]>
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 | */ | |
18ab4d4c DS |
64 | PLIST_HEAD(swap_active_head); |
65 | ||
66 | /* | |
67 | * all available (active, not full) swap_info_structs | |
68 | * protected with swap_avail_lock, ordered by priority. | |
69 | * This is used by get_swap_page() instead of swap_active_head | |
70 | * because swap_active_head includes all swap_info_structs, | |
71 | * but get_swap_page() doesn't need to look at full ones. | |
72 | * This uses its own lock instead of swap_lock because when a | |
73 | * swap_info_struct changes between not-full/full, it needs to | |
74 | * add/remove itself to/from this list, but the swap_info_struct->lock | |
75 | * is held and the locking order requires swap_lock to be taken | |
76 | * before any swap_info_struct->lock. | |
77 | */ | |
78 | static PLIST_HEAD(swap_avail_head); | |
79 | static DEFINE_SPINLOCK(swap_avail_lock); | |
1da177e4 | 80 | |
38b5faf4 | 81 | struct swap_info_struct *swap_info[MAX_SWAPFILES]; |
1da177e4 | 82 | |
fc0abb14 | 83 | static DEFINE_MUTEX(swapon_mutex); |
1da177e4 | 84 | |
66d7dd51 KS |
85 | static DECLARE_WAIT_QUEUE_HEAD(proc_poll_wait); |
86 | /* Activity counter to indicate that a swapon or swapoff has occurred */ | |
87 | static atomic_t proc_poll_event = ATOMIC_INIT(0); | |
88 | ||
8d69aaee | 89 | static inline unsigned char swap_count(unsigned char ent) |
355cfa73 | 90 | { |
570a335b | 91 | return ent & ~SWAP_HAS_CACHE; /* may include SWAP_HAS_CONT flag */ |
355cfa73 KH |
92 | } |
93 | ||
efa90a98 | 94 | /* returns 1 if swap entry is freed */ |
c9e44410 KH |
95 | static int |
96 | __try_to_reclaim_swap(struct swap_info_struct *si, unsigned long offset) | |
97 | { | |
efa90a98 | 98 | swp_entry_t entry = swp_entry(si->type, offset); |
c9e44410 KH |
99 | struct page *page; |
100 | int ret = 0; | |
101 | ||
33806f06 | 102 | page = find_get_page(swap_address_space(entry), entry.val); |
c9e44410 KH |
103 | if (!page) |
104 | return 0; | |
105 | /* | |
106 | * This function is called from scan_swap_map() and it's called | |
107 | * by vmscan.c at reclaiming pages. So, we hold a lock on a page, here. | |
108 | * We have to use trylock for avoiding deadlock. This is a special | |
109 | * case and you should use try_to_free_swap() with explicit lock_page() | |
110 | * in usual operations. | |
111 | */ | |
112 | if (trylock_page(page)) { | |
113 | ret = try_to_free_swap(page); | |
114 | unlock_page(page); | |
115 | } | |
116 | page_cache_release(page); | |
117 | return ret; | |
118 | } | |
355cfa73 | 119 | |
6a6ba831 HD |
120 | /* |
121 | * swapon tell device that all the old swap contents can be discarded, | |
122 | * to allow the swap device to optimize its wear-levelling. | |
123 | */ | |
124 | static int discard_swap(struct swap_info_struct *si) | |
125 | { | |
126 | struct swap_extent *se; | |
9625a5f2 HD |
127 | sector_t start_block; |
128 | sector_t nr_blocks; | |
6a6ba831 HD |
129 | int err = 0; |
130 | ||
9625a5f2 HD |
131 | /* Do not discard the swap header page! */ |
132 | se = &si->first_swap_extent; | |
133 | start_block = (se->start_block + 1) << (PAGE_SHIFT - 9); | |
134 | nr_blocks = ((sector_t)se->nr_pages - 1) << (PAGE_SHIFT - 9); | |
135 | if (nr_blocks) { | |
136 | err = blkdev_issue_discard(si->bdev, start_block, | |
dd3932ed | 137 | nr_blocks, GFP_KERNEL, 0); |
9625a5f2 HD |
138 | if (err) |
139 | return err; | |
140 | cond_resched(); | |
141 | } | |
6a6ba831 | 142 | |
9625a5f2 HD |
143 | list_for_each_entry(se, &si->first_swap_extent.list, list) { |
144 | start_block = se->start_block << (PAGE_SHIFT - 9); | |
145 | nr_blocks = (sector_t)se->nr_pages << (PAGE_SHIFT - 9); | |
6a6ba831 HD |
146 | |
147 | err = blkdev_issue_discard(si->bdev, start_block, | |
dd3932ed | 148 | nr_blocks, GFP_KERNEL, 0); |
6a6ba831 HD |
149 | if (err) |
150 | break; | |
151 | ||
152 | cond_resched(); | |
153 | } | |
154 | return err; /* That will often be -EOPNOTSUPP */ | |
155 | } | |
156 | ||
7992fde7 HD |
157 | /* |
158 | * swap allocation tell device that a cluster of swap can now be discarded, | |
159 | * to allow the swap device to optimize its wear-levelling. | |
160 | */ | |
161 | static void discard_swap_cluster(struct swap_info_struct *si, | |
162 | pgoff_t start_page, pgoff_t nr_pages) | |
163 | { | |
164 | struct swap_extent *se = si->curr_swap_extent; | |
165 | int found_extent = 0; | |
166 | ||
167 | while (nr_pages) { | |
168 | struct list_head *lh; | |
169 | ||
170 | if (se->start_page <= start_page && | |
171 | start_page < se->start_page + se->nr_pages) { | |
172 | pgoff_t offset = start_page - se->start_page; | |
173 | sector_t start_block = se->start_block + offset; | |
858a2990 | 174 | sector_t nr_blocks = se->nr_pages - offset; |
7992fde7 HD |
175 | |
176 | if (nr_blocks > nr_pages) | |
177 | nr_blocks = nr_pages; | |
178 | start_page += nr_blocks; | |
179 | nr_pages -= nr_blocks; | |
180 | ||
181 | if (!found_extent++) | |
182 | si->curr_swap_extent = se; | |
183 | ||
184 | start_block <<= PAGE_SHIFT - 9; | |
185 | nr_blocks <<= PAGE_SHIFT - 9; | |
186 | if (blkdev_issue_discard(si->bdev, start_block, | |
dd3932ed | 187 | nr_blocks, GFP_NOIO, 0)) |
7992fde7 HD |
188 | break; |
189 | } | |
190 | ||
191 | lh = se->list.next; | |
7992fde7 HD |
192 | se = list_entry(lh, struct swap_extent, list); |
193 | } | |
194 | } | |
195 | ||
048c27fd HD |
196 | #define SWAPFILE_CLUSTER 256 |
197 | #define LATENCY_LIMIT 256 | |
198 | ||
2a8f9449 SL |
199 | static inline void cluster_set_flag(struct swap_cluster_info *info, |
200 | unsigned int flag) | |
201 | { | |
202 | info->flags = flag; | |
203 | } | |
204 | ||
205 | static inline unsigned int cluster_count(struct swap_cluster_info *info) | |
206 | { | |
207 | return info->data; | |
208 | } | |
209 | ||
210 | static inline void cluster_set_count(struct swap_cluster_info *info, | |
211 | unsigned int c) | |
212 | { | |
213 | info->data = c; | |
214 | } | |
215 | ||
216 | static inline void cluster_set_count_flag(struct swap_cluster_info *info, | |
217 | unsigned int c, unsigned int f) | |
218 | { | |
219 | info->flags = f; | |
220 | info->data = c; | |
221 | } | |
222 | ||
223 | static inline unsigned int cluster_next(struct swap_cluster_info *info) | |
224 | { | |
225 | return info->data; | |
226 | } | |
227 | ||
228 | static inline void cluster_set_next(struct swap_cluster_info *info, | |
229 | unsigned int n) | |
230 | { | |
231 | info->data = n; | |
232 | } | |
233 | ||
234 | static inline void cluster_set_next_flag(struct swap_cluster_info *info, | |
235 | unsigned int n, unsigned int f) | |
236 | { | |
237 | info->flags = f; | |
238 | info->data = n; | |
239 | } | |
240 | ||
241 | static inline bool cluster_is_free(struct swap_cluster_info *info) | |
242 | { | |
243 | return info->flags & CLUSTER_FLAG_FREE; | |
244 | } | |
245 | ||
246 | static inline bool cluster_is_null(struct swap_cluster_info *info) | |
247 | { | |
248 | return info->flags & CLUSTER_FLAG_NEXT_NULL; | |
249 | } | |
250 | ||
251 | static inline void cluster_set_null(struct swap_cluster_info *info) | |
252 | { | |
253 | info->flags = CLUSTER_FLAG_NEXT_NULL; | |
254 | info->data = 0; | |
255 | } | |
256 | ||
815c2c54 SL |
257 | /* Add a cluster to discard list and schedule it to do discard */ |
258 | static void swap_cluster_schedule_discard(struct swap_info_struct *si, | |
259 | unsigned int idx) | |
260 | { | |
261 | /* | |
262 | * If scan_swap_map() can't find a free cluster, it will check | |
263 | * si->swap_map directly. To make sure the discarding cluster isn't | |
264 | * taken by scan_swap_map(), mark the swap entries bad (occupied). It | |
265 | * will be cleared after discard | |
266 | */ | |
267 | memset(si->swap_map + idx * SWAPFILE_CLUSTER, | |
268 | SWAP_MAP_BAD, SWAPFILE_CLUSTER); | |
269 | ||
270 | if (cluster_is_null(&si->discard_cluster_head)) { | |
271 | cluster_set_next_flag(&si->discard_cluster_head, | |
272 | idx, 0); | |
273 | cluster_set_next_flag(&si->discard_cluster_tail, | |
274 | idx, 0); | |
275 | } else { | |
276 | unsigned int tail = cluster_next(&si->discard_cluster_tail); | |
277 | cluster_set_next(&si->cluster_info[tail], idx); | |
278 | cluster_set_next_flag(&si->discard_cluster_tail, | |
279 | idx, 0); | |
280 | } | |
281 | ||
282 | schedule_work(&si->discard_work); | |
283 | } | |
284 | ||
285 | /* | |
286 | * Doing discard actually. After a cluster discard is finished, the cluster | |
287 | * will be added to free cluster list. caller should hold si->lock. | |
288 | */ | |
289 | static void swap_do_scheduled_discard(struct swap_info_struct *si) | |
290 | { | |
291 | struct swap_cluster_info *info; | |
292 | unsigned int idx; | |
293 | ||
294 | info = si->cluster_info; | |
295 | ||
296 | while (!cluster_is_null(&si->discard_cluster_head)) { | |
297 | idx = cluster_next(&si->discard_cluster_head); | |
298 | ||
299 | cluster_set_next_flag(&si->discard_cluster_head, | |
300 | cluster_next(&info[idx]), 0); | |
301 | if (cluster_next(&si->discard_cluster_tail) == idx) { | |
302 | cluster_set_null(&si->discard_cluster_head); | |
303 | cluster_set_null(&si->discard_cluster_tail); | |
304 | } | |
305 | spin_unlock(&si->lock); | |
306 | ||
307 | discard_swap_cluster(si, idx * SWAPFILE_CLUSTER, | |
308 | SWAPFILE_CLUSTER); | |
309 | ||
310 | spin_lock(&si->lock); | |
311 | cluster_set_flag(&info[idx], CLUSTER_FLAG_FREE); | |
312 | if (cluster_is_null(&si->free_cluster_head)) { | |
313 | cluster_set_next_flag(&si->free_cluster_head, | |
314 | idx, 0); | |
315 | cluster_set_next_flag(&si->free_cluster_tail, | |
316 | idx, 0); | |
317 | } else { | |
318 | unsigned int tail; | |
319 | ||
320 | tail = cluster_next(&si->free_cluster_tail); | |
321 | cluster_set_next(&info[tail], idx); | |
322 | cluster_set_next_flag(&si->free_cluster_tail, | |
323 | idx, 0); | |
324 | } | |
325 | memset(si->swap_map + idx * SWAPFILE_CLUSTER, | |
326 | 0, SWAPFILE_CLUSTER); | |
327 | } | |
328 | } | |
329 | ||
330 | static void swap_discard_work(struct work_struct *work) | |
331 | { | |
332 | struct swap_info_struct *si; | |
333 | ||
334 | si = container_of(work, struct swap_info_struct, discard_work); | |
335 | ||
336 | spin_lock(&si->lock); | |
337 | swap_do_scheduled_discard(si); | |
338 | spin_unlock(&si->lock); | |
339 | } | |
340 | ||
2a8f9449 SL |
341 | /* |
342 | * The cluster corresponding to page_nr will be used. The cluster will be | |
343 | * removed from free cluster list and its usage counter will be increased. | |
344 | */ | |
345 | static void inc_cluster_info_page(struct swap_info_struct *p, | |
346 | struct swap_cluster_info *cluster_info, unsigned long page_nr) | |
347 | { | |
348 | unsigned long idx = page_nr / SWAPFILE_CLUSTER; | |
349 | ||
350 | if (!cluster_info) | |
351 | return; | |
352 | if (cluster_is_free(&cluster_info[idx])) { | |
353 | VM_BUG_ON(cluster_next(&p->free_cluster_head) != idx); | |
354 | cluster_set_next_flag(&p->free_cluster_head, | |
355 | cluster_next(&cluster_info[idx]), 0); | |
356 | if (cluster_next(&p->free_cluster_tail) == idx) { | |
357 | cluster_set_null(&p->free_cluster_tail); | |
358 | cluster_set_null(&p->free_cluster_head); | |
359 | } | |
360 | cluster_set_count_flag(&cluster_info[idx], 0, 0); | |
361 | } | |
362 | ||
363 | VM_BUG_ON(cluster_count(&cluster_info[idx]) >= SWAPFILE_CLUSTER); | |
364 | cluster_set_count(&cluster_info[idx], | |
365 | cluster_count(&cluster_info[idx]) + 1); | |
366 | } | |
367 | ||
368 | /* | |
369 | * The cluster corresponding to page_nr decreases one usage. If the usage | |
370 | * counter becomes 0, which means no page in the cluster is in using, we can | |
371 | * optionally discard the cluster and add it to free cluster list. | |
372 | */ | |
373 | static void dec_cluster_info_page(struct swap_info_struct *p, | |
374 | struct swap_cluster_info *cluster_info, unsigned long page_nr) | |
375 | { | |
376 | unsigned long idx = page_nr / SWAPFILE_CLUSTER; | |
377 | ||
378 | if (!cluster_info) | |
379 | return; | |
380 | ||
381 | VM_BUG_ON(cluster_count(&cluster_info[idx]) == 0); | |
382 | cluster_set_count(&cluster_info[idx], | |
383 | cluster_count(&cluster_info[idx]) - 1); | |
384 | ||
385 | if (cluster_count(&cluster_info[idx]) == 0) { | |
815c2c54 SL |
386 | /* |
387 | * If the swap is discardable, prepare discard the cluster | |
388 | * instead of free it immediately. The cluster will be freed | |
389 | * after discard. | |
390 | */ | |
edfe23da SL |
391 | if ((p->flags & (SWP_WRITEOK | SWP_PAGE_DISCARD)) == |
392 | (SWP_WRITEOK | SWP_PAGE_DISCARD)) { | |
815c2c54 SL |
393 | swap_cluster_schedule_discard(p, idx); |
394 | return; | |
395 | } | |
396 | ||
2a8f9449 SL |
397 | cluster_set_flag(&cluster_info[idx], CLUSTER_FLAG_FREE); |
398 | if (cluster_is_null(&p->free_cluster_head)) { | |
399 | cluster_set_next_flag(&p->free_cluster_head, idx, 0); | |
400 | cluster_set_next_flag(&p->free_cluster_tail, idx, 0); | |
401 | } else { | |
402 | unsigned int tail = cluster_next(&p->free_cluster_tail); | |
403 | cluster_set_next(&cluster_info[tail], idx); | |
404 | cluster_set_next_flag(&p->free_cluster_tail, idx, 0); | |
405 | } | |
406 | } | |
407 | } | |
408 | ||
409 | /* | |
410 | * It's possible scan_swap_map() uses a free cluster in the middle of free | |
411 | * cluster list. Avoiding such abuse to avoid list corruption. | |
412 | */ | |
ebc2a1a6 SL |
413 | static bool |
414 | scan_swap_map_ssd_cluster_conflict(struct swap_info_struct *si, | |
2a8f9449 SL |
415 | unsigned long offset) |
416 | { | |
ebc2a1a6 SL |
417 | struct percpu_cluster *percpu_cluster; |
418 | bool conflict; | |
419 | ||
2a8f9449 | 420 | offset /= SWAPFILE_CLUSTER; |
ebc2a1a6 | 421 | conflict = !cluster_is_null(&si->free_cluster_head) && |
2a8f9449 SL |
422 | offset != cluster_next(&si->free_cluster_head) && |
423 | cluster_is_free(&si->cluster_info[offset]); | |
ebc2a1a6 SL |
424 | |
425 | if (!conflict) | |
426 | return false; | |
427 | ||
428 | percpu_cluster = this_cpu_ptr(si->percpu_cluster); | |
429 | cluster_set_null(&percpu_cluster->index); | |
430 | return true; | |
431 | } | |
432 | ||
433 | /* | |
434 | * Try to get a swap entry from current cpu's swap entry pool (a cluster). This | |
435 | * might involve allocating a new cluster for current CPU too. | |
436 | */ | |
437 | static void scan_swap_map_try_ssd_cluster(struct swap_info_struct *si, | |
438 | unsigned long *offset, unsigned long *scan_base) | |
