Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/mm/swapfile.c | |
3 | * | |
4 | * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds | |
5 | * Swap reorganised 29.12.95, Stephen Tweedie | |
6 | */ | |
7 | ||
1da177e4 LT |
8 | #include <linux/mm.h> |
9 | #include <linux/hugetlb.h> | |
10 | #include <linux/mman.h> | |
11 | #include <linux/slab.h> | |
12 | #include <linux/kernel_stat.h> | |
13 | #include <linux/swap.h> | |
14 | #include <linux/vmalloc.h> | |
15 | #include <linux/pagemap.h> | |
16 | #include <linux/namei.h> | |
072441e2 | 17 | #include <linux/shmem_fs.h> |
1da177e4 | 18 | #include <linux/blkdev.h> |
20137a49 | 19 | #include <linux/random.h> |
1da177e4 LT |
20 | #include <linux/writeback.h> |
21 | #include <linux/proc_fs.h> | |
22 | #include <linux/seq_file.h> | |
23 | #include <linux/init.h> | |
5ad64688 | 24 | #include <linux/ksm.h> |
1da177e4 LT |
25 | #include <linux/rmap.h> |
26 | #include <linux/security.h> | |
27 | #include <linux/backing-dev.h> | |
fc0abb14 | 28 | #include <linux/mutex.h> |
c59ede7b | 29 | #include <linux/capability.h> |
1da177e4 | 30 | #include <linux/syscalls.h> |
8a9f3ccd | 31 | #include <linux/memcontrol.h> |
66d7dd51 | 32 | #include <linux/poll.h> |
72788c38 | 33 | #include <linux/oom.h> |
38b5faf4 DM |
34 | #include <linux/frontswap.h> |
35 | #include <linux/swapfile.h> | |
f981c595 | 36 | #include <linux/export.h> |
1da177e4 LT |
37 | |
38 | #include <asm/pgtable.h> | |
39 | #include <asm/tlbflush.h> | |
40 | #include <linux/swapops.h> | |
5d1ea48b | 41 | #include <linux/swap_cgroup.h> |
1da177e4 | 42 | |
570a335b HD |
43 | static bool swap_count_continued(struct swap_info_struct *, pgoff_t, |
44 | unsigned char); | |
45 | static void free_swap_count_continuations(struct swap_info_struct *); | |
d4906e1a | 46 | static sector_t map_swap_entry(swp_entry_t, struct block_device**); |
570a335b | 47 | |
38b5faf4 | 48 | DEFINE_SPINLOCK(swap_lock); |
7c363b8c | 49 | static unsigned int nr_swapfiles; |
ec8acf20 | 50 | atomic_long_t nr_swap_pages; |
fb0fec50 CW |
51 | /* |
52 | * Some modules use swappable objects and may try to swap them out under | |
53 | * memory pressure (via the shrinker). Before doing so, they may wish to | |
54 | * check to see if any swap space is available. | |
55 | */ | |
56 | EXPORT_SYMBOL_GPL(nr_swap_pages); | |
ec8acf20 | 57 | /* protected with swap_lock. reading in vm_swap_full() doesn't need lock */ |
1da177e4 | 58 | long total_swap_pages; |
78ecba08 | 59 | static int least_priority; |
1da177e4 | 60 | |
1da177e4 LT |
61 | static const char Bad_file[] = "Bad swap file entry "; |
62 | static const char Unused_file[] = "Unused swap file entry "; | |
63 | static const char Bad_offset[] = "Bad swap offset entry "; | |
64 | static const char Unused_offset[] = "Unused swap offset entry "; | |
65 | ||
adfab836 DS |
66 | /* |
67 | * all active swap_info_structs | |
68 | * protected with swap_lock, and ordered by priority. | |
69 | */ | |
18ab4d4c DS |
70 | PLIST_HEAD(swap_active_head); |
71 | ||
72 | /* | |
73 | * all available (active, not full) swap_info_structs | |
74 | * protected with swap_avail_lock, ordered by priority. | |
75 | * This is used by get_swap_page() instead of swap_active_head | |
76 | * because swap_active_head includes all swap_info_structs, | |
77 | * but get_swap_page() doesn't need to look at full ones. | |
78 | * This uses its own lock instead of swap_lock because when a | |
79 | * swap_info_struct changes between not-full/full, it needs to | |
80 | * add/remove itself to/from this list, but the swap_info_struct->lock | |
81 | * is held and the locking order requires swap_lock to be taken | |
82 | * before any swap_info_struct->lock. | |
83 | */ | |
84 | static PLIST_HEAD(swap_avail_head); | |
85 | static DEFINE_SPINLOCK(swap_avail_lock); | |
1da177e4 | 86 | |
38b5faf4 | 87 | struct swap_info_struct *swap_info[MAX_SWAPFILES]; |
1da177e4 | 88 | |
fc0abb14 | 89 | static DEFINE_MUTEX(swapon_mutex); |
1da177e4 | 90 | |
66d7dd51 KS |
91 | static DECLARE_WAIT_QUEUE_HEAD(proc_poll_wait); |
92 | /* Activity counter to indicate that a swapon or swapoff has occurred */ | |
93 | static atomic_t proc_poll_event = ATOMIC_INIT(0); | |
94 | ||
8d69aaee | 95 | static inline unsigned char swap_count(unsigned char ent) |
355cfa73 | 96 | { |
570a335b | 97 | return ent & ~SWAP_HAS_CACHE; /* may include SWAP_HAS_CONT flag */ |
355cfa73 KH |
98 | } |
99 | ||
efa90a98 | 100 | /* returns 1 if swap entry is freed */ |
c9e44410 KH |
101 | static int |
102 | __try_to_reclaim_swap(struct swap_info_struct *si, unsigned long offset) | |
103 | { | |
efa90a98 | 104 | swp_entry_t entry = swp_entry(si->type, offset); |
c9e44410 KH |
105 | struct page *page; |
106 | int ret = 0; | |
107 | ||
33806f06 | 108 | page = find_get_page(swap_address_space(entry), entry.val); |
c9e44410 KH |
109 | if (!page) |
110 | return 0; | |
111 | /* | |
112 | * This function is called from scan_swap_map() and it's called | |
113 | * by vmscan.c at reclaiming pages. So, we hold a lock on a page, here. | |
114 | * We have to use trylock for avoiding deadlock. This is a special | |
115 | * case and you should use try_to_free_swap() with explicit lock_page() | |
116 | * in usual operations. | |
117 | */ | |
118 | if (trylock_page(page)) { | |
119 | ret = try_to_free_swap(page); | |
120 | unlock_page(page); | |
121 | } | |
09cbfeaf | 122 | put_page(page); |
c9e44410 KH |
123 | return ret; |
124 | } | |
355cfa73 | 125 | |
6a6ba831 HD |
126 | /* |
127 | * swapon tell device that all the old swap contents can be discarded, | |
128 | * to allow the swap device to optimize its wear-levelling. | |
129 | */ | |
130 | static int discard_swap(struct swap_info_struct *si) | |
131 | { | |
132 | struct swap_extent *se; | |
9625a5f2 HD |
133 | sector_t start_block; |
134 | sector_t nr_blocks; | |
6a6ba831 HD |
135 | int err = 0; |
136 | ||
9625a5f2 HD |
137 | /* Do not discard the swap header page! */ |
138 | se = &si->first_swap_extent; | |
139 | start_block = (se->start_block + 1) << (PAGE_SHIFT - 9); | |
140 | nr_blocks = ((sector_t)se->nr_pages - 1) << (PAGE_SHIFT - 9); | |
141 | if (nr_blocks) { | |
142 | err = blkdev_issue_discard(si->bdev, start_block, | |
dd3932ed | 143 | nr_blocks, GFP_KERNEL, 0); |
9625a5f2 HD |
144 | if (err) |
145 | return err; | |
146 | cond_resched(); | |
147 | } | |
6a6ba831 | 148 | |
9625a5f2 HD |
149 | list_for_each_entry(se, &si->first_swap_extent.list, list) { |
150 | start_block = se->start_block << (PAGE_SHIFT - 9); | |
151 | nr_blocks = (sector_t)se->nr_pages << (PAGE_SHIFT - 9); | |
6a6ba831 HD |
152 | |
153 | err = blkdev_issue_discard(si->bdev, start_block, | |
dd3932ed | 154 | nr_blocks, GFP_KERNEL, 0); |
6a6ba831 HD |
155 | if (err) |
156 | break; | |
157 | ||
158 | cond_resched(); | |
159 | } | |
160 | return err; /* That will often be -EOPNOTSUPP */ | |
161 | } | |
162 | ||
7992fde7 HD |
163 | /* |
164 | * swap allocation tell device that a cluster of swap can now be discarded, | |
165 | * to allow the swap device to optimize its wear-levelling. | |
166 | */ | |
167 | static void discard_swap_cluster(struct swap_info_struct *si, | |
168 | pgoff_t start_page, pgoff_t nr_pages) | |
169 | { | |
170 | struct swap_extent *se = si->curr_swap_extent; | |
171 | int found_extent = 0; | |
172 | ||
173 | while (nr_pages) { | |
7992fde7 HD |
174 | if (se->start_page <= start_page && |
175 | start_page < se->start_page + se->nr_pages) { | |
176 | pgoff_t offset = start_page - se->start_page; | |
177 | sector_t start_block = se->start_block + offset; | |
858a2990 | 178 | sector_t nr_blocks = se->nr_pages - offset; |
7992fde7 HD |
179 | |
180 | if (nr_blocks > nr_pages) | |
181 | nr_blocks = nr_pages; | |
182 | start_page += nr_blocks; | |
183 | nr_pages -= nr_blocks; | |
184 | ||
185 | if (!found_extent++) | |
186 | si->curr_swap_extent = se; | |
187 | ||
188 | start_block <<= PAGE_SHIFT - 9; | |
189 | nr_blocks <<= PAGE_SHIFT - 9; | |
190 | if (blkdev_issue_discard(si->bdev, start_block, | |
dd3932ed | 191 | nr_blocks, GFP_NOIO, 0)) |
7992fde7 HD |
192 | break; |
193 | } | |
194 | ||
a8ae4991 | 195 | se = list_next_entry(se, list); |
7992fde7 HD |
196 | } |
197 | } | |
198 | ||
048c27fd HD |
199 | #define SWAPFILE_CLUSTER 256 |
200 | #define LATENCY_LIMIT 256 | |
201 | ||
2a8f9449 SL |
202 | static inline void cluster_set_flag(struct swap_cluster_info *info, |
203 | unsigned int flag) | |
204 | { | |
205 | info->flags = flag; | |
206 | } | |
207 | ||
208 | static inline unsigned int cluster_count(struct swap_cluster_info *info) | |
209 | { | |
210 | return info->data; | |
211 | } | |
212 | ||
213 | static inline void cluster_set_count(struct swap_cluster_info *info, | |
214 | unsigned int c) | |
215 | { | |
216 | info->data = c; | |
217 | } | |
218 | ||
219 | static inline void cluster_set_count_flag(struct swap_cluster_info *info, | |
220 | unsigned int c, unsigned int f) | |
221 | { | |
222 | info->flags = f; | |
223 | info->data = c; | |
224 | } | |
225 | ||
226 | static inline unsigned int cluster_next(struct swap_cluster_info *info) | |
227 | { | |
228 | return info->data; | |
229 | } | |
230 | ||
231 | static inline void cluster_set_next(struct swap_cluster_info *info, | |
232 | unsigned int n) | |
233 | { | |
234 | info->data = n; | |
235 | } | |
236 | ||
237 | static inline void cluster_set_next_flag(struct swap_cluster_info *info, | |
238 | unsigned int n, unsigned int f) | |
239 | { | |
240 | info->flags = f; | |
241 | info->data = n; | |
242 | } | |
243 | ||
244 | static inline bool cluster_is_free(struct swap_cluster_info *info) | |
245 | { | |
246 | return info->flags & CLUSTER_FLAG_FREE; | |
247 | } | |
248 | ||
249 | static inline bool cluster_is_null(struct swap_cluster_info *info) | |
250 | { | |
251 | return info->flags & CLUSTER_FLAG_NEXT_NULL; | |
252 | } | |
253 | ||
254 | static inline void cluster_set_null(struct swap_cluster_info *info) | |
255 | { | |
256 | info->flags = CLUSTER_FLAG_NEXT_NULL; | |
257 | info->data = 0; | |
258 | } | |
259 | ||
815c2c54 SL |
260 | /* Add a cluster to discard list and schedule it to do discard */ |
261 | static void swap_cluster_schedule_discard(struct swap_info_struct *si, | |
262 | unsigned int idx) | |
263 | { | |
264 | /* | |
265 | * If scan_swap_map() can't find a free cluster, it will check | |
266 | * si->swap_map directly. To make sure the discarding cluster isn't | |
267 | * taken by scan_swap_map(), mark the swap entries bad (occupied). It | |
268 | * will be cleared after discard | |
269 | */ | |
270 | memset(si->swap_map + idx * SWAPFILE_CLUSTER, | |
271 | SWAP_MAP_BAD, SWAPFILE_CLUSTER); | |
272 | ||
273 | if (cluster_is_null(&si->discard_cluster_head)) { | |
274 | cluster_set_next_flag(&si->discard_cluster_head, | |
275 | idx, 0); | |
276 | cluster_set_next_flag(&si->discard_cluster_tail, | |
277 | idx, 0); | |
278 | } else { | |
279 | unsigned int tail = cluster_next(&si->discard_cluster_tail); | |
280 | cluster_set_next(&si->cluster_info[tail], idx); | |
281 | cluster_set_next_flag(&si->discard_cluster_tail, | |
282 | idx, 0); | |
283 | } | |
284 | ||
285 | schedule_work(&si->discard_work); | |
286 | } | |
287 | ||
288 | /* | |
289 | * Doing discard actually. After a cluster discard is finished, the cluster | |
290 | * will be added to free cluster list. caller should hold si->lock. | |
291 | */ | |
292 | static void swap_do_scheduled_discard(struct swap_info_struct *si) | |
293 | { | |
294 | struct swap_cluster_info *info; | |
295 | unsigned int idx; | |
296 | ||
297 | info = si->cluster_info; | |
298 | ||
299 | while (!cluster_is_null(&si->discard_cluster_head)) { | |
300 | idx = cluster_next(&si->discard_cluster_head); | |
301 | ||
302 | cluster_set_next_flag(&si->discard_cluster_head, | |
303 | cluster_next(&info[idx]), 0); | |
304 | if (cluster_next(&si->discard_cluster_tail) == idx) { | |
305 | cluster_set_null(&si->discard_cluster_head); | |
306 | cluster_set_null(&si->discard_cluster_tail); | |
307 | } | |
308 | spin_unlock(&si->lock); | |
309 | ||
310 | discard_swap_cluster(si, idx * SWAPFILE_CLUSTER, | |
311 | SWAPFILE_CLUSTER); | |
312 | ||
313 | spin_lock(&si->lock); | |
314 | cluster_set_flag(&info[idx], CLUSTER_FLAG_FREE); | |
315 | if (cluster_is_null(&si->free_cluster_head)) { | |
316 | cluster_set_next_flag(&si->free_cluster_head, | |
317 | idx, 0); | |
318 | cluster_set_next_flag(&si->free_cluster_tail, | |
319 | idx, 0); | |
320 | } else { | |
321 | unsigned int tail; | |
322 | ||
323 | tail = cluster_next(&si->free_cluster_tail); | |
324 | cluster_set_next(&info[tail], idx); | |
325 | cluster_set_next_flag(&si->free_cluster_tail, | |
326 | idx, 0); | |
327 | } | |
328 | memset(si->swap_map + idx * SWAPFILE_CLUSTER, | |
329 | 0, SWAPFILE_CLUSTER); | |
330 | } | |
331 | } | |
332 | ||
333 | static void swap_discard_work(struct work_struct *work) | |
334 | { | |
335 | struct swap_info_struct *si; | |
336 | ||
337 | si = container_of(work, struct swap_info_struct, discard_work); | |
338 | ||
339 | spin_lock(&si->lock); | |
340 | swap_do_scheduled_discard(si); | |
341 | spin_unlock(&si->lock); | |
342 | } | |
343 | ||
2a8f9449 SL |
344 | /* |
345 | * The cluster corresponding to page_nr will be used. The cluster will be | |
346 | * removed from free cluster list and its usage counter will be increased. | |
347 | */ | |
348 | static void inc_cluster_info_page(struct swap_info_struct *p, | |
349 | struct swap_cluster_info *cluster_info, unsigned long page_nr) | |
350 | { | |
351 | unsigned long idx = page_nr / SWAPFILE_CLUSTER; | |
352 | ||
353 | if (!cluster_info) | |
354 | return; | |
355 | if (cluster_is_free(&cluster_info[idx])) { | |
356 | VM_BUG_ON(cluster_next(&p->free_cluster_head) != idx); | |
357 | cluster_set_next_flag(&p->free_cluster_head, | |
358 | cluster_next(&cluster_info[idx]), 0); | |
359 | if (cluster_next(&p->free_cluster_tail) == idx) { | |
360 | cluster_set_null(&p->free_cluster_tail); | |
361 | cluster_set_null(&p->free_cluster_head); | |
362 | } | |
363 | cluster_set_count_flag(&cluster_info[idx], 0, 0); | |
364 | } | |
365 | ||
366 | VM_BUG_ON(cluster_count(&cluster_info[idx]) >= SWAPFILE_CLUSTER); | |
367 | cluster_set_count(&cluster_info[idx], | |
368 | cluster_count(&cluster_info[idx]) + 1); | |
369 | } | |
370 | ||
371 | /* | |
372 | * The cluster corresponding to page_nr decreases one usage. If the usage | |
373 | * counter becomes 0, which means no page in the cluster is in using, we can | |
374 | * optionally discard the cluster and add it to free cluster list. | |
375 | */ | |
376 | static void dec_cluster_info_page(struct swap_info_struct *p, | |
377 | struct swap_cluster_info *cluster_info, unsigned long page_nr) | |
378 | { | |
