xref: /freebsd/sys/geom/raid/g_raid.c (revision 2b06b2013c82ee7f744ac5b6a413edede3eeb8cd)
1 /*-
2  * Copyright (c) 2010 Alexander Motin <mav@FreeBSD.org>
3  * All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  *
14  * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND
15  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE
18  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
24  * SUCH DAMAGE.
25  */
26 
27 #include <sys/cdefs.h>
28 __FBSDID("$FreeBSD$");
29 
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 <sys/proc.h>
45 #include <sys/kthread.h>
46 #include <sys/sched.h>
47 #include <geom/raid/g_raid.h>
48 #include "g_raid_md_if.h"
49 #include "g_raid_tr_if.h"
50 
51 static MALLOC_DEFINE(M_RAID, "raid_data", "GEOM_RAID Data");
52 
53 SYSCTL_DECL(_kern_geom);
54 SYSCTL_NODE(_kern_geom, OID_AUTO, raid, CTLFLAG_RW, 0, "GEOM_RAID stuff");
55 int g_raid_enable = 1;
56 TUNABLE_INT("kern.geom.raid.enable", &g_raid_enable);
57 SYSCTL_INT(_kern_geom_raid, OID_AUTO, enable, CTLFLAG_RW,
58     &g_raid_enable, 0, "Enable on-disk metadata taste");
59 u_int g_raid_aggressive_spare = 0;
60 TUNABLE_INT("kern.geom.raid.aggressive_spare", &g_raid_aggressive_spare);
61 SYSCTL_UINT(_kern_geom_raid, OID_AUTO, aggressive_spare, CTLFLAG_RW,
62     &g_raid_aggressive_spare, 0, "Use disks without metadata as spare");
63 u_int g_raid_debug = 0;
64 TUNABLE_INT("kern.geom.raid.debug", &g_raid_debug);
65 SYSCTL_UINT(_kern_geom_raid, OID_AUTO, debug, CTLFLAG_RW, &g_raid_debug, 0,
66     "Debug level");
67 int g_raid_read_err_thresh = 10;
68 TUNABLE_INT("kern.geom.raid.read_err_thresh", &g_raid_read_err_thresh);
69 SYSCTL_UINT(_kern_geom_raid, OID_AUTO, read_err_thresh, CTLFLAG_RW,
70     &g_raid_read_err_thresh, 0,
71     "Number of read errors equated to disk failure");
72 u_int g_raid_start_timeout = 30;
73 TUNABLE_INT("kern.geom.raid.start_timeout", &g_raid_start_timeout);
74 SYSCTL_UINT(_kern_geom_raid, OID_AUTO, start_timeout, CTLFLAG_RW,
75     &g_raid_start_timeout, 0,
76     "Time to wait for all array components");
77 static u_int g_raid_clean_time = 5;
78 TUNABLE_INT("kern.geom.raid.clean_time", &g_raid_clean_time);
79 SYSCTL_UINT(_kern_geom_raid, OID_AUTO, clean_time, CTLFLAG_RW,
80     &g_raid_clean_time, 0, "Mark volume as clean when idling");
81 static u_int g_raid_disconnect_on_failure = 1;
82 TUNABLE_INT("kern.geom.raid.disconnect_on_failure",
83     &g_raid_disconnect_on_failure);
84 SYSCTL_UINT(_kern_geom_raid, OID_AUTO, disconnect_on_failure, CTLFLAG_RW,
85     &g_raid_disconnect_on_failure, 0, "Disconnect component on I/O failure.");
86 static u_int g_raid_name_format = 0;
87 TUNABLE_INT("kern.geom.raid.name_format", &g_raid_name_format);
88 SYSCTL_UINT(_kern_geom_raid, OID_AUTO, name_format, CTLFLAG_RW,
89     &g_raid_name_format, 0, "Providers name format.");
90 static u_int g_raid_idle_threshold = 1000000;
91 TUNABLE_INT("kern.geom.raid.idle_threshold", &g_raid_idle_threshold);
92 SYSCTL_UINT(_kern_geom_raid, OID_AUTO, idle_threshold, CTLFLAG_RW,
93     &g_raid_idle_threshold, 1000000,
94     "Time in microseconds to consider a volume idle.");
95 static u_int ar_legacy_aliases = 1;
96 SYSCTL_INT(_kern_geom_raid, OID_AUTO, legacy_aliases, CTLFLAG_RW,
97            &ar_legacy_aliases, 0, "Create aliases named as the legacy ataraid style.");
98 TUNABLE_INT("kern.geom_raid.legacy_aliases", &ar_legacy_aliases);
99 
100 
101 #define	MSLEEP(rv, ident, mtx, priority, wmesg, timeout)	do {	\
102 	G_RAID_DEBUG(4, "%s: Sleeping %p.", __func__, (ident));		\
103 	rv = msleep((ident), (mtx), (priority), (wmesg), (timeout));	\
104 	G_RAID_DEBUG(4, "%s: Woken up %p.", __func__, (ident));		\
105 } while (0)
106 
107 LIST_HEAD(, g_raid_md_class) g_raid_md_classes =
108     LIST_HEAD_INITIALIZER(g_raid_md_classes);
109 
110 LIST_HEAD(, g_raid_tr_class) g_raid_tr_classes =
111     LIST_HEAD_INITIALIZER(g_raid_tr_classes);
112 
113 LIST_HEAD(, g_raid_volume) g_raid_volumes =
114     LIST_HEAD_INITIALIZER(g_raid_volumes);
115 
116 static eventhandler_tag g_raid_post_sync = NULL;
117 static int g_raid_started = 0;
118 static int g_raid_shutdown = 0;
119 
120 static int g_raid_destroy_geom(struct gctl_req *req, struct g_class *mp,
121     struct g_geom *gp);
122 static g_taste_t g_raid_taste;
123 static void g_raid_init(struct g_class *mp);
124 static void g_raid_fini(struct g_class *mp);
125 
126 struct g_class g_raid_class = {
127 	.name = G_RAID_CLASS_NAME,
128 	.version = G_VERSION,
129 	.ctlreq = g_raid_ctl,
130 	.taste = g_raid_taste,
131 	.destroy_geom = g_raid_destroy_geom,
132 	.init = g_raid_init,
133 	.fini = g_raid_fini
134 };
135 
136 static void g_raid_destroy_provider(struct g_raid_volume *vol);
137 static int g_raid_update_disk(struct g_raid_disk *disk, u_int event);
138 static int g_raid_update_subdisk(struct g_raid_subdisk *subdisk, u_int event);
139 static int g_raid_update_volume(struct g_raid_volume *vol, u_int event);
140 static int g_raid_update_node(struct g_raid_softc *sc, u_int event);
141 static void g_raid_dumpconf(struct sbuf *sb, const char *indent,
142     struct g_geom *gp, struct g_consumer *cp, struct g_provider *pp);
143 static void g_raid_start(struct bio *bp);
144 static void g_raid_start_request(struct bio *bp);
145 static void g_raid_disk_done(struct bio *bp);
146 static void g_raid_poll(struct g_raid_softc *sc);
147 
148 static const char *
149 g_raid_node_event2str(int event)
150 {
151 
152 	switch (event) {
153 	case G_RAID_NODE_E_WAKE:
154 		return ("WAKE");
155 	case G_RAID_NODE_E_START:
156 		return ("START");
157 	default:
158 		return ("INVALID");
159 	}
160 }
161 
162 const char *
163 g_raid_disk_state2str(int state)
164 {
165 
166 	switch (state) {
167 	case G_RAID_DISK_S_NONE:
168 		return ("NONE");
169 	case G_RAID_DISK_S_OFFLINE:
170 		return ("OFFLINE");
171 	case G_RAID_DISK_S_DISABLED:
172 		return ("DISABLED");
173 	case G_RAID_DISK_S_FAILED:
174 		return ("FAILED");
175 	case G_RAID_DISK_S_STALE_FAILED:
176 		return ("STALE_FAILED");
177 	case G_RAID_DISK_S_SPARE:
178 		return ("SPARE");
179 	case G_RAID_DISK_S_STALE:
180 		return ("STALE");
181 	case G_RAID_DISK_S_ACTIVE:
182 		return ("ACTIVE");
183 	default:
184 		return ("INVALID");
185 	}
186 }
187 
188 static const char *
189 g_raid_disk_event2str(int event)
190 {
191 
192 	switch (event) {
193 	case G_RAID_DISK_E_DISCONNECTED:
194 		return ("DISCONNECTED");
195 	default:
196 		return ("INVALID");
197 	}
198 }
199 
200 const char *
201 g_raid_subdisk_state2str(int state)
202 {
203 
204 	switch (state) {
205 	case G_RAID_SUBDISK_S_NONE:
206 		return ("NONE");
207 	case G_RAID_SUBDISK_S_FAILED:
208 		return ("FAILED");
209 	case G_RAID_SUBDISK_S_NEW:
210 		return ("NEW");
211 	case G_RAID_SUBDISK_S_REBUILD:
212 		return ("REBUILD");
213 	case G_RAID_SUBDISK_S_UNINITIALIZED:
214 		return ("UNINITIALIZED");
215 	case G_RAID_SUBDISK_S_STALE:
216 		return ("STALE");
217 	case G_RAID_SUBDISK_S_RESYNC:
218 		return ("RESYNC");
219 	case G_RAID_SUBDISK_S_ACTIVE:
220 		return ("ACTIVE");
221 	default:
222 		return ("INVALID");
223 	}
224 }
225 
226 static const char *
227 g_raid_subdisk_event2str(int event)
228 {
229 
230 	switch (event) {
231 	case G_RAID_SUBDISK_E_NEW:
232 		return ("NEW");
233 	case G_RAID_SUBDISK_E_FAILED:
234 		return ("FAILED");
235 	case G_RAID_SUBDISK_E_DISCONNECTED:
236 		return ("DISCONNECTED");
237 	default:
238 		return ("INVALID");
239 	}
240 }
241 
242 const char *
243 g_raid_volume_state2str(int state)
244 {
245 
246 	switch (state) {
247 	case G_RAID_VOLUME_S_STARTING:
248 		return ("STARTING");
249 	case G_RAID_VOLUME_S_BROKEN:
250 		return ("BROKEN");
251 	case G_RAID_VOLUME_S_DEGRADED:
252 		return ("DEGRADED");
253 	case G_RAID_VOLUME_S_SUBOPTIMAL:
254 		return ("SUBOPTIMAL");
255 	case G_RAID_VOLUME_S_OPTIMAL:
256 		return ("OPTIMAL");
257 	case G_RAID_VOLUME_S_UNSUPPORTED:
258 		return ("UNSUPPORTED");
259 	case G_RAID_VOLUME_S_STOPPED:
260 		return ("STOPPED");
261 	default:
262 		return ("INVALID");
263 	}
264 }
265 
266 static const char *
267 g_raid_volume_event2str(int event)
268 {
269 
270 	switch (event) {
271 	case G_RAID_VOLUME_E_UP:
272 		return ("UP");
273 	case G_RAID_VOLUME_E_DOWN:
274 		return ("DOWN");
275 	case G_RAID_VOLUME_E_START:
276 		return ("START");
277 	case G_RAID_VOLUME_E_STARTMD:
278 		return ("STARTMD");
279 	default:
280 		return ("INVALID");
281 	}
282 }
283 
284 const char *
285 g_raid_volume_level2str(int level, int qual)
286 {
287 
288 	switch (level) {
289 	case G_RAID_VOLUME_RL_RAID0:
290 		return ("RAID0");
291 	case G_RAID_VOLUME_RL_RAID1:
292 		return ("RAID1");
293 	case G_RAID_VOLUME_RL_RAID3:
294 		if (qual == G_RAID_VOLUME_RLQ_R3P0)
295 			return ("RAID3-P0");
296 		if (qual == G_RAID_VOLUME_RLQ_R3PN)
297 			return ("RAID3-PN");
298 		return ("RAID3");
299 	case G_RAID_VOLUME_RL_RAID4:
300 		if (qual == G_RAID_VOLUME_RLQ_R4P0)
301 			return ("RAID4-P0");
302 		if (qual == G_RAID_VOLUME_RLQ_R4PN)
303 			return ("RAID4-PN");
304 		return ("RAID4");
305 	case G_RAID_VOLUME_RL_RAID5:
306 		if (qual == G_RAID_VOLUME_RLQ_R5RA)
307 			return ("RAID5-RA");
308 		if (qual == G_RAID_VOLUME_RLQ_R5RS)
309 			return ("RAID5-RS");
310 		if (qual == G_RAID_VOLUME_RLQ_R5LA)
311 			return ("RAID5-LA");
312 		if (qual == G_RAID_VOLUME_RLQ_R5LS)
313 			return ("RAID5-LS");
314 		return ("RAID5");
315 	case G_RAID_VOLUME_RL_RAID6:
316 		if (qual == G_RAID_VOLUME_RLQ_R6RA)
317 			return ("RAID6-RA");
318 		if (qual == G_RAID_VOLUME_RLQ_R6RS)
319 			return ("RAID6-RS");
320 		if (qual == G_RAID_VOLUME_RLQ_R6LA)
321 			return ("RAID6-LA");
322 		if (qual == G_RAID_VOLUME_RLQ_R6LS)
323 			return ("RAID6-LS");
324 		return ("RAID6");
325 	case G_RAID_VOLUME_RL_RAIDMDF:
326 		if (qual == G_RAID_VOLUME_RLQ_RMDFRA)
327 			return ("RAIDMDF-RA");
328 		if (qual == G_RAID_VOLUME_RLQ_RMDFRS)
329 			return ("RAIDMDF-RS");
330 		if (qual == G_RAID_VOLUME_RLQ_RMDFLA)
331 			return ("RAIDMDF-LA");
332 		if (qual == G_RAID_VOLUME_RLQ_RMDFLS)
333 			return ("RAIDMDF-LS");
