 1367 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
1da177e4 LT |
1368 | |
1369 | if (!PageSwapCache(page)) | |
1370 | return 0; | |
1371 | if (PageWriteback(page)) | |
1372 | return 0; | |
a2c43eed | 1373 | if (page_swapcount(page)) |
1da177e4 LT |
1374 | return 0; |
1375 | ||
b73d7fce HD |
1376 | /* |
1377 | * Once hibernation has begun to create its image of memory, | |
1378 | * there's a danger that one of the calls to try_to_free_swap() | |
1379 | * - most probably a call from __try_to_reclaim_swap() while | |
1380 | * hibernation is allocating its own swap pages for the image, | |
1381 | * but conceivably even a call from memory reclaim - will free | |
1382 | * the swap from a page which has already been recorded in the | |
1383 | * image as a clean swapcache page, and then reuse its swap for | |
1384 | * another page of the image. On waking from hibernation, the | |
1385 | * original page might be freed under memory pressure, then | |
1386 | * later read back in from swap, now with the wrong data. | |
1387 | * | |
2de1a7e4 | 1388 | * Hibernation suspends storage while it is writing the image |
f90ac398 | 1389 | * to disk so check that here. |
b73d7fce | 1390 | */ |
f90ac398 | 1391 | if (pm_suspended_storage()) |
b73d7fce HD |
1392 | return 0; |
1393 | ||
a2c43eed HD |
1394 | delete_from_swap_cache(page); |
1395 | SetPageDirty(page); | |
1396 | return 1; | |
68a22394 RR |
1397 | } |
1398 | ||
1da177e4 LT |
1399 | /* |
1400 | * Free the swap entry like above, but also try to | |
1401 | * free the page cache entry if it is the last user. | |
1402 | */ | |
2509ef26 | 1403 | int free_swap_and_cache(swp_entry_t entry) |
1da177e4 | 1404 | { |
2509ef26 | 1405 | struct swap_info_struct *p; |
1da177e4 | 1406 | struct page *page = NULL; |
7c00bafe | 1407 | unsigned char count; |
1da177e4 | 1408 | |
a7420aa5 | 1409 | if (non_swap_entry(entry)) |
2509ef26 | 1410 | return 1; |
0697212a | 1411 | |
7c00bafe | 1412 | p = _swap_info_get(entry); |
1da177e4 | 1413 | if (p) { |
7c00bafe TC |
1414 | count = __swap_entry_free(p, entry, 1); |
1415 | if (count == SWAP_HAS_CACHE) { | |
33806f06 | 1416 | page = find_get_page(swap_address_space(entry), |
f6ab1f7f | 1417 | swp_offset(entry)); |
8413ac9d | 1418 | if (page && !trylock_page(page)) { |
09cbfeaf | 1419 | put_page(page); |
93fac704 NP |
1420 | page = NULL; |
1421 | } | |
7c00bafe | 1422 | } else if (!count) |
67afa38e | 1423 | free_swap_slot(entry); |
1da177e4 LT |
1424 | } |
1425 | if (page) { | |
a2c43eed HD |
1426 | /* |
1427 | * Not mapped elsewhere, or swap space full? Free it! | |
1428 | * Also recheck PageSwapCache now page is locked (above). | |
1429 | */ | |
93fac704 | 1430 | if (PageSwapCache(page) && !PageWriteback(page) && |
322b8afe HY |
1431 | (!page_mapped(page) || mem_cgroup_swap_full(page)) && |
1432 | !swap_swapcount(p, entry)) { | |
1da177e4 LT |
1433 | delete_from_swap_cache(page); |
1434 | SetPageDirty(page); | |
1435 | } | |
1436 | unlock_page(page); | |
09cbfeaf | 1437 | put_page(page); |
1da177e4 | 1438 | } |
2509ef26 | 1439 | return p != NULL; |
1da177e4 LT |
1440 | } |
1441 | ||
b0cb1a19 | 1442 | #ifdef CONFIG_HIBERNATION |
f577eb30 | 1443 | /* |
915bae9e | 1444 | * Find the swap type that corresponds to given device (if any). |
f577eb30 | 1445 | * |
915bae9e RW |
1446 | * @offset - number of the PAGE_SIZE-sized block of the device, starting |
1447 | * from 0, in which the swap header is expected to be located. | |
1448 | * | |
1449 | * This is needed for the suspend to disk (aka swsusp). | |
f577eb30 | 1450 | */ |
7bf23687 | 1451 | int swap_type_of(dev_t device, sector_t offset, struct block_device **bdev_p) |
f577eb30 | 1452 | { |
915bae9e | 1453 | struct block_device *bdev = NULL; |
efa90a98 | 1454 | int type; |
f577eb30 | 1455 | |
915bae9e RW |
1456 | if (device) |
1457 | bdev = bdget(device); | |
1458 | ||
f577eb30 | 1459 | spin_lock(&swap_lock); |
efa90a98 HD |
1460 | for (type = 0; type < nr_swapfiles; type++) { |
1461 | struct swap_info_struct *sis = swap_info[type]; | |
f577eb30 | 1462 | |
915bae9e | 1463 | if (!(sis->flags & SWP_WRITEOK)) |
f577eb30 | 1464 | continue; |
b6b5bce3 | 1465 | |
915bae9e | 1466 | if (!bdev) { |
7bf23687 | 1467 | if (bdev_p) |
dddac6a7 | 1468 | *bdev_p = bdgrab(sis->bdev); |
7bf23687 | 1469 | |
6e1819d6 | 1470 | spin_unlock(&swap_lock); |
efa90a98 | 1471 | return type; |
6e1819d6 | 1472 | } |
915bae9e | 1473 | if (bdev == sis->bdev) { |
9625a5f2 | 1474 | struct swap_extent *se = &sis->first_swap_extent; |
915bae9e | 1475 | |
915bae9e | 1476 | if (se->start_block == offset) { |
7bf23687 | 1477 | if (bdev_p) |
dddac6a7 | 1478 | *bdev_p = bdgrab(sis->bdev); |
7bf23687 | 1479 | |
915bae9e RW |
1480 | spin_unlock(&swap_lock); |
1481 | bdput(bdev); | |
efa90a98 | 1482 | return type; |
915bae9e | 1483 | } |
f577eb30 RW |
1484 | } |
1485 | } | |
1486 | spin_unlock(&swap_lock); | |
915bae9e RW |
1487 | if (bdev) |
1488 | bdput(bdev); | |
1489 | ||
f577eb30 RW |
1490 | return -ENODEV; |
1491 | } | |
1492 | ||
73c34b6a HD |
1493 | /* |
1494 | * Get the (PAGE_SIZE) block corresponding to given offset on the swapdev | |
1495 | * corresponding to given index in swap_info (swap type). | |
1496 | */ | |
1497 | sector_t swapdev_block(int type, pgoff_t offset) | |
1498 | { | |
1499 | struct block_device *bdev; | |
1500 | ||
1501 | if ((unsigned int)type >= nr_swapfiles) | |
1502 | return 0; | |
1503 | if (!(swap_info[type]->flags & SWP_WRITEOK)) | |
1504 | return 0; | |
d4906e1a | 1505 | return map_swap_entry(swp_entry(type, offset), &bdev); |
73c34b6a HD |
1506 | } |
1507 | ||
f577eb30 RW |
1508 | /* |
1509 | * Return either the total number of swap pages of given type, or the number | |
1510 | * of free pages of that type (depending on @free) | |
1511 | * | |
1512 | * This is needed for software suspend | |
1513 | */ | |
1514 | unsigned int count_swap_pages(int type, int free) | |
1515 | { | |
1516 | unsigned int n = 0; | |
1517 | ||
efa90a98 HD |
1518 | spin_lock(&swap_lock); |
1519 | if ((unsigned int)type < nr_swapfiles) { | |
1520 | struct swap_info_struct *sis = swap_info[type]; | |
1521 | ||
ec8acf20 | 1522 | spin_lock(&sis->lock); |
efa90a98 HD |
1523 | if (sis->flags & SWP_WRITEOK) { |
1524 | n = sis->pages; | |
f577eb30 | 1525 | if (free) |
efa90a98 | 1526 | n -= sis->inuse_pages; |
f577eb30 | 1527 | } |
ec8acf20 | 1528 | spin_unlock(&sis->lock); |
f577eb30 | 1529 | } |
efa90a98 | 1530 | spin_unlock(&swap_lock); |
f577eb30 RW |
1531 | return n; |
1532 | } | |
73c34b6a | 1533 | #endif /* CONFIG_HIBERNATION */ |
f577eb30 | 1534 | |
9f8bdb3f | 1535 | static inline int pte_same_as_swp(pte_t pte, pte_t swp_pte) |
179ef71c | 1536 | { |
9f8bdb3f | 1537 | return pte_same(pte_swp_clear_soft_dirty(pte), swp_pte); |
179ef71c CG |
1538 | } |
1539 | ||
1da177e4 | 1540 | /* |
72866f6f HD |
1541 | * No need to decide whether this PTE shares the swap entry with others, |
1542 | * just let do_wp_page work it out if a write is requested later - to | |
1543 | * force COW, vm_page_prot omits write permission from any private vma. | |
1da177e4 | 1544 | */ |
044d66c1 | 1545 | static int unuse_pte(struct vm_area_struct *vma, pmd_t *pmd, |
1da177e4 LT |
1546 | unsigned long addr, swp_entry_t entry, struct page *page) |
1547 | { | |
9e16b7fb | 1548 | struct page *swapcache; |
72835c86 | 1549 | struct mem_cgroup *memcg; |
044d66c1 HD |
1550 | spinlock_t *ptl; |
1551 | pte_t *pte; | |
1552 | int ret = 1; | |
1553 | ||
9e16b7fb HD |
1554 | swapcache = page; |
1555 | page = ksm_might_need_to_copy(page, vma, addr); | |
1556 | if (unlikely(!page)) | |
1557 | return -ENOMEM; | |
1558 | ||
f627c2f5 KS |
1559 | if (mem_cgroup_try_charge(page, vma->vm_mm, GFP_KERNEL, |
1560 | &memcg, false)) { | |
044d66c1 | 1561 | ret = -ENOMEM; |
85d9fc89 KH |
1562 | goto out_nolock; |
1563 | } | |
044d66c1 HD |
1564 | |
1565 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); | |
9f8bdb3f | 1566 | if (unlikely(!pte_same_as_swp(*pte, swp_entry_to_pte(entry)))) { |
f627c2f5 | 1567 | mem_cgroup_cancel_charge(page, memcg, false); |
044d66c1 HD |
1568 | ret = 0; |
1569 | goto out; | |
1570 | } | |
8a9f3ccd | 1571 | |
b084d435 | 1572 | dec_mm_counter(vma->vm_mm, MM_SWAPENTS); |
d559db08 | 1573 | inc_mm_counter(vma->vm_mm, MM_ANONPAGES); |
1da177e4 LT |
1574 | get_page(page); |
1575 | set_pte_at(vma->vm_mm, addr, pte, | |
1576 | pte_mkold(mk_pte(page, vma->vm_page_prot))); | |
00501b53 | 1577 | if (page == swapcache) { |
d281ee61 | 1578 | page_add_anon_rmap(page, vma, addr, false); |
f627c2f5 | 1579 | mem_cgroup_commit_charge(page, memcg, true, false); |
00501b53 | 1580 | } else { /* ksm created a completely new copy */ |
d281ee61 | 1581 | page_add_new_anon_rmap(page, vma, addr, false); |
f627c2f5 | 1582 | mem_cgroup_commit_charge(page, memcg, false, false); |
00501b53 JW |
1583 | lru_cache_add_active_or_unevictable(page, vma); |
1584 | } | |
1da177e4 LT |
1585 | swap_free(entry); |
1586 | /* | |
1587 | * Move the page to the active list so it is not | |
1588 | * immediately swapped out again after swapon. | |
1589 | */ | |
1590 | activate_page(page); | |
044d66c1 HD |
1591 | out: |
1592 | pte_unmap_unlock(pte, ptl); | |
85d9fc89 | 1593 | out_nolock: |
9e16b7fb HD |
1594 | if (page != swapcache) { |
1595 | unlock_page(page); | |
1596 | put_page(page); | |
1597 | } | |
044d66c1 | 1598 | return ret; |
1da177e4 LT |
1599 | } |
1600 | ||
1601 | static int unuse_pte_range(struct vm_area_struct *vma, pmd_t *pmd, | |
1602 | unsigned long addr, unsigned long end, | |
1603 | swp_entry_t entry, struct page *page) | |
1604 | { | |
1da177e4 | 1605 | pte_t swp_pte = swp_entry_to_pte(entry); |
705e87c0 | 1606 | pte_t *pte; |
8a9f3ccd | 1607 | int ret = 0; |
1da177e4 | 1608 | |
044d66c1 HD |
1609 | /* |
1610 | * We don't actually need pte lock while scanning for swp_pte: since | |
1611 | * we hold page lock and mmap_sem, swp_pte cannot be inserted into the | |
1612 | * page table while we're scanning; though it could get zapped, and on | |
1613 | * some architectures (e.g. x86_32 with PAE) we might catch a glimpse | |
1614 | * of unmatched parts which look like swp_pte, so unuse_pte must | |
1615 | * recheck under pte lock. Scanning without pte lock lets it be | |
2de1a7e4 | 1616 | * preemptable whenever CONFIG_PREEMPT but not CONFIG_HIGHPTE. |
044d66c1 HD |
1617 | */ |
1618 | pte = pte_offset_map(pmd, addr); | |
1da177e4 LT |
1619 | do { |
1620 | /* | |
1621 | * swapoff spends a _lot_ of time in this loop! | |
