| 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 | #include <linux/sec_debug.h> |
| 45 | |
| 46 | #ifdef CONFIG_ZSWAP_SAME_PAGE_SHARING |
| 47 | #include <linux/jhash.h> |
| 48 | #endif |
| 49 | |
| 50 | /********************************* |
| 51 | * statistics |
| 52 | **********************************/ |
| 53 | /* Total bytes used by the compressed storage */ |
| 54 | static u64 zswap_pool_total_size; |
| 55 | /* Number of memory pages used by the compressed pool */ |
| 56 | u64 zswap_pool_pages; |
| 57 | /* The number of compressed pages currently stored in zswap */ |
| 58 | atomic_t zswap_stored_pages = ATOMIC_INIT(0); |
| 59 | |
| 60 | #ifdef CONFIG_ZSWAP_SAME_PAGE_SHARING |
| 61 | /* The number of swapped out pages which are identified as duplicate |
| 62 | to the existing zswap pages. Compression and storing of these |
| 63 | pages is avoided */ |
| 64 | static atomic_t zswap_duplicate_pages = ATOMIC_INIT(0); |
| 65 | #endif |
| 66 | |
| 67 | /* |
| 68 | * The statistics below are not protected from concurrent access for |
| 69 | * performance reasons so they may not be a 100% accurate. However, |
| 70 | * they do provide useful information on roughly how many times a |
| 71 | * certain event is occurring. |
| 72 | */ |
| 73 | |
| 74 | /* Pool limit was hit (see zswap_max_pool_percent) */ |
| 75 | static u64 zswap_pool_limit_hit; |
| 76 | /* Pages written back when pool limit was reached */ |
| 77 | static u64 zswap_written_back_pages; |
| 78 | /* Store failed due to a reclaim failure after pool limit was reached */ |
| 79 | static u64 zswap_reject_reclaim_fail; |
| 80 | /* Compressed page was too big for the allocator to (optimally) store */ |
| 81 | static u64 zswap_reject_compress_poor; |
| 82 | /* Store failed because underlying allocator could not get memory */ |
| 83 | static u64 zswap_reject_alloc_fail; |
| 84 | /* Store failed because the entry metadata could not be allocated (rare) */ |
| 85 | static u64 zswap_reject_kmemcache_fail; |
| 86 | /* Duplicate store was encountered (rare) */ |
| 87 | static u64 zswap_duplicate_entry; |
| 88 | |
| 89 | /* The number of zero pages currently stored in zswap */ |
| 90 | static atomic_t zswap_zero_pages = ATOMIC_INIT(0); |
| 91 | |
| 92 | /********************************* |
| 93 | * tunables |
| 94 | **********************************/ |
| 95 | |
| 96 | /* Enable/disable zswap (disabled by default) */ |
| 97 | static bool zswap_enabled = 1; |
| 98 | static int zswap_enabled_param_set(const char *, |
| 99 | const struct kernel_param *); |
| 100 | static struct kernel_param_ops zswap_enabled_param_ops = { |
| 101 | .set = zswap_enabled_param_set, |
| 102 | .get = param_get_bool, |
| 103 | }; |
| 104 | module_param_cb(enabled, &zswap_enabled_param_ops, &zswap_enabled, 0644); |
| 105 | |
| 106 | /* Crypto compressor to use */ |
| 107 | #define ZSWAP_COMPRESSOR_DEFAULT "lzo" |
| 108 | #define ZSWAP_COMPRESSOR "lz4" |
| 109 | static char *zswap_compressor = ZSWAP_COMPRESSOR; |
| 110 | static int zswap_compressor_param_set(const char *, |
| 111 | const struct kernel_param *); |
| 112 | static struct kernel_param_ops zswap_compressor_param_ops = { |
| 113 | .set = zswap_compressor_param_set, |
| 114 | .get = param_get_charp, |
| 115 | .free = param_free_charp, |
| 116 | }; |
| 117 | module_param_cb(compressor, &zswap_compressor_param_ops, |
| 118 | &zswap_compressor, 0644); |
| 119 | |
| 120 | /* Compressed storage zpool to use */ |
| 121 | #define ZSWAP_ZPOOL_DEFAULT "zsmalloc" |
| 122 | static char *zswap_zpool_type = ZSWAP_ZPOOL_DEFAULT; |
| 123 | static int zswap_zpool_param_set(const char *, const struct kernel_param *); |
| 124 | static struct kernel_param_ops zswap_zpool_param_ops = { |
| 125 | .set = zswap_zpool_param_set, |
| 126 | .get = param_get_charp, |
| 127 | .free = param_free_charp, |
| 128 | }; |
| 129 | module_param_cb(zpool, &zswap_zpool_param_ops, &zswap_zpool_type, 0644); |
| 130 | |
| 131 | /* The maximum percentage of memory that the compressed pool can occupy */ |
| 132 | static unsigned int zswap_max_pool_percent = 50; |
| 133 | module_param_named(max_pool_percent, zswap_max_pool_percent, uint, 0644); |
| 134 | |
| 135 | /********************************* |
| 136 | * data structures |
| 137 | **********************************/ |
| 138 | #ifdef CONFIG_ZSWAP_SAME_PAGE_SHARING |
| 139 | /* |
| 140 | * struct zswap_handle |
| 141 | * This structure contains the metadata for tracking single zpool |
| 142 | * allocation. |
| 143 | * |
| 144 | * rbnode - links the handle into red-black tree |
| 145 | * checksum - 32-bit checksum value of the page swapped to zswap |
| 146 | * ref_count - number of pages sharing this handle |
| 147 | * length - the length in bytes of the compressed page data. |
| 148 | * Needed during decompression. |
| 149 | * handle - zpool allocation handle that stores the compressed page data |
| 150 | */ |
| 151 | struct zswap_handle { |
| 152 | struct rb_node rbnode; |
| 153 | u32 checksum; |
| 154 | u32 ref_count; |
| 155 | unsigned int length; |
| 156 | unsigned long handle; |
| 157 | }; |
| 158 | #endif |
| 159 | |
| 160 | struct zswap_pool { |
| 161 | struct zpool *zpool; |
| 162 | struct crypto_comp * __percpu *tfm; |
| 163 | struct kref kref; |
| 164 | struct list_head list; |
| 165 | struct rcu_head rcu_head; |
| 166 | struct notifier_block notifier; |
| 167 | char tfm_name[CRYPTO_MAX_ALG_NAME]; |
| 168 | }; |
| 169 | |
| 170 | /* |
| 171 | * struct zswap_entry |
| 172 | * |
| 173 | * This structure contains the metadata for tracking a single compressed |
| 174 | * page within zswap. |
| 175 | * |
| 176 | * #ifndef CONFIG_ZSWAP_SAME_PAGE_SHARING |
| 177 | * rbnode - links the entry into red-black tree for the appropriate swap type |
| 178 | * offset - the swap offset for the entry. Index into the red-black tree. |
| 179 | * refcount - the number of outstanding reference to the entry. This is needed |
| 180 | * to protect against premature freeing of the entry by code |
| 181 | * concurrent calls to load, invalidate, and writeback. The lock |
| 182 | * for the zswap_tree structure that contains the entry must |
| 183 | * be held while changing the refcount. Since the lock must |
| 184 | * be held, there is no reason to also make refcount atomic. |
| 185 | * length - the length in bytes of the compressed page data. Needed during |
| 186 | * decompression |
| 187 | * pool - the zswap_pool the entry's data is in |
| 188 | * handle - zpool allocation handle that stores the compressed page data |
| 189 | * zero_flag - the flag indicating the page for the zswap_entry is a zero page. |
| 190 | * zswap does not store the page during compression. |
| 191 | * It memsets the page with 0 during decompression. |
| 192 | * #else |
| 193 | * zhandle - pointer to struct zswap_handle where length and handle are moved into. |
| 194 | * #endif |
| 195 | */ |
| 196 | #ifndef CONFIG_ZSWAP_SAME_PAGE_SHARING |
| 197 | struct zswap_entry { |
| 198 | struct rb_node rbnode; |
| 199 | pgoff_t offset; |
| 200 | int refcount; |
| 201 | unsigned int length; |
| 202 | struct zswap_pool *pool; |
| 203 | unsigned long handle; |
| 204 | unsigned char zero_flag; |
| 205 | }; |
| 206 | #else |
| 207 | struct zswap_entry { |
| 208 | struct rb_node rbnode; |
| 209 | pgoff_t offset; |
| 210 | int refcount; |
| 211 | struct zswap_pool *pool; |
| 212 | struct zswap_handle *zhandle; |
| 213 | unsigned char zero_flag; |
| 214 | }; |
| 215 | #endif |
| 216 | |
| 217 | #ifdef CONFIG_ZSWAP_ENABLE_WRITEBACK |
| 218 | struct zswap_header { |
| 219 | swp_entry_t swpentry; |
| 220 | }; |
| 221 | #endif |
| 222 | |
