Root/
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 | |
8 | #include <linux/mm.h> |
9 | #include <linux/hugetlb.h> |
10 | #include <linux/mman.h> |
11 | #include <linux/slab.h> |
12 | #include <linux/kernel_stat.h> |
13 | #include <linux/swap.h> |
14 | #include <linux/vmalloc.h> |
15 | #include <linux/pagemap.h> |
16 | #include <linux/namei.h> |
17 | #include <linux/shm.h> |
18 | #include <linux/blkdev.h> |
19 | #include <linux/random.h> |
20 | #include <linux/writeback.h> |
21 | #include <linux/proc_fs.h> |
22 | #include <linux/seq_file.h> |
23 | #include <linux/init.h> |
24 | #include <linux/module.h> |
25 | #include <linux/ksm.h> |
26 | #include <linux/rmap.h> |
27 | #include <linux/security.h> |
28 | #include <linux/backing-dev.h> |
29 | #include <linux/mutex.h> |
30 | #include <linux/capability.h> |
31 | #include <linux/syscalls.h> |
32 | #include <linux/memcontrol.h> |
33 | #include <linux/poll.h> |
34 | |
35 | #include <asm/pgtable.h> |
36 | #include <asm/tlbflush.h> |
37 | #include <linux/swapops.h> |
38 | #include <linux/page_cgroup.h> |
39 | |
40 | static bool swap_count_continued(struct swap_info_struct *, pgoff_t, |
41 | unsigned char); |
42 | static void free_swap_count_continuations(struct swap_info_struct *); |
43 | static sector_t map_swap_entry(swp_entry_t, struct block_device**); |
44 | |
45 | static DEFINE_SPINLOCK(swap_lock); |
46 | static unsigned int nr_swapfiles; |
47 | long nr_swap_pages; |
48 | long total_swap_pages; |
49 | static int least_priority; |
50 | |
51 | static const char Bad_file[] = "Bad swap file entry "; |
52 | static const char Unused_file[] = "Unused swap file entry "; |
53 | static const char Bad_offset[] = "Bad swap offset entry "; |
54 | static const char Unused_offset[] = "Unused swap offset entry "; |
55 | |
56 | static struct swap_list_t swap_list = {-1, -1}; |
57 | |
58 | static struct swap_info_struct *swap_info[MAX_SWAPFILES]; |
59 | |
60 | static DEFINE_MUTEX(swapon_mutex); |
61 | |
62 | static DECLARE_WAIT_QUEUE_HEAD(proc_poll_wait); |
63 | /* Activity counter to indicate that a swapon or swapoff has occurred */ |
64 | static atomic_t proc_poll_event = ATOMIC_INIT(0); |
65 | |
66 | static inline unsigned char swap_count(unsigned char ent) |
67 | { |
68 | return ent & ~SWAP_HAS_CACHE; /* may include SWAP_HAS_CONT flag */ |
69 | } |
70 | |
71 | /* returns 1 if swap entry is freed */ |
72 | static int |
73 | __try_to_reclaim_swap(struct swap_info_struct *si, unsigned long offset) |
74 | { |
75 | swp_entry_t entry = swp_entry(si->type, offset); |
76 | struct page *page; |
77 | int ret = 0; |
78 | |
79 | page = find_get_page(&swapper_space, entry.val); |
80 | if (!page) |
81 | return 0; |
82 | /* |
83 | * This function is called from scan_swap_map() and it's called |
84 | * by vmscan.c at reclaiming pages. So, we hold a lock on a page, here. |
85 | * We have to use trylock for avoiding deadlock. This is a special |
86 | * case and you should use try_to_free_swap() with explicit lock_page() |
87 | * in usual operations. |
88 | */ |
89 | if (trylock_page(page)) { |
90 | ret = try_to_free_swap(page); |
91 | unlock_page(page); |
92 | } |
93 | page_cache_release(page); |
94 | return ret; |
95 | } |
96 | |
97 | /* |
98 | * We need this because the bdev->unplug_fn can sleep and we cannot |
99 | * hold swap_lock while calling the unplug_fn. And swap_lock |
100 | * cannot be turned into a mutex. |
101 | */ |
102 | static DECLARE_RWSEM(swap_unplug_sem); |
103 | |
104 | void swap_unplug_io_fn(struct backing_dev_info *unused_bdi, struct page *page) |
105 | { |
106 | swp_entry_t entry; |
107 | |
108 | down_read(&swap_unplug_sem); |
109 | entry.val = page_private(page); |
110 | if (PageSwapCache(page)) { |
111 | struct block_device *bdev = swap_info[swp_type(entry)]->bdev; |
112 | struct backing_dev_info *bdi; |
113 | |
114 | /* |
115 | * If the page is removed from swapcache from under us (with a |
116 | * racy try_to_unuse/swapoff) we need an additional reference |
117 | * count to avoid reading garbage from page_private(page) above. |
118 | * If the WARN_ON triggers during a swapoff it maybe the race |
119 | * condition and it's harmless. However if it triggers without |
120 | * swapoff it signals a problem. |
121 | */ |
122 | WARN_ON(page_count(page) <= 1); |
123 | |
124 | bdi = bdev->bd_inode->i_mapping->backing_dev_info; |
125 | blk_run_backing_dev(bdi, page); |
126 | } |
127 | up_read(&swap_unplug_sem); |
128 | } |
129 | |
130 | /* |
131 | * swapon tell device that all the old swap contents can be discarded, |
132 | * to allow the swap device to optimize its wear-levelling. |
133 | */ |
134 | static int discard_swap(struct swap_info_struct *si) |
135 | { |
136 | struct swap_extent *se; |
137 | sector_t start_block; |
138 | sector_t nr_blocks; |
139 | int err = 0; |
140 | |
141 | /* Do not discard the swap header page! */ |
142 | se = &si->first_swap_extent; |
143 | start_block = (se->start_block + 1) << (PAGE_SHIFT - 9); |
144 | nr_blocks = ((sector_t)se->nr_pages - 1) << (PAGE_SHIFT - 9); |
145 | if (nr_blocks) { |
146 | err = blkdev_issue_discard(si->bdev, start_block, |
147 | nr_blocks, GFP_KERNEL, 0); |
148 | if (err) |
149 | return err; |
150 | cond_resched(); |
151 | } |
152 | |
153 | list_for_each_entry(se, &si->first_swap_extent.list, list) { |
154 | start_block = se->start_block << (PAGE_SHIFT - 9); |
155 | nr_blocks = (sector_t)se->nr_pages << (PAGE_SHIFT - 9); |
156 | |
157 | err = blkdev_issue_discard(si->bdev, start_block, |
158 | nr_blocks, GFP_KERNEL, 0); |
159 | if (err) |
160 | break; |
161 | |
162 | cond_resched(); |
163 | } |
164 | return err; /* That will often be -EOPNOTSUPP */ |
165 | } |
166 | |
167 | /* |
168 | * swap allocation tell device that a cluster of swap can now be discarded, |
169 | * to allow the swap device to optimize its wear-levelling. |
170 | */ |
171 | static void discard_swap_cluster(struct swap_info_struct *si, |
172 | pgoff_t start_page, pgoff_t nr_pages) |
173 | { |
174 | struct swap_extent *se = si->curr_swap_extent; |
175 | int found_extent = 0; |
176 | |
177 | while (nr_pages) { |
178 | struct list_head *lh; |
179 | |
180 | if (se->start_page <= start_page && |
181 | start_page < se->start_page + se->nr_pages) { |
182 | pgoff_t offset = start_page - se->start_page; |
183 | sector_t start_block = se->start_block + offset; |
184 | sector_t nr_blocks = se->nr_pages - offset; |
185 | |
186 | if (nr_blocks > nr_pages) |
187 | nr_blocks = nr_pages; |
188 | start_page += nr_blocks; |
189 | nr_pages -= nr_blocks; |
190 | |
191 | if (!found_extent++) |
192 | si->curr_swap_extent = se; |
193 | |
194 | start_block <<= PAGE_SHIFT - 9; |
195 | nr_blocks <<= PAGE_SHIFT - 9; |
196 | if (blkdev_issue_discard(si->bdev, start_block, |
197 | nr_blocks, GFP_NOIO, 0)) |
198 | break; |
199 | } |
200 | |
201 | lh = se->list.next; |
202 | se = list_entry(lh, struct swap_extent, list); |
203 | } |
204 | } |
205 | |
206 | static int wait_for_discard(void *word) |
207 | { |
208 | schedule(); |
209 | return 0; |
210 | } |
211 | |
212 | #define SWAPFILE_CLUSTER 256 |
213 | #define LATENCY_LIMIT 256 |
214 | |
215 | static inline unsigned long scan_swap_map(struct swap_info_struct *si, |
216 | unsigned char usage) |
217 | { |
218 | unsigned long offset; |
219 | unsigned long scan_base; |
220 | unsigned long last_in_cluster = 0; |
221 | int latency_ration = LATENCY_LIMIT; |
222 | int found_free_cluster = 0; |
223 | |
224 | /* |
225 | * We try to cluster swap pages by allocating them sequentially |
226 | * in swap. Once we've allocated SWAPFILE_CLUSTER pages this |
227 | * way, however, we resort to first-free allocation, starting |
228 | * a new cluster. This prevents us from scattering swap pages |
229 | * all over the entire swap partition, so that we reduce |
230 | * overall disk seek times between swap pages. -- sct |
231 | * But we do now try to find an empty cluster. -Andrea |
232 | * And we let swap pages go all over an SSD partition. Hugh |
233 | */ |
234 | |
235 | si->flags += SWP_SCANNING; |
236 | scan_base = offset = si->cluster_next; |
237 | |
238 | if (unlikely(!si->cluster_nr--)) { |
239 | if (si->pages - si->inuse_pages < SWAPFILE_CLUSTER) { |
240 | si->cluster_nr = SWAPFILE_CLUSTER - 1; |
241 | goto checks; |
242 | } |
243 | if (si->flags & SWP_DISCARDABLE) { |
244 | /* |
245 | * Start range check on racing allocations, in case |
246 | * they overlap the cluster we eventually decide on |
247 | * (we scan without swap_lock to allow preemption). |
248 | * It's hardly conceivable that cluster_nr could be |
249 | * wrapped during our scan, but don't depend on it. |
250 | */ |
251 | if (si->lowest_alloc) |
252 | goto checks; |
253 | si->lowest_alloc = si->max; |
254 | si->highest_alloc = 0; |
255 | } |
256 | spin_unlock(&swap_lock); |
257 | |
258 | /* |
259 | * If seek is expensive, start searching for new cluster from |
260 | * start of partition, to minimize the span of allocated swap. |
261 | * But if seek is cheap, search from our current position, so |
262 | * that swap is allocated from all over the partition: if the |
263 | * Flash Translation Layer only remaps within limited zones, |
264 | * we don't want to wear out the first zone too quickly. |
265 | */ |
266 | if (!(si->flags & SWP_SOLIDSTATE)) |
267 | scan_base = offset = si->lowest_bit; |
268 | last_in_cluster = offset + SWAPFILE_CLUSTER - 1; |
269 | |
270 | /* Locate the first empty (unaligned) cluster */ |
271 | for (; last_in_cluster <= si->highest_bit; offset++) { |
272 | if (si->swap_map[offset]) |
273 | last_in_cluster = offset + SWAPFILE_CLUSTER; |
274 | else if (offset == last_in_cluster) { |
275 | spin_lock(&swap_lock); |
276 | offset -= SWAPFILE_CLUSTER - 1; |
277 | si->cluster_next = offset; |
278 | si->cluster_nr = SWAPFILE_CLUSTER - 1; |
279 | found_free_cluster = 1; |
280 | goto checks; |
281 | } |
282 | if (unlikely(--latency_ration < 0)) { |
283 | cond_resched(); |
284 | latency_ration = LATENCY_LIMIT; |
285 | } |
286 | } |
287 | |
288 | offset = si->lowest_bit; |
289 | last_in_cluster = offset + SWAPFILE_CLUSTER - 1; |
290 | |
291 | /* Locate the first empty (unaligned) cluster */ |
292 | for (; last_in_cluster < scan_base; offset++) { |
293 | if (si->swap_map[offset]) |
294 | last_in_cluster = offset + SWAPFILE_CLUSTER; |
295 | else if (offset == last_in_cluster) { |
296 | spin_lock(&swap_lock); |
297 | offset -= SWAPFILE_CLUSTER - 1; |
298 | si->cluster_next = offset; |
299 | si->cluster_nr = SWAPFILE_CLUSTER - 1; |
300 | found_free_cluster = 1; |
301 | goto checks; |
302 | } |
303 | if (unlikely(--latency_ration < 0)) { |
304 | cond_resched(); |
