/*
* FUNCTION: meta_replicaslice()
* INPUT: dnp - the name of the drive to check
* OUTPUT: slicep - pointer to slice number
* ep - pointer to an md_error_t structure in which
* to return errors to the caller
* RETURNS: int - 0 - value pointed to by slicep is valid
* -1 - otherwise
*
* PURPOSE: Determine which slice of the specified drive to
* reserve, presumably for metadb replica usage.
*
* NOTE: If slicep is NULL, the return code will indicate
* whether or not the slice number could be determined
*/
int
meta_replicaslice(
mddrivename_t *dnp,
uint_t *slicep,
md_error_t *ep
)
{
int err = 0;
int ioctl_return;
int fd;
char *rname;
struct dk_geom geom;
rname = dnp->rname;
if ((fd = open(rname, (O_RDONLY|O_NDELAY), 0)) < 0) {
char *n;
int open_errno;
size_t len;
if (errno != ENOENT)
return (mdsyserror(ep, errno, rname));
len = strlen(rname) + 3;
n = Zalloc(len);
(void) snprintf(n, len, "%ss0", rname);
fd = open(n, (O_RDONLY|O_NDELAY), 0);
open_errno = errno;
Free(n);
if (fd < 0) {
return (mdsyserror(ep, open_errno, rname));
}
}
/*
* if our drivenamep points to a device not supporting
* DKIOCGGEOM, we have an EFI label.
*/
errno = 0;
ioctl_return = ioctl(fd, DKIOCGGEOM, &geom);
err = errno;
(void) close(fd);
/*
* If the DKIOCGGEOM ioctl succeeded, then the device has a
* VTOC style label. In this case, we use slice 7.
*/
if (ioctl_return == 0) {
if (slicep != NULL) {
*slicep = MD_SLICE7;
}
return (0);
}
/*
* ENOTSUP indicates an EFI style label, in which case slice 7
* cannot be used because its minor number is reserved. In
* this case, use slice 6.
*/
if (err == ENOTSUP) {
if (slicep != NULL) {
*slicep = MD_SLICE6;
}
return (0);
}
/*
* Those are the only two cases we know how to deal with;
* either the drivenamep didn't point to a disk, or the ioctl
* failed for some other reason.
*/
if (err == ENOTTY) {
return (mddeverror(ep, MDE_NOT_DISK, NODEV, rname));
}
return (mdsyserror(ep, err, rname));
}
/*
* FUNCTION: meta_repartition_drive()
* INPUT: sp - the set name for the device to check
* dnp - the name of the drive to partition
* options - options (see NOTES)
* OUTPUT: vtocp - pointer to an mdvtoc_t structure in which
* to return the new VTOC to the caller
* ep - pointer to an md_error_t structure in which
* to return errors to the caller
* RETURNS: int - 0 - drive was or can be repartitioned
* -1 - drive could not or should not be
* repartitioned
* PURPOSE: Repartition a disk for use in a disk set or in order
* to create soft partitions on it. Alternatively,
* return the VTOC that the disk would have if it were
* repartitioned without actually repartitioning it.
*
* NOTES:
*
* This routine will repartition a drive to make it suitable for
* inclusion in a diskset. Specifically, it will create a
* proposed VTOC that specifies a replica slice that begins at the
* first valid lba, is large enough to hold a label and a metadb
* replica, does not overlap any other slices, and is unmountable.
* If the current replica slice already satisfies those criteria,
* the routine will neither create a proposed VTOC nor repartition
* the drive unless the MD_REPART_FORCE flag is passed into the
* routine in the options argument. If the routine does create a
* proposed VTOC, it will return the proposed VTOC in *vtocp if
* vtocp isn't NULL.
*
* The slice to be used as the replica slice is determined by the
* function meta_replicaslice().
*
* If the replica slice does not satisfy the above criteria or the
* MD_REPART_FORCE flag is set, the proposed VTOC will specify a
* replica slice that satisfies the above criteria, a slice zero
* that contains the remaining space on the disk, and no other
* slices. If that repartitioning would cause the replica slice
* to move or shrink, and the MD_REPART_LEAVE_REP option is set,
* the routine will return -1 without creating or returning a
* proposed vtoc, and without repartitioning the disk. Otherwise
* the routine will repartition the disk unless the
* MD_REPART_DONT_LABEL flag is set in the options argument.
*
* If the MD_REPART_DONT_LABEL flag is set in the options argument,
* but the routine would otherwise repartition the drive, the
* routine won't repartition the drive, but will create a proposed
* VTOC that satisfies the criteria defined above and return it
* it in *vtocp if vtocp isn't NULL, The MD_REPART_DONT_LABEL
* option allows calling routines to determine what the contents of
* the drive's VTOC would be if the drive were repartitioned without
* actually repartitioning the drive.
