include/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, 2015 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 #include <sys/zfs_ratelimit.h>
  42
  43 #ifdef  __cplusplus
  44 extern "C" {
  45 #endif
  46
  47 /*
  48  * Virtual device descriptors.
  49  *
  50  * All storage pool operations go through the virtual device framework,
  51  * which provides data replication and I/O scheduling.
  52  */
  53
  54 /*
  55  * Forward declarations that lots of things need.
  56  */
  57 typedef struct vdev_queue vdev_queue_t;
  58 typedef struct vdev_cache vdev_cache_t;
  59 typedef struct vdev_cache_entry vdev_cache_entry_t;
  60 struct abd;
  61
  62 extern int zfs_vdev_queue_depth_pct;
  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 boolean_t vdev_need_resilver_func_t(vdev_t *vd, uint64_t, size_t);
  76 typedef void    vdev_hold_func_t(vdev_t *vd);
  77 typedef void    vdev_rele_func_t(vdev_t *vd);
  78
  79 typedef void    vdev_remap_cb_t(uint64_t inner_offset, vdev_t *vd,
  80     uint64_t offset, uint64_t size, void *arg);
  81 typedef void    vdev_remap_func_t(vdev_t *vd, uint64_t offset, uint64_t size,
  82     vdev_remap_cb_t callback, void *arg);
  83
  84 typedef const struct vdev_ops {
  85         vdev_open_func_t                *vdev_op_open;
  86         vdev_close_func_t               *vdev_op_close;
  87         vdev_asize_func_t               *vdev_op_asize;
  88         vdev_io_start_func_t            *vdev_op_io_start;
  89         vdev_io_done_func_t             *vdev_op_io_done;
  90         vdev_state_change_func_t        *vdev_op_state_change;
  91         vdev_need_resilver_func_t       *vdev_op_need_resilver;
  92         vdev_hold_func_t                *vdev_op_hold;
  93         vdev_rele_func_t                *vdev_op_rele;
  94         vdev_remap_func_t               *vdev_op_remap;
  95         char                            vdev_op_type[16];
  96         boolean_t                       vdev_op_leaf;
  97 } vdev_ops_t;
  98
  99 /*
 100  * Virtual device properties
 101  */
 102 struct vdev_cache_entry {
 103         struct abd      *ve_abd;
 104         uint64_t        ve_offset;
 105         clock_t         ve_lastused;
 106         avl_node_t      ve_offset_node;
 107         avl_node_t      ve_lastused_node;
 108         uint32_t        ve_hits;
 109         uint16_t        ve_missed_update;
 110         zio_t           *ve_fill_io;
 111 };
 112
 113 struct vdev_cache {
 114         avl_tree_t      vc_offset_tree;
 115         avl_tree_t      vc_lastused_tree;
 116         kmutex_t        vc_lock;
 117 };
 118
 119 typedef struct vdev_queue_class {
 120         uint32_t        vqc_active;
 121
 122         /*
 123          * Sorted by offset or timestamp, depending on if the queue is
 124          * LBA-ordered vs FIFO.
 125          */
 126         avl_tree_t      vqc_queued_tree;
 127 } vdev_queue_class_t;
 128
 129 struct vdev_queue {
 130         vdev_t          *vq_vdev;
 131         vdev_queue_class_t vq_class[ZIO_PRIORITY_NUM_QUEUEABLE];
 132         avl_tree_t      vq_active_tree;
 133         avl_tree_t      vq_read_offset_tree;
 134         avl_tree_t      vq_write_offset_tree;
 135         uint64_t        vq_last_offset;
 136         hrtime_t        vq_io_complete_ts; /* time last i/o completed */
 137         hrtime_t        vq_io_delta_ts;
 138         zio_t           vq_io_search; /* used as local for stack reduction */
 139         kmutex_t        vq_lock;
 140 };
 141
 142 /*
 143  * On-disk indirect vdev state.
 144  *
 145  * An indirect vdev is described exclusively in the MOS config of a pool.
 146  * The config for an indirect vdev includes several fields, which are
 147  * accessed in memory by a vdev_indirect_config_t.
 148  */
 149 typedef struct vdev_indirect_config {
 150         /*
 151          * Object (in MOS) which contains the indirect mapping. This object
 152          * contains an array of vdev_indirect_mapping_entry_phys_t ordered by
 153          * vimep_src. The bonus buffer for this object is a
 154          * vdev_indirect_mapping_phys_t. This object is allocated when a vdev
 155          * removal is initiated.
 156          *
 157          * Note that this object can be empty if none of the data on the vdev
 158          * has been copied yet.
