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/zbud.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 /* Number of memory pages used by the compressed pool */
  49 static u64 zswap_pool_pages;
  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, 0);
  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, 0);
  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 /*********************************
  93 * compression functions
  94 **********************************/
  95 /* per-cpu compression transforms */
  96 static struct crypto_comp * __percpu *zswap_comp_pcpu_tfms;
  97
  98 enum comp_op {
  99         ZSWAP_COMPOP_COMPRESS,
 100         ZSWAP_COMPOP_DECOMPRESS
 101 };
 102
 103 static int zswap_comp_op(enum comp_op op, const u8 *src, unsigned int slen,
 104                                 u8 *dst, unsigned int *dlen)
 105 {
 106         struct crypto_comp *tfm;
 107         int ret;
 108
 109         tfm = *per_cpu_ptr(zswap_comp_pcpu_tfms, get_cpu());
 110         switch (op) {
 111         case ZSWAP_COMPOP_COMPRESS:
 112                 ret = crypto_comp_compress(tfm, src, slen, dst, dlen);
 113                 break;
 114         case ZSWAP_COMPOP_DECOMPRESS:
 115                 ret = crypto_comp_decompress(tfm, src, slen, dst, dlen);
 116                 break;
 117         default:
 118                 ret = -EINVAL;
 119         }
 120
 121         put_cpu();
 122         return ret;
 123 }
 124
 125 static int __init zswap_comp_init(void)
 126 {
 127         if (!crypto_has_comp(zswap_compressor, 0, 0)) {
 128                 pr_info("%s compressor not available\n", zswap_compressor);
 129                 /* fall back to default compressor */
 130                 zswap_compressor = ZSWAP_COMPRESSOR_DEFAULT;
 131                 if (!crypto_has_comp(zswap_compressor, 0, 0))
 132                         /* can't even load the default compressor */
 133                         return -ENODEV;
 134         }
 135         pr_info("using %s compressor\n", zswap_compressor);
 136
 137         /* alloc percpu transforms */
 138         zswap_comp_pcpu_tfms = alloc_percpu(struct crypto_comp *);
 139         if (!zswap_comp_pcpu_tfms)
 140                 return -ENOMEM;
 141         return 0;
 142 }
 143
 144 static void zswap_comp_exit(void)
 145 {
 146         /* free percpu transforms */
 147         if (zswap_comp_pcpu_tfms)
 148                 free_percpu(zswap_comp_pcpu_tfms);
 149 }
 150
 151 /*********************************
 152 * data structures
 153 **********************************/
 154 /*
 155  * struct zswap_entry
 156  *
 157  * This structure contains the metadata for tracking a single compressed
 158  * page within zswap.
 159  *
 160  * rbnode - links the entry into red-black tree for the appropriate swap type
 161  * refcount - the number of outstanding reference to the entry. This is needed
 162  *            to protect against premature freeing of the entry by code
 163  *            concurent calls to load, invalidate, and writeback.  The lock
 164  *            for the zswap_tree structure that contains the entry must
 165  *            be held while changing the refcount.  Since the lock must
 166  *            be held, there is no reason to also make refcount atomic.
 167  * offset - the swap offset for the entry.  Index into the red-black tree.