379 | unsigned long idx = page_nr / SWAPFILE_CLUSTER; | |
380 | ||
381 | if (!cluster_info) | |
382 | return; | |
383 | ||
384 | VM_BUG_ON(cluster_count(&cluster_info[idx]) == 0); | |
385 | cluster_set_count(&cluster_info[idx], | |
386 | cluster_count(&cluster_info[idx]) - 1); | |
387 | ||
388 | if (cluster_count(&cluster_info[idx]) == 0) { | |
815c2c54 SL |
389 | /* |
390 | * If the swap is discardable, prepare discard the cluster | |
391 | * instead of free it immediately. The cluster will be freed | |
392 | * after discard. | |
393 | */ | |
edfe23da SL |
394 | if ((p->flags & (SWP_WRITEOK | SWP_PAGE_DISCARD)) == |
395 | (SWP_WRITEOK | SWP_PAGE_DISCARD)) { | |
815c2c54 SL |
396 | swap_cluster_schedule_discard(p, idx); |
397 | return; | |
398 | } | |
399 | ||
2a8f9449 SL |
400 | cluster_set_flag(&cluster_info[idx], CLUSTER_FLAG_FREE); |
401 | if (cluster_is_null(&p->free_cluster_head)) { | |
402 | cluster_set_next_flag(&p->free_cluster_head, idx, 0); | |
403 | cluster_set_next_flag(&p->free_cluster_tail, idx, 0); | |
404 | } else { | |
405 | unsigned int tail = cluster_next(&p->free_cluster_tail); | |
406 | cluster_set_next(&cluster_info[tail], idx); | |
407 | cluster_set_next_flag(&p->free_cluster_tail, idx, 0); | |
408 | } | |
409 | } | |
410 | } | |
411 | ||
412 | /* | |
413 | * It's possible scan_swap_map() uses a free cluster in the middle of free | |
414 | * cluster list. Avoiding such abuse to avoid list corruption. | |
415 | */ | |
ebc2a1a6 SL |
416 | static bool |
417 | scan_swap_map_ssd_cluster_conflict(struct swap_info_struct *si, | |
2a8f9449 SL |
418 | unsigned long offset) |
419 | { | |
ebc2a1a6 SL |
420 | struct percpu_cluster *percpu_cluster; |
421 | bool conflict; | |
422 | ||
2a8f9449 | 423 | offset /= SWAPFILE_CLUSTER; |
ebc2a1a6 | 424 | conflict = !cluster_is_null(&si->free_cluster_head) && |
2a8f9449 SL |
425 | offset != cluster_next(&si->free_cluster_head) && |
426 | cluster_is_free(&si->cluster_info[offset]); | |
ebc2a1a6 SL |
427 | |
428 | if (!conflict) | |
429 | return false; | |
430 | ||
431 | percpu_cluster = this_cpu_ptr(si->percpu_cluster); | |
432 | cluster_set_null(&percpu_cluster->index); | |
433 | return true; | |
434 | } | |
435 | ||
436 | /* | |
437 | * Try to get a swap entry from current cpu's swap entry pool (a cluster). This | |
438 | * might involve allocating a new cluster for current CPU too. | |
439 | */ | |
440 | static void scan_swap_map_try_ssd_cluster(struct swap_info_struct *si, | |
441 | unsigned long *offset, unsigned long *scan_base) | |
442 | { | |
443 | struct percpu_cluster *cluster; | |
444 | bool found_free; | |
445 | unsigned long tmp; | |
446 | ||
447 | new_cluster: | |
448 | cluster = this_cpu_ptr(si->percpu_cluster); | |
449 | if (cluster_is_null(&cluster->index)) { | |
450 | if (!cluster_is_null(&si->free_cluster_head)) { | |
451 | cluster->index = si->free_cluster_head; | |
452 | cluster->next = cluster_next(&cluster->index) * | |
453 | SWAPFILE_CLUSTER; | |
454 | } else if (!cluster_is_null(&si->discard_cluster_head)) { | |
455 | /* | |
456 | * we don't have free cluster but have some clusters in | |
457 | * discarding, do discard now and reclaim them | |
458 | */ | |
459 | swap_do_scheduled_discard(si); | |
460 | *scan_base = *offset = si->cluster_next; | |
461 | goto new_cluster; | |
462 | } else | |
463 | return; | |
464 | } | |
465 | ||
466 | found_free = false; | |
467 | ||
468 | /* | |
469 | * Other CPUs can use our cluster if they can't find a free cluster, | |
470 | * check if there is still free entry in the cluster | |
471 | */ | |
472 | tmp = cluster->next; | |
473 | while (tmp < si->max && tmp < (cluster_next(&cluster->index) + 1) * | |
474 | SWAPFILE_CLUSTER) { | |
475 | if (!si->swap_map[tmp]) { | |
476 | found_free = true; | |
477 | break; | |
478 | } | |
479 | tmp++; | |
480 | } | |
481 | if (!found_free) { | |
482 | cluster_set_null(&cluster->index); | |
483 | goto new_cluster; | |
484 | } | |
485 | cluster->next = tmp + 1; | |
486 | *offset = tmp; | |
487 | *scan_base = tmp; | |
2a8f9449 SL |
488 | } |
489 | ||
24b8ff7c CEB |
490 | static unsigned long scan_swap_map(struct swap_info_struct *si, |
491 | unsigned char usage) | |
1da177e4 | 492 | { |
ebebbbe9 | 493 | unsigned long offset; |
c60aa176 | 494 | unsigned long scan_base; |
7992fde7 | 495 | unsigned long last_in_cluster = 0; |
048c27fd | 496 | int latency_ration = LATENCY_LIMIT; |
7dfad418 | 497 | |
886bb7e9 | 498 | /* |
7dfad418 HD |
499 | * We try to cluster swap pages by allocating them sequentially |
500 | * in swap. Once we've allocated SWAPFILE_CLUSTER pages this | |
501 | * way, however, we resort to first-free allocation, starting | |
502 | * a new cluster. This prevents us from scattering swap pages | |
503 | * all over the entire swap partition, so that we reduce | |
504 | * overall disk seek times between swap pages. -- sct | |
505 | * But we do now try to find an empty cluster. -Andrea | |
c60aa176 | 506 | * And we let swap pages go all over an SSD partition. Hugh |
7dfad418 HD |
507 | */ |
508 | ||
52b7efdb | 509 | si->flags += SWP_SCANNING; |
c60aa176 | 510 | scan_base = offset = si->cluster_next; |
ebebbbe9 | 511 | |
ebc2a1a6 SL |
512 | /* SSD algorithm */ |
513 | if (si->cluster_info) { | |
514 | scan_swap_map_try_ssd_cluster(si, &offset, &scan_base); | |
515 | goto checks; | |
516 | } | |
517 | ||
ebebbbe9 HD |
518 | if (unlikely(!si->cluster_nr--)) { |
519 | if (si->pages - si->inuse_pages < SWAPFILE_CLUSTER) { | |
520 | si->cluster_nr = SWAPFILE_CLUSTER - 1; | |
521 | goto checks; | |
522 | } | |
2a8f9449 | 523 | |
ec8acf20 | 524 | spin_unlock(&si->lock); |
7dfad418 | 525 | |
c60aa176 HD |
526 | /* |
527 | * If seek is expensive, start searching for new cluster from | |
528 | * start of partition, to minimize the span of allocated swap. | |
50088c44 CY |
529 | * If seek is cheap, that is the SWP_SOLIDSTATE si->cluster_info |
530 | * case, just handled by scan_swap_map_try_ssd_cluster() above. | |
c60aa176 | 531 | */ |
50088c44 | 532 | scan_base = offset = si->lowest_bit; |
7dfad418 HD |
533 | last_in_cluster = offset + SWAPFILE_CLUSTER - 1; |
534 | ||
535 | /* Locate the first empty (unaligned) cluster */ | |
536 | for (; last_in_cluster <= si->highest_bit; offset++) { | |
1da177e4 | 537 | if (si->swap_map[offset]) |
7dfad418 HD |
538 | last_in_cluster = offset + SWAPFILE_CLUSTER; |
539 | else if (offset == last_in_cluster) { | |
ec8acf20 | 540 | spin_lock(&si->lock); |
ebebbbe9 HD |
541 | offset -= SWAPFILE_CLUSTER - 1; |
542 | si->cluster_next = offset; | |
543 | si->cluster_nr = SWAPFILE_CLUSTER - 1; | |
c60aa176 HD |
544 | goto checks; |
545 | } | |
546 | if (unlikely(--latency_ration < 0)) { | |
547 | cond_resched(); | |
548 | latency_ration = LATENCY_LIMIT; | |
549 | } | |
550 | } | |
551 | ||
552 | offset = scan_base; | |
ec8acf20 | 553 | spin_lock(&si->lock); |
ebebbbe9 | 554 | si->cluster_nr = SWAPFILE_CLUSTER - 1; |
1da177e4 | 555 | } |
7dfad418 | 556 | |
ebebbbe9 | 557 | checks: |
ebc2a1a6 SL |
558 | if (si->cluster_info) { |
559 | while (scan_swap_map_ssd_cluster_conflict(si, offset)) | |
560 | scan_swap_map_try_ssd_cluster(si, &offset, &scan_base); | |
561 | } | |
ebebbbe9 | 562 | if (!(si->flags & SWP_WRITEOK)) |
52b7efdb | 563 | goto no_page; |
7dfad418 HD |
564 | if (!si->highest_bit) |
565 | goto no_page; | |
ebebbbe9 | 566 | if (offset > si->highest_bit) |
c60aa176 | 567 | scan_base = offset = si->lowest_bit; |
c9e44410 | 568 | |
b73d7fce HD |
569 | /* reuse swap entry of cache-only swap if not busy. */ |
570 | if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) { | |
c9e44410 | 571 | int swap_was_freed; |
ec8acf20 | 572 | spin_unlock(&si->lock); |
c9e44410 | 573 | swap_was_freed = __try_to_reclaim_swap(si, offset); |
ec8acf20 | 574 | spin_lock(&si->lock); |
c9e44410 KH |
575 | /* entry was freed successfully, try to use this again */ |
576 | if (swap_was_freed) | |
577 | goto checks; | |
578 | goto scan; /* check next one */ | |
579 | } | |
580 | ||
ebebbbe9 HD |
581 | if (si->swap_map[offset]) |
582 | goto scan; | |
583 | ||
584 | if (offset == si->lowest_bit) | |
585 | si->lowest_bit++; | |
586 | if (offset == si->highest_bit) | |
587 | si->highest_bit--; | |
588 | si->inuse_pages++; | |
589 | if (si->inuse_pages == si->pages) { | |
590 | si->lowest_bit = si->max; | |
591 | si->highest_bit = 0; | |
18ab4d4c DS |
592 | spin_lock(&swap_avail_lock); |
593 | plist_del(&si->avail_list, &swap_avail_head); | |
594 | spin_unlock(&swap_avail_lock); | |
1da177e4 | 595 | } |
253d553b | 596 | si->swap_map[offset] = usage; |
2a8f9449 | 597 | inc_cluster_info_page(si, si->cluster_info, offset); |
ebebbbe9 HD |
598 | si->cluster_next = offset + 1; |
599 | si->flags -= SWP_SCANNING; | |
7992fde7 | 600 | |
ebebbbe9 | 601 | return offset; |
7dfad418 | 602 | |
ebebbbe9 | 603 | scan: |
ec8acf20 | 604 | spin_unlock(&si->lock); |
7dfad418 | 605 | while (++offset <= si->highest_bit) { |
52b7efdb | 606 | if (!si->swap_map[offset]) { |
ec8acf20 | 607 | spin_lock(&si->lock); |
52b7efdb HD |
608 | goto checks; |
609 | } | |
c9e44410 | 610 | if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) { |
ec8acf20 | 611 | spin_lock(&si->lock); |
c9e44410 KH |
612 | goto checks; |
613 | } | |
048c27fd HD |
614 | if (unlikely(--latency_ration < 0)) { |
615 | cond_resched(); | |
616 | latency_ration = LATENCY_LIMIT; | |
617 | } | |
7dfad418 | 618 | } |
c60aa176 | 619 | offset = si->lowest_bit; |
a5998061 | 620 | while (offset < scan_base) { |
c60aa176 | 621 | if (!si->swap_map[offset]) { |
ec8acf20 | 622 | spin_lock(&si->lock); |
c60aa176 HD |
623 | goto checks; |
624 | } | |
c9e44410 | 625 | if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) { |
ec8acf20 | 626 | spin_lock(&si->lock); |
c9e44410 KH |
627 | goto checks; |
628 | } | |
c60aa176 HD |
629 | if (unlikely(--latency_ration < 0)) { |
630 | cond_resched(); | |
631 | latency_ration = LATENCY_LIMIT; | |
632 | } | |
a5998061 | 633 | offset++; |
c60aa176 | 634 | } |
ec8acf20 | 635 | spin_lock(&si->lock); |
7dfad418 HD |
636 | |
637 | no_page: | |
52b7efdb | 638 | si->flags -= SWP_SCANNING; |
1da177e4 LT |
639 | return 0; |
640 | } | |
641 | ||
642 | swp_entry_t get_swap_page(void) | |
643 | { | |
adfab836 | 644 | struct swap_info_struct *si, *next; |
fb4f88dc | 645 | pgoff_t offset; |
1da177e4 | 646 | |
ec8acf20 | 647 | if (atomic_long_read(&nr_swap_pages) <= 0) |
fb4f88dc | 648 | goto noswap; |
ec8acf20 | 649 | atomic_long_dec(&nr_swap_pages); |
fb4f88dc | 650 | |
18ab4d4c DS |
651 | spin_lock(&swap_avail_lock); |
652 | ||
653 | start_over: | |
654 | plist_for_each_entry_safe(si, next, &swap_avail_head, avail_list) { | |
655 | /* requeue si to after same-priority siblings */ | |
656 | plist_requeue(&si->avail_list, &swap_avail_head); | |
657 | spin_unlock(&swap_avail_lock); | |
ec8acf20 | 658 | spin_lock(&si->lock); |
adfab836 | 659 | if (!si->highest_bit || !(si->flags & SWP_WRITEOK)) { |
18ab4d4c DS |
660 | spin_lock(&swap_avail_lock); |
661 | if (plist_node_empty(&si->avail_list)) { | |
662 | spin_unlock(&si->lock); | |
663 | goto nextsi; | |
664 | } | |
665 | WARN(!si->highest_bit, | |
666 | "swap_info %d in list but !highest_bit\n", | |
667 | si->type); | |
668 | WARN(!(si->flags & SWP_WRITEOK), | |
669 | "swap_info %d in list but !SWP_WRITEOK\n", | |
670 | si->type); | |
671 | plist_del(&si->avail_list, &swap_avail_head); | |
ec8acf20 | 672 | spin_unlock(&si->lock); |
18ab4d4c | 673 | goto nextsi; |
ec8acf20 | 674 | } |
fb4f88dc | 675 | |
355cfa73 | 676 | /* This is called for allocating swap entry for cache */ |
253d553b | 677 | offset = scan_swap_map(si, SWAP_HAS_CACHE); |
ec8acf20 SL |
678 | spin_unlock(&si->lock); |
679 | if (offset) | |
adfab836 | 680 | return swp_entry(si->type, offset); |
18ab4d4c DS |
681 | pr_debug("scan_swap_map of si %d failed to find offset\n", |
682 | si->type); | |
683 | spin_lock(&swap_avail_lock); | |
684 | nextsi: | |
adfab836 DS |
685 | /* |
686 | * if we got here, it's likely that si was almost full before, | |
687 | * and since scan_swap_map() can drop the si->lock, multiple | |
688 | * callers probably all tried to get a page from the same si | |
18ab4d4c DS |
689 | * and it filled up before we could get one; or, the si filled |
690 | * up between us dropping swap_avail_lock and taking si->lock. | |
691 | * Since we dropped the swap_avail_lock, the swap_avail_head | |
692 | * list may have been modified; so if next is still in the | |
693 | * swap_avail_head list then try it, otherwise start over. | |
adfab836 | 694 | */ |
18ab4d4c DS |
695 | if (plist_node_empty(&next->avail_list)) |
696 | goto start_over; | |
1da177e4 | 697 | } |
fb4f88dc | 698 | |
18ab4d4c DS |
699 | spin_unlock(&swap_avail_lock); |
700 | ||
ec8acf20 | 701 | atomic_long_inc(&nr_swap_pages); |
fb4f88dc | 702 | noswap: |
fb4f88dc | 703 | return (swp_entry_t) {0}; |
1da177e4 LT |
704 | } |
705 | ||
2de1a7e4 | 706 | /* The only caller of this function is now suspend routine */ |
910321ea HD |
707 | swp_entry_t get_swap_page_of_type(int type) |
708 | { | |
709 | struct swap_info_struct *si; | |
710 | pgoff_t offset; | |
711 | ||
910321ea | 712 | si = swap_info[type]; |
ec8acf20 | 713 | spin_lock(&si->lock); |
910321ea | 714 | if (si && (si->flags & SWP_WRITEOK)) { |
ec8acf20 | 715 | atomic_long_dec(&nr_swap_pages); |
910321ea HD |
716 | /* This is called for allocating swap entry, not cache */ |
717 | offset = scan_swap_map(si, 1); | |
718 | if (offset) { | |
ec8acf20 | 719 | spin_unlock(&si->lock); |
910321ea HD |
720 | return swp_entry(type, offset); |
721 | } | |
ec8acf20 | 722 | atomic_long_inc(&nr_swap_pages); |
910321ea | 723 | } |
ec8acf20 | 724 | spin_unlock(&si->lock); |
910321ea HD |
725 | return (swp_entry_t) {0}; |
726 | } | |
727 | ||
73c34b6a | 728 | static struct swap_info_struct *swap_info_get(swp_entry_t entry) |
1da177e4 | 729 | { |
73c34b6a | 730 | struct swap_info_struct *p; |
1da177e4 LT |
731 | unsigned long offset, type; |
732 | ||
733 | if (!entry.val) | |
734 | goto out; | |
735 | type = swp_type(entry); | |
736 | if (type >= nr_swapfiles) | |
737 | goto bad_nofile; | |
efa90a98 | 738 | p = swap_info[type]; |
1da177e4 LT |
739 | if (!(p->flags & SWP_USED)) |
740 | goto bad_device; | |
741 | offset = swp_offset(entry); | |
742 | if (offset >= p->max) | |
743 | goto bad_offset; | |
744 | if (!p->swap_map[offset]) | |
745 | goto bad_free; | |
ec8acf20 | 746 | spin_lock(&p->lock); |
1da177e4 LT |
747 | return p; |
748 | ||
749 | bad_free: | |
465c47fd | 750 | pr_err("swap_free: %s%08lx\n", Unused_offset, entry.val); |
1da177e4 LT |
751 | goto out; |
752 | bad_offset: | |
465c47fd | 753 | pr_err("swap_free: %s%08lx\n", Bad_offset, entry.val); |
1da177e4 LT |
754 | goto out; |
755 | bad_device: | |