1622 | * Test inline before going to call unuse_pte. | |
1623 | */ | |
9f8bdb3f | 1624 | if (unlikely(pte_same_as_swp(*pte, swp_pte))) { |
044d66c1 HD |
1625 | pte_unmap(pte); |
1626 | ret = unuse_pte(vma, pmd, addr, entry, page); | |
1627 | if (ret) | |
1628 | goto out; | |
1629 | pte = pte_offset_map(pmd, addr); | |
1da177e4 LT |
1630 | } |
1631 | } while (pte++, addr += PAGE_SIZE, addr != end); | |
044d66c1 HD |
1632 | pte_unmap(pte - 1); |
1633 | out: | |
8a9f3ccd | 1634 | return ret; |
1da177e4 LT |
1635 | } |
1636 | ||
1637 | static inline int unuse_pmd_range(struct vm_area_struct *vma, pud_t *pud, | |
1638 | unsigned long addr, unsigned long end, | |
1639 | swp_entry_t entry, struct page *page) | |
1640 | { | |
1641 | pmd_t *pmd; | |
1642 | unsigned long next; | |
8a9f3ccd | 1643 | int ret; |
1da177e4 LT |
1644 | |
1645 | pmd = pmd_offset(pud, addr); | |
1646 | do { | |
dc644a07 | 1647 | cond_resched(); |
1da177e4 | 1648 | next = pmd_addr_end(addr, end); |
1a5a9906 | 1649 | if (pmd_none_or_trans_huge_or_clear_bad(pmd)) |
1da177e4 | 1650 | continue; |
8a9f3ccd BS |
1651 | ret = unuse_pte_range(vma, pmd, addr, next, entry, page); |
1652 | if (ret) | |
1653 | return ret; | |
1da177e4 LT |
1654 | } while (pmd++, addr = next, addr != end); |
1655 | return 0; | |
1656 | } | |
1657 | ||
c2febafc | 1658 | static inline int unuse_pud_range(struct vm_area_struct *vma, p4d_t *p4d, |
1da177e4 LT |
1659 | unsigned long addr, unsigned long end, |
1660 | swp_entry_t entry, struct page *page) | |
1661 | { | |
1662 | pud_t *pud; | |
1663 | unsigned long next; | |
8a9f3ccd | 1664 | int ret; |
1da177e4 | 1665 | |
c2febafc | 1666 | pud = pud_offset(p4d, addr); |
1da177e4 LT |
1667 | do { |
1668 | next = pud_addr_end(addr, end); | |
1669 | if (pud_none_or_clear_bad(pud)) | |
1670 | continue; | |
8a9f3ccd BS |
1671 | ret = unuse_pmd_range(vma, pud, addr, next, entry, page); |
1672 | if (ret) | |
1673 | return ret; | |
1da177e4 LT |
1674 | } while (pud++, addr = next, addr != end); |
1675 | return 0; | |
1676 | } | |
1677 | ||
c2febafc KS |
1678 | static inline int unuse_p4d_range(struct vm_area_struct *vma, pgd_t *pgd, |
1679 | unsigned long addr, unsigned long end, | |
1680 | swp_entry_t entry, struct page *page) | |
1681 | { | |
1682 | p4d_t *p4d; | |
1683 | unsigned long next; | |
1684 | int ret; | |
1685 | ||
1686 | p4d = p4d_offset(pgd, addr); | |
1687 | do { | |
1688 | next = p4d_addr_end(addr, end); | |
1689 | if (p4d_none_or_clear_bad(p4d)) | |
1690 | continue; | |
1691 | ret = unuse_pud_range(vma, p4d, addr, next, entry, page); | |
1692 | if (ret) | |
1693 | return ret; | |
1694 | } while (p4d++, addr = next, addr != end); | |
1695 | return 0; | |
1696 | } | |
1697 | ||
1da177e4 LT |
1698 | static int unuse_vma(struct vm_area_struct *vma, |
1699 | swp_entry_t entry, struct page *page) | |
1700 | { | |
1701 | pgd_t *pgd; | |
1702 | unsigned long addr, end, next; | |
8a9f3ccd | 1703 | int ret; |
1da177e4 | 1704 | |
3ca7b3c5 | 1705 | if (page_anon_vma(page)) { |
1da177e4 LT |
1706 | addr = page_address_in_vma(page, vma); |
1707 | if (addr == -EFAULT) | |
1708 | return 0; | |
1709 | else | |
1710 | end = addr + PAGE_SIZE; | |
1711 | } else { | |
1712 | addr = vma->vm_start; | |
1713 | end = vma->vm_end; | |
1714 | } | |
1715 | ||
1716 | pgd = pgd_offset(vma->vm_mm, addr); | |
1717 | do { | |
1718 | next = pgd_addr_end(addr, end); | |
1719 | if (pgd_none_or_clear_bad(pgd)) | |
1720 | continue; | |
c2febafc | 1721 | ret = unuse_p4d_range(vma, pgd, addr, next, entry, page); |
8a9f3ccd BS |
1722 | if (ret) |
1723 | return ret; | |
1da177e4 LT |
1724 | } while (pgd++, addr = next, addr != end); |
1725 | return 0; | |
1726 | } | |
1727 | ||
1728 | static int unuse_mm(struct mm_struct *mm, | |
1729 | swp_entry_t entry, struct page *page) | |
1730 | { | |
1731 | struct vm_area_struct *vma; | |
8a9f3ccd | 1732 | int ret = 0; |
1da177e4 LT |
1733 | |
1734 | if (!down_read_trylock(&mm->mmap_sem)) { | |
1735 | /* | |
7d03431c FLVC |
1736 | * Activate page so shrink_inactive_list is unlikely to unmap |
1737 | * its ptes while lock is dropped, so swapoff can make progress. | |
1da177e4 | 1738 | */ |
c475a8ab | 1739 | activate_page(page); |
1da177e4 LT |
1740 | unlock_page(page); |
1741 | down_read(&mm->mmap_sem); | |
1742 | lock_page(page); | |
1743 | } | |
1da177e4 | 1744 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
8a9f3ccd | 1745 | if (vma->anon_vma && (ret = unuse_vma(vma, entry, page))) |
1da177e4 | 1746 | break; |
dc644a07 | 1747 | cond_resched(); |
1da177e4 | 1748 | } |
1da177e4 | 1749 | up_read(&mm->mmap_sem); |
8a9f3ccd | 1750 | return (ret < 0)? ret: 0; |
1da177e4 LT |
1751 | } |
1752 | ||
1753 | /* | |
38b5faf4 DM |
1754 | * Scan swap_map (or frontswap_map if frontswap parameter is true) |
1755 | * from current position to next entry still in use. | |
1da177e4 LT |
1756 | * Recycle to start on reaching the end, returning 0 when empty. |
1757 | */ | |
6eb396dc | 1758 | static unsigned int find_next_to_unuse(struct swap_info_struct *si, |
38b5faf4 | 1759 | unsigned int prev, bool frontswap) |
1da177e4 | 1760 | { |
6eb396dc HD |
1761 | unsigned int max = si->max; |
1762 | unsigned int i = prev; | |
8d69aaee | 1763 | unsigned char count; |
1da177e4 LT |
1764 | |
1765 | /* | |
5d337b91 | 1766 | * No need for swap_lock here: we're just looking |
1da177e4 LT |
1767 | * for whether an entry is in use, not modifying it; false |
1768 | * hits are okay, and sys_swapoff() has already prevented new | |
5d337b91 | 1769 | * allocations from this area (while holding swap_lock). |
1da177e4 LT |
1770 | */ |
1771 | for (;;) { | |
1772 | if (++i >= max) { | |
1773 | if (!prev) { | |
1774 | i = 0; | |
1775 | break; | |
1776 | } | |
1777 | /* | |
1778 | * No entries in use at top of swap_map, | |
1779 | * loop back to start and recheck there. | |
1780 | */ | |
1781 | max = prev + 1; | |
1782 | prev = 0; | |
1783 | i = 1; | |
1784 | } | |
4db0c3c2 | 1785 | count = READ_ONCE(si->swap_map[i]); |
355cfa73 | 1786 | if (count && swap_count(count) != SWAP_MAP_BAD) |
dc644a07 HD |
1787 | if (!frontswap || frontswap_test(si, i)) |
1788 | break; | |
1789 | if ((i % LATENCY_LIMIT) == 0) | |
1790 | cond_resched(); | |
1da177e4 LT |
1791 | } |
1792 | return i; | |
1793 | } | |
1794 | ||
1795 | /* | |
1796 | * We completely avoid races by reading each swap page in advance, | |
1797 | * and then search for the process using it. All the necessary | |
1798 | * page table adjustments can then be made atomically. | |
38b5faf4 DM |
1799 | * |
1800 | * if the boolean frontswap is true, only unuse pages_to_unuse pages; | |
1801 | * pages_to_unuse==0 means all pages; ignored if frontswap is false | |
1da177e4 | 1802 | */ |
38b5faf4 DM |
1803 | int try_to_unuse(unsigned int type, bool frontswap, |
1804 | unsigned long pages_to_unuse) | |
1da177e4 | 1805 | { |
efa90a98 | 1806 | struct swap_info_struct *si = swap_info[type]; |
1da177e4 | 1807 | struct mm_struct *start_mm; |
edfe23da SL |
1808 | volatile unsigned char *swap_map; /* swap_map is accessed without |
1809 | * locking. Mark it as volatile | |
1810 | * to prevent compiler doing | |
1811 | * something odd. | |
1812 | */ | |
8d69aaee | 1813 | unsigned char swcount; |
1da177e4 LT |
1814 | struct page *page; |
1815 | swp_entry_t entry; | |
6eb396dc | 1816 | unsigned int i = 0; |
1da177e4 | 1817 | int retval = 0; |
1da177e4 LT |
1818 | |
1819 | /* | |
1820 | * When searching mms for an entry, a good strategy is to | |
1821 | * start at the first mm we freed the previous entry from | |
1822 | * (though actually we don't notice whether we or coincidence | |
1823 | * freed the entry). Initialize this start_mm with a hold. | |
1824 | * | |
1825 | * A simpler strategy would be to start at the last mm we | |
1826 | * freed the previous entry from; but that would take less | |
1827 | * advantage of mmlist ordering, which clusters forked mms | |
1828 | * together, child after parent. If we race with dup_mmap(), we | |
1829 | * prefer to resolve parent before child, lest we miss entries | |
1830 | * duplicated after we scanned child: using last mm would invert | |
570a335b | 1831 | * that. |
1da177e4 LT |
1832 | */ |
1833 | start_mm = &init_mm; | |
3fce371b | 1834 | mmget(&init_mm); |
1da177e4 LT |
1835 | |
1836 | /* | |
1837 | * Keep on scanning until all entries have gone. Usually, | |
1838 | * one pass through swap_map is enough, but not necessarily: | |
1839 | * there are races when an instance of an entry might be missed. | |
1840 | */ | |
38b5faf4 | 1841 | while ((i = find_next_to_unuse(si, i, frontswap)) != 0) { |
1da177e4 LT |
1842 | if (signal_pending(current)) { |
1843 | retval = -EINTR; | |
1844 | break; | |
1845 | } | |
1846 | ||
886bb7e9 | 1847 | /* |
1da177e4 LT |
1848 | * Get a page for the entry, using the existing swap |
1849 | * cache page if there is one. Otherwise, get a clean | |
886bb7e9 | 1850 | * page and read the swap into it. |
1da177e4 LT |
1851 | */ |
1852 | swap_map = &si->swap_map[i]; | |
1853 | entry = swp_entry(type, i); | |
02098fea HD |
1854 | page = read_swap_cache_async(entry, |
1855 | GFP_HIGHUSER_MOVABLE, NULL, 0); | |
1da177e4 LT |
1856 | if (!page) { |
1857 | /* | |
1858 | * Either swap_duplicate() failed because entry | |
1859 | * has been freed independently, and will not be | |
1860 | * reused since sys_swapoff() already disabled | |
1861 | * allocation from here, or alloc_page() failed. | |
1862 | */ | |
edfe23da SL |
1863 | swcount = *swap_map; |
1864 | /* | |
1865 | * We don't hold lock here, so the swap entry could be | |
1866 | * SWAP_MAP_BAD (when the cluster is discarding). | |
1867 | * Instead of fail out, We can just skip the swap | |
1868 | * entry because swapoff will wait for discarding | |
1869 | * finish anyway. | |
1870 | */ | |
1871 | if (!swcount || swcount == SWAP_MAP_BAD) | |
1da177e4 LT |
1872 | continue; |
1873 | retval = -ENOMEM; | |
1874 | break; | |
1875 | } | |
1876 | ||
1877 | /* | |
1878 | * Don't hold on to start_mm if it looks like exiting. | |
1879 | */ | |
1880 | if (atomic_read(&start_mm->mm_users) == 1) { | |
1881 | mmput(start_mm); | |
1882 | start_mm = &init_mm; | |
3fce371b | 1883 | mmget(&init_mm); |
1da177e4 LT |
1884 | } |
1885 | ||
1886 | /* | |
1887 | * Wait for and lock page. When do_swap_page races with | |
1888 | * try_to_unuse, do_swap_page can handle the fault much | |
1889 | * faster than try_to_unuse can locate the entry. This | |
1890 | * apparently redundant "wait_on_page_locked" lets try_to_unuse | |
