1 /*-
2 * Copyright (c) 1997, 1998
3 * Nan Yang Computer Services Limited. All rights reserved.
4 *
5 * This software is distributed under the so-called ``Berkeley
6 * License'':
7 *
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
10 * are met:
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 * 3. All advertising materials mentioning features or use of this software
17 * must display the following acknowledgement:
18 * This product includes software developed by Nan Yang Computer
19 * Services Limited.
20 * 4. Neither the name of the Company nor the names of its contributors
21 * may be used to endorse or promote products derived from this software
22 * without specific prior written permission.
23 *
24 * This software is provided ``as is'', and any express or implied
25 * warranties, including, but not limited to, the implied warranties of
26 * merchantability and fitness for a particular purpose are disclaimed.
27 * In no event shall the company or contributors be liable for any
28 * direct, indirect, incidental, special, exemplary, or consequential
29 * damages (including, but not limited to, procurement of substitute
30 * goods or services; loss of use, data, or profits; or business
31 * interruption) however caused and on any theory of liability, whether
32 * in contract, strict liability, or tort (including negligence or
33 * otherwise) arising in any way out of the use of this software, even if
34 * advised of the possibility of such damage.
35 *
36 * $Id: vinumio.c,v 1.1.1.1 2003/10/10 03:08:25 grog Exp $
37 * $FreeBSD$
38 */
39
40 #include <dev/vinum/vinumhdr.h>
41 #include <dev/vinum/request.h>
42
43 static char *sappend(char *txt, char *s);
44 static int drivecmp(const void *va, const void *vb);
45
46 /* Open the device associated with the drive, and set drive's vp.
47 * Return an error number */
48 int
49 open_drive(struct drive *drive, struct proc *p, int verbose)
50 {
51 struct nameidata nd;
52 struct vattr va;
53 int error;
54
55 if (drive->devicename[0] == '\0') /* no device name */
56 sprintf(drive->devicename, "/dev/%s", drive->label.name); /* get it from the drive name */
57 NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, drive->devicename, p);
58 error = vn_open(&nd, FREAD | FWRITE, 0); /* open the device */
59 if (error != 0) { /* can't open? */
60 set_drive_state(drive->driveno, drive_down, setstate_force);
61 drive->lasterror = error;
62 if (verbose)
63 printf("vinum open_drive %s: failed with error %d\n", drive->devicename, error); /* XXX */
64 return error;
65 }
66 drive->flags |= VF_OPEN; /* mark the drive as open */
67 drive->vp = nd.ni_vp;
68 drive->p = p;
69
70 error = VOP_GETATTR(drive->vp, &va, NOCRED, drive->p);
71 if (error) {
72 VOP_UNLOCK(drive->vp, 0);
73 close_drive(drive);
74 set_drive_state(drive->driveno, drive_down, setstate_force);
75 drive->lasterror = error;
76 if (verbose)
77 printf("vinum open_drive %s: GETAATTR returns error %d\n", drive->devicename, error); /* XXX */
78 return error;
79 }
80 drive->dev = va.va_rdev; /* device */
81
82 if (va.va_type != VBLK) { /* only consider block devices */
83 VOP_UNLOCK(drive->vp, 0);
84 close_drive(drive);
85 set_drive_state(drive->driveno, drive_down, setstate_force); /* this also closes the drive */
86 drive->lasterror = ENOTBLK;
87 if (verbose)
88 printf("vinum open_drive %s: Not a block device\n", drive->devicename); /* XXX */
89 return ENOTBLK;
90 }
91 drive->vp->v_numoutput = 0;
92 VOP_UNLOCK(drive->vp, 0);
93 return 0;
94 }
95
96
97 /*
98 * Set some variables in the drive struct in more
99 * convenient form. Return error indication.
100 */
101 int
102 set_drive_parms(struct drive *drive)
103 {
104 drive->blocksize = BLKDEV_IOSIZE; /* do we need this? */
105 drive->secsperblock = drive->blocksize /* number of sectors per block */
106 / drive->partinfo.disklab->d_secsize;
107
108 /* Now update the label part */
109 bcopy(hostname, drive->label.sysname, VINUMHOSTNAMELEN); /* put in host name */
110 microtime(&drive->label.date_of_birth); /* and current time */
111 drive->label.drive_size = ((u_int64_t) drive->partinfo.part->p_size) /* size of the drive in bytes */
112 *((u_int64_t) drive->partinfo.disklab->d_secsize);
113 #ifdef VINUMDEBUG
114 if (debug & DEBUG_BIGDRIVE) /* pretend we're 100 times as big */
115 drive->label.drive_size *= 100;
116 #endif
117
118 /* number of sectors available for subdisks */
119 drive->sectors_available = drive->label.drive_size / DEV_BSIZE - DATASTART;
120
121 /*
122 * Bug in 3.0 as of January 1998: you can open
123 * non-existent slices. They have a length of 0.
