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> | |
6fa3eb70 | 12 | #include <linux/export.h> |
1da177e4 LT |
13 | #include <linux/kernel_stat.h> |
14 | #include <linux/swap.h> | |
15 | #include <linux/vmalloc.h> | |
16 | #include <linux/pagemap.h> | |
17 | #include <linux/namei.h> | |
072441e2 | 18 | #include <linux/shmem_fs.h> |
1da177e4 | 19 | #include <linux/blkdev.h> |
20137a49 | 20 | #include <linux/random.h> |
1da177e4 LT |
21 | #include <linux/writeback.h> |
22 | #include <linux/proc_fs.h> | |
23 | #include <linux/seq_file.h> | |
24 | #include <linux/init.h> | |
5ad64688 | 25 | #include <linux/ksm.h> |
1da177e4 LT |
26 | #include <linux/rmap.h> |
27 | #include <linux/security.h> | |
28 | #include <linux/backing-dev.h> | |
fc0abb14 | 29 | #include <linux/mutex.h> |
c59ede7b | 30 | #include <linux/capability.h> |
1da177e4 | 31 | #include <linux/syscalls.h> |
8a9f3ccd | 32 | #include <linux/memcontrol.h> |
66d7dd51 | 33 | #include <linux/poll.h> |
72788c38 | 34 | #include <linux/oom.h> |
38b5faf4 DM |
35 | #include <linux/frontswap.h> |
36 | #include <linux/swapfile.h> | |
f981c595 | 37 | #include <linux/export.h> |
1da177e4 LT |
38 | |
39 | #include <asm/pgtable.h> | |
40 | #include <asm/tlbflush.h> | |
41 | #include <linux/swapops.h> | |
27a7faa0 | 42 | #include <linux/page_cgroup.h> |
1da177e4 | 43 | |
570a335b HD |
44 | static bool swap_count_continued(struct swap_info_struct *, pgoff_t, |
45 | unsigned char); | |
46 | static void free_swap_count_continuations(struct swap_info_struct *); | |
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 | ||
24b8ff7c CEB |
187 | static unsigned long scan_swap_map(struct swap_info_struct *si, |
188 | unsigned char usage) | |
1da177e4 | 189 | { |
ebebbbe9 | 190 | unsigned long offset; |
c60aa176 | 191 | unsigned long scan_base; |
7992fde7 | 192 | unsigned long last_in_cluster = 0; |
048c27fd | 193 | int latency_ration = LATENCY_LIMIT; |
7992fde7 | 194 | int found_free_cluster = 0; |
7dfad418 | 195 | |
886bb7e9 | 196 | /* |
7dfad418 HD |
197 | * We try to cluster swap pages by allocating them sequentially |
198 | * in swap. Once we've allocated SWAPFILE_CLUSTER pages this | |
199 | * way, however, we resort to first-free allocation, starting | |
200 | * a new cluster. This prevents us from scattering swap pages | |
201 | * all over the entire swap partition, so that we reduce | |
202 | * overall disk seek times between swap pages. -- sct | |
203 | * But we do now try to find an empty cluster. -Andrea | |
c60aa176 | 204 | * And we let swap pages go all over an SSD partition. Hugh |
7dfad418 HD |
205 | */ |
206 | ||
52b7efdb | 207 | si->flags += SWP_SCANNING; |
c60aa176 | 208 | scan_base = offset = si->cluster_next; |
ebebbbe9 HD |
209 | |
210 | if (unlikely(!si->cluster_nr--)) { | |
211 | if (si->pages - si->inuse_pages < SWAPFILE_CLUSTER) { | |
212 | si->cluster_nr = SWAPFILE_CLUSTER - 1; | |
213 | goto checks; | |
214 | } | |
7992fde7 HD |
215 | if (si->flags & SWP_DISCARDABLE) { |
216 | /* | |
217 | * Start range check on racing allocations, in case | |
218 | * they overlap the cluster we eventually decide on | |
219 | * (we scan without swap_lock to allow preemption). | |
220 | * It's hardly conceivable that cluster_nr could be | |
221 | * wrapped during our scan, but don't depend on it. | |
222 | */ | |
223 | if (si->lowest_alloc) | |
224 | goto checks; | |
225 | si->lowest_alloc = si->max; | |
226 | si->highest_alloc = 0; | |
227 | } | |
ec8acf20 | 228 | spin_unlock(&si->lock); |
7dfad418 | 229 | |
c60aa176 HD |
230 | /* |
231 | * If seek is expensive, start searching for new cluster from | |
232 | * start of partition, to minimize the span of allocated swap. | |
233 | * But if seek is cheap, search from our current position, so | |
234 | * that swap is allocated from all over the partition: if the | |
235 | * Flash Translation Layer only remaps within limited zones, | |
236 | * we don't want to wear out the first zone too quickly. | |
237 | */ | |
238 | if (!(si->flags & SWP_SOLIDSTATE)) | |
239 | scan_base = offset = si->lowest_bit; | |
7dfad418 HD |
240 | last_in_cluster = offset + SWAPFILE_CLUSTER - 1; |
241 | ||
242 | /* Locate the first empty (unaligned) cluster */ | |
243 | for (; last_in_cluster <= si->highest_bit; offset++) { | |
1da177e4 | 244 | if (si->swap_map[offset]) |
7dfad418 HD |
245 | last_in_cluster = offset + SWAPFILE_CLUSTER; |
246 | else if (offset == last_in_cluster) { | |
ec8acf20 | 247 | spin_lock(&si->lock); |
ebebbbe9 HD |
248 | offset -= SWAPFILE_CLUSTER - 1; |
249 | si->cluster_next = offset; | |
250 | si->cluster_nr = SWAPFILE_CLUSTER - 1; | |
7992fde7 | 251 | found_free_cluster = 1; |
ebebbbe9 | 252 | goto checks; |
1da177e4 | 253 | } |
048c27fd HD |
254 | if (unlikely(--latency_ration < 0)) { |
255 | cond_resched(); | |
256 | latency_ration = LATENCY_LIMIT; | |
257 | } | |
7dfad418 | 258 | } |
ebebbbe9 HD |
259 | |
260 | offset = si->lowest_bit; | |
c60aa176 HD |
261 | last_in_cluster = offset + SWAPFILE_CLUSTER - 1; |
262 | ||
263 | /* Locate the first empty (unaligned) cluster */ | |
264 | for (; last_in_cluster < scan_base; offset++) { | |
265 | if (si->swap_map[offset]) | |
266 | last_in_cluster = offset + SWAPFILE_CLUSTER; | |
267 | else if (offset == last_in_cluster) { | |
ec8acf20 | 268 | spin_lock(&si->lock); |
c60aa176 HD |
269 | offset -= SWAPFILE_CLUSTER - 1; |
270 | si->cluster_next = offset; | |
271 | si->cluster_nr = SWAPFILE_CLUSTER - 1; | |
272 | found_free_cluster = 1; | |
273 | goto checks; | |
274 | } | |
275 | if (unlikely(--latency_ration < 0)) { | |
276 | cond_resched(); | |
277 | latency_ration = LATENCY_LIMIT; | |
278 | } | |
279 | } | |
280 | ||
281 | offset = scan_base; | |
ec8acf20 | 282 | spin_lock(&si->lock); |
ebebbbe9 | 283 | si->cluster_nr = SWAPFILE_CLUSTER - 1; |
7992fde7 | 284 | si->lowest_alloc = 0; |
1da177e4 | 285 | } |
7dfad418 | 286 | |
ebebbbe9 HD |
287 | checks: |
288 | if (!(si->flags & SWP_WRITEOK)) | |
52b7efdb | 289 | goto no_page; |
7dfad418 HD |
290 | if (!si->highest_bit) |
291 | goto no_page; | |
ebebbbe9 | 292 | if (offset > si->highest_bit) |
c60aa176 | 293 | scan_base = offset = si->lowest_bit; |
c9e44410 | 294 | |
b73d7fce HD |
295 | /* reuse swap entry of cache-only swap if not busy. */ |
296 | if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) { | |
c9e44410 | 297 | int swap_was_freed; |
ec8acf20 | 298 | spin_unlock(&si->lock); |
c9e44410 | 299 | swap_was_freed = __try_to_reclaim_swap(si, offset); |
ec8acf20 | 300 | spin_lock(&si->lock); |
c9e44410 KH |
301 | /* entry was freed successfully, try to use this again */ |
302 | if (swap_was_freed) | |
303 | goto checks; | |
304 | goto scan; /* check next one */ | |
305 | } | |
306 | ||
ebebbbe9 HD |
307 | if (si->swap_map[offset]) |
308 | goto scan; | |
309 | ||
310 | if (offset == si->lowest_bit) | |
311 | si->lowest_bit++; | |
312 | if (offset == si->highest_bit) | |
313 | si->highest_bit--; | |
314 | si->inuse_pages++; | |
315 | if (si->inuse_pages == si->pages) { | |
316 | si->lowest_bit = si->max; | |
317 | si->highest_bit = 0; | |
1da177e4 | 318 | } |
253d553b | 319 | si->swap_map[offset] = usage; |
ebebbbe9 HD |
320 | si->cluster_next = offset + 1; |
321 | si->flags -= SWP_SCANNING; | |
7992fde7 HD |
322 | |
323 | if (si->lowest_alloc) { | |
324 | /* | |
325 | * Only set when SWP_DISCARDABLE, and there's a scan | |
326 | * for a free cluster in progress or just completed. | |
327 | */ | |
328 | if (found_free_cluster) { | |
329 | /* | |
330 | * To optimize wear-levelling, discard the | |
331 | * old data of the cluster, taking care not to | |
332 | * discard any of its pages that have already | |
333 | * been allocated by racing tasks (offset has | |
334 | * already stepped over any at the beginning). | |
335 | */ | |
336 | if (offset < si->highest_alloc && | |
337 | si->lowest_alloc <= last_in_cluster) | |
338 | last_in_cluster = si->lowest_alloc - 1; | |
339 | si->flags |= SWP_DISCARDING; | |
ec8acf20 | 340 | spin_unlock(&si->lock); |
7992fde7 HD |
341 | |
342 | if (offset < last_in_cluster) | |
343 | discard_swap_cluster(si, offset, | |
344 | last_in_cluster - offset + 1); | |
345 | ||
ec8acf20 | 346 | spin_lock(&si->lock); |
7992fde7 HD |
347 | si->lowest_alloc = 0; |
348 | si->flags &= ~SWP_DISCARDING; | |
349 | ||
350 | smp_mb(); /* wake_up_bit advises this */ | |
351 | wake_up_bit(&si->flags, ilog2(SWP_DISCARDING)); | |
352 | ||
353 | } else if (si->flags & SWP_DISCARDING) { | |
354 | /* | |
355 | * Delay using pages allocated by racing tasks | |
356 | * until the whole discard has been issued. We | |
357 | * could defer that delay until swap_writepage, | |
358 | * but it's easier to keep this self-contained. | |
359 | */ | |
ec8acf20 | 360 | spin_unlock(&si->lock); |
7992fde7 HD |
361 | wait_on_bit(&si->flags, ilog2(SWP_DISCARDING), |
362 | wait_for_discard, TASK_UNINTERRUPTIBLE); | |
ec8acf20 | 363 | spin_lock(&si->lock); |
7992fde7 HD |
364 | } else { |
365 | /* | |
366 | * Note pages allocated by racing tasks while | |
367 | * scan for a free cluster is in progress, so | |
368 | * that its final discard can exclude them. | |
369 | */ | |
370 | if (offset < si->lowest_alloc) | |
371 | si->lowest_alloc = offset; | |
372 | if (offset > si->highest_alloc) | |
373 | si->highest_alloc = offset; | |
374 | } | |
375 | } | |
ebebbbe9 | 376 | return offset; |
7dfad418 | 377 | |
ebebbbe9 | 378 | scan: |
ec8acf20 | 379 | spin_unlock(&si->lock); |
7dfad418 | 380 | while (++offset <= si->highest_bit) { |
52b7efdb | 381 | if (!si->swap_map[offset]) { |
ec8acf20 | 382 | spin_lock(&si->lock); |
52b7efdb HD |
383 | goto checks; |
384 | } | |
c9e44410 | 385 | if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) { |
ec8acf20 | 386 | spin_lock(&si->lock); |
c9e44410 KH |
387 | goto checks; |
388 | } | |
048c27fd HD |
389 | if (unlikely(--latency_ration < 0)) { |
390 | cond_resched(); | |
391 | latency_ration = LATENCY_LIMIT; | |
392 | } | |
7dfad418 | 393 | } |
c60aa176 HD |
394 | offset = si->lowest_bit; |
395 | while (++offset < scan_base) { | |
396 | if (!si->swap_map[offset]) { | |
ec8acf20 | 397 | spin_lock(&si->lock); |
c60aa176 HD |
398 | goto checks; |
399 | } | |
c9e44410 | 400 | if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) { |
ec8acf20 | 401 | spin_lock(&si->lock); |
c9e44410 KH |
402 | goto checks; |
403 | } | |
c60aa176 HD |
404 | if (unlikely(--latency_ration < 0)) { |
405 | cond_resched(); | |
406 | latency_ration = LATENCY_LIMIT; | |
407 | } | |
408 | } | |
ec8acf20 | 409 | spin_lock(&si->lock); |
7dfad418 HD |
410 | |
411 | no_page: | |
52b7efdb | 412 | si->flags -= SWP_SCANNING; |
1da177e4 LT |
413 | return 0; |
414 | } | |
415 | ||
416 | swp_entry_t get_swap_page(void) | |
417 | { | |
fb4f88dc HD |
418 | struct swap_info_struct *si; |
419 | pgoff_t offset; | |
420 | int type, next; | |
421 | int wrapped = 0; | |
ec8acf20 | 422 | int hp_index; |
1da177e4 | 423 | |
5d337b91 | 424 | spin_lock(&swap_lock); |
ec8acf20 | 425 | if (atomic_long_read(&nr_swap_pages) <= 0) |
fb4f88dc | 426 | goto noswap; |
ec8acf20 | 427 | atomic_long_dec(&nr_swap_pages); |
fb4f88dc HD |
428 | |
429 | for (type = swap_list.next; type >= 0 && wrapped < 2; type = next) { | |
ec8acf20 SL |
430 | hp_index = atomic_xchg(&highest_priority_index, -1); |
431 | /* | |
432 | * highest_priority_index records current highest priority swap | |
433 | * type which just frees swap entries. If its priority is | |
434 | * higher than that of swap_list.next swap type, we use it. It | |
435 | * isn't protected by swap_lock, so it can be an invalid value | |
436 | * if the corresponding swap type is swapoff. We double check | |
437 | * the flags here. It's even possible the swap type is swapoff | |
438 | * and swapon again and its priority is changed. In such rare | |
439 | * case, low prority swap type might be used, but eventually | |
440 | * high priority swap will be used after several rounds of | |
441 | * swap. | |
442 | */ | |
443 | if (hp_index != -1 && hp_index != type && | |
444 | swap_info[type]->prio < swap_info[hp_index]->prio && | |
445 | (swap_info[hp_index]->flags & SWP_WRITEOK)) { | |
446 | type = hp_index; | |
447 | swap_list.next = type; | |
448 | } | |
449 | ||
efa90a98 | 450 | si = swap_info[type]; |
fb4f88dc HD |
451 | next = si->next; |
452 | if (next < 0 || | |
efa90a98 | 453 | (!wrapped && si->prio != swap_info[next]->prio)) { |
fb4f88dc HD |
454 | next = swap_list.head; |