*/
int
meta_repartition_drive(
mdsetname_t *sp,
mddrivename_t *dnp,
int options,
mdvtoc_t *vtocp,
md_error_t *ep
)
{
uint_t replicaslice;
diskaddr_t first_lba, last_lba;
int round_sizes = 1;
unsigned long long cylsize;
unsigned long long drvsize;
int i;
mdgeom_t *mdgp;
mdvtoc_t *mdvp;
mdvtoc_t proposed_vtoc;
uint_t reservedcyl;
ushort_t resflag;
mdname_t *resnp;
unsigned long long ressize;
md_set_desc *sd;
daddr_t dbsize;
diskaddr_t replica_start;
diskaddr_t replica_size;
diskaddr_t replica_end;
diskaddr_t data_start;
diskaddr_t data_size;
if (meta_replicaslice(dnp, &replicaslice, ep) != 0) {
return (-1);
}
/* Don't round for EFI disks */
if (replicaslice == MD_SLICE6)
round_sizes = 0;
/*
* We took as argument a drive name pointer, but we need a
* slice name pointer to retrieve vtoc information. So get
* the name pointer for slice zero first, then use it to get
* the vtoc info for the disk.
*/
if ((resnp = metaslicename(dnp, MD_SLICE0, ep)) == NULL)
return (-1);
if ((mdvp = metagetvtoc(resnp, FALSE, NULL, ep)) == NULL)
return (-1);
/*
* Determine the metadb size.
*/
dbsize = MD_DBSIZE;
if (!metaislocalset(sp)) {
if ((sd = metaget_setdesc(sp, ep)) == NULL)
return (-1);
if (MD_MNSET_DESC(sd))
dbsize = MD_MN_DBSIZE;
}
/* If we've got an efi disk, we better have lba info */
first_lba = mdvp->first_lba;
last_lba = mdvp->last_lba;
ASSERT((round_sizes != 0) || (last_lba > 0));
/*
* At this point, ressize is used as a minimum value. Later
* it will be rounded up to a cylinder boundary if
* appropriate. ressize is in units of disk sectors.
*/
ressize = dbsize + VTOC_SIZE;
resflag = V_UNMNT;
/*
* If we're forcing the repartition, we can skip the replica
* slice and overlap tests.
*/
if (options & MD_REPART_FORCE) {
goto do_repartition;
}
/*
* Replica slice tests: it must begin at first_lba, be long
* enough, have the right flags, and not overlap any other
* slices. If any of these conditions is violated, we need to
* repartition the disk.
*/
if (mdvp->parts[replicaslice].start != first_lba) {
goto do_repartition;
}
if (mdvp->parts[replicaslice].size < ressize) {
goto do_repartition;
}
if (mdvp->parts[replicaslice].flag != resflag) {
goto do_repartition;
}
/*
* Check for overlap: this test should use the actual size of
* the replica slice, as contained in the vtoc, and NOT the
* minimum size calculated above.
*/
replica_end = first_lba + mdvp->parts[replicaslice].size;
for (i = 0; i < mdvp->nparts; i++) {
if (i != replicaslice) {
if ((mdvp->parts[i].size > 0) &&
(mdvp->parts[i].start < replica_end)) {
goto do_repartition;
}
}
}
/*
* If we passed the above tests, then the disk is already
* partitioned appropriately, and we're not being told to
* force a change.
*/
return (0);
do_repartition:
/* Retrieve disk geometry info and round to cylinder sizes */
if (round_sizes != 0) {
if ((mdgp = metagetgeom(resnp, ep)) == NULL)
return (-1);
/*
* Both cylsize and drvsize are in units of disk
* sectors.
*
* The intended results are of type unsigned long
* long. Since each operand of the first
* multiplication is of type unsigned int, we risk
* overflow by multiplying and then converting the
* result. Therefore we explicitly cast (at least)
* one of the operands, forcing conversion BEFORE
* multiplication, and avoiding overflow. The second
* assignment is OK, since one of the operands is
* already of the desired type.
*/
cylsize =
((unsigned long long)mdgp->nhead) * mdgp->nsect;
drvsize = cylsize * mdgp->ncyl;
/*
* How many cylinders must we reserve for the replica
* slice to ensure that it meets the previously
* calculated minimum size?
*/
reservedcyl = (ressize + cylsize - 1) / cylsize;
ressize = reservedcyl * cylsize;
} else {
drvsize = last_lba - first_lba;
}
/* Would this require a forbidden change? */
if (options & MD_REPART_LEAVE_REP) {
if ((mdvp->parts[replicaslice].start != first_lba) ||
(mdvp->parts[replicaslice].size < ressize)) {
return (mddeverror(ep, MDE_REPART_REPLICA,
resnp->dev, NULL));
}
}
/*
* It seems unlikely that someone would pass us too small a
* disk, but it's still worth checking for...