305 | latency_ration = LATENCY_LIMIT; |
306 | } |
307 | } |
308 | |
309 | offset = scan_base; |
310 | spin_lock(&swap_lock); |
311 | si->cluster_nr = SWAPFILE_CLUSTER - 1; |
312 | si->lowest_alloc = 0; |
313 | } |
314 | |
315 | checks: |
316 | if (!(si->flags & SWP_WRITEOK)) |
317 | goto no_page; |
318 | if (!si->highest_bit) |
319 | goto no_page; |
320 | if (offset > si->highest_bit) |
321 | scan_base = offset = si->lowest_bit; |
322 | |
323 | /* reuse swap entry of cache-only swap if not busy. */ |
324 | if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) { |
325 | int swap_was_freed; |
326 | spin_unlock(&swap_lock); |
327 | swap_was_freed = __try_to_reclaim_swap(si, offset); |
328 | spin_lock(&swap_lock); |
329 | /* entry was freed successfully, try to use this again */ |
330 | if (swap_was_freed) |
331 | goto checks; |
332 | goto scan; /* check next one */ |
333 | } |
334 | |
335 | if (si->swap_map[offset]) |
336 | goto scan; |
337 | |
338 | if (offset == si->lowest_bit) |
339 | si->lowest_bit++; |
340 | if (offset == si->highest_bit) |
341 | si->highest_bit--; |
342 | si->inuse_pages++; |
343 | if (si->inuse_pages == si->pages) { |
344 | si->lowest_bit = si->max; |
345 | si->highest_bit = 0; |
346 | } |
347 | si->swap_map[offset] = usage; |
348 | si->cluster_next = offset + 1; |
349 | si->flags -= SWP_SCANNING; |
350 | |
351 | if (si->lowest_alloc) { |
352 | /* |
353 | * Only set when SWP_DISCARDABLE, and there's a scan |
354 | * for a free cluster in progress or just completed. |
355 | */ |
356 | if (found_free_cluster) { |
357 | /* |
358 | * To optimize wear-levelling, discard the |
359 | * old data of the cluster, taking care not to |
360 | * discard any of its pages that have already |
361 | * been allocated by racing tasks (offset has |
362 | * already stepped over any at the beginning). |
363 | */ |
364 | if (offset < si->highest_alloc && |
365 | si->lowest_alloc <= last_in_cluster) |
366 | last_in_cluster = si->lowest_alloc - 1; |
367 | si->flags |= SWP_DISCARDING; |
368 | spin_unlock(&swap_lock); |
369 | |
370 | if (offset < last_in_cluster) |
371 | discard_swap_cluster(si, offset, |
372 | last_in_cluster - offset + 1); |
373 | |
374 | spin_lock(&swap_lock); |
375 | si->lowest_alloc = 0; |
376 | si->flags &= ~SWP_DISCARDING; |
377 | |
378 | smp_mb(); /* wake_up_bit advises this */ |
379 | wake_up_bit(&si->flags, ilog2(SWP_DISCARDING)); |
380 | |
381 | } else if (si->flags & SWP_DISCARDING) { |
382 | /* |
383 | * Delay using pages allocated by racing tasks |
384 | * until the whole discard has been issued. We |
385 | * could defer that delay until swap_writepage, |
386 | * but it's easier to keep this self-contained. |
387 | */ |
388 | spin_unlock(&swap_lock); |
389 | wait_on_bit(&si->flags, ilog2(SWP_DISCARDING), |
390 | wait_for_discard, TASK_UNINTERRUPTIBLE); |
391 | spin_lock(&swap_lock); |
392 | } else { |
393 | /* |
394 | * Note pages allocated by racing tasks while |
395 | * scan for a free cluster is in progress, so |
396 | * that its final discard can exclude them. |
397 | */ |
398 | if (offset < si->lowest_alloc) |
399 | si->lowest_alloc = offset; |
400 | if (offset > si->highest_alloc) |
401 | si->highest_alloc = offset; |
402 | } |
403 | } |
404 | return offset; |
405 | |
406 | scan: |
407 | spin_unlock(&swap_lock); |
408 | while (++offset <= si->highest_bit) { |
409 | if (!si->swap_map[offset]) { |
410 | spin_lock(&swap_lock); |
411 | goto checks; |
412 | } |
413 | if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) { |
414 | spin_lock(&swap_lock); |
415 | goto checks; |
416 | } |
417 | if (unlikely(--latency_ration < 0)) { |
418 | cond_resched(); |
419 | latency_ration = LATENCY_LIMIT; |
420 | } |
421 | } |
422 | offset = si->lowest_bit; |
423 | while (++offset < scan_base) { |
424 | if (!si->swap_map[offset]) { |
425 | spin_lock(&swap_lock); |
426 | goto checks; |
427 | } |
428 | if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) { |
429 | spin_lock(&swap_lock); |
430 | goto checks; |
431 | } |
432 | if (unlikely(--latency_ration < 0)) { |
433 | cond_resched(); |
434 | latency_ration = LATENCY_LIMIT; |
435 | } |
436 | } |
437 | spin_lock(&swap_lock); |
438 | |
439 | no_page: |
440 | si->flags -= SWP_SCANNING; |
441 | return 0; |
442 | } |
443 | |
444 | swp_entry_t get_swap_page(void) |
445 | { |
446 | struct swap_info_struct *si; |
447 | pgoff_t offset; |
448 | int type, next; |
449 | int wrapped = 0; |
450 | |
451 | spin_lock(&swap_lock); |
452 | if (nr_swap_pages <= 0) |
453 | goto noswap; |
454 | nr_swap_pages--; |
455 | |
456 | for (type = swap_list.next; type >= 0 && wrapped < 2; type = next) { |
457 | si = swap_info[type]; |
458 | next = si->next; |
459 | if (next < 0 || |
460 | (!wrapped && si->prio != swap_info[next]->prio)) { |
461 | next = swap_list.head; |
462 | wrapped++; |
463 | } |
464 | |
465 | if (!si->highest_bit) |
466 | continue; |
467 | if (!(si->flags & SWP_WRITEOK)) |
468 | continue; |
469 | |
470 | swap_list.next = next; |
471 | /* This is called for allocating swap entry for cache */ |
472 | offset = scan_swap_map(si, SWAP_HAS_CACHE); |
473 | if (offset) { |
474 | spin_unlock(&swap_lock); |
475 | return swp_entry(type, offset); |
476 | } |
477 | next = swap_list.next; |
478 | } |
479 | |
480 | nr_swap_pages++; |
481 | noswap: |
482 | spin_unlock(&swap_lock); |
483 | return (swp_entry_t) {0}; |
484 | } |
485 | |
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 | |
492 | spin_lock(&swap_lock); |
493 | si = swap_info[type]; |
494 | if (si && (si->flags & SWP_WRITEOK)) { |
495 | nr_swap_pages--; |
496 | /* This is called for allocating swap entry, not cache */ |
497 | offset = scan_swap_map(si, 1); |
498 | if (offset) { |
499 | spin_unlock(&swap_lock); |
500 | return swp_entry(type, offset); |
501 | } |
502 | nr_swap_pages++; |
503 | } |
504 | spin_unlock(&swap_lock); |
505 | return (swp_entry_t) {0}; |
506 | } |
507 | |
508 | static struct swap_info_struct *swap_info_get(swp_entry_t entry) |
509 | { |
510 | struct swap_info_struct *p; |
511 | unsigned long offset, type; |
512 | |
513 | if (!entry.val) |
514 | goto out; |
515 | type = swp_type(entry); |
516 | if (type >= nr_swapfiles) |
517 | goto bad_nofile; |
518 | p = swap_info[type]; |
519 | if (!(p->flags & SWP_USED)) |
520 | goto bad_device; |
521 | offset = swp_offset(entry); |
522 | if (offset >= p->max) |
523 | goto bad_offset; |
524 | if (!p->swap_map[offset]) |
525 | goto bad_free; |
526 | spin_lock(&swap_lock); |
527 | return p; |
528 | |
529 | bad_free: |
530 | printk(KERN_ERR "swap_free: %s%08lx\n", Unused_offset, entry.val); |
531 | goto out; |
532 | bad_offset: |
533 | printk(KERN_ERR "swap_free: %s%08lx\n", Bad_offset, entry.val); |
534 | goto out; |
535 | bad_device: |
536 | printk(KERN_ERR "swap_free: %s%08lx\n", Unused_file, entry.val); |
537 | goto out; |
538 | bad_nofile: |
539 | printk(KERN_ERR "swap_free: %s%08lx\n", Bad_file, entry.val); |
540 | out: |
541 | return NULL; |
542 | } |
543 | |
544 | static unsigned char swap_entry_free(struct swap_info_struct *p, |
545 | swp_entry_t entry, unsigned char usage) |
546 | { |
547 | unsigned long offset = swp_offset(entry); |
548 | unsigned char count; |
549 | unsigned char has_cache; |
550 | |
551 | count = p->swap_map[offset]; |
552 | has_cache = count & SWAP_HAS_CACHE; |
553 | count &= ~SWAP_HAS_CACHE; |
554 | |
555 | if (usage == SWAP_HAS_CACHE) { |
556 | VM_BUG_ON(!has_cache); |
557 | has_cache = 0; |
558 | } else if (count == SWAP_MAP_SHMEM) { |
559 | /* |
560 | * Or we could insist on shmem.c using a special |
561 | * swap_shmem_free() and free_shmem_swap_and_cache()... |
562 | */ |
563 | count = 0; |
564 | } else if ((count & ~COUNT_CONTINUED) <= SWAP_MAP_MAX) { |
565 | if (count == COUNT_CONTINUED) { |
566 | if (swap_count_continued(p, offset, count)) |
567 | count = SWAP_MAP_MAX | COUNT_CONTINUED; |
568 | else |
569 | count = SWAP_MAP_MAX; |
570 | } else |
571 | count--; |
572 | } |
573 | |
574 | if (!count) |
575 | mem_cgroup_uncharge_swap(entry); |
576 | |
577 | usage = count | has_cache; |
578 | p->swap_map[offset] = usage; |
579 | |
580 | /* free if no reference */ |
581 | if (!usage) { |
582 | struct gendisk *disk = p->bdev->bd_disk; |
583 | if (offset < p->lowest_bit) |
584 | p->lowest_bit = offset; |
585 | if (offset > p->highest_bit) |
586 | p->highest_bit = offset; |
587 | if (swap_list.next >= 0 && |
588 | p->prio > swap_info[swap_list.next]->prio) |
589 | swap_list.next = p->type; |
590 | nr_swap_pages++; |
591 | p->inuse_pages--; |
592 | if ((p->flags & SWP_BLKDEV) && |
593 | disk->fops->swap_slot_free_notify) |
594 | disk->fops->swap_slot_free_notify(p->bdev, offset); |
595 | } |
596 | |
597 | return usage; |
598 | } |
599 | |
600 | /* |
601 | * Caller has made sure that the swapdevice corresponding to entry |
602 | * is still around or has not been recycled. |
603 | */ |
604 | void swap_free(swp_entry_t entry) |
605 | { |
606 | struct swap_info_struct *p; |
607 | |
608 | p = swap_info_get(entry); |
609 | if (p) { |
610 | swap_entry_free(p, entry, 1); |
611 | spin_unlock(&swap_lock); |
612 | } |
613 | } |
614 | |
615 | /* |
616 | * Called after dropping swapcache to decrease refcnt to swap entries. |
617 | */ |
618 | void swapcache_free(swp_entry_t entry, struct page *page) |
619 | { |
620 | struct swap_info_struct *p; |
621 | unsigned char count; |
622 | |
623 | p = swap_info_get(entry); |
624 | if (p) { |
625 | count = swap_entry_free(p, entry, SWAP_HAS_CACHE); |
626 | if (page) |
627 | mem_cgroup_uncharge_swapcache(page, entry, count != 0); |
628 | spin_unlock(&swap_lock); |
629 | } |
630 | } |
631 | |
632 | /* |
633 | * How many references to page are currently swapped out? |
634 | * This does not give an exact answer when swap count is continued, |
635 | * but does include the high COUNT_CONTINUED flag to allow for that. |
636 | */ |
637 | static inline int page_swapcount(struct page *page) |
638 | { |
639 | int count = 0; |
640 | struct swap_info_struct *p; |
641 | swp_entry_t entry; |
642 | |
643 | entry.val = page_private(page); |
644 | p = swap_info_get(entry); |
645 | if (p) { |
646 | count = swap_count(p->swap_map[swp_offset(entry)]); |
647 | spin_unlock(&swap_lock); |
648 | } |
649 | return count; |
650 | } |
651 | |
652 | /* |
653 | * We can write to an anon page without COW if there are no other references |
654 | * to it. And as a side-effect, free up its swap: because the old content |
655 | * on disk will never be read, and seeking back there to write new content |
656 | * later would only waste time away from clustering. |
657 | */ |
658 | int reuse_swap_page(struct page *page) |