1891 | * defer to do_swap_page in such a case - in some tests, | |
1892 | * do_swap_page and try_to_unuse repeatedly compete. | |
1893 | */ | |
1894 | wait_on_page_locked(page); | |
1895 | wait_on_page_writeback(page); | |
1896 | lock_page(page); | |
1897 | wait_on_page_writeback(page); | |
1898 | ||
1899 | /* | |
1900 | * Remove all references to entry. | |
1da177e4 | 1901 | */ |
1da177e4 | 1902 | swcount = *swap_map; |
aaa46865 HD |
1903 | if (swap_count(swcount) == SWAP_MAP_SHMEM) { |
1904 | retval = shmem_unuse(entry, page); | |
1905 | /* page has already been unlocked and released */ | |
1906 | if (retval < 0) | |
1907 | break; | |
1908 | continue; | |
1da177e4 | 1909 | } |
aaa46865 HD |
1910 | if (swap_count(swcount) && start_mm != &init_mm) |
1911 | retval = unuse_mm(start_mm, entry, page); | |
1912 | ||
355cfa73 | 1913 | if (swap_count(*swap_map)) { |
1da177e4 LT |
1914 | int set_start_mm = (*swap_map >= swcount); |
1915 | struct list_head *p = &start_mm->mmlist; | |
1916 | struct mm_struct *new_start_mm = start_mm; | |
1917 | struct mm_struct *prev_mm = start_mm; | |
1918 | struct mm_struct *mm; | |
1919 | ||
3fce371b VN |
1920 | mmget(new_start_mm); |
1921 | mmget(prev_mm); | |
1da177e4 | 1922 | spin_lock(&mmlist_lock); |
aaa46865 | 1923 | while (swap_count(*swap_map) && !retval && |
1da177e4 LT |
1924 | (p = p->next) != &start_mm->mmlist) { |
1925 | mm = list_entry(p, struct mm_struct, mmlist); | |
388f7934 | 1926 | if (!mmget_not_zero(mm)) |
1da177e4 | 1927 | continue; |
1da177e4 LT |
1928 | spin_unlock(&mmlist_lock); |
1929 | mmput(prev_mm); | |
1930 | prev_mm = mm; | |
1931 | ||
1932 | cond_resched(); | |
1933 | ||
1934 | swcount = *swap_map; | |
355cfa73 | 1935 | if (!swap_count(swcount)) /* any usage ? */ |
1da177e4 | 1936 | ; |
aaa46865 | 1937 | else if (mm == &init_mm) |
1da177e4 | 1938 | set_start_mm = 1; |
aaa46865 | 1939 | else |
1da177e4 | 1940 | retval = unuse_mm(mm, entry, page); |
355cfa73 | 1941 | |
32c5fc10 | 1942 | if (set_start_mm && *swap_map < swcount) { |
1da177e4 | 1943 | mmput(new_start_mm); |
3fce371b | 1944 | mmget(mm); |
1da177e4 LT |
1945 | new_start_mm = mm; |
1946 | set_start_mm = 0; | |
1947 | } | |
1948 | spin_lock(&mmlist_lock); | |
1949 | } | |
1950 | spin_unlock(&mmlist_lock); | |
1951 | mmput(prev_mm); | |
1952 | mmput(start_mm); | |
1953 | start_mm = new_start_mm; | |
1954 | } | |
1955 | if (retval) { | |
1956 | unlock_page(page); | |
09cbfeaf | 1957 | put_page(page); |
1da177e4 LT |
1958 | break; |
1959 | } | |
1960 | ||
1da177e4 LT |
1961 | /* |
1962 | * If a reference remains (rare), we would like to leave | |
1963 | * the page in the swap cache; but try_to_unmap could | |
1964 | * then re-duplicate the entry once we drop page lock, | |
1965 | * so we might loop indefinitely; also, that page could | |
1966 | * not be swapped out to other storage meanwhile. So: | |
1967 | * delete from cache even if there's another reference, | |
1968 | * after ensuring that the data has been saved to disk - | |
1969 | * since if the reference remains (rarer), it will be | |
1970 | * read from disk into another page. Splitting into two | |
1971 | * pages would be incorrect if swap supported "shared | |
1972 | * private" pages, but they are handled by tmpfs files. | |
5ad64688 HD |
1973 | * |
1974 | * Given how unuse_vma() targets one particular offset | |
1975 | * in an anon_vma, once the anon_vma has been determined, | |
1976 | * this splitting happens to be just what is needed to | |
1977 | * handle where KSM pages have been swapped out: re-reading | |
1978 | * is unnecessarily slow, but we can fix that later on. | |
1da177e4 | 1979 | */ |
355cfa73 KH |
1980 | if (swap_count(*swap_map) && |
1981 | PageDirty(page) && PageSwapCache(page)) { | |
1da177e4 LT |
1982 | struct writeback_control wbc = { |
1983 | .sync_mode = WB_SYNC_NONE, | |
1984 | }; | |
1985 | ||
1986 | swap_writepage(page, &wbc); | |
1987 | lock_page(page); | |
1988 | wait_on_page_writeback(page); | |
1989 | } | |
68bdc8d6 HD |
1990 | |
1991 | /* | |
1992 | * It is conceivable that a racing task removed this page from | |
1993 | * swap cache just before we acquired the page lock at the top, | |
1994 | * or while we dropped it in unuse_mm(). The page might even | |
1995 | * be back in swap cache on another swap area: that we must not | |
1996 | * delete, since it may not have been written out to swap yet. | |
1997 | */ | |
1998 | if (PageSwapCache(page) && | |
1999 | likely(page_private(page) == entry.val)) | |
2e0e26c7 | 2000 | delete_from_swap_cache(page); |
1da177e4 LT |
2001 | |
2002 | /* | |
2003 | * So we could skip searching mms once swap count went | |
2004 | * to 1, we did not mark any present ptes as dirty: must | |
2706a1b8 | 2005 | * mark page dirty so shrink_page_list will preserve it. |
1da177e4 LT |
2006 | */ |
2007 | SetPageDirty(page); | |
2008 | unlock_page(page); | |
09cbfeaf | 2009 | put_page(page); |
1da177e4 LT |
2010 | |
2011 | /* | |
2012 | * Make sure that we aren't completely killing | |
2013 | * interactive performance. | |
2014 | */ | |
2015 | cond_resched(); | |
38b5faf4 DM |
2016 | if (frontswap && pages_to_unuse > 0) { |
2017 | if (!--pages_to_unuse) | |
2018 | break; | |
2019 | } | |
1da177e4 LT |
2020 | } |
2021 | ||
2022 | mmput(start_mm); | |
1da177e4 LT |
2023 | return retval; |
2024 | } | |
2025 | ||
2026 | /* | |
5d337b91 HD |
2027 | * After a successful try_to_unuse, if no swap is now in use, we know |
2028 | * we can empty the mmlist. swap_lock must be held on entry and exit. | |
2029 | * Note that mmlist_lock nests inside swap_lock, and an mm must be | |
1da177e4 LT |
2030 | * added to the mmlist just after page_duplicate - before would be racy. |
2031 | */ | |
2032 | static void drain_mmlist(void) | |
2033 | { | |
2034 | struct list_head *p, *next; | |
efa90a98 | 2035 | unsigned int type; |
1da177e4 | 2036 | |
efa90a98 HD |
2037 | for (type = 0; type < nr_swapfiles; type++) |
2038 | if (swap_info[type]->inuse_pages) | |
1da177e4 LT |
2039 | return; |
2040 | spin_lock(&mmlist_lock); | |
2041 | list_for_each_safe(p, next, &init_mm.mmlist) | |
2042 | list_del_init(p); | |
2043 | spin_unlock(&mmlist_lock); | |
2044 | } | |
2045 | ||
2046 | /* | |
2047 | * Use this swapdev's extent info to locate the (PAGE_SIZE) block which | |
d4906e1a LS |
2048 | * corresponds to page offset for the specified swap entry. |
2049 | * Note that the type of this function is sector_t, but it returns page offset | |
2050 | * into the bdev, not sector offset. | |
1da177e4 | 2051 | */ |
d4906e1a | 2052 | static sector_t map_swap_entry(swp_entry_t entry, struct block_device **bdev) |
1da177e4 | 2053 | { |
f29ad6a9 HD |
2054 | struct swap_info_struct *sis; |
2055 | struct swap_extent *start_se; | |
2056 | struct swap_extent *se; | |
2057 | pgoff_t offset; | |
2058 | ||
efa90a98 | 2059 | sis = swap_info[swp_type(entry)]; |
f29ad6a9 HD |
2060 | *bdev = sis->bdev; |
2061 | ||
2062 | offset = swp_offset(entry); | |
2063 | start_se = sis->curr_swap_extent; | |
2064 | se = start_se; | |
1da177e4 LT |
2065 | |
2066 | for ( ; ; ) { | |
1da177e4 LT |
2067 | if (se->start_page <= offset && |
2068 | offset < (se->start_page + se->nr_pages)) { | |
2069 | return se->start_block + (offset - se->start_page); | |
2070 | } | |
a8ae4991 | 2071 | se = list_next_entry(se, list); |
1da177e4 LT |
2072 | sis->curr_swap_extent = se; |
2073 | BUG_ON(se == start_se); /* It *must* be present */ | |
2074 | } | |
2075 | } | |
2076 | ||
d4906e1a LS |
2077 | /* |
2078 | * Returns the page offset into bdev for the specified page's swap entry. | |
2079 | */ | |
2080 | sector_t map_swap_page(struct page *page, struct block_device **bdev) | |
2081 | { | |
2082 | swp_entry_t entry; | |
2083 | entry.val = page_private(page); | |
2084 | return map_swap_entry(entry, bdev); | |
2085 | } | |
2086 | ||
1da177e4 LT |
2087 | /* |
2088 | * Free all of a swapdev's extent information | |
2089 | */ | |
2090 | static void destroy_swap_extents(struct swap_info_struct *sis) | |
2091 | { | |
9625a5f2 | 2092 | while (!list_empty(&sis->first_swap_extent.list)) { |
1da177e4 LT |
2093 | struct swap_extent *se; |
2094 | ||
a8ae4991 | 2095 | se = list_first_entry(&sis->first_swap_extent.list, |
1da177e4 LT |
2096 | struct swap_extent, list); |
2097 | list_del(&se->list); | |
2098 | kfree(se); | |
2099 | } | |
62c230bc MG |
2100 | |
2101 | if (sis->flags & SWP_FILE) { | |
2102 | struct file *swap_file = sis->swap_file; | |
2103 | struct address_space *mapping = swap_file->f_mapping; | |
2104 | ||
2105 | sis->flags &= ~SWP_FILE; | |
2106 | mapping->a_ops->swap_deactivate(swap_file); | |
2107 | } | |
1da177e4 LT |
2108 | } |
2109 | ||
2110 | /* | |
2111 | * Add a block range (and the corresponding page range) into this swapdev's | |
11d31886 | 2112 | * extent list. The extent list is kept sorted in page order. |
1da177e4 | 2113 | * |
11d31886 | 2114 | * This function rather assumes that it is called in ascending page order. |
1da177e4 | 2115 | */ |
a509bc1a | 2116 | int |
1da177e4 LT |
2117 | add_swap_extent(struct swap_info_struct *sis, unsigned long start_page, |
2118 | unsigned long nr_pages, sector_t start_block) | |
2119 | { | |
2120 | struct swap_extent *se; | |
2121 | struct swap_extent *new_se; | |
2122 | struct list_head *lh; | |
2123 | ||
9625a5f2 HD |
2124 | if (start_page == 0) { |
2125 | se = &sis->first_swap_extent; | |
2126 | sis->curr_swap_extent = se; | |
2127 | se->start_page = 0; | |
2128 | se->nr_pages = nr_pages; | |
2129 | se->start_block = start_block; | |
2130 | return 1; | |
2131 | } else { | |
2132 | lh = sis->first_swap_extent.list.prev; /* Highest extent */ | |
1da177e4 | 2133 | se = list_entry(lh, struct swap_extent, list); |
11d31886 HD |
2134 | BUG_ON(se->start_page + se->nr_pages != start_page); |
2135 | if (se->start_block + se->nr_pages == start_block) { | |
1da177e4 LT |
2136 | /* Merge it */ |
2137 | se->nr_pages += nr_pages; | |
2138 | return 0; | |
2139 | } | |
1da177e4 LT |
2140 | } |
2141 | ||
2142 | /* | |
2143 | * No merge. Insert a new extent, preserving ordering. | |
2144 | */ | |
2145 | new_se = kmalloc(sizeof(*se), GFP_KERNEL); | |
2146 | if (new_se == NULL) | |
2147 | return -ENOMEM; | |
2148 | new_se->start_page = start_page; | |
2149 | new_se->nr_pages = nr_pages; | |
2150 | new_se->start_block = start_block; | |
2151 | ||
9625a5f2 | 2152 | list_add_tail(&new_se->list, &sis->first_swap_extent.list); |
53092a74 | 2153 | return 1; |
1da177e4 LT |
2154 | } |
