Linux Audio

Check our new training course

Real-Time Linux with PREEMPT_RT training

Feb 18-20, 2025
Register
Loading...
v5.4
   1// SPDX-License-Identifier: GPL-2.0-only
   2/*
   3 *  linux/mm/swapfile.c
   4 *
   5 *  Copyright (C) 1991, 1992, 1993, 1994  Linus Torvalds
   6 *  Swap reorganised 29.12.95, Stephen Tweedie
   7 */
   8
   9#include <linux/mm.h>
  10#include <linux/sched/mm.h>
  11#include <linux/sched/task.h>
  12#include <linux/hugetlb.h>
  13#include <linux/mman.h>
  14#include <linux/slab.h>
  15#include <linux/kernel_stat.h>
  16#include <linux/swap.h>
  17#include <linux/vmalloc.h>
  18#include <linux/pagemap.h>
  19#include <linux/namei.h>
  20#include <linux/shmem_fs.h>
  21#include <linux/blkdev.h>
  22#include <linux/random.h>
  23#include <linux/writeback.h>
  24#include <linux/proc_fs.h>
  25#include <linux/seq_file.h>
  26#include <linux/init.h>
  27#include <linux/ksm.h>
  28#include <linux/rmap.h>
  29#include <linux/security.h>
  30#include <linux/backing-dev.h>
  31#include <linux/mutex.h>
  32#include <linux/capability.h>
  33#include <linux/syscalls.h>
  34#include <linux/memcontrol.h>
  35#include <linux/poll.h>
  36#include <linux/oom.h>
  37#include <linux/frontswap.h>
  38#include <linux/swapfile.h>
  39#include <linux/export.h>
  40#include <linux/swap_slots.h>
  41#include <linux/sort.h>
  42
  43#include <asm/pgtable.h>
  44#include <asm/tlbflush.h>
  45#include <linux/swapops.h>
  46#include <linux/swap_cgroup.h>
  47
  48static bool swap_count_continued(struct swap_info_struct *, pgoff_t,
  49				 unsigned char);
  50static void free_swap_count_continuations(struct swap_info_struct *);
  51static sector_t map_swap_entry(swp_entry_t, struct block_device**);
  52
  53DEFINE_SPINLOCK(swap_lock);
  54static unsigned int nr_swapfiles;
  55atomic_long_t nr_swap_pages;
  56/*
  57 * Some modules use swappable objects and may try to swap them out under
  58 * memory pressure (via the shrinker). Before doing so, they may wish to
  59 * check to see if any swap space is available.
  60 */
  61EXPORT_SYMBOL_GPL(nr_swap_pages);
  62/* protected with swap_lock. reading in vm_swap_full() doesn't need lock */
  63long total_swap_pages;
  64static int least_priority = -1;
  65
  66static const char Bad_file[] = "Bad swap file entry ";
  67static const char Unused_file[] = "Unused swap file entry ";
  68static const char Bad_offset[] = "Bad swap offset entry ";
  69static const char Unused_offset[] = "Unused swap offset entry ";
  70
  71/*
  72 * all active swap_info_structs
  73 * protected with swap_lock, and ordered by priority.
  74 */
  75PLIST_HEAD(swap_active_head);
  76
  77/*
  78 * all available (active, not full) swap_info_structs
  79 * protected with swap_avail_lock, ordered by priority.
  80 * This is used by get_swap_page() instead of swap_active_head
  81 * because swap_active_head includes all swap_info_structs,
  82 * but get_swap_page() doesn't need to look at full ones.
  83 * This uses its own lock instead of swap_lock because when a
  84 * swap_info_struct changes between not-full/full, it needs to
  85 * add/remove itself to/from this list, but the swap_info_struct->lock
  86 * is held and the locking order requires swap_lock to be taken
  87 * before any swap_info_struct->lock.
  88 */
  89static struct plist_head *swap_avail_heads;
  90static DEFINE_SPINLOCK(swap_avail_lock);
  91
  92struct swap_info_struct *swap_info[MAX_SWAPFILES];
  93
  94static DEFINE_MUTEX(swapon_mutex);
  95
  96static DECLARE_WAIT_QUEUE_HEAD(proc_poll_wait);
  97/* Activity counter to indicate that a swapon or swapoff has occurred */
  98static atomic_t proc_poll_event = ATOMIC_INIT(0);
  99
 100atomic_t nr_rotate_swap = ATOMIC_INIT(0);
 101
 102static struct swap_info_struct *swap_type_to_swap_info(int type)
 103{
 104	if (type >= READ_ONCE(nr_swapfiles))
 105		return NULL;
 106
 107	smp_rmb();	/* Pairs with smp_wmb in alloc_swap_info. */
 108	return READ_ONCE(swap_info[type]);
 109}
 110
 111static inline unsigned char swap_count(unsigned char ent)
 112{
 113	return ent & ~SWAP_HAS_CACHE;	/* may include COUNT_CONTINUED flag */
 114}
 115
 116/* Reclaim the swap entry anyway if possible */
 117#define TTRS_ANYWAY		0x1
 118/*
 119 * Reclaim the swap entry if there are no more mappings of the
 120 * corresponding page
 121 */
 122#define TTRS_UNMAPPED		0x2
 123/* Reclaim the swap entry if swap is getting full*/
 124#define TTRS_FULL		0x4
 125
 126/* returns 1 if swap entry is freed */
 127static int __try_to_reclaim_swap(struct swap_info_struct *si,
 128				 unsigned long offset, unsigned long flags)
 129{
 130	swp_entry_t entry = swp_entry(si->type, offset);
 131	struct page *page;
 132	int ret = 0;
 133
 134	page = find_get_page(swap_address_space(entry), offset);
 135	if (!page)
 136		return 0;
 137	/*
 138	 * When this function is called from scan_swap_map_slots() and it's
 139	 * called by vmscan.c at reclaiming pages. So, we hold a lock on a page,
 140	 * here. We have to use trylock for avoiding deadlock. This is a special
 141	 * case and you should use try_to_free_swap() with explicit lock_page()
 142	 * in usual operations.
 143	 */
 144	if (trylock_page(page)) {
 145		if ((flags & TTRS_ANYWAY) ||
 146		    ((flags & TTRS_UNMAPPED) && !page_mapped(page)) ||
 147		    ((flags & TTRS_FULL) && mem_cgroup_swap_full(page)))
 148			ret = try_to_free_swap(page);
 149		unlock_page(page);
 150	}
 151	put_page(page);
 152	return ret;
 153}
 154
 155static inline struct swap_extent *first_se(struct swap_info_struct *sis)
 156{
 157	struct rb_node *rb = rb_first(&sis->swap_extent_root);
 158	return rb_entry(rb, struct swap_extent, rb_node);
 159}
 160
 161static inline struct swap_extent *next_se(struct swap_extent *se)
 162{
 163	struct rb_node *rb = rb_next(&se->rb_node);
 164	return rb ? rb_entry(rb, struct swap_extent, rb_node) : NULL;
 165}
 166
 167/*
 168 * swapon tell device that all the old swap contents can be discarded,
 169 * to allow the swap device to optimize its wear-levelling.
 170 */
 171static int discard_swap(struct swap_info_struct *si)
 172{
 173	struct swap_extent *se;
 174	sector_t start_block;
 175	sector_t nr_blocks;
 176	int err = 0;
 177
 178	/* Do not discard the swap header page! */
 179	se = first_se(si);
 180	start_block = (se->start_block + 1) << (PAGE_SHIFT - 9);
 181	nr_blocks = ((sector_t)se->nr_pages - 1) << (PAGE_SHIFT - 9);
 182	if (nr_blocks) {
 183		err = blkdev_issue_discard(si->bdev, start_block,
 184				nr_blocks, GFP_KERNEL, 0);
 185		if (err)
 186			return err;
 187		cond_resched();
 188	}
 189
 190	for (se = next_se(se); se; se = next_se(se)) {
 191		start_block = se->start_block << (PAGE_SHIFT - 9);
 192		nr_blocks = (sector_t)se->nr_pages << (PAGE_SHIFT - 9);
 193
 194		err = blkdev_issue_discard(si->bdev, start_block,
 195				nr_blocks, GFP_KERNEL, 0);
 196		if (err)
 197			break;
 198
 199		cond_resched();
 200	}
 201	return err;		/* That will often be -EOPNOTSUPP */
 202}
 203
 204static struct swap_extent *
 205offset_to_swap_extent(struct swap_info_struct *sis, unsigned long offset)
 206{
 207	struct swap_extent *se;
 208	struct rb_node *rb;
 209
 210	rb = sis->swap_extent_root.rb_node;
 211	while (rb) {
 212		se = rb_entry(rb, struct swap_extent, rb_node);
 213		if (offset < se->start_page)
 214			rb = rb->rb_left;
 215		else if (offset >= se->start_page + se->nr_pages)
 216			rb = rb->rb_right;
 217		else
 218			return se;
 219	}
 220	/* It *must* be present */
 221	BUG();
 222}
 223
 224/*
 225 * swap allocation tell device that a cluster of swap can now be discarded,
 226 * to allow the swap device to optimize its wear-levelling.
 227 */
 228static void discard_swap_cluster(struct swap_info_struct *si,
 229				 pgoff_t start_page, pgoff_t nr_pages)
 230{
 231	struct swap_extent *se = offset_to_swap_extent(si, start_page);
 232
 233	while (nr_pages) {
 234		pgoff_t offset = start_page - se->start_page;
 235		sector_t start_block = se->start_block + offset;
 236		sector_t nr_blocks = se->nr_pages - offset;
 237
 238		if (nr_blocks > nr_pages)
 239			nr_blocks = nr_pages;
 240		start_page += nr_blocks;
 241		nr_pages -= nr_blocks;
 242
 243		start_block <<= PAGE_SHIFT - 9;
 244		nr_blocks <<= PAGE_SHIFT - 9;
 245		if (blkdev_issue_discard(si->bdev, start_block,
 246					nr_blocks, GFP_NOIO, 0))
 247			break;
 248
 249		se = next_se(se);
 250	}
 251}
 252
 253#ifdef CONFIG_THP_SWAP
 254#define SWAPFILE_CLUSTER	HPAGE_PMD_NR
 255
 256#define swap_entry_size(size)	(size)
 257#else
 258#define SWAPFILE_CLUSTER	256
 259
 260/*
 261 * Define swap_entry_size() as constant to let compiler to optimize
 262 * out some code if !CONFIG_THP_SWAP
 263 */
 264#define swap_entry_size(size)	1
 265#endif
 266#define LATENCY_LIMIT		256
 267
 268static inline void cluster_set_flag(struct swap_cluster_info *info,
 269	unsigned int flag)
 270{
 271	info->flags = flag;
 272}
 273
 274static inline unsigned int cluster_count(struct swap_cluster_info *info)
 275{
 276	return info->data;
 277}
 278
 279static inline void cluster_set_count(struct swap_cluster_info *info,
 280				     unsigned int c)
 281{
 282	info->data = c;
 283}
 284
 285static inline void cluster_set_count_flag(struct swap_cluster_info *info,
 286					 unsigned int c, unsigned int f)
 287{
 288	info->flags = f;
 289	info->data = c;
 290}
 291
 292static inline unsigned int cluster_next(struct swap_cluster_info *info)
 293{
 294	return info->data;
 295}
 296
 297static inline void cluster_set_next(struct swap_cluster_info *info,
 298				    unsigned int n)
 299{
 300	info->data = n;
 301}
 302
 303static inline void cluster_set_next_flag(struct swap_cluster_info *info,
 304					 unsigned int n, unsigned int f)
 305{
 306	info->flags = f;
 307	info->data = n;
 308}
 309
 310static inline bool cluster_is_free(struct swap_cluster_info *info)
 311{
 312	return info->flags & CLUSTER_FLAG_FREE;
 313}
 314
 315static inline bool cluster_is_null(struct swap_cluster_info *info)
 316{
 317	return info->flags & CLUSTER_FLAG_NEXT_NULL;
 318}
 319
 320static inline void cluster_set_null(struct swap_cluster_info *info)
 321{
 322	info->flags = CLUSTER_FLAG_NEXT_NULL;
 323	info->data = 0;
 324}
 325
 326static inline bool cluster_is_huge(struct swap_cluster_info *info)
 327{
 328	if (IS_ENABLED(CONFIG_THP_SWAP))
 329		return info->flags & CLUSTER_FLAG_HUGE;
 330	return false;
 331}
 332
 333static inline void cluster_clear_huge(struct swap_cluster_info *info)
 334{
 335	info->flags &= ~CLUSTER_FLAG_HUGE;
 336}
 337
 338static inline struct swap_cluster_info *lock_cluster(struct swap_info_struct *si,
 339						     unsigned long offset)
 340{
 341	struct swap_cluster_info *ci;
 342
 343	ci = si->cluster_info;
 344	if (ci) {
 345		ci += offset / SWAPFILE_CLUSTER;
 346		spin_lock(&ci->lock);
 347	}
 348	return ci;
 349}
 350
 351static inline void unlock_cluster(struct swap_cluster_info *ci)
 352{
 353	if (ci)
 354		spin_unlock(&ci->lock);
 355}
 356
 357/*
 358 * Determine the locking method in use for this device.  Return
 359 * swap_cluster_info if SSD-style cluster-based locking is in place.
 360 */
 361static inline struct swap_cluster_info *lock_cluster_or_swap_info(
 362		struct swap_info_struct *si, unsigned long offset)
 363{
 364	struct swap_cluster_info *ci;
 365
 366	/* Try to use fine-grained SSD-style locking if available: */
 367	ci = lock_cluster(si, offset);
 368	/* Otherwise, fall back to traditional, coarse locking: */
 369	if (!ci)
 370		spin_lock(&si->lock);
 371
 372	return ci;
 373}
 374
 375static inline void unlock_cluster_or_swap_info(struct swap_info_struct *si,
 376					       struct swap_cluster_info *ci)
 377{
 378	if (ci)
 379		unlock_cluster(ci);
 380	else
 381		spin_unlock(&si->lock);
 382}
 383
 384static inline bool cluster_list_empty(struct swap_cluster_list *list)
 385{
 386	return cluster_is_null(&list->head);
 387}
 388
 389static inline unsigned int cluster_list_first(struct swap_cluster_list *list)
 390{
 391	return cluster_next(&list->head);
 392}
 393
 394static void cluster_list_init(struct swap_cluster_list *list)
 395{
 396	cluster_set_null(&list->head);
 397	cluster_set_null(&list->tail);
 398}
 399
 400static void cluster_list_add_tail(struct swap_cluster_list *list,
 401				  struct swap_cluster_info *ci,
 402				  unsigned int idx)
 403{
 404	if (cluster_list_empty(list)) {
 405		cluster_set_next_flag(&list->head, idx, 0);
 406		cluster_set_next_flag(&list->tail, idx, 0);
 407	} else {
 408		struct swap_cluster_info *ci_tail;
 409		unsigned int tail = cluster_next(&list->tail);
 410
 411		/*
 412		 * Nested cluster lock, but both cluster locks are
 413		 * only acquired when we held swap_info_struct->lock
 414		 */
 415		ci_tail = ci + tail;
 416		spin_lock_nested(&ci_tail->lock, SINGLE_DEPTH_NESTING);
 417		cluster_set_next(ci_tail, idx);
 418		spin_unlock(&ci_tail->lock);
 419		cluster_set_next_flag(&list->tail, idx, 0);
 420	}
 421}
 422
 423static unsigned int cluster_list_del_first(struct swap_cluster_list *list,
 424					   struct swap_cluster_info *ci)
 425{
 426	unsigned int idx;
 427
 428	idx = cluster_next(&list->head);
 429	if (cluster_next(&list->tail) == idx) {
 430		cluster_set_null(&list->head);
 431		cluster_set_null(&list->tail);
 432	} else
 433		cluster_set_next_flag(&list->head,
 434				      cluster_next(&ci[idx]), 0);
 435
 436	return idx;
 437}
 438
 439/* Add a cluster to discard list and schedule it to do discard */
 440static void swap_cluster_schedule_discard(struct swap_info_struct *si,
 441		unsigned int idx)
 442{
 443	/*
 444	 * If scan_swap_map() can't find a free cluster, it will check
 445	 * si->swap_map directly. To make sure the discarding cluster isn't
 446	 * taken by scan_swap_map(), mark the swap entries bad (occupied). It
 447	 * will be cleared after discard
 448	 */
 449	memset(si->swap_map + idx * SWAPFILE_CLUSTER,
 450			SWAP_MAP_BAD, SWAPFILE_CLUSTER);
 451
 452	cluster_list_add_tail(&si->discard_clusters, si->cluster_info, idx);
 453
 454	schedule_work(&si->discard_work);
 455}
 456
 457static void __free_cluster(struct swap_info_struct *si, unsigned long idx)
 458{
 459	struct swap_cluster_info *ci = si->cluster_info;
 460
 461	cluster_set_flag(ci + idx, CLUSTER_FLAG_FREE);
 462	cluster_list_add_tail(&si->free_clusters, ci, idx);
 463}
 464
 465/*
 466 * Doing discard actually. After a cluster discard is finished, the cluster
 467 * will be added to free cluster list. caller should hold si->lock.
 468*/
 469static void swap_do_scheduled_discard(struct swap_info_struct *si)
 470{
 471	struct swap_cluster_info *info, *ci;
 472	unsigned int idx;
 473
 474	info = si->cluster_info;
 475
 476	while (!cluster_list_empty(&si->discard_clusters)) {
 477		idx = cluster_list_del_first(&si->discard_clusters, info);
 478		spin_unlock(&si->lock);
 479
 480		discard_swap_cluster(si, idx * SWAPFILE_CLUSTER,
 481				SWAPFILE_CLUSTER);
 482
 483		spin_lock(&si->lock);
 484		ci = lock_cluster(si, idx * SWAPFILE_CLUSTER);
 485		__free_cluster(si, idx);
 486		memset(si->swap_map + idx * SWAPFILE_CLUSTER,
 487				0, SWAPFILE_CLUSTER);
 488		unlock_cluster(ci);
 489	}
 490}
 491
 492static void swap_discard_work(struct work_struct *work)
 493{
 494	struct swap_info_struct *si;
 495
 496	si = container_of(work, struct swap_info_struct, discard_work);
 497
 498	spin_lock(&si->lock);
 499	swap_do_scheduled_discard(si);
 500	spin_unlock(&si->lock);
 501}
 502
 503static void alloc_cluster(struct swap_info_struct *si, unsigned long idx)
 504{
 505	struct swap_cluster_info *ci = si->cluster_info;
 506
 507	VM_BUG_ON(cluster_list_first(&si->free_clusters) != idx);
 508	cluster_list_del_first(&si->free_clusters, ci);
 509	cluster_set_count_flag(ci + idx, 0, 0);
 510}
 511
 512static void free_cluster(struct swap_info_struct *si, unsigned long idx)
 513{
 514	struct swap_cluster_info *ci = si->cluster_info + idx;
 515
 516	VM_BUG_ON(cluster_count(ci) != 0);
 517	/*
 518	 * If the swap is discardable, prepare discard the cluster
 519	 * instead of free it immediately. The cluster will be freed
 520	 * after discard.
 521	 */
 522	if ((si->flags & (SWP_WRITEOK | SWP_PAGE_DISCARD)) ==
 523	    (SWP_WRITEOK | SWP_PAGE_DISCARD)) {
 524		swap_cluster_schedule_discard(si, idx);
 525		return;
 526	}
 527
 528	__free_cluster(si, idx);
 529}
 530
 531/*
 532 * The cluster corresponding to page_nr will be used. The cluster will be
 533 * removed from free cluster list and its usage counter will be increased.
 534 */
 535static void inc_cluster_info_page(struct swap_info_struct *p,
 536	struct swap_cluster_info *cluster_info, unsigned long page_nr)
 537{
 538	unsigned long idx = page_nr / SWAPFILE_CLUSTER;
 539
 540	if (!cluster_info)
 541		return;
 542	if (cluster_is_free(&cluster_info[idx]))
 543		alloc_cluster(p, idx);
 544
 545	VM_BUG_ON(cluster_count(&cluster_info[idx]) >= SWAPFILE_CLUSTER);
 546	cluster_set_count(&cluster_info[idx],
 547		cluster_count(&cluster_info[idx]) + 1);
 548}
 549
 550/*
 551 * The cluster corresponding to page_nr decreases one usage. If the usage
 552 * counter becomes 0, which means no page in the cluster is in using, we can
 553 * optionally discard the cluster and add it to free cluster list.
 554 */
 555static void dec_cluster_info_page(struct swap_info_struct *p,
 556	struct swap_cluster_info *cluster_info, unsigned long page_nr)
 557{
 558	unsigned long idx = page_nr / SWAPFILE_CLUSTER;
 559
 560	if (!cluster_info)
 561		return;
 562
 563	VM_BUG_ON(cluster_count(&cluster_info[idx]) == 0);
 564	cluster_set_count(&cluster_info[idx],
 565		cluster_count(&cluster_info[idx]) - 1);
 566
 567	if (cluster_count(&cluster_info[idx]) == 0)
 568		free_cluster(p, idx);
 569}
 570
 571/*
 572 * It's possible scan_swap_map() uses a free cluster in the middle of free
 573 * cluster list. Avoiding such abuse to avoid list corruption.
 574 */
 575static bool
 576scan_swap_map_ssd_cluster_conflict(struct swap_info_struct *si,
 577	unsigned long offset)
 578{
 579	struct percpu_cluster *percpu_cluster;
 580	bool conflict;
 581
 582	offset /= SWAPFILE_CLUSTER;
 583	conflict = !cluster_list_empty(&si->free_clusters) &&
 584		offset != cluster_list_first(&si->free_clusters) &&
 585		cluster_is_free(&si->cluster_info[offset]);
 586
 587	if (!conflict)
 588		return false;
 589
 590	percpu_cluster = this_cpu_ptr(si->percpu_cluster);
 591	cluster_set_null(&percpu_cluster->index);
 592	return true;
 593}
 594
 595/*
 596 * Try to get a swap entry from current cpu's swap entry pool (a cluster). This
 597 * might involve allocating a new cluster for current CPU too.
 598 */
 599static bool scan_swap_map_try_ssd_cluster(struct swap_info_struct *si,
 600	unsigned long *offset, unsigned long *scan_base)
 601{
 602	struct percpu_cluster *cluster;
 603	struct swap_cluster_info *ci;
 604	bool found_free;
 605	unsigned long tmp, max;
 606
 607new_cluster:
 608	cluster = this_cpu_ptr(si->percpu_cluster);
 609	if (cluster_is_null(&cluster->index)) {
 610		if (!cluster_list_empty(&si->free_clusters)) {
 611			cluster->index = si->free_clusters.head;
 612			cluster->next = cluster_next(&cluster->index) *
 613					SWAPFILE_CLUSTER;
 614		} else if (!cluster_list_empty(&si->discard_clusters)) {
 615			/*
 616			 * we don't have free cluster but have some clusters in
 617			 * discarding, do discard now and reclaim them
 
 618			 */
 619			swap_do_scheduled_discard(si);
 620			*scan_base = *offset = si->cluster_next;
 
 621			goto new_cluster;
 622		} else
 623			return false;
 624	}
 625
 626	found_free = false;
 627
 628	/*
 629	 * Other CPUs can use our cluster if they can't find a free cluster,
 630	 * check if there is still free entry in the cluster
 631	 */
 632	tmp = cluster->next;
 633	max = min_t(unsigned long, si->max,
 634		    (cluster_next(&cluster->index) + 1) * SWAPFILE_CLUSTER);
 635	if (tmp >= max) {
 636		cluster_set_null(&cluster->index);
 637		goto new_cluster;
 638	}
 639	ci = lock_cluster(si, tmp);
 640	while (tmp < max) {
 641		if (!si->swap_map[tmp]) {
 642			found_free = true;
 643			break;
 644		}
 645		tmp++;
 646	}
 647	unlock_cluster(ci);
 648	if (!found_free) {
 649		cluster_set_null(&cluster->index);
 650		goto new_cluster;
 651	}
 652	cluster->next = tmp + 1;
 653	*offset = tmp;
 654	*scan_base = tmp;
 655	return found_free;
 656}
 657
 658static void __del_from_avail_list(struct swap_info_struct *p)
 659{
 660	int nid;
 661
 662	for_each_node(nid)
 663		plist_del(&p->avail_lists[nid], &swap_avail_heads[nid]);
 664}
 665
 666static void del_from_avail_list(struct swap_info_struct *p)
 667{
 668	spin_lock(&swap_avail_lock);
 669	__del_from_avail_list(p);
 670	spin_unlock(&swap_avail_lock);
 671}
 672
 673static void swap_range_alloc(struct swap_info_struct *si, unsigned long offset,
 674			     unsigned int nr_entries)
 675{
 676	unsigned int end = offset + nr_entries - 1;
 677
 678	if (offset == si->lowest_bit)
 679		si->lowest_bit += nr_entries;
 680	if (end == si->highest_bit)
 681		si->highest_bit -= nr_entries;
 682	si->inuse_pages += nr_entries;
 683	if (si->inuse_pages == si->pages) {
 684		si->lowest_bit = si->max;
 685		si->highest_bit = 0;
 686		del_from_avail_list(si);
 687	}
 688}
 689
 690static void add_to_avail_list(struct swap_info_struct *p)
 691{
 692	int nid;
 693
 694	spin_lock(&swap_avail_lock);
 695	for_each_node(nid) {
 696		WARN_ON(!plist_node_empty(&p->avail_lists[nid]));
 697		plist_add(&p->avail_lists[nid], &swap_avail_heads[nid]);
 698	}
 699	spin_unlock(&swap_avail_lock);
 700}
 701
 702static void swap_range_free(struct swap_info_struct *si, unsigned long offset,
 703			    unsigned int nr_entries)
 704{
 
 705	unsigned long end = offset + nr_entries - 1;
 706	void (*swap_slot_free_notify)(struct block_device *, unsigned long);
 707
 708	if (offset < si->lowest_bit)
 709		si->lowest_bit = offset;
 710	if (end > si->highest_bit) {
 711		bool was_full = !si->highest_bit;
 712
 713		si->highest_bit = end;
 714		if (was_full && (si->flags & SWP_WRITEOK))
 715			add_to_avail_list(si);
 716	}
 717	atomic_long_add(nr_entries, &nr_swap_pages);
 718	si->inuse_pages -= nr_entries;
 719	if (si->flags & SWP_BLKDEV)
 720		swap_slot_free_notify =
 721			si->bdev->bd_disk->fops->swap_slot_free_notify;
 722	else
 723		swap_slot_free_notify = NULL;
 724	while (offset <= end) {
 725		frontswap_invalidate_page(si->type, offset);
 726		if (swap_slot_free_notify)
 727			swap_slot_free_notify(si->bdev, offset);
 728		offset++;
 729	}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 730}
 731
 732static int scan_swap_map_slots(struct swap_info_struct *si,
 733			       unsigned char usage, int nr,
 734			       swp_entry_t slots[])
 735{
 736	struct swap_cluster_info *ci;
 737	unsigned long offset;
 738	unsigned long scan_base;
 739	unsigned long last_in_cluster = 0;
 740	int latency_ration = LATENCY_LIMIT;
 741	int n_ret = 0;
 742
 743	if (nr > SWAP_BATCH)
 744		nr = SWAP_BATCH;
 745
 746	/*
 747	 * We try to cluster swap pages by allocating them sequentially
 748	 * in swap.  Once we've allocated SWAPFILE_CLUSTER pages this
 749	 * way, however, we resort to first-free allocation, starting
 750	 * a new cluster.  This prevents us from scattering swap pages
 751	 * all over the entire swap partition, so that we reduce
 752	 * overall disk seek times between swap pages.  -- sct
 753	 * But we do now try to find an empty cluster.  -Andrea
 754	 * And we let swap pages go all over an SSD partition.  Hugh
 755	 */
 756
 757	si->flags += SWP_SCANNING;
 758	scan_base = offset = si->cluster_next;
 
 
 
 
 
 
 
 
 
 759
 760	/* SSD algorithm */
 761	if (si->cluster_info) {
 762		if (scan_swap_map_try_ssd_cluster(si, &offset, &scan_base))
 763			goto checks;
 764		else
 765			goto scan;
 766	}
 767
 768	if (unlikely(!si->cluster_nr--)) {
 769		if (si->pages - si->inuse_pages < SWAPFILE_CLUSTER) {
 770			si->cluster_nr = SWAPFILE_CLUSTER - 1;
 771			goto checks;
 772		}
 773
 774		spin_unlock(&si->lock);
 775
 776		/*
 777		 * If seek is expensive, start searching for new cluster from
 778		 * start of partition, to minimize the span of allocated swap.
 779		 * If seek is cheap, that is the SWP_SOLIDSTATE si->cluster_info
 780		 * case, just handled by scan_swap_map_try_ssd_cluster() above.
 781		 */
 782		scan_base = offset = si->lowest_bit;
 783		last_in_cluster = offset + SWAPFILE_CLUSTER - 1;
 784
 785		/* Locate the first empty (unaligned) cluster */
 786		for (; last_in_cluster <= si->highest_bit; offset++) {
 787			if (si->swap_map[offset])
 788				last_in_cluster = offset + SWAPFILE_CLUSTER;
 789			else if (offset == last_in_cluster) {
 790				spin_lock(&si->lock);
 791				offset -= SWAPFILE_CLUSTER - 1;
 792				si->cluster_next = offset;
 793				si->cluster_nr = SWAPFILE_CLUSTER - 1;
 794				goto checks;
 795			}
 796			if (unlikely(--latency_ration < 0)) {
 797				cond_resched();
 798				latency_ration = LATENCY_LIMIT;
 799			}
 800		}
 801
 802		offset = scan_base;
 803		spin_lock(&si->lock);
 804		si->cluster_nr = SWAPFILE_CLUSTER - 1;
 805	}
 806
 807checks:
 808	if (si->cluster_info) {
 809		while (scan_swap_map_ssd_cluster_conflict(si, offset)) {
 810		/* take a break if we already got some slots */
 811			if (n_ret)
 812				goto done;
 813			if (!scan_swap_map_try_ssd_cluster(si, &offset,
 814							&scan_base))
 815				goto scan;
 816		}
 817	}
 818	if (!(si->flags & SWP_WRITEOK))
 819		goto no_page;
 820	if (!si->highest_bit)
 821		goto no_page;
 822	if (offset > si->highest_bit)
 823		scan_base = offset = si->lowest_bit;
 824
 825	ci = lock_cluster(si, offset);
 826	/* reuse swap entry of cache-only swap if not busy. */
 827	if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) {
 828		int swap_was_freed;
 829		unlock_cluster(ci);
 830		spin_unlock(&si->lock);
 831		swap_was_freed = __try_to_reclaim_swap(si, offset, TTRS_ANYWAY);
 832		spin_lock(&si->lock);
 833		/* entry was freed successfully, try to use this again */
 834		if (swap_was_freed)
 835			goto checks;
 836		goto scan; /* check next one */
 837	}
 838
 839	if (si->swap_map[offset]) {
 840		unlock_cluster(ci);
 841		if (!n_ret)
 842			goto scan;
 843		else
 844			goto done;
 845	}
 846	si->swap_map[offset] = usage;
 847	inc_cluster_info_page(si, si->cluster_info, offset);
 848	unlock_cluster(ci);
 849
 850	swap_range_alloc(si, offset, 1);
 851	si->cluster_next = offset + 1;
 852	slots[n_ret++] = swp_entry(si->type, offset);
 853
 854	/* got enough slots or reach max slots? */
 855	if ((n_ret == nr) || (offset >= si->highest_bit))
 856		goto done;
 857
 858	/* search for next available slot */
 859
 860	/* time to take a break? */
 861	if (unlikely(--latency_ration < 0)) {
 862		if (n_ret)
 863			goto done;
 864		spin_unlock(&si->lock);
 865		cond_resched();
 866		spin_lock(&si->lock);
 867		latency_ration = LATENCY_LIMIT;
 868	}
 869
 870	/* try to get more slots in cluster */
 871	if (si->cluster_info) {
 872		if (scan_swap_map_try_ssd_cluster(si, &offset, &scan_base))
 873			goto checks;
 874		else
 875			goto done;
 876	}
 877	/* non-ssd case */
 878	++offset;
 879
 880	/* non-ssd case, still more slots in cluster? */
 881	if (si->cluster_nr && !si->swap_map[offset]) {
 882		--si->cluster_nr;
 883		goto checks;
 884	}
 885
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 886done:
 
 887	si->flags -= SWP_SCANNING;
 888	return n_ret;
 889
 890scan:
 891	spin_unlock(&si->lock);
 892	while (++offset <= si->highest_bit) {
 893		if (!si->swap_map[offset]) {
 894			spin_lock(&si->lock);
 895			goto checks;
 896		}
 897		if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) {
 
 898			spin_lock(&si->lock);
 899			goto checks;
 900		}
 901		if (unlikely(--latency_ration < 0)) {
 902			cond_resched();
 903			latency_ration = LATENCY_LIMIT;
 
 904		}
 905	}
 906	offset = si->lowest_bit;
 907	while (offset < scan_base) {
 908		if (!si->swap_map[offset]) {
 909			spin_lock(&si->lock);
 910			goto checks;
 911		}
 912		if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) {
 
 913			spin_lock(&si->lock);
 914			goto checks;
 915		}
 916		if (unlikely(--latency_ration < 0)) {
 917			cond_resched();
 918			latency_ration = LATENCY_LIMIT;
 
