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 * Given a target vdev, translates the logical range "in" to the physical 84 * range "res" 85 */ 86 typedef void vdev_xlation_func_t(vdev_t *cvd, const range_seg_t *in, 87 range_seg_t *res); 88 89 typedef struct vdev_ops { 90 vdev_open_func_t *vdev_op_open; 91 vdev_close_func_t *vdev_op_close; 92 vdev_asize_func_t *vdev_op_asize; 93 vdev_io_start_func_t *vdev_op_io_start; 94 vdev_io_done_func_t *vdev_op_io_done; 95 vdev_state_change_func_t *vdev_op_state_change; 96 vdev_hold_func_t *vdev_op_hold; 97 vdev_rele_func_t *vdev_op_rele; 98 vdev_remap_func_t *vdev_op_remap; 99 /* 100 * For translating ranges from non-leaf vdevs (e.g. raidz) to leaves. 101 * Used when initializing vdevs. Isn't used by leaf ops. 102 */ 103 vdev_xlation_func_t *vdev_op_xlate; 104 char vdev_op_type[16]; 105 boolean_t vdev_op_leaf; 106 } vdev_ops_t; 107 108 /* 109 * Virtual device properties 110 */ 111 struct vdev_cache_entry { 112 struct abd *ve_abd; 113 uint64_t ve_offset; 114 uint64_t ve_lastused; 115 avl_node_t ve_offset_node; 116 avl_node_t ve_lastused_node; 117 uint32_t ve_hits; 118 uint16_t ve_missed_update; 119 zio_t *ve_fill_io; 120 }; 121 122 struct vdev_cache { 123 avl_tree_t vc_offset_tree; 124 avl_tree_t vc_lastused_tree; 125 kmutex_t vc_lock; 126 }; 127 128 typedef struct vdev_queue_class { 129 uint32_t vqc_active; 130 131 /* 132 * Sorted by offset or timestamp, depending on if the queue is 133 * LBA-ordered vs FIFO. 134 */ 135 avl_tree_t vqc_queued_tree; 136 } vdev_queue_class_t; 137 138 struct vdev_queue { 139 vdev_t *vq_vdev; 140 vdev_queue_class_t vq_class[ZIO_PRIORITY_NUM_QUEUEABLE]; 141 avl_tree_t vq_active_tree; 142 avl_tree_t vq_read_offset_tree; 143 avl_tree_t vq_write_offset_tree; 144 uint64_t vq_last_offset; 145 hrtime_t vq_io_complete_ts; /* time last i/o completed */ 146 kmutex_t vq_lock; 147 }; 148 149 /* 150 * On-disk indirect vdev state. 151 * 152 * An indirect vdev is described exclusively in the MOS config of a pool. 153 * The config for an indirect vdev includes several fields, which are 154 * accessed in memory by a vdev_indirect_config_t. 155 */ 156 typedef struct vdev_indirect_config { 157 /* 158 * Object (in MOS) which contains the indirect mapping. This object 159 * contains an array of vdev_indirect_mapping_entry_phys_t ordered by 160 * vimep_src. The bonus buffer for this object is a 161 * vdev_indirect_mapping_phys_t. This object is allocated when a vdev 162 * removal is initiated. 163 * 164 * Note that this object can be empty if none of the data on the vdev 165 * has been copied yet. 166 */ 167 uint64_t vic_mapping_object; 168 169 /* 170 * Object (in MOS) which contains the birth times for the mapping 171 * entries. This object contains an array of 172 * vdev_indirect_birth_entry_phys_t sorted by vibe_offset. The bonus 173 * buffer for this object is a vdev_indirect_birth_phys_t. This object 174 * is allocated when a vdev removal is initiated. 175 * 176 * Note that this object can be empty if none of the vdev has yet been 177 * copied. 