| 223 | /* |
| 224 | * The tree lock in the zswap_tree struct protects a few things: |
| 225 | * - the rbtree |
| 226 | * - the refcount field of each entry in the tree |
| 227 | */ |
| 228 | struct zswap_tree { |
| 229 | struct rb_root rbroot; |
| 230 | #ifdef CONFIG_ZSWAP_SAME_PAGE_SHARING |
| 231 | struct rb_root zhandleroot; |
| 232 | void *buffer; |
| 233 | #endif |
| 234 | spinlock_t lock; |
| 235 | }; |
| 236 | |
| 237 | static struct zswap_tree *zswap_trees[MAX_SWAPFILES]; |
| 238 | |
| 239 | /* RCU-protected iteration */ |
| 240 | static LIST_HEAD(zswap_pools); |
| 241 | /* protects zswap_pools list modification */ |
| 242 | static DEFINE_SPINLOCK(zswap_pools_lock); |
| 243 | /* pool counter to provide unique names to zpool */ |
| 244 | static atomic_t zswap_pools_count = ATOMIC_INIT(0); |
| 245 | |
| 246 | /* used by param callback function */ |
| 247 | static bool zswap_init_started; |
| 248 | |
| 249 | /* fatal error during init */ |
| 250 | static bool zswap_init_failed; |
| 251 | |
| 252 | /********************************* |
| 253 | * helpers and fwd declarations |
| 254 | **********************************/ |
| 255 | |
| 256 | #define zswap_pool_debug(msg, p) \ |
| 257 | pr_debug("%s pool %s/%s\n", msg, (p)->tfm_name, \ |
| 258 | zpool_get_type((p)->zpool)) |
| 259 | |
| 260 | static int zswap_writeback_entry(struct zpool *pool, unsigned long handle); |
| 261 | static int zswap_pool_get(struct zswap_pool *pool); |
| 262 | static void zswap_pool_put(struct zswap_pool *pool); |
| 263 | |
| 264 | static const struct zpool_ops zswap_zpool_ops = { |
| 265 | .evict = zswap_writeback_entry |
| 266 | }; |
| 267 | |
| 268 | static bool zswap_is_full(void) |
| 269 | { |
| 270 | return totalram_pages * zswap_max_pool_percent / 100 < |
| 271 | DIV_ROUND_UP(zswap_pool_total_size, PAGE_SIZE); |
| 272 | } |
| 273 | |
| 274 | static void zswap_update_total_size(void) |
| 275 | { |
| 276 | struct zswap_pool *pool; |
| 277 | u64 total = 0; |
| 278 | |
| 279 | rcu_read_lock(); |
| 280 | |
| 281 | list_for_each_entry_rcu(pool, &zswap_pools, list) |
| 282 | total += zpool_get_total_size(pool->zpool); |
| 283 | |
| 284 | rcu_read_unlock(); |
| 285 | |
| 286 | zswap_pool_total_size = total; |
| 287 | zswap_pool_pages = zswap_pool_total_size >> PAGE_SHIFT; |
| 288 | } |
| 289 | |
| 290 | /********************************* |
| 291 | * zswap entry functions |
| 292 | **********************************/ |
| 293 | static struct kmem_cache *zswap_entry_cache; |
| 294 | |
| 295 | static int __init zswap_entry_cache_create(void) |
| 296 | { |
| 297 | zswap_entry_cache = KMEM_CACHE(zswap_entry, 0); |
| 298 | return zswap_entry_cache == NULL; |
| 299 | } |
| 300 | |
| 301 | static void __init zswap_entry_cache_destroy(void) |
| 302 | { |
| 303 | kmem_cache_destroy(zswap_entry_cache); |
| 304 | } |
| 305 | |
| 306 | static struct zswap_entry *zswap_entry_cache_alloc(gfp_t gfp) |
| 307 | { |
| 308 | struct zswap_entry *entry; |
| 309 | entry = kmem_cache_alloc(zswap_entry_cache, gfp); |
| 310 | if (!entry) |
| 311 | return NULL; |
| 312 | entry->refcount = 1; |
| 313 | entry->zero_flag = 0; |
| 314 | #ifdef CONFIG_ZSWAP_SAME_PAGE_SHARING |
| 315 | entry->zhandle = NULL; |
| 316 | #endif |
| 317 | RB_CLEAR_NODE(&entry->rbnode); |
| 318 | return entry; |
| 319 | } |
| 320 | |
| 321 | static void zswap_entry_cache_free(struct zswap_entry *entry) |
| 322 | { |
| 323 | kmem_cache_free(zswap_entry_cache, entry); |
| 324 | } |
| 325 | |
| 326 | #ifdef CONFIG_ZSWAP_SAME_PAGE_SHARING |
| 327 | /********************************* |
| 328 | * zswap handle functions |
| 329 | **********************************/ |
| 330 | static struct kmem_cache *zswap_handle_cache; |
| 331 | |
| 332 | static int zswap_handle_cache_create(void) |
| 333 | { |
| 334 | zswap_handle_cache = KMEM_CACHE(zswap_handle, 0); |
| 335 | return zswap_handle_cache == NULL; |
| 336 | } |
| 337 | |
| 338 | static void __init zswap_handle_cache_destroy(void) |
| 339 | { |
| 340 | kmem_cache_destroy(zswap_handle_cache); |
| 341 | } |
| 342 | |
| 343 | static struct zswap_handle *zswap_handle_cache_alloc(gfp_t gfp) |
| 344 | { |
| 345 | struct zswap_handle *zhandle; |
| 346 | zhandle = kmem_cache_alloc(zswap_handle_cache, gfp); |
| 347 | if (!zhandle) |
| 348 | return NULL; |
| 349 | zhandle->ref_count = 1; |
| 350 | RB_CLEAR_NODE(&zhandle->rbnode); |
| 351 | return zhandle; |
| 352 | } |
| 353 | |
| 354 | static void zswap_handle_cache_free(struct zswap_handle *zhandle) |
| 355 | { |
| 356 | kmem_cache_free(zswap_handle_cache, zhandle); |
| 357 | } |
| 358 | #endif |
| 359 | |
| 360 | /********************************* |
| 361 | * rbtree functions |
| 362 | **********************************/ |
| 363 | static struct zswap_entry *zswap_rb_search(struct rb_root *root, pgoff_t offset) |
| 364 | { |
| 365 | struct rb_node *node = root->rb_node; |
| 366 | struct zswap_entry *entry; |
| 367 | |
| 368 | while (node) { |
| 369 | entry = rb_entry(node, struct zswap_entry, rbnode); |
| 370 | if (entry->offset > offset) |
| 371 | node = node->rb_left; |
| 372 | else if (entry->offset < offset) |
| 373 | node = node->rb_right; |
| 374 | else |
| 375 | return entry; |
| 376 | } |
| 377 | return NULL; |
| 378 | } |
| 379 | |
| 380 | /* |
| 381 | * In the case that a entry with the same offset is found, a pointer to |
| 382 | * the existing entry is stored in dupentry and the function returns -EEXIST |
| 383 | */ |
| 384 | static int zswap_rb_insert(struct rb_root *root, struct zswap_entry *entry, |
| 385 | struct zswap_entry **dupentry) |
| 386 | { |
| 387 | struct rb_node **link = &root->rb_node, *parent = NULL; |
| 388 | struct zswap_entry *myentry; |
| 389 | |
| 390 | while (*link) { |
| 391 | parent = *link; |
| 392 | myentry = rb_entry(parent, struct zswap_entry, rbnode); |
| 393 | if (myentry->offset > entry->offset) |
| 394 | link = &(*link)->rb_left; |
| 395 | else if (myentry->offset < entry->offset) |
| 396 | link = &(*link)->rb_right; |
| 397 | else { |
| 398 | *dupentry = myentry; |
| 399 | return -EEXIST; |
| 400 | } |
| 401 | } |
| 402 | rb_link_node(&entry->rbnode, parent, link); |
| 403 | rb_insert_color(&entry->rbnode, root); |
| 404 | return 0; |
| 405 | } |
| 406 | |
| 407 | static void zswap_rb_erase(struct rb_root *root, struct zswap_entry *entry) |
| 408 | { |
| 409 | if (!RB_EMPTY_NODE(&entry->rbnode)) { |
| 410 | rb_erase(&entry->rbnode, root); |
| 411 | RB_CLEAR_NODE(&entry->rbnode); |
| 412 | } |
| 413 | } |
| 414 | |
| 415 | #ifdef CONFIG_ZSWAP_SAME_PAGE_SHARING |
| 416 | static struct zswap_handle *zswap_handle_rb_search(struct rb_root *root, |
| 417 | u32 checksum) |
| 418 | { |
| 419 | struct rb_node *node = root->rb_node; |
| 420 | struct zswap_handle *zhandle; |
| 421 | |
| 422 | while (node) { |
| 423 | zhandle = rb_entry(node, struct zswap_handle, rbnode); |
| 424 | if (zhandle->checksum > checksum) |
| 425 | node = node->rb_left; |
| 426 | else if (zhandle->checksum < checksum) |
| 427 | node = node->rb_right; |
| 428 | else |
| 429 | return zhandle; |
| 430 | } |
| 431 | return NULL; |
| 432 | } |
| 433 | |
| 434 | /* |
| 435 | * In the case that zhandle with the same checksum is found, a pointer to |
| 436 | * the existing zhandle is stored in duphandle and the function returns -EEXIST |
| 437 | */ |
| 438 | static int zswap_handle_rb_insert(struct rb_root *root, |
| 439 | struct zswap_handle *zhandle, |
| 440 | struct zswap_handle **duphandle) |
| 441 | { |
| 442 | struct rb_node **link = &root->rb_node, *parent = NULL; |
| 443 | struct zswap_handle *myhandle; |
| 444 | |
| 445 | while (*link) { |
