2 * Copyright (c) 1997, 1998, 1999
3 * Nan Yang Computer Services Limited. All rights reserved.
5 * Parts copyright (c) 1997, 1998 Cybernet Corporation, NetMAX project.
7 * Written by Greg Lehey
9 * This software is distributed under the so-called ``Berkeley
12 * Redistribution and use in source and binary forms, with or without
13 * modification, are permitted provided that the following conditions
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
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.
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.
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 1.4 2003/11/15 21:05:42 dillon Exp $
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).
58 revive_block(int sdno)
60 int s; /* priority level */
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 */
74 plexblkno = 0; /* to keep the compiler happy */
77 if (sd->plexno < 0) /* no plex? */
79 plex = &PLEX[sd->plexno]; /* point to plex */
81 vol = &VOL[plex->volno];
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 */
93 /* Now decide where to read from */
94 switch (plex->organization) {
96 plexblkno = sd->revived + sd->plexoffset; /* corresponding address in plex */
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 */
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 */
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 */
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 */
125 * Now adjust for the strangenesses
126 * in RAID-4 and RAID-5 striping.
128 if (sd->plexsdno > psd) /* beyond the parity stripe, */
129 plexblkno -= plex->stripesize; /* one stripe less */
131 plexblkno -= plex->stripesize * sd->plexsdno; /* go back to the beginning of the band */
134 case plex_disorg: /* to keep the compiler happy */
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 */
144 bp = geteblk(size); /* Get a buffer */
150 * Amount to transfer: block size, unless it
151 * would overlap the end.
154 bp->b_resid = bp->b_bcount;
155 bp->b_blkno = plexblkno; /* 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, */
160 * First, read the data from the volume. We
161 * don't care which plex, that's bre's job.
163 bp->b_dev = VINUMDEV(plex->volno, 0, 0, VINUM_VOLUME_TYPE); /* create the device number */
164 else /* it's an unattached plex */
165 bp->b_dev = VINUM_PLEX(sd->plexno); /* create the device number */
167 bp->b_flags = B_READ; /* either way, read it */
172 if (bp->b_flags & B_ERROR)
175 /* Now write to the subdisk */
177 bp->b_dev = VINUM_SD(sdno); /* create the device number */
178 bp->b_flags = B_ORDERED | B_WRITE; /* and make this an ordered write */
179 bp->b_resid = bp->b_bcount;
180 bp->b_blkno = sd->revived; /* write it to here */
181 sdio(bp); /* perform the I/O */
183 if (bp->b_flags & B_ERROR)
186 sd->revived += bp->b_bcount >> DEV_BSHIFT; /* moved this much further down */
187 if (sd->revived >= sd->sectors) { /* finished */
189 set_sd_state(sdno, sd_up, setstate_force); /* bring the sd up */
190 log(LOG_INFO, "vinum: %s is %s\n", sd->name, sd_state(sd->state));
191 save_config(); /* and save the updated configuration */
192 error = 0; /* we're done */
195 if (lock) /* we took a lock, */
196 unlockrange(sd->plexno, lock); /* give it back */
197 while (sd->waitlist) { /* we have waiting requests */
199 struct request *rq = sd->waitlist;
201 if (debug & DEBUG_REVIVECONFLICT)
203 "Relaunch revive conflict sd %d: %p\n%s dev %d.%d, offset 0x%x, length %ld\n",
206 rq->bp->b_flags & B_READ ? "Read" : "Write",
207 major(rq->bp->b_dev),
208 minor(rq->bp->b_dev),
212 launch_requests(sd->waitlist, 1); /* do them now */
213 sd->waitlist = sd->waitlist->next; /* and move on to the next */
216 if (bp->b_qindex == 0) { /* not on a queue, */
217 bp->b_flags |= B_INVAL;
218 bp->b_flags &= ~B_ERROR;
219 brelse(bp); /* is this kosher? */
225 * Check or rebuild the parity blocks of a RAID-4
228 * The variables plex->checkblock and
229 * plex->rebuildblock represent the
230 * subdisk-relative address of the stripe we're
231 * looking at, not the plex-relative address. We
232 * store it in the plex and not as a local
233 * variable because this function could be
234 * stopped, and we don't want to repeat the part
235 * we've already done. This is also the reason
236 * why we don't initialize it here except at the
237 * end. It gets initialized with the plex on
240 * Each call to this function processes at most
241 * one stripe. We can't loop in this function,
242 * because we're unstoppable, so we have to be
243 * called repeatedly from userland.