124 */
125 if (drive->label.drive_size < MINVINUMSLICE) { /* too small to worry about */
126 set_drive_state(drive->driveno, drive_down, setstate_force);
127 drive->lasterror = ENOSPC;
128 return ENOSPC;
129 }
130 drive->freelist_size = INITIAL_DRIVE_FREELIST; /* initial number of entries */
131 drive->freelist = (struct drive_freelist *)
132 Malloc(INITIAL_DRIVE_FREELIST * sizeof(struct drive_freelist));
133 if (drive->freelist == NULL) /* can't malloc, dammit */
134 return ENOSPC;
135 drive->freelist_entries = 1; /* just (almost) the complete drive */
136 drive->freelist[0].offset = DATASTART; /* starts here */
137 drive->freelist[0].sectors = (drive->label.drive_size >> DEV_BSHIFT) - DATASTART; /* and it's this long */
138 if (drive->label.name[0] != '\0') /* got a name */
139 set_drive_state(drive->driveno, drive_up, setstate_force); /* our drive is accessible */
140 else /* we know about it, but that's all */
141 drive->state = drive_referenced;
142 return 0;
143 }
144
145 /*
146 * Initialize a drive: open the device and add
147 * device information.
148 */
149 int
150 init_drive(struct drive *drive, int verbose)
151 {
152 int error;
153 struct disklabel label;
154 const struct bdevsw *bdev;
155
156 if (drive->devicename[0] != '/') {
157 drive->lasterror = EINVAL;
158 log(LOG_ERR, "vinum: Can't open drive without drive name\n");
159 return EINVAL;
160 }
161 drive->lasterror = open_drive(drive, curproc, verbose); /* open the drive */
162 if (drive->lasterror)
163 return drive->lasterror;
164
165 bdev = bdevsw_lookup(drive->dev);
166 if (bdev == NULL)
167 return ENXIO;
168 drive->lasterror = (*bdev->d_ioctl) (drive->dev,
169 DIOCGDINFO,
170 (caddr_t) & label,
171 FREAD,
172 curproc);
173 if (drive->lasterror == 0) {
174 if (label.d_type == DTYPE_FLOPPY) {
175 drive->partinfo.disklab->d_secsize = label.d_secsize;
176 drive->partinfo.part->p_size = label.d_secsize * label.d_secperunit;
177 drive->partinfo.part->p_fstype = FS_VINUM; /* force Vinum */
178 } else
179 /* XXX should use VOP interface, like FreeBSD */
180 drive->lasterror = (*bdev->d_ioctl) (drive->dev, /* get the partition information */
181 DIOCGPART,
182 (caddr_t) & drive->partinfo,
183 FREAD,
184 curproc);
185 }
186 if (drive->lasterror) {
187 if (verbose)
188 log(LOG_WARNING,
189 "vinum open_drive %s: Can't get partition information, drive->lasterror %d\n",
190 drive->devicename,
191 drive->lasterror);
192 error = drive->lasterror; /* clobbered by close_drive */
193 close_drive(drive);
194 return error;
195 }
196 if (drive->partinfo.part->p_fstype != FS_VINUM) { /* not Vinum */
197 drive->lasterror = EFTYPE;
198 if (verbose)
199 log(LOG_WARNING,
200 "vinum open_drive %s: Wrong partition type for vinum\n",
201 drive->devicename);
202 close_drive(drive);
203 return EFTYPE;
204 }
205 return set_drive_parms(drive); /* set various odds and ends */
206 }
207
208 /* Close a drive if it's open. */
209 void
210 close_drive(struct drive *drive)
211 {
212 LOCKDRIVE(drive); /* keep the daemon out */
213 if (drive->flags & VF_OPEN)
214 close_locked_drive(drive); /* and close it */
215 if (drive->state > drive_down) /* if it's up */
216 drive->state = drive_down; /* make sure it's down */
217 unlockdrive(drive);
218 }
219
220 /*
221 * Real drive close code, called with drive already locked.
222 * We have also checked that the drive is open. No errors.
223 */
224 void
225 close_locked_drive(struct drive *drive)
226 {
227 /*
228 * If we can't access the drive, we can't flush
229 * the queues, which spec_close() will try to
230 * do. Get rid of them here first.
231 */
232
233 const struct bdevsw *bdev = bdevsw_lookup(drive->dev);
234
235 if (bdev == NULL)
236 panic("no bdevsw"); /* XXX */
237 drive->lasterror = (*bdev->d_close) (drive->dev, 0, 0, NULL);
238 drive->flags &= ~VF_OPEN; /* no longer open */
239 }
240
241 /*
242 * Remove drive from the configuration.
243 * Caller must ensure that it isn't active.
