xref: /dragonfly/sys/dev/raid/vinum/vinumrevive.c (revision 38c2ea22)
1 /*-
2  * Copyright (c) 1997, 1998, 1999
3  *	Nan Yang Computer Services Limited.  All rights reserved.
4  *
5  *  Parts copyright (c) 1997, 1998 Cybernet Corporation, NetMAX project.
6  *
7  *  Written by Greg Lehey
8  *
9  *  This software is distributed under the so-called ``Berkeley
10  *  License'':
11  *
12  * Redistribution and use in source and binary forms, with or without
13  * modification, are permitted provided that the following conditions
14  * are met:
15  * 1. Redistributions of source code must retain the above copyright
16  *    notice, this list of conditions and the following disclaimer.
17  * 2. Redistributions in binary form must reproduce the above copyright
18  *    notice, this list of conditions and the following disclaimer in the
19  *    documentation and/or other materials provided with the distribution.
20  * 3. All advertising materials mentioning features or use of this software
21  *    must display the following acknowledgement:
22  *	This product includes software developed by Nan Yang Computer
23  *      Services Limited.
24  * 4. Neither the name of the Company nor the names of its contributors
25  *    may be used to endorse or promote products derived from this software
26  *    without specific prior written permission.
27  *
28  * This software is provided ``as is'', and any express or implied
29  * warranties, including, but not limited to, the implied warranties of
30  * merchantability and fitness for a particular purpose are disclaimed.
31  * In no event shall the company or contributors be liable for any
32  * direct, indirect, incidental, special, exemplary, or consequential
33  * damages (including, but not limited to, procurement of substitute
34  * goods or services; loss of use, data, or profits; or business
35  * interruption) however caused and on any theory of liability, whether
36  * in contract, strict liability, or tort (including negligence or
37  * otherwise) arising in any way out of the use of this software, even if
38  * advised of the possibility of such damage.
39  *
40  * $Id: vinumrevive.c,v 1.14 2000/12/21 01:55:11 grog Exp grog $
41  * $FreeBSD: src/sys/dev/vinum/vinumrevive.c,v 1.22.2.5 2001/03/13 02:59:43 grog Exp $
42  */
43 
44 #include "vinumhdr.h"
45 #include "request.h"
46 
47 /*
48  * Revive a block of a subdisk.  Return an error
49  * indication.  EAGAIN means successful copy, but
50  * that more blocks remain to be copied.  EINVAL
51  * means that the subdisk isn't associated with a
52  * plex (which means a programming error if we get
53  * here at all; FIXME).
54  */
55 
56 int
57 revive_block(int sdno)
58 {
59     struct sd *sd;
60     struct plex *plex;
61     struct volume *vol;
62     struct buf *bp;
63     cdev_t dev;
64     int error = EAGAIN;
65     int size;						    /* size of revive block, bytes */
66     vinum_off_t plexblkno;					    /* lblkno in plex */
67     int psd;						    /* parity subdisk number */
68     u_int64_t stripe;					    /* stripe number */
69     int paritysd = 0;					    /* set if this is the parity stripe */
70     struct rangelock *lock;				    /* for locking */
71     vinum_off_t stripeoffset;				    /* offset in stripe */
72 
73     plexblkno = 0;					    /* to keep the compiler happy */
74     sd = &SD[sdno];
75     lock = NULL;
76     if (sd->plexno < 0)					    /* no plex? */
77 	return EINVAL;
78     plex = &PLEX[sd->plexno];				    /* point to plex */
79     if (plex->volno >= 0)
80 	vol = &VOL[plex->volno];
81     else
82 	vol = NULL;
83 
