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