178 */ 179 uint64_t vic_births_object; 180 181 /* 182 * This is the vdev ID which was removed previous to this vdev, or 183 * UINT64_MAX if there are no previously removed vdevs. 184 */ 185 uint64_t vic_prev_indirect_vdev; 186 } vdev_indirect_config_t; 187 188 /* 189 * Virtual device descriptor 190 */ 191 struct vdev { 192 /* 193 * Common to all vdev types. 194 */ 195 uint64_t vdev_id; /* child number in vdev parent */ 196 uint64_t vdev_guid; /* unique ID for this vdev */ 197 uint64_t vdev_guid_sum; /* self guid + all child guids */ 198 uint64_t vdev_orig_guid; /* orig. guid prior to remove */ 199 uint64_t vdev_asize; /* allocatable device capacity */ 200 uint64_t vdev_min_asize; /* min acceptable asize */ 201 uint64_t vdev_max_asize; /* max acceptable asize */ 202 uint64_t vdev_ashift; /* block alignment shift */ 203 uint64_t vdev_state; /* see VDEV_STATE_* #defines */ 204 uint64_t vdev_prevstate; /* used when reopening a vdev */ 205 vdev_ops_t *vdev_ops; /* vdev operations */ 206 spa_t *vdev_spa; /* spa for this vdev */ 207 void *vdev_tsd; /* type-specific data */ 208 vnode_t *vdev_name_vp; /* vnode for pathname */ 209 vnode_t *vdev_devid_vp; /* vnode for devid */ 210 vdev_t *vdev_top; /* top-level vdev */ 211 vdev_t *vdev_parent; /* parent vdev */ 212 vdev_t **vdev_child; /* array of children */ 213 uint64_t vdev_children; /* number of children */ 214 vdev_stat_t vdev_stat; /* virtual device statistics */ 215 boolean_t vdev_expanding; /* expand the vdev? */ 216 boolean_t vdev_reopening; /* reopen in progress? */ 217 int vdev_open_error; /* error on last open */ 218 kthread_t *vdev_open_thread; /* thread opening children */ 219 uint64_t vdev_crtxg; /* txg when top-level was added */ 220 221 /* 222 * Top-level vdev state. 223 */ 224 uint64_t vdev_ms_array; /* metaslab array object */ 225 uint64_t vdev_ms_shift; /* metaslab size shift */ 226 uint64_t vdev_ms_count; /* number of metaslabs */ 227 metaslab_group_t *vdev_mg; /* metaslab group */ 228 metaslab_t **vdev_ms; /* metaslab array */ 229 txg_list_t vdev_ms_list; /* per-txg dirty metaslab lists */ 230 txg_list_t vdev_dtl_list; /* per-txg dirty DTL lists */ 231 txg_node_t vdev_txg_node; /* per-txg dirty vdev linkage */ 232 boolean_t vdev_remove_wanted; /* async remove wanted? */ 233 boolean_t vdev_probe_wanted; /* async probe wanted? */ 234 list_node_t vdev_config_dirty_node; /* config dirty list */ 235 list_node_t vdev_state_dirty_node; /* state dirty list */ 236 uint64_t vdev_deflate_ratio; /* deflation ratio (x512) */ 237 uint64_t vdev_islog; /* is an intent log device */ 238 uint64_t vdev_removing; /* device is being removed? */ 239 boolean_t vdev_ishole; /* is a hole in the namespace */ 240 kmutex_t vdev_queue_lock; /* protects vdev_queue_depth */ 241 uint64_t vdev_top_zap; 242 243 /* pool checkpoint related */ 244 space_map_t *vdev_checkpoint_sm; /* contains reserved blocks */ 245 246 boolean_t vdev_initialize_exit_wanted; 247 vdev_initializing_state_t vdev_initialize_state; 248 kthread_t *vdev_initialize_thread; 