usr/src/uts/common/fs/zfs/sys/vdev_impl.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) 2011, 2018 by Delphix. All rights reserved.
  24  */
  25
  26 #ifndef _SYS_VDEV_IMPL_H
  27 #define _SYS_VDEV_IMPL_H
  28
  29 #include <sys/avl.h>
  30 #include <sys/bpobj.h>
  31 #include <sys/dmu.h>
  32 #include <sys/metaslab.h>
  33 #include <sys/nvpair.h>
  34 #include <sys/space_map.h>
  35 #include <sys/vdev.h>
  36 #include <sys/dkio.h>
  37 #include <sys/uberblock_impl.h>
  38 #include <sys/vdev_indirect_mapping.h>
  39 #include <sys/vdev_indirect_births.h>
  40 #include <sys/vdev_removal.h>
  41
  42 #ifdef  __cplusplus
  43 extern "C" {
  44 #endif
  45
  46 /*
  47  * Virtual device descriptors.
  48  *
  49  * All storage pool operations go through the virtual device framework,
  50  * which provides data replication and I/O scheduling.
  51  */
  52
  53 /*
  54  * Forward declarations that lots of things need.
  55  */
  56 typedef struct vdev_queue vdev_queue_t;
  57 typedef struct vdev_cache vdev_cache_t;
  58 typedef struct vdev_cache_entry vdev_cache_entry_t;
  59 struct abd;
  60
  61 extern int zfs_vdev_queue_depth_pct;
  62 extern int zfs_vdev_def_queue_depth;
  63 extern uint32_t zfs_vdev_async_write_max_active;
  64
  65 /*
  66  * Virtual device operations
  67  */
  68 typedef int     vdev_open_func_t(vdev_t *vd, uint64_t *size, uint64_t *max_size,
  69     uint64_t *ashift);
  70 typedef void    vdev_close_func_t(vdev_t *vd);
  71 typedef uint64_t vdev_asize_func_t(vdev_t *vd, uint64_t psize);
  72 typedef void    vdev_io_start_func_t(zio_t *zio);
  73 typedef void    vdev_io_done_func_t(zio_t *zio);
  74 typedef void    vdev_state_change_func_t(vdev_t *vd, int, int);
  75 typedef void    vdev_hold_func_t(vdev_t *vd);
  76 typedef void    vdev_rele_func_t(vdev_t *vd);
  77
  78 typedef void    vdev_remap_cb_t(uint64_t inner_offset, vdev_t *vd,
  79     uint64_t offset, uint64_t size, void *arg);
  80 typedef void    vdev_remap_func_t(vdev_t *vd, uint64_t offset, uint64_t size,
  81     vdev_remap_cb_t callback, void *arg);
  82
  83 typedef struct vdev_ops {
  84         vdev_open_func_t                *vdev_op_open;
  85         vdev_close_func_t               *vdev_op_close;
  86         vdev_asize_func_t               *vdev_op_asize;
  87         vdev_io_start_func_t            *vdev_op_io_start;
  88         vdev_io_done_func_t             *vdev_op_io_done;
  89         vdev_state_change_func_t        *vdev_op_state_change;
  90         vdev_hold_func_t                *vdev_op_hold;
  91         vdev_rele_func_t                *vdev_op_rele;
  92         vdev_remap_func_t               *vdev_op_remap;
  93         char                            vdev_op_type[16];
  94         boolean_t                       vdev_op_leaf;
  95 } vdev_ops_t;
  96
  97 /*
  98  * Virtual device properties
  99  */
 100 struct vdev_cache_entry {
 101         struct abd      *ve_abd;
 102         uint64_t        ve_offset;
 103         uint64_t        ve_lastused;
 104         avl_node_t      ve_offset_node;
 105         avl_node_t      ve_lastused_node;
 106         uint32_t        ve_hits;
 107         uint16_t        ve_missed_update;
 108         zio_t           *ve_fill_io;
 109 };
 110
 111 struct vdev_cache {
 112         avl_tree_t      vc_offset_tree;
 113         avl_tree_t      vc_lastused_tree;
 114         kmutex_t        vc_lock;
 115 };
 116
 117 typedef struct vdev_queue_class {
 118         uint32_t        vqc_active;
 119
 120         /*
 121          * Sorted by offset or timestamp, depending on if the queue is
 122          * LBA-ordered vs FIFO.
