int
slm_rmc_handle_getinode(struct pscrpc_request *rq)
{
const struct srm_get_inode_req *mq;
struct srm_get_inode_rep *mp;
struct fidc_membh *f;
SL_RSX_ALLOCREP(rq, mq, mp);
mp->rc = -slm_fcmh_get(&mq->fg, &f);
if (mp->rc)
return (0);
slm_pack_inode(f, &mp->ino);
fcmh_op_done(f);
return (0);
}
int
slm_rmc_handle_link(struct pscrpc_request *rq)
{
struct fidc_membh *p = NULL, *c = NULL;
struct srm_link_req *mq;
struct srm_link_rep *mp;
int vfsid;
SL_RSX_ALLOCREP(rq, mq, mp);
mp->rc = slfid_to_vfsid(mq->fg.fg_fid, &vfsid);
if (mp->rc)
PFL_GOTOERR(out, mp->rc);
mp->rc = -slm_fcmh_get(&mq->pfg, &p);
if (mp->rc)
PFL_GOTOERR(out, mp->rc);
mp->rc = -slm_fcmh_get(&mq->fg, &c);
if (mp->rc)
PFL_GOTOERR(out, mp->rc);
mq->name[sizeof(mq->name) - 1] = '\0';
mds_reserve_slot(1);
mp->rc = mdsio_link(vfsid, fcmh_2_mfid(c), fcmh_2_mfid(p),
mq->name, &rootcreds, mdslog_namespace);
mds_unreserve_slot(1);
mdsio_fcmh_refreshattr(c, &mp->cattr);
mdsio_fcmh_refreshattr(p, &mp->pattr);
out:
if (c)
fcmh_op_done(c);
if (p)
fcmh_op_done(p);
return (0);
}
/*
* Handle a BMAPCHWRMODE request to upgrade a client bmap lease from
* READ-only to READ+WRITE.
* @rq: RPC request.
*/
int
slm_rmc_handle_bmap_chwrmode(struct pscrpc_request *rq)
{
struct bmap_mds_lease *bml = NULL;
struct srm_bmap_chwrmode_req *mq;
struct srm_bmap_chwrmode_rep *mp;
struct fidc_membh *f = NULL;
struct bmapc_memb *b = NULL;
struct bmap_mds_info *bmi;
SL_RSX_ALLOCREP(rq, mq, mp);
mp->rc = -slm_fcmh_get(&mq->sbd.sbd_fg, &f);
if (mp->rc)
PFL_GOTOERR(out, mp->rc);
mp->rc = bmap_lookup(f, mq->sbd.sbd_bmapno, &b);
if (mp->rc)
PFL_GOTOERR(out, mp->rc);
bmi = bmap_2_bmi(b);
bml = mds_bmap_getbml(b, mq->sbd.sbd_seq,
mq->sbd.sbd_nid, mq->sbd.sbd_pid);
if (bml == NULL)
PFL_GOTOERR(out, mp->rc = -EINVAL);
mp->rc = mds_bmap_bml_chwrmode(bml, mq->prefios[0]);
if (mp->rc == -PFLERR_ALREADY)
mp->rc = 0;
else if (mp->rc)
PFL_GOTOERR(out, mp->rc);
mp->sbd = mq->sbd;
mp->sbd.sbd_seq = bml->bml_seq;
mp->sbd.sbd_key = bmi->bmi_assign->odtr_crc;
psc_assert(bmi->bmi_wr_ion);
mp->sbd.sbd_ios = rmmi2resm(bmi->bmi_wr_ion)->resm_res_id;
out:
if (bml)
mds_bmap_bml_release(bml);
if (b)
bmap_op_done(b);
if (f)
fcmh_op_done(f);
return (0);
}
int
slm_rmc_handle_lookup(struct pscrpc_request *rq)
{
struct fidc_membh *p = NULL;
struct srm_lookup_req *mq;
struct srm_lookup_rep *mp;
int vfsid;
SL_RSX_ALLOCREP(rq, mq, mp);
mp->rc = slfid_to_vfsid(mq->pfg.fg_fid, &vfsid);
if (mp->rc)
PFL_GOTOERR(out, mp->rc);
mp->rc = -slm_fcmh_get(&mq->pfg, &p);
if (mp->rc)
PFL_GOTOERR(out, mp->rc);
mq->name[sizeof(mq->name) - 1] = '\0';
psclog_diag("lookup: pfid="SLPRI_FID" name=%s", fcmh_2_mfid(p),
mq->name);
if (fcmh_2_mfid(p) == SLFID_ROOT &&
