rf_netbsdkintf.c revision 1.79.2.1 1 /* $NetBSD: rf_netbsdkintf.c,v 1.79.2.1 2000/06/22 17:07:55 minoura Exp $ */
2 /*-
3 * Copyright (c) 1996, 1997, 1998 The NetBSD Foundation, Inc.
4 * All rights reserved.
5 *
6 * This code is derived from software contributed to The NetBSD Foundation
7 * by Greg Oster; Jason R. Thorpe.
8 *
9 * Redistribution and use in source and binary forms, with or without
10 * modification, are permitted provided that the following conditions
11 * are met:
12 * 1. Redistributions of source code must retain the above copyright
13 * notice, this list of conditions and the following disclaimer.
14 * 2. Redistributions in binary form must reproduce the above copyright
15 * notice, this list of conditions and the following disclaimer in the
16 * documentation and/or other materials provided with the distribution.
17 * 3. All advertising materials mentioning features or use of this software
18 * must display the following acknowledgement:
19 * This product includes software developed by the NetBSD
20 * Foundation, Inc. and its contributors.
21 * 4. Neither the name of The NetBSD Foundation nor the names of its
22 * contributors may be used to endorse or promote products derived
23 * from this software without specific prior written permission.
24 *
25 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
26 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
27 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
28 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
29 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
30 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
31 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
32 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
33 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
34 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
35 * POSSIBILITY OF SUCH DAMAGE.
36 */
37
38 /*
39 * Copyright (c) 1988 University of Utah.
40 * Copyright (c) 1990, 1993
41 * The Regents of the University of California. All rights reserved.
42 *
43 * This code is derived from software contributed to Berkeley by
44 * the Systems Programming Group of the University of Utah Computer
45 * Science Department.
46 *
47 * Redistribution and use in source and binary forms, with or without
48 * modification, are permitted provided that the following conditions
49 * are met:
50 * 1. Redistributions of source code must retain the above copyright
51 * notice, this list of conditions and the following disclaimer.
52 * 2. Redistributions in binary form must reproduce the above copyright
53 * notice, this list of conditions and the following disclaimer in the
54 * documentation and/or other materials provided with the distribution.
55 * 3. All advertising materials mentioning features or use of this software
56 * must display the following acknowledgement:
57 * This product includes software developed by the University of
58 * California, Berkeley and its contributors.
59 * 4. Neither the name of the University nor the names of its contributors
60 * may be used to endorse or promote products derived from this software
61 * without specific prior written permission.
62 *
63 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
64 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
65 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
66 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
67 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
68 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
69 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
70 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
71 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
72 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
73 * SUCH DAMAGE.
74 *
75 * from: Utah $Hdr: cd.c 1.6 90/11/28$
76 *
77 * @(#)cd.c 8.2 (Berkeley) 11/16/93
78 */
79
80
81
82
83 /*
84 * Copyright (c) 1995 Carnegie-Mellon University.
85 * All rights reserved.
86 *
87 * Authors: Mark Holland, Jim Zelenka
88 *
89 * Permission to use, copy, modify and distribute this software and
90 * its documentation is hereby granted, provided that both the copyright
91 * notice and this permission notice appear in all copies of the
92 * software, derivative works or modified versions, and any portions
93 * thereof, and that both notices appear in supporting documentation.
94 *
95 * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
96 * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
97 * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
98 *
99 * Carnegie Mellon requests users of this software to return to
100 *
101 * Software Distribution Coordinator or Software.Distribution (at) CS.CMU.EDU
102 * School of Computer Science
103 * Carnegie Mellon University
104 * Pittsburgh PA 15213-3890
105 *
106 * any improvements or extensions that they make and grant Carnegie the
107 * rights to redistribute these changes.
108 */
109
110 /***********************************************************
111 *
112 * rf_kintf.c -- the kernel interface routines for RAIDframe
113 *
114 ***********************************************************/
115
116 #include <sys/errno.h>
117 #include <sys/param.h>
118 #include <sys/pool.h>
119 #include <sys/queue.h>
120 #include <sys/disk.h>
121 #include <sys/device.h>
122 #include <sys/stat.h>
123 #include <sys/ioctl.h>
124 #include <sys/fcntl.h>
125 #include <sys/systm.h>
126 #include <sys/namei.h>
127 #include <sys/vnode.h>
128 #include <sys/param.h>
129 #include <sys/types.h>
130 #include <machine/types.h>
131 #include <sys/disklabel.h>
132 #include <sys/conf.h>
133 #include <sys/lock.h>
134 #include <sys/buf.h>
135 #include <sys/user.h>
136 #include <sys/reboot.h>
137
138 #include "raid.h"
139 #include "opt_raid_autoconfig.h"
140 #include "rf_raid.h"
141 #include "rf_raidframe.h"
142 #include "rf_copyback.h"
143 #include "rf_dag.h"
144 #include "rf_dagflags.h"
145 #include "rf_diskqueue.h"
146 #include "rf_acctrace.h"
147 #include "rf_etimer.h"
148 #include "rf_general.h"
149 #include "rf_debugMem.h"
150 #include "rf_kintf.h"
151 #include "rf_options.h"
152 #include "rf_driver.h"
153 #include "rf_parityscan.h"
154 #include "rf_debugprint.h"
155 #include "rf_threadstuff.h"
156 #include "rf_configure.h"
157
158 int rf_kdebug_level = 0;
159
160 #ifdef DEBUG
161 #define db1_printf(a) if (rf_kdebug_level > 0) printf a
162 #else /* DEBUG */
163 #define db1_printf(a) { }
164 #endif /* DEBUG */
165
166 static RF_Raid_t **raidPtrs; /* global raid device descriptors */
167
168 RF_DECLARE_STATIC_MUTEX(rf_sparet_wait_mutex)
169
170 static RF_SparetWait_t *rf_sparet_wait_queue; /* requests to install a
171 * spare table */
172 static RF_SparetWait_t *rf_sparet_resp_queue; /* responses from
173 * installation process */
174
175 /* prototypes */
176 static void KernelWakeupFunc(struct buf * bp);
177 static void InitBP(struct buf * bp, struct vnode *, unsigned rw_flag,
178 dev_t dev, RF_SectorNum_t startSect,
179 RF_SectorCount_t numSect, caddr_t buf,
180 void (*cbFunc) (struct buf *), void *cbArg,
181 int logBytesPerSector, struct proc * b_proc);
182 static void raidinit __P((RF_Raid_t *));
183
184 void raidattach __P((int));
185 int raidsize __P((dev_t));
186 int raidopen __P((dev_t, int, int, struct proc *));
187 int raidclose __P((dev_t, int, int, struct proc *));
188 int raidioctl __P((dev_t, u_long, caddr_t, int, struct proc *));
189 int raidwrite __P((dev_t, struct uio *, int));
190 int raidread __P((dev_t, struct uio *, int));
191 void raidstrategy __P((struct buf *));
192 int raiddump __P((dev_t, daddr_t, caddr_t, size_t));
193
194 /*
195 * Pilfered from ccd.c
196 */
197
198 struct raidbuf {
199 struct buf rf_buf; /* new I/O buf. MUST BE FIRST!!! */
200 struct buf *rf_obp; /* ptr. to original I/O buf */
201 int rf_flags; /* misc. flags */
202 RF_DiskQueueData_t *req;/* the request that this was part of.. */
203 };
204
205
206 #define RAIDGETBUF(rs) pool_get(&(rs)->sc_cbufpool, PR_NOWAIT)
207 #define RAIDPUTBUF(rs, cbp) pool_put(&(rs)->sc_cbufpool, cbp)
208
209 /* XXX Not sure if the following should be replacing the raidPtrs above,
210 or if it should be used in conjunction with that...
211 */
212
213 struct raid_softc {
214 int sc_flags; /* flags */
215 int sc_cflags; /* configuration flags */
216 size_t sc_size; /* size of the raid device */
217 char sc_xname[20]; /* XXX external name */
218 struct disk sc_dkdev; /* generic disk device info */
219 struct pool sc_cbufpool; /* component buffer pool */
220 struct buf_queue buf_queue; /* used for the device queue */
221 };
222 /* sc_flags */
223 #define RAIDF_INITED 0x01 /* unit has been initialized */
224 #define RAIDF_WLABEL 0x02 /* label area is writable */
225 #define RAIDF_LABELLING 0x04 /* unit is currently being labelled */
226 #define RAIDF_WANTED 0x40 /* someone is waiting to obtain a lock */
227 #define RAIDF_LOCKED 0x80 /* unit is locked */
228
229 #define raidunit(x) DISKUNIT(x)
230 int numraid = 0;
231
232 /*
233 * Allow RAIDOUTSTANDING number of simultaneous IO's to this RAID device.
234 * Be aware that large numbers can allow the driver to consume a lot of
235 * kernel memory, especially on writes, and in degraded mode reads.
236 *
237 * For example: with a stripe width of 64 blocks (32k) and 5 disks,
238 * a single 64K write will typically require 64K for the old data,
239 * 64K for the old parity, and 64K for the new parity, for a total
240 * of 192K (if the parity buffer is not re-used immediately).
241 * Even it if is used immedately, that's still 128K, which when multiplied
242 * by say 10 requests, is 1280K, *on top* of the 640K of incoming data.
243 *
244 * Now in degraded mode, for example, a 64K read on the above setup may
245 * require data reconstruction, which will require *all* of the 4 remaining
246 * disks to participate -- 4 * 32K/disk == 128K again.
247 */
248
249 #ifndef RAIDOUTSTANDING
250 #define RAIDOUTSTANDING 6
251 #endif
252
253 #define RAIDLABELDEV(dev) \
254 (MAKEDISKDEV(major((dev)), raidunit((dev)), RAW_PART))
255
256 /* declared here, and made public, for the benefit of KVM stuff.. */
257 struct raid_softc *raid_softc;
258
259 static void raidgetdefaultlabel __P((RF_Raid_t *, struct raid_softc *,
260 struct disklabel *));
261 static void raidgetdisklabel __P((dev_t));
262 static void raidmakedisklabel __P((struct raid_softc *));
263
264 static int raidlock __P((struct raid_softc *));
265 static void raidunlock __P((struct raid_softc *));
266
267 static void rf_markalldirty __P((RF_Raid_t *));
268 void rf_mountroot_hook __P((struct device *));
269
270 struct device *raidrootdev;
271
272 void rf_ReconThread __P((struct rf_recon_req *));
273 /* XXX what I want is: */
274 /*void rf_ReconThread __P((RF_Raid_t *raidPtr)); */
275 void rf_RewriteParityThread __P((RF_Raid_t *raidPtr));
276 void rf_CopybackThread __P((RF_Raid_t *raidPtr));
277 void rf_ReconstructInPlaceThread __P((struct rf_recon_req *));
278 void rf_buildroothack __P((void *));
279
280 RF_AutoConfig_t *rf_find_raid_components __P((void));
281 RF_ConfigSet_t *rf_create_auto_sets __P((RF_AutoConfig_t *));
282 static int rf_does_it_fit __P((RF_ConfigSet_t *,RF_AutoConfig_t *));
283 static int rf_reasonable_label __P((RF_ComponentLabel_t *));
284 void rf_create_configuration __P((RF_AutoConfig_t *,RF_Config_t *,
285 RF_Raid_t *));
286 int rf_set_autoconfig __P((RF_Raid_t *, int));
287 int rf_set_rootpartition __P((RF_Raid_t *, int));
288 void rf_release_all_vps __P((RF_ConfigSet_t *));
289 void rf_cleanup_config_set __P((RF_ConfigSet_t *));
290 int rf_have_enough_components __P((RF_ConfigSet_t *));
291 int rf_auto_config_set __P((RF_ConfigSet_t *, int *));
292
293 static int raidautoconfig = 0; /* Debugging, mostly. Set to 0 to not
294 allow autoconfig to take place.
295 Note that this is overridden by having
296 RAID_AUTOCONFIG as an option in the
297 kernel config file. */
298
299 void
300 raidattach(num)
301 int num;
302 {
303 int raidID;
304 int i, rc;
305 RF_AutoConfig_t *ac_list; /* autoconfig list */
306 RF_ConfigSet_t *config_sets;
307
308 #ifdef DEBUG
309 printf("raidattach: Asked for %d units\n", num);
310 #endif
311
312 if (num <= 0) {
313 #ifdef DIAGNOSTIC
314 panic("raidattach: count <= 0");
315 #endif
316 return;
317 }
318 /* This is where all the initialization stuff gets done. */
319
320 numraid = num;
321
322 /* Make some space for requested number of units... */
323
324 RF_Calloc(raidPtrs, num, sizeof(RF_Raid_t *), (RF_Raid_t **));
325 if (raidPtrs == NULL) {
326 panic("raidPtrs is NULL!!\n");
327 }
328
329 rc = rf_mutex_init(&rf_sparet_wait_mutex);
330 if (rc) {
331 RF_PANIC();
332 }
333
334 rf_sparet_wait_queue = rf_sparet_resp_queue = NULL;
335
336 for (i = 0; i < num; i++)
337 raidPtrs[i] = NULL;
338 rc = rf_BootRaidframe();
339 if (rc == 0)
340 printf("Kernelized RAIDframe activated\n");
341 else
342 panic("Serious error booting RAID!!\n");
343
344 /* put together some datastructures like the CCD device does.. This
345 * lets us lock the device and what-not when it gets opened. */
346
347 raid_softc = (struct raid_softc *)
348 malloc(num * sizeof(struct raid_softc),
349 M_RAIDFRAME, M_NOWAIT);
350 if (raid_softc == NULL) {
351 printf("WARNING: no memory for RAIDframe driver\n");
352 return;
353 }
354
355 bzero(raid_softc, num * sizeof(struct raid_softc));
356
357 raidrootdev = (struct device *)malloc(num * sizeof(struct device),
358 M_RAIDFRAME, M_NOWAIT);
359 if (raidrootdev == NULL) {
360 panic("No memory for RAIDframe driver!!?!?!\n");
361 }
362
363 for (raidID = 0; raidID < num; raidID++) {
364 BUFQ_INIT(&raid_softc[raidID].buf_queue);
365
366 raidrootdev[raidID].dv_class = DV_DISK;
367 raidrootdev[raidID].dv_cfdata = NULL;
368 raidrootdev[raidID].dv_unit = raidID;
369 raidrootdev[raidID].dv_parent = NULL;
370 raidrootdev[raidID].dv_flags = 0;
371 sprintf(raidrootdev[raidID].dv_xname,"raid%d",raidID);
372
373 RF_Calloc(raidPtrs[raidID], 1, sizeof(RF_Raid_t),
374 (RF_Raid_t *));
375 if (raidPtrs[raidID] == NULL) {
376 printf("WARNING: raidPtrs[%d] is NULL\n", raidID);
377 numraid = raidID;
378 return;
379 }
380 }
381
382 #if RAID_AUTOCONFIG
383 raidautoconfig = 1;
384 #endif
385
386 if (raidautoconfig) {
387 /* 1. locate all RAID components on the system */
388
389 #if DEBUG
390 printf("Searching for raid components...\n");
391 #endif
392 ac_list = rf_find_raid_components();
393
394 /* 2. sort them into their respective sets */
395
396 config_sets = rf_create_auto_sets(ac_list);
397
398 /* 3. evaluate each set and configure the valid ones
399 This gets done in rf_buildroothack() */
400
401 /* schedule the creation of the thread to do the
402 "/ on RAID" stuff */
403
404 kthread_create(rf_buildroothack,config_sets);
405
406 #if 0
407 mountroothook_establish(rf_mountroot_hook, &raidrootdev[0]);
408 #endif
409 }
410
411 }
412
413 void
414 rf_buildroothack(arg)
415 void *arg;
416 {
417 RF_ConfigSet_t *config_sets = arg;
418 RF_ConfigSet_t *cset;
419 RF_ConfigSet_t *next_cset;
420 int retcode;
421 int raidID;
422 int rootID;
423 int num_root;
424
425 num_root = 0;
426 cset = config_sets;
427 while(cset != NULL ) {
428 next_cset = cset->next;
429 if (rf_have_enough_components(cset) &&
430 cset->ac->clabel->autoconfigure==1) {
431 retcode = rf_auto_config_set(cset,&raidID);
432 if (!retcode) {
433 if (cset->rootable) {
434 rootID = raidID;
435 num_root++;
436 }
437 } else {
438 /* The autoconfig didn't work :( */
439 #if DEBUG
440 printf("Autoconfig failed with code %d for raid%d\n", retcode, raidID);
441 #endif
442 rf_release_all_vps(cset);
443 }
444 } else {
445 /* we're not autoconfiguring this set...