334 		return ("RAIDMDF");
335 	case G_RAID_VOLUME_RL_RAID1E:
336 		if (qual == G_RAID_VOLUME_RLQ_R1EA)
337 			return ("RAID1E-A");
338 		if (qual == G_RAID_VOLUME_RLQ_R1EO)
339 			return ("RAID1E-O");
340 		return ("RAID1E");
341 	case G_RAID_VOLUME_RL_SINGLE:
342 		return ("SINGLE");
343 	case G_RAID_VOLUME_RL_CONCAT:
344 		return ("CONCAT");
345 	case G_RAID_VOLUME_RL_RAID5E:
346 		if (qual == G_RAID_VOLUME_RLQ_R5ERA)
347 			return ("RAID5E-RA");
348 		if (qual == G_RAID_VOLUME_RLQ_R5ERS)
349 			return ("RAID5E-RS");
350 		if (qual == G_RAID_VOLUME_RLQ_R5ELA)
351 			return ("RAID5E-LA");
352 		if (qual == G_RAID_VOLUME_RLQ_R5ELS)
353 			return ("RAID5E-LS");
354 		return ("RAID5E");
355 	case G_RAID_VOLUME_RL_RAID5EE:
356 		if (qual == G_RAID_VOLUME_RLQ_R5EERA)
357 			return ("RAID5EE-RA");
358 		if (qual == G_RAID_VOLUME_RLQ_R5EERS)
359 			return ("RAID5EE-RS");
360 		if (qual == G_RAID_VOLUME_RLQ_R5EELA)
361 			return ("RAID5EE-LA");
362 		if (qual == G_RAID_VOLUME_RLQ_R5EELS)
363 			return ("RAID5EE-LS");
364 		return ("RAID5EE");
365 	case G_RAID_VOLUME_RL_RAID5R:
366 		if (qual == G_RAID_VOLUME_RLQ_R5RRA)
367 			return ("RAID5R-RA");
368 		if (qual == G_RAID_VOLUME_RLQ_R5RRS)
369 			return ("RAID5R-RS");
370 		if (qual == G_RAID_VOLUME_RLQ_R5RLA)
371 			return ("RAID5R-LA");
372 		if (qual == G_RAID_VOLUME_RLQ_R5RLS)
373 			return ("RAID5R-LS");
374 		return ("RAID5E");
375 	default:
376 		return ("UNKNOWN");
377 	}
378 }
379 
380 int
381 g_raid_volume_str2level(const char *str, int *level, int *qual)
382 {
383 
384 	*level = G_RAID_VOLUME_RL_UNKNOWN;
385 	*qual = G_RAID_VOLUME_RLQ_NONE;
386 	if (strcasecmp(str, "RAID0") == 0)
387 		*level = G_RAID_VOLUME_RL_RAID0;
388 	else if (strcasecmp(str, "RAID1") == 0)
389 		*level = G_RAID_VOLUME_RL_RAID1;
390 	else if (strcasecmp(str, "RAID3-P0") == 0) {
391 		*level = G_RAID_VOLUME_RL_RAID3;
392 		*qual = G_RAID_VOLUME_RLQ_R3P0;
393 	} else if (strcasecmp(str, "RAID3-PN") == 0 ||
394 		   strcasecmp(str, "RAID3") == 0) {
395 		*level = G_RAID_VOLUME_RL_RAID3;
396 		*qual = G_RAID_VOLUME_RLQ_R3PN;
397 	} else if (strcasecmp(str, "RAID4-P0") == 0) {
398 		*level = G_RAID_VOLUME_RL_RAID4;
399 		*qual = G_RAID_VOLUME_RLQ_R4P0;
400 	} else if (strcasecmp(str, "RAID4-PN") == 0 ||
401 		   strcasecmp(str, "RAID4") == 0) {
402 		*level = G_RAID_VOLUME_RL_RAID4;
403 		*qual = G_RAID_VOLUME_RLQ_R4PN;
404 	} else if (strcasecmp(str, "RAID5-RA") == 0) {
405 		*level = G_RAID_VOLUME_RL_RAID5;
406 		*qual = G_RAID_VOLUME_RLQ_R5RA;
407 	} else if (strcasecmp(str, "RAID5-RS") == 0) {
408 		*level = G_RAID_VOLUME_RL_RAID5;
409 		*qual = G_RAID_VOLUME_RLQ_R5RS;
410 	} else if (strcasecmp(str, "RAID5") == 0 ||
411 		   strcasecmp(str, "RAID5-LA") == 0) {
412 		*level = G_RAID_VOLUME_RL_RAID5;
413 		*qual = G_RAID_VOLUME_RLQ_R5LA;
414 	} else if (strcasecmp(str, "RAID5-LS") == 0) {
415 		*level = G_RAID_VOLUME_RL_RAID5;
416 		*qual = G_RAID_VOLUME_RLQ_R5LS;
417 	} else if (strcasecmp(str, "RAID6-RA") == 0) {
418 		*level = G_RAID_VOLUME_RL_RAID6;
419 		*qual = G_RAID_VOLUME_RLQ_R6RA;
420 	} else if (strcasecmp(str, "RAID6-RS") == 0) {
421 		*level = G_RAID_VOLUME_RL_RAID6;
422 		*qual = G_RAID_VOLUME_RLQ_R6RS;
423 	} else if (strcasecmp(str, "RAID6") == 0 ||
424 		   strcasecmp(str, "RAID6-LA") == 0) {
425 		*level = G_RAID_VOLUME_RL_RAID6;
426 		*qual = G_RAID_VOLUME_RLQ_R6LA;
427 	} else if (strcasecmp(str, "RAID6-LS") == 0) {
428 		*level = G_RAID_VOLUME_RL_RAID6;
429 		*qual = G_RAID_VOLUME_RLQ_R6LS;
430 	} else if (strcasecmp(str, "RAIDMDF-RA") == 0) {
431 		*level = G_RAID_VOLUME_RL_RAIDMDF;
432 		*qual = G_RAID_VOLUME_RLQ_RMDFRA;
433 	} else if (strcasecmp(str, "RAIDMDF-RS") == 0) {
434 		*level = G_RAID_VOLUME_RL_RAIDMDF;
435 		*qual = G_RAID_VOLUME_RLQ_RMDFRS;
436 	} else if (strcasecmp(str, "RAIDMDF") == 0 ||
437 		   strcasecmp(str, "RAIDMDF-LA") == 0) {
438 		*level = G_RAID_VOLUME_RL_RAIDMDF;
439 		*qual = G_RAID_VOLUME_RLQ_RMDFLA;
440 	} else if (strcasecmp(str, "RAIDMDF-LS") == 0) {
441 		*level = G_RAID_VOLUME_RL_RAIDMDF;
442 		*qual = G_RAID_VOLUME_RLQ_RMDFLS;
443 	} else if (strcasecmp(str, "RAID10") == 0 ||
444 		   strcasecmp(str, "RAID1E") == 0 ||
445 		   strcasecmp(str, "RAID1E-A") == 0) {
446 		*level = G_RAID_VOLUME_RL_RAID1E;
447 		*qual = G_RAID_VOLUME_RLQ_R1EA;
448 	} else if (strcasecmp(str, "RAID1E-O") == 0) {
449 		*level = G_RAID_VOLUME_RL_RAID1E;
450 		*qual = G_RAID_VOLUME_RLQ_R1EO;
451 	} else if (strcasecmp(str, "SINGLE") == 0)
452 		*level = G_RAID_VOLUME_RL_SINGLE;
453 	else if (strcasecmp(str, "CONCAT") == 0)
454 		*level = G_RAID_VOLUME_RL_CONCAT;
455 	else if (strcasecmp(str, "RAID5E-RA") == 0) {
456 		*level = G_RAID_VOLUME_RL_RAID5E;
457 		*qual = G_RAID_VOLUME_RLQ_R5ERA;
458 	} else if (strcasecmp(str, "RAID5E-RS") == 0) {
459 		*level = G_RAID_VOLUME_RL_RAID5E;
460 		*qual = G_RAID_VOLUME_RLQ_R5ERS;
461 	} else if (strcasecmp(str, "RAID5E") == 0 ||
462 		   strcasecmp(str, "RAID5E-LA") == 0) {
463 		*level = G_RAID_VOLUME_RL_RAID5E;
464 		*qual = G_RAID_VOLUME_RLQ_R5ELA;
465 	} else if (strcasecmp(str, "RAID5E-LS") == 0) {
466 		*level = G_RAID_VOLUME_RL_RAID5E;
467 		*qual = G_RAID_VOLUME_RLQ_R5ELS;
468 	} else if (strcasecmp(str, "RAID5EE-RA") == 0) {
469 		*level = G_RAID_VOLUME_RL_RAID5EE;
470 		*qual = G_RAID_VOLUME_RLQ_R5EERA;
471 	} else if (strcasecmp(str, "RAID5EE-RS") == 0) {
472 		*level = G_RAID_VOLUME_RL_RAID5EE;
473 		*qual = G_RAID_VOLUME_RLQ_R5EERS;
474 	} else if (strcasecmp(str, "RAID5EE") == 0 ||
475 		   strcasecmp(str, "RAID5EE-LA") == 0) {
476 		*level = G_RAID_VOLUME_RL_RAID5EE;
477 		*qual = G_RAID_VOLUME_RLQ_R5EELA;
478 	} else if (strcasecmp(str, "RAID5EE-LS") == 0) {
479 		*level = G_RAID_VOLUME_RL_RAID5EE;
480 		*qual = G_RAID_VOLUME_RLQ_R5EELS;
481 	} else if (strcasecmp(str, "RAID5R-RA") == 0) {
482 		*level = G_RAID_VOLUME_RL_RAID5R;
483 		*qual = G_RAID_VOLUME_RLQ_R5RRA;
484 	} else if (strcasecmp(str, "RAID5R-RS") == 0) {
485 		*level = G_RAID_VOLUME_RL_RAID5R;
486 		*qual = G_RAID_VOLUME_RLQ_R5RRS;
487 	} else if (strcasecmp(str, "RAID5R") == 0 ||
488 		   strcasecmp(str, "RAID5R-LA") == 0) {
489 		*level = G_RAID_VOLUME_RL_RAID5R;
490 		*qual = G_RAID_VOLUME_RLQ_R5RLA;
491 	} else if (strcasecmp(str, "RAID5R-LS") == 0) {
492 		*level = G_RAID_VOLUME_RL_RAID5R;
493 		*qual = G_RAID_VOLUME_RLQ_R5RLS;
494 	} else
495 		return (-1);
496 	return (0);
497 }
498 
499 const char *
500 g_raid_get_diskname(struct g_raid_disk *disk)
501 {
502 
503 	if (disk->d_consumer == NULL || disk->d_consumer->provider == NULL)
504 		return ("[unknown]");
505 	return (disk->d_consumer->provider->name);
506 }
507 
508 void
509 g_raid_get_disk_info(struct g_raid_disk *disk)
510 {
511 	struct g_consumer *cp = disk->d_consumer;
512 	int error, len;
513 
514 	/* Read kernel dumping information. */
515 	disk->d_kd.offset = 0;
516 	disk->d_kd.length = OFF_MAX;
517 	len = sizeof(disk->d_kd);
518 	error = g_io_getattr("GEOM::kerneldump", cp, &len, &disk->d_kd);
519 	if (error)
520 		disk->d_kd.di.dumper = NULL;
521 	if (disk->d_kd.di.dumper == NULL)
522 		G_RAID_DEBUG1(2, disk->d_softc,
523 		    "Dumping not supported by %s: %d.",
524 		    cp->provider->name, error);
525 
526 	/* Read BIO_DELETE support. */
527 	error = g_getattr("GEOM::candelete", cp, &disk->d_candelete);
528 	if (error)
529 		disk->d_candelete = 0;
530 	if (!disk->d_candelete)
531 		G_RAID_DEBUG1(2, disk->d_softc,
532 		    "BIO_DELETE not supported by %s: %d.",
533 		    cp->provider->name, error);
534 }
535 
536 void
537 g_raid_report_disk_state(struct g_raid_disk *disk)
538 {
539 	struct g_raid_subdisk *sd;
540 	int len, state;
541 	uint32_t s;
542 
543 	if (disk->d_consumer == NULL)
544 		return;
545 	if (disk->d_state == G_RAID_DISK_S_DISABLED) {
546 		s = G_STATE_ACTIVE; /* XXX */
547 	} else if (disk->d_state == G_RAID_DISK_S_FAILED ||
548 	    disk->d_state == G_RAID_DISK_S_STALE_FAILED) {
549 		s = G_STATE_FAILED;
550 	} else {
551 		state = G_RAID_SUBDISK_S_ACTIVE;
552 		TAILQ_FOREACH(sd, &disk->d_subdisks, sd_next) {
553 			if (sd->sd_state < state)
554 				state = sd->sd_state;
555 		}
556 		if (state == G_RAID_SUBDISK_S_FAILED)
557 			s = G_STATE_FAILED;
558 		else if (state == G_RAID_SUBDISK_S_NEW ||
559 		    state == G_RAID_SUBDISK_S_REBUILD)
560 			s = G_STATE_REBUILD;
561 		else if (state == G_RAID_SUBDISK_S_STALE ||
562 		    state == G_RAID_SUBDISK_S_RESYNC)
563 			s = G_STATE_RESYNC;
564 		else
565 			s = G_STATE_ACTIVE;
566 	}
567 	len = sizeof(s);
568 	g_io_getattr("GEOM::setstate", disk->d_consumer, &len, &s);
569 	G_RAID_DEBUG1(2, disk->d_softc, "Disk %s state reported as %d.",
570 	    g_raid_get_diskname(disk), s);
571 }
572 
573 void
574 g_raid_change_disk_state(struct g_raid_disk *disk, int state)
575 {
576 
577 	G_RAID_DEBUG1(0, disk->d_softc, "Disk %s state changed from %s to %s.",
578 	    g_raid_get_diskname(disk),
579 	    g_raid_disk_state2str(disk->d_state),
580 	    g_raid_disk_state2str(state));
581 	disk->d_state = state;
582 	g_raid_report_disk_state(disk);
583 }
584 
585 void
586 g_raid_change_subdisk_state(struct g_raid_subdisk *sd, int state)
587 {
588 
589 	G_RAID_DEBUG1(0, sd->sd_softc,
590 	    "Subdisk %s:%d-%s state changed from %s to %s.",
591 	    sd->sd_volume->v_name, sd->sd_pos,
592 	    sd->sd_disk ? g_raid_get_diskname(sd->sd_disk) : "[none]",
593 	    g_raid_subdisk_state2str(sd->sd_state),
594 	    g_raid_subdisk_state2str(state));
595 	sd->sd_state = state;
596 	if (sd->sd_disk)
597 		g_raid_report_disk_state(sd->sd_disk);
598 }
599 
600 void
601 g_raid_change_volume_state(struct g_raid_volume *vol, int state)
602 {
603 
604 	G_RAID_DEBUG1(0, vol->v_softc,
605 	    "Volume %s state changed from %s to %s.",
606 	    vol->v_name,
607 	    g_raid_volume_state2str(vol->v_state),
608 	    g_raid_volume_state2str(state));
609 	vol->v_state = state;
610 }
611 
612 /*
613  * --- Events handling functions ---
614  * Events in geom_raid are used to maintain subdisks and volumes status
615  * from one thread to simplify locking.
