sys/geom/raid/tr_raid5.c

   1 /*-
   2  * Copyright (c) 2012 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/limits.h>
  36 #include <sys/lock.h>
  37 #include <sys/malloc.h>
  38 #include <sys/mutex.h>
  39 #include <sys/sysctl.h>
  40 #include <sys/systm.h>
  41 #include <geom/geom.h>
  42 #include "geom/raid/g_raid.h"
  43 #include "g_raid_tr_if.h"
  44
  45 static MALLOC_DEFINE(M_TR_RAID5, "tr_raid5_data", "GEOM_RAID RAID5 data");
  46
  47 #define TR_RAID5_NONE 0
  48 #define TR_RAID5_REBUILD 1
  49 #define TR_RAID5_RESYNC 2
  50
  51 #define TR_RAID5_F_DOING_SOME   0x1
  52 #define TR_RAID5_F_LOCKED       0x2
  53 #define TR_RAID5_F_ABORT        0x4
  54
  55 struct g_raid_tr_raid5_object {
  56         struct g_raid_tr_object  trso_base;
  57         int                      trso_starting;
  58         int                      trso_stopping;
  59         int                      trso_type;
  60         int                      trso_recover_slabs; /* slabs before rest */
  61         int                      trso_fair_io;
  62         int                      trso_meta_update;
  63         int                      trso_flags;
  64         struct g_raid_subdisk   *trso_failed_sd; /* like per volume */
  65         void                    *trso_buffer;    /* Buffer space */
  66         struct bio               trso_bio;
  67 };
  68
  69 static g_raid_tr_taste_t g_raid_tr_taste_raid5;
  70 static g_raid_tr_event_t g_raid_tr_event_raid5;
  71 static g_raid_tr_start_t g_raid_tr_start_raid5;
  72 static g_raid_tr_stop_t g_raid_tr_stop_raid5;
  73 static g_raid_tr_iostart_t g_raid_tr_iostart_raid5;
  74 static g_raid_tr_iodone_t g_raid_tr_iodone_raid5;
  75 static g_raid_tr_kerneldump_t g_raid_tr_kerneldump_raid5;
  76 static g_raid_tr_locked_t g_raid_tr_locked_raid5;
  77 static g_raid_tr_free_t g_raid_tr_free_raid5;
  78
  79 static kobj_method_t g_raid_tr_raid5_methods[] = {
  80         KOBJMETHOD(g_raid_tr_taste,     g_raid_tr_taste_raid5),
  81         KOBJMETHOD(g_raid_tr_event,     g_raid_tr_event_raid5),
  82         KOBJMETHOD(g_raid_tr_start,     g_raid_tr_start_raid5),
  83         KOBJMETHOD(g_raid_tr_stop,      g_raid_tr_stop_raid5),
  84         KOBJMETHOD(g_raid_tr_iostart,   g_raid_tr_iostart_raid5),
  85         KOBJMETHOD(g_raid_tr_iodone,    g_raid_tr_iodone_raid5),
  86         KOBJMETHOD(g_raid_tr_kerneldump, g_raid_tr_kerneldump_raid5),
  87         KOBJMETHOD(g_raid_tr_locked,    g_raid_tr_locked_raid5),
  88         KOBJMETHOD(g_raid_tr_free,      g_raid_tr_free_raid5),
  89         { 0, 0 }
  90 };
  91
  92 static struct g_raid_tr_class g_raid_tr_raid5_class = {
  93         "RAID5",
  94         g_raid_tr_raid5_methods,
  95         sizeof(struct g_raid_tr_raid5_object),
  96         .trc_enable = 1,
  97         .trc_priority = 100
  98 };
  99
 100 static int
 101 g_raid_tr_taste_raid5(struct g_raid_tr_object *tr, struct g_raid_volume *vol)
 102 {
 103         struct g_raid_tr_raid5_object *trs;
 104         u_int qual;
 105
 106         trs = (struct g_raid_tr_raid5_object *)tr;
 107         qual = tr->tro_volume->v_raid_level_qualifier;
 108         if (tr->tro_volume->v_raid_level == G_RAID_VOLUME_RL_RAID4 &&
 109             qual >= 0 && qual <= 1) {
 110                 /* RAID4 */
 111         } else if ((tr->tro_volume->v_raid_level == G_RAID_VOLUME_RL_RAID5 ||
 112              tr->tro_volume->v_raid_level == G_RAID_VOLUME_RL_RAID5E ||
 113              tr->tro_volume->v_raid_level == G_RAID_VOLUME_RL_RAID5EE ||
 114              tr->tro_volume->v_raid_level == G_RAID_VOLUME_RL_RAID5R ||
 115              tr->tro_volume->v_raid_level == G_RAID_VOLUME_RL_RAID6 ||
 116              tr->tro_volume->v_raid_level == G_RAID_VOLUME_RL_RAIDMDF) &&
 117             qual >= 0 && qual <= 3) {
 118                 /* RAID5/5E/5EE/5R/6/MDF */
 119         } else
 120                 return (G_RAID_TR_TASTE_FAIL);
