rf_copyback.c

RAIDFrame是个非常好的磁盘阵列RAID仿真工具

/*
 * Copyright (c) 1995 Carnegie-Mellon University.
 * All rights reserved.
 *
 * Author: Mark Holland
 *
 * Permission to use, copy, modify and distribute this software and
 * its documentation is hereby granted, provided that both the copyright
 * notice and this permission notice appear in all copies of the
 * software, derivative works or modified versions, and any portions
 * thereof, and that both notices appear in supporting documentation.
 *
 * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
 * CONDITION.  CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
 * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
 *
 * Carnegie Mellon requests users of this software to return to
 *
 *  Software Distribution Coordinator  or  [e-mail]
 *  School of Computer Science
 *  Carnegie Mellon University
 *  Pittsburgh PA 15213-3890
 *
 * any improvements or extensions that they make and grant Carnegie the
 * rights to redistribute these changes.
 */

/*****************************************************************************************
 *
 * copyback.c -- code to copy reconstructed data back from spare space to
 *               the replaced disk.
 *
 * the code operates using callbacks on the I/Os to continue with the next 
 * unit to be copied back.  We do this because a simple loop containing blocking I/Os
 * will not work in the simulator.
 *
 ****************************************************************************************/

/*
 * $Locker:  $
 * $Log: rf_copyback.c,v $
 * Revision 1.26  1996/08/06  22:26:00  jimz
 * don't include sys/buf.h on linux
 *
 * Revision 1.25  1996/07/30  03:30:40  jimz
 * include rf_types.h first
 *
 * Revision 1.24  1996/07/27  18:39:52  jimz
 * cleanup sweep
 *
 * Revision 1.23  1996/07/18  22:57:14  jimz
 * port simulator to AIX
 *
 * Revision 1.22  1996/07/11  19:08:00  jimz
 * generalize reconstruction mechanism
 * allow raid1 reconstructs via copyback (done with array
 * quiesced, not online, therefore not disk-directed)
 *
 * Revision 1.21  1996/07/11  16:03:47  jimz
 * fixed hanging bug in rf_CopybackWriteDoneProc()
 *
 * Revision 1.20  1996/06/10  11:55:47  jimz
 * Straightened out some per-array/not-per-array distinctions, fixed
 * a couple bugs related to confusion. Added shutdown lists. Removed
 * layout shutdown function (now subsumed by shutdown lists).
 *
 * Revision 1.19  1996/06/09  02:36:46  jimz
 * lots of little crufty cleanup- fixup whitespace
 * issues, comment #ifdefs, improve typing in some
 * places (esp size-related)
 *
 * Revision 1.18  1996/06/07  21:33:04  jimz
 * begin using consistent types for sector numbers,
 * stripe numbers, row+col numbers, recon unit numbers
 *
 * Revision 1.17  1996/06/05  18:06:02  jimz
 * Major code cleanup. The Great Renaming is now done.
 * Better modularity. Better typing. Fixed a bunch of
 * synchronization bugs. Made a lot of global stuff
 * per-desc or per-array. Removed dead code.
 *
 * Revision 1.16  1996/06/03  23:28:26  jimz
 * more bugfixes
 * check in tree to sync for IPDS runs with current bugfixes
 * there still may be a problem with threads in the script test
 * getting I/Os stuck- not trivially reproducible (runs ~50 times
 * in a row without getting stuck)
 *
 * Revision 1.15  1996/06/02  17:31:48  jimz
 * Moved a lot of global stuff into array structure, where it belongs.
 * Fixed up paritylogging, pss modules in this manner. Some general
 * code cleanup. Removed lots of dead code, some dead files.
 *
 * Revision 1.14  1996/05/31  22:26:54  jimz
 * fix a lot of mapping problems, memory allocation problems
 * found some weird lock issues, fixed 'em
 * more code cleanup
 *
 * Revision 1.13  1996/05/30  11:29:41  jimz
 * Numerous bug fixes. Stripe lock release code disagreed with the taking code
 * about when stripes should be locked (I made it consistent: no parity, no lock)
 * There was a lot of extra serialization of I/Os which I've removed- a lot of
 * it was to calculate values for the cache code, which is no longer with us.
 * More types, function, macro cleanup. Added code to properly quiesce the array
 * on shutdown. Made a lot of stuff array-specific which was (bogusly) general
 * before. Fixed memory allocation, freeing bugs.
 *
 * Revision 1.12  1996/05/27  18:56:37  jimz
 * more code cleanup
 * better typing
 * compiles in all 3 environments
 *
 * Revision 1.11  1996/05/24  22:17:04  jimz
 * continue code + namespace cleanup
 * typed a bunch of flags
 *
 * Revision 1.10  1996/05/24  01:59:45  jimz
 * another checkpoint in code cleanup for release
 * time to sync kernel tree
 *
 * Revision 1.9  1996/05/23  21:46:35  jimz
 * checkpoint in code cleanup (release prep)
 * lots of types, function names have been fixed
 *
 * Revision 1.8  1996/05/23  00:33:23  jimz
 * code cleanup: move all debug decls to rf_options.c, all extern
 * debug decls to rf_options.h, all debug vars preceded by rf_
 *
 * Revision 1.7  1996/05/18  19:51:34  jimz
 * major code cleanup- fix syntax, make some types consistent,
 * add prototypes, clean out dead code, et cetera
 *
 * Revision 1.6  1995/12/12  18:10:06  jimz
 * MIN -> RF_MIN, MAX -> RF_MAX, ASSERT -> RF_ASSERT
 * fix 80-column brain damage in comments
 *
 * Revision 1.5  1995/12/01  15:15:31  root
 * added copyright info
 *
 * Revision 1.4  1995/06/23  13:41:36  robby
 * updeated to prototypes in rf_layout.h
 *
 */

