xref: /freebsd/sys/geom/raid/g_raid.c (revision 61ba55bcf70f2340f9c943c9571113b3fd8eda69)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause
3  *
4  * Copyright (c) 2010 Alexander Motin <mav@FreeBSD.org>
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  *
16  * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND
17  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE
20  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26  * SUCH DAMAGE.
27  */
28 
29 #include <sys/cdefs.h>
30 #include <sys/param.h>
31 #include <sys/systm.h>
32 #include <sys/kernel.h>
33 #include <sys/module.h>
34 #include <sys/limits.h>
35 #include <sys/lock.h>
36 #include <sys/mutex.h>
37 #include <sys/bio.h>
38 #include <sys/sbuf.h>
39 #include <sys/sysctl.h>
40 #include <sys/malloc.h>
41 #include <sys/eventhandler.h>
42 #include <vm/uma.h>
43 #include <geom/geom.h>
44 #include <geom/geom_dbg.h>
45 #include <sys/proc.h>
46 #include <sys/kthread.h>
47 #include <sys/sched.h>
48 #include <geom/raid/g_raid.h>
49 #include "g_raid_md_if.h"
50 #include "g_raid_tr_if.h"
51 
52 static MALLOC_DEFINE(M_RAID, "raid_data", "GEOM_RAID Data");
53 
54 SYSCTL_DECL(_kern_geom);
55 SYSCTL_NODE(_kern_geom, OID_AUTO, raid, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
56     "GEOM_RAID stuff");
57 int g_raid_enable = 1;
58 SYSCTL_INT(_kern_geom_raid, OID_AUTO, enable, CTLFLAG_RWTUN,
59     &g_raid_enable, 0, "Enable on-disk metadata taste");
60 u_int g_raid_aggressive_spare = 0;
61 SYSCTL_UINT(_kern_geom_raid, OID_AUTO, aggressive_spare, CTLFLAG_RWTUN,
62     &g_raid_aggressive_spare, 0, "Use disks without metadata as spare");
63 u_int g_raid_debug = 0;
64 SYSCTL_UINT(_kern_geom_raid, OID_AUTO, debug, CTLFLAG_RWTUN, &g_raid_debug, 0,
65     "Debug level");
66 int g_raid_read_err_thresh = 10;
67 SYSCTL_UINT(_kern_geom_raid, OID_AUTO, read_err_thresh, CTLFLAG_RWTUN,
68     &g_raid_read_err_thresh, 0,
69     "Number of read errors equated to disk failure");
70 u_int g_raid_start_timeout = 30;
71 SYSCTL_UINT(_kern_geom_raid, OID_AUTO, start_timeout, CTLFLAG_RWTUN,
72     &g_raid_start_timeout, 0,
73     "Time to wait for all array components");
74 static u_int g_raid_clean_time = 5;
75 SYSCTL_UINT(_kern_geom_raid, OID_AUTO, clean_time, CTLFLAG_RWTUN,
76     &g_raid_clean_time, 0, "Mark volume as clean when idling");
77 static u_int g_raid_disconnect_on_failure = 1;
78 SYSCTL_UINT(_kern_geom_raid, OID_AUTO, disconnect_on_failure, CTLFLAG_RWTUN,
79     &g_raid_disconnect_on_failure, 0, "Disconnect component on I/O failure.");
80 static u_int g_raid_name_format = 0;
81 SYSCTL_UINT(_kern_geom_raid, OID_AUTO, name_format, CTLFLAG_RWTUN,
82     &g_raid_name_format, 0, "Providers name format.");
83 static u_int g_raid_idle_threshold = 1000000;
84 SYSCTL_UINT(_kern_geom_raid, OID_AUTO, idle_threshold, CTLFLAG_RWTUN,
85     &g_raid_idle_threshold, 1000000,
86     "Time in microseconds to consider a volume idle.");
87 
88 #define	MSLEEP(rv, ident, mtx, priority, wmesg, timeout)	do {	\
89 	G_RAID_DEBUG(4, "%s: Sleeping %p.", __func__, (ident));		\
90 	rv = msleep((ident), (mtx), (priority), (wmesg), (timeout));	\
91 	G_RAID_DEBUG(4, "%s: Woken up %p.", __func__, (ident));		\
92 } while (0)
93 
94 LIST_HEAD(, g_raid_md_class) g_raid_md_classes =
95     LIST_HEAD_INITIALIZER(g_raid_md_classes);
96 
97 LIST_HEAD(, g_raid_tr_class) g_raid_tr_classes =
98     LIST_HEAD_INITIALIZER(g_raid_tr_classes);
99 
100 LIST_HEAD(, g_raid_volume) g_raid_volumes =
101     LIST_HEAD_INITIALIZER(g_raid_volumes);
102 
103 static eventhandler_tag g_raid_post_sync = NULL;
104 static int g_raid_started = 0;
105 static int g_raid_shutdown = 0;
106 
107 static int g_raid_destroy_geom(struct gctl_req *req, struct g_class *mp,
108     struct g_geom *gp);
109 static g_taste_t g_raid_taste;
110 static void g_raid_init(struct g_class *mp);
111 static void g_raid_fini(struct g_class *mp);
112 
113 struct g_class g_raid_class = {
114 	.name = G_RAID_CLASS_NAME,
115 	.version = G_VERSION,
116 	.ctlreq = g_raid_ctl,
117 	.taste = g_raid_taste,
118 	.destroy_geom = g_raid_destroy_geom,
119 	.init = g_raid_init,
120 	.fini = g_raid_fini
121 };
122 
123 static void g_raid_destroy_provider(struct g_raid_volume *vol);
124 static int g_raid_update_disk(struct g_raid_disk *disk, u_int event);
125 static int g_raid_update_subdisk(struct g_raid_subdisk *subdisk, u_int event);
126 static int g_raid_update_volume(struct g_raid_volume *vol, u_int event);
127 static int g_raid_update_node(struct g_raid_softc *sc, u_int event);
128 static void g_raid_dumpconf(struct sbuf *sb, const char *indent,
129     struct g_geom *gp, struct g_consumer *cp, struct g_provider *pp);
130 static void g_raid_start(struct bio *bp);
131 static void g_raid_start_request(struct bio *bp);
132 static void g_raid_disk_done(struct bio *bp);
133 static void g_raid_poll(struct g_raid_softc *sc);
134 
135 static const char *
136 g_raid_node_event2str(int event)
137 {
138 
139 	switch (event) {
140 	case G_RAID_NODE_E_WAKE:
141 		return ("WAKE");
142 	case G_RAID_NODE_E_START:
143 		return ("START");
144 	default:
145 		return ("INVALID");
146 	}
147 }
148 
149 const char *
150 g_raid_disk_state2str(int state)
151 {
152 
153 	switch (state) {
154 	case G_RAID_DISK_S_NONE:
155 		return ("NONE");
156 	case G_RAID_DISK_S_OFFLINE:
157 		return ("OFFLINE");
158 	case G_RAID_DISK_S_DISABLED:
159 		return ("DISABLED");
160 	case G_RAID_DISK_S_FAILED:
161 		return ("FAILED");
162 	case G_RAID_DISK_S_STALE_FAILED:
163 		return ("STALE_FAILED");
164 	case G_RAID_DISK_S_SPARE:
165 		return ("SPARE");
166 	case G_RAID_DISK_S_STALE:
167 		return ("STALE");
168 	case G_RAID_DISK_S_ACTIVE:
169 		return ("ACTIVE");
170 	default:
171 		return ("INVALID");
172 	}
173 }
174 
175 static const char *
176 g_raid_disk_event2str(int event)
177 {
178 
179 	switch (event) {
180 	case G_RAID_DISK_E_DISCONNECTED:
181 		return ("DISCONNECTED");
182 	default:
183 		return ("INVALID");
184 	}
185 }
186 
187 const char *
188 g_raid_subdisk_state2str(int state)
189 {
190 
191 	switch (state) {
192 	case G_RAID_SUBDISK_S_NONE:
193 		return ("NONE");
194 	case G_RAID_SUBDISK_S_FAILED:
195 		return ("FAILED");
196 	case G_RAID_SUBDISK_S_NEW:
197 		return ("NEW");
198 	case G_RAID_SUBDISK_S_REBUILD:
199 		return ("REBUILD");
200 	case G_RAID_SUBDISK_S_UNINITIALIZED:
201 		return ("UNINITIALIZED");
202 	case G_RAID_SUBDISK_S_STALE:
203 		return ("STALE");
204 	case G_RAID_SUBDISK_S_RESYNC:
205 		return ("RESYNC");
206 	case G_RAID_SUBDISK_S_ACTIVE:
207 		return ("ACTIVE");
208 	default:
209 		return ("INVALID");
210 	}
211 }
212 
213 static const char *
214 g_raid_subdisk_event2str(int event)
215 {
216 
217 	switch (event) {
218 	case G_RAID_SUBDISK_E_NEW:
219 		return ("NEW");
220 	case G_RAID_SUBDISK_E_FAILED:
221 		return ("FAILED");
222 	case G_RAID_SUBDISK_E_DISCONNECTED:
223 		return ("DISCONNECTED");
224 	default:
225 		return ("INVALID");
226 	}
227 }
228 
229 const char *
230 g_raid_volume_state2str(int state)
231 {
232 
233 	switch (state) {
234 	case G_RAID_VOLUME_S_STARTING:
235 		return ("STARTING");
236 	case G_RAID_VOLUME_S_BROKEN:
237 		return ("BROKEN");
238 	case G_RAID_VOLUME_S_DEGRADED:
239 		return ("DEGRADED");
240 	case G_RAID_VOLUME_S_SUBOPTIMAL:
241 		return ("SUBOPTIMAL");
242 	case G_RAID_VOLUME_S_OPTIMAL:
243 		return ("OPTIMAL");
244 	case G_RAID_VOLUME_S_UNSUPPORTED:
245 		return ("UNSUPPORTED");
246 	case G_RAID_VOLUME_S_STOPPED:
247 		return ("STOPPED");
248 	default:
249 		return ("INVALID");
250 	}
251 }
252 
253 static const char *
254 g_raid_volume_event2str(int event)
255 {
256 
257 	switch (event) {
258 	case G_RAID_VOLUME_E_UP:
259 		return ("UP");
260 	case G_RAID_VOLUME_E_DOWN:
261 		return ("DOWN");
262 	case G_RAID_VOLUME_E_START:
263 		return ("START");
264 	case G_RAID_VOLUME_E_STARTMD:
265 		return ("STARTMD");
266 	default:
267 		return ("INVALID");
268 	}
269 }
270 
271 const char *
272 g_raid_volume_level2str(int level, int qual)
273 {
274 
275 	switch (level) {
276 	case G_RAID_VOLUME_RL_RAID0:
277 		return ("RAID0");
278 	case G_RAID_VOLUME_RL_RAID1:
279 		return ("RAID1");
280 	case G_RAID_VOLUME_RL_RAID3:
281 		if (qual == G_RAID_VOLUME_RLQ_R3P0)
282 			return ("RAID3-P0");
283 		if (qual == G_RAID_VOLUME_RLQ_R3PN)
284 			return ("RAID3-PN");
285 		return ("RAID3");
286 	case G_RAID_VOLUME_RL_RAID4:
287 		if (qual == G_RAID_VOLUME_RLQ_R4P0)
288 			return ("RAID4-P0");
289 		if (qual == G_RAID_VOLUME_RLQ_R4PN)
290 			return ("RAID4-PN");
291 		return ("RAID4");
292 	case G_RAID_VOLUME_RL_RAID5:
293 		if (qual == G_RAID_VOLUME_RLQ_R5RA)
294 			return ("RAID5-RA");
295 		if (qual == G_RAID_VOLUME_RLQ_R5RS)
296 			return ("RAID5-RS");
297 		if (qual == G_RAID_VOLUME_RLQ_R5LA)
298 			return ("RAID5-LA");
299 		if (qual == G_RAID_VOLUME_RLQ_R5LS)
300 			return ("RAID5-LS");
301 		return ("RAID5");
302 	case G_RAID_VOLUME_RL_RAID6:
303 		if (qual == G_RAID_VOLUME_RLQ_R6RA)
304 			return ("RAID6-RA");
305 		if (qual == G_RAID_VOLUME_RLQ_R6RS)
306 			return ("RAID6-RS");
307 		if (qual == G_RAID_VOLUME_RLQ_R6LA)
308 			return ("RAID6-LA");
309 		if (qual == G_RAID_VOLUME_RLQ_R6LS)
310 			return ("RAID6-LS");
311 		return ("RAID6");
312 	case G_RAID_VOLUME_RL_RAIDMDF:
313 		if (qual == G_RAID_VOLUME_RLQ_RMDFRA)
314 			return ("RAIDMDF-RA");
315 		if (qual == G_RAID_VOLUME_RLQ_RMDFRS)
316 			return ("RAIDMDF-RS");
317 		if (qual == G_RAID_VOLUME_RLQ_RMDFLA)
318 			return ("RAIDMDF-LA");
319 		if (qual == G_RAID_VOLUME_RLQ_RMDFLS)
320 			return ("RAIDMDF-LS");
321 		return ("RAIDMDF");
322 	case G_RAID_VOLUME_RL_RAID1E:
323 		if (qual == G_RAID_VOLUME_RLQ_R1EA)
324 			return ("RAID1E-A");
325 		if (qual == G_RAID_VOLUME_RLQ_R1EO)
326 			return ("RAID1E-O");
327 		return ("RAID1E");
