]>
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> | |
27a7faa0 | 41 | #include <linux/page_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 SL |
50 | atomic_long_t nr_swap_pages; |
51 | /* protected with swap_lock. reading in vm_swap_full() doesn't need lock */ | |
1da177e4 | 52 | long total_swap_pages; |
78ecba08 | 53 | static int least_priority; |
1da177e4 | 54 | |
1da177e4 LT |
55 | static const char Bad_file[] = "Bad swap file entry "; |
56 | static const char Unused_file[] = "Unused swap file entry "; | |
57 | static const char Bad_offset[] = "Bad swap offset entry "; | |
58 | static const char Unused_offset[] = "Unused swap offset entry "; | |
59 | ||
adfab836 DS |
60 | /* |
61 | * all active swap_info_structs | |
62 | * protected with swap_lock, and ordered by priority. | |
63 | */ | |
64 | LIST_HEAD(swap_list_head); | |
1da177e4 | 65 | |
38b5faf4 | 66 | struct swap_info_struct *swap_info[MAX_SWAPFILES]; |
1da177e4 | 67 | |
fc0abb14 | 68 | static DEFINE_MUTEX(swapon_mutex); |
1da177e4 | 69 | |
66d7dd51 KS |
70 | static DECLARE_WAIT_QUEUE_HEAD(proc_poll_wait); |
71 | /* Activity counter to indicate that a swapon or swapoff has occurred */ | |
72 | static atomic_t proc_poll_event = ATOMIC_INIT(0); | |
73 | ||
8d69aaee | 74 | static inline unsigned char swap_count(unsigned char ent) |
355cfa73 | 75 | { |
570a335b | 76 | return ent & ~SWAP_HAS_CACHE; /* may include SWAP_HAS_CONT flag */ |
355cfa73 KH |
77 | } |
78 | ||
efa90a98 | 79 | /* returns 1 if swap entry is freed */ |
c9e44410 KH |
80 | static int |
81 | __try_to_reclaim_swap(struct swap_info_struct *si, unsigned long offset) | |
82 | { | |
efa90a98 | 83 | swp_entry_t entry = swp_entry(si->type, offset); |
c9e44410 KH |
84 | struct page *page; |
85 | int ret = 0; | |
86 | ||
33806f06 | 87 | page = find_get_page(swap_address_space(entry), entry.val); |
c9e44410 KH |
88 | if (!page) |
89 | return 0; | |
90 | /* | |
91 | * This function is called from scan_swap_map() and it's called | |
92 | * by vmscan.c at reclaiming pages. So, we hold a lock on a page, here. | |
93 | * We have to use trylock for avoiding deadlock. This is a special | |
94 | * case and you should use try_to_free_swap() with explicit lock_page() | |
95 | * in usual operations. | |
96 | */ | |
97 | if (trylock_page(page)) { | |
98 | ret = try_to_free_swap(page); | |
99 | unlock_page(page); | |
100 | } | |
101 | page_cache_release(page); | |
102 | return ret; | |
103 | } | |
355cfa73 | 104 | |
6a6ba831 HD |
105 | /* |
106 | * swapon tell device that all the old swap contents can be discarded, | |
107 | * to allow the swap device to optimize its wear-levelling. | |
108 | */ | |
109 | static int discard_swap(struct swap_info_struct *si) | |
110 | { | |
111 | struct swap_extent *se; | |
9625a5f2 HD |
112 | sector_t start_block; |
113 | sector_t nr_blocks; | |
6a6ba831 HD |
114 | int err = 0; |
115 | ||
9625a5f2 HD |
116 | /* Do not discard the swap header page! */ |
117 | se = &si->first_swap_extent; | |
118 | start_block = (se->start_block + 1) << (PAGE_SHIFT - 9); | |
119 | nr_blocks = ((sector_t)se->nr_pages - 1) << (PAGE_SHIFT - 9); | |
120 | if (nr_blocks) { | |
121 | err = blkdev_issue_discard(si->bdev, start_block, | |
dd3932ed | 122 | nr_blocks, GFP_KERNEL, 0); |
9625a5f2 HD |
123 | if (err) |
124 | return err; | |
125 | cond_resched(); | |
126 | } | |
6a6ba831 | 127 | |
9625a5f2 HD |
128 | list_for_each_entry(se, &si->first_swap_extent.list, list) { |
129 | start_block = se->start_block << (PAGE_SHIFT - 9); | |
130 | nr_blocks = (sector_t)se->nr_pages << (PAGE_SHIFT - 9); | |
6a6ba831 HD |
131 | |
132 | err = blkdev_issue_discard(si->bdev, start_block, | |
dd3932ed | 133 | nr_blocks, GFP_KERNEL, 0); |
6a6ba831 HD |
134 | if (err) |
135 | break; | |
136 | ||
137 | cond_resched(); | |
138 | } | |
139 | return err; /* That will often be -EOPNOTSUPP */ | |
140 | } | |
141 | ||
7992fde7 HD |
142 | /* |
143 | * swap allocation tell device that a cluster of swap can now be discarded, | |
144 | * to allow the swap device to optimize its wear-levelling. | |
145 | */ | |
146 | static void discard_swap_cluster(struct swap_info_struct *si, | |
147 | pgoff_t start_page, pgoff_t nr_pages) | |
148 | { | |
149 | struct swap_extent *se = si->curr_swap_extent; | |
150 | int found_extent = 0; | |
151 | ||
152 | while (nr_pages) { | |
153 | struct list_head *lh; | |
154 | ||
155 | if (se->start_page <= start_page && | |
156 | start_page < se->start_page + se->nr_pages) { | |
157 | pgoff_t offset = start_page - se->start_page; | |
158 | sector_t start_block = se->start_block + offset; | |
858a2990 | 159 | sector_t nr_blocks = se->nr_pages - offset; |
7992fde7 HD |
160 | |
161 | if (nr_blocks > nr_pages) | |
162 | nr_blocks = nr_pages; | |
163 | start_page += nr_blocks; | |
164 | nr_pages -= nr_blocks; | |
165 | ||
166 | if (!found_extent++) | |
167 | si->curr_swap_extent = se; | |
168 | ||
169 | start_block <<= PAGE_SHIFT - 9; | |
170 | nr_blocks <<= PAGE_SHIFT - 9; | |
171 | if (blkdev_issue_discard(si->bdev, start_block, | |
dd3932ed | 172 | nr_blocks, GFP_NOIO, 0)) |
7992fde7 HD |
173 | break; |
174 | } | |
175 | ||
176 | lh = se->list.next; | |
7992fde7 HD |
177 | se = list_entry(lh, struct swap_extent, list); |
178 | } | |
179 | } | |
180 | ||
048c27fd HD |
181 | #define SWAPFILE_CLUSTER 256 |
182 | #define LATENCY_LIMIT 256 | |
183 | ||
2a8f9449 SL |
184 | static inline void cluster_set_flag(struct swap_cluster_info *info, |
185 | unsigned int flag) | |
186 | { | |
187 | info->flags = flag; | |
188 | } | |
189 | ||
190 | static inline unsigned int cluster_count(struct swap_cluster_info *info) | |
191 | { | |
192 | return info->data; | |
193 | } | |
194 | ||
195 | static inline void cluster_set_count(struct swap_cluster_info *info, | |
196 | unsigned int c) | |
197 | { | |
198 | info->data = c; | |
199 | } | |
200 | ||
201 | static inline void cluster_set_count_flag(struct swap_cluster_info *info, | |
202 | unsigned int c, unsigned int f) | |
203 | { | |
204 | info->flags = f; | |
205 | info->data = c; | |
206 | } | |
207 | ||
208 | static inline unsigned int cluster_next(struct swap_cluster_info *info) | |
209 | { | |
210 | return info->data; | |
211 | } | |
212 | ||
213 | static inline void cluster_set_next(struct swap_cluster_info *info, | |
214 | unsigned int n) | |
215 | { | |
216 | info->data = n; | |
217 | } | |
218 | ||
219 | static inline void cluster_set_next_flag(struct swap_cluster_info *info, | |
220 | unsigned int n, unsigned int f) | |
221 | { | |
222 | info->flags = f; | |
223 | info->data = n; | |
224 | } | |
225 | ||
226 | static inline bool cluster_is_free(struct swap_cluster_info *info) | |
227 | { | |
228 | return info->flags & CLUSTER_FLAG_FREE; | |
229 | } | |
230 | ||
231 | static inline bool cluster_is_null(struct swap_cluster_info *info) | |
232 | { | |
233 | return info->flags & CLUSTER_FLAG_NEXT_NULL; | |
234 | } | |
235 | ||
236 | static inline void cluster_set_null(struct swap_cluster_info *info) | |
237 | { | |
238 | info->flags = CLUSTER_FLAG_NEXT_NULL; | |
239 | info->data = 0; | |
240 | } | |
241 | ||
815c2c54 SL |
242 | /* Add a cluster to discard list and schedule it to do discard */ |
243 | static void swap_cluster_schedule_discard(struct swap_info_struct *si, | |
244 | unsigned int idx) | |
245 | { | |
246 | /* | |
247 | * If scan_swap_map() can't find a free cluster, it will check | |
248 | * si->swap_map directly. To make sure the discarding cluster isn't | |
249 | * taken by scan_swap_map(), mark the swap entries bad (occupied). It | |
250 | * will be cleared after discard | |
251 | */ | |
252 | memset(si->swap_map + idx * SWAPFILE_CLUSTER, | |
253 | SWAP_MAP_BAD, SWAPFILE_CLUSTER); | |
254 | ||
255 | if (cluster_is_null(&si->discard_cluster_head)) { | |
256 | cluster_set_next_flag(&si->discard_cluster_head, | |
257 | idx, 0); | |
258 | cluster_set_next_flag(&si->discard_cluster_tail, | |
259 | idx, 0); | |
260 | } else { | |
261 | unsigned int tail = cluster_next(&si->discard_cluster_tail); | |
262 | cluster_set_next(&si->cluster_info[tail], idx); | |
263 | cluster_set_next_flag(&si->discard_cluster_tail, | |
264 | idx, 0); | |
265 | } | |
266 | ||
267 | schedule_work(&si->discard_work); | |
268 | } | |
269 | ||
270 | /* | |
271 | * Doing discard actually. After a cluster discard is finished, the cluster | |
272 | * will be added to free cluster list. caller should hold si->lock. | |
273 | */ | |
274 | static void swap_do_scheduled_discard(struct swap_info_struct *si) | |
275 | { | |
276 | struct swap_cluster_info *info; | |
277 | unsigned int idx; | |
278 | ||
279 | info = si->cluster_info; | |
280 | ||
281 | while (!cluster_is_null(&si->discard_cluster_head)) { | |
282 | idx = cluster_next(&si->discard_cluster_head); | |
283 | ||
284 | cluster_set_next_flag(&si->discard_cluster_head, | |
285 | cluster_next(&info[idx]), 0); | |
286 | if (cluster_next(&si->discard_cluster_tail) == idx) { | |
287 | cluster_set_null(&si->discard_cluster_head); | |
288 | cluster_set_null(&si->discard_cluster_tail); | |
289 | } | |
290 | spin_unlock(&si->lock); | |
291 | ||
292 | discard_swap_cluster(si, idx * SWAPFILE_CLUSTER, | |
293 | SWAPFILE_CLUSTER); | |
294 | ||
295 | spin_lock(&si->lock); | |
296 | cluster_set_flag(&info[idx], CLUSTER_FLAG_FREE); | |
297 | if (cluster_is_null(&si->free_cluster_head)) { | |
298 | cluster_set_next_flag(&si->free_cluster_head, | |
299 | idx, 0); | |
300 | cluster_set_next_flag(&si->free_cluster_tail, | |
301 | idx, 0); | |
302 | } else { | |
303 | unsigned int tail; | |
304 | ||
305 | tail = cluster_next(&si->free_cluster_tail); | |
306 | cluster_set_next(&info[tail], idx); | |
307 | cluster_set_next_flag(&si->free_cluster_tail, | |
308 | idx, 0); | |
309 | } | |
310 | memset(si->swap_map + idx * SWAPFILE_CLUSTER, | |
311 | 0, SWAPFILE_CLUSTER); | |
312 | } | |
313 | } | |
314 | ||
315 | static void swap_discard_work(struct work_struct *work) | |
316 | { | |
317 | struct swap_info_struct *si; | |
318 | ||
319 | si = container_of(work, struct swap_info_struct, discard_work); | |
320 | ||
321 | spin_lock(&si->lock); | |
322 | swap_do_scheduled_discard(si); | |
323 | spin_unlock(&si->lock); | |
324 | } | |
325 | ||
2a8f9449 SL |
326 | /* |
327 | * The cluster corresponding to page_nr will be used. The cluster will be | |
328 | * removed from free cluster list and its usage counter will be increased. | |
329 | */ | |
330 | static void inc_cluster_info_page(struct swap_info_struct *p, | |
331 | struct swap_cluster_info *cluster_info, unsigned long page_nr) | |
332 | { | |
333 | unsigned long idx = page_nr / SWAPFILE_CLUSTER; | |
334 | ||
335 | if (!cluster_info) | |
336 | return; | |
337 | if (cluster_is_free(&cluster_info[idx])) { | |
338 | VM_BUG_ON(cluster_next(&p->free_cluster_head) != idx); | |
339 | cluster_set_next_flag(&p->free_cluster_head, | |
340 | cluster_next(&cluster_info[idx]), 0); | |
341 | if (cluster_next(&p->free_cluster_tail) == idx) { | |
342 | cluster_set_null(&p->free_cluster_tail); | |
343 | cluster_set_null(&p->free_cluster_head); | |
344 | } | |
345 | cluster_set_count_flag(&cluster_info[idx], 0, 0); | |
346 | } | |
347 | ||
348 | VM_BUG_ON(cluster_count(&cluster_info[idx]) >= SWAPFILE_CLUSTER); | |
349 | cluster_set_count(&cluster_info[idx], | |
350 | cluster_count(&cluster_info[idx]) + 1); | |
351 | } | |
352 | ||
353 | /* | |
354 | * The cluster corresponding to page_nr decreases one usage. If the usage | |
355 | * counter becomes 0, which means no page in the cluster is in using, we can | |
356 | * optionally discard the cluster and add it to free cluster list. | |
357 | */ | |
358 | static void dec_cluster_info_page(struct swap_info_struct *p, | |
359 | struct swap_cluster_info *cluster_info, unsigned long page_nr) | |
360 | { | |
361 | unsigned long idx = page_nr / SWAPFILE_CLUSTER; | |
362 | ||
363 | if (!cluster_info) | |
364 | return; | |
365 | ||
366 | VM_BUG_ON(cluster_count(&cluster_info[idx]) == 0); | |
367 | cluster_set_count(&cluster_info[idx], | |
368 | cluster_count(&cluster_info[idx]) - 1); | |
369 | ||
370 | if (cluster_count(&cluster_info[idx]) == 0) { | |
815c2c54 SL |
371 | /* |
372 | * If the swap is discardable, prepare discard the cluster | |
373 | * instead of free it immediately. The cluster will be freed | |
374 | * after discard. | |
375 | */ | |
edfe23da SL |
376 | if ((p->flags & (SWP_WRITEOK | SWP_PAGE_DISCARD)) == |
377 | (SWP_WRITEOK | SWP_PAGE_DISCARD)) { | |
815c2c54 SL |
378 | swap_cluster_schedule_discard(p, idx); |
379 | return; | |
380 | } | |
381 | ||
2a8f9449 SL |
382 | cluster_set_flag(&cluster_info[idx], CLUSTER_FLAG_FREE); |
383 | if (cluster_is_null(&p->free_cluster_head)) { | |
384 | cluster_set_next_flag(&p->free_cluster_head, idx, 0); | |
385 | cluster_set_next_flag(&p->free_cluster_tail, idx, 0); | |
386 | } else { | |
387 | unsigned int tail = cluster_next(&p->free_cluster_tail); | |
388 | cluster_set_next(&cluster_info[tail], idx); | |
389 | cluster_set_next_flag(&p->free_cluster_tail, idx, 0); | |
390 | } | |
391 | } | |
392 | } | |
393 | ||
394 | /* | |
395 | * It's possible scan_swap_map() uses a free cluster in the middle of free | |
396 | * cluster list. Avoiding such abuse to avoid list corruption. | |
397 | */ | |
ebc2a1a6 SL |
398 | static bool |
399 | scan_swap_map_ssd_cluster_conflict(struct swap_info_struct *si, | |
2a8f9449 SL |
400 | unsigned long offset) |
401 | { | |
ebc2a1a6 SL |
402 | struct percpu_cluster *percpu_cluster; |
403 | bool conflict; | |
404 | ||
2a8f9449 | 405 | offset /= SWAPFILE_CLUSTER; |
ebc2a1a6 | 406 | conflict = !cluster_is_null(&si->free_cluster_head) && |
2a8f9449 SL |
407 | offset != cluster_next(&si->free_cluster_head) && |
408 | cluster_is_free(&si->cluster_info[offset]); | |
ebc2a1a6 SL |
409 | |
410 | if (!conflict) | |
411 | return false; | |
412 | ||
413 | percpu_cluster = this_cpu_ptr(si->percpu_cluster); | |
414 | cluster_set_null(&percpu_cluster->index); | |
415 | return true; | |
416 | } | |
417 | ||
418 | /* | |
419 | * Try to get a swap entry from current cpu's swap entry pool (a cluster). This | |
420 | * might involve allocating a new cluster for current CPU too. | |
421 | */ | |