 919		}
 920		offset++;
 921	}
 922	spin_lock(&si->lock);
 923
 924no_page:
 925	si->flags -= SWP_SCANNING;
 926	return n_ret;
 927}
 928
 929static int swap_alloc_cluster(struct swap_info_struct *si, swp_entry_t *slot)
 930{
 931	unsigned long idx;
 932	struct swap_cluster_info *ci;
 933	unsigned long offset, i;
 934	unsigned char *map;
 935
 936	/*
 937	 * Should not even be attempting cluster allocations when huge
 938	 * page swap is disabled.  Warn and fail the allocation.
 939	 */
 940	if (!IS_ENABLED(CONFIG_THP_SWAP)) {
 941		VM_WARN_ON_ONCE(1);
 942		return 0;
 943	}
 944
 945	if (cluster_list_empty(&si->free_clusters))
 946		return 0;
 947
 948	idx = cluster_list_first(&si->free_clusters);
 949	offset = idx * SWAPFILE_CLUSTER;
 950	ci = lock_cluster(si, offset);
 951	alloc_cluster(si, idx);
 952	cluster_set_count_flag(ci, SWAPFILE_CLUSTER, CLUSTER_FLAG_HUGE);
 953
 954	map = si->swap_map + offset;
 955	for (i = 0; i < SWAPFILE_CLUSTER; i++)
 956		map[i] = SWAP_HAS_CACHE;
 957	unlock_cluster(ci);
 958	swap_range_alloc(si, offset, SWAPFILE_CLUSTER);
 959	*slot = swp_entry(si->type, offset);
 960
 961	return 1;
 962}
 963
 964static void swap_free_cluster(struct swap_info_struct *si, unsigned long idx)
 965{
 966	unsigned long offset = idx * SWAPFILE_CLUSTER;
 967	struct swap_cluster_info *ci;
 968
 969	ci = lock_cluster(si, offset);
 970	memset(si->swap_map + offset, 0, SWAPFILE_CLUSTER);
 971	cluster_set_count_flag(ci, 0, 0);
 972	free_cluster(si, idx);
 973	unlock_cluster(ci);
 974	swap_range_free(si, offset, SWAPFILE_CLUSTER);
 975}
 976
 977static unsigned long scan_swap_map(struct swap_info_struct *si,
 978				   unsigned char usage)
 979{
 980	swp_entry_t entry;
 981	int n_ret;
 982
 983	n_ret = scan_swap_map_slots(si, usage, 1, &entry);
 984
 985	if (n_ret)
 986		return swp_offset(entry);
 987	else
 988		return 0;
 989
 990}
 991
 992int get_swap_pages(int n_goal, swp_entry_t swp_entries[], int entry_size)
 993{
 994	unsigned long size = swap_entry_size(entry_size);
 995	struct swap_info_struct *si, *next;
 996	long avail_pgs;
 997	int n_ret = 0;
 998	int node;
 999
1000	/* Only single cluster request supported */
1001	WARN_ON_ONCE(n_goal > 1 && size == SWAPFILE_CLUSTER);
1002
1003	avail_pgs = atomic_long_read(&nr_swap_pages) / size;
1004	if (avail_pgs <= 0)
1005		goto noswap;
1006
1007	if (n_goal > SWAP_BATCH)
1008		n_goal = SWAP_BATCH;
1009
1010	if (n_goal > avail_pgs)
1011		n_goal = avail_pgs;
1012
1013	atomic_long_sub(n_goal * size, &nr_swap_pages);
1014
1015	spin_lock(&swap_avail_lock);
1016
1017start_over:
1018	node = numa_node_id();
1019	plist_for_each_entry_safe(si, next, &swap_avail_heads[node], avail_lists[node]) {
1020		/* requeue si to after same-priority siblings */
1021		plist_requeue(&si->avail_lists[node], &swap_avail_heads[node]);
1022		spin_unlock(&swap_avail_lock);
1023		spin_lock(&si->lock);
1024		if (!si->highest_bit || !(si->flags & SWP_WRITEOK)) {
1025			spin_lock(&swap_avail_lock);
1026			if (plist_node_empty(&si->avail_lists[node])) {
1027				spin_unlock(&si->lock);
1028				goto nextsi;
1029			}
1030			WARN(!si->highest_bit,
1031			     "swap_info %d in list but !highest_bit\n",
1032			     si->type);
1033			WARN(!(si->flags & SWP_WRITEOK),
1034			     "swap_info %d in list but !SWP_WRITEOK\n",
1035			     si->type);
1036			__del_from_avail_list(si);
1037			spin_unlock(&si->lock);
1038			goto nextsi;
1039		}
1040		if (size == SWAPFILE_CLUSTER) {
1041			if (!(si->flags & SWP_FS))
1042				n_ret = swap_alloc_cluster(si, swp_entries);
1043		} else
1044			n_ret = scan_swap_map_slots(si, SWAP_HAS_CACHE,
1045						    n_goal, swp_entries);
1046		spin_unlock(&si->lock);
1047		if (n_ret || size == SWAPFILE_CLUSTER)
1048			goto check_out;
1049		pr_debug("scan_swap_map of si %d failed to find offset\n",
1050			si->type);
1051
1052		spin_lock(&swap_avail_lock);
1053nextsi:
1054		/*
1055		 * if we got here, it's likely that si was almost full before,
1056		 * and since scan_swap_map() can drop the si->lock, multiple
1057		 * callers probably all tried to get a page from the same si
1058		 * and it filled up before we could get one; or, the si filled
1059		 * up between us dropping swap_avail_lock and taking si->lock.
1060		 * Since we dropped the swap_avail_lock, the swap_avail_head
1061		 * list may have been modified; so if next is still in the
1062		 * swap_avail_head list then try it, otherwise start over
1063		 * if we have not gotten any slots.
1064		 */
1065		if (plist_node_empty(&next->avail_lists[node]))
1066			goto start_over;
1067	}
1068
1069	spin_unlock(&swap_avail_lock);
1070
1071check_out:
1072	if (n_ret < n_goal)
1073		atomic_long_add((long)(n_goal - n_ret) * size,
1074				&nr_swap_pages);
1075noswap:
1076	return n_ret;
1077}
1078
1079/* The only caller of this function is now suspend routine */
1080swp_entry_t get_swap_page_of_type(int type)
1081{
1082	struct swap_info_struct *si = swap_type_to_swap_info(type);
1083	pgoff_t offset;
1084
1085	if (!si)
1086		goto fail;
1087
1088	spin_lock(&si->lock);
1089	if (si->flags & SWP_WRITEOK) {
1090		atomic_long_dec(&nr_swap_pages);
1091		/* This is called for allocating swap entry, not cache */
1092		offset = scan_swap_map(si, 1);
1093		if (offset) {
1094			spin_unlock(&si->lock);
1095			return swp_entry(type, offset);
1096		}
1097		atomic_long_inc(&nr_swap_pages);
1098	}
1099	spin_unlock(&si->lock);
1100fail:
1101	return (swp_entry_t) {0};
1102}
1103
1104static struct swap_info_struct *__swap_info_get(swp_entry_t entry)
1105{
1106	struct swap_info_struct *p;
1107	unsigned long offset;
1108
1109	if (!entry.val)
1110		goto out;
1111	p = swp_swap_info(entry);
1112	if (!p)
1113		goto bad_nofile;
1114	if (!(p->flags & SWP_USED))
1115		goto bad_device;
1116	offset = swp_offset(entry);
1117	if (offset >= p->max)
1118		goto bad_offset;
1119	return p;
1120
1121bad_offset:
1122	pr_err("swap_info_get: %s%08lx\n", Bad_offset, entry.val);
1123	goto out;
1124bad_device:
1125	pr_err("swap_info_get: %s%08lx\n", Unused_file, entry.val);
1126	goto out;
1127bad_nofile:
1128	pr_err("swap_info_get: %s%08lx\n", Bad_file, entry.val);
1129out:
1130	return NULL;
1131}
1132
1133static struct swap_info_struct *_swap_info_get(swp_entry_t entry)
1134{
1135	struct swap_info_struct *p;
1136
1137	p = __swap_info_get(entry);
1138	if (!p)
1139		goto out;
1140	if (!p->swap_map[swp_offset(entry)])
1141		goto bad_free;
1142	return p;
1143
1144bad_free:
1145	pr_err("swap_info_get: %s%08lx\n", Unused_offset, entry.val);
1146	goto out;
1147out:
1148	return NULL;
1149}
1150
1151static struct swap_info_struct *swap_info_get(swp_entry_t entry)
1152{
1153	struct swap_info_struct *p;
1154
1155	p = _swap_info_get(entry);
1156	if (p)
1157		spin_lock(&p->lock);
1158	return p;
1159}
1160
1161static struct swap_info_struct *swap_info_get_cont(swp_entry_t entry,
1162					struct swap_info_struct *q)
1163{
1164	struct swap_info_struct *p;
1165
1166	p = _swap_info_get(entry);
1167
1168	if (p != q) {
1169		if (q != NULL)
1170			spin_unlock(&q->lock);
1171		if (p != NULL)
1172			spin_lock(&p->lock);
1173	}
1174	return p;
1175}
1176
1177static unsigned char __swap_entry_free_locked(struct swap_info_struct *p,
1178					      unsigned long offset,
1179					      unsigned char usage)
1180{
1181	unsigned char count;
1182	unsigned char has_cache;
1183
1184	count = p->swap_map[offset];
1185
1186	has_cache = count & SWAP_HAS_CACHE;
1187	count &= ~SWAP_HAS_CACHE;
1188
1189	if (usage == SWAP_HAS_CACHE) {
1190		VM_BUG_ON(!has_cache);
1191		has_cache = 0;
1192	} else if (count == SWAP_MAP_SHMEM) {
1193		/*
1194		 * Or we could insist on shmem.c using a special
1195		 * swap_shmem_free() and free_shmem_swap_and_cache()...
1196		 */
1197		count = 0;
1198	} else if ((count & ~COUNT_CONTINUED) <= SWAP_MAP_MAX) {
1199		if (count == COUNT_CONTINUED) {
1200			if (swap_count_continued(p, offset, count))
1201				count = SWAP_MAP_MAX | COUNT_CONTINUED;
1202			else
1203				count = SWAP_MAP_MAX;
1204		} else
1205			count--;
1206	}
1207
1208	usage = count | has_cache;
1209	p->swap_map[offset] = usage ? : SWAP_HAS_CACHE;
 
 
 
1210
1211	return usage;
1212}
1213
1214/*
1215 * Check whether swap entry is valid in the swap device.  If so,
1216 * return pointer to swap_info_struct, and keep the swap entry valid
1217 * via preventing the swap device from being swapoff, until
1218 * put_swap_device() is called.  Otherwise return NULL.
1219 *
1220 * The entirety of the RCU read critical section must come before the
1221 * return from or after the call to synchronize_rcu() in
1222 * enable_swap_info() or swapoff().  So if "si->flags & SWP_VALID" is
1223 * true, the si->map, si->cluster_info, etc. must be valid in the
1224 * critical section.
1225 *
1226 * Notice that swapoff or swapoff+swapon can still happen before the
1227 * rcu_read_lock() in get_swap_device() or after the rcu_read_unlock()
1228 * in put_swap_device() if there isn't any other way to prevent
1229 * swapoff, such as page lock, page table lock, etc.  The caller must
1230 * be prepared for that.  For example, the following situation is
1231 * possible.
1232 *
1233 *   CPU1				CPU2
1234 *   do_swap_page()
1235 *     ...				swapoff+swapon
1236 *     __read_swap_cache_async()
1237 *       swapcache_prepare()
1238 *         __swap_duplicate()
1239 *           // check swap_map
1240 *     // verify PTE not changed
1241 *
1242 * In __swap_duplicate(), the swap_map need to be checked before
1243 * changing partly because the specified swap entry may be for another
1244 * swap device which has been swapoff.  And in do_swap_page(), after
1245 * the page is read from the swap device, the PTE is verified not
1246 * changed with the page table locked to check whether the swap device
1247 * has been swapoff or swapoff+swapon.
1248 */
1249struct swap_info_struct *get_swap_device(swp_entry_t entry)
1250{
1251	struct swap_info_struct *si;
1252	unsigned long offset;
1253
1254	if (!entry.val)
1255		goto out;
1256	si = swp_swap_info(entry);
1257	if (!si)
1258		goto bad_nofile;
1259
1260	rcu_read_lock();
1261	if (!(si->flags & SWP_VALID))
1262		goto unlock_out;
1263	offset = swp_offset(entry);
1264	if (offset >= si->max)
1265		goto unlock_out;
1266
1267	return si;
1268bad_nofile:
1269	pr_err("%s: %s%08lx\n", __func__, Bad_file, entry.val);
1270out:
1271	return NULL;
1272unlock_out:
1273	rcu_read_unlock();
1274	return NULL;
1275}
1276
1277static unsigned char __swap_entry_free(struct swap_info_struct *p,
1278				       swp_entry_t entry, unsigned char usage)
1279{
1280	struct swap_cluster_info *ci;
1281	unsigned long offset = swp_offset(entry);
 
1282
1283	ci = lock_cluster_or_swap_info(p, offset);
1284	usage = __swap_entry_free_locked(p, offset, usage);
1285	unlock_cluster_or_swap_info(p, ci);
1286	if (!usage)
1287		free_swap_slot(entry);
1288
1289	return usage;
1290}
1291
1292static void swap_entry_free(struct swap_info_struct *p, swp_entry_t entry)
1293{
1294	struct swap_cluster_info *ci;
1295	unsigned long offset = swp_offset(entry);
1296	unsigned char count;
1297
1298	ci = lock_cluster(p, offset);
1299	count = p->swap_map[offset];
1300	VM_BUG_ON(count != SWAP_HAS_CACHE);
1301	p->swap_map[offset] = 0;
1302	dec_cluster_info_page(p, p->cluster_info, offset);
1303	unlock_cluster(ci);
1304
1305	mem_cgroup_uncharge_swap(entry, 1);
1306	swap_range_free(p, offset, 1);
1307}
1308
1309/*
1310 * Caller has made sure that the swap device corresponding to entry
1311 * is still around or has not been recycled.
1312 */
1313void swap_free(swp_entry_t entry)
1314{
1315	struct swap_info_struct *p;
1316
1317	p = _swap_info_get(entry);
1318	if (p)
1319		__swap_entry_free(p, entry, 1);
1320}
1321
1322/*
1323 * Called after dropping swapcache to decrease refcnt to swap entries.
1324 */
1325void put_swap_page(struct page *page, swp_entry_t entry)
1326{
1327	unsigned long offset = swp_offset(entry);
1328	unsigned long idx = offset / SWAPFILE_CLUSTER;
1329	struct swap_cluster_info *ci;
1330	struct swap_info_struct *si;
1331	unsigned char *map;
1332	unsigned int i, free_entries = 0;
1333	unsigned char val;
1334	int size = swap_entry_size(hpage_nr_pages(page));
1335
1336	si = _swap_info_get(entry);
1337	if (!si)
1338		return;
1339
1340	ci = lock_cluster_or_swap_info(si, offset);
1341	if (size == SWAPFILE_CLUSTER) {
1342		VM_BUG_ON(!cluster_is_huge(ci));
1343		map = si->swap_map + offset;
1344		for (i = 0; i < SWAPFILE_CLUSTER; i++) {
1345			val = map[i];
1346			VM_BUG_ON(!(val & SWAP_HAS_CACHE));
1347			if (val == SWAP_HAS_CACHE)
1348				free_entries++;
1349		}
1350		cluster_clear_huge(ci);
1351		if (free_entries == SWAPFILE_CLUSTER) {
1352			unlock_cluster_or_swap_info(si, ci);
1353			spin_lock(&si->lock);
1354			mem_cgroup_uncharge_swap(entry, SWAPFILE_CLUSTER);
1355			swap_free_cluster(si, idx);
1356			spin_unlock(&si->lock);
1357			return;
1358		}
1359	}
1360	for (i = 0; i < size; i++, entry.val++) {
1361		if (!__swap_entry_free_locked(si, offset + i, SWAP_HAS_CACHE)) {
1362			unlock_cluster_or_swap_info(si, ci);
1363			free_swap_slot(entry);
1364			if (i == size - 1)
1365				return;
1366			lock_cluster_or_swap_info(si, offset);
1367		}
1368	}
1369	unlock_cluster_or_swap_info(si, ci);
1370}
1371
1372#ifdef CONFIG_THP_SWAP
1373int split_swap_cluster(swp_entry_t entry)
1374{
1375	struct swap_info_struct *si;
1376	struct swap_cluster_info *ci;
1377	unsigned long offset = swp_offset(entry);
1378
1379	si = _swap_info_get(entry);
1380	if (!si)
1381		return -EBUSY;
1382	ci = lock_cluster(si, offset);
1383	cluster_clear_huge(ci);
1384	unlock_cluster(ci);
1385	return 0;
1386}
1387#endif
1388
1389static int swp_entry_cmp(const void *ent1, const void *ent2)
1390{
1391	const swp_entry_t *e1 = ent1, *e2 = ent2;
1392
1393	return (int)swp_type(*e1) - (int)swp_type(*e2);
1394}
1395
1396void swapcache_free_entries(swp_entry_t *entries, int n)
1397{
1398	struct swap_info_struct *p, *prev;
1399	int i;
1400
1401	if (n <= 0)
1402		return;
1403
1404	prev = NULL;
1405	p = NULL;
1406
1407	/*
1408	 * Sort swap entries by swap device, so each lock is only taken once.
1409	 * nr_swapfiles isn't absolutely correct, but the overhead of sort() is
1410	 * so low that it isn't necessary to optimize further.
1411	 */
1412	if (nr_swapfiles > 1)
1413		sort(entries, n, sizeof(entries[0]), swp_entry_cmp, NULL);
1414	for (i = 0; i < n; ++i) {
1415		p = swap_info_get_cont(entries[i], prev);
1416		if (p)
1417			swap_entry_free(p, entries[i]);
1418		prev = p;
1419	}
1420	if (p)
1421		spin_unlock(&p->lock);
1422}
1423
1424/*
1425 * How many references to page are currently swapped out?
1426 * This does not give an exact answer when swap count is continued,
1427 * but does include the high COUNT_CONTINUED flag to allow for that.
1428 */
1429int page_swapcount(struct page *page)
1430{
1431	int count = 0;
1432	struct swap_info_struct *p;
1433	struct swap_cluster_info *ci;
1434	swp_entry_t entry;
1435	unsigned long offset;
1436
1437	entry.val = page_private(page);
1438	p = _swap_info_get(entry);
1439	if (p) {
1440		offset = swp_offset(entry);
1441		ci = lock_cluster_or_swap_info(p, offset);
1442		count = swap_count(p->swap_map[offset]);
1443		unlock_cluster_or_swap_info(p, ci);
1444	}
1445	return count;
1446}
1447
1448int __swap_count(swp_entry_t entry)
1449{
1450	struct swap_info_struct *si;
1451	pgoff_t offset = swp_offset(entry);
1452	int count = 0;
1453
1454	si = get_swap_device(entry);
1455	if (si) {
1456		count = swap_count(si->swap_map[offset]);
1457		put_swap_device(si);
1458	}
1459	return count;
1460}
1461
1462static int swap_swapcount(struct swap_info_struct *si, swp_entry_t entry)
1463{
1464	int count = 0;
1465	pgoff_t offset = swp_offset(entry);
1466	struct swap_cluster_info *ci;
1467
1468	ci = lock_cluster_or_swap_info(si, offset);
1469	count = swap_count(si->swap_map[offset]);
1470	unlock_cluster_or_swap_info(si, ci);
1471	return count;
1472}
1473
1474/*
1475 * How many references to @entry are currently swapped out?
1476 * This does not give an exact answer when swap count is continued,
1477 * but does include the high COUNT_CONTINUED flag to allow for that.
1478 */
1479int __swp_swapcount(swp_entry_t entry)
1480{
1481	int count = 0;
1482	struct swap_info_struct *si;
1483
1484	si = get_swap_device(entry);
1485	if (si) {
1486		count = swap_swapcount(si, entry);
1487		put_swap_device(si);
1488	}
1489	return count;
1490}
1491
1492/*
1493 * How many references to @entry are currently swapped out?
1494 * This considers COUNT_CONTINUED so it returns exact answer.
1495 */
1496int swp_swapcount(swp_entry_t entry)
1497{
1498	int count, tmp_count, n;
1499	struct swap_info_struct *p;
1500	struct swap_cluster_info *ci;
1501	struct page *page;
1502	pgoff_t offset;
1503	unsigned char *map;
1504
1505	p = _swap_info_get(entry);
1506	if (!p)
1507		return 0;
1508
1509	offset = swp_offset(entry);
1510
1511	ci = lock_cluster_or_swap_info(p, offset);
1512
1513	count = swap_count(p->swap_map[offset]);
1514	if (!(count & COUNT_CONTINUED))
1515		goto out;
1516
1517	count &= ~COUNT_CONTINUED;
1518	n = SWAP_MAP_MAX + 1;
1519
1520	page = vmalloc_to_page(p->swap_map + offset);
1521	offset &= ~PAGE_MASK;
1522	VM_BUG_ON(page_private(page) != SWP_CONTINUED);
1523
1524	do {
1525		page = list_next_entry(page, lru);
1526		map = kmap_atomic(page);
1527		tmp_count = map[offset];
1528		kunmap_atomic(map);
1529
1530		count += (tmp_count & ~COUNT_CONTINUED) * n;
1531		n *= (SWAP_CONT_MAX + 1);
1532	} while (tmp_count & COUNT_CONTINUED);
1533out:
1534	unlock_cluster_or_swap_info(p, ci);
1535	return count;
1536}
1537
1538static bool swap_page_trans_huge_swapped(struct swap_info_struct *si,
1539					 swp_entry_t entry)
1540{
1541	struct swap_cluster_info *ci;
1542	unsigned char *map = si->swap_map;
1543	unsigned long roffset = swp_offset(entry);
1544	unsigned long offset = round_down(roffset, SWAPFILE_CLUSTER);
1545	int i;
1546	bool ret = false;
1547
1548	ci = lock_cluster_or_swap_info(si, offset);
1549	if (!ci || !cluster_is_huge(ci)) {
1550		if (swap_count(map[roffset]))
1551			ret = true;
1552		goto unlock_out;
1553	}
1554	for (i = 0; i < SWAPFILE_CLUSTER; i++) {
1555		if (swap_count(map[offset + i])) {
1556			ret = true;
1557			break;
1558		}
1559	}
1560unlock_out:
1561	unlock_cluster_or_swap_info(si, ci);
1562	return ret;
1563}
1564
1565static bool page_swapped(struct page *page)
1566{
1567	swp_entry_t entry;
1568	struct swap_info_struct *si;
1569
1570	if (!IS_ENABLED(CONFIG_THP_SWAP) || likely(!PageTransCompound(page)))
1571		return page_swapcount(page) != 0;
1572
1573	page = compound_head(page);
1574	entry.val = page_private(page);
1575	si = _swap_info_get(entry);
1576	if (si)
1577		return swap_page_trans_huge_swapped(si, entry);
1578	return false;
1579}
1580
1581static int page_trans_huge_map_swapcount(struct page *page, int *total_mapcount,
1582					 int *total_swapcount)
1583{
1584	int i, map_swapcount, _total_mapcount, _total_swapcount;
1585	unsigned long offset = 0;
1586	struct swap_info_struct *si;
1587	struct swap_cluster_info *ci = NULL;
1588	unsigned char *map = NULL;
1589	int mapcount, swapcount = 0;
1590
1591	/* hugetlbfs shouldn't call it */
1592	VM_BUG_ON_PAGE(PageHuge(page), page);
1593
1594	if (!IS_ENABLED(CONFIG_THP_SWAP) || likely(!PageTransCompound(page))) {
1595		mapcount = page_trans_huge_mapcount(page, total_mapcount);
1596		if (PageSwapCache(page))
1597			swapcount = page_swapcount(page);
1598		if (total_swapcount)
1599			*total_swapcount = swapcount;
1600		return mapcount + swapcount;
1601	}
1602
1603	page = compound_head(page);
1604
1605	_total_mapcount = _total_swapcount = map_swapcount = 0;
1606	if (PageSwapCache(page)) {
1607		swp_entry_t entry;
1608
1609		entry.val = page_private(page);
1610		si = _swap_info_get(entry);
1611		if (si) {
1612			map = si->swap_map;
1613			offset = swp_offset(entry);
1614		}
1615	}
1616	if (map)
1617		ci = lock_cluster(si, offset);
1618	for (i = 0; i < HPAGE_PMD_NR; i++) {
1619		mapcount = atomic_read(&page[i]._mapcount) + 1;
1620		_total_mapcount += mapcount;
1621		if (map) {
1622			swapcount = swap_count(map[offset + i]);
1623			_total_swapcount += swapcount;
1624		}
1625		map_swapcount = max(map_swapcount, mapcount + swapcount);
1626	}
1627	unlock_cluster(ci);
1628	if (PageDoubleMap(page)) {
1629		map_swapcount -= 1;
1630		_total_mapcount -= HPAGE_PMD_NR;
1631	}
1632	mapcount = compound_mapcount(page);
1633	map_swapcount += mapcount;
1634	_total_mapcount += mapcount;
1635	if (total_mapcount)
1636		*total_mapcount = _total_mapcount;
1637	if (total_swapcount)
1638		*total_swapcount = _total_swapcount;
1639
1640	return map_swapcount;
1641}
1642
1643/*
1644 * We can write to an anon page without COW if there are no other references
1645 * to it.  And as a side-effect, free up its swap: because the old content
1646 * on disk will never be read, and seeking back there to write new content
1647 * later would only waste time away from clustering.
1648 *
1649 * NOTE: total_map_swapcount should not be relied upon by the caller if
1650 * reuse_swap_page() returns false, but it may be always overwritten
1651 * (see the other implementation for CONFIG_SWAP=n).
1652 */
1653bool reuse_swap_page(struct page *page, int *total_map_swapcount)
1654{
1655	int count, total_mapcount, total_swapcount;
1656
1657	VM_BUG_ON_PAGE(!PageLocked(page), page);
1658	if (unlikely(PageKsm(page)))
1659		return false;
1660	count = page_trans_huge_map_swapcount(page, &total_mapcount,
1661					      &total_swapcount);
1662	if (total_map_swapcount)
1663		*total_map_swapcount = total_mapcount + total_swapcount;
1664	if (count == 1 && PageSwapCache(page) &&
1665	    (likely(!PageTransCompound(page)) ||
1666	     /* The remaining swap count will be freed soon */
1667	     total_swapcount == page_swapcount(page))) {
1668		if (!PageWriteback(page)) {
1669			page = compound_head(page);
1670			delete_from_swap_cache(page);
1671			SetPageDirty(page);
1672		} else {
1673			swp_entry_t entry;
1674			struct swap_info_struct *p;
1675
1676			entry.val = page_private(page);
1677			p = swap_info_get(entry);
1678			if (p->flags & SWP_STABLE_WRITES) {
1679				spin_unlock(&p->lock);
1680				return false;
1681			}
1682			spin_unlock(&p->lock);
1683		}
1684	}
1685
1686	return count <= 1;
1687}
1688
1689/*
1690 * If swap is getting full, or if there are no more mappings of this page,
1691 * then try_to_free_swap is called to free its swap space.
1692 */
1693int try_to_free_swap(struct page *page)
1694{
1695	VM_BUG_ON_PAGE(!PageLocked(page), page);
1696
1697	if (!PageSwapCache(page))
1698		return 0;
1699	if (PageWriteback(page))
1700		return 0;
1701	if (page_swapped(page))
1702		return 0;
1703
1704	/*
1705	 * Once hibernation has begun to create its image of memory,
1706	 * there's a danger that one of the calls to try_to_free_swap()
1707	 * - most probably a call from __try_to_reclaim_swap() while
1708	 * hibernation is allocating its own swap pages for the image,
1709	 * but conceivably even a call from memory reclaim - will free
1710	 * the swap from a page which has already been recorded in the
1711	 * image as a clean swapcache page, and then reuse its swap for
1712	 * another page of the image.  On waking from hibernation, the
1713	 * original page might be freed under memory pressure, then
1714	 * later read back in from swap, now with the wrong data.
1715	 *
1716	 * Hibernation suspends storage while it is writing the image
1717	 * to disk so check that here.
1718	 */
1719	if (pm_suspended_storage())
1720		return 0;
1721
1722	page = compound_head(page);
1723	delete_from_swap_cache(page);
1724	SetPageDirty(page);
1725	return 1;
1726}
1727
1728/*
1729 * Free the swap entry like above, but also try to
1730 * free the page cache entry if it is the last user.
1731 */
1732int free_swap_and_cache(swp_entry_t entry)
1733{
1734	struct swap_info_struct *p;
1735	unsigned char count;
1736
1737	if (non_swap_entry(entry))
1738		return 1;
1739
1740	p = _swap_info_get(entry);
1741	if (p) {
1742		count = __swap_entry_free(p, entry, 1);
1743		if (count == SWAP_HAS_CACHE &&
1744		    !swap_page_trans_huge_swapped(p, entry))
1745			__try_to_reclaim_swap(p, swp_offset(entry),
1746					      TTRS_UNMAPPED | TTRS_FULL);
1747	}
1748	return p != NULL;
1749}
1750
1751#ifdef CONFIG_HIBERNATION
1752/*
1753 * Find the swap type that corresponds to given device (if any).
1754 *
1755 * @offset - number of the PAGE_SIZE-sized block of the device, starting
1756 * from 0, in which the swap header is expected to be located.
1757 *
1758 * This is needed for the suspend to disk (aka swsusp).
1759 */
1760int swap_type_of(dev_t device, sector_t offset, struct block_device **bdev_p)
1761{
1762	struct block_device *bdev = NULL;
1763	int type;
1764
1765	if (device)
1766		bdev = bdget(device);
1767
1768	spin_lock(&swap_lock);
1769	for (type = 0; type < nr_swapfiles; type++) {
1770		struct swap_info_struct *sis = swap_info[type];
1771
1772		if (!(sis->flags & SWP_WRITEOK))
1773			continue;
1774
1775		if (!bdev) {
1776			if (bdev_p)
1777				*bdev_p = bdgrab(sis->bdev);
1778
1779			spin_unlock(&swap_lock);
1780			return type;
1781		}
1782		if (bdev == sis->bdev) {
1783			struct swap_extent *se = first_se(sis);
1784
1785			if (se->start_block == offset) {
1786				if (bdev_p)
1787					*bdev_p = bdgrab(sis->bdev);
1788
1789				spin_unlock(&swap_lock);
1790				bdput(bdev);
1791				return type;
1792			}
1793		}
1794	}
1795	spin_unlock(&swap_lock);
1796	if (bdev)
1797		bdput(bdev);
1798
1799	return -ENODEV;
1800}
1801
1802/*
1803 * Get the (PAGE_SIZE) block corresponding to given offset on the swapdev
1804 * corresponding to given index in swap_info (swap type).
1805 */
1806sector_t swapdev_block(int type, pgoff_t offset)
1807{
1808	struct block_device *bdev;
1809	struct swap_info_struct *si = swap_type_to_swap_info(type);
1810
1811	if (!si || !(si->flags & SWP_WRITEOK))
1812		return 0;
1813	return map_swap_entry(swp_entry(type, offset), &bdev);
1814}
1815
1816/*
1817 * Return either the total number of swap pages of given type, or the number
1818 * of free pages of that type (depending on @free)
1819 *
1820 * This is needed for software suspend
1821 */
1822unsigned int count_swap_pages(int type, int free)
1823{
1824	unsigned int n = 0;
1825
1826	spin_lock(&swap_lock);
1827	if ((unsigned int)type < nr_swapfiles) {
1828		struct swap_info_struct *sis = swap_info[type];
1829
1830		spin_lock(&sis->lock);
1831		if (sis->flags & SWP_WRITEOK) {
1832			n = sis->pages;
1833			if (free)
1834				n -= sis->inuse_pages;
1835		}
1836		spin_unlock(&sis->lock);
1837	}
1838	spin_unlock(&swap_lock);
1839	return n;
1840}
1841#endif /* CONFIG_HIBERNATION */
1842
1843static inline int pte_same_as_swp(pte_t pte, pte_t swp_pte)
1844{
1845	return pte_same(pte_swp_clear_soft_dirty(pte), swp_pte);
1846}
1847
1848/*
1849 * No need to decide whether this PTE shares the swap entry with others,
1850 * just let do_wp_page work it out if a write is requested later - to
1851 * force COW, vm_page_prot omits write permission from any private vma.
1852 */
1853static int unuse_pte(struct vm_area_struct *vma, pmd_t *pmd,
1854		unsigned long addr, swp_entry_t entry, struct page *page)
1855{
1856	struct page *swapcache;
1857	struct mem_cgroup *memcg;
1858	spinlock_t *ptl;
1859	pte_t *pte;
1860	int ret = 1;
1861
1862	swapcache = page;
1863	page = ksm_might_need_to_copy(page, vma, addr);
1864	if (unlikely(!page))
1865		return -ENOMEM;
1866
1867	if (mem_cgroup_try_charge(page, vma->vm_mm, GFP_KERNEL,
1868				&memcg, false)) {
1869		ret = -ENOMEM;
1870		goto out_nolock;
1871	}
1872
1873	pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
1874	if (unlikely(!pte_same_as_swp(*pte, swp_entry_to_pte(entry)))) {
1875		mem_cgroup_cancel_charge(page, memcg, false);
1876		ret = 0;
1877		goto out;
1878	}
1879
1880	dec_mm_counter(vma->vm_mm, MM_SWAPENTS);
1881	inc_mm_counter(vma->vm_mm, MM_ANONPAGES);
1882	get_page(page);
1883	set_pte_at(vma->vm_mm, addr, pte,
1884		   pte_mkold(mk_pte(page, vma->vm_page_prot)));
1885	if (page == swapcache) {
1886		page_add_anon_rmap(page, vma, addr, false);
1887		mem_cgroup_commit_charge(page, memcg, true, false);
1888	} else { /* ksm created a completely new copy */
1889		page_add_new_anon_rmap(page, vma, addr, false);
1890		mem_cgroup_commit_charge(page, memcg, false, false);
1891		lru_cache_add_active_or_unevictable(page, vma);
1892	}
1893	swap_free(entry);
1894	/*
1895	 * Move the page to the active list so it is not
1896	 * immediately swapped out again after swapon.
1897	 */
1898	activate_page(page);
1899out:
1900	pte_unmap_unlock(pte, ptl);
1901out_nolock:
1902	if (page != swapcache) {
1903		unlock_page(page);
1904		put_page(page);
1905	}
1906	return ret;
1907}
1908
1909static int unuse_pte_range(struct vm_area_struct *vma, pmd_t *pmd,
1910			unsigned long addr, unsigned long end,
1911			unsigned int type, bool frontswap,
1912			unsigned long *fs_pages_to_unuse)
1913{
1914	struct page *page;
1915	swp_entry_t entry;
1916	pte_t *pte;
1917	struct swap_info_struct *si;
1918	unsigned long offset;
1919	int ret = 0;
1920	volatile unsigned char *swap_map;
1921
1922	si = swap_info[type];
1923	pte = pte_offset_map(pmd, addr);
1924	do {
1925		struct vm_fault vmf;
1926
1927		if (!is_swap_pte(*pte))
1928			continue;
1929
1930		entry = pte_to_swp_entry(*pte);
1931		if (swp_type(entry) != type)
1932			continue;
1933
1934		offset = swp_offset(entry);
1935		if (frontswap && !frontswap_test(si, offset))
1936			continue;
1937
1938		pte_unmap(pte);
1939		swap_map = &si->swap_map[offset];
1940		vmf.vma = vma;
1941		vmf.address = addr;
1942		vmf.pmd = pmd;
1943		page = swapin_readahead(entry, GFP_HIGHUSER_MOVABLE, &vmf);
 
