xref: /dragonfly/sys/dev/raid/vinum/vinumrevive.c (revision fe76c4fb)
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  * $DragonFly: src/sys/dev/raid/vinum/vinumrevive.c,v 1.10 2006/04/30 17:22:17 dillon Exp $
43  */
44 
45 #include "vinumhdr.h"
46 #include "request.h"
47 
48 /*
49  * Revive a block of a subdisk.  Return an error
50  * indication.  EAGAIN means successful copy, but
51  * that more blocks remain to be copied.  EINVAL
52  * means that the subdisk isn't associated with a
53  * plex (which means a programming error if we get
54  * here at all; FIXME).
55  */
56 
57 int
58 revive_block(int sdno)
59 {
60     struct sd *sd;
61     struct plex *plex;
62     struct volume *vol;
63     struct buf *bp;
64     dev_t dev;
65     int error = EAGAIN;
66     int size;						    /* size of revive block, bytes */
67     daddr_t plexblkno;					    /* lblkno in plex */
68     int psd;						    /* parity subdisk number */
69     u_int64_t stripe;					    /* stripe number */
70     int paritysd = 0;					    /* set if this is the parity stripe */
71     struct rangelock *lock;				    /* for locking */
72     daddr_t stripeoffset;				    /* offset in stripe */
73 
74     plexblkno = 0;					    /* to keep the compiler happy */
75     sd = &SD[sdno];
76     lock = NULL;
77     if (sd->plexno < 0)					    /* no plex? */
78 	return EINVAL;
79     plex = &PLEX[sd->plexno];				    /* point to plex */
80     if (plex->volno >= 0)
81 	vol = &VOL[plex->volno];
82     else
83 	vol = NULL;
84 
85     if ((sd->revive_blocksize == 0)			    /* no block size */
86     ||(sd->revive_blocksize & ((1 << DEV_BSHIFT) - 1)))	    /* or invalid block size */
87 	sd->revive_blocksize = DEFAULT_REVIVE_BLOCKSIZE;
88     else if (sd->revive_blocksize > MAX_REVIVE_BLOCKSIZE)
89 	sd->revive_blocksize = MAX_REVIVE_BLOCKSIZE;
90     size = min(sd->revive_blocksize >> DEV_BSHIFT, sd->sectors - sd->revived) << DEV_BSHIFT;
91     sd->reviver = curproc->p_pid;			    /* note who last had a bash at it */
92 
93     /* Now decide where to read from */
94     switch (plex->organization) {
95     case plex_concat:
96 	plexblkno = sd->revived + sd->plexoffset;	    /* corresponding address in plex */
97 	break;
98 
99     case plex_striped:
100 	stripeoffset = sd->revived % plex->stripesize;	    /* offset from beginning of stripe */
101 	if (stripeoffset + (size >> DEV_BSHIFT) > plex->stripesize)
102 	    size = (plex->stripesize - stripeoffset) << DEV_BSHIFT;
103 	plexblkno = sd->plexoffset			    /* base */
104 	    + (sd->revived - stripeoffset) * plex->subdisks /* offset to beginning of stripe */
105 	    + stripeoffset;				    /* offset from beginning of stripe */
106 	break;
107 
108     case plex_raid4:
109     case plex_raid5:
110 	stripeoffset = sd->revived % plex->stripesize;	    /* offset from beginning of stripe */
111 	plexblkno = sd->plexoffset			    /* base */
112 	    + (sd->revived - stripeoffset) * (plex->subdisks - 1) /* offset to beginning of stripe */
113 	    +stripeoffset;				    /* offset from beginning of stripe */
114 	stripe = (sd->revived / plex->stripesize);	    /* stripe number */
115 
116 	/* Make sure we don't go beyond the end of the band. */
117 	size = min(size, (plex->stripesize - stripeoffset) << DEV_BSHIFT);
118 	if (plex->organization == plex_raid4)
119 	    psd = plex->subdisks - 1;			    /* parity subdisk for this stripe */
120 	else
121 	    psd = plex->subdisks - 1 - stripe % plex->subdisks;	/* parity subdisk for this stripe */
122 	paritysd = plex->sdnos[psd] == sdno;		    /* note if it's the parity subdisk */
123 
124 	/*
125 	 * Now adjust for the strangenesses
126 	 * in RAID-4 and RAID-5 striping.