 168  * handle - zsmalloc allocation handle that stores the compressed page data
 169  * length - the length in bytes of the compressed page data.  Needed during
 170  *           decompression
 171  */
 172 struct zswap_entry {
 173         struct rb_node rbnode;
 174         pgoff_t offset;
 175         int refcount;
 176         unsigned int length;
 177         unsigned long handle;
 178 };
 179
 180 struct zswap_header {
 181         swp_entry_t swpentry;
 182 };
 183
 184 /*
 185  * The tree lock in the zswap_tree struct protects a few things:
 186  * - the rbtree
 187  * - the refcount field of each entry in the tree
 188  */
 189 struct zswap_tree {
 190         struct rb_root rbroot;
 191         spinlock_t lock;
 192         struct zbud_pool *pool;
 193 };
 194
 195 static struct zswap_tree *zswap_trees[MAX_SWAPFILES];
 196
 197 /*********************************
 198 * zswap entry functions
 199 **********************************/
 200 static struct kmem_cache *zswap_entry_cache;
 201
 202 static int zswap_entry_cache_create(void)
 203 {
 204         zswap_entry_cache = KMEM_CACHE(zswap_entry, 0);
 205         return (zswap_entry_cache == NULL);
 206 }
 207
 208 static void zswap_entry_cache_destory(void)
 209 {
 210         kmem_cache_destroy(zswap_entry_cache);
 211 }
 212
 213 static struct zswap_entry *zswap_entry_cache_alloc(gfp_t gfp)
 214 {
 215         struct zswap_entry *entry;
 216         entry = kmem_cache_alloc(zswap_entry_cache, gfp);
 217         if (!entry)
 218                 return NULL;
 219         entry->refcount = 1;
 220         RB_CLEAR_NODE(&entry->rbnode);
 221         return entry;
 222 }
 223
 224 static void zswap_entry_cache_free(struct zswap_entry *entry)
 225 {
 226         kmem_cache_free(zswap_entry_cache, entry);
 227 }
 228
 229 /*********************************
 230 * rbtree functions
 231 **********************************/
 232 static struct zswap_entry *zswap_rb_search(struct rb_root *root, pgoff_t offset)
 233 {
 234         struct rb_node *node = root->rb_node;
 235         struct zswap_entry *entry;
 236
 237         while (node) {
 238                 entry = rb_entry(node, struct zswap_entry, rbnode);
 239                 if (entry->offset > offset)
 240                         node = node->rb_left;
 241                 else if (entry->offset < offset)
 242                         node = node->rb_right;
 243                 else
 244                         return entry;
 245         }
 246         return NULL;
 247 }
 248
 249 /*
 250  * In the case that a entry with the same offset is found, a pointer to
 251  * the existing entry is stored in dupentry and the function returns -EEXIST
 252  */
 253 static int zswap_rb_insert(struct rb_root *root, struct zswap_entry *entry,
 254                         struct zswap_entry **dupentry)
 255 {
 256         struct rb_node **link = &root->rb_node, *parent = NULL;
 257         struct zswap_entry *myentry;
 258
 259         while (*link) {
 260                 parent = *link;
 261                 myentry = rb_entry(parent, struct zswap_entry, rbnode);
 262                 if (myentry->offset > entry->offset)
 263                         link = &(*link)->rb_left;
 264                 else if (myentry->offset < entry->offset)
 265                         link = &(*link)->rb_right;
 266                 else {
 267                         *dupentry = myentry;
 268                         return -EEXIST;
 269                 }
 270         }
 271         rb_link_node(&entry->rbnode, parent, link);
 272         rb_insert_color(&entry->rbnode, root);
 273         return 0;
 274 }
 275
 276 static void zswap_rb_erase(struct rb_root *root, struct zswap_entry *entry)
 277 {
 278         if (!RB_EMPTY_NODE(&entry->rbnode)) {
 279                 rb_erase(&entry->rbnode, root);
 280                 RB_CLEAR_NODE(&entry->rbnode);
 281         }
 282 }
 283
 284 /*
 285  * Carries out the common pattern of freeing and entry's zsmalloc allocation,
 286  * freeing the entry itself, and decrementing the number of stored pages.
