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 https://opensource.org/licenses/CDDL-1.0.
  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 } 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 } while (0)
  65
  66 /* The l2size in the header is only used by L2 cache */
  67 #define HDR_SET_L2SIZE(hdr, x) do { \
  68         ASSERT(IS_P2ALIGNED((x), 1U << SPA_MINBLOCKSHIFT)); \
  69         (hdr)->b_l2size = ((x) >> SPA_MINBLOCKSHIFT); \
  70 } while (0)
  71
  72 #define HDR_GET_LSIZE(hdr)      ((hdr)->b_lsize << SPA_MINBLOCKSHIFT)
  73 #define HDR_GET_PSIZE(hdr)      ((hdr)->b_psize << SPA_MINBLOCKSHIFT)
  74 #define HDR_GET_L2SIZE(hdr)     ((hdr)->b_l2size << SPA_MINBLOCKSHIFT)
  75
  76 typedef struct arc_buf_hdr arc_buf_hdr_t;
  77 typedef struct arc_buf arc_buf_t;
  78 typedef struct arc_prune arc_prune_t;
  79
  80 /*
  81  * Because the ARC can store encrypted data, errors (not due to bugs) may arise
  82  * while transforming data into its desired format - specifically, when
  83  * decrypting, the key may not be present, or the HMAC may not be correct
  84  * which signifies deliberate tampering with the on-disk state
  85  * (assuming that the checksum was correct). If any error occurs, the "buf"
  86  * parameter will be NULL.
  87  */
  88 typedef void arc_read_done_func_t(zio_t *zio, const zbookmark_phys_t *zb,
  89     const blkptr_t *bp, arc_buf_t *buf, void *priv);
  90 typedef void arc_write_done_func_t(zio_t *zio, arc_buf_t *buf, void *priv);
  91 typedef void arc_prune_func_t(uint64_t bytes, void *priv);
  92
  93 /* Shared module parameters */
  94 extern uint_t zfs_arc_average_blocksize;
  95 extern int l2arc_exclude_special;
  96
  97 /* generic arc_done_func_t's which you can use */
  98 arc_read_done_func_t arc_bcopy_func;
  99 arc_read_done_func_t arc_getbuf_func;
 100
 101 /* generic arc_prune_func_t wrapper for callbacks */
 102 struct arc_prune {
 103         arc_prune_func_t        *p_pfunc;
 104         void                    *p_private;
 105         uint64_t                p_adjust;
 106         list_node_t             p_node;
 107         zfs_refcount_t          p_refcnt;
 108 };
 109
 110 typedef enum arc_strategy {
 111         ARC_STRATEGY_META_ONLY          = 0, /* Evict only meta data buffers */
 112         ARC_STRATEGY_META_BALANCED      = 1, /* Evict data buffers if needed */
 113 } arc_strategy_t;
 114
 115 typedef enum arc_flags
 116 {
 117         /*
 118          * Public flags that can be passed into the ARC by external consumers.
 119          */
 120         ARC_FLAG_WAIT                   = 1 << 0,       /* perform sync I/O */
 121         ARC_FLAG_NOWAIT                 = 1 << 1,       /* perform async I/O */
 122         ARC_FLAG_PREFETCH               = 1 << 2,       /* I/O is a prefetch */
 123         ARC_FLAG_CACHED                 = 1 << 3,       /* I/O was in cache */
 124         ARC_FLAG_L2CACHE                = 1 << 4,       /* cache in L2ARC */
 125         ARC_FLAG_UNCACHED               = 1 << 5,       /* evict after use */
 126         ARC_FLAG_PRESCIENT_PREFETCH     = 1 << 6,       /* long min lifespan */
 127
 128         /*
 129          * Private ARC flags.  These flags are private ARC only flags that
 130          * will show up in b_flags in the arc_buf_hdr_t. These flags should
 131          * only be set by ARC code.
