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  */
  24
  25 #ifndef _SYS_VDEV_IMPL_H
  26 #define _SYS_VDEV_IMPL_H
  27
  28 #include <sys/avl.h>
  29 #include <sys/dmu.h>
  30 #include <sys/metaslab.h>
  31 #include <sys/nvpair.h>
  32 #include <sys/space_map.h>
  33 #include <sys/vdev.h>
  34 #include <sys/dkio.h>
  35 #include <sys/uberblock_impl.h>
  36
  37 #ifdef  __cplusplus
  38 extern "C" {
  39 #endif
  40
  41 /*
  42  * Virtual device descriptors.
  43  *
  44  * All storage pool operations go through the virtual device framework,
  45  * which provides data replication and I/O scheduling.
  46  */
  47
  48 /*
  49  * Forward declarations that lots of things need.
  50  */
  51 typedef struct vdev_queue vdev_queue_t;
  52 typedef struct vdev_cache vdev_cache_t;
  53 typedef struct vdev_cache_entry vdev_cache_entry_t;
  54
  55 /*
  56  * Virtual device operations
  57  */
  58 typedef int     vdev_open_func_t(vdev_t *vd, uint64_t *size, uint64_t *ashift);
  59 typedef void    vdev_close_func_t(vdev_t *vd);
  60 typedef uint64_t vdev_asize_func_t(vdev_t *vd, uint64_t psize);
  61 typedef int     vdev_io_start_func_t(zio_t *zio);
  62 typedef void    vdev_io_done_func_t(zio_t *zio);
  63 typedef void    vdev_state_change_func_t(vdev_t *vd, int, int);
  64 typedef void    vdev_hold_func_t(vdev_t *vd);
  65 typedef void    vdev_rele_func_t(vdev_t *vd);
  66
  67 typedef struct vdev_ops {
  68         vdev_open_func_t                *vdev_op_open;
  69         vdev_close_func_t               *vdev_op_close;
  70         vdev_asize_func_t               *vdev_op_asize;
  71         vdev_io_start_func_t            *vdev_op_io_start;
  72         vdev_io_done_func_t             *vdev_op_io_done;
  73         vdev_state_change_func_t        *vdev_op_state_change;
  74         vdev_hold_func_t                *vdev_op_hold;
  75         vdev_rele_func_t                *vdev_op_rele;
  76         char                            vdev_op_type[16];
  77         boolean_t                       vdev_op_leaf;
  78 } vdev_ops_t;
  79
  80 /*
  81  * Virtual device properties
  82  */
  83 struct vdev_cache_entry {
  84         char            *ve_data;
  85         uint64_t        ve_offset;
  86         uint64_t        ve_lastused;
  87         avl_node_t      ve_offset_node;
  88         avl_node_t      ve_lastused_node;
  89         uint32_t        ve_hits;
  90         uint16_t        ve_missed_update;
  91         zio_t           *ve_fill_io;
  92 };
  93
  94 struct vdev_cache {
  95         avl_tree_t      vc_offset_tree;
  96         avl_tree_t      vc_lastused_tree;
  97         kmutex_t        vc_lock;
  98 };
  99
 100 struct vdev_queue {
 101         avl_tree_t      vq_deadline_tree;
 102         avl_tree_t      vq_read_tree;
 103         avl_tree_t      vq_write_tree;
 104         avl_tree_t      vq_pending_tree;
 105         kmutex_t        vq_lock;
 106 };
 107
 108 /*
 109  * Virtual device descriptor
 110  */
 111 struct vdev {
 112         /*
 113          * Common to all vdev types.
