1 /* $NetBSD: rf_raid.h,v 1.14 2001/10/04 15:58:55 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 46 #include "rf_alloclist.h" 47 #include "rf_stripelocks.h" 48 #include "rf_layout.h" 49 #include "rf_disks.h" 50 #include "rf_debugMem.h" 51 #include "rf_diskqueue.h" 52 #include "rf_reconstruct.h" 53 #include "rf_acctrace.h" 54 55 #if RF_INCLUDE_PARITYLOGGING > 0 56 #include "rf_paritylog.h" 57 #endif /* RF_INCLUDE_PARITYLOGGING > 0 */ 58 59 #define RF_COMPONENT_LABEL_VERSION_1 1 60 #define RF_COMPONENT_LABEL_VERSION 2 61 #define RF_RAID_DIRTY 0 62 #define RF_RAID_CLEAN 1 63 64 65 /* 66 * Each row in the array is a distinct parity group, so 67 * each has it's own status, which is one of the following. 68 */ 69 typedef enum RF_RowStatus_e { 70 rf_rs_optimal, 71 rf_rs_degraded, 72 rf_rs_reconstructing, 73 rf_rs_reconfigured 74 } RF_RowStatus_t; 75 76 struct RF_CumulativeStats_s { 77 struct timeval start; /* the time when the stats were last started */ 78 struct timeval stop; /* the time when the stats were last stopped */ 79 long sum_io_us; /* sum of all user response times (us) */ 80 long num_ios; /* total number of I/Os serviced */ 81 long num_sect_moved; /* total number of sectors read or written */ 82 }; 83 84 struct RF_ThroughputStats_s { 85 RF_DECLARE_MUTEX(mutex) /* a mutex used to lock the configuration 86 * stuff */ 87 struct timeval start; /* timer started when numOutstandingRequests 88 * moves from 0 to 1 */ 89 struct timeval stop; /* timer stopped when numOutstandingRequests 90 * moves from 1 to 0 */ 91 RF_uint64 sum_io_us; /* total time timer is enabled */ 92 RF_uint64 num_ios; /* total number of ios processed by RAIDframe */ 93 long num_out_ios; /* number of outstanding ios */ 94 }; 95 96 struct RF_Raid_s { 97 /* This portion never changes, and can be accessed without locking */ 98 /* an exception is Disks[][].status, which requires locking when it is 99 * changed. XXX this is no longer true. numSpare and friends can 100 * change now. 101 */ 102 u_int numRow; /* number of rows of disks, typically == # of 103 * ranks */ 104 u_int numCol; /* number of columns of disks, typically == # 105 * of disks/rank */ 106 u_int numSpare; /* number of spare disks */ 107 int maxQueueDepth; /* max disk queue depth */ 108 RF_SectorCount_t totalSectors; /* total number of sectors in the 109 * array */ 110 RF_SectorCount_t sectorsPerDisk; /* number of sectors on each 111 * disk */ 112 u_int logBytesPerSector; /* base-2 log of the number of bytes 113 * in a sector */ 114 u_int bytesPerSector; /* bytes in a sector */ 115 RF_int32 sectorMask; /* mask of bytes-per-sector */ 116 117 RF_RaidLayout_t Layout; /* all information related to layout */ 118 RF_RaidDisk_t **Disks; /* all information related to physical disks */ 119 RF_DiskQueue_t **Queues;/* all information related to disk queues */ 120 RF_DiskQueueSW_t *qType;/* pointer to the DiskQueueSW used for the 121 component queues. */ 122 /* NOTE: This is an anchor point via which the queues can be 123 * accessed, but the enqueue/dequeue routines in diskqueue.c use a 124 * local copy of this pointer for the actual accesses. */ 125 /* The remainder of the structure can change, and therefore requires 126 * locking on reads and updates */ 127 RF_DECLARE_MUTEX(mutex) /* mutex used to serialize access to 128 * the fields below */ 129 RF_RowStatus_t *status; /* the status of each row in the array */ 130 int valid; /* indicates successful configuration */ 131 RF_LockTableEntry_t *lockTable; /* stripe-lock table */ 132 RF_LockTableEntry_t *quiesceLock; /* quiesnce table */ 133 int numFailures; /* total number of failures in the array */ 134 int numNewFailures; /* number of *new* failures (that havn't 135 caused a mod_counter update */ 136 137 int parity_good; /* !0 if parity is known to be correct */ 138 int serial_number; /* a "serial number" for this set */ 139 int mod_counter; /* modification counter for component labels */ 140 int clean; /* the clean bit for this array. */ 141 142 int openings; /* Number of IO's which can be scheduled 143 simultaneously (high-level - not a 144 per-component limit)*/ 145 146 int maxOutstanding; /* maxOutstanding requests (per-component) */ 147 int autoconfigure; /* automatically configure this RAID set. 148 0 == no, 1 == yes */ 149 int root_partition; /* Use this set as / 150 0 == no, 1 == yes*/ 151 int last_unit; /* last unit number (e.g. 0 for /dev/raid0) 152 of this component. Used for autoconfigure 153 only. */ 154 int config_order; /* 0 .. n. The order in which the component 155 should be auto-configured. E.g. 0 is will 156 done first, (and would become raid0). 