328 	case G_RAID_VOLUME_RL_SINGLE:
329 		return ("SINGLE");
330 	case G_RAID_VOLUME_RL_CONCAT:
331 		return ("CONCAT");
332 	case G_RAID_VOLUME_RL_RAID5E:
333 		if (qual == G_RAID_VOLUME_RLQ_R5ERA)
334 			return ("RAID5E-RA");
335 		if (qual == G_RAID_VOLUME_RLQ_R5ERS)
336 			return ("RAID5E-RS");
337 		if (qual == G_RAID_VOLUME_RLQ_R5ELA)
338 			return ("RAID5E-LA");
339 		if (qual == G_RAID_VOLUME_RLQ_R5ELS)
340 			return ("RAID5E-LS");
341 		return ("RAID5E");
342 	case G_RAID_VOLUME_RL_RAID5EE:
343 		if (qual == G_RAID_VOLUME_RLQ_R5EERA)
344 			return ("RAID5EE-RA");
345 		if (qual == G_RAID_VOLUME_RLQ_R5EERS)
346 			return ("RAID5EE-RS");
347 		if (qual == G_RAID_VOLUME_RLQ_R5EELA)
348 			return ("RAID5EE-LA");
349 		if (qual == G_RAID_VOLUME_RLQ_R5EELS)
350 			return ("RAID5EE-LS");
351 		return ("RAID5EE");
352 	case G_RAID_VOLUME_RL_RAID5R:
353 		if (qual == G_RAID_VOLUME_RLQ_R5RRA)
354 			return ("RAID5R-RA");
355 		if (qual == G_RAID_VOLUME_RLQ_R5RRS)
356 			return ("RAID5R-RS");
357 		if (qual == G_RAID_VOLUME_RLQ_R5RLA)
358 			return ("RAID5R-LA");
359 		if (qual == G_RAID_VOLUME_RLQ_R5RLS)
360 			return ("RAID5R-LS");
361 		return ("RAID5E");
362 	default:
363 		return ("UNKNOWN");
364 	}
365 }
366 
367 int
368 g_raid_volume_str2level(const char *str, int *level, int *qual)
369 {
370 
371 	*level = G_RAID_VOLUME_RL_UNKNOWN;
372 	*qual = G_RAID_VOLUME_RLQ_NONE;
373 	if (strcasecmp(str, "RAID0") == 0)
374 		*level = G_RAID_VOLUME_RL_RAID0;
375 	else if (strcasecmp(str, "RAID1") == 0)
376 		*level = G_RAID_VOLUME_RL_RAID1;
377 	else if (strcasecmp(str, "RAID3-P0") == 0) {
378 		*level = G_RAID_VOLUME_RL_RAID3;
379 		*qual = G_RAID_VOLUME_RLQ_R3P0;
380 	} else if (strcasecmp(str, "RAID3-PN") == 0 ||
381 		   strcasecmp(str, "RAID3") == 0) {
382 		*level = G_RAID_VOLUME_RL_RAID3;
383 		*qual = G_RAID_VOLUME_RLQ_R3PN;
384 	} else if (strcasecmp(str, "RAID4-P0") == 0) {
385 		*level = G_RAID_VOLUME_RL_RAID4;
386 		*qual = G_RAID_VOLUME_RLQ_R4P0;
387 	} else if (strcasecmp(str, "RAID4-PN") == 0 ||
388 		   strcasecmp(str, "RAID4") == 0) {
389 		*level = G_RAID_VOLUME_RL_RAID4;
390 		*qual = G_RAID_VOLUME_RLQ_R4PN;
391 	} else if (strcasecmp(str, "RAID5-RA") == 0) {
392 		*level = G_RAID_VOLUME_RL_RAID5;
393 		*qual = G_RAID_VOLUME_RLQ_R5RA;
394 	} else if (strcasecmp(str, "RAID5-RS") == 0) {
395 		*level = G_RAID_VOLUME_RL_RAID5;
396 		*qual = G_RAID_VOLUME_RLQ_R5RS;
397 	} else if (strcasecmp(str, "RAID5") == 0 ||
398 		   strcasecmp(str, "RAID5-LA") == 0) {
399 		*level = G_RAID_VOLUME_RL_RAID5;
400 		*qual = G_RAID_VOLUME_RLQ_R5LA;
401 	} else if (strcasecmp(str, "RAID5-LS") == 0) {
402 		*level = G_RAID_VOLUME_RL_RAID5;
403 		*qual = G_RAID_VOLUME_RLQ_R5LS;
404 	} else if (strcasecmp(str, "RAID6-RA") == 0) {
405 		*level = G_RAID_VOLUME_RL_RAID6;
406 		*qual = G_RAID_VOLUME_RLQ_R6RA;
407 	} else if (strcasecmp(str, "RAID6-RS") == 0) {
408 		*level = G_RAID_VOLUME_RL_RAID6;
409 		*qual = G_RAID_VOLUME_RLQ_R6RS;
410 	} else if (strcasecmp(str, "RAID6") == 0 ||
411 		   strcasecmp(str, "RAID6-LA") == 0) {
412 		*level = G_RAID_VOLUME_RL_RAID6;
413 		*qual = G_RAID_VOLUME_RLQ_R6LA;
414 	} else if (strcasecmp(str, "RAID6-LS") == 0) {
415 		*level = G_RAID_VOLUME_RL_RAID6;
416 		*qual = G_RAID_VOLUME_RLQ_R6LS;
417 	} else if (strcasecmp(str, "RAIDMDF-RA") == 0) {
418 		*level = G_RAID_VOLUME_RL_RAIDMDF;
419 		*qual = G_RAID_VOLUME_RLQ_RMDFRA;
420 	} else if (strcasecmp(str, "RAIDMDF-RS") == 0) {
421 		*level = G_RAID_VOLUME_RL_RAIDMDF;
422 		*qual = G_RAID_VOLUME_RLQ_RMDFRS;
423 	} else if (strcasecmp(str, "RAIDMDF") == 0 ||
424 		   strcasecmp(str, "RAIDMDF-LA") == 0) {
425 		*level = G_RAID_VOLUME_RL_RAIDMDF;
426 		*qual = G_RAID_VOLUME_RLQ_RMDFLA;
427 	} else if (strcasecmp(str, "RAIDMDF-LS") == 0) {
428 		*level = G_RAID_VOLUME_RL_RAIDMDF;
429 		*qual = G_RAID_VOLUME_RLQ_RMDFLS;
430 	} else if (strcasecmp(str, "RAID10") == 0 ||
431 		   strcasecmp(str, "RAID1E") == 0 ||
432 		   strcasecmp(str, "RAID1E-A") == 0) {
433 		*level = G_RAID_VOLUME_RL_RAID1E;
434 		*qual = G_RAID_VOLUME_RLQ_R1EA;
435 	} else if (strcasecmp(str, "RAID1E-O") == 0) {
436 		*level = G_RAID_VOLUME_RL_RAID1E;
437 		*qual = G_RAID_VOLUME_RLQ_R1EO;
438 	} else if (strcasecmp(str, "SINGLE") == 0)
439 		*level = G_RAID_VOLUME_RL_SINGLE;
440 	else if (strcasecmp(str, "CONCAT") == 0)
441 		*level = G_RAID_VOLUME_RL_CONCAT;
442 	else if (strcasecmp(str, "RAID5E-RA") == 0) {
443 		*level = G_RAID_VOLUME_RL_RAID5E;
444 		*qual = G_RAID_VOLUME_RLQ_R5ERA;
445 	} else if (strcasecmp(str, "RAID5E-RS") == 0) {
446 		*level = G_RAID_VOLUME_RL_RAID5E;
447 		*qual = G_RAID_VOLUME_RLQ_R5ERS;
448 	} else if (strcasecmp(str, "RAID5E") == 0 ||
449 		   strcasecmp(str, "RAID5E-LA") == 0) {
450 		*level = G_RAID_VOLUME_RL_RAID5E;
451 		*qual = G_RAID_VOLUME_RLQ_R5ELA;
452 	} else if (strcasecmp(str, "RAID5E-LS") == 0) {
453 		*level = G_RAID_VOLUME_RL_RAID5E;
454 		*qual = G_RAID_VOLUME_RLQ_R5ELS;
455 	} else if (strcasecmp(str, "RAID5EE-RA") == 0) {
456 		*level = G_RAID_VOLUME_RL_RAID5EE;
457 		*qual = G_RAID_VOLUME_RLQ_R5EERA;
458 	} else if (strcasecmp(str, "RAID5EE-RS") == 0) {
459 		*level = G_RAID_VOLUME_RL_RAID5EE;
460 		*qual = G_RAID_VOLUME_RLQ_R5EERS;
461 	} else if (strcasecmp(str, "RAID5EE") == 0 ||
462 		   strcasecmp(str, "RAID5EE-LA") == 0) {
463 		*level = G_RAID_VOLUME_RL_RAID5EE;
464 		*qual = G_RAID_VOLUME_RLQ_R5EELA;
465 	} else if (strcasecmp(str, "RAID5EE-LS") == 0) {
466 		*level = G_RAID_VOLUME_RL_RAID5EE;
467 		*qual = G_RAID_VOLUME_RLQ_R5EELS;
468 	} else if (strcasecmp(str, "RAID5R-RA") == 0) {
469 		*level = G_RAID_VOLUME_RL_RAID5R;
470 		*qual = G_RAID_VOLUME_RLQ_R5RRA;
471 	} else if (strcasecmp(str, "RAID5R-RS") == 0) {
472 		*level = G_RAID_VOLUME_RL_RAID5R;
473 		*qual = G_RAID_VOLUME_RLQ_R5RRS;
474 	} else if (strcasecmp(str, "RAID5R") == 0 ||
475 		   strcasecmp(str, "RAID5R-LA") == 0) {
476 		*level = G_RAID_VOLUME_RL_RAID5R;
477 		*qual = G_RAID_VOLUME_RLQ_R5RLA;
478 	} else if (strcasecmp(str, "RAID5R-LS") == 0) {
479 		*level = G_RAID_VOLUME_RL_RAID5R;
480 		*qual = G_RAID_VOLUME_RLQ_R5RLS;
481 	} else
482 		return (-1);
483 	return (0);
484 }
485 
486 const char *
487 g_raid_get_diskname(struct g_raid_disk *disk)
488 {
489 
490 	if (disk->d_consumer == NULL || disk->d_consumer->provider == NULL)
491 		return ("[unknown]");
492 	return (disk->d_consumer->provider->name);
493 }
494 
495 void
496 g_raid_get_disk_info(struct g_raid_disk *disk)
497 {
498 	struct g_consumer *cp = disk->d_consumer;
499 	int error, len;
500 
501 	/* Read kernel dumping information. */
502 	disk->d_kd.offset = 0;
503 	disk->d_kd.length = OFF_MAX;
504 	len = sizeof(disk->d_kd);
505 	error = g_io_getattr("GEOM::kerneldump", cp, &len, &disk->d_kd);
506 	if (error)
507 		disk->d_kd.di.dumper = NULL;
508 	if (disk->d_kd.di.dumper == NULL)
509 		G_RAID_DEBUG1(2, disk->d_softc,
510 		    "Dumping not supported by %s: %d.",
511 		    cp->provider->name, error);
512 
513 	/* Read BIO_DELETE support. */
514 	error = g_getattr("GEOM::candelete", cp, &disk->d_candelete);
515 	if (error)
516 		disk->d_candelete = 0;
517 	if (!disk->d_candelete)
518 		G_RAID_DEBUG1(2, disk->d_softc,
519 		    "BIO_DELETE not supported by %s: %d.",
520 		    cp->provider->name, error);
521 }
522 
523 void
524 g_raid_report_disk_state(struct g_raid_disk *disk)
525 {
526 	struct g_raid_subdisk *sd;
527 	int len, state;
528 	uint32_t s;
529 
530 	if (disk->d_consumer == NULL)
531 		return;
532 	if (disk->d_state == G_RAID_DISK_S_DISABLED) {
533 		s = G_STATE_ACTIVE; /* XXX */
534 	} else if (disk->d_state == G_RAID_DISK_S_FAILED ||
535 	    disk->d_state == G_RAID_DISK_S_STALE_FAILED) {
536 		s = G_STATE_FAILED;
537 	} else {
538 		state = G_RAID_SUBDISK_S_ACTIVE;
539 		TAILQ_FOREACH(sd, &disk->d_subdisks, sd_next) {
540 			if (sd->sd_state < state)
541 				state = sd->sd_state;
542 		}
543 		if (state == G_RAID_SUBDISK_S_FAILED)
544 			s = G_STATE_FAILED;
545 		else if (state == G_RAID_SUBDISK_S_NEW ||
546 		    state == G_RAID_SUBDISK_S_REBUILD)
547 			s = G_STATE_REBUILD;
548 		else if (state == G_RAID_SUBDISK_S_STALE ||
549 		    state == G_RAID_SUBDISK_S_RESYNC)
550 			s = G_STATE_RESYNC;
551 		else
552 			s = G_STATE_ACTIVE;
553 	}
554 	len = sizeof(s);
555 	g_io_getattr("GEOM::setstate", disk->d_consumer, &len, &s);
556 	G_RAID_DEBUG1(2, disk->d_softc, "Disk %s state reported as %d.",
557 	    g_raid_get_diskname(disk), s);
558 }
559 
560 void
561 g_raid_change_disk_state(struct g_raid_disk *disk, int state)
562 {
563 
564 	G_RAID_DEBUG1(0, disk->d_softc, "Disk %s state changed from %s to %s.",
565 	    g_raid_get_diskname(disk),
566 	    g_raid_disk_state2str(disk->d_state),
567 	    g_raid_disk_state2str(state));
568 	disk->d_state = state;
569 	g_raid_report_disk_state(disk);
570 }
571 
572 void
573 g_raid_change_subdisk_state(struct g_raid_subdisk *sd, int state)
574 {
575 
576 	G_RAID_DEBUG1(0, sd->sd_softc,
577 	    "Subdisk %s:%d-%s state changed from %s to %s.",
578 	    sd->sd_volume->v_name, sd->sd_pos,
579 	    sd->sd_disk ? g_raid_get_diskname(sd->sd_disk) : "[none]",
580 	    g_raid_subdisk_state2str(sd->sd_state),
581 	    g_raid_subdisk_state2str(state));
582 	sd->sd_state = state;
583 	if (sd->sd_disk)
584 		g_raid_report_disk_state(sd->sd_disk);
585 }
586 
587 void
588 g_raid_change_volume_state(struct g_raid_volume *vol, int state)
589 {
590 
591 	G_RAID_DEBUG1(0, vol->v_softc,
592 	    "Volume %s state changed from %s to %s.",
593 	    vol->v_name,
594 	    g_raid_volume_state2str(vol->v_state),
595 	    g_raid_volume_state2str(state));
596 	vol->v_state = state;
597 }
598 
599 /*
600  * --- Events handling functions ---
601  * Events in geom_raid are used to maintain subdisks and volumes status
602  * from one thread to simplify locking.
