linux/mm/swapfile.c
<<
>>
Prefs
   1/*
   2 *  linux/mm/swapfile.c
   3 *
   4 *  Copyright (C) 1991, 1992, 1993, 1994  Linus Torvalds
   5 *  Swap reorganised 29.12.95, Stephen Tweedie
   6 */
   7
   8#include <linux/mm.h>
   9#include <linux/hugetlb.h>
  10#include <linux/mman.h>
  11#include <linux/slab.h>
  12#include <linux/kernel_stat.h>
  13#include <linux/swap.h>
  14#include <linux/vmalloc.h>
  15#include <linux/pagemap.h>
  16#include <linux/namei.h>
  17#include <linux/shmem_fs.h>
  18#include <linux/blkdev.h>
  19#include <linux/random.h>
  20#include <linux/writeback.h>
  21#include <linux/proc_fs.h>
  22#include <linux/seq_file.h>
  23#include <linux/init.h>
  24#include <linux/ksm.h>
  25#include <linux/rmap.h>
  26#include <linux/security.h>
  27#include <linux/backing-dev.h>
  28#include <linux/mutex.h>
  29#include <linux/capability.h>
  30#include <linux/syscalls.h>
  31#include <linux/memcontrol.h>
  32#include <linux/poll.h>
  33#include <linux/oom.h>
  34#include <linux/frontswap.h>
  35#include <linux/swapfile.h>
  36#include <linux/export.h>
  37
  38#include <asm/pgtable.h>
  39#include <asm/tlbflush.h>
  40#include <linux/swapops.h>
  41#include <linux/page_cgroup.h>
  42
  43static bool swap_count_continued(struct swap_info_struct *, pgoff_t,
  44                                 unsigned char);
  45static void free_swap_count_continuations(struct swap_info_struct *);
  46static sector_t map_swap_entry(swp_entry_t, struct block_device**);
  47
  48DEFINE_SPINLOCK(swap_lock);
  49static unsigned int nr_swapfiles;
  50atomic_long_t nr_swap_pages;
  51/* protected with swap_lock. reading in vm_swap_full() doesn't need lock */
  52long total_swap_pages;
  53static int least_priority;
  54static atomic_t highest_priority_index = ATOMIC_INIT(-1);
  55
  56static const char Bad_file[] = "Bad swap file entry ";
  57static const char Unused_file[] = "Unused swap file entry ";
  58static const char Bad_offset[] = "Bad swap offset entry ";
  59static const char Unused_offset[] = "Unused swap offset entry ";
  60
  61struct swap_list_t swap_list = {-1, -1};
  62
  63struct swap_info_struct *swap_info[MAX_SWAPFILES];
  64
  65static DEFINE_MUTEX(swapon_mutex);
  66
  67static DECLARE_WAIT_QUEUE_HEAD(proc_poll_wait);
  68/* Activity counter to indicate that a swapon or swapoff has occurred */
  69static atomic_t proc_poll_event = ATOMIC_INIT(0);
  70
  71static inline unsigned char swap_count(unsigned char ent)
  72{
  73        return ent & ~SWAP_HAS_CACHE;   /* may include SWAP_HAS_CONT flag */
  74}
  75
  76/* returns 1 if swap entry is freed */
  77static int
  78__try_to_reclaim_swap(struct swap_info_struct *si, unsigned long offset)
  79{
  80        swp_entry_t entry = swp_entry(si->type, offset);
  81        struct page *page;
  82        int ret = 0;
  83
  84        page = find_get_page(swap_address_space(entry), entry.val);
  85        if (!page)
  86                return 0;
  87        /*
  88         * This function is called from scan_swap_map() and it's called
  89         * by vmscan.c at reclaiming pages. So, we hold a lock on a page, here.
  90         * We have to use trylock for avoiding deadlock. This is a special
  91         * case and you should use try_to_free_swap() with explicit lock_page()
  92         * in usual operations.
  93         */
  94        if (trylock_page(page)) {
  95                ret = try_to_free_swap(page);
  96                unlock_page(page);
  97        }
  98        page_cache_release(page);
  99        return ret;
 100}
 101
 102/*
 103 * swapon tell device that all the old swap contents can be discarded,
 104 * to allow the swap device to optimize its wear-levelling.
 105 */
 106static int discard_swap(struct swap_info_struct *si)
 107{
 108        struct swap_extent *se;
 109        sector_t start_block;
 110        sector_t nr_blocks;
 111        int err = 0;
 112
 113        /* Do not discard the swap header page! */
 114        se = &si->first_swap_extent;
 115        start_block = (se->start_block + 1) << (PAGE_SHIFT - 9);
 116        nr_blocks = ((sector_t)se->nr_pages - 1) << (PAGE_SHIFT - 9);
 117        if (nr_blocks) {
 118                err = blkdev_issue_discard(si->bdev, start_block,
 119                                nr_blocks, GFP_KERNEL, 0);
 120                if (err)
 121                        return err;
 122                cond_resched();
 123        }
 124
 125        list_for_each_entry(se, &si->first_swap_extent.list, list) {
 126                start_block = se->start_block << (PAGE_SHIFT - 9);
 127                nr_blocks = (sector_t)se->nr_pages << (PAGE_SHIFT - 9);
 128
 129                err = blkdev_issue_discard(si->bdev, start_block,
 130                                nr_blocks, GFP_KERNEL, 0);
 131                if (err)
 132                        break;
 133
 134                cond_resched();
 135        }
 136        return err;             /* That will often be -EOPNOTSUPP */
 137}
 138
 139/*
 140 * swap allocation tell device that a cluster of swap can now be discarded,
 141 * to allow the swap device to optimize its wear-levelling.
 142 */
 143static void discard_swap_cluster(struct swap_info_struct *si,
 144                                 pgoff_t start_page, pgoff_t nr_pages)
 145{
 146        struct swap_extent *se = si->curr_swap_extent;
 147        int found_extent = 0;
 148
 149        while (nr_pages) {
 150                struct list_head *lh;
 151
 152                if (se->start_page <= start_page &&
 153                    start_page < se->start_page + se->nr_pages) {
 154                        pgoff_t offset = start_page - se->start_page;
 155                        sector_t start_block = se->start_block + offset;
 156                        sector_t nr_blocks = se->nr_pages - offset;
 157
 158                        if (nr_blocks > nr_pages)
 159                                nr_blocks = nr_pages;
 160                        start_page += nr_blocks;
 161                        nr_pages -= nr_blocks;
 162
 163                        if (!found_extent++)
 164                                si->curr_swap_extent = se;
 165
 166                        start_block <<= PAGE_SHIFT - 9;
 167                        nr_blocks <<= PAGE_SHIFT - 9;
 168                        if (blkdev_issue_discard(si->bdev, start_block,
 169                                    nr_blocks, GFP_NOIO, 0))
 170                                break;
 171                }
 172
 173                lh = se->list.next;
 174                se = list_entry(lh, struct swap_extent, list);
 175        }
 176}
 177
 178#define SWAPFILE_CLUSTER        256
 179#define LATENCY_LIMIT           256
 180
 181static inline void cluster_set_flag(struct swap_cluster_info *info,
 182        unsigned int flag)
 183{
 184        info->flags = flag;
 185}
 186
 187static inline unsigned int cluster_count(struct swap_cluster_info *info)
 188{
 189        return info->data;
 190}
 191
 192static inline void cluster_set_count(struct swap_cluster_info *info,
 193                                     unsigned int c)
 194{
 195        info->data = c;
 196}
 197
 198static inline void cluster_set_count_flag(struct swap_cluster_info *info,
 199                                         unsigned int c, unsigned int f)
 200{
 201        info->flags = f;
 202        info->data = c;
 203}
 204
 205static inline unsigned int cluster_next(struct swap_cluster_info *info)
 206{
 207        return info->data;
 208}
 209
 210static inline void cluster_set_next(struct swap_cluster_info *info,
 211                                    unsigned int n)
 212{
 213        info->data = n;
 214}
 215
 216static inline void cluster_set_next_flag(struct swap_cluster_info *info,
 217                                         unsigned int n, unsigned int f)
 218{
 219        info->flags = f;
 220        info->data = n;
 221}
 222
 223static inline bool cluster_is_free(struct swap_cluster_info *info)
 224{
 225        return info->flags & CLUSTER_FLAG_FREE;
 226}
 227
 228static inline bool cluster_is_null(struct swap_cluster_info *info)
 229{
 230        return info->flags & CLUSTER_FLAG_NEXT_NULL;
 231}
 232
 233static inline void cluster_set_null(struct swap_cluster_info *info)
 234{
 235        info->flags = CLUSTER_FLAG_NEXT_NULL;
 236        info->data = 0;
 237}
 238
 239/* Add a cluster to discard list and schedule it to do discard */
 240static void swap_cluster_schedule_discard(struct swap_info_struct *si,
 241                unsigned int idx)
 242{
 243        /*
 244         * If scan_swap_map() can't find a free cluster, it will check
 245         * si->swap_map directly. To make sure the discarding cluster isn't
 246         * taken by scan_swap_map(), mark the swap entries bad (occupied). It
 247         * will be cleared after discard
 248         */
 249        memset(si->swap_map + idx * SWAPFILE_CLUSTER,
 250                        SWAP_MAP_BAD, SWAPFILE_CLUSTER);
 251
 252        if (cluster_is_null(&si->discard_cluster_head)) {
 253                cluster_set_next_flag(&si->discard_cluster_head,
 254                                                idx, 0);
 255                cluster_set_next_flag(&si->discard_cluster_tail,
 256                                                idx, 0);
 257        } else {
 258                unsigned int tail = cluster_next(&si->discard_cluster_tail);
 259                cluster_set_next(&si->cluster_info[tail], idx);
 260                cluster_set_next_flag(&si->discard_cluster_tail,
 261                                                idx, 0);
 262        }
 263
 264        schedule_work(&si->discard_work);
 265}
 266
 267/*
 268 * Doing discard actually. After a cluster discard is finished, the cluster
 269 * will be added to free cluster list. caller should hold si->lock.
 270*/
 271static void swap_do_scheduled_discard(struct swap_info_struct *si)
 272{
 273        struct swap_cluster_info *info;
 274        unsigned int idx;
 275
 276        info = si->cluster_info;
 277
 278        while (!cluster_is_null(&si->discard_cluster_head)) {
 279                idx = cluster_next(&si->discard_cluster_head);
 280
 281                cluster_set_next_flag(&si->discard_cluster_head,
 282                                                cluster_next(&info[idx]), 0);
 283                if (cluster_next(&si->discard_cluster_tail) == idx) {
 284                        cluster_set_null(&si->discard_cluster_head);
 285                        cluster_set_null(&si->discard_cluster_tail);
 286                }
 287                spin_unlock(&si->lock);
 288
 289                discard_swap_cluster(si, idx * SWAPFILE_CLUSTER,
 290                                SWAPFILE_CLUSTER);
 291
 292                spin_lock(&si->lock);
 293                cluster_set_flag(&info[idx], CLUSTER_FLAG_FREE);
 294                if (cluster_is_null(&si->free_cluster_head)) {
 295                        cluster_set_next_flag(&si->free_cluster_head,
 296                                                idx, 0);
 297                        cluster_set_next_flag(&si->free_cluster_tail,
 298                                                idx, 0);
 299                } else {
 300                        unsigned int tail;
 301
 302                        tail = cluster_next(&si->free_cluster_tail);
 303                        cluster_set_next(&info[tail], idx);
 304                        cluster_set_next_flag(&si->free_cluster_tail,
 305                                                idx, 0);
 306                }
 307                memset(si->swap_map + idx * SWAPFILE_CLUSTER,
 308                                0, SWAPFILE_CLUSTER);
 309        }
 310}
 311
 312static void swap_discard_work(struct work_struct *work)
 313{
 314        struct swap_info_struct *si;
 315
 316        si = container_of(work, struct swap_info_struct, discard_work);
 317
 318        spin_lock(&si->lock);
 319        swap_do_scheduled_discard(si);
 320        spin_unlock(&si->lock);
 321}
 322
 323/*
 324 * The cluster corresponding to page_nr will be used. The cluster will be
 325 * removed from free cluster list and its usage counter will be increased.
 326 */
 327static void inc_cluster_info_page(struct swap_info_struct *p,
 328        struct swap_cluster_info *cluster_info, unsigned long page_nr)
 329{
 330        unsigned long idx = page_nr / SWAPFILE_CLUSTER;
 331
 332        if (!cluster_info)
 333                return;
 334        if (cluster_is_free(&cluster_info[idx])) {
 335                VM_BUG_ON(cluster_next(&p->free_cluster_head) != idx);
 336                cluster_set_next_flag(&p->free_cluster_head,
 337                        cluster_next(&cluster_info[idx]), 0);
 338                if (cluster_next(&p->free_cluster_tail) == idx) {
 339                        cluster_set_null(&p->free_cluster_tail);
 340                        cluster_set_null(&p->free_cluster_head);
 341                }
 342                cluster_set_count_flag(&cluster_info[idx], 0, 0);
 343        }
 344
 345        VM_BUG_ON(cluster_count(&cluster_info[idx]) >= SWAPFILE_CLUSTER);
 346        cluster_set_count(&cluster_info[idx],
 347                cluster_count(&cluster_info[idx]) + 1);
 348}
 349
 350/*
 351 * The cluster corresponding to page_nr decreases one usage. If the usage
 352 * counter becomes 0, which means no page in the cluster is in using, we can
 353 * optionally discard the cluster and add it to free cluster list.
