xref: /titanic_41/usr/src/uts/common/io/lvm/md/md.c (revision b885580b43755ee4ea1e280b85428893d2ba9291)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 
22 /*
23  * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
24  * Use is subject to license terms.
25  */
26 
27 /*
28  * Md - is the meta-disk driver.   It sits below the UFS file system
29  * but above the 'real' disk drivers, xy, id, sd etc.
30  *
31  * To the UFS software, md looks like a normal driver, since it has
32  * the normal kinds of entries in the bdevsw and cdevsw arrays. So
33  * UFS accesses md in the usual ways.  In particular, the strategy
34  * routine, mdstrategy(), gets called by fbiwrite(), ufs_getapage(),
35  * and ufs_writelbn().
36  *
37  * Md maintains an array of minor devices (meta-partitions).   Each
38  * meta partition stands for a matrix of real partitions, in rows
39  * which are not necessarily of equal length.	Md maintains a table,
40  * with one entry for each meta-partition,  which lists the rows and
41  * columns of actual partitions, and the job of the strategy routine
42  * is to translate from the meta-partition device and block numbers
43  * known to UFS into the actual partitions' device and block numbers.
44  *
45  * See below, in mdstrategy(), mdreal(), and mddone() for details of
46  * this translation.
47  */
48 
49 /*
50  * Driver for Virtual Disk.
51  */
52 
53 #include <sys/user.h>
54 #include <sys/sysmacros.h>
55 #include <sys/conf.h>
56 #include <sys/stat.h>
57 #include <sys/errno.h>
58 #include <sys/param.h>
59 #include <sys/systm.h>
60 #include <sys/file.h>
61 #include <sys/open.h>
62 #include <sys/dkio.h>
63 #include <sys/vtoc.h>
64 #include <sys/cmn_err.h>
65 #include <sys/ddi.h>
66 #include <sys/sunddi.h>
67 #include <sys/debug.h>
68 #include <sys/utsname.h>
69 #include <sys/lvm/mdvar.h>
70 #include <sys/lvm/md_names.h>
71 #include <sys/lvm/md_mddb.h>
72 #include <sys/lvm/md_sp.h>
73 #include <sys/types.h>
74 #include <sys/kmem.h>
75 #include <sys/cladm.h>
76 #include <sys/priv_names.h>
77 #include <sys/modhash.h>
78 
79 #ifndef	lint
80 char 		_depends_on[] = "strmod/rpcmod";
81 #endif	/* lint */
82 int		md_init_debug	= 0;	/* module binding debug */
83 
84 /*
85  * Tunable to turn off the failfast behavior.
86  */
87 int		md_ff_disable = 0;
88 
89 /*
90  * dynamically allocated list of non FF driver names - needs to
91  * be freed when md is detached.
92  */
93 char	**non_ff_drivers = NULL;
94 
95 md_krwlock_t	md_unit_array_rw;	/* protects all unit arrays */
96 md_krwlock_t	nm_lock;		/* protects all the name spaces */
97 
98 md_resync_t	md_cpr_resync;
99 
100 extern char	svm_bootpath[];
101 #define	SVM_PSEUDO_STR	"/pseudo/md@0:"
102 
103 #define		VERSION_LENGTH	6
104 #define		VERSION		"1.0"
105 
106 /*
107  * Keep track of possible 'orphan' entries in the name space
108  */
109 int		*md_nm_snarfed = NULL;
110 
111 /*
112  * Global tunable giving the percentage of free space left in replica during
113  * conversion of non-devid style replica to devid style replica.
114  */
115 int		md_conv_perc = MDDB_DEVID_CONV_PERC;
116 
117 #ifdef	DEBUG
118 /* debug code to verify framework exclusion guarantees */
119 int		md_in;
120 kmutex_t	md_in_mx;			/* used to md global stuff */
121 #define	IN_INIT		0x01
122 #define	IN_FINI		0x02
123 #define	IN_ATTACH	0x04
124 #define	IN_DETACH	0x08
125 #define	IN_OPEN		0x10
126 #define	MD_SET_IN(x) {						\
127 	mutex_enter(&md_in_mx);					\
128 	if (md_in)						\
129 		debug_enter("MD_SET_IN exclusion lost");	\
130 	if (md_in & x)						\
131 		debug_enter("MD_SET_IN already set");		\
132 	md_in |= x;						\
133 	mutex_exit(&md_in_mx);					\
134 }
135 
136 #define	MD_CLR_IN(x) {						\
137 	mutex_enter(&md_in_mx);					\
138 	if (md_in & ~(x))					\
139 		debug_enter("MD_CLR_IN exclusion lost");	\
140 	if (!(md_in & x))					\
141 		debug_enter("MD_CLR_IN already clr");		\
142 	md_in &= ~x;						\
143 	mutex_exit(&md_in_mx);					\
144 }
145 #else	/* DEBUG */
146 #define	MD_SET_IN(x)
147 #define	MD_CLR_IN(x)
148 #endif	/* DEBUG */
149 hrtime_t savetime1, savetime2;
150 
151 
152 /*
153  * list things protected by md_mx even if they aren't
154  * used in this file.
155  */
156 kmutex_t	md_mx;			/* used to md global stuff */
157 kcondvar_t	md_cv;			/* md_status events */
158 int		md_status = 0;		/* global status for the meta-driver */
159 int		md_num_daemons = 0;
160 int		md_ioctl_cnt = 0;
161 int		md_mtioctl_cnt = 0;	/* multithreaded ioctl cnt */
162 uint_t		md_mdelay = 10;		/* variable so can be patched */
163 
164 int		(*mdv_strategy_tstpnt)(buf_t *, int, void*);
165 
166 major_t		md_major, md_major_targ;
167 
168 unit_t		md_nunits = MD_MAXUNITS;
169 set_t		md_nsets = MD_MAXSETS;
170 int		md_nmedh = 0;
171 char		*md_med_trans_lst = NULL;
172 md_set_t	md_set[MD_MAXSETS];
173 md_set_io_t	md_set_io[MD_MAXSETS];
174 
175 md_krwlock_t	hsp_rwlp;		/* protects hot_spare_interface */
176 md_krwlock_t	ni_rwlp;		/* protects notify_interface */
177 md_ops_t	**md_ops = NULL;
178 ddi_modhandle_t	*md_mods = NULL;
179 md_ops_t	*md_opslist;
180 clock_t		md_hz;
181 md_event_queue_t	*md_event_queue = NULL;
182 
183 int		md_in_upgrade;
184 int		md_keep_repl_state;
185 int		md_devid_destroy;
186 
187 /* for sending messages thru a door to userland */
188 door_handle_t	mdmn_door_handle = NULL;
189 int		mdmn_door_did = -1;
190 
191 dev_info_t		*md_devinfo = NULL;
192 
193 md_mn_nodeid_t	md_mn_mynode_id = ~0u;	/* My node id (for multi-node sets) */
194 
195 static	uint_t		md_ocnt[OTYPCNT];
196 
197 static int		mdinfo(dev_info_t *, ddi_info_cmd_t, void *, void **);
198 static int		mdattach(dev_info_t *, ddi_attach_cmd_t);
199 static int		mddetach(dev_info_t *, ddi_detach_cmd_t);
200 static int		mdopen(dev_t *, int, int, cred_t *);
201 static int		mdclose(dev_t, int, int, cred_t *);
202 static int		mddump(dev_t, caddr_t, daddr_t, int);
203 static int		mdread(dev_t, struct uio *, cred_t *);
204 static int		mdwrite(dev_t, struct uio *, cred_t *);
205 static int		mdaread(dev_t, struct aio_req *, cred_t *);
206 static int		mdawrite(dev_t, struct aio_req *, cred_t *);
207 static int		mdioctl(dev_t, int, intptr_t, int, cred_t *, int *);
208 static int		mdprop_op(dev_t, dev_info_t *,
209 				ddi_prop_op_t, int, char *, caddr_t, int *);
210 
211 static struct cb_ops md_cb_ops = {
212 	mdopen,			/* open */
213 	mdclose,		/* close */
214 	mdstrategy,		/* strategy */
215 				/* print routine -- none yet */
216 	(int(*)(dev_t, char *))nulldev,
217 	mddump,			/* dump */
218 	mdread,			/* read */
219 	mdwrite,		/* write */
220 	mdioctl,		/* ioctl */
221 				/* devmap */
222 	(int(*)(dev_t, devmap_cookie_t, offset_t, size_t, size_t *,
223 			uint_t))nodev,
224 				/* mmap */
225 	(int(*)(dev_t, off_t, int))nodev,
226 				/* segmap */
227 	(int(*)(dev_t, off_t, struct as *, caddr_t *, off_t, unsigned,
228 		unsigned, unsigned, cred_t *))nodev,
229 	nochpoll,		/* poll */
230 	mdprop_op,		/* prop_op */
231 	0,			/* streamtab */
232 	(D_64BIT|D_MP|D_NEW),	/* driver compatibility flag */
233 	CB_REV,			/* cb_ops version */
234 	mdaread,		/* aread */
235 	mdawrite,		/* awrite */
236 };
237 
238 static struct dev_ops md_devops = {
239 	DEVO_REV,		/* dev_ops version */
240 	0,			/* device reference count */
241 	mdinfo,			/* info routine */
242 	nulldev,		/* identify routine */
243 	nulldev,		/* probe - not defined */
244 	mdattach,		/* attach routine */
245 	mddetach,		/* detach routine */
246 	nodev,			/* reset - not defined */
247 	&md_cb_ops,		/* driver operations */
248 	NULL,			/* bus operations */
249 	nodev,			/* power management */
250 	ddi_quiesce_not_needed,		/* quiesce */
251 };
252 
253 /*
254  * loadable module wrapper
255  */
256 #include <sys/modctl.h>
257 
258 static struct modldrv modldrv = {
259 	&mod_driverops,			/* type of module -- a pseudodriver */
260 	"Solaris Volume Manager base module", /* name of the module */
261 	&md_devops,			/* driver ops */
262 };
263 
264 static struct modlinkage modlinkage = {
265 	MODREV_1,
266 	(void *)&modldrv,
267 	NULL
268 };
269 
270 
271 /* md_medd.c */
272 extern	void	med_init(void);
273 extern	void	med_fini(void);
274 extern  void	md_devid_cleanup(set_t, uint_t);
275 
276 /* md_names.c */
277 extern void			*lookup_entry(struct nm_next_hdr *, set_t,
278 					side_t, mdkey_t, md_dev64_t, int);
279 extern struct nm_next_hdr	*get_first_record(set_t, int, int);
280 extern int			remove_entry(struct nm_next_hdr *,
281 					side_t, mdkey_t, int);
282 
283 int		md_maxphys	= 0;	/* maximum io size in bytes */
284 #define		MD_MAXBCOUNT	(1024 * 1024)
285 unsigned	md_maxbcount	= 0;	/* maximum physio size in bytes */
286 
287 /*
288  * Some md ioctls trigger io framework device tree operations.  An
289  * example is md ioctls that call md_resolve_bydevid(): which uses the
290  * io framework to resolve a devid. Such operations result in acquiring
291  * io framework locks (like ndi_devi_enter() of "/") while holding
292  * driver locks (like md_unit_writerlock()).