659 | { |
660 | int count; |
661 | |
662 | VM_BUG_ON(!PageLocked(page)); |
663 | if (unlikely(PageKsm(page))) |
664 | return 0; |
665 | count = page_mapcount(page); |
666 | if (count <= 1 && PageSwapCache(page)) { |
667 | count += page_swapcount(page); |
668 | if (count == 1 && !PageWriteback(page)) { |
669 | delete_from_swap_cache(page); |
670 | SetPageDirty(page); |
671 | } |
672 | } |
673 | return count <= 1; |
674 | } |
675 | |
676 | /* |
677 | * If swap is getting full, or if there are no more mappings of this page, |
678 | * then try_to_free_swap is called to free its swap space. |
679 | */ |
680 | int try_to_free_swap(struct page *page) |
681 | { |
682 | VM_BUG_ON(!PageLocked(page)); |
683 | |
684 | if (!PageSwapCache(page)) |
685 | return 0; |
686 | if (PageWriteback(page)) |
687 | return 0; |
688 | if (page_swapcount(page)) |
689 | return 0; |
690 | |
691 | /* |
692 | * Once hibernation has begun to create its image of memory, |
693 | * there's a danger that one of the calls to try_to_free_swap() |
694 | * - most probably a call from __try_to_reclaim_swap() while |
695 | * hibernation is allocating its own swap pages for the image, |
696 | * but conceivably even a call from memory reclaim - will free |
697 | * the swap from a page which has already been recorded in the |
698 | * image as a clean swapcache page, and then reuse its swap for |
699 | * another page of the image. On waking from hibernation, the |
700 | * original page might be freed under memory pressure, then |
701 | * later read back in from swap, now with the wrong data. |
702 | * |
703 | * Hibernation clears bits from gfp_allowed_mask to prevent |
704 | * memory reclaim from writing to disk, so check that here. |
705 | */ |
706 | if (!(gfp_allowed_mask & __GFP_IO)) |
707 | return 0; |
708 | |
709 | delete_from_swap_cache(page); |
710 | SetPageDirty(page); |
711 | return 1; |
712 | } |
713 | |
714 | /* |
715 | * Free the swap entry like above, but also try to |
716 | * free the page cache entry if it is the last user. |
717 | */ |
718 | int free_swap_and_cache(swp_entry_t entry) |
719 | { |
720 | struct swap_info_struct *p; |
721 | struct page *page = NULL; |
722 | |
723 | if (non_swap_entry(entry)) |
724 | return 1; |
725 | |
726 | p = swap_info_get(entry); |
727 | if (p) { |
728 | if (swap_entry_free(p, entry, 1) == SWAP_HAS_CACHE) { |
729 | page = find_get_page(&swapper_space, entry.val); |
730 | if (page && !trylock_page(page)) { |
731 | page_cache_release(page); |
732 | page = NULL; |
733 | } |
734 | } |
735 | spin_unlock(&swap_lock); |
736 | } |
737 | if (page) { |
738 | /* |
739 | * Not mapped elsewhere, or swap space full? Free it! |
740 | * Also recheck PageSwapCache now page is locked (above). |
741 | */ |
742 | if (PageSwapCache(page) && !PageWriteback(page) && |
743 | (!page_mapped(page) || vm_swap_full())) { |
744 | delete_from_swap_cache(page); |
745 | SetPageDirty(page); |
746 | } |
747 | unlock_page(page); |
748 | page_cache_release(page); |
749 | } |
750 | return p != NULL; |
751 | } |
752 | |
753 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR |
754 | /** |
755 | * mem_cgroup_count_swap_user - count the user of a swap entry |
756 | * @ent: the swap entry to be checked |
757 | * @pagep: the pointer for the swap cache page of the entry to be stored |
758 | * |
759 | * Returns the number of the user of the swap entry. The number is valid only |
760 | * for swaps of anonymous pages. |
761 | * If the entry is found on swap cache, the page is stored to pagep with |
762 | * refcount of it being incremented. |
763 | */ |
764 | int mem_cgroup_count_swap_user(swp_entry_t ent, struct page **pagep) |
765 | { |
766 | struct page *page; |
767 | struct swap_info_struct *p; |
768 | int count = 0; |
769 | |
770 | page = find_get_page(&swapper_space, ent.val); |
771 | if (page) |
772 | count += page_mapcount(page); |
773 | p = swap_info_get(ent); |
774 | if (p) { |
775 | count += swap_count(p->swap_map[swp_offset(ent)]); |
776 | spin_unlock(&swap_lock); |
777 | } |
778 | |
779 | *pagep = page; |
780 | return count; |
781 | } |
782 | #endif |
783 | |
784 | #ifdef CONFIG_HIBERNATION |
785 | /* |
786 | * Find the swap type that corresponds to given device (if any). |
787 | * |
788 | * @offset - number of the PAGE_SIZE-sized block of the device, starting |
789 | * from 0, in which the swap header is expected to be located. |
790 | * |
791 | * This is needed for the suspend to disk (aka swsusp). |
792 | */ |
793 | int swap_type_of(dev_t device, sector_t offset, struct block_device **bdev_p) |
794 | { |
795 | struct block_device *bdev = NULL; |
796 | int type; |
797 | |
798 | if (device) |
799 | bdev = bdget(device); |
800 | |
801 | spin_lock(&swap_lock); |
802 | for (type = 0; type < nr_swapfiles; type++) { |
803 | struct swap_info_struct *sis = swap_info[type]; |
804 | |
805 | if (!(sis->flags & SWP_WRITEOK)) |
806 | continue; |
807 | |
808 | if (!bdev) { |
809 | if (bdev_p) |
810 | *bdev_p = bdgrab(sis->bdev); |
811 | |
812 | spin_unlock(&swap_lock); |
813 | return type; |
814 | } |
815 | if (bdev == sis->bdev) { |
816 | struct swap_extent *se = &sis->first_swap_extent; |
817 | |
818 | if (se->start_block == offset) { |
819 | if (bdev_p) |
820 | *bdev_p = bdgrab(sis->bdev); |
821 | |
822 | spin_unlock(&swap_lock); |
823 | bdput(bdev); |
824 | return type; |
825 | } |
826 | } |
827 | } |
828 | spin_unlock(&swap_lock); |
829 | if (bdev) |
830 | bdput(bdev); |
831 | |
832 | return -ENODEV; |
833 | } |
834 | |
835 | /* |
836 | * Get the (PAGE_SIZE) block corresponding to given offset on the swapdev |
837 | * corresponding to given index in swap_info (swap type). |
838 | */ |
839 | sector_t swapdev_block(int type, pgoff_t offset) |
840 | { |
841 | struct block_device *bdev; |
842 | |
843 | if ((unsigned int)type >= nr_swapfiles) |
844 | return 0; |
845 | if (!(swap_info[type]->flags & SWP_WRITEOK)) |
846 | return 0; |
847 | return map_swap_entry(swp_entry(type, offset), &bdev); |
848 | } |
849 | |
850 | /* |
851 | * Return either the total number of swap pages of given type, or the number |
852 | * of free pages of that type (depending on @free) |
853 | * |
854 | * This is needed for software suspend |
855 | */ |
856 | unsigned int count_swap_pages(int type, int free) |
857 | { |
858 | unsigned int n = 0; |
859 | |
860 | spin_lock(&swap_lock); |
861 | if ((unsigned int)type < nr_swapfiles) { |
862 | struct swap_info_struct *sis = swap_info[type]; |
863 | |
864 | if (sis->flags & SWP_WRITEOK) { |
865 | n = sis->pages; |
866 | if (free) |
867 | n -= sis->inuse_pages; |
868 | } |
869 | } |
870 | spin_unlock(&swap_lock); |
871 | return n; |
872 | } |
873 | #endif /* CONFIG_HIBERNATION */ |
874 | |
875 | /* |
876 | * No need to decide whether this PTE shares the swap entry with others, |
877 | * just let do_wp_page work it out if a write is requested later - to |
878 | * force COW, vm_page_prot omits write permission from any private vma. |
879 | */ |
880 | static int unuse_pte(struct vm_area_struct *vma, pmd_t *pmd, |
881 | unsigned long addr, swp_entry_t entry, struct page *page) |
882 | { |
883 | struct mem_cgroup *ptr = NULL; |
884 | spinlock_t *ptl; |
885 | pte_t *pte; |
886 | int ret = 1; |
887 | |
888 | if (mem_cgroup_try_charge_swapin(vma->vm_mm, page, GFP_KERNEL, &ptr)) { |
889 | ret = -ENOMEM; |
890 | goto out_nolock; |
891 | } |
892 | |
893 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
894 | if (unlikely(!pte_same(*pte, swp_entry_to_pte(entry)))) { |
895 | if (ret > 0) |
896 | mem_cgroup_cancel_charge_swapin(ptr); |
897 | ret = 0; |
898 | goto out; |
899 | } |
900 | |
901 | dec_mm_counter(vma->vm_mm, MM_SWAPENTS); |
902 | inc_mm_counter(vma->vm_mm, MM_ANONPAGES); |
903 | get_page(page); |
904 | set_pte_at(vma->vm_mm, addr, pte, |
905 | pte_mkold(mk_pte(page, vma->vm_page_prot))); |
906 | page_add_anon_rmap(page, vma, addr); |
907 | mem_cgroup_commit_charge_swapin(page, ptr); |
908 | swap_free(entry); |
909 | /* |
910 | * Move the page to the active list so it is not |
911 | * immediately swapped out again after swapon. |
912 | */ |
913 | activate_page(page); |
914 | out: |
915 | pte_unmap_unlock(pte, ptl); |
916 | out_nolock: |
917 | return ret; |
918 | } |
919 | |
920 | static int unuse_pte_range(struct vm_area_struct *vma, pmd_t *pmd, |
921 | unsigned long addr, unsigned long end, |
922 | swp_entry_t entry, struct page *page) |
923 | { |
924 | pte_t swp_pte = swp_entry_to_pte(entry); |
925 | pte_t *pte; |
926 | int ret = 0; |
927 | |
928 | /* |
929 | * We don't actually need pte lock while scanning for swp_pte: since |
930 | * we hold page lock and mmap_sem, swp_pte cannot be inserted into the |
931 | * page table while we're scanning; though it could get zapped, and on |
932 | * some architectures (e.g. x86_32 with PAE) we might catch a glimpse |
933 | * of unmatched parts which look like swp_pte, so unuse_pte must |
934 | * recheck under pte lock. Scanning without pte lock lets it be |
935 | * preemptible whenever CONFIG_PREEMPT but not CONFIG_HIGHPTE. |
936 | */ |
937 | pte = pte_offset_map(pmd, addr); |
938 | do { |
939 | /* |
940 | * swapoff spends a _lot_ of time in this loop! |
941 | * Test inline before going to call unuse_pte. |
942 | */ |
943 | if (unlikely(pte_same(*pte, swp_pte))) { |
944 | pte_unmap(pte); |
945 | ret = unuse_pte(vma, pmd, addr, entry, page); |
946 | if (ret) |
947 | goto out; |
948 | pte = pte_offset_map(pmd, addr); |
949 | } |
950 | } while (pte++, addr += PAGE_SIZE, addr != end); |
951 | pte_unmap(pte - 1); |
952 | out: |
953 | return ret; |
954 | } |
955 | |
956 | static inline int unuse_pmd_range(struct vm_area_struct *vma, pud_t *pud, |
957 | unsigned long addr, unsigned long end, |
958 | swp_entry_t entry, struct page *page) |
959 | { |
960 | pmd_t *pmd; |
961 | unsigned long next; |
962 | int ret; |
963 | |
964 | pmd = pmd_offset(pud, addr); |
965 | do { |
966 | next = pmd_addr_end(addr, end); |
967 | if (pmd_none_or_clear_bad(pmd)) |
968 | continue; |
969 | ret = unuse_pte_range(vma, pmd, addr, next, entry, page); |
970 | if (ret) |
971 | return ret; |
972 | } while (pmd++, addr = next, addr != end); |
973 | return 0; |
974 | } |
975 | |
976 | static inline int unuse_pud_range(struct vm_area_struct *vma, pgd_t *pgd, |
977 | unsigned long addr, unsigned long end, |
978 | swp_entry_t entry, struct page *page) |
979 | { |
980 | pud_t *pud; |
981 | unsigned long next; |
982 | int ret; |
983 | |
984 | pud = pud_offset(pgd, addr); |
985 | do { |
986 | next = pud_addr_end(addr, end); |
987 | if (pud_none_or_clear_bad(pud)) |
988 | continue; |
989 | ret = unuse_pmd_range(vma, pud, addr, next, entry, page); |
