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 1.15 2006/12/22 23:26:24 swildner 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)
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 */
143 bp = getpbuf(&vinum_conf.physbufs); /* Get a buffer */
144 bp->b_data = Malloc(size);
147 * Amount to transfer: block size, unless it
148 * would overlap the end.
151 bp->b_resid = bp->b_bcount;
152 bp->b_bio1.bio_offset = (off_t)plexblkno << DEV_BSHIFT; /* start here */
153 bp->b_bio1.bio_done = biodone_sync;
154 bp->b_bio1.bio_flags |= BIO_SYNC;
155 if (isstriped(plex)) /* we need to lock striped plexes */
156 lock = lockrange(plexblkno << DEV_BSHIFT, bp, plex); /* lock it */
157 if (vol != NULL) /* it's part of a volume, */
159 * First, read the data from the volume. We
160 * don't care which plex, that's bre's job.
162 dev = VINUMDEV(plex->volno, 0, 0, VINUM_VOLUME_TYPE); /* create the device number */
163 else /* it's an unattached plex */
164 dev = VINUM_PLEX(sd->plexno); /* create the device number */
166 bp->b_cmd = BUF_CMD_READ;
167 vinumstart(dev, &bp->b_bio1, 1);
168 biowait(&bp->b_bio1, "drvrd");
171 if (bp->b_flags & B_ERROR)
174 /* Now write to the subdisk */
176 dev = VINUM_SD(sdno); /* create the device number */
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 sdio(&bp->b_bio1); /* perform the I/O */
183 biowait(&bp->b_bio1, "drvwr");
184 if (bp->b_flags & B_ERROR)
187 sd->revived += bp->b_bcount >> DEV_BSHIFT; /* moved this much further down */
188 if (sd->revived >= sd->sectors) { /* finished */
190 set_sd_state(sdno, sd_up, setstate_force); /* bring the sd up */
191 log(LOG_INFO, "vinum: %s is %s\n", sd->name, sd_state(sd->state));
192 save_config(); /* and save the updated configuration */
193 error = 0; /* we're done */
196 if (lock) /* we took a lock, */
197 unlockrange(sd->plexno, lock); /* give it back */
198 while (sd->waitlist) { /* we have waiting requests */
200 struct request *rq = sd->waitlist;
203 if (debug & DEBUG_REVIVECONFLICT) {
204 dev = rq->bio->bio_driver_info;
206 "Relaunch revive conflict sd %d: %p\n%s dev %d.%d, offset 0x%llx, length %d\n",
209 (rq->bio->bio_buf->b_cmd == BUF_CMD_READ) ? "Read" : "Write",
213 rq->bio->bio_buf->b_bcount);
216 launch_requests(sd->waitlist, 1); /* do them now */
217 sd->waitlist = sd->waitlist->next; /* and move on to the next */
221 relpbuf(bp, &vinum_conf.physbufs);
226 * Check or rebuild the parity blocks of a RAID-4
229 * The variables plex->checkblock and
230 * plex->rebuildblock represent the
231 * subdisk-relative address of the stripe we're
232 * looking at, not the plex-relative address. We
233 * store it in the plex and not as a local
234 * variable because this function could be
235 * stopped, and we don't want to repeat the part
236 * we've already done. This is also the reason
237 * why we don't initialize it here except at the
238 * end. It gets initialized with the plex on
241 * Each call to this function processes at most
242 * one stripe. We can't loop in this function,
243 * because we're unstoppable, so we have to be
244 * called repeatedly from userland.
247 parityops(struct vinum_ioctl_msg *data)
251 int size; /* I/O transfer size, bytes */
252 int stripe; /* stripe number in plex */
253 int psd; /* parity subdisk number */
254 struct rangelock *lock; /* lock on stripe */
255 struct _ioctl_reply *reply;
256 off_t pstripe; /* pointer to our stripe counter */
258 off_t errorloc; /* offset of parity error */
259 enum parityop op; /* operation to perform */
261 plexno = data->index;
264 reply = (struct _ioctl_reply *) data;
265 reply->error = EAGAIN; /* expect to repeat this call */
266 plex = &PLEX[plexno];
267 if (!isparity(plex)) { /* not RAID-4 or RAID-5 */
268 reply->error = EINVAL;
270 } else if (plex->state < plex_flaky) {
272 strcpy(reply->msg, "Plex is not completely accessible\n");
275 pstripe = data->offset;
276 stripe = pstripe / plex->stripesize; /* stripe number */
277 psd = plex->subdisks - 1 - stripe % plex->subdisks; /* parity subdisk for this stripe */
278 size = min(DEFAULT_REVIVE_BLOCKSIZE, /* one block at a time */
279 plex->stripesize << DEV_BSHIFT);
281 pbp = parityrebuild(plex, pstripe, size, op, &lock, &errorloc); /* do the grunt work */
282 if (pbp == NULL) { /* no buffer space */
283 reply->error = ENOMEM;
284 return; /* chicken out */
287 * Now we have a result in the data buffer of
288 * the parity buffer header, which we have kept.