#include "rf_types.h"
#include <sys/time.h>
#ifndef LINUX
#include <sys/buf.h>
#endif /* !LINUX */
#include "rf_raid.h"
#include "rf_threadid.h"
#include "rf_mcpair.h"
#include "rf_acctrace.h"
#include "rf_etimer.h"
#include "rf_general.h"
#include "rf_utils.h"
#include "rf_copyback.h"
#include "rf_camlayer.h"
#include "rf_decluster.h"
#include "rf_driver.h"
#include "rf_shutdown.h"
#include "rf_sys.h"

#define RF_COPYBACK_DATA   0
#define RF_COPYBACK_PARITY 1

int rf_copyback_in_progress;

static int  rf_CopybackReadDoneProc(RF_CopybackDesc_t *desc, int status);
static int  rf_CopybackWriteDoneProc(RF_CopybackDesc_t *desc, int status);
static void rf_CopybackOne(RF_CopybackDesc_t *desc, int typ,
	RF_RaidAddr_t addr, RF_RowCol_t testRow, RF_RowCol_t testCol,
	RF_SectorNum_t testOffs);
static void rf_CopybackComplete(RF_CopybackDesc_t *desc, int status);

int rf_ConfigureCopyback(listp)
  RF_ShutdownList_t  **listp;
{
  rf_copyback_in_progress = 0;
  return(0);
}

/* do a complete copyback */
void rf_CopybackReconstructedData(raidPtr)
  RF_Raid_t  *raidPtr;
{
  int bus, targ, lun, done, retcode;
  RF_CopybackDesc_t *desc;
  RF_RowCol_t frow, fcol;
  RF_RaidDisk_t *badDisk;
  RF_DiskOp_t *tur_op;
  char *databuf;

  done = 0;
  fcol = 0;
  for (frow=0; frow<raidPtr->numRow; frow++) {
    for (fcol=0; fcol<raidPtr->numCol; fcol++) {
      if (raidPtr->Disks[frow][fcol].status == rf_ds_dist_spared
        || raidPtr->Disks[frow][fcol].status == rf_ds_spared)
      {
        done = 1;
        break;
      }
    }
    if (done)
      break;
  }

  if (frow == raidPtr->numRow) {
    printf("COPYBACK:  no disks need copyback\n");
    return;
  }

  badDisk = &raidPtr->Disks[frow][fcol];
#ifndef SIMULATE
  if (rf_extract_ids(badDisk->devname, &bus, &targ, &lun)) {
    printf("COPYBACK: unable to extract bus, target, lun from devname %s\n",
      badDisk->devname);
    return;
  }

  /* TUR the disk that's marked as bad to be sure that it's actually alive */
  rf_SCSI_AllocTUR(&tur_op);
  retcode = rf_SCSI_DoTUR(tur_op, bus, targ, lun, badDisk->dev);
  rf_SCSI_FreeDiskOp(tur_op, 0);
  if (retcode) {
    printf("COPYBACK: target disk failed TUR\n");
    return;
  }
#endif /* !SIMULATE */

  /* get a buffer to hold one SU  */
  RF_Malloc(databuf, rf_RaidAddressToByte(raidPtr, raidPtr->Layout.sectorsPerStripeUnit), (char *));

  /* create a descriptor */
  RF_Malloc(desc, sizeof(*desc), (RF_CopybackDesc_t *));
  desc->raidPtr       = raidPtr;
  desc->status        = 0;