]>
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; |
ec8acf20 | 54 | static atomic_t highest_priority_index = ATOMIC_INIT(-1); |
1da177e4 | 55 | |
1da177e4 LT |
56 | static const char Bad_file[] = "Bad swap file entry "; |
57 | static const char Unused_file[] = "Unused swap file entry "; | |
58 | static const char Bad_offset[] = "Bad swap offset entry "; | |
59 | static const char Unused_offset[] = "Unused swap offset entry "; | |
60 | ||
38b5faf4 | 61 | struct swap_list_t swap_list = {-1, -1}; |
1da177e4 | 62 | |
38b5faf4 | 63 | struct swap_info_struct *swap_info[MAX_SWAPFILES]; |
1da177e4 | 64 | |
fc0abb14 | 65 | static DEFINE_MUTEX(swapon_mutex); |
1da177e4 | 66 | |
66d7dd51 KS |
67 | static DECLARE_WAIT_QUEUE_HEAD(proc_poll_wait); |
68 | /* Activity counter to indicate that a swapon or swapoff has occurred */ | |
69 | static atomic_t proc_poll_event = ATOMIC_INIT(0); | |
70 | ||
8d69aaee | 71 | static inline unsigned char swap_count(unsigned char ent) |
355cfa73 | 72 | { |
570a335b | 73 | return ent & ~SWAP_HAS_CACHE; /* may include SWAP_HAS_CONT flag */ |
355cfa73 KH |
74 | } |
75 | ||
efa90a98 | 76 | /* returns 1 if swap entry is freed */ |
c9e44410 KH |
77 | static int |
78 | __try_to_reclaim_swap(struct swap_info_struct *si, unsigned long offset) | |
79 | { | |
efa90a98 | 80 | swp_entry_t entry = swp_entry(si->type, offset); |
c9e44410 KH |
81 | struct page *page; |
82 | int ret = 0; | |
83 | ||
33806f06 | 84 | page = find_get_page(swap_address_space(entry), entry.val); |
c9e44410 KH |
85 | if (!page) |
86 | return 0; | |
87 | /* | |
88 | * This function is called from scan_swap_map() and it's called | |
89 | * by vmscan.c at reclaiming pages. So, we hold a lock on a page, here. | |
90 | * We have to use trylock for avoiding deadlock. This is a special | |
91 | * case and you should use try_to_free_swap() with explicit lock_page() | |
92 | * in usual operations. | |
93 | */ | |
94 | if (trylock_page(page)) { | |
95 | ret = try_to_free_swap(page); | |
96 | unlock_page(page); | |
97 | } | |
98 | page_cache_release(page); | |
99 | return ret; | |
100 | } | |
355cfa73 | 101 | |
6a6ba831 HD |
102 | /* |
103 | * swapon tell device that all the old swap contents can be discarded, | |
104 | * to allow the swap device to optimize its wear-levelling. | |
105 | */ | |
106 | static int discard_swap(struct swap_info_struct *si) | |
107 | { | |
108 | struct swap_extent *se; | |
9625a5f2 HD |
109 | sector_t start_block; |
110 | sector_t nr_blocks; | |
6a6ba831 HD |
111 | int err = 0; |
112 | ||
9625a5f2 HD |
113 | /* Do not discard the swap header page! */ |
114 | se = &si->first_swap_extent; | |
115 | start_block = (se->start_block + 1) << (PAGE_SHIFT - 9); | |
116 | nr_blocks = ((sector_t)se->nr_pages - 1) << (PAGE_SHIFT - 9); | |
117 | if (nr_blocks) { | |
118 | err = blkdev_issue_discard(si->bdev, start_block, | |
dd3932ed | 119 | nr_blocks, GFP_KERNEL, 0); |
9625a5f2 HD |
120 | if (err) |
121 | return err; | |
122 | cond_resched(); | |
123 | } | |
6a6ba831 | 124 | |
9625a5f2 HD |
125 | list_for_each_entry(se, &si->first_swap_extent.list, list) { |
126 | start_block = se->start_block << (PAGE_SHIFT - 9); | |
127 | nr_blocks = (sector_t)se->nr_pages << (PAGE_SHIFT - 9); | |
6a6ba831 HD |
128 | |
129 | err = blkdev_issue_discard(si->bdev, start_block, | |
dd3932ed | 130 | nr_blocks, GFP_KERNEL, 0); |
6a6ba831 HD |
131 | if (err) |
132 | break; | |
133 | ||
134 | cond_resched(); | |
135 | } | |
136 | return err; /* That will often be -EOPNOTSUPP */ | |
137 | } | |
138 | ||
7992fde7 HD |
139 | /* |
140 | * swap allocation tell device that a cluster of swap can now be discarded, | |
141 | * to allow the swap device to optimize its wear-levelling. | |
142 | */ | |
143 | static void discard_swap_cluster(struct swap_info_struct *si, | |
144 | pgoff_t start_page, pgoff_t nr_pages) | |
145 | { | |
146 | struct swap_extent *se = si->curr_swap_extent; | |
147 | int found_extent = 0; | |
148 | ||
149 | while (nr_pages) { | |
150 | struct list_head *lh; | |
151 | ||
152 | if (se->start_page <= start_page && | |
153 | start_page < se->start_page + se->nr_pages) { | |
154 | pgoff_t offset = start_page - se->start_page; | |
155 | sector_t start_block = se->start_block + offset; | |
858a2990 | 156 | sector_t nr_blocks = se->nr_pages - offset; |
7992fde7 HD |
157 | |
158 | if (nr_blocks > nr_pages) | |
159 | nr_blocks = nr_pages; | |
160 | start_page += nr_blocks; | |
161 | nr_pages -= nr_blocks; | |
162 | ||
163 | if (!found_extent++) | |
164 | si->curr_swap_extent = se; | |
165 | ||
166 | start_block <<= PAGE_SHIFT - 9; | |
167 | nr_blocks <<= PAGE_SHIFT - 9; | |
168 | if (blkdev_issue_discard(si->bdev, start_block, | |
dd3932ed | 169 | nr_blocks, GFP_NOIO, 0)) |
7992fde7 HD |
170 | break; |
171 | } | |
172 | ||
173 | lh = se->list.next; | |
7992fde7 HD |
174 | se = list_entry(lh, struct swap_extent, list); |
175 | } | |
176 | } | |
177 | ||
178 | static int wait_for_discard(void *word) | |
179 | { | |
180 | schedule(); | |
181 | return 0; | |
182 | } | |
183 | ||
048c27fd HD |
184 | #define SWAPFILE_CLUSTER 256 |
185 | #define LATENCY_LIMIT 256 | |
186 | ||
2a8f9449 SL |
187 | static inline void cluster_set_flag(struct swap_cluster_info *info, |
188 | unsigned int flag) | |
189 | { | |
190 | info->flags = flag; | |
191 | } | |
192 | ||
193 | static inline unsigned int cluster_count(struct swap_cluster_info *info) | |
194 | { | |
195 | return info->data; | |
196 | } | |
197 | ||
198 | static inline void cluster_set_count(struct swap_cluster_info *info, | |
199 | unsigned int c) | |
200 | { | |
201 | info->data = c; | |
202 | } | |
203 | ||
204 | static inline void cluster_set_count_flag(struct swap_cluster_info *info, | |
205 | unsigned int c, unsigned int f) | |
206 | { | |
207 | info->flags = f; | |
208 | info->data = c; | |
209 | } | |
210 | ||
211 | static inline unsigned int cluster_next(struct swap_cluster_info *info) | |
212 | { | |
213 | return info->data; | |
214 | } | |
215 | ||
216 | static inline void cluster_set_next(struct swap_cluster_info *info, | |
217 | unsigned int n) | |
218 | { | |
219 | info->data = n; | |
220 | } | |
221 | ||
222 | static inline void cluster_set_next_flag(struct swap_cluster_info *info, | |
223 | unsigned int n, unsigned int f) | |
224 | { | |
225 | info->flags = f; | |
226 | info->data = n; | |
227 | } | |
228 | ||
229 | static inline bool cluster_is_free(struct swap_cluster_info *info) | |
230 | { | |
231 | return info->flags & CLUSTER_FLAG_FREE; | |
232 | } | |
233 | ||
234 | static inline bool cluster_is_null(struct swap_cluster_info *info) | |
235 | { | |
236 | return info->flags & CLUSTER_FLAG_NEXT_NULL; | |
237 | } | |
238 | ||
239 | static inline void cluster_set_null(struct swap_cluster_info *info) | |
240 | { | |
241 | info->flags = CLUSTER_FLAG_NEXT_NULL; | |
242 | info->data = 0; | |
243 | } | |
244 | ||
245 | /* | |
246 | * The cluster corresponding to page_nr will be used. The cluster will be | |
247 | * removed from free cluster list and its usage counter will be increased. | |
248 | */ | |
249 | static void inc_cluster_info_page(struct swap_info_struct *p, | |
250 | struct swap_cluster_info *cluster_info, unsigned long page_nr) | |
251 | { | |
252 | unsigned long idx = page_nr / SWAPFILE_CLUSTER; | |
253 | ||
254 | if (!cluster_info) | |
255 | return; | |
256 | if (cluster_is_free(&cluster_info[idx])) { | |
257 | VM_BUG_ON(cluster_next(&p->free_cluster_head) != idx); | |
258 | cluster_set_next_flag(&p->free_cluster_head, | |
259 | cluster_next(&cluster_info[idx]), 0); | |
260 | if (cluster_next(&p->free_cluster_tail) == idx) { | |
261 | cluster_set_null(&p->free_cluster_tail); | |
262 | cluster_set_null(&p->free_cluster_head); | |
263 | } | |
264 | cluster_set_count_flag(&cluster_info[idx], 0, 0); | |
265 | } | |
266 | ||
267 | VM_BUG_ON(cluster_count(&cluster_info[idx]) >= SWAPFILE_CLUSTER); | |
268 | cluster_set_count(&cluster_info[idx], | |
269 | cluster_count(&cluster_info[idx]) + 1); | |
270 | } | |
271 | ||
272 | /* | |
273 | * The cluster corresponding to page_nr decreases one usage. If the usage | |
274 | * counter becomes 0, which means no page in the cluster is in using, we can | |
275 | * optionally discard the cluster and add it to free cluster list. | |
276 | */ | |
277 | static void dec_cluster_info_page(struct swap_info_struct *p, | |
278 | struct swap_cluster_info *cluster_info, unsigned long page_nr) | |
279 | { | |
280 | unsigned long idx = page_nr / SWAPFILE_CLUSTER; | |
281 | ||
282 | if (!cluster_info) | |
283 | return; | |
284 | ||
285 | VM_BUG_ON(cluster_count(&cluster_info[idx]) == 0); | |
286 | cluster_set_count(&cluster_info[idx], | |
287 | cluster_count(&cluster_info[idx]) - 1); | |
288 | ||
289 | if (cluster_count(&cluster_info[idx]) == 0) { | |
290 | cluster_set_flag(&cluster_info[idx], CLUSTER_FLAG_FREE); | |
291 | if (cluster_is_null(&p->free_cluster_head)) { | |
292 | cluster_set_next_flag(&p->free_cluster_head, idx, 0); | |
293 | cluster_set_next_flag(&p->free_cluster_tail, idx, 0); | |
294 | } else { | |
295 | unsigned int tail = cluster_next(&p->free_cluster_tail); | |
296 | cluster_set_next(&cluster_info[tail], idx); | |
297 | cluster_set_next_flag(&p->free_cluster_tail, idx, 0); | |
298 | } | |
299 | } | |
300 | } | |
301 | ||
302 | /* | |
303 | * It's possible scan_swap_map() uses a free cluster in the middle of free | |
304 | * cluster list. Avoiding such abuse to avoid list corruption. | |
305 | */ | |
306 | static inline bool scan_swap_map_recheck_cluster(struct swap_info_struct *si, | |
307 | unsigned long offset) | |
308 | { | |
309 | offset /= SWAPFILE_CLUSTER; | |
310 | return !cluster_is_null(&si->free_cluster_head) && | |
311 | offset != cluster_next(&si->free_cluster_head) && | |
312 | cluster_is_free(&si->cluster_info[offset]); | |
313 | } | |
314 | ||
24b8ff7c CEB |
315 | static unsigned long scan_swap_map(struct swap_info_struct *si, |
316 | unsigned char usage) | |
1da177e4 | 317 | { |
ebebbbe9 | 318 | unsigned long offset; |
c60aa176 | 319 | unsigned long scan_base; |
7992fde7 | 320 | unsigned long last_in_cluster = 0; |
048c27fd | 321 | int latency_ration = LATENCY_LIMIT; |
7992fde7 | 322 | int found_free_cluster = 0; |
7dfad418 | 323 | |
886bb7e9 | 324 | /* |
7dfad418 HD |
325 | * We try to cluster swap pages by allocating them sequentially |
326 | * in swap. Once we've allocated SWAPFILE_CLUSTER pages this | |
327 | * way, however, we resort to first-free allocation, starting | |
328 | * a new cluster. This prevents us from scattering swap pages | |
329 | * all over the entire swap partition, so that we reduce | |
330 | * overall disk seek times between swap pages. -- sct | |
331 | * But we do now try to find an empty cluster. -Andrea | |
c60aa176 | 332 | * And we let swap pages go all over an SSD partition. Hugh |
7dfad418 HD |
333 | */ |
334 | ||
52b7efdb | 335 | si->flags += SWP_SCANNING; |
c60aa176 | 336 | scan_base = offset = si->cluster_next; |
ebebbbe9 HD |
337 | |
338 | if (unlikely(!si->cluster_nr--)) { | |
339 | if (si->pages - si->inuse_pages < SWAPFILE_CLUSTER) { | |
340 | si->cluster_nr = SWAPFILE_CLUSTER - 1; | |
341 | goto checks; | |
342 | } | |
dcf6b7dd | 343 | if (si->flags & SWP_PAGE_DISCARD) { |
7992fde7 HD |
344 | /* |
345 | * Start range check on racing allocations, in case | |
346 | * they overlap the cluster we eventually decide on | |
347 | * (we scan without swap_lock to allow preemption). | |
348 | * It's hardly conceivable that cluster_nr could be | |
349 | * wrapped during our scan, but don't depend on it. | |
350 | */ | |
351 | if (si->lowest_alloc) | |
352 | goto checks; | |
353 | si->lowest_alloc = si->max; | |
354 | si->highest_alloc = 0; | |
355 | } | |
2a8f9449 SL |
356 | check_cluster: |
357 | if (!cluster_is_null(&si->free_cluster_head)) { | |
358 | offset = cluster_next(&si->free_cluster_head) * | |
359 | SWAPFILE_CLUSTER; | |
360 | last_in_cluster = offset + SWAPFILE_CLUSTER - 1; | |
361 | si->cluster_next = offset; | |
362 | si->cluster_nr = SWAPFILE_CLUSTER - 1; | |
363 | found_free_cluster = 1; | |
364 | goto checks; | |
365 | } else if (si->cluster_info) { | |
366 | /* | |
367 | * Checking free cluster is fast enough, we can do the | |
368 | * check every time | |
369 | */ | |
370 | si->cluster_nr = 0; | |
371 | si->lowest_alloc = 0; | |
372 | goto checks; | |
373 | } | |
374 | ||
ec8acf20 | 375 | spin_unlock(&si->lock); |
7dfad418 | 376 | |
c60aa176 HD |
377 | /* |
378 | * If seek is expensive, start searching for new cluster from | |
379 | * start of partition, to minimize the span of allocated swap. | |
380 | * But if seek is cheap, search from our current position, so | |
381 | * that swap is allocated from all over the partition: if the | |
382 | * Flash Translation Layer only remaps within limited zones, | |
383 | * we don't want to wear out the first zone too quickly. | |
384 | */ | |
385 | if (!(si->flags & SWP_SOLIDSTATE)) | |
386 | scan_base = offset = si->lowest_bit; | |
7dfad418 HD |
387 | last_in_cluster = offset + SWAPFILE_CLUSTER - 1; |
388 | ||
389 | /* Locate the first empty (unaligned) cluster */ | |
390 | for (; last_in_cluster <= si->highest_bit; offset++) { | |
1da177e4 | 391 | if (si->swap_map[offset]) |
7dfad418 HD |
392 | last_in_cluster = offset + SWAPFILE_CLUSTER; |
393 | else if (offset == last_in_cluster) { | |
ec8acf20 | 394 | spin_lock(&si->lock); |
ebebbbe9 HD |
395 | offset -= SWAPFILE_CLUSTER - 1; |
396 | si->cluster_next = offset; | |
397 | si->cluster_nr = SWAPFILE_CLUSTER - 1; | |
7992fde7 | 398 | found_free_cluster = 1; |
ebebbbe9 | 399 | goto checks; |
1da177e4 | 400 | } |
048c27fd HD |