249 /* Protects vdev_initialize_thread and vdev_initialize_state. */ 250 kmutex_t vdev_initialize_lock; 251 kcondvar_t vdev_initialize_cv; 252 uint64_t vdev_initialize_offset[TXG_SIZE]; 253 uint64_t vdev_initialize_last_offset; 254 range_tree_t *vdev_initialize_tree; /* valid while initializing */ 255 uint64_t vdev_initialize_bytes_est; 256 uint64_t vdev_initialize_bytes_done; 257 time_t vdev_initialize_action_time; /* start and end time */ 258 259 /* for limiting outstanding I/Os */ 260 kmutex_t vdev_initialize_io_lock; 261 kcondvar_t vdev_initialize_io_cv; 262 uint64_t vdev_initialize_inflight; 263 264 /* 265 * Values stored in the config for an indirect or removing vdev. 266 */ 267 vdev_indirect_config_t vdev_indirect_config; 268 269 /* 270 * The vdev_indirect_rwlock protects the vdev_indirect_mapping 271 * pointer from changing on indirect vdevs (when it is condensed). 272 * Note that removing (not yet indirect) vdevs have different 273 * access patterns (the mapping is not accessed from open context, 274 * e.g. from zio_read) and locking strategy (e.g. svr_lock). 275 */ 276 krwlock_t vdev_indirect_rwlock; 277 vdev_indirect_mapping_t *vdev_indirect_mapping; 278 vdev_indirect_births_t *vdev_indirect_births; 279 280 /* 281 * In memory data structures used to manage the obsolete sm, for 282 * indirect or removing vdevs. 283 * 284 * The vdev_obsolete_segments is the in-core record of the segments 285 * that are no longer referenced anywhere in the pool (due to 286 * being freed or remapped and not referenced by any snapshots). 287 * During a sync, segments are added to vdev_obsolete_segments 288 * via vdev_indirect_mark_obsolete(); at the end of each sync 289 * pass, this is appended to vdev_obsolete_sm via 290 * vdev_indirect_sync_obsolete(). The vdev_obsolete_lock 291 * protects against concurrent modifications of vdev_obsolete_segments 292 * from multiple zio threads. 293 */ 294 kmutex_t vdev_obsolete_lock; 295 range_tree_t *vdev_obsolete_segments; 296 space_map_t *vdev_obsolete_sm; 297 298 /* 299 * The queue depth parameters determine how many async writes are 300 * still pending (i.e. allocated but not yet issued to disk) per 301 * top-level (vdev_async_write_queue_depth) and the maximum allowed 302 * (vdev_max_async_write_queue_depth). These values only apply to 303 * top-level vdevs. 304 */ 305 uint64_t vdev_async_write_queue_depth; 306 uint64_t vdev_max_async_write_queue_depth; 307 308 /* 309 * Leaf vdev state. 310 */ 311 range_tree_t *vdev_dtl[DTL_TYPES]; /* dirty time logs */ 312 space_map_t *vdev_dtl_sm; /* dirty time log space map */ 313 txg_node_t vdev_dtl_node; /* per-txg dirty DTL linkage */ 314 uint64_t vdev_dtl_object; /* DTL object */ 315 uint64_t vdev_psize; /* physical device capacity */ 316 uint64_t vdev_wholedisk; /* true if this is a whole disk */ 317 uint64_t vdev_offline; /* persistent offline state */ 318 uint64_t vdev_faulted; /* persistent faulted state */ 319 uint64_t vdev_degraded; /* persistent degraded state */ 320 uint64_t vdev_removed; /* persistent