293  *
294  * The prop_op(9E) entry point is called from the devinfo driver with
295  * an active ndi_devi_enter of "/". To avoid deadlock, md's prop_op
296  * implementation must avoid taking a lock that is held per above md
297  * ioctl description: i.e. mdprop_op(9E) can't call md_unit_readerlock()
298  * without risking deadlock.
299  *
300  * To service "size" requests without risking deadlock, we maintain a
301  * "mnum->nblocks" sizemap (protected by a short-term global mutex).
302  */
303 static kmutex_t		md_nblocks_mutex;
304 static mod_hash_t	*md_nblocksmap;		/* mnum -> nblocks */
305 int			md_nblocksmap_size = 512;
306 
307 /*
308  * Maintain "mnum->nblocks" sizemap for mdprop_op use:
309  *
310  * Create: any code that establishes a unit's un_total_blocks needs the
311  * following type of call to establish nblocks for mdprop_op():
312  *	md_nblocks_set(mnum, un->c.un_total_blocks);"
313  *	NOTE: locate via cscope md_create_minor_node/md_create_unit_incore
314  *		...or  "MD_UNIT..*="
315  *
316  * Change: any code that changes a unit's un_total_blocks needs the
317  * following type of call to sync nblocks for mdprop_op():
318  *	md_nblocks_set(mnum, un->c.un_total_blocks);"
319  *	NOTE: locate via cscope for "un_total_blocks[ \t]*="
320  *
321  * Destroy: any code that deletes a unit needs the following type of call
322  * to sync nblocks for mdprop_op():
323  *	md_nblocks_set(mnum, -1ULL);
324  *	NOTE: locate via cscope md_remove_minor_node/md_destroy_unit_incore
325  *		...or  "MD_UNIT..*="
326  */
327 void
328 md_nblocks_set(minor_t mnum, uint64_t nblocks)
329 {
330 	mutex_enter(&md_nblocks_mutex);
331 	if (nblocks == -1ULL)
332 		(void) mod_hash_destroy(md_nblocksmap,
333 		    (mod_hash_key_t)(intptr_t)mnum);
334 	else
335 		(void) mod_hash_replace(md_nblocksmap,
336 		    (mod_hash_key_t)(intptr_t)mnum,
337 		    (mod_hash_val_t)(intptr_t)nblocks);
338 	mutex_exit(&md_nblocks_mutex);
339 }
340 
341 /* get the size of a mnum from "mnum->nblocks" sizemap */
342 uint64_t
343 md_nblocks_get(minor_t mnum)
344 {
345 	mod_hash_val_t	hv;
346 
347 	mutex_enter(&md_nblocks_mutex);
348 	if (mod_hash_find(md_nblocksmap,
349 	    (mod_hash_key_t)(intptr_t)mnum, &hv) == 0) {
350 		mutex_exit(&md_nblocks_mutex);
351 		return ((uint64_t)(intptr_t)hv);
352 	}
353 	mutex_exit(&md_nblocks_mutex);
354 	return (0);
355 }
356 
357 /* allocate/free dynamic space associated with driver globals */
358 void
359 md_global_alloc_free(int alloc)
360 {
361 	set_t	s;
362 
363 	if (alloc) {
364 		/* initialize driver global locks */
365 		cv_init(&md_cv, NULL, CV_DEFAULT, NULL);
366 		mutex_init(&md_mx, NULL, MUTEX_DEFAULT, NULL);
367 		rw_init(&md_unit_array_rw.lock, NULL, RW_DEFAULT, NULL);
368 		rw_init(&nm_lock.lock, NULL, RW_DEFAULT, NULL);
369 		rw_init(&ni_rwlp.lock, NULL, RW_DRIVER, NULL);
370 		rw_init(&hsp_rwlp.lock, NULL, RW_DRIVER, NULL);
371 		mutex_init(&md_cpr_resync.md_resync_mutex, NULL,
372 		    MUTEX_DEFAULT, NULL);
373 		mutex_init(&md_nblocks_mutex, NULL, MUTEX_DEFAULT, NULL);
374 
375 		/* initialize per set driver global locks */
376 		for (s = 0; s < MD_MAXSETS; s++) {
377 			/* initialize per set driver globals locks */
378 			mutex_init(&md_set[s].s_dbmx,
379 			    NULL, MUTEX_DEFAULT, NULL);
380 			mutex_init(&md_set_io[s].md_io_mx,
381 			    NULL, MUTEX_DEFAULT, NULL);
382 			cv_init(&md_set_io[s].md_io_cv,
383 			    NULL, CV_DEFAULT, NULL);
384 		}
385 	} else {
386 		/* destroy per set driver global locks */
387 		for (s = 0; s < MD_MAXSETS; s++) {
388 			cv_destroy(&md_set_io[s].md_io_cv);
389 			mutex_destroy(&md_set_io[s].md_io_mx);
390 			mutex_destroy(&md_set[s].s_dbmx);
391 		}
392 
393 		/* destroy driver global locks */
394 		mutex_destroy(&md_nblocks_mutex);
395 		mutex_destroy(&md_cpr_resync.md_resync_mutex);
396 		rw_destroy(&hsp_rwlp.lock);
397 		rw_destroy(&ni_rwlp.lock);
398 		rw_destroy(&nm_lock.lock);
399 		rw_destroy(&md_unit_array_rw.lock);
400 		mutex_destroy(&md_mx);
401 		cv_destroy(&md_cv);
402 	}
403 }
404 
405 int
406 _init(void)
407 {
408 	set_t	s;
409 	int	err;
410 
411 	MD_SET_IN(IN_INIT);
412 
413 	/* allocate dynamic space associated with driver globals */
414 	md_global_alloc_free(1);
415 
416 	/* initialize driver globals */
417 	md_major = ddi_name_to_major("md");
418 	md_hz = drv_usectohz(NUM_USEC_IN_SEC);
419 
420 	/* initialize tunable globals */
421 	if (md_maxphys == 0)		/* maximum io size in bytes */
422 		md_maxphys = maxphys;
423 	if (md_maxbcount == 0)		/* maximum physio size in bytes */
424 		md_maxbcount = MD_MAXBCOUNT;
425 
426 	/* initialize per set driver globals */
427 	for (s = 0; s < MD_MAXSETS; s++)
428 		md_set_io[s].io_state = MD_SET_ACTIVE;
429 
430 	/*
431 	 * NOTE: the framework does not currently guarantee exclusion
432 	 * between _init and attach after calling mod_install.
433 	 */
434 	MD_CLR_IN(IN_INIT);
435 	if ((err = mod_install(&modlinkage))) {
436 		MD_SET_IN(IN_INIT);
437 		md_global_alloc_free(0);	/* free dynamic space */
438 		MD_CLR_IN(IN_INIT);
439 	}
440 	return (err);
441 }
442 
443 int
444 _fini(void)
445 {
446 	int	err;
447 
448 	/*
449 	 * NOTE: the framework currently does not guarantee exclusion
450 	 * with attach until after mod_remove returns 0.