401 | if (unlikely(--latency_ration < 0)) { |
402 | cond_resched(); | |
403 | latency_ration = LATENCY_LIMIT; | |
404 | } | |
7dfad418 | 405 | } |
ebebbbe9 HD |
406 | |
407 | offset = si->lowest_bit; | |
c60aa176 HD |
408 | last_in_cluster = offset + SWAPFILE_CLUSTER - 1; |
409 | ||
410 | /* Locate the first empty (unaligned) cluster */ | |
411 | for (; last_in_cluster < scan_base; offset++) { | |
412 | if (si->swap_map[offset]) | |
413 | last_in_cluster = offset + SWAPFILE_CLUSTER; | |
414 | else if (offset == last_in_cluster) { | |
ec8acf20 | 415 | spin_lock(&si->lock); |
c60aa176 HD |
416 | offset -= SWAPFILE_CLUSTER - 1; |
417 | si->cluster_next = offset; | |
418 | si->cluster_nr = SWAPFILE_CLUSTER - 1; | |
419 | found_free_cluster = 1; | |
420 | goto checks; | |
421 | } | |
422 | if (unlikely(--latency_ration < 0)) { | |
423 | cond_resched(); | |
424 | latency_ration = LATENCY_LIMIT; | |
425 | } | |
426 | } | |
427 | ||
428 | offset = scan_base; | |
ec8acf20 | 429 | spin_lock(&si->lock); |
ebebbbe9 | 430 | si->cluster_nr = SWAPFILE_CLUSTER - 1; |
7992fde7 | 431 | si->lowest_alloc = 0; |
1da177e4 | 432 | } |
7dfad418 | 433 | |
ebebbbe9 | 434 | checks: |
2a8f9449 SL |
435 | if (scan_swap_map_recheck_cluster(si, offset)) |
436 | goto check_cluster; | |
ebebbbe9 | 437 | if (!(si->flags & SWP_WRITEOK)) |
52b7efdb | 438 | goto no_page; |
7dfad418 HD |
439 | if (!si->highest_bit) |
440 | goto no_page; | |
ebebbbe9 | 441 | if (offset > si->highest_bit) |
c60aa176 | 442 | scan_base = offset = si->lowest_bit; |
c9e44410 | 443 | |
b73d7fce HD |
444 | /* reuse swap entry of cache-only swap if not busy. */ |
445 | if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) { | |
c9e44410 | 446 | int swap_was_freed; |
ec8acf20 | 447 | spin_unlock(&si->lock); |
c9e44410 | 448 | swap_was_freed = __try_to_reclaim_swap(si, offset); |
ec8acf20 | 449 | spin_lock(&si->lock); |
c9e44410 KH |
450 | /* entry was freed successfully, try to use this again */ |
451 | if (swap_was_freed) | |
452 | goto checks; | |
453 | goto scan; /* check next one */ | |
454 | } | |
455 | ||
ebebbbe9 HD |
456 | if (si->swap_map[offset]) |
457 | goto scan; | |
458 | ||
459 | if (offset == si->lowest_bit) | |
460 | si->lowest_bit++; | |
461 | if (offset == si->highest_bit) | |
462 | si->highest_bit--; | |
463 | si->inuse_pages++; | |
464 | if (si->inuse_pages == si->pages) { | |
465 | si->lowest_bit = si->max; | |
466 | si->highest_bit = 0; | |
1da177e4 | 467 | } |
253d553b | 468 | si->swap_map[offset] = usage; |
2a8f9449 | 469 | inc_cluster_info_page(si, si->cluster_info, offset); |
ebebbbe9 HD |
470 | si->cluster_next = offset + 1; |
471 | si->flags -= SWP_SCANNING; | |
7992fde7 HD |
472 | |
473 | if (si->lowest_alloc) { | |
474 | /* | |
dcf6b7dd | 475 | * Only set when SWP_PAGE_DISCARD, and there's a scan |
7992fde7 HD |
476 | * for a free cluster in progress or just completed. |
477 | */ | |
478 | if (found_free_cluster) { | |
479 | /* | |
480 | * To optimize wear-levelling, discard the | |
481 | * old data of the cluster, taking care not to | |
482 | * discard any of its pages that have already | |
483 | * been allocated by racing tasks (offset has | |
484 | * already stepped over any at the beginning). | |
485 | */ | |
486 | if (offset < si->highest_alloc && | |
487 | si->lowest_alloc <= last_in_cluster) | |
488 | last_in_cluster = si->lowest_alloc - 1; | |
489 | si->flags |= SWP_DISCARDING; | |
ec8acf20 | 490 | spin_unlock(&si->lock); |
7992fde7 HD |
491 | |
492 | if (offset < last_in_cluster) | |
493 | discard_swap_cluster(si, offset, | |
494 | last_in_cluster - offset + 1); | |
495 | ||
ec8acf20 | 496 | spin_lock(&si->lock); |
7992fde7 HD |
497 | si->lowest_alloc = 0; |
498 | si->flags &= ~SWP_DISCARDING; | |
499 | ||
500 | smp_mb(); /* wake_up_bit advises this */ | |
501 | wake_up_bit(&si->flags, ilog2(SWP_DISCARDING)); | |
502 | ||
503 | } else if (si->flags & SWP_DISCARDING) { | |
504 | /* | |
505 | * Delay using pages allocated by racing tasks | |
506 | * until the whole discard has been issued. We | |
507 | * could defer that delay until swap_writepage, | |
508 | * but it's easier to keep this self-contained. | |
509 | */ | |
ec8acf20 | 510 | spin_unlock(&si->lock); |
7992fde7 HD |
511 | wait_on_bit(&si->flags, ilog2(SWP_DISCARDING), |
512 | wait_for_discard, TASK_UNINTERRUPTIBLE); | |
ec8acf20 | 513 | spin_lock(&si->lock); |
7992fde7 HD |
514 | } else { |
515 | /* | |
516 | * Note pages allocated by racing tasks while | |
517 | * scan for a free cluster is in progress, so | |
518 | * that its final discard can exclude them. | |
519 | */ | |
520 | if (offset < si->lowest_alloc) | |
521 | si->lowest_alloc = offset; | |
522 | if (offset > si->highest_alloc) | |
523 | si->highest_alloc = offset; | |
524 | } | |
525 | } | |
ebebbbe9 | 526 | return offset; |
7dfad418 | 527 | |
ebebbbe9 | 528 | scan: |
ec8acf20 | 529 | spin_unlock(&si->lock); |
7dfad418 | 530 | while (++offset <= si->highest_bit) { |
52b7efdb | 531 | if (!si->swap_map[offset]) { |
ec8acf20 | 532 | spin_lock(&si->lock); |
52b7efdb HD |
533 | goto checks; |
534 | } | |
c9e44410 | 535 | if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) { |
ec8acf20 | 536 | spin_lock(&si->lock); |
c9e44410 KH |
537 | goto checks; |
538 | } | |
048c27fd HD |
539 | if (unlikely(--latency_ration < 0)) { |
540 | cond_resched(); | |
541 | latency_ration = LATENCY_LIMIT; | |
542 | } | |
7dfad418 | 543 | } |
c60aa176 HD |
544 | offset = si->lowest_bit; |
545 | while (++offset < scan_base) { | |
546 | if (!si->swap_map[offset]) { | |
ec8acf20 | 547 | spin_lock(&si->lock); |
c60aa176 HD |
548 | goto checks; |
549 | } | |
c9e44410 | 550 | if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) { |
ec8acf20 | 551 | spin_lock(&si->lock); |
c9e44410 KH |
552 | goto checks; |
553 | } | |
c60aa176 HD |
554 | if (unlikely(--latency_ration < 0)) { |
555 | cond_resched(); | |
556 | latency_ration = LATENCY_LIMIT; | |
557 | } | |
558 | } | |
ec8acf20 | 559 | spin_lock(&si->lock); |
7dfad418 HD |
560 | |
561 | no_page: | |
52b7efdb | 562 | si->flags -= SWP_SCANNING; |
1da177e4 LT |
563 | return 0; |
564 | } | |
565 | ||
566 | swp_entry_t get_swap_page(void) | |
567 | { | |
fb4f88dc HD |
568 | struct swap_info_struct *si; |
569 | pgoff_t offset; | |
570 | int type, next; | |
571 | int wrapped = 0; | |
ec8acf20 | 572 | int hp_index; |
1da177e4 | 573 | |
5d337b91 | 574 | spin_lock(&swap_lock); |
ec8acf20 | 575 | if (atomic_long_read(&nr_swap_pages) <= 0) |
fb4f88dc | 576 | goto noswap; |
ec8acf20 | 577 | atomic_long_dec(&nr_swap_pages); |
fb4f88dc HD |
578 | |
579 | for (type = swap_list.next; type >= 0 && wrapped < 2; type = next) { | |
ec8acf20 SL |
580 | hp_index = atomic_xchg(&highest_priority_index, -1); |
581 | /* | |
582 | * highest_priority_index records current highest priority swap | |
583 | * type which just frees swap entries. If its priority is | |
584 | * higher than that of swap_list.next swap type, we use it. It | |
585 | * isn't protected by swap_lock, so it can be an invalid value | |
586 | * if the corresponding swap type is swapoff. We double check | |
587 | * the flags here. It's even possible the swap type is swapoff | |
588 | * and swapon again and its priority is changed. In such rare | |
589 | * case, low prority swap type might be used, but eventually | |
590 | * high priority swap will be used after several rounds of | |
591 | * swap. | |
592 | */ | |
593 | if (hp_index != -1 && hp_index != type && | |
594 | swap_info[type]->prio < swap_info[hp_index]->prio && | |
595 | (swap_info[hp_index]->flags & SWP_WRITEOK)) { | |
596 | type = hp_index; | |
597 | swap_list.next = type; | |
598 | } | |
599 | ||
efa90a98 | 600 | si = swap_info[type]; |
fb4f88dc HD |
601 | next = si->next; |
602 | if (next < 0 || | |
efa90a98 | 603 | (!wrapped && si->prio != swap_info[next]->prio)) { |
fb4f88dc HD |
604 | next = swap_list.head; |
605 | wrapped++; | |
1da177e4 | 606 | } |
fb4f88dc | 607 | |
ec8acf20 SL |
608 | spin_lock(&si->lock); |
609 | if (!si->highest_bit) { | |
610 | spin_unlock(&si->lock); | |
fb4f88dc | 611 | continue; |
ec8acf20 SL |
612 | } |
613 | if (!(si->flags & SWP_WRITEOK)) { | |
614 | spin_unlock(&si->lock); | |
fb4f88dc | 615 | continue; |
ec8acf20 | 616 | } |
fb4f88dc HD |
617 | |
618 | swap_list.next = next; | |
ec8acf20 SL |
619 | |
620 | spin_unlock(&swap_lock); | |
355cfa73 | 621 | /* This is called for allocating swap entry for cache */ |
253d553b | 622 | offset = scan_swap_map(si, SWAP_HAS_CACHE); |
ec8acf20 SL |
623 | spin_unlock(&si->lock); |
624 | if (offset) | |
fb4f88dc | 625 | return swp_entry(type, offset); |
ec8acf20 | 626 | spin_lock(&swap_lock); |
fb4f88dc | 627 | next = swap_list.next; |
1da177e4 | 628 | } |
fb4f88dc | 629 | |
ec8acf20 | 630 | atomic_long_inc(&nr_swap_pages); |
fb4f88dc | 631 | noswap: |
5d337b91 | 632 | spin_unlock(&swap_lock); |
fb4f88dc | 633 | return (swp_entry_t) {0}; |
1da177e4 LT |
634 | } |
635 | ||
910321ea HD |
636 | /* The only caller of this function is now susupend routine */ |
637 | swp_entry_t get_swap_page_of_type(int type) | |
638 | { | |
639 | struct swap_info_struct *si; | |
640 | pgoff_t offset; | |
641 | ||
910321ea | 642 | si = swap_info[type]; |
ec8acf20 | 643 | spin_lock(&si->lock); |
910321ea | 644 | if (si && (si->flags & SWP_WRITEOK)) { |
ec8acf20 | 645 | atomic_long_dec(&nr_swap_pages); |
910321ea HD |
646 | /* This is called for allocating swap entry, not cache */ |
647 | offset = scan_swap_map(si, 1); | |
648 | if (offset) { | |
ec8acf20 | 649 | spin_unlock(&si->lock); |
910321ea HD |
650 | return swp_entry(type, offset); |
651 | } | |
ec8acf20 | 652 | atomic_long_inc(&nr_swap_pages); |
910321ea | 653 | } |
ec8acf20 | 654 | spin_unlock(&si->lock); |
910321ea HD |
655 | return (swp_entry_t) {0}; |
656 | } | |
657 | ||
73c34b6a | 658 | static struct swap_info_struct *swap_info_get(swp_entry_t entry) |
1da177e4 | 659 | { |
73c34b6a | 660 | struct swap_info_struct *p; |
1da177e4 LT |
661 | unsigned long offset, type; |
662 | ||
663 | if (!entry.val) | |
664 | goto out; | |
665 | type = swp_type(entry); | |
666 | if (type >= nr_swapfiles) | |
667 | goto bad_nofile; | |
efa90a98 | 668 | p = swap_info[type]; |
1da177e4 LT |
669 | if (!(p->flags & SWP_USED)) |
670 | goto bad_device; | |
671 | offset = swp_offset(entry); | |
672 | if (offset >= p->max) | |
673 | goto bad_offset; | |
674 | if (!p->swap_map[offset]) | |
675 | goto bad_free; | |
ec8acf20 | 676 | spin_lock(&p->lock); |
1da177e4 LT |
677 | return p; |
678 | ||
679 | bad_free: | |
465c47fd | 680 | pr_err("swap_free: %s%08lx\n", Unused_offset, entry.val); |
1da177e4 LT |
681 | goto out; |
682 | bad_offset: | |
465c47fd | 683 | pr_err("swap_free: %s%08lx\n", Bad_offset, entry.val); |
1da177e4 LT |
684 | goto out; |
685 | bad_device: | |
465c47fd | 686 | pr_err("swap_free: %s%08lx\n", Unused_file, entry.val); |
1da177e4 LT |
687 | goto out; |
688 | bad_nofile: | |
465c47fd | 689 | pr_err("swap_free: %s%08lx\n", Bad_file, entry.val); |
1da177e4 LT |
690 | out: |
691 | return NULL; | |
886bb7e9 | 692 | } |
1da177e4 | 693 | |
ec8acf20 SL |
694 | /* |
695 | * This swap type frees swap entry, check if it is the highest priority swap | |
696 | * type which just frees swap entry. get_swap_page() uses | |
697 | * highest_priority_index to search highest priority swap type. The | |
698 | * swap_info_struct.lock can't protect us if there are multiple swap types | |
699 | * active, so we use atomic_cmpxchg. | |
700 | */ | |
701 | static void set_highest_priority_index(int type) | |
702 | { | |
703 | int old_hp_index, new_hp_index; | |
704 | ||
705 | do { | |
706 | old_hp_index = atomic_read(&highest_priority_index); | |
707 | if (old_hp_index != -1 && | |
708 | swap_info[old_hp_index]->prio >= swap_info[type]->prio) | |
709 | break; | |
710 | new_hp_index = type; | |
711 | } while (atomic_cmpxchg(&highest_priority_index, | |
712 | old_hp_index, new_hp_index) != old_hp_index); | |
713 | } | |
714 | ||
8d69aaee HD |
715 | static unsigned char swap_entry_free(struct swap_info_struct *p, |
716 | swp_entry_t entry, unsigned char usage) | |
1da177e4 | 717 | { |
253d553b | 718 | unsigned long offset = swp_offset(entry); |
8d69aaee HD |
719 | unsigned char count; |
720 | unsigned char has_cache; | |
355cfa73 | 721 | |
253d553b HD |
722 | count = p->swap_map[offset]; |
723 | has_cache = count & SWAP_HAS_CACHE; | |
724 | count &= ~SWAP_HAS_CACHE; | |
355cfa73 | 725 | |
253d553b | 726 | if (usage == SWAP_HAS_CACHE) { |
355cfa73 | 727 | VM_BUG_ON(!has_cache); |
253d553b | 728 | has_cache = 0; |
aaa46865 HD |
729 | } else if (count == SWAP_MAP_SHMEM) { |
730 | /* | |
731 | * Or we could insist on shmem.c using a special | |
732 | * swap_shmem_free() and free_shmem_swap_and_cache()... | |
733 | */ | |
734 | count = 0; | |
570a335b HD |
735 | } else if ((count & ~COUNT_CONTINUED) <= SWAP_MAP_MAX) { |
736 | if (count == COUNT_CONTINUED) { | |
737 | if (swap_count_continued(p, offset, count)) | |
738 | count = SWAP_MAP_MAX | COUNT_CONTINUED; | |
739 | else | |
740 | count = SWAP_MAP_MAX; | |
741 | } else | |
742 | count--; | |
743 | } | |
253d553b HD |
744 | |
745 | if (!count) | |
746 | mem_cgroup_uncharge_swap(entry); | |
747 | ||
748 | usage = count | has_cache; | |
749 | p->swap_map[offset] = usage; | |
355cfa73 | 750 | |
355cfa73 | 751 | /* free if no reference */ |
253d553b | 752 | if (!usage) { |
2a8f9449 | 753 | dec_cluster_info_page(p, p->cluster_info, offset); |
355cfa73 KH |
754 | if (offset < p->lowest_bit) |
755 | p->lowest_bit = offset; | |
756 | if (offset > p->highest_bit) | |
757 | p->highest_bit = offset; | |
ec8acf20 SL |