127 	 */
128 	if (sd->plexsdno > psd)				    /* beyond the parity stripe, */
129 	    plexblkno -= plex->stripesize;		    /* one stripe less */
130 	else if (paritysd)
131 	    plexblkno -= plex->stripesize * sd->plexsdno;   /* go back to the beginning of the band */
132 	break;
133 
134     case plex_disorg:					    /* to keep the compiler happy */
135 	break;
136     }
137 
138     if (paritysd) {					    /* we're reviving a parity block, */
139 	bp = parityrebuild(plex, sd->revived, size, rebuildparity, &lock, NULL); /* do the grunt work */
140 	if (bp == NULL)					    /* no buffer space */
141 	    return ENOMEM;				    /* chicken out */
142     } else {						    /* data block */
143 	crit_enter();
144 	bp = geteblk(size);				    /* Get a buffer */
145 	crit_exit();
146 	if (bp == NULL)
147 	    return ENOMEM;
148 
149 	/*
150 	 * Amount to transfer: block size, unless it
151 	 * would overlap the end.
152 	 */
153 	bp->b_bcount = size;
154 	bp->b_resid = bp->b_bcount;
155 	bp->b_bio1.bio_offset = (off_t)plexblkno << DEV_BSHIFT;		    /* start here */
156 	if (isstriped(plex))				    /* we need to lock striped plexes */
157 	    lock = lockrange(plexblkno << DEV_BSHIFT, bp, plex); /* lock it */
158 	if (vol != NULL)				    /* it's part of a volume, */
159 	    /*
160 	       * First, read the data from the volume.  We
161 	       * don't care which plex, that's bre's job.
162 	     */
163 	    dev = VINUMDEV(plex->volno, 0, 0, VINUM_VOLUME_TYPE);	/* create the device number */
164 	else						    /* it's an unattached plex */
165 	    dev = VINUM_PLEX(sd->plexno);		    /* create the device number */
166 
167 	bp->b_flags = B_PAGING;		    /* either way, read it */
168 	bp->b_cmd = BUF_CMD_READ;
169 	vinumstart(dev, &bp->b_bio1, 1);
170 	biowait(bp);
171     }
172 
173     if (bp->b_flags & B_ERROR)
174 	error = bp->b_error;
175     else
176 	/* Now write to the subdisk */
177     {
178 	dev = VINUM_SD(sdno);			    /* create the device number */
179 	bp->b_flags = B_ORDERED | B_PAGING;    /* and make this an ordered write */
180 	bp->b_cmd = BUF_CMD_WRITE;
181 	bp->b_resid = bp->b_bcount;
182 	bp->b_bio1.bio_offset = (off_t)sd->revived << DEV_BSHIFT;		    /* write it to here */
183 	bp->b_bio1.bio_driver_info = dev;
184 	sdio(&bp->b_bio1);				    /* perform the I/O */
185 	biowait(bp);
186 	if (bp->b_flags & B_ERROR)
187 	    error = bp->b_error;
188 	else {
189 	    sd->revived += bp->b_bcount >> DEV_BSHIFT;	    /* moved this much further down */
190 	    if (sd->revived >= sd->sectors) {		    /* finished */
191 		sd->revived = 0;
192 		set_sd_state(sdno, sd_up, setstate_force);  /* bring the sd up */
193 		log(LOG_INFO, "vinum: %s is %s\n", sd->name, sd_state(sd->state));
194 		save_config();				    /* and save the updated configuration */
195 		error = 0;				    /* we're done */
196 	    }
197 	}
198 	if (lock)					    /* we took a lock, */
199 	    unlockrange(sd->plexno, lock);		    /* give it back */
200 	while (sd->waitlist) {				    /* we have waiting requests */
201 #if VINUMDEBUG
202 	    struct request *rq = sd->waitlist;
203 	    dev_t dev;
204 
205 	    if (debug & DEBUG_REVIVECONFLICT) {
206 		dev = rq->bio->bio_driver_info;
207 		log(LOG_DEBUG,
208 		    "Relaunch revive conflict sd %d: %p\n%s dev %d.%d, offset 0x%llx, length %d\n",
209 		    rq->sdno,
210 		    rq,
211 		    (rq->bio->bio_buf->b_cmd == BUF_CMD_READ) ? "Read" : "Write",
212 		    major(dev),
213 		    minor(dev),
214 		    rq->bio->bio_offset,
215 		    rq->bio->bio_buf->b_bcount);
216 	    }
217 #endif
218 	    launch_requests(sd->waitlist, 1);		    /* do them now */
219 	    sd->waitlist = sd->waitlist->next;		    /* and move on to the next */
220 	}
221     }
222     if (bp->b_qindex == 0) {				    /* not on a queue, */
223 	bp->b_flags |= B_INVAL;
224 	bp->b_flags &= ~B_ERROR;
225 	brelse(bp);					    /* is this kosher? */
226     }
227     return error;
228 }
229 
230 /*
231  * Check or rebuild the parity blocks of a RAID-4
232  * or RAID-5 plex.