422 | static void scan_swap_map_try_ssd_cluster(struct swap_info_struct *si, | |
423 | unsigned long *offset, unsigned long *scan_base) | |
424 | { | |
425 | struct percpu_cluster *cluster; | |
426 | bool found_free; | |
427 | unsigned long tmp; | |
428 | ||
429 | new_cluster: | |
430 | cluster = this_cpu_ptr(si->percpu_cluster); | |
431 | if (cluster_is_null(&cluster->index)) { | |
432 | if (!cluster_is_null(&si->free_cluster_head)) { | |
433 | cluster->index = si->free_cluster_head; | |
434 | cluster->next = cluster_next(&cluster->index) * | |
435 | SWAPFILE_CLUSTER; | |
436 | } else if (!cluster_is_null(&si->discard_cluster_head)) { | |
437 | /* | |
438 | * we don't have free cluster but have some clusters in | |
439 | * discarding, do discard now and reclaim them | |
440 | */ | |
441 | swap_do_scheduled_discard(si); | |
442 | *scan_base = *offset = si->cluster_next; | |
443 | goto new_cluster; | |
444 | } else | |
445 | return; | |
446 | } | |
447 | ||
448 | found_free = false; | |
449 | ||
450 | /* | |
451 | * Other CPUs can use our cluster if they can't find a free cluster, | |
452 | * check if there is still free entry in the cluster | |
453 | */ | |
454 | tmp = cluster->next; | |
455 | while (tmp < si->max && tmp < (cluster_next(&cluster->index) + 1) * | |
456 | SWAPFILE_CLUSTER) { | |
457 | if (!si->swap_map[tmp]) { | |
458 | found_free = true; | |
459 | break; | |
460 | } | |
461 | tmp++; | |
462 | } | |
463 | if (!found_free) { | |
464 | cluster_set_null(&cluster->index); | |
465 | goto new_cluster; | |
466 | } | |
467 | cluster->next = tmp + 1; | |
468 | *offset = tmp; | |
469 | *scan_base = tmp; | |
2a8f9449 SL |
470 | } |
471 | ||
24b8ff7c CEB |
472 | static unsigned long scan_swap_map(struct swap_info_struct *si, |
473 | unsigned char usage) | |
1da177e4 | 474 | { |
ebebbbe9 | 475 | unsigned long offset; |
c60aa176 | 476 | unsigned long scan_base; |
7992fde7 | 477 | unsigned long last_in_cluster = 0; |
048c27fd | 478 | int latency_ration = LATENCY_LIMIT; |
7dfad418 | 479 | |
886bb7e9 | 480 | /* |
7dfad418 HD |
481 | * We try to cluster swap pages by allocating them sequentially |
482 | * in swap. Once we've allocated SWAPFILE_CLUSTER pages this | |
483 | * way, however, we resort to first-free allocation, starting | |
484 | * a new cluster. This prevents us from scattering swap pages | |
485 | * all over the entire swap partition, so that we reduce | |
486 | * overall disk seek times between swap pages. -- sct | |
487 | * But we do now try to find an empty cluster. -Andrea | |
c60aa176 | 488 | * And we let swap pages go all over an SSD partition. Hugh |
7dfad418 HD |
489 | */ |
490 | ||
52b7efdb | 491 | si->flags += SWP_SCANNING; |
c60aa176 | 492 | scan_base = offset = si->cluster_next; |
ebebbbe9 | 493 | |
ebc2a1a6 SL |
494 | /* SSD algorithm */ |
495 | if (si->cluster_info) { | |
496 | scan_swap_map_try_ssd_cluster(si, &offset, &scan_base); | |
497 | goto checks; | |
498 | } | |
499 | ||
ebebbbe9 HD |
500 | if (unlikely(!si->cluster_nr--)) { |
501 | if (si->pages - si->inuse_pages < SWAPFILE_CLUSTER) { | |
502 | si->cluster_nr = SWAPFILE_CLUSTER - 1; | |
503 | goto checks; | |
504 | } | |
2a8f9449 | 505 | |
ec8acf20 | 506 | spin_unlock(&si->lock); |
7dfad418 | 507 | |
c60aa176 HD |
508 | /* |
509 | * If seek is expensive, start searching for new cluster from | |
510 | * start of partition, to minimize the span of allocated swap. | |
511 | * But if seek is cheap, search from our current position, so | |
512 | * that swap is allocated from all over the partition: if the | |
513 | * Flash Translation Layer only remaps within limited zones, | |
514 | * we don't want to wear out the first zone too quickly. | |
515 | */ | |
516 | if (!(si->flags & SWP_SOLIDSTATE)) | |
517 | scan_base = offset = si->lowest_bit; | |
7dfad418 HD |
518 | last_in_cluster = offset + SWAPFILE_CLUSTER - 1; |
519 | ||
520 | /* Locate the first empty (unaligned) cluster */ | |
521 | for (; last_in_cluster <= si->highest_bit; offset++) { | |
1da177e4 | 522 | if (si->swap_map[offset]) |
7dfad418 HD |
523 | last_in_cluster = offset + SWAPFILE_CLUSTER; |
524 | else if (offset == last_in_cluster) { | |
ec8acf20 | 525 | spin_lock(&si->lock); |
ebebbbe9 HD |
526 | offset -= SWAPFILE_CLUSTER - 1; |
527 | si->cluster_next = offset; | |
528 | si->cluster_nr = SWAPFILE_CLUSTER - 1; | |
529 | goto checks; | |
1da177e4 | 530 | } |
048c27fd HD |
531 | if (unlikely(--latency_ration < 0)) { |
532 | cond_resched(); | |
533 | latency_ration = LATENCY_LIMIT; | |
534 | } | |
7dfad418 | 535 | } |
ebebbbe9 HD |
536 | |
537 | offset = si->lowest_bit; | |
c60aa176 HD |
538 | last_in_cluster = offset + SWAPFILE_CLUSTER - 1; |
539 | ||
540 | /* Locate the first empty (unaligned) cluster */ | |
541 | for (; last_in_cluster < scan_base; offset++) { | |
542 | if (si->swap_map[offset]) | |
543 | last_in_cluster = offset + SWAPFILE_CLUSTER; | |
544 | else if (offset == last_in_cluster) { | |
ec8acf20 | 545 | spin_lock(&si->lock); |
c60aa176 HD |
546 | offset -= SWAPFILE_CLUSTER - 1; |
547 | si->cluster_next = offset; | |
548 | si->cluster_nr = SWAPFILE_CLUSTER - 1; | |
c60aa176 HD |
549 | goto checks; |
550 | } | |
551 | if (unlikely(--latency_ration < 0)) { | |
552 | cond_resched(); | |
553 | latency_ration = LATENCY_LIMIT; | |
554 | } | |
555 | } | |
556 | ||
557 | offset = scan_base; | |
ec8acf20 | 558 | spin_lock(&si->lock); |
ebebbbe9 | 559 | si->cluster_nr = SWAPFILE_CLUSTER - 1; |
1da177e4 | 560 | } |
7dfad418 | 561 | |
ebebbbe9 | 562 | checks: |
ebc2a1a6 SL |
563 | if (si->cluster_info) { |
564 | while (scan_swap_map_ssd_cluster_conflict(si, offset)) | |
565 | scan_swap_map_try_ssd_cluster(si, &offset, &scan_base); | |
566 | } | |
ebebbbe9 | 567 | if (!(si->flags & SWP_WRITEOK)) |
52b7efdb | 568 | goto no_page; |
7dfad418 HD |
569 | if (!si->highest_bit) |
570 | goto no_page; | |
ebebbbe9 | 571 | if (offset > si->highest_bit) |
c60aa176 | 572 | scan_base = offset = si->lowest_bit; |
c9e44410 | 573 | |
b73d7fce HD |
574 | /* reuse swap entry of cache-only swap if not busy. */ |
575 | if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) { | |
c9e44410 | 576 | int swap_was_freed; |
ec8acf20 | 577 | spin_unlock(&si->lock); |
c9e44410 | 578 | swap_was_freed = __try_to_reclaim_swap(si, offset); |
ec8acf20 | 579 | spin_lock(&si->lock); |
c9e44410 KH |
580 | /* entry was freed successfully, try to use this again */ |
581 | if (swap_was_freed) | |
582 | goto checks; | |
583 | goto scan; /* check next one */ | |
584 | } | |
585 | ||
ebebbbe9 HD |
586 | if (si->swap_map[offset]) |
587 | goto scan; | |
588 | ||
589 | if (offset == si->lowest_bit) | |
590 | si->lowest_bit++; | |
591 | if (offset == si->highest_bit) | |
592 | si->highest_bit--; | |
593 | si->inuse_pages++; | |
594 | if (si->inuse_pages == si->pages) { | |
595 | si->lowest_bit = si->max; | |
596 | si->highest_bit = 0; | |
1da177e4 | 597 | } |
253d553b | 598 | si->swap_map[offset] = usage; |
2a8f9449 | 599 | inc_cluster_info_page(si, si->cluster_info, offset); |
ebebbbe9 HD |
600 | si->cluster_next = offset + 1; |
601 | si->flags -= SWP_SCANNING; | |
7992fde7 | 602 | |
ebebbbe9 | 603 | return offset; |
7dfad418 | 604 | |
ebebbbe9 | 605 | scan: |
ec8acf20 | 606 | spin_unlock(&si->lock); |
7dfad418 | 607 | while (++offset <= si->highest_bit) { |
52b7efdb | 608 | if (!si->swap_map[offset]) { |
ec8acf20 | 609 | spin_lock(&si->lock); |
52b7efdb HD |
610 | goto checks; |
611 | } | |
c9e44410 | 612 | if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) { |
ec8acf20 | 613 | spin_lock(&si->lock); |
c9e44410 KH |
614 | goto checks; |
615 | } | |
048c27fd HD |
616 | if (unlikely(--latency_ration < 0)) { |
617 | cond_resched(); | |
618 | latency_ration = LATENCY_LIMIT; | |
619 | } | |
7dfad418 | 620 | } |
c60aa176 | 621 | offset = si->lowest_bit; |
a5998061 | 622 | while (offset < scan_base) { |
c60aa176 | 623 | if (!si->swap_map[offset]) { |
ec8acf20 | 624 | spin_lock(&si->lock); |
c60aa176 HD |
625 | goto checks; |
626 | } | |
c9e44410 | 627 | if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) { |
ec8acf20 | 628 | spin_lock(&si->lock); |
c9e44410 KH |
629 | goto checks; |
630 | } | |
c60aa176 HD |
631 | if (unlikely(--latency_ration < 0)) { |
632 | cond_resched(); | |
633 | latency_ration = LATENCY_LIMIT; | |
634 | } | |
a5998061 | 635 | offset++; |
c60aa176 | 636 | } |
ec8acf20 | 637 | spin_lock(&si->lock); |
7dfad418 HD |
638 | |
639 | no_page: | |
52b7efdb | 640 | si->flags -= SWP_SCANNING; |
1da177e4 LT |
641 | return 0; |
642 | } | |
643 | ||
644 | swp_entry_t get_swap_page(void) | |
645 | { | |
adfab836 | 646 | struct swap_info_struct *si, *next; |
fb4f88dc | 647 | pgoff_t offset; |
adfab836 | 648 | struct list_head *tmp; |
1da177e4 | 649 | |
5d337b91 | 650 | spin_lock(&swap_lock); |
ec8acf20 | 651 | if (atomic_long_read(&nr_swap_pages) <= 0) |
fb4f88dc | 652 | goto noswap; |
ec8acf20 | 653 | atomic_long_dec(&nr_swap_pages); |
fb4f88dc | 654 | |
adfab836 DS |
655 | list_for_each(tmp, &swap_list_head) { |
656 | si = list_entry(tmp, typeof(*si), list); | |
ec8acf20 | 657 | spin_lock(&si->lock); |
adfab836 | 658 | if (!si->highest_bit || !(si->flags & SWP_WRITEOK)) { |
ec8acf20 | 659 | spin_unlock(&si->lock); |
fb4f88dc | 660 | continue; |
ec8acf20 | 661 | } |
fb4f88dc | 662 | |
adfab836 DS |
663 | /* |
664 | * rotate the current swap_info that we're going to use | |
665 | * to after any other swap_info that have the same prio, | |
666 | * so that all equal-priority swap_info get used equally | |
667 | */ | |
668 | next = si; | |
669 | list_for_each_entry_continue(next, &swap_list_head, list) { | |
670 | if (si->prio != next->prio) | |
671 | break; | |
672 | list_rotate_left(&si->list); | |
673 | next = si; | |
674 | } | |
ec8acf20 SL |
675 | |
676 | spin_unlock(&swap_lock); | |
355cfa73 | 677 | /* This is called for allocating swap entry for cache */ |
253d553b | 678 | offset = scan_swap_map(si, SWAP_HAS_CACHE); |
ec8acf20 SL |
679 | spin_unlock(&si->lock); |
680 | if (offset) | |
adfab836 | 681 | return swp_entry(si->type, offset); |
ec8acf20 | 682 | spin_lock(&swap_lock); |
adfab836 DS |
683 | /* |
684 | * if we got here, it's likely that si was almost full before, | |
685 | * and since scan_swap_map() can drop the si->lock, multiple | |
686 | * callers probably all tried to get a page from the same si | |
687 | * and it filled up before we could get one. So we need to | |
688 | * try again. Since we dropped the swap_lock, there may now | |
689 | * be non-full higher priority swap_infos, and this si may have | |
690 | * even been removed from the list (although very unlikely). | |
691 | * Let's start over. | |
692 | */ | |
693 | tmp = &swap_list_head; | |
1da177e4 | 694 | } |
fb4f88dc | 695 | |
ec8acf20 | 696 | atomic_long_inc(&nr_swap_pages); |
fb4f88dc | 697 | noswap: |
5d337b91 | 698 | spin_unlock(&swap_lock); |
fb4f88dc | 699 | return (swp_entry_t) {0}; |
1da177e4 LT |
700 | } |
701 | ||
2de1a7e4 | 702 | /* The only caller of this function is now suspend routine */ |
910321ea HD |
703 | swp_entry_t get_swap_page_of_type(int type) |
704 | { | |
705 | struct swap_info_struct *si; | |
706 | pgoff_t offset; | |
707 | ||
910321ea | 708 | si = swap_info[type]; |
ec8acf20 | 709 | spin_lock(&si->lock); |
910321ea | 710 | if (si && (si->flags & SWP_WRITEOK)) { |
ec8acf20 | 711 | atomic_long_dec(&nr_swap_pages); |
910321ea HD |
712 | /* This is called for allocating swap entry, not cache */ |
713 | offset = scan_swap_map(si, 1); | |
714 | if (offset) { | |
ec8acf20 | 715 | spin_unlock(&si->lock); |
910321ea HD |
716 | return swp_entry(type, offset); |
717 | } | |
ec8acf20 | 718 | atomic_long_inc(&nr_swap_pages); |
910321ea | 719 | } |
ec8acf20 | 720 | spin_unlock(&si->lock); |
910321ea HD |
721 | return (swp_entry_t) {0}; |
722 | } | |
723 | ||
73c34b6a | 724 | static struct swap_info_struct *swap_info_get(swp_entry_t entry) |
1da177e4 | 725 | { |
73c34b6a | 726 | struct swap_info_struct *p; |
1da177e4 LT |
727 | unsigned long offset, type; |
728 | ||
729 | if (!entry.val) | |
730 | goto out; | |
731 | type = swp_type(entry); | |
732 | if (type >= nr_swapfiles) | |
733 | goto bad_nofile; | |
efa90a98 | 734 | p = swap_info[type]; |
1da177e4 LT |
735 | if (!(p->flags & SWP_USED)) |
736 | goto bad_device; | |
737 | offset = swp_offset(entry); | |
738 | if (offset >= p->max) | |
739 | goto bad_offset; | |
740 | if (!p->swap_map[offset]) | |
741 | goto bad_free; | |
ec8acf20 | 742 | spin_lock(&p->lock); |
1da177e4 LT |
743 | return p; |
744 | ||
745 | bad_free: | |
465c47fd | 746 | pr_err("swap_free: %s%08lx\n", Unused_offset, entry.val); |
1da177e4 LT |
747 | goto out; |
748 | bad_offset: | |
465c47fd | 749 | pr_err("swap_free: %s%08lx\n", Bad_offset, entry.val); |
1da177e4 LT |
750 | goto out; |
751 | bad_device: | |
465c47fd | 752 | pr_err("swap_free: %s%08lx\n", Unused_file, entry.val); |
1da177e4 LT |
753 | goto out; |
754 | bad_nofile: | |
465c47fd | 755 | pr_err("swap_free: %s%08lx\n", Bad_file, entry.val); |
1da177e4 LT |
756 | out: |
757 | return NULL; | |
886bb7e9 | 758 | } |
1da177e4 | 759 | |
8d69aaee HD |
760 | static unsigned char swap_entry_free(struct swap_info_struct *p, |
761 | swp_entry_t entry, unsigned char usage) | |
1da177e4 | 762 | { |
253d553b | 763 | unsigned long offset = swp_offset(entry); |
8d69aaee HD |
764 | unsigned char count; |
765 | unsigned char has_cache; | |
355cfa73 | 766 | |
253d553b HD |
767 | count = p->swap_map[offset]; |
768 | has_cache = count & SWAP_HAS_CACHE; | |
769 | count &= ~SWAP_HAS_CACHE; | |
355cfa73 | 770 | |
253d553b | 771 | if (usage == SWAP_HAS_CACHE) { |
355cfa73 | 772 | VM_BUG_ON(!has_cache); |
253d553b | 773 | has_cache = 0; |
aaa46865 HD |
774 | } else if (count == SWAP_MAP_SHMEM) { |
775 | /* | |
776 | * Or we could insist on shmem.c using a special | |
777 | * swap_shmem_free() and free_shmem_swap_and_cache()... | |
778 | */ | |
779 | count = 0; | |
570a335b HD |
780 | } else if ((count & ~COUNT_CONTINUED) <= SWAP_MAP_MAX) { |
781 | if (count == COUNT_CONTINUED) { | |
782 | if (swap_count_continued(p, offset, count)) | |
783 | count = SWAP_MAP_MAX | COUNT_CONTINUED; | |
784 | else | |
785 | count = SWAP_MAP_MAX; | |
786 | } else | |