 354 */
 355static void dec_cluster_info_page(struct swap_info_struct *p,
 356        struct swap_cluster_info *cluster_info, unsigned long page_nr)
 357{
 358        unsigned long idx = page_nr / SWAPFILE_CLUSTER;
 359
 360        if (!cluster_info)
 361                return;
 362
 363        VM_BUG_ON(cluster_count(&cluster_info[idx]) == 0);
 364        cluster_set_count(&cluster_info[idx],
 365                cluster_count(&cluster_info[idx]) - 1);
 366
 367        if (cluster_count(&cluster_info[idx]) == 0) {
 368                /*
 369                 * If the swap is discardable, prepare discard the cluster
 370                 * instead of free it immediately. The cluster will be freed
 371                 * after discard.
 372                 */
 373                if ((p->flags & (SWP_WRITEOK | SWP_PAGE_DISCARD)) ==
 374                                 (SWP_WRITEOK | SWP_PAGE_DISCARD)) {
 375                        swap_cluster_schedule_discard(p, idx);
 376                        return;
 377                }
 378
 379                cluster_set_flag(&cluster_info[idx], CLUSTER_FLAG_FREE);
 380                if (cluster_is_null(&p->free_cluster_head)) {
 381                        cluster_set_next_flag(&p->free_cluster_head, idx, 0);
 382                        cluster_set_next_flag(&p->free_cluster_tail, idx, 0);
 383                } else {
 384                        unsigned int tail = cluster_next(&p->free_cluster_tail);
 385                        cluster_set_next(&cluster_info[tail], idx);
 386                        cluster_set_next_flag(&p->free_cluster_tail, idx, 0);
 387                }
 388        }
 389}
 390
 391/*
 392 * It's possible scan_swap_map() uses a free cluster in the middle of free
 393 * cluster list. Avoiding such abuse to avoid list corruption.
 394 */
 395static bool
 396scan_swap_map_ssd_cluster_conflict(struct swap_info_struct *si,
 397        unsigned long offset)
 398{
 399        struct percpu_cluster *percpu_cluster;
 400        bool conflict;
 401
 402        offset /= SWAPFILE_CLUSTER;
 403        conflict = !cluster_is_null(&si->free_cluster_head) &&
 404                offset != cluster_next(&si->free_cluster_head) &&
 405                cluster_is_free(&si->cluster_info[offset]);
 406
 407        if (!conflict)
 408                return false;
 409
 410        percpu_cluster = this_cpu_ptr(si->percpu_cluster);
 411        cluster_set_null(&percpu_cluster->index);
 412        return true;
 413}
 414
 415/*
 416 * Try to get a swap entry from current cpu's swap entry pool (a cluster). This
 417 * might involve allocating a new cluster for current CPU too.
 418 */
 419static void scan_swap_map_try_ssd_cluster(struct swap_info_struct *si,
 420        unsigned long *offset, unsigned long *scan_base)
 421{
 422        struct percpu_cluster *cluster;
 423        bool found_free;
 424        unsigned long tmp;
 425
 426new_cluster:
 427        cluster = this_cpu_ptr(si->percpu_cluster);
 428        if (cluster_is_null(&cluster->index)) {
 429                if (!cluster_is_null(&si->free_cluster_head)) {
 430                        cluster->index = si->free_cluster_head;
 431                        cluster->next = cluster_next(&cluster->index) *
 432                                        SWAPFILE_CLUSTER;
 433                } else if (!cluster_is_null(&si->discard_cluster_head)) {
 434                        /*
 435                         * we don't have free cluster but have some clusters in
 436                         * discarding, do discard now and reclaim them
 437                         */
 438                        swap_do_scheduled_discard(si);
 439                        *scan_base = *offset = si->cluster_next;
 440                        goto new_cluster;
 441                } else
 442                        return;
 443        }
 444
 445        found_free = false;
 446
 447        /*
 448         * Other CPUs can use our cluster if they can't find a free cluster,
 449         * check if there is still free entry in the cluster
 450         */
 451        tmp = cluster->next;
 452        while (tmp < si->max && tmp < (cluster_next(&cluster->index) + 1) *
 453               SWAPFILE_CLUSTER) {
 454                if (!si->swap_map[tmp]) {
 455                        found_free = true;
 456                        break;
 457                }
 458                tmp++;
 459        }
 460        if (!found_free) {
 461                cluster_set_null(&cluster->index);
 462                goto new_cluster;
 463        }
 464        cluster->next = tmp + 1;
 465        *offset = tmp;
 466        *scan_base = tmp;
 467}
 468
 469static unsigned long scan_swap_map(struct swap_info_struct *si,
 470                                   unsigned char usage)
 471{
 472        unsigned long offset;
 473        unsigned long scan_base;
 474        unsigned long last_in_cluster = 0;
 475        int latency_ration = LATENCY_LIMIT;
 476
 477        /*
 478         * We try to cluster swap pages by allocating them sequentially
 479         * in swap.  Once we've allocated SWAPFILE_CLUSTER pages this
 480         * way, however, we resort to first-free allocation, starting
 481         * a new cluster.  This prevents us from scattering swap pages
 482         * all over the entire swap partition, so that we reduce
 483         * overall disk seek times between swap pages.  -- sct
 484         * But we do now try to find an empty cluster.  -Andrea
 485         * And we let swap pages go all over an SSD partition.  Hugh
 486         */
 487
 488        si->flags += SWP_SCANNING;
 489        scan_base = offset = si->cluster_next;
 490
 491        /* SSD algorithm */
 492        if (si->cluster_info) {
 493                scan_swap_map_try_ssd_cluster(si, &offset, &scan_base);
 494                goto checks;
 495        }
 496
 497        if (unlikely(!si->cluster_nr--)) {
 498                if (si->pages - si->inuse_pages < SWAPFILE_CLUSTER) {
 499                        si->cluster_nr = SWAPFILE_CLUSTER - 1;
 500                        goto checks;
 501                }
 502
 503                spin_unlock(&si->lock);
 504
 505                /*
 506                 * If seek is expensive, start searching for new cluster from
 507                 * start of partition, to minimize the span of allocated swap.
 508                 * But if seek is cheap, search from our current position, so
 509                 * that swap is allocated from all over the partition: if the
 510                 * Flash Translation Layer only remaps within limited zones,
 511                 * we don't want to wear out the first zone too quickly.
 512                 */
 513                if (!(si->flags & SWP_SOLIDSTATE))
 514                        scan_base = offset = si->lowest_bit;
 515                last_in_cluster = offset + SWAPFILE_CLUSTER - 1;
 516
 517                /* Locate the first empty (unaligned) cluster */
 518                for (; last_in_cluster <= si->highest_bit; offset++) {
 519                        if (si->swap_map[offset])
 520                                last_in_cluster = offset + SWAPFILE_CLUSTER;
 521                        else if (offset == last_in_cluster) {
 522                                spin_lock(&si->lock);
 523                                offset -= SWAPFILE_CLUSTER - 1;
 524                                si->cluster_next = offset;
 525                                si->cluster_nr = SWAPFILE_CLUSTER - 1;
 526                                goto checks;
 527                        }
 528                        if (unlikely(--latency_ration < 0)) {
 529                                cond_resched();
 530                                latency_ration = LATENCY_LIMIT;
 531                        }
 532                }
 533
 534                offset = si->lowest_bit;
 535                last_in_cluster = offset + SWAPFILE_CLUSTER - 1;
 536
 537                /* Locate the first empty (unaligned) cluster */
 538                for (; last_in_cluster < scan_base; offset++) {
 539                        if (si->swap_map[offset])
 540                                last_in_cluster = offset + SWAPFILE_CLUSTER;
 541                        else if (offset == last_in_cluster) {
 542                                spin_lock(&si->lock);
 543                                offset -= SWAPFILE_CLUSTER - 1;
 544                                si->cluster_next = offset;
 545                                si->cluster_nr = SWAPFILE_CLUSTER - 1;
 546                                goto checks;
 547                        }
 548                        if (unlikely(--latency_ration < 0)) {
 549                                cond_resched();
 550                                latency_ration = LATENCY_LIMIT;
 551                        }
 552                }
 553
 554                offset = scan_base;
 555                spin_lock(&si->lock);
 556                si->cluster_nr = SWAPFILE_CLUSTER - 1;
 557        }
 558
 559checks:
 560        if (si->cluster_info) {
 561                while (scan_swap_map_ssd_cluster_conflict(si, offset))
 562                        scan_swap_map_try_ssd_cluster(si, &offset, &scan_base);
 563        }
 564        if (!(si->flags & SWP_WRITEOK))
 565                goto no_page;
 566        if (!si->highest_bit)
 567                goto no_page;
 568        if (offset > si->highest_bit)
 569                scan_base = offset = si->lowest_bit;
 570
 571        /* reuse swap entry of cache-only swap if not busy. */
 572        if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) {
 573                int swap_was_freed;
 574                spin_unlock(&si->lock);
 575                swap_was_freed = __try_to_reclaim_swap(si, offset);
 576                spin_lock(&si->lock);
 577                /* entry was freed successfully, try to use this again */
 578                if (swap_was_freed)
 579                        goto checks;
 580                goto scan; /* check next one */
 581        }
 582
 583        if (si->swap_map[offset])
 584                goto scan;
 585
 586        if (offset == si->lowest_bit)
 587                si->lowest_bit++;
 588        if (offset == si->highest_bit)
 589                si->highest_bit--;
 590        si->inuse_pages++;
 591        if (si->inuse_pages == si->pages) {
 592                si->lowest_bit = si->max;
 593                si->highest_bit = 0;
 594        }
 595        si->swap_map[offset] = usage;
 596        inc_cluster_info_page(si, si->cluster_info, offset);
 597        si->cluster_next = offset + 1;
 598        si->flags -= SWP_SCANNING;
 599
 600        return offset;
 601
 602scan:
 603        spin_unlock(&si->lock);
 604        while (++offset <= si->highest_bit) {
 605                if (!si->swap_map[offset]) {
 606                        spin_lock(&si->lock);
 607                        goto checks;
 608                }
 609                if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) {
 610                        spin_lock(&si->lock);
 611                        goto checks;
 612                }
 613                if (unlikely(--latency_ration < 0)) {
 614                        cond_resched();
 615                        latency_ration = LATENCY_LIMIT;
 616                }
 617        }
 618        offset = si->lowest_bit;
 619        while (++offset < scan_base) {
 620                if (!si->swap_map[offset]) {
 621                        spin_lock(&si->lock);
 622                        goto checks;
 623                }
 624                if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) {
 625                        spin_lock(&si->lock);
 626                        goto checks;
 627                }
 628                if (unlikely(--latency_ration < 0)) {
 629                        cond_resched();
 630                        latency_ration = LATENCY_LIMIT;
 631                }
 632        }
 633        spin_lock(&si->lock);
 634
 635no_page:
 636        si->flags -= SWP_SCANNING;
 637        return 0;
 638}
 639
 640swp_entry_t get_swap_page(void)
 641{
 642        struct swap_info_struct *si;
 643        pgoff_t offset;
 644        int type, next;
 645        int wrapped = 0;
 646        int hp_index;
 647
 648        spin_lock(&swap_lock);
 649        if (atomic_long_read(&nr_swap_pages) <= 0)
 650                goto noswap;
 651        atomic_long_dec(&nr_swap_pages);
 652
 653        for (type = swap_list.next; type >= 0 && wrapped < 2; type = next) {
 654                hp_index = atomic_xchg(&highest_priority_index, -1);
 655                /*
 656                 * highest_priority_index records current highest priority swap
 657                 * type which just frees swap entries. If its priority is
 658                 * higher than that of swap_list.next swap type, we use it.  It
 659                 * isn't protected by swap_lock, so it can be an invalid value
 660                 * if the corresponding swap type is swapoff. We double check
 661                 * the flags here. It's even possible the swap type is swapoff
 662                 * and swapon again and its priority is changed. In such rare
 663                 * case, low prority swap type might be used, but eventually
 664                 * high priority swap will be used after several rounds of
 665                 * swap.