*/
if (((round_sizes != 0) && (reservedcyl >= (int)mdgp->ncyl)) ||
((round_sizes == 0) && (ressize + first_lba >= last_lba))) {
return (mdmddberror(ep, MDE_DB_TOOSMALL,
meta_getminor(resnp->dev), sp->setno, 0, NULL));
}
replica_start = first_lba;
replica_size = ressize;
data_start = first_lba + ressize;
data_size = drvsize - ressize;
/*
* Create the proposed VTOC. First copy the current VTOC
* into the proposed VTOC to duplicate the values that don't
* need to change. Then change the partition table and set
* the flag value for the replica slice to resflag to reserve it
* for metadata.
*/
proposed_vtoc = *mdvp;
/* We need at least replicaslice partitions in the proposed vtoc */
if (replicaslice >= proposed_vtoc.nparts) {
proposed_vtoc.nparts = replicaslice + 1;
}
for (i = 0; i < proposed_vtoc.nparts; i++) {
/* don't change the reserved partition of an EFI device */
if (proposed_vtoc.parts[i].tag == V_RESERVED)
data_size = proposed_vtoc.parts[i].start - data_start;
else
(void) memset(&proposed_vtoc.parts[i], '\0',
sizeof (proposed_vtoc.parts[i]));
}
proposed_vtoc.parts[MD_SLICE0].start = data_start;
proposed_vtoc.parts[MD_SLICE0].size = data_size;
proposed_vtoc.parts[MD_SLICE0].tag = V_USR;
proposed_vtoc.parts[replicaslice].start = replica_start;
proposed_vtoc.parts[replicaslice].size = replica_size;
proposed_vtoc.parts[replicaslice].flag = resflag;
proposed_vtoc.parts[replicaslice].tag = V_USR;
if (!(options & MD_REPART_DONT_LABEL)) {
/*
* Label the disk with the proposed VTOC.
*/
*mdvp = proposed_vtoc;
if (metasetvtoc(resnp, ep) != 0) {
return (-1);
}
}
if (vtocp != NULL) {
/*
* Return the proposed VTOC.
*/
*vtocp = proposed_vtoc;
}
return (0);
}
static int
seths_enable(set_hs_params_t *shs)
{
hot_spare_t *hs;
mddb_recid_t recids[2];
set_t setno = shs->md_driver.md_setno;
mdkey_t key_old;
int num_keys_old = 0;
/*
* Find device by using key associated with shs_component_old.
* If unable to find a unique key for shs_component_old
* then fail since namespace has multiple entries
* for this old component and we're unable to determine
* which key is the valid match for shs_component_old.
* This failure keeps a hotspare from being enabled on a slice
* that may already be in use by another metadevice.
*/
if (md_getkeyfromdev(setno, mddb_getsidenum(setno),
shs->shs_component_old, &key_old, &num_keys_old) != 0) {
return (mddeverror(&shs->mde, MDE_NAME_SPACE,
shs->shs_component_old));
}
/*
* If more than one key matches given old_dev - fail command
* since unable to determine which key is correct.
*/
if (num_keys_old > 1) {
return (mddeverror(&shs->mde, MDE_MULTNM,
shs->shs_component_old));
}
/*
* If there is no key for this entry then fail since
* a key for this entry should exist.
*/
if (num_keys_old == 0) {
return (mddeverror(&shs->mde, MDE_INVAL_HS,
shs->shs_component_old));
}
/* Scan the hot spare list for the hs */
hs = (hot_spare_t *)md_set[setno].s_hs;
while (hs) {
/*
* Since component may or may not be currently in the system,
* use the keys to find a match (not the devt).