603  */
604 static void
605 g_raid_event_free(struct g_raid_event *ep)
606 {
607 
608 	free(ep, M_RAID);
609 }
610 
611 int
612 g_raid_event_send(void *arg, int event, int flags)
613 {
614 	struct g_raid_softc *sc;
615 	struct g_raid_event *ep;
616 	int error;
617 
618 	if ((flags & G_RAID_EVENT_VOLUME) != 0) {
619 		sc = ((struct g_raid_volume *)arg)->v_softc;
620 	} else if ((flags & G_RAID_EVENT_DISK) != 0) {
621 		sc = ((struct g_raid_disk *)arg)->d_softc;
622 	} else if ((flags & G_RAID_EVENT_SUBDISK) != 0) {
623 		sc = ((struct g_raid_subdisk *)arg)->sd_softc;
624 	} else {
625 		sc = arg;
626 	}
627 	ep = malloc(sizeof(*ep), M_RAID,
628 	    sx_xlocked(&sc->sc_lock) ? M_WAITOK : M_NOWAIT);
629 	if (ep == NULL)
630 		return (ENOMEM);
631 	ep->e_tgt = arg;
632 	ep->e_event = event;
633 	ep->e_flags = flags;
634 	ep->e_error = 0;
635 	G_RAID_DEBUG1(4, sc, "Sending event %p. Waking up %p.", ep, sc);
636 	mtx_lock(&sc->sc_queue_mtx);
637 	TAILQ_INSERT_TAIL(&sc->sc_events, ep, e_next);
638 	mtx_unlock(&sc->sc_queue_mtx);
639 	wakeup(sc);
640 
641 	if ((flags & G_RAID_EVENT_WAIT) == 0)
642 		return (0);
643 
644 	sx_assert(&sc->sc_lock, SX_XLOCKED);
645 	G_RAID_DEBUG1(4, sc, "Sleeping on %p.", ep);
646 	sx_xunlock(&sc->sc_lock);
647 	while ((ep->e_flags & G_RAID_EVENT_DONE) == 0) {
648 		mtx_lock(&sc->sc_queue_mtx);
649 		MSLEEP(error, ep, &sc->sc_queue_mtx, PRIBIO | PDROP, "m:event",
650 		    hz * 5);
651 	}
652 	error = ep->e_error;
653 	g_raid_event_free(ep);
654 	sx_xlock(&sc->sc_lock);
655 	return (error);
656 }
657 
658 static void
659 g_raid_event_cancel(struct g_raid_softc *sc, void *tgt)
660 {
661 	struct g_raid_event *ep, *tmpep;
662 
663 	sx_assert(&sc->sc_lock, SX_XLOCKED);
664 
665 	mtx_lock(&sc->sc_queue_mtx);
666 	TAILQ_FOREACH_SAFE(ep, &sc->sc_events, e_next, tmpep) {
667 		if (ep->e_tgt != tgt)
668 			continue;
669 		TAILQ_REMOVE(&sc->sc_events, ep, e_next);
670 		if ((ep->e_flags & G_RAID_EVENT_WAIT) == 0)
671 			g_raid_event_free(ep);
672 		else {
673 			ep->e_error = ECANCELED;
674 			wakeup(ep);
675 		}
676 	}
677 	mtx_unlock(&sc->sc_queue_mtx);
678 }
679 
680 static int
681 g_raid_event_check(struct g_raid_softc *sc, void *tgt)
682 {
683 	struct g_raid_event *ep;
684 	int	res = 0;
685 
686 	sx_assert(&sc->sc_lock, SX_XLOCKED);
687 
688 	mtx_lock(&sc->sc_queue_mtx);
689 	TAILQ_FOREACH(ep, &sc->sc_events, e_next) {
690 		if (ep->e_tgt != tgt)
691 			continue;
692 		res = 1;
693 		break;
694 	}
695 	mtx_unlock(&sc->sc_queue_mtx);
696 	return (res);
697 }
698 
699 /*
700  * Return the number of disks in given state.
701  * If state is equal to -1, count all connected disks.
702  */
703 u_int
704 g_raid_ndisks(struct g_raid_softc *sc, int state)
705 {
706 	struct g_raid_disk *disk;
707 	u_int n;
708 
709 	sx_assert(&sc->sc_lock, SX_LOCKED);
710 
711 	n = 0;
712 	TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
713 		if (disk->d_state == state || state == -1)
714 			n++;
715 	}
716 	return (n);
717 }
718 
719 /*
720  * Return the number of subdisks in given state.
721  * If state is equal to -1, count all connected disks.
722  */
723 u_int
724 g_raid_nsubdisks(struct g_raid_volume *vol, int state)
725 {
726 	struct g_raid_subdisk *subdisk;
727 	struct g_raid_softc *sc __diagused;
728 	u_int i, n ;
729 
730 	sc = vol->v_softc;
731 	sx_assert(&sc->sc_lock, SX_LOCKED);
732 
733 	n = 0;
734 	for (i = 0; i < vol->v_disks_count; i++) {
735 		subdisk = &vol->v_subdisks[i];
736 		if ((state == -1 &&
737 		     subdisk->sd_state != G_RAID_SUBDISK_S_NONE) ||
738 		    subdisk->sd_state == state)
739 			n++;
740 	}
741 	return (n);
742 }
743 
744 /*
745  * Return the first subdisk in given state.
746  * If state is equal to -1, then the first connected disks.
747  */
748 struct g_raid_subdisk *
749 g_raid_get_subdisk(struct g_raid_volume *vol, int state)
750 {
751 	struct g_raid_subdisk *sd;
752 	struct g_raid_softc *sc __diagused;
753 	u_int i;
754 
755 	sc = vol->v_softc;
756 	sx_assert(&sc->sc_lock, SX_LOCKED);
757 
758 	for (i = 0; i < vol->v_disks_count; i++) {
759 		sd = &vol->v_subdisks[i];
760 		if ((state == -1 &&
761 		     sd->sd_state != G_RAID_SUBDISK_S_NONE) ||
762 		    sd->sd_state == state)
763 			return (sd);
764 	}
765 	return (NULL);
766 }
767 
768 struct g_consumer *
769 g_raid_open_consumer(struct g_raid_softc *sc, const char *name)
770 {
771 	struct g_consumer *cp;
772 	struct g_provider *pp;
773 
774 	g_topology_assert();
775 
776 	if (strncmp(name, _PATH_DEV, 5) == 0)
777 		name += 5;
778 	pp = g_provider_by_name(name);
779 	if (pp == NULL)
780 		return (NULL);
781 	cp = g_new_consumer(sc->sc_geom);
782 	cp->flags |= G_CF_DIRECT_RECEIVE;
783 	if (g_attach(cp, pp) != 0) {
784 		g_destroy_consumer(cp);
785 		return (NULL);
786 	}
787 	if (g_access(cp, 1, 1, 1) != 0) {
788 		g_detach(cp);
789 		g_destroy_consumer(cp);
790 		return (NULL);
791 	}
792 	return (cp);
793 }
794 
795 static u_int
796 g_raid_nrequests(struct g_raid_softc *sc, struct g_consumer *cp)
797 {
798 	struct bio *bp;
799 	u_int nreqs = 0;
800 
801 	mtx_lock(&sc->sc_queue_mtx);
802 	TAILQ_FOREACH(bp, &sc->sc_queue.queue, bio_queue) {
803 		if (bp->bio_from == cp)
804 			nreqs++;
805 	}
806 	mtx_unlock(&sc->sc_queue_mtx);
807 	return (nreqs);
808 }
809 
810 u_int
811 g_raid_nopens(struct g_raid_softc *sc)
812 {
813 	struct g_raid_volume *vol;
814 	u_int opens;
815 
816 	opens = 0;
817 	TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) {
818 		if (vol->v_provider_open != 0)
819 			opens++;
820 	}
821 	return (opens);
822 }
823 
824 static int
825 g_raid_consumer_is_busy(struct g_raid_softc *sc, struct g_consumer *cp)
826 {
827 
828 	if (cp->index > 0) {
829 		G_RAID_DEBUG1(2, sc,
830 		    "I/O requests for %s exist, can't destroy it now.",
831 		    cp->provider->name);
832 		return (1);
833 	}
834 	if (g_raid_nrequests(sc, cp) > 0) {
835 		G_RAID_DEBUG1(2, sc,
836 		    "I/O requests for %s in queue, can't destroy it now.",
837 		    cp->provider->name);
838 		return (1);
839 	}
840 	return (0);
841 }
842 
843 static void
844 g_raid_destroy_consumer(void *arg, int flags __unused)
845 {
846 	struct g_consumer *cp;
847 
848 	g_topology_assert();
849 
850 	cp = arg;
851 	G_RAID_DEBUG(1, "Consumer %s destroyed.", cp->provider->name);
852 	g_detach(cp);
853 	g_destroy_consumer(cp);
854 }
855 
856 void
857 g_raid_kill_consumer(struct g_raid_softc *sc, struct g_consumer *cp)
858 {
859 	struct g_provider *pp;
860 	int retaste_wait;
861 
862 	g_topology_assert_not();
863 
864 	g_topology_lock();
865 	cp->private = NULL;
866 	if (g_raid_consumer_is_busy(sc, cp))
867 		goto out;
868 	pp = cp->provider;
869 	retaste_wait = 0;
870 	if (cp->acw == 1) {
871 		if ((pp->geom->flags & G_GEOM_WITHER) == 0)
872 			retaste_wait = 1;
873 	}
874 	if (cp->acr > 0 || cp->acw > 0 || cp->ace > 0)
875 		g_access(cp, -cp->acr, -cp->acw, -cp->ace);
876 	if (retaste_wait) {
877 		/*
878 		 * After retaste event was send (inside g_access()), we can send
879 		 * event to detach and destroy consumer.
880 		 * A class, which has consumer to the given provider connected
881 		 * will not receive retaste event for the provider.
882 		 * This is the way how I ignore retaste events when I close
883 		 * consumers opened for write: I detach and destroy consumer
884 		 * after retaste event is sent.
885 		 */
886 		g_post_event(g_raid_destroy_consumer, cp, M_WAITOK, NULL);
887 		goto out;
888 	}
889 	G_RAID_DEBUG(1, "Consumer %s destroyed.", pp->name);
890 	g_detach(cp);
891 	g_destroy_consumer(cp);
892 out:
893 	g_topology_unlock();
894 }
895 
896 static void
897 g_raid_orphan(struct g_consumer *cp)
898 {
899 	struct g_raid_disk *disk;
900 
901 	g_topology_assert();
902 
903 	disk = cp->private;
904 	if (disk == NULL)
905 		return;
906 	g_raid_event_send(disk, G_RAID_DISK_E_DISCONNECTED,
907 	    G_RAID_EVENT_DISK);
908 }
909 
910 static void
911 g_raid_clean(struct g_raid_volume *vol, int acw)
912 {
913 	struct g_raid_softc *sc;
914 	int timeout;
915 
916 	sc = vol->v_softc;
917 	g_topology_assert_not();
918 	sx_assert(&sc->sc_lock, SX_XLOCKED);
919 
920 //	if ((sc->sc_flags & G_RAID_DEVICE_FLAG_NOFAILSYNC) != 0)
921 //		return;
922 	if (!vol->v_dirty)
923 		return;
924 	if (vol->v_writes > 0)
925 		return;
926 	if (acw > 0 || (acw == -1 &&
927 	    vol->v_provider != NULL && vol->v_provider->acw > 0)) {
928 		timeout = g_raid_clean_time - (time_uptime - vol->v_last_write);
929 		if (!g_raid_shutdown && timeout > 0)
930 			return;
931 	}
932 	vol->v_dirty = 0;
933 	G_RAID_DEBUG1(1, sc, "Volume %s marked as clean.",
934 	    vol->v_name);
935 	g_raid_write_metadata(sc, vol, NULL, NULL);
936 }
937 
938 static void
939 g_raid_dirty(struct g_raid_volume *vol)
940 {
941 	struct g_raid_softc *sc;
942 
943 	sc = vol->v_softc;
944 	g_topology_assert_not();
945 	sx_assert(&sc->sc_lock, SX_XLOCKED);
946 
947 //	if ((sc->sc_flags & G_RAID_DEVICE_FLAG_NOFAILSYNC) != 0)
948 //		return;
949 	vol->v_dirty = 1;
950 	G_RAID_DEBUG1(1, sc, "Volume %s marked as dirty.",
951 	    vol->v_name);
952 	g_raid_write_metadata(sc, vol, NULL, NULL);
953 }
954 
955 void
956 g_raid_tr_flush_common(struct g_raid_tr_object *tr, struct bio *bp)
957 {
958 	struct g_raid_volume *vol;
959 	struct g_raid_subdisk *sd;
960 	struct bio_queue_head queue;
961 	struct bio *cbp;
962 	int i;
963 
964 	vol = tr->tro_volume;
965 
966 	/*
967 	 * Allocate all bios before sending any request, so we can return
968 	 * ENOMEM in nice and clean way.
