1 /* $NetBSD: rf_copyback.c,v 1.50 2014/06/14 07:39:00 hannken 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 *
31 * copyback.c -- code to copy reconstructed data back from spare space to
32 * the replaced disk.
33 *
34 * the code operates using callbacks on the I/Os to continue with the
35 * next unit to be copied back. We do this because a simple loop
36 * containing blocking I/Os will not work in the simulator.
37 *
38 ****************************************************************************/
39
40 #include <sys/cdefs.h>
41 __KERNEL_RCSID(0, "$NetBSD: rf_copyback.c,v 1.50 2014/06/14 07:39:00 hannken Exp $");
42
43 #include <dev/raidframe/raidframevar.h>
44
45 #include <sys/time.h>
46 #include <sys/buf.h>
47 #include "rf_raid.h"
48 #include "rf_mcpair.h"
49 #include "rf_acctrace.h"
50 #include "rf_etimer.h"
51 #include "rf_general.h"
52 #include "rf_utils.h"
53 #include "rf_copyback.h"
54 #include "rf_decluster.h"
55 #include "rf_driver.h"
56 #include "rf_shutdown.h"
57 #include "rf_kintf.h"
58
59 #define RF_COPYBACK_DATA 0
60 #define RF_COPYBACK_PARITY 1
61
62 int rf_copyback_in_progress;
63
64 static int rf_CopybackReadDoneProc(RF_CopybackDesc_t * desc, int status);
65 static int rf_CopybackWriteDoneProc(RF_CopybackDesc_t * desc, int status);
66 static void rf_CopybackOne(RF_CopybackDesc_t * desc, int typ,
67 RF_RaidAddr_t addr, RF_RowCol_t testCol,
68 RF_SectorNum_t testOffs);
69 static void rf_CopybackComplete(RF_CopybackDesc_t * desc, int status);
70
71 int
rf_ConfigureCopyback(RF_ShutdownList_t ** listp)72 rf_ConfigureCopyback(RF_ShutdownList_t **listp)
73 {
74 rf_copyback_in_progress = 0;
75 return (0);
76 }
77
78 #include <sys/param.h>
79 #include <sys/systm.h>
80 #include <sys/proc.h>
81 #include <sys/ioctl.h>
82 #include <sys/fcntl.h>
83 #include <sys/vnode.h>
84 #include <sys/namei.h> /* for pathbuf */
85
86 #include <miscfs/specfs/specdev.h> /* for v_rdev */
87
88 /* do a complete copyback */
89 void
rf_CopybackReconstructedData(RF_Raid_t * raidPtr)90 rf_CopybackReconstructedData(RF_Raid_t *raidPtr)
91 {
92 RF_ComponentLabel_t *c_label;
93 int found, retcode;
94 RF_CopybackDesc_t *desc;
95 RF_RowCol_t fcol;
96 RF_RaidDisk_t *badDisk;
97 char *databuf;
98
99 struct pathbuf *dev_pb;
100 struct vnode *vp;
101
102 int ac;
103
104 fcol = 0;
105 found = 0;
106 for (fcol = 0; fcol < raidPtr->numCol; fcol++) {
107 if (raidPtr->Disks[fcol].status == rf_ds_dist_spared
108 || raidPtr->Disks[fcol].status == rf_ds_spared) {
109 found = 1;
110 break;
111 }
112 }
113
114 if (!found) {
115 printf("raid%d: no disks need copyback\n", raidPtr->raidid);
116 return;
117 }
118
119 badDisk = &raidPtr->Disks[fcol];
120
121 /* This device may have been opened successfully the first time. Close
122 * it before trying to open it again.. */
123
124 if (raidPtr->raid_cinfo[fcol].ci_vp != NULL) {
125 printf("Closed the open device: %s\n",
126 raidPtr->Disks[fcol].devname);
127 vp = raidPtr->raid_cinfo[fcol].ci_vp;
128 ac = raidPtr->Disks[fcol].auto_configured;
129 rf_close_component(raidPtr, vp, ac);
130 raidPtr->raid_cinfo[fcol].ci_vp = NULL;
131
132 }
133 /* note that this disk was *not* auto_configured (any longer) */
134 raidPtr->Disks[fcol].auto_configured = 0;
135
136 printf("About to (re-)open the device: %s\n",