465c47fd | 756 | pr_err("swap_free: %s%08lx\n", Unused_file, entry.val); |
1da177e4 LT |
757 | goto out; |
758 | bad_nofile: | |
465c47fd | 759 | pr_err("swap_free: %s%08lx\n", Bad_file, entry.val); |
1da177e4 LT |
760 | out: |
761 | return NULL; | |
886bb7e9 | 762 | } |
1da177e4 | 763 | |
8d69aaee HD |
764 | static unsigned char swap_entry_free(struct swap_info_struct *p, |
765 | swp_entry_t entry, unsigned char usage) | |
1da177e4 | 766 | { |
253d553b | 767 | unsigned long offset = swp_offset(entry); |
8d69aaee HD |
768 | unsigned char count; |
769 | unsigned char has_cache; | |
355cfa73 | 770 | |
253d553b HD |
771 | count = p->swap_map[offset]; |
772 | has_cache = count & SWAP_HAS_CACHE; | |
773 | count &= ~SWAP_HAS_CACHE; | |
355cfa73 | 774 | |
253d553b | 775 | if (usage == SWAP_HAS_CACHE) { |
355cfa73 | 776 | VM_BUG_ON(!has_cache); |
253d553b | 777 | has_cache = 0; |
aaa46865 HD |
778 | } else if (count == SWAP_MAP_SHMEM) { |
779 | /* | |
780 | * Or we could insist on shmem.c using a special | |
781 | * swap_shmem_free() and free_shmem_swap_and_cache()... | |
782 | */ | |
783 | count = 0; | |
570a335b HD |
784 | } else if ((count & ~COUNT_CONTINUED) <= SWAP_MAP_MAX) { |
785 | if (count == COUNT_CONTINUED) { | |
786 | if (swap_count_continued(p, offset, count)) | |
787 | count = SWAP_MAP_MAX | COUNT_CONTINUED; | |
788 | else | |
789 | count = SWAP_MAP_MAX; | |
790 | } else | |
791 | count--; | |
792 | } | |
253d553b | 793 | |
253d553b HD |
794 | usage = count | has_cache; |
795 | p->swap_map[offset] = usage; | |
355cfa73 | 796 | |
355cfa73 | 797 | /* free if no reference */ |
253d553b | 798 | if (!usage) { |
37e84351 | 799 | mem_cgroup_uncharge_swap(entry); |
2a8f9449 | 800 | dec_cluster_info_page(p, p->cluster_info, offset); |
355cfa73 KH |
801 | if (offset < p->lowest_bit) |
802 | p->lowest_bit = offset; | |
18ab4d4c DS |
803 | if (offset > p->highest_bit) { |
804 | bool was_full = !p->highest_bit; | |
355cfa73 | 805 | p->highest_bit = offset; |
18ab4d4c DS |
806 | if (was_full && (p->flags & SWP_WRITEOK)) { |
807 | spin_lock(&swap_avail_lock); | |
808 | WARN_ON(!plist_node_empty(&p->avail_list)); | |
809 | if (plist_node_empty(&p->avail_list)) | |
810 | plist_add(&p->avail_list, | |
811 | &swap_avail_head); | |
812 | spin_unlock(&swap_avail_lock); | |
813 | } | |
814 | } | |
ec8acf20 | 815 | atomic_long_inc(&nr_swap_pages); |
355cfa73 | 816 | p->inuse_pages--; |
38b5faf4 | 817 | frontswap_invalidate_page(p->type, offset); |
73744923 MG |
818 | if (p->flags & SWP_BLKDEV) { |
819 | struct gendisk *disk = p->bdev->bd_disk; | |
820 | if (disk->fops->swap_slot_free_notify) | |
821 | disk->fops->swap_slot_free_notify(p->bdev, | |
822 | offset); | |
823 | } | |
1da177e4 | 824 | } |
253d553b HD |
825 | |
826 | return usage; | |
1da177e4 LT |
827 | } |
828 | ||
829 | /* | |
2de1a7e4 | 830 | * Caller has made sure that the swap device corresponding to entry |
1da177e4 LT |
831 | * is still around or has not been recycled. |
832 | */ | |
833 | void swap_free(swp_entry_t entry) | |
834 | { | |
73c34b6a | 835 | struct swap_info_struct *p; |
1da177e4 LT |
836 | |
837 | p = swap_info_get(entry); | |
838 | if (p) { | |
253d553b | 839 | swap_entry_free(p, entry, 1); |
ec8acf20 | 840 | spin_unlock(&p->lock); |
1da177e4 LT |
841 | } |
842 | } | |
843 | ||
cb4b86ba KH |
844 | /* |
845 | * Called after dropping swapcache to decrease refcnt to swap entries. | |
846 | */ | |
0a31bc97 | 847 | void swapcache_free(swp_entry_t entry) |
cb4b86ba | 848 | { |
355cfa73 KH |
849 | struct swap_info_struct *p; |
850 | ||
355cfa73 KH |
851 | p = swap_info_get(entry); |
852 | if (p) { | |
0a31bc97 | 853 | swap_entry_free(p, entry, SWAP_HAS_CACHE); |
ec8acf20 | 854 | spin_unlock(&p->lock); |
355cfa73 | 855 | } |
cb4b86ba KH |
856 | } |
857 | ||
1da177e4 | 858 | /* |
c475a8ab | 859 | * How many references to page are currently swapped out? |
570a335b HD |
860 | * This does not give an exact answer when swap count is continued, |
861 | * but does include the high COUNT_CONTINUED flag to allow for that. | |
1da177e4 | 862 | */ |
bde05d1c | 863 | int page_swapcount(struct page *page) |
1da177e4 | 864 | { |
c475a8ab HD |
865 | int count = 0; |
866 | struct swap_info_struct *p; | |
1da177e4 LT |
867 | swp_entry_t entry; |
868 | ||
4c21e2f2 | 869 | entry.val = page_private(page); |
1da177e4 LT |
870 | p = swap_info_get(entry); |
871 | if (p) { | |
355cfa73 | 872 | count = swap_count(p->swap_map[swp_offset(entry)]); |
ec8acf20 | 873 | spin_unlock(&p->lock); |
1da177e4 | 874 | } |
c475a8ab | 875 | return count; |
1da177e4 LT |
876 | } |
877 | ||
8334b962 MK |
878 | /* |
879 | * How many references to @entry are currently swapped out? | |
880 | * This considers COUNT_CONTINUED so it returns exact answer. | |
881 | */ | |
882 | int swp_swapcount(swp_entry_t entry) | |
883 | { | |
884 | int count, tmp_count, n; | |
885 | struct swap_info_struct *p; | |
886 | struct page *page; | |
887 | pgoff_t offset; | |
888 | unsigned char *map; | |
889 | ||
890 | p = swap_info_get(entry); | |
891 | if (!p) | |
892 | return 0; | |
893 | ||
894 | count = swap_count(p->swap_map[swp_offset(entry)]); | |
895 | if (!(count & COUNT_CONTINUED)) | |
896 | goto out; | |
897 | ||
898 | count &= ~COUNT_CONTINUED; | |
899 | n = SWAP_MAP_MAX + 1; | |
900 | ||
901 | offset = swp_offset(entry); | |
902 | page = vmalloc_to_page(p->swap_map + offset); | |
903 | offset &= ~PAGE_MASK; | |
904 | VM_BUG_ON(page_private(page) != SWP_CONTINUED); | |
905 | ||
906 | do { | |
a8ae4991 | 907 | page = list_next_entry(page, lru); |
8334b962 MK |
908 | map = kmap_atomic(page); |
909 | tmp_count = map[offset]; | |
910 | kunmap_atomic(map); | |
911 | ||
912 | count += (tmp_count & ~COUNT_CONTINUED) * n; | |
913 | n *= (SWAP_CONT_MAX + 1); | |
914 | } while (tmp_count & COUNT_CONTINUED); | |
915 | out: | |
916 | spin_unlock(&p->lock); | |
917 | return count; | |
918 | } | |
919 | ||
1da177e4 | 920 | /* |
7b1fe597 HD |
921 | * We can write to an anon page without COW if there are no other references |
922 | * to it. And as a side-effect, free up its swap: because the old content | |
923 | * on disk will never be read, and seeking back there to write new content | |
924 | * later would only waste time away from clustering. | |
6d0a07ed AA |
925 | * |
926 | * NOTE: total_mapcount should not be relied upon by the caller if | |
927 | * reuse_swap_page() returns false, but it may be always overwritten | |
928 | * (see the other implementation for CONFIG_SWAP=n). | |
1da177e4 | 929 | */ |
6d0a07ed | 930 | bool reuse_swap_page(struct page *page, int *total_mapcount) |
1da177e4 | 931 | { |
c475a8ab HD |
932 | int count; |
933 | ||
309381fe | 934 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
5ad64688 | 935 | if (unlikely(PageKsm(page))) |
6d0a07ed AA |
936 | return false; |
937 | count = page_trans_huge_mapcount(page, total_mapcount); | |
7b1fe597 | 938 | if (count <= 1 && PageSwapCache(page)) { |
c475a8ab | 939 | count += page_swapcount(page); |
7b1fe597 HD |
940 | if (count == 1 && !PageWriteback(page)) { |
941 | delete_from_swap_cache(page); | |
942 | SetPageDirty(page); | |
943 | } | |
944 | } | |
5ad64688 | 945 | return count <= 1; |
1da177e4 LT |
946 | } |
947 | ||
948 | /* | |
a2c43eed HD |
949 | * If swap is getting full, or if there are no more mappings of this page, |
950 | * then try_to_free_swap is called to free its swap space. | |
1da177e4 | 951 | */ |
a2c43eed | 952 | int try_to_free_swap(struct page *page) |
1da177e4 | 953 | { |
309381fe | 954 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
1da177e4 LT |
955 | |
956 | if (!PageSwapCache(page)) | |
957 | return 0; | |
958 | if (PageWriteback(page)) | |
959 | return 0; | |
a2c43eed | 960 | if (page_swapcount(page)) |
1da177e4 LT |
961 | return 0; |
962 | ||
b73d7fce HD |
963 | /* |
964 | * Once hibernation has begun to create its image of memory, | |
965 | * there's a danger that one of the calls to try_to_free_swap() | |
966 | * - most probably a call from __try_to_reclaim_swap() while | |
967 | * hibernation is allocating its own swap pages for the image, | |
968 | * but conceivably even a call from memory reclaim - will free | |
969 | * the swap from a page which has already been recorded in the | |
970 | * image as a clean swapcache page, and then reuse its swap for | |
971 | * another page of the image. On waking from hibernation, the | |
972 | * original page might be freed under memory pressure, then | |
973 | * later read back in from swap, now with the wrong data. | |
974 | * | |
2de1a7e4 | 975 | * Hibernation suspends storage while it is writing the image |
f90ac398 | 976 | * to disk so check that here. |
b73d7fce | 977 | */ |
f90ac398 | 978 | if (pm_suspended_storage()) |
b73d7fce HD |
979 | return 0; |
980 | ||
a2c43eed HD |
981 | delete_from_swap_cache(page); |
982 | SetPageDirty(page); | |
983 | return 1; | |
68a22394 RR |
984 | } |
985 | ||
1da177e4 LT |
986 | /* |
987 | * Free the swap entry like above, but also try to | |
988 | * free the page cache entry if it is the last user. | |
989 | */ | |
2509ef26 | 990 | int free_swap_and_cache(swp_entry_t entry) |
1da177e4 | 991 | { |
2509ef26 | 992 | struct swap_info_struct *p; |
1da177e4 LT |
993 | struct page *page = NULL; |
994 | ||
a7420aa5 | 995 | if (non_swap_entry(entry)) |
2509ef26 | 996 | return 1; |
0697212a | 997 | |
1da177e4 LT |
998 | p = swap_info_get(entry); |
999 | if (p) { | |
253d553b | 1000 | if (swap_entry_free(p, entry, 1) == SWAP_HAS_CACHE) { |
33806f06 SL |
1001 | page = find_get_page(swap_address_space(entry), |
1002 | entry.val); | |
8413ac9d | 1003 | if (page && !trylock_page(page)) { |
09cbfeaf | 1004 | put_page(page); |
93fac704 NP |
1005 | page = NULL; |
1006 | } | |
1007 | } | |
ec8acf20 | 1008 | spin_unlock(&p->lock); |
1da177e4 LT |
1009 | } |
1010 | if (page) { | |
a2c43eed HD |
1011 | /* |
1012 | * Not mapped elsewhere, or swap space full? Free it! | |
1013 | * Also recheck PageSwapCache now page is locked (above). | |
1014 | */ | |
93fac704 | 1015 | if (PageSwapCache(page) && !PageWriteback(page) && |
5ccc5aba | 1016 | (!page_mapped(page) || mem_cgroup_swap_full(page))) { |
1da177e4 LT |
1017 | delete_from_swap_cache(page); |
1018 | SetPageDirty(page); | |
1019 | } | |
1020 | unlock_page(page); | |
09cbfeaf | 1021 | put_page(page); |
1da177e4 | 1022 | } |
2509ef26 | 1023 | return p != NULL; |
1da177e4 LT |
1024 | } |
1025 | ||
b0cb1a19 | 1026 | #ifdef CONFIG_HIBERNATION |
f577eb30 | 1027 | /* |
915bae9e | 1028 | * Find the swap type that corresponds to given device (if any). |
f577eb30 | 1029 | * |
915bae9e RW |
1030 | * @offset - number of the PAGE_SIZE-sized block of the device, starting |
1031 | * from 0, in which the swap header is expected to be located. | |
1032 | * | |
1033 | * This is needed for the suspend to disk (aka swsusp). | |
f577eb30 | 1034 | */ |
7bf23687 | 1035 | int swap_type_of(dev_t device, sector_t offset, struct block_device **bdev_p) |
f577eb30 | 1036 | { |
915bae9e | 1037 | struct block_device *bdev = NULL; |
efa90a98 | 1038 | int type; |
f577eb30 | 1039 | |
915bae9e RW |
1040 | if (device) |
1041 | bdev = bdget(device); | |
1042 | ||
f577eb30 | 1043 | spin_lock(&swap_lock); |
efa90a98 HD |
1044 | for (type = 0; type < nr_swapfiles; type++) { |
1045 | struct swap_info_struct *sis = swap_info[type]; | |
f577eb30 | 1046 | |
915bae9e | 1047 | if (!(sis->flags & SWP_WRITEOK)) |
f577eb30 | 1048 | continue; |
b6b5bce3 | 1049 | |
915bae9e | 1050 | if (!bdev) { |
7bf23687 | 1051 | if (bdev_p) |
dddac6a7 | 1052 | *bdev_p = bdgrab(sis->bdev); |
7bf23687 | 1053 | |
6e1819d6 | 1054 | spin_unlock(&swap_lock); |
efa90a98 | 1055 | return type; |
6e1819d6 | 1056 | } |
915bae9e | 1057 | if (bdev == sis->bdev) { |
9625a5f2 | 1058 | struct swap_extent *se = &sis->first_swap_extent; |
915bae9e | 1059 | |
915bae9e | 1060 | if (se->start_block == offset) { |
7bf23687 | 1061 | if (bdev_p) |
dddac6a7 | 1062 | *bdev_p = bdgrab(sis->bdev); |
7bf23687 | 1063 | |
915bae9e RW |
1064 | spin_unlock(&swap_lock); |
1065 | bdput(bdev); | |
efa90a98 | 1066 | return type; |
915bae9e | 1067 | } |
f577eb30 RW |
1068 | } |
1069 | } | |
1070 | spin_unlock(&swap_lock); | |
915bae9e RW |
1071 | if (bdev) |
1072 | bdput(bdev); | |
1073 | ||
f577eb30 RW |
1074 | return -ENODEV; |
1075 | } | |
1076 | ||
73c34b6a HD |
1077 | /* |
1078 | * Get the (PAGE_SIZE) block corresponding to given offset on the swapdev | |
1079 | * corresponding to given index in swap_info (swap type). | |
1080 | */ | |
1081 | sector_t swapdev_block(int type, pgoff_t offset) | |
1082 | { | |
1083 | struct block_device *bdev; | |
1084 | ||
1085 | if ((unsigned int)type >= nr_swapfiles) | |
1086 | return 0; | |
1087 | if (!(swap_info[type]->flags & SWP_WRITEOK)) | |
1088 | return 0; | |
d4906e1a | 1089 | return map_swap_entry(swp_entry(type, offset), &bdev); |
73c34b6a HD |
1090 | } |
1091 | ||
f577eb30 RW |
1092 | /* |
1093 | * Return either the total number of swap pages of given type, or the number | |
1094 | * of free pages of that type (depending on @free) | |
1095 | * | |
1096 | * This is needed for software suspend | |
1097 | */ | |
1098 | unsigned int count_swap_pages(int type, int free) | |
1099 | { | |
1100 | unsigned int n = 0; | |
1101 | ||
efa90a98 HD |
1102 | spin_lock(&swap_lock); |
1103 | if ((unsigned int)type < nr_swapfiles) { | |
1104 | struct swap_info_struct *sis = swap_info[type]; | |
1105 | ||
ec8acf20 | 1106 | spin_lock(&sis->lock); |
efa90a98 HD |
1107 | if (sis->flags & SWP_WRITEOK) { |
1108 | n = sis->pages; | |
f577eb30 | 1109 | if (free) |
efa90a98 | 1110 | n -= sis->inuse_pages; |
f577eb30 | 1111 | } |
ec8acf20 | 1112 | spin_unlock(&sis->lock); |
f577eb30 | 1113 | } |
efa90a98 | 1114 | spin_unlock(&swap_lock); |
f577eb30 RW |
1115 | return n; |
1116 | } | |
73c34b6a | 1117 | #endif /* CONFIG_HIBERNATION */ |
f577eb30 | 1118 | |
9f8bdb3f | 1119 | static inline int pte_same_as_swp(pte_t pte, pte_t swp_pte) |
179ef71c | 1120 | { |
9f8bdb3f | 1121 | return pte_same(pte_swp_clear_soft_dirty(pte), swp_pte); |
179ef71c CG |
1122 | } |
1123 | ||
1da177e4 | 1124 | /* |
72866f6f HD |
1125 | * No need to decide whether this PTE shares the swap entry with others, |
1126 | * just let do_wp_page work it out if a write is requested later - to | |
1127 | * force COW, vm_page_prot omits write permission from any private vma. | |
1da177e4 | 1128 | */ |
044d66c1 | 1129 | static int unuse_pte(struct vm_area_struct *vma, pmd_t *pmd, |
1da177e4 LT |
1130 | unsigned long addr, swp_entry_t entry, struct page *page) |
1131 | { | |
9e16b7fb | 1132 | struct page *swapcache; |
72835c86 | 1133 | struct mem_cgroup *memcg; |
044d66c1 HD |
1134 | spinlock_t *ptl; |
1135 | pte_t *pte; | |