969 	 */
970 	bioq_init(&queue);
971 	for (i = 0; i < vol->v_disks_count; i++) {
972 		sd = &vol->v_subdisks[i];
973 		if (sd->sd_state == G_RAID_SUBDISK_S_NONE ||
974 		    sd->sd_state == G_RAID_SUBDISK_S_FAILED)
975 			continue;
976 		cbp = g_clone_bio(bp);
977 		if (cbp == NULL)
978 			goto failure;
979 		cbp->bio_caller1 = sd;
980 		bioq_insert_tail(&queue, cbp);
981 	}
982 	while ((cbp = bioq_takefirst(&queue)) != NULL) {
983 		sd = cbp->bio_caller1;
984 		cbp->bio_caller1 = NULL;
985 		g_raid_subdisk_iostart(sd, cbp);
986 	}
987 	return;
988 failure:
989 	while ((cbp = bioq_takefirst(&queue)) != NULL)
990 		g_destroy_bio(cbp);
991 	if (bp->bio_error == 0)
992 		bp->bio_error = ENOMEM;
993 	g_raid_iodone(bp, bp->bio_error);
994 }
995 
996 static void
997 g_raid_tr_kerneldump_common_done(struct bio *bp)
998 {
999 
1000 	bp->bio_flags |= BIO_DONE;
1001 }
1002 
1003 int
1004 g_raid_tr_kerneldump_common(struct g_raid_tr_object *tr,
1005     void *virtual, vm_offset_t physical, off_t offset, size_t length)
1006 {
1007 	struct g_raid_softc *sc;
1008 	struct g_raid_volume *vol;
1009 	struct bio bp;
1010 
1011 	vol = tr->tro_volume;
1012 	sc = vol->v_softc;
1013 
1014 	g_reset_bio(&bp);
1015 	bp.bio_cmd = BIO_WRITE;
1016 	bp.bio_done = g_raid_tr_kerneldump_common_done;
1017 	bp.bio_attribute = NULL;
1018 	bp.bio_offset = offset;
1019 	bp.bio_length = length;
1020 	bp.bio_data = virtual;
1021 	bp.bio_to = vol->v_provider;
1022 
1023 	g_raid_start(&bp);
1024 	while (!(bp.bio_flags & BIO_DONE)) {
1025 		G_RAID_DEBUG1(4, sc, "Poll...");
1026 		g_raid_poll(sc);
1027 		DELAY(10);
1028 	}
1029 
1030 	return (bp.bio_error != 0 ? EIO : 0);
1031 }
1032 
1033 static int
1034 g_raid_dump(void *arg, void *virtual, off_t offset, size_t length)
1035 {
1036 	struct g_raid_volume *vol;
1037 	int error;
1038 
1039 	vol = (struct g_raid_volume *)arg;
1040 	G_RAID_DEBUG1(3, vol->v_softc, "Dumping at off %llu len %llu.",
1041 	    (long long unsigned)offset, (long long unsigned)length);
1042 
1043 	error = G_RAID_TR_KERNELDUMP(vol->v_tr, virtual, offset, length);
1044 	return (error);
1045 }
1046 
1047 static void
1048 g_raid_kerneldump(struct g_raid_softc *sc, struct bio *bp)
1049 {
1050 	struct g_kerneldump *gkd;
1051 	struct g_provider *pp;
1052 	struct g_raid_volume *vol;
1053 
1054 	gkd = (struct g_kerneldump*)bp->bio_data;
1055 	pp = bp->bio_to;
1056 	vol = pp->private;
1057 	g_trace(G_T_TOPOLOGY, "g_raid_kerneldump(%s, %jd, %jd)",
1058 		pp->name, (intmax_t)gkd->offset, (intmax_t)gkd->length);
1059 	gkd->di.dumper = g_raid_dump;
1060 	gkd->di.priv = vol;
1061 	gkd->di.blocksize = vol->v_sectorsize;
1062 	gkd->di.maxiosize = DFLTPHYS;
1063 	gkd->di.mediaoffset = gkd->offset;
1064 	if ((gkd->offset + gkd->length) > vol->v_mediasize)
1065 		gkd->length = vol->v_mediasize - gkd->offset;
1066 	gkd->di.mediasize = gkd->length;
1067 	g_io_deliver(bp, 0);
1068 }
1069 
1070 static void
1071 g_raid_candelete(struct g_raid_softc *sc, struct bio *bp)
1072 {
1073 	struct g_provider *pp;
1074 	struct g_raid_volume *vol;
1075 	struct g_raid_subdisk *sd;
1076 	int i, val;
1077 
1078 	pp = bp->bio_to;
1079 	vol = pp->private;
1080 	for (i = 0; i < vol->v_disks_count; i++) {
1081 		sd = &vol->v_subdisks[i];
1082 		if (sd->sd_state == G_RAID_SUBDISK_S_NONE)
1083 			continue;
1084 		if (sd->sd_disk->d_candelete)
1085 			break;
1086 	}
1087 	val = i < vol->v_disks_count;
1088 	g_handleattr(bp, "GEOM::candelete", &val, sizeof(val));
1089 }
1090 
1091 static void
1092 g_raid_start(struct bio *bp)
1093 {
1094 	struct g_raid_softc *sc;
1095 
1096 	sc = bp->bio_to->geom->softc;
1097 	/*
1098 	 * If sc == NULL or there are no valid disks, provider's error
1099 	 * should be set and g_raid_start() should not be called at all.
1100 	 */
1101 //	KASSERT(sc != NULL && sc->sc_state == G_RAID_VOLUME_S_RUNNING,
1102 //	    ("Provider's error should be set (error=%d)(mirror=%s).",
1103 //	    bp->bio_to->error, bp->bio_to->name));
1104 	G_RAID_LOGREQ(3, bp, "Request received.");
1105 
1106 	switch (bp->bio_cmd) {
1107 	case BIO_READ:
1108 	case BIO_WRITE:
1109 	case BIO_DELETE:
1110 	case BIO_FLUSH:
1111 	case BIO_SPEEDUP:
1112 		break;
1113 	case BIO_GETATTR:
1114 		if (!strcmp(bp->bio_attribute, "GEOM::candelete"))
1115 			g_raid_candelete(sc, bp);
1116 		else if (!strcmp(bp->bio_attribute, "GEOM::kerneldump"))
1117 			g_raid_kerneldump(sc, bp);
1118 		else
1119 			g_io_deliver(bp, EOPNOTSUPP);
1120 		return;
1121 	default:
1122 		g_io_deliver(bp, EOPNOTSUPP);
1123 		return;
1124 	}
1125 	mtx_lock(&sc->sc_queue_mtx);
1126 	bioq_insert_tail(&sc->sc_queue, bp);
1127 	mtx_unlock(&sc->sc_queue_mtx);
1128 	if (!dumping) {
1129 		G_RAID_DEBUG1(4, sc, "Waking up %p.", sc);
1130 		wakeup(sc);
1131 	}
1132 }
1133 
1134 static int
1135 g_raid_bio_overlaps(const struct bio *bp, off_t lstart, off_t len)
1136 {
1137 	/*
1138 	 * 5 cases:
1139 	 * (1) bp entirely below NO
1140 	 * (2) bp entirely above NO
1141 	 * (3) bp start below, but end in range YES
1142 	 * (4) bp entirely within YES
1143 	 * (5) bp starts within, ends above YES
1144 	 *
1145 	 * lock range 10-19 (offset 10 length 10)
1146 	 * (1) 1-5: first if kicks it out
1147 	 * (2) 30-35: second if kicks it out
1148 	 * (3) 5-15: passes both ifs
1149 	 * (4) 12-14: passes both ifs
1150 	 * (5) 19-20: passes both
1151 	 */
1152 	off_t lend = lstart + len - 1;
1153 	off_t bstart = bp->bio_offset;
1154 	off_t bend = bp->bio_offset + bp->bio_length - 1;
1155 
1156 	if (bend < lstart)
1157 		return (0);
1158 	if (lend < bstart)
1159 		return (0);
1160 	return (1);
1161 }
1162 
1163 static int
1164 g_raid_is_in_locked_range(struct g_raid_volume *vol, const struct bio *bp)
1165 {
1166 	struct g_raid_lock *lp;
1167 
1168 	sx_assert(&vol->v_softc->sc_lock, SX_LOCKED);
1169 
1170 	LIST_FOREACH(lp, &vol->v_locks, l_next) {
1171 		if (g_raid_bio_overlaps(bp, lp->l_offset, lp->l_length))
1172 			return (1);
1173 	}
1174 	return (0);
1175 }
1176 
1177 static void
1178 g_raid_start_request(struct bio *bp)
1179 {
1180 	struct g_raid_softc *sc __diagused;
1181 	struct g_raid_volume *vol;
1182 
1183 	sc = bp->bio_to->geom->softc;
1184 	sx_assert(&sc->sc_lock, SX_LOCKED);
1185 	vol = bp->bio_to->private;
1186 
1187 	/*
1188 	 * Check to see if this item is in a locked range.  If so,
1189 	 * queue it to our locked queue and return.  We'll requeue
1190 	 * it when the range is unlocked.  Internal I/O for the
1191 	 * rebuild/rescan/recovery process is excluded from this
1192 	 * check so we can actually do the recovery.
1193 	 */
1194 	if (!(bp->bio_cflags & G_RAID_BIO_FLAG_SPECIAL) &&
1195 	    g_raid_is_in_locked_range(vol, bp)) {
1196 		G_RAID_LOGREQ(3, bp, "Defer request.");
1197 		bioq_insert_tail(&vol->v_locked, bp);
1198 		return;
1199 	}
1200 
1201 	/*
1202 	 * If we're actually going to do the write/delete, then
1203 	 * update the idle stats for the volume.
1204 	 */
1205 	if (bp->bio_cmd == BIO_WRITE || bp->bio_cmd == BIO_DELETE) {
1206 		if (!vol->v_dirty)
1207 			g_raid_dirty(vol);
1208 		vol->v_writes++;
1209 	}
1210 
1211 	/*
1212 	 * Put request onto inflight queue, so we can check if new
1213 	 * synchronization requests don't collide with it.  Then tell
1214 	 * the transformation layer to start the I/O.
1215 	 */
1216 	bioq_insert_tail(&vol->v_inflight, bp);
1217 	G_RAID_LOGREQ(4, bp, "Request started");
1218 	G_RAID_TR_IOSTART(vol->v_tr, bp);
1219 }
1220 
1221 static void
1222 g_raid_finish_with_locked_ranges(struct g_raid_volume *vol, struct bio *bp)
1223 {
1224 	off_t off, len;
1225 	struct bio *nbp;
1226 	struct g_raid_lock *lp;
1227 
1228 	vol->v_pending_lock = 0;
1229 	LIST_FOREACH(lp, &vol->v_locks, l_next) {
1230 		if (lp->l_pending) {
1231 			off = lp->l_offset;
1232 			len = lp->l_length;
1233 			lp->l_pending = 0;
1234 			TAILQ_FOREACH(nbp, &vol->v_inflight.queue, bio_queue) {
1235 				if (g_raid_bio_overlaps(nbp, off, len))
1236 					lp->l_pending++;
1237 			}
1238 			if (lp->l_pending) {
1239 				vol->v_pending_lock = 1;
1240 				G_RAID_DEBUG1(4, vol->v_softc,
1241 				    "Deferred lock(%jd, %jd) has %d pending",
1242 				    (intmax_t)off, (intmax_t)(off + len),
1243 				    lp->l_pending);
1244 				continue;
1245 			}
1246 			G_RAID_DEBUG1(4, vol->v_softc,
1247 			    "Deferred lock of %jd to %jd completed",
1248 			    (intmax_t)off, (intmax_t)(off + len));
1249 			G_RAID_TR_LOCKED(vol->v_tr, lp->l_callback_arg);
1250 		}
1251 	}
1252 }
1253 
1254 void
1255 g_raid_iodone(struct bio *bp, int error)
1256 {
1257 	struct g_raid_softc *sc __diagused;
1258 	struct g_raid_volume *vol;
1259 
1260 	sc = bp->bio_to->geom->softc;
1261 	sx_assert(&sc->sc_lock, SX_LOCKED);
1262 	vol = bp->bio_to->private;
1263 	G_RAID_LOGREQ(3, bp, "Request done: %d.", error);
1264 
1265 	/* Update stats if we done write/delete. */
1266 	if (bp->bio_cmd == BIO_WRITE || bp->bio_cmd == BIO_DELETE) {
1267 		vol->v_writes--;
1268 		vol->v_last_write = time_uptime;
1269 	}
1270 
1271 	bioq_remove(&vol->v_inflight, bp);
1272 	if (vol->v_pending_lock && g_raid_is_in_locked_range(vol, bp))
1273 		g_raid_finish_with_locked_ranges(vol, bp);
1274 	getmicrouptime(&vol->v_last_done);
1275 	g_io_deliver(bp, error);
1276 }
1277 
1278 int
1279 g_raid_lock_range(struct g_raid_volume *vol, off_t off, off_t len,
1280     struct bio *ignore, void *argp)
1281 {
1282 	struct g_raid_softc *sc;
1283 	struct g_raid_lock *lp;
1284 	struct bio *bp;
1285 
1286 	sc = vol->v_softc;
1287 	lp = malloc(sizeof(*lp), M_RAID, M_WAITOK | M_ZERO);
1288 	LIST_INSERT_HEAD(&vol->v_locks, lp, l_next);
1289 	lp->l_offset = off;
1290 	lp->l_length = len;
1291 	lp->l_callback_arg = argp;
1292 
1293 	lp->l_pending = 0;
1294 	TAILQ_FOREACH(bp, &vol->v_inflight.queue, bio_queue) {
1295 		if (bp != ignore && g_raid_bio_overlaps(bp, off, len))
1296 			lp->l_pending++;
1297 	}
1298 
1299 	/*
1300 	 * If there are any writes that are pending, we return EBUSY.  All
1301 	 * callers will have to wait until all pending writes clear.
