1 // SPDX-License-Identifier: CDDL-1.0 2 /* 3 * This file and its contents are supplied under the terms of the 4 * Common Development and Distribution License ("CDDL"), version 1.0. 5 * You may only use this file in accordance with the terms of version 6 * 1.0 of the CDDL. 7 * 8 * A full copy of the text of the CDDL should have accompanied this 9 * source. A copy of the CDDL is also available via the Internet at 10 * https://opensource.org/license/CDDL-1.0. 11 */ 12 /* 13 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. 14 * Copyright (c) 2012, 2016 by Delphix. All rights reserved. 15 * Copyright (c) 2013 by Saso Kiselkov. All rights reserved. 16 * Copyright (c) 2019, Allan Jude 17 * Copyright (c) 2019, Klara Inc. 18 */ 19 20 #ifndef _SYS_ARC_H 21 #define _SYS_ARC_H 22 23 #include <sys/zfs_context.h> 24 25 #ifdef __cplusplus 26 extern "C" { 27 #endif 28 29 #include <sys/zio.h> 30 #include <sys/dmu.h> 31 #include <sys/spa.h> 32 #include <sys/zfs_refcount.h> 33 34 /* 35 * Used by arc_flush() to inform arc_evict_state() that it should evict 36 * all available buffers from the arc state being passed in. 37 */ 38 #define ARC_EVICT_ALL UINT64_MAX 39 40 /* 41 * ZFS gets very unhappy when the maximum ARC size is smaller than the maximum 42 * block size and a larger block is written. To leave some safety margin, we 43 * limit the minimum for zfs_arc_max to the maximium transaction size. 44 */ 45 #define MIN_ARC_MAX DMU_MAX_ACCESS 46 47 #define HDR_SET_LSIZE(hdr, x) do { \ 48 ASSERT(IS_P2ALIGNED(x, 1U << SPA_MINBLOCKSHIFT)); \ 49 (hdr)->b_lsize = ((x) >> SPA_MINBLOCKSHIFT); \ 50 } while (0) 51 52 #define HDR_SET_PSIZE(hdr, x) do { \ 53 ASSERT(IS_P2ALIGNED((x), 1U << SPA_MINBLOCKSHIFT)); \ 54 (hdr)->b_psize = ((x) >> SPA_MINBLOCKSHIFT); \ 55 } while (0) 56 57 /* The l2size in the header is only used by L2 cache */ 58 #define HDR_SET_L2SIZE(hdr, x) do { \ 59 ASSERT(IS_P2ALIGNED((x), 1U << SPA_MINBLOCKSHIFT)); \ 60 (hdr)->b_l2size = ((x) >> SPA_MINBLOCKSHIFT); \ 61 } while (0) 62 63 #define HDR_GET_LSIZE(hdr) ((hdr)->b_lsize << SPA_MINBLOCKSHIFT) 64 #define HDR_GET_PSIZE(hdr) ((hdr)->b_psize << SPA_MINBLOCKSHIFT) 65 #define HDR_GET_L2SIZE(hdr) ((hdr)->b_l2size << SPA_MINBLOCKSHIFT) 66 67 typedef struct arc_buf_hdr arc_buf_hdr_t; 68 typedef struct arc_buf arc_buf_t; 69 typedef struct arc_prune arc_prune_t; 70 71 /* 72 * Because the ARC can store encrypted data, errors (not due to bugs) may arise 73 * while transforming data into its desired format - specifically, when 74 * decrypting, the key may not be present, or the HMAC may not be correct 75 * which signifies deliberate tampering with the on-disk state 76 * (assuming that the checksum was correct). If any error occurs, the "buf" 77 * parameter will be NULL. 78 */ 79 typedef void arc_read_done_func_t(zio_t *zio, const zbookmark_phys_t *zb, 80 const blkptr_t *bp, arc_buf_t *buf, void *priv); 81 typedef void arc_write_done_func_t(zio_t *zio, arc_buf_t *buf, void *priv); 82 typedef void arc_prune_func_t(uint64_t bytes, void *priv); 83 84 /* Shared module parameters */ 85 extern uint_t zfs_arc_average_blocksize; 86 extern int l2arc_exclude_special; 87 88 /* generic arc_done_func_t which can be used */ 89 arc_read_done_func_t arc_getbuf_func; 90 91 /* generic arc_prune_func_t wrapper for callbacks */ 92 struct arc_prune { 93 arc_prune_func_t *p_pfunc; 94 void *p_private; 95 uint64_t p_adjust; 96 list_node_t p_node; 97 zfs_refcount_t p_refcnt; 98 }; 99 100 typedef enum arc_strategy { 101 ARC_STRATEGY_META_ONLY = 0, /* Evict only meta data buffers */ 102 ARC_STRATEGY_META_BALANCED = 1, /* Evict data buffers if needed */ 103 } arc_strategy_t; 104 105 typedef enum arc_flags 106 { 107 /* 108 * Public flags that can be passed into the ARC by external consumers. 109 */ 110 ARC_FLAG_WAIT = 1 << 0, /* perform sync I/O */ 111 ARC_FLAG_NOWAIT = 1 << 1, /* perform async I/O */ 112 ARC_FLAG_PREFETCH = 1 << 2, /* I/O is a prefetch */ 113 ARC_FLAG_CACHED = 1 << 3, /* I/O was in cache */ 114 ARC_FLAG_L2CACHE = 1 << 4, /* cache in L2ARC */ 115 ARC_FLAG_UNCACHED = 1 << 5, /* evict after use */ 116 ARC_FLAG_PRESCIENT_PREFETCH = 1 << 6, /* long min lifespan */ 117 118 /* 119 * Private ARC flags. These flags are private ARC only flags that 120 * will show up in b_flags in the arc_buf_hdr_t. These flags should 121 * only be set by ARC code. 122 */ 123 ARC_FLAG_IN_HASH_TABLE = 1 << 7, /* buffer is hashed */ 124 ARC_FLAG_IO_IN_PROGRESS = 1 << 8, /* I/O in progress */ 125 ARC_FLAG_IO_ERROR = 1 << 9, /* I/O failed for buf */ 126 ARC_FLAG_INDIRECT = 1 << 10, /* indirect block */ 127 /* Indicates that block was read with ASYNC priority. */ 128 ARC_FLAG_PRIO_ASYNC_READ = 1 << 11, 129 ARC_FLAG_L2_WRITING = 1 << 12, /* write in progress */ 130 ARC_FLAG_L2_EVICTED = 1 << 13, /* evicted during I/O */ 131 ARC_FLAG_L2_WRITE_HEAD = 1 << 14, /* head of write list */ 132 /* 133 * Encrypted or authenticated on disk (may be plaintext in memory). 134 * This header has b_crypt_hdr allocated. Does not include indirect 135 * blocks with checksums of MACs which will also have their X 136 * (encrypted) bit set in the bp. 137 */ 138 ARC_FLAG_PROTECTED = 1 << 15, 139 /* data has not been authenticated yet */ 140 ARC_FLAG_NOAUTH = 1 << 16, 141 /* indicates that the buffer contains metadata (otherwise, data) */ 142 ARC_FLAG_BUFC_METADATA = 1 << 17, 143 144 /* Flags specifying whether optional hdr struct fields are defined */ 145 ARC_FLAG_HAS_L1HDR = 1 << 18, 146 ARC_FLAG_HAS_L2HDR = 1 << 19, 147 148 /* 149 * Indicates the arc_buf_hdr_t's b_pdata matches the on-disk data. 150 * This allows the l2arc to use the blkptr's checksum to verify 151 * the data without having to store the checksum in the hdr. 152 */ 153 ARC_FLAG_COMPRESSED_ARC = 1 << 20, 154 ARC_FLAG_SHARED_DATA = 1 << 21, 155 156 /* 157 * Fail this arc_read() (with ENOENT) if the data is not already present 158 * in cache. 159 */ 160 ARC_FLAG_CACHED_ONLY = 1 << 22, 161 162 /* 163 * Don't instantiate an arc_buf_t for arc_read_done. 164 */ 165 ARC_FLAG_NO_BUF = 1 << 23, 166 167 /* 168 * The arc buffer's compression mode is stored in the top 7 bits of the 169 * flags field, so these dummy flags are included so that MDB can 170 * interpret the enum properly. 