1 /* $OpenBSD: rf_dagffrd.c,v 1.4 2002/12/16 07:01:03 tdeval Exp $ */
2 /* $NetBSD: rf_dagffrd.c,v 1.4 2000/01/07 03:40:58 oster Exp $ */
3
4 /*
5 * Copyright (c) 1995 Carnegie-Mellon University.
6 * All rights reserved.
7 *
8 * Author: Mark Holland, Daniel Stodolsky, William V. Courtright II
9 *
10 * Permission to use, copy, modify and distribute this software and
11 * its documentation is hereby granted, provided that both the copyright
12 * notice and this permission notice appear in all copies of the
13 * software, derivative works or modified versions, and any portions
14 * thereof, and that both notices appear in supporting documentation.
15 *
16 * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
17 * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
18 * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
19 *
20 * Carnegie Mellon requests users of this software to return to
21 *
22 * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU
23 * School of Computer Science
24 * Carnegie Mellon University
25 * Pittsburgh PA 15213-3890
26 *
27 * any improvements or extensions that they make and grant Carnegie the
28 * rights to redistribute these changes.
29 */
30
31 /*
32 * rf_dagffrd.c
33 *
34 * Code for creating fault-free read DAGs.
35 *
36 */
37
38 #include "rf_types.h"
39 #include "rf_raid.h"
40 #include "rf_dag.h"
41 #include "rf_dagutils.h"
42 #include "rf_dagfuncs.h"
43 #include "rf_debugMem.h"
44 #include "rf_memchunk.h"
45 #include "rf_general.h"
46 #include "rf_dagffrd.h"
47
48 void rf_CreateMirrorReadDAG( RF_Raid_t *, RF_AccessStripeMap_t *,
49 RF_DagHeader_t *, void *, RF_RaidAccessFlags_t, RF_AllocListElem_t *,
50 int (*) (RF_DagNode_t *));
51
52 /*****************************************************************************
53 *
54 * General comments on DAG creation:
55 *
56 * All DAGs in this file use roll-away error recovery. Each DAG has a single
57 * commit node, usually called "Cmt." If an error occurs before the Cmt node
58 * is reached, the execution engine will halt forward execution and work
59 * backward through the graph, executing the undo functions. Assuming that
60 * each node in the graph prior to the Cmt node are undoable and atomic - or -
61 * does not make changes to permanent state, the graph will fail atomically.
62 * If an error occurs after the Cmt node executes, the engine will roll-forward
63 * through the graph, blindly executing nodes until it reaches the end.
64 * If a graph reaches the end, it is assumed to have completed successfully.
65 *
66 * A graph has only 1 Cmt node.
67 *
68 *****************************************************************************/
69
70
71 /*****************************************************************************
72 *
73 * The following wrappers map the standard DAG creation interface to the
74 * DAG creation routines. Additionally, these wrappers enable experimentation
75 * with new DAG structures by providing an extra level of indirection, allowing
76 * the DAG creation routines to be replaced at this single point.
77 *
78 *****************************************************************************/
79
80 void
81 rf_CreateFaultFreeReadDAG(
82 RF_Raid_t *raidPtr,
83 RF_AccessStripeMap_t *asmap,
84 RF_DagHeader_t *dag_h,
85 void *bp,
86 RF_RaidAccessFlags_t flags,
87 RF_AllocListElem_t *allocList
88 )
89 {
90 rf_CreateNonredundantDAG(raidPtr, asmap, dag_h, bp, flags, allocList,
91 RF_IO_TYPE_READ);
92 }
93
94
95 /*****************************************************************************
96 *
97 * DAG creation code begins here.
98 *
99 *****************************************************************************/
100
101 /*****************************************************************************
102 *
103 * Creates a DAG to perform a nonredundant read or write of data within one
104 * stripe.