1302 	 */
1303 	if (lp->l_pending > 0) {
1304 		vol->v_pending_lock = 1;
1305 		G_RAID_DEBUG1(4, sc, "Locking range %jd to %jd deferred %d pend",
1306 		    (intmax_t)off, (intmax_t)(off+len), lp->l_pending);
1307 		return (EBUSY);
1308 	}
1309 	G_RAID_DEBUG1(4, sc, "Locking range %jd to %jd",
1310 	    (intmax_t)off, (intmax_t)(off+len));
1311 	G_RAID_TR_LOCKED(vol->v_tr, lp->l_callback_arg);
1312 	return (0);
1313 }
1314 
1315 int
1316 g_raid_unlock_range(struct g_raid_volume *vol, off_t off, off_t len)
1317 {
1318 	struct g_raid_lock *lp;
1319 	struct g_raid_softc *sc;
1320 	struct bio *bp;
1321 
1322 	sc = vol->v_softc;
1323 	LIST_FOREACH(lp, &vol->v_locks, l_next) {
1324 		if (lp->l_offset == off && lp->l_length == len) {
1325 			LIST_REMOVE(lp, l_next);
1326 			/* XXX
1327 			 * Right now we just put them all back on the queue
1328 			 * and hope for the best.  We hope this because any
1329 			 * locked ranges will go right back on this list
1330 			 * when the worker thread runs.
1331 			 * XXX
1332 			 */
1333 			G_RAID_DEBUG1(4, sc, "Unlocked %jd to %jd",
1334 			    (intmax_t)lp->l_offset,
1335 			    (intmax_t)(lp->l_offset+lp->l_length));
1336 			mtx_lock(&sc->sc_queue_mtx);
1337 			while ((bp = bioq_takefirst(&vol->v_locked)) != NULL)
1338 				bioq_insert_tail(&sc->sc_queue, bp);
1339 			mtx_unlock(&sc->sc_queue_mtx);
1340 			free(lp, M_RAID);
1341 			return (0);
1342 		}
1343 	}
1344 	return (EINVAL);
1345 }
1346 
1347 void
1348 g_raid_subdisk_iostart(struct g_raid_subdisk *sd, struct bio *bp)
1349 {
1350 	struct g_consumer *cp;
1351 	struct g_raid_disk *disk, *tdisk;
1352 
1353 	bp->bio_caller1 = sd;
1354 
1355 	/*
1356 	 * Make sure that the disk is present. Generally it is a task of
1357 	 * transformation layers to not send requests to absent disks, but
1358 	 * it is better to be safe and report situation then sorry.
1359 	 */
1360 	if (sd->sd_disk == NULL) {
1361 		G_RAID_LOGREQ(0, bp, "Warning! I/O request to an absent disk!");
1362 nodisk:
1363 		bp->bio_from = NULL;
1364 		bp->bio_to = NULL;
1365 		bp->bio_error = ENXIO;
1366 		g_raid_disk_done(bp);
1367 		return;
1368 	}
1369 	disk = sd->sd_disk;
1370 	if (disk->d_state != G_RAID_DISK_S_ACTIVE &&
1371 	    disk->d_state != G_RAID_DISK_S_FAILED) {
1372 		G_RAID_LOGREQ(0, bp, "Warning! I/O request to a disk in a "
1373 		    "wrong state (%s)!", g_raid_disk_state2str(disk->d_state));
1374 		goto nodisk;
1375 	}
1376 
1377 	cp = disk->d_consumer;
1378 	bp->bio_from = cp;
1379 	bp->bio_to = cp->provider;
1380 	cp->index++;
1381 
1382 	/* Update average disks load. */
1383 	TAILQ_FOREACH(tdisk, &sd->sd_softc->sc_disks, d_next) {
1384 		if (tdisk->d_consumer == NULL)
1385 			tdisk->d_load = 0;
1386 		else
1387 			tdisk->d_load = (tdisk->d_consumer->index *
1388 			    G_RAID_SUBDISK_LOAD_SCALE + tdisk->d_load * 7) / 8;
1389 	}
1390 
1391 	disk->d_last_offset = bp->bio_offset + bp->bio_length;
1392 	if (dumping) {
1393 		G_RAID_LOGREQ(3, bp, "Sending dumping request.");
1394 		if (bp->bio_cmd == BIO_WRITE) {
1395 			bp->bio_error = g_raid_subdisk_kerneldump(sd,
1396 			    bp->bio_data, bp->bio_offset, bp->bio_length);
1397 		} else
1398 			bp->bio_error = EOPNOTSUPP;
1399 		g_raid_disk_done(bp);
1400 	} else {
1401 		bp->bio_done = g_raid_disk_done;
1402 		bp->bio_offset += sd->sd_offset;
1403 		G_RAID_LOGREQ(3, bp, "Sending request.");
1404 		g_io_request(bp, cp);
1405 	}
1406 }
1407 
1408 int
1409 g_raid_subdisk_kerneldump(struct g_raid_subdisk *sd, void *virtual,
1410     off_t offset, size_t length)
1411 {
1412 
1413 	if (sd->sd_disk == NULL)
1414 		return (ENXIO);
1415 	if (sd->sd_disk->d_kd.di.dumper == NULL)
1416 		return (EOPNOTSUPP);
1417 	return (dump_write(&sd->sd_disk->d_kd.di, virtual,
1418 	    sd->sd_disk->d_kd.di.mediaoffset + sd->sd_offset + offset, length));
1419 }
1420 
1421 static void
1422 g_raid_disk_done(struct bio *bp)
1423 {
1424 	struct g_raid_softc *sc;
1425 	struct g_raid_subdisk *sd;
1426 
1427 	sd = bp->bio_caller1;
1428 	sc = sd->sd_softc;
1429 	mtx_lock(&sc->sc_queue_mtx);
1430 	bioq_insert_tail(&sc->sc_queue, bp);
1431 	mtx_unlock(&sc->sc_queue_mtx);
1432 	if (!dumping)
1433 		wakeup(sc);
1434 }
1435 
1436 static void
1437 g_raid_disk_done_request(struct bio *bp)
1438 {
1439 	struct g_raid_softc *sc;
1440 	struct g_raid_disk *disk;
1441 	struct g_raid_subdisk *sd;
1442 	struct g_raid_volume *vol;
1443 
1444 	g_topology_assert_not();
1445 
1446 	G_RAID_LOGREQ(3, bp, "Disk request done: %d.", bp->bio_error);
1447 	sd = bp->bio_caller1;
1448 	sc = sd->sd_softc;
1449 	vol = sd->sd_volume;
1450 	if (bp->bio_from != NULL) {
1451 		bp->bio_from->index--;
1452 		disk = bp->bio_from->private;
1453 		if (disk == NULL)
1454 			g_raid_kill_consumer(sc, bp->bio_from);
1455 	}
1456 	bp->bio_offset -= sd->sd_offset;
1457 
1458 	G_RAID_TR_IODONE(vol->v_tr, sd, bp);
1459 }
1460 
1461 static void
1462 g_raid_handle_event(struct g_raid_softc *sc, struct g_raid_event *ep)
1463 {
1464 
1465 	if ((ep->e_flags & G_RAID_EVENT_VOLUME) != 0)
1466 		ep->e_error = g_raid_update_volume(ep->e_tgt, ep->e_event);
1467 	else if ((ep->e_flags & G_RAID_EVENT_DISK) != 0)
1468 		ep->e_error = g_raid_update_disk(ep->e_tgt, ep->e_event);
1469 	else if ((ep->e_flags & G_RAID_EVENT_SUBDISK) != 0)
1470 		ep->e_error = g_raid_update_subdisk(ep->e_tgt, ep->e_event);
1471 	else
1472 		ep->e_error = g_raid_update_node(ep->e_tgt, ep->e_event);
1473 	if ((ep->e_flags & G_RAID_EVENT_WAIT) == 0) {
1474 		KASSERT(ep->e_error == 0,
1475 		    ("Error cannot be handled."));
1476 		g_raid_event_free(ep);
1477 	} else {
1478 		ep->e_flags |= G_RAID_EVENT_DONE;
1479 		G_RAID_DEBUG1(4, sc, "Waking up %p.", ep);
1480 		mtx_lock(&sc->sc_queue_mtx);
1481 		wakeup(ep);
1482 		mtx_unlock(&sc->sc_queue_mtx);
1483 	}
1484 }
1485 
1486 /*
1487  * Worker thread.
1488  */
1489 static void
1490 g_raid_worker(void *arg)
1491 {
1492 	struct g_raid_softc *sc;
1493 	struct g_raid_event *ep;
1494 	struct g_raid_volume *vol;
1495 	struct bio *bp;
1496 	struct timeval now, t;
1497 	int timeout, rv;
1498 
1499 	sc = arg;
1500 	thread_lock(curthread);
1501 	sched_prio(curthread, PRIBIO);
1502 	thread_unlock(curthread);
1503 
1504 	sx_xlock(&sc->sc_lock);
1505 	for (;;) {
1506 		mtx_lock(&sc->sc_queue_mtx);
1507 		/*
1508 		 * First take a look at events.
1509 		 * This is important to handle events before any I/O requests.
1510 		 */
1511 		bp = NULL;
1512 		vol = NULL;
1513 		rv = 0;
1514 		ep = TAILQ_FIRST(&sc->sc_events);
1515 		if (ep != NULL)
1516 			TAILQ_REMOVE(&sc->sc_events, ep, e_next);
1517 		else if ((bp = bioq_takefirst(&sc->sc_queue)) != NULL)
1518 			;
1519 		else {
1520 			getmicrouptime(&now);
1521 			t = now;
1522 			TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) {
1523 				if (bioq_first(&vol->v_inflight) == NULL &&
1524 				    vol->v_tr &&
1525 				    timevalcmp(&vol->v_last_done, &t, < ))
1526 					t = vol->v_last_done;
1527 			}
1528 			timevalsub(&t, &now);
1529 			timeout = g_raid_idle_threshold +
1530 			    t.tv_sec * 1000000 + t.tv_usec;
1531 			if (timeout > 0) {
1532 				/*
1533 				 * Two steps to avoid overflows at HZ=1000
1534 				 * and idle timeouts > 2.1s.  Some rounding
1535 				 * errors can occur, but they are < 1tick,
1536 				 * which is deemed to be close enough for
1537 				 * this purpose.
1538 				 */
1539 				int micpertic = 1000000 / hz;
1540 				timeout = (timeout + micpertic - 1) / micpertic;
1541 				sx_xunlock(&sc->sc_lock);
1542 				MSLEEP(rv, sc, &sc->sc_queue_mtx,
1543 				    PRIBIO | PDROP, "-", timeout);
1544 				sx_xlock(&sc->sc_lock);
1545 				goto process;
1546 			} else
1547 				rv = EWOULDBLOCK;
1548 		}
1549 		mtx_unlock(&sc->sc_queue_mtx);
1550 process:
1551 		if (ep != NULL) {
1552 			g_raid_handle_event(sc, ep);
1553 		} else if (bp != NULL) {
1554 			if (bp->bio_to != NULL &&
1555 			    bp->bio_to->geom == sc->sc_geom)
1556 				g_raid_start_request(bp);
1557 			else
1558 				g_raid_disk_done_request(bp);
1559 		} else if (rv == EWOULDBLOCK) {
1560 			TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) {
1561 				g_raid_clean(vol, -1);
1562 				if (bioq_first(&vol->v_inflight) == NULL &&
1563 				    vol->v_tr) {
1564 					t.tv_sec = g_raid_idle_threshold / 1000000;
1565 					t.tv_usec = g_raid_idle_threshold % 1000000;
1566 					timevaladd(&t, &vol->v_last_done);
1567 					getmicrouptime(&now);
1568 					if (timevalcmp(&t, &now, <= )) {
1569 						G_RAID_TR_IDLE(vol->v_tr);
1570 						vol->v_last_done = now;
1571 					}
1572 				}
1573 			}
1574 		}
1575 		if (sc->sc_stopping == G_RAID_DESTROY_HARD)
1576 			g_raid_destroy_node(sc, 1);	/* May not return. */
1577 	}
1578 }
1579 
1580 static void
1581 g_raid_poll(struct g_raid_softc *sc)
1582 {
1583 	struct g_raid_event *ep;
1584 	struct bio *bp;
1585 
1586 	sx_xlock(&sc->sc_lock);
1587 	mtx_lock(&sc->sc_queue_mtx);
1588 	/*
1589 	 * First take a look at events.
1590 	 * This is important to handle events before any I/O requests.
