rf_layout.h revision 1.2 1 1.2 oster /* $NetBSD: rf_layout.h,v 1.2 1999/01/26 02:33:58 oster Exp $ */
2 1.1 oster /*
3 1.1 oster * Copyright (c) 1995 Carnegie-Mellon University.
4 1.1 oster * All rights reserved.
5 1.1 oster *
6 1.1 oster * Author: Mark Holland
7 1.1 oster *
8 1.1 oster * Permission to use, copy, modify and distribute this software and
9 1.1 oster * its documentation is hereby granted, provided that both the copyright
10 1.1 oster * notice and this permission notice appear in all copies of the
11 1.1 oster * software, derivative works or modified versions, and any portions
12 1.1 oster * thereof, and that both notices appear in supporting documentation.
13 1.1 oster *
14 1.1 oster * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
15 1.1 oster * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
16 1.1 oster * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
17 1.1 oster *
18 1.1 oster * Carnegie Mellon requests users of this software to return to
19 1.1 oster *
20 1.1 oster * Software Distribution Coordinator or Software.Distribution (at) CS.CMU.EDU
21 1.1 oster * School of Computer Science
22 1.1 oster * Carnegie Mellon University
23 1.1 oster * Pittsburgh PA 15213-3890
24 1.1 oster *
25 1.1 oster * any improvements or extensions that they make and grant Carnegie the
26 1.1 oster * rights to redistribute these changes.
27 1.1 oster */
28 1.1 oster
29 1.1 oster /* rf_layout.h -- header file defining layout data structures
30 1.1 oster */
31 1.1 oster
32 1.1 oster #ifndef _RF__RF_LAYOUT_H_
33 1.1 oster #define _RF__RF_LAYOUT_H_
34 1.1 oster
35 1.1 oster #include "rf_types.h"
36 1.1 oster #include "rf_archs.h"
37 1.1 oster #include "rf_alloclist.h"
38 1.1 oster
39 1.2 oster #ifndef _KERNEL
40 1.2 oster #include <stdio.h>
41 1.2 oster #endif
42 1.2 oster
43 1.1 oster /*****************************************************************************************
44 1.1 oster *
45 1.1 oster * This structure identifies all layout-specific operations and parameters.
46 1.1 oster *
47 1.1 oster ****************************************************************************************/
48 1.1 oster
49 1.1 oster typedef struct RF_LayoutSW_s {
50 1.1 oster RF_ParityConfig_t parityConfig;
51 1.1 oster char *configName;
52 1.1 oster
53 1.2 oster #ifndef _KERNEL
54 1.1 oster /* layout-specific parsing */
55 1.1 oster int (*MakeLayoutSpecific)(FILE *fp, RF_Config_t *cfgPtr, void *arg);
56 1.1 oster void *makeLayoutSpecificArg;
57 1.1 oster #endif /* !KERNEL */
58 1.1 oster
59 1.1 oster #if RF_UTILITY == 0
60 1.1 oster /* initialization routine */
61 1.1 oster int (*Configure)(RF_ShutdownList_t **shutdownListp, RF_Raid_t *raidPtr, RF_Config_t *cfgPtr);
62 1.1 oster
63 1.1 oster /* routine to map RAID sector address -> physical (row, col, offset) */
64 1.1 oster void (*MapSector)(RF_Raid_t *raidPtr, RF_RaidAddr_t raidSector,
65 1.1 oster RF_RowCol_t *row, RF_RowCol_t *col, RF_SectorNum_t *diskSector, int remap);
66 1.1 oster
67 1.1 oster /* routine to map RAID sector address -> physical (r,c,o) of parity unit */
68 1.1 oster void (*MapParity)(RF_Raid_t *raidPtr, RF_RaidAddr_t raidSector,
69 1.1 oster RF_RowCol_t *row, RF_RowCol_t *col, RF_SectorNum_t *diskSector, int remap);
70 1.1 oster
71 1.1 oster /* routine to map RAID sector address -> physical (r,c,o) of Q unit */
72 1.1 oster void (*MapQ)(RF_Raid_t *raidPtr, RF_RaidAddr_t raidSector, RF_RowCol_t *row,
