xref: /dragonfly/sys/dev/raid/vinum/vinum.c (revision ae24b5e0)
1 /*-
2  * Copyright (c) 1997, 1998
3  *	Nan Yang Computer Services Limited.  All rights reserved.
4  *
5  *  Written by Greg Lehey
6  *
7  *  This software is distributed under the so-called ``Berkeley
8  *  License'':
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. All advertising materials mentioning features or use of this software
19  *    must display the following acknowledgement:
20  *	This product includes software developed by Nan Yang Computer
21  *      Services Limited.
22  * 4. Neither the name of the Company nor the names of its contributors
23  *    may be used to endorse or promote products derived from this software
24  *    without specific prior written permission.
25  *
26  * This software is provided ``as is'', and any express or implied
27  * warranties, including, but not limited to, the implied warranties of
28  * merchantability and fitness for a particular purpose are disclaimed.
29  * In no event shall the company or contributors be liable for any
30  * direct, indirect, incidental, special, exemplary, or consequential
31  * damages (including, but not limited to, procurement of substitute
32  * goods or services; loss of use, data, or profits; or business
33  * interruption) however caused and on any theory of liability, whether
34  * in contract, strict liability, or tort (including negligence or
35  * otherwise) arising in any way out of the use of this software, even if
36  * advised of the possibility of such damage.
37  *
38  * $Id: vinum.c,v 1.33 2001/01/09 06:19:15 grog Exp grog $
39  * $FreeBSD: src/sys/dev/vinum/vinum.c,v 1.38.2.3 2003/01/07 12:14:16 joerg Exp $
40  */
41 
42 #define STATIC static					    /* nothing while we're testing XXX */
43 
44 #include "vinumhdr.h"
45 #include <sys/sysproto.h>				    /* for sync(2) */
46 #include <sys/poll.h>					    /* XXX: poll ops used in kq filters */
47 #include <sys/event.h>
48 #include <sys/udev.h>
49 #ifdef VINUMDEBUG
50 #include <sys/reboot.h>
51 int debug = 0;
52 extern int total_malloced;
53 extern int malloccount;
54 extern struct mc malloced[];
55 #endif
56 #include "request.h"
57 
58 struct dev_ops vinum_ops =
59 {
60 	{ "vinum", 0, D_DISK },
61 	.d_open =	vinumopen,
62 	.d_close =	vinumclose,
63 	.d_read =	physread,
64 	.d_write =	physwrite,
65 	.d_ioctl =	vinumioctl,
66 	.d_kqfilter =	vinumkqfilter,
67 	.d_strategy =	vinumstrategy,
68 	.d_dump =	vinumdump,
69 	.d_psize =	vinumsize,
70 };
71 
72 /* Called by main() during pseudo-device attachment. */
73 STATIC void vinumattach(void *);
74 
75 STATIC int vinum_modevent(module_t mod, modeventtype_t type, void *unused);
76 STATIC void vinum_initconf(void);
77 
78 struct _vinum_conf vinum_conf;				    /* configuration information */
79 cdev_t	vinum_super_dev;
80 cdev_t	vinum_wsuper_dev;
81 cdev_t	vinum_daemon_dev;
82 
83 /*
84  * Called by main() during pseudo-device attachment.  All we need
85  * to do is allocate enough space for devices to be configured later, and
86  * add devsw entries.
87  */
88 void
89 vinumattach(void *dummy)
90 {
91     char *cp, *cp1, *cp2, **drives;
92     int i, rv;
93     struct volume *vol;
94 
95     /* modload should prevent multiple loads, so this is worth a panic */
96     if ((vinum_conf.flags & VF_LOADED) != 0)
97 	panic("vinum: already loaded");
98 
99     log(LOG_INFO, "vinum: loaded\n");
100     vinum_conf.flags |= VF_LOADED;			    /* we're loaded now */
101 
102     daemonq = NULL;					    /* initialize daemon's work queue */
103     dqend = NULL;
104 
105 #if 0
106     dev_ops_add(&vinum_ops, 0, 0);
107 #endif
108 
109     vinum_initconf();
110 
111     /*
112      * Create superdev, wrongsuperdev, and controld devices.