 123          */
 124         avl_tree_t      vqc_queued_tree;
 125 } vdev_queue_class_t;
 126
 127 struct vdev_queue {
 128         vdev_t          *vq_vdev;
 129         vdev_queue_class_t vq_class[ZIO_PRIORITY_NUM_QUEUEABLE];
 130         avl_tree_t      vq_active_tree;
 131         avl_tree_t      vq_read_offset_tree;
 132         avl_tree_t      vq_write_offset_tree;
 133         uint64_t        vq_last_offset;
 134         hrtime_t        vq_io_complete_ts; /* time last i/o completed */
 135         kmutex_t        vq_lock;
 136 };
 137
 138 /*
 139  * On-disk indirect vdev state.
 140  *
 141  * An indirect vdev is described exclusively in the MOS config of a pool.
 142  * The config for an indirect vdev includes several fields, which are
 143  * accessed in memory by a vdev_indirect_config_t.
 144  */
 145 typedef struct vdev_indirect_config {
 146         /*
 147          * Object (in MOS) which contains the indirect mapping. This object
 148          * contains an array of vdev_indirect_mapping_entry_phys_t ordered by
 149          * vimep_src. The bonus buffer for this object is a
 150          * vdev_indirect_mapping_phys_t. This object is allocated when a vdev
 151          * removal is initiated.
 152          *
 153          * Note that this object can be empty if none of the data on the vdev
 154          * has been copied yet.
 155          */
 156         uint64_t        vic_mapping_object;
 157
 158         /*
 159          * Object (in MOS) which contains the birth times for the mapping
 160          * entries. This object contains an array of
 161          * vdev_indirect_birth_entry_phys_t sorted by vibe_offset. The bonus
 162          * buffer for this object is a vdev_indirect_birth_phys_t. This object
 163          * is allocated when a vdev removal is initiated.
 164          *
 165          * Note that this object can be empty if none of the vdev has yet been
 166          * copied.
 167          */
 168         uint64_t        vic_births_object;
 169
 170         /*
 171          * This is the vdev ID which was removed previous to this vdev, or
 172          * UINT64_MAX if there are no previously removed vdevs.
 173          */
 174         uint64_t        vic_prev_indirect_vdev;
 175 } vdev_indirect_config_t;
 176
 177 /*
 178  * Virtual device descriptor
 179  */
 180 struct vdev {
 181         /*
 182          * Common to all vdev types.
 183          */
 184         uint64_t        vdev_id;        /* child number in vdev parent  */
 185         uint64_t        vdev_guid;      /* unique ID for this vdev      */
 186         uint64_t        vdev_guid_sum;  /* self guid + all child guids  */
 187         uint64_t        vdev_orig_guid; /* orig. guid prior to remove   */
 188         uint64_t        vdev_asize;     /* allocatable device capacity  */
 189         uint64_t        vdev_min_asize; /* min acceptable asize         */
 190         uint64_t        vdev_max_asize; /* max acceptable asize         */
 191         uint64_t        vdev_ashift;    /* block alignment shift        */
 192         uint64_t        vdev_state;     /* see VDEV_STATE_* #defines    */
 193         uint64_t        vdev_prevstate; /* used when reopening a vdev   */
 194         vdev_ops_t      *vdev_ops;      /* vdev operations              */
 195         spa_t           *vdev_spa;      /* spa for this vdev            */
 196         void            *vdev_tsd;      /* type-specific data           */
 197         vnode_t         *vdev_name_vp;  /* vnode for pathname           */
 198         vnode_t         *vdev_devid_vp; /* vnode for devid              */
 199         vdev_t          *vdev_top;      /* top-level vdev               */
 200         vdev_t          *vdev_parent;   /* parent vdev                  */
 201         vdev_t          **vdev_child;   /* array of children            */
 202         uint64_t        vdev_children;  /* number of children           */
 203         vdev_stat_t     vdev_stat;      /* virtual device statistics    */
 204         boolean_t       vdev_expanding; /* expand the vdev?             */
 205         boolean_t       vdev_reopening; /* reopen in progress?          */
 206         int             vdev_open_error; /* error on last open          */
 207         kthread_t       *vdev_open_thread; /* thread opening children   */
 208         uint64_t        vdev_crtxg;     /* txg when top-level was added */
 209
 210         /*
 211          * Top-level vdev state.