| 446 | parent = *link; |
| 447 | myhandle = rb_entry(parent, struct zswap_handle, rbnode); |
| 448 | if (myhandle->checksum > zhandle->checksum) |
| 449 | link = &parent->rb_left; |
| 450 | else if (myhandle->checksum < zhandle->checksum) |
| 451 | link = &parent->rb_right; |
| 452 | else { |
| 453 | *duphandle = myhandle; |
| 454 | return -EEXIST; |
| 455 | } |
| 456 | } |
| 457 | rb_link_node(&zhandle->rbnode, parent, link); |
| 458 | rb_insert_color(&zhandle->rbnode, root); |
| 459 | return 0; |
| 460 | } |
| 461 | |
| 462 | static void zswap_handle_erase(struct rb_root *root, |
| 463 | struct zswap_handle *zhandle) |
| 464 | { |
| 465 | if (!RB_EMPTY_NODE(&zhandle->rbnode)) { |
| 466 | rb_erase(&zhandle->rbnode, root); |
| 467 | RB_CLEAR_NODE(&zhandle->rbnode); |
| 468 | } |
| 469 | } |
| 470 | |
| 471 | static void zswap_free_handle(struct zswap_pool *pool, struct zswap_handle *zhandle) |
| 472 | { |
| 473 | zpool_free(pool->zpool, zhandle->handle); |
| 474 | zswap_handle_cache_free(zhandle); |
| 475 | } |
| 476 | |
| 477 | /* This function searches for the same page in the zhandle RB-Tree based on the |
| 478 | * checksum value of the new page. If the same page is found the zhandle of that |
| 479 | * page is returned. |
| 480 | */ |
| 481 | static struct zswap_handle *zswap_same_page_search(struct zswap_pool *pool, struct zswap_tree *tree, |
| 482 | u8 *uncmem, u32 checksum) |
| 483 | { |
| 484 | int ret = 0; |
| 485 | unsigned int dlen = PAGE_SIZE; |
| 486 | u8 *src = NULL, *dst = NULL; |
| 487 | struct zswap_handle *myhandle = NULL; |
| 488 | struct crypto_comp *tfm; |
| 489 | |
| 490 | myhandle = zswap_handle_rb_search(&tree->zhandleroot, checksum); |
| 491 | if (myhandle) { |
| 492 | /* Compare memory contents */ |
| 493 | dst = (u8 *)tree->buffer; |
| 494 | src = (u8 *)zpool_map_handle(pool->zpool, |
| 495 | myhandle->handle, ZPOOL_MM_RO_NOWAIT); |
| 496 | if (!src) |
| 497 | return NULL; |
| 498 | |
| 499 | if (myhandle->length == PAGE_SIZE) |
| 500 | copy_page(dst, src); |
| 501 | else { |
| 502 | tfm = *get_cpu_ptr(pool->tfm); |
| 503 | ret = crypto_comp_decompress(tfm, src, myhandle->length, dst, &dlen); |
| 504 | put_cpu_ptr(pool->tfm); |
| 505 | } |
| 506 | |
| 507 | zpool_unmap_handle(pool->zpool, myhandle->handle); |
| 508 | BUG_ON(ret); |
| 509 | |
| 510 | ret = memcmp(dst, uncmem, PAGE_SIZE); |
| 511 | if (ret) |
| 512 | myhandle = NULL; |
| 513 | } |
| 514 | return myhandle; |
| 515 | } |
| 516 | #endif |
| 517 | |
| 518 | /* |
| 519 | * Carries out the common pattern of freeing and entry's zpool allocation, |
| 520 | * freeing the entry itself, and decrementing the number of stored pages. |
| 521 | */ |
| 522 | static void zswap_free_entry(struct zswap_entry *entry) |
| 523 | { |
| 524 | if (entry->zero_flag == 1) { |
| 525 | atomic_dec(&zswap_zero_pages); |
| 526 | goto zeropage_out; |
| 527 | } |
| 528 | #ifdef CONFIG_ZSWAP_SAME_PAGE_SHARING |
| 529 | entry->zhandle->ref_count--; |
| 530 | if (!entry->zhandle->ref_count) |
| 531 | zswap_free_handle(entry->pool, entry->zhandle); |
| 532 | else |
| 533 | atomic_dec(&zswap_duplicate_pages); |
| 534 | #else |
| 535 | zpool_free(entry->pool->zpool, entry->handle); |
| 536 | #endif |
| 537 | zeropage_out: |
| 538 | zswap_pool_put(entry->pool); |
| 539 | zswap_entry_cache_free(entry); |
| 540 | atomic_dec(&zswap_stored_pages); |
| 541 | zswap_update_total_size(); |
| 542 | } |
| 543 | |
| 544 | /* caller must hold the tree lock */ |
| 545 | static void zswap_entry_get(struct zswap_entry *entry) |
| 546 | { |
| 547 | entry->refcount++; |
| 548 | } |
| 549 | |
| 550 | /* caller must hold the tree lock |
| 551 | * remove from the tree and free it, if nobody reference the entry |
| 552 | */ |
| 553 | static void zswap_entry_put(struct zswap_tree *tree, |
| 554 | struct zswap_entry *entry) |
| 555 | { |
| 556 | int refcount = --entry->refcount; |
| 557 | |
| 558 | BUG_ON(refcount < 0); |
| 559 | if (refcount == 0) { |
| 560 | #ifdef CONFIG_ZSWAP_SAME_PAGE_SHARING |
| 561 | if (entry->zhandle && entry->zhandle->ref_count == 1) |
| 562 | zswap_handle_erase(&tree->zhandleroot, entry->zhandle); |
| 563 | #endif |
| 564 | zswap_rb_erase(&tree->rbroot, entry); |
| 565 | zswap_free_entry(entry); |
| 566 | } |
| 567 | } |
| 568 | |
| 569 | /* caller must hold the tree lock */ |
| 570 | static struct zswap_entry *zswap_entry_find_get(struct rb_root *root, |
| 571 | pgoff_t offset) |
| 572 | { |
| 573 | struct zswap_entry *entry; |
| 574 | |
| 575 | entry = zswap_rb_search(root, offset); |
| 576 | if (entry) |
| 577 | zswap_entry_get(entry); |
| 578 | |
| 579 | return entry; |
| 580 | } |
| 581 | |
| 582 | /********************************* |
| 583 | * per-cpu code |
| 584 | **********************************/ |
| 585 | static DEFINE_PER_CPU(u8 *, zswap_dstmem); |
| 586 | |
| 587 | static int __zswap_cpu_dstmem_notifier(unsigned long action, unsigned long cpu) |
| 588 | { |
| 589 | u8 *dst; |
| 590 | |
| 591 | switch (action) { |
| 592 | case CPU_UP_PREPARE: |
| 593 | dst = kmalloc_node(PAGE_SIZE * 2, GFP_KERNEL, cpu_to_node(cpu)); |
| 594 | if (!dst) { |
| 595 | pr_err("can't allocate compressor buffer\n"); |
| 596 | return NOTIFY_BAD; |
| 597 | } |
| 598 | per_cpu(zswap_dstmem, cpu) = dst; |
| 599 | break; |
| 600 | case CPU_DEAD: |
| 601 | case CPU_UP_CANCELED: |
| 602 | dst = per_cpu(zswap_dstmem, cpu); |
| 603 | kfree(dst); |
| 604 | per_cpu(zswap_dstmem, cpu) = NULL; |
| 605 | break; |
| 606 | default: |
| 607 | break; |
| 608 | } |
| 609 | return NOTIFY_OK; |
| 610 | } |
| 611 | |
| 612 | static int zswap_cpu_dstmem_notifier(struct notifier_block *nb, |
| 613 | unsigned long action, void *pcpu) |
| 614 | { |
| 615 | return __zswap_cpu_dstmem_notifier(action, (unsigned long)pcpu); |
| 616 | } |
| 617 | |
| 618 | static struct notifier_block zswap_dstmem_notifier = { |
| 619 | .notifier_call = zswap_cpu_dstmem_notifier, |
| 620 | }; |
| 621 | |
| 622 | static int __init zswap_cpu_dstmem_init(void) |
| 623 | { |
| 624 | unsigned long cpu; |
| 625 | |
| 626 | cpu_notifier_register_begin(); |
| 627 | for_each_online_cpu(cpu) |
| 628 | if (__zswap_cpu_dstmem_notifier(CPU_UP_PREPARE, cpu) == |
| 629 | NOTIFY_BAD) |
| 630 | goto cleanup; |
| 631 | __register_cpu_notifier(&zswap_dstmem_notifier); |
| 632 | cpu_notifier_register_done(); |
| 633 | return 0; |
| 634 | |
| 635 | cleanup: |
| 636 | for_each_online_cpu(cpu) |
| 637 | __zswap_cpu_dstmem_notifier(CPU_UP_CANCELED, cpu); |
| 638 | cpu_notifier_register_done(); |
| 639 | return -ENOMEM; |
| 640 | } |
| 641 | |
| 642 | static void zswap_cpu_dstmem_destroy(void) |
| 643 | { |
| 644 | unsigned long cpu; |
| 645 | |
| 646 | cpu_notifier_register_begin(); |
| 647 | for_each_online_cpu(cpu) |
| 648 | __zswap_cpu_dstmem_notifier(CPU_UP_CANCELED, cpu); |
| 649 | __unregister_cpu_notifier(&zswap_dstmem_notifier); |
| 650 | cpu_notifier_register_done(); |
| 651 | } |
| 652 | |
| 653 | static int __zswap_cpu_comp_notifier(struct zswap_pool *pool, |
| 654 | unsigned long action, unsigned long cpu) |
| 655 | { |
| 656 | struct crypto_comp *tfm; |
| 657 | |
| 658 | switch (action) { |
| 659 | case CPU_UP_PREPARE: |
| 660 | if (WARN_ON(*per_cpu_ptr(pool->tfm, cpu))) |
| 661 | break; |
| 662 | tfm = crypto_alloc_comp(pool->tfm_name, 0, 0); |
| 663 | if (IS_ERR_OR_NULL(tfm)) { |
| 664 | pr_err("could not alloc crypto comp %s : %ld\n", |
| 665 | pool->tfm_name, PTR_ERR(tfm)); |
| 666 | return NOTIFY_BAD; |
| 667 | } |
| 668 | *per_cpu_ptr(pool->tfm, cpu) = tfm; |
| 669 | break; |
| 670 | case CPU_DEAD: |
| 671 | case CPU_UP_CANCELED: |
| 672 | tfm = *per_cpu_ptr(pool->tfm, cpu); |