strcmp(mq->name, SL_RPATH_META_DIR) == 0)
PFL_GOTOERR(out, mp->rc = -EINVAL);
if (mq->pfg.fg_fid == SLFID_ROOT && use_global_mount) {
uint64_t fid;
struct sl_site *site;
mp->rc = -ENOENT;
CONF_LOCK();
CONF_FOREACH_SITE(site) {
if (strcmp(mq->name, site->site_name) != 0)
continue;
fid = SLFID_ROOT;
FID_SET_SITEID(fid, site->site_id);
mp->xattrsize = 0;
mp->attr.sst_fg.fg_fid = fid;
mp->attr.sst_fg.fg_gen = 2;
slm_root_attributes(&mp->attr);
mp->rc = 0;
break;
}
CONF_ULOCK();
goto out;
}
int
slm_rmc_handle_extendbmapls(struct pscrpc_request *rq)
{
struct srm_leasebmapext_req *mq;
struct srm_leasebmapext_rep *mp;
struct fidc_membh *f;
SL_RSX_ALLOCREP(rq, mq, mp);
mp->rc = -slm_fcmh_get(&mq->sbd.sbd_fg, &f);
if (mp->rc)
return (0);
mp->rc = mds_lease_renew(f, &mq->sbd, &mp->sbd, rq->rq_export);
fcmh_op_done(f);
return (0);
}
int
slm_rmc_handle_reassignbmapls(struct pscrpc_request *rq)
{
struct srm_reassignbmap_req *mq;
struct srm_reassignbmap_rep *mp;
struct fidc_membh *f;
SL_RSX_ALLOCREP(rq, mq, mp);
mp->rc = -slm_fcmh_get(&mq->sbd.sbd_fg, &f);
if (mp->rc)
return (0);
mp->rc = mds_lease_reassign(f, &mq->sbd, mq->pios,
mq->prev_sliods, mq->nreassigns, &mp->sbd, rq->rq_export);
fcmh_op_done(f);
return (0);
}
int
slm_rmc_handle_getattr(struct pscrpc_request *rq)
{
const struct srm_getattr_req *mq;
struct srm_getattr_rep *mp;
struct fidc_membh *f;
int vfsid;
SL_RSX_ALLOCREP(rq, mq, mp);
psclog_diag("pfid="SLPRI_FID, mq->fg.fg_fid);
if (mq->fg.fg_fid == SLFID_ROOT && use_global_mount) {
mp->attr.sst_fg.fg_fid = SLFID_ROOT;
mp->attr.sst_fg.fg_gen = FGEN_ANY-1;
slm_root_attributes(&mp->attr);
return (0);
}
mp->rc = -slm_fcmh_get(&mq->fg, &f);
if (mp->rc)
PFL_GOTOERR(out, mp->rc);
mp->rc = slfid_to_vfsid(mq->fg.fg_fid, &vfsid);
if (mp->rc)
PFL_GOTOERR(out, mp->rc);
mp->xattrsize = mdsio_hasxattrs(vfsid, &rootcreds,
fcmh_2_mfid(f));
FCMH_LOCK(f);
mp->attr = f->fcmh_sstb;
out:
if (f)
fcmh_op_done(f);
return (0);
}
int
slm_rmc_handle_getbmap(struct pscrpc_request *rq)
{
const struct srm_leasebmap_req *mq;
struct srm_leasebmap_rep *mp;
struct fidc_membh *f;
int rc = 0;
SL_RSX_ALLOCREP(rq, mq, mp);
if (mq->rw == SL_WRITE)
OPSTAT_INCR("getbmap-lease-write");
else if (mq->rw == SL_READ)
OPSTAT_INCR("getbmap-lease-read");
else {
mp->rc = -EINVAL;
return (0);
}
mp->rc = -slm_fcmh_get(&mq->fg, &f);
if (mp->rc)
return (0);
mp->flags = mq->flags;
mp->rc = mds_bmap_load_cli(f, mq->bmapno, mq->flags, mq->rw,
mq->prefios[0], &mp->sbd, rq->rq_export, mp->repls, 0);
if (mp->rc)
PFL_GOTOERR(out, mp->rc);
if (mp->flags & SRM_LEASEBMAPF_GETINODE)
slm_pack_inode(f, &mp->ino);
out:
fcmh_op_done(f);
return (rc ? rc : mp->rc);
}
/*
* Handle a NAMESPACE_FORWARD request from another MDS.
