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 2009 Sun Microsystems, Inc. All rights reserved. 23 * Use is subject to license terms. 24 */ 25 26 /* 27 * Copyright (c) 2011, 2018 by Delphix. All rights reserved. 28 */ 29 30 #ifndef _SYS_METASLAB_IMPL_H 31 #define _SYS_METASLAB_IMPL_H 32 33 #include <sys/metaslab.h> 34 #include <sys/space_map.h> 35 #include <sys/range_tree.h> 36 #include <sys/vdev.h> 37 #include <sys/txg.h> 38 #include <sys/avl.h> 39 40 #ifdef __cplusplus 41 extern "C" { 42 #endif 43 44 /* 45 * Metaslab allocation tracing record. 46 */ 47 typedef struct metaslab_alloc_trace { 48 list_node_t mat_list_node; 49 metaslab_group_t *mat_mg; 50 metaslab_t *mat_msp; 51 uint64_t mat_size; 52 uint64_t mat_weight; 53 uint32_t mat_dva_id; 54 uint64_t mat_offset; 55 int mat_allocator; 56 } metaslab_alloc_trace_t; 57 58 /* 59 * Used by the metaslab allocation tracing facility to indicate 60 * error conditions. These errors are stored to the offset member 61 * of the metaslab_alloc_trace_t record and displayed by mdb. 62 */ 63 typedef enum trace_alloc_type { 64 TRACE_ALLOC_FAILURE = -1ULL, 65 TRACE_TOO_SMALL = -2ULL, 66 TRACE_FORCE_GANG = -3ULL, 67 TRACE_NOT_ALLOCATABLE = -4ULL, 68 TRACE_GROUP_FAILURE = -5ULL, 69 TRACE_ENOSPC = -6ULL, 70 TRACE_CONDENSING = -7ULL, 71 TRACE_VDEV_ERROR = -8ULL 72 } trace_alloc_type_t; 73 74 #define METASLAB_WEIGHT_PRIMARY (1ULL << 63) 75 #define METASLAB_WEIGHT_SECONDARY (1ULL << 62) 76 #define METASLAB_WEIGHT_CLAIM (1ULL << 61) 77 #define METASLAB_WEIGHT_TYPE (1ULL << 60) 78 #define METASLAB_ACTIVE_MASK \ 79 (METASLAB_WEIGHT_PRIMARY | METASLAB_WEIGHT_SECONDARY | \ 80 METASLAB_WEIGHT_CLAIM) 81 82 /* 83 * The metaslab weight is used to encode the amount of free space in a 84 * metaslab, such that the "best" metaslab appears first when sorting the 85 * metaslabs by weight. The weight (and therefore the "best" metaslab) can 86 * be determined in two different ways: by computing a weighted sum of all 87 * the free space in the metaslab (a space based weight) or by counting only 88 * the free segments of the largest size (a segment based weight). We prefer 89 * the segment based weight because it reflects how the free space is 90 * comprised, but we cannot always use it -- legacy pools do not have the 91 * space map histogram information necessary to determine the largest 92 * contiguous regions. Pools that have the space map histogram determine 93 * the segment weight by looking at each bucket in the histogram and 94 * determining the free space whose size in bytes is in the range: 95 * [2^i, 2^(i+1)) 96 * We then encode the largest index, i, that contains regions into the 97 * segment-weighted value. 98 * 99 * Space-based weight: 100 * 101 * 64 56 48 40 32 24 16 8 0 102 * +-------+-------+-------+-------+-------+-------+-------+-------+ 103 * |PSC1| weighted-free space | 104 * +-------+-------+-------+-------+-------+-------+-------+-------+ 105 * 106 * PS - indicates primary and secondary activation 107 * C - indicates activation for claimed block zio 108 * space - the fragmentation-weighted space 109 * 110 * Segment-based weight: 111 * 112 * 64 56 48 40 32 24 16 8 0 113 * +-------+-------+-------+-------+-------+-------+-------+-------+ 114 * |PSC0| idx| count of segments in region | 115 * +-------+-------+-------+-------+-------+-------+-------+-------+ 116 * 117 * PS - indicates primary and secondary activation 118 * C - indicates activation for claimed block zio 119 * idx - index for the highest bucket in the histogram 120 * count - number of segments in the specified bucket 121 */ 122 #define WEIGHT_GET_ACTIVE(weight) BF64_GET((weight), 61, 3) 123 #define WEIGHT_SET_ACTIVE(weight, x) BF64_SET((weight), 61, 3, x) 124 125 #define WEIGHT_IS_SPACEBASED(weight) \ 126 ((weight) == 0 || BF64_GET((weight), 60, 1)) 127 #define WEIGHT_SET_SPACEBASED(weight) BF64_SET((weight), 60, 1, 1) 128 129 /* 130 * These macros are only applicable to segment-based weighting. 