113      */
114     vinum_super_dev =	make_dev(&vinum_ops, VINUM_SUPERDEV,
115 				 UID_ROOT, GID_WHEEL, 0600,
116 				 VINUM_SUPERDEV_BASE);
117     vinum_wsuper_dev =	make_dev(&vinum_ops, VINUM_WRONGSUPERDEV,
118 				 UID_ROOT, GID_WHEEL, 0600,
119 				 VINUM_WRONGSUPERDEV_BASE);
120     vinum_daemon_dev = 	make_dev(&vinum_ops, VINUM_DAEMON_DEV,
121 				 UID_ROOT, GID_WHEEL, 0600,
122 				 VINUM_DAEMON_DEV_BASE);
123 
124     /*
125      * See if the loader has passed us a disk to
126      * read the initial configuration from.
127      */
128     if ((cp = kgetenv("vinum.drives")) != NULL) {
129 	for (cp1 = cp, i = 0, drives = NULL; *cp1 != '\0'; i++) {
130 	    cp2 = cp1;
131 	    while (*cp1 != '\0' && *cp1 != ',' && *cp1 != ' ')
132 		cp1++;
133 	    if (*cp1 != '\0')
134 		*cp1++ = '\0';
135 	    drives = krealloc(drives, (unsigned long)((i + 1) * sizeof(char *)),
136 			     M_TEMP, M_WAITOK);
137 	    drives[i] = cp2;
138 	}
139 	if (i == 0)
140 	    goto bailout;
141 	rv = vinum_scandisk(drives, i);
142 	if (rv)
143 	    log(LOG_NOTICE, "vinum_scandisk() returned %d", rv);
144     bailout:
145 	kfree(drives, M_TEMP);
146     }
147     if ((cp = kgetenv("vinum.root")) != NULL) {
148 	for (i = 0; i < vinum_conf.volumes_used; i++) {
149 	    vol = &vinum_conf.volume[i];
150 	    if ((vol->state == volume_up)
151 		&& (strcmp (vol->name, cp) == 0)
152 	    ) {
153 		rootdev = make_dev(&vinum_ops, i, UID_ROOT, GID_OPERATOR,
154 				0640, VINUM_BASE "vinumroot");
155 		udev_dict_set_cstr(rootdev, "subsystem", "raid");
156 		udev_dict_set_cstr(rootdev, "disk-type", "raid");
157 		log(LOG_INFO, "vinum: using volume %s for root device\n", cp);
158 		break;
159 	    }
160 	}
161     }
162 }
163 
164 /*
165  * Check if we have anything open.  If confopen is != 0,
166  * that goes for the super device as well, otherwise
167  * only for volumes.
168  *
169  * Return 0 if not inactive, 1 if inactive.
170  */
171 int
172 vinum_inactive(int confopen)
173 {
174     int i;
175     int can_do = 1;					    /* assume we can do it */
176 
177     if (confopen && (vinum_conf.flags & VF_OPEN))	    /* open by vinum(8)? */
178 	return 0;					    /* can't do it while we're open */
179     lock_config();
180     for (i = 0; i < vinum_conf.volumes_allocated; i++) {
181 	if ((VOL[i].state > volume_down)
182 	    && (VOL[i].flags & VF_OPEN)) {		    /* volume is open */
183 	    can_do = 0;
184 	    break;
185 	}
186     }
187     unlock_config();
188     return can_do;
189 }
190 
191 /*
192  * Free all structures.
193  * If cleardrive is 0, save the configuration; otherwise
194  * remove the configuration from the drive.
195  *
196  * Before coming here, ensure that no volumes are open.
