static int
stripe_reset(md_i_reset_t *mirp)
{
minor_t mnum = mirp->mnum;
ms_unit_t *un;
mdi_unit_t *ui;
set_t setno = MD_MIN2SET(mnum);
mdclrerror(&mirp->mde);
if ((setno >= md_nsets) || (MD_MIN2UNIT(mnum) >= md_nunits))
return (mdmderror(&mirp->mde, MDE_INVAL_UNIT, mnum));
if (md_get_setstatus(setno) & MD_SET_STALE)
return (mdmddberror(&mirp->mde, MDE_DB_STALE, mnum, setno));
un = MD_UNIT(mnum);
if (un == NULL) {
return (mdmderror(&mirp->mde, MDE_UNIT_NOT_SETUP, mnum));
}
/* This prevents new opens */
rw_enter(&md_unit_array_rw.lock, RW_WRITER);
if (MD_HAS_PARENT(un->c.un_parent)) {
rw_exit(&md_unit_array_rw.lock);
return (mdmderror(&mirp->mde, MDE_IN_USE, mnum));
}
/* single thread */
ui = MDI_UNIT(mnum);
un = md_unit_openclose_enter(ui);
if (md_unit_isopen(ui)) {
md_unit_openclose_exit(ui);
rw_exit(&md_unit_array_rw.lock);
return (mdmderror(&mirp->mde, MDE_IS_OPEN, mnum));
}
md_unit_openclose_exit(ui);
reset_stripe(un, mnum, 1);
/*
* Update unit availability
*/
md_set[setno].s_un_avail++;
/*
* If MN set, reset s_un_next so all nodes can have
* the same view of the next available slot when
* nodes are -w and -j
*/
if (MD_MNSET_SETNO(setno)) {
(void) md_upd_set_unnext(setno, MD_MIN2UNIT(mnum));
}
rw_exit(&md_unit_array_rw.lock);
return (0);
}
/*
* NAME: raid_resync_unit
*
* DESCRIPTION: RAID metadevice specific resync routine.
* Open the unit and start resync_unit as a separate thread.
*
* PARAMETERS: minor_t mnum - minor number identity of metadevice
* md_error_t *ep - output error parameter
*
* RETURN: On error return 1 or set ep to nonzero, otherwise return 0.
*
* LOCKS: Acquires and releases Unit Writer Lock.
*/
int
raid_resync_unit(
minor_t mnum,
md_error_t *ep
)
{
mdi_unit_t *ui;
set_t setno = MD_MIN2SET(mnum);
mr_unit_t *un;
ui = MDI_UNIT(mnum);
un = MD_UNIT(mnum);
if (md_get_setstatus(setno) & MD_SET_STALE)
return (mdmddberror(ep, MDE_DB_STALE, mnum, setno));
ASSERT(un->un_column[un->un_resync_index].un_devflags &
(MD_RAID_COPY_RESYNC | MD_RAID_REGEN_RESYNC));
/* Don't start a resync if the device is not available */
if ((ui == NULL) || (ui->ui_tstate & MD_DEV_ERRORED)) {
return (mdmderror(ep, MDE_RAID_OPEN_FAILURE, mnum));
}
if (raid_internal_open(mnum, FREAD | FWRITE, OTYP_LYR, 0)) {
(void) md_unit_writerlock(ui);
release_resync_request(mnum);
md_unit_writerexit(ui);
SE_NOTIFY(EC_SVM_STATE, ESC_SVM_OPEN_FAIL, SVM_TAG_METADEVICE,
setno, MD_SID(un));
return (mdmderror(ep, MDE_RAID_OPEN_FAILURE, mnum));
}
/* start resync_unit thread */
(void) thread_create(NULL, 0, resync_unit, (void *)(uintptr_t)mnum,
0, &p0, TS_RUN, minclsyspri);
return (0);
}
static int
set_hs(
set_hs_params_t *shs
)
{
mdclrerror(&shs->mde);
if (md_get_setstatus(shs->md_driver.md_setno) & MD_SET_STALE)
return (mdmddberror(&shs->mde, MDE_DB_STALE, NODEV32,
shs->md_driver.md_setno));
switch (shs->shs_cmd) {
case ADD_HOT_SPARE:
return (seths_add(shs));
case DELETE_HOT_SPARE:
return (seths_delete(shs));
case REPLACE_HOT_SPARE:
return (seths_replace(shs));
case FIX_HOT_SPARE:
return (seths_enable(shs));
default:
return (mderror(&shs->mde, MDE_INVAL_HSOP));
}
}
static int
stripe_change(
md_stripe_params_t *msp,
IOLOCK *lock
)
{
ms_params_t *pp = &msp->params;
minor_t mnum = msp->mnum;
ms_unit_t *un;