446 release the associated resources */
447 rf_release_all_vps(cset);
448 }
449 /* cleanup */
450 rf_cleanup_config_set(cset);
451 cset = next_cset;
452 }
453 if (boothowto & RB_ASKNAME) {
454 /* We don't auto-config... */
455 } else {
456 /* They didn't ask, and we found something bootable... */
457
458 if (num_root == 1) {
459 booted_device = &raidrootdev[rootID];
460 } else if (num_root > 1) {
461 /* we can't guess.. require the user to answer... */
462 boothowto |= RB_ASKNAME;
463 }
464 }
465 }
466
467
468 int
469 raidsize(dev)
470 dev_t dev;
471 {
472 struct raid_softc *rs;
473 struct disklabel *lp;
474 int part, unit, omask, size;
475
476 unit = raidunit(dev);
477 if (unit >= numraid)
478 return (-1);
479 rs = &raid_softc[unit];
480
481 if ((rs->sc_flags & RAIDF_INITED) == 0)
482 return (-1);
483
484 part = DISKPART(dev);
485 omask = rs->sc_dkdev.dk_openmask & (1 << part);
486 lp = rs->sc_dkdev.dk_label;
487
488 if (omask == 0 && raidopen(dev, 0, S_IFBLK, curproc))
489 return (-1);
490
491 if (lp->d_partitions[part].p_fstype != FS_SWAP)
492 size = -1;
493 else
494 size = lp->d_partitions[part].p_size *
495 (lp->d_secsize / DEV_BSIZE);
496
497 if (omask == 0 && raidclose(dev, 0, S_IFBLK, curproc))
498 return (-1);
499
500 return (size);
501
502 }
503
504 int
505 raiddump(dev, blkno, va, size)
506 dev_t dev;
507 daddr_t blkno;
508 caddr_t va;
509 size_t size;
510 {
511 /* Not implemented. */
512 return ENXIO;
513 }
514 /* ARGSUSED */
515 int
516 raidopen(dev, flags, fmt, p)
517 dev_t dev;
518 int flags, fmt;
519 struct proc *p;
520 {
521 int unit = raidunit(dev);
522 struct raid_softc *rs;
523 struct disklabel *lp;
524 int part, pmask;
525 int error = 0;
526
527 if (unit >= numraid)
528 return (ENXIO);
529 rs = &raid_softc[unit];
530
531 if ((error = raidlock(rs)) != 0)
532 return (error);
533 lp = rs->sc_dkdev.dk_label;
534
535 part = DISKPART(dev);
536 pmask = (1 << part);
537
538 db1_printf(("Opening raid device number: %d partition: %d\n",
539 unit, part));
540
541
542 if ((rs->sc_flags & RAIDF_INITED) &&
543 (rs->sc_dkdev.dk_openmask == 0))
544 raidgetdisklabel(dev);
545
546 /* make sure that this partition exists */
547
548 if (part != RAW_PART) {
549 db1_printf(("Not a raw partition..\n"));
550 if (((rs->sc_flags & RAIDF_INITED) == 0) ||
551 ((part >= lp->d_npartitions) ||
552 (lp->d_partitions[part].p_fstype == FS_UNUSED))) {
553 error = ENXIO;
554 raidunlock(rs);
555 db1_printf(("Bailing out...\n"));
556 return (error);
557 }
558 }
559 /* Prevent this unit from being unconfigured while open. */
560 switch (fmt) {
561 case S_IFCHR:
562 rs->sc_dkdev.dk_copenmask |= pmask;
563 break;
564
565 case S_IFBLK:
566 rs->sc_dkdev.dk_bopenmask |= pmask;
567 break;
568 }
569
570 if ((rs->sc_dkdev.dk_openmask == 0) &&
571 ((rs->sc_flags & RAIDF_INITED) != 0)) {
572 /* First one... mark things as dirty... Note that we *MUST*
573 have done a configure before this. I DO NOT WANT TO BE
574 SCRIBBLING TO RANDOM COMPONENTS UNTIL IT'S BEEN DETERMINED
575 THAT THEY BELONG TOGETHER!!!!! */
576 /* XXX should check to see if we're only open for reading
577 here... If so, we needn't do this, but then need some
578 other way of keeping track of what's happened.. */
579
580 rf_markalldirty( raidPtrs[unit] );
581 }
582
583
584 rs->sc_dkdev.dk_openmask =
585 rs->sc_dkdev.dk_copenmask | rs->sc_dkdev.dk_bopenmask;
586
587 raidunlock(rs);
588
589 return (error);
590
591
592 }
593 /* ARGSUSED */
594 int
595 raidclose(dev, flags, fmt, p)
596 dev_t dev;
597 int flags, fmt;
598 struct proc *p;
599 {
600 int unit = raidunit(dev);
601 struct raid_softc *rs;
602 int error = 0;
603 int part;
604
605 if (unit >= numraid)
606 return (ENXIO);
607 rs = &raid_softc[unit];
608
609 if ((error = raidlock(rs)) != 0)
610 return (error);
611
612 part = DISKPART(dev);
613
614 /* ...that much closer to allowing unconfiguration... */
615 switch (fmt) {
616 case S_IFCHR:
617 rs->sc_dkdev.dk_copenmask &= ~(1 << part);
618 break;
619
620 case S_IFBLK:
621 rs->sc_dkdev.dk_bopenmask &= ~(1 << part);
622 break;
623 }
624 rs->sc_dkdev.dk_openmask =
625 rs->sc_dkdev.dk_copenmask | rs->sc_dkdev.dk_bopenmask;
626
627 if ((rs->sc_dkdev.dk_openmask == 0) &&
628 ((rs->sc_flags & RAIDF_INITED) != 0)) {
629 /* Last one... device is not unconfigured yet.
630 Device shutdown has taken care of setting the
631 clean bits if RAIDF_INITED is not set
632 mark things as clean... */
633 #if 0
634 printf("Last one on raid%d. Updating status.\n",unit);
635 #endif
636 rf_update_component_labels(raidPtrs[unit],
637 RF_FINAL_COMPONENT_UPDATE);
638 }
639
640 raidunlock(rs);
641 return (0);
642
643 }
644
645 void
646 raidstrategy(bp)
647 struct buf *bp;
648 {
649 int s;
650
651 unsigned int raidID = raidunit(bp->b_dev);
652 RF_Raid_t *raidPtr;
653 struct raid_softc *rs = &raid_softc[raidID];
654 struct disklabel *lp;
655 int wlabel;
656
657 if ((rs->sc_flags & RAIDF_INITED) ==0) {
658 bp->b_error = ENXIO;
659 bp->b_flags = B_ERROR;
660 bp->b_resid = bp->b_bcount;
661 biodone(bp);
662 return;
663 }
664 if (raidID >= numraid || !raidPtrs[raidID]) {
665 bp->b_error = ENODEV;
666 bp->b_flags |= B_ERROR;
667 bp->b_resid = bp->b_bcount;
668 biodone(bp);
669 return;
670 }
671 raidPtr = raidPtrs[raidID];
672 if (!raidPtr->valid) {
673 bp->b_error = ENODEV;
674 bp->b_flags |= B_ERROR;
675 bp->b_resid = bp->b_bcount;
676 biodone(bp);
677 return;
678 }
679 if (bp->b_bcount == 0) {
680 db1_printf(("b_bcount is zero..\n"));
681 biodone(bp);
682 return;
683 }
684 lp = rs->sc_dkdev.dk_label;
685
686 /*
687 * Do bounds checking and adjust transfer. If there's an
688 * error, the bounds check will flag that for us.
689 */
690
691 wlabel = rs->sc_flags & (RAIDF_WLABEL | RAIDF_LABELLING);
692 if (DISKPART(bp->b_dev) != RAW_PART)
693 if (bounds_check_with_label(bp, lp, wlabel) <= 0) {
694 db1_printf(("Bounds check failed!!:%d %d\n",
695 (int) bp->b_blkno, (int) wlabel));
696 biodone(bp);
697 return;
698 }
699 s = splbio();
700
701 bp->b_resid = 0;
702
703 /* stuff it onto our queue */
704 BUFQ_INSERT_TAIL(&rs->buf_queue, bp);
705
706 raidstart(raidPtrs[raidID]);
707
708 splx(s);
709 }
710 /* ARGSUSED */
711 int
712 raidread(dev, uio, flags)
713 dev_t dev;
714 struct uio *uio;
715 int flags;
716 {
717 int unit = raidunit(dev);
718 struct raid_softc *rs;
719 int part;
720
721 if (unit >= numraid)
722 return (ENXIO);
723 rs = &raid_softc[unit];
724
725 if ((rs->sc_flags & RAIDF_INITED) == 0)
726 return (ENXIO);
727 part = DISKPART(dev);
728
729 db1_printf(("raidread: unit: %d partition: %d\n", unit, part));
730
731 return (physio(raidstrategy, NULL, dev, B_READ, minphys, uio));
732
733 }
734 /* ARGSUSED */
735 int
736 raidwrite(dev, uio, flags)
737 dev_t dev;
738 struct uio *uio;
739 int flags;
740 {
741 int unit = raidunit(dev);
742 struct raid_softc *rs;
743
744 if (unit >= numraid)
745 return (ENXIO);
746 rs = &raid_softc[unit];
747
748 if ((rs->sc_flags & RAIDF_INITED) == 0)
749 return (ENXIO);
750 db1_printf(("raidwrite\n"));
751 return (physio(raidstrategy, NULL, dev, B_WRITE, minphys, uio));
752
753 }
754
755 int
756 raidioctl(dev, cmd, data, flag, p)
757 dev_t dev;
758 u_long cmd;
759 caddr_t data;
760 int flag;
761 struct proc *p;
762 {
763 int unit = raidunit(dev);
764 int error = 0;
765 int part, pmask;
766 struct raid_softc *rs;
767 RF_Config_t *k_cfg, *u_cfg;
768 RF_Raid_t *raidPtr;
769 RF_RaidDisk_t *diskPtr;
770 RF_AccTotals_t *totals;
771 RF_DeviceConfig_t *d_cfg, **ucfgp;
772 u_char *specific_buf;
773 int retcode = 0;
774 int row;
775 int column;
776 struct rf_recon_req *rrcopy, *rr;
777 RF_ComponentLabel_t *clabel;
778 RF_ComponentLabel_t ci_label;
779 RF_ComponentLabel_t **clabel_ptr;
780 RF_SingleComponent_t *sparePtr,*componentPtr;
781 RF_SingleComponent_t hot_spare;
782 RF_SingleComponent_t component;
783 RF_ProgressInfo_t progressInfo, **progressInfoPtr;
784 int i, j, d;
785
786 if (unit >= numraid)
787 return (ENXIO);
788 rs = &raid_softc[unit];
789 raidPtr = raidPtrs[unit];
790
791 db1_printf(("raidioctl: %d %d %d %d\n", (int) dev,
792 (int) DISKPART(dev), (int) unit, (int) cmd));
793
794 /* Must be open for writes for these commands... */
795 switch (cmd) {
796 case DIOCSDINFO:
797 case DIOCWDINFO:
798 case DIOCWLABEL:
799 if ((flag & FWRITE) == 0)
800 return (EBADF);
801 }
802
803 /* Must be initialized for these... */
804 switch (cmd) {
805 case DIOCGDINFO:
806 case DIOCSDINFO:
807 case DIOCWDINFO:
808 case DIOCGPART:
809 case DIOCWLABEL:
810 case DIOCGDEFLABEL:
811 case RAIDFRAME_SHUTDOWN:
812 case RAIDFRAME_REWRITEPARITY:
813 case RAIDFRAME_GET_INFO:
814 case RAIDFRAME_RESET_ACCTOTALS:
815 case RAIDFRAME_GET_ACCTOTALS:
816 case RAIDFRAME_KEEP_ACCTOTALS:
817 case RAIDFRAME_GET_SIZE:
818 case RAIDFRAME_FAIL_DISK:
819 case RAIDFRAME_COPYBACK:
820 case RAIDFRAME_CHECK_RECON_STATUS:
821 case RAIDFRAME_CHECK_RECON_STATUS_EXT:
822 case RAIDFRAME_GET_COMPONENT_LABEL:
823 case RAIDFRAME_SET_COMPONENT_LABEL:
824 case RAIDFRAME_ADD_HOT_SPARE:
825 case RAIDFRAME_REMOVE_HOT_SPARE:
826 case RAIDFRAME_INIT_LABELS:
827 case RAIDFRAME_REBUILD_IN_PLACE:
828 case RAIDFRAME_CHECK_PARITY:
829 case RAIDFRAME_CHECK_PARITYREWRITE_STATUS:
830 case RAIDFRAME_CHECK_PARITYREWRITE_STATUS_EXT:
831 case RAIDFRAME_CHECK_COPYBACK_STATUS:
832 case RAIDFRAME_CHECK_COPYBACK_STATUS_EXT:
833 case RAIDFRAME_SET_AUTOCONFIG:
834 case RAIDFRAME_SET_ROOT:
835 case RAIDFRAME_DELETE_COMPONENT:
836 case RAIDFRAME_INCORPORATE_HOT_SPARE:
837 if ((rs->sc_flags & RAIDF_INITED) == 0)
838 return (ENXIO);
839 }
840
841 switch (cmd) {
842
843 /* configure the system */
844 case RAIDFRAME_CONFIGURE:
845
846 if (raidPtr->valid) {
847 /* There is a valid RAID set running on this unit! */
848 printf("raid%d: Device already configured!\n",unit);
849 return(EINVAL);
850 }
851
852 /* copy-in the configuration information */
853 /* data points to a pointer to the configuration structure */
854
855 u_cfg = *((RF_Config_t **) data);
856 RF_Malloc(k_cfg, sizeof(RF_Config_t), (RF_Config_t *));
857 if (k_cfg == NULL) {
858 return (ENOMEM);
859 }
860 retcode = copyin((caddr_t) u_cfg, (caddr_t) k_cfg,
861 sizeof(RF_Config_t));
862 if (retcode) {
863 RF_Free(k_cfg, sizeof(RF_Config_t));
864 db1_printf(("rf_ioctl: retcode=%d copyin.1\n",
865 retcode));
866 return (retcode);
867 }
868 /* allocate a buffer for the layout-specific data, and copy it
869 * in */
870 if (k_cfg->layoutSpecificSize) {
871 if (k_cfg->layoutSpecificSize > 10000) {
872 /* sanity check */
873 RF_Free(k_cfg, sizeof(RF_Config_t));
874 return (EINVAL);
875 }
876 RF_Malloc(specific_buf, k_cfg->layoutSpecificSize,
877 (u_char *));
878 if (specific_buf == NULL) {
879 RF_Free(k_cfg, sizeof(RF_Config_t));
880 return (ENOMEM);
881 }
882 retcode = copyin(k_cfg->layoutSpecific,
883 (caddr_t) specific_buf,
884 k_cfg->layoutSpecificSize);
885 if (retcode) {
886 RF_Free(k_cfg, sizeof(RF_Config_t));
887 RF_Free(specific_buf,
888 k_cfg->layoutSpecificSize);
889 db1_printf(("rf_ioctl: retcode=%d copyin.2\n",
890 retcode));
891 return (retcode);
892 }
893 } else
894 specific_buf = NULL;
895 k_cfg->layoutSpecific = specific_buf;
896
897 /* should do some kind of sanity check on the configuration.