787 | count--; | |
788 | } | |
253d553b HD |
789 | |
790 | if (!count) | |
791 | mem_cgroup_uncharge_swap(entry); | |
792 | ||
793 | usage = count | has_cache; | |
794 | p->swap_map[offset] = usage; | |
355cfa73 | 795 | |
355cfa73 | 796 | /* free if no reference */ |
253d553b | 797 | if (!usage) { |
2a8f9449 | 798 | dec_cluster_info_page(p, p->cluster_info, offset); |
355cfa73 KH |
799 | if (offset < p->lowest_bit) |
800 | p->lowest_bit = offset; | |
801 | if (offset > p->highest_bit) | |
802 | p->highest_bit = offset; | |
ec8acf20 | 803 | atomic_long_inc(&nr_swap_pages); |
355cfa73 | 804 | p->inuse_pages--; |
38b5faf4 | 805 | frontswap_invalidate_page(p->type, offset); |
73744923 MG |
806 | if (p->flags & SWP_BLKDEV) { |
807 | struct gendisk *disk = p->bdev->bd_disk; | |
808 | if (disk->fops->swap_slot_free_notify) | |
809 | disk->fops->swap_slot_free_notify(p->bdev, | |
810 | offset); | |
811 | } | |
1da177e4 | 812 | } |
253d553b HD |
813 | |
814 | return usage; | |
1da177e4 LT |
815 | } |
816 | ||
817 | /* | |
2de1a7e4 | 818 | * Caller has made sure that the swap device corresponding to entry |
1da177e4 LT |
819 | * is still around or has not been recycled. |
820 | */ | |
821 | void swap_free(swp_entry_t entry) | |
822 | { | |
73c34b6a | 823 | struct swap_info_struct *p; |
1da177e4 LT |
824 | |
825 | p = swap_info_get(entry); | |
826 | if (p) { | |
253d553b | 827 | swap_entry_free(p, entry, 1); |
ec8acf20 | 828 | spin_unlock(&p->lock); |
1da177e4 LT |
829 | } |
830 | } | |
831 | ||
cb4b86ba KH |
832 | /* |
833 | * Called after dropping swapcache to decrease refcnt to swap entries. | |
834 | */ | |
835 | void swapcache_free(swp_entry_t entry, struct page *page) | |
836 | { | |
355cfa73 | 837 | struct swap_info_struct *p; |
8d69aaee | 838 | unsigned char count; |
355cfa73 | 839 | |
355cfa73 KH |
840 | p = swap_info_get(entry); |
841 | if (p) { | |
253d553b HD |
842 | count = swap_entry_free(p, entry, SWAP_HAS_CACHE); |
843 | if (page) | |
844 | mem_cgroup_uncharge_swapcache(page, entry, count != 0); | |
ec8acf20 | 845 | spin_unlock(&p->lock); |
355cfa73 | 846 | } |
cb4b86ba KH |
847 | } |
848 | ||
1da177e4 | 849 | /* |
c475a8ab | 850 | * How many references to page are currently swapped out? |
570a335b HD |
851 | * This does not give an exact answer when swap count is continued, |
852 | * but does include the high COUNT_CONTINUED flag to allow for that. | |
1da177e4 | 853 | */ |
bde05d1c | 854 | int page_swapcount(struct page *page) |
1da177e4 | 855 | { |
c475a8ab HD |
856 | int count = 0; |
857 | struct swap_info_struct *p; | |
1da177e4 LT |
858 | swp_entry_t entry; |
859 | ||
4c21e2f2 | 860 | entry.val = page_private(page); |
1da177e4 LT |
861 | p = swap_info_get(entry); |
862 | if (p) { | |
355cfa73 | 863 | count = swap_count(p->swap_map[swp_offset(entry)]); |
ec8acf20 | 864 | spin_unlock(&p->lock); |
1da177e4 | 865 | } |
c475a8ab | 866 | return count; |
1da177e4 LT |
867 | } |
868 | ||
869 | /* | |
7b1fe597 HD |
870 | * We can write to an anon page without COW if there are no other references |
871 | * to it. And as a side-effect, free up its swap: because the old content | |
872 | * on disk will never be read, and seeking back there to write new content | |
873 | * later would only waste time away from clustering. | |
1da177e4 | 874 | */ |
7b1fe597 | 875 | int reuse_swap_page(struct page *page) |
1da177e4 | 876 | { |
c475a8ab HD |
877 | int count; |
878 | ||
309381fe | 879 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
5ad64688 HD |
880 | if (unlikely(PageKsm(page))) |
881 | return 0; | |
c475a8ab | 882 | count = page_mapcount(page); |
7b1fe597 | 883 | if (count <= 1 && PageSwapCache(page)) { |
c475a8ab | 884 | count += page_swapcount(page); |
7b1fe597 HD |
885 | if (count == 1 && !PageWriteback(page)) { |
886 | delete_from_swap_cache(page); | |
887 | SetPageDirty(page); | |
888 | } | |
889 | } | |
5ad64688 | 890 | return count <= 1; |
1da177e4 LT |
891 | } |
892 | ||
893 | /* | |
a2c43eed HD |
894 | * If swap is getting full, or if there are no more mappings of this page, |
895 | * then try_to_free_swap is called to free its swap space. | |
1da177e4 | 896 | */ |
a2c43eed | 897 | int try_to_free_swap(struct page *page) |
1da177e4 | 898 | { |
309381fe | 899 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
1da177e4 LT |
900 | |
901 | if (!PageSwapCache(page)) | |
902 | return 0; | |
903 | if (PageWriteback(page)) | |
904 | return 0; | |
a2c43eed | 905 | if (page_swapcount(page)) |
1da177e4 LT |
906 | return 0; |
907 | ||
b73d7fce HD |
908 | /* |
909 | * Once hibernation has begun to create its image of memory, | |
910 | * there's a danger that one of the calls to try_to_free_swap() | |
911 | * - most probably a call from __try_to_reclaim_swap() while | |
912 | * hibernation is allocating its own swap pages for the image, | |
913 | * but conceivably even a call from memory reclaim - will free | |
914 | * the swap from a page which has already been recorded in the | |
915 | * image as a clean swapcache page, and then reuse its swap for | |
916 | * another page of the image. On waking from hibernation, the | |
917 | * original page might be freed under memory pressure, then | |
918 | * later read back in from swap, now with the wrong data. | |
919 | * | |
2de1a7e4 | 920 | * Hibernation suspends storage while it is writing the image |
f90ac398 | 921 | * to disk so check that here. |
b73d7fce | 922 | */ |
f90ac398 | 923 | if (pm_suspended_storage()) |
b73d7fce HD |
924 | return 0; |
925 | ||
a2c43eed HD |
926 | delete_from_swap_cache(page); |
927 | SetPageDirty(page); | |
928 | return 1; | |
68a22394 RR |
929 | } |
930 | ||
1da177e4 LT |
931 | /* |
932 | * Free the swap entry like above, but also try to | |
933 | * free the page cache entry if it is the last user. | |
934 | */ | |
2509ef26 | 935 | int free_swap_and_cache(swp_entry_t entry) |
1da177e4 | 936 | { |
2509ef26 | 937 | struct swap_info_struct *p; |
1da177e4 LT |
938 | struct page *page = NULL; |
939 | ||
a7420aa5 | 940 | if (non_swap_entry(entry)) |
2509ef26 | 941 | return 1; |
0697212a | 942 | |
1da177e4 LT |
943 | p = swap_info_get(entry); |
944 | if (p) { | |
253d553b | 945 | if (swap_entry_free(p, entry, 1) == SWAP_HAS_CACHE) { |
33806f06 SL |
946 | page = find_get_page(swap_address_space(entry), |
947 | entry.val); | |
8413ac9d | 948 | if (page && !trylock_page(page)) { |
93fac704 NP |
949 | page_cache_release(page); |
950 | page = NULL; | |
951 | } | |
952 | } | |
ec8acf20 | 953 | spin_unlock(&p->lock); |
1da177e4 LT |
954 | } |
955 | if (page) { | |
a2c43eed HD |
956 | /* |
957 | * Not mapped elsewhere, or swap space full? Free it! | |
958 | * Also recheck PageSwapCache now page is locked (above). | |
959 | */ | |
93fac704 | 960 | if (PageSwapCache(page) && !PageWriteback(page) && |
a2c43eed | 961 | (!page_mapped(page) || vm_swap_full())) { |
1da177e4 LT |
962 | delete_from_swap_cache(page); |
963 | SetPageDirty(page); | |
964 | } | |
965 | unlock_page(page); | |
966 | page_cache_release(page); | |
967 | } | |
2509ef26 | 968 | return p != NULL; |
1da177e4 LT |
969 | } |
970 | ||
b0cb1a19 | 971 | #ifdef CONFIG_HIBERNATION |
f577eb30 | 972 | /* |
915bae9e | 973 | * Find the swap type that corresponds to given device (if any). |
f577eb30 | 974 | * |
915bae9e RW |
975 | * @offset - number of the PAGE_SIZE-sized block of the device, starting |
976 | * from 0, in which the swap header is expected to be located. | |
977 | * | |
978 | * This is needed for the suspend to disk (aka swsusp). | |
f577eb30 | 979 | */ |
7bf23687 | 980 | int swap_type_of(dev_t device, sector_t offset, struct block_device **bdev_p) |
f577eb30 | 981 | { |
915bae9e | 982 | struct block_device *bdev = NULL; |
efa90a98 | 983 | int type; |
f577eb30 | 984 | |
915bae9e RW |
985 | if (device) |
986 | bdev = bdget(device); | |
987 | ||
f577eb30 | 988 | spin_lock(&swap_lock); |
efa90a98 HD |
989 | for (type = 0; type < nr_swapfiles; type++) { |
990 | struct swap_info_struct *sis = swap_info[type]; | |
f577eb30 | 991 | |
915bae9e | 992 | if (!(sis->flags & SWP_WRITEOK)) |
f577eb30 | 993 | continue; |
b6b5bce3 | 994 | |
915bae9e | 995 | if (!bdev) { |
7bf23687 | 996 | if (bdev_p) |
dddac6a7 | 997 | *bdev_p = bdgrab(sis->bdev); |
7bf23687 | 998 | |
6e1819d6 | 999 | spin_unlock(&swap_lock); |
efa90a98 | 1000 | return type; |
6e1819d6 | 1001 | } |
915bae9e | 1002 | if (bdev == sis->bdev) { |
9625a5f2 | 1003 | struct swap_extent *se = &sis->first_swap_extent; |
915bae9e | 1004 | |
915bae9e | 1005 | if (se->start_block == offset) { |
7bf23687 | 1006 | if (bdev_p) |
dddac6a7 | 1007 | *bdev_p = bdgrab(sis->bdev); |
7bf23687 | 1008 | |
915bae9e RW |
1009 | spin_unlock(&swap_lock); |
1010 | bdput(bdev); | |
efa90a98 | 1011 | return type; |
915bae9e | 1012 | } |
f577eb30 RW |
1013 | } |
1014 | } | |
1015 | spin_unlock(&swap_lock); | |
915bae9e RW |
1016 | if (bdev) |
1017 | bdput(bdev); | |
1018 | ||
f577eb30 RW |
1019 | return -ENODEV; |
1020 | } | |
1021 | ||
73c34b6a HD |
1022 | /* |
1023 | * Get the (PAGE_SIZE) block corresponding to given offset on the swapdev | |
1024 | * corresponding to given index in swap_info (swap type). | |
1025 | */ | |
1026 | sector_t swapdev_block(int type, pgoff_t offset) | |
1027 | { | |
1028 | struct block_device *bdev; | |
1029 | ||
1030 | if ((unsigned int)type >= nr_swapfiles) | |
1031 | return 0; | |
1032 | if (!(swap_info[type]->flags & SWP_WRITEOK)) | |
1033 | return 0; | |
d4906e1a | 1034 | return map_swap_entry(swp_entry(type, offset), &bdev); |
73c34b6a HD |
1035 | } |
1036 | ||
f577eb30 RW |
1037 | /* |
1038 | * Return either the total number of swap pages of given type, or the number | |
1039 | * of free pages of that type (depending on @free) | |
1040 | * | |
1041 | * This is needed for software suspend | |
1042 | */ | |
1043 | unsigned int count_swap_pages(int type, int free) | |
1044 | { | |
1045 | unsigned int n = 0; | |
1046 | ||
efa90a98 HD |
1047 | spin_lock(&swap_lock); |
1048 | if ((unsigned int)type < nr_swapfiles) { | |
1049 | struct swap_info_struct *sis = swap_info[type]; | |
1050 | ||
ec8acf20 | 1051 | spin_lock(&sis->lock); |
efa90a98 HD |
1052 | if (sis->flags & SWP_WRITEOK) { |
1053 | n = sis->pages; | |
f577eb30 | 1054 | if (free) |
efa90a98 | 1055 | n -= sis->inuse_pages; |
f577eb30 | 1056 | } |
ec8acf20 | 1057 | spin_unlock(&sis->lock); |
f577eb30 | 1058 | } |
efa90a98 | 1059 | spin_unlock(&swap_lock); |
f577eb30 RW |
1060 | return n; |
1061 | } | |
73c34b6a | 1062 | #endif /* CONFIG_HIBERNATION */ |
f577eb30 | 1063 | |
179ef71c CG |
1064 | static inline int maybe_same_pte(pte_t pte, pte_t swp_pte) |
1065 | { | |
1066 | #ifdef CONFIG_MEM_SOFT_DIRTY | |
1067 | /* | |
1068 | * When pte keeps soft dirty bit the pte generated | |
1069 | * from swap entry does not has it, still it's same | |
1070 | * pte from logical point of view. | |
1071 | */ | |
1072 | pte_t swp_pte_dirty = pte_swp_mksoft_dirty(swp_pte); | |
1073 | return pte_same(pte, swp_pte) || pte_same(pte, swp_pte_dirty); | |
1074 | #else | |
1075 | return pte_same(pte, swp_pte); | |
1076 | #endif | |
1077 | } | |
1078 | ||
1da177e4 | 1079 | /* |
72866f6f HD |
1080 | * No need to decide whether this PTE shares the swap entry with others, |
1081 | * just let do_wp_page work it out if a write is requested later - to | |
1082 | * force COW, vm_page_prot omits write permission from any private vma. | |
1da177e4 | 1083 | */ |
044d66c1 | 1084 | static int unuse_pte(struct vm_area_struct *vma, pmd_t *pmd, |
1da177e4 LT |
1085 | unsigned long addr, swp_entry_t entry, struct page *page) |
1086 | { | |
9e16b7fb | 1087 | struct page *swapcache; |
72835c86 | 1088 | struct mem_cgroup *memcg; |
044d66c1 HD |
1089 | spinlock_t *ptl; |
1090 | pte_t *pte; | |
1091 | int ret = 1; | |
1092 | ||
9e16b7fb HD |
1093 | swapcache = page; |
1094 | page = ksm_might_need_to_copy(page, vma, addr); | |
1095 | if (unlikely(!page)) | |
1096 | return -ENOMEM; | |
1097 | ||
72835c86 JW |
1098 | if (mem_cgroup_try_charge_swapin(vma->vm_mm, page, |
1099 | GFP_KERNEL, &memcg)) { | |
044d66c1 | 1100 | ret = -ENOMEM; |
85d9fc89 KH |
1101 | goto out_nolock; |
1102 | } | |
044d66c1 HD |
1103 | |
1104 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); | |
179ef71c | 1105 | if (unlikely(!maybe_same_pte(*pte, swp_entry_to_pte(entry)))) { |
5d84c776 | 1106 | mem_cgroup_cancel_charge_swapin(memcg); |
044d66c1 HD |
1107 | ret = 0; |
1108 | goto out; | |
1109 | } | |
8a9f3ccd | 1110 | |
b084d435 | 1111 | dec_mm_counter(vma->vm_mm, MM_SWAPENTS); |
d559db08 | 1112 | inc_mm_counter(vma->vm_mm, MM_ANONPAGES); |
1da177e4 LT |
1113 | get_page(page); |
1114 | set_pte_at(vma->vm_mm, addr, pte, | |
1115 | pte_mkold(mk_pte(page, vma->vm_page_prot))); | |
9e16b7fb HD |
1116 | if (page == swapcache) |
1117 | page_add_anon_rmap(page, vma, addr); | |
1118 | else /* ksm created a completely new copy */ | |
1119 | page_add_new_anon_rmap(page, vma, addr); | |
72835c86 | 1120 | mem_cgroup_commit_charge_swapin(page, memcg); |
1da177e4 LT |
1121 | swap_free(entry); |
1122 | /* | |
1123 | * Move the page to the active list so it is not | |
1124 | * immediately swapped out again after swapon. | |
1125 | */ | |
1126 | activate_page(page); | |
044d66c1 HD |
1127 | out: |
1128 | pte_unmap_unlock(pte, ptl); | |
85d9fc89 | 1129 | out_nolock: |
9e16b7fb HD |
1130 | if (page != swapcache) { |
1131 | unlock_page(page); | |
1132 | put_page(page); | |
1133 | } | |
044d66c1 | 1134 | return ret; |
1da177e4 LT |
1135 | } |
1136 | ||
1137 | static int unuse_pte_range(struct vm_area_struct *vma, pmd_t *pmd, | |
1138 | unsigned long addr, unsigned long end, | |
1139 | swp_entry_t entry, struct page *page) | |
1140 | { | |
1da177e4 | 1141 | pte_t swp_pte = swp_entry_to_pte(entry); |
705e87c0 | 1142 | pte_t *pte; |
8a9f3ccd | 1143 | int ret = 0; |
1da177e4 | 1144 | |
044d66c1 HD |
1145 | /* |
1146 | * We don't actually need pte lock while scanning for swp_pte: since | |
1147 | * we hold page lock and mmap_sem, swp_pte cannot be inserted into the | |
1148 | * page table while we're scanning; though it could get zapped, and on | |
1149 | * some architectures (e.g. x86_32 with PAE) we might catch a glimpse | |
1150 | * of unmatched parts which look like swp_pte, so unuse_pte must | |