244 */
245 void
246 remove_drive(int driveno)
247 {
248 struct drive *drive = &vinum_conf.drive[driveno];
249 struct vinum_hdr *vhdr; /* buffer for header */
250 int error;
251
252 if (drive->state > drive_referenced) { /* real drive */
253 if (drive->state == drive_up) {
254 vhdr = (struct vinum_hdr *) Malloc(VINUMHEADERLEN); /* allocate buffer */
255 CHECKALLOC(vhdr, "Can't allocate memory");
256 error = read_drive(drive, (void *) vhdr, VINUMHEADERLEN, VINUM_LABEL_OFFSET);
257 if (error)
258 drive->lasterror = error;
259 else {
260 vhdr->magic = VINUM_NOMAGIC; /* obliterate the magic, but leave the rest */
261 write_drive(drive, (void *) vhdr, VINUMHEADERLEN, VINUM_LABEL_OFFSET);
262 }
263 Free(vhdr);
264 }
265 free_drive(drive); /* close it and free resources */
266 save_config(); /* and save the updated configuration */
267 }
268 }
269
270 /*
271 * Transfer drive data. Usually called from one of these defines;
272 * #define read_drive(a, b, c, d) driveio (a, b, c, d, B_READ)
273 * #define write_drive(a, b, c, d) driveio (a, b, c, d, B_WRITE)
274 *
275 * length and offset are in bytes, but must be multiples of sector
276 * size. The function *does not check* for this condition, and
277 * truncates ruthlessly.
278 * Return error number.
279 */
280 int
281 driveio(struct drive *drive, char *buf, size_t length, off_t offset, int flag)
282 {
283 int error;
284 struct buf *bp;
285 const struct bdevsw *bdev;
286
287 error = 0; /* to keep the compiler happy */
288 while (length) { /* divide into small enough blocks */
289 int len = min(length, MAXBSIZE); /* maximum block device transfer is MAXBSIZE */
290
291 bp = geteblk(len); /* get a buffer header */
292 bp->b_flags = flag | B_BUSY;
293 bp->b_dev = drive->dev; /* device */
294 bp->b_blkno = offset / drive->partinfo.disklab->d_secsize; /* block number */
295 bp->b_saveaddr = bp->b_data;
296 bp->b_data = buf;
297 bp->b_bcount = len;
298 bdev = bdevsw_lookup(bp->b_dev);
299 if (bdev == NULL)
300 return ENXIO;
301 (*bdev->d_strategy) (bp);
302 error = biowait(bp);
303 bp->b_data = bp->b_saveaddr;
304 bp->b_flags |= B_INVAL | B_AGE;
305 bp->b_flags &= ~B_ERROR;
306 brelse(bp);
307 if (error)
308 break;
309 length -= len; /* update pointers */
310 buf += len;
311 offset += len;
312 }
313 return error;
314 }
315
316 /*
317 * Check a drive for a vinum header. If found,
318 * update the drive information. We come here
319 * with a partially populated drive structure
320 * which includes the device name.
321 *
322 * Return information on what we found.
323 *
324 * This function is called from two places: check_drive,
325 * which wants to find out whether the drive is a
326 * Vinum drive, and config_drive, which asserts that
327 * it is a vinum drive. In the first case, we don't
328 * print error messages (verbose==0), in the second
329 * we do (verbose==1).
330 */
331 enum drive_label_info
332 read_drive_label(struct drive *drive, int verbose)
333 {
334 int error;
335 int result; /* result of our search */
336 struct vinum_hdr *vhdr; /* and as header */
337
338 error = init_drive(drive, 0); /* find the drive */
339 if (error) /* find the drive */
340 return DL_CANT_OPEN; /* not ours */
341
342 vhdr = (struct vinum_hdr *) Malloc(VINUMHEADERLEN); /* allocate buffers */
343 CHECKALLOC(vhdr, "Can't allocate memory");
344
345 drive->state = drive_up; /* be optimistic */
346 error = read_drive(drive, (void *) vhdr, VINUMHEADERLEN, VINUM_LABEL_OFFSET);
347 if (vhdr->magic == VINUM_MAGIC) { /* ours! */
348 if (drive->label.name[0] /* we have a name for this drive */
349 &&(strcmp(drive->label.name, vhdr->label.name))) { /* but it doesn't match the real name */
350 drive->lasterror = EINVAL;
351 result = DL_WRONG_DRIVE; /* it's the wrong drive */
352 drive->state = drive_unallocated; /* put it back, it's not ours */
353 } else
354 result = DL_OURS;
355 /*
356 * We copy the drive anyway so that we have
357 * the correct name in the drive info. This
358 * may not be the name specified
359 */
360 drive->label = vhdr->label; /* put in the label information */
361 } else if (vhdr->magic == VINUM_NOMAGIC) /* was ours, but we gave it away */
362 result = DL_DELETED_LABEL; /* and return the info */
363 else
364 result = DL_NOT_OURS; /* we could have it, but we don't yet */
365 Free(vhdr); /* that's all. */
366 return result;
367 }
368
369 /*
370 * Check a drive for a vinum header. If found,
371 * read configuration information from the drive and
372 * incorporate the data into the configuration.
