void
slm_fcmh_dtor(struct fidc_membh *f)
{
struct fcmh_mds_info *fmi;
int rc, vfsid;
fmi = fcmh_2_fmi(f);
if (fcmh_isreg(f)) {
psc_assert(psc_dynarray_len(&fmi->fmi_ptrunc_clients) == 0);
psc_dynarray_free(&fmi->fmi_ptrunc_clients);
}
if (fcmh_isreg(f) || fcmh_isdir(f)) {
/* XXX Need to worry about other modes here */
if (!fmi->fmi_ctor_rc) {
slfid_to_vfsid(fcmh_2_fid(f), &vfsid);
rc = mdsio_release(vfsid, &rootcreds,
fcmh_2_mfh(f));
psc_assert(rc == 0);
}
}
if (fcmh_isdir(f)) {
slfid_to_vfsid(fcmh_2_fid(f), &vfsid);
rc = mdsio_release(vfsid, &rootcreds,
fcmh_2_dino_mfh(f));
psc_assert(rc == 0);
}
if (fmi->fmi_inodeh.inoh_extras)
PSCFREE(fmi->fmi_inodeh.inoh_extras);
}
/*
* Update the high-level app stat(2)-like attribute buffer for a FID
* cache member.
* @f: FID cache member to update.
* @sstb: incoming stat attributes.
* @flags: behavioral flags.
* Notes:
* (1) if SAVELOCAL has been specified, save local field values:
* (o) file size
* (o) mtime
* (2) This function should only be used by a client.
*/
void
slc_fcmh_setattrf(struct fidc_membh *f, struct srt_stat *sstb,
int flags)
{
uidmap_int_stat(sstb);
if (flags & FCMH_SETATTRF_HAVELOCK)
FCMH_LOCK_ENSURE(f);
else
FCMH_LOCK(f);
if (fcmh_2_gen(f) == FGEN_ANY)
fcmh_2_gen(f) = sstb->sst_gen;
if ((FID_GET_INUM(fcmh_2_fid(f))) != SLFID_ROOT &&
fcmh_2_gen(f) > sstb->sst_gen) {
OPSTAT_INCR("msl.generation-backwards");
DEBUG_FCMH(PLL_DIAG, f, "attempt to set attr with "
"gen %"PRIu64" from old gen %"PRIu64,
fcmh_2_gen(f), sstb->sst_gen);
goto out;
}
/*
* If we don't have stat attributes, how can we save our local
* updates?
*/
if ((f->fcmh_flags & FCMH_HAVE_ATTRS) == 0)
flags |= FCMH_SETATTRF_CLOBBER;
/*
* Always update for roots because we might have faked them
* with readdir at the super root.
*/
if ((FID_GET_INUM(fcmh_2_fid(f))) == SLFID_ROOT)
flags |= FCMH_SETATTRF_CLOBBER;
psc_assert(sstb->sst_gen != FGEN_ANY);
psc_assert(f->fcmh_fg.fg_fid == sstb->sst_fid);
/*
* The default behavior is to save st_size and st_mtim since we
* might have done I/O that the MDS does not know about.
*/
if ((flags & FCMH_SETATTRF_CLOBBER) == 0 &&
fcmh_isreg(f)) {
/*
* If generation numbers match, take the highest of the
* values. Otherwise, disregard local values and
* blindly accept whatever the MDS tells us.