455 | wrapped++; | |
1da177e4 | 456 | } |
fb4f88dc | 457 | |
ec8acf20 SL |
458 | spin_lock(&si->lock); |
459 | if (!si->highest_bit) { | |
460 | spin_unlock(&si->lock); | |
fb4f88dc | 461 | continue; |
ec8acf20 SL |
462 | } |
463 | if (!(si->flags & SWP_WRITEOK)) { | |
464 | spin_unlock(&si->lock); | |
fb4f88dc | 465 | continue; |
ec8acf20 | 466 | } |
fb4f88dc HD |
467 | |
468 | swap_list.next = next; | |
ec8acf20 SL |
469 | |
470 | spin_unlock(&swap_lock); | |
355cfa73 | 471 | /* This is called for allocating swap entry for cache */ |
253d553b | 472 | offset = scan_swap_map(si, SWAP_HAS_CACHE); |
ec8acf20 SL |
473 | spin_unlock(&si->lock); |
474 | if (offset) | |
fb4f88dc | 475 | return swp_entry(type, offset); |
ec8acf20 | 476 | spin_lock(&swap_lock); |
fb4f88dc | 477 | next = swap_list.next; |
1da177e4 | 478 | } |
fb4f88dc | 479 | |
ec8acf20 | 480 | atomic_long_inc(&nr_swap_pages); |
fb4f88dc | 481 | noswap: |
5d337b91 | 482 | spin_unlock(&swap_lock); |
fb4f88dc | 483 | return (swp_entry_t) {0}; |
1da177e4 LT |
484 | } |
485 | ||
910321ea HD |
486 | /* The only caller of this function is now susupend routine */ |
487 | swp_entry_t get_swap_page_of_type(int type) | |
488 | { | |
489 | struct swap_info_struct *si; | |
490 | pgoff_t offset; | |
491 | ||
910321ea | 492 | si = swap_info[type]; |
ec8acf20 | 493 | spin_lock(&si->lock); |
910321ea | 494 | if (si && (si->flags & SWP_WRITEOK)) { |
ec8acf20 | 495 | atomic_long_dec(&nr_swap_pages); |
910321ea HD |
496 | /* This is called for allocating swap entry, not cache */ |
497 | offset = scan_swap_map(si, 1); | |
498 | if (offset) { | |
ec8acf20 | 499 | spin_unlock(&si->lock); |
910321ea HD |
500 | return swp_entry(type, offset); |
501 | } | |
ec8acf20 | 502 | atomic_long_inc(&nr_swap_pages); |
910321ea | 503 | } |
ec8acf20 | 504 | spin_unlock(&si->lock); |
910321ea HD |
505 | return (swp_entry_t) {0}; |
506 | } | |
6fa3eb70 | 507 | EXPORT_SYMBOL_GPL(get_swap_page_of_type); |
910321ea | 508 | |
6fa3eb70 S |
509 | #ifdef CONFIG_MEMCG |
510 | swp_entry_t get_swap_page_by_memcg(struct page *page) | |
511 | { | |
512 | struct swap_info_struct *si; | |
513 | int type, next, wrapped; | |
514 | pgoff_t offset; | |
515 | ||
516 | /* Go to original get_swap_page if we have only 1 or no swap area. */ | |
517 | if (nr_swapfiles <= 1) | |
518 | return get_swap_page(); | |
519 | ||
520 | spin_lock(&swap_lock); | |
521 | if (atomic_long_read(&nr_swap_pages) <= 0) | |
522 | goto noswap; | |
523 | atomic_long_dec(&nr_swap_pages); | |
524 | ||
525 | /* In which memcg */ | |
526 | if (memcg_is_root(page)) { | |
527 | /* High to low priority */ | |
528 | wrapped = 1; | |
529 | type = swap_list.head; | |
530 | } else { | |
531 | /* Low to high priority */ | |
532 | wrapped = 0; | |
533 | si = swap_info[swap_list.head]; | |
534 | type = si->next; | |
535 | } | |
536 | ||
537 | /* Scan for an empty swap entry */ | |
538 | for (; (type >= 0) || (wrapped == 0); type = next) { | |
539 | if (type < 0) { | |
540 | wrapped++; | |
541 | next = swap_list.head; | |
542 | continue; | |
543 | } | |
544 | ||
545 | si = swap_info[type]; | |
546 | next = si->next; | |
547 | spin_lock(&si->lock); | |
548 | if (!si->highest_bit) { | |
549 | spin_unlock(&si->lock); | |
550 | continue; | |
551 | } | |
552 | if (!(si->flags & SWP_WRITEOK)) { | |
553 | spin_unlock(&si->lock); | |
554 | continue; | |
555 | } | |
556 | spin_unlock(&swap_lock); | |
557 | /* This is called for allocating swap entry for cache */ | |
558 | offset = scan_swap_map(si, SWAP_HAS_CACHE); | |
559 | spin_unlock(&si->lock); | |
560 | if (offset) | |
561 | return swp_entry(type, offset); | |
562 | spin_lock(&swap_lock); | |
563 | } | |
564 | ||
565 | atomic_long_inc(&nr_swap_pages); | |
566 | noswap: | |
567 | spin_unlock(&swap_lock); | |
568 | return (swp_entry_t) {0}; | |
569 | } | |
570 | #endif | |
571 | ||
572 | static unsigned int find_next_to_unuse(struct swap_info_struct *si, | |
573 | unsigned int prev, bool frontswap); | |
574 | ||
575 | void get_swap_range_of_type(int type, swp_entry_t *start, swp_entry_t *end, | |
576 | unsigned int limit) | |
577 | { | |
578 | struct swap_info_struct *si; | |
579 | pgoff_t start_at; | |
580 | unsigned int i; | |
581 | ||
582 | *start = swp_entry(0, 0); | |
583 | *end = swp_entry(0, 0); | |
584 | si = swap_info[type]; | |
585 | spin_lock(&si->lock); | |
586 | if (si && (si->flags & SWP_WRITEOK)) { | |
587 | /* This is called for allocating swap entry, not cache */ | |
588 | start_at = scan_swap_map(si, 1); | |
589 | if (start_at) { | |
590 | unsigned long stop_at = find_next_to_unuse(si, start_at, 0); | |
591 | if (stop_at > start_at) | |
592 | stop_at--; | |
593 | else | |
594 | stop_at = si->max - 1; | |
595 | if (stop_at - start_at + 1 > limit) | |
596 | stop_at = min_t(unsigned int, | |
597 | start_at + limit - 1, | |
598 | si->max - 1); | |
599 | /* Mark them used */ | |
600 | for (i = start_at; i <= stop_at; i++) | |
601 | si->swap_map[i] = 1; | |
602 | /* first page already done above */ | |
603 | si->inuse_pages += stop_at - start_at; | |
604 | ||
605 | atomic_long_sub(stop_at - start_at + 1, &nr_swap_pages); | |
606 | if (start_at + 1 == si->lowest_bit) | |
607 | si->lowest_bit = stop_at + 1; | |
608 | if (si->inuse_pages == si->pages) { | |
609 | si->lowest_bit = si->max; | |
610 | si->highest_bit = 0; | |
611 | } | |
612 | si->cluster_next = stop_at + 1; | |
613 | *start = swp_entry(type, start_at); | |
614 | *end = swp_entry(type, stop_at); | |
615 | } | |
616 | } | |
617 | spin_unlock(&si->lock); | |
618 | } | |
619 | EXPORT_SYMBOL_GPL(get_swap_range_of_type); | |
620 | ||
621 | struct swap_info_struct *swap_info_get(swp_entry_t entry) | |
1da177e4 | 622 | { |
73c34b6a | 623 | struct swap_info_struct *p; |
1da177e4 LT |
624 | unsigned long offset, type; |
625 | ||
626 | if (!entry.val) | |
627 | goto out; | |
628 | type = swp_type(entry); | |
629 | if (type >= nr_swapfiles) | |
630 | goto bad_nofile; | |
efa90a98 | 631 | p = swap_info[type]; |
1da177e4 LT |
632 | if (!(p->flags & SWP_USED)) |
633 | goto bad_device; | |
634 | offset = swp_offset(entry); | |
635 | if (offset >= p->max) | |
636 | goto bad_offset; | |
637 | if (!p->swap_map[offset]) | |
638 | goto bad_free; | |
ec8acf20 | 639 | spin_lock(&p->lock); |
1da177e4 LT |
640 | return p; |
641 | ||
642 | bad_free: | |
643 | printk(KERN_ERR "swap_free: %s%08lx\n", Unused_offset, entry.val); | |
644 | goto out; | |
645 | bad_offset: | |
646 | printk(KERN_ERR "swap_free: %s%08lx\n", Bad_offset, entry.val); | |
647 | goto out; | |
648 | bad_device: | |
649 | printk(KERN_ERR "swap_free: %s%08lx\n", Unused_file, entry.val); | |
650 | goto out; | |
651 | bad_nofile: | |
652 | printk(KERN_ERR "swap_free: %s%08lx\n", Bad_file, entry.val); | |
653 | out: | |
654 | return NULL; | |
886bb7e9 | 655 | } |
1da177e4 | 656 | |
6fa3eb70 S |
657 | void swap_info_unlock(struct swap_info_struct *si) |
658 | { | |
659 | spin_unlock(&si->lock); | |
660 | } | |
661 | ||
ec8acf20 SL |
662 | /* |
663 | * This swap type frees swap entry, check if it is the highest priority swap | |
664 | * type which just frees swap entry. get_swap_page() uses | |
665 | * highest_priority_index to search highest priority swap type. The | |
666 | * swap_info_struct.lock can't protect us if there are multiple swap types | |
667 | * active, so we use atomic_cmpxchg. | |
668 | */ | |
669 | static void set_highest_priority_index(int type) | |
670 | { | |
671 | int old_hp_index, new_hp_index; | |
672 | ||
673 | do { | |
674 | old_hp_index = atomic_read(&highest_priority_index); | |
675 | if (old_hp_index != -1 && | |
676 | swap_info[old_hp_index]->prio >= swap_info[type]->prio) | |
677 | break; | |
678 | new_hp_index = type; | |
679 | } while (atomic_cmpxchg(&highest_priority_index, | |
680 | old_hp_index, new_hp_index) != old_hp_index); | |
681 | } | |
682 | ||
8d69aaee HD |
683 | static unsigned char swap_entry_free(struct swap_info_struct *p, |
684 | swp_entry_t entry, unsigned char usage) | |
1da177e4 | 685 | { |
253d553b | 686 | unsigned long offset = swp_offset(entry); |
8d69aaee HD |
687 | unsigned char count; |
688 | unsigned char has_cache; | |
355cfa73 | 689 | |
253d553b HD |
690 | count = p->swap_map[offset]; |
691 | has_cache = count & SWAP_HAS_CACHE; | |
692 | count &= ~SWAP_HAS_CACHE; | |
355cfa73 | 693 | |
253d553b | 694 | if (usage == SWAP_HAS_CACHE) { |
355cfa73 | 695 | VM_BUG_ON(!has_cache); |
253d553b | 696 | has_cache = 0; |
aaa46865 HD |
697 | } else if (count == SWAP_MAP_SHMEM) { |
698 | /* | |
699 | * Or we could insist on shmem.c using a special | |
700 | * swap_shmem_free() and free_shmem_swap_and_cache()... | |
701 | */ | |
702 | count = 0; | |
570a335b HD |
703 | } else if ((count & ~COUNT_CONTINUED) <= SWAP_MAP_MAX) { |
704 | if (count == COUNT_CONTINUED) { | |
705 | if (swap_count_continued(p, offset, count)) | |
706 | count = SWAP_MAP_MAX | COUNT_CONTINUED; | |
707 | else | |
708 | count = SWAP_MAP_MAX; | |
709 | } else | |
710 | count--; | |
711 | } | |
253d553b HD |
712 | |
713 | if (!count) | |
714 | mem_cgroup_uncharge_swap(entry); | |
715 | ||
716 | usage = count | has_cache; | |
717 | p->swap_map[offset] = usage; | |
355cfa73 | 718 | |
355cfa73 | 719 | /* free if no reference */ |
253d553b | 720 | if (!usage) { |
355cfa73 KH |
721 | if (offset < p->lowest_bit) |
722 | p->lowest_bit = offset; | |
723 | if (offset > p->highest_bit) | |
724 | p->highest_bit = offset; | |
ec8acf20 SL |
725 | set_highest_priority_index(p->type); |
726 | atomic_long_inc(&nr_swap_pages); | |
355cfa73 | 727 | p->inuse_pages--; |
38b5faf4 | 728 | frontswap_invalidate_page(p->type, offset); |
73744923 MG |
729 | if (p->flags & SWP_BLKDEV) { |
730 | struct gendisk *disk = p->bdev->bd_disk; | |
731 | if (disk->fops->swap_slot_free_notify) | |
732 | disk->fops->swap_slot_free_notify(p->bdev, | |
733 | offset); | |
734 | } | |
1da177e4 | 735 | } |
253d553b HD |
736 | |
737 | return usage; | |
1da177e4 LT |
738 | } |
739 | ||
740 | /* | |
741 | * Caller has made sure that the swapdevice corresponding to entry | |
742 | * is still around or has not been recycled. | |
743 | */ | |
744 | void swap_free(swp_entry_t entry) | |
745 | { | |
73c34b6a | 746 | struct swap_info_struct *p; |
1da177e4 LT |
747 | |
748 | p = swap_info_get(entry); | |
749 | if (p) { | |
253d553b | 750 | swap_entry_free(p, entry, 1); |
ec8acf20 | 751 | spin_unlock(&p->lock); |
1da177e4 LT |
752 | } |
753 | } | |
754 | ||
cb4b86ba KH |
755 | /* |
756 | * Called after dropping swapcache to decrease refcnt to swap entries. | |
757 | */ | |
758 | void swapcache_free(swp_entry_t entry, struct page *page) | |
759 | { | |
355cfa73 | 760 | struct swap_info_struct *p; |
8d69aaee | 761 | unsigned char count; |
355cfa73 | 762 | |
355cfa73 KH |
763 | p = swap_info_get(entry); |
764 | if (p) { | |
253d553b HD |
765 | count = swap_entry_free(p, entry, SWAP_HAS_CACHE); |
766 | if (page) | |
767 | mem_cgroup_uncharge_swapcache(page, entry, count != 0); | |
ec8acf20 | 768 | spin_unlock(&p->lock); |
355cfa73 | 769 | } |
cb4b86ba | 770 | } |
6fa3eb70 | 771 | EXPORT_SYMBOL_GPL(swap_free); |
cb4b86ba | 772 | |
1da177e4 | 773 | /* |
c475a8ab | 774 | * How many references to page are currently swapped out? |
570a335b HD |
775 | * This does not give an exact answer when swap count is continued, |
776 | * but does include the high COUNT_CONTINUED flag to allow for that. | |
1da177e4 | 777 | */ |
bde05d1c | 778 | int page_swapcount(struct page *page) |
1da177e4 | 779 | { |
c475a8ab HD |
780 | int count = 0; |
781 | struct swap_info_struct *p; | |
1da177e4 LT |
782 | swp_entry_t entry; |
783 | ||
4c21e2f2 | 784 | entry.val = page_private(page); |
1da177e4 LT |
785 | p = swap_info_get(entry); |
786 | if (p) { | |
355cfa73 | 787 | count = swap_count(p->swap_map[swp_offset(entry)]); |
ec8acf20 | 788 | spin_unlock(&p->lock); |
1da177e4 | 789 | } |
c475a8ab | 790 | return count; |
1da177e4 LT |
791 | } |
792 | ||
793 | /* | |
7b1fe597 HD |
794 | * We can write to an anon page without COW if there are no other references |
795 | * to it. And as a side-effect, free up its swap: because the old content | |
796 | * on disk will never be read, and seeking back there to write new content | |
797 | * later would only waste time away from clustering. | |
1da177e4 | 798 | */ |
7b1fe597 | 799 | int reuse_swap_page(struct page *page) |
1da177e4 | 800 | { |
c475a8ab HD |
801 | int count; |
