stats = exp->exp_nid_stats;
LASSERT(stats != NULL);
- num_stats = NUM_OBD_STATS + LPROC_OFD_LAST;
+ num_stats = NUM_OBD_STATS + LPROC_OFD_STATS_LAST;
+
stats->nid_stats = lprocfs_alloc_stats(num_stats,
LPROCFS_STATS_FLAG_NOPERCPU);
if (stats->nid_stats == NULL)
return -ENOMEM;
- lprocfs_init_ops_stats(LPROC_OFD_LAST, stats->nid_stats);
- lprocfs_counter_init(stats->nid_stats, LPROC_OFD_READ_BYTES,
- LPROCFS_CNTR_AVGMINMAX, "read_bytes", "bytes");
- lprocfs_counter_init(stats->nid_stats, LPROC_OFD_WRITE_BYTES,
- LPROCFS_CNTR_AVGMINMAX, "write_bytes", "bytes");
-
+ lprocfs_init_ops_stats(LPROC_OFD_STATS_LAST, stats->nid_stats);
+ ofd_stats_counter_init(stats->nid_stats);
rc = lprocfs_register_stats(stats->nid_proc, "stats",
stats->nid_stats);
if (rc)
fed->fed_group = data->ocd_group;
data->ocd_connect_flags &= OST_CONNECT_SUPPORTED;
- exp->exp_connect_data = *data;
data->ocd_version = LUSTRE_VERSION_CODE;
/* Kindly make sure the SKIP_ORPHAN flag is from MDS. */
data->ocd_grant_extent = ofd->ofd_dt_conf.ddp_grant_frag >> 10;
}
- if (exp_connect_flags(exp) & OBD_CONNECT_GRANT)
+ if (data->ocd_connect_flags & OBD_CONNECT_GRANT)
data->ocd_grant = ofd_grant_connect(env, exp, data->ocd_grant,
new_connection);
ofd = ofd_dev(obd->obd_lu_dev);
- rc = lu_env_refill((struct lu_env *)env);
- if (rc != 0) {
- CERROR("Failure to refill session: '%d'\n", rc);
- RETURN(rc);
- }
-
- ofd_info_init(env, exp);
rc = ofd_parse_connect_data(env, exp, data, false);
if (rc == 0)
ofd_export_stats_init(ofd, exp, localdata);
exp = class_conn2export(&conn);
LASSERT(exp != NULL);
- rc = lu_env_refill((struct lu_env *)env);
- if (rc != 0) {
- CERROR("Failure to refill session: '%d'\n", rc);
- GOTO(out, rc);
- }
-
- ofd_info_init(env, exp);
-
rc = ofd_parse_connect_data(env, exp, data, true);
if (rc)
GOTO(out, rc);
ofd_grant_discard(exp);
- rc = lu_env_init(&env, LCT_DT_THREAD);
- if (rc)
- RETURN(rc);
-
/* Do not erase record for recoverable client. */
if (exp->exp_obd->obd_replayable &&
- (!exp->exp_obd->obd_fail || exp->exp_failed))
- tgt_client_del(&env, exp);
- lu_env_fini(&env);
+ (!exp->exp_obd->obd_fail || exp->exp_failed)) {
+ rc = lu_env_init(&env, LCT_DT_THREAD);
+ if (rc)
+ GOTO(out, rc);
+ tgt_client_del(&env, exp);
+ lu_env_fini(&env);
+ }
+out:
class_export_put(exp);
RETURN(rc);
}
spin_lock_init(&exp->exp_filter_data.fed_lock);
CFS_INIT_LIST_HEAD(&exp->exp_filter_data.fed_mod_list);
+ atomic_set(&exp->exp_filter_data.fed_soft_sync_count, 0);
spin_lock(&exp->exp_lock);
exp->exp_connecting = 1;
spin_unlock(&exp->exp_lock);
GOTO(out, rc = 0);
}
- rc = ostid_to_fid(&fid, &fm_key->oa.o_oi, 0);
+ rc = ostid_to_fid(&fid, &fm_key->oa.o_oi,
+ ofd->ofd_lut.lut_lsd.lsd_osd_index);
if (rc != 0)
GOTO(out, rc);
CDEBUG(D_INODE, "get FIEMAP of object "DFID"\n",
ofd_read_unlock(env, fo);
ofd_object_put(env, fo);
}
- } else if (KEY_IS(KEY_SYNC_LOCK_CANCEL)) {
- *((__u32 *) val) = ofd->ofd_sync_lock_cancel;
- *vallen = sizeof(__u32);
} else if (KEY_IS(KEY_LAST_FID)) {
struct ofd_device *ofd = ofd_exp(exp);
struct ofd_seq *oseq;
GOTO(out, rc = PTR_ERR(oseq));
rc = ostid_to_fid(fid, &oseq->os_oi,
- ofd->ofd_lut.lut_lsd.lsd_osd_index);
+ ofd->ofd_lut.lut_lsd.lsd_osd_index);
if (rc != 0)
GOTO(out_put, rc);
return 0;
}
-static int ofd_statfs(const struct lu_env *env, struct obd_export *exp,
- struct obd_statfs *osfs, __u64 max_age, __u32 flags)
+int ofd_statfs(const struct lu_env *env, struct obd_export *exp,
+ struct obd_statfs *osfs, __u64 max_age, __u32 flags)
{
struct obd_device *obd = class_exp2obd(exp);
struct ofd_device *ofd = ofd_dev(exp->exp_obd->obd_lu_dev);
info = ofd_info_init(env, exp);
ofd_oti2info(info, oti);
- rc = ostid_to_fid(&info->fti_fid, &oinfo->oi_oa->o_oi, 0);
- if (rc != 0)
- RETURN(rc);
-
+ info->fti_fid = oinfo->oi_oa->o_oi.oi_fid;
ost_fid_build_resid(&info->fti_fid, &info->fti_resid);