105 * For reads, this DAG is as follows:
106 *
107 * /---- read ----\
108 * Header -- Block ---- read ---- Commit -- Terminate
109 * \---- read ----/
110 *
111 * For writes, this DAG is as follows:
112 *
113 * /---- write ----\
114 * Header -- Commit ---- write ---- Block -- Terminate
115 * \---- write ----/
116 *
117 * There is one disk node per stripe unit accessed, and all disk nodes are in
118 * parallel.
119 *
120 * Tricky point here: The first disk node (read or write) is created
121 * normally. Subsequent disk nodes are created by copying the first one,
122 * and modifying a few params. The "succedents" and "antecedents" fields are
123 * _not_ re-created in each node, but rather left pointing to the same array
124 * that was malloc'd when the first node was created. Thus, it's essential
125 * that when this DAG is freed, the succedents and antecedents fields be freed
126 * in ONLY ONE of the read nodes. This does not apply to the "params" field
127 * because it is recreated for each READ node.
128 *
129 * Note that normal-priority accesses do not need to be tagged with their
130 * parity stripe ID, because they will never be promoted. Hence, I've
131 * commented-out the code to do this, and marked it with UNNEEDED.
132 *
133 *****************************************************************************/
134
135 void
136 rf_CreateNonredundantDAG(
137 RF_Raid_t *raidPtr,
138 RF_AccessStripeMap_t *asmap,
139 RF_DagHeader_t *dag_h,
140 void *bp,
141 RF_RaidAccessFlags_t flags,
142 RF_AllocListElem_t *allocList,
143 RF_IoType_t type
144 )
145 {
146 RF_DagNode_t *nodes, *diskNodes, *blockNode, *commitNode, *termNode;
147 RF_PhysDiskAddr_t *pda = asmap->physInfo;
148 int (*doFunc) (RF_DagNode_t *), (*undoFunc) (RF_DagNode_t *);
149 int i, n, totalNumNodes;
150 char *name;
151
152 n = asmap->numStripeUnitsAccessed;
153 dag_h->creator = "NonredundantDAG";
154
155 RF_ASSERT(RF_IO_IS_R_OR_W(type));
156 switch (type) {
157 case RF_IO_TYPE_READ:
158 doFunc = rf_DiskReadFunc;
159 undoFunc = rf_DiskReadUndoFunc;
160 name = "R ";
161 if (rf_dagDebug)
162 printf("[Creating non-redundant read DAG]\n");
163 break;
164 case RF_IO_TYPE_WRITE:
165 doFunc = rf_DiskWriteFunc;
166 undoFunc = rf_DiskWriteUndoFunc;
167 name = "W ";
168 if (rf_dagDebug)
169 printf("[Creating non-redundant write DAG]\n");
170 break;
171 default:
172 RF_PANIC();
173 }
174
175 /*
176 * For reads, the dag can not commit until the block node is reached.
177 * For writes, the dag commits immediately.
