mm/zswap.c

   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 */
  49 static u64 zswap_pool_total_size;
  50 /* The number of compressed pages currently stored in zswap */
  51 static atomic_t zswap_stored_pages = ATOMIC_INIT(0);
  52
  53 /*
  54  * The statistics below are not protected from concurrent access for
  55  * performance reasons so they may not be a 100% accurate.  However,
  56  * they do provide useful information on roughly how many times a
  57  * certain event is occurring.
  58 */
  59
  60 /* Pool limit was hit (see zswap_max_pool_percent) */
  61 static u64 zswap_pool_limit_hit;
  62 /* Pages written back when pool limit was reached */
  63 static u64 zswap_written_back_pages;
  64 /* Store failed due to a reclaim failure after pool limit was reached */
  65 static u64 zswap_reject_reclaim_fail;
  66 /* Compressed page was too big for the allocator to (optimally) store */
  67 static u64 zswap_reject_compress_poor;
  68 /* Store failed because underlying allocator could not get memory */
  69 static u64 zswap_reject_alloc_fail;
  70 /* Store failed because the entry metadata could not be allocated (rare) */
  71 static u64 zswap_reject_kmemcache_fail;
  72 /* Duplicate store was encountered (rare) */
  73 static u64 zswap_duplicate_entry;
  74
  75 /*********************************
  76 * tunables
  77 **********************************/
  78 /* Enable/disable zswap (disabled by default, fixed at boot for now) */
  79 static bool zswap_enabled __read_mostly;
  80 module_param_named(enabled, zswap_enabled, bool, 0444);
  81
  82 /* Compressor to be used by zswap (fixed at boot for now) */
  83 #define ZSWAP_COMPRESSOR_DEFAULT "lzo"
  84 static char *zswap_compressor = ZSWAP_COMPRESSOR_DEFAULT;
  85 module_param_named(compressor, zswap_compressor, charp, 0444);
  86
  87 /* The maximum percentage of memory that the compressed pool can occupy */
  88 static unsigned int zswap_max_pool_percent = 20;
  89 module_param_named(max_pool_percent,
  90                         zswap_max_pool_percent, uint, 0644);
  91
  92 /* Compressed storage to use */
  93 #define ZSWAP_ZPOOL_DEFAULT "zbud"
  94 static char *zswap_zpool_type = ZSWAP_ZPOOL_DEFAULT;
  95 module_param_named(zpool, zswap_zpool_type, charp, 0444);
  96
  97 /* zpool is shared by all of zswap backend  */
  98 static struct zpool *zswap_pool;
  99
 100 /*********************************
 101 * compression functions
 102 **********************************/
 103 /* per-cpu compression transforms */
 104 static struct crypto_comp * __percpu *zswap_comp_pcpu_tfms;
 105
 106 enum comp_op {
 107         ZSWAP_COMPOP_COMPRESS,
 108         ZSWAP_COMPOP_DECOMPRESS
 109 };
 110
 111 static int zswap_comp_op(enum comp_op op, const u8 *src, unsigned int slen,
 112                                 u8 *dst, unsigned int *dlen)
 113 {
 114         struct crypto_comp *tfm;
 115         int ret;
 116
 117         tfm = *per_cpu_ptr(zswap_comp_pcpu_tfms, get_cpu());
 118         switch (op) {
 119         case ZSWAP_COMPOP_COMPRESS:
 120                 ret = crypto_comp_compress(tfm, src, slen, dst, dlen);
 121                 break;
 122         case ZSWAP_COMPOP_DECOMPRESS:
 123                 ret = crypto_comp_decompress(tfm, src, slen, dst, dlen);
 124                 break;
 125         default:
 126                 ret = -EINVAL;
 127         }
 128
 129         put_cpu();
 130         return ret;
 131 }
 132
 133 static int __init zswap_comp_init(void)
 134 {
 135         if (!crypto_has_comp(zswap_compressor, 0, 0)) {
 136                 pr_info("%s compressor not available\n", zswap_compressor);
 137                 /* fall back to default compressor */
 138                 zswap_compressor = ZSWAP_COMPRESSOR_DEFAULT;
 139                 if (!crypto_has_comp(zswap_compressor, 0, 0))
 140                         /* can't even load the default compressor */
 141                         return -ENODEV;
 142         }
 143         pr_info("using %s compressor\n", zswap_compressor);
 144
 145         /* alloc percpu transforms */
 146         zswap_comp_pcpu_tfms = alloc_percpu(struct crypto_comp *);
 147         if (!zswap_comp_pcpu_tfms)
 148                 return -ENOMEM;
 149         return 0;
 150 }
 151
 152 static void __init zswap_comp_exit(void)
 153 {
 154         /* free percpu transforms */
 155         free_percpu(zswap_comp_pcpu_tfms);
 156 }
 157
 158 /*********************************
 159 * data structures
 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  * refcount - the number of outstanding reference to the entry. This is needed
 169  *            to protect against premature freeing of the entry by code
 170  *            concurrent calls to load, invalidate, and writeback.  The lock
 171  *            for the zswap_tree structure that contains the entry must
 172  *            be held while changing the refcount.  Since the lock must
 173  *            be held, there is no reason to also make refcount atomic.