*/
int
slm_rmm_handle_namespace_forward(struct pscrpc_request *rq)
{
char *from, *to, *name, *linkname;
struct fidc_membh *p = NULL, *op = NULL, *np = NULL;
struct srm_forward_req *mq;
struct srm_forward_rep *mp;
struct slash_creds cr;
struct srt_stat sstb;
void *mfh;
int vfsid;
p = op = np = NULL;
SL_RSX_ALLOCREP(rq, mq, mp);
if (mq->op != SLM_FORWARD_MKDIR &&
mq->op != SLM_FORWARD_RMDIR &&
mq->op != SLM_FORWARD_CREATE &&
mq->op != SLM_FORWARD_UNLINK &&
mq->op != SLM_FORWARD_SYMLINK &&
mq->op != SLM_FORWARD_RENAME &&
mq->op != SLM_FORWARD_SETATTR) {
mp->rc = -EINVAL;
return (0);
}
psclog_info("op=%d, name=%s", mq->op, mq->req.name);
mp->rc = slfid_to_vfsid(mq->fg.fg_fid, &vfsid);
if (mp->rc)
return (0);
if (current_vfsid != vfsid) {
mp->rc = -EINVAL;
return (0);
}
cr.scr_uid = mq->creds.scr_uid;
cr.scr_gid = mq->creds.scr_gid;
mds_reserve_slot(2);
switch (mq->op) {
case SLM_FORWARD_MKDIR:
mp->rc = slm_fcmh_get(&mq->fg, &p);
if (mp->rc)
break;
sstb.sst_mode = mq->mode;
sstb.sst_uid = mq->creds.scr_uid;
sstb.sst_gid = mq->creds.scr_gid;
mp->rc = -mdsio_mkdir(vfsid, fcmh_2_mfid(p),
mq->req.name, &sstb, 0, 0, &mp->attr, NULL,
mdslog_namespace, slm_get_next_slashfid, 0);
break;
case SLM_FORWARD_CREATE:
mp->rc = slm_fcmh_get(&mq->fg, &p);
if (mp->rc)
break;
mp->rc = mdsio_opencreate(vfsid, fcmh_2_mfid(p),
&cr, O_CREAT | O_EXCL | O_RDWR, mq->mode,
mq->req.name, NULL, &mp->attr, &mfh,
mdslog_namespace, slm_get_next_slashfid, 0);
if (!mp->rc)
mdsio_release(vfsid, &rootcreds, mfh);
break;
case SLM_FORWARD_RMDIR:
mp->rc = slm_fcmh_get(&mq->fg, &p);
if (mp->rc)
break;
mp->rc = mdsio_rmdir(vfsid, fcmh_2_mfid(p), NULL,
mq->req.name, &rootcreds, mdslog_namespace);
break;
case SLM_FORWARD_UNLINK:
mp->rc = slm_fcmh_get(&mq->fg, &p);
if (mp->rc)
break;
mp->rc = -mdsio_unlink(vfsid, fcmh_2_mfid(p), NULL,
mq->req.name, &rootcreds, mdslog_namespace,
&mp->attr);
break;
case SLM_FORWARD_RENAME:
mp->rc = slm_fcmh_get(&mq->fg, &op);
if (mp->rc)
break;
mp->rc = slm_fcmh_get(&mq->nfg, &np);
if (mp->rc)
break;
from = mq->req.name;
to = mq->req.name + strlen(mq->req.name) + 1;
mp->rc = mdsio_rename(vfsid, fcmh_2_mfid(op), from,
fcmh_2_mfid(np), to, &rootcreds,
mdslog_namespace, &mp->attr);
break;
case SLM_FORWARD_SETATTR:
/*
* This is tough because we have some logic at the fcmh
* layer dealing with (partial) truncates. It is not a
* pure namespace operation.