 287  */
 288 static void zswap_free_entry(struct zswap_tree *tree,
 289                         struct zswap_entry *entry)
 290 {
 291         zbud_free(tree->pool, entry->handle);
 292         zswap_entry_cache_free(entry);
 293         atomic_dec(&zswap_stored_pages);
 294         zswap_pool_pages = zbud_get_pool_size(tree->pool);
 295 }
 296
 297 /* caller must hold the tree lock */
 298 static void zswap_entry_get(struct zswap_entry *entry)
 299 {
 300         entry->refcount++;
 301 }
 302
 303 /* caller must hold the tree lock
 304 * remove from the tree and free it, if nobody reference the entry
 305 */
 306 static void zswap_entry_put(struct zswap_tree *tree,
 307                         struct zswap_entry *entry)
 308 {
 309         int refcount = --entry->refcount;
 310
 311         BUG_ON(refcount < 0);
 312         if (refcount == 0) {
 313                 zswap_rb_erase(&tree->rbroot, entry);
 314                 zswap_free_entry(tree, entry);
 315         }
 316 }
 317
 318 /* caller must hold the tree lock */
 319 static struct zswap_entry *zswap_entry_find_get(struct rb_root *root,
 320                                 pgoff_t offset)
 321 {
 322         struct zswap_entry *entry = NULL;
 323
 324         entry = zswap_rb_search(root, offset);
 325         if (entry)
 326                 zswap_entry_get(entry);
 327
 328         return entry;
 329 }
 330
 331 /*********************************
 332 * per-cpu code
 333 **********************************/
 334 static DEFINE_PER_CPU(u8 *, zswap_dstmem);
 335
 336 static int __zswap_cpu_notifier(unsigned long action, unsigned long cpu)
 337 {
 338         struct crypto_comp *tfm;
 339         u8 *dst;
 340
 341         switch (action) {
 342         case CPU_UP_PREPARE:
 343                 tfm = crypto_alloc_comp(zswap_compressor, 0, 0);
 344                 if (IS_ERR(tfm)) {
 345                         pr_err("can't allocate compressor transform\n");
 346                         return NOTIFY_BAD;
 347                 }
 348                 *per_cpu_ptr(zswap_comp_pcpu_tfms, cpu) = tfm;
 349                 dst = kmalloc(PAGE_SIZE * 2, GFP_KERNEL);
 350                 if (!dst) {
 351                         pr_err("can't allocate compressor buffer\n");
 352                         crypto_free_comp(tfm);
 353                         *per_cpu_ptr(zswap_comp_pcpu_tfms, cpu) = NULL;
 354                         return NOTIFY_BAD;
 355                 }
 356                 per_cpu(zswap_dstmem, cpu) = dst;
 357                 break;
 358         case CPU_DEAD:
 359         case CPU_UP_CANCELED:
 360                 tfm = *per_cpu_ptr(zswap_comp_pcpu_tfms, cpu);
 361                 if (tfm) {
 362                         crypto_free_comp(tfm);
 363                         *per_cpu_ptr(zswap_comp_pcpu_tfms, cpu) = NULL;
 364                 }
 365                 dst = per_cpu(zswap_dstmem, cpu);
 366                 kfree(dst);
 367                 per_cpu(zswap_dstmem, cpu) = NULL;
 368                 break;
 369         default:
 370                 break;
 371         }
 372         return NOTIFY_OK;
 373 }
 374
 375 static int zswap_cpu_notifier(struct notifier_block *nb,
 376                                 unsigned long action, void *pcpu)
 377 {
 378         unsigned long cpu = (unsigned long)pcpu;
 379         return __zswap_cpu_notifier(action, cpu);
 380 }
 381
 382 static struct notifier_block zswap_cpu_notifier_block = {
 383         .notifier_call = zswap_cpu_notifier
 384 };
 385
 386 static int zswap_cpu_init(void)
 387 {
 388         unsigned long cpu;
 389
 390         get_online_cpus();
 391         for_each_online_cpu(cpu)
 392                 if (__zswap_cpu_notifier(CPU_UP_PREPARE, cpu) != NOTIFY_OK)
 393                         goto cleanup;
 394         register_cpu_notifier(&zswap_cpu_notifier_block);
 395         put_online_cpus();
 396         return 0;
 397
 398 cleanup:
 399         for_each_online_cpu(cpu)
 400                 __zswap_cpu_notifier(CPU_UP_CANCELED, cpu);
 401         put_online_cpus();
 402         return -ENOMEM;
 403 }
 404
 405 /*********************************
 406 * helpers
 407 **********************************/
 408 static bool zswap_is_full(void)
 409 {
 410         return (totalram_pages * zswap_max_pool_percent / 100 <
 411                 zswap_pool_pages);
 412 }
 413
 414 /*********************************
 415 * writeback code
 416 **********************************/
 417 /* return enum for zswap_get_swap_cache_page */
 418 enum zswap_get_swap_ret {
 419         ZSWAP_SWAPCACHE_NEW,
 420         ZSWAP_SWAPCACHE_EXIST,
 421         ZSWAP_SWAPCACHE_FAIL,
 422 };
 423
 424 /*
 425  * zswap_get_swap_cache_page
 426  *
 427  * This is an adaption of read_swap_cache_async()
 428  *
 429  * This function tries to find a page with the given swap entry
 430  * in the swapper_space address space (the swap cache).  If the page
 431  * is found, it is returned in retpage.  Otherwise, a page is allocated,
 432  * added to the swap cache, and returned in retpage.