157 This may be in conflict with last_unit!!?! */ 158 /* Not currently used. */ 159 160 /* 161 * Cleanup stuff 162 */ 163 RF_ShutdownList_t *shutdownList; /* shutdown activities */ 164 RF_AllocListElem_t *cleanupList; /* memory to be freed at 165 * shutdown time */ 166 167 /* 168 * Recon stuff 169 */ 170 RF_HeadSepLimit_t headSepLimit; 171 int numFloatingReconBufs; 172 int reconInProgress; 173 RF_DECLARE_COND(waitForReconCond) 174 RF_RaidReconDesc_t *reconDesc; /* reconstruction descriptor */ 175 RF_ReconCtrl_t **reconControl; /* reconstruction control structure 176 * pointers for each row in the array */ 177 178 /* 179 * Array-quiescence stuff 180 */ 181 RF_DECLARE_MUTEX(access_suspend_mutex) 182 RF_DECLARE_COND(quiescent_cond) 183 RF_IoCount_t accesses_suspended; 184 RF_IoCount_t accs_in_flight; 185 int access_suspend_release; 186 int waiting_for_quiescence; 187 RF_CallbackDesc_t *quiesce_wait_list; 188 189 /* 190 * Statistics 191 */ 192 #if !defined(_KERNEL) && !defined(SIMULATE) 193 RF_ThroughputStats_t throughputstats; 194 #endif /* !KERNEL && !SIMULATE */ 195 RF_CumulativeStats_t userstats; 196 int parity_rewrite_stripes_done; 197 int recon_stripes_done; 198 int copyback_stripes_done; 199 200 int recon_in_progress; 201 int parity_rewrite_in_progress; 202 int copyback_in_progress; 203 204 /* 205 * Engine thread control 206 */ 207 RF_DECLARE_MUTEX(node_queue_mutex) 208 RF_DECLARE_COND(node_queue_cond) 209 RF_DagNode_t *node_queue; 210 RF_Thread_t parity_rewrite_thread; 211 RF_Thread_t copyback_thread; 212 RF_Thread_t engine_thread; 213 RF_Thread_t recon_thread; 214 RF_ThreadGroup_t engine_tg; 215 int shutdown_engine; 216 int dags_in_flight; /* debug */ 217 218 /* 219 * PSS (Parity Stripe Status) stuff 220 */ 221 RF_FreeList_t *pss_freelist; 222 long pssTableSize; 223 224 /* 225 * Reconstruction stuff 226 */ 227 int procsInBufWait; 228 int numFullReconBuffers; 229 RF_AccTraceEntry_t *recon_tracerecs; 230 unsigned long accumXorTimeUs; 231 RF_ReconDoneProc_t *recon_done_procs; 232 RF_DECLARE_MUTEX(recon_done_proc_mutex) 233 /* 234 * nAccOutstanding, waitShutdown protected by desc freelist lock 235 * (This may seem strange, since that's a central serialization point 236 * for a per-array piece of data, but otherwise, it'd be an extra 237 * per-array lock, and that'd only be less efficient...) 238 */ 239 RF_DECLARE_COND(outstandingCond) 240 int waitShutdown; 241 int nAccOutstanding; 242 243 RF_DiskId_t **diskids; 244 RF_DiskId_t *sparediskids; 245 246 int raidid; 247 RF_AccTotals_t acc_totals; 248 int keep_acc_totals; 249 250 struct raidcinfo **raid_cinfo; /* array of component info */ 251 252 int terminate_disk_queues; 253 254 /* 255 * XXX 256 * 257 * config-specific information should be moved 258 * somewhere else, or at least hung off this 259 * in some generic way 260 */ 261 262 /* used by rf_compute_workload_shift */ 263 RF_RowCol_t hist_diskreq[RF_MAXROW][RF_MAXCOL]; 264 265 /* used by declustering */ 266 int noRotate; 267 268 #if RF_INCLUDE_PARITYLOGGING > 0 269 /* used by parity logging */ 270 RF_SectorCount_t regionLogCapacity; 271 RF_ParityLogQueue_t parityLogPool; /* pool of unused parity logs */ 272 RF_RegionInfo_t *regionInfo; /* array of region state */ 273 int numParityLogs; 274 int numSectorsPerLog; 275 int regionParityRange; 276 int logsInUse; /* debugging */ 277 RF_ParityLogDiskQueue_t parityLogDiskQueue; /* state of parity 278 * logging disk work */ 279 RF_RegionBufferQueue_t regionBufferPool; /* buffers for holding 280 * region log */ 281 RF_RegionBufferQueue_t parityBufferPool; /* buffers for holding 282 * parity */ 283 caddr_t parityLogBufferHeap; /* pool of unused parity logs */ 284 RF_Thread_t pLogDiskThreadHandle; 285 286 #endif /* RF_INCLUDE_PARITYLOGGING > 0 */ 287 }; 288 #endif /* !_RF__RF_RAID_H_ */ 289