]>
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 LT |
8 | #include <linux/mm.h> |
9 | #include <linux/hugetlb.h> | |
10 | #include <linux/mman.h> | |
11 | #include <linux/slab.h> | |
12 | #include <linux/kernel_stat.h> | |
13 | #include <linux/swap.h> | |
14 | #include <linux/vmalloc.h> | |
15 | #include <linux/pagemap.h> | |
16 | #include <linux/namei.h> | |
072441e2 | 17 | #include <linux/shmem_fs.h> |
1da177e4 | 18 | #include <linux/blkdev.h> |
20137a49 | 19 | #include <linux/random.h> |
1da177e4 LT |
20 | #include <linux/writeback.h> |
21 | #include <linux/proc_fs.h> | |
22 | #include <linux/seq_file.h> | |
23 | #include <linux/init.h> | |
5ad64688 | 24 | #include <linux/ksm.h> |
1da177e4 LT |
25 | #include <linux/rmap.h> |
26 | #include <linux/security.h> | |
27 | #include <linux/backing-dev.h> | |
fc0abb14 | 28 | #include <linux/mutex.h> |
c59ede7b | 29 | #include <linux/capability.h> |
1da177e4 | 30 | #include <linux/syscalls.h> |
8a9f3ccd | 31 | #include <linux/memcontrol.h> |
66d7dd51 | 32 | #include <linux/poll.h> |
72788c38 | 33 | #include <linux/oom.h> |
38b5faf4 DM |
34 | #include <linux/frontswap.h> |
35 | #include <linux/swapfile.h> | |
f981c595 | 36 | #include <linux/export.h> |
1da177e4 LT |
37 | |
38 | #include <asm/pgtable.h> | |
39 | #include <asm/tlbflush.h> | |
40 | #include <linux/swapops.h> | |
5d1ea48b | 41 | #include <linux/swap_cgroup.h> |
1da177e4 | 42 | |
570a335b HD |
43 | static bool swap_count_continued(struct swap_info_struct *, pgoff_t, |
44 | unsigned char); | |
45 | static void free_swap_count_continuations(struct swap_info_struct *); | |
d4906e1a | 46 | static sector_t map_swap_entry(swp_entry_t, struct block_device**); |
570a335b | 47 | |
38b5faf4 | 48 | DEFINE_SPINLOCK(swap_lock); |
7c363b8c | 49 | static unsigned int nr_swapfiles; |
ec8acf20 | 50 | atomic_long_t nr_swap_pages; |
fb0fec50 CW |
51 | /* |
52 | * Some modules use swappable objects and may try to swap them out under | |
53 | * memory pressure (via the shrinker). Before doing so, they may wish to | |
54 | * check to see if any swap space is available. | |
55 | */ | |
56 | EXPORT_SYMBOL_GPL(nr_swap_pages); | |
ec8acf20 | 57 | /* protected with swap_lock. reading in vm_swap_full() doesn't need lock */ |
1da177e4 | 58 | long total_swap_pages; |
78ecba08 | 59 | static int least_priority; |
1da177e4 | 60 | |
1da177e4 LT |
61 | static const char Bad_file[] = "Bad swap file entry "; |
62 | static const char Unused_file[] = "Unused swap file entry "; | |
63 | static const char Bad_offset[] = "Bad swap offset entry "; | |
64 | static const char Unused_offset[] = "Unused swap offset entry "; | |
65 | ||
adfab836 DS |
66 | /* |
67 | * all active swap_info_structs | |
68 | * protected with swap_lock, and ordered by priority. | |
69 | */ | |
18ab4d4c DS |
70 | PLIST_HEAD(swap_active_head); |
71 | ||
72 | /* | |
73 | * all available (active, not full) swap_info_structs | |
74 | * protected with swap_avail_lock, ordered by priority. | |
75 | * This is used by get_swap_page() instead of swap_active_head | |
76 | * because swap_active_head includes all swap_info_structs, | |
77 | * but get_swap_page() doesn't need to look at full ones. | |
78 | * This uses its own lock instead of swap_lock because when a | |
79 | * swap_info_struct changes between not-full/full, it needs to | |
80 | * add/remove itself to/from this list, but the swap_info_struct->lock | |
81 | * is held and the locking order requires swap_lock to be taken | |
82 | * before any swap_info_struct->lock. | |
83 | */ | |
84 | static PLIST_HEAD(swap_avail_head); | |
85 | static DEFINE_SPINLOCK(swap_avail_lock); | |
1da177e4 | 86 | |
38b5faf4 | 87 | struct swap_info_struct *swap_info[MAX_SWAPFILES]; |
1da177e4 | 88 | |
fc0abb14 | 89 | static DEFINE_MUTEX(swapon_mutex); |
1da177e4 | 90 | |
66d7dd51 KS |
91 | static DECLARE_WAIT_QUEUE_HEAD(proc_poll_wait); |
92 | /* Activity counter to indicate that a swapon or swapoff has occurred */ | |
93 | static atomic_t proc_poll_event = ATOMIC_INIT(0); | |
94 | ||
8d69aaee | 95 | static inline unsigned char swap_count(unsigned char ent) |
355cfa73 | 96 | { |
570a335b | 97 | return ent & ~SWAP_HAS_CACHE; /* may include SWAP_HAS_CONT flag */ |
355cfa73 KH |
98 | } |
99 | ||
efa90a98 | 100 | /* returns 1 if swap entry is freed */ |
c9e44410 KH |
101 | static int |
102 | __try_to_reclaim_swap(struct swap_info_struct *si, unsigned long offset) | |
103 | { | |
efa90a98 | 104 | swp_entry_t entry = swp_entry(si->type, offset); |
c9e44410 KH |
105 | struct page *page; |
106 | int ret = 0; | |
107 | ||
f6ab1f7f | 108 | page = find_get_page(swap_address_space(entry), swp_offset(entry)); |
c9e44410 KH |
109 | if (!page) |
110 | return 0; | |
111 | /* | |
112 | * This function is called from scan_swap_map() and it's called | |
113 | * by vmscan.c at reclaiming pages. So, we hold a lock on a page, here. | |
114 | * We have to use trylock for avoiding deadlock. This is a special | |
115 | * case and you should use try_to_free_swap() with explicit lock_page() | |
116 | * in usual operations. | |
117 | */ | |
118 | if (trylock_page(page)) { | |
119 | ret = try_to_free_swap(page); | |
120 | unlock_page(page); | |
121 | } | |
09cbfeaf | 122 | put_page(page); |
c9e44410 KH |
123 | return ret; |
124 | } | |
355cfa73 | 125 | |
6a6ba831 HD |
126 | /* |
127 | * swapon tell device that all the old swap contents can be discarded, | |
128 | * to allow the swap device to optimize its wear-levelling. | |
129 | */ | |
130 | static int discard_swap(struct swap_info_struct *si) | |
131 | { | |
132 | struct swap_extent *se; | |
9625a5f2 HD |
133 | sector_t start_block; |
134 | sector_t nr_blocks; | |
6a6ba831 HD |
135 | int err = 0; |
136 | ||
9625a5f2 HD |
137 | /* Do not discard the swap header page! */ |
138 | se = &si->first_swap_extent; | |
139 | start_block = (se->start_block + 1) << (PAGE_SHIFT - 9); | |
140 | nr_blocks = ((sector_t)se->nr_pages - 1) << (PAGE_SHIFT - 9); | |
141 | if (nr_blocks) { | |
142 | err = blkdev_issue_discard(si->bdev, start_block, | |
dd3932ed | 143 | nr_blocks, GFP_KERNEL, 0); |
9625a5f2 HD |
144 | if (err) |
145 | return err; | |
146 | cond_resched(); | |
147 | } | |
6a6ba831 | 148 | |
9625a5f2 HD |
149 | list_for_each_entry(se, &si->first_swap_extent.list, list) { |
150 | start_block = se->start_block << (PAGE_SHIFT - 9); | |
151 | nr_blocks = (sector_t)se->nr_pages << (PAGE_SHIFT - 9); | |
6a6ba831 HD |
152 | |
153 | err = blkdev_issue_discard(si->bdev, start_block, | |
dd3932ed | 154 | nr_blocks, GFP_KERNEL, 0); |
6a6ba831 HD |
155 | if (err) |
156 | break; | |
157 | ||
158 | cond_resched(); | |
159 | } | |
160 | return err; /* That will often be -EOPNOTSUPP */ | |
161 | } | |
162 | ||
7992fde7 HD |
163 | /* |
164 | * swap allocation tell device that a cluster of swap can now be discarded, | |
165 | * to allow the swap device to optimize its wear-levelling. | |
166 | */ | |
167 | static void discard_swap_cluster(struct swap_info_struct *si, | |
168 | pgoff_t start_page, pgoff_t nr_pages) | |
169 | { | |
170 | struct swap_extent *se = si->curr_swap_extent; | |
171 | int found_extent = 0; | |
172 | ||
173 | while (nr_pages) { | |
7992fde7 HD |
174 | if (se->start_page <= start_page && |
175 | start_page < se->start_page + se->nr_pages) { | |
176 | pgoff_t offset = start_page - se->start_page; | |
177 | sector_t start_block = se->start_block + offset; | |
858a2990 | 178 | sector_t nr_blocks = se->nr_pages - offset; |
7992fde7 HD |
179 | |
180 | if (nr_blocks > nr_pages) | |
181 | nr_blocks = nr_pages; | |
182 | start_page += nr_blocks; | |
183 | nr_pages -= nr_blocks; | |
184 | ||
185 | if (!found_extent++) | |
186 | si->curr_swap_extent = se; | |
187 | ||
188 | start_block <<= PAGE_SHIFT - 9; | |
189 | nr_blocks <<= PAGE_SHIFT - 9; | |
190 | if (blkdev_issue_discard(si->bdev, start_block, | |
dd3932ed | 191 | nr_blocks, GFP_NOIO, 0)) |
7992fde7 HD |
192 | break; |
193 | } | |
194 | ||
a8ae4991 | 195 | se = list_next_entry(se, list); |
7992fde7 HD |
196 | } |
197 | } | |
198 | ||
048c27fd HD |
199 | #define SWAPFILE_CLUSTER 256 |
200 | #define LATENCY_LIMIT 256 | |
201 | ||
2a8f9449 SL |
202 | static inline void cluster_set_flag(struct swap_cluster_info *info, |
203 | unsigned int flag) | |
204 | { | |
205 | info->flags = flag; | |
206 | } | |
207 | ||
208 | static inline unsigned int cluster_count(struct swap_cluster_info *info) | |
209 | { | |
210 | return info->data; | |
211 | } | |
212 | ||
213 | static inline void cluster_set_count(struct swap_cluster_info *info, | |
214 | unsigned int c) | |
215 | { | |
216 | info->data = c; | |
217 | } | |
218 | ||
219 | static inline void cluster_set_count_flag(struct swap_cluster_info *info, | |
220 | unsigned int c, unsigned int f) | |
221 | { | |
222 | info->flags = f; | |
223 | info->data = c; | |
224 | } | |
225 | ||
226 | static inline unsigned int cluster_next(struct swap_cluster_info *info) | |
227 | { | |
228 | return info->data; | |
229 | } | |
230 | ||
231 | static inline void cluster_set_next(struct swap_cluster_info *info, | |
232 | unsigned int n) | |
233 | { | |
234 | info->data = n; | |
235 | } | |
236 | ||
237 | static inline void cluster_set_next_flag(struct swap_cluster_info *info, | |
238 | unsigned int n, unsigned int f) | |
239 | { | |
240 | info->flags = f; | |
241 | info->data = n; | |
242 | } | |
243 | ||
244 | static inline bool cluster_is_free(struct swap_cluster_info *info) | |
245 | { | |
246 | return info->flags & CLUSTER_FLAG_FREE; | |
247 | } | |
248 | ||
249 | static inline bool cluster_is_null(struct swap_cluster_info *info) | |
250 | { | |
251 | return info->flags & CLUSTER_FLAG_NEXT_NULL; | |
252 | } | |
253 | ||
254 | static inline void cluster_set_null(struct swap_cluster_info *info) | |
255 | { | |
256 | info->flags = CLUSTER_FLAG_NEXT_NULL; | |
257 | info->data = 0; | |
258 | } | |
259 | ||
235b6217 HY |
260 | static inline struct swap_cluster_info *lock_cluster(struct swap_info_struct *si, |
261 | unsigned long offset) | |
262 | { | |
263 | struct swap_cluster_info *ci; | |
264 | ||
265 | ci = si->cluster_info; | |
266 | if (ci) { | |
267 | ci += offset / SWAPFILE_CLUSTER; | |
268 | spin_lock(&ci->lock); | |
269 | } | |
270 | return ci; | |
271 | } | |
272 | ||
273 | static inline void unlock_cluster(struct swap_cluster_info *ci) | |
274 | { | |
275 | if (ci) | |
276 | spin_unlock(&ci->lock); | |
277 | } | |
278 | ||
279 | static inline struct swap_cluster_info *lock_cluster_or_swap_info( | |
280 | struct swap_info_struct *si, | |
281 | unsigned long offset) | |
282 | { | |
283 | struct swap_cluster_info *ci; | |
284 | ||
285 | ci = lock_cluster(si, offset); | |
286 | if (!ci) | |
287 | spin_lock(&si->lock); | |
288 | ||
289 | return ci; | |
290 | } | |
291 | ||
292 | static inline void unlock_cluster_or_swap_info(struct swap_info_struct *si, | |
293 | struct swap_cluster_info *ci) | |
294 | { | |
295 | if (ci) | |
296 | unlock_cluster(ci); | |
297 | else | |
298 | spin_unlock(&si->lock); | |
299 | } | |
300 | ||
6b534915 HY |
301 | static inline bool cluster_list_empty(struct swap_cluster_list *list) |
302 | { | |
303 | return cluster_is_null(&list->head); | |
304 | } | |
305 | ||
306 | static inline unsigned int cluster_list_first(struct swap_cluster_list *list) | |
307 | { | |
308 | return cluster_next(&list->head); | |
309 | } | |
310 | ||
311 | static void cluster_list_init(struct swap_cluster_list *list) | |
312 | { | |
313 | cluster_set_null(&list->head); | |
314 | cluster_set_null(&list->tail); | |
315 | } | |
316 | ||
317 | static void cluster_list_add_tail(struct swap_cluster_list *list, | |
318 | struct swap_cluster_info *ci, | |
319 | unsigned int idx) | |
320 | { | |
321 | if (cluster_list_empty(list)) { | |
322 | cluster_set_next_flag(&list->head, idx, 0); | |
323 | cluster_set_next_flag(&list->tail, idx, 0); | |
324 | } else { | |
235b6217 | 325 | struct swap_cluster_info *ci_tail; |
6b534915 HY |
326 | unsigned int tail = cluster_next(&list->tail); |
327 | ||
235b6217 HY |
328 | /* |
329 | * Nested cluster lock, but both cluster locks are | |
330 | * only acquired when we held swap_info_struct->lock | |
331 | */ | |
332 | ci_tail = ci + tail; | |
333 | spin_lock_nested(&ci_tail->lock, SINGLE_DEPTH_NESTING); | |
334 | cluster_set_next(ci_tail, idx); | |
335 | unlock_cluster(ci_tail); | |
6b534915 HY |
336 | cluster_set_next_flag(&list->tail, idx, 0); |
337 | } | |
338 | } | |
339 | ||
340 | static unsigned int cluster_list_del_first(struct swap_cluster_list *list, | |
341 | struct swap_cluster_info *ci) | |
342 | { | |
343 | unsigned int idx; | |
344 | ||
345 | idx = cluster_next(&list->head); | |
346 | if (cluster_next(&list->tail) == idx) { | |
347 | cluster_set_null(&list->head); | |
348 | cluster_set_null(&list->tail); | |
349 | } else | |
350 | cluster_set_next_flag(&list->head, | |
351 | cluster_next(&ci[idx]), 0); | |
352 | ||
353 | return idx; | |
354 | } | |
355 | ||
815c2c54 SL |
356 | /* Add a cluster to discard list and schedule it to do discard */ |
357 | static void swap_cluster_schedule_discard(struct swap_info_struct *si, | |
358 | unsigned int idx) | |
359 | { | |
360 | /* | |
361 | * If scan_swap_map() can't find a free cluster, it will check | |
362 | * si->swap_map directly. To make sure the discarding cluster isn't | |
363 | * taken by scan_swap_map(), mark the swap entries bad (occupied). It | |
364 | * will be cleared after discard | |
365 | */ | |
366 | memset(si->swap_map + idx * SWAPFILE_CLUSTER, | |
367 | SWAP_MAP_BAD, SWAPFILE_CLUSTER); | |
368 | ||
6b534915 | 369 | cluster_list_add_tail(&si->discard_clusters, si->cluster_info, idx); |
815c2c54 SL |
370 | |
371 | schedule_work(&si->discard_work); | |
372 | } | |
373 | ||
374 | /* | |
375 | * Doing discard actually. After a cluster discard is finished, the cluster | |
376 | * will be added to free cluster list. caller should hold si->lock. | |
377 | */ | |
378 | static void swap_do_scheduled_discard(struct swap_info_struct *si) | |
379 | { | |
235b6217 | 380 | struct swap_cluster_info *info, *ci; |
815c2c54 SL |
381 | unsigned int idx; |
382 | ||
383 | info = si->cluster_info; | |
384 | ||
6b534915 HY |
385 | while (!cluster_list_empty(&si->discard_clusters)) { |
386 | idx = cluster_list_del_first(&si->discard_clusters, info); | |
815c2c54 SL |
387 | spin_unlock(&si->lock); |
388 | ||
389 | discard_swap_cluster(si, idx * SWAPFILE_CLUSTER, | |
390 | SWAPFILE_CLUSTER); | |
391 | ||
392 | spin_lock(&si->lock); | |
235b6217 HY |
393 | ci = lock_cluster(si, idx * SWAPFILE_CLUSTER); |
394 | cluster_set_flag(ci, CLUSTER_FLAG_FREE); | |
395 | unlock_cluster(ci); | |
6b534915 | 396 | cluster_list_add_tail(&si->free_clusters, info, idx); |
235b6217 | 397 | ci = lock_cluster(si, idx * SWAPFILE_CLUSTER); |
815c2c54 SL |
398 | memset(si->swap_map + idx * SWAPFILE_CLUSTER, |
399 | 0, SWAPFILE_CLUSTER); | |
235b6217 | 400 | unlock_cluster(ci); |
815c2c54 SL |
401 | } |
402 | } | |
403 | ||
404 | static void swap_discard_work(struct work_struct *work) | |
405 | { | |
406 | struct swap_info_struct *si; | |
407 | ||
408 | si = container_of(work, struct swap_info_struct, discard_work); | |
409 | ||
410 | spin_lock(&si->lock); | |
411 | swap_do_scheduled_discard(si); | |
412 | spin_unlock(&si->lock); | |
413 | } | |
414 | ||
2a8f9449 SL |
415 | /* |
416 | * The cluster corresponding to page_nr will be used. The cluster will be | |