373 *
374 * Return drive number.
375 */
376 struct drive *
377 check_drive(char *devicename)
378 {
379 int driveno;
380 int i;
381 struct drive *drive;
382
383 driveno = find_drive_by_name(devicename, 1); /* if entry doesn't exist, create it */
384 drive = &vinum_conf.drive[driveno]; /* and get a pointer */
385
386 if (read_drive_label(drive, 0) == DL_OURS) { /* one of ours */
387 for (i = 0; i < vinum_conf.drives_allocated; i++) { /* see if the name already exists */
388 if ((i != driveno) /* not this drive */
389 &&(DRIVE[i].state != drive_unallocated) /* and it's allocated */
390 &&(strcmp(DRIVE[i].label.name,
391 DRIVE[driveno].label.name) == 0)) { /* and it has the same name */
392 struct drive *mydrive = &DRIVE[i];
393
394 if (mydrive->devicename[0] == '/') { /* we know a device name for it */
395 /*
396 * set an error, but don't take the
397 * drive down: that would cause unneeded
398 * error messages.
399 */
400 drive->lasterror = EEXIST;
401 break;
402 } else { /* it's just a place holder, */
403 int sdno;
404
405 for (sdno = 0; sdno < vinum_conf.subdisks_allocated; sdno++) { /* look at each subdisk */
406 if ((SD[sdno].driveno == i) /* it's pointing to this one, */
407 &&(SD[sdno].state != sd_unallocated)) { /* and it's a real subdisk */
408 SD[sdno].driveno = drive->driveno; /* point to the one we found */
409 update_sd_state(sdno); /* and update its state */
410 }
411 }
412 bzero(mydrive, sizeof(struct drive)); /* don't deallocate it, just remove it */
413 }
414 }
415 }
416 } else {
417 if (drive->lasterror == 0)
418 drive->lasterror = ENODEV;
419 close_drive(drive);
420 drive->state = drive_down;
421 }
422 return drive;
423 }
424
425 static char *
426 sappend(char *txt, char *s)
427 {
428 while ((*s++ = *txt++) != 0);
429 return s - 1;
430 }
431
432 /* Kludge: kernel printf doesn't handle quads correctly XXX */
433 static char *lltoa(long long l, char *s);
434
435 static char *
436 lltoa(long long l, char *s)
437 {
438 if (l < 0) {
439 *s++ = '-';
440 l = -l;
441 }
442 if (l > 9) {
443 s = lltoa(l / 10, s);
444 l %= 10;
445 }
446 *s++ = l + '';
447 return s;
448 }
449 /*
450 * Format the configuration in text form into the buffer
451 * at config. Don't go beyond len bytes
452 * XXX this stinks. Fix soon.
453 */
454 void
455 format_config(char *config, int len)
456 {
457 int i;
458 int j;
459 char *s = config;
460
461 bzero(config, len);
462
463 /* First, the volume configuration */
464 for (i = 0; i < vinum_conf.volumes_allocated; i++) {
465 struct volume *vol;
466
467 vol = &vinum_conf.volume[i];
468 if ((vol->state > volume_uninit)
469 && (vol->name[0] != '\0')) { /* paranoia */
470 if (vol->preferred_plex >= 0) /* preferences, */
471 sprintf(s,
472 "volume %s state %s readpol prefer %s",
473 vol->name,
474 volume_state(vol->state),
475 vinum_conf.plex[vol->preferred_plex].name);
476 else /* default round-robin */
477 sprintf(s,
478 "volume %s state %s",
479 vol->name,
480 volume_state(vol->state));
481 while (*s)
482 s++; /* find the end */
483 s = sappend("\n", s);
484 if (s > &config[len - 80]) {
485 log(LOG_ERR, "vinum: configuration data overflow\n");
486 return;
487 }
488 }
489 }
490
491 /* Then the plex configuration */
492 for (i = 0; i < vinum_conf.plexes_allocated; i++) {
493 struct plex *plex;
494
495 plex = &vinum_conf.plex[i];
496 if ((plex->state != plex_referenced)
497 && (plex->name[0] != '\0')) { /* paranoia */
