1// SPDX-License-Identifier: GPL-2.0
2/*
3 * linux/mm/swap_state.c
4 *
5 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
6 * Swap reorganised 29.12.95, Stephen Tweedie
7 *
8 * Rewritten to use page cache, (C) 1998 Stephen Tweedie
9 */
10#include <linux/mm.h>
11#include <linux/gfp.h>
12#include <linux/kernel_stat.h>
13#include <linux/swap.h>
14#include <linux/swapops.h>
15#include <linux/init.h>
16#include <linux/pagemap.h>
17#include <linux/backing-dev.h>
18#include <linux/blkdev.h>
19#include <linux/pagevec.h>
20#include <linux/migrate.h>
21#include <linux/vmalloc.h>
22#include <linux/swap_slots.h>
23#include <linux/huge_mm.h>
24
25#include <asm/pgtable.h>
26
27/*
28 * swapper_space is a fiction, retained to simplify the path through
29 * vmscan's shrink_page_list.
30 */
31static const struct address_space_operations swap_aops = {
32 .writepage = swap_writepage,
33 .set_page_dirty = swap_set_page_dirty,
34#ifdef CONFIG_MIGRATION
35 .migratepage = migrate_page,
36#endif
37};
38
39struct address_space *swapper_spaces[MAX_SWAPFILES] __read_mostly;
40static unsigned int nr_swapper_spaces[MAX_SWAPFILES] __read_mostly;
41static bool enable_vma_readahead __read_mostly = true;
42
43#define SWAP_RA_WIN_SHIFT (PAGE_SHIFT / 2)
44#define SWAP_RA_HITS_MASK ((1UL << SWAP_RA_WIN_SHIFT) - 1)
45#define SWAP_RA_HITS_MAX SWAP_RA_HITS_MASK
46#define SWAP_RA_WIN_MASK (~PAGE_MASK & ~SWAP_RA_HITS_MASK)
47
48#define SWAP_RA_HITS(v) ((v) & SWAP_RA_HITS_MASK)
49#define SWAP_RA_WIN(v) (((v) & SWAP_RA_WIN_MASK) >> SWAP_RA_WIN_SHIFT)
50#define SWAP_RA_ADDR(v) ((v) & PAGE_MASK)
51
52#define SWAP_RA_VAL(addr, win, hits) \
53 (((addr) & PAGE_MASK) | \
54 (((win) << SWAP_RA_WIN_SHIFT) & SWAP_RA_WIN_MASK) | \
55 ((hits) & SWAP_RA_HITS_MASK))
56
57/* Initial readahead hits is 4 to start up with a small window */
58#define GET_SWAP_RA_VAL(vma) \
59 (atomic_long_read(&(vma)->swap_readahead_info) ? : 4)
60
61#define INC_CACHE_INFO(x) do { swap_cache_info.x++; } while (0)
62#define ADD_CACHE_INFO(x, nr) do { swap_cache_info.x += (nr); } while (0)
63
64static struct {
65 unsigned long add_total;
66 unsigned long del_total;
67 unsigned long find_success;
68 unsigned long find_total;
69} swap_cache_info;
70
71unsigned long total_swapcache_pages(void)
72{
73 unsigned int i, j, nr;
74 unsigned long ret = 0;
75 struct address_space *spaces;
76
77 rcu_read_lock();
78 for (i = 0; i < MAX_SWAPFILES; i++) {
79 /*
80 * The corresponding entries in nr_swapper_spaces and
81 * swapper_spaces will be reused only after at least
82 * one grace period. So it is impossible for them
83 * belongs to different usage.
84 */
85 nr = nr_swapper_spaces[i];
86 spaces = rcu_dereference(swapper_spaces[i]);
87 if (!nr || !spaces)
88 continue;
89 for (j = 0; j < nr; j++)
90 ret += spaces[j].nrpages;
91 }
92 rcu_read_unlock();
93 return ret;
94}
95
96static atomic_t swapin_readahead_hits = ATOMIC_INIT(4);
97
98void show_swap_cache_info(void)
99{
100 printk("%lu pages in swap cache\n", total_swapcache_pages());
101 printk("Swap cache stats: add %lu, delete %lu, find %lu/%lu\n",
102 swap_cache_info.add_total, swap_cache_info.del_total,
103 swap_cache_info.find_success, swap_cache_info.find_total);
104 printk("Free swap = %ldkB\n",
105 get_nr_swap_pages() << (PAGE_SHIFT - 10));
106 printk("Total swap = %lukB\n", total_swap_pages << (PAGE_SHIFT - 10));
107}
108
109/*
110 * add_to_swap_cache resembles add_to_page_cache_locked on swapper_space,
111 * but sets SwapCache flag and private instead of mapping and index.
