static int
get_hs(
get_hs_params_t *ghs
)
{
hot_spare_t *hs;
set_t setno = ghs->md_driver.md_setno;
mdclrerror(&ghs->mde);
/* Scan the hot spare list for the hs */
hs = (hot_spare_t *)md_set[setno].s_hs;
while (hs) {
if (hs->hs_key == ghs->ghs_key) {
break;
}
hs = hs->hs_next;
}
if (hs == NULL) {
return (mddeverror(&ghs->mde, MDE_INVAL_HS,
ghs->ghs_devnum));
}
ghs->ghs_start_blk = hs->hs_start_blk;
ghs->ghs_number_blks = hs->hs_number_blks;
ghs->ghs_state = hs->hs_state;
ghs->ghs_timestamp = hs->hs_timestamp;
ghs->ghs_revision = hs->hs_revision;
return (0);
}
/*
* check to see if a device is in a metadevice
*/
int
meta_check_inmeta(
mdsetname_t *sp,
mdname_t *np,
mdchkopts_t options,
diskaddr_t slblk,
diskaddr_t nblks,
md_error_t *ep
)
{
uint_t partno;
/* see if replica slice is ok, only applies to disks in sets */
if (! (options & MDCHK_ALLOW_REPSLICE) &&
! metaislocalset(sp)) {
uint_t rep_slice;
if (metagetvtoc(np, FALSE, &partno, ep) == NULL)
return (-1);
if (meta_replicaslice(np->drivenamep, &rep_slice, ep)
!= 0)
return (-1);
if (partno == rep_slice)
return (mddeverror(ep, MDE_REPCOMP_INVAL, np->dev,
np->cname));
}
/* check for databases */
if (meta_check_inreplica(sp, np, slblk, nblks, ep) != 0) {
if (mdisuseerror(ep, MDE_ALREADY)) {
if (options & MDCHK_ALLOW_MDDB) {
mdclrerror(ep);
} else {
return (mddeverror(ep, MDE_HAS_MDDB,
np->dev, np->cname));
}
} else {
return (-1);
}
}
/* check metadevices */
if (meta_check_instripe(sp, np, slblk, nblks, ep) != 0)
return (-1);
if (meta_check_inmirror(sp, np, slblk, nblks, ep) != 0)
return (-1);
if (meta_check_intrans(sp, np, options, slblk, nblks, ep) != 0)
return (-1);
if (meta_check_insp(sp, np, slblk, nblks, ep) != 0)
return (-1);
if (! (options & MDCHK_ALLOW_HS)) {
if (meta_check_inhsp(sp, np, slblk, nblks, ep) != 0)
return (-1);
}
if (meta_check_inraid(sp, np, slblk, nblks, ep) != 0)
return (-1);
/* return success */
return (0);
}
/*
* check to see if a device is in its set
*/
int
meta_check_inset(
mdsetname_t *sp,
mdname_t *np,
md_error_t *ep
)
{
mdsetname_t *npsp;
int bypass_daemon = FALSE;
/* check devices set */
if (metaislocalset(sp))
bypass_daemon = TRUE;
if ((npsp = metagetset(np, bypass_daemon, ep)) == NULL) {
if ((! metaismeta(np)) &&
(metaislocalset(sp)) &&
(mdismddberror(ep, MDE_DB_NODB))) {
mdclrerror(ep);
npsp = sp;
} else {
return (-1);
}
}
/* check set */
if (metaissameset(sp, npsp))
return (0);
/* return appropriate error */
if (metaislocalset(sp))
return (mddeverror(ep, MDE_IN_SHARED_SET, np->dev, np->cname));
else
return (mddeverror(ep, MDE_NOT_IN_SET, np->dev, np->cname));
}
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);
}
/*
* Is a file system currently mounted on this disk drive?
*/
int
meta_check_drivemounted(
mdsetname_t *sp,
mddrivename_t *dnp,
md_error_t *ep
)
{
FILE *mfp;
struct mnttab m;
int rval = 0;
char mountp[MNT_LINE_MAX];
char mnt_special[MNT_LINE_MAX];
/* should have a set */
assert(sp != NULL);
/* look in mnttab */
if ((mfp = open_mnttab()) == NULL)
return (mdsyserror(ep, errno, MNTTAB));
while ((getmntent(mfp, &m) == 0) && (rval == 0)) {
char **fstype = skip_these_mntents;
int skipit = 0;
mdname_t *mnp;
if ((m.mnt_special == NULL) || (m.mnt_mountp == NULL))
continue;
if (m.mnt_mountp[0] != '/')
continue;
while (*fstype != NULL)
if (strcmp(m.mnt_fstype, *fstype++) == 0) {
skipit++;
break;
}
if (skipit == 1)
continue;
(void) strcpy(mountp, m.mnt_mountp);
(void) strcpy(mnt_special, m.mnt_special);
if ((mnp = metaname(&sp, mnt_special,
LOGICAL_DEVICE, ep)) == NULL) {
mdclrerror(ep);
continue;
}
if (strcmp(dnp->cname, mnp->drivenamep->cname) == 0) {
rval = mduseerror(ep, MDE_IS_MOUNTED, NODEV64,
mountp, dnp->cname);
}
}
/* return success */
return (rval);
}
int
meta_check_driveinset(mdsetname_t *sp, mddrivename_t *dn, md_error_t *ep)
{
set_t setno;
set_t max_sets;
if ((max_sets = get_max_sets(ep)) == 0)
return (-1);
for (setno = 1; setno < max_sets; setno++) {
mdsetname_t *sp1;
int is_it;
if (setno == sp->setno)
continue;
if ((sp1 = metasetnosetname(setno, ep)) == NULL) {
if (mdismddberror(ep, MDE_DB_NODB)) {
mdclrerror(ep);
return (0);
}
if (mdiserror(ep, MDE_NO_SET)) {
mdclrerror(ep);
continue;
}
return (-1);
}
metaflushsetname(sp1);
if ((is_it = meta_is_drive_in_thisset(sp1, dn, FALSE, ep))
== -1)
return (-1);
if (is_it)
return (mddserror(ep, MDE_DS_DRIVEINSET, sp->setno,
sp1->setname, dn->cname, sp->setname));
}
return (0);
}
void
create_diskset_links()
{
int max_sets;
int i;
md_error_t error = mdnullerror;
/*
* Resolve the function pointers for libsds_sc so that we can
* snarf the set records.