 433  *
 434  * If success, the swap cache page is returned in retpage
 435  * Returns ZSWAP_SWAPCACHE_EXIST if page was already in the swap cache
 436  * Returns ZSWAP_SWAPCACHE_NEW if the new page needs to be populated,
 437  *     the new page is added to swapcache and locked
 438  * Returns ZSWAP_SWAPCACHE_FAIL on error
 439  */
 440 static int zswap_get_swap_cache_page(swp_entry_t entry,
 441                                 struct page **retpage)
 442 {
 443         struct page *found_page, *new_page = NULL;
 444         struct address_space *swapper_space = swap_address_space(entry);
 445         int err;
 446
 447         *retpage = NULL;
 448         do {
 449                 /*
 450                  * First check the swap cache.  Since this is normally
 451                  * called after lookup_swap_cache() failed, re-calling
 452                  * that would confuse statistics.
 453                  */
 454                 found_page = find_get_page(swapper_space, entry.val);
 455                 if (found_page)
 456                         break;
 457
 458                 /*
 459                  * Get a new page to read into from swap.
 460                  */
 461                 if (!new_page) {
 462                         new_page = alloc_page(GFP_KERNEL);
 463                         if (!new_page)
 464                                 break; /* Out of memory */
 465                 }
 466
 467                 /*
 468                  * call radix_tree_preload() while we can wait.
 469                  */
 470                 err = radix_tree_preload(GFP_KERNEL);
 471                 if (err)
 472                         break;
 473
 474                 /*
 475                  * Swap entry may have been freed since our caller observed it.
 476                  */
 477                 err = swapcache_prepare(entry);
 478                 if (err == -EEXIST) { /* seems racy */
 479                         radix_tree_preload_end();
 480                         continue;
 481                 }
 482                 if (err) { /* swp entry is obsolete ? */
 483                         radix_tree_preload_end();
 484                         break;
 485                 }
 486
 487                 /* May fail (-ENOMEM) if radix-tree node allocation failed. */
 488                 __set_page_locked(new_page);
 489                 SetPageSwapBacked(new_page);
 490                 err = __add_to_swap_cache(new_page, entry);
 491                 if (likely(!err)) {
 492                         radix_tree_preload_end();
 493                         lru_cache_add_anon(new_page);
 494                         *retpage = new_page;
 495                         return ZSWAP_SWAPCACHE_NEW;
 496                 }
 497                 radix_tree_preload_end();
 498                 ClearPageSwapBacked(new_page);
 499                 __clear_page_locked(new_page);
 500                 /*
 501                  * add_to_swap_cache() doesn't return -EEXIST, so we can safely
 502                  * clear SWAP_HAS_CACHE flag.
 503                  */
 504                 swapcache_free(entry, NULL);
 505         } while (err != -ENOMEM);
 506
 507         if (new_page)
 508                 page_cache_release(new_page);
 509         if (!found_page)
 510                 return ZSWAP_SWAPCACHE_FAIL;
 511         *retpage = found_page;
 512         return ZSWAP_SWAPCACHE_EXIST;
 513 }
 514
 515 /*
 516  * Attempts to free an entry by adding a page to the swap cache,
 517  * decompressing the entry data into the page, and issuing a
 518  * bio write to write the page back to the swap device.