417 | * removed from free cluster list and its usage counter will be increased. | |
418 | */ | |
419 | static void inc_cluster_info_page(struct swap_info_struct *p, | |
420 | struct swap_cluster_info *cluster_info, unsigned long page_nr) | |
421 | { | |
422 | unsigned long idx = page_nr / SWAPFILE_CLUSTER; | |
423 | ||
424 | if (!cluster_info) | |
425 | return; | |
426 | if (cluster_is_free(&cluster_info[idx])) { | |
6b534915 HY |
427 | VM_BUG_ON(cluster_list_first(&p->free_clusters) != idx); |
428 | cluster_list_del_first(&p->free_clusters, cluster_info); | |
2a8f9449 SL |
429 | cluster_set_count_flag(&cluster_info[idx], 0, 0); |
430 | } | |
431 | ||
432 | VM_BUG_ON(cluster_count(&cluster_info[idx]) >= SWAPFILE_CLUSTER); | |
433 | cluster_set_count(&cluster_info[idx], | |
434 | cluster_count(&cluster_info[idx]) + 1); | |
435 | } | |
436 | ||
437 | /* | |
438 | * The cluster corresponding to page_nr decreases one usage. If the usage | |
439 | * counter becomes 0, which means no page in the cluster is in using, we can | |
440 | * optionally discard the cluster and add it to free cluster list. | |
441 | */ | |
442 | static void dec_cluster_info_page(struct swap_info_struct *p, | |
443 | struct swap_cluster_info *cluster_info, unsigned long page_nr) | |
444 | { | |
445 | unsigned long idx = page_nr / SWAPFILE_CLUSTER; | |
446 | ||
447 | if (!cluster_info) | |
448 | return; | |
449 | ||
450 | VM_BUG_ON(cluster_count(&cluster_info[idx]) == 0); | |
451 | cluster_set_count(&cluster_info[idx], | |
452 | cluster_count(&cluster_info[idx]) - 1); | |
453 | ||
454 | if (cluster_count(&cluster_info[idx]) == 0) { | |
815c2c54 SL |
455 | /* |
456 | * If the swap is discardable, prepare discard the cluster | |
457 | * instead of free it immediately. The cluster will be freed | |
458 | * after discard. | |
459 | */ | |
edfe23da SL |
460 | if ((p->flags & (SWP_WRITEOK | SWP_PAGE_DISCARD)) == |
461 | (SWP_WRITEOK | SWP_PAGE_DISCARD)) { | |
815c2c54 SL |
462 | swap_cluster_schedule_discard(p, idx); |
463 | return; | |
464 | } | |
465 | ||
2a8f9449 | 466 | cluster_set_flag(&cluster_info[idx], CLUSTER_FLAG_FREE); |
6b534915 | 467 | cluster_list_add_tail(&p->free_clusters, cluster_info, idx); |
2a8f9449 SL |
468 | } |
469 | } | |
470 | ||
471 | /* | |
472 | * It's possible scan_swap_map() uses a free cluster in the middle of free | |
473 | * cluster list. Avoiding such abuse to avoid list corruption. | |
474 | */ | |
ebc2a1a6 SL |
475 | static bool |
476 | scan_swap_map_ssd_cluster_conflict(struct swap_info_struct *si, | |
2a8f9449 SL |
477 | unsigned long offset) |
478 | { | |
ebc2a1a6 SL |
479 | struct percpu_cluster *percpu_cluster; |
480 | bool conflict; | |
481 | ||
2a8f9449 | 482 | offset /= SWAPFILE_CLUSTER; |
6b534915 HY |
483 | conflict = !cluster_list_empty(&si->free_clusters) && |
484 | offset != cluster_list_first(&si->free_clusters) && | |
2a8f9449 | 485 | cluster_is_free(&si->cluster_info[offset]); |
ebc2a1a6 SL |
486 | |
487 | if (!conflict) | |
488 | return false; | |
489 | ||
490 | percpu_cluster = this_cpu_ptr(si->percpu_cluster); | |
491 | cluster_set_null(&percpu_cluster->index); | |
492 | return true; | |
493 | } | |
494 | ||
495 | /* | |
496 | * Try to get a swap entry from current cpu's swap entry pool (a cluster). This | |
497 | * might involve allocating a new cluster for current CPU too. | |
498 | */ | |
499 | static void scan_swap_map_try_ssd_cluster(struct swap_info_struct *si, | |
500 | unsigned long *offset, unsigned long *scan_base) | |
501 | { | |
502 | struct percpu_cluster *cluster; | |
235b6217 | 503 | struct swap_cluster_info *ci; |
ebc2a1a6 | 504 | bool found_free; |
235b6217 | 505 | unsigned long tmp, max; |
ebc2a1a6 SL |
506 | |
507 | new_cluster: | |
508 | cluster = this_cpu_ptr(si->percpu_cluster); | |
509 | if (cluster_is_null(&cluster->index)) { | |
6b534915 HY |
510 | if (!cluster_list_empty(&si->free_clusters)) { |
511 | cluster->index = si->free_clusters.head; | |
ebc2a1a6 SL |
512 | cluster->next = cluster_next(&cluster->index) * |
513 | SWAPFILE_CLUSTER; | |
6b534915 | 514 | } else if (!cluster_list_empty(&si->discard_clusters)) { |
ebc2a1a6 SL |
515 | /* |
516 | * we don't have free cluster but have some clusters in | |
517 | * discarding, do discard now and reclaim them | |
518 | */ | |
519 | swap_do_scheduled_discard(si); | |
520 | *scan_base = *offset = si->cluster_next; | |
521 | goto new_cluster; | |
522 | } else | |
523 | return; | |
524 | } | |
525 | ||
526 | found_free = false; | |
527 | ||
528 | /* | |
529 | * Other CPUs can use our cluster if they can't find a free cluster, | |
530 | * check if there is still free entry in the cluster | |
531 | */ | |
532 | tmp = cluster->next; | |
235b6217 HY |
533 | max = min_t(unsigned long, si->max, |
534 | (cluster_next(&cluster->index) + 1) * SWAPFILE_CLUSTER); | |
535 | if (tmp >= max) { | |
536 | cluster_set_null(&cluster->index); | |
537 | goto new_cluster; | |
538 | } | |
539 | ci = lock_cluster(si, tmp); | |
540 | while (tmp < max) { | |
ebc2a1a6 SL |
541 | if (!si->swap_map[tmp]) { |
542 | found_free = true; | |
543 | break; | |
544 | } | |
545 | tmp++; | |
546 | } | |
235b6217 | 547 | unlock_cluster(ci); |
ebc2a1a6 SL |
548 | if (!found_free) { |
549 | cluster_set_null(&cluster->index); | |
550 | goto new_cluster; | |
551 | } | |
552 | cluster->next = tmp + 1; | |
553 | *offset = tmp; | |
554 | *scan_base = tmp; | |
2a8f9449 SL |
555 | } |
556 | ||
24b8ff7c CEB |
557 | static unsigned long scan_swap_map(struct swap_info_struct *si, |
558 | unsigned char usage) | |
1da177e4 | 559 | { |
235b6217 | 560 | struct swap_cluster_info *ci; |
ebebbbe9 | 561 | unsigned long offset; |
c60aa176 | 562 | unsigned long scan_base; |
7992fde7 | 563 | unsigned long last_in_cluster = 0; |
048c27fd | 564 | int latency_ration = LATENCY_LIMIT; |
7dfad418 | 565 | |
886bb7e9 | 566 | /* |
7dfad418 HD |
567 | * We try to cluster swap pages by allocating them sequentially |
568 | * in swap. Once we've allocated SWAPFILE_CLUSTER pages this | |
569 | * way, however, we resort to first-free allocation, starting | |
570 | * a new cluster. This prevents us from scattering swap pages | |
571 | * all over the entire swap partition, so that we reduce | |
572 | * overall disk seek times between swap pages. -- sct | |
573 | * But we do now try to find an empty cluster. -Andrea | |
c60aa176 | 574 | * And we let swap pages go all over an SSD partition. Hugh |
7dfad418 HD |
575 | */ |
576 | ||
52b7efdb | 577 | si->flags += SWP_SCANNING; |
c60aa176 | 578 | scan_base = offset = si->cluster_next; |
ebebbbe9 | 579 | |
ebc2a1a6 SL |
580 | /* SSD algorithm */ |
581 | if (si->cluster_info) { | |
582 | scan_swap_map_try_ssd_cluster(si, &offset, &scan_base); | |
583 | goto checks; | |
584 | } | |
585 | ||
ebebbbe9 HD |
586 | if (unlikely(!si->cluster_nr--)) { |
587 | if (si->pages - si->inuse_pages < SWAPFILE_CLUSTER) { | |
588 | si->cluster_nr = SWAPFILE_CLUSTER - 1; | |
589 | goto checks; | |
590 | } | |
2a8f9449 | 591 | |
ec8acf20 | 592 | spin_unlock(&si->lock); |
7dfad418 | 593 | |
c60aa176 HD |
594 | /* |
595 | * If seek is expensive, start searching for new cluster from | |
596 | * start of partition, to minimize the span of allocated swap. | |
50088c44 CY |
597 | * If seek is cheap, that is the SWP_SOLIDSTATE si->cluster_info |
598 | * case, just handled by scan_swap_map_try_ssd_cluster() above. | |
c60aa176 | 599 | */ |
50088c44 | 600 | scan_base = offset = si->lowest_bit; |
7dfad418 HD |
601 | last_in_cluster = offset + SWAPFILE_CLUSTER - 1; |
602 | ||
603 | /* Locate the first empty (unaligned) cluster */ | |
604 | for (; last_in_cluster <= si->highest_bit; offset++) { | |
1da177e4 | 605 | if (si->swap_map[offset]) |
7dfad418 HD |
606 | last_in_cluster = offset + SWAPFILE_CLUSTER; |
607 | else if (offset == last_in_cluster) { | |
ec8acf20 | 608 | spin_lock(&si->lock); |
ebebbbe9 HD |
609 | offset -= SWAPFILE_CLUSTER - 1; |
610 | si->cluster_next = offset; | |
611 | si->cluster_nr = SWAPFILE_CLUSTER - 1; | |
c60aa176 HD |
612 | goto checks; |
613 | } | |
614 | if (unlikely(--latency_ration < 0)) { | |
615 | cond_resched(); | |
616 | latency_ration = LATENCY_LIMIT; | |
617 | } | |
618 | } | |
619 | ||
620 | offset = scan_base; | |
ec8acf20 | 621 | spin_lock(&si->lock); |
ebebbbe9 | 622 | si->cluster_nr = SWAPFILE_CLUSTER - 1; |
1da177e4 | 623 | } |
7dfad418 | 624 | |
ebebbbe9 | 625 | checks: |
ebc2a1a6 SL |
626 | if (si->cluster_info) { |
627 | while (scan_swap_map_ssd_cluster_conflict(si, offset)) | |
628 | scan_swap_map_try_ssd_cluster(si, &offset, &scan_base); | |
629 | } | |
ebebbbe9 | 630 | if (!(si->flags & SWP_WRITEOK)) |
52b7efdb | 631 | goto no_page; |
7dfad418 HD |
632 | if (!si->highest_bit) |
633 | goto no_page; | |
ebebbbe9 | 634 | if (offset > si->highest_bit) |
c60aa176 | 635 | scan_base = offset = si->lowest_bit; |
c9e44410 | 636 | |
235b6217 | 637 | ci = lock_cluster(si, offset); |
b73d7fce HD |
638 | /* reuse swap entry of cache-only swap if not busy. */ |
639 | if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) { | |
c9e44410 | 640 | int swap_was_freed; |
235b6217 | 641 | unlock_cluster(ci); |
ec8acf20 | 642 | spin_unlock(&si->lock); |
c9e44410 | 643 | swap_was_freed = __try_to_reclaim_swap(si, offset); |
ec8acf20 | 644 | spin_lock(&si->lock); |
c9e44410 KH |
645 | /* entry was freed successfully, try to use this again */ |
646 | if (swap_was_freed) | |
647 | goto checks; | |
648 | goto scan; /* check next one */ | |
649 | } | |
650 | ||
235b6217 HY |
651 | if (si->swap_map[offset]) { |
652 | unlock_cluster(ci); | |
ebebbbe9 | 653 | goto scan; |
235b6217 | 654 | } |
ebebbbe9 HD |
655 | |
656 | if (offset == si->lowest_bit) | |
657 | si->lowest_bit++; | |
658 | if (offset == si->highest_bit) | |
659 | si->highest_bit--; | |
660 | si->inuse_pages++; | |
661 | if (si->inuse_pages == si->pages) { | |
662 | si->lowest_bit = si->max; | |
663 | si->highest_bit = 0; | |
18ab4d4c DS |
664 | spin_lock(&swap_avail_lock); |
665 | plist_del(&si->avail_list, &swap_avail_head); | |
666 | spin_unlock(&swap_avail_lock); | |
1da177e4 | 667 | } |
253d553b | 668 | si->swap_map[offset] = usage; |
2a8f9449 | 669 | inc_cluster_info_page(si, si->cluster_info, offset); |
235b6217 | 670 | unlock_cluster(ci); |
ebebbbe9 HD |
671 | si->cluster_next = offset + 1; |
672 | si->flags -= SWP_SCANNING; | |
7992fde7 | 673 | |
ebebbbe9 | 674 | return offset; |
7dfad418 | 675 | |
ebebbbe9 | 676 | scan: |
ec8acf20 | 677 | spin_unlock(&si->lock); |
7dfad418 | 678 | while (++offset <= si->highest_bit) { |
52b7efdb | 679 | if (!si->swap_map[offset]) { |
ec8acf20 | 680 | spin_lock(&si->lock); |
52b7efdb HD |
681 | goto checks; |
682 | } | |
c9e44410 | 683 | if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) { |
ec8acf20 | 684 | spin_lock(&si->lock); |
c9e44410 KH |
685 | goto checks; |
686 | } | |
048c27fd HD |
687 | if (unlikely(--latency_ration < 0)) { |
688 | cond_resched(); | |
689 | latency_ration = LATENCY_LIMIT; | |
690 | } | |
7dfad418 | 691 | } |
c60aa176 | 692 | offset = si->lowest_bit; |
a5998061 | 693 | while (offset < scan_base) { |
c60aa176 | 694 | if (!si->swap_map[offset]) { |
ec8acf20 | 695 | spin_lock(&si->lock); |
c60aa176 HD |
696 | goto checks; |
697 | } | |
c9e44410 | 698 | if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) { |
ec8acf20 | 699 | spin_lock(&si->lock); |
c9e44410 KH |
700 | goto checks; |
701 | } | |
c60aa176 HD |
702 | if (unlikely(--latency_ration < 0)) { |
703 | cond_resched(); | |
704 | latency_ration = LATENCY_LIMIT; | |
705 | } | |
a5998061 | 706 | offset++; |
c60aa176 | 707 | } |
ec8acf20 | 708 | spin_lock(&si->lock); |
7dfad418 HD |
709 | |
710 | no_page: | |
52b7efdb | 711 | si->flags -= SWP_SCANNING; |
1da177e4 LT |
712 | return 0; |
713 | } | |
714 | ||
715 | swp_entry_t get_swap_page(void) | |
716 | { | |
adfab836 | 717 | struct swap_info_struct *si, *next; |
fb4f88dc | 718 | pgoff_t offset; |
1da177e4 | 719 | |
ec8acf20 | 720 | if (atomic_long_read(&nr_swap_pages) <= 0) |
fb4f88dc | 721 | goto noswap; |
ec8acf20 | 722 | atomic_long_dec(&nr_swap_pages); |
fb4f88dc | 723 | |
18ab4d4c DS |
724 | spin_lock(&swap_avail_lock); |
725 | ||
726 | start_over: | |
727 | plist_for_each_entry_safe(si, next, &swap_avail_head, avail_list) { | |
728 | /* requeue si to after same-priority siblings */ | |
729 | plist_requeue(&si->avail_list, &swap_avail_head); | |
730 | spin_unlock(&swap_avail_lock); | |
ec8acf20 | 731 | spin_lock(&si->lock); |
adfab836 | 732 | if (!si->highest_bit || !(si->flags & SWP_WRITEOK)) { |
18ab4d4c DS |
733 | spin_lock(&swap_avail_lock); |
734 | if (plist_node_empty(&si->avail_list)) { | |
735 | spin_unlock(&si->lock); | |
736 | goto nextsi; | |
737 | } | |
738 | WARN(!si->highest_bit, | |
739 | "swap_info %d in list but !highest_bit\n", | |
740 | si->type); | |
741 | WARN(!(si->flags & SWP_WRITEOK), | |
742 | "swap_info %d in list but !SWP_WRITEOK\n", | |
743 | si->type); | |
744 | plist_del(&si->avail_list, &swap_avail_head); | |
ec8acf20 | 745 | spin_unlock(&si->lock); |
18ab4d4c | 746 | goto nextsi; |
ec8acf20 | 747 | } |
fb4f88dc | 748 | |
355cfa73 | 749 | /* This is called for allocating swap entry for cache */ |
253d553b | 750 | offset = scan_swap_map(si, SWAP_HAS_CACHE); |
ec8acf20 SL |
751 | spin_unlock(&si->lock); |
752 | if (offset) | |
adfab836 | 753 | return swp_entry(si->type, offset); |
18ab4d4c DS |
754 | pr_debug("scan_swap_map of si %d failed to find offset\n", |
755 | si->type); | |
756 | spin_lock(&swap_avail_lock); | |
757 | nextsi: | |
adfab836 DS |
758 | /* |
759 | * if we got here, it's likely that si was almost full before, | |
760 | * and since scan_swap_map() can drop the si->lock, multiple | |
761 | * callers probably all tried to get a page from the same si | |
18ab4d4c DS |
762 | * and it filled up before we could get one; or, the si filled |
763 | * up between us dropping swap_avail_lock and taking si->lock. | |
764 | * Since we dropped the swap_avail_lock, the swap_avail_head | |
765 | * list may have been modified; so if next is still in the | |
766 | * swap_avail_head list then try it, otherwise start over. | |
adfab836 | 767 | */ |
18ab4d4c DS |
768 | if (plist_node_empty(&next->avail_list)) |
769 | goto start_over; | |
1da177e4 | 770 | } |
fb4f88dc | 771 | |
18ab4d4c DS |
772 | spin_unlock(&swap_avail_lock); |
773 | ||
ec8acf20 | 774 | atomic_long_inc(&nr_swap_pages); |
fb4f88dc | 775 | noswap: |
fb4f88dc | 776 | return (swp_entry_t) {0}; |
1da177e4 LT |
777 | } |
778 | ||
2de1a7e4 | 779 | /* The only caller of this function is now suspend routine */ |
910321ea HD |
780 | swp_entry_t get_swap_page_of_type(int type) |
781 | { | |
782 | struct swap_info_struct *si; | |
783 | pgoff_t offset; | |
784 | ||
910321ea | 785 | si = swap_info[type]; |
ec8acf20 | 786 | spin_lock(&si->lock); |
910321ea | 787 | if (si && (si->flags & SWP_WRITEOK)) { |
ec8acf20 | 788 | atomic_long_dec(&nr_swap_pages); |
910321ea HD |
789 | /* This is called for allocating swap entry, not cache */ |
790 | offset = scan_swap_map(si, 1); | |
791 | if (offset) { | |
ec8acf20 | 792 | spin_unlock(&si->lock); |
910321ea HD |
793 | return swp_entry(type, offset); |
794 | } | |
ec8acf20 | 795 | atomic_long_inc(&nr_swap_pages); |
910321ea | 796 | } |
ec8acf20 | 797 | spin_unlock(&si->lock); |
910321ea HD |
798 | return (swp_entry_t) {0}; |
799 | } | |
800 | ||
