Steve Kondik | f94910d | 2016-04-14 02:52:21 -0700 | [diff] [blame] | 1 | /* |
| 2 | * linux/mm/zcache.c |
| 3 | * |
| 4 | * A cleancache backend for file pages compression. |
| 5 | * Concepts based on original zcache by Dan Magenheimer. |
| 6 | * Copyright (C) 2013 Bob Liu <bob.liu@xxxxxxxxxx> |
| 7 | * |
| 8 | * With zcache, active file pages can be compressed in memory during page |
| 9 | * reclaiming. When their data is needed again the I/O reading operation is |
| 10 | * avoided. This results in a significant performance gain under memory pressure |
| 11 | * for systems with many file pages. |
| 12 | * |
| 13 | * This program is free software; you can redistribute it and/or |
| 14 | * modify it under the terms of the GNU General Public License |
| 15 | * as published by the Free Software Foundation; either version 2 |
| 16 | * of the License, or (at your option) any later version. |
| 17 | * |
| 18 | * This program is distributed in the hope that it will be useful, |
| 19 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 20 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
| 21 | * GNU General Public License for more details. |
| 22 | */ |
| 23 | |
| 24 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt |
| 25 | |
| 26 | #include <linux/atomic.h> |
| 27 | #include <linux/cleancache.h> |
| 28 | #include <linux/cpu.h> |
| 29 | #include <linux/crypto.h> |
| 30 | #include <linux/page-flags.h> |
| 31 | #include <linux/pagemap.h> |
| 32 | #include <linux/highmem.h> |
| 33 | #include <linux/mm_types.h> |
| 34 | #include <linux/module.h> |
| 35 | #include <linux/slab.h> |
| 36 | #include <linux/spinlock.h> |
| 37 | #include <linux/radix-tree.h> |
| 38 | #include <linux/rbtree.h> |
| 39 | #include <linux/types.h> |
| 40 | #include <linux/zbud.h> |
| 41 | |
| 42 | /* |
| 43 | * Enable/disable zcache (disabled by default) |
| 44 | */ |
| 45 | static bool zcache_enabled __read_mostly; |
| 46 | module_param_named(enabled, zcache_enabled, bool, 0); |
| 47 | |
| 48 | /* |
| 49 | * Compressor to be used by zcache |
| 50 | */ |
| 51 | #define ZCACHE_COMPRESSOR_DEFAULT "lzo" |
| 52 | static char *zcache_compressor = ZCACHE_COMPRESSOR_DEFAULT; |
| 53 | module_param_named(compressor, zcache_compressor, charp, 0); |
| 54 | |
| 55 | /* |
| 56 | * The maximum percentage of memory that the compressed pool can occupy. |
| 57 | */ |
| 58 | static unsigned int zcache_max_pool_percent = 10; |
| 59 | module_param_named(max_pool_percent, zcache_max_pool_percent, uint, 0644); |
| 60 | |
| 61 | static unsigned int zcache_clear_percent = 4; |
| 62 | module_param_named(clear_percent, zcache_clear_percent, uint, 0644); |
| 63 | /* |
| 64 | * zcache statistics |
| 65 | */ |
| 66 | static u64 zcache_pool_limit_hit; |
| 67 | static u64 zcache_dup_entry; |
| 68 | static u64 zcache_zbud_alloc_fail; |
| 69 | static u64 zcache_evict_zpages; |
| 70 | static u64 zcache_evict_filepages; |
| 71 | static u64 zcache_inactive_pages_refused; |
| 72 | static u64 zcache_reclaim_fail; |
| 73 | static u64 zcache_pool_shrink; |
| 74 | static u64 zcache_pool_shrink_fail; |
| 75 | static u64 zcache_pool_shrink_pages; |
| 76 | static u64 zcache_store_failed; |
| 77 | static atomic_t zcache_stored_pages = ATOMIC_INIT(0); |
| 78 | static atomic_t zcache_stored_zero_pages = ATOMIC_INIT(0); |
| 79 | |
| 80 | #define GFP_ZCACHE \ |
| 81 | (__GFP_FS | __GFP_NORETRY | __GFP_NOWARN | \ |
| 82 | __GFP_NOMEMALLOC | __GFP_NO_KSWAPD | __GFP_ZERO) |
| 83 | |
| 84 | /* |
| 85 | * Make sure this is different from radix tree |
| 86 | * indirect ptr or exceptional entry. |
| 87 | */ |
| 88 | #define ZERO_HANDLE ((void *)~(~0UL >> 1)) |
| 89 | |
| 90 | /* |
| 91 | * Zcache receives pages for compression through the Cleancache API and is able |
| 92 | * to evict pages from its own compressed pool on an LRU basis in the case that |
| 93 | * the compressed pool is full. |
| 94 | * |
| 95 | * Zcache makes use of zbud for the managing the compressed memory pool. Each |
| 96 | * allocation in zbud is not directly accessible by address. Rather, a handle |
| 97 | * (zaddr) is return by the allocation routine and that handle(zaddr must be |
| 98 | * mapped before being accessed. The compressed memory pool grows on demand and |
| 99 | * shrinks as compressed pages are freed. |
| 100 | * |
| 101 | * When a file page is passed from cleancache to zcache, zcache maintains a |
| 102 | * mapping of the <filesystem_type, inode_number, page_index> to the zbud |
| 103 | * address that references that compressed file page. This mapping is achieved |
| 104 | * with a red-black tree per filesystem type, plus a radix tree per red-black |
| 105 | * node. |
| 106 | * |
| 107 | * A zcache pool with pool_id as the index is created when a filesystem mounted |
| 108 | * Each zcache pool has a red-black tree, the inode number(rb_index) is the |
| 109 | * search key. Each red-black tree node has a radix tree which use |
| 110 | * page->index(ra_index) as the index. Each radix tree slot points to the zbud |