451 	 */
452 	if ((err = mod_remove(&modlinkage)))
453 		return (err);
454 
455 	MD_SET_IN(IN_FINI);
456 	md_global_alloc_free(0);	/* free dynamic space */
457 	MD_CLR_IN(IN_FINI);
458 	return (err);
459 }
460 
461 int
462 _info(struct modinfo *modinfop)
463 {
464 	return (mod_info(&modlinkage, modinfop));
465 }
466 
467 /* ARGSUSED */
468 static int
469 mdattach(dev_info_t *dip, ddi_attach_cmd_t cmd)
470 {
471 	int	len;
472 	unit_t	i;
473 	size_t	sz;
474 	char	ver[VERSION_LENGTH];
475 	char	**maj_str_array;
476 	char	*str, *str2;
477 
478 	MD_SET_IN(IN_ATTACH);
479 	md_in_upgrade = 0;
480 	md_keep_repl_state = 0;
481 	md_devid_destroy = 0;
482 
483 	if (cmd != DDI_ATTACH) {
484 		MD_CLR_IN(IN_ATTACH);
485 		return (DDI_FAILURE);
486 	}
487 
488 	if (md_devinfo != NULL) {
489 		MD_CLR_IN(IN_ATTACH);
490 		return (DDI_FAILURE);
491 	}
492 
493 	mddb_init();
494 
495 	if (md_start_daemons(TRUE)) {
496 		MD_CLR_IN(IN_ATTACH);
497 		mddb_unload();		/* undo mddb_init() allocations */
498 		return (DDI_FAILURE);
499 	}
500 
501 	/* clear the halted state */
502 	md_clr_status(MD_GBL_HALTED);
503 
504 	/* see if the diagnostic switch is on */
505 	if (ddi_prop_get_int(DDI_DEV_T_ANY, dip,
506 	    DDI_PROP_DONTPASS, "md_init_debug", 0))
507 		md_init_debug++;
508 
509 	/* see if the failfast disable switch is on */
510 	if (ddi_prop_get_int(DDI_DEV_T_ANY, dip,
511 	    DDI_PROP_DONTPASS, "md_ff_disable", 0))
512 		md_ff_disable++;
513 
514 	/* try and get the md_nmedh property */
515 	md_nmedh = ddi_prop_get_int(DDI_DEV_T_ANY, dip,
516 	    DDI_PROP_DONTPASS, "md_nmedh", MED_DEF_HOSTS);
517 	if ((md_nmedh <= 0) || (md_nmedh > MED_MAX_HOSTS))
518 		md_nmedh = MED_DEF_HOSTS;
519 
520 	/* try and get the md_med_trans_lst property */
521 	len = 0;
522 	if (ddi_prop_op(DDI_DEV_T_ANY, dip, PROP_LEN,
523 	    0, "md_med_trans_lst", NULL, &len) != DDI_PROP_SUCCESS ||
524 	    len == 0) {
525 		md_med_trans_lst = md_strdup("tcp");
526 	} else {
527 		md_med_trans_lst = kmem_zalloc((size_t)len, KM_SLEEP);
528 		if (ddi_prop_op(DDI_DEV_T_ANY, dip, PROP_LEN_AND_VAL_BUF,
529 		    0, "md_med_trans_lst", md_med_trans_lst, &len) !=
530 		    DDI_PROP_SUCCESS) {
531 			kmem_free(md_med_trans_lst, (size_t)len);
532 			md_med_trans_lst = md_strdup("tcp");
533 		}
534 	}
535 
536 	/*
537 	 * Must initialize the internal data structures before the
538 	 * any possible calls to 'goto attach_failure' as _fini
539 	 * routine references them.
540 	 */
541 	med_init();
542 
543 	md_ops = (md_ops_t **)kmem_zalloc(
544 	    sizeof (md_ops_t *) * MD_NOPS, KM_SLEEP);
545 	md_mods = (ddi_modhandle_t *)kmem_zalloc(
546 	    sizeof (ddi_modhandle_t) * MD_NOPS, KM_SLEEP);
547 
548 	/* try and get the md_xlate property */
549 	/* Should we only do this if upgrade? */
550 	len = sizeof (char) * 5;
551 	if (ddi_prop_op(DDI_DEV_T_ANY, dip, PROP_LEN_AND_VAL_BUF,
552 	    0, "md_xlate_ver", ver, &len) == DDI_PROP_SUCCESS) {
553 		if (strcmp(ver, VERSION) == 0) {
554 			len = 0;
555 			if (ddi_prop_op(DDI_DEV_T_ANY, dip,
556 			    PROP_LEN_AND_VAL_ALLOC, 0, "md_xlate",
557 			    (caddr_t)&md_tuple_table, &len) !=
558 			    DDI_PROP_SUCCESS) {
559 				if (md_init_debug)
560 					cmn_err(CE_WARN,
561 					    "md_xlate ddi_prop_op failed");
562 				goto attach_failure;
563 			} else {
564 				md_tuple_length =
565 				    len/(2 * ((int)sizeof (dev32_t)));
566 				md_in_upgrade = 1;
567 			}
568 
569 			/* Get target's name to major table */
570 			if (ddi_prop_lookup_string_array(DDI_DEV_T_ANY,
571 			    dip, DDI_PROP_DONTPASS,
572 			    "md_targ_nm_table", &maj_str_array,
573 			    &md_majortab_len) != DDI_PROP_SUCCESS) {
574 				md_majortab_len = 0;
575 				if (md_init_debug)
576 					cmn_err(CE_WARN, "md_targ_nm_table "
577 					    "ddi_prop_lookup_string_array "
578 					    "failed");
579 				goto attach_failure;
580 			}
581 
582 			md_major_tuple_table =
583 			    (struct md_xlate_major_table *)
584 			    kmem_zalloc(md_majortab_len *
585 			    sizeof (struct md_xlate_major_table), KM_SLEEP);
586 
587 			for (i = 0; i < md_majortab_len; i++) {
588 				/* Getting major name */
589 				str = strchr(maj_str_array[i], ' ');
590 				if (str == NULL)
591 					continue;
592 				*str = '\0';
593 				md_major_tuple_table[i].drv_name =
594 				    md_strdup(maj_str_array[i]);
595 
596 				/* Simplified atoi to get major number */
597 				str2 = str + 1;
598 				md_major_tuple_table[i].targ_maj = 0;
599 				while ((*str2 >= '0') && (*str2 <= '9')) {
600 					md_major_tuple_table[i].targ_maj *= 10;
601 					md_major_tuple_table[i].targ_maj +=
602 					    *str2++ - '0';
603 				}
604 				*str = ' ';
605 			}
606 			ddi_prop_free((void *)maj_str_array);
607 		} else {
608 			if (md_init_debug)
609 				cmn_err(CE_WARN, "md_xlate_ver is incorrect");
610 			goto attach_failure;
611 		}
612 	}
613 
614 	/*
615 	 * Check for properties:
616 	 * 	md_keep_repl_state and md_devid_destroy
617 	 * and set globals if these exist.
618 	 */
619 	md_keep_repl_state = ddi_getprop(DDI_DEV_T_ANY, dip,
620 	    0, "md_keep_repl_state", 0);
621 
622 	md_devid_destroy = ddi_getprop(DDI_DEV_T_ANY, dip,
623 	    0, "md_devid_destroy", 0);
624 
625 	if (MD_UPGRADE)
626 		md_major_targ = md_targ_name_to_major("md");
627 	else
628 		md_major_targ = 0;
629 
630 	/* allocate admin device node */
631 	if (ddi_create_priv_minor_node(dip, "admin", S_IFCHR,
632 	    MD_ADM_MINOR, DDI_PSEUDO, 0, NULL, PRIV_SYS_CONFIG, 0640))
633 		goto attach_failure;
634 
635 	if (ddi_prop_create(DDI_DEV_T_NONE, dip, DDI_PROP_CANSLEEP,
636 	    DDI_KERNEL_IOCTL, NULL, 0) != DDI_SUCCESS)
637 		goto attach_failure;
638 
639 	if (ddi_prop_update_int(DDI_DEV_T_NONE, dip,
640 	    "ddi-abrwrite-supported", 1) != DDI_SUCCESS)
641 		goto attach_failure;
642 
643 	/* these could have been cleared by a detach */
644 	md_nunits = MD_MAXUNITS;
645 	md_nsets = MD_MAXSETS;
646 
647 	sz = sizeof (void *) * MD_MAXUNITS;
648 	if (md_set[0].s_un == NULL)
649 		md_set[0].s_un = kmem_zalloc(sz, KM_SLEEP);
650 	if (md_set[0].s_ui == NULL)
651 		md_set[0].s_ui = kmem_zalloc(sz, KM_SLEEP);
652 
653 	md_devinfo = dip;
654 
655 	/*
656 	 * Only allocate device node for root mirror metadevice.
657 	 * Don't pre-allocate unnecessary device nodes (thus slowing down a
658 	 * boot when we attach).
659 	 * We can't read the mddbs in attach.  The mddbs will be read
660 	 * by metainit during the boot process when it is doing the
661 	 * auto-take processing and any other minor nodes will be
662 	 * allocated at that point.
663 	 *
664 	 * There are two scenarios to be aware of here:
665 	 * 1) when we are booting from a mirrored root we need the root
666 	 *    metadevice to exist very early (during vfs_mountroot processing)
667 	 * 2) we need all of the nodes to be created so that any mnttab entries
668 	 *    will succeed (handled by metainit reading the mddb during boot).
669 	 */
670 	if (strncmp(SVM_PSEUDO_STR, svm_bootpath, sizeof (SVM_PSEUDO_STR) - 1)
671 	    == 0) {
672 		char *p;
673 		int mnum = 0;
674 
675 		/*
676 		 * The svm_bootpath string looks something like
677 		 * /pseudo/md@0:0,150,blk where 150 is the minor number
678 		 * in this example so we need to set the pointer p onto
679 		 * the first digit of the minor number and convert it
680 		 * from ascii.
681 		 */
682 		for (p = svm_bootpath + sizeof (SVM_PSEUDO_STR) + 1;
683 		    *p >= '0' && *p <= '9'; p++) {
684 			mnum *= 10;
685 			mnum += *p - '0';
686 		}
687 
688 		if (md_create_minor_node(0, mnum)) {
689 			kmem_free(md_set[0].s_un, sz);
690 			kmem_free(md_set[0].s_ui, sz);
691 			goto attach_failure;
692 		}
693 	}
694 
695 	/* create the hash to store the meta device sizes */
696 	md_nblocksmap = mod_hash_create_idhash("md_nblocksmap",
697 	    md_nblocksmap_size, mod_hash_null_valdtor);
698 
699 	MD_CLR_IN(IN_ATTACH);
700 	return (DDI_SUCCESS);
701 
702 attach_failure:
703 	/*
704 	 * Use our own detach routine to toss any stuff we allocated above.
705 	 * NOTE: detach will call md_halt to free the mddb_init allocations.
706 	 */
707 	MD_CLR_IN(IN_ATTACH);
708 	if (mddetach(dip, DDI_DETACH) != DDI_SUCCESS)
709 		cmn_err(CE_WARN, "detach from attach failed");
710 	return (DDI_FAILURE);
711 }
712 
713 /* ARGSUSED */
714 static int
715 mddetach(dev_info_t *dip, ddi_detach_cmd_t cmd)
716 {
717 	extern int	check_active_locators();
718 	set_t		s;
719 	size_t		sz;
720 	int		len;
721 
722 	MD_SET_IN(IN_DETACH);
723 
724 	/* check command */
725 	if (cmd != DDI_DETACH) {
726 		MD_CLR_IN(IN_DETACH);
727 		return (DDI_FAILURE);
728 	}
729 
730 	/*
731 	 * if we have not already halted yet we have no active config
732 	 * then automatically initiate a halt so we can detach.
