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 2010 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 #include <sys/avl.h> 30 #include <sys/dmu.h> 31 #include <sys/metaslab.h> 32 #include <sys/nvpair.h> 33 #include <sys/space_map.h> 34 #include <sys/vdev.h> 35 #include <sys/dkio.h> 36 #include <sys/uberblock_impl.h> 37 38 #ifdef __cplusplus 39 extern "C" { 40 #endif 41 42 /* 43 * Virtual device descriptors. 44 * 45 * All storage pool operations go through the virtual device framework, 46 * which provides data replication and I/O scheduling. 47 */ 48 49 /* 50 * Forward declarations that lots of things need. 51 */ 52 typedef struct vdev_queue vdev_queue_t; 53 typedef struct vdev_cache vdev_cache_t; 54 typedef struct vdev_cache_entry vdev_cache_entry_t; 55 56 /* 57 * Virtual device operations 58 */ 59 typedef int vdev_open_func_t(vdev_t *vd, uint64_t *size, uint64_t *ashift); 60 typedef void vdev_close_func_t(vdev_t *vd); 61 typedef uint64_t vdev_asize_func_t(vdev_t *vd, uint64_t psize); 62 typedef int vdev_io_start_func_t(zio_t *zio); 63 typedef void vdev_io_done_func_t(zio_t *zio); 64 typedef void vdev_state_change_func_t(vdev_t *vd, int, int); 65 typedef void vdev_hold_func_t(vdev_t *vd); 66 typedef void vdev_rele_func_t(vdev_t *vd); 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 vdev_hold_func_t *vdev_op_hold; 76 vdev_rele_func_t *vdev_op_rele; 77 char vdev_op_type[16]; 78 boolean_t vdev_op_leaf; 79 } vdev_ops_t; 80 81 /* 82 * Virtual device properties 83 */ 84 struct vdev_cache_entry { 85 char *ve_data; 86 uint64_t ve_offset; 87 uint64_t ve_lastused; 88 avl_node_t ve_offset_node; 89 avl_node_t ve_lastused_node; 90 uint32_t ve_hits; 91 uint16_t ve_missed_update; 92 zio_t *ve_fill_io; 93 }; 94 95 struct vdev_cache { 96 avl_tree_t vc_offset_tree; 97 avl_tree_t vc_lastused_tree; 98 kmutex_t vc_lock; 99 }; 100 101 struct vdev_queue { 102 avl_tree_t vq_deadline_tree; 103 avl_tree_t vq_read_tree; 104 avl_tree_t vq_write_tree; 105 avl_tree_t vq_pending_tree; 106 kmutex_t vq_lock; 107 }; 108 109 /* 110 * Virtual device descriptor 111 */ 112 struct vdev { 113 /* 114 * Common to all vdev types. 115 */ 116 uint64_t vdev_id; /* child number in vdev parent */ 117 uint64_t vdev_guid; /* unique ID for this vdev */ 118 uint64_t vdev_guid_sum; /* self guid + all child guids */ 119 uint64_t vdev_orig_guid; /* orig. guid prior to remove */ 120 uint64_t vdev_asize; /* allocatable device capacity */ 121 uint64_t vdev_min_asize; /* min acceptable asize */ 122 uint64_t vdev_ashift; /* block alignment shift */ 123 uint64_t vdev_state; /* see VDEV_STATE_* #defines */ 124 uint64_t vdev_prevstate; /* used when reopening a vdev */ 125 vdev_ops_t *vdev_ops; /* vdev operations */ 126 spa_t *vdev_spa; /* spa for this vdev */ 127 void *vdev_tsd; /* type-specific data */ 128 vnode_t *vdev_name_vp; /* vnode for pathname */ 129 vnode_t *vdev_devid_vp; /* vnode for devid */ 130 vdev_t *vdev_top; /* top-level vdev */ 131 vdev_t *vdev_parent; /* parent vdev */ 132 vdev_t **vdev_child; /* array of children */ 133 uint64_t vdev_children; /* number of children */ 134 space_map_t vdev_dtl[DTL_TYPES]; /* in-core dirty time logs */ 135 vdev_stat_t vdev_stat; /* virtual device statistics */ 136 boolean_t vdev_expanding; /* expand the vdev? */ 137 boolean_t vdev_reopening; /* reopen in progress? */ 138 int vdev_open_error; /* error on last open */ 139 kthread_t *vdev_open_thread; /* thread opening children */ 140 uint64_t vdev_crtxg; /* txg when top-level was added */ 141 142 /* 143 * Top-level vdev state. 