197  */
198 void
199 free_vinum(int cleardrive)
200 {
201     union daemoninfo di = { .nothing = 0 };
202     int i;
203     int drives_allocated = vinum_conf.drives_allocated;
204 
205     if (DRIVE != NULL) {
206 	if (cleardrive) {				    /* remove the vinum config */
207 	    for (i = 0; i < drives_allocated; i++)
208 		remove_drive(i);			    /* remove the drive */
209 	} else {					    /* keep the config */
210 	    for (i = 0; i < drives_allocated; i++)
211 		free_drive(&DRIVE[i]);			    /* close files and things */
212 	}
213 	Free(DRIVE);
214     }
215     while ((vinum_conf.flags & (VF_STOPPING | VF_DAEMONOPEN))
216 	== (VF_STOPPING | VF_DAEMONOPEN)) {		    /* at least one daemon open, we're stopping */
217 	queue_daemon_request(daemonrq_return, di);	    /* stop the daemon */
218 	tsleep(&vinumclose, 0, "vstop", 1);		    /* and wait for it */
219     }
220     if (SD != NULL) {
221 	for (i = 0; i < vinum_conf.subdisks_allocated; i++) {
222 	    struct sd *sd = &vinum_conf.sd[i];
223 	    if (sd->sd_dev) {
224 		destroy_dev(sd->sd_dev);
225 		sd->sd_dev = NULL;
226 	    }
227 	}
228 	Free(SD);
229     }
230     if (PLEX != NULL) {
231 	for (i = 0; i < vinum_conf.plexes_allocated; i++) {
232 	    struct plex *plex = &vinum_conf.plex[i];
233 
234 	    if (plex->plex_dev) {
235 		destroy_dev(plex->plex_dev);
236 		plex->plex_dev = NULL;
237 	    }
238 
239 	    if (plex->state != plex_unallocated) {	    /* we have real data there */
240 		if (plex->sdnos)
241 		    Free(plex->sdnos);
242 	    }
243 	}
244 	Free(PLEX);
245     }
246     if (VOL != NULL) {
247 	for (i = 0; i < vinum_conf.volumes_allocated; i++) {
248 	    struct volume *vol = &vinum_conf.volume[i];
249 
250 	    if (vol->vol_dev) {
251 		destroy_dev(vol->vol_dev);
252 		vol->vol_dev = NULL;
253 	    }
254 	}
255 	Free(VOL);
256     }
257     bzero(&vinum_conf, sizeof(vinum_conf));
258     vinum_initconf();
259 }
260 
261 STATIC void
262 vinum_initconf(void)
263 {
264     vinum_conf.physbufs = nswbuf_kva / 2 + 1;
265 
266     /* allocate space: drives... */
267     DRIVE = (struct drive *) Malloc(sizeof(struct drive) * INITIAL_DRIVES);
268     CHECKALLOC(DRIVE, "vinum: no memory\n");
269     bzero(DRIVE, sizeof(struct drive) * INITIAL_DRIVES);
270     vinum_conf.drives_allocated = INITIAL_DRIVES;
271     vinum_conf.drives_used = 0;
272 
273     /* volumes, ... */
274     VOL = (struct volume *) Malloc(sizeof(struct volume) * INITIAL_VOLUMES);
275     CHECKALLOC(VOL, "vinum: no memory\n");
276     bzero(VOL, sizeof(struct volume) * INITIAL_VOLUMES);
277     vinum_conf.volumes_allocated = INITIAL_VOLUMES;
278     vinum_conf.volumes_used = 0;
279 
280     /* plexes, ... */
281     PLEX = (struct plex *) Malloc(sizeof(struct plex) * INITIAL_PLEXES);
282     CHECKALLOC(PLEX, "vinum: no memory\n");
283     bzero(PLEX, sizeof(struct plex) * INITIAL_PLEXES);
284     vinum_conf.plexes_allocated = INITIAL_PLEXES;
285     vinum_conf.plexes_used = 0;
286 
287     /* and subdisks */
288     SD = (struct sd *) Malloc(sizeof(struct sd) * INITIAL_SUBDISKS);
289     CHECKALLOC(SD, "vinum: no memory\n");
290     bzero(SD, sizeof(struct sd) * INITIAL_SUBDISKS);
291     vinum_conf.subdisks_allocated = INITIAL_SUBDISKS;
292     vinum_conf.subdisks_used = 0;
293 }
294 
295 STATIC int
296 vinum_modevent(module_t mod, modeventtype_t type, void *unused)
297 {
298     switch (type) {
299     case MOD_LOAD:
300 	vinumattach(NULL);
301 	return 0;					    /* OK */
302     case MOD_UNLOAD:
303 	if (!vinum_inactive(1))				    /* is anything open? */
304 	    return EBUSY;				    /* yes, we can't do it */
305 	vinum_conf.flags |= VF_STOPPING;		    /* note that we want to stop */
306 	sys_sync(NULL);			    /* write out buffers */
307 	free_vinum(0);					    /* clean up */
308 
309 	if (vinum_super_dev) {
310 	    destroy_dev(vinum_super_dev);
311 	    vinum_super_dev = NULL;
312 	}
313 	if (vinum_wsuper_dev) {
314 	    destroy_dev(vinum_wsuper_dev);
315 	    vinum_wsuper_dev = NULL;
316 	}
317 	if (vinum_daemon_dev) {
318 	    destroy_dev(vinum_daemon_dev);
319 	    vinum_daemon_dev = NULL;
320 	}
321 
322 	sync_devs();
323 #ifdef VINUMDEBUG
324 	if (total_malloced) {
325 	    int i;
326 #ifdef INVARIANTS
327 	    int *poke;
328 #endif
329 
330 	    for (i = 0; i < malloccount; i++) {
331 		if (debug & DEBUG_WARNINGS)		    /* want to hear about them */
332 		    log(LOG_WARNING,
333 			"vinum: exiting with %d bytes malloced from %s:%d\n",
334 			malloced[i].size,
335 			malloced[i].file,
336 			malloced[i].line);
337 #ifdef INVARIANTS
338 		poke = &((int *) malloced[i].address)
339 		    [malloced[i].size / (2 * sizeof(int))]; /* middle of the area */
340 		if (*poke == 0xdeadc0de)		    /* already freed */
341 		    log(LOG_ERR,
342 			"vinum: exiting with malloc table inconsistency at %p from %s:%d\n",
343 			malloced[i].address,
344 			malloced[i].file,
345 			malloced[i].line);
346 #endif
347 		Free(malloced[i].address);
348 	    }
349 	}
350 #endif
351 	dev_ops_remove_all(&vinum_ops);
352 	log(LOG_INFO, "vinum: unloaded\n");		    /* tell the world */
353 	return 0;
354     default:
355 	break;
356     }
357     return 0;
358 }
359 
360 moduledata_t vinum_mod =
361 {
362     "vinum",
363     (modeventhand_t) vinum_modevent,
364     0
365 };
366 DECLARE_MODULE(vinum, vinum_mod, SI_SUB_RAID, SI_ORDER_MIDDLE);
367 
368 /* ARGSUSED */
369 /* Open a vinum object */
370 int
371 vinumopen(struct dev_open_args *ap)
372 {
373     cdev_t dev = ap->a_head.a_dev;
374     int error;
375     unsigned int index;
376     struct volume *vol;
377     struct plex *plex;
378     struct sd *sd;
379     int devminor;					    /* minor number */
380 
381     devminor = minor(dev);
382     error = 0;
383     /* First, decide what we're looking at */
384     switch (DEVTYPE(dev)) {
385     case VINUM_VOLUME_TYPE:
386 	index = Volno(dev);
387 	if (index >= vinum_conf.volumes_allocated)
388 	    return ENXIO;				    /* no such device */
389 	vol = &VOL[index];
390 
391 	switch (vol->state) {
392 	case volume_unallocated:
393 	case volume_uninit:
394 	    return ENXIO;
395 
396 	case volume_up:
397 	    vol->flags |= VF_OPEN;			    /* note we're open */
398 	    return 0;
399 
400 	case volume_down:
401 	    return EIO;
402 
403 	default:
404 	    return EINVAL;
405 	}
406 
407     case VINUM_PLEX_TYPE:
408 	if (Volno(dev) >= vinum_conf.volumes_allocated)
409 	    return ENXIO;
410 	/* FALLTHROUGH */
411 
412     case VINUM_RAWPLEX_TYPE:
413 	index = Plexno(dev);				    /* get plex index in vinum_conf */
414 	if (index >= vinum_conf.plexes_allocated)
415 	    return ENXIO;				    /* no such device */
416 	plex = &PLEX[index];
417 
418 	switch (plex->state) {
419 	case plex_referenced:
420 	case plex_unallocated:
421 	    return EINVAL;
422 
423 	default:
424 	    plex->flags |= VF_OPEN;			    /* note we're open */
425 	    return 0;
426 	}
427 
428     case VINUM_SD_TYPE:
429 	if ((Volno(dev) >= vinum_conf.volumes_allocated)    /* no such volume */