rc = ofd_auth_capa(exp, &info->fti_fid, ostid_seq(&oa->o_oi),
oinfo_capa(oinfo), CAPA_OPC_META_WRITE);
if (rc)
GOTO(out_unlock, rc);
- res = ldlm_resource_get(ns, NULL, &info->fti_resid, LDLM_EXTENT, 0);
- if (res != NULL) {
- ldlm_res_lvbo_update(res, NULL, 0);
- ldlm_resource_putref(res);
- }
-
- oinfo->oi_oa->o_valid = OBD_MD_FLID;
-
- /* Quota release needs uid/gid info */
- rc = ofd_attr_get(env, fo, &info->fti_attr);
- obdo_from_la(oinfo->oi_oa, &info->fti_attr,
- OFD_VALID_FLAGS | LA_UID | LA_GID);
ofd_info2oti(info, oti);
ofd_counter_incr(exp, LPROC_OFD_STATS_SETATTR, oti->oti_jobid, 1);
out_unlock:
ofd_object_put(env, fo);
out:
+ if (rc == 0) {
+ /* we do not call this before to avoid lu_object_find() in
+ * ->lvbo_update() holding another reference on the object.
+ * otherwise concurrent destroy can make the object unavailable
+ * for 2nd lu_object_find() waiting for the first reference
+ * to go... deadlock! */
+ res = ldlm_resource_get(ns, NULL, &info->fti_resid, LDLM_EXTENT, 0);
+ if (res != NULL) {
+ ldlm_res_lvbo_update(res, NULL, 0);
+ ldlm_resource_putref(res);
+ }
+ }
+
return rc;
}
info = ofd_info_init(env, exp);
ofd_oti2info(info, oti);
- rc = ostid_to_fid(&info->fti_fid, &oinfo->oi_oa->o_oi, 0);
- if (rc != 0)
- RETURN(rc);
+ info->fti_fid = oinfo->oi_oa->o_oi.oi_fid;
ost_fid_build_resid(&info->fti_fid, &info->fti_resid);
CDEBUG(D_INODE, "calling punch for object "DFID", valid = "LPX64
return rc;
}
-static int ofd_destroy_by_fid(const struct lu_env *env,
- struct ofd_device *ofd,
- const struct lu_fid *fid, int orphan)
+int ofd_destroy_by_fid(const struct lu_env *env, struct ofd_device *ofd,
+ const struct lu_fid *fid, int orphan)
{
struct ofd_thread_info *info = ofd_info(env);
struct lustre_handle lockh;
{
struct ofd_device *ofd = ofd_exp(exp);
struct ofd_thread_info *info;
+ struct lu_fid *fid;
+ obd_id oid;
obd_count count;
int rc = 0;
info = ofd_info_init(env, exp);
ofd_oti2info(info, oti);
+ fid = &info->fti_fid;
if (!(oa->o_valid & OBD_MD_FLGROUP))
ostid_set_seq_mdt0(&oa->o_oi);
+ *fid = oa->o_oi.oi_fid;
+ oid = ostid_id(&oa->o_oi);
+ LASSERT(oid != 0);
+
/* check that o_misc makes sense */
if (oa->o_valid & OBD_MD_FLOBJCOUNT)
count = oa->o_misc;
else
count = 1; /* default case - single destroy */
- /**
- * There can be sequence of objects to destroy. Therefore this request
- * may have multiple transaction involved in. It is OK, we need only
- * the highest used transno to be reported back in reply but not for
- * replays, they must report their transno
- */
- if (info->fti_transno == 0) /* not replay */
- info->fti_mult_trans = 1;
-
- CDEBUG(D_HA, "%s: Destroy object "DOSTID" count %d\n", ofd_name(ofd),
+ CDEBUG(D_INODE, "%s: Destroy object "DOSTID" count %d\n", ofd_name(ofd),
POSTID(&oa->o_oi), count);
+
while (count > 0) {
int lrc;
- lrc = ostid_to_fid(&info->fti_fid, &oa->o_oi, 0);
- if (lrc != 0) {
- if (rc == 0)
- rc = lrc;
- GOTO(out, rc);
- }
- lrc = ofd_destroy_by_fid(env, ofd, &info->fti_fid, 0);
+ lrc = ofd_destroy_by_fid(env, ofd, fid, 0);
if (lrc == -ENOENT) {
CDEBUG(D_INODE,
"%s: destroying non-existent object "DFID"\n",
- ofd_obd(ofd)->obd_name, PFID(&info->fti_fid));
+ ofd_obd(ofd)->obd_name, PFID(fid));
/* rewrite rc with -ENOENT only if it is 0 */
if (rc == 0)
rc = lrc;
} else if (lrc != 0) {
CERROR("%s: error destroying object "DFID": %d\n",
- ofd_obd(ofd)->obd_name, PFID(&info->fti_fid),
+ ofd_obd(ofd)->obd_name, PFID(fid),
rc);
rc = lrc;
}
+
count--;
- ostid_inc_id(&oa->o_oi);
+ oid++;
+ lrc = fid_set_id(fid, oid);
+ if (unlikely(lrc != 0 && count > 0))
+ GOTO(out, rc = lrc);
}
- /* if we have transaction then there were some deletions, we don't
- * need to return ENOENT in that case because it will not wait
- * for commit of these deletions. The ENOENT must be returned only
- * if there were no transations.