178 */
179 dag_h->numCommitNodes = 1;
180 dag_h->numCommits = 0;
181 dag_h->numSuccedents = 1;
182
183 /*
184 * Node count:
185 * 1 block node
186 * n data reads (or writes)
187 * 1 commit node
188 * 1 terminator node
189 */
190 RF_ASSERT(n > 0);
191 totalNumNodes = n + 3;
192 RF_CallocAndAdd(nodes, totalNumNodes, sizeof(RF_DagNode_t),
193 (RF_DagNode_t *), allocList);
194 i = 0;
195 diskNodes = &nodes[i];
196 i += n;
197 blockNode = &nodes[i];
198 i += 1;
199 commitNode = &nodes[i];
200 i += 1;
201 termNode = &nodes[i];
202 i += 1;
203 RF_ASSERT(i == totalNumNodes);
204
205 /* Initialize nodes. */
206 switch (type) {
207 case RF_IO_TYPE_READ:
208 rf_InitNode(blockNode, rf_wait, RF_FALSE, rf_NullNodeFunc,
209 rf_NullNodeUndoFunc, NULL, n, 0, 0, 0, dag_h, "Nil",
210 allocList);
211 rf_InitNode(commitNode, rf_wait, RF_TRUE, rf_NullNodeFunc,
212 rf_NullNodeUndoFunc, NULL, 1, n, 0, 0, dag_h, "Cmt",
213 allocList);
214 rf_InitNode(termNode, rf_wait, RF_FALSE, rf_TerminateFunc,
215 rf_TerminateUndoFunc, NULL, 0, 1, 0, 0, dag_h, "Trm",
216 allocList);
217 break;
218 case RF_IO_TYPE_WRITE:
219 rf_InitNode(blockNode, rf_wait, RF_FALSE, rf_NullNodeFunc,
220 rf_NullNodeUndoFunc, NULL, 1, 0, 0, 0, dag_h, "Nil",
221 allocList);
222 rf_InitNode(commitNode, rf_wait, RF_TRUE, rf_NullNodeFunc,
223 rf_NullNodeUndoFunc, NULL, n, 1, 0, 0, dag_h, "Cmt",
224 allocList);
225 rf_InitNode(termNode, rf_wait, RF_FALSE, rf_TerminateFunc,
226 rf_TerminateUndoFunc, NULL, 0, n, 0, 0, dag_h, "Trm",
227 allocList);
228 break;
229 default:
230 RF_PANIC();
231 }
232
233 for (i = 0; i < n; i++) {
234 RF_ASSERT(pda != NULL);
235 rf_InitNode(&diskNodes[i], rf_wait, RF_FALSE, doFunc, undoFunc,
236 rf_GenericWakeupFunc, 1, 1, 4, 0, dag_h, name, allocList);
237 diskNodes[i].params[0].p = pda;
238 diskNodes[i].params[1].p = pda->bufPtr;
239 /* Parity stripe id is not necessary. */
240 diskNodes[i].params[2].v = 0;
241 diskNodes[i].params[3].v =
242 RF_CREATE_PARAM3(RF_IO_NORMAL_PRIORITY, 0, 0, 0);
243 pda = pda->next;
244 }
245
246 /*
247 * Connect nodes.
248 */
249
250 /* Connect hdr to block node. */
251 RF_ASSERT(blockNode->numAntecedents == 0);
252 dag_h->succedents[0] = blockNode;
253
254 if (type == RF_IO_TYPE_READ) {
255 /* Connecting a nonredundant read DAG. */
256 RF_ASSERT(blockNode->numSuccedents == n);
257 RF_ASSERT(commitNode->numAntecedents == n);
258 for (i = 0; i < n; i++) {
259 /* Connect block node to each read node. */
260 RF_ASSERT(diskNodes[i].numAntecedents == 1);
261 blockNode->succedents[i] = &diskNodes[i];
262 diskNodes[i].antecedents[0] = blockNode;
263 diskNodes[i].antType[0] = rf_control;
264
265 /* Connect each read node to the commit node. */
266 RF_ASSERT(diskNodes[i].numSuccedents == 1);
267 diskNodes[i].succedents[0] = commitNode;
268 commitNode->antecedents[i] = &diskNodes[i];
269 commitNode->antType[i] = rf_control;
270 }
271 /* Connect the commit node to the term node. */
272 RF_ASSERT(commitNode->numSuccedents == 1);
273 RF_ASSERT(termNode->numAntecedents == 1);
274 RF_ASSERT(termNode->numSuccedents == 0);
275 commitNode->succedents[0] = termNode;
276 termNode->antecedents[0] = commitNode;
277 termNode->antType[0] = rf_control;
278 } else {
279 /* Connecting a nonredundant write DAG. */
280 /* Connect the block node to the commit node. */
281 RF_ASSERT(blockNode->numSuccedents == 1);
282 RF_ASSERT(commitNode->numAntecedents == 1);
283 blockNode->succedents[0] = commitNode;
284 commitNode->antecedents[0] = blockNode;
285 commitNode->antType[0] = rf_control;
286
287 RF_ASSERT(commitNode->numSuccedents == n);
288 RF_ASSERT(termNode->numAntecedents == n);
289 RF_ASSERT(termNode->numSuccedents == 0);
290 for (i = 0; i < n; i++) {
291 /* Connect the commit node to each write node. */
292 RF_ASSERT(diskNodes[i].numAntecedents == 1);
293 commitNode->succedents[i] = &diskNodes[i];
294 diskNodes[i].antecedents[0] = commitNode;
295 diskNodes[i].antType[0] = rf_control;
296
297 /* Connect each write node to the term node. */
298 RF_ASSERT(diskNodes[i].numSuccedents == 1);
299 diskNodes[i].succedents[0] = termNode;
300 termNode->antecedents[i] = &diskNodes[i];
301 termNode->antType[i] = rf_control;
302 }
303 }
304 }
305 /*****************************************************************************
306 * Create a fault-free read DAG for RAID level 1.