1591 	 */
1592 	ep = TAILQ_FIRST(&sc->sc_events);
1593 	if (ep != NULL) {
1594 		TAILQ_REMOVE(&sc->sc_events, ep, e_next);
1595 		mtx_unlock(&sc->sc_queue_mtx);
1596 		g_raid_handle_event(sc, ep);
1597 		goto out;
1598 	}
1599 	bp = bioq_takefirst(&sc->sc_queue);
1600 	if (bp != NULL) {
1601 		mtx_unlock(&sc->sc_queue_mtx);
1602 		if (bp->bio_from == NULL ||
1603 		    bp->bio_from->geom != sc->sc_geom)
1604 			g_raid_start_request(bp);
1605 		else
1606 			g_raid_disk_done_request(bp);
1607 	}
1608 out:
1609 	sx_xunlock(&sc->sc_lock);
1610 }
1611 
1612 static void
1613 g_raid_launch_provider(struct g_raid_volume *vol)
1614 {
1615 	struct g_raid_disk *disk;
1616 	struct g_raid_subdisk *sd;
1617 	struct g_raid_softc *sc;
1618 	struct g_provider *pp;
1619 	char name[G_RAID_MAX_VOLUMENAME];
1620 	off_t off;
1621 	int i;
1622 
1623 	sc = vol->v_softc;
1624 	sx_assert(&sc->sc_lock, SX_LOCKED);
1625 
1626 	g_topology_lock();
1627 	/* Try to name provider with volume name. */
1628 	snprintf(name, sizeof(name), "raid/%s", vol->v_name);
1629 	if (g_raid_name_format == 0 || vol->v_name[0] == 0 ||
1630 	    g_provider_by_name(name) != NULL) {
1631 		/* Otherwise use sequential volume number. */
1632 		snprintf(name, sizeof(name), "raid/r%d", vol->v_global_id);
1633 	}
1634 
1635 	pp = g_new_providerf(sc->sc_geom, "%s", name);
1636 	pp->flags |= G_PF_DIRECT_RECEIVE;
1637 	if (vol->v_tr->tro_class->trc_accept_unmapped) {
1638 		pp->flags |= G_PF_ACCEPT_UNMAPPED;
1639 		for (i = 0; i < vol->v_disks_count; i++) {
1640 			sd = &vol->v_subdisks[i];
1641 			if (sd->sd_state == G_RAID_SUBDISK_S_NONE)
1642 				continue;
1643 			if ((sd->sd_disk->d_consumer->provider->flags &
1644 			    G_PF_ACCEPT_UNMAPPED) == 0)
1645 				pp->flags &= ~G_PF_ACCEPT_UNMAPPED;
1646 		}
1647 	}
1648 	pp->private = vol;
1649 	pp->mediasize = vol->v_mediasize;
1650 	pp->sectorsize = vol->v_sectorsize;
1651 	pp->stripesize = 0;
1652 	pp->stripeoffset = 0;
1653 	if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID1 ||
1654 	    vol->v_raid_level == G_RAID_VOLUME_RL_RAID3 ||
1655 	    vol->v_raid_level == G_RAID_VOLUME_RL_SINGLE ||
1656 	    vol->v_raid_level == G_RAID_VOLUME_RL_CONCAT) {
1657 		if ((disk = vol->v_subdisks[0].sd_disk) != NULL &&
1658 		    disk->d_consumer != NULL &&
1659 		    disk->d_consumer->provider != NULL) {
1660 			pp->stripesize = disk->d_consumer->provider->stripesize;
1661 			off = disk->d_consumer->provider->stripeoffset;
1662 			pp->stripeoffset = off + vol->v_subdisks[0].sd_offset;
1663 			if (off > 0)
1664 				pp->stripeoffset %= off;
1665 		}
1666 		if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID3) {
1667 			pp->stripesize *= (vol->v_disks_count - 1);
1668 			pp->stripeoffset *= (vol->v_disks_count - 1);
1669 		}
1670 	} else
1671 		pp->stripesize = vol->v_strip_size;
1672 	vol->v_provider = pp;
1673 	g_error_provider(pp, 0);
1674 	g_topology_unlock();
1675 	G_RAID_DEBUG1(0, sc, "Provider %s for volume %s created.",
1676 	    pp->name, vol->v_name);
1677 }
1678 
1679 static void
1680 g_raid_destroy_provider(struct g_raid_volume *vol)
1681 {
1682 	struct g_raid_softc *sc;
1683 	struct g_provider *pp;
1684 	struct bio *bp, *tmp;
1685 
1686 	g_topology_assert_not();
1687 	sc = vol->v_softc;
1688 	pp = vol->v_provider;
1689 	KASSERT(pp != NULL, ("NULL provider (volume=%s).", vol->v_name));
1690 
1691 	g_topology_lock();
1692 	g_error_provider(pp, ENXIO);
1693 	mtx_lock(&sc->sc_queue_mtx);
1694 	TAILQ_FOREACH_SAFE(bp, &sc->sc_queue.queue, bio_queue, tmp) {
1695 		if (bp->bio_to != pp)
1696 			continue;
1697 		bioq_remove(&sc->sc_queue, bp);
1698 		g_io_deliver(bp, ENXIO);
1699 	}
1700 	mtx_unlock(&sc->sc_queue_mtx);
1701 	G_RAID_DEBUG1(0, sc, "Provider %s for volume %s destroyed.",
1702 	    pp->name, vol->v_name);
1703 	g_wither_provider(pp, ENXIO);
1704 	g_topology_unlock();
1705 	vol->v_provider = NULL;
1706 }
1707 
1708 /*
1709  * Update device state.
1710  */
1711 static int
1712 g_raid_update_volume(struct g_raid_volume *vol, u_int event)
1713 {
1714 	struct g_raid_softc *sc;
1715 
1716 	sc = vol->v_softc;
1717 	sx_assert(&sc->sc_lock, SX_XLOCKED);
1718 
1719 	G_RAID_DEBUG1(2, sc, "Event %s for volume %s.",
1720 	    g_raid_volume_event2str(event),
1721 	    vol->v_name);
1722 	switch (event) {
1723 	case G_RAID_VOLUME_E_DOWN:
1724 		if (vol->v_provider != NULL)
1725 			g_raid_destroy_provider(vol);
1726 		break;
1727 	case G_RAID_VOLUME_E_UP:
1728 		if (vol->v_provider == NULL)
1729 			g_raid_launch_provider(vol);
1730 		break;
1731 	case G_RAID_VOLUME_E_START:
1732 		if (vol->v_tr)
1733 			G_RAID_TR_START(vol->v_tr);
1734 		return (0);
1735 	default:
1736 		if (sc->sc_md)
1737 			G_RAID_MD_VOLUME_EVENT(sc->sc_md, vol, event);
1738 		return (0);
1739 	}
1740 
1741 	/* Manage root mount release. */
1742 	if (vol->v_starting) {
1743 		vol->v_starting = 0;
1744 		G_RAID_DEBUG1(1, sc, "root_mount_rel %p", vol->v_rootmount);
1745 		root_mount_rel(vol->v_rootmount);
1746 		vol->v_rootmount = NULL;
1747 	}
1748 	if (vol->v_stopping && vol->v_provider_open == 0)
1749 		g_raid_destroy_volume(vol);
1750 	return (0);
1751 }
1752 
1753 /*
1754  * Update subdisk state.
1755  */
1756 static int
1757 g_raid_update_subdisk(struct g_raid_subdisk *sd, u_int event)
1758 {
1759 	struct g_raid_softc *sc;
1760 	struct g_raid_volume *vol;
1761 
1762 	sc = sd->sd_softc;
1763 	vol = sd->sd_volume;
1764 	sx_assert(&sc->sc_lock, SX_XLOCKED);
1765 
1766 	G_RAID_DEBUG1(2, sc, "Event %s for subdisk %s:%d-%s.",
1767 	    g_raid_subdisk_event2str(event),
1768 	    vol->v_name, sd->sd_pos,
1769 	    sd->sd_disk ? g_raid_get_diskname(sd->sd_disk) : "[none]");
1770 	if (vol->v_tr)
1771 		G_RAID_TR_EVENT(vol->v_tr, sd, event);
1772 
1773 	return (0);
1774 }
1775 
1776 /*
1777  * Update disk state.
1778  */
1779 static int
1780 g_raid_update_disk(struct g_raid_disk *disk, u_int event)
1781 {
1782 	struct g_raid_softc *sc;
1783 
1784 	sc = disk->d_softc;
1785 	sx_assert(&sc->sc_lock, SX_XLOCKED);
1786 
1787 	G_RAID_DEBUG1(2, sc, "Event %s for disk %s.",
1788 	    g_raid_disk_event2str(event),
1789 	    g_raid_get_diskname(disk));
1790 
1791 	if (sc->sc_md)
1792 		G_RAID_MD_EVENT(sc->sc_md, disk, event);
1793 	return (0);
1794 }
1795 
1796 /*
1797  * Node event.
1798  */
1799 static int
1800 g_raid_update_node(struct g_raid_softc *sc, u_int event)
1801 {
1802 	sx_assert(&sc->sc_lock, SX_XLOCKED);
1803 
1804 	G_RAID_DEBUG1(2, sc, "Event %s for the array.",
1805 	    g_raid_node_event2str(event));
1806 
1807 	if (event == G_RAID_NODE_E_WAKE)
1808 		return (0);
1809 	if (sc->sc_md)
1810 		G_RAID_MD_EVENT(sc->sc_md, NULL, event);
1811 	return (0);
1812 }
1813 
1814 static int
1815 g_raid_access(struct g_provider *pp, int acr, int acw, int ace)
1816 {
1817 	struct g_raid_volume *vol;
1818 	struct g_raid_softc *sc;
1819 	int dcw, opens, error = 0;
1820 
1821 	g_topology_assert();
1822 	sc = pp->geom->softc;
1823 	vol = pp->private;
1824 	KASSERT(sc != NULL, ("NULL softc (provider=%s).", pp->name));
1825 	KASSERT(vol != NULL, ("NULL volume (provider=%s).", pp->name));
1826 
1827 	G_RAID_DEBUG1(2, sc, "Access request for %s: r%dw%de%d.", pp->name,
1828 	    acr, acw, ace);
1829 	dcw = pp->acw + acw;
1830 
1831 	g_topology_unlock();
1832 	sx_xlock(&sc->sc_lock);
1833 	/* Deny new opens while dying. */
1834 	if (sc->sc_stopping != 0 && (acr > 0 || acw > 0 || ace > 0)) {
1835 		error = ENXIO;
1836 		goto out;
1837 	}
1838 	/* Deny write opens for read-only volumes. */
1839 	if (vol->v_read_only && acw > 0) {
1840 		error = EROFS;
1841 		goto out;
1842 	}
1843 	if (dcw == 0)
1844 		g_raid_clean(vol, dcw);
1845 	vol->v_provider_open += acr + acw + ace;
1846 	/* Handle delayed node destruction. */
1847 	if (sc->sc_stopping == G_RAID_DESTROY_DELAYED &&
1848 	    vol->v_provider_open == 0) {
1849 		/* Count open volumes. */
1850 		opens = g_raid_nopens(sc);
1851 		if (opens == 0) {
1852 			sc->sc_stopping = G_RAID_DESTROY_HARD;
1853 			/* Wake up worker to make it selfdestruct. */
1854 			g_raid_event_send(sc, G_RAID_NODE_E_WAKE, 0);
1855 		}
1856 	}
1857 	/* Handle open volume destruction. */
1858 	if (vol->v_stopping && vol->v_provider_open == 0)
1859 		g_raid_destroy_volume(vol);
1860 out:
1861 	sx_xunlock(&sc->sc_lock);
1862 	g_topology_lock();
1863 	return (error);
1864 }
1865 
1866 struct g_raid_softc *
1867 g_raid_create_node(struct g_class *mp,
1868     const char *name, struct g_raid_md_object *md)
1869 {
1870 	struct g_raid_softc *sc;
1871 	struct g_geom *gp;
1872 	int error;
1873 
1874 	g_topology_assert();
1875 	G_RAID_DEBUG(1, "Creating array %s.", name);
1876 
1877 	gp = g_new_geomf(mp, "%s", name);
1878 	sc = malloc(sizeof(*sc), M_RAID, M_WAITOK | M_ZERO);
1879 	gp->start = g_raid_start;
1880 	gp->orphan = g_raid_orphan;
1881 	gp->access = g_raid_access;
1882 	gp->dumpconf = g_raid_dumpconf;
1883 
1884 	sc->sc_md = md;
1885 	sc->sc_geom = gp;
1886 	sc->sc_flags = 0;
1887 	TAILQ_INIT(&sc->sc_volumes);
1888 	TAILQ_INIT(&sc->sc_disks);
1889 	sx_init(&sc->sc_lock, "graid:lock");
1890 	mtx_init(&sc->sc_queue_mtx, "graid:queue", NULL, MTX_DEF);
1891 	TAILQ_INIT(&sc->sc_events);
1892 	bioq_init(&sc->sc_queue);
1893 	gp->softc = sc;
1894 	error = kproc_create(g_raid_worker, sc, &sc->sc_worker, 0, 0,
1895 	    "g_raid %s", name);
1896 	if (error != 0) {
1897 		G_RAID_DEBUG(0, "Cannot create kernel thread for %s.", name);
1898 		mtx_destroy(&sc->sc_queue_mtx);
1899 		sx_destroy(&sc->sc_lock);
1900 		g_destroy_geom(sc->sc_geom);
1901 		free(sc, M_RAID);
1902 		return (NULL);
1903 	}
1904 
1905 	G_RAID_DEBUG1(0, sc, "Array %s created.", name);
1906 	return (sc);
1907 }
1908 
1909 struct g_raid_volume *