| 111 | * address combining with some extra information(zcache_ra_handle). |
| 112 | */ |
| 113 | #define MAX_ZCACHE_POOLS 32 |
| 114 | /* |
| 115 | * One zcache_pool per (cleancache aware) filesystem mount instance |
| 116 | */ |
| 117 | struct zcache_pool { |
| 118 | struct rb_root rbtree; |
| 119 | rwlock_t rb_lock; /* Protects rbtree */ |
| 120 | u64 size; |
| 121 | struct zbud_pool *pool; /* Zbud pool used */ |
| 122 | }; |
| 123 | |
| 124 | /* |
| 125 | * Manage all zcache pools |
| 126 | */ |
| 127 | struct _zcache { |
| 128 | struct zcache_pool *pools[MAX_ZCACHE_POOLS]; |
| 129 | u32 num_pools; /* Current no. of zcache pools */ |
| 130 | spinlock_t pool_lock; /* Protects pools[] and num_pools */ |
| 131 | }; |
| 132 | struct _zcache zcache; |
| 133 | |
| 134 | /* |
| 135 | * Redblack tree node, each node has a page index radix-tree. |
| 136 | * Indexed by inode nubmer. |
| 137 | */ |
| 138 | struct zcache_rbnode { |
| 139 | struct rb_node rb_node; |
| 140 | int rb_index; |
| 141 | struct radix_tree_root ratree; /* Page radix tree per inode rbtree */ |
| 142 | spinlock_t ra_lock; /* Protects radix tree */ |
| 143 | struct kref refcount; |
| 144 | }; |
| 145 | |
| 146 | /* |
| 147 | * Radix-tree leaf, indexed by page->index |
| 148 | */ |
| 149 | struct zcache_ra_handle { |
| 150 | int rb_index; /* Redblack tree index */ |
| 151 | int ra_index; /* Radix tree index */ |
| 152 | int zlen; /* Compressed page size */ |
| 153 | struct zcache_pool *zpool; /* Finding zcache_pool during evict */ |
| 154 | }; |
| 155 | |
| 156 | u64 zcache_pages(void) |
| 157 | { |
| 158 | int i; |
| 159 | u64 count = 0; |
| 160 | |
| 161 | for (i = 0; (i < MAX_ZCACHE_POOLS) && zcache.pools[i]; i++) |
| 162 | count += zcache.pools[i]->size; |
| 163 | |
| 164 | return count; |
| 165 | } |
| 166 | |
| 167 | static struct kmem_cache *zcache_rbnode_cache; |
| 168 | static int zcache_rbnode_cache_create(void) |
| 169 | { |
| 170 | zcache_rbnode_cache = KMEM_CACHE(zcache_rbnode, 0); |
| 171 | return zcache_rbnode_cache == NULL; |
| 172 | } |
| 173 | static void zcache_rbnode_cache_destroy(void) |
| 174 | { |
| 175 | kmem_cache_destroy(zcache_rbnode_cache); |
| 176 | } |
| 177 | |
| 178 | static int zcache_shrink(struct shrinker *s, struct shrink_control *sc) |
| 179 | { |
| 180 | unsigned long active_file; |
| 181 | unsigned long file; |
| 182 | long file_gap; |
| 183 | unsigned long freed = 0; |
| 184 | unsigned long pool; |
| 185 | static bool running; |
| 186 | int i = 0; |
| 187 | int retries; |
| 188 | |
| 189 | if (running) |
| 190 | goto end; |
| 191 | |
| 192 | running = true; |
| 193 | active_file = global_page_state(NR_ACTIVE_FILE); |
| 194 | file = global_page_state(NR_FILE_PAGES); |
| 195 | pool = zcache_pages(); |
| 196 | |
| 197 | file_gap = pool - file; |
| 198 | |
| 199 | if ((file_gap >= 0) && |
| 200 | (totalram_pages * zcache_clear_percent / 100 > file)) { |
| 201 | file_gap = pool; |
| 202 | zcache_pool_shrink++; |
| 203 | goto reclaim; |
| 204 | } |
| 205 | |
| 206 | /* |
| 207 | * file_gap == 0 means that the number of pages |
| 208 | * stored by zcache is around twice as many as the |
| 209 | * number of active file pages. |
| 210 | */ |
| 211 | file_gap = pool - active_file; |
| 212 | if (file_gap < 0) |
| 213 | file_gap = 0; |
| 214 | else |
| 215 | zcache_pool_shrink++; |
| 216 | |
| 217 | reclaim: |
| 218 | retries = file_gap; |
| 219 | while ((file_gap > 0) && retries) { |
| 220 | struct zcache_pool *zpool = |
| 221 | zcache.pools[i++ % MAX_ZCACHE_POOLS]; |
| 222 | if (!zpool || !zpool->size) |
| 223 | continue; |
| 224 | if (zbud_reclaim_page(zpool->pool, 8)) { |
| 225 | zcache_pool_shrink_fail++; |
| 226 | retries--; |
| 227 | continue; |
| 228 | } |
| 229 | freed++; |
| 230 | file_gap--; |
| 231 | } |
| 232 | |
| 233 | zcache_pool_shrink_pages += freed; |
| 234 | for (i = 0; (i < MAX_ZCACHE_POOLS) && zcache.pools[i]; i++) |
| 235 | zcache.pools[i]->size = |
| 236 | zbud_get_pool_size(zcache.pools[i]->pool); |
| 237 | |
| 238 | running = false; |
| 239 | end: |
| 240 | return freed; |
| 241 | } |
| 242 | |
| 243 | static struct shrinker zcache_shrinker = { |
| 244 | .shrink = zcache_shrink, |
| 245 | .seeks = DEFAULT_SEEKS * 16 |
| 246 | }; |
| 247 | |
| 248 | /* |
| 249 | * Compression functions |
| 250 | * (Below functions are copyed from zswap!) |
| 251 | */ |
| 252 | static struct crypto_comp * __percpu *zcache_comp_pcpu_tfms; |
| 253 | |
| 254 | enum comp_op { |
| 255 | ZCACHE_COMPOP_COMPRESS, |
| 256 | ZCACHE_COMPOP_DECOMPRESS |
| 257 | }; |
| 258 | |
| 259 | static int zcache_comp_op(enum comp_op op, const u8 *src, unsigned int slen, |
| 260 | u8 *dst, unsigned int *dlen) |
| 261 | { |
| 262 | struct crypto_comp *tfm; |
| 263 | int ret; |
| 264 | |
| 265 | tfm = *per_cpu_ptr(zcache_comp_pcpu_tfms, get_cpu()); |
| 266 | switch (op) { |
| 267 | case ZCACHE_COMPOP_COMPRESS: |
| 268 | ret = crypto_comp_compress(tfm, src, slen, dst, dlen); |
| 269 | break; |
