include/sys/arc.h

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