990 | if (ret) |
991 | return ret; |
992 | } while (pud++, addr = next, addr != end); |
993 | return 0; |
994 | } |
995 | |
996 | static int unuse_vma(struct vm_area_struct *vma, |
997 | swp_entry_t entry, struct page *page) |
998 | { |
999 | pgd_t *pgd; |
1000 | unsigned long addr, end, next; |
1001 | int ret; |
1002 | |
1003 | if (page_anon_vma(page)) { |
1004 | addr = page_address_in_vma(page, vma); |
1005 | if (addr == -EFAULT) |
1006 | return 0; |
1007 | else |
1008 | end = addr + PAGE_SIZE; |
1009 | } else { |
1010 | addr = vma->vm_start; |
1011 | end = vma->vm_end; |
1012 | } |
1013 | |
1014 | pgd = pgd_offset(vma->vm_mm, addr); |
1015 | do { |
1016 | next = pgd_addr_end(addr, end); |
1017 | if (pgd_none_or_clear_bad(pgd)) |
1018 | continue; |
1019 | ret = unuse_pud_range(vma, pgd, addr, next, entry, page); |
1020 | if (ret) |
1021 | return ret; |
1022 | } while (pgd++, addr = next, addr != end); |
1023 | return 0; |
1024 | } |
1025 | |
1026 | static int unuse_mm(struct mm_struct *mm, |
1027 | swp_entry_t entry, struct page *page) |
1028 | { |
1029 | struct vm_area_struct *vma; |
1030 | int ret = 0; |
1031 | |
1032 | if (!down_read_trylock(&mm->mmap_sem)) { |
1033 | /* |
1034 | * Activate page so shrink_inactive_list is unlikely to unmap |
1035 | * its ptes while lock is dropped, so swapoff can make progress. |
1036 | */ |
1037 | activate_page(page); |
1038 | unlock_page(page); |
1039 | down_read(&mm->mmap_sem); |
1040 | lock_page(page); |
1041 | } |
1042 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
1043 | if (vma->anon_vma && (ret = unuse_vma(vma, entry, page))) |
1044 | break; |
1045 | } |
1046 | up_read(&mm->mmap_sem); |
1047 | return (ret < 0)? ret: 0; |
1048 | } |
1049 | |
1050 | /* |
1051 | * Scan swap_map from current position to next entry still in use. |
1052 | * Recycle to start on reaching the end, returning 0 when empty. |
1053 | */ |
1054 | static unsigned int find_next_to_unuse(struct swap_info_struct *si, |
1055 | unsigned int prev) |
1056 | { |
1057 | unsigned int max = si->max; |
1058 | unsigned int i = prev; |
1059 | unsigned char count; |
1060 | |
1061 | /* |
1062 | * No need for swap_lock here: we're just looking |
1063 | * for whether an entry is in use, not modifying it; false |
1064 | * hits are okay, and sys_swapoff() has already prevented new |
1065 | * allocations from this area (while holding swap_lock). |
1066 | */ |
1067 | for (;;) { |
1068 | if (++i >= max) { |
1069 | if (!prev) { |
1070 | i = 0; |
1071 | break; |
1072 | } |
1073 | /* |
1074 | * No entries in use at top of swap_map, |
1075 | * loop back to start and recheck there. |
1076 | */ |
1077 | max = prev + 1; |
1078 | prev = 0; |
1079 | i = 1; |
1080 | } |
1081 | count = si->swap_map[i]; |
1082 | if (count && swap_count(count) != SWAP_MAP_BAD) |
1083 | break; |
1084 | } |
1085 | return i; |
1086 | } |
1087 | |
1088 | /* |
1089 | * We completely avoid races by reading each swap page in advance, |
1090 | * and then search for the process using it. All the necessary |
1091 | * page table adjustments can then be made atomically. |
1092 | */ |
1093 | static int try_to_unuse(unsigned int type) |
1094 | { |
1095 | struct swap_info_struct *si = swap_info[type]; |
1096 | struct mm_struct *start_mm; |
1097 | unsigned char *swap_map; |
1098 | unsigned char swcount; |
1099 | struct page *page; |
1100 | swp_entry_t entry; |
1101 | unsigned int i = 0; |
1102 | int retval = 0; |
1103 | |
1104 | /* |
1105 | * When searching mms for an entry, a good strategy is to |
1106 | * start at the first mm we freed the previous entry from |
1107 | * (though actually we don't notice whether we or coincidence |
1108 | * freed the entry). Initialize this start_mm with a hold. |
1109 | * |
1110 | * A simpler strategy would be to start at the last mm we |
1111 | * freed the previous entry from; but that would take less |
1112 | * advantage of mmlist ordering, which clusters forked mms |
1113 | * together, child after parent. If we race with dup_mmap(), we |
1114 | * prefer to resolve parent before child, lest we miss entries |
1115 | * duplicated after we scanned child: using last mm would invert |
1116 | * that. |
1117 | */ |
1118 | start_mm = &init_mm; |
1119 | atomic_inc(&init_mm.mm_users); |
1120 | |
1121 | /* |
1122 | * Keep on scanning until all entries have gone. Usually, |
1123 | * one pass through swap_map is enough, but not necessarily: |
1124 | * there are races when an instance of an entry might be missed. |
1125 | */ |
1126 | while ((i = find_next_to_unuse(si, i)) != 0) { |
1127 | if (signal_pending(current)) { |
1128 | retval = -EINTR; |
1129 | break; |
1130 | } |
1131 | |
1132 | /* |
1133 | * Get a page for the entry, using the existing swap |
1134 | * cache page if there is one. Otherwise, get a clean |
1135 | * page and read the swap into it. |
1136 | */ |
1137 | swap_map = &si->swap_map[i]; |
1138 | entry = swp_entry(type, i); |
1139 | page = read_swap_cache_async(entry, |
1140 | GFP_HIGHUSER_MOVABLE, NULL, 0); |
1141 | if (!page) { |
1142 | /* |
1143 | * Either swap_duplicate() failed because entry |
1144 | * has been freed independently, and will not be |
1145 | * reused since sys_swapoff() already disabled |
1146 | * allocation from here, or alloc_page() failed. |
1147 | */ |
1148 | if (!*swap_map) |
1149 | continue; |
1150 | retval = -ENOMEM; |
1151 | break; |
1152 | } |
1153 | |
1154 | /* |
1155 | * Don't hold on to start_mm if it looks like exiting. |
1156 | */ |
1157 | if (atomic_read(&start_mm->mm_users) == 1) { |
1158 | mmput(start_mm); |
1159 | start_mm = &init_mm; |
1160 | atomic_inc(&init_mm.mm_users); |
1161 | } |
1162 | |
1163 | /* |
1164 | * Wait for and lock page. When do_swap_page races with |
1165 | * try_to_unuse, do_swap_page can handle the fault much |
1166 | * faster than try_to_unuse can locate the entry. This |
1167 | * apparently redundant "wait_on_page_locked" lets try_to_unuse |
1168 | * defer to do_swap_page in such a case - in some tests, |
1169 | * do_swap_page and try_to_unuse repeatedly compete. |
1170 | */ |
1171 | wait_on_page_locked(page); |
1172 | wait_on_page_writeback(page); |
1173 | lock_page(page); |
1174 | wait_on_page_writeback(page); |
1175 | |
1176 | /* |
1177 | * Remove all references to entry. |
1178 | */ |
1179 | swcount = *swap_map; |
1180 | if (swap_count(swcount) == SWAP_MAP_SHMEM) { |
1181 | retval = shmem_unuse(entry, page); |
1182 | /* page has already been unlocked and released */ |
1183 | if (retval < 0) |
1184 | break; |
1185 | continue; |
1186 | } |
1187 | if (swap_count(swcount) && start_mm != &init_mm) |
1188 | retval = unuse_mm(start_mm, entry, page); |
1189 | |
1190 | if (swap_count(*swap_map)) { |
1191 | int set_start_mm = (*swap_map >= swcount); |
1192 | struct list_head *p = &start_mm->mmlist; |
1193 | struct mm_struct *new_start_mm = start_mm; |
1194 | struct mm_struct *prev_mm = start_mm; |
1195 | struct mm_struct *mm; |
1196 | |
1197 | atomic_inc(&new_start_mm->mm_users); |
1198 | atomic_inc(&prev_mm->mm_users); |
1199 | spin_lock(&mmlist_lock); |
1200 | while (swap_count(*swap_map) && !retval && |
1201 | (p = p->next) != &start_mm->mmlist) { |
1202 | mm = list_entry(p, struct mm_struct, mmlist); |
1203 | if (!atomic_inc_not_zero(&mm->mm_users)) |
1204 | continue; |
1205 | spin_unlock(&mmlist_lock); |
1206 | mmput(prev_mm); |
1207 | prev_mm = mm; |
1208 | |
1209 | cond_resched(); |
1210 | |
1211 | swcount = *swap_map; |
1212 | if (!swap_count(swcount)) /* any usage ? */ |
1213 | ; |
1214 | else if (mm == &init_mm) |
1215 | set_start_mm = 1; |
1216 | else |
1217 | retval = unuse_mm(mm, entry, page); |
1218 | |
1219 | if (set_start_mm && *swap_map < swcount) { |
1220 | mmput(new_start_mm); |
1221 | atomic_inc(&mm->mm_users); |
1222 | new_start_mm = mm; |
1223 | set_start_mm = 0; |
1224 | } |
1225 | spin_lock(&mmlist_lock); |
1226 | } |
1227 | spin_unlock(&mmlist_lock); |
1228 | mmput(prev_mm); |
1229 | mmput(start_mm); |
1230 | start_mm = new_start_mm; |
1231 | } |
1232 | if (retval) { |
1233 | unlock_page(page); |
1234 | page_cache_release(page); |
1235 | break; |
1236 | } |
1237 | |
1238 | /* |
1239 | * If a reference remains (rare), we would like to leave |
1240 | * the page in the swap cache; but try_to_unmap could |
1241 | * then re-duplicate the entry once we drop page lock, |
1242 | * so we might loop indefinitely; also, that page could |
1243 | * not be swapped out to other storage meanwhile. So: |
1244 | * delete from cache even if there's another reference, |
1245 | * after ensuring that the data has been saved to disk - |
1246 | * since if the reference remains (rarer), it will be |
1247 | * read from disk into another page. Splitting into two |
1248 | * pages would be incorrect if swap supported "shared |
1249 | * private" pages, but they are handled by tmpfs files. |
1250 | * |
1251 | * Given how unuse_vma() targets one particular offset |
1252 | * in an anon_vma, once the anon_vma has been determined, |
1253 | * this splitting happens to be just what is needed to |
1254 | * handle where KSM pages have been swapped out: re-reading |
1255 | * is unnecessarily slow, but we can fix that later on. |
1256 | */ |
1257 | if (swap_count(*swap_map) && |
1258 | PageDirty(page) && PageSwapCache(page)) { |
1259 | struct writeback_control wbc = { |
1260 | .sync_mode = WB_SYNC_NONE, |
1261 | }; |
1262 | |
1263 | swap_writepage(page, &wbc); |
1264 | lock_page(page); |
1265 | wait_on_page_writeback(page); |
1266 | } |
1267 | |
1268 | /* |
1269 | * It is conceivable that a racing task removed this page from |
1270 | * swap cache just before we acquired the page lock at the top, |
1271 | * or while we dropped it in unuse_mm(). The page might even |
1272 | * be back in swap cache on another swap area: that we must not |
1273 | * delete, since it may not have been written out to swap yet. |
1274 | */ |
1275 | if (PageSwapCache(page) && |
1276 | likely(page_private(page) == entry.val)) |
1277 | delete_from_swap_cache(page); |
1278 | |
1279 | /* |
1280 | * So we could skip searching mms once swap count went |
1281 | * to 1, we did not mark any present ptes as dirty: must |
1282 | * mark page dirty so shrink_page_list will preserve it. |
1283 | */ |
1284 | SetPageDirty(page); |
1285 | unlock_page(page); |
1286 | page_cache_release(page); |
1287 | |
1288 | /* |
1289 | * Make sure that we aren't completely killing |
1290 | * interactive performance. |
1291 | */ |
1292 | cond_resched(); |
1293 | } |
1294 | |
1295 | mmput(start_mm); |
1296 | return retval; |
1297 | } |
1298 | |
1299 | /* |
1300 | * After a successful try_to_unuse, if no swap is now in use, we know |