mdi_unit_t *ui;
int r, c, i;
struct ms_row *mdr;
ms_comp_t *mdcomp, *mdc;
mddb_recid_t recids[4];
int irecid;
int inc_new_hsp = 0;
int err;
set_t setno = MD_MIN2SET(mnum);
mdclrerror(&msp->mde);
if ((setno >= md_nsets) || (MD_MIN2UNIT(mnum) >= md_nunits))
return (mdmderror(&msp->mde, MDE_INVAL_UNIT, mnum));
if (md_get_setstatus(setno) & MD_SET_STALE)
return (mdmddberror(&msp->mde, MDE_DB_STALE, mnum, setno));
if ((ui = MDI_UNIT(mnum)) == NULL) {
return (mdmderror(&msp->mde, MDE_UNIT_NOT_SETUP, mnum));
}
if (!pp->change_hsp_id)
return (0);
un = (ms_unit_t *)md_ioctl_writerlock(lock, ui);
/* verify that no hot spares are in use */
mdcomp = (struct ms_comp *)((void *)&((char *)un)[un->un_ocomp]);
for (r = 0; r < un->un_nrows; r++) {
mdr = &un->un_row[r];
for (c = 0, i = mdr->un_icomp; c < mdr->un_ncomp; c++) {
mdc = &mdcomp[i++];
if (mdc->un_mirror.ms_hs_id != 0) {
return (mdmderror(&msp->mde, MDE_HS_IN_USE,
mnum));
}
}
}
recids[1] = 0;
recids[2] = 0;
irecid = 1;
if (pp->hsp_id != -1) {
/* increment the reference count of the new hsp */
err = md_hot_spare_ifc(HSP_INCREF, pp->hsp_id, 0, 0,
&recids[1], NULL, NULL, NULL);
if (err) {
return (mdhsperror(&msp->mde, MDE_INVAL_HSP,
pp->hsp_id));
}
inc_new_hsp = 1;
irecid++;
}
if (un->un_hsp_id != -1) {
/* decrement the reference count of the old hsp */
err = md_hot_spare_ifc(HSP_DECREF, un->un_hsp_id, 0, 0,
&recids[irecid], NULL, NULL, NULL);
if (err) {
err = mdhsperror(&msp->mde, MDE_INVAL_HSP,
pp->hsp_id);
if (inc_new_hsp) {
(void) md_hot_spare_ifc(HSP_DECREF,
pp->hsp_id, 0, 0,
&recids[1], NULL, NULL, NULL);
/*
* Don't need to commit the record,
* cause it never got commit before
*/
}
return (err);
}
}
un->un_hsp_id = pp->hsp_id;
recids[0] = un->c.un_record_id;
recids[3] = 0;
mddb_commitrecs_wrapper(recids);
SE_NOTIFY(EC_SVM_STATE, ESC_SVM_CHANGE, SVM_TAG_METADEVICE,
MD_UN2SET(un), MD_SID(un));
return (0);
}
static int
stripe_grow(void *d, int mode, IOLOCK *lockp)
{
minor_t mnum;
ms_unit_t *un, *new_un;
mdi_unit_t *ui;
minor_t *par = NULL;
IOLOCK *plock = NULL;
ms_comp_t *mdcomp, *new_comp;
int row, i, c;
mddb_recid_t ms_recid;
mddb_recid_t old_vtoc = 0;
mddb_recid_t *recids;
md_create_rec_option_t options;
mddb_type_t typ1;
int err;
int64_t tb, atb;
uint_t nr, oc;
int opened;
int rval = 0;
set_t setno;
md_error_t *mdep;
int npar;
int rid;
int num_recs;
u_longlong_t rev;
md_grow_params_t *mgp = d;
mnum = mgp->mnum;
mdep = &mgp->mde;
setno = MD_MIN2SET(mnum);
npar = mgp->npar;
mdclrerror(mdep);
if ((setno >= md_nsets) || (MD_MIN2UNIT(mnum) >= md_nunits))
return (mdmderror(mdep, MDE_INVAL_UNIT, mnum));
if (md_get_setstatus(setno) & MD_SET_STALE)
return (mdmddberror(mdep, MDE_DB_STALE, mnum, setno));
ui = MDI_UNIT(mnum);
if (ui == NULL) {
return (mdmderror(mdep, MDE_UNIT_NOT_SETUP, mnum));
}
if (npar >= 1) {
ASSERT((minor_t *)(uintptr_t)mgp->par != NULL);
par = kmem_alloc(npar * sizeof (*par), KM_SLEEP);
plock = kmem_alloc(npar * sizeof (*plock), KM_SLEEP);
if (ddi_copyin((caddr_t)(uintptr_t)mgp->par, (caddr_t)par,
(npar * sizeof (*par)), mode) != 0) {
kmem_free(par, npar * sizeof (*par));
kmem_free(plock, npar * sizeof (*plock));
return (EFAULT);
}
}
/*
* we grab unit reader/writer first, then parent locks,
* then our own.