 212          */
 213         uint64_t        vdev_ms_array;  /* metaslab array object        */
 214         uint64_t        vdev_ms_shift;  /* metaslab size shift          */
 215         uint64_t        vdev_ms_count;  /* number of metaslabs          */
 216         metaslab_group_t *vdev_mg;      /* metaslab group               */
 217         metaslab_t      **vdev_ms;      /* metaslab array               */
 218         txg_list_t      vdev_ms_list;   /* per-txg dirty metaslab lists */
 219         txg_list_t      vdev_dtl_list;  /* per-txg dirty DTL lists      */
 220         txg_node_t      vdev_txg_node;  /* per-txg dirty vdev linkage   */
 221         boolean_t       vdev_remove_wanted; /* async remove wanted?     */
 222         boolean_t       vdev_probe_wanted; /* async probe wanted?       */
 223         list_node_t     vdev_config_dirty_node; /* config dirty list    */
 224         list_node_t     vdev_state_dirty_node; /* state dirty list      */
 225         uint64_t        vdev_deflate_ratio; /* deflation ratio (x512)   */
 226         uint64_t        vdev_islog;     /* is an intent log device      */
 227         uint64_t        vdev_removing;  /* device is being removed?     */
 228         boolean_t       vdev_ishole;    /* is a hole in the namespace   */
 229         kmutex_t        vdev_queue_lock; /* protects vdev_queue_depth   */
 230         uint64_t        vdev_top_zap;
 231
 232         /* pool checkpoint related */
 233         space_map_t     *vdev_checkpoint_sm;    /* contains reserved blocks */
 234
 235         /*
 236          * Values stored in the config for an indirect or removing vdev.
 237          */
 238         vdev_indirect_config_t  vdev_indirect_config;
 239
 240         /*
 241          * The vdev_indirect_rwlock protects the vdev_indirect_mapping
 242          * pointer from changing on indirect vdevs (when it is condensed).
 243          * Note that removing (not yet indirect) vdevs have different
 244          * access patterns (the mapping is not accessed from open context,
 245          * e.g. from zio_read) and locking strategy (e.g. svr_lock).
 246          */
 247         krwlock_t vdev_indirect_rwlock;
 248         vdev_indirect_mapping_t *vdev_indirect_mapping;
 249         vdev_indirect_births_t *vdev_indirect_births;
 250
 251         /*
 252          * In memory data structures used to manage the obsolete sm, for
 253          * indirect or removing vdevs.
 254          *
 255          * The vdev_obsolete_segments is the in-core record of the segments
 256          * that are no longer referenced anywhere in the pool (due to
 257          * being freed or remapped and not referenced by any snapshots).
 258          * During a sync, segments are added to vdev_obsolete_segments
 259          * via vdev_indirect_mark_obsolete(); at the end of each sync
 260          * pass, this is appended to vdev_obsolete_sm via
 261          * vdev_indirect_sync_obsolete().  The vdev_obsolete_lock
 262          * protects against concurrent modifications of vdev_obsolete_segments
 263          * from multiple zio threads.