 159          */
 160         uint64_t        vic_mapping_object;
 161
 162         /*
 163          * Object (in MOS) which contains the birth times for the mapping
 164          * entries. This object contains an array of
 165          * vdev_indirect_birth_entry_phys_t sorted by vibe_offset. The bonus
 166          * buffer for this object is a vdev_indirect_birth_phys_t. This object
 167          * is allocated when a vdev removal is initiated.
 168          *
 169          * Note that this object can be empty if none of the vdev has yet been
 170          * copied.
 171          */
 172         uint64_t        vic_births_object;
 173
 174         /*
 175          * This is the vdev ID which was removed previous to this vdev, or
 176          * UINT64_MAX if there are no previously removed vdevs.
 177          */
 178         uint64_t        vic_prev_indirect_vdev;
 179 } vdev_indirect_config_t;
 180
 181 /*
 182  * Virtual device descriptor
 183  */
 184 struct vdev {
 185         /*
 186          * Common to all vdev types.
 187          */
 188         uint64_t        vdev_id;        /* child number in vdev parent  */
 189         uint64_t        vdev_guid;      /* unique ID for this vdev      */
 190         uint64_t        vdev_guid_sum;  /* self guid + all child guids  */
 191         uint64_t        vdev_orig_guid; /* orig. guid prior to remove   */
 192         uint64_t        vdev_asize;     /* allocatable device capacity  */
 193         uint64_t        vdev_min_asize; /* min acceptable asize         */
 194         uint64_t        vdev_max_asize; /* max acceptable asize         */
 195         uint64_t        vdev_ashift;    /* block alignment shift        */
 196         uint64_t        vdev_state;     /* see VDEV_STATE_* #defines    */
 197         uint64_t        vdev_prevstate; /* used when reopening a vdev   */
 198         vdev_ops_t      *vdev_ops;      /* vdev operations              */
 199         spa_t           *vdev_spa;      /* spa for this vdev            */
 200         void            *vdev_tsd;      /* type-specific data           */
 201         vnode_t         *vdev_name_vp;  /* vnode for pathname           */
 202         vnode_t         *vdev_devid_vp; /* vnode for devid              */
 203         vdev_t          *vdev_top;      /* top-level vdev               */
 204         vdev_t          *vdev_parent;   /* parent vdev                  */
 205         vdev_t          **vdev_child;   /* array of children            */
 206         uint64_t        vdev_children;  /* number of children           */
 207         vdev_stat_t     vdev_stat;      /* virtual device statistics    */
 208         vdev_stat_ex_t  vdev_stat_ex;   /* extended statistics          */
 209         boolean_t       vdev_expanding; /* expand the vdev?             */
 210         boolean_t       vdev_reopening; /* reopen in progress?          */
 211         boolean_t       vdev_nonrot;    /* true if solid state          */
 212         int             vdev_open_error; /* error on last open          */
 213         kthread_t       *vdev_open_thread; /* thread opening children   */
 214         uint64_t        vdev_crtxg;     /* txg when top-level was added */
 215
 216         /*
 217          * Top-level vdev state.
 218          */
 219         uint64_t        vdev_ms_array;  /* metaslab array object        */
 220         uint64_t        vdev_ms_shift;  /* metaslab size shift          */
 221         uint64_t        vdev_ms_count;  /* number of metaslabs          */
 222         metaslab_group_t *vdev_mg;      /* metaslab group               */
 223         metaslab_t      **vdev_ms;      /* metaslab array               */
 224         uint64_t        vdev_pending_fastwrite; /* allocated fastwrites */
 225         txg_list_t      vdev_ms_list;   /* per-txg dirty metaslab lists */
 226         txg_list_t      vdev_dtl_list;  /* per-txg dirty DTL lists      */
 227         txg_node_t      vdev_txg_node;  /* per-txg dirty vdev linkage   */
 228         boolean_t       vdev_remove_wanted; /* async remove wanted?     */
 229         boolean_t       vdev_probe_wanted; /* async probe wanted?       */
 230         list_node_t     vdev_config_dirty_node; /* config dirty list    */
 231         list_node_t     vdev_state_dirty_node; /* state dirty list      */
 232         uint64_t        vdev_deflate_ratio; /* deflation ratio (x512)   */
 233         uint64_t        vdev_islog;     /* is an intent log device      */
 234         uint64_t        vdev_removing;  /* device is being removed?     */
 235         boolean_t       vdev_ishole;    /* is a hole in the namespace   */
 236         kmutex_t        vdev_queue_lock; /* protects vdev_queue_depth   */
 237         uint64_t        vdev_top_zap;
 238
 239         /*
 240          * Values stored in the config for an indirect or removing vdev.