| 270 | case ZCACHE_COMPOP_DECOMPRESS: |
| 271 | ret = crypto_comp_decompress(tfm, src, slen, dst, dlen); |
| 272 | break; |
| 273 | default: |
| 274 | ret = -EINVAL; |
| 275 | } |
| 276 | |
| 277 | put_cpu(); |
| 278 | return ret; |
| 279 | } |
| 280 | |
| 281 | static int __init zcache_comp_init(void) |
| 282 | { |
| 283 | if (!crypto_has_comp(zcache_compressor, 0, 0)) { |
| 284 | pr_info("%s compressor not available\n", zcache_compressor); |
| 285 | /* fall back to default compressor */ |
| 286 | zcache_compressor = ZCACHE_COMPRESSOR_DEFAULT; |
| 287 | if (!crypto_has_comp(zcache_compressor, 0, 0)) |
| 288 | /* can't even load the default compressor */ |
| 289 | return -ENODEV; |
| 290 | } |
| 291 | pr_info("using %s compressor\n", zcache_compressor); |
| 292 | |
| 293 | /* alloc percpu transforms */ |
| 294 | zcache_comp_pcpu_tfms = alloc_percpu(struct crypto_comp *); |
| 295 | if (!zcache_comp_pcpu_tfms) |
| 296 | return -ENOMEM; |
| 297 | return 0; |
| 298 | } |
| 299 | |
| 300 | static void zcache_comp_exit(void) |
| 301 | { |
| 302 | /* free percpu transforms */ |
| 303 | if (zcache_comp_pcpu_tfms) |
| 304 | free_percpu(zcache_comp_pcpu_tfms); |
| 305 | } |
| 306 | |
| 307 | /* |
| 308 | * Per-cpu code |
| 309 | * (Below functions are also copyed from zswap!) |
| 310 | */ |
| 311 | static DEFINE_PER_CPU(u8 *, zcache_dstmem); |
| 312 | |
| 313 | static int __zcache_cpu_notifier(unsigned long action, unsigned long cpu) |
| 314 | { |
| 315 | struct crypto_comp *tfm; |
| 316 | u8 *dst; |
| 317 | |
| 318 | switch (action) { |
| 319 | case CPU_UP_PREPARE: |
| 320 | tfm = crypto_alloc_comp(zcache_compressor, 0, 0); |
| 321 | if (IS_ERR(tfm)) { |
| 322 | pr_err("can't allocate compressor transform\n"); |
| 323 | return NOTIFY_BAD; |
| 324 | } |
| 325 | *per_cpu_ptr(zcache_comp_pcpu_tfms, cpu) = tfm; |
| 326 | dst = kmalloc(PAGE_SIZE * 2, GFP_KERNEL); |
| 327 | if (!dst) { |
| 328 | pr_err("can't allocate compressor buffer\n"); |
| 329 | crypto_free_comp(tfm); |
| 330 | *per_cpu_ptr(zcache_comp_pcpu_tfms, cpu) = NULL; |
| 331 | return NOTIFY_BAD; |
| 332 | } |
| 333 | per_cpu(zcache_dstmem, cpu) = dst; |
| 334 | break; |
| 335 | case CPU_DEAD: |
| 336 | case CPU_UP_CANCELED: |
| 337 | tfm = *per_cpu_ptr(zcache_comp_pcpu_tfms, cpu); |
| 338 | if (tfm) { |
| 339 | crypto_free_comp(tfm); |
| 340 | *per_cpu_ptr(zcache_comp_pcpu_tfms, cpu) = NULL; |
| 341 | } |
| 342 | dst = per_cpu(zcache_dstmem, cpu); |
| 343 | kfree(dst); |
| 344 | per_cpu(zcache_dstmem, cpu) = NULL; |
| 345 | break; |
| 346 | default: |
| 347 | break; |
| 348 | } |
| 349 | return NOTIFY_OK; |
| 350 | } |
| 351 | |
| 352 | static int zcache_cpu_notifier(struct notifier_block *nb, |
| 353 | unsigned long action, void *pcpu) |
| 354 | { |
| 355 | unsigned long cpu = (unsigned long)pcpu; |
| 356 | |
| 357 | return __zcache_cpu_notifier(action, cpu); |
| 358 | } |
| 359 | |
| 360 | static struct notifier_block zcache_cpu_notifier_block = { |
| 361 | .notifier_call = zcache_cpu_notifier |
| 362 | }; |
| 363 | |
| 364 | static int zcache_cpu_init(void) |
| 365 | { |
| 366 | unsigned long cpu; |
| 367 | |
| 368 | get_online_cpus(); |
| 369 | for_each_online_cpu(cpu) |
| 370 | if (__zcache_cpu_notifier(CPU_UP_PREPARE, cpu) != NOTIFY_OK) |
| 371 | goto cleanup; |
| 372 | register_cpu_notifier(&zcache_cpu_notifier_block); |
| 373 | put_online_cpus(); |
| 374 | return 0; |
| 375 | |
| 376 | cleanup: |
| 377 | for_each_online_cpu(cpu) |
| 378 | __zcache_cpu_notifier(CPU_UP_CANCELED, cpu); |
| 379 | put_online_cpus(); |
| 380 | return -ENOMEM; |
| 381 | } |
| 382 | |
| 383 | /* |
| 384 | * Zcache helpers |
| 385 | */ |
| 386 | static bool zcache_is_full(void) |
| 387 | { |
| 388 | long file = global_page_state(NR_FILE_PAGES); |
| 389 | |
| 390 | return ((totalram_pages * zcache_max_pool_percent / 100 < |
| 391 | zcache_pages()) || |
| 392 | (totalram_pages * zcache_clear_percent / 100 > |
| 393 | file)); |
| 394 | } |
| 395 | |
| 396 | /* |
| 397 | * The caller must hold zpool->rb_lock at least |
| 398 | */ |
| 399 | static struct zcache_rbnode *zcache_find_rbnode(struct rb_root *rbtree, |
| 400 | int index, struct rb_node **rb_parent, struct rb_node ***rb_link) |
| 401 | { |
| 402 | struct zcache_rbnode *entry; |
| 403 | struct rb_node **__rb_link, *__rb_parent, *rb_prev; |
| 404 | |
| 405 | __rb_link = &rbtree->rb_node; |
| 406 | rb_prev = __rb_parent = NULL; |
| 407 | |
| 408 | while (*__rb_link) { |
| 409 | __rb_parent = *__rb_link; |
| 410 | entry = rb_entry(__rb_parent, struct zcache_rbnode, rb_node); |
| 411 | if (entry->rb_index > index) |
| 412 | __rb_link = &__rb_parent->rb_left; |
| 413 | else if (entry->rb_index < index) { |
| 414 | rb_prev = __rb_parent; |
| 415 | __rb_link = &__rb_parent->rb_right; |
| 416 | } else |
| 417 | return entry; |
| 418 | } |
| 419 | |
| 420 | if (rb_parent) |
| 421 | *rb_parent = __rb_parent; |
| 422 | if (rb_link) |
| 423 | *rb_link = __rb_link; |
| 424 | return NULL; |
| 425 | } |
| 426 | |
