Linux Audio

Check our new training course

Loading...
v5.9
   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/mempool.h>
  28#include <linux/zpool.h>
  29
  30#include <linux/mm_types.h>
  31#include <linux/page-flags.h>
  32#include <linux/swapops.h>
  33#include <linux/writeback.h>
  34#include <linux/pagemap.h>
  35#include <linux/workqueue.h>
  36
  37/*********************************
  38* statistics
  39**********************************/
  40/* Total bytes used by the compressed storage */
  41static u64 zswap_pool_total_size;
  42/* The number of compressed pages currently stored in zswap */
  43static atomic_t zswap_stored_pages = ATOMIC_INIT(0);
  44/* The number of same-value filled pages currently stored in zswap */
  45static atomic_t zswap_same_filled_pages = ATOMIC_INIT(0);
  46
  47/*
  48 * The statistics below are not protected from concurrent access for
  49 * performance reasons so they may not be a 100% accurate.  However,
  50 * they do provide useful information on roughly how many times a
  51 * certain event is occurring.
  52*/
  53
  54/* Pool limit was hit (see zswap_max_pool_percent) */
  55static u64 zswap_pool_limit_hit;
  56/* Pages written back when pool limit was reached */
  57static u64 zswap_written_back_pages;
  58/* Store failed due to a reclaim failure after pool limit was reached */
  59static u64 zswap_reject_reclaim_fail;
  60/* Compressed page was too big for the allocator to (optimally) store */
  61static u64 zswap_reject_compress_poor;
  62/* Store failed because underlying allocator could not get memory */
  63static u64 zswap_reject_alloc_fail;
  64/* Store failed because the entry metadata could not be allocated (rare) */
  65static u64 zswap_reject_kmemcache_fail;
  66/* Duplicate store was encountered (rare) */
  67static u64 zswap_duplicate_entry;
  68
  69/* Shrinker work queue */
  70static struct workqueue_struct *shrink_wq;
  71/* Pool limit was hit, we need to calm down */
  72static bool zswap_pool_reached_full;
  73
  74/*********************************
  75* tunables
  76**********************************/
  77
  78#define ZSWAP_PARAM_UNSET ""
  79
  80/* Enable/disable zswap */
  81static bool zswap_enabled = IS_ENABLED(CONFIG_ZSWAP_DEFAULT_ON);
  82static int zswap_enabled_param_set(const char *,
  83				   const struct kernel_param *);
  84static struct kernel_param_ops zswap_enabled_param_ops = {
  85	.set =		zswap_enabled_param_set,
  86	.get =		param_get_bool,
  87};
  88module_param_cb(enabled, &zswap_enabled_param_ops, &zswap_enabled, 0644);
  89
  90/* Crypto compressor to use */
  91static char *zswap_compressor = CONFIG_ZSWAP_COMPRESSOR_DEFAULT;
 
  92static int zswap_compressor_param_set(const char *,
  93				      const struct kernel_param *);
  94static struct kernel_param_ops zswap_compressor_param_ops = {
  95	.set =		zswap_compressor_param_set,
  96	.get =		param_get_charp,
  97	.free =		param_free_charp,
  98};
  99module_param_cb(compressor, &zswap_compressor_param_ops,
 100		&zswap_compressor, 0644);
 101
 102/* Compressed storage zpool to use */
 103static char *zswap_zpool_type = CONFIG_ZSWAP_ZPOOL_DEFAULT;
 