*/
mp->rc = slm_fcmh_get(&mq->fg, &p);
if (mp->rc)
break;
mp->rc = -mdsio_setattr(vfsid, fcmh_2_mfid(p),
&mq->req.sstb, mq->to_set, &rootcreds, &mp->attr,
fcmh_2_mfh(p), mdslog_namespace);
break;
case SLM_FORWARD_SYMLINK:
mp->rc = slm_fcmh_get(&mq->fg, &p);
if (mp->rc)
break;
name = mq->req.name;
linkname = mq->req.name + strlen(mq->req.name) + 1;
mp->rc = mdsio_symlink(vfsid, linkname,
fcmh_2_mfid(p), name, &cr, &mp->attr, NULL,
NULL, slm_get_next_slashfid, 0);
break;
}
mds_unreserve_slot(2);
if (p)
fcmh_op_done(p);
if (op)
fcmh_op_done(op);
if (np)
fcmh_op_done(np);
return (0);
}
int
mds_repl_delrq(const struct sl_fidgen *fgp, sl_bmapno_t bmapno,
sl_bmapno_t *nbmaps, sl_replica_t *iosv, int nios)
{
int tract[NBREPLST], rc, iosidx[SL_MAX_REPLICAS], flags;
sl_bmapno_t nbmaps_processed = 0;
struct slm_repl_valid replv;
struct fidc_membh *f = NULL;
struct bmap *b;
if (nios < 1 || nios > SL_MAX_REPLICAS || *nbmaps == 0)
return (-EINVAL);
rc = slm_fcmh_get(fgp, &f);
if (rc)
return (-rc);
FCMH_LOCK(f);
if (fcmh_isdir(f))
flags = IOSV_LOOKUPF_DEL;
else
flags = IOSV_LOOKUPF_LOOKUP;
/* Find replica IOS indexes. */
rc = -_mds_repl_iosv_lookup(current_vfsid, fcmh_2_inoh(f), iosv,
iosidx, nios, flags);
if (fcmh_isdir(f) || rc)
PFL_GOTOERR(out, rc);
replv.nios = nios;
replv.idx = iosidx;
brepls_init(tract, -1);
tract[BREPLST_REPL_QUEUED] = BREPLST_GARBAGE_QUEUED;
tract[BREPLST_REPL_SCHED] = BREPLST_GARBAGE_QUEUED;
tract[BREPLST_VALID] = BREPLST_GARBAGE_QUEUED;
/* Wildcards shouldn't result in errors on zero-length files. */
if (*nbmaps != (sl_bmapno_t)-1)
rc = -SLERR_BMAP_INVALID;
/*
* The following loop will bail out on the very first error.
* However, its previous action, if any, has already taken
* effect.
*/
for (; *nbmaps && bmapno < fcmh_nvalidbmaps(f);
bmapno++, --*nbmaps, nbmaps_processed++) {
if (nbmaps_processed >= SLM_REPLRQ_NBMAPS_MAX)
PFL_GOTOERR(out, rc = -PFLERR_WOULDBLOCK);
rc = -bmap_get(f, bmapno, SL_WRITE, &b);
if (rc)
PFL_GOTOERR(out, rc);
/*
* Before blindly doing the transition, we have to check
* to ensure this operation would retain at least one
* valid replica.
*/
replv.n = 0;
mds_repl_bmap_walkcb(b, NULL, NULL, 0,
slm_repl_countvalid_cb, &replv);
flags = 0;
if (replv.n == 0)
rc = -SLERR_LASTREPL;
else {
rc = _mds_repl_bmap_walk(b, tract, NULL, 0, iosidx,
nios, slm_repl_delrq_cb, &flags);
psc_assert(!rc);
if (flags & FLAG_DIRTY)
rc = mds_bmap_write_logrepls(b);
}
bmap_op_done_type(b, BMAP_OPCNT_LOOKUP);
if (rc)
PFL_GOTOERR(out, rc);
}
out:
if (f)
fcmh_op_done(f);
*nbmaps = nbmaps_processed;
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);
}
