/*
* For the given bmap, change the status of all its replicas marked
* "valid" to "invalid" except for the replica specified.
*
* This is a high-level convenience call provided to easily update
* status after an ION has received some new I/O, which would make all
* other existing copies of the bmap on any other replicas old.
* @b: the bmap.
* @iosidx: the index of the only ION resource in the inode replica
* table that should be marked "valid".
*
* Note: All callers must journal log these bmap replica changes
* themselves. In addition, they must log any changes to the inode
* _before_ the bmap changes. Otherwise, we could end up actually
* having bmap replicas that are not recognized by the information
* stored in the inode during log replay.
*/
int
mds_repl_inv_except(struct bmap *b, int iosidx)
{
int rc, logit = 0, tract[NBREPLST], retifset[NBREPLST];
uint32_t policy;
/* Ensure replica on active IOS is marked valid. */
brepls_init(tract, -1);
tract[BREPLST_INVALID] = BREPLST_VALID;
tract[BREPLST_GARBAGE_SCHED] = BREPLST_VALID;
tract[BREPLST_GARBAGE_QUEUED] = BREPLST_VALID;
/*
* The old state for this bmap on the given IOS is
* either valid or invalid.
*/
brepls_init_idx(retifset);
retifset[BREPLST_INVALID] = 0;
retifset[BREPLST_VALID] = 0;
/*
* XXX on full truncate, the metafile will exist, which means
* the bmap states will exist, which means a new IOS will be
* selected which will probably be GARBAGE after truncate
* happens a few times.
*/
rc = mds_repl_bmap_walk(b, tract, retifset, 0, &iosidx, 1);
if (rc) {
psclog_errorx("bcs_repls has active IOS marked in a "
"weird state while invalidating other replicas; "
"fid="SLPRI_FID" bmap=%d iosidx=%d state=%d",
fcmh_2_fid(b->bcm_fcmh), b->bcm_bmapno, iosidx, rc);
}
policy = bmap_2_replpol(b);
/*
* Invalidate all other replicas.
* Note: if the status is SCHED here, don't do anything; once
* the replication status update comes from the ION, we will
* know he copied an old bmap and mark it OLD then.
*/
brepls_init(tract, -1);
tract[BREPLST_VALID] = policy == BRPOL_PERSIST ?
BREPLST_REPL_QUEUED : BREPLST_GARBAGE_QUEUED;
tract[BREPLST_REPL_SCHED] = BREPLST_REPL_QUEUED;
brepls_init(retifset, 0);
retifset[BREPLST_VALID] = 1;
retifset[BREPLST_REPL_SCHED] = 1;
if (_mds_repl_bmap_walk(b, tract, retifset, REPL_WALKF_MODOTH,
&iosidx, 1, NULL, NULL)) {
logit = 1;
BHGEN_INCREMENT(b);
}
if (logit)
rc = mds_bmap_write_logrepls(b);
else
rc = mds_bmap_write(b, NULL, NULL);
/*
* If this bmap is marked for persistent replication, the repl
* request must exist and should be marked such that the
* replication monitors do not release it in the midst of
* processing it as this activity now means they have more to
* do.
*/
if (policy == BRPOL_PERSIST)
upsch_enqueue(&bmap_2_bmi(b)->bmi_upd);
return (rc);
}
int
mds_inode_dump(int vfsid, struct sl_ino_compat *sic,
struct slash_inode_handle *ih, void *readh)
{
struct fidc_membh *f;
struct bmapc_memb *b;
struct mio_fh *fh;
sl_bmapno_t i;
int rc, fl;
void *th;
f = inoh_2_fcmh(ih);
th = inoh_2_mfh(ih);
fh = inoh_2_mfh(ih);
fl = BMAPGETF_CREATE | BMAPGETF_NOAUTOINST;
if (sic)
fl |= BMAPGETF_NORETRIEVE;
for (i = 0; ; i++) {
fh->fh = readh;
rc = bmap_getf(f, i, SL_WRITE, fl, &b);
fh->fh = th;
if (rc == SLERR_BMAP_INVALID) {
(void)INOH_RLOCK(ih);
break;
}
if (rc)
return (rc);
if (sic) {
rc = sic->sic_read_bmap(b, readh);
if (rc) {
bmap_op_done(b);
(void)INOH_RLOCK(ih);
if (rc == SLERR_BMAP_INVALID)
break;
return (rc);
}
}
rc = mds_bmap_write(b, NULL, NULL);
bmap_op_done(b);
(void)INOH_RLOCK(ih);
if (rc)
return (rc);
}
rc = mds_inox_write(vfsid, ih, NULL, NULL);
if (rc)
return (rc);
rc = mds_inode_write(vfsid, ih, NULL, NULL);
if (rc)
return (rc);
mdsio_fsync(vfsid, &rootcreds, 1, th);
return (0);
}