 114          */
 115         uint64_t        vdev_id;        /* child number in vdev parent  */
 116         uint64_t        vdev_guid;      /* unique ID for this vdev      */
 117         uint64_t        vdev_guid_sum;  /* self guid + all child guids  */
 118         uint64_t        vdev_orig_guid; /* orig. guid prior to remove   */
 119         uint64_t        vdev_asize;     /* allocatable device capacity  */
 120         uint64_t        vdev_min_asize; /* min acceptable asize         */
 121         uint64_t        vdev_ashift;    /* block alignment shift        */
 122         uint64_t        vdev_state;     /* see VDEV_STATE_* #defines    */
 123         uint64_t        vdev_prevstate; /* used when reopening a vdev   */
 124         vdev_ops_t      *vdev_ops;      /* vdev operations              */
 125         spa_t           *vdev_spa;      /* spa for this vdev            */
 126         void            *vdev_tsd;      /* type-specific data           */
 127         vnode_t         *vdev_name_vp;  /* vnode for pathname           */
 128         vnode_t         *vdev_devid_vp; /* vnode for devid              */
 129         vdev_t          *vdev_top;      /* top-level vdev               */
 130         vdev_t          *vdev_parent;   /* parent vdev                  */
 131         vdev_t          **vdev_child;   /* array of children            */
 132         uint64_t        vdev_children;  /* number of children           */
 133         space_map_t     vdev_dtl[DTL_TYPES]; /* in-core dirty time logs */
 134         vdev_stat_t     vdev_stat;      /* virtual device statistics    */
 135         boolean_t       vdev_expanding; /* expand the vdev?             */
 136         boolean_t       vdev_reopening; /* reopen in progress?          */
 137         int             vdev_open_error; /* error on last open          */
 138         kthread_t       *vdev_open_thread; /* thread opening children   */
 139         uint64_t        vdev_crtxg;     /* txg when top-level was added */
 140
 141         /*
 142          * Top-level vdev state.
 143          */
 144         uint64_t        vdev_ms_array;  /* metaslab array object        */
 145         uint64_t        vdev_ms_shift;  /* metaslab size shift          */
 146         uint64_t        vdev_ms_count;  /* number of metaslabs          */
 147         metaslab_group_t *vdev_mg;      /* metaslab group               */
 148         metaslab_t      **vdev_ms;      /* metaslab array               */
 149         txg_list_t      vdev_ms_list;   /* per-txg dirty metaslab lists */
 150         txg_list_t      vdev_dtl_list;  /* per-txg dirty DTL lists      */
 151         txg_node_t      vdev_txg_node;  /* per-txg dirty vdev linkage   */
 152         boolean_t       vdev_remove_wanted; /* async remove wanted?     */
 153         boolean_t       vdev_probe_wanted; /* async probe wanted?       */
 154         uint64_t        vdev_removing;  /* device is being removed?     */
 155         list_node_t     vdev_config_dirty_node; /* config dirty list    */
 156         list_node_t     vdev_state_dirty_node; /* state dirty list      */
 157         uint64_t        vdev_deflate_ratio; /* deflation ratio (x512)   */
 158         uint64_t        vdev_islog;     /* is an intent log device      */
 159         uint64_t        vdev_ishole;    /* is a hole in the namespace   */
 160
 161         /*
 162          * Leaf vdev state.
