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