131 */ 132 #define WEIGHT_GET_INDEX(weight) BF64_GET((weight), 54, 6) 133 #define WEIGHT_SET_INDEX(weight, x) BF64_SET((weight), 54, 6, x) 134 #define WEIGHT_GET_COUNT(weight) BF64_GET((weight), 0, 54) 135 #define WEIGHT_SET_COUNT(weight, x) BF64_SET((weight), 0, 54, x) 136 137 /* 138 * A metaslab class encompasses a category of allocatable top-level vdevs. 139 * Each top-level vdev is associated with a metaslab group which defines 140 * the allocatable region for that vdev. Examples of these categories include 141 * "normal" for data block allocations (i.e. main pool allocations) or "log" 142 * for allocations designated for intent log devices (i.e. slog devices). 143 * When a block allocation is requested from the SPA it is associated with a 144 * metaslab_class_t, and only top-level vdevs (i.e. metaslab groups) belonging 145 * to the class can be used to satisfy that request. Allocations are done 146 * by traversing the metaslab groups that are linked off of the mc_rotor field. 147 * This rotor points to the next metaslab group where allocations will be 148 * attempted. Allocating a block is a 3 step process -- select the metaslab 149 * group, select the metaslab, and then allocate the block. The metaslab 150 * class defines the low-level block allocator that will be used as the 151 * final step in allocation. These allocators are pluggable allowing each class 152 * to use a block allocator that best suits that class. 153 */ 154 struct metaslab_class { 155 kmutex_t mc_lock; 156 spa_t *mc_spa; 157 metaslab_group_t *mc_rotor; 158 metaslab_ops_t *mc_ops; 159 uint64_t mc_aliquot; 160 161 /* 162 * Track the number of metaslab groups that have been initialized 163 * and can accept allocations. An initialized metaslab group is 164 * one has been completely added to the config (i.e. we have 165 * updated the MOS config and the space has been added to the pool). 166 */ 167 uint64_t mc_groups; 168 169 /* 170 * Toggle to enable/disable the allocation throttle. 171 */ 172 boolean_t mc_alloc_throttle_enabled; 173 174 /* 175 * The allocation throttle works on a reservation system. Whenever 176 * an asynchronous zio wants to perform an allocation it must 177 * first reserve the number of blocks that it wants to allocate. 178 * If there aren't sufficient slots available for the pending zio 179 * then that I/O is throttled until more slots free up. The current 180 * number of reserved allocations is maintained by the mc_alloc_slots 181 * refcount. The mc_alloc_max_slots value determines the maximum 182 * number of allocations that the system allows. Gang blocks are 183 * allowed to reserve slots even if we've reached the maximum 184 * number of allocations allowed. 185 */ 186 uint64_t *mc_alloc_max_slots; 187 refcount_t *mc_alloc_slots; 188 189 uint64_t mc_alloc_groups; /* # of allocatable groups */ 190 191 uint64_t mc_alloc; /* total allocated space */ 192 uint64_t mc_deferred; /* total deferred frees */ 193 uint64_t mc_space; /* total space (alloc + free) */ 194 uint64_t mc_dspace; /* total deflated space */ 195 uint64_t mc_histogram[RANGE_TREE_HISTOGRAM_SIZE]; 196 }; 197 198 /* 199 * Metaslab groups encapsulate all the allocatable regions (i.e. metaslabs) 200 * of a top-level vdev. They are linked togther to form a circular linked 201 * list and can belong to only one metaslab class. Metaslab groups may become 202 * ineligible for allocations for a number of reasons such as limited free 203 * space, fragmentation, or going offline. When this happens the allocator will 204 * simply find the next metaslab group in the linked list and attempt 205 * to allocate from that group instead. 206 */ 207 struct metaslab_group { 208 kmutex_t mg_lock; 209 metaslab_t **mg_primaries; 210 metaslab_t **mg_secondaries; 211 avl_tree_t mg_metaslab_tree; 212 uint64_t mg_aliquot; 213 boolean_t mg_allocatable; /* can we allocate? */ 214 uint64_t mg_ms_ready; 215 216 /* 217 * A metaslab group is considered to be initialized only after 218 * we have updated the MOS config and added the space to the pool. 219 * We only allow allocation attempts to a metaslab group if it 220 * has been initialized. 