246 parityops(struct vinum_ioctl_msg *data)
250 int size; /* I/O transfer size, bytes */
251 int stripe; /* stripe number in plex */
252 int psd; /* parity subdisk number */
253 struct rangelock *lock; /* lock on stripe */
254 struct _ioctl_reply *reply;
255 off_t pstripe; /* pointer to our stripe counter */
257 off_t errorloc; /* offset of parity error */
258 enum parityop op; /* operation to perform */
260 plexno = data->index;
263 reply = (struct _ioctl_reply *) data;
264 reply->error = EAGAIN; /* expect to repeat this call */
265 plex = &PLEX[plexno];
266 if (!isparity(plex)) { /* not RAID-4 or RAID-5 */
267 reply->error = EINVAL;
269 } else if (plex->state < plex_flaky) {
271 strcpy(reply->msg, "Plex is not completely accessible\n");
274 pstripe = data->offset;
275 stripe = pstripe / plex->stripesize; /* stripe number */
276 psd = plex->subdisks - 1 - stripe % plex->subdisks; /* parity subdisk for this stripe */
277 size = min(DEFAULT_REVIVE_BLOCKSIZE, /* one block at a time */
278 plex->stripesize << DEV_BSHIFT);
280 pbp = parityrebuild(plex, pstripe, size, op, &lock, &errorloc); /* do the grunt work */
281 if (pbp == NULL) { /* no buffer space */
282 reply->error = ENOMEM;
283 return; /* chicken out */
286 * Now we have a result in the data buffer of
287 * the parity buffer header, which we have kept.
288 * Decide what to do with it.
290 reply->msg[0] = '\0'; /* until shown otherwise */
291 if ((pbp->b_flags & B_ERROR) == 0) { /* no error */
292 if ((op == rebuildparity)
293 || (op == rebuildandcheckparity)) {
294 pbp->b_flags &= ~B_READ;
295 pbp->b_resid = pbp->b_bcount;
296 sdio(pbp); /* write the parity block */
299 if (((op == checkparity)
300 || (op == rebuildandcheckparity))
301 && (errorloc != -1)) {
302 if (op == checkparity)
305 "Parity incorrect at offset 0x%llx\n",
308 if (reply->error == EAGAIN) { /* still OK, */
309 plex->checkblock = pstripe + (pbp->b_bcount >> DEV_BSHIFT); /* moved this much further down */
310 if (plex->checkblock >= SD[plex->sdnos[0]].sectors) { /* finished */
311 plex->checkblock = 0;
316 if (pbp->b_flags & B_ERROR)
317 reply->error = pbp->b_error;
318 pbp->b_flags |= B_INVAL;
319 pbp->b_flags &= ~B_ERROR;
321 unlockrange(plexno, lock);
325 * Rebuild a parity stripe. Return pointer to
326 * parity bp. On return,
328 * 1. The band is locked. The caller must unlock
329 * the band and release the buffer header.
331 * 2. All buffer headers except php have been
332 * released. The caller must release pbp.
334 * 3. For checkparity and rebuildandcheckparity,
335 * the parity is compared with the current
336 * parity block. If it's different, the
337 * offset of the error is returned to
338 * errorloc. The caller can set the value of
339 * the pointer to NULL if this is called for
342 * pstripe is the subdisk-relative base address of
343 * the data to be reconstructed, size is the size
344 * of the transfer in bytes.
347 parityrebuild(struct plex *plex,
351 struct rangelock **lockp,
357 u_int64_t stripe; /* stripe number */
358 int *parity_buf; /* buffer address for current parity block */
359 int *newparity_buf; /* and for new parity block */
360 int mysize; /* I/O transfer size for this transfer */
361 int isize; /* mysize in ints */
363 int psd; /* parity subdisk number */
364 int newpsd; /* and "subdisk number" of new parity */
365 struct buf **bpp; /* pointers to our bps */
366 struct buf *pbp; /* buffer header for parity stripe */
368 int bufcount; /* number of buffers we need */
370 stripe = pstripe / plex->stripesize; /* stripe number */
371 psd = plex->subdisks - 1 - stripe % plex->subdisks; /* parity subdisk for this stripe */
372 parity_buf = NULL; /* to keep the compiler happy */
376 * It's possible that the default transfer size
377 * we chose is not a factor of the stripe size.