 174  * offset - the swap offset for the entry.  Index into the red-black tree.
 175  * handle - zpool allocation handle that stores the compressed page data
 176  * length - the length in bytes of the compressed page data.  Needed during
 177  *          decompression
 178  */
 179 struct zswap_entry {
 180         struct rb_node rbnode;
 181         pgoff_t offset;
 182         int refcount;
 183         unsigned int length;
 184         unsigned long handle;
 185 };
 186
 187 struct zswap_header {
 188         swp_entry_t swpentry;
 189 };
 190
 191 /*
 192  * The tree lock in the zswap_tree struct protects a few things:
 193  * - the rbtree
 194  * - the refcount field of each entry in the tree
 195  */
 196 struct zswap_tree {
 197         struct rb_root rbroot;
 198         spinlock_t lock;
 199 };
 200
 201 static struct zswap_tree *zswap_trees[MAX_SWAPFILES];
 202
 203 /*********************************
 204 * zswap entry functions
 205 **********************************/
 206 static struct kmem_cache *zswap_entry_cache;
 207
 208 static int __init zswap_entry_cache_create(void)
 209 {
 210         zswap_entry_cache = KMEM_CACHE(zswap_entry, 0);
 211         return zswap_entry_cache == NULL;
 212 }
 213
 214 static void __init zswap_entry_cache_destroy(void)
 215 {
 216         kmem_cache_destroy(zswap_entry_cache);
 217 }
 218
 219 static struct zswap_entry *zswap_entry_cache_alloc(gfp_t gfp)
 220 {
 221         struct zswap_entry *entry;
 222         entry = kmem_cache_alloc(zswap_entry_cache, gfp);
 223         if (!entry)
 224                 return NULL;
 225         entry->refcount = 1;
 226         RB_CLEAR_NODE(&entry->rbnode);
 227         return entry;
 228 }
 229
 230 static void zswap_entry_cache_free(struct zswap_entry *entry)
 231 {
 232         kmem_cache_free(zswap_entry_cache, entry);
 233 }
 234
 235 /*********************************
 236 * rbtree functions
 237 **********************************/
 238 static struct zswap_entry *zswap_rb_search(struct rb_root *root, pgoff_t offset)
 239 {
 240         struct rb_node *node = root->rb_node;
 241         struct zswap_entry *entry;
 242
 243         while (node) {
 244                 entry = rb_entry(node, struct zswap_entry, rbnode);
 245                 if (entry->offset > offset)
 246                         node = node->rb_left;
 247                 else if (entry->offset < offset)
 248                         node = node->rb_right;
 249                 else
 250                         return entry;
 251         }
 252         return NULL;
 253 }
 254
 255 /*
 256  * In the case that a entry with the same offset is found, a pointer to
 257  * the existing entry is stored in dupentry and the function returns -EEXIST
 258  */
 259 static int zswap_rb_insert(struct rb_root *root, struct zswap_entry *entry,
 260                         struct zswap_entry **dupentry)
 261 {
 262         struct rb_node **link = &root->rb_node, *parent = NULL;
 263         struct zswap_entry *myentry;
 264
 265         while (*link) {
 266                 parent = *link;
 267                 myentry = rb_entry(parent, struct zswap_entry, rbnode);
 268                 if (myentry->offset > entry->offset)
 269                         link = &(*link)->rb_left;
 270                 else if (myentry->offset < entry->offset)
 271                         link = &(*link)->rb_right;
 272                 else {
 273                         *dupentry = myentry;
 274                         return -EEXIST;
 275                 }
 276         }
 277         rb_link_node(&entry->rbnode, parent, link);
 278         rb_insert_color(&entry->rbnode, root);
 279         return 0;
 280 }
 281
 282 static void zswap_rb_erase(struct rb_root *root, struct zswap_entry *entry)
 283 {
 284         if (!RB_EMPTY_NODE(&entry->rbnode)) {
 285                 rb_erase(&entry->rbnode, root);
 286                 RB_CLEAR_NODE(&entry->rbnode);
 287         }
 288 }
 289
 290 /*
 291  * Carries out the common pattern of freeing and entry's zpool allocation,
 292  * freeing the entry itself, and decrementing the number of stored pages.