 121         trs->trso_starting = 1;
 122         return (G_RAID_TR_TASTE_SUCCEED);
 123 }
 124
 125 static int
 126 g_raid_tr_update_state_raid5(struct g_raid_volume *vol,
 127     struct g_raid_subdisk *sd)
 128 {
 129         struct g_raid_tr_raid5_object *trs;
 130         struct g_raid_softc *sc;
 131         u_int s;
 132         int na, ns, nu;
 133
 134         sc = vol->v_softc;
 135         trs = (struct g_raid_tr_raid5_object *)vol->v_tr;
 136         if (trs->trso_stopping &&
 137             (trs->trso_flags & TR_RAID5_F_DOING_SOME) == 0)
 138                 s = G_RAID_VOLUME_S_STOPPED;
 139         else if (trs->trso_starting)
 140                 s = G_RAID_VOLUME_S_STARTING;
 141         else {
 142                 na = g_raid_nsubdisks(vol, G_RAID_SUBDISK_S_ACTIVE);
 143                 ns = g_raid_nsubdisks(vol, G_RAID_SUBDISK_S_STALE) +
 144                      g_raid_nsubdisks(vol, G_RAID_SUBDISK_S_RESYNC);
 145                 nu = g_raid_nsubdisks(vol, G_RAID_SUBDISK_S_UNINITIALIZED);
 146                 if (na == vol->v_disks_count)
 147                         s = G_RAID_VOLUME_S_OPTIMAL;
 148                 else if (na + ns == vol->v_disks_count ||
 149                     na + ns + nu == vol->v_disks_count /* XXX: Temporary. */)
 150                         s = G_RAID_VOLUME_S_SUBOPTIMAL;
 151                 else if (na == vol->v_disks_count - 1 ||
 152                     na + ns + nu == vol->v_disks_count)
 153                         s = G_RAID_VOLUME_S_DEGRADED;
 154                 else
 155                         s = G_RAID_VOLUME_S_BROKEN;
 156         }
 157         if (s != vol->v_state) {
 158                 g_raid_event_send(vol, G_RAID_VOLUME_S_ALIVE(s) ?
 159                     G_RAID_VOLUME_E_UP : G_RAID_VOLUME_E_DOWN,
 160                     G_RAID_EVENT_VOLUME);
 161                 g_raid_change_volume_state(vol, s);
 162                 if (!trs->trso_starting && !trs->trso_stopping)
 163                         g_raid_write_metadata(sc, vol, NULL, NULL);
 164         }
 165         return (0);
 166 }
 167
 168 static int
 169 g_raid_tr_event_raid5(struct g_raid_tr_object *tr,
 170     struct g_raid_subdisk *sd, u_int event)
 171 {
 172
 173         g_raid_tr_update_state_raid5(tr->tro_volume, sd);
 174         return (0);
 175 }
 176
 177 static int
 178 g_raid_tr_start_raid5(struct g_raid_tr_object *tr)
 179 {
 180         struct g_raid_tr_raid5_object *trs;
 181         struct g_raid_volume *vol;
 182
 183         trs = (struct g_raid_tr_raid5_object *)tr;
 184         vol = tr->tro_volume;
 185         trs->trso_starting = 0;
 186         g_raid_tr_update_state_raid5(vol, NULL);
 187         return (0);
 188 }
 189
 190 static int
 191 g_raid_tr_stop_raid5(struct g_raid_tr_object *tr)
 192 {
 193         struct g_raid_tr_raid5_object *trs;
 194         struct g_raid_volume *vol;
 195
 196         trs = (struct g_raid_tr_raid5_object *)tr;
 197         vol = tr->tro_volume;
 198         trs->trso_starting = 0;
 199         trs->trso_stopping = 1;
 200         g_raid_tr_update_state_raid5(vol, NULL);
 201         return (0);
 202 }
 203
 204 static void
 205 g_raid_tr_iostart_raid5_read(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, start, length, nstripe, remain;
 213         int no, pno, ddisks, pdisks, protate, pleft;
 214         u_int strip_size, lvl, qual;
 215
 216         vol = tr->tro_volume;
 217         addr = bp->bio_data;
 218         strip_size = vol->v_strip_size;
 219         lvl = tr->tro_volume->v_raid_level;
 220         qual = tr->tro_volume->v_raid_level_qualifier;
 221         protate = tr->tro_volume->v_rotate_parity;
 222
 223         /* Stripe number. */
 224         nstripe = bp->bio_offset / strip_size;
 225         /* Start position in stripe. */
 226         start = bp->bio_offset % strip_size;
 227         /* Number of data and parity disks. */
 228         if (lvl == G_RAID_VOLUME_RL_RAIDMDF)