73 1.1 oster RF_RowCol_t *col, RF_SectorNum_t *diskSector, int remap);
74 1.1 oster
75 1.1 oster /* routine to identify the disks comprising a stripe */
76 1.1 oster void (*IdentifyStripe)(RF_Raid_t *raidPtr, RF_RaidAddr_t addr,
77 1.1 oster RF_RowCol_t **diskids, RF_RowCol_t *outRow);
78 1.1 oster
79 1.1 oster /* routine to select a dag */
80 1.1 oster void (*SelectionFunc)(RF_Raid_t *raidPtr, RF_IoType_t type,
81 1.1 oster RF_AccessStripeMap_t *asmap,
82 1.1 oster RF_VoidFuncPtr *);
83 1.1 oster #if 0
84 1.1 oster void (**createFunc)(RF_Raid_t *,
85 1.1 oster RF_AccessStripeMap_t *,
86 1.1 oster RF_DagHeader_t *, void *,
87 1.1 oster RF_RaidAccessFlags_t,
88 1.1 oster RF_AllocListElem_t *));
89 1.1 oster
90 1.1 oster #endif
91 1.1 oster
92 1.1 oster /* map a stripe ID to a parity stripe ID. This is typically the identity mapping */
93 1.1 oster void (*MapSIDToPSID)(RF_RaidLayout_t *layoutPtr, RF_StripeNum_t stripeID,
94 1.1 oster RF_StripeNum_t *psID, RF_ReconUnitNum_t *which_ru);
95 1.1 oster
96 1.1 oster /* get default head separation limit (may be NULL) */
97 1.1 oster RF_HeadSepLimit_t (*GetDefaultHeadSepLimit)(RF_Raid_t *raidPtr);
98 1.1 oster
99 1.1 oster /* get default num recon buffers (may be NULL) */
100 1.1 oster int (*GetDefaultNumFloatingReconBuffers)(RF_Raid_t *raidPtr);
101 1.1 oster
102 1.1 oster /* get number of spare recon units (may be NULL) */
103 1.1 oster RF_ReconUnitCount_t (*GetNumSpareRUs)(RF_Raid_t *raidPtr);
104 1.1 oster
105 1.1 oster /* spare table installation (may be NULL) */
106 1.1 oster int (*InstallSpareTable)(RF_Raid_t *raidPtr, RF_RowCol_t frow, RF_RowCol_t fcol);
107 1.1 oster
108 1.1 oster /* recon buffer submission function */
109 1.1 oster int (*SubmitReconBuffer)(RF_ReconBuffer_t *rbuf, int keep_it,
110 1.1 oster int use_committed);
111 1.1 oster
112 1.1 oster /*
113 1.1 oster * verify that parity information for a stripe is correct
114 1.1 oster * see rf_parityscan.h for return vals
115 1.1 oster */
116 1.1 oster int (*VerifyParity)(RF_Raid_t *raidPtr, RF_RaidAddr_t raidAddr,
117 1.1 oster RF_PhysDiskAddr_t *parityPDA, int correct_it, RF_RaidAccessFlags_t flags);
118 1.1 oster
119 1.1 oster /* number of faults tolerated by this mapping */
120 1.1 oster int faultsTolerated;
121 1.1 oster
122 1.1 oster /* states to step through in an access. Must end with "LastState".
123 1.1 oster * The default is DefaultStates in rf_layout.c */
124 1.1 oster RF_AccessState_t *states;
125 1.1 oster
126 1.1 oster RF_AccessStripeMapFlags_t flags;
127 1.1 oster #endif /* RF_UTILITY == 0 */
128 1.1 oster } RF_LayoutSW_t;
129 1.1 oster
130 1.1 oster /* enables remapping to spare location under dist sparing */
131 1.1 oster #define RF_REMAP 1
132 1.1 oster #define RF_DONT_REMAP 0
133 1.1 oster
134 1.1 oster /*
135 1.1 oster * Flags values for RF_AccessStripeMapFlags_t
136 1.1 oster */
137 1.1 oster #define RF_NO_STRIPE_LOCKS 0x0001 /* suppress stripe locks */
138 1.1 oster #define RF_DISTRIBUTE_SPARE 0x0002 /* distribute spare space in archs that support it */
139 1.1 oster #define RF_BD_DECLUSTERED 0x0004 /* declustering uses block designs */
140 1.1 oster
141 1.1 oster /*************************************************************************
142 1.1 oster *
143 1.1 oster * this structure forms the layout component of the main Raid
144 1.1 oster * structure. It describes everything needed to define and perform
145 1.1 oster * the mapping of logical RAID addresses <-> physical disk addresses.
