root/dev/raidframe/rf_copyback.c

DEFINITIONS

1. rf_ConfigureCopyback
2. rf_CopybackReconstructedData
3. rf_ContinueCopyback
4. rf_CopybackOne
5. rf_CopybackReadDoneProc
6. rf_CopybackWriteDoneProc
7. rf_CopybackComplete

    1 /*      $OpenBSD: rf_copyback.c,v 1.8 2007/06/05 00:38:22 deraadt Exp $ */
    2 /*      $NetBSD: rf_copyback.c,v 1.14 2000/03/07 02:59:50 oster Exp $   */
    4 /*
    5  * Copyright (c) 1995 Carnegie-Mellon University.
    6  * All rights reserved.
    7  *
    8  * Author: Mark Holland
    9  *
   10  * Permission to use, copy, modify and distribute this software and
   11  * its documentation is hereby granted, provided that both the copyright
   12  * notice and this permission notice appear in all copies of the
   13  * software, derivative works or modified versions, and any portions
   14  * thereof, and that both notices appear in supporting documentation.
   15  *
   16  * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
   17  * CONDITION.  CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
   18  * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
   19  *
   20  * Carnegie Mellon requests users of this software to return to
   21  *
   22  *  Software Distribution Coordinator  or  Software.Distribution@CS.CMU.EDU
   23  *  School of Computer Science
   24  *  Carnegie Mellon University
   25  *  Pittsburgh PA 15213-3890
   26  *
   27  * any improvements or extensions that they make and grant Carnegie the
   28  * rights to redistribute these changes.
   29  */
   32 /*****************************************************************************
   33  *
   34  * copyback.c -- Code to copy reconstructed data back from spare space to
   35  *               the replaced disk.
   36  *
   37  * The code operates using callbacks on the I/Os to continue with the next
   38  * unit to be copied back. We do this because a simple loop containing
   39  * blocking I/Os will not work in the simulator.
   40  *
   41  *****************************************************************************/
   43 #include "rf_types.h"
   45 #include <sys/time.h>
   46 #include <sys/buf.h>
   47 #include "rf_raid.h"
   48 #include "rf_mcpair.h"
   49 #include "rf_acctrace.h"
   50 #include "rf_etimer.h"
   51 #include "rf_general.h"
   52 #include "rf_utils.h"
   53 #include "rf_copyback.h"
   54 #include "rf_decluster.h"
   55 #include "rf_driver.h"
   56 #include "rf_shutdown.h"
   57 #include "rf_kintf.h"
   59 #define RF_COPYBACK_DATA        0
   60 #define RF_COPYBACK_PARITY      1
   62 int     rf_copyback_in_progress;
   64 int  rf_CopybackReadDoneProc(RF_CopybackDesc_t *, int);
   65 int  rf_CopybackWriteDoneProc(RF_CopybackDesc_t *, int);
   66 void rf_CopybackOne(RF_CopybackDesc_t *, int, RF_RaidAddr_t,
RF_RowCol_t, RF_RowCol_t, RF_SectorNum_t);
   68 void rf_CopybackComplete(RF_CopybackDesc_t *, int);
   70 int
rf_ConfigureCopyback(RF_ShutdownList_t **listp)
   72 {
rf_copyback_in_progress = 0;
   74         return (0);
   75 }
   77 #include <sys/types.h>
   78 #include <sys/param.h>
   79 #include <sys/systm.h>
   80 #include <sys/proc.h>
   81 #include <sys/ioctl.h>
   82 #include <sys/fcntl.h>
   83 #ifdef  __NETBSD__
   84 #include <sys/vnode.h>
   85 #endif
   88 /* Do a complete copyback. */
   89 void
rf_CopybackReconstructedData(RF_Raid_t *raidPtr)
   91 {
RF_ComponentLabel_t c_label;
   93         int done, retcode;
RF_CopybackDesc_t *desc;
RF_RowCol_t frow, fcol;
RF_RaidDisk_t *badDisk;
   97         char *databuf;
   99         struct partinfo dpart;
  100         struct vnode *vp;
  101         struct vattr va;
  102         struct proc *proc;
  104         int ac;
done = 0;
fcol = 0;
  108         for (frow = 0; frow < raidPtr->numRow; frow++) {
  109                 for (fcol = 0; fcol < raidPtr->numCol; fcol++) {
  110                         if (raidPtr->Disks[frow][fcol].status ==
rf_ds_dist_spared ||
raidPtr->Disks[frow][fcol].status ==
rf_ds_spared) {
done = 1;
  115                                 break;
  116                         }
  117                 }
  118                 if (done)
  119                         break;
  120         }
  122         if (frow == raidPtr->numRow) {
printf("COPYBACK: No disks need copyback.\n");
  124                 return;
  125         }
badDisk = &raidPtr->Disks[frow][fcol];
proc = raidPtr->engine_thread;
  130         /*
  131          * This device may have been opened successfully the first time.