1136 | int ret = 1; | |
1137 | ||
9e16b7fb HD |
1138 | swapcache = page; |
1139 | page = ksm_might_need_to_copy(page, vma, addr); | |
1140 | if (unlikely(!page)) | |
1141 | return -ENOMEM; | |
1142 | ||
f627c2f5 KS |
1143 | if (mem_cgroup_try_charge(page, vma->vm_mm, GFP_KERNEL, |
1144 | &memcg, false)) { | |
044d66c1 | 1145 | ret = -ENOMEM; |
85d9fc89 KH |
1146 | goto out_nolock; |
1147 | } | |
044d66c1 HD |
1148 | |
1149 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); | |
9f8bdb3f | 1150 | if (unlikely(!pte_same_as_swp(*pte, swp_entry_to_pte(entry)))) { |
f627c2f5 | 1151 | mem_cgroup_cancel_charge(page, memcg, false); |
044d66c1 HD |
1152 | ret = 0; |
1153 | goto out; | |
1154 | } | |
8a9f3ccd | 1155 | |
b084d435 | 1156 | dec_mm_counter(vma->vm_mm, MM_SWAPENTS); |
d559db08 | 1157 | inc_mm_counter(vma->vm_mm, MM_ANONPAGES); |
1da177e4 LT |
1158 | get_page(page); |
1159 | set_pte_at(vma->vm_mm, addr, pte, | |
1160 | pte_mkold(mk_pte(page, vma->vm_page_prot))); | |
00501b53 | 1161 | if (page == swapcache) { |
d281ee61 | 1162 | page_add_anon_rmap(page, vma, addr, false); |
f627c2f5 | 1163 | mem_cgroup_commit_charge(page, memcg, true, false); |
00501b53 | 1164 | } else { /* ksm created a completely new copy */ |
d281ee61 | 1165 | page_add_new_anon_rmap(page, vma, addr, false); |
f627c2f5 | 1166 | mem_cgroup_commit_charge(page, memcg, false, false); |
00501b53 JW |
1167 | lru_cache_add_active_or_unevictable(page, vma); |
1168 | } | |
1da177e4 LT |
1169 | swap_free(entry); |
1170 | /* | |
1171 | * Move the page to the active list so it is not | |
1172 | * immediately swapped out again after swapon. | |
1173 | */ | |
1174 | activate_page(page); | |
044d66c1 HD |
1175 | out: |
1176 | pte_unmap_unlock(pte, ptl); | |
85d9fc89 | 1177 | out_nolock: |
9e16b7fb HD |
1178 | if (page != swapcache) { |
1179 | unlock_page(page); | |
1180 | put_page(page); | |
1181 | } | |
044d66c1 | 1182 | return ret; |
1da177e4 LT |
1183 | } |
1184 | ||
1185 | static int unuse_pte_range(struct vm_area_struct *vma, pmd_t *pmd, | |
1186 | unsigned long addr, unsigned long end, | |
1187 | swp_entry_t entry, struct page *page) | |
1188 | { | |
1da177e4 | 1189 | pte_t swp_pte = swp_entry_to_pte(entry); |
705e87c0 | 1190 | pte_t *pte; |
8a9f3ccd | 1191 | int ret = 0; |
1da177e4 | 1192 | |
044d66c1 HD |
1193 | /* |
1194 | * We don't actually need pte lock while scanning for swp_pte: since | |
1195 | * we hold page lock and mmap_sem, swp_pte cannot be inserted into the | |
1196 | * page table while we're scanning; though it could get zapped, and on | |
1197 | * some architectures (e.g. x86_32 with PAE) we might catch a glimpse | |
1198 | * of unmatched parts which look like swp_pte, so unuse_pte must | |
1199 | * recheck under pte lock. Scanning without pte lock lets it be | |
2de1a7e4 | 1200 | * preemptable whenever CONFIG_PREEMPT but not CONFIG_HIGHPTE. |
044d66c1 HD |
1201 | */ |
1202 | pte = pte_offset_map(pmd, addr); | |
1da177e4 LT |
1203 | do { |
1204 | /* | |
1205 | * swapoff spends a _lot_ of time in this loop! | |
1206 | * Test inline before going to call unuse_pte. | |
1207 | */ | |
9f8bdb3f | 1208 | if (unlikely(pte_same_as_swp(*pte, swp_pte))) { |
044d66c1 HD |
1209 | pte_unmap(pte); |
1210 | ret = unuse_pte(vma, pmd, addr, entry, page); | |
1211 | if (ret) | |
1212 | goto out; | |
1213 | pte = pte_offset_map(pmd, addr); | |
1da177e4 LT |
1214 | } |
1215 | } while (pte++, addr += PAGE_SIZE, addr != end); | |
044d66c1 HD |
1216 | pte_unmap(pte - 1); |
1217 | out: | |
8a9f3ccd | 1218 | return ret; |
1da177e4 LT |
1219 | } |
1220 | ||
1221 | static inline int unuse_pmd_range(struct vm_area_struct *vma, pud_t *pud, | |
1222 | unsigned long addr, unsigned long end, | |
1223 | swp_entry_t entry, struct page *page) | |
1224 | { | |
1225 | pmd_t *pmd; | |
1226 | unsigned long next; | |
8a9f3ccd | 1227 | int ret; |
1da177e4 LT |
1228 | |
1229 | pmd = pmd_offset(pud, addr); | |
1230 | do { | |
1231 | next = pmd_addr_end(addr, end); | |
1a5a9906 | 1232 | if (pmd_none_or_trans_huge_or_clear_bad(pmd)) |
1da177e4 | 1233 | continue; |
8a9f3ccd BS |
1234 | ret = unuse_pte_range(vma, pmd, addr, next, entry, page); |
1235 | if (ret) | |
1236 | return ret; | |
1da177e4 LT |
1237 | } while (pmd++, addr = next, addr != end); |
1238 | return 0; | |
1239 | } | |
1240 | ||
1241 | static inline int unuse_pud_range(struct vm_area_struct *vma, pgd_t *pgd, | |
1242 | unsigned long addr, unsigned long end, | |
1243 | swp_entry_t entry, struct page *page) | |
1244 | { | |
1245 | pud_t *pud; | |
1246 | unsigned long next; | |
8a9f3ccd | 1247 | int ret; |
1da177e4 LT |
1248 | |
1249 | pud = pud_offset(pgd, addr); | |
1250 | do { | |
1251 | next = pud_addr_end(addr, end); | |
1252 | if (pud_none_or_clear_bad(pud)) | |
1253 | continue; | |
8a9f3ccd BS |
1254 | ret = unuse_pmd_range(vma, pud, addr, next, entry, page); |
1255 | if (ret) | |
1256 | return ret; | |
1da177e4 LT |
1257 | } while (pud++, addr = next, addr != end); |
1258 | return 0; | |
1259 | } | |
1260 | ||
1261 | static int unuse_vma(struct vm_area_struct *vma, | |
1262 | swp_entry_t entry, struct page *page) | |
1263 | { | |
1264 | pgd_t *pgd; | |
1265 | unsigned long addr, end, next; | |
8a9f3ccd | 1266 | int ret; |
1da177e4 | 1267 | |
3ca7b3c5 | 1268 | if (page_anon_vma(page)) { |
1da177e4 LT |
1269 | addr = page_address_in_vma(page, vma); |
1270 | if (addr == -EFAULT) | |
1271 | return 0; | |
1272 | else | |
1273 | end = addr + PAGE_SIZE; | |
1274 | } else { | |
1275 | addr = vma->vm_start; | |
1276 | end = vma->vm_end; | |
1277 | } | |
1278 | ||
1279 | pgd = pgd_offset(vma->vm_mm, addr); | |
1280 | do { | |
1281 | next = pgd_addr_end(addr, end); | |
1282 | if (pgd_none_or_clear_bad(pgd)) | |
1283 | continue; | |
8a9f3ccd BS |
1284 | ret = unuse_pud_range(vma, pgd, addr, next, entry, page); |
1285 | if (ret) | |
1286 | return ret; | |
1da177e4 LT |
1287 | } while (pgd++, addr = next, addr != end); |
1288 | return 0; | |
1289 | } | |
1290 | ||
1291 | static int unuse_mm(struct mm_struct *mm, | |
1292 | swp_entry_t entry, struct page *page) | |
1293 | { | |
1294 | struct vm_area_struct *vma; | |
8a9f3ccd | 1295 | int ret = 0; |
1da177e4 LT |
1296 | |
1297 | if (!down_read_trylock(&mm->mmap_sem)) { | |
1298 | /* | |
7d03431c FLVC |
1299 | * Activate page so shrink_inactive_list is unlikely to unmap |
1300 | * its ptes while lock is dropped, so swapoff can make progress. | |
1da177e4 | 1301 | */ |
c475a8ab | 1302 | activate_page(page); |
1da177e4 LT |
1303 | unlock_page(page); |
1304 | down_read(&mm->mmap_sem); | |
1305 | lock_page(page); | |
1306 | } | |
1da177e4 | 1307 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
8a9f3ccd | 1308 | if (vma->anon_vma && (ret = unuse_vma(vma, entry, page))) |
1da177e4 LT |
1309 | break; |
1310 | } | |
1da177e4 | 1311 | up_read(&mm->mmap_sem); |
8a9f3ccd | 1312 | return (ret < 0)? ret: 0; |
1da177e4 LT |
1313 | } |
1314 | ||
1315 | /* | |
38b5faf4 DM |
1316 | * Scan swap_map (or frontswap_map if frontswap parameter is true) |
1317 | * from current position to next entry still in use. | |
1da177e4 LT |
1318 | * Recycle to start on reaching the end, returning 0 when empty. |
1319 | */ | |
6eb396dc | 1320 | static unsigned int find_next_to_unuse(struct swap_info_struct *si, |
38b5faf4 | 1321 | unsigned int prev, bool frontswap) |
1da177e4 | 1322 | { |
6eb396dc HD |
1323 | unsigned int max = si->max; |
1324 | unsigned int i = prev; | |
8d69aaee | 1325 | unsigned char count; |
1da177e4 LT |
1326 | |
1327 | /* | |
5d337b91 | 1328 | * No need for swap_lock here: we're just looking |
1da177e4 LT |
1329 | * for whether an entry is in use, not modifying it; false |
1330 | * hits are okay, and sys_swapoff() has already prevented new | |
5d337b91 | 1331 | * allocations from this area (while holding swap_lock). |
1da177e4 LT |
1332 | */ |
1333 | for (;;) { | |
1334 | if (++i >= max) { | |
1335 | if (!prev) { | |
1336 | i = 0; | |
1337 | break; | |
1338 | } | |
1339 | /* | |
1340 | * No entries in use at top of swap_map, | |
1341 | * loop back to start and recheck there. | |
1342 | */ | |
1343 | max = prev + 1; | |
1344 | prev = 0; | |
1345 | i = 1; | |
1346 | } | |
38b5faf4 DM |
1347 | if (frontswap) { |
1348 | if (frontswap_test(si, i)) | |
1349 | break; | |
1350 | else | |
1351 | continue; | |
1352 | } | |
4db0c3c2 | 1353 | count = READ_ONCE(si->swap_map[i]); |
355cfa73 | 1354 | if (count && swap_count(count) != SWAP_MAP_BAD) |
1da177e4 LT |
1355 | break; |
1356 | } | |
1357 | return i; | |
1358 | } | |
1359 | ||
1360 | /* | |
1361 | * We completely avoid races by reading each swap page in advance, | |
1362 | * and then search for the process using it. All the necessary | |
1363 | * page table adjustments can then be made atomically. | |
38b5faf4 DM |
1364 | * |
1365 | * if the boolean frontswap is true, only unuse pages_to_unuse pages; | |
1366 | * pages_to_unuse==0 means all pages; ignored if frontswap is false | |
1da177e4 | 1367 | */ |
38b5faf4 DM |
1368 | int try_to_unuse(unsigned int type, bool frontswap, |
1369 | unsigned long pages_to_unuse) | |
1da177e4 | 1370 | { |
efa90a98 | 1371 | struct swap_info_struct *si = swap_info[type]; |
1da177e4 | 1372 | struct mm_struct *start_mm; |
edfe23da SL |
1373 | volatile unsigned char *swap_map; /* swap_map is accessed without |
1374 | * locking. Mark it as volatile | |
1375 | * to prevent compiler doing | |
1376 | * something odd. | |
1377 | */ | |
8d69aaee | 1378 | unsigned char swcount; |
1da177e4 LT |
1379 | struct page *page; |
1380 | swp_entry_t entry; | |
6eb396dc | 1381 | unsigned int i = 0; |
1da177e4 | 1382 | int retval = 0; |
1da177e4 LT |
1383 | |
1384 | /* | |
1385 | * When searching mms for an entry, a good strategy is to | |
1386 | * start at the first mm we freed the previous entry from | |
1387 | * (though actually we don't notice whether we or coincidence | |
1388 | * freed the entry). Initialize this start_mm with a hold. | |
1389 | * | |
1390 | * A simpler strategy would be to start at the last mm we | |
1391 | * freed the previous entry from; but that would take less | |
1392 | * advantage of mmlist ordering, which clusters forked mms | |
1393 | * together, child after parent. If we race with dup_mmap(), we | |
1394 | * prefer to resolve parent before child, lest we miss entries | |
1395 | * duplicated after we scanned child: using last mm would invert | |
570a335b | 1396 | * that. |
1da177e4 LT |
1397 | */ |
1398 | start_mm = &init_mm; | |
1399 | atomic_inc(&init_mm.mm_users); | |
1400 | ||
1401 | /* | |
1402 | * Keep on scanning until all entries have gone. Usually, | |
1403 | * one pass through swap_map is enough, but not necessarily: | |
1404 | * there are races when an instance of an entry might be missed. | |
1405 | */ | |
38b5faf4 | 1406 | while ((i = find_next_to_unuse(si, i, frontswap)) != 0) { |
1da177e4 LT |
1407 | if (signal_pending(current)) { |
1408 | retval = -EINTR; | |
1409 | break; | |
1410 | } | |
1411 | ||
886bb7e9 | 1412 | /* |
1da177e4 LT |
1413 | * Get a page for the entry, using the existing swap |
1414 | * cache page if there is one. Otherwise, get a clean | |
886bb7e9 | 1415 | * page and read the swap into it. |
1da177e4 LT |
1416 | */ |
1417 | swap_map = &si->swap_map[i]; | |
1418 | entry = swp_entry(type, i); | |
02098fea HD |
1419 | page = read_swap_cache_async(entry, |
1420 | GFP_HIGHUSER_MOVABLE, NULL, 0); | |
1da177e4 LT |
1421 | if (!page) { |
1422 | /* | |
1423 | * Either swap_duplicate() failed because entry | |
1424 | * has been freed independently, and will not be | |
1425 | * reused since sys_swapoff() already disabled | |
1426 | * allocation from here, or alloc_page() failed. | |
1427 | */ | |
edfe23da SL |
1428 | swcount = *swap_map; |
1429 | /* | |
1430 | * We don't hold lock here, so the swap entry could be | |
1431 | * SWAP_MAP_BAD (when the cluster is discarding). | |
1432 | * Instead of fail out, We can just skip the swap | |
1433 | * entry because swapoff will wait for discarding | |
1434 | * finish anyway. | |
1435 | */ | |
1436 | if (!swcount || swcount == SWAP_MAP_BAD) | |
1da177e4 LT |
1437 | continue; |
1438 | retval = -ENOMEM; | |
1439 | break; | |
1440 | } | |
1441 | ||
1442 | /* | |
1443 | * Don't hold on to start_mm if it looks like exiting. | |
1444 | */ | |
1445 | if (atomic_read(&start_mm->mm_users) == 1) { | |
1446 | mmput(start_mm); | |
1447 | start_mm = &init_mm; | |
1448 | atomic_inc(&init_mm.mm_users); | |
1449 | } | |
1450 | ||
1451 | /* | |
1452 | * Wait for and lock page. When do_swap_page races with | |
1453 | * try_to_unuse, do_swap_page can handle the fault much | |
1454 | * faster than try_to_unuse can locate the entry. This | |
1455 | * apparently redundant "wait_on_page_locked" lets try_to_unuse | |
1456 | * defer to do_swap_page in such a case - in some tests, | |
1457 | * do_swap_page and try_to_unuse repeatedly compete. | |
1458 | */ | |
1459 | wait_on_page_locked(page); | |
1460 | wait_on_page_writeback(page); | |
1461 | lock_page(page); | |
1462 | wait_on_page_writeback(page); | |
1463 | ||
1464 | /* | |
1465 | * Remove all references to entry. | |
1da177e4 | 1466 | */ |
1da177e4 | 1467 | swcount = *swap_map; |
aaa46865 HD |
1468 | if (swap_count(swcount) == SWAP_MAP_SHMEM) { |
1469 | retval = shmem_unuse(entry, page); | |
1470 | /* page has already been unlocked and released */ | |
1471 | if (retval < 0) | |
1472 | break; | |
1473 | continue; | |
1da177e4 | 1474 | } |
aaa46865 HD |
1475 | if (swap_count(swcount) && start_mm != &init_mm) |
1476 | retval = unuse_mm(start_mm, entry, page); | |
1477 | ||
355cfa73 | 1478 | if (swap_count(*swap_map)) { |
1da177e4 LT |
1479 | int set_start_mm = (*swap_map >= swcount); |
1480 | struct list_head *p = &start_mm->mmlist; | |
1481 | struct mm_struct *new_start_mm = start_mm; | |
1482 | struct mm_struct *prev_mm = start_mm; | |
1483 | struct mm_struct *mm; | |
1484 | ||
1485 | atomic_inc(&new_start_mm->mm_users); | |
1486 | atomic_inc(&prev_mm->mm_users); | |
1487 | spin_lock(&mmlist_lock); | |
aaa46865 | 1488 | while (swap_count(*swap_map) && !retval && |
1da177e4 LT |
1489 | (p = p->next) != &start_mm->mmlist) { |
1490 | mm = list_entry(p, struct mm_struct, mmlist); | |
70af7c5c | 1491 | if (!atomic_inc_not_zero(&mm->mm_users)) |
1da177e4 | 1492 | continue; |
1da177e4 LT |
1493 | spin_unlock(&mmlist_lock); |
1494 | mmput(prev_mm); | |
1495 | prev_mm = mm; | |
1496 | ||
1497 | cond_resched(); | |
1498 | ||
1499 | swcount = *swap_map; | |
355cfa73 | 1500 | if (!swap_count(swcount)) /* any usage ? */ |
1da177e4 | 1501 | ; |
aaa46865 | 1502 | else if (mm == &init_mm) |
1da177e4 | 1503 | set_start_mm = 1; |
aaa46865 | 1504 | else |
1da177e4 | 1505 | retval = unuse_mm(mm, entry, page); |
355cfa73 | 1506 | |