removed state */ 321 uint64_t vdev_resilver_txg; /* persistent resilvering state */ 322 uint64_t vdev_nparity; /* number of parity devices for raidz */ 323 char *vdev_path; /* vdev path (if any) */ 324 char *vdev_devid; /* vdev devid (if any) */ 325 char *vdev_physpath; /* vdev device path (if any) */ 326 char *vdev_fru; /* physical FRU location */ 327 uint64_t vdev_not_present; /* not present during import */ 328 uint64_t vdev_unspare; /* unspare when resilvering done */ 329 boolean_t vdev_nowritecache; /* true if flushwritecache failed */ 330 boolean_t vdev_checkremove; /* temporary online test */ 331 boolean_t vdev_forcefault; /* force online fault */ 332 boolean_t vdev_splitting; /* split or repair in progress */ 333 boolean_t vdev_delayed_close; /* delayed device close? */ 334 boolean_t vdev_tmpoffline; /* device taken offline temporarily? */ 335 boolean_t vdev_detached; /* device detached? */ 336 boolean_t vdev_cant_read; /* vdev is failing all reads */ 337 boolean_t vdev_cant_write; /* vdev is failing all writes */ 338 boolean_t vdev_isspare; /* was a hot spare */ 339 boolean_t vdev_isl2cache; /* was a l2cache device */ 340 vdev_queue_t vdev_queue; /* I/O deadline schedule queue */ 341 vdev_cache_t vdev_cache; /* physical block cache */ 342 spa_aux_vdev_t *vdev_aux; /* for l2cache and spares vdevs */ 343 zio_t *vdev_probe_zio; /* root of current probe */ 344 vdev_aux_t vdev_label_aux; /* on-disk aux state */ 345 uint64_t vdev_leaf_zap; 346 347 /* 348 * For DTrace to work in userland (libzpool) context, these fields must 349 * remain at the end of the structure. DTrace will use the kernel's 350 * CTF definition for 'struct vdev', and since the size of a kmutex_t is 351 * larger in userland, the offsets for the rest of the fields would be 352 * incorrect. 353 */ 354 kmutex_t vdev_dtl_lock; /* vdev_dtl_{map,resilver} */ 355 kmutex_t vdev_stat_lock; /* vdev_stat */ 356 kmutex_t vdev_probe_lock; /* protects vdev_probe_zio */ 357 }; 358 359 #define VDEV_RAIDZ_MAXPARITY 3 360 361 #define VDEV_PAD_SIZE (8 << 10) 362 /* 2 padding areas (vl_pad1 and vl_pad2) to skip */ 363 #define VDEV_SKIP_SIZE VDEV_PAD_SIZE * 2 364 #define VDEV_PHYS_SIZE (112 << 10) 365 #define VDEV_UBERBLOCK_RING (128 << 10) 366 367 /* The largest uberblock we support is 8k. */ 368 #define MAX_UBERBLOCK_SHIFT (13) 369 #define VDEV_UBERBLOCK_SHIFT(vd) \ 370 MIN(MAX((vd)->vdev_top->vdev_ashift, UBERBLOCK_SHIFT), \ 371 MAX_UBERBLOCK_SHIFT) 372 #define VDEV_UBERBLOCK_COUNT(vd) \ 373 (VDEV_UBERBLOCK_RING >> VDEV_UBERBLOCK_SHIFT(vd)) 374 #define VDEV_UBERBLOCK_OFFSET(vd, n) \ 375 offsetof(vdev_label_t, vl_uberblock[(n) << VDEV_UBERBLOCK_SHIFT(vd)]) 376 #define VDEV_UBERBLOCK_SIZE(vd) (1ULL << VDEV_UBERBLOCK_SHIFT(vd)) 377 378 typedef struct vdev_phys { 379 char vp_nvlist[VDEV_PHYS_SIZE - sizeof (zio_eck_t)]; 380 zio_eck_t vp_zbt; 381 } vdev_phys_t; 382 383 typedef struct vdev_label { 384 char vl_pad1[VDEV_PAD_SIZE]; /* 8K */ 385 char vl_pad2[VDEV_PAD_SIZE]; /* 8K */ 386 vdev_phys_t vl_vdev_phys; /* 112K */ 387 char vl_uberblock[VDEV_UBERBLOCK_RING]; /* 128K */ 388 } vdev_label_t; /* 256K total */ 389 390 /* 391 * vdev_dirty() flags 392 */ 393 #define VDD_METASLAB 0x01 394 #define VDD_DTL 0x02 395 396 /* Offset of embedded boot loader region on each label */ 397 #define VDEV_BOOT_OFFSET (2 * sizeof (vdev_label_t)) 398 /* 399 * Size of embedded boot loader region on each label. 