137 raidPtr->Disks[fcol].devname);
138
139 dev_pb = pathbuf_create(raidPtr->Disks[fcol].devname);
140 if (dev_pb == NULL) {
141 /* shouldn't happen unless maybe the system is OOMing */
142 printf("raid%d: copyback: pathbuf_create on device: %s failed: %d!\n",
143 raidPtr->raidid, raidPtr->Disks[fcol].devname,
144 ENOMEM);
145 return;
146 }
147 retcode = dk_lookup(dev_pb, curlwp, &vp);
148 pathbuf_destroy(dev_pb);
149
150 if (retcode) {
151 printf("raid%d: copyback: dk_lookup on device: %s failed: %d!\n",
152 raidPtr->raidid, raidPtr->Disks[fcol].devname,
153 retcode);
154
155 /* XXX the component isn't responding properly... must be
156 * still dead :-( */
157 return;
158
159 } else {
160
161 /* Ok, so we can at least do a lookup... How about actually
162 * getting a vp for it? */
163
164 retcode = rf_getdisksize(vp, &raidPtr->Disks[fcol]);
165 if (retcode) {
166 return;
167 }
168
169 raidPtr->raid_cinfo[fcol].ci_vp = vp;
170 raidPtr->raid_cinfo[fcol].ci_dev = vp->v_rdev;
171
172 raidPtr->Disks[fcol].dev = vp->v_rdev; /* XXX or the above? */
173
174 /* we allow the user to specify that only a fraction of the
175 * disks should be used this is just for debug: it speeds up
176 * the parity scan */
177 raidPtr->Disks[fcol].numBlocks =
178 raidPtr->Disks[fcol].numBlocks *
179 rf_sizePercentage / 100;
180 }
181
182 if (retcode) {
183 printf("raid%d: copyback: target disk failed TUR\n",
184 raidPtr->raidid);
185 return;
186 }
187 /* get a buffer to hold one SU */
188 RF_Malloc(databuf, rf_RaidAddressToByte(raidPtr, raidPtr->Layout.sectorsPerStripeUnit), (char *));
189
190 /* create a descriptor */
191 RF_Malloc(desc, sizeof(*desc), (RF_CopybackDesc_t *));
192 desc->raidPtr = raidPtr;
193 desc->status = 0;
194 desc->fcol = fcol;
195 desc->spCol = badDisk->spareCol;
196 desc->stripeAddr = 0;
197 desc->sectPerSU = raidPtr->Layout.sectorsPerStripeUnit;
198 desc->sectPerStripe = raidPtr->Layout.sectorsPerStripeUnit * raidPtr->Layout.numDataCol;
199 desc->databuf = databuf;
200 desc->mcpair = rf_AllocMCPair();
201
202 /* quiesce the array, since we don't want to code support for user
203 * accs here */
204 rf_SuspendNewRequestsAndWait(raidPtr);
205
206 /* adjust state of the array and of the disks */
207 rf_lock_mutex2(raidPtr->mutex);
208 raidPtr->Disks[desc->fcol].status = rf_ds_optimal;
209 raidPtr->status = rf_rs_optimal;
210 rf_copyback_in_progress = 1; /* debug only */
211 rf_unlock_mutex2(raidPtr->mutex);
212
213 RF_GETTIME(desc->starttime);
214 rf_ContinueCopyback(desc);
215
216 /* Data has been restored. Fix up the component label. */
217 /* Don't actually need the read here.. */
218
219 c_label = raidget_component_label(raidPtr, fcol);
220 raid_init_component_label(raidPtr, c_label);
221
222 c_label->row = 0;
223 c_label->column = fcol;
224 rf_component_label_set_partitionsize(c_label,
225 raidPtr->Disks[fcol].partitionSize);
226
227 raidflush_component_label(raidPtr, fcol);
228
229 /* XXXjld why is this here? */
230 rf_update_component_labels(raidPtr, RF_NORMAL_COMPONENT_UPDATE);
231 }
232
233
234 /*
235 * invoked via callback after a copyback I/O has completed to
236 * continue on with the next one
237 */
238 void
rf_ContinueCopyback(RF_CopybackDesc_t * desc)239 rf_ContinueCopyback(RF_CopybackDesc_t *desc)