| 673 | if (!IS_ERR_OR_NULL(tfm)) |
| 674 | crypto_free_comp(tfm); |
| 675 | *per_cpu_ptr(pool->tfm, cpu) = NULL; |
| 676 | break; |
| 677 | default: |
| 678 | break; |
| 679 | } |
| 680 | return NOTIFY_OK; |
| 681 | } |
| 682 | |
| 683 | static int zswap_cpu_comp_notifier(struct notifier_block *nb, |
| 684 | unsigned long action, void *pcpu) |
| 685 | { |
| 686 | unsigned long cpu = (unsigned long)pcpu; |
| 687 | struct zswap_pool *pool = container_of(nb, typeof(*pool), notifier); |
| 688 | |
| 689 | return __zswap_cpu_comp_notifier(pool, action, cpu); |
| 690 | } |
| 691 | |
| 692 | static int zswap_cpu_comp_init(struct zswap_pool *pool) |
| 693 | { |
| 694 | unsigned long cpu; |
| 695 | |
| 696 | memset(&pool->notifier, 0, sizeof(pool->notifier)); |
| 697 | pool->notifier.notifier_call = zswap_cpu_comp_notifier; |
| 698 | |
| 699 | cpu_notifier_register_begin(); |
| 700 | for_each_online_cpu(cpu) |
| 701 | if (__zswap_cpu_comp_notifier(pool, CPU_UP_PREPARE, cpu) == |
| 702 | NOTIFY_BAD) |
| 703 | goto cleanup; |
| 704 | __register_cpu_notifier(&pool->notifier); |
| 705 | cpu_notifier_register_done(); |
| 706 | return 0; |
| 707 | |
| 708 | cleanup: |
| 709 | for_each_online_cpu(cpu) |
| 710 | __zswap_cpu_comp_notifier(pool, CPU_UP_CANCELED, cpu); |
| 711 | cpu_notifier_register_done(); |
| 712 | return -ENOMEM; |
| 713 | } |
| 714 | |
| 715 | static void zswap_cpu_comp_destroy(struct zswap_pool *pool) |
| 716 | { |
| 717 | unsigned long cpu; |
| 718 | |
| 719 | cpu_notifier_register_begin(); |
| 720 | for_each_online_cpu(cpu) |
| 721 | __zswap_cpu_comp_notifier(pool, CPU_UP_CANCELED, cpu); |
| 722 | __unregister_cpu_notifier(&pool->notifier); |
| 723 | cpu_notifier_register_done(); |
| 724 | } |
| 725 | |
| 726 | /********************************* |
| 727 | * pool functions |
| 728 | **********************************/ |
| 729 | |
| 730 | static struct zswap_pool *__zswap_pool_current(void) |
| 731 | { |
| 732 | struct zswap_pool *pool; |
| 733 | |
| 734 | pool = list_first_or_null_rcu(&zswap_pools, typeof(*pool), list); |
| 735 | WARN_ON(!pool); |
| 736 | |
| 737 | return pool; |
| 738 | } |
| 739 | |
| 740 | static struct zswap_pool *zswap_pool_current(void) |
| 741 | { |
| 742 | assert_spin_locked(&zswap_pools_lock); |
| 743 | |
| 744 | return __zswap_pool_current(); |
| 745 | } |
| 746 | |
| 747 | static struct zswap_pool *zswap_pool_current_get(void) |
| 748 | { |
| 749 | struct zswap_pool *pool; |
| 750 | |
| 751 | rcu_read_lock(); |
| 752 | |
| 753 | pool = __zswap_pool_current(); |
| 754 | if (!pool || !zswap_pool_get(pool)) |
| 755 | pool = NULL; |
| 756 | |
| 757 | rcu_read_unlock(); |
| 758 | |
| 759 | return pool; |
| 760 | } |
| 761 | |
| 762 | #ifdef CONFIG_ZSWAP_ENABLE_WRITEBACK |
| 763 | static struct zswap_pool *zswap_pool_last_get(void) |
| 764 | { |
| 765 | struct zswap_pool *pool, *last = NULL; |
| 766 | |
| 767 | rcu_read_lock(); |
| 768 | |
| 769 | list_for_each_entry_rcu(pool, &zswap_pools, list) |
| 770 | last = pool; |
| 771 | if (!WARN_ON(!last) && !zswap_pool_get(last)) |
| 772 | last = NULL; |
| 773 | |
| 774 | rcu_read_unlock(); |
| 775 | |
| 776 | return last; |
| 777 | } |
| 778 | #endif |
| 779 | |
| 780 | /* type and compressor must be null-terminated */ |
| 781 | static struct zswap_pool *zswap_pool_find_get(char *type, char *compressor) |
| 782 | { |
| 783 | struct zswap_pool *pool; |
| 784 | |
| 785 | assert_spin_locked(&zswap_pools_lock); |
| 786 | |
| 787 | list_for_each_entry_rcu(pool, &zswap_pools, list) { |
| 788 | if (strcmp(pool->tfm_name, compressor)) |
| 789 | continue; |
| 790 | if (strcmp(zpool_get_type(pool->zpool), type)) |
| 791 | continue; |
| 792 | /* if we can't get it, it's about to be destroyed */ |
| 793 | if (!zswap_pool_get(pool)) |
| 794 | continue; |
| 795 | return pool; |
| 796 | } |
| 797 | |
| 798 | return NULL; |
| 799 | } |
| 800 | |
| 801 | static struct zswap_pool *zswap_pool_create(char *type, char *compressor) |
| 802 | { |
| 803 | struct zswap_pool *pool; |
| 804 | char name[38]; /* 'zswap' + 32 char (max) num + \0 */ |
| 805 | #ifdef CONFIG_ZSWAP_MIGRATION_SUPPORT |
| 806 | gfp_t gfp = __GFP_NORETRY | __GFP_NOWARN | __GFP_KSWAPD_RECLAIM | __GFP_HIGHMEM | __GFP_MOVABLE; |
| 807 | #else |
| 808 | gfp_t gfp = __GFP_NORETRY | __GFP_NOWARN | __GFP_KSWAPD_RECLAIM | __GFP_HIGHMEM; |
| 809 | #endif |
| 810 | |
| 811 | pool = kzalloc(sizeof(*pool), GFP_KERNEL); |
| 812 | if (!pool) { |
| 813 | pr_err("pool alloc failed\n"); |
| 814 | return NULL; |
| 815 | } |
| 816 | |
| 817 | /* unique name for each pool specifically required by zsmalloc */ |
| 818 | snprintf(name, 38, "zswap%x", atomic_inc_return(&zswap_pools_count)); |
| 819 | |
| 820 | pool->zpool = zpool_create_pool(type, name, gfp, &zswap_zpool_ops); |
| 821 | if (!pool->zpool) { |
| 822 | pr_err("%s zpool not available\n", type); |
| 823 | goto error; |
| 824 | } |
| 825 | pr_debug("using %s zpool\n", zpool_get_type(pool->zpool)); |
| 826 | |
| 827 | strlcpy(pool->tfm_name, compressor, sizeof(pool->tfm_name)); |
| 828 | pool->tfm = alloc_percpu(struct crypto_comp *); |
| 829 | if (!pool->tfm) { |
| 830 | pr_err("percpu alloc failed\n"); |
| 831 | goto error; |
| 832 | } |
| 833 | |
| 834 | if (zswap_cpu_comp_init(pool)) |
| 835 | goto error; |
| 836 | pr_debug("using %s compressor\n", pool->tfm_name); |
| 837 | |
| 838 | /* being the current pool takes 1 ref; this func expects the |
| 839 | * caller to always add the new pool as the current pool |
| 840 | */ |
| 841 | kref_init(&pool->kref); |
| 842 | INIT_LIST_HEAD(&pool->list); |
| 843 | |
| 844 | zswap_pool_debug("created", pool); |
| 845 | |
| 846 | return pool; |
| 847 | |
| 848 | error: |
| 849 | free_percpu(pool->tfm); |
| 850 | if (pool->zpool) |
| 851 | zpool_destroy_pool(pool->zpool); |
| 852 | kfree(pool); |
| 853 | return NULL; |
| 854 | } |
| 855 | |
| 856 | static __init struct zswap_pool *__zswap_pool_create_fallback(void) |
| 857 | { |
| 858 | if (!crypto_has_comp(zswap_compressor, 0, 0)) { |
| 859 | if (!strcmp(zswap_compressor, ZSWAP_COMPRESSOR_DEFAULT)) { |
| 860 | pr_err("default compressor %s not available\n", |
| 861 | zswap_compressor); |
| 862 | return NULL; |
| 863 | } |
| 864 | pr_err("compressor %s not available, using default %s\n", |
| 865 | zswap_compressor, ZSWAP_COMPRESSOR_DEFAULT); |
| 866 | param_free_charp(&zswap_compressor); |
| 867 | zswap_compressor = ZSWAP_COMPRESSOR_DEFAULT; |
| 868 | } |
| 869 | if (!zpool_has_pool(zswap_zpool_type)) { |
| 870 | if (!strcmp(zswap_zpool_type, ZSWAP_ZPOOL_DEFAULT)) { |
| 871 | pr_err("default zpool %s not available\n", |
| 872 | zswap_zpool_type); |
| 873 | return NULL; |
| 874 | } |
| 875 | pr_err("zpool %s not available, using default %s\n", |
| 876 | zswap_zpool_type, ZSWAP_ZPOOL_DEFAULT); |
| 877 | param_free_charp(&zswap_zpool_type); |
| 878 | zswap_zpool_type = ZSWAP_ZPOOL_DEFAULT; |
| 879 | } |
| 880 | |
| 881 | return zswap_pool_create(zswap_zpool_type, zswap_compressor); |
| 882 | } |
| 883 | |
| 884 | static void zswap_pool_destroy(struct zswap_pool *pool) |
| 885 | { |
| 886 | zswap_pool_debug("destroying", pool); |
| 887 | |
| 888 | zswap_cpu_comp_destroy(pool); |
| 889 | free_percpu(pool->tfm); |
| 890 | zpool_destroy_pool(pool->zpool); |
| 891 | kfree(pool); |
| 892 | } |
| 893 | |
| 894 | static int __must_check zswap_pool_get(struct zswap_pool *pool) |
| 895 | { |
| 896 | return kref_get_unless_zero(&pool->kref); |
| 897 | } |
| 898 | |
| 899 | static void __zswap_pool_release(struct rcu_head *head) |
| 900 | { |
| 901 | struct zswap_pool *pool = container_of(head, typeof(*pool), rcu_head); |
| 902 | |