*/
if (hs->hs_key == key_old) {
break;
}
hs = hs->hs_next;
}
if (hs == NULL) {
return (mddeverror(&shs->mde, MDE_INVAL_HS,
shs->shs_component_old));
}
/* make sure it's broken */
if (hs->hs_state != HSS_BROKEN) {
return (mddeverror(&shs->mde, MDE_FIX_INVAL_HS_STATE,
hs->hs_devnum));
}
/* In case of a dryrun, we're done here */
if (shs->shs_options & HS_OPT_DRYRUN) {
return (0);
}
/* fix it */
set_hot_spare_state(hs, HSS_AVAILABLE);
hs->hs_start_blk = shs->shs_start_blk;
hs->hs_has_label = shs->shs_has_label;
hs->hs_number_blks = shs->shs_number_blks;
/* commit the db records */
recids[0] = hs->hs_record_id;
recids[1] = 0;
mddb_commitrecs_wrapper(recids);
SE_NOTIFY(EC_SVM_STATE, ESC_SVM_ENABLE, SVM_TAG_HS, setno,
shs->shs_component_old);
return (0);
}
static int
seths_replace(set_hs_params_t *shs)
{
hot_spare_t *hs;
hot_spare_t *prev_hs;
hot_spare_t *new_hs;
hot_spare_pool_t *hsp;
int new_found = 0;
mddb_recid_t recid;
mddb_recid_t recids[5];
int i;
sv_dev_t sv;
int delete_hs = 0;
set_t setno;
mddb_type_t typ1;
mdkey_t key_old;
int num_keys_old = 0;
setno = HSP_SET(shs->shs_hot_spare_pool);
typ1 = (mddb_type_t)md_getshared_key(setno,
hotspares_md_ops.md_driver.md_drivername);
/* Scan the hot spare list */
hs = (hot_spare_t *)md_set[setno].s_hs;
prev_hs = (hot_spare_t *)0;
while (hs) {
if (hs->hs_devnum == shs->shs_component_old) {
break;
}
prev_hs = hs;
hs = hs->hs_next;
}
if (hs == NULL) {
/*
* Unable to find device using devnum so use
* key associated with shs_component_old instead.
* If unable to find a unique key for shs_component_old
* then fail since namespace has multiple entries
* for this old component and we're unable to determine
* which key is the valid match for shs_component_old.
*
* Only need to compare keys when hs_devnum is NODEV.
*/
if (md_getkeyfromdev(setno, mddb_getsidenum(setno),
shs->shs_component_old, &key_old, &num_keys_old) != 0) {
return (mddeverror(&shs->mde, MDE_NAME_SPACE,
shs->shs_component_old));
}
/*
* If more than one key matches given old_dev - fail command
* since unable to determine which key is correct.
*/
if (num_keys_old > 1) {
return (mddeverror(&shs->mde, MDE_MULTNM,
shs->shs_component_old));
}
/*
* If there is no key for this entry then fail since
* a key for this entry should exist.
*/
if (num_keys_old == 0) {
return (mddeverror(&shs->mde, MDE_INVAL_HS,
shs->shs_component_old));
}
/* Scan the hot spare list again */
hs = (hot_spare_t *)md_set[setno].s_hs;
prev_hs = (hot_spare_t *)0;
while (hs) {
/*
* Only need to compare keys when hs_devnum is NODEV.
*/
if ((hs->hs_devnum == NODEV64) &&
(hs->hs_key == key_old)) {
break;
}
prev_hs = hs;
hs = hs->hs_next;
}
}
if (hs == NULL) {
return (mddeverror(&shs->mde, MDE_INVAL_HS,
shs->shs_component_old));
}
/* check the force flag and the state of the hot spare */
if (((shs->shs_options & HS_OPT_FORCE) == 0) &&
(hs->hs_state == HSS_RESERVED)) {
return (mdhserror(&shs->mde, MDE_HS_RESVD,
shs->shs_hot_spare_pool,
hs->hs_devnum));
}
/* Scan the hot spare pool list */
hsp = find_hot_spare_pool(setno, shs->shs_hot_spare_pool);
if (hsp == (hot_spare_pool_t *)0) {
return (mdhsperror(&shs->mde, MDE_INVAL_HSP,
shs->shs_hot_spare_pool));
}
/*
* Make sure the old device is in the pool.
*/
for (i = 0; i < hsp->hsp_nhotspares; i++) {
if (hsp->hsp_hotspares[i] == hs->hs_record_id) {
break;
}
}
if (i >= hsp->hsp_nhotspares) {
return (mddeverror(&shs->mde, MDE_INVAL_HS,
hs->hs_devnum));
}
/* Scan the hot spare list for the new hs */
new_hs = (hot_spare_t *)md_set[setno].s_hs;
new_found = 0;
while (new_hs) {
if (new_hs->hs_devnum == shs->shs_component_new) {
new_found = 1;
break;
}
new_hs = new_hs->hs_next;
}
/*
* Make sure the new device is not already in the pool.
* We don't have to search the hs in this hsp, if the
* new hs was just created. Only if the hot spare was found.