898 * Store the sum of all the bytes in the last byte? */
899
900 /* configure the system */
901
902 /*
903 * Clear the entire RAID descriptor, just to make sure
904 * there is no stale data left in the case of a
905 * reconfiguration
906 */
907 bzero((char *) raidPtr, sizeof(RF_Raid_t));
908 raidPtr->raidid = unit;
909
910 retcode = rf_Configure(raidPtr, k_cfg, NULL);
911
912 if (retcode == 0) {
913
914 /* allow this many simultaneous IO's to
915 this RAID device */
916 raidPtr->openings = RAIDOUTSTANDING;
917
918 raidinit(raidPtr);
919 rf_markalldirty(raidPtr);
920 }
921 /* free the buffers. No return code here. */
922 if (k_cfg->layoutSpecificSize) {
923 RF_Free(specific_buf, k_cfg->layoutSpecificSize);
924 }
925 RF_Free(k_cfg, sizeof(RF_Config_t));
926
927 return (retcode);
928
929 /* shutdown the system */
930 case RAIDFRAME_SHUTDOWN:
931
932 if ((error = raidlock(rs)) != 0)
933 return (error);
934
935 /*
936 * If somebody has a partition mounted, we shouldn't
937 * shutdown.
938 */
939
940 part = DISKPART(dev);
941 pmask = (1 << part);
942 if ((rs->sc_dkdev.dk_openmask & ~pmask) ||
943 ((rs->sc_dkdev.dk_bopenmask & pmask) &&
944 (rs->sc_dkdev.dk_copenmask & pmask))) {
945 raidunlock(rs);
946 return (EBUSY);
947 }
948
949 retcode = rf_Shutdown(raidPtr);
950
951 pool_destroy(&rs->sc_cbufpool);
952
953 /* It's no longer initialized... */
954 rs->sc_flags &= ~RAIDF_INITED;
955
956 /* Detach the disk. */
957 disk_detach(&rs->sc_dkdev);
958
959 raidunlock(rs);
960
961 return (retcode);
962 case RAIDFRAME_GET_COMPONENT_LABEL:
963 clabel_ptr = (RF_ComponentLabel_t **) data;
964 /* need to read the component label for the disk indicated
965 by row,column in clabel */
966
967 /* For practice, let's get it directly fromdisk, rather
968 than from the in-core copy */
969 RF_Malloc( clabel, sizeof( RF_ComponentLabel_t ),
970 (RF_ComponentLabel_t *));
971 if (clabel == NULL)
972 return (ENOMEM);
973
974 bzero((char *) clabel, sizeof(RF_ComponentLabel_t));
975
976 retcode = copyin( *clabel_ptr, clabel,
977 sizeof(RF_ComponentLabel_t));
978
979 if (retcode) {
980 RF_Free( clabel, sizeof(RF_ComponentLabel_t));
981 return(retcode);
982 }
983
984 row = clabel->row;
985 column = clabel->column;
986
987 if ((row < 0) || (row >= raidPtr->numRow) ||
988 (column < 0) || (column >= raidPtr->numCol +
989 raidPtr->numSpare)) {
990 RF_Free( clabel, sizeof(RF_ComponentLabel_t));
991 return(EINVAL);
992 }
993
994 raidread_component_label(raidPtr->Disks[row][column].dev,
995 raidPtr->raid_cinfo[row][column].ci_vp,
996 clabel );
997
998 retcode = copyout((caddr_t) clabel,
999 (caddr_t) *clabel_ptr,
1000 sizeof(RF_ComponentLabel_t));
1001 RF_Free( clabel, sizeof(RF_ComponentLabel_t));
1002 return (retcode);
1003
1004 case RAIDFRAME_SET_COMPONENT_LABEL:
1005 clabel = (RF_ComponentLabel_t *) data;
1006
1007 /* XXX check the label for valid stuff... */
1008 /* Note that some things *should not* get modified --
1009 the user should be re-initing the labels instead of
1010 trying to patch things.
1011 */
1012
1013 printf("Got component label:\n");
1014 printf("Version: %d\n",clabel->version);
1015 printf("Serial Number: %d\n",clabel->serial_number);
1016 printf("Mod counter: %d\n",clabel->mod_counter);
1017 printf("Row: %d\n", clabel->row);
1018 printf("Column: %d\n", clabel->column);
1019 printf("Num Rows: %d\n", clabel->num_rows);
1020 printf("Num Columns: %d\n", clabel->num_columns);
1021 printf("Clean: %d\n", clabel->clean);
1022 printf("Status: %d\n", clabel->status);
1023
1024 row = clabel->row;
1025 column = clabel->column;
1026
1027 if ((row < 0) || (row >= raidPtr->numRow) ||
1028 (column < 0) || (column >= raidPtr->numCol)) {
1029 return(EINVAL);
1030 }
1031
1032 /* XXX this isn't allowed to do anything for now :-) */
1033
1034 /* XXX and before it is, we need to fill in the rest
1035 of the fields!?!?!?! */
1036 #if 0
1037 raidwrite_component_label(
1038 raidPtr->Disks[row][column].dev,
1039 raidPtr->raid_cinfo[row][column].ci_vp,
1040 clabel );
1041 #endif
1042 return (0);
1043
1044 case RAIDFRAME_INIT_LABELS:
1045 clabel = (RF_ComponentLabel_t *) data;
1046 /*
1047 we only want the serial number from
1048 the above. We get all the rest of the information
1049 from the config that was used to create this RAID
1050 set.
1051 */
1052
1053 raidPtr->serial_number = clabel->serial_number;
1054
1055 raid_init_component_label(raidPtr, &ci_label);
1056 ci_label.serial_number = clabel->serial_number;
1057
1058 for(row=0;row<raidPtr->numRow;row++) {
1059 ci_label.row = row;
1060 for(column=0;column<raidPtr->numCol;column++) {
1061 diskPtr = &raidPtr->Disks[row][column];
1062 ci_label.partitionSize = diskPtr->partitionSize;
1063 ci_label.column = column;
1064 raidwrite_component_label(
1065 raidPtr->Disks[row][column].dev,
1066 raidPtr->raid_cinfo[row][column].ci_vp,
1067 &ci_label );
1068 }
1069 }
1070
1071 return (retcode);
1072 case RAIDFRAME_SET_AUTOCONFIG:
1073 d = rf_set_autoconfig(raidPtr, *(int *) data);
1074 printf("New autoconfig value is: %d\n", d);
1075 *(int *) data = d;
1076 return (retcode);
1077
1078 case RAIDFRAME_SET_ROOT:
1079 d = rf_set_rootpartition(raidPtr, *(int *) data);
1080 printf("New rootpartition value is: %d\n", d);
1081 *(int *) data = d;
1082 return (retcode);
1083
1084 /* initialize all parity */
1085 case RAIDFRAME_REWRITEPARITY:
1086
1087 if (raidPtr->Layout.map->faultsTolerated == 0) {
1088 /* Parity for RAID 0 is trivially correct */
1089 raidPtr->parity_good = RF_RAID_CLEAN;
1090 return(0);
1091 }
1092
1093 if (raidPtr->parity_rewrite_in_progress == 1) {
1094 /* Re-write is already in progress! */
1095 return(EINVAL);
1096 }
1097
1098 retcode = RF_CREATE_THREAD(raidPtr->parity_rewrite_thread,
1099 rf_RewriteParityThread,
1100 raidPtr,"raid_parity");
1101 return (retcode);
1102
1103
1104 case RAIDFRAME_ADD_HOT_SPARE:
1105 sparePtr = (RF_SingleComponent_t *) data;
1106 memcpy( &hot_spare, sparePtr, sizeof(RF_SingleComponent_t));
1107 retcode = rf_add_hot_spare(raidPtr, &hot_spare);
1108 return(retcode);
1109
1110 case RAIDFRAME_REMOVE_HOT_SPARE:
1111 return(retcode);
1112
1113 case RAIDFRAME_DELETE_COMPONENT:
1114 componentPtr = (RF_SingleComponent_t *)data;
1115 memcpy( &component, componentPtr,
1116 sizeof(RF_SingleComponent_t));
1117 retcode = rf_delete_component(raidPtr, &component);
1118 return(retcode);
1119
1120 case RAIDFRAME_INCORPORATE_HOT_SPARE:
1121 componentPtr = (RF_SingleComponent_t *)data;
1122 memcpy( &component, componentPtr,
1123 sizeof(RF_SingleComponent_t));
1124 retcode = rf_incorporate_hot_spare(raidPtr, &component);
1125 return(retcode);
1126
1127 case RAIDFRAME_REBUILD_IN_PLACE:
1128
1129 if (raidPtr->Layout.map->faultsTolerated == 0) {
1130 /* Can't do this on a RAID 0!! */
1131 return(EINVAL);
1132 }
1133
1134 if (raidPtr->recon_in_progress == 1) {
1135 /* a reconstruct is already in progress! */
1136 return(EINVAL);
1137 }
1138
1139 componentPtr = (RF_SingleComponent_t *) data;
1140 memcpy( &component, componentPtr,
1141 sizeof(RF_SingleComponent_t));
1142 row = component.row;
1143 column = component.column;
1144 printf("Rebuild: %d %d\n",row, column);
1145 if ((row < 0) || (row >= raidPtr->numRow) ||
1146 (column < 0) || (column >= raidPtr->numCol)) {
1147 return(EINVAL);
1148 }
1149
1150 RF_Malloc(rrcopy, sizeof(*rrcopy), (struct rf_recon_req *));
1151 if (rrcopy == NULL)
1152 return(ENOMEM);
1153
1154 rrcopy->raidPtr = (void *) raidPtr;
1155 rrcopy->row = row;
1156 rrcopy->col = column;
1157
1158 retcode = RF_CREATE_THREAD(raidPtr->recon_thread,
1159 rf_ReconstructInPlaceThread,
1160 rrcopy,"raid_reconip");
1161 return(retcode);
1162
1163 case RAIDFRAME_GET_INFO:
1164 if (!raidPtr->valid)
1165 return (ENODEV);
1166 ucfgp = (RF_DeviceConfig_t **) data;
1167 RF_Malloc(d_cfg, sizeof(RF_DeviceConfig_t),
1168 (RF_DeviceConfig_t *));
1169 if (d_cfg == NULL)
1170 return (ENOMEM);
1171 bzero((char *) d_cfg, sizeof(RF_DeviceConfig_t));
1172 d_cfg->rows = raidPtr->numRow;
1173 d_cfg->cols = raidPtr->numCol;
1174 d_cfg->ndevs = raidPtr->numRow * raidPtr->numCol;
1175 if (d_cfg->ndevs >= RF_MAX_DISKS) {
1176 RF_Free(d_cfg, sizeof(RF_DeviceConfig_t));
1177 return (ENOMEM);
1178 }
1179 d_cfg->nspares = raidPtr->numSpare;
1180 if (d_cfg->nspares >= RF_MAX_DISKS) {
1181 RF_Free(d_cfg, sizeof(RF_DeviceConfig_t));
1182 return (ENOMEM);
1183 }
1184 d_cfg->maxqdepth = raidPtr->maxQueueDepth;
1185 d = 0;
1186 for (i = 0; i < d_cfg->rows; i++) {
1187 for (j = 0; j < d_cfg->cols; j++) {
1188 d_cfg->devs[d] = raidPtr->Disks[i][j];
1189 d++;
1190 }
1191 }
1192 for (j = d_cfg->cols, i = 0; i < d_cfg->nspares; i++, j++) {
1193 d_cfg->spares[i] = raidPtr->Disks[0][j];
1194 }
1195 retcode = copyout((caddr_t) d_cfg, (caddr_t) * ucfgp,
1196 sizeof(RF_DeviceConfig_t));
1197 RF_Free(d_cfg, sizeof(RF_DeviceConfig_t));
1198
1199 return (retcode);
1200
1201 case RAIDFRAME_CHECK_PARITY:
1202 *(int *) data = raidPtr->parity_good;
1203 return (0);
1204
1205 case RAIDFRAME_RESET_ACCTOTALS:
1206 bzero(&raidPtr->acc_totals, sizeof(raidPtr->acc_totals));
1207 return (0);
1208
1209 case RAIDFRAME_GET_ACCTOTALS:
1210 totals = (RF_AccTotals_t *) data;
1211 *totals = raidPtr->acc_totals;
1212 return (0);
1213
1214 case RAIDFRAME_KEEP_ACCTOTALS:
1215 raidPtr->keep_acc_totals = *(int *)data;
1216 return (0);
1217
1218 case RAIDFRAME_GET_SIZE:
1219 *(int *) data = raidPtr->totalSectors;
1220 return (0);
1221
1222 /* fail a disk & optionally start reconstruction */
1223 case RAIDFRAME_FAIL_DISK:
1224
1225 if (raidPtr->Layout.map->faultsTolerated == 0) {
1226 /* Can't do this on a RAID 0!! */
1227 return(EINVAL);
1228 }
1229
1230 rr = (struct rf_recon_req *) data;
1231
1232 if (rr->row < 0 || rr->row >= raidPtr->numRow
1233 || rr->col < 0 || rr->col >= raidPtr->numCol)
1234 return (EINVAL);
1235
1236 printf("raid%d: Failing the disk: row: %d col: %d\n",
1237 unit, rr->row, rr->col);
1238
1239 /* make a copy of the recon request so that we don't rely on
1240 * the user's buffer */
1241 RF_Malloc(rrcopy, sizeof(*rrcopy), (struct rf_recon_req *));
1242 if (rrcopy == NULL)
1243 return(ENOMEM);
1244 bcopy(rr, rrcopy, sizeof(*rr));
1245 rrcopy->raidPtr = (void *) raidPtr;
1246
1247 retcode = RF_CREATE_THREAD(raidPtr->recon_thread,
1248 rf_ReconThread,
1249 rrcopy,"raid_recon");
1250 return (0);
1251
1252 /* invoke a copyback operation after recon on whatever disk
1253 * needs it, if any */
1254 case RAIDFRAME_COPYBACK:
1255
1256 if (raidPtr->Layout.map->faultsTolerated == 0) {
1257 /* This makes no sense on a RAID 0!! */
1258 return(EINVAL);
1259 }
1260
1261 if (raidPtr->copyback_in_progress == 1) {
1262 /* Copyback is already in progress! */
1263 return(EINVAL);
1264 }
1265
1266 retcode = RF_CREATE_THREAD(raidPtr->copyback_thread,
1267 rf_CopybackThread,
1268 raidPtr,"raid_copyback");
1269 return (retcode);
1270
1271 /* return the percentage completion of reconstruction */