*/
(void) sdssc_bind_library();
(void) init_metalib();
if ((max_sets = get_max_sets(&error)) == 0) {
debug_printf("create_diskset_links(): get_max_sets failed\n");
mdclrerror(&error);
return;
}
for (i = 1; i < max_sets; i++) {
md_set_record *sr;
char setname[MAXPATHLEN];
char setnum[MAXPATHLEN];
if ((sr = metad_getsetbynum(i, &error)) == NULL) {
mdclrerror(&error);
continue;
}
(void) snprintf(setname, MAXPATHLEN, "/dev/md/%s",
sr->sr_setname);
(void) snprintf(setnum, MAXPATHLEN, "shared/%d", i);
/*
* Ignore failures to create the symlink. This could
* happen because suninstall is restartable so the
* symlink might have already been created.
*/
(void) symlink(setnum, setname);
}
}
/*
* free disk status list
*/
void
meta_free_disk_status_list(
md_disk_status_list_t *dslp
)
{
md_disk_status_list_t *next = NULL;
for (/* void */; (dslp != NULL); dslp = next) {
next = dslp->next;
mdclrerror(&dslp->status);
Free(dslp);
}
}
/*
* NAME: meta_raid_resync_all
* DESCRIPTION: loop through the RAID devices synch'ing all
* PARAMETERS: char *sp - the set to synch
* daddr_t size - resync size
* md_error_t *ep - return error info
*
*/
int
meta_raid_resync_all(
mdsetname_t *sp,
daddr_t size,
md_error_t *ep
)
{
mdnamelist_t *nlp = NULL;
mdnamelist_t *p;
int rval = 0, fval;
/* should have a set */
assert(sp != NULL);
/* get raids */
if (meta_get_raid_names(sp, &nlp, 0, ep) < 0)
return (-1);
/* fork a process */
if ((fval = md_daemonize(sp, ep)) != 0) {
/*
* md_daemonize forks off a process to do the work. This
* is the parent or errror.
*/
if (fval > 0) {
if (nlp != NULL)
metafreenamelist(nlp);
return (0);
}
mdclrerror(ep);
}
assert((fval == 0) || (fval == -1));
/* resync each raid */
for (p = nlp; (p != NULL); p = p->next) {
mdname_t *raidnp = p->namep;
if (meta_raid_resync(sp, raidnp, size, ep) != 0)
rval = -1;
}
/* cleanup, return success */
if (nlp != NULL)
metafreenamelist(nlp);
if (fval == 0)
exit(0);
return (rval);
}
static int
get_hsp(
void *d,
int mode
)
{
hot_spare_pool_t *hsp;
get_hsp_t *ghsp;
size_t size;
set_t setno;
int err = 0;
md_i_get_t *migp = (md_i_get_t *)d;
setno = migp->md_driver.md_setno;
mdclrerror(&migp->mde);
/* Scan the hot spare pool list */
hsp = find_hot_spare_pool(setno, migp->id);
if (hsp == NULL) {
return (mdhsperror(&migp->mde, MDE_INVAL_HSP,
migp->id));
}
size = (sizeof (ghsp->ghsp_hs_keys[0]) * (hsp->hsp_nhotspares - 1)) +
sizeof (get_hsp_t);
if (migp->size == 0) {
migp->size = (int)size;
return (0);
}
if (migp->size < size)
return (EFAULT);
ghsp = kmem_alloc(size, KM_SLEEP);
ghsp->ghsp_id = hsp->hsp_self_id;
ghsp->ghsp_refcount = hsp->hsp_refcount;
ghsp->ghsp_nhotspares = hsp->hsp_nhotspares;
build_key_list(setno, hsp, ghsp->ghsp_hs_keys);
if (ddi_copyout(ghsp, (caddr_t)(uintptr_t)migp->mdp, size, mode))
err = EFAULT;
kmem_free(ghsp, size);
return (err);
}
/*
* check whether device is swapped on
*/
static int
meta_check_swapped(
mdsetname_t *sp,
mdname_t *np,
md_error_t *ep
)
{
struct swaptable *swtp;
int nswap;
int i;
int rval = 0;
/* should have a set */
assert(sp != NULL);
/* get swap info */
if (get_swapinfo(&swtp, &nswap, ep) != 0)
return (-1);
/* look for match */
for (i = 0; ((i < nswap) && (rval == 0)); ++i) {
mdname_t *snp;
if ((snp = metaname(&sp, swtp->swt_ent[i].ste_path,
UNKNOWN, ep)) == NULL) {
mdclrerror(ep);
continue;
}
if (np->dev == snp->dev) {
rval = mddeverror(ep, MDE_IS_SWAPPED,
np->dev, np->cname);
} else { /* not swap - does it overlap */
rval = meta_check_overlap(snp->cname, np, 0, -1,
snp, 0, -1, ep);
if (rval != 0) {
(void) mdoverlaperror(ep, MDE_OVERLAP_SWAP,
np->cname, NULL, snp->cname);
}
}
}
free_swapinfo(swtp);
/* return success */
return (rval);
}
/*
* check whether device is a dump device
*/
static int
meta_check_dump(
mdsetname_t *sp,
mdname_t *np,
md_error_t *ep
)
{
int rval = 0;
int dump_fd;
char device[MAXPATHLEN];
if ((dump_fd = open("/dev/dump", O_RDONLY)) < 0)
return (mdsyserror(ep, errno, "/dev/dump"));
if (ioctl(dump_fd, DIOCGETDEV, device) != -1) {
mdname_t *dump_np;
if ((dump_np = metaname(&sp, device, UNKNOWN, ep)) == NULL) {
mdclrerror(ep);
(void) close(dump_fd);
return (0);
}
if (np->dev == dump_np->dev) {
rval = mddeverror(ep, MDE_IS_DUMP,
np->dev, np->cname);
} else { /* not a dump device - but does it overlap? */
rval = meta_check_overlap(dump_np->cname, np, 0, -1,
dump_np, 0, -1, ep);
if (rval != 0) {
(void) mdoverlaperror(ep, MDE_OVERLAP_DUMP,
np->cname, NULL, dump_np->cname);
}
}
}
(void) close(dump_fd);
return (rval);
}
/*
* Is a driver currently swapped on?