32c5fc10 | 1507 | if (set_start_mm && *swap_map < swcount) { |
1da177e4 LT |
1508 | mmput(new_start_mm); |
1509 | atomic_inc(&mm->mm_users); | |
1510 | new_start_mm = mm; | |
1511 | set_start_mm = 0; | |
1512 | } | |
1513 | spin_lock(&mmlist_lock); | |
1514 | } | |
1515 | spin_unlock(&mmlist_lock); | |
1516 | mmput(prev_mm); | |
1517 | mmput(start_mm); | |
1518 | start_mm = new_start_mm; | |
1519 | } | |
1520 | if (retval) { | |
1521 | unlock_page(page); | |
09cbfeaf | 1522 | put_page(page); |
1da177e4 LT |
1523 | break; |
1524 | } | |
1525 | ||
1da177e4 LT |
1526 | /* |
1527 | * If a reference remains (rare), we would like to leave | |
1528 | * the page in the swap cache; but try_to_unmap could | |
1529 | * then re-duplicate the entry once we drop page lock, | |
1530 | * so we might loop indefinitely; also, that page could | |
1531 | * not be swapped out to other storage meanwhile. So: | |
1532 | * delete from cache even if there's another reference, | |
1533 | * after ensuring that the data has been saved to disk - | |
1534 | * since if the reference remains (rarer), it will be | |
1535 | * read from disk into another page. Splitting into two | |
1536 | * pages would be incorrect if swap supported "shared | |
1537 | * private" pages, but they are handled by tmpfs files. | |
5ad64688 HD |
1538 | * |
1539 | * Given how unuse_vma() targets one particular offset | |
1540 | * in an anon_vma, once the anon_vma has been determined, | |
1541 | * this splitting happens to be just what is needed to | |
1542 | * handle where KSM pages have been swapped out: re-reading | |
1543 | * is unnecessarily slow, but we can fix that later on. | |
1da177e4 | 1544 | */ |
355cfa73 KH |
1545 | if (swap_count(*swap_map) && |
1546 | PageDirty(page) && PageSwapCache(page)) { | |
1da177e4 LT |
1547 | struct writeback_control wbc = { |
1548 | .sync_mode = WB_SYNC_NONE, | |
1549 | }; | |
1550 | ||
1551 | swap_writepage(page, &wbc); | |
1552 | lock_page(page); | |
1553 | wait_on_page_writeback(page); | |
1554 | } | |
68bdc8d6 HD |
1555 | |
1556 | /* | |
1557 | * It is conceivable that a racing task removed this page from | |
1558 | * swap cache just before we acquired the page lock at the top, | |
1559 | * or while we dropped it in unuse_mm(). The page might even | |
1560 | * be back in swap cache on another swap area: that we must not | |
1561 | * delete, since it may not have been written out to swap yet. | |
1562 | */ | |
1563 | if (PageSwapCache(page) && | |
1564 | likely(page_private(page) == entry.val)) | |
2e0e26c7 | 1565 | delete_from_swap_cache(page); |
1da177e4 LT |
1566 | |
1567 | /* | |
1568 | * So we could skip searching mms once swap count went | |
1569 | * to 1, we did not mark any present ptes as dirty: must | |
2706a1b8 | 1570 | * mark page dirty so shrink_page_list will preserve it. |
1da177e4 LT |
1571 | */ |
1572 | SetPageDirty(page); | |
1573 | unlock_page(page); | |
09cbfeaf | 1574 | put_page(page); |
1da177e4 LT |
1575 | |
1576 | /* | |
1577 | * Make sure that we aren't completely killing | |
1578 | * interactive performance. | |
1579 | */ | |
1580 | cond_resched(); | |
38b5faf4 DM |
1581 | if (frontswap && pages_to_unuse > 0) { |
1582 | if (!--pages_to_unuse) | |
1583 | break; | |
1584 | } | |
1da177e4 LT |
1585 | } |
1586 | ||
1587 | mmput(start_mm); | |
1da177e4 LT |
1588 | return retval; |
1589 | } | |
1590 | ||
1591 | /* | |
5d337b91 HD |
1592 | * After a successful try_to_unuse, if no swap is now in use, we know |
1593 | * we can empty the mmlist. swap_lock must be held on entry and exit. | |
1594 | * Note that mmlist_lock nests inside swap_lock, and an mm must be | |
1da177e4 LT |
1595 | * added to the mmlist just after page_duplicate - before would be racy. |
1596 | */ | |
1597 | static void drain_mmlist(void) | |
1598 | { | |
1599 | struct list_head *p, *next; | |
efa90a98 | 1600 | unsigned int type; |
1da177e4 | 1601 | |
efa90a98 HD |
1602 | for (type = 0; type < nr_swapfiles; type++) |
1603 | if (swap_info[type]->inuse_pages) | |
1da177e4 LT |
1604 | return; |
1605 | spin_lock(&mmlist_lock); | |
1606 | list_for_each_safe(p, next, &init_mm.mmlist) | |
1607 | list_del_init(p); | |
1608 | spin_unlock(&mmlist_lock); | |
1609 | } | |
1610 | ||
1611 | /* | |
1612 | * Use this swapdev's extent info to locate the (PAGE_SIZE) block which | |
d4906e1a LS |
1613 | * corresponds to page offset for the specified swap entry. |
1614 | * Note that the type of this function is sector_t, but it returns page offset | |
1615 | * into the bdev, not sector offset. | |
1da177e4 | 1616 | */ |
d4906e1a | 1617 | static sector_t map_swap_entry(swp_entry_t entry, struct block_device **bdev) |
1da177e4 | 1618 | { |
f29ad6a9 HD |
1619 | struct swap_info_struct *sis; |
1620 | struct swap_extent *start_se; | |
1621 | struct swap_extent *se; | |
1622 | pgoff_t offset; | |
1623 | ||
efa90a98 | 1624 | sis = swap_info[swp_type(entry)]; |
f29ad6a9 HD |
1625 | *bdev = sis->bdev; |
1626 | ||
1627 | offset = swp_offset(entry); | |
1628 | start_se = sis->curr_swap_extent; | |
1629 | se = start_se; | |
1da177e4 LT |
1630 | |
1631 | for ( ; ; ) { | |
1da177e4 LT |
1632 | if (se->start_page <= offset && |
1633 | offset < (se->start_page + se->nr_pages)) { | |
1634 | return se->start_block + (offset - se->start_page); | |
1635 | } | |
a8ae4991 | 1636 | se = list_next_entry(se, list); |
1da177e4 LT |
1637 | sis->curr_swap_extent = se; |
1638 | BUG_ON(se == start_se); /* It *must* be present */ | |
1639 | } | |
1640 | } | |
1641 | ||
d4906e1a LS |
1642 | /* |
1643 | * Returns the page offset into bdev for the specified page's swap entry. | |
1644 | */ | |
1645 | sector_t map_swap_page(struct page *page, struct block_device **bdev) | |
1646 | { | |
1647 | swp_entry_t entry; | |
1648 | entry.val = page_private(page); | |
1649 | return map_swap_entry(entry, bdev); | |
1650 | } | |
1651 | ||
1da177e4 LT |
1652 | /* |
1653 | * Free all of a swapdev's extent information | |
1654 | */ | |
1655 | static void destroy_swap_extents(struct swap_info_struct *sis) | |
1656 | { | |
9625a5f2 | 1657 | while (!list_empty(&sis->first_swap_extent.list)) { |
1da177e4 LT |
1658 | struct swap_extent *se; |
1659 | ||
a8ae4991 | 1660 | se = list_first_entry(&sis->first_swap_extent.list, |
1da177e4 LT |
1661 | struct swap_extent, list); |
1662 | list_del(&se->list); | |
1663 | kfree(se); | |
1664 | } | |
62c230bc MG |
1665 | |
1666 | if (sis->flags & SWP_FILE) { | |
1667 | struct file *swap_file = sis->swap_file; | |
1668 | struct address_space *mapping = swap_file->f_mapping; | |
1669 | ||
1670 | sis->flags &= ~SWP_FILE; | |
1671 | mapping->a_ops->swap_deactivate(swap_file); | |
1672 | } | |
1da177e4 LT |
1673 | } |
1674 | ||
1675 | /* | |
1676 | * Add a block range (and the corresponding page range) into this swapdev's | |
11d31886 | 1677 | * extent list. The extent list is kept sorted in page order. |
1da177e4 | 1678 | * |
11d31886 | 1679 | * This function rather assumes that it is called in ascending page order. |
1da177e4 | 1680 | */ |
a509bc1a | 1681 | int |
1da177e4 LT |
1682 | add_swap_extent(struct swap_info_struct *sis, unsigned long start_page, |
1683 | unsigned long nr_pages, sector_t start_block) | |
1684 | { | |
1685 | struct swap_extent *se; | |
1686 | struct swap_extent *new_se; | |
1687 | struct list_head *lh; | |
1688 | ||
9625a5f2 HD |
1689 | if (start_page == 0) { |
1690 | se = &sis->first_swap_extent; | |
1691 | sis->curr_swap_extent = se; | |
1692 | se->start_page = 0; | |
1693 | se->nr_pages = nr_pages; | |
1694 | se->start_block = start_block; | |
1695 | return 1; | |
1696 | } else { | |
1697 | lh = sis->first_swap_extent.list.prev; /* Highest extent */ | |
1da177e4 | 1698 | se = list_entry(lh, struct swap_extent, list); |
11d31886 HD |
1699 | BUG_ON(se->start_page + se->nr_pages != start_page); |
1700 | if (se->start_block + se->nr_pages == start_block) { | |
1da177e4 LT |
1701 | /* Merge it */ |
1702 | se->nr_pages += nr_pages; | |
1703 | return 0; | |
1704 | } | |
1da177e4 LT |
1705 | } |
1706 | ||
1707 | /* | |
1708 | * No merge. Insert a new extent, preserving ordering. | |
1709 | */ | |
1710 | new_se = kmalloc(sizeof(*se), GFP_KERNEL); | |
1711 | if (new_se == NULL) | |
1712 | return -ENOMEM; | |
1713 | new_se->start_page = start_page; | |
1714 | new_se->nr_pages = nr_pages; | |
1715 | new_se->start_block = start_block; | |
1716 | ||
9625a5f2 | 1717 | list_add_tail(&new_se->list, &sis->first_swap_extent.list); |
53092a74 | 1718 | return 1; |
1da177e4 LT |
1719 | } |
1720 | ||
1721 | /* | |
1722 | * A `swap extent' is a simple thing which maps a contiguous range of pages | |
1723 | * onto a contiguous range of disk blocks. An ordered list of swap extents | |
1724 | * is built at swapon time and is then used at swap_writepage/swap_readpage | |
1725 | * time for locating where on disk a page belongs. | |
1726 | * | |
1727 | * If the swapfile is an S_ISBLK block device, a single extent is installed. | |
1728 | * This is done so that the main operating code can treat S_ISBLK and S_ISREG | |
1729 | * swap files identically. | |
1730 | * | |
1731 | * Whether the swapdev is an S_ISREG file or an S_ISBLK blockdev, the swap | |
1732 | * extent list operates in PAGE_SIZE disk blocks. Both S_ISREG and S_ISBLK | |
1733 | * swapfiles are handled *identically* after swapon time. | |
1734 | * | |
1735 | * For S_ISREG swapfiles, setup_swap_extents() will walk all the file's blocks | |
1736 | * and will parse them into an ordered extent list, in PAGE_SIZE chunks. If | |
1737 | * some stray blocks are found which do not fall within the PAGE_SIZE alignment | |
1738 | * requirements, they are simply tossed out - we will never use those blocks | |
1739 | * for swapping. | |
1740 | * | |
b0d9bcd4 | 1741 | * For S_ISREG swapfiles we set S_SWAPFILE across the life of the swapon. This |
1da177e4 LT |
1742 | * prevents root from shooting her foot off by ftruncating an in-use swapfile, |
1743 | * which will scribble on the fs. | |
1744 | * | |
1745 | * The amount of disk space which a single swap extent represents varies. | |
1746 | * Typically it is in the 1-4 megabyte range. So we can have hundreds of | |
1747 | * extents in the list. To avoid much list walking, we cache the previous | |
1748 | * search location in `curr_swap_extent', and start new searches from there. | |
1749 | * This is extremely effective. The average number of iterations in | |
1750 | * map_swap_page() has been measured at about 0.3 per page. - akpm. | |
1751 | */ | |
53092a74 | 1752 | static int setup_swap_extents(struct swap_info_struct *sis, sector_t *span) |
1da177e4 | 1753 | { |
62c230bc MG |
1754 | struct file *swap_file = sis->swap_file; |
1755 | struct address_space *mapping = swap_file->f_mapping; | |
1756 | struct inode *inode = mapping->host; | |
1da177e4 LT |
1757 | int ret; |
1758 | ||
1da177e4 LT |
1759 | if (S_ISBLK(inode->i_mode)) { |
1760 | ret = add_swap_extent(sis, 0, sis->max, 0); | |
53092a74 | 1761 | *span = sis->pages; |
a509bc1a | 1762 | return ret; |
1da177e4 LT |
1763 | } |
1764 | ||
62c230bc | 1765 | if (mapping->a_ops->swap_activate) { |
a509bc1a | 1766 | ret = mapping->a_ops->swap_activate(sis, swap_file, span); |
62c230bc MG |
1767 | if (!ret) { |
1768 | sis->flags |= SWP_FILE; | |
1769 | ret = add_swap_extent(sis, 0, sis->max, 0); | |
1770 | *span = sis->pages; | |
1771 | } | |
a509bc1a | 1772 | return ret; |
62c230bc MG |
1773 | } |
1774 | ||
a509bc1a | 1775 | return generic_swapfile_activate(sis, swap_file, span); |
1da177e4 LT |
1776 | } |
1777 | ||
cf0cac0a | 1778 | static void _enable_swap_info(struct swap_info_struct *p, int prio, |
2a8f9449 SL |
1779 | unsigned char *swap_map, |
1780 | struct swap_cluster_info *cluster_info) | |
40531542 | 1781 | { |
40531542 CEB |
1782 | if (prio >= 0) |
1783 | p->prio = prio; | |
1784 | else | |
1785 | p->prio = --least_priority; | |
18ab4d4c DS |
1786 | /* |
1787 | * the plist prio is negated because plist ordering is | |
1788 | * low-to-high, while swap ordering is high-to-low | |
1789 | */ | |
1790 | p->list.prio = -p->prio; | |
1791 | p->avail_list.prio = -p->prio; | |
40531542 | 1792 | p->swap_map = swap_map; |
2a8f9449 | 1793 | p->cluster_info = cluster_info; |
40531542 | 1794 | p->flags |= SWP_WRITEOK; |
ec8acf20 | 1795 | atomic_long_add(p->pages, &nr_swap_pages); |
40531542 CEB |
1796 | total_swap_pages += p->pages; |
1797 | ||
adfab836 | 1798 | assert_spin_locked(&swap_lock); |
adfab836 | 1799 | /* |
18ab4d4c DS |
1800 | * both lists are plists, and thus priority ordered. |
1801 | * swap_active_head needs to be priority ordered for swapoff(), | |
1802 | * which on removal of any swap_info_struct with an auto-assigned | |
1803 | * (i.e. negative) priority increments the auto-assigned priority | |
1804 | * of any lower-priority swap_info_structs. | |
1805 | * swap_avail_head needs to be priority ordered for get_swap_page(), | |
1806 | * which allocates swap pages from the highest available priority | |
1807 | * swap_info_struct. | |
adfab836 | 1808 | */ |
18ab4d4c DS |
1809 | plist_add(&p->list, &swap_active_head); |
1810 | spin_lock(&swap_avail_lock); | |
1811 | plist_add(&p->avail_list, &swap_avail_head); | |
1812 | spin_unlock(&swap_avail_lock); | |
cf0cac0a CEB |
1813 | } |
1814 | ||
1815 | static void enable_swap_info(struct swap_info_struct *p, int prio, | |
1816 | unsigned char *swap_map, | |
2a8f9449 | 1817 | struct swap_cluster_info *cluster_info, |
cf0cac0a CEB |
1818 | unsigned long *frontswap_map) |
1819 | { | |
4f89849d | 1820 | frontswap_init(p->type, frontswap_map); |
cf0cac0a | 1821 | spin_lock(&swap_lock); |
ec8acf20 | 1822 | spin_lock(&p->lock); |
2a8f9449 | 1823 | _enable_swap_info(p, prio, swap_map, cluster_info); |
ec8acf20 | 1824 | spin_unlock(&p->lock); |
cf0cac0a CEB |
1825 | spin_unlock(&swap_lock); |
1826 | } | |
1827 | ||
1828 | static void reinsert_swap_info(struct swap_info_struct *p) | |
1829 | { | |
1830 | spin_lock(&swap_lock); | |
ec8acf20 | 1831 | spin_lock(&p->lock); |
2a8f9449 | 1832 | _enable_swap_info(p, p->prio, p->swap_map, p->cluster_info); |
ec8acf20 | 1833 | spin_unlock(&p->lock); |
40531542 CEB |
1834 | spin_unlock(&swap_lock); |
1835 | } | |
1836 | ||
c4ea37c2 | 1837 | SYSCALL_DEFINE1(swapoff, const char __user *, specialfile) |
1da177e4 | 1838 | { |
73c34b6a | 1839 | struct swap_info_struct *p = NULL; |
8d69aaee | 1840 | unsigned char *swap_map; |
2a8f9449 | 1841 | struct swap_cluster_info *cluster_info; |
4f89849d | 1842 | unsigned long *frontswap_map; |
1da177e4 LT |
1843 | struct file *swap_file, *victim; |
1844 | struct address_space *mapping; | |
1845 | struct inode *inode; | |
91a27b2a | 1846 | struct filename *pathname; |
adfab836 | 1847 | int err, found = 0; |