 264          */
 265         kmutex_t        vdev_obsolete_lock;
 266         range_tree_t    *vdev_obsolete_segments;
 267         space_map_t     *vdev_obsolete_sm;
 268
 269         /*
 270          * The queue depth parameters determine how many async writes are
 271          * still pending (i.e. allocated by net yet issued to disk) per
 272          * top-level (vdev_async_write_queue_depth) and the maximum allowed
 273          * (vdev_max_async_write_queue_depth). These values only apply to
 274          * top-level vdevs.
 275          */
 276         uint64_t        vdev_async_write_queue_depth;
 277         uint64_t        vdev_max_async_write_queue_depth;
 278
 279         /*
 280          * Leaf vdev state.
 281          */
 282         range_tree_t    *vdev_dtl[DTL_TYPES]; /* dirty time logs        */
 283         space_map_t     *vdev_dtl_sm;   /* dirty time log space map     */
 284         txg_node_t      vdev_dtl_node;  /* per-txg dirty DTL linkage    */
 285         uint64_t        vdev_dtl_object; /* DTL object                  */
 286         uint64_t        vdev_psize;     /* physical device capacity     */
 287         uint64_t        vdev_wholedisk; /* true if this is a whole disk */
 288         uint64_t        vdev_offline;   /* persistent offline state     */
 289         uint64_t        vdev_faulted;   /* persistent faulted state     */
 290         uint64_t        vdev_degraded;  /* persistent degraded state    */
 291         uint64_t        vdev_removed;   /* persistent removed state     */
 292         uint64_t        vdev_resilver_txg; /* persistent resilvering state */
 293         uint64_t        vdev_nparity;   /* number of parity devices for raidz */
 294         char            *vdev_path;     /* vdev path (if any)           */
 295         char            *vdev_devid;    /* vdev devid (if any)          */
 296         char            *vdev_physpath; /* vdev device path (if any)    */
 297         char            *vdev_fru;      /* physical FRU location        */
 298         uint64_t        vdev_not_present; /* not present during import  */
 299         uint64_t        vdev_unspare;   /* unspare when resilvering done */
 300         boolean_t       vdev_nowritecache; /* true if flushwritecache failed */
 301         boolean_t       vdev_checkremove; /* temporary online test      */
 302         boolean_t       vdev_forcefault; /* force online fault          */
 303         boolean_t       vdev_splitting; /* split or repair in progress  */
 304         boolean_t       vdev_delayed_close; /* delayed device close?    */
 305         boolean_t       vdev_tmpoffline; /* device taken offline temporarily? */
 306         boolean_t       vdev_detached;  /* device detached?             */
 307         boolean_t       vdev_cant_read; /* vdev is failing all reads    */
 308         boolean_t       vdev_cant_write; /* vdev is failing all writes  */
 309         boolean_t       vdev_isspare;   /* was a hot spare              */
 310         boolean_t       vdev_isl2cache; /* was a l2cache device         */
 311         vdev_queue_t    vdev_queue;     /* I/O deadline schedule queue  */
 312         vdev_cache_t    vdev_cache;     /* physical block cache         */
 313         spa_aux_vdev_t  *vdev_aux;      /* for l2cache and spares vdevs */
 314         zio_t           *vdev_probe_zio; /* root of current probe       */
 315         vdev_aux_t      vdev_label_aux; /* on-disk aux state            */
 316         uint64_t        vdev_leaf_zap;
 317
 318         /*
 319          * For DTrace to work in userland (libzpool) context, these fields must
 320          * remain at the end of the structure.  DTrace will use the kernel's
 321          * CTF definition for 'struct vdev', and since the size of a kmutex_t is
 322          * larger in userland, the offsets for the rest of the fields would be
 323          * incorrect.