112 */
113int add_to_swap_cache(struct page *page, swp_entry_t entry, gfp_t gfp)
114{
115 struct address_space *address_space = swap_address_space(entry);
116 pgoff_t idx = swp_offset(entry);
117 XA_STATE_ORDER(xas, &address_space->i_pages, idx, compound_order(page));
118 unsigned long i, nr = 1UL << compound_order(page);
119
120 VM_BUG_ON_PAGE(!PageLocked(page), page);
121 VM_BUG_ON_PAGE(PageSwapCache(page), page);
122 VM_BUG_ON_PAGE(!PageSwapBacked(page), page);
123
124 page_ref_add(page, nr);
125 SetPageSwapCache(page);
126
127 do {
128 xas_lock_irq(&xas);
129 xas_create_range(&xas);
130 if (xas_error(&xas))
131 goto unlock;
132 for (i = 0; i < nr; i++) {
133 VM_BUG_ON_PAGE(xas.xa_index != idx + i, page);
134 set_page_private(page + i, entry.val + i);
135 xas_store(&xas, page + i);
136 xas_next(&xas);
137 }
138 address_space->nrpages += nr;
139 __mod_node_page_state(page_pgdat(page), NR_FILE_PAGES, nr);
140 ADD_CACHE_INFO(add_total, nr);
141unlock:
142 xas_unlock_irq(&xas);
143 } while (xas_nomem(&xas, gfp));
144
145 if (!xas_error(&xas))
146 return 0;
147
148 ClearPageSwapCache(page);
149 page_ref_sub(page, nr);
150 return xas_error(&xas);
151}
152
153/*
154 * This must be called only on pages that have
155 * been verified to be in the swap cache.
156 */
157void __delete_from_swap_cache(struct page *page, swp_entry_t entry)
158{
159 struct address_space *address_space = swap_address_space(entry);
160 int i, nr = hpage_nr_pages(page);
161 pgoff_t idx = swp_offset(entry);
162 XA_STATE(xas, &address_space->i_pages, idx);
163
164 VM_BUG_ON_PAGE(!PageLocked(page), page);
165 VM_BUG_ON_PAGE(!PageSwapCache(page), page);
166 VM_BUG_ON_PAGE(PageWriteback(page), page);
167
168 for (i = 0; i < nr; i++) {
169 void *entry = xas_store(&xas, NULL);
170 VM_BUG_ON_PAGE(entry != page + i, entry);
171 set_page_private(page + i, 0);
172 xas_next(&xas);
173 }
174 ClearPageSwapCache(page);
175 address_space->nrpages -= nr;
176 __mod_node_page_state(page_pgdat(page), NR_FILE_PAGES, -nr);
177 ADD_CACHE_INFO(del_total, nr);
178}
179
180/**
181 * add_to_swap - allocate swap space for a page
182 * @page: page we want to move to swap
183 *
184 * Allocate swap space for the page and add the page to the
185 * swap cache. Caller needs to hold the page lock.
186 */
187int add_to_swap(struct page *page)
188{
189 swp_entry_t entry;
190 int err;
191
192 VM_BUG_ON_PAGE(!PageLocked(page), page);
193 VM_BUG_ON_PAGE(!PageUptodate(page), page);
194
195 entry = get_swap_page(page);
196 if (!entry.val)
197 return 0;
198
199 /*
200 * XArray node allocations from PF_MEMALLOC contexts could
201 * completely exhaust the page allocator. __GFP_NOMEMALLOC
202 * stops emergency reserves from being allocated.
203 *
204 * TODO: this could cause a theoretical memory reclaim
205 * deadlock in the swap out path.
206 */
207 /*
208 * Add it to the swap cache.
209 */
210 err = add_to_swap_cache(page, entry,
211 __GFP_HIGH|__GFP_NOMEMALLOC|__GFP_NOWARN);
212 if (err)
213 /*
214 * add_to_swap_cache() doesn't return -EEXIST, so we can safely
215 * clear SWAP_HAS_CACHE flag.