 132          */
 133         ARC_FLAG_IN_HASH_TABLE          = 1 << 7,       /* buffer is hashed */
 134         ARC_FLAG_IO_IN_PROGRESS         = 1 << 8,       /* I/O in progress */
 135         ARC_FLAG_IO_ERROR               = 1 << 9,       /* I/O failed for buf */
 136         ARC_FLAG_INDIRECT               = 1 << 10,      /* indirect block */
 137         /* Indicates that block was read with ASYNC priority. */
 138         ARC_FLAG_PRIO_ASYNC_READ        = 1 << 11,
 139         ARC_FLAG_L2_WRITING             = 1 << 12,      /* write in progress */
 140         ARC_FLAG_L2_EVICTED             = 1 << 13,      /* evicted during I/O */
 141         ARC_FLAG_L2_WRITE_HEAD          = 1 << 14,      /* head of write list */
 142         /*
 143          * Encrypted or authenticated on disk (may be plaintext in memory).
 144          * This header has b_crypt_hdr allocated. Does not include indirect
 145          * blocks with checksums of MACs which will also have their X
 146          * (encrypted) bit set in the bp.
 147          */
 148         ARC_FLAG_PROTECTED              = 1 << 15,
 149         /* data has not been authenticated yet */
 150         ARC_FLAG_NOAUTH                 = 1 << 16,
 151         /* indicates that the buffer contains metadata (otherwise, data) */
 152         ARC_FLAG_BUFC_METADATA          = 1 << 17,
 153
 154         /* Flags specifying whether optional hdr struct fields are defined */
 155         ARC_FLAG_HAS_L1HDR              = 1 << 18,
 156         ARC_FLAG_HAS_L2HDR              = 1 << 19,
 157
 158         /*
 159          * Indicates the arc_buf_hdr_t's b_pdata matches the on-disk data.
 160          * This allows the l2arc to use the blkptr's checksum to verify
 161          * the data without having to store the checksum in the hdr.
 162          */
 163         ARC_FLAG_COMPRESSED_ARC         = 1 << 20,
 164         ARC_FLAG_SHARED_DATA            = 1 << 21,
 165
 166         /*
 167          * Fail this arc_read() (with ENOENT) if the data is not already present
 168          * in cache.
 169          */
 170         ARC_FLAG_CACHED_ONLY            = 1 << 22,
 171
 172         /*
 173          * Don't instantiate an arc_buf_t for arc_read_done.
 174          */
 175         ARC_FLAG_NO_BUF                 = 1 << 23,
 176
 177         /*
 178          * The arc buffer's compression mode is stored in the top 7 bits of the
 179          * flags field, so these dummy flags are included so that MDB can
 180          * interpret the enum properly.
 181          */
 182         ARC_FLAG_COMPRESS_0             = 1 << 24,
 183         ARC_FLAG_COMPRESS_1             = 1 << 25,
 184         ARC_FLAG_COMPRESS_2             = 1 << 26,
 185         ARC_FLAG_COMPRESS_3             = 1 << 27,
 186         ARC_FLAG_COMPRESS_4             = 1 << 28,
 187         ARC_FLAG_COMPRESS_5             = 1 << 29,
 188         ARC_FLAG_COMPRESS_6             = 1 << 30
 189 } arc_flags_t;
 190
 191 typedef enum arc_buf_flags {
 192         ARC_BUF_FLAG_SHARED             = 1 << 0,
 193         ARC_BUF_FLAG_COMPRESSED         = 1 << 1,
 194         /*
 195          * indicates whether this arc_buf_t is encrypted, regardless of
 196          * state on-disk
 197          */
 198         ARC_BUF_FLAG_ENCRYPTED          = 1 << 2
 199 } arc_buf_flags_t;
 200
 201 struct arc_buf {
 202         arc_buf_hdr_t           *b_hdr;
 203         arc_buf_t               *b_next;
 204         void                    *b_data;
 205         arc_buf_flags_t         b_flags;
 206 };
 207
 208 typedef enum arc_buf_contents {
 209         ARC_BUFC_DATA,                          /* buffer contains data */
 210         ARC_BUFC_METADATA,                      /* buffer contains metadata */
 211         ARC_BUFC_NUMTYPES
 212 } arc_buf_contents_t;
 213
 214 /*
 215  * The following breakdowns of arc_size exist for kstat only.