144 */ 145 uint64_t vdev_ms_array; /* metaslab array object */ 146 uint64_t vdev_ms_shift; /* metaslab size shift */ 147 uint64_t vdev_ms_count; /* number of metaslabs */ 148 metaslab_group_t *vdev_mg; /* metaslab group */ 149 metaslab_t **vdev_ms; /* metaslab array */ 150 txg_list_t vdev_ms_list; /* per-txg dirty metaslab lists */ 151 txg_list_t vdev_dtl_list; /* per-txg dirty DTL lists */ 152 txg_node_t vdev_txg_node; /* per-txg dirty vdev linkage */ 153 boolean_t vdev_remove_wanted; /* async remove wanted? */ 154 boolean_t vdev_probe_wanted; /* async probe wanted? */ 155 boolean_t vdev_removing; /* device is being removed? */ 156 list_node_t vdev_config_dirty_node; /* config dirty list */ 157 list_node_t vdev_state_dirty_node; /* state dirty list */ 158 uint64_t vdev_deflate_ratio; /* deflation ratio (x512) */ 159 uint64_t vdev_islog; /* is an intent log device */ 160 uint64_t vdev_ishole; /* is a hole in the namespace */ 161 162 /* 163 * Leaf vdev state. 164 */ 165 uint64_t vdev_psize; /* physical device capacity */ 166 space_map_obj_t vdev_dtl_smo; /* dirty time log space map obj */ 167 txg_node_t vdev_dtl_node; /* per-txg dirty DTL linkage */ 168 uint64_t vdev_wholedisk; /* true if this is a whole disk */ 169 uint64_t vdev_offline; /* persistent offline state */ 170 uint64_t vdev_faulted; /* persistent faulted state */ 171 uint64_t vdev_degraded; /* persistent degraded state */ 172 uint64_t vdev_removed; /* persistent removed 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 uint8_t vdev_tmpoffline; /* device taken offline temporarily? */ 186 uint8_t vdev_detached; /* device detached? */ 187 uint8_t vdev_cant_read; /* vdev is failing all reads */ 188 uint8_t vdev_cant_write; /* vdev is failing all writes */ 189 uint64_t vdev_isspare; /* was a hot spare */ 190 uint64_t vdev_isl2cache; /* was a l2cache device */ 191 vdev_queue_t vdev_queue; /* I/O deadline schedule queue */ 192 vdev_cache_t vdev_cache; /* physical block cache */ 193 spa_aux_vdev_t *vdev_aux; /* for l2cache vdevs */ 194 zio_t *vdev_probe_zio; /* root of current probe */ 195 vdev_aux_t vdev_label_aux; /* on-disk aux state */ 196 197 /* 198 * For DTrace to work in userland (libzpool) context, these fields must 199 * remain at the end of the structure. DTrace will use the kernel's 200 * CTF definition for 'struct vdev', and since the size of a kmutex_t is 201 * larger in userland, the offsets for the rest fields would be 202 * incorrect. 203 */ 204 kmutex_t vdev_dtl_lock; /* vdev_dtl_{map,resilver} */ 205 kmutex_t vdev_stat_lock; /* vdev_stat */ 206 kmutex_t vdev_probe_lock; /* protects vdev_probe_zio */ 207 }; 208 209 #define VDEV_RAIDZ_MAXPARITY 3 210 211 #define VDEV_PAD_SIZE (8 << 10) 212 /* 2 padding areas (vl_pad1 and vl_pad2) to skip */ 213 #define VDEV_SKIP_SIZE VDEV_PAD_SIZE * 2 214 #define VDEV_PHYS_SIZE (112 << 10) 215 #define VDEV_UBERBLOCK_RING (128 << 10) 216 217 #define VDEV_UBERBLOCK_SHIFT(vd) \ 218 MAX((vd)->vdev_top->vdev_ashift, UBERBLOCK_SHIFT) 219 #define