171 */ 172 ARC_FLAG_COMPRESS_0 = 1 << 24, 173 ARC_FLAG_COMPRESS_1 = 1 << 25, 174 ARC_FLAG_COMPRESS_2 = 1 << 26, 175 ARC_FLAG_COMPRESS_3 = 1 << 27, 176 ARC_FLAG_COMPRESS_4 = 1 << 28, 177 ARC_FLAG_COMPRESS_5 = 1 << 29, 178 ARC_FLAG_COMPRESS_6 = 1 << 30 179 } arc_flags_t; 180 181 typedef enum arc_buf_flags { 182 ARC_BUF_FLAG_SHARED = 1 << 0, 183 ARC_BUF_FLAG_COMPRESSED = 1 << 1, 184 /* 185 * indicates whether this arc_buf_t is encrypted, regardless of 186 * state on-disk 187 */ 188 ARC_BUF_FLAG_ENCRYPTED = 1 << 2 189 } arc_buf_flags_t; 190 191 struct arc_buf { 192 arc_buf_hdr_t *b_hdr; 193 arc_buf_t *b_next; 194 void *b_data; 195 arc_buf_flags_t b_flags; 196 }; 197 198 typedef enum arc_buf_contents { 199 ARC_BUFC_DATA, /* buffer contains data */ 200 ARC_BUFC_METADATA, /* buffer contains metadata */ 201 ARC_BUFC_NUMTYPES 202 } arc_buf_contents_t; 203 204 /* 205 * The following breakdowns of arc_size exist for kstat only. 206 */ 207 typedef enum arc_space_type { 208 ARC_SPACE_DATA, 209 ARC_SPACE_META, 210 ARC_SPACE_HDRS, 211 ARC_SPACE_L2HDRS, 212 ARC_SPACE_DBUF, 213 ARC_SPACE_DNODE, 214 ARC_SPACE_BONUS, 215 ARC_SPACE_ABD_CHUNK_WASTE, 216 ARC_SPACE_NUMTYPES 217 } arc_space_type_t; 218 219 typedef enum arc_state_type { 220 ARC_STATE_ANON, 221 ARC_STATE_MRU, 222 ARC_STATE_MRU_GHOST, 223 ARC_STATE_MFU, 224 ARC_STATE_MFU_GHOST, 225 ARC_STATE_L2C_ONLY, 226 ARC_STATE_UNCACHED, 227 ARC_STATE_NUMTYPES 228 } arc_state_type_t; 229 230 typedef struct arc_buf_info { 231 arc_state_type_t abi_state_type; 232 arc_buf_contents_t abi_state_contents; 233 uint32_t abi_flags; 234 uint32_t abi_bufcnt; 235 uint64_t abi_size; 236 uint64_t abi_spa; 237 uint64_t abi_access; 238 uint32_t abi_mru_hits; 239 uint32_t abi_mru_ghost_hits; 240 uint32_t abi_mfu_hits; 241 uint32_t abi_mfu_ghost_hits; 242 uint32_t abi_l2arc_hits; 243 uint32_t abi_holds; 244 uint64_t abi_l2arc_dattr; 245 uint64_t abi_l2arc_asize; 246 enum zio_compress abi_l2arc_compress; 247 } arc_buf_info_t; 248 249 /* 250 * Flags returned by arc_cached; describes which part of the arc 251 * the block is cached in. 252 */ 253 #define ARC_CACHED_EMBEDDED (1U << 0) 254 #define ARC_CACHED_IN_L1 (1U << 1) 255 #define ARC_CACHED_IN_MRU (1U << 2) 256 #define ARC_CACHED_IN_MFU (1U << 3) 257 #define ARC_CACHED_IN_L2 (1U << 4) 258 259 void arc_space_consume(uint64_t space, arc_space_type_t type); 260 void arc_space_return(uint64_t space, arc_space_type_t type); 261 boolean_t arc_is_metadata(arc_buf_t *buf); 262 boolean_t arc_is_encrypted(arc_buf_t *buf); 263 boolean_t arc_is_unauthenticated(arc_buf_t *buf); 264 enum zio_compress arc_get_compression(arc_buf_t *buf); 265 void arc_get_raw_params(arc_buf_t *buf, boolean_t *byteorder, uint8_t *salt, 266 uint8_t *iv, uint8_t *mac); 267 int arc_untransform(arc_buf_t *buf, spa_t *spa, const zbookmark_phys_t *zb, 268 boolean_t in_place); 269 void arc_convert_to_raw(arc_buf_t *buf, uint64_t dsobj, boolean_t byteorder, 270 dmu_object_type_t ot, const uint8_t *salt, const uint8_t *iv, 271 const uint8_t *mac); 272 arc_buf_t *arc_alloc_buf(spa_t *spa, const void *tag, arc_buf_contents_t type, 273 int32_t size); 274 arc_buf_t *arc_alloc_compressed_buf(spa_t *spa, const void *tag, 275 uint64_t psize, uint64_t lsize, enum zio_compress compression_type, 276 uint8_t complevel); 277 arc_buf_t *arc_alloc_raw_buf(spa_t *spa, const void *tag, uint64_t dsobj, 278 boolean_t byteorder, const uint8_t *salt, const uint8_t *iv, 279 const uint8_t *mac, dmu_object_type_t ot, uint64_t psize, uint64_t lsize, 280 enum zio_compress