 216  */
 217 typedef enum arc_space_type {
 218         ARC_SPACE_DATA,
 219         ARC_SPACE_META,
 220         ARC_SPACE_HDRS,
 221         ARC_SPACE_L2HDRS,
 222         ARC_SPACE_DBUF,
 223         ARC_SPACE_DNODE,
 224         ARC_SPACE_BONUS,
 225         ARC_SPACE_ABD_CHUNK_WASTE,
 226         ARC_SPACE_NUMTYPES
 227 } arc_space_type_t;
 228
 229 typedef enum arc_state_type {
 230         ARC_STATE_ANON,
 231         ARC_STATE_MRU,
 232         ARC_STATE_MRU_GHOST,
 233         ARC_STATE_MFU,
 234         ARC_STATE_MFU_GHOST,
 235         ARC_STATE_L2C_ONLY,
 236         ARC_STATE_UNCACHED,
 237         ARC_STATE_NUMTYPES
 238 } arc_state_type_t;
 239
 240 typedef struct arc_buf_info {
 241         arc_state_type_t        abi_state_type;
 242         arc_buf_contents_t      abi_state_contents;
 243         uint32_t                abi_flags;
 244         uint32_t                abi_bufcnt;
 245         uint64_t                abi_size;
 246         uint64_t                abi_spa;
 247         uint64_t                abi_access;
 248         uint32_t                abi_mru_hits;
 249         uint32_t                abi_mru_ghost_hits;
 250         uint32_t                abi_mfu_hits;
 251         uint32_t                abi_mfu_ghost_hits;
 252         uint32_t                abi_l2arc_hits;
 253         uint32_t                abi_holds;
 254         uint64_t                abi_l2arc_dattr;
 255         uint64_t                abi_l2arc_asize;
 256         enum zio_compress       abi_l2arc_compress;
 257 } arc_buf_info_t;
 258
 259 /*
 260  * Flags returned by arc_cached; describes which part of the arc
 261  * the block is cached in.
 262  */
 263 #define ARC_CACHED_EMBEDDED     (1U << 0)
 264 #define ARC_CACHED_IN_L1        (1U << 1)
 265 #define ARC_CACHED_IN_MRU       (1U << 2)
 266 #define ARC_CACHED_IN_MFU       (1U << 3)
 267 #define ARC_CACHED_IN_L2        (1U << 4)
 268
 269 void arc_space_consume(uint64_t space, arc_space_type_t type);
 270 void arc_space_return(uint64_t space, arc_space_type_t type);
 271 boolean_t arc_is_metadata(arc_buf_t *buf);
 272 boolean_t arc_is_encrypted(arc_buf_t *buf);
 273 boolean_t arc_is_unauthenticated(arc_buf_t *buf);
 274 enum zio_compress arc_get_compression(arc_buf_t *buf);
 275 void arc_get_raw_params(arc_buf_t *buf, boolean_t *byteorder, uint8_t *salt,
 276     uint8_t *iv, uint8_t *mac);
 277 int arc_untransform(arc_buf_t *buf, spa_t *spa, const zbookmark_phys_t *zb,
 278     boolean_t in_place);
 279 void arc_convert_to_raw(arc_buf_t *buf, uint64_t dsobj, boolean_t byteorder,
 280     dmu_object_type_t ot, const uint8_t *salt, const uint8_t *iv,
 281     const uint8_t *mac);
 282 arc_buf_t *arc_alloc_buf(spa_t *spa, const void *tag, arc_buf_contents_t type,
 283     int32_t size);
 284 arc_buf_t *arc_alloc_compressed_buf(spa_t *spa, const void *tag,
 285     uint64_t psize, uint64_t lsize, enum zio_compress compression_type,
 286     uint8_t complevel);
 287 arc_buf_t *arc_alloc_raw_buf(spa_t *spa, const void *tag, uint64_t dsobj,
 288     boolean_t byteorder, const uint8_t *salt, const uint8_t *iv,
 289     const uint8_t *mac, dmu_object_type_t ot, uint64_t psize, uint64_t lsize,
 290     enum zio_compress compression_type, uint8_t complevel);
 291 uint8_t arc_get_complevel(arc_buf_t *buf);