 293  */
 294 static void zswap_free_entry(struct zswap_entry *entry)
 295 {
 296         zpool_free(zswap_pool, entry->handle);
 297         zswap_entry_cache_free(entry);
 298         atomic_dec(&zswap_stored_pages);
 299         zswap_pool_total_size = zpool_get_total_size(zswap_pool);
 300 }
 301
 302 /* caller must hold the tree lock */
 303 static void zswap_entry_get(struct zswap_entry *entry)
 304 {
 305         entry->refcount++;
 306 }
 307
 308 /* caller must hold the tree lock
 309 * remove from the tree and free it, if nobody reference the entry
 310 */
 311 static void zswap_entry_put(struct zswap_tree *tree,
 312                         struct zswap_entry *entry)
 313 {
 314         int refcount = --entry->refcount;
 315
 316         BUG_ON(refcount < 0);
 317         if (refcount == 0) {
 318                 zswap_rb_erase(&tree->rbroot, entry);
 319                 zswap_free_entry(entry);
 320         }
 321 }
 322
 323 /* caller must hold the tree lock */
 324 static struct zswap_entry *zswap_entry_find_get(struct rb_root *root,
 325                                 pgoff_t offset)
 326 {
 327         struct zswap_entry *entry = NULL;
 328
 329         entry = zswap_rb_search(root, offset);
 330         if (entry)
 331                 zswap_entry_get(entry);
 332
 333         return entry;
 334 }
 335
 336 /*********************************
 337 * per-cpu code
 338 **********************************/
 339 static DEFINE_PER_CPU(u8 *, zswap_dstmem);
 340
 341 static int __zswap_cpu_notifier(unsigned long action, unsigned long cpu)
 342 {
 343         struct crypto_comp *tfm;
 344         u8 *dst;
 345
 346         switch (action) {
 347         case CPU_UP_PREPARE:
 348                 tfm = crypto_alloc_comp(zswap_compressor, 0, 0);
 349                 if (IS_ERR(tfm)) {
 350                         pr_err("can't allocate compressor transform\n");
 351                         return NOTIFY_BAD;
 352                 }
 353                 *per_cpu_ptr(zswap_comp_pcpu_tfms, cpu) = tfm;
 354                 dst = kmalloc_node(PAGE_SIZE * 2, GFP_KERNEL, cpu_to_node(cpu));
 355                 if (!dst) {
 356                         pr_err("can't allocate compressor buffer\n");
 357                         crypto_free_comp(tfm);
 358                         *per_cpu_ptr(zswap_comp_pcpu_tfms, cpu) = NULL;
 359                         return NOTIFY_BAD;
 360                 }
 361                 per_cpu(zswap_dstmem, cpu) = dst;
 362                 break;
 363         case CPU_DEAD:
 364         case CPU_UP_CANCELED:
 365                 tfm = *per_cpu_ptr(zswap_comp_pcpu_tfms, cpu);
 366                 if (tfm) {
 367                         crypto_free_comp(tfm);
 368                         *per_cpu_ptr(zswap_comp_pcpu_tfms, cpu) = NULL;
 369                 }
 370                 dst = per_cpu(zswap_dstmem, cpu);
 371                 kfree(dst);
 372                 per_cpu(zswap_dstmem, cpu) = NULL;
 373                 break;
 374         default:
 375                 break;
 376         }
 377         return NOTIFY_OK;
 378 }
 379
 380 static int zswap_cpu_notifier(struct notifier_block *nb,
 381                                 unsigned long action, void *pcpu)
 382 {
 383         unsigned long cpu = (unsigned long)pcpu;
 384         return __zswap_cpu_notifier(action, cpu);
 385 }
 386
 387 static struct notifier_block zswap_cpu_notifier_block = {
 388         .notifier_call = zswap_cpu_notifier
 389 };
 390
 391 static int __init zswap_cpu_init(void)
 392 {
 393         unsigned long cpu;
 394
 395         cpu_notifier_register_begin();
 396         for_each_online_cpu(cpu)
 397                 if (__zswap_cpu_notifier(CPU_UP_PREPARE, cpu) != NOTIFY_OK)
 398                         goto cleanup;
 399         __register_cpu_notifier(&zswap_cpu_notifier_block);
 400         cpu_notifier_register_done();
 401         return 0;
 402
 403 cleanup:
 404         for_each_online_cpu(cpu)
 405                 __zswap_cpu_notifier(CPU_UP_CANCELED, cpu);
 406         cpu_notifier_register_done();
