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