216 */
217 goto fail;
218 /*
219 * Normally the page will be dirtied in unmap because its pte should be
220 * dirty. A special case is MADV_FREE page. The page'e pte could have
221 * dirty bit cleared but the page's SwapBacked bit is still set because
222 * clearing the dirty bit and SwapBacked bit has no lock protected. For
223 * such page, unmap will not set dirty bit for it, so page reclaim will
224 * not write the page out. This can cause data corruption when the page
225 * is swap in later. Always setting the dirty bit for the page solves
226 * the problem.
227 */
228 set_page_dirty(page);
229
230 return 1;
231
232fail:
233 put_swap_page(page, entry);
234 return 0;
235}
236
237/*
238 * This must be called only on pages that have
239 * been verified to be in the swap cache and locked.
240 * It will never put the page into the free list,
241 * the caller has a reference on the page.
242 */
243void delete_from_swap_cache(struct page *page)
244{
245 swp_entry_t entry = { .val = page_private(page) };
246 struct address_space *address_space = swap_address_space(entry);
247
248 xa_lock_irq(&address_space->i_pages);
249 __delete_from_swap_cache(page, entry);
250 xa_unlock_irq(&address_space->i_pages);
251
252 put_swap_page(page, entry);
253 page_ref_sub(page, hpage_nr_pages(page));
254}
255
256/*
257 * If we are the only user, then try to free up the swap cache.
258 *
259 * Its ok to check for PageSwapCache without the page lock
260 * here because we are going to recheck again inside
261 * try_to_free_swap() _with_ the lock.
262 * - Marcelo
263 */
264static inline void free_swap_cache(struct page *page)
265{
266 if (PageSwapCache(page) && !page_mapped(page) && trylock_page(page)) {
267 try_to_free_swap(page);
268 unlock_page(page);
269 }
270}
271
272/*
273 * Perform a free_page(), also freeing any swap cache associated with
274 * this page if it is the last user of the page.
275 */
276void free_page_and_swap_cache(struct page *page)
277{
278 free_swap_cache(page);
279 if (!is_huge_zero_page(page))
280 put_page(page);
281}
282
283/*
284 * Passed an array of pages, drop them all from swapcache and then release
285 * them. They are removed from the LRU and freed if this is their last use.
286 */
287void free_pages_and_swap_cache(struct page **pages, int nr)
288{
289 struct page **pagep = pages;
290 int i;
291
292 lru_add_drain();
293 for (i = 0; i < nr; i++)
294 free_swap_cache(pagep[i]);
295 release_pages(pagep, nr);
296}
297
298static inline bool swap_use_vma_readahead(void)
299{
300 return READ_ONCE(enable_vma_readahead) && !atomic_read(&nr_rotate_swap);
301}
302
303/*
304 * Lookup a swap entry in the swap cache. A found page will be returned
305 * unlocked and with its refcount incremented - we rely on the kernel
306 * lock getting page table operations atomic even if we drop the page
307 * lock before returning.
308 */
309struct page *lookup_swap_cache(swp_entry_t entry, struct vm_area_struct *vma,
310 unsigned long addr)
311{
312 struct page *page;
313
314 page = find_get_page(swap_address_space(entry), swp_offset(entry));
315
316 INC_CACHE_INFO(find_total);
317 if (page) {
318 bool vma_ra = swap_use_vma_readahead();
319 bool readahead;
320
321 INC_CACHE_INFO(find_success);
322 /*
323 * At the moment, we don't support PG_readahead for anon THP
324 * so let's bail out rather than confusing the readahead stat.
325 */
326 if (unlikely(PageTransCompound(page)))
327 return page;
328
329 readahead = TestClearPageReadahead(page);
330 if (vma && vma_ra) {
331 unsigned long ra_val;
332 int win, hits;
333
334 ra_val = GET_SWAP_RA_VAL(vma);
335 win = SWAP_RA_WIN(ra_val);
336 hits = SWAP_RA_HITS(ra_val);
337 if (readahead)
338 hits = min_t(int, hits + 1, SWAP_RA_HITS_MAX);
339 atomic_long_set(&vma->swap_readahead_info,
340 SWAP_RA_VAL(addr, win, hits));
341 }
342
343 if (readahead) {
344 count_vm_event(SWAP_RA_HIT);
345 if (!vma || !vma_ra)
346 atomic_inc(&swapin_readahead_hits);
347 }
348 }
349
350 return page;
351}
352
353struct page *__read_swap_cache_async(swp_entry_t entry, gfp_t gfp_mask,
354 struct vm_area_struct *vma, unsigned long addr,
355 bool *new_page_allocated)
356{
357 struct page *found_page, *new_page = NULL;
358 struct address_space *swapper_space = swap_address_space(entry);
359 int err;
360 *new_page_allocated = false;
361
362 do {
363 /*
364 * First check the swap cache. Since this is normally
365 * called after lookup_swap_cache() failed, re-calling
366 * that would confuse statistics.