 519  *
 520  * This can be thought of as a "resumed writeback" of the page
 521  * to the swap device.  We are basically resuming the same swap
 522  * writeback path that was intercepted with the frontswap_store()
 523  * in the first place.  After the page has been decompressed into
 524  * the swap cache, the compressed version stored by zswap can be
 525  * freed.
 526  */
 527 static int zswap_writeback_entry(struct zbud_pool *pool, unsigned long handle)
 528 {
 529         struct zswap_header *zhdr;
 530         swp_entry_t swpentry;
 531         struct zswap_tree *tree;
 532         pgoff_t offset;
 533         struct zswap_entry *entry;
 534         struct page *page;
 535         u8 *src, *dst;
 536         unsigned int dlen;
 537         int ret;
 538         struct writeback_control wbc = {
 539                 .sync_mode = WB_SYNC_NONE,
 540         };
 541
 542         /* extract swpentry from data */
 543         zhdr = zbud_map(pool, handle);
 544         swpentry = zhdr->swpentry; /* here */
 545         zbud_unmap(pool, handle);
 546         tree = zswap_trees[swp_type(swpentry)];
 547         offset = swp_offset(swpentry);
 548         BUG_ON(pool != tree->pool);
 549
 550         /* find and ref zswap entry */
 551         spin_lock(&tree->lock);
 552         entry = zswap_entry_find_get(&tree->rbroot, offset);
 553         if (!entry) {
 554                 /* entry was invalidated */
 555                 spin_unlock(&tree->lock);
 556                 return 0;
 557         }
 558         spin_unlock(&tree->lock);
 559         BUG_ON(offset != entry->offset);
 560
 561         /* try to allocate swap cache page */
 562         switch (zswap_get_swap_cache_page(swpentry, &page)) {
 563         case ZSWAP_SWAPCACHE_FAIL: /* no memory or invalidate happened */
 564                 ret = -ENOMEM;
 565                 goto fail;
 566
 567         case ZSWAP_SWAPCACHE_EXIST:
 568                 /* page is already in the swap cache, ignore for now */
 569                 page_cache_release(page);
 570                 ret = -EEXIST;
 571                 goto fail;
 572
 573         case ZSWAP_SWAPCACHE_NEW: /* page is locked */
 574                 /* decompress */
 575                 dlen = PAGE_SIZE;
 576                 src = (u8 *)zbud_map(tree->pool, entry->handle) +
 577                         sizeof(struct zswap_header);
 578                 dst = kmap_atomic(page);
 579                 ret = zswap_comp_op(ZSWAP_COMPOP_DECOMPRESS, src,
 580                                 entry->length, dst, &dlen);
 581                 kunmap_atomic(dst);
 582                 zbud_unmap(tree->pool, entry->handle);
 583                 BUG_ON(ret);
 584                 BUG_ON(dlen != PAGE_SIZE);
 585
 586                 /* page is up to date */
 587                 SetPageUptodate(page);
 588         }
 589
 590         /* move it to the tail of the inactive list after end_writeback */
 591         SetPageReclaim(page);
 592
 593         /* start writeback */
 594         __swap_writepage(page, &wbc, end_swap_bio_write);
 595         page_cache_release(page);
 596         zswap_written_back_pages++;
 597
 598         spin_lock(&tree->lock);
 599         /* drop local reference */
 600         zswap_entry_put(tree, entry);
 601
 602         /*
 603         * There are two possible situations for entry here:
 604         * (1) refcount is 1(normal case),  entry is valid and on the tree
 605         * (2) refcount is 0, entry is freed and not on the tree
 606         *     because invalidate happened during writeback
 607         *  search the tree and free the entry if find entry
 608         */
 609         if (entry == zswap_rb_search(&tree->rbroot, offset))
 610                 zswap_entry_put(tree, entry);