289 * Decide what to do with it.
291 reply->msg[0] = '\0'; /* until shown otherwise */
292 if ((pbp->b_flags & B_ERROR) == 0) { /* no error */
293 if ((op == rebuildparity)
294 || (op == rebuildandcheckparity)) {
295 pbp->b_cmd = BUF_CMD_WRITE;
296 pbp->b_resid = pbp->b_bcount;
297 sdio(&pbp->b_bio1); /* write the parity block */
298 biowait(&pbp->b_bio1, "drvwr");
300 if (((op == checkparity)
301 || (op == rebuildandcheckparity))
302 && (errorloc != -1)) {
303 if (op == checkparity)
306 "Parity incorrect at offset 0x%llx\n",
309 if (reply->error == EAGAIN) { /* still OK, */
310 plex->checkblock = pstripe + (pbp->b_bcount >> DEV_BSHIFT); /* moved this much further down */
311 if (plex->checkblock >= SD[plex->sdnos[0]].sectors) { /* finished */
312 plex->checkblock = 0;
317 if (pbp->b_flags & B_ERROR)
318 reply->error = pbp->b_error;
320 relpbuf(pbp, &vinum_conf.physbufs);
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,
356 u_int64_t stripe; /* stripe number */
357 int *parity_buf; /* buffer address for current parity block */
358 int *newparity_buf; /* and for new parity block */
359 int mysize; /* I/O transfer size for this transfer */
360 int isize; /* mysize in ints */
362 int psd; /* parity subdisk number */
363 int newpsd; /* and "subdisk number" of new parity */
364 struct buf **bpp; /* pointers to our bps */
365 struct buf *pbp; /* buffer header for parity stripe */
367 int bufcount; /* number of buffers we need */
369 stripe = pstripe / plex->stripesize; /* stripe number */
370 psd = plex->subdisks - 1 - stripe % plex->subdisks; /* parity subdisk for this stripe */
371 parity_buf = NULL; /* to keep the compiler happy */
375 * It's possible that the default transfer size
376 * we chose is not a factor of the stripe size.
377 * We *must* limit this operation to a single
378 * stripe, at least for RAID-5 rebuild, since
379 * the parity subdisk changes between stripes,
380 * so in this case we need to perform a short
381 * transfer. Set variable mysize to reflect
384 mysize = min(size, (plex->stripesize * (stripe + 1) - pstripe) << DEV_BSHIFT);
385 isize = mysize / (sizeof(int)); /* number of ints in the buffer */
386 bufcount = plex->subdisks + 1; /* sd buffers plus result buffer */
387 newpsd = plex->subdisks;
388 bpp = (struct buf **) Malloc(bufcount * sizeof(struct buf *)); /* array of pointers to bps */
390 /* First, build requests for all subdisks */
391 for (sdno = 0; sdno < bufcount; sdno++) { /* for each subdisk */
392 if ((sdno != psd) || (op != rebuildparity)) {
393 /* Get a buffer header and initialize it. */
394 bpp[sdno] = getpbuf(&vinum_conf.physbufs); /* Get a buffer */
395 bpp[sdno]->b_data = Malloc(mysize);
397 parity_buf = (int *) bpp[sdno]->b_data;
398 if (sdno == newpsd) /* the new one? */
399 bpp[sdno]->b_bio1.bio_driver_info = VINUM_SD(plex->sdnos[psd]); /* write back to the parity SD */
401 bpp[sdno]->b_bio1.bio_driver_info = VINUM_SD(plex->sdnos[sdno]); /* device number */
402 bpp[sdno]->b_cmd = BUF_CMD_READ; /* either way, read it */
403 bpp[sdno]->b_bcount = mysize;
404 bpp[sdno]->b_resid = bpp[sdno]->b_bcount;
405 bpp[sdno]->b_bio1.bio_offset = (off_t)pstripe << DEV_BSHIFT; /* transfer from here */
409 /* Initialize result buffer */
411 newparity_buf = (int *) bpp[newpsd]->b_data;
412 bzero(newparity_buf, mysize);
415 * Now lock the stripe with the first non-parity
418 *lockp = lockrange(pstripe * plex->stripesize * (plex->subdisks - 1),
423 * Then issue requests for all subdisks in
424 * parallel. Don't transfer the parity stripe
425 * if we're rebuilding parity, unless we also
428 for (sdno = 0; sdno < plex->subdisks; sdno++) { /* for each real subdisk */
429 if ((sdno != psd) || (op != rebuildparity)) {
430 sdio(&bpp[sdno]->b_bio1);
435 * Next, wait for the requests to complete.