616  */
617 static void
618 g_raid_event_free(struct g_raid_event *ep)
619 {
620 
621 	free(ep, M_RAID);
622 }
623 
624 int
625 g_raid_event_send(void *arg, int event, int flags)
626 {
627 	struct g_raid_softc *sc;
628 	struct g_raid_event *ep;
629 	int error;
630 
631 	if ((flags & G_RAID_EVENT_VOLUME) != 0) {
632 		sc = ((struct g_raid_volume *)arg)->v_softc;
633 	} else if ((flags & G_RAID_EVENT_DISK) != 0) {
634 		sc = ((struct g_raid_disk *)arg)->d_softc;
635 	} else if ((flags & G_RAID_EVENT_SUBDISK) != 0) {
636 		sc = ((struct g_raid_subdisk *)arg)->sd_softc;
637 	} else {
638 		sc = arg;
639 	}
640 	ep = malloc(sizeof(*ep), M_RAID,
641 	    sx_xlocked(&sc->sc_lock) ? M_WAITOK : M_NOWAIT);
642 	if (ep == NULL)
643 		return (ENOMEM);
644 	ep->e_tgt = arg;
645 	ep->e_event = event;
646 	ep->e_flags = flags;
647 	ep->e_error = 0;
648 	G_RAID_DEBUG1(4, sc, "Sending event %p. Waking up %p.", ep, sc);
649 	mtx_lock(&sc->sc_queue_mtx);
650 	TAILQ_INSERT_TAIL(&sc->sc_events, ep, e_next);
651 	mtx_unlock(&sc->sc_queue_mtx);
652 	wakeup(sc);
653 
654 	if ((flags & G_RAID_EVENT_WAIT) == 0)
655 		return (0);
656 
657 	sx_assert(&sc->sc_lock, SX_XLOCKED);
658 	G_RAID_DEBUG1(4, sc, "Sleeping on %p.", ep);
659 	sx_xunlock(&sc->sc_lock);
660 	while ((ep->e_flags & G_RAID_EVENT_DONE) == 0) {
661 		mtx_lock(&sc->sc_queue_mtx);
662 		MSLEEP(error, ep, &sc->sc_queue_mtx, PRIBIO | PDROP, "m:event",
663 		    hz * 5);
664 	}
665 	error = ep->e_error;
666 	g_raid_event_free(ep);
667 	sx_xlock(&sc->sc_lock);
668 	return (error);
669 }
670 
671 static void
672 g_raid_event_cancel(struct g_raid_softc *sc, void *tgt)
673 {
674 	struct g_raid_event *ep, *tmpep;
675 
676 	sx_assert(&sc->sc_lock, SX_XLOCKED);
677 
678 	mtx_lock(&sc->sc_queue_mtx);
679 	TAILQ_FOREACH_SAFE(ep, &sc->sc_events, e_next, tmpep) {
680 		if (ep->e_tgt != tgt)
681 			continue;
682 		TAILQ_REMOVE(&sc->sc_events, ep, e_next);
683 		if ((ep->e_flags & G_RAID_EVENT_WAIT) == 0)
684 			g_raid_event_free(ep);
685 		else {
686 			ep->e_error = ECANCELED;
687 			wakeup(ep);
688 		}
689 	}
690 	mtx_unlock(&sc->sc_queue_mtx);
691 }
692 
693 static int
694 g_raid_event_check(struct g_raid_softc *sc, void *tgt)
695 {
696 	struct g_raid_event *ep;
697 	int	res = 0;
698 
699 	sx_assert(&sc->sc_lock, SX_XLOCKED);
700 
701 	mtx_lock(&sc->sc_queue_mtx);
702 	TAILQ_FOREACH(ep, &sc->sc_events, e_next) {
703 		if (ep->e_tgt != tgt)
704 			continue;
705 		res = 1;
706 		break;
707 	}
708 	mtx_unlock(&sc->sc_queue_mtx);
709 	return (res);
710 }
711 
712 /*
713  * Return the number of disks in given state.
714  * If state is equal to -1, count all connected disks.
715  */
716 u_int
717 g_raid_ndisks(struct g_raid_softc *sc, int state)
718 {
719 	struct g_raid_disk *disk;
720 	u_int n;
721 
722 	sx_assert(&sc->sc_lock, SX_LOCKED);
723 
724 	n = 0;
725 	TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
726 		if (disk->d_state == state || state == -1)
727 			n++;
728 	}
729 	return (n);
730 }
731 
732 /*
733  * Return the number of subdisks in given state.
734  * If state is equal to -1, count all connected disks.
735  */
736 u_int
737 g_raid_nsubdisks(struct g_raid_volume *vol, int state)
738 {
739 	struct g_raid_subdisk *subdisk;
740 	struct g_raid_softc *sc;
741 	u_int i, n ;
742 
743 	sc = vol->v_softc;
744 	sx_assert(&sc->sc_lock, SX_LOCKED);
745 
746 	n = 0;
747 	for (i = 0; i < vol->v_disks_count; i++) {
748 		subdisk = &vol->v_subdisks[i];
749 		if ((state == -1 &&
750 		     subdisk->sd_state != G_RAID_SUBDISK_S_NONE) ||
751 		    subdisk->sd_state == state)
752 			n++;
753 	}
754 	return (n);
755 }
756 
757 /*
758  * Return the first subdisk in given state.
759  * If state is equal to -1, then the first connected disks.
760  */
761 struct g_raid_subdisk *
762 g_raid_get_subdisk(struct g_raid_volume *vol, int state)
763 {
764 	struct g_raid_subdisk *sd;
765 	struct g_raid_softc *sc;
766 	u_int i;
767 
768 	sc = vol->v_softc;
769 	sx_assert(&sc->sc_lock, SX_LOCKED);
770 
771 	for (i = 0; i < vol->v_disks_count; i++) {
772 		sd = &vol->v_subdisks[i];
773 		if ((state == -1 &&
774 		     sd->sd_state != G_RAID_SUBDISK_S_NONE) ||
775 		    sd->sd_state == state)
776 			return (sd);
777 	}
778 	return (NULL);
779 }
780 
781 struct g_consumer *
782 g_raid_open_consumer(struct g_raid_softc *sc, const char *name)
783 {
784 	struct g_consumer *cp;
785 	struct g_provider *pp;
786 
787 	g_topology_assert();
788 
789 	if (strncmp(name, "/dev/", 5) == 0)
790 		name += 5;
791 	pp = g_provider_by_name(name);
792 	if (pp == NULL)
793 		return (NULL);
794 	cp = g_new_consumer(sc->sc_geom);
795 	if (g_attach(cp, pp) != 0) {
796 		g_destroy_consumer(cp);
797 		return (NULL);
798 	}
799 	if (g_access(cp, 1, 1, 1) != 0) {
800 		g_detach(cp);
801 		g_destroy_consumer(cp);
802 		return (NULL);
803 	}
804 	return (cp);
805 }
806 
807 static u_int
808 g_raid_nrequests(struct g_raid_softc *sc, struct g_consumer *cp)
809 {
810 	struct bio *bp;
811 	u_int nreqs = 0;
812 
813 	mtx_lock(&sc->sc_queue_mtx);
814 	TAILQ_FOREACH(bp, &sc->sc_queue.queue, bio_queue) {
815 		if (bp->bio_from == cp)
816 			nreqs++;
817 	}
818 	mtx_unlock(&sc->sc_queue_mtx);
819 	return (nreqs);
820 }
821 
822 u_int
823 g_raid_nopens(struct g_raid_softc *sc)
824 {
825 	struct g_raid_volume *vol;
826 	u_int opens;
827 
828 	opens = 0;
829 	TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) {
830 		if (vol->v_provider_open != 0)
831 			opens++;
832 	}
833 	return (opens);
834 }
835 
836 static int
837 g_raid_consumer_is_busy(struct g_raid_softc *sc, struct g_consumer *cp)
838 {
839 
840 	if (cp->index > 0) {
841 		G_RAID_DEBUG1(2, sc,
842 		    "I/O requests for %s exist, can't destroy it now.",
843 		    cp->provider->name);
844 		return (1);
845 	}
846 	if (g_raid_nrequests(sc, cp) > 0) {
847 		G_RAID_DEBUG1(2, sc,
848 		    "I/O requests for %s in queue, can't destroy it now.",
849 		    cp->provider->name);
850 		return (1);
851 	}
852 	return (0);
853 }
854 
855 static void
856 g_raid_destroy_consumer(void *arg, int flags __unused)
857 {
858 	struct g_consumer *cp;
859 
860 	g_topology_assert();
861 
862 	cp = arg;
863 	G_RAID_DEBUG(1, "Consumer %s destroyed.", cp->provider->name);
864 	g_detach(cp);
865 	g_destroy_consumer(cp);
866 }
867 
868 void
869 g_raid_kill_consumer(struct g_raid_softc *sc, struct g_consumer *cp)
870 {
871 	struct g_provider *pp;
872 	int retaste_wait;
873 
874 	g_topology_assert_not();
875 
876 	g_topology_lock();
877 	cp->private = NULL;
878 	if (g_raid_consumer_is_busy(sc, cp))
879 		goto out;
880 	pp = cp->provider;
881 	retaste_wait = 0;
882 	if (cp->acw == 1) {
883 		if ((pp->geom->flags & G_GEOM_WITHER) == 0)
884 			retaste_wait = 1;
885 	}
886 	if (cp->acr > 0 || cp->acw > 0 || cp->ace > 0)
887 		g_access(cp, -cp->acr, -cp->acw, -cp->ace);
888 	if (retaste_wait) {
889 		/*
890 		 * After retaste event was send (inside g_access()), we can send
891 		 * event to detach and destroy consumer.
892 		 * A class, which has consumer to the given provider connected
893 		 * will not receive retaste event for the provider.
894 		 * This is the way how I ignore retaste events when I close
895 		 * consumers opened for write: I detach and destroy consumer
896 		 * after retaste event is sent.
897 		 */
898 		g_post_event(g_raid_destroy_consumer, cp, M_WAITOK, NULL);
899 		goto out;
900 	}
901 	G_RAID_DEBUG(1, "Consumer %s destroyed.", pp->name);
902 	g_detach(cp);
903 	g_destroy_consumer(cp);
904 out:
905 	g_topology_unlock();
906 }
907 
908 static void
909 g_raid_orphan(struct g_consumer *cp)
910 {
911 	struct g_raid_disk *disk;
912 
913 	g_topology_assert();
914 
915 	disk = cp->private;
916 	if (disk == NULL)
917 		return;
918 	g_raid_event_send(disk, G_RAID_DISK_E_DISCONNECTED,
919 	    G_RAID_EVENT_DISK);
920 }
921 
922 static void
923 g_raid_clean(struct g_raid_volume *vol, int acw)
924 {
925 	struct g_raid_softc *sc;
926 	int timeout;
927 
928 	sc = vol->v_softc;
929 	g_topology_assert_not();
930 	sx_assert(&sc->sc_lock, SX_XLOCKED);
931 
932 //	if ((sc->sc_flags & G_RAID_DEVICE_FLAG_NOFAILSYNC) != 0)
933 //		return;
934 	if (!vol->v_dirty)
935 		return;
936 	if (vol->v_writes > 0)
937 		return;
938 	if (acw > 0 || (acw == -1 &&
939 	    vol->v_provider != NULL && vol->v_provider->acw > 0)) {
940 		timeout = g_raid_clean_time - (time_uptime - vol->v_last_write);
941 		if (!g_raid_shutdown && timeout > 0)
942 			return;
943 	}
944 	vol->v_dirty = 0;
945 	G_RAID_DEBUG1(1, sc, "Volume %s marked as clean.",
946 	    vol->v_name);
947 	g_raid_write_metadata(sc, vol, NULL, NULL);
948 }
949 
950 static void
951 g_raid_dirty(struct g_raid_volume *vol)
952 {
953 	struct g_raid_softc *sc;
954 
955 	sc = vol->v_softc;
956 	g_topology_assert_not();
957 	sx_assert(&sc->sc_lock, SX_XLOCKED);
958 
959 //	if ((sc->sc_flags & G_RAID_DEVICE_FLAG_NOFAILSYNC) != 0)
960 //		return;
961 	vol->v_dirty = 1;
962 	G_RAID_DEBUG1(1, sc, "Volume %s marked as dirty.",
963 	    vol->v_name);
964 	g_raid_write_metadata(sc, vol, NULL, NULL);
965 }
966 
967 void
968 g_raid_tr_flush_common(struct g_raid_tr_object *tr, struct bio *bp)
969 {
970 	struct g_raid_softc *sc;
971 	struct g_raid_volume *vol;
972 	struct g_raid_subdisk *sd;
973 	struct bio_queue_head queue;
974 	struct bio *cbp;
975 	int i;
976 
977 	vol = tr->tro_volume;
978 	sc = vol->v_softc;
979 
980 	/*
981 	 * Allocate all bios before sending any request, so we can return
982 	 * ENOMEM in nice and clean way.