 611         spin_unlock(&tree->lock);
 612
 613         goto end;
 614
 615         /*
 616         * if we get here due to ZSWAP_SWAPCACHE_EXIST
 617         * a load may happening concurrently
 618         * it is safe and okay to not free the entry
 619         * if we free the entry in the following put
 620         * it it either okay to return !0
 621         */
 622 fail:
 623         spin_lock(&tree->lock);
 624         zswap_entry_put(tree, entry);
 625         spin_unlock(&tree->lock);
 626
 627 end:
 628         return ret;
 629 }
 630
 631 /*********************************
 632 * frontswap hooks
 633 **********************************/
 634 /* attempts to compress and store an single page */
 635 static int zswap_frontswap_store(unsigned type, pgoff_t offset,
 636                                 struct page *page)
 637 {
 638         struct zswap_tree *tree = zswap_trees[type];
 639         struct zswap_entry *entry, *dupentry;
 640         int ret;
 641         unsigned int dlen = PAGE_SIZE, len;
 642         unsigned long handle;
 643         char *buf;
 644         u8 *src, *dst;
 645         struct zswap_header *zhdr;
 646
 647         if (!tree) {
 648                 ret = -ENODEV;
 649                 goto reject;
 650         }
 651
 652         /* reclaim space if needed */
 653         if (zswap_is_full()) {
 654                 zswap_pool_limit_hit++;
 655                 if (zbud_reclaim_page(tree->pool, 8)) {
 656                         zswap_reject_reclaim_fail++;
 657                         ret = -ENOMEM;
 658                         goto reject;
 659                 }
 660         }
 661
 662         /* allocate entry */
 663         entry = zswap_entry_cache_alloc(GFP_KERNEL);
 664         if (!entry) {
 665                 zswap_reject_kmemcache_fail++;
 666                 ret = -ENOMEM;
 667                 goto reject;
 668         }
 669
 670         /* compress */
 671         dst = get_cpu_var(zswap_dstmem);
 672         src = kmap_atomic(page);
 673         ret = zswap_comp_op(ZSWAP_COMPOP_COMPRESS, src, PAGE_SIZE, dst, &dlen);
 674         kunmap_atomic(src);
 675         if (ret) {
 676                 ret = -EINVAL;
 677                 goto freepage;
 678         }
 679
 680         /* store */
 681         len = dlen + sizeof(struct zswap_header);
 682         ret = zbud_alloc(tree->pool, len, __GFP_NORETRY | __GFP_NOWARN,
 683                 &handle);
 684         if (ret == -ENOSPC) {
 685                 zswap_reject_compress_poor++;
 686                 goto freepage;
 687         }
 688         if (ret) {
 689                 zswap_reject_alloc_fail++;
 690                 goto freepage;
 691         }
 692         zhdr = zbud_map(tree->pool, handle);
 693         zhdr->swpentry = swp_entry(type, offset);
 694         buf = (u8 *)(zhdr + 1);
 695         memcpy(buf, dst, dlen);
 696         zbud_unmap(tree->pool, handle);
 697         put_cpu_var(zswap_dstmem);
 698
 699         /* populate entry */
 700         entry->offset = offset;
 701         entry->handle = handle;
 702         entry->length = dlen;
 703
 704         /* map */
 705         spin_lock(&tree->lock);
 706         do {
 707                 ret = zswap_rb_insert(&tree->rbroot, entry, &dupentry);
 708                 if (ret == -EEXIST) {
 709                         zswap_duplicate_entry++;
 710                         /* remove from rbtree */
 711                         zswap_rb_erase(&tree->rbroot, dupentry);
 712                         zswap_entry_put(tree, dupentry);
 713                 }
 714         } while (ret == -EEXIST);
 715         spin_unlock(&tree->lock);
 716
 717         /* update stats */
 718         atomic_inc(&zswap_stored_pages);
 719         zswap_pool_pages = zbud_get_pool_size(tree->pool);