1910 g_raid_create_volume(struct g_raid_softc *sc, const char *name, int id)
1911 {
1912 	struct g_raid_volume	*vol, *vol1;
1913 	int i;
1914 
1915 	G_RAID_DEBUG1(1, sc, "Creating volume %s.", name);
1916 	vol = malloc(sizeof(*vol), M_RAID, M_WAITOK | M_ZERO);
1917 	vol->v_softc = sc;
1918 	strlcpy(vol->v_name, name, G_RAID_MAX_VOLUMENAME);
1919 	vol->v_state = G_RAID_VOLUME_S_STARTING;
1920 	vol->v_raid_level = G_RAID_VOLUME_RL_UNKNOWN;
1921 	vol->v_raid_level_qualifier = G_RAID_VOLUME_RLQ_UNKNOWN;
1922 	vol->v_rotate_parity = 1;
1923 	bioq_init(&vol->v_inflight);
1924 	bioq_init(&vol->v_locked);
1925 	LIST_INIT(&vol->v_locks);
1926 	for (i = 0; i < G_RAID_MAX_SUBDISKS; i++) {
1927 		vol->v_subdisks[i].sd_softc = sc;
1928 		vol->v_subdisks[i].sd_volume = vol;
1929 		vol->v_subdisks[i].sd_pos = i;
1930 		vol->v_subdisks[i].sd_state = G_RAID_DISK_S_NONE;
1931 	}
1932 
1933 	/* Find free ID for this volume. */
1934 	g_topology_lock();
1935 	vol1 = vol;
1936 	if (id >= 0) {
1937 		LIST_FOREACH(vol1, &g_raid_volumes, v_global_next) {
1938 			if (vol1->v_global_id == id)
1939 				break;
1940 		}
1941 	}
1942 	if (vol1 != NULL) {
1943 		for (id = 0; ; id++) {
1944 			LIST_FOREACH(vol1, &g_raid_volumes, v_global_next) {
1945 				if (vol1->v_global_id == id)
1946 					break;
1947 			}
1948 			if (vol1 == NULL)
1949 				break;
1950 		}
1951 	}
1952 	vol->v_global_id = id;
1953 	LIST_INSERT_HEAD(&g_raid_volumes, vol, v_global_next);
1954 	g_topology_unlock();
1955 
1956 	/* Delay root mounting. */
1957 	vol->v_rootmount = root_mount_hold("GRAID");
1958 	G_RAID_DEBUG1(1, sc, "root_mount_hold %p", vol->v_rootmount);
1959 	vol->v_starting = 1;
1960 	TAILQ_INSERT_TAIL(&sc->sc_volumes, vol, v_next);
1961 	return (vol);
1962 }
1963 
1964 struct g_raid_disk *
1965 g_raid_create_disk(struct g_raid_softc *sc)
1966 {
1967 	struct g_raid_disk	*disk;
1968 
1969 	G_RAID_DEBUG1(1, sc, "Creating disk.");
1970 	disk = malloc(sizeof(*disk), M_RAID, M_WAITOK | M_ZERO);
1971 	disk->d_softc = sc;
1972 	disk->d_state = G_RAID_DISK_S_NONE;
1973 	TAILQ_INIT(&disk->d_subdisks);
1974 	TAILQ_INSERT_TAIL(&sc->sc_disks, disk, d_next);
1975 	return (disk);
1976 }
1977 
1978 int g_raid_start_volume(struct g_raid_volume *vol)
1979 {
1980 	struct g_raid_tr_class *class;
1981 	struct g_raid_tr_object *obj;
1982 	int status;
1983 
1984 	G_RAID_DEBUG1(2, vol->v_softc, "Starting volume %s.", vol->v_name);
1985 	LIST_FOREACH(class, &g_raid_tr_classes, trc_list) {
1986 		if (!class->trc_enable)
1987 			continue;
1988 		G_RAID_DEBUG1(2, vol->v_softc,
1989 		    "Tasting volume %s for %s transformation.",
1990 		    vol->v_name, class->name);
1991 		obj = (void *)kobj_create((kobj_class_t)class, M_RAID,
1992 		    M_WAITOK);
1993 		obj->tro_class = class;
1994 		obj->tro_volume = vol;
1995 		status = G_RAID_TR_TASTE(obj, vol);
1996 		if (status != G_RAID_TR_TASTE_FAIL)
1997 			break;
1998 		kobj_delete((kobj_t)obj, M_RAID);
1999 	}
2000 	if (class == NULL) {
2001 		G_RAID_DEBUG1(0, vol->v_softc,
2002 		    "No transformation module found for %s.",
2003 		    vol->v_name);
2004 		vol->v_tr = NULL;
2005 		g_raid_change_volume_state(vol, G_RAID_VOLUME_S_UNSUPPORTED);
2006 		g_raid_event_send(vol, G_RAID_VOLUME_E_DOWN,
2007 		    G_RAID_EVENT_VOLUME);
2008 		return (-1);
2009 	}
2010 	G_RAID_DEBUG1(2, vol->v_softc,
2011 	    "Transformation module %s chosen for %s.",
2012 	    class->name, vol->v_name);
2013 	vol->v_tr = obj;
2014 	return (0);
2015 }
2016 
2017 int
2018 g_raid_destroy_node(struct g_raid_softc *sc, int worker)
2019 {
2020 	struct g_raid_volume *vol, *tmpv;
2021 	struct g_raid_disk *disk, *tmpd;
2022 	int error = 0;
2023 
2024 	sc->sc_stopping = G_RAID_DESTROY_HARD;
2025 	TAILQ_FOREACH_SAFE(vol, &sc->sc_volumes, v_next, tmpv) {
2026 		if (g_raid_destroy_volume(vol))
2027 			error = EBUSY;
2028 	}
2029 	if (error)
2030 		return (error);
2031 	TAILQ_FOREACH_SAFE(disk, &sc->sc_disks, d_next, tmpd) {
2032 		if (g_raid_destroy_disk(disk))
2033 			error = EBUSY;
2034 	}
2035 	if (error)
2036 		return (error);
2037 	if (sc->sc_md) {
2038 		G_RAID_MD_FREE(sc->sc_md);
2039 		kobj_delete((kobj_t)sc->sc_md, M_RAID);
2040 		sc->sc_md = NULL;
2041 	}
2042 	if (sc->sc_geom != NULL) {
2043 		G_RAID_DEBUG1(0, sc, "Array %s destroyed.", sc->sc_name);
2044 		g_topology_lock();
2045 		sc->sc_geom->softc = NULL;
2046 		g_wither_geom(sc->sc_geom, ENXIO);
2047 		g_topology_unlock();
2048 		sc->sc_geom = NULL;
2049 	} else
2050 		G_RAID_DEBUG(1, "Array destroyed.");
2051 	if (worker) {
2052 		g_raid_event_cancel(sc, sc);
2053 		mtx_destroy(&sc->sc_queue_mtx);
2054 		sx_xunlock(&sc->sc_lock);
2055 		sx_destroy(&sc->sc_lock);
2056 		wakeup(&sc->sc_stopping);
2057 		free(sc, M_RAID);
2058 		curthread->td_pflags &= ~TDP_GEOM;
2059 		G_RAID_DEBUG(1, "Thread exiting.");
2060 		kproc_exit(0);
2061 	} else {
2062 		/* Wake up worker to make it selfdestruct. */
2063 		g_raid_event_send(sc, G_RAID_NODE_E_WAKE, 0);
2064 	}
2065 	return (0);
2066 }
2067 
2068 int
2069 g_raid_destroy_volume(struct g_raid_volume *vol)
2070 {
2071 	struct g_raid_softc *sc;
2072 	struct g_raid_disk *disk;
2073 	int i;
2074 
2075 	sc = vol->v_softc;
2076 	G_RAID_DEBUG1(2, sc, "Destroying volume %s.", vol->v_name);
2077 	vol->v_stopping = 1;
2078 	if (vol->v_state != G_RAID_VOLUME_S_STOPPED) {
2079 		if (vol->v_tr) {
2080 			G_RAID_TR_STOP(vol->v_tr);
2081 			return (EBUSY);
2082 		} else
2083 			vol->v_state = G_RAID_VOLUME_S_STOPPED;
2084 	}
2085 	if (g_raid_event_check(sc, vol) != 0)
2086 		return (EBUSY);
2087 	if (vol->v_provider != NULL)
2088 		return (EBUSY);
2089 	if (vol->v_provider_open != 0)
2090 		return (EBUSY);
2091 	if (vol->v_tr) {
2092 		G_RAID_TR_FREE(vol->v_tr);
2093 		kobj_delete((kobj_t)vol->v_tr, M_RAID);
2094 		vol->v_tr = NULL;
2095 	}
2096 	if (vol->v_rootmount)
2097 		root_mount_rel(vol->v_rootmount);
2098 	g_topology_lock();
2099 	LIST_REMOVE(vol, v_global_next);
2100 	g_topology_unlock();
2101 	TAILQ_REMOVE(&sc->sc_volumes, vol, v_next);
2102 	for (i = 0; i < G_RAID_MAX_SUBDISKS; i++) {
2103 		g_raid_event_cancel(sc, &vol->v_subdisks[i]);
2104 		disk = vol->v_subdisks[i].sd_disk;
2105 		if (disk == NULL)
2106 			continue;
2107 		TAILQ_REMOVE(&disk->d_subdisks, &vol->v_subdisks[i], sd_next);
2108 	}
2109 	G_RAID_DEBUG1(2, sc, "Volume %s destroyed.", vol->v_name);
2110 	if (sc->sc_md)
2111 		G_RAID_MD_FREE_VOLUME(sc->sc_md, vol);
2112 	g_raid_event_cancel(sc, vol);
2113 	free(vol, M_RAID);
2114 	if (sc->sc_stopping == G_RAID_DESTROY_HARD) {
2115 		/* Wake up worker to let it selfdestruct. */
2116 		g_raid_event_send(sc, G_RAID_NODE_E_WAKE, 0);
2117 	}
2118 	return (0);
2119 }
2120 
2121 int
2122 g_raid_destroy_disk(struct g_raid_disk *disk)
2123 {
2124 	struct g_raid_softc *sc;
2125 	struct g_raid_subdisk *sd, *tmp;
2126 
2127 	sc = disk->d_softc;
2128 	G_RAID_DEBUG1(2, sc, "Destroying disk.");
2129 	if (disk->d_consumer) {
2130 		g_raid_kill_consumer(sc, disk->d_consumer);
2131 		disk->d_consumer = NULL;
2132 	}
2133 	TAILQ_FOREACH_SAFE(sd, &disk->d_subdisks, sd_next, tmp) {
2134 		g_raid_change_subdisk_state(sd, G_RAID_SUBDISK_S_NONE);
2135 		g_raid_event_send(sd, G_RAID_SUBDISK_E_DISCONNECTED,
2136 		    G_RAID_EVENT_SUBDISK);
2137 		TAILQ_REMOVE(&disk->d_subdisks, sd, sd_next);
2138 		sd->sd_disk = NULL;
2139 	}
2140 	TAILQ_REMOVE(&sc->sc_disks, disk, d_next);
2141 	if (sc->sc_md)
2142 		G_RAID_MD_FREE_DISK(sc->sc_md, disk);
2143 	g_raid_event_cancel(sc, disk);
2144 	free(disk, M_RAID);
2145 	return (0);
2146 }
2147 
2148 int
2149 g_raid_destroy(struct g_raid_softc *sc, int how)
2150 {
2151 	int error, opens;
2152 
2153 	g_topology_assert_not();
2154 	if (sc == NULL)
2155 		return (ENXIO);
2156 	sx_assert(&sc->sc_lock, SX_XLOCKED);
2157 
2158 	/* Count open volumes. */
2159 	opens = g_raid_nopens(sc);
2160 
2161 	/* React on some opened volumes. */
2162 	if (opens > 0) {
2163 		switch (how) {
2164 		case G_RAID_DESTROY_SOFT:
2165 			G_RAID_DEBUG1(1, sc,
2166 			    "%d volumes are still open.",
2167 			    opens);
2168 			sx_xunlock(&sc->sc_lock);
2169 			return (EBUSY);
2170 		case G_RAID_DESTROY_DELAYED:
2171 			G_RAID_DEBUG1(1, sc,
2172 			    "Array will be destroyed on last close.");
2173 			sc->sc_stopping = G_RAID_DESTROY_DELAYED;
2174 			sx_xunlock(&sc->sc_lock);
2175 			return (EBUSY);
2176 		case G_RAID_DESTROY_HARD:
2177 			G_RAID_DEBUG1(1, sc,
2178 			    "%d volumes are still open.",
2179 			    opens);
2180 		}
2181 	}
2182 
2183 	/* Mark node for destruction. */
2184 	sc->sc_stopping = G_RAID_DESTROY_HARD;
2185 	/* Wake up worker to let it selfdestruct. */
2186 	g_raid_event_send(sc, G_RAID_NODE_E_WAKE, 0);
2187 	/* Sleep until node destroyed. */
2188 	error = sx_sleep(&sc->sc_stopping, &sc->sc_lock,
2189 	    PRIBIO | PDROP, "r:destroy", hz * 3);
2190 	return (error == EWOULDBLOCK ? EBUSY : 0);
2191 }
2192 
2193 static void
2194 g_raid_taste_orphan(struct g_consumer *cp)
2195 {
2196 
2197 	KASSERT(1 == 0, ("%s called while tasting %s.", __func__,
2198 	    cp->provider->name));
2199 }
2200 
2201 static struct g_geom *
2202 g_raid_taste(struct g_class *mp, struct g_provider *pp, int flags __unused)
2203 {
2204 	struct g_consumer *cp;
2205 	struct g_geom *gp, *geom;
2206 	struct g_raid_md_class *class;
2207 	struct g_raid_md_object *obj;
2208 	int status;
2209 
2210 	g_topology_assert();
2211 	g_trace(G_T_TOPOLOGY, "%s(%s, %s)", __func__, mp->name, pp->name);
2212 	if (!g_raid_enable)
2213 		return (NULL);
2214 	G_RAID_DEBUG(2, "Tasting provider %s.", pp->name);
2215 
2216 	geom = NULL;
2217 	status = G_RAID_MD_TASTE_FAIL;
2218 	gp = g_new_geomf(mp, "raid:taste");
2219 	/*
2220 	 * This orphan function should be never called.