1301 | * we can empty the mmlist. swap_lock must be held on entry and exit. |
1302 | * Note that mmlist_lock nests inside swap_lock, and an mm must be |
1303 | * added to the mmlist just after page_duplicate - before would be racy. |
1304 | */ |
1305 | static void drain_mmlist(void) |
1306 | { |
1307 | struct list_head *p, *next; |
1308 | unsigned int type; |
1309 | |
1310 | for (type = 0; type < nr_swapfiles; type++) |
1311 | if (swap_info[type]->inuse_pages) |
1312 | return; |
1313 | spin_lock(&mmlist_lock); |
1314 | list_for_each_safe(p, next, &init_mm.mmlist) |
1315 | list_del_init(p); |
1316 | spin_unlock(&mmlist_lock); |
1317 | } |
1318 | |
1319 | /* |
1320 | * Use this swapdev's extent info to locate the (PAGE_SIZE) block which |
1321 | * corresponds to page offset for the specified swap entry. |
1322 | * Note that the type of this function is sector_t, but it returns page offset |
1323 | * into the bdev, not sector offset. |
1324 | */ |
1325 | static sector_t map_swap_entry(swp_entry_t entry, struct block_device **bdev) |
1326 | { |
1327 | struct swap_info_struct *sis; |
1328 | struct swap_extent *start_se; |
1329 | struct swap_extent *se; |
1330 | pgoff_t offset; |
1331 | |
1332 | sis = swap_info[swp_type(entry)]; |
1333 | *bdev = sis->bdev; |
1334 | |
1335 | offset = swp_offset(entry); |
1336 | start_se = sis->curr_swap_extent; |
1337 | se = start_se; |
1338 | |
1339 | for ( ; ; ) { |
1340 | struct list_head *lh; |
1341 | |
1342 | if (se->start_page <= offset && |
1343 | offset < (se->start_page + se->nr_pages)) { |
1344 | return se->start_block + (offset - se->start_page); |
1345 | } |
1346 | lh = se->list.next; |
1347 | se = list_entry(lh, struct swap_extent, list); |
1348 | sis->curr_swap_extent = se; |
1349 | BUG_ON(se == start_se); /* It *must* be present */ |
1350 | } |
1351 | } |
1352 | |
1353 | /* |
1354 | * Returns the page offset into bdev for the specified page's swap entry. |
1355 | */ |
1356 | sector_t map_swap_page(struct page *page, struct block_device **bdev) |
1357 | { |
1358 | swp_entry_t entry; |
1359 | entry.val = page_private(page); |
1360 | return map_swap_entry(entry, bdev); |
1361 | } |
1362 | |
1363 | /* |
1364 | * Free all of a swapdev's extent information |
1365 | */ |
1366 | static void destroy_swap_extents(struct swap_info_struct *sis) |
1367 | { |
1368 | while (!list_empty(&sis->first_swap_extent.list)) { |
1369 | struct swap_extent *se; |
1370 | |
1371 | se = list_entry(sis->first_swap_extent.list.next, |
1372 | struct swap_extent, list); |
1373 | list_del(&se->list); |
1374 | kfree(se); |
1375 | } |
1376 | } |
1377 | |
1378 | /* |
1379 | * Add a block range (and the corresponding page range) into this swapdev's |
1380 | * extent list. The extent list is kept sorted in page order. |
1381 | * |
1382 | * This function rather assumes that it is called in ascending page order. |
1383 | */ |
1384 | static int |
1385 | add_swap_extent(struct swap_info_struct *sis, unsigned long start_page, |
1386 | unsigned long nr_pages, sector_t start_block) |
1387 | { |
1388 | struct swap_extent *se; |
1389 | struct swap_extent *new_se; |
1390 | struct list_head *lh; |
1391 | |
1392 | if (start_page == 0) { |
1393 | se = &sis->first_swap_extent; |
1394 | sis->curr_swap_extent = se; |
1395 | se->start_page = 0; |
1396 | se->nr_pages = nr_pages; |
1397 | se->start_block = start_block; |
1398 | return 1; |
1399 | } else { |
1400 | lh = sis->first_swap_extent.list.prev; /* Highest extent */ |
1401 | se = list_entry(lh, struct swap_extent, list); |
1402 | BUG_ON(se->start_page + se->nr_pages != start_page); |
1403 | if (se->start_block + se->nr_pages == start_block) { |
1404 | /* Merge it */ |
1405 | se->nr_pages += nr_pages; |
1406 | return 0; |
1407 | } |
1408 | } |
1409 | |
1410 | /* |
1411 | * No merge. Insert a new extent, preserving ordering. |
1412 | */ |
1413 | new_se = kmalloc(sizeof(*se), GFP_KERNEL); |
1414 | if (new_se == NULL) |
1415 | return -ENOMEM; |
1416 | new_se->start_page = start_page; |
1417 | new_se->nr_pages = nr_pages; |
1418 | new_se->start_block = start_block; |
1419 | |
1420 | list_add_tail(&new_se->list, &sis->first_swap_extent.list); |
1421 | return 1; |
1422 | } |
1423 | |
1424 | /* |
1425 | * A `swap extent' is a simple thing which maps a contiguous range of pages |
1426 | * onto a contiguous range of disk blocks. An ordered list of swap extents |
1427 | * is built at swapon time and is then used at swap_writepage/swap_readpage |
1428 | * time for locating where on disk a page belongs. |
1429 | * |
1430 | * If the swapfile is an S_ISBLK block device, a single extent is installed. |
1431 | * This is done so that the main operating code can treat S_ISBLK and S_ISREG |
1432 | * swap files identically. |
1433 | * |
1434 | * Whether the swapdev is an S_ISREG file or an S_ISBLK blockdev, the swap |
1435 | * extent list operates in PAGE_SIZE disk blocks. Both S_ISREG and S_ISBLK |
1436 | * swapfiles are handled *identically* after swapon time. |
1437 | * |
1438 | * For S_ISREG swapfiles, setup_swap_extents() will walk all the file's blocks |
1439 | * and will parse them into an ordered extent list, in PAGE_SIZE chunks. If |
1440 | * some stray blocks are found which do not fall within the PAGE_SIZE alignment |
1441 | * requirements, they are simply tossed out - we will never use those blocks |
1442 | * for swapping. |
1443 | * |
1444 | * For S_ISREG swapfiles we set S_SWAPFILE across the life of the swapon. This |
1445 | * prevents root from shooting her foot off by ftruncating an in-use swapfile, |
1446 | * which will scribble on the fs. |
1447 | * |
1448 | * The amount of disk space which a single swap extent represents varies. |
1449 | * Typically it is in the 1-4 megabyte range. So we can have hundreds of |
1450 | * extents in the list. To avoid much list walking, we cache the previous |
1451 | * search location in `curr_swap_extent', and start new searches from there. |
1452 | * This is extremely effective. The average number of iterations in |
1453 | * map_swap_page() has been measured at about 0.3 per page. - akpm. |
1454 | */ |
1455 | static int setup_swap_extents(struct swap_info_struct *sis, sector_t *span) |
1456 | { |
1457 | struct inode *inode; |
1458 | unsigned blocks_per_page; |
1459 | unsigned long page_no; |
1460 | unsigned blkbits; |
1461 | sector_t probe_block; |
1462 | sector_t last_block; |
1463 | sector_t lowest_block = -1; |
1464 | sector_t highest_block = 0; |
1465 | int nr_extents = 0; |
1466 | int ret; |
1467 | |
1468 | inode = sis->swap_file->f_mapping->host; |
1469 | if (S_ISBLK(inode->i_mode)) { |
1470 | ret = add_swap_extent(sis, 0, sis->max, 0); |
1471 | *span = sis->pages; |
1472 | goto out; |
1473 | } |
1474 | |
1475 | blkbits = inode->i_blkbits; |
1476 | blocks_per_page = PAGE_SIZE >> blkbits; |
1477 | |
1478 | /* |
1479 | * Map all the blocks into the extent list. This code doesn't try |
1480 | * to be very smart. |
1481 | */ |
1482 | probe_block = 0; |
1483 | page_no = 0; |
1484 | last_block = i_size_read(inode) >> blkbits; |
1485 | while ((probe_block + blocks_per_page) <= last_block && |
1486 | page_no < sis->max) { |
1487 | unsigned block_in_page; |
1488 | sector_t first_block; |
1489 | |
1490 | first_block = bmap(inode, probe_block); |
1491 | if (first_block == 0) |
1492 | goto bad_bmap; |
1493 | |
1494 | /* |
1495 | * It must be PAGE_SIZE aligned on-disk |
1496 | */ |
1497 | if (first_block & (blocks_per_page - 1)) { |
1498 | probe_block++; |
1499 | goto reprobe; |
1500 | } |
1501 | |
1502 | for (block_in_page = 1; block_in_page < blocks_per_page; |
1503 | block_in_page++) { |
1504 | sector_t block; |
1505 | |
1506 | block = bmap(inode, probe_block + block_in_page); |
1507 | if (block == 0) |
1508 | goto bad_bmap; |
1509 | if (block != first_block + block_in_page) { |
1510 | /* Discontiguity */ |
1511 | probe_block++; |
1512 | goto reprobe; |
1513 | } |
1514 | } |
1515 | |
1516 | first_block >>= (PAGE_SHIFT - blkbits); |
1517 | if (page_no) { /* exclude the header page */ |
1518 | if (first_block < lowest_block) |
1519 | lowest_block = first_block; |
1520 | if (first_block > highest_block) |
1521 | highest_block = first_block; |
1522 | } |
1523 | |
1524 | /* |
1525 | * We found a PAGE_SIZE-length, PAGE_SIZE-aligned run of blocks |
1526 | */ |
1527 | ret = add_swap_extent(sis, page_no, 1, first_block); |
1528 | if (ret < 0) |
1529 | goto out; |
1530 | nr_extents += ret; |
1531 | page_no++; |
1532 | probe_block += blocks_per_page; |
1533 | reprobe: |
1534 | continue; |
1535 | } |
1536 | ret = nr_extents; |
1537 | *span = 1 + highest_block - lowest_block; |
1538 | if (page_no == 0) |
1539 | page_no = 1; /* force Empty message */ |
1540 | sis->max = page_no; |
1541 | sis->pages = page_no - 1; |
1542 | sis->highest_bit = page_no - 1; |
1543 | out: |
1544 | return ret; |
1545 | bad_bmap: |
1546 | printk(KERN_ERR "swapon: swapfile has holes\n"); |
1547 | ret = -EINVAL; |
1548 | goto out; |
1549 | } |
1550 | |
1551 | SYSCALL_DEFINE1(swapoff, const char __user *, specialfile) |
1552 | { |
1553 | struct swap_info_struct *p = NULL; |
1554 | unsigned char *swap_map; |
1555 | struct file *swap_file, *victim; |
1556 | struct address_space *mapping; |
1557 | struct inode *inode; |
1558 | char *pathname; |
1559 | int i, type, prev; |
1560 | int err; |
1561 | |
1562 | if (!capable(CAP_SYS_ADMIN)) |
1563 | return -EPERM; |
1564 | |
1565 | pathname = getname(specialfile); |
1566 | err = PTR_ERR(pathname); |
1567 | if (IS_ERR(pathname)) |
1568 | goto out; |
1569 | |
1570 | victim = filp_open(pathname, O_RDWR|O_LARGEFILE, 0); |
1571 | putname(pathname); |
1572 | err = PTR_ERR(victim); |
1573 | if (IS_ERR(victim)) |
1574 | goto out; |
1575 | |
1576 | mapping = victim->f_mapping; |
1577 | prev = -1; |
1578 | spin_lock(&swap_lock); |
1579 | for (type = swap_list.head; type >= 0; type = swap_info[type]->next) { |
1580 | p = swap_info[type]; |
1581 | if (p->flags & SWP_WRITEOK) { |
1582 | if (p->swap_file->f_mapping == mapping) |
1583 | break; |
1584 | } |
1585 | prev = type; |
1586 | } |
1587 | if (type < 0) { |
1588 | err = -EINVAL; |
1589 | spin_unlock(&swap_lock); |
1590 | goto out_dput; |
1591 | } |
1592 | if (!security_vm_enough_memory(p->pages)) |
1593 | vm_unacct_memory(p->pages); |
1594 | else { |
1595 | err = -ENOMEM; |
1596 | spin_unlock(&swap_lock); |
1597 | goto out_dput; |
1598 | } |
1599 | if (prev < 0) |
1600 | swap_list.head = p->next; |
1601 | else |
1602 | swap_info[prev]->next = p->next; |
1603 | if (type == swap_list.next) { |
1604 | /* just pick something that's safe... */ |