 292 arc_buf_t *arc_loan_buf(spa_t *spa, boolean_t is_metadata, int size);
 293 arc_buf_t *arc_loan_compressed_buf(spa_t *spa, uint64_t psize, uint64_t lsize,
 294     enum zio_compress compression_type, uint8_t complevel);
 295 arc_buf_t *arc_loan_raw_buf(spa_t *spa, uint64_t dsobj, boolean_t byteorder,
 296     const uint8_t *salt, const uint8_t *iv, const uint8_t *mac,
 297     dmu_object_type_t ot, uint64_t psize, uint64_t lsize,
 298     enum zio_compress compression_type, uint8_t complevel);
 299 void arc_return_buf(arc_buf_t *buf, const void *tag);
 300 void arc_loan_inuse_buf(arc_buf_t *buf, const void *tag);
 301 void arc_buf_destroy(arc_buf_t *buf, const void *tag);
 302 void arc_buf_info(arc_buf_t *buf, arc_buf_info_t *abi, int state_index);
 303 uint64_t arc_buf_size(arc_buf_t *buf);
 304 uint64_t arc_buf_lsize(arc_buf_t *buf);
 305 void arc_buf_access(arc_buf_t *buf);
 306 void arc_release(arc_buf_t *buf, const void *tag);
 307 int arc_released(arc_buf_t *buf);
 308 void arc_buf_sigsegv(int sig, siginfo_t *si, void *unused);
 309 void arc_buf_freeze(arc_buf_t *buf);
 310 void arc_buf_thaw(arc_buf_t *buf);
 311 #ifdef ZFS_DEBUG
 312 int arc_referenced(arc_buf_t *buf);
 313 #else
 314 #define arc_referenced(buf) ((void) sizeof (buf), 0)
 315 #endif
 316
 317 int arc_read(zio_t *pio, spa_t *spa, const blkptr_t *bp,
 318     arc_read_done_func_t *done, void *priv, zio_priority_t priority,
 319     int flags, arc_flags_t *arc_flags, const zbookmark_phys_t *zb);
 320 zio_t *arc_write(zio_t *pio, spa_t *spa, uint64_t txg, blkptr_t *bp,
 321     arc_buf_t *buf, boolean_t uncached, boolean_t l2arc, const zio_prop_t *zp,
 322     arc_write_done_func_t *ready, arc_write_done_func_t *child_ready,
 323     arc_write_done_func_t *done, void *priv, zio_priority_t priority,
 324     int zio_flags, const zbookmark_phys_t *zb);
 325
 326 arc_prune_t *arc_add_prune_callback(arc_prune_func_t *func, void *priv);
 327 void arc_remove_prune_callback(arc_prune_t *p);
 328 void arc_freed(spa_t *spa, const blkptr_t *bp);
 329 int arc_cached(spa_t *spa, const blkptr_t *bp);
 330
 331 void arc_flush(spa_t *spa, boolean_t retry);
 332 void arc_flush_async(spa_t *spa);
 333 void arc_tempreserve_clear(uint64_t reserve);
 334 int arc_tempreserve_space(spa_t *spa, uint64_t reserve, uint64_t txg);
 335 boolean_t arc_async_flush_guid_inuse(uint64_t load_guid);
 336
 337 uint64_t arc_all_memory(void);
 338 uint64_t arc_default_max(uint64_t min, uint64_t allmem);
 339 uint64_t arc_target_bytes(void);
 340 void arc_set_limits(uint64_t);
 341 void arc_init(void);
 342 void arc_fini(void);
 343
 344 /*
 345  * Level 2 ARC
 346  */
 347
 348 void l2arc_add_vdev(spa_t *spa, vdev_t *vd);
 349 void l2arc_remove_vdev(vdev_t *vd);
 350 boolean_t l2arc_vdev_present(vdev_t *vd);
 351 void l2arc_rebuild_vdev(vdev_t *vd, boolean_t reopen);
 352 boolean_t l2arc_range_check_overlap(uint64_t bottom, uint64_t top,
 353     uint64_t check);
 354 void l2arc_init(void);
 355 void l2arc_fini(void);
 356 void l2arc_start(void);
 357 void l2arc_stop(void);
 358 void l2arc_spa_rebuild_start(spa_t *spa);
 359 void l2arc_spa_rebuild_stop(spa_t *spa);
 360
 361 #ifndef _KERNEL
 362 extern boolean_t arc_watch;
 363 #endif
 364
 365 #ifdef  __cplusplus
 366 }
 367 #endif
 368
 369 #endif /* _SYS_ARC_H */