733 	 */
734 	if (!(md_get_status() & MD_GBL_HALTED)) {
735 		if (check_active_locators() == 0) {
736 			/*
737 			 * NOTE: a successful md_halt will have done the
738 			 * mddb_unload to free allocations done in mddb_init
739 			 */
740 			if (md_halt(MD_NO_GBL_LOCKS_HELD)) {
741 				cmn_err(CE_NOTE, "md:detach: "
742 				    "Could not halt Solaris Volume Manager");
743 				MD_CLR_IN(IN_DETACH);
744 				return (DDI_FAILURE);
745 			}
746 		}
747 
748 		/* fail detach if we have not halted */
749 		if (!(md_get_status() & MD_GBL_HALTED)) {
750 			MD_CLR_IN(IN_DETACH);
751 			return (DDI_FAILURE);
752 		}
753 	}
754 
755 	/* must be in halted state, this will be cleared on next attach */
756 	ASSERT(md_get_status() & MD_GBL_HALTED);
757 
758 	/* cleanup attach allocations and initializations */
759 	md_major_targ = 0;
760 
761 	sz = sizeof (void *) * md_nunits;
762 	for (s = 0; s < md_nsets; s++) {
763 		if (md_set[s].s_un != NULL) {
764 			kmem_free(md_set[s].s_un, sz);
765 			md_set[s].s_un = NULL;
766 		}
767 
768 		if (md_set[s].s_ui != NULL) {
769 			kmem_free(md_set[s].s_ui, sz);
770 			md_set[s].s_ui = NULL;
771 		}
772 	}
773 	md_nunits = 0;
774 	md_nsets = 0;
775 	md_nmedh = 0;
776 
777 	if (non_ff_drivers != NULL) {
778 		int	i;
779 
780 		for (i = 0; non_ff_drivers[i] != NULL; i++)
781 			kmem_free(non_ff_drivers[i],
782 			    strlen(non_ff_drivers[i]) + 1);
783 
784 		/* free i+1 entries because there is a null entry at list end */
785 		kmem_free(non_ff_drivers, (i + 1) * sizeof (char *));
786 		non_ff_drivers = NULL;
787 	}
788 
789 	if (md_med_trans_lst != NULL) {
790 		kmem_free(md_med_trans_lst, strlen(md_med_trans_lst) + 1);
791 		md_med_trans_lst = NULL;
792 	}
793 
794 	if (md_mods != NULL) {
795 		kmem_free(md_mods, sizeof (ddi_modhandle_t) * MD_NOPS);
796 		md_mods = NULL;
797 	}
798 
799 	if (md_ops != NULL) {
800 		kmem_free(md_ops, sizeof (md_ops_t *) * MD_NOPS);
801 		md_ops = NULL;
802 	}
803 
804 	if (MD_UPGRADE) {
805 		len = md_tuple_length * (2 * ((int)sizeof (dev32_t)));
806 		md_in_upgrade = 0;
807 		md_xlate_free(len);
808 		md_majortab_free();
809 	}
810 
811 	/*
812 	 * Undo what we did in mdattach, freeing resources
813 	 * and removing things we installed.  The system
814 	 * framework guarantees we are not active with this devinfo
815 	 * node in any other entry points at this time.
816 	 */
817 	ddi_prop_remove_all(dip);
818 	ddi_remove_minor_node(dip, NULL);
819 
820 	med_fini();
821 
822 	mod_hash_destroy_idhash(md_nblocksmap);
823 
824 	md_devinfo = NULL;
825 
826 	MD_CLR_IN(IN_DETACH);
827 	return (DDI_SUCCESS);
828 }
829 
830 
831 /*
832  * Given the device number return the devinfo pointer
833  * given to md via md_attach
834  */
835 /*ARGSUSED*/
836 static int
837 mdinfo(dev_info_t *dip, ddi_info_cmd_t infocmd, void *arg, void **result)
838 {
839 	int		error = DDI_FAILURE;
840 
841 	switch (infocmd) {
842 	case DDI_INFO_DEVT2DEVINFO:
843 		if (md_devinfo) {
844 			*result = (void *)md_devinfo;
845 			error = DDI_SUCCESS;
846 		}
847 		break;
848 
849 	case DDI_INFO_DEVT2INSTANCE:
850 		*result = (void *)0;
851 		error = DDI_SUCCESS;
852 		break;
853 	}
854 	return (error);
855 }
856 
857 /*
858  * property operation routine.  return the number of blocks for the partition
859  * in question or forward the request to the property facilities.
860  */
861 static int
862 mdprop_op(
863 	dev_t dev,		/* device number associated with device */
864 	dev_info_t *dip,	/* device info struct for this device */
865 	ddi_prop_op_t prop_op,	/* property operator */
866 	int mod_flags,		/* property flags */
867 	char *name,		/* name of property */
868 	caddr_t valuep,		/* where to put property value */
869 	int *lengthp)		/* put length of property here */
870 {
871 	return (ddi_prop_op_nblocks(dev, dip, prop_op, mod_flags,
872 	    name, valuep, lengthp, md_nblocks_get(getminor(dev))));
873 }
874 
875 static void
876 snarf_user_data(set_t setno)
877 {
878 	mddb_recid_t		recid;
879 	mddb_recstatus_t	status;
880 
881 	recid = mddb_makerecid(setno, 0);
882 	while ((recid = mddb_getnextrec(recid, MDDB_USER, 0)) > 0) {
883 		if (mddb_getrecprivate(recid) & MD_PRV_GOTIT)
884 			continue;
885 
886 		status = mddb_getrecstatus(recid);
887 		if (status == MDDB_STALE)
888 			continue;
889 
890 		if (status == MDDB_NODATA) {
891 			mddb_setrecprivate(recid, MD_PRV_PENDDEL);
892 			continue;
893 		}
894 
895 		ASSERT(status == MDDB_OK);
896 
897 		mddb_setrecprivate(recid, MD_PRV_GOTIT);
898 	}
899 }
900 
901 static void
902 md_print_block_usage(mddb_set_t *s, uint_t blks)
903 {
904 	uint_t		ib;
905 	int		li;
906 	mddb_mb_ic_t	*mbip;
907 	uint_t		max_blk_needed;
908 	mddb_lb_t	*lbp;
909 	mddb_sidelocator_t	*slp;
910 	int		drv_index;
911 	md_splitname	sn;
912 	char		*name;
913 	char		*suffix;
914 	size_t		prefixlen;
915 	size_t		suffixlen;
916 	int		alloc_sz;
917 
918 
919 	max_blk_needed = s->s_totalblkcnt - s->s_freeblkcnt + blks;
920 
921 	cmn_err(CE_WARN, "Blocks in Metadevice State Database: %d\n"
922 	    "            Additional Blocks Needed:            %d\n\n"
923 	    "            Increase size of following replicas for\n"
924 	    "            device relocatability by deleting listed\n"
925 	    "            replica and re-adding replica with\n"
926 	    "            increased size (see metadb(1M)):\n"
927 	    "                Replica                   Increase By",
928 	    s->s_totalblkcnt, (blks - s->s_freeblkcnt));
929 
930 	lbp = s->s_lbp;
931 
932 	for (li = 0; li < lbp->lb_loccnt; li++) {
933 		if (lbp->lb_locators[li].l_flags & MDDB_F_DELETED)
934 			continue;
935 		ib = 0;
936 		for (mbip = s->s_mbiarray[li]; mbip != NULL;
937 		    mbip = mbip->mbi_next) {
938 			ib += (uint_t)mbip->mbi_mddb_mb.mb_blkcnt;
939 		}
940 		if (ib == 0)
941 			continue;
942 		if (ib < max_blk_needed) {
943 			slp = &lbp->lb_sidelocators[s->s_sideno][li];
944 			drv_index = slp->l_drvnm_index;
945 			mddb_locatorblock2splitname(s->s_lnp, li, s->s_sideno,
946 			    &sn);
947 			prefixlen = SPN_PREFIX(&sn).pre_len;
948 			suffixlen = SPN_SUFFIX(&sn).suf_len;
949 			alloc_sz = (int)(prefixlen + suffixlen + 2);
950 			name = (char *)kmem_alloc(alloc_sz, KM_SLEEP);
951 			(void) strncpy(name, SPN_PREFIX(&sn).pre_data,
952 			    prefixlen);
953 			name[prefixlen] = '/';
954 			suffix = name + (prefixlen + 1);
955 			(void) strncpy(suffix, SPN_SUFFIX(&sn).suf_data,
956 			    suffixlen);
957 			name[prefixlen + suffixlen + 1] = '\0';
958 			cmn_err(CE_WARN,
959 			    "  %s (%s:%d:%d)   %d blocks",
960 			    name, lbp->lb_drvnm[drv_index].dn_data,
961 			    slp->l_mnum, lbp->lb_locators[li].l_blkno,
962 			    (max_blk_needed - ib));
963 			kmem_free(name, alloc_sz);
964 		}
965 	}
966 }
967 
968 /*
969  * md_create_minor_node:
970  *	Create the minor device for the given set and un_self_id.
971  *
972  * Input:
973  *	setno	- set number
974  *	mnum	- selfID of unit
975  *
976  * Output:
977  *	None.
978  *
979  * Returns 0 for success, 1 for failure.
980  *
981  * Side-effects:
982  *	None.
983  */
984 int
985 md_create_minor_node(set_t setno, minor_t mnum)
986 {
987 	char		name[20];
988 
989 	/* Check for valid arguments */
990 	if (setno >= MD_MAXSETS || MD_MIN2UNIT(mnum) >= MD_MAXUNITS)
991 		return (1);
992 
993 	(void) snprintf(name, 20, "%u,%u,blk",
994 	    (unsigned)setno, (unsigned)MD_MIN2UNIT(mnum));
995 
996 	if (ddi_create_minor_node(md_devinfo, name, S_IFBLK,
997 	    MD_MKMIN(setno, mnum), DDI_PSEUDO, 0))
998 		return (1);
999 
1000 	(void) snprintf(name, 20, "%u,%u,raw",
1001 	    (unsigned)setno, (unsigned)MD_MIN2UNIT(mnum));
1002 
1003 	if (ddi_create_minor_node(md_devinfo, name, S_IFCHR,
1004 	    MD_MKMIN(setno, mnum), DDI_PSEUDO, 0))
1005 		return (1);
1006 
1007 	return (0);
1008 }
1009 
1010 /*
1011  * For a given key check if it is an orphaned record.