439 | { | |
440 | struct percpu_cluster *cluster; | |
441 | bool found_free; | |
442 | unsigned long tmp; | |
443 | ||
444 | new_cluster: | |
445 | cluster = this_cpu_ptr(si->percpu_cluster); | |
446 | if (cluster_is_null(&cluster->index)) { | |
447 | if (!cluster_is_null(&si->free_cluster_head)) { | |
448 | cluster->index = si->free_cluster_head; | |
449 | cluster->next = cluster_next(&cluster->index) * | |
450 | SWAPFILE_CLUSTER; | |
451 | } else if (!cluster_is_null(&si->discard_cluster_head)) { | |
452 | /* | |
453 | * we don't have free cluster but have some clusters in | |
454 | * discarding, do discard now and reclaim them | |
455 | */ | |
456 | swap_do_scheduled_discard(si); | |
457 | *scan_base = *offset = si->cluster_next; | |
458 | goto new_cluster; | |
459 | } else | |
460 | return; | |
461 | } | |
462 | ||
463 | found_free = false; | |
464 | ||
465 | /* | |
466 | * Other CPUs can use our cluster if they can't find a free cluster, | |
467 | * check if there is still free entry in the cluster | |
468 | */ | |
469 | tmp = cluster->next; | |
470 | while (tmp < si->max && tmp < (cluster_next(&cluster->index) + 1) * | |
471 | SWAPFILE_CLUSTER) { | |
472 | if (!si->swap_map[tmp]) { | |
473 | found_free = true; | |
474 | break; | |
475 | } | |
476 | tmp++; | |
477 | } | |
478 | if (!found_free) { | |
479 | cluster_set_null(&cluster->index); | |
480 | goto new_cluster; | |
481 | } | |
482 | cluster->next = tmp + 1; | |
483 | *offset = tmp; | |
484 | *scan_base = tmp; | |
2a8f9449 SL |
485 | } |
486 | ||
24b8ff7c CEB |
487 | static unsigned long scan_swap_map(struct swap_info_struct *si, |
488 | unsigned char usage) | |
1da177e4 | 489 | { |
ebebbbe9 | 490 | unsigned long offset; |
c60aa176 | 491 | unsigned long scan_base; |
7992fde7 | 492 | unsigned long last_in_cluster = 0; |
048c27fd | 493 | int latency_ration = LATENCY_LIMIT; |
7dfad418 | 494 | |
886bb7e9 | 495 | /* |
7dfad418 HD |
496 | * We try to cluster swap pages by allocating them sequentially |
497 | * in swap. Once we've allocated SWAPFILE_CLUSTER pages this | |
498 | * way, however, we resort to first-free allocation, starting | |
499 | * a new cluster. This prevents us from scattering swap pages | |
500 | * all over the entire swap partition, so that we reduce | |
501 | * overall disk seek times between swap pages. -- sct | |
502 | * But we do now try to find an empty cluster. -Andrea | |
c60aa176 | 503 | * And we let swap pages go all over an SSD partition. Hugh |
7dfad418 HD |
504 | */ |
505 | ||
52b7efdb | 506 | si->flags += SWP_SCANNING; |
c60aa176 | 507 | scan_base = offset = si->cluster_next; |
ebebbbe9 | 508 | |
ebc2a1a6 SL |
509 | /* SSD algorithm */ |
510 | if (si->cluster_info) { | |
511 | scan_swap_map_try_ssd_cluster(si, &offset, &scan_base); | |
512 | goto checks; | |
513 | } | |
514 | ||
ebebbbe9 HD |
515 | if (unlikely(!si->cluster_nr--)) { |
516 | if (si->pages - si->inuse_pages < SWAPFILE_CLUSTER) { | |
517 | si->cluster_nr = SWAPFILE_CLUSTER - 1; | |
518 | goto checks; | |
519 | } | |
2a8f9449 | 520 | |
ec8acf20 | 521 | spin_unlock(&si->lock); |
7dfad418 | 522 | |
c60aa176 HD |
523 | /* |
524 | * If seek is expensive, start searching for new cluster from | |
525 | * start of partition, to minimize the span of allocated swap. | |
526 | * But if seek is cheap, search from our current position, so | |
527 | * that swap is allocated from all over the partition: if the | |
528 | * Flash Translation Layer only remaps within limited zones, | |
529 | * we don't want to wear out the first zone too quickly. | |
530 | */ | |
531 | if (!(si->flags & SWP_SOLIDSTATE)) | |
532 | scan_base = offset = si->lowest_bit; | |
7dfad418 HD |
533 | last_in_cluster = offset + SWAPFILE_CLUSTER - 1; |
534 | ||
535 | /* Locate the first empty (unaligned) cluster */ | |
536 | for (; last_in_cluster <= si->highest_bit; offset++) { | |
1da177e4 | 537 | if (si->swap_map[offset]) |
7dfad418 HD |
538 | last_in_cluster = offset + SWAPFILE_CLUSTER; |
539 | else if (offset == last_in_cluster) { | |
ec8acf20 | 540 | spin_lock(&si->lock); |
ebebbbe9 HD |
541 | offset -= SWAPFILE_CLUSTER - 1; |
542 | si->cluster_next = offset; | |
543 | si->cluster_nr = SWAPFILE_CLUSTER - 1; | |
544 | goto checks; | |
1da177e4 | 545 | } |
048c27fd HD |
546 | if (unlikely(--latency_ration < 0)) { |
547 | cond_resched(); | |
548 | latency_ration = LATENCY_LIMIT; | |
549 | } | |
7dfad418 | 550 | } |
ebebbbe9 HD |
551 | |
552 | offset = si->lowest_bit; | |
c60aa176 HD |
553 | last_in_cluster = offset + SWAPFILE_CLUSTER - 1; |
554 | ||
555 | /* Locate the first empty (unaligned) cluster */ | |
556 | for (; last_in_cluster < scan_base; offset++) { | |
557 | if (si->swap_map[offset]) | |
558 | last_in_cluster = offset + SWAPFILE_CLUSTER; | |
559 | else if (offset == last_in_cluster) { | |
ec8acf20 | 560 | spin_lock(&si->lock); |
c60aa176 HD |
561 | offset -= SWAPFILE_CLUSTER - 1; |
562 | si->cluster_next = offset; | |
563 | si->cluster_nr = SWAPFILE_CLUSTER - 1; | |
c60aa176 HD |
564 | goto checks; |
565 | } | |
566 | if (unlikely(--latency_ration < 0)) { | |
567 | cond_resched(); | |
568 | latency_ration = LATENCY_LIMIT; | |
569 | } | |
570 | } | |
571 | ||
572 | offset = scan_base; | |
ec8acf20 | 573 | spin_lock(&si->lock); |
ebebbbe9 | 574 | si->cluster_nr = SWAPFILE_CLUSTER - 1; |
1da177e4 | 575 | } |
7dfad418 | 576 | |
ebebbbe9 | 577 | checks: |
ebc2a1a6 SL |
578 | if (si->cluster_info) { |
579 | while (scan_swap_map_ssd_cluster_conflict(si, offset)) | |
580 | scan_swap_map_try_ssd_cluster(si, &offset, &scan_base); | |
581 | } | |
ebebbbe9 | 582 | if (!(si->flags & SWP_WRITEOK)) |
52b7efdb | 583 | goto no_page; |
7dfad418 HD |
584 | if (!si->highest_bit) |
585 | goto no_page; | |
ebebbbe9 | 586 | if (offset > si->highest_bit) |
c60aa176 | 587 | scan_base = offset = si->lowest_bit; |
c9e44410 | 588 | |
b73d7fce HD |
589 | /* reuse swap entry of cache-only swap if not busy. */ |
590 | if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) { | |
c9e44410 | 591 | int swap_was_freed; |
ec8acf20 | 592 | spin_unlock(&si->lock); |
c9e44410 | 593 | swap_was_freed = __try_to_reclaim_swap(si, offset); |
ec8acf20 | 594 | spin_lock(&si->lock); |
c9e44410 KH |
595 | /* entry was freed successfully, try to use this again */ |
596 | if (swap_was_freed) | |
597 | goto checks; | |
598 | goto scan; /* check next one */ | |
599 | } | |
600 | ||
ebebbbe9 HD |
601 | if (si->swap_map[offset]) |
602 | goto scan; | |
603 | ||
604 | if (offset == si->lowest_bit) | |
605 | si->lowest_bit++; | |
606 | if (offset == si->highest_bit) | |
607 | si->highest_bit--; | |
608 | si->inuse_pages++; | |
609 | if (si->inuse_pages == si->pages) { | |
610 | si->lowest_bit = si->max; | |
611 | si->highest_bit = 0; | |
18ab4d4c DS |
612 | spin_lock(&swap_avail_lock); |
613 | plist_del(&si->avail_list, &swap_avail_head); | |
614 | spin_unlock(&swap_avail_lock); | |
1da177e4 | 615 | } |
253d553b | 616 | si->swap_map[offset] = usage; |
2a8f9449 | 617 | inc_cluster_info_page(si, si->cluster_info, offset); |
ebebbbe9 HD |
618 | si->cluster_next = offset + 1; |
619 | si->flags -= SWP_SCANNING; | |
7992fde7 | 620 | |
ebebbbe9 | 621 | return offset; |
7dfad418 | 622 | |
ebebbbe9 | 623 | scan: |
ec8acf20 | 624 | spin_unlock(&si->lock); |
7dfad418 | 625 | while (++offset <= si->highest_bit) { |
52b7efdb | 626 | if (!si->swap_map[offset]) { |
ec8acf20 | 627 | spin_lock(&si->lock); |
52b7efdb HD |
628 | goto checks; |
629 | } | |
c9e44410 | 630 | if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) { |
ec8acf20 | 631 | spin_lock(&si->lock); |
c9e44410 KH |
632 | goto checks; |
633 | } | |
048c27fd HD |
634 | if (unlikely(--latency_ration < 0)) { |
635 | cond_resched(); | |
636 | latency_ration = LATENCY_LIMIT; | |
637 | } | |
7dfad418 | 638 | } |
c60aa176 | 639 | offset = si->lowest_bit; |
a5998061 | 640 | while (offset < scan_base) { |
c60aa176 | 641 | if (!si->swap_map[offset]) { |
ec8acf20 | 642 | spin_lock(&si->lock); |
c60aa176 HD |
643 | goto checks; |
644 | } | |
c9e44410 | 645 | if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) { |
ec8acf20 | 646 | spin_lock(&si->lock); |
c9e44410 KH |
647 | goto checks; |
648 | } | |
c60aa176 HD |
649 | if (unlikely(--latency_ration < 0)) { |
650 | cond_resched(); | |
651 | latency_ration = LATENCY_LIMIT; | |
652 | } | |
a5998061 | 653 | offset++; |
c60aa176 | 654 | } |
ec8acf20 | 655 | spin_lock(&si->lock); |
7dfad418 HD |
656 | |
657 | no_page: | |
52b7efdb | 658 | si->flags -= SWP_SCANNING; |
1da177e4 LT |
659 | return 0; |
660 | } | |
661 | ||
662 | swp_entry_t get_swap_page(void) | |
663 | { | |
adfab836 | 664 | struct swap_info_struct *si, *next; |
fb4f88dc | 665 | pgoff_t offset; |
1da177e4 | 666 | |
ec8acf20 | 667 | if (atomic_long_read(&nr_swap_pages) <= 0) |
fb4f88dc | 668 | goto noswap; |
ec8acf20 | 669 | atomic_long_dec(&nr_swap_pages); |
fb4f88dc | 670 | |
18ab4d4c DS |
671 | spin_lock(&swap_avail_lock); |
672 | ||
673 | start_over: | |
674 | plist_for_each_entry_safe(si, next, &swap_avail_head, avail_list) { | |
675 | /* requeue si to after same-priority siblings */ | |
676 | plist_requeue(&si->avail_list, &swap_avail_head); | |
677 | spin_unlock(&swap_avail_lock); | |
ec8acf20 | 678 | spin_lock(&si->lock); |
adfab836 | 679 | if (!si->highest_bit || !(si->flags & SWP_WRITEOK)) { |
18ab4d4c DS |
680 | spin_lock(&swap_avail_lock); |
681 | if (plist_node_empty(&si->avail_list)) { | |
682 | spin_unlock(&si->lock); | |
683 | goto nextsi; | |
684 | } | |
685 | WARN(!si->highest_bit, | |
686 | "swap_info %d in list but !highest_bit\n", | |
687 | si->type); | |
688 | WARN(!(si->flags & SWP_WRITEOK), | |
689 | "swap_info %d in list but !SWP_WRITEOK\n", | |
690 | si->type); | |
691 | plist_del(&si->avail_list, &swap_avail_head); | |
ec8acf20 | 692 | spin_unlock(&si->lock); |
18ab4d4c | 693 | goto nextsi; |
ec8acf20 | 694 | } |
fb4f88dc | 695 | |
355cfa73 | 696 | /* This is called for allocating swap entry for cache */ |
253d553b | 697 | offset = scan_swap_map(si, SWAP_HAS_CACHE); |
ec8acf20 SL |
698 | spin_unlock(&si->lock); |
699 | if (offset) | |
adfab836 | 700 | return swp_entry(si->type, offset); |
18ab4d4c DS |
701 | pr_debug("scan_swap_map of si %d failed to find offset\n", |
702 | si->type); | |
703 | spin_lock(&swap_avail_lock); | |
704 | nextsi: | |
adfab836 DS |
705 | /* |
706 | * if we got here, it's likely that si was almost full before, | |
707 | * and since scan_swap_map() can drop the si->lock, multiple | |
708 | * callers probably all tried to get a page from the same si | |
18ab4d4c DS |
709 | * and it filled up before we could get one; or, the si filled |
710 | * up between us dropping swap_avail_lock and taking si->lock. | |
711 | * Since we dropped the swap_avail_lock, the swap_avail_head | |
712 | * list may have been modified; so if next is still in the | |
713 | * swap_avail_head list then try it, otherwise start over. | |
adfab836 | 714 | */ |
18ab4d4c DS |
715 | if (plist_node_empty(&next->avail_list)) |
716 | goto start_over; | |
1da177e4 | 717 | } |
fb4f88dc | 718 | |
18ab4d4c DS |
719 | spin_unlock(&swap_avail_lock); |
720 | ||
ec8acf20 | 721 | atomic_long_inc(&nr_swap_pages); |
fb4f88dc | 722 | noswap: |
fb4f88dc | 723 | return (swp_entry_t) {0}; |
1da177e4 LT |
724 | } |
725 | ||
2de1a7e4 | 726 | /* The only caller of this function is now suspend routine */ |
910321ea HD |
727 | swp_entry_t get_swap_page_of_type(int type) |
728 | { | |
729 | struct swap_info_struct *si; | |
730 | pgoff_t offset; | |
731 | ||
910321ea | 732 | si = swap_info[type]; |
ec8acf20 | 733 | spin_lock(&si->lock); |
910321ea | 734 | if (si && (si->flags & SWP_WRITEOK)) { |
ec8acf20 | 735 | atomic_long_dec(&nr_swap_pages); |
910321ea HD |
736 | /* This is called for allocating swap entry, not cache */ |
737 | offset = scan_swap_map(si, 1); | |
738 | if (offset) { | |
ec8acf20 | 739 | spin_unlock(&si->lock); |
910321ea HD |
740 | return swp_entry(type, offset); |
741 | } | |
ec8acf20 | 742 | atomic_long_inc(&nr_swap_pages); |
910321ea | 743 | } |
ec8acf20 | 744 | spin_unlock(&si->lock); |
910321ea HD |
745 | return (swp_entry_t) {0}; |
746 | } | |
747 | ||
73c34b6a | 748 | static struct swap_info_struct *swap_info_get(swp_entry_t entry) |
1da177e4 | 749 | { |
73c34b6a | 750 | struct swap_info_struct *p; |
1da177e4 LT |
751 | unsigned long offset, type; |
752 | ||
753 | if (!entry.val) | |
754 | goto out; | |
755 | type = swp_type(entry); | |
756 | if (type >= nr_swapfiles) | |
757 | goto bad_nofile; | |
efa90a98 | 758 | p = swap_info[type]; |
1da177e4 LT |
759 | if (!(p->flags & SWP_USED)) |
760 | goto bad_device; | |
761 | offset = swp_offset(entry); | |
762 | if (offset >= p->max) | |
763 | goto bad_offset; | |
764 | if (!p->swap_map[offset]) | |
765 | goto bad_free; | |
ec8acf20 | 766 | spin_lock(&p->lock); |
1da177e4 LT |
767 | return p; |
768 | ||
769 | bad_free: | |
465c47fd | 770 | pr_err("swap_free: %s%08lx\n", Unused_offset, entry.val); |
1da177e4 LT |
771 | goto out; |
772 | bad_offset: | |
465c47fd | 773 | pr_err("swap_free: %s%08lx\n", Bad_offset, entry.val); |
1da177e4 LT |
774 | goto out; |
775 | bad_device: | |
465c47fd | 776 | pr_err("swap_free: %s%08lx\n", Unused_file, entry.val); |
1da177e4 LT |
777 | goto out; |
778 | bad_nofile: | |
465c47fd | 779 | pr_err("swap_free: %s%08lx\n", Bad_file, entry.val); |
1da177e4 LT |
780 | out: |
781 | return NULL; | |
886bb7e9 | 782 | } |
1da177e4 | 783 | |
8d69aaee HD |
784 | static unsigned char swap_entry_free(struct swap_info_struct *p, |
785 | swp_entry_t entry, unsigned char usage) | |
1da177e4 | 786 | { |
253d553b | 787 | unsigned long offset = swp_offset(entry); |
8d69aaee HD |
788 | unsigned char count; |
789 | unsigned char has_cache; | |
355cfa73 | 790 | |
253d553b HD |
791 | count = p->swap_map[offset]; |
792 | has_cache = count & SWAP_HAS_CACHE; | |
793 | count &= ~SWAP_HAS_CACHE; | |
355cfa73 | 794 | |
253d553b | 795 | if (usage == SWAP_HAS_CACHE) { |
355cfa73 | 796 | VM_BUG_ON(!has_cache); |
253d553b | 797 | has_cache = 0; |
aaa46865 HD |
798 | } else if (count == SWAP_MAP_SHMEM) { |
799 | /* | |
800 | * Or we could insist on shmem.c using a special | |
801 | * swap_shmem_free() and free_shmem_swap_and_cache()... | |
802 | */ | |
803 | count = 0; | |
570a335b HD |
804 | } else if ((count & ~COUNT_CONTINUED) <= SWAP_MAP_MAX) { |