2155 | ||
2156 | /* | |
2157 | * A `swap extent' is a simple thing which maps a contiguous range of pages | |
2158 | * onto a contiguous range of disk blocks. An ordered list of swap extents | |
2159 | * is built at swapon time and is then used at swap_writepage/swap_readpage | |
2160 | * time for locating where on disk a page belongs. | |
2161 | * | |
2162 | * If the swapfile is an S_ISBLK block device, a single extent is installed. | |
2163 | * This is done so that the main operating code can treat S_ISBLK and S_ISREG | |
2164 | * swap files identically. | |
2165 | * | |
2166 | * Whether the swapdev is an S_ISREG file or an S_ISBLK blockdev, the swap | |
2167 | * extent list operates in PAGE_SIZE disk blocks. Both S_ISREG and S_ISBLK | |
2168 | * swapfiles are handled *identically* after swapon time. | |
2169 | * | |
2170 | * For S_ISREG swapfiles, setup_swap_extents() will walk all the file's blocks | |
2171 | * and will parse them into an ordered extent list, in PAGE_SIZE chunks. If | |
2172 | * some stray blocks are found which do not fall within the PAGE_SIZE alignment | |
2173 | * requirements, they are simply tossed out - we will never use those blocks | |
2174 | * for swapping. | |
2175 | * | |
b0d9bcd4 | 2176 | * For S_ISREG swapfiles we set S_SWAPFILE across the life of the swapon. This |
1da177e4 LT |
2177 | * prevents root from shooting her foot off by ftruncating an in-use swapfile, |
2178 | * which will scribble on the fs. | |
2179 | * | |
2180 | * The amount of disk space which a single swap extent represents varies. | |
2181 | * Typically it is in the 1-4 megabyte range. So we can have hundreds of | |
2182 | * extents in the list. To avoid much list walking, we cache the previous | |
2183 | * search location in `curr_swap_extent', and start new searches from there. | |
2184 | * This is extremely effective. The average number of iterations in | |
2185 | * map_swap_page() has been measured at about 0.3 per page. - akpm. | |
2186 | */ | |
53092a74 | 2187 | static int setup_swap_extents(struct swap_info_struct *sis, sector_t *span) |
1da177e4 | 2188 | { |
62c230bc MG |
2189 | struct file *swap_file = sis->swap_file; |
2190 | struct address_space *mapping = swap_file->f_mapping; | |
2191 | struct inode *inode = mapping->host; | |
1da177e4 LT |
2192 | int ret; |
2193 | ||
1da177e4 LT |
2194 | if (S_ISBLK(inode->i_mode)) { |
2195 | ret = add_swap_extent(sis, 0, sis->max, 0); | |
53092a74 | 2196 | *span = sis->pages; |
a509bc1a | 2197 | return ret; |
1da177e4 LT |
2198 | } |
2199 | ||
62c230bc | 2200 | if (mapping->a_ops->swap_activate) { |
a509bc1a | 2201 | ret = mapping->a_ops->swap_activate(sis, swap_file, span); |
62c230bc MG |
2202 | if (!ret) { |
2203 | sis->flags |= SWP_FILE; | |
2204 | ret = add_swap_extent(sis, 0, sis->max, 0); | |
2205 | *span = sis->pages; | |
2206 | } | |
a509bc1a | 2207 | return ret; |
62c230bc MG |
2208 | } |
2209 | ||
a509bc1a | 2210 | return generic_swapfile_activate(sis, swap_file, span); |
1da177e4 LT |
2211 | } |
2212 | ||
cf0cac0a | 2213 | static void _enable_swap_info(struct swap_info_struct *p, int prio, |
2a8f9449 SL |
2214 | unsigned char *swap_map, |
2215 | struct swap_cluster_info *cluster_info) | |
40531542 | 2216 | { |
40531542 CEB |
2217 | if (prio >= 0) |
2218 | p->prio = prio; | |
2219 | else | |
2220 | p->prio = --least_priority; | |
18ab4d4c DS |
2221 | /* |
2222 | * the plist prio is negated because plist ordering is | |
2223 | * low-to-high, while swap ordering is high-to-low | |
2224 | */ | |
2225 | p->list.prio = -p->prio; | |
2226 | p->avail_list.prio = -p->prio; | |
40531542 | 2227 | p->swap_map = swap_map; |
2a8f9449 | 2228 | p->cluster_info = cluster_info; |
40531542 | 2229 | p->flags |= SWP_WRITEOK; |
ec8acf20 | 2230 | atomic_long_add(p->pages, &nr_swap_pages); |
40531542 CEB |
2231 | total_swap_pages += p->pages; |
2232 | ||
adfab836 | 2233 | assert_spin_locked(&swap_lock); |
adfab836 | 2234 | /* |
18ab4d4c DS |
2235 | * both lists are plists, and thus priority ordered. |
2236 | * swap_active_head needs to be priority ordered for swapoff(), | |
2237 | * which on removal of any swap_info_struct with an auto-assigned | |
2238 | * (i.e. negative) priority increments the auto-assigned priority | |
2239 | * of any lower-priority swap_info_structs. | |
2240 | * swap_avail_head needs to be priority ordered for get_swap_page(), | |
2241 | * which allocates swap pages from the highest available priority | |
2242 | * swap_info_struct. | |
adfab836 | 2243 | */ |
18ab4d4c DS |
2244 | plist_add(&p->list, &swap_active_head); |
2245 | spin_lock(&swap_avail_lock); | |
2246 | plist_add(&p->avail_list, &swap_avail_head); | |
2247 | spin_unlock(&swap_avail_lock); | |
cf0cac0a CEB |
2248 | } |
2249 | ||
2250 | static void enable_swap_info(struct swap_info_struct *p, int prio, | |
2251 | unsigned char *swap_map, | |
2a8f9449 | 2252 | struct swap_cluster_info *cluster_info, |
cf0cac0a CEB |
2253 | unsigned long *frontswap_map) |
2254 | { | |
4f89849d | 2255 | frontswap_init(p->type, frontswap_map); |
cf0cac0a | 2256 | spin_lock(&swap_lock); |
ec8acf20 | 2257 | spin_lock(&p->lock); |
2a8f9449 | 2258 | _enable_swap_info(p, prio, swap_map, cluster_info); |
ec8acf20 | 2259 | spin_unlock(&p->lock); |
cf0cac0a CEB |
2260 | spin_unlock(&swap_lock); |
2261 | } | |
2262 | ||
2263 | static void reinsert_swap_info(struct swap_info_struct *p) | |
2264 | { | |
2265 | spin_lock(&swap_lock); | |
ec8acf20 | 2266 | spin_lock(&p->lock); |
2a8f9449 | 2267 | _enable_swap_info(p, p->prio, p->swap_map, p->cluster_info); |
ec8acf20 | 2268 | spin_unlock(&p->lock); |
40531542 CEB |
2269 | spin_unlock(&swap_lock); |
2270 | } | |
2271 | ||
67afa38e TC |
2272 | bool has_usable_swap(void) |
2273 | { | |
2274 | bool ret = true; | |
2275 | ||
2276 | spin_lock(&swap_lock); | |
2277 | if (plist_head_empty(&swap_active_head)) | |
2278 | ret = false; | |
2279 | spin_unlock(&swap_lock); | |
2280 | return ret; | |
2281 | } | |
2282 | ||
c4ea37c2 | 2283 | SYSCALL_DEFINE1(swapoff, const char __user *, specialfile) |
1da177e4 | 2284 | { |
73c34b6a | 2285 | struct swap_info_struct *p = NULL; |
8d69aaee | 2286 | unsigned char *swap_map; |
2a8f9449 | 2287 | struct swap_cluster_info *cluster_info; |
4f89849d | 2288 | unsigned long *frontswap_map; |
1da177e4 LT |
2289 | struct file *swap_file, *victim; |
2290 | struct address_space *mapping; | |
2291 | struct inode *inode; | |
91a27b2a | 2292 | struct filename *pathname; |
adfab836 | 2293 | int err, found = 0; |
5b808a23 | 2294 | unsigned int old_block_size; |
886bb7e9 | 2295 | |
1da177e4 LT |
2296 | if (!capable(CAP_SYS_ADMIN)) |
2297 | return -EPERM; | |
2298 | ||
191c5424 AV |
2299 | BUG_ON(!current->mm); |
2300 | ||
1da177e4 | 2301 | pathname = getname(specialfile); |
1da177e4 | 2302 | if (IS_ERR(pathname)) |
f58b59c1 | 2303 | return PTR_ERR(pathname); |
1da177e4 | 2304 | |
669abf4e | 2305 | victim = file_open_name(pathname, O_RDWR|O_LARGEFILE, 0); |
1da177e4 LT |
2306 | err = PTR_ERR(victim); |
2307 | if (IS_ERR(victim)) | |
2308 | goto out; | |
2309 | ||
2310 | mapping = victim->f_mapping; | |
5d337b91 | 2311 | spin_lock(&swap_lock); |
18ab4d4c | 2312 | plist_for_each_entry(p, &swap_active_head, list) { |
22c6f8fd | 2313 | if (p->flags & SWP_WRITEOK) { |
adfab836 DS |
2314 | if (p->swap_file->f_mapping == mapping) { |
2315 | found = 1; | |
1da177e4 | 2316 | break; |
adfab836 | 2317 | } |
1da177e4 | 2318 | } |
1da177e4 | 2319 | } |
adfab836 | 2320 | if (!found) { |
1da177e4 | 2321 | err = -EINVAL; |
5d337b91 | 2322 | spin_unlock(&swap_lock); |
1da177e4 LT |
2323 | goto out_dput; |
2324 | } | |
191c5424 | 2325 | if (!security_vm_enough_memory_mm(current->mm, p->pages)) |
1da177e4 LT |
2326 | vm_unacct_memory(p->pages); |
2327 | else { | |
2328 | err = -ENOMEM; | |
5d337b91 | 2329 | spin_unlock(&swap_lock); |
1da177e4 LT |
2330 | goto out_dput; |
2331 | } | |
18ab4d4c DS |
2332 | spin_lock(&swap_avail_lock); |
2333 | plist_del(&p->avail_list, &swap_avail_head); | |
2334 | spin_unlock(&swap_avail_lock); | |
ec8acf20 | 2335 | spin_lock(&p->lock); |
78ecba08 | 2336 | if (p->prio < 0) { |
adfab836 DS |
2337 | struct swap_info_struct *si = p; |
2338 | ||
18ab4d4c | 2339 | plist_for_each_entry_continue(si, &swap_active_head, list) { |
adfab836 | 2340 | si->prio++; |
18ab4d4c DS |
2341 | si->list.prio--; |
2342 | si->avail_list.prio--; | |
adfab836 | 2343 | } |
78ecba08 HD |
2344 | least_priority++; |
2345 | } | |
18ab4d4c | 2346 | plist_del(&p->list, &swap_active_head); |
ec8acf20 | 2347 | atomic_long_sub(p->pages, &nr_swap_pages); |
1da177e4 LT |
2348 | total_swap_pages -= p->pages; |
2349 | p->flags &= ~SWP_WRITEOK; | |
ec8acf20 | 2350 | spin_unlock(&p->lock); |
5d337b91 | 2351 | spin_unlock(&swap_lock); |
fb4f88dc | 2352 | |
039939a6 TC |
2353 | disable_swap_slots_cache_lock(); |
2354 | ||
e1e12d2f | 2355 | set_current_oom_origin(); |
adfab836 | 2356 | err = try_to_unuse(p->type, false, 0); /* force unuse all pages */ |
e1e12d2f | 2357 | clear_current_oom_origin(); |
1da177e4 | 2358 | |
1da177e4 LT |
2359 | if (err) { |
2360 | /* re-insert swap space back into swap_list */ | |
cf0cac0a | 2361 | reinsert_swap_info(p); |
039939a6 | 2362 | reenable_swap_slots_cache_unlock(); |
1da177e4 LT |
2363 | goto out_dput; |
2364 | } | |
52b7efdb | 2365 | |
039939a6 TC |
2366 | reenable_swap_slots_cache_unlock(); |
2367 | ||
815c2c54 SL |
2368 | flush_work(&p->discard_work); |
2369 | ||
5d337b91 | 2370 | destroy_swap_extents(p); |
570a335b HD |
2371 | if (p->flags & SWP_CONTINUED) |
2372 | free_swap_count_continuations(p); | |
2373 | ||
fc0abb14 | 2374 | mutex_lock(&swapon_mutex); |
5d337b91 | 2375 | spin_lock(&swap_lock); |
ec8acf20 | 2376 | spin_lock(&p->lock); |
5d337b91 HD |
2377 | drain_mmlist(); |
2378 | ||
52b7efdb | 2379 | /* wait for anyone still in scan_swap_map */ |
52b7efdb HD |
2380 | p->highest_bit = 0; /* cuts scans short */ |
2381 | while (p->flags >= SWP_SCANNING) { | |
ec8acf20 | 2382 | spin_unlock(&p->lock); |
5d337b91 | 2383 | spin_unlock(&swap_lock); |
13e4b57f | 2384 | schedule_timeout_uninterruptible(1); |
5d337b91 | 2385 | spin_lock(&swap_lock); |
ec8acf20 | 2386 | spin_lock(&p->lock); |
52b7efdb | 2387 | } |
52b7efdb | 2388 | |
1da177e4 | 2389 | swap_file = p->swap_file; |
5b808a23 | 2390 | old_block_size = p->old_block_size; |
1da177e4 LT |
2391 | p->swap_file = NULL; |
2392 | p->max = 0; | |
2393 | swap_map = p->swap_map; | |
2394 | p->swap_map = NULL; | |
2a8f9449 SL |
2395 | cluster_info = p->cluster_info; |