*/
if (new_found) {
for (i = 0; i < hsp->hsp_nhotspares; i++)
if (hsp->hsp_hotspares[i] == new_hs->hs_record_id) {
return (mdhserror(&shs->mde, MDE_HS_INUSE,
shs->shs_hot_spare_pool,
new_hs->hs_devnum));
}
}
/* In case of a dryrun, we're done here */
if (shs->shs_options & HS_OPT_DRYRUN) {
return (0);
}
/*
* Create the new hotspare
*/
if (!new_found) {
/* create a hot spare record */
if (shs->shs_size_option & MD_CRO_64BIT) {
#if defined(_ILP32)
return (mdhserror(&shs->mde, MDE_HS_UNIT_TOO_LARGE,
shs->shs_hot_spare_pool, shs->shs_component_new));
#else
recid = mddb_createrec(HS_ONDSK_STR_SIZE, typ1, HS_REC,
MD_CRO_64BIT | MD_CRO_HOTSPARE, setno);
#endif
} else {
recid = mddb_createrec(HS_ONDSK_STR_SIZE, typ1, HS_REC,
MD_CRO_32BIT | MD_CRO_HOTSPARE, setno);
}
if (recid < 0) {
return (mdhserror(&shs->mde, MDE_HS_CREATE_FAILURE,
shs->shs_hot_spare_pool,
shs->shs_component_new));
}
/* get the addr */
new_hs = (hot_spare_t *)mddb_getrecaddr_resize(recid,
sizeof (*new_hs), 0);
new_hs->hs_record_id = recid;
new_hs->hs_devnum = shs->shs_component_new;
new_hs->hs_key = shs->shs_key_new;
new_hs->hs_start_blk = shs->shs_start_blk;
new_hs->hs_has_label = shs->shs_has_label;
new_hs->hs_number_blks = shs->shs_number_blks;
set_hot_spare_state(new_hs, HSS_AVAILABLE);
new_hs->hs_refcount = 0;
new_hs->hs_isopen = 1;
}
/* lock the db records */
recids[0] = hs->hs_record_id;
recids[1] = new_hs->hs_record_id;
recids[2] = hsp->hsp_record_id;
recids[3] = 0;
sv.setno = setno;
sv.key = hs->hs_key;
hs->hs_refcount--;
if (hs->hs_refcount == 0) {
/*
* NOTE: We do not commit the previous hot spare record.
* There is no need, the link we get rebuilt at boot time.
*/
if (prev_hs) {
prev_hs->hs_next = hs->hs_next;
} else
md_set[setno].s_hs = (void *) hs->hs_next;
/* mark hs to be deleted in the correct order */
delete_hs = 1;
recids[0] = new_hs->hs_record_id;
recids[1] = hsp->hsp_record_id;
recids[2] = 0;
}
/* link into the hs list */
new_hs->hs_refcount++;
if (!new_found) {
/* do this AFTER the old dev is possibly removed */
new_hs->hs_next = (hot_spare_t *)md_set[setno].s_hs;
md_set[setno].s_hs = (void *) new_hs;
}
/* find the location of the old hs in the hsp */
for (i = 0; i < hsp->hsp_nhotspares; i++) {
if (hsp->hsp_hotspares[i] == hs->hs_record_id) {
hsp->hsp_hotspares[i] = new_hs->hs_record_id;
break;
}
}
if (shs->shs_size_option & MD_CRO_64BIT) {
new_hs->hs_revision |= MD_64BIT_META_DEV;
} else {
new_hs->hs_revision &= ~MD_64BIT_META_DEV;
}
/* commit the db records */
mddb_commitrecs_wrapper(recids);
if (delete_hs)
mddb_deleterec_wrapper(hs->hs_record_id);
md_rem_names(&sv, 1);
SE_NOTIFY(EC_SVM_CONFIG, ESC_SVM_REPLACE, SVM_TAG_HSP, setno,
md_expldev(hsp->hsp_self_id));
return (0);
}
static int
seths_delete(set_hs_params_t *shs)
{
hot_spare_t *hs;
hot_spare_t *prev_hs;
hot_spare_pool_t *hsp;
mddb_recid_t recids[4];
int i;
set_t setno;
sv_dev_t sv;
int delete_hs = 0;
mdkey_t key_old;
int num_keys_old = 0;
/* delete the hot spare pool */
if (shs->shs_options & HS_OPT_POOL) {
return (seths_delete_hsp(shs));
}
setno = HSP_SET(shs->shs_hot_spare_pool);
/* Scan the hot spare list */
hs = (hot_spare_t *)md_set[setno].s_hs;
prev_hs = (hot_spare_t *)0;
while (hs) {
if (hs->hs_devnum == shs->shs_component_old) {
break;
}
prev_hs = hs;
hs = hs->hs_next;
}
if (hs == NULL) {
/*
* Unable to find device using devnum so use
* key associated with shs_component_old instead.
* If unable to find a unique key for shs_component_old
* then fail since namespace has multiple entries
* for this old component and we're unable to determine
* which key is the valid match for shs_component_old.
*
* Only need to compare keys when hs_devnum is NODEV.
*/
if (md_getkeyfromdev(setno, mddb_getsidenum(setno),
shs->shs_component_old, &key_old, &num_keys_old) != 0) {
return (mddeverror(&shs->mde, MDE_NAME_SPACE,
shs->shs_component_old));
}
/*
* If more than one key matches given old_dev - fail command
* since shouldn't add new hotspare if namespace has
* multiple entries.