*/
int
meta_check_driveswapped(
mdsetname_t *sp,
mddrivename_t *dnp,
md_error_t *ep
)
{
struct swaptable *swtp;
int nswap;
int i;
int rval = 0;
/* should have a set */
assert(sp != NULL);
/* get swap info */
if (get_swapinfo(&swtp, &nswap, ep) != 0)
return (-1);
/* look for match */
for (i = 0; (i < nswap); ++i) {
mdname_t *snp;
if ((snp = metaname(&sp, swtp->swt_ent[i].ste_path,
LOGICAL_DEVICE, ep)) == NULL) {
mdclrerror(ep);
continue;
}
if (strcmp(dnp->cname, snp->drivenamep->cname) == 0) {
rval = mddeverror(ep, MDE_IS_SWAPPED, NODEV64,
dnp->cname);
}
}
free_swapinfo(swtp);
/* return success */
return (rval);
}
/*ARGSUSED*/
static int
stripe_get(void *d, int mode, IOLOCK *lock)
{
minor_t mnum;
mdi_unit_t *ui;
ms_unit_t *un;
md_error_t *mdep;
md_i_get_t *migp = d;
mnum = migp->id;
mdep = &migp->mde;
mdclrerror(mdep);
if ((MD_MIN2SET(mnum) >= md_nsets) || (MD_MIN2UNIT(mnum) >= md_nunits))
return (mdmderror(mdep, MDE_INVAL_UNIT, mnum));
if ((ui = MDI_UNIT(mnum)) == NULL) {
return (mdmderror(mdep, MDE_UNIT_NOT_SETUP, mnum));
}
un = (ms_unit_t *)md_ioctl_readerlock(lock, ui);
if (migp->size == 0) {
migp->size = un->c.un_size;
return (0);
}
if (migp->size < un->c.un_size) {
return (EFAULT);
}
if (ddi_copyout(un, (void *)(uintptr_t)migp->mdp,
un->c.un_size, mode))
return (EFAULT);
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_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);
}
int
get_mdcomponents(char *uname, svm_info_t **svmpp)
{
svm_info_t *svmp;
md_error_t status, *ep;
mdname_t *namep;
mdnamelist_t *nlp = NULL;
mdnamelist_t *p;
mdsetname_t *sp = NULL;
char *strp = NULL;
int rval, cnt;
rval = RET_SUCCESS;
cnt = 0;
status = mdnullerror;
ep = &status;
svmp = *svmpp;
(void) init_metalib();
debug_printf("get_mdcomponents(): Enter unit name %s\n", uname);
if (((namep = metaname(&sp, uname, META_DEVICE, ep)) == NULL) ||
(metachkmeta(namep, ep) != 0)) {
debug_printf("get_mdcomponents(): "
"metaname or metachkmeta failed\n");
mdclrerror(ep);
return (RET_ERROR);
}
debug_printf("get_mdcomponents(): meta_getdevs %s\n", namep->cname);
if ((meta_getdevs(sp, namep, &nlp, ep)) < 0) {
debug_printf("get_mdcomponents(): "
"comp %s - meta_getdevs failed\n", uname);
metafreenamelist(nlp);
mdclrerror(ep);
return (RET_ERROR);
}
/* compute the number of devices */
for (p = nlp, cnt = 0; p != NULL; p = p->next, cnt++)
;
/*
* Need to add n -1 components since slvmp already has space
* for one device.
*/
svmp = (svm_info_t *)realloc(svmp, sizeof (svm_info_t) +
(sizeof (char *) * (cnt - 1)));
if (svmp == NULL) {
debug_printf("get_mdcomponents(): realloc of svmp failed\n");
metafreenamelist(nlp);
return (RET_ERROR);
}
for (p = nlp, cnt = 0; p != NULL; p = p->next, cnt++) {
mdname_t *devnp = p->namep;
if ((strp = strdup(devnp->cname)) == NULL) {
rval = RET_ERROR;
break;
}
svmp->md_comps[cnt] = strp;
}
/* count is set to the number of devices in the list */
svmp->count = cnt;
svmp->root_md = strdup(uname);
if (rval == RET_SUCCESS && svmp->root_md != NULL) {
debug_printf("get_mdcomponents(): root_md %s count %d \n",
svmp->root_md, svmp->count);
for (cnt = 0; cnt < svmp->count; cnt++)
debug_printf("get_mdcomponents(): %s\n",
svmp->md_comps[cnt]);
} else {
rval = RET_ERROR;
svm_free(svmp);
svmp = NULL;
debug_printf("get_mdcomponents(): malloc failed\n");
}
metafreenamelist(nlp);
*svmpp = svmp;
return (rval);
}
/*
* check for same drive
*
* Differentiate between matching on name/dev_t and devid. In the latter
* case it is correct to fail but misleading to give the same error msg as
* for an overlapping slice.