84     if ((sd->revive_blocksize == 0)			    /* no block size */
85     ||(sd->revive_blocksize & ((1 << DEV_BSHIFT) - 1)))	    /* or invalid block size */
86 	sd->revive_blocksize = DEFAULT_REVIVE_BLOCKSIZE;
87     else if (sd->revive_blocksize > MAX_REVIVE_BLOCKSIZE)
88 	sd->revive_blocksize = MAX_REVIVE_BLOCKSIZE;
89     size = u64min(sd->revive_blocksize >> DEV_BSHIFT, sd->sectors - sd->revived) << DEV_BSHIFT;
90     sd->reviver = curproc->p_pid;			    /* note who last had a bash at it */
91 
92     /* Now decide where to read from */
93     switch (plex->organization) {
94     case plex_concat:
95 	plexblkno = sd->revived + sd->plexoffset;	    /* corresponding address in plex */
96 	break;
97 
98     case plex_striped:
99 	stripeoffset = sd->revived % plex->stripesize;	    /* offset from beginning of stripe */
100 	if (stripeoffset + (size >> DEV_BSHIFT) > plex->stripesize)
101 	    size = (plex->stripesize - stripeoffset) << DEV_BSHIFT;
102 	plexblkno = sd->plexoffset			    /* base */
103 	    + (sd->revived - stripeoffset) * plex->subdisks /* offset to beginning of stripe */
104 	    + stripeoffset;				    /* offset from beginning of stripe */
105 	break;
106 
107     case plex_raid4:
108     case plex_raid5:
109 	stripeoffset = sd->revived % plex->stripesize;	    /* offset from beginning of stripe */
110 	plexblkno = sd->plexoffset			    /* base */
111 	    + (sd->revived - stripeoffset) * (plex->subdisks - 1) /* offset to beginning of stripe */
112 	    +stripeoffset;				    /* offset from beginning of stripe */
113 	stripe = (sd->revived / plex->stripesize);	    /* stripe number */
114 
115 	/* Make sure we don't go beyond the end of the band. */
116 	size = u64min(size, (plex->stripesize - stripeoffset) << DEV_BSHIFT);
117 	if (plex->organization == plex_raid4)
118 	    psd = plex->subdisks - 1;			    /* parity subdisk for this stripe */
119 	else
120 	    psd = plex->subdisks - 1 - stripe % plex->subdisks;	/* parity subdisk for this stripe */
121 	paritysd = plex->sdnos[psd] == sdno;		    /* note if it's the parity subdisk */
122 
123 	/*
124 	 * Now adjust for the strangenesses
125 	 * in RAID-4 and RAID-5 striping.
126 	 */
127 	if (sd->plexsdno > psd)				    /* beyond the parity stripe, */
128 	    plexblkno -= plex->stripesize;		    /* one stripe less */
129 	else if (paritysd)
130 	    plexblkno -= plex->stripesize * sd->plexsdno;   /* go back to the beginning of the band */
131 	break;
132 
133     case plex_disorg:					    /* to keep the compiler happy */
134 	break;
135     }
136 
137     if (paritysd) {					    /* we're reviving a parity block, */
138 	bp = parityrebuild(plex, sd->revived, size, rebuildparity, &lock, NULL); /* do the grunt work */
139 	if (bp == NULL)					    /* no buffer space */
140 	    return ENOMEM;				    /* chicken out */
141     } else {						    /* data block */
142 	bp = getpbuf(&vinum_conf.physbufs);		    /* Get a buffer */
143 	bp->b_data = Malloc(size);
144 
145 	/*
146 	 * Amount to transfer: block size, unless it
147 	 * would overlap the end.
148 	 */
149 	bp->b_bcount = size;
150 	bp->b_resid = bp->b_bcount;
151 	bp->b_bio1.bio_offset = (off_t)plexblkno << DEV_BSHIFT;		    /* start here */
152 	bp->b_bio1.bio_done = biodone_sync;
153 	bp->b_bio1.bio_flags |= BIO_SYNC;
154 	if (isstriped(plex))				    /* we need to lock striped plexes */
155 	    lock = lockrange(plexblkno << DEV_BSHIFT, bp, plex); /* lock it */
156 	if (vol != NULL)				    /* it's part of a volume, */
157 	    /*
158 	       * First, read the data from the volume.  We
159 	       * don't care which plex, that's bre's job.