| 903 | /* nobody should have been able to get a kref... */ |
| 904 | WARN_ON(kref_get_unless_zero(&pool->kref)); |
| 905 | |
| 906 | /* pool is now off zswap_pools list and has no references. */ |
| 907 | zswap_pool_destroy(pool); |
| 908 | } |
| 909 | |
| 910 | static void __zswap_pool_empty(struct kref *kref) |
| 911 | { |
| 912 | struct zswap_pool *pool; |
| 913 | |
| 914 | pool = container_of(kref, typeof(*pool), kref); |
| 915 | |
| 916 | spin_lock(&zswap_pools_lock); |
| 917 | |
| 918 | WARN_ON(pool == zswap_pool_current()); |
| 919 | |
| 920 | list_del_rcu(&pool->list); |
| 921 | call_rcu(&pool->rcu_head, __zswap_pool_release); |
| 922 | |
| 923 | spin_unlock(&zswap_pools_lock); |
| 924 | } |
| 925 | |
| 926 | static void zswap_pool_put(struct zswap_pool *pool) |
| 927 | { |
| 928 | kref_put(&pool->kref, __zswap_pool_empty); |
| 929 | } |
| 930 | |
| 931 | /********************************* |
| 932 | * param callbacks |
| 933 | **********************************/ |
| 934 | |
| 935 | /* val must be a null-terminated string */ |
| 936 | static int __zswap_param_set(const char *val, const struct kernel_param *kp, |
| 937 | char *type, char *compressor) |
| 938 | { |
| 939 | struct zswap_pool *pool, *put_pool = NULL; |
| 940 | char *s = strstrip((char *)val); |
| 941 | int ret; |
| 942 | |
| 943 | if (zswap_init_failed) { |
| 944 | pr_err("can't set param, initialization failed\n"); |
| 945 | return -ENODEV; |
| 946 | } |
| 947 | |
| 948 | /* no change required */ |
| 949 | if (!strcmp(s, *(char **)kp->arg)) |
| 950 | return 0; |
| 951 | |
| 952 | /* if this is load-time (pre-init) param setting, |
| 953 | * don't create a pool; that's done during init. |
| 954 | */ |
| 955 | if (!zswap_init_started) |
| 956 | return param_set_charp(s, kp); |
| 957 | |
| 958 | if (!type) { |
| 959 | if (!zpool_has_pool(s)) { |
| 960 | pr_err("zpool %s not available\n", s); |
| 961 | return -ENOENT; |
| 962 | } |
| 963 | type = s; |
| 964 | } else if (!compressor) { |
| 965 | if (!crypto_has_comp(s, 0, 0)) { |
| 966 | pr_err("compressor %s not available\n", s); |
| 967 | return -ENOENT; |
| 968 | } |
| 969 | compressor = s; |
| 970 | } else { |
| 971 | WARN_ON(1); |
| 972 | return -EINVAL; |
| 973 | } |
| 974 | |
| 975 | spin_lock(&zswap_pools_lock); |
| 976 | |
| 977 | pool = zswap_pool_find_get(type, compressor); |
| 978 | if (pool) { |
| 979 | zswap_pool_debug("using existing", pool); |
| 980 | list_del_rcu(&pool->list); |
| 981 | } else { |
| 982 | spin_unlock(&zswap_pools_lock); |
| 983 | pool = zswap_pool_create(type, compressor); |
| 984 | spin_lock(&zswap_pools_lock); |
| 985 | } |
| 986 | |
| 987 | if (pool) |
| 988 | ret = param_set_charp(s, kp); |
| 989 | else |
| 990 | ret = -EINVAL; |
| 991 | |
| 992 | if (!ret) { |
| 993 | put_pool = zswap_pool_current(); |
| 994 | list_add_rcu(&pool->list, &zswap_pools); |
| 995 | } else if (pool) { |
| 996 | /* add the possibly pre-existing pool to the end of the pools |
| 997 | * list; if it's new (and empty) then it'll be removed and |
| 998 | * destroyed by the put after we drop the lock |
| 999 | */ |
| 1000 | list_add_tail_rcu(&pool->list, &zswap_pools); |
| 1001 | put_pool = pool; |
| 1002 | } |
| 1003 | |
| 1004 | spin_unlock(&zswap_pools_lock); |
| 1005 | |
| 1006 | /* drop the ref from either the old current pool, |
| 1007 | * or the new pool we failed to add |
| 1008 | */ |
| 1009 | if (put_pool) |
| 1010 | zswap_pool_put(put_pool); |
| 1011 | |
| 1012 | return ret; |
| 1013 | } |
| 1014 | |
| 1015 | static int zswap_compressor_param_set(const char *val, |
| 1016 | const struct kernel_param *kp) |
| 1017 | { |
| 1018 | return __zswap_param_set(val, kp, zswap_zpool_type, NULL); |
| 1019 | } |
| 1020 | |
| 1021 | static int zswap_zpool_param_set(const char *val, |
| 1022 | const struct kernel_param *kp) |
| 1023 | { |
| 1024 | return __zswap_param_set(val, kp, NULL, zswap_compressor); |
| 1025 | } |
| 1026 | |
| 1027 | static int zswap_enabled_param_set(const char *val, |
| 1028 | const struct kernel_param *kp) |
| 1029 | { |
| 1030 | if (zswap_init_failed) { |
| 1031 | pr_err("can't enable, initialization failed\n"); |
| 1032 | return -ENODEV; |
| 1033 | } |
| 1034 | |
| 1035 | return param_set_bool(val, kp); |
| 1036 | } |
| 1037 | |
| 1038 | /********************************* |
| 1039 | * writeback code |
| 1040 | **********************************/ |
| 1041 | /* return enum for zswap_get_swap_cache_page */ |
| 1042 | enum zswap_get_swap_ret { |
| 1043 | ZSWAP_SWAPCACHE_NEW, |
| 1044 | ZSWAP_SWAPCACHE_EXIST, |
| 1045 | ZSWAP_SWAPCACHE_FAIL, |
| 1046 | }; |
| 1047 | |
| 1048 | #ifdef CONFIG_ZSWAP_ENABLE_WRITEBACK |
| 1049 | /* |
| 1050 | * zswap_get_swap_cache_page |
| 1051 | * |
| 1052 | * This is an adaption of read_swap_cache_async() |
| 1053 | * |
| 1054 | * This function tries to find a page with the given swap entry |
| 1055 | * in the swapper_space address space (the swap cache). If the page |
| 1056 | * is found, it is returned in retpage. Otherwise, a page is allocated, |
| 1057 | * added to the swap cache, and returned in retpage. |
| 1058 | * |
| 1059 | * If success, the swap cache page is returned in retpage |
| 1060 | * Returns ZSWAP_SWAPCACHE_EXIST if page was already in the swap cache |
| 1061 | * Returns ZSWAP_SWAPCACHE_NEW if the new page needs to be populated, |
| 1062 | * the new page is added to swapcache and locked |
| 1063 | * Returns ZSWAP_SWAPCACHE_FAIL on error |
| 1064 | */ |
| 1065 | static int zswap_get_swap_cache_page(swp_entry_t entry, |
| 1066 | struct page **retpage) |
| 1067 | { |
| 1068 | bool page_was_allocated; |
| 1069 | |
| 1070 | *retpage = __read_swap_cache_async(entry, GFP_KERNEL, |
| 1071 | NULL, 0, &page_was_allocated); |
| 1072 | if (page_was_allocated) |
| 1073 | return ZSWAP_SWAPCACHE_NEW; |
| 1074 | if (!*retpage) |
| 1075 | return ZSWAP_SWAPCACHE_FAIL; |
| 1076 | return ZSWAP_SWAPCACHE_EXIST; |
| 1077 | } |
| 1078 | |
| 1079 | /* |
| 1080 | * Attempts to free an entry by adding a page to the swap cache, |
| 1081 | * decompressing the entry data into the page, and issuing a |
| 1082 | * bio write to write the page back to the swap device. |
| 1083 | * |
| 1084 | * This can be thought of as a "resumed writeback" of the page |
| 1085 | * to the swap device. We are basically resuming the same swap |
| 1086 | * writeback path that was intercepted with the frontswap_store() |
| 1087 | * in the first place. After the page has been decompressed into |
| 1088 | * the swap cache, the compressed version stored by zswap can be |
| 1089 | * freed. |
| 1090 | */ |
| 1091 | static int zswap_writeback_entry(struct zpool *pool, unsigned long handle) |
| 1092 | { |
| 1093 | struct zswap_header *zhdr; |
| 1094 | swp_entry_t swpentry; |
| 1095 | struct zswap_tree *tree; |
| 1096 | pgoff_t offset; |
| 1097 | struct zswap_entry *entry; |
| 1098 | struct page *page; |
| 1099 | struct crypto_comp *tfm; |
| 1100 | u8 *src, *dst; |
| 1101 | unsigned int dlen; |
| 1102 | int ret; |
| 1103 | struct writeback_control wbc = { |
| 1104 | .sync_mode = WB_SYNC_NONE, |
| 1105 | }; |
| 1106 | |
| 1107 | /* extract swpentry from data */ |
| 1108 | zhdr = zpool_map_handle(pool, handle, ZPOOL_MM_RO); |
| 1109 | swpentry = zhdr->swpentry; /* here */ |
| 1110 | zpool_unmap_handle(pool, handle); |
| 1111 | tree = zswap_trees[swp_type(swpentry)]; |
| 1112 | offset = swp_offset(swpentry); |
| 1113 | |
| 1114 | /* find and ref zswap entry */ |
| 1115 | spin_lock(&tree->lock); |
| 1116 | entry = zswap_entry_find_get(&tree->rbroot, offset); |
| 1117 | if (!entry) { |
| 1118 | /* entry was invalidated */ |