* we expect parent units to be sorted to avoid deadlock
*/
rw_enter(&md_unit_array_rw.lock, RW_WRITER);
for (i = 0; i < npar; ++i) {
(void) md_ioctl_writerlock(&plock[i],
MDI_UNIT(par[i]));
}
un = (ms_unit_t *)md_ioctl_writerlock(lockp, ui);
if (un->un_nrows != mgp->nrows) {
rval = EINVAL;
goto out;
}
typ1 = (mddb_type_t)md_getshared_key(setno,
stripe_md_ops.md_driver.md_drivername);
/*
* Preserve the friendly name nature of growing device.
*/
options = MD_CRO_STRIPE;
if (un->c.un_revision & MD_FN_META_DEV)
options |= MD_CRO_FN;
if (mgp->options & MD_CRO_64BIT) {
#if defined(_ILP32)
rval = mdmderror(mdep, MDE_UNIT_TOO_LARGE, mnum);
goto out;
#else
ms_recid = mddb_createrec((size_t)mgp->size, typ1, 0,
MD_CRO_64BIT | options, setno);
#endif
} else {
ms_recid = mddb_createrec((size_t)mgp->size, typ1, 0,
MD_CRO_32BIT | options, setno);
}
if (ms_recid < 0) {
rval = mddbstatus2error(mdep, (int)ms_recid, mnum, setno);
goto out;
}
/* get the address of the new unit */
new_un = (ms_unit_t *)mddb_getrecaddr(ms_recid);
/*
* It is okay that we muck with the new unit here,
* since no one else will know about the unit struct
* until we commit it. If we crash, the record will
* be automatically purged, since we haven't
* committed it yet and the old unit struct will be found.
*/
/* copy in the user's unit struct */
err = ddi_copyin((caddr_t)(uintptr_t)mgp->mdp, (caddr_t)new_un,
(size_t)mgp->size, mode);
if (err) {
mddb_deleterec_wrapper(ms_recid);
rval = EFAULT;
goto out;
}
if (options & MD_CRO_FN)
new_un->c.un_revision |= MD_FN_META_DEV;
/*
* allocate the real recids array. since we may have to
* commit underlying metadevice records, we need an
* array of size: total number of new components being
* attached + 2 (one for the stripe itself, one for the
* end marker).
*/
num_recs = 2;
rid = 0;
for (row = 0; row < new_un->un_nrows; row++) {
struct ms_row *mdr = &new_un->un_row[row];
num_recs += mdr->un_ncomp;
}
recids = kmem_alloc(num_recs * sizeof (mddb_recid_t), KM_SLEEP);
recids[rid++] = ms_recid;
/*
* Save a few of the new unit structs fields.
* Before they get clobbered.
*/
tb = new_un->c.un_total_blocks;
atb = new_un->c.un_actual_tb;
nr = new_un->un_nrows;
oc = new_un->un_ocomp;
rev = new_un->c.un_revision;
/*
* Copy the old unit struct (static stuff)
* into new unit struct
*/
bcopy((caddr_t)un, (caddr_t)new_un,
sizeof (ms_unit_t) + ((nr - 2) * (sizeof (struct ms_row))));
/*
* Restore the saved stuff.
*/
new_un->c.un_total_blocks = tb;
md_nblocks_set(mnum, new_un->c.un_total_blocks);
new_un->c.un_actual_tb = atb;
new_un->un_nrows = nr;
new_un->un_ocomp = oc;
new_un->c.un_revision = rev;
new_un->c.un_record_id = ms_recid;
new_un->c.un_size = mgp->size;
/* All 64 bit metadevices only support EFI labels. */
if (mgp->options & MD_CRO_64BIT) {
new_un->c.un_flag |= MD_EFILABEL;
/*
* If the device was previously smaller than a terabyte,
* and had a vtoc record attached to it, we remove the
* vtoc record, because the layout has changed completely.
*/
if (((un->c.un_revision & MD_64BIT_META_DEV) == 0) &&
(un->c.un_vtoc_id != 0)) {
old_vtoc = un->c.un_vtoc_id;
new_un->c.un_vtoc_id =
md_vtoc_to_efi_record(old_vtoc, setno);
}
}
/*
* Copy the old component structs into the new unit struct.