 241          */
 242         vdev_indirect_config_t  vdev_indirect_config;
 243
 244         /*
 245          * The vdev_indirect_rwlock protects the vdev_indirect_mapping
 246          * pointer from changing on indirect vdevs (when it is condensed).
 247          * Note that removing (not yet indirect) vdevs have different
 248          * access patterns (the mapping is not accessed from open context,
 249          * e.g. from zio_read) and locking strategy (e.g. svr_lock).
 250          */
 251         krwlock_t vdev_indirect_rwlock;
 252         vdev_indirect_mapping_t *vdev_indirect_mapping;
 253         vdev_indirect_births_t *vdev_indirect_births;
 254
 255         /*
 256          * In memory data structures used to manage the obsolete sm, for
 257          * indirect or removing vdevs.
 258          *
 259          * The vdev_obsolete_segments is the in-core record of the segments
 260          * that are no longer referenced anywhere in the pool (due to
 261          * being freed or remapped and not referenced by any snapshots).
 262          * During a sync, segments are added to vdev_obsolete_segments
 263          * via vdev_indirect_mark_obsolete(); at the end of each sync
 264          * pass, this is appended to vdev_obsolete_sm via
 265          * vdev_indirect_sync_obsolete().  The vdev_obsolete_lock
 266          * protects against concurrent modifications of vdev_obsolete_segments
 267          * from multiple zio threads.
 268          */
 269         kmutex_t        vdev_obsolete_lock;
 270         range_tree_t    *vdev_obsolete_segments;
 271         space_map_t     *vdev_obsolete_sm;
 272
 273         /*
 274          * The queue depth parameters determine how many async writes are
 275          * still pending (i.e. allocated by net yet issued to disk) per
 276          * top-level (vdev_async_write_queue_depth) and the maximum allowed
 277          * (vdev_max_async_write_queue_depth). These values only apply to
 278          * top-level vdevs.
 279          */
 280         uint64_t        vdev_async_write_queue_depth;
 281         uint64_t        vdev_max_async_write_queue_depth;
 282
 283         /*
 284          * Protects the vdev_scan_io_queue field itself as well as the
 285          * structure's contents (when present).
 286          */
 287         kmutex_t                        vdev_scan_io_queue_lock;
 288         struct dsl_scan_io_queue        *vdev_scan_io_queue;
 289
 290         /*
 291          * Leaf vdev state.
 292          */
 293         range_tree_t    *vdev_dtl[DTL_TYPES]; /* dirty time logs        */
 294         space_map_t     *vdev_dtl_sm;   /* dirty time log space map     */
 295         txg_node_t      vdev_dtl_node;  /* per-txg dirty DTL linkage    */
 296         uint64_t        vdev_dtl_object; /* DTL object                  */
 297         uint64_t        vdev_psize;     /* physical device capacity     */
 298         uint64_t        vdev_wholedisk; /* true if this is a whole disk */
 299         uint64_t        vdev_offline;   /* persistent offline state     */
 300         uint64_t        vdev_faulted;   /* persistent faulted state     */
 301         uint64_t        vdev_degraded;  /* persistent degraded state    */
 302         uint64_t        vdev_removed;   /* persistent removed state     */
 303         uint64_t        vdev_resilver_txg; /* persistent resilvering state */
 304         uint64_t        vdev_nparity;   /* number of parity devices for raidz */
 305         char            *vdev_path;     /* vdev path (if any)           */
 306         char            *vdev_devid;    /* vdev devid (if any)          */
 307         char            *vdev_physpath; /* vdev device path (if any)    */
 308         char            *vdev_enc_sysfs_path;   /* enclosure sysfs path */
 309         char            *vdev_fru;      /* physical FRU location        */
 310         uint64_t        vdev_not_present; /* not present during import  */
 311         uint64_t        vdev_unspare;   /* unspare when resilvering done */
 312         boolean_t       vdev_nowritecache; /* true if flushwritecache failed */
 313         boolean_t       vdev_checkremove; /* temporary online test      */
 314         boolean_t       vdev_forcefault; /* force online fault          */
 315         boolean_t       vdev_splitting; /* split or repair in progress  */
 316         boolean_t       vdev_delayed_close; /* delayed device close?    */
 317         boolean_t       vdev_tmpoffline; /* device taken offline temporarily? */
 318         boolean_t       vdev_detached;  /* device detached?             */
 319         boolean_t       vdev_cant_read; /* vdev is failing all reads    */
 320         boolean_t       vdev_cant_write; /* vdev is failing all writes  */
 321         boolean_t       vdev_isspare;   /* was a hot spare              */
 322         boolean_t       vdev_isl2cache; /* was a l2cache device         */
 323         boolean_t       vdev_copy_uberblocks;  /* post expand copy uberblocks */
 324         vdev_queue_t    vdev_queue;     /* I/O deadline schedule queue  */
 325         vdev_cache_t    vdev_cache;     /* physical block cache         */
 326         spa_aux_vdev_t  *vdev_aux;      /* for l2cache and spares vdevs */
 327         zio_t           *vdev_probe_zio; /* root of current probe       */
 328         vdev_aux_t      vdev_label_aux; /* on-disk aux state            */
 329         uint64_t        vdev_leaf_zap;
 330         hrtime_t        vdev_mmp_pending; /* 0 if write finished        */
 331         uint64_t        vdev_mmp_kstat_id;      /* to find kstat entry */
 332
 333         /*
 334          * For DTrace to work in userland (libzpool) context, these fields must
 335          * remain at the end of the structure.  DTrace will use the kernel's
 336          * CTF definition for 'struct vdev', and since the size of a kmutex_t is
 337          * larger in userland, the offsets for the rest of the fields would be
 338          * incorrect.