| 1119 | spin_unlock(&tree->lock); |
| 1120 | return 0; |
| 1121 | } |
| 1122 | spin_unlock(&tree->lock); |
| 1123 | BUG_ON(offset != entry->offset); |
| 1124 | |
| 1125 | /* try to allocate swap cache page */ |
| 1126 | switch (zswap_get_swap_cache_page(swpentry, &page)) { |
| 1127 | case ZSWAP_SWAPCACHE_FAIL: /* no memory or invalidate happened */ |
| 1128 | ret = -ENOMEM; |
| 1129 | goto fail; |
| 1130 | |
| 1131 | case ZSWAP_SWAPCACHE_EXIST: |
| 1132 | /* page is already in the swap cache, ignore for now */ |
| 1133 | page_cache_release(page); |
| 1134 | ret = -EEXIST; |
| 1135 | goto fail; |
| 1136 | |
| 1137 | case ZSWAP_SWAPCACHE_NEW: /* page is locked */ |
| 1138 | /* decompress */ |
| 1139 | dlen = PAGE_SIZE; |
| 1140 | src = (u8 *)zpool_map_handle(entry->pool->zpool, entry->handle, |
| 1141 | ZPOOL_MM_RO) + sizeof(struct zswap_header); |
| 1142 | dst = kmap_atomic(page); |
| 1143 | tfm = *get_cpu_ptr(entry->pool->tfm); |
| 1144 | ret = crypto_comp_decompress(tfm, src, entry->length, |
| 1145 | dst, &dlen); |
| 1146 | put_cpu_ptr(entry->pool->tfm); |
| 1147 | kunmap_atomic(dst); |
| 1148 | zpool_unmap_handle(entry->pool->zpool, entry->handle); |
| 1149 | BUG_ON(ret); |
| 1150 | BUG_ON(dlen != PAGE_SIZE); |
| 1151 | |
| 1152 | /* page is up to date */ |
| 1153 | SetPageUptodate(page); |
| 1154 | } |
| 1155 | |
| 1156 | /* move it to the tail of the inactive list after end_writeback */ |
| 1157 | SetPageReclaim(page); |
| 1158 | |
| 1159 | /* start writeback */ |
| 1160 | __swap_writepage(page, &wbc, end_swap_bio_write); |
| 1161 | page_cache_release(page); |
| 1162 | zswap_written_back_pages++; |
| 1163 | |
| 1164 | spin_lock(&tree->lock); |
| 1165 | /* drop local reference */ |
| 1166 | zswap_entry_put(tree, entry); |
| 1167 | |
| 1168 | /* |
| 1169 | * There are two possible situations for entry here: |
| 1170 | * (1) refcount is 1(normal case), entry is valid and on the tree |
| 1171 | * (2) refcount is 0, entry is freed and not on the tree |
| 1172 | * because invalidate happened during writeback |
| 1173 | * search the tree and free the entry if find entry |
| 1174 | */ |
| 1175 | if (entry == zswap_rb_search(&tree->rbroot, offset)) |
| 1176 | zswap_entry_put(tree, entry); |
| 1177 | spin_unlock(&tree->lock); |
| 1178 | |
| 1179 | goto end; |
| 1180 | |
| 1181 | /* |
| 1182 | * if we get here due to ZSWAP_SWAPCACHE_EXIST |
| 1183 | * a load may happening concurrently |
| 1184 | * it is safe and okay to not free the entry |
| 1185 | * if we free the entry in the following put |
| 1186 | * it it either okay to return !0 |
| 1187 | */ |
| 1188 | fail: |
| 1189 | spin_lock(&tree->lock); |
| 1190 | zswap_entry_put(tree, entry); |
| 1191 | spin_unlock(&tree->lock); |
| 1192 | |
| 1193 | end: |
| 1194 | return ret; |
| 1195 | } |
| 1196 | |
| 1197 | static int zswap_shrink(void) |
| 1198 | { |
| 1199 | struct zswap_pool *pool; |
| 1200 | int ret; |
| 1201 | |
| 1202 | pool = zswap_pool_last_get(); |
| 1203 | if (!pool) |
| 1204 | return -ENOENT; |
| 1205 | |
| 1206 | ret = zpool_shrink(pool->zpool, 1, NULL); |
| 1207 | |
| 1208 | zswap_pool_put(pool); |
| 1209 | |
| 1210 | return ret; |
| 1211 | } |
| 1212 | #else |
| 1213 | static int zswap_writeback_entry(struct zpool *pool, unsigned long handle) |
| 1214 | { |
| 1215 | return -EINVAL; |
| 1216 | } |
| 1217 | |
| 1218 | static int zswap_shrink(void) |
| 1219 | { |
| 1220 | return -EINVAL; |
| 1221 | } |
| 1222 | #endif /* CONFIG_ZSWAP_ENABLE_WRITEBACK */ |
| 1223 | |
| 1224 | static int page_zero_filled(void *ptr) |
| 1225 | { |
| 1226 | unsigned int pos; |
| 1227 | unsigned long *page; |
| 1228 | |
| 1229 | page = (unsigned long *)ptr; |
| 1230 | |
| 1231 | for (pos = 0; pos != PAGE_SIZE / sizeof(*page); pos++) { |
| 1232 | if (page[pos]) |
| 1233 | return 0; |
| 1234 | } |
| 1235 | |
| 1236 | return 1; |
| 1237 | } |
| 1238 | |
| 1239 | /********************************* |
| 1240 | * frontswap hooks |
| 1241 | **********************************/ |
| 1242 | /* attempts to compress and store an single page */ |
| 1243 | static int zswap_frontswap_store(unsigned type, pgoff_t offset, |
| 1244 | struct page *page) |
| 1245 | { |
| 1246 | struct zswap_tree *tree = zswap_trees[type]; |
| 1247 | struct zswap_entry *entry, *dupentry; |
| 1248 | struct crypto_comp *tfm; |
| 1249 | int ret; |
| 1250 | unsigned int dlen = PAGE_SIZE, len; |
| 1251 | unsigned long handle; |
| 1252 | char *buf; |
| 1253 | u8 *src, *dst; |
| 1254 | #ifdef CONFIG_ZSWAP_ENABLE_WRITEBACK |
| 1255 | struct zswap_header *zhdr; |
| 1256 | #endif |
| 1257 | #ifdef CONFIG_ZSWAP_SAME_PAGE_SHARING |
| 1258 | struct zswap_handle *zhandle = NULL, *duphandle = NULL; |
| 1259 | u32 checksum = 0; |
| 1260 | #endif |
| 1261 | |
| 1262 | if (!zswap_enabled || !tree) { |
| 1263 | ret = -ENODEV; |
| 1264 | goto reject; |
| 1265 | } |
| 1266 | |
| 1267 | /* if this page got EIO on pageout before, give up immediately */ |
| 1268 | if (PageError(page)) { |
| 1269 | ret = -ENOMEM; |
| 1270 | goto reject; |
| 1271 | } |
| 1272 | |
| 1273 | /* reclaim space if needed */ |
| 1274 | if (zswap_is_full()) { |
| 1275 | zswap_pool_limit_hit++; |
| 1276 | if (zswap_shrink()) { |
| 1277 | zswap_reject_reclaim_fail++; |
| 1278 | ret = -ENOMEM; |
| 1279 | goto reject; |
| 1280 | } |
| 1281 | } |
| 1282 | |
| 1283 | /* allocate entry */ |
| 1284 | entry = zswap_entry_cache_alloc(GFP_KERNEL); |
| 1285 | if (!entry) { |
| 1286 | zswap_reject_kmemcache_fail++; |
| 1287 | ret = -ENOMEM; |
| 1288 | goto reject; |
| 1289 | } |
| 1290 | |
| 1291 | /* if entry is successfully added, it keeps the reference */ |
| 1292 | entry->pool = zswap_pool_current_get(); |
| 1293 | if (!entry->pool) { |
| 1294 | ret = -EINVAL; |
| 1295 | goto freepage; |
| 1296 | } |
| 1297 | |
| 1298 | /* compress */ |
| 1299 | src = kmap_atomic(page); |
| 1300 | if (page_zero_filled(src)) { |
| 1301 | atomic_inc(&zswap_zero_pages); |
| 1302 | entry->zero_flag = 1; |
| 1303 | kunmap_atomic(src); |
| 1304 | |
| 1305 | #ifdef CONFIG_ZSWAP_SAME_PAGE_SHARING |
| 1306 | entry->offset = offset; |
| 1307 | goto insert_entry; |
| 1308 | #else |
| 1309 | handle = 0; |
| 1310 | dlen = PAGE_SIZE; |
| 1311 | goto zeropage_out; |
| 1312 | #endif |
| 1313 | } |
| 1314 | #ifdef CONFIG_ZSWAP_SAME_PAGE_SHARING |
| 1315 | checksum = jhash2((const u32 *)src, PAGE_SIZE / 4, 17); |
| 1316 | spin_lock(&tree->lock); |
| 1317 | zhandle = zswap_same_page_search(entry->pool, tree, src, checksum); |
| 1318 | if (zhandle) { |
| 1319 | entry->offset = offset; |
| 1320 | entry->zhandle = zhandle; |
| 1321 | entry->zhandle->ref_count++; |
| 1322 | spin_unlock(&tree->lock); |
| 1323 | kunmap_atomic(src); |
| 1324 | atomic_inc(&zswap_duplicate_pages); |
| 1325 | goto insert_entry; |
| 1326 | } |
| 1327 | spin_unlock(&tree->lock); |
| 1328 | #endif |
| 1329 | |
| 1330 | /* compress */ |
| 1331 | dst = get_cpu_var(zswap_dstmem); |
| 1332 | tfm = *get_cpu_ptr(entry->pool->tfm); |
| 1333 | ret = crypto_comp_compress(tfm, src, PAGE_SIZE, dst, &dlen); |
| 1334 | kunmap_atomic(src); |
| 1335 | put_cpu_ptr(entry->pool->tfm); |
| 1336 | if (ret) { |
| 1337 | ret = -EINVAL; |
| 1338 | goto put_dstmem; |
| 1339 | } |
| 1340 | |
| 1341 | /* store */ |
| 1342 | if (dlen > PAGE_SIZE) |
| 1343 | dlen = PAGE_SIZE; |
| 1344 | len = dlen; |
| 1345 | #ifdef CONFIG_ZSWAP_ENABLE_WRITEBACK |
| 1346 | len += sizeof(struct zswap_header); |
| 1347 | #endif |
| 1348 | ret = zpool_malloc(entry->pool->zpool, len, |
| 1349 | __GFP_NORETRY | __GFP_NOWARN | __GFP_KSWAPD_RECLAIM, |
| 1350 | &handle); |
| 1351 | if (ret == -ENOSPC) { |
| 1352 | zswap_reject_compress_poor++; |
| 1353 | goto put_dstmem; |