307 *
308 * Hdr -> Nil -> Rmir -> Cmt -> Trm
309 *
310 * The "Rmir" node schedules a read from the disk in the mirror pair with the
311 * shortest disk queue. The proper queue is selected at Rmir execution. This
312 * deferred mapping is unlike other archs in RAIDframe which generally fix
313 * mapping at DAG creation time.
314 *
315 * Parameters: raidPtr - description of the physical array
316 * asmap - logical & physical addresses for this access
317 * bp - buffer ptr (for holding read data)
318 * flags - general flags (e.g. disk locking)
319 * allocList - list of memory allocated in DAG creation
320 *****************************************************************************/
321
322 void
323 rf_CreateMirrorReadDAG(
324 RF_Raid_t *raidPtr,
325 RF_AccessStripeMap_t *asmap,
326 RF_DagHeader_t *dag_h,
327 void *bp,
328 RF_RaidAccessFlags_t flags,
329 RF_AllocListElem_t *allocList,
330 int (*readfunc) (RF_DagNode_t *)
331 )
332 {
333 RF_DagNode_t *readNodes, *nodes, *blockNode, *commitNode, *termNode;
334 RF_PhysDiskAddr_t *data_pda = asmap->physInfo;
335 RF_PhysDiskAddr_t *parity_pda = asmap->parityInfo;
336 int i, n, totalNumNodes;
337
338 n = asmap->numStripeUnitsAccessed;
339 dag_h->creator = "RaidOneReadDAG";
340 if (rf_dagDebug) {
341 printf("[Creating RAID level 1 read DAG]\n");
342 }
343 /*
344 * This dag can not commit until the commit node is reached.
345 * Errors prior to the commit point imply the dag has failed.