*
*/
int
meta_check_samedrive(
mdname_t *np1, /* first comp */
mdname_t *np2, /* second comp */
md_error_t *ep
)
{
mdcinfo_t *cinfop1, *cinfop2;
mdnmtype_t type1 = np1->drivenamep->type;
mdnmtype_t type2 = np2->drivenamep->type;
int l = 0;
char *name1 = NULL;
char *name2 = NULL;
int retval = CANT_TELL;
int fd1 = -1;
int fd2 = -1;
int rc1 = -2, rc2 = -2;
uint_t strl1 = 0, strl2 = 0;
int devid1_found = 0;
int devid2_found = 0;
ddi_devid_t devid1 = NULL;
ddi_devid_t devid2 = NULL;
dev_list_t *dnlp = NULL;
assert(type1 != MDT_FAST_META && type1 != MDT_FAST_COMP);
assert(type2 != MDT_FAST_META && type2 != MDT_FAST_COMP);
/*
* The process of determining if 2 names are the same drive is
* as follows:
*
* Case 1 - The filenames are identical
*
* Case 2 - Both devices have a devid
* get and compare the devids for the devices. If both
* devices have a devid then the compare will is all
* that is needed we are done.
*
* Case 3 - One or more devices does not have a devid
* start by doing a simple compare of the name, if they
* are the same just return.
*
* If the names differ then keep going and see if the
* may be the same underlying devic. First check to
* see if the sd name is the same (old code).
*
* Then check the major and minor numbers to see if
* they are the same. If they are then return (old code).
*
* Next compare the raw name and the component name and
* if they are the same then return.
*
* All else has failed so use the component name (cname)
* component number and unit number. If they all are
* equal then call them the same drive.
*
*/
if ((np1 == NULL) || (np2 == NULL))
return (NOT_SAMEDRIVE);
/* if the name structs are the same then the drives must be */
if (np1 == np2)
return (IDENTICAL_NAME_DEVT);
name1 = np1->bname;
name2 = np2->bname;
if ((name1 == NULL) || ((strl1 = strlen(name1)) == 0) ||
(name2 == NULL) || ((strl2 = strlen(name2)) == 0))
return (NOT_SAMEDRIVE);
if ((strl1 == strl2) && (strcmp(name1, name2) == 0)) {
/* names are identical */
return (IDENTICAL_NAME_DEVT);
}
if (is_metaname(name1) || is_metaname(name2))
return (NOT_SAMEDRIVE);
/*
* Check to see if the devicename is in the static list. If so,
* use its devid. Otherwise do the expensive operations
* of opening the device, getting the devid, and closing the
* device. Add the result into the static list.
*
* The case where this list will be useful is when there are soft
* partitions on multiple drives and a new soft partition is being
* created. In that situation the underlying physical device name
* for the new soft partition would be compared against each of the
* existing soft partititions. Without this static list that would
* involve 2 opens, closes, and devid gets for each existing soft
* partition
*/
for (dnlp = devnamelist; (dnlp != NULL) &&
!(devid1_found && devid2_found); dnlp = dnlp->dev_nxt) {
if (!devid1_found && (strcmp(dnlp->dev_name, name1) == 0)) {
devid1_found = 1;
devid1 = dnlp->devid;
if (devid1 == NULL)
rc1 = 1;
else
rc1 = 0;
continue;
}
if (!devid2_found && (strcmp(dnlp->dev_name, name2) == 0)) {
devid2_found = 1;
devid2 = dnlp->devid;
if (devid2 == NULL)
rc2 = 1;
else
rc2 = 0;
continue;
}
}
/*
* Start by checking if the device has a device id, and if they
* are equal. If they are there is no question there is a match.
*
* The process here is open each disk, get the devid for each
* disk. If they both have a devid compare them and return
* the results.
*/
if (!devid1_found) {
if ((fd1 = open(name1, O_RDONLY | O_NDELAY)) < 0) {
return (NOT_SAMEDRIVE);
}
rc1 = devid_get(fd1, &devid1);
(void) close(fd1);
/* add the name and devid to the cache */
add_to_devname_list(name1, devid1);
}
if (!devid2_found) {
if ((fd2 = open(name2, O_RDONLY | O_NDELAY)) < 0) {
return (NOT_SAMEDRIVE);
}
rc2 = devid_get(fd2, &devid2);
(void) close(fd2);
/* add the name and devid to the cache */
add_to_devname_list(name2, devid2);
}
if ((rc1 == 0) && (rc2 == 0)) {
if (devid_compare(devid1, devid2) == 0)
retval = IDENTICAL_DEVIDS; /* same devid */
else
retval = NOT_SAMEDRIVE; /* different drives */
}
if (retval >= 0) {
return (retval);
}
/*
* At this point in time one of the two drives did not have a
* device ID. Do not make the assumption that is one drive
* did have a device id and the other did not that they are not
* the same. One drive could be covered by a device and still
* be the same drive. This is a general flaw in the system at
* this time.