498 sprintf(s, "plex name %s state %s org %s ",
499 plex->name,
500 plex_state(plex->state),
501 plex_org(plex->organization));
502 while (*s)
503 s++; /* find the end */
504 if ((plex->organization == plex_striped)
505 || (plex->organization == plex_raid5)) {
506 sprintf(s, "%db ", (int) plex->stripesize);
507 while (*s)
508 s++; /* find the end */
509 }
510 if (plex->volno >= 0) /* we have a volume */
511 sprintf(s, "vol %s ", vinum_conf.volume[plex->volno].name);
512 while (*s)
513 s++; /* find the end */
514 for (j = 0; j < plex->subdisks; j++) {
515 sprintf(s, " sd %s", vinum_conf.sd[plex->sdnos[j]].name);
516 }
517 s = sappend("\n", s);
518 if (s > &config[len - 80]) {
519 log(LOG_ERR, "vinum: configuration data overflow\n");
520 return;
521 }
522 }
523 }
524
525 /* And finally the subdisk configuration */
526 for (i = 0; i < vinum_conf.subdisks_allocated; i++) {
527 struct sd *sd;
528
529 sd = &SD[i];
530 if ((sd->state != sd_referenced)
531 && (sd->name[0] != '\0')) { /* paranoia */
532 sprintf(s,
533 "sd name %s drive %s plex %s state %s len ",
534 sd->name,
535 vinum_conf.drive[sd->driveno].label.name,
536 vinum_conf.plex[sd->plexno].name,
537 sd_state(sd->state));
538 while (*s)
539 s++; /* find the end */
540 s = lltoa(sd->sectors, s);
541 s = sappend("b driveoffset ", s);
542 s = lltoa(sd->driveoffset, s);
543 s = sappend("b plexoffset ", s);
544 s = lltoa(sd->plexoffset, s);
545 s = sappend("b\n", s);
546 if (s > &config[len - 80]) {
547 log(LOG_ERR, "vinum: configuration data overflow\n");
548 return;
549 }
550 }
551 }
552 }
553
554 /*
555 * issue a save config request to the dæmon. The actual work
556 * is done in process context by daemon_save_config.
557 */
558 void
559 save_config(void)
560 {
561 queue_daemon_request(daemonrq_saveconfig, (union daemoninfo) 0);
562 }
563
564 /*
565 * Write the configuration to all vinum slices. This
566 * is performed by the daemon only.
567 */
568 void
569 daemon_save_config(void)
570 {
571 int error;
572 int written_config; /* set when we first write the config to disk */
573 int driveno;
574 struct drive *drive; /* point to current drive info */
575 struct vinum_hdr *vhdr; /* and as header */
576 char *config; /* point to config data */
577 int wlabel_on; /* to set writing label on/off */
578 const struct bdevsw *bdev;
579
580 /* don't save the configuration while we're still working on it */
581 if (vinum_conf.flags & VF_CONFIGURING)
582 return;
583 written_config = 0; /* no config written yet */
584 /* Build a volume header */
585 vhdr = (struct vinum_hdr *) Malloc(VINUMHEADERLEN); /* get space for the config data */
586 CHECKALLOC(vhdr, "Can't allocate config data");
587 vhdr->magic = VINUM_MAGIC; /* magic number */
588 vhdr->config_length = MAXCONFIG; /* length of following config info */
589
590 config = Malloc(MAXCONFIG); /* get space for the config data */
591 CHECKALLOC(config, "Can't allocate config data");
592
593 format_config(config, MAXCONFIG);
594 error = 0; /* no errors yet */
595 for (driveno = 0; driveno < vinum_conf.drives_allocated; driveno++) {
596 drive = &vinum_conf.drive[driveno]; /* point to drive */
597 if (drive->state > drive_referenced) {
598 LOCKDRIVE(drive); /* don't let it change */
599
600 /*
601 * First, do some drive consistency checks. Some
602 * of these are kludges, others require a process
603 * context and couldn't be done before.
