mm/zswap.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  * zswap.c - zswap driver file
   4  *
   5  * zswap is a backend for frontswap that takes pages that are in the process
   6  * of being swapped out and attempts to compress and store them in a
   7  * RAM-based memory pool.  This can result in a significant I/O reduction on
   8  * the swap device and, in the case where decompressing from RAM is faster
   9  * than reading from the swap device, can also improve workload performance.
  10  *
  11  * Copyright (C) 2012  Seth Jennings <sjenning@linux.vnet.ibm.com>
  12 */
  13
  14 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  15
  16 #include <linux/module.h>
  17 #include <linux/cpu.h>
  18 #include <linux/highmem.h>
  19 #include <linux/slab.h>
  20 #include <linux/spinlock.h>
  21 #include <linux/types.h>
  22 #include <linux/atomic.h>
  23 #include <linux/frontswap.h>
  24 #include <linux/rbtree.h>
  25 #include <linux/swap.h>
  26 #include <linux/crypto.h>
  27 #include <linux/scatterlist.h>
  28 #include <linux/mempool.h>
  29 #include <linux/zpool.h>
  30 #include <crypto/acompress.h>
  31
  32 #include <linux/mm_types.h>
  33 #include <linux/page-flags.h>
  34 #include <linux/swapops.h>
  35 #include <linux/writeback.h>
  36 #include <linux/pagemap.h>
  37 #include <linux/workqueue.h>
  38
  39 /*********************************
  40 * statistics
  41 **********************************/
  42 /* Total bytes used by the compressed storage */
  43 static u64 zswap_pool_total_size;
  44 /* The number of compressed pages currently stored in zswap */
  45 static atomic_t zswap_stored_pages = ATOMIC_INIT(0);
  46 /* The number of same-value filled pages currently stored in zswap */
  47 static atomic_t zswap_same_filled_pages = ATOMIC_INIT(0);
  48
  49 /*
  50  * The statistics below are not protected from concurrent access for
  51  * performance reasons so they may not be a 100% accurate.  However,
  52  * they do provide useful information on roughly how many times a
  53  * certain event is occurring.
  54 */
  55
  56 /* Pool limit was hit (see zswap_max_pool_percent) */
  57 static u64 zswap_pool_limit_hit;
  58 /* Pages written back when pool limit was reached */
  59 static u64 zswap_written_back_pages;
  60 /* Store failed due to a reclaim failure after pool limit was reached */
  61 static u64 zswap_reject_reclaim_fail;
  62 /* Compressed page was too big for the allocator to (optimally) store */
  63 static u64 zswap_reject_compress_poor;
  64 /* Store failed because underlying allocator could not get memory */
  65 static u64 zswap_reject_alloc_fail;
  66 /* Store failed because the entry metadata could not be allocated (rare) */
  67 static u64 zswap_reject_kmemcache_fail;
  68 /* Duplicate store was encountered (rare) */
  69 static u64 zswap_duplicate_entry;
  70
  71 /* Shrinker work queue */
  72 static struct workqueue_struct *shrink_wq;
  73 /* Pool limit was hit, we need to calm down */
  74 static bool zswap_pool_reached_full;
  75
  76 /*********************************
  77 * tunables
  78 **********************************/
  79
  80 #define ZSWAP_PARAM_UNSET ""
  81
  82 /* Enable/disable zswap */
  83 static bool zswap_enabled = IS_ENABLED(CONFIG_ZSWAP_DEFAULT_ON);
  84 static int zswap_enabled_param_set(const char *,
  85                                    const struct kernel_param *);
  86 static const struct kernel_param_ops zswap_enabled_param_ops = {
  87         .set =          zswap_enabled_param_set,
  88         .get =          param_get_bool,
  89 };
  90 module_param_cb(enabled, &zswap_enabled_param_ops, &zswap_enabled, 0644);
  91
  92 /* Crypto compressor to use */
  93 static char *zswap_compressor = CONFIG_ZSWAP_COMPRESSOR_DEFAULT;
  94 static int zswap_compressor_param_set(const char *,
  95                                       const struct kernel_param *);
  96 static const struct kernel_param_ops zswap_compressor_param_ops = {
  97         .set =          zswap_compressor_param_set,
  98         .get =          param_get_charp,
  99         .free =         param_free_charp,
 100 };
 101 module_param_cb(compressor, &zswap_compressor_param_ops,
 102                 &zswap_compressor, 0644);
 103
 104 /* Compressed storage zpool to use */
 105 static char *zswap_zpool_type = CONFIG_ZSWAP_ZPOOL_DEFAULT;
 106 static int zswap_zpool_param_set(const char *, const struct kernel_param *);
 107 static const struct kernel_param_ops zswap_zpool_param_ops = {
 108         .set =          zswap_zpool_param_set,
 109         .get =          param_get_charp,
 110         .free =         param_free_charp,
 111 };
 112 module_param_cb(zpool, &zswap_zpool_param_ops, &zswap_zpool_type, 0644);
 113
 114 /* The maximum percentage of memory that the compressed pool can occupy */
 115 static unsigned int zswap_max_pool_percent = 20;
 116 module_param_named(max_pool_percent, zswap_max_pool_percent, uint, 0644);
 117
 118 /* The threshold for accepting new pages after the max_pool_percent was hit */
 119 static unsigned int zswap_accept_thr_percent = 90; /* of max pool size */
 120 module_param_named(accept_threshold_percent, zswap_accept_thr_percent,
 121                    uint, 0644);
 122
 123 /*
 124  * Enable/disable handling same-value filled pages (enabled by default).
 125  * If disabled every page is considered non-same-value filled.
 126  */
 127 static bool zswap_same_filled_pages_enabled = true;
 128 module_param_named(same_filled_pages_enabled, zswap_same_filled_pages_enabled,
 129                    bool, 0644);
 130
 131 /* Enable/disable handling non-same-value filled pages (enabled by default) */
 132 static bool zswap_non_same_filled_pages_enabled = true;
 133 module_param_named(non_same_filled_pages_enabled, zswap_non_same_filled_pages_enabled,
 134                    bool, 0644);
 135
 136 /*********************************
 137 * data structures
 138 **********************************/
 139
 140 struct crypto_acomp_ctx {
 141         struct crypto_acomp *acomp;
 142         struct acomp_req *req;
 143         struct crypto_wait wait;
 144         u8 *dstmem;
 145         struct mutex *mutex;
 146 };
 147
 148 struct zswap_pool {
 149         struct zpool *zpool;
 150         struct crypto_acomp_ctx __percpu *acomp_ctx;
 151         struct kref kref;
 152         struct list_head list;
 153         struct work_struct release_work;
 154         struct work_struct shrink_work;
 155         struct hlist_node node;
 156         char tfm_name[CRYPTO_MAX_ALG_NAME];
 157 };
 158
 159 /*
 160  * struct zswap_entry
 161  *
 162  * This structure contains the metadata for tracking a single compressed
 163  * page within zswap.
 164  *
 165  * rbnode - links the entry into red-black tree for the appropriate swap type
 166  * offset - the swap offset for the entry.  Index into the red-black tree.
 167  * refcount - the number of outstanding reference to the entry. This is needed
 168  *            to protect against premature freeing of the entry by code
 169  *            concurrent calls to load, invalidate, and writeback.  The lock
 170  *            for the zswap_tree structure that contains the entry must
 171  *            be held while changing the refcount.  Since the lock must
 172  *            be held, there is no reason to also make refcount atomic.
