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 2006 Sun Microsystems, Inc. All rights reserved. 23 * Use is subject to license terms. 24 */ 25 26 #ifndef _SYS_VDEV_IMPL_H 27 #define _SYS_VDEV_IMPL_H 28 29 #pragma ident "%Z%%M% %I% %E% SMI" 30 31 #include <sys/avl.h> 32 #include <sys/dmu.h> 33 #include <sys/metaslab.h> 34 #include <sys/nvpair.h> 35 #include <sys/space_map.h> 36 #include <sys/vdev.h> 37 #include <sys/dkio.h> 38 #include <sys/uberblock_impl.h> 39 40 #ifdef __cplusplus 41 extern "C" { 42 #endif 43 44 /* 45 * Virtual device descriptors. 46 * 47 * All storage pool operations go through the virtual device framework, 48 * which provides data replication and I/O scheduling. 49 */ 50 51 /* 52 * Forward declarations that lots of things need. 53 */ 54 typedef struct vdev_queue vdev_queue_t; 55 typedef struct vdev_cache vdev_cache_t; 56 typedef struct vdev_cache_entry vdev_cache_entry_t; 57 58 /* 59 * Virtual device operations 60 */ 61 typedef int vdev_open_func_t(vdev_t *vd, uint64_t *size, uint64_t *ashift); 62 typedef void vdev_close_func_t(vdev_t *vd); 63 typedef uint64_t vdev_asize_func_t(vdev_t *vd, uint64_t psize); 64 typedef void vdev_io_start_func_t(zio_t *zio); 65 typedef void vdev_io_done_func_t(zio_t *zio); 66 typedef void vdev_state_change_func_t(vdev_t *vd, int, int); 67 68 typedef struct vdev_ops { 69 vdev_open_func_t *vdev_op_open; 70 vdev_close_func_t *vdev_op_close; 71 vdev_asize_func_t *vdev_op_asize; 72 vdev_io_start_func_t *vdev_op_io_start; 73 vdev_io_done_func_t *vdev_op_io_done; 74 vdev_state_change_func_t *vdev_op_state_change; 75 char vdev_op_type[16]; 76 boolean_t vdev_op_leaf; 77 } vdev_ops_t; 78 79 /* 80 * Virtual device properties 81 */ 82 struct vdev_cache_entry { 83 char *ve_data; 84 uint64_t ve_offset; 85 uint64_t ve_lastused; 86 avl_node_t ve_offset_node; 87 avl_node_t ve_lastused_node; 88 uint32_t ve_hits; 89 uint16_t ve_missed_update; 90 zio_t *ve_fill_io; 91 }; 92 93 struct vdev_cache { 94 uint64_t vc_size; 95 uint64_t vc_bshift; 96 uint64_t vc_blocksize; 97 uint64_t vc_max; 98 avl_tree_t vc_offset_tree; 99 avl_tree_t vc_lastused_tree; 100 kmutex_t vc_lock; 101 }; 102 103 struct vdev_queue { 104 uint64_t vq_min_pending; 105 uint64_t vq_max_pending; 106 uint64_t vq_scrub_limit; 107 uint64_t vq_agg_limit; 108 uint64_t vq_time_shift; 109 uint64_t vq_ramp_rate; 110 uint64_t vq_scrub_count; 111 avl_tree_t vq_deadline_tree; 112 avl_tree_t vq_read_tree; 113 avl_tree_t vq_write_tree; 114 avl_tree_t vq_pending_tree; 115 kmutex_t vq_lock; 116 }; 117 118 /* 119 * Virtual device descriptor 120 */ 121 struct vdev { 122 /* 123 * Common to all vdev types. 124 */ 125 uint64_t vdev_id; /* child number in vdev parent */ 126 uint64_t vdev_guid; /* unique ID for this vdev */ 127 uint64_t vdev_guid_sum; /* self guid + all child guids */ 128 uint64_t vdev_asize; /* allocatable device capacity */ 129 uint64_t vdev_ashift; /* block alignment shift */ 130 uint64_t vdev_state; /* see VDEV_STATE_* #defines */ 131 uint64_t vdev_prevstate; /* used when reopening a vdev */ 132 vdev_ops_t *vdev_ops; /* vdev operations */ 133 spa_t *vdev_spa; /* spa for this vdev */ 134 void *vdev_tsd; /* type-specific data */ 135 vdev_t *vdev_top; /* top-level vdev */ 136 vdev_t *vdev_parent; /* parent vdev */ 137 vdev_t **vdev_child; /* array of children */ 138 uint64_t vdev_children; /* number of children */ 139 space_map_t vdev_dtl_map; /* dirty time log in-core state */ 140 space_map_t vdev_dtl_scrub; /* DTL for scrub repair writes */ 141 vdev_stat_t vdev_stat; /* virtual device statistics */ 142 143 /* 144 * Top-level vdev state. 