1012  * The following conditions are used to determine an orphan.
1013  * 1. The device associated with that key is not a metadevice.
1014  * 2. If DEVID_STYLE then the physical device does not have a device Id
1015  * associated with it.
1016  *
1017  * If a key does not have an entry in the devid namespace it could be
1018  * a device that does not support device ids. Hence the record is not
1019  * deleted.
1020  */
1021 
1022 static int
1023 md_verify_orphaned_record(set_t setno, mdkey_t key)
1024 {
1025 	md_dev64_t	odev; /* orphaned dev */
1026 	mddb_set_t	*s;
1027 	side_t		side = 0;
1028 	struct nm_next_hdr	*did_nh = NULL;
1029 
1030 	s = (mddb_set_t *)md_set[setno].s_db;
1031 	if ((did_nh = get_first_record(setno, 1,  (NM_DEVID | NM_NOTSHARED)))
1032 	    == NULL)
1033 		return (0);
1034 	/*
1035 	 * If devid style is set then get the dev_t using MD_NOTRUST_DEVT
1036 	 */
1037 	if (s->s_lbp->lb_flags & MDDB_DEVID_STYLE) {
1038 		odev = md_getdevnum(setno, side, key, MD_NOTRUST_DEVT);
1039 		if ((odev == NODEV64) || (md_getmajor(odev) == md_major))
1040 			return (0);
1041 		if (lookup_entry(did_nh, setno, side, key, odev, NM_DEVID) ==
1042 		    NULL)
1043 			return (1);
1044 	}
1045 	return (0);
1046 }
1047 
1048 int
1049 md_snarf_db_set(set_t setno, md_error_t *ep)
1050 {
1051 	int			err = 0;
1052 	int			i;
1053 	mddb_recid_t		recid;
1054 	mddb_type_t		drvrid;
1055 	mddb_recstatus_t	status;
1056 	md_ops_t		*ops;
1057 	uint_t			privat;
1058 	mddb_set_t		*s;
1059 	uint_t			cvt_blks;
1060 	struct nm_next_hdr	*nh;
1061 	mdkey_t			key = MD_KEYWILD;
1062 	side_t			side = 0;
1063 	int			size;
1064 	int			devid_flag;
1065 	int			retval;
1066 	uint_t			un;
1067 	int			un_next_set = 0;
1068 
1069 	md_haltsnarf_enter(setno);
1070 
1071 	mutex_enter(&md_mx);
1072 	if (md_set[setno].s_status & MD_SET_SNARFED) {
1073 		mutex_exit(&md_mx);
1074 		md_haltsnarf_exit(setno);
1075 		return (0);
1076 	}
1077 	mutex_exit(&md_mx);
1078 
1079 	if (! (md_get_status() & MD_GBL_DAEMONS_LIVE)) {
1080 		if (md_start_daemons(TRUE)) {
1081 			if (ep != NULL)
1082 				(void) mdsyserror(ep, ENXIO);
1083 			err = -1;
1084 			goto out;
1085 		}
1086 	}
1087 
1088 
1089 	/*
1090 	 * Load the devid name space if it exists
1091 	 */
1092 	(void) md_load_namespace(setno, NULL, NM_DEVID);
1093 	if (!md_load_namespace(setno, ep, 0L)) {
1094 		/*
1095 		 * Unload the devid namespace
1096 		 */
1097 		(void) md_unload_namespace(setno, NM_DEVID);
1098 		err = -1;
1099 		goto out;
1100 	}
1101 
1102 	/*
1103 	 * If replica is in non-devid state, convert if:
1104 	 * 	- not in probe during upgrade (md_keep_repl_state = 0)
1105 	 * 	- enough space available in replica
1106 	 *	- local set
1107 	 *	- not a multi-node diskset
1108 	 *	- clustering is not present (for non-local set)
1109 	 */
1110 	s = (mddb_set_t *)md_set[setno].s_db;
1111 	devid_flag = 0;
1112 	if (!(s->s_lbp->lb_flags & MDDB_DEVID_STYLE) && !md_keep_repl_state)
1113 		devid_flag = 1;
1114 	if (cluster_bootflags & CLUSTER_CONFIGURED)
1115 		if (setno != MD_LOCAL_SET)
1116 			devid_flag = 0;
1117 	if (MD_MNSET_SETNO(setno))
1118 		devid_flag = 0;
1119 	if ((md_devid_destroy == 1) && (md_keep_repl_state == 1))
1120 		devid_flag = 0;
1121 
1122 	/*
1123 	 * if we weren't devid style before and md_keep_repl_state=1
1124 	 * we need to stay non-devid
1125 	 */
1126 	if ((md_keep_repl_state == 1) &&
1127 	    ((s->s_lbp->lb_flags & MDDB_DEVID_STYLE) == 0))
1128 		devid_flag = 0;
1129 	if (devid_flag) {
1130 		/*
1131 		 * Determine number of free blocks needed to convert
1132 		 * entire replica to device id format - locator blocks
1133 		 * and namespace.
1134 		 */
1135 		cvt_blks = 0;
1136 		if (mddb_lb_did_convert(s, 0, &cvt_blks) != 0) {
1137 			if (ep != NULL)
1138 				(void) mdsyserror(ep, EIO);
1139 			err = -1;
1140 			goto out;
1141 
1142 		}
1143 		cvt_blks += md_nm_did_chkspace(setno);
1144 
1145 		/* add MDDB_DEVID_CONV_PERC% */
1146 		if ((md_conv_perc > 0) && (md_conv_perc <= 100)) {
1147 			cvt_blks = cvt_blks * (100 + md_conv_perc) / 100;
1148 		}
1149 
1150 		if (cvt_blks <= s->s_freeblkcnt) {
1151 			if (mddb_lb_did_convert(s, 1, &cvt_blks) != 0) {
1152 				if (ep != NULL)
1153 					(void) mdsyserror(ep, EIO);
1154 				err = -1;
1155 				goto out;
1156 			}
1157 
1158 		} else {
1159 			/*
1160 			 * Print message that replica can't be converted for
1161 			 * lack of space.   No failure - just continue to
1162 			 * run without device ids.
1163 			 */
1164 			cmn_err(CE_WARN,
1165 			    "Unable to add Solaris Volume Manager device "
1166 			    "relocation data.\n"
1167 			    "          To use device relocation feature:\n"
1168 			    "          - Increase size of listed replicas\n"
1169 			    "          - Reboot");
1170 			md_print_block_usage(s, cvt_blks);
1171 			cmn_err(CE_WARN,
1172 			    "Loading set without device relocation data.\n"
1173 			    "          Solaris Volume Manager disk movement "
1174 			    "not tracked in local set.");
1175 		}
1176 	}
1177 
1178 	/*
1179 	 * go through and load any modules referenced in
1180 	 * data base
1181 	 */
1182 	recid = mddb_makerecid(setno, 0);
1183 	while ((recid = mddb_getnextrec(recid, MDDB_ALL, 0)) > 0) {
1184 		status = mddb_getrecstatus(recid);
1185 		if (status == MDDB_STALE) {
1186 			if (! (md_get_setstatus(setno) & MD_SET_STALE)) {
1187 				md_set_setstatus(setno, MD_SET_STALE);
1188 				cmn_err(CE_WARN,
1189 				    "md: state database is stale");
1190 			}
1191 		} else if (status == MDDB_NODATA) {
1192 			mddb_setrecprivate(recid, MD_PRV_PENDDEL);
1193 			continue;
1194 		}
1195 		drvrid = mddb_getrectype1(recid);
1196 		if (drvrid < MDDB_FIRST_MODID)
1197 			continue;
1198 		if (md_loadsubmod(setno, md_getshared_name(setno, drvrid),
1199 		    drvrid) < 0) {
1200 			cmn_err(CE_NOTE, "md: could not load misc/%s",
1201 			    md_getshared_name(setno, drvrid));
1202 		}
1203 	}
1204 
1205 	if (recid < 0)
1206 		goto out;
1207 
1208 	snarf_user_data(setno);
1209 
1210 	/*
1211 	 * Initialize the md_nm_snarfed array
1212 	 * this array is indexed by the key and
1213 	 * is set by md_getdevnum during the snarf time
1214 	 */
1215 	if ((nh = get_first_record(setno, 0, NM_NOTSHARED)) != NULL) {
1216 		size = (int)((((struct nm_rec_hdr *)nh->nmn_record)->
1217 		    r_next_key) * (sizeof (int)));
1218 		md_nm_snarfed = (int *)kmem_zalloc(size, KM_SLEEP);
1219 	}
1220 
1221 	/*
1222 	 * go through and snarf until nothing gets added
1223 	 */
1224 	do {
1225 		i = 0;
1226 		for (ops = md_opslist; ops != NULL; ops = ops->md_next) {
1227 			if (ops->md_snarf != NULL) {
1228 				retval = ops->md_snarf(MD_SNARF_DOIT, setno);
1229 				if (retval == -1) {
1230 					err = -1;
1231 					/* Don't know the failed unit */
1232 					(void) mdmderror(ep, MDE_RR_ALLOC_ERROR,
1233 					    0);
1234 					(void) md_halt_set(setno, MD_HALT_ALL);
1235 					(void) mddb_unload_set(setno);
1236 					md_haltsnarf_exit(setno);
1237 					return (err);
1238 				} else {
1239 					i += retval;
1240 				}
1241 			}
1242 		}
1243 	} while (i);
1244 
1245 	/*
1246 	 * Set the first available slot and availability
1247 	 */
1248 	md_set[setno].s_un_avail = 0;
1249 	for (un = 0; un < MD_MAXUNITS; un++) {
1250 		if (md_set[setno].s_un[un] != NULL) {
1251 			continue;
1252 		} else {
1253 			if (!un_next_set) {
1254 				md_set[setno].s_un_next = un;
1255 				un_next_set = 1;
1256 			}
1257 			md_set[setno].s_un_avail++;
1258 		}
1259 	}
1260 
1261 	md_set_setstatus(setno, MD_SET_SNARFED);
1262 
1263 	recid = mddb_makerecid(setno, 0);
1264 	while ((recid = mddb_getnextrec(recid, MDDB_ALL, 0)) > 0) {
1265 		privat = mddb_getrecprivate(recid);
1266 		if (privat & MD_PRV_COMMIT) {
1267 			if (mddb_commitrec(recid)) {
1268 				if (!(md_get_setstatus(setno) & MD_SET_STALE)) {
1269 					md_set_setstatus(setno, MD_SET_STALE);
1270 					cmn_err(CE_WARN,
1271 					    "md: state database is stale");
1272 				}
1273 			}
1274 			mddb_setrecprivate(recid, MD_PRV_GOTIT);
1275 		}
1276 	}
1277 
1278 	/* Deletes must happen after all the commits */
1279 	recid = mddb_makerecid(setno, 0);
1280 	while ((recid = mddb_getnextrec(recid, MDDB_ALL, 0)) > 0) {
1281 		privat = mddb_getrecprivate(recid);
1282 		if (privat & MD_PRV_DELETE) {
1283 			if (mddb_deleterec(recid)) {
1284 				if (!(md_get_setstatus(setno) & MD_SET_STALE)) {
1285 					md_set_setstatus(setno, MD_SET_STALE);
1286 					cmn_err(CE_WARN,
1287 					    "md: state database is stale");
1288 				}
1289 				mddb_setrecprivate(recid, MD_PRV_GOTIT);
1290 			}
1291 			recid = mddb_makerecid(setno, 0);
1292 		}
1293 	}
1294 
1295 	/*
1296 	 * go through and clean up records until nothing gets cleaned up.
