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 <[email protected]>
  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 #include "swap.h"
  40
  41 /*********************************
  42 * statistics
  43 **********************************/
  44 /* Total bytes used by the compressed storage */
  45 static u64 zswap_pool_total_size;
  46 /* The number of compressed pages currently stored in zswap */
  47 static atomic_t zswap_stored_pages = ATOMIC_INIT(0);
  48 /* The number of same-value filled pages currently stored in zswap */
  49 static atomic_t zswap_same_filled_pages = ATOMIC_INIT(0);
  50
  51 /*
  52  * The statistics below are not protected from concurrent access for
  53  * performance reasons so they may not be a 100% accurate.  However,
  54  * they do provide useful information on roughly how many times a
  55  * certain event is occurring.
  56 */
  57
  58 /* Pool limit was hit (see zswap_max_pool_percent) */
  59 static u64 zswap_pool_limit_hit;
  60 /* Pages written back when pool limit was reached */
  61 static u64 zswap_written_back_pages;
  62 /* Store failed due to a reclaim failure after pool limit was reached */
  63 static u64 zswap_reject_reclaim_fail;
  64 /* Compressed page was too big for the allocator to (optimally) store */
  65 static u64 zswap_reject_compress_poor;
  66 /* Store failed because underlying allocator could not get memory */
  67 static u64 zswap_reject_alloc_fail;
  68 /* Store failed because the entry metadata could not be allocated (rare) */
  69 static u64 zswap_reject_kmemcache_fail;
  70 /* Duplicate store was encountered (rare) */
  71 static u64 zswap_duplicate_entry;
  72
  73 /* Shrinker work queue */
  74 static struct workqueue_struct *shrink_wq;
  75 /* Pool limit was hit, we need to calm down */
  76 static bool zswap_pool_reached_full;
  77
  78 /*********************************
  79 * tunables
  80 **********************************/
  81
  82 #define ZSWAP_PARAM_UNSET ""
  83
  84 /* Enable/disable zswap */
  85 static bool zswap_enabled = IS_ENABLED(CONFIG_ZSWAP_DEFAULT_ON);
  86 static int zswap_enabled_param_set(const char *,
  87                                    const struct kernel_param *);
  88 static const struct kernel_param_ops zswap_enabled_param_ops = {
  89         .set =          zswap_enabled_param_set,
  90         .get =          param_get_bool,
  91 };
  92 module_param_cb(enabled, &zswap_enabled_param_ops, &zswap_enabled, 0644);
  93
  94 /* Crypto compressor to use */
  95 static char *zswap_compressor = CONFIG_ZSWAP_COMPRESSOR_DEFAULT;
  96 static int zswap_compressor_param_set(const char *,
  97                                       const struct kernel_param *);
  98 static const struct kernel_param_ops zswap_compressor_param_ops = {
  99         .set =          zswap_compressor_param_set,
 100         .get =          param_get_charp,
 101         .free =         param_free_charp,
 102 };
 103 module_param_cb(compressor, &zswap_compressor_param_ops,
 104                 &zswap_compressor, 0644);
 105
 106 /* Compressed storage zpool to use */
 107 static char *zswap_zpool_type = CONFIG_ZSWAP_ZPOOL_DEFAULT;
 108 static int zswap_zpool_param_set(const char *, const struct kernel_param *);
 109 static const struct kernel_param_ops zswap_zpool_param_ops = {
 110         .set =          zswap_zpool_param_set,
 111         .get =          param_get_charp,
 112         .free =         param_free_charp,
 113 };
 114 module_param_cb(zpool, &zswap_zpool_param_ops, &zswap_zpool_type, 0644);
 115
 116 /* The maximum percentage of memory that the compressed pool can occupy */
 117 static unsigned int zswap_max_pool_percent = 20;
 118 module_param_named(max_pool_percent, zswap_max_pool_percent, uint, 0644);
 119
 120 /* The threshold for accepting new pages after the max_pool_percent was hit */
 121 static unsigned int zswap_accept_thr_percent = 90; /* of max pool size */
 122 module_param_named(accept_threshold_percent, zswap_accept_thr_percent,
 123                    uint, 0644);
 124
 125 /*
 126  * Enable/disable handling same-value filled pages (enabled by default).
 127  * If disabled every page is considered non-same-value filled.
 128  */
 129 static bool zswap_same_filled_pages_enabled = true;
 130 module_param_named(same_filled_pages_enabled, zswap_same_filled_pages_enabled,
 131                    bool, 0644);
 132
 133 /* Enable/disable handling non-same-value filled pages (enabled by default) */
 134 static bool zswap_non_same_filled_pages_enabled = true;
 135 module_param_named(non_same_filled_pages_enabled, zswap_non_same_filled_pages_enabled,
 136                    bool, 0644);
 137
 138 /*********************************
 139 * data structures
 140 **********************************/
 141
 142 struct crypto_acomp_ctx {
 143         struct crypto_acomp *acomp;
 144         struct acomp_req *req;
 145         struct crypto_wait wait;
 146         u8 *dstmem;
 147         struct mutex *mutex;
 148 };
 149
 150 struct zswap_pool {
 151         struct zpool *zpool;
 152         struct crypto_acomp_ctx __percpu *acomp_ctx;
 153         struct kref kref;
 154         struct list_head list;
 155         struct work_struct release_work;
 156         struct work_struct shrink_work;
 157         struct hlist_node node;
 158         char tfm_name[CRYPTO_MAX_ALG_NAME];
 159 };
 160
 161 /*
 162  * struct zswap_entry
 163  *
 164  * This structure contains the metadata for tracking a single compressed
 165  * page within zswap.
 166  *
 167  * rbnode - links the entry into red-black tree for the appropriate swap type
 168  * offset - the swap offset for the entry.  Index into the red-black tree.
 169  * refcount - the number of outstanding reference to the entry. This is needed
 170  *            to protect against premature freeing of the entry by code
 171  *            concurrent calls to load, invalidate, and writeback.  The lock
 172  *            for the zswap_tree structure that contains the entry must
 173  *            be held while changing the refcount.  Since the lock must
 174  *            be held, there is no reason to also make refcount atomic.
 175  * length - the length in bytes of the compressed page data.  Needed during
 176  *          decompression. For a same value filled page length is 0.