 163          */
 164         uint64_t        vdev_psize;     /* physical device capacity     */
 165         space_map_obj_t vdev_dtl_smo;   /* dirty time log space map obj */
 166         txg_node_t      vdev_dtl_node;  /* per-txg dirty DTL linkage    */
 167         uint64_t        vdev_wholedisk; /* true if this is a whole disk */
 168         uint64_t        vdev_offline;   /* persistent offline state     */
 169         uint64_t        vdev_faulted;   /* persistent faulted state     */
 170         uint64_t        vdev_degraded;  /* persistent degraded state    */
 171         uint64_t        vdev_removed;   /* persistent removed state     */
 172         uint64_t        vdev_resilvering; /* persistent resilvering state */
 173         uint64_t        vdev_nparity;   /* number of parity devices for raidz */
 174         char            *vdev_path;     /* vdev path (if any)           */
 175         char            *vdev_devid;    /* vdev devid (if any)          */
 176         char            *vdev_physpath; /* vdev device path (if any)    */
 177         char            *vdev_fru;      /* physical FRU location        */
 178         uint64_t        vdev_not_present; /* not present during import  */
 179         uint64_t        vdev_unspare;   /* unspare when resilvering done */
 180         hrtime_t        vdev_last_try;  /* last reopen time             */
 181         boolean_t       vdev_nowritecache; /* true if flushwritecache failed */
 182         boolean_t       vdev_checkremove; /* temporary online test      */
 183         boolean_t       vdev_forcefault; /* force online fault          */
 184         boolean_t       vdev_splitting; /* split or repair in progress  */
 185         boolean_t       vdev_delayed_close; /* delayed device close?    */
 186         uint8_t         vdev_tmpoffline; /* device taken offline temporarily? */
 187         uint8_t         vdev_detached;  /* device detached?             */
 188         uint8_t         vdev_cant_read; /* vdev is failing all reads    */
 189         uint8_t         vdev_cant_write; /* vdev is failing all writes  */
 190         uint64_t        vdev_isspare;   /* was a hot spare              */
 191         uint64_t        vdev_isl2cache; /* was a l2cache device         */
 192         vdev_queue_t    vdev_queue;     /* I/O deadline schedule queue  */
 193         vdev_cache_t    vdev_cache;     /* physical block cache         */
 194         spa_aux_vdev_t  *vdev_aux;      /* for l2cache vdevs            */
 195         zio_t           *vdev_probe_zio; /* root of current probe       */
 196         vdev_aux_t      vdev_label_aux; /* on-disk aux state            */
 197
 198         /*
 199          * For DTrace to work in userland (libzpool) context, these fields must
 200          * remain at the end of the structure.  DTrace will use the kernel's
 201          * CTF definition for 'struct vdev', and since the size of a kmutex_t is
 202          * larger in userland, the offsets for the rest fields would be
 203          * incorrect.
 204          */
 205         kmutex_t        vdev_dtl_lock;  /* vdev_dtl_{map,resilver}      */
 206         kmutex_t        vdev_stat_lock; /* vdev_stat                    */
 207         kmutex_t        vdev_probe_lock; /* protects vdev_probe_zio     */
 208 };
 209
 210 #define VDEV_RAIDZ_MAXPARITY    3
 211
 212 #define VDEV_PAD_SIZE           (8 << 10)
 213 /* 2 padding areas (vl_pad1 and vl_pad2) to skip */
 214 #define VDEV_SKIP_SIZE          VDEV_PAD_SIZE * 2
 215 #define VDEV_PHYS_SIZE          (112 << 10)
 216 #define VDEV_UBERBLOCK_RING     (128 << 10)
 217
 218 #define VDEV_UBERBLOCK_SHIFT(vd)        \
 219         MAX((vd)->vdev_top->vdev_ashift, UBERBLOCK_SHIFT)
 220 #define VDEV_UBERBLOCK_COUNT(vd)        \
 221         (VDEV_UBERBLOCK_RING >> VDEV_UBERBLOCK_SHIFT(vd))
 222 #define VDEV_UBERBLOCK_OFFSET(vd, n)    \
 223         offsetof(vdev_label_t, vl_uberblock[(n) << VDEV_UBERBLOCK_SHIFT(vd)])
 224 #define VDEV_UBERBLOCK_SIZE(vd)         (1ULL << VDEV_UBERBLOCK_SHIFT(vd))
 225
 226 typedef struct vdev_phys {
 227         char            vp_nvlist[VDEV_PHYS_SIZE - sizeof (zio_eck_t)];
 228         zio_eck_t       vp_zbt;
 229 } vdev_phys_t;
 230
 231 typedef struct vdev_label {
 232         char            vl_pad1[VDEV_PAD_SIZE];                 /*  8K */
 233         char            vl_pad2[VDEV_PAD_SIZE];                 /*  8K */
 234         vdev_phys_t     vl_vdev_phys;                           /* 112K */
 235         char            vl_uberblock[VDEV_UBERBLOCK_RING];      /* 128K */
 236 } vdev_label_t;                                                 /* 256K total */
 237
 238 /*
 239  * vdev_dirty() flags
 240  */
 241 #define VDD_METASLAB    0x01
 242 #define VDD_DTL         0x02
 243
 244 /*
 245  * Size and offset of embedded boot loader region on each label.