1605 | swap_list.next = swap_list.head; |
1606 | } |
1607 | if (p->prio < 0) { |
1608 | for (i = p->next; i >= 0; i = swap_info[i]->next) |
1609 | swap_info[i]->prio = p->prio--; |
1610 | least_priority++; |
1611 | } |
1612 | nr_swap_pages -= p->pages; |
1613 | total_swap_pages -= p->pages; |
1614 | p->flags &= ~SWP_WRITEOK; |
1615 | spin_unlock(&swap_lock); |
1616 | |
1617 | current->flags |= PF_OOM_ORIGIN; |
1618 | err = try_to_unuse(type); |
1619 | current->flags &= ~PF_OOM_ORIGIN; |
1620 | |
1621 | if (err) { |
1622 | /* re-insert swap space back into swap_list */ |
1623 | spin_lock(&swap_lock); |
1624 | if (p->prio < 0) |
1625 | p->prio = --least_priority; |
1626 | prev = -1; |
1627 | for (i = swap_list.head; i >= 0; i = swap_info[i]->next) { |
1628 | if (p->prio >= swap_info[i]->prio) |
1629 | break; |
1630 | prev = i; |
1631 | } |
1632 | p->next = i; |
1633 | if (prev < 0) |
1634 | swap_list.head = swap_list.next = type; |
1635 | else |
1636 | swap_info[prev]->next = type; |
1637 | nr_swap_pages += p->pages; |
1638 | total_swap_pages += p->pages; |
1639 | p->flags |= SWP_WRITEOK; |
1640 | spin_unlock(&swap_lock); |
1641 | goto out_dput; |
1642 | } |
1643 | |
1644 | /* wait for any unplug function to finish */ |
1645 | down_write(&swap_unplug_sem); |
1646 | up_write(&swap_unplug_sem); |
1647 | |
1648 | destroy_swap_extents(p); |
1649 | if (p->flags & SWP_CONTINUED) |
1650 | free_swap_count_continuations(p); |
1651 | |
1652 | mutex_lock(&swapon_mutex); |
1653 | spin_lock(&swap_lock); |
1654 | drain_mmlist(); |
1655 | |
1656 | /* wait for anyone still in scan_swap_map */ |
1657 | p->highest_bit = 0; /* cuts scans short */ |
1658 | while (p->flags >= SWP_SCANNING) { |
1659 | spin_unlock(&swap_lock); |
1660 | schedule_timeout_uninterruptible(1); |
1661 | spin_lock(&swap_lock); |
1662 | } |
1663 | |
1664 | swap_file = p->swap_file; |
1665 | p->swap_file = NULL; |
1666 | p->max = 0; |
1667 | swap_map = p->swap_map; |
1668 | p->swap_map = NULL; |
1669 | p->flags = 0; |
1670 | spin_unlock(&swap_lock); |
1671 | mutex_unlock(&swapon_mutex); |
1672 | vfree(swap_map); |
1673 | /* Destroy swap account informatin */ |
1674 | swap_cgroup_swapoff(type); |
1675 | |
1676 | inode = mapping->host; |
1677 | if (S_ISBLK(inode->i_mode)) { |
1678 | struct block_device *bdev = I_BDEV(inode); |
1679 | set_blocksize(bdev, p->old_block_size); |
1680 | bd_release(bdev); |
1681 | } else { |
1682 | mutex_lock(&inode->i_mutex); |
1683 | inode->i_flags &= ~S_SWAPFILE; |
1684 | mutex_unlock(&inode->i_mutex); |
1685 | } |
1686 | filp_close(swap_file, NULL); |
1687 | err = 0; |
1688 | atomic_inc(&proc_poll_event); |
1689 | wake_up_interruptible(&proc_poll_wait); |
1690 | |
1691 | out_dput: |
1692 | filp_close(victim, NULL); |
1693 | out: |
1694 | return err; |
1695 | } |
1696 | |
1697 | #ifdef CONFIG_PROC_FS |
1698 | struct proc_swaps { |
1699 | struct seq_file seq; |
1700 | int event; |
1701 | }; |
1702 | |
1703 | static unsigned swaps_poll(struct file *file, poll_table *wait) |
1704 | { |
1705 | struct proc_swaps *s = file->private_data; |
1706 | |
1707 | poll_wait(file, &proc_poll_wait, wait); |
1708 | |
1709 | if (s->event != atomic_read(&proc_poll_event)) { |
1710 | s->event = atomic_read(&proc_poll_event); |
1711 | return POLLIN | POLLRDNORM | POLLERR | POLLPRI; |
1712 | } |
1713 | |
1714 | return POLLIN | POLLRDNORM; |
1715 | } |
1716 | |
1717 | /* iterator */ |
1718 | static void *swap_start(struct seq_file *swap, loff_t *pos) |
1719 | { |
1720 | struct swap_info_struct *si; |
1721 | int type; |
1722 | loff_t l = *pos; |
1723 | |
1724 | mutex_lock(&swapon_mutex); |
1725 | |
1726 | if (!l) |
1727 | return SEQ_START_TOKEN; |
1728 | |
1729 | for (type = 0; type < nr_swapfiles; type++) { |
1730 | smp_rmb(); /* read nr_swapfiles before swap_info[type] */ |
1731 | si = swap_info[type]; |
1732 | if (!(si->flags & SWP_USED) || !si->swap_map) |
1733 | continue; |
1734 | if (!--l) |
1735 | return si; |
1736 | } |
1737 | |
1738 | return NULL; |
1739 | } |
1740 | |
1741 | static void *swap_next(struct seq_file *swap, void *v, loff_t *pos) |
1742 | { |
1743 | struct swap_info_struct *si = v; |
1744 | int type; |
1745 | |
1746 | if (v == SEQ_START_TOKEN) |
1747 | type = 0; |
1748 | else |
1749 | type = si->type + 1; |
1750 | |
1751 | for (; type < nr_swapfiles; type++) { |
1752 | smp_rmb(); /* read nr_swapfiles before swap_info[type] */ |
1753 | si = swap_info[type]; |
1754 | if (!(si->flags & SWP_USED) || !si->swap_map) |
1755 | continue; |
1756 | ++*pos; |
1757 | return si; |
1758 | } |
1759 | |
1760 | return NULL; |
1761 | } |
1762 | |
1763 | static void swap_stop(struct seq_file *swap, void *v) |
1764 | { |
1765 | mutex_unlock(&swapon_mutex); |
1766 | } |
1767 | |
1768 | static int swap_show(struct seq_file *swap, void *v) |
1769 | { |
1770 | struct swap_info_struct *si = v; |
1771 | struct file *file; |
1772 | int len; |
1773 | |
1774 | if (si == SEQ_START_TOKEN) { |
1775 | seq_puts(swap,"Filename\t\t\t\tType\t\tSize\tUsed\tPriority\n"); |
1776 | return 0; |
1777 | } |
1778 | |
1779 | file = si->swap_file; |
1780 | len = seq_path(swap, &file->f_path, " \t\n\\"); |
1781 | seq_printf(swap, "%*s%s\t%u\t%u\t%d\n", |
1782 | len < 40 ? 40 - len : 1, " ", |
1783 | S_ISBLK(file->f_path.dentry->d_inode->i_mode) ? |
1784 | "partition" : "file\t", |
1785 | si->pages << (PAGE_SHIFT - 10), |
1786 | si->inuse_pages << (PAGE_SHIFT - 10), |
1787 | si->prio); |
1788 | return 0; |
1789 | } |
1790 | |
1791 | static const struct seq_operations swaps_op = { |
1792 | .start = swap_start, |
1793 | .next = swap_next, |
1794 | .stop = swap_stop, |
1795 | .show = swap_show |
1796 | }; |
1797 | |
1798 | static int swaps_open(struct inode *inode, struct file *file) |
1799 | { |
1800 | struct proc_swaps *s; |
1801 | int ret; |
1802 | |
1803 | s = kmalloc(sizeof(struct proc_swaps), GFP_KERNEL); |
1804 | if (!s) |
1805 | return -ENOMEM; |
1806 | |
1807 | file->private_data = s; |
1808 | |
1809 | ret = seq_open(file, &swaps_op); |
1810 | if (ret) { |
1811 | kfree(s); |
1812 | return ret; |
1813 | } |
1814 | |
1815 | s->seq.private = s; |
1816 | s->event = atomic_read(&proc_poll_event); |
1817 | return ret; |
1818 | } |
1819 | |
1820 | static const struct file_operations proc_swaps_operations = { |
1821 | .open = swaps_open, |
1822 | .read = seq_read, |
1823 | .llseek = seq_lseek, |
1824 | .release = seq_release, |
1825 | .poll = swaps_poll, |
1826 | }; |
1827 | |
1828 | static int __init procswaps_init(void) |
1829 | { |
1830 | proc_create("swaps", 0, NULL, &proc_swaps_operations); |
1831 | return 0; |
1832 | } |
1833 | __initcall(procswaps_init); |
1834 | #endif /* CONFIG_PROC_FS */ |
1835 | |
1836 | #ifdef MAX_SWAPFILES_CHECK |
1837 | static int __init max_swapfiles_check(void) |
1838 | { |
1839 | MAX_SWAPFILES_CHECK(); |
1840 | return 0; |
1841 | } |
1842 | late_initcall(max_swapfiles_check); |
1843 | #endif |
1844 | |
1845 | /* |
1846 | * Written 01/25/92 by Simmule Turner, heavily changed by Linus. |
1847 | * |
1848 | * The swapon system call |
1849 | */ |
1850 | SYSCALL_DEFINE2(swapon, const char __user *, specialfile, int, swap_flags) |
1851 | { |
1852 | struct swap_info_struct *p; |
1853 | char *name = NULL; |
1854 | struct block_device *bdev = NULL; |
1855 | struct file *swap_file = NULL; |
1856 | struct address_space *mapping; |
1857 | unsigned int type; |
1858 | int i, prev; |
1859 | int error; |
1860 | union swap_header *swap_header; |
1861 | unsigned int nr_good_pages; |
1862 | int nr_extents = 0; |
1863 | sector_t span; |
1864 | unsigned long maxpages; |
1865 | unsigned long swapfilepages; |
1866 | unsigned char *swap_map = NULL; |
1867 | struct page *page = NULL; |
1868 | struct inode *inode = NULL; |
1869 | int did_down = 0; |
1870 | |
1871 | if (!capable(CAP_SYS_ADMIN)) |
1872 | return -EPERM; |
1873 | |
1874 | p = kzalloc(sizeof(*p), GFP_KERNEL); |
1875 | if (!p) |
1876 | return -ENOMEM; |
1877 | |
1878 | spin_lock(&swap_lock); |
1879 | for (type = 0; type < nr_swapfiles; type++) { |
1880 | if (!(swap_info[type]->flags & SWP_USED)) |
1881 | break; |
1882 | } |
1883 | error = -EPERM; |
1884 | if (type >= MAX_SWAPFILES) { |
1885 | spin_unlock(&swap_lock); |
1886 | kfree(p); |
1887 | goto out; |
1888 | } |
1889 | if (type >= nr_swapfiles) { |
1890 | p->type = type; |
1891 | swap_info[type] = p; |
1892 | /* |
1893 | * Write swap_info[type] before nr_swapfiles, in case a |
1894 | * racing procfs swap_start() or swap_next() is reading them. |
1895 | * (We never shrink nr_swapfiles, we never free this entry.) |
1896 | */ |
1897 | smp_wmb(); |
1898 | nr_swapfiles++; |
1899 | } else { |
1900 | kfree(p); |
1901 | p = swap_info[type]; |
1902 | /* |
1903 | * Do not memset this entry: a racing procfs swap_next() |
1904 | * would be relying on p->type to remain valid. |
1905 | */ |
1906 | } |
1907 | INIT_LIST_HEAD(&p->first_swap_extent.list); |
1908 | p->flags = SWP_USED; |
1909 | p->next = -1; |
1910 | spin_unlock(&swap_lock); |
1911 | |
1912 | name = getname(specialfile); |
1913 | error = PTR_ERR(name); |
1914 | if (IS_ERR(name)) { |
1915 | name = NULL; |
1916 | goto bad_swap_2; |
1917 | } |
1918 | swap_file = filp_open(name, O_RDWR|O_LARGEFILE, 0); |
1919 | error = PTR_ERR(swap_file); |
1920 | if (IS_ERR(swap_file)) { |
1921 | swap_file = NULL; |
1922 | goto bad_swap_2; |
1923 | } |
1924 | |
1925 | p->swap_file = swap_file; |
1926 | mapping = swap_file->f_mapping; |
1927 | inode = mapping->host; |
1928 | |
1929 | error = -EBUSY; |
1930 | for (i = 0; i < nr_swapfiles; i++) { |
1931 | struct swap_info_struct *q = swap_info[i]; |
1932 | |
1933 | if (i == type || !q->swap_file) |
1934 | continue; |
1935 | if (mapping == q->swap_file->f_mapping) |
1936 | goto bad_swap; |
1937 | } |
1938 | |
1939 | error = -EINVAL; |
1940 | if (S_ISBLK(inode->i_mode)) { |
1941 | bdev = I_BDEV(inode); |
1942 | error = bd_claim(bdev, sys_swapon); |
1943 | if (error < 0) { |
1944 | bdev = NULL; |
1945 | error = -EINVAL; |
1946 | goto bad_swap; |
1947 | } |
1948 | p->old_block_size = block_size(bdev); |
1949 | error = set_blocksize(bdev, PAGE_SIZE); |
1950 | if (error < 0) |
1951 | goto bad_swap; |
1952 | p->bdev = bdev; |
1953 | p->flags |= SWP_BLKDEV; |
1954 | } else if (S_ISREG(inode->i_mode)) { |
1955 | p->bdev = inode->i_sb->s_bdev; |
1956 | mutex_lock(&inode->i_mutex); |
1957 | did_down = 1; |
1958 | if (IS_SWAPFILE(inode)) { |
1959 | error = -EBUSY; |
1960 | goto bad_swap; |
1961 | } |
1962 | } else { |
1963 | goto bad_swap; |
1964 | } |
1965 | |
1966 | swapfilepages = i_size_read(inode) >> PAGE_SHIFT; |
1967 | |
1968 | /* |
1969 | * Read the swap header. |
1970 | */ |