430 	||(Plexno(dev) >= vinum_conf.plexes_allocated))	    /* or no such plex */
431 	    return ENXIO;				    /* no such device */
432 
433 	/* FALLTHROUGH */
434 
435     case VINUM_RAWSD_TYPE:
436 	index = Sdno(dev);				    /* get the subdisk number */
437 	if ((index >= vinum_conf.subdisks_allocated)	    /* not a valid SD entry */
438 	||(SD[index].state < sd_init))			    /* or SD is not real */
439 	    return ENXIO;				    /* no such device */
440 	sd = &SD[index];
441 
442 	/*
443 	 * Opening a subdisk is always a special operation, so we
444 	 * ignore the state as long as it represents a real subdisk
445 	 */
446 	switch (sd->state) {
447 	case sd_unallocated:
448 	case sd_uninit:
449 	    return EINVAL;
450 
451 	default:
452 	    sd->flags |= VF_OPEN;			    /* note we're open */
453 	    return 0;
454 	}
455 
456     case VINUM_SUPERDEV_TYPE:
457 	error = priv_check_cred(ap->a_cred, PRIV_ROOT, 0);  /* are we root? */
458 	if (error == 0) {				    /* yes, can do */
459 	    if (devminor == VINUM_DAEMON_DEV)		    /* daemon device */
460 		vinum_conf.flags |= VF_DAEMONOPEN;	    /* we're open */
461 	    else if (devminor == VINUM_SUPERDEV)
462 		vinum_conf.flags |= VF_OPEN;		    /* we're open */
463 	    else
464 		error = ENODEV;				    /* nothing, maybe a debug mismatch */
465 	}
466 	return error;
467 
468 	/* Vinum drives are disks.  We already have a disk
469 	 * driver, so don't handle them here */
470     case VINUM_DRIVE_TYPE:
471     default:
472 	return ENODEV;					    /* don't know what to do with these */
473     }
474 }
475 
476 /* ARGSUSED */
477 int
478 vinumclose(struct dev_close_args *ap)
479 {
480     cdev_t dev = ap->a_head.a_dev;
481     unsigned int index;
482     struct volume *vol;
483     int devminor;
484 
485     devminor = minor(dev);
486     index = Volno(dev);
487     /* First, decide what we're looking at */
488     switch (DEVTYPE(dev)) {
489     case VINUM_VOLUME_TYPE:
490 	if (index >= vinum_conf.volumes_allocated)
491 	    return ENXIO;				    /* no such device */
492 	vol = &VOL[index];
493 
494 	switch (vol->state) {
495 	case volume_unallocated:
496 	case volume_uninit:
497 	    return ENXIO;
498 
499 	case volume_up:
500 	    vol->flags &= ~VF_OPEN;			    /* reset our flags */
501 	    return 0;
502 
503 	case volume_down:
504 	    return EIO;
505 
506 	default:
507 	    return EINVAL;
508 	}
509 
510     case VINUM_PLEX_TYPE:
511 	if (Volno(dev) >= vinum_conf.volumes_allocated)
512 	    return ENXIO;
513 	/* FALLTHROUGH */
514 
515     case VINUM_RAWPLEX_TYPE:
516 	index = Plexno(dev);				    /* get plex index in vinum_conf */
517 	if (index >= vinum_conf.plexes_allocated)
518 	    return ENXIO;				    /* no such device */
519 	PLEX[index].flags &= ~VF_OPEN;			    /* reset our flags */
520 	return 0;
521 
522     case VINUM_SD_TYPE:
523 	if ((Volno(dev) >= vinum_conf.volumes_allocated) || /* no such volume */
524 	    (Plexno(dev) >= vinum_conf.plexes_allocated))   /* or no such plex */
525 	    return ENXIO;				    /* no such device */
526 	/* FALLTHROUGH */
527 
528     case VINUM_RAWSD_TYPE:
529 	index = Sdno(dev);				    /* get the subdisk number */
530 	if (index >= vinum_conf.subdisks_allocated)
531 	    return ENXIO;				    /* no such device */
532 	SD[index].flags &= ~VF_OPEN;			    /* reset our flags */
533 	return 0;
534 
535     case VINUM_SUPERDEV_TYPE:
536 	/*
537 	 * don't worry about whether we're root:
538 	 * nobody else would get this far.