 407         return -ENOMEM;
 408 }
 409
 410 /*********************************
 411 * helpers
 412 **********************************/
 413 static bool zswap_is_full(void)
 414 {
 415         return totalram_pages * zswap_max_pool_percent / 100 <
 416                 DIV_ROUND_UP(zswap_pool_total_size, PAGE_SIZE);
 417 }
 418
 419 /*********************************
 420 * writeback code
 421 **********************************/
 422 /* return enum for zswap_get_swap_cache_page */
 423 enum zswap_get_swap_ret {
 424         ZSWAP_SWAPCACHE_NEW,
 425         ZSWAP_SWAPCACHE_EXIST,
 426         ZSWAP_SWAPCACHE_FAIL,
 427 };
 428
 429 /*
 430  * zswap_get_swap_cache_page
 431  *
 432  * This is an adaption of read_swap_cache_async()
 433  *
 434  * This function tries to find a page with the given swap entry
 435  * in the swapper_space address space (the swap cache).  If the page
 436  * is found, it is returned in retpage.  Otherwise, a page is allocated,
 437  * added to the swap cache, and returned in retpage.
 438  *
 439  * If success, the swap cache page is returned in retpage
 440  * Returns ZSWAP_SWAPCACHE_EXIST if page was already in the swap cache
 441  * Returns ZSWAP_SWAPCACHE_NEW if the new page needs to be populated,
 442  *     the new page is added to swapcache and locked
 443  * Returns ZSWAP_SWAPCACHE_FAIL on error
 444  */
 445 static int zswap_get_swap_cache_page(swp_entry_t entry,
 446                                 struct page **retpage)
 447 {
 448         struct page *found_page, *new_page = NULL;
 449         struct address_space *swapper_space = swap_address_space(entry);
 450         int err;
 451
 452         *retpage = NULL;
 453         do {
 454                 /*
 455                  * First check the swap cache.  Since this is normally
 456                  * called after lookup_swap_cache() failed, re-calling
 457                  * that would confuse statistics.
 458                  */
 459                 found_page = find_get_page(swapper_space, entry.val);
 460                 if (found_page)
 461                         break;
 462
 463                 /*
 464                  * Get a new page to read into from swap.
 465                  */
 466                 if (!new_page) {
 467                         new_page = alloc_page(GFP_KERNEL);
 468                         if (!new_page)
 469                                 break; /* Out of memory */
 470                 }
 471
 472                 /*
 473                  * call radix_tree_preload() while we can wait.
 474                  */
 475                 err = radix_tree_preload(GFP_KERNEL);
 476                 if (err)
 477                         break;
 478
 479                 /*
 480                  * Swap entry may have been freed since our caller observed it.
 481                  */
 482                 err = swapcache_prepare(entry);
 483                 if (err == -EEXIST) { /* seems racy */
 484                         radix_tree_preload_end();
 485                         continue;
 486                 }
 487                 if (err) { /* swp entry is obsolete ? */
 488                         radix_tree_preload_end();
 489                         break;
 490                 }
 491
 492                 /* May fail (-ENOMEM) if radix-tree node allocation failed. */
 493                 __set_page_locked(new_page);
 494                 SetPageSwapBacked(new_page);
 495                 err = __add_to_swap_cache(new_page, entry);
 496                 if (likely(!err)) {
 497                         radix_tree_preload_end();
 498                         lru_cache_add_anon(new_page);
 499                         *retpage = new_page;
 500                         return ZSWAP_SWAPCACHE_NEW;
 501                 }
 502                 radix_tree_preload_end();
 503                 ClearPageSwapBacked(new_page);
 504                 __clear_page_locked(new_page);
 505                 /*
 506                  * add_to_swap_cache() doesn't return -EEXIST, so we can safely
 507                  * clear SWAP_HAS_CACHE flag.