400 * The total size of the first two labels plus the boot area is 4MB. 401 */ 402 #define VDEV_BOOT_SIZE (7ULL << 19) /* 3.5M */ 403 404 /* 405 * Size of label regions at the start and end of each leaf device. 406 */ 407 #define VDEV_LABEL_START_SIZE (2 * sizeof (vdev_label_t) + VDEV_BOOT_SIZE) 408 #define VDEV_LABEL_END_SIZE (2 * sizeof (vdev_label_t)) 409 #define VDEV_LABELS 4 410 #define VDEV_BEST_LABEL VDEV_LABELS 411 412 #define VDEV_ALLOC_LOAD 0 413 #define VDEV_ALLOC_ADD 1 414 #define VDEV_ALLOC_SPARE 2 415 #define VDEV_ALLOC_L2CACHE 3 416 #define VDEV_ALLOC_ROOTPOOL 4 417 #define VDEV_ALLOC_SPLIT 5 418 #define VDEV_ALLOC_ATTACH 6 419 420 /* 421 * Allocate or free a vdev 422 */ 423 extern vdev_t *vdev_alloc_common(spa_t *spa, uint_t id, uint64_t guid, 424 vdev_ops_t *ops); 425 extern int vdev_alloc(spa_t *spa, vdev_t **vdp, nvlist_t *config, 426 vdev_t *parent, uint_t id, int alloctype); 427 extern void vdev_free(vdev_t *vd); 428 429 /* 430 * Add or remove children and parents 431 */ 432 extern void vdev_add_child(vdev_t *pvd, vdev_t *cvd); 433 extern void vdev_remove_child(vdev_t *pvd, vdev_t *cvd); 434 extern void vdev_compact_children(vdev_t *pvd); 435 extern vdev_t *vdev_add_parent(vdev_t *cvd, vdev_ops_t *ops); 436 extern void vdev_remove_parent(vdev_t *cvd); 437 438 /* 439 * vdev sync load and sync 440 */ 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 void vdev_default_xlate(vdev_t *vd, const range_seg_t *in, 467 range_seg_t *out); 468 extern uint64_t vdev_default_asize(vdev_t *vd, uint64_t psize); 469 extern uint64_t vdev_get_min_asize(vdev_t *vd); 470 extern void vdev_set_min_asize(vdev_t *vd); 471 472 /* 473 * Global variables 474 */ 475 extern int vdev_standard_sm_blksz; 476 /* zdb uses this tunable, so it must be declared here to make lint happy. */ 477 extern int zfs_vdev_cache_size; 478 479 /* 480 * Functions from vdev_indirect.c 481 */ 482 extern void vdev_indirect_sync_obsolete(vdev_t *vd, dmu_tx_t *tx); 483 extern boolean_t vdev_indirect_should_condense(vdev_t *vd); 484 extern void spa_condense_indirect_start_sync(vdev_t *vd, dmu_tx_t *tx); 485 extern int vdev_obsolete_sm_object(vdev_t *vd); 486 extern boolean_t vdev_obsolete_counts_are_precise(vdev_t *vd); 487 488 /* 489 * Other miscellaneous functions 490 */ 491 int vdev_checkpoint_sm_object(vdev_t *vd); 492 493 /* 494 * The vdev_buf_t is used to translate between zio_t and buf_t, and back again. 495 */ 496 typedef struct vdev_buf { 497 buf_t vb_buf; /* buffer that describes the io */ 498 zio_t *vb_io; /* pointer back to the original zio_t */ 499 } vdev_buf_t; 500 501 #ifdef __cplusplus 502 } 503 #endif 504 505 #endif /* _SYS_VDEV_IMPL_H */ 506