160 	     */
161 	    dev = vol->vol_dev;
162 	else						    /* it's an unattached plex */
163 	    dev = PLEX[sd->plexno].plex_dev;
164 
165 	bp->b_cmd = BUF_CMD_READ;
166 	vinumstart(dev, &bp->b_bio1, 1);
167 	biowait(&bp->b_bio1, "drvrd");
168     }
169 
170     if (bp->b_flags & B_ERROR)
171 	error = bp->b_error;
172     else
173 	/* Now write to the subdisk */
174     {
175 	dev = SD[sdno].sd_dev;
176 	KKASSERT(dev != NULL);
177 	bp->b_flags |= B_ORDERED;		    /* and make this an ordered write */
178 	bp->b_cmd = BUF_CMD_WRITE;
179 	bp->b_resid = bp->b_bcount;
180 	bp->b_bio1.bio_offset = (off_t)sd->revived << DEV_BSHIFT;		    /* write it to here */
181 	bp->b_bio1.bio_driver_info = dev;
182 	bp->b_bio1.bio_done = biodone_sync;
183 	sdio(&bp->b_bio1);				    /* perform the I/O */
184 	biowait(&bp->b_bio1, "drvwr");
185 	if (bp->b_flags & B_ERROR)
186 	    error = bp->b_error;
187 	else {
188 	    sd->revived += bp->b_bcount >> DEV_BSHIFT;	    /* moved this much further down */
189 	    if (sd->revived >= sd->sectors) {		    /* finished */
190 		sd->revived = 0;
191 		set_sd_state(sdno, sd_up, setstate_force);  /* bring the sd up */
192 		log(LOG_INFO, "vinum: %s is %s\n", sd->name, sd_state(sd->state));
193 		save_config();				    /* and save the updated configuration */
194 		error = 0;				    /* we're done */
195 	    }
196 	}
197 	if (lock)					    /* we took a lock, */
198 	    unlockrange(sd->plexno, lock);		    /* give it back */
199 	while (sd->waitlist) {				    /* we have waiting requests */
200 #if VINUMDEBUG
201 	    struct request *rq = sd->waitlist;
202 	    cdev_t dev;
203 
204 	    if (debug & DEBUG_REVIVECONFLICT) {
205 		dev = rq->bio->bio_driver_info;
206 		log(LOG_DEBUG,
207 		    "Relaunch revive conflict sd %d: %p\n%s dev %d.%d, offset 0x%llx, length %d\n",
208 		    rq->sdno,
209 		    rq,
210 		    (rq->bio->bio_buf->b_cmd == BUF_CMD_READ) ? "Read" : "Write",
211 		    major(dev),
212 		    minor(dev),
213 		    rq->bio->bio_offset,
214 		    rq->bio->bio_buf->b_bcount);
215 	    }
216 #endif
217 	    launch_requests(sd->waitlist, 1);		    /* do them now */
218 	    sd->waitlist = sd->waitlist->next;		    /* and move on to the next */
219 	}
220     }
221     Free(bp->b_data);
222     relpbuf(bp, &vinum_conf.physbufs);
223     return error;
224 }
225 
226 /*
227  * Check or rebuild the parity blocks of a RAID-4
228  * or RAID-5 plex.
229  *
230  * The variables plex->checkblock and
231  * plex->rebuildblock represent the
232  * subdisk-relative address of the stripe we're
233  * looking at, not the plex-relative address.  We
234  * store it in the plex and not as a local
235  * variable because this function could be
236  * stopped, and we don't want to repeat the part
237  * we've already done.  This is also the reason
238  * why we don't initialize it here except at the
239  * end.  It gets initialized with the plex on
240  * creation.
241  *
242  * Each call to this function processes at most
243  * one stripe.  We can't loop in this function,
244  * because we're unstoppable, so we have to be
245  * called repeatedly from userland.