 324          */
 325         kmutex_t        vdev_dtl_lock;  /* vdev_dtl_{map,resilver}      */
 326         kmutex_t        vdev_stat_lock; /* vdev_stat                    */
 327         kmutex_t        vdev_probe_lock; /* protects vdev_probe_zio     */
 328 };
 329
 330 #define VDEV_RAIDZ_MAXPARITY    3
 331
 332 #define VDEV_PAD_SIZE           (8 << 10)
 333 /* 2 padding areas (vl_pad1 and vl_pad2) to skip */
 334 #define VDEV_SKIP_SIZE          VDEV_PAD_SIZE * 2
 335 #define VDEV_PHYS_SIZE          (112 << 10)
 336 #define VDEV_UBERBLOCK_RING     (128 << 10)
 337
 338 /* The largest uberblock we support is 8k. */
 339 #define MAX_UBERBLOCK_SHIFT (13)
 340 #define VDEV_UBERBLOCK_SHIFT(vd)        \
 341         MIN(MAX((vd)->vdev_top->vdev_ashift, UBERBLOCK_SHIFT), \
 342             MAX_UBERBLOCK_SHIFT)
 343 #define VDEV_UBERBLOCK_COUNT(vd)        \
 344         (VDEV_UBERBLOCK_RING >> VDEV_UBERBLOCK_SHIFT(vd))
 345 #define VDEV_UBERBLOCK_OFFSET(vd, n)    \
 346         offsetof(vdev_label_t, vl_uberblock[(n) << VDEV_UBERBLOCK_SHIFT(vd)])
 347 #define VDEV_UBERBLOCK_SIZE(vd)         (1ULL << VDEV_UBERBLOCK_SHIFT(vd))
 348
 349 typedef struct vdev_phys {
 350         char            vp_nvlist[VDEV_PHYS_SIZE - sizeof (zio_eck_t)];
 351         zio_eck_t       vp_zbt;
 352 } vdev_phys_t;
 353
 354 typedef struct vdev_label {
 355         char            vl_pad1[VDEV_PAD_SIZE];                 /*  8K */
 356         char            vl_pad2[VDEV_PAD_SIZE];                 /*  8K */
 357         vdev_phys_t     vl_vdev_phys;                           /* 112K */
 358         char            vl_uberblock[VDEV_UBERBLOCK_RING];      /* 128K */
 359 } vdev_label_t;                                                 /* 256K total */
 360
 361 /*
 362  * vdev_dirty() flags
 363  */
 364 #define VDD_METASLAB    0x01
 365 #define VDD_DTL         0x02
 366
 367 /* Offset of embedded boot loader region on each label */
 368 #define VDEV_BOOT_OFFSET        (2 * sizeof (vdev_label_t))
 369 /*
 370  * Size of embedded boot loader region on each label.
 371  * The total size of the first two labels plus the boot area is 4MB.
 372  */
 373 #define VDEV_BOOT_SIZE          (7ULL << 19)                    /* 3.5M */
 374
 375 /*
 376  * Size of label regions at the start and end of each leaf device.