367 */
368 found_page = find_get_page(swapper_space, swp_offset(entry));
369 if (found_page)
370 break;
371
372 /*
373 * Just skip read ahead for unused swap slot.
374 * During swap_off when swap_slot_cache is disabled,
375 * we have to handle the race between putting
376 * swap entry in swap cache and marking swap slot
377 * as SWAP_HAS_CACHE. That's done in later part of code or
378 * else swap_off will be aborted if we return NULL.
379 */
380 if (!__swp_swapcount(entry) && swap_slot_cache_enabled)
381 break;
382
383 /*
384 * Get a new page to read into from swap.
385 */
386 if (!new_page) {
387 new_page = alloc_page_vma(gfp_mask, vma, addr);
388 if (!new_page)
389 break; /* Out of memory */
390 }
391
392 /*
393 * Swap entry may have been freed since our caller observed it.
394 */
395 err = swapcache_prepare(entry);
396 if (err == -EEXIST) {
397 /*
398 * We might race against get_swap_page() and stumble
399 * across a SWAP_HAS_CACHE swap_map entry whose page
400 * has not been brought into the swapcache yet.
401 */
402 cond_resched();
403 continue;
404 } else if (err) /* swp entry is obsolete ? */
405 break;
406
407 /* May fail (-ENOMEM) if XArray node allocation failed. */
408 __SetPageLocked(new_page);
409 __SetPageSwapBacked(new_page);
410 err = add_to_swap_cache(new_page, entry, gfp_mask & GFP_KERNEL);
411 if (likely(!err)) {
412 /* Initiate read into locked page */
413 SetPageWorkingset(new_page);
414 lru_cache_add_anon(new_page);
415 *new_page_allocated = true;
416 return new_page;
417 }
418 __ClearPageLocked(new_page);
419 /*
420 * add_to_swap_cache() doesn't return -EEXIST, so we can safely
421 * clear SWAP_HAS_CACHE flag.
422 */
423 put_swap_page(new_page, entry);
424 } while (err != -ENOMEM);
425
426 if (new_page)
427 put_page(new_page);
428 return found_page;
429}
430
431/*
432 * Locate a page of swap in physical memory, reserving swap cache space
433 * and reading the disk if it is not already cached.
434 * A failure return means that either the page allocation failed or that
435 * the swap entry is no longer in use.
436 */
437struct page *read_swap_cache_async(swp_entry_t entry, gfp_t gfp_mask,
438 struct vm_area_struct *vma, unsigned long addr, bool do_poll)
439{
440 bool page_was_allocated;
441 struct page *retpage = __read_swap_cache_async(entry, gfp_mask,
442 vma, addr, &page_was_allocated);
443
444 if (page_was_allocated)
445 swap_readpage(retpage, do_poll);
446
447 return retpage;
448}
449
450static unsigned int __swapin_nr_pages(unsigned long prev_offset,
451 unsigned long offset,
452 int hits,
453 int max_pages,
454 int prev_win)
455{
456 unsigned int pages, last_ra;
457
458 /*
459 * This heuristic has been found to work well on both sequential and
460 * random loads, swapping to hard disk or to SSD: please don't ask
461 * what the "+ 2" means, it just happens to work well, that's all.
462 */
463 pages = hits + 2;
464 if (pages == 2) {
465 /*
466 * We can have no readahead hits to judge by: but must not get
467 * stuck here forever, so check for an adjacent offset instead
468 * (and don't even bother to check whether swap type is same).