758 | set_highest_priority_index(p->type); |
759 | atomic_long_inc(&nr_swap_pages); | |
355cfa73 | 760 | p->inuse_pages--; |
38b5faf4 | 761 | frontswap_invalidate_page(p->type, offset); |
73744923 MG |
762 | if (p->flags & SWP_BLKDEV) { |
763 | struct gendisk *disk = p->bdev->bd_disk; | |
764 | if (disk->fops->swap_slot_free_notify) | |
765 | disk->fops->swap_slot_free_notify(p->bdev, | |
766 | offset); | |
767 | } | |
1da177e4 | 768 | } |
253d553b HD |
769 | |
770 | return usage; | |
1da177e4 LT |
771 | } |
772 | ||
773 | /* | |
774 | * Caller has made sure that the swapdevice corresponding to entry | |
775 | * is still around or has not been recycled. | |
776 | */ | |
777 | void swap_free(swp_entry_t entry) | |
778 | { | |
73c34b6a | 779 | struct swap_info_struct *p; |
1da177e4 LT |
780 | |
781 | p = swap_info_get(entry); | |
782 | if (p) { | |
253d553b | 783 | swap_entry_free(p, entry, 1); |
ec8acf20 | 784 | spin_unlock(&p->lock); |
1da177e4 LT |
785 | } |
786 | } | |
787 | ||
cb4b86ba KH |
788 | /* |
789 | * Called after dropping swapcache to decrease refcnt to swap entries. | |
790 | */ | |
791 | void swapcache_free(swp_entry_t entry, struct page *page) | |
792 | { | |
355cfa73 | 793 | struct swap_info_struct *p; |
8d69aaee | 794 | unsigned char count; |
355cfa73 | 795 | |
355cfa73 KH |
796 | p = swap_info_get(entry); |
797 | if (p) { | |
253d553b HD |
798 | count = swap_entry_free(p, entry, SWAP_HAS_CACHE); |
799 | if (page) | |
800 | mem_cgroup_uncharge_swapcache(page, entry, count != 0); | |
ec8acf20 | 801 | spin_unlock(&p->lock); |
355cfa73 | 802 | } |
cb4b86ba KH |
803 | } |
804 | ||
1da177e4 | 805 | /* |
c475a8ab | 806 | * How many references to page are currently swapped out? |
570a335b HD |
807 | * This does not give an exact answer when swap count is continued, |
808 | * but does include the high COUNT_CONTINUED flag to allow for that. | |
1da177e4 | 809 | */ |
bde05d1c | 810 | int page_swapcount(struct page *page) |
1da177e4 | 811 | { |
c475a8ab HD |
812 | int count = 0; |
813 | struct swap_info_struct *p; | |
1da177e4 LT |
814 | swp_entry_t entry; |
815 | ||
4c21e2f2 | 816 | entry.val = page_private(page); |
1da177e4 LT |
817 | p = swap_info_get(entry); |
818 | if (p) { | |
355cfa73 | 819 | count = swap_count(p->swap_map[swp_offset(entry)]); |
ec8acf20 | 820 | spin_unlock(&p->lock); |
1da177e4 | 821 | } |
c475a8ab | 822 | return count; |
1da177e4 LT |
823 | } |
824 | ||
825 | /* | |
7b1fe597 HD |
826 | * We can write to an anon page without COW if there are no other references |
827 | * to it. And as a side-effect, free up its swap: because the old content | |
828 | * on disk will never be read, and seeking back there to write new content | |
829 | * later would only waste time away from clustering. | |
1da177e4 | 830 | */ |
7b1fe597 | 831 | int reuse_swap_page(struct page *page) |
1da177e4 | 832 | { |
c475a8ab HD |
833 | int count; |
834 | ||
51726b12 | 835 | VM_BUG_ON(!PageLocked(page)); |
5ad64688 HD |
836 | if (unlikely(PageKsm(page))) |
837 | return 0; | |
c475a8ab | 838 | count = page_mapcount(page); |
7b1fe597 | 839 | if (count <= 1 && PageSwapCache(page)) { |
c475a8ab | 840 | count += page_swapcount(page); |
7b1fe597 HD |
841 | if (count == 1 && !PageWriteback(page)) { |
842 | delete_from_swap_cache(page); | |
843 | SetPageDirty(page); | |
844 | } | |
845 | } | |
5ad64688 | 846 | return count <= 1; |
1da177e4 LT |
847 | } |
848 | ||
849 | /* | |
a2c43eed HD |
850 | * If swap is getting full, or if there are no more mappings of this page, |
851 | * then try_to_free_swap is called to free its swap space. | |
1da177e4 | 852 | */ |
a2c43eed | 853 | int try_to_free_swap(struct page *page) |
1da177e4 | 854 | { |
51726b12 | 855 | VM_BUG_ON(!PageLocked(page)); |
1da177e4 LT |
856 | |
857 | if (!PageSwapCache(page)) | |
858 | return 0; | |
859 | if (PageWriteback(page)) | |
860 | return 0; | |
a2c43eed | 861 | if (page_swapcount(page)) |
1da177e4 LT |
862 | return 0; |
863 | ||
b73d7fce HD |
864 | /* |
865 | * Once hibernation has begun to create its image of memory, | |
866 | * there's a danger that one of the calls to try_to_free_swap() | |
867 | * - most probably a call from __try_to_reclaim_swap() while | |
868 | * hibernation is allocating its own swap pages for the image, | |
869 | * but conceivably even a call from memory reclaim - will free | |
870 | * the swap from a page which has already been recorded in the | |
871 | * image as a clean swapcache page, and then reuse its swap for | |
872 | * another page of the image. On waking from hibernation, the | |
873 | * original page might be freed under memory pressure, then | |
874 | * later read back in from swap, now with the wrong data. | |
875 | * | |
f90ac398 MG |
876 | * Hibration suspends storage while it is writing the image |
877 | * to disk so check that here. | |
b73d7fce | 878 | */ |
f90ac398 | 879 | if (pm_suspended_storage()) |
b73d7fce HD |
880 | return 0; |
881 | ||
a2c43eed HD |
882 | delete_from_swap_cache(page); |
883 | SetPageDirty(page); | |
884 | return 1; | |
68a22394 RR |
885 | } |
886 | ||
1da177e4 LT |
887 | /* |
888 | * Free the swap entry like above, but also try to | |
889 | * free the page cache entry if it is the last user. | |
890 | */ | |
2509ef26 | 891 | int free_swap_and_cache(swp_entry_t entry) |
1da177e4 | 892 | { |
2509ef26 | 893 | struct swap_info_struct *p; |
1da177e4 LT |
894 | struct page *page = NULL; |
895 | ||
a7420aa5 | 896 | if (non_swap_entry(entry)) |
2509ef26 | 897 | return 1; |
0697212a | 898 | |
1da177e4 LT |
899 | p = swap_info_get(entry); |
900 | if (p) { | |
253d553b | 901 | if (swap_entry_free(p, entry, 1) == SWAP_HAS_CACHE) { |
33806f06 SL |
902 | page = find_get_page(swap_address_space(entry), |
903 | entry.val); | |
8413ac9d | 904 | if (page && !trylock_page(page)) { |
93fac704 NP |
905 | page_cache_release(page); |
906 | page = NULL; | |
907 | } | |
908 | } | |
ec8acf20 | 909 | spin_unlock(&p->lock); |
1da177e4 LT |
910 | } |
911 | if (page) { | |
a2c43eed HD |
912 | /* |
913 | * Not mapped elsewhere, or swap space full? Free it! | |
914 | * Also recheck PageSwapCache now page is locked (above). | |
915 | */ | |
93fac704 | 916 | if (PageSwapCache(page) && !PageWriteback(page) && |
a2c43eed | 917 | (!page_mapped(page) || vm_swap_full())) { |
1da177e4 LT |
918 | delete_from_swap_cache(page); |
919 | SetPageDirty(page); | |
920 | } | |
921 | unlock_page(page); | |
922 | page_cache_release(page); | |
923 | } | |
2509ef26 | 924 | return p != NULL; |
1da177e4 LT |
925 | } |
926 | ||
b0cb1a19 | 927 | #ifdef CONFIG_HIBERNATION |
f577eb30 | 928 | /* |
915bae9e | 929 | * Find the swap type that corresponds to given device (if any). |
f577eb30 | 930 | * |
915bae9e RW |
931 | * @offset - number of the PAGE_SIZE-sized block of the device, starting |
932 | * from 0, in which the swap header is expected to be located. | |
933 | * | |
934 | * This is needed for the suspend to disk (aka swsusp). | |
f577eb30 | 935 | */ |
7bf23687 | 936 | int swap_type_of(dev_t device, sector_t offset, struct block_device **bdev_p) |
f577eb30 | 937 | { |
915bae9e | 938 | struct block_device *bdev = NULL; |
efa90a98 | 939 | int type; |
f577eb30 | 940 | |
915bae9e RW |
941 | if (device) |
942 | bdev = bdget(device); | |
943 | ||
f577eb30 | 944 | spin_lock(&swap_lock); |
efa90a98 HD |
945 | for (type = 0; type < nr_swapfiles; type++) { |
946 | struct swap_info_struct *sis = swap_info[type]; | |
f577eb30 | 947 | |
915bae9e | 948 | if (!(sis->flags & SWP_WRITEOK)) |
f577eb30 | 949 | continue; |
b6b5bce3 | 950 | |
915bae9e | 951 | if (!bdev) { |
7bf23687 | 952 | if (bdev_p) |
dddac6a7 | 953 | *bdev_p = bdgrab(sis->bdev); |
7bf23687 | 954 | |
6e1819d6 | 955 | spin_unlock(&swap_lock); |
efa90a98 | 956 | return type; |
6e1819d6 | 957 | } |
915bae9e | 958 | if (bdev == sis->bdev) { |
9625a5f2 | 959 | struct swap_extent *se = &sis->first_swap_extent; |
915bae9e | 960 | |
915bae9e | 961 | if (se->start_block == offset) { |
7bf23687 | 962 | if (bdev_p) |
dddac6a7 | 963 | *bdev_p = bdgrab(sis->bdev); |
7bf23687 | 964 | |
915bae9e RW |
965 | spin_unlock(&swap_lock); |
966 | bdput(bdev); | |
efa90a98 | 967 | return type; |
915bae9e | 968 | } |
f577eb30 RW |
969 | } |
970 | } | |
971 | spin_unlock(&swap_lock); | |
915bae9e RW |
972 | if (bdev) |
973 | bdput(bdev); | |
974 | ||
f577eb30 RW |
975 | return -ENODEV; |
976 | } | |
977 | ||
73c34b6a HD |
978 | /* |
979 | * Get the (PAGE_SIZE) block corresponding to given offset on the swapdev | |
980 | * corresponding to given index in swap_info (swap type). | |
981 | */ | |
982 | sector_t swapdev_block(int type, pgoff_t offset) | |
983 | { | |
984 | struct block_device *bdev; | |
985 | ||
986 | if ((unsigned int)type >= nr_swapfiles) | |
987 | return 0; | |
988 | if (!(swap_info[type]->flags & SWP_WRITEOK)) | |
989 | return 0; | |
d4906e1a | 990 | return map_swap_entry(swp_entry(type, offset), &bdev); |
73c34b6a HD |
991 | } |
992 | ||
f577eb30 RW |
993 | /* |
994 | * Return either the total number of swap pages of given type, or the number | |
995 | * of free pages of that type (depending on @free) | |
996 | * | |
997 | * This is needed for software suspend | |
998 | */ | |
999 | unsigned int count_swap_pages(int type, int free) | |
1000 | { | |
1001 | unsigned int n = 0; | |
1002 | ||
efa90a98 HD |
1003 | spin_lock(&swap_lock); |
1004 | if ((unsigned int)type < nr_swapfiles) { | |
1005 | struct swap_info_struct *sis = swap_info[type]; | |
1006 | ||
ec8acf20 | 1007 | spin_lock(&sis->lock); |
efa90a98 HD |
1008 | if (sis->flags & SWP_WRITEOK) { |
1009 | n = sis->pages; | |
f577eb30 | 1010 | if (free) |
efa90a98 | 1011 | n -= sis->inuse_pages; |
f577eb30 | 1012 | } |
ec8acf20 | 1013 | spin_unlock(&sis->lock); |
f577eb30 | 1014 | } |
efa90a98 | 1015 | spin_unlock(&swap_lock); |
f577eb30 RW |
1016 | return n; |
1017 | } | |
73c34b6a | 1018 | #endif /* CONFIG_HIBERNATION */ |
f577eb30 | 1019 | |
179ef71c CG |
1020 | static inline int maybe_same_pte(pte_t pte, pte_t swp_pte) |
1021 | { | |
1022 | #ifdef CONFIG_MEM_SOFT_DIRTY | |
1023 | /* | |
1024 | * When pte keeps soft dirty bit the pte generated | |
1025 | * from swap entry does not has it, still it's same | |
1026 | * pte from logical point of view. | |
1027 | */ | |
1028 | pte_t swp_pte_dirty = pte_swp_mksoft_dirty(swp_pte); | |
1029 | return pte_same(pte, swp_pte) || pte_same(pte, swp_pte_dirty); | |
1030 | #else | |
1031 | return pte_same(pte, swp_pte); | |
1032 | #endif | |
1033 | } | |
1034 | ||
1da177e4 | 1035 | /* |
72866f6f HD |
1036 | * No need to decide whether this PTE shares the swap entry with others, |
1037 | * just let do_wp_page work it out if a write is requested later - to | |
1038 | * force COW, vm_page_prot omits write permission from any private vma. | |
1da177e4 | 1039 | */ |
044d66c1 | 1040 | static int unuse_pte(struct vm_area_struct *vma, pmd_t *pmd, |
1da177e4 LT |
1041 | unsigned long addr, swp_entry_t entry, struct page *page) |
1042 | { | |
9e16b7fb | 1043 | struct page *swapcache; |
72835c86 | 1044 | struct mem_cgroup *memcg; |
044d66c1 HD |
1045 | spinlock_t *ptl; |
1046 | pte_t *pte; | |
1047 | int ret = 1; | |
1048 | ||
9e16b7fb HD |
1049 | swapcache = page; |
1050 | page = ksm_might_need_to_copy(page, vma, addr); | |
1051 | if (unlikely(!page)) | |
1052 | return -ENOMEM; | |
1053 | ||
72835c86 JW |
1054 | if (mem_cgroup_try_charge_swapin(vma->vm_mm, page, |
1055 | GFP_KERNEL, &memcg)) { | |
044d66c1 | 1056 | ret = -ENOMEM; |
85d9fc89 KH |
1057 | goto out_nolock; |
1058 | } | |
044d66c1 HD |
1059 | |
1060 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); | |
179ef71c | 1061 | if (unlikely(!maybe_same_pte(*pte, swp_entry_to_pte(entry)))) { |
5d84c776 | 1062 | mem_cgroup_cancel_charge_swapin(memcg); |
044d66c1 HD |
1063 | ret = 0; |
1064 | goto out; | |
1065 | } | |
8a9f3ccd | 1066 | |
b084d435 | 1067 | dec_mm_counter(vma->vm_mm, MM_SWAPENTS); |
d559db08 | 1068 | inc_mm_counter(vma->vm_mm, MM_ANONPAGES); |
1da177e4 LT |
1069 | get_page(page); |
1070 | set_pte_at(vma->vm_mm, addr, pte, | |
1071 | pte_mkold(mk_pte(page, vma->vm_page_prot))); | |
9e16b7fb HD |
1072 | if (page == swapcache) |
1073 | page_add_anon_rmap(page, vma, addr); | |
1074 | else /* ksm created a completely new copy */ | |
1075 | page_add_new_anon_rmap(page, vma, addr); | |
72835c86 | 1076 | mem_cgroup_commit_charge_swapin(page, memcg); |
1da177e4 LT |
1077 | swap_free(entry); |
1078 | /* | |
1079 | * Move the page to the active list so it is not | |
1080 | * immediately swapped out again after swapon. | |
1081 | */ | |
1082 | activate_page(page); | |
044d66c1 HD |
1083 | out: |
1084 | pte_unmap_unlock(pte, ptl); | |
85d9fc89 | 1085 | out_nolock: |
9e16b7fb HD |
1086 | if (page != swapcache) { |
1087 | unlock_page(page); | |
1088 | put_page(page); | |
1089 | } | |
044d66c1 | 1090 | return ret; |
1da177e4 LT |
1091 | } |
1092 | ||
1093 | static int unuse_pte_range(struct vm_area_struct *vma, pmd_t *pmd, | |
1094 | unsigned long addr, unsigned long end, | |
1095 | swp_entry_t entry, struct page *page) | |
1096 | { | |
1da177e4 | 1097 | pte_t swp_pte = swp_entry_to_pte(entry); |
705e87c0 | 1098 | pte_t *pte; |
8a9f3ccd | 1099 | int ret = 0; |
1da177e4 | 1100 | |
044d66c1 HD |
1101 | /* |
1102 | * We don't actually need pte lock while scanning for swp_pte: since | |
1103 | * we hold page lock and mmap_sem, swp_pte cannot be inserted into the | |
1104 | * page table while we're scanning; though it could get zapped, and on | |
1105 | * some architectures (e.g. x86_32 with PAE) we might catch a glimpse | |
1106 | * of unmatched parts which look like swp_pte, so unuse_pte must | |