| 427 | static struct zcache_rbnode *zcache_find_get_rbnode(struct zcache_pool *zpool, |
| 428 | int rb_index) |
| 429 | { |
| 430 | unsigned long flags; |
| 431 | struct zcache_rbnode *rbnode; |
| 432 | |
| 433 | read_lock_irqsave(&zpool->rb_lock, flags); |
| 434 | rbnode = zcache_find_rbnode(&zpool->rbtree, rb_index, 0, 0); |
| 435 | if (rbnode) |
| 436 | kref_get(&rbnode->refcount); |
| 437 | read_unlock_irqrestore(&zpool->rb_lock, flags); |
| 438 | return rbnode; |
| 439 | } |
| 440 | |
| 441 | /* |
| 442 | * kref_put callback for zcache_rbnode. |
| 443 | * |
| 444 | * The rbnode must have been isolated from rbtree already. |
| 445 | */ |
| 446 | static void zcache_rbnode_release(struct kref *kref) |
| 447 | { |
| 448 | struct zcache_rbnode *rbnode; |
| 449 | |
| 450 | rbnode = container_of(kref, struct zcache_rbnode, refcount); |
| 451 | BUG_ON(rbnode->ratree.rnode); |
| 452 | kmem_cache_free(zcache_rbnode_cache, rbnode); |
| 453 | } |
| 454 | |
| 455 | /* |
| 456 | * Check whether the radix-tree of this rbnode is empty. |
| 457 | * If that's true, then we can delete this zcache_rbnode from |
| 458 | * zcache_pool->rbtree |
| 459 | * |
| 460 | * Caller must hold zcache_rbnode->ra_lock |
| 461 | */ |
| 462 | static int zcache_rbnode_empty(struct zcache_rbnode *rbnode) |
| 463 | { |
| 464 | return rbnode->ratree.rnode == NULL; |
| 465 | } |
| 466 | |
| 467 | /* |
| 468 | * Remove zcache_rbnode from zpool->rbtree |
| 469 | * |
| 470 | * holded_rblock - whether the caller has holded zpool->rb_lock |
| 471 | */ |
| 472 | static void zcache_rbnode_isolate(struct zcache_pool *zpool, |
| 473 | struct zcache_rbnode *rbnode, bool holded_rblock) |
| 474 | { |
| 475 | unsigned long flags; |
| 476 | |
| 477 | if (!holded_rblock) |
| 478 | write_lock_irqsave(&zpool->rb_lock, flags); |
| 479 | /* |
| 480 | * Someone can get reference on this rbnode before we could |
| 481 | * acquire write lock above. |
| 482 | * We want to remove it from zpool->rbtree when only the caller and |
| 483 | * corresponding ratree holds a reference to this rbnode. |
| 484 | * Below check ensures that a racing zcache put will not end up adding |
| 485 | * a page to an isolated node and thereby losing that memory. |
| 486 | */ |
| 487 | if (atomic_read(&rbnode->refcount.refcount) == 2) { |
| 488 | rb_erase(&rbnode->rb_node, &zpool->rbtree); |
| 489 | RB_CLEAR_NODE(&rbnode->rb_node); |
| 490 | kref_put(&rbnode->refcount, zcache_rbnode_release); |
| 491 | } |
| 492 | if (!holded_rblock) |
| 493 | write_unlock_irqrestore(&zpool->rb_lock, flags); |
| 494 | } |
| 495 | |
| 496 | /* |
| 497 | * Store zaddr which allocated by zbud_alloc() to the hierarchy rbtree-ratree. |
| 498 | */ |
| 499 | static int zcache_store_zaddr(struct zcache_pool *zpool, |
| 500 | int ra_index, int rb_index, unsigned long zaddr) |
| 501 | { |
| 502 | unsigned long flags; |
| 503 | struct zcache_rbnode *rbnode, *tmp; |
| 504 | struct rb_node **link = NULL, *parent = NULL; |
| 505 | int ret; |
| 506 | void *dup_zaddr; |
| 507 | |
| 508 | rbnode = zcache_find_get_rbnode(zpool, rb_index); |
| 509 | if (!rbnode) { |
| 510 | /* alloc and init a new rbnode */ |
| 511 | rbnode = kmem_cache_alloc(zcache_rbnode_cache, |
| 512 | GFP_ZCACHE); |
| 513 | if (!rbnode) |
| 514 | return -ENOMEM; |
| 515 | |
| 516 | INIT_RADIX_TREE(&rbnode->ratree, GFP_ATOMIC|__GFP_NOWARN); |
| 517 | spin_lock_init(&rbnode->ra_lock); |
| 518 | rbnode->rb_index = rb_index; |
| 519 | kref_init(&rbnode->refcount); |
| 520 | RB_CLEAR_NODE(&rbnode->rb_node); |
| 521 | |
| 522 | /* add that rbnode to rbtree */ |
| 523 | write_lock_irqsave(&zpool->rb_lock, flags); |
| 524 | tmp = zcache_find_rbnode(&zpool->rbtree, rb_index, |
| 525 | &parent, &link); |
| 526 | if (tmp) { |
| 527 | /* somebody else allocated new rbnode */ |
| 528 | kmem_cache_free(zcache_rbnode_cache, rbnode); |
| 529 | rbnode = tmp; |
| 530 | } else { |
| 531 | rb_link_node(&rbnode->rb_node, parent, link); |
| 532 | rb_insert_color(&rbnode->rb_node, &zpool->rbtree); |
| 533 | } |
| 534 | |
| 535 | /* Inc the reference of this zcache_rbnode */ |
| 536 | kref_get(&rbnode->refcount); |
| 537 | write_unlock_irqrestore(&zpool->rb_lock, flags); |
| 538 | } |
| 539 | |
| 540 | /* Succfully got a zcache_rbnode when arriving here */ |
| 541 | spin_lock_irqsave(&rbnode->ra_lock, flags); |
| 542 | dup_zaddr = radix_tree_delete(&rbnode->ratree, ra_index); |
| 543 | if (unlikely(dup_zaddr)) { |
Steve Kondik | f94910d | 2016-04-14 02:52:21 -0700 | [diff] [blame] | 544 | if (dup_zaddr == ZERO_HANDLE) { |
| 545 | atomic_dec(&zcache_stored_zero_pages); |
| 546 | } else { |
| 547 | zbud_free(zpool->pool, (unsigned long)dup_zaddr); |
| 548 | atomic_dec(&zcache_stored_pages); |
| 549 | zpool->size = zbud_get_pool_size(zpool->pool); |
| 550 | } |
| 551 | zcache_dup_entry++; |
| 552 | } |
| 553 | |
| 554 | /* Insert zcache_ra_handle to ratree */ |
| 555 | ret = radix_tree_insert(&rbnode->ratree, ra_index, |
| 556 | (void *)zaddr); |
| 557 | spin_unlock_irqrestore(&rbnode->ra_lock, flags); |