802 | ||
51726b12 | 803 | VM_BUG_ON(!PageLocked(page)); |
5ad64688 HD |
804 | if (unlikely(PageKsm(page))) |
805 | return 0; | |
c475a8ab | 806 | count = page_mapcount(page); |
7b1fe597 | 807 | if (count <= 1 && PageSwapCache(page)) { |
c475a8ab | 808 | count += page_swapcount(page); |
7b1fe597 HD |
809 | if (count == 1 && !PageWriteback(page)) { |
810 | delete_from_swap_cache(page); | |
811 | SetPageDirty(page); | |
812 | } | |
813 | } | |
5ad64688 | 814 | return count <= 1; |
1da177e4 LT |
815 | } |
816 | ||
817 | /* | |
a2c43eed HD |
818 | * If swap is getting full, or if there are no more mappings of this page, |
819 | * then try_to_free_swap is called to free its swap space. | |
1da177e4 | 820 | */ |
a2c43eed | 821 | int try_to_free_swap(struct page *page) |
1da177e4 | 822 | { |
51726b12 | 823 | VM_BUG_ON(!PageLocked(page)); |
1da177e4 LT |
824 | |
825 | if (!PageSwapCache(page)) | |
826 | return 0; | |
827 | if (PageWriteback(page)) | |
828 | return 0; | |
a2c43eed | 829 | if (page_swapcount(page)) |
1da177e4 LT |
830 | return 0; |
831 | ||
b73d7fce HD |
832 | /* |
833 | * Once hibernation has begun to create its image of memory, | |
834 | * there's a danger that one of the calls to try_to_free_swap() | |
835 | * - most probably a call from __try_to_reclaim_swap() while | |
836 | * hibernation is allocating its own swap pages for the image, | |
837 | * but conceivably even a call from memory reclaim - will free | |
838 | * the swap from a page which has already been recorded in the | |
839 | * image as a clean swapcache page, and then reuse its swap for | |
840 | * another page of the image. On waking from hibernation, the | |
841 | * original page might be freed under memory pressure, then | |
842 | * later read back in from swap, now with the wrong data. | |
843 | * | |
f90ac398 MG |
844 | * Hibration suspends storage while it is writing the image |
845 | * to disk so check that here. | |
b73d7fce | 846 | */ |
f90ac398 | 847 | if (pm_suspended_storage()) |
b73d7fce HD |
848 | return 0; |
849 | ||
a2c43eed HD |
850 | delete_from_swap_cache(page); |
851 | SetPageDirty(page); | |
852 | return 1; | |
68a22394 RR |
853 | } |
854 | ||
1da177e4 LT |
855 | /* |
856 | * Free the swap entry like above, but also try to | |
857 | * free the page cache entry if it is the last user. | |
858 | */ | |
2509ef26 | 859 | int free_swap_and_cache(swp_entry_t entry) |
1da177e4 | 860 | { |
2509ef26 | 861 | struct swap_info_struct *p; |
1da177e4 LT |
862 | struct page *page = NULL; |
863 | ||
a7420aa5 | 864 | if (non_swap_entry(entry)) |
2509ef26 | 865 | return 1; |
0697212a | 866 | |
1da177e4 LT |
867 | p = swap_info_get(entry); |
868 | if (p) { | |
253d553b | 869 | if (swap_entry_free(p, entry, 1) == SWAP_HAS_CACHE) { |
33806f06 SL |
870 | page = find_get_page(swap_address_space(entry), |
871 | entry.val); | |
8413ac9d | 872 | if (page && !trylock_page(page)) { |
93fac704 NP |
873 | page_cache_release(page); |
874 | page = NULL; | |
875 | } | |
876 | } | |
ec8acf20 | 877 | spin_unlock(&p->lock); |
1da177e4 LT |
878 | } |
879 | if (page) { | |
a2c43eed HD |
880 | /* |
881 | * Not mapped elsewhere, or swap space full? Free it! | |
882 | * Also recheck PageSwapCache now page is locked (above). | |
883 | */ | |
93fac704 | 884 | if (PageSwapCache(page) && !PageWriteback(page) && |
a2c43eed | 885 | (!page_mapped(page) || vm_swap_full())) { |
1da177e4 LT |
886 | delete_from_swap_cache(page); |
887 | SetPageDirty(page); | |
888 | } | |
889 | unlock_page(page); | |
890 | page_cache_release(page); | |
891 | } | |
2509ef26 | 892 | return p != NULL; |
1da177e4 LT |
893 | } |
894 | ||
b0cb1a19 | 895 | #ifdef CONFIG_HIBERNATION |
f577eb30 | 896 | /* |
915bae9e | 897 | * Find the swap type that corresponds to given device (if any). |
f577eb30 | 898 | * |
915bae9e RW |
899 | * @offset - number of the PAGE_SIZE-sized block of the device, starting |
900 | * from 0, in which the swap header is expected to be located. | |
901 | * | |
902 | * This is needed for the suspend to disk (aka swsusp). | |
f577eb30 | 903 | */ |
7bf23687 | 904 | int swap_type_of(dev_t device, sector_t offset, struct block_device **bdev_p) |
f577eb30 | 905 | { |
915bae9e | 906 | struct block_device *bdev = NULL; |
efa90a98 | 907 | int type; |
f577eb30 | 908 | |
915bae9e RW |
909 | if (device) |
910 | bdev = bdget(device); | |
911 | ||
f577eb30 | 912 | spin_lock(&swap_lock); |
efa90a98 HD |
913 | for (type = 0; type < nr_swapfiles; type++) { |
914 | struct swap_info_struct *sis = swap_info[type]; | |
f577eb30 | 915 | |
915bae9e | 916 | if (!(sis->flags & SWP_WRITEOK)) |
f577eb30 | 917 | continue; |
b6b5bce3 | 918 | |
915bae9e | 919 | if (!bdev) { |
7bf23687 | 920 | if (bdev_p) |
dddac6a7 | 921 | *bdev_p = bdgrab(sis->bdev); |
7bf23687 | 922 | |
6e1819d6 | 923 | spin_unlock(&swap_lock); |
efa90a98 | 924 | return type; |
6e1819d6 | 925 | } |
915bae9e | 926 | if (bdev == sis->bdev) { |
9625a5f2 | 927 | struct swap_extent *se = &sis->first_swap_extent; |
915bae9e | 928 | |
915bae9e | 929 | if (se->start_block == offset) { |
7bf23687 | 930 | if (bdev_p) |
dddac6a7 | 931 | *bdev_p = bdgrab(sis->bdev); |
7bf23687 | 932 | |
915bae9e RW |
933 | spin_unlock(&swap_lock); |
934 | bdput(bdev); | |
efa90a98 | 935 | return type; |
915bae9e | 936 | } |
f577eb30 RW |
937 | } |
938 | } | |
939 | spin_unlock(&swap_lock); | |
915bae9e RW |
940 | if (bdev) |
941 | bdput(bdev); | |
942 | ||
f577eb30 RW |
943 | return -ENODEV; |
944 | } | |
945 | ||
73c34b6a HD |
946 | /* |
947 | * Get the (PAGE_SIZE) block corresponding to given offset on the swapdev | |
948 | * corresponding to given index in swap_info (swap type). | |
949 | */ | |
950 | sector_t swapdev_block(int type, pgoff_t offset) | |
951 | { | |
952 | struct block_device *bdev; | |
953 | ||
954 | if ((unsigned int)type >= nr_swapfiles) | |
955 | return 0; | |
956 | if (!(swap_info[type]->flags & SWP_WRITEOK)) | |
957 | return 0; | |
d4906e1a | 958 | return map_swap_entry(swp_entry(type, offset), &bdev); |
73c34b6a HD |
959 | } |
960 | ||
f577eb30 RW |
961 | /* |
962 | * Return either the total number of swap pages of given type, or the number | |
963 | * of free pages of that type (depending on @free) | |
964 | * | |
965 | * This is needed for software suspend | |
966 | */ | |
967 | unsigned int count_swap_pages(int type, int free) | |
968 | { | |
969 | unsigned int n = 0; | |
970 | ||
efa90a98 HD |
971 | spin_lock(&swap_lock); |
972 | if ((unsigned int)type < nr_swapfiles) { | |
973 | struct swap_info_struct *sis = swap_info[type]; | |
974 | ||
ec8acf20 | 975 | spin_lock(&sis->lock); |
efa90a98 HD |
976 | if (sis->flags & SWP_WRITEOK) { |
977 | n = sis->pages; | |
f577eb30 | 978 | if (free) |
efa90a98 | 979 | n -= sis->inuse_pages; |
f577eb30 | 980 | } |
ec8acf20 | 981 | spin_unlock(&sis->lock); |
f577eb30 | 982 | } |
efa90a98 | 983 | spin_unlock(&swap_lock); |
f577eb30 RW |
984 | return n; |
985 | } | |
73c34b6a | 986 | #endif /* CONFIG_HIBERNATION */ |
f577eb30 | 987 | |
1da177e4 | 988 | /* |
72866f6f HD |
989 | * No need to decide whether this PTE shares the swap entry with others, |
990 | * just let do_wp_page work it out if a write is requested later - to | |
991 | * force COW, vm_page_prot omits write permission from any private vma. | |
1da177e4 | 992 | */ |
044d66c1 | 993 | static int unuse_pte(struct vm_area_struct *vma, pmd_t *pmd, |
1da177e4 LT |
994 | unsigned long addr, swp_entry_t entry, struct page *page) |
995 | { | |
9e16b7fb | 996 | struct page *swapcache; |
72835c86 | 997 | struct mem_cgroup *memcg; |
044d66c1 HD |
998 | spinlock_t *ptl; |
999 | pte_t *pte; | |
1000 | int ret = 1; | |
1001 | ||
9e16b7fb HD |
1002 | swapcache = page; |
1003 | page = ksm_might_need_to_copy(page, vma, addr); | |
1004 | if (unlikely(!page)) | |
1005 | return -ENOMEM; | |
1006 | ||
72835c86 JW |
1007 | if (mem_cgroup_try_charge_swapin(vma->vm_mm, page, |
1008 | GFP_KERNEL, &memcg)) { | |
044d66c1 | 1009 | ret = -ENOMEM; |
85d9fc89 KH |
1010 | goto out_nolock; |
1011 | } | |
044d66c1 HD |
1012 | |
1013 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); | |
1014 | if (unlikely(!pte_same(*pte, swp_entry_to_pte(entry)))) { | |
5d84c776 | 1015 | mem_cgroup_cancel_charge_swapin(memcg); |
044d66c1 HD |
1016 | ret = 0; |
1017 | goto out; | |
1018 | } | |
8a9f3ccd | 1019 | |
b084d435 | 1020 | dec_mm_counter(vma->vm_mm, MM_SWAPENTS); |
d559db08 | 1021 | inc_mm_counter(vma->vm_mm, MM_ANONPAGES); |
1da177e4 LT |
1022 | get_page(page); |
1023 | set_pte_at(vma->vm_mm, addr, pte, | |
1024 | pte_mkold(mk_pte(page, vma->vm_page_prot))); | |
9e16b7fb HD |
1025 | if (page == swapcache) |
1026 | page_add_anon_rmap(page, vma, addr); | |
1027 | else /* ksm created a completely new copy */ | |
1028 | page_add_new_anon_rmap(page, vma, addr); | |
72835c86 | 1029 | mem_cgroup_commit_charge_swapin(page, memcg); |
1da177e4 LT |
1030 | swap_free(entry); |
1031 | /* | |
1032 | * Move the page to the active list so it is not | |
1033 | * immediately swapped out again after swapon. | |
1034 | */ | |
1035 | activate_page(page); | |
044d66c1 HD |
1036 | out: |
1037 | pte_unmap_unlock(pte, ptl); | |
85d9fc89 | 1038 | out_nolock: |
9e16b7fb HD |
1039 | if (page != swapcache) { |
1040 | unlock_page(page); | |
1041 | put_page(page); | |
1042 | } | |
044d66c1 | 1043 | return ret; |
1da177e4 LT |
1044 | } |
1045 | ||
1046 | static int unuse_pte_range(struct vm_area_struct *vma, pmd_t *pmd, | |
1047 | unsigned long addr, unsigned long end, | |
1048 | swp_entry_t entry, struct page *page) | |
1049 | { | |
1da177e4 | 1050 | pte_t swp_pte = swp_entry_to_pte(entry); |
705e87c0 | 1051 | pte_t *pte; |
8a9f3ccd | 1052 | int ret = 0; |
1da177e4 | 1053 | |
044d66c1 HD |
1054 | /* |
1055 | * We don't actually need pte lock while scanning for swp_pte: since | |
1056 | * we hold page lock and mmap_sem, swp_pte cannot be inserted into the | |
1057 | * page table while we're scanning; though it could get zapped, and on | |
1058 | * some architectures (e.g. x86_32 with PAE) we might catch a glimpse | |
1059 | * of unmatched parts which look like swp_pte, so unuse_pte must | |
1060 | * recheck under pte lock. Scanning without pte lock lets it be | |
1061 | * preemptible whenever CONFIG_PREEMPT but not CONFIG_HIGHPTE. | |
1062 | */ | |
1063 | pte = pte_offset_map(pmd, addr); | |
1da177e4 LT |
1064 | do { |
1065 | /* | |
1066 | * swapoff spends a _lot_ of time in this loop! | |
1067 | * Test inline before going to call unuse_pte. | |
1068 | */ | |
1069 | if (unlikely(pte_same(*pte, swp_pte))) { | |
044d66c1 HD |
1070 | pte_unmap(pte); |
1071 | ret = unuse_pte(vma, pmd, addr, entry, page); | |
1072 | if (ret) | |
1073 | goto out; | |
1074 | pte = pte_offset_map(pmd, addr); | |
1da177e4 LT |
1075 | } |
1076 | } while (pte++, addr += PAGE_SIZE, addr != end); | |
044d66c1 HD |
1077 | pte_unmap(pte - 1); |
1078 | out: | |
8a9f3ccd | 1079 | return ret; |
1da177e4 LT |
1080 | } |
1081 | ||
1082 | static inline int unuse_pmd_range(struct vm_area_struct *vma, pud_t *pud, | |
1083 | unsigned long addr, unsigned long end, | |
1084 | swp_entry_t entry, struct page *page) | |
1085 | { | |
1086 | pmd_t *pmd; | |
1087 | unsigned long next; | |
8a9f3ccd | 1088 | int ret; |
1da177e4 LT |
1089 | |
1090 | pmd = pmd_offset(pud, addr); | |
1091 | do { | |
1092 | next = pmd_addr_end(addr, end); | |
1a5a9906 | 1093 | if (pmd_none_or_trans_huge_or_clear_bad(pmd)) |
1da177e4 | 1094 | continue; |
8a9f3ccd BS |
1095 | ret = unuse_pte_range(vma, pmd, addr, next, entry, page); |
1096 | if (ret) | |
1097 | return ret; | |
1da177e4 LT |
1098 | } while (pmd++, addr = next, addr != end); |
1099 | return 0; | |
1100 | } | |
1101 | ||
1102 | static inline int unuse_pud_range(struct vm_area_struct *vma, pgd_t *pgd, | |
1103 | unsigned long addr, unsigned long end, | |
1104 | swp_entry_t entry, struct page *page) | |
1105 | { | |
1106 | pud_t *pud; | |
1107 | unsigned long next; | |
8a9f3ccd | 1108 | int ret; |
1da177e4 LT |
1109 | |
1110 | pud = pud_offset(pgd, addr); | |
1111 | do { | |
1112 | next = pud_addr_end(addr, end); | |
1113 | if (pud_none_or_clear_bad(pud)) | |
1114 | continue; | |
8a9f3ccd BS |
1115 | ret = unuse_pmd_range(vma, pud, addr, next, entry, page); |
1116 | if (ret) | |
1117 | return ret; | |
1da177e4 LT |
1118 | } while (pud++, addr = next, addr != end); |
1119 | return 0; | |
1120 | } | |
1121 | ||
1122 | static int unuse_vma(struct vm_area_struct *vma, | |