1107 | * recheck under pte lock. Scanning without pte lock lets it be | |
1108 | * preemptible whenever CONFIG_PREEMPT but not CONFIG_HIGHPTE. | |
1109 | */ | |
1110 | pte = pte_offset_map(pmd, addr); | |
1da177e4 LT |
1111 | do { |
1112 | /* | |
1113 | * swapoff spends a _lot_ of time in this loop! | |
1114 | * Test inline before going to call unuse_pte. | |
1115 | */ | |
179ef71c | 1116 | if (unlikely(maybe_same_pte(*pte, swp_pte))) { |
044d66c1 HD |
1117 | pte_unmap(pte); |
1118 | ret = unuse_pte(vma, pmd, addr, entry, page); | |
1119 | if (ret) | |
1120 | goto out; | |
1121 | pte = pte_offset_map(pmd, addr); | |
1da177e4 LT |
1122 | } |
1123 | } while (pte++, addr += PAGE_SIZE, addr != end); | |
044d66c1 HD |
1124 | pte_unmap(pte - 1); |
1125 | out: | |
8a9f3ccd | 1126 | return ret; |
1da177e4 LT |
1127 | } |
1128 | ||
1129 | static inline int unuse_pmd_range(struct vm_area_struct *vma, pud_t *pud, | |
1130 | unsigned long addr, unsigned long end, | |
1131 | swp_entry_t entry, struct page *page) | |
1132 | { | |
1133 | pmd_t *pmd; | |
1134 | unsigned long next; | |
8a9f3ccd | 1135 | int ret; |
1da177e4 LT |
1136 | |
1137 | pmd = pmd_offset(pud, addr); | |
1138 | do { | |
1139 | next = pmd_addr_end(addr, end); | |
1a5a9906 | 1140 | if (pmd_none_or_trans_huge_or_clear_bad(pmd)) |
1da177e4 | 1141 | continue; |
8a9f3ccd BS |
1142 | ret = unuse_pte_range(vma, pmd, addr, next, entry, page); |
1143 | if (ret) | |
1144 | return ret; | |
1da177e4 LT |
1145 | } while (pmd++, addr = next, addr != end); |
1146 | return 0; | |
1147 | } | |
1148 | ||
1149 | static inline int unuse_pud_range(struct vm_area_struct *vma, pgd_t *pgd, | |
1150 | unsigned long addr, unsigned long end, | |
1151 | swp_entry_t entry, struct page *page) | |
1152 | { | |
1153 | pud_t *pud; | |
1154 | unsigned long next; | |
8a9f3ccd | 1155 | int ret; |
1da177e4 LT |
1156 | |
1157 | pud = pud_offset(pgd, addr); | |
1158 | do { | |
1159 | next = pud_addr_end(addr, end); | |
1160 | if (pud_none_or_clear_bad(pud)) | |
1161 | continue; | |
8a9f3ccd BS |
1162 | ret = unuse_pmd_range(vma, pud, addr, next, entry, page); |
1163 | if (ret) | |
1164 | return ret; | |
1da177e4 LT |
1165 | } while (pud++, addr = next, addr != end); |
1166 | return 0; | |
1167 | } | |
1168 | ||
1169 | static int unuse_vma(struct vm_area_struct *vma, | |
1170 | swp_entry_t entry, struct page *page) | |
1171 | { | |
1172 | pgd_t *pgd; | |
1173 | unsigned long addr, end, next; | |
8a9f3ccd | 1174 | int ret; |
1da177e4 | 1175 | |
3ca7b3c5 | 1176 | if (page_anon_vma(page)) { |
1da177e4 LT |
1177 | addr = page_address_in_vma(page, vma); |
1178 | if (addr == -EFAULT) | |
1179 | return 0; | |
1180 | else | |
1181 | end = addr + PAGE_SIZE; | |
1182 | } else { | |
1183 | addr = vma->vm_start; | |
1184 | end = vma->vm_end; | |
1185 | } | |
1186 | ||
1187 | pgd = pgd_offset(vma->vm_mm, addr); | |
1188 | do { | |
1189 | next = pgd_addr_end(addr, end); | |
1190 | if (pgd_none_or_clear_bad(pgd)) | |
1191 | continue; | |
8a9f3ccd BS |
1192 | ret = unuse_pud_range(vma, pgd, addr, next, entry, page); |
1193 | if (ret) | |
1194 | return ret; | |
1da177e4 LT |
1195 | } while (pgd++, addr = next, addr != end); |
1196 | return 0; | |
1197 | } | |
1198 | ||
1199 | static int unuse_mm(struct mm_struct *mm, | |
1200 | swp_entry_t entry, struct page *page) | |
1201 | { | |
1202 | struct vm_area_struct *vma; | |
8a9f3ccd | 1203 | int ret = 0; |
1da177e4 LT |
1204 | |
1205 | if (!down_read_trylock(&mm->mmap_sem)) { | |
1206 | /* | |
7d03431c FLVC |
1207 | * Activate page so shrink_inactive_list is unlikely to unmap |
1208 | * its ptes while lock is dropped, so swapoff can make progress. | |
1da177e4 | 1209 | */ |
c475a8ab | 1210 | activate_page(page); |
1da177e4 LT |
1211 | unlock_page(page); |
1212 | down_read(&mm->mmap_sem); | |
1213 | lock_page(page); | |
1214 | } | |
1da177e4 | 1215 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
8a9f3ccd | 1216 | if (vma->anon_vma && (ret = unuse_vma(vma, entry, page))) |
1da177e4 LT |
1217 | break; |
1218 | } | |
1da177e4 | 1219 | up_read(&mm->mmap_sem); |
8a9f3ccd | 1220 | return (ret < 0)? ret: 0; |
1da177e4 LT |
1221 | } |
1222 | ||
1223 | /* | |
38b5faf4 DM |
1224 | * Scan swap_map (or frontswap_map if frontswap parameter is true) |
1225 | * from current position to next entry still in use. | |
1da177e4 LT |
1226 | * Recycle to start on reaching the end, returning 0 when empty. |
1227 | */ | |
6eb396dc | 1228 | static unsigned int find_next_to_unuse(struct swap_info_struct *si, |
38b5faf4 | 1229 | unsigned int prev, bool frontswap) |
1da177e4 | 1230 | { |
6eb396dc HD |
1231 | unsigned int max = si->max; |
1232 | unsigned int i = prev; | |
8d69aaee | 1233 | unsigned char count; |
1da177e4 LT |
1234 | |
1235 | /* | |
5d337b91 | 1236 | * No need for swap_lock here: we're just looking |
1da177e4 LT |
1237 | * for whether an entry is in use, not modifying it; false |
1238 | * hits are okay, and sys_swapoff() has already prevented new | |
5d337b91 | 1239 | * allocations from this area (while holding swap_lock). |
1da177e4 LT |
1240 | */ |
1241 | for (;;) { | |
1242 | if (++i >= max) { | |
1243 | if (!prev) { | |
1244 | i = 0; | |
1245 | break; | |
1246 | } | |
1247 | /* | |
1248 | * No entries in use at top of swap_map, | |
1249 | * loop back to start and recheck there. | |
1250 | */ | |
1251 | max = prev + 1; | |
1252 | prev = 0; | |
1253 | i = 1; | |
1254 | } | |
38b5faf4 DM |
1255 | if (frontswap) { |
1256 | if (frontswap_test(si, i)) | |
1257 | break; | |
1258 | else | |
1259 | continue; | |
1260 | } | |
1da177e4 | 1261 | count = si->swap_map[i]; |
355cfa73 | 1262 | if (count && swap_count(count) != SWAP_MAP_BAD) |
1da177e4 LT |
1263 | break; |
1264 | } | |
1265 | return i; | |
1266 | } | |
1267 | ||
1268 | /* | |
1269 | * We completely avoid races by reading each swap page in advance, | |
1270 | * and then search for the process using it. All the necessary | |
1271 | * page table adjustments can then be made atomically. | |
38b5faf4 DM |
1272 | * |
1273 | * if the boolean frontswap is true, only unuse pages_to_unuse pages; | |
1274 | * pages_to_unuse==0 means all pages; ignored if frontswap is false | |
1da177e4 | 1275 | */ |
38b5faf4 DM |
1276 | int try_to_unuse(unsigned int type, bool frontswap, |
1277 | unsigned long pages_to_unuse) | |
1da177e4 | 1278 | { |
efa90a98 | 1279 | struct swap_info_struct *si = swap_info[type]; |
1da177e4 | 1280 | struct mm_struct *start_mm; |
8d69aaee HD |
1281 | unsigned char *swap_map; |
1282 | unsigned char swcount; | |
1da177e4 LT |
1283 | struct page *page; |
1284 | swp_entry_t entry; | |
6eb396dc | 1285 | unsigned int i = 0; |
1da177e4 | 1286 | int retval = 0; |
1da177e4 LT |
1287 | |
1288 | /* | |
1289 | * When searching mms for an entry, a good strategy is to | |
1290 | * start at the first mm we freed the previous entry from | |
1291 | * (though actually we don't notice whether we or coincidence | |
1292 | * freed the entry). Initialize this start_mm with a hold. | |
1293 | * | |
1294 | * A simpler strategy would be to start at the last mm we | |
1295 | * freed the previous entry from; but that would take less | |
1296 | * advantage of mmlist ordering, which clusters forked mms | |
1297 | * together, child after parent. If we race with dup_mmap(), we | |
1298 | * prefer to resolve parent before child, lest we miss entries | |
1299 | * duplicated after we scanned child: using last mm would invert | |
570a335b | 1300 | * that. |
1da177e4 LT |
1301 | */ |
1302 | start_mm = &init_mm; | |
1303 | atomic_inc(&init_mm.mm_users); | |
1304 | ||
1305 | /* | |
1306 | * Keep on scanning until all entries have gone. Usually, | |
1307 | * one pass through swap_map is enough, but not necessarily: | |
1308 | * there are races when an instance of an entry might be missed. | |
1309 | */ | |
38b5faf4 | 1310 | while ((i = find_next_to_unuse(si, i, frontswap)) != 0) { |
1da177e4 LT |
1311 | if (signal_pending(current)) { |
1312 | retval = -EINTR; | |
1313 | break; | |
1314 | } | |
1315 | ||
886bb7e9 | 1316 | /* |
1da177e4 LT |
1317 | * Get a page for the entry, using the existing swap |
1318 | * cache page if there is one. Otherwise, get a clean | |
886bb7e9 | 1319 | * page and read the swap into it. |
1da177e4 LT |
1320 | */ |
1321 | swap_map = &si->swap_map[i]; | |
1322 | entry = swp_entry(type, i); | |
02098fea HD |
1323 | page = read_swap_cache_async(entry, |
1324 | GFP_HIGHUSER_MOVABLE, NULL, 0); | |
1da177e4 LT |
1325 | if (!page) { |
1326 | /* | |
1327 | * Either swap_duplicate() failed because entry | |
1328 | * has been freed independently, and will not be | |
1329 | * reused since sys_swapoff() already disabled | |
1330 | * allocation from here, or alloc_page() failed. | |
1331 | */ | |
1332 | if (!*swap_map) | |
1333 | continue; | |
1334 | retval = -ENOMEM; | |
1335 | break; | |
1336 | } | |
1337 | ||
1338 | /* | |
1339 | * Don't hold on to start_mm if it looks like exiting. | |
1340 | */ | |
1341 | if (atomic_read(&start_mm->mm_users) == 1) { | |
1342 | mmput(start_mm); | |
1343 | start_mm = &init_mm; | |
1344 | atomic_inc(&init_mm.mm_users); | |
1345 | } | |
1346 | ||
1347 | /* | |
1348 | * Wait for and lock page. When do_swap_page races with | |
1349 | * try_to_unuse, do_swap_page can handle the fault much | |
1350 | * faster than try_to_unuse can locate the entry. This | |
1351 | * apparently redundant "wait_on_page_locked" lets try_to_unuse | |
1352 | * defer to do_swap_page in such a case - in some tests, | |
1353 | * do_swap_page and try_to_unuse repeatedly compete. | |
1354 | */ | |
1355 | wait_on_page_locked(page); | |
1356 | wait_on_page_writeback(page); | |
1357 | lock_page(page); | |
1358 | wait_on_page_writeback(page); | |
1359 | ||
1360 | /* | |
1361 | * Remove all references to entry. | |
1da177e4 | 1362 | */ |
1da177e4 | 1363 | swcount = *swap_map; |
aaa46865 HD |
1364 | if (swap_count(swcount) == SWAP_MAP_SHMEM) { |
1365 | retval = shmem_unuse(entry, page); | |
1366 | /* page has already been unlocked and released */ | |
1367 | if (retval < 0) | |
1368 | break; | |
1369 | continue; | |
1da177e4 | 1370 | } |
aaa46865 HD |
1371 | if (swap_count(swcount) && start_mm != &init_mm) |
1372 | retval = unuse_mm(start_mm, entry, page); | |
1373 | ||
355cfa73 | 1374 | if (swap_count(*swap_map)) { |
1da177e4 LT |
1375 | int set_start_mm = (*swap_map >= swcount); |
1376 | struct list_head *p = &start_mm->mmlist; | |
1377 | struct mm_struct *new_start_mm = start_mm; | |
1378 | struct mm_struct *prev_mm = start_mm; | |
1379 | struct mm_struct *mm; | |
1380 | ||
1381 | atomic_inc(&new_start_mm->mm_users); | |
1382 | atomic_inc(&prev_mm->mm_users); | |
1383 | spin_lock(&mmlist_lock); | |
aaa46865 | 1384 | while (swap_count(*swap_map) && !retval && |
1da177e4 LT |
1385 | (p = p->next) != &start_mm->mmlist) { |
1386 | mm = list_entry(p, struct mm_struct, mmlist); | |
70af7c5c | 1387 | if (!atomic_inc_not_zero(&mm->mm_users)) |
1da177e4 | 1388 | continue; |
1da177e4 LT |
1389 | spin_unlock(&mmlist_lock); |
1390 | mmput(prev_mm); | |
1391 | prev_mm = mm; | |
1392 | ||
1393 | cond_resched(); | |
1394 | ||
1395 | swcount = *swap_map; | |
355cfa73 | 1396 | if (!swap_count(swcount)) /* any usage ? */ |
1da177e4 | 1397 | ; |
aaa46865 | 1398 | else if (mm == &init_mm) |
1da177e4 | 1399 | set_start_mm = 1; |
aaa46865 | 1400 | else |
1da177e4 | 1401 | retval = unuse_mm(mm, entry, page); |
355cfa73 | 1402 | |
32c5fc10 | 1403 | if (set_start_mm && *swap_map < swcount) { |
1da177e4 LT |
1404 | mmput(new_start_mm); |
1405 | atomic_inc(&mm->mm_users); | |
1406 | new_start_mm = mm; | |
1407 | set_start_mm = 0; | |
1408 | } | |
1409 | spin_lock(&mmlist_lock); | |
1410 | } | |
1411 | spin_unlock(&mmlist_lock); | |
1412 | mmput(prev_mm); | |
1413 | mmput(start_mm); | |
1414 | start_mm = new_start_mm; | |
1415 | } | |
1416 | if (retval) { | |
1417 | unlock_page(page); | |
1418 | page_cache_release(page); | |
1419 | break; | |
1420 | } | |
1421 | ||
1da177e4 LT |
1422 | /* |
1423 | * If a reference remains (rare), we would like to leave | |
1424 | * the page in the swap cache; but try_to_unmap could | |
1425 | * then re-duplicate the entry once we drop page lock, | |
1426 | * so we might loop indefinitely; also, that page could | |
1427 | * not be swapped out to other storage meanwhile. So: | |
1428 | * delete from cache even if there's another reference, | |
1429 | * after ensuring that the data has been saved to disk - | |
1430 | * since if the reference remains (rarer), it will be | |
1431 | * read from disk into another page. Splitting into two | |
1432 | * pages would be incorrect if swap supported "shared | |
1433 | * private" pages, but they are handled by tmpfs files. | |
5ad64688 HD |
1434 | * |
1435 | * Given how unuse_vma() targets one particular offset | |
1436 | * in an anon_vma, once the anon_vma has been determined, | |
1437 | * this splitting happens to be just what is needed to | |
1438 | * handle where KSM pages have been swapped out: re-reading | |
1439 | * is unnecessarily slow, but we can fix that later on. | |
1da177e4 | 1440 | */ |
355cfa73 KH |
1441 | if (swap_count(*swap_map) && |
1442 | PageDirty(page) && PageSwapCache(page)) { | |
1da177e4 LT |
1443 | struct writeback_control wbc = { |
1444 | .sync_mode = WB_SYNC_NONE, | |
1445 | }; | |
1446 | ||
1447 | swap_writepage(page, &wbc); | |
1448 | lock_page(page); | |
1449 | wait_on_page_writeback(page); | |
1450 | } | |
68bdc8d6 HD |
1451 | |
1452 | /* | |
1453 | * It is conceivable that a racing task removed this page from | |
1454 | * swap cache just before we acquired the page lock at the top, | |
1455 | * or while we dropped it in unuse_mm(). The page might even | |
