/****************************************************************************************
* set up the data structures describing the spare disks in the array
* recall from the above comment that the spare disk descriptors are stored
* in row zero, which is specially expanded to hold them.
***************************************************************************************/
int
rf_ConfigureSpareDisks(
RF_ShutdownList_t ** listp,
RF_Raid_t * raidPtr,
RF_Config_t * cfgPtr)
{
char buf[256];
int r, c, i, ret;
RF_DiskOp_t *rdcap_op = NULL, *tur_op = NULL;
unsigned bs;
RF_RaidDisk_t *disks;
int num_spares_done;
struct proc *proc;
#if !defined(__NetBSD__) && !defined(__OpenBSD__)
ret = rf_SCSI_AllocReadCapacity(&rdcap_op);
if (ret)
goto fail;
ret = rf_SCSI_AllocTUR(&tur_op);
if (ret)
goto fail;
#endif /* !__NetBSD__ && !__OpenBSD__ */
num_spares_done = 0;
proc = raidPtr->proc;
/* The space for the spares should have already been allocated by
* ConfigureDisks() */
disks = &raidPtr->Disks[0][raidPtr->numCol];
for (i = 0; i < raidPtr->numSpare; i++) {
ret = rf_ConfigureDisk(raidPtr, &cfgPtr->spare_names[i][0],
&disks[i], rdcap_op, tur_op,
cfgPtr->spare_devs[i], 0, raidPtr->numCol + i);
if (ret)
goto fail;
if (disks[i].status != rf_ds_optimal) {
RF_ERRORMSG1("Warning: spare disk %s failed TUR\n", buf);
} else {
disks[i].status = rf_ds_spare; /* change status to
* spare */
DPRINTF6("Spare Disk %d: dev %s numBlocks %ld blockSize %d (%ld MB)\n", i,
disks[i].devname,
(long int) disks[i].numBlocks, disks[i].blockSize,
(long int) disks[i].numBlocks * disks[i].blockSize / 1024 / 1024);
}
num_spares_done++;
}
#if (defined(__NetBSD__) || defined(__OpenBSD__)) && (_KERNEL)
#else
rf_SCSI_FreeDiskOp(rdcap_op, 1);
rdcap_op = NULL;
rf_SCSI_FreeDiskOp(tur_op, 0);
tur_op = NULL;
#endif
/* check sizes and block sizes on spare disks */
bs = 1 << raidPtr->logBytesPerSector;
for (i = 0; i < raidPtr->numSpare; i++) {
if (disks[i].blockSize != bs) {
RF_ERRORMSG3("Block size of %d on spare disk %s is not the same as on other disks (%d)\n", disks[i].blockSize, disks[i].devname, bs);
ret = EINVAL;
goto fail;
}
if (disks[i].numBlocks < raidPtr->sectorsPerDisk) {
RF_ERRORMSG3("Spare disk %s (%d blocks) is too small to serve as a spare (need %ld blocks)\n",
disks[i].devname, disks[i].blockSize, (long int) raidPtr->sectorsPerDisk);
ret = EINVAL;
goto fail;
} else
if (disks[i].numBlocks > raidPtr->sectorsPerDisk) {
RF_ERRORMSG2("Warning: truncating spare disk %s to %ld blocks\n", disks[i].devname, (long int) raidPtr->sectorsPerDisk);
disks[i].numBlocks = raidPtr->sectorsPerDisk;
}
}
return (0);
fail:
#if (defined(__NetBSD__) || defined(__OpenBSD__)) && defined(_KERNEL)
/* Release the hold on the main components. We've failed to allocate
* a spare, and since we're failing, we need to free things.. */
for (r = 0; r < raidPtr->numRow; r++) {
for (c = 0; c < raidPtr->numCol; c++) {
/* Cleanup.. */
#ifdef DEBUG
printf("Cleaning up row: %d col: %d\n", r, c);
#endif
if (raidPtr->raid_cinfo[r][c].ci_vp) {
(void) vn_close(raidPtr->raid_cinfo[r][c].ci_vp,
FREAD | FWRITE, proc->p_ucred, proc);
}
}
}
for (i = 0; i < raidPtr->numSpare; i++) {
/* Cleanup.. */
#ifdef DEBUG
printf("Cleaning up spare: %d\n", i);
#endif
if (raidPtr->raid_cinfo[0][raidPtr->numCol + i].ci_vp) {
(void) vn_close(raidPtr->raid_cinfo[0][raidPtr->numCol + i].ci_vp,
FREAD | FWRITE, proc->p_ucred, proc);
}
}
#else
if (rdcap_op)
rf_SCSI_FreeDiskOp(rdcap_op, 1);
if (tur_op)
rf_SCSI_FreeDiskOp(tur_op, 0);
#endif
return (ret);
}
/****************************************************************************************
*
* initialize the disks comprising the array
*
* We want the spare disks to have regular row,col numbers so that we can easily
* substitue a spare for a failed disk. But, the driver code assumes throughout
* that the array contains numRow by numCol _non-spare_ disks, so it's not clear
* how to fit in the spares. This is an unfortunate holdover from raidSim. The
* quick and dirty fix is to make row zero bigger than the rest, and put all the
* spares in it. This probably needs to get changed eventually.