983 	 */
984 	bioq_init(&queue);
985 	for (i = 0; i < vol->v_disks_count; i++) {
986 		sd = &vol->v_subdisks[i];
987 		if (sd->sd_state == G_RAID_SUBDISK_S_NONE ||
988 		    sd->sd_state == G_RAID_SUBDISK_S_FAILED)
989 			continue;
990 		cbp = g_clone_bio(bp);
991 		if (cbp == NULL)
992 			goto failure;
993 		cbp->bio_caller1 = sd;
994 		bioq_insert_tail(&queue, cbp);
995 	}
996 	while ((cbp = bioq_takefirst(&queue)) != NULL) {
997 		sd = cbp->bio_caller1;
998 		cbp->bio_caller1 = NULL;
999 		g_raid_subdisk_iostart(sd, cbp);
1000 	}
1001 	return;
1002 failure:
1003 	while ((cbp = bioq_takefirst(&queue)) != NULL)
1004 		g_destroy_bio(cbp);
1005 	if (bp->bio_error == 0)
1006 		bp->bio_error = ENOMEM;
1007 	g_raid_iodone(bp, bp->bio_error);
1008 }
1009 
1010 static void
1011 g_raid_tr_kerneldump_common_done(struct bio *bp)
1012 {
1013 
1014 	bp->bio_flags |= BIO_DONE;
1015 }
1016 
1017 int
1018 g_raid_tr_kerneldump_common(struct g_raid_tr_object *tr,
1019     void *virtual, vm_offset_t physical, off_t offset, size_t length)
1020 {
1021 	struct g_raid_softc *sc;
1022 	struct g_raid_volume *vol;
1023 	struct bio bp;
1024 
1025 	vol = tr->tro_volume;
1026 	sc = vol->v_softc;
1027 
1028 	bzero(&bp, sizeof(bp));
1029 	bp.bio_cmd = BIO_WRITE;
1030 	bp.bio_done = g_raid_tr_kerneldump_common_done;
1031 	bp.bio_attribute = NULL;
1032 	bp.bio_offset = offset;
1033 	bp.bio_length = length;
1034 	bp.bio_data = virtual;
1035 	bp.bio_to = vol->v_provider;
1036 
1037 	g_raid_start(&bp);
1038 	while (!(bp.bio_flags & BIO_DONE)) {
1039 		G_RAID_DEBUG1(4, sc, "Poll...");
1040 		g_raid_poll(sc);
1041 		DELAY(10);
1042 	}
1043 
1044 	return (bp.bio_error != 0 ? EIO : 0);
1045 }
1046 
1047 static int
1048 g_raid_dump(void *arg,
1049     void *virtual, vm_offset_t physical, off_t offset, size_t length)
1050 {
1051 	struct g_raid_volume *vol;
1052 	int error;
1053 
1054 	vol = (struct g_raid_volume *)arg;
1055 	G_RAID_DEBUG1(3, vol->v_softc, "Dumping at off %llu len %llu.",
1056 	    (long long unsigned)offset, (long long unsigned)length);
1057 
1058 	error = G_RAID_TR_KERNELDUMP(vol->v_tr,
1059 	    virtual, physical, offset, length);
1060 	return (error);
1061 }
1062 
1063 static void
1064 g_raid_kerneldump(struct g_raid_softc *sc, struct bio *bp)
1065 {
1066 	struct g_kerneldump *gkd;
1067 	struct g_provider *pp;
1068 	struct g_raid_volume *vol;
1069 
1070 	gkd = (struct g_kerneldump*)bp->bio_data;
1071 	pp = bp->bio_to;
1072 	vol = pp->private;
1073 	g_trace(G_T_TOPOLOGY, "g_raid_kerneldump(%s, %jd, %jd)",
1074 		pp->name, (intmax_t)gkd->offset, (intmax_t)gkd->length);
1075 	gkd->di.dumper = g_raid_dump;
1076 	gkd->di.priv = vol;
1077 	gkd->di.blocksize = vol->v_sectorsize;
1078 	gkd->di.maxiosize = DFLTPHYS;
1079 	gkd->di.mediaoffset = gkd->offset;
1080 	if ((gkd->offset + gkd->length) > vol->v_mediasize)
1081 		gkd->length = vol->v_mediasize - gkd->offset;
1082 	gkd->di.mediasize = gkd->length;
1083 	g_io_deliver(bp, 0);
1084 }
1085 
1086 static void
1087 g_raid_candelete(struct g_raid_softc *sc, struct bio *bp)
1088 {
1089 	struct g_provider *pp;
1090 	struct g_raid_volume *vol;
1091 	struct g_raid_subdisk *sd;
1092 	int *val;
1093 	int i;
1094 
1095 	val = (int *)bp->bio_data;
1096 	pp = bp->bio_to;
1097 	vol = pp->private;
1098 	*val = 0;
1099 	for (i = 0; i < vol->v_disks_count; i++) {
1100 		sd = &vol->v_subdisks[i];
1101 		if (sd->sd_state == G_RAID_SUBDISK_S_NONE)
1102 			continue;
1103 		if (sd->sd_disk->d_candelete) {
1104 			*val = 1;
1105 			break;
1106 		}
1107 	}
1108 	g_io_deliver(bp, 0);
1109 }
1110 
1111 static void
1112 g_raid_start(struct bio *bp)
1113 {
1114 	struct g_raid_softc *sc;
1115 
1116 	sc = bp->bio_to->geom->softc;
1117 	/*
1118 	 * If sc == NULL or there are no valid disks, provider's error
1119 	 * should be set and g_raid_start() should not be called at all.
1120 	 */
1121 //	KASSERT(sc != NULL && sc->sc_state == G_RAID_VOLUME_S_RUNNING,
1122 //	    ("Provider's error should be set (error=%d)(mirror=%s).",
1123 //	    bp->bio_to->error, bp->bio_to->name));
1124 	G_RAID_LOGREQ(3, bp, "Request received.");
1125 
1126 	switch (bp->bio_cmd) {
1127 	case BIO_READ:
1128 	case BIO_WRITE:
1129 	case BIO_DELETE:
1130 	case BIO_FLUSH:
1131 		break;
1132 	case BIO_GETATTR:
1133 		if (!strcmp(bp->bio_attribute, "GEOM::candelete"))
1134 			g_raid_candelete(sc, bp);
1135 		else if (!strcmp(bp->bio_attribute, "GEOM::kerneldump"))
1136 			g_raid_kerneldump(sc, bp);
1137 		else
1138 			g_io_deliver(bp, EOPNOTSUPP);
1139 		return;
1140 	default:
1141 		g_io_deliver(bp, EOPNOTSUPP);
1142 		return;
1143 	}
1144 	mtx_lock(&sc->sc_queue_mtx);
1145 	bioq_disksort(&sc->sc_queue, bp);
1146 	mtx_unlock(&sc->sc_queue_mtx);
1147 	if (!dumping) {
1148 		G_RAID_DEBUG1(4, sc, "Waking up %p.", sc);
1149 		wakeup(sc);
1150 	}
1151 }
1152 
1153 static int
1154 g_raid_bio_overlaps(const struct bio *bp, off_t lstart, off_t len)
1155 {
1156 	/*
1157 	 * 5 cases:
1158 	 * (1) bp entirely below NO
1159 	 * (2) bp entirely above NO
1160 	 * (3) bp start below, but end in range YES
1161 	 * (4) bp entirely within YES
1162 	 * (5) bp starts within, ends above YES
1163 	 *
1164 	 * lock range 10-19 (offset 10 length 10)
1165 	 * (1) 1-5: first if kicks it out
1166 	 * (2) 30-35: second if kicks it out
1167 	 * (3) 5-15: passes both ifs
1168 	 * (4) 12-14: passes both ifs
1169 	 * (5) 19-20: passes both
1170 	 */
1171 	off_t lend = lstart + len - 1;
1172 	off_t bstart = bp->bio_offset;
1173 	off_t bend = bp->bio_offset + bp->bio_length - 1;
1174 
1175 	if (bend < lstart)
1176 		return (0);
1177 	if (lend < bstart)
1178 		return (0);
1179 	return (1);
1180 }
1181 
1182 static int
1183 g_raid_is_in_locked_range(struct g_raid_volume *vol, const struct bio *bp)
1184 {
1185 	struct g_raid_lock *lp;
1186 
1187 	sx_assert(&vol->v_softc->sc_lock, SX_LOCKED);
1188 
1189 	LIST_FOREACH(lp, &vol->v_locks, l_next) {
1190 		if (g_raid_bio_overlaps(bp, lp->l_offset, lp->l_length))
1191 			return (1);
1192 	}
1193 	return (0);
1194 }
1195 
1196 static void
1197 g_raid_start_request(struct bio *bp)
1198 {
1199 	struct g_raid_softc *sc;
1200 	struct g_raid_volume *vol;
1201 
1202 	sc = bp->bio_to->geom->softc;
1203 	sx_assert(&sc->sc_lock, SX_LOCKED);
1204 	vol = bp->bio_to->private;
1205 
1206 	/*
1207 	 * Check to see if this item is in a locked range.  If so,
1208 	 * queue it to our locked queue and return.  We'll requeue
1209 	 * it when the range is unlocked.  Internal I/O for the
1210 	 * rebuild/rescan/recovery process is excluded from this
1211 	 * check so we can actually do the recovery.
1212 	 */
1213 	if (!(bp->bio_cflags & G_RAID_BIO_FLAG_SPECIAL) &&
1214 	    g_raid_is_in_locked_range(vol, bp)) {
1215 		G_RAID_LOGREQ(3, bp, "Defer request.");
1216 		bioq_insert_tail(&vol->v_locked, bp);
1217 		return;
1218 	}
1219 
1220 	/*
1221 	 * If we're actually going to do the write/delete, then
1222 	 * update the idle stats for the volume.
1223 	 */
1224 	if (bp->bio_cmd == BIO_WRITE || bp->bio_cmd == BIO_DELETE) {
1225 		if (!vol->v_dirty)
1226 			g_raid_dirty(vol);
1227 		vol->v_writes++;
1228 	}
1229 
1230 	/*
1231 	 * Put request onto inflight queue, so we can check if new
1232 	 * synchronization requests don't collide with it.  Then tell
1233 	 * the transformation layer to start the I/O.
1234 	 */
1235 	bioq_insert_tail(&vol->v_inflight, bp);
1236 	G_RAID_LOGREQ(4, bp, "Request started");
1237 	G_RAID_TR_IOSTART(vol->v_tr, bp);
1238 }
1239 
1240 static void
1241 g_raid_finish_with_locked_ranges(struct g_raid_volume *vol, struct bio *bp)
1242 {
1243 	off_t off, len;
1244 	struct bio *nbp;
1245 	struct g_raid_lock *lp;
1246 
1247 	vol->v_pending_lock = 0;
1248 	LIST_FOREACH(lp, &vol->v_locks, l_next) {
1249 		if (lp->l_pending) {
1250 			off = lp->l_offset;
1251 			len = lp->l_length;
1252 			lp->l_pending = 0;
1253 			TAILQ_FOREACH(nbp, &vol->v_inflight.queue, bio_queue) {
1254 				if (g_raid_bio_overlaps(nbp, off, len))
1255 					lp->l_pending++;
1256 			}
1257 			if (lp->l_pending) {
1258 				vol->v_pending_lock = 1;
1259 				G_RAID_DEBUG1(4, vol->v_softc,
1260 				    "Deferred lock(%jd, %jd) has %d pending",
1261 				    (intmax_t)off, (intmax_t)(off + len),
1262 				    lp->l_pending);
1263 				continue;
1264 			}
1265 			G_RAID_DEBUG1(4, vol->v_softc,
1266 			    "Deferred lock of %jd to %jd completed",
1267 			    (intmax_t)off, (intmax_t)(off + len));
1268 			G_RAID_TR_LOCKED(vol->v_tr, lp->l_callback_arg);
1269 		}
1270 	}
1271 }
1272 
1273 void
1274 g_raid_iodone(struct bio *bp, int error)
1275 {
1276 	struct g_raid_softc *sc;
1277 	struct g_raid_volume *vol;
1278 
1279 	sc = bp->bio_to->geom->softc;
1280 	sx_assert(&sc->sc_lock, SX_LOCKED);
1281 	vol = bp->bio_to->private;
1282 	G_RAID_LOGREQ(3, bp, "Request done: %d.", error);
1283 
1284 	/* Update stats if we done write/delete. */
1285 	if (bp->bio_cmd == BIO_WRITE || bp->bio_cmd == BIO_DELETE) {
1286 		vol->v_writes--;
1287 		vol->v_last_write = time_uptime;
1288 	}
1289 
1290 	bioq_remove(&vol->v_inflight, bp);
1291 	if (vol->v_pending_lock && g_raid_is_in_locked_range(vol, bp))
1292 		g_raid_finish_with_locked_ranges(vol, bp);
1293 	getmicrouptime(&vol->v_last_done);
1294 	g_io_deliver(bp, error);
1295 }
1296 
1297 int
1298 g_raid_lock_range(struct g_raid_volume *vol, off_t off, off_t len,
1299     struct bio *ignore, void *argp)
1300 {
1301 	struct g_raid_softc *sc;
1302 	struct g_raid_lock *lp;
1303 	struct bio *bp;
1304 
1305 	sc = vol->v_softc;
1306 	lp = malloc(sizeof(*lp), M_RAID, M_WAITOK | M_ZERO);
1307 	LIST_INSERT_HEAD(&vol->v_locks, lp, l_next);
1308 	lp->l_offset = off;
1309 	lp->l_length = len;
1310 	lp->l_callback_arg = argp;
1311 
1312 	lp->l_pending = 0;
1313 	TAILQ_FOREACH(bp, &vol->v_inflight.queue, bio_queue) {
1314 		if (bp != ignore && g_raid_bio_overlaps(bp, off, len))
1315 			lp->l_pending++;
1316 	}
1317 
1318 	/*
1319 	 * If there are any writes that are pending, we return EBUSY.  All
1320 	 * callers will have to wait until all pending writes clear.
