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