compression_type, uint8_t complevel); 281 uint8_t arc_get_complevel(arc_buf_t *buf); 282 arc_buf_t *arc_loan_buf(spa_t *spa, boolean_t is_metadata, int size); 283 arc_buf_t *arc_loan_compressed_buf(spa_t *spa, uint64_t psize, uint64_t lsize, 284 enum zio_compress compression_type, uint8_t complevel); 285 arc_buf_t *arc_loan_raw_buf(spa_t *spa, uint64_t dsobj, boolean_t byteorder, 286 const uint8_t *salt, const uint8_t *iv, const uint8_t *mac, 287 dmu_object_type_t ot, uint64_t psize, uint64_t lsize, 288 enum zio_compress compression_type, uint8_t complevel); 289 void arc_return_buf(arc_buf_t *buf, const void *tag); 290 void arc_loan_inuse_buf(arc_buf_t *buf, const void *tag); 291 void arc_buf_destroy(arc_buf_t *buf, const void *tag); 292 void arc_buf_info(arc_buf_t *buf, arc_buf_info_t *abi, int state_index); 293 uint64_t arc_buf_size(arc_buf_t *buf); 294 uint64_t arc_buf_lsize(arc_buf_t *buf); 295 void arc_buf_access(arc_buf_t *buf); 296 void arc_release(arc_buf_t *buf, const void *tag); 297 int arc_released(arc_buf_t *buf); 298 void arc_buf_sigsegv(int sig, siginfo_t *si, void *unused); 299 void arc_buf_freeze(arc_buf_t *buf); 300 void arc_buf_thaw(arc_buf_t *buf); 301 #ifdef ZFS_DEBUG 302 int arc_referenced(arc_buf_t *buf); 303 #else 304 #define arc_referenced(buf) ((void) sizeof (buf), 0) 305 #endif 306 307 int arc_read(zio_t *pio, spa_t *spa, const blkptr_t *bp, 308 arc_read_done_func_t *done, void *priv, zio_priority_t priority, 309 int flags, arc_flags_t *arc_flags, const zbookmark_phys_t *zb); 310 zio_t *arc_write(zio_t *pio, spa_t *spa, uint64_t txg, blkptr_t *bp, 311 arc_buf_t *buf, boolean_t uncached, boolean_t l2arc, const zio_prop_t *zp, 312 arc_write_done_func_t *ready, arc_write_done_func_t *child_ready, 313 arc_write_done_func_t *done, void *priv, zio_priority_t priority, 314 int zio_flags, const zbookmark_phys_t *zb); 315 316 arc_prune_t *arc_add_prune_callback(arc_prune_func_t *func, void *priv); 317 void arc_remove_prune_callback(arc_prune_t *p); 318 void arc_freed(spa_t *spa, const blkptr_t *bp); 319 int arc_cached(spa_t *spa, const blkptr_t *bp); 320 321 void arc_flush(spa_t *spa, boolean_t retry); 322 void arc_flush_async(spa_t *spa); 323 void arc_tempreserve_clear(uint64_t reserve); 324 int arc_tempreserve_space(spa_t *spa, uint64_t reserve, uint64_t txg); 325 boolean_t arc_async_flush_guid_inuse(uint64_t load_guid); 326 327 uint64_t arc_all_memory(void); 328 uint64_t arc_default_max(uint64_t min, uint64_t allmem); 329 uint64_t arc_target_bytes(void); 330 uint64_t arc_boot_target_bytes(void); 331 void arc_set_limits(uint64_t); 332 void arc_init(void); 333 void arc_fini(void); 334 335 /* 336 * Level 2 ARC 337 */ 338 339 void l2arc_add_vdev(spa_t *spa, vdev_t *vd); 340 void l2arc_remove_vdev(vdev_t *vd); 341 boolean_t l2arc_vdev_present(vdev_t *vd); 342 void l2arc_rebuild_vdev(vdev_t *vd, boolean_t reopen); 343 boolean_t l2arc_range_check_overlap(uint64_t bottom, uint64_t top, 344 uint64_t check); 345 void l2arc_init(void); 346 void l2arc_fini(void); 347 void l2arc_spa_rebuild_start(spa_t *spa); 348 void l2arc_spa_rebuild_stop(spa_t *spa); 349 350 #ifndef _KERNEL 351 extern boolean_t arc_watch; 352 #endif 353 354 #ifdef __cplusplus 355 } 356 #endif 357 358 #endif /* _SYS_ARC_H */ 359