 177  * pool - the zswap_pool the entry's data is in
 178  * handle - zpool allocation handle that stores the compressed page data
 179  * value - value of the same-value filled pages which have same content
 180  */
 181 struct zswap_entry {
 182         struct rb_node rbnode;
 183         pgoff_t offset;
 184         int refcount;
 185         unsigned int length;
 186         struct zswap_pool *pool;
 187         union {
 188                 unsigned long handle;
 189                 unsigned long value;
 190         };
 191 };
 192
 193 struct zswap_header {
 194         swp_entry_t swpentry;
 195 };
 196
 197 /*
 198  * The tree lock in the zswap_tree struct protects a few things:
 199  * - the rbtree
 200  * - the refcount field of each entry in the tree
 201  */
 202 struct zswap_tree {
 203         struct rb_root rbroot;
 204         spinlock_t lock;
 205 };
 206
 207 static struct zswap_tree *zswap_trees[MAX_SWAPFILES];
 208
 209 /* RCU-protected iteration */
 210 static LIST_HEAD(zswap_pools);
 211 /* protects zswap_pools list modification */
 212 static DEFINE_SPINLOCK(zswap_pools_lock);
 213 /* pool counter to provide unique names to zpool */
 214 static atomic_t zswap_pools_count = ATOMIC_INIT(0);
 215
 216 /* used by param callback function */
 217 static bool zswap_init_started;
 218
 219 /* fatal error during init */
 220 static bool zswap_init_failed;
 221
 222 /* init completed, but couldn't create the initial pool */
 223 static bool zswap_has_pool;
 224
 225 /*********************************
 226 * helpers and fwd declarations
 227 **********************************/
 228
 229 #define zswap_pool_debug(msg, p)                                \
 230         pr_debug("%s pool %s/%s\n", msg, (p)->tfm_name,         \
 231                  zpool_get_type((p)->zpool))
 232
 233 static int zswap_writeback_entry(struct zpool *pool, unsigned long handle);
 234 static int zswap_pool_get(struct zswap_pool *pool);
 235 static void zswap_pool_put(struct zswap_pool *pool);
 236
 237 static const struct zpool_ops zswap_zpool_ops = {
 238         .evict = zswap_writeback_entry
 239 };
 240
 241 static bool zswap_is_full(void)
 242 {
 243         return totalram_pages() * zswap_max_pool_percent / 100 <
 244                         DIV_ROUND_UP(zswap_pool_total_size, PAGE_SIZE);
 245 }
 246
 247 static bool zswap_can_accept(void)
 248 {
 249         return totalram_pages() * zswap_accept_thr_percent / 100 *
 250                                 zswap_max_pool_percent / 100 >
 251                         DIV_ROUND_UP(zswap_pool_total_size, PAGE_SIZE);
 252 }
 253
 254 static void zswap_update_total_size(void)
 255 {
 256         struct zswap_pool *pool;
 257         u64 total = 0;
 258
 259         rcu_read_lock();
 260
 261         list_for_each_entry_rcu(pool, &zswap_pools, list)
 262                 total += zpool_get_total_size(pool->zpool);
 263
 264         rcu_read_unlock();
 265
 266         zswap_pool_total_size = total;
 267 }
 268
 269 /*********************************
 270 * zswap entry functions
 271 **********************************/
 272 static struct kmem_cache *zswap_entry_cache;
 273
 274 static int __init zswap_entry_cache_create(void)
 275 {
 276         zswap_entry_cache = KMEM_CACHE(zswap_entry, 0);
 277         return zswap_entry_cache == NULL;
 278 }
 279
 280 static void __init zswap_entry_cache_destroy(void)
 281 {
 282         kmem_cache_destroy(zswap_entry_cache);
 283 }
 284
 285 static struct zswap_entry *zswap_entry_cache_alloc(gfp_t gfp)
 286 {
 287         struct zswap_entry *entry;
 288         entry = kmem_cache_alloc(zswap_entry_cache, gfp);
 289         if (!entry)
 290                 return NULL;
 291         entry->refcount = 1;
 292         RB_CLEAR_NODE(&entry->rbnode);
 293         return entry;
 294 }
 295
 296 static void zswap_entry_cache_free(struct zswap_entry *entry)
 297 {
 298         kmem_cache_free(zswap_entry_cache, entry);
 299 }
 300
 301 /*********************************
 302 * rbtree functions
 303 **********************************/
 304 static struct zswap_entry *zswap_rb_search(struct rb_root *root, pgoff_t offset)
 305 {
 306         struct rb_node *node = root->rb_node;
 307         struct zswap_entry *entry;
 308
 309         while (node) {
 310                 entry = rb_entry(node, struct zswap_entry, rbnode);
 311                 if (entry->offset > offset)
 312                         node = node->rb_left;
 313                 else if (entry->offset < offset)
 314                         node = node->rb_right;
 315                 else
 316                         return entry;
 317         }
 318         return NULL;
 319 }
 320
 321 /*
 322  * In the case that a entry with the same offset is found, a pointer to
 323  * the existing entry is stored in dupentry and the function returns -EEXIST
 324  */
 325 static int zswap_rb_insert(struct rb_root *root, struct zswap_entry *entry,
 326                         struct zswap_entry **dupentry)
 327 {
 328         struct rb_node **link = &root->rb_node, *parent = NULL;
 329         struct zswap_entry *myentry;
 330
 331         while (*link) {
 332                 parent = *link;
 333                 myentry = rb_entry(parent, struct zswap_entry, rbnode);
 334                 if (myentry->offset > entry->offset)
 335                         link = &(*link)->rb_left;
 336                 else if (myentry->offset < entry->offset)
 337                         link = &(*link)->rb_right;
 338                 else {
 339                         *dupentry = myentry;
 340                         return -EEXIST;
 341                 }
 342         }
 343         rb_link_node(&entry->rbnode, parent, link);
 344         rb_insert_color(&entry->rbnode, root);
 345         return 0;
 346 }
 347
 348 static void zswap_rb_erase(struct rb_root *root, struct zswap_entry *entry)
 349 {
 350         if (!RB_EMPTY_NODE(&entry->rbnode)) {
 351                 rb_erase(&entry->rbnode, root);
 352                 RB_CLEAR_NODE(&entry->rbnode);
 353         }
 354 }
 355
 356 /*
 357  * Carries out the common pattern of freeing and entry's zpool allocation,
 358  * freeing the entry itself, and decrementing the number of stored pages.
 359  */
 360 static void zswap_free_entry(struct zswap_entry *entry)
 361 {
 362         if (!entry->length)
 363                 atomic_dec(&zswap_same_filled_pages);
 364         else {
 365                 zpool_free(entry->pool->zpool, entry->handle);
 366                 zswap_pool_put(entry->pool);
 367         }
 368         zswap_entry_cache_free(entry);
 369         atomic_dec(&zswap_stored_pages);
 370         zswap_update_total_size();
 371 }
 372
 373 /* caller must hold the tree lock */
 374 static void zswap_entry_get(struct zswap_entry *entry)
 375 {
 376         entry->refcount++;
 377 }
 378
 379 /* caller must hold the tree lock
 380 * remove from the tree and free it, if nobody reference the entry
 381 */
 382 static void zswap_entry_put(struct zswap_tree *tree,
 383                         struct zswap_entry *entry)
 384 {
 385         int refcount = --entry->refcount;
 386
 387         BUG_ON(refcount < 0);
 388         if (refcount == 0) {
 389                 zswap_rb_erase(&tree->rbroot, entry);
 390                 zswap_free_entry(entry);
 391         }
 392 }
 393
 394 /* caller must hold the tree lock */
 395 static struct zswap_entry *zswap_entry_find_get(struct rb_root *root,
 396                                 pgoff_t offset)
 397 {
 398         struct zswap_entry *entry;
 399
 400         entry = zswap_rb_search(root, offset);
 401         if (entry)
 402                 zswap_entry_get(entry);
 403
 404         return entry;
 405 }
 406
 407 /*********************************
 408 * per-cpu code
 409 **********************************/
 410 static DEFINE_PER_CPU(u8 *, zswap_dstmem);
 411 /*
 412  * If users dynamically change the zpool type and compressor at runtime, i.e.