 
 
 
1944		if (!page) {
1945			if (*swap_map == 0 || *swap_map == SWAP_MAP_BAD)
1946				goto try_next;
1947			return -ENOMEM;
1948		}
1949
1950		lock_page(page);
1951		wait_on_page_writeback(page);
1952		ret = unuse_pte(vma, pmd, addr, entry, page);
1953		if (ret < 0) {
1954			unlock_page(page);
1955			put_page(page);
1956			goto out;
1957		}
1958
1959		try_to_free_swap(page);
1960		unlock_page(page);
1961		put_page(page);
1962
1963		if (*fs_pages_to_unuse && !--(*fs_pages_to_unuse)) {
1964			ret = FRONTSWAP_PAGES_UNUSED;
1965			goto out;
1966		}
1967try_next:
1968		pte = pte_offset_map(pmd, addr);
1969	} while (pte++, addr += PAGE_SIZE, addr != end);
1970	pte_unmap(pte - 1);
1971
1972	ret = 0;
1973out:
1974	return ret;
1975}
1976
1977static inline int unuse_pmd_range(struct vm_area_struct *vma, pud_t *pud,
1978				unsigned long addr, unsigned long end,
1979				unsigned int type, bool frontswap,
1980				unsigned long *fs_pages_to_unuse)
1981{
1982	pmd_t *pmd;
1983	unsigned long next;
1984	int ret;
1985
1986	pmd = pmd_offset(pud, addr);
1987	do {
1988		cond_resched();
1989		next = pmd_addr_end(addr, end);
1990		if (pmd_none_or_trans_huge_or_clear_bad(pmd))
1991			continue;
1992		ret = unuse_pte_range(vma, pmd, addr, next, type,
1993				      frontswap, fs_pages_to_unuse);
1994		if (ret)
1995			return ret;
1996	} while (pmd++, addr = next, addr != end);
1997	return 0;
1998}
1999
2000static inline int unuse_pud_range(struct vm_area_struct *vma, p4d_t *p4d,
2001				unsigned long addr, unsigned long end,
2002				unsigned int type, bool frontswap,
2003				unsigned long *fs_pages_to_unuse)
2004{
2005	pud_t *pud;
2006	unsigned long next;
2007	int ret;
2008
2009	pud = pud_offset(p4d, addr);
2010	do {
2011		next = pud_addr_end(addr, end);
2012		if (pud_none_or_clear_bad(pud))
2013			continue;
2014		ret = unuse_pmd_range(vma, pud, addr, next, type,
2015				      frontswap, fs_pages_to_unuse);
2016		if (ret)
2017			return ret;
2018	} while (pud++, addr = next, addr != end);
2019	return 0;
2020}
2021
2022static inline int unuse_p4d_range(struct vm_area_struct *vma, pgd_t *pgd,
2023				unsigned long addr, unsigned long end,
2024				unsigned int type, bool frontswap,
2025				unsigned long *fs_pages_to_unuse)
2026{
2027	p4d_t *p4d;
2028	unsigned long next;
2029	int ret;
2030
2031	p4d = p4d_offset(pgd, addr);
2032	do {
2033		next = p4d_addr_end(addr, end);
2034		if (p4d_none_or_clear_bad(p4d))
2035			continue;
2036		ret = unuse_pud_range(vma, p4d, addr, next, type,
2037				      frontswap, fs_pages_to_unuse);
2038		if (ret)
2039			return ret;
2040	} while (p4d++, addr = next, addr != end);
2041	return 0;
2042}
2043
2044static int unuse_vma(struct vm_area_struct *vma, unsigned int type,
2045		     bool frontswap, unsigned long *fs_pages_to_unuse)
2046{
2047	pgd_t *pgd;
2048	unsigned long addr, end, next;
2049	int ret;
2050
2051	addr = vma->vm_start;
2052	end = vma->vm_end;
2053
2054	pgd = pgd_offset(vma->vm_mm, addr);
2055	do {
2056		next = pgd_addr_end(addr, end);
2057		if (pgd_none_or_clear_bad(pgd))
2058			continue;
2059		ret = unuse_p4d_range(vma, pgd, addr, next, type,
2060				      frontswap, fs_pages_to_unuse);
2061		if (ret)
2062			return ret;
2063	} while (pgd++, addr = next, addr != end);
2064	return 0;
2065}
2066
2067static int unuse_mm(struct mm_struct *mm, unsigned int type,
2068		    bool frontswap, unsigned long *fs_pages_to_unuse)
2069{
2070	struct vm_area_struct *vma;
2071	int ret = 0;
2072
2073	down_read(&mm->mmap_sem);
2074	for (vma = mm->mmap; vma; vma = vma->vm_next) {
2075		if (vma->anon_vma) {
2076			ret = unuse_vma(vma, type, frontswap,
2077					fs_pages_to_unuse);
2078			if (ret)
2079				break;
2080		}
2081		cond_resched();
2082	}
2083	up_read(&mm->mmap_sem);
2084	return ret;
2085}
2086
2087/*
2088 * Scan swap_map (or frontswap_map if frontswap parameter is true)
2089 * from current position to next entry still in use. Return 0
2090 * if there are no inuse entries after prev till end of the map.
2091 */
2092static unsigned int find_next_to_unuse(struct swap_info_struct *si,
2093					unsigned int prev, bool frontswap)
2094{
2095	unsigned int i;
2096	unsigned char count;
2097
2098	/*
2099	 * No need for swap_lock here: we're just looking
2100	 * for whether an entry is in use, not modifying it; false
2101	 * hits are okay, and sys_swapoff() has already prevented new
2102	 * allocations from this area (while holding swap_lock).
2103	 */
2104	for (i = prev + 1; i < si->max; i++) {
2105		count = READ_ONCE(si->swap_map[i]);
2106		if (count && swap_count(count) != SWAP_MAP_BAD)
2107			if (!frontswap || frontswap_test(si, i))
2108				break;
2109		if ((i % LATENCY_LIMIT) == 0)
2110			cond_resched();
2111	}
2112
2113	if (i == si->max)
2114		i = 0;
2115
2116	return i;
2117}
2118
2119/*
2120 * If the boolean frontswap is true, only unuse pages_to_unuse pages;
2121 * pages_to_unuse==0 means all pages; ignored if frontswap is false
2122 */
2123int try_to_unuse(unsigned int type, bool frontswap,
2124		 unsigned long pages_to_unuse)
2125{
2126	struct mm_struct *prev_mm;
2127	struct mm_struct *mm;
2128	struct list_head *p;
2129	int retval = 0;
2130	struct swap_info_struct *si = swap_info[type];
2131	struct page *page;
2132	swp_entry_t entry;
2133	unsigned int i;
2134
2135	if (!si->inuse_pages)
2136		return 0;
2137
2138	if (!frontswap)
2139		pages_to_unuse = 0;
2140
2141retry:
2142	retval = shmem_unuse(type, frontswap, &pages_to_unuse);
2143	if (retval)
2144		goto out;
2145
2146	prev_mm = &init_mm;
2147	mmget(prev_mm);
2148
2149	spin_lock(&mmlist_lock);
2150	p = &init_mm.mmlist;
2151	while (si->inuse_pages &&
2152	       !signal_pending(current) &&
2153	       (p = p->next) != &init_mm.mmlist) {
2154
2155		mm = list_entry(p, struct mm_struct, mmlist);
2156		if (!mmget_not_zero(mm))
2157			continue;
2158		spin_unlock(&mmlist_lock);
2159		mmput(prev_mm);
2160		prev_mm = mm;
2161		retval = unuse_mm(mm, type, frontswap, &pages_to_unuse);
2162
2163		if (retval) {
2164			mmput(prev_mm);
2165			goto out;
2166		}
2167
2168		/*
2169		 * Make sure that we aren't completely killing
2170		 * interactive performance.
2171		 */
2172		cond_resched();
2173		spin_lock(&mmlist_lock);
2174	}
2175	spin_unlock(&mmlist_lock);
2176
2177	mmput(prev_mm);
2178
2179	i = 0;
2180	while (si->inuse_pages &&
2181	       !signal_pending(current) &&
2182	       (i = find_next_to_unuse(si, i, frontswap)) != 0) {
2183
2184		entry = swp_entry(type, i);
2185		page = find_get_page(swap_address_space(entry), i);
2186		if (!page)
2187			continue;
2188
2189		/*
2190		 * It is conceivable that a racing task removed this page from
2191		 * swap cache just before we acquired the page lock. The page
2192		 * might even be back in swap cache on another swap area. But
2193		 * that is okay, try_to_free_swap() only removes stale pages.
2194		 */
2195		lock_page(page);
2196		wait_on_page_writeback(page);
2197		try_to_free_swap(page);
2198		unlock_page(page);
2199		put_page(page);
2200
2201		/*
2202		 * For frontswap, we just need to unuse pages_to_unuse, if
2203		 * it was specified. Need not check frontswap again here as
2204		 * we already zeroed out pages_to_unuse if not frontswap.
2205		 */
2206		if (pages_to_unuse && --pages_to_unuse == 0)
2207			goto out;
2208	}
2209
2210	/*
2211	 * Lets check again to see if there are still swap entries in the map.
2212	 * If yes, we would need to do retry the unuse logic again.
2213	 * Under global memory pressure, swap entries can be reinserted back
2214	 * into process space after the mmlist loop above passes over them.
2215	 *
2216	 * Limit the number of retries? No: when mmget_not_zero() above fails,
2217	 * that mm is likely to be freeing swap from exit_mmap(), which proceeds
2218	 * at its own independent pace; and even shmem_writepage() could have
2219	 * been preempted after get_swap_page(), temporarily hiding that swap.
2220	 * It's easy and robust (though cpu-intensive) just to keep retrying.
2221	 */
2222	if (si->inuse_pages) {
2223		if (!signal_pending(current))
2224			goto retry;
2225		retval = -EINTR;
2226	}
2227out:
2228	return (retval == FRONTSWAP_PAGES_UNUSED) ? 0 : retval;
2229}
2230
2231/*
2232 * After a successful try_to_unuse, if no swap is now in use, we know
2233 * we can empty the mmlist.  swap_lock must be held on entry and exit.
2234 * Note that mmlist_lock nests inside swap_lock, and an mm must be
2235 * added to the mmlist just after page_duplicate - before would be racy.
2236 */
2237static void drain_mmlist(void)
2238{
2239	struct list_head *p, *next;
2240	unsigned int type;
2241
2242	for (type = 0; type < nr_swapfiles; type++)
2243		if (swap_info[type]->inuse_pages)
2244			return;
2245	spin_lock(&mmlist_lock);
2246	list_for_each_safe(p, next, &init_mm.mmlist)
2247		list_del_init(p);
2248	spin_unlock(&mmlist_lock);
2249}
2250
2251/*
2252 * Use this swapdev's extent info to locate the (PAGE_SIZE) block which
2253 * corresponds to page offset for the specified swap entry.
2254 * Note that the type of this function is sector_t, but it returns page offset
2255 * into the bdev, not sector offset.
2256 */
2257static sector_t map_swap_entry(swp_entry_t entry, struct block_device **bdev)
2258{
2259	struct swap_info_struct *sis;
2260	struct swap_extent *se;
2261	pgoff_t offset;
2262
2263	sis = swp_swap_info(entry);
2264	*bdev = sis->bdev;
2265
2266	offset = swp_offset(entry);
2267	se = offset_to_swap_extent(sis, offset);
2268	return se->start_block + (offset - se->start_page);
2269}
2270
2271/*
2272 * Returns the page offset into bdev for the specified page's swap entry.
2273 */
2274sector_t map_swap_page(struct page *page, struct block_device **bdev)
2275{
2276	swp_entry_t entry;
2277	entry.val = page_private(page);
2278	return map_swap_entry(entry, bdev);
2279}
2280
2281/*
2282 * Free all of a swapdev's extent information
2283 */
2284static void destroy_swap_extents(struct swap_info_struct *sis)
2285{
2286	while (!RB_EMPTY_ROOT(&sis->swap_extent_root)) {
2287		struct rb_node *rb = sis->swap_extent_root.rb_node;
2288		struct swap_extent *se = rb_entry(rb, struct swap_extent, rb_node);
2289
2290		rb_erase(rb, &sis->swap_extent_root);
2291		kfree(se);
2292	}
2293
2294	if (sis->flags & SWP_ACTIVATED) {
2295		struct file *swap_file = sis->swap_file;
2296		struct address_space *mapping = swap_file->f_mapping;
2297
2298		sis->flags &= ~SWP_ACTIVATED;
2299		if (mapping->a_ops->swap_deactivate)
2300			mapping->a_ops->swap_deactivate(swap_file);
2301	}
2302}
2303
2304/*
2305 * Add a block range (and the corresponding page range) into this swapdev's
2306 * extent tree.
2307 *
2308 * This function rather assumes that it is called in ascending page order.
2309 */
2310int
2311add_swap_extent(struct swap_info_struct *sis, unsigned long start_page,
2312		unsigned long nr_pages, sector_t start_block)
2313{
2314	struct rb_node **link = &sis->swap_extent_root.rb_node, *parent = NULL;
2315	struct swap_extent *se;
2316	struct swap_extent *new_se;
2317
2318	/*
2319	 * place the new node at the right most since the
2320	 * function is called in ascending page order.
2321	 */
2322	while (*link) {
2323		parent = *link;
2324		link = &parent->rb_right;
2325	}
2326
2327	if (parent) {
2328		se = rb_entry(parent, struct swap_extent, rb_node);
2329		BUG_ON(se->start_page + se->nr_pages != start_page);
2330		if (se->start_block + se->nr_pages == start_block) {
2331			/* Merge it */
2332			se->nr_pages += nr_pages;
2333			return 0;
2334		}
2335	}
2336
2337	/* No merge, insert a new extent. */
2338	new_se = kmalloc(sizeof(*se), GFP_KERNEL);
2339	if (new_se == NULL)
2340		return -ENOMEM;
2341	new_se->start_page = start_page;
2342	new_se->nr_pages = nr_pages;
2343	new_se->start_block = start_block;
2344
2345	rb_link_node(&new_se->rb_node, parent, link);
2346	rb_insert_color(&new_se->rb_node, &sis->swap_extent_root);
2347	return 1;
2348}
2349EXPORT_SYMBOL_GPL(add_swap_extent);
2350
2351/*
2352 * A `swap extent' is a simple thing which maps a contiguous range of pages
2353 * onto a contiguous range of disk blocks.  An ordered list of swap extents
2354 * is built at swapon time and is then used at swap_writepage/swap_readpage
2355 * time for locating where on disk a page belongs.
2356 *
2357 * If the swapfile is an S_ISBLK block device, a single extent is installed.
2358 * This is done so that the main operating code can treat S_ISBLK and S_ISREG
2359 * swap files identically.
2360 *
2361 * Whether the swapdev is an S_ISREG file or an S_ISBLK blockdev, the swap
2362 * extent list operates in PAGE_SIZE disk blocks.  Both S_ISREG and S_ISBLK
2363 * swapfiles are handled *identically* after swapon time.
2364 *
2365 * For S_ISREG swapfiles, setup_swap_extents() will walk all the file's blocks
2366 * and will parse them into an ordered extent list, in PAGE_SIZE chunks.  If
2367 * some stray blocks are found which do not fall within the PAGE_SIZE alignment
2368 * requirements, they are simply tossed out - we will never use those blocks
2369 * for swapping.
2370 *
2371 * For all swap devices we set S_SWAPFILE across the life of the swapon.  This
2372 * prevents users from writing to the swap device, which will corrupt memory.
2373 *
2374 * The amount of disk space which a single swap extent represents varies.
2375 * Typically it is in the 1-4 megabyte range.  So we can have hundreds of
2376 * extents in the list.  To avoid much list walking, we cache the previous
2377 * search location in `curr_swap_extent', and start new searches from there.
2378 * This is extremely effective.  The average number of iterations in
2379 * map_swap_page() has been measured at about 0.3 per page.  - akpm.
2380 */
2381static int setup_swap_extents(struct swap_info_struct *sis, sector_t *span)
2382{
2383	struct file *swap_file = sis->swap_file;
2384	struct address_space *mapping = swap_file->f_mapping;
2385	struct inode *inode = mapping->host;
2386	int ret;
2387
2388	if (S_ISBLK(inode->i_mode)) {
2389		ret = add_swap_extent(sis, 0, sis->max, 0);
2390		*span = sis->pages;
2391		return ret;
2392	}
2393
2394	if (mapping->a_ops->swap_activate) {
2395		ret = mapping->a_ops->swap_activate(sis, swap_file, span);
2396		if (ret >= 0)
2397			sis->flags |= SWP_ACTIVATED;
2398		if (!ret) {
2399			sis->flags |= SWP_FS;
2400			ret = add_swap_extent(sis, 0, sis->max, 0);
2401			*span = sis->pages;
2402		}
2403		return ret;
2404	}
2405
2406	return generic_swapfile_activate(sis, swap_file, span);
2407}
2408
2409static int swap_node(struct swap_info_struct *p)
2410{
2411	struct block_device *bdev;
2412
2413	if (p->bdev)
2414		bdev = p->bdev;
2415	else
2416		bdev = p->swap_file->f_inode->i_sb->s_bdev;
2417
2418	return bdev ? bdev->bd_disk->node_id : NUMA_NO_NODE;
2419}
2420
2421static void setup_swap_info(struct swap_info_struct *p, int prio,
2422			    unsigned char *swap_map,
2423			    struct swap_cluster_info *cluster_info)
2424{
2425	int i;
2426
2427	if (prio >= 0)
2428		p->prio = prio;
2429	else
2430		p->prio = --least_priority;
2431	/*
2432	 * the plist prio is negated because plist ordering is
2433	 * low-to-high, while swap ordering is high-to-low
2434	 */
2435	p->list.prio = -p->prio;
2436	for_each_node(i) {
2437		if (p->prio >= 0)
2438			p->avail_lists[i].prio = -p->prio;
2439		else {
2440			if (swap_node(p) == i)
2441				p->avail_lists[i].prio = 1;
2442			else
2443				p->avail_lists[i].prio = -p->prio;
2444		}
2445	}
2446	p->swap_map = swap_map;
2447	p->cluster_info = cluster_info;
2448}
2449
2450static void _enable_swap_info(struct swap_info_struct *p)
2451{
2452	p->flags |= SWP_WRITEOK | SWP_VALID;
2453	atomic_long_add(p->pages, &nr_swap_pages);
2454	total_swap_pages += p->pages;
2455
2456	assert_spin_locked(&swap_lock);
2457	/*
2458	 * both lists are plists, and thus priority ordered.
2459	 * swap_active_head needs to be priority ordered for swapoff(),
2460	 * which on removal of any swap_info_struct with an auto-assigned
2461	 * (i.e. negative) priority increments the auto-assigned priority
2462	 * of any lower-priority swap_info_structs.
2463	 * swap_avail_head needs to be priority ordered for get_swap_page(),
2464	 * which allocates swap pages from the highest available priority
2465	 * swap_info_struct.
2466	 */
2467	plist_add(&p->list, &swap_active_head);
2468	add_to_avail_list(p);
2469}
2470
2471static void enable_swap_info(struct swap_info_struct *p, int prio,
2472				unsigned char *swap_map,
2473				struct swap_cluster_info *cluster_info,
2474				unsigned long *frontswap_map)
2475{
2476	frontswap_init(p->type, frontswap_map);
2477	spin_lock(&swap_lock);
2478	spin_lock(&p->lock);
2479	setup_swap_info(p, prio, swap_map, cluster_info);
2480	spin_unlock(&p->lock);
2481	spin_unlock(&swap_lock);
2482	/*
2483	 * Guarantee swap_map, cluster_info, etc. fields are valid
2484	 * between get/put_swap_device() if SWP_VALID bit is set
2485	 */
2486	synchronize_rcu();
2487	spin_lock(&swap_lock);
2488	spin_lock(&p->lock);
2489	_enable_swap_info(p);
2490	spin_unlock(&p->lock);
2491	spin_unlock(&swap_lock);
2492}
2493
2494static void reinsert_swap_info(struct swap_info_struct *p)
2495{
2496	spin_lock(&swap_lock);
2497	spin_lock(&p->lock);
2498	setup_swap_info(p, p->prio, p->swap_map, p->cluster_info);
2499	_enable_swap_info(p);
2500	spin_unlock(&p->lock);
2501	spin_unlock(&swap_lock);
2502}
2503
2504bool has_usable_swap(void)
2505{
2506	bool ret = true;
2507
2508	spin_lock(&swap_lock);
2509	if (plist_head_empty(&swap_active_head))
2510		ret = false;
2511	spin_unlock(&swap_lock);
2512	return ret;
2513}
2514
2515SYSCALL_DEFINE1(swapoff, const char __user *, specialfile)
2516{
2517	struct swap_info_struct *p = NULL;
2518	unsigned char *swap_map;
2519	struct swap_cluster_info *cluster_info;
2520	unsigned long *frontswap_map;
2521	struct file *swap_file, *victim;
2522	struct address_space *mapping;
2523	struct inode *inode;
2524	struct filename *pathname;
2525	int err, found = 0;
2526	unsigned int old_block_size;
2527
2528	if (!capable(CAP_SYS_ADMIN))
2529		return -EPERM;
2530
2531	BUG_ON(!current->mm);
2532
2533	pathname = getname(specialfile);
2534	if (IS_ERR(pathname))
2535		return PTR_ERR(pathname);
2536
2537	victim = file_open_name(pathname, O_RDWR|O_LARGEFILE, 0);
2538	err = PTR_ERR(victim);
2539	if (IS_ERR(victim))
2540		goto out;
2541
2542	mapping = victim->f_mapping;
2543	spin_lock(&swap_lock);
2544	plist_for_each_entry(p, &swap_active_head, list) {
2545		if (p->flags & SWP_WRITEOK) {
2546			if (p->swap_file->f_mapping == mapping) {
2547				found = 1;
2548				break;
2549			}
2550		}
2551	}
2552	if (!found) {
2553		err = -EINVAL;
2554		spin_unlock(&swap_lock);
2555		goto out_dput;
2556	}
2557	if (!security_vm_enough_memory_mm(current->mm, p->pages))
2558		vm_unacct_memory(p->pages);
2559	else {
2560		err = -ENOMEM;
2561		spin_unlock(&swap_lock);
2562		goto out_dput;
2563	}
2564	del_from_avail_list(p);
2565	spin_lock(&p->lock);
2566	if (p->prio < 0) {
2567		struct swap_info_struct *si = p;
2568		int nid;
2569
2570		plist_for_each_entry_continue(si, &swap_active_head, list) {
2571			si->prio++;
2572			si->list.prio--;
2573			for_each_node(nid) {
2574				if (si->avail_lists[nid].prio != 1)
2575					si->avail_lists[nid].prio--;
2576			}
2577		}
2578		least_priority++;
2579	}
2580	plist_del(&p->list, &swap_active_head);
2581	atomic_long_sub(p->pages, &nr_swap_pages);
2582	total_swap_pages -= p->pages;
2583	p->flags &= ~SWP_WRITEOK;
2584	spin_unlock(&p->lock);
2585	spin_unlock(&swap_lock);
2586
2587	disable_swap_slots_cache_lock();
2588
2589	set_current_oom_origin();
2590	err = try_to_unuse(p->type, false, 0); /* force unuse all pages */
2591	clear_current_oom_origin();
2592
2593	if (err) {
2594		/* re-insert swap space back into swap_list */
2595		reinsert_swap_info(p);
2596		reenable_swap_slots_cache_unlock();
2597		goto out_dput;
2598	}
2599
2600	reenable_swap_slots_cache_unlock();
2601
2602	spin_lock(&swap_lock);
2603	spin_lock(&p->lock);
2604	p->flags &= ~SWP_VALID;		/* mark swap device as invalid */
2605	spin_unlock(&p->lock);
2606	spin_unlock(&swap_lock);
2607	/*
2608	 * wait for swap operations protected by get/put_swap_device()
2609	 * to complete
2610	 */
2611	synchronize_rcu();
2612
2613	flush_work(&p->discard_work);
2614
2615	destroy_swap_extents(p);
2616	if (p->flags & SWP_CONTINUED)
2617		free_swap_count_continuations(p);
2618
2619	if (!p->bdev || !blk_queue_nonrot(bdev_get_queue(p->bdev)))
2620		atomic_dec(&nr_rotate_swap);
2621
2622	mutex_lock(&swapon_mutex);
2623	spin_lock(&swap_lock);
2624	spin_lock(&p->lock);
2625	drain_mmlist();
2626
2627	/* wait for anyone still in scan_swap_map */
2628	p->highest_bit = 0;		/* cuts scans short */
2629	while (p->flags >= SWP_SCANNING) {
2630		spin_unlock(&p->lock);
2631		spin_unlock(&swap_lock);
2632		schedule_timeout_uninterruptible(1);
2633		spin_lock(&swap_lock);
2634		spin_lock(&p->lock);
2635	}
2636
2637	swap_file = p->swap_file;
2638	old_block_size = p->old_block_size;
2639	p->swap_file = NULL;
2640	p->max = 0;
2641	swap_map = p->swap_map;
2642	p->swap_map = NULL;
2643	cluster_info = p->cluster_info;
2644	p->cluster_info = NULL;
2645	frontswap_map = frontswap_map_get(p);
2646	spin_unlock(&p->lock);
2647	spin_unlock(&swap_lock);
2648	frontswap_invalidate_area(p->type);
2649	frontswap_map_set(p, NULL);
2650	mutex_unlock(&swapon_mutex);
2651	free_percpu(p->percpu_cluster);
2652	p->percpu_cluster = NULL;
 
 
2653	vfree(swap_map);
2654	kvfree(cluster_info);
2655	kvfree(frontswap_map);
2656	/* Destroy swap account information */
2657	swap_cgroup_swapoff(p->type);
2658	exit_swap_address_space(p->type);
2659
2660	inode = mapping->host;
2661	if (S_ISBLK(inode->i_mode)) {
2662		struct block_device *bdev = I_BDEV(inode);
2663
2664		set_blocksize(bdev, old_block_size);
2665		blkdev_put(bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
2666	}
2667
2668	inode_lock(inode);
2669	inode->i_flags &= ~S_SWAPFILE;
2670	inode_unlock(inode);
2671	filp_close(swap_file, NULL);
2672
2673	/*
2674	 * Clear the SWP_USED flag after all resources are freed so that swapon
2675	 * can reuse this swap_info in alloc_swap_info() safely.  It is ok to
2676	 * not hold p->lock after we cleared its SWP_WRITEOK.
2677	 */
2678	spin_lock(&swap_lock);
2679	p->flags = 0;
2680	spin_unlock(&swap_lock);
2681
2682	err = 0;
2683	atomic_inc(&proc_poll_event);
2684	wake_up_interruptible(&proc_poll_wait);
2685
2686out_dput:
2687	filp_close(victim, NULL);
2688out:
2689	putname(pathname);
2690	return err;
2691}
2692
2693#ifdef CONFIG_PROC_FS
2694static __poll_t swaps_poll(struct file *file, poll_table *wait)
2695{
2696	struct seq_file *seq = file->private_data;
2697
2698	poll_wait(file, &proc_poll_wait, wait);
2699
2700	if (seq->poll_event != atomic_read(&proc_poll_event)) {
2701		seq->poll_event = atomic_read(&proc_poll_event);
2702		return EPOLLIN | EPOLLRDNORM | EPOLLERR | EPOLLPRI;
2703	}
2704
2705	return EPOLLIN | EPOLLRDNORM;
2706}
2707
2708/* iterator */
2709static void *swap_start(struct seq_file *swap, loff_t *pos)
2710{
2711	struct swap_info_struct *si;
2712	int type;
2713	loff_t l = *pos;
2714
2715	mutex_lock(&swapon_mutex);
2716
2717	if (!l)
2718		return SEQ_START_TOKEN;
2719
2720	for (type = 0; (si = swap_type_to_swap_info(type)); type++) {
2721		if (!(si->flags & SWP_USED) || !si->swap_map)
2722			continue;
2723		if (!--l)
2724			return si;
2725	}
2726
2727	return NULL;
2728}
2729
2730static void *swap_next(struct seq_file *swap, void *v, loff_t *pos)
2731{
2732	struct swap_info_struct *si = v;
2733	int type;
2734
2735	if (v == SEQ_START_TOKEN)
2736		type = 0;
2737	else
2738		type = si->type + 1;
2739
 
2740	for (; (si = swap_type_to_swap_info(type)); type++) {
2741		if (!(si->flags & SWP_USED) || !si->swap_map)
2742			continue;
2743		++*pos;
2744		return si;
2745	}
2746
2747	return NULL;
2748}
2749
2750static void swap_stop(struct seq_file *swap, void *v)
2751{
2752	mutex_unlock(&swapon_mutex);
2753}
2754
2755static int swap_show(struct seq_file *swap, void *v)
2756{
2757	struct swap_info_struct *si = v;
2758	struct file *file;
2759	int len;
 
2760
2761	if (si == SEQ_START_TOKEN) {
2762		seq_puts(swap,"Filename\t\t\t\tType\t\tSize\tUsed\tPriority\n");
2763		return 0;
2764	}
2765
 
 
 
2766	file = si->swap_file;
2767	len = seq_file_path(swap, file, " \t\n\\");
2768	seq_printf(swap, "%*s%s\t%u\t%u\t%d\n",
2769			len < 40 ? 40 - len : 1, " ",
2770			S_ISBLK(file_inode(file)->i_mode) ?
2771				"partition" : "file\t",
2772			si->pages << (PAGE_SHIFT - 10),
2773			si->inuse_pages << (PAGE_SHIFT - 10),
2774			si->prio);
2775	return 0;
2776}
2777
2778static const struct seq_operations swaps_op = {
2779	.start =	swap_start,
2780	.next =		swap_next,
2781	.stop =		swap_stop,
2782	.show =		swap_show
2783};
2784
2785static int swaps_open(struct inode *inode, struct file *file)
2786{
2787	struct seq_file *seq;
2788	int ret;
2789
2790	ret = seq_open(file, &swaps_op);
2791	if (ret)
2792		return ret;
2793
2794	seq = file->private_data;
2795	seq->poll_event = atomic_read(&proc_poll_event);
2796	return 0;
2797}
2798
2799static const struct file_operations proc_swaps_operations = {
2800	.open		= swaps_open,
2801	.read		= seq_read,
2802	.llseek		= seq_lseek,
2803	.release	= seq_release,
2804	.poll		= swaps_poll,
 
2805};
2806
2807static int __init procswaps_init(void)
2808{
2809	proc_create("swaps", 0, NULL, &proc_swaps_operations);
2810	return 0;
2811}
2812__initcall(procswaps_init);
2813#endif /* CONFIG_PROC_FS */
2814
2815#ifdef MAX_SWAPFILES_CHECK
2816static int __init max_swapfiles_check(void)
2817{
2818	MAX_SWAPFILES_CHECK();
2819	return 0;
2820}
2821late_initcall(max_swapfiles_check);
2822#endif
2823
2824static struct swap_info_struct *alloc_swap_info(void)
2825{
2826	struct swap_info_struct *p;
2827	unsigned int type;
2828	int i;
2829
2830	p = kvzalloc(struct_size(p, avail_lists, nr_node_ids), GFP_KERNEL);
2831	if (!p)
2832		return ERR_PTR(-ENOMEM);
2833
2834	spin_lock(&swap_lock);
2835	for (type = 0; type < nr_swapfiles; type++) {
2836		if (!(swap_info[type]->flags & SWP_USED))
2837			break;
2838	}
2839	if (type >= MAX_SWAPFILES) {
2840		spin_unlock(&swap_lock);
2841		kvfree(p);
2842		return ERR_PTR(-EPERM);
2843	}
2844	if (type >= nr_swapfiles) {
2845		p->type = type;
2846		WRITE_ONCE(swap_info[type], p);
2847		/*
2848		 * Write swap_info[type] before nr_swapfiles, in case a
2849		 * racing procfs swap_start() or swap_next() is reading them.
2850		 * (We never shrink nr_swapfiles, we never free this entry.)
2851		 */
2852		smp_wmb();
2853		WRITE_ONCE(nr_swapfiles, nr_swapfiles + 1);
2854	} else {
2855		kvfree(p);
2856		p = swap_info[type];
2857		/*
2858		 * Do not memset this entry: a racing procfs swap_next()
2859		 * would be relying on p->type to remain valid.
2860		 */
2861	}
2862	p->swap_extent_root = RB_ROOT;
2863	plist_node_init(&p->list, 0);
2864	for_each_node(i)
2865		plist_node_init(&p->avail_lists[i], 0);
2866	p->flags = SWP_USED;
2867	spin_unlock(&swap_lock);
2868	spin_lock_init(&p->lock);
2869	spin_lock_init(&p->cont_lock);
2870
2871	return p;
2872}
2873
2874static int claim_swapfile(struct swap_info_struct *p, struct inode *inode)
2875{
2876	int error;
2877
2878	if (S_ISBLK(inode->i_mode)) {
2879		p->bdev = bdgrab(I_BDEV(inode));
2880		error = blkdev_get(p->bdev,
2881				   FMODE_READ | FMODE_WRITE | FMODE_EXCL, p);
2882		if (error < 0) {
2883			p->bdev = NULL;
2884			return error;
2885		}
2886		p->old_block_size = block_size(p->bdev);
2887		error = set_blocksize(p->bdev, PAGE_SIZE);
2888		if (error < 0)
2889			return error;
 
 
 
 
 
 
 