146 1.1 oster *
147 1.1 oster *************************************************************************/
148 1.1 oster struct RF_RaidLayout_s {
149 1.1 oster /* configuration parameters */
150 1.1 oster RF_SectorCount_t sectorsPerStripeUnit; /* number of sectors in one stripe unit */
151 1.1 oster RF_StripeCount_t SUsPerPU; /* stripe units per parity unit */
152 1.1 oster RF_StripeCount_t SUsPerRU; /* stripe units per reconstruction unit */
153 1.1 oster
154 1.1 oster /* redundant-but-useful info computed from the above, used in all layouts */
155 1.1 oster RF_StripeCount_t numStripe; /* total number of stripes in the array */
156 1.1 oster RF_SectorCount_t dataSectorsPerStripe;
157 1.1 oster RF_StripeCount_t dataStripeUnitsPerDisk;
158 1.1 oster u_int bytesPerStripeUnit;
159 1.1 oster u_int dataBytesPerStripe;
160 1.1 oster RF_StripeCount_t numDataCol; /* number of SUs of data per stripe (name here is a la RAID4) */
161 1.1 oster RF_StripeCount_t numParityCol; /* number of SUs of parity per stripe. Always 1 for now */
162 1.1 oster RF_StripeCount_t numParityLogCol; /* number of SUs of parity log per stripe. Always 1 for now */
163 1.1 oster RF_StripeCount_t stripeUnitsPerDisk;
164 1.1 oster
165 1.1 oster RF_LayoutSW_t *map; /* ptr to struct holding mapping fns and information */
166 1.1 oster void *layoutSpecificInfo; /* ptr to a structure holding layout-specific params */
167 1.1 oster };
168 1.1 oster
169 1.1 oster /*****************************************************************************************
170 1.1 oster *
171 1.1 oster * The mapping code returns a pointer to a list of AccessStripeMap structures, which
172 1.1 oster * describes all the mapping information about an access. The list contains one
173 1.1 oster * AccessStripeMap structure per stripe touched by the access. Each element in the list
174 1.1 oster * contains a stripe identifier and a pointer to a list of PhysDiskAddr structuress. Each
175 1.1 oster * element in this latter list describes the physical location of a stripe unit accessed
176 1.1 oster * within the corresponding stripe.
177 1.1 oster *
178 1.1 oster ****************************************************************************************/
179 1.1 oster
180 1.1 oster #define RF_PDA_TYPE_DATA 0
181 1.1 oster #define RF_PDA_TYPE_PARITY 1
182 1.1 oster #define RF_PDA_TYPE_Q 2
183 1.1 oster
184 1.1 oster struct RF_PhysDiskAddr_s {
185 1.1 oster RF_RowCol_t row,col; /* disk identifier */
186 1.1 oster RF_SectorNum_t startSector; /* sector offset into the disk */
187 1.1 oster RF_SectorCount_t numSector; /* number of sectors accessed */
188 1.1 oster int type; /* used by higher levels: currently, data, parity, or q */
189 1.1 oster caddr_t bufPtr; /* pointer to buffer supplying/receiving data */
190 1.1 oster RF_RaidAddr_t raidAddress; /* raid address corresponding to this physical disk address */
191 1.1 oster RF_PhysDiskAddr_t *next;
192 1.1 oster };
193 1.1 oster
194 1.1 oster #define RF_MAX_FAILED_PDA RF_MAXCOL
195 1.1 oster
196 1.1 oster struct RF_AccessStripeMap_s {
197 1.1 oster RF_StripeNum_t stripeID; /* the stripe index */
198 1.1 oster RF_RaidAddr_t raidAddress; /* the starting raid address within this stripe */
199 1.1 oster RF_RaidAddr_t endRaidAddress; /* raid address one sector past the end of the access */
200 1.1 oster RF_SectorCount_t totalSectorsAccessed; /* total num sectors identified in physInfo list */
201 1.1 oster RF_StripeCount_t numStripeUnitsAccessed; /* total num elements in physInfo list */
202 1.1 oster int numDataFailed; /* number of failed data disks accessed */
203 1.1 oster int numParityFailed; /* number of failed parity disks accessed (0 or 1) */