*/
/*
* The optimization can not happen if we are given an old style name
* in the form /dev/XXNN[a-h], since the name caches differently and
* allows overlaps to happen.
*/
if (! ((sscanf(np1->bname, "/dev/%*[^0-9/]%*u%*[a-h]%n", &l) == 0 &&
l == strlen(np1->bname)) ||
(sscanf(np2->bname, "/dev/%*[^0-9/]%*u%*[a-h]%n", &l) == 0 &&
l == strlen(np2->bname))) &&
((type1 == MDT_COMP) || (type1 == MDT_META)) &&
((type2 == MDT_COMP) || (type2 == MDT_META)))
if (np1->drivenamep == np2->drivenamep)
return (IDENTICAL_NAME_DEVT);
else
return (NOT_SAMEDRIVE);
/* check for same drive */
if (meta_getmajor(np1->dev) != meta_getmajor(np2->dev))
return (NOT_SAMEDRIVE); /* not same drive */
if (((cinfop1 = metagetcinfo(np1, ep)) == NULL) ||
((cinfop2 = metagetcinfo(np2, ep)) == NULL)) {
if ((strcmp(np1->drivenamep->cname,
np2->drivenamep->cname) != 0) &&
(strcmp(np1->drivenamep->rname,
np2->drivenamep->rname) != 0)) {
mdclrerror(ep);
return (NOT_SAMEDRIVE); /* not same drive */
} else {
return (CANT_TELL); /* can't tell */
}
} else if ((strncmp(cinfop1->cname, cinfop2->cname,
sizeof (cinfop1->cname)) != 0) ||
(cinfop1->cnum != cinfop2->cnum) ||
(cinfop1->unit != cinfop2->unit)) {
return (NOT_SAMEDRIVE); /* not same drive */
}
/* same drive */
return (IDENTICAL_NAME_DEVT);
}
/*
* check whether device is mounted
*/
static int
meta_check_mounted(
mdsetname_t *sp,
mdname_t *np,
md_error_t *ep
)
{
FILE *mfp;
struct mnttab m;
int rval = 0;
char mountp[MNT_LINE_MAX];
char mnt_special[MNT_LINE_MAX];
/* should have a set */
assert(sp != NULL);
/* look in mnttab */
if ((mfp = open_mnttab()) == NULL)
return (mdsyserror(ep, errno, MNTTAB));
while ((getmntent(mfp, &m) == 0) && (rval == 0)) {
char **fstype = skip_these_mntents;
int skipit = 0;
mdname_t *mnp;
if ((m.mnt_special == NULL) || (m.mnt_mountp == NULL))
continue;
if (m.mnt_mountp[0] != '/')
continue;
while (*fstype != NULL)
if (strcmp(m.mnt_fstype, *fstype++) == 0) {
skipit++;
break;
}
if (skipit == 1)
continue;
(void) strcpy(mountp, m.mnt_mountp);
(void) strcpy(mnt_special, m.mnt_special);
if ((mnp = metaname(&sp, mnt_special, UNKNOWN, ep)) == NULL) {
mdclrerror(ep);
continue;
}
if (np->dev == mnp->dev) {
rval = mduseerror(ep, MDE_IS_MOUNTED,
np->dev, mountp, np->cname);
} else { /* device isn't in mnttab - does it overlap? */
rval = meta_check_overlap(mnp->cname, np, 0, -1,
mnp, 0, -1, ep);
if (rval != 0) {
(void) mdoverlaperror(ep, MDE_OVERLAP_MOUNTED,
np->cname, mountp, mnp->cname);
}
}
}
/* return success */
return (rval);
}
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);
}
/*
* setup RPC service
*
* if can't authenticate return < 0
* any other error return > 0
*/
int
svc_init(
struct svc_req *rqstp, /* RPC stuff */
int amode, /* R_OK | W_OK */
md_error_t *ep /* returned status */
)
{
SVCXPRT *transp;
if (sdssc_bind_library() == SDSSC_ERROR) {
mdsyserror(ep, EACCES, "can't bind to cluster library");
return (1);
}
/*
* if we have no rpc service info, we must have been
* called recursively from within the daemon
*/
if (rqstp == NULL) {
mdclrerror(ep);
return (0); /* OK */
}
/*
* initialize
*/
transp = rqstp->rq_xprt;
assert(transp != NULL);
*ep = mdnullerror;
/*
* check credentials
*/
switch (rqstp->rq_cred.oa_flavor) {
/* UNIX flavor */
case AUTH_SYS:
{
if (check_sys(rqstp, amode, ep) != 0)
return (1); /* error */
break;
}
/* can't authenticate anything else */
default:
svcerr_weakauth(transp);
return (-1); /* weak authentication */
}
/*
* (re)initialize
*/
if (md_init_daemon("rpc.metad", ep) != 0)
return (1); /* error */
if (set_snarf(ep))
return (1);
sr_validate();
/* success */
return (0);
}
static int
init_entries(
mdsetname_t **spp,
md_tab_t *tabp,
mdcmdopts_t options,
uint_t flags,
bool_t called_thru_rpc,
md_error_t *ep
)
{
uint_t cnt = 0;
uint_t line;
int rval = 0;
int ret;
/* for all matching entries, which haven't already been done */
for (line = 0; (line < tabp->nlines); ++line) {
md_tab_line_t *linep = &tabp->lines[line];
char *uname = linep->argv[0];
/* see if already done */
if (linep->flags != DO_AGAIN)
continue;
/* clear the metadev/hsp caches between inits */
metaflushmetanames();
/* try it */
if ((called_thru_rpc == FALSE) &&
meta_is_mn_name(spp, uname, ep)) {
/*
* MN set, send command to all nodes
* Note that is sp is NULL, meta_is_mn_name() derives
* sp from linep->argv which is the metadevice arg
*/
ret = mn_send_command(spp, linep->argc, linep->argv,
options, flags, linep->context, ep);
} else {
char *cname = NULL;
cname = meta_name_getname(spp, uname, META_DEVICE, ep);
if (cname == NULL) {
mde_perror(ep, "");
mdclrerror(ep);
} else {
ret = meta_init_name(spp, linep->argc,