240 {
241 RF_SectorNum_t testOffs, stripeAddr;
242 RF_Raid_t *raidPtr = desc->raidPtr;
243 RF_RaidAddr_t addr;
244 RF_RowCol_t testCol;
245 #if RF_DEBUG_RECON
246 int old_pctg, new_pctg;
247 struct timeval t, diff;
248 #endif
249 int done;
250
251 #if RF_DEBUG_RECON
252 old_pctg = (-1);
253 #endif
254 while (1) {
255 stripeAddr = desc->stripeAddr;
256 desc->raidPtr->copyback_stripes_done = stripeAddr
257 / desc->sectPerStripe;
258 #if RF_DEBUG_RECON
259 if (rf_prReconSched) {
260 old_pctg = 100 * desc->stripeAddr / raidPtr->totalSectors;
261 }
262 #endif
263 desc->stripeAddr += desc->sectPerStripe;
264 #if RF_DEBUG_RECON
265 if (rf_prReconSched) {
266 new_pctg = 100 * desc->stripeAddr / raidPtr->totalSectors;
267 if (new_pctg != old_pctg) {
268 RF_GETTIME(t);
269 RF_TIMEVAL_DIFF(&desc->starttime, &t, &diff);
270 printf("%d %d.%06d\n", new_pctg, (int) diff.tv_sec, (int) diff.tv_usec);
271 }
272 }
273 #endif
274 if (stripeAddr >= raidPtr->totalSectors) {
275 rf_CopybackComplete(desc, 0);
276 return;
277 }
278 /* walk through the current stripe, su-by-su */
279 for (done = 0, addr = stripeAddr; addr < stripeAddr + desc->sectPerStripe; addr += desc->sectPerSU) {
280
281 /* map the SU, disallowing remap to spare space */
282 (raidPtr->Layout.map->MapSector) (raidPtr, addr, &testCol, &testOffs, RF_DONT_REMAP);
283
284 if (testCol == desc->fcol) {
285 rf_CopybackOne(desc, RF_COPYBACK_DATA, addr, testCol, testOffs);
286 done = 1;
287 break;
288 }
289 }
290
291 if (!done) {
292 /* we didn't find the failed disk in the data part.
293 * check parity. */
294
295 /* map the parity for this stripe, disallowing remap
296 * to spare space */
297 (raidPtr->Layout.map->MapParity) (raidPtr, stripeAddr, &testCol, &testOffs, RF_DONT_REMAP);
298
299 if (testCol == desc->fcol) {
300 rf_CopybackOne(desc, RF_COPYBACK_PARITY, stripeAddr, testCol, testOffs);
301 }
302 }
303 /* check to see if the last read/write pair failed */
304 if (desc->status) {
305 rf_CopybackComplete(desc, 1);
306 return;
307 }
308 /* we didn't find any units to copy back in this stripe.
309 * Continue with the next one */
310 }
311 }
312
313
314 /* copyback one unit */
315 static void
rf_CopybackOne(RF_CopybackDesc_t * desc,int typ,RF_RaidAddr_t addr,RF_RowCol_t testCol,RF_SectorNum_t testOffs)316 rf_CopybackOne(RF_CopybackDesc_t *desc, int typ, RF_RaidAddr_t addr,
317 RF_RowCol_t testCol, RF_SectorNum_t testOffs)
318 {
319 RF_SectorCount_t sectPerSU = desc->sectPerSU;
320 RF_Raid_t *raidPtr = desc->raidPtr;
321 RF_RowCol_t spCol = desc->spCol;
322 RF_SectorNum_t spOffs;
323
324 /* find the spare spare location for this SU */
325 if (raidPtr->Layout.map->flags & RF_DISTRIBUTE_SPARE) {
326 if (typ == RF_COPYBACK_DATA)
327 raidPtr->Layout.map->MapSector(raidPtr, addr, &spCol, &spOffs, RF_REMAP);
328 else
329 raidPtr->Layout.map->MapParity(raidPtr, addr, &spCol, &spOffs, RF_REMAP);
330 } else {
331 spOffs = testOffs;
332 }
333
334 /* create reqs to read the old location & write the new */
335 desc->readreq = rf_CreateDiskQueueData(RF_IO_TYPE_READ, spOffs,
336 sectPerSU, desc->databuf, 0L, 0,
337 (int (*) (void *, int)) rf_CopybackReadDoneProc, desc,
338 NULL, (void *) raidPtr, RF_DISKQUEUE_DATA_FLAGS_NONE, NULL,
339 PR_WAITOK);