235b6217 | 801 | static struct swap_info_struct *_swap_info_get(swp_entry_t entry) |
1da177e4 | 802 | { |
73c34b6a | 803 | struct swap_info_struct *p; |
1da177e4 LT |
804 | unsigned long offset, type; |
805 | ||
806 | if (!entry.val) | |
807 | goto out; | |
808 | type = swp_type(entry); | |
809 | if (type >= nr_swapfiles) | |
810 | goto bad_nofile; | |
efa90a98 | 811 | p = swap_info[type]; |
1da177e4 LT |
812 | if (!(p->flags & SWP_USED)) |
813 | goto bad_device; | |
814 | offset = swp_offset(entry); | |
815 | if (offset >= p->max) | |
816 | goto bad_offset; | |
817 | if (!p->swap_map[offset]) | |
818 | goto bad_free; | |
1da177e4 LT |
819 | return p; |
820 | ||
821 | bad_free: | |
6a991fc7 | 822 | pr_err("swap_info_get: %s%08lx\n", Unused_offset, entry.val); |
1da177e4 LT |
823 | goto out; |
824 | bad_offset: | |
6a991fc7 | 825 | pr_err("swap_info_get: %s%08lx\n", Bad_offset, entry.val); |
1da177e4 LT |
826 | goto out; |
827 | bad_device: | |
6a991fc7 | 828 | pr_err("swap_info_get: %s%08lx\n", Unused_file, entry.val); |
1da177e4 LT |
829 | goto out; |
830 | bad_nofile: | |
6a991fc7 | 831 | pr_err("swap_info_get: %s%08lx\n", Bad_file, entry.val); |
1da177e4 LT |
832 | out: |
833 | return NULL; | |
886bb7e9 | 834 | } |
1da177e4 | 835 | |
235b6217 HY |
836 | static struct swap_info_struct *swap_info_get(swp_entry_t entry) |
837 | { | |
838 | struct swap_info_struct *p; | |
839 | ||
840 | p = _swap_info_get(entry); | |
841 | if (p) | |
842 | spin_lock(&p->lock); | |
843 | return p; | |
844 | } | |
845 | ||
8d69aaee | 846 | static unsigned char swap_entry_free(struct swap_info_struct *p, |
235b6217 HY |
847 | swp_entry_t entry, unsigned char usage, |
848 | bool swap_info_locked) | |
1da177e4 | 849 | { |
235b6217 | 850 | struct swap_cluster_info *ci; |
253d553b | 851 | unsigned long offset = swp_offset(entry); |
8d69aaee HD |
852 | unsigned char count; |
853 | unsigned char has_cache; | |
235b6217 HY |
854 | bool lock_swap_info = false; |
855 | ||
856 | if (!swap_info_locked) { | |
857 | count = p->swap_map[offset]; | |
858 | if (!p->cluster_info || count == usage || count == SWAP_MAP_SHMEM) { | |
859 | lock_swap_info: | |
860 | swap_info_locked = true; | |
861 | lock_swap_info = true; | |
862 | spin_lock(&p->lock); | |
863 | } | |
864 | } | |
865 | ||
866 | ci = lock_cluster(p, offset); | |
355cfa73 | 867 | |
253d553b | 868 | count = p->swap_map[offset]; |
235b6217 HY |
869 | |
870 | if (!swap_info_locked && (count == usage || count == SWAP_MAP_SHMEM)) { | |
871 | unlock_cluster(ci); | |
872 | goto lock_swap_info; | |
873 | } | |
874 | ||
253d553b HD |
875 | has_cache = count & SWAP_HAS_CACHE; |
876 | count &= ~SWAP_HAS_CACHE; | |
355cfa73 | 877 | |
253d553b | 878 | if (usage == SWAP_HAS_CACHE) { |
355cfa73 | 879 | VM_BUG_ON(!has_cache); |
253d553b | 880 | has_cache = 0; |
aaa46865 HD |
881 | } else if (count == SWAP_MAP_SHMEM) { |
882 | /* | |
883 | * Or we could insist on shmem.c using a special | |
884 | * swap_shmem_free() and free_shmem_swap_and_cache()... | |
885 | */ | |
886 | count = 0; | |
570a335b HD |
887 | } else if ((count & ~COUNT_CONTINUED) <= SWAP_MAP_MAX) { |
888 | if (count == COUNT_CONTINUED) { | |
889 | if (swap_count_continued(p, offset, count)) | |
890 | count = SWAP_MAP_MAX | COUNT_CONTINUED; | |
891 | else | |
892 | count = SWAP_MAP_MAX; | |
893 | } else | |
894 | count--; | |
895 | } | |
253d553b | 896 | |
253d553b HD |
897 | usage = count | has_cache; |
898 | p->swap_map[offset] = usage; | |
355cfa73 | 899 | |
235b6217 HY |
900 | unlock_cluster(ci); |
901 | ||
355cfa73 | 902 | /* free if no reference */ |
253d553b | 903 | if (!usage) { |
235b6217 | 904 | VM_BUG_ON(!swap_info_locked); |
37e84351 | 905 | mem_cgroup_uncharge_swap(entry); |
235b6217 | 906 | ci = lock_cluster(p, offset); |
2a8f9449 | 907 | dec_cluster_info_page(p, p->cluster_info, offset); |
235b6217 | 908 | unlock_cluster(ci); |
355cfa73 KH |
909 | if (offset < p->lowest_bit) |
910 | p->lowest_bit = offset; | |
18ab4d4c DS |
911 | if (offset > p->highest_bit) { |
912 | bool was_full = !p->highest_bit; | |
355cfa73 | 913 | p->highest_bit = offset; |
18ab4d4c DS |
914 | if (was_full && (p->flags & SWP_WRITEOK)) { |
915 | spin_lock(&swap_avail_lock); | |
916 | WARN_ON(!plist_node_empty(&p->avail_list)); | |
917 | if (plist_node_empty(&p->avail_list)) | |
918 | plist_add(&p->avail_list, | |
919 | &swap_avail_head); | |
920 | spin_unlock(&swap_avail_lock); | |
921 | } | |
922 | } | |
ec8acf20 | 923 | atomic_long_inc(&nr_swap_pages); |
355cfa73 | 924 | p->inuse_pages--; |
38b5faf4 | 925 | frontswap_invalidate_page(p->type, offset); |
73744923 MG |
926 | if (p->flags & SWP_BLKDEV) { |
927 | struct gendisk *disk = p->bdev->bd_disk; | |
928 | if (disk->fops->swap_slot_free_notify) | |
929 | disk->fops->swap_slot_free_notify(p->bdev, | |
930 | offset); | |
931 | } | |
1da177e4 | 932 | } |
253d553b | 933 | |
235b6217 HY |
934 | if (lock_swap_info) |
935 | spin_unlock(&p->lock); | |
936 | ||
253d553b | 937 | return usage; |
1da177e4 LT |
938 | } |
939 | ||
940 | /* | |
2de1a7e4 | 941 | * Caller has made sure that the swap device corresponding to entry |
1da177e4 LT |
942 | * is still around or has not been recycled. |
943 | */ | |
944 | void swap_free(swp_entry_t entry) | |
945 | { | |
73c34b6a | 946 | struct swap_info_struct *p; |
1da177e4 | 947 | |
235b6217 HY |
948 | p = _swap_info_get(entry); |
949 | if (p) | |
950 | swap_entry_free(p, entry, 1, false); | |
1da177e4 LT |
951 | } |
952 | ||
cb4b86ba KH |
953 | /* |
954 | * Called after dropping swapcache to decrease refcnt to swap entries. | |
955 | */ | |
0a31bc97 | 956 | void swapcache_free(swp_entry_t entry) |
cb4b86ba | 957 | { |
355cfa73 KH |
958 | struct swap_info_struct *p; |
959 | ||
235b6217 HY |
960 | p = _swap_info_get(entry); |
961 | if (p) | |
962 | swap_entry_free(p, entry, SWAP_HAS_CACHE, false); | |
cb4b86ba KH |
963 | } |
964 | ||
1da177e4 | 965 | /* |
c475a8ab | 966 | * How many references to page are currently swapped out? |
570a335b HD |
967 | * This does not give an exact answer when swap count is continued, |
968 | * but does include the high COUNT_CONTINUED flag to allow for that. | |
1da177e4 | 969 | */ |
bde05d1c | 970 | int page_swapcount(struct page *page) |
1da177e4 | 971 | { |
c475a8ab HD |
972 | int count = 0; |
973 | struct swap_info_struct *p; | |
235b6217 | 974 | struct swap_cluster_info *ci; |
1da177e4 | 975 | swp_entry_t entry; |
235b6217 | 976 | unsigned long offset; |
1da177e4 | 977 | |
4c21e2f2 | 978 | entry.val = page_private(page); |
235b6217 | 979 | p = _swap_info_get(entry); |
1da177e4 | 980 | if (p) { |
235b6217 HY |
981 | offset = swp_offset(entry); |
982 | ci = lock_cluster_or_swap_info(p, offset); | |
983 | count = swap_count(p->swap_map[offset]); | |
984 | unlock_cluster_or_swap_info(p, ci); | |
1da177e4 | 985 | } |
c475a8ab | 986 | return count; |
1da177e4 LT |
987 | } |
988 | ||
8334b962 MK |
989 | /* |
990 | * How many references to @entry are currently swapped out? | |
991 | * This considers COUNT_CONTINUED so it returns exact answer. | |
992 | */ | |
993 | int swp_swapcount(swp_entry_t entry) | |
994 | { | |
995 | int count, tmp_count, n; | |
996 | struct swap_info_struct *p; | |
235b6217 | 997 | struct swap_cluster_info *ci; |
8334b962 MK |
998 | struct page *page; |
999 | pgoff_t offset; | |
1000 | unsigned char *map; | |
1001 | ||
235b6217 | 1002 | p = _swap_info_get(entry); |
8334b962 MK |
1003 | if (!p) |
1004 | return 0; | |
1005 | ||
235b6217 HY |
1006 | offset = swp_offset(entry); |
1007 | ||
1008 | ci = lock_cluster_or_swap_info(p, offset); | |
1009 | ||
1010 | count = swap_count(p->swap_map[offset]); | |
8334b962 MK |
1011 | if (!(count & COUNT_CONTINUED)) |
1012 | goto out; | |
1013 | ||
1014 | count &= ~COUNT_CONTINUED; | |
1015 | n = SWAP_MAP_MAX + 1; | |
1016 | ||
8334b962 MK |
1017 | page = vmalloc_to_page(p->swap_map + offset); |
1018 | offset &= ~PAGE_MASK; | |
1019 | VM_BUG_ON(page_private(page) != SWP_CONTINUED); | |
1020 | ||
1021 | do { | |
a8ae4991 | 1022 | page = list_next_entry(page, lru); |
8334b962 MK |
1023 | map = kmap_atomic(page); |
1024 | tmp_count = map[offset]; | |
1025 | kunmap_atomic(map); | |
1026 | ||
1027 | count += (tmp_count & ~COUNT_CONTINUED) * n; | |
1028 | n *= (SWAP_CONT_MAX + 1); | |
1029 | } while (tmp_count & COUNT_CONTINUED); | |
1030 | out: | |
235b6217 | 1031 | unlock_cluster_or_swap_info(p, ci); |
8334b962 MK |
1032 | return count; |
1033 | } | |
1034 | ||
1da177e4 | 1035 | /* |
7b1fe597 HD |
1036 | * We can write to an anon page without COW if there are no other references |
1037 | * to it. And as a side-effect, free up its swap: because the old content | |
1038 | * on disk will never be read, and seeking back there to write new content | |
1039 | * later would only waste time away from clustering. | |
6d0a07ed AA |
1040 | * |
1041 | * NOTE: total_mapcount should not be relied upon by the caller if | |
1042 | * reuse_swap_page() returns false, but it may be always overwritten | |
1043 | * (see the other implementation for CONFIG_SWAP=n). | |
1da177e4 | 1044 | */ |
6d0a07ed | 1045 | bool reuse_swap_page(struct page *page, int *total_mapcount) |
1da177e4 | 1046 | { |
c475a8ab HD |
1047 | int count; |
1048 | ||
309381fe | 1049 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
5ad64688 | 1050 | if (unlikely(PageKsm(page))) |
6d0a07ed AA |
1051 | return false; |
1052 | count = page_trans_huge_mapcount(page, total_mapcount); | |
7b1fe597 | 1053 | if (count <= 1 && PageSwapCache(page)) { |
c475a8ab | 1054 | count += page_swapcount(page); |
f0571429 MK |
1055 | if (count != 1) |
1056 | goto out; | |
1057 | if (!PageWriteback(page)) { | |
7b1fe597 HD |
1058 | delete_from_swap_cache(page); |
1059 | SetPageDirty(page); | |
f0571429 MK |
1060 | } else { |
1061 | swp_entry_t entry; | |
1062 | struct swap_info_struct *p; | |
1063 | ||
1064 | entry.val = page_private(page); | |
1065 | p = swap_info_get(entry); | |
1066 | if (p->flags & SWP_STABLE_WRITES) { | |
1067 | spin_unlock(&p->lock); | |
1068 | return false; | |
1069 | } | |
1070 | spin_unlock(&p->lock); | |
7b1fe597 HD |
1071 | } |
1072 | } | |
f0571429 | 1073 | out: |
5ad64688 | 1074 | return count <= 1; |
1da177e4 LT |
1075 | } |
1076 | ||
1077 | /* | |
a2c43eed HD |
1078 | * If swap is getting full, or if there are no more mappings of this page, |
1079 | * then try_to_free_swap is called to free its swap space. | |
1da177e4 | 1080 | */ |
a2c43eed | 1081 | int try_to_free_swap(struct page *page) |
1da177e4 | 1082 | { |
309381fe | 1083 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
1da177e4 LT |
1084 | |
1085 | if (!PageSwapCache(page)) | |
1086 | return 0; | |
1087 | if (PageWriteback(page)) | |
1088 | return 0; | |
a2c43eed | 1089 | if (page_swapcount(page)) |
1da177e4 LT |
1090 | return 0; |
1091 | ||
b73d7fce HD |
1092 | /* |
1093 | * Once hibernation has begun to create its image of memory, | |
1094 | * there's a danger that one of the calls to try_to_free_swap() | |
1095 | * - most probably a call from __try_to_reclaim_swap() while | |
1096 | * hibernation is allocating its own swap pages for the image, | |
1097 | * but conceivably even a call from memory reclaim - will free | |
1098 | * the swap from a page which has already been recorded in the | |
1099 | * image as a clean swapcache page, and then reuse its swap for | |
1100 | * another page of the image. On waking from hibernation, the | |
1101 | * original page might be freed under memory pressure, then | |
1102 | * later read back in from swap, now with the wrong data. | |
1103 | * | |
2de1a7e4 | 1104 | * Hibernation suspends storage while it is writing the image |
f90ac398 | 1105 | * to disk so check that here. |
b73d7fce | 1106 | */ |
f90ac398 | 1107 | if (pm_suspended_storage()) |
b73d7fce HD |
1108 | return 0; |
1109 | ||
a2c43eed HD |
1110 | delete_from_swap_cache(page); |
1111 | SetPageDirty(page); | |
1112 | return 1; | |
68a22394 RR |
1113 | } |
1114 | ||
1da177e4 LT |
1115 | /* |
1116 | * Free the swap entry like above, but also try to | |
1117 | * free the page cache entry if it is the last user. | |
1118 | */ | |
2509ef26 | 1119 | int free_swap_and_cache(swp_entry_t entry) |
1da177e4 | 1120 | { |
2509ef26 | 1121 | struct swap_info_struct *p; |
1da177e4 LT |
1122 | struct page *page = NULL; |
1123 | ||
a7420aa5 | 1124 | if (non_swap_entry(entry)) |
2509ef26 | 1125 | return 1; |
0697212a | 1126 | |
1da177e4 LT |
1127 | p = swap_info_get(entry); |
1128 | if (p) { | |
235b6217 | 1129 | if (swap_entry_free(p, entry, 1, true) == SWAP_HAS_CACHE) { |
33806f06 | 1130 | page = find_get_page(swap_address_space(entry), |
f6ab1f7f | 1131 | swp_offset(entry)); |
8413ac9d | 1132 | if (page && !trylock_page(page)) { |
09cbfeaf | 1133 | put_page(page); |
93fac704 NP |
1134 | page = NULL; |
1135 | } | |
1136 | } | |
ec8acf20 | 1137 | spin_unlock(&p->lock); |
1da177e4 LT |
1138 | } |
1139 | if (page) { | |
a2c43eed HD |
1140 | /* |
1141 | * Not mapped elsewhere, or swap space full? Free it! | |
1142 | * Also recheck PageSwapCache now page is locked (above). | |
1143 | */ | |
93fac704 | 1144 | if (PageSwapCache(page) && !PageWriteback(page) && |
5ccc5aba | 1145 | (!page_mapped(page) || mem_cgroup_swap_full(page))) { |
1da177e4 LT |
1146 | delete_from_swap_cache(page); |
1147 | SetPageDirty(page); | |
1148 | } | |
1149 | unlock_page(page); | |
09cbfeaf | 1150 | put_page(page); |
1da177e4 | 1151 | } |
2509ef26 | 1152 | return p != NULL; |
1da177e4 LT |
1153 | } |
1154 | ||
b0cb1a19 | 1155 | #ifdef CONFIG_HIBERNATION |
f577eb30 | 1156 | /* |
915bae9e | 1157 | * Find the swap type that corresponds to given device (if any). |
f577eb30 | 1158 | * |
915bae9e RW |
1159 | * @offset - number of the PAGE_SIZE-sized block of the device, starting |
1160 | * from 0, in which the swap header is expected to be located. | |
1161 | * | |
1162 | * This is needed for the suspend to disk (aka swsusp). | |
f577eb30 | 1163 | */ |
7bf23687 | 1164 | int swap_type_of(dev_t device, sector_t offset, struct block_device **bdev_p) |
f577eb30 | 1165 | { |
915bae9e | 1166 | struct block_device *bdev = NULL; |
efa90a98 | 1167 | int type; |
f577eb30 | 1168 | |
915bae9e RW |
1169 | if (device) |
1170 | bdev = bdget(device); | |
1171 | ||
f577eb30 | 1172 | spin_lock(&swap_lock); |
efa90a98 HD |
1173 | for (type = 0; type < nr_swapfiles; type++) { |
1174 | struct swap_info_struct *sis = swap_info[type]; | |
f577eb30 | 1175 | |
915bae9e | 1176 | if (!(sis->flags & SWP_WRITEOK)) |
f577eb30 | 1177 | continue; |
b6b5bce3 | 1178 | |
915bae9e | 1179 | if (!bdev) { |
7bf23687 | 1180 | if (bdev_p) |
dddac6a7 | 1181 | *bdev_p = bdgrab(sis->bdev); |
7bf23687 | 1182 | |
6e1819d6 | 1183 | spin_unlock(&swap_lock); |
efa90a98 | 1184 | return type; |
6e1819d6 | 1185 | } |
915bae9e | 1186 | if (bdev == sis->bdev) { |
9625a5f2 | 1187 | struct swap_extent *se = &sis->first_swap_extent; |
915bae9e | 1188 | |
915bae9e | 1189 | if (se->start_block == offset) { |
7bf23687 | 1190 | if (bdev_p) |
dddac6a7 | 1191 | *bdev_p = bdgrab(sis->bdev); |
7bf23687 | 1192 | |
915bae9e RW |
1193 | spin_unlock(&swap_lock); |
1194 | bdput(bdev); | |
efa90a98 | 1195 | return type; |
915bae9e | 1196 | } |
f577eb30 RW |
1197 | } |
1198 | } | |
1199 | spin_unlock(&swap_lock); | |
915bae9e RW |
1200 | if (bdev) |
1201 | bdput(bdev); | |
1202 | ||
f577eb30 RW |
1203 | return -ENODEV; |
1204 | } | |
1205 | ||
73c34b6a HD |