*/
mdcomp = (ms_comp_t *)((void *)&((char *)un)[un->un_ocomp]);
new_comp = (ms_comp_t *)((void *)&((char *)new_un)[new_un->un_ocomp]);
for (row = 0; row < un->un_nrows; row++) {
struct ms_row *mdr = &un->un_row[row];
for (i = 0, c = mdr->un_icomp; i < mdr->un_ncomp; i++, c++) {
bcopy((caddr_t)&mdcomp[c], (caddr_t)&new_comp[c],
sizeof (ms_comp_t));
}
}
opened = md_unit_isopen(ui);
/*
* Set parent on metadevices being added.
* Open the new devices being added.
* NOTE: currently soft partitions are the only metadevices
* which can appear within a stripe.
*/
for (row = un->un_nrows; row < new_un->un_nrows; row++) {
struct ms_row *mdr = &new_un->un_row[row];
for (i = 0, c = mdr->un_icomp; i < mdr->un_ncomp; i++) {
struct ms_comp *mdc = &new_comp[c++];
md_dev64_t comp_dev;
md_unit_t *comp_un;
comp_dev = mdc->un_dev;
/* set parent on any metadevices */
if (md_getmajor(comp_dev) == md_major) {
comp_un = MD_UNIT(md_getminor(comp_dev));
recids[rid++] = MD_RECID(comp_un);
md_set_parent(comp_dev, MD_SID(new_un));
}
if (opened) {
md_dev64_t tmpdev = mdc->un_dev;
/*
* Open by device id
* Check if this comp is hotspared and
* if it is then use the key for hotspare
*/
tmpdev = md_resolve_bydevid(mnum, tmpdev,
mdc->un_mirror.ms_hs_id ?
mdc->un_mirror.ms_hs_key : mdc->un_key);
(void) md_layered_open(mnum, &tmpdev,
MD_OFLG_NULL);
mdc->un_dev = tmpdev;
mdc->un_mirror.ms_flags |= MDM_S_ISOPEN;
}
}
}
/* set end marker */
recids[rid] = 0;
/* commit new unit struct */
mddb_commitrecs_wrapper(recids);
/* delete old unit struct */
mddb_deleterec_wrapper(un->c.un_record_id);
/* place new unit in in-core array */
md_nblocks_set(mnum, new_un->c.un_total_blocks);
MD_UNIT(mnum) = new_un;
/*
* If old_vtoc has a non zero value, we know:
* - This unit crossed the border from smaller to larger one TB
* - There was a vtoc record for the unit,
* - This vtoc record is no longer needed, because
* a new efi record has been created for this un.
*/
if (old_vtoc != 0) {
mddb_deleterec_wrapper(old_vtoc);
}
/* free recids array */
kmem_free(recids, num_recs * sizeof (mddb_recid_t));
SE_NOTIFY(EC_SVM_CONFIG, ESC_SVM_GROW, SVM_TAG_METADEVICE,
MD_UN2SET(new_un), MD_SID(new_un));
/* release locks, return success */
out:
for (i = npar - 1; (i >= 0); --i)
md_ioctl_writerexit(&plock[i]);
rw_exit(&md_unit_array_rw.lock);
if (plock != NULL)
kmem_free(plock, npar * sizeof (*plock));
if (par != NULL)
kmem_free(par, npar * sizeof (*par));
return (rval);
}
static int
stripe_set(void *d, int mode)
{
minor_t mnum;
ms_unit_t *un;
void *p;
mddb_recid_t ms_recid;
mddb_recid_t *recids;
mddb_type_t typ1;
int err;
set_t setno;
md_error_t *mdep;
struct ms_comp *mdcomp;
int row;
int rid;
int num_recs;
int i, c;
md_set_params_t *msp = d;
mnum = msp->mnum;
setno = MD_MIN2SET(mnum);
mdep = &msp->mde;
mdclrerror(mdep);
if ((setno >= md_nsets) || (MD_MIN2UNIT(mnum) >= md_nunits)) {
return (mdmderror(mdep, MDE_INVAL_UNIT, mnum));
}
if (md_get_setstatus(setno) & MD_SET_STALE)
return (mdmddberror(mdep, MDE_DB_STALE, mnum, setno));
un = MD_UNIT(mnum);
if (un != NULL) {
return (mdmderror(mdep, MDE_UNIT_ALREADY_SETUP, mnum));
}
typ1 = (mddb_type_t)md_getshared_key(setno,
stripe_md_ops.md_driver.md_drivername);
/* create the db record for this mdstruct */
if (msp->options & MD_CRO_64BIT) {
#if defined(_ILP32)
return (mdmderror(mdep, MDE_UNIT_TOO_LARGE, mnum));
#else
ms_recid = mddb_createrec((size_t)msp->size, typ1, 0,
MD_CRO_64BIT | MD_CRO_STRIPE | MD_CRO_FN, setno);
#endif
} else {
ms_recid = mddb_createrec((size_t)msp->size, typ1, 0,
MD_CRO_32BIT | MD_CRO_STRIPE | MD_CRO_FN, setno);
}
if (ms_recid < 0)
return (mddbstatus2error(mdep, ms_recid, mnum, setno));
/* get the address of the mdstruct */
p = (void *) mddb_getrecaddr(ms_recid);
/*
* It is okay that we muck with the mdstruct here,
* since no one else will know about the mdstruct
* until we commit it. If we crash, the record will
* be automatically purged, since we haven't
* committed it yet.