 666                 */
 667                if (hp_index != -1 && hp_index != type &&
 668                    swap_info[type]->prio < swap_info[hp_index]->prio &&
 669                    (swap_info[hp_index]->flags & SWP_WRITEOK)) {
 670                        type = hp_index;
 671                        swap_list.next = type;
 672                }
 673
 674                si = swap_info[type];
 675                next = si->next;
 676                if (next < 0 ||
 677                    (!wrapped && si->prio != swap_info[next]->prio)) {
 678                        next = swap_list.head;
 679                        wrapped++;
 680                }
 681
 682                spin_lock(&si->lock);
 683                if (!si->highest_bit) {
 684                        spin_unlock(&si->lock);
 685                        continue;
 686                }
 687                if (!(si->flags & SWP_WRITEOK)) {
 688                        spin_unlock(&si->lock);
 689                        continue;
 690                }
 691
 692                swap_list.next = next;
 693
 694                spin_unlock(&swap_lock);
 695                /* This is called for allocating swap entry for cache */
 696                offset = scan_swap_map(si, SWAP_HAS_CACHE);
 697                spin_unlock(&si->lock);
 698                if (offset)
 699                        return swp_entry(type, offset);
 700                spin_lock(&swap_lock);
 701                next = swap_list.next;
 702        }
 703
 704        atomic_long_inc(&nr_swap_pages);
 705noswap:
 706        spin_unlock(&swap_lock);
 707        return (swp_entry_t) {0};
 708}
 709
 710/* The only caller of this function is now susupend routine */
 711swp_entry_t get_swap_page_of_type(int type)
 712{
 713        struct swap_info_struct *si;
 714        pgoff_t offset;
 715
 716        si = swap_info[type];
 717        spin_lock(&si->lock);
 718        if (si && (si->flags & SWP_WRITEOK)) {
 719                atomic_long_dec(&nr_swap_pages);
 720                /* This is called for allocating swap entry, not cache */
 721                offset = scan_swap_map(si, 1);
 722                if (offset) {
 723                        spin_unlock(&si->lock);
 724                        return swp_entry(type, offset);
 725                }
 726                atomic_long_inc(&nr_swap_pages);
 727        }
 728        spin_unlock(&si->lock);
 729        return (swp_entry_t) {0};
 730}
 731
 732static struct swap_info_struct *swap_info_get(swp_entry_t entry)
 733{
 734        struct swap_info_struct *p;
 735        unsigned long offset, type;
 736
 737        if (!entry.val)
 738                goto out;
 739        type = swp_type(entry);
 740        if (type >= nr_swapfiles)
 741                goto bad_nofile;
 742        p = swap_info[type];
 743        if (!(p->flags & SWP_USED))
 744                goto bad_device;
 745        offset = swp_offset(entry);
 746        if (offset >= p->max)
 747                goto bad_offset;
 748        if (!p->swap_map[offset])
 749                goto bad_free;
 750        spin_lock(&p->lock);
 751        return p;
 752
 753bad_free:
 754        pr_err("swap_free: %s%08lx\n", Unused_offset, entry.val);
 755        goto out;
 756bad_offset:
 757        pr_err("swap_free: %s%08lx\n", Bad_offset, entry.val);
 758        goto out;
 759bad_device:
 760        pr_err("swap_free: %s%08lx\n", Unused_file, entry.val);
 761        goto out;
 762bad_nofile:
 763        pr_err("swap_free: %s%08lx\n", Bad_file, entry.val);
 764out:
 765        return NULL;
 766}
 767
 768/*
 769 * This swap type frees swap entry, check if it is the highest priority swap
 770 * type which just frees swap entry. get_swap_page() uses
 771 * highest_priority_index to search highest priority swap type. The
 772 * swap_info_struct.lock can't protect us if there are multiple swap types
 773 * active, so we use atomic_cmpxchg.
 774 */
 775static void set_highest_priority_index(int type)
 776{
 777        int old_hp_index, new_hp_index;
 778
 779        do {
 780                old_hp_index = atomic_read(&highest_priority_index);
 781                if (old_hp_index != -1 &&
 782                        swap_info[old_hp_index]->prio >= swap_info[type]->prio)
 783                        break;
 784                new_hp_index = type;
 785        } while (atomic_cmpxchg(&highest_priority_index,
 786                old_hp_index, new_hp_index) != old_hp_index);
 787}
 788
 789static unsigned char swap_entry_free(struct swap_info_struct *p,
 790                                     swp_entry_t entry, unsigned char usage)
 791{
 792        unsigned long offset = swp_offset(entry);
 793        unsigned char count;
 794        unsigned char has_cache;
 795
 796        count = p->swap_map[offset];
 797        has_cache = count & SWAP_HAS_CACHE;
 798        count &= ~SWAP_HAS_CACHE;
 799
 800        if (usage == SWAP_HAS_CACHE) {
 801                VM_BUG_ON(!has_cache);
 802                has_cache = 0;
 803        } else if (count == SWAP_MAP_SHMEM) {
 804                /*
 805                 * Or we could insist on shmem.c using a special
 806                 * swap_shmem_free() and free_shmem_swap_and_cache()...
 807                 */
 808                count = 0;
 809        } else if ((count & ~COUNT_CONTINUED) <= SWAP_MAP_MAX) {
 810                if (count == COUNT_CONTINUED) {
 811                        if (swap_count_continued(p, offset, count))
 812                                count = SWAP_MAP_MAX | COUNT_CONTINUED;
 813                        else
 814                                count = SWAP_MAP_MAX;
 815                } else
 816                        count--;
 817        }
 818
 819        if (!count)
 820                mem_cgroup_uncharge_swap(entry);
 821
 822        usage = count | has_cache;
 823        p->swap_map[offset] = usage;
 824
 825        /* free if no reference */
 826        if (!usage) {
 827                dec_cluster_info_page(p, p->cluster_info, offset);
 828                if (offset < p->lowest_bit)
 829                        p->lowest_bit = offset;
 830                if (offset > p->highest_bit)
 831                        p->highest_bit = offset;
 832                set_highest_priority_index(p->type);
 833                atomic_long_inc(&nr_swap_pages);
 834                p->inuse_pages--;
 835                frontswap_invalidate_page(p->type, offset);
 836                if (p->flags & SWP_BLKDEV) {
 837                        struct gendisk *disk = p->bdev->bd_disk;
 838                        if (disk->fops->swap_slot_free_notify)
 839                                disk->fops->swap_slot_free_notify(p->bdev,
 840                                                                  offset);
 841                }
 842        }
 843
 844        return usage;
 845}
 846
 847/*
 848 * Caller has made sure that the swapdevice corresponding to entry
 849 * is still around or has not been recycled.
 850 */
 851void swap_free(swp_entry_t entry)
 852{
 853        struct swap_info_struct *p;
 854
 855        p = swap_info_get(entry);
 856        if (p) {
 857                swap_entry_free(p, entry, 1);
 858                spin_unlock(&p->lock);
 859        }
 860}
 861
 862/*
 863 * Called after dropping swapcache to decrease refcnt to swap entries.
 864 */
 865void swapcache_free(swp_entry_t entry, struct page *page)
 866{
 867        struct swap_info_struct *p;
 868        unsigned char count;
 869
 870        p = swap_info_get(entry);
 871        if (p) {
 872                count = swap_entry_free(p, entry, SWAP_HAS_CACHE);
 873                if (page)
 874                        mem_cgroup_uncharge_swapcache(page, entry, count != 0);
 875                spin_unlock(&p->lock);
 876        }
 877}
 878
 879/*
 880 * How many references to page are currently swapped out?
 881 * This does not give an exact answer when swap count is continued,
 882 * but does include the high COUNT_CONTINUED flag to allow for that.
 883 */
 884int page_swapcount(struct page *page)
 885{
 886        int count = 0;
 887        struct swap_info_struct *p;
 888        swp_entry_t entry;
 889
 890        entry.val = page_private(page);
 891        p = swap_info_get(entry);
 892        if (p) {
 893                count = swap_count(p->swap_map[swp_offset(entry)]);
 894                spin_unlock(&p->lock);
 895        }
 896        return count;
 897}
 898
 899/*
 900 * We can write to an anon page without COW if there are no other references
 901 * to it.  And as a side-effect, free up its swap: because the old content
 902 * on disk will never be read, and seeking back there to write new content
 903 * later would only waste time away from clustering.
 904 */
 905int reuse_swap_page(struct page *page)
 906{
 907        int count;
 908
 909        VM_BUG_ON(!PageLocked(page));
 910        if (unlikely(PageKsm(page)))
 911                return 0;
 912        count = page_mapcount(page);
 913        if (count <= 1 && PageSwapCache(page)) {
 914                count += page_swapcount(page);
 915                if (count == 1 && !PageWriteback(page)) {
 916                        delete_from_swap_cache(page);
 917                        SetPageDirty(page);
 918                }
 919        }
 920        return count <= 1;
 921}
 922
 923/*
 924 * If swap is getting full, or if there are no more mappings of this page,
 925 * then try_to_free_swap is called to free its swap space.
 926 */
 927int try_to_free_swap(struct page *page)
 928{
 929        VM_BUG_ON(!PageLocked(page));
 930
 931        if (!PageSwapCache(page))
 932                return 0;
 933        if (PageWriteback(page))
 934                return 0;
 935        if (page_swapcount(page))
 936                return 0;
 937
 938        /*
 939         * Once hibernation has begun to create its image of memory,
 940         * there's a danger that one of the calls to try_to_free_swap()
 941         * - most probably a call from __try_to_reclaim_swap() while
 942         * hibernation is allocating its own swap pages for the image,
 943         * but conceivably even a call from memory reclaim - will free
 944         * the swap from a page which has already been recorded in the
 945         * image as a clean swapcache page, and then reuse its swap for
 946         * another page of the image.  On waking from hibernation, the
 947         * original page might be freed under memory pressure, then
 948         * later read back in from swap, now with the wrong data.
 949         *
 950         * Hibration suspends storage while it is writing the image
 951         * to disk so check that here.
 952         */
 953        if (pm_suspended_storage())
 954                return 0;
 955
 956        delete_from_swap_cache(page);
 957        SetPageDirty(page);
 958        return 1;
 959}
 960
 961/*
 962 * Free the swap entry like above, but also try to
 963 * free the page cache entry if it is the last user.
 964 */
 965int free_swap_and_cache(swp_entry_t entry)
 966{
 967        struct swap_info_struct *p;
 968        struct page *page = NULL;
 969
 970        if (non_swap_entry(entry))
 971                return 1;
 972
 973        p = swap_info_get(entry);
 974        if (p) {
 975                if (swap_entry_free(p, entry, 1) == SWAP_HAS_CACHE) {
 976                        page = find_get_page(swap_address_space(entry),
 977                                                entry.val);
 978                        if (page && !trylock_page(page)) {
 979                                page_cache_release(page);
 980                                page = NULL;
 981                        }
 982                }
 983                spin_unlock(&p->lock);
 984        }
 985        if (page) {
 986                /*
 987                 * Not mapped elsewhere, or swap space full? Free it!
 988                 * Also recheck PageSwapCache now page is locked (above).
 989                 */
 990                if (PageSwapCache(page) && !PageWriteback(page) &&
 991                                (!page_mapped(page) || vm_swap_full())) {
 992                        delete_from_swap_cache(page);
 993                        SetPageDirty(page);
 994                }
 995                unlock_page(page);
 996                page_cache_release(page);
 997        }
 998        return p != NULL;
 999}
1000
1001#ifdef CONFIG_HIBERNATION
1002/*
1003 * Find the swap type that corresponds to given device (if any).
1004 *
1005 * @offset - number of the PAGE_SIZE-sized block of the device, starting
1006 * from 0, in which the swap header is expected to be located.
1007 *
1008 * This is needed for the suspend to disk (aka swsusp).
1009 */
1010int swap_type_of(dev_t device, sector_t offset, struct block_device **bdev_p)
1011{
1012        struct block_device *bdev = NULL;
1013        int type;
1014
1015        if (device)
1016                bdev = bdget(device);
1017
1018        spin_lock(&swap_lock);
1019        for (type = 0; type < nr_swapfiles; type++) {
1020                struct swap_info_struct *sis = swap_info[type];
1021
1022                if (!(sis->flags & SWP_WRITEOK))
1023                        continue;
1024
1025                if (!bdev) {
1026                        if (bdev_p)
1027                                *bdev_p = bdgrab(sis->bdev);
1028
1029                        spin_unlock(&swap_lock);
1030                        return type;
1031                }
1032                if (bdev == sis->bdev) {
1033                        struct swap_extent *se = &sis->first_swap_extent;
1034
1035                        if (se->start_block == offset) {
1036                                if (bdev_p)
1037                                        *bdev_p = bdgrab(sis->bdev);
1038
1039                                spin_unlock(&swap_lock);
1040                                bdput(bdev);
1041                                return type;
1042                        }
1043                }
1044        }
1045        spin_unlock(&swap_lock);
1046        if (bdev)
1047                bdput(bdev);
1048
1049        return -ENODEV;
1050}
1051
1052/*
1053 * Get the (PAGE_SIZE) block corresponding to given offset on the swapdev
1054 * corresponding to given index in swap_info (swap type).
1055 */
1056sector_t swapdev_block(int type, pgoff_t offset)
1057{
1058        struct block_device *bdev;
1059
1060        if ((unsigned int)type >= nr_swapfiles)
1061                return 0;
1062        if (!(swap_info[type]->flags & SWP_WRITEOK))
1063                return 0;
1064        return map_swap_entry(swp_entry(type, offset), &bdev);
1065}
1066
1067/*
1068 * Return either the total number of swap pages of given type, or the number
1069 * of free pages of that type (depending on @free)
1070 *
1071 * This is needed for software suspend
1072 */
1073unsigned int count_swap_pages(int type, int free)
1074{
1075        unsigned int n = 0;
1076
1077        spin_lock(&swap_lock);
1078        if ((unsigned int)type < nr_swapfiles) {
1079                struct swap_info_struct *sis = swap_info[type];
1080
1081                spin_lock(&sis->lock);
1082                if (sis->flags & SWP_WRITEOK) {
1083                        n = sis->pages;
1084                        if (free)
1085                                n -= sis->inuse_pages;
1086                }
1087                spin_unlock(&sis->lock);
1088        }
1089        spin_unlock(&swap_lock);
1090        return n;
1091}
1092#endif /* CONFIG_HIBERNATION */
1093
1094static inline int maybe_same_pte(pte_t pte, pte_t swp_pte)
1095{
1096#ifdef CONFIG_MEM_SOFT_DIRTY
1097        /*
1098         * When pte keeps soft dirty bit the pte generated
1099         * from swap entry does not has it, still it's same
1100         * pte from logical point of view.
1101         */
1102        pte_t swp_pte_dirty = pte_swp_mksoft_dirty(swp_pte);
1103        return pte_same(pte, swp_pte) || pte_same(pte, swp_pte_dirty);
1104#else
1105        return pte_same(pte, swp_pte);
1106#endif
1107}
1108
1109/*
1110 * No need to decide whether this PTE shares the swap entry with others,
1111 * just let do_wp_page work it out if a write is requested later - to
1112 * force COW, vm_page_prot omits write permission from any private vma.