1297 	 */
1298 	do {
1299 		i = 0;
1300 		for (ops = md_opslist; ops != NULL; ops = ops->md_next)
1301 			if (ops->md_snarf != NULL)
1302 				i += ops->md_snarf(MD_SNARF_CLEANUP, setno);
1303 	} while (i);
1304 
1305 	if (md_nm_snarfed != NULL &&
1306 	    !(md_get_setstatus(setno) & MD_SET_STALE)) {
1307 		/*
1308 		 * go thru and cleanup the namespace and the device id
1309 		 * name space
1310 		 */
1311 		for (key = 1;
1312 		    key < ((struct nm_rec_hdr *)nh->nmn_record)->r_next_key;
1313 		    key++) {
1314 			/*
1315 			 * Is the entry an 'orphan'?
1316 			 */
1317 			if (lookup_entry(nh, setno, side, key, NODEV64, 0L) !=
1318 			    NULL) {
1319 				/*
1320 				 * If the value is not set then apparently
1321 				 * it is not part of the current configuration,
1322 				 * remove it this can happen when system panic
1323 				 * between the primary name space update and
1324 				 * the device id name space update
1325 				 */
1326 				if (md_nm_snarfed[key] == 0) {
1327 					if (md_verify_orphaned_record(setno,
1328 					    key) == 1)
1329 						(void) remove_entry(nh,
1330 						    side, key, 0L);
1331 				}
1332 			}
1333 		}
1334 	}
1335 
1336 	if (md_nm_snarfed != NULL) {
1337 		/*
1338 		 * Done and free the memory
1339 		 */
1340 		kmem_free(md_nm_snarfed, size);
1341 		md_nm_snarfed = NULL;
1342 	}
1343 
1344 	if (s->s_lbp->lb_flags & MDDB_DEVID_STYLE &&
1345 	    !(md_get_setstatus(setno) & MD_SET_STALE)) {
1346 		/*
1347 		 * if the destroy flag has been set and
1348 		 * the MD_SET_DIDCLUP bit is not set in
1349 		 * the set's status field, cleanup the
1350 		 * entire device id namespace
1351 		 */
1352 		if (md_devid_destroy &&
1353 		    !(md_get_setstatus(setno) & MD_SET_DIDCLUP)) {
1354 			(void) md_devid_cleanup(setno, 1);
1355 			md_set_setstatus(setno, MD_SET_DIDCLUP);
1356 		} else
1357 			(void) md_devid_cleanup(setno, 0);
1358 	}
1359 
1360 	/*
1361 	 * clear single threading on snarf, return success or error
1362 	 */
1363 out:
1364 	md_haltsnarf_exit(setno);
1365 	return (err);
1366 }
1367 
1368 void
1369 get_minfo(struct dk_minfo *info, minor_t mnum)
1370 {
1371 	md_unit_t	*un;
1372 	mdi_unit_t	*ui;
1373 
1374 	info->dki_capacity = 0;
1375 	info->dki_lbsize = 0;
1376 	info->dki_media_type = 0;
1377 
1378 	if ((ui = MDI_UNIT(mnum)) == NULL) {
1379 		return;
1380 	}
1381 	un = (md_unit_t *)md_unit_readerlock(ui);
1382 	info->dki_capacity = un->c.un_total_blocks;
1383 	md_unit_readerexit(ui);
1384 	info->dki_lbsize = DEV_BSIZE;
1385 	info->dki_media_type = DK_UNKNOWN;
1386 }
1387 
1388 
1389 void
1390 get_info(struct dk_cinfo *info, minor_t mnum)
1391 {
1392 	/*
1393 	 * Controller Information
1394 	 */
1395 	info->dki_ctype = DKC_MD;
1396 	info->dki_cnum = ddi_get_instance(ddi_get_parent(md_devinfo));
1397 	(void) strcpy(info->dki_cname,
1398 	    ddi_get_name(ddi_get_parent(md_devinfo)));
1399 	/*
1400 	 * Unit Information
1401 	 */
1402 	info->dki_unit = mnum;
1403 	info->dki_slave = 0;
1404 	(void) strcpy(info->dki_dname, ddi_driver_name(md_devinfo));
1405 	info->dki_flags = 0;
1406 	info->dki_partition = 0;
1407 	info->dki_maxtransfer = (ushort_t)(md_maxphys / DEV_BSIZE);
1408 
1409 	/*
1410 	 * We can't get from here to there yet
1411 	 */
1412 	info->dki_addr = 0;
1413 	info->dki_space = 0;
1414 	info->dki_prio = 0;
1415 	info->dki_vec = 0;
1416 }
1417 
1418 /*
1419  * open admin device
1420  */
1421 static int
1422 mdadminopen(
1423 	int	flag,
1424 	int	otyp)
1425 {
1426 	int	err = 0;
1427 
1428 	/* single thread */
1429 	mutex_enter(&md_mx);
1430 
1431 	/* check type and flags */
1432 	if ((otyp != OTYP_CHR) && (otyp != OTYP_LYR)) {
1433 		err = EINVAL;
1434 		goto out;
1435 	}
1436 	if (((flag & FEXCL) && (md_status & MD_GBL_OPEN)) ||
1437 	    (md_status & MD_GBL_EXCL)) {
1438 		err = EBUSY;
1439 		goto out;
1440 	}
1441 
1442 	/* count and flag open */
1443 	md_ocnt[otyp]++;
1444 	md_status |= MD_GBL_OPEN;
1445 	if (flag & FEXCL)
1446 		md_status |= MD_GBL_EXCL;
1447 
1448 	/* unlock return success */
1449 out:
1450 	mutex_exit(&md_mx);
1451 	return (err);
1452 }
1453 
1454 /*
1455  * open entry point
1456  */
1457 static int
1458 mdopen(
1459 	dev_t		*dev,
1460 	int		flag,
1461 	int		otyp,
1462 	cred_t		*cred_p)
1463 {
1464 	minor_t		mnum = getminor(*dev);
1465 	unit_t		unit = MD_MIN2UNIT(mnum);
1466 	set_t		setno = MD_MIN2SET(mnum);
1467 	mdi_unit_t	*ui = NULL;
1468 	int		err = 0;
1469 	md_parent_t	parent;
1470 
1471 	/* dispatch admin device opens */
1472 	if (mnum == MD_ADM_MINOR)
1473 		return (mdadminopen(flag, otyp));
1474 
1475 	/* lock, check status */
1476 	rw_enter(&md_unit_array_rw.lock, RW_READER);
1477 
1478 tryagain:
1479 	if (md_get_status() & MD_GBL_HALTED)  {
1480 		err = ENODEV;
1481 		goto out;
1482 	}
1483 
1484 	/* check minor */
1485 	if ((setno >= md_nsets) || (unit >= md_nunits)) {
1486 		err = ENXIO;
1487 		goto out;
1488 	}
1489 
1490 	/* make sure we're snarfed */
1491 	if ((md_get_setstatus(MD_LOCAL_SET) & MD_SET_SNARFED) == 0) {
1492 		if (md_snarf_db_set(MD_LOCAL_SET, NULL) != 0) {
1493 			err = ENODEV;
1494 			goto out;
1495 		}
1496 	}
1497 	if ((md_get_setstatus(setno) & MD_SET_SNARFED) == 0) {
1498 		err = ENODEV;
1499 		goto out;
1500 	}
1501 
1502 	/* check unit */
1503 	if ((ui = MDI_UNIT(mnum)) == NULL) {
1504 		err = ENXIO;
1505 		goto out;
1506 	}
1507 
1508 	/*
1509 	 * The softpart open routine may do an I/O during the open, in
1510 	 * which case the open routine will set the OPENINPROGRESS flag
1511 	 * and drop all locks during the I/O.  If this thread sees
1512 	 * the OPENINPROGRESS flag set, if should wait until the flag
1513 	 * is reset before calling the driver's open routine.  It must
1514 	 * also revalidate the world after it grabs the unit_array lock
1515 	 * since the set may have been released or the metadevice cleared
1516 	 * during the sleep.