1456 | * be back in swap cache on another swap area: that we must not | |
1457 | * delete, since it may not have been written out to swap yet. | |
1458 | */ | |
1459 | if (PageSwapCache(page) && | |
1460 | likely(page_private(page) == entry.val)) | |
2e0e26c7 | 1461 | delete_from_swap_cache(page); |
1da177e4 LT |
1462 | |
1463 | /* | |
1464 | * So we could skip searching mms once swap count went | |
1465 | * to 1, we did not mark any present ptes as dirty: must | |
2706a1b8 | 1466 | * mark page dirty so shrink_page_list will preserve it. |
1da177e4 LT |
1467 | */ |
1468 | SetPageDirty(page); | |
1469 | unlock_page(page); | |
1470 | page_cache_release(page); | |
1471 | ||
1472 | /* | |
1473 | * Make sure that we aren't completely killing | |
1474 | * interactive performance. | |
1475 | */ | |
1476 | cond_resched(); | |
38b5faf4 DM |
1477 | if (frontswap && pages_to_unuse > 0) { |
1478 | if (!--pages_to_unuse) | |
1479 | break; | |
1480 | } | |
1da177e4 LT |
1481 | } |
1482 | ||
1483 | mmput(start_mm); | |
1da177e4 LT |
1484 | return retval; |
1485 | } | |
1486 | ||
1487 | /* | |
5d337b91 HD |
1488 | * After a successful try_to_unuse, if no swap is now in use, we know |
1489 | * we can empty the mmlist. swap_lock must be held on entry and exit. | |
1490 | * Note that mmlist_lock nests inside swap_lock, and an mm must be | |
1da177e4 LT |
1491 | * added to the mmlist just after page_duplicate - before would be racy. |
1492 | */ | |
1493 | static void drain_mmlist(void) | |
1494 | { | |
1495 | struct list_head *p, *next; | |
efa90a98 | 1496 | unsigned int type; |
1da177e4 | 1497 | |
efa90a98 HD |
1498 | for (type = 0; type < nr_swapfiles; type++) |
1499 | if (swap_info[type]->inuse_pages) | |
1da177e4 LT |
1500 | return; |
1501 | spin_lock(&mmlist_lock); | |
1502 | list_for_each_safe(p, next, &init_mm.mmlist) | |
1503 | list_del_init(p); | |
1504 | spin_unlock(&mmlist_lock); | |
1505 | } | |
1506 | ||
1507 | /* | |
1508 | * Use this swapdev's extent info to locate the (PAGE_SIZE) block which | |
d4906e1a LS |
1509 | * corresponds to page offset for the specified swap entry. |
1510 | * Note that the type of this function is sector_t, but it returns page offset | |
1511 | * into the bdev, not sector offset. | |
1da177e4 | 1512 | */ |
d4906e1a | 1513 | static sector_t map_swap_entry(swp_entry_t entry, struct block_device **bdev) |
1da177e4 | 1514 | { |
f29ad6a9 HD |
1515 | struct swap_info_struct *sis; |
1516 | struct swap_extent *start_se; | |
1517 | struct swap_extent *se; | |
1518 | pgoff_t offset; | |
1519 | ||
efa90a98 | 1520 | sis = swap_info[swp_type(entry)]; |
f29ad6a9 HD |
1521 | *bdev = sis->bdev; |
1522 | ||
1523 | offset = swp_offset(entry); | |
1524 | start_se = sis->curr_swap_extent; | |
1525 | se = start_se; | |
1da177e4 LT |
1526 | |
1527 | for ( ; ; ) { | |
1528 | struct list_head *lh; | |
1529 | ||
1530 | if (se->start_page <= offset && | |
1531 | offset < (se->start_page + se->nr_pages)) { | |
1532 | return se->start_block + (offset - se->start_page); | |
1533 | } | |
11d31886 | 1534 | lh = se->list.next; |
1da177e4 LT |
1535 | se = list_entry(lh, struct swap_extent, list); |
1536 | sis->curr_swap_extent = se; | |
1537 | BUG_ON(se == start_se); /* It *must* be present */ | |
1538 | } | |
1539 | } | |
1540 | ||
d4906e1a LS |
1541 | /* |
1542 | * Returns the page offset into bdev for the specified page's swap entry. | |
1543 | */ | |
1544 | sector_t map_swap_page(struct page *page, struct block_device **bdev) | |
1545 | { | |
1546 | swp_entry_t entry; | |
1547 | entry.val = page_private(page); | |
1548 | return map_swap_entry(entry, bdev); | |
1549 | } | |
1550 | ||
1da177e4 LT |
1551 | /* |
1552 | * Free all of a swapdev's extent information | |
1553 | */ | |
1554 | static void destroy_swap_extents(struct swap_info_struct *sis) | |
1555 | { | |
9625a5f2 | 1556 | while (!list_empty(&sis->first_swap_extent.list)) { |
1da177e4 LT |
1557 | struct swap_extent *se; |
1558 | ||
9625a5f2 | 1559 | se = list_entry(sis->first_swap_extent.list.next, |
1da177e4 LT |
1560 | struct swap_extent, list); |
1561 | list_del(&se->list); | |
1562 | kfree(se); | |
1563 | } | |
62c230bc MG |
1564 | |
1565 | if (sis->flags & SWP_FILE) { | |
1566 | struct file *swap_file = sis->swap_file; | |
1567 | struct address_space *mapping = swap_file->f_mapping; | |
1568 | ||
1569 | sis->flags &= ~SWP_FILE; | |
1570 | mapping->a_ops->swap_deactivate(swap_file); | |
1571 | } | |
1da177e4 LT |
1572 | } |
1573 | ||
1574 | /* | |
1575 | * Add a block range (and the corresponding page range) into this swapdev's | |
11d31886 | 1576 | * extent list. The extent list is kept sorted in page order. |
1da177e4 | 1577 | * |
11d31886 | 1578 | * This function rather assumes that it is called in ascending page order. |
1da177e4 | 1579 | */ |
a509bc1a | 1580 | int |
1da177e4 LT |
1581 | add_swap_extent(struct swap_info_struct *sis, unsigned long start_page, |
1582 | unsigned long nr_pages, sector_t start_block) | |
1583 | { | |
1584 | struct swap_extent *se; | |
1585 | struct swap_extent *new_se; | |
1586 | struct list_head *lh; | |
1587 | ||
9625a5f2 HD |
1588 | if (start_page == 0) { |
1589 | se = &sis->first_swap_extent; | |
1590 | sis->curr_swap_extent = se; | |
1591 | se->start_page = 0; | |
1592 | se->nr_pages = nr_pages; | |
1593 | se->start_block = start_block; | |
1594 | return 1; | |
1595 | } else { | |
1596 | lh = sis->first_swap_extent.list.prev; /* Highest extent */ | |
1da177e4 | 1597 | se = list_entry(lh, struct swap_extent, list); |
11d31886 HD |
1598 | BUG_ON(se->start_page + se->nr_pages != start_page); |
1599 | if (se->start_block + se->nr_pages == start_block) { | |
1da177e4 LT |
1600 | /* Merge it */ |
1601 | se->nr_pages += nr_pages; | |
1602 | return 0; | |
1603 | } | |
1da177e4 LT |
1604 | } |
1605 | ||
1606 | /* | |
1607 | * No merge. Insert a new extent, preserving ordering. | |
1608 | */ | |
1609 | new_se = kmalloc(sizeof(*se), GFP_KERNEL); | |
1610 | if (new_se == NULL) | |
1611 | return -ENOMEM; | |
1612 | new_se->start_page = start_page; | |
1613 | new_se->nr_pages = nr_pages; | |
1614 | new_se->start_block = start_block; | |
1615 | ||
9625a5f2 | 1616 | list_add_tail(&new_se->list, &sis->first_swap_extent.list); |
53092a74 | 1617 | return 1; |
1da177e4 LT |
1618 | } |
1619 | ||
1620 | /* | |
1621 | * A `swap extent' is a simple thing which maps a contiguous range of pages | |
1622 | * onto a contiguous range of disk blocks. An ordered list of swap extents | |
1623 | * is built at swapon time and is then used at swap_writepage/swap_readpage | |
1624 | * time for locating where on disk a page belongs. | |
1625 | * | |
1626 | * If the swapfile is an S_ISBLK block device, a single extent is installed. | |
1627 | * This is done so that the main operating code can treat S_ISBLK and S_ISREG | |
1628 | * swap files identically. | |
1629 | * | |
1630 | * Whether the swapdev is an S_ISREG file or an S_ISBLK blockdev, the swap | |
1631 | * extent list operates in PAGE_SIZE disk blocks. Both S_ISREG and S_ISBLK | |
1632 | * swapfiles are handled *identically* after swapon time. | |
1633 | * | |
1634 | * For S_ISREG swapfiles, setup_swap_extents() will walk all the file's blocks | |
1635 | * and will parse them into an ordered extent list, in PAGE_SIZE chunks. If | |
1636 | * some stray blocks are found which do not fall within the PAGE_SIZE alignment | |
1637 | * requirements, they are simply tossed out - we will never use those blocks | |
1638 | * for swapping. | |
1639 | * | |
b0d9bcd4 | 1640 | * For S_ISREG swapfiles we set S_SWAPFILE across the life of the swapon. This |
1da177e4 LT |
1641 | * prevents root from shooting her foot off by ftruncating an in-use swapfile, |
1642 | * which will scribble on the fs. | |
1643 | * | |
1644 | * The amount of disk space which a single swap extent represents varies. | |
1645 | * Typically it is in the 1-4 megabyte range. So we can have hundreds of | |
1646 | * extents in the list. To avoid much list walking, we cache the previous | |
1647 | * search location in `curr_swap_extent', and start new searches from there. | |
1648 | * This is extremely effective. The average number of iterations in | |
1649 | * map_swap_page() has been measured at about 0.3 per page. - akpm. | |
1650 | */ | |
53092a74 | 1651 | static int setup_swap_extents(struct swap_info_struct *sis, sector_t *span) |
1da177e4 | 1652 | { |
62c230bc MG |
1653 | struct file *swap_file = sis->swap_file; |
1654 | struct address_space *mapping = swap_file->f_mapping; | |
1655 | struct inode *inode = mapping->host; | |
1da177e4 LT |
1656 | int ret; |
1657 | ||
1da177e4 LT |
1658 | if (S_ISBLK(inode->i_mode)) { |
1659 | ret = add_swap_extent(sis, 0, sis->max, 0); | |
53092a74 | 1660 | *span = sis->pages; |
a509bc1a | 1661 | return ret; |
1da177e4 LT |
1662 | } |
1663 | ||
62c230bc | 1664 | if (mapping->a_ops->swap_activate) { |
a509bc1a | 1665 | ret = mapping->a_ops->swap_activate(sis, swap_file, span); |
62c230bc MG |
1666 | if (!ret) { |
1667 | sis->flags |= SWP_FILE; | |
1668 | ret = add_swap_extent(sis, 0, sis->max, 0); | |
1669 | *span = sis->pages; | |
1670 | } | |
a509bc1a | 1671 | return ret; |
62c230bc MG |
1672 | } |
1673 | ||
a509bc1a | 1674 | return generic_swapfile_activate(sis, swap_file, span); |
1da177e4 LT |
1675 | } |
1676 | ||
cf0cac0a | 1677 | static void _enable_swap_info(struct swap_info_struct *p, int prio, |
2a8f9449 SL |
1678 | unsigned char *swap_map, |
1679 | struct swap_cluster_info *cluster_info) | |
40531542 CEB |
1680 | { |
1681 | int i, prev; | |
1682 | ||
40531542 CEB |
1683 | if (prio >= 0) |
1684 | p->prio = prio; | |
1685 | else | |
1686 | p->prio = --least_priority; | |
1687 | p->swap_map = swap_map; | |
2a8f9449 | 1688 | p->cluster_info = cluster_info; |
40531542 | 1689 | p->flags |= SWP_WRITEOK; |
ec8acf20 | 1690 | atomic_long_add(p->pages, &nr_swap_pages); |
40531542 CEB |
1691 | total_swap_pages += p->pages; |
1692 | ||
1693 | /* insert swap space into swap_list: */ | |
1694 | prev = -1; | |
1695 | for (i = swap_list.head; i >= 0; i = swap_info[i]->next) { | |
1696 | if (p->prio >= swap_info[i]->prio) | |
1697 | break; | |
1698 | prev = i; | |
1699 | } | |
1700 | p->next = i; | |
1701 | if (prev < 0) | |
1702 | swap_list.head = swap_list.next = p->type; | |
1703 | else | |
1704 | swap_info[prev]->next = p->type; | |
cf0cac0a CEB |
1705 | } |
1706 | ||
1707 | static void enable_swap_info(struct swap_info_struct *p, int prio, | |
1708 | unsigned char *swap_map, | |
2a8f9449 | 1709 | struct swap_cluster_info *cluster_info, |
cf0cac0a CEB |
1710 | unsigned long *frontswap_map) |
1711 | { | |
4f89849d | 1712 | frontswap_init(p->type, frontswap_map); |
cf0cac0a | 1713 | spin_lock(&swap_lock); |
ec8acf20 | 1714 | spin_lock(&p->lock); |
2a8f9449 | 1715 | _enable_swap_info(p, prio, swap_map, cluster_info); |
ec8acf20 | 1716 | spin_unlock(&p->lock); |
cf0cac0a CEB |
1717 | spin_unlock(&swap_lock); |
1718 | } | |
1719 | ||
1720 | static void reinsert_swap_info(struct swap_info_struct *p) | |
1721 | { | |
1722 | spin_lock(&swap_lock); | |
ec8acf20 | 1723 | spin_lock(&p->lock); |
2a8f9449 | 1724 | _enable_swap_info(p, p->prio, p->swap_map, p->cluster_info); |
ec8acf20 | 1725 | spin_unlock(&p->lock); |
40531542 CEB |
1726 | spin_unlock(&swap_lock); |
1727 | } | |
1728 | ||
c4ea37c2 | 1729 | SYSCALL_DEFINE1(swapoff, const char __user *, specialfile) |
1da177e4 | 1730 | { |
73c34b6a | 1731 | struct swap_info_struct *p = NULL; |
8d69aaee | 1732 | unsigned char *swap_map; |
2a8f9449 | 1733 | struct swap_cluster_info *cluster_info; |
4f89849d | 1734 | unsigned long *frontswap_map; |
1da177e4 LT |
1735 | struct file *swap_file, *victim; |
1736 | struct address_space *mapping; | |
1737 | struct inode *inode; | |
91a27b2a | 1738 | struct filename *pathname; |
1da177e4 LT |
1739 | int i, type, prev; |
1740 | int err; | |
886bb7e9 | 1741 | |
1da177e4 LT |
1742 | if (!capable(CAP_SYS_ADMIN)) |
1743 | return -EPERM; | |
1744 | ||
191c5424 AV |
1745 | BUG_ON(!current->mm); |
1746 | ||
1da177e4 | 1747 | pathname = getname(specialfile); |
1da177e4 | 1748 | if (IS_ERR(pathname)) |
f58b59c1 | 1749 | return PTR_ERR(pathname); |
1da177e4 | 1750 | |
669abf4e | 1751 | victim = file_open_name(pathname, O_RDWR|O_LARGEFILE, 0); |
1da177e4 LT |
1752 | err = PTR_ERR(victim); |
1753 | if (IS_ERR(victim)) | |
1754 | goto out; | |
1755 | ||
1756 | mapping = victim->f_mapping; | |
1757 | prev = -1; | |
5d337b91 | 1758 | spin_lock(&swap_lock); |
efa90a98 HD |
1759 | for (type = swap_list.head; type >= 0; type = swap_info[type]->next) { |
1760 | p = swap_info[type]; | |
22c6f8fd | 1761 | if (p->flags & SWP_WRITEOK) { |
1da177e4 LT |
1762 | if (p->swap_file->f_mapping == mapping) |
1763 | break; | |
1764 | } | |
1765 | prev = type; | |
1766 | } | |
1767 | if (type < 0) { | |
1768 | err = -EINVAL; | |
5d337b91 | 1769 | spin_unlock(&swap_lock); |
1da177e4 LT |
1770 | goto out_dput; |
1771 | } | |
191c5424 | 1772 | if (!security_vm_enough_memory_mm(current->mm, p->pages)) |
1da177e4 LT |
1773 | vm_unacct_memory(p->pages); |
1774 | else { | |
1775 | err = -ENOMEM; | |
5d337b91 | 1776 | spin_unlock(&swap_lock); |
1da177e4 LT |
1777 | goto out_dput; |
1778 | } | |
efa90a98 | 1779 | if (prev < 0) |
1da177e4 | 1780 | swap_list.head = p->next; |
efa90a98 HD |
1781 | else |
1782 | swap_info[prev]->next = p->next; | |
1da177e4 LT |
1783 | if (type == swap_list.next) { |
1784 | /* just pick something that's safe... */ | |
1785 | swap_list.next = swap_list.head; | |
1786 | } | |
ec8acf20 | 1787 | spin_lock(&p->lock); |
78ecba08 | 1788 | if (p->prio < 0) { |
efa90a98 HD |