246  */
247 void
248 parityops(struct vinum_ioctl_msg *data)
249 {
250     int plexno;
251     struct plex *plex;
252     int size;						    /* I/O transfer size, bytes */
253     vinum_off_t stripe;						    /* stripe number in plex */
254     int psd;						    /* parity subdisk number */
255     struct rangelock *lock;				    /* lock on stripe */
256     struct _ioctl_reply *reply;
257     off_t pstripe;					    /* pointer to our stripe counter */
258     struct buf *pbp;
259     off_t errorloc;					    /* offset of parity error */
260     enum parityop op;					    /* operation to perform */
261 
262     plexno = data->index;
263     op = data->op;
264     pbp = NULL;
265     reply = (struct _ioctl_reply *) data;
266     reply->error = EAGAIN;				    /* expect to repeat this call */
267     plex = &PLEX[plexno];
268     if (!isparity(plex)) {				    /* not RAID-4 or RAID-5 */
269 	reply->error = EINVAL;
270 	return;
271     } else if (plex->state < plex_flaky) {
272 	reply->error = EIO;
273 	strcpy(reply->msg, "Plex is not completely accessible\n");
274 	return;
275     }
276     pstripe = data->offset;
277     stripe = pstripe / plex->stripesize;		    /* stripe number */
278     psd = plex->subdisks - 1 - stripe % plex->subdisks;	    /* parity subdisk for this stripe */
279     size = imin(DEFAULT_REVIVE_BLOCKSIZE,		    /* one block at a time */
280 	plex->stripesize << DEV_BSHIFT);
281 
282     pbp = parityrebuild(plex, pstripe, size, op, &lock, &errorloc); /* do the grunt work */
283     if (pbp == NULL) {					    /* no buffer space */
284 	reply->error = ENOMEM;
285 	return;						    /* chicken out */
286     }
287     /*
288      * Now we have a result in the data buffer of
289      * the parity buffer header, which we have kept.
290      * Decide what to do with it.
291      */
292     reply->msg[0] = '\0';				    /* until shown otherwise */
293     if ((pbp->b_flags & B_ERROR) == 0) {		    /* no error */
294 	if ((op == rebuildparity)
295 	    || (op == rebuildandcheckparity)) {
296 	    pbp->b_cmd = BUF_CMD_WRITE;
297 	    pbp->b_resid = pbp->b_bcount;
298 	    pbp->b_bio1.bio_done = biodone_sync;
299 	    sdio(&pbp->b_bio1);				    /* write the parity block */
300 	    biowait(&pbp->b_bio1, "drvwr");
301 	}
302 	if (((op == checkparity)
303 		|| (op == rebuildandcheckparity))
304 	    && (errorloc != -1)) {
305 	    if (op == checkparity)
306 		reply->error = EIO;
307 	    ksprintf(reply->msg,
308 		"Parity incorrect at offset 0x%llx\n",
309 		(long long)errorloc);
310 	}
311 	if (reply->error == EAGAIN) {			    /* still OK, */
312 	    plex->checkblock = pstripe + (pbp->b_bcount >> DEV_BSHIFT);	/* moved this much further down */
313 	    if (plex->checkblock >= SD[plex->sdnos[0]].sectors) { /* finished */
314 		plex->checkblock = 0;
315 		reply->error = 0;
316 	    }
317 	}
318     }
319     if (pbp->b_flags & B_ERROR)
320 	reply->error = pbp->b_error;
321     Free(pbp->b_data);
322     relpbuf(pbp, &vinum_conf.physbufs);
323     unlockrange(plexno, lock);
324 }
325 
326 /*
327  * Rebuild a parity stripe.  Return pointer to
328  * parity bp.  On return,
329  *
330  * 1.  The band is locked.  The caller must unlock
331  *     the band and release the buffer header.
332  *
333  * 2.  All buffer headers except php have been
334  *     released.  The caller must release pbp.
335  *
336  * 3.  For checkparity and rebuildandcheckparity,
337  *     the parity is compared with the current
338  *     parity block.  If it's different, the
339  *     offset of the error is returned to
340  *     errorloc.  The caller can set the value of
341  *     the pointer to NULL if this is called for
342  *     rebuilding parity.