1113 */
1114static int unuse_pte(struct vm_area_struct *vma, pmd_t *pmd,
1115                unsigned long addr, swp_entry_t entry, struct page *page)
1116{
1117        struct page *swapcache;
1118        struct mem_cgroup *memcg;
1119        spinlock_t *ptl;
1120        pte_t *pte;
1121        int ret = 1;
1122
1123        swapcache = page;
1124        page = ksm_might_need_to_copy(page, vma, addr);
1125        if (unlikely(!page))
1126                return -ENOMEM;
1127
1128        if (mem_cgroup_try_charge_swapin(vma->vm_mm, page,
1129                                         GFP_KERNEL, &memcg)) {
1130                ret = -ENOMEM;
1131                goto out_nolock;
1132        }
1133
1134        pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
1135        if (unlikely(!maybe_same_pte(*pte, swp_entry_to_pte(entry)))) {
1136                mem_cgroup_cancel_charge_swapin(memcg);
1137                ret = 0;
1138                goto out;
1139        }
1140
1141        dec_mm_counter(vma->vm_mm, MM_SWAPENTS);
1142        inc_mm_counter(vma->vm_mm, MM_ANONPAGES);
1143        get_page(page);
1144        set_pte_at(vma->vm_mm, addr, pte,
1145                   pte_mkold(mk_pte(page, vma->vm_page_prot)));
1146        if (page == swapcache)
1147                page_add_anon_rmap(page, vma, addr);
1148        else /* ksm created a completely new copy */
1149                page_add_new_anon_rmap(page, vma, addr);
1150        mem_cgroup_commit_charge_swapin(page, memcg);
1151        swap_free(entry);
1152        /*
1153         * Move the page to the active list so it is not
1154         * immediately swapped out again after swapon.
1155         */
1156        activate_page(page);
1157out:
1158        pte_unmap_unlock(pte, ptl);
1159out_nolock:
1160        if (page != swapcache) {
1161                unlock_page(page);
1162                put_page(page);
1163        }
1164        return ret;
1165}
1166
1167static int unuse_pte_range(struct vm_area_struct *vma, pmd_t *pmd,
1168                                unsigned long addr, unsigned long end,
1169                                swp_entry_t entry, struct page *page)
1170{
1171        pte_t swp_pte = swp_entry_to_pte(entry);
1172        pte_t *pte;
1173        int ret = 0;
1174
1175        /*
1176         * We don't actually need pte lock while scanning for swp_pte: since
1177         * we hold page lock and mmap_sem, swp_pte cannot be inserted into the
1178         * page table while we're scanning; though it could get zapped, and on
1179         * some architectures (e.g. x86_32 with PAE) we might catch a glimpse
1180         * of unmatched parts which look like swp_pte, so unuse_pte must
1181         * recheck under pte lock.  Scanning without pte lock lets it be
1182         * preemptible whenever CONFIG_PREEMPT but not CONFIG_HIGHPTE.
1183         */
1184        pte = pte_offset_map(pmd, addr);
1185        do {
1186                /*
1187                 * swapoff spends a _lot_ of time in this loop!
1188                 * Test inline before going to call unuse_pte.
1189                 */
1190                if (unlikely(maybe_same_pte(*pte, swp_pte))) {
1191                        pte_unmap(pte);
1192                        ret = unuse_pte(vma, pmd, addr, entry, page);
1193                        if (ret)
1194                                goto out;
1195                        pte = pte_offset_map(pmd, addr);
1196                }
1197        } while (pte++, addr += PAGE_SIZE, addr != end);
1198        pte_unmap(pte - 1);
1199out:
1200        return ret;
1201}
1202
1203static inline int unuse_pmd_range(struct vm_area_struct *vma, pud_t *pud,
1204                                unsigned long addr, unsigned long end,
1205                                swp_entry_t entry, struct page *page)
1206{
1207        pmd_t *pmd;
1208        unsigned long next;
1209        int ret;
1210
1211        pmd = pmd_offset(pud, addr);
1212        do {
1213                next = pmd_addr_end(addr, end);
1214                if (pmd_none_or_trans_huge_or_clear_bad(pmd))
1215                        continue;
1216                ret = unuse_pte_range(vma, pmd, addr, next, entry, page);
1217                if (ret)
1218                        return ret;
1219        } while (pmd++, addr = next, addr != end);
1220        return 0;
1221}
1222
1223static inline int unuse_pud_range(struct vm_area_struct *vma, pgd_t *pgd,
1224                                unsigned long addr, unsigned long end,
1225                                swp_entry_t entry, struct page *page)
1226{
1227        pud_t *pud;
1228        unsigned long next;
1229        int ret;
1230
1231        pud = pud_offset(pgd, addr);
1232        do {
1233                next = pud_addr_end(addr, end);
1234                if (pud_none_or_clear_bad(pud))
1235                        continue;
1236                ret = unuse_pmd_range(vma, pud, addr, next, entry, page);
1237                if (ret)
1238                        return ret;
1239        } while (pud++, addr = next, addr != end);
1240        return 0;
1241}
1242
1243static int unuse_vma(struct vm_area_struct *vma,
1244                                swp_entry_t entry, struct page *page)
1245{
1246        pgd_t *pgd;
1247        unsigned long addr, end, next;
1248        int ret;
1249
1250        if (page_anon_vma(page)) {
1251                addr = page_address_in_vma(page, vma);
1252                if (addr == -EFAULT)
1253                        return 0;
1254                else
1255                        end = addr + PAGE_SIZE;
1256        } else {
1257                addr = vma->vm_start;
1258                end = vma->vm_end;
1259        }
1260
1261        pgd = pgd_offset(vma->vm_mm, addr);
1262        do {
1263                next = pgd_addr_end(addr, end);
1264                if (pgd_none_or_clear_bad(pgd))
1265                        continue;
1266                ret = unuse_pud_range(vma, pgd, addr, next, entry, page);
1267                if (ret)
1268                        return ret;
1269        } while (pgd++, addr = next, addr != end);
1270        return 0;
1271}
1272
1273static int unuse_mm(struct mm_struct *mm,
1274                                swp_entry_t entry, struct page *page)
1275{
1276        struct vm_area_struct *vma;
1277        int ret = 0;
1278
1279        if (!down_read_trylock(&mm->mmap_sem)) {
1280                /*
1281                 * Activate page so shrink_inactive_list is unlikely to unmap
1282                 * its ptes while lock is dropped, so swapoff can make progress.
1283                 */
1284                activate_page(page);
1285                unlock_page(page);
1286                down_read(&mm->mmap_sem);
1287                lock_page(page);
1288        }
1289        for (vma = mm->mmap; vma; vma = vma->vm_next) {
1290                if (vma->anon_vma && (ret = unuse_vma(vma, entry, page)))
1291                        break;
1292        }
1293        up_read(&mm->mmap_sem);
1294        return (ret < 0)? ret: 0;
1295}
1296
1297/*
1298 * Scan swap_map (or frontswap_map if frontswap parameter is true)
1299 * from current position to next entry still in use.
1300 * Recycle to start on reaching the end, returning 0 when empty.
1301 */
1302static unsigned int find_next_to_unuse(struct swap_info_struct *si,
1303                                        unsigned int prev, bool frontswap)
1304{
1305        unsigned int max = si->max;
1306        unsigned int i = prev;
1307        unsigned char count;
1308
1309        /*
1310         * No need for swap_lock here: we're just looking
1311         * for whether an entry is in use, not modifying it; false
1312         * hits are okay, and sys_swapoff() has already prevented new
1313         * allocations from this area (while holding swap_lock).
1314         */
1315        for (;;) {
1316                if (++i >= max) {
1317                        if (!prev) {
1318                                i = 0;
1319                                break;
1320                        }
1321                        /*
1322                         * No entries in use at top of swap_map,
1323                         * loop back to start and recheck there.
1324                         */
1325                        max = prev + 1;
1326                        prev = 0;
1327                        i = 1;
1328                }
1329                if (frontswap) {
1330                        if (frontswap_test(si, i))
1331                                break;
1332                        else
1333                                continue;
1334                }
1335                count = ACCESS_ONCE(si->swap_map[i]);
1336                if (count && swap_count(count) != SWAP_MAP_BAD)
1337                        break;
1338        }
1339        return i;
1340}
1341
1342/*
1343 * We completely avoid races by reading each swap page in advance,
1344 * and then search for the process using it.  All the necessary
1345 * page table adjustments can then be made atomically.
1346 *
1347 * if the boolean frontswap is true, only unuse pages_to_unuse pages;
1348 * pages_to_unuse==0 means all pages; ignored if frontswap is false
1349 */
1350int try_to_unuse(unsigned int type, bool frontswap,
1351                 unsigned long pages_to_unuse)
1352{
1353        struct swap_info_struct *si = swap_info[type];
1354        struct mm_struct *start_mm;
1355        volatile unsigned char *swap_map; /* swap_map is accessed without
1356                                           * locking. Mark it as volatile
1357                                           * to prevent compiler doing
1358                                           * something odd.
1359                                           */
1360        unsigned char swcount;
1361        struct page *page;
1362        swp_entry_t entry;
1363        unsigned int i = 0;
1364        int retval = 0;
1365
1366        /*
1367         * When searching mms for an entry, a good strategy is to
1368         * start at the first mm we freed the previous entry from
1369         * (though actually we don't notice whether we or coincidence
1370         * freed the entry).  Initialize this start_mm with a hold.
1371         *
1372         * A simpler strategy would be to start at the last mm we
1373         * freed the previous entry from; but that would take less
1374         * advantage of mmlist ordering, which clusters forked mms
1375         * together, child after parent.  If we race with dup_mmap(), we
1376         * prefer to resolve parent before child, lest we miss entries
1377         * duplicated after we scanned child: using last mm would invert
1378         * that.
1379         */
1380        start_mm = &init_mm;
1381        atomic_inc(&init_mm.mm_users);
1382
1383        /*
1384         * Keep on scanning until all entries have gone.  Usually,
1385         * one pass through swap_map is enough, but not necessarily:
1386         * there are races when an instance of an entry might be missed.
1387         */
1388        while ((i = find_next_to_unuse(si, i, frontswap)) != 0) {
1389                if (signal_pending(current)) {
1390                        retval = -EINTR;
1391                        break;
1392                }
1393
1394                /*
1395                 * Get a page for the entry, using the existing swap
1396                 * cache page if there is one.  Otherwise, get a clean
1397                 * page and read the swap into it.
1398                 */
1399                swap_map = &si->swap_map[i];
1400                entry = swp_entry(type, i);
1401                page = read_swap_cache_async(entry,
1402                                        GFP_HIGHUSER_MOVABLE, NULL, 0);
1403                if (!page) {
1404                        /*
1405                         * Either swap_duplicate() failed because entry
1406                         * has been freed independently, and will not be
1407                         * reused since sys_swapoff() already disabled
1408                         * allocation from here, or alloc_page() failed.
1409                         */
1410                        swcount = *swap_map;
1411                        /*
1412                         * We don't hold lock here, so the swap entry could be
1413                         * SWAP_MAP_BAD (when the cluster is discarding).
1414                         * Instead of fail out, We can just skip the swap
1415                         * entry because swapoff will wait for discarding
1416                         * finish anyway.
1417                         */
1418                        if (!swcount || swcount == SWAP_MAP_BAD)
1419                                continue;
1420                        retval = -ENOMEM;
1421                        break;
1422                }
1423
1424                /*
1425                 * Don't hold on to start_mm if it looks like exiting.
1426                 */
1427                if (atomic_read(&start_mm->mm_users) == 1) {
1428                        mmput(start_mm);
1429                        start_mm = &init_mm;
1430                        atomic_inc(&init_mm.mm_users);
1431                }
1432
1433                /*
1434                 * Wait for and lock page.  When do_swap_page races with
1435                 * try_to_unuse, do_swap_page can handle the fault much
1436                 * faster than try_to_unuse can locate the entry.  This
1437                 * apparently redundant "wait_on_page_locked" lets try_to_unuse
1438                 * defer to do_swap_page in such a case - in some tests,
1439                 * do_swap_page and try_to_unuse repeatedly compete.
1440                 */
1441                wait_on_page_locked(page);
1442                wait_on_page_writeback(page);
1443                lock_page(page);
1444                wait_on_page_writeback(page);
1445
1446                /*
1447                 * Remove all references to entry.
1448                 */
1449                swcount = *swap_map;
1450                if (swap_count(swcount) == SWAP_MAP_SHMEM) {
1451                        retval = shmem_unuse(entry, page);
1452                        /* page has already been unlocked and released */
1453                        if (retval < 0)
1454                                break;
1455                        continue;
1456                }
1457                if (swap_count(swcount) && start_mm != &init_mm)
1458                        retval = unuse_mm(start_mm, entry, page);
1459
1460                if (swap_count(*swap_map)) {
1461                        int set_start_mm = (*swap_map >= swcount);
1462                        struct list_head *p = &start_mm->mmlist;
1463                        struct mm_struct *new_start_mm = start_mm;
1464                        struct mm_struct *prev_mm = start_mm;
1465                        struct mm_struct *mm;
1466
1467                        atomic_inc(&new_start_mm->mm_users);
1468                        atomic_inc(&prev_mm->mm_users);
1469                        spin_lock(&mmlist_lock);
1470                        while (swap_count(*swap_map) && !retval &&
1471                                        (p = p->next) != &start_mm->mmlist) {
1472                                mm = list_entry(p, struct mm_struct, mmlist);
1473                                if (!atomic_inc_not_zero(&mm->mm_users))
1474                                        continue;
1475                                spin_unlock(&mmlist_lock);
1476                                mmput(prev_mm);
1477                                prev_mm = mm;
1478
1479                                cond_resched();
1480
1481                                swcount = *swap_map;
1482                                if (!swap_count(swcount)) /* any usage ? */
1483                                        ;
1484                                else if (mm == &init_mm)
1485                                        set_start_mm = 1;
1486                                else
1487                                        retval = unuse_mm(mm, entry, page);
1488
1489                                if (set_start_mm && *swap_map < swcount) {
1490                                        mmput(new_start_mm);
1491                                        atomic_inc(&mm->mm_users);
1492                                        new_start_mm = mm;
1493                                        set_start_mm = 0;
1494                                }
1495                                spin_lock(&mmlist_lock);
1496                        }
1497                        spin_unlock(&mmlist_lock);
1498                        mmput(prev_mm);
1499                        mmput(start_mm);
1500                        start_mm = new_start_mm;
1501                }
1502                if (retval) {
1503                        unlock_page(page);
1504                        page_cache_release(page);
1505                        break;
1506                }
1507
1508                /*
1509                 * If a reference remains (rare), we would like to leave
1510                 * the page in the swap cache; but try_to_unmap could
1511                 * then re-duplicate the entry once we drop page lock,
1512                 * so we might loop indefinitely; also, that page could
1513                 * not be swapped out to other storage meanwhile.  So:
1514                 * delete from cache even if there's another reference,
1515                 * after ensuring that the data has been saved to disk -
1516                 * since if the reference remains (rarer), it will be
1517                 * read from disk into another page.  Splitting into two
1518                 * pages would be incorrect if swap supported "shared
1519                 * private" pages, but they are handled by tmpfs files.