1123 | swp_entry_t entry, struct page *page) | |
1124 | { | |
1125 | pgd_t *pgd; | |
1126 | unsigned long addr, end, next; | |
8a9f3ccd | 1127 | int ret; |
1da177e4 | 1128 | |
3ca7b3c5 | 1129 | if (page_anon_vma(page)) { |
1da177e4 LT |
1130 | addr = page_address_in_vma(page, vma); |
1131 | if (addr == -EFAULT) | |
1132 | return 0; | |
1133 | else | |
1134 | end = addr + PAGE_SIZE; | |
1135 | } else { | |
1136 | addr = vma->vm_start; | |
1137 | end = vma->vm_end; | |
1138 | } | |
1139 | ||
1140 | pgd = pgd_offset(vma->vm_mm, addr); | |
1141 | do { | |
1142 | next = pgd_addr_end(addr, end); | |
1143 | if (pgd_none_or_clear_bad(pgd)) | |
1144 | continue; | |
8a9f3ccd BS |
1145 | ret = unuse_pud_range(vma, pgd, addr, next, entry, page); |
1146 | if (ret) | |
1147 | return ret; | |
1da177e4 LT |
1148 | } while (pgd++, addr = next, addr != end); |
1149 | return 0; | |
1150 | } | |
1151 | ||
1152 | static int unuse_mm(struct mm_struct *mm, | |
1153 | swp_entry_t entry, struct page *page) | |
1154 | { | |
1155 | struct vm_area_struct *vma; | |
8a9f3ccd | 1156 | int ret = 0; |
1da177e4 LT |
1157 | |
1158 | if (!down_read_trylock(&mm->mmap_sem)) { | |
1159 | /* | |
7d03431c FLVC |
1160 | * Activate page so shrink_inactive_list is unlikely to unmap |
1161 | * its ptes while lock is dropped, so swapoff can make progress. | |
1da177e4 | 1162 | */ |
c475a8ab | 1163 | activate_page(page); |
1da177e4 LT |
1164 | unlock_page(page); |
1165 | down_read(&mm->mmap_sem); | |
1166 | lock_page(page); | |
1167 | } | |
1da177e4 | 1168 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
8a9f3ccd | 1169 | if (vma->anon_vma && (ret = unuse_vma(vma, entry, page))) |
1da177e4 LT |
1170 | break; |
1171 | } | |
1da177e4 | 1172 | up_read(&mm->mmap_sem); |
8a9f3ccd | 1173 | return (ret < 0)? ret: 0; |
1da177e4 LT |
1174 | } |
1175 | ||
1176 | /* | |
38b5faf4 DM |
1177 | * Scan swap_map (or frontswap_map if frontswap parameter is true) |
1178 | * from current position to next entry still in use. | |
1da177e4 LT |
1179 | * Recycle to start on reaching the end, returning 0 when empty. |
1180 | */ | |
6eb396dc | 1181 | static unsigned int find_next_to_unuse(struct swap_info_struct *si, |
38b5faf4 | 1182 | unsigned int prev, bool frontswap) |
1da177e4 | 1183 | { |
6eb396dc HD |
1184 | unsigned int max = si->max; |
1185 | unsigned int i = prev; | |
8d69aaee | 1186 | unsigned char count; |
1da177e4 LT |
1187 | |
1188 | /* | |
5d337b91 | 1189 | * No need for swap_lock here: we're just looking |
1da177e4 LT |
1190 | * for whether an entry is in use, not modifying it; false |
1191 | * hits are okay, and sys_swapoff() has already prevented new | |
5d337b91 | 1192 | * allocations from this area (while holding swap_lock). |
1da177e4 LT |
1193 | */ |
1194 | for (;;) { | |
1195 | if (++i >= max) { | |
1196 | if (!prev) { | |
1197 | i = 0; | |
1198 | break; | |
1199 | } | |
1200 | /* | |
1201 | * No entries in use at top of swap_map, | |
1202 | * loop back to start and recheck there. | |
1203 | */ | |
1204 | max = prev + 1; | |
1205 | prev = 0; | |
1206 | i = 1; | |
1207 | } | |
38b5faf4 DM |
1208 | if (frontswap) { |
1209 | if (frontswap_test(si, i)) | |
1210 | break; | |
1211 | else | |
1212 | continue; | |
1213 | } | |
1da177e4 | 1214 | count = si->swap_map[i]; |
355cfa73 | 1215 | if (count && swap_count(count) != SWAP_MAP_BAD) |
1da177e4 LT |
1216 | break; |
1217 | } | |
1218 | return i; | |
1219 | } | |
1220 | ||
1221 | /* | |
1222 | * We completely avoid races by reading each swap page in advance, | |
1223 | * and then search for the process using it. All the necessary | |
1224 | * page table adjustments can then be made atomically. | |
38b5faf4 DM |
1225 | * |
1226 | * if the boolean frontswap is true, only unuse pages_to_unuse pages; | |
1227 | * pages_to_unuse==0 means all pages; ignored if frontswap is false | |
1da177e4 | 1228 | */ |
38b5faf4 DM |
1229 | int try_to_unuse(unsigned int type, bool frontswap, |
1230 | unsigned long pages_to_unuse) | |
1da177e4 | 1231 | { |
efa90a98 | 1232 | struct swap_info_struct *si = swap_info[type]; |
1da177e4 | 1233 | struct mm_struct *start_mm; |
8d69aaee HD |
1234 | unsigned char *swap_map; |
1235 | unsigned char swcount; | |
1da177e4 LT |
1236 | struct page *page; |
1237 | swp_entry_t entry; | |
6eb396dc | 1238 | unsigned int i = 0; |
1da177e4 | 1239 | int retval = 0; |
1da177e4 LT |
1240 | |
1241 | /* | |
1242 | * When searching mms for an entry, a good strategy is to | |
1243 | * start at the first mm we freed the previous entry from | |
1244 | * (though actually we don't notice whether we or coincidence | |
1245 | * freed the entry). Initialize this start_mm with a hold. | |
1246 | * | |
1247 | * A simpler strategy would be to start at the last mm we | |
1248 | * freed the previous entry from; but that would take less | |
1249 | * advantage of mmlist ordering, which clusters forked mms | |
1250 | * together, child after parent. If we race with dup_mmap(), we | |
1251 | * prefer to resolve parent before child, lest we miss entries | |
1252 | * duplicated after we scanned child: using last mm would invert | |
570a335b | 1253 | * that. |
1da177e4 LT |
1254 | */ |
1255 | start_mm = &init_mm; | |
1256 | atomic_inc(&init_mm.mm_users); | |
1257 | ||
1258 | /* | |
1259 | * Keep on scanning until all entries have gone. Usually, | |
1260 | * one pass through swap_map is enough, but not necessarily: | |
1261 | * there are races when an instance of an entry might be missed. | |
1262 | */ | |
38b5faf4 | 1263 | while ((i = find_next_to_unuse(si, i, frontswap)) != 0) { |
1da177e4 LT |
1264 | if (signal_pending(current)) { |
1265 | retval = -EINTR; | |
1266 | break; | |
1267 | } | |
1268 | ||
886bb7e9 | 1269 | /* |
1da177e4 LT |
1270 | * Get a page for the entry, using the existing swap |
1271 | * cache page if there is one. Otherwise, get a clean | |
886bb7e9 | 1272 | * page and read the swap into it. |
1da177e4 LT |
1273 | */ |
1274 | swap_map = &si->swap_map[i]; | |
1275 | entry = swp_entry(type, i); | |
02098fea HD |
1276 | page = read_swap_cache_async(entry, |
1277 | GFP_HIGHUSER_MOVABLE, NULL, 0); | |
1da177e4 LT |
1278 | if (!page) { |
1279 | /* | |
1280 | * Either swap_duplicate() failed because entry | |
1281 | * has been freed independently, and will not be | |
1282 | * reused since sys_swapoff() already disabled | |
1283 | * allocation from here, or alloc_page() failed. | |
1284 | */ | |
1285 | if (!*swap_map) | |
1286 | continue; | |
1287 | retval = -ENOMEM; | |
1288 | break; | |
1289 | } | |
1290 | ||
1291 | /* | |
1292 | * Don't hold on to start_mm if it looks like exiting. | |
1293 | */ | |
1294 | if (atomic_read(&start_mm->mm_users) == 1) { | |
1295 | mmput(start_mm); | |
1296 | start_mm = &init_mm; | |
1297 | atomic_inc(&init_mm.mm_users); | |
1298 | } | |
1299 | ||
1300 | /* | |
1301 | * Wait for and lock page. When do_swap_page races with | |
1302 | * try_to_unuse, do_swap_page can handle the fault much | |
1303 | * faster than try_to_unuse can locate the entry. This | |
1304 | * apparently redundant "wait_on_page_locked" lets try_to_unuse | |
1305 | * defer to do_swap_page in such a case - in some tests, | |
1306 | * do_swap_page and try_to_unuse repeatedly compete. | |
1307 | */ | |
1308 | wait_on_page_locked(page); | |
1309 | wait_on_page_writeback(page); | |
1310 | lock_page(page); | |
1311 | wait_on_page_writeback(page); | |
1312 | ||
1313 | /* | |
1314 | * Remove all references to entry. | |
1da177e4 | 1315 | */ |
1da177e4 | 1316 | swcount = *swap_map; |
aaa46865 HD |
1317 | if (swap_count(swcount) == SWAP_MAP_SHMEM) { |
1318 | retval = shmem_unuse(entry, page); | |
1319 | /* page has already been unlocked and released */ | |
1320 | if (retval < 0) | |
1321 | break; | |
1322 | continue; | |
1da177e4 | 1323 | } |
aaa46865 HD |
1324 | if (swap_count(swcount) && start_mm != &init_mm) |
1325 | retval = unuse_mm(start_mm, entry, page); | |
1326 | ||
355cfa73 | 1327 | if (swap_count(*swap_map)) { |
1da177e4 LT |
1328 | int set_start_mm = (*swap_map >= swcount); |
1329 | struct list_head *p = &start_mm->mmlist; | |
1330 | struct mm_struct *new_start_mm = start_mm; | |
1331 | struct mm_struct *prev_mm = start_mm; | |
1332 | struct mm_struct *mm; | |
1333 | ||
1334 | atomic_inc(&new_start_mm->mm_users); | |
1335 | atomic_inc(&prev_mm->mm_users); | |
1336 | spin_lock(&mmlist_lock); | |
aaa46865 | 1337 | while (swap_count(*swap_map) && !retval && |
1da177e4 LT |
1338 | (p = p->next) != &start_mm->mmlist) { |
1339 | mm = list_entry(p, struct mm_struct, mmlist); | |
70af7c5c | 1340 | if (!atomic_inc_not_zero(&mm->mm_users)) |
1da177e4 | 1341 | continue; |
1da177e4 LT |
1342 | spin_unlock(&mmlist_lock); |
1343 | mmput(prev_mm); | |
1344 | prev_mm = mm; | |
1345 | ||
1346 | cond_resched(); | |
1347 | ||
1348 | swcount = *swap_map; | |
355cfa73 | 1349 | if (!swap_count(swcount)) /* any usage ? */ |
1da177e4 | 1350 | ; |
aaa46865 | 1351 | else if (mm == &init_mm) |
1da177e4 | 1352 | set_start_mm = 1; |
aaa46865 | 1353 | else |
1da177e4 | 1354 | retval = unuse_mm(mm, entry, page); |
355cfa73 | 1355 | |
32c5fc10 | 1356 | if (set_start_mm && *swap_map < swcount) { |
1da177e4 LT |
1357 | mmput(new_start_mm); |
1358 | atomic_inc(&mm->mm_users); | |
1359 | new_start_mm = mm; | |
1360 | set_start_mm = 0; | |
1361 | } | |
1362 | spin_lock(&mmlist_lock); | |
1363 | } | |
1364 | spin_unlock(&mmlist_lock); | |
1365 | mmput(prev_mm); | |
1366 | mmput(start_mm); | |
1367 | start_mm = new_start_mm; | |
1368 | } | |
1369 | if (retval) { | |
1370 | unlock_page(page); | |
1371 | page_cache_release(page); | |
1372 | break; | |
1373 | } | |
1374 | ||
1da177e4 LT |
1375 | /* |
1376 | * If a reference remains (rare), we would like to leave | |
1377 | * the page in the swap cache; but try_to_unmap could | |
1378 | * then re-duplicate the entry once we drop page lock, | |
1379 | * so we might loop indefinitely; also, that page could | |
1380 | * not be swapped out to other storage meanwhile. So: | |
1381 | * delete from cache even if there's another reference, | |
1382 | * after ensuring that the data has been saved to disk - | |
1383 | * since if the reference remains (rarer), it will be | |
1384 | * read from disk into another page. Splitting into two | |
1385 | * pages would be incorrect if swap supported "shared | |
1386 | * private" pages, but they are handled by tmpfs files. | |
5ad64688 HD |
1387 | * |
1388 | * Given how unuse_vma() targets one particular offset | |
1389 | * in an anon_vma, once the anon_vma has been determined, | |
1390 | * this splitting happens to be just what is needed to | |
1391 | * handle where KSM pages have been swapped out: re-reading | |
1392 | * is unnecessarily slow, but we can fix that later on. | |
1da177e4 | 1393 | */ |
355cfa73 KH |
1394 | if (swap_count(*swap_map) && |
1395 | PageDirty(page) && PageSwapCache(page)) { | |
1da177e4 LT |
1396 | struct writeback_control wbc = { |
1397 | .sync_mode = WB_SYNC_NONE, | |
1398 | }; | |
1399 | ||
1400 | swap_writepage(page, &wbc); | |
1401 | lock_page(page); | |
1402 | wait_on_page_writeback(page); | |
1403 | } | |
68bdc8d6 HD |
1404 | |
1405 | /* | |
1406 | * It is conceivable that a racing task removed this page from | |
1407 | * swap cache just before we acquired the page lock at the top, | |
1408 | * or while we dropped it in unuse_mm(). The page might even | |
1409 | * be back in swap cache on another swap area: that we must not | |
1410 | * delete, since it may not have been written out to swap yet. | |
1411 | */ | |
1412 | if (PageSwapCache(page) && | |
1413 | likely(page_private(page) == entry.val)) | |
2e0e26c7 | 1414 | delete_from_swap_cache(page); |
1da177e4 LT |
1415 | |
1416 | /* | |
1417 | * So we could skip searching mms once swap count went | |
1418 | * to 1, we did not mark any present ptes as dirty: must | |
2706a1b8 | 1419 | * mark page dirty so shrink_page_list will preserve it. |
1da177e4 LT |
1420 | */ |
1421 | SetPageDirty(page); | |
1422 | unlock_page(page); | |
1423 | page_cache_release(page); | |
1424 | ||
1425 | /* | |
1426 | * Make sure that we aren't completely killing | |
1427 | * interactive performance. | |
1428 | */ | |
1429 | cond_resched(); | |
38b5faf4 DM |
1430 | if (frontswap && pages_to_unuse > 0) { |
1431 | if (!--pages_to_unuse) | |
1432 | break; | |
1433 | } | |
1da177e4 LT |
1434 | } |
1435 | ||
1436 | mmput(start_mm); | |
1da177e4 LT |
1437 | return retval; |