 173  * length - the length in bytes of the compressed page data.  Needed during
 174  *          decompression. For a same value filled page length is 0.
 175  * pool - the zswap_pool the entry's data is in
 176  * handle - zpool allocation handle that stores the compressed page data
 177  * value - value of the same-value filled pages which have same content
 178  */
 179 struct zswap_entry {
 180         struct rb_node rbnode;
 181         pgoff_t offset;
 182         int refcount;
 183         unsigned int length;
 184         struct zswap_pool *pool;
 185         union {
 186                 unsigned long handle;
 187                 unsigned long value;
 188         };
 189 };
 190
 191 struct zswap_header {
 192         swp_entry_t swpentry;
 193 };
 194
 195 /*
 196  * The tree lock in the zswap_tree struct protects a few things:
 197  * - the rbtree
 198  * - the refcount field of each entry in the tree
 199  */
 200 struct zswap_tree {
 201         struct rb_root rbroot;
 202         spinlock_t lock;
 203 };
 204
 205 static struct zswap_tree *zswap_trees[MAX_SWAPFILES];
 206
 207 /* RCU-protected iteration */
 208 static LIST_HEAD(zswap_pools);
 209 /* protects zswap_pools list modification */
 210 static DEFINE_SPINLOCK(zswap_pools_lock);
 211 /* pool counter to provide unique names to zpool */
 212 static atomic_t zswap_pools_count = ATOMIC_INIT(0);
 213
 214 /* used by param callback function */
 215 static bool zswap_init_started;
 216
 217 /* fatal error during init */
 218 static bool zswap_init_failed;
 219
 220 /* init completed, but couldn't create the initial pool */
 221 static bool zswap_has_pool;
 222
 223 /*********************************
 224 * helpers and fwd declarations
 225 **********************************/
 226
 227 #define zswap_pool_debug(msg, p)                                \
 228         pr_debug("%s pool %s/%s\n", msg, (p)->tfm_name,         \
 229                  zpool_get_type((p)->zpool))
 230
 231 static int zswap_writeback_entry(struct zpool *pool, unsigned long handle);
 232 static int zswap_pool_get(struct zswap_pool *pool);
 233 static void zswap_pool_put(struct zswap_pool *pool);
 234
 235 static const struct zpool_ops zswap_zpool_ops = {
 236         .evict = zswap_writeback_entry
 237 };
 238
 239 static bool zswap_is_full(void)
 240 {
 241         return totalram_pages() * zswap_max_pool_percent / 100 <
 242                         DIV_ROUND_UP(zswap_pool_total_size, PAGE_SIZE);
 243 }
 244
 245 static bool zswap_can_accept(void)
 246 {
 247         return totalram_pages() * zswap_accept_thr_percent / 100 *
 248                                 zswap_max_pool_percent / 100 >
 249                         DIV_ROUND_UP(zswap_pool_total_size, PAGE_SIZE);
 250 }
 251
 252 static void zswap_update_total_size(void)
 253 {
 254         struct zswap_pool *pool;
 255         u64 total = 0;
 256
 257         rcu_read_lock();
 258
 259         list_for_each_entry_rcu(pool, &zswap_pools, list)
 260                 total += zpool_get_total_size(pool->zpool);
 261
 262         rcu_read_unlock();
 263
 264         zswap_pool_total_size = total;
 265 }
 266
 267 /*********************************
 268 * zswap entry functions
 269 **********************************/
 270 static struct kmem_cache *zswap_entry_cache;
 271
 272 static int __init zswap_entry_cache_create(void)
 273 {
 274         zswap_entry_cache = KMEM_CACHE(zswap_entry, 0);
 275         return zswap_entry_cache == NULL;
 276 }
 277
 278 static void __init zswap_entry_cache_destroy(void)
 279 {
 280         kmem_cache_destroy(zswap_entry_cache);
 281 }
 282
 283 static struct zswap_entry *zswap_entry_cache_alloc(gfp_t gfp)
 284 {
 285         struct zswap_entry *entry;
 286         entry = kmem_cache_alloc(zswap_entry_cache, gfp);
 287         if (!entry)
 288                 return NULL;
 289         entry->refcount = 1;
 290         RB_CLEAR_NODE(&entry->rbnode);
 291         return entry;
 292 }
 293
 294 static void zswap_entry_cache_free(struct zswap_entry *entry)
 295 {
 296         kmem_cache_free(zswap_entry_cache, entry);
 297 }
 298
 299 /*********************************
 300 * rbtree functions
 301 **********************************/
 302 static struct zswap_entry *zswap_rb_search(struct rb_root *root, pgoff_t offset)
 303 {
 304         struct rb_node *node = root->rb_node;
 305         struct zswap_entry *entry;
 306
 307         while (node) {
 308                 entry = rb_entry(node, struct zswap_entry, rbnode);
 309                 if (entry->offset > offset)
 310                         node = node->rb_left;
 311                 else if (entry->offset < offset)
 312                         node = node->rb_right;
 313                 else
 314                         return entry;
 315         }
 316         return NULL;
 317 }
 318
 319 /*
 320  * In the case that a entry with the same offset is found, a pointer to
 321  * the existing entry is stored in dupentry and the function returns -EEXIST
 322  */
 323 static int zswap_rb_insert(struct rb_root *root, struct zswap_entry *entry,
 324                         struct zswap_entry **dupentry)
 325 {
 326         struct rb_node **link = &root->rb_node, *parent = NULL;
 327         struct zswap_entry *myentry;
 328
 329         while (*link) {
 330                 parent = *link;
 331                 myentry = rb_entry(parent, struct zswap_entry, rbnode);
 332                 if (myentry->offset > entry->offset)
 333                         link = &(*link)->rb_left;
 334                 else if (myentry->offset < entry->offset)
 335                         link = &(*link)->rb_right;
 336                 else {
 337                         *dupentry = myentry;
 338                         return -EEXIST;
 339                 }
 340         }
 341         rb_link_node(&entry->rbnode, parent, link);
 342         rb_insert_color(&entry->rbnode, root);
 343         return 0;
 344 }
 345
 346 static void zswap_rb_erase(struct rb_root *root, struct zswap_entry *entry)
 347 {
 348         if (!RB_EMPTY_NODE(&entry->rbnode)) {
 349                 rb_erase(&entry->rbnode, root);
 350                 RB_CLEAR_NODE(&entry->rbnode);
 351         }
 352 }
 353
 354 /*
 355  * Carries out the common pattern of freeing and entry's zpool allocation,
 356  * freeing the entry itself, and decrementing the number of stored pages.