1151 | * recheck under pte lock. Scanning without pte lock lets it be | |
2de1a7e4 | 1152 | * preemptable whenever CONFIG_PREEMPT but not CONFIG_HIGHPTE. |
044d66c1 HD |
1153 | */ |
1154 | pte = pte_offset_map(pmd, addr); | |
1da177e4 LT |
1155 | do { |
1156 | /* | |
1157 | * swapoff spends a _lot_ of time in this loop! | |
1158 | * Test inline before going to call unuse_pte. | |
1159 | */ | |
179ef71c | 1160 | if (unlikely(maybe_same_pte(*pte, swp_pte))) { |
044d66c1 HD |
1161 | pte_unmap(pte); |
1162 | ret = unuse_pte(vma, pmd, addr, entry, page); | |
1163 | if (ret) | |
1164 | goto out; | |
1165 | pte = pte_offset_map(pmd, addr); | |
1da177e4 LT |
1166 | } |
1167 | } while (pte++, addr += PAGE_SIZE, addr != end); | |
044d66c1 HD |
1168 | pte_unmap(pte - 1); |
1169 | out: | |
8a9f3ccd | 1170 | return ret; |
1da177e4 LT |
1171 | } |
1172 | ||
1173 | static inline int unuse_pmd_range(struct vm_area_struct *vma, pud_t *pud, | |
1174 | unsigned long addr, unsigned long end, | |
1175 | swp_entry_t entry, struct page *page) | |
1176 | { | |
1177 | pmd_t *pmd; | |
1178 | unsigned long next; | |
8a9f3ccd | 1179 | int ret; |
1da177e4 LT |
1180 | |
1181 | pmd = pmd_offset(pud, addr); | |
1182 | do { | |
1183 | next = pmd_addr_end(addr, end); | |
1a5a9906 | 1184 | if (pmd_none_or_trans_huge_or_clear_bad(pmd)) |
1da177e4 | 1185 | continue; |
8a9f3ccd BS |
1186 | ret = unuse_pte_range(vma, pmd, addr, next, entry, page); |
1187 | if (ret) | |
1188 | return ret; | |
1da177e4 LT |
1189 | } while (pmd++, addr = next, addr != end); |
1190 | return 0; | |
1191 | } | |
1192 | ||
1193 | static inline int unuse_pud_range(struct vm_area_struct *vma, pgd_t *pgd, | |
1194 | unsigned long addr, unsigned long end, | |
1195 | swp_entry_t entry, struct page *page) | |
1196 | { | |
1197 | pud_t *pud; | |
1198 | unsigned long next; | |
8a9f3ccd | 1199 | int ret; |
1da177e4 LT |
1200 | |
1201 | pud = pud_offset(pgd, addr); | |
1202 | do { | |
1203 | next = pud_addr_end(addr, end); | |
1204 | if (pud_none_or_clear_bad(pud)) | |
1205 | continue; | |
8a9f3ccd BS |
1206 | ret = unuse_pmd_range(vma, pud, addr, next, entry, page); |
1207 | if (ret) | |
1208 | return ret; | |
1da177e4 LT |
1209 | } while (pud++, addr = next, addr != end); |
1210 | return 0; | |
1211 | } | |
1212 | ||
1213 | static int unuse_vma(struct vm_area_struct *vma, | |
1214 | swp_entry_t entry, struct page *page) | |
1215 | { | |
1216 | pgd_t *pgd; | |
1217 | unsigned long addr, end, next; | |
8a9f3ccd | 1218 | int ret; |
1da177e4 | 1219 | |
3ca7b3c5 | 1220 | if (page_anon_vma(page)) { |
1da177e4 LT |
1221 | addr = page_address_in_vma(page, vma); |
1222 | if (addr == -EFAULT) | |
1223 | return 0; | |
1224 | else | |
1225 | end = addr + PAGE_SIZE; | |
1226 | } else { | |
1227 | addr = vma->vm_start; | |
1228 | end = vma->vm_end; | |
1229 | } | |
1230 | ||
1231 | pgd = pgd_offset(vma->vm_mm, addr); | |
1232 | do { | |
1233 | next = pgd_addr_end(addr, end); | |
1234 | if (pgd_none_or_clear_bad(pgd)) | |
1235 | continue; | |
8a9f3ccd BS |
1236 | ret = unuse_pud_range(vma, pgd, addr, next, entry, page); |
1237 | if (ret) | |
1238 | return ret; | |
1da177e4 LT |
1239 | } while (pgd++, addr = next, addr != end); |
1240 | return 0; | |
1241 | } | |
1242 | ||
1243 | static int unuse_mm(struct mm_struct *mm, | |
1244 | swp_entry_t entry, struct page *page) | |
1245 | { | |
1246 | struct vm_area_struct *vma; | |
8a9f3ccd | 1247 | int ret = 0; |
1da177e4 LT |
1248 | |
1249 | if (!down_read_trylock(&mm->mmap_sem)) { | |
1250 | /* | |
7d03431c FLVC |
1251 | * Activate page so shrink_inactive_list is unlikely to unmap |
1252 | * its ptes while lock is dropped, so swapoff can make progress. | |
1da177e4 | 1253 | */ |
c475a8ab | 1254 | activate_page(page); |
1da177e4 LT |
1255 | unlock_page(page); |
1256 | down_read(&mm->mmap_sem); | |
1257 | lock_page(page); | |
1258 | } | |
1da177e4 | 1259 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
8a9f3ccd | 1260 | if (vma->anon_vma && (ret = unuse_vma(vma, entry, page))) |
1da177e4 LT |
1261 | break; |
1262 | } | |
1da177e4 | 1263 | up_read(&mm->mmap_sem); |
8a9f3ccd | 1264 | return (ret < 0)? ret: 0; |
1da177e4 LT |
1265 | } |
1266 | ||
1267 | /* | |
38b5faf4 DM |
1268 | * Scan swap_map (or frontswap_map if frontswap parameter is true) |
1269 | * from current position to next entry still in use. | |
1da177e4 LT |
1270 | * Recycle to start on reaching the end, returning 0 when empty. |
1271 | */ | |
6eb396dc | 1272 | static unsigned int find_next_to_unuse(struct swap_info_struct *si, |
38b5faf4 | 1273 | unsigned int prev, bool frontswap) |
1da177e4 | 1274 | { |
6eb396dc HD |
1275 | unsigned int max = si->max; |
1276 | unsigned int i = prev; | |
8d69aaee | 1277 | unsigned char count; |
1da177e4 LT |
1278 | |
1279 | /* | |
5d337b91 | 1280 | * No need for swap_lock here: we're just looking |
1da177e4 LT |
1281 | * for whether an entry is in use, not modifying it; false |
1282 | * hits are okay, and sys_swapoff() has already prevented new | |
5d337b91 | 1283 | * allocations from this area (while holding swap_lock). |
1da177e4 LT |
1284 | */ |
1285 | for (;;) { | |
1286 | if (++i >= max) { | |
1287 | if (!prev) { | |
1288 | i = 0; | |
1289 | break; | |
1290 | } | |
1291 | /* | |
1292 | * No entries in use at top of swap_map, | |
1293 | * loop back to start and recheck there. | |
1294 | */ | |
1295 | max = prev + 1; | |
1296 | prev = 0; | |
1297 | i = 1; | |
1298 | } | |
38b5faf4 DM |
1299 | if (frontswap) { |
1300 | if (frontswap_test(si, i)) | |
1301 | break; | |
1302 | else | |
1303 | continue; | |
1304 | } | |
edfe23da | 1305 | count = ACCESS_ONCE(si->swap_map[i]); |
355cfa73 | 1306 | if (count && swap_count(count) != SWAP_MAP_BAD) |
1da177e4 LT |
1307 | break; |
1308 | } | |
1309 | return i; | |
1310 | } | |
1311 | ||
1312 | /* | |
1313 | * We completely avoid races by reading each swap page in advance, | |
1314 | * and then search for the process using it. All the necessary | |
1315 | * page table adjustments can then be made atomically. | |
38b5faf4 DM |
1316 | * |
1317 | * if the boolean frontswap is true, only unuse pages_to_unuse pages; | |
1318 | * pages_to_unuse==0 means all pages; ignored if frontswap is false | |
1da177e4 | 1319 | */ |
38b5faf4 DM |
1320 | int try_to_unuse(unsigned int type, bool frontswap, |
1321 | unsigned long pages_to_unuse) | |
1da177e4 | 1322 | { |
efa90a98 | 1323 | struct swap_info_struct *si = swap_info[type]; |
1da177e4 | 1324 | struct mm_struct *start_mm; |
edfe23da SL |
1325 | volatile unsigned char *swap_map; /* swap_map is accessed without |
1326 | * locking. Mark it as volatile | |
1327 | * to prevent compiler doing | |
1328 | * something odd. | |
1329 | */ | |
8d69aaee | 1330 | unsigned char swcount; |
1da177e4 LT |
1331 | struct page *page; |
1332 | swp_entry_t entry; | |
6eb396dc | 1333 | unsigned int i = 0; |
1da177e4 | 1334 | int retval = 0; |
1da177e4 LT |
1335 | |
1336 | /* | |
1337 | * When searching mms for an entry, a good strategy is to | |
1338 | * start at the first mm we freed the previous entry from | |
1339 | * (though actually we don't notice whether we or coincidence | |
1340 | * freed the entry). Initialize this start_mm with a hold. | |
1341 | * | |
1342 | * A simpler strategy would be to start at the last mm we | |
1343 | * freed the previous entry from; but that would take less | |
1344 | * advantage of mmlist ordering, which clusters forked mms | |
1345 | * together, child after parent. If we race with dup_mmap(), we | |
1346 | * prefer to resolve parent before child, lest we miss entries | |
1347 | * duplicated after we scanned child: using last mm would invert | |
570a335b | 1348 | * that. |
1da177e4 LT |
1349 | */ |
1350 | start_mm = &init_mm; | |
1351 | atomic_inc(&init_mm.mm_users); | |
1352 | ||
1353 | /* | |
1354 | * Keep on scanning until all entries have gone. Usually, | |
1355 | * one pass through swap_map is enough, but not necessarily: | |
1356 | * there are races when an instance of an entry might be missed. | |
1357 | */ | |
38b5faf4 | 1358 | while ((i = find_next_to_unuse(si, i, frontswap)) != 0) { |
1da177e4 LT |
1359 | if (signal_pending(current)) { |
1360 | retval = -EINTR; | |
1361 | break; | |
1362 | } | |
1363 | ||
886bb7e9 | 1364 | /* |
1da177e4 LT |
1365 | * Get a page for the entry, using the existing swap |
1366 | * cache page if there is one. Otherwise, get a clean | |
886bb7e9 | 1367 | * page and read the swap into it. |
1da177e4 LT |
1368 | */ |
1369 | swap_map = &si->swap_map[i]; | |
1370 | entry = swp_entry(type, i); | |
02098fea HD |
1371 | page = read_swap_cache_async(entry, |
1372 | GFP_HIGHUSER_MOVABLE, NULL, 0); | |
1da177e4 LT |
1373 | if (!page) { |
1374 | /* | |
1375 | * Either swap_duplicate() failed because entry | |
1376 | * has been freed independently, and will not be | |
1377 | * reused since sys_swapoff() already disabled | |
1378 | * allocation from here, or alloc_page() failed. | |
1379 | */ | |
edfe23da SL |
1380 | swcount = *swap_map; |
1381 | /* | |
1382 | * We don't hold lock here, so the swap entry could be | |
1383 | * SWAP_MAP_BAD (when the cluster is discarding). | |
1384 | * Instead of fail out, We can just skip the swap | |
1385 | * entry because swapoff will wait for discarding | |
1386 | * finish anyway. | |
1387 | */ | |
1388 | if (!swcount || swcount == SWAP_MAP_BAD) | |
1da177e4 LT |
1389 | continue; |
1390 | retval = -ENOMEM; | |
1391 | break; | |
1392 | } | |
1393 | ||
1394 | /* | |
1395 | * Don't hold on to start_mm if it looks like exiting. | |
1396 | */ | |
1397 | if (atomic_read(&start_mm->mm_users) == 1) { | |
1398 | mmput(start_mm); | |
1399 | start_mm = &init_mm; | |
1400 | atomic_inc(&init_mm.mm_users); | |
1401 | } | |
1402 | ||
1403 | /* | |
1404 | * Wait for and lock page. When do_swap_page races with | |
1405 | * try_to_unuse, do_swap_page can handle the fault much | |
1406 | * faster than try_to_unuse can locate the entry. This | |
1407 | * apparently redundant "wait_on_page_locked" lets try_to_unuse | |
1408 | * defer to do_swap_page in such a case - in some tests, | |
1409 | * do_swap_page and try_to_unuse repeatedly compete. | |
1410 | */ | |
1411 | wait_on_page_locked(page); | |
1412 | wait_on_page_writeback(page); | |
1413 | lock_page(page); | |
1414 | wait_on_page_writeback(page); | |
1415 | ||
1416 | /* | |
1417 | * Remove all references to entry. | |
1da177e4 | 1418 | */ |
1da177e4 | 1419 | swcount = *swap_map; |
aaa46865 HD |
1420 | if (swap_count(swcount) == SWAP_MAP_SHMEM) { |
1421 | retval = shmem_unuse(entry, page); | |
1422 | /* page has already been unlocked and released */ | |
1423 | if (retval < 0) | |
1424 | break; | |
1425 | continue; | |
1da177e4 | 1426 | } |
aaa46865 HD |
1427 | if (swap_count(swcount) && start_mm != &init_mm) |
1428 | retval = unuse_mm(start_mm, entry, page); | |
1429 | ||
355cfa73 | 1430 | if (swap_count(*swap_map)) { |
1da177e4 LT |
1431 | int set_start_mm = (*swap_map >= swcount); |
1432 | struct list_head *p = &start_mm->mmlist; | |
1433 | struct mm_struct *new_start_mm = start_mm; | |
1434 | struct mm_struct *prev_mm = start_mm; | |
1435 | struct mm_struct *mm; | |
1436 | ||
1437 | atomic_inc(&new_start_mm->mm_users); | |
1438 | atomic_inc(&prev_mm->mm_users); | |
1439 | spin_lock(&mmlist_lock); | |
aaa46865 | 1440 | while (swap_count(*swap_map) && !retval && |
1da177e4 LT |
1441 | (p = p->next) != &start_mm->mmlist) { |
1442 | mm = list_entry(p, struct mm_struct, mmlist); | |
70af7c5c | 1443 | if (!atomic_inc_not_zero(&mm->mm_users)) |
1da177e4 | 1444 | continue; |
1da177e4 LT |
1445 | spin_unlock(&mmlist_lock); |
1446 | mmput(prev_mm); | |
1447 | prev_mm = mm; | |
1448 | ||
1449 | cond_resched(); | |
1450 | ||
1451 | swcount = *swap_map; | |
355cfa73 | 1452 | if (!swap_count(swcount)) /* any usage ? */ |
1da177e4 | 1453 | ; |
aaa46865 | 1454 | else if (mm == &init_mm) |
1da177e4 | 1455 | set_start_mm = 1; |
aaa46865 | 1456 | else |
1da177e4 | 1457 | retval = unuse_mm(mm, entry, page); |
355cfa73 | 1458 | |
32c5fc10 | 1459 | if (set_start_mm && *swap_map < swcount) { |
1da177e4 LT |
1460 | mmput(new_start_mm); |
1461 | atomic_inc(&mm->mm_users); | |
1462 | new_start_mm = mm; | |
1463 | set_start_mm = 0; | |
1464 | } | |
1465 | spin_lock(&mmlist_lock); | |
1466 | } | |
1467 | spin_unlock(&mmlist_lock); | |
1468 | mmput(prev_mm); | |
1469 | mmput(start_mm); | |
1470 | start_mm = new_start_mm; | |
1471 | } | |
1472 | if (retval) { | |
1473 | unlock_page(page); | |
1474 | page_cache_release(page); | |
1475 | break; | |
1476 | } | |
1477 | ||
1da177e4 LT |
1478 | /* |
1479 | * If a reference remains (rare), we would like to leave | |
1480 | * the page in the swap cache; but try_to_unmap could | |
1481 | * then re-duplicate the entry once we drop page lock, | |
1482 | * so we might loop indefinitely; also, that page could | |
1483 | * not be swapped out to other storage meanwhile. So: | |
1484 | * delete from cache even if there's another reference, | |
1485 | * after ensuring that the data has been saved to disk - | |
1486 | * since if the reference remains (rarer), it will be | |
1487 | * read from disk into another page. Splitting into two | |
1488 | * pages would be incorrect if swap supported "shared | |
1489 | * private" pages, but they are handled by tmpfs files. | |
5ad64688 HD |
1490 | * |
1491 | * Given how unuse_vma() targets one particular offset | |
1492 | * in an anon_vma, once the anon_vma has been determined, | |
1493 | * this splitting happens to be just what is needed to | |
1494 | * handle where KSM pages have been swapped out: re-reading | |
1495 | * is unnecessarily slow, but we can fix that later on. | |
1da177e4 | 1496 | */ |
355cfa73 KH |
1497 | if (swap_count(*swap_map) && |
1498 | PageDirty(page) && PageSwapCache(page)) { | |
1da177e4 LT |
1499 | struct writeback_control wbc = { |
1500 | .sync_mode = WB_SYNC_NONE, | |
1501 | }; | |
1502 | ||
1503 | swap_writepage(page, &wbc); | |
1504 | lock_page(page); | |
1505 | wait_on_page_writeback(page); | |
1506 | } | |
68bdc8d6 HD |
1507 | |
1508 | /* | |
1509 | * It is conceivable that a racing task removed this page from | |