2396 | p->cluster_info = NULL; | |
4f89849d | 2397 | frontswap_map = frontswap_map_get(p); |
ec8acf20 | 2398 | spin_unlock(&p->lock); |
5d337b91 | 2399 | spin_unlock(&swap_lock); |
adfab836 | 2400 | frontswap_invalidate_area(p->type); |
58e97ba6 | 2401 | frontswap_map_set(p, NULL); |
fc0abb14 | 2402 | mutex_unlock(&swapon_mutex); |
ebc2a1a6 SL |
2403 | free_percpu(p->percpu_cluster); |
2404 | p->percpu_cluster = NULL; | |
1da177e4 | 2405 | vfree(swap_map); |
54f180d3 HY |
2406 | kvfree(cluster_info); |
2407 | kvfree(frontswap_map); | |
2de1a7e4 | 2408 | /* Destroy swap account information */ |
adfab836 | 2409 | swap_cgroup_swapoff(p->type); |
4b3ef9da | 2410 | exit_swap_address_space(p->type); |
27a7faa0 | 2411 | |
1da177e4 LT |
2412 | inode = mapping->host; |
2413 | if (S_ISBLK(inode->i_mode)) { | |
2414 | struct block_device *bdev = I_BDEV(inode); | |
5b808a23 | 2415 | set_blocksize(bdev, old_block_size); |
e525fd89 | 2416 | blkdev_put(bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL); |
1da177e4 | 2417 | } else { |
5955102c | 2418 | inode_lock(inode); |
1da177e4 | 2419 | inode->i_flags &= ~S_SWAPFILE; |
5955102c | 2420 | inode_unlock(inode); |
1da177e4 LT |
2421 | } |
2422 | filp_close(swap_file, NULL); | |
f893ab41 WY |
2423 | |
2424 | /* | |
2425 | * Clear the SWP_USED flag after all resources are freed so that swapon | |
2426 | * can reuse this swap_info in alloc_swap_info() safely. It is ok to | |
2427 | * not hold p->lock after we cleared its SWP_WRITEOK. | |
2428 | */ | |
2429 | spin_lock(&swap_lock); | |
2430 | p->flags = 0; | |
2431 | spin_unlock(&swap_lock); | |
2432 | ||
1da177e4 | 2433 | err = 0; |
66d7dd51 KS |
2434 | atomic_inc(&proc_poll_event); |
2435 | wake_up_interruptible(&proc_poll_wait); | |
1da177e4 LT |
2436 | |
2437 | out_dput: | |
2438 | filp_close(victim, NULL); | |
2439 | out: | |
f58b59c1 | 2440 | putname(pathname); |
1da177e4 LT |
2441 | return err; |
2442 | } | |
2443 | ||
2444 | #ifdef CONFIG_PROC_FS | |
66d7dd51 KS |
2445 | static unsigned swaps_poll(struct file *file, poll_table *wait) |
2446 | { | |
f1514638 | 2447 | struct seq_file *seq = file->private_data; |
66d7dd51 KS |
2448 | |
2449 | poll_wait(file, &proc_poll_wait, wait); | |
2450 | ||
f1514638 KS |
2451 | if (seq->poll_event != atomic_read(&proc_poll_event)) { |
2452 | seq->poll_event = atomic_read(&proc_poll_event); | |
66d7dd51 KS |
2453 | return POLLIN | POLLRDNORM | POLLERR | POLLPRI; |
2454 | } | |
2455 | ||
2456 | return POLLIN | POLLRDNORM; | |
2457 | } | |
2458 | ||
1da177e4 LT |
2459 | /* iterator */ |
2460 | static void *swap_start(struct seq_file *swap, loff_t *pos) | |
2461 | { | |
efa90a98 HD |
2462 | struct swap_info_struct *si; |
2463 | int type; | |
1da177e4 LT |
2464 | loff_t l = *pos; |
2465 | ||
fc0abb14 | 2466 | mutex_lock(&swapon_mutex); |
1da177e4 | 2467 | |
881e4aab SS |
2468 | if (!l) |
2469 | return SEQ_START_TOKEN; | |
2470 | ||
efa90a98 HD |
2471 | for (type = 0; type < nr_swapfiles; type++) { |
2472 | smp_rmb(); /* read nr_swapfiles before swap_info[type] */ | |
2473 | si = swap_info[type]; | |
2474 | if (!(si->flags & SWP_USED) || !si->swap_map) | |
1da177e4 | 2475 | continue; |
881e4aab | 2476 | if (!--l) |
efa90a98 | 2477 | return si; |
1da177e4 LT |
2478 | } |
2479 | ||
2480 | return NULL; | |
2481 | } | |
2482 | ||
2483 | static void *swap_next(struct seq_file *swap, void *v, loff_t *pos) | |
2484 | { | |
efa90a98 HD |
2485 | struct swap_info_struct *si = v; |
2486 | int type; | |
1da177e4 | 2487 | |
881e4aab | 2488 | if (v == SEQ_START_TOKEN) |
efa90a98 HD |
2489 | type = 0; |
2490 | else | |
2491 | type = si->type + 1; | |
881e4aab | 2492 | |
efa90a98 HD |
2493 | for (; type < nr_swapfiles; type++) { |
2494 | smp_rmb(); /* read nr_swapfiles before swap_info[type] */ | |
2495 | si = swap_info[type]; | |
2496 | if (!(si->flags & SWP_USED) || !si->swap_map) | |
1da177e4 LT |
2497 | continue; |
2498 | ++*pos; | |
efa90a98 | 2499 | return si; |
1da177e4 LT |
2500 | } |
2501 | ||
2502 | return NULL; | |
2503 | } | |
2504 | ||
2505 | static void swap_stop(struct seq_file *swap, void *v) | |
2506 | { | |
fc0abb14 | 2507 | mutex_unlock(&swapon_mutex); |
1da177e4 LT |
2508 | } |
2509 | ||
2510 | static int swap_show(struct seq_file *swap, void *v) | |
2511 | { | |
efa90a98 | 2512 | struct swap_info_struct *si = v; |
1da177e4 LT |
2513 | struct file *file; |
2514 | int len; | |
2515 | ||
efa90a98 | 2516 | if (si == SEQ_START_TOKEN) { |
881e4aab SS |
2517 | seq_puts(swap,"Filename\t\t\t\tType\t\tSize\tUsed\tPriority\n"); |
2518 | return 0; | |
2519 | } | |
1da177e4 | 2520 | |
efa90a98 | 2521 | file = si->swap_file; |
2726d566 | 2522 | len = seq_file_path(swap, file, " \t\n\\"); |
6eb396dc | 2523 | seq_printf(swap, "%*s%s\t%u\t%u\t%d\n", |
886bb7e9 | 2524 | len < 40 ? 40 - len : 1, " ", |
496ad9aa | 2525 | S_ISBLK(file_inode(file)->i_mode) ? |
1da177e4 | 2526 | "partition" : "file\t", |
efa90a98 HD |
2527 | si->pages << (PAGE_SHIFT - 10), |
2528 | si->inuse_pages << (PAGE_SHIFT - 10), | |
2529 | si->prio); | |
1da177e4 LT |
2530 | return 0; |
2531 | } | |
2532 | ||
15ad7cdc | 2533 | static const struct seq_operations swaps_op = { |
1da177e4 LT |
2534 | .start = swap_start, |
2535 | .next = swap_next, | |
2536 | .stop = swap_stop, | |
2537 | .show = swap_show | |
2538 | }; | |
2539 | ||
2540 | static int swaps_open(struct inode *inode, struct file *file) | |
2541 | { | |
f1514638 | 2542 | struct seq_file *seq; |
66d7dd51 KS |
2543 | int ret; |
2544 | ||
66d7dd51 | 2545 | ret = seq_open(file, &swaps_op); |
f1514638 | 2546 | if (ret) |
66d7dd51 | 2547 | return ret; |
66d7dd51 | 2548 | |
f1514638 KS |
2549 | seq = file->private_data; |
2550 | seq->poll_event = atomic_read(&proc_poll_event); | |
2551 | return 0; | |
1da177e4 LT |
2552 | } |
2553 | ||
15ad7cdc | 2554 | static const struct file_operations proc_swaps_operations = { |
1da177e4 LT |
2555 | .open = swaps_open, |
2556 | .read = seq_read, | |
2557 | .llseek = seq_lseek, | |
2558 | .release = seq_release, | |
66d7dd51 | 2559 | .poll = swaps_poll, |
1da177e4 LT |
2560 | }; |
2561 | ||
2562 | static int __init procswaps_init(void) | |
2563 | { | |
3d71f86f | 2564 | proc_create("swaps", 0, NULL, &proc_swaps_operations); |
1da177e4 LT |
2565 | return 0; |
2566 | } | |
2567 | __initcall(procswaps_init); | |
2568 | #endif /* CONFIG_PROC_FS */ | |
2569 | ||
1796316a JB |
2570 | #ifdef MAX_SWAPFILES_CHECK |
2571 | static int __init max_swapfiles_check(void) | |
2572 | { | |
2573 | MAX_SWAPFILES_CHECK(); | |
2574 | return 0; | |
2575 | } | |
2576 | late_initcall(max_swapfiles_check); | |
2577 | #endif | |
2578 | ||
53cbb243 | 2579 | static struct swap_info_struct *alloc_swap_info(void) |
1da177e4 | 2580 | { |
73c34b6a | 2581 | struct swap_info_struct *p; |
1da177e4 | 2582 | unsigned int type; |
efa90a98 HD |
2583 | |
2584 | p = kzalloc(sizeof(*p), GFP_KERNEL); | |
2585 | if (!p) | |
53cbb243 | 2586 | return ERR_PTR(-ENOMEM); |
efa90a98 | 2587 | |
5d337b91 | 2588 | spin_lock(&swap_lock); |
efa90a98 HD |
2589 | for (type = 0; type < nr_swapfiles; type++) { |
2590 | if (!(swap_info[type]->flags & SWP_USED)) | |
1da177e4 | 2591 | break; |
efa90a98 | 2592 | } |
0697212a | 2593 | if (type >= MAX_SWAPFILES) { |
5d337b91 | 2594 | spin_unlock(&swap_lock); |
efa90a98 | 2595 | kfree(p); |
730c0581 | 2596 | return ERR_PTR(-EPERM); |
1da177e4 | 2597 | } |
efa90a98 HD |
2598 | if (type >= nr_swapfiles) { |
2599 | p->type = type; | |
2600 | swap_info[type] = p; | |
2601 | /* | |
2602 | * Write swap_info[type] before nr_swapfiles, in case a | |
2603 | * racing procfs swap_start() or swap_next() is reading them. | |
2604 | * (We never shrink nr_swapfiles, we never free this entry.) | |
2605 | */ | |
2606 | smp_wmb(); | |
2607 | nr_swapfiles++; | |
2608 | } else { | |
2609 | kfree(p); | |
2610 | p = swap_info[type]; | |
2611 | /* | |
2612 | * Do not memset this entry: a racing procfs swap_next() | |
2613 | * would be relying on p->type to remain valid. | |
2614 | */ | |
2615 | } | |
9625a5f2 | 2616 | INIT_LIST_HEAD(&p->first_swap_extent.list); |
18ab4d4c DS |
2617 | plist_node_init(&p->list, 0); |
2618 | plist_node_init(&p->avail_list, 0); | |
1da177e4 | 2619 | p->flags = SWP_USED; |
5d337b91 | 2620 | spin_unlock(&swap_lock); |
ec8acf20 | 2621 | spin_lock_init(&p->lock); |
efa90a98 | 2622 | |
53cbb243 | 2623 | return p; |
53cbb243 CEB |
2624 | } |
2625 | ||
4d0e1e10 CEB |
2626 | static int claim_swapfile(struct swap_info_struct *p, struct inode *inode) |
2627 | { | |
2628 | int error; | |
2629 | ||
2630 | if (S_ISBLK(inode->i_mode)) { | |
2631 | p->bdev = bdgrab(I_BDEV(inode)); | |
2632 | error = blkdev_get(p->bdev, | |
6f179af8 | 2633 | FMODE_READ | FMODE_WRITE | FMODE_EXCL, p); |
4d0e1e10 CEB |
2634 | if (error < 0) { |
2635 | p->bdev = NULL; | |
6f179af8 | 2636 | return error; |
4d0e1e10 CEB |
2637 | } |
2638 | p->old_block_size = block_size(p->bdev); | |
2639 | error = set_blocksize(p->bdev, PAGE_SIZE); | |
2640 | if (error < 0) | |
87ade72a | 2641 | return error; |
4d0e1e10 CEB |
2642 | p->flags |= SWP_BLKDEV; |
2643 | } else if (S_ISREG(inode->i_mode)) { | |
2644 | p->bdev = inode->i_sb->s_bdev; | |
5955102c | 2645 | inode_lock(inode); |
87ade72a CEB |
2646 | if (IS_SWAPFILE(inode)) |
2647 | return -EBUSY; | |
2648 | } else | |
2649 | return -EINVAL; | |
4d0e1e10 CEB |
2650 | |
2651 | return 0; | |
4d0e1e10 CEB |
2652 | } |
2653 | ||
ca8bd38b CEB |
2654 | static unsigned long read_swap_header(struct swap_info_struct *p, |
2655 | union swap_header *swap_header, | |
2656 | struct inode *inode) | |
2657 | { | |
2658 | int i; | |
2659 | unsigned long maxpages; | |
2660 | unsigned long swapfilepages; | |
d6bbbd29 | 2661 | unsigned long last_page; |
ca8bd38b CEB |
2662 | |
2663 | if (memcmp("SWAPSPACE2", swap_header->magic.magic, 10)) { | |
465c47fd | 2664 | pr_err("Unable to find swap-space signature\n"); |
38719025 | 2665 | return 0; |
ca8bd38b CEB |
2666 | } |
2667 | ||
2668 | /* swap partition endianess hack... */ | |
2669 | if (swab32(swap_header->info.version) == 1) { | |
2670 | swab32s(&swap_header->info.version); | |
2671 | swab32s(&swap_header->info.last_page); | |
2672 | swab32s(&swap_header->info.nr_badpages); | |
dd111be6 JH |
2673 | if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES) |
2674 | return 0; | |
ca8bd38b CEB |
2675 | for (i = 0; i < swap_header->info.nr_badpages; i++) |
2676 | swab32s(&swap_header->info.badpages[i]); | |
2677 | } | |
2678 | /* Check the swap header's sub-version */ | |