5b808a23 | 1848 | unsigned int old_block_size; |
886bb7e9 | 1849 | |
1da177e4 LT |
1850 | if (!capable(CAP_SYS_ADMIN)) |
1851 | return -EPERM; | |
1852 | ||
191c5424 AV |
1853 | BUG_ON(!current->mm); |
1854 | ||
1da177e4 | 1855 | pathname = getname(specialfile); |
1da177e4 | 1856 | if (IS_ERR(pathname)) |
f58b59c1 | 1857 | return PTR_ERR(pathname); |
1da177e4 | 1858 | |
669abf4e | 1859 | victim = file_open_name(pathname, O_RDWR|O_LARGEFILE, 0); |
1da177e4 LT |
1860 | err = PTR_ERR(victim); |
1861 | if (IS_ERR(victim)) | |
1862 | goto out; | |
1863 | ||
1864 | mapping = victim->f_mapping; | |
5d337b91 | 1865 | spin_lock(&swap_lock); |
18ab4d4c | 1866 | plist_for_each_entry(p, &swap_active_head, list) { |
22c6f8fd | 1867 | if (p->flags & SWP_WRITEOK) { |
adfab836 DS |
1868 | if (p->swap_file->f_mapping == mapping) { |
1869 | found = 1; | |
1da177e4 | 1870 | break; |
adfab836 | 1871 | } |
1da177e4 | 1872 | } |
1da177e4 | 1873 | } |
adfab836 | 1874 | if (!found) { |
1da177e4 | 1875 | err = -EINVAL; |
5d337b91 | 1876 | spin_unlock(&swap_lock); |
1da177e4 LT |
1877 | goto out_dput; |
1878 | } | |
191c5424 | 1879 | if (!security_vm_enough_memory_mm(current->mm, p->pages)) |
1da177e4 LT |
1880 | vm_unacct_memory(p->pages); |
1881 | else { | |
1882 | err = -ENOMEM; | |
5d337b91 | 1883 | spin_unlock(&swap_lock); |
1da177e4 LT |
1884 | goto out_dput; |
1885 | } | |
18ab4d4c DS |
1886 | spin_lock(&swap_avail_lock); |
1887 | plist_del(&p->avail_list, &swap_avail_head); | |
1888 | spin_unlock(&swap_avail_lock); | |
ec8acf20 | 1889 | spin_lock(&p->lock); |
78ecba08 | 1890 | if (p->prio < 0) { |
adfab836 DS |
1891 | struct swap_info_struct *si = p; |
1892 | ||
18ab4d4c | 1893 | plist_for_each_entry_continue(si, &swap_active_head, list) { |
adfab836 | 1894 | si->prio++; |
18ab4d4c DS |
1895 | si->list.prio--; |
1896 | si->avail_list.prio--; | |
adfab836 | 1897 | } |
78ecba08 HD |
1898 | least_priority++; |
1899 | } | |
18ab4d4c | 1900 | plist_del(&p->list, &swap_active_head); |
ec8acf20 | 1901 | atomic_long_sub(p->pages, &nr_swap_pages); |
1da177e4 LT |
1902 | total_swap_pages -= p->pages; |
1903 | p->flags &= ~SWP_WRITEOK; | |
ec8acf20 | 1904 | spin_unlock(&p->lock); |
5d337b91 | 1905 | spin_unlock(&swap_lock); |
fb4f88dc | 1906 | |
e1e12d2f | 1907 | set_current_oom_origin(); |
adfab836 | 1908 | err = try_to_unuse(p->type, false, 0); /* force unuse all pages */ |
e1e12d2f | 1909 | clear_current_oom_origin(); |
1da177e4 | 1910 | |
1da177e4 LT |
1911 | if (err) { |
1912 | /* re-insert swap space back into swap_list */ | |
cf0cac0a | 1913 | reinsert_swap_info(p); |
1da177e4 LT |
1914 | goto out_dput; |
1915 | } | |
52b7efdb | 1916 | |
815c2c54 SL |
1917 | flush_work(&p->discard_work); |
1918 | ||
5d337b91 | 1919 | destroy_swap_extents(p); |
570a335b HD |
1920 | if (p->flags & SWP_CONTINUED) |
1921 | free_swap_count_continuations(p); | |
1922 | ||
fc0abb14 | 1923 | mutex_lock(&swapon_mutex); |
5d337b91 | 1924 | spin_lock(&swap_lock); |
ec8acf20 | 1925 | spin_lock(&p->lock); |
5d337b91 HD |
1926 | drain_mmlist(); |
1927 | ||
52b7efdb | 1928 | /* wait for anyone still in scan_swap_map */ |
52b7efdb HD |
1929 | p->highest_bit = 0; /* cuts scans short */ |
1930 | while (p->flags >= SWP_SCANNING) { | |
ec8acf20 | 1931 | spin_unlock(&p->lock); |
5d337b91 | 1932 | spin_unlock(&swap_lock); |
13e4b57f | 1933 | schedule_timeout_uninterruptible(1); |
5d337b91 | 1934 | spin_lock(&swap_lock); |
ec8acf20 | 1935 | spin_lock(&p->lock); |
52b7efdb | 1936 | } |
52b7efdb | 1937 | |
1da177e4 | 1938 | swap_file = p->swap_file; |
5b808a23 | 1939 | old_block_size = p->old_block_size; |
1da177e4 LT |
1940 | p->swap_file = NULL; |
1941 | p->max = 0; | |
1942 | swap_map = p->swap_map; | |
1943 | p->swap_map = NULL; | |
2a8f9449 SL |
1944 | cluster_info = p->cluster_info; |
1945 | p->cluster_info = NULL; | |
4f89849d | 1946 | frontswap_map = frontswap_map_get(p); |
ec8acf20 | 1947 | spin_unlock(&p->lock); |
5d337b91 | 1948 | spin_unlock(&swap_lock); |
adfab836 | 1949 | frontswap_invalidate_area(p->type); |
58e97ba6 | 1950 | frontswap_map_set(p, NULL); |
fc0abb14 | 1951 | mutex_unlock(&swapon_mutex); |
ebc2a1a6 SL |
1952 | free_percpu(p->percpu_cluster); |
1953 | p->percpu_cluster = NULL; | |
1da177e4 | 1954 | vfree(swap_map); |
2a8f9449 | 1955 | vfree(cluster_info); |
4f89849d | 1956 | vfree(frontswap_map); |
2de1a7e4 | 1957 | /* Destroy swap account information */ |
adfab836 | 1958 | swap_cgroup_swapoff(p->type); |
27a7faa0 | 1959 | |
1da177e4 LT |
1960 | inode = mapping->host; |
1961 | if (S_ISBLK(inode->i_mode)) { | |
1962 | struct block_device *bdev = I_BDEV(inode); | |
5b808a23 | 1963 | set_blocksize(bdev, old_block_size); |
e525fd89 | 1964 | blkdev_put(bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL); |
1da177e4 | 1965 | } else { |
5955102c | 1966 | inode_lock(inode); |
1da177e4 | 1967 | inode->i_flags &= ~S_SWAPFILE; |
5955102c | 1968 | inode_unlock(inode); |
1da177e4 LT |
1969 | } |
1970 | filp_close(swap_file, NULL); | |
f893ab41 WY |
1971 | |
1972 | /* | |
1973 | * Clear the SWP_USED flag after all resources are freed so that swapon | |
1974 | * can reuse this swap_info in alloc_swap_info() safely. It is ok to | |
1975 | * not hold p->lock after we cleared its SWP_WRITEOK. | |
1976 | */ | |
1977 | spin_lock(&swap_lock); | |
1978 | p->flags = 0; | |
1979 | spin_unlock(&swap_lock); | |
1980 | ||
1da177e4 | 1981 | err = 0; |
66d7dd51 KS |
1982 | atomic_inc(&proc_poll_event); |
1983 | wake_up_interruptible(&proc_poll_wait); | |
1da177e4 LT |
1984 | |
1985 | out_dput: | |
1986 | filp_close(victim, NULL); | |
1987 | out: | |
f58b59c1 | 1988 | putname(pathname); |
1da177e4 LT |
1989 | return err; |
1990 | } | |
1991 | ||
1992 | #ifdef CONFIG_PROC_FS | |
66d7dd51 KS |
1993 | static unsigned swaps_poll(struct file *file, poll_table *wait) |
1994 | { | |
f1514638 | 1995 | struct seq_file *seq = file->private_data; |
66d7dd51 KS |
1996 | |
1997 | poll_wait(file, &proc_poll_wait, wait); | |
1998 | ||
f1514638 KS |
1999 | if (seq->poll_event != atomic_read(&proc_poll_event)) { |
2000 | seq->poll_event = atomic_read(&proc_poll_event); | |
66d7dd51 KS |
2001 | return POLLIN | POLLRDNORM | POLLERR | POLLPRI; |
2002 | } | |
2003 | ||
2004 | return POLLIN | POLLRDNORM; | |
2005 | } | |
2006 | ||
1da177e4 LT |
2007 | /* iterator */ |
2008 | static void *swap_start(struct seq_file *swap, loff_t *pos) | |
2009 | { | |
efa90a98 HD |
2010 | struct swap_info_struct *si; |
2011 | int type; | |
1da177e4 LT |
2012 | loff_t l = *pos; |
2013 | ||
fc0abb14 | 2014 | mutex_lock(&swapon_mutex); |
1da177e4 | 2015 | |
881e4aab SS |
2016 | if (!l) |
2017 | return SEQ_START_TOKEN; | |
2018 | ||
efa90a98 HD |
2019 | for (type = 0; type < nr_swapfiles; type++) { |
2020 | smp_rmb(); /* read nr_swapfiles before swap_info[type] */ | |
2021 | si = swap_info[type]; | |
2022 | if (!(si->flags & SWP_USED) || !si->swap_map) | |
1da177e4 | 2023 | continue; |
881e4aab | 2024 | if (!--l) |
efa90a98 | 2025 | return si; |
1da177e4 LT |
2026 | } |
2027 | ||
2028 | return NULL; | |
2029 | } | |
2030 | ||
2031 | static void *swap_next(struct seq_file *swap, void *v, loff_t *pos) | |
2032 | { | |
efa90a98 HD |
2033 | struct swap_info_struct *si = v; |
2034 | int type; | |
1da177e4 | 2035 | |
881e4aab | 2036 | if (v == SEQ_START_TOKEN) |
efa90a98 HD |
2037 | type = 0; |
2038 | else | |
2039 | type = si->type + 1; | |
881e4aab | 2040 | |
efa90a98 HD |
2041 | for (; type < nr_swapfiles; type++) { |
2042 | smp_rmb(); /* read nr_swapfiles before swap_info[type] */ | |
2043 | si = swap_info[type]; | |
2044 | if (!(si->flags & SWP_USED) || !si->swap_map) | |
1da177e4 LT |
2045 | continue; |
2046 | ++*pos; | |
efa90a98 | 2047 | return si; |
1da177e4 LT |
2048 | } |
2049 | ||
2050 | return NULL; | |
2051 | } | |
2052 | ||
2053 | static void swap_stop(struct seq_file *swap, void *v) | |
2054 | { | |
fc0abb14 | 2055 | mutex_unlock(&swapon_mutex); |
1da177e4 LT |
2056 | } |
2057 | ||
2058 | static int swap_show(struct seq_file *swap, void *v) | |
2059 | { | |
efa90a98 | 2060 | struct swap_info_struct *si = v; |
1da177e4 LT |
2061 | struct file *file; |
2062 | int len; | |
2063 | ||
efa90a98 | 2064 | if (si == SEQ_START_TOKEN) { |
881e4aab SS |
2065 | seq_puts(swap,"Filename\t\t\t\tType\t\tSize\tUsed\tPriority\n"); |
2066 | return 0; | |
2067 | } | |
1da177e4 | 2068 | |
efa90a98 | 2069 | file = si->swap_file; |
2726d566 | 2070 | len = seq_file_path(swap, file, " \t\n\\"); |
6eb396dc | 2071 | seq_printf(swap, "%*s%s\t%u\t%u\t%d\n", |
886bb7e9 | 2072 | len < 40 ? 40 - len : 1, " ", |
496ad9aa | 2073 | S_ISBLK(file_inode(file)->i_mode) ? |
1da177e4 | 2074 | "partition" : "file\t", |
efa90a98 HD |
2075 | si->pages << (PAGE_SHIFT - 10), |
2076 | si->inuse_pages << (PAGE_SHIFT - 10), | |
2077 | si->prio); | |
1da177e4 LT |
2078 | return 0; |
2079 | } | |
2080 | ||
15ad7cdc | 2081 | static const struct seq_operations swaps_op = { |
1da177e4 LT |
2082 | .start = swap_start, |
2083 | .next = swap_next, | |
2084 | .stop = swap_stop, | |
2085 | .show = swap_show | |
2086 | }; | |
2087 | ||
2088 | static int swaps_open(struct inode *inode, struct file *file) | |
2089 | { | |
f1514638 | 2090 | struct seq_file *seq; |
66d7dd51 KS |
2091 | int ret; |
2092 | ||
66d7dd51 | 2093 | ret = seq_open(file, &swaps_op); |
f1514638 | 2094 | if (ret) |
66d7dd51 | 2095 | return ret; |
66d7dd51 | 2096 | |
f1514638 KS |
2097 | seq = file->private_data; |
2098 | seq->poll_event = atomic_read(&proc_poll_event); | |
2099 | return 0; | |
1da177e4 LT |
2100 | } |
2101 | ||
15ad7cdc | 2102 | static const struct file_operations proc_swaps_operations = { |
1da177e4 LT |
2103 | .open = swaps_open, |
2104 | .read = seq_read, | |
2105 | .llseek = seq_lseek, | |
2106 | .release = seq_release, | |
66d7dd51 | 2107 | .poll = swaps_poll, |
1da177e4 LT |
2108 | }; |
2109 | ||
2110 | static int __init procswaps_init(void) | |
2111 | { | |
3d71f86f | 2112 | proc_create("swaps", 0, NULL, &proc_swaps_operations); |
1da177e4 LT |
2113 | return 0; |
2114 | } | |
2115 | __initcall(procswaps_init); | |
2116 | #endif /* CONFIG_PROC_FS */ | |
2117 | ||
1796316a JB |
2118 | #ifdef MAX_SWAPFILES_CHECK |
2119 | static int __init max_swapfiles_check(void) | |
2120 | { | |
2121 | MAX_SWAPFILES_CHECK(); | |
2122 | return 0; | |
2123 | } | |
2124 | late_initcall(max_swapfiles_check); | |
2125 | #endif | |
2126 | ||
53cbb243 | 2127 | static struct swap_info_struct *alloc_swap_info(void) |
1da177e4 | 2128 | { |
73c34b6a | 2129 | struct swap_info_struct *p; |
1da177e4 | 2130 | unsigned int type; |
efa90a98 HD |
2131 | |
2132 | p = kzalloc(sizeof(*p), GFP_KERNEL); | |
2133 | if (!p) | |
53cbb243 | 2134 | return ERR_PTR(-ENOMEM); |
efa90a98 | 2135 | |
5d337b91 | 2136 | spin_lock(&swap_lock); |
efa90a98 HD |
2137 | for (type = 0; type < nr_swapfiles; type++) { |
2138 | if (!(swap_info[type]->flags & SWP_USED)) | |
1da177e4 | 2139 | break; |
efa90a98 | 2140 | } |
0697212a | 2141 | if (type >= MAX_SWAPFILES) { |
5d337b91 | 2142 | spin_unlock(&swap_lock); |
efa90a98 | 2143 | kfree(p); |
730c0581 | 2144 | return ERR_PTR(-EPERM); |
1da177e4 | 2145 | } |
efa90a98 HD |
2146 | if (type >= nr_swapfiles) { |
2147 | p->type = type; | |
2148 | swap_info[type] = p; | |
2149 | /* | |
2150 | * Write swap_info[type] before nr_swapfiles, in case a | |
2151 | * racing procfs swap_start() or swap_next() is reading them. | |
2152 | * (We never shrink nr_swapfiles, we never free this entry.) | |
2153 | */ | |
2154 | smp_wmb(); | |
2155 | nr_swapfiles++; | |
2156 | } else { | |
2157 | kfree(p); | |
2158 | p = swap_info[type]; | |
2159 | /* | |
2160 | * Do not memset this entry: a racing procfs swap_next() | |
2161 | * would be relying on p->type to remain valid. | |
2162 | */ | |
2163 | } | |
9625a5f2 | 2164 | INIT_LIST_HEAD(&p->first_swap_extent.list); |
18ab4d4c DS |
2165 | plist_node_init(&p->list, 0); |
2166 | plist_node_init(&p->avail_list, 0); | |
1da177e4 | 2167 | p->flags = SWP_USED; |
5d337b91 | 2168 | spin_unlock(&swap_lock); |
ec8acf20 | 2169 | spin_lock_init(&p->lock); |
efa90a98 | 2170 | |
53cbb243 | 2171 | return p; |
53cbb243 CEB |
2172 | } |
2173 | ||
4d0e1e10 CEB |
2174 | static int claim_swapfile(struct swap_info_struct *p, struct inode *inode) |
2175 | { | |
2176 | int error; | |
2177 | ||
2178 | if (S_ISBLK(inode->i_mode)) { | |
2179 | p->bdev = bdgrab(I_BDEV(inode)); | |
2180 | error = blkdev_get(p->bdev, | |
6f179af8 | 2181 | FMODE_READ | FMODE_WRITE | FMODE_EXCL, p); |
4d0e1e10 CEB |
2182 | if (error < 0) { |
2183 | p->bdev = NULL; | |
6f179af8 | 2184 | return error; |
4d0e1e10 CEB |
2185 | } |
2186 | p->old_block_size = block_size(p->bdev); | |
2187 | error = set_blocksize(p->bdev, PAGE_SIZE); | |
2188 | if (error < 0) | |
87ade72a | 2189 | return error; |
4d0e1e10 CEB |
2190 | p->flags |= SWP_BLKDEV; |
2191 | } else if (S_ISREG(inode->i_mode)) { | |
2192 | p->bdev = inode->i_sb->s_bdev; | |
5955102c | 2193 | inode_lock(inode); |
87ade72a CEB |
2194 | if (IS_SWAPFILE(inode)) |
2195 | return -EBUSY; | |
2196 | } else | |
2197 | return -EINVAL; | |
4d0e1e10 CEB |
2198 | |
2199 | return 0; | |
4d0e1e10 CEB |
2200 | } |
2201 | ||
ca8bd38b CEB |
2202 | static unsigned long read_swap_header(struct swap_info_struct *p, |
2203 | union swap_header *swap_header, | |
2204 | struct inode *inode) | |
2205 | { | |
2206 | int i; | |
2207 | unsigned long maxpages; | |
2208 | unsigned long swapfilepages; | |
d6bbbd29 | 2209 | unsigned long last_page; |
ca8bd38b CEB |
2210 | |
2211 | if (memcmp("SWAPSPACE2", swap_header->magic.magic, 10)) { | |
465c47fd | 2212 | pr_err("Unable to find swap-space signature\n"); |
38719025 | 2213 | return 0; |
ca8bd38b CEB |
2214 | } |
2215 | ||
2216 | /* swap partition endianess hack... */ | |
2217 | if (swab32(swap_header->info.version) == 1) { | |
2218 | swab32s(&swap_header->info.version); | |
2219 | swab32s(&swap_header->info.last_page); | |
2220 | swab32s(&swap_header->info.nr_badpages); | |
2221 | for (i = 0; i < swap_header->info.nr_badpages; i++) | |
2222 | swab32s(&swap_header->info.badpages[i]); | |
2223 | } | |
2224 | /* Check the swap header's sub-version */ | |
2225 | if (swap_header->info.version != 1) { | |
465c47fd AM |
2226 | pr_warn("Unable to handle swap header version %d\n", |
2227 | swap_header->info.version); | |
38719025 | 2228 | return 0; |
ca8bd38b CEB |
2229 | } |
2230 | ||
2231 | p->lowest_bit = 1; | |
2232 | p->cluster_next = 1; | |
2233 | p->cluster_nr = 0; | |
2234 | ||
2235 | /* | |
2236 | * Find out how many pages are allowed for a single swap | |
9b15b817 | 2237 | * device. There are two limiting factors: 1) the number |
a2c16d6c HD |
2238 | * of bits for the swap offset in the swp_entry_t type, and |
2239 | * 2) the number of bits in the swap pte as defined by the | |