 508                  */
 509                 swapcache_free(entry);
 510         } while (err != -ENOMEM);
 511
 512         if (new_page)
 513                 page_cache_release(new_page);
 514         if (!found_page)
 515                 return ZSWAP_SWAPCACHE_FAIL;
 516         *retpage = found_page;
 517         return ZSWAP_SWAPCACHE_EXIST;
 518 }
 519
 520 /*
 521  * Attempts to free an entry by adding a page to the swap cache,
 522  * decompressing the entry data into the page, and issuing a
 523  * bio write to write the page back to the swap device.
 524  *
 525  * This can be thought of as a "resumed writeback" of the page
 526  * to the swap device.  We are basically resuming the same swap
 527  * writeback path that was intercepted with the frontswap_store()
 528  * in the first place.  After the page has been decompressed into
 529  * the swap cache, the compressed version stored by zswap can be
 530  * freed.
 531  */
 532 static int zswap_writeback_entry(struct zpool *pool, unsigned long handle)
 533 {
 534         struct zswap_header *zhdr;
 535         swp_entry_t swpentry;
 536         struct zswap_tree *tree;
 537         pgoff_t offset;
 538         struct zswap_entry *entry;
 539         struct page *page;
 540         u8 *src, *dst;
 541         unsigned int dlen;
 542         int ret;
 543         struct writeback_control wbc = {
 544                 .sync_mode = WB_SYNC_NONE,
 545         };
 546
 547         /* extract swpentry from data */
 548         zhdr = zpool_map_handle(pool, handle, ZPOOL_MM_RO);
 549         swpentry = zhdr->swpentry; /* here */
 550         zpool_unmap_handle(pool, handle);
 551         tree = zswap_trees[swp_type(swpentry)];
 552         offset = swp_offset(swpentry);
 553
 554         /* find and ref zswap entry */
 555         spin_lock(&tree->lock);
 556         entry = zswap_entry_find_get(&tree->rbroot, offset);
 557         if (!entry) {
 558                 /* entry was invalidated */
 559                 spin_unlock(&tree->lock);
 560                 return 0;
 561         }
 562         spin_unlock(&tree->lock);
 563         BUG_ON(offset != entry->offset);
 564
 565         /* try to allocate swap cache page */
 566         switch (zswap_get_swap_cache_page(swpentry, &page)) {
 567         case ZSWAP_SWAPCACHE_FAIL: /* no memory or invalidate happened */
 568                 ret = -ENOMEM;
 569                 goto fail;
 570
 571         case ZSWAP_SWAPCACHE_EXIST:
 572                 /* page is already in the swap cache, ignore for now */
 573                 page_cache_release(page);
 574                 ret = -EEXIST;
 575                 goto fail;
 576
 577         case ZSWAP_SWAPCACHE_NEW: /* page is locked */
 578                 /* decompress */
 579                 dlen = PAGE_SIZE;
 580                 src = (u8 *)zpool_map_handle(zswap_pool, entry->handle,
 581                                 ZPOOL_MM_RO) + sizeof(struct zswap_header);
 582                 dst = kmap_atomic(page);
 583                 ret = zswap_comp_op(ZSWAP_COMPOP_DECOMPRESS, src,
 584                                 entry->length, dst, &dlen);
 585                 kunmap_atomic(dst);
 586                 zpool_unmap_handle(zswap_pool, entry->handle);
 587                 BUG_ON(ret);
 588                 BUG_ON(dlen != PAGE_SIZE);
 589
 590                 /* page is up to date */
 591                 SetPageUptodate(page);
 592         }
 593
 594         /* move it to the tail of the inactive list after end_writeback */
 595         SetPageReclaim(page);
 596
 597         /* start writeback */
 598         __swap_writepage(page, &wbc, end_swap_bio_write);
 599         page_cache_release(page);
 600         zswap_written_back_pages++;
 601
 602         spin_lock(&tree->lock);
 603         /* drop local reference */
 604         zswap_entry_put(tree, entry);
 605
 606         /*
 607         * There are two possible situations for entry here:
 608         * (1) refcount is 1(normal case),  entry is valid and on the tree
 609         * (2) refcount is 0, entry is freed and not on the tree