1206 | /* |
1207 | * Get the (PAGE_SIZE) block corresponding to given offset on the swapdev | |
1208 | * corresponding to given index in swap_info (swap type). | |
1209 | */ | |
1210 | sector_t swapdev_block(int type, pgoff_t offset) | |
1211 | { | |
1212 | struct block_device *bdev; | |
1213 | ||
1214 | if ((unsigned int)type >= nr_swapfiles) | |
1215 | return 0; | |
1216 | if (!(swap_info[type]->flags & SWP_WRITEOK)) | |
1217 | return 0; | |
d4906e1a | 1218 | return map_swap_entry(swp_entry(type, offset), &bdev); |
73c34b6a HD |
1219 | } |
1220 | ||
f577eb30 RW |
1221 | /* |
1222 | * Return either the total number of swap pages of given type, or the number | |
1223 | * of free pages of that type (depending on @free) | |
1224 | * | |
1225 | * This is needed for software suspend | |
1226 | */ | |
1227 | unsigned int count_swap_pages(int type, int free) | |
1228 | { | |
1229 | unsigned int n = 0; | |
1230 | ||
efa90a98 HD |
1231 | spin_lock(&swap_lock); |
1232 | if ((unsigned int)type < nr_swapfiles) { | |
1233 | struct swap_info_struct *sis = swap_info[type]; | |
1234 | ||
ec8acf20 | 1235 | spin_lock(&sis->lock); |
efa90a98 HD |
1236 | if (sis->flags & SWP_WRITEOK) { |
1237 | n = sis->pages; | |
f577eb30 | 1238 | if (free) |
efa90a98 | 1239 | n -= sis->inuse_pages; |
f577eb30 | 1240 | } |
ec8acf20 | 1241 | spin_unlock(&sis->lock); |
f577eb30 | 1242 | } |
efa90a98 | 1243 | spin_unlock(&swap_lock); |
f577eb30 RW |
1244 | return n; |
1245 | } | |
73c34b6a | 1246 | #endif /* CONFIG_HIBERNATION */ |
f577eb30 | 1247 | |
9f8bdb3f | 1248 | static inline int pte_same_as_swp(pte_t pte, pte_t swp_pte) |
179ef71c | 1249 | { |
9f8bdb3f | 1250 | return pte_same(pte_swp_clear_soft_dirty(pte), swp_pte); |
179ef71c CG |
1251 | } |
1252 | ||
1da177e4 | 1253 | /* |
72866f6f HD |
1254 | * No need to decide whether this PTE shares the swap entry with others, |
1255 | * just let do_wp_page work it out if a write is requested later - to | |
1256 | * force COW, vm_page_prot omits write permission from any private vma. | |
1da177e4 | 1257 | */ |
044d66c1 | 1258 | static int unuse_pte(struct vm_area_struct *vma, pmd_t *pmd, |
1da177e4 LT |
1259 | unsigned long addr, swp_entry_t entry, struct page *page) |
1260 | { | |
9e16b7fb | 1261 | struct page *swapcache; |
72835c86 | 1262 | struct mem_cgroup *memcg; |
044d66c1 HD |
1263 | spinlock_t *ptl; |
1264 | pte_t *pte; | |
1265 | int ret = 1; | |
1266 | ||
9e16b7fb HD |
1267 | swapcache = page; |
1268 | page = ksm_might_need_to_copy(page, vma, addr); | |
1269 | if (unlikely(!page)) | |
1270 | return -ENOMEM; | |
1271 | ||
f627c2f5 KS |
1272 | if (mem_cgroup_try_charge(page, vma->vm_mm, GFP_KERNEL, |
1273 | &memcg, false)) { | |
044d66c1 | 1274 | ret = -ENOMEM; |
85d9fc89 KH |
1275 | goto out_nolock; |
1276 | } | |
044d66c1 HD |
1277 | |
1278 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); | |
9f8bdb3f | 1279 | if (unlikely(!pte_same_as_swp(*pte, swp_entry_to_pte(entry)))) { |
f627c2f5 | 1280 | mem_cgroup_cancel_charge(page, memcg, false); |
044d66c1 HD |
1281 | ret = 0; |
1282 | goto out; | |
1283 | } | |
8a9f3ccd | 1284 | |
b084d435 | 1285 | dec_mm_counter(vma->vm_mm, MM_SWAPENTS); |
d559db08 | 1286 | inc_mm_counter(vma->vm_mm, MM_ANONPAGES); |
1da177e4 LT |
1287 | get_page(page); |
1288 | set_pte_at(vma->vm_mm, addr, pte, | |
1289 | pte_mkold(mk_pte(page, vma->vm_page_prot))); | |
00501b53 | 1290 | if (page == swapcache) { |
d281ee61 | 1291 | page_add_anon_rmap(page, vma, addr, false); |
f627c2f5 | 1292 | mem_cgroup_commit_charge(page, memcg, true, false); |
00501b53 | 1293 | } else { /* ksm created a completely new copy */ |
d281ee61 | 1294 | page_add_new_anon_rmap(page, vma, addr, false); |
f627c2f5 | 1295 | mem_cgroup_commit_charge(page, memcg, false, false); |
00501b53 JW |
1296 | lru_cache_add_active_or_unevictable(page, vma); |
1297 | } | |
1da177e4 LT |
1298 | swap_free(entry); |
1299 | /* | |
1300 | * Move the page to the active list so it is not | |
1301 | * immediately swapped out again after swapon. | |
1302 | */ | |
1303 | activate_page(page); | |
044d66c1 HD |
1304 | out: |
1305 | pte_unmap_unlock(pte, ptl); | |
85d9fc89 | 1306 | out_nolock: |
9e16b7fb HD |
1307 | if (page != swapcache) { |
1308 | unlock_page(page); | |
1309 | put_page(page); | |
1310 | } | |
044d66c1 | 1311 | return ret; |
1da177e4 LT |
1312 | } |
1313 | ||
1314 | static int unuse_pte_range(struct vm_area_struct *vma, pmd_t *pmd, | |
1315 | unsigned long addr, unsigned long end, | |
1316 | swp_entry_t entry, struct page *page) | |
1317 | { | |
1da177e4 | 1318 | pte_t swp_pte = swp_entry_to_pte(entry); |
705e87c0 | 1319 | pte_t *pte; |
8a9f3ccd | 1320 | int ret = 0; |
1da177e4 | 1321 | |
044d66c1 HD |
1322 | /* |
1323 | * We don't actually need pte lock while scanning for swp_pte: since | |
1324 | * we hold page lock and mmap_sem, swp_pte cannot be inserted into the | |
1325 | * page table while we're scanning; though it could get zapped, and on | |
1326 | * some architectures (e.g. x86_32 with PAE) we might catch a glimpse | |
1327 | * of unmatched parts which look like swp_pte, so unuse_pte must | |
1328 | * recheck under pte lock. Scanning without pte lock lets it be | |
2de1a7e4 | 1329 | * preemptable whenever CONFIG_PREEMPT but not CONFIG_HIGHPTE. |
044d66c1 HD |
1330 | */ |
1331 | pte = pte_offset_map(pmd, addr); | |
1da177e4 LT |
1332 | do { |
1333 | /* | |
1334 | * swapoff spends a _lot_ of time in this loop! | |
1335 | * Test inline before going to call unuse_pte. | |
1336 | */ | |
9f8bdb3f | 1337 | if (unlikely(pte_same_as_swp(*pte, swp_pte))) { |
044d66c1 HD |
1338 | pte_unmap(pte); |
1339 | ret = unuse_pte(vma, pmd, addr, entry, page); | |
1340 | if (ret) | |
1341 | goto out; | |
1342 | pte = pte_offset_map(pmd, addr); | |
1da177e4 LT |
1343 | } |
1344 | } while (pte++, addr += PAGE_SIZE, addr != end); | |
044d66c1 HD |
1345 | pte_unmap(pte - 1); |
1346 | out: | |
8a9f3ccd | 1347 | return ret; |
1da177e4 LT |
1348 | } |
1349 | ||
1350 | static inline int unuse_pmd_range(struct vm_area_struct *vma, pud_t *pud, | |
1351 | unsigned long addr, unsigned long end, | |
1352 | swp_entry_t entry, struct page *page) | |
1353 | { | |
1354 | pmd_t *pmd; | |
1355 | unsigned long next; | |
8a9f3ccd | 1356 | int ret; |
1da177e4 LT |
1357 | |
1358 | pmd = pmd_offset(pud, addr); | |
1359 | do { | |
dc644a07 | 1360 | cond_resched(); |
1da177e4 | 1361 | next = pmd_addr_end(addr, end); |
1a5a9906 | 1362 | if (pmd_none_or_trans_huge_or_clear_bad(pmd)) |
1da177e4 | 1363 | continue; |
8a9f3ccd BS |
1364 | ret = unuse_pte_range(vma, pmd, addr, next, entry, page); |
1365 | if (ret) | |
1366 | return ret; | |
1da177e4 LT |
1367 | } while (pmd++, addr = next, addr != end); |
1368 | return 0; | |
1369 | } | |
1370 | ||
1371 | static inline int unuse_pud_range(struct vm_area_struct *vma, pgd_t *pgd, | |
1372 | unsigned long addr, unsigned long end, | |
1373 | swp_entry_t entry, struct page *page) | |
1374 | { | |
1375 | pud_t *pud; | |
1376 | unsigned long next; | |
8a9f3ccd | 1377 | int ret; |
1da177e4 LT |
1378 | |
1379 | pud = pud_offset(pgd, addr); | |
1380 | do { | |
1381 | next = pud_addr_end(addr, end); | |
1382 | if (pud_none_or_clear_bad(pud)) | |
1383 | continue; | |
8a9f3ccd BS |
1384 | ret = unuse_pmd_range(vma, pud, addr, next, entry, page); |
1385 | if (ret) | |
1386 | return ret; | |
1da177e4 LT |
1387 | } while (pud++, addr = next, addr != end); |
1388 | return 0; | |
1389 | } | |
1390 | ||
1391 | static int unuse_vma(struct vm_area_struct *vma, | |
1392 | swp_entry_t entry, struct page *page) | |
1393 | { | |
1394 | pgd_t *pgd; | |
1395 | unsigned long addr, end, next; | |
8a9f3ccd | 1396 | int ret; |
1da177e4 | 1397 | |
3ca7b3c5 | 1398 | if (page_anon_vma(page)) { |
1da177e4 LT |
1399 | addr = page_address_in_vma(page, vma); |
1400 | if (addr == -EFAULT) | |
1401 | return 0; | |
1402 | else | |
1403 | end = addr + PAGE_SIZE; | |
1404 | } else { | |
1405 | addr = vma->vm_start; | |
1406 | end = vma->vm_end; | |
1407 | } | |
1408 | ||
1409 | pgd = pgd_offset(vma->vm_mm, addr); | |
1410 | do { | |
1411 | next = pgd_addr_end(addr, end); | |
1412 | if (pgd_none_or_clear_bad(pgd)) | |
1413 | continue; | |
8a9f3ccd BS |
1414 | ret = unuse_pud_range(vma, pgd, addr, next, entry, page); |
1415 | if (ret) | |
1416 | return ret; | |
1da177e4 LT |
1417 | } while (pgd++, addr = next, addr != end); |
1418 | return 0; | |
1419 | } | |
1420 | ||
1421 | static int unuse_mm(struct mm_struct *mm, | |
1422 | swp_entry_t entry, struct page *page) | |
1423 | { | |
1424 | struct vm_area_struct *vma; | |
8a9f3ccd | 1425 | int ret = 0; |
1da177e4 LT |
1426 | |
1427 | if (!down_read_trylock(&mm->mmap_sem)) { | |
1428 | /* | |
7d03431c FLVC |
1429 | * Activate page so shrink_inactive_list is unlikely to unmap |
1430 | * its ptes while lock is dropped, so swapoff can make progress. | |
1da177e4 | 1431 | */ |
c475a8ab | 1432 | activate_page(page); |
1da177e4 LT |
1433 | unlock_page(page); |
1434 | down_read(&mm->mmap_sem); | |
1435 | lock_page(page); | |
1436 | } | |
1da177e4 | 1437 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
8a9f3ccd | 1438 | if (vma->anon_vma && (ret = unuse_vma(vma, entry, page))) |
1da177e4 | 1439 | break; |
dc644a07 | 1440 | cond_resched(); |
1da177e4 | 1441 | } |
1da177e4 | 1442 | up_read(&mm->mmap_sem); |
8a9f3ccd | 1443 | return (ret < 0)? ret: 0; |
1da177e4 LT |
1444 | } |
1445 | ||
1446 | /* | |
38b5faf4 DM |
1447 | * Scan swap_map (or frontswap_map if frontswap parameter is true) |
1448 | * from current position to next entry still in use. | |
1da177e4 LT |
1449 | * Recycle to start on reaching the end, returning 0 when empty. |
1450 | */ | |
6eb396dc | 1451 | static unsigned int find_next_to_unuse(struct swap_info_struct *si, |
38b5faf4 | 1452 | unsigned int prev, bool frontswap) |
1da177e4 | 1453 | { |
6eb396dc HD |
1454 | unsigned int max = si->max; |
1455 | unsigned int i = prev; | |
8d69aaee | 1456 | unsigned char count; |
1da177e4 LT |
1457 | |
1458 | /* | |
5d337b91 | 1459 | * No need for swap_lock here: we're just looking |
1da177e4 LT |
1460 | * for whether an entry is in use, not modifying it; false |
1461 | * hits are okay, and sys_swapoff() has already prevented new | |
5d337b91 | 1462 | * allocations from this area (while holding swap_lock). |
1da177e4 LT |
1463 | */ |
1464 | for (;;) { | |
1465 | if (++i >= max) { | |
1466 | if (!prev) { | |
1467 | i = 0; | |
1468 | break; | |
1469 | } | |
1470 | /* | |
1471 | * No entries in use at top of swap_map, | |
1472 | * loop back to start and recheck there. | |
1473 | */ | |
1474 | max = prev + 1; | |
1475 | prev = 0; | |
1476 | i = 1; | |
1477 | } | |
4db0c3c2 | 1478 | count = READ_ONCE(si->swap_map[i]); |
355cfa73 | 1479 | if (count && swap_count(count) != SWAP_MAP_BAD) |
dc644a07 HD |
1480 | if (!frontswap || frontswap_test(si, i)) |
1481 | break; | |
1482 | if ((i % LATENCY_LIMIT) == 0) | |
1483 | cond_resched(); | |
1da177e4 LT |
1484 | } |
1485 | return i; | |
1486 | } | |
1487 | ||
1488 | /* | |
1489 | * We completely avoid races by reading each swap page in advance, | |
1490 | * and then search for the process using it. All the necessary | |
1491 | * page table adjustments can then be made atomically. | |
38b5faf4 DM |
1492 | * |
1493 | * if the boolean frontswap is true, only unuse pages_to_unuse pages; | |
1494 | * pages_to_unuse==0 means all pages; ignored if frontswap is false | |
1da177e4 | 1495 | */ |
38b5faf4 DM |
1496 | int try_to_unuse(unsigned int type, bool frontswap, |
1497 | unsigned long pages_to_unuse) | |
1da177e4 | 1498 | { |
efa90a98 | 1499 | struct swap_info_struct *si = swap_info[type]; |
1da177e4 | 1500 | struct mm_struct *start_mm; |
edfe23da SL |
1501 | volatile unsigned char *swap_map; /* swap_map is accessed without |
1502 | * locking. Mark it as volatile | |
1503 | * to prevent compiler doing | |
1504 | * something odd. | |
1505 | */ | |
8d69aaee | 1506 | unsigned char swcount; |
1da177e4 LT |
1507 | struct page *page; |
1508 | swp_entry_t entry; | |
6eb396dc | 1509 | unsigned int i = 0; |
1da177e4 | 1510 | int retval = 0; |
1da177e4 LT |
1511 | |
1512 | /* | |
1513 | * When searching mms for an entry, a good strategy is to | |
1514 | * start at the first mm we freed the previous entry from | |
1515 | * (though actually we don't notice whether we or coincidence | |
1516 | * freed the entry). Initialize this start_mm with a hold. | |
1517 | * | |
1518 | * A simpler strategy would be to start at the last mm we | |
1519 | * freed the previous entry from; but that would take less | |
1520 | * advantage of mmlist ordering, which clusters forked mms | |
1521 | * together, child after parent. If we race with dup_mmap(), we | |
1522 | * prefer to resolve parent before child, lest we miss entries | |
1523 | * duplicated after we scanned child: using last mm would invert | |
570a335b | 1524 | * that. |
1da177e4 LT |
1525 | */ |
1526 | start_mm = &init_mm; | |
1527 | atomic_inc(&init_mm.mm_users); | |
1528 | ||
1529 | /* | |
1530 | * Keep on scanning until all entries have gone. Usually, | |
1531 | * one pass through swap_map is enough, but not necessarily: | |
1532 | * there are races when an instance of an entry might be missed. | |
1533 | */ | |
38b5faf4 | 1534 | while ((i = find_next_to_unuse(si, i, frontswap)) != 0) { |
1da177e4 LT |
1535 | if (signal_pending(current)) { |
1536 | retval = -EINTR; | |
1537 | break; | |
1538 | } | |
1539 | ||
886bb7e9 | 1540 | /* |
1da177e4 LT |
1541 | * Get a page for the entry, using the existing swap |
1542 | * cache page if there is one. Otherwise, get a clean | |
886bb7e9 | 1543 | * page and read the swap into it. |
1da177e4 LT |
1544 | */ |
1545 | swap_map = &si->swap_map[i]; | |
1546 | entry = swp_entry(type, i); | |
02098fea HD |
1547 | page = read_swap_cache_async(entry, |
1548 | GFP_HIGHUSER_MOVABLE, NULL, 0); | |
1da177e4 LT |
1549 | if (!page) { |
1550 | /* | |
1551 | * Either swap_duplicate() failed because entry | |
1552 | * has been freed independently, and will not be | |
1553 | * reused since sys_swapoff() already disabled | |
1554 | * allocation from here, or alloc_page() failed. | |
1555 | */ | |
edfe23da SL |
1556 | swcount = *swap_map; |
1557 | /* | |
1558 | * We don't hold lock here, so the swap entry could be | |
1559 | * SWAP_MAP_BAD (when the cluster is discarding). | |
1560 | * Instead of fail out, We can just skip the swap | |
1561 | * entry because swapoff will wait for discarding | |
1562 | * finish anyway. | |
1563 | */ | |
1564 | if (!swcount || swcount == SWAP_MAP_BAD) | |
1da177e4 LT |
1565 | continue; |
1566 | retval = -ENOMEM; | |
1567 | break; | |
1568 | } | |
1569 | ||
1570 | /* | |
1571 | * Don't hold on to start_mm if it looks like exiting. | |
1572 | */ | |
1573 | if (atomic_read(&start_mm->mm_users) == 1) { | |
1574 | mmput(start_mm); | |
1575 | start_mm = &init_mm; | |
1576 | atomic_inc(&init_mm.mm_users); | |
1577 | } | |
1578 | ||
1579 | /* | |
1580 | * Wait for and lock page. When do_swap_page races with | |
1581 | * try_to_unuse, do_swap_page can handle the fault much | |
1582 | * faster than try_to_unuse can locate the entry. This | |
1583 | * apparently redundant "wait_on_page_locked" lets try_to_unuse | |
1584 | * defer to do_swap_page in such a case - in some tests, | |
1585 | * do_swap_page and try_to_unuse repeatedly compete. | |
1586 | */ | |
1587 | wait_on_page_locked(page); | |