| 558 | if (unlikely(ret)) { |
| 559 | write_lock_irqsave(&zpool->rb_lock, flags); |
| 560 | spin_lock(&rbnode->ra_lock); |
| 561 | |
| 562 | if (zcache_rbnode_empty(rbnode)) |
| 563 | zcache_rbnode_isolate(zpool, rbnode, 1); |
| 564 | |
| 565 | spin_unlock(&rbnode->ra_lock); |
| 566 | write_unlock_irqrestore(&zpool->rb_lock, flags); |
| 567 | } |
| 568 | |
| 569 | kref_put(&rbnode->refcount, zcache_rbnode_release); |
| 570 | return ret; |
| 571 | } |
| 572 | |
| 573 | /* |
| 574 | * Load zaddr and delete it from radix tree. |
| 575 | * If the radix tree of the corresponding rbnode is empty, delete the rbnode |
| 576 | * from zpool->rbtree also. |
| 577 | */ |
| 578 | static void *zcache_load_delete_zaddr(struct zcache_pool *zpool, |
| 579 | int rb_index, int ra_index) |
| 580 | { |
| 581 | struct zcache_rbnode *rbnode; |
| 582 | void *zaddr = NULL; |
| 583 | unsigned long flags; |
| 584 | |
| 585 | rbnode = zcache_find_get_rbnode(zpool, rb_index); |
| 586 | if (!rbnode) |
| 587 | goto out; |
| 588 | |
| 589 | BUG_ON(rbnode->rb_index != rb_index); |
| 590 | |
| 591 | spin_lock_irqsave(&rbnode->ra_lock, flags); |
| 592 | zaddr = radix_tree_delete(&rbnode->ratree, ra_index); |
| 593 | spin_unlock_irqrestore(&rbnode->ra_lock, flags); |
| 594 | |
| 595 | /* rb_lock and ra_lock must be taken again in the given sequence */ |
| 596 | write_lock_irqsave(&zpool->rb_lock, flags); |
| 597 | spin_lock(&rbnode->ra_lock); |
| 598 | if (zcache_rbnode_empty(rbnode)) |
| 599 | zcache_rbnode_isolate(zpool, rbnode, 1); |
| 600 | spin_unlock(&rbnode->ra_lock); |
| 601 | write_unlock_irqrestore(&zpool->rb_lock, flags); |
| 602 | |
| 603 | kref_put(&rbnode->refcount, zcache_rbnode_release); |
| 604 | out: |
| 605 | return zaddr; |
| 606 | } |
| 607 | |
| 608 | static bool zero_page(struct page *page) |
| 609 | { |
| 610 | unsigned long *ptr = kmap_atomic(page); |
| 611 | int i; |
| 612 | bool ret = false; |
| 613 | |
| 614 | for (i = 0; i < PAGE_SIZE / sizeof(*ptr); i++) { |
| 615 | if (ptr[i]) |
| 616 | goto out; |
| 617 | } |
| 618 | ret = true; |
| 619 | out: |
| 620 | kunmap_atomic(ptr); |
| 621 | return ret; |
| 622 | } |
| 623 | |
| 624 | static void zcache_store_page(int pool_id, struct cleancache_filekey key, |
| 625 | pgoff_t index, struct page *page) |
| 626 | { |
| 627 | struct zcache_ra_handle *zhandle; |
| 628 | u8 *zpage, *src, *dst; |
| 629 | /* Address of zhandle + compressed data(zpage) */ |
| 630 | unsigned long zaddr = 0; |
| 631 | unsigned int zlen = PAGE_SIZE; |
| 632 | bool zero = 0; |
| 633 | int ret; |
| 634 | |
| 635 | struct zcache_pool *zpool = zcache.pools[pool_id]; |
| 636 | |
| 637 | /* |
| 638 | * Zcache will be ineffective if the compressed memory pool is full with |
| 639 | * compressed inactive file pages and most of them will never be used |
| 640 | * again. |
| 641 | * So we refuse to compress pages that are not from active file list. |
| 642 | */ |
| 643 | if (!PageWasActive(page)) { |
| 644 | zcache_inactive_pages_refused++; |
| 645 | return; |
| 646 | } |
| 647 | |
| 648 | zero = zero_page(page); |
| 649 | if (zero) |
| 650 | goto zero; |
| 651 | |
| 652 | if (zcache_is_full()) { |
| 653 | zcache_pool_limit_hit++; |
| 654 | if (zbud_reclaim_page(zpool->pool, 8)) { |
| 655 | zcache_reclaim_fail++; |
| 656 | return; |
| 657 | } |
| 658 | /* |
| 659 | * Continue if reclaimed a page frame succ. |
| 660 | */ |
| 661 | zcache_evict_filepages++; |
| 662 | zpool->size = zbud_get_pool_size(zpool->pool); |
| 663 | } |
| 664 | |
| 665 | /* compress */ |
| 666 | dst = get_cpu_var(zcache_dstmem); |
| 667 | src = kmap_atomic(page); |
| 668 | ret = zcache_comp_op(ZCACHE_COMPOP_COMPRESS, src, PAGE_SIZE, dst, |
| 669 | &zlen); |
| 670 | kunmap_atomic(src); |
| 671 | if (ret) { |
| 672 | pr_err("zcache compress error ret %d\n", ret); |
| 673 | put_cpu_var(zcache_dstmem); |
| 674 | return; |
| 675 | } |
| 676 | |
| 677 | /* store zcache handle together with compressed page data */ |
| 678 | ret = zbud_alloc(zpool->pool, zlen + sizeof(struct zcache_ra_handle), |
| 679 | GFP_ZCACHE, &zaddr); |
| 680 | if (ret) { |
| 681 | zcache_zbud_alloc_fail++; |
| 682 | put_cpu_var(zcache_dstmem); |
| 683 | return; |
| 684 | } |
| 685 | |
| 686 | zhandle = (struct zcache_ra_handle *)zbud_map(zpool->pool, zaddr); |
| 687 | |
| 688 | /* Compressed page data stored at the end of zcache_ra_handle */ |
| 689 | zpage = (u8 *)(zhandle + 1); |
| 690 | memcpy(zpage, dst, zlen); |
| 691 | zbud_unmap(zpool->pool, zaddr); |
| 692 | put_cpu_var(zcache_dstmem); |
| 693 | |
| 694 | zero: |
| 695 | if (zero) |
| 696 | zaddr = (unsigned long)ZERO_HANDLE; |
| 697 | |
| 698 | /* store zcache handle */ |
| 699 | ret = zcache_store_zaddr(zpool, index, key.u.ino, zaddr); |
| 700 | if (ret) { |
| 701 | zcache_store_failed++; |
| 702 | if (!zero) |
| 703 | zbud_free(zpool->pool, zaddr); |
| 704 | return; |
| 705 | } |
| 706 | |
| 707 | /* update stats */ |
| 708 | if (zero) { |
| 709 | atomic_inc(&zcache_stored_zero_pages); |