1438 | } | |
1439 | ||
1440 | /* | |
5d337b91 HD |
1441 | * After a successful try_to_unuse, if no swap is now in use, we know |
1442 | * we can empty the mmlist. swap_lock must be held on entry and exit. | |
1443 | * Note that mmlist_lock nests inside swap_lock, and an mm must be | |
1da177e4 LT |
1444 | * added to the mmlist just after page_duplicate - before would be racy. |
1445 | */ | |
1446 | static void drain_mmlist(void) | |
1447 | { | |
1448 | struct list_head *p, *next; | |
efa90a98 | 1449 | unsigned int type; |
1da177e4 | 1450 | |
efa90a98 HD |
1451 | for (type = 0; type < nr_swapfiles; type++) |
1452 | if (swap_info[type]->inuse_pages) | |
1da177e4 LT |
1453 | return; |
1454 | spin_lock(&mmlist_lock); | |
1455 | list_for_each_safe(p, next, &init_mm.mmlist) | |
1456 | list_del_init(p); | |
1457 | spin_unlock(&mmlist_lock); | |
1458 | } | |
1459 | ||
1460 | /* | |
1461 | * Use this swapdev's extent info to locate the (PAGE_SIZE) block which | |
d4906e1a LS |
1462 | * corresponds to page offset for the specified swap entry. |
1463 | * Note that the type of this function is sector_t, but it returns page offset | |
1464 | * into the bdev, not sector offset. | |
1da177e4 | 1465 | */ |
6fa3eb70 | 1466 | sector_t map_swap_entry(swp_entry_t entry, struct block_device **bdev) |
1da177e4 | 1467 | { |
f29ad6a9 HD |
1468 | struct swap_info_struct *sis; |
1469 | struct swap_extent *start_se; | |
1470 | struct swap_extent *se; | |
1471 | pgoff_t offset; | |
1472 | ||
efa90a98 | 1473 | sis = swap_info[swp_type(entry)]; |
f29ad6a9 HD |
1474 | *bdev = sis->bdev; |
1475 | ||
1476 | offset = swp_offset(entry); | |
1477 | start_se = sis->curr_swap_extent; | |
1478 | se = start_se; | |
1da177e4 LT |
1479 | |
1480 | for ( ; ; ) { | |
1481 | struct list_head *lh; | |
1482 | ||
1483 | if (se->start_page <= offset && | |
1484 | offset < (se->start_page + se->nr_pages)) { | |
1485 | return se->start_block + (offset - se->start_page); | |
1486 | } | |
11d31886 | 1487 | lh = se->list.next; |
1da177e4 LT |
1488 | se = list_entry(lh, struct swap_extent, list); |
1489 | sis->curr_swap_extent = se; | |
1490 | BUG_ON(se == start_se); /* It *must* be present */ | |
1491 | } | |
1492 | } | |
6fa3eb70 | 1493 | EXPORT_SYMBOL_GPL(map_swap_entry); |
1da177e4 | 1494 | |
d4906e1a LS |
1495 | /* |
1496 | * Returns the page offset into bdev for the specified page's swap entry. | |
1497 | */ | |
1498 | sector_t map_swap_page(struct page *page, struct block_device **bdev) | |
1499 | { | |
1500 | swp_entry_t entry; | |
1501 | entry.val = page_private(page); | |
1502 | return map_swap_entry(entry, bdev); | |
1503 | } | |
1504 | ||
1da177e4 LT |
1505 | /* |
1506 | * Free all of a swapdev's extent information | |
1507 | */ | |
1508 | static void destroy_swap_extents(struct swap_info_struct *sis) | |
1509 | { | |
9625a5f2 | 1510 | while (!list_empty(&sis->first_swap_extent.list)) { |
1da177e4 LT |
1511 | struct swap_extent *se; |
1512 | ||
9625a5f2 | 1513 | se = list_entry(sis->first_swap_extent.list.next, |
1da177e4 LT |
1514 | struct swap_extent, list); |
1515 | list_del(&se->list); | |
1516 | kfree(se); | |
1517 | } | |
62c230bc MG |
1518 | |
1519 | if (sis->flags & SWP_FILE) { | |
1520 | struct file *swap_file = sis->swap_file; | |
1521 | struct address_space *mapping = swap_file->f_mapping; | |
1522 | ||
1523 | sis->flags &= ~SWP_FILE; | |
1524 | mapping->a_ops->swap_deactivate(swap_file); | |
1525 | } | |
1da177e4 LT |
1526 | } |
1527 | ||
1528 | /* | |
1529 | * Add a block range (and the corresponding page range) into this swapdev's | |
11d31886 | 1530 | * extent list. The extent list is kept sorted in page order. |
1da177e4 | 1531 | * |
11d31886 | 1532 | * This function rather assumes that it is called in ascending page order. |
1da177e4 | 1533 | */ |
a509bc1a | 1534 | int |
1da177e4 LT |
1535 | add_swap_extent(struct swap_info_struct *sis, unsigned long start_page, |
1536 | unsigned long nr_pages, sector_t start_block) | |
1537 | { | |
1538 | struct swap_extent *se; | |
1539 | struct swap_extent *new_se; | |
1540 | struct list_head *lh; | |
1541 | ||
9625a5f2 HD |
1542 | if (start_page == 0) { |
1543 | se = &sis->first_swap_extent; | |
1544 | sis->curr_swap_extent = se; | |
1545 | se->start_page = 0; | |
1546 | se->nr_pages = nr_pages; | |
1547 | se->start_block = start_block; | |
1548 | return 1; | |
1549 | } else { | |
1550 | lh = sis->first_swap_extent.list.prev; /* Highest extent */ | |
1da177e4 | 1551 | se = list_entry(lh, struct swap_extent, list); |
11d31886 HD |
1552 | BUG_ON(se->start_page + se->nr_pages != start_page); |
1553 | if (se->start_block + se->nr_pages == start_block) { | |
1da177e4 LT |
1554 | /* Merge it */ |
1555 | se->nr_pages += nr_pages; | |
1556 | return 0; | |
1557 | } | |
1da177e4 LT |
1558 | } |
1559 | ||
1560 | /* | |
1561 | * No merge. Insert a new extent, preserving ordering. | |
1562 | */ | |
1563 | new_se = kmalloc(sizeof(*se), GFP_KERNEL); | |
1564 | if (new_se == NULL) | |
1565 | return -ENOMEM; | |
1566 | new_se->start_page = start_page; | |
1567 | new_se->nr_pages = nr_pages; | |
1568 | new_se->start_block = start_block; | |
1569 | ||
9625a5f2 | 1570 | list_add_tail(&new_se->list, &sis->first_swap_extent.list); |
53092a74 | 1571 | return 1; |
1da177e4 LT |
1572 | } |
1573 | ||
1574 | /* | |
1575 | * A `swap extent' is a simple thing which maps a contiguous range of pages | |
1576 | * onto a contiguous range of disk blocks. An ordered list of swap extents | |
1577 | * is built at swapon time and is then used at swap_writepage/swap_readpage | |
1578 | * time for locating where on disk a page belongs. | |
1579 | * | |
1580 | * If the swapfile is an S_ISBLK block device, a single extent is installed. | |
1581 | * This is done so that the main operating code can treat S_ISBLK and S_ISREG | |
1582 | * swap files identically. | |
1583 | * | |
1584 | * Whether the swapdev is an S_ISREG file or an S_ISBLK blockdev, the swap | |
1585 | * extent list operates in PAGE_SIZE disk blocks. Both S_ISREG and S_ISBLK | |
1586 | * swapfiles are handled *identically* after swapon time. | |
1587 | * | |
1588 | * For S_ISREG swapfiles, setup_swap_extents() will walk all the file's blocks | |
1589 | * and will parse them into an ordered extent list, in PAGE_SIZE chunks. If | |
1590 | * some stray blocks are found which do not fall within the PAGE_SIZE alignment | |
1591 | * requirements, they are simply tossed out - we will never use those blocks | |
1592 | * for swapping. | |
1593 | * | |
b0d9bcd4 | 1594 | * For S_ISREG swapfiles we set S_SWAPFILE across the life of the swapon. This |
1da177e4 LT |
1595 | * prevents root from shooting her foot off by ftruncating an in-use swapfile, |
1596 | * which will scribble on the fs. | |
1597 | * | |
1598 | * The amount of disk space which a single swap extent represents varies. | |
1599 | * Typically it is in the 1-4 megabyte range. So we can have hundreds of | |
1600 | * extents in the list. To avoid much list walking, we cache the previous | |
1601 | * search location in `curr_swap_extent', and start new searches from there. | |
1602 | * This is extremely effective. The average number of iterations in | |
1603 | * map_swap_page() has been measured at about 0.3 per page. - akpm. | |
1604 | */ | |
53092a74 | 1605 | static int setup_swap_extents(struct swap_info_struct *sis, sector_t *span) |
1da177e4 | 1606 | { |
62c230bc MG |
1607 | struct file *swap_file = sis->swap_file; |
1608 | struct address_space *mapping = swap_file->f_mapping; | |
1609 | struct inode *inode = mapping->host; | |
1da177e4 LT |
1610 | int ret; |
1611 | ||
1da177e4 LT |
1612 | if (S_ISBLK(inode->i_mode)) { |
1613 | ret = add_swap_extent(sis, 0, sis->max, 0); | |
53092a74 | 1614 | *span = sis->pages; |
a509bc1a | 1615 | return ret; |
1da177e4 LT |
1616 | } |
1617 | ||
62c230bc | 1618 | if (mapping->a_ops->swap_activate) { |
a509bc1a | 1619 | ret = mapping->a_ops->swap_activate(sis, swap_file, span); |
62c230bc MG |
1620 | if (!ret) { |
1621 | sis->flags |= SWP_FILE; | |
1622 | ret = add_swap_extent(sis, 0, sis->max, 0); | |
1623 | *span = sis->pages; | |
1624 | } | |
a509bc1a | 1625 | return ret; |
62c230bc MG |
1626 | } |
1627 | ||
a509bc1a | 1628 | return generic_swapfile_activate(sis, swap_file, span); |
1da177e4 LT |
1629 | } |
1630 | ||
cf0cac0a | 1631 | static void _enable_swap_info(struct swap_info_struct *p, int prio, |
4f89849d | 1632 | unsigned char *swap_map) |
40531542 CEB |
1633 | { |
1634 | int i, prev; | |
1635 | ||
40531542 CEB |
1636 | if (prio >= 0) |
1637 | p->prio = prio; | |
1638 | else | |
1639 | p->prio = --least_priority; | |
1640 | p->swap_map = swap_map; | |
1641 | p->flags |= SWP_WRITEOK; | |
ec8acf20 | 1642 | atomic_long_add(p->pages, &nr_swap_pages); |
40531542 CEB |
1643 | total_swap_pages += p->pages; |
1644 | ||
1645 | /* insert swap space into swap_list: */ | |
1646 | prev = -1; | |
1647 | for (i = swap_list.head; i >= 0; i = swap_info[i]->next) { | |
1648 | if (p->prio >= swap_info[i]->prio) | |
1649 | break; | |
1650 | prev = i; | |
1651 | } | |
1652 | p->next = i; | |
1653 | if (prev < 0) | |
1654 | swap_list.head = swap_list.next = p->type; | |
1655 | else | |
1656 | swap_info[prev]->next = p->type; | |
cf0cac0a CEB |
1657 | } |
1658 | ||
1659 | static void enable_swap_info(struct swap_info_struct *p, int prio, | |
1660 | unsigned char *swap_map, | |
1661 | unsigned long *frontswap_map) | |
1662 | { | |
4f89849d | 1663 | frontswap_init(p->type, frontswap_map); |
cf0cac0a | 1664 | spin_lock(&swap_lock); |
ec8acf20 | 1665 | spin_lock(&p->lock); |
4f89849d | 1666 | _enable_swap_info(p, prio, swap_map); |
ec8acf20 | 1667 | spin_unlock(&p->lock); |
cf0cac0a CEB |
1668 | spin_unlock(&swap_lock); |
1669 | } | |
1670 | ||
1671 | static void reinsert_swap_info(struct swap_info_struct *p) | |
1672 | { | |
1673 | spin_lock(&swap_lock); | |
ec8acf20 | 1674 | spin_lock(&p->lock); |
4f89849d | 1675 | _enable_swap_info(p, p->prio, p->swap_map); |
ec8acf20 | 1676 | spin_unlock(&p->lock); |
40531542 CEB |
1677 | spin_unlock(&swap_lock); |
1678 | } | |
1679 | ||
c4ea37c2 | 1680 | SYSCALL_DEFINE1(swapoff, const char __user *, specialfile) |
1da177e4 | 1681 | { |
73c34b6a | 1682 | struct swap_info_struct *p = NULL; |
8d69aaee | 1683 | unsigned char *swap_map; |
4f89849d | 1684 | unsigned long *frontswap_map; |
1da177e4 LT |
1685 | struct file *swap_file, *victim; |
1686 | struct address_space *mapping; | |
1687 | struct inode *inode; | |
91a27b2a | 1688 | struct filename *pathname; |
1da177e4 LT |
1689 | int i, type, prev; |
1690 | int err; | |
886bb7e9 | 1691 | |
1da177e4 LT |
1692 | if (!capable(CAP_SYS_ADMIN)) |
1693 | return -EPERM; | |
1694 | ||
6fa3eb70 S |
1695 | #ifdef CONFIG_TOI |
1696 | // FIXME: Turn it off due to current->mm may be NULL in kernel space | |
1697 | // by calling sys_swapoff(swapfilename) in disable_swapfile() @ tuxonice_swap.c | |
1698 | pr_warn("[HIB/SWAP] [%s] file(%s) current(%p/%d/%s) current->mm(%p)\n", __func__, specialfile, current, current->pid, current->comm, current->mm); | |
1699 | WARN_ON(!current->mm); | |
1700 | #else | |
191c5424 | 1701 | BUG_ON(!current->mm); |
6fa3eb70 | 1702 | #endif |
191c5424 | 1703 | |
1da177e4 | 1704 | pathname = getname(specialfile); |
1da177e4 | 1705 | if (IS_ERR(pathname)) |
f58b59c1 | 1706 | return PTR_ERR(pathname); |
1da177e4 | 1707 | |
669abf4e | 1708 | victim = file_open_name(pathname, O_RDWR|O_LARGEFILE, 0); |
1da177e4 LT |
1709 | err = PTR_ERR(victim); |
1710 | if (IS_ERR(victim)) | |
1711 | goto out; | |
1712 | ||
1713 | mapping = victim->f_mapping; | |
1714 | prev = -1; | |
5d337b91 | 1715 | spin_lock(&swap_lock); |
efa90a98 HD |
1716 | for (type = swap_list.head; type >= 0; type = swap_info[type]->next) { |
1717 | p = swap_info[type]; | |
22c6f8fd | 1718 | if (p->flags & SWP_WRITEOK) { |
1da177e4 LT |
1719 | if (p->swap_file->f_mapping == mapping) |
1720 | break; | |
1721 | } | |
1722 | prev = type; | |
1723 | } | |
1724 | if (type < 0) { | |
1725 | err = -EINVAL; | |
5d337b91 | 1726 | spin_unlock(&swap_lock); |
1da177e4 LT |
1727 | goto out_dput; |
1728 | } | |
191c5424 | 1729 | if (!security_vm_enough_memory_mm(current->mm, p->pages)) |
1da177e4 LT |
1730 | vm_unacct_memory(p->pages); |
1731 | else { | |
1732 | err = -ENOMEM; | |
5d337b91 | 1733 | spin_unlock(&swap_lock); |
1da177e4 LT |
1734 | goto out_dput; |
1735 | } | |
efa90a98 | 1736 | if (prev < 0) |