1517 	 */
1518 	if (MD_MNSET_SETNO(setno)) {
1519 		mutex_enter(&ui->ui_mx);
1520 		if (ui->ui_lock & MD_UL_OPENINPROGRESS) {
1521 			rw_exit(&md_unit_array_rw.lock);
1522 			cv_wait(&ui->ui_cv, &ui->ui_mx);
1523 			rw_enter(&md_unit_array_rw.lock, RW_READER);
1524 			mutex_exit(&ui->ui_mx);
1525 			goto tryagain;
1526 		}
1527 		mutex_exit(&ui->ui_mx);
1528 	}
1529 
1530 	/* Test if device is openable */
1531 	if ((ui->ui_tstate & MD_NOTOPENABLE) != 0) {
1532 		err = ENXIO;
1533 		goto out;
1534 	}
1535 
1536 	/* don't allow opens w/WRITE flag if stale */
1537 	if ((flag & FWRITE) && (md_get_setstatus(setno) & MD_SET_STALE)) {
1538 		err = EROFS;
1539 		goto out;
1540 	}
1541 
1542 	/* don't allow writes to subdevices */
1543 	parent = md_get_parent(md_expldev(*dev));
1544 	if ((flag & FWRITE) && MD_HAS_PARENT(parent)) {
1545 		err = EROFS;
1546 		goto out;
1547 	}
1548 
1549 	/* open underlying driver */
1550 	if (md_ops[ui->ui_opsindex]->md_open != NULL) {
1551 		if ((err = (*md_ops[ui->ui_opsindex]->md_open)
1552 		    (dev, flag, otyp, cred_p, 0)) != 0)
1553 			goto out;
1554 	}
1555 
1556 	/* or do it ourselves */
1557 	else {
1558 		/* single thread */
1559 		(void) md_unit_openclose_enter(ui);
1560 		err = md_unit_incopen(mnum, flag, otyp);
1561 		md_unit_openclose_exit(ui);
1562 		if (err != 0)
1563 			goto out;
1564 	}
1565 
1566 	/* unlock, return status */
1567 out:
1568 	rw_exit(&md_unit_array_rw.lock);
1569 	return (err);
1570 }
1571 
1572 /*
1573  * close admin device
1574  */
1575 static int
1576 mdadminclose(
1577 	int	otyp)
1578 {
1579 	int	i;
1580 	int	err = 0;
1581 
1582 	/* single thread */
1583 	mutex_enter(&md_mx);
1584 
1585 	/* check type and flags */
1586 	if ((otyp < 0) || (otyp >= OTYPCNT)) {
1587 		err = EINVAL;
1588 		goto out;
1589 	} else if (md_ocnt[otyp] == 0) {
1590 		err = ENXIO;
1591 		goto out;
1592 	}
1593 
1594 	/* count and flag closed */
1595 	if (otyp == OTYP_LYR)
1596 		md_ocnt[otyp]--;
1597 	else
1598 		md_ocnt[otyp] = 0;
1599 	md_status &= ~MD_GBL_OPEN;
1600 	for (i = 0; (i < OTYPCNT); ++i)
1601 		if (md_ocnt[i] != 0)
1602 			md_status |= MD_GBL_OPEN;
1603 	if (! (md_status & MD_GBL_OPEN))
1604 		md_status &= ~MD_GBL_EXCL;
1605 
1606 	/* unlock return success */
1607 out:
1608 	mutex_exit(&md_mx);
1609 	return (err);
1610 }
1611 
1612 /*
1613  * close entry point
1614  */
1615 static int
1616 mdclose(
1617 	dev_t		dev,
1618 	int		flag,
1619 	int		otyp,
1620 	cred_t		*cred_p)
1621 {
1622 	minor_t		mnum = getminor(dev);
1623 	set_t		setno = MD_MIN2SET(mnum);
1624 	unit_t		unit = MD_MIN2UNIT(mnum);
1625 	mdi_unit_t	*ui = NULL;
1626 	int		err = 0;
1627 
1628 	/* dispatch admin device closes */
1629 	if (mnum == MD_ADM_MINOR)
1630 		return (mdadminclose(otyp));
1631 
1632 	/* check minor */
1633 	if ((setno >= md_nsets) || (unit >= md_nunits) ||
1634 	    ((ui = MDI_UNIT(mnum)) == NULL)) {
1635 		err = ENXIO;
1636 		goto out;
1637 	}
1638 
1639 	/* close underlying driver */
1640 	if (md_ops[ui->ui_opsindex]->md_close != NULL) {
1641 		if ((err = (*md_ops[ui->ui_opsindex]->md_close)
1642 		    (dev, flag, otyp, cred_p, 0)) != 0)
1643 			goto out;
1644 	}
1645 
1646 	/* or do it ourselves */
1647 	else {
1648 		/* single thread */
1649 		(void) md_unit_openclose_enter(ui);
1650 		err = md_unit_decopen(mnum, otyp);
1651 		md_unit_openclose_exit(ui);
1652 		if (err != 0)
1653 			goto out;
1654 	}
1655 
1656 	/* return success */
1657 out:
1658 	return (err);
1659 }
1660 
1661 
1662 /*
1663  * This routine performs raw read operations.  It is called from the
1664  * device switch at normal priority.
1665  *
1666  * The main catch is that the *uio struct which is passed to us may
1667  * specify a read which spans two buffers, which would be contiguous
1668  * on a single partition,  but not on a striped partition. This will
1669  * be handled by mdstrategy.
1670  */
1671 /*ARGSUSED*/
1672 static int
1673 mdread(dev_t dev, struct uio *uio, cred_t *credp)
1674 {
1675 	minor_t		mnum;
1676 	mdi_unit_t	*ui;
1677 	int		error;
1678 
1679 	if (((mnum = getminor(dev)) == MD_ADM_MINOR) ||
1680 	    (MD_MIN2SET(mnum) >= md_nsets) ||
1681 	    (MD_MIN2UNIT(mnum) >= md_nunits) ||
1682 	    ((ui = MDI_UNIT(mnum)) == NULL))
1683 		return (ENXIO);
1684 
1685 	if (md_ops[ui->ui_opsindex]->md_read  != NULL)
1686 		return ((*md_ops[ui->ui_opsindex]->md_read)
1687 		    (dev, uio, credp));
1688 
1689 	if ((error = md_chk_uio(uio)) != 0)
1690 		return (error);
1691 
1692 	return (physio(mdstrategy, NULL, dev, B_READ, md_minphys, uio));
1693 }
1694 
1695 /*
1696  * This routine performs async raw read operations.  It is called from the
1697  * device switch at normal priority.
1698  *
1699  * The main catch is that the *aio struct which is passed to us may
1700  * specify a read which spans two buffers, which would be contiguous
1701  * on a single partition,  but not on a striped partition. This will
1702  * be handled by mdstrategy.
1703  */
1704 /*ARGSUSED*/
1705 static int
1706 mdaread(dev_t dev, struct aio_req *aio, cred_t *credp)
1707 {
1708 	minor_t		mnum;
1709 	mdi_unit_t	*ui;
1710 	int		error;
1711 
1712 
1713 	if (((mnum = getminor(dev)) == MD_ADM_MINOR) ||
1714 	    (MD_MIN2SET(mnum) >= md_nsets) ||
1715 	    (MD_MIN2UNIT(mnum) >= md_nunits) ||
1716 	    ((ui = MDI_UNIT(mnum)) == NULL))
1717 		return (ENXIO);
1718 
1719 	if (md_ops[ui->ui_opsindex]->md_aread  != NULL)
1720 		return ((*md_ops[ui->ui_opsindex]->md_aread)
1721 		    (dev, aio, credp));
1722 
1723 	if ((error = md_chk_uio(aio->aio_uio)) != 0)
1724 		return (error);
1725 
1726 	return (aphysio(mdstrategy, anocancel, dev, B_READ, md_minphys, aio));
1727 }
1728 
1729 /*
1730  * This routine performs raw write operations.	It is called from the
1731  * device switch at normal priority.
1732  *
1733  * The main catch is that the *uio struct which is passed to us may
1734  * specify a write which spans two buffers, which would be contiguous
1735  * on a single partition,  but not on a striped partition. This is
1736  * handled by mdstrategy.
1737  *
1738  */
1739 /*ARGSUSED*/
1740 static int
1741 mdwrite(dev_t dev, struct uio *uio, cred_t *credp)
1742 {
1743 	minor_t		mnum;
1744 	mdi_unit_t	*ui;
1745 	int		error;
1746 
1747 	if (((mnum = getminor(dev)) == MD_ADM_MINOR) ||
1748 	    (MD_MIN2SET(mnum) >= md_nsets) ||
1749 	    (MD_MIN2UNIT(mnum) >= md_nunits) ||
1750 	    ((ui = MDI_UNIT(mnum)) == NULL))
1751 		return (ENXIO);
1752 
1753 	if (md_ops[ui->ui_opsindex]->md_write  != NULL)
1754 		return ((*md_ops[ui->ui_opsindex]->md_write)
1755 		    (dev, uio, credp));
1756 
1757 	if ((error = md_chk_uio(uio)) != 0)
1758 		return (error);
1759 
1760 	return (physio(mdstrategy, NULL, dev, B_WRITE, md_minphys, uio));
1761 }
1762 
1763 /*
1764  * This routine performs async raw write operations.  It is called from the
1765  * device switch at normal priority.
1766  *
1767  * The main catch is that the *aio struct which is passed to us may
1768  * specify a write which spans two buffers, which would be contiguous
1769  * on a single partition,  but not on a striped partition. This is
1770  * handled by mdstrategy.