1971 | if (!mapping->a_ops->readpage) { |
1972 | error = -EINVAL; |
1973 | goto bad_swap; |
1974 | } |
1975 | page = read_mapping_page(mapping, 0, swap_file); |
1976 | if (IS_ERR(page)) { |
1977 | error = PTR_ERR(page); |
1978 | goto bad_swap; |
1979 | } |
1980 | swap_header = kmap(page); |
1981 | |
1982 | if (memcmp("SWAPSPACE2", swap_header->magic.magic, 10)) { |
1983 | printk(KERN_ERR "Unable to find swap-space signature\n"); |
1984 | error = -EINVAL; |
1985 | goto bad_swap; |
1986 | } |
1987 | |
1988 | /* swap partition endianess hack... */ |
1989 | if (swab32(swap_header->info.version) == 1) { |
1990 | swab32s(&swap_header->info.version); |
1991 | swab32s(&swap_header->info.last_page); |
1992 | swab32s(&swap_header->info.nr_badpages); |
1993 | for (i = 0; i < swap_header->info.nr_badpages; i++) |
1994 | swab32s(&swap_header->info.badpages[i]); |
1995 | } |
1996 | /* Check the swap header's sub-version */ |
1997 | if (swap_header->info.version != 1) { |
1998 | printk(KERN_WARNING |
1999 | "Unable to handle swap header version %d\n", |
2000 | swap_header->info.version); |
2001 | error = -EINVAL; |
2002 | goto bad_swap; |
2003 | } |
2004 | |
2005 | p->lowest_bit = 1; |
2006 | p->cluster_next = 1; |
2007 | p->cluster_nr = 0; |
2008 | |
2009 | /* |
2010 | * Find out how many pages are allowed for a single swap |
2011 | * device. There are two limiting factors: 1) the number of |
2012 | * bits for the swap offset in the swp_entry_t type and |
2013 | * 2) the number of bits in the a swap pte as defined by |
2014 | * the different architectures. In order to find the |
2015 | * largest possible bit mask a swap entry with swap type 0 |
2016 | * and swap offset ~0UL is created, encoded to a swap pte, |
2017 | * decoded to a swp_entry_t again and finally the swap |
2018 | * offset is extracted. This will mask all the bits from |
2019 | * the initial ~0UL mask that can't be encoded in either |
2020 | * the swp_entry_t or the architecture definition of a |
2021 | * swap pte. |
2022 | */ |
2023 | maxpages = swp_offset(pte_to_swp_entry( |
2024 | swp_entry_to_pte(swp_entry(0, ~0UL)))) + 1; |
2025 | if (maxpages > swap_header->info.last_page) { |
2026 | maxpages = swap_header->info.last_page + 1; |
2027 | /* p->max is an unsigned int: don't overflow it */ |
2028 | if ((unsigned int)maxpages == 0) |
2029 | maxpages = UINT_MAX; |
2030 | } |
2031 | p->highest_bit = maxpages - 1; |
2032 | |
2033 | error = -EINVAL; |
2034 | if (!maxpages) |
2035 | goto bad_swap; |
2036 | if (swapfilepages && maxpages > swapfilepages) { |
2037 | printk(KERN_WARNING |
2038 | "Swap area shorter than signature indicates\n"); |
2039 | goto bad_swap; |
2040 | } |
2041 | if (swap_header->info.nr_badpages && S_ISREG(inode->i_mode)) |
2042 | goto bad_swap; |
2043 | if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES) |
2044 | goto bad_swap; |
2045 | |
2046 | /* OK, set up the swap map and apply the bad block list */ |
2047 | swap_map = vmalloc(maxpages); |
2048 | if (!swap_map) { |
2049 | error = -ENOMEM; |
2050 | goto bad_swap; |
2051 | } |
2052 | |
2053 | memset(swap_map, 0, maxpages); |
2054 | nr_good_pages = maxpages - 1; /* omit header page */ |
2055 | |
2056 | for (i = 0; i < swap_header->info.nr_badpages; i++) { |
2057 | unsigned int page_nr = swap_header->info.badpages[i]; |
2058 | if (page_nr == 0 || page_nr > swap_header->info.last_page) { |
2059 | error = -EINVAL; |
2060 | goto bad_swap; |
2061 | } |
2062 | if (page_nr < maxpages) { |
2063 | swap_map[page_nr] = SWAP_MAP_BAD; |
2064 | nr_good_pages--; |
2065 | } |
2066 | } |
2067 | |
2068 | error = swap_cgroup_swapon(type, maxpages); |
2069 | if (error) |
2070 | goto bad_swap; |
2071 | |
2072 | if (nr_good_pages) { |
2073 | swap_map[0] = SWAP_MAP_BAD; |
2074 | p->max = maxpages; |
2075 | p->pages = nr_good_pages; |
2076 | nr_extents = setup_swap_extents(p, &span); |
2077 | if (nr_extents < 0) { |
2078 | error = nr_extents; |
2079 | goto bad_swap; |
2080 | } |
2081 | nr_good_pages = p->pages; |
2082 | } |
2083 | if (!nr_good_pages) { |
2084 | printk(KERN_WARNING "Empty swap-file\n"); |
2085 | error = -EINVAL; |
2086 | goto bad_swap; |
2087 | } |
2088 | |
2089 | if (p->bdev) { |
2090 | if (blk_queue_nonrot(bdev_get_queue(p->bdev))) { |
2091 | p->flags |= SWP_SOLIDSTATE; |
2092 | p->cluster_next = 1 + (random32() % p->highest_bit); |
2093 | } |
2094 | if (discard_swap(p) == 0 && (swap_flags & SWAP_FLAG_DISCARD)) |
2095 | p->flags |= SWP_DISCARDABLE; |
2096 | } |
2097 | |
2098 | mutex_lock(&swapon_mutex); |
2099 | spin_lock(&swap_lock); |
2100 | if (swap_flags & SWAP_FLAG_PREFER) |
2101 | p->prio = |
2102 | (swap_flags & SWAP_FLAG_PRIO_MASK) >> SWAP_FLAG_PRIO_SHIFT; |
2103 | else |
2104 | p->prio = --least_priority; |
2105 | p->swap_map = swap_map; |
2106 | p->flags |= SWP_WRITEOK; |
2107 | nr_swap_pages += nr_good_pages; |
2108 | total_swap_pages += nr_good_pages; |
2109 | |
2110 | printk(KERN_INFO "Adding %uk swap on %s. " |
2111 | "Priority:%d extents:%d across:%lluk %s%s\n", |
2112 | nr_good_pages<<(PAGE_SHIFT-10), name, p->prio, |
2113 | nr_extents, (unsigned long long)span<<(PAGE_SHIFT-10), |
2114 | (p->flags & SWP_SOLIDSTATE) ? "SS" : "", |
2115 | (p->flags & SWP_DISCARDABLE) ? "D" : ""); |
2116 | |
2117 | /* insert swap space into swap_list: */ |
2118 | prev = -1; |
2119 | for (i = swap_list.head; i >= 0; i = swap_info[i]->next) { |
2120 | if (p->prio >= swap_info[i]->prio) |
2121 | break; |
2122 | prev = i; |
2123 | } |
2124 | p->next = i; |
2125 | if (prev < 0) |
2126 | swap_list.head = swap_list.next = type; |
2127 | else |
2128 | swap_info[prev]->next = type; |
2129 | spin_unlock(&swap_lock); |
2130 | mutex_unlock(&swapon_mutex); |
2131 | atomic_inc(&proc_poll_event); |
2132 | wake_up_interruptible(&proc_poll_wait); |
2133 | |
2134 | error = 0; |
2135 | goto out; |
2136 | bad_swap: |
2137 | if (bdev) { |
2138 | set_blocksize(bdev, p->old_block_size); |
2139 | bd_release(bdev); |
2140 | } |
2141 | destroy_swap_extents(p); |
2142 | swap_cgroup_swapoff(type); |
2143 | bad_swap_2: |
2144 | spin_lock(&swap_lock); |
2145 | p->swap_file = NULL; |
2146 | p->flags = 0; |
2147 | spin_unlock(&swap_lock); |
2148 | vfree(swap_map); |
2149 | if (swap_file) |
2150 | filp_close(swap_file, NULL); |
2151 | out: |
2152 | if (page && !IS_ERR(page)) { |
2153 | kunmap(page); |
2154 | page_cache_release(page); |
2155 | } |
2156 | if (name) |
2157 | putname(name); |
2158 | if (did_down) { |
2159 | if (!error) |
2160 | inode->i_flags |= S_SWAPFILE; |
2161 | mutex_unlock(&inode->i_mutex); |
2162 | } |
2163 | return error; |
2164 | } |
2165 | |
2166 | void si_swapinfo(struct sysinfo *val) |
2167 | { |
2168 | unsigned int type; |
2169 | unsigned long nr_to_be_unused = 0; |
2170 | |
2171 | spin_lock(&swap_lock); |
2172 | for (type = 0; type < nr_swapfiles; type++) { |
2173 | struct swap_info_struct *si = swap_info[type]; |
2174 | |
2175 | if ((si->flags & SWP_USED) && !(si->flags & SWP_WRITEOK)) |
2176 | nr_to_be_unused += si->inuse_pages; |
2177 | } |
2178 | val->freeswap = nr_swap_pages + nr_to_be_unused; |
2179 | val->totalswap = total_swap_pages + nr_to_be_unused; |
2180 | spin_unlock(&swap_lock); |
2181 | } |
2182 | |
2183 | /* |
2184 | * Verify that a swap entry is valid and increment its swap map count. |
2185 | * |
2186 | * Returns error code in following case. |
2187 | * - success -> 0 |
2188 | * - swp_entry is invalid -> EINVAL |
2189 | * - swp_entry is migration entry -> EINVAL |
2190 | * - swap-cache reference is requested but there is already one. -> EEXIST |
2191 | * - swap-cache reference is requested but the entry is not used. -> ENOENT |
2192 | * - swap-mapped reference requested but needs continued swap count. -> ENOMEM |
2193 | */ |
2194 | static int __swap_duplicate(swp_entry_t entry, unsigned char usage) |
2195 | { |
2196 | struct swap_info_struct *p; |
2197 | unsigned long offset, type; |
2198 | unsigned char count; |
2199 | unsigned char has_cache; |
2200 | int err = -EINVAL; |
2201 | |
2202 | if (non_swap_entry(entry)) |
2203 | goto out; |
2204 | |
2205 | type = swp_type(entry); |
2206 | if (type >= nr_swapfiles) |
2207 | goto bad_file; |
2208 | p = swap_info[type]; |
2209 | offset = swp_offset(entry); |
2210 | |
2211 | spin_lock(&swap_lock); |
2212 | if (unlikely(offset >= p->max)) |
2213 | goto unlock_out; |
2214 | |
2215 | count = p->swap_map[offset]; |
2216 | has_cache = count & SWAP_HAS_CACHE; |
2217 | count &= ~SWAP_HAS_CACHE; |
2218 | err = 0; |
2219 | |
2220 | if (usage == SWAP_HAS_CACHE) { |
2221 | |
2222 | /* set SWAP_HAS_CACHE if there is no cache and entry is used */ |
2223 | if (!has_cache && count) |
2224 | has_cache = SWAP_HAS_CACHE; |
2225 | else if (has_cache) /* someone else added cache */ |
2226 | err = -EEXIST; |
2227 | else /* no users remaining */ |
2228 | err = -ENOENT; |
2229 | |
2230 | } else if (count || has_cache) { |
2231 | |
2232 | if ((count & ~COUNT_CONTINUED) < SWAP_MAP_MAX) |
2233 | count += usage; |
2234 | else if ((count & ~COUNT_CONTINUED) > SWAP_MAP_MAX) |
2235 | err = -EINVAL; |
2236 | else if (swap_count_continued(p, offset, count)) |
2237 | count = COUNT_CONTINUED; |
2238 | else |
2239 | err = -ENOMEM; |
2240 | } else |
2241 | err = -ENOENT; /* unused swap entry */ |
2242 | |
2243 | p->swap_map[offset] = count | has_cache; |
2244 | |
2245 | unlock_out: |
2246 | spin_unlock(&swap_lock); |
2247 | out: |
2248 | return err; |
2249 | |
2250 | bad_file: |
2251 | printk(KERN_ERR "swap_dup: %s%08lx\n", Bad_file, entry.val); |
2252 | goto out; |
2253 | } |
2254 | |
2255 | /* |
2256 | * Help swapoff by noting that swap entry belongs to shmem/tmpfs |
2257 | * (in which case its reference count is never incremented). |
2258 | */ |
2259 | void swap_shmem_alloc(swp_entry_t entry) |
2260 | { |
2261 | __swap_duplicate(entry, SWAP_MAP_SHMEM); |
2262 | } |
2263 | |
2264 | /* |
2265 | * Increase reference count of swap entry by 1. |
2266 | * Returns 0 for success, or -ENOMEM if a swap_count_continuation is required |
2267 | * but could not be atomically allocated. Returns 0, just as if it succeeded, |
2268 | * if __swap_duplicate() fails for another reason (-EINVAL or -ENOENT), which |
2269 | * might occur if a page table entry has got corrupted. |
2270 | */ |
2271 | int swap_duplicate(swp_entry_t entry) |
2272 | { |
2273 | int err = 0; |
2274 | |
2275 | while (!err && __swap_duplicate(entry, 1) == -ENOMEM) |
2276 | err = add_swap_count_continuation(entry, GFP_ATOMIC); |
2277 | return err; |
2278 | } |
2279 | |
2280 | /* |
2281 | * @entry: swap entry for which we allocate swap cache. |
2282 | * |
2283 | * Called when allocating swap cache for existing swap entry, |