378 * We *must* limit this operation to a single
379 * stripe, at least for RAID-5 rebuild, since
380 * the parity subdisk changes between stripes,
381 * so in this case we need to perform a short
382 * transfer. Set variable mysize to reflect
385 mysize = min(size, (plex->stripesize * (stripe + 1) - pstripe) << DEV_BSHIFT);
386 isize = mysize / (sizeof(int)); /* number of ints in the buffer */
387 bufcount = plex->subdisks + 1; /* sd buffers plus result buffer */
388 newpsd = plex->subdisks;
389 bpp = (struct buf **) Malloc(bufcount * sizeof(struct buf *)); /* array of pointers to bps */
391 /* First, build requests for all subdisks */
392 for (sdno = 0; sdno < bufcount; sdno++) { /* for each subdisk */
393 if ((sdno != psd) || (op != rebuildparity)) {
394 /* Get a buffer header and initialize it. */
396 bpp[sdno] = geteblk(mysize); /* Get a buffer */
397 if (bpp[sdno] == NULL) {
398 while (sdno-- > 0) { /* release the ones we got */
399 bpp[sdno]->b_flags |= B_INVAL;
400 brelse(bpp[sdno]); /* give back our resources */
403 printf("vinum: can't allocate buffer space for parity op.\n");
404 return NULL; /* no bpps */
408 parity_buf = (int *) bpp[sdno]->b_data;
409 if (sdno == newpsd) /* the new one? */
410 bpp[sdno]->b_dev = VINUM_SD(plex->sdnos[psd]); /* write back to the parity SD */
412 bpp[sdno]->b_dev = VINUM_SD(plex->sdnos[sdno]); /* device number */
413 bpp[sdno]->b_flags = B_READ; /* either way, read it */
414 bpp[sdno]->b_bcount = mysize;
415 bpp[sdno]->b_resid = bpp[sdno]->b_bcount;
416 bpp[sdno]->b_blkno = pstripe; /* transfer from here */
420 /* Initialize result buffer */
422 newparity_buf = (int *) bpp[newpsd]->b_data;
423 bzero(newparity_buf, mysize);
426 * Now lock the stripe with the first non-parity
429 *lockp = lockrange(pstripe * plex->stripesize * (plex->subdisks - 1),
434 * Then issue requests for all subdisks in
435 * parallel. Don't transfer the parity stripe
436 * if we're rebuilding parity, unless we also
439 for (sdno = 0; sdno < plex->subdisks; sdno++) { /* for each real subdisk */
440 if ((sdno != psd) || (op != rebuildparity)) {
446 * Next, wait for the requests to complete.
447 * We wait in the order in which they were
448 * issued, which isn't necessarily the order in
449 * which they complete, but we don't have a
450 * convenient way of doing the latter, and the
453 for (sdno = 0; sdno < plex->subdisks; sdno++) { /* for each subdisk */
454 if ((sdno != psd) || (op != rebuildparity)) {
456 if (bpp[sdno]->b_flags & B_ERROR) /* can't read, */
457 error = bpp[sdno]->b_error;
458 else if (sdno != psd) { /* update parity */
459 sbuf = (int *) bpp[sdno]->b_data;
460 for (i = 0; i < isize; i++)
461 ((int *) newparity_buf)[i] ^= sbuf[i]; /* xor in the buffer */
464 if (sdno != psd) { /* release all bps except parity */
465 bpp[sdno]->b_flags |= B_INVAL;
466 brelse(bpp[sdno]); /* give back our resources */
471 * If we're checking, compare the calculated
472 * and the read parity block. If they're
473 * different, return the plex-relative offset;
474 * otherwise return -1.