 413  * zswap is running, zswap can have more than one zpool on one cpu, but they
 414  * are sharing dtsmem. So we need this mutex to be per-cpu.
 415  */
 416 static DEFINE_PER_CPU(struct mutex *, zswap_mutex);
 417
 418 static int zswap_dstmem_prepare(unsigned int cpu)
 419 {
 420         struct mutex *mutex;
 421         u8 *dst;
 422
 423         dst = kmalloc_node(PAGE_SIZE * 2, GFP_KERNEL, cpu_to_node(cpu));
 424         if (!dst)
 425                 return -ENOMEM;
 426
 427         mutex = kmalloc_node(sizeof(*mutex), GFP_KERNEL, cpu_to_node(cpu));
 428         if (!mutex) {
 429                 kfree(dst);
 430                 return -ENOMEM;
 431         }
 432
 433         mutex_init(mutex);
 434         per_cpu(zswap_dstmem, cpu) = dst;
 435         per_cpu(zswap_mutex, cpu) = mutex;
 436         return 0;
 437 }
 438
 439 static int zswap_dstmem_dead(unsigned int cpu)
 440 {
 441         struct mutex *mutex;
 442         u8 *dst;
 443
 444         mutex = per_cpu(zswap_mutex, cpu);
 445         kfree(mutex);
 446         per_cpu(zswap_mutex, cpu) = NULL;
 447
 448         dst = per_cpu(zswap_dstmem, cpu);
 449         kfree(dst);
 450         per_cpu(zswap_dstmem, cpu) = NULL;
 451
 452         return 0;
 453 }
 454
 455 static int zswap_cpu_comp_prepare(unsigned int cpu, struct hlist_node *node)
 456 {
 457         struct zswap_pool *pool = hlist_entry(node, struct zswap_pool, node);
 458         struct crypto_acomp_ctx *acomp_ctx = per_cpu_ptr(pool->acomp_ctx, cpu);
 459         struct crypto_acomp *acomp;
 460         struct acomp_req *req;
 461
 462         acomp = crypto_alloc_acomp_node(pool->tfm_name, 0, 0, cpu_to_node(cpu));
 463         if (IS_ERR(acomp)) {
 464                 pr_err("could not alloc crypto acomp %s : %ld\n",
 465                                 pool->tfm_name, PTR_ERR(acomp));
 466                 return PTR_ERR(acomp);
 467         }
 468         acomp_ctx->acomp = acomp;
 469
 470         req = acomp_request_alloc(acomp_ctx->acomp);
 471         if (!req) {
 472                 pr_err("could not alloc crypto acomp_request %s\n",
 473                        pool->tfm_name);
 474                 crypto_free_acomp(acomp_ctx->acomp);
 475                 return -ENOMEM;
 476         }
 477         acomp_ctx->req = req;
 478
 479         crypto_init_wait(&acomp_ctx->wait);
 480         /*
 481          * if the backend of acomp is async zip, crypto_req_done() will wakeup
 482          * crypto_wait_req(); if the backend of acomp is scomp, the callback
 483          * won't be called, crypto_wait_req() will return without blocking.
 484          */
 485         acomp_request_set_callback(req, CRYPTO_TFM_REQ_MAY_BACKLOG,
 486                                    crypto_req_done, &acomp_ctx->wait);
 487
 488         acomp_ctx->mutex = per_cpu(zswap_mutex, cpu);
 489         acomp_ctx->dstmem = per_cpu(zswap_dstmem, cpu);
 490
 491         return 0;
 492 }
 493
 494 static int zswap_cpu_comp_dead(unsigned int cpu, struct hlist_node *node)
 495 {
 496         struct zswap_pool *pool = hlist_entry(node, struct zswap_pool, node);
 497         struct crypto_acomp_ctx *acomp_ctx = per_cpu_ptr(pool->acomp_ctx, cpu);
 498
 499         if (!IS_ERR_OR_NULL(acomp_ctx)) {
 500                 if (!IS_ERR_OR_NULL(acomp_ctx->req))
 501                         acomp_request_free(acomp_ctx->req);
 502                 if (!IS_ERR_OR_NULL(acomp_ctx->acomp))
 503                         crypto_free_acomp(acomp_ctx->acomp);
 504         }
 505
 506         return 0;
 507 }
 508
 509 /*********************************
 510 * pool functions
 511 **********************************/
 512
 513 static struct zswap_pool *__zswap_pool_current(void)
 514 {
 515         struct zswap_pool *pool;
 516
 517         pool = list_first_or_null_rcu(&zswap_pools, typeof(*pool), list);
 518         WARN_ONCE(!pool && zswap_has_pool,
 519                   "%s: no page storage pool!\n", __func__);
 520
 521         return pool;
 522 }
 523
 524 static struct zswap_pool *zswap_pool_current(void)
 525 {
 526         assert_spin_locked(&zswap_pools_lock);
 527
 528         return __zswap_pool_current();
 529 }
 530
 531 static struct zswap_pool *zswap_pool_current_get(void)
 532 {
 533         struct zswap_pool *pool;
 534
 535         rcu_read_lock();
 536
 537         pool = __zswap_pool_current();
 538         if (!zswap_pool_get(pool))
 539                 pool = NULL;
 540
 541         rcu_read_unlock();
 542
 543         return pool;
 544 }
 545
 546 static struct zswap_pool *zswap_pool_last_get(void)
 547 {
 548         struct zswap_pool *pool, *last = NULL;
 549
 550         rcu_read_lock();
 551
 552         list_for_each_entry_rcu(pool, &zswap_pools, list)
 553                 last = pool;
 554         WARN_ONCE(!last && zswap_has_pool,
 555                   "%s: no page storage pool!\n", __func__);
 556         if (!zswap_pool_get(last))
 557                 last = NULL;
 558
 559         rcu_read_unlock();
 560
 561         return last;
 562 }
 563
 564 /* type and compressor must be null-terminated */
 565 static struct zswap_pool *zswap_pool_find_get(char *type, char *compressor)
 566 {
 567         struct zswap_pool *pool;
 568
 569         assert_spin_locked(&zswap_pools_lock);
 570
 571         list_for_each_entry_rcu(pool, &zswap_pools, list) {
 572                 if (strcmp(pool->tfm_name, compressor))
 573                         continue;
 574                 if (strcmp(zpool_get_type(pool->zpool), type))
 575                         continue;
 576                 /* if we can't get it, it's about to be destroyed */
 577                 if (!zswap_pool_get(pool))
 578                         continue;
 579                 return pool;
 580         }
 581
 582         return NULL;
 583 }
 584
 585 static void shrink_worker(struct work_struct *w)
 586 {
 587         struct zswap_pool *pool = container_of(w, typeof(*pool),
 588                                                 shrink_work);
 589
 590         if (zpool_shrink(pool->zpool, 1, NULL))
 591                 zswap_reject_reclaim_fail++;
 592         zswap_pool_put(pool);
 593 }
 594
 595 static struct zswap_pool *zswap_pool_create(char *type, char *compressor)
 596 {
 597         struct zswap_pool *pool;
 598         char name[38]; /* 'zswap' + 32 char (max) num + \0 */
 599         gfp_t gfp = __GFP_NORETRY | __GFP_NOWARN | __GFP_KSWAPD_RECLAIM;
 600         int ret;
 601
 602         if (!zswap_has_pool) {
 603                 /* if either are unset, pool initialization failed, and we
 604                  * need both params to be set correctly before trying to
 605                  * create a pool.