*
***************************************************************************************/
int
rf_ConfigureDisks(
RF_ShutdownList_t ** listp,
RF_Raid_t * raidPtr,
RF_Config_t * cfgPtr)
{
RF_RaidDisk_t **disks;
RF_SectorCount_t min_numblks = (RF_SectorCount_t) 0x7FFFFFFFFFFFLL;
RF_RowCol_t r, c;
int bs, ret;
unsigned i, count, foundone = 0, numFailuresThisRow;
RF_DiskOp_t *rdcap_op = NULL, *tur_op = NULL;
int num_rows_done, num_cols_done;
struct proc *proc = 0;
#if !defined(__NetBSD__) && !defined(__OpenBSD__)
ret = rf_SCSI_AllocReadCapacity(&rdcap_op);
if (ret)
goto fail;
ret = rf_SCSI_AllocTUR(&tur_op);
if (ret)
goto fail;
#endif /* !__NetBSD__ && !__OpenBSD__ */
num_rows_done = 0;
num_cols_done = 0;
RF_CallocAndAdd(disks, raidPtr->numRow, sizeof(RF_RaidDisk_t *), (RF_RaidDisk_t **), raidPtr->cleanupList);
if (disks == NULL) {
ret = ENOMEM;
goto fail;
}
raidPtr->Disks = disks;
proc = raidPtr->proc; /* Blah XXX */
/* get space for the device-specific stuff... */
RF_CallocAndAdd(raidPtr->raid_cinfo, raidPtr->numRow,
sizeof(struct raidcinfo *), (struct raidcinfo **),
raidPtr->cleanupList);
if (raidPtr->raid_cinfo == NULL) {
ret = ENOMEM;
goto fail;
}
for (r = 0; r < raidPtr->numRow; r++) {
numFailuresThisRow = 0;
RF_CallocAndAdd(disks[r], raidPtr->numCol + ((r == 0) ? raidPtr->numSpare : 0), sizeof(RF_RaidDisk_t), (RF_RaidDisk_t *), raidPtr->cleanupList);
if (disks[r] == NULL) {
ret = ENOMEM;
goto fail;
}
/* get more space for device specific stuff.. */
RF_CallocAndAdd(raidPtr->raid_cinfo[r],
raidPtr->numCol + ((r == 0) ? raidPtr->numSpare : 0),
sizeof(struct raidcinfo), (struct raidcinfo *),
raidPtr->cleanupList);
if (raidPtr->raid_cinfo[r] == NULL) {
ret = ENOMEM;
goto fail;
}
for (c = 0; c < raidPtr->numCol; c++) {
ret = rf_ConfigureDisk(raidPtr, &cfgPtr->devnames[r][c][0],
&disks[r][c], rdcap_op, tur_op,
cfgPtr->devs[r][c], r, c);
if (ret)
goto fail;
if (disks[r][c].status != rf_ds_optimal) {
numFailuresThisRow++;
} else {
if (disks[r][c].numBlocks < min_numblks)
min_numblks = disks[r][c].numBlocks;
DPRINTF7("Disk at row %d col %d: dev %s numBlocks %ld blockSize %d (%ld MB)\n",
r, c, disks[r][c].devname,
(long int) disks[r][c].numBlocks,
disks[r][c].blockSize,
(long int) disks[r][c].numBlocks * disks[r][c].blockSize / 1024 / 1024);
}
num_cols_done++;
}
/* XXX fix for n-fault tolerant */
if (numFailuresThisRow > 0)
raidPtr->status[r] = rf_rs_degraded;
num_rows_done++;
}
#if (defined(__NetBSD__) || defined(__OpenBSD__)) && defined(_KERNEL)
/* we do nothing */
#else
rf_SCSI_FreeDiskOp(rdcap_op, 1);
rdcap_op = NULL;
rf_SCSI_FreeDiskOp(tur_op, 0);
tur_op = NULL;
#endif
/* all disks must be the same size & have the same block size, bs must
* be a power of 2 */
bs = 0;
for (foundone = r = 0; !foundone && r < raidPtr->numRow; r++) {
for (c = 0; !foundone && c < raidPtr->numCol; c++) {
if (disks[r][c].status == rf_ds_optimal) {
bs = disks[r][c].blockSize;
foundone = 1;
}
}
}
if (!foundone) {
RF_ERRORMSG("RAIDFRAME: Did not find any live disks in the array.\n");
ret = EINVAL;
goto fail;
}
for (count = 0, i = 1; i; i <<= 1)
if (bs & i)
count++;
if (count != 1) {
RF_ERRORMSG1("Error: block size on disks (%d) must be a power of 2\n", bs);
ret = EINVAL;
goto fail;
}
for (r = 0; r < raidPtr->numRow; r++) {
for (c = 0; c < raidPtr->numCol; c++) {
if (disks[r][c].status == rf_ds_optimal) {
if (disks[r][c].blockSize != bs) {
