1 /* 2 * CDDL HEADER START 3 * 4 * The contents of this file are subject to the terms of the 5 * Common Development and Distribution License (the "License"). 6 * You may not use this file except in compliance with the License. 7 * 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9 * or http://www.opensolaris.org/os/licensing. 10 * See the License for the specific language governing permissions 11 * and limitations under the License. 12 * 13 * When distributing Covered Code, include this CDDL HEADER in each 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15 * If applicable, add the following below this CDDL HEADER, with the 16 * fields enclosed by brackets "[]" replaced with your own identifying 17 * information: Portions Copyright [yyyy] [name of copyright owner] 18 * 19 * CDDL HEADER END 20 */ 21 /* 22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. 23 * Copyright (c) 2012, 2016 by Delphix. All rights reserved. 24 * Copyright (c) 2013 by Saso Kiselkov. All rights reserved. 25 * Copyright (c) 2019, Allan Jude 26 * Copyright (c) 2019, Klara Inc. 27 */ 28 29 #ifndef _SYS_ARC_H 30 #define _SYS_ARC_H 31 32 #include <sys/zfs_context.h> 33 34 #ifdef __cplusplus 35 extern "C" { 36 #endif 37 38 #include <sys/zio.h> 39 #include <sys/dmu.h> 40 #include <sys/spa.h> 41 #include <sys/zfs_refcount.h> 42 43 /* 44 * Used by arc_flush() to inform arc_evict_state() that it should evict 45 * all available buffers from the arc state being passed in. 46 */ 47 #define ARC_EVICT_ALL UINT64_MAX 48 49 /* 50 * ZFS gets very unhappy when the maximum ARC size is smaller than the maximum 51 * block size and a larger block is written. To leave some safety margin, we 52 * limit the minimum for zfs_arc_max to the maximium transaction size. 53 */ 54 #define MIN_ARC_MAX DMU_MAX_ACCESS 55 56 #define HDR_SET_LSIZE(hdr, x) do { \ 57 ASSERT(IS_P2ALIGNED(x, 1U << SPA_MINBLOCKSHIFT)); \ 58 (hdr)->b_lsize = ((x) >> SPA_MINBLOCKSHIFT); \ 59 _NOTE(CONSTCOND) } while (0) 60 61 #define HDR_SET_PSIZE(hdr, x) do { \ 62 ASSERT(IS_P2ALIGNED((x), 1U << SPA_MINBLOCKSHIFT)); \ 63 (hdr)->b_psize = ((x) >> SPA_MINBLOCKSHIFT); \ 64 _NOTE(CONSTCOND) } while (0) 65 66 #define HDR_GET_LSIZE(hdr) ((hdr)->b_lsize << SPA_MINBLOCKSHIFT) 67 #define HDR_GET_PSIZE(hdr) ((hdr)->b_psize << SPA_MINBLOCKSHIFT) 68 69 typedef struct arc_buf_hdr arc_buf_hdr_t; 70 typedef struct arc_buf arc_buf_t; 71 typedef struct arc_prune arc_prune_t; 72 73 /* 74 * Because the ARC can store encrypted data, errors (not due to bugs) may arise 75 * while transforming data into its desired format - specifically, when 76 * decrypting, the key may not be present, or the HMAC may not be correct 77 * which signifies deliberate tampering with the on-disk state 78 * (assuming that the checksum was correct). If any error occurs, the "buf" 79 * parameter will be NULL. 80 */ 81 typedef void arc_read_done_func_t(zio_t *zio, const zbookmark_phys_t *zb, 82 const blkptr_t *bp, arc_buf_t *buf, void *priv); 83 typedef void arc_write_done_func_t(zio_t *zio, arc_buf_t *buf, void *priv); 84 typedef void arc_prune_func_t(int64_t bytes, void *priv); 85 86 /* Shared module parameters */ 87 extern int zfs_arc_average_blocksize; 88 89 /* generic arc_done_func_t's which you can use */ 90 arc_read_done_func_t arc_bcopy_func; 91 arc_read_done_func_t arc_getbuf_func; 92 93 /* generic arc_prune_func_t wrapper for callbacks */ 94 struct arc_prune { 95 arc_prune_func_t *p_pfunc; 96 void *p_private; 97 uint64_t p_adjust; 98 list_node_t p_node; 99 zfs_refcount_t p_refcnt; 100 }; 101 102 typedef enum arc_strategy { 103 ARC_STRATEGY_META_ONLY = 0, /* Evict only meta data buffers */ 104 ARC_STRATEGY_META_BALANCED = 1, /* Evict data buffers if needed */ 105 } arc_strategy_t; 106 107 typedef enum arc_flags 108 { 109 /* 110 * Public flags that can be passed into the ARC by external consumers. 111 */ 112 ARC_FLAG_WAIT = 1 << 0, /* perform sync I/O */ 113 ARC_FLAG_NOWAIT = 1 << 1, /* perform async I/O */ 114 ARC_FLAG_PREFETCH = 1 << 2, /* I/O is a prefetch */ 115 ARC_FLAG_CACHED = 1 << 3, /* I/O was in cache */ 116 ARC_FLAG_L2CACHE = 1 << 4, /* cache in L2ARC */ 117 ARC_FLAG_PREDICTIVE_PREFETCH = 1 << 5, /* I/O from zfetch */ 118 ARC_FLAG_PRESCIENT_PREFETCH = 1 << 6, /* long min lifespan */ 119 120 /* 121 * Private ARC flags. These flags are private ARC only flags that 122 * will show up in b_flags in the arc_hdr_buf_t. These flags should 123 * only be set by ARC code. 124 */ 125 ARC_FLAG_IN_HASH_TABLE = 1 << 7, /* buffer is hashed */ 126 ARC_FLAG_IO_IN_PROGRESS = 1 << 8, /* I/O in progress */ 127 ARC_FLAG_IO_ERROR = 1 << 9, /* I/O failed for buf */ 128 ARC_FLAG_INDIRECT = 1 << 10, /* indirect block */ 129 /* Indicates that block was read with ASYNC priority. */ 130 ARC_FLAG_PRIO_ASYNC_READ = 1 << 11, 131 ARC_FLAG_L2_WRITING = 1 << 12, /* write in progress */ 132 ARC_FLAG_L2_EVICTED = 1 << 13, /* evicted during I/O */ 133 ARC_FLAG_L2_WRITE_HEAD = 1 << 14, /* head of write list */ 134 /* 135 * Encrypted or authenticated on disk (may be plaintext in memory). 136 * This header has b_crypt_hdr allocated. Does not include indirect 137 * blocks with checksums of MACs which will also have their X 138 * (encrypted) bit set in the bp. 139 */ 140 ARC_FLAG_PROTECTED = 1 << 15, 141 /* data has not been authenticated yet */ 142 ARC_FLAG_NOAUTH = 1 << 16, 143 /* indicates that the buffer contains metadata (otherwise, data) */ 144 ARC_FLAG_BUFC_METADATA = 1 << 17, 145 146 /* Flags specifying whether optional hdr struct fields are defined */ 147 ARC_FLAG_HAS_L1HDR = 1 << 18, 148 ARC_FLAG_HAS_L2HDR = 1 << 19, 149 150 /* 151 * Indicates the arc_buf_hdr_t's b_pdata matches the on-disk data. 152 * This allows the l2arc to use the blkptr's checksum to verify 153 * the data without having to store the checksum in the hdr. 154 */ 155 ARC_FLAG_COMPRESSED_ARC = 1 << 20, 156 ARC_FLAG_SHARED_DATA = 1 << 21, 157 158 /* 159 * Fail this arc_read() (with ENOENT) if the data is not already present 160 * in cache. 161 */ 162 ARC_FLAG_CACHED_ONLY = 1 << 22, 163 164 /* 165 * Don't instantiate an arc_buf_t for arc_read_done. 166 */ 167 ARC_FLAG_NO_BUF = 1 << 23, 168 169 /* 170 * The arc buffer's compression mode is stored in the top 7 bits of the 171 * flags field, so these dummy flags are included so that MDB can 172 * interpret the enum properly. 