233  *
234  * The variables plex->checkblock and
235  * plex->rebuildblock represent the
236  * subdisk-relative address of the stripe we're
237  * looking at, not the plex-relative address.  We
238  * store it in the plex and not as a local
239  * variable because this function could be
240  * stopped, and we don't want to repeat the part
241  * we've already done.  This is also the reason
242  * why we don't initialize it here except at the
243  * end.  It gets initialized with the plex on
244  * creation.
245  *
246  * Each call to this function processes at most
247  * one stripe.  We can't loop in this function,
248  * because we're unstoppable, so we have to be
249  * called repeatedly from userland.
250  */
251 void
252 parityops(struct vinum_ioctl_msg *data)
253 {
254     int plexno;
255     struct plex *plex;
256     int size;						    /* I/O transfer size, bytes */
257     int stripe;						    /* stripe number in plex */
258     int psd;						    /* parity subdisk number */
259     struct rangelock *lock;				    /* lock on stripe */
260     struct _ioctl_reply *reply;
261     off_t pstripe;					    /* pointer to our stripe counter */
262     struct buf *pbp;
263     off_t errorloc;					    /* offset of parity error */
264     enum parityop op;					    /* operation to perform */
265 
266     plexno = data->index;
267     op = data->op;
268     pbp = NULL;
269     reply = (struct _ioctl_reply *) data;
270     reply->error = EAGAIN;				    /* expect to repeat this call */
271     plex = &PLEX[plexno];
272     if (!isparity(plex)) {				    /* not RAID-4 or RAID-5 */
273 	reply->error = EINVAL;
274 	return;
275     } else if (plex->state < plex_flaky) {
276 	reply->error = EIO;
277 	strcpy(reply->msg, "Plex is not completely accessible\n");
278 	return;
279     }
280     pstripe = data->offset;
281     stripe = pstripe / plex->stripesize;		    /* stripe number */
282     psd = plex->subdisks - 1 - stripe % plex->subdisks;	    /* parity subdisk for this stripe */
283     size = min(DEFAULT_REVIVE_BLOCKSIZE,		    /* one block at a time */
284 	plex->stripesize << DEV_BSHIFT);
285 
286     pbp = parityrebuild(plex, pstripe, size, op, &lock, &errorloc); /* do the grunt work */
287     if (pbp == NULL) {					    /* no buffer space */
288 	reply->error = ENOMEM;
289 	return;						    /* chicken out */
290     }
291     /*
292      * Now we have a result in the data buffer of
293      * the parity buffer header, which we have kept.
294      * Decide what to do with it.
295      */
296     reply->msg[0] = '\0';				    /* until shown otherwise */
297     if ((pbp->b_flags & B_ERROR) == 0) {		    /* no error */
298 	if ((op == rebuildparity)
299 	    || (op == rebuildandcheckparity)) {
300 	    pbp->b_cmd = BUF_CMD_WRITE;
301 	    pbp->b_resid = pbp->b_bcount;
302 	    sdio(&pbp->b_bio1);				    /* write the parity block */
303 	    biowait(pbp);
304 	}
305 	if (((op == checkparity)
306 		|| (op == rebuildandcheckparity))
307 	    && (errorloc != -1)) {
308 	    if (op == checkparity)
309 		reply->error = EIO;
310 	    sprintf(reply->msg,
311 		"Parity incorrect at offset 0x%llx\n",
312 		errorloc);
313 	}
314 	if (reply->error == EAGAIN) {			    /* still OK, */
315 	    plex->checkblock = pstripe + (pbp->b_bcount >> DEV_BSHIFT);	/* moved this much further down */
316 	    if (plex->checkblock >= SD[plex->sdnos[0]].sectors) { /* finished */
317 		plex->checkblock = 0;
318 		reply->error = 0;
319 	    }
320 	}
321     }
322     if (pbp->b_flags & B_ERROR)
323 	reply->error = pbp->b_error;
324     pbp->b_flags |= B_INVAL;
325     pbp->b_flags &= ~B_ERROR;
326     brelse(pbp);
327     unlockrange(plexno, lock);
328 }
329 
330 /*
331  * Rebuild a parity stripe.  Return pointer to
332  * parity bp.  On return,
333  *
334  * 1.  The band is locked.  The caller must unlock
335  *     the band and release the buffer header.