1588 | wait_on_page_writeback(page); | |
1589 | lock_page(page); | |
1590 | wait_on_page_writeback(page); | |
1591 | ||
1592 | /* | |
1593 | * Remove all references to entry. | |
1da177e4 | 1594 | */ |
1da177e4 | 1595 | swcount = *swap_map; |
aaa46865 HD |
1596 | if (swap_count(swcount) == SWAP_MAP_SHMEM) { |
1597 | retval = shmem_unuse(entry, page); | |
1598 | /* page has already been unlocked and released */ | |
1599 | if (retval < 0) | |
1600 | break; | |
1601 | continue; | |
1da177e4 | 1602 | } |
aaa46865 HD |
1603 | if (swap_count(swcount) && start_mm != &init_mm) |
1604 | retval = unuse_mm(start_mm, entry, page); | |
1605 | ||
355cfa73 | 1606 | if (swap_count(*swap_map)) { |
1da177e4 LT |
1607 | int set_start_mm = (*swap_map >= swcount); |
1608 | struct list_head *p = &start_mm->mmlist; | |
1609 | struct mm_struct *new_start_mm = start_mm; | |
1610 | struct mm_struct *prev_mm = start_mm; | |
1611 | struct mm_struct *mm; | |
1612 | ||
1613 | atomic_inc(&new_start_mm->mm_users); | |
1614 | atomic_inc(&prev_mm->mm_users); | |
1615 | spin_lock(&mmlist_lock); | |
aaa46865 | 1616 | while (swap_count(*swap_map) && !retval && |
1da177e4 LT |
1617 | (p = p->next) != &start_mm->mmlist) { |
1618 | mm = list_entry(p, struct mm_struct, mmlist); | |
70af7c5c | 1619 | if (!atomic_inc_not_zero(&mm->mm_users)) |
1da177e4 | 1620 | continue; |
1da177e4 LT |
1621 | spin_unlock(&mmlist_lock); |
1622 | mmput(prev_mm); | |
1623 | prev_mm = mm; | |
1624 | ||
1625 | cond_resched(); | |
1626 | ||
1627 | swcount = *swap_map; | |
355cfa73 | 1628 | if (!swap_count(swcount)) /* any usage ? */ |
1da177e4 | 1629 | ; |
aaa46865 | 1630 | else if (mm == &init_mm) |
1da177e4 | 1631 | set_start_mm = 1; |
aaa46865 | 1632 | else |
1da177e4 | 1633 | retval = unuse_mm(mm, entry, page); |
355cfa73 | 1634 | |
32c5fc10 | 1635 | if (set_start_mm && *swap_map < swcount) { |
1da177e4 LT |
1636 | mmput(new_start_mm); |
1637 | atomic_inc(&mm->mm_users); | |
1638 | new_start_mm = mm; | |
1639 | set_start_mm = 0; | |
1640 | } | |
1641 | spin_lock(&mmlist_lock); | |
1642 | } | |
1643 | spin_unlock(&mmlist_lock); | |
1644 | mmput(prev_mm); | |
1645 | mmput(start_mm); | |
1646 | start_mm = new_start_mm; | |
1647 | } | |
1648 | if (retval) { | |
1649 | unlock_page(page); | |
09cbfeaf | 1650 | put_page(page); |
1da177e4 LT |
1651 | break; |
1652 | } | |
1653 | ||
1da177e4 LT |
1654 | /* |
1655 | * If a reference remains (rare), we would like to leave | |
1656 | * the page in the swap cache; but try_to_unmap could | |
1657 | * then re-duplicate the entry once we drop page lock, | |
1658 | * so we might loop indefinitely; also, that page could | |
1659 | * not be swapped out to other storage meanwhile. So: | |
1660 | * delete from cache even if there's another reference, | |
1661 | * after ensuring that the data has been saved to disk - | |
1662 | * since if the reference remains (rarer), it will be | |
1663 | * read from disk into another page. Splitting into two | |
1664 | * pages would be incorrect if swap supported "shared | |
1665 | * private" pages, but they are handled by tmpfs files. | |
5ad64688 HD |
1666 | * |
1667 | * Given how unuse_vma() targets one particular offset | |
1668 | * in an anon_vma, once the anon_vma has been determined, | |
1669 | * this splitting happens to be just what is needed to | |
1670 | * handle where KSM pages have been swapped out: re-reading | |
1671 | * is unnecessarily slow, but we can fix that later on. | |
1da177e4 | 1672 | */ |
355cfa73 KH |
1673 | if (swap_count(*swap_map) && |
1674 | PageDirty(page) && PageSwapCache(page)) { | |
1da177e4 LT |
1675 | struct writeback_control wbc = { |
1676 | .sync_mode = WB_SYNC_NONE, | |
1677 | }; | |
1678 | ||
1679 | swap_writepage(page, &wbc); | |
1680 | lock_page(page); | |
1681 | wait_on_page_writeback(page); | |
1682 | } | |
68bdc8d6 HD |
1683 | |
1684 | /* | |
1685 | * It is conceivable that a racing task removed this page from | |
1686 | * swap cache just before we acquired the page lock at the top, | |
1687 | * or while we dropped it in unuse_mm(). The page might even | |
1688 | * be back in swap cache on another swap area: that we must not | |
1689 | * delete, since it may not have been written out to swap yet. | |
1690 | */ | |
1691 | if (PageSwapCache(page) && | |
1692 | likely(page_private(page) == entry.val)) | |
2e0e26c7 | 1693 | delete_from_swap_cache(page); |
1da177e4 LT |
1694 | |
1695 | /* | |
1696 | * So we could skip searching mms once swap count went | |
1697 | * to 1, we did not mark any present ptes as dirty: must | |
2706a1b8 | 1698 | * mark page dirty so shrink_page_list will preserve it. |
1da177e4 LT |
1699 | */ |
1700 | SetPageDirty(page); | |
1701 | unlock_page(page); | |
09cbfeaf | 1702 | put_page(page); |
1da177e4 LT |
1703 | |
1704 | /* | |
1705 | * Make sure that we aren't completely killing | |
1706 | * interactive performance. | |
1707 | */ | |
1708 | cond_resched(); | |
38b5faf4 DM |
1709 | if (frontswap && pages_to_unuse > 0) { |
1710 | if (!--pages_to_unuse) | |
1711 | break; | |
1712 | } | |
1da177e4 LT |
1713 | } |
1714 | ||
1715 | mmput(start_mm); | |
1da177e4 LT |
1716 | return retval; |
1717 | } | |
1718 | ||
1719 | /* | |
5d337b91 HD |
1720 | * After a successful try_to_unuse, if no swap is now in use, we know |
1721 | * we can empty the mmlist. swap_lock must be held on entry and exit. | |
1722 | * Note that mmlist_lock nests inside swap_lock, and an mm must be | |
1da177e4 LT |
1723 | * added to the mmlist just after page_duplicate - before would be racy. |
1724 | */ | |
1725 | static void drain_mmlist(void) | |
1726 | { | |
1727 | struct list_head *p, *next; | |
efa90a98 | 1728 | unsigned int type; |
1da177e4 | 1729 | |
efa90a98 HD |
1730 | for (type = 0; type < nr_swapfiles; type++) |
1731 | if (swap_info[type]->inuse_pages) | |
1da177e4 LT |
1732 | return; |
1733 | spin_lock(&mmlist_lock); | |
1734 | list_for_each_safe(p, next, &init_mm.mmlist) | |
1735 | list_del_init(p); | |
1736 | spin_unlock(&mmlist_lock); | |
1737 | } | |
1738 | ||
1739 | /* | |
1740 | * Use this swapdev's extent info to locate the (PAGE_SIZE) block which | |
d4906e1a LS |
1741 | * corresponds to page offset for the specified swap entry. |
1742 | * Note that the type of this function is sector_t, but it returns page offset | |
1743 | * into the bdev, not sector offset. | |
1da177e4 | 1744 | */ |
d4906e1a | 1745 | static sector_t map_swap_entry(swp_entry_t entry, struct block_device **bdev) |
1da177e4 | 1746 | { |
f29ad6a9 HD |
1747 | struct swap_info_struct *sis; |
1748 | struct swap_extent *start_se; | |
1749 | struct swap_extent *se; | |
1750 | pgoff_t offset; | |
1751 | ||
efa90a98 | 1752 | sis = swap_info[swp_type(entry)]; |
f29ad6a9 HD |
1753 | *bdev = sis->bdev; |
1754 | ||
1755 | offset = swp_offset(entry); | |
1756 | start_se = sis->curr_swap_extent; | |
1757 | se = start_se; | |
1da177e4 LT |
1758 | |
1759 | for ( ; ; ) { | |
1da177e4 LT |
1760 | if (se->start_page <= offset && |
1761 | offset < (se->start_page + se->nr_pages)) { | |
1762 | return se->start_block + (offset - se->start_page); | |
1763 | } | |
a8ae4991 | 1764 | se = list_next_entry(se, list); |
1da177e4 LT |
1765 | sis->curr_swap_extent = se; |
1766 | BUG_ON(se == start_se); /* It *must* be present */ | |
1767 | } | |
1768 | } | |
1769 | ||
d4906e1a LS |
1770 | /* |
1771 | * Returns the page offset into bdev for the specified page's swap entry. | |
1772 | */ | |
1773 | sector_t map_swap_page(struct page *page, struct block_device **bdev) | |
1774 | { | |
1775 | swp_entry_t entry; | |
1776 | entry.val = page_private(page); | |
1777 | return map_swap_entry(entry, bdev); | |
1778 | } | |
1779 | ||
1da177e4 LT |
1780 | /* |
1781 | * Free all of a swapdev's extent information | |
1782 | */ | |
1783 | static void destroy_swap_extents(struct swap_info_struct *sis) | |
1784 | { | |
9625a5f2 | 1785 | while (!list_empty(&sis->first_swap_extent.list)) { |
1da177e4 LT |
1786 | struct swap_extent *se; |
1787 | ||
a8ae4991 | 1788 | se = list_first_entry(&sis->first_swap_extent.list, |
1da177e4 LT |
1789 | struct swap_extent, list); |
1790 | list_del(&se->list); | |
1791 | kfree(se); | |
1792 | } | |
62c230bc MG |
1793 | |
1794 | if (sis->flags & SWP_FILE) { | |
1795 | struct file *swap_file = sis->swap_file; | |
1796 | struct address_space *mapping = swap_file->f_mapping; | |
1797 | ||
1798 | sis->flags &= ~SWP_FILE; | |
1799 | mapping->a_ops->swap_deactivate(swap_file); | |
1800 | } | |
1da177e4 LT |
1801 | } |
1802 | ||
1803 | /* | |
1804 | * Add a block range (and the corresponding page range) into this swapdev's | |
11d31886 | 1805 | * extent list. The extent list is kept sorted in page order. |
1da177e4 | 1806 | * |
11d31886 | 1807 | * This function rather assumes that it is called in ascending page order. |
1da177e4 | 1808 | */ |
a509bc1a | 1809 | int |
1da177e4 LT |
1810 | add_swap_extent(struct swap_info_struct *sis, unsigned long start_page, |
1811 | unsigned long nr_pages, sector_t start_block) | |
1812 | { | |
1813 | struct swap_extent *se; | |
1814 | struct swap_extent *new_se; | |
1815 | struct list_head *lh; | |
1816 | ||
9625a5f2 HD |
1817 | if (start_page == 0) { |
1818 | se = &sis->first_swap_extent; | |
1819 | sis->curr_swap_extent = se; | |
1820 | se->start_page = 0; | |
1821 | se->nr_pages = nr_pages; | |
1822 | se->start_block = start_block; | |
1823 | return 1; | |
1824 | } else { | |
1825 | lh = sis->first_swap_extent.list.prev; /* Highest extent */ | |
1da177e4 | 1826 | se = list_entry(lh, struct swap_extent, list); |
11d31886 HD |
1827 | BUG_ON(se->start_page + se->nr_pages != start_page); |
1828 | if (se->start_block + se->nr_pages == start_block) { | |
1da177e4 LT |
1829 | /* Merge it */ |
1830 | se->nr_pages += nr_pages; | |
1831 | return 0; | |
1832 | } | |
1da177e4 LT |
1833 | } |
1834 | ||
1835 | /* | |
1836 | * No merge. Insert a new extent, preserving ordering. | |
1837 | */ | |
1838 | new_se = kmalloc(sizeof(*se), GFP_KERNEL); | |
1839 | if (new_se == NULL) | |
1840 | return -ENOMEM; | |
1841 | new_se->start_page = start_page; | |
1842 | new_se->nr_pages = nr_pages; | |
1843 | new_se->start_block = start_block; | |
1844 | ||
9625a5f2 | 1845 | list_add_tail(&new_se->list, &sis->first_swap_extent.list); |
53092a74 | 1846 | return 1; |
1da177e4 LT |
1847 | } |
1848 | ||
1849 | /* | |
1850 | * A `swap extent' is a simple thing which maps a contiguous range of pages | |
1851 | * onto a contiguous range of disk blocks. An ordered list of swap extents | |
1852 | * is built at swapon time and is then used at swap_writepage/swap_readpage | |
1853 | * time for locating where on disk a page belongs. | |
1854 | * | |
1855 | * If the swapfile is an S_ISBLK block device, a single extent is installed. | |
1856 | * This is done so that the main operating code can treat S_ISBLK and S_ISREG | |
1857 | * swap files identically. | |
1858 | * | |
1859 | * Whether the swapdev is an S_ISREG file or an S_ISBLK blockdev, the swap | |
1860 | * extent list operates in PAGE_SIZE disk blocks. Both S_ISREG and S_ISBLK | |
1861 | * swapfiles are handled *identically* after swapon time. | |
1862 | * | |
1863 | * For S_ISREG swapfiles, setup_swap_extents() will walk all the file's blocks | |
1864 | * and will parse them into an ordered extent list, in PAGE_SIZE chunks. If | |
1865 | * some stray blocks are found which do not fall within the PAGE_SIZE alignment | |
1866 | * requirements, they are simply tossed out - we will never use those blocks | |
1867 | * for swapping. | |
1868 | * | |
b0d9bcd4 | 1869 | * For S_ISREG swapfiles we set S_SWAPFILE across the life of the swapon. This |
1da177e4 LT |
1870 | * prevents root from shooting her foot off by ftruncating an in-use swapfile, |
1871 | * which will scribble on the fs. | |
1872 | * | |
1873 | * The amount of disk space which a single swap extent represents varies. | |
1874 | * Typically it is in the 1-4 megabyte range. So we can have hundreds of | |
1875 | * extents in the list. To avoid much list walking, we cache the previous | |
1876 | * search location in `curr_swap_extent', and start new searches from there. | |
1877 | * This is extremely effective. The average number of iterations in | |
1878 | * map_swap_page() has been measured at about 0.3 per page. - akpm. | |
1879 | */ | |
53092a74 | 1880 | static int setup_swap_extents(struct swap_info_struct *sis, sector_t *span) |
1da177e4 | 1881 | { |
62c230bc MG |
1882 | struct file *swap_file = sis->swap_file; |
1883 | struct address_space *mapping = swap_file->f_mapping; | |
1884 | struct inode *inode = mapping->host; | |
1da177e4 LT |
1885 | int ret; |
1886 | ||
1da177e4 LT |
1887 | if (S_ISBLK(inode->i_mode)) { |
1888 | ret = add_swap_extent(sis, 0, sis->max, 0); | |
53092a74 | 1889 | *span = sis->pages; |
a509bc1a | 1890 | return ret; |
1da177e4 LT |
1891 | } |
1892 | ||
62c230bc | 1893 | if (mapping->a_ops->swap_activate) { |
a509bc1a | 1894 | ret = mapping->a_ops->swap_activate(sis, swap_file, span); |
62c230bc MG |
1895 | if (!ret) { |
1896 | sis->flags |= SWP_FILE; | |
1897 | ret = add_swap_extent(sis, 0, sis->max, 0); | |
1898 | *span = sis->pages; | |
1899 | } | |
a509bc1a | 1900 | return ret; |
62c230bc MG |
1901 | } |
1902 | ||
a509bc1a | 1903 | return generic_swapfile_activate(sis, swap_file, span); |
1da177e4 LT |
1904 | } |
1905 | ||
cf0cac0a | 1906 | static void _enable_swap_info(struct swap_info_struct *p, int prio, |
2a8f9449 SL |
1907 | unsigned char *swap_map, |
1908 | struct swap_cluster_info *cluster_info) | |
40531542 | 1909 | { |
40531542 CEB |
1910 | if (prio >= 0) |
1911 | p->prio = prio; | |
1912 | else | |
1913 | p->prio = --least_priority; | |
18ab4d4c DS |
1914 | /* |
1915 | * the plist prio is negated because plist ordering is | |
1916 | * low-to-high, while swap ordering is high-to-low | |
1917 | */ | |
1918 | p->list.prio = -p->prio; | |
1919 | p->avail_list.prio = -p->prio; | |
40531542 | 1920 | p->swap_map = swap_map; |
2a8f9449 | 1921 | p->cluster_info = cluster_info; |
40531542 | 1922 | p->flags |= SWP_WRITEOK; |
ec8acf20 | 1923 | atomic_long_add(p->pages, &nr_swap_pages); |
40531542 CEB |
1924 | total_swap_pages += p->pages; |
1925 | ||
adfab836 | 1926 | assert_spin_locked(&swap_lock); |
adfab836 | 1927 | /* |
18ab4d4c DS |
1928 | * both lists are plists, and thus priority ordered. |
1929 | * swap_active_head needs to be priority ordered for swapoff(), | |
1930 | * which on removal of any swap_info_struct with an auto-assigned | |
1931 | * (i.e. negative) priority increments the auto-assigned priority | |
1932 | * of any lower-priority swap_info_structs. | |
1933 | * swap_avail_head needs to be priority ordered for get_swap_page(), | |
1934 | * which allocates swap pages from the highest available priority | |
1935 | * swap_info_struct. | |
adfab836 | 1936 | */ |
18ab4d4c DS |
1937 | plist_add(&p->list, &swap_active_head); |
1938 | spin_lock(&swap_avail_lock); | |
1939 | plist_add(&p->avail_list, &swap_avail_head); | |
1940 | spin_unlock(&swap_avail_lock); | |
cf0cac0a CEB |
1941 | } |
1942 | ||
1943 | static void enable_swap_info(struct swap_info_struct *p, int prio, | |
1944 | unsigned char *swap_map, | |
2a8f9449 | 1945 | struct swap_cluster_info *cluster_info, |
cf0cac0a CEB |
1946 | unsigned long *frontswap_map) |
1947 | { | |
4f89849d | 1948 | frontswap_init(p->type, frontswap_map); |
cf0cac0a | 1949 | spin_lock(&swap_lock); |