 606                  */
 607                 if (!strcmp(type, ZSWAP_PARAM_UNSET))
 608                         return NULL;
 609                 if (!strcmp(compressor, ZSWAP_PARAM_UNSET))
 610                         return NULL;
 611         }
 612
 613         pool = kzalloc(sizeof(*pool), GFP_KERNEL);
 614         if (!pool)
 615                 return NULL;
 616
 617         /* unique name for each pool specifically required by zsmalloc */
 618         snprintf(name, 38, "zswap%x", atomic_inc_return(&zswap_pools_count));
 619
 620         pool->zpool = zpool_create_pool(type, name, gfp, &zswap_zpool_ops);
 621         if (!pool->zpool) {
 622                 pr_err("%s zpool not available\n", type);
 623                 goto error;
 624         }
 625         pr_debug("using %s zpool\n", zpool_get_type(pool->zpool));
 626
 627         strscpy(pool->tfm_name, compressor, sizeof(pool->tfm_name));
 628
 629         pool->acomp_ctx = alloc_percpu(*pool->acomp_ctx);
 630         if (!pool->acomp_ctx) {
 631                 pr_err("percpu alloc failed\n");
 632                 goto error;
 633         }
 634
 635         ret = cpuhp_state_add_instance(CPUHP_MM_ZSWP_POOL_PREPARE,
 636                                        &pool->node);
 637         if (ret)
 638                 goto error;
 639         pr_debug("using %s compressor\n", pool->tfm_name);
 640
 641         /* being the current pool takes 1 ref; this func expects the
 642          * caller to always add the new pool as the current pool
 643          */
 644         kref_init(&pool->kref);
 645         INIT_LIST_HEAD(&pool->list);
 646         INIT_WORK(&pool->shrink_work, shrink_worker);
 647
 648         zswap_pool_debug("created", pool);
 649
 650         return pool;
 651
 652 error:
 653         if (pool->acomp_ctx)
 654                 free_percpu(pool->acomp_ctx);
 655         if (pool->zpool)
 656                 zpool_destroy_pool(pool->zpool);
 657         kfree(pool);
 658         return NULL;
 659 }
 660
 661 static __init struct zswap_pool *__zswap_pool_create_fallback(void)
 662 {
 663         bool has_comp, has_zpool;
 664
 665         has_comp = crypto_has_acomp(zswap_compressor, 0, 0);
 666         if (!has_comp && strcmp(zswap_compressor,
 667                                 CONFIG_ZSWAP_COMPRESSOR_DEFAULT)) {
 668                 pr_err("compressor %s not available, using default %s\n",
 669                        zswap_compressor, CONFIG_ZSWAP_COMPRESSOR_DEFAULT);
 670                 param_free_charp(&zswap_compressor);
 671                 zswap_compressor = CONFIG_ZSWAP_COMPRESSOR_DEFAULT;
 672                 has_comp = crypto_has_acomp(zswap_compressor, 0, 0);
 673         }
 674         if (!has_comp) {
 675                 pr_err("default compressor %s not available\n",
 676                        zswap_compressor);
 677                 param_free_charp(&zswap_compressor);
 678                 zswap_compressor = ZSWAP_PARAM_UNSET;
 679         }
 680
 681         has_zpool = zpool_has_pool(zswap_zpool_type);
 682         if (!has_zpool && strcmp(zswap_zpool_type,
 683                                  CONFIG_ZSWAP_ZPOOL_DEFAULT)) {
 684                 pr_err("zpool %s not available, using default %s\n",
 685                        zswap_zpool_type, CONFIG_ZSWAP_ZPOOL_DEFAULT);
 686                 param_free_charp(&zswap_zpool_type);
 687                 zswap_zpool_type = CONFIG_ZSWAP_ZPOOL_DEFAULT;
 688                 has_zpool = zpool_has_pool(zswap_zpool_type);
 689         }
 690         if (!has_zpool) {
 691                 pr_err("default zpool %s not available\n",
 692                        zswap_zpool_type);
 693                 param_free_charp(&zswap_zpool_type);
 694                 zswap_zpool_type = ZSWAP_PARAM_UNSET;
 695         }
 696
 697         if (!has_comp || !has_zpool)
 698                 return NULL;
 699
 700         return zswap_pool_create(zswap_zpool_type, zswap_compressor);
 701 }
 702
 703 static void zswap_pool_destroy(struct zswap_pool *pool)
 704 {
 705         zswap_pool_debug("destroying", pool);
 706
 707         cpuhp_state_remove_instance(CPUHP_MM_ZSWP_POOL_PREPARE, &pool->node);
 708         free_percpu(pool->acomp_ctx);
 709         zpool_destroy_pool(pool->zpool);
 710         kfree(pool);
 711 }
 712
 713 static int __must_check zswap_pool_get(struct zswap_pool *pool)
 714 {
 715         if (!pool)
 716                 return 0;
 717
 718         return kref_get_unless_zero(&pool->kref);
 719 }
 720
 721 static void __zswap_pool_release(struct work_struct *work)
 722 {
 723         struct zswap_pool *pool = container_of(work, typeof(*pool),
 724                                                 release_work);
 725
 726         synchronize_rcu();
 727
 728         /* nobody should have been able to get a kref... */
 729         WARN_ON(kref_get_unless_zero(&pool->kref));
 730
 731         /* pool is now off zswap_pools list and has no references. */
 732         zswap_pool_destroy(pool);
 733 }
 734
 735 static void __zswap_pool_empty(struct kref *kref)
 736 {
 737         struct zswap_pool *pool;
 738
 739         pool = container_of(kref, typeof(*pool), kref);
 740
 741         spin_lock(&zswap_pools_lock);
 742
 743         WARN_ON(pool == zswap_pool_current());
 744
 745         list_del_rcu(&pool->list);
 746
 747         INIT_WORK(&pool->release_work, __zswap_pool_release);
 748         schedule_work(&pool->release_work);
 749
 750         spin_unlock(&zswap_pools_lock);
 751 }
 752
 753 static void zswap_pool_put(struct zswap_pool *pool)
 754 {
 755         kref_put(&pool->kref, __zswap_pool_empty);
 756 }
 757
 758 /*********************************
 759 * param callbacks
 760 **********************************/
 761
 762 /* val must be a null-terminated string */
 763 static int __zswap_param_set(const char *val, const struct kernel_param *kp,
 764                              char *type, char *compressor)
 765 {
 766         struct zswap_pool *pool, *put_pool = NULL;
 767         char *s = strstrip((char *)val);
 768         int ret;
 769
 770         if (zswap_init_failed) {
 771                 pr_err("can't set param, initialization failed\n");
 772                 return -ENODEV;
 773         }
 774
 775         /* no change required */
 776         if (!strcmp(s, *(char **)kp->arg) && zswap_has_pool)
 777                 return 0;
 778
 779         /* if this is load-time (pre-init) param setting,
 780          * don't create a pool; that's done during init.