145 */ 146 uint64_t vdev_ms_array; /* metaslab array object */ 147 uint64_t vdev_ms_shift; /* metaslab size shift */ 148 uint64_t vdev_ms_count; /* number of metaslabs */ 149 metaslab_group_t *vdev_mg; /* metaslab group */ 150 metaslab_t **vdev_ms; /* metaslab array */ 151 txg_list_t vdev_ms_list; /* per-txg dirty metaslab lists */ 152 txg_list_t vdev_dtl_list; /* per-txg dirty DTL lists */ 153 txg_node_t vdev_txg_node; /* per-txg dirty vdev linkage */ 154 uint8_t vdev_reopen_wanted; /* async reopen wanted? */ 155 list_node_t vdev_dirty_node; /* config dirty list */ 156 uint64_t vdev_deflate_ratio; /* deflation ratio (x512) */ 157 158 /* 159 * Leaf vdev state. 160 */ 161 uint64_t vdev_psize; /* physical device capacity */ 162 space_map_obj_t vdev_dtl; /* dirty time log on-disk state */ 163 txg_node_t vdev_dtl_node; /* per-txg dirty DTL linkage */ 164 uint64_t vdev_wholedisk; /* true if this is a whole disk */ 165 uint64_t vdev_offline; /* device taken offline? */ 166 uint64_t vdev_nparity; /* number of parity devices for raidz */ 167 char *vdev_path; /* vdev path (if any) */ 168 char *vdev_devid; /* vdev devid (if any) */ 169 uint64_t vdev_fault_arg; /* fault injection paramater */ 170 int vdev_fault_mask; /* zio types to fault */ 171 uint8_t vdev_fault_mode; /* fault injection mode */ 172 uint8_t vdev_cache_active; /* vdev_cache and vdev_queue */ 173 uint8_t vdev_tmpoffline; /* device taken offline temporarily? */ 174 uint8_t vdev_detached; /* device detached? */ 175 uint64_t vdev_isspare; /* was a hot spare */ 176 vdev_queue_t vdev_queue; /* I/O deadline schedule queue */ 177 vdev_cache_t vdev_cache; /* physical block cache */ 178 uint64_t vdev_not_present; /* not present during import */ 179 hrtime_t vdev_last_try; /* last reopen time */ 180 boolean_t vdev_nowritecache; /* true if flushwritecache failed */ 181 182 /* 183 * For DTrace to work in userland (libzpool) context, these fields must 184 * remain at the end of the structure. DTrace will use the kernel's 185 * CTF definition for 'struct vdev', and since the size of a kmutex_t is 186 * larger in userland, the offsets for the rest fields would be 187 * incorrect. 188 */ 189 kmutex_t vdev_dtl_lock; /* vdev_dtl_{map,resilver} */ 190 kmutex_t vdev_stat_lock; /* vdev_stat */ 191 }; 192 193 #define VDEV_SKIP_SIZE (8 << 10) 194 #define VDEV_BOOT_HEADER_SIZE (8 << 10) 195 #define VDEV_PHYS_SIZE (112 << 10) 196 #define VDEV_UBERBLOCK_RING (128 << 10) 197 198 #define VDEV_UBERBLOCK_SHIFT(vd) \ 199 MAX((vd)->vdev_top->vdev_ashift, UBERBLOCK_SHIFT) 200 #define VDEV_UBERBLOCK_COUNT(vd) \ 201 (VDEV_UBERBLOCK_RING >> VDEV_UBERBLOCK_SHIFT(vd)) 202 #define VDEV_UBERBLOCK_OFFSET(vd, n) \ 