1520                 *
1521                 * Given how unuse_vma() targets one particular offset
1522                 * in an anon_vma, once the anon_vma has been determined,
1523                 * this splitting happens to be just what is needed to
1524                 * handle where KSM pages have been swapped out: re-reading
1525                 * is unnecessarily slow, but we can fix that later on.
1526                 */
1527                if (swap_count(*swap_map) &&
1528                     PageDirty(page) && PageSwapCache(page)) {
1529                        struct writeback_control wbc = {
1530                                .sync_mode = WB_SYNC_NONE,
1531                        };
1532
1533                        swap_writepage(page, &wbc);
1534                        lock_page(page);
1535                        wait_on_page_writeback(page);
1536                }
1537
1538                /*
1539                 * It is conceivable that a racing task removed this page from
1540                 * swap cache just before we acquired the page lock at the top,
1541                 * or while we dropped it in unuse_mm().  The page might even
1542                 * be back in swap cache on another swap area: that we must not
1543                 * delete, since it may not have been written out to swap yet.
1544                 */
1545                if (PageSwapCache(page) &&
1546                    likely(page_private(page) == entry.val))
1547                        delete_from_swap_cache(page);
1548
1549                /*
1550                 * So we could skip searching mms once swap count went
1551                 * to 1, we did not mark any present ptes as dirty: must
1552                 * mark page dirty so shrink_page_list will preserve it.
1553                 */
1554                SetPageDirty(page);
1555                unlock_page(page);
1556                page_cache_release(page);
1557
1558                /*
1559                 * Make sure that we aren't completely killing
1560                 * interactive performance.
1561                 */
1562                cond_resched();
1563                if (frontswap && pages_to_unuse > 0) {
1564                        if (!--pages_to_unuse)
1565                                break;
1566                }
1567        }
1568
1569        mmput(start_mm);
1570        return retval;
1571}
1572
1573/*
1574 * After a successful try_to_unuse, if no swap is now in use, we know
1575 * we can empty the mmlist.  swap_lock must be held on entry and exit.
1576 * Note that mmlist_lock nests inside swap_lock, and an mm must be
1577 * added to the mmlist just after page_duplicate - before would be racy.
1578 */
1579static void drain_mmlist(void)
1580{
1581        struct list_head *p, *next;
1582        unsigned int type;
1583
1584        for (type = 0; type < nr_swapfiles; type++)
1585                if (swap_info[type]->inuse_pages)
1586                        return;
1587        spin_lock(&mmlist_lock);
1588        list_for_each_safe(p, next, &init_mm.mmlist)
1589                list_del_init(p);
1590        spin_unlock(&mmlist_lock);
1591}
1592
1593/*
1594 * Use this swapdev's extent info to locate the (PAGE_SIZE) block which
1595 * corresponds to page offset for the specified swap entry.
1596 * Note that the type of this function is sector_t, but it returns page offset
1597 * into the bdev, not sector offset.
1598 */
1599static sector_t map_swap_entry(swp_entry_t entry, struct block_device **bdev)
1600{
1601        struct swap_info_struct *sis;
1602        struct swap_extent *start_se;
1603        struct swap_extent *se;
1604        pgoff_t offset;
1605
1606        sis = swap_info[swp_type(entry)];
1607        *bdev = sis->bdev;
1608
1609        offset = swp_offset(entry);
1610        start_se = sis->curr_swap_extent;
1611        se = start_se;
1612
1613        for ( ; ; ) {
1614                struct list_head *lh;
1615
1616                if (se->start_page <= offset &&
1617                                offset < (se->start_page + se->nr_pages)) {
1618                        return se->start_block + (offset - se->start_page);
1619                }
1620                lh = se->list.next;
1621                se = list_entry(lh, struct swap_extent, list);
1622                sis->curr_swap_extent = se;
1623                BUG_ON(se == start_se);         /* It *must* be present */
1624        }
1625}
1626
1627/*
1628 * Returns the page offset into bdev for the specified page's swap entry.
1629 */
1630sector_t map_swap_page(struct page *page, struct block_device **bdev)
1631{
1632        swp_entry_t entry;
1633        entry.val = page_private(page);
1634        return map_swap_entry(entry, bdev);
1635}
1636
1637/*
1638 * Free all of a swapdev's extent information
1639 */
1640static void destroy_swap_extents(struct swap_info_struct *sis)
1641{
1642        while (!list_empty(&sis->first_swap_extent.list)) {
1643                struct swap_extent *se;
1644
1645                se = list_entry(sis->first_swap_extent.list.next,
1646                                struct swap_extent, list);
1647                list_del(&se->list);
1648                kfree(se);
1649        }
1650
1651        if (sis->flags & SWP_FILE) {
1652                struct file *swap_file = sis->swap_file;
1653                struct address_space *mapping = swap_file->f_mapping;
1654
1655                sis->flags &= ~SWP_FILE;
1656                mapping->a_ops->swap_deactivate(swap_file);
1657        }
1658}
1659
1660/*
1661 * Add a block range (and the corresponding page range) into this swapdev's
1662 * extent list.  The extent list is kept sorted in page order.
1663 *
1664 * This function rather assumes that it is called in ascending page order.
1665 */
1666int
1667add_swap_extent(struct swap_info_struct *sis, unsigned long start_page,
1668                unsigned long nr_pages, sector_t start_block)
1669{
1670        struct swap_extent *se;
1671        struct swap_extent *new_se;
1672        struct list_head *lh;
1673
1674        if (start_page == 0) {
1675                se = &sis->first_swap_extent;
1676                sis->curr_swap_extent = se;
1677                se->start_page = 0;
1678                se->nr_pages = nr_pages;
1679                se->start_block = start_block;
1680                return 1;
1681        } else {
1682                lh = sis->first_swap_extent.list.prev;  /* Highest extent */
1683                se = list_entry(lh, struct swap_extent, list);
1684                BUG_ON(se->start_page + se->nr_pages != start_page);
1685                if (se->start_block + se->nr_pages == start_block) {
1686                        /* Merge it */
1687                        se->nr_pages += nr_pages;
1688                        return 0;
1689                }
1690        }
1691
1692        /*
1693         * No merge.  Insert a new extent, preserving ordering.
1694         */
1695        new_se = kmalloc(sizeof(*se), GFP_KERNEL);
1696        if (new_se == NULL)
1697                return -ENOMEM;
1698        new_se->start_page = start_page;
1699        new_se->nr_pages = nr_pages;
1700        new_se->start_block = start_block;
1701
1702        list_add_tail(&new_se->list, &sis->first_swap_extent.list);
1703        return 1;
1704}
1705
1706/*
1707 * A `swap extent' is a simple thing which maps a contiguous range of pages
1708 * onto a contiguous range of disk blocks.  An ordered list of swap extents
1709 * is built at swapon time and is then used at swap_writepage/swap_readpage
1710 * time for locating where on disk a page belongs.
1711 *
1712 * If the swapfile is an S_ISBLK block device, a single extent is installed.
1713 * This is done so that the main operating code can treat S_ISBLK and S_ISREG
1714 * swap files identically.
1715 *
1716 * Whether the swapdev is an S_ISREG file or an S_ISBLK blockdev, the swap
1717 * extent list operates in PAGE_SIZE disk blocks.  Both S_ISREG and S_ISBLK
1718 * swapfiles are handled *identically* after swapon time.
1719 *
1720 * For S_ISREG swapfiles, setup_swap_extents() will walk all the file's blocks
1721 * and will parse them into an ordered extent list, in PAGE_SIZE chunks.  If
1722 * some stray blocks are found which do not fall within the PAGE_SIZE alignment
1723 * requirements, they are simply tossed out - we will never use those blocks
1724 * for swapping.
1725 *
1726 * For S_ISREG swapfiles we set S_SWAPFILE across the life of the swapon.  This
1727 * prevents root from shooting her foot off by ftruncating an in-use swapfile,
1728 * which will scribble on the fs.
1729 *
1730 * The amount of disk space which a single swap extent represents varies.
1731 * Typically it is in the 1-4 megabyte range.  So we can have hundreds of
1732 * extents in the list.  To avoid much list walking, we cache the previous
1733 * search location in `curr_swap_extent', and start new searches from there.
1734 * This is extremely effective.  The average number of iterations in
1735 * map_swap_page() has been measured at about 0.3 per page.  - akpm.