476 if ((op == checkparity)
477 || (op == rebuildandcheckparity)) {
478 *errorloc = -1; /* no error yet */
479 for (i = 0; i < isize; i++) {
480 if (parity_buf[i] != newparity_buf[i]) {
481 *errorloc = (off_t) (pstripe << DEV_BSHIFT) * (plex->subdisks - 1)
486 bpp[psd]->b_flags |= B_INVAL;
487 brelse(bpp[psd]); /* give back our resources */
489 /* release our resources */
492 pbp->b_flags |= B_ERROR;
493 pbp->b_error = error;
499 * Initialize a subdisk by writing zeroes to the
500 * complete address space. If verify is set,
501 * check each transfer for correctness.
503 * Each call to this function writes (and maybe
504 * checks) a single block.
507 initsd(int sdno, int verify)
509 int s; /* priority level */
515 int size; /* size of init block, bytes */
516 daddr_t plexblkno; /* lblkno in plex */
517 int verified; /* set when we're happy with what we wrote */
520 plexblkno = 0; /* to keep the compiler happy */
522 if (sd->plexno < 0) /* no plex? */
524 plex = &PLEX[sd->plexno]; /* point to plex */
525 if (plex->volno >= 0)
526 vol = &VOL[plex->volno];
530 if (sd->init_blocksize == 0) {
531 if (plex->stripesize != 0) /* we're striped, don't init more than */
532 sd->init_blocksize = min(DEFAULT_REVIVE_BLOCKSIZE, /* one block at a time */
533 plex->stripesize << DEV_BSHIFT);
535 sd->init_blocksize = DEFAULT_REVIVE_BLOCKSIZE;
536 } else if (sd->init_blocksize > MAX_REVIVE_BLOCKSIZE)
537 sd->init_blocksize = MAX_REVIVE_BLOCKSIZE;
539 size = min(sd->init_blocksize >> DEV_BSHIFT, sd->sectors - sd->initialized) << DEV_BSHIFT;
542 while (!verified) { /* until we're happy with it, */
544 bp = geteblk(size); /* Get a buffer */
550 bp->b_resid = bp->b_bcount;
551 bp->b_blkno = sd->initialized; /* write it to here */
552 bzero(bp->b_data, bp->b_bcount);
553 bp->b_dev = VINUM_SD(sdno); /* create the device number */
554 bp->b_flags &= ~B_READ;
555 sdio(bp); /* perform the I/O */
557 if (bp->b_flags & B_ERROR)
559 if (bp->b_qindex == 0) { /* not on a queue, */
560 bp->b_flags |= B_INVAL;
561 bp->b_flags &= ~B_ERROR;
562 brelse(bp); /* is this kosher? */
564 if ((error == 0) && verify) { /* check that it got there */
566 bp = geteblk(size); /* get a buffer */
572 bp->b_resid = bp->b_bcount;
573 bp->b_blkno = sd->initialized; /* read from here */
574 bp->b_dev = VINUM_SD(sdno); /* create the device number */
575 bp->b_flags |= B_READ; /* read it back */
580 * XXX Bug fix code. This is hopefully no
581 * longer needed (21 February 2000).
583 if (bp->b_flags & B_ERROR)
585 else if ((*bp->b_data != 0) /* first word spammed */
586 ||(bcmp(bp->b_data, &bp->b_data[1], bp->b_bcount - 1))) { /* or one of the others */
587 printf("vinum: init error on %s, offset 0x%llx sectors\n",
589 (long long) sd->initialized);
593 if (bp->b_qindex == 0) { /* not on a queue, */
594 bp->b_flags |= B_INVAL;
595 bp->b_flags &= ~B_ERROR;
596 brelse(bp); /* is this kosher? */
602 if (error == 0) { /* did it, */
603 sd->initialized += size >> DEV_BSHIFT; /* moved this much further down */
604 if (sd->initialized >= sd->sectors) { /* finished */
606 set_sd_state(sdno, sd_initialized, setstate_force); /* bring the sd up */
607 log(LOG_INFO, "vinum: %s is %s\n", sd->name, sd_state(sd->state));
608 save_config(); /* and save the updated configuration */
609 } else /* more to go, */
610 error = EAGAIN; /* ya'll come back, see? */
615 /* Local Variables: */
616 /* fill-column: 50 */