173 */ 174 ARC_FLAG_COMPRESS_0 = 1 << 24, 175 ARC_FLAG_COMPRESS_1 = 1 << 25, 176 ARC_FLAG_COMPRESS_2 = 1 << 26, 177 ARC_FLAG_COMPRESS_3 = 1 << 27, 178 ARC_FLAG_COMPRESS_4 = 1 << 28, 179 ARC_FLAG_COMPRESS_5 = 1 << 29, 180 ARC_FLAG_COMPRESS_6 = 1 << 30 181 182 } arc_flags_t; 183 184 typedef enum arc_buf_flags { 185 ARC_BUF_FLAG_SHARED = 1 << 0, 186 ARC_BUF_FLAG_COMPRESSED = 1 << 1, 187 /* 188 * indicates whether this arc_buf_t is encrypted, regardless of 189 * state on-disk 190 */ 191 ARC_BUF_FLAG_ENCRYPTED = 1 << 2 192 } arc_buf_flags_t; 193 194 struct arc_buf { 195 arc_buf_hdr_t *b_hdr; 196 arc_buf_t *b_next; 197 kmutex_t b_evict_lock; 198 void *b_data; 199 arc_buf_flags_t b_flags; 200 }; 201 202 typedef enum arc_buf_contents { 203 ARC_BUFC_INVALID, /* invalid type */ 204 ARC_BUFC_DATA, /* buffer contains data */ 205 ARC_BUFC_METADATA, /* buffer contains metadata */ 206 ARC_BUFC_NUMTYPES 207 } arc_buf_contents_t; 208 209 /* 210 * The following breakdowns of arc_size exist for kstat only. 211 */ 212 typedef enum arc_space_type { 213 ARC_SPACE_DATA, 214 ARC_SPACE_META, 215 ARC_SPACE_HDRS, 216 ARC_SPACE_L2HDRS, 217 ARC_SPACE_DBUF, 218 ARC_SPACE_DNODE, 219 ARC_SPACE_BONUS, 220 ARC_SPACE_ABD_CHUNK_WASTE, 221 ARC_SPACE_NUMTYPES 222 } arc_space_type_t; 223 224 typedef enum arc_state_type { 225 ARC_STATE_ANON, 226 ARC_STATE_MRU, 227 ARC_STATE_MRU_GHOST, 228 ARC_STATE_MFU, 229 ARC_STATE_MFU_GHOST, 230 ARC_STATE_L2C_ONLY, 231 ARC_STATE_NUMTYPES 232 } arc_state_type_t; 233 234 typedef struct arc_buf_info { 235 arc_state_type_t abi_state_type; 236 arc_buf_contents_t abi_state_contents; 237 uint32_t abi_flags; 238 uint32_t abi_bufcnt; 239 uint64_t abi_size; 240 uint64_t abi_spa; 241 uint64_t abi_access; 242 uint32_t abi_mru_hits; 243 uint32_t abi_mru_ghost_hits; 244 uint32_t abi_mfu_hits; 245 uint32_t abi_mfu_ghost_hits; 246 uint32_t abi_l2arc_hits; 247 uint32_t abi_holds; 248 uint64_t abi_l2arc_dattr; 249 uint64_t abi_l2arc_asize; 250 enum zio_compress abi_l2arc_compress; 251 } arc_buf_info_t; 252 253 void arc_space_consume(uint64_t space, arc_space_type_t type); 254 void arc_space_return(uint64_t space, arc_space_type_t type); 255 boolean_t arc_is_metadata(arc_buf_t *buf); 256 boolean_t arc_is_encrypted(arc_buf_t *buf); 257 boolean_t arc_is_unauthenticated(arc_buf_t *buf); 258 enum zio_compress arc_get_compression(arc_buf_t *buf); 259 void arc_get_raw_params(arc_buf_t *buf, boolean_t *byteorder, uint8_t *salt, 260 uint8_t *iv, uint8_t *mac); 261 int arc_untransform(arc_buf_t *buf, spa_t *spa, const zbookmark_phys_t *zb, 262 boolean_t in_place); 263 void arc_convert_to_raw(arc_buf_t *buf, uint64_t dsobj, boolean_t byteorder, 264 dmu_object_type_t ot, const uint8_t *salt, const uint8_t *iv, 265 const uint8_t *mac); 266 arc_buf_t *arc_alloc_buf(spa_t *spa, void *tag, arc_buf_contents_t type, 267 int32_t size); 268 arc_buf_t *arc_alloc_compressed_buf(spa_t *spa, void *tag, 269 uint64_t psize, uint64_t lsize, enum zio_compress compression_type, 270 uint8_t complevel); 271 arc_buf_t *arc_alloc_raw_buf(spa_t *spa, void *tag, uint64_t dsobj, 272 boolean_t byteorder, const