*/
if (fcmh_2_ptruncgen(f) == sstb->sst_ptruncgen &&
fcmh_2_gen(f) == sstb->sst_gen &&
fcmh_2_fsz(f) > sstb->sst_size)
sstb->sst_size = fcmh_2_fsz(f);
if (fcmh_2_utimgen(f) == sstb->sst_utimgen)
sstb->sst_mtim = f->fcmh_sstb.sst_mtim;
}
COPY_SSTB(sstb, &f->fcmh_sstb);
f->fcmh_flags |= FCMH_HAVE_ATTRS;
f->fcmh_flags &= ~FCMH_GETTING_ATTRS;
if (sl_fcmh_ops.sfop_postsetattr)
sl_fcmh_ops.sfop_postsetattr(f);
DEBUG_FCMH(PLL_DEBUG, f, "attr set");
out:
if (!(flags & FCMH_SETATTRF_HAVELOCK))
FCMH_ULOCK(f);
}
int
slm_fcmh_ctor(struct fidc_membh *f, __unusedx int flags)
{
struct fcmh_mds_info *fmi;
struct mio_fh *ino_mfh;
struct slm_inoh *ih;
mio_fid_t ino_mfid;
int rc, vfsid;
DEBUG_FCMH(PLL_DIAG, f, "ctor");
rc = slfid_to_vfsid(fcmh_2_fid(f), &vfsid);
if (rc) {
DEBUG_FCMH(PLL_WARN, f, "invalid file system ID; "
"rc=%d", rc);
return (rc);
}
fmi = fcmh_2_fmi(f);
memset(fmi, 0, sizeof(*fmi));
rc = mdsio_lookup_slfid(vfsid, fcmh_2_fid(f), &rootcreds,
&f->fcmh_sstb, &fcmh_2_mfid(f));
if (rc) {
fmi->fmi_ctor_rc = rc;
DEBUG_FCMH(PLL_WARN, f, "mdsio_lookup_slfid failed; "
"fid="SLPRI_FID" rc=%d",
fcmh_2_fid(f), rc);
return (rc);
}
ih = &fmi->fmi_inodeh;
ih->inoh_flags = INOH_INO_NOTLOADED;
ino_mfid = fcmh_2_mfid(f);
ino_mfh = fcmh_2_mfhp(f);
if (fcmh_isdir(f)) {
mio_fid_t pmfid;
char fn[24];
rc = mdsio_opendir(vfsid, fcmh_2_mfid(f), &rootcreds,
NULL, &fcmh_2_mfh(f));
if (rc) {
DEBUG_FCMH(PLL_WARN, f, "mdsio_opendir failed; "
"mio_fid=%"PRIx64" rc=%d", fcmh_2_mfid(f),
rc);
return (rc);
}
snprintf(fn, sizeof(fn), "%016"PRIx64".ino",
fcmh_2_fid(f));
pmfid = mdsio_getfidlinkdir(fcmh_2_fid(f));
rc = mdsio_lookup(vfsid, pmfid, fn,
&fcmh_2_dino_mfid(f), &rootcreds, NULL);
if (rc == ENOENT) {
struct slm_inox_od inox;
rc = mdsio_opencreatef(vfsid, pmfid, &rootcreds,
O_CREAT | O_EXCL | O_RDWR,
MDSIO_OPENCRF_NOLINK, 0644, fn,
&fcmh_2_dino_mfid(f), NULL,
&fcmh_2_dino_mfh(f), NULL, NULL, 0);
psc_assert(rc == 0);
INOH_LOCK(ih);
rc = mds_inode_write(vfsid, ih, NULL, NULL);
psc_assert(rc == 0);
memset(&inox, 0, sizeof(inox));
ih->inoh_extras = &inox;
rc = mds_inox_write(vfsid, ih, NULL, NULL);
ih->inoh_extras = NULL;
INOH_ULOCK(ih);
psc_assert(rc == 0);
mdsio_release(vfsid, &rootcreds,
fcmh_2_dino_mfh(f));
} else if (rc) {
fmi->fmi_ctor_rc = rc;
DEBUG_FCMH(PLL_WARN, f,
"mdsio_lookup failed; rc=%d", rc);
return (rc);
}
ino_mfid = fcmh_2_dino_mfid(f);
ino_mfh = fcmh_2_dino_mfhp(f);
}
if (fcmh_isreg(f))
psc_dynarray_init(&fmi->fmi_ptrunc_clients);
if (fcmh_isdir(f) || fcmh_isreg(f)) {
/*
* We shouldn't need O_LARGEFILE because SLASH2
* metafiles are small.
*
* I created a file with size of 8070450532247928832
* using dd by seeking to a large offset and writing one
* byte. Somehow, the ZFS size becomes 5119601018368.
* Without O_LARGEFILE, I got EOVERFLOW (75) here. The
* SLASH2 size is correct though.