539 	 */
540 	if (devminor == VINUM_SUPERDEV)			    /* normal superdev */
541 	    vinum_conf.flags &= ~VF_OPEN;		    /* no longer open */
542 	else if (devminor == VINUM_DAEMON_DEV) {	    /* the daemon device */
543 	    vinum_conf.flags &= ~VF_DAEMONOPEN;		    /* no longer open */
544 	    if (vinum_conf.flags & VF_STOPPING)		    /* we're stopping, */
545 		wakeup(&vinumclose);			    /* we can continue stopping now */
546 	}
547 	return 0;
548 
549     case VINUM_DRIVE_TYPE:
550     default:
551 	return ENODEV;					    /* don't know what to do with these */
552     }
553 }
554 
555 /* size routine */
556 int
557 vinumsize(struct dev_psize_args *ap)
558 {
559     cdev_t dev = ap->a_head.a_dev;
560     struct volume *vol;
561 
562     vol = &VOL[Volno(dev)];
563 
564     if (vol->state == volume_up) {
565 	ap->a_result = (int64_t)vol->size;
566 	return(0);
567     } else {
568 	return(ENXIO);
569     }
570 }
571 
572 int
573 vinumdump(struct dev_dump_args *ap)
574 {
575     /* Not implemented. */
576     return ENXIO;
577 }
578 
579 void
580 vinumfilt_detach(struct knote *kn) {}
581 
582 int
583 vinumfilt_rd(struct knote *kn, long hint)
584 {
585     cdev_t dev = (cdev_t)kn->kn_hook;
586 
587     if (seltrue(dev, POLLIN | POLLRDNORM))
588         return (1);
589 
590     return (0);
591 }
592 
593 int
594 vinumfilt_wr(struct knote *kn, long hint)
595 {
596     /* Writing is always OK */
597     return (1);
598 }
599 
600 struct filterops vinumfiltops_rd =
601     { FILTEROP_ISFD, NULL, vinumfilt_detach, vinumfilt_rd };
602 struct filterops vinumfiltops_wr =
603     { FILTEROP_ISFD, NULL, vinumfilt_detach, vinumfilt_wr };
604 
605 int
606 vinumkqfilter(struct dev_kqfilter_args *ap)
607 {
608     if (ap->a_kn->kn_filter == EVFILT_READ) {
609         ap->a_kn->kn_fop = &vinumfiltops_rd;
610         ap->a_kn->kn_hook = (caddr_t)ap->a_head.a_dev;
611         ap->a_result = 0;
612     } else if (ap->a_kn->kn_filter == EVFILT_WRITE) {
613 	ap->a_kn->kn_fop = &vinumfiltops_wr;
614         ap->a_result = 0;
615     } else {
616         ap->a_result = EOPNOTSUPP;
617     }
618 
619     return (0);
620 }
621