1272 case RAIDFRAME_CHECK_RECON_STATUS:
1273 if (raidPtr->Layout.map->faultsTolerated == 0) {
1274 /* This makes no sense on a RAID 0, so tell the
1275 user it's done. */
1276 *(int *) data = 100;
1277 return(0);
1278 }
1279 row = 0; /* XXX we only consider a single row... */
1280 if (raidPtr->status[row] != rf_rs_reconstructing)
1281 *(int *) data = 100;
1282 else
1283 *(int *) data = raidPtr->reconControl[row]->percentComplete;
1284 return (0);
1285 case RAIDFRAME_CHECK_RECON_STATUS_EXT:
1286 progressInfoPtr = (RF_ProgressInfo_t **) data;
1287 row = 0; /* XXX we only consider a single row... */
1288 if (raidPtr->status[row] != rf_rs_reconstructing) {
1289 progressInfo.remaining = 0;
1290 progressInfo.completed = 100;
1291 progressInfo.total = 100;
1292 } else {
1293 progressInfo.total =
1294 raidPtr->reconControl[row]->numRUsTotal;
1295 progressInfo.completed =
1296 raidPtr->reconControl[row]->numRUsComplete;
1297 progressInfo.remaining = progressInfo.total -
1298 progressInfo.completed;
1299 }
1300 retcode = copyout((caddr_t) &progressInfo,
1301 (caddr_t) *progressInfoPtr,
1302 sizeof(RF_ProgressInfo_t));
1303 return (retcode);
1304
1305 case RAIDFRAME_CHECK_PARITYREWRITE_STATUS:
1306 if (raidPtr->Layout.map->faultsTolerated == 0) {
1307 /* This makes no sense on a RAID 0, so tell the
1308 user it's done. */
1309 *(int *) data = 100;
1310 return(0);
1311 }
1312 if (raidPtr->parity_rewrite_in_progress == 1) {
1313 *(int *) data = 100 *
1314 raidPtr->parity_rewrite_stripes_done /
1315 raidPtr->Layout.numStripe;
1316 } else {
1317 *(int *) data = 100;
1318 }
1319 return (0);
1320
1321 case RAIDFRAME_CHECK_PARITYREWRITE_STATUS_EXT:
1322 progressInfoPtr = (RF_ProgressInfo_t **) data;
1323 if (raidPtr->parity_rewrite_in_progress == 1) {
1324 progressInfo.total = raidPtr->Layout.numStripe;
1325 progressInfo.completed =
1326 raidPtr->parity_rewrite_stripes_done;
1327 progressInfo.remaining = progressInfo.total -
1328 progressInfo.completed;
1329 } else {
1330 progressInfo.remaining = 0;
1331 progressInfo.completed = 100;
1332 progressInfo.total = 100;
1333 }
1334 retcode = copyout((caddr_t) &progressInfo,
1335 (caddr_t) *progressInfoPtr,
1336 sizeof(RF_ProgressInfo_t));
1337 return (retcode);
1338
1339 case RAIDFRAME_CHECK_COPYBACK_STATUS:
1340 if (raidPtr->Layout.map->faultsTolerated == 0) {
1341 /* This makes no sense on a RAID 0 */
1342 *(int *) data = 100;
1343 return(0);
1344 }
1345 if (raidPtr->copyback_in_progress == 1) {
1346 *(int *) data = 100 * raidPtr->copyback_stripes_done /
1347 raidPtr->Layout.numStripe;
1348 } else {
1349 *(int *) data = 100;
1350 }
1351 return (0);
1352
1353 case RAIDFRAME_CHECK_COPYBACK_STATUS_EXT:
1354 if (raidPtr->copyback_in_progress == 1) {
1355 progressInfo.total = raidPtr->Layout.numStripe;
1356 progressInfo.completed =
1357 raidPtr->parity_rewrite_stripes_done;
1358 progressInfo.remaining = progressInfo.total -
1359 progressInfo.completed;
1360 } else {
1361 progressInfo.remaining = 0;
1362 progressInfo.completed = 100;
1363 progressInfo.total = 100;
1364 }
1365 retcode = copyout((caddr_t) &progressInfo,
1366 (caddr_t) *progressInfoPtr,
1367 sizeof(RF_ProgressInfo_t));
1368 return (retcode);
1369
1370 /* the sparetable daemon calls this to wait for the kernel to
1371 * need a spare table. this ioctl does not return until a
1372 * spare table is needed. XXX -- calling mpsleep here in the
1373 * ioctl code is almost certainly wrong and evil. -- XXX XXX
1374 * -- I should either compute the spare table in the kernel,
1375 * or have a different -- XXX XXX -- interface (a different
1376 * character device) for delivering the table -- XXX */
1377 #if 0
1378 case RAIDFRAME_SPARET_WAIT:
1379 RF_LOCK_MUTEX(rf_sparet_wait_mutex);
1380 while (!rf_sparet_wait_queue)
1381 mpsleep(&rf_sparet_wait_queue, (PZERO + 1) | PCATCH, "sparet wait", 0, (void *) simple_lock_addr(rf_sparet_wait_mutex), MS_LOCK_SIMPLE);
1382 waitreq = rf_sparet_wait_queue;
1383 rf_sparet_wait_queue = rf_sparet_wait_queue->next;
1384 RF_UNLOCK_MUTEX(rf_sparet_wait_mutex);
1385
1386 /* structure assignment */
1387 *((RF_SparetWait_t *) data) = *waitreq;
1388
1389 RF_Free(waitreq, sizeof(*waitreq));
1390 return (0);
1391
1392 /* wakes up a process waiting on SPARET_WAIT and puts an error
1393 * code in it that will cause the dameon to exit */
1394 case RAIDFRAME_ABORT_SPARET_WAIT:
1395 RF_Malloc(waitreq, sizeof(*waitreq), (RF_SparetWait_t *));
1396 waitreq->fcol = -1;
1397 RF_LOCK_MUTEX(rf_sparet_wait_mutex);
1398 waitreq->next = rf_sparet_wait_queue;
1399 rf_sparet_wait_queue = waitreq;
1400 RF_UNLOCK_MUTEX(rf_sparet_wait_mutex);
1401 wakeup(&rf_sparet_wait_queue);
1402 return (0);
1403
1404 /* used by the spare table daemon to deliver a spare table
1405 * into the kernel */
1406 case RAIDFRAME_SEND_SPARET:
1407
1408 /* install the spare table */
1409 retcode = rf_SetSpareTable(raidPtr, *(void **) data);
1410
1411 /* respond to the requestor. the return status of the spare
1412 * table installation is passed in the "fcol" field */
1413 RF_Malloc(waitreq, sizeof(*waitreq), (RF_SparetWait_t *));
1414 waitreq->fcol = retcode;
1415 RF_LOCK_MUTEX(rf_sparet_wait_mutex);
1416 waitreq->next = rf_sparet_resp_queue;
1417 rf_sparet_resp_queue = waitreq;
1418 wakeup(&rf_sparet_resp_queue);
1419 RF_UNLOCK_MUTEX(rf_sparet_wait_mutex);
1420
1421 return (retcode);
1422 #endif
1423
1424 default:
1425 break; /* fall through to the os-specific code below */
1426
1427 }
1428
1429 if (!raidPtr->valid)
1430 return (EINVAL);
1431
1432 /*
1433 * Add support for "regular" device ioctls here.
1434 */
1435
1436 switch (cmd) {
1437 case DIOCGDINFO:
1438 *(struct disklabel *) data = *(rs->sc_dkdev.dk_label);
1439 break;
1440
1441 case DIOCGPART:
1442 ((struct partinfo *) data)->disklab = rs->sc_dkdev.dk_label;
1443 ((struct partinfo *) data)->part =
1444 &rs->sc_dkdev.dk_label->d_partitions[DISKPART(dev)];
1445 break;
1446
1447 case DIOCWDINFO:
1448 case DIOCSDINFO:
1449 if ((error = raidlock(rs)) != 0)
1450 return (error);
1451
1452 rs->sc_flags |= RAIDF_LABELLING;
1453
1454 error = setdisklabel(rs->sc_dkdev.dk_label,
1455 (struct disklabel *) data, 0, rs->sc_dkdev.dk_cpulabel);
1456 if (error == 0) {
1457 if (cmd == DIOCWDINFO)
1458 error = writedisklabel(RAIDLABELDEV(dev),
1459 raidstrategy, rs->sc_dkdev.dk_label,
1460 rs->sc_dkdev.dk_cpulabel);
1461 }
1462 rs->sc_flags &= ~RAIDF_LABELLING;
1463
1464 raidunlock(rs);
1465
1466 if (error)
1467 return (error);
1468 break;
1469
1470 case DIOCWLABEL:
1471 if (*(int *) data != 0)
1472 rs->sc_flags |= RAIDF_WLABEL;
1473 else
1474 rs->sc_flags &= ~RAIDF_WLABEL;
1475 break;
1476
1477 case DIOCGDEFLABEL:
1478 raidgetdefaultlabel(raidPtr, rs,
1479 (struct disklabel *) data);
1480 break;
1481
1482 default:
1483 retcode = ENOTTY;
1484 }
1485 return (retcode);
1486
1487 }
1488
1489
1490 /* raidinit -- complete the rest of the initialization for the
1491 RAIDframe device. */
1492
1493
1494 static void
1495 raidinit(raidPtr)
1496 RF_Raid_t *raidPtr;
1497 {
1498 struct raid_softc *rs;
1499 int unit;
1500
1501 unit = raidPtr->raidid;
1502
1503 rs = &raid_softc[unit];
1504 pool_init(&rs->sc_cbufpool, sizeof(struct raidbuf), 0,
1505 0, 0, "raidpl", 0, NULL, NULL, M_RAIDFRAME);
1506
1507
1508 /* XXX should check return code first... */
1509 rs->sc_flags |= RAIDF_INITED;
1510
1511 sprintf(rs->sc_xname, "raid%d", unit); /* XXX doesn't check bounds. */
1512
1513 rs->sc_dkdev.dk_name = rs->sc_xname;
1514
1515 /* disk_attach actually creates space for the CPU disklabel, among
1516 * other things, so it's critical to call this *BEFORE* we try putzing
1517 * with disklabels. */
1518
1519 disk_attach(&rs->sc_dkdev);
1520
1521 /* XXX There may be a weird interaction here between this, and
1522 * protectedSectors, as used in RAIDframe. */
1523
1524 rs->sc_size = raidPtr->totalSectors;
1525
1526 }
1527
1528 /* wake up the daemon & tell it to get us a spare table
1529 * XXX
1530 * the entries in the queues should be tagged with the raidPtr
1531 * so that in the extremely rare case that two recons happen at once,
1532 * we know for which device were requesting a spare table
1533 * XXX
1534 *
1535 * XXX This code is not currently used. GO
1536 */
1537 int
1538 rf_GetSpareTableFromDaemon(req)
1539 RF_SparetWait_t *req;
1540 {
1541 int retcode;
1542
1543 RF_LOCK_MUTEX(rf_sparet_wait_mutex);
1544 req->next = rf_sparet_wait_queue;
1545 rf_sparet_wait_queue = req;
1546 wakeup(&rf_sparet_wait_queue);
1547
1548 /* mpsleep unlocks the mutex */
1549 while (!rf_sparet_resp_queue) {
1550 tsleep(&rf_sparet_resp_queue, PRIBIO,
1551 "raidframe getsparetable", 0);
1552 }
1553 req = rf_sparet_resp_queue;
1554 rf_sparet_resp_queue = req->next;
1555 RF_UNLOCK_MUTEX(rf_sparet_wait_mutex);
1556
1557 retcode = req->fcol;
1558 RF_Free(req, sizeof(*req)); /* this is not the same req as we
1559 * alloc'd */
1560 return (retcode);
1561 }
1562
1563 /* a wrapper around rf_DoAccess that extracts appropriate info from the
1564 * bp & passes it down.
1565 * any calls originating in the kernel must use non-blocking I/O
1566 * do some extra sanity checking to return "appropriate" error values for
1567 * certain conditions (to make some standard utilities work)
1568 *
1569 * Formerly known as: rf_DoAccessKernel
1570 */
1571 void
1572 raidstart(raidPtr)
1573 RF_Raid_t *raidPtr;
1574 {
1575 RF_SectorCount_t num_blocks, pb, sum;
1576 RF_RaidAddr_t raid_addr;
1577 int retcode;
1578 struct partition *pp;
1579 daddr_t blocknum;
1580 int unit;
1581 struct raid_softc *rs;
1582 int do_async;
1583 struct buf *bp;
1584
1585 unit = raidPtr->raidid;
1586 rs = &raid_softc[unit];
1587
1588 /* quick check to see if anything has died recently */
1589 RF_LOCK_MUTEX(raidPtr->mutex);
1590 if (raidPtr->numNewFailures > 0) {
1591 rf_update_component_labels(raidPtr,
1592 RF_NORMAL_COMPONENT_UPDATE);
1593 raidPtr->numNewFailures--;
1594 }
1595 RF_UNLOCK_MUTEX(raidPtr->mutex);
1596
1597 /* Check to see if we're at the limit... */
1598 RF_LOCK_MUTEX(raidPtr->mutex);
1599 while (raidPtr->openings > 0) {
1600 RF_UNLOCK_MUTEX(raidPtr->mutex);
1601
1602 /* get the next item, if any, from the queue */
1603 if ((bp = BUFQ_FIRST(&rs->buf_queue)) == NULL) {
1604 /* nothing more to do */
1605 return;
1606 }
1607 BUFQ_REMOVE(&rs->buf_queue, bp);
1608
1609 /* Ok, for the bp we have here, bp->b_blkno is relative to the
1610 * partition.. Need to make it absolute to the underlying
1611 * device.. */
1612
1613 blocknum = bp->b_blkno;
1614 if (DISKPART(bp->b_dev) != RAW_PART) {
1615 pp = &rs->sc_dkdev.dk_label->d_partitions[DISKPART(bp->b_dev)];
1616 blocknum += pp->p_offset;
1617 }
1618
1619 db1_printf(("Blocks: %d, %d\n", (int) bp->b_blkno,
1620 (int) blocknum));
1621
1622 db1_printf(("bp->b_bcount = %d\n", (int) bp->b_bcount));
1623 db1_printf(("bp->b_resid = %d\n", (int) bp->b_resid));
1624
1625 /* *THIS* is where we adjust what block we're going to...