 104static int zswap_zpool_param_set(const char *, const struct kernel_param *);
 105static struct kernel_param_ops zswap_zpool_param_ops = {
 106	.set =		zswap_zpool_param_set,
 107	.get =		param_get_charp,
 108	.free =		param_free_charp,
 109};
 110module_param_cb(zpool, &zswap_zpool_param_ops, &zswap_zpool_type, 0644);
 111
 112/* The maximum percentage of memory that the compressed pool can occupy */
 113static unsigned int zswap_max_pool_percent = 20;
 114module_param_named(max_pool_percent, zswap_max_pool_percent, uint, 0644);
 115
 116/* The threshold for accepting new pages after the max_pool_percent was hit */
 117static unsigned int zswap_accept_thr_percent = 90; /* of max pool size */
 118module_param_named(accept_threshold_percent, zswap_accept_thr_percent,
 119		   uint, 0644);
 120
 121/* Enable/disable handling same-value filled pages (enabled by default) */
 122static bool zswap_same_filled_pages_enabled = true;
 123module_param_named(same_filled_pages_enabled, zswap_same_filled_pages_enabled,
 124		   bool, 0644);
 125
 126/*********************************
 127* data structures
 128**********************************/
 129
 130struct zswap_pool {
 131	struct zpool *zpool;
 132	struct crypto_comp * __percpu *tfm;
 133	struct kref kref;
 134	struct list_head list;
 135	struct work_struct release_work;
 136	struct work_struct shrink_work;
 137	struct hlist_node node;
 138	char tfm_name[CRYPTO_MAX_ALG_NAME];
 139};
 140
 141/*
 142 * struct zswap_entry
 143 *
 144 * This structure contains the metadata for tracking a single compressed
 145 * page within zswap.
 146 *
 147 * rbnode - links the entry into red-black tree for the appropriate swap type
 148 * offset - the swap offset for the entry.  Index into the red-black tree.
 149 * refcount - the number of outstanding reference to the entry. This is needed
 150 *            to protect against premature freeing of the entry by code
 151 *            concurrent calls to load, invalidate, and writeback.  The lock
 152 *            for the zswap_tree structure that contains the entry must
 153 *            be held while changing the refcount.  Since the lock must
 154 *            be held, there is no reason to also make refcount atomic.
 155 * length - the length in bytes of the compressed page data.  Needed during
 156 *          decompression. For a same value filled page length is 0.
 157 * pool - the zswap_pool the entry's data is in
 158 * handle - zpool allocation handle that stores the compressed page data
 159 * value - value of the same-value filled pages which have same content
 160 */
 161struct zswap_entry {
 162	struct rb_node rbnode;
 163	pgoff_t offset;
 164	int refcount;
 165	unsigned int length;
 166	struct zswap_pool *pool;
 167	union {
 168		unsigned long handle;
 169		unsigned long value;
 170	};
 171};
 172
 173struct zswap_header {
 174	swp_entry_t swpentry;
 175};
 176
 177/*
 178 * The tree lock in the zswap_tree struct protects a few things:
 179 * - the rbtree
 180 * - the refcount field of each entry in the tree
 181 */
 182struct zswap_tree {
 183	struct rb_root rbroot;
 184	spinlock_t lock;
 185};
 186
 187static struct zswap_tree *zswap_trees[MAX_SWAPFILES];
 188
 189/* RCU-protected iteration */
 190static LIST_HEAD(zswap_pools);
 191/* protects zswap_pools list modification */
 192static DEFINE_SPINLOCK(zswap_pools_lock);
 193/* pool counter to provide unique names to zpool */
 194static atomic_t zswap_pools_count = ATOMIC_INIT(0);
 195
 196/* used by param callback function */
 197static bool zswap_init_started;
 198
 199/* fatal error during init */
 200static bool zswap_init_failed;
 201
 202/* init completed, but couldn't create the initial pool */
 203static bool zswap_has_pool;
 204
 205/*********************************
 206* helpers and fwd declarations
 207**********************************/
 208
 209#define zswap_pool_debug(msg, p)				\
 210	pr_debug("%s pool %s/%s\n", msg, (p)->tfm_name,		\
 211		 zpool_get_type((p)->zpool))
 212
 213static int zswap_writeback_entry(struct zpool *pool, unsigned long handle);
 214static int zswap_pool_get(struct zswap_pool *pool);
 215static void zswap_pool_put(struct zswap_pool *pool);
 216
 217static const struct zpool_ops zswap_zpool_ops = {
 218	.evict = zswap_writeback_entry
 219};
 220
 221static bool zswap_is_full(void)
 222{
 223	return totalram_pages() * zswap_max_pool_percent / 100 <
 224			DIV_ROUND_UP(zswap_pool_total_size, PAGE_SIZE);
 225}
 226
 227static bool zswap_can_accept(void)
 228{
 229	return totalram_pages() * zswap_accept_thr_percent / 100 *
 230				zswap_max_pool_percent / 100 >
 231			DIV_ROUND_UP(zswap_pool_total_size, PAGE_SIZE);
 232}
 233
 234static void zswap_update_total_size(void)
 235{
 236	struct zswap_pool *pool;
 237	u64 total = 0;
 238
 239	rcu_read_lock();
 240
 241	list_for_each_entry_rcu(pool, &zswap_pools, list)
 242		total += zpool_get_total_size(pool->zpool);
 243
 244	rcu_read_unlock();
 245
 246	zswap_pool_total_size = total;
 247}
 248
 249/*********************************
 250* zswap entry functions
 251**********************************/
 252static struct kmem_cache *zswap_entry_cache;
 253
 254static int __init zswap_entry_cache_create(void)
 255{
 256	zswap_entry_cache = KMEM_CACHE(zswap_entry, 0);
 257	return zswap_entry_cache == NULL;
 258}
 259
 260static void __init zswap_entry_cache_destroy(void)
 261{
 262	kmem_cache_destroy(zswap_entry_cache);
 263}
 264
 265static struct zswap_entry *zswap_entry_cache_alloc(gfp_t gfp)
 266{
 267	struct zswap_entry *entry;
 268	entry = kmem_cache_alloc(zswap_entry_cache, gfp);
 269	if (!entry)
 270		return NULL;
 271	entry->refcount = 1;
 272	RB_CLEAR_NODE(&entry->rbnode);
 273	return entry;
 274}
 275
 276static void zswap_entry_cache_free(struct zswap_entry *entry)
 277{
 278	kmem_cache_free(zswap_entry_cache, entry);
 279}
 280
 281/*********************************
 282* rbtree functions
 283**********************************/
 284static struct zswap_entry *zswap_rb_search(struct rb_root *root, pgoff_t offset)
 285{
 286	struct rb_node *node = root->rb_node;
 287	struct zswap_entry *entry;
 288
 289	while (node) {
 290		entry = rb_entry(node, struct zswap_entry, rbnode);
 291		if (entry->offset > offset)
 292			node = node->rb_left;
 293		else if (entry->offset < offset)
 294			node = node->rb_right;
 295		else
 296			return entry;
 297	}
 298	return NULL;
 299}
 300
 301/*
 302 * In the case that a entry with the same offset is found, a pointer to
 303 * the existing entry is stored in dupentry and the function returns -EEXIST
 304 */
 305static int zswap_rb_insert(struct rb_root *root, struct zswap_entry *entry,
 306			struct zswap_entry **dupentry)
 307{
 308	struct rb_node **link = &root->rb_node, *parent = NULL;
 309	struct zswap_entry *myentry;
 310
 311	while (*link) {
 312		parent = *link;
 313		myentry = rb_entry(parent, struct zswap_entry, rbnode);
 314		if (myentry->offset > entry->offset)
 315			link = &(*link)->rb_left;
 316		else if (myentry->offset < entry->offset)
 317			link = &(*link)->rb_right;
 318		else {
 319			*dupentry = myentry;
 320			return -EEXIST;
 321		}
 322	}
 323	rb_link_node(&entry->rbnode, parent, link);
 324	rb_insert_color(&entry->rbnode, root);
 325	return 0;
 326}
 327
 328static void zswap_rb_erase(struct rb_root *root, struct zswap_entry *entry)
 329{
 330	if (!RB_EMPTY_NODE(&entry->rbnode)) {
 331		rb_erase(&entry->rbnode, root);
 332		RB_CLEAR_NODE(&entry->rbnode);
 333	}
 334}
 335
 336/*
 337 * Carries out the common pattern of freeing and entry's zpool allocation,
 338 * freeing the entry itself, and decrementing the number of stored pages.
 339 */
 340static void zswap_free_entry(struct zswap_entry *entry)
 341{
 342	if (!entry->length)
 343		atomic_dec(&zswap_same_filled_pages);
 344	else {
 345		zpool_free(entry->pool->zpool, entry->handle);
 346		zswap_pool_put(entry->pool);
 347	}
 348	zswap_entry_cache_free(entry);
 349	atomic_dec(&zswap_stored_pages);
 350	zswap_update_total_size();
 351}
 352
 353/* caller must hold the tree lock */
 354static void zswap_entry_get(struct zswap_entry *entry)
 355{
 356	entry->refcount++;
 357}
 358
 359/* caller must hold the tree lock
 360* remove from the tree and free it, if nobody reference the entry
 361*/
 362static void zswap_entry_put(struct zswap_tree *tree,
 363			struct zswap_entry *entry)
 364{
 365	int refcount = --entry->refcount;
 366
 367	BUG_ON(refcount < 0);
 368	if (refcount == 0) {
 369		zswap_rb_erase(&tree->rbroot, entry);
 370		zswap_free_entry(entry);
 371	}
 372}
 373
 374/* caller must hold the tree lock */
 375static struct zswap_entry *zswap_entry_find_get(struct rb_root *root,
 376				pgoff_t offset)
 377{
 378	struct zswap_entry *entry;
 379
 380	entry = zswap_rb_search(root, offset);
 381	if (entry)
 382		zswap_entry_get(entry);
 383
 384	return entry;
 385}
 386
 387/*********************************
 388* per-cpu code
 389**********************************/
 390static DEFINE_PER_CPU(u8 *, zswap_dstmem);
 391
 392static int zswap_dstmem_prepare(unsigned int cpu)
 393{
 394	u8 *dst;
 395
 396	dst = kmalloc_node(PAGE_SIZE * 2, GFP_KERNEL, cpu_to_node(cpu));
 397	if (!dst)
 398		return -ENOMEM;
 399
 400	per_cpu(zswap_dstmem, cpu) = dst;
 401	return 0;
 402}
 403
 404static int zswap_dstmem_dead(unsigned int cpu)
 405{
 406	u8 *dst;
 407
 408	dst = per_cpu(zswap_dstmem, cpu);
 409	kfree(dst);
 410	per_cpu(zswap_dstmem, cpu) = NULL;
 411
 412	return 0;
 413}
 414
 415static int zswap_cpu_comp_prepare(unsigned int cpu, struct hlist_node *node)
 416{
 417	struct zswap_pool *pool = hlist_entry(node, struct zswap_pool, node);
 418	struct crypto_comp *tfm;
 419
 420	if (WARN_ON(*per_cpu_ptr(pool->tfm, cpu)))
 421		return 0;
 422
 423	tfm = crypto_alloc_comp(pool->tfm_name, 0, 0);
 424	if (IS_ERR_OR_NULL(tfm)) {
 425		pr_err("could not alloc crypto comp %s : %ld\n",
 426		       pool->tfm_name, PTR_ERR(tfm));
 427		return -ENOMEM;
 428	}
 429	*per_cpu_ptr(pool->tfm, cpu) = tfm;
 430	return 0;
 431}
 432
 433static int zswap_cpu_comp_dead(unsigned int cpu, struct hlist_node *node)
 434{
 435	struct zswap_pool *pool = hlist_entry(node, struct zswap_pool, node);
 436	struct crypto_comp *tfm;
 437
 438	tfm = *per_cpu_ptr(pool->tfm, cpu);
 439	if (!IS_ERR_OR_NULL(tfm))
 440		crypto_free_comp(tfm);
 441	*per_cpu_ptr(pool->tfm, cpu) = NULL;
 442	return 0;
 443}
 444
 445/*********************************
 446* pool functions
 447**********************************/
 448
 449static struct zswap_pool *__zswap_pool_current(void)
 450{
 451	struct zswap_pool *pool;
 452
 453	pool = list_first_or_null_rcu(&zswap_pools, typeof(*pool), list);
 454	WARN_ONCE(!pool && zswap_has_pool,
 455		  "%s: no page storage pool!\n", __func__);
 456
 457	return pool;
 458}
 459
 460static struct zswap_pool *zswap_pool_current(void)
 461{
 462	assert_spin_locked(&zswap_pools_lock);
 463
 464	return __zswap_pool_current();
 465}
 466
 467static struct zswap_pool *zswap_pool_current_get(void)
 468{
 469	struct zswap_pool *pool;
 470
 471	rcu_read_lock();
 472
 473	pool = __zswap_pool_current();
 474	if (!zswap_pool_get(pool))
 475		pool = NULL;
 476
 477	rcu_read_unlock();
 478
 479	return pool;
 480}
 481
 482static struct zswap_pool *zswap_pool_last_get(void)
 483{
 484	struct zswap_pool *pool, *last = NULL;
 485
 486	rcu_read_lock();
 487
 488	list_for_each_entry_rcu(pool, &zswap_pools, list)
 489		last = pool;
 490	WARN_ONCE(!last && zswap_has_pool,
 491		  "%s: no page storage pool!\n", __func__);
 492	if (!zswap_pool_get(last))
 493		last = NULL;
 494
 495	rcu_read_unlock();
 496
 497	return last;
 498}
 499
 500/* type and compressor must be null-terminated */
 501static struct zswap_pool *zswap_pool_find_get(char *type, char *compressor)
 502{
 503	struct zswap_pool *pool;
 504
 505	assert_spin_locked(&zswap_pools_lock);
 506
 507	list_for_each_entry_rcu(pool, &zswap_pools, list) {
 508		if (strcmp(pool->tfm_name, compressor))
 509			continue;
 510		if (strcmp(zpool_get_type(pool->zpool), type))
 511			continue;
 512		/* if we can't get it, it's about to be destroyed */
 513		if (!zswap_pool_get(pool))
 514			continue;
 515		return pool;
 516	}
 517
 518	return NULL;
 519}
 520
 521static void shrink_worker(struct work_struct *w)
 522{
 523	struct zswap_pool *pool = container_of(w, typeof(*pool),
 524						shrink_work);
 525
 526	if (zpool_shrink(pool->zpool, 1, NULL))
 527		zswap_reject_reclaim_fail++;
 528	zswap_pool_put(pool);
 529}
 530
 531static struct zswap_pool *zswap_pool_create(char *type, char *compressor)
 532{
 533	struct zswap_pool *pool;
 534	char name[38]; /* 'zswap' + 32 char (max) num + \0 */
 535	gfp_t gfp = __GFP_NORETRY | __GFP_NOWARN | __GFP_KSWAPD_RECLAIM;
 536	int ret;
 537
 538	if (!zswap_has_pool) {
 539		/* if either are unset, pool initialization failed, and we
 540		 * need both params to be set correctly before trying to
 541		 * create a pool.
 542		 */
 543		if (!strcmp(type, ZSWAP_PARAM_UNSET))
 544			return NULL;
 545		if (!strcmp(compressor, ZSWAP_PARAM_UNSET))
 546			return NULL;
 547	}
 548
 549	pool = kzalloc(sizeof(*pool), GFP_KERNEL);
 550	if (!pool)
 551		return NULL;
 552
 553	/* unique name for each pool specifically required by zsmalloc */
 554	snprintf(name, 38, "zswap%x", atomic_inc_return(&zswap_pools_count));
 555
 556	pool->zpool = zpool_create_pool(type, name, gfp, &zswap_zpool_ops);
 557	if (!pool->zpool) {
 558		pr_err("%s zpool not available\n", type);
 559		goto error;
 560	}
 561	pr_debug("using %s zpool\n", zpool_get_type(pool->zpool));
 562
 563	strlcpy(pool->tfm_name, compressor, sizeof(pool->tfm_name));
 564	pool->tfm = alloc_percpu(struct crypto_comp *);
 565	if (!pool->tfm) {
 566		pr_err("percpu alloc failed\n");
 567		goto error;
 568	}
 569
 570	ret = cpuhp_state_add_instance(CPUHP_MM_ZSWP_POOL_PREPARE,
 571				       &pool->node);
 572	if (ret)
 573		goto error;
 574	pr_debug("using %s compressor\n", pool->tfm_name);
 575
 576	/* being the current pool takes 1 ref; this func expects the
 577	 * caller to always add the new pool as the current pool
 578	 */
 579	kref_init(&pool->kref);
 580	INIT_LIST_HEAD(&pool->list);
 581	INIT_WORK(&pool->shrink_work, shrink_worker);
 582
 583	zswap_pool_debug("created", pool);
 584
 585	return pool;
 586
 587error:
 588	free_percpu(pool->tfm);
 589	if (pool->zpool)
 590		zpool_destroy_pool(pool->zpool);
 591	kfree(pool);
 592	return NULL;
 593}
 594
 595static __init struct zswap_pool *__zswap_pool_create_fallback(void)
 596{
 597	bool has_comp, has_zpool;
 598
 599	has_comp = crypto_has_comp(zswap_compressor, 0, 0);
 600	if (!has_comp && strcmp(zswap_compressor,
 601				CONFIG_ZSWAP_COMPRESSOR_DEFAULT)) {
 602		pr_err("compressor %s not available, using default %s\n",
 603		       zswap_compressor, CONFIG_ZSWAP_COMPRESSOR_DEFAULT);
 604		param_free_charp(&zswap_compressor);
 605		zswap_compressor = CONFIG_ZSWAP_COMPRESSOR_DEFAULT;
 606		has_comp = crypto_has_comp(zswap_compressor, 0, 0);
 607	}
 608	if (!has_comp) {
 609		pr_err("default compressor %s not available\n",
 610		       zswap_compressor);
 611		param_free_charp(&zswap_compressor);
 612		zswap_compressor = ZSWAP_PARAM_UNSET;
 613	}
 614
 615	has_zpool = zpool_has_pool(zswap_zpool_type);
 616	if (!has_zpool && strcmp(zswap_zpool_type,
 617				 CONFIG_ZSWAP_ZPOOL_DEFAULT)) {
 618		pr_err("zpool %s not available, using default %s\n",
 619		       zswap_zpool_type, CONFIG_ZSWAP_ZPOOL_DEFAULT);
 620		param_free_charp(&zswap_zpool_type);
 621		zswap_zpool_type = CONFIG_ZSWAP_ZPOOL_DEFAULT;
 622		has_zpool = zpool_has_pool(zswap_zpool_type);
 623	}
 624	if (!has_zpool) {
 625		pr_err("default zpool %s not available\n",
 626		       zswap_zpool_type);
 627		param_free_charp(&zswap_zpool_type);
 628		zswap_zpool_type = ZSWAP_PARAM_UNSET;
 629	}
 630
 631	if (!has_comp || !has_zpool)
 632		return NULL;
 633
 634	return zswap_pool_create(zswap_zpool_type, zswap_compressor);
 635}
 636
 637static void zswap_pool_destroy(struct zswap_pool *pool)
 638{
 639	zswap_pool_debug("destroying", pool);
 640
 641	cpuhp_state_remove_instance(CPUHP_MM_ZSWP_POOL_PREPARE, &pool->node);
 642	free_percpu(pool->tfm);
 643	zpool_destroy_pool(pool->zpool);
 644	kfree(pool);
 645}
 646
 647static int __must_check zswap_pool_get(struct zswap_pool *pool)
 648{
 649	if (!pool)
 650		return 0;
 651
 652	return kref_get_unless_zero(&pool->kref);
 653}
 654
 655static void __zswap_pool_release(struct work_struct *work)
 656{
 657	struct zswap_pool *pool = container_of(work, typeof(*pool),
 658						release_work);
 659
 660	synchronize_rcu();
 661
 662	/* nobody should have been able to get a kref... */
 663	WARN_ON(kref_get_unless_zero(&pool->kref));
 664
 665	/* pool is now off zswap_pools list and has no references. */
 666	zswap_pool_destroy(pool);
 667}
 668
 669static void __zswap_pool_empty(struct kref *kref)
 670{
 671	struct zswap_pool *pool;
 672
 673	pool = container_of(kref, typeof(*pool), kref);
 674
 675	spin_lock(&zswap_pools_lock);
 676
 677	WARN_ON(pool == zswap_pool_current());
 678
 679	list_del_rcu(&pool->list);
 680
 681	INIT_WORK(&pool->release_work, __zswap_pool_release);
 682	schedule_work(&pool->release_work);
 683
 684	spin_unlock(&zswap_pools_lock);
 685}
 686
 687static void zswap_pool_put(struct zswap_pool *pool)
 688{
 689	kref_put(&pool->kref, __zswap_pool_empty);
 690}
 691
 692/*********************************
 693* param callbacks
 694**********************************/
 695
 696/* val must be a null-terminated string */
 697static int __zswap_param_set(const char *val, const struct kernel_param *kp,
 698			     char *type, char *compressor)
 699{
 700	struct zswap_pool *pool, *put_pool = NULL;
 701	char *s = strstrip((char *)val);
 702	int ret;
 703
 704	if (zswap_init_failed) {
 705		pr_err("can't set param, initialization failed\n");
 706		return -ENODEV;
 707	}
 708
 709	/* no change required */
 710	if (!strcmp(s, *(char **)kp->arg) && zswap_has_pool)
 711		return 0;
 712
 713	/* if this is load-time (pre-init) param setting,
 714	 * don't create a pool; that's done during init.
 715	 */
 716	if (!zswap_init_started)
 717		return param_set_charp(s, kp);
 718
 719	if (!type) {
 720		if (!zpool_has_pool(s)) {
 721			pr_err("zpool %s not available\n", s);
 722			return -ENOENT;
 723		}
 724		type = s;
 725	} else if (!compressor) {
 726		if (!crypto_has_comp(s, 0, 0)) {
 727			pr_err("compressor %s not available\n", s);
 728			return -ENOENT;
 729		}
 730		compressor = s;
 731	} else {
 732		WARN_ON(1);
 733		return -EINVAL;
 734	}
 735
 736	spin_lock(&zswap_pools_lock);
 737
 738	pool = zswap_pool_find_get(type, compressor);
 739	if (pool) {
 740		zswap_pool_debug("using existing", pool);
 741		WARN_ON(pool == zswap_pool_current());
 742		list_del_rcu(&pool->list);
 743	}
 744
 745	spin_unlock(&zswap_pools_lock);
 746
 747	if (!pool)
 748		pool = zswap_pool_create(type, compressor);
 749
 750	if (pool)
 751		ret = param_set_charp(s, kp);
 752	else
 753		ret = -EINVAL;
 754
 755	spin_lock(&zswap_pools_lock);
 756
 757	if (!ret) {
 758		put_pool = zswap_pool_current();
 759		list_add_rcu(&pool->list, &zswap_pools);
 760		zswap_has_pool = true;
 761	} else if (pool) {
 762		/* add the possibly pre-existing pool to the end of the pools
 763		 * list; if it's new (and empty) then it'll be removed and
 764		 * destroyed by the put after we drop the lock
 765		 */
 766		list_add_tail_rcu(&pool->list, &zswap_pools);
 767		put_pool = pool;
 768	}
 769
 770	spin_unlock(&zswap_pools_lock);
 771
 772	if (!zswap_has_pool && !pool) {
 773		/* if initial pool creation failed, and this pool creation also
 774		 * failed, maybe both compressor and zpool params were bad.
 775		 * Allow changing this param, so pool creation will succeed
 776		 * when the other param is changed. We already verified this
 777		 * param is ok in the zpool_has_pool() or crypto_has_comp()
 778		 * checks above.
 779		 */
 780		ret = param_set_charp(s, kp);
 781	}
 782
 783	/* drop the ref from either the old current pool,
 784	 * or the new pool we failed to add
 785	 */
 786	if (put_pool)
 787		zswap_pool_put(put_pool);
 788
 789	return ret;
 790}
 791
 792static int zswap_compressor_param_set(const char *val,
 793				      const struct kernel_param *kp)
 794{
 795	return __zswap_param_set(val, kp, zswap_zpool_type, NULL);
 796}
 797
 798static int zswap_zpool_param_set(const char *val,
 799				 const struct kernel_param *kp)
 800{
 801	return __zswap_param_set(val, kp, NULL, zswap_compressor);
 802}
 803
 804static int zswap_enabled_param_set(const char *val,
 805				   const struct kernel_param *kp)
 806{
 807	if (zswap_init_failed) {
 808		pr_err("can't enable, initialization failed\n");
 809		return -ENODEV;
 810	}
 811	if (!zswap_has_pool && zswap_init_started) {
 812		pr_err("can't enable, no pool configured\n");
 813		return -ENODEV;
 814	}
 815
 816	return param_set_bool(val, kp);
 817}
 818
 819/*********************************
 820* writeback code
 821**********************************/
 822/* return enum for zswap_get_swap_cache_page */
 823enum zswap_get_swap_ret {
 824	ZSWAP_SWAPCACHE_NEW,
 825	ZSWAP_SWAPCACHE_EXIST,
 826	ZSWAP_SWAPCACHE_FAIL,
 827};
 828
 829/*
 830 * zswap_get_swap_cache_page
 831 *
 832 * This is an adaption of read_swap_cache_async()
 833 *
 834 * This function tries to find a page with the given swap entry
 835 * in the swapper_space address space (the swap cache).  If the page
 836 * is found, it is returned in retpage.  Otherwise, a page is allocated,
 837 * added to the swap cache, and returned in retpage.
 838 *
 839 * If success, the swap cache page is returned in retpage
 840 * Returns ZSWAP_SWAPCACHE_EXIST if page was already in the swap cache
 841 * Returns ZSWAP_SWAPCACHE_NEW if the new page needs to be populated,
 842 *     the new page is added to swapcache and locked
 843 * Returns ZSWAP_SWAPCACHE_FAIL on error
 844 */
 845static int zswap_get_swap_cache_page(swp_entry_t entry,
 846				struct page **retpage)
 847{
 848	bool page_was_allocated;
 849
 850	*retpage = __read_swap_cache_async(entry, GFP_KERNEL,
 851			NULL, 0, &page_was_allocated);
 852	if (page_was_allocated)
 853		return ZSWAP_SWAPCACHE_NEW;
 854	if (!*retpage)
 855		return ZSWAP_SWAPCACHE_FAIL;
 856	return ZSWAP_SWAPCACHE_EXIST;
 857}
 858
 859/*
 860 * Attempts to free an entry by adding a page to the swap cache,
 861 * decompressing the entry data into the page, and issuing a
 862 * bio write to write the page back to the swap device.
 863 *
 864 * This can be thought of as a "resumed writeback" of the page
 865 * to the swap device.  We are basically resuming the same swap
 866 * writeback path that was intercepted with the frontswap_store()
 867 * in the first place.  After the page has been decompressed into
 868 * the swap cache, the compressed version stored by zswap can be
 869 * freed.
 870 */
 871static int zswap_writeback_entry(struct zpool *pool, unsigned long handle)
 872{
 873	struct zswap_header *zhdr;
 874	swp_entry_t swpentry;
 875	struct zswap_tree *tree;
 876	pgoff_t offset;
 877	struct zswap_entry *entry;
 878	struct page *page;
 879	struct crypto_comp *tfm;
 880	u8 *src, *dst;
 881	unsigned int dlen;
 882	int ret;
 883	struct writeback_control wbc = {
 884		.sync_mode = WB_SYNC_NONE,
 885	};
 886
 887	/* extract swpentry from data */
 888	zhdr = zpool_map_handle(pool, handle, ZPOOL_MM_RO);
 889	swpentry = zhdr->swpentry; /* here */
 