1736 */
1737static int setup_swap_extents(struct swap_info_struct *sis, sector_t *span)
1738{
1739        struct file *swap_file = sis->swap_file;
1740        struct address_space *mapping = swap_file->f_mapping;
1741        struct inode *inode = mapping->host;
1742        int ret;
1743
1744        if (S_ISBLK(inode->i_mode)) {
1745                ret = add_swap_extent(sis, 0, sis->max, 0);
1746                *span = sis->pages;
1747                return ret;
1748        }
1749
1750        if (mapping->a_ops->swap_activate) {
1751                ret = mapping->a_ops->swap_activate(sis, swap_file, span);
1752                if (!ret) {
1753                        sis->flags |= SWP_FILE;
1754                        ret = add_swap_extent(sis, 0, sis->max, 0);
1755                        *span = sis->pages;
1756                }
1757                return ret;
1758        }
1759
1760        return generic_swapfile_activate(sis, swap_file, span);
1761}
1762
1763static void _enable_swap_info(struct swap_info_struct *p, int prio,
1764                                unsigned char *swap_map,
1765                                struct swap_cluster_info *cluster_info)
1766{
1767        int i, prev;
1768
1769        if (prio >= 0)
1770                p->prio = prio;
1771        else
1772                p->prio = --least_priority;
1773        p->swap_map = swap_map;
1774        p->cluster_info = cluster_info;
1775        p->flags |= SWP_WRITEOK;
1776        atomic_long_add(p->pages, &nr_swap_pages);
1777        total_swap_pages += p->pages;
1778
1779        /* insert swap space into swap_list: */
1780        prev = -1;
1781        for (i = swap_list.head; i >= 0; i = swap_info[i]->next) {
1782                if (p->prio >= swap_info[i]->prio)
1783                        break;
1784                prev = i;
1785        }
1786        p->next = i;
1787        if (prev < 0)
1788                swap_list.head = swap_list.next = p->type;
1789        else
1790                swap_info[prev]->next = p->type;
1791}
1792
1793static void enable_swap_info(struct swap_info_struct *p, int prio,
1794                                unsigned char *swap_map,
1795                                struct swap_cluster_info *cluster_info,
1796                                unsigned long *frontswap_map)
1797{
1798        frontswap_init(p->type, frontswap_map);
1799        spin_lock(&swap_lock);
1800        spin_lock(&p->lock);
1801         _enable_swap_info(p, prio, swap_map, cluster_info);
1802        spin_unlock(&p->lock);
1803        spin_unlock(&swap_lock);
1804}
1805
1806static void reinsert_swap_info(struct swap_info_struct *p)
1807{
1808        spin_lock(&swap_lock);
1809        spin_lock(&p->lock);
1810        _enable_swap_info(p, p->prio, p->swap_map, p->cluster_info);
1811        spin_unlock(&p->lock);
1812        spin_unlock(&swap_lock);
1813}
1814
1815SYSCALL_DEFINE1(swapoff, const char __user *, specialfile)
1816{
1817        struct swap_info_struct *p = NULL;
1818        unsigned char *swap_map;
1819        struct swap_cluster_info *cluster_info;
1820        unsigned long *frontswap_map;
1821        struct file *swap_file, *victim;
1822        struct address_space *mapping;
1823        struct inode *inode;
1824        struct filename *pathname;
1825        int i, type, prev;
1826        int err;
1827        unsigned int old_block_size;
1828
1829        if (!capable(CAP_SYS_ADMIN))
1830                return -EPERM;
1831
1832        BUG_ON(!current->mm);
1833
1834        pathname = getname(specialfile);
1835        if (IS_ERR(pathname))
1836                return PTR_ERR(pathname);
1837
1838        victim = file_open_name(pathname, O_RDWR|O_LARGEFILE, 0);
1839        err = PTR_ERR(victim);
1840        if (IS_ERR(victim))
1841                goto out;
1842
1843        mapping = victim->f_mapping;
1844        prev = -1;
1845        spin_lock(&swap_lock);
1846        for (type = swap_list.head; type >= 0; type = swap_info[type]->next) {
1847                p = swap_info[type];
1848                if (p->flags & SWP_WRITEOK) {
1849                        if (p->swap_file->f_mapping == mapping)
1850                                break;
1851                }
1852                prev = type;
1853        }
1854        if (type < 0) {
1855                err = -EINVAL;
1856                spin_unlock(&swap_lock);
1857                goto out_dput;
1858        }
1859        if (!security_vm_enough_memory_mm(current->mm, p->pages))
1860                vm_unacct_memory(p->pages);
1861        else {
1862                err = -ENOMEM;
1863                spin_unlock(&swap_lock);
1864                goto out_dput;
1865        }
1866        if (prev < 0)
1867                swap_list.head = p->next;
1868        else
1869                swap_info[prev]->next = p->next;
1870        if (type == swap_list.next) {
1871                /* just pick something that's safe... */
1872                swap_list.next = swap_list.head;
1873        }
1874        spin_lock(&p->lock);
1875        if (p->prio < 0) {
1876                for (i = p->next; i >= 0; i = swap_info[i]->next)
1877                        swap_info[i]->prio = p->prio--;
1878                least_priority++;
1879        }
1880        atomic_long_sub(p->pages, &nr_swap_pages);
1881        total_swap_pages -= p->pages;
1882        p->flags &= ~SWP_WRITEOK;
1883        spin_unlock(&p->lock);
1884        spin_unlock(&swap_lock);
1885
1886        set_current_oom_origin();
1887        err = try_to_unuse(type, false, 0); /* force all pages to be unused */
1888        clear_current_oom_origin();
1889
1890        if (err) {
1891                /* re-insert swap space back into swap_list */
1892                reinsert_swap_info(p);
1893                goto out_dput;
1894        }
1895
1896        flush_work(&p->discard_work);
1897
1898        destroy_swap_extents(p);
1899        if (p->flags & SWP_CONTINUED)
1900                free_swap_count_continuations(p);
1901
1902        mutex_lock(&swapon_mutex);
1903        spin_lock(&swap_lock);
1904        spin_lock(&p->lock);
1905        drain_mmlist();
1906
1907        /* wait for anyone still in scan_swap_map */
1908        p->highest_bit = 0;             /* cuts scans short */
1909        while (p->flags >= SWP_SCANNING) {
1910                spin_unlock(&p->lock);
1911                spin_unlock(&swap_lock);
1912                schedule_timeout_uninterruptible(1);
1913                spin_lock(&swap_lock);
1914                spin_lock(&p->lock);
1915        }
1916
1917        swap_file = p->swap_file;
1918        old_block_size = p->old_block_size;
1919        p->swap_file = NULL;
1920        p->max = 0;
1921        swap_map = p->swap_map;
1922        p->swap_map = NULL;
1923        cluster_info = p->cluster_info;
1924        p->cluster_info = NULL;
1925        p->flags = 0;
1926        frontswap_map = frontswap_map_get(p);
1927        frontswap_map_set(p, NULL);
1928        spin_unlock(&p->lock);
1929        spin_unlock(&swap_lock);
1930        frontswap_invalidate_area(type);
1931        mutex_unlock(&swapon_mutex);
1932        free_percpu(p->percpu_cluster);
1933        p->percpu_cluster = NULL;
1934        vfree(swap_map);
1935        vfree(cluster_info);
1936        vfree(frontswap_map);
1937        /* Destroy swap account informatin */
1938        swap_cgroup_swapoff(type);
1939
1940        inode = mapping->host;
1941        if (S_ISBLK(inode->i_mode)) {
1942                struct block_device *bdev = I_BDEV(inode);
1943                set_blocksize(bdev, old_block_size);
1944                blkdev_put(bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
1945        } else {
1946                mutex_lock(&inode->i_mutex);
1947                inode->i_flags &= ~S_SWAPFILE;
1948                mutex_unlock(&inode->i_mutex);
1949        }
1950        filp_close(swap_file, NULL);
1951        err = 0;
1952        atomic_inc(&proc_poll_event);
1953        wake_up_interruptible(&proc_poll_wait);
1954
1955out_dput:
1956        filp_close(victim, NULL);
1957out:
1958        putname(pathname);
1959        return err;
1960}
1961
1962#ifdef CONFIG_PROC_FS
1963static unsigned swaps_poll(struct file *file, poll_table *wait)
1964{
1965        struct seq_file *seq = file->private_data;
1966
1967        poll_wait(file, &proc_poll_wait, wait);
1968
1969        if (seq->poll_event != atomic_read(&proc_poll_event)) {
1970                seq->poll_event = atomic_read(&proc_poll_event);
1971                return POLLIN | POLLRDNORM | POLLERR | POLLPRI;
1972        }
1973
1974        return POLLIN | POLLRDNORM;
1975}
1976
1977/* iterator */
1978static void *swap_start(struct seq_file *swap, loff_t *pos)
1979{
1980        struct swap_info_struct *si;
1981        int type;
1982        loff_t l = *pos;
1983
1984        mutex_lock(&swapon_mutex);
1985
1986        if (!l)
1987                return SEQ_START_TOKEN;
1988
1989        for (type = 0; type < nr_swapfiles; type++) {
1990                smp_rmb();      /* read nr_swapfiles before swap_info[type] */
1991                si = swap_info[type];
1992                if (!(si->flags & SWP_USED) || !si->swap_map)
1993                        continue;
1994                if (!--l)
1995                        return si;
1996        }
1997
1998        return NULL;
1999}
2000
2001static void *swap_next(struct seq_file *swap, void *v, loff_t *pos)
2002{
2003        struct swap_info_struct *si = v;
2004        int type;
2005
2006        if (v == SEQ_START_TOKEN)
2007                type = 0;
2008        else
2009                type = si->type + 1;
2010
2011        for (; type < nr_swapfiles; type++) {
2012                smp_rmb();      /* read nr_swapfiles before swap_info[type] */
2013                si = swap_info[type];
2014                if (!(si->flags & SWP_USED) || !si->swap_map)
2015                        continue;
2016                ++*pos;
2017                return si;
2018        }
2019
2020        return NULL;
2021}
2022
2023static void swap_stop(struct seq_file *swap, void *v)
2024{
2025        mutex_unlock(&swapon_mutex);
2026}
2027
2028static int swap_show(struct seq_file *swap, void *v)
2029{
2030        struct swap_info_struct *si = v;
2031        struct file *file;
2032        int len;
2033
2034        if (si == SEQ_START_TOKEN) {
2035                seq_puts(swap,"Filename\t\t\t\tType\t\tSize\tUsed\tPriority\n");
2036                return 0;
2037        }
2038
2039        file = si->swap_file;
2040        len = seq_path(swap, &file->f_path, " \t\n\\");
2041        seq_printf(swap, "%*s%s\t%u\t%u\t%d\n",
2042                        len < 40 ? 40 - len : 1, " ",
2043                        S_ISBLK(file_inode(file)->i_mode) ?
2044                                "partition" : "file\t",
2045                        si->pages << (PAGE_SHIFT - 10),
2046                        si->inuse_pages << (PAGE_SHIFT - 10),
2047                        si->prio);
2048        return 0;
2049}
2050
2051static const struct seq_operations swaps_op = {
2052        .start =        swap_start,
2053        .next =         swap_next,
2054        .stop =         swap_stop,
2055        .show =         swap_show
2056};
2057
2058static int swaps_open(struct inode *inode, struct file *file)
2059{
2060        struct seq_file *seq;
2061        int ret;
2062
2063        ret = seq_open(file, &swaps_op);
2064        if (ret)
2065                return ret;
2066
2067        seq = file->private_data;
2068        seq->poll_event = atomic_read(&proc_poll_event);
2069        return 0;
2070}
2071
2072static const struct file_operations proc_swaps_operations = {
2073        .open           = swaps_open,
2074        .read           = seq_read,
2075        .llseek         = seq_lseek,
2076        .release        = seq_release,
2077        .poll           = swaps_poll,
2078};
2079
2080static int __init procswaps_init(void)
2081{
2082        proc_create("swaps", 0, NULL, &proc_swaps_operations);
2083        return 0;
2084}
2085__initcall(procswaps_init);
2086#endif /* CONFIG_PROC_FS */
2087
2088#ifdef MAX_SWAPFILES_CHECK
2089static int __init max_swapfiles_check(void)
2090{
2091        MAX_SWAPFILES_CHECK();
2092        return 0;
2093}
2094late_initcall(max_swapfiles_check);
2095#endif
2096
2097static struct swap_info_struct *alloc_swap_info(void)
2098{
2099        struct swap_info_struct *p;
2100        unsigned int type;
2101
2102        p = kzalloc(sizeof(*p), GFP_KERNEL);
2103        if (!p)
2104                return ERR_PTR(-ENOMEM);
2105
2106        spin_lock(&swap_lock);
2107        for (type = 0; type < nr_swapfiles; type++) {
2108                if (!(swap_info[type]->flags & SWP_USED))
2109                        break;
2110        }
2111        if (type >= MAX_SWAPFILES) {
2112                spin_unlock(&swap_lock);
2113                kfree(p);
2114                return ERR_PTR(-EPERM);
2115        }
2116        if (type >= nr_swapfiles) {
2117                p->type = type;
2118                swap_info[type] = p;
2119                /*
2120                 * Write swap_info[type] before nr_swapfiles, in case a
2121                 * racing procfs swap_start() or swap_next() is reading them.
2122                 * (We never shrink nr_swapfiles, we never free this entry.)
2123                 */
2124                smp_wmb();
2125                nr_swapfiles++;
2126        } else {
2127                kfree(p);
2128                p = swap_info[type];
2129                /*
2130                 * Do not memset this entry: a racing procfs swap_next()
2131                 * would be relying on p->type to remain valid.
2132                 */
2133        }
2134        INIT_LIST_HEAD(&p->first_swap_extent.list);
2135        p->flags = SWP_USED;
2136        p->next = -1;
2137        spin_unlock(&swap_lock);
2138        spin_lock_init(&p->lock);
2139
2140        return p;
2141}
2142
2143static int claim_swapfile(struct swap_info_struct *p, struct inode *inode)
2144{
2145        int error;
2146
2147        if (S_ISBLK(inode->i_mode)) {
2148                p->bdev = bdgrab(I_BDEV(inode));
2149                error = blkdev_get(p->bdev,
2150                                   FMODE_READ | FMODE_WRITE | FMODE_EXCL,
2151                                   sys_swapon);
2152                if (error < 0) {
2153                        p->bdev = NULL;
2154                        return -EINVAL;
2155                }
2156                p->old_block_size = block_size(p->bdev);
2157                error = set_blocksize(p->bdev, PAGE_SIZE);
2158                if (error < 0)
2159                        return error;
2160                p->flags |= SWP_BLKDEV;
2161        } else if (S_ISREG(inode->i_mode)) {
2162                p->bdev = inode->i_sb->s_bdev;
2163                mutex_lock(&inode->i_mutex);
2164                if (IS_SWAPFILE(inode))
2165                        return -EBUSY;
2166        } else
2167                return -EINVAL;
2168
2169        return 0;
2170}
2171
2172static unsigned long read_swap_header(struct swap_info_struct *p,
2173                                        union swap_header *swap_header,
2174                                        struct inode *inode)
2175{
2176        int i;
2177        unsigned long maxpages;
2178        unsigned long swapfilepages;
2179        unsigned long last_page;
2180
2181        if (memcmp("SWAPSPACE2", swap_header->magic.magic, 10)) {
2182                pr_err("Unable to find swap-space signature\n");
2183                return 0;
2184        }
2185
2186        /* swap partition endianess hack... */
2187        if (swab32(swap_header->info.version) == 1) {
2188                swab32s(&swap_header->info.version);
2189                swab32s(&swap_header->info.last_page);
2190                swab32s(&swap_header->info.nr_badpages);
2191                for (i = 0; i < swap_header->info.nr_badpages; i++)
2192                        swab32s(&swap_header->info.badpages[i]);
2193        }
2194        /* Check the swap header's sub-version */
2195        if (swap_header->info.version != 1) {
2196                pr_warn("Unable to handle swap header version %d\n",
2197                        swap_header->info.version);
2198                return 0;
2199        }
2200
2201        p->lowest_bit  = 1;
2202        p->cluster_next = 1;
2203        p->cluster_nr = 0;
2204
2205        /*
2206         * Find out how many pages are allowed for a single swap
2207         * device. There are two limiting factors: 1) the number
2208         * of bits for the swap offset in the swp_entry_t type, and
2209         * 2) the number of bits in the swap pte as defined by the
2210         * different architectures. In order to find the
2211         * largest possible bit mask, a swap entry with swap type 0
2212         * and swap offset ~0UL is created, encoded to a swap pte,
2213         * decoded to a swp_entry_t again, and finally the swap
2214         * offset is extracted. This will mask all the bits from
2215         * the initial ~0UL mask that can't be encoded in either
2216         * the swp_entry_t or the architecture definition of a
2217         * swap pte.