805 | if (count == COUNT_CONTINUED) { | |
806 | if (swap_count_continued(p, offset, count)) | |
807 | count = SWAP_MAP_MAX | COUNT_CONTINUED; | |
808 | else | |
809 | count = SWAP_MAP_MAX; | |
810 | } else | |
811 | count--; | |
812 | } | |
253d553b HD |
813 | |
814 | if (!count) | |
815 | mem_cgroup_uncharge_swap(entry); | |
816 | ||
817 | usage = count | has_cache; | |
818 | p->swap_map[offset] = usage; | |
355cfa73 | 819 | |
355cfa73 | 820 | /* free if no reference */ |
253d553b | 821 | if (!usage) { |
2a8f9449 | 822 | dec_cluster_info_page(p, p->cluster_info, offset); |
355cfa73 KH |
823 | if (offset < p->lowest_bit) |
824 | p->lowest_bit = offset; | |
18ab4d4c DS |
825 | if (offset > p->highest_bit) { |
826 | bool was_full = !p->highest_bit; | |
355cfa73 | 827 | p->highest_bit = offset; |
18ab4d4c DS |
828 | if (was_full && (p->flags & SWP_WRITEOK)) { |
829 | spin_lock(&swap_avail_lock); | |
830 | WARN_ON(!plist_node_empty(&p->avail_list)); | |
831 | if (plist_node_empty(&p->avail_list)) | |
832 | plist_add(&p->avail_list, | |
833 | &swap_avail_head); | |
834 | spin_unlock(&swap_avail_lock); | |
835 | } | |
836 | } | |
ec8acf20 | 837 | atomic_long_inc(&nr_swap_pages); |
355cfa73 | 838 | p->inuse_pages--; |
38b5faf4 | 839 | frontswap_invalidate_page(p->type, offset); |
73744923 MG |
840 | if (p->flags & SWP_BLKDEV) { |
841 | struct gendisk *disk = p->bdev->bd_disk; | |
842 | if (disk->fops->swap_slot_free_notify) | |
843 | disk->fops->swap_slot_free_notify(p->bdev, | |
844 | offset); | |
845 | } | |
1da177e4 | 846 | } |
253d553b HD |
847 | |
848 | return usage; | |
1da177e4 LT |
849 | } |
850 | ||
851 | /* | |
2de1a7e4 | 852 | * Caller has made sure that the swap device corresponding to entry |
1da177e4 LT |
853 | * is still around or has not been recycled. |
854 | */ | |
855 | void swap_free(swp_entry_t entry) | |
856 | { | |
73c34b6a | 857 | struct swap_info_struct *p; |
1da177e4 LT |
858 | |
859 | p = swap_info_get(entry); | |
860 | if (p) { | |
253d553b | 861 | swap_entry_free(p, entry, 1); |
ec8acf20 | 862 | spin_unlock(&p->lock); |
1da177e4 LT |
863 | } |
864 | } | |
865 | ||
cb4b86ba KH |
866 | /* |
867 | * Called after dropping swapcache to decrease refcnt to swap entries. | |
868 | */ | |
869 | void swapcache_free(swp_entry_t entry, struct page *page) | |
870 | { | |
355cfa73 | 871 | struct swap_info_struct *p; |
8d69aaee | 872 | unsigned char count; |
355cfa73 | 873 | |
355cfa73 KH |
874 | p = swap_info_get(entry); |
875 | if (p) { | |
253d553b HD |
876 | count = swap_entry_free(p, entry, SWAP_HAS_CACHE); |
877 | if (page) | |
878 | mem_cgroup_uncharge_swapcache(page, entry, count != 0); | |
ec8acf20 | 879 | spin_unlock(&p->lock); |
355cfa73 | 880 | } |
cb4b86ba KH |
881 | } |
882 | ||
1da177e4 | 883 | /* |
c475a8ab | 884 | * How many references to page are currently swapped out? |
570a335b HD |
885 | * This does not give an exact answer when swap count is continued, |
886 | * but does include the high COUNT_CONTINUED flag to allow for that. | |
1da177e4 | 887 | */ |
bde05d1c | 888 | int page_swapcount(struct page *page) |
1da177e4 | 889 | { |
c475a8ab HD |
890 | int count = 0; |
891 | struct swap_info_struct *p; | |
1da177e4 LT |
892 | swp_entry_t entry; |
893 | ||
4c21e2f2 | 894 | entry.val = page_private(page); |
1da177e4 LT |
895 | p = swap_info_get(entry); |
896 | if (p) { | |
355cfa73 | 897 | count = swap_count(p->swap_map[swp_offset(entry)]); |
ec8acf20 | 898 | spin_unlock(&p->lock); |
1da177e4 | 899 | } |
c475a8ab | 900 | return count; |
1da177e4 LT |
901 | } |
902 | ||
903 | /* | |
7b1fe597 HD |
904 | * We can write to an anon page without COW if there are no other references |
905 | * to it. And as a side-effect, free up its swap: because the old content | |
906 | * on disk will never be read, and seeking back there to write new content | |
907 | * later would only waste time away from clustering. | |
1da177e4 | 908 | */ |
7b1fe597 | 909 | int reuse_swap_page(struct page *page) |
1da177e4 | 910 | { |
c475a8ab HD |
911 | int count; |
912 | ||
309381fe | 913 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
5ad64688 HD |
914 | if (unlikely(PageKsm(page))) |
915 | return 0; | |
c475a8ab | 916 | count = page_mapcount(page); |
7b1fe597 | 917 | if (count <= 1 && PageSwapCache(page)) { |
c475a8ab | 918 | count += page_swapcount(page); |
7b1fe597 HD |
919 | if (count == 1 && !PageWriteback(page)) { |
920 | delete_from_swap_cache(page); | |
921 | SetPageDirty(page); | |
922 | } | |
923 | } | |
5ad64688 | 924 | return count <= 1; |
1da177e4 LT |
925 | } |
926 | ||
927 | /* | |
a2c43eed HD |
928 | * If swap is getting full, or if there are no more mappings of this page, |
929 | * then try_to_free_swap is called to free its swap space. | |
1da177e4 | 930 | */ |
a2c43eed | 931 | int try_to_free_swap(struct page *page) |
1da177e4 | 932 | { |
309381fe | 933 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
1da177e4 LT |
934 | |
935 | if (!PageSwapCache(page)) | |
936 | return 0; | |
937 | if (PageWriteback(page)) | |
938 | return 0; | |
a2c43eed | 939 | if (page_swapcount(page)) |
1da177e4 LT |
940 | return 0; |
941 | ||
b73d7fce HD |
942 | /* |
943 | * Once hibernation has begun to create its image of memory, | |
944 | * there's a danger that one of the calls to try_to_free_swap() | |
945 | * - most probably a call from __try_to_reclaim_swap() while | |
946 | * hibernation is allocating its own swap pages for the image, | |
947 | * but conceivably even a call from memory reclaim - will free | |
948 | * the swap from a page which has already been recorded in the | |
949 | * image as a clean swapcache page, and then reuse its swap for | |
950 | * another page of the image. On waking from hibernation, the | |
951 | * original page might be freed under memory pressure, then | |
952 | * later read back in from swap, now with the wrong data. | |
953 | * | |
2de1a7e4 | 954 | * Hibernation suspends storage while it is writing the image |
f90ac398 | 955 | * to disk so check that here. |
b73d7fce | 956 | */ |
f90ac398 | 957 | if (pm_suspended_storage()) |
b73d7fce HD |
958 | return 0; |
959 | ||
a2c43eed HD |
960 | delete_from_swap_cache(page); |
961 | SetPageDirty(page); | |
962 | return 1; | |
68a22394 RR |
963 | } |
964 | ||
1da177e4 LT |
965 | /* |
966 | * Free the swap entry like above, but also try to | |
967 | * free the page cache entry if it is the last user. | |
968 | */ | |
2509ef26 | 969 | int free_swap_and_cache(swp_entry_t entry) |
1da177e4 | 970 | { |
2509ef26 | 971 | struct swap_info_struct *p; |
1da177e4 LT |
972 | struct page *page = NULL; |
973 | ||
a7420aa5 | 974 | if (non_swap_entry(entry)) |
2509ef26 | 975 | return 1; |
0697212a | 976 | |
1da177e4 LT |
977 | p = swap_info_get(entry); |
978 | if (p) { | |
253d553b | 979 | if (swap_entry_free(p, entry, 1) == SWAP_HAS_CACHE) { |
33806f06 SL |
980 | page = find_get_page(swap_address_space(entry), |
981 | entry.val); | |
8413ac9d | 982 | if (page && !trylock_page(page)) { |
93fac704 NP |
983 | page_cache_release(page); |
984 | page = NULL; | |
985 | } | |
986 | } | |
ec8acf20 | 987 | spin_unlock(&p->lock); |
1da177e4 LT |
988 | } |
989 | if (page) { | |
a2c43eed HD |
990 | /* |
991 | * Not mapped elsewhere, or swap space full? Free it! | |
992 | * Also recheck PageSwapCache now page is locked (above). | |
993 | */ | |
93fac704 | 994 | if (PageSwapCache(page) && !PageWriteback(page) && |
a2c43eed | 995 | (!page_mapped(page) || vm_swap_full())) { |
1da177e4 LT |
996 | delete_from_swap_cache(page); |
997 | SetPageDirty(page); | |
998 | } | |
999 | unlock_page(page); | |
1000 | page_cache_release(page); | |
1001 | } | |
2509ef26 | 1002 | return p != NULL; |
1da177e4 LT |
1003 | } |
1004 | ||
b0cb1a19 | 1005 | #ifdef CONFIG_HIBERNATION |
f577eb30 | 1006 | /* |
915bae9e | 1007 | * Find the swap type that corresponds to given device (if any). |
f577eb30 | 1008 | * |
915bae9e RW |
1009 | * @offset - number of the PAGE_SIZE-sized block of the device, starting |
1010 | * from 0, in which the swap header is expected to be located. | |
1011 | * | |
1012 | * This is needed for the suspend to disk (aka swsusp). | |
f577eb30 | 1013 | */ |
7bf23687 | 1014 | int swap_type_of(dev_t device, sector_t offset, struct block_device **bdev_p) |
f577eb30 | 1015 | { |
915bae9e | 1016 | struct block_device *bdev = NULL; |
efa90a98 | 1017 | int type; |
f577eb30 | 1018 | |
915bae9e RW |
1019 | if (device) |
1020 | bdev = bdget(device); | |
1021 | ||
f577eb30 | 1022 | spin_lock(&swap_lock); |
efa90a98 HD |
1023 | for (type = 0; type < nr_swapfiles; type++) { |
1024 | struct swap_info_struct *sis = swap_info[type]; | |
f577eb30 | 1025 | |
915bae9e | 1026 | if (!(sis->flags & SWP_WRITEOK)) |
f577eb30 | 1027 | continue; |
b6b5bce3 | 1028 | |
915bae9e | 1029 | if (!bdev) { |
7bf23687 | 1030 | if (bdev_p) |
dddac6a7 | 1031 | *bdev_p = bdgrab(sis->bdev); |
7bf23687 | 1032 | |
6e1819d6 | 1033 | spin_unlock(&swap_lock); |
efa90a98 | 1034 | return type; |
6e1819d6 | 1035 | } |
915bae9e | 1036 | if (bdev == sis->bdev) { |
9625a5f2 | 1037 | struct swap_extent *se = &sis->first_swap_extent; |
915bae9e | 1038 | |
915bae9e | 1039 | if (se->start_block == offset) { |
7bf23687 | 1040 | if (bdev_p) |
dddac6a7 | 1041 | *bdev_p = bdgrab(sis->bdev); |
7bf23687 | 1042 | |
915bae9e RW |
1043 | spin_unlock(&swap_lock); |
1044 | bdput(bdev); | |
efa90a98 | 1045 | return type; |
915bae9e | 1046 | } |
f577eb30 RW |
1047 | } |
1048 | } | |
1049 | spin_unlock(&swap_lock); | |
915bae9e RW |
1050 | if (bdev) |
1051 | bdput(bdev); | |
1052 | ||
f577eb30 RW |
1053 | return -ENODEV; |
1054 | } | |
1055 | ||
73c34b6a HD |
1056 | /* |
1057 | * Get the (PAGE_SIZE) block corresponding to given offset on the swapdev | |
1058 | * corresponding to given index in swap_info (swap type). | |
1059 | */ | |
1060 | sector_t swapdev_block(int type, pgoff_t offset) | |
1061 | { | |
1062 | struct block_device *bdev; | |
1063 | ||
1064 | if ((unsigned int)type >= nr_swapfiles) | |
1065 | return 0; | |
1066 | if (!(swap_info[type]->flags & SWP_WRITEOK)) | |
1067 | return 0; | |
d4906e1a | 1068 | return map_swap_entry(swp_entry(type, offset), &bdev); |
73c34b6a HD |
1069 | } |
1070 | ||
f577eb30 RW |
1071 | /* |
1072 | * Return either the total number of swap pages of given type, or the number | |
1073 | * of free pages of that type (depending on @free) | |
1074 | * | |
1075 | * This is needed for software suspend | |
1076 | */ | |
1077 | unsigned int count_swap_pages(int type, int free) | |
1078 | { | |
1079 | unsigned int n = 0; | |
1080 | ||
efa90a98 HD |
1081 | spin_lock(&swap_lock); |
1082 | if ((unsigned int)type < nr_swapfiles) { | |
1083 | struct swap_info_struct *sis = swap_info[type]; | |
1084 | ||
ec8acf20 | 1085 | spin_lock(&sis->lock); |
efa90a98 HD |
1086 | if (sis->flags & SWP_WRITEOK) { |
1087 | n = sis->pages; | |
f577eb30 | 1088 | if (free) |
efa90a98 | 1089 | n -= sis->inuse_pages; |
f577eb30 | 1090 | } |
ec8acf20 | 1091 | spin_unlock(&sis->lock); |
f577eb30 | 1092 | } |
efa90a98 | 1093 | spin_unlock(&swap_lock); |
f577eb30 RW |
1094 | return n; |
1095 | } | |
73c34b6a | 1096 | #endif /* CONFIG_HIBERNATION */ |
f577eb30 | 1097 | |
179ef71c CG |
1098 | static inline int maybe_same_pte(pte_t pte, pte_t swp_pte) |
1099 | { | |
1100 | #ifdef CONFIG_MEM_SOFT_DIRTY | |
1101 | /* | |
1102 | * When pte keeps soft dirty bit the pte generated | |
1103 | * from swap entry does not has it, still it's same | |
1104 | * pte from logical point of view. | |
1105 | */ | |
1106 | pte_t swp_pte_dirty = pte_swp_mksoft_dirty(swp_pte); | |
1107 | return pte_same(pte, swp_pte) || pte_same(pte, swp_pte_dirty); | |
1108 | #else | |
1109 | return pte_same(pte, swp_pte); | |
1110 | #endif | |
1111 | } | |
1112 | ||
1da177e4 | 1113 | /* |
72866f6f HD |
1114 | * No need to decide whether this PTE shares the swap entry with others, |
1115 | * just let do_wp_page work it out if a write is requested later - to | |
1116 | * force COW, vm_page_prot omits write permission from any private vma. | |
1da177e4 | 1117 | */ |
044d66c1 | 1118 | static int unuse_pte(struct vm_area_struct *vma, pmd_t *pmd, |
1da177e4 LT |
1119 | unsigned long addr, swp_entry_t entry, struct page *page) |
1120 | { | |
9e16b7fb | 1121 | struct page *swapcache; |
72835c86 | 1122 | struct mem_cgroup *memcg; |
044d66c1 HD |
1123 | spinlock_t *ptl; |
1124 | pte_t *pte; | |
1125 | int ret = 1; | |
1126 | ||
9e16b7fb HD |
1127 | swapcache = page; |
1128 | page = ksm_might_need_to_copy(page, vma, addr); | |
1129 | if (unlikely(!page)) | |
1130 | return -ENOMEM; | |
1131 | ||
72835c86 JW |
1132 | if (mem_cgroup_try_charge_swapin(vma->vm_mm, page, |
1133 | GFP_KERNEL, &memcg)) { | |
044d66c1 | 1134 | ret = -ENOMEM; |
85d9fc89 KH |
1135 | goto out_nolock; |
1136 | } | |
044d66c1 HD |
1137 | |
1138 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); | |
179ef71c | 1139 | if (unlikely(!maybe_same_pte(*pte, swp_entry_to_pte(entry)))) { |
5d84c776 | 1140 | mem_cgroup_cancel_charge_swapin(memcg); |
044d66c1 HD |
1141 | ret = 0; |
1142 | goto out; | |
1143 | } | |
8a9f3ccd | 1144 | |
b084d435 | 1145 | dec_mm_counter(vma->vm_mm, MM_SWAPENTS); |
d559db08 | 1146 | inc_mm_counter(vma->vm_mm, MM_ANONPAGES); |
1da177e4 LT |
1147 | get_page(page); |
1148 | set_pte_at(vma->vm_mm, addr, pte, | |
1149 | pte_mkold(mk_pte(page, vma->vm_page_prot))); | |
9e16b7fb HD |
1150 | if (page == swapcache) |
1151 | page_add_anon_rmap(page, vma, addr); | |
1152 | else /* ksm created a completely new copy */ | |
1153 | page_add_new_anon_rmap(page, vma, addr); | |
72835c86 | 1154 | mem_cgroup_commit_charge_swapin(page, memcg); |
1da177e4 LT |
1155 | swap_free(entry); |
1156 | /* | |
1157 | * Move the page to the active list so it is not | |
1158 | * immediately swapped out again after swapon. | |
1159 | */ | |
1160 | activate_page(page); | |
044d66c1 HD |
1161 | out: |
1162 | pte_unmap_unlock(pte, ptl); | |
85d9fc89 | 1163 | out_nolock: |
9e16b7fb HD |
1164 | if (page != swapcache) { |
1165 | unlock_page(page); | |
1166 | put_page(page); | |
1167 | } | |
044d66c1 | 1168 | return ret; |
1da177e4 LT |
1169 | } |
1170 | ||
1171 | static int unuse_pte_range(struct vm_area_struct *vma, pmd_t *pmd, | |
1172 | unsigned long addr, unsigned long end, | |
1173 | swp_entry_t entry, struct page *page) | |
1174 | { | |
1da177e4 | 1175 | pte_t swp_pte = swp_entry_to_pte(entry); |