1626 * but DO NOT TOUCH bp->b_blkno!!! */
1627 raid_addr = blocknum;
1628
1629 num_blocks = bp->b_bcount >> raidPtr->logBytesPerSector;
1630 pb = (bp->b_bcount & raidPtr->sectorMask) ? 1 : 0;
1631 sum = raid_addr + num_blocks + pb;
1632 if (1 || rf_debugKernelAccess) {
1633 db1_printf(("raid_addr=%d sum=%d num_blocks=%d(+%d) (%d)\n",
1634 (int) raid_addr, (int) sum, (int) num_blocks,
1635 (int) pb, (int) bp->b_resid));
1636 }
1637 if ((sum > raidPtr->totalSectors) || (sum < raid_addr)
1638 || (sum < num_blocks) || (sum < pb)) {
1639 bp->b_error = ENOSPC;
1640 bp->b_flags |= B_ERROR;
1641 bp->b_resid = bp->b_bcount;
1642 biodone(bp);
1643 RF_LOCK_MUTEX(raidPtr->mutex);
1644 continue;
1645 }
1646 /*
1647 * XXX rf_DoAccess() should do this, not just DoAccessKernel()
1648 */
1649
1650 if (bp->b_bcount & raidPtr->sectorMask) {
1651 bp->b_error = EINVAL;
1652 bp->b_flags |= B_ERROR;
1653 bp->b_resid = bp->b_bcount;
1654 biodone(bp);
1655 RF_LOCK_MUTEX(raidPtr->mutex);
1656 continue;
1657
1658 }
1659 db1_printf(("Calling DoAccess..\n"));
1660
1661
1662 RF_LOCK_MUTEX(raidPtr->mutex);
1663 raidPtr->openings--;
1664 RF_UNLOCK_MUTEX(raidPtr->mutex);
1665
1666 /*
1667 * Everything is async.
1668 */
1669 do_async = 1;
1670
1671 /* don't ever condition on bp->b_flags & B_WRITE.
1672 * always condition on B_READ instead */
1673
1674 /* XXX we're still at splbio() here... do we *really*
1675 need to be? */
1676
1677
1678 retcode = rf_DoAccess(raidPtr, (bp->b_flags & B_READ) ?
1679 RF_IO_TYPE_READ : RF_IO_TYPE_WRITE,
1680 do_async, raid_addr, num_blocks,
1681 bp->b_data, bp, NULL, NULL,
1682 RF_DAG_NONBLOCKING_IO, NULL, NULL, NULL);
1683
1684
1685 RF_LOCK_MUTEX(raidPtr->mutex);
1686 }
1687 RF_UNLOCK_MUTEX(raidPtr->mutex);
1688 }
1689
1690
1691
1692
1693 /* invoke an I/O from kernel mode. Disk queue should be locked upon entry */
1694
1695 int
1696 rf_DispatchKernelIO(queue, req)
1697 RF_DiskQueue_t *queue;
1698 RF_DiskQueueData_t *req;
1699 {
1700 int op = (req->type == RF_IO_TYPE_READ) ? B_READ : B_WRITE;
1701 struct buf *bp;
1702 struct raidbuf *raidbp = NULL;
1703 struct raid_softc *rs;
1704 int unit;
1705 int s;
1706
1707 s=0;
1708 /* s = splbio();*/ /* want to test this */
1709 /* XXX along with the vnode, we also need the softc associated with
1710 * this device.. */
1711
1712 req->queue = queue;
1713
1714 unit = queue->raidPtr->raidid;
1715
1716 db1_printf(("DispatchKernelIO unit: %d\n", unit));
1717
1718 if (unit >= numraid) {
1719 printf("Invalid unit number: %d %d\n", unit, numraid);
1720 panic("Invalid Unit number in rf_DispatchKernelIO\n");
1721 }
1722 rs = &raid_softc[unit];
1723
1724 /* XXX is this the right place? */
1725 disk_busy(&rs->sc_dkdev);
1726
1727 bp = req->bp;
1728 #if 1
1729 /* XXX when there is a physical disk failure, someone is passing us a
1730 * buffer that contains old stuff!! Attempt to deal with this problem
1731 * without taking a performance hit... (not sure where the real bug
1732 * is. It's buried in RAIDframe somewhere) :-( GO ) */
1733
1734 if (bp->b_flags & B_ERROR) {
1735 bp->b_flags &= ~B_ERROR;
1736 }
1737 if (bp->b_error != 0) {
1738 bp->b_error = 0;
1739 }
1740 #endif
1741 raidbp = RAIDGETBUF(rs);
1742
1743 raidbp->rf_flags = 0; /* XXX not really used anywhere... */
1744
1745 /*
1746 * context for raidiodone
1747 */
1748 raidbp->rf_obp = bp;
1749 raidbp->req = req;
1750
1751 LIST_INIT(&raidbp->rf_buf.b_dep);
1752
1753 switch (req->type) {
1754 case RF_IO_TYPE_NOP: /* used primarily to unlock a locked queue */
1755 /* XXX need to do something extra here.. */
1756 /* I'm leaving this in, as I've never actually seen it used,
1757 * and I'd like folks to report it... GO */
1758 printf(("WAKEUP CALLED\n"));
1759 queue->numOutstanding++;
1760
1761 /* XXX need to glue the original buffer into this?? */
1762
1763 KernelWakeupFunc(&raidbp->rf_buf);
1764 break;
1765
1766 case RF_IO_TYPE_READ:
1767 case RF_IO_TYPE_WRITE:
1768
1769 if (req->tracerec) {
1770 RF_ETIMER_START(req->tracerec->timer);
1771 }
1772 InitBP(&raidbp->rf_buf, queue->rf_cinfo->ci_vp,
1773 op | bp->b_flags, queue->rf_cinfo->ci_dev,
1774 req->sectorOffset, req->numSector,
1775 req->buf, KernelWakeupFunc, (void *) req,
1776 queue->raidPtr->logBytesPerSector, req->b_proc);
1777
1778 if (rf_debugKernelAccess) {
1779 db1_printf(("dispatch: bp->b_blkno = %ld\n",
1780 (long) bp->b_blkno));
1781 }
1782 queue->numOutstanding++;
1783 queue->last_deq_sector = req->sectorOffset;
1784 /* acc wouldn't have been let in if there were any pending
1785 * reqs at any other priority */
1786 queue->curPriority = req->priority;
1787
1788 db1_printf(("Going for %c to unit %d row %d col %d\n",
1789 req->type, unit, queue->row, queue->col));
1790 db1_printf(("sector %d count %d (%d bytes) %d\n",
1791 (int) req->sectorOffset, (int) req->numSector,
1792 (int) (req->numSector <<
1793 queue->raidPtr->logBytesPerSector),
1794 (int) queue->raidPtr->logBytesPerSector));
1795 if ((raidbp->rf_buf.b_flags & B_READ) == 0) {
1796 raidbp->rf_buf.b_vp->v_numoutput++;
1797 }
1798 VOP_STRATEGY(&raidbp->rf_buf);
1799
1800 break;
1801
1802 default:
1803 panic("bad req->type in rf_DispatchKernelIO");
1804 }
1805 db1_printf(("Exiting from DispatchKernelIO\n"));
1806 /* splx(s); */ /* want to test this */
1807 return (0);
1808 }
1809 /* this is the callback function associated with a I/O invoked from
1810 kernel code.
1811 */
1812 static void
1813 KernelWakeupFunc(vbp)
1814 struct buf *vbp;
1815 {
1816 RF_DiskQueueData_t *req = NULL;
1817 RF_DiskQueue_t *queue;
1818 struct raidbuf *raidbp = (struct raidbuf *) vbp;
1819 struct buf *bp;
1820 struct raid_softc *rs;
1821 int unit;
1822 int s;
1823
1824 s = splbio();
1825 db1_printf(("recovering the request queue:\n"));
1826 req = raidbp->req;
1827
1828 bp = raidbp->rf_obp;
1829
1830 queue = (RF_DiskQueue_t *) req->queue;
1831
1832 if (raidbp->rf_buf.b_flags & B_ERROR) {
1833 bp->b_flags |= B_ERROR;
1834 bp->b_error = raidbp->rf_buf.b_error ?
1835 raidbp->rf_buf.b_error : EIO;
1836 }
1837
1838 /* XXX methinks this could be wrong... */
1839 #if 1
1840 bp->b_resid = raidbp->rf_buf.b_resid;
1841 #endif
1842
1843 if (req->tracerec) {
1844 RF_ETIMER_STOP(req->tracerec->timer);
1845 RF_ETIMER_EVAL(req->tracerec->timer);
1846 RF_LOCK_MUTEX(rf_tracing_mutex);
1847 req->tracerec->diskwait_us += RF_ETIMER_VAL_US(req->tracerec->timer);
1848 req->tracerec->phys_io_us += RF_ETIMER_VAL_US(req->tracerec->timer);
1849 req->tracerec->num_phys_ios++;
1850 RF_UNLOCK_MUTEX(rf_tracing_mutex);
1851 }
1852 bp->b_bcount = raidbp->rf_buf.b_bcount; /* XXXX ?? */
1853
1854 unit = queue->raidPtr->raidid; /* *Much* simpler :-> */
1855
1856
1857 /* XXX Ok, let's get aggressive... If B_ERROR is set, let's go
1858 * ballistic, and mark the component as hosed... */
1859
1860 if (bp->b_flags & B_ERROR) {
1861 /* Mark the disk as dead */
1862 /* but only mark it once... */
1863 if (queue->raidPtr->Disks[queue->row][queue->col].status ==
1864 rf_ds_optimal) {
1865 printf("raid%d: IO Error. Marking %s as failed.\n",
1866 unit, queue->raidPtr->Disks[queue->row][queue->col].devname);
1867 queue->raidPtr->Disks[queue->row][queue->col].status =
1868 rf_ds_failed;
1869 queue->raidPtr->status[queue->row] = rf_rs_degraded;
1870 queue->raidPtr->numFailures++;
1871 queue->raidPtr->numNewFailures++;
1872 } else { /* Disk is already dead... */
1873 /* printf("Disk already marked as dead!\n"); */
1874 }
1875
1876 }
1877
1878 rs = &raid_softc[unit];
1879 RAIDPUTBUF(rs, raidbp);
1880
1881
1882 if (bp->b_resid == 0) {
1883 /* XXX is this the right place for a disk_unbusy()??!??!?!? */
1884 disk_unbusy(&rs->sc_dkdev, (bp->b_bcount - bp->b_resid));
1885 }
1886
1887 rf_DiskIOComplete(queue, req, (bp->b_flags & B_ERROR) ? 1 : 0);
1888 (req->CompleteFunc) (req->argument, (bp->b_flags & B_ERROR) ? 1 : 0);
1889
1890 splx(s);
1891 }
1892
1893
1894
1895 /*
1896 * initialize a buf structure for doing an I/O in the kernel.
1897 */
1898 static void
1899 InitBP(bp, b_vp, rw_flag, dev, startSect, numSect, buf, cbFunc, cbArg,
1900 logBytesPerSector, b_proc)
1901 struct buf *bp;
1902 struct vnode *b_vp;
1903 unsigned rw_flag;
1904 dev_t dev;
1905 RF_SectorNum_t startSect;
1906 RF_SectorCount_t numSect;
1907 caddr_t buf;
1908 void (*cbFunc) (struct buf *);
1909 void *cbArg;
1910 int logBytesPerSector;
1911 struct proc *b_proc;
1912 {
1913 /* bp->b_flags = B_PHYS | rw_flag; */
1914 bp->b_flags = B_CALL | rw_flag; /* XXX need B_PHYS here too??? */
1915 bp->b_bcount = numSect << logBytesPerSector;
1916 bp->b_bufsize = bp->b_bcount;
1917 bp->b_error = 0;
1918 bp->b_dev = dev;
1919 bp->b_data = buf;
1920 bp->b_blkno = startSect;
1921 bp->b_resid = bp->b_bcount; /* XXX is this right!??!?!! */
1922 if (bp->b_bcount == 0) {
1923 panic("bp->b_bcount is zero in InitBP!!\n");
1924 }
1925 bp->b_proc = b_proc;
1926 bp->b_iodone = cbFunc;
1927 bp->b_vp = b_vp;
1928
1929 }
1930
1931 static void
1932 raidgetdefaultlabel(raidPtr, rs, lp)
1933 RF_Raid_t *raidPtr;
1934 struct raid_softc *rs;
1935 struct disklabel *lp;
1936 {
1937 db1_printf(("Building a default label...\n"));
1938 bzero(lp, sizeof(*lp));
1939
1940 /* fabricate a label... */
1941 lp->d_secperunit = raidPtr->totalSectors;
1942 lp->d_secsize = raidPtr->bytesPerSector;
1943 lp->d_nsectors = raidPtr->Layout.dataSectorsPerStripe;
1944 lp->d_ntracks = 1;
1945 lp->d_ncylinders = raidPtr->totalSectors /
1946 (lp->d_nsectors * lp->d_ntracks);
1947 lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors;
1948
1949 strncpy(lp->d_typename, "raid", sizeof(lp->d_typename));
1950 lp->d_type = DTYPE_RAID;
1951 strncpy(lp->d_packname, "fictitious", sizeof(lp->d_packname));
1952 lp->d_rpm = 3600;
1953 lp->d_interleave = 1;
1954 lp->d_flags = 0;
1955
1956 lp->d_partitions[RAW_PART].p_offset = 0;
1957 lp->d_partitions[RAW_PART].p_size = raidPtr->totalSectors;
1958 lp->d_partitions[RAW_PART].p_fstype = FS_UNUSED;
1959 lp->d_npartitions = RAW_PART + 1;
1960
1961 lp->d_magic = DISKMAGIC;
1962 lp->d_magic2 = DISKMAGIC;
1963 lp->d_checksum = dkcksum(rs->sc_dkdev.dk_label);
1964
1965 }
1966 /*
1967 * Read the disklabel from the raid device. If one is not present, fake one
1968 * up.
1969 */
1970 static void
1971 raidgetdisklabel(dev)
1972 dev_t dev;
1973 {
1974 int unit = raidunit(dev);
1975 struct raid_softc *rs = &raid_softc[unit];
1976 char *errstring;
1977 struct disklabel *lp = rs->sc_dkdev.dk_label;
1978 struct cpu_disklabel *clp = rs->sc_dkdev.dk_cpulabel;
1979 RF_Raid_t *raidPtr;
1980
1981 db1_printf(("Getting the disklabel...\n"));
1982
1983 bzero(clp, sizeof(*clp));
1984
1985 raidPtr = raidPtrs[unit];
1986
1987 raidgetdefaultlabel(raidPtr, rs, lp);
1988
1989 /*
1990 * Call the generic disklabel extraction routine.
1991 */
1992 errstring = readdisklabel(RAIDLABELDEV(dev), raidstrategy,
1993 rs->sc_dkdev.dk_label, rs->sc_dkdev.dk_cpulabel);
1994 if (errstring)
1995 raidmakedisklabel(rs);
1996 else {
1997 int i;
1998 struct partition *pp;
1999
2000 /*
2001 * Sanity check whether the found disklabel is valid.
2002 *
2003 * This is necessary since total size of the raid device
2004 * may vary when an interleave is changed even though exactly
2005 * same componets are used, and old disklabel may used
2006 * if that is found.