 890	tree = zswap_trees[swp_type(swpentry)];
 891	offset = swp_offset(swpentry);
 892
 893	/* find and ref zswap entry */
 894	spin_lock(&tree->lock);
 895	entry = zswap_entry_find_get(&tree->rbroot, offset);
 896	if (!entry) {
 897		/* entry was invalidated */
 898		spin_unlock(&tree->lock);
 899		zpool_unmap_handle(pool, handle);
 900		return 0;
 901	}
 902	spin_unlock(&tree->lock);
 903	BUG_ON(offset != entry->offset);
 904
 905	/* try to allocate swap cache page */
 906	switch (zswap_get_swap_cache_page(swpentry, &page)) {
 907	case ZSWAP_SWAPCACHE_FAIL: /* no memory or invalidate happened */
 908		ret = -ENOMEM;
 909		goto fail;
 910
 911	case ZSWAP_SWAPCACHE_EXIST:
 912		/* page is already in the swap cache, ignore for now */
 913		put_page(page);
 914		ret = -EEXIST;
 915		goto fail;
 916
 917	case ZSWAP_SWAPCACHE_NEW: /* page is locked */
 918		/* decompress */
 919		dlen = PAGE_SIZE;
 920		src = (u8 *)zhdr + sizeof(struct zswap_header);
 
 921		dst = kmap_atomic(page);
 922		tfm = *get_cpu_ptr(entry->pool->tfm);
 923		ret = crypto_comp_decompress(tfm, src, entry->length,
 924					     dst, &dlen);
 925		put_cpu_ptr(entry->pool->tfm);
 926		kunmap_atomic(dst);
 