VDEV_UBERBLOCK_COUNT(vd) \ 220 (VDEV_UBERBLOCK_RING >> VDEV_UBERBLOCK_SHIFT(vd)) 221 #define VDEV_UBERBLOCK_OFFSET(vd, n) \ 222 offsetof(vdev_label_t, vl_uberblock[(n) << VDEV_UBERBLOCK_SHIFT(vd)]) 223 #define VDEV_UBERBLOCK_SIZE(vd) (1ULL << VDEV_UBERBLOCK_SHIFT(vd)) 224 225 typedef struct vdev_phys { 226 char vp_nvlist[VDEV_PHYS_SIZE - sizeof (zio_eck_t)]; 227 zio_eck_t vp_zbt; 228 } vdev_phys_t; 229 230 typedef struct vdev_label { 231 char vl_pad1[VDEV_PAD_SIZE]; /* 8K */ 232 char vl_pad2[VDEV_PAD_SIZE]; /* 8K */ 233 vdev_phys_t vl_vdev_phys; /* 112K */ 234 char vl_uberblock[VDEV_UBERBLOCK_RING]; /* 128K */ 235 } vdev_label_t; /* 256K total */ 236 237 /* 238 * vdev_dirty() flags 239 */ 240 #define VDD_METASLAB 0x01 241 #define VDD_DTL 0x02 242 243 /* 244 * Size and offset of embedded boot loader region on each label. 245 * The total size of the first two labels plus the boot area is 4MB. 246 */ 247 #define VDEV_BOOT_OFFSET (2 * sizeof (vdev_label_t)) 248 #define VDEV_BOOT_SIZE (7ULL << 19) /* 3.5M */ 249 250 /* 251 * Size of label regions at the start and end of each leaf device. 252 */ 253 #define VDEV_LABEL_START_SIZE (2 * sizeof (vdev_label_t) + VDEV_BOOT_SIZE) 254 #define VDEV_LABEL_END_SIZE (2 * sizeof (vdev_label_t)) 255 #define VDEV_LABELS 4 256 257 #define VDEV_ALLOC_LOAD 0 258 #define VDEV_ALLOC_ADD 1 259 #define VDEV_ALLOC_SPARE 2 260 #define VDEV_ALLOC_L2CACHE 3 261 #define VDEV_ALLOC_ROOTPOOL 4 262 #define VDEV_ALLOC_SPLIT 5 263 264 /* 265 * Allocate or free a vdev 266 */ 267 extern vdev_t *vdev_alloc_common(spa_t *spa, uint_t id, uint64_t guid, 268 vdev_ops_t *ops); 269 extern int vdev_alloc(spa_t *spa, vdev_t **vdp, nvlist_t *config, 270 vdev_t *parent, uint_t id, int alloctype); 271 extern void vdev_free(vdev_t *vd); 272 273 /* 274 * Add or remove children and parents 275 */ 276 extern void vdev_add_child(vdev_t *pvd, vdev_t *cvd); 277 extern void vdev_remove_child(vdev_t *pvd, vdev_t *cvd); 278 extern void vdev_compact_children(vdev_t *pvd); 279 extern vdev_t *vdev_add_parent(vdev_t *cvd, vdev_ops_t *ops); 280 extern void vdev_remove_parent(vdev_t *cvd); 281 282 /* 283 * vdev sync load and sync 284 */ 285 extern void vdev_load_log_state(vdev_t *nvd, vdev_t *ovd); 286 extern void vdev_load(vdev_t *vd); 287 extern void vdev_sync(vdev_t *vd, uint64_t txg); 288 extern void vdev_sync_done(vdev_t *vd, uint64_t txg); 289 extern void vdev_dirty(vdev_t *vd, int flags, void *arg, uint64_t txg); 290 291 /* 292 * Available vdev types. 293 */ 294 extern vdev_ops_t vdev_root_ops; 295 extern vdev_ops_t vdev_mirror_ops; 296 extern vdev_ops_t vdev_replacing_ops; 297 extern vdev_ops_t vdev_raidz_ops; 298 extern vdev_ops_t vdev_disk_ops; 299 extern vdev_ops_t vdev_file_ops; 300 extern vdev_ops_t vdev_missing_ops; 301 extern vdev_ops_t vdev_hole_ops; 302 extern vdev_ops_t vdev_spare_ops; 303 304 /* 305 * Common size functions 306 */ 307 extern uint64_t vdev_default_asize(vdev_t *vd, uint64_t psize); 308 extern uint64_t vdev_get_min_asize(vdev_t *vd); 309 extern void vdev_set_min_asize(vdev_t *vd); 310 311 /* 312 * zdb uses this tunable, so it must be declared here to make lint happy. 313 */ 314 extern int zfs_vdev_cache_size; 315 316 #ifdef __cplusplus 317 } 318 #endif 319 320 #endif /* _SYS_VDEV_IMPL_H */ 321