 357  */
 358 static void zswap_free_entry(struct zswap_entry *entry)
 359 {
 360         if (!entry->length)
 361                 atomic_dec(&zswap_same_filled_pages);
 362         else {
 363                 zpool_free(entry->pool->zpool, entry->handle);
 364                 zswap_pool_put(entry->pool);
 365         }
 366         zswap_entry_cache_free(entry);
 367         atomic_dec(&zswap_stored_pages);
 368         zswap_update_total_size();
 369 }
 370
 371 /* caller must hold the tree lock */
 372 static void zswap_entry_get(struct zswap_entry *entry)
 373 {
 374         entry->refcount++;
 375 }
 376
 377 /* caller must hold the tree lock
 378 * remove from the tree and free it, if nobody reference the entry
 379 */
 380 static void zswap_entry_put(struct zswap_tree *tree,
 381                         struct zswap_entry *entry)
 382 {
 383         int refcount = --entry->refcount;
 384
 385         BUG_ON(refcount < 0);
 386         if (refcount == 0) {
 387                 zswap_rb_erase(&tree->rbroot, entry);
 388                 zswap_free_entry(entry);
 389         }
 390 }
 391
 392 /* caller must hold the tree lock */
 393 static struct zswap_entry *zswap_entry_find_get(struct rb_root *root,
 394                                 pgoff_t offset)
 395 {
 396         struct zswap_entry *entry;
 397
 398         entry = zswap_rb_search(root, offset);
 399         if (entry)
 400                 zswap_entry_get(entry);
 401
 402         return entry;
 403 }
 404
 405 /*********************************
 406 * per-cpu code
 407 **********************************/
 408 static DEFINE_PER_CPU(u8 *, zswap_dstmem);
 409 /*
 410  * If users dynamically change the zpool type and compressor at runtime, i.e.
 411  * zswap is running, zswap can have more than one zpool on one cpu, but they
 412  * are sharing dtsmem. So we need this mutex to be per-cpu.
 413  */
 414 static DEFINE_PER_CPU(struct mutex *, zswap_mutex);
 415
 416 static int zswap_dstmem_prepare(unsigned int cpu)
 417 {
 418         struct mutex *mutex;
 419         u8 *dst;
 420
 421         dst = kmalloc_node(PAGE_SIZE * 2, GFP_KERNEL, cpu_to_node(cpu));
 422         if (!dst)
 423                 return -ENOMEM;
 424
 425         mutex = kmalloc_node(sizeof(*mutex), GFP_KERNEL, cpu_to_node(cpu));
 426         if (!mutex) {
 427                 kfree(dst);
 428                 return -ENOMEM;
 429         }
 430
 431         mutex_init(mutex);
 432         per_cpu(zswap_dstmem, cpu) = dst;
 433         per_cpu(zswap_mutex, cpu) = mutex;
 434         return 0;
 435 }
 436
 437 static int zswap_dstmem_dead(unsigned int cpu)
 438 {
 439         struct mutex *mutex;
 440         u8 *dst;
 441
 442         mutex = per_cpu(zswap_mutex, cpu);
 443         kfree(mutex);
 444         per_cpu(zswap_mutex, cpu) = NULL;
 445
 446         dst = per_cpu(zswap_dstmem, cpu);
 447         kfree(dst);
 448         per_cpu(zswap_dstmem, cpu) = NULL;
 449
 450         return 0;
 451 }
 452
 453 static int zswap_cpu_comp_prepare(unsigned int cpu, struct hlist_node *node)
 454 {
 455         struct zswap_pool *pool = hlist_entry(node, struct zswap_pool, node);
 456         struct crypto_acomp_ctx *acomp_ctx = per_cpu_ptr(pool->acomp_ctx, cpu);
 457         struct crypto_acomp *acomp;
 458         struct acomp_req *req;
 459
 460         acomp = crypto_alloc_acomp_node(pool->tfm_name, 0, 0, cpu_to_node(cpu));
 461         if (IS_ERR(acomp)) {
 462                 pr_err("could not alloc crypto acomp %s : %ld\n",
 463                                 pool->tfm_name, PTR_ERR(acomp));
 464                 return PTR_ERR(acomp);
 465         }
 466         acomp_ctx->acomp = acomp;
 467
 468         req = acomp_request_alloc(acomp_ctx->acomp);
 469         if (!req) {
 470                 pr_err("could not alloc crypto acomp_request %s\n",
 471                        pool->tfm_name);
 472                 crypto_free_acomp(acomp_ctx->acomp);
 473                 return -ENOMEM;
 474         }
 475         acomp_ctx->req = req;
 476
 477         crypto_init_wait(&acomp_ctx->wait);
 478         /*
 479          * if the backend of acomp is async zip, crypto_req_done() will wakeup
 480          * crypto_wait_req(); if the backend of acomp is scomp, the callback
 481          * won't be called, crypto_wait_req() will return without blocking.
 482          */
 483         acomp_request_set_callback(req, CRYPTO_TFM_REQ_MAY_BACKLOG,
 484                                    crypto_req_done, &acomp_ctx->wait);
 485
 486         acomp_ctx->mutex = per_cpu(zswap_mutex, cpu);
 487         acomp_ctx->dstmem = per_cpu(zswap_dstmem, cpu);
 488
 489         return 0;
 490 }
 491
 492 static int zswap_cpu_comp_dead(unsigned int cpu, struct hlist_node *node)
 493 {
 494         struct zswap_pool *pool = hlist_entry(node, struct zswap_pool, node);
 495         struct crypto_acomp_ctx *acomp_ctx = per_cpu_ptr(pool->acomp_ctx, cpu);
 496
 497         if (!IS_ERR_OR_NULL(acomp_ctx)) {
 498                 if (!IS_ERR_OR_NULL(acomp_ctx->req))
 499                         acomp_request_free(acomp_ctx->req);
 500                 if (!IS_ERR_OR_NULL(acomp_ctx->acomp))
 501                         crypto_free_acomp(acomp_ctx->acomp);
 502         }
 503
 504         return 0;
 505 }
 506
 507 /*********************************
 508 * pool functions
 509 **********************************/
 510
 511 static struct zswap_pool *__zswap_pool_current(void)
 512 {
 513         struct zswap_pool *pool;
 514
 515         pool = list_first_or_null_rcu(&zswap_pools, typeof(*pool), list);
 516         WARN_ONCE(!pool && zswap_has_pool,
 517                   "%s: no page storage pool!\n", __func__);
 518
 519         return pool;
 520 }
 521
 522 static struct zswap_pool *zswap_pool_current(void)
 523 {
 524         assert_spin_locked(&zswap_pools_lock);
 525
 526         return __zswap_pool_current();
 527 }
 528
 529 static struct zswap_pool *zswap_pool_current_get(void)
 530 {
 531         struct zswap_pool *pool;
 532
 533         rcu_read_lock();
 534
 535         pool = __zswap_pool_current();
 536         if (!zswap_pool_get(pool))
 537                 pool = NULL;
 538
 539         rcu_read_unlock();
 540
 541         return pool;
 542 }
 543
 544 static struct zswap_pool *zswap_pool_last_get(void)
 545 {
 546         struct zswap_pool *pool, *last = NULL;
 547
 548         rcu_read_lock();
 549
 550         list_for_each_entry_rcu(pool, &zswap_pools, list)
 551                 last = pool;
 552         WARN_ONCE(!last && zswap_has_pool,
 553                   "%s: no page storage pool!\n", __func__);
 554         if (!zswap_pool_get(last))
 555                 last = NULL;
 556
 557         rcu_read_unlock();
 558
 559         return last;
 560 }
 561
 562 /* type and compressor must be null-terminated */
 563 static struct zswap_pool *zswap_pool_find_get(char *type, char *compressor)
 564 {
 565         struct zswap_pool *pool;
 566
 567         assert_spin_locked(&zswap_pools_lock);
 568
 569         list_for_each_entry_rcu(pool, &zswap_pools, list) {
 570                 if (strcmp(pool->tfm_name, compressor))
 571                         continue;
 572                 if (strcmp(zpool_get_type(pool->zpool), type))
 573                         continue;
 574                 /* if we can't get it, it's about to be destroyed */
 575                 if (!zswap_pool_get(pool))
 576                         continue;
 577                 return pool;
 578         }
 579
 580         return NULL;
 581 }
 582
 583 static void shrink_worker(struct work_struct *w)
 584 {
 585         struct zswap_pool *pool = container_of(w, typeof(*pool),
 586                                                 shrink_work);
 587
 588         if (zpool_shrink(pool->zpool, 1, NULL))
 589                 zswap_reject_reclaim_fail++;
 590         zswap_pool_put(pool);
 591 }
 592
 593 static struct zswap_pool *zswap_pool_create(char *type, char *compressor)
 594 {
 595         struct zswap_pool *pool;
 596         char name[38]; /* 'zswap' + 32 char (max) num + \0 */
 597         gfp_t gfp = __GFP_NORETRY | __GFP_NOWARN | __GFP_KSWAPD_RECLAIM;
 598         int ret;
 599
 600         if (!zswap_has_pool) {
 601                 /* if either are unset, pool initialization failed, and we
 602                  * need both params to be set correctly before trying to
 603                  * create a pool.