 927		BUG_ON(ret);
 928		BUG_ON(dlen != PAGE_SIZE);
 929
 930		/* page is up to date */
 931		SetPageUptodate(page);
 932	}
 933
 934	/* move it to the tail of the inactive list after end_writeback */
 935	SetPageReclaim(page);
 936
 937	/* start writeback */
 938	__swap_writepage(page, &wbc, end_swap_bio_write);
 939	put_page(page);
 940	zswap_written_back_pages++;
 941
 942	spin_lock(&tree->lock);
 943	/* drop local reference */
 944	zswap_entry_put(tree, entry);
 945
 946	/*
 947	* There are two possible situations for entry here:
 948	* (1) refcount is 1(normal case),  entry is valid and on the tree
 949	* (2) refcount is 0, entry is freed and not on the tree
 950	*     because invalidate happened during writeback
 951	*  search the tree and free the entry if find entry
 952	*/
 953	if (entry == zswap_rb_search(&tree->rbroot, offset))
 954		zswap_entry_put(tree, entry);
 955	spin_unlock(&tree->lock);
 956
 957	goto end;
 958
 959	/*
 960	* if we get here due to ZSWAP_SWAPCACHE_EXIST
 961	* a load may happening concurrently
 962	* it is safe and okay to not free the entry
 963	* if we free the entry in the following put
 964	* it it either okay to return !0
 965	*/
 966fail:
 967	spin_lock(&tree->lock);
 968	zswap_entry_put(tree, entry);
 969	spin_unlock(&tree->lock);
 970
 971end:
 972	zpool_unmap_handle(pool, handle);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 973	return ret;
 974}
 975
 976static int zswap_is_page_same_filled(void *ptr, unsigned long *value)
 977{
 978	unsigned int pos;
 979	unsigned long *page;
 980
 981	page = (unsigned long *)ptr;
 982	for (pos = 1; pos < PAGE_SIZE / sizeof(*page); pos++) {
 983		if (page[pos] != page[0])
 984			return 0;
 985	}
 986	*value = page[0];
 987	return 1;
 988}
 989
 990static void zswap_fill_page(void *ptr, unsigned long value)
 991{
 992	unsigned long *page;
 993
 994	page = (unsigned long *)ptr;
 995	memset_l(page, value, PAGE_SIZE / sizeof(unsigned long));
 996}
 997
 998/*********************************
 999* frontswap hooks
1000**********************************/
1001/* attempts to compress and store an single page */
1002static int zswap_frontswap_store(unsigned type, pgoff_t offset,
1003				struct page *page)
1004{
1005	struct zswap_tree *tree = zswap_trees[type];
1006	struct zswap_entry *entry, *dupentry;
1007	struct crypto_comp *tfm;
1008	int ret;
1009	unsigned int hlen, dlen = PAGE_SIZE;
1010	unsigned long handle, value;
1011	char *buf;
1012	u8 *src, *dst;
1013	struct zswap_header zhdr = { .swpentry = swp_entry(type, offset) };
1014	gfp_t gfp;
1015
1016	/* THP isn't supported */
1017	if (PageTransHuge(page)) {
1018		ret = -EINVAL;
1019		goto reject;
1020	}
1021
1022	if (!zswap_enabled || !tree) {
1023		ret = -ENODEV;
1024		goto reject;
1025	}
1026
1027	/* reclaim space if needed */
1028	if (zswap_is_full()) {
1029		struct zswap_pool *pool;
1030
1031		zswap_pool_limit_hit++;
1032		zswap_pool_reached_full = true;
1033		pool = zswap_pool_last_get();
1034		if (pool)
1035			queue_work(shrink_wq, &pool->shrink_work);
1036		ret = -ENOMEM;
1037		goto reject;
1038	}
1039
1040	if (zswap_pool_reached_full) {
1041	       if (!zswap_can_accept()) {
1042			ret = -ENOMEM;
1043			goto reject;
1044		} else
1045			zswap_pool_reached_full = false;
1046	}
1047
1048	/* allocate entry */
1049	entry = zswap_entry_cache_alloc(GFP_KERNEL);
1050	if (!entry) {
1051		zswap_reject_kmemcache_fail++;
1052		ret = -ENOMEM;
1053		goto reject;
1054	}
1055
1056	if (zswap_same_filled_pages_enabled) {
1057		src = kmap_atomic(page);
1058		if (zswap_is_page_same_filled(src, &value)) {
1059			kunmap_atomic(src);
1060			entry->offset = offset;
1061			entry->length = 0;
1062			entry->value = value;
1063			atomic_inc(&zswap_same_filled_pages);
1064			goto insert_entry;
1065		}
1066		kunmap_atomic(src);
1067	}
1068
1069	/* if entry is successfully added, it keeps the reference */
1070	entry->pool = zswap_pool_current_get();
1071	if (!entry->pool) {
1072		ret = -EINVAL;
1073		goto freepage;
1074	}
1075
1076	/* compress */
1077	dst = get_cpu_var(zswap_dstmem);
1078	tfm = *get_cpu_ptr(entry->pool->tfm);
1079	src = kmap_atomic(page);
1080	ret = crypto_comp_compress(tfm, src, PAGE_SIZE, dst, &dlen);
1081	kunmap_atomic(src);
1082	put_cpu_ptr(entry->pool->tfm);
1083	if (ret) {
1084		ret = -EINVAL;
1085		goto put_dstmem;
1086	}
1087
1088	/* store */
1089	hlen = zpool_evictable(entry->pool->zpool) ? sizeof(zhdr) : 0;
1090	gfp = __GFP_NORETRY | __GFP_NOWARN | __GFP_KSWAPD_RECLAIM;
1091	if (zpool_malloc_support_movable(entry->pool->zpool))
1092		gfp |= __GFP_HIGHMEM | __GFP_MOVABLE;
1093	ret = zpool_malloc(entry->pool->zpool, hlen + dlen, gfp, &handle);
1094	if (ret == -ENOSPC) {
1095		zswap_reject_compress_poor++;
1096		goto put_dstmem;
1097	}
1098	if (ret) {
1099		zswap_reject_alloc_fail++;
1100		goto put_dstmem;
1101	}
1102	buf = zpool_map_handle(entry->pool->zpool, handle, ZPOOL_MM_RW);
1103	memcpy(buf, &zhdr, hlen);
1104	memcpy(buf + hlen, dst, dlen);
1105	zpool_unmap_handle(entry->pool->zpool, handle);
1106	put_cpu_var(zswap_dstmem);
1107
1108	/* populate entry */
1109	entry->offset = offset;
1110	entry->handle = handle;
1111	entry->length = dlen;
1112
1113insert_entry:
1114	/* map */
1115	spin_lock(&tree->lock);
1116	do {
1117		ret = zswap_rb_insert(&tree->rbroot, entry, &dupentry);
1118		if (ret == -EEXIST) {
1119			zswap_duplicate_entry++;
1120			/* remove from rbtree */
1121			zswap_rb_erase(&tree->rbroot, dupentry);
1122			zswap_entry_put(tree, dupentry);
1123		}
1124	} while (ret == -EEXIST);
1125	spin_unlock(&tree->lock);
1126
1127	/* update stats */
1128	atomic_inc(&zswap_stored_pages);
1129	zswap_update_total_size();
1130
1131	return 0;
1132
1133put_dstmem:
1134	put_cpu_var(zswap_dstmem);
1135	zswap_pool_put(entry->pool);
1136freepage:
1137	zswap_entry_cache_free(entry);
1138reject:
1139	return ret;
1140}
1141
1142/*
1143 * returns 0 if the page was successfully decompressed
1144 * return -1 on entry not found or error
1145*/
1146static int zswap_frontswap_load(unsigned type, pgoff_t offset,
1147				struct page *page)
1148{
1149	struct zswap_tree *tree = zswap_trees[type];
1150	struct zswap_entry *entry;
1151	struct crypto_comp *tfm;
1152	u8 *src, *dst;
1153	unsigned int dlen;
1154	int ret;
1155
1156	/* find */
1157	spin_lock(&tree->lock);
1158	entry = zswap_entry_find_get(&tree->rbroot, offset);
1159	if (!entry) {
1160		/* entry was written back */
1161		spin_unlock(&tree->lock);
1162		return -1;
1163	}
1164	spin_unlock(&tree->lock);
1165
1166	if (!entry->length) {
1167		dst = kmap_atomic(page);
1168		zswap_fill_page(dst, entry->value);
1169		kunmap_atomic(dst);
1170		goto freeentry;
1171	}
1172
1173	/* decompress */
1174	dlen = PAGE_SIZE;
1175	src = zpool_map_handle(entry->pool->zpool, entry->handle, ZPOOL_MM_RO);
1176	if (zpool_evictable(entry->pool->zpool))
1177		src += sizeof(struct zswap_header);
1178	dst = kmap_atomic(page);
1179	tfm = *get_cpu_ptr(entry->pool->tfm);
1180	ret = crypto_comp_decompress(tfm, src, entry->length, dst, &dlen);
1181	put_cpu_ptr(entry->pool->tfm);
1182	kunmap_atomic(dst);
1183	zpool_unmap_handle(entry->pool->zpool, entry->handle);
1184	BUG_ON(ret);
1185
1186freeentry:
1187	spin_lock(&tree->lock);
1188	zswap_entry_put(tree, entry);
1189	spin_unlock(&tree->lock);
1190
1191	return 0;
1192}
1193
1194/* frees an entry in zswap */
1195static void zswap_frontswap_invalidate_page(unsigned type, pgoff_t offset)
1196{
1197	struct zswap_tree *tree = zswap_trees[type];
1198	struct zswap_entry *entry;
1199
1200	/* find */
1201	spin_lock(&tree->lock);
1202	entry = zswap_rb_search(&tree->rbroot, offset);
1203	if (!entry) {
1204		/* entry was written back */
1205		spin_unlock(&tree->lock);
1206		return;
1207	}
1208
1209	/* remove from rbtree */
1210	zswap_rb_erase(&tree->rbroot, entry);
1211
1212	/* drop the initial reference from entry creation */
1213	zswap_entry_put(tree, entry);
1214
1215	spin_unlock(&tree->lock);
1216}
1217
1218/* frees all zswap entries for the given swap type */
1219static void zswap_frontswap_invalidate_area(unsigned type)
1220{
1221	struct zswap_tree *tree = zswap_trees[type];
1222	struct zswap_entry *entry, *n;
1223
1224	if (!tree)
1225		return;
1226
1227	/* walk the tree and free everything */
1228	spin_lock(&tree->lock);
1229	rbtree_postorder_for_each_entry_safe(entry, n, &tree->rbroot, rbnode)
1230		zswap_free_entry(entry);
1231	tree->rbroot = RB_ROOT;
1232	spin_unlock(&tree->lock);
1233	kfree(tree);
1234	zswap_trees[type] = NULL;
1235}
1236
1237static void zswap_frontswap_init(unsigned type)
1238{
1239	struct zswap_tree *tree;
1240
1241	tree = kzalloc(sizeof(*tree), GFP_KERNEL);
1242	if (!tree) {
1243		pr_err("alloc failed, zswap disabled for swap type %d\n", type);
1244		return;
1245	}
1246
1247	tree->rbroot = RB_ROOT;
1248	spin_lock_init(&tree->lock);
1249	zswap_trees[type] = tree;
1250}
1251
1252static struct frontswap_ops zswap_frontswap_ops = {
1253	.store = zswap_frontswap_store,
1254	.load = zswap_frontswap_load,
1255	.invalidate_page = zswap_frontswap_invalidate_page,
1256	.invalidate_area = zswap_frontswap_invalidate_area,
1257	.init = zswap_frontswap_init
1258};
1259
1260/*********************************
1261* debugfs functions
1262**********************************/
1263#ifdef CONFIG_DEBUG_FS
1264#include <linux/debugfs.h>
1265
1266static struct dentry *zswap_debugfs_root;
1267
1268static int __init zswap_debugfs_init(void)
1269{
1270	if (!debugfs_initialized())
1271		return -ENODEV;
1272
1273	zswap_debugfs_root = debugfs_create_dir("zswap", NULL);
 
 
1274
1275	debugfs_create_u64("pool_limit_hit", 0444,
1276			   zswap_debugfs_root, &zswap_pool_limit_hit);
1277	debugfs_create_u64("reject_reclaim_fail", 0444,
1278			   zswap_debugfs_root, &zswap_reject_reclaim_fail);
1279	debugfs_create_u64("reject_alloc_fail", 0444,
1280			   zswap_debugfs_root, &zswap_reject_alloc_fail);
1281	debugfs_create_u64("reject_kmemcache_fail", 0444,
1282			   zswap_debugfs_root, &zswap_reject_kmemcache_fail);
1283	debugfs_create_u64("reject_compress_poor", 0444,
1284			   zswap_debugfs_root, &zswap_reject_compress_poor);
1285	debugfs_create_u64("written_back_pages", 0444,
1286			   zswap_debugfs_root, &zswap_written_back_pages);
1287	debugfs_create_u64("duplicate_entry", 0444,
1288			   zswap_debugfs_root, &zswap_duplicate_entry);
1289	debugfs_create_u64("pool_total_size", 0444,
1290			   zswap_debugfs_root, &zswap_pool_total_size);
1291	debugfs_create_atomic_t("stored_pages", 0444,
1292				zswap_debugfs_root, &zswap_stored_pages);
1293	debugfs_create_atomic_t("same_filled_pages", 0444,
1294				zswap_debugfs_root, &zswap_same_filled_pages);
1295
1296	return 0;
1297}
1298
1299static void __exit zswap_debugfs_exit(void)
1300{
1301	debugfs_remove_recursive(zswap_debugfs_root);
1302}
1303#else
1304static int __init zswap_debugfs_init(void)
1305{
1306	return 0;
1307}
1308
1309static void __exit zswap_debugfs_exit(void) { }
1310#endif
1311
1312/*********************************
1313* module init and exit
1314**********************************/
1315static int __init init_zswap(void)
1316{
1317	struct zswap_pool *pool;
1318	int ret;
1319
1320	zswap_init_started = true;
1321
1322	if (zswap_entry_cache_create()) {
1323		pr_err("entry cache creation failed\n");
1324		goto cache_fail;
1325	}
1326
1327	ret = cpuhp_setup_state(CPUHP_MM_ZSWP_MEM_PREPARE, "mm/zswap:prepare",
1328				zswap_dstmem_prepare, zswap_dstmem_dead);
1329	if (ret) {
1330		pr_err("dstmem alloc failed\n");
1331		goto dstmem_fail;
1332	}
1333
1334	ret = cpuhp_setup_state_multi(CPUHP_MM_ZSWP_POOL_PREPARE,
1335				      "mm/zswap_pool:prepare",
1336				      zswap_cpu_comp_prepare,
1337				      zswap_cpu_comp_dead);
1338	if (ret)
1339		goto hp_fail;
1340
1341	pool = __zswap_pool_create_fallback();
1342	if (pool) {
1343		pr_info("loaded using pool %s/%s\n", pool->tfm_name,
1344			zpool_get_type(pool->zpool));
1345		list_add(&pool->list, &zswap_pools);
1346		zswap_has_pool = true;
1347	} else {
1348		pr_err("pool creation failed\n");
1349		zswap_enabled = false;
1350	}
1351
1352	shrink_wq = create_workqueue("zswap-shrink");
1353	if (!shrink_wq)
1354		goto fallback_fail;
1355
1356	frontswap_register_ops(&zswap_frontswap_ops);
1357	if (zswap_debugfs_init())
1358		pr_warn("debugfs initialization failed\n");
1359	return 0;
1360
1361fallback_fail:
1362	if (pool)
1363		zswap_pool_destroy(pool);
1364hp_fail:
1365	cpuhp_remove_state(CPUHP_MM_ZSWP_MEM_PREPARE);
1366dstmem_fail:
1367	zswap_entry_cache_destroy();
1368cache_fail:
1369	/* if built-in, we aren't unloaded on failure; don't allow use */
1370	zswap_init_failed = true;
1371	zswap_enabled = false;
1372	return -ENOMEM;
1373}
1374/* must be late so crypto has time to come up */
1375late_initcall(init_zswap);
1376
1377MODULE_LICENSE("GPL");
1378MODULE_AUTHOR("Seth Jennings <sjennings@variantweb.net>");
1379MODULE_DESCRIPTION("Compressed cache for swap pages");
v4.17
 