  132          * Close it before trying to open it again.
  133          */
  135         if (raidPtr->raid_cinfo[frow][fcol].ci_vp != NULL) {
printf("Close the opened device: %s.\n",
raidPtr->Disks[frow][fcol].devname);
vp = raidPtr->raid_cinfo[frow][fcol].ci_vp;
ac = raidPtr->Disks[frow][fcol].auto_configured;
rf_close_component(raidPtr, vp, ac);
raidPtr->raid_cinfo[frow][fcol].ci_vp = NULL;
  143         }
  144         /* Note that this disk was *not* auto_configured (any longer). */
raidPtr->Disks[frow][fcol].auto_configured = 0;
printf("About to (re-)open the device: %s.\n",
raidPtr->Disks[frow][fcol].devname);
retcode = raidlookup(raidPtr->Disks[frow][fcol].devname, proc, &vp);
  152         if (retcode) {
printf("COPYBACK: raidlookup on device: %s failed: %d !\n",
raidPtr->Disks[frow][fcol].devname, retcode);
  156                 /*
  157                  * XXX The component isn't responding properly... Must be
  158                  * still dead :-(
  159                  */
  160                 return;
  162         } else {
  164                 /*
  165                  * Ok, so we can at least do a lookup...
  166                  * How about actually getting a vp for it ?
  167                  */
  169                 if ((retcode = VOP_GETATTR(vp, &va, proc->p_ucred, proc)) != 0)
  170                 {
  171                         return;
  172                 }
retcode = VOP_IOCTL(vp, DIOCGPART, (caddr_t) &dpart, FREAD,
proc->p_ucred, proc);
  175                 if (retcode) {
  176                         return;
  177                 }
raidPtr->Disks[frow][fcol].blockSize = dpart.disklab->d_secsize;
raidPtr->Disks[frow][fcol].numBlocks = DL_GETPSIZE(dpart.part) -
rf_protectedSectors;
raidPtr->raid_cinfo[frow][fcol].ci_vp = vp;
raidPtr->raid_cinfo[frow][fcol].ci_dev = va.va_rdev;
  186                 /* XXX Or the above ? */
raidPtr->Disks[frow][fcol].dev = va.va_rdev;
  189                 /*
  190                  * We allow the user to specify that only a fraction of the
  191                  * disks should be used this is just for debug: it speeds up
  192                  * the parity scan.
  193                  */
raidPtr->Disks[frow][fcol].numBlocks =
raidPtr->Disks[frow][fcol].numBlocks *
rf_sizePercentage / 100;
  197         }
  198 #if 0
  199         /* This is the way it was done before the CAM stuff was removed. */
  201         if (rf_extract_ids(badDisk->devname, &bus, &targ, &lun)) {
printf("COPYBACK: unable to extract bus, target, lun from"
  203                     " devname %s.\n", badDisk->devname);
  204                 return;
  205         }
  206         /*
  207          * TUR the disk that's marked as bad to be sure that it's actually
  208          * alive.
  209          */
rf_SCSI_AllocTUR(&tur_op);
retcode = rf_SCSI_DoTUR(tur_op, bus, targ, lun, badDisk->dev);
rf_SCSI_FreeDiskOp(tur_op, 0);
  213 #endif
  215         if (retcode) {
printf("COPYBACK: target disk failed TUR.\n");
  217                 return;
  218         }
  219         /* Get a buffer to hold one SU. */
RF_Malloc(databuf, rf_RaidAddressToByte(raidPtr,
raidPtr->Layout.sectorsPerStripeUnit), (char *));
  223         /* Create a descriptor. */
RF_Malloc(desc, sizeof(*desc), (RF_CopybackDesc_t *));
desc->raidPtr = raidPtr;
desc->status = 0;
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;
desc->mcpair = rf_AllocMCPair();
printf("COPYBACK: Quiescing the array.\n");
  239         /*
  240          * Quiesce the array, since we don't want to code support for user
  241          * accs here.