1da177e4 | 1737 | swap_list.head = p->next; |
efa90a98 HD |
1738 | else |
1739 | swap_info[prev]->next = p->next; | |
1da177e4 LT |
1740 | if (type == swap_list.next) { |
1741 | /* just pick something that's safe... */ | |
1742 | swap_list.next = swap_list.head; | |
1743 | } | |
ec8acf20 | 1744 | spin_lock(&p->lock); |
78ecba08 | 1745 | if (p->prio < 0) { |
efa90a98 HD |
1746 | for (i = p->next; i >= 0; i = swap_info[i]->next) |
1747 | swap_info[i]->prio = p->prio--; | |
78ecba08 HD |
1748 | least_priority++; |
1749 | } | |
ec8acf20 | 1750 | atomic_long_sub(p->pages, &nr_swap_pages); |
1da177e4 LT |
1751 | total_swap_pages -= p->pages; |
1752 | p->flags &= ~SWP_WRITEOK; | |
ec8acf20 | 1753 | spin_unlock(&p->lock); |
5d337b91 | 1754 | spin_unlock(&swap_lock); |
fb4f88dc | 1755 | |
e1e12d2f | 1756 | set_current_oom_origin(); |
38b5faf4 | 1757 | err = try_to_unuse(type, false, 0); /* force all pages to be unused */ |
e1e12d2f | 1758 | clear_current_oom_origin(); |
1da177e4 | 1759 | |
1da177e4 LT |
1760 | if (err) { |
1761 | /* re-insert swap space back into swap_list */ | |
cf0cac0a | 1762 | reinsert_swap_info(p); |
1da177e4 LT |
1763 | goto out_dput; |
1764 | } | |
52b7efdb | 1765 | |
5d337b91 | 1766 | destroy_swap_extents(p); |
570a335b HD |
1767 | if (p->flags & SWP_CONTINUED) |
1768 | free_swap_count_continuations(p); | |
1769 | ||
fc0abb14 | 1770 | mutex_lock(&swapon_mutex); |
5d337b91 | 1771 | spin_lock(&swap_lock); |
ec8acf20 | 1772 | spin_lock(&p->lock); |
5d337b91 HD |
1773 | drain_mmlist(); |
1774 | ||
52b7efdb | 1775 | /* wait for anyone still in scan_swap_map */ |
52b7efdb HD |
1776 | p->highest_bit = 0; /* cuts scans short */ |
1777 | while (p->flags >= SWP_SCANNING) { | |
ec8acf20 | 1778 | spin_unlock(&p->lock); |
5d337b91 | 1779 | spin_unlock(&swap_lock); |
13e4b57f | 1780 | schedule_timeout_uninterruptible(1); |
5d337b91 | 1781 | spin_lock(&swap_lock); |
ec8acf20 | 1782 | spin_lock(&p->lock); |
52b7efdb | 1783 | } |
52b7efdb | 1784 | |
1da177e4 LT |
1785 | swap_file = p->swap_file; |
1786 | p->swap_file = NULL; | |
1787 | p->max = 0; | |
1788 | swap_map = p->swap_map; | |
1789 | p->swap_map = NULL; | |
1790 | p->flags = 0; | |
4f89849d MK |
1791 | frontswap_map = frontswap_map_get(p); |
1792 | frontswap_map_set(p, NULL); | |
ec8acf20 | 1793 | spin_unlock(&p->lock); |
5d337b91 | 1794 | spin_unlock(&swap_lock); |
4f89849d | 1795 | frontswap_invalidate_area(type); |
fc0abb14 | 1796 | mutex_unlock(&swapon_mutex); |
1da177e4 | 1797 | vfree(swap_map); |
4f89849d | 1798 | vfree(frontswap_map); |
27a7faa0 KH |
1799 | /* Destroy swap account informatin */ |
1800 | swap_cgroup_swapoff(type); | |
1801 | ||
1da177e4 LT |
1802 | inode = mapping->host; |
1803 | if (S_ISBLK(inode->i_mode)) { | |
1804 | struct block_device *bdev = I_BDEV(inode); | |
1805 | set_blocksize(bdev, p->old_block_size); | |
e525fd89 | 1806 | blkdev_put(bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL); |
1da177e4 | 1807 | } else { |
1b1dcc1b | 1808 | mutex_lock(&inode->i_mutex); |
1da177e4 | 1809 | inode->i_flags &= ~S_SWAPFILE; |
1b1dcc1b | 1810 | mutex_unlock(&inode->i_mutex); |
1da177e4 LT |
1811 | } |
1812 | filp_close(swap_file, NULL); | |
1813 | err = 0; | |
66d7dd51 KS |
1814 | atomic_inc(&proc_poll_event); |
1815 | wake_up_interruptible(&proc_poll_wait); | |
1da177e4 LT |
1816 | |
1817 | out_dput: | |
1818 | filp_close(victim, NULL); | |
1819 | out: | |
f58b59c1 | 1820 | putname(pathname); |
1da177e4 LT |
1821 | return err; |
1822 | } | |
6fa3eb70 | 1823 | EXPORT_SYMBOL_GPL(sys_swapoff); |
1da177e4 LT |
1824 | |
1825 | #ifdef CONFIG_PROC_FS | |
66d7dd51 KS |
1826 | static unsigned swaps_poll(struct file *file, poll_table *wait) |
1827 | { | |
f1514638 | 1828 | struct seq_file *seq = file->private_data; |
66d7dd51 KS |
1829 | |
1830 | poll_wait(file, &proc_poll_wait, wait); | |
1831 | ||
f1514638 KS |
1832 | if (seq->poll_event != atomic_read(&proc_poll_event)) { |
1833 | seq->poll_event = atomic_read(&proc_poll_event); | |
66d7dd51 KS |
1834 | return POLLIN | POLLRDNORM | POLLERR | POLLPRI; |
1835 | } | |
1836 | ||
1837 | return POLLIN | POLLRDNORM; | |
1838 | } | |
1839 | ||
1da177e4 LT |
1840 | /* iterator */ |
1841 | static void *swap_start(struct seq_file *swap, loff_t *pos) | |
1842 | { | |
efa90a98 HD |
1843 | struct swap_info_struct *si; |
1844 | int type; | |
1da177e4 LT |
1845 | loff_t l = *pos; |
1846 | ||
fc0abb14 | 1847 | mutex_lock(&swapon_mutex); |
1da177e4 | 1848 | |
881e4aab SS |
1849 | if (!l) |
1850 | return SEQ_START_TOKEN; | |
1851 | ||
efa90a98 HD |
1852 | for (type = 0; type < nr_swapfiles; type++) { |
1853 | smp_rmb(); /* read nr_swapfiles before swap_info[type] */ | |
1854 | si = swap_info[type]; | |
1855 | if (!(si->flags & SWP_USED) || !si->swap_map) | |
1da177e4 | 1856 | continue; |
881e4aab | 1857 | if (!--l) |
efa90a98 | 1858 | return si; |
1da177e4 LT |
1859 | } |
1860 | ||
1861 | return NULL; | |
1862 | } | |
1863 | ||
1864 | static void *swap_next(struct seq_file *swap, void *v, loff_t *pos) | |
1865 | { | |
efa90a98 HD |
1866 | struct swap_info_struct *si = v; |
1867 | int type; | |
1da177e4 | 1868 | |
881e4aab | 1869 | if (v == SEQ_START_TOKEN) |
efa90a98 HD |
1870 | type = 0; |
1871 | else | |
1872 | type = si->type + 1; | |
881e4aab | 1873 | |
efa90a98 HD |
1874 | for (; type < nr_swapfiles; type++) { |
1875 | smp_rmb(); /* read nr_swapfiles before swap_info[type] */ | |
1876 | si = swap_info[type]; | |
1877 | if (!(si->flags & SWP_USED) || !si->swap_map) | |
1da177e4 LT |
1878 | continue; |
1879 | ++*pos; | |
efa90a98 | 1880 | return si; |
1da177e4 LT |
1881 | } |
1882 | ||
1883 | return NULL; | |
1884 | } | |
1885 | ||
1886 | static void swap_stop(struct seq_file *swap, void *v) | |
1887 | { | |
fc0abb14 | 1888 | mutex_unlock(&swapon_mutex); |
1da177e4 LT |
1889 | } |
1890 | ||
1891 | static int swap_show(struct seq_file *swap, void *v) | |
1892 | { | |
efa90a98 | 1893 | struct swap_info_struct *si = v; |
1da177e4 LT |
1894 | struct file *file; |
1895 | int len; | |
1896 | ||
efa90a98 | 1897 | if (si == SEQ_START_TOKEN) { |
881e4aab SS |
1898 | seq_puts(swap,"Filename\t\t\t\tType\t\tSize\tUsed\tPriority\n"); |
1899 | return 0; | |
1900 | } | |
1da177e4 | 1901 | |
efa90a98 | 1902 | file = si->swap_file; |
c32c2f63 | 1903 | len = seq_path(swap, &file->f_path, " \t\n\\"); |
6eb396dc | 1904 | seq_printf(swap, "%*s%s\t%u\t%u\t%d\n", |
886bb7e9 | 1905 | len < 40 ? 40 - len : 1, " ", |
496ad9aa | 1906 | S_ISBLK(file_inode(file)->i_mode) ? |
1da177e4 | 1907 | "partition" : "file\t", |
efa90a98 HD |
1908 | si->pages << (PAGE_SHIFT - 10), |
1909 | si->inuse_pages << (PAGE_SHIFT - 10), | |
1910 | si->prio); | |
1da177e4 LT |
1911 | return 0; |
1912 | } | |
1913 | ||
15ad7cdc | 1914 | static const struct seq_operations swaps_op = { |
1da177e4 LT |
1915 | .start = swap_start, |
1916 | .next = swap_next, | |
1917 | .stop = swap_stop, | |
1918 | .show = swap_show | |
1919 | }; | |
1920 | ||
1921 | static int swaps_open(struct inode *inode, struct file *file) | |
1922 | { | |
f1514638 | 1923 | struct seq_file *seq; |
66d7dd51 KS |
1924 | int ret; |
1925 | ||
66d7dd51 | 1926 | ret = seq_open(file, &swaps_op); |
f1514638 | 1927 | if (ret) |
66d7dd51 | 1928 | return ret; |
66d7dd51 | 1929 | |
f1514638 KS |
1930 | seq = file->private_data; |
1931 | seq->poll_event = atomic_read(&proc_poll_event); | |
1932 | return 0; | |
1da177e4 LT |
1933 | } |
1934 | ||
15ad7cdc | 1935 | static const struct file_operations proc_swaps_operations = { |
1da177e4 LT |
1936 | .open = swaps_open, |
1937 | .read = seq_read, | |
1938 | .llseek = seq_lseek, | |
1939 | .release = seq_release, | |
66d7dd51 | 1940 | .poll = swaps_poll, |
1da177e4 LT |
1941 | }; |
1942 | ||
1943 | static int __init procswaps_init(void) | |
1944 | { | |
3d71f86f | 1945 | proc_create("swaps", 0, NULL, &proc_swaps_operations); |
1da177e4 LT |
1946 | return 0; |
1947 | } | |
1948 | __initcall(procswaps_init); | |
1949 | #endif /* CONFIG_PROC_FS */ | |
1950 | ||
1796316a JB |
1951 | #ifdef MAX_SWAPFILES_CHECK |
1952 | static int __init max_swapfiles_check(void) | |
1953 | { | |
1954 | MAX_SWAPFILES_CHECK(); | |
1955 | return 0; | |
1956 | } | |
1957 | late_initcall(max_swapfiles_check); | |
1958 | #endif | |
1959 | ||
53cbb243 | 1960 | static struct swap_info_struct *alloc_swap_info(void) |
1da177e4 | 1961 | { |
73c34b6a | 1962 | struct swap_info_struct *p; |
1da177e4 | 1963 | unsigned int type; |
efa90a98 HD |
1964 | |
1965 | p = kzalloc(sizeof(*p), GFP_KERNEL); | |
1966 | if (!p) | |
53cbb243 | 1967 | return ERR_PTR(-ENOMEM); |
efa90a98 | 1968 | |
5d337b91 | 1969 | spin_lock(&swap_lock); |
efa90a98 HD |
1970 | for (type = 0; type < nr_swapfiles; type++) { |
1971 | if (!(swap_info[type]->flags & SWP_USED)) | |
1da177e4 | 1972 | break; |
efa90a98 | 1973 | } |
0697212a | 1974 | if (type >= MAX_SWAPFILES) { |
5d337b91 | 1975 | spin_unlock(&swap_lock); |
efa90a98 | 1976 | kfree(p); |
730c0581 | 1977 | return ERR_PTR(-EPERM); |
1da177e4 | 1978 | } |
efa90a98 HD |
1979 | if (type >= nr_swapfiles) { |
1980 | p->type = type; | |
1981 | swap_info[type] = p; | |
1982 | /* | |
1983 | * Write swap_info[type] before nr_swapfiles, in case a | |
1984 | * racing procfs swap_start() or swap_next() is reading them. | |
1985 | * (We never shrink nr_swapfiles, we never free this entry.) | |
1986 | */ | |
1987 | smp_wmb(); | |
1988 | nr_swapfiles++; | |
1989 | } else { | |
1990 | kfree(p); | |
1991 | p = swap_info[type]; | |
1992 | /* | |
1993 | * Do not memset this entry: a racing procfs swap_next() | |
1994 | * would be relying on p->type to remain valid. | |
1995 | */ | |
1996 | } | |
9625a5f2 | 1997 | INIT_LIST_HEAD(&p->first_swap_extent.list); |
1da177e4 | 1998 | p->flags = SWP_USED; |
1da177e4 | 1999 | p->next = -1; |
5d337b91 | 2000 | spin_unlock(&swap_lock); |
ec8acf20 | 2001 | spin_lock_init(&p->lock); |
efa90a98 | 2002 | |
53cbb243 | 2003 | return p; |
53cbb243 CEB |
2004 | } |
2005 | ||
4d0e1e10 CEB |
2006 | static int claim_swapfile(struct swap_info_struct *p, struct inode *inode) |
2007 | { | |
2008 | int error; | |
2009 | ||
2010 | if (S_ISBLK(inode->i_mode)) { | |
2011 | p->bdev = bdgrab(I_BDEV(inode)); | |
2012 | error = blkdev_get(p->bdev, | |
2013 | FMODE_READ | FMODE_WRITE | FMODE_EXCL, | |
2014 | sys_swapon); | |
2015 | if (error < 0) { | |
2016 | p->bdev = NULL; | |
87ade72a | 2017 | return -EINVAL; |
4d0e1e10 CEB |
2018 | } |
2019 | p->old_block_size = block_size(p->bdev); | |
2020 | error = set_blocksize(p->bdev, PAGE_SIZE); | |
2021 | if (error < 0) | |
87ade72a | 2022 | return error; |
4d0e1e10 CEB |
2023 | p->flags |= SWP_BLKDEV; |
2024 | } else if (S_ISREG(inode->i_mode)) { | |
2025 | p->bdev = inode->i_sb->s_bdev; | |
2026 | mutex_lock(&inode->i_mutex); | |
87ade72a CEB |
2027 | if (IS_SWAPFILE(inode)) |
2028 | return -EBUSY; | |
2029 | } else | |
2030 | return -EINVAL; | |
4d0e1e10 CEB |
2031 | |
2032 | return 0; | |
4d0e1e10 CEB |
2033 | } |
2034 | ||
ca8bd38b CEB |
2035 | static unsigned long read_swap_header(struct swap_info_struct *p, |
2036 | union swap_header *swap_header, | |
2037 | struct inode *inode) | |
2038 | { | |
2039 | int i; | |
2040 | unsigned long maxpages; | |
2041 | unsigned long swapfilepages; | |
2042 | ||
2043 | if (memcmp("SWAPSPACE2", swap_header->magic.magic, 10)) { | |
2044 | printk(KERN_ERR "Unable to find swap-space signature\n"); | |
38719025 | 2045 | return 0; |
ca8bd38b CEB |
2046 | } |
2047 | ||
2048 | /* swap partition endianess hack... */ | |
2049 | if (swab32(swap_header->info.version) == 1) { | |
2050 | swab32s(&swap_header->info.version); | |
2051 | swab32s(&swap_header->info.last_page); | |
2052 | swab32s(&swap_header->info.nr_badpages); | |
2053 | for (i = 0; i < swap_header->info.nr_badpages; i++) | |
2054 | swab32s(&swap_header->info.badpages[i]); | |
2055 | } | |
2056 | /* Check the swap header's sub-version */ | |
2057 | if (swap_header->info.version != 1) { | |
2058 | printk(KERN_WARNING | |
2059 | "Unable to handle swap header version %d\n", | |
2060 | swap_header->info.version); | |
38719025 | 2061 | return 0; |
ca8bd38b CEB |
2062 | } |
2063 | ||
2064 | p->lowest_bit = 1; | |
2065 | p->cluster_next = 1; | |
2066 | p->cluster_nr = 0; | |
2067 | ||
2068 | /* | |
2069 | * Find out how many pages are allowed for a single swap | |
9b15b817 | 2070 | * device. There are two limiting factors: 1) the number |
a2c16d6c HD |
2071 | * of bits for the swap offset in the swp_entry_t type, and |
