sys/geom/raid/tr_concat.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/bio.h>
  32 #include <sys/endian.h>
  33 #include <sys/kernel.h>
  34 #include <sys/kobj.h>
  35 #include <sys/lock.h>
  36 #include <sys/malloc.h>
  37 #include <sys/mutex.h>
  38 #include <sys/systm.h>
  39 #include <geom/geom.h>
  40 #include "geom/raid/g_raid.h"
  41 #include "g_raid_tr_if.h"
  42
  43 static MALLOC_DEFINE(M_TR_CONCAT, "tr_concat_data", "GEOM_RAID CONCAT data");
  44
  45 struct g_raid_tr_concat_object {
  46         struct g_raid_tr_object  trso_base;
  47         int                      trso_starting;
  48         int                      trso_stopped;
  49 };
  50
  51 static g_raid_tr_taste_t g_raid_tr_taste_concat;
  52 static g_raid_tr_event_t g_raid_tr_event_concat;
  53 static g_raid_tr_start_t g_raid_tr_start_concat;
  54 static g_raid_tr_stop_t g_raid_tr_stop_concat;
  55 static g_raid_tr_iostart_t g_raid_tr_iostart_concat;
  56 static g_raid_tr_iodone_t g_raid_tr_iodone_concat;
  57 static g_raid_tr_kerneldump_t g_raid_tr_kerneldump_concat;
  58 static g_raid_tr_free_t g_raid_tr_free_concat;
  59
  60 static kobj_method_t g_raid_tr_concat_methods[] = {
  61         KOBJMETHOD(g_raid_tr_taste,     g_raid_tr_taste_concat),
  62         KOBJMETHOD(g_raid_tr_event,     g_raid_tr_event_concat),
  63         KOBJMETHOD(g_raid_tr_start,     g_raid_tr_start_concat),
  64         KOBJMETHOD(g_raid_tr_stop,      g_raid_tr_stop_concat),
  65         KOBJMETHOD(g_raid_tr_iostart,   g_raid_tr_iostart_concat),
  66         KOBJMETHOD(g_raid_tr_iodone,    g_raid_tr_iodone_concat),
  67         KOBJMETHOD(g_raid_tr_kerneldump,        g_raid_tr_kerneldump_concat),
  68         KOBJMETHOD(g_raid_tr_free,      g_raid_tr_free_concat),
  69         { 0, 0 }
  70 };
  71
  72 static struct g_raid_tr_class g_raid_tr_concat_class = {
  73         "CONCAT",
  74         g_raid_tr_concat_methods,
  75         sizeof(struct g_raid_tr_concat_object),
  76         .trc_enable = 1,
  77         .trc_priority = 50
  78 };
  79
  80 static int
  81 g_raid_tr_taste_concat(struct g_raid_tr_object *tr, struct g_raid_volume *volume)
  82 {
  83         struct g_raid_tr_concat_object *trs;
  84
  85         trs = (struct g_raid_tr_concat_object *)tr;
  86         if (tr->tro_volume->v_raid_level != G_RAID_VOLUME_RL_SINGLE &&
  87             tr->tro_volume->v_raid_level != G_RAID_VOLUME_RL_CONCAT &&
  88             !(tr->tro_volume->v_disks_count == 1 &&
  89               tr->tro_volume->v_raid_level != G_RAID_VOLUME_RL_UNKNOWN))
  90                 return (G_RAID_TR_TASTE_FAIL);
  91         trs->trso_starting = 1;
  92         return (G_RAID_TR_TASTE_SUCCEED);
  93 }
  94
  95 static int
  96 g_raid_tr_update_state_concat(struct g_raid_volume *vol)
  97 {
  98         struct g_raid_tr_concat_object *trs;
  99         struct g_raid_softc *sc;
 100         off_t size;
 101         u_int s;
 102         int i, n, f;
 103
 104         sc = vol->v_softc;
 105         trs = (struct g_raid_tr_concat_object *)vol->v_tr;
 106         if (trs->trso_stopped)
 107                 s = G_RAID_VOLUME_S_STOPPED;
 108         else if (trs->trso_starting)
 109                 s = G_RAID_VOLUME_S_STARTING;
 110         else {
 111                 n = g_raid_nsubdisks(vol, G_RAID_SUBDISK_S_ACTIVE);
 112                 f = g_raid_nsubdisks(vol, G_RAID_SUBDISK_S_FAILED);
 113                 if (n + f == vol->v_disks_count) {
 114                         if (f == 0)
 115                                 s = G_RAID_VOLUME_S_OPTIMAL;
 116                         else
 117                                 s = G_RAID_VOLUME_S_SUBOPTIMAL;
 118                 } else
 119                         s = G_RAID_VOLUME_S_BROKEN;
 120         }
 121         if (s != vol->v_state) {
 122
 123                 /*
 124                  * Some metadata modules may not know CONCAT volume
 125                  * mediasize until all disks connected. Recalculate.