2890		p->flags |= SWP_BLKDEV;
2891	} else if (S_ISREG(inode->i_mode)) {
2892		p->bdev = inode->i_sb->s_bdev;
2893	}
2894
2895	inode_lock(inode);
2896	if (IS_SWAPFILE(inode))
2897		return -EBUSY;
2898
2899	return 0;
2900}
2901
2902
2903/*
2904 * Find out how many pages are allowed for a single swap device. There
2905 * are two limiting factors:
2906 * 1) the number of bits for the swap offset in the swp_entry_t type, and
2907 * 2) the number of bits in the swap pte, as defined by the different
2908 * architectures.
2909 *
2910 * In order to find the largest possible bit mask, a swap entry with
2911 * swap type 0 and swap offset ~0UL is created, encoded to a swap pte,
2912 * decoded to a swp_entry_t again, and finally the swap offset is
2913 * extracted.
2914 *
2915 * This will mask all the bits from the initial ~0UL mask that can't
2916 * be encoded in either the swp_entry_t or the architecture definition
2917 * of a swap pte.
2918 */
2919unsigned long generic_max_swapfile_size(void)
2920{
2921	return swp_offset(pte_to_swp_entry(
2922			swp_entry_to_pte(swp_entry(0, ~0UL)))) + 1;
2923}
2924
2925/* Can be overridden by an architecture for additional checks. */
2926__weak unsigned long max_swapfile_size(void)
2927{
2928	return generic_max_swapfile_size();
2929}
2930
2931static unsigned long read_swap_header(struct swap_info_struct *p,
2932					union swap_header *swap_header,
2933					struct inode *inode)
2934{
2935	int i;
2936	unsigned long maxpages;
2937	unsigned long swapfilepages;
2938	unsigned long last_page;
2939
2940	if (memcmp("SWAPSPACE2", swap_header->magic.magic, 10)) {
2941		pr_err("Unable to find swap-space signature\n");
2942		return 0;
2943	}
2944
2945	/* swap partition endianess hack... */
2946	if (swab32(swap_header->info.version) == 1) {
2947		swab32s(&swap_header->info.version);
2948		swab32s(&swap_header->info.last_page);
2949		swab32s(&swap_header->info.nr_badpages);
2950		if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES)
2951			return 0;
2952		for (i = 0; i < swap_header->info.nr_badpages; i++)
2953			swab32s(&swap_header->info.badpages[i]);
2954	}
2955	/* Check the swap header's sub-version */
2956	if (swap_header->info.version != 1) {
2957		pr_warn("Unable to handle swap header version %d\n",
2958			swap_header->info.version);
2959		return 0;
2960	}
2961
2962	p->lowest_bit  = 1;
2963	p->cluster_next = 1;
2964	p->cluster_nr = 0;
2965
2966	maxpages = max_swapfile_size();
2967	last_page = swap_header->info.last_page;
2968	if (!last_page) {
2969		pr_warn("Empty swap-file\n");
2970		return 0;
2971	}
2972	if (last_page > maxpages) {
2973		pr_warn("Truncating oversized swap area, only using %luk out of %luk\n",
2974			maxpages << (PAGE_SHIFT - 10),
2975			last_page << (PAGE_SHIFT - 10));
2976	}
2977	if (maxpages > last_page) {
2978		maxpages = last_page + 1;
2979		/* p->max is an unsigned int: don't overflow it */
2980		if ((unsigned int)maxpages == 0)
2981			maxpages = UINT_MAX;
2982	}
2983	p->highest_bit = maxpages - 1;
2984
2985	if (!maxpages)
2986		return 0;
2987	swapfilepages = i_size_read(inode) >> PAGE_SHIFT;
2988	if (swapfilepages && maxpages > swapfilepages) {
2989		pr_warn("Swap area shorter than signature indicates\n");
2990		return 0;
2991	}
2992	if (swap_header->info.nr_badpages && S_ISREG(inode->i_mode))
2993		return 0;
2994	if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES)
2995		return 0;
2996
2997	return maxpages;
2998}
2999
3000#define SWAP_CLUSTER_INFO_COLS						\
3001	DIV_ROUND_UP(L1_CACHE_BYTES, sizeof(struct swap_cluster_info))
3002#define SWAP_CLUSTER_SPACE_COLS						\
3003	DIV_ROUND_UP(SWAP_ADDRESS_SPACE_PAGES, SWAPFILE_CLUSTER)
3004#define SWAP_CLUSTER_COLS						\
3005	max_t(unsigned int, SWAP_CLUSTER_INFO_COLS, SWAP_CLUSTER_SPACE_COLS)
3006
3007static int setup_swap_map_and_extents(struct swap_info_struct *p,
3008					union swap_header *swap_header,
3009					unsigned char *swap_map,
3010					struct swap_cluster_info *cluster_info,
3011					unsigned long maxpages,
3012					sector_t *span)
3013{
3014	unsigned int j, k;
3015	unsigned int nr_good_pages;
3016	int nr_extents;
3017	unsigned long nr_clusters = DIV_ROUND_UP(maxpages, SWAPFILE_CLUSTER);
3018	unsigned long col = p->cluster_next / SWAPFILE_CLUSTER % SWAP_CLUSTER_COLS;
3019	unsigned long i, idx;
3020
3021	nr_good_pages = maxpages - 1;	/* omit header page */
3022
3023	cluster_list_init(&p->free_clusters);
3024	cluster_list_init(&p->discard_clusters);
3025
3026	for (i = 0; i < swap_header->info.nr_badpages; i++) {
3027		unsigned int page_nr = swap_header->info.badpages[i];
3028		if (page_nr == 0 || page_nr > swap_header->info.last_page)
3029			return -EINVAL;
3030		if (page_nr < maxpages) {
3031			swap_map[page_nr] = SWAP_MAP_BAD;
3032			nr_good_pages--;
3033			/*
3034			 * Haven't marked the cluster free yet, no list
3035			 * operation involved
3036			 */
3037			inc_cluster_info_page(p, cluster_info, page_nr);
3038		}
3039	}
3040
3041	/* Haven't marked the cluster free yet, no list operation involved */
3042	for (i = maxpages; i < round_up(maxpages, SWAPFILE_CLUSTER); i++)
3043		inc_cluster_info_page(p, cluster_info, i);
3044
3045	if (nr_good_pages) {
3046		swap_map[0] = SWAP_MAP_BAD;
3047		/*
3048		 * Not mark the cluster free yet, no list
3049		 * operation involved
3050		 */
3051		inc_cluster_info_page(p, cluster_info, 0);
3052		p->max = maxpages;
3053		p->pages = nr_good_pages;
3054		nr_extents = setup_swap_extents(p, span);
3055		if (nr_extents < 0)
3056			return nr_extents;
3057		nr_good_pages = p->pages;
3058	}
3059	if (!nr_good_pages) {
3060		pr_warn("Empty swap-file\n");
3061		return -EINVAL;
3062	}
3063
3064	if (!cluster_info)
3065		return nr_extents;
3066
3067
3068	/*
3069	 * Reduce false cache line sharing between cluster_info and
3070	 * sharing same address space.
3071	 */
3072	for (k = 0; k < SWAP_CLUSTER_COLS; k++) {
3073		j = (k + col) % SWAP_CLUSTER_COLS;
3074		for (i = 0; i < DIV_ROUND_UP(nr_clusters, SWAP_CLUSTER_COLS); i++) {
3075			idx = i * SWAP_CLUSTER_COLS + j;
3076			if (idx >= nr_clusters)
3077				continue;
3078			if (cluster_count(&cluster_info[idx]))
3079				continue;
3080			cluster_set_flag(&cluster_info[idx], CLUSTER_FLAG_FREE);
3081			cluster_list_add_tail(&p->free_clusters, cluster_info,
3082					      idx);
3083		}
3084	}
3085	return nr_extents;
3086}
3087
3088/*
3089 * Helper to sys_swapon determining if a given swap
3090 * backing device queue supports DISCARD operations.
3091 */
3092static bool swap_discardable(struct swap_info_struct *si)
3093{
3094	struct request_queue *q = bdev_get_queue(si->bdev);
3095
3096	if (!q || !blk_queue_discard(q))
3097		return false;
3098
3099	return true;
3100}
3101
3102SYSCALL_DEFINE2(swapon, const char __user *, specialfile, int, swap_flags)
3103{
3104	struct swap_info_struct *p;
3105	struct filename *name;
3106	struct file *swap_file = NULL;
3107	struct address_space *mapping;
3108	int prio;
3109	int error;
3110	union swap_header *swap_header;
3111	int nr_extents;
3112	sector_t span;
3113	unsigned long maxpages;
3114	unsigned char *swap_map = NULL;
3115	struct swap_cluster_info *cluster_info = NULL;
3116	unsigned long *frontswap_map = NULL;
3117	struct page *page = NULL;
3118	struct inode *inode = NULL;
3119	bool inced_nr_rotate_swap = false;
3120
3121	if (swap_flags & ~SWAP_FLAGS_VALID)
3122		return -EINVAL;
3123
3124	if (!capable(CAP_SYS_ADMIN))
3125		return -EPERM;
3126
3127	if (!swap_avail_heads)
3128		return -ENOMEM;
3129
3130	p = alloc_swap_info();
3131	if (IS_ERR(p))
3132		return PTR_ERR(p);
3133
3134	INIT_WORK(&p->discard_work, swap_discard_work);
3135
3136	name = getname(specialfile);
3137	if (IS_ERR(name)) {
3138		error = PTR_ERR(name);
3139		name = NULL;
3140		goto bad_swap;
3141	}
3142	swap_file = file_open_name(name, O_RDWR|O_LARGEFILE, 0);
3143	if (IS_ERR(swap_file)) {
3144		error = PTR_ERR(swap_file);
3145		swap_file = NULL;
3146		goto bad_swap;
3147	}
3148
3149	p->swap_file = swap_file;
3150	mapping = swap_file->f_mapping;
3151	inode = mapping->host;
3152
3153	/* If S_ISREG(inode->i_mode) will do inode_lock(inode); */
3154	error = claim_swapfile(p, inode);
3155	if (unlikely(error))
3156		goto bad_swap;
3157
 
 
 
 
 
 
3158	/*
3159	 * Read the swap header.
3160	 */
3161	if (!mapping->a_ops->readpage) {
3162		error = -EINVAL;
3163		goto bad_swap;
3164	}
3165	page = read_mapping_page(mapping, 0, swap_file);
3166	if (IS_ERR(page)) {
3167		error = PTR_ERR(page);
3168		goto bad_swap;
3169	}
3170	swap_header = kmap(page);
3171
3172	maxpages = read_swap_header(p, swap_header, inode);
3173	if (unlikely(!maxpages)) {
3174		error = -EINVAL;
3175		goto bad_swap;
3176	}
3177
3178	/* OK, set up the swap map and apply the bad block list */
3179	swap_map = vzalloc(maxpages);
3180	if (!swap_map) {
3181		error = -ENOMEM;
3182		goto bad_swap;
3183	}
3184
3185	if (bdi_cap_stable_pages_required(inode_to_bdi(inode)))
3186		p->flags |= SWP_STABLE_WRITES;
3187
3188	if (bdi_cap_synchronous_io(inode_to_bdi(inode)))
3189		p->flags |= SWP_SYNCHRONOUS_IO;
3190
3191	if (p->bdev && blk_queue_nonrot(bdev_get_queue(p->bdev))) {
3192		int cpu;
3193		unsigned long ci, nr_cluster;
3194
3195		p->flags |= SWP_SOLIDSTATE;
 
 
 
 
 
3196		/*
3197		 * select a random position to start with to help wear leveling
3198		 * SSD
3199		 */
3200		p->cluster_next = 1 + (prandom_u32() % p->highest_bit);
 
 
 
3201		nr_cluster = DIV_ROUND_UP(maxpages, SWAPFILE_CLUSTER);
3202
3203		cluster_info = kvcalloc(nr_cluster, sizeof(*cluster_info),
3204					GFP_KERNEL);
3205		if (!cluster_info) {
3206			error = -ENOMEM;
3207			goto bad_swap;
3208		}
3209
3210		for (ci = 0; ci < nr_cluster; ci++)
3211			spin_lock_init(&((cluster_info + ci)->lock));
3212
3213		p->percpu_cluster = alloc_percpu(struct percpu_cluster);
3214		if (!p->percpu_cluster) {
3215			error = -ENOMEM;
3216			goto bad_swap;
3217		}
3218		for_each_possible_cpu(cpu) {
3219			struct percpu_cluster *cluster;
3220			cluster = per_cpu_ptr(p->percpu_cluster, cpu);
3221			cluster_set_null(&cluster->index);
3222		}
3223	} else {
3224		atomic_inc(&nr_rotate_swap);
3225		inced_nr_rotate_swap = true;
3226	}
3227
3228	error = swap_cgroup_swapon(p->type, maxpages);
3229	if (error)
3230		goto bad_swap;
3231
3232	nr_extents = setup_swap_map_and_extents(p, swap_header, swap_map,
3233		cluster_info, maxpages, &span);
3234	if (unlikely(nr_extents < 0)) {
3235		error = nr_extents;
3236		goto bad_swap;
3237	}
3238	/* frontswap enabled? set up bit-per-page map for frontswap */
3239	if (IS_ENABLED(CONFIG_FRONTSWAP))
3240		frontswap_map = kvcalloc(BITS_TO_LONGS(maxpages),
3241					 sizeof(long),
3242					 GFP_KERNEL);
3243
3244	if (p->bdev &&(swap_flags & SWAP_FLAG_DISCARD) && swap_discardable(p)) {
3245		/*
3246		 * When discard is enabled for swap with no particular
3247		 * policy flagged, we set all swap discard flags here in
3248		 * order to sustain backward compatibility with older
3249		 * swapon(8) releases.
3250		 */
3251		p->flags |= (SWP_DISCARDABLE | SWP_AREA_DISCARD |
3252			     SWP_PAGE_DISCARD);
3253
3254		/*
3255		 * By flagging sys_swapon, a sysadmin can tell us to
3256		 * either do single-time area discards only, or to just
3257		 * perform discards for released swap page-clusters.
3258		 * Now it's time to adjust the p->flags accordingly.
3259		 */
3260		if (swap_flags & SWAP_FLAG_DISCARD_ONCE)
3261			p->flags &= ~SWP_PAGE_DISCARD;
3262		else if (swap_flags & SWAP_FLAG_DISCARD_PAGES)
3263			p->flags &= ~SWP_AREA_DISCARD;
3264
3265		/* issue a swapon-time discard if it's still required */
3266		if (p->flags & SWP_AREA_DISCARD) {
3267			int err = discard_swap(p);
3268			if (unlikely(err))
3269				pr_err("swapon: discard_swap(%p): %d\n",
3270					p, err);
3271		}
3272	}
3273
3274	error = init_swap_address_space(p->type, maxpages);
3275	if (error)
3276		goto bad_swap;
3277
3278	/*
3279	 * Flush any pending IO and dirty mappings before we start using this
3280	 * swap device.
3281	 */
3282	inode->i_flags |= S_SWAPFILE;
3283	error = inode_drain_writes(inode);
3284	if (error) {
3285		inode->i_flags &= ~S_SWAPFILE;
3286		goto bad_swap;
3287	}
3288
3289	mutex_lock(&swapon_mutex);
3290	prio = -1;
3291	if (swap_flags & SWAP_FLAG_PREFER)
3292		prio =
3293		  (swap_flags & SWAP_FLAG_PRIO_MASK) >> SWAP_FLAG_PRIO_SHIFT;
3294	enable_swap_info(p, prio, swap_map, cluster_info, frontswap_map);
3295
3296	pr_info("Adding %uk swap on %s.  Priority:%d extents:%d across:%lluk %s%s%s%s%s\n",
3297		p->pages<<(PAGE_SHIFT-10), name->name, p->prio,
3298		nr_extents, (unsigned long long)span<<(PAGE_SHIFT-10),
3299		(p->flags & SWP_SOLIDSTATE) ? "SS" : "",
3300		(p->flags & SWP_DISCARDABLE) ? "D" : "",
3301		(p->flags & SWP_AREA_DISCARD) ? "s" : "",
3302		(p->flags & SWP_PAGE_DISCARD) ? "c" : "",
3303		(frontswap_map) ? "FS" : "");
3304
3305	mutex_unlock(&swapon_mutex);
3306	atomic_inc(&proc_poll_event);
3307	wake_up_interruptible(&proc_poll_wait);
3308
3309	error = 0;
3310	goto out;
 
 
3311bad_swap:
3312	free_percpu(p->percpu_cluster);
3313	p->percpu_cluster = NULL;
 
 
3314	if (inode && S_ISBLK(inode->i_mode) && p->bdev) {
3315		set_blocksize(p->bdev, p->old_block_size);
3316		blkdev_put(p->bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
3317	}
 
3318	destroy_swap_extents(p);
3319	swap_cgroup_swapoff(p->type);
3320	spin_lock(&swap_lock);
3321	p->swap_file = NULL;
3322	p->flags = 0;
3323	spin_unlock(&swap_lock);
3324	vfree(swap_map);
3325	kvfree(cluster_info);
3326	kvfree(frontswap_map);
3327	if (inced_nr_rotate_swap)
3328		atomic_dec(&nr_rotate_swap);
3329	if (swap_file) {
3330		if (inode) {
3331			inode_unlock(inode);
3332			inode = NULL;
3333		}
3334		filp_close(swap_file, NULL);
3335	}
3336out:
3337	if (page && !IS_ERR(page)) {
3338		kunmap(page);
3339		put_page(page);
3340	}
3341	if (name)
3342		putname(name);
3343	if (inode)
3344		inode_unlock(inode);
3345	if (!error)
3346		enable_swap_slots_cache();
3347	return error;
3348}
3349
3350void si_swapinfo(struct sysinfo *val)
3351{
3352	unsigned int type;
3353	unsigned long nr_to_be_unused = 0;
3354
3355	spin_lock(&swap_lock);
3356	for (type = 0; type < nr_swapfiles; type++) {
3357		struct swap_info_struct *si = swap_info[type];
3358
3359		if ((si->flags & SWP_USED) && !(si->flags & SWP_WRITEOK))
3360			nr_to_be_unused += si->inuse_pages;
3361	}
3362	val->freeswap = atomic_long_read(&nr_swap_pages) + nr_to_be_unused;
3363	val->totalswap = total_swap_pages + nr_to_be_unused;
3364	spin_unlock(&swap_lock);
3365}
3366
3367/*
3368 * Verify that a swap entry is valid and increment its swap map count.
3369 *
3370 * Returns error code in following case.
3371 * - success -> 0
3372 * - swp_entry is invalid -> EINVAL
3373 * - swp_entry is migration entry -> EINVAL
3374 * - swap-cache reference is requested but there is already one. -> EEXIST
3375 * - swap-cache reference is requested but the entry is not used. -> ENOENT
3376 * - swap-mapped reference requested but needs continued swap count. -> ENOMEM
3377 */
3378static int __swap_duplicate(swp_entry_t entry, unsigned char usage)
3379{
3380	struct swap_info_struct *p;
3381	struct swap_cluster_info *ci;
3382	unsigned long offset;
3383	unsigned char count;
3384	unsigned char has_cache;
3385	int err = -EINVAL;
3386
3387	p = get_swap_device(entry);
3388	if (!p)
3389		goto out;
3390
3391	offset = swp_offset(entry);
3392	ci = lock_cluster_or_swap_info(p, offset);
3393
3394	count = p->swap_map[offset];
3395
3396	/*
3397	 * swapin_readahead() doesn't check if a swap entry is valid, so the
3398	 * swap entry could be SWAP_MAP_BAD. Check here with lock held.
3399	 */
3400	if (unlikely(swap_count(count) == SWAP_MAP_BAD)) {
3401		err = -ENOENT;
3402		goto unlock_out;
3403	}
3404
3405	has_cache = count & SWAP_HAS_CACHE;
3406	count &= ~SWAP_HAS_CACHE;
3407	err = 0;
3408
3409	if (usage == SWAP_HAS_CACHE) {
3410
3411		/* set SWAP_HAS_CACHE if there is no cache and entry is used */
3412		if (!has_cache && count)
3413			has_cache = SWAP_HAS_CACHE;
3414		else if (has_cache)		/* someone else added cache */
3415			err = -EEXIST;
3416		else				/* no users remaining */
3417			err = -ENOENT;
3418
3419	} else if (count || has_cache) {
3420
3421		if ((count & ~COUNT_CONTINUED) < SWAP_MAP_MAX)
3422			count += usage;
3423		else if ((count & ~COUNT_CONTINUED) > SWAP_MAP_MAX)
3424			err = -EINVAL;
3425		else if (swap_count_continued(p, offset, count))
3426			count = COUNT_CONTINUED;
3427		else
3428			err = -ENOMEM;
3429	} else
3430		err = -ENOENT;			/* unused swap entry */
3431
3432	p->swap_map[offset] = count | has_cache;
3433
3434unlock_out:
3435	unlock_cluster_or_swap_info(p, ci);
3436out:
3437	if (p)
3438		put_swap_device(p);
3439	return err;
3440}
3441
3442/*
3443 * Help swapoff by noting that swap entry belongs to shmem/tmpfs
3444 * (in which case its reference count is never incremented).
3445 */
3446void swap_shmem_alloc(swp_entry_t entry)
3447{
3448	__swap_duplicate(entry, SWAP_MAP_SHMEM);
3449}
3450
3451/*
3452 * Increase reference count of swap entry by 1.
3453 * Returns 0 for success, or -ENOMEM if a swap_count_continuation is required
3454 * but could not be atomically allocated.  Returns 0, just as if it succeeded,
3455 * if __swap_duplicate() fails for another reason (-EINVAL or -ENOENT), which
3456 * might occur if a page table entry has got corrupted.
3457 */
3458int swap_duplicate(swp_entry_t entry)
3459{
3460	int err = 0;
3461
3462	while (!err && __swap_duplicate(entry, 1) == -ENOMEM)
3463		err = add_swap_count_continuation(entry, GFP_ATOMIC);
3464	return err;
3465}
3466
3467/*
3468 * @entry: swap entry for which we allocate swap cache.
3469 *
3470 * Called when allocating swap cache for existing swap entry,
3471 * This can return error codes. Returns 0 at success.
3472 * -EBUSY means there is a swap cache.
3473 * Note: return code is different from swap_duplicate().
3474 */
3475int swapcache_prepare(swp_entry_t entry)
3476{
3477	return __swap_duplicate(entry, SWAP_HAS_CACHE);
3478}
3479
3480struct swap_info_struct *swp_swap_info(swp_entry_t entry)
3481{
3482	return swap_type_to_swap_info(swp_type(entry));
3483}
3484
3485struct swap_info_struct *page_swap_info(struct page *page)
3486{
3487	swp_entry_t entry = { .val = page_private(page) };
3488	return swp_swap_info(entry);
3489}
3490
3491/*
3492 * out-of-line __page_file_ methods to avoid include hell.
3493 */
3494struct address_space *__page_file_mapping(struct page *page)
3495{
3496	return page_swap_info(page)->swap_file->f_mapping;
3497}
3498EXPORT_SYMBOL_GPL(__page_file_mapping);
3499
3500pgoff_t __page_file_index(struct page *page)
3501{
3502	swp_entry_t swap = { .val = page_private(page) };
3503	return swp_offset(swap);
3504}
3505EXPORT_SYMBOL_GPL(__page_file_index);
3506
3507/*
3508 * add_swap_count_continuation - called when a swap count is duplicated
3509 * beyond SWAP_MAP_MAX, it allocates a new page and links that to the entry's
3510 * page of the original vmalloc'ed swap_map, to hold the continuation count
3511 * (for that entry and for its neighbouring PAGE_SIZE swap entries).  Called
3512 * again when count is duplicated beyond SWAP_MAP_MAX * SWAP_CONT_MAX, etc.
3513 *
3514 * These continuation pages are seldom referenced: the common paths all work
3515 * on the original swap_map, only referring to a continuation page when the
3516 * low "digit" of a count is incremented or decremented through SWAP_MAP_MAX.
3517 *
3518 * add_swap_count_continuation(, GFP_ATOMIC) can be called while holding
3519 * page table locks; if it fails, add_swap_count_continuation(, GFP_KERNEL)
3520 * can be called after dropping locks.
3521 */
3522int add_swap_count_continuation(swp_entry_t entry, gfp_t gfp_mask)
3523{
3524	struct swap_info_struct *si;
3525	struct swap_cluster_info *ci;
3526	struct page *head;
3527	struct page *page;
3528	struct page *list_page;
3529	pgoff_t offset;
3530	unsigned char count;
3531	int ret = 0;
3532
3533	/*
3534	 * When debugging, it's easier to use __GFP_ZERO here; but it's better
3535	 * for latency not to zero a page while GFP_ATOMIC and holding locks.
3536	 */
3537	page = alloc_page(gfp_mask | __GFP_HIGHMEM);
3538
3539	si = get_swap_device(entry);
3540	if (!si) {
3541		/*
3542		 * An acceptable race has occurred since the failing
3543		 * __swap_duplicate(): the swap device may be swapoff
3544		 */
3545		goto outer;
3546	}
3547	spin_lock(&si->lock);
3548
3549	offset = swp_offset(entry);
3550
3551	ci = lock_cluster(si, offset);
3552
3553	count = si->swap_map[offset] & ~SWAP_HAS_CACHE;
3554
3555	if ((count & ~COUNT_CONTINUED) != SWAP_MAP_MAX) {
3556		/*
3557		 * The higher the swap count, the more likely it is that tasks
3558		 * will race to add swap count continuation: we need to avoid
3559		 * over-provisioning.
3560		 */
3561		goto out;
3562	}
3563
3564	if (!page) {
3565		ret = -ENOMEM;
3566		goto out;
3567	}
3568
3569	/*
3570	 * We are fortunate that although vmalloc_to_page uses pte_offset_map,
3571	 * no architecture is using highmem pages for kernel page tables: so it
3572	 * will not corrupt the GFP_ATOMIC caller's atomic page table kmaps.
3573	 */
3574	head = vmalloc_to_page(si->swap_map + offset);
3575	offset &= ~PAGE_MASK;
3576
3577	spin_lock(&si->cont_lock);
3578	/*
3579	 * Page allocation does not initialize the page's lru field,
3580	 * but it does always reset its private field.
3581	 */
3582	if (!page_private(head)) {
3583		BUG_ON(count & COUNT_CONTINUED);
3584		INIT_LIST_HEAD(&head->lru);
3585		set_page_private(head, SWP_CONTINUED);
3586		si->flags |= SWP_CONTINUED;
3587	}
3588
3589	list_for_each_entry(list_page, &head->lru, lru) {
3590		unsigned char *map;
3591
3592		/*
3593		 * If the previous map said no continuation, but we've found
3594		 * a continuation page, free our allocation and use this one.
3595		 */
3596		if (!(count & COUNT_CONTINUED))
3597			goto out_unlock_cont;
3598
3599		map = kmap_atomic(list_page) + offset;
3600		count = *map;
3601		kunmap_atomic(map);
3602
3603		/*
3604		 * If this continuation count now has some space in it,
3605		 * free our allocation and use this one.
3606		 */
3607		if ((count & ~COUNT_CONTINUED) != SWAP_CONT_MAX)
3608			goto out_unlock_cont;
3609	}
3610
3611	list_add_tail(&page->lru, &head->lru);
3612	page = NULL;			/* now it's attached, don't free it */
3613out_unlock_cont:
3614	spin_unlock(&si->cont_lock);
3615out:
3616	unlock_cluster(ci);
3617	spin_unlock(&si->lock);
3618	put_swap_device(si);
3619outer:
3620	if (page)
3621		__free_page(page);
3622	return ret;
3623}
3624
3625/*
3626 * swap_count_continued - when the original swap_map count is incremented
3627 * from SWAP_MAP_MAX, check if there is already a continuation page to carry
3628 * into, carry if so, or else fail until a new continuation page is allocated;
3629 * when the original swap_map count is decremented from 0 with continuation,
3630 * borrow from the continuation and report whether it still holds more.
3631 * Called while __swap_duplicate() or swap_entry_free() holds swap or cluster
3632 * lock.
3633 */
3634static bool swap_count_continued(struct swap_info_struct *si,
3635				 pgoff_t offset, unsigned char count)
3636{
3637	struct page *head;
3638	struct page *page;
3639	unsigned char *map;
3640	bool ret;
3641
3642	head = vmalloc_to_page(si->swap_map + offset);
3643	if (page_private(head) != SWP_CONTINUED) {
3644		BUG_ON(count & COUNT_CONTINUED);
3645		return false;		/* need to add count continuation */
3646	}
3647
3648	spin_lock(&si->cont_lock);
3649	offset &= ~PAGE_MASK;
3650	page = list_entry(head->lru.next, struct page, lru);
3651	map = kmap_atomic(page) + offset;
3652
3653	if (count == SWAP_MAP_MAX)	/* initial increment from swap_map */
3654		goto init_map;		/* jump over SWAP_CONT_MAX checks */
3655
3656	if (count == (SWAP_MAP_MAX | COUNT_CONTINUED)) { /* incrementing */
3657		/*
3658		 * Think of how you add 1 to 999
3659		 */
3660		while (*map == (SWAP_CONT_MAX | COUNT_CONTINUED)) {
3661			kunmap_atomic(map);
3662			page = list_entry(page->lru.next, struct page, lru);
3663			BUG_ON(page == head);
3664			map = kmap_atomic(page) + offset;
3665		}
3666		if (*map == SWAP_CONT_MAX) {
3667			kunmap_atomic(map);
3668			page = list_entry(page->lru.next, struct page, lru);
3669			if (page == head) {
3670				ret = false;	/* add count continuation */
3671				goto out;
3672			}
3673			map = kmap_atomic(page) + offset;
3674init_map:		*map = 0;		/* we didn't zero the page */
3675		}
3676		*map += 1;
3677		kunmap_atomic(map);
3678		page = list_entry(page->lru.prev, struct page, lru);
3679		while (page != head) {
3680			map = kmap_atomic(page) + offset;
3681			*map = COUNT_CONTINUED;
3682			kunmap_atomic(map);
3683			page = list_entry(page->lru.prev, struct page, lru);
3684		}
3685		ret = true;			/* incremented */
3686
3687	} else {				/* decrementing */
3688		/*
3689		 * Think of how you subtract 1 from 1000
3690		 */
3691		BUG_ON(count != COUNT_CONTINUED);
3692		while (*map == COUNT_CONTINUED) {
3693			kunmap_atomic(map);
3694			page = list_entry(page->lru.next, struct page, lru);
3695			BUG_ON(page == head);
3696			map = kmap_atomic(page) + offset;
3697		}
3698		BUG_ON(*map == 0);
3699		*map -= 1;
3700		if (*map == 0)
3701			count = 0;
3702		kunmap_atomic(map);
3703		page = list_entry(page->lru.prev, struct page, lru);
3704		while (page != head) {
3705			map = kmap_atomic(page) + offset;
3706			*map = SWAP_CONT_MAX | count;
3707			count = COUNT_CONTINUED;
3708			kunmap_atomic(map);
3709			page = list_entry(page->lru.prev, struct page, lru);
3710		}
3711		ret = count == COUNT_CONTINUED;
3712	}
3713out:
3714	spin_unlock(&si->cont_lock);
3715	return ret;
3716}
3717
3718/*
3719 * free_swap_count_continuations - swapoff free all the continuation pages
3720 * appended to the swap_map, after swap_map is quiesced, before vfree'ing it.
3721 */
3722static void free_swap_count_continuations(struct swap_info_struct *si)
3723{
3724	pgoff_t offset;
3725
3726	for (offset = 0; offset < si->max; offset += PAGE_SIZE) {
3727		struct page *head;
3728		head = vmalloc_to_page(si->swap_map + offset);
3729		if (page_private(head)) {
3730			struct page *page, *next;
3731
3732			list_for_each_entry_safe(page, next, &head->lru, lru) {
3733				list_del(&page->lru);
3734				__free_page(page);
3735			}
3736		}
3737	}
3738}
3739
3740#if defined(CONFIG_MEMCG) && defined(CONFIG_BLK_CGROUP)
3741void mem_cgroup_throttle_swaprate(struct mem_cgroup *memcg, int node,
3742				  gfp_t gfp_mask)
3743{
3744	struct swap_info_struct *si, *next;
3745	if (!(gfp_mask & __GFP_IO) || !memcg)
 
 
3746		return;
3747
3748	if (!blk_cgroup_congested())
3749		return;
3750
3751	/*
3752	 * We've already scheduled a throttle, avoid taking the global swap
3753	 * lock.
3754	 */
3755	if (current->throttle_queue)
3756		return;
3757
3758	spin_lock(&swap_avail_lock);
3759	plist_for_each_entry_safe(si, next, &swap_avail_heads[node],
3760				  avail_lists[node]) {
3761		if (si->bdev) {
3762			blkcg_schedule_throttle(bdev_get_queue(si->bdev),
3763						true);
3764			break;
3765		}
3766	}
3767	spin_unlock(&swap_avail_lock);
3768}
3769#endif
3770
3771static int __init swapfile_init(void)
3772{
3773	int nid;
3774
3775	swap_avail_heads = kmalloc_array(nr_node_ids, sizeof(struct plist_head),
3776					 GFP_KERNEL);
3777	if (!swap_avail_heads) {
3778		pr_emerg("Not enough memory for swap heads, swap is disabled\n");
3779		return -ENOMEM;
3780	}
3781
3782	for_each_node(nid)
3783		plist_head_init(&swap_avail_heads[nid]);
3784
3785	return 0;
3786}
3787subsys_initcall(swapfile_init);
v5.9
   1// SPDX-License-Identifier: GPL-2.0-only
   2/*
   3 *  linux/mm/swapfile.c
   4 *
   5 *  Copyright (C) 1991, 1992, 1993, 1994  Linus Torvalds
   6 *  Swap reorganised 29.12.95, Stephen Tweedie
   7 */
   8
   9#include <linux/mm.h>
  10#include <linux/sched/mm.h>
  11#include <linux/sched/task.h>
  12#include <linux/hugetlb.h>
  13#include <linux/mman.h>
  14#include <linux/slab.h>
  15#include <linux/kernel_stat.h>
  16#include <linux/swap.h>
  17#include <linux/vmalloc.h>
  18#include <linux/pagemap.h>
  19#include <linux/namei.h>
  20#include <linux/shmem_fs.h>
  21#include <linux/blkdev.h>
  22#include <linux/random.h>
  23#include <linux/writeback.h>
  24#include <linux/proc_fs.h>
  25#include <linux/seq_file.h>
  26#include <linux/init.h>
  27#include <linux/ksm.h>
  28#include <linux/rmap.h>
  29#include <linux/security.h>
  30#include <linux/backing-dev.h>
  31#include <linux/mutex.h>
  32#include <linux/capability.h>
  33#include <linux/syscalls.h>
  34#include <linux/memcontrol.h>
  35#include <linux/poll.h>
  36#include <linux/oom.h>
  37#include <linux/frontswap.h>
  38#include <linux/swapfile.h>
  39#include <linux/export.h>
  40#include <linux/swap_slots.h>
  41#include <linux/sort.h>
  42
 
