1 /* $NetBSD: rf_raid.h,v 1.17 2002/10/22 03:15:28 oster Exp $ */ 2 /* 3 * Copyright (c) 1995 Carnegie-Mellon University. 4 * All rights reserved. 5 * 6 * Author: Mark Holland 7 * 8 * Permission to use, copy, modify and distribute this software and 9 * its documentation is hereby granted, provided that both the copyright 10 * notice and this permission notice appear in all copies of the 11 * software, derivative works or modified versions, and any portions 12 * thereof, and that both notices appear in supporting documentation. 13 * 14 * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS" 15 * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND 16 * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE. 17 * 18 * Carnegie Mellon requests users of this software to return to 19 * 20 * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU 21 * School of Computer Science 22 * Carnegie Mellon University 23 * Pittsburgh PA 15213-3890 24 * 25 * any improvements or extensions that they make and grant Carnegie the 26 * rights to redistribute these changes. 27 */ 28 29 /********************************************** 30 * rf_raid.h -- main header file for RAID driver 31 **********************************************/ 32 33 34 #ifndef _RF__RF_RAID_H_ 35 #define _RF__RF_RAID_H_ 36 37 #include <dev/raidframe/raidframevar.h> 38 #include "rf_archs.h" 39 #include "rf_threadstuff.h" 40 41 #include "rf_netbsd.h" 42 43 #include <sys/disklabel.h> 44 #include <sys/types.h> 45 #include <sys/queue.h> 46 47 #include "rf_alloclist.h" 48 #include "rf_stripelocks.h" 49 #include "rf_layout.h" 50 #include "rf_disks.h" 51 #include "rf_debugMem.h" 52 #include "rf_diskqueue.h" 53 #include "rf_reconstruct.h" 54 #include "rf_acctrace.h" 55 #include "rf_fifo.h" 56 57 #if RF_INCLUDE_PARITYLOGGING > 0 58 #include "rf_paritylog.h" 59 #endif /* RF_INCLUDE_PARITYLOGGING > 0 */ 60 61 #define RF_COMPONENT_LABEL_VERSION_1 1 62 #define RF_COMPONENT_LABEL_VERSION 2 63 #define RF_RAID_DIRTY 0 64 #define RF_RAID_CLEAN 1 65 66 67 /* 68 * Each row in the array is a distinct parity group, so 69 * each has it's own status, which is one of the following. 70 */ 71 typedef enum RF_RowStatus_e { 72 rf_rs_optimal, 73 rf_rs_degraded, 74 rf_rs_reconstructing, 75 rf_rs_reconfigured 76 } RF_RowStatus_t; 77 78 struct RF_CumulativeStats_s { 79 struct timeval start; /* the time when the stats were last started */ 80 struct timeval stop; /* the time when the stats were last stopped */ 81 long sum_io_us; /* sum of all user response times (us) */ 82 long num_ios; /* total number of I/Os serviced */ 83 long num_sect_moved; /* total number of sectors read or written */ 84 }; 85 86 struct RF_ThroughputStats_s { 87 RF_DECLARE_MUTEX(mutex) /* a mutex used to lock the configuration 88 * stuff */ 89 struct timeval start; /* timer started when numOutstandingRequests 90 * moves from 0 to 1 */ 91 struct timeval stop; /* timer stopped when numOutstandingRequests 92 * moves from 1 to 0 */ 93 RF_uint64 sum_io_us; /* total time timer is enabled */ 94 RF_uint64 num_ios; /* total number of ios processed by RAIDframe */ 95 long num_out_ios; /* number of outstanding ios */ 96 }; 97 98 struct RF_Raid_s { 99 /* This portion never changes, and can be accessed without locking */ 100 /* an exception is Disks[][].status, which requires locking when it is 101 * changed. XXX this is no longer true. numSpare and friends can 102 * change now. 