336  *
337  * 2.  All buffer headers except php have been
338  *     released.  The caller must release pbp.
339  *
340  * 3.  For checkparity and rebuildandcheckparity,
341  *     the parity is compared with the current
342  *     parity block.  If it's different, the
343  *     offset of the error is returned to
344  *     errorloc.  The caller can set the value of
345  *     the pointer to NULL if this is called for
346  *     rebuilding parity.
347  *
348  * pstripe is the subdisk-relative base address of
349  * the data to be reconstructed, size is the size
350  * of the transfer in bytes.
351  */
352 struct buf *
353 parityrebuild(struct plex *plex,
354     u_int64_t pstripe,
355     int size,
356     enum parityop op,
357     struct rangelock **lockp,
358     off_t * errorloc)
359 {
360     int error;
361     int sdno;
362     u_int64_t stripe;					    /* stripe number */
363     int *parity_buf;					    /* buffer address for current parity block */
364     int *newparity_buf;					    /* and for new parity block */
365     int mysize;						    /* I/O transfer size for this transfer */
366     int isize;						    /* mysize in ints */
367     int i;
368     int psd;						    /* parity subdisk number */
369     int newpsd;						    /* and "subdisk number" of new parity */
370     struct buf **bpp;					    /* pointers to our bps */
371     struct buf *pbp;					    /* buffer header for parity stripe */
372     int *sbuf;
373     int bufcount;					    /* number of buffers we need */
374 
375     stripe = pstripe / plex->stripesize;		    /* stripe number */
376     psd = plex->subdisks - 1 - stripe % plex->subdisks;	    /* parity subdisk for this stripe */
377     parity_buf = NULL;					    /* to keep the compiler happy */
378     error = 0;
379 
380     /*
381      * It's possible that the default transfer size
382      * we chose is not a factor of the stripe size.
383      * We *must* limit this operation to a single
384      * stripe, at least for RAID-5 rebuild, since
385      * the parity subdisk changes between stripes,
386      * so in this case we need to perform a short
387      * transfer.  Set variable mysize to reflect
388      * this.
389      */
390     mysize = min(size, (plex->stripesize * (stripe + 1) - pstripe) << DEV_BSHIFT);
391     isize = mysize / (sizeof(int));			    /* number of ints in the buffer */
392     bufcount = plex->subdisks + 1;			    /* sd buffers plus result buffer */
393     newpsd = plex->subdisks;
394     bpp = (struct buf **) Malloc(bufcount * sizeof(struct buf *)); /* array of pointers to bps */
395 
396     /* First, build requests for all subdisks */
397     for (sdno = 0; sdno < bufcount; sdno++) {		    /* for each subdisk */
398 	if ((sdno != psd) || (op != rebuildparity)) {
399 	    /* Get a buffer header and initialize it. */
400 	    crit_enter();
401 	    bpp[sdno] = geteblk(mysize);		    /* Get a buffer */
402 	    if (bpp[sdno] == NULL) {
403 		while (sdno-- > 0) {			    /* release the ones we got */
404 		    bpp[sdno]->b_flags |= B_INVAL;
405 		    brelse(bpp[sdno]);			    /* give back our resources */
406 		}
407 		crit_exit();
408 		printf("vinum: can't allocate buffer space for parity op.\n");
409 		return NULL;				    /* no bpps */
410 	    }
411 	    crit_exit();
412 	    if (sdno == psd)
413 		parity_buf = (int *) bpp[sdno]->b_data;
414 	    if (sdno == newpsd)				    /* the new one? */
415 		bpp[sdno]->b_bio1.bio_driver_info = VINUM_SD(plex->sdnos[psd]); /* write back to the parity SD */
416 	    else
417 		bpp[sdno]->b_bio1.bio_driver_info = VINUM_SD(plex->sdnos[sdno]);	/* device number */
418 	    bpp[sdno]->b_cmd = BUF_CMD_READ;	    /* either way, read it */
419 	    bpp[sdno]->b_flags = B_PAGING;
420 	    bpp[sdno]->b_bcount = mysize;
421 	    bpp[sdno]->b_resid = bpp[sdno]->b_bcount;
422 	    bpp[sdno]->b_bio1.bio_offset = (off_t)pstripe << DEV_BSHIFT;	    /* transfer from here */
423 	}
424     }
425 
426     /* Initialize result buffer */
427     pbp = bpp[newpsd];
428     newparity_buf = (int *) bpp[newpsd]->b_data;
429     bzero(newparity_buf, mysize);
430 
431     /*
432      * Now lock the stripe with the first non-parity
433      * bp as locking bp.