 126                  */
 127                 if (G_RAID_VOLUME_S_ALIVE(s) &&
 128                     !G_RAID_VOLUME_S_ALIVE(vol->v_state)) {
 129                         size = 0;
 130                         for (i = 0; i < vol->v_disks_count; i++) {
 131                                 if (vol->v_subdisks[i].sd_state !=
 132                                     G_RAID_SUBDISK_S_NONE)
 133                                         size += vol->v_subdisks[i].sd_size;
 134                         }
 135                         vol->v_mediasize = size;
 136                 }
 137
 138                 g_raid_event_send(vol, G_RAID_VOLUME_S_ALIVE(s) ?
 139                     G_RAID_VOLUME_E_UP : G_RAID_VOLUME_E_DOWN,
 140                     G_RAID_EVENT_VOLUME);
 141                 g_raid_change_volume_state(vol, s);
 142                 if (!trs->trso_starting && !trs->trso_stopped)
 143                         g_raid_write_metadata(sc, vol, NULL, NULL);
 144         }
 145         return (0);
 146 }
 147
 148 static int
 149 g_raid_tr_event_concat(struct g_raid_tr_object *tr,
 150     struct g_raid_subdisk *sd, u_int event)
 151 {
 152         struct g_raid_tr_concat_object *trs;
 153         struct g_raid_softc *sc;
 154         struct g_raid_volume *vol;
 155         int state;
 156
 157         trs = (struct g_raid_tr_concat_object *)tr;
 158         vol = tr->tro_volume;
 159         sc = vol->v_softc;
 160
 161         state = sd->sd_state;
 162         if (state != G_RAID_SUBDISK_S_NONE &&
 163             state != G_RAID_SUBDISK_S_FAILED &&
 164             state != G_RAID_SUBDISK_S_ACTIVE) {
 165                 G_RAID_DEBUG1(1, sc,
 166                     "Promote subdisk %s:%d from %s to ACTIVE.",
 167                     vol->v_name, sd->sd_pos,
 168                     g_raid_subdisk_state2str(sd->sd_state));
 169                 g_raid_change_subdisk_state(sd, G_RAID_SUBDISK_S_ACTIVE);
 170         }
 171         if (state != sd->sd_state &&
 172             !trs->trso_starting && !trs->trso_stopped)
 173                 g_raid_write_metadata(sc, vol, sd, NULL);
 174         g_raid_tr_update_state_concat(vol);
 175         return (0);
 176 }
 177
 178 static int
 179 g_raid_tr_start_concat(struct g_raid_tr_object *tr)
 180 {
 181         struct g_raid_tr_concat_object *trs;
 182         struct g_raid_volume *vol;
 183
 184         trs = (struct g_raid_tr_concat_object *)tr;
 185         vol = tr->tro_volume;
 186         trs->trso_starting = 0;
 187         g_raid_tr_update_state_concat(vol);
 188         return (0);
 189 }
 190
 191 static int
 192 g_raid_tr_stop_concat(struct g_raid_tr_object *tr)
 193 {
 194         struct g_raid_tr_concat_object *trs;
 195         struct g_raid_volume *vol;
 196
 197         trs = (struct g_raid_tr_concat_object *)tr;
 198         vol = tr->tro_volume;
 199         trs->trso_starting = 0;
 200         trs->trso_stopped = 1;
 201         g_raid_tr_update_state_concat(vol);
 202         return (0);
 203 }
 204
 205 static void
 206 g_raid_tr_iostart_concat(struct g_raid_tr_object *tr, struct bio *bp)
 207 {
 208         struct g_raid_volume *vol;
 209         struct g_raid_subdisk *sd;
 210         struct bio_queue_head queue;
 211         struct bio *cbp;
 212         char *addr;
 213         off_t offset, length, remain;
 214         u_int no;
 215
 216         vol = tr->tro_volume;
 217         if (vol->v_state != G_RAID_VOLUME_S_OPTIMAL &&
 218             vol->v_state != G_RAID_VOLUME_S_SUBOPTIMAL) {
 219                 g_raid_iodone(bp, EIO);
 220                 return;
 221         }
 222         if (bp->bio_cmd == BIO_FLUSH) {
 223                 g_raid_tr_flush_common(tr, bp);
 224                 return;
 225         }
 226
 227         offset = bp->bio_offset;
 228         remain = bp->bio_length;
 229         addr = bp->bio_data;
 230         no = 0;
 231         while (no < vol->v_disks_count &&
 232             offset >= vol->v_subdisks[no].sd_size) {
 233                 offset -= vol->v_subdisks[no].sd_size;
 234                 no++;