705e87c0 | 1176 | pte_t *pte; |
8a9f3ccd | 1177 | int ret = 0; |
1da177e4 | 1178 | |
044d66c1 HD |
1179 | /* |
1180 | * We don't actually need pte lock while scanning for swp_pte: since | |
1181 | * we hold page lock and mmap_sem, swp_pte cannot be inserted into the | |
1182 | * page table while we're scanning; though it could get zapped, and on | |
1183 | * some architectures (e.g. x86_32 with PAE) we might catch a glimpse | |
1184 | * of unmatched parts which look like swp_pte, so unuse_pte must | |
1185 | * recheck under pte lock. Scanning without pte lock lets it be | |
2de1a7e4 | 1186 | * preemptable whenever CONFIG_PREEMPT but not CONFIG_HIGHPTE. |
044d66c1 HD |
1187 | */ |
1188 | pte = pte_offset_map(pmd, addr); | |
1da177e4 LT |
1189 | do { |
1190 | /* | |
1191 | * swapoff spends a _lot_ of time in this loop! | |
1192 | * Test inline before going to call unuse_pte. | |
1193 | */ | |
179ef71c | 1194 | if (unlikely(maybe_same_pte(*pte, swp_pte))) { |
044d66c1 HD |
1195 | pte_unmap(pte); |
1196 | ret = unuse_pte(vma, pmd, addr, entry, page); | |
1197 | if (ret) | |
1198 | goto out; | |
1199 | pte = pte_offset_map(pmd, addr); | |
1da177e4 LT |
1200 | } |
1201 | } while (pte++, addr += PAGE_SIZE, addr != end); | |
044d66c1 HD |
1202 | pte_unmap(pte - 1); |
1203 | out: | |
8a9f3ccd | 1204 | return ret; |
1da177e4 LT |
1205 | } |
1206 | ||
1207 | static inline int unuse_pmd_range(struct vm_area_struct *vma, pud_t *pud, | |
1208 | unsigned long addr, unsigned long end, | |
1209 | swp_entry_t entry, struct page *page) | |
1210 | { | |
1211 | pmd_t *pmd; | |
1212 | unsigned long next; | |
8a9f3ccd | 1213 | int ret; |
1da177e4 LT |
1214 | |
1215 | pmd = pmd_offset(pud, addr); | |
1216 | do { | |
1217 | next = pmd_addr_end(addr, end); | |
1a5a9906 | 1218 | if (pmd_none_or_trans_huge_or_clear_bad(pmd)) |
1da177e4 | 1219 | continue; |
8a9f3ccd BS |
1220 | ret = unuse_pte_range(vma, pmd, addr, next, entry, page); |
1221 | if (ret) | |
1222 | return ret; | |
1da177e4 LT |
1223 | } while (pmd++, addr = next, addr != end); |
1224 | return 0; | |
1225 | } | |
1226 | ||
1227 | static inline int unuse_pud_range(struct vm_area_struct *vma, pgd_t *pgd, | |
1228 | unsigned long addr, unsigned long end, | |
1229 | swp_entry_t entry, struct page *page) | |
1230 | { | |
1231 | pud_t *pud; | |
1232 | unsigned long next; | |
8a9f3ccd | 1233 | int ret; |
1da177e4 LT |
1234 | |
1235 | pud = pud_offset(pgd, addr); | |
1236 | do { | |
1237 | next = pud_addr_end(addr, end); | |
1238 | if (pud_none_or_clear_bad(pud)) | |
1239 | continue; | |
8a9f3ccd BS |
1240 | ret = unuse_pmd_range(vma, pud, addr, next, entry, page); |
1241 | if (ret) | |
1242 | return ret; | |
1da177e4 LT |
1243 | } while (pud++, addr = next, addr != end); |
1244 | return 0; | |
1245 | } | |
1246 | ||
1247 | static int unuse_vma(struct vm_area_struct *vma, | |
1248 | swp_entry_t entry, struct page *page) | |
1249 | { | |
1250 | pgd_t *pgd; | |
1251 | unsigned long addr, end, next; | |
8a9f3ccd | 1252 | int ret; |
1da177e4 | 1253 | |
3ca7b3c5 | 1254 | if (page_anon_vma(page)) { |
1da177e4 LT |
1255 | addr = page_address_in_vma(page, vma); |
1256 | if (addr == -EFAULT) | |
1257 | return 0; | |
1258 | else | |
1259 | end = addr + PAGE_SIZE; | |
1260 | } else { | |
1261 | addr = vma->vm_start; | |
1262 | end = vma->vm_end; | |
1263 | } | |
1264 | ||
1265 | pgd = pgd_offset(vma->vm_mm, addr); | |
1266 | do { | |
1267 | next = pgd_addr_end(addr, end); | |
1268 | if (pgd_none_or_clear_bad(pgd)) | |
1269 | continue; | |
8a9f3ccd BS |
1270 | ret = unuse_pud_range(vma, pgd, addr, next, entry, page); |
1271 | if (ret) | |
1272 | return ret; | |
1da177e4 LT |
1273 | } while (pgd++, addr = next, addr != end); |
1274 | return 0; | |
1275 | } | |
1276 | ||
1277 | static int unuse_mm(struct mm_struct *mm, | |
1278 | swp_entry_t entry, struct page *page) | |
1279 | { | |
1280 | struct vm_area_struct *vma; | |
8a9f3ccd | 1281 | int ret = 0; |
1da177e4 LT |
1282 | |
1283 | if (!down_read_trylock(&mm->mmap_sem)) { | |
1284 | /* | |
7d03431c FLVC |
1285 | * Activate page so shrink_inactive_list is unlikely to unmap |
1286 | * its ptes while lock is dropped, so swapoff can make progress. | |
1da177e4 | 1287 | */ |
c475a8ab | 1288 | activate_page(page); |
1da177e4 LT |
1289 | unlock_page(page); |
1290 | down_read(&mm->mmap_sem); | |
1291 | lock_page(page); | |
1292 | } | |
1da177e4 | 1293 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
8a9f3ccd | 1294 | if (vma->anon_vma && (ret = unuse_vma(vma, entry, page))) |
1da177e4 LT |
1295 | break; |
1296 | } | |
1da177e4 | 1297 | up_read(&mm->mmap_sem); |
8a9f3ccd | 1298 | return (ret < 0)? ret: 0; |
1da177e4 LT |
1299 | } |
1300 | ||
1301 | /* | |
38b5faf4 DM |
1302 | * Scan swap_map (or frontswap_map if frontswap parameter is true) |
1303 | * from current position to next entry still in use. | |
1da177e4 LT |
1304 | * Recycle to start on reaching the end, returning 0 when empty. |
1305 | */ | |
6eb396dc | 1306 | static unsigned int find_next_to_unuse(struct swap_info_struct *si, |
38b5faf4 | 1307 | unsigned int prev, bool frontswap) |
1da177e4 | 1308 | { |
6eb396dc HD |
1309 | unsigned int max = si->max; |
1310 | unsigned int i = prev; | |
8d69aaee | 1311 | unsigned char count; |
1da177e4 LT |
1312 | |
1313 | /* | |
5d337b91 | 1314 | * No need for swap_lock here: we're just looking |
1da177e4 LT |
1315 | * for whether an entry is in use, not modifying it; false |
1316 | * hits are okay, and sys_swapoff() has already prevented new | |
5d337b91 | 1317 | * allocations from this area (while holding swap_lock). |
1da177e4 LT |
1318 | */ |
1319 | for (;;) { | |
1320 | if (++i >= max) { | |
1321 | if (!prev) { | |
1322 | i = 0; | |
1323 | break; | |
1324 | } | |
1325 | /* | |
1326 | * No entries in use at top of swap_map, | |
1327 | * loop back to start and recheck there. | |
1328 | */ | |
1329 | max = prev + 1; | |
1330 | prev = 0; | |
1331 | i = 1; | |
1332 | } | |
38b5faf4 DM |
1333 | if (frontswap) { |
1334 | if (frontswap_test(si, i)) | |
1335 | break; | |
1336 | else | |
1337 | continue; | |
1338 | } | |
edfe23da | 1339 | count = ACCESS_ONCE(si->swap_map[i]); |
355cfa73 | 1340 | if (count && swap_count(count) != SWAP_MAP_BAD) |
1da177e4 LT |
1341 | break; |
1342 | } | |
1343 | return i; | |
1344 | } | |
1345 | ||
1346 | /* | |
1347 | * We completely avoid races by reading each swap page in advance, | |
1348 | * and then search for the process using it. All the necessary | |
1349 | * page table adjustments can then be made atomically. | |
38b5faf4 DM |
1350 | * |
1351 | * if the boolean frontswap is true, only unuse pages_to_unuse pages; | |
1352 | * pages_to_unuse==0 means all pages; ignored if frontswap is false | |
1da177e4 | 1353 | */ |
38b5faf4 DM |
1354 | int try_to_unuse(unsigned int type, bool frontswap, |
1355 | unsigned long pages_to_unuse) | |
1da177e4 | 1356 | { |
efa90a98 | 1357 | struct swap_info_struct *si = swap_info[type]; |
1da177e4 | 1358 | struct mm_struct *start_mm; |
edfe23da SL |
1359 | volatile unsigned char *swap_map; /* swap_map is accessed without |
1360 | * locking. Mark it as volatile | |
1361 | * to prevent compiler doing | |
1362 | * something odd. | |
1363 | */ | |
8d69aaee | 1364 | unsigned char swcount; |
1da177e4 LT |
1365 | struct page *page; |
1366 | swp_entry_t entry; | |
6eb396dc | 1367 | unsigned int i = 0; |
1da177e4 | 1368 | int retval = 0; |
1da177e4 LT |
1369 | |
1370 | /* | |
1371 | * When searching mms for an entry, a good strategy is to | |
1372 | * start at the first mm we freed the previous entry from | |
1373 | * (though actually we don't notice whether we or coincidence | |
1374 | * freed the entry). Initialize this start_mm with a hold. | |
1375 | * | |
1376 | * A simpler strategy would be to start at the last mm we | |
1377 | * freed the previous entry from; but that would take less | |
1378 | * advantage of mmlist ordering, which clusters forked mms | |
1379 | * together, child after parent. If we race with dup_mmap(), we | |
1380 | * prefer to resolve parent before child, lest we miss entries | |
1381 | * duplicated after we scanned child: using last mm would invert | |
570a335b | 1382 | * that. |
1da177e4 LT |
1383 | */ |
1384 | start_mm = &init_mm; | |
1385 | atomic_inc(&init_mm.mm_users); | |
1386 | ||
1387 | /* | |
1388 | * Keep on scanning until all entries have gone. Usually, | |
1389 | * one pass through swap_map is enough, but not necessarily: | |
1390 | * there are races when an instance of an entry might be missed. | |
1391 | */ | |
38b5faf4 | 1392 | while ((i = find_next_to_unuse(si, i, frontswap)) != 0) { |
1da177e4 LT |
1393 | if (signal_pending(current)) { |
1394 | retval = -EINTR; | |
1395 | break; | |
1396 | } | |
1397 | ||
886bb7e9 | 1398 | /* |
1da177e4 LT |
1399 | * Get a page for the entry, using the existing swap |
1400 | * cache page if there is one. Otherwise, get a clean | |
886bb7e9 | 1401 | * page and read the swap into it. |
1da177e4 LT |
1402 | */ |
1403 | swap_map = &si->swap_map[i]; | |
1404 | entry = swp_entry(type, i); | |
02098fea HD |
1405 | page = read_swap_cache_async(entry, |
1406 | GFP_HIGHUSER_MOVABLE, NULL, 0); | |
1da177e4 LT |
1407 | if (!page) { |
1408 | /* | |
1409 | * Either swap_duplicate() failed because entry | |
1410 | * has been freed independently, and will not be | |
1411 | * reused since sys_swapoff() already disabled | |
1412 | * allocation from here, or alloc_page() failed. | |
1413 | */ | |
edfe23da SL |
1414 | swcount = *swap_map; |
1415 | /* | |
1416 | * We don't hold lock here, so the swap entry could be | |
1417 | * SWAP_MAP_BAD (when the cluster is discarding). | |
1418 | * Instead of fail out, We can just skip the swap | |
1419 | * entry because swapoff will wait for discarding | |
1420 | * finish anyway. | |
1421 | */ | |
1422 | if (!swcount || swcount == SWAP_MAP_BAD) | |
1da177e4 LT |
1423 | continue; |
1424 | retval = -ENOMEM; | |
1425 | break; | |
1426 | } | |
1427 | ||
1428 | /* | |
1429 | * Don't hold on to start_mm if it looks like exiting. | |
1430 | */ | |
1431 | if (atomic_read(&start_mm->mm_users) == 1) { | |
1432 | mmput(start_mm); | |
1433 | start_mm = &init_mm; | |
1434 | atomic_inc(&init_mm.mm_users); | |
1435 | } | |
1436 | ||
1437 | /* | |
1438 | * Wait for and lock page. When do_swap_page races with | |
1439 | * try_to_unuse, do_swap_page can handle the fault much | |
1440 | * faster than try_to_unuse can locate the entry. This | |
1441 | * apparently redundant "wait_on_page_locked" lets try_to_unuse | |
1442 | * defer to do_swap_page in such a case - in some tests, | |
1443 | * do_swap_page and try_to_unuse repeatedly compete. | |
1444 | */ | |
1445 | wait_on_page_locked(page); | |
1446 | wait_on_page_writeback(page); | |
1447 | lock_page(page); | |
1448 | wait_on_page_writeback(page); | |
1449 | ||
1450 | /* | |
1451 | * Remove all references to entry. | |
1da177e4 | 1452 | */ |
1da177e4 | 1453 | swcount = *swap_map; |
aaa46865 HD |
1454 | if (swap_count(swcount) == SWAP_MAP_SHMEM) { |
1455 | retval = shmem_unuse(entry, page); | |
1456 | /* page has already been unlocked and released */ | |
1457 | if (retval < 0) | |
1458 | break; | |
1459 | continue; | |
1da177e4 | 1460 | } |
aaa46865 HD |
1461 | if (swap_count(swcount) && start_mm != &init_mm) |
1462 | retval = unuse_mm(start_mm, entry, page); | |
1463 | ||
355cfa73 | 1464 | if (swap_count(*swap_map)) { |
1da177e4 LT |
1465 | int set_start_mm = (*swap_map >= swcount); |
1466 | struct list_head *p = &start_mm->mmlist; | |
1467 | struct mm_struct *new_start_mm = start_mm; | |
1468 | struct mm_struct *prev_mm = start_mm; | |
1469 | struct mm_struct *mm; | |
1470 | ||
1471 | atomic_inc(&new_start_mm->mm_users); | |
1472 | atomic_inc(&prev_mm->mm_users); | |
1473 | spin_lock(&mmlist_lock); | |
aaa46865 | 1474 | while (swap_count(*swap_map) && !retval && |
1da177e4 LT |
1475 | (p = p->next) != &start_mm->mmlist) { |
1476 | mm = list_entry(p, struct mm_struct, mmlist); | |
70af7c5c | 1477 | if (!atomic_inc_not_zero(&mm->mm_users)) |
1da177e4 | 1478 | continue; |
1da177e4 LT |
1479 | spin_unlock(&mmlist_lock); |
1480 | mmput(prev_mm); | |
1481 | prev_mm = mm; | |
1482 | ||
1483 | cond_resched(); | |
1484 | ||
1485 | swcount = *swap_map; | |
355cfa73 | 1486 | if (!swap_count(swcount)) /* any usage ? */ |
1da177e4 | 1487 | ; |
aaa46865 | 1488 | else if (mm == &init_mm) |
1da177e4 | 1489 | set_start_mm = 1; |
aaa46865 | 1490 | else |
1da177e4 | 1491 | retval = unuse_mm(mm, entry, page); |
355cfa73 | 1492 | |
32c5fc10 | 1493 | if (set_start_mm && *swap_map < swcount) { |
1da177e4 LT |
1494 | mmput(new_start_mm); |
1495 | atomic_inc(&mm->mm_users); | |
1496 | new_start_mm = mm; | |
1497 | set_start_mm = 0; | |
1498 | } | |
1499 | spin_lock(&mmlist_lock); | |
1500 | } | |
1501 | spin_unlock(&mmlist_lock); | |
1502 | mmput(prev_mm); | |
1503 | mmput(start_mm); | |
1504 | start_mm = new_start_mm; | |
1505 | } | |
1506 | if (retval) { | |
1507 | unlock_page(page); | |
1508 | page_cache_release(page); | |
1509 | break; | |
1510 | } | |
1511 | ||
1da177e4 LT |
1512 | /* |
1513 | * If a reference remains (rare), we would like to leave | |
1514 | * the page in the swap cache; but try_to_unmap could | |
1515 | * then re-duplicate the entry once we drop page lock, | |
1516 | * so we might loop indefinitely; also, that page could | |
1517 | * not be swapped out to other storage meanwhile. So: | |
1518 | * delete from cache even if there's another reference, | |
1519 | * after ensuring that the data has been saved to disk - | |
1520 | * since if the reference remains (rarer), it will be | |
1521 | * read from disk into another page. Splitting into two | |
1522 | * pages would be incorrect if swap supported "shared | |
1523 | * private" pages, but they are handled by tmpfs files. | |
5ad64688 HD |
1524 | * |
1525 | * Given how unuse_vma() targets one particular offset | |
1526 | * in an anon_vma, once the anon_vma has been determined, | |
1527 | * this splitting happens to be just what is needed to | |
1528 | * handle where KSM pages have been swapped out: re-reading | |
1529 | * is unnecessarily slow, but we can fix that later on. | |
1da177e4 | 1530 | */ |
355cfa73 KH |
1531 | if (swap_count(*swap_map) && |
1532 | PageDirty(page) && PageSwapCache(page)) { | |
1da177e4 LT |
1533 | struct writeback_control wbc = { |
1534 | .sync_mode = WB_SYNC_NONE, | |