434      */
435     *lockp = lockrange(pstripe * plex->stripesize * (plex->subdisks - 1),
436 	bpp[psd ? 0 : 1],
437 	plex);
438 
439     /*
440      * Then issue requests for all subdisks in
441      * parallel.  Don't transfer the parity stripe
442      * if we're rebuilding parity, unless we also
443      * want to check it.
444      */
445     for (sdno = 0; sdno < plex->subdisks; sdno++) {	    /* for each real subdisk */
446 	if ((sdno != psd) || (op != rebuildparity)) {
447 	    sdio(&bpp[sdno]->b_bio1);
448 	}
449     }
450 
451     /*
452      * Next, wait for the requests to complete.
453      * We wait in the order in which they were
454      * issued, which isn't necessarily the order in
455      * which they complete, but we don't have a
456      * convenient way of doing the latter, and the
457      * delay is minimal.
458      */
459     for (sdno = 0; sdno < plex->subdisks; sdno++) {	    /* for each subdisk */
460 	if ((sdno != psd) || (op != rebuildparity)) {
461 	    biowait(bpp[sdno]);
462 	    if (bpp[sdno]->b_flags & B_ERROR)		    /* can't read, */
463 		error = bpp[sdno]->b_error;
464 	    else if (sdno != psd) {			    /* update parity */
465 		sbuf = (int *) bpp[sdno]->b_data;
466 		for (i = 0; i < isize; i++)
467 		    ((int *) newparity_buf)[i] ^= sbuf[i];  /* xor in the buffer */
468 	    }
469 	}
470 	if (sdno != psd) {				    /* release all bps except parity */
471 	    bpp[sdno]->b_flags |= B_INVAL;
472 	    brelse(bpp[sdno]);				    /* give back our resources */
473 	}
474     }
475 
476     /*
477      * If we're checking, compare the calculated
478      * and the read parity block.  If they're
479      * different, return the plex-relative offset;
480      * otherwise return -1.
481      */
482     if ((op == checkparity)
483 	|| (op == rebuildandcheckparity)) {
484 	*errorloc = -1;					    /* no error yet */
485 	for (i = 0; i < isize; i++) {
486 	    if (parity_buf[i] != newparity_buf[i]) {
487 		*errorloc = (off_t) (pstripe << DEV_BSHIFT) * (plex->subdisks - 1)
488 		    + i * sizeof(int);
489 		break;
490 	    }
491 	}
492 	bpp[psd]->b_flags |= B_INVAL;
493 	brelse(bpp[psd]);				    /* give back our resources */
494     }
495     /* release our resources */
496     Free(bpp);
497     if (error) {
498 	pbp->b_flags |= B_ERROR;
499 	pbp->b_error = error;
500     }
501     return pbp;
502 }
503 
504 /*
505  * Initialize a subdisk by writing zeroes to the
506  * complete address space.  If verify is set,
507  * check each transfer for correctness.
508  *
509  * Each call to this function writes (and maybe
510  * checks) a single block.