 604                  */
 605                 if (!strcmp(type, ZSWAP_PARAM_UNSET))
 606                         return NULL;
 607                 if (!strcmp(compressor, ZSWAP_PARAM_UNSET))
 608                         return NULL;
 609         }
 610
 611         pool = kzalloc(sizeof(*pool), GFP_KERNEL);
 612         if (!pool)
 613                 return NULL;
 614
 615         /* unique name for each pool specifically required by zsmalloc */
 616         snprintf(name, 38, "zswap%x", atomic_inc_return(&zswap_pools_count));
 617
 618         pool->zpool = zpool_create_pool(type, name, gfp, &zswap_zpool_ops);
 619         if (!pool->zpool) {
 620                 pr_err("%s zpool not available\n", type);
 621                 goto error;
 622         }
 623         pr_debug("using %s zpool\n", zpool_get_type(pool->zpool));
 624
 625         strscpy(pool->tfm_name, compressor, sizeof(pool->tfm_name));
 626
 627         pool->acomp_ctx = alloc_percpu(*pool->acomp_ctx);
 628         if (!pool->acomp_ctx) {
 629                 pr_err("percpu alloc failed\n");
 630                 goto error;
 631         }
 632
 633         ret = cpuhp_state_add_instance(CPUHP_MM_ZSWP_POOL_PREPARE,
 634                                        &pool->node);
 635         if (ret)
 636                 goto error;
 637         pr_debug("using %s compressor\n", pool->tfm_name);
 638
 639         /* being the current pool takes 1 ref; this func expects the
 640          * caller to always add the new pool as the current pool
 641          */
 642         kref_init(&pool->kref);
 643         INIT_LIST_HEAD(&pool->list);
 644         INIT_WORK(&pool->shrink_work, shrink_worker);
 645
 646         zswap_pool_debug("created", pool);
 647
 648         return pool;
 649
 650 error:
 651         if (pool->acomp_ctx)
 652                 free_percpu(pool->acomp_ctx);
 653         if (pool->zpool)
 654                 zpool_destroy_pool(pool->zpool);
 655         kfree(pool);
 656         return NULL;
 657 }
 658
 659 static __init struct zswap_pool *__zswap_pool_create_fallback(void)
 660 {
 661         bool has_comp, has_zpool;
 662
 663         has_comp = crypto_has_acomp(zswap_compressor, 0, 0);
 664         if (!has_comp && strcmp(zswap_compressor,
 665                                 CONFIG_ZSWAP_COMPRESSOR_DEFAULT)) {
 666                 pr_err("compressor %s not available, using default %s\n",
 667                        zswap_compressor, CONFIG_ZSWAP_COMPRESSOR_DEFAULT);
 668                 param_free_charp(&zswap_compressor);
 669                 zswap_compressor = CONFIG_ZSWAP_COMPRESSOR_DEFAULT;
 670                 has_comp = crypto_has_acomp(zswap_compressor, 0, 0);
 671         }
 672         if (!has_comp) {
 673                 pr_err("default compressor %s not available\n",
 674                        zswap_compressor);
 675                 param_free_charp(&zswap_compressor);
 676                 zswap_compressor = ZSWAP_PARAM_UNSET;
 677         }
 678
 679         has_zpool = zpool_has_pool(zswap_zpool_type);
 680         if (!has_zpool && strcmp(zswap_zpool_type,
 681                                  CONFIG_ZSWAP_ZPOOL_DEFAULT)) {
 682                 pr_err("zpool %s not available, using default %s\n",
 683                        zswap_zpool_type, CONFIG_ZSWAP_ZPOOL_DEFAULT);
 684                 param_free_charp(&zswap_zpool_type);
 685                 zswap_zpool_type = CONFIG_ZSWAP_ZPOOL_DEFAULT;
 686                 has_zpool = zpool_has_pool(zswap_zpool_type);
 687         }
 688         if (!has_zpool) {
 689                 pr_err("default zpool %s not available\n",
 690                        zswap_zpool_type);
 691                 param_free_charp(&zswap_zpool_type);
 692                 zswap_zpool_type = ZSWAP_PARAM_UNSET;
 693         }
 694
 695         if (!has_comp || !has_zpool)
 696                 return NULL;
 697
 698         return zswap_pool_create(zswap_zpool_type, zswap_compressor);
 699 }
 700
 701 static void zswap_pool_destroy(struct zswap_pool *pool)
 702 {
 703         zswap_pool_debug("destroying", pool);
 704
 705         cpuhp_state_remove_instance(CPUHP_MM_ZSWP_POOL_PREPARE, &pool->node);
 706         free_percpu(pool->acomp_ctx);
 707         zpool_destroy_pool(pool->zpool);
 708         kfree(pool);
 709 }
 710
 711 static int __must_check zswap_pool_get(struct zswap_pool *pool)
 712 {
 713         if (!pool)
 714                 return 0;
 715
 716         return kref_get_unless_zero(&pool->kref);
 717 }
 718
 719 static void __zswap_pool_release(struct work_struct *work)
 720 {
 721         struct zswap_pool *pool = container_of(work, typeof(*pool),
 722                                                 release_work);
 723
 724         synchronize_rcu();
 725
 726         /* nobody should have been able to get a kref... */
 727         WARN_ON(kref_get_unless_zero(&pool->kref));
 728
 729         /* pool is now off zswap_pools list and has no references. */
 730         zswap_pool_destroy(pool);
 731 }
 732
 733 static void __zswap_pool_empty(struct kref *kref)
 734 {
 735         struct zswap_pool *pool;
 736
 737         pool = container_of(kref, typeof(*pool), kref);
 738
 739         spin_lock(&zswap_pools_lock);
 740
 741         WARN_ON(pool == zswap_pool_current());
 742
 743         list_del_rcu(&pool->list);
 744
 745         INIT_WORK(&pool->release_work, __zswap_pool_release);
 746         schedule_work(&pool->release_work);
 747
 748         spin_unlock(&zswap_pools_lock);
 749 }
 750
 751 static void zswap_pool_put(struct zswap_pool *pool)
 752 {
 753         kref_put(&pool->kref, __zswap_pool_empty);
 754 }
 755
 756 /*********************************
 757 * param callbacks
 758 **********************************/
 759
 760 /* val must be a null-terminated string */
 761 static int __zswap_param_set(const char *val, const struct kernel_param *kp,
 762                              char *type, char *compressor)
 763 {
 764         struct zswap_pool *pool, *put_pool = NULL;
 765         char *s = strstrip((char *)val);
 766         int ret;
 767
 768         if (zswap_init_failed) {
 769                 pr_err("can't set param, initialization failed\n");
 770                 return -ENODEV;
 771         }
 772
 773         /* no change required */
 774         if (!strcmp(s, *(char **)kp->arg) && zswap_has_pool)
 775                 return 0;
 776
 777         /* if this is load-time (pre-init) param setting,
 778          * don't create a pool; that's done during init.