604 */
605 if ((drive->devicename[0] == '\0')
606 || (drive->label.name[0] == '\0')) {
607 unlockdrive(drive);
608 free_drive(drive); /* get rid of it */
609 break;
610 }
611 if (((drive->flags & VF_OPEN) == 0) /* drive not open */
612 &&(drive->state > drive_down)) { /* and it thinks it's not down */
613 unlockdrive(drive);
614 set_drive_state(driveno, drive_down, setstate_force); /* tell it what's what */
615 continue;
616 }
617 if ((drive->state == drive_down) /* it's down */
618 &&(drive->flags & VF_OPEN)) { /* but open, */
619 unlockdrive(drive);
620 close_drive(drive); /* close it */
621 } else if (drive->state > drive_down) {
622 microtime(&drive->label.last_update); /* time of last update is now */
623 bcopy((char *) &drive->label, /* and the label info from the drive structure */
624 (char *) &vhdr->label,
625 sizeof(vhdr->label));
626 if ((drive->state != drive_unallocated)
627 && (drive->state != drive_referenced)) { /* and it's a real drive */
628 wlabel_on = 1; /* enable writing the label */
629 bdev = bdevsw_lookup(drive->dev);
630 if (bdev == NULL)
631 /* XXX log this */
632 error = ENXIO;
633 else
634 error = (*bdev->d_ioctl) (drive->dev, /* make the label writeable */
635 DIOCWLABEL,
636 (caddr_t) & wlabel_on,
637 FWRITE,
638 curproc);
639 if (error == 0)
640 error = write_drive(drive, (char *) vhdr, VINUMHEADERLEN, VINUM_LABEL_OFFSET);
641 if (error == 0)
642 error = write_drive(drive, config, MAXCONFIG, VINUM_CONFIG_OFFSET); /* first config copy */
643 if (error == 0)
644 error = write_drive(drive, config, MAXCONFIG, VINUM_CONFIG_OFFSET + MAXCONFIG); /* second copy */
645 wlabel_on = 0; /* enable writing the label */
646 if (error == 0)
647 /* make the label non-writeable again */
648 error = (*bdev->d_ioctl) (drive->dev,
649 DIOCWLABEL,
650 (caddr_t) & wlabel_on,
651 FWRITE,
652 curproc);
653 unlockdrive(drive);
654 if (error) {
655 log(LOG_ERR,
656 "vinum: Can't write config to %s, error %d\n",
657 drive->devicename,
658 error);
659 set_drive_state(drive->driveno, drive_down, setstate_force);
660 } else
661 written_config = 1; /* we've written it on at least one drive */
662 }
663 } else /* not worth looking at, */
664 unlockdrive(drive); /* just unlock it again */
665 }
666 }
667 Free(vhdr);
668 Free(config);
669 }
670
671 /*
672 * Disk labels are a mess. The correct way to
673 * access them is with the DIOC[GSW]DINFO ioctls,
674 * but some programs, such as newfs, access the
675 * disk directly, so we have to write things
676 * there. We do this only on request. If a user
677 * request tries to read it directly, we fake up
678 * one on the fly.
679 */
680
681 /*
682 * get_volume_label returns a label structure to
683 * lp, which is allocated by the caller.
684 */
685 void
686 get_volume_label(char *name, int plexes, u_int64_t size, struct disklabel *lp)
687 {
688 bzero(lp, sizeof(struct disklabel));
689
690 strncpy(lp->d_typename, "vinum", sizeof(lp->d_typename));
691 lp->d_type = DTYPE_VINUM;
692 strncpy(lp->d_packname, name, min(sizeof(lp->d_packname), sizeof(name)));
693 lp->d_rpm = 14400 * plexes; /* to keep them guessing */
694 lp->d_interleave = 1;
695 lp->d_flags = 0;
696
697 /*
698 * A Vinum volume has a single track with all
699 * its sectors.
700 */
701 lp->d_secsize = DEV_BSIZE; /* bytes per sector */
702 lp->d_nsectors = size; /* data sectors per track */
703 lp->d_ntracks = 1; /* tracks per cylinder */
704 lp->d_ncylinders = 1; /* data cylinders per unit */
705 lp->d_secpercyl = size; /* data sectors per cylinder */
706 lp->d_secperunit = size; /* data sectors per unit */
707
708 lp->d_bbsize = 8192; /* fake boot block size */
709 lp->d_sbsize = 1024; /* NetBSD doesn't seem to have an SBSIZE XXX */
710 lp->d_magic = DISKMAGIC;
711 lp->d_magic2 = DISKMAGIC;
712
713 /*
714 * Set up partitions a, b and c to be identical
715 * and the size of the volume. a is UFS, b is
716 * swap, c is nothing.
717 */
718 lp->d_partitions[0].p_size = size;
719 lp->d_partitions[0].p_fsize = 1024;
720 lp->d_partitions[0].p_fstype = FS_BSDFFS; /* FreeBSD File System :-) */
721 lp->d_partitions[0].p_fsize = 1024; /* FS fragment size */
722 lp->d_partitions[0].p_frag = 8; /* and fragments per block */
723 lp->d_partitions[SWAP_PART].p_size = size;
724 lp->d_partitions[SWAP_PART].p_fstype = FS_SWAP; /* swap partition */
725 lp->d_partitions[LABEL_PART].p_size = size;
726 lp->d_npartitions = LABEL_PART + 1;
727 strncpy(lp->d_packname, name, min(sizeof(lp->d_packname), sizeof(name)));
728 lp->d_checksum = dkcksum(lp);
729 }
730
731 /* Write a volume label. This implements the VINUM_LABEL ioctl. */
732 int
733 write_volume_label(int volno)
734 {
735 struct disklabel *lp;
736 struct buf *bp;
737 struct disklabel *dlp;
738 struct volume *vol;
739 int error;
740
741 lp = (struct disklabel *) Malloc((sizeof(struct disklabel) + (DEV_BSIZE - 1)) & (DEV_BSIZE - 1));
742 if (lp == 0)
743 return ENOMEM;
744
745 if ((unsigned) (volno) >= (unsigned) vinum_conf.volumes_allocated) /* invalid volume */
746 return ENOENT;
747
748 vol = &VOL[volno]; /* volume in question */
749 if (vol->state <= volume_uninit) /* nothing there */
750 return ENXIO;
751 else if (vol->state < volume_up) /* not accessible */
752 return EIO; /* I/O error */
753
754 get_volume_label(vol->name, vol->plexes, vol->size, lp); /* get the label */
755
756 /*
757 * Now write to disk. This code is derived from the
758 * system writedisklabel (), which does silly things
759 * like reading the label and refusing to write
760 * unless it's already there.