2679 | if (swap_header->info.version != 1) { | |
465c47fd AM |
2680 | pr_warn("Unable to handle swap header version %d\n", |
2681 | swap_header->info.version); | |
38719025 | 2682 | return 0; |
ca8bd38b CEB |
2683 | } |
2684 | ||
2685 | p->lowest_bit = 1; | |
2686 | p->cluster_next = 1; | |
2687 | p->cluster_nr = 0; | |
2688 | ||
2689 | /* | |
2690 | * Find out how many pages are allowed for a single swap | |
9b15b817 | 2691 | * device. There are two limiting factors: 1) the number |
a2c16d6c HD |
2692 | * of bits for the swap offset in the swp_entry_t type, and |
2693 | * 2) the number of bits in the swap pte as defined by the | |
9b15b817 | 2694 | * different architectures. In order to find the |
a2c16d6c | 2695 | * largest possible bit mask, a swap entry with swap type 0 |
ca8bd38b | 2696 | * and swap offset ~0UL is created, encoded to a swap pte, |
a2c16d6c | 2697 | * decoded to a swp_entry_t again, and finally the swap |
ca8bd38b CEB |
2698 | * offset is extracted. This will mask all the bits from |
2699 | * the initial ~0UL mask that can't be encoded in either | |
2700 | * the swp_entry_t or the architecture definition of a | |
9b15b817 | 2701 | * swap pte. |
ca8bd38b CEB |
2702 | */ |
2703 | maxpages = swp_offset(pte_to_swp_entry( | |
9b15b817 | 2704 | swp_entry_to_pte(swp_entry(0, ~0UL)))) + 1; |
d6bbbd29 RJ |
2705 | last_page = swap_header->info.last_page; |
2706 | if (last_page > maxpages) { | |
465c47fd | 2707 | pr_warn("Truncating oversized swap area, only using %luk out of %luk\n", |
d6bbbd29 RJ |
2708 | maxpages << (PAGE_SHIFT - 10), |
2709 | last_page << (PAGE_SHIFT - 10)); | |
2710 | } | |
2711 | if (maxpages > last_page) { | |
2712 | maxpages = last_page + 1; | |
ca8bd38b CEB |
2713 | /* p->max is an unsigned int: don't overflow it */ |
2714 | if ((unsigned int)maxpages == 0) | |
2715 | maxpages = UINT_MAX; | |
2716 | } | |
2717 | p->highest_bit = maxpages - 1; | |
2718 | ||
2719 | if (!maxpages) | |
38719025 | 2720 | return 0; |
ca8bd38b CEB |
2721 | swapfilepages = i_size_read(inode) >> PAGE_SHIFT; |
2722 | if (swapfilepages && maxpages > swapfilepages) { | |
465c47fd | 2723 | pr_warn("Swap area shorter than signature indicates\n"); |
38719025 | 2724 | return 0; |
ca8bd38b CEB |
2725 | } |
2726 | if (swap_header->info.nr_badpages && S_ISREG(inode->i_mode)) | |
38719025 | 2727 | return 0; |
ca8bd38b | 2728 | if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES) |
38719025 | 2729 | return 0; |
ca8bd38b CEB |
2730 | |
2731 | return maxpages; | |
ca8bd38b CEB |
2732 | } |
2733 | ||
4b3ef9da | 2734 | #define SWAP_CLUSTER_INFO_COLS \ |
235b6217 | 2735 | DIV_ROUND_UP(L1_CACHE_BYTES, sizeof(struct swap_cluster_info)) |
4b3ef9da HY |
2736 | #define SWAP_CLUSTER_SPACE_COLS \ |
2737 | DIV_ROUND_UP(SWAP_ADDRESS_SPACE_PAGES, SWAPFILE_CLUSTER) | |
2738 | #define SWAP_CLUSTER_COLS \ | |
2739 | max_t(unsigned int, SWAP_CLUSTER_INFO_COLS, SWAP_CLUSTER_SPACE_COLS) | |
235b6217 | 2740 | |
915d4d7b CEB |
2741 | static int setup_swap_map_and_extents(struct swap_info_struct *p, |
2742 | union swap_header *swap_header, | |
2743 | unsigned char *swap_map, | |
2a8f9449 | 2744 | struct swap_cluster_info *cluster_info, |
915d4d7b CEB |
2745 | unsigned long maxpages, |
2746 | sector_t *span) | |
2747 | { | |
235b6217 | 2748 | unsigned int j, k; |
915d4d7b CEB |
2749 | unsigned int nr_good_pages; |
2750 | int nr_extents; | |
2a8f9449 | 2751 | unsigned long nr_clusters = DIV_ROUND_UP(maxpages, SWAPFILE_CLUSTER); |
235b6217 HY |
2752 | unsigned long col = p->cluster_next / SWAPFILE_CLUSTER % SWAP_CLUSTER_COLS; |
2753 | unsigned long i, idx; | |
915d4d7b CEB |
2754 | |
2755 | nr_good_pages = maxpages - 1; /* omit header page */ | |
2756 | ||
6b534915 HY |
2757 | cluster_list_init(&p->free_clusters); |
2758 | cluster_list_init(&p->discard_clusters); | |
2a8f9449 | 2759 | |
915d4d7b CEB |
2760 | for (i = 0; i < swap_header->info.nr_badpages; i++) { |
2761 | unsigned int page_nr = swap_header->info.badpages[i]; | |
bdb8e3f6 CEB |
2762 | if (page_nr == 0 || page_nr > swap_header->info.last_page) |
2763 | return -EINVAL; | |
915d4d7b CEB |
2764 | if (page_nr < maxpages) { |
2765 | swap_map[page_nr] = SWAP_MAP_BAD; | |
2766 | nr_good_pages--; | |
2a8f9449 SL |
2767 | /* |
2768 | * Haven't marked the cluster free yet, no list | |
2769 | * operation involved | |
2770 | */ | |
2771 | inc_cluster_info_page(p, cluster_info, page_nr); | |
915d4d7b CEB |
2772 | } |
2773 | } | |
2774 | ||
2a8f9449 SL |
2775 | /* Haven't marked the cluster free yet, no list operation involved */ |
2776 | for (i = maxpages; i < round_up(maxpages, SWAPFILE_CLUSTER); i++) | |
2777 | inc_cluster_info_page(p, cluster_info, i); | |
2778 | ||
915d4d7b CEB |
2779 | if (nr_good_pages) { |
2780 | swap_map[0] = SWAP_MAP_BAD; | |
2a8f9449 SL |
2781 | /* |
2782 | * Not mark the cluster free yet, no list | |
2783 | * operation involved | |
2784 | */ | |
2785 | inc_cluster_info_page(p, cluster_info, 0); | |
915d4d7b CEB |
2786 | p->max = maxpages; |
2787 | p->pages = nr_good_pages; | |
2788 | nr_extents = setup_swap_extents(p, span); | |
bdb8e3f6 CEB |
2789 | if (nr_extents < 0) |
2790 | return nr_extents; | |
915d4d7b CEB |
2791 | nr_good_pages = p->pages; |
2792 | } | |
2793 | if (!nr_good_pages) { | |
465c47fd | 2794 | pr_warn("Empty swap-file\n"); |
bdb8e3f6 | 2795 | return -EINVAL; |
915d4d7b CEB |
2796 | } |
2797 | ||
2a8f9449 SL |
2798 | if (!cluster_info) |
2799 | return nr_extents; | |
2800 | ||
235b6217 | 2801 | |
4b3ef9da HY |
2802 | /* |
2803 | * Reduce false cache line sharing between cluster_info and | |
2804 | * sharing same address space. | |
2805 | */ | |
235b6217 HY |
2806 | for (k = 0; k < SWAP_CLUSTER_COLS; k++) { |
2807 | j = (k + col) % SWAP_CLUSTER_COLS; | |
2808 | for (i = 0; i < DIV_ROUND_UP(nr_clusters, SWAP_CLUSTER_COLS); i++) { | |
2809 | idx = i * SWAP_CLUSTER_COLS + j; | |
2810 | if (idx >= nr_clusters) | |
2811 | continue; | |
2812 | if (cluster_count(&cluster_info[idx])) | |
2813 | continue; | |
2a8f9449 | 2814 | cluster_set_flag(&cluster_info[idx], CLUSTER_FLAG_FREE); |
6b534915 HY |
2815 | cluster_list_add_tail(&p->free_clusters, cluster_info, |
2816 | idx); | |
2a8f9449 | 2817 | } |
2a8f9449 | 2818 | } |
915d4d7b | 2819 | return nr_extents; |
915d4d7b CEB |
2820 | } |
2821 | ||
dcf6b7dd RA |
2822 | /* |
2823 | * Helper to sys_swapon determining if a given swap | |
2824 | * backing device queue supports DISCARD operations. | |
2825 | */ | |
2826 | static bool swap_discardable(struct swap_info_struct *si) | |
2827 | { | |
2828 | struct request_queue *q = bdev_get_queue(si->bdev); | |
2829 | ||
2830 | if (!q || !blk_queue_discard(q)) | |
2831 | return false; | |
2832 | ||
2833 | return true; | |
2834 | } | |
2835 | ||
53cbb243 CEB |
2836 | SYSCALL_DEFINE2(swapon, const char __user *, specialfile, int, swap_flags) |
2837 | { | |
2838 | struct swap_info_struct *p; | |
91a27b2a | 2839 | struct filename *name; |
53cbb243 CEB |
2840 | struct file *swap_file = NULL; |
2841 | struct address_space *mapping; | |
40531542 | 2842 | int prio; |
53cbb243 CEB |
2843 | int error; |
2844 | union swap_header *swap_header; | |
915d4d7b | 2845 | int nr_extents; |
53cbb243 CEB |
2846 | sector_t span; |
2847 | unsigned long maxpages; | |
53cbb243 | 2848 | unsigned char *swap_map = NULL; |
2a8f9449 | 2849 | struct swap_cluster_info *cluster_info = NULL; |
38b5faf4 | 2850 | unsigned long *frontswap_map = NULL; |
53cbb243 CEB |
2851 | struct page *page = NULL; |
2852 | struct inode *inode = NULL; | |
53cbb243 | 2853 | |
d15cab97 HD |
2854 | if (swap_flags & ~SWAP_FLAGS_VALID) |
2855 | return -EINVAL; | |
2856 | ||
53cbb243 CEB |
2857 | if (!capable(CAP_SYS_ADMIN)) |
2858 | return -EPERM; | |
2859 | ||
2860 | p = alloc_swap_info(); | |
2542e513 CEB |
2861 | if (IS_ERR(p)) |
2862 | return PTR_ERR(p); | |
53cbb243 | 2863 | |
815c2c54 SL |
2864 | INIT_WORK(&p->discard_work, swap_discard_work); |
2865 | ||
1da177e4 | 2866 | name = getname(specialfile); |
1da177e4 | 2867 | if (IS_ERR(name)) { |
7de7fb6b | 2868 | error = PTR_ERR(name); |
1da177e4 | 2869 | name = NULL; |
bd69010b | 2870 | goto bad_swap; |
1da177e4 | 2871 | } |
669abf4e | 2872 | swap_file = file_open_name(name, O_RDWR|O_LARGEFILE, 0); |
1da177e4 | 2873 | if (IS_ERR(swap_file)) { |
7de7fb6b | 2874 | error = PTR_ERR(swap_file); |
1da177e4 | 2875 | swap_file = NULL; |
bd69010b | 2876 | goto bad_swap; |
1da177e4 LT |
2877 | } |
2878 | ||
2879 | p->swap_file = swap_file; | |
2880 | mapping = swap_file->f_mapping; | |
2130781e | 2881 | inode = mapping->host; |
6f179af8 | 2882 | |
5955102c | 2883 | /* If S_ISREG(inode->i_mode) will do inode_lock(inode); */ |
4d0e1e10 CEB |
2884 | error = claim_swapfile(p, inode); |
2885 | if (unlikely(error)) | |
1da177e4 | 2886 | goto bad_swap; |
1da177e4 | 2887 | |
1da177e4 LT |
2888 | /* |
2889 | * Read the swap header. | |
2890 | */ | |
2891 | if (!mapping->a_ops->readpage) { | |
2892 | error = -EINVAL; | |
2893 | goto bad_swap; | |
2894 | } | |
090d2b18 | 2895 | page = read_mapping_page(mapping, 0, swap_file); |
1da177e4 LT |
2896 | if (IS_ERR(page)) { |
2897 | error = PTR_ERR(page); | |
2898 | goto bad_swap; | |
2899 | } | |
81e33971 | 2900 | swap_header = kmap(page); |
1da177e4 | 2901 | |
ca8bd38b CEB |
2902 | maxpages = read_swap_header(p, swap_header, inode); |
2903 | if (unlikely(!maxpages)) { | |
1da177e4 LT |
2904 | error = -EINVAL; |
2905 | goto bad_swap; | |
2906 | } | |
886bb7e9 | 2907 | |
81e33971 | 2908 | /* OK, set up the swap map and apply the bad block list */ |
803d0c83 | 2909 | swap_map = vzalloc(maxpages); |
81e33971 HD |
2910 | if (!swap_map) { |
2911 | error = -ENOMEM; | |
2912 | goto bad_swap; | |
2913 | } | |
f0571429 MK |
2914 | |
2915 | if (bdi_cap_stable_pages_required(inode_to_bdi(inode))) | |
2916 | p->flags |= SWP_STABLE_WRITES; | |
2917 | ||
2a8f9449 | 2918 | if (p->bdev && blk_queue_nonrot(bdev_get_queue(p->bdev))) { |
6f179af8 | 2919 | int cpu; |
235b6217 | 2920 | unsigned long ci, nr_cluster; |
6f179af8 | 2921 | |
2a8f9449 SL |
2922 | p->flags |= SWP_SOLIDSTATE; |
2923 | /* | |
2924 | * select a random position to start with to help wear leveling | |
2925 | * SSD | |
2926 | */ | |
2927 | p->cluster_next = 1 + (prandom_u32() % p->highest_bit); | |
235b6217 | 2928 | nr_cluster = DIV_ROUND_UP(maxpages, SWAPFILE_CLUSTER); |
2a8f9449 | 2929 | |
54f180d3 HY |
2930 | cluster_info = kvzalloc(nr_cluster * sizeof(*cluster_info), |
2931 | GFP_KERNEL); | |
2a8f9449 SL |
2932 | if (!cluster_info) { |