511  */
512 int
513 initsd(int sdno, int verify)
514 {
515     struct sd *sd;
516     struct plex *plex;
517     struct volume *vol;
518     struct buf *bp;
519     int error;
520     int size;						    /* size of init block, bytes */
521     daddr_t plexblkno;					    /* lblkno in plex */
522     int verified;					    /* set when we're happy with what we wrote */
523 
524     error = 0;
525     plexblkno = 0;					    /* to keep the compiler happy */
526     sd = &SD[sdno];
527     if (sd->plexno < 0)					    /* no plex? */
528 	return EINVAL;
529     plex = &PLEX[sd->plexno];				    /* point to plex */
530     if (plex->volno >= 0)
531 	vol = &VOL[plex->volno];
532     else
533 	vol = NULL;
534 
535     if (sd->init_blocksize == 0) {
536 	if (plex->stripesize != 0)			    /* we're striped, don't init more than */
537 	    sd->init_blocksize = min(DEFAULT_REVIVE_BLOCKSIZE, /* one block at a time */
538 		plex->stripesize << DEV_BSHIFT);
539 	else
540 	    sd->init_blocksize = DEFAULT_REVIVE_BLOCKSIZE;
541     } else if (sd->init_blocksize > MAX_REVIVE_BLOCKSIZE)
542 	sd->init_blocksize = MAX_REVIVE_BLOCKSIZE;
543 
544     size = min(sd->init_blocksize >> DEV_BSHIFT, sd->sectors - sd->initialized) << DEV_BSHIFT;
545 
546     verified = 0;
547     while (!verified) {					    /* until we're happy with it, */
548 	crit_enter();
549 	bp = geteblk(size);				    /* Get a buffer */
550 	crit_exit();
551 	if (bp == NULL)
552 	    return ENOMEM;
553 
554 	bp->b_bcount = size;
555 	bp->b_resid = bp->b_bcount;
556 	bp->b_bio1.bio_offset = (off_t)sd->initialized << DEV_BSHIFT;		    /* write it to here */
557 	bp->b_bio1.bio_driver_info = VINUM_SD(sdno);
558 	bzero(bp->b_data, bp->b_bcount);
559 	bp->b_cmd = BUF_CMD_WRITE;
560 	sdio(&bp->b_bio1);		    /* perform the I/O */
561 	biowait(bp);
562 	if (bp->b_flags & B_ERROR)
563 	    error = bp->b_error;
564 	if (bp->b_qindex == 0) {			    /* not on a queue, */
565 	    bp->b_flags |= B_INVAL;
566 	    bp->b_flags &= ~B_ERROR;
567 	    brelse(bp);					    /* is this kosher? */
568 	}
569 	if ((error == 0) && verify) {			    /* check that it got there */
570 	    crit_enter();
571 	    bp = geteblk(size);				    /* get a buffer */
572 	    if (bp == NULL) {
573 		crit_exit();
574 		error = ENOMEM;
575 	    } else {
576 		bp->b_bcount = size;
577 		bp->b_resid = bp->b_bcount;
578 		bp->b_bio1.bio_offset = (off_t)sd->initialized << DEV_BSHIFT;	    /* read from here */
579 		bp->b_bio1.bio_driver_info = VINUM_SD(sdno);
580 		bp->b_cmd = BUF_CMD_READ;		    /* read it back */
581 		crit_exit();
582 		sdio(&bp->b_bio1);
583 		biowait(bp);
584 		/*
585 		 * XXX Bug fix code.  This is hopefully no
586 		 * longer needed (21 February 2000).
587 		 */
588 		if (bp->b_flags & B_ERROR)
589 		    error = bp->b_error;
590 		else if ((*bp->b_data != 0)		    /* first word spammed */
591 		||(bcmp(bp->b_data, &bp->b_data[1], bp->b_bcount - 1))) { /* or one of the others */
592 		    printf("vinum: init error on %s, offset 0x%llx sectors\n",
593 			sd->name,
594 			(long long) sd->initialized);
595 		    verified = 0;
596 		} else
597 		    verified = 1;
598 		if (bp->b_qindex == 0) {		    /* not on a queue, */
599 		    bp->b_flags |= B_INVAL;
600 		    bp->b_flags &= ~B_ERROR;
601 		    brelse(bp);				    /* is this kosher? */
602 		}
603 	    }
604 	} else
605 	    verified = 1;
606     }
607     if (error == 0) {					    /* did it, */
608 	sd->initialized += size >> DEV_BSHIFT;		    /* moved this much further down */
609 	if (sd->initialized >= sd->sectors) {		    /* finished */
610 	    sd->initialized = 0;
611 	    set_sd_state(sdno, sd_initialized, setstate_force);	/* bring the sd up */
612 	    log(LOG_INFO, "vinum: %s is %s\n", sd->name, sd_state(sd->state));
613 	    save_config();				    /* and save the updated configuration */
614 	} else						    /* more to go, */
615 	    error = EAGAIN;				    /* ya'll come back, see? */
616     }
617     return error;
618 }
619 
620 /* Local Variables: */
621 /* fill-column: 50 */
622 /* End: */
623