1789 | for (i = p->next; i >= 0; i = swap_info[i]->next) |
1790 | swap_info[i]->prio = p->prio--; | |
78ecba08 HD |
1791 | least_priority++; |
1792 | } | |
ec8acf20 | 1793 | atomic_long_sub(p->pages, &nr_swap_pages); |
1da177e4 LT |
1794 | total_swap_pages -= p->pages; |
1795 | p->flags &= ~SWP_WRITEOK; | |
ec8acf20 | 1796 | spin_unlock(&p->lock); |
5d337b91 | 1797 | spin_unlock(&swap_lock); |
fb4f88dc | 1798 | |
e1e12d2f | 1799 | set_current_oom_origin(); |
38b5faf4 | 1800 | err = try_to_unuse(type, false, 0); /* force all pages to be unused */ |
e1e12d2f | 1801 | clear_current_oom_origin(); |
1da177e4 | 1802 | |
1da177e4 LT |
1803 | if (err) { |
1804 | /* re-insert swap space back into swap_list */ | |
cf0cac0a | 1805 | reinsert_swap_info(p); |
1da177e4 LT |
1806 | goto out_dput; |
1807 | } | |
52b7efdb | 1808 | |
5d337b91 | 1809 | destroy_swap_extents(p); |
570a335b HD |
1810 | if (p->flags & SWP_CONTINUED) |
1811 | free_swap_count_continuations(p); | |
1812 | ||
fc0abb14 | 1813 | mutex_lock(&swapon_mutex); |
5d337b91 | 1814 | spin_lock(&swap_lock); |
ec8acf20 | 1815 | spin_lock(&p->lock); |
5d337b91 HD |
1816 | drain_mmlist(); |
1817 | ||
52b7efdb | 1818 | /* wait for anyone still in scan_swap_map */ |
52b7efdb HD |
1819 | p->highest_bit = 0; /* cuts scans short */ |
1820 | while (p->flags >= SWP_SCANNING) { | |
ec8acf20 | 1821 | spin_unlock(&p->lock); |
5d337b91 | 1822 | spin_unlock(&swap_lock); |
13e4b57f | 1823 | schedule_timeout_uninterruptible(1); |
5d337b91 | 1824 | spin_lock(&swap_lock); |
ec8acf20 | 1825 | spin_lock(&p->lock); |
52b7efdb | 1826 | } |
52b7efdb | 1827 | |
1da177e4 LT |
1828 | swap_file = p->swap_file; |
1829 | p->swap_file = NULL; | |
1830 | p->max = 0; | |
1831 | swap_map = p->swap_map; | |
1832 | p->swap_map = NULL; | |
2a8f9449 SL |
1833 | cluster_info = p->cluster_info; |
1834 | p->cluster_info = NULL; | |
1da177e4 | 1835 | p->flags = 0; |
4f89849d MK |
1836 | frontswap_map = frontswap_map_get(p); |
1837 | frontswap_map_set(p, NULL); | |
ec8acf20 | 1838 | spin_unlock(&p->lock); |
5d337b91 | 1839 | spin_unlock(&swap_lock); |
4f89849d | 1840 | frontswap_invalidate_area(type); |
fc0abb14 | 1841 | mutex_unlock(&swapon_mutex); |
1da177e4 | 1842 | vfree(swap_map); |
2a8f9449 | 1843 | vfree(cluster_info); |
4f89849d | 1844 | vfree(frontswap_map); |
27a7faa0 KH |
1845 | /* Destroy swap account informatin */ |
1846 | swap_cgroup_swapoff(type); | |
1847 | ||
1da177e4 LT |
1848 | inode = mapping->host; |
1849 | if (S_ISBLK(inode->i_mode)) { | |
1850 | struct block_device *bdev = I_BDEV(inode); | |
1851 | set_blocksize(bdev, p->old_block_size); | |
e525fd89 | 1852 | blkdev_put(bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL); |
1da177e4 | 1853 | } else { |
1b1dcc1b | 1854 | mutex_lock(&inode->i_mutex); |
1da177e4 | 1855 | inode->i_flags &= ~S_SWAPFILE; |
1b1dcc1b | 1856 | mutex_unlock(&inode->i_mutex); |
1da177e4 LT |
1857 | } |
1858 | filp_close(swap_file, NULL); | |
1859 | err = 0; | |
66d7dd51 KS |
1860 | atomic_inc(&proc_poll_event); |
1861 | wake_up_interruptible(&proc_poll_wait); | |
1da177e4 LT |
1862 | |
1863 | out_dput: | |
1864 | filp_close(victim, NULL); | |
1865 | out: | |
f58b59c1 | 1866 | putname(pathname); |
1da177e4 LT |
1867 | return err; |
1868 | } | |
1869 | ||
1870 | #ifdef CONFIG_PROC_FS | |
66d7dd51 KS |
1871 | static unsigned swaps_poll(struct file *file, poll_table *wait) |
1872 | { | |
f1514638 | 1873 | struct seq_file *seq = file->private_data; |
66d7dd51 KS |
1874 | |
1875 | poll_wait(file, &proc_poll_wait, wait); | |
1876 | ||
f1514638 KS |
1877 | if (seq->poll_event != atomic_read(&proc_poll_event)) { |
1878 | seq->poll_event = atomic_read(&proc_poll_event); | |
66d7dd51 KS |
1879 | return POLLIN | POLLRDNORM | POLLERR | POLLPRI; |
1880 | } | |
1881 | ||
1882 | return POLLIN | POLLRDNORM; | |
1883 | } | |
1884 | ||
1da177e4 LT |
1885 | /* iterator */ |
1886 | static void *swap_start(struct seq_file *swap, loff_t *pos) | |
1887 | { | |
efa90a98 HD |
1888 | struct swap_info_struct *si; |
1889 | int type; | |
1da177e4 LT |
1890 | loff_t l = *pos; |
1891 | ||
fc0abb14 | 1892 | mutex_lock(&swapon_mutex); |
1da177e4 | 1893 | |
881e4aab SS |
1894 | if (!l) |
1895 | return SEQ_START_TOKEN; | |
1896 | ||
efa90a98 HD |
1897 | for (type = 0; type < nr_swapfiles; type++) { |
1898 | smp_rmb(); /* read nr_swapfiles before swap_info[type] */ | |
1899 | si = swap_info[type]; | |
1900 | if (!(si->flags & SWP_USED) || !si->swap_map) | |
1da177e4 | 1901 | continue; |
881e4aab | 1902 | if (!--l) |
efa90a98 | 1903 | return si; |
1da177e4 LT |
1904 | } |
1905 | ||
1906 | return NULL; | |
1907 | } | |
1908 | ||
1909 | static void *swap_next(struct seq_file *swap, void *v, loff_t *pos) | |
1910 | { | |
efa90a98 HD |
1911 | struct swap_info_struct *si = v; |
1912 | int type; | |
1da177e4 | 1913 | |
881e4aab | 1914 | if (v == SEQ_START_TOKEN) |
efa90a98 HD |
1915 | type = 0; |
1916 | else | |
1917 | type = si->type + 1; | |
881e4aab | 1918 | |
efa90a98 HD |
1919 | for (; type < nr_swapfiles; type++) { |
1920 | smp_rmb(); /* read nr_swapfiles before swap_info[type] */ | |
1921 | si = swap_info[type]; | |
1922 | if (!(si->flags & SWP_USED) || !si->swap_map) | |
1da177e4 LT |
1923 | continue; |
1924 | ++*pos; | |
efa90a98 | 1925 | return si; |
1da177e4 LT |
1926 | } |
1927 | ||
1928 | return NULL; | |
1929 | } | |
1930 | ||
1931 | static void swap_stop(struct seq_file *swap, void *v) | |
1932 | { | |
fc0abb14 | 1933 | mutex_unlock(&swapon_mutex); |
1da177e4 LT |
1934 | } |
1935 | ||
1936 | static int swap_show(struct seq_file *swap, void *v) | |
1937 | { | |
efa90a98 | 1938 | struct swap_info_struct *si = v; |
1da177e4 LT |
1939 | struct file *file; |
1940 | int len; | |
1941 | ||
efa90a98 | 1942 | if (si == SEQ_START_TOKEN) { |
881e4aab SS |
1943 | seq_puts(swap,"Filename\t\t\t\tType\t\tSize\tUsed\tPriority\n"); |
1944 | return 0; | |
1945 | } | |
1da177e4 | 1946 | |
efa90a98 | 1947 | file = si->swap_file; |
c32c2f63 | 1948 | len = seq_path(swap, &file->f_path, " \t\n\\"); |
6eb396dc | 1949 | seq_printf(swap, "%*s%s\t%u\t%u\t%d\n", |
886bb7e9 | 1950 | len < 40 ? 40 - len : 1, " ", |
496ad9aa | 1951 | S_ISBLK(file_inode(file)->i_mode) ? |
1da177e4 | 1952 | "partition" : "file\t", |
efa90a98 HD |
1953 | si->pages << (PAGE_SHIFT - 10), |
1954 | si->inuse_pages << (PAGE_SHIFT - 10), | |
1955 | si->prio); | |
1da177e4 LT |
1956 | return 0; |
1957 | } | |
1958 | ||
15ad7cdc | 1959 | static const struct seq_operations swaps_op = { |
1da177e4 LT |
1960 | .start = swap_start, |
1961 | .next = swap_next, | |
1962 | .stop = swap_stop, | |
1963 | .show = swap_show | |
1964 | }; | |
1965 | ||
1966 | static int swaps_open(struct inode *inode, struct file *file) | |
1967 | { | |
f1514638 | 1968 | struct seq_file *seq; |
66d7dd51 KS |
1969 | int ret; |
1970 | ||
66d7dd51 | 1971 | ret = seq_open(file, &swaps_op); |
f1514638 | 1972 | if (ret) |
66d7dd51 | 1973 | return ret; |
66d7dd51 | 1974 | |
f1514638 KS |
1975 | seq = file->private_data; |
1976 | seq->poll_event = atomic_read(&proc_poll_event); | |
1977 | return 0; | |
1da177e4 LT |
1978 | } |
1979 | ||
15ad7cdc | 1980 | static const struct file_operations proc_swaps_operations = { |
1da177e4 LT |
1981 | .open = swaps_open, |
1982 | .read = seq_read, | |
1983 | .llseek = seq_lseek, | |
1984 | .release = seq_release, | |
66d7dd51 | 1985 | .poll = swaps_poll, |
1da177e4 LT |
1986 | }; |
1987 | ||
1988 | static int __init procswaps_init(void) | |
1989 | { | |
3d71f86f | 1990 | proc_create("swaps", 0, NULL, &proc_swaps_operations); |
1da177e4 LT |
1991 | return 0; |
1992 | } | |
1993 | __initcall(procswaps_init); | |
1994 | #endif /* CONFIG_PROC_FS */ | |
1995 | ||
1796316a JB |
1996 | #ifdef MAX_SWAPFILES_CHECK |
1997 | static int __init max_swapfiles_check(void) | |
1998 | { | |
1999 | MAX_SWAPFILES_CHECK(); | |
2000 | return 0; | |
2001 | } | |
2002 | late_initcall(max_swapfiles_check); | |
2003 | #endif | |
2004 | ||
53cbb243 | 2005 | static struct swap_info_struct *alloc_swap_info(void) |
1da177e4 | 2006 | { |
73c34b6a | 2007 | struct swap_info_struct *p; |
1da177e4 | 2008 | unsigned int type; |
efa90a98 HD |
2009 | |
2010 | p = kzalloc(sizeof(*p), GFP_KERNEL); | |
2011 | if (!p) | |
53cbb243 | 2012 | return ERR_PTR(-ENOMEM); |
efa90a98 | 2013 | |
5d337b91 | 2014 | spin_lock(&swap_lock); |
efa90a98 HD |
2015 | for (type = 0; type < nr_swapfiles; type++) { |
2016 | if (!(swap_info[type]->flags & SWP_USED)) | |
1da177e4 | 2017 | break; |
efa90a98 | 2018 | } |
0697212a | 2019 | if (type >= MAX_SWAPFILES) { |
5d337b91 | 2020 | spin_unlock(&swap_lock); |
efa90a98 | 2021 | kfree(p); |
730c0581 | 2022 | return ERR_PTR(-EPERM); |
1da177e4 | 2023 | } |
efa90a98 HD |
2024 | if (type >= nr_swapfiles) { |
2025 | p->type = type; | |
2026 | swap_info[type] = p; | |
2027 | /* | |
2028 | * Write swap_info[type] before nr_swapfiles, in case a | |
2029 | * racing procfs swap_start() or swap_next() is reading them. | |
2030 | * (We never shrink nr_swapfiles, we never free this entry.) | |
2031 | */ | |
2032 | smp_wmb(); | |
2033 | nr_swapfiles++; | |
2034 | } else { | |
2035 | kfree(p); | |
2036 | p = swap_info[type]; | |
2037 | /* | |
2038 | * Do not memset this entry: a racing procfs swap_next() | |
2039 | * would be relying on p->type to remain valid. | |
2040 | */ | |
2041 | } | |
9625a5f2 | 2042 | INIT_LIST_HEAD(&p->first_swap_extent.list); |
1da177e4 | 2043 | p->flags = SWP_USED; |
1da177e4 | 2044 | p->next = -1; |
5d337b91 | 2045 | spin_unlock(&swap_lock); |
ec8acf20 | 2046 | spin_lock_init(&p->lock); |
efa90a98 | 2047 | |
53cbb243 | 2048 | return p; |
53cbb243 CEB |
2049 | } |
2050 | ||
4d0e1e10 CEB |
2051 | static int claim_swapfile(struct swap_info_struct *p, struct inode *inode) |
2052 | { | |
2053 | int error; | |
2054 | ||
2055 | if (S_ISBLK(inode->i_mode)) { | |
2056 | p->bdev = bdgrab(I_BDEV(inode)); | |
2057 | error = blkdev_get(p->bdev, | |
2058 | FMODE_READ | FMODE_WRITE | FMODE_EXCL, | |
2059 | sys_swapon); | |
2060 | if (error < 0) { | |
2061 | p->bdev = NULL; | |
87ade72a | 2062 | return -EINVAL; |
4d0e1e10 CEB |
2063 | } |
2064 | p->old_block_size = block_size(p->bdev); | |
2065 | error = set_blocksize(p->bdev, PAGE_SIZE); | |
2066 | if (error < 0) | |
87ade72a | 2067 | return error; |
4d0e1e10 CEB |
2068 | p->flags |= SWP_BLKDEV; |
2069 | } else if (S_ISREG(inode->i_mode)) { | |
2070 | p->bdev = inode->i_sb->s_bdev; | |
2071 | mutex_lock(&inode->i_mutex); | |
87ade72a CEB |
2072 | if (IS_SWAPFILE(inode)) |
2073 | return -EBUSY; | |
2074 | } else | |
2075 | return -EINVAL; | |
4d0e1e10 CEB |
2076 | |
2077 | return 0; | |
4d0e1e10 CEB |
2078 | } |
2079 | ||
ca8bd38b CEB |
2080 | static unsigned long read_swap_header(struct swap_info_struct *p, |
2081 | union swap_header *swap_header, | |
2082 | struct inode *inode) | |
2083 | { | |
2084 | int i; | |
2085 | unsigned long maxpages; | |
2086 | unsigned long swapfilepages; | |
d6bbbd29 | 2087 | unsigned long last_page; |
ca8bd38b CEB |
2088 | |
2089 | if (memcmp("SWAPSPACE2", swap_header->magic.magic, 10)) { | |
465c47fd | 2090 | pr_err("Unable to find swap-space signature\n"); |
38719025 | 2091 | return 0; |
ca8bd38b CEB |
2092 | } |
2093 | ||
2094 | /* swap partition endianess hack... */ | |
2095 | if (swab32(swap_header->info.version) == 1) { | |
2096 | swab32s(&swap_header->info.version); | |
2097 | swab32s(&swap_header->info.last_page); | |
2098 | swab32s(&swap_header->info.nr_badpages); | |
2099 | for (i = 0; i < swap_header->info.nr_badpages; i++) | |
2100 | swab32s(&swap_header->info.badpages[i]); | |
2101 | } | |
2102 | /* Check the swap header's sub-version */ | |
2103 | if (swap_header->info.version != 1) { | |
465c47fd AM |
2104 | pr_warn("Unable to handle swap header version %d\n", |
2105 | swap_header->info.version); | |
38719025 | 2106 | return 0; |
ca8bd38b CEB |
2107 | } |
2108 | ||
2109 | p->lowest_bit = 1; | |
2110 | p->cluster_next = 1; | |
2111 | p->cluster_nr = 0; | |
2112 | ||
2113 | /* | |
2114 | * Find out how many pages are allowed for a single swap | |
9b15b817 | 2115 | * device. There are two limiting factors: 1) the number |
a2c16d6c HD |
2116 | * of bits for the swap offset in the swp_entry_t type, and |
2117 | * 2) the number of bits in the swap pte as defined by the | |
9b15b817 | 2118 | * different architectures. In order to find the |
a2c16d6c | 2119 | * largest possible bit mask, a swap entry with swap type 0 |
ca8bd38b | 2120 | * and swap offset ~0UL is created, encoded to a swap pte, |
a2c16d6c | 2121 | * decoded to a swp_entry_t again, and finally the swap |
ca8bd38b CEB |
2122 | * offset is extracted. This will mask all the bits from |
2123 | * the initial ~0UL mask that can't be encoded in either | |
2124 | * the swp_entry_t or the architecture definition of a | |
9b15b817 | 2125 | * swap pte. |
ca8bd38b CEB |
2126 | */ |
2127 | maxpages = swp_offset(pte_to_swp_entry( | |
9b15b817 | 2128 | swp_entry_to_pte(swp_entry(0, ~0UL)))) + 1; |
d6bbbd29 RJ |
2129 | last_page = swap_header->info.last_page; |
2130 | if (last_page > maxpages) { | |
465c47fd | 2131 | pr_warn("Truncating oversized swap area, only using %luk out of %luk\n", |
d6bbbd29 RJ |
2132 | maxpages << (PAGE_SHIFT - 10), |
2133 | last_page << (PAGE_SHIFT - 10)); | |
2134 | } | |
2135 | if (maxpages > last_page) { | |
2136 | maxpages = last_page + 1; | |
ca8bd38b CEB |
2137 | /* p->max is an unsigned int: don't overflow it */ |
2138 | if ((unsigned int)maxpages == 0) | |
2139 | maxpages = UINT_MAX; | |
2140 | } | |
2141 | p->highest_bit = maxpages - 1; | |
2142 | ||
2143 | if (!maxpages) | |
38719025 | 2144 | return 0; |
ca8bd38b CEB |