2007 */
2008 if (lp->d_secperunit != rs->sc_size)
2009 printf("WARNING: %s: "
2010 "total sector size in disklabel (%d) != "
2011 "the size of raid (%ld)\n", rs->sc_xname,
2012 lp->d_secperunit, (long) rs->sc_size);
2013 for (i = 0; i < lp->d_npartitions; i++) {
2014 pp = &lp->d_partitions[i];
2015 if (pp->p_offset + pp->p_size > rs->sc_size)
2016 printf("WARNING: %s: end of partition `%c' "
2017 "exceeds the size of raid (%ld)\n",
2018 rs->sc_xname, 'a' + i, (long) rs->sc_size);
2019 }
2020 }
2021
2022 }
2023 /*
2024 * Take care of things one might want to take care of in the event
2025 * that a disklabel isn't present.
2026 */
2027 static void
2028 raidmakedisklabel(rs)
2029 struct raid_softc *rs;
2030 {
2031 struct disklabel *lp = rs->sc_dkdev.dk_label;
2032 db1_printf(("Making a label..\n"));
2033
2034 /*
2035 * For historical reasons, if there's no disklabel present
2036 * the raw partition must be marked FS_BSDFFS.
2037 */
2038
2039 lp->d_partitions[RAW_PART].p_fstype = FS_BSDFFS;
2040
2041 strncpy(lp->d_packname, "default label", sizeof(lp->d_packname));
2042
2043 lp->d_checksum = dkcksum(lp);
2044 }
2045 /*
2046 * Lookup the provided name in the filesystem. If the file exists,
2047 * is a valid block device, and isn't being used by anyone else,
2048 * set *vpp to the file's vnode.
2049 * You'll find the original of this in ccd.c
2050 */
2051 int
2052 raidlookup(path, p, vpp)
2053 char *path;
2054 struct proc *p;
2055 struct vnode **vpp; /* result */
2056 {
2057 struct nameidata nd;
2058 struct vnode *vp;
2059 struct vattr va;
2060 int error;
2061
2062 NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, path, p);
2063 if ((error = vn_open(&nd, FREAD | FWRITE, 0)) != 0) {
2064 #ifdef DEBUG
2065 printf("RAIDframe: vn_open returned %d\n", error);
2066 #endif
2067 return (error);
2068 }
2069 vp = nd.ni_vp;
2070 if (vp->v_usecount > 1) {
2071 VOP_UNLOCK(vp, 0);
2072 (void) vn_close(vp, FREAD | FWRITE, p->p_ucred, p);
2073 return (EBUSY);
2074 }
2075 if ((error = VOP_GETATTR(vp, &va, p->p_ucred, p)) != 0) {
2076 VOP_UNLOCK(vp, 0);
2077 (void) vn_close(vp, FREAD | FWRITE, p->p_ucred, p);
2078 return (error);
2079 }
2080 /* XXX: eventually we should handle VREG, too. */
2081 if (va.va_type != VBLK) {
2082 VOP_UNLOCK(vp, 0);
2083 (void) vn_close(vp, FREAD | FWRITE, p->p_ucred, p);
2084 return (ENOTBLK);
2085 }
2086 VOP_UNLOCK(vp, 0);
2087 *vpp = vp;
2088 return (0);
2089 }
2090 /*
2091 * Wait interruptibly for an exclusive lock.
2092 *
2093 * XXX
2094 * Several drivers do this; it should be abstracted and made MP-safe.
2095 * (Hmm... where have we seen this warning before :-> GO )
2096 */
2097 static int
2098 raidlock(rs)
2099 struct raid_softc *rs;
2100 {
2101 int error;
2102
2103 while ((rs->sc_flags & RAIDF_LOCKED) != 0) {
2104 rs->sc_flags |= RAIDF_WANTED;
2105 if ((error =
2106 tsleep(rs, PRIBIO | PCATCH, "raidlck", 0)) != 0)
2107 return (error);
2108 }
2109 rs->sc_flags |= RAIDF_LOCKED;
2110 return (0);
2111 }
2112 /*
2113 * Unlock and wake up any waiters.
2114 */
2115 static void
2116 raidunlock(rs)
2117 struct raid_softc *rs;
2118 {
2119
2120 rs->sc_flags &= ~RAIDF_LOCKED;
2121 if ((rs->sc_flags & RAIDF_WANTED) != 0) {
2122 rs->sc_flags &= ~RAIDF_WANTED;
2123 wakeup(rs);
2124 }
2125 }
2126
2127
2128 #define RF_COMPONENT_INFO_OFFSET 16384 /* bytes */
2129 #define RF_COMPONENT_INFO_SIZE 1024 /* bytes */
2130
2131 int
2132 raidmarkclean(dev_t dev, struct vnode *b_vp, int mod_counter)
2133 {
2134 RF_ComponentLabel_t clabel;
2135 raidread_component_label(dev, b_vp, &clabel);
2136 clabel.mod_counter = mod_counter;
2137 clabel.clean = RF_RAID_CLEAN;
2138 raidwrite_component_label(dev, b_vp, &clabel);
2139 return(0);
2140 }
2141
2142
2143 int
2144 raidmarkdirty(dev_t dev, struct vnode *b_vp, int mod_counter)
2145 {
2146 RF_ComponentLabel_t clabel;
2147 raidread_component_label(dev, b_vp, &clabel);
2148 clabel.mod_counter = mod_counter;
2149 clabel.clean = RF_RAID_DIRTY;
2150 raidwrite_component_label(dev, b_vp, &clabel);
2151 return(0);
2152 }
2153
2154 /* ARGSUSED */
2155 int
2156 raidread_component_label(dev, b_vp, clabel)
2157 dev_t dev;
2158 struct vnode *b_vp;
2159 RF_ComponentLabel_t *clabel;
2160 {
2161 struct buf *bp;
2162 int error;
2163
2164 /* XXX should probably ensure that we don't try to do this if
2165 someone has changed rf_protected_sectors. */
2166
2167 /* get a block of the appropriate size... */
2168 bp = geteblk((int)RF_COMPONENT_INFO_SIZE);
2169 bp->b_dev = dev;
2170
2171 /* get our ducks in a row for the read */
2172 bp->b_blkno = RF_COMPONENT_INFO_OFFSET / DEV_BSIZE;
2173 bp->b_bcount = RF_COMPONENT_INFO_SIZE;
2174 bp->b_flags = B_BUSY | B_READ;
2175 bp->b_resid = RF_COMPONENT_INFO_SIZE / DEV_BSIZE;
2176
2177 (*bdevsw[major(bp->b_dev)].d_strategy)(bp);
2178
2179 error = biowait(bp);
2180
2181 if (!error) {
2182 memcpy(clabel, bp->b_data,
2183 sizeof(RF_ComponentLabel_t));
2184 #if 0
2185 rf_print_component_label( clabel );
2186 #endif
2187 } else {
2188 #if 0
2189 printf("Failed to read RAID component label!\n");
2190 #endif
2191 }
2192
2193 bp->b_flags = B_INVAL | B_AGE;
2194 brelse(bp);
2195 return(error);
2196 }
2197 /* ARGSUSED */
2198 int
2199 raidwrite_component_label(dev, b_vp, clabel)
2200 dev_t dev;
2201 struct vnode *b_vp;
2202 RF_ComponentLabel_t *clabel;
2203 {
2204 struct buf *bp;
2205 int error;
2206
2207 /* get a block of the appropriate size... */
2208 bp = geteblk((int)RF_COMPONENT_INFO_SIZE);
2209 bp->b_dev = dev;
2210
2211 /* get our ducks in a row for the write */
2212 bp->b_blkno = RF_COMPONENT_INFO_OFFSET / DEV_BSIZE;
2213 bp->b_bcount = RF_COMPONENT_INFO_SIZE;
2214 bp->b_flags = B_BUSY | B_WRITE;
2215 bp->b_resid = RF_COMPONENT_INFO_SIZE / DEV_BSIZE;
2216
2217 memset(bp->b_data, 0, RF_COMPONENT_INFO_SIZE );
2218
2219 memcpy(bp->b_data, clabel, sizeof(RF_ComponentLabel_t));
2220
2221 (*bdevsw[major(bp->b_dev)].d_strategy)(bp);
2222 error = biowait(bp);
2223 bp->b_flags = B_INVAL | B_AGE;
2224 brelse(bp);
2225 if (error) {
2226 #if 1
2227 printf("Failed to write RAID component info!\n");
2228 #endif
2229 }
2230
2231 return(error);
2232 }
2233
2234 void
2235 rf_markalldirty(raidPtr)
2236 RF_Raid_t *raidPtr;
2237 {
2238 RF_ComponentLabel_t clabel;
2239 int r,c;
2240
2241 raidPtr->mod_counter++;
2242 for (r = 0; r < raidPtr->numRow; r++) {
2243 for (c = 0; c < raidPtr->numCol; c++) {
2244 if (raidPtr->Disks[r][c].status != rf_ds_failed) {
2245 raidread_component_label(
2246 raidPtr->Disks[r][c].dev,
2247 raidPtr->raid_cinfo[r][c].ci_vp,
2248 &clabel);
2249 if (clabel.status == rf_ds_spared) {
2250 /* XXX do something special...
2251 but whatever you do, don't
2252 try to access it!! */
2253 } else {
2254 #if 0
2255 clabel.status =
2256 raidPtr->Disks[r][c].status;
2257 raidwrite_component_label(
2258 raidPtr->Disks[r][c].dev,
2259 raidPtr->raid_cinfo[r][c].ci_vp,
2260 &clabel);
2261 #endif
2262 raidmarkdirty(
2263 raidPtr->Disks[r][c].dev,
2264 raidPtr->raid_cinfo[r][c].ci_vp,
2265 raidPtr->mod_counter);
2266 }
2267 }
2268 }
2269 }
2270 /* printf("Component labels marked dirty.\n"); */
2271 #if 0
2272 for( c = 0; c < raidPtr->numSpare ; c++) {
2273 sparecol = raidPtr->numCol + c;
2274 if (raidPtr->Disks[r][sparecol].status == rf_ds_used_spare) {
2275 /*
2276
2277 XXX this is where we get fancy and map this spare
2278 into it's correct spot in the array.
2279
2280 */
2281 /*
2282
2283 we claim this disk is "optimal" if it's
2284 rf_ds_used_spare, as that means it should be
2285 directly substitutable for the disk it replaced.
2286 We note that too...
2287
2288 */
2289
2290 for(i=0;i<raidPtr->numRow;i++) {
2291 for(j=0;j<raidPtr->numCol;j++) {
2292 if ((raidPtr->Disks[i][j].spareRow ==
2293 r) &&
2294 (raidPtr->Disks[i][j].spareCol ==
2295 sparecol)) {
2296 srow = r;
2297 scol = sparecol;
2298 break;
2299 }
2300 }
2301 }
2302
2303 raidread_component_label(
2304 raidPtr->Disks[r][sparecol].dev,
2305 raidPtr->raid_cinfo[r][sparecol].ci_vp,
2306 &clabel);
2307 /* make sure status is noted */
2308 clabel.version = RF_COMPONENT_LABEL_VERSION;
2309 clabel.mod_counter = raidPtr->mod_counter;
2310 clabel.serial_number = raidPtr->serial_number;
2311 clabel.row = srow;
2312 clabel.column = scol;
2313 clabel.num_rows = raidPtr->numRow;
2314 clabel.num_columns = raidPtr->numCol;
2315 clabel.clean = RF_RAID_DIRTY; /* changed in a bit*/
2316 clabel.status = rf_ds_optimal;
2317 raidwrite_component_label(
2318 raidPtr->Disks[r][sparecol].dev,
2319 raidPtr->raid_cinfo[r][sparecol].ci_vp,
2320 &clabel);
2321 raidmarkclean( raidPtr->Disks[r][sparecol].dev,
2322 raidPtr->raid_cinfo[r][sparecol].ci_vp);
2323 }
2324 }
2325
2326 #endif
2327 }
2328
2329
2330 void
2331 rf_update_component_labels(raidPtr, final)
2332 RF_Raid_t *raidPtr;
2333 int final;
2334 {
2335 RF_ComponentLabel_t clabel;
2336 int sparecol;
2337 int r,c;
2338 int i,j;
2339 int srow, scol;
2340
2341 srow = -1;
2342 scol = -1;
2343
2344 /* XXX should do extra checks to make sure things really are clean,
2345 rather than blindly setting the clean bit... */
2346
2347 raidPtr->mod_counter++;
2348
2349 for (r = 0; r < raidPtr->numRow; r++) {
2350 for (c = 0; c < raidPtr->numCol; c++) {
2351 if (raidPtr->Disks[r][c].status == rf_ds_optimal) {
2352 raidread_component_label(
2353 raidPtr->Disks[r][c].dev,
2354 raidPtr->raid_cinfo[r][c].ci_vp,
2355 &clabel);
2356 /* make sure status is noted */
2357 clabel.status = rf_ds_optimal;
2358 /* bump the counter */
2359 clabel.mod_counter = raidPtr->mod_counter;
2360
2361 raidwrite_component_label(
2362 raidPtr->Disks[r][c].dev,
2363 raidPtr->raid_cinfo[r][c].ci_vp,
2364 &clabel);
2365 if (final == RF_FINAL_COMPONENT_UPDATE) {
2366 if (raidPtr->parity_good == RF_RAID_CLEAN) {
2367 raidmarkclean(
2368 raidPtr->Disks[r][c].dev,
2369 raidPtr->raid_cinfo[r][c].ci_vp,
2370 raidPtr->mod_counter);
2371 }
2372 }
2373 }
2374 /* else we don't touch it.. */
2375 }
2376 }
2377
2378 for( c = 0; c < raidPtr->numSpare ; c++) {
2379 sparecol = raidPtr->numCol + c;
2380 if (raidPtr->Disks[0][sparecol].status == rf_ds_used_spare) {
2381 /*
2382
2383 we claim this disk is "optimal" if it's
2384 rf_ds_used_spare, as that means it should be
2385 directly substitutable for the disk it replaced.