linep->argv, cname, options, ep);
Free(cname);
if (ret != 0) {
if (!(flags & MD_IGNORE_STDERR)) {
mderrorextra(ep, linep->context);
mde_perror(ep, "");
rval = -1;
}
mdclrerror(ep);
}
}
}
if (ret == 0) {
linep->flags = IS_DONE;
++cnt;
}
}
/* return success */
if (rval != 0)
return (rval);
return (cnt);
}
char *
obj2devname(uint32_t tag, set_t setno, md_dev64_t dev)
{
char *setname;
char *uname;
char name[MD_MAX_CTDLEN];
mdsetname_t *sp;
md_error_t status = mdnullerror;
md_set_record *md_sr;
minor_t mnum = meta_getminor(dev);
int rtn = 0;
setname = NULL;
if ((setno != MD_SET_BAD) &&
((sp = metasetnosetname(setno, &status)) != NULL)) {
setname = sp->setname;
}
name[0] = '\0';
switch (tag) {
case SVM_TAG_HS:
case SVM_TAG_METADEVICE:
case SVM_TAG_MIRROR:
case SVM_TAG_RAID5:
case SVM_TAG_STRIPE:
case SVM_TAG_TRANS:
uname = get_mdname(sp, mnum);
if (uname == NULL)
return (NULL);
(void) strcpy(name, uname);
break;
case SVM_TAG_HSP:
uname = get_hspname(sp, mnum);
if (uname == NULL)
return (NULL);
(void) strcpy(name, uname);
break;
case SVM_TAG_DRIVE:
(void) sprintf(name, "drive");
break;
case SVM_TAG_HOST:
md_sr = NULL;
if (setname != NULL) {
md_sr = getsetbyname(setname, &status);
}
if ((md_sr != NULL) && (md_sr->sr_nodes[mnum] != NULL)) {
/*
* Get the host data from the node array.
*/
rtn = snprintf(name, sizeof (name), "%s",
md_sr->sr_nodes[mnum]);
}
if ((name[0] == '\0') || (rtn >= sizeof (name))) {
(void) sprintf(name, "host");
rtn = 0;
}
break;
case SVM_TAG_SET:
if (setname == NULL) {
(void) sprintf(name, "diskset");
} else {
rtn = snprintf(name, sizeof (name), "%s", setname);
}
break;
default:
if ((setname = get_devname(setno, dev)) != NULL) {
rtn = snprintf(name, sizeof (name), "%s", setname);
}
break;
}
mdclrerror(&status);
/* Check if we got any rubbish for any of the snprintf's */
if ((name[0] == '\0') || (rtn >= sizeof (name))) {
return (NULL);
}
return (strdup(name));
}
/*
* initialize all devices in set
*/
static int
init_all(
mdsetname_t **spp,
mdcmdopts_t options,
bool_t called_thru_rpc,
md_error_t *ep
)
{
md_tab_t *tabp = NULL;
size_t setlen;
uint_t more;
int done;
int eval = -1;
/*
* Only take the lock if this is not a MN set
* We can only enter this code for a MN set if we are the initiator
* and in this case, we don't want to take locks.
*/
if (meta_is_mn_set((*spp), ep) == 0) {
/* grab set lock */
if (meta_lock(*spp, TRUE, ep)) {
mde_perror(ep, "");
mdclrerror(ep);
return (eval);
}
/* check for ownership */
if (meta_check_ownership(*spp, ep) != 0) {
mde_perror(ep, "");
mdclrerror(ep);
return (eval);
}
/* lock is held across init_entries */
options |= MDCMD_NOLOCK;
}
/* get md.tab, preen entries */
if ((tabp = meta_tab_parse(NULL, ep)) == NULL) {
mde_perror(ep, "");
mdclrerror(ep);
return (eval);
}
setlen = strlen((*spp)->setname);
for (more = 0; (more < tabp->nlines); ++more) {
md_tab_line_t *linep = &tabp->lines[more];
char *cname = linep->cname;
char *p;
size_t len;
/* better have args */
assert((linep->argc > 0) && (linep->argv[0] != NULL));
/* only do metadevices and hotspare pools in set */
if (linep->type & TAB_MD_HSP) {
if ((p = strrchr(cname, '/')) == NULL) {
len = 0;
} else {
len = p - cname;
}
if ((len == setlen) &&
(strncmp(cname, (*spp)->setname, len) == 0)) {
linep->flags = DO_AGAIN;
} else {
linep->flags = DONT_DO;
}
} else {
linep->flags = DONT_DO;
}
}
eval = 1;
/* while more devices get made */
do {
done = init_entries(spp, tabp, options,
MD_IGNORE_STDERR|MD_RETRY_BUSY, called_thru_rpc, ep);
} while (done > 0);
/* now do it and report errors */
if (init_entries(spp, tabp, options, MD_RETRY_BUSY,
called_thru_rpc, ep) >= 0)
eval = 0; /* success */
mdclrerror(ep);
/* cleanup, return success */
out:
meta_tab_free(tabp);
return (eval);
}
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);
}
static int
stripe_replace(replace_params_t *params)
{
minor_t mnum = params->mnum;
ms_unit_t *un;
mddb_recid_t recids[6];
ms_new_dev_t nd;
ms_cd_info_t cd;
int ci;
int cmpcnt;
void *repl_data;
md_dev64_t fake_devt;
void (*repl_done)();
mdclrerror(¶ms->mde);
un = (ms_unit_t *)MD_UNIT(mnum);
if (MD_STATUS(un) & MD_UN_RESYNC_ACTIVE) {
return (mdmderror(¶ms->mde, MDE_RESYNC_ACTIVE, mnum));
}
nd.nd_dev = params->new_dev;
nd.nd_key = params->new_key;
nd.nd_nblks = params->number_blks;
nd.nd_start_blk = params->start_blk;
nd.nd_labeled = params->has_label;
nd.nd_hs_id = 0;
/*
* stripe_component_count and stripe_get_dev only care about the
* minor number associated with the first argument which is a
* md_dev64_t
*
* The comments section for these two routines have been updated
* to indicate that this routine calls with fake major numbers.