| 1354 | } |
| 1355 | if (ret) { |
| 1356 | zswap_reject_alloc_fail++; |
| 1357 | goto put_dstmem; |
| 1358 | } |
| 1359 | #ifdef CONFIG_ZSWAP_ENABLE_WRITEBACK |
| 1360 | zhdr = zpool_map_handle(entry->pool->zpool, handle, ZPOOL_MM_RW); |
| 1361 | zhdr->swpentry = swp_entry(type, offset); |
| 1362 | buf = (u8 *)(zhdr + 1); |
| 1363 | memcpy(buf, dst, dlen); |
| 1364 | #else |
| 1365 | buf = (u8 *)zpool_map_handle(entry->pool->zpool, handle, ZPOOL_MM_RW); |
| 1366 | if (dlen == PAGE_SIZE) { |
| 1367 | src = kmap_atomic(page); |
| 1368 | copy_page(buf, src); |
| 1369 | kunmap_atomic(src); |
| 1370 | } else |
| 1371 | memcpy(buf, dst, dlen); |
| 1372 | #endif |
| 1373 | zpool_unmap_handle(entry->pool->zpool, handle); |
| 1374 | put_cpu_var(zswap_dstmem); |
| 1375 | |
| 1376 | #ifndef CONFIG_ZSWAP_SAME_PAGE_SHARING |
| 1377 | zeropage_out: |
| 1378 | #endif |
| 1379 | /* populate entry */ |
| 1380 | entry->offset = offset; |
| 1381 | #ifdef CONFIG_ZSWAP_SAME_PAGE_SHARING |
| 1382 | zhandle = zswap_handle_cache_alloc(GFP_KERNEL); |
| 1383 | if (!zhandle) { |
| 1384 | ret = -ENOMEM; |
| 1385 | goto freeentry; |
| 1386 | } |
| 1387 | entry->zhandle = zhandle; |
| 1388 | entry->zhandle->handle = handle; |
| 1389 | entry->zhandle->length = dlen; |
| 1390 | entry->zhandle->checksum = checksum; |
| 1391 | #else |
| 1392 | entry->handle = handle; |
| 1393 | entry->length = dlen; |
| 1394 | #endif |
| 1395 | |
| 1396 | #ifdef CONFIG_ZSWAP_SAME_PAGE_SHARING |
| 1397 | spin_lock(&tree->lock); |
| 1398 | ret = zswap_handle_rb_insert(&tree->zhandleroot, |
| 1399 | entry->zhandle, &duphandle); |
| 1400 | spin_unlock(&tree->lock); |
| 1401 | |
| 1402 | insert_entry: |
| 1403 | #endif |
| 1404 | /* map */ |
| 1405 | spin_lock(&tree->lock); |
| 1406 | do { |
| 1407 | ret = zswap_rb_insert(&tree->rbroot, entry, &dupentry); |
| 1408 | if (ret == -EEXIST) { |
| 1409 | zswap_duplicate_entry++; |
| 1410 | /* remove from rbtree */ |
| 1411 | zswap_rb_erase(&tree->rbroot, dupentry); |
| 1412 | zswap_entry_put(tree, dupentry); |
| 1413 | } |
| 1414 | } while (ret == -EEXIST); |
| 1415 | spin_unlock(&tree->lock); |
| 1416 | |
| 1417 | /* update stats */ |
| 1418 | atomic_inc(&zswap_stored_pages); |
| 1419 | zswap_update_total_size(); |
| 1420 | |
| 1421 | return 0; |
| 1422 | |
| 1423 | put_dstmem: |
| 1424 | put_cpu_var(zswap_dstmem); |
| 1425 | #ifdef CONFIG_ZSWAP_SAME_PAGE_SHARING |
| 1426 | freeentry: |
| 1427 | #endif |
| 1428 | zswap_pool_put(entry->pool); |
| 1429 | freepage: |
| 1430 | zswap_entry_cache_free(entry); |
| 1431 | reject: |
| 1432 | return ret; |
| 1433 | } |
| 1434 | |
| 1435 | static void hexdump(char *title, u8 *data, int len) |
| 1436 | { |
| 1437 | int i; |
| 1438 | |
| 1439 | printk("%s: length = %d @ %p\n", title, len, data); |
| 1440 | for (i = 0; i < len; i++) { |
| 1441 | printk("%02x ", data[i]); |
| 1442 | if ((i & 0xf) == 0xf) |
| 1443 | printk("\n"); |
| 1444 | } |
| 1445 | printk("\n"); |
| 1446 | } |
| 1447 | |
| 1448 | static void check_single_bitflip(struct zswap_entry *entry, |
| 1449 | unsigned char *src, size_t src_size, unsigned char *dst) |
| 1450 | { |
| 1451 | struct crypto_comp *tfm; |
| 1452 | unsigned int dlen; |
| 1453 | int byte_offset, bit_offset, type; |
| 1454 | int i, j, ret, success = 0; |
| 1455 | char debug_msg[32]; |
| 1456 | |
| 1457 | for (i = 0; i < src_size; i++) { |
| 1458 | for (j = 0; j < 8; j++) { |
| 1459 | unsigned char org_byte = src[i]; |
| 1460 | unsigned char tmp_byte = src[i]; |
| 1461 | |
| 1462 | if (((tmp_byte >> j) & 0x1) == 0x1) |
| 1463 | tmp_byte &= ~(1 << j); |
| 1464 | else |
| 1465 | tmp_byte |= (1 << j); |
| 1466 | src[i] = tmp_byte; |
| 1467 | |
| 1468 | dlen = PAGE_SIZE; |
| 1469 | tfm = *get_cpu_ptr(entry->pool->tfm); |
| 1470 | ret = crypto_comp_decompress(tfm, src, src_size, dst, &dlen); |
| 1471 | put_cpu_ptr(entry->pool->tfm); |
| 1472 | |
| 1473 | if (ret == 0) { |
| 1474 | if (success == 0) { |
| 1475 | type = ((org_byte >> j) & 0x1); |
| 1476 | byte_offset = (i / 4) * 4; |
| 1477 | bit_offset = ((i % 4) * 8) + j; |
| 1478 | } |
| 1479 | if (success < 99) |
| 1480 | success++; |
| 1481 | } |
| 1482 | src[i] = org_byte; |
| 1483 | } |
| 1484 | } |
| 1485 | |
| 1486 | if (success) |
| 1487 | sprintf(debug_msg, "%02d/0x%16p/%02d/%d", success, src + byte_offset, bit_offset, type); |
| 1488 | else |
| 1489 | sprintf(debug_msg, "%02d/0x%16p", success, src); |
| 1490 | |
| 1491 | sec_debug_set_extra_info_zswap(debug_msg); |
| 1492 | |
| 1493 | return; |
| 1494 | } |
| 1495 | |
| 1496 | /* |
| 1497 | * returns 0 if the page was successfully decompressed |
| 1498 | * return -1 on entry not found or error |
| 1499 | */ |
| 1500 | static int zswap_frontswap_load(unsigned type, pgoff_t offset, |
| 1501 | struct page *page) |
| 1502 | { |
| 1503 | struct zswap_tree *tree = zswap_trees[type]; |
| 1504 | struct zswap_entry *entry; |
| 1505 | struct crypto_comp *tfm; |
| 1506 | u8 *src, *dst; |
| 1507 | unsigned int dlen; |
| 1508 | int ret = 0; |
| 1509 | |
| 1510 | /* find */ |
| 1511 | spin_lock(&tree->lock); |
| 1512 | entry = zswap_entry_find_get(&tree->rbroot, offset); |
| 1513 | if (!entry) { |
| 1514 | /* entry was written back */ |
| 1515 | spin_unlock(&tree->lock); |
| 1516 | return -1; |
| 1517 | } |
| 1518 | spin_unlock(&tree->lock); |
| 1519 | |
| 1520 | if (entry->zero_flag == 1) { |
| 1521 | dst = kmap_atomic(page); |
| 1522 | memset(dst, 0, PAGE_SIZE); |
| 1523 | kunmap_atomic(dst); |
| 1524 | goto zeropage_out; |
| 1525 | } |
| 1526 | |
| 1527 | /* decompress */ |
| 1528 | dlen = PAGE_SIZE; |
| 1529 | #ifdef CONFIG_ZSWAP_SAME_PAGE_SHARING |
| 1530 | src = (u8 *)zpool_map_handle(entry->pool->zpool, entry->zhandle->handle, |
| 1531 | ZPOOL_MM_RO); |
| 1532 | #else |
| 1533 | src = (u8 *)zpool_map_handle(entry->pool->zpool, entry->handle, |
| 1534 | ZPOOL_MM_RO); |
| 1535 | #endif |
| 1536 | dst = kmap_atomic(page); |
| 1537 | #ifdef CONFIG_ZSWAP_ENABLE_WRITEBACK |
| 1538 | src += sizeof(struct zswap_header); |
| 1539 | tfm = *get_cpu_ptr(entry->pool->tfm); |
| 1540 | ret = crypto_comp_decompress(tfm, src, entry->length, dst, &dlen); |
| 1541 | put_cpu_ptr(entry->pool->tfm); |
| 1542 | #else |
| 1543 | #ifdef CONFIG_ZSWAP_SAME_PAGE_SHARING |
| 1544 | if (entry->zhandle->length == PAGE_SIZE) |
| 1545 | copy_page(dst, src); |
| 1546 | else { |
| 1547 | tfm = *get_cpu_ptr(entry->pool->tfm); |
| 1548 | ret = crypto_comp_decompress(tfm, src, entry->zhandle->length, dst, &dlen); |
| 1549 | put_cpu_ptr(entry->pool->tfm); |
| 1550 | } |
| 1551 | #else |
| 1552 | if (entry->length == PAGE_SIZE) |
| 1553 | copy_page(dst, src); |
| 1554 | else { |
| 1555 | tfm = *get_cpu_ptr(entry->pool->tfm); |
| 1556 | ret = crypto_comp_decompress(tfm, src, entry->length, dst, &dlen); |
| 1557 | put_cpu_ptr(entry->pool->tfm); |
| 1558 | } |
| 1559 | #endif |
| 1560 | #endif |
| 1561 | if (ret) { |
| 1562 | #ifdef CONFIG_ZSWAP_SAME_PAGE_SHARING |
| 1563 | hexdump("src buffer", src, entry->zhandle->length); |
| 1564 | #else |
| 1565 | hexdump("src buffer", src, entry->length); |
| 1566 | #endif |
| 1567 | if (dlen) |
| 1568 | hexdump("dest buffer", dst, dlen); |
| 1569 | printk("zswap_comp_op returned %d\n", ret); |
| 1570 | #ifdef CONFIG_ZSWAP_SAME_PAGE_SHARING |
| 1571 | check_single_bitflip(entry, src, entry->zhandle->length, dst); |
| 1572 | #else |
| 1573 | check_single_bitflip(entry, src, entry->length, dst); |
| 1574 | #endif |
| 1575 | } |
| 1576 | |
| 1577 | kunmap_atomic(dst); |
| 1578 | #ifdef CONFIG_ZSWAP_SAME_PAGE_SHARING |
| 1579 | zpool_unmap_handle(entry->pool->zpool, entry->zhandle->handle); |