204 1.1 oster int numQFailed; /* number of failed Q units accessed (0 or 1) */
205 1.1 oster RF_AccessStripeMapFlags_t flags; /* various flags */
206 1.1 oster #if 0
207 1.1 oster RF_PhysDiskAddr_t *failedPDA; /* points to the PDA that has failed */
208 1.1 oster RF_PhysDiskAddr_t *failedPDAtwo; /* points to the second PDA that has failed, if any */
209 1.1 oster #else
210 1.1 oster int numFailedPDAs; /* number of failed phys addrs */
211 1.1 oster RF_PhysDiskAddr_t *failedPDAs[RF_MAX_FAILED_PDA]; /* array of failed phys addrs */
212 1.1 oster #endif
213 1.1 oster RF_PhysDiskAddr_t *physInfo; /* a list of PhysDiskAddr structs */
214 1.1 oster RF_PhysDiskAddr_t *parityInfo; /* list of physical addrs for the parity (P of P + Q ) */
215 1.1 oster RF_PhysDiskAddr_t *qInfo; /* list of physical addrs for the Q of P + Q */
216 1.1 oster RF_LockReqDesc_t lockReqDesc; /* used for stripe locking */
217 1.1 oster RF_RowCol_t origRow; /* the original row: we may redirect the acc to a different row */
218 1.1 oster RF_AccessStripeMap_t *next;
219 1.1 oster };
220 1.1 oster
221 1.1 oster /* flag values */
222 1.1 oster #define RF_ASM_REDIR_LARGE_WRITE 0x00000001 /* allows large-write creation code to redirect failed accs */
223 1.1 oster #define RF_ASM_BAILOUT_DAG_USED 0x00000002 /* allows us to detect recursive calls to the bailout write dag */
224 1.1 oster #define RF_ASM_FLAGS_LOCK_TRIED 0x00000004 /* we've acquired the lock on the first parity range in this parity stripe */
225 1.1 oster #define RF_ASM_FLAGS_LOCK_TRIED2 0x00000008 /* we've acquired the lock on the 2nd parity range in this parity stripe */
226 1.1 oster #define RF_ASM_FLAGS_FORCE_TRIED 0x00000010 /* we've done the force-recon call on this parity stripe */
227 1.1 oster #define RF_ASM_FLAGS_RECON_BLOCKED 0x00000020 /* we blocked recon => we must unblock it later */
228 1.1 oster
229 1.1 oster struct RF_AccessStripeMapHeader_s {
230 1.1 oster RF_StripeCount_t numStripes; /* total number of stripes touched by this acc */
231 1.1 oster RF_AccessStripeMap_t *stripeMap; /* pointer to the actual map. Also used for making lists */
232 1.1 oster RF_AccessStripeMapHeader_t *next;
233 1.1 oster };
234 1.1 oster
235 1.1 oster /*****************************************************************************************
236 1.1 oster *
237 1.1 oster * various routines mapping addresses in the RAID address space. These work across
238 1.1 oster * all layouts. DON'T PUT ANY LAYOUT-SPECIFIC CODE HERE.
239 1.1 oster *
240 1.1 oster ****************************************************************************************/
241 1.1 oster
242 1.1 oster /* return the identifier of the stripe containing the given address */
243 1.1 oster #define rf_RaidAddressToStripeID(_layoutPtr_, _addr_) \
244 1.1 oster ( ((_addr_) / (_layoutPtr_)->sectorsPerStripeUnit) / (_layoutPtr_)->numDataCol )
245 1.1 oster
246 1.1 oster /* return the raid address of the start of the indicates stripe ID */
247 1.1 oster #define rf_StripeIDToRaidAddress(_layoutPtr_, _sid_) \
248 1.1 oster ( ((_sid_) * (_layoutPtr_)->sectorsPerStripeUnit) * (_layoutPtr_)->numDataCol )
249 1.1 oster
250 1.1 oster /* return the identifier of the stripe containing the given stripe unit id */
251 1.1 oster #define rf_StripeUnitIDToStripeID(_layoutPtr_, _addr_) \
252 1.1 oster ( (_addr_) / (_layoutPtr_)->numDataCol )
253 1.1 oster
254 1.1 oster /* return the identifier of the stripe unit containing the given address */
255 1.1 oster #define rf_RaidAddressToStripeUnitID(_layoutPtr_, _addr_) \
256 1.1 oster ( ((_addr_) / (_layoutPtr_)->sectorsPerStripeUnit) )
257 1.1 oster
258 1.1 oster /* return the RAID address of next stripe boundary beyond the given address */