ec8acf20 | 1950 | spin_lock(&p->lock); |
2a8f9449 | 1951 | _enable_swap_info(p, prio, swap_map, cluster_info); |
ec8acf20 | 1952 | spin_unlock(&p->lock); |
cf0cac0a CEB |
1953 | spin_unlock(&swap_lock); |
1954 | } | |
1955 | ||
1956 | static void reinsert_swap_info(struct swap_info_struct *p) | |
1957 | { | |
1958 | spin_lock(&swap_lock); | |
ec8acf20 | 1959 | spin_lock(&p->lock); |
2a8f9449 | 1960 | _enable_swap_info(p, p->prio, p->swap_map, p->cluster_info); |
ec8acf20 | 1961 | spin_unlock(&p->lock); |
40531542 CEB |
1962 | spin_unlock(&swap_lock); |
1963 | } | |
1964 | ||
c4ea37c2 | 1965 | SYSCALL_DEFINE1(swapoff, const char __user *, specialfile) |
1da177e4 | 1966 | { |
73c34b6a | 1967 | struct swap_info_struct *p = NULL; |
8d69aaee | 1968 | unsigned char *swap_map; |
2a8f9449 | 1969 | struct swap_cluster_info *cluster_info; |
4f89849d | 1970 | unsigned long *frontswap_map; |
1da177e4 LT |
1971 | struct file *swap_file, *victim; |
1972 | struct address_space *mapping; | |
1973 | struct inode *inode; | |
91a27b2a | 1974 | struct filename *pathname; |
adfab836 | 1975 | int err, found = 0; |
5b808a23 | 1976 | unsigned int old_block_size; |
886bb7e9 | 1977 | |
1da177e4 LT |
1978 | if (!capable(CAP_SYS_ADMIN)) |
1979 | return -EPERM; | |
1980 | ||
191c5424 AV |
1981 | BUG_ON(!current->mm); |
1982 | ||
1da177e4 | 1983 | pathname = getname(specialfile); |
1da177e4 | 1984 | if (IS_ERR(pathname)) |
f58b59c1 | 1985 | return PTR_ERR(pathname); |
1da177e4 | 1986 | |
669abf4e | 1987 | victim = file_open_name(pathname, O_RDWR|O_LARGEFILE, 0); |
1da177e4 LT |
1988 | err = PTR_ERR(victim); |
1989 | if (IS_ERR(victim)) | |
1990 | goto out; | |
1991 | ||
1992 | mapping = victim->f_mapping; | |
5d337b91 | 1993 | spin_lock(&swap_lock); |
18ab4d4c | 1994 | plist_for_each_entry(p, &swap_active_head, list) { |
22c6f8fd | 1995 | if (p->flags & SWP_WRITEOK) { |
adfab836 DS |
1996 | if (p->swap_file->f_mapping == mapping) { |
1997 | found = 1; | |
1da177e4 | 1998 | break; |
adfab836 | 1999 | } |
1da177e4 | 2000 | } |
1da177e4 | 2001 | } |
adfab836 | 2002 | if (!found) { |
1da177e4 | 2003 | err = -EINVAL; |
5d337b91 | 2004 | spin_unlock(&swap_lock); |
1da177e4 LT |
2005 | goto out_dput; |
2006 | } | |
191c5424 | 2007 | if (!security_vm_enough_memory_mm(current->mm, p->pages)) |
1da177e4 LT |
2008 | vm_unacct_memory(p->pages); |
2009 | else { | |
2010 | err = -ENOMEM; | |
5d337b91 | 2011 | spin_unlock(&swap_lock); |
1da177e4 LT |
2012 | goto out_dput; |
2013 | } | |
18ab4d4c DS |
2014 | spin_lock(&swap_avail_lock); |
2015 | plist_del(&p->avail_list, &swap_avail_head); | |
2016 | spin_unlock(&swap_avail_lock); | |
ec8acf20 | 2017 | spin_lock(&p->lock); |
78ecba08 | 2018 | if (p->prio < 0) { |
adfab836 DS |
2019 | struct swap_info_struct *si = p; |
2020 | ||
18ab4d4c | 2021 | plist_for_each_entry_continue(si, &swap_active_head, list) { |
adfab836 | 2022 | si->prio++; |
18ab4d4c DS |
2023 | si->list.prio--; |
2024 | si->avail_list.prio--; | |
adfab836 | 2025 | } |
78ecba08 HD |
2026 | least_priority++; |
2027 | } | |
18ab4d4c | 2028 | plist_del(&p->list, &swap_active_head); |
ec8acf20 | 2029 | atomic_long_sub(p->pages, &nr_swap_pages); |
1da177e4 LT |
2030 | total_swap_pages -= p->pages; |
2031 | p->flags &= ~SWP_WRITEOK; | |
ec8acf20 | 2032 | spin_unlock(&p->lock); |
5d337b91 | 2033 | spin_unlock(&swap_lock); |
fb4f88dc | 2034 | |
e1e12d2f | 2035 | set_current_oom_origin(); |
adfab836 | 2036 | err = try_to_unuse(p->type, false, 0); /* force unuse all pages */ |
e1e12d2f | 2037 | clear_current_oom_origin(); |
1da177e4 | 2038 | |
1da177e4 LT |
2039 | if (err) { |
2040 | /* re-insert swap space back into swap_list */ | |
cf0cac0a | 2041 | reinsert_swap_info(p); |
1da177e4 LT |
2042 | goto out_dput; |
2043 | } | |
52b7efdb | 2044 | |
815c2c54 SL |
2045 | flush_work(&p->discard_work); |
2046 | ||
5d337b91 | 2047 | destroy_swap_extents(p); |
570a335b HD |
2048 | if (p->flags & SWP_CONTINUED) |
2049 | free_swap_count_continuations(p); | |
2050 | ||
fc0abb14 | 2051 | mutex_lock(&swapon_mutex); |
5d337b91 | 2052 | spin_lock(&swap_lock); |
ec8acf20 | 2053 | spin_lock(&p->lock); |
5d337b91 HD |
2054 | drain_mmlist(); |
2055 | ||
52b7efdb | 2056 | /* wait for anyone still in scan_swap_map */ |
52b7efdb HD |
2057 | p->highest_bit = 0; /* cuts scans short */ |
2058 | while (p->flags >= SWP_SCANNING) { | |
ec8acf20 | 2059 | spin_unlock(&p->lock); |
5d337b91 | 2060 | spin_unlock(&swap_lock); |
13e4b57f | 2061 | schedule_timeout_uninterruptible(1); |
5d337b91 | 2062 | spin_lock(&swap_lock); |
ec8acf20 | 2063 | spin_lock(&p->lock); |
52b7efdb | 2064 | } |
52b7efdb | 2065 | |
1da177e4 | 2066 | swap_file = p->swap_file; |
5b808a23 | 2067 | old_block_size = p->old_block_size; |
1da177e4 LT |
2068 | p->swap_file = NULL; |
2069 | p->max = 0; | |
2070 | swap_map = p->swap_map; | |
2071 | p->swap_map = NULL; | |
2a8f9449 SL |
2072 | cluster_info = p->cluster_info; |
2073 | p->cluster_info = NULL; | |
4f89849d | 2074 | frontswap_map = frontswap_map_get(p); |
ec8acf20 | 2075 | spin_unlock(&p->lock); |
5d337b91 | 2076 | spin_unlock(&swap_lock); |
adfab836 | 2077 | frontswap_invalidate_area(p->type); |
58e97ba6 | 2078 | frontswap_map_set(p, NULL); |
fc0abb14 | 2079 | mutex_unlock(&swapon_mutex); |
ebc2a1a6 SL |
2080 | free_percpu(p->percpu_cluster); |
2081 | p->percpu_cluster = NULL; | |
1da177e4 | 2082 | vfree(swap_map); |
2a8f9449 | 2083 | vfree(cluster_info); |
4f89849d | 2084 | vfree(frontswap_map); |
2de1a7e4 | 2085 | /* Destroy swap account information */ |
adfab836 | 2086 | swap_cgroup_swapoff(p->type); |
27a7faa0 | 2087 | |
1da177e4 LT |
2088 | inode = mapping->host; |
2089 | if (S_ISBLK(inode->i_mode)) { | |
2090 | struct block_device *bdev = I_BDEV(inode); | |
5b808a23 | 2091 | set_blocksize(bdev, old_block_size); |
e525fd89 | 2092 | blkdev_put(bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL); |
1da177e4 | 2093 | } else { |
5955102c | 2094 | inode_lock(inode); |
1da177e4 | 2095 | inode->i_flags &= ~S_SWAPFILE; |
5955102c | 2096 | inode_unlock(inode); |
1da177e4 LT |
2097 | } |
2098 | filp_close(swap_file, NULL); | |
f893ab41 WY |
2099 | |
2100 | /* | |
2101 | * Clear the SWP_USED flag after all resources are freed so that swapon | |
2102 | * can reuse this swap_info in alloc_swap_info() safely. It is ok to | |
2103 | * not hold p->lock after we cleared its SWP_WRITEOK. | |
2104 | */ | |
2105 | spin_lock(&swap_lock); | |
2106 | p->flags = 0; | |
2107 | spin_unlock(&swap_lock); | |
2108 | ||
1da177e4 | 2109 | err = 0; |
66d7dd51 KS |
2110 | atomic_inc(&proc_poll_event); |
2111 | wake_up_interruptible(&proc_poll_wait); | |
1da177e4 LT |
2112 | |
2113 | out_dput: | |
2114 | filp_close(victim, NULL); | |
2115 | out: | |
f58b59c1 | 2116 | putname(pathname); |
1da177e4 LT |
2117 | return err; |
2118 | } | |
2119 | ||
2120 | #ifdef CONFIG_PROC_FS | |
66d7dd51 KS |
2121 | static unsigned swaps_poll(struct file *file, poll_table *wait) |
2122 | { | |
f1514638 | 2123 | struct seq_file *seq = file->private_data; |
66d7dd51 KS |
2124 | |
2125 | poll_wait(file, &proc_poll_wait, wait); | |
2126 | ||
f1514638 KS |
2127 | if (seq->poll_event != atomic_read(&proc_poll_event)) { |
2128 | seq->poll_event = atomic_read(&proc_poll_event); | |
66d7dd51 KS |
2129 | return POLLIN | POLLRDNORM | POLLERR | POLLPRI; |
2130 | } | |
2131 | ||
2132 | return POLLIN | POLLRDNORM; | |
2133 | } | |
2134 | ||
1da177e4 LT |
2135 | /* iterator */ |
2136 | static void *swap_start(struct seq_file *swap, loff_t *pos) | |
2137 | { | |
efa90a98 HD |
2138 | struct swap_info_struct *si; |
2139 | int type; | |
1da177e4 LT |
2140 | loff_t l = *pos; |
2141 | ||
fc0abb14 | 2142 | mutex_lock(&swapon_mutex); |
1da177e4 | 2143 | |
881e4aab SS |
2144 | if (!l) |
2145 | return SEQ_START_TOKEN; | |
2146 | ||
efa90a98 HD |
2147 | for (type = 0; type < nr_swapfiles; type++) { |
2148 | smp_rmb(); /* read nr_swapfiles before swap_info[type] */ | |
2149 | si = swap_info[type]; | |
2150 | if (!(si->flags & SWP_USED) || !si->swap_map) | |
1da177e4 | 2151 | continue; |
881e4aab | 2152 | if (!--l) |
efa90a98 | 2153 | return si; |
1da177e4 LT |
2154 | } |
2155 | ||
2156 | return NULL; | |
2157 | } | |
2158 | ||
2159 | static void *swap_next(struct seq_file *swap, void *v, loff_t *pos) | |
2160 | { | |
efa90a98 HD |
2161 | struct swap_info_struct *si = v; |
2162 | int type; | |
1da177e4 | 2163 | |
881e4aab | 2164 | if (v == SEQ_START_TOKEN) |
efa90a98 HD |
2165 | type = 0; |
2166 | else | |
2167 | type = si->type + 1; | |
881e4aab | 2168 | |
efa90a98 HD |
2169 | for (; type < nr_swapfiles; type++) { |
2170 | smp_rmb(); /* read nr_swapfiles before swap_info[type] */ | |
2171 | si = swap_info[type]; | |
2172 | if (!(si->flags & SWP_USED) || !si->swap_map) | |
1da177e4 LT |
2173 | continue; |
2174 | ++*pos; | |
efa90a98 | 2175 | return si; |
1da177e4 LT |
2176 | } |
2177 | ||
2178 | return NULL; | |
2179 | } | |
2180 | ||
2181 | static void swap_stop(struct seq_file *swap, void *v) | |
2182 | { | |
fc0abb14 | 2183 | mutex_unlock(&swapon_mutex); |
1da177e4 LT |
2184 | } |
2185 | ||
2186 | static int swap_show(struct seq_file *swap, void *v) | |
2187 | { | |
efa90a98 | 2188 | struct swap_info_struct *si = v; |
1da177e4 LT |
2189 | struct file *file; |
2190 | int len; | |
2191 | ||
efa90a98 | 2192 | if (si == SEQ_START_TOKEN) { |
881e4aab SS |
2193 | seq_puts(swap,"Filename\t\t\t\tType\t\tSize\tUsed\tPriority\n"); |
2194 | return 0; | |
2195 | } | |
1da177e4 | 2196 | |
efa90a98 | 2197 | file = si->swap_file; |
2726d566 | 2198 | len = seq_file_path(swap, file, " \t\n\\"); |
6eb396dc | 2199 | seq_printf(swap, "%*s%s\t%u\t%u\t%d\n", |
886bb7e9 | 2200 | len < 40 ? 40 - len : 1, " ", |
496ad9aa | 2201 | S_ISBLK(file_inode(file)->i_mode) ? |
1da177e4 | 2202 | "partition" : "file\t", |
efa90a98 HD |
2203 | si->pages << (PAGE_SHIFT - 10), |
2204 | si->inuse_pages << (PAGE_SHIFT - 10), | |
2205 | si->prio); | |
1da177e4 LT |
2206 | return 0; |
2207 | } | |
2208 | ||
15ad7cdc | 2209 | static const struct seq_operations swaps_op = { |
1da177e4 LT |
2210 | .start = swap_start, |
2211 | .next = swap_next, | |
2212 | .stop = swap_stop, | |
2213 | .show = swap_show | |
2214 | }; | |
2215 | ||
2216 | static int swaps_open(struct inode *inode, struct file *file) | |
2217 | { | |
f1514638 | 2218 | struct seq_file *seq; |
66d7dd51 KS |
2219 | int ret; |
2220 | ||
66d7dd51 | 2221 | ret = seq_open(file, &swaps_op); |
f1514638 | 2222 | if (ret) |
66d7dd51 | 2223 | return ret; |
66d7dd51 | 2224 | |
f1514638 KS |
2225 | seq = file->private_data; |
2226 | seq->poll_event = atomic_read(&proc_poll_event); | |
2227 | return 0; | |
1da177e4 LT |
2228 | } |
2229 | ||
15ad7cdc | 2230 | static const struct file_operations proc_swaps_operations = { |
1da177e4 LT |
2231 | .open = swaps_open, |
2232 | .read = seq_read, | |
2233 | .llseek = seq_lseek, | |
2234 | .release = seq_release, | |
66d7dd51 | 2235 | .poll = swaps_poll, |
1da177e4 LT |
2236 | }; |
2237 | ||
2238 | static int __init procswaps_init(void) | |
2239 | { | |
3d71f86f | 2240 | proc_create("swaps", 0, NULL, &proc_swaps_operations); |
1da177e4 LT |
2241 | return 0; |
2242 | } | |
2243 | __initcall(procswaps_init); | |
2244 | #endif /* CONFIG_PROC_FS */ | |
2245 | ||
1796316a JB |
2246 | #ifdef MAX_SWAPFILES_CHECK |
2247 | static int __init max_swapfiles_check(void) | |
2248 | { | |
2249 | MAX_SWAPFILES_CHECK(); | |
2250 | return 0; | |
2251 | } | |
2252 | late_initcall(max_swapfiles_check); | |
2253 | #endif | |
2254 | ||
53cbb243 | 2255 | static struct swap_info_struct *alloc_swap_info(void) |
1da177e4 | 2256 | { |
73c34b6a | 2257 | struct swap_info_struct *p; |
1da177e4 | 2258 | unsigned int type; |
efa90a98 HD |
2259 | |
2260 | p = kzalloc(sizeof(*p), GFP_KERNEL); | |
2261 | if (!p) | |
53cbb243 | 2262 | return ERR_PTR(-ENOMEM); |
efa90a98 | 2263 | |
5d337b91 | 2264 | spin_lock(&swap_lock); |
efa90a98 HD |
2265 | for (type = 0; type < nr_swapfiles; type++) { |
2266 | if (!(swap_info[type]->flags & SWP_USED)) | |
1da177e4 | 2267 | break; |
efa90a98 | 2268 | } |
0697212a | 2269 | if (type >= MAX_SWAPFILES) { |
5d337b91 | 2270 | spin_unlock(&swap_lock); |
efa90a98 | 2271 | kfree(p); |
730c0581 | 2272 | return ERR_PTR(-EPERM); |
1da177e4 | 2273 | } |
efa90a98 HD |
2274 | if (type >= nr_swapfiles) { |
2275 | p->type = type; | |
2276 | swap_info[type] = p; | |
2277 | /* | |
2278 | * Write swap_info[type] before nr_swapfiles, in case a | |
2279 | * racing procfs swap_start() or swap_next() is reading them. | |
2280 | * (We never shrink nr_swapfiles, we never free this entry.) | |
2281 | */ | |
2282 | smp_wmb(); | |
2283 | nr_swapfiles++; | |
2284 | } else { | |
2285 | kfree(p); | |
2286 | p = swap_info[type]; | |
2287 | /* | |
2288 | * Do not memset this entry: a racing procfs swap_next() | |
2289 | * would be relying on p->type to remain valid. | |
2290 | */ | |
2291 | } | |
9625a5f2 | 2292 | INIT_LIST_HEAD(&p->first_swap_extent.list); |
18ab4d4c DS |
2293 | plist_node_init(&p->list, 0); |
2294 | plist_node_init(&p->avail_list, 0); | |
1da177e4 | 2295 | p->flags = SWP_USED; |
5d337b91 | 2296 | spin_unlock(&swap_lock); |
ec8acf20 | 2297 | spin_lock_init(&p->lock); |
efa90a98 | 2298 | |
53cbb243 | 2299 | return p; |
53cbb243 CEB |
2300 | } |
2301 | ||
4d0e1e10 CEB |
2302 | static int claim_swapfile(struct swap_info_struct *p, struct inode *inode) |
2303 | { | |
2304 | int error; | |
2305 | ||
2306 | if (S_ISBLK(inode->i_mode)) { | |
2307 | p->bdev = bdgrab(I_BDEV(inode)); | |
2308 | error = blkdev_get(p->bdev, | |
6f179af8 | 2309 | FMODE_READ | FMODE_WRITE | FMODE_EXCL, p); |
4d0e1e10 CEB |
2310 | if (error < 0) { |
2311 | p->bdev = NULL; | |
6f179af8 | 2312 | return error; |
4d0e1e10 CEB |
2313 | } |
2314 | p->old_block_size = block_size(p->bdev); | |
2315 | error = set_blocksize(p->bdev, PAGE_SIZE); | |
2316 | if (error < 0) | |
87ade72a | 2317 | return error; |
4d0e1e10 CEB |
2318 | p->flags |= SWP_BLKDEV; |
2319 | } else if (S_ISREG(inode->i_mode)) { | |
2320 | p->bdev = inode->i_sb->s_bdev; | |
5955102c | 2321 | inode_lock(inode); |
87ade72a CEB |
2322 | if (IS_SWAPFILE(inode)) |
2323 | return -EBUSY; | |
2324 | } else | |
2325 | return -EINVAL; | |
4d0e1e10 CEB |
2326 | |
2327 | return 0; | |
4d0e1e10 CEB |
2328 | } |
2329 | ||
ca8bd38b CEB |
2330 | static unsigned long read_swap_header(struct swap_info_struct *p, |
2331 | union swap_header *swap_header, | |
2332 | struct inode *inode) | |
2333 | { | |
2334 | int i; | |
2335 | unsigned long maxpages; | |
2336 | unsigned long swapfilepages; | |
d6bbbd29 | 2337 | unsigned long last_page; |
ca8bd38b CEB |
2338 | |
2339 | if (memcmp("SWAPSPACE2", swap_header->magic.magic, 10)) { | |
465c47fd | 2340 | pr_err("Unable to find swap-space signature\n"); |
38719025 | 2341 | return 0; |
ca8bd38b CEB |
2342 | } |
2343 | ||
2344 | /* swap partition endianess hack... */ | |
2345 | if (swab32(swap_header->info.version) == 1) { | |
2346 | swab32s(&swap_header->info.version); | |
2347 | swab32s(&swap_header->info.last_page); | |
2348 | swab32s(&swap_header->info.nr_badpages); | |
dd111be6 JH |
2349 | if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES) |
2350 | return 0; | |
ca8bd38b CEB |
2351 | for (i = 0; i < swap_header->info.nr_badpages; i++) |
2352 | swab32s(&swap_header->info.badpages[i]); | |
2353 | } | |
2354 | /* Check the swap header's sub-version */ | |
2355 | if (swap_header->info.version != 1) { | |