761 */
762 bp = geteblk((int) lp->d_secsize); /* get a buffer */
763 bp->b_dev = makedev(VINUM_CDEV_MAJOR, vol->volno); /* our own raw volume */
764 bp->b_blkno = LABELSECTOR * ((int) lp->d_secsize / DEV_BSIZE);
765 bp->b_bcount = lp->d_secsize;
766 bzero(bp->b_data, lp->d_secsize);
767 dlp = (struct disklabel *) bp->b_data;
768 *dlp = *lp;
769 bp->b_flags &= ~B_INVAL;
770 bp->b_flags |= B_BUSY | B_WRITE;
771 vinumstrategy(bp);
772 error = biowait(bp);
773 bp->b_flags |= B_INVAL | B_AGE;
774 bp->b_flags &= ~B_ERROR;
775
776 brelse(bp);
777 return error;
778 }
779
780 /* Look at all disks on the system for vinum slices. */
781 int
782 vinum_scandisk(char *devicename[], int drives)
783 {
784 struct drive *volatile drive;
785 volatile int driveno;
786 int firstdrive; /* first drive in this list */
787 volatile int gooddrives; /* number of usable drives found */
788 int firsttime; /* set if we have never configured before */
789 int error;
790 char *config_text; /* read the config info from disk into here */
791 char *volatile cptr; /* pointer into config information */
792 char *eptr; /* end pointer into config information */
793 char *config_line; /* copy the config line to */
794 volatile int status;
795 int *volatile drivelist; /* list of drive indices */
796 #define DRIVENAMELEN 64
797 #define DRIVEPARTS 35 /* max partitions per drive, excluding c */
798 char partname[DRIVENAMELEN]; /* for creating partition names */
799
800 status = 0; /* success indication */
801 vinum_conf.flags |= VF_READING_CONFIG; /* reading config from disk */
802
803 gooddrives = 0; /* number of usable drives found */
804 firstdrive = vinum_conf.drives_used; /* the first drive */
805 firsttime = vinum_conf.drives_used == 0; /* are we a virgin? */
806
807 /* allocate a drive pointer list */
808 drivelist = (int *) Malloc(drives * DRIVEPARTS * sizeof(int));
809 CHECKALLOC(drivelist, "Can't allocate memory");
810
811 /* Open all drives and find which was modified most recently */
812 for (driveno = 0; driveno < drives; driveno++) {
813 char part; /* UNIX partition */
814 int slice;
815 int founddrive; /* flag when we find a vinum drive */
816
817 founddrive = 0; /* no vinum drive found yet on this spindle */
818 /* first try the partition table */
819 for (slice = 1; slice < 5; slice++)
820 for (part = 'a'; part < 'i'; part++) {
821 if (part != 'c') { /* don't do the c partition */
822 snprintf(partname,
823 DRIVENAMELEN,
824 "%ss%d%c",
825 devicename[driveno],
826 slice,
827 part);
828 drive = check_drive(partname); /* try to open it */
829 if ((drive->lasterror != 0) /* didn't work, */
830 ||(drive->state != drive_up))
831 free_drive(drive); /* get rid of it */
832 else if (drive->flags & VF_CONFIGURED) /* already read this config, */
833 log(LOG_WARNING,
834 "vinum: already read config from %s\n", /* say so */
835 drive->label.name);
836 else {
837 drivelist[gooddrives] = drive->driveno; /* keep the drive index */
838 drive->flags &= ~VF_NEWBORN; /* which is no longer newly born */
839 gooddrives++;
840 founddrive++;
841 }
842 }
843 }
844 if (founddrive == 0) { /* didn't find anything, */
845 for (part = 'a'; part < 'i'; part++) /* try the compatibility partition */
846 if (part != 'c') { /* don't do the c partition */
847 snprintf(partname, /* /dev/sd0a */
848 DRIVENAMELEN,
849 "%s%c",
850 devicename[driveno],
851 part);
852 drive = check_drive(partname); /* try to open it */
853 if ((drive->lasterror != 0) /* didn't work, */
854 ||(drive->state != drive_up))
855 free_drive(drive); /* get rid of it */
856 else if (drive->flags & VF_CONFIGURED) /* already read this config, */
857 log(LOG_WARNING,
858 "vinum: already read config from %s\n", /* say so */
859 drive->label.name);
860 else {
861 drivelist[gooddrives] = drive->driveno; /* keep the drive index */
862 drive->flags &= ~VF_NEWBORN; /* which is no longer newly born */
863 gooddrives++;
864 }
865 }
866 }
867 }
868
869 if (gooddrives == 0) {
870 if (firsttime)
871 log(LOG_WARNING, "vinum: no drives found\n");
872 else
873 log(LOG_INFO, "vinum: no additional drives found\n");
874 return ENOENT;
875 }
876 /*
877 * We now have at least one drive open. Sort
878 * them in order of config time and merge the
879 * config info with what we have already.