uint8_t *salt, const uint8_t *iv, 273 const uint8_t *mac, dmu_object_type_t ot, uint64_t psize, uint64_t lsize, 274 enum zio_compress compression_type, uint8_t complevel); 275 uint8_t arc_get_complevel(arc_buf_t *buf); 276 arc_buf_t *arc_loan_buf(spa_t *spa, boolean_t is_metadata, int size); 277 arc_buf_t *arc_loan_compressed_buf(spa_t *spa, uint64_t psize, uint64_t lsize, 278 enum zio_compress compression_type, uint8_t complevel); 279 arc_buf_t *arc_loan_raw_buf(spa_t *spa, uint64_t dsobj, boolean_t byteorder, 280 const uint8_t *salt, const uint8_t *iv, const uint8_t *mac, 281 dmu_object_type_t ot, uint64_t psize, uint64_t lsize, 282 enum zio_compress compression_type, uint8_t complevel); 283 void arc_return_buf(arc_buf_t *buf, void *tag); 284 void arc_loan_inuse_buf(arc_buf_t *buf, void *tag); 285 void arc_buf_destroy(arc_buf_t *buf, void *tag); 286 void arc_buf_info(arc_buf_t *buf, arc_buf_info_t *abi, int state_index); 287 uint64_t arc_buf_size(arc_buf_t *buf); 288 uint64_t arc_buf_lsize(arc_buf_t *buf); 289 void arc_buf_access(arc_buf_t *buf); 290 void arc_release(arc_buf_t *buf, void *tag); 291 int arc_released(arc_buf_t *buf); 292 void arc_buf_sigsegv(int sig, siginfo_t *si, void *unused); 293 void arc_buf_freeze(arc_buf_t *buf); 294 void arc_buf_thaw(arc_buf_t *buf); 295 #ifdef ZFS_DEBUG 296 int arc_referenced(arc_buf_t *buf); 297 #endif 298 299 int arc_read(zio_t *pio, spa_t *spa, const blkptr_t *bp, 300 arc_read_done_func_t *done, void *priv, zio_priority_t priority, 301 int flags, arc_flags_t *arc_flags, const zbookmark_phys_t *zb); 302 zio_t *arc_write(zio_t *pio, spa_t *spa, uint64_t txg, 303 blkptr_t *bp, arc_buf_t *buf, boolean_t l2arc, const zio_prop_t *zp, 304 arc_write_done_func_t *ready, arc_write_done_func_t *child_ready, 305 arc_write_done_func_t *physdone, arc_write_done_func_t *done, 306 void *priv, zio_priority_t priority, int zio_flags, 307 const zbookmark_phys_t *zb); 308 309 arc_prune_t *arc_add_prune_callback(arc_prune_func_t *func, void *priv); 310 void arc_remove_prune_callback(arc_prune_t *p); 311 void arc_freed(spa_t *spa, const blkptr_t *bp); 312 313 void arc_flush(spa_t *spa, boolean_t retry); 314 void arc_tempreserve_clear(uint64_t reserve); 315 int arc_tempreserve_space(spa_t *spa, uint64_t reserve, uint64_t txg); 316 317 uint64_t arc_all_memory(void); 318 uint64_t arc_default_max(uint64_t min, uint64_t allmem); 319 uint64_t arc_target_bytes(void); 320 void arc_set_limits(uint64_t); 321 void arc_init(void); 322 void arc_fini(void); 323 324 /* 325 * Level 2 ARC 326 */ 327 328 void l2arc_add_vdev(spa_t *spa, vdev_t *vd); 329 void l2arc_remove_vdev(vdev_t *vd); 330 boolean_t l2arc_vdev_present(vdev_t *vd); 331 void l2arc_rebuild_vdev(vdev_t *vd, boolean_t reopen); 332 boolean_t l2arc_range_check_overlap(uint64_t bottom, uint64_t top, 333 uint64_t check); 334 void l2arc_init(void); 335 void l2arc_fini(void); 336 void l2arc_start(void); 337 void l2arc_stop(void); 338 void l2arc_spa_rebuild_start(spa_t *spa); 339 340 #ifndef _KERNEL 341 extern boolean_t arc_watch; 342 #endif 343 344 #ifdef __cplusplus 345 } 346 #endif 347 348 #endif /* _SYS_ARC_H */ 349