  242          */
rf_SuspendNewRequestsAndWait(raidPtr);
  245         /* 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);
  256         /*
  257          * Data has been restored.
  258          * Fix up the component label.
  259          * Don't actually need the read here.
  260          */
raidread_component_label(raidPtr->raid_cinfo[frow][fcol].ci_dev,
raidPtr->raid_cinfo[frow][fcol].ci_vp,
  263                                  &c_label);
raid_init_component_label(raidPtr, &c_label);
c_label.row = frow;
c_label.column = fcol;
raidwrite_component_label(raidPtr->raid_cinfo[frow][fcol].ci_dev,
raidPtr->raid_cinfo[frow][fcol].ci_vp,
  272                                   &c_label);
  273 }
  276 /*
  277  * Invoked via callback after a copyback I/O has completed to
  278  * continue on with the next one.
  279  */
  280 void
rf_ContinueCopyback(RF_CopybackDesc_t *desc)
  282 {
RF_SectorNum_t testOffs, stripeAddr;
RF_Raid_t *raidPtr = desc->raidPtr;
RF_RaidAddr_t addr;
RF_RowCol_t testRow, testCol;
  287         int old_pctg, new_pctg, done;
  288         struct timeval t, diff;
old_pctg = (-1);
  291         while (1) {
stripeAddr = desc->stripeAddr;
desc->raidPtr->copyback_stripes_done = stripeAddr /
desc->sectPerStripe;
  295                 if (rf_prReconSched) {
old_pctg = 100 * desc->stripeAddr /
raidPtr->totalSectors;
  298                 }
desc->stripeAddr += desc->sectPerStripe;
  300                 if (rf_prReconSched) {
new_pctg = 100 * desc->stripeAddr /
raidPtr->totalSectors;
  303                         if (new_pctg != old_pctg) {
RF_GETTIME(t);
RF_TIMEVAL_DIFF(&desc->starttime, &t, &diff);
printf("%d %d.%06d\n", new_pctg,
  307                                     (int) diff.tv_sec, (int) diff.tv_usec);
  308                         }
  309                 }
  310                 if (stripeAddr >= raidPtr->totalSectors) {
rf_CopybackComplete(desc, 0);
  312                         return;
  313                 }
  314                 /* Walk through the current stripe, su-by-su. */
  315                 for (done = 0, addr = stripeAddr;
addr < stripeAddr + desc->sectPerStripe;
addr += desc->sectPerSU) {
  319                         /* Map the SU, disallowing remap to spare space. */
  320                         (raidPtr->Layout.map->MapSector) (raidPtr, addr,
  321                             &testRow, &testCol, &testOffs, RF_DONT_REMAP);
  323                         if (testRow == desc->frow && testCol == desc->fcol) {
rf_CopybackOne(desc, RF_COPYBACK_DATA, addr,
testRow, testCol, testOffs);
done = 1;
  327                                 break;
  328                         }
  329                 }
  331                 if (!done) {
  332                         /*
  333                          * We didn't find the failed disk in the data part,
  334                          * check parity.
  335                          */
  337                         /*
  338                          * Map the parity for this stripe, disallowing remap
  339                          * to spare space.
  340                          */
  341                         (raidPtr->Layout.map->MapParity) (raidPtr, stripeAddr,
  342                             &testRow, &testCol, &testOffs, RF_DONT_REMAP);
  344                         if (testRow == desc->frow && testCol == desc->fcol) {
rf_CopybackOne(desc, RF_COPYBACK_PARITY,
stripeAddr, testRow, testCol, testOffs);
  347                         }
  348                 }
  349                 /* Check to see if the last read/write pair failed. */
  350                 if (desc->status) {
rf_CopybackComplete(desc, 1);
  352                         return;
  353                 }
  354                 /*
  355                  * We didn't find any units to copy back in this stripe.
  356                  * Continue with the next one.