2145 | swapfilepages = i_size_read(inode) >> PAGE_SHIFT; |
2146 | if (swapfilepages && maxpages > swapfilepages) { | |
465c47fd | 2147 | pr_warn("Swap area shorter than signature indicates\n"); |
38719025 | 2148 | return 0; |
ca8bd38b CEB |
2149 | } |
2150 | if (swap_header->info.nr_badpages && S_ISREG(inode->i_mode)) | |
38719025 | 2151 | return 0; |
ca8bd38b | 2152 | if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES) |
38719025 | 2153 | return 0; |
ca8bd38b CEB |
2154 | |
2155 | return maxpages; | |
ca8bd38b CEB |
2156 | } |
2157 | ||
915d4d7b CEB |
2158 | static int setup_swap_map_and_extents(struct swap_info_struct *p, |
2159 | union swap_header *swap_header, | |
2160 | unsigned char *swap_map, | |
2a8f9449 | 2161 | struct swap_cluster_info *cluster_info, |
915d4d7b CEB |
2162 | unsigned long maxpages, |
2163 | sector_t *span) | |
2164 | { | |
2165 | int i; | |
915d4d7b CEB |
2166 | unsigned int nr_good_pages; |
2167 | int nr_extents; | |
2a8f9449 SL |
2168 | unsigned long nr_clusters = DIV_ROUND_UP(maxpages, SWAPFILE_CLUSTER); |
2169 | unsigned long idx = p->cluster_next / SWAPFILE_CLUSTER; | |
915d4d7b CEB |
2170 | |
2171 | nr_good_pages = maxpages - 1; /* omit header page */ | |
2172 | ||
2a8f9449 SL |
2173 | cluster_set_null(&p->free_cluster_head); |
2174 | cluster_set_null(&p->free_cluster_tail); | |
2175 | ||
915d4d7b CEB |
2176 | for (i = 0; i < swap_header->info.nr_badpages; i++) { |
2177 | unsigned int page_nr = swap_header->info.badpages[i]; | |
bdb8e3f6 CEB |
2178 | if (page_nr == 0 || page_nr > swap_header->info.last_page) |
2179 | return -EINVAL; | |
915d4d7b CEB |
2180 | if (page_nr < maxpages) { |
2181 | swap_map[page_nr] = SWAP_MAP_BAD; | |
2182 | nr_good_pages--; | |
2a8f9449 SL |
2183 | /* |
2184 | * Haven't marked the cluster free yet, no list | |
2185 | * operation involved | |
2186 | */ | |
2187 | inc_cluster_info_page(p, cluster_info, page_nr); | |
915d4d7b CEB |
2188 | } |
2189 | } | |
2190 | ||
2a8f9449 SL |
2191 | /* Haven't marked the cluster free yet, no list operation involved */ |
2192 | for (i = maxpages; i < round_up(maxpages, SWAPFILE_CLUSTER); i++) | |
2193 | inc_cluster_info_page(p, cluster_info, i); | |
2194 | ||
915d4d7b CEB |
2195 | if (nr_good_pages) { |
2196 | swap_map[0] = SWAP_MAP_BAD; | |
2a8f9449 SL |
2197 | /* |
2198 | * Not mark the cluster free yet, no list | |
2199 | * operation involved | |
2200 | */ | |
2201 | inc_cluster_info_page(p, cluster_info, 0); | |
915d4d7b CEB |
2202 | p->max = maxpages; |
2203 | p->pages = nr_good_pages; | |
2204 | nr_extents = setup_swap_extents(p, span); | |
bdb8e3f6 CEB |
2205 | if (nr_extents < 0) |
2206 | return nr_extents; | |
915d4d7b CEB |
2207 | nr_good_pages = p->pages; |
2208 | } | |
2209 | if (!nr_good_pages) { | |
465c47fd | 2210 | pr_warn("Empty swap-file\n"); |
bdb8e3f6 | 2211 | return -EINVAL; |
915d4d7b CEB |
2212 | } |
2213 | ||
2a8f9449 SL |
2214 | if (!cluster_info) |
2215 | return nr_extents; | |
2216 | ||
2217 | for (i = 0; i < nr_clusters; i++) { | |
2218 | if (!cluster_count(&cluster_info[idx])) { | |
2219 | cluster_set_flag(&cluster_info[idx], CLUSTER_FLAG_FREE); | |
2220 | if (cluster_is_null(&p->free_cluster_head)) { | |
2221 | cluster_set_next_flag(&p->free_cluster_head, | |
2222 | idx, 0); | |
2223 | cluster_set_next_flag(&p->free_cluster_tail, | |
2224 | idx, 0); | |
2225 | } else { | |
2226 | unsigned int tail; | |
2227 | ||
2228 | tail = cluster_next(&p->free_cluster_tail); | |
2229 | cluster_set_next(&cluster_info[tail], idx); | |
2230 | cluster_set_next_flag(&p->free_cluster_tail, | |
2231 | idx, 0); | |
2232 | } | |
2233 | } | |
2234 | idx++; | |
2235 | if (idx == nr_clusters) | |
2236 | idx = 0; | |
2237 | } | |
915d4d7b | 2238 | return nr_extents; |
915d4d7b CEB |
2239 | } |
2240 | ||
dcf6b7dd RA |
2241 | /* |
2242 | * Helper to sys_swapon determining if a given swap | |
2243 | * backing device queue supports DISCARD operations. | |
2244 | */ | |
2245 | static bool swap_discardable(struct swap_info_struct *si) | |
2246 | { | |
2247 | struct request_queue *q = bdev_get_queue(si->bdev); | |
2248 | ||
2249 | if (!q || !blk_queue_discard(q)) | |
2250 | return false; | |
2251 | ||
2252 | return true; | |
2253 | } | |
2254 | ||
53cbb243 CEB |
2255 | SYSCALL_DEFINE2(swapon, const char __user *, specialfile, int, swap_flags) |
2256 | { | |
2257 | struct swap_info_struct *p; | |
91a27b2a | 2258 | struct filename *name; |
53cbb243 CEB |
2259 | struct file *swap_file = NULL; |
2260 | struct address_space *mapping; | |
40531542 CEB |
2261 | int i; |
2262 | int prio; | |
53cbb243 CEB |
2263 | int error; |
2264 | union swap_header *swap_header; | |
915d4d7b | 2265 | int nr_extents; |
53cbb243 CEB |
2266 | sector_t span; |
2267 | unsigned long maxpages; | |
53cbb243 | 2268 | unsigned char *swap_map = NULL; |
2a8f9449 | 2269 | struct swap_cluster_info *cluster_info = NULL; |
38b5faf4 | 2270 | unsigned long *frontswap_map = NULL; |
53cbb243 CEB |
2271 | struct page *page = NULL; |
2272 | struct inode *inode = NULL; | |
53cbb243 | 2273 | |
d15cab97 HD |
2274 | if (swap_flags & ~SWAP_FLAGS_VALID) |
2275 | return -EINVAL; | |
2276 | ||
53cbb243 CEB |
2277 | if (!capable(CAP_SYS_ADMIN)) |
2278 | return -EPERM; | |
2279 | ||
2280 | p = alloc_swap_info(); | |
2542e513 CEB |
2281 | if (IS_ERR(p)) |
2282 | return PTR_ERR(p); | |
53cbb243 | 2283 | |
1da177e4 | 2284 | name = getname(specialfile); |
1da177e4 | 2285 | if (IS_ERR(name)) { |
7de7fb6b | 2286 | error = PTR_ERR(name); |
1da177e4 | 2287 | name = NULL; |
bd69010b | 2288 | goto bad_swap; |
1da177e4 | 2289 | } |
669abf4e | 2290 | swap_file = file_open_name(name, O_RDWR|O_LARGEFILE, 0); |
1da177e4 | 2291 | if (IS_ERR(swap_file)) { |
7de7fb6b | 2292 | error = PTR_ERR(swap_file); |
1da177e4 | 2293 | swap_file = NULL; |
bd69010b | 2294 | goto bad_swap; |
1da177e4 LT |
2295 | } |
2296 | ||
2297 | p->swap_file = swap_file; | |
2298 | mapping = swap_file->f_mapping; | |
1da177e4 | 2299 | |
1da177e4 | 2300 | for (i = 0; i < nr_swapfiles; i++) { |
efa90a98 | 2301 | struct swap_info_struct *q = swap_info[i]; |
1da177e4 | 2302 | |
e8e6c2ec | 2303 | if (q == p || !q->swap_file) |
1da177e4 | 2304 | continue; |
7de7fb6b CEB |
2305 | if (mapping == q->swap_file->f_mapping) { |
2306 | error = -EBUSY; | |
1da177e4 | 2307 | goto bad_swap; |
7de7fb6b | 2308 | } |
1da177e4 LT |
2309 | } |
2310 | ||
2130781e CEB |
2311 | inode = mapping->host; |
2312 | /* If S_ISREG(inode->i_mode) will do mutex_lock(&inode->i_mutex); */ | |
4d0e1e10 CEB |
2313 | error = claim_swapfile(p, inode); |
2314 | if (unlikely(error)) | |
1da177e4 | 2315 | goto bad_swap; |
1da177e4 | 2316 | |
1da177e4 LT |
2317 | /* |
2318 | * Read the swap header. | |
2319 | */ | |
2320 | if (!mapping->a_ops->readpage) { | |
2321 | error = -EINVAL; | |
2322 | goto bad_swap; | |
2323 | } | |
090d2b18 | 2324 | page = read_mapping_page(mapping, 0, swap_file); |
1da177e4 LT |
2325 | if (IS_ERR(page)) { |
2326 | error = PTR_ERR(page); | |
2327 | goto bad_swap; | |
2328 | } | |
81e33971 | 2329 | swap_header = kmap(page); |
1da177e4 | 2330 | |
ca8bd38b CEB |
2331 | maxpages = read_swap_header(p, swap_header, inode); |
2332 | if (unlikely(!maxpages)) { | |
1da177e4 LT |
2333 | error = -EINVAL; |
2334 | goto bad_swap; | |
2335 | } | |
886bb7e9 | 2336 | |
81e33971 | 2337 | /* OK, set up the swap map and apply the bad block list */ |
803d0c83 | 2338 | swap_map = vzalloc(maxpages); |
81e33971 HD |
2339 | if (!swap_map) { |
2340 | error = -ENOMEM; | |
2341 | goto bad_swap; | |
2342 | } | |
2a8f9449 SL |
2343 | if (p->bdev && blk_queue_nonrot(bdev_get_queue(p->bdev))) { |
2344 | p->flags |= SWP_SOLIDSTATE; | |
2345 | /* | |
2346 | * select a random position to start with to help wear leveling | |
2347 | * SSD | |
2348 | */ | |
2349 | p->cluster_next = 1 + (prandom_u32() % p->highest_bit); | |
2350 | ||
2351 | cluster_info = vzalloc(DIV_ROUND_UP(maxpages, | |
2352 | SWAPFILE_CLUSTER) * sizeof(*cluster_info)); | |
2353 | if (!cluster_info) { | |
2354 | error = -ENOMEM; | |
2355 | goto bad_swap; | |
2356 | } | |
2357 | } | |
1da177e4 | 2358 | |
1421ef3c CEB |
2359 | error = swap_cgroup_swapon(p->type, maxpages); |
2360 | if (error) | |
2361 | goto bad_swap; | |
2362 | ||
915d4d7b | 2363 | nr_extents = setup_swap_map_and_extents(p, swap_header, swap_map, |
2a8f9449 | 2364 | cluster_info, maxpages, &span); |
915d4d7b CEB |
2365 | if (unlikely(nr_extents < 0)) { |
2366 | error = nr_extents; | |
1da177e4 LT |
2367 | goto bad_swap; |
2368 | } | |
38b5faf4 DM |
2369 | /* frontswap enabled? set up bit-per-page map for frontswap */ |
2370 | if (frontswap_enabled) | |
7b57976d | 2371 | frontswap_map = vzalloc(BITS_TO_LONGS(maxpages) * sizeof(long)); |
1da177e4 | 2372 | |
2a8f9449 SL |
2373 | if (p->bdev &&(swap_flags & SWAP_FLAG_DISCARD) && swap_discardable(p)) { |
2374 | /* | |
2375 | * When discard is enabled for swap with no particular | |
2376 | * policy flagged, we set all swap discard flags here in | |
2377 | * order to sustain backward compatibility with older | |
2378 | * swapon(8) releases. | |
2379 | */ | |
2380 | p->flags |= (SWP_DISCARDABLE | SWP_AREA_DISCARD | | |
2381 | SWP_PAGE_DISCARD); | |
dcf6b7dd | 2382 | |
2a8f9449 SL |
2383 | /* |
2384 | * By flagging sys_swapon, a sysadmin can tell us to | |
2385 | * either do single-time area discards only, or to just | |
2386 | * perform discards for released swap page-clusters. | |
2387 | * Now it's time to adjust the p->flags accordingly. | |
2388 | */ | |
2389 | if (swap_flags & SWAP_FLAG_DISCARD_ONCE) | |
2390 | p->flags &= ~SWP_PAGE_DISCARD; | |
2391 | else if (swap_flags & SWAP_FLAG_DISCARD_PAGES) | |
2392 | p->flags &= ~SWP_AREA_DISCARD; | |
2393 | ||
2394 | /* issue a swapon-time discard if it's still required */ | |
2395 | if (p->flags & SWP_AREA_DISCARD) { | |
2396 | int err = discard_swap(p); | |
2397 | if (unlikely(err)) | |
2398 | pr_err("swapon: discard_swap(%p): %d\n", | |
2399 | p, err); | |
dcf6b7dd | 2400 | } |
20137a49 | 2401 | } |
6a6ba831 | 2402 | |
fc0abb14 | 2403 | mutex_lock(&swapon_mutex); |
40531542 | 2404 | prio = -1; |
78ecba08 | 2405 | if (swap_flags & SWAP_FLAG_PREFER) |
40531542 | 2406 | prio = |
78ecba08 | 2407 | (swap_flags & SWAP_FLAG_PRIO_MASK) >> SWAP_FLAG_PRIO_SHIFT; |
2a8f9449 | 2408 | enable_swap_info(p, prio, swap_map, cluster_info, frontswap_map); |
c69dbfb8 | 2409 | |
465c47fd | 2410 | pr_info("Adding %uk swap on %s. " |
dcf6b7dd | 2411 | "Priority:%d extents:%d across:%lluk %s%s%s%s%s\n", |
91a27b2a | 2412 | p->pages<<(PAGE_SHIFT-10), name->name, p->prio, |
c69dbfb8 CEB |
2413 | nr_extents, (unsigned long long)span<<(PAGE_SHIFT-10), |
2414 | (p->flags & SWP_SOLIDSTATE) ? "SS" : "", | |
38b5faf4 | 2415 | (p->flags & SWP_DISCARDABLE) ? "D" : "", |
dcf6b7dd RA |
2416 | (p->flags & SWP_AREA_DISCARD) ? "s" : "", |
2417 | (p->flags & SWP_PAGE_DISCARD) ? "c" : "", | |
38b5faf4 | 2418 | (frontswap_map) ? "FS" : ""); |
c69dbfb8 | 2419 | |
fc0abb14 | 2420 | mutex_unlock(&swapon_mutex); |
66d7dd51 KS |
2421 | atomic_inc(&proc_poll_event); |
2422 | wake_up_interruptible(&proc_poll_wait); | |
2423 | ||
9b01c350 CEB |
2424 | if (S_ISREG(inode->i_mode)) |
2425 | inode->i_flags |= S_SWAPFILE; | |
1da177e4 LT |
2426 | error = 0; |
2427 | goto out; | |
2428 | bad_swap: | |
bd69010b | 2429 | if (inode && S_ISBLK(inode->i_mode) && p->bdev) { |
f2090d2d CEB |
2430 | set_blocksize(p->bdev, p->old_block_size); |
2431 | blkdev_put(p->bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL); | |
1da177e4 | 2432 | } |
4cd3bb10 | 2433 | destroy_swap_extents(p); |
e8e6c2ec | 2434 | swap_cgroup_swapoff(p->type); |
5d337b91 | 2435 | spin_lock(&swap_lock); |
1da177e4 | 2436 | p->swap_file = NULL; |
1da177e4 | 2437 | p->flags = 0; |
5d337b91 | 2438 | spin_unlock(&swap_lock); |
1da177e4 | 2439 | vfree(swap_map); |
2a8f9449 | 2440 | vfree(cluster_info); |
52c50567 | 2441 | if (swap_file) { |
2130781e | 2442 | if (inode && S_ISREG(inode->i_mode)) { |
52c50567 | 2443 | mutex_unlock(&inode->i_mutex); |
2130781e CEB |
2444 | inode = NULL; |
2445 | } | |
1da177e4 | 2446 | filp_close(swap_file, NULL); |
52c50567 | 2447 | } |
1da177e4 LT |
2448 | out: |
2449 | if (page && !IS_ERR(page)) { | |
2450 | kunmap(page); | |
2451 | page_cache_release(page); | |
2452 | } | |
2453 | if (name) | |
2454 | putname(name); | |
9b01c350 | 2455 | if (inode && S_ISREG(inode->i_mode)) |
1b1dcc1b | 2456 | mutex_unlock(&inode->i_mutex); |
1da177e4 LT |
2457 | return error; |
2458 | } | |
2459 | ||
2460 | void si_swapinfo(struct sysinfo *val) | |
2461 | { | |
efa90a98 | 2462 | unsigned int type; |
1da177e4 LT |
2463 | unsigned long nr_to_be_unused = 0; |
2464 | ||
5d337b91 | 2465 | spin_lock(&swap_lock); |
efa90a98 HD |
2466 | for (type = 0; type < nr_swapfiles; type++) { |
2467 | struct swap_info_struct *si = swap_info[type]; | |
2468 | ||
2469 | if ((si->flags & SWP_USED) && !(si->flags & SWP_WRITEOK)) | |
2470 | nr_to_be_unused += si->inuse_pages; | |
1da177e4 | 2471 | } |
ec8acf20 | 2472 | val->freeswap = atomic_long_read(&nr_swap_pages) + nr_to_be_unused; |
1da177e4 | 2473 | val->totalswap = total_swap_pages + nr_to_be_unused; |
5d337b91 | 2474 | spin_unlock(&swap_lock); |
1da177e4 LT |
2475 | } |
2476 | ||
2477 | /* | |
2478 | * Verify that a swap entry is valid and increment its swap map count. | |
2479 | * | |
355cfa73 KH |
2480 | * Returns error code in following case. |
2481 | * - success -> 0 | |
2482 | * - swp_entry is invalid -> EINVAL | |
2483 | * - swp_entry is migration entry -> EINVAL | |
2484 | * - swap-cache reference is requested but there is already one. -> EEXIST | |
2485 | * - swap-cache reference is requested but the entry is not used. -> ENOENT | |