103 */ 104 u_int numRow; /* number of rows of disks, typically == # of 105 * ranks */ 106 u_int numCol; /* number of columns of disks, typically == # 107 * of disks/rank */ 108 u_int numSpare; /* number of spare disks */ 109 int maxQueueDepth; /* max disk queue depth */ 110 RF_SectorCount_t totalSectors; /* total number of sectors in the 111 * array */ 112 RF_SectorCount_t sectorsPerDisk; /* number of sectors on each 113 * disk */ 114 u_int logBytesPerSector; /* base-2 log of the number of bytes 115 * in a sector */ 116 u_int bytesPerSector; /* bytes in a sector */ 117 RF_int32 sectorMask; /* mask of bytes-per-sector */ 118 119 RF_RaidLayout_t Layout; /* all information related to layout */ 120 RF_RaidDisk_t **Disks; /* all information related to physical disks */ 121 RF_DiskQueue_t **Queues;/* all information related to disk queues */ 122 RF_DiskQueueSW_t *qType;/* pointer to the DiskQueueSW used for the 123 component queues. */ 124 /* NOTE: This is an anchor point via which the queues can be 125 * accessed, but the enqueue/dequeue routines in diskqueue.c use a 126 * local copy of this pointer for the actual accesses. */ 127 /* The remainder of the structure can change, and therefore requires 128 * locking on reads and updates */ 129 RF_DECLARE_MUTEX(mutex) /* mutex used to serialize access to 130 * the fields below */ 131 RF_RowStatus_t *status; /* the status of each row in the array */ 132 int valid; /* indicates successful configuration */ 133 RF_LockTableEntry_t *lockTable; /* stripe-lock table */ 134 RF_LockTableEntry_t *quiesceLock; /* quiesnce table */ 135 int numFailures; /* total number of failures in the array */ 136 int numNewFailures; /* number of *new* failures (that havn't 137 caused a mod_counter update */ 138 139 int parity_good; /* !0 if parity is known to be correct */ 140 int serial_number; /* a "serial number" for this set */ 141 int mod_counter; /* modification counter for component labels */ 142 int clean; /* the clean bit for this array. */ 143 144 int openings; /* Number of IO's which can be scheduled 145 simultaneously (high-level - not a 146 per-component limit)*/ 147 148 int maxOutstanding; /* maxOutstanding requests (per-component) */ 149 int autoconfigure; /* automatically configure this RAID set. 150 0 == no, 1 == yes */ 151 int root_partition; /* Use this set as / 152 0 == no, 1 == yes*/ 153 int last_unit; /* last unit number (e.g. 0 for /dev/raid0) 154 of this component. Used for autoconfigure 155 only. */ 156 int config_order; /* 0 .. n. The order in which the component 157 should be auto-configured. E.g. 0 is will 158 done first, (and would become raid0). 159 This may be in conflict with last_unit!!?! */ 160 /* Not currently used. */ 161 162 /* queue to gather up requests from KernelWakeupFunc() and let 163 a kernel thread deal with calling rf_DiskIOComplete and any 164 callback functions. */ 165 TAILQ_HEAD(iodone_q,RF_DiskQueueData_s) iodone; 166 /* and a lock to protect it */ 167 struct simplelock iodone_lock; 168 169 /* 170 * Cleanup stuff 171 */ 172 RF_ShutdownList_t *shutdownList; /* shutdown activities */ 173 RF_AllocListElem_t *cleanupList; /* memory to be freed at 174 * shutdown time */ 175 176 /* 177 * Recon stuff 178 */ 179 RF_HeadSepLimit_t headSepLimit; 180 int numFloatingReconBufs; 181 int reconInProgress; 182 RF_DECLARE_COND(waitForReconCond) 183 RF_RaidReconDesc_t *reconDesc; /* reconstruction descriptor */ 