 377  */
 378 #define VDEV_LABEL_START_SIZE   (2 * sizeof (vdev_label_t) + VDEV_BOOT_SIZE)
 379 #define VDEV_LABEL_END_SIZE     (2 * sizeof (vdev_label_t))
 380 #define VDEV_LABELS             4
 381 #define VDEV_BEST_LABEL         VDEV_LABELS
 382
 383 #define VDEV_ALLOC_LOAD         0
 384 #define VDEV_ALLOC_ADD          1
 385 #define VDEV_ALLOC_SPARE        2
 386 #define VDEV_ALLOC_L2CACHE      3
 387 #define VDEV_ALLOC_ROOTPOOL     4
 388 #define VDEV_ALLOC_SPLIT        5
 389 #define VDEV_ALLOC_ATTACH       6
 390
 391 /*
 392  * Allocate or free a vdev
 393  */
 394 extern vdev_t *vdev_alloc_common(spa_t *spa, uint_t id, uint64_t guid,
 395     vdev_ops_t *ops);
 396 extern int vdev_alloc(spa_t *spa, vdev_t **vdp, nvlist_t *config,
 397     vdev_t *parent, uint_t id, int alloctype);
 398 extern void vdev_free(vdev_t *vd);
 399
 400 /*
 401  * Add or remove children and parents
 402  */
 403 extern void vdev_add_child(vdev_t *pvd, vdev_t *cvd);
 404 extern void vdev_remove_child(vdev_t *pvd, vdev_t *cvd);
 405 extern void vdev_compact_children(vdev_t *pvd);
 406 extern vdev_t *vdev_add_parent(vdev_t *cvd, vdev_ops_t *ops);
 407 extern void vdev_remove_parent(vdev_t *cvd);
 408
 409 /*
 410  * vdev sync load and sync
 411  */
 412 extern boolean_t vdev_log_state_valid(vdev_t *vd);
 413 extern int vdev_load(vdev_t *vd);
 414 extern int vdev_dtl_load(vdev_t *vd);
 415 extern void vdev_sync(vdev_t *vd, uint64_t txg);
 416 extern void vdev_sync_done(vdev_t *vd, uint64_t txg);
 417 extern void vdev_dirty(vdev_t *vd, int flags, void *arg, uint64_t txg);
 418 extern void vdev_dirty_leaves(vdev_t *vd, int flags, uint64_t txg);
 419
 420 /*
 421  * Available vdev types.
 422  */
 423 extern vdev_ops_t vdev_root_ops;
 424 extern vdev_ops_t vdev_mirror_ops;
 425 extern vdev_ops_t vdev_replacing_ops;
 426 extern vdev_ops_t vdev_raidz_ops;
 427 extern vdev_ops_t vdev_disk_ops;
 428 extern vdev_ops_t vdev_file_ops;
 429 extern vdev_ops_t vdev_missing_ops;
 430 extern vdev_ops_t vdev_hole_ops;
 431 extern vdev_ops_t vdev_spare_ops;
 432 extern vdev_ops_t vdev_indirect_ops;
 433
 434 /*
 435  * Common size functions
 436  */
 437 extern uint64_t vdev_default_asize(vdev_t *vd, uint64_t psize);
 438 extern uint64_t vdev_get_min_asize(vdev_t *vd);
 439 extern void vdev_set_min_asize(vdev_t *vd);
 440
 441 /*
 442  * Global variables
 443  */
 444 extern int vdev_standard_sm_blksz;
 445 /* zdb uses this tunable, so it must be declared here to make lint happy. */
 446 extern int zfs_vdev_cache_size;
 447
 448 /*
 449  * Functions from vdev_indirect.c
 450  */
 451 extern void vdev_indirect_sync_obsolete(vdev_t *vd, dmu_tx_t *tx);
 452 extern boolean_t vdev_indirect_should_condense(vdev_t *vd);
 453 extern void spa_condense_indirect_start_sync(vdev_t *vd, dmu_tx_t *tx);
 454 extern int vdev_obsolete_sm_object(vdev_t *vd);
 455 extern boolean_t vdev_obsolete_counts_are_precise(vdev_t *vd);
 456
 457 /*
 458  * Other miscellaneous functions
 459  */
 460 int vdev_checkpoint_sm_object(vdev_t *vd);
 461
 462 /*
 463  * The vdev_buf_t is used to translate between zio_t and buf_t, and back again.
 464  */
 465 typedef struct vdev_buf {
 466         buf_t   vb_buf;         /* buffer that describes the io */
 467         zio_t   *vb_io;         /* pointer back to the original zio_t */
 468 } vdev_buf_t;
 469
 470 #ifdef  __cplusplus
 471 }
 472 #endif
 473
 474 #endif  /* _SYS_VDEV_IMPL_H */