 339          */
 340         kmutex_t        vdev_dtl_lock;  /* vdev_dtl_{map,resilver}      */
 341         kmutex_t        vdev_stat_lock; /* vdev_stat                    */
 342         kmutex_t        vdev_probe_lock; /* protects vdev_probe_zio     */
 343
 344         /*
 345          * We rate limit ZIO delay and ZIO checksum events, since they
 346          * can flood ZED with tons of events when a drive is acting up.
 347          */
 348         zfs_ratelimit_t vdev_delay_rl;
 349         zfs_ratelimit_t vdev_checksum_rl;
 350 };
 351
 352 #define VDEV_RAIDZ_MAXPARITY    3
 353
 354 #define VDEV_PAD_SIZE           (8 << 10)
 355 /* 2 padding areas (vl_pad1 and vl_pad2) to skip */
 356 #define VDEV_SKIP_SIZE          VDEV_PAD_SIZE * 2
 357 #define VDEV_PHYS_SIZE          (112 << 10)
 358 #define VDEV_UBERBLOCK_RING     (128 << 10)
 359
 360 /*
 361  * MMP blocks occupy the last MMP_BLOCKS_PER_LABEL slots in the uberblock
 362  * ring when MMP is enabled.
 363  */
 364 #define MMP_BLOCKS_PER_LABEL    1
 365
 366 /* The largest uberblock we support is 8k. */
 367 #define MAX_UBERBLOCK_SHIFT (13)
 368 #define VDEV_UBERBLOCK_SHIFT(vd)        \
 369         MIN(MAX((vd)->vdev_top->vdev_ashift, UBERBLOCK_SHIFT), \
 370             MAX_UBERBLOCK_SHIFT)
 371 #define VDEV_UBERBLOCK_COUNT(vd)        \
 372         (VDEV_UBERBLOCK_RING >> VDEV_UBERBLOCK_SHIFT(vd))
 373 #define VDEV_UBERBLOCK_OFFSET(vd, n)    \
 374         offsetof(vdev_label_t, vl_uberblock[(n) << VDEV_UBERBLOCK_SHIFT(vd)])
 375 #define VDEV_UBERBLOCK_SIZE(vd)         (1ULL << VDEV_UBERBLOCK_SHIFT(vd))
 376
 377 typedef struct vdev_phys {
 378         char            vp_nvlist[VDEV_PHYS_SIZE - sizeof (zio_eck_t)];
 379         zio_eck_t       vp_zbt;
 380 } vdev_phys_t;
 381
 382 typedef struct vdev_label {
 383         char            vl_pad1[VDEV_PAD_SIZE];                 /*  8K */
 384         char            vl_pad2[VDEV_PAD_SIZE];                 /*  8K */
 385         vdev_phys_t     vl_vdev_phys;                           /* 112K */
 386         char            vl_uberblock[VDEV_UBERBLOCK_RING];      /* 128K */
 387 } vdev_label_t;                                                 /* 256K total */
 388
 389 /*
 390  * vdev_dirty() flags
 391  */
 392 #define VDD_METASLAB    0x01
 393 #define VDD_DTL         0x02
 394
 395 /* Offset of embedded boot loader region on each label */
 396 #define VDEV_BOOT_OFFSET        (2 * sizeof (vdev_label_t))
 397 /*
 398  * Size of embedded boot loader region on each label.