 720
 721         return 0;
 722
 723 freepage:
 724         put_cpu_var(zswap_dstmem);
 725         zswap_entry_cache_free(entry);
 726 reject:
 727         return ret;
 728 }
 729
 730 /*
 731  * returns 0 if the page was successfully decompressed
 732  * return -1 on entry not found or error
 733 */
 734 static int zswap_frontswap_load(unsigned type, pgoff_t offset,
 735                                 struct page *page)
 736 {
 737         struct zswap_tree *tree = zswap_trees[type];
 738         struct zswap_entry *entry;
 739         u8 *src, *dst;
 740         unsigned int dlen;
 741         int ret;
 742
 743         /* find */
 744         spin_lock(&tree->lock);
 745         entry = zswap_entry_find_get(&tree->rbroot, offset);
 746         if (!entry) {
 747                 /* entry was written back */
 748                 spin_unlock(&tree->lock);
 749                 return -1;
 750         }
 751         spin_unlock(&tree->lock);
 752
 753         /* decompress */
 754         dlen = PAGE_SIZE;
 755         src = (u8 *)zbud_map(tree->pool, entry->handle) +
 756                         sizeof(struct zswap_header);
 757         dst = kmap_atomic(page);
 758         ret = zswap_comp_op(ZSWAP_COMPOP_DECOMPRESS, src, entry->length,
 759                 dst, &dlen);
 760         kunmap_atomic(dst);
 761         zbud_unmap(tree->pool, entry->handle);
 762         BUG_ON(ret);
 763
 764         spin_lock(&tree->lock);
 765         zswap_entry_put(tree, entry);
 766         spin_unlock(&tree->lock);
 767
 768         return 0;
 769 }
 770
 771 /* frees an entry in zswap */
 772 static void zswap_frontswap_invalidate_page(unsigned type, pgoff_t offset)
 773 {
 774         struct zswap_tree *tree = zswap_trees[type];
 775         struct zswap_entry *entry;
 776
 777         /* find */
 778         spin_lock(&tree->lock);
 779         entry = zswap_rb_search(&tree->rbroot, offset);
 780         if (!entry) {
 781                 /* entry was written back */
 782                 spin_unlock(&tree->lock);
 783                 return;
 784         }
 785
 786         /* remove from rbtree */
 787         zswap_rb_erase(&tree->rbroot, entry);
 788
 789         /* drop the initial reference from entry creation */
 790         zswap_entry_put(tree, entry);
 791
 792         spin_unlock(&tree->lock);
 793 }
 794
 795 /* frees all zswap entries for the given swap type */
 796 static void zswap_frontswap_invalidate_area(unsigned type)
 797 {
 798         struct zswap_tree *tree = zswap_trees[type];
 799         struct zswap_entry *entry, *n;
 800
 801         if (!tree)
 802                 return;
 803
 804         /* walk the tree and free everything */
 805         spin_lock(&tree->lock);
 806         rbtree_postorder_for_each_entry_safe(entry, n, &tree->rbroot, rbnode)
 807                 zswap_free_entry(tree, entry);
 808         tree->rbroot = RB_ROOT;
 809         spin_unlock(&tree->lock);
 810
 811         zbud_destroy_pool(tree->pool);
 812         kfree(tree);
 813         zswap_trees[type] = NULL;
 814 }
 815
 816 static struct zbud_ops zswap_zbud_ops = {
 817         .evict = zswap_writeback_entry
 818 };
 819
 820 static void zswap_frontswap_init(unsigned type)
 821 {
 822         struct zswap_tree *tree;
 823
 824         tree = kzalloc(sizeof(struct zswap_tree), GFP_KERNEL);
 825         if (!tree)
 826                 goto err;
 827         tree->pool = zbud_create_pool(GFP_KERNEL, &zswap_zbud_ops);
 828         if (!tree->pool)
 829                 goto freetree;
 830         tree->rbroot = RB_ROOT;
 831         spin_lock_init(&tree->lock);
 832         zswap_trees[type] = tree;
 833         return;
 834
 835 freetree:
 836         kfree(tree);
 837 err:
