sys/geom/raid/g_raid.c

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