465c47fd AM |
2356 | pr_warn("Unable to handle swap header version %d\n", |
2357 | swap_header->info.version); | |
38719025 | 2358 | return 0; |
ca8bd38b CEB |
2359 | } |
2360 | ||
2361 | p->lowest_bit = 1; | |
2362 | p->cluster_next = 1; | |
2363 | p->cluster_nr = 0; | |
2364 | ||
2365 | /* | |
2366 | * Find out how many pages are allowed for a single swap | |
9b15b817 | 2367 | * device. There are two limiting factors: 1) the number |
a2c16d6c HD |
2368 | * of bits for the swap offset in the swp_entry_t type, and |
2369 | * 2) the number of bits in the swap pte as defined by the | |
9b15b817 | 2370 | * different architectures. In order to find the |
a2c16d6c | 2371 | * largest possible bit mask, a swap entry with swap type 0 |
ca8bd38b | 2372 | * and swap offset ~0UL is created, encoded to a swap pte, |
a2c16d6c | 2373 | * decoded to a swp_entry_t again, and finally the swap |
ca8bd38b CEB |
2374 | * offset is extracted. This will mask all the bits from |
2375 | * the initial ~0UL mask that can't be encoded in either | |
2376 | * the swp_entry_t or the architecture definition of a | |
9b15b817 | 2377 | * swap pte. |
ca8bd38b CEB |
2378 | */ |
2379 | maxpages = swp_offset(pte_to_swp_entry( | |
9b15b817 | 2380 | swp_entry_to_pte(swp_entry(0, ~0UL)))) + 1; |
d6bbbd29 RJ |
2381 | last_page = swap_header->info.last_page; |
2382 | if (last_page > maxpages) { | |
465c47fd | 2383 | pr_warn("Truncating oversized swap area, only using %luk out of %luk\n", |
d6bbbd29 RJ |
2384 | maxpages << (PAGE_SHIFT - 10), |
2385 | last_page << (PAGE_SHIFT - 10)); | |
2386 | } | |
2387 | if (maxpages > last_page) { | |
2388 | maxpages = last_page + 1; | |
ca8bd38b CEB |
2389 | /* p->max is an unsigned int: don't overflow it */ |
2390 | if ((unsigned int)maxpages == 0) | |
2391 | maxpages = UINT_MAX; | |
2392 | } | |
2393 | p->highest_bit = maxpages - 1; | |
2394 | ||
2395 | if (!maxpages) | |
38719025 | 2396 | return 0; |
ca8bd38b CEB |
2397 | swapfilepages = i_size_read(inode) >> PAGE_SHIFT; |
2398 | if (swapfilepages && maxpages > swapfilepages) { | |
465c47fd | 2399 | pr_warn("Swap area shorter than signature indicates\n"); |
38719025 | 2400 | return 0; |
ca8bd38b CEB |
2401 | } |
2402 | if (swap_header->info.nr_badpages && S_ISREG(inode->i_mode)) | |
38719025 | 2403 | return 0; |
ca8bd38b | 2404 | if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES) |
38719025 | 2405 | return 0; |
ca8bd38b CEB |
2406 | |
2407 | return maxpages; | |
ca8bd38b CEB |
2408 | } |
2409 | ||
235b6217 HY |
2410 | #define SWAP_CLUSTER_COLS \ |
2411 | DIV_ROUND_UP(L1_CACHE_BYTES, sizeof(struct swap_cluster_info)) | |
2412 | ||
915d4d7b CEB |
2413 | static int setup_swap_map_and_extents(struct swap_info_struct *p, |
2414 | union swap_header *swap_header, | |
2415 | unsigned char *swap_map, | |
2a8f9449 | 2416 | struct swap_cluster_info *cluster_info, |
915d4d7b CEB |
2417 | unsigned long maxpages, |
2418 | sector_t *span) | |
2419 | { | |
235b6217 | 2420 | unsigned int j, k; |
915d4d7b CEB |
2421 | unsigned int nr_good_pages; |
2422 | int nr_extents; | |
2a8f9449 | 2423 | unsigned long nr_clusters = DIV_ROUND_UP(maxpages, SWAPFILE_CLUSTER); |
235b6217 HY |
2424 | unsigned long col = p->cluster_next / SWAPFILE_CLUSTER % SWAP_CLUSTER_COLS; |
2425 | unsigned long i, idx; | |
915d4d7b CEB |
2426 | |
2427 | nr_good_pages = maxpages - 1; /* omit header page */ | |
2428 | ||
6b534915 HY |
2429 | cluster_list_init(&p->free_clusters); |
2430 | cluster_list_init(&p->discard_clusters); | |
2a8f9449 | 2431 | |
915d4d7b CEB |
2432 | for (i = 0; i < swap_header->info.nr_badpages; i++) { |
2433 | unsigned int page_nr = swap_header->info.badpages[i]; | |
bdb8e3f6 CEB |
2434 | if (page_nr == 0 || page_nr > swap_header->info.last_page) |
2435 | return -EINVAL; | |
915d4d7b CEB |
2436 | if (page_nr < maxpages) { |
2437 | swap_map[page_nr] = SWAP_MAP_BAD; | |
2438 | nr_good_pages--; | |
2a8f9449 SL |
2439 | /* |
2440 | * Haven't marked the cluster free yet, no list | |
2441 | * operation involved | |
2442 | */ | |
2443 | inc_cluster_info_page(p, cluster_info, page_nr); | |
915d4d7b CEB |
2444 | } |
2445 | } | |
2446 | ||
2a8f9449 SL |
2447 | /* Haven't marked the cluster free yet, no list operation involved */ |
2448 | for (i = maxpages; i < round_up(maxpages, SWAPFILE_CLUSTER); i++) | |
2449 | inc_cluster_info_page(p, cluster_info, i); | |
2450 | ||
915d4d7b CEB |
2451 | if (nr_good_pages) { |
2452 | swap_map[0] = SWAP_MAP_BAD; | |
2a8f9449 SL |
2453 | /* |
2454 | * Not mark the cluster free yet, no list | |
2455 | * operation involved | |
2456 | */ | |
2457 | inc_cluster_info_page(p, cluster_info, 0); | |
915d4d7b CEB |
2458 | p->max = maxpages; |
2459 | p->pages = nr_good_pages; | |
2460 | nr_extents = setup_swap_extents(p, span); | |
bdb8e3f6 CEB |
2461 | if (nr_extents < 0) |
2462 | return nr_extents; | |
915d4d7b CEB |
2463 | nr_good_pages = p->pages; |
2464 | } | |
2465 | if (!nr_good_pages) { | |
465c47fd | 2466 | pr_warn("Empty swap-file\n"); |
bdb8e3f6 | 2467 | return -EINVAL; |
915d4d7b CEB |
2468 | } |
2469 | ||
2a8f9449 SL |
2470 | if (!cluster_info) |
2471 | return nr_extents; | |
2472 | ||
235b6217 HY |
2473 | |
2474 | /* Reduce false cache line sharing between cluster_info */ | |
2475 | for (k = 0; k < SWAP_CLUSTER_COLS; k++) { | |
2476 | j = (k + col) % SWAP_CLUSTER_COLS; | |
2477 | for (i = 0; i < DIV_ROUND_UP(nr_clusters, SWAP_CLUSTER_COLS); i++) { | |
2478 | idx = i * SWAP_CLUSTER_COLS + j; | |
2479 | if (idx >= nr_clusters) | |
2480 | continue; | |
2481 | if (cluster_count(&cluster_info[idx])) | |
2482 | continue; | |
2a8f9449 | 2483 | cluster_set_flag(&cluster_info[idx], CLUSTER_FLAG_FREE); |
6b534915 HY |
2484 | cluster_list_add_tail(&p->free_clusters, cluster_info, |
2485 | idx); | |
2a8f9449 | 2486 | } |
2a8f9449 | 2487 | } |
915d4d7b | 2488 | return nr_extents; |
915d4d7b CEB |
2489 | } |
2490 | ||
dcf6b7dd RA |
2491 | /* |
2492 | * Helper to sys_swapon determining if a given swap | |
2493 | * backing device queue supports DISCARD operations. | |
2494 | */ | |
2495 | static bool swap_discardable(struct swap_info_struct *si) | |
2496 | { | |
2497 | struct request_queue *q = bdev_get_queue(si->bdev); | |
2498 | ||
2499 | if (!q || !blk_queue_discard(q)) | |
2500 | return false; | |
2501 | ||
2502 | return true; | |
2503 | } | |
2504 | ||
53cbb243 CEB |
2505 | SYSCALL_DEFINE2(swapon, const char __user *, specialfile, int, swap_flags) |
2506 | { | |
2507 | struct swap_info_struct *p; | |
91a27b2a | 2508 | struct filename *name; |
53cbb243 CEB |
2509 | struct file *swap_file = NULL; |
2510 | struct address_space *mapping; | |
40531542 | 2511 | int prio; |
53cbb243 CEB |
2512 | int error; |
2513 | union swap_header *swap_header; | |
915d4d7b | 2514 | int nr_extents; |
53cbb243 CEB |
2515 | sector_t span; |
2516 | unsigned long maxpages; | |
53cbb243 | 2517 | unsigned char *swap_map = NULL; |
2a8f9449 | 2518 | struct swap_cluster_info *cluster_info = NULL; |
38b5faf4 | 2519 | unsigned long *frontswap_map = NULL; |
53cbb243 CEB |
2520 | struct page *page = NULL; |
2521 | struct inode *inode = NULL; | |
53cbb243 | 2522 | |
d15cab97 HD |
2523 | if (swap_flags & ~SWAP_FLAGS_VALID) |
2524 | return -EINVAL; | |
2525 | ||
53cbb243 CEB |
2526 | if (!capable(CAP_SYS_ADMIN)) |
2527 | return -EPERM; | |
2528 | ||
2529 | p = alloc_swap_info(); | |
2542e513 CEB |
2530 | if (IS_ERR(p)) |
2531 | return PTR_ERR(p); | |
53cbb243 | 2532 | |
815c2c54 SL |
2533 | INIT_WORK(&p->discard_work, swap_discard_work); |
2534 | ||
1da177e4 | 2535 | name = getname(specialfile); |
1da177e4 | 2536 | if (IS_ERR(name)) { |
7de7fb6b | 2537 | error = PTR_ERR(name); |
1da177e4 | 2538 | name = NULL; |
bd69010b | 2539 | goto bad_swap; |
1da177e4 | 2540 | } |
669abf4e | 2541 | swap_file = file_open_name(name, O_RDWR|O_LARGEFILE, 0); |
1da177e4 | 2542 | if (IS_ERR(swap_file)) { |
7de7fb6b | 2543 | error = PTR_ERR(swap_file); |
1da177e4 | 2544 | swap_file = NULL; |
bd69010b | 2545 | goto bad_swap; |
1da177e4 LT |
2546 | } |
2547 | ||
2548 | p->swap_file = swap_file; | |
2549 | mapping = swap_file->f_mapping; | |
2130781e | 2550 | inode = mapping->host; |
6f179af8 | 2551 | |
5955102c | 2552 | /* If S_ISREG(inode->i_mode) will do inode_lock(inode); */ |
4d0e1e10 CEB |
2553 | error = claim_swapfile(p, inode); |
2554 | if (unlikely(error)) | |
1da177e4 | 2555 | goto bad_swap; |
1da177e4 | 2556 | |
1da177e4 LT |
2557 | /* |
2558 | * Read the swap header. | |
2559 | */ | |
2560 | if (!mapping->a_ops->readpage) { | |
2561 | error = -EINVAL; | |
2562 | goto bad_swap; | |
2563 | } | |
090d2b18 | 2564 | page = read_mapping_page(mapping, 0, swap_file); |
1da177e4 LT |
2565 | if (IS_ERR(page)) { |
2566 | error = PTR_ERR(page); | |
2567 | goto bad_swap; | |
2568 | } | |
81e33971 | 2569 | swap_header = kmap(page); |
1da177e4 | 2570 | |
ca8bd38b CEB |
2571 | maxpages = read_swap_header(p, swap_header, inode); |
2572 | if (unlikely(!maxpages)) { | |
1da177e4 LT |
2573 | error = -EINVAL; |
2574 | goto bad_swap; | |
2575 | } | |
886bb7e9 | 2576 | |
81e33971 | 2577 | /* OK, set up the swap map and apply the bad block list */ |
803d0c83 | 2578 | swap_map = vzalloc(maxpages); |
81e33971 HD |
2579 | if (!swap_map) { |
2580 | error = -ENOMEM; | |
2581 | goto bad_swap; | |
2582 | } | |
f0571429 MK |
2583 | |
2584 | if (bdi_cap_stable_pages_required(inode_to_bdi(inode))) | |
2585 | p->flags |= SWP_STABLE_WRITES; | |
2586 | ||
2a8f9449 | 2587 | if (p->bdev && blk_queue_nonrot(bdev_get_queue(p->bdev))) { |
6f179af8 | 2588 | int cpu; |
235b6217 | 2589 | unsigned long ci, nr_cluster; |
6f179af8 | 2590 | |
2a8f9449 SL |
2591 | p->flags |= SWP_SOLIDSTATE; |
2592 | /* | |
2593 | * select a random position to start with to help wear leveling | |
2594 | * SSD | |
2595 | */ | |
2596 | p->cluster_next = 1 + (prandom_u32() % p->highest_bit); | |
235b6217 | 2597 | nr_cluster = DIV_ROUND_UP(maxpages, SWAPFILE_CLUSTER); |
2a8f9449 | 2598 | |
235b6217 | 2599 | cluster_info = vzalloc(nr_cluster * sizeof(*cluster_info)); |
2a8f9449 SL |
2600 | if (!cluster_info) { |
2601 | error = -ENOMEM; | |
2602 | goto bad_swap; | |
2603 | } | |
235b6217 HY |
2604 | |
2605 | for (ci = 0; ci < nr_cluster; ci++) | |
2606 | spin_lock_init(&((cluster_info + ci)->lock)); | |
2607 | ||
ebc2a1a6 SL |
2608 | p->percpu_cluster = alloc_percpu(struct percpu_cluster); |
2609 | if (!p->percpu_cluster) { | |
2610 | error = -ENOMEM; | |
2611 | goto bad_swap; | |
2612 | } | |
6f179af8 | 2613 | for_each_possible_cpu(cpu) { |
ebc2a1a6 | 2614 | struct percpu_cluster *cluster; |
6f179af8 | 2615 | cluster = per_cpu_ptr(p->percpu_cluster, cpu); |
ebc2a1a6 SL |
2616 | cluster_set_null(&cluster->index); |
2617 | } | |
2a8f9449 | 2618 | } |
1da177e4 | 2619 | |
1421ef3c CEB |
2620 | error = swap_cgroup_swapon(p->type, maxpages); |
2621 | if (error) | |
2622 | goto bad_swap; | |
2623 | ||
915d4d7b | 2624 | nr_extents = setup_swap_map_and_extents(p, swap_header, swap_map, |
2a8f9449 | 2625 | cluster_info, maxpages, &span); |
915d4d7b CEB |
2626 | if (unlikely(nr_extents < 0)) { |
2627 | error = nr_extents; | |
1da177e4 LT |
2628 | goto bad_swap; |
2629 | } | |
38b5faf4 | 2630 | /* frontswap enabled? set up bit-per-page map for frontswap */ |
8ea1d2a1 | 2631 | if (IS_ENABLED(CONFIG_FRONTSWAP)) |
7b57976d | 2632 | frontswap_map = vzalloc(BITS_TO_LONGS(maxpages) * sizeof(long)); |
1da177e4 | 2633 | |
2a8f9449 SL |
2634 | if (p->bdev &&(swap_flags & SWAP_FLAG_DISCARD) && swap_discardable(p)) { |
2635 | /* | |
2636 | * When discard is enabled for swap with no particular | |
2637 | * policy flagged, we set all swap discard flags here in | |
2638 | * order to sustain backward compatibility with older | |
2639 | * swapon(8) releases. | |
2640 | */ | |
2641 | p->flags |= (SWP_DISCARDABLE | SWP_AREA_DISCARD | | |
2642 | SWP_PAGE_DISCARD); | |
dcf6b7dd | 2643 | |
2a8f9449 SL |
2644 | /* |
2645 | * By flagging sys_swapon, a sysadmin can tell us to | |
2646 | * either do single-time area discards only, or to just | |
2647 | * perform discards for released swap page-clusters. | |
2648 | * Now it's time to adjust the p->flags accordingly. | |
2649 | */ | |
2650 | if (swap_flags & SWAP_FLAG_DISCARD_ONCE) | |
2651 | p->flags &= ~SWP_PAGE_DISCARD; | |
2652 | else if (swap_flags & SWAP_FLAG_DISCARD_PAGES) | |
2653 | p->flags &= ~SWP_AREA_DISCARD; | |
2654 | ||
2655 | /* issue a swapon-time discard if it's still required */ | |
2656 | if (p->flags & SWP_AREA_DISCARD) { | |
2657 | int err = discard_swap(p); | |
2658 | if (unlikely(err)) | |
2659 | pr_err("swapon: discard_swap(%p): %d\n", | |
2660 | p, err); | |
dcf6b7dd | 2661 | } |
20137a49 | 2662 | } |
6a6ba831 | 2663 | |
fc0abb14 | 2664 | mutex_lock(&swapon_mutex); |
40531542 | 2665 | prio = -1; |
78ecba08 | 2666 | if (swap_flags & SWAP_FLAG_PREFER) |
40531542 | 2667 | prio = |
78ecba08 | 2668 | (swap_flags & SWAP_FLAG_PRIO_MASK) >> SWAP_FLAG_PRIO_SHIFT; |
2a8f9449 | 2669 | enable_swap_info(p, prio, swap_map, cluster_info, frontswap_map); |
c69dbfb8 | 2670 | |
756a025f | 2671 | pr_info("Adding %uk swap on %s. Priority:%d extents:%d across:%lluk %s%s%s%s%s\n", |
91a27b2a | 2672 | p->pages<<(PAGE_SHIFT-10), name->name, p->prio, |
c69dbfb8 CEB |
2673 | nr_extents, (unsigned long long)span<<(PAGE_SHIFT-10), |
2674 | (p->flags & SWP_SOLIDSTATE) ? "SS" : "", | |
38b5faf4 | 2675 | (p->flags & SWP_DISCARDABLE) ? "D" : "", |
dcf6b7dd RA |
2676 | (p->flags & SWP_AREA_DISCARD) ? "s" : "", |
2677 | (p->flags & SWP_PAGE_DISCARD) ? "c" : "", | |
38b5faf4 | 2678 | (frontswap_map) ? "FS" : ""); |
c69dbfb8 | 2679 | |
fc0abb14 | 2680 | mutex_unlock(&swapon_mutex); |
66d7dd51 KS |
2681 | atomic_inc(&proc_poll_event); |
2682 | wake_up_interruptible(&proc_poll_wait); | |
2683 | ||
9b01c350 CEB |
2684 | if (S_ISREG(inode->i_mode)) |
2685 | inode->i_flags |= S_SWAPFILE; | |
1da177e4 LT |
2686 | error = 0; |
2687 | goto out; | |
2688 | bad_swap: | |
ebc2a1a6 SL |
2689 | free_percpu(p->percpu_cluster); |
2690 | p->percpu_cluster = NULL; | |
bd69010b | 2691 | if (inode && S_ISBLK(inode->i_mode) && p->bdev) { |
f2090d2d CEB |
2692 | set_blocksize(p->bdev, p->old_block_size); |
2693 | blkdev_put(p->bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL); | |
1da177e4 | 2694 | } |
4cd3bb10 | 2695 | destroy_swap_extents(p); |
e8e6c2ec | 2696 | swap_cgroup_swapoff(p->type); |
5d337b91 | 2697 | spin_lock(&swap_lock); |
1da177e4 | 2698 | p->swap_file = NULL; |
1da177e4 | 2699 | p->flags = 0; |
5d337b91 | 2700 | spin_unlock(&swap_lock); |
1da177e4 | 2701 | vfree(swap_map); |
2a8f9449 | 2702 | vfree(cluster_info); |
52c50567 | 2703 | if (swap_file) { |
2130781e | 2704 | if (inode && S_ISREG(inode->i_mode)) { |
5955102c | 2705 | inode_unlock(inode); |
2130781e CEB |
2706 | inode = NULL; |
2707 | } | |
1da177e4 | 2708 | filp_close(swap_file, NULL); |
52c50567 | 2709 | } |
1da177e4 LT |
2710 | out: |
2711 | if (page && !IS_ERR(page)) { | |
2712 | kunmap(page); | |
09cbfeaf | 2713 | put_page(page); |
1da177e4 LT |
2714 | } |
2715 | if (name) | |
2716 | putname(name); | |
9b01c350 | 2717 | if (inode && S_ISREG(inode->i_mode)) |
5955102c | 2718 | inode_unlock(inode); |
1da177e4 LT |
2719 | return error; |
2720 | } | |
2721 | ||
2722 | void si_swapinfo(struct sysinfo *val) | |
2723 | { | |
efa90a98 | 2724 | unsigned int type; |
1da177e4 LT |
2725 | unsigned long nr_to_be_unused = 0; |
2726 | ||
5d337b91 | 2727 | spin_lock(&swap_lock); |
