rf_copyback.c

RAIDFrame是个非常好的磁盘阵列RAID仿真工具

  desc->frow          = frow;
  desc->fcol          = fcol;
  desc->spRow         = badDisk->spareRow;
  desc->spCol         = badDisk->spareCol;
  desc->stripeAddr    = 0;
  desc->sectPerSU     = raidPtr->Layout.sectorsPerStripeUnit;
  desc->sectPerStripe = raidPtr->Layout.sectorsPerStripeUnit * raidPtr->Layout.numDataCol;
  desc->databuf       = databuf;
#ifndef SIMULATE
  desc->mcpair        = rf_AllocMCPair();
#endif /* !SIMULATE */

  printf("COPYBACK: Quiescing the array\n");
  /* quiesce the array, since we don't want to code support for user accs here */
  rf_SuspendNewRequestsAndWait(raidPtr);

  /* adjust state of the array and of the disks */
  RF_LOCK_MUTEX(raidPtr->mutex);
  raidPtr->Disks[desc->frow][desc->fcol].status = rf_ds_optimal;
  raidPtr->status[desc->frow] = rf_rs_optimal;
  rf_copyback_in_progress = 1;                   /* debug only */
  RF_UNLOCK_MUTEX(raidPtr->mutex);
  
  printf("COPYBACK: Beginning\n");
  RF_GETTIME(desc->starttime);
  rf_ContinueCopyback(desc);
}


/*
 * invoked via callback after a copyback I/O has completed to
 * continue on with the next one
 */
void rf_ContinueCopyback(desc)
  RF_CopybackDesc_t  *desc;
{
  RF_SectorNum_t testOffs, stripeAddr;
  RF_Raid_t *raidPtr = desc->raidPtr;
  RF_RaidAddr_t addr;
  RF_RowCol_t testRow, testCol; 
  int old_pctg, new_pctg, done;
  struct timeval t, diff;

  old_pctg = (-1);
  while (1) {
    stripeAddr = desc->stripeAddr;
    if (rf_prReconSched) {
      old_pctg = 100 * desc->stripeAddr / raidPtr->totalSectors;
    }
    desc->stripeAddr += desc->sectPerStripe;
    if (rf_prReconSched) {
      new_pctg = 100 * desc->stripeAddr / raidPtr->totalSectors;
      if (new_pctg != old_pctg) {
        RF_GETTIME(t);
        RF_TIMEVAL_DIFF(&desc->starttime, &t, &diff);
        printf("%d %d.%06d\n",new_pctg, diff.tv_sec, diff.tv_usec);
      }
    }
    
    if (stripeAddr >= raidPtr->totalSectors) {
      rf_CopybackComplete(desc, 0);
      return;
    }
    
    /* walk through the current stripe, su-by-su */
    for (done=0, addr = stripeAddr; addr < stripeAddr+desc->sectPerStripe; addr += desc->sectPerSU) {
      
      /* map the SU, disallowing remap to spare space */
      (raidPtr->Layout.map->MapSector)(raidPtr, addr, &testRow, &testCol, &testOffs, RF_DONT_REMAP);
      
      if (testRow == desc->frow && testCol == desc->fcol) {
        rf_CopybackOne(desc, RF_COPYBACK_DATA, addr, testRow, testCol, testOffs);
#ifdef SIMULATE
        return;
#else /* SIMULATE */
        done = 1;
        break;
#endif /* SIMULATE */
      }
    }

    if (!done) {
      /* we didn't find the failed disk in the data part.  check parity. */
    
      /* map the parity for this stripe, disallowing remap to spare space */
      (raidPtr->Layout.map->MapParity)(raidPtr, stripeAddr, &testRow, &testCol, &testOffs, RF_DONT_REMAP);
    
      if (testRow == desc->frow && testCol == desc->fcol) {
        rf_CopybackOne(desc, RF_COPYBACK_PARITY, stripeAddr, testRow, testCol, testOffs);
#ifdef SIMULATE
        return;
#endif /* SIMULATE */
      }
    }

    /* check to see if the last read/write pair failed */
    if (desc->status) {
      rf_CopybackComplete(desc, 1);
      return;
    }
    
    /* we didn't find any units to copy back in this stripe.  Continue with the next one */
  }
}


/* copyback one unit */
static void rf_CopybackOne(desc, typ, addr, testRow, testCol, testOffs)
  RF_CopybackDesc_t  *desc;
  int                 typ;
  RF_RaidAddr_t       addr;
  RF_RowCol_t         testRow;
  RF_RowCol_t         testCol;
  RF_SectorNum_t      testOffs;
{
  RF_SectorCount_t sectPerSU = desc->sectPerSU;
  RF_Raid_t *raidPtr = desc->raidPtr;
  RF_RowCol_t spRow     = desc->spRow;
  RF_RowCol_t spCol     = desc->spCol;
  RF_SectorNum_t spOffs;