| 710 | } else { |
| 711 | zhandle->ra_index = index; |
| 712 | zhandle->rb_index = key.u.ino; |
| 713 | zhandle->zlen = zlen; |
| 714 | zhandle->zpool = zpool; |
| 715 | atomic_inc(&zcache_stored_pages); |
| 716 | zpool->size = zbud_get_pool_size(zpool->pool); |
| 717 | } |
| 718 | |
| 719 | return; |
| 720 | } |
| 721 | |
| 722 | static int zcache_load_page(int pool_id, struct cleancache_filekey key, |
| 723 | pgoff_t index, struct page *page) |
| 724 | { |
| 725 | int ret = 0; |
| 726 | u8 *src, *dst; |
| 727 | void *zaddr; |
| 728 | unsigned int dlen = PAGE_SIZE; |
| 729 | struct zcache_ra_handle *zhandle; |
| 730 | struct zcache_pool *zpool = zcache.pools[pool_id]; |
| 731 | |
| 732 | zaddr = zcache_load_delete_zaddr(zpool, key.u.ino, index); |
| 733 | if (!zaddr) |
| 734 | return -ENOENT; |
| 735 | else if (zaddr == ZERO_HANDLE) |
| 736 | goto map; |
| 737 | |
| 738 | zhandle = (struct zcache_ra_handle *)zbud_map(zpool->pool, |
| 739 | (unsigned long)zaddr); |
| 740 | /* Compressed page data stored at the end of zcache_ra_handle */ |
| 741 | src = (u8 *)(zhandle + 1); |
| 742 | |
| 743 | /* decompress */ |
| 744 | map: |
| 745 | dst = kmap_atomic(page); |
| 746 | if (zaddr != ZERO_HANDLE) { |
| 747 | ret = zcache_comp_op(ZCACHE_COMPOP_DECOMPRESS, src, |
| 748 | zhandle->zlen, dst, &dlen); |
| 749 | } else { |
| 750 | memset(dst, 0, PAGE_SIZE); |
| 751 | kunmap_atomic(dst); |
| 752 | flush_dcache_page(page); |
| 753 | atomic_dec(&zcache_stored_zero_pages); |
| 754 | goto out; |
| 755 | } |
| 756 | kunmap_atomic(dst); |
| 757 | zbud_unmap(zpool->pool, (unsigned long)zaddr); |
| 758 | zbud_free(zpool->pool, (unsigned long)zaddr); |
| 759 | |
| 760 | BUG_ON(ret); |
| 761 | BUG_ON(dlen != PAGE_SIZE); |
| 762 | |
| 763 | /* update stats */ |
| 764 | atomic_dec(&zcache_stored_pages); |
| 765 | zpool->size = zbud_get_pool_size(zpool->pool); |
| 766 | out: |
| 767 | SetPageWasActive(page); |
| 768 | return ret; |
| 769 | } |
| 770 | |
| 771 | static void zcache_flush_page(int pool_id, struct cleancache_filekey key, |
| 772 | pgoff_t index) |
| 773 | { |
| 774 | struct zcache_pool *zpool = zcache.pools[pool_id]; |
| 775 | void *zaddr = NULL; |
| 776 | |
| 777 | zaddr = zcache_load_delete_zaddr(zpool, key.u.ino, index); |
| 778 | if (zaddr && (zaddr != ZERO_HANDLE)) { |
| 779 | zbud_free(zpool->pool, (unsigned long)zaddr); |
| 780 | atomic_dec(&zcache_stored_pages); |
| 781 | zpool->size = zbud_get_pool_size(zpool->pool); |
| 782 | } else if (zaddr == ZERO_HANDLE) { |
| 783 | atomic_dec(&zcache_stored_zero_pages); |
| 784 | } |
| 785 | } |
| 786 | |
| 787 | #define FREE_BATCH 16 |
| 788 | /* |
| 789 | * Callers must hold the lock |
| 790 | */ |
| 791 | static void zcache_flush_ratree(struct zcache_pool *zpool, |
| 792 | struct zcache_rbnode *rbnode) |
| 793 | { |
| 794 | unsigned long index = 0; |
| 795 | int count, i; |
| 796 | struct zcache_ra_handle *zhandle; |
| 797 | void *zaddr = NULL; |
| 798 | |
| 799 | do { |
| 800 | void *zaddrs[FREE_BATCH]; |
| 801 | unsigned long indices[FREE_BATCH]; |
| 802 | |
| 803 | count = radix_tree_gang_lookup_index(&rbnode->ratree, |
| 804 | (void **)zaddrs, indices, |
| 805 | index, FREE_BATCH); |
| 806 | |
| 807 | for (i = 0; i < count; i++) { |
| 808 | if (zaddrs[i] == ZERO_HANDLE) { |
| 809 | zaddr = radix_tree_delete(&rbnode->ratree, |
| 810 | indices[i]); |
| 811 | if (zaddr) |
| 812 | atomic_dec(&zcache_stored_zero_pages); |
| 813 | continue; |
| 814 | } |
| 815 | zhandle = (struct zcache_ra_handle *)zbud_map( |
| 816 | zpool->pool, (unsigned long)zaddrs[i]); |
| 817 | index = zhandle->ra_index; |
| 818 | zaddr = radix_tree_delete(&rbnode->ratree, index); |
| 819 | if (!zaddr) |
| 820 | continue; |
| 821 | zbud_unmap(zpool->pool, (unsigned long)zaddrs[i]); |
| 822 | zbud_free(zpool->pool, (unsigned long)zaddrs[i]); |
| 823 | atomic_dec(&zcache_stored_pages); |
| 824 | zpool->size = zbud_get_pool_size(zpool->pool); |
| 825 | } |
| 826 | |
| 827 | index++; |
| 828 | } while (count == FREE_BATCH); |
| 829 | } |
| 830 | |
| 831 | static void zcache_flush_inode(int pool_id, struct cleancache_filekey key) |
| 832 | { |
| 833 | struct zcache_rbnode *rbnode; |
| 834 | unsigned long flags1, flags2; |
| 835 | struct zcache_pool *zpool = zcache.pools[pool_id]; |
| 836 | |
| 837 | /* |
| 838 | * Refuse new pages added in to the same rbinode, so get rb_lock at |
| 839 | * first. |
| 840 | */ |
| 841 | write_lock_irqsave(&zpool->rb_lock, flags1); |
| 842 | rbnode = zcache_find_rbnode(&zpool->rbtree, key.u.ino, 0, 0); |
| 843 | if (!rbnode) { |
| 844 | write_unlock_irqrestore(&zpool->rb_lock, flags1); |
| 845 | return; |
| 846 | } |
| 847 | |
| 848 | kref_get(&rbnode->refcount); |
| 849 | spin_lock_irqsave(&rbnode->ra_lock, flags2); |
| 850 | |
| 851 | zcache_flush_ratree(zpool, rbnode); |
| 852 | if (zcache_rbnode_empty(rbnode)) |
| 853 | /* When arrvied here, we already hold rb_lock */ |
| 854 | zcache_rbnode_isolate(zpool, rbnode, 1); |
| 855 | |