221 */ 222 boolean_t mg_initialized; 223 224 uint64_t mg_free_capacity; /* percentage free */ 225 int64_t mg_bias; 226 int64_t mg_activation_count; 227 metaslab_class_t *mg_class; 228 vdev_t *mg_vd; 229 taskq_t *mg_taskq; 230 metaslab_group_t *mg_prev; 231 metaslab_group_t *mg_next; 232 233 /* 234 * In order for the allocation throttle to function properly, we cannot 235 * have too many IOs going to each disk by default; the throttle 236 * operates by allocating more work to disks that finish quickly, so 237 * allocating larger chunks to each disk reduces its effectiveness. 238 * However, if the number of IOs going to each allocator is too small, 239 * we will not perform proper aggregation at the vdev_queue layer, 240 * also resulting in decreased performance. Therefore, we will use a 241 * ramp-up strategy. 242 * 243 * Each allocator in each metaslab group has a current queue depth 244 * (mg_alloc_queue_depth[allocator]) and a current max queue depth 245 * (mg_cur_max_alloc_queue_depth[allocator]), and each metaslab group 246 * has an absolute max queue depth (mg_max_alloc_queue_depth). We 247 * add IOs to an allocator until the mg_alloc_queue_depth for that 248 * allocator hits the cur_max. Every time an IO completes for a given 249 * allocator on a given metaslab group, we increment its cur_max until 250 * it reaches mg_max_alloc_queue_depth. The cur_max resets every txg to 251 * help protect against disks that decrease in performance over time. 252 * 253 * It's possible for an allocator to handle more allocations than 254 * its max. This can occur when gang blocks are required or when other 255 * groups are unable to handle their share of allocations. 256 */ 257 uint64_t mg_max_alloc_queue_depth; 258 uint64_t *mg_cur_max_alloc_queue_depth; 259 refcount_t *mg_alloc_queue_depth; 260 int mg_allocators; 261 /* 262 * A metalab group that can no longer allocate the minimum block 263 * size will set mg_no_free_space. Once a metaslab group is out 264 * of space then its share of work must be distributed to other 265 * groups. 266 */ 267 boolean_t mg_no_free_space; 268 269 uint64_t mg_allocations; 270 uint64_t mg_failed_allocations; 271 uint64_t mg_fragmentation; 272 uint64_t mg_histogram[RANGE_TREE_HISTOGRAM_SIZE]; 273 }; 274 275 /* 276 * This value defines the number of elements in the ms_lbas array. The value 277 * of 64 was chosen as it covers all power of 2 buckets up to UINT64_MAX. 278 * This is the equivalent of highbit(UINT64_MAX). 279 */ 280 #define MAX_LBAS 64 281 282 /* 283 * Each metaslab maintains a set of in-core trees to track metaslab 284 * operations. The in-core free tree (ms_allocatable) contains the list of 285 * free segments which are eligible for allocation. As blocks are 286 * allocated, the allocated segment are removed from the ms_allocatable and 287 * added to a per txg allocation tree (ms_allocating). As blocks are 288 * freed, they are added to the free tree (ms_freeing). These trees 289 * allow us to process all allocations and frees in syncing context 290 * where it is safe to update the on-disk space maps. An additional set 291 * of in-core trees is maintained to track deferred frees 292 * (ms_defer). Once a block is freed it will move from the 293 * ms_freed to the ms_defer tree. A deferred free means that a block 294 * has been freed but cannot be used by the pool until TXG_DEFER_SIZE 295 * transactions groups later. For example, a block that is freed in txg 296 * 50 will not be available for reallocation until txg 52 (50 + 297 * TXG_DEFER_SIZE). This provides a safety net for uberblock rollback. 298 * A pool could be safely rolled back TXG_DEFERS_SIZE transactions 299 * groups and ensure that no block has been reallocated. 