1535 | }; | |
1536 | ||
1537 | swap_writepage(page, &wbc); | |
1538 | lock_page(page); | |
1539 | wait_on_page_writeback(page); | |
1540 | } | |
68bdc8d6 HD |
1541 | |
1542 | /* | |
1543 | * It is conceivable that a racing task removed this page from | |
1544 | * swap cache just before we acquired the page lock at the top, | |
1545 | * or while we dropped it in unuse_mm(). The page might even | |
1546 | * be back in swap cache on another swap area: that we must not | |
1547 | * delete, since it may not have been written out to swap yet. | |
1548 | */ | |
1549 | if (PageSwapCache(page) && | |
1550 | likely(page_private(page) == entry.val)) | |
2e0e26c7 | 1551 | delete_from_swap_cache(page); |
1da177e4 LT |
1552 | |
1553 | /* | |
1554 | * So we could skip searching mms once swap count went | |
1555 | * to 1, we did not mark any present ptes as dirty: must | |
2706a1b8 | 1556 | * mark page dirty so shrink_page_list will preserve it. |
1da177e4 LT |
1557 | */ |
1558 | SetPageDirty(page); | |
1559 | unlock_page(page); | |
1560 | page_cache_release(page); | |
1561 | ||
1562 | /* | |
1563 | * Make sure that we aren't completely killing | |
1564 | * interactive performance. | |
1565 | */ | |
1566 | cond_resched(); | |
38b5faf4 DM |
1567 | if (frontswap && pages_to_unuse > 0) { |
1568 | if (!--pages_to_unuse) | |
1569 | break; | |
1570 | } | |
1da177e4 LT |
1571 | } |
1572 | ||
1573 | mmput(start_mm); | |
1da177e4 LT |
1574 | return retval; |
1575 | } | |
1576 | ||
1577 | /* | |
5d337b91 HD |
1578 | * After a successful try_to_unuse, if no swap is now in use, we know |
1579 | * we can empty the mmlist. swap_lock must be held on entry and exit. | |
1580 | * Note that mmlist_lock nests inside swap_lock, and an mm must be | |
1da177e4 LT |
1581 | * added to the mmlist just after page_duplicate - before would be racy. |
1582 | */ | |
1583 | static void drain_mmlist(void) | |
1584 | { | |
1585 | struct list_head *p, *next; | |
efa90a98 | 1586 | unsigned int type; |
1da177e4 | 1587 | |
efa90a98 HD |
1588 | for (type = 0; type < nr_swapfiles; type++) |
1589 | if (swap_info[type]->inuse_pages) | |
1da177e4 LT |
1590 | return; |
1591 | spin_lock(&mmlist_lock); | |
1592 | list_for_each_safe(p, next, &init_mm.mmlist) | |
1593 | list_del_init(p); | |
1594 | spin_unlock(&mmlist_lock); | |
1595 | } | |
1596 | ||
1597 | /* | |
1598 | * Use this swapdev's extent info to locate the (PAGE_SIZE) block which | |
d4906e1a LS |
1599 | * corresponds to page offset for the specified swap entry. |
1600 | * Note that the type of this function is sector_t, but it returns page offset | |
1601 | * into the bdev, not sector offset. | |
1da177e4 | 1602 | */ |
d4906e1a | 1603 | static sector_t map_swap_entry(swp_entry_t entry, struct block_device **bdev) |
1da177e4 | 1604 | { |
f29ad6a9 HD |
1605 | struct swap_info_struct *sis; |
1606 | struct swap_extent *start_se; | |
1607 | struct swap_extent *se; | |
1608 | pgoff_t offset; | |
1609 | ||
efa90a98 | 1610 | sis = swap_info[swp_type(entry)]; |
f29ad6a9 HD |
1611 | *bdev = sis->bdev; |
1612 | ||
1613 | offset = swp_offset(entry); | |
1614 | start_se = sis->curr_swap_extent; | |
1615 | se = start_se; | |
1da177e4 LT |
1616 | |
1617 | for ( ; ; ) { | |
1618 | struct list_head *lh; | |
1619 | ||
1620 | if (se->start_page <= offset && | |
1621 | offset < (se->start_page + se->nr_pages)) { | |
1622 | return se->start_block + (offset - se->start_page); | |
1623 | } | |
11d31886 | 1624 | lh = se->list.next; |
1da177e4 LT |
1625 | se = list_entry(lh, struct swap_extent, list); |
1626 | sis->curr_swap_extent = se; | |
1627 | BUG_ON(se == start_se); /* It *must* be present */ | |
1628 | } | |
1629 | } | |
1630 | ||
d4906e1a LS |
1631 | /* |
1632 | * Returns the page offset into bdev for the specified page's swap entry. | |
1633 | */ | |
1634 | sector_t map_swap_page(struct page *page, struct block_device **bdev) | |
1635 | { | |
1636 | swp_entry_t entry; | |
1637 | entry.val = page_private(page); | |
1638 | return map_swap_entry(entry, bdev); | |
1639 | } | |
1640 | ||
1da177e4 LT |
1641 | /* |
1642 | * Free all of a swapdev's extent information | |
1643 | */ | |
1644 | static void destroy_swap_extents(struct swap_info_struct *sis) | |
1645 | { | |
9625a5f2 | 1646 | while (!list_empty(&sis->first_swap_extent.list)) { |
1da177e4 LT |
1647 | struct swap_extent *se; |
1648 | ||
9625a5f2 | 1649 | se = list_entry(sis->first_swap_extent.list.next, |
1da177e4 LT |
1650 | struct swap_extent, list); |
1651 | list_del(&se->list); | |
1652 | kfree(se); | |
1653 | } | |
62c230bc MG |
1654 | |
1655 | if (sis->flags & SWP_FILE) { | |
1656 | struct file *swap_file = sis->swap_file; | |
1657 | struct address_space *mapping = swap_file->f_mapping; | |
1658 | ||
1659 | sis->flags &= ~SWP_FILE; | |
1660 | mapping->a_ops->swap_deactivate(swap_file); | |
1661 | } | |
1da177e4 LT |
1662 | } |
1663 | ||
1664 | /* | |
1665 | * Add a block range (and the corresponding page range) into this swapdev's | |
11d31886 | 1666 | * extent list. The extent list is kept sorted in page order. |
1da177e4 | 1667 | * |
11d31886 | 1668 | * This function rather assumes that it is called in ascending page order. |
1da177e4 | 1669 | */ |
a509bc1a | 1670 | int |
1da177e4 LT |
1671 | add_swap_extent(struct swap_info_struct *sis, unsigned long start_page, |
1672 | unsigned long nr_pages, sector_t start_block) | |
1673 | { | |
1674 | struct swap_extent *se; | |
1675 | struct swap_extent *new_se; | |
1676 | struct list_head *lh; | |
1677 | ||
9625a5f2 HD |
1678 | if (start_page == 0) { |
1679 | se = &sis->first_swap_extent; | |
1680 | sis->curr_swap_extent = se; | |
1681 | se->start_page = 0; | |
1682 | se->nr_pages = nr_pages; | |
1683 | se->start_block = start_block; | |
1684 | return 1; | |
1685 | } else { | |
1686 | lh = sis->first_swap_extent.list.prev; /* Highest extent */ | |
1da177e4 | 1687 | se = list_entry(lh, struct swap_extent, list); |
11d31886 HD |
1688 | BUG_ON(se->start_page + se->nr_pages != start_page); |
1689 | if (se->start_block + se->nr_pages == start_block) { | |
1da177e4 LT |
1690 | /* Merge it */ |
1691 | se->nr_pages += nr_pages; | |
1692 | return 0; | |
1693 | } | |
1da177e4 LT |
1694 | } |
1695 | ||
1696 | /* | |
1697 | * No merge. Insert a new extent, preserving ordering. | |
1698 | */ | |
1699 | new_se = kmalloc(sizeof(*se), GFP_KERNEL); | |
1700 | if (new_se == NULL) | |
1701 | return -ENOMEM; | |
1702 | new_se->start_page = start_page; | |
1703 | new_se->nr_pages = nr_pages; | |
1704 | new_se->start_block = start_block; | |
1705 | ||
9625a5f2 | 1706 | list_add_tail(&new_se->list, &sis->first_swap_extent.list); |
53092a74 | 1707 | return 1; |
1da177e4 LT |
1708 | } |
1709 | ||
1710 | /* | |
1711 | * A `swap extent' is a simple thing which maps a contiguous range of pages | |
1712 | * onto a contiguous range of disk blocks. An ordered list of swap extents | |
1713 | * is built at swapon time and is then used at swap_writepage/swap_readpage | |
1714 | * time for locating where on disk a page belongs. | |
1715 | * | |
1716 | * If the swapfile is an S_ISBLK block device, a single extent is installed. | |
1717 | * This is done so that the main operating code can treat S_ISBLK and S_ISREG | |
1718 | * swap files identically. | |
1719 | * | |
1720 | * Whether the swapdev is an S_ISREG file or an S_ISBLK blockdev, the swap | |
1721 | * extent list operates in PAGE_SIZE disk blocks. Both S_ISREG and S_ISBLK | |
1722 | * swapfiles are handled *identically* after swapon time. | |
1723 | * | |
1724 | * For S_ISREG swapfiles, setup_swap_extents() will walk all the file's blocks | |
1725 | * and will parse them into an ordered extent list, in PAGE_SIZE chunks. If | |
1726 | * some stray blocks are found which do not fall within the PAGE_SIZE alignment | |
1727 | * requirements, they are simply tossed out - we will never use those blocks | |
1728 | * for swapping. | |
1729 | * | |
b0d9bcd4 | 1730 | * For S_ISREG swapfiles we set S_SWAPFILE across the life of the swapon. This |
1da177e4 LT |
1731 | * prevents root from shooting her foot off by ftruncating an in-use swapfile, |
1732 | * which will scribble on the fs. | |
1733 | * | |
1734 | * The amount of disk space which a single swap extent represents varies. | |
1735 | * Typically it is in the 1-4 megabyte range. So we can have hundreds of | |
1736 | * extents in the list. To avoid much list walking, we cache the previous | |
1737 | * search location in `curr_swap_extent', and start new searches from there. | |
1738 | * This is extremely effective. The average number of iterations in | |
1739 | * map_swap_page() has been measured at about 0.3 per page. - akpm. | |
1740 | */ | |
53092a74 | 1741 | static int setup_swap_extents(struct swap_info_struct *sis, sector_t *span) |
1da177e4 | 1742 | { |
62c230bc MG |
1743 | struct file *swap_file = sis->swap_file; |
1744 | struct address_space *mapping = swap_file->f_mapping; | |
1745 | struct inode *inode = mapping->host; | |
1da177e4 LT |
1746 | int ret; |
1747 | ||
1da177e4 LT |
1748 | if (S_ISBLK(inode->i_mode)) { |
1749 | ret = add_swap_extent(sis, 0, sis->max, 0); | |
53092a74 | 1750 | *span = sis->pages; |
a509bc1a | 1751 | return ret; |
1da177e4 LT |
1752 | } |
1753 | ||
62c230bc | 1754 | if (mapping->a_ops->swap_activate) { |
a509bc1a | 1755 | ret = mapping->a_ops->swap_activate(sis, swap_file, span); |
62c230bc MG |
1756 | if (!ret) { |
1757 | sis->flags |= SWP_FILE; | |
1758 | ret = add_swap_extent(sis, 0, sis->max, 0); | |
1759 | *span = sis->pages; | |
1760 | } | |
a509bc1a | 1761 | return ret; |
62c230bc MG |
1762 | } |
1763 | ||
a509bc1a | 1764 | return generic_swapfile_activate(sis, swap_file, span); |
1da177e4 LT |
1765 | } |
1766 | ||
cf0cac0a | 1767 | static void _enable_swap_info(struct swap_info_struct *p, int prio, |
2a8f9449 SL |
1768 | unsigned char *swap_map, |
1769 | struct swap_cluster_info *cluster_info) | |
40531542 | 1770 | { |
40531542 CEB |
1771 | if (prio >= 0) |
1772 | p->prio = prio; | |
1773 | else | |
1774 | p->prio = --least_priority; | |
18ab4d4c DS |
1775 | /* |
1776 | * the plist prio is negated because plist ordering is | |
1777 | * low-to-high, while swap ordering is high-to-low | |
1778 | */ | |
1779 | p->list.prio = -p->prio; | |
1780 | p->avail_list.prio = -p->prio; | |
40531542 | 1781 | p->swap_map = swap_map; |
2a8f9449 | 1782 | p->cluster_info = cluster_info; |
40531542 | 1783 | p->flags |= SWP_WRITEOK; |
ec8acf20 | 1784 | atomic_long_add(p->pages, &nr_swap_pages); |
40531542 CEB |
1785 | total_swap_pages += p->pages; |
1786 | ||
adfab836 | 1787 | assert_spin_locked(&swap_lock); |
adfab836 | 1788 | /* |
18ab4d4c DS |
1789 | * both lists are plists, and thus priority ordered. |
1790 | * swap_active_head needs to be priority ordered for swapoff(), | |
1791 | * which on removal of any swap_info_struct with an auto-assigned | |
1792 | * (i.e. negative) priority increments the auto-assigned priority | |
1793 | * of any lower-priority swap_info_structs. | |
1794 | * swap_avail_head needs to be priority ordered for get_swap_page(), | |
1795 | * which allocates swap pages from the highest available priority | |
1796 | * swap_info_struct. | |
adfab836 | 1797 | */ |
18ab4d4c DS |
1798 | plist_add(&p->list, &swap_active_head); |
1799 | spin_lock(&swap_avail_lock); | |
1800 | plist_add(&p->avail_list, &swap_avail_head); | |
1801 | spin_unlock(&swap_avail_lock); | |
cf0cac0a CEB |
1802 | } |
1803 | ||
1804 | static void enable_swap_info(struct swap_info_struct *p, int prio, | |
1805 | unsigned char *swap_map, | |
2a8f9449 | 1806 | struct swap_cluster_info *cluster_info, |
cf0cac0a CEB |
1807 | unsigned long *frontswap_map) |
1808 | { | |
4f89849d | 1809 | frontswap_init(p->type, frontswap_map); |
cf0cac0a | 1810 | spin_lock(&swap_lock); |
ec8acf20 | 1811 | spin_lock(&p->lock); |
2a8f9449 | 1812 | _enable_swap_info(p, prio, swap_map, cluster_info); |
ec8acf20 | 1813 | spin_unlock(&p->lock); |
cf0cac0a CEB |
1814 | spin_unlock(&swap_lock); |
1815 | } | |
1816 | ||
1817 | static void reinsert_swap_info(struct swap_info_struct *p) | |
1818 | { | |
1819 | spin_lock(&swap_lock); | |
ec8acf20 | 1820 | spin_lock(&p->lock); |
2a8f9449 | 1821 | _enable_swap_info(p, p->prio, p->swap_map, p->cluster_info); |
ec8acf20 | 1822 | spin_unlock(&p->lock); |
40531542 CEB |
1823 | spin_unlock(&swap_lock); |
1824 | } | |
1825 | ||
c4ea37c2 | 1826 | SYSCALL_DEFINE1(swapoff, const char __user *, specialfile) |
1da177e4 | 1827 | { |
73c34b6a | 1828 | struct swap_info_struct *p = NULL; |
8d69aaee | 1829 | unsigned char *swap_map; |
2a8f9449 | 1830 | struct swap_cluster_info *cluster_info; |
4f89849d | 1831 | unsigned long *frontswap_map; |
1da177e4 LT |
1832 | struct file *swap_file, *victim; |
1833 | struct address_space *mapping; | |
1834 | struct inode *inode; | |
91a27b2a | 1835 | struct filename *pathname; |
adfab836 | 1836 | int err, found = 0; |
5b808a23 | 1837 | unsigned int old_block_size; |
886bb7e9 | 1838 | |
1da177e4 LT |
1839 | if (!capable(CAP_SYS_ADMIN)) |
1840 | return -EPERM; | |
1841 | ||
191c5424 AV |
1842 | BUG_ON(!current->mm); |
1843 | ||
1da177e4 | 1844 | pathname = getname(specialfile); |
1da177e4 | 1845 | if (IS_ERR(pathname)) |
f58b59c1 | 1846 | return PTR_ERR(pathname); |
1da177e4 | 1847 | |
669abf4e | 1848 | victim = file_open_name(pathname, O_RDWR|O_LARGEFILE, 0); |
1da177e4 LT |
1849 | err = PTR_ERR(victim); |
1850 | if (IS_ERR(victim)) | |
1851 | goto out; | |
1852 | ||
1853 | mapping = victim->f_mapping; | |
5d337b91 | 1854 | spin_lock(&swap_lock); |
18ab4d4c | 1855 | plist_for_each_entry(p, &swap_active_head, list) { |
22c6f8fd | 1856 | if (p->flags & SWP_WRITEOK) { |
adfab836 DS |
1857 | if (p->swap_file->f_mapping == mapping) { |
1858 | found = 1; | |
1da177e4 | 1859 | break; |
adfab836 | 1860 | } |
1da177e4 | 1861 | } |
1da177e4 | 1862 | } |
adfab836 | 1863 | if (!found) { |
1da177e4 | 1864 | err = -EINVAL; |
5d337b91 | 1865 | spin_unlock(&swap_lock); |
1da177e4 LT |
1866 | goto out_dput; |
1867 | } | |
191c5424 | 1868 | if (!security_vm_enough_memory_mm(current->mm, p->pages)) |
1da177e4 LT |
1869 | vm_unacct_memory(p->pages); |
1870 | else { | |
1871 | err = -ENOMEM; | |
5d337b91 | 1872 | spin_unlock(&swap_lock); |
1da177e4 LT |
1873 | goto out_dput; |
1874 | } | |
18ab4d4c DS |
1875 | spin_lock(&swap_avail_lock); |
1876 | plist_del(&p->avail_list, &swap_avail_head); | |