 779          */
 780         if (!zswap_init_started)
 781                 return param_set_charp(s, kp);
 782
 783         if (!type) {
 784                 if (!zpool_has_pool(s)) {
 785                         pr_err("zpool %s not available\n", s);
 786                         return -ENOENT;
 787                 }
 788                 type = s;
 789         } else if (!compressor) {
 790                 if (!crypto_has_acomp(s, 0, 0)) {
 791                         pr_err("compressor %s not available\n", s);
 792                         return -ENOENT;
 793                 }
 794                 compressor = s;
 795         } else {
 796                 WARN_ON(1);
 797                 return -EINVAL;
 798         }
 799
 800         spin_lock(&zswap_pools_lock);
 801
 802         pool = zswap_pool_find_get(type, compressor);
 803         if (pool) {
 804                 zswap_pool_debug("using existing", pool);
 805                 WARN_ON(pool == zswap_pool_current());
 806                 list_del_rcu(&pool->list);
 807         }
 808
 809         spin_unlock(&zswap_pools_lock);
 810
 811         if (!pool)
 812                 pool = zswap_pool_create(type, compressor);
 813
 814         if (pool)
 815                 ret = param_set_charp(s, kp);
 816         else
 817                 ret = -EINVAL;
 818
 819         spin_lock(&zswap_pools_lock);
 820
 821         if (!ret) {
 822                 put_pool = zswap_pool_current();
 823                 list_add_rcu(&pool->list, &zswap_pools);
 824                 zswap_has_pool = true;
 825         } else if (pool) {
 826                 /* add the possibly pre-existing pool to the end of the pools
 827                  * list; if it's new (and empty) then it'll be removed and
 828                  * destroyed by the put after we drop the lock
 829                  */
 830                 list_add_tail_rcu(&pool->list, &zswap_pools);
 831                 put_pool = pool;
 832         }
 833
 834         spin_unlock(&zswap_pools_lock);
 835
 836         if (!zswap_has_pool && !pool) {
 837                 /* if initial pool creation failed, and this pool creation also
 838                  * failed, maybe both compressor and zpool params were bad.
 839                  * Allow changing this param, so pool creation will succeed
 840                  * when the other param is changed. We already verified this
 841                  * param is ok in the zpool_has_pool() or crypto_has_acomp()
 842                  * checks above.
 843                  */
 844                 ret = param_set_charp(s, kp);
 845         }
 846
 847         /* drop the ref from either the old current pool,
 848          * or the new pool we failed to add
 849          */
 850         if (put_pool)
 851                 zswap_pool_put(put_pool);
 852
 853         return ret;
 854 }
 855
 856 static int zswap_compressor_param_set(const char *val,
 857                                       const struct kernel_param *kp)
 858 {
 859         return __zswap_param_set(val, kp, zswap_zpool_type, NULL);
 860 }
 861
 862 static int zswap_zpool_param_set(const char *val,
 863                                  const struct kernel_param *kp)
 864 {
 865         return __zswap_param_set(val, kp, NULL, zswap_compressor);
 866 }
 867
 868 static int zswap_enabled_param_set(const char *val,
 869                                    const struct kernel_param *kp)
 870 {
 871         if (zswap_init_failed) {
 872                 pr_err("can't enable, initialization failed\n");
 873                 return -ENODEV;
 874         }
 875         if (!zswap_has_pool && zswap_init_started) {
 876                 pr_err("can't enable, no pool configured\n");
 877                 return -ENODEV;
 878         }
 879
 880         return param_set_bool(val, kp);
 881 }
 882
 883 /*********************************
 884 * writeback code
 885 **********************************/
 886 /* return enum for zswap_get_swap_cache_page */
 887 enum zswap_get_swap_ret {
 888         ZSWAP_SWAPCACHE_NEW,
 889         ZSWAP_SWAPCACHE_EXIST,
 890         ZSWAP_SWAPCACHE_FAIL,
 891 };
 892
 893 /*
 894  * zswap_get_swap_cache_page
 895  *
 896  * This is an adaption of read_swap_cache_async()
 897  *
 898  * This function tries to find a page with the given swap entry
 899  * in the swapper_space address space (the swap cache).  If the page
 900  * is found, it is returned in retpage.  Otherwise, a page is allocated,
 901  * added to the swap cache, and returned in retpage.
 902  *
 903  * If success, the swap cache page is returned in retpage
 904  * Returns ZSWAP_SWAPCACHE_EXIST if page was already in the swap cache
 905  * Returns ZSWAP_SWAPCACHE_NEW if the new page needs to be populated,
 906  *     the new page is added to swapcache and locked
 907  * Returns ZSWAP_SWAPCACHE_FAIL on error
 908  */
 909 static int zswap_get_swap_cache_page(swp_entry_t entry,
 910                                 struct page **retpage)
 911 {
 912         bool page_was_allocated;
 913
 914         *retpage = __read_swap_cache_async(entry, GFP_KERNEL,
 915                         NULL, 0, &page_was_allocated);
 916         if (page_was_allocated)
 917                 return ZSWAP_SWAPCACHE_NEW;
 918         if (!*retpage)
 919                 return ZSWAP_SWAPCACHE_FAIL;
 920         return ZSWAP_SWAPCACHE_EXIST;
 921 }
 922
 923 /*
 924  * Attempts to free an entry by adding a page to the swap cache,
 925  * decompressing the entry data into the page, and issuing a
 926  * bio write to write the page back to the swap device.
 927  *
 928  * This can be thought of as a "resumed writeback" of the page
 929  * to the swap device.  We are basically resuming the same swap
 930  * writeback path that was intercepted with the frontswap_store()
 931  * in the first place.  After the page has been decompressed into
 932  * the swap cache, the compressed version stored by zswap can be
 933  * freed.