| 856 | spin_unlock_irqrestore(&rbnode->ra_lock, flags2); |
| 857 | write_unlock_irqrestore(&zpool->rb_lock, flags1); |
| 858 | kref_put(&rbnode->refcount, zcache_rbnode_release); |
| 859 | } |
| 860 | |
| 861 | static void zcache_destroy_pool(struct zcache_pool *zpool); |
| 862 | static void zcache_flush_fs(int pool_id) |
| 863 | { |
| 864 | struct zcache_rbnode *z_rbnode = NULL; |
| 865 | struct rb_node *rbnode; |
| 866 | unsigned long flags1, flags2; |
| 867 | struct zcache_pool *zpool; |
| 868 | |
| 869 | if (pool_id < 0) |
| 870 | return; |
| 871 | |
| 872 | zpool = zcache.pools[pool_id]; |
| 873 | if (!zpool) |
| 874 | return; |
| 875 | |
| 876 | /* |
| 877 | * Refuse new pages added in, so get rb_lock at first. |
| 878 | */ |
| 879 | write_lock_irqsave(&zpool->rb_lock, flags1); |
| 880 | |
| 881 | rbnode = rb_first(&zpool->rbtree); |
| 882 | while (rbnode) { |
| 883 | z_rbnode = rb_entry(rbnode, struct zcache_rbnode, rb_node); |
| 884 | rbnode = rb_next(rbnode); |
| 885 | if (z_rbnode) { |
| 886 | kref_get(&z_rbnode->refcount); |
| 887 | spin_lock_irqsave(&z_rbnode->ra_lock, flags2); |
| 888 | zcache_flush_ratree(zpool, z_rbnode); |
| 889 | if (zcache_rbnode_empty(z_rbnode)) |
| 890 | zcache_rbnode_isolate(zpool, z_rbnode, 1); |
| 891 | spin_unlock_irqrestore(&z_rbnode->ra_lock, flags2); |
| 892 | kref_put(&z_rbnode->refcount, zcache_rbnode_release); |
| 893 | } |
| 894 | } |
| 895 | |
| 896 | write_unlock_irqrestore(&zpool->rb_lock, flags1); |
| 897 | zcache_destroy_pool(zpool); |
| 898 | } |
| 899 | |
| 900 | /* |
| 901 | * Evict compressed pages from zcache pool on an LRU basis after the compressed |
| 902 | * pool is full. |
| 903 | */ |
| 904 | static int zcache_evict_zpage(struct zbud_pool *pool, unsigned long zaddr) |
| 905 | { |
| 906 | struct zcache_pool *zpool; |
| 907 | struct zcache_ra_handle *zhandle; |
| 908 | void *zaddr_intree; |
| 909 | |
| 910 | BUG_ON(zaddr == (unsigned long)ZERO_HANDLE); |
| 911 | |
| 912 | zhandle = (struct zcache_ra_handle *)zbud_map(pool, zaddr); |
| 913 | |
| 914 | zpool = zhandle->zpool; |
| 915 | /* There can be a race with zcache store */ |
| 916 | if (!zpool) |
| 917 | return -EINVAL; |
| 918 | |
| 919 | BUG_ON(pool != zpool->pool); |
| 920 | |
| 921 | zaddr_intree = zcache_load_delete_zaddr(zpool, zhandle->rb_index, |
| 922 | zhandle->ra_index); |
| 923 | if (zaddr_intree) { |
| 924 | BUG_ON((unsigned long)zaddr_intree != zaddr); |
| 925 | zbud_unmap(pool, zaddr); |
| 926 | zbud_free(pool, zaddr); |
| 927 | atomic_dec(&zcache_stored_pages); |
| 928 | zpool->size = zbud_get_pool_size(pool); |
| 929 | zcache_evict_zpages++; |
| 930 | } |
| 931 | return 0; |
| 932 | } |
| 933 | |
| 934 | static struct zbud_ops zcache_zbud_ops = { |
| 935 | .evict = zcache_evict_zpage |
| 936 | }; |
| 937 | |
| 938 | /* Return pool id */ |
| 939 | static int zcache_create_pool(void) |
| 940 | { |
| 941 | int ret; |
| 942 | struct zcache_pool *zpool; |
| 943 | |
| 944 | zpool = kzalloc(sizeof(*zpool), GFP_KERNEL); |
| 945 | if (!zpool) { |
| 946 | ret = -ENOMEM; |
| 947 | goto out; |
| 948 | } |
| 949 | |
| 950 | zpool->pool = zbud_create_pool(GFP_KERNEL, &zcache_zbud_ops); |
| 951 | if (!zpool->pool) { |
| 952 | kfree(zpool); |
| 953 | ret = -ENOMEM; |
| 954 | goto out; |
| 955 | } |
| 956 | |
| 957 | spin_lock(&zcache.pool_lock); |
| 958 | if (zcache.num_pools == MAX_ZCACHE_POOLS) { |
| 959 | pr_err("Cannot create new pool (limit:%u)\n", MAX_ZCACHE_POOLS); |
| 960 | zbud_destroy_pool(zpool->pool); |
| 961 | kfree(zpool); |
| 962 | ret = -EPERM; |
| 963 | goto out_unlock; |
| 964 | } |
| 965 | |
| 966 | rwlock_init(&zpool->rb_lock); |
| 967 | zpool->rbtree = RB_ROOT; |
| 968 | /* Add to pool list */ |
| 969 | for (ret = 0; ret < MAX_ZCACHE_POOLS; ret++) |
| 970 | if (!zcache.pools[ret]) |
| 971 | break; |
| 972 | zcache.pools[ret] = zpool; |
| 973 | zcache.num_pools++; |
| 974 | pr_info("New pool created id:%d\n", ret); |
| 975 | |
| 976 | out_unlock: |
| 977 | spin_unlock(&zcache.pool_lock); |
| 978 | out: |
| 979 | return ret; |
| 980 | } |
| 981 | |
| 982 | static void zcache_destroy_pool(struct zcache_pool *zpool) |
| 983 | { |
| 984 | int i; |
| 985 | |
| 986 | if (!zpool) |
| 987 | return; |
| 988 | |
| 989 | spin_lock(&zcache.pool_lock); |
| 990 | zcache.num_pools--; |
| 991 | for (i = 0; i < MAX_ZCACHE_POOLS; i++) |
| 992 | if (zcache.pools[i] == zpool) |
| 993 | break; |
| 994 | zcache.pools[i] = NULL; |
| 995 | spin_unlock(&zcache.pool_lock); |
| 996 | |
| 997 | if (!RB_EMPTY_ROOT(&zpool->rbtree)) |
| 998 | WARN_ON("Memory leak detected. Freeing non-empty pool!\n"); |
| 999 | |
| 1000 | zbud_destroy_pool(zpool->pool); |
| 1001 | kfree(zpool); |
| 1002 | } |
| 1003 | |
| 1004 | static int zcache_init_fs(size_t pagesize) |
| 1005 | { |
| 1006 | int ret; |
| 1007 | |
| 1008 | if (pagesize != PAGE_SIZE) { |
| 1009 | pr_info("Unsupported page size: %zu", pagesize); |
| 1010 | ret = -EINVAL; |
| 1011 | goto out; |
| 1012 | } |
| 1013 | |
| 1014 | ret = zcache_create_pool(); |