1321 	 */
1322 	if (lp->l_pending > 0) {
1323 		vol->v_pending_lock = 1;
1324 		G_RAID_DEBUG1(4, sc, "Locking range %jd to %jd deferred %d pend",
1325 		    (intmax_t)off, (intmax_t)(off+len), lp->l_pending);
1326 		return (EBUSY);
1327 	}
1328 	G_RAID_DEBUG1(4, sc, "Locking range %jd to %jd",
1329 	    (intmax_t)off, (intmax_t)(off+len));
1330 	G_RAID_TR_LOCKED(vol->v_tr, lp->l_callback_arg);
1331 	return (0);
1332 }
1333 
1334 int
1335 g_raid_unlock_range(struct g_raid_volume *vol, off_t off, off_t len)
1336 {
1337 	struct g_raid_lock *lp;
1338 	struct g_raid_softc *sc;
1339 	struct bio *bp;
1340 
1341 	sc = vol->v_softc;
1342 	LIST_FOREACH(lp, &vol->v_locks, l_next) {
1343 		if (lp->l_offset == off && lp->l_length == len) {
1344 			LIST_REMOVE(lp, l_next);
1345 			/* XXX
1346 			 * Right now we just put them all back on the queue
1347 			 * and hope for the best.  We hope this because any
1348 			 * locked ranges will go right back on this list
1349 			 * when the worker thread runs.
1350 			 * XXX
1351 			 */
1352 			G_RAID_DEBUG1(4, sc, "Unlocked %jd to %jd",
1353 			    (intmax_t)lp->l_offset,
1354 			    (intmax_t)(lp->l_offset+lp->l_length));
1355 			mtx_lock(&sc->sc_queue_mtx);
1356 			while ((bp = bioq_takefirst(&vol->v_locked)) != NULL)
1357 				bioq_disksort(&sc->sc_queue, bp);
1358 			mtx_unlock(&sc->sc_queue_mtx);
1359 			free(lp, M_RAID);
1360 			return (0);
1361 		}
1362 	}
1363 	return (EINVAL);
1364 }
1365 
1366 void
1367 g_raid_subdisk_iostart(struct g_raid_subdisk *sd, struct bio *bp)
1368 {
1369 	struct g_consumer *cp;
1370 	struct g_raid_disk *disk, *tdisk;
1371 
1372 	bp->bio_caller1 = sd;
1373 
1374 	/*
1375 	 * Make sure that the disk is present. Generally it is a task of
1376 	 * transformation layers to not send requests to absent disks, but
1377 	 * it is better to be safe and report situation then sorry.
1378 	 */
1379 	if (sd->sd_disk == NULL) {
1380 		G_RAID_LOGREQ(0, bp, "Warning! I/O request to an absent disk!");
1381 nodisk:
1382 		bp->bio_from = NULL;
1383 		bp->bio_to = NULL;
1384 		bp->bio_error = ENXIO;
1385 		g_raid_disk_done(bp);
1386 		return;
1387 	}
1388 	disk = sd->sd_disk;
1389 	if (disk->d_state != G_RAID_DISK_S_ACTIVE &&
1390 	    disk->d_state != G_RAID_DISK_S_FAILED) {
1391 		G_RAID_LOGREQ(0, bp, "Warning! I/O request to a disk in a "
1392 		    "wrong state (%s)!", g_raid_disk_state2str(disk->d_state));
1393 		goto nodisk;
1394 	}
1395 
1396 	cp = disk->d_consumer;
1397 	bp->bio_from = cp;
1398 	bp->bio_to = cp->provider;
1399 	cp->index++;
1400 
1401 	/* Update average disks load. */
1402 	TAILQ_FOREACH(tdisk, &sd->sd_softc->sc_disks, d_next) {
1403 		if (tdisk->d_consumer == NULL)
1404 			tdisk->d_load = 0;
1405 		else
1406 			tdisk->d_load = (tdisk->d_consumer->index *
1407 			    G_RAID_SUBDISK_LOAD_SCALE + tdisk->d_load * 7) / 8;
1408 	}
1409 
1410 	disk->d_last_offset = bp->bio_offset + bp->bio_length;
1411 	if (dumping) {
1412 		G_RAID_LOGREQ(3, bp, "Sending dumping request.");
1413 		if (bp->bio_cmd == BIO_WRITE) {
1414 			bp->bio_error = g_raid_subdisk_kerneldump(sd,
1415 			    bp->bio_data, 0, bp->bio_offset, bp->bio_length);
1416 		} else
1417 			bp->bio_error = EOPNOTSUPP;
1418 		g_raid_disk_done(bp);
1419 	} else {
1420 		bp->bio_done = g_raid_disk_done;
1421 		bp->bio_offset += sd->sd_offset;
1422 		G_RAID_LOGREQ(3, bp, "Sending request.");
1423 		g_io_request(bp, cp);
1424 	}
1425 }
1426 
1427 int
1428 g_raid_subdisk_kerneldump(struct g_raid_subdisk *sd,
1429     void *virtual, vm_offset_t physical, off_t offset, size_t length)
1430 {
1431 
1432 	if (sd->sd_disk == NULL)
1433 		return (ENXIO);
1434 	if (sd->sd_disk->d_kd.di.dumper == NULL)
1435 		return (EOPNOTSUPP);
1436 	return (dump_write(&sd->sd_disk->d_kd.di,
1437 	    virtual, physical,
1438 	    sd->sd_disk->d_kd.di.mediaoffset + sd->sd_offset + offset,
1439 	    length));
1440 }
1441 
1442 static void
1443 g_raid_disk_done(struct bio *bp)
1444 {
1445 	struct g_raid_softc *sc;
1446 	struct g_raid_subdisk *sd;
1447 
1448 	sd = bp->bio_caller1;
1449 	sc = sd->sd_softc;
1450 	mtx_lock(&sc->sc_queue_mtx);
1451 	bioq_disksort(&sc->sc_queue, bp);
1452 	mtx_unlock(&sc->sc_queue_mtx);
1453 	if (!dumping)
1454 		wakeup(sc);
1455 }
1456 
1457 static void
1458 g_raid_disk_done_request(struct bio *bp)
1459 {
1460 	struct g_raid_softc *sc;
1461 	struct g_raid_disk *disk;
1462 	struct g_raid_subdisk *sd;
1463 	struct g_raid_volume *vol;
1464 
1465 	g_topology_assert_not();
1466 
1467 	G_RAID_LOGREQ(3, bp, "Disk request done: %d.", bp->bio_error);
1468 	sd = bp->bio_caller1;
1469 	sc = sd->sd_softc;
1470 	vol = sd->sd_volume;
1471 	if (bp->bio_from != NULL) {
1472 		bp->bio_from->index--;
1473 		disk = bp->bio_from->private;
1474 		if (disk == NULL)
1475 			g_raid_kill_consumer(sc, bp->bio_from);
1476 	}
1477 	bp->bio_offset -= sd->sd_offset;
1478 
1479 	G_RAID_TR_IODONE(vol->v_tr, sd, bp);
1480 }
1481 
1482 static void
1483 g_raid_handle_event(struct g_raid_softc *sc, struct g_raid_event *ep)
1484 {
1485 
1486 	if ((ep->e_flags & G_RAID_EVENT_VOLUME) != 0)
1487 		ep->e_error = g_raid_update_volume(ep->e_tgt, ep->e_event);
1488 	else if ((ep->e_flags & G_RAID_EVENT_DISK) != 0)
1489 		ep->e_error = g_raid_update_disk(ep->e_tgt, ep->e_event);
1490 	else if ((ep->e_flags & G_RAID_EVENT_SUBDISK) != 0)
1491 		ep->e_error = g_raid_update_subdisk(ep->e_tgt, ep->e_event);
1492 	else
1493 		ep->e_error = g_raid_update_node(ep->e_tgt, ep->e_event);
1494 	if ((ep->e_flags & G_RAID_EVENT_WAIT) == 0) {
1495 		KASSERT(ep->e_error == 0,
1496 		    ("Error cannot be handled."));
1497 		g_raid_event_free(ep);
1498 	} else {
1499 		ep->e_flags |= G_RAID_EVENT_DONE;
1500 		G_RAID_DEBUG1(4, sc, "Waking up %p.", ep);
1501 		mtx_lock(&sc->sc_queue_mtx);
1502 		wakeup(ep);
1503 		mtx_unlock(&sc->sc_queue_mtx);
1504 	}
1505 }
1506 
1507 /*
1508  * Worker thread.
1509  */
1510 static void
1511 g_raid_worker(void *arg)
1512 {
1513 	struct g_raid_softc *sc;
1514 	struct g_raid_event *ep;
1515 	struct g_raid_volume *vol;
1516 	struct bio *bp;
1517 	struct timeval now, t;
1518 	int timeout, rv;
1519 
1520 	sc = arg;
1521 	thread_lock(curthread);
1522 	sched_prio(curthread, PRIBIO);
1523 	thread_unlock(curthread);
1524 
1525 	sx_xlock(&sc->sc_lock);
1526 	for (;;) {
1527 		mtx_lock(&sc->sc_queue_mtx);
1528 		/*
1529 		 * First take a look at events.
1530 		 * This is important to handle events before any I/O requests.
1531 		 */
1532 		bp = NULL;
1533 		vol = NULL;
1534 		rv = 0;
1535 		ep = TAILQ_FIRST(&sc->sc_events);
1536 		if (ep != NULL)
1537 			TAILQ_REMOVE(&sc->sc_events, ep, e_next);
1538 		else if ((bp = bioq_takefirst(&sc->sc_queue)) != NULL)
1539 			;
1540 		else {
1541 			getmicrouptime(&now);
1542 			t = now;
1543 			TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) {
1544 				if (bioq_first(&vol->v_inflight) == NULL &&
1545 				    vol->v_tr &&
1546 				    timevalcmp(&vol->v_last_done, &t, < ))
1547 					t = vol->v_last_done;
1548 			}
1549 			timevalsub(&t, &now);
1550 			timeout = g_raid_idle_threshold +
1551 			    t.tv_sec * 1000000 + t.tv_usec;
1552 			if (timeout > 0) {
1553 				/*
1554 				 * Two steps to avoid overflows at HZ=1000
1555 				 * and idle timeouts > 2.1s.  Some rounding
1556 				 * errors can occur, but they are < 1tick,
1557 				 * which is deemed to be close enough for
1558 				 * this purpose.
1559 				 */
1560 				int micpertic = 1000000 / hz;
1561 				timeout = (timeout + micpertic - 1) / micpertic;
1562 				sx_xunlock(&sc->sc_lock);
1563 				MSLEEP(rv, sc, &sc->sc_queue_mtx,
1564 				    PRIBIO | PDROP, "-", timeout);
1565 				sx_xlock(&sc->sc_lock);
1566 				goto process;
1567 			} else
1568 				rv = EWOULDBLOCK;
1569 		}
1570 		mtx_unlock(&sc->sc_queue_mtx);
1571 process:
1572 		if (ep != NULL) {
1573 			g_raid_handle_event(sc, ep);
1574 		} else if (bp != NULL) {
1575 			if (bp->bio_to != NULL &&
1576 			    bp->bio_to->geom == sc->sc_geom)
1577 				g_raid_start_request(bp);
1578 			else
1579 				g_raid_disk_done_request(bp);
1580 		} else if (rv == EWOULDBLOCK) {
1581 			TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) {
1582 				g_raid_clean(vol, -1);
1583 				if (bioq_first(&vol->v_inflight) == NULL &&
1584 				    vol->v_tr) {
1585 					t.tv_sec = g_raid_idle_threshold / 1000000;
1586 					t.tv_usec = g_raid_idle_threshold % 1000000;
1587 					timevaladd(&t, &vol->v_last_done);
1588 					getmicrouptime(&now);
1589 					if (timevalcmp(&t, &now, <= )) {
1590 						G_RAID_TR_IDLE(vol->v_tr);
1591 						vol->v_last_done = now;
1592 					}
1593 				}
1594 			}
1595 		}
1596 		if (sc->sc_stopping == G_RAID_DESTROY_HARD)
1597 			g_raid_destroy_node(sc, 1);	/* May not return. */
1598 	}
1599 }
1600 
1601 static void
1602 g_raid_poll(struct g_raid_softc *sc)
1603 {
1604 	struct g_raid_event *ep;
1605 	struct bio *bp;
1606 
1607 	sx_xlock(&sc->sc_lock);
1608 	mtx_lock(&sc->sc_queue_mtx);
1609 	/*
1610 	 * First take a look at events.
1611 	 * This is important to handle events before any I/O requests.
1612 	 */
1613 	ep = TAILQ_FIRST(&sc->sc_events);
1614 	if (ep != NULL) {
1615 		TAILQ_REMOVE(&sc->sc_events, ep, e_next);
1616 		mtx_unlock(&sc->sc_queue_mtx);
1617 		g_raid_handle_event(sc, ep);
1618 		goto out;
1619 	}
1620 	bp = bioq_takefirst(&sc->sc_queue);
1621 	if (bp != NULL) {
1622 		mtx_unlock(&sc->sc_queue_mtx);
1623 		if (bp->bio_from == NULL ||
1624 		    bp->bio_from->geom != sc->sc_geom)
1625 			g_raid_start_request(bp);
1626 		else
1627 			g_raid_disk_done_request(bp);
1628 	}
1629 out:
1630 	sx_xunlock(&sc->sc_lock);
1631 }
1632 
1633 static void
1634 g_raid_launch_provider(struct g_raid_volume *vol)
1635 {
1636 	struct g_raid_disk *disk;
1637 	struct g_raid_subdisk *sd;
1638 	struct g_raid_softc *sc;
1639 	struct g_provider *pp;
1640 	char name[G_RAID_MAX_VOLUMENAME];
1641 	char   announce_buf[80], buf1[32];
1642 	off_t off;
1643 	int i;
1644 
1645 	sc = vol->v_softc;
1646 	sx_assert(&sc->sc_lock, SX_LOCKED);
1647 
1648 	g_topology_lock();
1649 	/* Try to name provider with volume name. */
1650 	snprintf(name, sizeof(name), "raid/%s", vol->v_name);
1651 	if (g_raid_name_format == 0 || vol->v_name[0] == 0 ||
1652 	    g_provider_by_name(name) != NULL) {
1653 		/* Otherwise use sequential volume number. */
1654 		snprintf(name, sizeof(name), "raid/r%d", vol->v_global_id);
1655 	}
1656 
1657 	/*
1658 	 * Create a /dev/ar%d that the old ataraid(4) stack once
1659 	 * created as an alias for /dev/raid/r%d if requested.