2933 | error = -ENOMEM; | |
2934 | goto bad_swap; | |
2935 | } | |
235b6217 HY |
2936 | |
2937 | for (ci = 0; ci < nr_cluster; ci++) | |
2938 | spin_lock_init(&((cluster_info + ci)->lock)); | |
2939 | ||
ebc2a1a6 SL |
2940 | p->percpu_cluster = alloc_percpu(struct percpu_cluster); |
2941 | if (!p->percpu_cluster) { | |
2942 | error = -ENOMEM; | |
2943 | goto bad_swap; | |
2944 | } | |
6f179af8 | 2945 | for_each_possible_cpu(cpu) { |
ebc2a1a6 | 2946 | struct percpu_cluster *cluster; |
6f179af8 | 2947 | cluster = per_cpu_ptr(p->percpu_cluster, cpu); |
ebc2a1a6 SL |
2948 | cluster_set_null(&cluster->index); |
2949 | } | |
2a8f9449 | 2950 | } |
1da177e4 | 2951 | |
1421ef3c CEB |
2952 | error = swap_cgroup_swapon(p->type, maxpages); |
2953 | if (error) | |
2954 | goto bad_swap; | |
2955 | ||
915d4d7b | 2956 | nr_extents = setup_swap_map_and_extents(p, swap_header, swap_map, |
2a8f9449 | 2957 | cluster_info, maxpages, &span); |
915d4d7b CEB |
2958 | if (unlikely(nr_extents < 0)) { |
2959 | error = nr_extents; | |
1da177e4 LT |
2960 | goto bad_swap; |
2961 | } | |
38b5faf4 | 2962 | /* frontswap enabled? set up bit-per-page map for frontswap */ |
8ea1d2a1 | 2963 | if (IS_ENABLED(CONFIG_FRONTSWAP)) |
54f180d3 HY |
2964 | frontswap_map = kvzalloc(BITS_TO_LONGS(maxpages) * sizeof(long), |
2965 | GFP_KERNEL); | |
1da177e4 | 2966 | |
2a8f9449 SL |
2967 | if (p->bdev &&(swap_flags & SWAP_FLAG_DISCARD) && swap_discardable(p)) { |
2968 | /* | |
2969 | * When discard is enabled for swap with no particular | |
2970 | * policy flagged, we set all swap discard flags here in | |
2971 | * order to sustain backward compatibility with older | |
2972 | * swapon(8) releases. | |
2973 | */ | |
2974 | p->flags |= (SWP_DISCARDABLE | SWP_AREA_DISCARD | | |
2975 | SWP_PAGE_DISCARD); | |
dcf6b7dd | 2976 | |
2a8f9449 SL |
2977 | /* |
2978 | * By flagging sys_swapon, a sysadmin can tell us to | |
2979 | * either do single-time area discards only, or to just | |
2980 | * perform discards for released swap page-clusters. | |
2981 | * Now it's time to adjust the p->flags accordingly. | |
2982 | */ | |
2983 | if (swap_flags & SWAP_FLAG_DISCARD_ONCE) | |
2984 | p->flags &= ~SWP_PAGE_DISCARD; | |
2985 | else if (swap_flags & SWAP_FLAG_DISCARD_PAGES) | |
2986 | p->flags &= ~SWP_AREA_DISCARD; | |
2987 | ||
2988 | /* issue a swapon-time discard if it's still required */ | |
2989 | if (p->flags & SWP_AREA_DISCARD) { | |
2990 | int err = discard_swap(p); | |
2991 | if (unlikely(err)) | |
2992 | pr_err("swapon: discard_swap(%p): %d\n", | |
2993 | p, err); | |
dcf6b7dd | 2994 | } |
20137a49 | 2995 | } |
6a6ba831 | 2996 | |
4b3ef9da HY |
2997 | error = init_swap_address_space(p->type, maxpages); |
2998 | if (error) | |
2999 | goto bad_swap; | |
3000 | ||
fc0abb14 | 3001 | mutex_lock(&swapon_mutex); |
40531542 | 3002 | prio = -1; |
78ecba08 | 3003 | if (swap_flags & SWAP_FLAG_PREFER) |
40531542 | 3004 | prio = |
78ecba08 | 3005 | (swap_flags & SWAP_FLAG_PRIO_MASK) >> SWAP_FLAG_PRIO_SHIFT; |
2a8f9449 | 3006 | enable_swap_info(p, prio, swap_map, cluster_info, frontswap_map); |
c69dbfb8 | 3007 | |
756a025f | 3008 | pr_info("Adding %uk swap on %s. Priority:%d extents:%d across:%lluk %s%s%s%s%s\n", |
91a27b2a | 3009 | p->pages<<(PAGE_SHIFT-10), name->name, p->prio, |
c69dbfb8 CEB |
3010 | nr_extents, (unsigned long long)span<<(PAGE_SHIFT-10), |
3011 | (p->flags & SWP_SOLIDSTATE) ? "SS" : "", | |
38b5faf4 | 3012 | (p->flags & SWP_DISCARDABLE) ? "D" : "", |
dcf6b7dd RA |
3013 | (p->flags & SWP_AREA_DISCARD) ? "s" : "", |
3014 | (p->flags & SWP_PAGE_DISCARD) ? "c" : "", | |
38b5faf4 | 3015 | (frontswap_map) ? "FS" : ""); |
c69dbfb8 | 3016 | |
fc0abb14 | 3017 | mutex_unlock(&swapon_mutex); |
66d7dd51 KS |
3018 | atomic_inc(&proc_poll_event); |
3019 | wake_up_interruptible(&proc_poll_wait); | |
3020 | ||
9b01c350 CEB |
3021 | if (S_ISREG(inode->i_mode)) |
3022 | inode->i_flags |= S_SWAPFILE; | |
1da177e4 LT |
3023 | error = 0; |
3024 | goto out; | |
3025 | bad_swap: | |
ebc2a1a6 SL |
3026 | free_percpu(p->percpu_cluster); |
3027 | p->percpu_cluster = NULL; | |
bd69010b | 3028 | if (inode && S_ISBLK(inode->i_mode) && p->bdev) { |
f2090d2d CEB |
3029 | set_blocksize(p->bdev, p->old_block_size); |
3030 | blkdev_put(p->bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL); | |
1da177e4 | 3031 | } |
4cd3bb10 | 3032 | destroy_swap_extents(p); |
e8e6c2ec | 3033 | swap_cgroup_swapoff(p->type); |
5d337b91 | 3034 | spin_lock(&swap_lock); |
1da177e4 | 3035 | p->swap_file = NULL; |
1da177e4 | 3036 | p->flags = 0; |
5d337b91 | 3037 | spin_unlock(&swap_lock); |
1da177e4 | 3038 | vfree(swap_map); |
2a8f9449 | 3039 | vfree(cluster_info); |
52c50567 | 3040 | if (swap_file) { |
2130781e | 3041 | if (inode && S_ISREG(inode->i_mode)) { |
5955102c | 3042 | inode_unlock(inode); |
2130781e CEB |
3043 | inode = NULL; |
3044 | } | |
1da177e4 | 3045 | filp_close(swap_file, NULL); |
52c50567 | 3046 | } |
1da177e4 LT |
3047 | out: |
3048 | if (page && !IS_ERR(page)) { | |
3049 | kunmap(page); | |
09cbfeaf | 3050 | put_page(page); |
1da177e4 LT |
3051 | } |
3052 | if (name) | |
3053 | putname(name); | |
9b01c350 | 3054 | if (inode && S_ISREG(inode->i_mode)) |
5955102c | 3055 | inode_unlock(inode); |
039939a6 TC |
3056 | if (!error) |
3057 | enable_swap_slots_cache(); | |
1da177e4 LT |
3058 | return error; |
3059 | } | |
3060 | ||
3061 | void si_swapinfo(struct sysinfo *val) | |
3062 | { | |
efa90a98 | 3063 | unsigned int type; |
1da177e4 LT |
3064 | unsigned long nr_to_be_unused = 0; |
3065 | ||
5d337b91 | 3066 | spin_lock(&swap_lock); |
efa90a98 HD |
3067 | for (type = 0; type < nr_swapfiles; type++) { |
3068 | struct swap_info_struct *si = swap_info[type]; | |
3069 | ||
3070 | if ((si->flags & SWP_USED) && !(si->flags & SWP_WRITEOK)) | |
3071 | nr_to_be_unused += si->inuse_pages; | |
1da177e4 | 3072 | } |
ec8acf20 | 3073 | val->freeswap = atomic_long_read(&nr_swap_pages) + nr_to_be_unused; |
1da177e4 | 3074 | val->totalswap = total_swap_pages + nr_to_be_unused; |
5d337b91 | 3075 | spin_unlock(&swap_lock); |
1da177e4 LT |
3076 | } |
3077 | ||
3078 | /* | |
3079 | * Verify that a swap entry is valid and increment its swap map count. | |
3080 | * | |
355cfa73 KH |
3081 | * Returns error code in following case. |
3082 | * - success -> 0 | |
3083 | * - swp_entry is invalid -> EINVAL | |
3084 | * - swp_entry is migration entry -> EINVAL | |
3085 | * - swap-cache reference is requested but there is already one. -> EEXIST | |
3086 | * - swap-cache reference is requested but the entry is not used. -> ENOENT | |
570a335b | 3087 | * - swap-mapped reference requested but needs continued swap count. -> ENOMEM |
1da177e4 | 3088 | */ |
8d69aaee | 3089 | static int __swap_duplicate(swp_entry_t entry, unsigned char usage) |
1da177e4 | 3090 | { |
73c34b6a | 3091 | struct swap_info_struct *p; |
235b6217 | 3092 | struct swap_cluster_info *ci; |
1da177e4 | 3093 | unsigned long offset, type; |
8d69aaee HD |
3094 | unsigned char count; |
3095 | unsigned char has_cache; | |
253d553b | 3096 | int err = -EINVAL; |
1da177e4 | 3097 | |
a7420aa5 | 3098 | if (non_swap_entry(entry)) |
253d553b | 3099 | goto out; |
0697212a | 3100 | |
1da177e4 LT |
3101 | type = swp_type(entry); |
3102 | if (type >= nr_swapfiles) | |
3103 | goto bad_file; | |
efa90a98 | 3104 | p = swap_info[type]; |
1da177e4 | 3105 | offset = swp_offset(entry); |
355cfa73 | 3106 | if (unlikely(offset >= p->max)) |
235b6217 HY |
3107 | goto out; |
3108 | ||
3109 | ci = lock_cluster_or_swap_info(p, offset); | |
355cfa73 | 3110 | |
253d553b | 3111 | count = p->swap_map[offset]; |
edfe23da SL |
3112 | |
3113 | /* | |
3114 | * swapin_readahead() doesn't check if a swap entry is valid, so the | |
3115 | * swap entry could be SWAP_MAP_BAD. Check here with lock held. | |
3116 | */ | |
3117 | if (unlikely(swap_count(count) == SWAP_MAP_BAD)) { | |
3118 | err = -ENOENT; | |
3119 | goto unlock_out; | |
3120 | } | |
3121 | ||
253d553b HD |
3122 | has_cache = count & SWAP_HAS_CACHE; |
3123 | count &= ~SWAP_HAS_CACHE; | |
3124 | err = 0; | |
355cfa73 | 3125 | |
253d553b | 3126 | if (usage == SWAP_HAS_CACHE) { |
355cfa73 KH |
3127 | |
3128 | /* set SWAP_HAS_CACHE if there is no cache and entry is used */ | |
253d553b HD |
3129 | if (!has_cache && count) |
3130 | has_cache = SWAP_HAS_CACHE; | |
3131 | else if (has_cache) /* someone else added cache */ | |
3132 | err = -EEXIST; | |
3133 | else /* no users remaining */ | |
3134 | err = -ENOENT; | |
355cfa73 KH |
3135 | |
3136 | } else if (count || has_cache) { | |
253d553b | 3137 | |
570a335b HD |
3138 | if ((count & ~COUNT_CONTINUED) < SWAP_MAP_MAX) |
3139 | count += usage; | |
3140 | else if ((count & ~COUNT_CONTINUED) > SWAP_MAP_MAX) | |
253d553b | 3141 | err = -EINVAL; |
570a335b HD |
3142 | else if (swap_count_continued(p, offset, count)) |
3143 | count = COUNT_CONTINUED; | |
3144 | else | |
3145 | err = -ENOMEM; | |
355cfa73 | 3146 | } else |
253d553b HD |
3147 | err = -ENOENT; /* unused swap entry */ |
3148 | ||
3149 | p->swap_map[offset] = count | has_cache; | |
3150 | ||
355cfa73 | 3151 | unlock_out: |
235b6217 | 3152 | unlock_cluster_or_swap_info(p, ci); |
1da177e4 | 3153 | out: |
253d553b | 3154 | return err; |
1da177e4 LT |
3155 | |
3156 | bad_file: | |
465c47fd | 3157 | pr_err("swap_dup: %s%08lx\n", Bad_file, entry.val); |
1da177e4 LT |
3158 | goto out; |
3159 | } | |
253d553b | 3160 | |
aaa46865 HD |
3161 | /* |
3162 | * Help swapoff by noting that swap entry belongs to shmem/tmpfs | |
3163 | * (in which case its reference count is never incremented). | |
3164 | */ | |
3165 | void swap_shmem_alloc(swp_entry_t entry) | |
3166 | { | |
3167 | __swap_duplicate(entry, SWAP_MAP_SHMEM); | |
3168 | } | |
3169 | ||
355cfa73 | 3170 | /* |
08259d58 HD |
3171 | * Increase reference count of swap entry by 1. |
3172 | * Returns 0 for success, or -ENOMEM if a swap_count_continuation is required | |
3173 | * but could not be atomically allocated. Returns 0, just as if it succeeded, | |
3174 | * if __swap_duplicate() fails for another reason (-EINVAL or -ENOENT), which | |
3175 | * might occur if a page table entry has got corrupted. | |
355cfa73 | 3176 | */ |
570a335b | 3177 | int swap_duplicate(swp_entry_t entry) |
355cfa73 | 3178 | { |
570a335b HD |
3179 | int err = 0; |
3180 | ||
3181 | while (!err && __swap_duplicate(entry, 1) == -ENOMEM) | |
3182 | err = add_swap_count_continuation(entry, GFP_ATOMIC); | |
3183 | return err; | |
355cfa73 | 3184 | } |
1da177e4 | 3185 | |