1510 | * swap cache just before we acquired the page lock at the top, | |
1511 | * or while we dropped it in unuse_mm(). The page might even | |
1512 | * be back in swap cache on another swap area: that we must not | |
1513 | * delete, since it may not have been written out to swap yet. | |
1514 | */ | |
1515 | if (PageSwapCache(page) && | |
1516 | likely(page_private(page) == entry.val)) | |
2e0e26c7 | 1517 | delete_from_swap_cache(page); |
1da177e4 LT |
1518 | |
1519 | /* | |
1520 | * So we could skip searching mms once swap count went | |
1521 | * to 1, we did not mark any present ptes as dirty: must | |
2706a1b8 | 1522 | * mark page dirty so shrink_page_list will preserve it. |
1da177e4 LT |
1523 | */ |
1524 | SetPageDirty(page); | |
1525 | unlock_page(page); | |
1526 | page_cache_release(page); | |
1527 | ||
1528 | /* | |
1529 | * Make sure that we aren't completely killing | |
1530 | * interactive performance. | |
1531 | */ | |
1532 | cond_resched(); | |
38b5faf4 DM |
1533 | if (frontswap && pages_to_unuse > 0) { |
1534 | if (!--pages_to_unuse) | |
1535 | break; | |
1536 | } | |
1da177e4 LT |
1537 | } |
1538 | ||
1539 | mmput(start_mm); | |
1da177e4 LT |
1540 | return retval; |
1541 | } | |
1542 | ||
1543 | /* | |
5d337b91 HD |
1544 | * After a successful try_to_unuse, if no swap is now in use, we know |
1545 | * we can empty the mmlist. swap_lock must be held on entry and exit. | |
1546 | * Note that mmlist_lock nests inside swap_lock, and an mm must be | |
1da177e4 LT |
1547 | * added to the mmlist just after page_duplicate - before would be racy. |
1548 | */ | |
1549 | static void drain_mmlist(void) | |
1550 | { | |
1551 | struct list_head *p, *next; | |
efa90a98 | 1552 | unsigned int type; |
1da177e4 | 1553 | |
efa90a98 HD |
1554 | for (type = 0; type < nr_swapfiles; type++) |
1555 | if (swap_info[type]->inuse_pages) | |
1da177e4 LT |
1556 | return; |
1557 | spin_lock(&mmlist_lock); | |
1558 | list_for_each_safe(p, next, &init_mm.mmlist) | |
1559 | list_del_init(p); | |
1560 | spin_unlock(&mmlist_lock); | |
1561 | } | |
1562 | ||
1563 | /* | |
1564 | * Use this swapdev's extent info to locate the (PAGE_SIZE) block which | |
d4906e1a LS |
1565 | * corresponds to page offset for the specified swap entry. |
1566 | * Note that the type of this function is sector_t, but it returns page offset | |
1567 | * into the bdev, not sector offset. | |
1da177e4 | 1568 | */ |
d4906e1a | 1569 | static sector_t map_swap_entry(swp_entry_t entry, struct block_device **bdev) |
1da177e4 | 1570 | { |
f29ad6a9 HD |
1571 | struct swap_info_struct *sis; |
1572 | struct swap_extent *start_se; | |
1573 | struct swap_extent *se; | |
1574 | pgoff_t offset; | |
1575 | ||
efa90a98 | 1576 | sis = swap_info[swp_type(entry)]; |
f29ad6a9 HD |
1577 | *bdev = sis->bdev; |
1578 | ||
1579 | offset = swp_offset(entry); | |
1580 | start_se = sis->curr_swap_extent; | |
1581 | se = start_se; | |
1da177e4 LT |
1582 | |
1583 | for ( ; ; ) { | |
1584 | struct list_head *lh; | |
1585 | ||
1586 | if (se->start_page <= offset && | |
1587 | offset < (se->start_page + se->nr_pages)) { | |
1588 | return se->start_block + (offset - se->start_page); | |
1589 | } | |
11d31886 | 1590 | lh = se->list.next; |
1da177e4 LT |
1591 | se = list_entry(lh, struct swap_extent, list); |
1592 | sis->curr_swap_extent = se; | |
1593 | BUG_ON(se == start_se); /* It *must* be present */ | |
1594 | } | |
1595 | } | |
1596 | ||
d4906e1a LS |
1597 | /* |
1598 | * Returns the page offset into bdev for the specified page's swap entry. | |
1599 | */ | |
1600 | sector_t map_swap_page(struct page *page, struct block_device **bdev) | |
1601 | { | |
1602 | swp_entry_t entry; | |
1603 | entry.val = page_private(page); | |
1604 | return map_swap_entry(entry, bdev); | |
1605 | } | |
1606 | ||
1da177e4 LT |
1607 | /* |
1608 | * Free all of a swapdev's extent information | |
1609 | */ | |
1610 | static void destroy_swap_extents(struct swap_info_struct *sis) | |
1611 | { | |
9625a5f2 | 1612 | while (!list_empty(&sis->first_swap_extent.list)) { |
1da177e4 LT |
1613 | struct swap_extent *se; |
1614 | ||
9625a5f2 | 1615 | se = list_entry(sis->first_swap_extent.list.next, |
1da177e4 LT |
1616 | struct swap_extent, list); |
1617 | list_del(&se->list); | |
1618 | kfree(se); | |
1619 | } | |
62c230bc MG |
1620 | |
1621 | if (sis->flags & SWP_FILE) { | |
1622 | struct file *swap_file = sis->swap_file; | |
1623 | struct address_space *mapping = swap_file->f_mapping; | |
1624 | ||
1625 | sis->flags &= ~SWP_FILE; | |
1626 | mapping->a_ops->swap_deactivate(swap_file); | |
1627 | } | |
1da177e4 LT |
1628 | } |
1629 | ||
1630 | /* | |
1631 | * Add a block range (and the corresponding page range) into this swapdev's | |
11d31886 | 1632 | * extent list. The extent list is kept sorted in page order. |
1da177e4 | 1633 | * |
11d31886 | 1634 | * This function rather assumes that it is called in ascending page order. |
1da177e4 | 1635 | */ |
a509bc1a | 1636 | int |
1da177e4 LT |
1637 | add_swap_extent(struct swap_info_struct *sis, unsigned long start_page, |
1638 | unsigned long nr_pages, sector_t start_block) | |
1639 | { | |
1640 | struct swap_extent *se; | |
1641 | struct swap_extent *new_se; | |
1642 | struct list_head *lh; | |
1643 | ||
9625a5f2 HD |
1644 | if (start_page == 0) { |
1645 | se = &sis->first_swap_extent; | |
1646 | sis->curr_swap_extent = se; | |
1647 | se->start_page = 0; | |
1648 | se->nr_pages = nr_pages; | |
1649 | se->start_block = start_block; | |
1650 | return 1; | |
1651 | } else { | |
1652 | lh = sis->first_swap_extent.list.prev; /* Highest extent */ | |
1da177e4 | 1653 | se = list_entry(lh, struct swap_extent, list); |
11d31886 HD |
1654 | BUG_ON(se->start_page + se->nr_pages != start_page); |
1655 | if (se->start_block + se->nr_pages == start_block) { | |
1da177e4 LT |
1656 | /* Merge it */ |
1657 | se->nr_pages += nr_pages; | |
1658 | return 0; | |
1659 | } | |
1da177e4 LT |
1660 | } |
1661 | ||
1662 | /* | |
1663 | * No merge. Insert a new extent, preserving ordering. | |
1664 | */ | |
1665 | new_se = kmalloc(sizeof(*se), GFP_KERNEL); | |
1666 | if (new_se == NULL) | |
1667 | return -ENOMEM; | |
1668 | new_se->start_page = start_page; | |
1669 | new_se->nr_pages = nr_pages; | |
1670 | new_se->start_block = start_block; | |
1671 | ||
9625a5f2 | 1672 | list_add_tail(&new_se->list, &sis->first_swap_extent.list); |
53092a74 | 1673 | return 1; |
1da177e4 LT |
1674 | } |
1675 | ||
1676 | /* | |
1677 | * A `swap extent' is a simple thing which maps a contiguous range of pages | |
1678 | * onto a contiguous range of disk blocks. An ordered list of swap extents | |
1679 | * is built at swapon time and is then used at swap_writepage/swap_readpage | |
1680 | * time for locating where on disk a page belongs. | |
1681 | * | |
1682 | * If the swapfile is an S_ISBLK block device, a single extent is installed. | |
1683 | * This is done so that the main operating code can treat S_ISBLK and S_ISREG | |
1684 | * swap files identically. | |
1685 | * | |
1686 | * Whether the swapdev is an S_ISREG file or an S_ISBLK blockdev, the swap | |
1687 | * extent list operates in PAGE_SIZE disk blocks. Both S_ISREG and S_ISBLK | |
1688 | * swapfiles are handled *identically* after swapon time. | |
1689 | * | |
1690 | * For S_ISREG swapfiles, setup_swap_extents() will walk all the file's blocks | |
1691 | * and will parse them into an ordered extent list, in PAGE_SIZE chunks. If | |
1692 | * some stray blocks are found which do not fall within the PAGE_SIZE alignment | |
1693 | * requirements, they are simply tossed out - we will never use those blocks | |
1694 | * for swapping. | |
1695 | * | |
b0d9bcd4 | 1696 | * For S_ISREG swapfiles we set S_SWAPFILE across the life of the swapon. This |
1da177e4 LT |
1697 | * prevents root from shooting her foot off by ftruncating an in-use swapfile, |
1698 | * which will scribble on the fs. | |
1699 | * | |
1700 | * The amount of disk space which a single swap extent represents varies. | |
1701 | * Typically it is in the 1-4 megabyte range. So we can have hundreds of | |
1702 | * extents in the list. To avoid much list walking, we cache the previous | |
1703 | * search location in `curr_swap_extent', and start new searches from there. | |
1704 | * This is extremely effective. The average number of iterations in | |
1705 | * map_swap_page() has been measured at about 0.3 per page. - akpm. | |
1706 | */ | |
53092a74 | 1707 | static int setup_swap_extents(struct swap_info_struct *sis, sector_t *span) |
1da177e4 | 1708 | { |
62c230bc MG |
1709 | struct file *swap_file = sis->swap_file; |
1710 | struct address_space *mapping = swap_file->f_mapping; | |
1711 | struct inode *inode = mapping->host; | |
1da177e4 LT |
1712 | int ret; |
1713 | ||
1da177e4 LT |
1714 | if (S_ISBLK(inode->i_mode)) { |
1715 | ret = add_swap_extent(sis, 0, sis->max, 0); | |
53092a74 | 1716 | *span = sis->pages; |
a509bc1a | 1717 | return ret; |
1da177e4 LT |
1718 | } |
1719 | ||
62c230bc | 1720 | if (mapping->a_ops->swap_activate) { |
a509bc1a | 1721 | ret = mapping->a_ops->swap_activate(sis, swap_file, span); |
62c230bc MG |
1722 | if (!ret) { |
1723 | sis->flags |= SWP_FILE; | |
1724 | ret = add_swap_extent(sis, 0, sis->max, 0); | |
1725 | *span = sis->pages; | |
1726 | } | |
a509bc1a | 1727 | return ret; |
62c230bc MG |
1728 | } |
1729 | ||
a509bc1a | 1730 | return generic_swapfile_activate(sis, swap_file, span); |
1da177e4 LT |
1731 | } |
1732 | ||
cf0cac0a | 1733 | static void _enable_swap_info(struct swap_info_struct *p, int prio, |
2a8f9449 SL |
1734 | unsigned char *swap_map, |
1735 | struct swap_cluster_info *cluster_info) | |
40531542 | 1736 | { |
adfab836 | 1737 | struct swap_info_struct *si; |
40531542 | 1738 | |
40531542 CEB |
1739 | if (prio >= 0) |
1740 | p->prio = prio; | |
1741 | else | |
1742 | p->prio = --least_priority; | |
1743 | p->swap_map = swap_map; | |
2a8f9449 | 1744 | p->cluster_info = cluster_info; |
40531542 | 1745 | p->flags |= SWP_WRITEOK; |
ec8acf20 | 1746 | atomic_long_add(p->pages, &nr_swap_pages); |
40531542 CEB |
1747 | total_swap_pages += p->pages; |
1748 | ||
adfab836 DS |
1749 | assert_spin_locked(&swap_lock); |
1750 | BUG_ON(!list_empty(&p->list)); | |
1751 | /* | |
1752 | * insert into swap list; the list is in priority order, | |
1753 | * so that get_swap_page() can get a page from the highest | |
1754 | * priority swap_info_struct with available page(s), and | |
1755 | * swapoff can adjust the auto-assigned (i.e. negative) prio | |
1756 | * values for any lower-priority swap_info_structs when | |
1757 | * removing a negative-prio swap_info_struct | |
1758 | */ | |
1759 | list_for_each_entry(si, &swap_list_head, list) { | |
1760 | if (p->prio >= si->prio) { | |
1761 | list_add_tail(&p->list, &si->list); | |
1762 | return; | |
1763 | } | |
40531542 | 1764 | } |
adfab836 DS |
1765 | /* |
1766 | * this covers two cases: | |
1767 | * 1) p->prio is less than all existing prio | |
1768 | * 2) the swap list is empty | |
1769 | */ | |
1770 | list_add_tail(&p->list, &swap_list_head); | |
cf0cac0a CEB |
1771 | } |
1772 | ||
1773 | static void enable_swap_info(struct swap_info_struct *p, int prio, | |
1774 | unsigned char *swap_map, | |
2a8f9449 | 1775 | struct swap_cluster_info *cluster_info, |
cf0cac0a CEB |
1776 | unsigned long *frontswap_map) |
1777 | { | |
4f89849d | 1778 | frontswap_init(p->type, frontswap_map); |
cf0cac0a | 1779 | spin_lock(&swap_lock); |
ec8acf20 | 1780 | spin_lock(&p->lock); |
2a8f9449 | 1781 | _enable_swap_info(p, prio, swap_map, cluster_info); |
ec8acf20 | 1782 | spin_unlock(&p->lock); |
cf0cac0a CEB |
1783 | spin_unlock(&swap_lock); |
1784 | } | |
1785 | ||
1786 | static void reinsert_swap_info(struct swap_info_struct *p) | |
1787 | { | |
1788 | spin_lock(&swap_lock); | |
ec8acf20 | 1789 | spin_lock(&p->lock); |
2a8f9449 | 1790 | _enable_swap_info(p, p->prio, p->swap_map, p->cluster_info); |
ec8acf20 | 1791 | spin_unlock(&p->lock); |
40531542 CEB |
1792 | spin_unlock(&swap_lock); |
1793 | } | |
1794 | ||
c4ea37c2 | 1795 | SYSCALL_DEFINE1(swapoff, const char __user *, specialfile) |
1da177e4 | 1796 | { |
73c34b6a | 1797 | struct swap_info_struct *p = NULL; |
8d69aaee | 1798 | unsigned char *swap_map; |
2a8f9449 | 1799 | struct swap_cluster_info *cluster_info; |
4f89849d | 1800 | unsigned long *frontswap_map; |
1da177e4 LT |
1801 | struct file *swap_file, *victim; |
1802 | struct address_space *mapping; | |
1803 | struct inode *inode; | |
91a27b2a | 1804 | struct filename *pathname; |
adfab836 | 1805 | int err, found = 0; |
5b808a23 | 1806 | unsigned int old_block_size; |
886bb7e9 | 1807 | |
1da177e4 LT |
1808 | if (!capable(CAP_SYS_ADMIN)) |
1809 | return -EPERM; | |
1810 | ||
191c5424 AV |
1811 | BUG_ON(!current->mm); |
1812 | ||
1da177e4 | 1813 | pathname = getname(specialfile); |
1da177e4 | 1814 | if (IS_ERR(pathname)) |
f58b59c1 | 1815 | return PTR_ERR(pathname); |
1da177e4 | 1816 | |
669abf4e | 1817 | victim = file_open_name(pathname, O_RDWR|O_LARGEFILE, 0); |
1da177e4 LT |
1818 | err = PTR_ERR(victim); |
1819 | if (IS_ERR(victim)) | |
1820 | goto out; | |
1821 | ||
1822 | mapping = victim->f_mapping; | |
5d337b91 | 1823 | spin_lock(&swap_lock); |
adfab836 | 1824 | list_for_each_entry(p, &swap_list_head, list) { |
22c6f8fd | 1825 | if (p->flags & SWP_WRITEOK) { |
adfab836 DS |
1826 | if (p->swap_file->f_mapping == mapping) { |
1827 | found = 1; | |
1da177e4 | 1828 | break; |
adfab836 | 1829 | } |
1da177e4 | 1830 | } |
1da177e4 | 1831 | } |
adfab836 | 1832 | if (!found) { |
1da177e4 | 1833 | err = -EINVAL; |
5d337b91 | 1834 | spin_unlock(&swap_lock); |
1da177e4 LT |
1835 | goto out_dput; |
1836 | } | |
191c5424 | 1837 | if (!security_vm_enough_memory_mm(current->mm, p->pages)) |
1da177e4 LT |
1838 | vm_unacct_memory(p->pages); |
1839 | else { | |
1840 | err = -ENOMEM; | |
5d337b91 | 1841 | spin_unlock(&swap_lock); |
1da177e4 LT |
1842 | goto out_dput; |
1843 | } | |
ec8acf20 | 1844 | spin_lock(&p->lock); |
78ecba08 | 1845 | if (p->prio < 0) { |
adfab836 DS |
1846 | struct swap_info_struct *si = p; |
1847 | ||
1848 | list_for_each_entry_continue(si, &swap_list_head, list) { | |
1849 | si->prio++; | |
1850 | } | |
78ecba08 HD |
1851 | least_priority++; |
1852 | } | |