*/
fake_devt = md_makedevice(0, mnum);
cmpcnt = stripe_component_count(fake_devt, NULL);
for (ci = 0; ci < cmpcnt; ci++) {
(void) stripe_get_dev(fake_devt, NULL, ci, &cd);
if ((cd.cd_dev == params->old_dev) ||
(cd.cd_orig_dev == params->old_dev))
break;
}
if (ci == cmpcnt) {
return (EINVAL);
}
/* In case of a dryrun we're done here */
if (params->options & MDIOCTL_DRYRUN) {
return (0);
}
(void) stripe_replace_dev(fake_devt, 0, ci, &nd, recids, 6,
&repl_done, &repl_data);
mddb_commitrecs_wrapper(recids);
(*repl_done)(fake_devt, repl_data);
SE_NOTIFY(EC_SVM_CONFIG, ESC_SVM_REPLACE, SVM_TAG_METADEVICE,
MD_UN2SET(un), MD_SID(un));
return (0);
}
static int
stripe_getdevs(
void *d,
int mode,
IOLOCK *lock
)
{
minor_t mnum;
mdi_unit_t *ui;
ms_unit_t *un;
struct ms_row *mdr;
ms_comp_t *mdcomp, *mdc;
int r, c, i;
int cnt;
md_error_t *mdep;
md_dev64_t *devsp;
md_dev64_t unit_dev;
md_getdevs_params_t *mgdp = d;
mnum = mgdp->mnum;
mdep = &mgdp->mde;
/* check out unit */
mdclrerror(mdep);
if ((MD_MIN2SET(mnum) >= md_nsets) || (MD_MIN2UNIT(mnum) >= md_nunits))
return (mdmderror(mdep, MDE_INVAL_UNIT, mnum));
if ((ui = MDI_UNIT(mnum)) == NULL) {
return (mdmderror(mdep, MDE_UNIT_NOT_SETUP, mnum));
}
un = (ms_unit_t *)md_ioctl_readerlock(lock, ui);
mdcomp = (struct ms_comp *)((void *)&((char *)un)[un->un_ocomp]);
devsp = (md_dev64_t *)(uintptr_t)mgdp->devs;
for (cnt = 0, 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 (cnt < mgdp->cnt) {
unit_dev = mdc->un_dev;
if (md_getmajor(unit_dev) != md_major) {
if ((unit_dev = md_xlate_mini_2_targ
(unit_dev)) == NODEV64)
return (ENODEV);
}
if (ddi_copyout((caddr_t)&unit_dev, devsp,
sizeof (*devsp), mode) != 0)
return (EFAULT);
++devsp;
}
++cnt;
}
}
mgdp->cnt = cnt;
return (0);
}
/*ARGSUSED*/
int
meta_get_drive_names(
mdsetname_t *sp,
mddrivenamelist_t **dnlpp,
int options,
md_error_t *ep
)
{
mhd_did_flags_t flags = MHD_DID_SERIAL;
mhd_drive_info_list_t list;
mhd_drive_info_t *mp;
uint_t i;
unsigned cnt = 0;
int rval = -1;
mddrivenamelist_t **tailpp = dnlpp;
/* must have a set */
assert(sp != NULL);
load_paths_to_metamhd();
(void) memset(&list, 0, sizeof (list));
if ((meta_list_drives(NULL, NULL, flags, &list, ep)) < 0)
return (-1);
/* find drives in set */
for (i = 0; (i < list.mhd_drive_info_list_t_len); ++i) {
mddrivename_t *dnp;
mdname_t *np;
mp = &list.mhd_drive_info_list_t_val[i];
if (mp->dif_id.did_flags & MHD_DID_DUPLICATE)
continue;
/* quietly skip drives which don't conform */
if ((dnp = metadrivename(&sp, mp->dif_name, ep)) == NULL) {
mdclrerror(ep);
continue;
}
/* check in set */
if ((np = metaslicename(dnp, MD_SLICE0, ep)) == NULL)
goto out;
if (meta_check_inset(sp, np, ep) != 0) {
mdclrerror(ep);
continue;
}
/*
* Add the drivename struct to the end of the
* drivenamelist but keep a pointer to the last
* element so that we don't incur the overhead
* of traversing the list each time
*/
tailpp = meta_drivenamelist_append_wrapper(tailpp, dnp);
++cnt;
}
rval = cnt;
/* cleanup, return error */
out:
meta_free_drive_info_list(&list);
return (rval);
}
/*
* This routine is called from preenlib the first time. It is then
* recursively called through preen_subdev.