1877 | spin_unlock(&swap_avail_lock); | |
ec8acf20 | 1878 | spin_lock(&p->lock); |
78ecba08 | 1879 | if (p->prio < 0) { |
adfab836 DS |
1880 | struct swap_info_struct *si = p; |
1881 | ||
18ab4d4c | 1882 | plist_for_each_entry_continue(si, &swap_active_head, list) { |
adfab836 | 1883 | si->prio++; |
18ab4d4c DS |
1884 | si->list.prio--; |
1885 | si->avail_list.prio--; | |
adfab836 | 1886 | } |
78ecba08 HD |
1887 | least_priority++; |
1888 | } | |
18ab4d4c | 1889 | plist_del(&p->list, &swap_active_head); |
ec8acf20 | 1890 | atomic_long_sub(p->pages, &nr_swap_pages); |
1da177e4 LT |
1891 | total_swap_pages -= p->pages; |
1892 | p->flags &= ~SWP_WRITEOK; | |
ec8acf20 | 1893 | spin_unlock(&p->lock); |
5d337b91 | 1894 | spin_unlock(&swap_lock); |
fb4f88dc | 1895 | |
e1e12d2f | 1896 | set_current_oom_origin(); |
adfab836 | 1897 | err = try_to_unuse(p->type, false, 0); /* force unuse all pages */ |
e1e12d2f | 1898 | clear_current_oom_origin(); |
1da177e4 | 1899 | |
1da177e4 LT |
1900 | if (err) { |
1901 | /* re-insert swap space back into swap_list */ | |
cf0cac0a | 1902 | reinsert_swap_info(p); |
1da177e4 LT |
1903 | goto out_dput; |
1904 | } | |
52b7efdb | 1905 | |
815c2c54 SL |
1906 | flush_work(&p->discard_work); |
1907 | ||
5d337b91 | 1908 | destroy_swap_extents(p); |
570a335b HD |
1909 | if (p->flags & SWP_CONTINUED) |
1910 | free_swap_count_continuations(p); | |
1911 | ||
fc0abb14 | 1912 | mutex_lock(&swapon_mutex); |
5d337b91 | 1913 | spin_lock(&swap_lock); |
ec8acf20 | 1914 | spin_lock(&p->lock); |
5d337b91 HD |
1915 | drain_mmlist(); |
1916 | ||
52b7efdb | 1917 | /* wait for anyone still in scan_swap_map */ |
52b7efdb HD |
1918 | p->highest_bit = 0; /* cuts scans short */ |
1919 | while (p->flags >= SWP_SCANNING) { | |
ec8acf20 | 1920 | spin_unlock(&p->lock); |
5d337b91 | 1921 | spin_unlock(&swap_lock); |
13e4b57f | 1922 | schedule_timeout_uninterruptible(1); |
5d337b91 | 1923 | spin_lock(&swap_lock); |
ec8acf20 | 1924 | spin_lock(&p->lock); |
52b7efdb | 1925 | } |
52b7efdb | 1926 | |
1da177e4 | 1927 | swap_file = p->swap_file; |
5b808a23 | 1928 | old_block_size = p->old_block_size; |
1da177e4 LT |
1929 | p->swap_file = NULL; |
1930 | p->max = 0; | |
1931 | swap_map = p->swap_map; | |
1932 | p->swap_map = NULL; | |
2a8f9449 SL |
1933 | cluster_info = p->cluster_info; |
1934 | p->cluster_info = NULL; | |
4f89849d | 1935 | frontswap_map = frontswap_map_get(p); |
ec8acf20 | 1936 | spin_unlock(&p->lock); |
5d337b91 | 1937 | spin_unlock(&swap_lock); |
adfab836 | 1938 | frontswap_invalidate_area(p->type); |
58e97ba6 | 1939 | frontswap_map_set(p, NULL); |
fc0abb14 | 1940 | mutex_unlock(&swapon_mutex); |
ebc2a1a6 SL |
1941 | free_percpu(p->percpu_cluster); |
1942 | p->percpu_cluster = NULL; | |
1da177e4 | 1943 | vfree(swap_map); |
2a8f9449 | 1944 | vfree(cluster_info); |
4f89849d | 1945 | vfree(frontswap_map); |
2de1a7e4 | 1946 | /* Destroy swap account information */ |
adfab836 | 1947 | swap_cgroup_swapoff(p->type); |
27a7faa0 | 1948 | |
1da177e4 LT |
1949 | inode = mapping->host; |
1950 | if (S_ISBLK(inode->i_mode)) { | |
1951 | struct block_device *bdev = I_BDEV(inode); | |
5b808a23 | 1952 | set_blocksize(bdev, old_block_size); |
e525fd89 | 1953 | blkdev_put(bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL); |
1da177e4 | 1954 | } else { |
1b1dcc1b | 1955 | mutex_lock(&inode->i_mutex); |
1da177e4 | 1956 | inode->i_flags &= ~S_SWAPFILE; |
1b1dcc1b | 1957 | mutex_unlock(&inode->i_mutex); |
1da177e4 LT |
1958 | } |
1959 | filp_close(swap_file, NULL); | |
f893ab41 WY |
1960 | |
1961 | /* | |
1962 | * Clear the SWP_USED flag after all resources are freed so that swapon | |
1963 | * can reuse this swap_info in alloc_swap_info() safely. It is ok to | |
1964 | * not hold p->lock after we cleared its SWP_WRITEOK. | |
1965 | */ | |
1966 | spin_lock(&swap_lock); | |
1967 | p->flags = 0; | |
1968 | spin_unlock(&swap_lock); | |
1969 | ||
1da177e4 | 1970 | err = 0; |
66d7dd51 KS |
1971 | atomic_inc(&proc_poll_event); |
1972 | wake_up_interruptible(&proc_poll_wait); | |
1da177e4 LT |
1973 | |
1974 | out_dput: | |
1975 | filp_close(victim, NULL); | |
1976 | out: | |
f58b59c1 | 1977 | putname(pathname); |
1da177e4 LT |
1978 | return err; |
1979 | } | |
1980 | ||
1981 | #ifdef CONFIG_PROC_FS | |
66d7dd51 KS |
1982 | static unsigned swaps_poll(struct file *file, poll_table *wait) |
1983 | { | |
f1514638 | 1984 | struct seq_file *seq = file->private_data; |
66d7dd51 KS |
1985 | |
1986 | poll_wait(file, &proc_poll_wait, wait); | |
1987 | ||
f1514638 KS |
1988 | if (seq->poll_event != atomic_read(&proc_poll_event)) { |
1989 | seq->poll_event = atomic_read(&proc_poll_event); | |
66d7dd51 KS |
1990 | return POLLIN | POLLRDNORM | POLLERR | POLLPRI; |
1991 | } | |
1992 | ||
1993 | return POLLIN | POLLRDNORM; | |
1994 | } | |
1995 | ||
1da177e4 LT |
1996 | /* iterator */ |
1997 | static void *swap_start(struct seq_file *swap, loff_t *pos) | |
1998 | { | |
efa90a98 HD |
1999 | struct swap_info_struct *si; |
2000 | int type; | |
1da177e4 LT |
2001 | loff_t l = *pos; |
2002 | ||
fc0abb14 | 2003 | mutex_lock(&swapon_mutex); |
1da177e4 | 2004 | |
881e4aab SS |
2005 | if (!l) |
2006 | return SEQ_START_TOKEN; | |
2007 | ||
efa90a98 HD |
2008 | for (type = 0; type < nr_swapfiles; type++) { |
2009 | smp_rmb(); /* read nr_swapfiles before swap_info[type] */ | |
2010 | si = swap_info[type]; | |
2011 | if (!(si->flags & SWP_USED) || !si->swap_map) | |
1da177e4 | 2012 | continue; |
881e4aab | 2013 | if (!--l) |
efa90a98 | 2014 | return si; |
1da177e4 LT |
2015 | } |
2016 | ||
2017 | return NULL; | |
2018 | } | |
2019 | ||
2020 | static void *swap_next(struct seq_file *swap, void *v, loff_t *pos) | |
2021 | { | |
efa90a98 HD |
2022 | struct swap_info_struct *si = v; |
2023 | int type; | |
1da177e4 | 2024 | |
881e4aab | 2025 | if (v == SEQ_START_TOKEN) |
efa90a98 HD |
2026 | type = 0; |
2027 | else | |
2028 | type = si->type + 1; | |
881e4aab | 2029 | |
efa90a98 HD |
2030 | for (; type < nr_swapfiles; type++) { |
2031 | smp_rmb(); /* read nr_swapfiles before swap_info[type] */ | |
2032 | si = swap_info[type]; | |
2033 | if (!(si->flags & SWP_USED) || !si->swap_map) | |
1da177e4 LT |
2034 | continue; |
2035 | ++*pos; | |
efa90a98 | 2036 | return si; |
1da177e4 LT |
2037 | } |
2038 | ||
2039 | return NULL; | |
2040 | } | |
2041 | ||
2042 | static void swap_stop(struct seq_file *swap, void *v) | |
2043 | { | |
fc0abb14 | 2044 | mutex_unlock(&swapon_mutex); |
1da177e4 LT |
2045 | } |
2046 | ||
2047 | static int swap_show(struct seq_file *swap, void *v) | |
2048 | { | |
efa90a98 | 2049 | struct swap_info_struct *si = v; |
1da177e4 LT |
2050 | struct file *file; |
2051 | int len; | |
2052 | ||
efa90a98 | 2053 | if (si == SEQ_START_TOKEN) { |
881e4aab SS |
2054 | seq_puts(swap,"Filename\t\t\t\tType\t\tSize\tUsed\tPriority\n"); |
2055 | return 0; | |
2056 | } | |
1da177e4 | 2057 | |
efa90a98 | 2058 | file = si->swap_file; |
c32c2f63 | 2059 | len = seq_path(swap, &file->f_path, " \t\n\\"); |
6eb396dc | 2060 | seq_printf(swap, "%*s%s\t%u\t%u\t%d\n", |
886bb7e9 | 2061 | len < 40 ? 40 - len : 1, " ", |
496ad9aa | 2062 | S_ISBLK(file_inode(file)->i_mode) ? |
1da177e4 | 2063 | "partition" : "file\t", |
efa90a98 HD |
2064 | si->pages << (PAGE_SHIFT - 10), |
2065 | si->inuse_pages << (PAGE_SHIFT - 10), | |
2066 | si->prio); | |
1da177e4 LT |
2067 | return 0; |
2068 | } | |
2069 | ||
15ad7cdc | 2070 | static const struct seq_operations swaps_op = { |
1da177e4 LT |
2071 | .start = swap_start, |
2072 | .next = swap_next, | |
2073 | .stop = swap_stop, | |
2074 | .show = swap_show | |
2075 | }; | |
2076 | ||
2077 | static int swaps_open(struct inode *inode, struct file *file) | |
2078 | { | |
f1514638 | 2079 | struct seq_file *seq; |
66d7dd51 KS |
2080 | int ret; |
2081 | ||
66d7dd51 | 2082 | ret = seq_open(file, &swaps_op); |
f1514638 | 2083 | if (ret) |
66d7dd51 | 2084 | return ret; |
66d7dd51 | 2085 | |
f1514638 KS |
2086 | seq = file->private_data; |
2087 | seq->poll_event = atomic_read(&proc_poll_event); | |
2088 | return 0; | |
1da177e4 LT |
2089 | } |
2090 | ||
15ad7cdc | 2091 | static const struct file_operations proc_swaps_operations = { |
1da177e4 LT |
2092 | .open = swaps_open, |
2093 | .read = seq_read, | |
2094 | .llseek = seq_lseek, | |
2095 | .release = seq_release, | |
66d7dd51 | 2096 | .poll = swaps_poll, |
1da177e4 LT |
2097 | }; |
2098 | ||
2099 | static int __init procswaps_init(void) | |
2100 | { | |
3d71f86f | 2101 | proc_create("swaps", 0, NULL, &proc_swaps_operations); |
1da177e4 LT |
2102 | return 0; |
2103 | } | |
2104 | __initcall(procswaps_init); | |
2105 | #endif /* CONFIG_PROC_FS */ | |
2106 | ||
1796316a JB |
2107 | #ifdef MAX_SWAPFILES_CHECK |
2108 | static int __init max_swapfiles_check(void) | |
2109 | { | |
2110 | MAX_SWAPFILES_CHECK(); | |
2111 | return 0; | |
2112 | } | |
2113 | late_initcall(max_swapfiles_check); | |
2114 | #endif | |
2115 | ||
53cbb243 | 2116 | static struct swap_info_struct *alloc_swap_info(void) |
1da177e4 | 2117 | { |
73c34b6a | 2118 | struct swap_info_struct *p; |
1da177e4 | 2119 | unsigned int type; |
efa90a98 HD |
2120 | |
2121 | p = kzalloc(sizeof(*p), GFP_KERNEL); | |
2122 | if (!p) | |
53cbb243 | 2123 | return ERR_PTR(-ENOMEM); |
efa90a98 | 2124 | |
5d337b91 | 2125 | spin_lock(&swap_lock); |
efa90a98 HD |
2126 | for (type = 0; type < nr_swapfiles; type++) { |
2127 | if (!(swap_info[type]->flags & SWP_USED)) | |
1da177e4 | 2128 | break; |
efa90a98 | 2129 | } |
0697212a | 2130 | if (type >= MAX_SWAPFILES) { |
5d337b91 | 2131 | spin_unlock(&swap_lock); |
efa90a98 | 2132 | kfree(p); |
730c0581 | 2133 | return ERR_PTR(-EPERM); |
1da177e4 | 2134 | } |
efa90a98 HD |
2135 | if (type >= nr_swapfiles) { |
2136 | p->type = type; | |
2137 | swap_info[type] = p; | |
2138 | /* | |
2139 | * Write swap_info[type] before nr_swapfiles, in case a | |
2140 | * racing procfs swap_start() or swap_next() is reading them. | |
2141 | * (We never shrink nr_swapfiles, we never free this entry.) | |
2142 | */ | |
2143 | smp_wmb(); | |
2144 | nr_swapfiles++; | |
2145 | } else { | |
2146 | kfree(p); | |
2147 | p = swap_info[type]; | |
2148 | /* | |
2149 | * Do not memset this entry: a racing procfs swap_next() | |
2150 | * would be relying on p->type to remain valid. | |
2151 | */ | |
2152 | } | |
9625a5f2 | 2153 | INIT_LIST_HEAD(&p->first_swap_extent.list); |
18ab4d4c DS |
2154 | plist_node_init(&p->list, 0); |
2155 | plist_node_init(&p->avail_list, 0); | |
1da177e4 | 2156 | p->flags = SWP_USED; |
5d337b91 | 2157 | spin_unlock(&swap_lock); |
ec8acf20 | 2158 | spin_lock_init(&p->lock); |
efa90a98 | 2159 | |
53cbb243 | 2160 | return p; |
53cbb243 CEB |
2161 | } |
2162 | ||
4d0e1e10 CEB |
2163 | static int claim_swapfile(struct swap_info_struct *p, struct inode *inode) |
2164 | { | |
2165 | int error; | |
2166 | ||
2167 | if (S_ISBLK(inode->i_mode)) { | |
2168 | p->bdev = bdgrab(I_BDEV(inode)); | |
2169 | error = blkdev_get(p->bdev, | |
2170 | FMODE_READ | FMODE_WRITE | FMODE_EXCL, | |
2171 | sys_swapon); | |
2172 | if (error < 0) { | |
2173 | p->bdev = NULL; | |
87ade72a | 2174 | return -EINVAL; |
4d0e1e10 CEB |
2175 | } |
2176 | p->old_block_size = block_size(p->bdev); | |
2177 | error = set_blocksize(p->bdev, PAGE_SIZE); | |
2178 | if (error < 0) | |
87ade72a | 2179 | return error; |
4d0e1e10 CEB |
2180 | p->flags |= SWP_BLKDEV; |
2181 | } else if (S_ISREG(inode->i_mode)) { | |
2182 | p->bdev = inode->i_sb->s_bdev; | |
2183 | mutex_lock(&inode->i_mutex); | |
87ade72a CEB |
2184 | if (IS_SWAPFILE(inode)) |
2185 | return -EBUSY; | |
2186 | } else | |
2187 | return -EINVAL; | |
4d0e1e10 CEB |
2188 | |
2189 | return 0; | |
4d0e1e10 CEB |
2190 | } |
2191 | ||
ca8bd38b CEB |
2192 | static unsigned long read_swap_header(struct swap_info_struct *p, |
2193 | union swap_header *swap_header, | |
2194 | struct inode *inode) | |
2195 | { | |
2196 | int i; | |
2197 | unsigned long maxpages; | |
2198 | unsigned long swapfilepages; | |
d6bbbd29 | 2199 | unsigned long last_page; |
ca8bd38b CEB |
2200 | |
2201 | if (memcmp("SWAPSPACE2", swap_header->magic.magic, 10)) { | |
465c47fd | 2202 | pr_err("Unable to find swap-space signature\n"); |
38719025 | 2203 | return 0; |
ca8bd38b CEB |
2204 | } |
2205 | ||
2206 | /* swap partition endianess hack... */ | |
2207 | if (swab32(swap_header->info.version) == 1) { | |
2208 | swab32s(&swap_header->info.version); | |
2209 | swab32s(&swap_header->info.last_page); | |
2210 | swab32s(&swap_header->info.nr_badpages); | |
2211 | for (i = 0; i < swap_header->info.nr_badpages; i++) | |
2212 | swab32s(&swap_header->info.badpages[i]); | |
2213 | } | |
2214 | /* Check the swap header's sub-version */ | |
2215 | if (swap_header->info.version != 1) { | |
465c47fd AM |
2216 | pr_warn("Unable to handle swap header version %d\n", |
2217 | swap_header->info.version); | |
38719025 | 2218 | return 0; |
ca8bd38b CEB |
2219 | } |
2220 | ||
2221 | p->lowest_bit = 1; | |
2222 | p->cluster_next = 1; | |
2223 | p->cluster_nr = 0; | |
2224 | ||
2225 | /* | |
2226 | * Find out how many pages are allowed for a single swap | |
9b15b817 | 2227 | * device. There are two limiting factors: 1) the number |
a2c16d6c HD |
2228 | * of bits for the swap offset in the swp_entry_t type, and |
2229 | * 2) the number of bits in the swap pte as defined by the | |
9b15b817 | 2230 | * different architectures. In order to find the |
a2c16d6c | 2231 | * largest possible bit mask, a swap entry with swap type 0 |
ca8bd38b | 2232 | * and swap offset ~0UL is created, encoded to a swap pte, |
a2c16d6c | 2233 | * decoded to a swp_entry_t again, and finally the swap |
ca8bd38b CEB |
2234 | * offset is extracted. This will mask all the bits from |
2235 | * the initial ~0UL mask that can't be encoded in either | |
2236 | * the swp_entry_t or the architecture definition of a | |
9b15b817 | 2237 | * swap pte. |
ca8bd38b CEB |
2238 | */ |
2239 | maxpages = swp_offset(pte_to_swp_entry( | |
9b15b817 | 2240 | swp_entry_to_pte(swp_entry(0, ~0UL)))) + 1; |
d6bbbd29 RJ |
2241 | last_page = swap_header->info.last_page; |
2242 | if (last_page > maxpages) { | |
465c47fd | 2243 | pr_warn("Truncating oversized swap area, only using %luk out of %luk\n", |
d6bbbd29 RJ |
2244 | maxpages << (PAGE_SHIFT - 10), |
2245 | last_page << (PAGE_SHIFT - 10)); | |
2246 | } | |
2247 | if (maxpages > last_page) { | |
2248 | maxpages = last_page + 1; | |