  43#include <asm/tlbflush.h>
  44#include <linux/swapops.h>
  45#include <linux/swap_cgroup.h>
  46
  47static bool swap_count_continued(struct swap_info_struct *, pgoff_t,
  48				 unsigned char);
  49static void free_swap_count_continuations(struct swap_info_struct *);
  50static sector_t map_swap_entry(swp_entry_t, struct block_device**);
  51
  52DEFINE_SPINLOCK(swap_lock);
  53static unsigned int nr_swapfiles;
  54atomic_long_t nr_swap_pages;
  55/*
  56 * Some modules use swappable objects and may try to swap them out under
  57 * memory pressure (via the shrinker). Before doing so, they may wish to
  58 * check to see if any swap space is available.
  59 */
  60EXPORT_SYMBOL_GPL(nr_swap_pages);
  61/* protected with swap_lock. reading in vm_swap_full() doesn't need lock */
  62long total_swap_pages;
  63static int least_priority = -1;
  64
  65static const char Bad_file[] = "Bad swap file entry ";
  66static const char Unused_file[] = "Unused swap file entry ";
  67static const char Bad_offset[] = "Bad swap offset entry ";
  68static const char Unused_offset[] = "Unused swap offset entry ";
  69
  70/*
  71 * all active swap_info_structs
  72 * protected with swap_lock, and ordered by priority.
  73 */
  74PLIST_HEAD(swap_active_head);
  75
  76/*
  77 * all available (active, not full) swap_info_structs
  78 * protected with swap_avail_lock, ordered by priority.
  79 * This is used by get_swap_page() instead of swap_active_head
  80 * because swap_active_head includes all swap_info_structs,
  81 * but get_swap_page() doesn't need to look at full ones.
  82 * This uses its own lock instead of swap_lock because when a
  83 * swap_info_struct changes between not-full/full, it needs to
  84 * add/remove itself to/from this list, but the swap_info_struct->lock
  85 * is held and the locking order requires swap_lock to be taken
  86 * before any swap_info_struct->lock.
  87 */
  88static struct plist_head *swap_avail_heads;
  89static DEFINE_SPINLOCK(swap_avail_lock);
  90
  91struct swap_info_struct *swap_info[MAX_SWAPFILES];
  92
  93static DEFINE_MUTEX(swapon_mutex);
  94
  95static DECLARE_WAIT_QUEUE_HEAD(proc_poll_wait);
  96/* Activity counter to indicate that a swapon or swapoff has occurred */
  97static atomic_t proc_poll_event = ATOMIC_INIT(0);
  98
  99atomic_t nr_rotate_swap = ATOMIC_INIT(0);
 100
 101static struct swap_info_struct *swap_type_to_swap_info(int type)
 102{
 103	if (type >= READ_ONCE(nr_swapfiles))
 104		return NULL;
 105
 106	smp_rmb();	/* Pairs with smp_wmb in alloc_swap_info. */
 107	return READ_ONCE(swap_info[type]);
 108}
 109
 110static inline unsigned char swap_count(unsigned char ent)
 111{
 112	return ent & ~SWAP_HAS_CACHE;	/* may include COUNT_CONTINUED flag */
 113}
 114
 115/* Reclaim the swap entry anyway if possible */
 116#define TTRS_ANYWAY		0x1
 117/*
 118 * Reclaim the swap entry if there are no more mappings of the
 119 * corresponding page
 120 */
 121#define TTRS_UNMAPPED		0x2
 122/* Reclaim the swap entry if swap is getting full*/
 123#define TTRS_FULL		0x4
 124
 125/* returns 1 if swap entry is freed */
 126static int __try_to_reclaim_swap(struct swap_info_struct *si,
 127				 unsigned long offset, unsigned long flags)
 128{
 129	swp_entry_t entry = swp_entry(si->type, offset);
 130	struct page *page;
 131	int ret = 0;
 132
 133	page = find_get_page(swap_address_space(entry), offset);
 134	if (!page)
 135		return 0;
 136	/*
 137	 * When this function is called from scan_swap_map_slots() and it's
 138	 * called by vmscan.c at reclaiming pages. So, we hold a lock on a page,
 139	 * here. We have to use trylock for avoiding deadlock. This is a special
 140	 * case and you should use try_to_free_swap() with explicit lock_page()
 141	 * in usual operations.
 142	 */
 143	if (trylock_page(page)) {
 144		if ((flags & TTRS_ANYWAY) ||
 145		    ((flags & TTRS_UNMAPPED) && !page_mapped(page)) ||
 146		    ((flags & TTRS_FULL) && mem_cgroup_swap_full(page)))
 147			ret = try_to_free_swap(page);
 148		unlock_page(page);
 149	}
 150	put_page(page);
 151	return ret;
 152}
 153
 154static inline struct swap_extent *first_se(struct swap_info_struct *sis)
 155{
 156	struct rb_node *rb = rb_first(&sis->swap_extent_root);
 157	return rb_entry(rb, struct swap_extent, rb_node);
 158}
 159
 160static inline struct swap_extent *next_se(struct swap_extent *se)
 161{
 162	struct rb_node *rb = rb_next(&se->rb_node);
 163	return rb ? rb_entry(rb, struct swap_extent, rb_node) : NULL;
 164}
 165
 166/*
 167 * swapon tell device that all the old swap contents can be discarded,
 168 * to allow the swap device to optimize its wear-levelling.
 169 */
 170static int discard_swap(struct swap_info_struct *si)
 171{
 172	struct swap_extent *se;
 173	sector_t start_block;
 174	sector_t nr_blocks;
 175	int err = 0;
 176
 177	/* Do not discard the swap header page! */
 178	se = first_se(si);
 179	start_block = (se->start_block + 1) << (PAGE_SHIFT - 9);
 180	nr_blocks = ((sector_t)se->nr_pages - 1) << (PAGE_SHIFT - 9);
 181	if (nr_blocks) {
 182		err = blkdev_issue_discard(si->bdev, start_block,
 183				nr_blocks, GFP_KERNEL, 0);
 184		if (err)
 185			return err;
 186		cond_resched();
 187	}
 188
 189	for (se = next_se(se); se; se = next_se(se)) {
 190		start_block = se->start_block << (PAGE_SHIFT - 9);
 191		nr_blocks = (sector_t)se->nr_pages << (PAGE_SHIFT - 9);
 192
 193		err = blkdev_issue_discard(si->bdev, start_block,
 194				nr_blocks, GFP_KERNEL, 0);
 195		if (err)
 196			break;
 197
 198		cond_resched();
 199	}
 200	return err;		/* That will often be -EOPNOTSUPP */
 201}
 202
 203static struct swap_extent *
 204offset_to_swap_extent(struct swap_info_struct *sis, unsigned long offset)
 205{
 206	struct swap_extent *se;
 207	struct rb_node *rb;
 208
 209	rb = sis->swap_extent_root.rb_node;
 210	while (rb) {
 211		se = rb_entry(rb, struct swap_extent, rb_node);
 212		if (offset < se->start_page)
 213			rb = rb->rb_left;
 214		else if (offset >= se->start_page + se->nr_pages)
 215			rb = rb->rb_right;
 216		else
 217			return se;
 218	}
 219	/* It *must* be present */
 220	BUG();
 221}
 222
 223/*
 224 * swap allocation tell device that a cluster of swap can now be discarded,
 225 * to allow the swap device to optimize its wear-levelling.
 226 */
 227static void discard_swap_cluster(struct swap_info_struct *si,
 228				 pgoff_t start_page, pgoff_t nr_pages)
 229{
 230	struct swap_extent *se = offset_to_swap_extent(si, start_page);
 231
 232	while (nr_pages) {
 233		pgoff_t offset = start_page - se->start_page;
 234		sector_t start_block = se->start_block + offset;
 235		sector_t nr_blocks = se->nr_pages - offset;
 236
 237		if (nr_blocks > nr_pages)
 238			nr_blocks = nr_pages;
 239		start_page += nr_blocks;
 240		nr_pages -= nr_blocks;
 241
 242		start_block <<= PAGE_SHIFT - 9;
 243		nr_blocks <<= PAGE_SHIFT - 9;
 244		if (blkdev_issue_discard(si->bdev, start_block,
 245					nr_blocks, GFP_NOIO, 0))
 246			break;
 247
 248		se = next_se(se);
 249	}
 250}
 251
 252#ifdef CONFIG_THP_SWAP
 253#define SWAPFILE_CLUSTER	HPAGE_PMD_NR
 254
 255#define swap_entry_size(size)	(size)
 256#else
 257#define SWAPFILE_CLUSTER	256
 258
 259/*
 260 * Define swap_entry_size() as constant to let compiler to optimize
 261 * out some code if !CONFIG_THP_SWAP
 262 */
 263#define swap_entry_size(size)	1
 264#endif
 265#define LATENCY_LIMIT		256
 266
 267static inline void cluster_set_flag(struct swap_cluster_info *info,
 268	unsigned int flag)
 269{
 270	info->flags = flag;
 271}
 272
 273static inline unsigned int cluster_count(struct swap_cluster_info *info)
 274{
 275	return info->data;
 276}
 277
 278static inline void cluster_set_count(struct swap_cluster_info *info,
 279				     unsigned int c)
 280{
 281	info->data = c;
 282}
 283
 284static inline void cluster_set_count_flag(struct swap_cluster_info *info,
 285					 unsigned int c, unsigned int f)
 286{
 287	info->flags = f;
 288	info->data = c;
 289}
 290
 291static inline unsigned int cluster_next(struct swap_cluster_info *info)
 292{
 293	return info->data;
 294}
 295
 296static inline void cluster_set_next(struct swap_cluster_info *info,
 297				    unsigned int n)
 298{
 299	info->data = n;
 300}
 301
 302static inline void cluster_set_next_flag(struct swap_cluster_info *info,
 303					 unsigned int n, unsigned int f)
 304{
 305	info->flags = f;
 306	info->data = n;
 307}
 308
 309static inline bool cluster_is_free(struct swap_cluster_info *info)
 310{
 311	return info->flags & CLUSTER_FLAG_FREE;
 312}
 313
 314static inline bool cluster_is_null(struct swap_cluster_info *info)
 315{
 316	return info->flags & CLUSTER_FLAG_NEXT_NULL;
 317}
 318
 319static inline void cluster_set_null(struct swap_cluster_info *info)
 320{
 321	info->flags = CLUSTER_FLAG_NEXT_NULL;
 322	info->data = 0;
 323}
 324
 325static inline bool cluster_is_huge(struct swap_cluster_info *info)
 326{
 327	if (IS_ENABLED(CONFIG_THP_SWAP))
 328		return info->flags & CLUSTER_FLAG_HUGE;
 329	return false;
 330}
 331
 332static inline void cluster_clear_huge(struct swap_cluster_info *info)
 333{
 334	info->flags &= ~CLUSTER_FLAG_HUGE;
 335}
 336
 337static inline struct swap_cluster_info *lock_cluster(struct swap_info_struct *si,
 338						     unsigned long offset)
 339{
 340	struct swap_cluster_info *ci;
 341
 342	ci = si->cluster_info;
 343	if (ci) {
 344		ci += offset / SWAPFILE_CLUSTER;
 345		spin_lock(&ci->lock);
 346	}
 347	return ci;
 348}
 349
 350static inline void unlock_cluster(struct swap_cluster_info *ci)
 351{
 352	if (ci)
 353		spin_unlock(&ci->lock);
 354}
 355
 356/*
 357 * Determine the locking method in use for this device.  Return
 358 * swap_cluster_info if SSD-style cluster-based locking is in place.
 359 */
 360static inline struct swap_cluster_info *lock_cluster_or_swap_info(
 361		struct swap_info_struct *si, unsigned long offset)
 362{
 363	struct swap_cluster_info *ci;
 364
 365	/* Try to use fine-grained SSD-style locking if available: */
 366	ci = lock_cluster(si, offset);
 367	/* Otherwise, fall back to traditional, coarse locking: */
 368	if (!ci)
 369		spin_lock(&si->lock);
 370
 371	return ci;
 372}
 373
 374static inline void unlock_cluster_or_swap_info(struct swap_info_struct *si,
 375					       struct swap_cluster_info *ci)
 376{
 377	if (ci)
 378		unlock_cluster(ci);
 379	else
 380		spin_unlock(&si->lock);
 381}
 382
 383static inline bool cluster_list_empty(struct swap_cluster_list *list)
 384{
 385	return cluster_is_null(&list->head);
 386}
 387
 388static inline unsigned int cluster_list_first(struct swap_cluster_list *list)
 389{
 390	return cluster_next(&list->head);
 391}
 392
 393static void cluster_list_init(struct swap_cluster_list *list)
 394{
 395	cluster_set_null(&list->head);
 396	cluster_set_null(&list->tail);
 397}
 398
 399static void cluster_list_add_tail(struct swap_cluster_list *list,
 400				  struct swap_cluster_info *ci,
 401				  unsigned int idx)
 402{
 403	if (cluster_list_empty(list)) {
 404		cluster_set_next_flag(&list->head, idx, 0);
 405		cluster_set_next_flag(&list->tail, idx, 0);
 406	} else {
 407		struct swap_cluster_info *ci_tail;
 408		unsigned int tail = cluster_next(&list->tail);
 409
 410		/*
 411		 * Nested cluster lock, but both cluster locks are
 412		 * only acquired when we held swap_info_struct->lock
 413		 */
 414		ci_tail = ci + tail;
 415		spin_lock_nested(&ci_tail->lock, SINGLE_DEPTH_NESTING);
 416		cluster_set_next(ci_tail, idx);
 417		spin_unlock(&ci_tail->lock);
 418		cluster_set_next_flag(&list->tail, idx, 0);
 419	}
 420}
 421
 422static unsigned int cluster_list_del_first(struct swap_cluster_list *list,
 423					   struct swap_cluster_info *ci)
 424{
 425	unsigned int idx;
 426
 427	idx = cluster_next(&list->head);
 428	if (cluster_next(&list->tail) == idx) {
 429		cluster_set_null(&list->head);
 430		cluster_set_null(&list->tail);
 431	} else
 432		cluster_set_next_flag(&list->head,
 433				      cluster_next(&ci[idx]), 0);
 434
 435	return idx;
 436}
 437
 438/* Add a cluster to discard list and schedule it to do discard */
 439static void swap_cluster_schedule_discard(struct swap_info_struct *si,
 440		unsigned int idx)
 441{
 442	/*
 443	 * If scan_swap_map() can't find a free cluster, it will check
 444	 * si->swap_map directly. To make sure the discarding cluster isn't
 445	 * taken by scan_swap_map(), mark the swap entries bad (occupied). It
 446	 * will be cleared after discard
 447	 */
 448	memset(si->swap_map + idx * SWAPFILE_CLUSTER,
 449			SWAP_MAP_BAD, SWAPFILE_CLUSTER);
 450
 451	cluster_list_add_tail(&si->discard_clusters, si->cluster_info, idx);
 452
 453	schedule_work(&si->discard_work);
 454}
 455
 456static void __free_cluster(struct swap_info_struct *si, unsigned long idx)
 457{
 458	struct swap_cluster_info *ci = si->cluster_info;
 459
 460	cluster_set_flag(ci + idx, CLUSTER_FLAG_FREE);
 461	cluster_list_add_tail(&si->free_clusters, ci, idx);
 462}
 463
 464/*
 465 * Doing discard actually. After a cluster discard is finished, the cluster
 466 * will be added to free cluster list. caller should hold si->lock.
 467*/
 468static void swap_do_scheduled_discard(struct swap_info_struct *si)
 469{
 470	struct swap_cluster_info *info, *ci;
 471	unsigned int idx;
 472
 473	info = si->cluster_info;
 474
 475	while (!cluster_list_empty(&si->discard_clusters)) {
 476		idx = cluster_list_del_first(&si->discard_clusters, info);
 477		spin_unlock(&si->lock);
 478
 479		discard_swap_cluster(si, idx * SWAPFILE_CLUSTER,
 480				SWAPFILE_CLUSTER);
 481
 482		spin_lock(&si->lock);
 483		ci = lock_cluster(si, idx * SWAPFILE_CLUSTER);
 484		__free_cluster(si, idx);
 485		memset(si->swap_map + idx * SWAPFILE_CLUSTER,
 486				0, SWAPFILE_CLUSTER);
 487		unlock_cluster(ci);
 488	}
 489}
 490
 491static void swap_discard_work(struct work_struct *work)
 492{
 493	struct swap_info_struct *si;
 494
 495	si = container_of(work, struct swap_info_struct, discard_work);
 496
 497	spin_lock(&si->lock);
 498	swap_do_scheduled_discard(si);
 499	spin_unlock(&si->lock);
 500}
 501
 502static void alloc_cluster(struct swap_info_struct *si, unsigned long idx)
 503{
 504	struct swap_cluster_info *ci = si->cluster_info;
 505
 506	VM_BUG_ON(cluster_list_first(&si->free_clusters) != idx);
 507	cluster_list_del_first(&si->free_clusters, ci);
 508	cluster_set_count_flag(ci + idx, 0, 0);
 509}
 510
 511static void free_cluster(struct swap_info_struct *si, unsigned long idx)
 512{
 513	struct swap_cluster_info *ci = si->cluster_info + idx;
 514
 515	VM_BUG_ON(cluster_count(ci) != 0);
 516	/*
 517	 * If the swap is discardable, prepare discard the cluster
 518	 * instead of free it immediately. The cluster will be freed
 519	 * after discard.
 520	 */
 521	if ((si->flags & (SWP_WRITEOK | SWP_PAGE_DISCARD)) ==
 522	    (SWP_WRITEOK | SWP_PAGE_DISCARD)) {
 523		swap_cluster_schedule_discard(si, idx);
 524		return;
 525	}
 526
 527	__free_cluster(si, idx);
 528}
 529
 530/*
 531 * The cluster corresponding to page_nr will be used. The cluster will be
 532 * removed from free cluster list and its usage counter will be increased.
 533 */
 534static void inc_cluster_info_page(struct swap_info_struct *p,
 535	struct swap_cluster_info *cluster_info, unsigned long page_nr)
 536{
 537	unsigned long idx = page_nr / SWAPFILE_CLUSTER;
 538
 539	if (!cluster_info)
 540		return;
 541	if (cluster_is_free(&cluster_info[idx]))
 542		alloc_cluster(p, idx);
 543
 544	VM_BUG_ON(cluster_count(&cluster_info[idx]) >= SWAPFILE_CLUSTER);
 545	cluster_set_count(&cluster_info[idx],
 546		cluster_count(&cluster_info[idx]) + 1);
 547}
 548
 549/*
 550 * The cluster corresponding to page_nr decreases one usage. If the usage
 551 * counter becomes 0, which means no page in the cluster is in using, we can
 552 * optionally discard the cluster and add it to free cluster list.
 553 */
 554static void dec_cluster_info_page(struct swap_info_struct *p,
 555	struct swap_cluster_info *cluster_info, unsigned long page_nr)
 556{
 557	unsigned long idx = page_nr / SWAPFILE_CLUSTER;
 558
 559	if (!cluster_info)
 560		return;
 561
 562	VM_BUG_ON(cluster_count(&cluster_info[idx]) == 0);
 563	cluster_set_count(&cluster_info[idx],
 564		cluster_count(&cluster_info[idx]) - 1);
 565
 566	if (cluster_count(&cluster_info[idx]) == 0)
 567		free_cluster(p, idx);
 568}
 569
 570/*
 571 * It's possible scan_swap_map() uses a free cluster in the middle of free
 572 * cluster list. Avoiding such abuse to avoid list corruption.
 573 */
 574static bool
 575scan_swap_map_ssd_cluster_conflict(struct swap_info_struct *si,
 576	unsigned long offset)
 577{
 578	struct percpu_cluster *percpu_cluster;
 579	bool conflict;
 580
 581	offset /= SWAPFILE_CLUSTER;
 582	conflict = !cluster_list_empty(&si->free_clusters) &&
 583		offset != cluster_list_first(&si->free_clusters) &&
 584		cluster_is_free(&si->cluster_info[offset]);
 585
 586	if (!conflict)
 587		return false;
 588
 589	percpu_cluster = this_cpu_ptr(si->percpu_cluster);
 590	cluster_set_null(&percpu_cluster->index);
 591	return true;
 592}
 593
 594/*
 595 * Try to get a swap entry from current cpu's swap entry pool (a cluster). This
 596 * might involve allocating a new cluster for current CPU too.
 597 */
 598static bool scan_swap_map_try_ssd_cluster(struct swap_info_struct *si,
 599	unsigned long *offset, unsigned long *scan_base)
 600{
 601	struct percpu_cluster *cluster;
 602	struct swap_cluster_info *ci;
 
 603	unsigned long tmp, max;
 604
 605new_cluster:
 606	cluster = this_cpu_ptr(si->percpu_cluster);
 607	if (cluster_is_null(&cluster->index)) {
 608		if (!cluster_list_empty(&si->free_clusters)) {
 609			cluster->index = si->free_clusters.head;
 610			cluster->next = cluster_next(&cluster->index) *
 611					SWAPFILE_CLUSTER;
 612		} else if (!cluster_list_empty(&si->discard_clusters)) {
 613			/*
 614			 * we don't have free cluster but have some clusters in
 615			 * discarding, do discard now and reclaim them, then
 616			 * reread cluster_next_cpu since we dropped si->lock
 617			 */
 618			swap_do_scheduled_discard(si);
 619			*scan_base = this_cpu_read(*si->cluster_next_cpu);
 620			*offset = *scan_base;
 621			goto new_cluster;
 622		} else
 623			return false;
 624	}
 625
 
 
 626	/*
 627	 * Other CPUs can use our cluster if they can't find a free cluster,
 628	 * check if there is still free entry in the cluster
 629	 */
 630	tmp = cluster->next;
 631	max = min_t(unsigned long, si->max,
 632		    (cluster_next(&cluster->index) + 1) * SWAPFILE_CLUSTER);
 633	if (tmp < max) {
 634		ci = lock_cluster(si, tmp);
 635		while (tmp < max) {
 636			if (!si->swap_map[tmp])
 637				break;
 638			tmp++;
 
 
 
 639		}
 640		unlock_cluster(ci);
 641	}
 642	if (tmp >= max) {
 
 643		cluster_set_null(&cluster->index);
 644		goto new_cluster;
 645	}
 646	cluster->next = tmp + 1;
 647	*offset = tmp;
 648	*scan_base = tmp;
 649	return true;
 650}
 651
 652static void __del_from_avail_list(struct swap_info_struct *p)
 653{
 654	int nid;
 655
 656	for_each_node(nid)
 657		plist_del(&p->avail_lists[nid], &swap_avail_heads[nid]);
 658}
 659
 660static void del_from_avail_list(struct swap_info_struct *p)
 661{
 662	spin_lock(&swap_avail_lock);
 663	__del_from_avail_list(p);
 664	spin_unlock(&swap_avail_lock);
 665}
 666
 667static void swap_range_alloc(struct swap_info_struct *si, unsigned long offset,
 668			     unsigned int nr_entries)
 669{
 670	unsigned int end = offset + nr_entries - 1;
 671
 672	if (offset == si->lowest_bit)
 673		si->lowest_bit += nr_entries;
 674	if (end == si->highest_bit)
 675		WRITE_ONCE(si->highest_bit, si->highest_bit - nr_entries);
 676	si->inuse_pages += nr_entries;
 677	if (si->inuse_pages == si->pages) {
 678		si->lowest_bit = si->max;
 679		si->highest_bit = 0;
 680		del_from_avail_list(si);
 681	}
 682}
 683
 684static void add_to_avail_list(struct swap_info_struct *p)
 685{
 686	int nid;
 687
 688	spin_lock(&swap_avail_lock);
 689	for_each_node(nid) {
 690		WARN_ON(!plist_node_empty(&p->avail_lists[nid]));
 691		plist_add(&p->avail_lists[nid], &swap_avail_heads[nid]);
 692	}
 693	spin_unlock(&swap_avail_lock);
 694}
 695
 696static void swap_range_free(struct swap_info_struct *si, unsigned long offset,
 697			    unsigned int nr_entries)
 698{
 699	unsigned long begin = offset;
 700	unsigned long end = offset + nr_entries - 1;
 701	void (*swap_slot_free_notify)(struct block_device *, unsigned long);
 702
 703	if (offset < si->lowest_bit)
 704		si->lowest_bit = offset;
 705	if (end > si->highest_bit) {
 706		bool was_full = !si->highest_bit;
 707
 708		WRITE_ONCE(si->highest_bit, end);
 709		if (was_full && (si->flags & SWP_WRITEOK))
 710			add_to_avail_list(si);
 711	}
 712	atomic_long_add(nr_entries, &nr_swap_pages);
 713	si->inuse_pages -= nr_entries;
 714	if (si->flags & SWP_BLKDEV)
 715		swap_slot_free_notify =
 716			si->bdev->bd_disk->fops->swap_slot_free_notify;
 717	else
 718		swap_slot_free_notify = NULL;
 719	while (offset <= end) {
 720		frontswap_invalidate_page(si->type, offset);
 721		if (swap_slot_free_notify)
 722			swap_slot_free_notify(si->bdev, offset);
 723		offset++;
 724	}
 725	clear_shadow_from_swap_cache(si->type, begin, end);
 726}
 727
 728static void set_cluster_next(struct swap_info_struct *si, unsigned long next)
 729{
 730	unsigned long prev;
 731
 732	if (!(si->flags & SWP_SOLIDSTATE)) {
 733		si->cluster_next = next;
 734		return;
 735	}
 736
 737	prev = this_cpu_read(*si->cluster_next_cpu);
 738	/*
 739	 * Cross the swap address space size aligned trunk, choose
 740	 * another trunk randomly to avoid lock contention on swap
 741	 * address space if possible.
 742	 */
 743	if ((prev >> SWAP_ADDRESS_SPACE_SHIFT) !=
 744	    (next >> SWAP_ADDRESS_SPACE_SHIFT)) {
 745		/* No free swap slots available */
 746		if (si->highest_bit <= si->lowest_bit)
 747			return;
 748		next = si->lowest_bit +
 749			prandom_u32_max(si->highest_bit - si->lowest_bit + 1);
 750		next = ALIGN_DOWN(next, SWAP_ADDRESS_SPACE_PAGES);
 751		next = max_t(unsigned int, next, si->lowest_bit);
 752	}
 753	this_cpu_write(*si->cluster_next_cpu, next);
 754}
 755
 756static int scan_swap_map_slots(struct swap_info_struct *si,
 757			       unsigned char usage, int nr,
 758			       swp_entry_t slots[])
 759{
 760	struct swap_cluster_info *ci;
 761	unsigned long offset;
 762	unsigned long scan_base;
 763	unsigned long last_in_cluster = 0;
 764	int latency_ration = LATENCY_LIMIT;
 765	int n_ret = 0;
 766	bool scanned_many = false;
 
 
 767
 768	/*
 769	 * We try to cluster swap pages by allocating them sequentially
 770	 * in swap.  Once we've allocated SWAPFILE_CLUSTER pages this
 771	 * way, however, we resort to first-free allocation, starting
 772	 * a new cluster.  This prevents us from scattering swap pages
 773	 * all over the entire swap partition, so that we reduce
 774	 * overall disk seek times between swap pages.  -- sct
 775	 * But we do now try to find an empty cluster.  -Andrea
 776	 * And we let swap pages go all over an SSD partition.  Hugh
 777	 */
 778
 779	si->flags += SWP_SCANNING;
 780	/*
 781	 * Use percpu scan base for SSD to reduce lock contention on
 782	 * cluster and swap cache.  For HDD, sequential access is more
 783	 * important.
 784	 */
 785	if (si->flags & SWP_SOLIDSTATE)
 786		scan_base = this_cpu_read(*si->cluster_next_cpu);
 787	else
 788		scan_base = si->cluster_next;
 789	offset = scan_base;
 790
 791	/* SSD algorithm */
 792	if (si->cluster_info) {
 793		if (!scan_swap_map_try_ssd_cluster(si, &offset, &scan_base))
 
 
 794			goto scan;
 795	} else if (unlikely(!si->cluster_nr--)) {
 
 
 796		if (si->pages - si->inuse_pages < SWAPFILE_CLUSTER) {
 797			si->cluster_nr = SWAPFILE_CLUSTER - 1;
 798			goto checks;
 799		}
 800
 801		spin_unlock(&si->lock);
 802
 803		/*
 804		 * If seek is expensive, start searching for new cluster from
 805		 * start of partition, to minimize the span of allocated swap.
 806		 * If seek is cheap, that is the SWP_SOLIDSTATE si->cluster_info
 807		 * case, just handled by scan_swap_map_try_ssd_cluster() above.
 808		 */
 809		scan_base = offset = si->lowest_bit;
 810		last_in_cluster = offset + SWAPFILE_CLUSTER - 1;
 811
 812		/* Locate the first empty (unaligned) cluster */
 813		for (; last_in_cluster <= si->highest_bit; offset++) {
 814			if (si->swap_map[offset])
 815				last_in_cluster = offset + SWAPFILE_CLUSTER;
 816			else if (offset == last_in_cluster) {
 817				spin_lock(&si->lock);
 818				offset -= SWAPFILE_CLUSTER - 1;
 819				si->cluster_next = offset;
 820				si->cluster_nr = SWAPFILE_CLUSTER - 1;
 821				goto checks;
 822			}
 823			if (unlikely(--latency_ration < 0)) {
 824				cond_resched();
 825				latency_ration = LATENCY_LIMIT;
 826			}
 827		}
 828
 829		offset = scan_base;
 830		spin_lock(&si->lock);
 831		si->cluster_nr = SWAPFILE_CLUSTER - 1;
 832	}
 833
 834checks:
 835	if (si->cluster_info) {
 836		while (scan_swap_map_ssd_cluster_conflict(si, offset)) {
 837		/* take a break if we already got some slots */
 838			if (n_ret)
 839				goto done;
 840			if (!scan_swap_map_try_ssd_cluster(si, &offset,
 841							&scan_base))
 842				goto scan;
 843		}
 844	}
 845	if (!(si->flags & SWP_WRITEOK))
 846		goto no_page;
 847	if (!si->highest_bit)
 848		goto no_page;
 849	if (offset > si->highest_bit)
 850		scan_base = offset = si->lowest_bit;
 851
 852	ci = lock_cluster(si, offset);
 853	/* reuse swap entry of cache-only swap if not busy. */
 854	if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) {
 855		int swap_was_freed;
 856		unlock_cluster(ci);
 857		spin_unlock(&si->lock);
 858		swap_was_freed = __try_to_reclaim_swap(si, offset, TTRS_ANYWAY);
 859		spin_lock(&si->lock);
 860		/* entry was freed successfully, try to use this again */
 861		if (swap_was_freed)
 862			goto checks;
 863		goto scan; /* check next one */
 864	}
 865
 866	if (si->swap_map[offset]) {
 867		unlock_cluster(ci);
 868		if (!n_ret)
 869			goto scan;
 870		else
 871			goto done;
 872	}
 873	WRITE_ONCE(si->swap_map[offset], usage);
 874	inc_cluster_info_page(si, si->cluster_info, offset);
 875	unlock_cluster(ci);
 876
 877	swap_range_alloc(si, offset, 1);
 
 878	slots[n_ret++] = swp_entry(si->type, offset);
 879
 880	/* got enough slots or reach max slots? */
 881	if ((n_ret == nr) || (offset >= si->highest_bit))
 882		goto done;
 883
 884	/* search for next available slot */
 885
 886	/* time to take a break? */
 887	if (unlikely(--latency_ration < 0)) {
 888		if (n_ret)
 889			goto done;
 890		spin_unlock(&si->lock);
 891		cond_resched();
 892		spin_lock(&si->lock);
 893		latency_ration = LATENCY_LIMIT;
 894	}
 895
 896	/* try to get more slots in cluster */
 897	if (si->cluster_info) {
 898		if (scan_swap_map_try_ssd_cluster(si, &offset, &scan_base))
 899			goto checks;
 900	} else if (si->cluster_nr && !si->swap_map[++offset]) {
 901		/* non-ssd case, still more slots in cluster? */
 
 
 
 
 