203 offsetof(vdev_label_t, vl_uberblock[(n) << VDEV_UBERBLOCK_SHIFT(vd)]) 204 #define VDEV_UBERBLOCK_SIZE(vd) (1ULL << VDEV_UBERBLOCK_SHIFT(vd)) 205 206 /* ZFS boot block */ 207 #define VDEV_BOOT_MAGIC 0x2f5b007b10cULL 208 #define VDEV_BOOT_VERSION 1 /* version number */ 209 210 typedef struct vdev_boot_header { 211 uint64_t vb_magic; /* VDEV_BOOT_MAGIC */ 212 uint64_t vb_version; /* VDEV_BOOT_VERSION */ 213 uint64_t vb_offset; /* start offset (bytes) */ 214 uint64_t vb_size; /* size (bytes) */ 215 char vb_pad[VDEV_BOOT_HEADER_SIZE - 4 * sizeof (uint64_t)]; 216 } vdev_boot_header_t; 217 218 typedef struct vdev_phys { 219 char vp_nvlist[VDEV_PHYS_SIZE - sizeof (zio_block_tail_t)]; 220 zio_block_tail_t vp_zbt; 221 } vdev_phys_t; 222 223 typedef struct vdev_label { 224 char vl_pad[VDEV_SKIP_SIZE]; /* 8K */ 225 vdev_boot_header_t vl_boot_header; /* 8K */ 226 vdev_phys_t vl_vdev_phys; /* 112K */ 227 char vl_uberblock[VDEV_UBERBLOCK_RING]; /* 128K */ 228 } vdev_label_t; /* 256K total */ 229 230 /* 231 * vdev_dirty() flags 232 */ 233 #define VDD_METASLAB 0x01 234 #define VDD_DTL 0x02 235 236 /* 237 * Size and offset of embedded boot loader region on each label. 238 * The total size of the first two labels plus the boot area is 4MB. 239 */ 240 #define VDEV_BOOT_OFFSET (2 * sizeof (vdev_label_t)) 241 #define VDEV_BOOT_SIZE (7ULL << 19) /* 3.5M */ 242 243 /* 244 * Size of label regions at the start and end of each leaf device. 245 */ 246 #define VDEV_LABEL_START_SIZE (2 * sizeof (vdev_label_t) + VDEV_BOOT_SIZE) 247 #define VDEV_LABEL_END_SIZE (2 * sizeof (vdev_label_t)) 248 #define VDEV_LABELS 4 249 250 #define VDEV_ALLOC_LOAD 0 251 #define VDEV_ALLOC_ADD 1 252 #define VDEV_ALLOC_SPARE 2 253 254 /* 255 * Allocate or free a vdev 256 */ 257 extern int vdev_alloc(spa_t *spa, vdev_t **vdp, nvlist_t *config, 258 vdev_t *parent, uint_t id, int alloctype); 259 extern void vdev_free(vdev_t *vd); 260 261 /* 262 * Add or remove children and parents 263 */ 264 extern void vdev_add_child(vdev_t *pvd, vdev_t *cvd); 265 extern void vdev_remove_child(vdev_t *pvd, vdev_t *cvd); 266 extern void vdev_compact_children(vdev_t *pvd); 267 extern vdev_t *vdev_add_parent(vdev_t *cvd, vdev_ops_t *ops); 268 extern void vdev_remove_parent(vdev_t *cvd); 269 270 /* 271 * vdev sync load and sync 272 */ 273 extern void vdev_load(vdev_t *vd); 274 extern void vdev_sync(vdev_t *vd, uint64_t txg); 275 extern void vdev_sync_done(vdev_t *vd, uint64_t txg); 276 extern void vdev_dirty(vdev_t *vd, int flags, void *arg, uint64_t txg); 277 278 /* 279 * Available vdev types. 280 */ 281 extern vdev_ops_t vdev_root_ops; 282 extern vdev_ops_t vdev_mirror_ops; 283 extern vdev_ops_t vdev_replacing_ops; 284 extern vdev_ops_t vdev_raidz_ops; 285 extern vdev_ops_t vdev_disk_ops; 286 extern vdev_ops_t vdev_file_ops; 287 extern vdev_ops_t vdev_missing_ops; 288 extern vdev_ops_t vdev_spare_ops; 289 290 /* 291 * Common size functions 292 */ 293 extern uint64_t vdev_default_asize(vdev_t *vd, uint64_t psize); 294 extern uint64_t vdev_get_rsize(vdev_t *vd); 295 296 #ifdef __cplusplus 297 } 298 #endif 299 300 #endif /* _SYS_VDEV_IMPL_H */ 301