 235         }
 236         KASSERT(no < vol->v_disks_count,
 237             ("Request starts after volume end (%ju)", bp->bio_offset));
 238         bioq_init(&queue);
 239         do {
 240                 sd = &vol->v_subdisks[no];
 241                 length = MIN(sd->sd_size - offset, remain);
 242                 cbp = g_clone_bio(bp);
 243                 if (cbp == NULL)
 244                         goto failure;
 245                 cbp->bio_offset = offset;
 246                 cbp->bio_data = addr;
 247                 cbp->bio_length = length;
 248                 cbp->bio_caller1 = sd;
 249                 bioq_insert_tail(&queue, cbp);
 250                 remain -= length;
 251                 addr += length;
 252                 offset = 0;
 253                 no++;
 254                 KASSERT(no < vol->v_disks_count || remain == 0,
 255                     ("Request ends after volume end (%ju, %ju)",
 256                         bp->bio_offset, bp->bio_length));
 257         } while (remain > 0);
 258         for (cbp = bioq_first(&queue); cbp != NULL;
 259             cbp = bioq_first(&queue)) {
 260                 bioq_remove(&queue, cbp);
 261                 sd = cbp->bio_caller1;
 262                 cbp->bio_caller1 = NULL;
 263                 g_raid_subdisk_iostart(sd, cbp);
 264         }
 265         return;
 266 failure:
 267         for (cbp = bioq_first(&queue); cbp != NULL;
 268             cbp = bioq_first(&queue)) {
 269                 bioq_remove(&queue, cbp);
 270                 g_destroy_bio(cbp);
 271         }
 272         if (bp->bio_error == 0)
 273                 bp->bio_error = ENOMEM;
 274         g_raid_iodone(bp, bp->bio_error);
 275 }
 276
 277 static int
 278 g_raid_tr_kerneldump_concat(struct g_raid_tr_object *tr,
 279     void *virtual, vm_offset_t physical, off_t boffset, size_t blength)
 280 {
 281         struct g_raid_volume *vol;
 282         struct g_raid_subdisk *sd;
 283         char *addr;
 284         off_t offset, length, remain;
 285         int error, no;
 286
 287         vol = tr->tro_volume;
 288         if (vol->v_state != G_RAID_VOLUME_S_OPTIMAL)
 289                 return (ENXIO);
 290
 291         offset = boffset;
 292         remain = blength;
 293         addr = virtual;
 294         no = 0;
 295         while (no < vol->v_disks_count &&
 296             offset >= vol->v_subdisks[no].sd_size) {
 297                 offset -= vol->v_subdisks[no].sd_size;
 298                 no++;
 299         }
 300         KASSERT(no < vol->v_disks_count,
 301             ("Request starts after volume end (%ju)", boffset));
 302         do {
 303                 sd = &vol->v_subdisks[no];
 304                 length = MIN(sd->sd_size - offset, remain);
 305                 error = g_raid_subdisk_kerneldump(&vol->v_subdisks[no],
 306                     addr, 0, offset, length);
 307                 if (error != 0)
 308                         return (error);
 309                 remain -= length;
 310                 addr += length;
 311                 offset = 0;
 312                 no++;
 313                 KASSERT(no < vol->v_disks_count || remain == 0,
 314                     ("Request ends after volume end (%ju, %zu)",
 315                         boffset, blength));
 316         } while (remain > 0);
 317         return (0);
 318 }
 319
 320 static void
 321 g_raid_tr_iodone_concat(struct g_raid_tr_object *tr,
 322     struct g_raid_subdisk *sd,struct bio *bp)
 323 {
 324         struct bio *pbp;
 325
 326         pbp = bp->bio_parent;
 327         if (pbp->bio_error == 0)
 328                 pbp->bio_error = bp->bio_error;
 329         g_destroy_bio(bp);
 330         pbp->bio_inbed++;
 331         if (pbp->bio_children == pbp->bio_inbed) {
 332                 pbp->bio_completed = pbp->bio_length;
 333                 g_raid_iodone(pbp, bp->bio_error);
 334         }
 335 }
 336
 337 static int
 338 g_raid_tr_free_concat(struct g_raid_tr_object *tr)
 339 {
 340
 341         return (0);
 342 }
 343
 344 G_RAID_TR_DECLARE(concat, "CONCAT");