 
 902		--si->cluster_nr;
 903		goto checks;
 904	}
 905
 906	/*
 907	 * Even if there's no free clusters available (fragmented),
 908	 * try to scan a little more quickly with lock held unless we
 909	 * have scanned too many slots already.
 910	 */
 911	if (!scanned_many) {
 912		unsigned long scan_limit;
 913
 914		if (offset < scan_base)
 915			scan_limit = scan_base;
 916		else
 917			scan_limit = si->highest_bit;
 918		for (; offset <= scan_limit && --latency_ration > 0;
 919		     offset++) {
 920			if (!si->swap_map[offset])
 921				goto checks;
 922		}
 923	}
 924
 925done:
 926	set_cluster_next(si, offset + 1);
 927	si->flags -= SWP_SCANNING;
 928	return n_ret;
 929
 930scan:
 931	spin_unlock(&si->lock);
 932	while (++offset <= READ_ONCE(si->highest_bit)) {
 933		if (data_race(!si->swap_map[offset])) {
 934			spin_lock(&si->lock);
 935			goto checks;
 936		}
 937		if (vm_swap_full() &&
 938		    READ_ONCE(si->swap_map[offset]) == SWAP_HAS_CACHE) {
 939			spin_lock(&si->lock);
 940			goto checks;
 941		}
 942		if (unlikely(--latency_ration < 0)) {
 943			cond_resched();
 944			latency_ration = LATENCY_LIMIT;
 945			scanned_many = true;
 946		}
 947	}
 948	offset = si->lowest_bit;
 949	while (offset < scan_base) {
 950		if (data_race(!si->swap_map[offset])) {
 951			spin_lock(&si->lock);
 952			goto checks;
 953		}
 954		if (vm_swap_full() &&
 955		    READ_ONCE(si->swap_map[offset]) == SWAP_HAS_CACHE) {
 956			spin_lock(&si->lock);
 957			goto checks;
 958		}
 959		if (unlikely(--latency_ration < 0)) {
 960			cond_resched();
 961			latency_ration = LATENCY_LIMIT;
 962			scanned_many = true;
 963		}
 964		offset++;
 965	}
 966	spin_lock(&si->lock);
 967
 968no_page:
 969	si->flags -= SWP_SCANNING;
 970	return n_ret;
 971}
 972
 973static int swap_alloc_cluster(struct swap_info_struct *si, swp_entry_t *slot)
 974{
 975	unsigned long idx;
 976	struct swap_cluster_info *ci;
 977	unsigned long offset, i;
 978	unsigned char *map;
 979
 980	/*
 981	 * Should not even be attempting cluster allocations when huge
 982	 * page swap is disabled.  Warn and fail the allocation.
 983	 */
 984	if (!IS_ENABLED(CONFIG_THP_SWAP)) {
 985		VM_WARN_ON_ONCE(1);
 986		return 0;
 987	}
 988
 989	if (cluster_list_empty(&si->free_clusters))
 990		return 0;
 991
 992	idx = cluster_list_first(&si->free_clusters);
 993	offset = idx * SWAPFILE_CLUSTER;
 994	ci = lock_cluster(si, offset);
 995	alloc_cluster(si, idx);
 996	cluster_set_count_flag(ci, SWAPFILE_CLUSTER, CLUSTER_FLAG_HUGE);
 997
 998	map = si->swap_map + offset;
 999	for (i = 0; i < SWAPFILE_CLUSTER; i++)
1000		map[i] = SWAP_HAS_CACHE;
1001	unlock_cluster(ci);
1002	swap_range_alloc(si, offset, SWAPFILE_CLUSTER);
1003	*slot = swp_entry(si->type, offset);
1004
1005	return 1;
1006}
1007
1008static void swap_free_cluster(struct swap_info_struct *si, unsigned long idx)
1009{
1010	unsigned long offset = idx * SWAPFILE_CLUSTER;
1011	struct swap_cluster_info *ci;
1012
1013	ci = lock_cluster(si, offset);
1014	memset(si->swap_map + offset, 0, SWAPFILE_CLUSTER);
1015	cluster_set_count_flag(ci, 0, 0);
1016	free_cluster(si, idx);
1017	unlock_cluster(ci);
1018	swap_range_free(si, offset, SWAPFILE_CLUSTER);
1019}
1020
1021static unsigned long scan_swap_map(struct swap_info_struct *si,
1022				   unsigned char usage)
1023{
1024	swp_entry_t entry;
1025	int n_ret;
1026
1027	n_ret = scan_swap_map_slots(si, usage, 1, &entry);
1028
1029	if (n_ret)
1030		return swp_offset(entry);
1031	else
1032		return 0;
1033
1034}
1035
1036int get_swap_pages(int n_goal, swp_entry_t swp_entries[], int entry_size)
1037{
1038	unsigned long size = swap_entry_size(entry_size);
1039	struct swap_info_struct *si, *next;
1040	long avail_pgs;
1041	int n_ret = 0;
1042	int node;
1043
1044	/* Only single cluster request supported */
1045	WARN_ON_ONCE(n_goal > 1 && size == SWAPFILE_CLUSTER);
1046
1047	avail_pgs = atomic_long_read(&nr_swap_pages) / size;
1048	if (avail_pgs <= 0)
1049		goto noswap;
1050
1051	n_goal = min3((long)n_goal, (long)SWAP_BATCH, avail_pgs);
 
 
 
 
1052
1053	atomic_long_sub(n_goal * size, &nr_swap_pages);
1054
1055	spin_lock(&swap_avail_lock);
1056
1057start_over:
1058	node = numa_node_id();
1059	plist_for_each_entry_safe(si, next, &swap_avail_heads[node], avail_lists[node]) {
1060		/* requeue si to after same-priority siblings */
1061		plist_requeue(&si->avail_lists[node], &swap_avail_heads[node]);
1062		spin_unlock(&swap_avail_lock);
1063		spin_lock(&si->lock);
1064		if (!si->highest_bit || !(si->flags & SWP_WRITEOK)) {
1065			spin_lock(&swap_avail_lock);
1066			if (plist_node_empty(&si->avail_lists[node])) {
1067				spin_unlock(&si->lock);
1068				goto nextsi;
1069			}
1070			WARN(!si->highest_bit,
1071			     "swap_info %d in list but !highest_bit\n",
1072			     si->type);
1073			WARN(!(si->flags & SWP_WRITEOK),
1074			     "swap_info %d in list but !SWP_WRITEOK\n",
1075			     si->type);
1076			__del_from_avail_list(si);
1077			spin_unlock(&si->lock);
1078			goto nextsi;
1079		}
1080		if (size == SWAPFILE_CLUSTER) {
1081			if (si->flags & SWP_BLKDEV)
1082				n_ret = swap_alloc_cluster(si, swp_entries);
1083		} else
1084			n_ret = scan_swap_map_slots(si, SWAP_HAS_CACHE,
1085						    n_goal, swp_entries);
1086		spin_unlock(&si->lock);
1087		if (n_ret || size == SWAPFILE_CLUSTER)
1088			goto check_out;
1089		pr_debug("scan_swap_map of si %d failed to find offset\n",
1090			si->type);
1091
1092		spin_lock(&swap_avail_lock);
1093nextsi:
1094		/*
1095		 * if we got here, it's likely that si was almost full before,
1096		 * and since scan_swap_map() can drop the si->lock, multiple
1097		 * callers probably all tried to get a page from the same si
1098		 * and it filled up before we could get one; or, the si filled
1099		 * up between us dropping swap_avail_lock and taking si->lock.
1100		 * Since we dropped the swap_avail_lock, the swap_avail_head
1101		 * list may have been modified; so if next is still in the
1102		 * swap_avail_head list then try it, otherwise start over
1103		 * if we have not gotten any slots.
1104		 */
1105		if (plist_node_empty(&next->avail_lists[node]))
1106			goto start_over;
1107	}
1108
1109	spin_unlock(&swap_avail_lock);
1110
1111check_out:
1112	if (n_ret < n_goal)
1113		atomic_long_add((long)(n_goal - n_ret) * size,
1114				&nr_swap_pages);
1115noswap:
1116	return n_ret;
1117}
1118
1119/* The only caller of this function is now suspend routine */
1120swp_entry_t get_swap_page_of_type(int type)
1121{
1122	struct swap_info_struct *si = swap_type_to_swap_info(type);
1123	pgoff_t offset;
1124
1125	if (!si)
1126		goto fail;
1127
1128	spin_lock(&si->lock);
1129	if (si->flags & SWP_WRITEOK) {
1130		atomic_long_dec(&nr_swap_pages);
1131		/* This is called for allocating swap entry, not cache */
1132		offset = scan_swap_map(si, 1);
1133		if (offset) {
1134			spin_unlock(&si->lock);
1135			return swp_entry(type, offset);
1136		}
1137		atomic_long_inc(&nr_swap_pages);
1138	}
1139	spin_unlock(&si->lock);
1140fail:
1141	return (swp_entry_t) {0};
1142}
1143
1144static struct swap_info_struct *__swap_info_get(swp_entry_t entry)
1145{
1146	struct swap_info_struct *p;
1147	unsigned long offset;
1148
1149	if (!entry.val)
1150		goto out;
1151	p = swp_swap_info(entry);
1152	if (!p)
1153		goto bad_nofile;
1154	if (data_race(!(p->flags & SWP_USED)))
1155		goto bad_device;
1156	offset = swp_offset(entry);
1157	if (offset >= p->max)
1158		goto bad_offset;
1159	return p;
1160
1161bad_offset:
1162	pr_err("swap_info_get: %s%08lx\n", Bad_offset, entry.val);
1163	goto out;
1164bad_device:
1165	pr_err("swap_info_get: %s%08lx\n", Unused_file, entry.val);
1166	goto out;
1167bad_nofile:
1168	pr_err("swap_info_get: %s%08lx\n", Bad_file, entry.val);
1169out:
1170	return NULL;
1171}
1172
1173static struct swap_info_struct *_swap_info_get(swp_entry_t entry)
1174{
1175	struct swap_info_struct *p;
1176
1177	p = __swap_info_get(entry);
1178	if (!p)
1179		goto out;
1180	if (data_race(!p->swap_map[swp_offset(entry)]))
1181		goto bad_free;
1182	return p;
1183
1184bad_free:
1185	pr_err("swap_info_get: %s%08lx\n", Unused_offset, entry.val);
1186	goto out;
1187out:
1188	return NULL;
1189}
1190
1191static struct swap_info_struct *swap_info_get(swp_entry_t entry)
1192{
1193	struct swap_info_struct *p;
1194
1195	p = _swap_info_get(entry);
1196	if (p)
1197		spin_lock(&p->lock);
1198	return p;
1199}
1200
1201static struct swap_info_struct *swap_info_get_cont(swp_entry_t entry,
1202					struct swap_info_struct *q)
1203{
1204	struct swap_info_struct *p;
1205
1206	p = _swap_info_get(entry);
1207
1208	if (p != q) {
1209		if (q != NULL)
1210			spin_unlock(&q->lock);
1211		if (p != NULL)
1212			spin_lock(&p->lock);
1213	}
1214	return p;
1215}
1216
1217static unsigned char __swap_entry_free_locked(struct swap_info_struct *p,
1218					      unsigned long offset,
1219					      unsigned char usage)
1220{
1221	unsigned char count;
1222	unsigned char has_cache;
1223
1224	count = p->swap_map[offset];
1225
1226	has_cache = count & SWAP_HAS_CACHE;
1227	count &= ~SWAP_HAS_CACHE;
1228
1229	if (usage == SWAP_HAS_CACHE) {
1230		VM_BUG_ON(!has_cache);
1231		has_cache = 0;
1232	} else if (count == SWAP_MAP_SHMEM) {
1233		/*
1234		 * Or we could insist on shmem.c using a special
1235		 * swap_shmem_free() and free_shmem_swap_and_cache()...
1236		 */
1237		count = 0;
1238	} else if ((count & ~COUNT_CONTINUED) <= SWAP_MAP_MAX) {
1239		if (count == COUNT_CONTINUED) {
1240			if (swap_count_continued(p, offset, count))
1241				count = SWAP_MAP_MAX | COUNT_CONTINUED;
1242			else
1243				count = SWAP_MAP_MAX;
1244		} else
1245			count--;
1246	}
1247
1248	usage = count | has_cache;
1249	if (usage)
1250		WRITE_ONCE(p->swap_map[offset], usage);
1251	else
1252		WRITE_ONCE(p->swap_map[offset], SWAP_HAS_CACHE);
1253
1254	return usage;
1255}
1256
1257/*
1258 * Check whether swap entry is valid in the swap device.  If so,
1259 * return pointer to swap_info_struct, and keep the swap entry valid
1260 * via preventing the swap device from being swapoff, until
1261 * put_swap_device() is called.  Otherwise return NULL.
1262 *
1263 * The entirety of the RCU read critical section must come before the
1264 * return from or after the call to synchronize_rcu() in
1265 * enable_swap_info() or swapoff().  So if "si->flags & SWP_VALID" is
1266 * true, the si->map, si->cluster_info, etc. must be valid in the
1267 * critical section.
1268 *
1269 * Notice that swapoff or swapoff+swapon can still happen before the
1270 * rcu_read_lock() in get_swap_device() or after the rcu_read_unlock()
1271 * in put_swap_device() if there isn't any other way to prevent
1272 * swapoff, such as page lock, page table lock, etc.  The caller must
1273 * be prepared for that.  For example, the following situation is
1274 * possible.
1275 *
1276 *   CPU1				CPU2
1277 *   do_swap_page()
1278 *     ...				swapoff+swapon
1279 *     __read_swap_cache_async()
1280 *       swapcache_prepare()
1281 *         __swap_duplicate()
1282 *           // check swap_map
1283 *     // verify PTE not changed
1284 *
1285 * In __swap_duplicate(), the swap_map need to be checked before
1286 * changing partly because the specified swap entry may be for another
1287 * swap device which has been swapoff.  And in do_swap_page(), after
1288 * the page is read from the swap device, the PTE is verified not
1289 * changed with the page table locked to check whether the swap device
1290 * has been swapoff or swapoff+swapon.
1291 */
1292struct swap_info_struct *get_swap_device(swp_entry_t entry)
1293{
1294	struct swap_info_struct *si;
1295	unsigned long offset;
1296
1297	if (!entry.val)
1298		goto out;
1299	si = swp_swap_info(entry);
1300	if (!si)
1301		goto bad_nofile;
1302
1303	rcu_read_lock();
1304	if (data_race(!(si->flags & SWP_VALID)))
1305		goto unlock_out;
1306	offset = swp_offset(entry);
1307	if (offset >= si->max)
1308		goto unlock_out;
1309
1310	return si;
1311bad_nofile:
1312	pr_err("%s: %s%08lx\n", __func__, Bad_file, entry.val);
1313out:
1314	return NULL;
1315unlock_out:
1316	rcu_read_unlock();
1317	return NULL;
1318}
1319
1320static unsigned char __swap_entry_free(struct swap_info_struct *p,
1321				       swp_entry_t entry)
1322{
1323	struct swap_cluster_info *ci;
1324	unsigned long offset = swp_offset(entry);
1325	unsigned char usage;
1326
1327	ci = lock_cluster_or_swap_info(p, offset);
1328	usage = __swap_entry_free_locked(p, offset, 1);
1329	unlock_cluster_or_swap_info(p, ci);
1330	if (!usage)
1331		free_swap_slot(entry);
1332
1333	return usage;
1334}
1335
1336static void swap_entry_free(struct swap_info_struct *p, swp_entry_t entry)
1337{
1338	struct swap_cluster_info *ci;
1339	unsigned long offset = swp_offset(entry);
1340	unsigned char count;
1341
1342	ci = lock_cluster(p, offset);
1343	count = p->swap_map[offset];
1344	VM_BUG_ON(count != SWAP_HAS_CACHE);
1345	p->swap_map[offset] = 0;
1346	dec_cluster_info_page(p, p->cluster_info, offset);
1347	unlock_cluster(ci);
1348
1349	mem_cgroup_uncharge_swap(entry, 1);
1350	swap_range_free(p, offset, 1);
1351}
1352
1353/*
1354 * Caller has made sure that the swap device corresponding to entry
1355 * is still around or has not been recycled.
1356 */
1357void swap_free(swp_entry_t entry)
1358{
1359	struct swap_info_struct *p;
1360
1361	p = _swap_info_get(entry);
1362	if (p)
1363		__swap_entry_free(p, entry);
1364}
1365
1366/*
1367 * Called after dropping swapcache to decrease refcnt to swap entries.
1368 */
1369void put_swap_page(struct page *page, swp_entry_t entry)
1370{
1371	unsigned long offset = swp_offset(entry);
1372	unsigned long idx = offset / SWAPFILE_CLUSTER;
1373	struct swap_cluster_info *ci;
1374	struct swap_info_struct *si;
1375	unsigned char *map;
1376	unsigned int i, free_entries = 0;
1377	unsigned char val;
1378	int size = swap_entry_size(thp_nr_pages(page));
1379
1380	si = _swap_info_get(entry);
1381	if (!si)
1382		return;
1383
1384	ci = lock_cluster_or_swap_info(si, offset);
1385	if (size == SWAPFILE_CLUSTER) {
1386		VM_BUG_ON(!cluster_is_huge(ci));
1387		map = si->swap_map + offset;
1388		for (i = 0; i < SWAPFILE_CLUSTER; i++) {
1389			val = map[i];
1390			VM_BUG_ON(!(val & SWAP_HAS_CACHE));
1391			if (val == SWAP_HAS_CACHE)
1392				free_entries++;
1393		}
1394		cluster_clear_huge(ci);
1395		if (free_entries == SWAPFILE_CLUSTER) {
1396			unlock_cluster_or_swap_info(si, ci);
1397			spin_lock(&si->lock);
1398			mem_cgroup_uncharge_swap(entry, SWAPFILE_CLUSTER);
1399			swap_free_cluster(si, idx);
1400			spin_unlock(&si->lock);
1401			return;
1402		}
1403	}
1404	for (i = 0; i < size; i++, entry.val++) {
1405		if (!__swap_entry_free_locked(si, offset + i, SWAP_HAS_CACHE)) {
1406			unlock_cluster_or_swap_info(si, ci);
1407			free_swap_slot(entry);
1408			if (i == size - 1)
1409				return;
1410			lock_cluster_or_swap_info(si, offset);
1411		}
1412	}
1413	unlock_cluster_or_swap_info(si, ci);
1414}
1415
1416#ifdef CONFIG_THP_SWAP
1417int split_swap_cluster(swp_entry_t entry)
1418{
1419	struct swap_info_struct *si;
1420	struct swap_cluster_info *ci;
1421	unsigned long offset = swp_offset(entry);
1422
1423	si = _swap_info_get(entry);
1424	if (!si)
1425		return -EBUSY;
1426	ci = lock_cluster(si, offset);
1427	cluster_clear_huge(ci);
1428	unlock_cluster(ci);
1429	return 0;
1430}
1431#endif
1432
1433static int swp_entry_cmp(const void *ent1, const void *ent2)
1434{
1435	const swp_entry_t *e1 = ent1, *e2 = ent2;
1436
1437	return (int)swp_type(*e1) - (int)swp_type(*e2);
1438}
1439
1440void swapcache_free_entries(swp_entry_t *entries, int n)
1441{
1442	struct swap_info_struct *p, *prev;
1443	int i;
1444
1445	if (n <= 0)
1446		return;
1447
1448	prev = NULL;
1449	p = NULL;
1450
1451	/*
1452	 * Sort swap entries by swap device, so each lock is only taken once.
1453	 * nr_swapfiles isn't absolutely correct, but the overhead of sort() is
1454	 * so low that it isn't necessary to optimize further.
1455	 */
1456	if (nr_swapfiles > 1)
1457		sort(entries, n, sizeof(entries[0]), swp_entry_cmp, NULL);
1458	for (i = 0; i < n; ++i) {
1459		p = swap_info_get_cont(entries[i], prev);
1460		if (p)
1461			swap_entry_free(p, entries[i]);
1462		prev = p;
1463	}
1464	if (p)
1465		spin_unlock(&p->lock);
1466}
1467
1468/*
1469 * How many references to page are currently swapped out?
1470 * This does not give an exact answer when swap count is continued,
1471 * but does include the high COUNT_CONTINUED flag to allow for that.
1472 */
1473int page_swapcount(struct page *page)
1474{
1475	int count = 0;
1476	struct swap_info_struct *p;
1477	struct swap_cluster_info *ci;
1478	swp_entry_t entry;
1479	unsigned long offset;
1480
1481	entry.val = page_private(page);
1482	p = _swap_info_get(entry);
1483	if (p) {
1484		offset = swp_offset(entry);
1485		ci = lock_cluster_or_swap_info(p, offset);
1486		count = swap_count(p->swap_map[offset]);
1487		unlock_cluster_or_swap_info(p, ci);
1488	}
1489	return count;
1490}
1491
1492int __swap_count(swp_entry_t entry)
1493{
1494	struct swap_info_struct *si;
1495	pgoff_t offset = swp_offset(entry);
1496	int count = 0;
1497
1498	si = get_swap_device(entry);
1499	if (si) {
1500		count = swap_count(si->swap_map[offset]);
1501		put_swap_device(si);
1502	}
1503	return count;
1504}
1505
1506static int swap_swapcount(struct swap_info_struct *si, swp_entry_t entry)
1507{
1508	int count = 0;
1509	pgoff_t offset = swp_offset(entry);
1510	struct swap_cluster_info *ci;
1511
1512	ci = lock_cluster_or_swap_info(si, offset);
1513	count = swap_count(si->swap_map[offset]);
1514	unlock_cluster_or_swap_info(si, ci);
1515	return count;
1516}
1517
1518/*
1519 * How many references to @entry are currently swapped out?
1520 * This does not give an exact answer when swap count is continued,
1521 * but does include the high COUNT_CONTINUED flag to allow for that.
1522 */
1523int __swp_swapcount(swp_entry_t entry)
1524{
1525	int count = 0;
1526	struct swap_info_struct *si;
1527
1528	si = get_swap_device(entry);
1529	if (si) {
1530		count = swap_swapcount(si, entry);
1531		put_swap_device(si);
1532	}
1533	return count;
1534}
1535
1536/*
1537 * How many references to @entry are currently swapped out?
1538 * This considers COUNT_CONTINUED so it returns exact answer.
1539 */
1540int swp_swapcount(swp_entry_t entry)
1541{
1542	int count, tmp_count, n;
1543	struct swap_info_struct *p;
1544	struct swap_cluster_info *ci;
1545	struct page *page;
1546	pgoff_t offset;
1547	unsigned char *map;
1548
1549	p = _swap_info_get(entry);
1550	if (!p)
1551		return 0;
1552
1553	offset = swp_offset(entry);
1554
1555	ci = lock_cluster_or_swap_info(p, offset);
1556
1557	count = swap_count(p->swap_map[offset]);
1558	if (!(count & COUNT_CONTINUED))
1559		goto out;
1560
1561	count &= ~COUNT_CONTINUED;
1562	n = SWAP_MAP_MAX + 1;
1563
1564	page = vmalloc_to_page(p->swap_map + offset);
1565	offset &= ~PAGE_MASK;
1566	VM_BUG_ON(page_private(page) != SWP_CONTINUED);
1567
1568	do {
1569		page = list_next_entry(page, lru);
1570		map = kmap_atomic(page);
1571		tmp_count = map[offset];
1572		kunmap_atomic(map);
1573
1574		count += (tmp_count & ~COUNT_CONTINUED) * n;
1575		n *= (SWAP_CONT_MAX + 1);
1576	} while (tmp_count & COUNT_CONTINUED);
1577out:
1578	unlock_cluster_or_swap_info(p, ci);
1579	return count;
1580}
1581
1582static bool swap_page_trans_huge_swapped(struct swap_info_struct *si,
1583					 swp_entry_t entry)
1584{
1585	struct swap_cluster_info *ci;
1586	unsigned char *map = si->swap_map;
1587	unsigned long roffset = swp_offset(entry);
1588	unsigned long offset = round_down(roffset, SWAPFILE_CLUSTER);
1589	int i;
1590	bool ret = false;
1591
1592	ci = lock_cluster_or_swap_info(si, offset);
1593	if (!ci || !cluster_is_huge(ci)) {
1594		if (swap_count(map[roffset]))
1595			ret = true;
1596		goto unlock_out;
1597	}
1598	for (i = 0; i < SWAPFILE_CLUSTER; i++) {
1599		if (swap_count(map[offset + i])) {
1600			ret = true;
1601			break;
1602		}
1603	}
1604unlock_out:
1605	unlock_cluster_or_swap_info(si, ci);
1606	return ret;
1607}
1608
1609static bool page_swapped(struct page *page)
1610{
1611	swp_entry_t entry;
1612	struct swap_info_struct *si;
1613
1614	if (!IS_ENABLED(CONFIG_THP_SWAP) || likely(!PageTransCompound(page)))
1615		return page_swapcount(page) != 0;
1616
1617	page = compound_head(page);
1618	entry.val = page_private(page);
1619	si = _swap_info_get(entry);
1620	if (si)
1621		return swap_page_trans_huge_swapped(si, entry);
1622	return false;
1623}
1624
1625static int page_trans_huge_map_swapcount(struct page *page, int *total_mapcount,
1626					 int *total_swapcount)
1627{
1628	int i, map_swapcount, _total_mapcount, _total_swapcount;
1629	unsigned long offset = 0;
1630	struct swap_info_struct *si;
1631	struct swap_cluster_info *ci = NULL;
1632	unsigned char *map = NULL;
1633	int mapcount, swapcount = 0;
1634
1635	/* hugetlbfs shouldn't call it */
1636	VM_BUG_ON_PAGE(PageHuge(page), page);
1637
1638	if (!IS_ENABLED(CONFIG_THP_SWAP) || likely(!PageTransCompound(page))) {
1639		mapcount = page_trans_huge_mapcount(page, total_mapcount);
1640		if (PageSwapCache(page))
1641			swapcount = page_swapcount(page);
1642		if (total_swapcount)
1643			*total_swapcount = swapcount;
1644		return mapcount + swapcount;
1645	}
1646
1647	page = compound_head(page);
1648
1649	_total_mapcount = _total_swapcount = map_swapcount = 0;
1650	if (PageSwapCache(page)) {
1651		swp_entry_t entry;
1652
1653		entry.val = page_private(page);
1654		si = _swap_info_get(entry);
1655		if (si) {
1656			map = si->swap_map;
1657			offset = swp_offset(entry);
1658		}
1659	}
1660	if (map)
1661		ci = lock_cluster(si, offset);
1662	for (i = 0; i < HPAGE_PMD_NR; i++) {
1663		mapcount = atomic_read(&page[i]._mapcount) + 1;
1664		_total_mapcount += mapcount;
1665		if (map) {
1666			swapcount = swap_count(map[offset + i]);
1667			_total_swapcount += swapcount;
1668		}
1669		map_swapcount = max(map_swapcount, mapcount + swapcount);
1670	}
1671	unlock_cluster(ci);
1672	if (PageDoubleMap(page)) {
1673		map_swapcount -= 1;
1674		_total_mapcount -= HPAGE_PMD_NR;
1675	}
1676	mapcount = compound_mapcount(page);
1677	map_swapcount += mapcount;
1678	_total_mapcount += mapcount;
1679	if (total_mapcount)
1680		*total_mapcount = _total_mapcount;
1681	if (total_swapcount)
1682		*total_swapcount = _total_swapcount;
1683
1684	return map_swapcount;
1685}
1686
1687/*
1688 * We can write to an anon page without COW if there are no other references
1689 * to it.  And as a side-effect, free up its swap: because the old content
1690 * on disk will never be read, and seeking back there to write new content
1691 * later would only waste time away from clustering.
1692 *
1693 * NOTE: total_map_swapcount should not be relied upon by the caller if
1694 * reuse_swap_page() returns false, but it may be always overwritten
1695 * (see the other implementation for CONFIG_SWAP=n).
1696 */
1697bool reuse_swap_page(struct page *page, int *total_map_swapcount)
1698{
1699	int count, total_mapcount, total_swapcount;
1700
1701	VM_BUG_ON_PAGE(!PageLocked(page), page);
1702	if (unlikely(PageKsm(page)))
1703		return false;
1704	count = page_trans_huge_map_swapcount(page, &total_mapcount,
1705					      &total_swapcount);
1706	if (total_map_swapcount)
1707		*total_map_swapcount = total_mapcount + total_swapcount;
1708	if (count == 1 && PageSwapCache(page) &&
1709	    (likely(!PageTransCompound(page)) ||
1710	     /* The remaining swap count will be freed soon */
1711	     total_swapcount == page_swapcount(page))) {
1712		if (!PageWriteback(page)) {
1713			page = compound_head(page);
1714			delete_from_swap_cache(page);
1715			SetPageDirty(page);
1716		} else {
1717			swp_entry_t entry;
1718			struct swap_info_struct *p;
1719
1720			entry.val = page_private(page);
1721			p = swap_info_get(entry);
1722			if (p->flags & SWP_STABLE_WRITES) {
1723				spin_unlock(&p->lock);
1724				return false;
1725			}
1726			spin_unlock(&p->lock);
1727		}
1728	}
1729
1730	return count <= 1;
1731}
1732
1733/*
1734 * If swap is getting full, or if there are no more mappings of this page,
1735 * then try_to_free_swap is called to free its swap space.
1736 */
1737int try_to_free_swap(struct page *page)
1738{
1739	VM_BUG_ON_PAGE(!PageLocked(page), page);
1740
1741	if (!PageSwapCache(page))
1742		return 0;
1743	if (PageWriteback(page))
1744		return 0;
1745	if (page_swapped(page))
1746		return 0;
1747
1748	/*
1749	 * Once hibernation has begun to create its image of memory,
1750	 * there's a danger that one of the calls to try_to_free_swap()
1751	 * - most probably a call from __try_to_reclaim_swap() while
1752	 * hibernation is allocating its own swap pages for the image,
1753	 * but conceivably even a call from memory reclaim - will free
1754	 * the swap from a page which has already been recorded in the
1755	 * image as a clean swapcache page, and then reuse its swap for
1756	 * another page of the image.  On waking from hibernation, the
1757	 * original page might be freed under memory pressure, then
1758	 * later read back in from swap, now with the wrong data.
1759	 *
1760	 * Hibernation suspends storage while it is writing the image
1761	 * to disk so check that here.
1762	 */
1763	if (pm_suspended_storage())
1764		return 0;
1765
1766	page = compound_head(page);
1767	delete_from_swap_cache(page);
1768	SetPageDirty(page);
1769	return 1;
1770}
1771
1772/*
1773 * Free the swap entry like above, but also try to
1774 * free the page cache entry if it is the last user.
1775 */
1776int free_swap_and_cache(swp_entry_t entry)
1777{
1778	struct swap_info_struct *p;
1779	unsigned char count;
1780
1781	if (non_swap_entry(entry))
1782		return 1;
1783
1784	p = _swap_info_get(entry);
1785	if (p) {
1786		count = __swap_entry_free(p, entry);
1787		if (count == SWAP_HAS_CACHE &&
1788		    !swap_page_trans_huge_swapped(p, entry))
1789			__try_to_reclaim_swap(p, swp_offset(entry),
1790					      TTRS_UNMAPPED | TTRS_FULL);
1791	}
1792	return p != NULL;
1793}
1794
1795#ifdef CONFIG_HIBERNATION
1796/*
1797 * Find the swap type that corresponds to given device (if any).
1798 *
1799 * @offset - number of the PAGE_SIZE-sized block of the device, starting
1800 * from 0, in which the swap header is expected to be located.
1801 *
1802 * This is needed for the suspend to disk (aka swsusp).
1803 */
1804int swap_type_of(dev_t device, sector_t offset, struct block_device **bdev_p)
1805{
1806	struct block_device *bdev = NULL;
1807	int type;
1808
1809	if (device)
1810		bdev = bdget(device);
1811
1812	spin_lock(&swap_lock);
1813	for (type = 0; type < nr_swapfiles; type++) {
1814		struct swap_info_struct *sis = swap_info[type];
1815
1816		if (!(sis->flags & SWP_WRITEOK))
1817			continue;
1818
1819		if (!bdev) {
1820			if (bdev_p)
1821				*bdev_p = bdgrab(sis->bdev);
1822
1823			spin_unlock(&swap_lock);
1824			return type;
1825		}
1826		if (bdev == sis->bdev) {
1827			struct swap_extent *se = first_se(sis);
1828
1829			if (se->start_block == offset) {
1830				if (bdev_p)
1831					*bdev_p = bdgrab(sis->bdev);
1832
1833				spin_unlock(&swap_lock);
1834				bdput(bdev);
1835				return type;
1836			}
1837		}
1838	}
1839	spin_unlock(&swap_lock);
1840	if (bdev)
1841		bdput(bdev);
1842
1843	return -ENODEV;
1844}
1845
1846/*
1847 * Get the (PAGE_SIZE) block corresponding to given offset on the swapdev
1848 * corresponding to given index in swap_info (swap type).
1849 */
1850sector_t swapdev_block(int type, pgoff_t offset)
1851{
1852	struct block_device *bdev;
1853	struct swap_info_struct *si = swap_type_to_swap_info(type);
1854
1855	if (!si || !(si->flags & SWP_WRITEOK))
1856		return 0;
1857	return map_swap_entry(swp_entry(type, offset), &bdev);
1858}
1859
1860/*
1861 * Return either the total number of swap pages of given type, or the number
1862 * of free pages of that type (depending on @free)
1863 *
1864 * This is needed for software suspend
1865 */
1866unsigned int count_swap_pages(int type, int free)
1867{
1868	unsigned int n = 0;
1869
1870	spin_lock(&swap_lock);
1871	if ((unsigned int)type < nr_swapfiles) {
1872		struct swap_info_struct *sis = swap_info[type];
1873
1874		spin_lock(&sis->lock);
1875		if (sis->flags & SWP_WRITEOK) {
1876			n = sis->pages;
1877			if (free)
1878				n -= sis->inuse_pages;
1879		}
1880		spin_unlock(&sis->lock);
1881	}
1882	spin_unlock(&swap_lock);
1883	return n;
1884}
1885#endif /* CONFIG_HIBERNATION */
1886
1887static inline int pte_same_as_swp(pte_t pte, pte_t swp_pte)
1888{
1889	return pte_same(pte_swp_clear_soft_dirty(pte), swp_pte);
1890}
1891
1892/*
1893 * No need to decide whether this PTE shares the swap entry with others,
1894 * just let do_wp_page work it out if a write is requested later - to
1895 * force COW, vm_page_prot omits write permission from any private vma.
1896 */
1897static int unuse_pte(struct vm_area_struct *vma, pmd_t *pmd,
1898		unsigned long addr, swp_entry_t entry, struct page *page)
1899{
1900	struct page *swapcache;
 
1901	spinlock_t *ptl;
1902	pte_t *pte;
1903	int ret = 1;
1904
1905	swapcache = page;
1906	page = ksm_might_need_to_copy(page, vma, addr);
1907	if (unlikely(!page))
1908		return -ENOMEM;
1909
 
 
 
 
 
 
1910	pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
1911	if (unlikely(!pte_same_as_swp(*pte, swp_entry_to_pte(entry)))) {
 
1912		ret = 0;
1913		goto out;
1914	}
1915
1916	dec_mm_counter(vma->vm_mm, MM_SWAPENTS);
1917	inc_mm_counter(vma->vm_mm, MM_ANONPAGES);
1918	get_page(page);
1919	set_pte_at(vma->vm_mm, addr, pte,
1920		   pte_mkold(mk_pte(page, vma->vm_page_prot)));
1921	if (page == swapcache) {
1922		page_add_anon_rmap(page, vma, addr, false);
 
1923	} else { /* ksm created a completely new copy */
1924		page_add_new_anon_rmap(page, vma, addr, false);
1925		lru_cache_add_inactive_or_unevictable(page, vma);
 
1926	}
1927	swap_free(entry);
1928	/*
1929	 * Move the page to the active list so it is not
1930	 * immediately swapped out again after swapon.
1931	 */
1932	activate_page(page);
1933out:
1934	pte_unmap_unlock(pte, ptl);
 