 934  */
 935 static int zswap_writeback_entry(struct zpool *pool, unsigned long handle)
 936 {
 937         struct zswap_header *zhdr;
 938         swp_entry_t swpentry;
 939         struct zswap_tree *tree;
 940         pgoff_t offset;
 941         struct zswap_entry *entry;
 942         struct page *page;
 943         struct scatterlist input, output;
 944         struct crypto_acomp_ctx *acomp_ctx;
 945
 946         u8 *src, *tmp = NULL;
 947         unsigned int dlen;
 948         int ret;
 949         struct writeback_control wbc = {
 950                 .sync_mode = WB_SYNC_NONE,
 951         };
 952
 953         if (!zpool_can_sleep_mapped(pool)) {
 954                 tmp = kmalloc(PAGE_SIZE, GFP_ATOMIC);
 955                 if (!tmp)
 956                         return -ENOMEM;
 957         }
 958
 959         /* extract swpentry from data */
 960         zhdr = zpool_map_handle(pool, handle, ZPOOL_MM_RO);
 961         swpentry = zhdr->swpentry; /* here */
 962         tree = zswap_trees[swp_type(swpentry)];
 963         offset = swp_offset(swpentry);
 964
 965         /* find and ref zswap entry */
 966         spin_lock(&tree->lock);
 967         entry = zswap_entry_find_get(&tree->rbroot, offset);
 968         if (!entry) {
 969                 /* entry was invalidated */
 970                 spin_unlock(&tree->lock);
 971                 zpool_unmap_handle(pool, handle);
 972                 kfree(tmp);
 973                 return 0;
 974         }
 975         spin_unlock(&tree->lock);
 976         BUG_ON(offset != entry->offset);
 977
 978         src = (u8 *)zhdr + sizeof(struct zswap_header);
 979         if (!zpool_can_sleep_mapped(pool)) {
 980                 memcpy(tmp, src, entry->length);
 981                 src = tmp;
 982                 zpool_unmap_handle(pool, handle);
 983         }
 984
 985         /* try to allocate swap cache page */
 986         switch (zswap_get_swap_cache_page(swpentry, &page)) {
 987         case ZSWAP_SWAPCACHE_FAIL: /* no memory or invalidate happened */
 988                 ret = -ENOMEM;
 989                 goto fail;
 990
 991         case ZSWAP_SWAPCACHE_EXIST:
 992                 /* page is already in the swap cache, ignore for now */
 993                 put_page(page);
 994                 ret = -EEXIST;
 995                 goto fail;
 996
 997         case ZSWAP_SWAPCACHE_NEW: /* page is locked */
 998                 /* decompress */
 999                 acomp_ctx = raw_cpu_ptr(entry->pool->acomp_ctx);
1000                 dlen = PAGE_SIZE;
1001
1002                 mutex_lock(acomp_ctx->mutex);
1003                 sg_init_one(&input, src, entry->length);
1004                 sg_init_table(&output, 1);
1005                 sg_set_page(&output, page, PAGE_SIZE, 0);
1006                 acomp_request_set_params(acomp_ctx->req, &input, &output, entry->length, dlen);
1007                 ret = crypto_wait_req(crypto_acomp_decompress(acomp_ctx->req), &acomp_ctx->wait);
1008                 dlen = acomp_ctx->req->dlen;
1009                 mutex_unlock(acomp_ctx->mutex);
1010
1011                 BUG_ON(ret);
1012                 BUG_ON(dlen != PAGE_SIZE);
1013
1014                 /* page is up to date */
1015                 SetPageUptodate(page);
1016         }
1017
1018         /* move it to the tail of the inactive list after end_writeback */
1019         SetPageReclaim(page);
1020
1021         /* start writeback */
1022         __swap_writepage(page, &wbc, end_swap_bio_write);
1023         put_page(page);
1024         zswap_written_back_pages++;
1025
1026         spin_lock(&tree->lock);
1027         /* drop local reference */
1028         zswap_entry_put(tree, entry);
1029
1030         /*
1031         * There are two possible situations for entry here:
1032         * (1) refcount is 1(normal case),  entry is valid and on the tree
1033         * (2) refcount is 0, entry is freed and not on the tree
1034         *     because invalidate happened during writeback
1035         *  search the tree and free the entry if find entry
1036         */
1037         if (entry == zswap_rb_search(&tree->rbroot, offset))
1038                 zswap_entry_put(tree, entry);
1039         spin_unlock(&tree->lock);
1040
1041         goto end;
1042
1043         /*
1044         * if we get here due to ZSWAP_SWAPCACHE_EXIST
1045         * a load may be happening concurrently.
1046         * it is safe and okay to not free the entry.
1047         * if we free the entry in the following put
1048         * it is also okay to return !0
1049         */
1050 fail:
1051         spin_lock(&tree->lock);
1052         zswap_entry_put(tree, entry);
1053         spin_unlock(&tree->lock);
1054
1055 end:
1056         if (zpool_can_sleep_mapped(pool))
1057                 zpool_unmap_handle(pool, handle);
1058         else
1059                 kfree(tmp);
1060
1061         return ret;
1062 }
1063
1064 static int zswap_is_page_same_filled(void *ptr, unsigned long *value)
1065 {
1066         unsigned int pos;
1067         unsigned long *page;
1068
1069         page = (unsigned long *)ptr;
1070         for (pos = 1; pos < PAGE_SIZE / sizeof(*page); pos++) {
1071                 if (page[pos] != page[0])
1072                         return 0;
1073         }
1074         *value = page[0];
1075         return 1;
1076 }
1077
1078 static void zswap_fill_page(void *ptr, unsigned long value)
1079 {
1080         unsigned long *page;
1081
1082         page = (unsigned long *)ptr;
1083         memset_l(page, value, PAGE_SIZE / sizeof(unsigned long));
1084 }
1085
1086 /*********************************
1087 * frontswap hooks
1088 **********************************/
1089 /* attempts to compress and store an single page */
1090 static int zswap_frontswap_store(unsigned type, pgoff_t offset,
1091                                 struct page *page)
1092 {
1093         struct zswap_tree *tree = zswap_trees[type];
1094         struct zswap_entry *entry, *dupentry;
1095         struct scatterlist input, output;
1096         struct crypto_acomp_ctx *acomp_ctx;
1097         int ret;
1098         unsigned int hlen, dlen = PAGE_SIZE;
1099         unsigned long handle, value;
1100         char *buf;
1101         u8 *src, *dst;
1102         struct zswap_header zhdr = { .swpentry = swp_entry(type, offset) };
1103         gfp_t gfp;
1104
1105         /* THP isn't supported */
1106         if (PageTransHuge(page)) {
1107                 ret = -EINVAL;
1108                 goto reject;
1109         }
1110
1111         if (!zswap_enabled || !tree) {
1112                 ret = -ENODEV;
1113                 goto reject;
1114         }
1115
1116         /* reclaim space if needed */
1117         if (zswap_is_full()) {
1118                 struct zswap_pool *pool;
1119
1120                 zswap_pool_limit_hit++;
1121                 zswap_pool_reached_full = true;
1122                 pool = zswap_pool_last_get();
1123                 if (pool)
1124                         queue_work(shrink_wq, &pool->shrink_work);
1125                 ret = -ENOMEM;
1126                 goto reject;
1127         }
1128
1129         if (zswap_pool_reached_full) {
1130                if (!zswap_can_accept()) {
1131                         ret = -ENOMEM;
1132                         goto reject;
1133                 } else
1134                         zswap_pool_reached_full = false;
1135         }
1136
1137         /* allocate entry */
1138         entry = zswap_entry_cache_alloc(GFP_KERNEL);
1139         if (!entry) {
1140                 zswap_reject_kmemcache_fail++;
1141                 ret = -ENOMEM;
1142                 goto reject;
1143         }
1144
1145         if (zswap_same_filled_pages_enabled) {
1146                 src = kmap_atomic(page);
1147                 if (zswap_is_page_same_filled(src, &value)) {
1148                         kunmap_atomic(src);
1149                         entry->offset = offset;
1150                         entry->length = 0;
1151                         entry->value = value;
1152                         atomic_inc(&zswap_same_filled_pages);
1153                         goto insert_entry;
1154                 }
1155                 kunmap_atomic(src);
1156         }
1157
1158         if (!zswap_non_same_filled_pages_enabled) {
1159                 ret = -EINVAL;
1160                 goto freepage;
1161         }
1162
1163         /* if entry is successfully added, it keeps the reference */
1164         entry->pool = zswap_pool_current_get();
1165         if (!entry->pool) {
1166                 ret = -EINVAL;
1167                 goto freepage;
1168         }
1169
1170         /* compress */
1171         acomp_ctx = raw_cpu_ptr(entry->pool->acomp_ctx);
1172
1173         mutex_lock(acomp_ctx->mutex);
1174
1175         dst = acomp_ctx->dstmem;
1176         sg_init_table(&input, 1);
1177         sg_set_page(&input, page, PAGE_SIZE, 0);
1178
1179         /* zswap_dstmem is of size (PAGE_SIZE * 2). Reflect same in sg_list */
1180         sg_init_one(&output, dst, PAGE_SIZE * 2);
1181         acomp_request_set_params(acomp_ctx->req, &input, &output, PAGE_SIZE, dlen);
1182         /*
1183          * it maybe looks a little bit silly that we send an asynchronous request,
1184          * then wait for its completion synchronously. This makes the process look
1185          * synchronous in fact.