efa90a98 HD |
2728 | for (type = 0; type < nr_swapfiles; type++) { |
2729 | struct swap_info_struct *si = swap_info[type]; | |
2730 | ||
2731 | if ((si->flags & SWP_USED) && !(si->flags & SWP_WRITEOK)) | |
2732 | nr_to_be_unused += si->inuse_pages; | |
1da177e4 | 2733 | } |
ec8acf20 | 2734 | val->freeswap = atomic_long_read(&nr_swap_pages) + nr_to_be_unused; |
1da177e4 | 2735 | val->totalswap = total_swap_pages + nr_to_be_unused; |
5d337b91 | 2736 | spin_unlock(&swap_lock); |
1da177e4 LT |
2737 | } |
2738 | ||
2739 | /* | |
2740 | * Verify that a swap entry is valid and increment its swap map count. | |
2741 | * | |
355cfa73 KH |
2742 | * Returns error code in following case. |
2743 | * - success -> 0 | |
2744 | * - swp_entry is invalid -> EINVAL | |
2745 | * - swp_entry is migration entry -> EINVAL | |
2746 | * - swap-cache reference is requested but there is already one. -> EEXIST | |
2747 | * - swap-cache reference is requested but the entry is not used. -> ENOENT | |
570a335b | 2748 | * - swap-mapped reference requested but needs continued swap count. -> ENOMEM |
1da177e4 | 2749 | */ |
8d69aaee | 2750 | static int __swap_duplicate(swp_entry_t entry, unsigned char usage) |
1da177e4 | 2751 | { |
73c34b6a | 2752 | struct swap_info_struct *p; |
235b6217 | 2753 | struct swap_cluster_info *ci; |
1da177e4 | 2754 | unsigned long offset, type; |
8d69aaee HD |
2755 | unsigned char count; |
2756 | unsigned char has_cache; | |
253d553b | 2757 | int err = -EINVAL; |
1da177e4 | 2758 | |
a7420aa5 | 2759 | if (non_swap_entry(entry)) |
253d553b | 2760 | goto out; |
0697212a | 2761 | |
1da177e4 LT |
2762 | type = swp_type(entry); |
2763 | if (type >= nr_swapfiles) | |
2764 | goto bad_file; | |
efa90a98 | 2765 | p = swap_info[type]; |
1da177e4 | 2766 | offset = swp_offset(entry); |
355cfa73 | 2767 | if (unlikely(offset >= p->max)) |
235b6217 HY |
2768 | goto out; |
2769 | ||
2770 | ci = lock_cluster_or_swap_info(p, offset); | |
355cfa73 | 2771 | |
253d553b | 2772 | count = p->swap_map[offset]; |
edfe23da SL |
2773 | |
2774 | /* | |
2775 | * swapin_readahead() doesn't check if a swap entry is valid, so the | |
2776 | * swap entry could be SWAP_MAP_BAD. Check here with lock held. | |
2777 | */ | |
2778 | if (unlikely(swap_count(count) == SWAP_MAP_BAD)) { | |
2779 | err = -ENOENT; | |
2780 | goto unlock_out; | |
2781 | } | |
2782 | ||
253d553b HD |
2783 | has_cache = count & SWAP_HAS_CACHE; |
2784 | count &= ~SWAP_HAS_CACHE; | |
2785 | err = 0; | |
355cfa73 | 2786 | |
253d553b | 2787 | if (usage == SWAP_HAS_CACHE) { |
355cfa73 KH |
2788 | |
2789 | /* set SWAP_HAS_CACHE if there is no cache and entry is used */ | |
253d553b HD |
2790 | if (!has_cache && count) |
2791 | has_cache = SWAP_HAS_CACHE; | |
2792 | else if (has_cache) /* someone else added cache */ | |
2793 | err = -EEXIST; | |
2794 | else /* no users remaining */ | |
2795 | err = -ENOENT; | |
355cfa73 KH |
2796 | |
2797 | } else if (count || has_cache) { | |
253d553b | 2798 | |
570a335b HD |
2799 | if ((count & ~COUNT_CONTINUED) < SWAP_MAP_MAX) |
2800 | count += usage; | |
2801 | else if ((count & ~COUNT_CONTINUED) > SWAP_MAP_MAX) | |
253d553b | 2802 | err = -EINVAL; |
570a335b HD |
2803 | else if (swap_count_continued(p, offset, count)) |
2804 | count = COUNT_CONTINUED; | |
2805 | else | |
2806 | err = -ENOMEM; | |
355cfa73 | 2807 | } else |
253d553b HD |
2808 | err = -ENOENT; /* unused swap entry */ |
2809 | ||
2810 | p->swap_map[offset] = count | has_cache; | |
2811 | ||
355cfa73 | 2812 | unlock_out: |
235b6217 | 2813 | unlock_cluster_or_swap_info(p, ci); |
1da177e4 | 2814 | out: |
253d553b | 2815 | return err; |
1da177e4 LT |
2816 | |
2817 | bad_file: | |
465c47fd | 2818 | pr_err("swap_dup: %s%08lx\n", Bad_file, entry.val); |
1da177e4 LT |
2819 | goto out; |
2820 | } | |
253d553b | 2821 | |
aaa46865 HD |
2822 | /* |
2823 | * Help swapoff by noting that swap entry belongs to shmem/tmpfs | |
2824 | * (in which case its reference count is never incremented). | |
2825 | */ | |
2826 | void swap_shmem_alloc(swp_entry_t entry) | |
2827 | { | |
2828 | __swap_duplicate(entry, SWAP_MAP_SHMEM); | |
2829 | } | |
2830 | ||
355cfa73 | 2831 | /* |
08259d58 HD |
2832 | * Increase reference count of swap entry by 1. |
2833 | * Returns 0 for success, or -ENOMEM if a swap_count_continuation is required | |
2834 | * but could not be atomically allocated. Returns 0, just as if it succeeded, | |
2835 | * if __swap_duplicate() fails for another reason (-EINVAL or -ENOENT), which | |
2836 | * might occur if a page table entry has got corrupted. | |
355cfa73 | 2837 | */ |
570a335b | 2838 | int swap_duplicate(swp_entry_t entry) |
355cfa73 | 2839 | { |
570a335b HD |
2840 | int err = 0; |
2841 | ||
2842 | while (!err && __swap_duplicate(entry, 1) == -ENOMEM) | |
2843 | err = add_swap_count_continuation(entry, GFP_ATOMIC); | |
2844 | return err; | |
355cfa73 | 2845 | } |
1da177e4 | 2846 | |
cb4b86ba | 2847 | /* |
355cfa73 KH |
2848 | * @entry: swap entry for which we allocate swap cache. |
2849 | * | |
73c34b6a | 2850 | * Called when allocating swap cache for existing swap entry, |
355cfa73 KH |
2851 | * This can return error codes. Returns 0 at success. |
2852 | * -EBUSY means there is a swap cache. | |
2853 | * Note: return code is different from swap_duplicate(). | |
cb4b86ba KH |
2854 | */ |
2855 | int swapcache_prepare(swp_entry_t entry) | |
2856 | { | |
253d553b | 2857 | return __swap_duplicate(entry, SWAP_HAS_CACHE); |
cb4b86ba KH |
2858 | } |
2859 | ||
f981c595 MG |
2860 | struct swap_info_struct *page_swap_info(struct page *page) |
2861 | { | |
2862 | swp_entry_t swap = { .val = page_private(page) }; | |
f981c595 MG |
2863 | return swap_info[swp_type(swap)]; |
2864 | } | |
2865 | ||
2866 | /* | |
2867 | * out-of-line __page_file_ methods to avoid include hell. | |
2868 | */ | |
2869 | struct address_space *__page_file_mapping(struct page *page) | |
2870 | { | |
309381fe | 2871 | VM_BUG_ON_PAGE(!PageSwapCache(page), page); |
f981c595 MG |
2872 | return page_swap_info(page)->swap_file->f_mapping; |
2873 | } | |
2874 | EXPORT_SYMBOL_GPL(__page_file_mapping); | |
2875 | ||
2876 | pgoff_t __page_file_index(struct page *page) | |
2877 | { | |
2878 | swp_entry_t swap = { .val = page_private(page) }; | |
309381fe | 2879 | VM_BUG_ON_PAGE(!PageSwapCache(page), page); |
f981c595 MG |
2880 | return swp_offset(swap); |
2881 | } | |
2882 | EXPORT_SYMBOL_GPL(__page_file_index); | |
2883 | ||
570a335b HD |
2884 | /* |
2885 | * add_swap_count_continuation - called when a swap count is duplicated | |
2886 | * beyond SWAP_MAP_MAX, it allocates a new page and links that to the entry's | |
2887 | * page of the original vmalloc'ed swap_map, to hold the continuation count | |
2888 | * (for that entry and for its neighbouring PAGE_SIZE swap entries). Called | |
2889 | * again when count is duplicated beyond SWAP_MAP_MAX * SWAP_CONT_MAX, etc. | |
2890 | * | |
2891 | * These continuation pages are seldom referenced: the common paths all work | |
2892 | * on the original swap_map, only referring to a continuation page when the | |
2893 | * low "digit" of a count is incremented or decremented through SWAP_MAP_MAX. | |
2894 | * | |
2895 | * add_swap_count_continuation(, GFP_ATOMIC) can be called while holding | |
2896 | * page table locks; if it fails, add_swap_count_continuation(, GFP_KERNEL) | |
2897 | * can be called after dropping locks. | |
2898 | */ | |
2899 | int add_swap_count_continuation(swp_entry_t entry, gfp_t gfp_mask) | |
2900 | { | |
2901 | struct swap_info_struct *si; | |
235b6217 | 2902 | struct swap_cluster_info *ci; |
570a335b HD |
2903 | struct page *head; |
2904 | struct page *page; | |
2905 | struct page *list_page; | |
2906 | pgoff_t offset; | |
2907 | unsigned char count; | |
2908 | ||
2909 | /* | |
2910 | * When debugging, it's easier to use __GFP_ZERO here; but it's better | |
2911 | * for latency not to zero a page while GFP_ATOMIC and holding locks. | |
2912 | */ | |
2913 | page = alloc_page(gfp_mask | __GFP_HIGHMEM); | |
2914 | ||
2915 | si = swap_info_get(entry); | |
2916 | if (!si) { | |
2917 | /* | |
2918 | * An acceptable race has occurred since the failing | |
2919 | * __swap_duplicate(): the swap entry has been freed, | |
2920 | * perhaps even the whole swap_map cleared for swapoff. | |
2921 | */ | |
2922 | goto outer; | |
2923 | } | |
2924 | ||
2925 | offset = swp_offset(entry); | |
235b6217 HY |
2926 | |
2927 | ci = lock_cluster(si, offset); | |
2928 | ||
570a335b HD |
2929 | count = si->swap_map[offset] & ~SWAP_HAS_CACHE; |
2930 | ||
2931 | if ((count & ~COUNT_CONTINUED) != SWAP_MAP_MAX) { | |
2932 | /* | |
2933 | * The higher the swap count, the more likely it is that tasks | |
2934 | * will race to add swap count continuation: we need to avoid | |
2935 | * over-provisioning. | |
2936 | */ | |
2937 | goto out; | |
2938 | } | |
2939 | ||
2940 | if (!page) { | |
235b6217 | 2941 | unlock_cluster(ci); |
ec8acf20 | 2942 | spin_unlock(&si->lock); |
570a335b HD |
2943 | return -ENOMEM; |
2944 | } | |
2945 | ||
2946 | /* | |
2947 | * We are fortunate that although vmalloc_to_page uses pte_offset_map, | |
2de1a7e4 SJ |
2948 | * no architecture is using highmem pages for kernel page tables: so it |
2949 | * will not corrupt the GFP_ATOMIC caller's atomic page table kmaps. | |
570a335b HD |
2950 | */ |
2951 | head = vmalloc_to_page(si->swap_map + offset); | |
2952 | offset &= ~PAGE_MASK; | |
2953 | ||
2954 | /* | |
2955 | * Page allocation does not initialize the page's lru field, | |
2956 | * but it does always reset its private field. | |
2957 | */ | |
2958 | if (!page_private(head)) { | |
2959 | BUG_ON(count & COUNT_CONTINUED); | |
2960 | INIT_LIST_HEAD(&head->lru); | |
2961 | set_page_private(head, SWP_CONTINUED); | |
2962 | si->flags |= SWP_CONTINUED; | |
2963 | } | |
2964 | ||
2965 | list_for_each_entry(list_page, &head->lru, lru) { | |
2966 | unsigned char *map; | |
2967 | ||
2968 | /* | |
2969 | * If the previous map said no continuation, but we've found | |
2970 | * a continuation page, free our allocation and use this one. | |
2971 | */ | |
2972 | if (!(count & COUNT_CONTINUED)) | |
2973 | goto out; | |
2974 | ||
9b04c5fe | 2975 | map = kmap_atomic(list_page) + offset; |
570a335b | 2976 | count = *map; |
9b04c5fe | 2977 | kunmap_atomic(map); |
570a335b HD |
2978 | |
2979 | /* | |
2980 | * If this continuation count now has some space in it, | |
2981 | * free our allocation and use this one. | |
2982 | */ | |
2983 | if ((count & ~COUNT_CONTINUED) != SWAP_CONT_MAX) | |
2984 | goto out; | |
2985 | } | |
2986 | ||
2987 | list_add_tail(&page->lru, &head->lru); | |
2988 | page = NULL; /* now it's attached, don't free it */ | |
2989 | out: | |
235b6217 | 2990 | unlock_cluster(ci); |
ec8acf20 | 2991 | spin_unlock(&si->lock); |
570a335b HD |
2992 | outer: |
2993 | if (page) | |
2994 | __free_page(page); | |
2995 | return 0; | |
2996 | } | |
2997 | ||
2998 | /* | |
2999 | * swap_count_continued - when the original swap_map count is incremented | |
3000 | * from SWAP_MAP_MAX, check if there is already a continuation page to carry | |
3001 | * into, carry if so, or else fail until a new continuation page is allocated; | |
3002 | * when the original swap_map count is decremented from 0 with continuation, | |
3003 | * borrow from the continuation and report whether it still holds more. | |
235b6217 HY |
3004 | * Called while __swap_duplicate() or swap_entry_free() holds swap or cluster |
3005 | * lock. | |
570a335b HD |
3006 | */ |
3007 | static bool swap_count_continued(struct swap_info_struct *si, | |
3008 | pgoff_t offset, unsigned char count) | |
3009 | { | |
3010 | struct page *head; | |
3011 | struct page *page; | |
3012 | unsigned char *map; | |
3013 | ||
3014 | head = vmalloc_to_page(si->swap_map + offset); | |
3015 | if (page_private(head) != SWP_CONTINUED) { | |
3016 | BUG_ON(count & COUNT_CONTINUED); | |
3017 | return false; /* need to add count continuation */ | |
3018 | } | |
3019 | ||
3020 | offset &= ~PAGE_MASK; | |
3021 | page = list_entry(head->lru.next, struct page, lru); | |
9b04c5fe | 3022 | map = kmap_atomic(page) + offset; |
570a335b HD |
3023 | |
3024 | if (count == SWAP_MAP_MAX) /* initial increment from swap_map */ | |
3025 | goto init_map; /* jump over SWAP_CONT_MAX checks */ | |
3026 | ||
3027 | if (count == (SWAP_MAP_MAX | COUNT_CONTINUED)) { /* incrementing */ | |
3028 | /* | |
3029 | * Think of how you add 1 to 999 | |
3030 | */ | |
3031 | while (*map == (SWAP_CONT_MAX | COUNT_CONTINUED)) { | |
9b04c5fe | 3032 | kunmap_atomic(map); |
570a335b HD |
3033 | page = list_entry(page->lru.next, struct page, lru); |
3034 | BUG_ON(page == head); | |
9b04c5fe | 3035 | map = kmap_atomic(page) + offset; |
570a335b HD |
3036 | } |
3037 | if (*map == SWAP_CONT_MAX) { | |
9b04c5fe | 3038 | kunmap_atomic(map); |
570a335b HD |
3039 | page = list_entry(page->lru.next, struct page, lru); |
3040 | if (page == head) | |
3041 | return false; /* add count continuation */ | |
9b04c5fe | 3042 | map = kmap_atomic(page) + offset; |
570a335b HD |
3043 | init_map: *map = 0; /* we didn't zero the page */ |
3044 | } | |
3045 | *map += 1; | |
9b04c5fe | 3046 | kunmap_atomic(map); |
570a335b HD |
3047 | page = list_entry(page->lru.prev, struct page, lru); |
3048 | while (page != head) { | |
9b04c5fe | 3049 | map = kmap_atomic(page) + offset; |
570a335b | 3050 | *map = COUNT_CONTINUED; |
9b04c5fe | 3051 | kunmap_atomic(map); |
570a335b HD |
3052 | page = list_entry(page->lru.prev, struct page, lru); |
3053 | } | |
3054 | return true; /* incremented */ | |
3055 | ||
3056 | } else { /* decrementing */ | |
3057 | /* | |
3058 | * Think of how you subtract 1 from 1000 | |
3059 | */ | |
3060 | BUG_ON(count != COUNT_CONTINUED); | |
3061 | while (*map == COUNT_CONTINUED) { | |
9b04c5fe | 3062 | kunmap_atomic(map); |
570a335b HD |
3063 | page = list_entry(page->lru.next, struct page, lru); |
3064 | BUG_ON(page == head); | |
9b04c5fe | 3065 | map = kmap_atomic(page) + offset; |
570a335b HD |
3066 | } |
3067 | BUG_ON(*map == 0); | |
3068 | *map -= 1; | |
3069 | if (*map == 0) | |
3070 | count = 0; | |
9b04c5fe | 3071 | kunmap_atomic(map); |
570a335b HD |
3072 | page = list_entry(page->lru.prev, struct page, lru); |
3073 | while (page != head) { | |
9b04c5fe | 3074 | map = kmap_atomic(page) + offset; |
570a335b HD |
3075 | *map = SWAP_CONT_MAX | count; |
3076 | count = COUNT_CONTINUED; | |
9b04c5fe | 3077 | kunmap_atomic(map); |
570a335b HD |
3078 | page = list_entry(page->lru.prev, struct page, lru); |
3079 | } | |
3080 | return count == COUNT_CONTINUED; | |
3081 | } | |
3082 | } | |
3083 | ||
3084 | /* | |
3085 | * free_swap_count_continuations - swapoff free all the continuation pages | |
3086 | * appended to the swap_map, after swap_map is quiesced, before vfree'ing it. | |
3087 | */ | |
3088 | static void free_swap_count_continuations(struct swap_info_struct *si) | |
3089 | { | |
3090 | pgoff_t offset; | |
3091 | ||
3092 | for (offset = 0; offset < si->max; offset += PAGE_SIZE) { | |
3093 | struct page *head; | |
3094 | head = vmalloc_to_page(si->swap_map + offset); | |
3095 | if (page_private(head)) { | |
0d576d20 GT |
3096 | struct page *page, *next; |
3097 | ||
3098 | list_for_each_entry_safe(page, next, &head->lru, lru) { | |
3099 | list_del(&page->lru); | |
570a335b HD |
3100 | __free_page(page); |
3101 | } | |
3102 | } | |
3103 | } | |
3104 | } |