   1/*
   2 * zswap.c - zswap driver file
   3 *
   4 * zswap is a backend for frontswap that takes pages that are in the process
   5 * of being swapped out and attempts to compress and store them in a
   6 * RAM-based memory pool.  This can result in a significant I/O reduction on
   7 * the swap device and, in the case where decompressing from RAM is faster
   8 * than reading from the swap device, can also improve workload performance.
   9 *
  10 * Copyright (C) 2012  Seth Jennings <sjenning@linux.vnet.ibm.com>
  11 *
  12 * This program is free software; you can redistribute it and/or
  13 * modify it under the terms of the GNU General Public License
  14 * as published by the Free Software Foundation; either version 2
  15 * of the License, or (at your option) any later version.
  16 *
  17 * This program is distributed in the hope that it will be useful,
  18 * but WITHOUT ANY WARRANTY; without even the implied warranty of
  19 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  20 * GNU General Public License for more details.
  21*/
  22
  23#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  24
  25#include <linux/module.h>
  26#include <linux/cpu.h>
  27#include <linux/highmem.h>
  28#include <linux/slab.h>
  29#include <linux/spinlock.h>
  30#include <linux/types.h>
  31#include <linux/atomic.h>
  32#include <linux/frontswap.h>
  33#include <linux/rbtree.h>
  34#include <linux/swap.h>
  35#include <linux/crypto.h>
  36#include <linux/mempool.h>
  37#include <linux/zpool.h>
  38
  39#include <linux/mm_types.h>
  40#include <linux/page-flags.h>
  41#include <linux/swapops.h>
  42#include <linux/writeback.h>
  43#include <linux/pagemap.h>
 
  44
  45/*********************************
  46* statistics
  47**********************************/
  48/* Total bytes used by the compressed storage */
  49static u64 zswap_pool_total_size;
  50/* The number of compressed pages currently stored in zswap */
  51static atomic_t zswap_stored_pages = ATOMIC_INIT(0);
  52/* The number of same-value filled pages currently stored in zswap */
  53static atomic_t zswap_same_filled_pages = ATOMIC_INIT(0);
  54
  55/*
  56 * The statistics below are not protected from concurrent access for
  57 * performance reasons so they may not be a 100% accurate.  However,
  58 * they do provide useful information on roughly how many times a
  59 * certain event is occurring.
  60*/
  61
  62/* Pool limit was hit (see zswap_max_pool_percent) */
  63static u64 zswap_pool_limit_hit;
  64/* Pages written back when pool limit was reached */
  65static u64 zswap_written_back_pages;
  66/* Store failed due to a reclaim failure after pool limit was reached */
  67static u64 zswap_reject_reclaim_fail;
  68/* Compressed page was too big for the allocator to (optimally) store */
  69static u64 zswap_reject_compress_poor;
  70/* Store failed because underlying allocator could not get memory */
  71static u64 zswap_reject_alloc_fail;
  72/* Store failed because the entry metadata could not be allocated (rare) */
  73static u64 zswap_reject_kmemcache_fail;
  74/* Duplicate store was encountered (rare) */
  75static u64 zswap_duplicate_entry;
  76
 
 
 
 
 
  77/*********************************
  78* tunables
  79**********************************/
  80
  81#define ZSWAP_PARAM_UNSET ""
  82
  83/* Enable/disable zswap (disabled by default) */
  84static bool zswap_enabled;
  85static int zswap_enabled_param_set(const char *,
  86				   const struct kernel_param *);
  87static struct kernel_param_ops zswap_enabled_param_ops = {
  88	.set =		zswap_enabled_param_set,
  89	.get =		param_get_bool,
  90};
  91module_param_cb(enabled, &zswap_enabled_param_ops, &zswap_enabled, 0644);
  92
  93/* Crypto compressor to use */
  94#define ZSWAP_COMPRESSOR_DEFAULT "lzo"
  95static char *zswap_compressor = ZSWAP_COMPRESSOR_DEFAULT;
  96static int zswap_compressor_param_set(const char *,
  97				      const struct kernel_param *);
  98static 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};
 103module_param_cb(compressor, &zswap_compressor_param_ops,
 104		&zswap_compressor, 0644);
 105
 106/* Compressed storage zpool to use */
 107#define ZSWAP_ZPOOL_DEFAULT "zbud"
 108static char *zswap_zpool_type = ZSWAP_ZPOOL_DEFAULT;
 109static int zswap_zpool_param_set(const char *, const struct kernel_param *);
 110static struct kernel_param_ops zswap_zpool_param_ops = {
 111	.set =		zswap_zpool_param_set,
 112	.get =		param_get_charp,
 113	.free =		param_free_charp,
 114};
 115module_param_cb(zpool, &zswap_zpool_param_ops, &zswap_zpool_type, 0644);
 116
 117/* The maximum percentage of memory that the compressed pool can occupy */
 118static unsigned int zswap_max_pool_percent = 20;
 119module_param_named(max_pool_percent, zswap_max_pool_percent, uint, 0644);
 120
 
 
 
 
 
 121/* Enable/disable handling same-value filled pages (enabled by default) */
 122static bool zswap_same_filled_pages_enabled = true;
 123module_param_named(same_filled_pages_enabled, zswap_same_filled_pages_enabled,
 124		   bool, 0644);
 125
 126/*********************************
 127* data structures
 128**********************************/
 129
 130struct zswap_pool {
 131	struct zpool *zpool;
 132	struct crypto_comp * __percpu *tfm;
 133	struct kref kref;
 134	struct list_head list;
 135	struct work_struct work;
 
 136	struct hlist_node node;
 137	char tfm_name[CRYPTO_MAX_ALG_NAME];
 138};
 139
 140/*
 141 * struct zswap_entry
 142 *
 143 * This structure contains the metadata for tracking a single compressed
 144 * page within zswap.
 145 *
 146 * rbnode - links the entry into red-black tree for the appropriate swap type
 147 * offset - the swap offset for the entry.  Index into the red-black tree.
 148 * refcount - the number of outstanding reference to the entry. This is needed
 149 *            to protect against premature freeing of the entry by code
 150 *            concurrent calls to load, invalidate, and writeback.  The lock
 151 *            for the zswap_tree structure that contains the entry must
 152 *            be held while changing the refcount.  Since the lock must
 153 *            be held, there is no reason to also make refcount atomic.
 154 * length - the length in bytes of the compressed page data.  Needed during
 155 *          decompression. For a same value filled page length is 0.
 156 * pool - the zswap_pool the entry's data is in
 157 * handle - zpool allocation handle that stores the compressed page data
 158 * value - value of the same-value filled pages which have same content
 159 */
 160struct zswap_entry {
 161	struct rb_node rbnode;
 162	pgoff_t offset;
 163	int refcount;
 164	unsigned int length;
 165	struct zswap_pool *pool;
 166	union {
 167		unsigned long handle;
 168		unsigned long value;
 169	};
 170};
 171
 172struct zswap_header {
 173	swp_entry_t swpentry;
 174};
 175
 176/*
 177 * The tree lock in the zswap_tree struct protects a few things:
 178 * - the rbtree
 179 * - the refcount field of each entry in the tree
 180 */
 181struct zswap_tree {
 182	struct rb_root rbroot;
 183	spinlock_t lock;
 184};
 185
 186static struct zswap_tree *zswap_trees[MAX_SWAPFILES];
 187
 188/* RCU-protected iteration */
 189static LIST_HEAD(zswap_pools);
 190/* protects zswap_pools list modification */
 191static DEFINE_SPINLOCK(zswap_pools_lock);
 192/* pool counter to provide unique names to zpool */
 193static atomic_t zswap_pools_count = ATOMIC_INIT(0);
 194
 195/* used by param callback function */
 196static bool zswap_init_started;
 197
 198/* fatal error during init */
 199static bool zswap_init_failed;
 200
 201/* init completed, but couldn't create the initial pool */
 202static bool zswap_has_pool;
 203
 204/*********************************
 205* helpers and fwd declarations
 206**********************************/
 207
 208#define zswap_pool_debug(msg, p)				\
 209	pr_debug("%s pool %s/%s\n", msg, (p)->tfm_name,		\
 210		 zpool_get_type((p)->zpool))
 211
 212static int zswap_writeback_entry(struct zpool *pool, unsigned long handle);
 213static int zswap_pool_get(struct zswap_pool *pool);
 214static void zswap_pool_put(struct zswap_pool *pool);
 215
 216static const struct zpool_ops zswap_zpool_ops = {
 217	.evict = zswap_writeback_entry
 218};
 219
 220static bool zswap_is_full(void)
 221{
 222	return totalram_pages * zswap_max_pool_percent / 100 <
 223		DIV_ROUND_UP(zswap_pool_total_size, PAGE_SIZE);
 
 
 
 
 
 
 