adfab836 | 1853 | list_del_init(&p->list); |
ec8acf20 | 1854 | atomic_long_sub(p->pages, &nr_swap_pages); |
1da177e4 LT |
1855 | total_swap_pages -= p->pages; |
1856 | p->flags &= ~SWP_WRITEOK; | |
ec8acf20 | 1857 | spin_unlock(&p->lock); |
5d337b91 | 1858 | spin_unlock(&swap_lock); |
fb4f88dc | 1859 | |
e1e12d2f | 1860 | set_current_oom_origin(); |
adfab836 | 1861 | err = try_to_unuse(p->type, false, 0); /* force unuse all pages */ |
e1e12d2f | 1862 | clear_current_oom_origin(); |
1da177e4 | 1863 | |
1da177e4 LT |
1864 | if (err) { |
1865 | /* re-insert swap space back into swap_list */ | |
cf0cac0a | 1866 | reinsert_swap_info(p); |
1da177e4 LT |
1867 | goto out_dput; |
1868 | } | |
52b7efdb | 1869 | |
815c2c54 SL |
1870 | flush_work(&p->discard_work); |
1871 | ||
5d337b91 | 1872 | destroy_swap_extents(p); |
570a335b HD |
1873 | if (p->flags & SWP_CONTINUED) |
1874 | free_swap_count_continuations(p); | |
1875 | ||
fc0abb14 | 1876 | mutex_lock(&swapon_mutex); |
5d337b91 | 1877 | spin_lock(&swap_lock); |
ec8acf20 | 1878 | spin_lock(&p->lock); |
5d337b91 HD |
1879 | drain_mmlist(); |
1880 | ||
52b7efdb | 1881 | /* wait for anyone still in scan_swap_map */ |
52b7efdb HD |
1882 | p->highest_bit = 0; /* cuts scans short */ |
1883 | while (p->flags >= SWP_SCANNING) { | |
ec8acf20 | 1884 | spin_unlock(&p->lock); |
5d337b91 | 1885 | spin_unlock(&swap_lock); |
13e4b57f | 1886 | schedule_timeout_uninterruptible(1); |
5d337b91 | 1887 | spin_lock(&swap_lock); |
ec8acf20 | 1888 | spin_lock(&p->lock); |
52b7efdb | 1889 | } |
52b7efdb | 1890 | |
1da177e4 | 1891 | swap_file = p->swap_file; |
5b808a23 | 1892 | old_block_size = p->old_block_size; |
1da177e4 LT |
1893 | p->swap_file = NULL; |
1894 | p->max = 0; | |
1895 | swap_map = p->swap_map; | |
1896 | p->swap_map = NULL; | |
2a8f9449 SL |
1897 | cluster_info = p->cluster_info; |
1898 | p->cluster_info = NULL; | |
4f89849d | 1899 | frontswap_map = frontswap_map_get(p); |
ec8acf20 | 1900 | spin_unlock(&p->lock); |
5d337b91 | 1901 | spin_unlock(&swap_lock); |
adfab836 | 1902 | frontswap_invalidate_area(p->type); |
58e97ba6 | 1903 | frontswap_map_set(p, NULL); |
fc0abb14 | 1904 | mutex_unlock(&swapon_mutex); |
ebc2a1a6 SL |
1905 | free_percpu(p->percpu_cluster); |
1906 | p->percpu_cluster = NULL; | |
1da177e4 | 1907 | vfree(swap_map); |
2a8f9449 | 1908 | vfree(cluster_info); |
4f89849d | 1909 | vfree(frontswap_map); |
2de1a7e4 | 1910 | /* Destroy swap account information */ |
adfab836 | 1911 | swap_cgroup_swapoff(p->type); |
27a7faa0 | 1912 | |
1da177e4 LT |
1913 | inode = mapping->host; |
1914 | if (S_ISBLK(inode->i_mode)) { | |
1915 | struct block_device *bdev = I_BDEV(inode); | |
5b808a23 | 1916 | set_blocksize(bdev, old_block_size); |
e525fd89 | 1917 | blkdev_put(bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL); |
1da177e4 | 1918 | } else { |
1b1dcc1b | 1919 | mutex_lock(&inode->i_mutex); |
1da177e4 | 1920 | inode->i_flags &= ~S_SWAPFILE; |
1b1dcc1b | 1921 | mutex_unlock(&inode->i_mutex); |
1da177e4 LT |
1922 | } |
1923 | filp_close(swap_file, NULL); | |
f893ab41 WY |
1924 | |
1925 | /* | |
1926 | * Clear the SWP_USED flag after all resources are freed so that swapon | |
1927 | * can reuse this swap_info in alloc_swap_info() safely. It is ok to | |
1928 | * not hold p->lock after we cleared its SWP_WRITEOK. | |
1929 | */ | |
1930 | spin_lock(&swap_lock); | |
1931 | p->flags = 0; | |
1932 | spin_unlock(&swap_lock); | |
1933 | ||
1da177e4 | 1934 | err = 0; |
66d7dd51 KS |
1935 | atomic_inc(&proc_poll_event); |
1936 | wake_up_interruptible(&proc_poll_wait); | |
1da177e4 LT |
1937 | |
1938 | out_dput: | |
1939 | filp_close(victim, NULL); | |
1940 | out: | |
f58b59c1 | 1941 | putname(pathname); |
1da177e4 LT |
1942 | return err; |
1943 | } | |
1944 | ||
1945 | #ifdef CONFIG_PROC_FS | |
66d7dd51 KS |
1946 | static unsigned swaps_poll(struct file *file, poll_table *wait) |
1947 | { | |
f1514638 | 1948 | struct seq_file *seq = file->private_data; |
66d7dd51 KS |
1949 | |
1950 | poll_wait(file, &proc_poll_wait, wait); | |
1951 | ||
f1514638 KS |
1952 | if (seq->poll_event != atomic_read(&proc_poll_event)) { |
1953 | seq->poll_event = atomic_read(&proc_poll_event); | |
66d7dd51 KS |
1954 | return POLLIN | POLLRDNORM | POLLERR | POLLPRI; |
1955 | } | |
1956 | ||
1957 | return POLLIN | POLLRDNORM; | |
1958 | } | |
1959 | ||
1da177e4 LT |
1960 | /* iterator */ |
1961 | static void *swap_start(struct seq_file *swap, loff_t *pos) | |
1962 | { | |
efa90a98 HD |
1963 | struct swap_info_struct *si; |
1964 | int type; | |
1da177e4 LT |
1965 | loff_t l = *pos; |
1966 | ||
fc0abb14 | 1967 | mutex_lock(&swapon_mutex); |
1da177e4 | 1968 | |
881e4aab SS |
1969 | if (!l) |
1970 | return SEQ_START_TOKEN; | |
1971 | ||
efa90a98 HD |
1972 | for (type = 0; type < nr_swapfiles; type++) { |
1973 | smp_rmb(); /* read nr_swapfiles before swap_info[type] */ | |
1974 | si = swap_info[type]; | |
1975 | if (!(si->flags & SWP_USED) || !si->swap_map) | |
1da177e4 | 1976 | continue; |
881e4aab | 1977 | if (!--l) |
efa90a98 | 1978 | return si; |
1da177e4 LT |
1979 | } |
1980 | ||
1981 | return NULL; | |
1982 | } | |
1983 | ||
1984 | static void *swap_next(struct seq_file *swap, void *v, loff_t *pos) | |
1985 | { | |
efa90a98 HD |
1986 | struct swap_info_struct *si = v; |
1987 | int type; | |
1da177e4 | 1988 | |
881e4aab | 1989 | if (v == SEQ_START_TOKEN) |
efa90a98 HD |
1990 | type = 0; |
1991 | else | |
1992 | type = si->type + 1; | |
881e4aab | 1993 | |
efa90a98 HD |
1994 | for (; type < nr_swapfiles; type++) { |
1995 | smp_rmb(); /* read nr_swapfiles before swap_info[type] */ | |
1996 | si = swap_info[type]; | |
1997 | if (!(si->flags & SWP_USED) || !si->swap_map) | |
1da177e4 LT |
1998 | continue; |
1999 | ++*pos; | |
efa90a98 | 2000 | return si; |
1da177e4 LT |
2001 | } |
2002 | ||
2003 | return NULL; | |
2004 | } | |
2005 | ||
2006 | static void swap_stop(struct seq_file *swap, void *v) | |
2007 | { | |
fc0abb14 | 2008 | mutex_unlock(&swapon_mutex); |
1da177e4 LT |
2009 | } |
2010 | ||
2011 | static int swap_show(struct seq_file *swap, void *v) | |
2012 | { | |
efa90a98 | 2013 | struct swap_info_struct *si = v; |
1da177e4 LT |
2014 | struct file *file; |
2015 | int len; | |
2016 | ||
efa90a98 | 2017 | if (si == SEQ_START_TOKEN) { |
881e4aab SS |
2018 | seq_puts(swap,"Filename\t\t\t\tType\t\tSize\tUsed\tPriority\n"); |
2019 | return 0; | |
2020 | } | |
1da177e4 | 2021 | |
efa90a98 | 2022 | file = si->swap_file; |
c32c2f63 | 2023 | len = seq_path(swap, &file->f_path, " \t\n\\"); |
6eb396dc | 2024 | seq_printf(swap, "%*s%s\t%u\t%u\t%d\n", |
886bb7e9 | 2025 | len < 40 ? 40 - len : 1, " ", |
496ad9aa | 2026 | S_ISBLK(file_inode(file)->i_mode) ? |
1da177e4 | 2027 | "partition" : "file\t", |
efa90a98 HD |
2028 | si->pages << (PAGE_SHIFT - 10), |
2029 | si->inuse_pages << (PAGE_SHIFT - 10), | |
2030 | si->prio); | |
1da177e4 LT |
2031 | return 0; |
2032 | } | |
2033 | ||
15ad7cdc | 2034 | static const struct seq_operations swaps_op = { |
1da177e4 LT |
2035 | .start = swap_start, |
2036 | .next = swap_next, | |
2037 | .stop = swap_stop, | |
2038 | .show = swap_show | |
2039 | }; | |
2040 | ||
2041 | static int swaps_open(struct inode *inode, struct file *file) | |
2042 | { | |
f1514638 | 2043 | struct seq_file *seq; |
66d7dd51 KS |
2044 | int ret; |
2045 | ||
66d7dd51 | 2046 | ret = seq_open(file, &swaps_op); |
f1514638 | 2047 | if (ret) |
66d7dd51 | 2048 | return ret; |
66d7dd51 | 2049 | |
f1514638 KS |
2050 | seq = file->private_data; |
2051 | seq->poll_event = atomic_read(&proc_poll_event); | |
2052 | return 0; | |
1da177e4 LT |
2053 | } |
2054 | ||
15ad7cdc | 2055 | static const struct file_operations proc_swaps_operations = { |
1da177e4 LT |
2056 | .open = swaps_open, |
2057 | .read = seq_read, | |
2058 | .llseek = seq_lseek, | |
2059 | .release = seq_release, | |
66d7dd51 | 2060 | .poll = swaps_poll, |
1da177e4 LT |
2061 | }; |
2062 | ||
2063 | static int __init procswaps_init(void) | |
2064 | { | |
3d71f86f | 2065 | proc_create("swaps", 0, NULL, &proc_swaps_operations); |
1da177e4 LT |
2066 | return 0; |
2067 | } | |
2068 | __initcall(procswaps_init); | |
2069 | #endif /* CONFIG_PROC_FS */ | |
2070 | ||
1796316a JB |
2071 | #ifdef MAX_SWAPFILES_CHECK |
2072 | static int __init max_swapfiles_check(void) | |
2073 | { | |
2074 | MAX_SWAPFILES_CHECK(); | |
2075 | return 0; | |
2076 | } | |
2077 | late_initcall(max_swapfiles_check); | |
2078 | #endif | |
2079 | ||
53cbb243 | 2080 | static struct swap_info_struct *alloc_swap_info(void) |
1da177e4 | 2081 | { |
73c34b6a | 2082 | struct swap_info_struct *p; |
1da177e4 | 2083 | unsigned int type; |
efa90a98 HD |
2084 | |
2085 | p = kzalloc(sizeof(*p), GFP_KERNEL); | |
2086 | if (!p) | |
53cbb243 | 2087 | return ERR_PTR(-ENOMEM); |
efa90a98 | 2088 | |
5d337b91 | 2089 | spin_lock(&swap_lock); |
efa90a98 HD |
2090 | for (type = 0; type < nr_swapfiles; type++) { |
2091 | if (!(swap_info[type]->flags & SWP_USED)) | |
1da177e4 | 2092 | break; |
efa90a98 | 2093 | } |
0697212a | 2094 | if (type >= MAX_SWAPFILES) { |
5d337b91 | 2095 | spin_unlock(&swap_lock); |
efa90a98 | 2096 | kfree(p); |
730c0581 | 2097 | return ERR_PTR(-EPERM); |
1da177e4 | 2098 | } |
efa90a98 HD |
2099 | if (type >= nr_swapfiles) { |
2100 | p->type = type; | |
2101 | swap_info[type] = p; | |
2102 | /* | |
2103 | * Write swap_info[type] before nr_swapfiles, in case a | |
2104 | * racing procfs swap_start() or swap_next() is reading them. | |
2105 | * (We never shrink nr_swapfiles, we never free this entry.) | |
2106 | */ | |
2107 | smp_wmb(); | |
2108 | nr_swapfiles++; | |
2109 | } else { | |
2110 | kfree(p); | |
2111 | p = swap_info[type]; | |
2112 | /* | |
2113 | * Do not memset this entry: a racing procfs swap_next() | |
2114 | * would be relying on p->type to remain valid. | |
2115 | */ | |
2116 | } | |
9625a5f2 | 2117 | INIT_LIST_HEAD(&p->first_swap_extent.list); |
adfab836 | 2118 | INIT_LIST_HEAD(&p->list); |
1da177e4 | 2119 | p->flags = SWP_USED; |
5d337b91 | 2120 | spin_unlock(&swap_lock); |
ec8acf20 | 2121 | spin_lock_init(&p->lock); |
efa90a98 | 2122 | |
53cbb243 | 2123 | return p; |
53cbb243 CEB |
2124 | } |
2125 | ||
4d0e1e10 CEB |
2126 | static int claim_swapfile(struct swap_info_struct *p, struct inode *inode) |
2127 | { | |
2128 | int error; | |
2129 | ||
2130 | if (S_ISBLK(inode->i_mode)) { | |
2131 | p->bdev = bdgrab(I_BDEV(inode)); | |
2132 | error = blkdev_get(p->bdev, | |
2133 | FMODE_READ | FMODE_WRITE | FMODE_EXCL, | |
2134 | sys_swapon); | |
2135 | if (error < 0) { | |
2136 | p->bdev = NULL; | |
87ade72a | 2137 | return -EINVAL; |
4d0e1e10 CEB |
2138 | } |
2139 | p->old_block_size = block_size(p->bdev); | |
2140 | error = set_blocksize(p->bdev, PAGE_SIZE); | |
2141 | if (error < 0) | |
87ade72a | 2142 | return error; |
4d0e1e10 CEB |
2143 | p->flags |= SWP_BLKDEV; |
2144 | } else if (S_ISREG(inode->i_mode)) { | |
2145 | p->bdev = inode->i_sb->s_bdev; | |
2146 | mutex_lock(&inode->i_mutex); | |
87ade72a CEB |
2147 | if (IS_SWAPFILE(inode)) |
2148 | return -EBUSY; | |
2149 | } else | |
2150 | return -EINVAL; | |
4d0e1e10 CEB |
2151 | |
2152 | return 0; | |
4d0e1e10 CEB |
2153 | } |
2154 | ||
ca8bd38b CEB |
2155 | static unsigned long read_swap_header(struct swap_info_struct *p, |
2156 | union swap_header *swap_header, | |
2157 | struct inode *inode) | |
2158 | { | |
2159 | int i; | |
2160 | unsigned long maxpages; | |
2161 | unsigned long swapfilepages; | |
d6bbbd29 | 2162 | unsigned long last_page; |
ca8bd38b CEB |
2163 | |
2164 | if (memcmp("SWAPSPACE2", swap_header->magic.magic, 10)) { | |
465c47fd | 2165 | pr_err("Unable to find swap-space signature\n"); |
38719025 | 2166 | return 0; |
ca8bd38b CEB |
2167 | } |
2168 | ||
2169 | /* swap partition endianess hack... */ | |
2170 | if (swab32(swap_header->info.version) == 1) { | |
2171 | swab32s(&swap_header->info.version); | |
2172 | swab32s(&swap_header->info.last_page); | |
2173 | swab32s(&swap_header->info.nr_badpages); | |
2174 | for (i = 0; i < swap_header->info.nr_badpages; i++) | |
2175 | swab32s(&swap_header->info.badpages[i]); | |
2176 | } | |
2177 | /* Check the swap header's sub-version */ | |
2178 | if (swap_header->info.version != 1) { | |
465c47fd AM |
2179 | pr_warn("Unable to handle swap header version %d\n", |
2180 | swap_header->info.version); | |
38719025 | 2181 | return 0; |
ca8bd38b CEB |
2182 | } |
2183 | ||
2184 | p->lowest_bit = 1; | |
2185 | p->cluster_next = 1; | |
2186 | p->cluster_nr = 0; | |
2187 | ||
2188 | /* | |
2189 | * Find out how many pages are allowed for a single swap | |
9b15b817 | 2190 | * device. There are two limiting factors: 1) the number |
a2c16d6c HD |
2191 | * of bits for the swap offset in the swp_entry_t type, and |
2192 | * 2) the number of bits in the swap pte as defined by the | |
9b15b817 | 2193 | * different architectures. In order to find the |
a2c16d6c | 2194 | * largest possible bit mask, a swap entry with swap type 0 |
ca8bd38b | 2195 | * and swap offset ~0UL is created, encoded to a swap pte, |
a2c16d6c | 2196 | * decoded to a swp_entry_t again, and finally the swap |
ca8bd38b CEB |
2197 | * offset is extracted. This will mask all the bits from |
2198 | * the initial ~0UL mask that can't be encoded in either | |
2199 | * the swp_entry_t or the architecture definition of a | |
9b15b817 | 2200 | * swap pte. |
ca8bd38b CEB |
2201 | */ |
2202 | maxpages = swp_offset(pte_to_swp_entry( | |
9b15b817 | 2203 | swp_entry_to_pte(swp_entry(0, ~0UL)))) + 1; |
d6bbbd29 RJ |
2204 | last_page = swap_header->info.last_page; |
2205 | if (last_page > maxpages) { | |
465c47fd | 2206 | pr_warn("Truncating oversized swap area, only using %luk out of %luk\n", |
d6bbbd29 RJ |
2207 | maxpages << (PAGE_SHIFT - 10), |
2208 | last_page << (PAGE_SHIFT - 10)); | |
2209 | } | |
2210 | if (maxpages > last_page) { | |
2211 | maxpages = last_page + 1; | |
ca8bd38b CEB |
2212 | /* p->max is an unsigned int: don't overflow it */ |
2213 | if ((unsigned int)maxpages == 0) | |
2214 | maxpages = UINT_MAX; | |
2215 | } | |
2216 | p->highest_bit = maxpages - 1; | |
2217 | ||
2218 | if (!maxpages) | |
38719025 | 2219 | return 0; |
ca8bd38b CEB |
2220 | swapfilepages = i_size_read(inode) >> PAGE_SHIFT; |