 229                 pdisks = tr->tro_volume->v_mdf_pdisks;
 230         else if (lvl == G_RAID_VOLUME_RL_RAID5EE ||
 231             lvl == G_RAID_VOLUME_RL_RAID6)
 232                 pdisks = 2;
 233         else
 234                 pdisks = 1;
 235         ddisks = vol->v_disks_count - pdisks;
 236         /* Parity disk number. */
 237         if (lvl == G_RAID_VOLUME_RL_RAID4) {
 238                 if (qual == 0)          /* P0 */
 239                         pno = 0;
 240                 else                    /* PN */
 241                         pno = ddisks;
 242                 pleft = -1;
 243         } else {
 244                 pno = (nstripe / (ddisks * protate)) % vol->v_disks_count;
 245                 pleft = protate - (nstripe / ddisks) % protate;
 246                 if (qual >= 2) {        /* PN/Left */
 247                         pno = ddisks - pno;
 248                         if (pno < 0)
 249                                 pno += vol->v_disks_count;
 250                 }
 251         }
 252         /* Data disk number. */
 253         no = nstripe % ddisks;
 254         if (lvl == G_RAID_VOLUME_RL_RAID4) {
 255                 if (qual == 0)
 256                         no += pdisks;
 257         } else if (qual & 1) {  /* Continuation/Symmetric */
 258                 no = (pno + pdisks + no) % vol->v_disks_count;
 259         } else if (no >= pno)   /* Restart/Asymmetric */
 260                 no += pdisks;
 261         else
 262                 no += imax(0, pno + pdisks - vol->v_disks_count);
 263         /* Stripe start position in disk. */
 264         offset = (nstripe / ddisks) * strip_size;
 265         /* Length of data to operate. */
 266         remain = bp->bio_length;
 267
 268         bioq_init(&queue);
 269         do {
 270                 length = MIN(strip_size - start, remain);
 271                 cbp = g_clone_bio(bp);
 272                 if (cbp == NULL)
 273                         goto failure;
 274                 cbp->bio_offset = offset + start;
 275                 cbp->bio_data = addr;
 276                 cbp->bio_length = length;
 277                 cbp->bio_caller1 = &vol->v_subdisks[no];
 278                 bioq_insert_tail(&queue, cbp);
 279                 no++;
 280                 if (lvl == G_RAID_VOLUME_RL_RAID4) {
 281                         no %= vol->v_disks_count;
 282                         if (no == pno)
 283                                 no = (no + pdisks) % vol->v_disks_count;
 284                 } else if (qual & 1) {  /* Continuation/Symmetric */
 285                         no %= vol->v_disks_count;
 286                         if (no == pno) {
 287                                 if ((--pleft) <= 0) {
 288                                         pleft += protate;
 289                                         if (qual < 2)   /* P0/Right */
 290                                                 pno++;
 291                                         else            /* PN/Left */
 292                                                 pno += vol->v_disks_count - 1;
 293                                         pno %= vol->v_disks_count;
 294                                 }
 295                                 no = (pno + pdisks) % vol->v_disks_count;
 296                                 offset += strip_size;
 297                         }
 298                 } else {                /* Restart/Asymmetric */
 299                         if (no == pno)
 300                                 no += pdisks;
 301                         if (no >= vol->v_disks_count) {
 302                                 no -= vol->v_disks_count;
 303                                 if ((--pleft) <= 0) {
 304                                         pleft += protate;
 305                                         if (qual < 2)   /* P0/Right */
 306                                                 pno++;