1935	if (page != swapcache) {
1936		unlock_page(page);
1937		put_page(page);
1938	}
1939	return ret;
1940}
1941
1942static int unuse_pte_range(struct vm_area_struct *vma, pmd_t *pmd,
1943			unsigned long addr, unsigned long end,
1944			unsigned int type, bool frontswap,
1945			unsigned long *fs_pages_to_unuse)
1946{
1947	struct page *page;
1948	swp_entry_t entry;
1949	pte_t *pte;
1950	struct swap_info_struct *si;
1951	unsigned long offset;
1952	int ret = 0;
1953	volatile unsigned char *swap_map;
1954
1955	si = swap_info[type];
1956	pte = pte_offset_map(pmd, addr);
1957	do {
1958		struct vm_fault vmf;
1959
1960		if (!is_swap_pte(*pte))
1961			continue;
1962
1963		entry = pte_to_swp_entry(*pte);
1964		if (swp_type(entry) != type)
1965			continue;
1966
1967		offset = swp_offset(entry);
1968		if (frontswap && !frontswap_test(si, offset))
1969			continue;
1970
1971		pte_unmap(pte);
1972		swap_map = &si->swap_map[offset];
1973		page = lookup_swap_cache(entry, vma, addr);
1974		if (!page) {
1975			vmf.vma = vma;
1976			vmf.address = addr;
1977			vmf.pmd = pmd;
1978			page = swapin_readahead(entry, GFP_HIGHUSER_MOVABLE,
1979						&vmf);
1980		}
1981		if (!page) {
1982			if (*swap_map == 0 || *swap_map == SWAP_MAP_BAD)
1983				goto try_next;
1984			return -ENOMEM;
1985		}
1986
1987		lock_page(page);
1988		wait_on_page_writeback(page);
1989		ret = unuse_pte(vma, pmd, addr, entry, page);
1990		if (ret < 0) {
1991			unlock_page(page);
1992			put_page(page);
1993			goto out;
1994		}
1995
1996		try_to_free_swap(page);
1997		unlock_page(page);
1998		put_page(page);
1999
2000		if (*fs_pages_to_unuse && !--(*fs_pages_to_unuse)) {
2001			ret = FRONTSWAP_PAGES_UNUSED;
2002			goto out;
2003		}
2004try_next:
2005		pte = pte_offset_map(pmd, addr);
2006	} while (pte++, addr += PAGE_SIZE, addr != end);
2007	pte_unmap(pte - 1);
2008
2009	ret = 0;
2010out:
2011	return ret;
2012}
2013
2014static inline int unuse_pmd_range(struct vm_area_struct *vma, pud_t *pud,
2015				unsigned long addr, unsigned long end,
2016				unsigned int type, bool frontswap,
2017				unsigned long *fs_pages_to_unuse)
2018{
2019	pmd_t *pmd;
2020	unsigned long next;
2021	int ret;
2022
2023	pmd = pmd_offset(pud, addr);
2024	do {
2025		cond_resched();
2026		next = pmd_addr_end(addr, end);
2027		if (pmd_none_or_trans_huge_or_clear_bad(pmd))
2028			continue;
2029		ret = unuse_pte_range(vma, pmd, addr, next, type,
2030				      frontswap, fs_pages_to_unuse);
2031		if (ret)
2032			return ret;
2033	} while (pmd++, addr = next, addr != end);
2034	return 0;
2035}
2036
2037static inline int unuse_pud_range(struct vm_area_struct *vma, p4d_t *p4d,
2038				unsigned long addr, unsigned long end,
2039				unsigned int type, bool frontswap,
2040				unsigned long *fs_pages_to_unuse)
2041{
2042	pud_t *pud;
2043	unsigned long next;
2044	int ret;
2045
2046	pud = pud_offset(p4d, addr);
2047	do {
2048		next = pud_addr_end(addr, end);
2049		if (pud_none_or_clear_bad(pud))
2050			continue;
2051		ret = unuse_pmd_range(vma, pud, addr, next, type,
2052				      frontswap, fs_pages_to_unuse);
2053		if (ret)
2054			return ret;
2055	} while (pud++, addr = next, addr != end);
2056	return 0;
2057}
2058
2059static inline int unuse_p4d_range(struct vm_area_struct *vma, pgd_t *pgd,
2060				unsigned long addr, unsigned long end,
2061				unsigned int type, bool frontswap,
2062				unsigned long *fs_pages_to_unuse)
2063{
2064	p4d_t *p4d;
2065	unsigned long next;
2066	int ret;
2067
2068	p4d = p4d_offset(pgd, addr);
2069	do {
2070		next = p4d_addr_end(addr, end);
2071		if (p4d_none_or_clear_bad(p4d))
2072			continue;
2073		ret = unuse_pud_range(vma, p4d, addr, next, type,
2074				      frontswap, fs_pages_to_unuse);
2075		if (ret)
2076			return ret;
2077	} while (p4d++, addr = next, addr != end);
2078	return 0;
2079}
2080
2081static int unuse_vma(struct vm_area_struct *vma, unsigned int type,
2082		     bool frontswap, unsigned long *fs_pages_to_unuse)
2083{
2084	pgd_t *pgd;
2085	unsigned long addr, end, next;
2086	int ret;
2087
2088	addr = vma->vm_start;
2089	end = vma->vm_end;
2090
2091	pgd = pgd_offset(vma->vm_mm, addr);
2092	do {
2093		next = pgd_addr_end(addr, end);
2094		if (pgd_none_or_clear_bad(pgd))
2095			continue;
2096		ret = unuse_p4d_range(vma, pgd, addr, next, type,
2097				      frontswap, fs_pages_to_unuse);
2098		if (ret)
2099			return ret;
2100	} while (pgd++, addr = next, addr != end);
2101	return 0;
2102}
2103
2104static int unuse_mm(struct mm_struct *mm, unsigned int type,
2105		    bool frontswap, unsigned long *fs_pages_to_unuse)
2106{
2107	struct vm_area_struct *vma;
2108	int ret = 0;
2109
2110	mmap_read_lock(mm);
2111	for (vma = mm->mmap; vma; vma = vma->vm_next) {
2112		if (vma->anon_vma) {
2113			ret = unuse_vma(vma, type, frontswap,
2114					fs_pages_to_unuse);
2115			if (ret)
2116				break;
2117		}
2118		cond_resched();
2119	}
2120	mmap_read_unlock(mm);
2121	return ret;
2122}
2123
2124/*
2125 * Scan swap_map (or frontswap_map if frontswap parameter is true)
2126 * from current position to next entry still in use. Return 0
2127 * if there are no inuse entries after prev till end of the map.
2128 */
2129static unsigned int find_next_to_unuse(struct swap_info_struct *si,
2130					unsigned int prev, bool frontswap)
2131{
2132	unsigned int i;
2133	unsigned char count;
2134
2135	/*
2136	 * No need for swap_lock here: we're just looking
2137	 * for whether an entry is in use, not modifying it; false
2138	 * hits are okay, and sys_swapoff() has already prevented new
2139	 * allocations from this area (while holding swap_lock).
2140	 */
2141	for (i = prev + 1; i < si->max; i++) {
2142		count = READ_ONCE(si->swap_map[i]);
2143		if (count && swap_count(count) != SWAP_MAP_BAD)
2144			if (!frontswap || frontswap_test(si, i))
2145				break;
2146		if ((i % LATENCY_LIMIT) == 0)
2147			cond_resched();
2148	}
2149
2150	if (i == si->max)
2151		i = 0;
2152
2153	return i;
2154}
2155
2156/*
2157 * If the boolean frontswap is true, only unuse pages_to_unuse pages;
2158 * pages_to_unuse==0 means all pages; ignored if frontswap is false
2159 */
2160int try_to_unuse(unsigned int type, bool frontswap,
2161		 unsigned long pages_to_unuse)
2162{
2163	struct mm_struct *prev_mm;
2164	struct mm_struct *mm;
2165	struct list_head *p;
2166	int retval = 0;
2167	struct swap_info_struct *si = swap_info[type];
2168	struct page *page;
2169	swp_entry_t entry;
2170	unsigned int i;
2171
2172	if (!READ_ONCE(si->inuse_pages))
2173		return 0;
2174
2175	if (!frontswap)
2176		pages_to_unuse = 0;
2177
2178retry:
2179	retval = shmem_unuse(type, frontswap, &pages_to_unuse);
2180	if (retval)
2181		goto out;
2182
2183	prev_mm = &init_mm;
2184	mmget(prev_mm);
2185
2186	spin_lock(&mmlist_lock);
2187	p = &init_mm.mmlist;
2188	while (READ_ONCE(si->inuse_pages) &&
2189	       !signal_pending(current) &&
2190	       (p = p->next) != &init_mm.mmlist) {
2191
2192		mm = list_entry(p, struct mm_struct, mmlist);
2193		if (!mmget_not_zero(mm))
2194			continue;
2195		spin_unlock(&mmlist_lock);
2196		mmput(prev_mm);
2197		prev_mm = mm;
2198		retval = unuse_mm(mm, type, frontswap, &pages_to_unuse);
2199
2200		if (retval) {
2201			mmput(prev_mm);
2202			goto out;
2203		}
2204
2205		/*
2206		 * Make sure that we aren't completely killing
2207		 * interactive performance.
2208		 */
2209		cond_resched();
2210		spin_lock(&mmlist_lock);
2211	}
2212	spin_unlock(&mmlist_lock);
2213
2214	mmput(prev_mm);
2215
2216	i = 0;
2217	while (READ_ONCE(si->inuse_pages) &&
2218	       !signal_pending(current) &&
2219	       (i = find_next_to_unuse(si, i, frontswap)) != 0) {
2220
2221		entry = swp_entry(type, i);
2222		page = find_get_page(swap_address_space(entry), i);
2223		if (!page)
2224			continue;
2225
2226		/*
2227		 * It is conceivable that a racing task removed this page from
2228		 * swap cache just before we acquired the page lock. The page
2229		 * might even be back in swap cache on another swap area. But
2230		 * that is okay, try_to_free_swap() only removes stale pages.
2231		 */
2232		lock_page(page);
2233		wait_on_page_writeback(page);
2234		try_to_free_swap(page);
2235		unlock_page(page);
2236		put_page(page);
2237
2238		/*
2239		 * For frontswap, we just need to unuse pages_to_unuse, if
2240		 * it was specified. Need not check frontswap again here as
2241		 * we already zeroed out pages_to_unuse if not frontswap.
2242		 */
2243		if (pages_to_unuse && --pages_to_unuse == 0)
2244			goto out;
2245	}
2246
2247	/*
2248	 * Lets check again to see if there are still swap entries in the map.
2249	 * If yes, we would need to do retry the unuse logic again.
2250	 * Under global memory pressure, swap entries can be reinserted back
2251	 * into process space after the mmlist loop above passes over them.
2252	 *
2253	 * Limit the number of retries? No: when mmget_not_zero() above fails,
2254	 * that mm is likely to be freeing swap from exit_mmap(), which proceeds
2255	 * at its own independent pace; and even shmem_writepage() could have
2256	 * been preempted after get_swap_page(), temporarily hiding that swap.
2257	 * It's easy and robust (though cpu-intensive) just to keep retrying.
2258	 */
2259	if (READ_ONCE(si->inuse_pages)) {
2260		if (!signal_pending(current))
2261			goto retry;
2262		retval = -EINTR;
2263	}
2264out:
2265	return (retval == FRONTSWAP_PAGES_UNUSED) ? 0 : retval;
2266}
2267
2268/*
2269 * After a successful try_to_unuse, if no swap is now in use, we know
2270 * we can empty the mmlist.  swap_lock must be held on entry and exit.
2271 * Note that mmlist_lock nests inside swap_lock, and an mm must be
2272 * added to the mmlist just after page_duplicate - before would be racy.
2273 */
2274static void drain_mmlist(void)
2275{
2276	struct list_head *p, *next;
2277	unsigned int type;
2278
2279	for (type = 0; type < nr_swapfiles; type++)
2280		if (swap_info[type]->inuse_pages)
2281			return;
2282	spin_lock(&mmlist_lock);
2283	list_for_each_safe(p, next, &init_mm.mmlist)
2284		list_del_init(p);
2285	spin_unlock(&mmlist_lock);
2286}
2287
2288/*
2289 * Use this swapdev's extent info to locate the (PAGE_SIZE) block which
2290 * corresponds to page offset for the specified swap entry.
2291 * Note that the type of this function is sector_t, but it returns page offset
2292 * into the bdev, not sector offset.
2293 */
2294static sector_t map_swap_entry(swp_entry_t entry, struct block_device **bdev)
2295{
2296	struct swap_info_struct *sis;
2297	struct swap_extent *se;
2298	pgoff_t offset;
2299
2300	sis = swp_swap_info(entry);
2301	*bdev = sis->bdev;
2302
2303	offset = swp_offset(entry);
2304	se = offset_to_swap_extent(sis, offset);
2305	return se->start_block + (offset - se->start_page);
2306}
2307
2308/*
2309 * Returns the page offset into bdev for the specified page's swap entry.
2310 */
2311sector_t map_swap_page(struct page *page, struct block_device **bdev)
2312{
2313	swp_entry_t entry;
2314	entry.val = page_private(page);
2315	return map_swap_entry(entry, bdev);
2316}
2317
2318/*
2319 * Free all of a swapdev's extent information
2320 */
2321static void destroy_swap_extents(struct swap_info_struct *sis)
2322{
2323	while (!RB_EMPTY_ROOT(&sis->swap_extent_root)) {
2324		struct rb_node *rb = sis->swap_extent_root.rb_node;
2325		struct swap_extent *se = rb_entry(rb, struct swap_extent, rb_node);
2326
2327		rb_erase(rb, &sis->swap_extent_root);
2328		kfree(se);
2329	}
2330
2331	if (sis->flags & SWP_ACTIVATED) {
2332		struct file *swap_file = sis->swap_file;
2333		struct address_space *mapping = swap_file->f_mapping;
2334
2335		sis->flags &= ~SWP_ACTIVATED;
2336		if (mapping->a_ops->swap_deactivate)
2337			mapping->a_ops->swap_deactivate(swap_file);
2338	}
2339}
2340
2341/*
2342 * Add a block range (and the corresponding page range) into this swapdev's
2343 * extent tree.
2344 *
2345 * This function rather assumes that it is called in ascending page order.
2346 */
2347int
2348add_swap_extent(struct swap_info_struct *sis, unsigned long start_page,
2349		unsigned long nr_pages, sector_t start_block)
2350{
2351	struct rb_node **link = &sis->swap_extent_root.rb_node, *parent = NULL;
2352	struct swap_extent *se;
2353	struct swap_extent *new_se;
2354
2355	/*
2356	 * place the new node at the right most since the
2357	 * function is called in ascending page order.
2358	 */
2359	while (*link) {
2360		parent = *link;
2361		link = &parent->rb_right;
2362	}
2363
2364	if (parent) {
2365		se = rb_entry(parent, struct swap_extent, rb_node);
2366		BUG_ON(se->start_page + se->nr_pages != start_page);
2367		if (se->start_block + se->nr_pages == start_block) {
2368			/* Merge it */
2369			se->nr_pages += nr_pages;
2370			return 0;
2371		}
2372	}
2373
2374	/* No merge, insert a new extent. */
2375	new_se = kmalloc(sizeof(*se), GFP_KERNEL);
2376	if (new_se == NULL)
2377		return -ENOMEM;
2378	new_se->start_page = start_page;
2379	new_se->nr_pages = nr_pages;
2380	new_se->start_block = start_block;
2381
2382	rb_link_node(&new_se->rb_node, parent, link);
2383	rb_insert_color(&new_se->rb_node, &sis->swap_extent_root);
2384	return 1;
2385}
2386EXPORT_SYMBOL_GPL(add_swap_extent);
2387
2388/*
2389 * A `swap extent' is a simple thing which maps a contiguous range of pages
2390 * onto a contiguous range of disk blocks.  An ordered list of swap extents
2391 * is built at swapon time and is then used at swap_writepage/swap_readpage
2392 * time for locating where on disk a page belongs.
2393 *
2394 * If the swapfile is an S_ISBLK block device, a single extent is installed.
2395 * This is done so that the main operating code can treat S_ISBLK and S_ISREG
2396 * swap files identically.
2397 *
2398 * Whether the swapdev is an S_ISREG file or an S_ISBLK blockdev, the swap
2399 * extent list operates in PAGE_SIZE disk blocks.  Both S_ISREG and S_ISBLK
2400 * swapfiles are handled *identically* after swapon time.
2401 *
2402 * For S_ISREG swapfiles, setup_swap_extents() will walk all the file's blocks
2403 * and will parse them into an ordered extent list, in PAGE_SIZE chunks.  If
2404 * some stray blocks are found which do not fall within the PAGE_SIZE alignment
2405 * requirements, they are simply tossed out - we will never use those blocks
2406 * for swapping.
2407 *
2408 * For all swap devices we set S_SWAPFILE across the life of the swapon.  This
2409 * prevents users from writing to the swap device, which will corrupt memory.
2410 *
2411 * The amount of disk space which a single swap extent represents varies.
2412 * Typically it is in the 1-4 megabyte range.  So we can have hundreds of
2413 * extents in the list.  To avoid much list walking, we cache the previous
2414 * search location in `curr_swap_extent', and start new searches from there.
2415 * This is extremely effective.  The average number of iterations in
2416 * map_swap_page() has been measured at about 0.3 per page.  - akpm.
2417 */
2418static int setup_swap_extents(struct swap_info_struct *sis, sector_t *span)
2419{
2420	struct file *swap_file = sis->swap_file;
2421	struct address_space *mapping = swap_file->f_mapping;
2422	struct inode *inode = mapping->host;
2423	int ret;
2424
2425	if (S_ISBLK(inode->i_mode)) {
2426		ret = add_swap_extent(sis, 0, sis->max, 0);
2427		*span = sis->pages;
2428		return ret;
2429	}
2430
2431	if (mapping->a_ops->swap_activate) {
2432		ret = mapping->a_ops->swap_activate(sis, swap_file, span);
2433		if (ret >= 0)
2434			sis->flags |= SWP_ACTIVATED;
2435		if (!ret) {
2436			sis->flags |= SWP_FS;
2437			ret = add_swap_extent(sis, 0, sis->max, 0);
2438			*span = sis->pages;
2439		}
2440		return ret;
2441	}
2442
2443	return generic_swapfile_activate(sis, swap_file, span);
2444}
2445
2446static int swap_node(struct swap_info_struct *p)
2447{
2448	struct block_device *bdev;
2449
2450	if (p->bdev)
2451		bdev = p->bdev;
2452	else
2453		bdev = p->swap_file->f_inode->i_sb->s_bdev;
2454
2455	return bdev ? bdev->bd_disk->node_id : NUMA_NO_NODE;
2456}
2457
2458static void setup_swap_info(struct swap_info_struct *p, int prio,
2459			    unsigned char *swap_map,
2460			    struct swap_cluster_info *cluster_info)
2461{
2462	int i;
2463
2464	if (prio >= 0)
2465		p->prio = prio;
2466	else
2467		p->prio = --least_priority;
2468	/*
2469	 * the plist prio is negated because plist ordering is
2470	 * low-to-high, while swap ordering is high-to-low
2471	 */
2472	p->list.prio = -p->prio;
2473	for_each_node(i) {
2474		if (p->prio >= 0)
2475			p->avail_lists[i].prio = -p->prio;
2476		else {
2477			if (swap_node(p) == i)
2478				p->avail_lists[i].prio = 1;
2479			else
2480				p->avail_lists[i].prio = -p->prio;
2481		}
2482	}
2483	p->swap_map = swap_map;
2484	p->cluster_info = cluster_info;
2485}
2486
2487static void _enable_swap_info(struct swap_info_struct *p)
2488{
2489	p->flags |= SWP_WRITEOK | SWP_VALID;
2490	atomic_long_add(p->pages, &nr_swap_pages);
2491	total_swap_pages += p->pages;
2492
2493	assert_spin_locked(&swap_lock);
2494	/*
2495	 * both lists are plists, and thus priority ordered.
2496	 * swap_active_head needs to be priority ordered for swapoff(),
2497	 * which on removal of any swap_info_struct with an auto-assigned
2498	 * (i.e. negative) priority increments the auto-assigned priority
2499	 * of any lower-priority swap_info_structs.
2500	 * swap_avail_head needs to be priority ordered for get_swap_page(),
2501	 * which allocates swap pages from the highest available priority
2502	 * swap_info_struct.
2503	 */
2504	plist_add(&p->list, &swap_active_head);
2505	add_to_avail_list(p);
2506}
2507
2508static void enable_swap_info(struct swap_info_struct *p, int prio,
2509				unsigned char *swap_map,
2510				struct swap_cluster_info *cluster_info,
2511				unsigned long *frontswap_map)
2512{
2513	frontswap_init(p->type, frontswap_map);
2514	spin_lock(&swap_lock);
2515	spin_lock(&p->lock);
2516	setup_swap_info(p, prio, swap_map, cluster_info);
2517	spin_unlock(&p->lock);
2518	spin_unlock(&swap_lock);
2519	/*
2520	 * Guarantee swap_map, cluster_info, etc. fields are valid
2521	 * between get/put_swap_device() if SWP_VALID bit is set
2522	 */
2523	synchronize_rcu();
2524	spin_lock(&swap_lock);
2525	spin_lock(&p->lock);
2526	_enable_swap_info(p);
2527	spin_unlock(&p->lock);
2528	spin_unlock(&swap_lock);
2529}
2530
2531static void reinsert_swap_info(struct swap_info_struct *p)
2532{
2533	spin_lock(&swap_lock);
2534	spin_lock(&p->lock);
2535	setup_swap_info(p, p->prio, p->swap_map, p->cluster_info);
2536	_enable_swap_info(p);
2537	spin_unlock(&p->lock);
2538	spin_unlock(&swap_lock);
2539}
2540
2541bool has_usable_swap(void)
2542{
2543	bool ret = true;
2544
2545	spin_lock(&swap_lock);
2546	if (plist_head_empty(&swap_active_head))
2547		ret = false;
2548	spin_unlock(&swap_lock);
2549	return ret;
2550}
2551
2552SYSCALL_DEFINE1(swapoff, const char __user *, specialfile)
2553{
2554	struct swap_info_struct *p = NULL;
2555	unsigned char *swap_map;
2556	struct swap_cluster_info *cluster_info;
2557	unsigned long *frontswap_map;
2558	struct file *swap_file, *victim;
2559	struct address_space *mapping;
2560	struct inode *inode;
2561	struct filename *pathname;
2562	int err, found = 0;
2563	unsigned int old_block_size;
2564
2565	if (!capable(CAP_SYS_ADMIN))
2566		return -EPERM;
2567
2568	BUG_ON(!current->mm);
2569
2570	pathname = getname(specialfile);
2571	if (IS_ERR(pathname))
2572		return PTR_ERR(pathname);
2573
2574	victim = file_open_name(pathname, O_RDWR|O_LARGEFILE, 0);
2575	err = PTR_ERR(victim);
2576	if (IS_ERR(victim))
2577		goto out;
2578
2579	mapping = victim->f_mapping;
2580	spin_lock(&swap_lock);
2581	plist_for_each_entry(p, &swap_active_head, list) {
2582		if (p->flags & SWP_WRITEOK) {
2583			if (p->swap_file->f_mapping == mapping) {
2584				found = 1;
2585				break;
2586			}
2587		}
2588	}
2589	if (!found) {
2590		err = -EINVAL;
2591		spin_unlock(&swap_lock);
2592		goto out_dput;
2593	}
2594	if (!security_vm_enough_memory_mm(current->mm, p->pages))
2595		vm_unacct_memory(p->pages);
2596	else {
2597		err = -ENOMEM;
2598		spin_unlock(&swap_lock);
2599		goto out_dput;
2600	}
2601	del_from_avail_list(p);
2602	spin_lock(&p->lock);
2603	if (p->prio < 0) {
2604		struct swap_info_struct *si = p;
2605		int nid;
2606
2607		plist_for_each_entry_continue(si, &swap_active_head, list) {
2608			si->prio++;
2609			si->list.prio--;
2610			for_each_node(nid) {
2611				if (si->avail_lists[nid].prio != 1)
2612					si->avail_lists[nid].prio--;
2613			}
2614		}
2615		least_priority++;
2616	}
2617	plist_del(&p->list, &swap_active_head);
2618	atomic_long_sub(p->pages, &nr_swap_pages);
2619	total_swap_pages -= p->pages;
2620	p->flags &= ~SWP_WRITEOK;
2621	spin_unlock(&p->lock);
2622	spin_unlock(&swap_lock);
2623
2624	disable_swap_slots_cache_lock();
2625
2626	set_current_oom_origin();
2627	err = try_to_unuse(p->type, false, 0); /* force unuse all pages */
2628	clear_current_oom_origin();
2629
2630	if (err) {
2631		/* re-insert swap space back into swap_list */
2632		reinsert_swap_info(p);
2633		reenable_swap_slots_cache_unlock();
2634		goto out_dput;
2635	}
2636
2637	reenable_swap_slots_cache_unlock();
2638
2639	spin_lock(&swap_lock);
2640	spin_lock(&p->lock);
2641	p->flags &= ~SWP_VALID;		/* mark swap device as invalid */
2642	spin_unlock(&p->lock);
2643	spin_unlock(&swap_lock);
2644	/*
2645	 * wait for swap operations protected by get/put_swap_device()
2646	 * to complete
2647	 */
2648	synchronize_rcu();
2649
2650	flush_work(&p->discard_work);
2651
2652	destroy_swap_extents(p);
2653	if (p->flags & SWP_CONTINUED)
2654		free_swap_count_continuations(p);
2655
2656	if (!p->bdev || !blk_queue_nonrot(bdev_get_queue(p->bdev)))
2657		atomic_dec(&nr_rotate_swap);
2658
2659	mutex_lock(&swapon_mutex);
2660	spin_lock(&swap_lock);
2661	spin_lock(&p->lock);
2662	drain_mmlist();
2663
2664	/* wait for anyone still in scan_swap_map */
2665	p->highest_bit = 0;		/* cuts scans short */
2666	while (p->flags >= SWP_SCANNING) {
2667		spin_unlock(&p->lock);
2668		spin_unlock(&swap_lock);
2669		schedule_timeout_uninterruptible(1);
2670		spin_lock(&swap_lock);
2671		spin_lock(&p->lock);
2672	}
2673
2674	swap_file = p->swap_file;
2675	old_block_size = p->old_block_size;
2676	p->swap_file = NULL;
2677	p->max = 0;
2678	swap_map = p->swap_map;
2679	p->swap_map = NULL;
2680	cluster_info = p->cluster_info;
2681	p->cluster_info = NULL;
2682	frontswap_map = frontswap_map_get(p);
2683	spin_unlock(&p->lock);
2684	spin_unlock(&swap_lock);
2685	frontswap_invalidate_area(p->type);
2686	frontswap_map_set(p, NULL);
2687	mutex_unlock(&swapon_mutex);
2688	free_percpu(p->percpu_cluster);
2689	p->percpu_cluster = NULL;
2690	free_percpu(p->cluster_next_cpu);
2691	p->cluster_next_cpu = NULL;
2692	vfree(swap_map);
2693	kvfree(cluster_info);
2694	kvfree(frontswap_map);
2695	/* Destroy swap account information */
2696	swap_cgroup_swapoff(p->type);
2697	exit_swap_address_space(p->type);
2698
2699	inode = mapping->host;
2700	if (S_ISBLK(inode->i_mode)) {
2701		struct block_device *bdev = I_BDEV(inode);
2702
2703		set_blocksize(bdev, old_block_size);
2704		blkdev_put(bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
2705	}
2706
2707	inode_lock(inode);
2708	inode->i_flags &= ~S_SWAPFILE;
2709	inode_unlock(inode);
2710	filp_close(swap_file, NULL);
2711
2712	/*
2713	 * Clear the SWP_USED flag after all resources are freed so that swapon
2714	 * can reuse this swap_info in alloc_swap_info() safely.  It is ok to
2715	 * not hold p->lock after we cleared its SWP_WRITEOK.
2716	 */
2717	spin_lock(&swap_lock);
2718	p->flags = 0;
2719	spin_unlock(&swap_lock);
2720
2721	err = 0;
2722	atomic_inc(&proc_poll_event);
2723	wake_up_interruptible(&proc_poll_wait);
2724
2725out_dput:
2726	filp_close(victim, NULL);
2727out:
2728	putname(pathname);
2729	return err;
2730}
2731
2732#ifdef CONFIG_PROC_FS
2733static __poll_t swaps_poll(struct file *file, poll_table *wait)
2734{
2735	struct seq_file *seq = file->private_data;
2736
2737	poll_wait(file, &proc_poll_wait, wait);
2738
2739	if (seq->poll_event != atomic_read(&proc_poll_event)) {
2740		seq->poll_event = atomic_read(&proc_poll_event);
2741		return EPOLLIN | EPOLLRDNORM | EPOLLERR | EPOLLPRI;
2742	}
2743
2744	return EPOLLIN | EPOLLRDNORM;
2745}
2746
2747/* iterator */
2748static void *swap_start(struct seq_file *swap, loff_t *pos)
2749{
2750	struct swap_info_struct *si;
2751	int type;
2752	loff_t l = *pos;
2753
2754	mutex_lock(&swapon_mutex);
2755
2756	if (!l)
2757		return SEQ_START_TOKEN;
2758
2759	for (type = 0; (si = swap_type_to_swap_info(type)); type++) {
2760		if (!(si->flags & SWP_USED) || !si->swap_map)
2761			continue;
2762		if (!--l)
2763			return si;
2764	}
2765
2766	return NULL;
2767}
2768
2769static void *swap_next(struct seq_file *swap, void *v, loff_t *pos)
2770{
2771	struct swap_info_struct *si = v;
2772	int type;
2773
2774	if (v == SEQ_START_TOKEN)
2775		type = 0;
2776	else
2777		type = si->type + 1;
2778
2779	++(*pos);
2780	for (; (si = swap_type_to_swap_info(type)); type++) {
2781		if (!(si->flags & SWP_USED) || !si->swap_map)
2782			continue;
 
2783		return si;
2784	}
2785
2786	return NULL;
2787}
2788
2789static void swap_stop(struct seq_file *swap, void *v)
2790{
2791	mutex_unlock(&swapon_mutex);
2792}
2793
2794static int swap_show(struct seq_file *swap, void *v)
2795{
2796	struct swap_info_struct *si = v;
2797	struct file *file;
2798	int len;
2799	unsigned int bytes, inuse;
2800
2801	if (si == SEQ_START_TOKEN) {
2802		seq_puts(swap,"Filename\t\t\t\tType\t\tSize\t\tUsed\t\tPriority\n");
2803		return 0;
2804	}
2805
2806	bytes = si->pages << (PAGE_SHIFT - 10);
2807	inuse = si->inuse_pages << (PAGE_SHIFT - 10);
2808
2809	file = si->swap_file;
2810	len = seq_file_path(swap, file, " \t\n\\");
2811	seq_printf(swap, "%*s%s\t%u\t%s%u\t%s%d\n",
2812			len < 40 ? 40 - len : 1, " ",
2813			S_ISBLK(file_inode(file)->i_mode) ?
2814				"partition" : "file\t",
2815			bytes, bytes < 10000000 ? "\t" : "",
2816			inuse, inuse < 10000000 ? "\t" : "",
2817			si->prio);
2818	return 0;
2819}
2820
2821static const struct seq_operations swaps_op = {
2822	.start =	swap_start,
2823	.next =		swap_next,
2824	.stop =		swap_stop,
2825	.show =		swap_show
2826};
2827
2828static int swaps_open(struct inode *inode, struct file *file)
2829{
2830	struct seq_file *seq;
2831	int ret;
2832
2833	ret = seq_open(file, &swaps_op);
2834	if (ret)
2835		return ret;
2836
2837	seq = file->private_data;
2838	seq->poll_event = atomic_read(&proc_poll_event);
2839	return 0;
2840}
2841
2842static const struct proc_ops swaps_proc_ops = {
2843	.proc_flags	= PROC_ENTRY_PERMANENT,
2844	.proc_open	= swaps_open,
2845	.proc_read	= seq_read,
2846	.proc_lseek	= seq_lseek,
2847	.proc_release	= seq_release,
2848	.proc_poll	= swaps_poll,
2849};
2850
2851static int __init procswaps_init(void)
2852{
2853	proc_create("swaps", 0, NULL, &swaps_proc_ops);
2854	return 0;
2855}
2856__initcall(procswaps_init);
2857#endif /* CONFIG_PROC_FS */
2858
2859#ifdef MAX_SWAPFILES_CHECK
2860static int __init max_swapfiles_check(void)
2861{
2862	MAX_SWAPFILES_CHECK();
2863	return 0;
2864}
2865late_initcall(max_swapfiles_check);
2866#endif
2867
2868static struct swap_info_struct *alloc_swap_info(void)
2869{
2870	struct swap_info_struct *p;
2871	unsigned int type;
2872	int i;
2873
2874	p = kvzalloc(struct_size(p, avail_lists, nr_node_ids), GFP_KERNEL);
2875	if (!p)
2876		return ERR_PTR(-ENOMEM);
2877
2878	spin_lock(&swap_lock);
2879	for (type = 0; type < nr_swapfiles; type++) {
2880		if (!(swap_info[type]->flags & SWP_USED))
2881			break;
2882	}
2883	if (type >= MAX_SWAPFILES) {
2884		spin_unlock(&swap_lock);
2885		kvfree(p);
2886		return ERR_PTR(-EPERM);
2887	}
2888	if (type >= nr_swapfiles) {
2889		p->type = type;
2890		WRITE_ONCE(swap_info[type], p);
2891		/*
2892		 * Write swap_info[type] before nr_swapfiles, in case a
2893		 * racing procfs swap_start() or swap_next() is reading them.
2894		 * (We never shrink nr_swapfiles, we never free this entry.)
2895		 */
2896		smp_wmb();
2897		WRITE_ONCE(nr_swapfiles, nr_swapfiles + 1);
2898	} else {
2899		kvfree(p);
2900		p = swap_info[type];
2901		/*
2902		 * Do not memset this entry: a racing procfs swap_next()
2903		 * would be relying on p->type to remain valid.
2904		 */
2905	}
2906	p->swap_extent_root = RB_ROOT;
2907	plist_node_init(&p->list, 0);
2908	for_each_node(i)
2909		plist_node_init(&p->avail_lists[i], 0);
2910	p->flags = SWP_USED;
2911	spin_unlock(&swap_lock);
2912	spin_lock_init(&p->lock);
2913	spin_lock_init(&p->cont_lock);
2914
2915	return p;
2916}
2917
2918static int claim_swapfile(struct swap_info_struct *p, struct inode *inode)
2919{
2920	int error;
2921
2922	if (S_ISBLK(inode->i_mode)) {
2923		p->bdev = bdgrab(I_BDEV(inode));
2924		error = blkdev_get(p->bdev,
2925				   FMODE_READ | FMODE_WRITE | FMODE_EXCL, p);
2926		if (error < 0) {
2927			p->bdev = NULL;
2928			return error;
2929		}
2930		p->old_block_size = block_size(p->bdev);
2931		error = set_blocksize(p->bdev, PAGE_SIZE);
2932		if (error < 0)
2933			return error;
2934		/*
2935		 * Zoned block devices contain zones that have a sequential
2936		 * write only restriction.  Hence zoned block devices are not
2937		 * suitable for swapping.  Disallow them here.
2938		 */
2939		if (blk_queue_is_zoned(p->bdev->bd_disk->queue))
2940			return -EINVAL;
2941		p->flags |= SWP_BLKDEV;
2942	} else if (S_ISREG(inode->i_mode)) {
2943		p->bdev = inode->i_sb->s_bdev;
2944	}
2945
 