2386 We note that too...
2387
2388 */
2389
2390 for(i=0;i<raidPtr->numRow;i++) {
2391 for(j=0;j<raidPtr->numCol;j++) {
2392 if ((raidPtr->Disks[i][j].spareRow ==
2393 0) &&
2394 (raidPtr->Disks[i][j].spareCol ==
2395 sparecol)) {
2396 srow = i;
2397 scol = j;
2398 break;
2399 }
2400 }
2401 }
2402
2403 /* XXX shouldn't *really* need this... */
2404 raidread_component_label(
2405 raidPtr->Disks[0][sparecol].dev,
2406 raidPtr->raid_cinfo[0][sparecol].ci_vp,
2407 &clabel);
2408 /* make sure status is noted */
2409
2410 raid_init_component_label(raidPtr, &clabel);
2411
2412 clabel.mod_counter = raidPtr->mod_counter;
2413 clabel.row = srow;
2414 clabel.column = scol;
2415 clabel.status = rf_ds_optimal;
2416
2417 raidwrite_component_label(
2418 raidPtr->Disks[0][sparecol].dev,
2419 raidPtr->raid_cinfo[0][sparecol].ci_vp,
2420 &clabel);
2421 if (final == RF_FINAL_COMPONENT_UPDATE) {
2422 if (raidPtr->parity_good == RF_RAID_CLEAN) {
2423 raidmarkclean( raidPtr->Disks[0][sparecol].dev,
2424 raidPtr->raid_cinfo[0][sparecol].ci_vp,
2425 raidPtr->mod_counter);
2426 }
2427 }
2428 }
2429 }
2430 /* printf("Component labels updated\n"); */
2431 }
2432
2433 void
2434 rf_close_component(raidPtr, vp, auto_configured)
2435 RF_Raid_t *raidPtr;
2436 struct vnode *vp;
2437 int auto_configured;
2438 {
2439 struct proc *p;
2440
2441 p = raidPtr->engine_thread;
2442
2443 if (vp != NULL) {
2444 if (auto_configured == 1) {
2445 VOP_CLOSE(vp, FREAD, NOCRED, 0);
2446 vput(vp);
2447
2448 } else {
2449 VOP_UNLOCK(vp, 0);
2450 (void) vn_close(vp, FREAD | FWRITE, p->p_ucred, p);
2451 }
2452 } else {
2453 printf("vnode was NULL\n");
2454 }
2455 }
2456
2457
2458 void
2459 rf_UnconfigureVnodes(raidPtr)
2460 RF_Raid_t *raidPtr;
2461 {
2462 int r,c;
2463 struct proc *p;
2464 struct vnode *vp;
2465 int acd;
2466
2467
2468 /* We take this opportunity to close the vnodes like we should.. */
2469
2470 p = raidPtr->engine_thread;
2471
2472 for (r = 0; r < raidPtr->numRow; r++) {
2473 for (c = 0; c < raidPtr->numCol; c++) {
2474 printf("Closing vnode for row: %d col: %d\n", r, c);
2475 vp = raidPtr->raid_cinfo[r][c].ci_vp;
2476 acd = raidPtr->Disks[r][c].auto_configured;
2477 rf_close_component(raidPtr, vp, acd);
2478 raidPtr->raid_cinfo[r][c].ci_vp = NULL;
2479 raidPtr->Disks[r][c].auto_configured = 0;
2480 }
2481 }
2482 for (r = 0; r < raidPtr->numSpare; r++) {
2483 printf("Closing vnode for spare: %d\n", r);
2484 vp = raidPtr->raid_cinfo[0][raidPtr->numCol + r].ci_vp;
2485 acd = raidPtr->Disks[0][raidPtr->numCol + r].auto_configured;
2486 rf_close_component(raidPtr, vp, acd);
2487 raidPtr->raid_cinfo[0][raidPtr->numCol + r].ci_vp = NULL;
2488 raidPtr->Disks[0][raidPtr->numCol + r].auto_configured = 0;
2489 }
2490 }
2491
2492
2493 void
2494 rf_ReconThread(req)
2495 struct rf_recon_req *req;
2496 {
2497 int s;
2498 RF_Raid_t *raidPtr;
2499
2500 s = splbio();
2501 raidPtr = (RF_Raid_t *) req->raidPtr;
2502 raidPtr->recon_in_progress = 1;
2503
2504 rf_FailDisk((RF_Raid_t *) req->raidPtr, req->row, req->col,
2505 ((req->flags & RF_FDFLAGS_RECON) ? 1 : 0));
2506
2507 /* XXX get rid of this! we don't need it at all.. */
2508 RF_Free(req, sizeof(*req));
2509
2510 raidPtr->recon_in_progress = 0;
2511 splx(s);
2512
2513 /* That's all... */
2514 kthread_exit(0); /* does not return */
2515 }
2516
2517 void
2518 rf_RewriteParityThread(raidPtr)
2519 RF_Raid_t *raidPtr;
2520 {
2521 int retcode;
2522 int s;
2523
2524 raidPtr->parity_rewrite_in_progress = 1;
2525 s = splbio();
2526 retcode = rf_RewriteParity(raidPtr);
2527 splx(s);
2528 if (retcode) {
2529 printf("raid%d: Error re-writing parity!\n",raidPtr->raidid);
2530 } else {
2531 /* set the clean bit! If we shutdown correctly,
2532 the clean bit on each component label will get
2533 set */
2534 raidPtr->parity_good = RF_RAID_CLEAN;
2535 }
2536 raidPtr->parity_rewrite_in_progress = 0;
2537
2538 /* Anyone waiting for us to stop? If so, inform them... */
2539 if (raidPtr->waitShutdown) {
2540 wakeup(&raidPtr->parity_rewrite_in_progress);
2541 }
2542
2543 /* That's all... */
2544 kthread_exit(0); /* does not return */
2545 }
2546
2547
2548 void
2549 rf_CopybackThread(raidPtr)
2550 RF_Raid_t *raidPtr;
2551 {
2552 int s;
2553
2554 raidPtr->copyback_in_progress = 1;
2555 s = splbio();
2556 rf_CopybackReconstructedData(raidPtr);
2557 splx(s);
2558 raidPtr->copyback_in_progress = 0;
2559
2560 /* That's all... */
2561 kthread_exit(0); /* does not return */
2562 }
2563
2564
2565 void
2566 rf_ReconstructInPlaceThread(req)
2567 struct rf_recon_req *req;
2568 {
2569 int retcode;
2570 int s;
2571 RF_Raid_t *raidPtr;
2572
2573 s = splbio();
2574 raidPtr = req->raidPtr;
2575 raidPtr->recon_in_progress = 1;
2576 retcode = rf_ReconstructInPlace(raidPtr, req->row, req->col);
2577 RF_Free(req, sizeof(*req));
2578 raidPtr->recon_in_progress = 0;
2579 splx(s);
2580
2581 /* That's all... */
2582 kthread_exit(0); /* does not return */
2583 }
2584
2585 void
2586 rf_mountroot_hook(dev)
2587 struct device *dev;
2588 {
2589
2590 }
2591
2592
2593 RF_AutoConfig_t *
2594 rf_find_raid_components()
2595 {
2596 struct devnametobdevmaj *dtobdm;
2597 struct vnode *vp;
2598 struct disklabel label;
2599 struct device *dv;
2600 char *cd_name;
2601 dev_t dev;
2602 int error;
2603 int i;
2604 int good_one;
2605 RF_ComponentLabel_t *clabel;
2606 RF_AutoConfig_t *ac_list;
2607 RF_AutoConfig_t *ac;
2608
2609
2610 /* initialize the AutoConfig list */
2611 ac_list = NULL;
2612
2613 if (raidautoconfig) {
2614
2615 /* we begin by trolling through *all* the devices on the system */
2616
2617 for (dv = alldevs.tqh_first; dv != NULL;
2618 dv = dv->dv_list.tqe_next) {
2619
2620 /* we are only interested in disks... */
2621 if (dv->dv_class != DV_DISK)
2622 continue;
2623
2624 /* we don't care about floppies... */
2625 if (!strcmp(dv->dv_cfdata->cf_driver->cd_name,"fd")) {
2626 continue;
2627 }
2628
2629 /* need to find the device_name_to_block_device_major stuff */
2630 cd_name = dv->dv_cfdata->cf_driver->cd_name;
2631 dtobdm = dev_name2blk;
2632 while (dtobdm->d_name && strcmp(dtobdm->d_name, cd_name)) {
2633 dtobdm++;
2634 }
2635
2636 /* get a vnode for the raw partition of this disk */
2637
2638 dev = MAKEDISKDEV(dtobdm->d_maj, dv->dv_unit, RAW_PART);
2639 if (bdevvp(dev, &vp))
2640 panic("RAID can't alloc vnode");
2641
2642 error = VOP_OPEN(vp, FREAD, NOCRED, 0);
2643
2644 if (error) {
2645 /* "Who cares." Continue looking
2646 for something that exists*/
2647 vput(vp);
2648 continue;
2649 }
2650
2651 /* Ok, the disk exists. Go get the disklabel. */
2652 error = VOP_IOCTL(vp, DIOCGDINFO, (caddr_t)&label,
2653 FREAD, NOCRED, 0);
2654 if (error) {
2655 /*
2656 * XXX can't happen - open() would
2657 * have errored out (or faked up one)
2658 */
2659 printf("can't get label for dev %s%c (%d)!?!?\n",
2660 dv->dv_xname, 'a' + RAW_PART, error);
2661 }
2662
2663 /* don't need this any more. We'll allocate it again
2664 a little later if we really do... */
2665 VOP_CLOSE(vp, FREAD, NOCRED, 0);
2666 vput(vp);
2667
2668 for (i=0; i < label.d_npartitions; i++) {
2669 /* We only support partitions marked as RAID */
2670 if (label.d_partitions[i].p_fstype != FS_RAID)
2671 continue;
2672
2673 dev = MAKEDISKDEV(dtobdm->d_maj, dv->dv_unit, i);
2674 if (bdevvp(dev, &vp))
2675 panic("RAID can't alloc vnode");
2676
2677 error = VOP_OPEN(vp, FREAD, NOCRED, 0);
2678 if (error) {
2679 /* Whatever... */
2680 vput(vp);
2681 continue;
2682 }
2683
2684 good_one = 0;
2685
2686 clabel = (RF_ComponentLabel_t *)
2687 malloc(sizeof(RF_ComponentLabel_t),
2688 M_RAIDFRAME, M_NOWAIT);
2689 if (clabel == NULL) {
2690 /* XXX CLEANUP HERE */
2691 printf("RAID auto config: out of memory!\n");
2692 return(NULL); /* XXX probably should panic? */
2693 }
2694
2695 if (!raidread_component_label(dev, vp, clabel)) {
2696 /* Got the label. Does it look reasonable? */
2697 if (rf_reasonable_label(clabel) &&
2698 (clabel->partitionSize <=
2699 label.d_partitions[i].p_size)) {
2700 #if DEBUG
2701 printf("Component on: %s%c: %d\n",
2702 dv->dv_xname, 'a'+i,
2703 label.d_partitions[i].p_size);
2704 rf_print_component_label(clabel);
2705 #endif
2706 /* if it's reasonable, add it,
2707 else ignore it. */
2708 ac = (RF_AutoConfig_t *)
2709 malloc(sizeof(RF_AutoConfig_t),
2710 M_RAIDFRAME,
2711 M_NOWAIT);
2712 if (ac == NULL) {
2713 /* XXX should panic?? */
2714 return(NULL);
2715 }
2716
2717 sprintf(ac->devname, "%s%c",
2718 dv->dv_xname, 'a'+i);
2719 ac->dev = dev;
2720 ac->vp = vp;
2721 ac->clabel = clabel;
2722 ac->next = ac_list;
2723 ac_list = ac;
2724 good_one = 1;
2725 }
2726 }
2727 if (!good_one) {
2728 /* cleanup */
2729 free(clabel, M_RAIDFRAME);
2730 VOP_CLOSE(vp, FREAD, NOCRED, 0);
2731 vput(vp);
2732 }
2733 }
2734 }
2735 }
2736 return(ac_list);
2737 }
2738
2739 static int
2740 rf_reasonable_label(clabel)
2741 RF_ComponentLabel_t *clabel;
2742 {
2743
2744 if (((clabel->version==RF_COMPONENT_LABEL_VERSION_1) ||
2745 (clabel->version==RF_COMPONENT_LABEL_VERSION)) &&
2746 ((clabel->clean == RF_RAID_CLEAN) ||
2747 (clabel->clean == RF_RAID_DIRTY)) &&
2748 clabel->row >=0 &&
2749 clabel->column >= 0 &&
2750 clabel->num_rows > 0 &&
2751 clabel->num_columns > 0 &&
2752 clabel->row < clabel->num_rows &&
2753 clabel->column < clabel->num_columns &&
2754 clabel->blockSize > 0 &&
2755 clabel->numBlocks > 0) {
2756 /* label looks reasonable enough... */
2757 return(1);
2758 }
2759 return(0);
2760 }
2761
2762
2763 void
2764 rf_print_component_label(clabel)
2765 RF_ComponentLabel_t *clabel;
2766 {
2767 printf(" Row: %d Column: %d Num Rows: %d Num Columns: %d\n",
2768 clabel->row, clabel->column,
2769 clabel->num_rows, clabel->num_columns);
2770 printf(" Version: %d Serial Number: %d Mod Counter: %d\n",
2771 clabel->version, clabel->serial_number,
2772 clabel->mod_counter);
2773 printf(" Clean: %s Status: %d\n",
2774 clabel->clean ? "Yes" : "No", clabel->status );
2775 printf(" sectPerSU: %d SUsPerPU: %d SUsPerRU: %d\n",
2776 clabel->sectPerSU, clabel->SUsPerPU, clabel->SUsPerRU);
2777 printf(" RAID Level: %c blocksize: %d numBlocks: %d\n",
2778 (char) clabel->parityConfig, clabel->blockSize,
2779 clabel->numBlocks);
2780 printf(" Autoconfig: %s\n", clabel->autoconfigure ? "Yes" : "No" );
2781 printf(" Contains root partition: %s\n",
2782 clabel->root_partition ? "Yes" : "No" );
2783 printf(" Last configured as: raid%d\n", clabel->last_unit );
2784 #if 0
2785 printf(" Config order: %d\n", clabel->config_order);
2786 #endif
2787
2788 }
2789
2790 RF_ConfigSet_t *
2791 rf_create_auto_sets(ac_list)
2792 RF_AutoConfig_t *ac_list;
2793 {
2794 RF_AutoConfig_t *ac;
2795 RF_ConfigSet_t *config_sets;
2796 RF_ConfigSet_t *cset;
2797 RF_AutoConfig_t *ac_next;
2798
2799
2800 config_sets = NULL;
2801
2802 /* Go through the AutoConfig list, and figure out which components
2803 belong to what sets. */
2804 ac = ac_list;
2805 while(ac!=NULL) {
2806 /* we're going to putz with ac->next, so save it here
2807 for use at the end of the loop */
2808 ac_next = ac->next;
2809
2810 if (config_sets == NULL) {
2811 /* will need at least this one... */
2812 config_sets = (RF_ConfigSet_t *)
2813 malloc(sizeof(RF_ConfigSet_t),
2814 M_RAIDFRAME, M_NOWAIT);
2815 if (config_sets == NULL) {
2816 panic("rf_create_auto_sets: No memory!\n");
2817 }
2818 /* this one is easy :) */
2819 config_sets->ac = ac;
2820 config_sets->next = NULL;
2821 config_sets->rootable = 0;
2822 ac->next = NULL;
2823 } else {
2824 /* which set does this component fit into? */
2825 cset = config_sets;
2826 while(cset!=NULL) {
2827 if (rf_does_it_fit(cset, ac)) {
2828 /* looks like it matches... */
2829 ac->next = cset->ac;
2830 cset->ac = ac;
2831 break;
2832 }
2833 cset = cset->next;
2834 }
2835 if (cset==NULL) {
2836 /* didn't find a match above... new set..*/
2837 cset = (RF_ConfigSet_t *)
2838 malloc(sizeof(RF_ConfigSet_t),
2839 M_RAIDFRAME, M_NOWAIT);
2840 if (cset == NULL) {
2841 panic("rf_create_auto_sets: No memory!\n");
2842 }
2843 cset->ac = ac;
2844 ac->next = NULL;
2845 cset->next = config_sets;
2846 cset->rootable = 0;
2847 config_sets = cset;
2848 }
2849 }
2850 ac = ac_next;
2851 }
2852
2853
2854 return(config_sets);
2855 }
2856
2857 static int
2858 rf_does_it_fit(cset, ac)
2859 RF_ConfigSet_t *cset;
2860 RF_AutoConfig_t *ac;
2861 {
2862 RF_ComponentLabel_t *clabel1, *clabel2;
2863
2864 /* If this one matches the *first* one in the set, that's good
2865 enough, since the other members of the set would have been
2866 through here too... */
2867 /* note that we are not checking partitionSize here..