*
* The argument passed in (uname) starts with the special device from
* /etc/vfstab. Recursive calls pass in the underlying physical device
* names.
*/
void
preen_build_devs(
char *uname, /* name of metadevice */
struct dk_cinfo *dkiop, /* associated controller info */
void *dp /* magic info */
)
{
char *setname = NULL;
char *tname = NULL;
mdsetname_t *sp;
mdname_t *namep; /* metadevice name */
mdnamelist_t *nlp = NULL; /* list of real devices */
mdnamelist_t *p;
devid_nmlist_t *nm_list = NULL;
md_error_t status = mdnullerror;
md_error_t *ep = &status;
int ep_valid = 0; /* does ep contain a real error */
struct stat statb;
static int md_major = -1;
side_t sideno;
/*
* The rest of the code in this library can't work within a
* non-global zone so we just return the top level metadevice back
* to be fscked.
*/
if (getzoneid() != GLOBAL_ZONEID) {
preen_addunit(dp, dkiop->dki_dname, NULL, NULL,
dkiop->dki_unit);
return;
}
if (stat(uname, &statb) != 0)
return;
if (md_major == -1 &&
get_major_from_n2m(MD_MODULE, &md_major) != 0)
return;
/*
* If the path passed in is not a metadevice, then add that
* device to the list (preen_addunit) since it has to be a
* physical device.
*/
if (major(statb.st_rdev) != md_major) {
preen_addunit(dp, dkiop->dki_dname, NULL, NULL,
dkiop->dki_unit);
return;
}
/*
* Bind to the cluster library
*/
if (sdssc_bind_library() == SDSSC_ERROR)
return;
if (md_init_daemon("fsck", ep) != 0) {
ep_valid = 1;
goto out;
}
/*
* parse the path name to get the diskset name.
*/
parse_device(NULL, uname, &tname, &setname);
Free(tname);
if ((sp = metasetname(setname, ep)) == NULL) {
ep_valid = 1;
goto out;
}
/* check for ownership */
if (meta_check_ownership(sp, ep) != 0) {
/*
* Don't own the set but we are here implies
* that this is a clustered proxy device. Simply add
* the unit.
*/
preen_addunit(dp, dkiop->dki_dname, NULL, NULL,
dkiop->dki_unit);
ep_valid = 1;
goto out;
}
/*
* get list of underlying physical devices.
*/
if ((namep = metaname(&sp, uname, UNKNOWN, ep)) == NULL) {
ep_valid = 1;
goto out;
}
if (namep->dev == NODEV64) {
goto out;
}
if (meta_getdevs(sp, namep, &nlp, ep) != 0) {
ep_valid = 1;
goto out;
}
if ((sideno = getmyside(sp, ep)) == MD_SIDEWILD) {
ep_valid = 1;
goto out;
}
/* gather and add the underlying devs */
for (p = nlp; (p != NULL); p = p->next) {
mdname_t *devnp = p->namep;
int fd;
struct dk_cinfo cinfo;
ddi_devid_t md_did;
char *devname;
char *minor_name = NULL;
char mname[MAXPATHLEN];
/*
* we don't want to use the rname anymore because
* that may have changed. Use the device id information
* to find the correct ctd name and open based on that.
* If there isn't a devid or we have a did device, then
* use the rname. In clustering, it's corrected for us.
* If no devid it's at least worth a try.
*/
if (((md_did = meta_getdidbykey(sp->setno, sideno,
devnp->key, ep)) == NULL) || ((minor_name =
meta_getdidminorbykey(sp->setno, sideno,
devnp->key, ep)) == NULL)) {
devname = devnp->rname;
if (md_did)
Free(md_did);
} else {
if (strstr(minor_name, ",raw") == NULL) {
(void) snprintf(mname, MAXPATHLEN, "%s,raw",
minor_name);
} else {
(void) snprintf(mname, MAXPATHLEN, "%s",
minor_name);
}
/*
* We need to make sure we call this with a specific
* mname (raw mname) so that we get the exact slice
* with the given device id. Otherwise we could try
* to open a slice that doesn't really exist.
*/
if (meta_deviceid_to_nmlist("/dev", md_did,
mname, &nm_list) != 0) {
(void) mdsyserror(ep, errno, devnp->rname);
ep_valid = 1;
Free(md_did);
Free(minor_name);
goto out;
}
devname = Strdup(nm_list->devname);
Free(md_did);
Free(minor_name);
devid_free_nmlist(nm_list);
}
/* get device name and (real) cinfo */
if ((fd = open(devname, O_RDONLY, 0)) < 0) {
(void) mdsyserror(ep, errno, devname);
ep_valid = 1;
/*
* We need to scan all sub devices even if some fail
* since exit here would end up in not finishing fsck
* on all devices and prevent us from going into
* multiuser mode.
*/
continue;
}
if (ioctl(fd, DKIOCINFO, &cinfo) != 0) {
(void) mdsyserror(ep, errno, devname);
(void) close(fd);
ep_valid = 1;
/* Continue here too. See comment from before */
continue;
}
(void) close(fd); /* sd/ssd bug */
/*
* preen_subdev fails when the device name has been
* resolved to the physical layer. Hence it is added
* to preen_addunit.
*/
if (preen_subdev(devname, &cinfo, dp) != 0) {
preen_addunit(dp, cinfo.dki_dname, NULL, NULL,
cinfo.dki_unit);
}
}
/* cleanup, if we fail, just add this composite device to the list */
out:
if (setname != NULL)
Free(setname);
if (ep_valid != 0) {
mde_perror(&status, "");
mdclrerror(&status);
}
metafreenamelist(nlp);
}