1660 	 * This helps going from stable/7 ataraid devices to newer
1661 	 * FreeBSD releases. sbruno 07 MAY 2013
1662 	 */
1663 
1664         if (ar_legacy_aliases) {
1665 		snprintf(announce_buf, sizeof(announce_buf),
1666                         "kern.devalias.%s", name);
1667                 snprintf(buf1, sizeof(buf1),
1668                         "ar%d", vol->v_global_id);
1669                 setenv(announce_buf, buf1);
1670         }
1671 
1672 	pp = g_new_providerf(sc->sc_geom, "%s", name);
1673 	if (vol->v_tr->tro_class->trc_accept_unmapped) {
1674 		pp->flags |= G_PF_ACCEPT_UNMAPPED;
1675 		for (i = 0; i < vol->v_disks_count; i++) {
1676 			sd = &vol->v_subdisks[i];
1677 			if (sd->sd_state == G_RAID_SUBDISK_S_NONE)
1678 				continue;
1679 			if ((sd->sd_disk->d_consumer->provider->flags &
1680 			    G_PF_ACCEPT_UNMAPPED) == 0)
1681 				pp->flags &= ~G_PF_ACCEPT_UNMAPPED;
1682 		}
1683 	}
1684 	pp->private = vol;
1685 	pp->mediasize = vol->v_mediasize;
1686 	pp->sectorsize = vol->v_sectorsize;
1687 	pp->stripesize = 0;
1688 	pp->stripeoffset = 0;
1689 	if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID1 ||
1690 	    vol->v_raid_level == G_RAID_VOLUME_RL_RAID3 ||
1691 	    vol->v_raid_level == G_RAID_VOLUME_RL_SINGLE ||
1692 	    vol->v_raid_level == G_RAID_VOLUME_RL_CONCAT) {
1693 		if ((disk = vol->v_subdisks[0].sd_disk) != NULL &&
1694 		    disk->d_consumer != NULL &&
1695 		    disk->d_consumer->provider != NULL) {
1696 			pp->stripesize = disk->d_consumer->provider->stripesize;
1697 			off = disk->d_consumer->provider->stripeoffset;
1698 			pp->stripeoffset = off + vol->v_subdisks[0].sd_offset;
1699 			if (off > 0)
1700 				pp->stripeoffset %= off;
1701 		}
1702 		if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID3) {
1703 			pp->stripesize *= (vol->v_disks_count - 1);
1704 			pp->stripeoffset *= (vol->v_disks_count - 1);
1705 		}
1706 	} else
1707 		pp->stripesize = vol->v_strip_size;
1708 	vol->v_provider = pp;
1709 	g_error_provider(pp, 0);
1710 	g_topology_unlock();
1711 	G_RAID_DEBUG1(0, sc, "Provider %s for volume %s created.",
1712 	    pp->name, vol->v_name);
1713 }
1714 
1715 static void
1716 g_raid_destroy_provider(struct g_raid_volume *vol)
1717 {
1718 	struct g_raid_softc *sc;
1719 	struct g_provider *pp;
1720 	struct bio *bp, *tmp;
1721 
1722 	g_topology_assert_not();
1723 	sc = vol->v_softc;
1724 	pp = vol->v_provider;
1725 	KASSERT(pp != NULL, ("NULL provider (volume=%s).", vol->v_name));
1726 
1727 	g_topology_lock();
1728 	g_error_provider(pp, ENXIO);
1729 	mtx_lock(&sc->sc_queue_mtx);
1730 	TAILQ_FOREACH_SAFE(bp, &sc->sc_queue.queue, bio_queue, tmp) {
1731 		if (bp->bio_to != pp)
1732 			continue;
1733 		bioq_remove(&sc->sc_queue, bp);
1734 		g_io_deliver(bp, ENXIO);
1735 	}
1736 	mtx_unlock(&sc->sc_queue_mtx);
1737 	G_RAID_DEBUG1(0, sc, "Provider %s for volume %s destroyed.",
1738 	    pp->name, vol->v_name);
1739 	g_wither_provider(pp, ENXIO);
1740 	g_topology_unlock();
1741 	vol->v_provider = NULL;
1742 }
1743 
1744 /*
1745  * Update device state.
1746  */
1747 static int
1748 g_raid_update_volume(struct g_raid_volume *vol, u_int event)
1749 {
1750 	struct g_raid_softc *sc;
1751 
1752 	sc = vol->v_softc;
1753 	sx_assert(&sc->sc_lock, SX_XLOCKED);
1754 
1755 	G_RAID_DEBUG1(2, sc, "Event %s for volume %s.",
1756 	    g_raid_volume_event2str(event),
1757 	    vol->v_name);
1758 	switch (event) {
1759 	case G_RAID_VOLUME_E_DOWN:
1760 		if (vol->v_provider != NULL)
1761 			g_raid_destroy_provider(vol);
1762 		break;
1763 	case G_RAID_VOLUME_E_UP:
1764 		if (vol->v_provider == NULL)
1765 			g_raid_launch_provider(vol);
1766 		break;
1767 	case G_RAID_VOLUME_E_START:
1768 		if (vol->v_tr)
1769 			G_RAID_TR_START(vol->v_tr);
1770 		return (0);
1771 	default:
1772 		if (sc->sc_md)
1773 			G_RAID_MD_VOLUME_EVENT(sc->sc_md, vol, event);
1774 		return (0);
1775 	}
1776 
1777 	/* Manage root mount release. */
1778 	if (vol->v_starting) {
1779 		vol->v_starting = 0;
1780 		G_RAID_DEBUG1(1, sc, "root_mount_rel %p", vol->v_rootmount);
1781 		root_mount_rel(vol->v_rootmount);
1782 		vol->v_rootmount = NULL;
1783 	}
1784 	if (vol->v_stopping && vol->v_provider_open == 0)
1785 		g_raid_destroy_volume(vol);
1786 	return (0);
1787 }
1788 
1789 /*
1790  * Update subdisk state.
1791  */
1792 static int
1793 g_raid_update_subdisk(struct g_raid_subdisk *sd, u_int event)
1794 {
1795 	struct g_raid_softc *sc;
1796 	struct g_raid_volume *vol;
1797 
1798 	sc = sd->sd_softc;
1799 	vol = sd->sd_volume;
1800 	sx_assert(&sc->sc_lock, SX_XLOCKED);
1801 
1802 	G_RAID_DEBUG1(2, sc, "Event %s for subdisk %s:%d-%s.",
1803 	    g_raid_subdisk_event2str(event),
1804 	    vol->v_name, sd->sd_pos,
1805 	    sd->sd_disk ? g_raid_get_diskname(sd->sd_disk) : "[none]");
1806 	if (vol->v_tr)
1807 		G_RAID_TR_EVENT(vol->v_tr, sd, event);
1808 
1809 	return (0);
1810 }
1811 
1812 /*
1813  * Update disk state.
1814  */
1815 static int
1816 g_raid_update_disk(struct g_raid_disk *disk, u_int event)
1817 {
1818 	struct g_raid_softc *sc;
1819 
1820 	sc = disk->d_softc;
1821 	sx_assert(&sc->sc_lock, SX_XLOCKED);
1822 
1823 	G_RAID_DEBUG1(2, sc, "Event %s for disk %s.",
1824 	    g_raid_disk_event2str(event),
1825 	    g_raid_get_diskname(disk));
1826 
1827 	if (sc->sc_md)
1828 		G_RAID_MD_EVENT(sc->sc_md, disk, event);
1829 	return (0);
1830 }
1831 
1832 /*
1833  * Node event.
1834  */
1835 static int
1836 g_raid_update_node(struct g_raid_softc *sc, u_int event)
1837 {
1838 	sx_assert(&sc->sc_lock, SX_XLOCKED);
1839 
1840 	G_RAID_DEBUG1(2, sc, "Event %s for the array.",
1841 	    g_raid_node_event2str(event));
1842 
1843 	if (event == G_RAID_NODE_E_WAKE)
1844 		return (0);
1845 	if (sc->sc_md)
1846 		G_RAID_MD_EVENT(sc->sc_md, NULL, event);
1847 	return (0);
1848 }
1849 
1850 static int
1851 g_raid_access(struct g_provider *pp, int acr, int acw, int ace)
1852 {
1853 	struct g_raid_volume *vol;
1854 	struct g_raid_softc *sc;
1855 	int dcw, opens, error = 0;
1856 
1857 	g_topology_assert();
1858 	sc = pp->geom->softc;
1859 	vol = pp->private;
1860 	KASSERT(sc != NULL, ("NULL softc (provider=%s).", pp->name));
1861 	KASSERT(vol != NULL, ("NULL volume (provider=%s).", pp->name));
1862 
1863 	G_RAID_DEBUG1(2, sc, "Access request for %s: r%dw%de%d.", pp->name,
1864 	    acr, acw, ace);
1865 	dcw = pp->acw + acw;
1866 
1867 	g_topology_unlock();
1868 	sx_xlock(&sc->sc_lock);
1869 	/* Deny new opens while dying. */
1870 	if (sc->sc_stopping != 0 && (acr > 0 || acw > 0 || ace > 0)) {
1871 		error = ENXIO;
1872 		goto out;
1873 	}
1874 	/* Deny write opens for read-only volumes. */
1875 	if (vol->v_read_only && acw > 0) {
1876 		error = EROFS;
1877 		goto out;
1878 	}
1879 	if (dcw == 0)
1880 		g_raid_clean(vol, dcw);
1881 	vol->v_provider_open += acr + acw + ace;
1882 	/* Handle delayed node destruction. */
1883 	if (sc->sc_stopping == G_RAID_DESTROY_DELAYED &&
1884 	    vol->v_provider_open == 0) {
1885 		/* Count open volumes. */
1886 		opens = g_raid_nopens(sc);
1887 		if (opens == 0) {
1888 			sc->sc_stopping = G_RAID_DESTROY_HARD;
1889 			/* Wake up worker to make it selfdestruct. */
1890 			g_raid_event_send(sc, G_RAID_NODE_E_WAKE, 0);
1891 		}
1892 	}
1893 	/* Handle open volume destruction. */
1894 	if (vol->v_stopping && vol->v_provider_open == 0)
1895 		g_raid_destroy_volume(vol);
1896 out:
1897 	sx_xunlock(&sc->sc_lock);
1898 	g_topology_lock();
1899 	return (error);
1900 }
1901 
1902 struct g_raid_softc *
1903 g_raid_create_node(struct g_class *mp,
1904     const char *name, struct g_raid_md_object *md)
1905 {
1906 	struct g_raid_softc *sc;
1907 	struct g_geom *gp;
1908 	int error;
1909 
1910 	g_topology_assert();
1911 	G_RAID_DEBUG(1, "Creating array %s.", name);
1912 
1913 	gp = g_new_geomf(mp, "%s", name);
1914 	sc = malloc(sizeof(*sc), M_RAID, M_WAITOK | M_ZERO);
1915 	gp->start = g_raid_start;
1916 	gp->orphan = g_raid_orphan;
1917 	gp->access = g_raid_access;
1918 	gp->dumpconf = g_raid_dumpconf;
1919 
1920 	sc->sc_md = md;
1921 	sc->sc_geom = gp;
1922 	sc->sc_flags = 0;
1923 	TAILQ_INIT(&sc->sc_volumes);
1924 	TAILQ_INIT(&sc->sc_disks);
1925 	sx_init(&sc->sc_lock, "graid:lock");
1926 	mtx_init(&sc->sc_queue_mtx, "graid:queue", NULL, MTX_DEF);
1927 	TAILQ_INIT(&sc->sc_events);
1928 	bioq_init(&sc->sc_queue);
1929 	gp->softc = sc;
1930 	error = kproc_create(g_raid_worker, sc, &sc->sc_worker, 0, 0,
1931 	    "g_raid %s", name);
1932 	if (error != 0) {
1933 		G_RAID_DEBUG(0, "Cannot create kernel thread for %s.", name);
1934 		mtx_destroy(&sc->sc_queue_mtx);
1935 		sx_destroy(&sc->sc_lock);
1936 		g_destroy_geom(sc->sc_geom);
1937 		free(sc, M_RAID);
1938 		return (NULL);
1939 	}
1940 
1941 	G_RAID_DEBUG1(0, sc, "Array %s created.", name);
1942 	return (sc);
1943 }
1944 
1945 struct g_raid_volume *
1946 g_raid_create_volume(struct g_raid_softc *sc, const char *name, int id)
1947 {
1948 	struct g_raid_volume	*vol, *vol1;
1949 	int i;
1950 
1951 	G_RAID_DEBUG1(1, sc, "Creating volume %s.", name);
1952 	vol = malloc(sizeof(*vol), M_RAID, M_WAITOK | M_ZERO);
1953 	vol->v_softc = sc;
1954 	strlcpy(vol->v_name, name, G_RAID_MAX_VOLUMENAME);
1955 	vol->v_state = G_RAID_VOLUME_S_STARTING;
1956 	vol->v_raid_level = G_RAID_VOLUME_RL_UNKNOWN;