2868
2869 Note that we are also not checking the mod_counters here.
2870 If everything else matches execpt the mod_counter, that's
2871 good enough for this test. We will deal with the mod_counters
2872 a little later in the autoconfiguration process.
2873
2874 (clabel1->mod_counter == clabel2->mod_counter) &&
2875
2876 The reason we don't check for this is that failed disks
2877 will have lower modification counts. If those disks are
2878 not added to the set they used to belong to, then they will
2879 form their own set, which may result in 2 different sets,
2880 for example, competing to be configured at raid0, and
2881 perhaps competing to be the root filesystem set. If the
2882 wrong ones get configured, or both attempt to become /,
2883 weird behaviour and or serious lossage will occur. Thus we
2884 need to bring them into the fold here, and kick them out at
2885 a later point.
2886
2887 */
2888
2889 clabel1 = cset->ac->clabel;
2890 clabel2 = ac->clabel;
2891 if ((clabel1->version == clabel2->version) &&
2892 (clabel1->serial_number == clabel2->serial_number) &&
2893 (clabel1->num_rows == clabel2->num_rows) &&
2894 (clabel1->num_columns == clabel2->num_columns) &&
2895 (clabel1->sectPerSU == clabel2->sectPerSU) &&
2896 (clabel1->SUsPerPU == clabel2->SUsPerPU) &&
2897 (clabel1->SUsPerRU == clabel2->SUsPerRU) &&
2898 (clabel1->parityConfig == clabel2->parityConfig) &&
2899 (clabel1->maxOutstanding == clabel2->maxOutstanding) &&
2900 (clabel1->blockSize == clabel2->blockSize) &&
2901 (clabel1->numBlocks == clabel2->numBlocks) &&
2902 (clabel1->autoconfigure == clabel2->autoconfigure) &&
2903 (clabel1->root_partition == clabel2->root_partition) &&
2904 (clabel1->last_unit == clabel2->last_unit) &&
2905 (clabel1->config_order == clabel2->config_order)) {
2906 /* if it get's here, it almost *has* to be a match */
2907 } else {
2908 /* it's not consistent with somebody in the set..
2909 punt */
2910 return(0);
2911 }
2912 /* all was fine.. it must fit... */
2913 return(1);
2914 }
2915
2916 int
2917 rf_have_enough_components(cset)
2918 RF_ConfigSet_t *cset;
2919 {
2920 RF_AutoConfig_t *ac;
2921 RF_AutoConfig_t *auto_config;
2922 RF_ComponentLabel_t *clabel;
2923 int r,c;
2924 int num_rows;
2925 int num_cols;
2926 int num_missing;
2927 int mod_counter;
2928 int mod_counter_found;
2929 int even_pair_failed;
2930 char parity_type;
2931
2932
2933 /* check to see that we have enough 'live' components
2934 of this set. If so, we can configure it if necessary */
2935
2936 num_rows = cset->ac->clabel->num_rows;
2937 num_cols = cset->ac->clabel->num_columns;
2938 parity_type = cset->ac->clabel->parityConfig;
2939
2940 /* XXX Check for duplicate components!?!?!? */
2941
2942 /* Determine what the mod_counter is supposed to be for this set. */
2943
2944 mod_counter_found = 0;
2945 ac = cset->ac;
2946 while(ac!=NULL) {
2947 if (mod_counter_found==0) {
2948 mod_counter = ac->clabel->mod_counter;
2949 mod_counter_found = 1;
2950 } else {
2951 if (ac->clabel->mod_counter > mod_counter) {
2952 mod_counter = ac->clabel->mod_counter;
2953 }
2954 }
2955 ac = ac->next;
2956 }
2957
2958 num_missing = 0;
2959 auto_config = cset->ac;
2960
2961 for(r=0; r<num_rows; r++) {
2962 even_pair_failed = 0;
2963 for(c=0; c<num_cols; c++) {
2964 ac = auto_config;
2965 while(ac!=NULL) {
2966 if ((ac->clabel->row == r) &&
2967 (ac->clabel->column == c) &&
2968 (ac->clabel->mod_counter == mod_counter)) {
2969 /* it's this one... */
2970 #if DEBUG
2971 printf("Found: %s at %d,%d\n",
2972 ac->devname,r,c);
2973 #endif
2974 break;
2975 }
2976 ac=ac->next;
2977 }
2978 if (ac==NULL) {
2979 /* Didn't find one here! */
2980 /* special case for RAID 1, especially
2981 where there are more than 2
2982 components (where RAIDframe treats
2983 things a little differently :( ) */
2984 if (parity_type == '1') {
2985 if (c%2 == 0) { /* even component */
2986 even_pair_failed = 1;
2987 } else { /* odd component. If
2988 we're failed, and
2989 so is the even
2990 component, it's
2991 "Good Night, Charlie" */
2992 if (even_pair_failed == 1) {
2993 return(0);
2994 }
2995 }
2996 } else {
2997 /* normal accounting */
2998 num_missing++;
2999 }
3000 }
3001 if ((parity_type == '1') && (c%2 == 1)) {
3002 /* Just did an even component, and we didn't
3003 bail.. reset the even_pair_failed flag,
3004 and go on to the next component.... */
3005 even_pair_failed = 0;
3006 }
3007 }
3008 }
3009
3010 clabel = cset->ac->clabel;
3011
3012 if (((clabel->parityConfig == '0') && (num_missing > 0)) ||
3013 ((clabel->parityConfig == '4') && (num_missing > 1)) ||
3014 ((clabel->parityConfig == '5') && (num_missing > 1))) {
3015 /* XXX this needs to be made *much* more general */
3016 /* Too many failures */
3017 return(0);
3018 }
3019 /* otherwise, all is well, and we've got enough to take a kick
3020 at autoconfiguring this set */
3021 return(1);
3022 }
3023
3024 void
3025 rf_create_configuration(ac,config,raidPtr)
3026 RF_AutoConfig_t *ac;
3027 RF_Config_t *config;
3028 RF_Raid_t *raidPtr;
3029 {
3030 RF_ComponentLabel_t *clabel;
3031 int i;
3032
3033 clabel = ac->clabel;
3034
3035 /* 1. Fill in the common stuff */
3036 config->numRow = clabel->num_rows;
3037 config->numCol = clabel->num_columns;
3038 config->numSpare = 0; /* XXX should this be set here? */
3039 config->sectPerSU = clabel->sectPerSU;
3040 config->SUsPerPU = clabel->SUsPerPU;
3041 config->SUsPerRU = clabel->SUsPerRU;
3042 config->parityConfig = clabel->parityConfig;
3043 /* XXX... */
3044 strcpy(config->diskQueueType,"fifo");
3045 config->maxOutstandingDiskReqs = clabel->maxOutstanding;
3046 config->layoutSpecificSize = 0; /* XXX ?? */
3047
3048 while(ac!=NULL) {
3049 /* row/col values will be in range due to the checks
3050 in reasonable_label() */
3051 strcpy(config->devnames[ac->clabel->row][ac->clabel->column],
3052 ac->devname);
3053 ac = ac->next;
3054 }
3055
3056 for(i=0;i<RF_MAXDBGV;i++) {
3057 config->debugVars[i][0] = NULL;
3058 }
3059 }
3060
3061 int
3062 rf_set_autoconfig(raidPtr, new_value)
3063 RF_Raid_t *raidPtr;
3064 int new_value;
3065 {
3066 RF_ComponentLabel_t clabel;
3067 struct vnode *vp;
3068 dev_t dev;
3069 int row, column;
3070
3071 raidPtr->autoconfigure = new_value;
3072 for(row=0; row<raidPtr->numRow; row++) {
3073 for(column=0; column<raidPtr->numCol; column++) {
3074 if (raidPtr->Disks[row][column].status ==
3075 rf_ds_optimal) {
3076 dev = raidPtr->Disks[row][column].dev;
3077 vp = raidPtr->raid_cinfo[row][column].ci_vp;
3078 raidread_component_label(dev, vp, &clabel);
3079 clabel.autoconfigure = new_value;
3080 raidwrite_component_label(dev, vp, &clabel);
3081 }
3082 }
3083 }
3084 return(new_value);
3085 }
3086
3087 int
3088 rf_set_rootpartition(raidPtr, new_value)
3089 RF_Raid_t *raidPtr;
3090 int new_value;
3091 {
3092 RF_ComponentLabel_t clabel;
3093 struct vnode *vp;
3094 dev_t dev;
3095 int row, column;
3096
3097 raidPtr->root_partition = new_value;
3098 for(row=0; row<raidPtr->numRow; row++) {
3099 for(column=0; column<raidPtr->numCol; column++) {
3100 if (raidPtr->Disks[row][column].status ==
3101 rf_ds_optimal) {
3102 dev = raidPtr->Disks[row][column].dev;
3103 vp = raidPtr->raid_cinfo[row][column].ci_vp;
3104 raidread_component_label(dev, vp, &clabel);
3105 clabel.root_partition = new_value;
3106 raidwrite_component_label(dev, vp, &clabel);
3107 }
3108 }
3109 }
3110 return(new_value);
3111 }
3112
3113 void
3114 rf_release_all_vps(cset)
3115 RF_ConfigSet_t *cset;
3116 {
3117 RF_AutoConfig_t *ac;
3118
3119 ac = cset->ac;
3120 while(ac!=NULL) {
3121 /* Close the vp, and give it back */
3122 if (ac->vp) {
3123 VOP_CLOSE(ac->vp, FREAD, NOCRED, 0);
3124 vput(ac->vp);
3125 ac->vp = NULL;
3126 }
3127 ac = ac->next;
3128 }
3129 }
3130
3131
3132 void
3133 rf_cleanup_config_set(cset)
3134 RF_ConfigSet_t *cset;
3135 {
3136 RF_AutoConfig_t *ac;
3137 RF_AutoConfig_t *next_ac;
3138
3139 ac = cset->ac;
3140 while(ac!=NULL) {
3141 next_ac = ac->next;
3142 /* nuke the label */
3143 free(ac->clabel, M_RAIDFRAME);
3144 /* cleanup the config structure */
3145 free(ac, M_RAIDFRAME);
3146 /* "next.." */
3147 ac = next_ac;
3148 }
3149 /* and, finally, nuke the config set */
3150 free(cset, M_RAIDFRAME);
3151 }
3152
3153
3154 void
3155 raid_init_component_label(raidPtr, clabel)
3156 RF_Raid_t *raidPtr;
3157 RF_ComponentLabel_t *clabel;
3158 {
3159 /* current version number */
3160 clabel->version = RF_COMPONENT_LABEL_VERSION;
3161 clabel->serial_number = raidPtr->serial_number;
3162 clabel->mod_counter = raidPtr->mod_counter;
3163 clabel->num_rows = raidPtr->numRow;
3164 clabel->num_columns = raidPtr->numCol;
3165 clabel->clean = RF_RAID_DIRTY; /* not clean */
3166 clabel->status = rf_ds_optimal; /* "It's good!" */
3167
3168 clabel->sectPerSU = raidPtr->Layout.sectorsPerStripeUnit;
3169 clabel->SUsPerPU = raidPtr->Layout.SUsPerPU;
3170 clabel->SUsPerRU = raidPtr->Layout.SUsPerRU;
3171
3172 clabel->blockSize = raidPtr->bytesPerSector;
3173 clabel->numBlocks = raidPtr->sectorsPerDisk;
3174
3175 /* XXX not portable */
3176 clabel->parityConfig = raidPtr->Layout.map->parityConfig;
3177 clabel->maxOutstanding = raidPtr->maxOutstanding;
3178 clabel->autoconfigure = raidPtr->autoconfigure;
3179 clabel->root_partition = raidPtr->root_partition;
3180 clabel->last_unit = raidPtr->raidid;
3181 clabel->config_order = raidPtr->config_order;
3182 }
3183
3184 int
3185 rf_auto_config_set(cset,unit)
3186 RF_ConfigSet_t *cset;
3187 int *unit;
3188 {
3189 RF_Raid_t *raidPtr;
3190 RF_Config_t *config;
3191 int raidID;
3192 int retcode;
3193
3194 printf("RAID autoconfigure\n");
3195
3196 retcode = 0;
3197 *unit = -1;
3198
3199 /* 1. Create a config structure */
3200
3201 config = (RF_Config_t *)malloc(sizeof(RF_Config_t),
3202 M_RAIDFRAME,
3203 M_NOWAIT);
3204 if (config==NULL) {
3205 printf("Out of mem!?!?\n");
3206 /* XXX do something more intelligent here. */
3207 return(1);
3208 }
3209
3210 memset(config, 0, sizeof(RF_Config_t));
3211
3212 /* XXX raidID needs to be set correctly.. */
3213
3214 /*
3215 2. Figure out what RAID ID this one is supposed to live at
3216 See if we can get the same RAID dev that it was configured
3217 on last time..
3218 */
3219
3220 raidID = cset->ac->clabel->last_unit;
3221 if ((raidID < 0) || (raidID >= numraid)) {
3222 /* let's not wander off into lala land. */
3223 raidID = numraid - 1;
3224 }
3225 if (raidPtrs[raidID]->valid != 0) {
3226
3227 /*
3228 Nope... Go looking for an alternative...
3229 Start high so we don't immediately use raid0 if that's
3230 not taken.
3231 */
3232
3233 for(raidID = numraid; raidID >= 0; raidID--) {
3234 if (raidPtrs[raidID]->valid == 0) {
3235 /* can use this one! */
3236 break;
3237 }
3238 }
3239 }
3240
3241 if (raidID < 0) {
3242 /* punt... */
3243 printf("Unable to auto configure this set!\n");
3244 printf("(Out of RAID devs!)\n");
3245 return(1);
3246 }
3247 printf("Configuring raid%d:\n",raidID);
3248 raidPtr = raidPtrs[raidID];
3249
3250 /* XXX all this stuff should be done SOMEWHERE ELSE! */
3251 raidPtr->raidid = raidID;
3252 raidPtr->openings = RAIDOUTSTANDING;
3253
3254 /* 3. Build the configuration structure */
3255 rf_create_configuration(cset->ac, config, raidPtr);
3256
3257 /* 4. Do the configuration */
3258 retcode = rf_Configure(raidPtr, config, cset->ac);
3259
3260 if (retcode == 0) {
3261
3262 raidinit(raidPtrs[raidID]);
3263
3264 rf_markalldirty(raidPtrs[raidID]);
3265 raidPtrs[raidID]->autoconfigure = 1; /* XXX do this here? */
3266 if (cset->ac->clabel->root_partition==1) {
3267 /* everything configured just fine. Make a note
3268 that this set is eligible to be root. */
3269 cset->rootable = 1;
3270 /* XXX do this here? */
3271 raidPtrs[raidID]->root_partition = 1;
3272 }
3273 }
3274
3275 /* 5. Cleanup */
3276 free(config, M_RAIDFRAME);
3277
3278 *unit = raidID;
3279 return(retcode);
3280 }
3281