 224}
 225
 226static void zswap_update_total_size(void)
 227{
 228	struct zswap_pool *pool;
 229	u64 total = 0;
 230
 231	rcu_read_lock();
 232
 233	list_for_each_entry_rcu(pool, &zswap_pools, list)
 234		total += zpool_get_total_size(pool->zpool);
 235
 236	rcu_read_unlock();
 237
 238	zswap_pool_total_size = total;
 239}
 240
 241/*********************************
 242* zswap entry functions
 243**********************************/
 244static struct kmem_cache *zswap_entry_cache;
 245
 246static int __init zswap_entry_cache_create(void)
 247{
 248	zswap_entry_cache = KMEM_CACHE(zswap_entry, 0);
 249	return zswap_entry_cache == NULL;
 250}
 251
 252static void __init zswap_entry_cache_destroy(void)
 253{
 254	kmem_cache_destroy(zswap_entry_cache);
 255}
 256
 257static struct zswap_entry *zswap_entry_cache_alloc(gfp_t gfp)
 258{
 259	struct zswap_entry *entry;
 260	entry = kmem_cache_alloc(zswap_entry_cache, gfp);
 261	if (!entry)
 262		return NULL;
 263	entry->refcount = 1;
 264	RB_CLEAR_NODE(&entry->rbnode);
 265	return entry;
 266}
 267
 268static void zswap_entry_cache_free(struct zswap_entry *entry)
 269{
 270	kmem_cache_free(zswap_entry_cache, entry);
 271}
 272
 273/*********************************
 274* rbtree functions
 275**********************************/
 276static struct zswap_entry *zswap_rb_search(struct rb_root *root, pgoff_t offset)
 277{
 278	struct rb_node *node = root->rb_node;
 279	struct zswap_entry *entry;
 280
 281	while (node) {
 282		entry = rb_entry(node, struct zswap_entry, rbnode);
 283		if (entry->offset > offset)
 284			node = node->rb_left;
 285		else if (entry->offset < offset)
 286			node = node->rb_right;
 287		else
 288			return entry;
 289	}
 290	return NULL;
 291}
 292
 293/*
 294 * In the case that a entry with the same offset is found, a pointer to
 295 * the existing entry is stored in dupentry and the function returns -EEXIST
 296 */
 297static int zswap_rb_insert(struct rb_root *root, struct zswap_entry *entry,
 298			struct zswap_entry **dupentry)
 299{
 300	struct rb_node **link = &root->rb_node, *parent = NULL;
 301	struct zswap_entry *myentry;
 302
 303	while (*link) {
 304		parent = *link;
 305		myentry = rb_entry(parent, struct zswap_entry, rbnode);
 306		if (myentry->offset > entry->offset)
 307			link = &(*link)->rb_left;
 308		else if (myentry->offset < entry->offset)
 309			link = &(*link)->rb_right;
 310		else {
 311			*dupentry = myentry;
 312			return -EEXIST;
 313		}
 314	}
 315	rb_link_node(&entry->rbnode, parent, link);
 316	rb_insert_color(&entry->rbnode, root);
 317	return 0;
 318}
 319
 320static void zswap_rb_erase(struct rb_root *root, struct zswap_entry *entry)
 321{
 322	if (!RB_EMPTY_NODE(&entry->rbnode)) {
 323		rb_erase(&entry->rbnode, root);
 324		RB_CLEAR_NODE(&entry->rbnode);
 325	}
 326}
 327
 328/*
 329 * Carries out the common pattern of freeing and entry's zpool allocation,
 330 * freeing the entry itself, and decrementing the number of stored pages.
 331 */
 332static void zswap_free_entry(struct zswap_entry *entry)
 333{
 334	if (!entry->length)
 335		atomic_dec(&zswap_same_filled_pages);
 336	else {
 337		zpool_free(entry->pool->zpool, entry->handle);
 338		zswap_pool_put(entry->pool);
 339	}
 340	zswap_entry_cache_free(entry);
 341	atomic_dec(&zswap_stored_pages);
 342	zswap_update_total_size();
 343}
 344
 345/* caller must hold the tree lock */
 346static void zswap_entry_get(struct zswap_entry *entry)
 347{
 348	entry->refcount++;
 349}
 350
 351/* caller must hold the tree lock
 352* remove from the tree and free it, if nobody reference the entry
 353*/
 354static void zswap_entry_put(struct zswap_tree *tree,
 355			struct zswap_entry *entry)
 356{
 357	int refcount = --entry->refcount;
 358
 359	BUG_ON(refcount < 0);
 360	if (refcount == 0) {
 361		zswap_rb_erase(&tree->rbroot, entry);
 362		zswap_free_entry(entry);
 363	}
 364}
 365
 366/* caller must hold the tree lock */
 367static struct zswap_entry *zswap_entry_find_get(struct rb_root *root,
 368				pgoff_t offset)
 369{
 370	struct zswap_entry *entry;
 371
 372	entry = zswap_rb_search(root, offset);
 373	if (entry)
 374		zswap_entry_get(entry);
 375
 376	return entry;
 377}
 378
 379/*********************************
 380* per-cpu code
 381**********************************/
 382static DEFINE_PER_CPU(u8 *, zswap_dstmem);
 383
 384static int zswap_dstmem_prepare(unsigned int cpu)
 385{
 386	u8 *dst;
 387
 388	dst = kmalloc_node(PAGE_SIZE * 2, GFP_KERNEL, cpu_to_node(cpu));
 389	if (!dst)
 390		return -ENOMEM;
 391
 392	per_cpu(zswap_dstmem, cpu) = dst;
 393	return 0;
 394}
 395
 396static int zswap_dstmem_dead(unsigned int cpu)
 397{
 398	u8 *dst;
 399
 400	dst = per_cpu(zswap_dstmem, cpu);
 401	kfree(dst);
 402	per_cpu(zswap_dstmem, cpu) = NULL;
 403
 404	return 0;
 405}
 406
 407static int zswap_cpu_comp_prepare(unsigned int cpu, struct hlist_node *node)
 408{
 409	struct zswap_pool *pool = hlist_entry(node, struct zswap_pool, node);
 410	struct crypto_comp *tfm;
 411
 412	if (WARN_ON(*per_cpu_ptr(pool->tfm, cpu)))
 413		return 0;
 414
 415	tfm = crypto_alloc_comp(pool->tfm_name, 0, 0);
 416	if (IS_ERR_OR_NULL(tfm)) {
 417		pr_err("could not alloc crypto comp %s : %ld\n",
 418		       pool->tfm_name, PTR_ERR(tfm));
 419		return -ENOMEM;
 420	}
 421	*per_cpu_ptr(pool->tfm, cpu) = tfm;
 422	return 0;
 423}
 424
 425static int zswap_cpu_comp_dead(unsigned int cpu, struct hlist_node *node)
 426{
 427	struct zswap_pool *pool = hlist_entry(node, struct zswap_pool, node);
 428	struct crypto_comp *tfm;
 429
 430	tfm = *per_cpu_ptr(pool->tfm, cpu);
 431	if (!IS_ERR_OR_NULL(tfm))
 432		crypto_free_comp(tfm);
 433	*per_cpu_ptr(pool->tfm, cpu) = NULL;
 434	return 0;
 435}
 436
 437/*********************************
 438* pool functions
 439**********************************/
 440
 441static struct zswap_pool *__zswap_pool_current(void)
 442{
 443	struct zswap_pool *pool;
 444
 445	pool = list_first_or_null_rcu(&zswap_pools, typeof(*pool), list);
 446	WARN_ONCE(!pool && zswap_has_pool,
 447		  "%s: no page storage pool!\n", __func__);
 448
 449	return pool;
 450}
 451
 452static struct zswap_pool *zswap_pool_current(void)
 453{
 454	assert_spin_locked(&zswap_pools_lock);
 455
 456	return __zswap_pool_current();
 457}
 458
 459static struct zswap_pool *zswap_pool_current_get(void)
 460{
 461	struct zswap_pool *pool;
 462
 463	rcu_read_lock();
 464
 465	pool = __zswap_pool_current();
 466	if (!zswap_pool_get(pool))
 467		pool = NULL;
 468
 469	rcu_read_unlock();
 470
 471	return pool;
 472}
 473
 474static struct zswap_pool *zswap_pool_last_get(void)
 475{
 476	struct zswap_pool *pool, *last = NULL;
 477
 478	rcu_read_lock();
 479
 480	list_for_each_entry_rcu(pool, &zswap_pools, list)
 481		last = pool;
 482	WARN_ONCE(!last && zswap_has_pool,
 483		  "%s: no page storage pool!\n", __func__);
 484	if (!zswap_pool_get(last))
 485		last = NULL;
 486
 487	rcu_read_unlock();
 488
 489	return last;
 490}
 491
 492/* type and compressor must be null-terminated */
 493static struct zswap_pool *zswap_pool_find_get(char *type, char *compressor)
 494{
 495	struct zswap_pool *pool;
 496
 497	assert_spin_locked(&zswap_pools_lock);
 498
 499	list_for_each_entry_rcu(pool, &zswap_pools, list) {
 500		if (strcmp(pool->tfm_name, compressor))
 501			continue;
 502		if (strcmp(zpool_get_type(pool->zpool), type))
 503			continue;
 504		/* if we can't get it, it's about to be destroyed */
 505		if (!zswap_pool_get(pool))
 506			continue;
 507		return pool;
 508	}
 509
 510	return NULL;
 511}
 512
 
 
 
 
 
 
 
 
 
 
 513static struct zswap_pool *zswap_pool_create(char *type, char *compressor)
 514{
 515	struct zswap_pool *pool;
 516	char name[38]; /* 'zswap' + 32 char (max) num + \0 */
 517	gfp_t gfp = __GFP_NORETRY | __GFP_NOWARN | __GFP_KSWAPD_RECLAIM;
 518	int ret;
 519
 520	if (!zswap_has_pool) {
 521		/* if either are unset, pool initialization failed, and we
 522		 * need both params to be set correctly before trying to
 523		 * create a pool.
 524		 */
 525		if (!strcmp(type, ZSWAP_PARAM_UNSET))
 526			return NULL;
 527		if (!strcmp(compressor, ZSWAP_PARAM_UNSET))
 528			return NULL;
 529	}
 530
 531	pool = kzalloc(sizeof(*pool), GFP_KERNEL);
 532	if (!pool)
 533		return NULL;
 534
 535	/* unique name for each pool specifically required by zsmalloc */
 536	snprintf(name, 38, "zswap%x", atomic_inc_return(&zswap_pools_count));
 537
 538	pool->zpool = zpool_create_pool(type, name, gfp, &zswap_zpool_ops);
 539	if (!pool->zpool) {
 540		pr_err("%s zpool not available\n", type);
 541		goto error;
 542	}
 543	pr_debug("using %s zpool\n", zpool_get_type(pool->zpool));
 544
 545	strlcpy(pool->tfm_name, compressor, sizeof(pool->tfm_name));
 546	pool->tfm = alloc_percpu(struct crypto_comp *);
 547	if (!pool->tfm) {
 548		pr_err("percpu alloc failed\n");
 549		goto error;
 550	}
 551
 552	ret = cpuhp_state_add_instance(CPUHP_MM_ZSWP_POOL_PREPARE,
 553				       &pool->node);
 554	if (ret)
 555		goto error;
 556	pr_debug("using %s compressor\n", pool->tfm_name);
 557
 558	/* being the current pool takes 1 ref; this func expects the
 559	 * caller to always add the new pool as the current pool
 560	 */
 561	kref_init(&pool->kref);
 562	INIT_LIST_HEAD(&pool->list);
 
 563
 564	zswap_pool_debug("created", pool);
 565
 566	return pool;
 567
 568error:
 569	free_percpu(pool->tfm);
 570	if (pool->zpool)
 571		zpool_destroy_pool(pool->zpool);
 572	kfree(pool);
 573	return NULL;
 574}
 575
 576static __init struct zswap_pool *__zswap_pool_create_fallback(void)
 577{
 578	bool has_comp, has_zpool;
 579
 580	has_comp = crypto_has_comp(zswap_compressor, 0, 0);
 581	if (!has_comp && strcmp(zswap_compressor, ZSWAP_COMPRESSOR_DEFAULT)) {
 
 582		pr_err("compressor %s not available, using default %s\n",
 583		       zswap_compressor, ZSWAP_COMPRESSOR_DEFAULT);
 584		param_free_charp(&zswap_compressor);
 585		zswap_compressor = ZSWAP_COMPRESSOR_DEFAULT;
 586		has_comp = crypto_has_comp(zswap_compressor, 0, 0);
 587	}
 588	if (!has_comp) {
 589		pr_err("default compressor %s not available\n",
 590		       zswap_compressor);
 591		param_free_charp(&zswap_compressor);
 592		zswap_compressor = ZSWAP_PARAM_UNSET;
 593	}
 594
 595	has_zpool = zpool_has_pool(zswap_zpool_type);
 596	if (!has_zpool && strcmp(zswap_zpool_type, ZSWAP_ZPOOL_DEFAULT)) {
 
 597		pr_err("zpool %s not available, using default %s\n",
 598		       zswap_zpool_type, ZSWAP_ZPOOL_DEFAULT);
 599		param_free_charp(&zswap_zpool_type);
 600		zswap_zpool_type = ZSWAP_ZPOOL_DEFAULT;
 601		has_zpool = zpool_has_pool(zswap_zpool_type);
 602	}
 603	if (!has_zpool) {
 604		pr_err("default zpool %s not available\n",
 605		       zswap_zpool_type);
 606		param_free_charp(&zswap_zpool_type);
 607		zswap_zpool_type = ZSWAP_PARAM_UNSET;
 608	}
 609
 610	if (!has_comp || !has_zpool)
 611		return NULL;
 612
 613	return zswap_pool_create(zswap_zpool_type, zswap_compressor);
 614}
 615
 616static void zswap_pool_destroy(struct zswap_pool *pool)
 617{
 618	zswap_pool_debug("destroying", pool);
 619
 620	cpuhp_state_remove_instance(CPUHP_MM_ZSWP_POOL_PREPARE, &pool->node);
 621	free_percpu(pool->tfm);
 622	zpool_destroy_pool(pool->zpool);
 623	kfree(pool);
 624}
 625
 626static int __must_check zswap_pool_get(struct zswap_pool *pool)
 627{
 628	if (!pool)
 629		return 0;
 630
 631	return kref_get_unless_zero(&pool->kref);
 632}
 633
 634static void __zswap_pool_release(struct work_struct *work)
 635{
 636	struct zswap_pool *pool = container_of(work, typeof(*pool), work);
 