 838         pr_err("alloc failed, zswap disabled for swap type %d\n", type);
 839 }
 840
 841 static struct frontswap_ops zswap_frontswap_ops = {
 842         .store = zswap_frontswap_store,
 843         .load = zswap_frontswap_load,
 844         .invalidate_page = zswap_frontswap_invalidate_page,
 845         .invalidate_area = zswap_frontswap_invalidate_area,
 846         .init = zswap_frontswap_init
 847 };
 848
 849 /*********************************
 850 * debugfs functions
 851 **********************************/
 852 #ifdef CONFIG_DEBUG_FS
 853 #include <linux/debugfs.h>
 854
 855 static struct dentry *zswap_debugfs_root;
 856
 857 static int __init zswap_debugfs_init(void)
 858 {
 859         if (!debugfs_initialized())
 860                 return -ENODEV;
 861
 862         zswap_debugfs_root = debugfs_create_dir("zswap", NULL);
 863         if (!zswap_debugfs_root)
 864                 return -ENOMEM;
 865
 866         debugfs_create_u64("pool_limit_hit", S_IRUGO,
 867                         zswap_debugfs_root, &zswap_pool_limit_hit);
 868         debugfs_create_u64("reject_reclaim_fail", S_IRUGO,
 869                         zswap_debugfs_root, &zswap_reject_reclaim_fail);
 870         debugfs_create_u64("reject_alloc_fail", S_IRUGO,
 871                         zswap_debugfs_root, &zswap_reject_alloc_fail);
 872         debugfs_create_u64("reject_kmemcache_fail", S_IRUGO,
 873                         zswap_debugfs_root, &zswap_reject_kmemcache_fail);
 874         debugfs_create_u64("reject_compress_poor", S_IRUGO,
 875                         zswap_debugfs_root, &zswap_reject_compress_poor);
 876         debugfs_create_u64("written_back_pages", S_IRUGO,
 877                         zswap_debugfs_root, &zswap_written_back_pages);
 878         debugfs_create_u64("duplicate_entry", S_IRUGO,
 879                         zswap_debugfs_root, &zswap_duplicate_entry);
 880         debugfs_create_u64("pool_pages", S_IRUGO,
 881                         zswap_debugfs_root, &zswap_pool_pages);
 882         debugfs_create_atomic_t("stored_pages", S_IRUGO,
 883                         zswap_debugfs_root, &zswap_stored_pages);
 884
 885         return 0;
 886 }
 887
 888 static void __exit zswap_debugfs_exit(void)
 889 {
 890         debugfs_remove_recursive(zswap_debugfs_root);
 891 }
 892 #else
 893 static int __init zswap_debugfs_init(void)
 894 {
 895         return 0;
 896 }
 897
 898 static void __exit zswap_debugfs_exit(void) { }
 899 #endif
 900
 901 /*********************************
 902 * module init and exit
 903 **********************************/
 904 static int __init init_zswap(void)
 905 {
 906         if (!zswap_enabled)
 907                 return 0;
 908
 909         pr_info("loading zswap\n");
 910         if (zswap_entry_cache_create()) {
 911                 pr_err("entry cache creation failed\n");
 912                 goto error;
 913         }
 914         if (zswap_comp_init()) {
 915                 pr_err("compressor initialization failed\n");
 916                 goto compfail;
 917         }
 918         if (zswap_cpu_init()) {
 919                 pr_err("per-cpu initialization failed\n");
 920                 goto pcpufail;
 921         }
 922         frontswap_register_ops(&zswap_frontswap_ops);
 923         if (zswap_debugfs_init())
 924                 pr_warn("debugfs initialization failed\n");
 925         return 0;
 926 pcpufail:
 927         zswap_comp_exit();
 928 compfail:
 929         zswap_entry_cache_destory();
 930 error:
 931         return -ENOMEM;
 932 }
 933 /* must be late so crypto has time to come up */
 934 late_initcall(init_zswap);
 935
 936 MODULE_LICENSE("GPL");
 937 MODULE_AUTHOR("Seth Jennings <sjenning@linux.vnet.ibm.com>");
 938 MODULE_DESCRIPTION("Compressed cache for swap pages");