 
 
 
2946	return 0;
2947}
2948
2949
2950/*
2951 * Find out how many pages are allowed for a single swap device. There
2952 * are two limiting factors:
2953 * 1) the number of bits for the swap offset in the swp_entry_t type, and
2954 * 2) the number of bits in the swap pte, as defined by the different
2955 * architectures.
2956 *
2957 * In order to find the largest possible bit mask, a swap entry with
2958 * swap type 0 and swap offset ~0UL is created, encoded to a swap pte,
2959 * decoded to a swp_entry_t again, and finally the swap offset is
2960 * extracted.
2961 *
2962 * This will mask all the bits from the initial ~0UL mask that can't
2963 * be encoded in either the swp_entry_t or the architecture definition
2964 * of a swap pte.
2965 */
2966unsigned long generic_max_swapfile_size(void)
2967{
2968	return swp_offset(pte_to_swp_entry(
2969			swp_entry_to_pte(swp_entry(0, ~0UL)))) + 1;
2970}
2971
2972/* Can be overridden by an architecture for additional checks. */
2973__weak unsigned long max_swapfile_size(void)
2974{
2975	return generic_max_swapfile_size();
2976}
2977
2978static unsigned long read_swap_header(struct swap_info_struct *p,
2979					union swap_header *swap_header,
2980					struct inode *inode)
2981{
2982	int i;
2983	unsigned long maxpages;
2984	unsigned long swapfilepages;
2985	unsigned long last_page;
2986
2987	if (memcmp("SWAPSPACE2", swap_header->magic.magic, 10)) {
2988		pr_err("Unable to find swap-space signature\n");
2989		return 0;
2990	}
2991
2992	/* swap partition endianess hack... */
2993	if (swab32(swap_header->info.version) == 1) {
2994		swab32s(&swap_header->info.version);
2995		swab32s(&swap_header->info.last_page);
2996		swab32s(&swap_header->info.nr_badpages);
2997		if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES)
2998			return 0;
2999		for (i = 0; i < swap_header->info.nr_badpages; i++)
3000			swab32s(&swap_header->info.badpages[i]);
3001	}
3002	/* Check the swap header's sub-version */
3003	if (swap_header->info.version != 1) {
3004		pr_warn("Unable to handle swap header version %d\n",
3005			swap_header->info.version);
3006		return 0;
3007	}
3008
3009	p->lowest_bit  = 1;
3010	p->cluster_next = 1;
3011	p->cluster_nr = 0;
3012
3013	maxpages = max_swapfile_size();
3014	last_page = swap_header->info.last_page;
3015	if (!last_page) {
3016		pr_warn("Empty swap-file\n");
3017		return 0;
3018	}
3019	if (last_page > maxpages) {
3020		pr_warn("Truncating oversized swap area, only using %luk out of %luk\n",
3021			maxpages << (PAGE_SHIFT - 10),
3022			last_page << (PAGE_SHIFT - 10));
3023	}
3024	if (maxpages > last_page) {
3025		maxpages = last_page + 1;
3026		/* p->max is an unsigned int: don't overflow it */
3027		if ((unsigned int)maxpages == 0)
3028			maxpages = UINT_MAX;
3029	}
3030	p->highest_bit = maxpages - 1;
3031
3032	if (!maxpages)
3033		return 0;
3034	swapfilepages = i_size_read(inode) >> PAGE_SHIFT;
3035	if (swapfilepages && maxpages > swapfilepages) {
3036		pr_warn("Swap area shorter than signature indicates\n");
3037		return 0;
3038	}
3039	if (swap_header->info.nr_badpages && S_ISREG(inode->i_mode))
3040		return 0;
3041	if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES)
3042		return 0;
3043
3044	return maxpages;
3045}
3046
3047#define SWAP_CLUSTER_INFO_COLS						\
3048	DIV_ROUND_UP(L1_CACHE_BYTES, sizeof(struct swap_cluster_info))
3049#define SWAP_CLUSTER_SPACE_COLS						\
3050	DIV_ROUND_UP(SWAP_ADDRESS_SPACE_PAGES, SWAPFILE_CLUSTER)
3051#define SWAP_CLUSTER_COLS						\
3052	max_t(unsigned int, SWAP_CLUSTER_INFO_COLS, SWAP_CLUSTER_SPACE_COLS)
3053
3054static int setup_swap_map_and_extents(struct swap_info_struct *p,
3055					union swap_header *swap_header,
3056					unsigned char *swap_map,
3057					struct swap_cluster_info *cluster_info,
3058					unsigned long maxpages,
3059					sector_t *span)
3060{
3061	unsigned int j, k;
3062	unsigned int nr_good_pages;
3063	int nr_extents;
3064	unsigned long nr_clusters = DIV_ROUND_UP(maxpages, SWAPFILE_CLUSTER);
3065	unsigned long col = p->cluster_next / SWAPFILE_CLUSTER % SWAP_CLUSTER_COLS;
3066	unsigned long i, idx;
3067
3068	nr_good_pages = maxpages - 1;	/* omit header page */
3069
3070	cluster_list_init(&p->free_clusters);
3071	cluster_list_init(&p->discard_clusters);
3072
3073	for (i = 0; i < swap_header->info.nr_badpages; i++) {
3074		unsigned int page_nr = swap_header->info.badpages[i];
3075		if (page_nr == 0 || page_nr > swap_header->info.last_page)
3076			return -EINVAL;
3077		if (page_nr < maxpages) {
3078			swap_map[page_nr] = SWAP_MAP_BAD;
3079			nr_good_pages--;
3080			/*
3081			 * Haven't marked the cluster free yet, no list
3082			 * operation involved
3083			 */
3084			inc_cluster_info_page(p, cluster_info, page_nr);
3085		}
3086	}
3087
3088	/* Haven't marked the cluster free yet, no list operation involved */
3089	for (i = maxpages; i < round_up(maxpages, SWAPFILE_CLUSTER); i++)
3090		inc_cluster_info_page(p, cluster_info, i);
3091
3092	if (nr_good_pages) {
3093		swap_map[0] = SWAP_MAP_BAD;
3094		/*
3095		 * Not mark the cluster free yet, no list
3096		 * operation involved
3097		 */
3098		inc_cluster_info_page(p, cluster_info, 0);
3099		p->max = maxpages;
3100		p->pages = nr_good_pages;
3101		nr_extents = setup_swap_extents(p, span);
3102		if (nr_extents < 0)
3103			return nr_extents;
3104		nr_good_pages = p->pages;
3105	}
3106	if (!nr_good_pages) {
3107		pr_warn("Empty swap-file\n");
3108		return -EINVAL;
3109	}
3110
3111	if (!cluster_info)
3112		return nr_extents;
3113
3114
3115	/*
3116	 * Reduce false cache line sharing between cluster_info and
3117	 * sharing same address space.
3118	 */
3119	for (k = 0; k < SWAP_CLUSTER_COLS; k++) {
3120		j = (k + col) % SWAP_CLUSTER_COLS;
3121		for (i = 0; i < DIV_ROUND_UP(nr_clusters, SWAP_CLUSTER_COLS); i++) {
3122			idx = i * SWAP_CLUSTER_COLS + j;
3123			if (idx >= nr_clusters)
3124				continue;
3125			if (cluster_count(&cluster_info[idx]))
3126				continue;
3127			cluster_set_flag(&cluster_info[idx], CLUSTER_FLAG_FREE);
3128			cluster_list_add_tail(&p->free_clusters, cluster_info,
3129					      idx);
3130		}
3131	}
3132	return nr_extents;
3133}
3134
3135/*
3136 * Helper to sys_swapon determining if a given swap
3137 * backing device queue supports DISCARD operations.
3138 */
3139static bool swap_discardable(struct swap_info_struct *si)
3140{
3141	struct request_queue *q = bdev_get_queue(si->bdev);
3142
3143	if (!q || !blk_queue_discard(q))
3144		return false;
3145
3146	return true;
3147}
3148
3149SYSCALL_DEFINE2(swapon, const char __user *, specialfile, int, swap_flags)
3150{
3151	struct swap_info_struct *p;
3152	struct filename *name;
3153	struct file *swap_file = NULL;
3154	struct address_space *mapping;
3155	int prio;
3156	int error;
3157	union swap_header *swap_header;
3158	int nr_extents;
3159	sector_t span;
3160	unsigned long maxpages;
3161	unsigned char *swap_map = NULL;
3162	struct swap_cluster_info *cluster_info = NULL;
3163	unsigned long *frontswap_map = NULL;
3164	struct page *page = NULL;
3165	struct inode *inode = NULL;
3166	bool inced_nr_rotate_swap = false;
3167
3168	if (swap_flags & ~SWAP_FLAGS_VALID)
3169		return -EINVAL;
3170
3171	if (!capable(CAP_SYS_ADMIN))
3172		return -EPERM;
3173
3174	if (!swap_avail_heads)
3175		return -ENOMEM;
3176
3177	p = alloc_swap_info();
3178	if (IS_ERR(p))
3179		return PTR_ERR(p);
3180
3181	INIT_WORK(&p->discard_work, swap_discard_work);
3182
3183	name = getname(specialfile);
3184	if (IS_ERR(name)) {
3185		error = PTR_ERR(name);
3186		name = NULL;
3187		goto bad_swap;
3188	}
3189	swap_file = file_open_name(name, O_RDWR|O_LARGEFILE, 0);
3190	if (IS_ERR(swap_file)) {
3191		error = PTR_ERR(swap_file);
3192		swap_file = NULL;
3193		goto bad_swap;
3194	}
3195
3196	p->swap_file = swap_file;
3197	mapping = swap_file->f_mapping;
3198	inode = mapping->host;
3199
 
3200	error = claim_swapfile(p, inode);
3201	if (unlikely(error))
3202		goto bad_swap;
3203
3204	inode_lock(inode);
3205	if (IS_SWAPFILE(inode)) {
3206		error = -EBUSY;
3207		goto bad_swap_unlock_inode;
3208	}
3209
3210	/*
3211	 * Read the swap header.
3212	 */
3213	if (!mapping->a_ops->readpage) {
3214		error = -EINVAL;
3215		goto bad_swap_unlock_inode;
3216	}
3217	page = read_mapping_page(mapping, 0, swap_file);
3218	if (IS_ERR(page)) {
3219		error = PTR_ERR(page);
3220		goto bad_swap_unlock_inode;
3221	}
3222	swap_header = kmap(page);
3223
3224	maxpages = read_swap_header(p, swap_header, inode);
3225	if (unlikely(!maxpages)) {
3226		error = -EINVAL;
3227		goto bad_swap_unlock_inode;
3228	}
3229
3230	/* OK, set up the swap map and apply the bad block list */
3231	swap_map = vzalloc(maxpages);
3232	if (!swap_map) {
3233		error = -ENOMEM;
3234		goto bad_swap_unlock_inode;
3235	}
3236
3237	if (bdi_cap_stable_pages_required(inode_to_bdi(inode)))
3238		p->flags |= SWP_STABLE_WRITES;
3239
3240	if (bdi_cap_synchronous_io(inode_to_bdi(inode)))
3241		p->flags |= SWP_SYNCHRONOUS_IO;
3242
3243	if (p->bdev && blk_queue_nonrot(bdev_get_queue(p->bdev))) {
3244		int cpu;
3245		unsigned long ci, nr_cluster;
3246
3247		p->flags |= SWP_SOLIDSTATE;
3248		p->cluster_next_cpu = alloc_percpu(unsigned int);
3249		if (!p->cluster_next_cpu) {
3250			error = -ENOMEM;
3251			goto bad_swap_unlock_inode;
3252		}
3253		/*
3254		 * select a random position to start with to help wear leveling
3255		 * SSD
3256		 */
3257		for_each_possible_cpu(cpu) {
3258			per_cpu(*p->cluster_next_cpu, cpu) =
3259				1 + prandom_u32_max(p->highest_bit);
3260		}
3261		nr_cluster = DIV_ROUND_UP(maxpages, SWAPFILE_CLUSTER);
3262
3263		cluster_info = kvcalloc(nr_cluster, sizeof(*cluster_info),
3264					GFP_KERNEL);
3265		if (!cluster_info) {
3266			error = -ENOMEM;
3267			goto bad_swap_unlock_inode;
3268		}
3269
3270		for (ci = 0; ci < nr_cluster; ci++)
3271			spin_lock_init(&((cluster_info + ci)->lock));
3272
3273		p->percpu_cluster = alloc_percpu(struct percpu_cluster);
3274		if (!p->percpu_cluster) {
3275			error = -ENOMEM;
3276			goto bad_swap_unlock_inode;
3277		}
3278		for_each_possible_cpu(cpu) {
3279			struct percpu_cluster *cluster;
3280			cluster = per_cpu_ptr(p->percpu_cluster, cpu);
3281			cluster_set_null(&cluster->index);
3282		}
3283	} else {
3284		atomic_inc(&nr_rotate_swap);
3285		inced_nr_rotate_swap = true;
3286	}
3287
3288	error = swap_cgroup_swapon(p->type, maxpages);
3289	if (error)
3290		goto bad_swap_unlock_inode;
3291
3292	nr_extents = setup_swap_map_and_extents(p, swap_header, swap_map,
3293		cluster_info, maxpages, &span);
3294	if (unlikely(nr_extents < 0)) {
3295		error = nr_extents;
3296		goto bad_swap_unlock_inode;
3297	}
3298	/* frontswap enabled? set up bit-per-page map for frontswap */
3299	if (IS_ENABLED(CONFIG_FRONTSWAP))
3300		frontswap_map = kvcalloc(BITS_TO_LONGS(maxpages),
3301					 sizeof(long),
3302					 GFP_KERNEL);
3303
3304	if (p->bdev &&(swap_flags & SWAP_FLAG_DISCARD) && swap_discardable(p)) {
3305		/*
3306		 * When discard is enabled for swap with no particular
3307		 * policy flagged, we set all swap discard flags here in
3308		 * order to sustain backward compatibility with older
3309		 * swapon(8) releases.
3310		 */
3311		p->flags |= (SWP_DISCARDABLE | SWP_AREA_DISCARD |
3312			     SWP_PAGE_DISCARD);
3313
3314		/*
3315		 * By flagging sys_swapon, a sysadmin can tell us to
3316		 * either do single-time area discards only, or to just
3317		 * perform discards for released swap page-clusters.
3318		 * Now it's time to adjust the p->flags accordingly.
3319		 */
3320		if (swap_flags & SWAP_FLAG_DISCARD_ONCE)
3321			p->flags &= ~SWP_PAGE_DISCARD;
3322		else if (swap_flags & SWAP_FLAG_DISCARD_PAGES)
3323			p->flags &= ~SWP_AREA_DISCARD;
3324
3325		/* issue a swapon-time discard if it's still required */
3326		if (p->flags & SWP_AREA_DISCARD) {
3327			int err = discard_swap(p);
3328			if (unlikely(err))
3329				pr_err("swapon: discard_swap(%p): %d\n",
3330					p, err);
3331		}
3332	}
3333
3334	error = init_swap_address_space(p->type, maxpages);
3335	if (error)
3336		goto bad_swap_unlock_inode;
3337
3338	/*
3339	 * Flush any pending IO and dirty mappings before we start using this
3340	 * swap device.
3341	 */
3342	inode->i_flags |= S_SWAPFILE;
3343	error = inode_drain_writes(inode);
3344	if (error) {
3345		inode->i_flags &= ~S_SWAPFILE;
3346		goto bad_swap_unlock_inode;
3347	}
3348
3349	mutex_lock(&swapon_mutex);
3350	prio = -1;
3351	if (swap_flags & SWAP_FLAG_PREFER)
3352		prio =
3353		  (swap_flags & SWAP_FLAG_PRIO_MASK) >> SWAP_FLAG_PRIO_SHIFT;
3354	enable_swap_info(p, prio, swap_map, cluster_info, frontswap_map);
3355
3356	pr_info("Adding %uk swap on %s.  Priority:%d extents:%d across:%lluk %s%s%s%s%s\n",
3357		p->pages<<(PAGE_SHIFT-10), name->name, p->prio,
3358		nr_extents, (unsigned long long)span<<(PAGE_SHIFT-10),
3359		(p->flags & SWP_SOLIDSTATE) ? "SS" : "",
3360		(p->flags & SWP_DISCARDABLE) ? "D" : "",
3361		(p->flags & SWP_AREA_DISCARD) ? "s" : "",
3362		(p->flags & SWP_PAGE_DISCARD) ? "c" : "",
3363		(frontswap_map) ? "FS" : "");
3364
3365	mutex_unlock(&swapon_mutex);
3366	atomic_inc(&proc_poll_event);
3367	wake_up_interruptible(&proc_poll_wait);
3368
3369	error = 0;
3370	goto out;
3371bad_swap_unlock_inode:
3372	inode_unlock(inode);
3373bad_swap:
3374	free_percpu(p->percpu_cluster);
3375	p->percpu_cluster = NULL;
3376	free_percpu(p->cluster_next_cpu);
3377	p->cluster_next_cpu = NULL;
3378	if (inode && S_ISBLK(inode->i_mode) && p->bdev) {
3379		set_blocksize(p->bdev, p->old_block_size);
3380		blkdev_put(p->bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
3381	}
3382	inode = NULL;
3383	destroy_swap_extents(p);
3384	swap_cgroup_swapoff(p->type);
3385	spin_lock(&swap_lock);
3386	p->swap_file = NULL;
3387	p->flags = 0;
3388	spin_unlock(&swap_lock);
3389	vfree(swap_map);
3390	kvfree(cluster_info);
3391	kvfree(frontswap_map);
3392	if (inced_nr_rotate_swap)
3393		atomic_dec(&nr_rotate_swap);
3394	if (swap_file)
 
 
 
 
3395		filp_close(swap_file, NULL);
 
3396out:
3397	if (page && !IS_ERR(page)) {
3398		kunmap(page);
3399		put_page(page);
3400	}
3401	if (name)
3402		putname(name);
3403	if (inode)
3404		inode_unlock(inode);
3405	if (!error)
3406		enable_swap_slots_cache();
3407	return error;
3408}
3409
3410void si_swapinfo(struct sysinfo *val)
3411{
3412	unsigned int type;
3413	unsigned long nr_to_be_unused = 0;
3414
3415	spin_lock(&swap_lock);
3416	for (type = 0; type < nr_swapfiles; type++) {
3417		struct swap_info_struct *si = swap_info[type];
3418
3419		if ((si->flags & SWP_USED) && !(si->flags & SWP_WRITEOK))
3420			nr_to_be_unused += si->inuse_pages;
3421	}
3422	val->freeswap = atomic_long_read(&nr_swap_pages) + nr_to_be_unused;
3423	val->totalswap = total_swap_pages + nr_to_be_unused;
3424	spin_unlock(&swap_lock);
3425}
3426
3427/*
3428 * Verify that a swap entry is valid and increment its swap map count.
3429 *
3430 * Returns error code in following case.
3431 * - success -> 0
3432 * - swp_entry is invalid -> EINVAL
3433 * - swp_entry is migration entry -> EINVAL
3434 * - swap-cache reference is requested but there is already one. -> EEXIST
3435 * - swap-cache reference is requested but the entry is not used. -> ENOENT
3436 * - swap-mapped reference requested but needs continued swap count. -> ENOMEM
3437 */
3438static int __swap_duplicate(swp_entry_t entry, unsigned char usage)
3439{
3440	struct swap_info_struct *p;
3441	struct swap_cluster_info *ci;
3442	unsigned long offset;
3443	unsigned char count;
3444	unsigned char has_cache;
3445	int err = -EINVAL;
3446
3447	p = get_swap_device(entry);
3448	if (!p)
3449		goto out;
3450
3451	offset = swp_offset(entry);
3452	ci = lock_cluster_or_swap_info(p, offset);
3453
3454	count = p->swap_map[offset];
3455
3456	/*
3457	 * swapin_readahead() doesn't check if a swap entry is valid, so the
3458	 * swap entry could be SWAP_MAP_BAD. Check here with lock held.
3459	 */
3460	if (unlikely(swap_count(count) == SWAP_MAP_BAD)) {
3461		err = -ENOENT;
3462		goto unlock_out;
3463	}
3464
3465	has_cache = count & SWAP_HAS_CACHE;
3466	count &= ~SWAP_HAS_CACHE;
3467	err = 0;
3468
3469	if (usage == SWAP_HAS_CACHE) {
3470
3471		/* set SWAP_HAS_CACHE if there is no cache and entry is used */
3472		if (!has_cache && count)
3473			has_cache = SWAP_HAS_CACHE;
3474		else if (has_cache)		/* someone else added cache */
3475			err = -EEXIST;
3476		else				/* no users remaining */
3477			err = -ENOENT;
3478
3479	} else if (count || has_cache) {
3480
3481		if ((count & ~COUNT_CONTINUED) < SWAP_MAP_MAX)
3482			count += usage;
3483		else if ((count & ~COUNT_CONTINUED) > SWAP_MAP_MAX)
3484			err = -EINVAL;
3485		else if (swap_count_continued(p, offset, count))
3486			count = COUNT_CONTINUED;
3487		else
3488			err = -ENOMEM;
3489	} else
3490		err = -ENOENT;			/* unused swap entry */
3491
3492	WRITE_ONCE(p->swap_map[offset], count | has_cache);
3493
3494unlock_out:
3495	unlock_cluster_or_swap_info(p, ci);
3496out:
3497	if (p)
3498		put_swap_device(p);
3499	return err;
3500}
3501
3502/*
3503 * Help swapoff by noting that swap entry belongs to shmem/tmpfs
3504 * (in which case its reference count is never incremented).
3505 */
3506void swap_shmem_alloc(swp_entry_t entry)
3507{
3508	__swap_duplicate(entry, SWAP_MAP_SHMEM);
3509}
3510
3511/*
3512 * Increase reference count of swap entry by 1.
3513 * Returns 0 for success, or -ENOMEM if a swap_count_continuation is required
3514 * but could not be atomically allocated.  Returns 0, just as if it succeeded,
3515 * if __swap_duplicate() fails for another reason (-EINVAL or -ENOENT), which
3516 * might occur if a page table entry has got corrupted.
3517 */
3518int swap_duplicate(swp_entry_t entry)
3519{
3520	int err = 0;
3521
3522	while (!err && __swap_duplicate(entry, 1) == -ENOMEM)
3523		err = add_swap_count_continuation(entry, GFP_ATOMIC);
3524	return err;
3525}
3526
3527/*
3528 * @entry: swap entry for which we allocate swap cache.
3529 *
3530 * Called when allocating swap cache for existing swap entry,
3531 * This can return error codes. Returns 0 at success.
3532 * -EEXIST means there is a swap cache.
3533 * Note: return code is different from swap_duplicate().
3534 */
3535int swapcache_prepare(swp_entry_t entry)
3536{
3537	return __swap_duplicate(entry, SWAP_HAS_CACHE);
3538}
3539
3540struct swap_info_struct *swp_swap_info(swp_entry_t entry)
3541{
3542	return swap_type_to_swap_info(swp_type(entry));
3543}
3544
3545struct swap_info_struct *page_swap_info(struct page *page)
3546{
3547	swp_entry_t entry = { .val = page_private(page) };
3548	return swp_swap_info(entry);
3549}
3550
3551/*
3552 * out-of-line __page_file_ methods to avoid include hell.
3553 */
3554struct address_space *__page_file_mapping(struct page *page)
3555{
3556	return page_swap_info(page)->swap_file->f_mapping;
3557}
3558EXPORT_SYMBOL_GPL(__page_file_mapping);
3559
3560pgoff_t __page_file_index(struct page *page)
3561{
3562	swp_entry_t swap = { .val = page_private(page) };
3563	return swp_offset(swap);
3564}
3565EXPORT_SYMBOL_GPL(__page_file_index);
3566
3567/*
3568 * add_swap_count_continuation - called when a swap count is duplicated
3569 * beyond SWAP_MAP_MAX, it allocates a new page and links that to the entry's
3570 * page of the original vmalloc'ed swap_map, to hold the continuation count
3571 * (for that entry and for its neighbouring PAGE_SIZE swap entries).  Called
3572 * again when count is duplicated beyond SWAP_MAP_MAX * SWAP_CONT_MAX, etc.
3573 *
3574 * These continuation pages are seldom referenced: the common paths all work
3575 * on the original swap_map, only referring to a continuation page when the
3576 * low "digit" of a count is incremented or decremented through SWAP_MAP_MAX.
3577 *
3578 * add_swap_count_continuation(, GFP_ATOMIC) can be called while holding
3579 * page table locks; if it fails, add_swap_count_continuation(, GFP_KERNEL)
3580 * can be called after dropping locks.
3581 */
3582int add_swap_count_continuation(swp_entry_t entry, gfp_t gfp_mask)
3583{
3584	struct swap_info_struct *si;
3585	struct swap_cluster_info *ci;
3586	struct page *head;
3587	struct page *page;
3588	struct page *list_page;
3589	pgoff_t offset;
3590	unsigned char count;
3591	int ret = 0;
3592
3593	/*
3594	 * When debugging, it's easier to use __GFP_ZERO here; but it's better
3595	 * for latency not to zero a page while GFP_ATOMIC and holding locks.
3596	 */
3597	page = alloc_page(gfp_mask | __GFP_HIGHMEM);
3598
3599	si = get_swap_device(entry);
3600	if (!si) {
3601		/*
3602		 * An acceptable race has occurred since the failing
3603		 * __swap_duplicate(): the swap device may be swapoff
3604		 */
3605		goto outer;
3606	}
3607	spin_lock(&si->lock);
3608
3609	offset = swp_offset(entry);
3610
3611	ci = lock_cluster(si, offset);
3612
3613	count = si->swap_map[offset] & ~SWAP_HAS_CACHE;
3614
3615	if ((count & ~COUNT_CONTINUED) != SWAP_MAP_MAX) {
3616		/*
3617		 * The higher the swap count, the more likely it is that tasks
3618		 * will race to add swap count continuation: we need to avoid
3619		 * over-provisioning.
3620		 */
3621		goto out;
3622	}
3623
3624	if (!page) {
3625		ret = -ENOMEM;
3626		goto out;
3627	}
3628
3629	/*
3630	 * We are fortunate that although vmalloc_to_page uses pte_offset_map,
3631	 * no architecture is using highmem pages for kernel page tables: so it
3632	 * will not corrupt the GFP_ATOMIC caller's atomic page table kmaps.
3633	 */
3634	head = vmalloc_to_page(si->swap_map + offset);
3635	offset &= ~PAGE_MASK;
3636
3637	spin_lock(&si->cont_lock);
3638	/*
3639	 * Page allocation does not initialize the page's lru field,
3640	 * but it does always reset its private field.
3641	 */
3642	if (!page_private(head)) {
3643		BUG_ON(count & COUNT_CONTINUED);
3644		INIT_LIST_HEAD(&head->lru);
3645		set_page_private(head, SWP_CONTINUED);
3646		si->flags |= SWP_CONTINUED;
3647	}
3648
3649	list_for_each_entry(list_page, &head->lru, lru) {
3650		unsigned char *map;
3651
3652		/*
3653		 * If the previous map said no continuation, but we've found
3654		 * a continuation page, free our allocation and use this one.
3655		 */
3656		if (!(count & COUNT_CONTINUED))
3657			goto out_unlock_cont;
3658
3659		map = kmap_atomic(list_page) + offset;
3660		count = *map;
3661		kunmap_atomic(map);
3662
3663		/*
3664		 * If this continuation count now has some space in it,
3665		 * free our allocation and use this one.
3666		 */
3667		if ((count & ~COUNT_CONTINUED) != SWAP_CONT_MAX)
3668			goto out_unlock_cont;
3669	}
3670
3671	list_add_tail(&page->lru, &head->lru);
3672	page = NULL;			/* now it's attached, don't free it */
3673out_unlock_cont:
3674	spin_unlock(&si->cont_lock);
3675out:
3676	unlock_cluster(ci);
3677	spin_unlock(&si->lock);
3678	put_swap_device(si);
3679outer:
3680	if (page)
3681		__free_page(page);
3682	return ret;
3683}
3684
3685/*
3686 * swap_count_continued - when the original swap_map count is incremented
3687 * from SWAP_MAP_MAX, check if there is already a continuation page to carry
3688 * into, carry if so, or else fail until a new continuation page is allocated;
3689 * when the original swap_map count is decremented from 0 with continuation,
3690 * borrow from the continuation and report whether it still holds more.
3691 * Called while __swap_duplicate() or swap_entry_free() holds swap or cluster
3692 * lock.
3693 */
3694static bool swap_count_continued(struct swap_info_struct *si,
3695				 pgoff_t offset, unsigned char count)
3696{
3697	struct page *head;
3698	struct page *page;
3699	unsigned char *map;
3700	bool ret;
3701
3702	head = vmalloc_to_page(si->swap_map + offset);
3703	if (page_private(head) != SWP_CONTINUED) {
3704		BUG_ON(count & COUNT_CONTINUED);
3705		return false;		/* need to add count continuation */
3706	}
3707
3708	spin_lock(&si->cont_lock);
3709	offset &= ~PAGE_MASK;
3710	page = list_next_entry(head, lru);
3711	map = kmap_atomic(page) + offset;
3712
3713	if (count == SWAP_MAP_MAX)	/* initial increment from swap_map */
3714		goto init_map;		/* jump over SWAP_CONT_MAX checks */
3715
3716	if (count == (SWAP_MAP_MAX | COUNT_CONTINUED)) { /* incrementing */
3717		/*
3718		 * Think of how you add 1 to 999
3719		 */
3720		while (*map == (SWAP_CONT_MAX | COUNT_CONTINUED)) {
3721			kunmap_atomic(map);
3722			page = list_next_entry(page, lru);
3723			BUG_ON(page == head);
3724			map = kmap_atomic(page) + offset;
3725		}
3726		if (*map == SWAP_CONT_MAX) {
3727			kunmap_atomic(map);
3728			page = list_next_entry(page, lru);
3729			if (page == head) {
3730				ret = false;	/* add count continuation */
3731				goto out;
3732			}
3733			map = kmap_atomic(page) + offset;
3734init_map:		*map = 0;		/* we didn't zero the page */
3735		}
3736		*map += 1;
3737		kunmap_atomic(map);
3738		while ((page = list_prev_entry(page, lru)) != head) {
 
3739			map = kmap_atomic(page) + offset;
3740			*map = COUNT_CONTINUED;
3741			kunmap_atomic(map);
 
3742		}
3743		ret = true;			/* incremented */
3744
3745	} else {				/* decrementing */
3746		/*
3747		 * Think of how you subtract 1 from 1000
3748		 */
3749		BUG_ON(count != COUNT_CONTINUED);
3750		while (*map == COUNT_CONTINUED) {
3751			kunmap_atomic(map);
3752			page = list_next_entry(page, lru);
3753			BUG_ON(page == head);
3754			map = kmap_atomic(page) + offset;
3755		}
3756		BUG_ON(*map == 0);
3757		*map -= 1;
3758		if (*map == 0)
3759			count = 0;
3760		kunmap_atomic(map);
3761		while ((page = list_prev_entry(page, lru)) != head) {
 
3762			map = kmap_atomic(page) + offset;
3763			*map = SWAP_CONT_MAX | count;
3764			count = COUNT_CONTINUED;
3765			kunmap_atomic(map);
 
3766		}
3767		ret = count == COUNT_CONTINUED;
3768	}
3769out:
3770	spin_unlock(&si->cont_lock);
3771	return ret;
3772}
3773
3774/*
3775 * free_swap_count_continuations - swapoff free all the continuation pages
3776 * appended to the swap_map, after swap_map is quiesced, before vfree'ing it.
3777 */
3778static void free_swap_count_continuations(struct swap_info_struct *si)
3779{
3780	pgoff_t offset;
3781
3782	for (offset = 0; offset < si->max; offset += PAGE_SIZE) {
3783		struct page *head;
3784		head = vmalloc_to_page(si->swap_map + offset);
3785		if (page_private(head)) {
3786			struct page *page, *next;
3787
3788			list_for_each_entry_safe(page, next, &head->lru, lru) {
3789				list_del(&page->lru);
3790				__free_page(page);
3791			}
3792		}
3793	}
3794}
3795
3796#if defined(CONFIG_MEMCG) && defined(CONFIG_BLK_CGROUP)
3797void cgroup_throttle_swaprate(struct page *page, gfp_t gfp_mask)
 
3798{
3799	struct swap_info_struct *si, *next;
3800	int nid = page_to_nid(page);
3801
3802	if (!(gfp_mask & __GFP_IO))
3803		return;
3804
3805	if (!blk_cgroup_congested())
3806		return;
3807
3808	/*
3809	 * We've already scheduled a throttle, avoid taking the global swap
3810	 * lock.
3811	 */
3812	if (current->throttle_queue)
3813		return;
3814
3815	spin_lock(&swap_avail_lock);
3816	plist_for_each_entry_safe(si, next, &swap_avail_heads[nid],
3817				  avail_lists[nid]) {
3818		if (si->bdev) {
3819			blkcg_schedule_throttle(bdev_get_queue(si->bdev), true);
 
3820			break;
3821		}
3822	}
3823	spin_unlock(&swap_avail_lock);
3824}
3825#endif
3826
3827static int __init swapfile_init(void)
3828{
3829	int nid;
3830
3831	swap_avail_heads = kmalloc_array(nr_node_ids, sizeof(struct plist_head),
3832					 GFP_KERNEL);
3833	if (!swap_avail_heads) {
3834		pr_emerg("Not enough memory for swap heads, swap is disabled\n");
3835		return -ENOMEM;
3836	}
3837
3838	for_each_node(nid)
3839		plist_head_init(&swap_avail_heads[nid]);
3840
3841	return 0;
3842}
3843subsys_initcall(swapfile_init);