 637
 638	synchronize_rcu();
 639
 640	/* nobody should have been able to get a kref... */
 641	WARN_ON(kref_get_unless_zero(&pool->kref));
 642
 643	/* pool is now off zswap_pools list and has no references. */
 644	zswap_pool_destroy(pool);
 645}
 646
 647static void __zswap_pool_empty(struct kref *kref)
 648{
 649	struct zswap_pool *pool;
 650
 651	pool = container_of(kref, typeof(*pool), kref);
 652
 653	spin_lock(&zswap_pools_lock);
 654
 655	WARN_ON(pool == zswap_pool_current());
 656
 657	list_del_rcu(&pool->list);
 658
 659	INIT_WORK(&pool->work, __zswap_pool_release);
 660	schedule_work(&pool->work);
 661
 662	spin_unlock(&zswap_pools_lock);
 663}
 664
 665static void zswap_pool_put(struct zswap_pool *pool)
 666{
 667	kref_put(&pool->kref, __zswap_pool_empty);
 668}
 669
 670/*********************************
 671* param callbacks
 672**********************************/
 673
 674/* val must be a null-terminated string */
 675static int __zswap_param_set(const char *val, const struct kernel_param *kp,
 676			     char *type, char *compressor)
 677{
 678	struct zswap_pool *pool, *put_pool = NULL;
 679	char *s = strstrip((char *)val);
 680	int ret;
 681
 682	if (zswap_init_failed) {
 683		pr_err("can't set param, initialization failed\n");
 684		return -ENODEV;
 685	}
 686
 687	/* no change required */
 688	if (!strcmp(s, *(char **)kp->arg) && zswap_has_pool)
 689		return 0;
 690
 691	/* if this is load-time (pre-init) param setting,
 692	 * don't create a pool; that's done during init.
 693	 */
 694	if (!zswap_init_started)
 695		return param_set_charp(s, kp);
 696
 697	if (!type) {
 698		if (!zpool_has_pool(s)) {
 699			pr_err("zpool %s not available\n", s);
 700			return -ENOENT;
 701		}
 702		type = s;
 703	} else if (!compressor) {
 704		if (!crypto_has_comp(s, 0, 0)) {
 705			pr_err("compressor %s not available\n", s);
 706			return -ENOENT;
 707		}
 708		compressor = s;
 709	} else {
 710		WARN_ON(1);
 711		return -EINVAL;
 712	}
 713
 714	spin_lock(&zswap_pools_lock);
 715
 716	pool = zswap_pool_find_get(type, compressor);
 717	if (pool) {
 718		zswap_pool_debug("using existing", pool);
 719		WARN_ON(pool == zswap_pool_current());
 720		list_del_rcu(&pool->list);
 721	}
 722
 723	spin_unlock(&zswap_pools_lock);
 724
 725	if (!pool)
 726		pool = zswap_pool_create(type, compressor);
 727
 728	if (pool)
 729		ret = param_set_charp(s, kp);
 730	else
 731		ret = -EINVAL;
 732
 733	spin_lock(&zswap_pools_lock);
 734
 735	if (!ret) {
 736		put_pool = zswap_pool_current();
 737		list_add_rcu(&pool->list, &zswap_pools);
 738		zswap_has_pool = true;
 739	} else if (pool) {
 740		/* add the possibly pre-existing pool to the end of the pools
 741		 * list; if it's new (and empty) then it'll be removed and
 742		 * destroyed by the put after we drop the lock
 743		 */
 744		list_add_tail_rcu(&pool->list, &zswap_pools);
 745		put_pool = pool;
 746	}
 747
 748	spin_unlock(&zswap_pools_lock);
 749
 750	if (!zswap_has_pool && !pool) {
 751		/* if initial pool creation failed, and this pool creation also
 752		 * failed, maybe both compressor and zpool params were bad.
 753		 * Allow changing this param, so pool creation will succeed
 754		 * when the other param is changed. We already verified this
 755		 * param is ok in the zpool_has_pool() or crypto_has_comp()
 756		 * checks above.
 757		 */
 758		ret = param_set_charp(s, kp);
 759	}
 760
 761	/* drop the ref from either the old current pool,
 762	 * or the new pool we failed to add
 763	 */
 764	if (put_pool)
 765		zswap_pool_put(put_pool);
 766
 767	return ret;
 768}
 769
 770static int zswap_compressor_param_set(const char *val,
 771				      const struct kernel_param *kp)
 772{
 773	return __zswap_param_set(val, kp, zswap_zpool_type, NULL);
 774}
 775
 776static int zswap_zpool_param_set(const char *val,
 777				 const struct kernel_param *kp)
 778{
 779	return __zswap_param_set(val, kp, NULL, zswap_compressor);
 780}
 781
 782static int zswap_enabled_param_set(const char *val,
 783				   const struct kernel_param *kp)
 784{
 785	if (zswap_init_failed) {
 786		pr_err("can't enable, initialization failed\n");
 787		return -ENODEV;
 788	}
 789	if (!zswap_has_pool && zswap_init_started) {
 790		pr_err("can't enable, no pool configured\n");
 791		return -ENODEV;
 792	}
 793
 794	return param_set_bool(val, kp);
 795}
 796
 797/*********************************
 798* writeback code
 799**********************************/
 800/* return enum for zswap_get_swap_cache_page */
 801enum zswap_get_swap_ret {
 802	ZSWAP_SWAPCACHE_NEW,
 803	ZSWAP_SWAPCACHE_EXIST,
 804	ZSWAP_SWAPCACHE_FAIL,
 805};
 806
 807/*
 808 * zswap_get_swap_cache_page
 809 *
 810 * This is an adaption of read_swap_cache_async()
 811 *
 812 * This function tries to find a page with the given swap entry
 813 * in the swapper_space address space (the swap cache).  If the page
 814 * is found, it is returned in retpage.  Otherwise, a page is allocated,
 815 * added to the swap cache, and returned in retpage.
 816 *
 817 * If success, the swap cache page is returned in retpage
 818 * Returns ZSWAP_SWAPCACHE_EXIST if page was already in the swap cache
 819 * Returns ZSWAP_SWAPCACHE_NEW if the new page needs to be populated,
 820 *     the new page is added to swapcache and locked
 821 * Returns ZSWAP_SWAPCACHE_FAIL on error
 822 */
 823static int zswap_get_swap_cache_page(swp_entry_t entry,
 824				struct page **retpage)
 825{
 826	bool page_was_allocated;
 827
 828	*retpage = __read_swap_cache_async(entry, GFP_KERNEL,
 829			NULL, 0, &page_was_allocated);
 830	if (page_was_allocated)
 831		return ZSWAP_SWAPCACHE_NEW;
 832	if (!*retpage)
 833		return ZSWAP_SWAPCACHE_FAIL;
 834	return ZSWAP_SWAPCACHE_EXIST;
 835}
 836
 837/*
 838 * Attempts to free an entry by adding a page to the swap cache,
 839 * decompressing the entry data into the page, and issuing a
 840 * bio write to write the page back to the swap device.
 841 *
 842 * This can be thought of as a "resumed writeback" of the page
 843 * to the swap device.  We are basically resuming the same swap
 844 * writeback path that was intercepted with the frontswap_store()
 845 * in the first place.  After the page has been decompressed into
 846 * the swap cache, the compressed version stored by zswap can be
 847 * freed.
 848 */
 849static int zswap_writeback_entry(struct zpool *pool, unsigned long handle)
 850{
 851	struct zswap_header *zhdr;
 852	swp_entry_t swpentry;
 853	struct zswap_tree *tree;
 854	pgoff_t offset;
 855	struct zswap_entry *entry;
 856	struct page *page;
 857	struct crypto_comp *tfm;
 858	u8 *src, *dst;
 859	unsigned int dlen;
 860	int ret;
 861	struct writeback_control wbc = {
 862		.sync_mode = WB_SYNC_NONE,
 863	};
 864
 865	/* extract swpentry from data */
 866	zhdr = zpool_map_handle(pool, handle, ZPOOL_MM_RO);
 867	swpentry = zhdr->swpentry; /* here */
 868	zpool_unmap_handle(pool, handle);
 869	tree = zswap_trees[swp_type(swpentry)];
 870	offset = swp_offset(swpentry);
 871
 872	/* find and ref zswap entry */
 873	spin_lock(&tree->lock);
 874	entry = zswap_entry_find_get(&tree->rbroot, offset);
 875	if (!entry) {
 876		/* entry was invalidated */
 877		spin_unlock(&tree->lock);
 
 878		return 0;
 879	}
 880	spin_unlock(&tree->lock);
 881	BUG_ON(offset != entry->offset);
 882
 883	/* try to allocate swap cache page */
 884	switch (zswap_get_swap_cache_page(swpentry, &page)) {
 885	case ZSWAP_SWAPCACHE_FAIL: /* no memory or invalidate happened */
 886		ret = -ENOMEM;
 887		goto fail;
 888
 889	case ZSWAP_SWAPCACHE_EXIST:
 890		/* page is already in the swap cache, ignore for now */
 891		put_page(page);
 892		ret = -EEXIST;
 893		goto fail;
 894
 895	case ZSWAP_SWAPCACHE_NEW: /* page is locked */
 896		/* decompress */
 897		dlen = PAGE_SIZE;
 898		src = (u8 *)zpool_map_handle(entry->pool->zpool, entry->handle,
 899				ZPOOL_MM_RO) + sizeof(struct zswap_header);
 900		dst = kmap_atomic(page);
 901		tfm = *get_cpu_ptr(entry->pool->tfm);
 902		ret = crypto_comp_decompress(tfm, src, entry->length,
 903					     dst, &dlen);
 904		put_cpu_ptr(entry->pool->tfm);
 905		kunmap_atomic(dst);
 906		zpool_unmap_handle(entry->pool->zpool, entry->handle);
 907		BUG_ON(ret);
 908		BUG_ON(dlen != PAGE_SIZE);
 909
 910		/* page is up to date */
 911		SetPageUptodate(page);
 912	}
 913
 914	/* move it to the tail of the inactive list after end_writeback */
 915	SetPageReclaim(page);
 916
 917	/* start writeback */
 918	__swap_writepage(page, &wbc, end_swap_bio_write);
 919	put_page(page);
 920	zswap_written_back_pages++;
 921
 922	spin_lock(&tree->lock);
 923	/* drop local reference */
 924	zswap_entry_put(tree, entry);
 925
 926	/*
 927	* There are two possible situations for entry here:
 928	* (1) refcount is 1(normal case),  entry is valid and on the tree
 929	* (2) refcount is 0, entry is freed and not on the tree
 930	*     because invalidate happened during writeback
 931	*  search the tree and free the entry if find entry
 932	*/
 933	if (entry == zswap_rb_search(&tree->rbroot, offset))
 934		zswap_entry_put(tree, entry);
 935	spin_unlock(&tree->lock);
 936
 937	goto end;
 938
 939	/*
 940	* if we get here due to ZSWAP_SWAPCACHE_EXIST
 941	* a load may happening concurrently
 942	* it is safe and okay to not free the entry
 943	* if we free the entry in the following put
 944	* it it either okay to return !0
 945	*/
 946fail:
 947	spin_lock(&tree->lock);
 948	zswap_entry_put(tree, entry);
 949	spin_unlock(&tree->lock);
 950
 951end:
 952	return ret;
 953}
 954
 955static int zswap_shrink(void)
 956{
 957	struct zswap_pool *pool;
 958	int ret;
 959
 960	pool = zswap_pool_last_get();
 961	if (!pool)
 962		return -ENOENT;
 963
 964	ret = zpool_shrink(pool->zpool, 1, NULL);
 965
 966	zswap_pool_put(pool);
 967
 968	return ret;
 969}
 970
 971static int zswap_is_page_same_filled(void *ptr, unsigned long *value)
 972{
 973	unsigned int pos;
 974	unsigned long *page;
 975
 976	page = (unsigned long *)ptr;
 977	for (pos = 1; pos < PAGE_SIZE / sizeof(*page); pos++) {
 978		if (page[pos] != page[0])
 979			return 0;
 980	}
 981	*value = page[0];
 982	return 1;
 983}
 984
 985static void zswap_fill_page(void *ptr, unsigned long value)
 986{
 987	unsigned long *page;
 988
 989	page = (unsigned long *)ptr;
 990	memset_l(page, value, PAGE_SIZE / sizeof(unsigned long));
 991}
 992
 993/*********************************
 994* frontswap hooks
 995**********************************/
 996/* attempts to compress and store an single page */
 997static int zswap_frontswap_store(unsigned type, pgoff_t offset,
 998				struct page *page)
 999{
1000	struct zswap_tree *tree = zswap_trees[type];
1001	struct zswap_entry *entry, *dupentry;
1002	struct crypto_comp *tfm;
1003	int ret;
1004	unsigned int hlen, dlen = PAGE_SIZE;
1005	unsigned long handle, value;
1006	char *buf;
1007	u8 *src, *dst;
1008	struct zswap_header zhdr = { .swpentry = swp_entry(type, offset) };
 
1009
1010	/* THP isn't supported */
1011	if (PageTransHuge(page)) {
1012		ret = -EINVAL;
1013		goto reject;
1014	}
1015
1016	if (!zswap_enabled || !tree) {
1017		ret = -ENODEV;
1018		goto reject;
1019	}
1020
1021	/* reclaim space if needed */
1022	if (zswap_is_full()) {
 
 
1023		zswap_pool_limit_hit++;
1024		if (zswap_shrink()) {
1025			zswap_reject_reclaim_fail++;
 
 
 
 
 
 
 