346 */
347 dag_h->numCommitNodes = 1;
348 dag_h->numCommits = 0;
349 dag_h->numSuccedents = 1;
350
351 /*
352 * Node count:
353 * n data reads
354 * 1 block node
355 * 1 commit node
356 * 1 terminator node
357 */
358 RF_ASSERT(n > 0);
359 totalNumNodes = n + 3;
360 RF_CallocAndAdd(nodes, totalNumNodes, sizeof(RF_DagNode_t),
361 (RF_DagNode_t *), allocList);
362 i = 0;
363 readNodes = &nodes[i];
364 i += n;
365 blockNode = &nodes[i];
366 i += 1;
367 commitNode = &nodes[i];
368 i += 1;
369 termNode = &nodes[i];
370 i += 1;
371 RF_ASSERT(i == totalNumNodes);
372
373 /* Initialize nodes. */
374 rf_InitNode(blockNode, rf_wait, RF_FALSE, rf_NullNodeFunc,
375 rf_NullNodeUndoFunc, NULL, n, 0, 0, 0, dag_h, "Nil", allocList);
376 rf_InitNode(commitNode, rf_wait, RF_TRUE, rf_NullNodeFunc,
377 rf_NullNodeUndoFunc, NULL, 1, n, 0, 0, dag_h, "Cmt", allocList);
378 rf_InitNode(termNode, rf_wait, RF_FALSE, rf_TerminateFunc,
379 rf_TerminateUndoFunc, NULL, 0, 1, 0, 0, dag_h, "Trm", allocList);
380
381 for (i = 0; i < n; i++) {
382 RF_ASSERT(data_pda != NULL);
383 RF_ASSERT(parity_pda != NULL);
384 rf_InitNode(&readNodes[i], rf_wait, RF_FALSE, readfunc,
385 rf_DiskReadMirrorUndoFunc, rf_GenericWakeupFunc, 1, 1, 5,
386 0, dag_h, "Rmir", allocList);
387 readNodes[i].params[0].p = data_pda;
388 readNodes[i].params[1].p = data_pda->bufPtr;
389 /* Parity stripe id is not necessary. */
390 readNodes[i].params[2].p = 0;
391 readNodes[i].params[3].v =
392 RF_CREATE_PARAM3(RF_IO_NORMAL_PRIORITY, 0, 0, 0);
393 readNodes[i].params[4].p = parity_pda;
394 data_pda = data_pda->next;
395 parity_pda = parity_pda->next;
396 }
397
398 /*
399 * Connect nodes.
400 */
401
402 /* Connect hdr to block node. */
403 RF_ASSERT(blockNode->numAntecedents == 0);
404 dag_h->succedents[0] = blockNode;
405
406 /* Connect block node to read nodes. */
407 RF_ASSERT(blockNode->numSuccedents == n);
408 for (i = 0; i < n; i++) {
409 RF_ASSERT(readNodes[i].numAntecedents == 1);
410 blockNode->succedents[i] = &readNodes[i];
411 readNodes[i].antecedents[0] = blockNode;
412 readNodes[i].antType[0] = rf_control;
413 }
414
415 /* Connect read nodes to commit node. */
416 RF_ASSERT(commitNode->numAntecedents == n);
417 for (i = 0; i < n; i++) {
418 RF_ASSERT(readNodes[i].numSuccedents == 1);
419 readNodes[i].succedents[0] = commitNode;
420 commitNode->antecedents[i] = &readNodes[i];
421 commitNode->antType[i] = rf_control;
422 }
423
424 /* Connect commit node to term node. */
425 RF_ASSERT(commitNode->numSuccedents == 1);
426 RF_ASSERT(termNode->numAntecedents == 1);
427 RF_ASSERT(termNode->numSuccedents == 0);
428 commitNode->succedents[0] = termNode;
429 termNode->antecedents[0] = commitNode;
430 termNode->antType[0] = rf_control;
431 }
432
433 void
434 rf_CreateMirrorIdleReadDAG(
435 RF_Raid_t *raidPtr,
436 RF_AccessStripeMap_t *asmap,
437 RF_DagHeader_t *dag_h,
438 void *bp,
439 RF_RaidAccessFlags_t flags,
440 RF_AllocListElem_t *allocList
441 )
442 {
443 rf_CreateMirrorReadDAG(raidPtr, asmap, dag_h, bp, flags, allocList,
444 rf_DiskReadMirrorIdleFunc);
445 }
446
447 void
448 rf_CreateMirrorPartitionReadDAG(
449 RF_Raid_t *raidPtr,
450 RF_AccessStripeMap_t *asmap,
451 RF_DagHeader_t *dag_h,
452 void *bp,
453 RF_RaidAccessFlags_t flags,
454 RF_AllocListElem_t *allocList
455 )
456 {
457 rf_CreateMirrorReadDAG(raidPtr, asmap, dag_h, bp, flags, allocList,
458 rf_DiskReadMirrorPartitionFunc);
459 }