*/
rc = mdsio_opencreate(vfsid, ino_mfid, &rootcreds,
O_RDWR, 0, NULL, NULL, NULL, &ino_mfh->fh, NULL,
NULL, 0);
if (rc == 0) {
rc = mds_inode_read(&fmi->fmi_inodeh);
if (rc)
DEBUG_FCMH(PLL_WARN, f,
"could not load inode; "
"mfid=%"PRIx64" rc=%d",
ino_mfid, rc);
} else {
fmi->fmi_ctor_rc = rc;
DEBUG_FCMH(PLL_WARN, f,
"mdsio_opencreate failed; "
"mfid=%"PRIx64" rc=%d",
ino_mfid, rc);
}
} else
DEBUG_FCMH(PLL_DIAG, f, "special file, no zfs obj");
return (rc);
}
/*
* Handle a request to do replication from a client. May also
* reinitialize some parameters of the replication, such as priority, if
* the request already exists in the system.
*/
int
mds_repl_addrq(const struct sl_fidgen *fgp, sl_bmapno_t bmapno,
sl_bmapno_t *nbmaps, sl_replica_t *iosv, int nios, int sys_prio,
int usr_prio)
{
int tract[NBREPLST], ret_hasvalid[NBREPLST];
int iosidx[SL_MAX_REPLICAS], rc, flags;
sl_bmapno_t nbmaps_processed = 0;
struct fidc_membh *f = NULL;
struct bmap *b;
/* Perform sanity checks on request. */
if (nios < 1 || nios > SL_MAX_REPLICAS || *nbmaps == 0)
return (-EINVAL);
rc = slm_fcmh_get(fgp, &f);
if (rc)
return (-rc);
if (!fcmh_isdir(f) && !fcmh_isreg(f))
PFL_GOTOERR(out, rc = -PFLERR_NOTSUP);
/* Lookup replica(s)' indexes in our replica table. */
rc = -mds_repl_iosv_lookup_add(current_vfsid, fcmh_2_inoh(f),
iosv, iosidx, nios);
if (rc)
PFL_GOTOERR(out, rc);
/*
* If we are modifying a directory, we are done as just the
* replica table needs to be updated.
*/
if (fcmh_isdir(f))
PFL_GOTOERR(out, 0);
/*
* Setup structure to ensure at least one VALID replica exists.
*/
brepls_init(ret_hasvalid, 0);
ret_hasvalid[BREPLST_VALID] = 1;
/*
* Setup transitions to enqueue a replication.
*/
brepls_init(tract, -1);
tract[BREPLST_INVALID] = BREPLST_REPL_QUEUED;
tract[BREPLST_GARBAGE_SCHED] = BREPLST_REPL_QUEUED;
tract[BREPLST_GARBAGE_QUEUED] = BREPLST_REPL_QUEUED;
/* Wildcards shouldn't result in errors on zero-length files. */
if (*nbmaps != (sl_bmapno_t)-1)
rc = -SLERR_BMAP_INVALID;
for (; *nbmaps && bmapno < fcmh_nvalidbmaps(f);
bmapno++, --*nbmaps, nbmaps_processed++) {
if (nbmaps_processed >= SLM_REPLRQ_NBMAPS_MAX) {
rc = -PFLERR_WOULDBLOCK;
break;
}
rc = -bmap_get(f, bmapno, SL_WRITE, &b);
if (rc)
PFL_GOTOERR(out, rc);
/*
* If no VALID replicas exist, the bmap must be
* uninitialized/all zeroes; skip it.
*/
if (mds_repl_bmap_walk_all(b, NULL, ret_hasvalid,
REPL_WALKF_SCIRCUIT) == 0) {
bmap_op_done(b);
continue;
}
/*
* We do not follow the standard "retifset" API here
* because we need to preserve DIRTY if it gets set
* instead of some other state getting returned.
*/
flags = 0;
_mds_repl_bmap_walk(b, tract, NULL, 0, iosidx, nios,
slm_repl_addrq_cb, &flags);
/* both default to -1 in parse_replrq() */
bmap_2_bmi(b)->bmi_sys_prio = sys_prio;
bmap_2_bmi(b)->bmi_usr_prio = usr_prio;
if (flags & FLAG_DIRTY)
mds_bmap_write_logrepls(b);
else if (sys_prio != -1 || usr_prio != -1)
slm_repl_upd_write(b, 0);
bmap_op_done_type(b, BMAP_OPCNT_LOOKUP);
if (flags & FLAG_REPLICA_STATE_INVALID) {
/* See pfl_register_errno() */
rc = -SLERR_REPLICA_STATE_INVALID;
break;
}
}
out:
if (f)
fcmh_op_done(f);
*nbmaps = nbmaps_processed;
return (rc);
}