343  *
344  * pstripe is the subdisk-relative base address of
345  * the data to be reconstructed, size is the size
346  * of the transfer in bytes.
347  */
348 struct buf *
349 parityrebuild(struct plex *plex,
350     vinum_off_t pstripe,
351     int size,
352     enum parityop op,
353     struct rangelock **lockp,
354     off_t * errorloc)
355 {
356     int error;
357     int sdno;
358     u_int64_t stripe;					    /* stripe number */
359     int *parity_buf;					    /* buffer address for current parity block */
360     int *newparity_buf;					    /* and for new parity block */
361     int mysize;						    /* I/O transfer size for this transfer */
362     int isize;						    /* mysize in ints */
363     int i;
364     int psd;						    /* parity subdisk number */
365     int newpsd;						    /* and "subdisk number" of new parity */
366     struct buf **bpp;					    /* pointers to our bps */
367     struct buf *pbp;					    /* buffer header for parity stripe */
368     int *sbuf;
369     int bufcount;					    /* number of buffers we need */
370 
371     stripe = pstripe / plex->stripesize;		    /* stripe number */
372     psd = plex->subdisks - 1 - stripe % plex->subdisks;	    /* parity subdisk for this stripe */
373     parity_buf = NULL;					    /* to keep the compiler happy */
374     error = 0;
375 
376     /*
377      * It's possible that the default transfer size
378      * we chose is not a factor of the stripe size.
379      * We *must* limit this operation to a single
380      * stripe, at least for RAID-5 rebuild, since
381      * the parity subdisk changes between stripes,
382      * so in this case we need to perform a short
383      * transfer.  Set variable mysize to reflect
384      * this.
385      */
386     mysize = u64min(size, (plex->stripesize * (stripe + 1) - pstripe) << DEV_BSHIFT);
387     isize = mysize / (sizeof(int));			    /* number of ints in the buffer */
388     bufcount = plex->subdisks + 1;			    /* sd buffers plus result buffer */
389     newpsd = plex->subdisks;
390     bpp = (struct buf **) Malloc(bufcount * sizeof(struct buf *)); /* array of pointers to bps */
391 
392     /* First, build requests for all subdisks */
393     for (sdno = 0; sdno < bufcount; sdno++) {		    /* for each subdisk */
394 	if ((sdno != psd) || (op != rebuildparity)) {
395 	    /* Get a buffer header and initialize it. */
396 	    bpp[sdno] = getpbuf(&vinum_conf.physbufs);	    /* Get a buffer */
397 	    bpp[sdno]->b_data = Malloc(mysize);
398 	    if (sdno == psd)
399 		parity_buf = (int *) bpp[sdno]->b_data;
400 	    if (sdno == newpsd)				    /* the new one? */
401 		bpp[sdno]->b_bio1.bio_driver_info = SD[plex->sdnos[psd]].sd_dev; /* write back to the parity SD */
402 	    else
403 		bpp[sdno]->b_bio1.bio_driver_info = SD[plex->sdnos[sdno]].sd_dev;	/* device number */
404 	    KKASSERT(bpp[sdno]->b_bio1.bio_driver_info);
405 	    bpp[sdno]->b_cmd = BUF_CMD_READ;	    /* either way, read it */
406 	    bpp[sdno]->b_bcount = mysize;
407 	    bpp[sdno]->b_resid = bpp[sdno]->b_bcount;
408 	    bpp[sdno]->b_bio1.bio_offset = (off_t)pstripe << DEV_BSHIFT;	    /* transfer from here */
409 	    bpp[sdno]->b_bio1.bio_done = biodone_sync;
410 	}
411     }
412 
413     /* Initialize result buffer */
414     pbp = bpp[newpsd];
415     newparity_buf = (int *) bpp[newpsd]->b_data;
416     bzero(newparity_buf, mysize);
417 
418     /*
419      * Now lock the stripe with the first non-parity
420      * bp as locking bp.