 781          */
 782         if (!zswap_init_started)
 783                 return param_set_charp(s, kp);
 784
 785         if (!type) {
 786                 if (!zpool_has_pool(s)) {
 787                         pr_err("zpool %s not available\n", s);
 788                         return -ENOENT;
 789                 }
 790                 type = s;
 791         } else if (!compressor) {
 792                 if (!crypto_has_acomp(s, 0, 0)) {
 793                         pr_err("compressor %s not available\n", s);
 794                         return -ENOENT;
 795                 }
 796                 compressor = s;
 797         } else {
 798                 WARN_ON(1);
 799                 return -EINVAL;
 800         }
 801
 802         spin_lock(&zswap_pools_lock);
 803
 804         pool = zswap_pool_find_get(type, compressor);
 805         if (pool) {
 806                 zswap_pool_debug("using existing", pool);
 807                 WARN_ON(pool == zswap_pool_current());
 808                 list_del_rcu(&pool->list);
 809         }
 810
 811         spin_unlock(&zswap_pools_lock);
 812
 813         if (!pool)
 814                 pool = zswap_pool_create(type, compressor);
 815
 816         if (pool)
 817                 ret = param_set_charp(s, kp);
 818         else
 819                 ret = -EINVAL;
 820
 821         spin_lock(&zswap_pools_lock);
 822
 823         if (!ret) {
 824                 put_pool = zswap_pool_current();
 825                 list_add_rcu(&pool->list, &zswap_pools);
 826                 zswap_has_pool = true;
 827         } else if (pool) {
 828                 /* add the possibly pre-existing pool to the end of the pools
 829                  * list; if it's new (and empty) then it'll be removed and
 830                  * destroyed by the put after we drop the lock
 831                  */
 832                 list_add_tail_rcu(&pool->list, &zswap_pools);
 833                 put_pool = pool;
 834         }
 835
 836         spin_unlock(&zswap_pools_lock);
 837
 838         if (!zswap_has_pool && !pool) {
 839                 /* if initial pool creation failed, and this pool creation also
 840                  * failed, maybe both compressor and zpool params were bad.
 841                  * Allow changing this param, so pool creation will succeed
 842                  * when the other param is changed. We already verified this
 843                  * param is ok in the zpool_has_pool() or crypto_has_acomp()
 844                  * checks above.
 845                  */
 846                 ret = param_set_charp(s, kp);
 847         }
 848
 849         /* drop the ref from either the old current pool,
 850          * or the new pool we failed to add
 851          */
 852         if (put_pool)
 853                 zswap_pool_put(put_pool);
 854
 855         return ret;
 856 }
 857
 858 static int zswap_compressor_param_set(const char *val,
 859                                       const struct kernel_param *kp)
 860 {
 861         return __zswap_param_set(val, kp, zswap_zpool_type, NULL);
 862 }
 863
 864 static int zswap_zpool_param_set(const char *val,
 865                                  const struct kernel_param *kp)
 866 {
 867         return __zswap_param_set(val, kp, NULL, zswap_compressor);
 868 }
 869
 870 static int zswap_enabled_param_set(const char *val,
 871                                    const struct kernel_param *kp)
 872 {
 873         if (zswap_init_failed) {
 874                 pr_err("can't enable, initialization failed\n");
 875                 return -ENODEV;
 876         }
 877         if (!zswap_has_pool && zswap_init_started) {
 878                 pr_err("can't enable, no pool configured\n");
 879                 return -ENODEV;
 880         }
 881
 882         return param_set_bool(val, kp);
 883 }
 884
 885 /*********************************
 886 * writeback code
 887 **********************************/
 888 /* return enum for zswap_get_swap_cache_page */
 889 enum zswap_get_swap_ret {
 890         ZSWAP_SWAPCACHE_NEW,
 891         ZSWAP_SWAPCACHE_EXIST,
 892         ZSWAP_SWAPCACHE_FAIL,
 893 };
 894
 895 /*
 896  * zswap_get_swap_cache_page
 897  *
 898  * This is an adaption of read_swap_cache_async()
 899  *
 900  * This function tries to find a page with the given swap entry
 901  * in the swapper_space address space (the swap cache).  If the page
 902  * is found, it is returned in retpage.  Otherwise, a page is allocated,
 903  * added to the swap cache, and returned in retpage.
 904  *
 905  * If success, the swap cache page is returned in retpage
 906  * Returns ZSWAP_SWAPCACHE_EXIST if page was already in the swap cache
 907  * Returns ZSWAP_SWAPCACHE_NEW if the new page needs to be populated,
 908  *     the new page is added to swapcache and locked
 909  * Returns ZSWAP_SWAPCACHE_FAIL on error
 910  */
 911 static int zswap_get_swap_cache_page(swp_entry_t entry,
 912                                 struct page **retpage)
 913 {
 914         bool page_was_allocated;
 915
 916         *retpage = __read_swap_cache_async(entry, GFP_KERNEL,
 917                         NULL, 0, &page_was_allocated);
 918         if (page_was_allocated)
 919                 return ZSWAP_SWAPCACHE_NEW;
 920         if (!*retpage)
 921                 return ZSWAP_SWAPCACHE_FAIL;
 922         return ZSWAP_SWAPCACHE_EXIST;
 923 }
 924
 925 /*
 926  * Attempts to free an entry by adding a page to the swap cache,
 927  * decompressing the entry data into the page, and issuing a
 928  * bio write to write the page back to the swap device.
 929  *
 930  * This can be thought of as a "resumed writeback" of the page
 931  * to the swap device.  We are basically resuming the same swap
 932  * writeback path that was intercepted with the frontswap_store()
 933  * in the first place.  After the page has been decompressed into
 934  * the swap cache, the compressed version stored by zswap can be
 935  * freed.