cb4b86ba | 3186 | /* |
355cfa73 KH |
3187 | * @entry: swap entry for which we allocate swap cache. |
3188 | * | |
73c34b6a | 3189 | * Called when allocating swap cache for existing swap entry, |
355cfa73 KH |
3190 | * This can return error codes. Returns 0 at success. |
3191 | * -EBUSY means there is a swap cache. | |
3192 | * Note: return code is different from swap_duplicate(). | |
cb4b86ba KH |
3193 | */ |
3194 | int swapcache_prepare(swp_entry_t entry) | |
3195 | { | |
253d553b | 3196 | return __swap_duplicate(entry, SWAP_HAS_CACHE); |
cb4b86ba KH |
3197 | } |
3198 | ||
f981c595 MG |
3199 | struct swap_info_struct *page_swap_info(struct page *page) |
3200 | { | |
3201 | swp_entry_t swap = { .val = page_private(page) }; | |
f981c595 MG |
3202 | return swap_info[swp_type(swap)]; |
3203 | } | |
3204 | ||
3205 | /* | |
3206 | * out-of-line __page_file_ methods to avoid include hell. | |
3207 | */ | |
3208 | struct address_space *__page_file_mapping(struct page *page) | |
3209 | { | |
309381fe | 3210 | VM_BUG_ON_PAGE(!PageSwapCache(page), page); |
f981c595 MG |
3211 | return page_swap_info(page)->swap_file->f_mapping; |
3212 | } | |
3213 | EXPORT_SYMBOL_GPL(__page_file_mapping); | |
3214 | ||
3215 | pgoff_t __page_file_index(struct page *page) | |
3216 | { | |
3217 | swp_entry_t swap = { .val = page_private(page) }; | |
309381fe | 3218 | VM_BUG_ON_PAGE(!PageSwapCache(page), page); |
f981c595 MG |
3219 | return swp_offset(swap); |
3220 | } | |
3221 | EXPORT_SYMBOL_GPL(__page_file_index); | |
3222 | ||
570a335b HD |
3223 | /* |
3224 | * add_swap_count_continuation - called when a swap count is duplicated | |
3225 | * beyond SWAP_MAP_MAX, it allocates a new page and links that to the entry's | |
3226 | * page of the original vmalloc'ed swap_map, to hold the continuation count | |
3227 | * (for that entry and for its neighbouring PAGE_SIZE swap entries). Called | |
3228 | * again when count is duplicated beyond SWAP_MAP_MAX * SWAP_CONT_MAX, etc. | |
3229 | * | |
3230 | * These continuation pages are seldom referenced: the common paths all work | |
3231 | * on the original swap_map, only referring to a continuation page when the | |
3232 | * low "digit" of a count is incremented or decremented through SWAP_MAP_MAX. | |
3233 | * | |
3234 | * add_swap_count_continuation(, GFP_ATOMIC) can be called while holding | |
3235 | * page table locks; if it fails, add_swap_count_continuation(, GFP_KERNEL) | |
3236 | * can be called after dropping locks. | |
3237 | */ | |
3238 | int add_swap_count_continuation(swp_entry_t entry, gfp_t gfp_mask) | |
3239 | { | |
3240 | struct swap_info_struct *si; | |
235b6217 | 3241 | struct swap_cluster_info *ci; |
570a335b HD |
3242 | struct page *head; |
3243 | struct page *page; | |
3244 | struct page *list_page; | |
3245 | pgoff_t offset; | |
3246 | unsigned char count; | |
3247 | ||
3248 | /* | |
3249 | * When debugging, it's easier to use __GFP_ZERO here; but it's better | |
3250 | * for latency not to zero a page while GFP_ATOMIC and holding locks. | |
3251 | */ | |
3252 | page = alloc_page(gfp_mask | __GFP_HIGHMEM); | |
3253 | ||
3254 | si = swap_info_get(entry); | |
3255 | if (!si) { | |
3256 | /* | |
3257 | * An acceptable race has occurred since the failing | |
3258 | * __swap_duplicate(): the swap entry has been freed, | |
3259 | * perhaps even the whole swap_map cleared for swapoff. | |
3260 | */ | |
3261 | goto outer; | |
3262 | } | |
3263 | ||
3264 | offset = swp_offset(entry); | |
235b6217 HY |
3265 | |
3266 | ci = lock_cluster(si, offset); | |
3267 | ||
570a335b HD |
3268 | count = si->swap_map[offset] & ~SWAP_HAS_CACHE; |
3269 | ||
3270 | if ((count & ~COUNT_CONTINUED) != SWAP_MAP_MAX) { | |
3271 | /* | |
3272 | * The higher the swap count, the more likely it is that tasks | |
3273 | * will race to add swap count continuation: we need to avoid | |
3274 | * over-provisioning. | |
3275 | */ | |
3276 | goto out; | |
3277 | } | |
3278 | ||
3279 | if (!page) { | |
235b6217 | 3280 | unlock_cluster(ci); |
ec8acf20 | 3281 | spin_unlock(&si->lock); |
570a335b HD |
3282 | return -ENOMEM; |
3283 | } | |
3284 | ||
3285 | /* | |
3286 | * We are fortunate that although vmalloc_to_page uses pte_offset_map, | |
2de1a7e4 SJ |
3287 | * no architecture is using highmem pages for kernel page tables: so it |
3288 | * will not corrupt the GFP_ATOMIC caller's atomic page table kmaps. | |
570a335b HD |
3289 | */ |
3290 | head = vmalloc_to_page(si->swap_map + offset); | |
3291 | offset &= ~PAGE_MASK; | |
3292 | ||
3293 | /* | |
3294 | * Page allocation does not initialize the page's lru field, | |
3295 | * but it does always reset its private field. | |
3296 | */ | |
3297 | if (!page_private(head)) { | |
3298 | BUG_ON(count & COUNT_CONTINUED); | |
3299 | INIT_LIST_HEAD(&head->lru); | |
3300 | set_page_private(head, SWP_CONTINUED); | |
3301 | si->flags |= SWP_CONTINUED; | |
3302 | } | |
3303 | ||
3304 | list_for_each_entry(list_page, &head->lru, lru) { | |
3305 | unsigned char *map; | |
3306 | ||
3307 | /* | |
3308 | * If the previous map said no continuation, but we've found | |
3309 | * a continuation page, free our allocation and use this one. | |
3310 | */ | |
3311 | if (!(count & COUNT_CONTINUED)) | |
3312 | goto out; | |
3313 | ||
9b04c5fe | 3314 | map = kmap_atomic(list_page) + offset; |
570a335b | 3315 | count = *map; |
9b04c5fe | 3316 | kunmap_atomic(map); |
570a335b HD |
3317 | |
3318 | /* | |
3319 | * If this continuation count now has some space in it, | |
3320 | * free our allocation and use this one. | |
3321 | */ | |
3322 | if ((count & ~COUNT_CONTINUED) != SWAP_CONT_MAX) | |
3323 | goto out; | |
3324 | } | |
3325 | ||
3326 | list_add_tail(&page->lru, &head->lru); | |
3327 | page = NULL; /* now it's attached, don't free it */ | |
3328 | out: | |
235b6217 | 3329 | unlock_cluster(ci); |
ec8acf20 | 3330 | spin_unlock(&si->lock); |
570a335b HD |
3331 | outer: |
3332 | if (page) | |
3333 | __free_page(page); | |
3334 | return 0; | |
3335 | } | |
3336 | ||
3337 | /* | |
3338 | * swap_count_continued - when the original swap_map count is incremented | |
3339 | * from SWAP_MAP_MAX, check if there is already a continuation page to carry | |
3340 | * into, carry if so, or else fail until a new continuation page is allocated; | |
3341 | * when the original swap_map count is decremented from 0 with continuation, | |
3342 | * borrow from the continuation and report whether it still holds more. | |
235b6217 HY |
3343 | * Called while __swap_duplicate() or swap_entry_free() holds swap or cluster |
3344 | * lock. | |
570a335b HD |
3345 | */ |
3346 | static bool swap_count_continued(struct swap_info_struct *si, | |
3347 | pgoff_t offset, unsigned char count) | |
3348 | { | |
3349 | struct page *head; | |
3350 | struct page *page; | |
3351 | unsigned char *map; | |
3352 | ||
3353 | head = vmalloc_to_page(si->swap_map + offset); | |
3354 | if (page_private(head) != SWP_CONTINUED) { | |
3355 | BUG_ON(count & COUNT_CONTINUED); | |
3356 | return false; /* need to add count continuation */ | |
3357 | } | |
3358 | ||
3359 | offset &= ~PAGE_MASK; | |
3360 | page = list_entry(head->lru.next, struct page, lru); | |
9b04c5fe | 3361 | map = kmap_atomic(page) + offset; |
570a335b HD |
3362 | |
3363 | if (count == SWAP_MAP_MAX) /* initial increment from swap_map */ | |
3364 | goto init_map; /* jump over SWAP_CONT_MAX checks */ | |
3365 | ||
3366 | if (count == (SWAP_MAP_MAX | COUNT_CONTINUED)) { /* incrementing */ | |
3367 | /* | |
3368 | * Think of how you add 1 to 999 | |
3369 | */ | |
3370 | while (*map == (SWAP_CONT_MAX | COUNT_CONTINUED)) { | |
9b04c5fe | 3371 | kunmap_atomic(map); |
570a335b HD |
3372 | page = list_entry(page->lru.next, struct page, lru); |
3373 | BUG_ON(page == head); | |
9b04c5fe | 3374 | map = kmap_atomic(page) + offset; |
570a335b HD |
3375 | } |
3376 | if (*map == SWAP_CONT_MAX) { | |
9b04c5fe | 3377 | kunmap_atomic(map); |
570a335b HD |
3378 | page = list_entry(page->lru.next, struct page, lru); |
3379 | if (page == head) | |
3380 | return false; /* add count continuation */ | |
9b04c5fe | 3381 | map = kmap_atomic(page) + offset; |
570a335b HD |
3382 | init_map: *map = 0; /* we didn't zero the page */ |
3383 | } | |
3384 | *map += 1; | |
9b04c5fe | 3385 | kunmap_atomic(map); |
570a335b HD |
3386 | page = list_entry(page->lru.prev, struct page, lru); |
3387 | while (page != head) { | |
9b04c5fe | 3388 | map = kmap_atomic(page) + offset; |
570a335b | 3389 | *map = COUNT_CONTINUED; |
9b04c5fe | 3390 | kunmap_atomic(map); |
570a335b HD |
3391 | page = list_entry(page->lru.prev, struct page, lru); |
3392 | } | |
3393 | return true; /* incremented */ | |
3394 | ||
3395 | } else { /* decrementing */ | |
3396 | /* | |
3397 | * Think of how you subtract 1 from 1000 | |
3398 | */ | |
3399 | BUG_ON(count != COUNT_CONTINUED); | |
3400 | while (*map == COUNT_CONTINUED) { | |
9b04c5fe | 3401 | kunmap_atomic(map); |
570a335b HD |
3402 | page = list_entry(page->lru.next, struct page, lru); |
3403 | BUG_ON(page == head); | |
9b04c5fe | 3404 | map = kmap_atomic(page) + offset; |
570a335b HD |
3405 | } |
3406 | BUG_ON(*map == 0); | |
3407 | *map -= 1; | |
3408 | if (*map == 0) | |
3409 | count = 0; | |
9b04c5fe | 3410 | kunmap_atomic(map); |
570a335b HD |
3411 | page = list_entry(page->lru.prev, struct page, lru); |
3412 | while (page != head) { | |
9b04c5fe | 3413 | map = kmap_atomic(page) + offset; |
570a335b HD |
3414 | *map = SWAP_CONT_MAX | count; |
3415 | count = COUNT_CONTINUED; | |
9b04c5fe | 3416 | kunmap_atomic(map); |
570a335b HD |
3417 | page = list_entry(page->lru.prev, struct page, lru); |
3418 | } | |
3419 | return count == COUNT_CONTINUED; | |
3420 | } | |
3421 | } | |
3422 | ||
3423 | /* | |
3424 | * free_swap_count_continuations - swapoff free all the continuation pages | |
3425 | * appended to the swap_map, after swap_map is quiesced, before vfree'ing it. | |
3426 | */ | |
3427 | static void free_swap_count_continuations(struct swap_info_struct *si) | |
3428 | { | |
3429 | pgoff_t offset; | |
3430 | ||
3431 | for (offset = 0; offset < si->max; offset += PAGE_SIZE) { | |
3432 | struct page *head; | |
3433 | head = vmalloc_to_page(si->swap_map + offset); | |
3434 | if (page_private(head)) { | |
0d576d20 GT |
3435 | struct page *page, *next; |
3436 | ||
3437 | list_for_each_entry_safe(page, next, &head->lru, lru) { | |
3438 | list_del(&page->lru); | |
570a335b HD |
3439 | __free_page(page); |
3440 | } | |
3441 | } | |
3442 | } | |
3443 | } |