2072 | * 2) the number of bits in the swap pte as defined by the | |
9b15b817 | 2073 | * different architectures. In order to find the |
a2c16d6c | 2074 | * largest possible bit mask, a swap entry with swap type 0 |
ca8bd38b | 2075 | * and swap offset ~0UL is created, encoded to a swap pte, |
a2c16d6c | 2076 | * decoded to a swp_entry_t again, and finally the swap |
ca8bd38b CEB |
2077 | * offset is extracted. This will mask all the bits from |
2078 | * the initial ~0UL mask that can't be encoded in either | |
2079 | * the swp_entry_t or the architecture definition of a | |
9b15b817 | 2080 | * swap pte. |
ca8bd38b CEB |
2081 | */ |
2082 | maxpages = swp_offset(pte_to_swp_entry( | |
9b15b817 | 2083 | swp_entry_to_pte(swp_entry(0, ~0UL)))) + 1; |
ca8bd38b CEB |
2084 | if (maxpages > swap_header->info.last_page) { |
2085 | maxpages = swap_header->info.last_page + 1; | |
2086 | /* p->max is an unsigned int: don't overflow it */ | |
2087 | if ((unsigned int)maxpages == 0) | |
2088 | maxpages = UINT_MAX; | |
2089 | } | |
2090 | p->highest_bit = maxpages - 1; | |
2091 | ||
2092 | if (!maxpages) | |
38719025 | 2093 | return 0; |
ca8bd38b CEB |
2094 | swapfilepages = i_size_read(inode) >> PAGE_SHIFT; |
2095 | if (swapfilepages && maxpages > swapfilepages) { | |
2096 | printk(KERN_WARNING | |
2097 | "Swap area shorter than signature indicates\n"); | |
38719025 | 2098 | return 0; |
ca8bd38b CEB |
2099 | } |
2100 | if (swap_header->info.nr_badpages && S_ISREG(inode->i_mode)) | |
38719025 | 2101 | return 0; |
ca8bd38b | 2102 | if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES) |
38719025 | 2103 | return 0; |
ca8bd38b CEB |
2104 | |
2105 | return maxpages; | |
ca8bd38b CEB |
2106 | } |
2107 | ||
915d4d7b CEB |
2108 | static int setup_swap_map_and_extents(struct swap_info_struct *p, |
2109 | union swap_header *swap_header, | |
2110 | unsigned char *swap_map, | |
2111 | unsigned long maxpages, | |
2112 | sector_t *span) | |
2113 | { | |
2114 | int i; | |
915d4d7b CEB |
2115 | unsigned int nr_good_pages; |
2116 | int nr_extents; | |
2117 | ||
2118 | nr_good_pages = maxpages - 1; /* omit header page */ | |
2119 | ||
2120 | for (i = 0; i < swap_header->info.nr_badpages; i++) { | |
2121 | unsigned int page_nr = swap_header->info.badpages[i]; | |
bdb8e3f6 CEB |
2122 | if (page_nr == 0 || page_nr > swap_header->info.last_page) |
2123 | return -EINVAL; | |
915d4d7b CEB |
2124 | if (page_nr < maxpages) { |
2125 | swap_map[page_nr] = SWAP_MAP_BAD; | |
2126 | nr_good_pages--; | |
2127 | } | |
2128 | } | |
2129 | ||
2130 | if (nr_good_pages) { | |
2131 | swap_map[0] = SWAP_MAP_BAD; | |
2132 | p->max = maxpages; | |
2133 | p->pages = nr_good_pages; | |
2134 | nr_extents = setup_swap_extents(p, span); | |
bdb8e3f6 CEB |
2135 | if (nr_extents < 0) |
2136 | return nr_extents; | |
915d4d7b CEB |
2137 | nr_good_pages = p->pages; |
2138 | } | |
2139 | if (!nr_good_pages) { | |
2140 | printk(KERN_WARNING "Empty swap-file\n"); | |
bdb8e3f6 | 2141 | return -EINVAL; |
915d4d7b CEB |
2142 | } |
2143 | ||
2144 | return nr_extents; | |
915d4d7b CEB |
2145 | } |
2146 | ||
53cbb243 CEB |
2147 | SYSCALL_DEFINE2(swapon, const char __user *, specialfile, int, swap_flags) |
2148 | { | |
2149 | struct swap_info_struct *p; | |
91a27b2a | 2150 | struct filename *name; |
53cbb243 CEB |
2151 | struct file *swap_file = NULL; |
2152 | struct address_space *mapping; | |
40531542 CEB |
2153 | int i; |
2154 | int prio; | |
53cbb243 CEB |
2155 | int error; |
2156 | union swap_header *swap_header; | |
915d4d7b | 2157 | int nr_extents; |
53cbb243 CEB |
2158 | sector_t span; |
2159 | unsigned long maxpages; | |
53cbb243 | 2160 | unsigned char *swap_map = NULL; |
38b5faf4 | 2161 | unsigned long *frontswap_map = NULL; |
53cbb243 CEB |
2162 | struct page *page = NULL; |
2163 | struct inode *inode = NULL; | |
53cbb243 | 2164 | |
d15cab97 HD |
2165 | if (swap_flags & ~SWAP_FLAGS_VALID) |
2166 | return -EINVAL; | |
2167 | ||
53cbb243 CEB |
2168 | if (!capable(CAP_SYS_ADMIN)) |
2169 | return -EPERM; | |
2170 | ||
2171 | p = alloc_swap_info(); | |
2542e513 CEB |
2172 | if (IS_ERR(p)) |
2173 | return PTR_ERR(p); | |
53cbb243 | 2174 | |
1da177e4 | 2175 | name = getname(specialfile); |
1da177e4 | 2176 | if (IS_ERR(name)) { |
7de7fb6b | 2177 | error = PTR_ERR(name); |
1da177e4 | 2178 | name = NULL; |
bd69010b | 2179 | goto bad_swap; |
1da177e4 | 2180 | } |
669abf4e | 2181 | swap_file = file_open_name(name, O_RDWR|O_LARGEFILE, 0); |
1da177e4 | 2182 | if (IS_ERR(swap_file)) { |
7de7fb6b | 2183 | error = PTR_ERR(swap_file); |
1da177e4 | 2184 | swap_file = NULL; |
bd69010b | 2185 | goto bad_swap; |
1da177e4 LT |
2186 | } |
2187 | ||
2188 | p->swap_file = swap_file; | |
2189 | mapping = swap_file->f_mapping; | |
1da177e4 | 2190 | |
1da177e4 | 2191 | for (i = 0; i < nr_swapfiles; i++) { |
efa90a98 | 2192 | struct swap_info_struct *q = swap_info[i]; |
1da177e4 | 2193 | |
e8e6c2ec | 2194 | if (q == p || !q->swap_file) |
1da177e4 | 2195 | continue; |
7de7fb6b CEB |
2196 | if (mapping == q->swap_file->f_mapping) { |
2197 | error = -EBUSY; | |
1da177e4 | 2198 | goto bad_swap; |
7de7fb6b | 2199 | } |
1da177e4 LT |
2200 | } |
2201 | ||
2130781e CEB |
2202 | inode = mapping->host; |
2203 | /* If S_ISREG(inode->i_mode) will do mutex_lock(&inode->i_mutex); */ | |
4d0e1e10 CEB |
2204 | error = claim_swapfile(p, inode); |
2205 | if (unlikely(error)) | |
1da177e4 | 2206 | goto bad_swap; |
1da177e4 | 2207 | |
1da177e4 LT |
2208 | /* |
2209 | * Read the swap header. | |
2210 | */ | |
2211 | if (!mapping->a_ops->readpage) { | |
2212 | error = -EINVAL; | |
2213 | goto bad_swap; | |
2214 | } | |
090d2b18 | 2215 | page = read_mapping_page(mapping, 0, swap_file); |
1da177e4 LT |
2216 | if (IS_ERR(page)) { |
2217 | error = PTR_ERR(page); | |
2218 | goto bad_swap; | |
2219 | } | |
81e33971 | 2220 | swap_header = kmap(page); |
1da177e4 | 2221 | |
ca8bd38b CEB |
2222 | maxpages = read_swap_header(p, swap_header, inode); |
2223 | if (unlikely(!maxpages)) { | |
1da177e4 LT |
2224 | error = -EINVAL; |
2225 | goto bad_swap; | |
2226 | } | |
886bb7e9 | 2227 | |
81e33971 | 2228 | /* OK, set up the swap map and apply the bad block list */ |
803d0c83 | 2229 | swap_map = vzalloc(maxpages); |
81e33971 HD |
2230 | if (!swap_map) { |
2231 | error = -ENOMEM; | |
2232 | goto bad_swap; | |
2233 | } | |
1da177e4 | 2234 | |
1421ef3c CEB |
2235 | error = swap_cgroup_swapon(p->type, maxpages); |
2236 | if (error) | |
2237 | goto bad_swap; | |
2238 | ||
915d4d7b CEB |
2239 | nr_extents = setup_swap_map_and_extents(p, swap_header, swap_map, |
2240 | maxpages, &span); | |
2241 | if (unlikely(nr_extents < 0)) { | |
2242 | error = nr_extents; | |
1da177e4 LT |
2243 | goto bad_swap; |
2244 | } | |
38b5faf4 DM |
2245 | /* frontswap enabled? set up bit-per-page map for frontswap */ |
2246 | if (frontswap_enabled) | |
7b57976d | 2247 | frontswap_map = vzalloc(BITS_TO_LONGS(maxpages) * sizeof(long)); |
1da177e4 | 2248 | |
3bd0f0c7 SJ |
2249 | if (p->bdev) { |
2250 | if (blk_queue_nonrot(bdev_get_queue(p->bdev))) { | |
2251 | p->flags |= SWP_SOLIDSTATE; | |
d3d30417 | 2252 | p->cluster_next = 1 + (prandom_u32() % p->highest_bit); |
3bd0f0c7 | 2253 | } |
052b1987 | 2254 | if ((swap_flags & SWAP_FLAG_DISCARD) && discard_swap(p) == 0) |
3bd0f0c7 | 2255 | p->flags |= SWP_DISCARDABLE; |
20137a49 | 2256 | } |
6a6ba831 | 2257 | |
fc0abb14 | 2258 | mutex_lock(&swapon_mutex); |
40531542 | 2259 | prio = -1; |
78ecba08 | 2260 | if (swap_flags & SWAP_FLAG_PREFER) |
40531542 | 2261 | prio = |
78ecba08 | 2262 | (swap_flags & SWAP_FLAG_PRIO_MASK) >> SWAP_FLAG_PRIO_SHIFT; |
38b5faf4 | 2263 | enable_swap_info(p, prio, swap_map, frontswap_map); |
c69dbfb8 CEB |
2264 | |
2265 | printk(KERN_INFO "Adding %uk swap on %s. " | |
38b5faf4 | 2266 | "Priority:%d extents:%d across:%lluk %s%s%s\n", |
91a27b2a | 2267 | p->pages<<(PAGE_SHIFT-10), name->name, p->prio, |
c69dbfb8 CEB |
2268 | nr_extents, (unsigned long long)span<<(PAGE_SHIFT-10), |
2269 | (p->flags & SWP_SOLIDSTATE) ? "SS" : "", | |
38b5faf4 DM |
2270 | (p->flags & SWP_DISCARDABLE) ? "D" : "", |
2271 | (frontswap_map) ? "FS" : ""); | |
c69dbfb8 | 2272 | |
fc0abb14 | 2273 | mutex_unlock(&swapon_mutex); |
66d7dd51 KS |
2274 | atomic_inc(&proc_poll_event); |
2275 | wake_up_interruptible(&proc_poll_wait); | |
2276 | ||
9b01c350 CEB |
2277 | if (S_ISREG(inode->i_mode)) |
2278 | inode->i_flags |= S_SWAPFILE; | |
1da177e4 LT |
2279 | error = 0; |
2280 | goto out; | |
2281 | bad_swap: | |
bd69010b | 2282 | if (inode && S_ISBLK(inode->i_mode) && p->bdev) { |
f2090d2d CEB |
2283 | set_blocksize(p->bdev, p->old_block_size); |
2284 | blkdev_put(p->bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL); | |
1da177e4 | 2285 | } |
4cd3bb10 | 2286 | destroy_swap_extents(p); |
e8e6c2ec | 2287 | swap_cgroup_swapoff(p->type); |
5d337b91 | 2288 | spin_lock(&swap_lock); |
1da177e4 | 2289 | p->swap_file = NULL; |
1da177e4 | 2290 | p->flags = 0; |
5d337b91 | 2291 | spin_unlock(&swap_lock); |
1da177e4 | 2292 | vfree(swap_map); |
52c50567 | 2293 | if (swap_file) { |
2130781e | 2294 | if (inode && S_ISREG(inode->i_mode)) { |
52c50567 | 2295 | mutex_unlock(&inode->i_mutex); |
2130781e CEB |
2296 | inode = NULL; |
2297 | } | |
1da177e4 | 2298 | filp_close(swap_file, NULL); |
52c50567 | 2299 | } |
1da177e4 LT |
2300 | out: |
2301 | if (page && !IS_ERR(page)) { | |
2302 | kunmap(page); | |
2303 | page_cache_release(page); | |
2304 | } | |
2305 | if (name) | |
2306 | putname(name); | |
9b01c350 | 2307 | if (inode && S_ISREG(inode->i_mode)) |
1b1dcc1b | 2308 | mutex_unlock(&inode->i_mutex); |
1da177e4 LT |
2309 | return error; |
2310 | } | |
6fa3eb70 | 2311 | EXPORT_SYMBOL_GPL(sys_swapon); |
1da177e4 LT |
2312 | |
2313 | void si_swapinfo(struct sysinfo *val) | |
2314 | { | |
efa90a98 | 2315 | unsigned int type; |
1da177e4 LT |
2316 | unsigned long nr_to_be_unused = 0; |
2317 | ||
5d337b91 | 2318 | spin_lock(&swap_lock); |
efa90a98 HD |
2319 | for (type = 0; type < nr_swapfiles; type++) { |
2320 | struct swap_info_struct *si = swap_info[type]; | |
2321 | ||
2322 | if ((si->flags & SWP_USED) && !(si->flags & SWP_WRITEOK)) | |
2323 | nr_to_be_unused += si->inuse_pages; | |
1da177e4 | 2324 | } |
ec8acf20 | 2325 | val->freeswap = atomic_long_read(&nr_swap_pages) + nr_to_be_unused; |
1da177e4 | 2326 | val->totalswap = total_swap_pages + nr_to_be_unused; |
5d337b91 | 2327 | spin_unlock(&swap_lock); |
1da177e4 | 2328 | } |
6fa3eb70 | 2329 | EXPORT_SYMBOL_GPL(si_swapinfo); |
1da177e4 LT |
2330 | |
2331 | /* | |
2332 | * Verify that a swap entry is valid and increment its swap map count. | |
2333 | * | |
355cfa73 KH |
2334 | * Returns error code in following case. |
2335 | * - success -> 0 | |
2336 | * - swp_entry is invalid -> EINVAL | |
2337 | * - swp_entry is migration entry -> EINVAL | |
2338 | * - swap-cache reference is requested but there is already one. -> EEXIST | |
2339 | * - swap-cache reference is requested but the entry is not used. -> ENOENT | |
570a335b | 2340 | * - swap-mapped reference requested but needs continued swap count. -> ENOMEM |
1da177e4 | 2341 | */ |
8d69aaee | 2342 | static int __swap_duplicate(swp_entry_t entry, unsigned char usage) |
1da177e4 | 2343 | { |
73c34b6a | 2344 | struct swap_info_struct *p; |
1da177e4 | 2345 | unsigned long offset, type; |
8d69aaee HD |
2346 | unsigned char count; |
2347 | unsigned char has_cache; | |
253d553b | 2348 | int err = -EINVAL; |
1da177e4 | 2349 | |
a7420aa5 | 2350 | if (non_swap_entry(entry)) |
253d553b | 2351 | goto out; |
0697212a | 2352 | |
1da177e4 LT |
2353 | type = swp_type(entry); |
2354 | if (type >= nr_swapfiles) | |
2355 | goto bad_file; | |
efa90a98 | 2356 | p = swap_info[type]; |
1da177e4 LT |
2357 | offset = swp_offset(entry); |
2358 | ||
ec8acf20 | 2359 | spin_lock(&p->lock); |
355cfa73 KH |
2360 | if (unlikely(offset >= p->max)) |
2361 | goto unlock_out; | |
2362 | ||
253d553b HD |
2363 | count = p->swap_map[offset]; |
2364 | has_cache = count & SWAP_HAS_CACHE; | |
2365 | count &= ~SWAP_HAS_CACHE; | |
2366 | err = 0; | |
355cfa73 | 2367 | |
253d553b | 2368 | if (usage == SWAP_HAS_CACHE) { |
355cfa73 KH |
2369 | |
2370 | /* set SWAP_HAS_CACHE if there is no cache and entry is used */ | |
253d553b HD |
2371 | if (!has_cache && count) |
2372 | has_cache = SWAP_HAS_CACHE; | |
2373 | else if (has_cache) /* someone else added cache */ | |
2374 | err = -EEXIST; | |
2375 | else /* no users remaining */ | |