 936  */
 937 static int zswap_writeback_entry(struct zpool *pool, unsigned long handle)
 938 {
 939         struct zswap_header *zhdr;
 940         swp_entry_t swpentry;
 941         struct zswap_tree *tree;
 942         pgoff_t offset;
 943         struct zswap_entry *entry;
 944         struct page *page;
 945         struct scatterlist input, output;
 946         struct crypto_acomp_ctx *acomp_ctx;
 947
 948         u8 *src, *tmp = NULL;
 949         unsigned int dlen;
 950         int ret;
 951         struct writeback_control wbc = {
 952                 .sync_mode = WB_SYNC_NONE,
 953         };
 954
 955         if (!zpool_can_sleep_mapped(pool)) {
 956                 tmp = kmalloc(PAGE_SIZE, GFP_ATOMIC);
 957                 if (!tmp)
 958                         return -ENOMEM;
 959         }
 960
 961         /* extract swpentry from data */
 962         zhdr = zpool_map_handle(pool, handle, ZPOOL_MM_RO);
 963         swpentry = zhdr->swpentry; /* here */
 964         tree = zswap_trees[swp_type(swpentry)];
 965         offset = swp_offset(swpentry);
 966
 967         /* find and ref zswap entry */
 968         spin_lock(&tree->lock);
 969         entry = zswap_entry_find_get(&tree->rbroot, offset);
 970         if (!entry) {
 971                 /* entry was invalidated */
 972                 spin_unlock(&tree->lock);
 973                 zpool_unmap_handle(pool, handle);
 974                 kfree(tmp);
 975                 return 0;
 976         }
 977         spin_unlock(&tree->lock);
 978         BUG_ON(offset != entry->offset);
 979
 980         src = (u8 *)zhdr + sizeof(struct zswap_header);
 981         if (!zpool_can_sleep_mapped(pool)) {
 982                 memcpy(tmp, src, entry->length);
 983                 src = tmp;
 984                 zpool_unmap_handle(pool, handle);
 985         }
 986
 987         /* try to allocate swap cache page */
 988         switch (zswap_get_swap_cache_page(swpentry, &page)) {
 989         case ZSWAP_SWAPCACHE_FAIL: /* no memory or invalidate happened */
 990                 ret = -ENOMEM;
 991                 goto fail;
 992
 993         case ZSWAP_SWAPCACHE_EXIST:
 994                 /* page is already in the swap cache, ignore for now */
 995                 put_page(page);
 996                 ret = -EEXIST;
 997                 goto fail;
 998
 999         case ZSWAP_SWAPCACHE_NEW: /* page is locked */
1000                 /* decompress */
1001                 acomp_ctx = raw_cpu_ptr(entry->pool->acomp_ctx);
1002                 dlen = PAGE_SIZE;
1003
1004                 mutex_lock(acomp_ctx->mutex);
1005                 sg_init_one(&input, src, entry->length);
1006                 sg_init_table(&output, 1);
1007                 sg_set_page(&output, page, PAGE_SIZE, 0);
1008                 acomp_request_set_params(acomp_ctx->req, &input, &output, entry->length, dlen);
1009                 ret = crypto_wait_req(crypto_acomp_decompress(acomp_ctx->req), &acomp_ctx->wait);
1010                 dlen = acomp_ctx->req->dlen;
1011                 mutex_unlock(acomp_ctx->mutex);
1012
1013                 BUG_ON(ret);
1014                 BUG_ON(dlen != PAGE_SIZE);
1015
1016                 /* page is up to date */
1017                 SetPageUptodate(page);
1018         }
1019
1020         /* move it to the tail of the inactive list after end_writeback */
1021         SetPageReclaim(page);
1022
1023         /* start writeback */
1024         __swap_writepage(page, &wbc, end_swap_bio_write);
1025         put_page(page);
1026         zswap_written_back_pages++;
1027
1028         spin_lock(&tree->lock);
1029         /* drop local reference */
1030         zswap_entry_put(tree, entry);
1031
1032         /*
1033         * There are two possible situations for entry here:
1034         * (1) refcount is 1(normal case),  entry is valid and on the tree
1035         * (2) refcount is 0, entry is freed and not on the tree
1036         *     because invalidate happened during writeback
1037         *  search the tree and free the entry if find entry
1038         */
1039         if (entry == zswap_rb_search(&tree->rbroot, offset))
1040                 zswap_entry_put(tree, entry);
1041         spin_unlock(&tree->lock);
1042
1043         goto end;
1044
1045         /*
1046         * if we get here due to ZSWAP_SWAPCACHE_EXIST
1047         * a load may be happening concurrently.
1048         * it is safe and okay to not free the entry.
1049         * if we free the entry in the following put
1050         * it is also okay to return !0
1051         */
1052 fail:
1053         spin_lock(&tree->lock);
1054         zswap_entry_put(tree, entry);
1055         spin_unlock(&tree->lock);
1056
1057 end:
1058         if (zpool_can_sleep_mapped(pool))
1059                 zpool_unmap_handle(pool, handle);
1060         else
1061                 kfree(tmp);
1062
1063         return ret;
1064 }
1065
1066 static int zswap_is_page_same_filled(void *ptr, unsigned long *value)
1067 {
1068         unsigned int pos;
1069         unsigned long *page;
1070
1071         page = (unsigned long *)ptr;
1072         for (pos = 1; pos < PAGE_SIZE / sizeof(*page); pos++) {
1073                 if (page[pos] != page[0])
1074                         return 0;
1075         }
1076         *value = page[0];
1077         return 1;
1078 }
1079
1080 static void zswap_fill_page(void *ptr, unsigned long value)
1081 {
1082         unsigned long *page;
1083
1084         page = (unsigned long *)ptr;
1085         memset_l(page, value, PAGE_SIZE / sizeof(unsigned long));
1086 }
1087
1088 /*********************************
1089 * frontswap hooks
1090 **********************************/
1091 /* attempts to compress and store an single page */
1092 static int zswap_frontswap_store(unsigned type, pgoff_t offset,
1093                                 struct page *page)
1094 {
1095         struct zswap_tree *tree = zswap_trees[type];
1096         struct zswap_entry *entry, *dupentry;
1097         struct scatterlist input, output;
1098         struct crypto_acomp_ctx *acomp_ctx;
1099         int ret;
1100         unsigned int hlen, dlen = PAGE_SIZE;
1101         unsigned long handle, value;
1102         char *buf;
1103         u8 *src, *dst;
1104         struct zswap_header zhdr = { .swpentry = swp_entry(type, offset) };
1105         gfp_t gfp;
1106
1107         /* THP isn't supported */
1108         if (PageTransHuge(page)) {
1109                 ret = -EINVAL;
1110                 goto reject;
1111         }
1112
1113         if (!zswap_enabled || !tree) {
1114                 ret = -ENODEV;
1115                 goto reject;
1116         }
1117
1118         /* reclaim space if needed */
1119         if (zswap_is_full()) {
1120                 struct zswap_pool *pool;
1121
1122                 zswap_pool_limit_hit++;
1123                 zswap_pool_reached_full = true;
1124                 pool = zswap_pool_last_get();
1125                 if (pool)
1126                         queue_work(shrink_wq, &pool->shrink_work);
1127                 ret = -ENOMEM;
1128                 goto reject;
1129         }
1130
1131         if (zswap_pool_reached_full) {
1132                if (!zswap_can_accept()) {
1133                         ret = -ENOMEM;
1134                         goto reject;
1135                 } else
1136                         zswap_pool_reached_full = false;
1137         }
1138
1139         /* allocate entry */
1140         entry = zswap_entry_cache_alloc(GFP_KERNEL);
1141         if (!entry) {
1142                 zswap_reject_kmemcache_fail++;
1143                 ret = -ENOMEM;
1144                 goto reject;
1145         }
1146
1147         if (zswap_same_filled_pages_enabled) {
1148                 src = kmap_atomic(page);
1149                 if (zswap_is_page_same_filled(src, &value)) {
1150                         kunmap_atomic(src);
1151                         entry->offset = offset;
1152                         entry->length = 0;
1153                         entry->value = value;
1154                         atomic_inc(&zswap_same_filled_pages);
1155                         goto insert_entry;
1156                 }
1157                 kunmap_atomic(src);
1158         }
1159
1160         if (!zswap_non_same_filled_pages_enabled) {
1161                 ret = -EINVAL;
1162                 goto freepage;
1163         }
1164
1165         /* if entry is successfully added, it keeps the reference */
1166         entry->pool = zswap_pool_current_get();
1167         if (!entry->pool) {
1168                 ret = -EINVAL;
1169                 goto freepage;
1170         }
1171
1172         /* compress */
1173         acomp_ctx = raw_cpu_ptr(entry->pool->acomp_ctx);
1174
1175         mutex_lock(acomp_ctx->mutex);
1176
1177         dst = acomp_ctx->dstmem;
1178         sg_init_table(&input, 1);
1179         sg_set_page(&input, page, PAGE_SIZE, 0);
1180
1181         /* zswap_dstmem is of size (PAGE_SIZE * 2). Reflect same in sg_list */
1182         sg_init_one(&output, dst, PAGE_SIZE * 2);
1183         acomp_request_set_params(acomp_ctx->req, &input, &output, PAGE_SIZE, dlen);
1184         /*
1185          * it maybe looks a little bit silly that we send an asynchronous request,
1186          * then wait for its completion synchronously. This makes the process look
1187          * synchronous in fact.