2284 | * This can return error codes. Returns 0 at success. |
2285 | * -EBUSY means there is a swap cache. |
2286 | * Note: return code is different from swap_duplicate(). |
2287 | */ |
2288 | int swapcache_prepare(swp_entry_t entry) |
2289 | { |
2290 | return __swap_duplicate(entry, SWAP_HAS_CACHE); |
2291 | } |
2292 | |
2293 | /* |
2294 | * swap_lock prevents swap_map being freed. Don't grab an extra |
2295 | * reference on the swaphandle, it doesn't matter if it becomes unused. |
2296 | */ |
2297 | int valid_swaphandles(swp_entry_t entry, unsigned long *offset) |
2298 | { |
2299 | struct swap_info_struct *si; |
2300 | int our_page_cluster = page_cluster; |
2301 | pgoff_t target, toff; |
2302 | pgoff_t base, end; |
2303 | int nr_pages = 0; |
2304 | |
2305 | if (!our_page_cluster) /* no readahead */ |
2306 | return 0; |
2307 | |
2308 | si = swap_info[swp_type(entry)]; |
2309 | target = swp_offset(entry); |
2310 | base = (target >> our_page_cluster) << our_page_cluster; |
2311 | end = base + (1 << our_page_cluster); |
2312 | if (!base) /* first page is swap header */ |
2313 | base++; |
2314 | |
2315 | spin_lock(&swap_lock); |
2316 | if (end > si->max) /* don't go beyond end of map */ |
2317 | end = si->max; |
2318 | |
2319 | /* Count contiguous allocated slots above our target */ |
2320 | for (toff = target; ++toff < end; nr_pages++) { |
2321 | /* Don't read in free or bad pages */ |
2322 | if (!si->swap_map[toff]) |
2323 | break; |
2324 | if (swap_count(si->swap_map[toff]) == SWAP_MAP_BAD) |
2325 | break; |
2326 | } |
2327 | /* Count contiguous allocated slots below our target */ |
2328 | for (toff = target; --toff >= base; nr_pages++) { |
2329 | /* Don't read in free or bad pages */ |
2330 | if (!si->swap_map[toff]) |
2331 | break; |
2332 | if (swap_count(si->swap_map[toff]) == SWAP_MAP_BAD) |
2333 | break; |
2334 | } |
2335 | spin_unlock(&swap_lock); |
2336 | |
2337 | /* |
2338 | * Indicate starting offset, and return number of pages to get: |
2339 | * if only 1, say 0, since there's then no readahead to be done. |
2340 | */ |
2341 | *offset = ++toff; |
2342 | return nr_pages? ++nr_pages: 0; |
2343 | } |
2344 | |
2345 | /* |
2346 | * add_swap_count_continuation - called when a swap count is duplicated |
2347 | * beyond SWAP_MAP_MAX, it allocates a new page and links that to the entry's |
2348 | * page of the original vmalloc'ed swap_map, to hold the continuation count |
2349 | * (for that entry and for its neighbouring PAGE_SIZE swap entries). Called |
2350 | * again when count is duplicated beyond SWAP_MAP_MAX * SWAP_CONT_MAX, etc. |
2351 | * |
2352 | * These continuation pages are seldom referenced: the common paths all work |
2353 | * on the original swap_map, only referring to a continuation page when the |
2354 | * low "digit" of a count is incremented or decremented through SWAP_MAP_MAX. |
2355 | * |
2356 | * add_swap_count_continuation(, GFP_ATOMIC) can be called while holding |
2357 | * page table locks; if it fails, add_swap_count_continuation(, GFP_KERNEL) |
2358 | * can be called after dropping locks. |
2359 | */ |
2360 | int add_swap_count_continuation(swp_entry_t entry, gfp_t gfp_mask) |
2361 | { |
2362 | struct swap_info_struct *si; |
2363 | struct page *head; |
2364 | struct page *page; |
2365 | struct page *list_page; |
2366 | pgoff_t offset; |
2367 | unsigned char count; |
2368 | |
2369 | /* |
2370 | * When debugging, it's easier to use __GFP_ZERO here; but it's better |
2371 | * for latency not to zero a page while GFP_ATOMIC and holding locks. |
2372 | */ |
2373 | page = alloc_page(gfp_mask | __GFP_HIGHMEM); |
2374 | |
2375 | si = swap_info_get(entry); |
2376 | if (!si) { |
2377 | /* |
2378 | * An acceptable race has occurred since the failing |
2379 | * __swap_duplicate(): the swap entry has been freed, |
2380 | * perhaps even the whole swap_map cleared for swapoff. |
2381 | */ |
2382 | goto outer; |
2383 | } |
2384 | |
2385 | offset = swp_offset(entry); |
2386 | count = si->swap_map[offset] & ~SWAP_HAS_CACHE; |
2387 | |
2388 | if ((count & ~COUNT_CONTINUED) != SWAP_MAP_MAX) { |
2389 | /* |
2390 | * The higher the swap count, the more likely it is that tasks |
2391 | * will race to add swap count continuation: we need to avoid |
2392 | * over-provisioning. |
2393 | */ |
2394 | goto out; |
2395 | } |
2396 | |
2397 | if (!page) { |
2398 | spin_unlock(&swap_lock); |
2399 | return -ENOMEM; |
2400 | } |
2401 | |
2402 | /* |
2403 | * We are fortunate that although vmalloc_to_page uses pte_offset_map, |
2404 | * no architecture is using highmem pages for kernel pagetables: so it |
2405 | * will not corrupt the GFP_ATOMIC caller's atomic pagetable kmaps. |
2406 | */ |
2407 | head = vmalloc_to_page(si->swap_map + offset); |
2408 | offset &= ~PAGE_MASK; |
2409 | |
2410 | /* |
2411 | * Page allocation does not initialize the page's lru field, |
2412 | * but it does always reset its private field. |
2413 | */ |
2414 | if (!page_private(head)) { |
2415 | BUG_ON(count & COUNT_CONTINUED); |
2416 | INIT_LIST_HEAD(&head->lru); |
2417 | set_page_private(head, SWP_CONTINUED); |
2418 | si->flags |= SWP_CONTINUED; |
2419 | } |
2420 | |
2421 | list_for_each_entry(list_page, &head->lru, lru) { |
2422 | unsigned char *map; |
2423 | |
2424 | /* |
2425 | * If the previous map said no continuation, but we've found |
2426 | * a continuation page, free our allocation and use this one. |
2427 | */ |
2428 | if (!(count & COUNT_CONTINUED)) |
2429 | goto out; |
2430 | |
2431 | map = kmap_atomic(list_page, KM_USER0) + offset; |
2432 | count = *map; |
2433 | kunmap_atomic(map, KM_USER0); |
2434 | |
2435 | /* |
2436 | * If this continuation count now has some space in it, |
2437 | * free our allocation and use this one. |
2438 | */ |
2439 | if ((count & ~COUNT_CONTINUED) != SWAP_CONT_MAX) |
2440 | goto out; |
2441 | } |
2442 | |
2443 | list_add_tail(&page->lru, &head->lru); |
2444 | page = NULL; /* now it's attached, don't free it */ |
2445 | out: |
2446 | spin_unlock(&swap_lock); |
2447 | outer: |
2448 | if (page) |
2449 | __free_page(page); |
2450 | return 0; |
2451 | } |
2452 | |
2453 | /* |
2454 | * swap_count_continued - when the original swap_map count is incremented |
2455 | * from SWAP_MAP_MAX, check if there is already a continuation page to carry |
2456 | * into, carry if so, or else fail until a new continuation page is allocated; |
2457 | * when the original swap_map count is decremented from 0 with continuation, |
2458 | * borrow from the continuation and report whether it still holds more. |
2459 | * Called while __swap_duplicate() or swap_entry_free() holds swap_lock. |
2460 | */ |
2461 | static bool swap_count_continued(struct swap_info_struct *si, |
2462 | pgoff_t offset, unsigned char count) |
2463 | { |
2464 | struct page *head; |
2465 | struct page *page; |
2466 | unsigned char *map; |
2467 | |
2468 | head = vmalloc_to_page(si->swap_map + offset); |
2469 | if (page_private(head) != SWP_CONTINUED) { |
2470 | BUG_ON(count & COUNT_CONTINUED); |
2471 | return false; /* need to add count continuation */ |
2472 | } |
2473 | |
2474 | offset &= ~PAGE_MASK; |
2475 | page = list_entry(head->lru.next, struct page, lru); |
2476 | map = kmap_atomic(page, KM_USER0) + offset; |
2477 | |
2478 | if (count == SWAP_MAP_MAX) /* initial increment from swap_map */ |
2479 | goto init_map; /* jump over SWAP_CONT_MAX checks */ |
2480 | |
2481 | if (count == (SWAP_MAP_MAX | COUNT_CONTINUED)) { /* incrementing */ |
2482 | /* |
2483 | * Think of how you add 1 to 999 |
2484 | */ |
2485 | while (*map == (SWAP_CONT_MAX | COUNT_CONTINUED)) { |
2486 | kunmap_atomic(map, KM_USER0); |
2487 | page = list_entry(page->lru.next, struct page, lru); |
2488 | BUG_ON(page == head); |
2489 | map = kmap_atomic(page, KM_USER0) + offset; |
2490 | } |
2491 | if (*map == SWAP_CONT_MAX) { |
2492 | kunmap_atomic(map, KM_USER0); |
2493 | page = list_entry(page->lru.next, struct page, lru); |
2494 | if (page == head) |
2495 | return false; /* add count continuation */ |
2496 | map = kmap_atomic(page, KM_USER0) + offset; |
2497 | init_map: *map = 0; /* we didn't zero the page */ |
2498 | } |
2499 | *map += 1; |
2500 | kunmap_atomic(map, KM_USER0); |
2501 | page = list_entry(page->lru.prev, struct page, lru); |
2502 | while (page != head) { |
2503 | map = kmap_atomic(page, KM_USER0) + offset; |
2504 | *map = COUNT_CONTINUED; |
2505 | kunmap_atomic(map, KM_USER0); |
2506 | page = list_entry(page->lru.prev, struct page, lru); |
2507 | } |
2508 | return true; /* incremented */ |
2509 | |
2510 | } else { /* decrementing */ |
2511 | /* |
2512 | * Think of how you subtract 1 from 1000 |
2513 | */ |
2514 | BUG_ON(count != COUNT_CONTINUED); |
2515 | while (*map == COUNT_CONTINUED) { |
2516 | kunmap_atomic(map, KM_USER0); |
2517 | page = list_entry(page->lru.next, struct page, lru); |
2518 | BUG_ON(page == head); |
2519 | map = kmap_atomic(page, KM_USER0) + offset; |
2520 | } |
2521 | BUG_ON(*map == 0); |
2522 | *map -= 1; |
2523 | if (*map == 0) |
2524 | count = 0; |
2525 | kunmap_atomic(map, KM_USER0); |
2526 | page = list_entry(page->lru.prev, struct page, lru); |
2527 | while (page != head) { |
2528 | map = kmap_atomic(page, KM_USER0) + offset; |
2529 | *map = SWAP_CONT_MAX | count; |
2530 | count = COUNT_CONTINUED; |
2531 | kunmap_atomic(map, KM_USER0); |
2532 | page = list_entry(page->lru.prev, struct page, lru); |
2533 | } |
2534 | return count == COUNT_CONTINUED; |
2535 | } |
2536 | } |
2537 | |
2538 | /* |
2539 | * free_swap_count_continuations - swapoff free all the continuation pages |
2540 | * appended to the swap_map, after swap_map is quiesced, before vfree'ing it. |
2541 | */ |
2542 | static void free_swap_count_continuations(struct swap_info_struct *si) |
2543 | { |
2544 | pgoff_t offset; |
2545 | |
2546 | for (offset = 0; offset < si->max; offset += PAGE_SIZE) { |
2547 | struct page *head; |
2548 | head = vmalloc_to_page(si->swap_map + offset); |
2549 | if (page_private(head)) { |
2550 | struct list_head *this, *next; |
2551 | list_for_each_safe(this, next, &head->lru) { |
2552 | struct page *page; |
2553 | page = list_entry(this, struct page, lru); |
2554 | list_del(this); |
2555 | __free_page(page); |
2556 | } |
2557 | } |
2558 | } |
2559 | } |
2560 |
Branches:
ben-wpan
ben-wpan-stefan
javiroman/ks7010
jz-2.6.34
jz-2.6.34-rc5
jz-2.6.34-rc6
jz-2.6.34-rc7
jz-2.6.35
jz-2.6.36
jz-2.6.37
jz-2.6.38
jz-2.6.39
jz-3.0
jz-3.1
jz-3.11
jz-3.12
jz-3.13
jz-3.15
jz-3.16
jz-3.18-dt
jz-3.2
jz-3.3
jz-3.4
jz-3.5
jz-3.6
jz-3.6-rc2-pwm
jz-3.9
jz-3.9-clk
jz-3.9-rc8
jz47xx
jz47xx-2.6.38
master
Tags:
od-2011-09-04
od-2011-09-18
v2.6.34-rc5
v2.6.34-rc6
v2.6.34-rc7
v3.9