1957 	vol->v_raid_level_qualifier = G_RAID_VOLUME_RLQ_UNKNOWN;
1958 	vol->v_rotate_parity = 1;
1959 	bioq_init(&vol->v_inflight);
1960 	bioq_init(&vol->v_locked);
1961 	LIST_INIT(&vol->v_locks);
1962 	for (i = 0; i < G_RAID_MAX_SUBDISKS; i++) {
1963 		vol->v_subdisks[i].sd_softc = sc;
1964 		vol->v_subdisks[i].sd_volume = vol;
1965 		vol->v_subdisks[i].sd_pos = i;
1966 		vol->v_subdisks[i].sd_state = G_RAID_DISK_S_NONE;
1967 	}
1968 
1969 	/* Find free ID for this volume. */
1970 	g_topology_lock();
1971 	vol1 = vol;
1972 	if (id >= 0) {
1973 		LIST_FOREACH(vol1, &g_raid_volumes, v_global_next) {
1974 			if (vol1->v_global_id == id)
1975 				break;
1976 		}
1977 	}
1978 	if (vol1 != NULL) {
1979 		for (id = 0; ; id++) {
1980 			LIST_FOREACH(vol1, &g_raid_volumes, v_global_next) {
1981 				if (vol1->v_global_id == id)
1982 					break;
1983 			}
1984 			if (vol1 == NULL)
1985 				break;
1986 		}
1987 	}
1988 	vol->v_global_id = id;
1989 	LIST_INSERT_HEAD(&g_raid_volumes, vol, v_global_next);
1990 	g_topology_unlock();
1991 
1992 	/* Delay root mounting. */
1993 	vol->v_rootmount = root_mount_hold("GRAID");
1994 	G_RAID_DEBUG1(1, sc, "root_mount_hold %p", vol->v_rootmount);
1995 	vol->v_starting = 1;
1996 	TAILQ_INSERT_TAIL(&sc->sc_volumes, vol, v_next);
1997 	return (vol);
1998 }
1999 
2000 struct g_raid_disk *
2001 g_raid_create_disk(struct g_raid_softc *sc)
2002 {
2003 	struct g_raid_disk	*disk;
2004 
2005 	G_RAID_DEBUG1(1, sc, "Creating disk.");
2006 	disk = malloc(sizeof(*disk), M_RAID, M_WAITOK | M_ZERO);
2007 	disk->d_softc = sc;
2008 	disk->d_state = G_RAID_DISK_S_NONE;
2009 	TAILQ_INIT(&disk->d_subdisks);
2010 	TAILQ_INSERT_TAIL(&sc->sc_disks, disk, d_next);
2011 	return (disk);
2012 }
2013 
2014 int g_raid_start_volume(struct g_raid_volume *vol)
2015 {
2016 	struct g_raid_tr_class *class;
2017 	struct g_raid_tr_object *obj;
2018 	int status;
2019 
2020 	G_RAID_DEBUG1(2, vol->v_softc, "Starting volume %s.", vol->v_name);
2021 	LIST_FOREACH(class, &g_raid_tr_classes, trc_list) {
2022 		if (!class->trc_enable)
2023 			continue;
2024 		G_RAID_DEBUG1(2, vol->v_softc,
2025 		    "Tasting volume %s for %s transformation.",
2026 		    vol->v_name, class->name);
2027 		obj = (void *)kobj_create((kobj_class_t)class, M_RAID,
2028 		    M_WAITOK);
2029 		obj->tro_class = class;
2030 		obj->tro_volume = vol;
2031 		status = G_RAID_TR_TASTE(obj, vol);
2032 		if (status != G_RAID_TR_TASTE_FAIL)
2033 			break;
2034 		kobj_delete((kobj_t)obj, M_RAID);
2035 	}
2036 	if (class == NULL) {
2037 		G_RAID_DEBUG1(0, vol->v_softc,
2038 		    "No transformation module found for %s.",
2039 		    vol->v_name);
2040 		vol->v_tr = NULL;
2041 		g_raid_change_volume_state(vol, G_RAID_VOLUME_S_UNSUPPORTED);
2042 		g_raid_event_send(vol, G_RAID_VOLUME_E_DOWN,
2043 		    G_RAID_EVENT_VOLUME);
2044 		return (-1);
2045 	}
2046 	G_RAID_DEBUG1(2, vol->v_softc,
2047 	    "Transformation module %s chosen for %s.",
2048 	    class->name, vol->v_name);
2049 	vol->v_tr = obj;
2050 	return (0);
2051 }
2052 
2053 int
2054 g_raid_destroy_node(struct g_raid_softc *sc, int worker)
2055 {
2056 	struct g_raid_volume *vol, *tmpv;
2057 	struct g_raid_disk *disk, *tmpd;
2058 	int error = 0;
2059 
2060 	sc->sc_stopping = G_RAID_DESTROY_HARD;
2061 	TAILQ_FOREACH_SAFE(vol, &sc->sc_volumes, v_next, tmpv) {
2062 		if (g_raid_destroy_volume(vol))
2063 			error = EBUSY;
2064 	}
2065 	if (error)
2066 		return (error);
2067 	TAILQ_FOREACH_SAFE(disk, &sc->sc_disks, d_next, tmpd) {
2068 		if (g_raid_destroy_disk(disk))
2069 			error = EBUSY;
2070 	}
2071 	if (error)
2072 		return (error);
2073 	if (sc->sc_md) {
2074 		G_RAID_MD_FREE(sc->sc_md);
2075 		kobj_delete((kobj_t)sc->sc_md, M_RAID);
2076 		sc->sc_md = NULL;
2077 	}
2078 	if (sc->sc_geom != NULL) {
2079 		G_RAID_DEBUG1(0, sc, "Array %s destroyed.", sc->sc_name);
2080 		g_topology_lock();
2081 		sc->sc_geom->softc = NULL;
2082 		g_wither_geom(sc->sc_geom, ENXIO);
2083 		g_topology_unlock();
2084 		sc->sc_geom = NULL;
2085 	} else
2086 		G_RAID_DEBUG(1, "Array destroyed.");
2087 	if (worker) {
2088 		g_raid_event_cancel(sc, sc);
2089 		mtx_destroy(&sc->sc_queue_mtx);
2090 		sx_xunlock(&sc->sc_lock);
2091 		sx_destroy(&sc->sc_lock);
2092 		wakeup(&sc->sc_stopping);
2093 		free(sc, M_RAID);
2094 		curthread->td_pflags &= ~TDP_GEOM;
2095 		G_RAID_DEBUG(1, "Thread exiting.");
2096 		kproc_exit(0);
2097 	} else {
2098 		/* Wake up worker to make it selfdestruct. */
2099 		g_raid_event_send(sc, G_RAID_NODE_E_WAKE, 0);
2100 	}
2101 	return (0);
2102 }
2103 
2104 int
2105 g_raid_destroy_volume(struct g_raid_volume *vol)
2106 {
2107 	struct g_raid_softc *sc;
2108 	struct g_raid_disk *disk;
2109 	int i;
2110 
2111 	sc = vol->v_softc;
2112 	G_RAID_DEBUG1(2, sc, "Destroying volume %s.", vol->v_name);
2113 	vol->v_stopping = 1;
2114 	if (vol->v_state != G_RAID_VOLUME_S_STOPPED) {
2115 		if (vol->v_tr) {
2116 			G_RAID_TR_STOP(vol->v_tr);
2117 			return (EBUSY);
2118 		} else
2119 			vol->v_state = G_RAID_VOLUME_S_STOPPED;
2120 	}
2121 	if (g_raid_event_check(sc, vol) != 0)
2122 		return (EBUSY);
2123 	if (vol->v_provider != NULL)
2124 		return (EBUSY);
2125 	if (vol->v_provider_open != 0)
2126 		return (EBUSY);
2127 	if (vol->v_tr) {
2128 		G_RAID_TR_FREE(vol->v_tr);
2129 		kobj_delete((kobj_t)vol->v_tr, M_RAID);
2130 		vol->v_tr = NULL;
2131 	}
2132 	if (vol->v_rootmount)
2133 		root_mount_rel(vol->v_rootmount);
2134 	g_topology_lock();
2135 	LIST_REMOVE(vol, v_global_next);
2136 	g_topology_unlock();
2137 	TAILQ_REMOVE(&sc->sc_volumes, vol, v_next);
2138 	for (i = 0; i < G_RAID_MAX_SUBDISKS; i++) {
2139 		g_raid_event_cancel(sc, &vol->v_subdisks[i]);
2140 		disk = vol->v_subdisks[i].sd_disk;
2141 		if (disk == NULL)
2142 			continue;
2143 		TAILQ_REMOVE(&disk->d_subdisks, &vol->v_subdisks[i], sd_next);
2144 	}
2145 	G_RAID_DEBUG1(2, sc, "Volume %s destroyed.", vol->v_name);
2146 	if (sc->sc_md)
2147 		G_RAID_MD_FREE_VOLUME(sc->sc_md, vol);
2148 	g_raid_event_cancel(sc, vol);
2149 	free(vol, M_RAID);
2150 	if (sc->sc_stopping == G_RAID_DESTROY_HARD) {
2151 		/* Wake up worker to let it selfdestruct. */
2152 		g_raid_event_send(sc, G_RAID_NODE_E_WAKE, 0);
2153 	}
2154 	return (0);
2155 }
2156 
2157 int
2158 g_raid_destroy_disk(struct g_raid_disk *disk)
2159 {
2160 	struct g_raid_softc *sc;
2161 	struct g_raid_subdisk *sd, *tmp;
2162 
2163 	sc = disk->d_softc;
2164 	G_RAID_DEBUG1(2, sc, "Destroying disk.");
2165 	if (disk->d_consumer) {
2166 		g_raid_kill_consumer(sc, disk->d_consumer);
2167 		disk->d_consumer = NULL;
2168 	}
2169 	TAILQ_FOREACH_SAFE(sd, &disk->d_subdisks, sd_next, tmp) {
2170 		g_raid_change_subdisk_state(sd, G_RAID_SUBDISK_S_NONE);
2171 		g_raid_event_send(sd, G_RAID_SUBDISK_E_DISCONNECTED,
2172 		    G_RAID_EVENT_SUBDISK);
2173 		TAILQ_REMOVE(&disk->d_subdisks, sd, sd_next);
2174 		sd->sd_disk = NULL;
2175 	}
2176 	TAILQ_REMOVE(&sc->sc_disks, disk, d_next);
2177 	if (sc->sc_md)
2178 		G_RAID_MD_FREE_DISK(sc->sc_md, disk);
2179 	g_raid_event_cancel(sc, disk);
2180 	free(disk, M_RAID);
2181 	return (0);
2182 }
2183 
2184 int
2185 g_raid_destroy(struct g_raid_softc *sc, int how)
2186 {
2187 	int error, opens;
2188 
2189 	g_topology_assert_not();
2190 	if (sc == NULL)
2191 		return (ENXIO);
2192 	sx_assert(&sc->sc_lock, SX_XLOCKED);
2193 
2194 	/* Count open volumes. */
2195 	opens = g_raid_nopens(sc);
2196 
2197 	/* React on some opened volumes. */
2198 	if (opens > 0) {
2199 		switch (how) {
2200 		case G_RAID_DESTROY_SOFT:
2201 			G_RAID_DEBUG1(1, sc,
2202 			    "%d volumes are still open.",
2203 			    opens);
2204 			sx_xunlock(&sc->sc_lock);
2205 			return (EBUSY);
2206 		case G_RAID_DESTROY_DELAYED:
2207 			G_RAID_DEBUG1(1, sc,
2208 			    "Array will be destroyed on last close.");
2209 			sc->sc_stopping = G_RAID_DESTROY_DELAYED;
2210 			sx_xunlock(&sc->sc_lock);
2211 			return (EBUSY);
2212 		case G_RAID_DESTROY_HARD:
2213 			G_RAID_DEBUG1(1, sc,
2214 			    "%d volumes are still open.",
2215 			    opens);
2216 		}
2217 	}
2218 
2219 	/* Mark node for destruction. */
2220 	sc->sc_stopping = G_RAID_DESTROY_HARD;
2221 	/* Wake up worker to let it selfdestruct. */
2222 	g_raid_event_send(sc, G_RAID_NODE_E_WAKE, 0);
2223 	/* Sleep until node destroyed. */
2224 	error = sx_sleep(&sc->sc_stopping, &sc->sc_lock,
2225 	    PRIBIO | PDROP, "r:destroy", hz * 3);
2226 	return (error == EWOULDBLOCK ? EBUSY : 0);
2227 }
2228 
2229 static void
2230 g_raid_taste_orphan(struct g_consumer *cp)
2231 {
2232 
2233 	KASSERT(1 == 0, ("%s called while tasting %s.", __func__,
2234 	    cp->provider->name));
2235 }
2236 
2237 static struct g_geom *
2238 g_raid_taste(struct g_class *mp, struct g_provider *pp, int flags __unused)
2239 {
2240 	struct g_consumer *cp;
2241 	struct g_geom *gp, *geom;
2242 	struct g_raid_md_class *class;
2243 	struct g_raid_md_object *obj;
2244 	int status;
2245 
2246 	g_topology_assert();
2247 	g_trace(G_T_TOPOLOGY, "%s(%s, %s)", __func__, mp->name, pp->name);
2248 	if (!g_raid_enable)
2249 		return (NULL);
2250 	G_RAID_DEBUG(2, "Tasting provider %s.", pp->name);
2251 
2252 	gp = g_new_geomf(mp, "raid:taste");
2253 	/*
2254 	 * This orphan function should be never called.