 
1026			ret = -ENOMEM;
1027			goto reject;
1028		}
 
1029	}
1030
1031	/* allocate entry */
1032	entry = zswap_entry_cache_alloc(GFP_KERNEL);
1033	if (!entry) {
1034		zswap_reject_kmemcache_fail++;
1035		ret = -ENOMEM;
1036		goto reject;
1037	}
1038
1039	if (zswap_same_filled_pages_enabled) {
1040		src = kmap_atomic(page);
1041		if (zswap_is_page_same_filled(src, &value)) {
1042			kunmap_atomic(src);
1043			entry->offset = offset;
1044			entry->length = 0;
1045			entry->value = value;
1046			atomic_inc(&zswap_same_filled_pages);
1047			goto insert_entry;
1048		}
1049		kunmap_atomic(src);
1050	}
1051
1052	/* if entry is successfully added, it keeps the reference */
1053	entry->pool = zswap_pool_current_get();
1054	if (!entry->pool) {
1055		ret = -EINVAL;
1056		goto freepage;
1057	}
1058
1059	/* compress */
1060	dst = get_cpu_var(zswap_dstmem);
1061	tfm = *get_cpu_ptr(entry->pool->tfm);
1062	src = kmap_atomic(page);
1063	ret = crypto_comp_compress(tfm, src, PAGE_SIZE, dst, &dlen);
1064	kunmap_atomic(src);
1065	put_cpu_ptr(entry->pool->tfm);
1066	if (ret) {
1067		ret = -EINVAL;
1068		goto put_dstmem;
1069	}
1070
1071	/* store */
1072	hlen = zpool_evictable(entry->pool->zpool) ? sizeof(zhdr) : 0;
1073	ret = zpool_malloc(entry->pool->zpool, hlen + dlen,
1074			   __GFP_NORETRY | __GFP_NOWARN | __GFP_KSWAPD_RECLAIM,
1075			   &handle);
 
1076	if (ret == -ENOSPC) {
1077		zswap_reject_compress_poor++;
1078		goto put_dstmem;
1079	}
1080	if (ret) {
1081		zswap_reject_alloc_fail++;
1082		goto put_dstmem;
1083	}
1084	buf = zpool_map_handle(entry->pool->zpool, handle, ZPOOL_MM_RW);
1085	memcpy(buf, &zhdr, hlen);
1086	memcpy(buf + hlen, dst, dlen);
1087	zpool_unmap_handle(entry->pool->zpool, handle);
1088	put_cpu_var(zswap_dstmem);
1089
1090	/* populate entry */
1091	entry->offset = offset;
1092	entry->handle = handle;
1093	entry->length = dlen;
1094
1095insert_entry:
1096	/* map */
1097	spin_lock(&tree->lock);
1098	do {
1099		ret = zswap_rb_insert(&tree->rbroot, entry, &dupentry);
1100		if (ret == -EEXIST) {
1101			zswap_duplicate_entry++;
1102			/* remove from rbtree */
1103			zswap_rb_erase(&tree->rbroot, dupentry);
1104			zswap_entry_put(tree, dupentry);
1105		}
1106	} while (ret == -EEXIST);
1107	spin_unlock(&tree->lock);
1108
1109	/* update stats */
1110	atomic_inc(&zswap_stored_pages);
1111	zswap_update_total_size();
1112
1113	return 0;
1114
1115put_dstmem:
1116	put_cpu_var(zswap_dstmem);
1117	zswap_pool_put(entry->pool);
1118freepage:
1119	zswap_entry_cache_free(entry);
1120reject:
1121	return ret;
1122}
1123
1124/*
1125 * returns 0 if the page was successfully decompressed
1126 * return -1 on entry not found or error
1127*/
1128static int zswap_frontswap_load(unsigned type, pgoff_t offset,
1129				struct page *page)
1130{
1131	struct zswap_tree *tree = zswap_trees[type];
1132	struct zswap_entry *entry;
1133	struct crypto_comp *tfm;
1134	u8 *src, *dst;
1135	unsigned int dlen;
1136	int ret;
1137
1138	/* find */
1139	spin_lock(&tree->lock);
1140	entry = zswap_entry_find_get(&tree->rbroot, offset);
1141	if (!entry) {
1142		/* entry was written back */
1143		spin_unlock(&tree->lock);
1144		return -1;
1145	}
1146	spin_unlock(&tree->lock);
1147
1148	if (!entry->length) {
1149		dst = kmap_atomic(page);
1150		zswap_fill_page(dst, entry->value);
1151		kunmap_atomic(dst);
1152		goto freeentry;
1153	}
1154
1155	/* decompress */
1156	dlen = PAGE_SIZE;
1157	src = zpool_map_handle(entry->pool->zpool, entry->handle, ZPOOL_MM_RO);
1158	if (zpool_evictable(entry->pool->zpool))
1159		src += sizeof(struct zswap_header);
1160	dst = kmap_atomic(page);
1161	tfm = *get_cpu_ptr(entry->pool->tfm);
1162	ret = crypto_comp_decompress(tfm, src, entry->length, dst, &dlen);
1163	put_cpu_ptr(entry->pool->tfm);
1164	kunmap_atomic(dst);
1165	zpool_unmap_handle(entry->pool->zpool, entry->handle);
1166	BUG_ON(ret);
1167
1168freeentry:
1169	spin_lock(&tree->lock);
1170	zswap_entry_put(tree, entry);
1171	spin_unlock(&tree->lock);
1172
1173	return 0;
1174}
1175
1176/* frees an entry in zswap */
1177static void zswap_frontswap_invalidate_page(unsigned type, pgoff_t offset)
1178{
1179	struct zswap_tree *tree = zswap_trees[type];
1180	struct zswap_entry *entry;
1181
1182	/* find */
1183	spin_lock(&tree->lock);
1184	entry = zswap_rb_search(&tree->rbroot, offset);
1185	if (!entry) {
1186		/* entry was written back */
1187		spin_unlock(&tree->lock);
1188		return;
1189	}
1190
1191	/* remove from rbtree */
1192	zswap_rb_erase(&tree->rbroot, entry);
1193
1194	/* drop the initial reference from entry creation */
1195	zswap_entry_put(tree, entry);
1196
1197	spin_unlock(&tree->lock);
1198}
1199
1200/* frees all zswap entries for the given swap type */
1201static void zswap_frontswap_invalidate_area(unsigned type)
1202{
1203	struct zswap_tree *tree = zswap_trees[type];
1204	struct zswap_entry *entry, *n;
1205
1206	if (!tree)
1207		return;
1208
1209	/* walk the tree and free everything */
1210	spin_lock(&tree->lock);
1211	rbtree_postorder_for_each_entry_safe(entry, n, &tree->rbroot, rbnode)
1212		zswap_free_entry(entry);
1213	tree->rbroot = RB_ROOT;
1214	spin_unlock(&tree->lock);
1215	kfree(tree);
1216	zswap_trees[type] = NULL;
1217}
1218
1219static void zswap_frontswap_init(unsigned type)
1220{
1221	struct zswap_tree *tree;
1222
1223	tree = kzalloc(sizeof(*tree), GFP_KERNEL);
1224	if (!tree) {
1225		pr_err("alloc failed, zswap disabled for swap type %d\n", type);
1226		return;
1227	}
1228
1229	tree->rbroot = RB_ROOT;
1230	spin_lock_init(&tree->lock);
1231	zswap_trees[type] = tree;
1232}
1233
1234static struct frontswap_ops zswap_frontswap_ops = {
1235	.store = zswap_frontswap_store,
1236	.load = zswap_frontswap_load,
1237	.invalidate_page = zswap_frontswap_invalidate_page,
1238	.invalidate_area = zswap_frontswap_invalidate_area,
1239	.init = zswap_frontswap_init
1240};
1241
1242/*********************************
1243* debugfs functions
1244**********************************/
1245#ifdef CONFIG_DEBUG_FS
1246#include <linux/debugfs.h>
1247
1248static struct dentry *zswap_debugfs_root;
1249
1250static int __init zswap_debugfs_init(void)
1251{
1252	if (!debugfs_initialized())
1253		return -ENODEV;
1254
1255	zswap_debugfs_root = debugfs_create_dir("zswap", NULL);
1256	if (!zswap_debugfs_root)
1257		return -ENOMEM;
1258
1259	debugfs_create_u64("pool_limit_hit", S_IRUGO,
1260			zswap_debugfs_root, &zswap_pool_limit_hit);
1261	debugfs_create_u64("reject_reclaim_fail", S_IRUGO,
1262			zswap_debugfs_root, &zswap_reject_reclaim_fail);
1263	debugfs_create_u64("reject_alloc_fail", S_IRUGO,
1264			zswap_debugfs_root, &zswap_reject_alloc_fail);
1265	debugfs_create_u64("reject_kmemcache_fail", S_IRUGO,
1266			zswap_debugfs_root, &zswap_reject_kmemcache_fail);
1267	debugfs_create_u64("reject_compress_poor", S_IRUGO,
1268			zswap_debugfs_root, &zswap_reject_compress_poor);
1269	debugfs_create_u64("written_back_pages", S_IRUGO,
1270			zswap_debugfs_root, &zswap_written_back_pages);
1271	debugfs_create_u64("duplicate_entry", S_IRUGO,
1272			zswap_debugfs_root, &zswap_duplicate_entry);
1273	debugfs_create_u64("pool_total_size", S_IRUGO,
1274			zswap_debugfs_root, &zswap_pool_total_size);
1275	debugfs_create_atomic_t("stored_pages", S_IRUGO,
1276			zswap_debugfs_root, &zswap_stored_pages);
1277	debugfs_create_atomic_t("same_filled_pages", 0444,
1278			zswap_debugfs_root, &zswap_same_filled_pages);
1279
1280	return 0;
1281}
1282
1283static void __exit zswap_debugfs_exit(void)
1284{
1285	debugfs_remove_recursive(zswap_debugfs_root);
1286}
1287#else
1288static int __init zswap_debugfs_init(void)
1289{
1290	return 0;
1291}
1292
1293static void __exit zswap_debugfs_exit(void) { }
1294#endif
1295
1296/*********************************
1297* module init and exit
1298**********************************/
1299static int __init init_zswap(void)
1300{
1301	struct zswap_pool *pool;
1302	int ret;
1303
1304	zswap_init_started = true;
1305
1306	if (zswap_entry_cache_create()) {
1307		pr_err("entry cache creation failed\n");
1308		goto cache_fail;
1309	}
1310
1311	ret = cpuhp_setup_state(CPUHP_MM_ZSWP_MEM_PREPARE, "mm/zswap:prepare",
1312				zswap_dstmem_prepare, zswap_dstmem_dead);
1313	if (ret) {
1314		pr_err("dstmem alloc failed\n");
1315		goto dstmem_fail;
1316	}
1317
1318	ret = cpuhp_setup_state_multi(CPUHP_MM_ZSWP_POOL_PREPARE,
1319				      "mm/zswap_pool:prepare",
1320				      zswap_cpu_comp_prepare,
1321				      zswap_cpu_comp_dead);
1322	if (ret)
1323		goto hp_fail;
1324
1325	pool = __zswap_pool_create_fallback();
1326	if (pool) {
1327		pr_info("loaded using pool %s/%s\n", pool->tfm_name,
1328			zpool_get_type(pool->zpool));
1329		list_add(&pool->list, &zswap_pools);
1330		zswap_has_pool = true;
1331	} else {
1332		pr_err("pool creation failed\n");
1333		zswap_enabled = false;
1334	}
1335
 
 
 
 
1336	frontswap_register_ops(&zswap_frontswap_ops);
1337	if (zswap_debugfs_init())
1338		pr_warn("debugfs initialization failed\n");
1339	return 0;
1340
 
 
 
1341hp_fail:
1342	cpuhp_remove_state(CPUHP_MM_ZSWP_MEM_PREPARE);
1343dstmem_fail:
1344	zswap_entry_cache_destroy();
1345cache_fail:
1346	/* if built-in, we aren't unloaded on failure; don't allow use */
1347	zswap_init_failed = true;
1348	zswap_enabled = false;
1349	return -ENOMEM;
1350}
1351/* must be late so crypto has time to come up */
1352late_initcall(init_zswap);
1353
1354MODULE_LICENSE("GPL");
1355MODULE_AUTHOR("Seth Jennings <sjennings@variantweb.net>");
1356MODULE_DESCRIPTION("Compressed cache for swap pages");