1188          * Theoretically, acomp supports users send multiple acomp requests in one
1189          * acomp instance, then get those requests done simultaneously. but in this
1190          * case, frontswap actually does store and load page by page, there is no
1191          * existing method to send the second page before the first page is done
1192          * in one thread doing frontswap.
1193          * but in different threads running on different cpu, we have different
1194          * acomp instance, so multiple threads can do (de)compression in parallel.
1195          */
1196         ret = crypto_wait_req(crypto_acomp_compress(acomp_ctx->req), &acomp_ctx->wait);
1197         dlen = acomp_ctx->req->dlen;
1198
1199         if (ret) {
1200                 ret = -EINVAL;
1201                 goto put_dstmem;
1202         }
1203
1204         /* store */
1205         hlen = zpool_evictable(entry->pool->zpool) ? sizeof(zhdr) : 0;
1206         gfp = __GFP_NORETRY | __GFP_NOWARN | __GFP_KSWAPD_RECLAIM;
1207         if (zpool_malloc_support_movable(entry->pool->zpool))
1208                 gfp |= __GFP_HIGHMEM | __GFP_MOVABLE;
1209         ret = zpool_malloc(entry->pool->zpool, hlen + dlen, gfp, &handle);
1210         if (ret == -ENOSPC) {
1211                 zswap_reject_compress_poor++;
1212                 goto put_dstmem;
1213         }
1214         if (ret) {
1215                 zswap_reject_alloc_fail++;
1216                 goto put_dstmem;
1217         }
1218         buf = zpool_map_handle(entry->pool->zpool, handle, ZPOOL_MM_WO);
1219         memcpy(buf, &zhdr, hlen);
1220         memcpy(buf + hlen, dst, dlen);
1221         zpool_unmap_handle(entry->pool->zpool, handle);
1222         mutex_unlock(acomp_ctx->mutex);
1223
1224         /* populate entry */
1225         entry->offset = offset;
1226         entry->handle = handle;
1227         entry->length = dlen;
1228
1229 insert_entry:
1230         /* map */
1231         spin_lock(&tree->lock);
1232         do {
1233                 ret = zswap_rb_insert(&tree->rbroot, entry, &dupentry);
1234                 if (ret == -EEXIST) {
1235                         zswap_duplicate_entry++;
1236                         /* remove from rbtree */
1237                         zswap_rb_erase(&tree->rbroot, dupentry);
1238                         zswap_entry_put(tree, dupentry);
1239                 }
1240         } while (ret == -EEXIST);
1241         spin_unlock(&tree->lock);
1242
1243         /* update stats */
1244         atomic_inc(&zswap_stored_pages);
1245         zswap_update_total_size();
1246
1247         return 0;
1248
1249 put_dstmem:
1250         mutex_unlock(acomp_ctx->mutex);
1251         zswap_pool_put(entry->pool);
1252 freepage:
1253         zswap_entry_cache_free(entry);
1254 reject:
1255         return ret;
1256 }
1257
1258 /*
1259  * returns 0 if the page was successfully decompressed
1260  * return -1 on entry not found or error
1261 */
1262 static int zswap_frontswap_load(unsigned type, pgoff_t offset,
1263                                 struct page *page)
1264 {
1265         struct zswap_tree *tree = zswap_trees[type];
1266         struct zswap_entry *entry;
1267         struct scatterlist input, output;
1268         struct crypto_acomp_ctx *acomp_ctx;
1269         u8 *src, *dst, *tmp;
1270         unsigned int dlen;
1271         int ret;
1272
1273         /* find */
1274         spin_lock(&tree->lock);
1275         entry = zswap_entry_find_get(&tree->rbroot, offset);
1276         if (!entry) {
1277                 /* entry was written back */
1278                 spin_unlock(&tree->lock);
1279                 return -1;
1280         }
1281         spin_unlock(&tree->lock);
1282
1283         if (!entry->length) {
1284                 dst = kmap_atomic(page);
1285                 zswap_fill_page(dst, entry->value);
1286                 kunmap_atomic(dst);
1287                 ret = 0;
1288                 goto freeentry;
1289         }
1290
1291         if (!zpool_can_sleep_mapped(entry->pool->zpool)) {
1292
1293                 tmp = kmalloc(entry->length, GFP_ATOMIC);
1294                 if (!tmp) {
1295                         ret = -ENOMEM;
1296                         goto freeentry;
1297                 }
1298         }
1299
1300         /* decompress */
1301         dlen = PAGE_SIZE;
1302         src = zpool_map_handle(entry->pool->zpool, entry->handle, ZPOOL_MM_RO);
1303         if (zpool_evictable(entry->pool->zpool))
1304                 src += sizeof(struct zswap_header);
1305
1306         if (!zpool_can_sleep_mapped(entry->pool->zpool)) {
1307
1308                 memcpy(tmp, src, entry->length);
1309                 src = tmp;
1310
1311                 zpool_unmap_handle(entry->pool->zpool, entry->handle);
1312         }
1313
1314         acomp_ctx = raw_cpu_ptr(entry->pool->acomp_ctx);
1315         mutex_lock(acomp_ctx->mutex);
1316         sg_init_one(&input, src, entry->length);
1317         sg_init_table(&output, 1);
1318         sg_set_page(&output, page, PAGE_SIZE, 0);
1319         acomp_request_set_params(acomp_ctx->req, &input, &output, entry->length, dlen);
1320         ret = crypto_wait_req(crypto_acomp_decompress(acomp_ctx->req), &acomp_ctx->wait);
1321         mutex_unlock(acomp_ctx->mutex);
1322
1323         if (zpool_can_sleep_mapped(entry->pool->zpool))
1324                 zpool_unmap_handle(entry->pool->zpool, entry->handle);
1325         else
1326                 kfree(tmp);
1327
1328         BUG_ON(ret);
1329
1330 freeentry:
1331         spin_lock(&tree->lock);
1332         zswap_entry_put(tree, entry);
1333         spin_unlock(&tree->lock);
1334
1335         return ret;
1336 }
1337
1338 /* frees an entry in zswap */
1339 static void zswap_frontswap_invalidate_page(unsigned type, pgoff_t offset)
1340 {
1341         struct zswap_tree *tree = zswap_trees[type];
1342         struct zswap_entry *entry;
1343
1344         /* find */
1345         spin_lock(&tree->lock);
1346         entry = zswap_rb_search(&tree->rbroot, offset);
1347         if (!entry) {
1348                 /* entry was written back */
1349                 spin_unlock(&tree->lock);
1350                 return;
1351         }
1352
1353         /* remove from rbtree */