1771  *
1772  */
1773 /*ARGSUSED*/
1774 static int
1775 mdawrite(dev_t dev, struct aio_req *aio, cred_t *credp)
1776 {
1777 	minor_t		mnum;
1778 	mdi_unit_t	*ui;
1779 	int		error;
1780 
1781 
1782 	if (((mnum = getminor(dev)) == MD_ADM_MINOR) ||
1783 	    (MD_MIN2SET(mnum) >= md_nsets) ||
1784 	    (MD_MIN2UNIT(mnum) >= md_nunits) ||
1785 	    ((ui = MDI_UNIT(mnum)) == NULL))
1786 		return (ENXIO);
1787 
1788 	if (md_ops[ui->ui_opsindex]->md_awrite  != NULL)
1789 		return ((*md_ops[ui->ui_opsindex]->md_awrite)
1790 		    (dev, aio, credp));
1791 
1792 	if ((error = md_chk_uio(aio->aio_uio)) != 0)
1793 		return (error);
1794 
1795 	return (aphysio(mdstrategy, anocancel, dev, B_WRITE, md_minphys, aio));
1796 }
1797 
1798 int
1799 mdstrategy(struct buf *bp)
1800 {
1801 	minor_t		mnum;
1802 	mdi_unit_t	*ui;
1803 
1804 	ASSERT((bp->b_flags & B_DONE) == 0);
1805 
1806 	if (panicstr)
1807 		md_clr_status(MD_GBL_DAEMONS_LIVE);
1808 
1809 	if (((mnum = getminor(bp->b_edev)) == MD_ADM_MINOR) ||
1810 	    (MD_MIN2SET(mnum) >= md_nsets) ||
1811 	    (MD_MIN2UNIT(mnum) >= md_nunits) ||
1812 	    ((ui = MDI_UNIT(mnum)) == NULL)) {
1813 		bp->b_flags |= B_ERROR;
1814 		bp->b_error = ENXIO;
1815 		bp->b_resid = bp->b_bcount;
1816 		biodone(bp);
1817 		return (0);
1818 	}
1819 
1820 	bp->b_flags &= ~(B_ERROR | B_DONE);
1821 	if (md_ops[ui->ui_opsindex]->md_strategy  != NULL) {
1822 		(*md_ops[ui->ui_opsindex]->md_strategy) (bp, 0, NULL);
1823 	} else {
1824 		(void) errdone(ui, bp, ENXIO);
1825 	}
1826 	return (0);
1827 }
1828 
1829 /*
1830  * Return true if the ioctl is allowed to be multithreaded.
1831  * All the ioctls with MN are sent only from the message handlers through
1832  * rpc.mdcommd, which (via it's own locking mechanism) takes care that not two
1833  * ioctl for the same metadevice are issued at the same time.
1834  * So we are safe here.
1835  * The other ioctls do not mess with any metadevice structures and therefor
1836  * are harmless too, if called multiple times at the same time.
1837  */
1838 static boolean_t
1839 is_mt_ioctl(int cmd) {
1840 
1841 	switch (cmd) {
1842 	case MD_IOCGUNIQMSGID:
1843 	case MD_IOCGVERSION:
1844 	case MD_IOCISOPEN:
1845 	case MD_MN_SET_MM_OWNER:
1846 	case MD_MN_SET_STATE:
1847 	case MD_MN_SUSPEND_WRITES:
1848 	case MD_MN_ALLOCATE_HOTSPARE:
1849 	case MD_MN_SET_SETFLAGS:
1850 	case MD_MN_GET_SETFLAGS:
1851 	case MD_MN_MDDB_OPTRECFIX:
1852 	case MD_MN_MDDB_PARSE:
1853 	case MD_MN_MDDB_BLOCK:
1854 	case MD_MN_DB_USERREQ:
1855 	case MD_IOC_SPSTATUS:
1856 	case MD_MN_COMMD_ERR:
1857 	case MD_MN_SET_COMMD_RUNNING:
1858 	case MD_MN_RESYNC:
1859 	case MD_MN_SETSYNC:
1860 	case MD_MN_POKE_HOTSPARES:
1861 	case MD_MN_RR_DIRTY:
1862 	case MD_MN_RR_CLEAN:
1863 	case MD_MN_IOC_SPUPDATEWM:
1864 		return (1);
1865 	default:
1866 		return (0);
1867 	}
1868 }
1869 
1870 /*
1871  * This routine implements the ioctl calls for the Virtual Disk System.
1872  * It is called from the device switch at normal priority.
1873  */
1874 /* ARGSUSED */
1875 static int
1876 mdioctl(dev_t dev, int cmd, intptr_t data, int mode, cred_t *cred_p,
1877 	int *rval_p)
1878 {
1879 	minor_t		mnum = getminor(dev);
1880 	mdi_unit_t	*ui;
1881 	IOLOCK		lock;
1882 	int		err;
1883 
1884 	/*
1885 	 * For multinode disksets  number of ioctls are allowed to be
1886 	 * multithreaded.
1887 	 * A fundamental assumption made in this implementation is that
1888 	 * ioctls either do not interact with other md structures  or the
1889 	 * ioctl to the admin device can only occur if the metadevice
1890 	 * device is open. i.e. avoid a race between metaclear and the
1891 	 * progress of a multithreaded ioctl.
1892 	 */
1893 
1894 	if (!is_mt_ioctl(cmd) && md_ioctl_lock_enter() == EINTR) {
1895 		return (EINTR);
1896 	}
1897 
1898 	/*
1899 	 * initialize lock tracker
1900 	 */
1901 	IOLOCK_INIT(&lock);
1902 
1903 	/* Flag to indicate that MD_GBL_IOCTL_LOCK is not acquired */
1904 
1905 	if (is_mt_ioctl(cmd)) {
1906 		/* increment the md_mtioctl_cnt */
1907 		mutex_enter(&md_mx);
1908 		md_mtioctl_cnt++;
1909 		mutex_exit(&md_mx);
1910 		lock.l_flags |= MD_MT_IOCTL;
1911 	}
1912 
1913 	/*
1914 	 * this has been added to prevent notification from re-snarfing
1915 	 * so metaunload will work.  It may interfere with other modules
1916 	 * halt process.
1917 	 */
1918 	if (md_get_status() & (MD_GBL_HALTED | MD_GBL_DAEMONS_DIE))
1919 		return (IOLOCK_RETURN(ENXIO, &lock));
1920 
1921 	/*
1922 	 * admin device ioctls
1923 	 */
1924 	if (mnum == MD_ADM_MINOR) {
1925 		err = md_admin_ioctl(md_expldev(dev), cmd, (void *) data,
1926 		    mode, &lock);
1927 	}
1928 
1929 	/*
1930 	 * metadevice ioctls
1931 	 */
1932 	else if ((MD_MIN2SET(mnum) >= md_nsets) ||
1933 	    (MD_MIN2UNIT(mnum) >= md_nunits) ||
1934 	    (md_set[MD_MIN2SET(mnum)].s_ui == NULL) ||
1935 	    ((ui = MDI_UNIT(mnum)) == NULL)) {
1936 		err = ENXIO;
1937 	} else if (md_ops[ui->ui_opsindex]->md_ioctl == NULL) {
1938 		err = ENOTTY;
1939 	} else {
1940 		err = (*md_ops[ui->ui_opsindex]->md_ioctl)
1941 		    (dev, cmd, (void *) data, mode, &lock);
1942 	}
1943 
1944 	/*
1945 	 * drop any locks we grabbed
1946 	 */
1947 	return (IOLOCK_RETURN_IOCTLEND(err, &lock));
1948 }
1949 
1950 static int
1951 mddump(dev_t dev, caddr_t addr, daddr_t blkno, int nblk)
1952 {
1953 	minor_t		mnum;
1954 	set_t		setno;
1955 	mdi_unit_t	*ui;
1956 
1957 	if ((mnum = getminor(dev)) == MD_ADM_MINOR)
1958 		return (ENXIO);
1959 
1960 	setno = MD_MIN2SET(mnum);
1961 
1962 	if ((setno >= md_nsets) || (MD_MIN2UNIT(mnum) >= md_nunits) ||
1963 	    ((ui = MDI_UNIT(mnum)) == NULL))
1964 		return (ENXIO);
1965 
1966 
1967 	if ((md_get_setstatus(setno) & MD_SET_SNARFED) == 0)
1968 		return (ENXIO);
1969 
1970 	if (md_ops[ui->ui_opsindex]->md_dump  != NULL)
1971 		return ((*md_ops[ui->ui_opsindex]->md_dump)
1972 		    (dev, addr, blkno, nblk));
1973 
1974 	return (ENXIO);
1975 }
1976 
1977 /*
1978  * Metadevice unit number dispatcher
1979  * When this routine is called it will scan the
1980  * incore unit array and return the avail slot
1981  * hence the unit number to the caller
1982  *
1983  * Return -1 if there is nothing available
1984  */
1985 unit_t
1986 md_get_nextunit(set_t setno)
1987 {
1988 	unit_t	un, start;
1989 
1990 	/*
1991 	 * If nothing available
1992 	 */
1993 	if (md_set[setno].s_un_avail == 0) {
1994 		return (MD_UNITBAD);
1995 	}
1996 
1997 	mutex_enter(&md_mx);
1998 	start = un = md_set[setno].s_un_next;
1999 
2000 	/* LINTED: E_CONSTANT_CONDITION */
2001 	while (1) {
2002 		if (md_set[setno].s_un[un] == NULL) {
2003 			/*
2004 			 * Advance the starting index for the next
2005 			 * md_get_nextunit call
2006 			 */
2007 			if (un == MD_MAXUNITS - 1) {
2008 				md_set[setno].s_un_next = 0;
2009 			} else {
2010 				md_set[setno].s_un_next = un + 1;
2011 			}
2012 			break;
2013 		}
2014 
2015 		un = ((un == MD_MAXUNITS - 1) ? 0 : un + 1);
2016 
2017 		if (un == start) {
2018 			un = MD_UNITBAD;
2019 			break;
2020 		}
2021 
2022 	}
2023 
2024 	mutex_exit(&md_mx);
2025 	return (un);
2026 }
2027