469 */
470 if (offset != prev_offset + 1 && offset != prev_offset - 1)
471 pages = 1;
472 } else {
473 unsigned int roundup = 4;
474 while (roundup < pages)
475 roundup <<= 1;
476 pages = roundup;
477 }
478
479 if (pages > max_pages)
480 pages = max_pages;
481
482 /* Don't shrink readahead too fast */
483 last_ra = prev_win / 2;
484 if (pages < last_ra)
485 pages = last_ra;
486
487 return pages;
488}
489
490static unsigned long swapin_nr_pages(unsigned long offset)
491{
492 static unsigned long prev_offset;
493 unsigned int hits, pages, max_pages;
494 static atomic_t last_readahead_pages;
495
496 max_pages = 1 << READ_ONCE(page_cluster);
497 if (max_pages <= 1)
498 return 1;
499
500 hits = atomic_xchg(&swapin_readahead_hits, 0);
501 pages = __swapin_nr_pages(prev_offset, offset, hits, max_pages,
502 atomic_read(&last_readahead_pages));
503 if (!hits)
504 prev_offset = offset;
505 atomic_set(&last_readahead_pages, pages);
506
507 return pages;
508}
509
510/**
511 * swap_cluster_readahead - swap in pages in hope we need them soon
512 * @entry: swap entry of this memory
513 * @gfp_mask: memory allocation flags
514 * @vmf: fault information
515 *
516 * Returns the struct page for entry and addr, after queueing swapin.
517 *
518 * Primitive swap readahead code. We simply read an aligned block of
519 * (1 << page_cluster) entries in the swap area. This method is chosen
520 * because it doesn't cost us any seek time. We also make sure to queue
521 * the 'original' request together with the readahead ones...
522 *
523 * This has been extended to use the NUMA policies from the mm triggering
524 * the readahead.
525 *
526 * Caller must hold read mmap_sem if vmf->vma is not NULL.
527 */
528struct page *swap_cluster_readahead(swp_entry_t entry, gfp_t gfp_mask,
529 struct vm_fault *vmf)
530{
531 struct page *page;
532 unsigned long entry_offset = swp_offset(entry);
533 unsigned long offset = entry_offset;
534 unsigned long start_offset, end_offset;
535 unsigned long mask;
536 struct swap_info_struct *si = swp_swap_info(entry);
537 struct blk_plug plug;
538 bool do_poll = true, page_allocated;
539 struct vm_area_struct *vma = vmf->vma;
540 unsigned long addr = vmf->address;
541
542 mask = swapin_nr_pages(offset) - 1;
543 if (!mask)
544 goto skip;
545
546 /* Test swap type to make sure the dereference is safe */
547 if (likely(si->flags & (SWP_BLKDEV | SWP_FS))) {
548 struct inode *inode = si->swap_file->f_mapping->host;
549 if (inode_read_congested(inode))
550 goto skip;
551 }
552
553 do_poll = false;
554 /* Read a page_cluster sized and aligned cluster around offset. */
555 start_offset = offset & ~mask;
556 end_offset = offset | mask;
557 if (!start_offset) /* First page is swap header. */
558 start_offset++;
559 if (end_offset >= si->max)
560 end_offset = si->max - 1;
561
562 blk_start_plug(&plug);
563 for (offset = start_offset; offset <= end_offset ; offset++) {
564 /* Ok, do the async read-ahead now */
565 page = __read_swap_cache_async(
566 swp_entry(swp_type(entry), offset),
567 gfp_mask, vma, addr, &page_allocated);
568 if (!page)
569 continue;
570 if (page_allocated) {
571 swap_readpage(page, false);
572 if (offset != entry_offset) {
573 SetPageReadahead(page);
574 count_vm_event(SWAP_RA);
575 }
576 }
577 put_page(page);
578 }
579 blk_finish_plug(&plug);
580
581 lru_add_drain(); /* Push any new pages onto the LRU now */
582skip:
583 return read_swap_cache_async(entry, gfp_mask, vma, addr, do_poll);
584}
585
586int init_swap_address_space(unsigned int type, unsigned long nr_pages)
587{