 246  * The total size of the first two labels plus the boot area is 4MB.
 247  */
 248 #define VDEV_BOOT_OFFSET        (2 * sizeof (vdev_label_t))
 249 #define VDEV_BOOT_SIZE          (7ULL << 19)                    /* 3.5M */
 250
 251 /*
 252  * Size of label regions at the start and end of each leaf device.
 253  */
 254 #define VDEV_LABEL_START_SIZE   (2 * sizeof (vdev_label_t) + VDEV_BOOT_SIZE)
 255 #define VDEV_LABEL_END_SIZE     (2 * sizeof (vdev_label_t))
 256 #define VDEV_LABELS             4
 257
 258 #define VDEV_ALLOC_LOAD         0
 259 #define VDEV_ALLOC_ADD          1
 260 #define VDEV_ALLOC_SPARE        2
 261 #define VDEV_ALLOC_L2CACHE      3
 262 #define VDEV_ALLOC_ROOTPOOL     4
 263 #define VDEV_ALLOC_SPLIT        5
 264
 265 /*
 266  * Allocate or free a vdev
 267  */
 268 extern vdev_t *vdev_alloc_common(spa_t *spa, uint_t id, uint64_t guid,
 269     vdev_ops_t *ops);
 270 extern int vdev_alloc(spa_t *spa, vdev_t **vdp, nvlist_t *config,
 271     vdev_t *parent, uint_t id, int alloctype);
 272 extern void vdev_free(vdev_t *vd);
 273
 274 /*
 275  * Add or remove children and parents
 276  */
 277 extern void vdev_add_child(vdev_t *pvd, vdev_t *cvd);
 278 extern void vdev_remove_child(vdev_t *pvd, vdev_t *cvd);
 279 extern void vdev_compact_children(vdev_t *pvd);
 280 extern vdev_t *vdev_add_parent(vdev_t *cvd, vdev_ops_t *ops);
 281 extern void vdev_remove_parent(vdev_t *cvd);
 282
 283 /*
 284  * vdev sync load and sync
 285  */
 286 extern void vdev_load_log_state(vdev_t *nvd, vdev_t *ovd);
 287 extern boolean_t vdev_log_state_valid(vdev_t *vd);
 288 extern void vdev_load(vdev_t *vd);
 289 extern void vdev_sync(vdev_t *vd, uint64_t txg);
 290 extern void vdev_sync_done(vdev_t *vd, uint64_t txg);
 291 extern void vdev_dirty(vdev_t *vd, int flags, void *arg, uint64_t txg);
 292
 293 /*
 294  * Available vdev types.
 295  */
 296 extern vdev_ops_t vdev_root_ops;
 297 extern vdev_ops_t vdev_mirror_ops;
 298 extern vdev_ops_t vdev_replacing_ops;
 299 extern vdev_ops_t vdev_raidz_ops;
 300 extern vdev_ops_t vdev_disk_ops;
 301 extern vdev_ops_t vdev_file_ops;
 302 extern vdev_ops_t vdev_missing_ops;
 303 extern vdev_ops_t vdev_hole_ops;
 304 extern vdev_ops_t vdev_spare_ops;
 305
 306 /*
 307  * Common size functions
 308  */
 309 extern uint64_t vdev_default_asize(vdev_t *vd, uint64_t psize);
 310 extern uint64_t vdev_get_min_asize(vdev_t *vd);
 311 extern void vdev_set_min_asize(vdev_t *vd);
 312
 313 /*
 314  * zdb uses this tunable, so it must be declared here to make lint happy.
 315  */
 316 extern int zfs_vdev_cache_size;
 317
 318 #ifdef  __cplusplus
 319 }
 320 #endif
 321
 322 #endif  /* _SYS_VDEV_IMPL_H */