1186          * Theoretically, acomp supports users send multiple acomp requests in one
1187          * acomp instance, then get those requests done simultaneously. but in this
1188          * case, frontswap actually does store and load page by page, there is no
1189          * existing method to send the second page before the first page is done
1190          * in one thread doing frontswap.
1191          * but in different threads running on different cpu, we have different
1192          * acomp instance, so multiple threads can do (de)compression in parallel.
1193          */
1194         ret = crypto_wait_req(crypto_acomp_compress(acomp_ctx->req), &acomp_ctx->wait);
1195         dlen = acomp_ctx->req->dlen;
1196
1197         if (ret) {
1198                 ret = -EINVAL;
1199                 goto put_dstmem;
1200         }
1201
1202         /* store */
1203         hlen = zpool_evictable(entry->pool->zpool) ? sizeof(zhdr) : 0;
1204         gfp = __GFP_NORETRY | __GFP_NOWARN | __GFP_KSWAPD_RECLAIM;
1205         if (zpool_malloc_support_movable(entry->pool->zpool))
1206                 gfp |= __GFP_HIGHMEM | __GFP_MOVABLE;
1207         ret = zpool_malloc(entry->pool->zpool, hlen + dlen, gfp, &handle);
1208         if (ret == -ENOSPC) {
1209                 zswap_reject_compress_poor++;
1210                 goto put_dstmem;
1211         }
1212         if (ret) {
1213                 zswap_reject_alloc_fail++;
1214                 goto put_dstmem;
1215         }
1216         buf = zpool_map_handle(entry->pool->zpool, handle, ZPOOL_MM_WO);
1217         memcpy(buf, &zhdr, hlen);
1218         memcpy(buf + hlen, dst, dlen);
1219         zpool_unmap_handle(entry->pool->zpool, handle);
1220         mutex_unlock(acomp_ctx->mutex);
1221
1222         /* populate entry */
1223         entry->offset = offset;
1224         entry->handle = handle;
1225         entry->length = dlen;
1226
1227 insert_entry:
1228         /* map */
1229         spin_lock(&tree->lock);
1230         do {
1231                 ret = zswap_rb_insert(&tree->rbroot, entry, &dupentry);
1232                 if (ret == -EEXIST) {
1233                         zswap_duplicate_entry++;
1234                         /* remove from rbtree */
1235                         zswap_rb_erase(&tree->rbroot, dupentry);
1236                         zswap_entry_put(tree, dupentry);
1237                 }
1238         } while (ret == -EEXIST);
1239         spin_unlock(&tree->lock);
1240
1241         /* update stats */
1242         atomic_inc(&zswap_stored_pages);
1243         zswap_update_total_size();
1244
1245         return 0;
1246
1247 put_dstmem:
1248         mutex_unlock(acomp_ctx->mutex);
1249         zswap_pool_put(entry->pool);
1250 freepage:
1251         zswap_entry_cache_free(entry);
1252 reject:
1253         return ret;
1254 }
1255
1256 /*
1257  * returns 0 if the page was successfully decompressed
1258  * return -1 on entry not found or error
1259 */
1260 static int zswap_frontswap_load(unsigned type, pgoff_t offset,
1261                                 struct page *page)
1262 {
1263         struct zswap_tree *tree = zswap_trees[type];
1264         struct zswap_entry *entry;
1265         struct scatterlist input, output;
1266         struct crypto_acomp_ctx *acomp_ctx;
1267         u8 *src, *dst, *tmp;
1268         unsigned int dlen;
1269         int ret;
1270
1271         /* find */
1272         spin_lock(&tree->lock);
1273         entry = zswap_entry_find_get(&tree->rbroot, offset);
1274         if (!entry) {
1275                 /* entry was written back */
1276                 spin_unlock(&tree->lock);
1277                 return -1;
1278         }
1279         spin_unlock(&tree->lock);
1280
1281         if (!entry->length) {
1282                 dst = kmap_atomic(page);
1283                 zswap_fill_page(dst, entry->value);
1284                 kunmap_atomic(dst);
1285                 ret = 0;
1286                 goto freeentry;
1287         }
1288
1289         if (!zpool_can_sleep_mapped(entry->pool->zpool)) {
1290
1291                 tmp = kmalloc(entry->length, GFP_ATOMIC);
1292                 if (!tmp) {
1293                         ret = -ENOMEM;
1294                         goto freeentry;
1295                 }
1296         }
1297
1298         /* decompress */
1299         dlen = PAGE_SIZE;
1300         src = zpool_map_handle(entry->pool->zpool, entry->handle, ZPOOL_MM_RO);
1301         if (zpool_evictable(entry->pool->zpool))
1302                 src += sizeof(struct zswap_header);
1303
1304         if (!zpool_can_sleep_mapped(entry->pool->zpool)) {
1305
1306                 memcpy(tmp, src, entry->length);
1307                 src = tmp;
1308
1309                 zpool_unmap_handle(entry->pool->zpool, entry->handle);
1310         }
1311
1312         acomp_ctx = raw_cpu_ptr(entry->pool->acomp_ctx);
1313         mutex_lock(acomp_ctx->mutex);
1314         sg_init_one(&input, src, entry->length);
1315         sg_init_table(&output, 1);
1316         sg_set_page(&output, page, PAGE_SIZE, 0);
1317         acomp_request_set_params(acomp_ctx->req, &input, &output, entry->length, dlen);
1318         ret = crypto_wait_req(crypto_acomp_decompress(acomp_ctx->req), &acomp_ctx->wait);
1319         mutex_unlock(acomp_ctx->mutex);
1320
1321         if (zpool_can_sleep_mapped(entry->pool->zpool))
1322                 zpool_unmap_handle(entry->pool->zpool, entry->handle);
1323         else
1324                 kfree(tmp);
1325
1326         BUG_ON(ret);
1327
1328 freeentry:
1329         spin_lock(&tree->lock);
1330         zswap_entry_put(tree, entry);
1331         spin_unlock(&tree->lock);
1332
1333         return ret;
1334 }
1335
1336 /* frees an entry in zswap */
1337 static void zswap_frontswap_invalidate_page(unsigned type, pgoff_t offset)
1338 {
1339         struct zswap_tree *tree = zswap_trees[type];
1340         struct zswap_entry *entry;
1341
1342         /* find */
1343         spin_lock(&tree->lock);
1344         entry = zswap_rb_search(&tree->rbroot, offset);
1345         if (!entry) {
1346                 /* entry was written back */
1347                 spin_unlock(&tree->lock);
1348                 return;
1349         }
1350
1351         /* remove from rbtree */