436 * We wait in the order in which they were
437 * issued, which isn't necessarily the order in
438 * which they complete, but we don't have a
439 * convenient way of doing the latter, and the
442 for (sdno = 0; sdno < plex->subdisks; sdno++) { /* for each subdisk */
443 if ((sdno != psd) || (op != rebuildparity)) {
444 biowait(&bpp[sdno]->b_bio1, "drvio");
445 if (bpp[sdno]->b_flags & B_ERROR) /* can't read, */
446 error = bpp[sdno]->b_error;
447 else if (sdno != psd) { /* update parity */
448 sbuf = (int *) bpp[sdno]->b_data;
449 for (i = 0; i < isize; i++)
450 ((int *) newparity_buf)[i] ^= sbuf[i]; /* xor in the buffer */
453 if (sdno != psd) { /* release all bps except parity */
454 Free(bpp[sdno]->b_data);
455 relpbuf(bpp[sdno], &vinum_conf.physbufs); /* give back our resources */
460 * If we're checking, compare the calculated
461 * and the read parity block. If they're
462 * different, return the plex-relative offset;
463 * otherwise return -1.
465 if ((op == checkparity)
466 || (op == rebuildandcheckparity)) {
467 *errorloc = -1; /* no error yet */
468 for (i = 0; i < isize; i++) {
469 if (parity_buf[i] != newparity_buf[i]) {
470 *errorloc = (off_t) (pstripe << DEV_BSHIFT) * (plex->subdisks - 1)
475 Free(bpp[psd]->b_data);
476 relpbuf(bpp[psd], &vinum_conf.physbufs); /* give back our resources */
478 /* release our resources */
481 pbp->b_flags |= B_ERROR;
482 pbp->b_error = error;
488 * Initialize a subdisk by writing zeroes to the
489 * complete address space. If verify is set,
490 * check each transfer for correctness.
492 * Each call to this function writes (and maybe
493 * checks) a single block.
496 initsd(int sdno, int verify)
503 int size; /* size of init block, bytes */
504 daddr_t plexblkno; /* lblkno in plex */
505 int verified; /* set when we're happy with what we wrote */
508 plexblkno = 0; /* to keep the compiler happy */
510 if (sd->plexno < 0) /* no plex? */
512 plex = &PLEX[sd->plexno]; /* point to plex */
513 if (plex->volno >= 0)
514 vol = &VOL[plex->volno];
518 if (sd->init_blocksize == 0) {
519 if (plex->stripesize != 0) /* we're striped, don't init more than */
520 sd->init_blocksize = min(DEFAULT_REVIVE_BLOCKSIZE, /* one block at a time */
521 plex->stripesize << DEV_BSHIFT);
523 sd->init_blocksize = DEFAULT_REVIVE_BLOCKSIZE;
524 } else if (sd->init_blocksize > MAX_REVIVE_BLOCKSIZE)
525 sd->init_blocksize = MAX_REVIVE_BLOCKSIZE;
527 size = min(sd->init_blocksize >> DEV_BSHIFT, sd->sectors - sd->initialized) << DEV_BSHIFT;
529 bp = getpbuf(&vinum_conf.physbufs); /* Get a buffer */
530 bp->b_data = Malloc(size);
533 while (!verified) { /* until we're happy with it, */
535 bp->b_resid = bp->b_bcount;
536 bp->b_bio1.bio_offset = (off_t)sd->initialized << DEV_BSHIFT; /* write it to here */
537 bp->b_bio1.bio_driver_info = VINUM_SD(sdno);
538 bzero(bp->b_data, bp->b_bcount);
539 bp->b_cmd = BUF_CMD_WRITE;
540 sdio(&bp->b_bio1); /* perform the I/O */
541 biowait(&bp->b_bio1, "drvwr");
542 if (bp->b_flags & B_ERROR)
544 if ((error == 0) && verify) { /* check that it got there */
546 bp->b_resid = bp->b_bcount;
547 bp->b_bio1.bio_offset = (off_t)sd->initialized << DEV_BSHIFT; /* read from here */
548 bp->b_bio1.bio_driver_info = VINUM_SD(sdno);
549 bp->b_cmd = BUF_CMD_READ; /* read it back */
551 biowait(&bp->b_bio1, "drvrd");
553 * XXX Bug fix code. This is hopefully no
554 * longer needed (21 February 2000).
556 if (bp->b_flags & B_ERROR)
558 else if ((*bp->b_data != 0) /* first word spammed */
559 ||(bcmp(bp->b_data, &bp->b_data[1], bp->b_bcount - 1))) { /* or one of the others */
560 kprintf("vinum: init error on %s, offset 0x%llx sectors\n",
562 (long long) sd->initialized);
570 relpbuf(bp, &vinum_conf.physbufs);
571 if (error == 0) { /* did it, */
572 sd->initialized += size >> DEV_BSHIFT; /* moved this much further down */
573 if (sd->initialized >= sd->sectors) { /* finished */
575 set_sd_state(sdno, sd_initialized, setstate_force); /* bring the sd up */
576 log(LOG_INFO, "vinum: %s is %s\n", sd->name, sd_state(sd->state));
577 save_config(); /* and save the updated configuration */
578 } else /* more to go, */
579 error = EAGAIN; /* ya'll come back, see? */
584 /* Local Variables: */
585 /* fill-column: 50 */