*/
if (num_keys_old > 1) {
return (mddeverror(&shs->mde, MDE_MULTNM,
shs->shs_component_old));
}
/*
* If there is no key for this entry then fail since
* a key for this entry should exist.
*/
if (num_keys_old == 0) {
return (mddeverror(&shs->mde, MDE_INVAL_HS,
shs->shs_component_old));
}
/* Scan the hot spare list again */
hs = (hot_spare_t *)md_set[setno].s_hs;
prev_hs = (hot_spare_t *)0;
while (hs) {
/*
* Only need to compare keys when hs_devnum is NODEV.
*/
if ((hs->hs_devnum == NODEV64) &&
(hs->hs_key == key_old)) {
break;
}
prev_hs = hs;
hs = hs->hs_next;
}
}
if (hs == NULL) {
return (mddeverror(&shs->mde, MDE_INVAL_HS,
shs->shs_component_old));
}
/* Scan the hot spare pool list */
hsp = find_hot_spare_pool(setno, shs->shs_hot_spare_pool);
if (hsp == (hot_spare_pool_t *)0) {
return (mdhsperror(&shs->mde, MDE_INVAL_HSP,
shs->shs_hot_spare_pool));
}
/* check for force flag and state of hot spare */
if (((shs->shs_options & HS_OPT_FORCE) == 0) &&
(hs->hs_state == HSS_RESERVED)) {
return (mdhserror(&shs->mde, MDE_HS_RESVD,
shs->shs_hot_spare_pool, shs->shs_component_old));
}
if (hsp->hsp_refcount && (hs->hs_state == HSS_RESERVED)) {
return (mdhserror(&shs->mde, MDE_HS_RESVD,
shs->shs_hot_spare_pool, shs->shs_component_old));
}
/*
* Make sure the device is in the pool.
*/
for (i = 0; i < hsp->hsp_nhotspares; i++) {
if (hsp->hsp_hotspares[i] == hs->hs_record_id) {
break;
}
}
if (i >= hsp->hsp_nhotspares) {
return (mddeverror(&shs->mde, MDE_INVAL_HS,
hs->hs_devnum));
}
/* In case of a dryrun, we're done here */
if (shs->shs_options & HS_OPT_DRYRUN) {
return (0);
}
/* lock the db records */
recids[0] = hs->hs_record_id;
recids[1] = hsp->hsp_record_id;
recids[2] = 0;
sv.setno = setno;
sv.key = hs->hs_key;
hs->hs_refcount--;
if (hs->hs_refcount == 0) {
/*
* NOTE: We do not commit the previous hot spare record.
* There is no need, the link we get rebuilt at boot time.
*/
if (prev_hs) {
prev_hs->hs_next = hs->hs_next;
} else
md_set[setno].s_hs = (void *) hs->hs_next;
/* mark the hot spare to be deleted */
delete_hs = 1;
recids[0] = hsp->hsp_record_id;
recids[1] = 0;
}
/* find the location of the hs in the hsp */
for (i = 0; i < hsp->hsp_nhotspares; i++) {
if (hsp->hsp_hotspares[i] == hs->hs_record_id)
break;
}
/* remove the hs from the hsp */
for (i++; i < hsp->hsp_nhotspares; i++)
hsp->hsp_hotspares[i - 1] = hsp->hsp_hotspares[i];
hsp->hsp_nhotspares--;
/* commit the db records */
mddb_commitrecs_wrapper(recids);
if (delete_hs)
mddb_deleterec_wrapper(hs->hs_record_id);
md_rem_names(&sv, 1);
SE_NOTIFY(EC_SVM_CONFIG, ESC_SVM_REMOVE, SVM_TAG_HSP, setno,
md_expldev(hsp->hsp_self_id));
return (0);
}
static int
seths_add(set_hs_params_t *shs)
{
hot_spare_t *hs;
hot_spare_pool_t *hsp;
hot_spare_pool_t *prev_hsp;
hot_spare_pool_t *new_hsp;
hot_spare_pool_t *old_hsp;
md_create_rec_option_t options;
mddb_recid_t recid;
mddb_recid_t recids[5];
size_t new_size;
int i;
int delete_hsp = 0;
int irecid;
set_t setno;
mddb_type_t typ1;
int hsp_created = 0;
mdkey_t key_old;
int num_keys_old = 0;
/* Not much to do here in case of a dryrun */
if (shs->shs_options & HS_OPT_DRYRUN) {
return (0);
}
/* create an empty hot spare pool */
if (shs->shs_options & HS_OPT_POOL) {
return (seths_create_hsp(shs));
}
setno = HSP_SET(shs->shs_hot_spare_pool);
typ1 = (mddb_type_t)md_getshared_key(setno,
hotspares_md_ops.md_driver.md_drivername);
/* Scan the hot spare list */
hs = (hot_spare_t *)md_set[setno].s_hs;
while (hs) {
if (hs->hs_devnum == shs->shs_component_old) {
break;
}
hs = hs->hs_next;
}
if (hs == NULL) {
/*
* Did not find match for device using devnum so use
* key associated with shs_component_old just
* in case there is a match but the match's dev is NODEV.