259 1.1 oster #define rf_RaidAddressOfNextStripeBoundary(_layoutPtr_, _addr_) \
260 1.1 oster ( (((_addr_)/(_layoutPtr_)->dataSectorsPerStripe)+1) * (_layoutPtr_)->dataSectorsPerStripe )
261 1.1 oster
262 1.1 oster /* return the RAID address of the start of the stripe containing the given address */
263 1.1 oster #define rf_RaidAddressOfPrevStripeBoundary(_layoutPtr_, _addr_) \
264 1.1 oster ( (((_addr_)/(_layoutPtr_)->dataSectorsPerStripe)+0) * (_layoutPtr_)->dataSectorsPerStripe )
265 1.1 oster
266 1.1 oster /* return the RAID address of next stripe unit boundary beyond the given address */
267 1.1 oster #define rf_RaidAddressOfNextStripeUnitBoundary(_layoutPtr_, _addr_) \
268 1.1 oster ( (((_addr_)/(_layoutPtr_)->sectorsPerStripeUnit)+1L)*(_layoutPtr_)->sectorsPerStripeUnit )
269 1.1 oster
270 1.1 oster /* return the RAID address of the start of the stripe unit containing RAID address _addr_ */
271 1.1 oster #define rf_RaidAddressOfPrevStripeUnitBoundary(_layoutPtr_, _addr_) \
272 1.1 oster ( (((_addr_)/(_layoutPtr_)->sectorsPerStripeUnit)+0)*(_layoutPtr_)->sectorsPerStripeUnit )
273 1.1 oster
274 1.1 oster /* returns the offset into the stripe. used by RaidAddressStripeAligned */
275 1.1 oster #define rf_RaidAddressStripeOffset(_layoutPtr_, _addr_) \
276 1.1 oster ( (_addr_) % ((_layoutPtr_)->dataSectorsPerStripe) )
277 1.1 oster
278 1.1 oster /* returns the offset into the stripe unit. */
279 1.1 oster #define rf_StripeUnitOffset(_layoutPtr_, _addr_) \
280 1.1 oster ( (_addr_) % ((_layoutPtr_)->sectorsPerStripeUnit) )
281 1.1 oster
282 1.1 oster /* returns nonzero if the given RAID address is stripe-aligned */
283 1.1 oster #define rf_RaidAddressStripeAligned( __layoutPtr__, __addr__ ) \
284 1.1 oster ( rf_RaidAddressStripeOffset(__layoutPtr__, __addr__) == 0 )
285 1.1 oster
286 1.1 oster /* returns nonzero if the given address is stripe-unit aligned */
287 1.1 oster #define rf_StripeUnitAligned( __layoutPtr__, __addr__ ) \
288 1.1 oster ( rf_StripeUnitOffset(__layoutPtr__, __addr__) == 0 )
289 1.1 oster
290 1.1 oster /* convert an address expressed in RAID blocks to/from an addr expressed in bytes */
291 1.1 oster #define rf_RaidAddressToByte(_raidPtr_, _addr_) \
292 1.1 oster ( (_addr_) << ( (_raidPtr_)->logBytesPerSector ) )
293 1.1 oster
294 1.1 oster #define rf_ByteToRaidAddress(_raidPtr_, _addr_) \
295 1.1 oster ( (_addr_) >> ( (_raidPtr_)->logBytesPerSector ) )
296 1.1 oster
297 1.1 oster /* convert a raid address to/from a parity stripe ID. Conversion to raid address is easy,
298 1.1 oster * since we're asking for the address of the first sector in the parity stripe. Conversion to a
299 1.1 oster * parity stripe ID is more complex, since stripes are not contiguously allocated in
300 1.1 oster * parity stripes.
301 1.1 oster */
302 1.1 oster #define rf_RaidAddressToParityStripeID(_layoutPtr_, _addr_, _ru_num_) \
303 1.1 oster rf_MapStripeIDToParityStripeID( (_layoutPtr_), rf_RaidAddressToStripeID( (_layoutPtr_), (_addr_) ), (_ru_num_) )
304 1.1 oster
305 1.1 oster #define rf_ParityStripeIDToRaidAddress(_layoutPtr_, _psid_) \
306 1.1 oster ( (_psid_) * (_layoutPtr_)->SUsPerPU * (_layoutPtr_)->numDataCol * (_layoutPtr_)->sectorsPerStripeUnit )
307 1.1 oster
308 1.1 oster RF_LayoutSW_t *rf_GetLayout(RF_ParityConfig_t parityConfig);
309 1.1 oster int rf_ConfigureLayout(RF_ShutdownList_t **listp, RF_Raid_t *raidPtr,
310 1.1 oster RF_Config_t *cfgPtr);
311 1.1 oster RF_StripeNum_t rf_MapStripeIDToParityStripeID(RF_RaidLayout_t *layoutPtr,
312 1.1 oster RF_StripeNum_t stripeID, RF_ReconUnitNum_t *which_ru);
313 1.1 oster
314 1.1 oster #endif /* !_RF__RF_LAYOUT_H_ */
315