 307                                         else            /* PN/Left */
 308                                                 pno += vol->v_disks_count - 1;
 309                                         pno %= vol->v_disks_count;
 310                                 }
 311                                 if (no == pno)
 312                                         no += pdisks;
 313                                 else
 314                                         no += imax(0, pno + pdisks - vol->v_disks_count);
 315                                 offset += strip_size;
 316                         }
 317                 }
 318                 remain -= length;
 319                 addr += length;
 320                 start = 0;
 321         } while (remain > 0);
 322         for (cbp = bioq_first(&queue); cbp != NULL;
 323             cbp = bioq_first(&queue)) {
 324                 bioq_remove(&queue, cbp);
 325                 sd = cbp->bio_caller1;
 326                 cbp->bio_caller1 = NULL;
 327                 g_raid_subdisk_iostart(sd, cbp);
 328         }
 329         return;
 330 failure:
 331         for (cbp = bioq_first(&queue); cbp != NULL;
 332             cbp = bioq_first(&queue)) {
 333                 bioq_remove(&queue, cbp);
 334                 g_destroy_bio(cbp);
 335         }
 336         if (bp->bio_error == 0)
 337                 bp->bio_error = ENOMEM;
 338         g_raid_iodone(bp, bp->bio_error);
 339 }
 340
 341 static void
 342 g_raid_tr_iostart_raid5(struct g_raid_tr_object *tr, struct bio *bp)
 343 {
 344         struct g_raid_volume *vol;
 345         struct g_raid_tr_raid5_object *trs;
 346
 347         vol = tr->tro_volume;
 348         trs = (struct g_raid_tr_raid5_object *)tr;
 349         if (vol->v_state < G_RAID_VOLUME_S_SUBOPTIMAL) {
 350                 g_raid_iodone(bp, EIO);
 351                 return;
 352         }
 353         switch (bp->bio_cmd) {
 354         case BIO_READ:
 355                 g_raid_tr_iostart_raid5_read(tr, bp);
 356                 break;
 357         case BIO_WRITE:
 358         case BIO_DELETE:
 359         case BIO_FLUSH:
 360                 g_raid_iodone(bp, ENODEV);
 361                 break;
 362         default:
 363                 KASSERT(1 == 0, ("Invalid command here: %u (volume=%s)",
 364                     bp->bio_cmd, vol->v_name));
 365                 break;
 366         }
 367 }
 368
 369 static void
 370 g_raid_tr_iodone_raid5(struct g_raid_tr_object *tr,
 371     struct g_raid_subdisk *sd, struct bio *bp)
 372 {
 373         struct bio *pbp;
 374         int error;
 375
 376         pbp = bp->bio_parent;
 377         pbp->bio_inbed++;
 378         error = bp->bio_error;
 379         g_destroy_bio(bp);
 380         if (pbp->bio_children == pbp->bio_inbed) {
 381                 pbp->bio_completed = pbp->bio_length;
 382                 g_raid_iodone(pbp, error);
 383         }
 384 }
 385
 386 static int
 387 g_raid_tr_kerneldump_raid5(struct g_raid_tr_object *tr,
 388     void *virtual, vm_offset_t physical, off_t offset, size_t length)
 389 {
 390
 391         return (ENODEV);
 392 }
 393
 394 static int
 395 g_raid_tr_locked_raid5(struct g_raid_tr_object *tr, void *argp)
 396 {
 397         struct bio *bp;
 398         struct g_raid_subdisk *sd;
 399
 400         bp = (struct bio *)argp;
 401         sd = (struct g_raid_subdisk *)bp->bio_caller1;
 402         g_raid_subdisk_iostart(sd, bp);
 403
 404         return (0);
 405 }
 406
 407 static int
 408 g_raid_tr_free_raid5(struct g_raid_tr_object *tr)
 409 {
 410         struct g_raid_tr_raid5_object *trs;
 411
 412         trs = (struct g_raid_tr_raid5_object *)tr;
 413
 414         if (trs->trso_buffer != NULL) {
 415                 free(trs->trso_buffer, M_TR_RAID5);
 416                 trs->trso_buffer = NULL;
 417         }
 418         return (0);
 419 }
 420
 421 G_RAID_TR_DECLARE(raid5, "RAID5");