1352         zswap_rb_erase(&tree->rbroot, entry);
1353
1354         /* drop the initial reference from entry creation */
1355         zswap_entry_put(tree, entry);
1356
1357         spin_unlock(&tree->lock);
1358 }
1359
1360 /* frees all zswap entries for the given swap type */
1361 static void zswap_frontswap_invalidate_area(unsigned type)
1362 {
1363         struct zswap_tree *tree = zswap_trees[type];
1364         struct zswap_entry *entry, *n;
1365
1366         if (!tree)
1367                 return;
1368
1369         /* walk the tree and free everything */
1370         spin_lock(&tree->lock);
1371         rbtree_postorder_for_each_entry_safe(entry, n, &tree->rbroot, rbnode)
1372                 zswap_free_entry(entry);
1373         tree->rbroot = RB_ROOT;
1374         spin_unlock(&tree->lock);
1375         kfree(tree);
1376         zswap_trees[type] = NULL;
1377 }
1378
1379 static void zswap_frontswap_init(unsigned type)
1380 {
1381         struct zswap_tree *tree;
1382
1383         tree = kzalloc(sizeof(*tree), GFP_KERNEL);
1384         if (!tree) {
1385                 pr_err("alloc failed, zswap disabled for swap type %d\n", type);
1386                 return;
1387         }
1388
1389         tree->rbroot = RB_ROOT;
1390         spin_lock_init(&tree->lock);
1391         zswap_trees[type] = tree;
1392 }
1393
1394 static const struct frontswap_ops zswap_frontswap_ops = {
1395         .store = zswap_frontswap_store,
1396         .load = zswap_frontswap_load,
1397         .invalidate_page = zswap_frontswap_invalidate_page,
1398         .invalidate_area = zswap_frontswap_invalidate_area,
1399         .init = zswap_frontswap_init
1400 };
1401
1402 /*********************************
1403 * debugfs functions
1404 **********************************/
1405 #ifdef CONFIG_DEBUG_FS
1406 #include <linux/debugfs.h>
1407
1408 static struct dentry *zswap_debugfs_root;
1409
1410 static int __init zswap_debugfs_init(void)
1411 {
1412         if (!debugfs_initialized())
1413                 return -ENODEV;
1414
1415         zswap_debugfs_root = debugfs_create_dir("zswap", NULL);
1416
1417         debugfs_create_u64("pool_limit_hit", 0444,
1418                            zswap_debugfs_root, &zswap_pool_limit_hit);
1419         debugfs_create_u64("reject_reclaim_fail", 0444,
1420                            zswap_debugfs_root, &zswap_reject_reclaim_fail);
1421         debugfs_create_u64("reject_alloc_fail", 0444,
1422                            zswap_debugfs_root, &zswap_reject_alloc_fail);
1423         debugfs_create_u64("reject_kmemcache_fail", 0444,
1424                            zswap_debugfs_root, &zswap_reject_kmemcache_fail);
1425         debugfs_create_u64("reject_compress_poor", 0444,
1426                            zswap_debugfs_root, &zswap_reject_compress_poor);
1427         debugfs_create_u64("written_back_pages", 0444,
1428                            zswap_debugfs_root, &zswap_written_back_pages);
1429         debugfs_create_u64("duplicate_entry", 0444,
1430                            zswap_debugfs_root, &zswap_duplicate_entry);
1431         debugfs_create_u64("pool_total_size", 0444,
1432                            zswap_debugfs_root, &zswap_pool_total_size);
1433         debugfs_create_atomic_t("stored_pages", 0444,
1434                                 zswap_debugfs_root, &zswap_stored_pages);
1435         debugfs_create_atomic_t("same_filled_pages", 0444,
1436                                 zswap_debugfs_root, &zswap_same_filled_pages);
1437
1438         return 0;
1439 }
1440 #else
1441 static int __init zswap_debugfs_init(void)
1442 {
1443         return 0;
1444 }
1445 #endif
1446
1447 /*********************************
1448 * module init and exit
1449 **********************************/
1450 static int __init init_zswap(void)
1451 {
1452         struct zswap_pool *pool;
1453         int ret;
1454
1455         zswap_init_started = true;
1456
1457         if (zswap_entry_cache_create()) {
1458                 pr_err("entry cache creation failed\n");
1459                 goto cache_fail;
1460         }
1461
1462         ret = cpuhp_setup_state(CPUHP_MM_ZSWP_MEM_PREPARE, "mm/zswap:prepare",
1463                                 zswap_dstmem_prepare, zswap_dstmem_dead);
1464         if (ret) {
1465                 pr_err("dstmem alloc failed\n");
1466                 goto dstmem_fail;
1467         }
1468
1469         ret = cpuhp_setup_state_multi(CPUHP_MM_ZSWP_POOL_PREPARE,
1470                                       "mm/zswap_pool:prepare",
1471                                       zswap_cpu_comp_prepare,
1472                                       zswap_cpu_comp_dead);
1473         if (ret)
1474                 goto hp_fail;
1475
1476         pool = __zswap_pool_create_fallback();
1477         if (pool) {
1478                 pr_info("loaded using pool %s/%s\n", pool->tfm_name,
1479                         zpool_get_type(pool->zpool));
1480                 list_add(&pool->list, &zswap_pools);
1481                 zswap_has_pool = true;
1482         } else {
1483                 pr_err("pool creation failed\n");
1484                 zswap_enabled = false;
1485         }
1486
1487         shrink_wq = create_workqueue("zswap-shrink");
1488         if (!shrink_wq)
1489                 goto fallback_fail;
1490
1491         ret = frontswap_register_ops(&zswap_frontswap_ops);
1492         if (ret)
1493                 goto destroy_wq;
1494         if (zswap_debugfs_init())
1495                 pr_warn("debugfs initialization failed\n");
1496         return 0;
1497
1498 destroy_wq:
1499         destroy_workqueue(shrink_wq);
1500 fallback_fail:
1501         if (pool)
1502                 zswap_pool_destroy(pool);
1503 hp_fail:
1504         cpuhp_remove_state(CPUHP_MM_ZSWP_MEM_PREPARE);
1505 dstmem_fail:
1506         zswap_entry_cache_destroy();
1507 cache_fail:
1508         /* if built-in, we aren't unloaded on failure; don't allow use */
1509         zswap_init_failed = true;
1510         zswap_enabled = false;
1511         return -ENOMEM;
1512 }
1513 /* must be late so crypto has time to come up */
1514 late_initcall(init_zswap);
1515
1516 MODULE_LICENSE("GPL");
1517 MODULE_AUTHOR("Seth Jennings <sjennings@variantweb.net>");
1518 MODULE_DESCRIPTION("Compressed cache for swap pages");