2376 | err = -ENOENT; | |
355cfa73 KH |
2377 | |
2378 | } else if (count || has_cache) { | |
253d553b | 2379 | |
570a335b HD |
2380 | if ((count & ~COUNT_CONTINUED) < SWAP_MAP_MAX) |
2381 | count += usage; | |
2382 | else if ((count & ~COUNT_CONTINUED) > SWAP_MAP_MAX) | |
253d553b | 2383 | err = -EINVAL; |
570a335b HD |
2384 | else if (swap_count_continued(p, offset, count)) |
2385 | count = COUNT_CONTINUED; | |
2386 | else | |
2387 | err = -ENOMEM; | |
355cfa73 | 2388 | } else |
253d553b HD |
2389 | err = -ENOENT; /* unused swap entry */ |
2390 | ||
2391 | p->swap_map[offset] = count | has_cache; | |
2392 | ||
355cfa73 | 2393 | unlock_out: |
ec8acf20 | 2394 | spin_unlock(&p->lock); |
1da177e4 | 2395 | out: |
253d553b | 2396 | return err; |
1da177e4 LT |
2397 | |
2398 | bad_file: | |
2399 | printk(KERN_ERR "swap_dup: %s%08lx\n", Bad_file, entry.val); | |
2400 | goto out; | |
2401 | } | |
253d553b | 2402 | |
aaa46865 HD |
2403 | /* |
2404 | * Help swapoff by noting that swap entry belongs to shmem/tmpfs | |
2405 | * (in which case its reference count is never incremented). | |
2406 | */ | |
2407 | void swap_shmem_alloc(swp_entry_t entry) | |
2408 | { | |
2409 | __swap_duplicate(entry, SWAP_MAP_SHMEM); | |
2410 | } | |
2411 | ||
355cfa73 | 2412 | /* |
08259d58 HD |
2413 | * Increase reference count of swap entry by 1. |
2414 | * Returns 0 for success, or -ENOMEM if a swap_count_continuation is required | |
2415 | * but could not be atomically allocated. Returns 0, just as if it succeeded, | |
2416 | * if __swap_duplicate() fails for another reason (-EINVAL or -ENOENT), which | |
2417 | * might occur if a page table entry has got corrupted. | |
355cfa73 | 2418 | */ |
570a335b | 2419 | int swap_duplicate(swp_entry_t entry) |
355cfa73 | 2420 | { |
570a335b HD |
2421 | int err = 0; |
2422 | ||
2423 | while (!err && __swap_duplicate(entry, 1) == -ENOMEM) | |
2424 | err = add_swap_count_continuation(entry, GFP_ATOMIC); | |
2425 | return err; | |
355cfa73 | 2426 | } |
1da177e4 | 2427 | |
cb4b86ba | 2428 | /* |
355cfa73 KH |
2429 | * @entry: swap entry for which we allocate swap cache. |
2430 | * | |
73c34b6a | 2431 | * Called when allocating swap cache for existing swap entry, |
355cfa73 KH |
2432 | * This can return error codes. Returns 0 at success. |
2433 | * -EBUSY means there is a swap cache. | |
2434 | * Note: return code is different from swap_duplicate(). | |
cb4b86ba KH |
2435 | */ |
2436 | int swapcache_prepare(swp_entry_t entry) | |
2437 | { | |
253d553b | 2438 | return __swap_duplicate(entry, SWAP_HAS_CACHE); |
cb4b86ba KH |
2439 | } |
2440 | ||
f981c595 MG |
2441 | struct swap_info_struct *page_swap_info(struct page *page) |
2442 | { | |
2443 | swp_entry_t swap = { .val = page_private(page) }; | |
2444 | BUG_ON(!PageSwapCache(page)); | |
2445 | return swap_info[swp_type(swap)]; | |
2446 | } | |
2447 | ||
2448 | /* | |
2449 | * out-of-line __page_file_ methods to avoid include hell. | |
2450 | */ | |
2451 | struct address_space *__page_file_mapping(struct page *page) | |
2452 | { | |
2453 | VM_BUG_ON(!PageSwapCache(page)); | |
2454 | return page_swap_info(page)->swap_file->f_mapping; | |
2455 | } | |
2456 | EXPORT_SYMBOL_GPL(__page_file_mapping); | |
2457 | ||
2458 | pgoff_t __page_file_index(struct page *page) | |
2459 | { | |
2460 | swp_entry_t swap = { .val = page_private(page) }; | |
2461 | VM_BUG_ON(!PageSwapCache(page)); | |
2462 | return swp_offset(swap); | |
2463 | } | |
6fa3eb70 | 2464 | |
f981c595 MG |
2465 | EXPORT_SYMBOL_GPL(__page_file_index); |
2466 | ||
6fa3eb70 S |
2467 | struct swap_info_struct *get_swap_info_struct(unsigned type) |
2468 | { | |
2469 | return swap_info[type]; | |
2470 | } | |
2471 | EXPORT_SYMBOL_GPL(get_swap_info_struct); | |
2472 | ||
570a335b HD |
2473 | /* |
2474 | * add_swap_count_continuation - called when a swap count is duplicated | |
2475 | * beyond SWAP_MAP_MAX, it allocates a new page and links that to the entry's | |
2476 | * page of the original vmalloc'ed swap_map, to hold the continuation count | |
2477 | * (for that entry and for its neighbouring PAGE_SIZE swap entries). Called | |
2478 | * again when count is duplicated beyond SWAP_MAP_MAX * SWAP_CONT_MAX, etc. | |
2479 | * | |
2480 | * These continuation pages are seldom referenced: the common paths all work | |
2481 | * on the original swap_map, only referring to a continuation page when the | |
2482 | * low "digit" of a count is incremented or decremented through SWAP_MAP_MAX. | |
2483 | * | |
2484 | * add_swap_count_continuation(, GFP_ATOMIC) can be called while holding | |
2485 | * page table locks; if it fails, add_swap_count_continuation(, GFP_KERNEL) | |
2486 | * can be called after dropping locks. | |
2487 | */ | |
2488 | int add_swap_count_continuation(swp_entry_t entry, gfp_t gfp_mask) | |
2489 | { | |
2490 | struct swap_info_struct *si; | |
2491 | struct page *head; | |
2492 | struct page *page; | |
2493 | struct page *list_page; | |
2494 | pgoff_t offset; | |
2495 | unsigned char count; | |
2496 | ||
2497 | /* | |
2498 | * When debugging, it's easier to use __GFP_ZERO here; but it's better | |
2499 | * for latency not to zero a page while GFP_ATOMIC and holding locks. | |
2500 | */ | |
2501 | page = alloc_page(gfp_mask | __GFP_HIGHMEM); | |
2502 | ||
2503 | si = swap_info_get(entry); | |
2504 | if (!si) { | |
2505 | /* | |
2506 | * An acceptable race has occurred since the failing | |
2507 | * __swap_duplicate(): the swap entry has been freed, | |
2508 | * perhaps even the whole swap_map cleared for swapoff. | |
2509 | */ | |
2510 | goto outer; | |
2511 | } | |
2512 | ||
2513 | offset = swp_offset(entry); | |
2514 | count = si->swap_map[offset] & ~SWAP_HAS_CACHE; | |
2515 | ||
2516 | if ((count & ~COUNT_CONTINUED) != SWAP_MAP_MAX) { | |
2517 | /* | |
2518 | * The higher the swap count, the more likely it is that tasks | |
2519 | * will race to add swap count continuation: we need to avoid | |
2520 | * over-provisioning. | |
2521 | */ | |
2522 | goto out; | |
2523 | } | |
2524 | ||
2525 | if (!page) { | |
ec8acf20 | 2526 | spin_unlock(&si->lock); |
570a335b HD |
2527 | return -ENOMEM; |
2528 | } | |
2529 | ||
2530 | /* | |
2531 | * We are fortunate that although vmalloc_to_page uses pte_offset_map, | |
2532 | * no architecture is using highmem pages for kernel pagetables: so it | |
2533 | * will not corrupt the GFP_ATOMIC caller's atomic pagetable kmaps. | |
2534 | */ | |
2535 | head = vmalloc_to_page(si->swap_map + offset); | |
2536 | offset &= ~PAGE_MASK; | |
2537 | ||
2538 | /* | |
2539 | * Page allocation does not initialize the page's lru field, | |
2540 | * but it does always reset its private field. | |
2541 | */ | |
2542 | if (!page_private(head)) { | |
2543 | BUG_ON(count & COUNT_CONTINUED); | |
2544 | INIT_LIST_HEAD(&head->lru); | |
2545 | set_page_private(head, SWP_CONTINUED); | |
2546 | si->flags |= SWP_CONTINUED; | |
2547 | } | |
2548 | ||
2549 | list_for_each_entry(list_page, &head->lru, lru) { | |
2550 | unsigned char *map; | |
2551 | ||
2552 | /* | |
2553 | * If the previous map said no continuation, but we've found | |
2554 | * a continuation page, free our allocation and use this one. | |
2555 | */ | |
2556 | if (!(count & COUNT_CONTINUED)) | |
2557 | goto out; | |
2558 | ||
9b04c5fe | 2559 | map = kmap_atomic(list_page) + offset; |
570a335b | 2560 | count = *map; |
9b04c5fe | 2561 | kunmap_atomic(map); |
570a335b HD |
2562 | |
2563 | /* | |
2564 | * If this continuation count now has some space in it, | |
2565 | * free our allocation and use this one. | |
2566 | */ | |
2567 | if ((count & ~COUNT_CONTINUED) != SWAP_CONT_MAX) | |
2568 | goto out; | |
2569 | } | |
2570 | ||
2571 | list_add_tail(&page->lru, &head->lru); | |
2572 | page = NULL; /* now it's attached, don't free it */ | |
2573 | out: | |
ec8acf20 | 2574 | spin_unlock(&si->lock); |
570a335b HD |
2575 | outer: |
2576 | if (page) | |
2577 | __free_page(page); | |
2578 | return 0; | |
2579 | } | |
2580 | ||
2581 | /* | |
2582 | * swap_count_continued - when the original swap_map count is incremented | |
2583 | * from SWAP_MAP_MAX, check if there is already a continuation page to carry | |
2584 | * into, carry if so, or else fail until a new continuation page is allocated; | |
2585 | * when the original swap_map count is decremented from 0 with continuation, | |
2586 | * borrow from the continuation and report whether it still holds more. | |
2587 | * Called while __swap_duplicate() or swap_entry_free() holds swap_lock. | |
2588 | */ | |
2589 | static bool swap_count_continued(struct swap_info_struct *si, | |
2590 | pgoff_t offset, unsigned char count) | |
2591 | { | |
2592 | struct page *head; | |
2593 | struct page *page; | |
2594 | unsigned char *map; | |
2595 | ||
2596 | head = vmalloc_to_page(si->swap_map + offset); | |
2597 | if (page_private(head) != SWP_CONTINUED) { | |
2598 | BUG_ON(count & COUNT_CONTINUED); | |
2599 | return false; /* need to add count continuation */ | |
2600 | } | |
2601 | ||
2602 | offset &= ~PAGE_MASK; | |
2603 | page = list_entry(head->lru.next, struct page, lru); | |
9b04c5fe | 2604 | map = kmap_atomic(page) + offset; |
570a335b HD |
2605 | |
2606 | if (count == SWAP_MAP_MAX) /* initial increment from swap_map */ | |
2607 | goto init_map; /* jump over SWAP_CONT_MAX checks */ | |
2608 | ||
2609 | if (count == (SWAP_MAP_MAX | COUNT_CONTINUED)) { /* incrementing */ | |
2610 | /* | |
2611 | * Think of how you add 1 to 999 | |
2612 | */ | |
2613 | while (*map == (SWAP_CONT_MAX | COUNT_CONTINUED)) { | |
9b04c5fe | 2614 | kunmap_atomic(map); |
570a335b HD |
2615 | page = list_entry(page->lru.next, struct page, lru); |
2616 | BUG_ON(page == head); | |
9b04c5fe | 2617 | map = kmap_atomic(page) + offset; |
570a335b HD |
2618 | } |
2619 | if (*map == SWAP_CONT_MAX) { | |
9b04c5fe | 2620 | kunmap_atomic(map); |
570a335b HD |
2621 | page = list_entry(page->lru.next, struct page, lru); |
2622 | if (page == head) | |
2623 | return false; /* add count continuation */ | |
9b04c5fe | 2624 | map = kmap_atomic(page) + offset; |
570a335b HD |
2625 | init_map: *map = 0; /* we didn't zero the page */ |
2626 | } | |
2627 | *map += 1; | |
9b04c5fe | 2628 | kunmap_atomic(map); |
570a335b HD |
2629 | page = list_entry(page->lru.prev, struct page, lru); |
2630 | while (page != head) { | |
9b04c5fe | 2631 | map = kmap_atomic(page) + offset; |
570a335b | 2632 | *map = COUNT_CONTINUED; |
9b04c5fe | 2633 | kunmap_atomic(map); |
570a335b HD |
2634 | page = list_entry(page->lru.prev, struct page, lru); |
2635 | } | |
2636 | return true; /* incremented */ | |
2637 | ||
2638 | } else { /* decrementing */ | |
2639 | /* | |
2640 | * Think of how you subtract 1 from 1000 | |
2641 | */ | |
2642 | BUG_ON(count != COUNT_CONTINUED); | |
2643 | while (*map == COUNT_CONTINUED) { | |
9b04c5fe | 2644 | kunmap_atomic(map); |
570a335b HD |
2645 | page = list_entry(page->lru.next, struct page, lru); |
2646 | BUG_ON(page == head); | |
9b04c5fe | 2647 | map = kmap_atomic(page) + offset; |
570a335b HD |
2648 | } |
2649 | BUG_ON(*map == 0); | |
2650 | *map -= 1; | |
2651 | if (*map == 0) | |
2652 | count = 0; | |
9b04c5fe | 2653 | kunmap_atomic(map); |
570a335b HD |
2654 | page = list_entry(page->lru.prev, struct page, lru); |
2655 | while (page != head) { | |
9b04c5fe | 2656 | map = kmap_atomic(page) + offset; |
570a335b HD |
2657 | *map = SWAP_CONT_MAX | count; |
2658 | count = COUNT_CONTINUED; | |
9b04c5fe | 2659 | kunmap_atomic(map); |
570a335b HD |
2660 | page = list_entry(page->lru.prev, struct page, lru); |
2661 | } | |
2662 | return count == COUNT_CONTINUED; | |
2663 | } | |
2664 | } | |
2665 | ||
2666 | /* | |
2667 | * free_swap_count_continuations - swapoff free all the continuation pages | |
2668 | * appended to the swap_map, after swap_map is quiesced, before vfree'ing it. | |
2669 | */ | |
2670 | static void free_swap_count_continuations(struct swap_info_struct *si) | |
2671 | { | |
2672 | pgoff_t offset; | |
2673 | ||
2674 | for (offset = 0; offset < si->max; offset += PAGE_SIZE) { | |
2675 | struct page *head; | |
2676 | head = vmalloc_to_page(si->swap_map + offset); | |
2677 | if (page_private(head)) { | |
2678 | struct list_head *this, *next; | |
2679 | list_for_each_safe(this, next, &head->lru) { | |
2680 | struct page *page; | |
2681 | page = list_entry(this, struct page, lru); | |
2682 | list_del(this); | |
2683 | __free_page(page); | |
2684 | } | |
2685 | } | |
2686 | } | |
2687 | } |