2221 | if (swapfilepages && maxpages > swapfilepages) { | |
465c47fd | 2222 | pr_warn("Swap area shorter than signature indicates\n"); |
38719025 | 2223 | return 0; |
ca8bd38b CEB |
2224 | } |
2225 | if (swap_header->info.nr_badpages && S_ISREG(inode->i_mode)) | |
38719025 | 2226 | return 0; |
ca8bd38b | 2227 | if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES) |
38719025 | 2228 | return 0; |
ca8bd38b CEB |
2229 | |
2230 | return maxpages; | |
ca8bd38b CEB |
2231 | } |
2232 | ||
915d4d7b CEB |
2233 | static int setup_swap_map_and_extents(struct swap_info_struct *p, |
2234 | union swap_header *swap_header, | |
2235 | unsigned char *swap_map, | |
2a8f9449 | 2236 | struct swap_cluster_info *cluster_info, |
915d4d7b CEB |
2237 | unsigned long maxpages, |
2238 | sector_t *span) | |
2239 | { | |
2240 | int i; | |
915d4d7b CEB |
2241 | unsigned int nr_good_pages; |
2242 | int nr_extents; | |
2a8f9449 SL |
2243 | unsigned long nr_clusters = DIV_ROUND_UP(maxpages, SWAPFILE_CLUSTER); |
2244 | unsigned long idx = p->cluster_next / SWAPFILE_CLUSTER; | |
915d4d7b CEB |
2245 | |
2246 | nr_good_pages = maxpages - 1; /* omit header page */ | |
2247 | ||
2a8f9449 SL |
2248 | cluster_set_null(&p->free_cluster_head); |
2249 | cluster_set_null(&p->free_cluster_tail); | |
815c2c54 SL |
2250 | cluster_set_null(&p->discard_cluster_head); |
2251 | cluster_set_null(&p->discard_cluster_tail); | |
2a8f9449 | 2252 | |
915d4d7b CEB |
2253 | for (i = 0; i < swap_header->info.nr_badpages; i++) { |
2254 | unsigned int page_nr = swap_header->info.badpages[i]; | |
bdb8e3f6 CEB |
2255 | if (page_nr == 0 || page_nr > swap_header->info.last_page) |
2256 | return -EINVAL; | |
915d4d7b CEB |
2257 | if (page_nr < maxpages) { |
2258 | swap_map[page_nr] = SWAP_MAP_BAD; | |
2259 | nr_good_pages--; | |
2a8f9449 SL |
2260 | /* |
2261 | * Haven't marked the cluster free yet, no list | |
2262 | * operation involved | |
2263 | */ | |
2264 | inc_cluster_info_page(p, cluster_info, page_nr); | |
915d4d7b CEB |
2265 | } |
2266 | } | |
2267 | ||
2a8f9449 SL |
2268 | /* Haven't marked the cluster free yet, no list operation involved */ |
2269 | for (i = maxpages; i < round_up(maxpages, SWAPFILE_CLUSTER); i++) | |
2270 | inc_cluster_info_page(p, cluster_info, i); | |
2271 | ||
915d4d7b CEB |
2272 | if (nr_good_pages) { |
2273 | swap_map[0] = SWAP_MAP_BAD; | |
2a8f9449 SL |
2274 | /* |
2275 | * Not mark the cluster free yet, no list | |
2276 | * operation involved | |
2277 | */ | |
2278 | inc_cluster_info_page(p, cluster_info, 0); | |
915d4d7b CEB |
2279 | p->max = maxpages; |
2280 | p->pages = nr_good_pages; | |
2281 | nr_extents = setup_swap_extents(p, span); | |
bdb8e3f6 CEB |
2282 | if (nr_extents < 0) |
2283 | return nr_extents; | |
915d4d7b CEB |
2284 | nr_good_pages = p->pages; |
2285 | } | |
2286 | if (!nr_good_pages) { | |
465c47fd | 2287 | pr_warn("Empty swap-file\n"); |
bdb8e3f6 | 2288 | return -EINVAL; |
915d4d7b CEB |
2289 | } |
2290 | ||
2a8f9449 SL |
2291 | if (!cluster_info) |
2292 | return nr_extents; | |
2293 | ||
2294 | for (i = 0; i < nr_clusters; i++) { | |
2295 | if (!cluster_count(&cluster_info[idx])) { | |
2296 | cluster_set_flag(&cluster_info[idx], CLUSTER_FLAG_FREE); | |
2297 | if (cluster_is_null(&p->free_cluster_head)) { | |
2298 | cluster_set_next_flag(&p->free_cluster_head, | |
2299 | idx, 0); | |
2300 | cluster_set_next_flag(&p->free_cluster_tail, | |
2301 | idx, 0); | |
2302 | } else { | |
2303 | unsigned int tail; | |
2304 | ||
2305 | tail = cluster_next(&p->free_cluster_tail); | |
2306 | cluster_set_next(&cluster_info[tail], idx); | |
2307 | cluster_set_next_flag(&p->free_cluster_tail, | |
2308 | idx, 0); | |
2309 | } | |
2310 | } | |
2311 | idx++; | |
2312 | if (idx == nr_clusters) | |
2313 | idx = 0; | |
2314 | } | |
915d4d7b | 2315 | return nr_extents; |
915d4d7b CEB |
2316 | } |
2317 | ||
dcf6b7dd RA |
2318 | /* |
2319 | * Helper to sys_swapon determining if a given swap | |
2320 | * backing device queue supports DISCARD operations. | |
2321 | */ | |
2322 | static bool swap_discardable(struct swap_info_struct *si) | |
2323 | { | |
2324 | struct request_queue *q = bdev_get_queue(si->bdev); | |
2325 | ||
2326 | if (!q || !blk_queue_discard(q)) | |
2327 | return false; | |
2328 | ||
2329 | return true; | |
2330 | } | |
2331 | ||
53cbb243 CEB |
2332 | SYSCALL_DEFINE2(swapon, const char __user *, specialfile, int, swap_flags) |
2333 | { | |
2334 | struct swap_info_struct *p; | |
91a27b2a | 2335 | struct filename *name; |
53cbb243 CEB |
2336 | struct file *swap_file = NULL; |
2337 | struct address_space *mapping; | |
40531542 CEB |
2338 | int i; |
2339 | int prio; | |
53cbb243 CEB |
2340 | int error; |
2341 | union swap_header *swap_header; | |
915d4d7b | 2342 | int nr_extents; |
53cbb243 CEB |
2343 | sector_t span; |
2344 | unsigned long maxpages; | |
53cbb243 | 2345 | unsigned char *swap_map = NULL; |
2a8f9449 | 2346 | struct swap_cluster_info *cluster_info = NULL; |
38b5faf4 | 2347 | unsigned long *frontswap_map = NULL; |
53cbb243 CEB |
2348 | struct page *page = NULL; |
2349 | struct inode *inode = NULL; | |
53cbb243 | 2350 | |
d15cab97 HD |
2351 | if (swap_flags & ~SWAP_FLAGS_VALID) |
2352 | return -EINVAL; | |
2353 | ||
53cbb243 CEB |
2354 | if (!capable(CAP_SYS_ADMIN)) |
2355 | return -EPERM; | |
2356 | ||
2357 | p = alloc_swap_info(); | |
2542e513 CEB |
2358 | if (IS_ERR(p)) |
2359 | return PTR_ERR(p); | |
53cbb243 | 2360 | |
815c2c54 SL |
2361 | INIT_WORK(&p->discard_work, swap_discard_work); |
2362 | ||
1da177e4 | 2363 | name = getname(specialfile); |
1da177e4 | 2364 | if (IS_ERR(name)) { |
7de7fb6b | 2365 | error = PTR_ERR(name); |
1da177e4 | 2366 | name = NULL; |
bd69010b | 2367 | goto bad_swap; |
1da177e4 | 2368 | } |
669abf4e | 2369 | swap_file = file_open_name(name, O_RDWR|O_LARGEFILE, 0); |
1da177e4 | 2370 | if (IS_ERR(swap_file)) { |
7de7fb6b | 2371 | error = PTR_ERR(swap_file); |
1da177e4 | 2372 | swap_file = NULL; |
bd69010b | 2373 | goto bad_swap; |
1da177e4 LT |
2374 | } |
2375 | ||
2376 | p->swap_file = swap_file; | |
2377 | mapping = swap_file->f_mapping; | |
1da177e4 | 2378 | |
1da177e4 | 2379 | for (i = 0; i < nr_swapfiles; i++) { |
efa90a98 | 2380 | struct swap_info_struct *q = swap_info[i]; |
1da177e4 | 2381 | |
e8e6c2ec | 2382 | if (q == p || !q->swap_file) |
1da177e4 | 2383 | continue; |
7de7fb6b CEB |
2384 | if (mapping == q->swap_file->f_mapping) { |
2385 | error = -EBUSY; | |
1da177e4 | 2386 | goto bad_swap; |
7de7fb6b | 2387 | } |
1da177e4 LT |
2388 | } |
2389 | ||
2130781e CEB |
2390 | inode = mapping->host; |
2391 | /* If S_ISREG(inode->i_mode) will do mutex_lock(&inode->i_mutex); */ | |
4d0e1e10 CEB |
2392 | error = claim_swapfile(p, inode); |
2393 | if (unlikely(error)) | |
1da177e4 | 2394 | goto bad_swap; |
1da177e4 | 2395 | |
1da177e4 LT |
2396 | /* |
2397 | * Read the swap header. | |
2398 | */ | |
2399 | if (!mapping->a_ops->readpage) { | |
2400 | error = -EINVAL; | |
2401 | goto bad_swap; | |
2402 | } | |
090d2b18 | 2403 | page = read_mapping_page(mapping, 0, swap_file); |
1da177e4 LT |
2404 | if (IS_ERR(page)) { |
2405 | error = PTR_ERR(page); | |
2406 | goto bad_swap; | |
2407 | } | |
81e33971 | 2408 | swap_header = kmap(page); |
1da177e4 | 2409 | |
ca8bd38b CEB |
2410 | maxpages = read_swap_header(p, swap_header, inode); |
2411 | if (unlikely(!maxpages)) { | |
1da177e4 LT |
2412 | error = -EINVAL; |
2413 | goto bad_swap; | |
2414 | } | |
886bb7e9 | 2415 | |
81e33971 | 2416 | /* OK, set up the swap map and apply the bad block list */ |
803d0c83 | 2417 | swap_map = vzalloc(maxpages); |
81e33971 HD |
2418 | if (!swap_map) { |
2419 | error = -ENOMEM; | |
2420 | goto bad_swap; | |
2421 | } | |
2a8f9449 SL |
2422 | if (p->bdev && blk_queue_nonrot(bdev_get_queue(p->bdev))) { |
2423 | p->flags |= SWP_SOLIDSTATE; | |
2424 | /* | |
2425 | * select a random position to start with to help wear leveling | |
2426 | * SSD | |
2427 | */ | |
2428 | p->cluster_next = 1 + (prandom_u32() % p->highest_bit); | |
2429 | ||
2430 | cluster_info = vzalloc(DIV_ROUND_UP(maxpages, | |
2431 | SWAPFILE_CLUSTER) * sizeof(*cluster_info)); | |
2432 | if (!cluster_info) { | |
2433 | error = -ENOMEM; | |
2434 | goto bad_swap; | |
2435 | } | |
ebc2a1a6 SL |
2436 | p->percpu_cluster = alloc_percpu(struct percpu_cluster); |
2437 | if (!p->percpu_cluster) { | |
2438 | error = -ENOMEM; | |
2439 | goto bad_swap; | |
2440 | } | |
2441 | for_each_possible_cpu(i) { | |
2442 | struct percpu_cluster *cluster; | |
2443 | cluster = per_cpu_ptr(p->percpu_cluster, i); | |
2444 | cluster_set_null(&cluster->index); | |
2445 | } | |
2a8f9449 | 2446 | } |
1da177e4 | 2447 | |
1421ef3c CEB |
2448 | error = swap_cgroup_swapon(p->type, maxpages); |
2449 | if (error) | |
2450 | goto bad_swap; | |
2451 | ||
915d4d7b | 2452 | nr_extents = setup_swap_map_and_extents(p, swap_header, swap_map, |
2a8f9449 | 2453 | cluster_info, maxpages, &span); |
915d4d7b CEB |
2454 | if (unlikely(nr_extents < 0)) { |
2455 | error = nr_extents; | |
1da177e4 LT |
2456 | goto bad_swap; |
2457 | } | |
38b5faf4 DM |
2458 | /* frontswap enabled? set up bit-per-page map for frontswap */ |
2459 | if (frontswap_enabled) | |
7b57976d | 2460 | frontswap_map = vzalloc(BITS_TO_LONGS(maxpages) * sizeof(long)); |
1da177e4 | 2461 | |
2a8f9449 SL |
2462 | if (p->bdev &&(swap_flags & SWAP_FLAG_DISCARD) && swap_discardable(p)) { |
2463 | /* | |
2464 | * When discard is enabled for swap with no particular | |
2465 | * policy flagged, we set all swap discard flags here in | |
2466 | * order to sustain backward compatibility with older | |
2467 | * swapon(8) releases. | |
2468 | */ | |
2469 | p->flags |= (SWP_DISCARDABLE | SWP_AREA_DISCARD | | |
2470 | SWP_PAGE_DISCARD); | |
dcf6b7dd | 2471 | |
2a8f9449 SL |
2472 | /* |
2473 | * By flagging sys_swapon, a sysadmin can tell us to | |
2474 | * either do single-time area discards only, or to just | |
2475 | * perform discards for released swap page-clusters. | |
2476 | * Now it's time to adjust the p->flags accordingly. | |
2477 | */ | |
2478 | if (swap_flags & SWAP_FLAG_DISCARD_ONCE) | |
2479 | p->flags &= ~SWP_PAGE_DISCARD; | |
2480 | else if (swap_flags & SWAP_FLAG_DISCARD_PAGES) | |
2481 | p->flags &= ~SWP_AREA_DISCARD; | |
2482 | ||
2483 | /* issue a swapon-time discard if it's still required */ | |
2484 | if (p->flags & SWP_AREA_DISCARD) { | |
2485 | int err = discard_swap(p); | |
2486 | if (unlikely(err)) | |
2487 | pr_err("swapon: discard_swap(%p): %d\n", | |
2488 | p, err); | |
dcf6b7dd | 2489 | } |
20137a49 | 2490 | } |
6a6ba831 | 2491 | |
fc0abb14 | 2492 | mutex_lock(&swapon_mutex); |
40531542 | 2493 | prio = -1; |
78ecba08 | 2494 | if (swap_flags & SWAP_FLAG_PREFER) |
40531542 | 2495 | prio = |
78ecba08 | 2496 | (swap_flags & SWAP_FLAG_PRIO_MASK) >> SWAP_FLAG_PRIO_SHIFT; |
2a8f9449 | 2497 | enable_swap_info(p, prio, swap_map, cluster_info, frontswap_map); |
c69dbfb8 | 2498 | |
465c47fd | 2499 | pr_info("Adding %uk swap on %s. " |
dcf6b7dd | 2500 | "Priority:%d extents:%d across:%lluk %s%s%s%s%s\n", |
91a27b2a | 2501 | p->pages<<(PAGE_SHIFT-10), name->name, p->prio, |
c69dbfb8 CEB |
2502 | nr_extents, (unsigned long long)span<<(PAGE_SHIFT-10), |
2503 | (p->flags & SWP_SOLIDSTATE) ? "SS" : "", | |
38b5faf4 | 2504 | (p->flags & SWP_DISCARDABLE) ? "D" : "", |
dcf6b7dd RA |
2505 | (p->flags & SWP_AREA_DISCARD) ? "s" : "", |
2506 | (p->flags & SWP_PAGE_DISCARD) ? "c" : "", | |
38b5faf4 | 2507 | (frontswap_map) ? "FS" : ""); |
c69dbfb8 | 2508 | |
fc0abb14 | 2509 | mutex_unlock(&swapon_mutex); |
66d7dd51 KS |
2510 | atomic_inc(&proc_poll_event); |
2511 | wake_up_interruptible(&proc_poll_wait); | |
2512 | ||
9b01c350 CEB |
2513 | if (S_ISREG(inode->i_mode)) |
2514 | inode->i_flags |= S_SWAPFILE; | |
1da177e4 LT |
2515 | error = 0; |
2516 | goto out; | |
2517 | bad_swap: | |
ebc2a1a6 SL |
2518 | free_percpu(p->percpu_cluster); |
2519 | p->percpu_cluster = NULL; | |
bd69010b | 2520 | if (inode && S_ISBLK(inode->i_mode) && p->bdev) { |
f2090d2d CEB |
2521 | set_blocksize(p->bdev, p->old_block_size); |
2522 | blkdev_put(p->bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL); | |
1da177e4 | 2523 | } |
4cd3bb10 | 2524 | destroy_swap_extents(p); |
e8e6c2ec | 2525 | swap_cgroup_swapoff(p->type); |
5d337b91 | 2526 | spin_lock(&swap_lock); |
1da177e4 | 2527 | p->swap_file = NULL; |
1da177e4 | 2528 | p->flags = 0; |
5d337b91 | 2529 | spin_unlock(&swap_lock); |
1da177e4 | 2530 | vfree(swap_map); |
2a8f9449 | 2531 | vfree(cluster_info); |
52c50567 | 2532 | if (swap_file) { |
2130781e | 2533 | if (inode && S_ISREG(inode->i_mode)) { |
52c50567 | 2534 | mutex_unlock(&inode->i_mutex); |
2130781e CEB |
2535 | inode = NULL; |
2536 | } | |
1da177e4 | 2537 | filp_close(swap_file, NULL); |
52c50567 | 2538 | } |
1da177e4 LT |
2539 | out: |
2540 | if (page && !IS_ERR(page)) { | |
2541 | kunmap(page); | |
2542 | page_cache_release(page); | |
2543 | } | |
2544 | if (name) | |
2545 | putname(name); | |
9b01c350 | 2546 | if (inode && S_ISREG(inode->i_mode)) |
1b1dcc1b | 2547 | mutex_unlock(&inode->i_mutex); |
1da177e4 LT |
2548 | return error; |
2549 | } | |
2550 | ||
2551 | void si_swapinfo(struct sysinfo *val) | |
2552 | { | |
efa90a98 | 2553 | unsigned int type; |
1da177e4 LT |
2554 | unsigned long nr_to_be_unused = 0; |
2555 | ||
5d337b91 | 2556 | spin_lock(&swap_lock); |
efa90a98 HD |
2557 | for (type = 0; type < nr_swapfiles; type++) { |
2558 | struct swap_info_struct *si = swap_info[type]; | |
2559 | ||
2560 | if ((si->flags & SWP_USED) && !(si->flags & SWP_WRITEOK)) | |
2561 | nr_to_be_unused += si->inuse_pages; | |
1da177e4 | 2562 | } |
ec8acf20 | 2563 | val->freeswap = atomic_long_read(&nr_swap_pages) + nr_to_be_unused; |
1da177e4 | 2564 | val->totalswap = total_swap_pages + nr_to_be_unused; |
5d337b91 | 2565 | spin_unlock(&swap_lock); |
1da177e4 LT |
2566 | } |
2567 | ||
2568 | /* | |
2569 | * Verify that a swap entry is valid and increment its swap map count. | |
2570 | * | |
355cfa73 KH |
2571 | * Returns error code in following case. |
2572 | * - success -> 0 | |
2573 | * - swp_entry is invalid -> EINVAL | |
2574 | * - swp_entry is migration entry -> EINVAL | |
2575 | * - swap-cache reference is requested but there is already one. -> EEXIST | |