9b15b817 | 2240 | * different architectures. In order to find the |
a2c16d6c | 2241 | * largest possible bit mask, a swap entry with swap type 0 |
ca8bd38b | 2242 | * and swap offset ~0UL is created, encoded to a swap pte, |
a2c16d6c | 2243 | * decoded to a swp_entry_t again, and finally the swap |
ca8bd38b CEB |
2244 | * offset is extracted. This will mask all the bits from |
2245 | * the initial ~0UL mask that can't be encoded in either | |
2246 | * the swp_entry_t or the architecture definition of a | |
9b15b817 | 2247 | * swap pte. |
ca8bd38b CEB |
2248 | */ |
2249 | maxpages = swp_offset(pte_to_swp_entry( | |
9b15b817 | 2250 | swp_entry_to_pte(swp_entry(0, ~0UL)))) + 1; |
d6bbbd29 RJ |
2251 | last_page = swap_header->info.last_page; |
2252 | if (last_page > maxpages) { | |
465c47fd | 2253 | pr_warn("Truncating oversized swap area, only using %luk out of %luk\n", |
d6bbbd29 RJ |
2254 | maxpages << (PAGE_SHIFT - 10), |
2255 | last_page << (PAGE_SHIFT - 10)); | |
2256 | } | |
2257 | if (maxpages > last_page) { | |
2258 | maxpages = last_page + 1; | |
ca8bd38b CEB |
2259 | /* p->max is an unsigned int: don't overflow it */ |
2260 | if ((unsigned int)maxpages == 0) | |
2261 | maxpages = UINT_MAX; | |
2262 | } | |
2263 | p->highest_bit = maxpages - 1; | |
2264 | ||
2265 | if (!maxpages) | |
38719025 | 2266 | return 0; |
ca8bd38b CEB |
2267 | swapfilepages = i_size_read(inode) >> PAGE_SHIFT; |
2268 | if (swapfilepages && maxpages > swapfilepages) { | |
465c47fd | 2269 | pr_warn("Swap area shorter than signature indicates\n"); |
38719025 | 2270 | return 0; |
ca8bd38b CEB |
2271 | } |
2272 | if (swap_header->info.nr_badpages && S_ISREG(inode->i_mode)) | |
38719025 | 2273 | return 0; |
ca8bd38b | 2274 | if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES) |
38719025 | 2275 | return 0; |
ca8bd38b CEB |
2276 | |
2277 | return maxpages; | |
ca8bd38b CEB |
2278 | } |
2279 | ||
915d4d7b CEB |
2280 | static int setup_swap_map_and_extents(struct swap_info_struct *p, |
2281 | union swap_header *swap_header, | |
2282 | unsigned char *swap_map, | |
2a8f9449 | 2283 | struct swap_cluster_info *cluster_info, |
915d4d7b CEB |
2284 | unsigned long maxpages, |
2285 | sector_t *span) | |
2286 | { | |
2287 | int i; | |
915d4d7b CEB |
2288 | unsigned int nr_good_pages; |
2289 | int nr_extents; | |
2a8f9449 SL |
2290 | unsigned long nr_clusters = DIV_ROUND_UP(maxpages, SWAPFILE_CLUSTER); |
2291 | unsigned long idx = p->cluster_next / SWAPFILE_CLUSTER; | |
915d4d7b CEB |
2292 | |
2293 | nr_good_pages = maxpages - 1; /* omit header page */ | |
2294 | ||
2a8f9449 SL |
2295 | cluster_set_null(&p->free_cluster_head); |
2296 | cluster_set_null(&p->free_cluster_tail); | |
815c2c54 SL |
2297 | cluster_set_null(&p->discard_cluster_head); |
2298 | cluster_set_null(&p->discard_cluster_tail); | |
2a8f9449 | 2299 | |
915d4d7b CEB |
2300 | for (i = 0; i < swap_header->info.nr_badpages; i++) { |
2301 | unsigned int page_nr = swap_header->info.badpages[i]; | |
bdb8e3f6 CEB |
2302 | if (page_nr == 0 || page_nr > swap_header->info.last_page) |
2303 | return -EINVAL; | |
915d4d7b CEB |
2304 | if (page_nr < maxpages) { |
2305 | swap_map[page_nr] = SWAP_MAP_BAD; | |
2306 | nr_good_pages--; | |
2a8f9449 SL |
2307 | /* |
2308 | * Haven't marked the cluster free yet, no list | |
2309 | * operation involved | |
2310 | */ | |
2311 | inc_cluster_info_page(p, cluster_info, page_nr); | |
915d4d7b CEB |
2312 | } |
2313 | } | |
2314 | ||
2a8f9449 SL |
2315 | /* Haven't marked the cluster free yet, no list operation involved */ |
2316 | for (i = maxpages; i < round_up(maxpages, SWAPFILE_CLUSTER); i++) | |
2317 | inc_cluster_info_page(p, cluster_info, i); | |
2318 | ||
915d4d7b CEB |
2319 | if (nr_good_pages) { |
2320 | swap_map[0] = SWAP_MAP_BAD; | |
2a8f9449 SL |
2321 | /* |
2322 | * Not mark the cluster free yet, no list | |
2323 | * operation involved | |
2324 | */ | |
2325 | inc_cluster_info_page(p, cluster_info, 0); | |
915d4d7b CEB |
2326 | p->max = maxpages; |
2327 | p->pages = nr_good_pages; | |
2328 | nr_extents = setup_swap_extents(p, span); | |
bdb8e3f6 CEB |
2329 | if (nr_extents < 0) |
2330 | return nr_extents; | |
915d4d7b CEB |
2331 | nr_good_pages = p->pages; |
2332 | } | |
2333 | if (!nr_good_pages) { | |
465c47fd | 2334 | pr_warn("Empty swap-file\n"); |
bdb8e3f6 | 2335 | return -EINVAL; |
915d4d7b CEB |
2336 | } |
2337 | ||
2a8f9449 SL |
2338 | if (!cluster_info) |
2339 | return nr_extents; | |
2340 | ||
2341 | for (i = 0; i < nr_clusters; i++) { | |
2342 | if (!cluster_count(&cluster_info[idx])) { | |
2343 | cluster_set_flag(&cluster_info[idx], CLUSTER_FLAG_FREE); | |
2344 | if (cluster_is_null(&p->free_cluster_head)) { | |
2345 | cluster_set_next_flag(&p->free_cluster_head, | |
2346 | idx, 0); | |
2347 | cluster_set_next_flag(&p->free_cluster_tail, | |
2348 | idx, 0); | |
2349 | } else { | |
2350 | unsigned int tail; | |
2351 | ||
2352 | tail = cluster_next(&p->free_cluster_tail); | |
2353 | cluster_set_next(&cluster_info[tail], idx); | |
2354 | cluster_set_next_flag(&p->free_cluster_tail, | |
2355 | idx, 0); | |
2356 | } | |
2357 | } | |
2358 | idx++; | |
2359 | if (idx == nr_clusters) | |
2360 | idx = 0; | |
2361 | } | |
915d4d7b | 2362 | return nr_extents; |
915d4d7b CEB |
2363 | } |
2364 | ||
dcf6b7dd RA |
2365 | /* |
2366 | * Helper to sys_swapon determining if a given swap | |
2367 | * backing device queue supports DISCARD operations. | |
2368 | */ | |
2369 | static bool swap_discardable(struct swap_info_struct *si) | |
2370 | { | |
2371 | struct request_queue *q = bdev_get_queue(si->bdev); | |
2372 | ||
2373 | if (!q || !blk_queue_discard(q)) | |
2374 | return false; | |
2375 | ||
2376 | return true; | |
2377 | } | |
2378 | ||
53cbb243 CEB |
2379 | SYSCALL_DEFINE2(swapon, const char __user *, specialfile, int, swap_flags) |
2380 | { | |
2381 | struct swap_info_struct *p; | |
91a27b2a | 2382 | struct filename *name; |
53cbb243 CEB |
2383 | struct file *swap_file = NULL; |
2384 | struct address_space *mapping; | |
40531542 | 2385 | int prio; |
53cbb243 CEB |
2386 | int error; |
2387 | union swap_header *swap_header; | |
915d4d7b | 2388 | int nr_extents; |
53cbb243 CEB |
2389 | sector_t span; |
2390 | unsigned long maxpages; | |
53cbb243 | 2391 | unsigned char *swap_map = NULL; |
2a8f9449 | 2392 | struct swap_cluster_info *cluster_info = NULL; |
38b5faf4 | 2393 | unsigned long *frontswap_map = NULL; |
53cbb243 CEB |
2394 | struct page *page = NULL; |
2395 | struct inode *inode = NULL; | |
53cbb243 | 2396 | |
d15cab97 HD |
2397 | if (swap_flags & ~SWAP_FLAGS_VALID) |
2398 | return -EINVAL; | |
2399 | ||
53cbb243 CEB |
2400 | if (!capable(CAP_SYS_ADMIN)) |
2401 | return -EPERM; | |
2402 | ||
2403 | p = alloc_swap_info(); | |
2542e513 CEB |
2404 | if (IS_ERR(p)) |
2405 | return PTR_ERR(p); | |
53cbb243 | 2406 | |
815c2c54 SL |
2407 | INIT_WORK(&p->discard_work, swap_discard_work); |
2408 | ||
1da177e4 | 2409 | name = getname(specialfile); |
1da177e4 | 2410 | if (IS_ERR(name)) { |
7de7fb6b | 2411 | error = PTR_ERR(name); |
1da177e4 | 2412 | name = NULL; |
bd69010b | 2413 | goto bad_swap; |
1da177e4 | 2414 | } |
669abf4e | 2415 | swap_file = file_open_name(name, O_RDWR|O_LARGEFILE, 0); |
1da177e4 | 2416 | if (IS_ERR(swap_file)) { |
7de7fb6b | 2417 | error = PTR_ERR(swap_file); |
1da177e4 | 2418 | swap_file = NULL; |
bd69010b | 2419 | goto bad_swap; |
1da177e4 LT |
2420 | } |
2421 | ||
2422 | p->swap_file = swap_file; | |
2423 | mapping = swap_file->f_mapping; | |
2130781e | 2424 | inode = mapping->host; |
6f179af8 | 2425 | |
5955102c | 2426 | /* If S_ISREG(inode->i_mode) will do inode_lock(inode); */ |
4d0e1e10 CEB |
2427 | error = claim_swapfile(p, inode); |
2428 | if (unlikely(error)) | |
1da177e4 | 2429 | goto bad_swap; |
1da177e4 | 2430 | |
1da177e4 LT |
2431 | /* |
2432 | * Read the swap header. | |
2433 | */ | |
2434 | if (!mapping->a_ops->readpage) { | |
2435 | error = -EINVAL; | |
2436 | goto bad_swap; | |
2437 | } | |
090d2b18 | 2438 | page = read_mapping_page(mapping, 0, swap_file); |
1da177e4 LT |
2439 | if (IS_ERR(page)) { |
2440 | error = PTR_ERR(page); | |
2441 | goto bad_swap; | |
2442 | } | |
81e33971 | 2443 | swap_header = kmap(page); |
1da177e4 | 2444 | |
ca8bd38b CEB |
2445 | maxpages = read_swap_header(p, swap_header, inode); |
2446 | if (unlikely(!maxpages)) { | |
1da177e4 LT |
2447 | error = -EINVAL; |
2448 | goto bad_swap; | |
2449 | } | |
886bb7e9 | 2450 | |
81e33971 | 2451 | /* OK, set up the swap map and apply the bad block list */ |
803d0c83 | 2452 | swap_map = vzalloc(maxpages); |
81e33971 HD |
2453 | if (!swap_map) { |
2454 | error = -ENOMEM; | |
2455 | goto bad_swap; | |
2456 | } | |
2a8f9449 | 2457 | if (p->bdev && blk_queue_nonrot(bdev_get_queue(p->bdev))) { |
6f179af8 HD |
2458 | int cpu; |
2459 | ||
2a8f9449 SL |
2460 | p->flags |= SWP_SOLIDSTATE; |
2461 | /* | |
2462 | * select a random position to start with to help wear leveling | |
2463 | * SSD | |
2464 | */ | |
2465 | p->cluster_next = 1 + (prandom_u32() % p->highest_bit); | |
2466 | ||
2467 | cluster_info = vzalloc(DIV_ROUND_UP(maxpages, | |
2468 | SWAPFILE_CLUSTER) * sizeof(*cluster_info)); | |
2469 | if (!cluster_info) { | |
2470 | error = -ENOMEM; | |
2471 | goto bad_swap; | |
2472 | } | |
ebc2a1a6 SL |
2473 | p->percpu_cluster = alloc_percpu(struct percpu_cluster); |
2474 | if (!p->percpu_cluster) { | |
2475 | error = -ENOMEM; | |
2476 | goto bad_swap; | |
2477 | } | |
6f179af8 | 2478 | for_each_possible_cpu(cpu) { |
ebc2a1a6 | 2479 | struct percpu_cluster *cluster; |
6f179af8 | 2480 | cluster = per_cpu_ptr(p->percpu_cluster, cpu); |
ebc2a1a6 SL |
2481 | cluster_set_null(&cluster->index); |
2482 | } | |
2a8f9449 | 2483 | } |
1da177e4 | 2484 | |
1421ef3c CEB |
2485 | error = swap_cgroup_swapon(p->type, maxpages); |
2486 | if (error) | |
2487 | goto bad_swap; | |
2488 | ||
915d4d7b | 2489 | nr_extents = setup_swap_map_and_extents(p, swap_header, swap_map, |
2a8f9449 | 2490 | cluster_info, maxpages, &span); |
915d4d7b CEB |
2491 | if (unlikely(nr_extents < 0)) { |
2492 | error = nr_extents; | |
1da177e4 LT |
2493 | goto bad_swap; |
2494 | } | |
38b5faf4 | 2495 | /* frontswap enabled? set up bit-per-page map for frontswap */ |
8ea1d2a1 | 2496 | if (IS_ENABLED(CONFIG_FRONTSWAP)) |
7b57976d | 2497 | frontswap_map = vzalloc(BITS_TO_LONGS(maxpages) * sizeof(long)); |
1da177e4 | 2498 | |
2a8f9449 SL |
2499 | if (p->bdev &&(swap_flags & SWAP_FLAG_DISCARD) && swap_discardable(p)) { |
2500 | /* | |
2501 | * When discard is enabled for swap with no particular | |
2502 | * policy flagged, we set all swap discard flags here in | |
2503 | * order to sustain backward compatibility with older | |
2504 | * swapon(8) releases. | |
2505 | */ | |
2506 | p->flags |= (SWP_DISCARDABLE | SWP_AREA_DISCARD | | |
2507 | SWP_PAGE_DISCARD); | |
dcf6b7dd | 2508 | |
2a8f9449 SL |
2509 | /* |
2510 | * By flagging sys_swapon, a sysadmin can tell us to | |
2511 | * either do single-time area discards only, or to just | |
2512 | * perform discards for released swap page-clusters. | |
2513 | * Now it's time to adjust the p->flags accordingly. | |
2514 | */ | |
2515 | if (swap_flags & SWAP_FLAG_DISCARD_ONCE) | |
2516 | p->flags &= ~SWP_PAGE_DISCARD; | |
2517 | else if (swap_flags & SWAP_FLAG_DISCARD_PAGES) | |
2518 | p->flags &= ~SWP_AREA_DISCARD; | |
2519 | ||
2520 | /* issue a swapon-time discard if it's still required */ | |
2521 | if (p->flags & SWP_AREA_DISCARD) { | |
2522 | int err = discard_swap(p); | |
2523 | if (unlikely(err)) | |
2524 | pr_err("swapon: discard_swap(%p): %d\n", | |
2525 | p, err); | |
dcf6b7dd | 2526 | } |
20137a49 | 2527 | } |
6a6ba831 | 2528 | |
fc0abb14 | 2529 | mutex_lock(&swapon_mutex); |
40531542 | 2530 | prio = -1; |
78ecba08 | 2531 | if (swap_flags & SWAP_FLAG_PREFER) |
40531542 | 2532 | prio = |
78ecba08 | 2533 | (swap_flags & SWAP_FLAG_PRIO_MASK) >> SWAP_FLAG_PRIO_SHIFT; |
2a8f9449 | 2534 | enable_swap_info(p, prio, swap_map, cluster_info, frontswap_map); |
c69dbfb8 | 2535 | |
756a025f | 2536 | pr_info("Adding %uk swap on %s. Priority:%d extents:%d across:%lluk %s%s%s%s%s\n", |
91a27b2a | 2537 | p->pages<<(PAGE_SHIFT-10), name->name, p->prio, |
c69dbfb8 CEB |
2538 | nr_extents, (unsigned long long)span<<(PAGE_SHIFT-10), |
2539 | (p->flags & SWP_SOLIDSTATE) ? "SS" : "", | |
38b5faf4 | 2540 | (p->flags & SWP_DISCARDABLE) ? "D" : "", |
dcf6b7dd RA |
2541 | (p->flags & SWP_AREA_DISCARD) ? "s" : "", |
2542 | (p->flags & SWP_PAGE_DISCARD) ? "c" : "", | |
38b5faf4 | 2543 | (frontswap_map) ? "FS" : ""); |
c69dbfb8 | 2544 | |
fc0abb14 | 2545 | mutex_unlock(&swapon_mutex); |
66d7dd51 KS |
2546 | atomic_inc(&proc_poll_event); |
2547 | wake_up_interruptible(&proc_poll_wait); | |
2548 | ||
9b01c350 CEB |
2549 | if (S_ISREG(inode->i_mode)) |
2550 | inode->i_flags |= S_SWAPFILE; | |
1da177e4 LT |
2551 | error = 0; |
2552 | goto out; | |
2553 | bad_swap: | |
ebc2a1a6 SL |
2554 | free_percpu(p->percpu_cluster); |
2555 | p->percpu_cluster = NULL; | |
bd69010b | 2556 | if (inode && S_ISBLK(inode->i_mode) && p->bdev) { |
f2090d2d CEB |
2557 | set_blocksize(p->bdev, p->old_block_size); |
2558 | blkdev_put(p->bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL); | |
1da177e4 | 2559 | } |
4cd3bb10 | 2560 | destroy_swap_extents(p); |
e8e6c2ec | 2561 | swap_cgroup_swapoff(p->type); |
5d337b91 | 2562 | spin_lock(&swap_lock); |
1da177e4 | 2563 | p->swap_file = NULL; |
1da177e4 | 2564 | p->flags = 0; |
5d337b91 | 2565 | spin_unlock(&swap_lock); |
1da177e4 | 2566 | vfree(swap_map); |
2a8f9449 | 2567 | vfree(cluster_info); |
52c50567 | 2568 | if (swap_file) { |
2130781e | 2569 | if (inode && S_ISREG(inode->i_mode)) { |
5955102c | 2570 | inode_unlock(inode); |
2130781e CEB |
2571 | inode = NULL; |
2572 | } | |
1da177e4 | 2573 | filp_close(swap_file, NULL); |
52c50567 | 2574 | } |
1da177e4 LT |
2575 | out: |
2576 | if (page && !IS_ERR(page)) { | |
2577 | kunmap(page); | |
09cbfeaf | 2578 | put_page(page); |
1da177e4 LT |
2579 | } |
2580 | if (name) | |
2581 | putname(name); | |
9b01c350 | 2582 | if (inode && S_ISREG(inode->i_mode)) |
5955102c | 2583 | inode_unlock(inode); |
1da177e4 LT |
2584 | return error; |
2585 | } | |
2586 | ||
2587 | void si_swapinfo(struct sysinfo *val) | |
2588 | { | |
efa90a98 | 2589 | unsigned int type; |
1da177e4 LT |
2590 | unsigned long nr_to_be_unused = 0; |
2591 | ||
5d337b91 | 2592 | spin_lock(&swap_lock); |
efa90a98 HD |
2593 | for (type = 0; type < nr_swapfiles; type++) { |
2594 | struct swap_info_struct *si = swap_info[type]; | |
2595 | ||
2596 | if ((si->flags & SWP_USED) && !(si->flags & SWP_WRITEOK)) | |
2597 | nr_to_be_unused += si->inuse_pages; | |
1da177e4 | 2598 | } |
ec8acf20 | 2599 | val->freeswap = atomic_long_read(&nr_swap_pages) + nr_to_be_unused; |