2255 	 */
2256 	gp->orphan = g_raid_taste_orphan;
2257 	cp = g_new_consumer(gp);
2258 	g_attach(cp, pp);
2259 
2260 	geom = NULL;
2261 	LIST_FOREACH(class, &g_raid_md_classes, mdc_list) {
2262 		if (!class->mdc_enable)
2263 			continue;
2264 		G_RAID_DEBUG(2, "Tasting provider %s for %s metadata.",
2265 		    pp->name, class->name);
2266 		obj = (void *)kobj_create((kobj_class_t)class, M_RAID,
2267 		    M_WAITOK);
2268 		obj->mdo_class = class;
2269 		status = G_RAID_MD_TASTE(obj, mp, cp, &geom);
2270 		if (status != G_RAID_MD_TASTE_NEW)
2271 			kobj_delete((kobj_t)obj, M_RAID);
2272 		if (status != G_RAID_MD_TASTE_FAIL)
2273 			break;
2274 	}
2275 
2276 	g_detach(cp);
2277 	g_destroy_consumer(cp);
2278 	g_destroy_geom(gp);
2279 	G_RAID_DEBUG(2, "Tasting provider %s done.", pp->name);
2280 	return (geom);
2281 }
2282 
2283 int
2284 g_raid_create_node_format(const char *format, struct gctl_req *req,
2285     struct g_geom **gp)
2286 {
2287 	struct g_raid_md_class *class;
2288 	struct g_raid_md_object *obj;
2289 	int status;
2290 
2291 	G_RAID_DEBUG(2, "Creating array for %s metadata.", format);
2292 	LIST_FOREACH(class, &g_raid_md_classes, mdc_list) {
2293 		if (strcasecmp(class->name, format) == 0)
2294 			break;
2295 	}
2296 	if (class == NULL) {
2297 		G_RAID_DEBUG(1, "No support for %s metadata.", format);
2298 		return (G_RAID_MD_TASTE_FAIL);
2299 	}
2300 	obj = (void *)kobj_create((kobj_class_t)class, M_RAID,
2301 	    M_WAITOK);
2302 	obj->mdo_class = class;
2303 	status = G_RAID_MD_CREATE_REQ(obj, &g_raid_class, req, gp);
2304 	if (status != G_RAID_MD_TASTE_NEW)
2305 		kobj_delete((kobj_t)obj, M_RAID);
2306 	return (status);
2307 }
2308 
2309 static int
2310 g_raid_destroy_geom(struct gctl_req *req __unused,
2311     struct g_class *mp __unused, struct g_geom *gp)
2312 {
2313 	struct g_raid_softc *sc;
2314 	int error;
2315 
2316 	g_topology_unlock();
2317 	sc = gp->softc;
2318 	sx_xlock(&sc->sc_lock);
2319 	g_cancel_event(sc);
2320 	error = g_raid_destroy(gp->softc, G_RAID_DESTROY_SOFT);
2321 	g_topology_lock();
2322 	return (error);
2323 }
2324 
2325 void g_raid_write_metadata(struct g_raid_softc *sc, struct g_raid_volume *vol,
2326     struct g_raid_subdisk *sd, struct g_raid_disk *disk)
2327 {
2328 
2329 	if (sc->sc_stopping == G_RAID_DESTROY_HARD)
2330 		return;
2331 	if (sc->sc_md)
2332 		G_RAID_MD_WRITE(sc->sc_md, vol, sd, disk);
2333 }
2334 
2335 void g_raid_fail_disk(struct g_raid_softc *sc,
2336     struct g_raid_subdisk *sd, struct g_raid_disk *disk)
2337 {
2338 
2339 	if (disk == NULL)
2340 		disk = sd->sd_disk;
2341 	if (disk == NULL) {
2342 		G_RAID_DEBUG1(0, sc, "Warning! Fail request to an absent disk!");
2343 		return;
2344 	}
2345 	if (disk->d_state != G_RAID_DISK_S_ACTIVE) {
2346 		G_RAID_DEBUG1(0, sc, "Warning! Fail request to a disk in a "
2347 		    "wrong state (%s)!", g_raid_disk_state2str(disk->d_state));
2348 		return;
2349 	}
2350 	if (sc->sc_md)
2351 		G_RAID_MD_FAIL_DISK(sc->sc_md, sd, disk);
2352 }
2353 
2354 static void
2355 g_raid_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp,
2356     struct g_consumer *cp, struct g_provider *pp)
2357 {
2358 	struct g_raid_softc *sc;
2359 	struct g_raid_volume *vol;
2360 	struct g_raid_subdisk *sd;
2361 	struct g_raid_disk *disk;
2362 	int i, s;
2363 
2364 	g_topology_assert();
2365 
2366 	sc = gp->softc;
2367 	if (sc == NULL)
2368 		return;
2369 	if (pp != NULL) {
2370 		vol = pp->private;
2371 		g_topology_unlock();
2372 		sx_xlock(&sc->sc_lock);
2373 		sbuf_printf(sb, "%s<descr>%s %s volume</descr>\n", indent,
2374 		    sc->sc_md->mdo_class->name,
2375 		    g_raid_volume_level2str(vol->v_raid_level,
2376 		    vol->v_raid_level_qualifier));
2377 		sbuf_printf(sb, "%s<Label>%s</Label>\n", indent,
2378 		    vol->v_name);
2379 		sbuf_printf(sb, "%s<RAIDLevel>%s</RAIDLevel>\n", indent,
2380 		    g_raid_volume_level2str(vol->v_raid_level,
2381 		    vol->v_raid_level_qualifier));
2382 		sbuf_printf(sb,
2383 		    "%s<Transformation>%s</Transformation>\n", indent,
2384 		    vol->v_tr ? vol->v_tr->tro_class->name : "NONE");
2385 		sbuf_printf(sb, "%s<Components>%u</Components>\n", indent,
2386 		    vol->v_disks_count);
2387 		sbuf_printf(sb, "%s<Strip>%u</Strip>\n", indent,
2388 		    vol->v_strip_size);
2389 		sbuf_printf(sb, "%s<State>%s</State>\n", indent,
2390 		    g_raid_volume_state2str(vol->v_state));
2391 		sbuf_printf(sb, "%s<Dirty>%s</Dirty>\n", indent,
2392 		    vol->v_dirty ? "Yes" : "No");
2393 		sbuf_printf(sb, "%s<Subdisks>", indent);
2394 		for (i = 0; i < vol->v_disks_count; i++) {
2395 			sd = &vol->v_subdisks[i];
2396 			if (sd->sd_disk != NULL &&
2397 			    sd->sd_disk->d_consumer != NULL) {
2398 				sbuf_printf(sb, "%s ",
2399 				    g_raid_get_diskname(sd->sd_disk));
2400 			} else {
2401 				sbuf_printf(sb, "NONE ");
2402 			}
2403 			sbuf_printf(sb, "(%s",
2404 			    g_raid_subdisk_state2str(sd->sd_state));
2405 			if (sd->sd_state == G_RAID_SUBDISK_S_REBUILD ||
2406 			    sd->sd_state == G_RAID_SUBDISK_S_RESYNC) {
2407 				sbuf_printf(sb, " %d%%",
2408 				    (int)(sd->sd_rebuild_pos * 100 /
2409 				     sd->sd_size));
2410 			}
2411 			sbuf_printf(sb, ")");
2412 			if (i + 1 < vol->v_disks_count)
2413 				sbuf_printf(sb, ", ");
2414 		}
2415 		sbuf_printf(sb, "</Subdisks>\n");
2416 		sx_xunlock(&sc->sc_lock);
2417 		g_topology_lock();
2418 	} else if (cp != NULL) {
2419 		disk = cp->private;
2420 		if (disk == NULL)
2421 			return;
2422 		g_topology_unlock();
2423 		sx_xlock(&sc->sc_lock);
2424 		sbuf_printf(sb, "%s<State>%s", indent,
2425 		    g_raid_disk_state2str(disk->d_state));
2426 		if (!TAILQ_EMPTY(&disk->d_subdisks)) {
2427 			sbuf_printf(sb, " (");
2428 			TAILQ_FOREACH(sd, &disk->d_subdisks, sd_next) {
2429 				sbuf_printf(sb, "%s",
2430 				    g_raid_subdisk_state2str(sd->sd_state));
2431 				if (sd->sd_state == G_RAID_SUBDISK_S_REBUILD ||
2432 				    sd->sd_state == G_RAID_SUBDISK_S_RESYNC) {
2433 					sbuf_printf(sb, " %d%%",
2434 					    (int)(sd->sd_rebuild_pos * 100 /
2435 					     sd->sd_size));
2436 				}
2437 				if (TAILQ_NEXT(sd, sd_next))
2438 					sbuf_printf(sb, ", ");
2439 			}
2440 			sbuf_printf(sb, ")");
2441 		}
2442 		sbuf_printf(sb, "</State>\n");
2443 		sbuf_printf(sb, "%s<Subdisks>", indent);
2444 		TAILQ_FOREACH(sd, &disk->d_subdisks, sd_next) {
2445 			sbuf_printf(sb, "r%d(%s):%d@%ju",
2446 			    sd->sd_volume->v_global_id,
2447 			    sd->sd_volume->v_name,
2448 			    sd->sd_pos, sd->sd_offset);
2449 			if (TAILQ_NEXT(sd, sd_next))
2450 				sbuf_printf(sb, ", ");
2451 		}
2452 		sbuf_printf(sb, "</Subdisks>\n");
2453 		sbuf_printf(sb, "%s<ReadErrors>%d</ReadErrors>\n", indent,
2454 		    disk->d_read_errs);
2455 		sx_xunlock(&sc->sc_lock);
2456 		g_topology_lock();
2457 	} else {
2458 		g_topology_unlock();
2459 		sx_xlock(&sc->sc_lock);
2460 		if (sc->sc_md) {
2461 			sbuf_printf(sb, "%s<Metadata>%s</Metadata>\n", indent,
2462 			    sc->sc_md->mdo_class->name);
2463 		}
2464 		if (!TAILQ_EMPTY(&sc->sc_volumes)) {
2465 			s = 0xff;
2466 			TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) {
2467 				if (vol->v_state < s)
2468 					s = vol->v_state;
2469 			}
2470 			sbuf_printf(sb, "%s<State>%s</State>\n", indent,
2471 			    g_raid_volume_state2str(s));
2472 		}
2473 		sx_xunlock(&sc->sc_lock);
2474 		g_topology_lock();
2475 	}
2476 }
2477 
2478 static void
2479 g_raid_shutdown_post_sync(void *arg, int howto)
2480 {
2481 	struct g_class *mp;
2482 	struct g_geom *gp, *gp2;
2483 	struct g_raid_softc *sc;
2484 	struct g_raid_volume *vol;
2485 
2486 	mp = arg;
2487 	DROP_GIANT();
2488 	g_topology_lock();
2489 	g_raid_shutdown = 1;
2490 	LIST_FOREACH_SAFE(gp, &mp->geom, geom, gp2) {
2491 		if ((sc = gp->softc) == NULL)
2492 			continue;
2493 		g_topology_unlock();
2494 		sx_xlock(&sc->sc_lock);
2495 		TAILQ_FOREACH(vol, &sc->sc_volumes, v_next)
2496 			g_raid_clean(vol, -1);
2497 		g_cancel_event(sc);
2498 		g_raid_destroy(sc, G_RAID_DESTROY_DELAYED);
2499 		g_topology_lock();
2500 	}
2501 	g_topology_unlock();
2502 	PICKUP_GIANT();
2503 }
2504 
2505 static void
2506 g_raid_init(struct g_class *mp)
2507 {
2508 
2509 	g_raid_post_sync = EVENTHANDLER_REGISTER(shutdown_post_sync,
2510 	    g_raid_shutdown_post_sync, mp, SHUTDOWN_PRI_FIRST);
2511 	if (g_raid_post_sync == NULL)
2512 		G_RAID_DEBUG(0, "Warning! Cannot register shutdown event.");
2513 	g_raid_started = 1;
2514 }
2515 
2516 static void
2517 g_raid_fini(struct g_class *mp)
2518 {
2519 
2520 	if (g_raid_post_sync != NULL)
2521 		EVENTHANDLER_DEREGISTER(shutdown_post_sync, g_raid_post_sync);
2522 	g_raid_started = 0;
2523 }
2524 
2525 int
2526 g_raid_md_modevent(module_t mod, int type, void *arg)
2527 {
2528 	struct g_raid_md_class *class, *c, *nc;
2529 	int error;
2530 
2531 	error = 0;
2532 	class = arg;
2533 	switch (type) {
2534 	case MOD_LOAD:
2535 		c = LIST_FIRST(&g_raid_md_classes);
2536 		if (c == NULL || c->mdc_priority > class->mdc_priority)
2537 			LIST_INSERT_HEAD(&g_raid_md_classes, class, mdc_list);
2538 		else {
2539 			while ((nc = LIST_NEXT(c, mdc_list)) != NULL &&
2540 			    nc->mdc_priority < class->mdc_priority)
2541 				c = nc;
2542 			LIST_INSERT_AFTER(c, class, mdc_list);
2543 		}
2544 		if (g_raid_started)
2545 			g_retaste(&g_raid_class);
2546 		break;
2547 	case MOD_UNLOAD:
2548 		LIST_REMOVE(class, mdc_list);
2549 		break;
2550 	default:
2551 		error = EOPNOTSUPP;
2552 		break;
2553 	}
2554 
2555 	return (error);
2556 }
2557 
2558 int
2559 g_raid_tr_modevent(module_t mod, int type, void *arg)
2560 {
2561 	struct g_raid_tr_class *class, *c, *nc;
2562 	int error;
2563 
2564 	error = 0;
2565 	class = arg;
2566 	switch (type) {
2567 	case MOD_LOAD:
2568 		c = LIST_FIRST(&g_raid_tr_classes);
2569 		if (c == NULL || c->trc_priority > class->trc_priority)
2570 			LIST_INSERT_HEAD(&g_raid_tr_classes, class, trc_list);
2571 		else {
2572 			while ((nc = LIST_NEXT(c, trc_list)) != NULL &&
2573 			    nc->trc_priority < class->trc_priority)
2574 				c = nc;
2575 			LIST_INSERT_AFTER(c, class, trc_list);
2576 		}
2577 		break;
2578 	case MOD_UNLOAD:
2579 		LIST_REMOVE(class, trc_list);
2580 		break;
2581 	default:
2582 		error = EOPNOTSUPP;
2583 		break;
2584 	}
2585 
2586 	return (error);
2587 }
2588 
2589 /*
2590  * Use local implementation of DECLARE_GEOM_CLASS(g_raid_class, g_raid)
2591  * to reduce module priority, allowing submodules to register them first.
2592  */
2593 static moduledata_t g_raid_mod = {
2594 	"g_raid",
2595 	g_modevent,
2596 	&g_raid_class
2597 };
2598 DECLARE_MODULE(g_raid, g_raid_mod, SI_SUB_DRIVERS, SI_ORDER_THIRD);
2599 MODULE_VERSION(geom_raid, 0);
2600