 610         *     because invalidate happened during writeback
 611         *  search the tree and free the entry if find entry
 612         */
 613         if (entry == zswap_rb_search(&tree->rbroot, offset))
 614                 zswap_entry_put(tree, entry);
 615         spin_unlock(&tree->lock);
 616
 617         goto end;
 618
 619         /*
 620         * if we get here due to ZSWAP_SWAPCACHE_EXIST
 621         * a load may happening concurrently
 622         * it is safe and okay to not free the entry
 623         * if we free the entry in the following put
 624         * it it either okay to return !0
 625         */
 626 fail:
 627         spin_lock(&tree->lock);
 628         zswap_entry_put(tree, entry);
 629         spin_unlock(&tree->lock);
 630
 631 end:
 632         return ret;
 633 }
 634
 635 /*********************************
 636 * frontswap hooks
 637 **********************************/
 638 /* attempts to compress and store an single page */
 639 static int zswap_frontswap_store(unsigned type, pgoff_t offset,
 640                                 struct page *page)
 641 {
 642         struct zswap_tree *tree = zswap_trees[type];
 643         struct zswap_entry *entry, *dupentry;
 644         int ret;
 645         unsigned int dlen = PAGE_SIZE, len;
 646         unsigned long handle;
 647         char *buf;
 648         u8 *src, *dst;
 649         struct zswap_header *zhdr;
 650
 651         if (!tree) {
 652                 ret = -ENODEV;
 653                 goto reject;
 654         }
 655
 656         /* reclaim space if needed */
 657         if (zswap_is_full()) {
 658                 zswap_pool_limit_hit++;
 659                 if (zpool_shrink(zswap_pool, 1, NULL)) {
 660                         zswap_reject_reclaim_fail++;
 661                         ret = -ENOMEM;
 662                         goto reject;
 663                 }
 664         }
 665
 666         /* allocate entry */
 667         entry = zswap_entry_cache_alloc(GFP_KERNEL);
 668         if (!entry) {
 669                 zswap_reject_kmemcache_fail++;
 670                 ret = -ENOMEM;
 671                 goto reject;
 672         }
 673
 674         /* compress */
 675         dst = get_cpu_var(zswap_dstmem);
 676         src = kmap_atomic(page);
 677         ret = zswap_comp_op(ZSWAP_COMPOP_COMPRESS, src, PAGE_SIZE, dst, &dlen);
 678         kunmap_atomic(src);
 679         if (ret) {
 680                 ret = -EINVAL;
 681                 goto freepage;
 682         }
 683
 684         /* store */
 685         len = dlen + sizeof(struct zswap_header);
 686         ret = zpool_malloc(zswap_pool, len, __GFP_NORETRY | __GFP_NOWARN,
 687                 &handle);
 688         if (ret == -ENOSPC) {
 689                 zswap_reject_compress_poor++;
 690                 goto freepage;
 691         }
 692         if (ret) {
 693                 zswap_reject_alloc_fail++;
 694                 goto freepage;
 695         }
 696         zhdr = zpool_map_handle(zswap_pool, handle, ZPOOL_MM_RW);
 697         zhdr->swpentry = swp_entry(type, offset);
 698         buf = (u8 *)(zhdr + 1);
 699         memcpy(buf, dst, dlen);
 700         zpool_unmap_handle(zswap_pool, handle);
 701         put_cpu_var(zswap_dstmem);
 702
 703         /* populate entry */
 704         entry->offset = offset;
 705         entry->handle = handle;
 706         entry->length = dlen;
 707
 708         /* map */
 709         spin_lock(&tree->lock);
 710         do {
 711                 ret = zswap_rb_insert(&tree->rbroot, entry, &dupentry);
 712                 if (ret == -EEXIST) {
 713                         zswap_duplicate_entry++;
 714                         /* remove from rbtree */
 715                         zswap_rb_erase(&tree->rbroot, dupentry);
 716                         zswap_entry_put(tree, dupentry);
 717                 }
 718         } while (ret == -EEXIST);
 719         spin_unlock(&tree->lock);
 720
 721         /* update stats */
 722         atomic_inc(&zswap_stored_pages);
 723         zswap_pool_total_size = zpool_get_total_size(zswap_pool);