1354         zswap_rb_erase(&tree->rbroot, entry);
1355
1356         /* drop the initial reference from entry creation */
1357         zswap_entry_put(tree, entry);
1358
1359         spin_unlock(&tree->lock);
1360 }
1361
1362 /* frees all zswap entries for the given swap type */
1363 static void zswap_frontswap_invalidate_area(unsigned type)
1364 {
1365         struct zswap_tree *tree = zswap_trees[type];
1366         struct zswap_entry *entry, *n;
1367
1368         if (!tree)
1369                 return;
1370
1371         /* walk the tree and free everything */
1372         spin_lock(&tree->lock);
1373         rbtree_postorder_for_each_entry_safe(entry, n, &tree->rbroot, rbnode)
1374                 zswap_free_entry(entry);
1375         tree->rbroot = RB_ROOT;
1376         spin_unlock(&tree->lock);
1377         kfree(tree);
1378         zswap_trees[type] = NULL;
1379 }
1380
1381 static void zswap_frontswap_init(unsigned type)
1382 {
1383         struct zswap_tree *tree;
1384
1385         tree = kzalloc(sizeof(*tree), GFP_KERNEL);
1386         if (!tree) {
1387                 pr_err("alloc failed, zswap disabled for swap type %d\n", type);
1388                 return;
1389         }
1390
1391         tree->rbroot = RB_ROOT;
1392         spin_lock_init(&tree->lock);
1393         zswap_trees[type] = tree;
1394 }
1395
1396 static const struct frontswap_ops zswap_frontswap_ops = {
1397         .store = zswap_frontswap_store,
1398         .load = zswap_frontswap_load,
1399         .invalidate_page = zswap_frontswap_invalidate_page,
1400         .invalidate_area = zswap_frontswap_invalidate_area,
1401         .init = zswap_frontswap_init
1402 };
1403
1404 /*********************************
1405 * debugfs functions
1406 **********************************/
1407 #ifdef CONFIG_DEBUG_FS
1408 #include <linux/debugfs.h>
1409
1410 static struct dentry *zswap_debugfs_root;
1411
1412 static int __init zswap_debugfs_init(void)
1413 {
1414         if (!debugfs_initialized())
1415                 return -ENODEV;
1416
1417         zswap_debugfs_root = debugfs_create_dir("zswap", NULL);
1418
1419         debugfs_create_u64("pool_limit_hit", 0444,
1420                            zswap_debugfs_root, &zswap_pool_limit_hit);
1421         debugfs_create_u64("reject_reclaim_fail", 0444,
1422                            zswap_debugfs_root, &zswap_reject_reclaim_fail);
1423         debugfs_create_u64("reject_alloc_fail", 0444,
1424                            zswap_debugfs_root, &zswap_reject_alloc_fail);
1425         debugfs_create_u64("reject_kmemcache_fail", 0444,
1426                            zswap_debugfs_root, &zswap_reject_kmemcache_fail);
1427         debugfs_create_u64("reject_compress_poor", 0444,
1428                            zswap_debugfs_root, &zswap_reject_compress_poor);
1429         debugfs_create_u64("written_back_pages", 0444,
1430                            zswap_debugfs_root, &zswap_written_back_pages);
1431         debugfs_create_u64("duplicate_entry", 0444,
1432                            zswap_debugfs_root, &zswap_duplicate_entry);
1433         debugfs_create_u64("pool_total_size", 0444,
1434                            zswap_debugfs_root, &zswap_pool_total_size);
1435         debugfs_create_atomic_t("stored_pages", 0444,
1436                                 zswap_debugfs_root, &zswap_stored_pages);
1437         debugfs_create_atomic_t("same_filled_pages", 0444,
1438                                 zswap_debugfs_root, &zswap_same_filled_pages);
1439
1440         return 0;
1441 }
1442 #else
1443 static int __init zswap_debugfs_init(void)
1444 {
1445         return 0;
1446 }
1447 #endif
1448
1449 /*********************************
1450 * module init and exit
1451 **********************************/
1452 static int __init init_zswap(void)
1453 {
1454         struct zswap_pool *pool;
1455         int ret;
1456
1457         zswap_init_started = true;
1458
1459         if (zswap_entry_cache_create()) {
1460                 pr_err("entry cache creation failed\n");
1461                 goto cache_fail;
1462         }
1463
1464         ret = cpuhp_setup_state(CPUHP_MM_ZSWP_MEM_PREPARE, "mm/zswap:prepare",
1465                                 zswap_dstmem_prepare, zswap_dstmem_dead);
1466         if (ret) {
1467                 pr_err("dstmem alloc failed\n");
1468                 goto dstmem_fail;
1469         }
1470
1471         ret = cpuhp_setup_state_multi(CPUHP_MM_ZSWP_POOL_PREPARE,
1472                                       "mm/zswap_pool:prepare",
1473                                       zswap_cpu_comp_prepare,
1474                                       zswap_cpu_comp_dead);
1475         if (ret)
1476                 goto hp_fail;
1477
1478         pool = __zswap_pool_create_fallback();
1479         if (pool) {
1480                 pr_info("loaded using pool %s/%s\n", pool->tfm_name,
1481                         zpool_get_type(pool->zpool));
1482                 list_add(&pool->list, &zswap_pools);
1483                 zswap_has_pool = true;
1484         } else {
1485                 pr_err("pool creation failed\n");
1486                 zswap_enabled = false;
1487         }
1488
1489         shrink_wq = create_workqueue("zswap-shrink");
1490         if (!shrink_wq)
1491                 goto fallback_fail;
1492
1493         ret = frontswap_register_ops(&zswap_frontswap_ops);
1494         if (ret)
1495                 goto destroy_wq;
1496         if (zswap_debugfs_init())
1497                 pr_warn("debugfs initialization failed\n");
1498         return 0;
1499
1500 destroy_wq:
1501         destroy_workqueue(shrink_wq);
1502 fallback_fail:
1503         if (pool)
1504                 zswap_pool_destroy(pool);
1505 hp_fail:
1506         cpuhp_remove_state(CPUHP_MM_ZSWP_MEM_PREPARE);
1507 dstmem_fail:
1508         zswap_entry_cache_destroy();
1509 cache_fail:
1510         /* if built-in, we aren't unloaded on failure; don't allow use */
1511         zswap_init_failed = true;
1512         zswap_enabled = false;
1513         return -ENOMEM;
1514 }
1515 /* must be late so crypto has time to come up */
1516 late_initcall(init_zswap);
1517
1518 MODULE_LICENSE("GPL");
1519 MODULE_AUTHOR("Seth Jennings <[email protected]>");
1520 MODULE_DESCRIPTION("Compressed cache for swap pages");