*/
/* copy in the user's mdstruct */
if (err = ddi_copyin((caddr_t)(uintptr_t)msp->mdp, (caddr_t)p,
(size_t)msp->size, mode)) {
mddb_deleterec_wrapper(ms_recid);
return (EFAULT);
}
un = (ms_unit_t *)p;
/* All 64 bit metadevices only support EFI labels. */
if (msp->options & MD_CRO_64BIT) {
un->c.un_flag |= MD_EFILABEL;
}
/*
* allocate the real recids array. since we may have to commit
* underlying metadevice records, we need an array
* of size: total number of components in stripe + 3
* (1 for the stripe itself, one for the hotspare, one
* for the end marker).
*/
num_recs = 3;
rid = 0;
for (row = 0; row < un->un_nrows; row++) {
struct ms_row *mdr = &un->un_row[row];
num_recs += mdr->un_ncomp;
}
recids = kmem_alloc(num_recs * sizeof (mddb_recid_t), KM_SLEEP);
recids[rid++] = ms_recid;
MD_SID(un) = mnum;
MD_RECID(un) = recids[0];
MD_CAPAB(un) = MD_CAN_PARENT | MD_CAN_SUB_MIRROR | MD_CAN_SP;
MD_PARENT(un) = MD_NO_PARENT;
un->c.un_revision |= MD_FN_META_DEV;
if (err = stripe_build_incore(p, 0)) {
md_nblocks_set(mnum, -1ULL);
MD_UNIT(mnum) = NULL;
mddb_deleterec_wrapper(recids[0]);
kmem_free(recids, num_recs * sizeof (mddb_recid_t));
return (err);
}
/*
* Update unit availability
*/
md_set[setno].s_un_avail--;
recids[rid] = 0;
if (un->un_hsp_id != -1)
err = md_hot_spare_ifc(HSP_INCREF, un->un_hsp_id, 0, 0,
&recids[rid++], NULL, NULL, NULL);
if (err) {
md_nblocks_set(mnum, -1ULL);
MD_UNIT(mnum) = NULL;
mddb_deleterec_wrapper(recids[0]);
kmem_free(recids, num_recs * sizeof (mddb_recid_t));
return (mdhsperror(mdep, MDE_INVAL_HSP, un->un_hsp_id));
}
/*
* set the parent on any metadevice components.
* NOTE: currently soft partitions are the only metadevices
* which can appear within a stripe.
*/
mdcomp = (ms_comp_t *)((void *)&((char *)un)[un->un_ocomp]);
for (row = 0; row < un->un_nrows; row++) {
struct ms_row *mdr = &un->un_row[row];
for (i = 0, c = mdr->un_icomp; i < mdr->un_ncomp; i++) {
ms_comp_t *mdc = &mdcomp[c++];
md_dev64_t comp_dev;
md_unit_t *comp_un;
comp_dev = mdc->un_dev;
if (md_getmajor(comp_dev) == md_major) {
/* set parent and disallow soft partitioning */
comp_un = MD_UNIT(md_getminor(comp_dev));
recids[rid++] = MD_RECID(comp_un);
md_set_parent(mdc->un_dev, MD_SID(un));
}
}
}
/* set end marker */
recids[rid] = 0;
mddb_commitrecs_wrapper(recids);
md_create_unit_incore(mnum, &stripe_md_ops, 0);
kmem_free(recids, (num_recs * sizeof (mddb_recid_t)));
SE_NOTIFY(EC_SVM_CONFIG, ESC_SVM_CREATE, SVM_TAG_METADEVICE,
MD_UN2SET(un), MD_SID(un));
return (0);
}
/*
* 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);
}