421      */
422     *lockp = lockrange(pstripe * plex->stripesize * (plex->subdisks - 1),
423 	bpp[psd ? 0 : 1],
424 	plex);
425 
426     /*
427      * Then issue requests for all subdisks in
428      * parallel.  Don't transfer the parity stripe
429      * if we're rebuilding parity, unless we also
430      * want to check it.
431      */
432     for (sdno = 0; sdno < plex->subdisks; sdno++) {	    /* for each real subdisk */
433 	if ((sdno != psd) || (op != rebuildparity)) {
434 	    sdio(&bpp[sdno]->b_bio1);
435 	}
436     }
437 
438     /*
439      * Next, wait for the requests to complete.
440      * We wait in the order in which they were
441      * issued, which isn't necessarily the order in
442      * which they complete, but we don't have a
443      * convenient way of doing the latter, and the
444      * delay is minimal.
445      */
446     for (sdno = 0; sdno < plex->subdisks; sdno++) {	    /* for each subdisk */
447 	if ((sdno != psd) || (op != rebuildparity)) {
448 	    biowait(&bpp[sdno]->b_bio1, "drvio");
449 	    if (bpp[sdno]->b_flags & B_ERROR)		    /* can't read, */
450 		error = bpp[sdno]->b_error;
451 	    else if (sdno != psd) {			    /* update parity */
452 		sbuf = (int *) bpp[sdno]->b_data;
453 		for (i = 0; i < isize; i++)
454 		    newparity_buf[i] ^= sbuf[i];	    /* xor in the buffer */
455 	    }
456 	}
457 	if (sdno != psd) {				    /* release all bps except parity */
458 	    Free(bpp[sdno]->b_data);
459 	    relpbuf(bpp[sdno], &vinum_conf.physbufs);	    /* give back our resources */
460 	}
461     }
462 
463     /*
464      * If we're checking, compare the calculated
465      * and the read parity block.  If they're
466      * different, return the plex-relative offset;
467      * otherwise return -1.
468      */
469     if ((op == checkparity)
470 	|| (op == rebuildandcheckparity)) {
471 	*errorloc = -1;					    /* no error yet */
472 	for (i = 0; i < isize; i++) {
473 	    if (parity_buf[i] != newparity_buf[i]) {
474 		*errorloc = (off_t) (pstripe << DEV_BSHIFT) * (plex->subdisks - 1)
475 		    + i * sizeof(int);
476 		break;
477 	    }
478 	}
479 	Free(bpp[psd]->b_data);
480 	relpbuf(bpp[psd], &vinum_conf.physbufs);	    /* give back our resources */
481     }
482     /* release our resources */
483     Free(bpp);
484     if (error) {
485 	pbp->b_flags |= B_ERROR;
486 	pbp->b_error = error;
487     }
488     return pbp;
489 }
490 
491 /*
492  * Initialize a subdisk by writing zeroes to the
493  * complete address space.  If verify is set,
494  * check each transfer for correctness.
495  *
496  * Each call to this function writes (and maybe
497  * checks) a single block.