* If unable to find a unique key for shs_component_old
* then fail since namespace has multiple entries
* for this old component and we shouldn't allow
* an addition of a hotspare in this case.
*/
if (md_getkeyfromdev(setno, mddb_getsidenum(setno),
shs->shs_component_old, &key_old, &num_keys_old) != 0) {
return (mddeverror(&shs->mde, MDE_NAME_SPACE,
shs->shs_component_old));
}
/*
* If more than one key matches given old_dev - fail command
* since shouldn't add new hotspare if namespace has
* multiple entries.
*/
if (num_keys_old > 1) {
return (mddeverror(&shs->mde, MDE_MULTNM,
shs->shs_component_old));
}
/*
* If there is no key for this entry then fail since
* a key for this entry should exist.
*/
if (num_keys_old == 0) {
return (mddeverror(&shs->mde, MDE_INVAL_HS,
shs->shs_component_old));
}
/* Scan the hot spare list again */
hs = (hot_spare_t *)md_set[setno].s_hs;
while (hs) {
/*
* Only need to compare keys when hs_devnum is NODEV.
*/
if ((hs->hs_devnum == NODEV64) &&
(hs->hs_key == key_old)) {
break;
}
hs = hs->hs_next;
}
}
if (hs == NULL) {
/* create a hot spare record */
if (shs->shs_size_option & MD_CRO_64BIT) {
#if defined(_ILP32)
return (mdhserror(&shs->mde, MDE_HS_UNIT_TOO_LARGE,
shs->shs_hot_spare_pool, shs->shs_component_old));
#else
recid = mddb_createrec(HS_ONDSK_STR_SIZE, typ1, HS_REC,
MD_CRO_64BIT | MD_CRO_HOTSPARE, setno);
#endif
} else {
recid = mddb_createrec(HS_ONDSK_STR_SIZE, typ1, HS_REC,
MD_CRO_32BIT | MD_CRO_HOTSPARE, setno);
}
if (recid < 0) {
return (mdhserror(&shs->mde, MDE_HS_CREATE_FAILURE,
shs->shs_hot_spare_pool,
shs->shs_component_old));
}
/* get the addr */
hs = (hot_spare_t *)mddb_getrecaddr_resize(recid, sizeof (*hs),
0);
hs->hs_record_id = recid;
hs->hs_devnum = shs->shs_component_old;
hs->hs_key = shs->shs_key_old;
hs->hs_start_blk = shs->shs_start_blk;
hs->hs_has_label = shs->shs_has_label;
hs->hs_number_blks = shs->shs_number_blks;
set_hot_spare_state(hs, HSS_AVAILABLE);
hs->hs_refcount = 0;
hs->hs_next = (hot_spare_t *)md_set[setno].s_hs;
md_set[setno].s_hs = (void *) hs;
}
/* Scan the hot spare pool list */
hsp = (hot_spare_pool_t *)md_set[setno].s_hsp;
prev_hsp = (hot_spare_pool_t *)0;
while (hsp) {
if (hsp->hsp_self_id == shs->shs_hot_spare_pool) {
break;
}
prev_hsp = hsp;
hsp = hsp->hsp_next;
}
if (hsp == NULL) {
/* create a hot spare pool record */
recid = mddb_createrec(sizeof (hot_spare_pool_ond_t),
typ1, HSP_REC,
MD_CRO_32BIT | MD_CRO_HOTSPARE_POOL | MD_CRO_FN, setno);
if (recid < 0) {
return (mdhsperror(&shs->mde, MDE_HSP_CREATE_FAILURE,
shs->shs_hot_spare_pool));
}
/* get the record addr */
hsp = (hot_spare_pool_t *)mddb_getrecaddr_resize(recid,
sizeof (*hsp), HSP_ONDSK_STR_OFF);
hsp->hsp_self_id = shs->shs_hot_spare_pool;
hsp->hsp_record_id = recid;
hsp->hsp_next = (hot_spare_pool_t *)md_set[setno].s_hsp;
hsp->hsp_refcount = 0;
hsp->hsp_nhotspares = 0;
hsp->hsp_revision |= MD_FN_META_DEV;
/* force prev_hsp to NULL, this will cause hsp to be linked */
prev_hsp = (hot_spare_pool_t *)0;
rw_enter(&hotspares_md_ops.md_link_rw.lock, RW_WRITER);
hsp->hsp_link.ln_next = hotspares_md_ops.md_head;
hsp->hsp_link.ln_setno = setno;
hsp->hsp_link.ln_id = hsp->hsp_self_id;
hotspares_md_ops.md_head = &hsp->hsp_link;
rw_exit(&hotspares_md_ops.md_link_rw.lock);
hsp_created = 1;
} else {
/*
* Make sure the hot spare is not already in the pool.