2576 | * - swap-cache reference is requested but the entry is not used. -> ENOENT | |
570a335b | 2577 | * - swap-mapped reference requested but needs continued swap count. -> ENOMEM |
1da177e4 | 2578 | */ |
8d69aaee | 2579 | static int __swap_duplicate(swp_entry_t entry, unsigned char usage) |
1da177e4 | 2580 | { |
73c34b6a | 2581 | struct swap_info_struct *p; |
1da177e4 | 2582 | unsigned long offset, type; |
8d69aaee HD |
2583 | unsigned char count; |
2584 | unsigned char has_cache; | |
253d553b | 2585 | int err = -EINVAL; |
1da177e4 | 2586 | |
a7420aa5 | 2587 | if (non_swap_entry(entry)) |
253d553b | 2588 | goto out; |
0697212a | 2589 | |
1da177e4 LT |
2590 | type = swp_type(entry); |
2591 | if (type >= nr_swapfiles) | |
2592 | goto bad_file; | |
efa90a98 | 2593 | p = swap_info[type]; |
1da177e4 LT |
2594 | offset = swp_offset(entry); |
2595 | ||
ec8acf20 | 2596 | spin_lock(&p->lock); |
355cfa73 KH |
2597 | if (unlikely(offset >= p->max)) |
2598 | goto unlock_out; | |
2599 | ||
253d553b | 2600 | count = p->swap_map[offset]; |
edfe23da SL |
2601 | |
2602 | /* | |
2603 | * swapin_readahead() doesn't check if a swap entry is valid, so the | |
2604 | * swap entry could be SWAP_MAP_BAD. Check here with lock held. | |
2605 | */ | |
2606 | if (unlikely(swap_count(count) == SWAP_MAP_BAD)) { | |
2607 | err = -ENOENT; | |
2608 | goto unlock_out; | |
2609 | } | |
2610 | ||
253d553b HD |
2611 | has_cache = count & SWAP_HAS_CACHE; |
2612 | count &= ~SWAP_HAS_CACHE; | |
2613 | err = 0; | |
355cfa73 | 2614 | |
253d553b | 2615 | if (usage == SWAP_HAS_CACHE) { |
355cfa73 KH |
2616 | |
2617 | /* set SWAP_HAS_CACHE if there is no cache and entry is used */ | |
253d553b HD |
2618 | if (!has_cache && count) |
2619 | has_cache = SWAP_HAS_CACHE; | |
2620 | else if (has_cache) /* someone else added cache */ | |
2621 | err = -EEXIST; | |
2622 | else /* no users remaining */ | |
2623 | err = -ENOENT; | |
355cfa73 KH |
2624 | |
2625 | } else if (count || has_cache) { | |
253d553b | 2626 | |
570a335b HD |
2627 | if ((count & ~COUNT_CONTINUED) < SWAP_MAP_MAX) |
2628 | count += usage; | |
2629 | else if ((count & ~COUNT_CONTINUED) > SWAP_MAP_MAX) | |
253d553b | 2630 | err = -EINVAL; |
570a335b HD |
2631 | else if (swap_count_continued(p, offset, count)) |
2632 | count = COUNT_CONTINUED; | |
2633 | else | |
2634 | err = -ENOMEM; | |
355cfa73 | 2635 | } else |
253d553b HD |
2636 | err = -ENOENT; /* unused swap entry */ |
2637 | ||
2638 | p->swap_map[offset] = count | has_cache; | |
2639 | ||
355cfa73 | 2640 | unlock_out: |
ec8acf20 | 2641 | spin_unlock(&p->lock); |
1da177e4 | 2642 | out: |
253d553b | 2643 | return err; |
1da177e4 LT |
2644 | |
2645 | bad_file: | |
465c47fd | 2646 | pr_err("swap_dup: %s%08lx\n", Bad_file, entry.val); |
1da177e4 LT |
2647 | goto out; |
2648 | } | |
253d553b | 2649 | |
aaa46865 HD |
2650 | /* |
2651 | * Help swapoff by noting that swap entry belongs to shmem/tmpfs | |
2652 | * (in which case its reference count is never incremented). | |
2653 | */ | |
2654 | void swap_shmem_alloc(swp_entry_t entry) | |
2655 | { | |
2656 | __swap_duplicate(entry, SWAP_MAP_SHMEM); | |
2657 | } | |
2658 | ||
355cfa73 | 2659 | /* |
08259d58 HD |
2660 | * Increase reference count of swap entry by 1. |
2661 | * Returns 0 for success, or -ENOMEM if a swap_count_continuation is required | |
2662 | * but could not be atomically allocated. Returns 0, just as if it succeeded, | |
2663 | * if __swap_duplicate() fails for another reason (-EINVAL or -ENOENT), which | |
2664 | * might occur if a page table entry has got corrupted. | |
355cfa73 | 2665 | */ |
570a335b | 2666 | int swap_duplicate(swp_entry_t entry) |
355cfa73 | 2667 | { |
570a335b HD |
2668 | int err = 0; |
2669 | ||
2670 | while (!err && __swap_duplicate(entry, 1) == -ENOMEM) | |
2671 | err = add_swap_count_continuation(entry, GFP_ATOMIC); | |
2672 | return err; | |
355cfa73 | 2673 | } |
1da177e4 | 2674 | |
cb4b86ba | 2675 | /* |
355cfa73 KH |
2676 | * @entry: swap entry for which we allocate swap cache. |
2677 | * | |
73c34b6a | 2678 | * Called when allocating swap cache for existing swap entry, |
355cfa73 KH |
2679 | * This can return error codes. Returns 0 at success. |
2680 | * -EBUSY means there is a swap cache. | |
2681 | * Note: return code is different from swap_duplicate(). | |
cb4b86ba KH |
2682 | */ |
2683 | int swapcache_prepare(swp_entry_t entry) | |
2684 | { | |
253d553b | 2685 | return __swap_duplicate(entry, SWAP_HAS_CACHE); |
cb4b86ba KH |
2686 | } |
2687 | ||
f981c595 MG |
2688 | struct swap_info_struct *page_swap_info(struct page *page) |
2689 | { | |
2690 | swp_entry_t swap = { .val = page_private(page) }; | |
2691 | BUG_ON(!PageSwapCache(page)); | |
2692 | return swap_info[swp_type(swap)]; | |
2693 | } | |
2694 | ||
2695 | /* | |
2696 | * out-of-line __page_file_ methods to avoid include hell. | |
2697 | */ | |
2698 | struct address_space *__page_file_mapping(struct page *page) | |
2699 | { | |
309381fe | 2700 | VM_BUG_ON_PAGE(!PageSwapCache(page), page); |
f981c595 MG |
2701 | return page_swap_info(page)->swap_file->f_mapping; |
2702 | } | |
2703 | EXPORT_SYMBOL_GPL(__page_file_mapping); | |
2704 | ||
2705 | pgoff_t __page_file_index(struct page *page) | |
2706 | { | |
2707 | swp_entry_t swap = { .val = page_private(page) }; | |
309381fe | 2708 | VM_BUG_ON_PAGE(!PageSwapCache(page), page); |
f981c595 MG |
2709 | return swp_offset(swap); |
2710 | } | |
2711 | EXPORT_SYMBOL_GPL(__page_file_index); | |
2712 | ||
570a335b HD |
2713 | /* |
2714 | * add_swap_count_continuation - called when a swap count is duplicated | |
2715 | * beyond SWAP_MAP_MAX, it allocates a new page and links that to the entry's | |
2716 | * page of the original vmalloc'ed swap_map, to hold the continuation count | |
2717 | * (for that entry and for its neighbouring PAGE_SIZE swap entries). Called | |
2718 | * again when count is duplicated beyond SWAP_MAP_MAX * SWAP_CONT_MAX, etc. | |
2719 | * | |
2720 | * These continuation pages are seldom referenced: the common paths all work | |
2721 | * on the original swap_map, only referring to a continuation page when the | |
2722 | * low "digit" of a count is incremented or decremented through SWAP_MAP_MAX. | |
2723 | * | |
2724 | * add_swap_count_continuation(, GFP_ATOMIC) can be called while holding | |
2725 | * page table locks; if it fails, add_swap_count_continuation(, GFP_KERNEL) | |
2726 | * can be called after dropping locks. | |
2727 | */ | |
2728 | int add_swap_count_continuation(swp_entry_t entry, gfp_t gfp_mask) | |
2729 | { | |
2730 | struct swap_info_struct *si; | |
2731 | struct page *head; | |
2732 | struct page *page; | |
2733 | struct page *list_page; | |
2734 | pgoff_t offset; | |
2735 | unsigned char count; | |
2736 | ||
2737 | /* | |
2738 | * When debugging, it's easier to use __GFP_ZERO here; but it's better | |
2739 | * for latency not to zero a page while GFP_ATOMIC and holding locks. | |
2740 | */ | |
2741 | page = alloc_page(gfp_mask | __GFP_HIGHMEM); | |
2742 | ||
2743 | si = swap_info_get(entry); | |
2744 | if (!si) { | |
2745 | /* | |
2746 | * An acceptable race has occurred since the failing | |
2747 | * __swap_duplicate(): the swap entry has been freed, | |
2748 | * perhaps even the whole swap_map cleared for swapoff. | |
2749 | */ | |
2750 | goto outer; | |
2751 | } | |
2752 | ||
2753 | offset = swp_offset(entry); | |
2754 | count = si->swap_map[offset] & ~SWAP_HAS_CACHE; | |
2755 | ||
2756 | if ((count & ~COUNT_CONTINUED) != SWAP_MAP_MAX) { | |
2757 | /* | |
2758 | * The higher the swap count, the more likely it is that tasks | |
2759 | * will race to add swap count continuation: we need to avoid | |
2760 | * over-provisioning. | |
2761 | */ | |
2762 | goto out; | |
2763 | } | |
2764 | ||
2765 | if (!page) { | |
ec8acf20 | 2766 | spin_unlock(&si->lock); |
570a335b HD |
2767 | return -ENOMEM; |
2768 | } | |
2769 | ||
2770 | /* | |
2771 | * We are fortunate that although vmalloc_to_page uses pte_offset_map, | |
2de1a7e4 SJ |
2772 | * no architecture is using highmem pages for kernel page tables: so it |
2773 | * will not corrupt the GFP_ATOMIC caller's atomic page table kmaps. | |
570a335b HD |
2774 | */ |
2775 | head = vmalloc_to_page(si->swap_map + offset); | |
2776 | offset &= ~PAGE_MASK; | |
2777 | ||
2778 | /* | |
2779 | * Page allocation does not initialize the page's lru field, | |
2780 | * but it does always reset its private field. | |
2781 | */ | |
2782 | if (!page_private(head)) { | |
2783 | BUG_ON(count & COUNT_CONTINUED); | |
2784 | INIT_LIST_HEAD(&head->lru); | |
2785 | set_page_private(head, SWP_CONTINUED); | |
2786 | si->flags |= SWP_CONTINUED; | |
2787 | } | |
2788 | ||
2789 | list_for_each_entry(list_page, &head->lru, lru) { | |
2790 | unsigned char *map; | |
2791 | ||
2792 | /* | |
2793 | * If the previous map said no continuation, but we've found | |
2794 | * a continuation page, free our allocation and use this one. | |
2795 | */ | |
2796 | if (!(count & COUNT_CONTINUED)) | |
2797 | goto out; | |
2798 | ||
9b04c5fe | 2799 | map = kmap_atomic(list_page) + offset; |
570a335b | 2800 | count = *map; |
9b04c5fe | 2801 | kunmap_atomic(map); |
570a335b HD |
2802 | |
2803 | /* | |
2804 | * If this continuation count now has some space in it, | |
2805 | * free our allocation and use this one. | |
2806 | */ | |
2807 | if ((count & ~COUNT_CONTINUED) != SWAP_CONT_MAX) | |
2808 | goto out; | |
2809 | } | |
2810 | ||
2811 | list_add_tail(&page->lru, &head->lru); | |
2812 | page = NULL; /* now it's attached, don't free it */ | |
2813 | out: | |
ec8acf20 | 2814 | spin_unlock(&si->lock); |
570a335b HD |
2815 | outer: |
2816 | if (page) | |
2817 | __free_page(page); | |
2818 | return 0; | |
2819 | } | |
2820 | ||
2821 | /* | |
2822 | * swap_count_continued - when the original swap_map count is incremented | |
2823 | * from SWAP_MAP_MAX, check if there is already a continuation page to carry | |
2824 | * into, carry if so, or else fail until a new continuation page is allocated; | |
2825 | * when the original swap_map count is decremented from 0 with continuation, | |
2826 | * borrow from the continuation and report whether it still holds more. | |
2827 | * Called while __swap_duplicate() or swap_entry_free() holds swap_lock. | |
2828 | */ | |
2829 | static bool swap_count_continued(struct swap_info_struct *si, | |
2830 | pgoff_t offset, unsigned char count) | |
2831 | { | |
2832 | struct page *head; | |
2833 | struct page *page; | |
2834 | unsigned char *map; | |
2835 | ||
2836 | head = vmalloc_to_page(si->swap_map + offset); | |
2837 | if (page_private(head) != SWP_CONTINUED) { | |
2838 | BUG_ON(count & COUNT_CONTINUED); | |
2839 | return false; /* need to add count continuation */ | |
2840 | } | |
2841 | ||
2842 | offset &= ~PAGE_MASK; | |
2843 | page = list_entry(head->lru.next, struct page, lru); | |
9b04c5fe | 2844 | map = kmap_atomic(page) + offset; |
570a335b HD |
2845 | |
2846 | if (count == SWAP_MAP_MAX) /* initial increment from swap_map */ | |
2847 | goto init_map; /* jump over SWAP_CONT_MAX checks */ | |
2848 | ||
2849 | if (count == (SWAP_MAP_MAX | COUNT_CONTINUED)) { /* incrementing */ | |
2850 | /* | |
2851 | * Think of how you add 1 to 999 | |
2852 | */ | |
2853 | while (*map == (SWAP_CONT_MAX | COUNT_CONTINUED)) { | |
9b04c5fe | 2854 | kunmap_atomic(map); |
570a335b HD |
2855 | page = list_entry(page->lru.next, struct page, lru); |
2856 | BUG_ON(page == head); | |
9b04c5fe | 2857 | map = kmap_atomic(page) + offset; |
570a335b HD |
2858 | } |
2859 | if (*map == SWAP_CONT_MAX) { | |
9b04c5fe | 2860 | kunmap_atomic(map); |
570a335b HD |
2861 | page = list_entry(page->lru.next, struct page, lru); |
2862 | if (page == head) | |
2863 | return false; /* add count continuation */ | |
9b04c5fe | 2864 | map = kmap_atomic(page) + offset; |
570a335b HD |
2865 | init_map: *map = 0; /* we didn't zero the page */ |
2866 | } | |
2867 | *map += 1; | |
9b04c5fe | 2868 | kunmap_atomic(map); |
570a335b HD |
2869 | page = list_entry(page->lru.prev, struct page, lru); |
2870 | while (page != head) { | |
9b04c5fe | 2871 | map = kmap_atomic(page) + offset; |
570a335b | 2872 | *map = COUNT_CONTINUED; |
9b04c5fe | 2873 | kunmap_atomic(map); |
570a335b HD |
2874 | page = list_entry(page->lru.prev, struct page, lru); |
2875 | } | |
2876 | return true; /* incremented */ | |
2877 | ||
2878 | } else { /* decrementing */ | |
2879 | /* | |
2880 | * Think of how you subtract 1 from 1000 | |
2881 | */ | |
2882 | BUG_ON(count != COUNT_CONTINUED); | |
2883 | while (*map == COUNT_CONTINUED) { | |
9b04c5fe | 2884 | kunmap_atomic(map); |
570a335b HD |
2885 | page = list_entry(page->lru.next, struct page, lru); |
2886 | BUG_ON(page == head); | |
9b04c5fe | 2887 | map = kmap_atomic(page) + offset; |
570a335b HD |
2888 | } |
2889 | BUG_ON(*map == 0); | |
2890 | *map -= 1; | |
2891 | if (*map == 0) | |
2892 | count = 0; | |
9b04c5fe | 2893 | kunmap_atomic(map); |
570a335b HD |
2894 | page = list_entry(page->lru.prev, struct page, lru); |
2895 | while (page != head) { | |
9b04c5fe | 2896 | map = kmap_atomic(page) + offset; |
570a335b HD |
2897 | *map = SWAP_CONT_MAX | count; |
2898 | count = COUNT_CONTINUED; | |
9b04c5fe | 2899 | kunmap_atomic(map); |
570a335b HD |
2900 | page = list_entry(page->lru.prev, struct page, lru); |
2901 | } | |
2902 | return count == COUNT_CONTINUED; | |
2903 | } | |
2904 | } | |
2905 | ||
2906 | /* | |
2907 | * free_swap_count_continuations - swapoff free all the continuation pages | |
2908 | * appended to the swap_map, after swap_map is quiesced, before vfree'ing it. | |
2909 | */ | |
2910 | static void free_swap_count_continuations(struct swap_info_struct *si) | |
2911 | { | |
2912 | pgoff_t offset; | |
2913 | ||
2914 | for (offset = 0; offset < si->max; offset += PAGE_SIZE) { | |
2915 | struct page *head; | |
2916 | head = vmalloc_to_page(si->swap_map + offset); | |
2917 | if (page_private(head)) { | |
2918 | struct list_head *this, *next; | |
2919 | list_for_each_safe(this, next, &head->lru) { | |
2920 | struct page *page; | |
2921 | page = list_entry(this, struct page, lru); | |
2922 | list_del(this); | |
2923 | __free_page(page); | |
2924 | } | |
2925 | } | |
2926 | } | |
2927 | } |