880 */
881 qsort(drivelist, gooddrives, sizeof(int), drivecmp);
882 config_text = (char *) Malloc(MAXCONFIG * 2); /* allocate buffers */
883 CHECKALLOC(config_text, "Can't allocate memory");
884 config_line = (char *) Malloc(MAXCONFIGLINE * 2); /* allocate buffers */
885 CHECKALLOC(config_line, "Can't allocate memory");
886 for (driveno = 0; driveno < gooddrives; driveno++) { /* now include the config */
887 drive = &DRIVE[drivelist[driveno]]; /* point to the drive */
888
889 if (firsttime && (driveno == 0)) /* we've never configured before, */
890 log(LOG_INFO, "vinum: reading configuration from %s\n", drive->devicename);
891 else
892 log(LOG_INFO, "vinum: updating configuration from %s\n", drive->devicename);
893
894 if (drive->state == drive_up)
895 /* Read in both copies of the configuration information */
896 error = read_drive(drive, config_text, MAXCONFIG * 2, VINUM_CONFIG_OFFSET);
897 else {
898 error = EIO;
899 printf("vinum_scandisk: %s is %s\n", drive->devicename, drive_state(drive->state));
900 }
901
902 if (error != 0) {
903 log(LOG_ERR, "vinum: Can't read device %s, error %d\n", drive->devicename, error);
904 free_drive(drive); /* give it back */
905 status = error;
906 }
907 /*
908 * At this point, check that the two copies
909 * are the same, and do something useful if
910 * not. In particular, consider which is
911 * newer, and what this means for the
912 * integrity of the data on the drive.
913 */
914 else {
915 vinum_conf.drives_used++; /* another drive in use */
916 /* Parse the configuration, and add it to the global configuration */
917 for (cptr = config_text; *cptr != '\0';) { /* love this style(9) */
918 volatile int parse_status; /* return value from parse_config */
919
920 for (eptr = config_line; (*cptr != '\n') && (*cptr != '\0');) /* until the end of the line */
921 *eptr++ = *cptr++;
922 *eptr = '\0'; /* and delimit */
923 if (setjmp(&command_fail) == 0) { /* come back here on error and continue */
924 parse_status = parse_config(config_line, &keyword_set, 1); /* parse the config line */
925 if (parse_status < 0) { /* error in config */
926 /*
927 * This config should have been parsed
928 * in user space. If we run into
929 * problems here, something serious is
930 * afoot. Complain and let the user
931 * snarf the config to see what's
932 * wrong.
933 */
934 log(LOG_ERR,
935 "vinum: Config error on %s, aborting integration\n",
936 drive->devicename);
937 free_drive(drive); /* give it back */
938 status = EINVAL;
939 }
940 }
941 while (*cptr == '\n')
942 cptr++; /* skip to next line */
943 }
944 }
945 drive->flags |= VF_CONFIGURED; /* read this drive's configuration */
946 }
947
948 Free(config_text);
949 Free(drivelist);
950 vinum_conf.flags &= ~VF_READING_CONFIG; /* no longer reading from disk */
951 if (status != 0)
952 printf("vinum: couldn't read configuration");
953 else
954 updateconfig(VF_READING_CONFIG); /* update from disk config */
955 return status;
956 }
957
958 /*
959 * Compare the modification dates of the drives, for qsort.
960 * Return 1 if a < b, 0 if a == b, 01 if a > b: in other
961 * words, sort backwards.
962 */
963 int
964 drivecmp(const void *va, const void *vb)
965 {
966 const struct drive *a = &DRIVE[*(const int *) va];
967 const struct drive *b = &DRIVE[*(const int *) vb];
968
969 if ((a->label.last_update.tv_sec == b->label.last_update.tv_sec)
970 && (a->label.last_update.tv_usec == b->label.last_update.tv_usec))
971 return 0;
972 else if ((a->label.last_update.tv_sec > b->label.last_update.tv_sec)
973 || ((a->label.last_update.tv_sec == b->label.last_update.tv_sec)
974 && (a->label.last_update.tv_usec > b->label.last_update.tv_usec)))
975 return -1;
976 else
977 return 1;
978 }
979 /* Local Variables: */
980 /* fill-column: 50 */
981 /* End: */
Cache object: 2bda8c639bfc6e2bb109feb0e78892e8
|