int
mds_bmap_read_v1(struct bmapc_memb *b, void *readh)
{
struct {
uint8_t crcstates[128];
uint8_t repls[24];
uint64_t crcs[128];
uint32_t gen;
uint32_t replpol;
} bod;
struct fidc_membh *f;
struct iovec iovs[2];
uint64_t crc, od_crc;
int i, rc, vfsid;
size_t nb, bsz;
struct bmap_mds_info *bmi = bmap_2_bmi(b);
bsz = sizeof(bod) + sizeof(crc);
iovs[0].iov_base = &bod;
iovs[0].iov_len = sizeof(bod);
iovs[1].iov_base = &od_crc;
iovs[1].iov_len = sizeof(od_crc);
f = b->bcm_fcmh;
slfid_to_vfsid(fcmh_2_fid(f), &vfsid);
rc = mdsio_preadv(vfsid, &rootcreds, iovs, nitems(iovs), &nb,
bsz * b->bcm_bmapno + 0x1000, readh);
if (rc)
return (rc);
if (nb == 0)
return (SLERR_BMAP_INVALID);
if (nb != bsz)
return (SLERR_SHORTIO);
psc_crc64_calc(&crc, &bod, sizeof(bod));
if (crc != od_crc)
return (PFLERR_BADCRC);
for (i = 0; i < 128; i++)
bmi->bmi_crcstates[i] = bod.crcstates[i];
for (i = 0; i < 24; i++)
bmi->bmi_repls[i] = bod.repls[i];
for (i = 0; i < 128; i++)
bmap_2_crcs(b, i) = bod.crcs[i];
bmap_2_bgen(b) = bod.gen;
bmap_2_replpol(b) = bod.replpol;
return (0);
}
/*
* 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);
}