2218         */
2219        maxpages = swp_offset(pte_to_swp_entry(
2220                        swp_entry_to_pte(swp_entry(0, ~0UL)))) + 1;
2221        last_page = swap_header->info.last_page;
2222        if (last_page > maxpages) {
2223                pr_warn("Truncating oversized swap area, only using %luk out of %luk\n",
2224                        maxpages << (PAGE_SHIFT - 10),
2225                        last_page << (PAGE_SHIFT - 10));
2226        }
2227        if (maxpages > last_page) {
2228                maxpages = last_page + 1;
2229                /* p->max is an unsigned int: don't overflow it */
2230                if ((unsigned int)maxpages == 0)
2231                        maxpages = UINT_MAX;
2232        }
2233        p->highest_bit = maxpages - 1;
2234
2235        if (!maxpages)
2236                return 0;
2237        swapfilepages = i_size_read(inode) >> PAGE_SHIFT;
2238        if (swapfilepages && maxpages > swapfilepages) {
2239                pr_warn("Swap area shorter than signature indicates\n");
2240                return 0;
2241        }
2242        if (swap_header->info.nr_badpages && S_ISREG(inode->i_mode))
2243                return 0;
2244        if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES)
2245                return 0;
2246
2247        return maxpages;
2248}
2249
2250static int setup_swap_map_and_extents(struct swap_info_struct *p,
2251                                        union swap_header *swap_header,
2252                                        unsigned char *swap_map,
2253                                        struct swap_cluster_info *cluster_info,
2254                                        unsigned long maxpages,
2255                                        sector_t *span)
2256{
2257        int i;
2258        unsigned int nr_good_pages;
2259        int nr_extents;
2260        unsigned long nr_clusters = DIV_ROUND_UP(maxpages, SWAPFILE_CLUSTER);
2261        unsigned long idx = p->cluster_next / SWAPFILE_CLUSTER;
2262
2263        nr_good_pages = maxpages - 1;   /* omit header page */
2264
2265        cluster_set_null(&p->free_cluster_head);
2266        cluster_set_null(&p->free_cluster_tail);
2267        cluster_set_null(&p->discard_cluster_head);
2268        cluster_set_null(&p->discard_cluster_tail);
2269
2270        for (i = 0; i < swap_header->info.nr_badpages; i++) {
2271                unsigned int page_nr = swap_header->info.badpages[i];
2272                if (page_nr == 0 || page_nr > swap_header->info.last_page)
2273                        return -EINVAL;
2274                if (page_nr < maxpages) {
2275                        swap_map[page_nr] = SWAP_MAP_BAD;
2276                        nr_good_pages--;
2277                        /*
2278                         * Haven't marked the cluster free yet, no list
2279                         * operation involved
2280                         */
2281                        inc_cluster_info_page(p, cluster_info, page_nr);
2282                }
2283        }
2284
2285        /* Haven't marked the cluster free yet, no list operation involved */
2286        for (i = maxpages; i < round_up(maxpages, SWAPFILE_CLUSTER); i++)
2287                inc_cluster_info_page(p, cluster_info, i);
2288
2289        if (nr_good_pages) {
2290                swap_map[0] = SWAP_MAP_BAD;
2291                /*
2292                 * Not mark the cluster free yet, no list
2293                 * operation involved
2294                 */
2295                inc_cluster_info_page(p, cluster_info, 0);
2296                p->max = maxpages;
2297                p->pages = nr_good_pages;
2298                nr_extents = setup_swap_extents(p, span);
2299                if (nr_extents < 0)
2300                        return nr_extents;
2301                nr_good_pages = p->pages;
2302        }
2303        if (!nr_good_pages) {
2304                pr_warn("Empty swap-file\n");
2305                return -EINVAL;
2306        }
2307
2308        if (!cluster_info)
2309                return nr_extents;
2310
2311        for (i = 0; i < nr_clusters; i++) {
2312                if (!cluster_count(&cluster_info[idx])) {
2313                        cluster_set_flag(&cluster_info[idx], CLUSTER_FLAG_FREE);
2314                        if (cluster_is_null(&p->free_cluster_head)) {
2315                                cluster_set_next_flag(&p->free_cluster_head,
2316                                                                idx, 0);
2317                                cluster_set_next_flag(&p->free_cluster_tail,
2318                                                                idx, 0);
2319                        } else {
2320                                unsigned int tail;
2321
2322                                tail = cluster_next(&p->free_cluster_tail);
2323                                cluster_set_next(&cluster_info[tail], idx);
2324                                cluster_set_next_flag(&p->free_cluster_tail,
2325                                                                idx, 0);
2326                        }
2327                }
2328                idx++;
2329                if (idx == nr_clusters)
2330                        idx = 0;
2331        }
2332        return nr_extents;
2333}
2334
2335/*
2336 * Helper to sys_swapon determining if a given swap
2337 * backing device queue supports DISCARD operations.
2338 */
2339static bool swap_discardable(struct swap_info_struct *si)
2340{
2341        struct request_queue *q = bdev_get_queue(si->bdev);
2342
2343        if (!q || !blk_queue_discard(q))
2344                return false;
2345
2346        return true;
2347}
2348
2349SYSCALL_DEFINE2(swapon, const char __user *, specialfile, int, swap_flags)
2350{
2351        struct swap_info_struct *p;
2352        struct filename *name;
2353        struct file *swap_file = NULL;
2354        struct address_space *mapping;
2355        int i;
2356        int prio;
2357        int error;
2358        union swap_header *swap_header;
2359        int nr_extents;
2360        sector_t span;
2361        unsigned long maxpages;
2362        unsigned char *swap_map = NULL;
2363        struct swap_cluster_info *cluster_info = NULL;
2364        unsigned long *frontswap_map = NULL;
2365        struct page *page = NULL;
2366        struct inode *inode = NULL;
2367
2368        if (swap_flags & ~SWAP_FLAGS_VALID)
2369                return -EINVAL;
2370
2371        if (!capable(CAP_SYS_ADMIN))
2372                return -EPERM;
2373
2374        p = alloc_swap_info();
2375        if (IS_ERR(p))
2376                return PTR_ERR(p);
2377
2378        INIT_WORK(&p->discard_work, swap_discard_work);
2379
2380        name = getname(specialfile);
2381        if (IS_ERR(name)) {
2382                error = PTR_ERR(name);
2383                name = NULL;
2384                goto bad_swap;
2385        }
2386        swap_file = file_open_name(name, O_RDWR|O_LARGEFILE, 0);
2387        if (IS_ERR(swap_file)) {
2388                error = PTR_ERR(swap_file);
2389                swap_file = NULL;
2390                goto bad_swap;
2391        }
2392
2393        p->swap_file = swap_file;
2394        mapping = swap_file->f_mapping;
2395
2396        for (i = 0; i < nr_swapfiles; i++) {
2397                struct swap_info_struct *q = swap_info[i];
2398
2399                if (q == p || !q->swap_file)
2400                        continue;
2401                if (mapping == q->swap_file->f_mapping) {
2402                        error = -EBUSY;
2403                        goto bad_swap;
2404                }
2405        }
2406
2407        inode = mapping->host;
2408        /* If S_ISREG(inode->i_mode) will do mutex_lock(&inode->i_mutex); */
2409        error = claim_swapfile(p, inode);
2410        if (unlikely(error))
2411                goto bad_swap;
2412
2413        /*
2414         * Read the swap header.
2415         */
2416        if (!mapping->a_ops->readpage) {
2417                error = -EINVAL;
2418                goto bad_swap;
2419        }
2420        page = read_mapping_page(mapping, 0, swap_file);
2421        if (IS_ERR(page)) {
2422                error = PTR_ERR(page);
2423                goto bad_swap;
2424        }
2425        swap_header = kmap(page);
2426
2427        maxpages = read_swap_header(p, swap_header, inode);
2428        if (unlikely(!maxpages)) {
2429                error = -EINVAL;
2430                goto bad_swap;
2431        }
2432
2433        /* OK, set up the swap map and apply the bad block list */
2434        swap_map = vzalloc(maxpages);
2435        if (!swap_map) {
2436                error = -ENOMEM;
2437                goto bad_swap;
2438        }
2439        if (p->bdev && blk_queue_nonrot(bdev_get_queue(p->bdev))) {
2440                p->flags |= SWP_SOLIDSTATE;
2441                /*
2442                 * select a random position to start with to help wear leveling
2443                 * SSD
2444                 */
2445                p->cluster_next = 1 + (prandom_u32() % p->highest_bit);
2446
2447                cluster_info = vzalloc(DIV_ROUND_UP(maxpages,
2448                        SWAPFILE_CLUSTER) * sizeof(*cluster_info));
2449                if (!cluster_info) {
2450                        error = -ENOMEM;
2451                        goto bad_swap;
2452                }
2453                p->percpu_cluster = alloc_percpu(struct percpu_cluster);
2454                if (!p->percpu_cluster) {
2455                        error = -ENOMEM;
2456                        goto bad_swap;
2457                }
2458                for_each_possible_cpu(i) {
2459                        struct percpu_cluster *cluster;
2460                        cluster = per_cpu_ptr(p->percpu_cluster, i);
2461                        cluster_set_null(&cluster->index);
2462                }
2463        }
2464
2465        error = swap_cgroup_swapon(p->type, maxpages);
2466        if (error)
2467                goto bad_swap;
2468
2469        nr_extents = setup_swap_map_and_extents(p, swap_header, swap_map,
2470                cluster_info, maxpages, &span);
2471        if (unlikely(nr_extents < 0)) {
2472                error = nr_extents;
2473                goto bad_swap;
2474        }
2475        /* frontswap enabled? set up bit-per-page map for frontswap */
2476        if (frontswap_enabled)
2477                frontswap_map = vzalloc(BITS_TO_LONGS(maxpages) * sizeof(long));
2478
2479        if (p->bdev &&(swap_flags & SWAP_FLAG_DISCARD) && swap_discardable(p)) {
2480                /*
2481                 * When discard is enabled for swap with no particular
2482                 * policy flagged, we set all swap discard flags here in
2483                 * order to sustain backward compatibility with older
2484                 * swapon(8) releases.
2485                 */
2486                p->flags |= (SWP_DISCARDABLE | SWP_AREA_DISCARD |
2487                             SWP_PAGE_DISCARD);
2488
2489                /*
2490                 * By flagging sys_swapon, a sysadmin can tell us to
2491                 * either do single-time area discards only, or to just
2492                 * perform discards for released swap page-clusters.
2493                 * Now it's time to adjust the p->flags accordingly.
2494                 */
2495                if (swap_flags & SWAP_FLAG_DISCARD_ONCE)
2496                        p->flags &= ~SWP_PAGE_DISCARD;
2497                else if (swap_flags & SWAP_FLAG_DISCARD_PAGES)
2498                        p->flags &= ~SWP_AREA_DISCARD;
2499
2500                /* issue a swapon-time discard if it's still required */
2501                if (p->flags & SWP_AREA_DISCARD) {
2502                        int err = discard_swap(p);
2503                        if (unlikely(err))
2504                                pr_err("swapon: discard_swap(%p): %d\n",
2505                                        p, err);
2506                }
2507        }
2508
2509        mutex_lock(&swapon_mutex);
2510        prio = -1;
2511        if (swap_flags & SWAP_FLAG_PREFER)
2512                prio =
2513                  (swap_flags & SWAP_FLAG_PRIO_MASK) >> SWAP_FLAG_PRIO_SHIFT;
2514        enable_swap_info(p, prio, swap_map, cluster_info, frontswap_map);
2515
2516        pr_info("Adding %uk swap on %s.  "
2517                        "Priority:%d extents:%d across:%lluk %s%s%s%s%s\n",
2518                p->pages<<(PAGE_SHIFT-10), name->name, p->prio,
2519                nr_extents, (unsigned long long)span<<(PAGE_SHIFT-10),
2520                (p->flags & SWP_SOLIDSTATE) ? "SS" : "",
2521                (p->flags & SWP_DISCARDABLE) ? "D" : "",
2522                (p->flags & SWP_AREA_DISCARD) ? "s" : "",
2523                (p->flags & SWP_PAGE_DISCARD) ? "c" : "",
2524                (frontswap_map) ? "FS" : "");
2525
2526        mutex_unlock(&swapon_mutex);
2527        atomic_inc(&proc_poll_event);
2528        wake_up_interruptible(&proc_poll_wait);
2529
2530        if (S_ISREG(inode->i_mode))
2531                inode->i_flags |= S_SWAPFILE;
2532        error = 0;
2533        goto out;
2534bad_swap:
2535        free_percpu(p->percpu_cluster);
2536        p->percpu_cluster = NULL;
2537        if (inode && S_ISBLK(inode->i_mode) && p->bdev) {
2538                set_blocksize(p->bdev, p->old_block_size);
2539                blkdev_put(p->bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
2540        }
2541        destroy_swap_extents(p);
2542        swap_cgroup_swapoff(p->type);
2543        spin_lock(&swap_lock);
2544        p->swap_file = NULL;
2545        p->flags = 0;
2546        spin_unlock(&swap_lock);
2547        vfree(swap_map);
2548        vfree(cluster_info);
2549        if (swap_file) {
2550                if (inode && S_ISREG(inode->i_mode)) {
2551                        mutex_unlock(&inode->i_mutex);
2552                        inode = NULL;
2553                }
2554                filp_close(swap_file, NULL);
2555        }
2556out:
2557        if (page && !IS_ERR(page)) {
2558                kunmap(page);
2559                page_cache_release(page);
2560        }
2561        if (name)
2562                putname(name);
2563        if (inode && S_ISREG(inode->i_mode))
2564                mutex_unlock(&inode->i_mutex);
2565        return error;
2566}
2567
2568void si_swapinfo(struct sysinfo *val)
2569{
2570        unsigned int type;
2571        unsigned long nr_to_be_unused = 0;
2572
2573        spin_lock(&swap_lock);
2574        for (type = 0; type < nr_swapfiles; type++) {
2575                struct swap_info_struct *si = swap_info[type];
2576
2577                if ((si->flags & SWP_USED) && !(si->flags & SWP_WRITEOK))
2578                        nr_to_be_unused += si->inuse_pages;
2579        }
2580        val->freeswap = atomic_long_read(&nr_swap_pages) + nr_to_be_unused;
2581        val->totalswap = total_swap_pages + nr_to_be_unused;
2582        spin_unlock(&swap_lock);
2583}
2584
2585/*
2586 * Verify that a swap entry is valid and increment its swap map count.