570a335b | 2486 | * - swap-mapped reference requested but needs continued swap count. -> ENOMEM |
1da177e4 | 2487 | */ |
8d69aaee | 2488 | static int __swap_duplicate(swp_entry_t entry, unsigned char usage) |
1da177e4 | 2489 | { |
73c34b6a | 2490 | struct swap_info_struct *p; |
1da177e4 | 2491 | unsigned long offset, type; |
8d69aaee HD |
2492 | unsigned char count; |
2493 | unsigned char has_cache; | |
253d553b | 2494 | int err = -EINVAL; |
1da177e4 | 2495 | |
a7420aa5 | 2496 | if (non_swap_entry(entry)) |
253d553b | 2497 | goto out; |
0697212a | 2498 | |
1da177e4 LT |
2499 | type = swp_type(entry); |
2500 | if (type >= nr_swapfiles) | |
2501 | goto bad_file; | |
efa90a98 | 2502 | p = swap_info[type]; |
1da177e4 LT |
2503 | offset = swp_offset(entry); |
2504 | ||
ec8acf20 | 2505 | spin_lock(&p->lock); |
355cfa73 KH |
2506 | if (unlikely(offset >= p->max)) |
2507 | goto unlock_out; | |
2508 | ||
253d553b HD |
2509 | count = p->swap_map[offset]; |
2510 | has_cache = count & SWAP_HAS_CACHE; | |
2511 | count &= ~SWAP_HAS_CACHE; | |
2512 | err = 0; | |
355cfa73 | 2513 | |
253d553b | 2514 | if (usage == SWAP_HAS_CACHE) { |
355cfa73 KH |
2515 | |
2516 | /* set SWAP_HAS_CACHE if there is no cache and entry is used */ | |
253d553b HD |
2517 | if (!has_cache && count) |
2518 | has_cache = SWAP_HAS_CACHE; | |
2519 | else if (has_cache) /* someone else added cache */ | |
2520 | err = -EEXIST; | |
2521 | else /* no users remaining */ | |
2522 | err = -ENOENT; | |
355cfa73 KH |
2523 | |
2524 | } else if (count || has_cache) { | |
253d553b | 2525 | |
570a335b HD |
2526 | if ((count & ~COUNT_CONTINUED) < SWAP_MAP_MAX) |
2527 | count += usage; | |
2528 | else if ((count & ~COUNT_CONTINUED) > SWAP_MAP_MAX) | |
253d553b | 2529 | err = -EINVAL; |
570a335b HD |
2530 | else if (swap_count_continued(p, offset, count)) |
2531 | count = COUNT_CONTINUED; | |
2532 | else | |
2533 | err = -ENOMEM; | |
355cfa73 | 2534 | } else |
253d553b HD |
2535 | err = -ENOENT; /* unused swap entry */ |
2536 | ||
2537 | p->swap_map[offset] = count | has_cache; | |
2538 | ||
355cfa73 | 2539 | unlock_out: |
ec8acf20 | 2540 | spin_unlock(&p->lock); |
1da177e4 | 2541 | out: |
253d553b | 2542 | return err; |
1da177e4 LT |
2543 | |
2544 | bad_file: | |
465c47fd | 2545 | pr_err("swap_dup: %s%08lx\n", Bad_file, entry.val); |
1da177e4 LT |
2546 | goto out; |
2547 | } | |
253d553b | 2548 | |
aaa46865 HD |
2549 | /* |
2550 | * Help swapoff by noting that swap entry belongs to shmem/tmpfs | |
2551 | * (in which case its reference count is never incremented). | |
2552 | */ | |
2553 | void swap_shmem_alloc(swp_entry_t entry) | |
2554 | { | |
2555 | __swap_duplicate(entry, SWAP_MAP_SHMEM); | |
2556 | } | |
2557 | ||
355cfa73 | 2558 | /* |
08259d58 HD |
2559 | * Increase reference count of swap entry by 1. |
2560 | * Returns 0 for success, or -ENOMEM if a swap_count_continuation is required | |
2561 | * but could not be atomically allocated. Returns 0, just as if it succeeded, | |
2562 | * if __swap_duplicate() fails for another reason (-EINVAL or -ENOENT), which | |
2563 | * might occur if a page table entry has got corrupted. | |
355cfa73 | 2564 | */ |
570a335b | 2565 | int swap_duplicate(swp_entry_t entry) |
355cfa73 | 2566 | { |
570a335b HD |
2567 | int err = 0; |
2568 | ||
2569 | while (!err && __swap_duplicate(entry, 1) == -ENOMEM) | |
2570 | err = add_swap_count_continuation(entry, GFP_ATOMIC); | |
2571 | return err; | |
355cfa73 | 2572 | } |
1da177e4 | 2573 | |
cb4b86ba | 2574 | /* |
355cfa73 KH |
2575 | * @entry: swap entry for which we allocate swap cache. |
2576 | * | |
73c34b6a | 2577 | * Called when allocating swap cache for existing swap entry, |
355cfa73 KH |
2578 | * This can return error codes. Returns 0 at success. |
2579 | * -EBUSY means there is a swap cache. | |
2580 | * Note: return code is different from swap_duplicate(). | |
cb4b86ba KH |
2581 | */ |
2582 | int swapcache_prepare(swp_entry_t entry) | |
2583 | { | |
253d553b | 2584 | return __swap_duplicate(entry, SWAP_HAS_CACHE); |
cb4b86ba KH |
2585 | } |
2586 | ||
f981c595 MG |
2587 | struct swap_info_struct *page_swap_info(struct page *page) |
2588 | { | |
2589 | swp_entry_t swap = { .val = page_private(page) }; | |
2590 | BUG_ON(!PageSwapCache(page)); | |
2591 | return swap_info[swp_type(swap)]; | |
2592 | } | |
2593 | ||
2594 | /* | |
2595 | * out-of-line __page_file_ methods to avoid include hell. | |
2596 | */ | |
2597 | struct address_space *__page_file_mapping(struct page *page) | |
2598 | { | |
2599 | VM_BUG_ON(!PageSwapCache(page)); | |
2600 | return page_swap_info(page)->swap_file->f_mapping; | |
2601 | } | |
2602 | EXPORT_SYMBOL_GPL(__page_file_mapping); | |
2603 | ||
2604 | pgoff_t __page_file_index(struct page *page) | |
2605 | { | |
2606 | swp_entry_t swap = { .val = page_private(page) }; | |
2607 | VM_BUG_ON(!PageSwapCache(page)); | |
2608 | return swp_offset(swap); | |
2609 | } | |
2610 | EXPORT_SYMBOL_GPL(__page_file_index); | |
2611 | ||
570a335b HD |
2612 | /* |
2613 | * add_swap_count_continuation - called when a swap count is duplicated | |
2614 | * beyond SWAP_MAP_MAX, it allocates a new page and links that to the entry's | |
2615 | * page of the original vmalloc'ed swap_map, to hold the continuation count | |
2616 | * (for that entry and for its neighbouring PAGE_SIZE swap entries). Called | |
2617 | * again when count is duplicated beyond SWAP_MAP_MAX * SWAP_CONT_MAX, etc. | |
2618 | * | |
2619 | * These continuation pages are seldom referenced: the common paths all work | |
2620 | * on the original swap_map, only referring to a continuation page when the | |
2621 | * low "digit" of a count is incremented or decremented through SWAP_MAP_MAX. | |
2622 | * | |
2623 | * add_swap_count_continuation(, GFP_ATOMIC) can be called while holding | |
2624 | * page table locks; if it fails, add_swap_count_continuation(, GFP_KERNEL) | |
2625 | * can be called after dropping locks. | |
2626 | */ | |
2627 | int add_swap_count_continuation(swp_entry_t entry, gfp_t gfp_mask) | |
2628 | { | |
2629 | struct swap_info_struct *si; | |
2630 | struct page *head; | |
2631 | struct page *page; | |
2632 | struct page *list_page; | |
2633 | pgoff_t offset; | |
2634 | unsigned char count; | |
2635 | ||
2636 | /* | |
2637 | * When debugging, it's easier to use __GFP_ZERO here; but it's better | |
2638 | * for latency not to zero a page while GFP_ATOMIC and holding locks. | |
2639 | */ | |
2640 | page = alloc_page(gfp_mask | __GFP_HIGHMEM); | |
2641 | ||
2642 | si = swap_info_get(entry); | |
2643 | if (!si) { | |
2644 | /* | |
2645 | * An acceptable race has occurred since the failing | |
2646 | * __swap_duplicate(): the swap entry has been freed, | |
2647 | * perhaps even the whole swap_map cleared for swapoff. | |
2648 | */ | |
2649 | goto outer; | |
2650 | } | |
2651 | ||
2652 | offset = swp_offset(entry); | |
2653 | count = si->swap_map[offset] & ~SWAP_HAS_CACHE; | |
2654 | ||
2655 | if ((count & ~COUNT_CONTINUED) != SWAP_MAP_MAX) { | |
2656 | /* | |
2657 | * The higher the swap count, the more likely it is that tasks | |
2658 | * will race to add swap count continuation: we need to avoid | |
2659 | * over-provisioning. | |
2660 | */ | |
2661 | goto out; | |
2662 | } | |
2663 | ||
2664 | if (!page) { | |
ec8acf20 | 2665 | spin_unlock(&si->lock); |
570a335b HD |
2666 | return -ENOMEM; |
2667 | } | |
2668 | ||
2669 | /* | |
2670 | * We are fortunate that although vmalloc_to_page uses pte_offset_map, | |
2671 | * no architecture is using highmem pages for kernel pagetables: so it | |
2672 | * will not corrupt the GFP_ATOMIC caller's atomic pagetable kmaps. | |
2673 | */ | |
2674 | head = vmalloc_to_page(si->swap_map + offset); | |
2675 | offset &= ~PAGE_MASK; | |
2676 | ||
2677 | /* | |
2678 | * Page allocation does not initialize the page's lru field, | |
2679 | * but it does always reset its private field. | |
2680 | */ | |
2681 | if (!page_private(head)) { | |
2682 | BUG_ON(count & COUNT_CONTINUED); | |
2683 | INIT_LIST_HEAD(&head->lru); | |
2684 | set_page_private(head, SWP_CONTINUED); | |
2685 | si->flags |= SWP_CONTINUED; | |
2686 | } | |
2687 | ||
2688 | list_for_each_entry(list_page, &head->lru, lru) { | |
2689 | unsigned char *map; | |
2690 | ||
2691 | /* | |
2692 | * If the previous map said no continuation, but we've found | |
2693 | * a continuation page, free our allocation and use this one. | |
2694 | */ | |
2695 | if (!(count & COUNT_CONTINUED)) | |
2696 | goto out; | |
2697 | ||
9b04c5fe | 2698 | map = kmap_atomic(list_page) + offset; |
570a335b | 2699 | count = *map; |
9b04c5fe | 2700 | kunmap_atomic(map); |
570a335b HD |
2701 | |
2702 | /* | |
2703 | * If this continuation count now has some space in it, | |
2704 | * free our allocation and use this one. | |
2705 | */ | |
2706 | if ((count & ~COUNT_CONTINUED) != SWAP_CONT_MAX) | |
2707 | goto out; | |
2708 | } | |
2709 | ||
2710 | list_add_tail(&page->lru, &head->lru); | |
2711 | page = NULL; /* now it's attached, don't free it */ | |
2712 | out: | |
ec8acf20 | 2713 | spin_unlock(&si->lock); |
570a335b HD |
2714 | outer: |
2715 | if (page) | |
2716 | __free_page(page); | |
2717 | return 0; | |
2718 | } | |
2719 | ||
2720 | /* | |
2721 | * swap_count_continued - when the original swap_map count is incremented | |
2722 | * from SWAP_MAP_MAX, check if there is already a continuation page to carry | |
2723 | * into, carry if so, or else fail until a new continuation page is allocated; | |
2724 | * when the original swap_map count is decremented from 0 with continuation, | |
2725 | * borrow from the continuation and report whether it still holds more. | |
2726 | * Called while __swap_duplicate() or swap_entry_free() holds swap_lock. | |
2727 | */ | |
2728 | static bool swap_count_continued(struct swap_info_struct *si, | |
2729 | pgoff_t offset, unsigned char count) | |
2730 | { | |
2731 | struct page *head; | |
2732 | struct page *page; | |
2733 | unsigned char *map; | |
2734 | ||
2735 | head = vmalloc_to_page(si->swap_map + offset); | |
2736 | if (page_private(head) != SWP_CONTINUED) { | |
2737 | BUG_ON(count & COUNT_CONTINUED); | |
2738 | return false; /* need to add count continuation */ | |
2739 | } | |
2740 | ||
2741 | offset &= ~PAGE_MASK; | |
2742 | page = list_entry(head->lru.next, struct page, lru); | |
9b04c5fe | 2743 | map = kmap_atomic(page) + offset; |
570a335b HD |
2744 | |
2745 | if (count == SWAP_MAP_MAX) /* initial increment from swap_map */ | |
2746 | goto init_map; /* jump over SWAP_CONT_MAX checks */ | |
2747 | ||
2748 | if (count == (SWAP_MAP_MAX | COUNT_CONTINUED)) { /* incrementing */ | |
2749 | /* | |
2750 | * Think of how you add 1 to 999 | |
2751 | */ | |
2752 | while (*map == (SWAP_CONT_MAX | COUNT_CONTINUED)) { | |
9b04c5fe | 2753 | kunmap_atomic(map); |
570a335b HD |
2754 | page = list_entry(page->lru.next, struct page, lru); |
2755 | BUG_ON(page == head); | |
9b04c5fe | 2756 | map = kmap_atomic(page) + offset; |
570a335b HD |
2757 | } |
2758 | if (*map == SWAP_CONT_MAX) { | |
9b04c5fe | 2759 | kunmap_atomic(map); |
570a335b HD |
2760 | page = list_entry(page->lru.next, struct page, lru); |
2761 | if (page == head) | |
2762 | return false; /* add count continuation */ | |
9b04c5fe | 2763 | map = kmap_atomic(page) + offset; |
570a335b HD |
2764 | init_map: *map = 0; /* we didn't zero the page */ |
2765 | } | |
2766 | *map += 1; | |
9b04c5fe | 2767 | kunmap_atomic(map); |
570a335b HD |
2768 | page = list_entry(page->lru.prev, struct page, lru); |
2769 | while (page != head) { | |
9b04c5fe | 2770 | map = kmap_atomic(page) + offset; |
570a335b | 2771 | *map = COUNT_CONTINUED; |
9b04c5fe | 2772 | kunmap_atomic(map); |
570a335b HD |
2773 | page = list_entry(page->lru.prev, struct page, lru); |
2774 | } | |
2775 | return true; /* incremented */ | |
2776 | ||
2777 | } else { /* decrementing */ | |
2778 | /* | |
2779 | * Think of how you subtract 1 from 1000 | |
2780 | */ | |
2781 | BUG_ON(count != COUNT_CONTINUED); | |
2782 | while (*map == COUNT_CONTINUED) { | |
9b04c5fe | 2783 | kunmap_atomic(map); |
570a335b HD |
2784 | page = list_entry(page->lru.next, struct page, lru); |
2785 | BUG_ON(page == head); | |
9b04c5fe | 2786 | map = kmap_atomic(page) + offset; |
570a335b HD |
2787 | } |
2788 | BUG_ON(*map == 0); | |
2789 | *map -= 1; | |
2790 | if (*map == 0) | |
2791 | count = 0; | |
9b04c5fe | 2792 | kunmap_atomic(map); |
570a335b HD |
2793 | page = list_entry(page->lru.prev, struct page, lru); |
2794 | while (page != head) { | |
9b04c5fe | 2795 | map = kmap_atomic(page) + offset; |
570a335b HD |
2796 | *map = SWAP_CONT_MAX | count; |
2797 | count = COUNT_CONTINUED; | |
9b04c5fe | 2798 | kunmap_atomic(map); |
570a335b HD |
2799 | page = list_entry(page->lru.prev, struct page, lru); |
2800 | } | |
2801 | return count == COUNT_CONTINUED; | |
2802 | } | |
2803 | } | |
2804 | ||
2805 | /* | |
2806 | * free_swap_count_continuations - swapoff free all the continuation pages | |
2807 | * appended to the swap_map, after swap_map is quiesced, before vfree'ing it. | |
2808 | */ | |
2809 | static void free_swap_count_continuations(struct swap_info_struct *si) | |
2810 | { | |
2811 | pgoff_t offset; | |
2812 | ||
2813 | for (offset = 0; offset < si->max; offset += PAGE_SIZE) { | |
2814 | struct page *head; | |
2815 | head = vmalloc_to_page(si->swap_map + offset); | |
2816 | if (page_private(head)) { | |
2817 | struct list_head *this, *next; | |
2818 | list_for_each_safe(this, next, &head->lru) { | |
2819 | struct page *page; | |
2820 | page = list_entry(this, struct page, lru); | |
2821 | list_del(this); | |
2822 | __free_page(page); | |
2823 | } | |
2824 | } | |
2825 | } | |
2826 | } |