340 desc->writereq = rf_CreateDiskQueueData(RF_IO_TYPE_WRITE, testOffs,
341 sectPerSU, desc->databuf, 0L, 0,
342 (int (*) (void *, int)) rf_CopybackWriteDoneProc, desc,
343 NULL, (void *) raidPtr, RF_DISKQUEUE_DATA_FLAGS_NONE, NULL,
344 PR_WAITOK);
345 desc->fcol = testCol;
346
347 /* enqueue the read. the write will go out as part of the callback on
348 * the read. at user-level & in the kernel, wait for the read-write
349 * pair to complete. in the simulator, just return, since everything
350 * will happen as callbacks */
351
352 RF_LOCK_MCPAIR(desc->mcpair);
353 desc->mcpair->flag = 0;
354 RF_UNLOCK_MCPAIR(desc->mcpair);
355
356 rf_DiskIOEnqueue(&raidPtr->Queues[spCol], desc->readreq, RF_IO_NORMAL_PRIORITY);
357
358 RF_LOCK_MCPAIR(desc->mcpair);
359 while (!desc->mcpair->flag) {
360 RF_WAIT_MCPAIR(desc->mcpair);
361 }
362 RF_UNLOCK_MCPAIR(desc->mcpair);
363 rf_FreeDiskQueueData(desc->readreq);
364 rf_FreeDiskQueueData(desc->writereq);
365
366 }
367
368
369 /* called at interrupt context when the read has completed. just send out the write */
370 static int
rf_CopybackReadDoneProc(RF_CopybackDesc_t * desc,int status)371 rf_CopybackReadDoneProc(RF_CopybackDesc_t *desc, int status)
372 {
373 if (status) { /* invoke the callback with bad status */
374 printf("raid%d: copyback read failed. Aborting.\n",
375 desc->raidPtr->raidid);
376 (desc->writereq->CompleteFunc) (desc, -100);
377 } else {
378 rf_DiskIOEnqueue(&(desc->raidPtr->Queues[desc->fcol]), desc->writereq, RF_IO_NORMAL_PRIORITY);
379 }
380 return (0);
381 }
382 /* called at interrupt context when the write has completed.
383 * at user level & in the kernel, wake up the copyback thread.
384 * in the simulator, invoke the next copyback directly.
385 * can't free diskqueuedata structs in the kernel b/c we're at interrupt context.
386 */
387 static int
rf_CopybackWriteDoneProc(RF_CopybackDesc_t * desc,int status)388 rf_CopybackWriteDoneProc(RF_CopybackDesc_t *desc, int status)
389 {
390 if (status && status != -100) {
391 printf("raid%d: copyback write failed. Aborting.\n",
392 desc->raidPtr->raidid);
393 }
394 desc->status = status;
395 rf_MCPairWakeupFunc(desc->mcpair);
396 return (0);
397 }
398 /* invoked when the copyback has completed */
399 static void
rf_CopybackComplete(RF_CopybackDesc_t * desc,int status)400 rf_CopybackComplete(RF_CopybackDesc_t *desc, int status)
401 {
402 RF_Raid_t *raidPtr = desc->raidPtr;
403 struct timeval t, diff;
404
405 if (!status) {
406 rf_lock_mutex2(raidPtr->mutex);
407 if (raidPtr->Layout.map->flags & RF_DISTRIBUTE_SPARE) {
408 RF_ASSERT(raidPtr->Layout.map->parityConfig == 'D');
409 rf_FreeSpareTable(raidPtr);
410 } else {
411 raidPtr->Disks[desc->spCol].status = rf_ds_spare;
412 }
413 rf_unlock_mutex2(raidPtr->mutex);
414
415 RF_GETTIME(t);
416 RF_TIMEVAL_DIFF(&desc->starttime, &t, &diff);
417 #if 0
418 printf("Copyback time was %d.%06d seconds\n",
419 (int) diff.tv_sec, (int) diff.tv_usec);
420 #endif
421 } else
422 printf("raid%d: Copyback failure. Status: %d\n",
423 raidPtr->raidid, status);
424
425 RF_Free(desc->databuf, rf_RaidAddressToByte(raidPtr, desc->sectPerSU));
426 rf_FreeMCPair(desc->mcpair);
427 RF_Free(desc, sizeof(*desc));
428
429 rf_copyback_in_progress = 0;
430 rf_ResumeNewRequests(raidPtr);
431 }
432