2587 *
2588 * Returns error code in following case.
2589 * - success -> 0
2590 * - swp_entry is invalid -> EINVAL
2591 * - swp_entry is migration entry -> EINVAL
2592 * - swap-cache reference is requested but there is already one. -> EEXIST
2593 * - swap-cache reference is requested but the entry is not used. -> ENOENT
2594 * - swap-mapped reference requested but needs continued swap count. -> ENOMEM
2595 */
2596static int __swap_duplicate(swp_entry_t entry, unsigned char usage)
2597{
2598        struct swap_info_struct *p;
2599        unsigned long offset, type;
2600        unsigned char count;
2601        unsigned char has_cache;
2602        int err = -EINVAL;
2603
2604        if (non_swap_entry(entry))
2605                goto out;
2606
2607        type = swp_type(entry);
2608        if (type >= nr_swapfiles)
2609                goto bad_file;
2610        p = swap_info[type];
2611        offset = swp_offset(entry);
2612
2613        spin_lock(&p->lock);
2614        if (unlikely(offset >= p->max))
2615                goto unlock_out;
2616
2617        count = p->swap_map[offset];
2618
2619        /*
2620         * swapin_readahead() doesn't check if a swap entry is valid, so the
2621         * swap entry could be SWAP_MAP_BAD. Check here with lock held.
2622         */
2623        if (unlikely(swap_count(count) == SWAP_MAP_BAD)) {
2624                err = -ENOENT;
2625                goto unlock_out;
2626        }
2627
2628        has_cache = count & SWAP_HAS_CACHE;
2629        count &= ~SWAP_HAS_CACHE;
2630        err = 0;
2631
2632        if (usage == SWAP_HAS_CACHE) {
2633
2634                /* set SWAP_HAS_CACHE if there is no cache and entry is used */
2635                if (!has_cache && count)
2636                        has_cache = SWAP_HAS_CACHE;
2637                else if (has_cache)             /* someone else added cache */
2638                        err = -EEXIST;
2639                else                            /* no users remaining */
2640                        err = -ENOENT;
2641
2642        } else if (count || has_cache) {
2643
2644                if ((count & ~COUNT_CONTINUED) < SWAP_MAP_MAX)
2645                        count += usage;
2646                else if ((count & ~COUNT_CONTINUED) > SWAP_MAP_MAX)
2647                        err = -EINVAL;
2648                else if (swap_count_continued(p, offset, count))
2649                        count = COUNT_CONTINUED;
2650                else
2651                        err = -ENOMEM;
2652        } else
2653                err = -ENOENT;                  /* unused swap entry */
2654
2655        p->swap_map[offset] = count | has_cache;
2656
2657unlock_out:
2658        spin_unlock(&p->lock);
2659out:
2660        return err;
2661
2662bad_file:
2663        pr_err("swap_dup: %s%08lx\n", Bad_file, entry.val);
2664        goto out;
2665}
2666
2667/*
2668 * Help swapoff by noting that swap entry belongs to shmem/tmpfs
2669 * (in which case its reference count is never incremented).
2670 */
2671void swap_shmem_alloc(swp_entry_t entry)
2672{
2673        __swap_duplicate(entry, SWAP_MAP_SHMEM);
2674}
2675
2676/*
2677 * Increase reference count of swap entry by 1.
2678 * Returns 0 for success, or -ENOMEM if a swap_count_continuation is required
2679 * but could not be atomically allocated.  Returns 0, just as if it succeeded,
2680 * if __swap_duplicate() fails for another reason (-EINVAL or -ENOENT), which
2681 * might occur if a page table entry has got corrupted.
2682 */
2683int swap_duplicate(swp_entry_t entry)
2684{
2685        int err = 0;
2686
2687        while (!err && __swap_duplicate(entry, 1) == -ENOMEM)
2688                err = add_swap_count_continuation(entry, GFP_ATOMIC);
2689        return err;
2690}
2691
2692/*
2693 * @entry: swap entry for which we allocate swap cache.
2694 *
2695 * Called when allocating swap cache for existing swap entry,
2696 * This can return error codes. Returns 0 at success.
2697 * -EBUSY means there is a swap cache.
2698 * Note: return code is different from swap_duplicate().
2699 */
2700int swapcache_prepare(swp_entry_t entry)
2701{
2702        return __swap_duplicate(entry, SWAP_HAS_CACHE);
2703}
2704
2705struct swap_info_struct *page_swap_info(struct page *page)
2706{
2707        swp_entry_t swap = { .val = page_private(page) };
2708        BUG_ON(!PageSwapCache(page));
2709        return swap_info[swp_type(swap)];
2710}
2711
2712/*
2713 * out-of-line __page_file_ methods to avoid include hell.
2714 */
2715struct address_space *__page_file_mapping(struct page *page)
2716{
2717        VM_BUG_ON(!PageSwapCache(page));
2718        return page_swap_info(page)->swap_file->f_mapping;
2719}
2720EXPORT_SYMBOL_GPL(__page_file_mapping);
2721
2722pgoff_t __page_file_index(struct page *page)
2723{
2724        swp_entry_t swap = { .val = page_private(page) };
2725        VM_BUG_ON(!PageSwapCache(page));
2726        return swp_offset(swap);
2727}
2728EXPORT_SYMBOL_GPL(__page_file_index);
2729
2730/*
2731 * add_swap_count_continuation - called when a swap count is duplicated
2732 * beyond SWAP_MAP_MAX, it allocates a new page and links that to the entry's
2733 * page of the original vmalloc'ed swap_map, to hold the continuation count
2734 * (for that entry and for its neighbouring PAGE_SIZE swap entries).  Called
2735 * again when count is duplicated beyond SWAP_MAP_MAX * SWAP_CONT_MAX, etc.
2736 *
2737 * These continuation pages are seldom referenced: the common paths all work
2738 * on the original swap_map, only referring to a continuation page when the
2739 * low "digit" of a count is incremented or decremented through SWAP_MAP_MAX.
2740 *
2741 * add_swap_count_continuation(, GFP_ATOMIC) can be called while holding
2742 * page table locks; if it fails, add_swap_count_continuation(, GFP_KERNEL)
2743 * can be called after dropping locks.
2744 */
2745int add_swap_count_continuation(swp_entry_t entry, gfp_t gfp_mask)
2746{
2747        struct swap_info_struct *si;
2748        struct page *head;
2749        struct page *page;
2750        struct page *list_page;
2751        pgoff_t offset;
2752        unsigned char count;
2753
2754        /*
2755         * When debugging, it's easier to use __GFP_ZERO here; but it's better
2756         * for latency not to zero a page while GFP_ATOMIC and holding locks.
2757         */
2758        page = alloc_page(gfp_mask | __GFP_HIGHMEM);
2759
2760        si = swap_info_get(entry);
2761        if (!si) {
2762                /*
2763                 * An acceptable race has occurred since the failing
2764                 * __swap_duplicate(): the swap entry has been freed,
2765                 * perhaps even the whole swap_map cleared for swapoff.
2766                 */
2767                goto outer;
2768        }
2769
2770        offset = swp_offset(entry);
2771        count = si->swap_map[offset] & ~SWAP_HAS_CACHE;
2772
2773        if ((count & ~COUNT_CONTINUED) != SWAP_MAP_MAX) {
2774                /*
2775                 * The higher the swap count, the more likely it is that tasks
2776                 * will race to add swap count continuation: we need to avoid
2777                 * over-provisioning.
2778                 */
2779                goto out;
2780        }
2781
2782        if (!page) {
2783                spin_unlock(&si->lock);
2784                return -ENOMEM;
2785        }
2786
2787        /*
2788         * We are fortunate that although vmalloc_to_page uses pte_offset_map,
2789         * no architecture is using highmem pages for kernel pagetables: so it
2790         * will not corrupt the GFP_ATOMIC caller's atomic pagetable kmaps.
2791         */
2792        head = vmalloc_to_page(si->swap_map + offset);
2793        offset &= ~PAGE_MASK;
2794
2795        /*
2796         * Page allocation does not initialize the page's lru field,
2797         * but it does always reset its private field.
2798         */
2799        if (!page_private(head)) {
2800                BUG_ON(count & COUNT_CONTINUED);
2801                INIT_LIST_HEAD(&head->lru);
2802                set_page_private(head, SWP_CONTINUED);
2803                si->flags |= SWP_CONTINUED;
2804        }
2805
2806        list_for_each_entry(list_page, &head->lru, lru) {
2807                unsigned char *map;
2808
2809                /*
2810                 * If the previous map said no continuation, but we've found
2811                 * a continuation page, free our allocation and use this one.
2812                 */
2813                if (!(count & COUNT_CONTINUED))
2814                        goto out;
2815
2816                map = kmap_atomic(list_page) + offset;
2817                count = *map;
2818                kunmap_atomic(map);
2819
2820                /*
2821                 * If this continuation count now has some space in it,
2822                 * free our allocation and use this one.
2823                 */
2824                if ((count & ~COUNT_CONTINUED) != SWAP_CONT_MAX)
2825                        goto out;
2826        }
2827
2828        list_add_tail(&page->lru, &head->lru);
2829        page = NULL;                    /* now it's attached, don't free it */
2830out:
2831        spin_unlock(&si->lock);
2832outer:
2833        if (page)
2834                __free_page(page);
2835        return 0;
2836}
2837
2838/*
2839 * swap_count_continued - when the original swap_map count is incremented
2840 * from SWAP_MAP_MAX, check if there is already a continuation page to carry
2841 * into, carry if so, or else fail until a new continuation page is allocated;
2842 * when the original swap_map count is decremented from 0 with continuation,
2843 * borrow from the continuation and report whether it still holds more.
2844 * Called while __swap_duplicate() or swap_entry_free() holds swap_lock.
2845 */
2846static bool swap_count_continued(struct swap_info_struct *si,
2847                                 pgoff_t offset, unsigned char count)
2848{
2849        struct page *head;
2850        struct page *page;
2851        unsigned char *map;
2852
2853        head = vmalloc_to_page(si->swap_map + offset);
2854        if (page_private(head) != SWP_CONTINUED) {
2855                BUG_ON(count & COUNT_CONTINUED);
2856                return false;           /* need to add count continuation */
2857        }
2858
2859        offset &= ~PAGE_MASK;
2860        page = list_entry(head->lru.next, struct page, lru);
2861        map = kmap_atomic(page) + offset;
2862
2863        if (count == SWAP_MAP_MAX)      /* initial increment from swap_map */
2864                goto init_map;          /* jump over SWAP_CONT_MAX checks */
2865
2866        if (count == (SWAP_MAP_MAX | COUNT_CONTINUED)) { /* incrementing */
2867                /*
2868                 * Think of how you add 1 to 999
2869                 */
2870                while (*map == (SWAP_CONT_MAX | COUNT_CONTINUED)) {
2871                        kunmap_atomic(map);
2872                        page = list_entry(page->lru.next, struct page, lru);
2873                        BUG_ON(page == head);
2874                        map = kmap_atomic(page) + offset;
2875                }
2876                if (*map == SWAP_CONT_MAX) {
2877                        kunmap_atomic(map);
2878                        page = list_entry(page->lru.next, struct page, lru);
2879                        if (page == head)
2880                                return false;   /* add count continuation */
2881                        map = kmap_atomic(page) + offset;
2882init_map:               *map = 0;               /* we didn't zero the page */
2883                }
2884                *map += 1;
2885                kunmap_atomic(map);
2886                page = list_entry(page->lru.prev, struct page, lru);
2887                while (page != head) {
2888                        map = kmap_atomic(page) + offset;
2889                        *map = COUNT_CONTINUED;
2890                        kunmap_atomic(map);
2891                        page = list_entry(page->lru.prev, struct page, lru);
2892                }
2893                return true;                    /* incremented */
2894
2895        } else {                                /* decrementing */
2896                /*
2897                 * Think of how you subtract 1 from 1000
2898                 */
2899                BUG_ON(count != COUNT_CONTINUED);
2900                while (*map == COUNT_CONTINUED) {
2901                        kunmap_atomic(map);
2902                        page = list_entry(page->lru.next, struct page, lru);
2903                        BUG_ON(page == head);
2904                        map = kmap_atomic(page) + offset;
2905                }
2906                BUG_ON(*map == 0);
2907                *map -= 1;
2908                if (*map == 0)
2909                        count = 0;
2910                kunmap_atomic(map);
2911                page = list_entry(page->lru.prev, struct page, lru);
2912                while (page != head) {
2913                        map = kmap_atomic(page) + offset;
2914                        *map = SWAP_CONT_MAX | count;
2915                        count = COUNT_CONTINUED;
2916                        kunmap_atomic(map);
2917                        page = list_entry(page->lru.prev, struct page, lru);
2918                }
2919                return count == COUNT_CONTINUED;
2920        }
2921}
2922
2923/*
2924 * free_swap_count_continuations - swapoff free all the continuation pages
2925 * appended to the swap_map, after swap_map is quiesced, before vfree'ing it.
2926 */
2927static void free_swap_count_continuations(struct swap_info_struct *si)
2928{
2929        pgoff_t offset;
2930
2931        for (offset = 0; offset < si->max; offset += PAGE_SIZE) {
2932                struct page *head;
2933                head = vmalloc_to_page(si->swap_map + offset);
2934                if (page_private(head)) {
2935                        struct list_head *this, *next;
2936                        list_for_each_safe(this, next, &head->lru) {
2937                                struct page *page;
2938                                page = list_entry(this, struct page, lru);
2939                                list_del(this);
2940                                __free_page(page);
2941                        }
2942                }
2943        }
2944}
2945
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.