184 RF_ReconCtrl_t **reconControl; /* reconstruction control structure 185 * pointers for each row in the array */ 186 187 /* 188 * Array-quiescence stuff 189 */ 190 RF_DECLARE_MUTEX(access_suspend_mutex) 191 RF_DECLARE_COND(quiescent_cond) 192 RF_IoCount_t accesses_suspended; 193 RF_IoCount_t accs_in_flight; 194 int access_suspend_release; 195 int waiting_for_quiescence; 196 RF_CallbackDesc_t *quiesce_wait_list; 197 198 /* 199 * Statistics 200 */ 201 #if !defined(_KERNEL) && !defined(SIMULATE) 202 RF_ThroughputStats_t throughputstats; 203 #endif /* !KERNEL && !SIMULATE */ 204 RF_CumulativeStats_t userstats; 205 int parity_rewrite_stripes_done; 206 int recon_stripes_done; 207 int copyback_stripes_done; 208 209 int recon_in_progress; 210 int parity_rewrite_in_progress; 211 int copyback_in_progress; 212 int adding_hot_spare; 213 214 /* 215 * Engine thread control 216 */ 217 RF_DECLARE_MUTEX(node_queue_mutex) 218 RF_DECLARE_COND(node_queue_cond) 219 RF_DagNode_t *node_queue; 220 RF_Thread_t parity_rewrite_thread; 221 RF_Thread_t copyback_thread; 222 RF_Thread_t engine_thread; 223 RF_Thread_t engine_helper_thread; 224 RF_Thread_t recon_thread; 225 RF_ThreadGroup_t engine_tg; 226 int shutdown_engine; 227 int shutdown_raidio; 228 int dags_in_flight; /* debug */ 229 230 /* 231 * PSS (Parity Stripe Status) stuff 232 */ 233 long pssTableSize; 234 struct pool pss_pool; 235 struct pool pss_issued_pool; 236 237 /* 238 * Reconstruction stuff 239 */ 240 int procsInBufWait; 241 int numFullReconBuffers; 242 RF_AccTraceEntry_t *recon_tracerecs; 243 unsigned long accumXorTimeUs; 244 RF_ReconDoneProc_t *recon_done_procs; 245 RF_DECLARE_MUTEX(recon_done_proc_mutex) 246 /* 247 * nAccOutstanding, waitShutdown protected by desc freelist lock 248 * (This may seem strange, since that's a central serialization point 249 * for a per-array piece of data, but otherwise, it'd be an extra 250 * per-array lock, and that'd only be less efficient...) 251 */ 252 RF_DECLARE_COND(outstandingCond) 253 int waitShutdown; 254 int nAccOutstanding; 255 256 RF_DiskId_t **diskids; 257 RF_DiskId_t *sparediskids; 258 259 int raidid; 260 RF_AccTotals_t acc_totals; 261 int keep_acc_totals; 262 263 struct raidcinfo **raid_cinfo; /* array of component info */ 264 265 int terminate_disk_queues; 266 267 /* 268 * XXX 269 * 270 * config-specific information should be moved 271 * somewhere else, or at least hung off this 272 * in some generic way 273 */ 274 275 /* used by rf_compute_workload_shift */ 276 RF_RowCol_t hist_diskreq[RF_MAXROW][RF_MAXCOL]; 277 278 /* used by declustering */ 279 int noRotate; 280 281 #if RF_INCLUDE_PARITYLOGGING > 0 282 /* used by parity logging */ 283 RF_SectorCount_t regionLogCapacity; 284 RF_ParityLogQueue_t parityLogPool; /* pool of unused parity logs */ 285 RF_RegionInfo_t *regionInfo; /* array of region state */ 286 int numParityLogs; 287 int numSectorsPerLog; 288 int regionParityRange; 289 int logsInUse; /* debugging */ 290 RF_ParityLogDiskQueue_t parityLogDiskQueue; /* state of parity 291 * logging disk work */ 292 RF_RegionBufferQueue_t regionBufferPool; /* buffers for holding 293 * region log */ 294 RF_RegionBufferQueue_t parityBufferPool; /* buffers for holding 295 * parity */ 296 caddr_t parityLogBufferHeap; /* pool of unused parity logs */ 297 RF_Thread_t pLogDiskThreadHandle; 298 299 #endif /* RF_INCLUDE_PARITYLOGGING > 0 */ 300 }; 301 #endif /* !_RF__RF_RAID_H_ */ 302