 399  * The total size of the first two labels plus the boot area is 4MB.
 400  */
 401 #define VDEV_BOOT_SIZE          (7ULL << 19)                    /* 3.5M */
 402
 403 /*
 404  * Size of label regions at the start and end of each leaf device.
 405  */
 406 #define VDEV_LABEL_START_SIZE   (2 * sizeof (vdev_label_t) + VDEV_BOOT_SIZE)
 407 #define VDEV_LABEL_END_SIZE     (2 * sizeof (vdev_label_t))
 408 #define VDEV_LABELS             4
 409 #define VDEV_BEST_LABEL         VDEV_LABELS
 410
 411 #define VDEV_ALLOC_LOAD         0
 412 #define VDEV_ALLOC_ADD          1
 413 #define VDEV_ALLOC_SPARE        2
 414 #define VDEV_ALLOC_L2CACHE      3
 415 #define VDEV_ALLOC_ROOTPOOL     4
 416 #define VDEV_ALLOC_SPLIT        5
 417 #define VDEV_ALLOC_ATTACH       6
 418
 419 /*
 420  * Allocate or free a vdev
 421  */
 422 extern vdev_t *vdev_alloc_common(spa_t *spa, uint_t id, uint64_t guid,
 423     vdev_ops_t *ops);
 424 extern int vdev_alloc(spa_t *spa, vdev_t **vdp, nvlist_t *config,
 425     vdev_t *parent, uint_t id, int alloctype);
 426 extern void vdev_free(vdev_t *vd);
 427
 428 /*
 429  * Add or remove children and parents
 430  */
 431 extern void vdev_add_child(vdev_t *pvd, vdev_t *cvd);
 432 extern void vdev_remove_child(vdev_t *pvd, vdev_t *cvd);
 433 extern void vdev_compact_children(vdev_t *pvd);
 434 extern vdev_t *vdev_add_parent(vdev_t *cvd, vdev_ops_t *ops);
 435 extern void vdev_remove_parent(vdev_t *cvd);
 436
 437 /*
 438  * vdev sync load and sync
 439  */
 440 extern void vdev_load_log_state(vdev_t *nvd, vdev_t *ovd);
 441 extern boolean_t vdev_log_state_valid(vdev_t *vd);
 442 extern int vdev_load(vdev_t *vd);
 443 extern int vdev_dtl_load(vdev_t *vd);
 444 extern void vdev_sync(vdev_t *vd, uint64_t txg);
 445 extern void vdev_sync_done(vdev_t *vd, uint64_t txg);
 446 extern void vdev_dirty(vdev_t *vd, int flags, void *arg, uint64_t txg);
 447 extern void vdev_dirty_leaves(vdev_t *vd, int flags, uint64_t txg);
 448
 449 /*
 450  * Available vdev types.
 451  */
 452 extern vdev_ops_t vdev_root_ops;
 453 extern vdev_ops_t vdev_mirror_ops;
 454 extern vdev_ops_t vdev_replacing_ops;
 455 extern vdev_ops_t vdev_raidz_ops;
 456 extern vdev_ops_t vdev_disk_ops;
 457 extern vdev_ops_t vdev_file_ops;
 458 extern vdev_ops_t vdev_missing_ops;
 459 extern vdev_ops_t vdev_hole_ops;
 460 extern vdev_ops_t vdev_spare_ops;
 461 extern vdev_ops_t vdev_indirect_ops;
 462
 463 /*
 464  * Common size functions
 465  */
 466 extern uint64_t vdev_default_asize(vdev_t *vd, uint64_t psize);
 467 extern uint64_t vdev_get_min_asize(vdev_t *vd);
 468 extern void vdev_set_min_asize(vdev_t *vd);
 469
 470 /*
 471  * Global variables
 472  */
 473 /* zdb uses this tunable, so it must be declared here to make lint happy. */
 474 extern int zfs_vdev_cache_size;
 475
 476 /*
 477  * Functions from vdev_indirect.c
 478  */
 479 extern void vdev_indirect_sync_obsolete(vdev_t *vd, dmu_tx_t *tx);
 480 extern boolean_t vdev_indirect_should_condense(vdev_t *vd);
 481 extern void spa_condense_indirect_start_sync(vdev_t *vd, dmu_tx_t *tx);
 482 extern int vdev_obsolete_sm_object(vdev_t *vd);
 483 extern boolean_t vdev_obsolete_counts_are_precise(vdev_t *vd);
 484
 485 #ifdef  __cplusplus
 486 }
 487 #endif
 488
 489 #endif  /* _SYS_VDEV_IMPL_H */