  357                  */
  358         }
  359 }
  362 /* Copyback one unit. */
  363 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)
  366 {
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;
  373         /* Find the spare location for this SU. */
  374         if (raidPtr->Layout.map->flags & RF_DISTRIBUTE_SPARE) {
  375                 if (typ == RF_COPYBACK_DATA)
raidPtr->Layout.map->MapSector(raidPtr, addr, &spRow,
  377                             &spCol, &spOffs, RF_REMAP);
  378                 else
raidPtr->Layout.map->MapParity(raidPtr, addr, &spRow,
  380                             &spCol, &spOffs, RF_REMAP);
  381         } else {
spOffs = testOffs;
  383         }
  385         /* Create reqs to read the old location & write the new. */
desc->readreq = rf_CreateDiskQueueData(RF_IO_TYPE_READ, spOffs,
sectPerSU, desc->databuf, 0L, 0, (int (*) (void *, int))
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, (int (*) (void *, int))
rf_CopybackWriteDoneProc, desc, NULL, NULL, (void *) raidPtr,
RF_DISKQUEUE_DATA_FLAGS_NONE, NULL);
desc->frow = testRow;
desc->fcol = testCol;
  397         /*
  398          * Enqueue the read. The write will go out as part of the callback on
  399          * the read. At user-level & in the kernel, wait for the read-write
  400          * pair to complete. In the simulator, just return, since everything
  401          * will happen as callbacks.
  402          */
RF_LOCK_MUTEX(desc->mcpair->mutex);
desc->mcpair->flag = 0;
rf_DiskIOEnqueue(&raidPtr->Queues[spRow][spCol], desc->readreq,
RF_IO_NORMAL_PRIORITY);
  410         while (!desc->mcpair->flag) {
RF_WAIT_MCPAIR(desc->mcpair);
  412         }
RF_UNLOCK_MUTEX(desc->mcpair->mutex);
rf_FreeDiskQueueData(desc->readreq);
rf_FreeDiskQueueData(desc->writereq);
  417 }
  420 /*
  421  * Called at interrupt context when the read has completed.
  422  * Just send out the write.
  423  */
  424 int
rf_CopybackReadDoneProc(RF_CopybackDesc_t *desc, int status)
  426 {
  427         if (status) {           /* Invoke the callback with bad status. */
printf("COPYBACK: copyback read failed. Aborting.\n");
  429                 (desc->writereq->CompleteFunc) (desc, -100);
  430         } else {
rf_DiskIOEnqueue(&(desc->raidPtr
  432                     ->Queues[desc->frow][desc->fcol]),
desc->writereq, RF_IO_NORMAL_PRIORITY);
  434         }
  435         return (0);
  436 }
  439 /*
  440  * Called at interrupt context when the write has completed.
  441  * At user level & in the kernel, wake up the copyback thread.
  442  * In the simulator, invoke the next copyback directly.
  443  * Can't free diskqueuedata structs in the kernel because we're at
  444  * interrupt context.
  445  */
  446 int
rf_CopybackWriteDoneProc(RF_CopybackDesc_t *desc, int status)
  448 {
  449         if (status && status != -100) {
printf("COPYBACK: copyback write failed. Aborting.\n");
  451         }
desc->status = status;
rf_MCPairWakeupFunc(desc->mcpair);
  454         return (0);
  455 }
  458 /* Invoked when the copyback has completed. */
  459 void
rf_CopybackComplete(RF_CopybackDesc_t *desc, int status)
  461 {
RF_Raid_t *raidPtr = desc->raidPtr;
  463         struct timeval t, diff;
  465         if (!status) {
RF_LOCK_MUTEX(raidPtr->mutex);
  467                 if (raidPtr->Layout.map->flags & RF_DISTRIBUTE_SPARE) {
RF_ASSERT(raidPtr->Layout.map->parityConfig == 'D');
rf_FreeSpareTable(raidPtr);
  470                 } else {
raidPtr->Disks[desc->spRow][desc->spCol].status =
rf_ds_spare;
  473                 }
RF_UNLOCK_MUTEX(raidPtr->mutex);
RF_GETTIME(t);
RF_TIMEVAL_DIFF(&desc->starttime, &t, &diff);
printf("Copyback time was %d.%06d seconds.\n",
  479                     (int) diff.tv_sec, (int) diff.tv_usec);
  480         } else
printf("COPYBACK: Failure.\n");
RF_Free(desc->databuf, rf_RaidAddressToByte(raidPtr, desc->sectPerSU));
rf_FreeMCPair(desc->mcpair);
RF_Free(desc, sizeof(*desc));
rf_copyback_in_progress = 0;
rf_ResumeNewRequests(raidPtr);
  489 }