 724
 725         return 0;
 726
 727 freepage:
 728         put_cpu_var(zswap_dstmem);
 729         zswap_entry_cache_free(entry);
 730 reject:
 731         return ret;
 732 }
 733
 734 /*
 735  * returns 0 if the page was successfully decompressed
 736  * return -1 on entry not found or error
 737 */
 738 static int zswap_frontswap_load(unsigned type, pgoff_t offset,
 739                                 struct page *page)
 740 {
 741         struct zswap_tree *tree = zswap_trees[type];
 742         struct zswap_entry *entry;
 743         u8 *src, *dst;
 744         unsigned int dlen;
 745         int ret;
 746
 747         /* find */
 748         spin_lock(&tree->lock);
 749         entry = zswap_entry_find_get(&tree->rbroot, offset);
 750         if (!entry) {
 751                 /* entry was written back */
 752                 spin_unlock(&tree->lock);
 753                 return -1;
 754         }
 755         spin_unlock(&tree->lock);
 756
 757         /* decompress */
 758         dlen = PAGE_SIZE;
 759         src = (u8 *)zpool_map_handle(zswap_pool, entry->handle,
 760                         ZPOOL_MM_RO) + sizeof(struct zswap_header);
 761         dst = kmap_atomic(page);
 762         ret = zswap_comp_op(ZSWAP_COMPOP_DECOMPRESS, src, entry->length,
 763                 dst, &dlen);
 764         kunmap_atomic(dst);
 765         zpool_unmap_handle(zswap_pool, entry->handle);
 766         BUG_ON(ret);
 767
 768         spin_lock(&tree->lock);
 769         zswap_entry_put(tree, entry);
 770         spin_unlock(&tree->lock);
 771
 772         return 0;
 773 }
 774
 775 /* frees an entry in zswap */
 776 static void zswap_frontswap_invalidate_page(unsigned type, pgoff_t offset)
 777 {
 778         struct zswap_tree *tree = zswap_trees[type];
 779         struct zswap_entry *entry;
 780
 781         /* find */
 782         spin_lock(&tree->lock);
 783         entry = zswap_rb_search(&tree->rbroot, offset);
 784         if (!entry) {
 785                 /* entry was written back */
 786                 spin_unlock(&tree->lock);
 787                 return;
 788         }
 789
 790         /* remove from rbtree */
 791         zswap_rb_erase(&tree->rbroot, entry);
 792
 793         /* drop the initial reference from entry creation */
 794         zswap_entry_put(tree, entry);
 795
 796         spin_unlock(&tree->lock);
 797 }
 798
 799 /* frees all zswap entries for the given swap type */
 800 static void zswap_frontswap_invalidate_area(unsigned type)
 801 {
 802         struct zswap_tree *tree = zswap_trees[type];
 803         struct zswap_entry *entry, *n;
 804
 805         if (!tree)
 806                 return;
 807
 808         /* walk the tree and free everything */
 809         spin_lock(&tree->lock);
 810         rbtree_postorder_for_each_entry_safe(entry, n, &tree->rbroot, rbnode)
 811                 zswap_free_entry(entry);
 812         tree->rbroot = RB_ROOT;
 813         spin_unlock(&tree->lock);
 814         kfree(tree);
 815         zswap_trees[type] = NULL;
 816 }
 817
 818 static struct zpool_ops zswap_zpool_ops = {
 819         .evict = zswap_writeback_entry
 820 };
 821
 822 static void zswap_frontswap_init(unsigned type)
 823 {
 824         struct zswap_tree *tree;
 825
 826         tree = kzalloc(sizeof(struct zswap_tree), GFP_KERNEL);
 827         if (!tree) {
 828                 pr_err("alloc failed, zswap disabled for swap type %d\n", type);
 829                 return;
 830         }
 831
 832         tree->rbroot = RB_ROOT;
 833         spin_lock_init(&tree->lock);
 834         zswap_trees[type] = tree;
 835 }
 836
 837 static struct frontswap_ops zswap_frontswap_ops = {
 838         .store = zswap_frontswap_store,
 839         .load = zswap_frontswap_load,
 840         .invalidate_page = zswap_frontswap_invalidate_page,
 841         .invalidate_area = zswap_frontswap_invalidate_area,
 842         .init = zswap_frontswap_init
 843 };
 844
 845 /*********************************
 846 * debugfs functions
 847 **********************************/
 848 #ifdef CONFIG_DEBUG_FS
 849 #include <linux/debugfs.h>