RF_ERRORMSG2("Error: block size of disk at r %d c %d different from disk at r 0 c 0\n", r, c);
ret = EINVAL;
goto fail;
}
if (disks[r][c].numBlocks != min_numblks) {
RF_ERRORMSG3("WARNING: truncating disk at r %d c %d to %d blocks\n",
r, c, (int) min_numblks);
disks[r][c].numBlocks = min_numblks;
}
}
}
}
raidPtr->sectorsPerDisk = min_numblks;
raidPtr->logBytesPerSector = ffs(bs) - 1;
raidPtr->bytesPerSector = bs;
raidPtr->sectorMask = bs - 1;
return (0);
fail:
#if (defined(__NetBSD__) || defined(__OpenBSD__)) && defined(_KERNEL)
for (r = 0; r < raidPtr->numRow; r++) {
for (c = 0; c < raidPtr->numCol; c++) {
/* Cleanup.. */
#ifdef DEBUG
printf("Cleaning up row: %d col: %d\n", r, c);
#endif
if (raidPtr->raid_cinfo[r][c].ci_vp) {
(void) vn_close(raidPtr->raid_cinfo[r][c].ci_vp,
FREAD | FWRITE, proc->p_ucred, proc);
}
}
}
/* Space allocated for raid_vpp will get cleaned up at some other
* point */
/* XXX Need more #ifdefs in the above... */
#else
if (rdcap_op)
rf_SCSI_FreeDiskOp(rdcap_op, 1);
if (tur_op)
rf_SCSI_FreeDiskOp(tur_op, 0);
#endif
return (ret);
}
/* Do a complete copyback. */
void
rf_CopybackReconstructedData(RF_Raid_t *raidPtr)
{
RF_ComponentLabel_t c_label;
int done, retcode;
RF_CopybackDesc_t *desc;
RF_RowCol_t frow, fcol;
RF_RaidDisk_t *badDisk;
char *databuf;
struct partinfo dpart;
struct vnode *vp;
struct vattr va;
struct proc *proc;
int ac;
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];
proc = raidPtr->engine_thread;
/*
* This device may have been opened successfully the first time.
* Close it before trying to open it again.
*/
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;
}
/* 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);
if (retcode) {
printf("COPYBACK: raidlookup on device: %s failed: %d !\n",
raidPtr->Disks[frow][fcol].devname, retcode);
/*
* XXX The component isn't responding properly... Must be
* still dead :-(
*/
return;
} else {
/*
* Ok, so we can at least do a lookup...
* How about actually getting a vp for it ?
*/
if ((retcode = VOP_GETATTR(vp, &va, proc->p_ucred, proc)) != 0)
{
return;
}
retcode = VOP_IOCTL(vp, DIOCGPART, (caddr_t) &dpart, FREAD,
proc->p_ucred, proc);
if (retcode) {
return;
}
raidPtr->Disks[frow][fcol].blockSize = dpart.disklab->d_secsize;
raidPtr->Disks[frow][fcol].numBlocks = dpart.part->p_size -
rf_protectedSectors;
raidPtr->raid_cinfo[frow][fcol].ci_vp = vp;
raidPtr->raid_cinfo[frow][fcol].ci_dev = va.va_rdev;
/* XXX Or the above ? */
raidPtr->Disks[frow][fcol].dev = va.va_rdev;
/*
* We allow the user to specify that only a fraction of the
* disks should be used this is just for debug: it speeds up
* the parity scan.
*/
raidPtr->Disks[frow][fcol].numBlocks =
raidPtr->Disks[frow][fcol].numBlocks *
rf_sizePercentage / 100;
}
#if 0
/* This is the way it was done before the CAM stuff was removed. */
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);
#endif
if (retcode) {
printf("COPYBACK: target disk failed TUR.\n");
return;
}
/* 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;
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");
/*
* 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);
/*
* Data has been restored.
* Fix up the component label.
* Don't actually need the read here.
*/
raidread_component_label(raidPtr->raid_cinfo[frow][fcol].ci_dev,
raidPtr->raid_cinfo[frow][fcol].ci_vp,
&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,
&c_label);
}