ca8bd38b CEB |
2249 | /* p->max is an unsigned int: don't overflow it */ |
2250 | if ((unsigned int)maxpages == 0) | |
2251 | maxpages = UINT_MAX; | |
2252 | } | |
2253 | p->highest_bit = maxpages - 1; | |
2254 | ||
2255 | if (!maxpages) | |
38719025 | 2256 | return 0; |
ca8bd38b CEB |
2257 | swapfilepages = i_size_read(inode) >> PAGE_SHIFT; |
2258 | if (swapfilepages && maxpages > swapfilepages) { | |
465c47fd | 2259 | pr_warn("Swap area shorter than signature indicates\n"); |
38719025 | 2260 | return 0; |
ca8bd38b CEB |
2261 | } |
2262 | if (swap_header->info.nr_badpages && S_ISREG(inode->i_mode)) | |
38719025 | 2263 | return 0; |
ca8bd38b | 2264 | if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES) |
38719025 | 2265 | return 0; |
ca8bd38b CEB |
2266 | |
2267 | return maxpages; | |
ca8bd38b CEB |
2268 | } |
2269 | ||
915d4d7b CEB |
2270 | static int setup_swap_map_and_extents(struct swap_info_struct *p, |
2271 | union swap_header *swap_header, | |
2272 | unsigned char *swap_map, | |
2a8f9449 | 2273 | struct swap_cluster_info *cluster_info, |
915d4d7b CEB |
2274 | unsigned long maxpages, |
2275 | sector_t *span) | |
2276 | { | |
2277 | int i; | |
915d4d7b CEB |
2278 | unsigned int nr_good_pages; |
2279 | int nr_extents; | |
2a8f9449 SL |
2280 | unsigned long nr_clusters = DIV_ROUND_UP(maxpages, SWAPFILE_CLUSTER); |
2281 | unsigned long idx = p->cluster_next / SWAPFILE_CLUSTER; | |
915d4d7b CEB |
2282 | |
2283 | nr_good_pages = maxpages - 1; /* omit header page */ | |
2284 | ||
2a8f9449 SL |
2285 | cluster_set_null(&p->free_cluster_head); |
2286 | cluster_set_null(&p->free_cluster_tail); | |
815c2c54 SL |
2287 | cluster_set_null(&p->discard_cluster_head); |
2288 | cluster_set_null(&p->discard_cluster_tail); | |
2a8f9449 | 2289 | |
915d4d7b CEB |
2290 | for (i = 0; i < swap_header->info.nr_badpages; i++) { |
2291 | unsigned int page_nr = swap_header->info.badpages[i]; | |
bdb8e3f6 CEB |
2292 | if (page_nr == 0 || page_nr > swap_header->info.last_page) |
2293 | return -EINVAL; | |
915d4d7b CEB |
2294 | if (page_nr < maxpages) { |
2295 | swap_map[page_nr] = SWAP_MAP_BAD; | |
2296 | nr_good_pages--; | |
2a8f9449 SL |
2297 | /* |
2298 | * Haven't marked the cluster free yet, no list | |
2299 | * operation involved | |
2300 | */ | |
2301 | inc_cluster_info_page(p, cluster_info, page_nr); | |
915d4d7b CEB |
2302 | } |
2303 | } | |
2304 | ||
2a8f9449 SL |
2305 | /* Haven't marked the cluster free yet, no list operation involved */ |
2306 | for (i = maxpages; i < round_up(maxpages, SWAPFILE_CLUSTER); i++) | |
2307 | inc_cluster_info_page(p, cluster_info, i); | |
2308 | ||
915d4d7b CEB |
2309 | if (nr_good_pages) { |
2310 | swap_map[0] = SWAP_MAP_BAD; | |
2a8f9449 SL |
2311 | /* |
2312 | * Not mark the cluster free yet, no list | |
2313 | * operation involved | |
2314 | */ | |
2315 | inc_cluster_info_page(p, cluster_info, 0); | |
915d4d7b CEB |
2316 | p->max = maxpages; |
2317 | p->pages = nr_good_pages; | |
2318 | nr_extents = setup_swap_extents(p, span); | |
bdb8e3f6 CEB |
2319 | if (nr_extents < 0) |
2320 | return nr_extents; | |
915d4d7b CEB |
2321 | nr_good_pages = p->pages; |
2322 | } | |
2323 | if (!nr_good_pages) { | |
465c47fd | 2324 | pr_warn("Empty swap-file\n"); |
bdb8e3f6 | 2325 | return -EINVAL; |
915d4d7b CEB |
2326 | } |
2327 | ||
2a8f9449 SL |
2328 | if (!cluster_info) |
2329 | return nr_extents; | |
2330 | ||
2331 | for (i = 0; i < nr_clusters; i++) { | |
2332 | if (!cluster_count(&cluster_info[idx])) { | |
2333 | cluster_set_flag(&cluster_info[idx], CLUSTER_FLAG_FREE); | |
2334 | if (cluster_is_null(&p->free_cluster_head)) { | |
2335 | cluster_set_next_flag(&p->free_cluster_head, | |
2336 | idx, 0); | |
2337 | cluster_set_next_flag(&p->free_cluster_tail, | |
2338 | idx, 0); | |
2339 | } else { | |
2340 | unsigned int tail; | |
2341 | ||
2342 | tail = cluster_next(&p->free_cluster_tail); | |
2343 | cluster_set_next(&cluster_info[tail], idx); | |
2344 | cluster_set_next_flag(&p->free_cluster_tail, | |
2345 | idx, 0); | |
2346 | } | |
2347 | } | |
2348 | idx++; | |
2349 | if (idx == nr_clusters) | |
2350 | idx = 0; | |
2351 | } | |
915d4d7b | 2352 | return nr_extents; |
915d4d7b CEB |
2353 | } |
2354 | ||
dcf6b7dd RA |
2355 | /* |
2356 | * Helper to sys_swapon determining if a given swap | |
2357 | * backing device queue supports DISCARD operations. | |
2358 | */ | |
2359 | static bool swap_discardable(struct swap_info_struct *si) | |
2360 | { | |
2361 | struct request_queue *q = bdev_get_queue(si->bdev); | |
2362 | ||
2363 | if (!q || !blk_queue_discard(q)) | |
2364 | return false; | |
2365 | ||
2366 | return true; | |
2367 | } | |
2368 | ||
53cbb243 CEB |
2369 | SYSCALL_DEFINE2(swapon, const char __user *, specialfile, int, swap_flags) |
2370 | { | |
2371 | struct swap_info_struct *p; | |
91a27b2a | 2372 | struct filename *name; |
53cbb243 CEB |
2373 | struct file *swap_file = NULL; |
2374 | struct address_space *mapping; | |
40531542 CEB |
2375 | int i; |
2376 | int prio; | |
53cbb243 CEB |
2377 | int error; |
2378 | union swap_header *swap_header; | |
915d4d7b | 2379 | int nr_extents; |
53cbb243 CEB |
2380 | sector_t span; |
2381 | unsigned long maxpages; | |
53cbb243 | 2382 | unsigned char *swap_map = NULL; |
2a8f9449 | 2383 | struct swap_cluster_info *cluster_info = NULL; |
38b5faf4 | 2384 | unsigned long *frontswap_map = NULL; |
53cbb243 CEB |
2385 | struct page *page = NULL; |
2386 | struct inode *inode = NULL; | |
53cbb243 | 2387 | |
d15cab97 HD |
2388 | if (swap_flags & ~SWAP_FLAGS_VALID) |
2389 | return -EINVAL; | |
2390 | ||
53cbb243 CEB |
2391 | if (!capable(CAP_SYS_ADMIN)) |
2392 | return -EPERM; | |
2393 | ||
2394 | p = alloc_swap_info(); | |
2542e513 CEB |
2395 | if (IS_ERR(p)) |
2396 | return PTR_ERR(p); | |
53cbb243 | 2397 | |
815c2c54 SL |
2398 | INIT_WORK(&p->discard_work, swap_discard_work); |
2399 | ||
1da177e4 | 2400 | name = getname(specialfile); |
1da177e4 | 2401 | if (IS_ERR(name)) { |
7de7fb6b | 2402 | error = PTR_ERR(name); |
1da177e4 | 2403 | name = NULL; |
bd69010b | 2404 | goto bad_swap; |
1da177e4 | 2405 | } |
669abf4e | 2406 | swap_file = file_open_name(name, O_RDWR|O_LARGEFILE, 0); |
1da177e4 | 2407 | if (IS_ERR(swap_file)) { |
7de7fb6b | 2408 | error = PTR_ERR(swap_file); |
1da177e4 | 2409 | swap_file = NULL; |
bd69010b | 2410 | goto bad_swap; |
1da177e4 LT |
2411 | } |
2412 | ||
2413 | p->swap_file = swap_file; | |
2414 | mapping = swap_file->f_mapping; | |
1da177e4 | 2415 | |
1da177e4 | 2416 | for (i = 0; i < nr_swapfiles; i++) { |
efa90a98 | 2417 | struct swap_info_struct *q = swap_info[i]; |
1da177e4 | 2418 | |
e8e6c2ec | 2419 | if (q == p || !q->swap_file) |
1da177e4 | 2420 | continue; |
7de7fb6b CEB |
2421 | if (mapping == q->swap_file->f_mapping) { |
2422 | error = -EBUSY; | |
1da177e4 | 2423 | goto bad_swap; |
7de7fb6b | 2424 | } |
1da177e4 LT |
2425 | } |
2426 | ||
2130781e CEB |
2427 | inode = mapping->host; |
2428 | /* If S_ISREG(inode->i_mode) will do mutex_lock(&inode->i_mutex); */ | |
4d0e1e10 CEB |
2429 | error = claim_swapfile(p, inode); |
2430 | if (unlikely(error)) | |
1da177e4 | 2431 | goto bad_swap; |
1da177e4 | 2432 | |
1da177e4 LT |
2433 | /* |
2434 | * Read the swap header. | |
2435 | */ | |
2436 | if (!mapping->a_ops->readpage) { | |
2437 | error = -EINVAL; | |
2438 | goto bad_swap; | |
2439 | } | |
090d2b18 | 2440 | page = read_mapping_page(mapping, 0, swap_file); |
1da177e4 LT |
2441 | if (IS_ERR(page)) { |
2442 | error = PTR_ERR(page); | |
2443 | goto bad_swap; | |
2444 | } | |
81e33971 | 2445 | swap_header = kmap(page); |
1da177e4 | 2446 | |
ca8bd38b CEB |
2447 | maxpages = read_swap_header(p, swap_header, inode); |
2448 | if (unlikely(!maxpages)) { | |
1da177e4 LT |
2449 | error = -EINVAL; |
2450 | goto bad_swap; | |
2451 | } | |
886bb7e9 | 2452 | |
81e33971 | 2453 | /* OK, set up the swap map and apply the bad block list */ |
803d0c83 | 2454 | swap_map = vzalloc(maxpages); |
81e33971 HD |
2455 | if (!swap_map) { |
2456 | error = -ENOMEM; | |
2457 | goto bad_swap; | |
2458 | } | |
2a8f9449 SL |
2459 | if (p->bdev && blk_queue_nonrot(bdev_get_queue(p->bdev))) { |
2460 | p->flags |= SWP_SOLIDSTATE; | |
2461 | /* | |
2462 | * select a random position to start with to help wear leveling | |
2463 | * SSD | |
2464 | */ | |
2465 | p->cluster_next = 1 + (prandom_u32() % p->highest_bit); | |
2466 | ||
2467 | cluster_info = vzalloc(DIV_ROUND_UP(maxpages, | |
2468 | SWAPFILE_CLUSTER) * sizeof(*cluster_info)); | |
2469 | if (!cluster_info) { | |
2470 | error = -ENOMEM; | |
2471 | goto bad_swap; | |
2472 | } | |
ebc2a1a6 SL |
2473 | p->percpu_cluster = alloc_percpu(struct percpu_cluster); |
2474 | if (!p->percpu_cluster) { | |
2475 | error = -ENOMEM; | |
2476 | goto bad_swap; | |
2477 | } | |
2478 | for_each_possible_cpu(i) { | |
2479 | struct percpu_cluster *cluster; | |
2480 | cluster = per_cpu_ptr(p->percpu_cluster, i); | |
2481 | cluster_set_null(&cluster->index); | |
2482 | } | |
2a8f9449 | 2483 | } |
1da177e4 | 2484 | |
1421ef3c CEB |
2485 | error = swap_cgroup_swapon(p->type, maxpages); |
2486 | if (error) | |
2487 | goto bad_swap; | |
2488 | ||
915d4d7b | 2489 | nr_extents = setup_swap_map_and_extents(p, swap_header, swap_map, |
2a8f9449 | 2490 | cluster_info, maxpages, &span); |
915d4d7b CEB |
2491 | if (unlikely(nr_extents < 0)) { |
2492 | error = nr_extents; | |
1da177e4 LT |
2493 | goto bad_swap; |
2494 | } | |
38b5faf4 DM |
2495 | /* frontswap enabled? set up bit-per-page map for frontswap */ |
2496 | if (frontswap_enabled) | |
7b57976d | 2497 | frontswap_map = vzalloc(BITS_TO_LONGS(maxpages) * sizeof(long)); |
1da177e4 | 2498 | |
2a8f9449 SL |
2499 | if (p->bdev &&(swap_flags & SWAP_FLAG_DISCARD) && swap_discardable(p)) { |
2500 | /* | |
2501 | * When discard is enabled for swap with no particular | |
2502 | * policy flagged, we set all swap discard flags here in | |
2503 | * order to sustain backward compatibility with older | |
2504 | * swapon(8) releases. | |
2505 | */ | |
2506 | p->flags |= (SWP_DISCARDABLE | SWP_AREA_DISCARD | | |
2507 | SWP_PAGE_DISCARD); | |
dcf6b7dd | 2508 | |
2a8f9449 SL |
2509 | /* |
2510 | * By flagging sys_swapon, a sysadmin can tell us to | |
2511 | * either do single-time area discards only, or to just | |
2512 | * perform discards for released swap page-clusters. | |
2513 | * Now it's time to adjust the p->flags accordingly. | |
2514 | */ | |
2515 | if (swap_flags & SWAP_FLAG_DISCARD_ONCE) | |
2516 | p->flags &= ~SWP_PAGE_DISCARD; | |
2517 | else if (swap_flags & SWAP_FLAG_DISCARD_PAGES) | |
2518 | p->flags &= ~SWP_AREA_DISCARD; | |
2519 | ||
2520 | /* issue a swapon-time discard if it's still required */ | |
2521 | if (p->flags & SWP_AREA_DISCARD) { | |
2522 | int err = discard_swap(p); | |
2523 | if (unlikely(err)) | |
2524 | pr_err("swapon: discard_swap(%p): %d\n", | |
2525 | p, err); | |
dcf6b7dd | 2526 | } |
20137a49 | 2527 | } |
6a6ba831 | 2528 | |
fc0abb14 | 2529 | mutex_lock(&swapon_mutex); |
40531542 | 2530 | prio = -1; |
78ecba08 | 2531 | if (swap_flags & SWAP_FLAG_PREFER) |
40531542 | 2532 | prio = |
78ecba08 | 2533 | (swap_flags & SWAP_FLAG_PRIO_MASK) >> SWAP_FLAG_PRIO_SHIFT; |
2a8f9449 | 2534 | enable_swap_info(p, prio, swap_map, cluster_info, frontswap_map); |
c69dbfb8 | 2535 | |
465c47fd | 2536 | pr_info("Adding %uk swap on %s. " |
dcf6b7dd | 2537 | "Priority:%d extents:%d across:%lluk %s%s%s%s%s\n", |
91a27b2a | 2538 | p->pages<<(PAGE_SHIFT-10), name->name, p->prio, |
c69dbfb8 CEB |
2539 | nr_extents, (unsigned long long)span<<(PAGE_SHIFT-10), |
2540 | (p->flags & SWP_SOLIDSTATE) ? "SS" : "", | |
38b5faf4 | 2541 | (p->flags & SWP_DISCARDABLE) ? "D" : "", |
dcf6b7dd RA |
2542 | (p->flags & SWP_AREA_DISCARD) ? "s" : "", |
2543 | (p->flags & SWP_PAGE_DISCARD) ? "c" : "", | |
38b5faf4 | 2544 | (frontswap_map) ? "FS" : ""); |
c69dbfb8 | 2545 | |
fc0abb14 | 2546 | mutex_unlock(&swapon_mutex); |
66d7dd51 KS |
2547 | atomic_inc(&proc_poll_event); |
2548 | wake_up_interruptible(&proc_poll_wait); | |
2549 | ||
9b01c350 CEB |
2550 | if (S_ISREG(inode->i_mode)) |
2551 | inode->i_flags |= S_SWAPFILE; | |
1da177e4 LT |
2552 | error = 0; |
2553 | goto out; | |
2554 | bad_swap: | |
ebc2a1a6 SL |
2555 | free_percpu(p->percpu_cluster); |
2556 | p->percpu_cluster = NULL; | |
bd69010b | 2557 | if (inode && S_ISBLK(inode->i_mode) && p->bdev) { |
f2090d2d CEB |
2558 | set_blocksize(p->bdev, p->old_block_size); |
2559 | blkdev_put(p->bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL); | |
1da177e4 | 2560 | } |
4cd3bb10 | 2561 | destroy_swap_extents(p); |
e8e6c2ec | 2562 | swap_cgroup_swapoff(p->type); |
5d337b91 | 2563 | spin_lock(&swap_lock); |
1da177e4 | 2564 | p->swap_file = NULL; |
1da177e4 | 2565 | p->flags = 0; |
5d337b91 | 2566 | spin_unlock(&swap_lock); |
1da177e4 | 2567 | vfree(swap_map); |
2a8f9449 | 2568 | vfree(cluster_info); |
52c50567 | 2569 | if (swap_file) { |
2130781e | 2570 | if (inode && S_ISREG(inode->i_mode)) { |
52c50567 | 2571 | mutex_unlock(&inode->i_mutex); |
2130781e CEB |
2572 | inode = NULL; |
2573 | } | |
1da177e4 | 2574 | filp_close(swap_file, NULL); |
52c50567 | 2575 | } |
1da177e4 LT |
2576 | out: |
2577 | if (page && !IS_ERR(page)) { | |
2578 | kunmap(page); | |
2579 | page_cache_release(page); | |
2580 | } | |
2581 | if (name) | |
2582 | putname(name); | |
9b01c350 | 2583 | if (inode && S_ISREG(inode->i_mode)) |
1b1dcc1b | 2584 | mutex_unlock(&inode->i_mutex); |
1da177e4 LT |
2585 | return error; |
2586 | } | |
2587 | ||
2588 | void si_swapinfo(struct sysinfo *val) | |
2589 | { | |
efa90a98 | 2590 | unsigned int type; |
1da177e4 LT |
2591 | unsigned long nr_to_be_unused = 0; |
2592 | ||
5d337b91 | 2593 | spin_lock(&swap_lock); |
efa90a98 HD |
2594 | for (type = 0; type < nr_swapfiles; type++) { |
2595 | struct swap_info_struct *si = swap_info[type]; | |
2596 | ||
2597 | if ((si->flags & SWP_USED) && !(si->flags & SWP_WRITEOK)) | |
2598 | nr_to_be_unused += si->inuse_pages; | |
1da177e4 | 2599 | } |
ec8acf20 | 2600 | val->freeswap = atomic_long_read(&nr_swap_pages) + nr_to_be_unused; |
1da177e4 | 2601 | val->totalswap = total_swap_pages + nr_to_be_unused; |
5d337b91 | 2602 | spin_unlock(&swap_lock); |
1da177e4 LT |
2603 | } |
2604 | ||
2605 | /* | |
2606 | * Verify that a swap entry is valid and increment its swap map count. | |
2607 | * | |
355cfa73 KH |
2608 | * Returns error code in following case. |
2609 | * - success -> 0 | |