  /* find the spare spare location for this SU */
  if (raidPtr->Layout.map->flags & RF_DISTRIBUTE_SPARE) {
    if (typ == RF_COPYBACK_DATA)
      raidPtr->Layout.map->MapSector(raidPtr, addr, &spRow, &spCol, &spOffs, RF_REMAP);
    else
      raidPtr->Layout.map->MapParity(raidPtr, addr, &spRow, &spCol, &spOffs, RF_REMAP);      
  } else {
    spOffs = testOffs;
  }

  /* create reqs to read the old location & write the new */
  desc->readreq  = rf_CreateDiskQueueData(RF_IO_TYPE_READ, spOffs,
    sectPerSU, desc->databuf, 0L, 0, rf_CopybackReadDoneProc, desc,
    NULL, NULL, (void *) raidPtr, RF_DISKQUEUE_DATA_FLAGS_NONE, NULL);
  desc->writereq = rf_CreateDiskQueueData(RF_IO_TYPE_WRITE, testOffs,
    sectPerSU, desc->databuf, 0L, 0, rf_CopybackWriteDoneProc, desc,
    NULL, NULL, (void *) raidPtr, RF_DISKQUEUE_DATA_FLAGS_NONE, NULL);
  desc->frow = testRow;
  desc->fcol = testCol;
  
  /* enqueue the read.  the write will go out as part of the callback on the read.
   * at user-level & in the kernel, wait for the read-write pair to complete.
   * in the simulator, just return, since everything will happen as callbacks 
   */
#ifndef SIMULATE
  RF_LOCK_MUTEX(desc->mcpair->mutex);
  desc->mcpair->flag = 0;
#endif /* !SIMULATE */

  rf_DiskIOEnqueue(&raidPtr->Queues[spRow][spCol], desc->readreq, RF_IO_NORMAL_PRIORITY);
  
#ifndef SIMULATE
  while (!desc->mcpair->flag) {
    RF_WAIT_MCPAIR(desc->mcpair);
  }
  RF_UNLOCK_MUTEX(desc->mcpair->mutex);
  rf_FreeDiskQueueData(desc->readreq);
  rf_FreeDiskQueueData(desc->writereq);
#endif /* !SIMULATE */
}


/* called at interrupt context when the read has completed.  just send out the write */
static int rf_CopybackReadDoneProc(desc, status)
  RF_CopybackDesc_t  *desc;
  int                 status;
{
  if (status) {                                              /* invoke the callback with bad status */
    printf("COPYBACK: copyback read failed.  Aborting.\n");
    (desc->writereq->CompleteFunc)(desc, -100);
  }
  else {
    rf_DiskIOEnqueue(&(desc->raidPtr->Queues[desc->frow][desc->fcol]), desc->writereq, RF_IO_NORMAL_PRIORITY);
  }
  return(0);
}

/* called at interrupt context when the write has completed.  
 * at user level & in the kernel, wake up the copyback thread.
 * in the simulator, invoke the next copyback directly.
 * can't free diskqueuedata structs in the kernel b/c we're at interrupt context.
 */
static int rf_CopybackWriteDoneProc(desc, status)
  RF_CopybackDesc_t  *desc;
  int                 status;
{
  if (status && status != -100) {
    printf("COPYBACK: copyback write failed.  Aborting.\n");
  }

#ifdef SIMULATE
  rf_FreeDiskQueueData(desc->readreq);
  rf_FreeDiskQueueData(desc->writereq);
  if (!status)
    rf_ContinueCopyback(desc);
  else
    rf_CopybackComplete(desc, 1);
#else /* SIMULATE */
  desc->status = status;
  rf_MCPairWakeupFunc(desc->mcpair);
#endif /* SIMULATE */
  return(0);
}  

/* invoked when the copyback has completed */
static void rf_CopybackComplete(desc, status)
  RF_CopybackDesc_t  *desc;
  int                 status;
{
  RF_Raid_t *raidPtr = desc->raidPtr;
  struct timeval t, diff;

  if (!status) {
    RF_LOCK_MUTEX(raidPtr->mutex);
    if (raidPtr->Layout.map->flags & RF_DISTRIBUTE_SPARE) {
      RF_ASSERT(raidPtr->Layout.map->parityConfig == 'D');
      rf_FreeSpareTable(raidPtr);
    } else {
      raidPtr->Disks[desc->spRow][desc->spCol].status = rf_ds_spare;
    }
    RF_UNLOCK_MUTEX(raidPtr->mutex);

    RF_GETTIME(t);
    RF_TIMEVAL_DIFF(&desc->starttime, &t, &diff);
    printf("Copyback time was %d.%06d seconds\n",diff.tv_sec, diff.tv_usec);
  } else printf("COPYBACK: Failure.\n");

  RF_Free(desc->databuf, rf_RaidAddressToByte(raidPtr, desc->sectPerSU));
#ifndef SIMULATE
  rf_FreeMCPair(desc->mcpair);
#endif /* !SIMULATE */
  RF_Free(desc, sizeof(*desc));

  rf_copyback_in_progress = 0;
  rf_ResumeNewRequests(raidPtr);
}