588 struct address_space *spaces, *space;
589 unsigned int i, nr;
590
591 nr = DIV_ROUND_UP(nr_pages, SWAP_ADDRESS_SPACE_PAGES);
592 spaces = kvcalloc(nr, sizeof(struct address_space), GFP_KERNEL);
593 if (!spaces)
594 return -ENOMEM;
595 for (i = 0; i < nr; i++) {
596 space = spaces + i;
597 xa_init_flags(&space->i_pages, XA_FLAGS_LOCK_IRQ);
598 atomic_set(&space->i_mmap_writable, 0);
599 space->a_ops = &swap_aops;
600 /* swap cache doesn't use writeback related tags */
601 mapping_set_no_writeback_tags(space);
602 }
603 nr_swapper_spaces[type] = nr;
604 rcu_assign_pointer(swapper_spaces[type], spaces);
605
606 return 0;
607}
608
609void exit_swap_address_space(unsigned int type)
610{
611 struct address_space *spaces;
612
613 spaces = swapper_spaces[type];
614 nr_swapper_spaces[type] = 0;
615 rcu_assign_pointer(swapper_spaces[type], NULL);
616 synchronize_rcu();
617 kvfree(spaces);
618}
619
620static inline void swap_ra_clamp_pfn(struct vm_area_struct *vma,
621 unsigned long faddr,
622 unsigned long lpfn,
623 unsigned long rpfn,
624 unsigned long *start,
625 unsigned long *end)
626{
627 *start = max3(lpfn, PFN_DOWN(vma->vm_start),
628 PFN_DOWN(faddr & PMD_MASK));
629 *end = min3(rpfn, PFN_DOWN(vma->vm_end),
630 PFN_DOWN((faddr & PMD_MASK) + PMD_SIZE));
631}
632
633static void swap_ra_info(struct vm_fault *vmf,
634 struct vma_swap_readahead *ra_info)
635{
636 struct vm_area_struct *vma = vmf->vma;
637 unsigned long ra_val;
638 swp_entry_t entry;
639 unsigned long faddr, pfn, fpfn;
640 unsigned long start, end;
641 pte_t *pte, *orig_pte;
642 unsigned int max_win, hits, prev_win, win, left;
643#ifndef CONFIG_64BIT
644 pte_t *tpte;
645#endif
646
647 max_win = 1 << min_t(unsigned int, READ_ONCE(page_cluster),
648 SWAP_RA_ORDER_CEILING);
649 if (max_win == 1) {
650 ra_info->win = 1;
651 return;
652 }
653
654 faddr = vmf->address;
655 orig_pte = pte = pte_offset_map(vmf->pmd, faddr);
656 entry = pte_to_swp_entry(*pte);
657 if ((unlikely(non_swap_entry(entry)))) {
658 pte_unmap(orig_pte);
659 return;
660 }
661
662 fpfn = PFN_DOWN(faddr);
663 ra_val = GET_SWAP_RA_VAL(vma);
664 pfn = PFN_DOWN(SWAP_RA_ADDR(ra_val));
665 prev_win = SWAP_RA_WIN(ra_val);
666 hits = SWAP_RA_HITS(ra_val);
667 ra_info->win = win = __swapin_nr_pages(pfn, fpfn, hits,
668 max_win, prev_win);
669 atomic_long_set(&vma->swap_readahead_info,
670 SWAP_RA_VAL(faddr, win, 0));
671
672 if (win == 1) {
673 pte_unmap(orig_pte);
674 return;
675 }
676
677 /* Copy the PTEs because the page table may be unmapped */
678 if (fpfn == pfn + 1)
679 swap_ra_clamp_pfn(vma, faddr, fpfn, fpfn + win, &start, &end);
680 else if (pfn == fpfn + 1)
681 swap_ra_clamp_pfn(vma, faddr, fpfn - win + 1, fpfn + 1,
682 &start, &end);
683 else {
684 left = (win - 1) / 2;
685 swap_ra_clamp_pfn(vma, faddr, fpfn - left, fpfn + win - left,
686 &start, &end);
687 }
688 ra_info->nr_pte = end - start;
689 ra_info->offset = fpfn - start;
690 pte -= ra_info->offset;
691#ifdef CONFIG_64BIT
692 ra_info->ptes = pte;
693#else
694 tpte = ra_info->ptes;
695 for (pfn = start; pfn != end; pfn++)
696 *tpte++ = *pte++;
697#endif
698 pte_unmap(orig_pte);
699}
700
701/**
702 * swap_vma_readahead - swap in pages in hope we need them soon
703 * @entry: swap entry of this memory
704 * @gfp_mask: memory allocation flags
705 * @vmf: fault information
706 *
707 * Returns the struct page for entry and addr, after queueing swapin.
708 *
709 * Primitive swap readahead code. We simply read in a few pages whoes
710 * virtual addresses are around the fault address in the same vma.