498  */
499 int
500 initsd(int sdno, int verify)
501 {
502     struct sd *sd;
503     struct plex *plex;
504     struct volume *vol;
505     struct buf *bp;
506     int error;
507     int size;						    /* size of init block, bytes */
508     vinum_off_t plexblkno;					    /* lblkno in plex */
509     int verified;					    /* set when we're happy with what we wrote */
510 
511     error = 0;
512     plexblkno = 0;					    /* to keep the compiler happy */
513     sd = &SD[sdno];
514     if (sd->plexno < 0)					    /* no plex? */
515 	return EINVAL;
516     plex = &PLEX[sd->plexno];				    /* point to plex */
517     if (plex->volno >= 0)
518 	vol = &VOL[plex->volno];
519     else
520 	vol = NULL;
521 
522     if (sd->init_blocksize == 0) {
523 	if (plex->stripesize != 0)			    /* we're striped, don't init more than */
524 	    sd->init_blocksize = u64min(DEFAULT_REVIVE_BLOCKSIZE, /* one block at a time */
525 		plex->stripesize << DEV_BSHIFT);
526 	else
527 	    sd->init_blocksize = DEFAULT_REVIVE_BLOCKSIZE;
528     } else if (sd->init_blocksize > MAX_REVIVE_BLOCKSIZE)
529 	sd->init_blocksize = MAX_REVIVE_BLOCKSIZE;
530 
531     size = u64min(sd->init_blocksize >> DEV_BSHIFT, sd->sectors - sd->initialized) << DEV_BSHIFT;
532 
533     bp = getpbuf(&vinum_conf.physbufs);		    /* Get a buffer */
534     bp->b_data = Malloc(size);
535 
536     verified = 0;
537     while (!verified) {					    /* until we're happy with it, */
538 	bp->b_bcount = size;
539 	bp->b_resid = bp->b_bcount;
540 	bp->b_bio1.bio_offset = (off_t)sd->initialized << DEV_BSHIFT;		    /* write it to here */
541 	bp->b_bio1.bio_driver_info = SD[sdno].sd_dev;
542 	bp->b_bio1.bio_done = biodone_sync;
543 	KKASSERT(bp->b_bio1.bio_driver_info);
544 	bzero(bp->b_data, bp->b_bcount);
545 	bp->b_cmd = BUF_CMD_WRITE;
546 	sdio(&bp->b_bio1);		    /* perform the I/O */
547 	biowait(&bp->b_bio1, "drvwr");
548 	if (bp->b_flags & B_ERROR)
549 	    error = bp->b_error;
550 	if ((error == 0) && verify) {			    /* check that it got there */
551 	    bp->b_bcount = size;
552 	    bp->b_resid = bp->b_bcount;
553 	    bp->b_bio1.bio_offset = (off_t)sd->initialized << DEV_BSHIFT;	    /* read from here */
554 	    bp->b_bio1.bio_driver_info = SD[sdno].sd_dev;
555 	    bp->b_bio1.bio_done = biodone_sync;
556 	    KKASSERT(bp->b_bio1.bio_driver_info);
557 	    bp->b_cmd = BUF_CMD_READ;		    /* read it back */
558 	    sdio(&bp->b_bio1);
559 	    biowait(&bp->b_bio1, "drvrd");
560 	    /*
561 	     * XXX Bug fix code.  This is hopefully no
562 	     * longer needed (21 February 2000).
563 	     */
564 	    if (bp->b_flags & B_ERROR)
565 		error = bp->b_error;
566 	    else if ((*bp->b_data != 0)		    /* first word spammed */
567 	    ||(bcmp(bp->b_data, &bp->b_data[1], bp->b_bcount - 1))) { /* or one of the others */
568 		kprintf("vinum: init error on %s, offset 0x%llx sectors\n",
569 		    sd->name,
570 		    (long long) sd->initialized);
571 		verified = 0;
572 	    } else
573 		verified = 1;
574 	} else
575 	    verified = 1;
576     }
577     Free(bp->b_data);
578     relpbuf(bp, &vinum_conf.physbufs);
579     if (error == 0) {					    /* did it, */
580 	sd->initialized += size >> DEV_BSHIFT;		    /* moved this much further down */
581 	if (sd->initialized >= sd->sectors) {		    /* finished */
582 	    sd->initialized = 0;
583 	    set_sd_state(sdno, sd_initialized, setstate_force);	/* bring the sd up */
584 	    log(LOG_INFO, "vinum: %s is %s\n", sd->name, sd_state(sd->state));
585 	    save_config();				    /* and save the updated configuration */
586 	} else						    /* more to go, */
587 	    error = EAGAIN;				    /* ya'll come back, see? */
588     }
589     return error;
590 }
591 
592 /* Local Variables: */
593 /* fill-column: 50 */
594 /* End: */
595