300 * 301 * The simplified transition diagram looks like this: 302 * 303 * 304 * ALLOCATE 305 * | 306 * V 307 * free segment (ms_allocatable) -> ms_allocating[4] -> (write to space map) 308 * ^ 309 * | ms_freeing <--- FREE 310 * | | 311 * | v 312 * | ms_freed 313 * | | 314 * +-------- ms_defer[2] <-------+-------> (write to space map) 315 * 316 * 317 * Each metaslab's space is tracked in a single space map in the MOS, 318 * which is only updated in syncing context. Each time we sync a txg, 319 * we append the allocs and frees from that txg to the space map. The 320 * pool space is only updated once all metaslabs have finished syncing. 321 * 322 * To load the in-core free tree we read the space map from disk. This 323 * object contains a series of alloc and free records that are combined 324 * to make up the list of all free segments in this metaslab. These 325 * segments are represented in-core by the ms_allocatable and are stored 326 * in an AVL tree. 327 * 328 * As the space map grows (as a result of the appends) it will 329 * eventually become space-inefficient. When the metaslab's in-core 330 * free tree is zfs_condense_pct/100 times the size of the minimal 331 * on-disk representation, we rewrite it in its minimized form. If a 332 * metaslab needs to condense then we must set the ms_condensing flag to 333 * ensure that allocations are not performed on the metaslab that is 334 * being written. 335 */ 336 struct metaslab { 337 kmutex_t ms_lock; 338 kmutex_t ms_sync_lock; 339 kcondvar_t ms_load_cv; 340 space_map_t *ms_sm; 341 uint64_t ms_id; 342 uint64_t ms_start; 343 uint64_t ms_size; 344 uint64_t ms_fragmentation; 345 346 range_tree_t *ms_allocating[TXG_SIZE]; 347 range_tree_t *ms_allocatable; 348 349 /* 350 * The following range trees are accessed only from syncing context. 351 * ms_free*tree only have entries while syncing, and are empty 352 * between syncs. 353 */ 354 range_tree_t *ms_freeing; /* to free this syncing txg */ 355 range_tree_t *ms_freed; /* already freed this syncing txg */ 356 range_tree_t *ms_defer[TXG_DEFER_SIZE]; 357 range_tree_t *ms_checkpointing; /* to add to the checkpoint */ 358 359 boolean_t ms_condensing; /* condensing? */ 360 boolean_t ms_condense_wanted; 361 uint64_t ms_condense_checked_txg; 362 363 /* 364 * We must hold both ms_lock and ms_group->mg_lock in order to 365 * modify ms_loaded. 366 */ 367 boolean_t ms_loaded; 368 boolean_t ms_loading; 369 370 int64_t ms_deferspace; /* sum of ms_defermap[] space */ 371 uint64_t ms_weight; /* weight vs. others in group */ 372 uint64_t ms_activation_weight; /* activation weight */ 373 374 /* 375 * Track of whenever a metaslab is selected for loading or allocation. 376 * We use this value to determine how long the metaslab should 377 * stay cached. 378 */ 379 uint64_t ms_selected_txg; 380 381 uint64_t ms_alloc_txg; /* last successful alloc (debug only) */ 382 uint64_t ms_max_size; /* maximum allocatable size */ 383 384 /* 385 * -1 if it's not active in an allocator, otherwise set to the allocator 386 * this metaslab is active for. 387 */ 388 int ms_allocator; 389 boolean_t ms_primary; /* Only valid if ms_allocator is not -1 */ 390 391 /* 392 * The metaslab block allocators can optionally use a size-ordered 393 * range tree and/or an array of LBAs. Not all allocators use 394 * this functionality. The ms_allocatable_by_size should always 395 * contain the same number of segments as the ms_allocatable. The 396 * only difference is that the ms_allocatable_by_size is ordered by 397 * segment sizes. 398 */ 399 avl_tree_t ms_allocatable_by_size; 400 uint64_t ms_lbas[MAX_LBAS]; 401 402 metaslab_group_t *ms_group; /* metaslab group */ 403 avl_node_t ms_group_node; /* node in metaslab group tree */ 404 txg_node_t ms_txg_node; /* per-txg dirty metaslab links */ 405 406 boolean_t ms_new; 407 }; 408 409 #ifdef __cplusplus 410 } 411 #endif 412 413 #endif /* _SYS_METASLAB_IMPL_H */ 414