 850
 851 static struct dentry *zswap_debugfs_root;
 852
 853 static int __init zswap_debugfs_init(void)
 854 {
 855         if (!debugfs_initialized())
 856                 return -ENODEV;
 857
 858         zswap_debugfs_root = debugfs_create_dir("zswap", NULL);
 859         if (!zswap_debugfs_root)
 860                 return -ENOMEM;
 861
 862         debugfs_create_u64("pool_limit_hit", S_IRUGO,
 863                         zswap_debugfs_root, &zswap_pool_limit_hit);
 864         debugfs_create_u64("reject_reclaim_fail", S_IRUGO,
 865                         zswap_debugfs_root, &zswap_reject_reclaim_fail);
 866         debugfs_create_u64("reject_alloc_fail", S_IRUGO,
 867                         zswap_debugfs_root, &zswap_reject_alloc_fail);
 868         debugfs_create_u64("reject_kmemcache_fail", S_IRUGO,
 869                         zswap_debugfs_root, &zswap_reject_kmemcache_fail);
 870         debugfs_create_u64("reject_compress_poor", S_IRUGO,
 871                         zswap_debugfs_root, &zswap_reject_compress_poor);
 872         debugfs_create_u64("written_back_pages", S_IRUGO,
 873                         zswap_debugfs_root, &zswap_written_back_pages);
 874         debugfs_create_u64("duplicate_entry", S_IRUGO,
 875                         zswap_debugfs_root, &zswap_duplicate_entry);
 876         debugfs_create_u64("pool_total_size", S_IRUGO,
 877                         zswap_debugfs_root, &zswap_pool_total_size);
 878         debugfs_create_atomic_t("stored_pages", S_IRUGO,
 879                         zswap_debugfs_root, &zswap_stored_pages);
 880
 881         return 0;
 882 }
 883
 884 static void __exit zswap_debugfs_exit(void)
 885 {
 886         debugfs_remove_recursive(zswap_debugfs_root);
 887 }
 888 #else
 889 static int __init zswap_debugfs_init(void)
 890 {
 891         return 0;
 892 }
 893
 894 static void __exit zswap_debugfs_exit(void) { }
 895 #endif
 896
 897 /*********************************
 898 * module init and exit
 899 **********************************/
 900 static int __init init_zswap(void)
 901 {
 902         gfp_t gfp = __GFP_NORETRY | __GFP_NOWARN;
 903
 904         if (!zswap_enabled)
 905                 return 0;
 906
 907         pr_info("loading zswap\n");
 908
 909         zswap_pool = zpool_create_pool(zswap_zpool_type, gfp, &zswap_zpool_ops);
 910         if (!zswap_pool && strcmp(zswap_zpool_type, ZSWAP_ZPOOL_DEFAULT)) {
 911                 pr_info("%s zpool not available\n", zswap_zpool_type);
 912                 zswap_zpool_type = ZSWAP_ZPOOL_DEFAULT;
 913                 zswap_pool = zpool_create_pool(zswap_zpool_type, gfp,
 914                                         &zswap_zpool_ops);
 915         }
 916         if (!zswap_pool) {
 917                 pr_err("%s zpool not available\n", zswap_zpool_type);
 918                 pr_err("zpool creation failed\n");
 919                 goto error;
 920         }
 921         pr_info("using %s pool\n", zswap_zpool_type);
 922
 923         if (zswap_entry_cache_create()) {
 924                 pr_err("entry cache creation failed\n");
 925                 goto cachefail;
 926         }
 927         if (zswap_comp_init()) {
 928                 pr_err("compressor initialization failed\n");
 929                 goto compfail;
 930         }
 931         if (zswap_cpu_init()) {
 932                 pr_err("per-cpu initialization failed\n");
 933                 goto pcpufail;
 934         }
 935
 936         frontswap_register_ops(&zswap_frontswap_ops);
 937         if (zswap_debugfs_init())
 938                 pr_warn("debugfs initialization failed\n");
 939         return 0;
 940 pcpufail:
 941         zswap_comp_exit();
 942 compfail:
 943         zswap_entry_cache_destroy();
 944 cachefail:
 945         zpool_destroy_pool(zswap_pool);
 946 error:
 947         return -ENOMEM;
 948 }
 949 /* must be late so crypto has time to come up */
 950 late_initcall(init_zswap);
 951
 952 MODULE_LICENSE("GPL");
 953 MODULE_AUTHOR("Seth Jennings <sjennings@variantweb.net>");
 954 MODULE_DESCRIPTION("Compressed cache for swap pages");