711 *
712 * Caller must hold read mmap_sem if vmf->vma is not NULL.
713 *
714 */
715static struct page *swap_vma_readahead(swp_entry_t fentry, gfp_t gfp_mask,
716 struct vm_fault *vmf)
717{
718 struct blk_plug plug;
719 struct vm_area_struct *vma = vmf->vma;
720 struct page *page;
721 pte_t *pte, pentry;
722 swp_entry_t entry;
723 unsigned int i;
724 bool page_allocated;
725 struct vma_swap_readahead ra_info = {0,};
726
727 swap_ra_info(vmf, &ra_info);
728 if (ra_info.win == 1)
729 goto skip;
730
731 blk_start_plug(&plug);
732 for (i = 0, pte = ra_info.ptes; i < ra_info.nr_pte;
733 i++, pte++) {
734 pentry = *pte;
735 if (pte_none(pentry))
736 continue;
737 if (pte_present(pentry))
738 continue;
739 entry = pte_to_swp_entry(pentry);
740 if (unlikely(non_swap_entry(entry)))
741 continue;
742 page = __read_swap_cache_async(entry, gfp_mask, vma,
743 vmf->address, &page_allocated);
744 if (!page)
745 continue;
746 if (page_allocated) {
747 swap_readpage(page, false);
748 if (i != ra_info.offset) {
749 SetPageReadahead(page);
750 count_vm_event(SWAP_RA);
751 }
752 }
753 put_page(page);
754 }
755 blk_finish_plug(&plug);
756 lru_add_drain();
757skip:
758 return read_swap_cache_async(fentry, gfp_mask, vma, vmf->address,
759 ra_info.win == 1);
760}
761
762/**
763 * swapin_readahead - swap in pages in hope we need them soon
764 * @entry: swap entry of this memory
765 * @gfp_mask: memory allocation flags
766 * @vmf: fault information
767 *
768 * Returns the struct page for entry and addr, after queueing swapin.
769 *
770 * It's a main entry function for swap readahead. By the configuration,
771 * it will read ahead blocks by cluster-based(ie, physical disk based)
772 * or vma-based(ie, virtual address based on faulty address) readahead.
773 */
774struct page *swapin_readahead(swp_entry_t entry, gfp_t gfp_mask,
775 struct vm_fault *vmf)
776{
777 return swap_use_vma_readahead() ?
778 swap_vma_readahead(entry, gfp_mask, vmf) :
779 swap_cluster_readahead(entry, gfp_mask, vmf);
780}
781
782#ifdef CONFIG_SYSFS
783static ssize_t vma_ra_enabled_show(struct kobject *kobj,
784 struct kobj_attribute *attr, char *buf)
785{
786 return sprintf(buf, "%s\n", enable_vma_readahead ? "true" : "false");
787}
788static ssize_t vma_ra_enabled_store(struct kobject *kobj,
789 struct kobj_attribute *attr,
790 const char *buf, size_t count)
791{
792 if (!strncmp(buf, "true", 4) || !strncmp(buf, "1", 1))
793 enable_vma_readahead = true;
794 else if (!strncmp(buf, "false", 5) || !strncmp(buf, "0", 1))
795 enable_vma_readahead = false;
796 else
797 return -EINVAL;
798
799 return count;
800}
801static struct kobj_attribute vma_ra_enabled_attr =
802 __ATTR(vma_ra_enabled, 0644, vma_ra_enabled_show,
803 vma_ra_enabled_store);
804
805static struct attribute *swap_attrs[] = {
806 &vma_ra_enabled_attr.attr,
807 NULL,
808};
809
810static struct attribute_group swap_attr_group = {
811 .attrs = swap_attrs,
812};
813
814static int __init swap_init_sysfs(void)
815{
816 int err;
817 struct kobject *swap_kobj;
818
819 swap_kobj = kobject_create_and_add("swap", mm_kobj);
820 if (!swap_kobj) {
821 pr_err("failed to create swap kobject\n");
822 return -ENOMEM;
823 }
824 err = sysfs_create_group(swap_kobj, &swap_attr_group);
825 if (err) {
826 pr_err("failed to register swap group\n");
827 goto delete_obj;
828 }
829 return 0;
830
831delete_obj:
832 kobject_put(swap_kobj);
833 return err;
834}
835subsys_initcall(swap_init_sysfs);
836#endif
837