2221 	 */
2222 	gp->orphan = g_raid_taste_orphan;
2223 	cp = g_new_consumer(gp);
2224 	cp->flags |= G_CF_DIRECT_SEND | G_CF_DIRECT_RECEIVE;
2225 	if (g_attach(cp, pp) != 0)
2226 		goto ofail2;
2227 	if (g_access(cp, 1, 0, 0) != 0)
2228 		goto ofail;
2229 
2230 	LIST_FOREACH(class, &g_raid_md_classes, mdc_list) {
2231 		if (!class->mdc_enable)
2232 			continue;
2233 		G_RAID_DEBUG(2, "Tasting provider %s for %s metadata.",
2234 		    pp->name, class->name);
2235 		obj = (void *)kobj_create((kobj_class_t)class, M_RAID,
2236 		    M_WAITOK);
2237 		obj->mdo_class = class;
2238 		status = G_RAID_MD_TASTE(obj, mp, cp, &geom);
2239 		if (status != G_RAID_MD_TASTE_NEW)
2240 			kobj_delete((kobj_t)obj, M_RAID);
2241 		if (status != G_RAID_MD_TASTE_FAIL)
2242 			break;
2243 	}
2244 
2245 	if (status == G_RAID_MD_TASTE_FAIL)
2246 		(void)g_access(cp, -1, 0, 0);
2247 ofail:
2248 	g_detach(cp);
2249 ofail2:
2250 	g_destroy_consumer(cp);
2251 	g_destroy_geom(gp);
2252 	G_RAID_DEBUG(2, "Tasting provider %s done.", pp->name);
2253 	return (geom);
2254 }
2255 
2256 int
2257 g_raid_create_node_format(const char *format, struct gctl_req *req,
2258     struct g_geom **gp)
2259 {
2260 	struct g_raid_md_class *class;
2261 	struct g_raid_md_object *obj;
2262 	int status;
2263 
2264 	G_RAID_DEBUG(2, "Creating array for %s metadata.", format);
2265 	LIST_FOREACH(class, &g_raid_md_classes, mdc_list) {
2266 		if (strcasecmp(class->name, format) == 0)
2267 			break;
2268 	}
2269 	if (class == NULL) {
2270 		G_RAID_DEBUG(1, "No support for %s metadata.", format);
2271 		return (G_RAID_MD_TASTE_FAIL);
2272 	}
2273 	obj = (void *)kobj_create((kobj_class_t)class, M_RAID,
2274 	    M_WAITOK);
2275 	obj->mdo_class = class;
2276 	status = G_RAID_MD_CREATE_REQ(obj, &g_raid_class, req, gp);
2277 	if (status != G_RAID_MD_TASTE_NEW)
2278 		kobj_delete((kobj_t)obj, M_RAID);
2279 	return (status);
2280 }
2281 
2282 static int
2283 g_raid_destroy_geom(struct gctl_req *req __unused,
2284     struct g_class *mp __unused, struct g_geom *gp)
2285 {
2286 	struct g_raid_softc *sc;
2287 	int error;
2288 
2289 	g_topology_unlock();
2290 	sc = gp->softc;
2291 	sx_xlock(&sc->sc_lock);
2292 	g_cancel_event(sc);
2293 	error = g_raid_destroy(gp->softc, G_RAID_DESTROY_SOFT);
2294 	g_topology_lock();
2295 	return (error);
2296 }
2297 
2298 void g_raid_write_metadata(struct g_raid_softc *sc, struct g_raid_volume *vol,
2299     struct g_raid_subdisk *sd, struct g_raid_disk *disk)
2300 {
2301 
2302 	if (sc->sc_stopping == G_RAID_DESTROY_HARD)
2303 		return;
2304 	if (sc->sc_md)
2305 		G_RAID_MD_WRITE(sc->sc_md, vol, sd, disk);
2306 }
2307 
2308 void g_raid_fail_disk(struct g_raid_softc *sc,
2309     struct g_raid_subdisk *sd, struct g_raid_disk *disk)
2310 {
2311 
2312 	if (disk == NULL)
2313 		disk = sd->sd_disk;
2314 	if (disk == NULL) {
2315 		G_RAID_DEBUG1(0, sc, "Warning! Fail request to an absent disk!");
2316 		return;
2317 	}
2318 	if (disk->d_state != G_RAID_DISK_S_ACTIVE) {
2319 		G_RAID_DEBUG1(0, sc, "Warning! Fail request to a disk in a "
2320 		    "wrong state (%s)!", g_raid_disk_state2str(disk->d_state));
2321 		return;
2322 	}
2323 	if (sc->sc_md)
2324 		G_RAID_MD_FAIL_DISK(sc->sc_md, sd, disk);
2325 }
2326 
2327 static void
2328 g_raid_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp,
2329     struct g_consumer *cp, struct g_provider *pp)
2330 {
2331 	struct g_raid_softc *sc;
2332 	struct g_raid_volume *vol;
2333 	struct g_raid_subdisk *sd;
2334 	struct g_raid_disk *disk;
2335 	int i, s;
2336 
2337 	g_topology_assert();
2338 
2339 	sc = gp->softc;
2340 	if (sc == NULL)
2341 		return;
2342 	if (pp != NULL) {
2343 		vol = pp->private;
2344 		g_topology_unlock();
2345 		sx_xlock(&sc->sc_lock);
2346 		sbuf_printf(sb, "%s<descr>%s %s volume</descr>\n", indent,
2347 		    sc->sc_md->mdo_class->name,
2348 		    g_raid_volume_level2str(vol->v_raid_level,
2349 		    vol->v_raid_level_qualifier));
2350 		sbuf_printf(sb, "%s<Label>%s</Label>\n", indent,
2351 		    vol->v_name);
2352 		sbuf_printf(sb, "%s<RAIDLevel>%s</RAIDLevel>\n", indent,
2353 		    g_raid_volume_level2str(vol->v_raid_level,
2354 		    vol->v_raid_level_qualifier));
2355 		sbuf_printf(sb,
2356 		    "%s<Transformation>%s</Transformation>\n", indent,
2357 		    vol->v_tr ? vol->v_tr->tro_class->name : "NONE");
2358 		sbuf_printf(sb, "%s<Components>%u</Components>\n", indent,
2359 		    vol->v_disks_count);
2360 		sbuf_printf(sb, "%s<Strip>%u</Strip>\n", indent,
2361 		    vol->v_strip_size);
2362 		sbuf_printf(sb, "%s<State>%s</State>\n", indent,
2363 		    g_raid_volume_state2str(vol->v_state));
2364 		sbuf_printf(sb, "%s<Dirty>%s</Dirty>\n", indent,
2365 		    vol->v_dirty ? "Yes" : "No");
2366 		sbuf_printf(sb, "%s<Subdisks>", indent);
2367 		for (i = 0; i < vol->v_disks_count; i++) {
2368 			sd = &vol->v_subdisks[i];
2369 			if (sd->sd_disk != NULL &&
2370 			    sd->sd_disk->d_consumer != NULL) {
2371 				sbuf_printf(sb, "%s ",
2372 				    g_raid_get_diskname(sd->sd_disk));
2373 			} else {
2374 				sbuf_cat(sb, "NONE ");
2375 			}
2376 			sbuf_printf(sb, "(%s",
2377 			    g_raid_subdisk_state2str(sd->sd_state));
2378 			if (sd->sd_state == G_RAID_SUBDISK_S_REBUILD ||
2379 			    sd->sd_state == G_RAID_SUBDISK_S_RESYNC) {
2380 				sbuf_printf(sb, " %d%%",
2381 				    (int)(sd->sd_rebuild_pos * 100 /
2382 				     sd->sd_size));
2383 			}
2384 			sbuf_cat(sb, ")");
2385 			if (i + 1 < vol->v_disks_count)
2386 				sbuf_cat(sb, ", ");
2387 		}
2388 		sbuf_cat(sb, "</Subdisks>\n");
2389 		sx_xunlock(&sc->sc_lock);
2390 		g_topology_lock();
2391 	} else if (cp != NULL) {
2392 		disk = cp->private;
2393 		if (disk == NULL)
2394 			return;
2395 		g_topology_unlock();
2396 		sx_xlock(&sc->sc_lock);
2397 		sbuf_printf(sb, "%s<State>%s", indent,
2398 		    g_raid_disk_state2str(disk->d_state));
2399 		if (!TAILQ_EMPTY(&disk->d_subdisks)) {
2400 			sbuf_cat(sb, " (");
2401 			TAILQ_FOREACH(sd, &disk->d_subdisks, sd_next) {
2402 				sbuf_printf(sb, "%s",
2403 				    g_raid_subdisk_state2str(sd->sd_state));
2404 				if (sd->sd_state == G_RAID_SUBDISK_S_REBUILD ||
2405 				    sd->sd_state == G_RAID_SUBDISK_S_RESYNC) {
2406 					sbuf_printf(sb, " %d%%",
2407 					    (int)(sd->sd_rebuild_pos * 100 /
2408 					     sd->sd_size));
2409 				}
2410 				if (TAILQ_NEXT(sd, sd_next))
2411 					sbuf_cat(sb, ", ");
2412 			}
2413 			sbuf_cat(sb, ")");
2414 		}
2415 		sbuf_cat(sb, "</State>\n");
2416 		sbuf_printf(sb, "%s<Subdisks>", indent);
2417 		TAILQ_FOREACH(sd, &disk->d_subdisks, sd_next) {
2418 			sbuf_printf(sb, "r%d(%s):%d@%ju",
2419 			    sd->sd_volume->v_global_id,
2420 			    sd->sd_volume->v_name,
2421 			    sd->sd_pos, (uintmax_t)sd->sd_offset);
2422 			if (TAILQ_NEXT(sd, sd_next))
2423 				sbuf_cat(sb, ", ");
2424 		}
2425 		sbuf_cat(sb, "</Subdisks>\n");
2426 		sbuf_printf(sb, "%s<ReadErrors>%d</ReadErrors>\n", indent,
2427 		    disk->d_read_errs);
2428 		sx_xunlock(&sc->sc_lock);
2429 		g_topology_lock();
2430 	} else {
2431 		g_topology_unlock();
2432 		sx_xlock(&sc->sc_lock);
2433 		if (sc->sc_md) {
2434 			sbuf_printf(sb, "%s<Metadata>%s</Metadata>\n", indent,
2435 			    sc->sc_md->mdo_class->name);
2436 		}
2437 		if (!TAILQ_EMPTY(&sc->sc_volumes)) {
2438 			s = 0xff;
2439 			TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) {
2440 				if (vol->v_state < s)
2441 					s = vol->v_state;
2442 			}
2443 			sbuf_printf(sb, "%s<State>%s</State>\n", indent,
2444 			    g_raid_volume_state2str(s));
2445 		}
2446 		sx_xunlock(&sc->sc_lock);
2447 		g_topology_lock();
2448 	}
2449 }
2450 
2451 static void
2452 g_raid_shutdown_post_sync(void *arg, int howto)
2453 {
2454 	struct g_class *mp;
2455 	struct g_geom *gp, *gp2;
2456 	struct g_raid_softc *sc;
2457 	struct g_raid_volume *vol;
2458 
2459 	mp = arg;
2460 	g_topology_lock();
2461 	g_raid_shutdown = 1;
2462 	LIST_FOREACH_SAFE(gp, &mp->geom, geom, gp2) {
2463 		if ((sc = gp->softc) == NULL)
2464 			continue;
2465 		g_topology_unlock();
2466 		sx_xlock(&sc->sc_lock);
2467 		TAILQ_FOREACH(vol, &sc->sc_volumes, v_next)
2468 			g_raid_clean(vol, -1);
2469 		g_cancel_event(sc);
2470 		g_raid_destroy(sc, G_RAID_DESTROY_DELAYED);
2471 		g_topology_lock();
2472 	}
2473 	g_topology_unlock();
2474 }
2475 
2476 static void
2477 g_raid_init(struct g_class *mp)
2478 {
2479 
2480 	g_raid_post_sync = EVENTHANDLER_REGISTER(shutdown_post_sync,
2481 	    g_raid_shutdown_post_sync, mp, SHUTDOWN_PRI_FIRST);
2482 	if (g_raid_post_sync == NULL)
2483 		G_RAID_DEBUG(0, "Warning! Cannot register shutdown event.");
2484 	g_raid_started = 1;
2485 }
2486 
2487 static void
2488 g_raid_fini(struct g_class *mp)
2489 {
2490 
2491 	if (g_raid_post_sync != NULL)
2492 		EVENTHANDLER_DEREGISTER(shutdown_post_sync, g_raid_post_sync);
2493 	g_raid_started = 0;
2494 }
2495 
2496 int
2497 g_raid_md_modevent(module_t mod, int type, void *arg)
2498 {
2499 	struct g_raid_md_class *class, *c, *nc;
2500 	int error;
2501 
2502 	error = 0;
2503 	class = arg;
2504 	switch (type) {
2505 	case MOD_LOAD:
2506 		c = LIST_FIRST(&g_raid_md_classes);
2507 		if (c == NULL || c->mdc_priority > class->mdc_priority)
2508 			LIST_INSERT_HEAD(&g_raid_md_classes, class, mdc_list);
2509 		else {
2510 			while ((nc = LIST_NEXT(c, mdc_list)) != NULL &&
2511 			    nc->mdc_priority < class->mdc_priority)
2512 				c = nc;
2513 			LIST_INSERT_AFTER(c, class, mdc_list);
2514 		}
2515 		if (g_raid_started)
2516 			g_retaste(&g_raid_class);
2517 		break;
2518 	case MOD_UNLOAD:
2519 		LIST_REMOVE(class, mdc_list);
2520 		break;
2521 	default:
2522 		error = EOPNOTSUPP;
2523 		break;
2524 	}
2525 
2526 	return (error);
2527 }
2528 
2529 int
2530 g_raid_tr_modevent(module_t mod, int type, void *arg)
2531 {
2532 	struct g_raid_tr_class *class, *c, *nc;
2533 	int error;
2534 
2535 	error = 0;
2536 	class = arg;
2537 	switch (type) {
2538 	case MOD_LOAD:
2539 		c = LIST_FIRST(&g_raid_tr_classes);
2540 		if (c == NULL || c->trc_priority > class->trc_priority)
2541 			LIST_INSERT_HEAD(&g_raid_tr_classes, class, trc_list);
2542 		else {
2543 			while ((nc = LIST_NEXT(c, trc_list)) != NULL &&
2544 			    nc->trc_priority < class->trc_priority)
2545 				c = nc;
2546 			LIST_INSERT_AFTER(c, class, trc_list);
2547 		}
2548 		break;
2549 	case MOD_UNLOAD:
2550 		LIST_REMOVE(class, trc_list);
2551 		break;
2552 	default:
2553 		error = EOPNOTSUPP;
2554 		break;
2555 	}
2556 
2557 	return (error);
2558 }
2559 
2560 /*
2561  * Use local implementation of DECLARE_GEOM_CLASS(g_raid_class, g_raid)
2562  * to reduce module priority, allowing submodules to register them first.
2563  */
2564 static moduledata_t g_raid_mod = {
2565 	"g_raid",
2566 	g_modevent,
2567 	&g_raid_class
2568 };
2569 DECLARE_MODULE(g_raid, g_raid_mod, SI_SUB_DRIVERS, SI_ORDER_THIRD);
2570 MODULE_VERSION(geom_raid, 0);
2571