*/
for (i = 0; i < hsp->hsp_nhotspares; i++)
if (hsp->hsp_hotspares[i] == hs->hs_record_id) {
return (mdhserror(&shs->mde, MDE_HS_INUSE,
shs->shs_hot_spare_pool,
hs->hs_devnum));
}
/*
* Create a new hot spare pool record
* This gives us the one extra hs slot,
* because there is one slot in the
* hot_spare_pool struct
*/
new_size = sizeof (hot_spare_pool_ond_t) +
(sizeof (mddb_recid_t) * hsp->hsp_nhotspares);
/*
* The Friendly Name status of the new HSP should duplicate
* the status of the existing one.
*/
if (hsp->hsp_revision & MD_FN_META_DEV) {
options =
MD_CRO_32BIT | MD_CRO_HOTSPARE_POOL | MD_CRO_FN;
} else {
options = MD_CRO_32BIT | MD_CRO_HOTSPARE_POOL;
}
recid = mddb_createrec(new_size, typ1, HSP_REC, options, setno);
if (recid < 0) {
return (mdhsperror(&shs->mde, MDE_HSP_CREATE_FAILURE,
hsp->hsp_self_id));
}
new_size = sizeof (hot_spare_pool_t) +
(sizeof (mddb_recid_t) * hsp->hsp_nhotspares);
/* get the record addr */
new_hsp = (hot_spare_pool_t *)mddb_getrecaddr_resize(recid,
new_size, HSP_ONDSK_STR_OFF);
/* copy the old record into the new one */
bcopy((caddr_t)hsp, (caddr_t)new_hsp,
(size_t)((sizeof (hot_spare_pool_t) +
(sizeof (mddb_recid_t) * hsp->hsp_nhotspares)
- sizeof (mddb_recid_t))));
new_hsp->hsp_record_id = recid;
md_rem_link(setno, hsp->hsp_self_id,
&hotspares_md_ops.md_link_rw.lock,
&hotspares_md_ops.md_head);
rw_enter(&hotspares_md_ops.md_link_rw.lock, RW_WRITER);
new_hsp->hsp_link.ln_next = hotspares_md_ops.md_head;
new_hsp->hsp_link.ln_setno = setno;
new_hsp->hsp_link.ln_id = new_hsp->hsp_self_id;
hotspares_md_ops.md_head = &new_hsp->hsp_link;
rw_exit(&hotspares_md_ops.md_link_rw.lock);
/* mark the old hsp to be deleted */
delete_hsp = 1;
old_hsp = hsp;
hsp = new_hsp;
}
if (shs->shs_size_option & MD_CRO_64BIT) {
hs->hs_revision |= MD_64BIT_META_DEV;
} else {
hs->hs_revision &= ~MD_64BIT_META_DEV;
}
/* lock the db records */
recids[0] = hs->hs_record_id;
recids[1] = hsp->hsp_record_id;
irecid = 2;
if (delete_hsp)
recids[irecid++] = old_hsp->hsp_record_id;
recids[irecid] = 0;
/* increment the reference count */
hs->hs_refcount++;
/* add the hs at the end of the hot spare pool */
hsp->hsp_hotspares[hsp->hsp_nhotspares] = hs->hs_record_id;
hsp->hsp_nhotspares++;
/*
* NOTE: We do not commit the previous hot spare pool record.
* There is no need, the link gets rebuilt at boot time.
*/
if (prev_hsp)
prev_hsp->hsp_next = hsp;
else
md_set[setno].s_hsp = (void *) hsp;
if (delete_hsp)
old_hsp->hsp_self_id = MD_HSP_NONE;
/* commit the db records */
mddb_commitrecs_wrapper(recids);
if (delete_hsp) {
/* delete the old hot spare pool record */
mddb_deleterec_wrapper(old_hsp->hsp_record_id);
}
if (hsp_created) {
SE_NOTIFY(EC_SVM_CONFIG, ESC_SVM_CREATE, SVM_TAG_HSP, setno,
md_expldev(hsp->hsp_self_id));
}
SE_NOTIFY(EC_SVM_CONFIG, ESC_SVM_ADD, SVM_TAG_HSP, setno,
md_expldev(hsp->hsp_self_id));
return (0);
}