| 1580 | #else |
| 1581 | zpool_unmap_handle(entry->pool->zpool, entry->handle); |
| 1582 | #endif |
| 1583 | BUG_ON(ret); |
| 1584 | |
| 1585 | zeropage_out: |
| 1586 | spin_lock(&tree->lock); |
| 1587 | zswap_entry_put(tree, entry); |
| 1588 | spin_unlock(&tree->lock); |
| 1589 | |
| 1590 | return 0; |
| 1591 | } |
| 1592 | |
| 1593 | /* frees an entry in zswap */ |
| 1594 | static void zswap_frontswap_invalidate_page(unsigned type, pgoff_t offset) |
| 1595 | { |
| 1596 | struct zswap_tree *tree = zswap_trees[type]; |
| 1597 | struct zswap_entry *entry; |
| 1598 | |
| 1599 | /* find */ |
| 1600 | spin_lock(&tree->lock); |
| 1601 | entry = zswap_rb_search(&tree->rbroot, offset); |
| 1602 | if (!entry) { |
| 1603 | /* entry was written back */ |
| 1604 | spin_unlock(&tree->lock); |
| 1605 | return; |
| 1606 | } |
| 1607 | |
| 1608 | /* remove from rbtree */ |
| 1609 | zswap_rb_erase(&tree->rbroot, entry); |
| 1610 | |
| 1611 | /* drop the initial reference from entry creation */ |
| 1612 | zswap_entry_put(tree, entry); |
| 1613 | |
| 1614 | spin_unlock(&tree->lock); |
| 1615 | } |
| 1616 | |
| 1617 | /* frees all zswap entries for the given swap type */ |
| 1618 | static void zswap_frontswap_invalidate_area(unsigned type) |
| 1619 | { |
| 1620 | struct zswap_tree *tree = zswap_trees[type]; |
| 1621 | struct zswap_entry *entry, *n; |
| 1622 | |
| 1623 | if (!tree) |
| 1624 | return; |
| 1625 | |
| 1626 | /* walk the tree and free everything */ |
| 1627 | spin_lock(&tree->lock); |
| 1628 | rbtree_postorder_for_each_entry_safe(entry, n, &tree->rbroot, rbnode) |
| 1629 | zswap_free_entry(entry); |
| 1630 | tree->rbroot = RB_ROOT; |
| 1631 | #ifdef CONFIG_ZSWAP_SAME_PAGE_SHARING |
| 1632 | tree->zhandleroot = RB_ROOT; |
| 1633 | #endif |
| 1634 | spin_unlock(&tree->lock); |
| 1635 | #ifdef CONFIG_ZSWAP_SAME_PAGE_SHARING |
| 1636 | free_page((unsigned long)tree->buffer); |
| 1637 | #endif |
| 1638 | kfree(tree); |
| 1639 | zswap_trees[type] = NULL; |
| 1640 | } |
| 1641 | |
| 1642 | static void zswap_frontswap_init(unsigned type) |
| 1643 | { |
| 1644 | struct zswap_tree *tree; |
| 1645 | |
| 1646 | tree = kzalloc(sizeof(struct zswap_tree), GFP_KERNEL); |
| 1647 | if (!tree) { |
| 1648 | pr_err("alloc failed, zswap disabled for swap type %d\n", type); |
| 1649 | return; |
| 1650 | } |
| 1651 | #ifdef CONFIG_ZSWAP_SAME_PAGE_SHARING |
| 1652 | tree->buffer = (void *)__get_free_page(GFP_KERNEL | __GFP_ZERO); |
| 1653 | if (!tree->buffer) { |
| 1654 | pr_err("zswap: Error allocating compressor buffer\n"); |
| 1655 | kfree(tree); |
| 1656 | return; |
| 1657 | } |
| 1658 | tree->zhandleroot = RB_ROOT; |
| 1659 | #endif |
| 1660 | |
| 1661 | tree->rbroot = RB_ROOT; |
| 1662 | spin_lock_init(&tree->lock); |
| 1663 | zswap_trees[type] = tree; |
| 1664 | } |
| 1665 | |
| 1666 | static struct frontswap_ops zswap_frontswap_ops = { |
| 1667 | .store = zswap_frontswap_store, |
| 1668 | .load = zswap_frontswap_load, |
| 1669 | .invalidate_page = zswap_frontswap_invalidate_page, |
| 1670 | .invalidate_area = zswap_frontswap_invalidate_area, |
| 1671 | .init = zswap_frontswap_init |
| 1672 | }; |
| 1673 | |
| 1674 | /********************************* |
| 1675 | * debugfs functions |
| 1676 | **********************************/ |
| 1677 | #ifdef CONFIG_DEBUG_FS |
| 1678 | #include <linux/debugfs.h> |
| 1679 | |
| 1680 | static struct dentry *zswap_debugfs_root; |
| 1681 | |
| 1682 | static int __init zswap_debugfs_init(void) |
| 1683 | { |
| 1684 | if (!debugfs_initialized()) |
| 1685 | return -ENODEV; |
| 1686 | |
| 1687 | zswap_debugfs_root = debugfs_create_dir("zswap", NULL); |
| 1688 | if (!zswap_debugfs_root) |
| 1689 | return -ENOMEM; |
| 1690 | |
| 1691 | debugfs_create_u64("pool_limit_hit", S_IRUGO, |
| 1692 | zswap_debugfs_root, &zswap_pool_limit_hit); |
| 1693 | debugfs_create_u64("reject_reclaim_fail", S_IRUGO, |
| 1694 | zswap_debugfs_root, &zswap_reject_reclaim_fail); |
| 1695 | debugfs_create_u64("reject_alloc_fail", S_IRUGO, |
| 1696 | zswap_debugfs_root, &zswap_reject_alloc_fail); |
| 1697 | debugfs_create_u64("reject_kmemcache_fail", S_IRUGO, |
| 1698 | zswap_debugfs_root, &zswap_reject_kmemcache_fail); |
| 1699 | debugfs_create_u64("reject_compress_poor", S_IRUGO, |
| 1700 | zswap_debugfs_root, &zswap_reject_compress_poor); |
| 1701 | debugfs_create_u64("written_back_pages", S_IRUGO, |
| 1702 | zswap_debugfs_root, &zswap_written_back_pages); |
| 1703 | debugfs_create_u64("duplicate_entry", S_IRUGO, |
| 1704 | zswap_debugfs_root, &zswap_duplicate_entry); |
| 1705 | debugfs_create_u64("pool_total_size", S_IRUGO, |
| 1706 | zswap_debugfs_root, &zswap_pool_total_size); |
| 1707 | debugfs_create_u64("pool_pages", S_IRUGO, |
| 1708 | zswap_debugfs_root, &zswap_pool_pages); |
| 1709 | debugfs_create_atomic_t("stored_pages", S_IRUGO, |
| 1710 | zswap_debugfs_root, &zswap_stored_pages); |
| 1711 | #ifdef CONFIG_ZSWAP_SAME_PAGE_SHARING |
| 1712 | debugfs_create_atomic_t("duplicate_pages", S_IRUGO, |
| 1713 | zswap_debugfs_root, &zswap_duplicate_pages); |
| 1714 | #endif |
| 1715 | debugfs_create_atomic_t("zero_pages", S_IRUGO, |
| 1716 | zswap_debugfs_root, &zswap_zero_pages); |
| 1717 | |
| 1718 | return 0; |
| 1719 | } |
| 1720 | |
| 1721 | static void __exit zswap_debugfs_exit(void) |
| 1722 | { |
| 1723 | debugfs_remove_recursive(zswap_debugfs_root); |
| 1724 | } |
| 1725 | #else |
| 1726 | static int __init zswap_debugfs_init(void) |
| 1727 | { |
| 1728 | return 0; |
| 1729 | } |
| 1730 | |
| 1731 | static void __exit zswap_debugfs_exit(void) { } |
| 1732 | #endif |
| 1733 | |
| 1734 | /********************************* |
| 1735 | * module init and exit |
| 1736 | **********************************/ |
| 1737 | static int __init init_zswap(void) |
| 1738 | { |
| 1739 | struct zswap_pool *pool; |
| 1740 | |
| 1741 | zswap_init_started = true; |
| 1742 | |
| 1743 | if (zswap_entry_cache_create()) { |
| 1744 | pr_err("entry cache creation failed\n"); |
| 1745 | goto cache_fail; |
| 1746 | } |
| 1747 | |
| 1748 | #ifdef CONFIG_ZSWAP_SAME_PAGE_SHARING |
| 1749 | if (zswap_handle_cache_create()) { |
| 1750 | pr_err("handle cache creation failed\n"); |
| 1751 | goto handlecachefail; |
| 1752 | } |
| 1753 | #endif |
| 1754 | if (zswap_cpu_dstmem_init()) { |
| 1755 | pr_err("dstmem alloc failed\n"); |
| 1756 | goto dstmem_fail; |
| 1757 | } |
| 1758 | |
| 1759 | pool = __zswap_pool_create_fallback(); |
| 1760 | if (!pool) { |
| 1761 | pr_err("pool creation failed\n"); |
| 1762 | goto pool_fail; |
| 1763 | } |
| 1764 | pr_info("loaded using pool %s/%s\n", pool->tfm_name, |
| 1765 | zpool_get_type(pool->zpool)); |
| 1766 | |
| 1767 | list_add(&pool->list, &zswap_pools); |
| 1768 | |
| 1769 | frontswap_register_ops(&zswap_frontswap_ops); |
| 1770 | if (zswap_debugfs_init()) |
| 1771 | pr_warn("debugfs initialization failed\n"); |
| 1772 | return 0; |
| 1773 | |
| 1774 | pool_fail: |
| 1775 | zswap_cpu_dstmem_destroy(); |
| 1776 | dstmem_fail: |
| 1777 | #ifdef CONFIG_ZSWAP_SAME_PAGE_SHARING |
| 1778 | zswap_handle_cache_destroy(); |
| 1779 | handlecachefail: |
| 1780 | #endif |
| 1781 | zswap_entry_cache_destroy(); |
| 1782 | cache_fail: |
| 1783 | /* if built-in, we aren't unloaded on failure; don't allow use */ |
| 1784 | zswap_init_failed = true; |
| 1785 | zswap_enabled = false; |
| 1786 | return -ENOMEM; |
| 1787 | } |
| 1788 | /* must be late so crypto has time to come up */ |
| 1789 | late_initcall(init_zswap); |
| 1790 | |
| 1791 | MODULE_LICENSE("GPL"); |
| 1792 | MODULE_AUTHOR("Seth Jennings <sjennings@variantweb.net>"); |
| 1793 | MODULE_DESCRIPTION("Compressed cache for swap pages"); |