| 1015 | if (ret < 0) { |
| 1016 | pr_info("Failed to create new pool\n"); |
| 1017 | ret = -ENOMEM; |
| 1018 | goto out; |
| 1019 | } |
| 1020 | out: |
| 1021 | return ret; |
| 1022 | } |
| 1023 | |
| 1024 | static int zcache_init_shared_fs(char *uuid, size_t pagesize) |
| 1025 | { |
| 1026 | /* shared pools are unsupported and map to private */ |
| 1027 | return zcache_init_fs(pagesize); |
| 1028 | } |
| 1029 | |
| 1030 | static struct cleancache_ops zcache_ops = { |
| 1031 | .put_page = zcache_store_page, |
| 1032 | .get_page = zcache_load_page, |
| 1033 | .invalidate_page = zcache_flush_page, |
| 1034 | .invalidate_inode = zcache_flush_inode, |
| 1035 | .invalidate_fs = zcache_flush_fs, |
| 1036 | .init_shared_fs = zcache_init_shared_fs, |
| 1037 | .init_fs = zcache_init_fs |
| 1038 | }; |
| 1039 | |
| 1040 | /* |
| 1041 | * Debugfs functions |
| 1042 | */ |
| 1043 | #ifdef CONFIG_DEBUG_FS |
| 1044 | #include <linux/debugfs.h> |
| 1045 | |
| 1046 | static int pool_pages_get(void *_data, u64 *val) |
| 1047 | { |
| 1048 | *val = zcache_pages(); |
| 1049 | return 0; |
| 1050 | } |
| 1051 | |
| 1052 | DEFINE_SIMPLE_ATTRIBUTE(pool_page_fops, pool_pages_get, NULL, "%llu\n"); |
| 1053 | |
| 1054 | static struct dentry *zcache_debugfs_root; |
| 1055 | |
| 1056 | static int __init zcache_debugfs_init(void) |
| 1057 | { |
| 1058 | if (!debugfs_initialized()) |
| 1059 | return -ENODEV; |
| 1060 | |
| 1061 | zcache_debugfs_root = debugfs_create_dir("zcache", NULL); |
| 1062 | if (!zcache_debugfs_root) |
| 1063 | return -ENOMEM; |
| 1064 | |
| 1065 | debugfs_create_u64("pool_limit_hit", S_IRUGO, zcache_debugfs_root, |
| 1066 | &zcache_pool_limit_hit); |
| 1067 | debugfs_create_u64("reject_alloc_fail", S_IRUGO, zcache_debugfs_root, |
| 1068 | &zcache_zbud_alloc_fail); |
| 1069 | debugfs_create_u64("duplicate_entry", S_IRUGO, zcache_debugfs_root, |
| 1070 | &zcache_dup_entry); |
| 1071 | debugfs_create_file("pool_pages", S_IRUGO, zcache_debugfs_root, NULL, |
| 1072 | &pool_page_fops); |
| 1073 | debugfs_create_atomic_t("stored_pages", S_IRUGO, zcache_debugfs_root, |
| 1074 | &zcache_stored_pages); |
| 1075 | debugfs_create_atomic_t("stored_zero_pages", S_IRUGO, |
| 1076 | zcache_debugfs_root, &zcache_stored_zero_pages); |
| 1077 | debugfs_create_u64("evicted_zpages", S_IRUGO, zcache_debugfs_root, |
| 1078 | &zcache_evict_zpages); |
| 1079 | debugfs_create_u64("evicted_filepages", S_IRUGO, zcache_debugfs_root, |
| 1080 | &zcache_evict_filepages); |
| 1081 | debugfs_create_u64("reclaim_fail", S_IRUGO, zcache_debugfs_root, |
| 1082 | &zcache_reclaim_fail); |
| 1083 | debugfs_create_u64("inactive_pages_refused", S_IRUGO, |
| 1084 | zcache_debugfs_root, &zcache_inactive_pages_refused); |
| 1085 | debugfs_create_u64("pool_shrink_count", S_IRUGO, |
| 1086 | zcache_debugfs_root, &zcache_pool_shrink); |
| 1087 | debugfs_create_u64("pool_shrink_fail", S_IRUGO, |
| 1088 | zcache_debugfs_root, &zcache_pool_shrink_fail); |
| 1089 | debugfs_create_u64("pool_shrink_pages", S_IRUGO, |
| 1090 | zcache_debugfs_root, &zcache_pool_shrink_pages); |
| 1091 | debugfs_create_u64("store_fail", S_IRUGO, |
| 1092 | zcache_debugfs_root, &zcache_store_failed); |
| 1093 | return 0; |
| 1094 | } |
| 1095 | |
| 1096 | static void __exit zcache_debugfs_exit(void) |
| 1097 | { |
| 1098 | debugfs_remove_recursive(zcache_debugfs_root); |
| 1099 | } |
| 1100 | #else |
| 1101 | static int __init zcache_debugfs_init(void) |
| 1102 | { |
| 1103 | return 0; |
| 1104 | } |
| 1105 | static void __exit zcache_debugfs_exit(void) |
| 1106 | { |
| 1107 | } |
| 1108 | #endif |
| 1109 | |
| 1110 | /* |
| 1111 | * zcache init and exit |
| 1112 | */ |
| 1113 | static int __init init_zcache(void) |
| 1114 | { |
| 1115 | if (!zcache_enabled) |
| 1116 | return 0; |
| 1117 | |
| 1118 | pr_info("loading zcache..\n"); |
| 1119 | if (zcache_rbnode_cache_create()) { |
| 1120 | pr_err("entry cache creation failed\n"); |
| 1121 | goto error; |
| 1122 | } |
| 1123 | |
| 1124 | if (zcache_comp_init()) { |
| 1125 | pr_err("compressor initialization failed\n"); |
| 1126 | goto compfail; |
| 1127 | } |
| 1128 | if (zcache_cpu_init()) { |
| 1129 | pr_err("per-cpu initialization failed\n"); |
| 1130 | goto pcpufail; |
| 1131 | } |
| 1132 | |
| 1133 | spin_lock_init(&zcache.pool_lock); |
| 1134 | cleancache_register_ops(&zcache_ops); |
| 1135 | |
| 1136 | if (zcache_debugfs_init()) |
| 1137 | pr_warn("debugfs initialization failed\n"); |
| 1138 | register_shrinker(&zcache_shrinker); |
| 1139 | return 0; |
| 1140 | pcpufail: |
| 1141 | zcache_comp_exit(); |
| 1142 | compfail: |
| 1143 | zcache_rbnode_cache_destroy(); |
| 1144 | error: |
| 1145 | return -ENOMEM; |
| 1146 | } |
| 1147 | |
| 1148 | /* must be late so crypto has time to come up */ |
| 1149 | late_initcall(init_zcache); |
| 1150 | |
| 1151 | MODULE_LICENSE("GPL"); |
| 1152 | MODULE_AUTHOR("Bob Liu <bob.liu@xxxxxxxxxx>"); |
| 1153 | MODULE_DESCRIPTION("Compressed cache for clean file pages"); |
| 1154 | |