1da177e4 | 2600 | val->totalswap = total_swap_pages + nr_to_be_unused; |
5d337b91 | 2601 | spin_unlock(&swap_lock); |
1da177e4 LT |
2602 | } |
2603 | ||
2604 | /* | |
2605 | * Verify that a swap entry is valid and increment its swap map count. | |
2606 | * | |
355cfa73 KH |
2607 | * Returns error code in following case. |
2608 | * - success -> 0 | |
2609 | * - swp_entry is invalid -> EINVAL | |
2610 | * - swp_entry is migration entry -> EINVAL | |
2611 | * - swap-cache reference is requested but there is already one. -> EEXIST | |
2612 | * - swap-cache reference is requested but the entry is not used. -> ENOENT | |
570a335b | 2613 | * - swap-mapped reference requested but needs continued swap count. -> ENOMEM |
1da177e4 | 2614 | */ |
8d69aaee | 2615 | static int __swap_duplicate(swp_entry_t entry, unsigned char usage) |
1da177e4 | 2616 | { |
73c34b6a | 2617 | struct swap_info_struct *p; |
1da177e4 | 2618 | unsigned long offset, type; |
8d69aaee HD |
2619 | unsigned char count; |
2620 | unsigned char has_cache; | |
253d553b | 2621 | int err = -EINVAL; |
1da177e4 | 2622 | |
a7420aa5 | 2623 | if (non_swap_entry(entry)) |
253d553b | 2624 | goto out; |
0697212a | 2625 | |
1da177e4 LT |
2626 | type = swp_type(entry); |
2627 | if (type >= nr_swapfiles) | |
2628 | goto bad_file; | |
efa90a98 | 2629 | p = swap_info[type]; |
1da177e4 LT |
2630 | offset = swp_offset(entry); |
2631 | ||
ec8acf20 | 2632 | spin_lock(&p->lock); |
355cfa73 KH |
2633 | if (unlikely(offset >= p->max)) |
2634 | goto unlock_out; | |
2635 | ||
253d553b | 2636 | count = p->swap_map[offset]; |
edfe23da SL |
2637 | |
2638 | /* | |
2639 | * swapin_readahead() doesn't check if a swap entry is valid, so the | |
2640 | * swap entry could be SWAP_MAP_BAD. Check here with lock held. | |
2641 | */ | |
2642 | if (unlikely(swap_count(count) == SWAP_MAP_BAD)) { | |
2643 | err = -ENOENT; | |
2644 | goto unlock_out; | |
2645 | } | |
2646 | ||
253d553b HD |
2647 | has_cache = count & SWAP_HAS_CACHE; |
2648 | count &= ~SWAP_HAS_CACHE; | |
2649 | err = 0; | |
355cfa73 | 2650 | |
253d553b | 2651 | if (usage == SWAP_HAS_CACHE) { |
355cfa73 KH |
2652 | |
2653 | /* set SWAP_HAS_CACHE if there is no cache and entry is used */ | |
253d553b HD |
2654 | if (!has_cache && count) |
2655 | has_cache = SWAP_HAS_CACHE; | |
2656 | else if (has_cache) /* someone else added cache */ | |
2657 | err = -EEXIST; | |
2658 | else /* no users remaining */ | |
2659 | err = -ENOENT; | |
355cfa73 KH |
2660 | |
2661 | } else if (count || has_cache) { | |
253d553b | 2662 | |
570a335b HD |
2663 | if ((count & ~COUNT_CONTINUED) < SWAP_MAP_MAX) |
2664 | count += usage; | |
2665 | else if ((count & ~COUNT_CONTINUED) > SWAP_MAP_MAX) | |
253d553b | 2666 | err = -EINVAL; |
570a335b HD |
2667 | else if (swap_count_continued(p, offset, count)) |
2668 | count = COUNT_CONTINUED; | |
2669 | else | |
2670 | err = -ENOMEM; | |
355cfa73 | 2671 | } else |
253d553b HD |
2672 | err = -ENOENT; /* unused swap entry */ |
2673 | ||
2674 | p->swap_map[offset] = count | has_cache; | |
2675 | ||
355cfa73 | 2676 | unlock_out: |
ec8acf20 | 2677 | spin_unlock(&p->lock); |
1da177e4 | 2678 | out: |
253d553b | 2679 | return err; |
1da177e4 LT |
2680 | |
2681 | bad_file: | |
465c47fd | 2682 | pr_err("swap_dup: %s%08lx\n", Bad_file, entry.val); |
1da177e4 LT |
2683 | goto out; |
2684 | } | |
253d553b | 2685 | |
aaa46865 HD |
2686 | /* |
2687 | * Help swapoff by noting that swap entry belongs to shmem/tmpfs | |
2688 | * (in which case its reference count is never incremented). | |
2689 | */ | |
2690 | void swap_shmem_alloc(swp_entry_t entry) | |
2691 | { | |
2692 | __swap_duplicate(entry, SWAP_MAP_SHMEM); | |
2693 | } | |
2694 | ||
355cfa73 | 2695 | /* |
08259d58 HD |
2696 | * Increase reference count of swap entry by 1. |
2697 | * Returns 0 for success, or -ENOMEM if a swap_count_continuation is required | |
2698 | * but could not be atomically allocated. Returns 0, just as if it succeeded, | |
2699 | * if __swap_duplicate() fails for another reason (-EINVAL or -ENOENT), which | |
2700 | * might occur if a page table entry has got corrupted. | |
355cfa73 | 2701 | */ |
570a335b | 2702 | int swap_duplicate(swp_entry_t entry) |
355cfa73 | 2703 | { |
570a335b HD |
2704 | int err = 0; |
2705 | ||
2706 | while (!err && __swap_duplicate(entry, 1) == -ENOMEM) | |
2707 | err = add_swap_count_continuation(entry, GFP_ATOMIC); | |
2708 | return err; | |
355cfa73 | 2709 | } |
1da177e4 | 2710 | |
cb4b86ba | 2711 | /* |
355cfa73 KH |
2712 | * @entry: swap entry for which we allocate swap cache. |
2713 | * | |
73c34b6a | 2714 | * Called when allocating swap cache for existing swap entry, |
355cfa73 KH |
2715 | * This can return error codes. Returns 0 at success. |
2716 | * -EBUSY means there is a swap cache. | |
2717 | * Note: return code is different from swap_duplicate(). | |
cb4b86ba KH |
2718 | */ |
2719 | int swapcache_prepare(swp_entry_t entry) | |
2720 | { | |
253d553b | 2721 | return __swap_duplicate(entry, SWAP_HAS_CACHE); |
cb4b86ba KH |
2722 | } |
2723 | ||
f981c595 MG |
2724 | struct swap_info_struct *page_swap_info(struct page *page) |
2725 | { | |
2726 | swp_entry_t swap = { .val = page_private(page) }; | |
2727 | BUG_ON(!PageSwapCache(page)); | |
2728 | return swap_info[swp_type(swap)]; | |
2729 | } | |
2730 | ||
2731 | /* | |
2732 | * out-of-line __page_file_ methods to avoid include hell. | |
2733 | */ | |
2734 | struct address_space *__page_file_mapping(struct page *page) | |
2735 | { | |
309381fe | 2736 | VM_BUG_ON_PAGE(!PageSwapCache(page), page); |
f981c595 MG |
2737 | return page_swap_info(page)->swap_file->f_mapping; |
2738 | } | |
2739 | EXPORT_SYMBOL_GPL(__page_file_mapping); | |
2740 | ||
2741 | pgoff_t __page_file_index(struct page *page) | |
2742 | { | |
2743 | swp_entry_t swap = { .val = page_private(page) }; | |
309381fe | 2744 | VM_BUG_ON_PAGE(!PageSwapCache(page), page); |
f981c595 MG |
2745 | return swp_offset(swap); |
2746 | } | |
2747 | EXPORT_SYMBOL_GPL(__page_file_index); | |
2748 | ||
570a335b HD |
2749 | /* |
2750 | * add_swap_count_continuation - called when a swap count is duplicated | |
2751 | * beyond SWAP_MAP_MAX, it allocates a new page and links that to the entry's | |
2752 | * page of the original vmalloc'ed swap_map, to hold the continuation count | |
2753 | * (for that entry and for its neighbouring PAGE_SIZE swap entries). Called | |
2754 | * again when count is duplicated beyond SWAP_MAP_MAX * SWAP_CONT_MAX, etc. | |
2755 | * | |
2756 | * These continuation pages are seldom referenced: the common paths all work | |
2757 | * on the original swap_map, only referring to a continuation page when the | |
2758 | * low "digit" of a count is incremented or decremented through SWAP_MAP_MAX. | |
2759 | * | |
2760 | * add_swap_count_continuation(, GFP_ATOMIC) can be called while holding | |
2761 | * page table locks; if it fails, add_swap_count_continuation(, GFP_KERNEL) | |
2762 | * can be called after dropping locks. | |
2763 | */ | |
2764 | int add_swap_count_continuation(swp_entry_t entry, gfp_t gfp_mask) | |
2765 | { | |
2766 | struct swap_info_struct *si; | |
2767 | struct page *head; | |
2768 | struct page *page; | |
2769 | struct page *list_page; | |
2770 | pgoff_t offset; | |
2771 | unsigned char count; | |
2772 | ||
2773 | /* | |
2774 | * When debugging, it's easier to use __GFP_ZERO here; but it's better | |
2775 | * for latency not to zero a page while GFP_ATOMIC and holding locks. | |
2776 | */ | |
2777 | page = alloc_page(gfp_mask | __GFP_HIGHMEM); | |
2778 | ||
2779 | si = swap_info_get(entry); | |
2780 | if (!si) { | |
2781 | /* | |
2782 | * An acceptable race has occurred since the failing | |
2783 | * __swap_duplicate(): the swap entry has been freed, | |
2784 | * perhaps even the whole swap_map cleared for swapoff. | |
2785 | */ | |
2786 | goto outer; | |
2787 | } | |
2788 | ||
2789 | offset = swp_offset(entry); | |
2790 | count = si->swap_map[offset] & ~SWAP_HAS_CACHE; | |
2791 | ||
2792 | if ((count & ~COUNT_CONTINUED) != SWAP_MAP_MAX) { | |
2793 | /* | |
2794 | * The higher the swap count, the more likely it is that tasks | |
2795 | * will race to add swap count continuation: we need to avoid | |
2796 | * over-provisioning. | |
2797 | */ | |
2798 | goto out; | |
2799 | } | |
2800 | ||
2801 | if (!page) { | |
ec8acf20 | 2802 | spin_unlock(&si->lock); |
570a335b HD |
2803 | return -ENOMEM; |
2804 | } | |
2805 | ||
2806 | /* | |
2807 | * We are fortunate that although vmalloc_to_page uses pte_offset_map, | |
2de1a7e4 SJ |
2808 | * no architecture is using highmem pages for kernel page tables: so it |
2809 | * will not corrupt the GFP_ATOMIC caller's atomic page table kmaps. | |
570a335b HD |
2810 | */ |
2811 | head = vmalloc_to_page(si->swap_map + offset); | |
2812 | offset &= ~PAGE_MASK; | |
2813 | ||
2814 | /* | |
2815 | * Page allocation does not initialize the page's lru field, | |
2816 | * but it does always reset its private field. | |
2817 | */ | |
2818 | if (!page_private(head)) { | |
2819 | BUG_ON(count & COUNT_CONTINUED); | |
2820 | INIT_LIST_HEAD(&head->lru); | |
2821 | set_page_private(head, SWP_CONTINUED); | |
2822 | si->flags |= SWP_CONTINUED; | |
2823 | } | |
2824 | ||
2825 | list_for_each_entry(list_page, &head->lru, lru) { | |
2826 | unsigned char *map; | |
2827 | ||
2828 | /* | |
2829 | * If the previous map said no continuation, but we've found | |
2830 | * a continuation page, free our allocation and use this one. | |
2831 | */ | |
2832 | if (!(count & COUNT_CONTINUED)) | |
2833 | goto out; | |
2834 | ||
9b04c5fe | 2835 | map = kmap_atomic(list_page) + offset; |
570a335b | 2836 | count = *map; |
9b04c5fe | 2837 | kunmap_atomic(map); |
570a335b HD |
2838 | |
2839 | /* | |
2840 | * If this continuation count now has some space in it, | |
2841 | * free our allocation and use this one. | |
2842 | */ | |
2843 | if ((count & ~COUNT_CONTINUED) != SWAP_CONT_MAX) | |
2844 | goto out; | |
2845 | } | |
2846 | ||
2847 | list_add_tail(&page->lru, &head->lru); | |
2848 | page = NULL; /* now it's attached, don't free it */ | |
2849 | out: | |
ec8acf20 | 2850 | spin_unlock(&si->lock); |
570a335b HD |
2851 | outer: |
2852 | if (page) | |
2853 | __free_page(page); | |
2854 | return 0; | |
2855 | } | |
2856 | ||
2857 | /* | |
2858 | * swap_count_continued - when the original swap_map count is incremented | |
2859 | * from SWAP_MAP_MAX, check if there is already a continuation page to carry | |
2860 | * into, carry if so, or else fail until a new continuation page is allocated; | |
2861 | * when the original swap_map count is decremented from 0 with continuation, | |
2862 | * borrow from the continuation and report whether it still holds more. | |
2863 | * Called while __swap_duplicate() or swap_entry_free() holds swap_lock. | |
2864 | */ | |
2865 | static bool swap_count_continued(struct swap_info_struct *si, | |
2866 | pgoff_t offset, unsigned char count) | |
2867 | { | |
2868 | struct page *head; | |
2869 | struct page *page; | |
2870 | unsigned char *map; | |
2871 | ||
2872 | head = vmalloc_to_page(si->swap_map + offset); | |
2873 | if (page_private(head) != SWP_CONTINUED) { | |
2874 | BUG_ON(count & COUNT_CONTINUED); | |
2875 | return false; /* need to add count continuation */ | |
2876 | } | |
2877 | ||
2878 | offset &= ~PAGE_MASK; | |
2879 | page = list_entry(head->lru.next, struct page, lru); | |
9b04c5fe | 2880 | map = kmap_atomic(page) + offset; |
570a335b HD |
2881 | |
2882 | if (count == SWAP_MAP_MAX) /* initial increment from swap_map */ | |
2883 | goto init_map; /* jump over SWAP_CONT_MAX checks */ | |
2884 | ||
2885 | if (count == (SWAP_MAP_MAX | COUNT_CONTINUED)) { /* incrementing */ | |
2886 | /* | |
2887 | * Think of how you add 1 to 999 | |
2888 | */ | |
2889 | while (*map == (SWAP_CONT_MAX | COUNT_CONTINUED)) { | |
9b04c5fe | 2890 | kunmap_atomic(map); |
570a335b HD |
2891 | page = list_entry(page->lru.next, struct page, lru); |
2892 | BUG_ON(page == head); | |
9b04c5fe | 2893 | map = kmap_atomic(page) + offset; |
570a335b HD |
2894 | } |
2895 | if (*map == SWAP_CONT_MAX) { | |
9b04c5fe | 2896 | kunmap_atomic(map); |
570a335b HD |
2897 | page = list_entry(page->lru.next, struct page, lru); |
2898 | if (page == head) | |
2899 | return false; /* add count continuation */ | |
9b04c5fe | 2900 | map = kmap_atomic(page) + offset; |
570a335b HD |
2901 | init_map: *map = 0; /* we didn't zero the page */ |
2902 | } | |
2903 | *map += 1; | |
9b04c5fe | 2904 | kunmap_atomic(map); |
570a335b HD |
2905 | page = list_entry(page->lru.prev, struct page, lru); |
2906 | while (page != head) { | |
9b04c5fe | 2907 | map = kmap_atomic(page) + offset; |
570a335b | 2908 | *map = COUNT_CONTINUED; |
9b04c5fe | 2909 | kunmap_atomic(map); |
570a335b HD |
2910 | page = list_entry(page->lru.prev, struct page, lru); |
2911 | } | |
2912 | return true; /* incremented */ | |
2913 | ||
2914 | } else { /* decrementing */ | |
2915 | /* | |
2916 | * Think of how you subtract 1 from 1000 | |
2917 | */ | |
2918 | BUG_ON(count != COUNT_CONTINUED); | |
2919 | while (*map == COUNT_CONTINUED) { | |
9b04c5fe | 2920 | kunmap_atomic(map); |
570a335b HD |
2921 | page = list_entry(page->lru.next, struct page, lru); |
2922 | BUG_ON(page == head); | |
9b04c5fe | 2923 | map = kmap_atomic(page) + offset; |
570a335b HD |
2924 | } |
2925 | BUG_ON(*map == 0); | |
2926 | *map -= 1; | |
2927 | if (*map == 0) | |
2928 | count = 0; | |
9b04c5fe | 2929 | kunmap_atomic(map); |
570a335b HD |
2930 | page = list_entry(page->lru.prev, struct page, lru); |
2931 | while (page != head) { | |
9b04c5fe | 2932 | map = kmap_atomic(page) + offset; |
570a335b HD |
2933 | *map = SWAP_CONT_MAX | count; |
2934 | count = COUNT_CONTINUED; | |
9b04c5fe | 2935 | kunmap_atomic(map); |
570a335b HD |
2936 | page = list_entry(page->lru.prev, struct page, lru); |
2937 | } | |
2938 | return count == COUNT_CONTINUED; | |
2939 | } | |
2940 | } | |
2941 | ||
2942 | /* | |
2943 | * free_swap_count_continuations - swapoff free all the continuation pages | |
2944 | * appended to the swap_map, after swap_map is quiesced, before vfree'ing it. | |
2945 | */ | |
2946 | static void free_swap_count_continuations(struct swap_info_struct *si) | |
2947 | { | |
2948 | pgoff_t offset; | |
2949 | ||
2950 | for (offset = 0; offset < si->max; offset += PAGE_SIZE) { | |
2951 | struct page *head; | |
2952 | head = vmalloc_to_page(si->swap_map + offset); | |
2953 | if (page_private(head)) { | |
0d576d20 GT |
2954 | struct page *page, *next; |
2955 | ||
2956 | list_for_each_entry_safe(page, next, &head->lru, lru) { | |
2957 | list_del(&page->lru); | |
570a335b HD |
2958 | __free_page(page); |
2959 | } | |
2960 | } | |
2961 | } | |
2962 | } |