2610 | * - swp_entry is invalid -> EINVAL | |
2611 | * - swp_entry is migration entry -> EINVAL | |
2612 | * - swap-cache reference is requested but there is already one. -> EEXIST | |
2613 | * - swap-cache reference is requested but the entry is not used. -> ENOENT | |
570a335b | 2614 | * - swap-mapped reference requested but needs continued swap count. -> ENOMEM |
1da177e4 | 2615 | */ |
8d69aaee | 2616 | static int __swap_duplicate(swp_entry_t entry, unsigned char usage) |
1da177e4 | 2617 | { |
73c34b6a | 2618 | struct swap_info_struct *p; |
1da177e4 | 2619 | unsigned long offset, type; |
8d69aaee HD |
2620 | unsigned char count; |
2621 | unsigned char has_cache; | |
253d553b | 2622 | int err = -EINVAL; |
1da177e4 | 2623 | |
a7420aa5 | 2624 | if (non_swap_entry(entry)) |
253d553b | 2625 | goto out; |
0697212a | 2626 | |
1da177e4 LT |
2627 | type = swp_type(entry); |
2628 | if (type >= nr_swapfiles) | |
2629 | goto bad_file; | |
efa90a98 | 2630 | p = swap_info[type]; |
1da177e4 LT |
2631 | offset = swp_offset(entry); |
2632 | ||
ec8acf20 | 2633 | spin_lock(&p->lock); |
355cfa73 KH |
2634 | if (unlikely(offset >= p->max)) |
2635 | goto unlock_out; | |
2636 | ||
253d553b | 2637 | count = p->swap_map[offset]; |
edfe23da SL |
2638 | |
2639 | /* | |
2640 | * swapin_readahead() doesn't check if a swap entry is valid, so the | |
2641 | * swap entry could be SWAP_MAP_BAD. Check here with lock held. | |
2642 | */ | |
2643 | if (unlikely(swap_count(count) == SWAP_MAP_BAD)) { | |
2644 | err = -ENOENT; | |
2645 | goto unlock_out; | |
2646 | } | |
2647 | ||
253d553b HD |
2648 | has_cache = count & SWAP_HAS_CACHE; |
2649 | count &= ~SWAP_HAS_CACHE; | |
2650 | err = 0; | |
355cfa73 | 2651 | |
253d553b | 2652 | if (usage == SWAP_HAS_CACHE) { |
355cfa73 KH |
2653 | |
2654 | /* set SWAP_HAS_CACHE if there is no cache and entry is used */ | |
253d553b HD |
2655 | if (!has_cache && count) |
2656 | has_cache = SWAP_HAS_CACHE; | |
2657 | else if (has_cache) /* someone else added cache */ | |
2658 | err = -EEXIST; | |
2659 | else /* no users remaining */ | |
2660 | err = -ENOENT; | |
355cfa73 KH |
2661 | |
2662 | } else if (count || has_cache) { | |
253d553b | 2663 | |
570a335b HD |
2664 | if ((count & ~COUNT_CONTINUED) < SWAP_MAP_MAX) |
2665 | count += usage; | |
2666 | else if ((count & ~COUNT_CONTINUED) > SWAP_MAP_MAX) | |
253d553b | 2667 | err = -EINVAL; |
570a335b HD |
2668 | else if (swap_count_continued(p, offset, count)) |
2669 | count = COUNT_CONTINUED; | |
2670 | else | |
2671 | err = -ENOMEM; | |
355cfa73 | 2672 | } else |
253d553b HD |
2673 | err = -ENOENT; /* unused swap entry */ |
2674 | ||
2675 | p->swap_map[offset] = count | has_cache; | |
2676 | ||
355cfa73 | 2677 | unlock_out: |
ec8acf20 | 2678 | spin_unlock(&p->lock); |
1da177e4 | 2679 | out: |
253d553b | 2680 | return err; |
1da177e4 LT |
2681 | |
2682 | bad_file: | |
465c47fd | 2683 | pr_err("swap_dup: %s%08lx\n", Bad_file, entry.val); |
1da177e4 LT |
2684 | goto out; |
2685 | } | |
253d553b | 2686 | |
aaa46865 HD |
2687 | /* |
2688 | * Help swapoff by noting that swap entry belongs to shmem/tmpfs | |
2689 | * (in which case its reference count is never incremented). | |
2690 | */ | |
2691 | void swap_shmem_alloc(swp_entry_t entry) | |
2692 | { | |
2693 | __swap_duplicate(entry, SWAP_MAP_SHMEM); | |
2694 | } | |
2695 | ||
355cfa73 | 2696 | /* |
08259d58 HD |
2697 | * Increase reference count of swap entry by 1. |
2698 | * Returns 0 for success, or -ENOMEM if a swap_count_continuation is required | |
2699 | * but could not be atomically allocated. Returns 0, just as if it succeeded, | |
2700 | * if __swap_duplicate() fails for another reason (-EINVAL or -ENOENT), which | |
2701 | * might occur if a page table entry has got corrupted. | |
355cfa73 | 2702 | */ |
570a335b | 2703 | int swap_duplicate(swp_entry_t entry) |
355cfa73 | 2704 | { |
570a335b HD |
2705 | int err = 0; |
2706 | ||
2707 | while (!err && __swap_duplicate(entry, 1) == -ENOMEM) | |
2708 | err = add_swap_count_continuation(entry, GFP_ATOMIC); | |
2709 | return err; | |
355cfa73 | 2710 | } |
1da177e4 | 2711 | |
cb4b86ba | 2712 | /* |
355cfa73 KH |
2713 | * @entry: swap entry for which we allocate swap cache. |
2714 | * | |
73c34b6a | 2715 | * Called when allocating swap cache for existing swap entry, |
355cfa73 KH |
2716 | * This can return error codes. Returns 0 at success. |
2717 | * -EBUSY means there is a swap cache. | |
2718 | * Note: return code is different from swap_duplicate(). | |
cb4b86ba KH |
2719 | */ |
2720 | int swapcache_prepare(swp_entry_t entry) | |
2721 | { | |
253d553b | 2722 | return __swap_duplicate(entry, SWAP_HAS_CACHE); |
cb4b86ba KH |
2723 | } |
2724 | ||
f981c595 MG |
2725 | struct swap_info_struct *page_swap_info(struct page *page) |
2726 | { | |
2727 | swp_entry_t swap = { .val = page_private(page) }; | |
2728 | BUG_ON(!PageSwapCache(page)); | |
2729 | return swap_info[swp_type(swap)]; | |
2730 | } | |
2731 | ||
2732 | /* | |
2733 | * out-of-line __page_file_ methods to avoid include hell. | |
2734 | */ | |
2735 | struct address_space *__page_file_mapping(struct page *page) | |
2736 | { | |
309381fe | 2737 | VM_BUG_ON_PAGE(!PageSwapCache(page), page); |
f981c595 MG |
2738 | return page_swap_info(page)->swap_file->f_mapping; |
2739 | } | |
2740 | EXPORT_SYMBOL_GPL(__page_file_mapping); | |
2741 | ||
2742 | pgoff_t __page_file_index(struct page *page) | |
2743 | { | |
2744 | swp_entry_t swap = { .val = page_private(page) }; | |
309381fe | 2745 | VM_BUG_ON_PAGE(!PageSwapCache(page), page); |
f981c595 MG |
2746 | return swp_offset(swap); |
2747 | } | |
2748 | EXPORT_SYMBOL_GPL(__page_file_index); | |
2749 | ||
570a335b HD |
2750 | /* |
2751 | * add_swap_count_continuation - called when a swap count is duplicated | |
2752 | * beyond SWAP_MAP_MAX, it allocates a new page and links that to the entry's | |
2753 | * page of the original vmalloc'ed swap_map, to hold the continuation count | |
2754 | * (for that entry and for its neighbouring PAGE_SIZE swap entries). Called | |
2755 | * again when count is duplicated beyond SWAP_MAP_MAX * SWAP_CONT_MAX, etc. | |
2756 | * | |
2757 | * These continuation pages are seldom referenced: the common paths all work | |
2758 | * on the original swap_map, only referring to a continuation page when the | |
2759 | * low "digit" of a count is incremented or decremented through SWAP_MAP_MAX. | |
2760 | * | |
2761 | * add_swap_count_continuation(, GFP_ATOMIC) can be called while holding | |
2762 | * page table locks; if it fails, add_swap_count_continuation(, GFP_KERNEL) | |
2763 | * can be called after dropping locks. | |
2764 | */ | |
2765 | int add_swap_count_continuation(swp_entry_t entry, gfp_t gfp_mask) | |
2766 | { | |
2767 | struct swap_info_struct *si; | |
2768 | struct page *head; | |
2769 | struct page *page; | |
2770 | struct page *list_page; | |
2771 | pgoff_t offset; | |
2772 | unsigned char count; | |
2773 | ||
2774 | /* | |
2775 | * When debugging, it's easier to use __GFP_ZERO here; but it's better | |
2776 | * for latency not to zero a page while GFP_ATOMIC and holding locks. | |
2777 | */ | |
2778 | page = alloc_page(gfp_mask | __GFP_HIGHMEM); | |
2779 | ||
2780 | si = swap_info_get(entry); | |
2781 | if (!si) { | |
2782 | /* | |
2783 | * An acceptable race has occurred since the failing | |
2784 | * __swap_duplicate(): the swap entry has been freed, | |
2785 | * perhaps even the whole swap_map cleared for swapoff. | |
2786 | */ | |
2787 | goto outer; | |
2788 | } | |
2789 | ||
2790 | offset = swp_offset(entry); | |
2791 | count = si->swap_map[offset] & ~SWAP_HAS_CACHE; | |
2792 | ||
2793 | if ((count & ~COUNT_CONTINUED) != SWAP_MAP_MAX) { | |
2794 | /* | |
2795 | * The higher the swap count, the more likely it is that tasks | |
2796 | * will race to add swap count continuation: we need to avoid | |
2797 | * over-provisioning. | |
2798 | */ | |
2799 | goto out; | |
2800 | } | |
2801 | ||
2802 | if (!page) { | |
ec8acf20 | 2803 | spin_unlock(&si->lock); |
570a335b HD |
2804 | return -ENOMEM; |
2805 | } | |
2806 | ||
2807 | /* | |
2808 | * We are fortunate that although vmalloc_to_page uses pte_offset_map, | |
2de1a7e4 SJ |
2809 | * no architecture is using highmem pages for kernel page tables: so it |
2810 | * will not corrupt the GFP_ATOMIC caller's atomic page table kmaps. | |
570a335b HD |
2811 | */ |
2812 | head = vmalloc_to_page(si->swap_map + offset); | |
2813 | offset &= ~PAGE_MASK; | |
2814 | ||
2815 | /* | |
2816 | * Page allocation does not initialize the page's lru field, | |
2817 | * but it does always reset its private field. | |
2818 | */ | |
2819 | if (!page_private(head)) { | |
2820 | BUG_ON(count & COUNT_CONTINUED); | |
2821 | INIT_LIST_HEAD(&head->lru); | |
2822 | set_page_private(head, SWP_CONTINUED); | |
2823 | si->flags |= SWP_CONTINUED; | |
2824 | } | |
2825 | ||
2826 | list_for_each_entry(list_page, &head->lru, lru) { | |
2827 | unsigned char *map; | |
2828 | ||
2829 | /* | |
2830 | * If the previous map said no continuation, but we've found | |
2831 | * a continuation page, free our allocation and use this one. | |
2832 | */ | |
2833 | if (!(count & COUNT_CONTINUED)) | |
2834 | goto out; | |
2835 | ||
9b04c5fe | 2836 | map = kmap_atomic(list_page) + offset; |
570a335b | 2837 | count = *map; |
9b04c5fe | 2838 | kunmap_atomic(map); |
570a335b HD |
2839 | |
2840 | /* | |
2841 | * If this continuation count now has some space in it, | |
2842 | * free our allocation and use this one. | |
2843 | */ | |
2844 | if ((count & ~COUNT_CONTINUED) != SWAP_CONT_MAX) | |
2845 | goto out; | |
2846 | } | |
2847 | ||
2848 | list_add_tail(&page->lru, &head->lru); | |
2849 | page = NULL; /* now it's attached, don't free it */ | |
2850 | out: | |
ec8acf20 | 2851 | spin_unlock(&si->lock); |
570a335b HD |
2852 | outer: |
2853 | if (page) | |
2854 | __free_page(page); | |
2855 | return 0; | |
2856 | } | |
2857 | ||
2858 | /* | |
2859 | * swap_count_continued - when the original swap_map count is incremented | |
2860 | * from SWAP_MAP_MAX, check if there is already a continuation page to carry | |
2861 | * into, carry if so, or else fail until a new continuation page is allocated; | |
2862 | * when the original swap_map count is decremented from 0 with continuation, | |
2863 | * borrow from the continuation and report whether it still holds more. | |
2864 | * Called while __swap_duplicate() or swap_entry_free() holds swap_lock. | |
2865 | */ | |
2866 | static bool swap_count_continued(struct swap_info_struct *si, | |
2867 | pgoff_t offset, unsigned char count) | |
2868 | { | |
2869 | struct page *head; | |
2870 | struct page *page; | |
2871 | unsigned char *map; | |
2872 | ||
2873 | head = vmalloc_to_page(si->swap_map + offset); | |
2874 | if (page_private(head) != SWP_CONTINUED) { | |
2875 | BUG_ON(count & COUNT_CONTINUED); | |
2876 | return false; /* need to add count continuation */ | |
2877 | } | |
2878 | ||
2879 | offset &= ~PAGE_MASK; | |
2880 | page = list_entry(head->lru.next, struct page, lru); | |
9b04c5fe | 2881 | map = kmap_atomic(page) + offset; |
570a335b HD |
2882 | |
2883 | if (count == SWAP_MAP_MAX) /* initial increment from swap_map */ | |
2884 | goto init_map; /* jump over SWAP_CONT_MAX checks */ | |
2885 | ||
2886 | if (count == (SWAP_MAP_MAX | COUNT_CONTINUED)) { /* incrementing */ | |
2887 | /* | |
2888 | * Think of how you add 1 to 999 | |
2889 | */ | |
2890 | while (*map == (SWAP_CONT_MAX | COUNT_CONTINUED)) { | |
9b04c5fe | 2891 | kunmap_atomic(map); |
570a335b HD |
2892 | page = list_entry(page->lru.next, struct page, lru); |
2893 | BUG_ON(page == head); | |
9b04c5fe | 2894 | map = kmap_atomic(page) + offset; |
570a335b HD |
2895 | } |
2896 | if (*map == SWAP_CONT_MAX) { | |
9b04c5fe | 2897 | kunmap_atomic(map); |
570a335b HD |
2898 | page = list_entry(page->lru.next, struct page, lru); |
2899 | if (page == head) | |
2900 | return false; /* add count continuation */ | |
9b04c5fe | 2901 | map = kmap_atomic(page) + offset; |
570a335b HD |
2902 | init_map: *map = 0; /* we didn't zero the page */ |
2903 | } | |
2904 | *map += 1; | |
9b04c5fe | 2905 | kunmap_atomic(map); |
570a335b HD |
2906 | page = list_entry(page->lru.prev, struct page, lru); |
2907 | while (page != head) { | |
9b04c5fe | 2908 | map = kmap_atomic(page) + offset; |
570a335b | 2909 | *map = COUNT_CONTINUED; |
9b04c5fe | 2910 | kunmap_atomic(map); |
570a335b HD |
2911 | page = list_entry(page->lru.prev, struct page, lru); |
2912 | } | |
2913 | return true; /* incremented */ | |
2914 | ||
2915 | } else { /* decrementing */ | |
2916 | /* | |
2917 | * Think of how you subtract 1 from 1000 | |
2918 | */ | |
2919 | BUG_ON(count != COUNT_CONTINUED); | |
2920 | while (*map == COUNT_CONTINUED) { | |
9b04c5fe | 2921 | kunmap_atomic(map); |
570a335b HD |
2922 | page = list_entry(page->lru.next, struct page, lru); |
2923 | BUG_ON(page == head); | |
9b04c5fe | 2924 | map = kmap_atomic(page) + offset; |
570a335b HD |
2925 | } |
2926 | BUG_ON(*map == 0); | |
2927 | *map -= 1; | |
2928 | if (*map == 0) | |
2929 | count = 0; | |
9b04c5fe | 2930 | kunmap_atomic(map); |
570a335b HD |
2931 | page = list_entry(page->lru.prev, struct page, lru); |
2932 | while (page != head) { | |
9b04c5fe | 2933 | map = kmap_atomic(page) + offset; |
570a335b HD |
2934 | *map = SWAP_CONT_MAX | count; |
2935 | count = COUNT_CONTINUED; | |
9b04c5fe | 2936 | kunmap_atomic(map); |
570a335b HD |
2937 | page = list_entry(page->lru.prev, struct page, lru); |
2938 | } | |
2939 | return count == COUNT_CONTINUED; | |
2940 | } | |
2941 | } | |
2942 | ||
2943 | /* | |
2944 | * free_swap_count_continuations - swapoff free all the continuation pages | |
2945 | * appended to the swap_map, after swap_map is quiesced, before vfree'ing it. | |
2946 | */ | |
2947 | static void free_swap_count_continuations(struct swap_info_struct *si) | |
2948 | { | |
2949 | pgoff_t offset; | |
2950 | ||
2951 | for (offset = 0; offset < si->max; offset += PAGE_SIZE) { | |
2952 | struct page *head; | |
2953 | head = vmalloc_to_page(si->swap_map + offset); | |
2954 | if (page_private(head)) { | |
2955 | struct list_head *this, *next; | |
2956 | list_for_each_safe(this, next, &head->lru) { | |
2957 | struct page *page; | |
2958 | page = list_entry(this, struct page, lru); | |
2959 | list_del(this); | |
2960 | __free_page(page); | |
2961 | } | |
2962 | } | |
2963 | } | |
2964 | } |