- */
- if (rc == -ENOENT) {
- if (info->fti_transno != 0)
- rc = 0;
- } else if (rc != 0) {
- /*
- * If we have at least one transaction then llog record
- * on server will be removed upon commit, so for rc != 0
- * we return no transno and llog record will be reprocessed.
- */
- info->fti_transno = 0;
- }
- ofd_info2oti(info, oti);
+ GOTO(out, rc);
+
out:
- RETURN(rc);
+ ofd_info2oti(info, oti);
+ fid_to_ostid(fid, &oa->o_oi);
+ return rc;
}
-static int ofd_orphans_destroy(const struct lu_env *env,
- struct obd_export *exp, struct ofd_device *ofd,
- struct obdo *oa)
+int ofd_orphans_destroy(const struct lu_env *env, struct obd_export *exp,
+ struct ofd_device *ofd, struct obdo *oa)
{
- struct ofd_thread_info *info = ofd_info(env);
+ struct ofd_thread_info *info = ofd_info(env);
+ struct lu_fid *fid = &info->fti_fid;
+ struct ost_id *oi = &oa->o_oi;
+ struct ofd_seq *oseq;
+ obd_seq seq = ostid_seq(oi);
+ obd_id end_id = ostid_id(oi);
obd_id last;
+ obd_id oid;
int skip_orphan;
- int rc = 0;
- struct ost_id oi = oa->o_oi;
- __u64 end_id = ostid_id(&oa->o_oi);
- struct ofd_seq *oseq;
-
+ int rc = 0;
ENTRY;
- oseq = ofd_seq_get(ofd, ostid_seq(&oa->o_oi));
+ oseq = ofd_seq_get(ofd, seq);
if (oseq == NULL) {
CERROR("%s: Can not find seq for "DOSTID"\n",
- ofd_name(ofd), POSTID(&oa->o_oi));
+ ofd_name(ofd), POSTID(oi));
RETURN(-EINVAL);
}
+ *fid = oi->oi_fid;
+ last = ofd_seq_last_oid(oseq);
+ oid = last;
+
LASSERT(exp != NULL);
skip_orphan = !!(exp_connect_flags(exp) & OBD_CONNECT_SKIP_ORPHAN);
- last = ofd_seq_last_oid(oseq);
LCONSOLE(D_INFO, "%s: deleting orphan objects from "DOSTID
- " to "DOSTID"\n", ofd_name(ofd), ostid_seq(&oa->o_oi),
- end_id + 1, ostid_seq(&oa->o_oi), last);
+ " to "DOSTID"\n", ofd_name(ofd), seq, end_id + 1, seq, last);
- for (ostid_set_id(&oi, last); ostid_id(&oi) > end_id;
- ostid_dec_id(&oi)) {
- rc = ostid_to_fid(&info->fti_fid, &oi, 0);
- if (rc != 0)
+ while (oid > end_id) {
+ rc = fid_set_id(fid, oid);
+ if (unlikely(rc != 0))
GOTO(out_put, rc);
- rc = ofd_destroy_by_fid(env, ofd, &info->fti_fid, 1);
- if (rc && rc != -ENOENT) /* this is pretty fatal... */
- CEMERG("%s: error destroying precreated id "DOSTID
- ": rc = %d\n", ofd_name(ofd), POSTID(&oi), rc);
+
+ rc = ofd_destroy_by_fid(env, ofd, fid, 1);
+ if (rc != 0 && rc != -ENOENT) /* this is pretty fatal... */
+ CEMERG("%s: error destroying precreated id "DFID
+ ": rc = %d\n", ofd_name(ofd), PFID(fid), rc);
+
+ oid--;
if (!skip_orphan) {
- ofd_seq_last_oid_set(oseq, ostid_id(&oi) - 1);
+ ofd_seq_last_oid_set(oseq, oid);
/* update last_id on disk periodically so that if we
* restart * we don't need to re-scan all of the just
* deleted objects. */
- if ((ostid_id(&oi) & 511) == 0)
+ if ((oid & 511) == 0)
ofd_seq_last_oid_write(env, ofd, oseq);
}
}
+
CDEBUG(D_HA, "%s: after destroy: set last_id to "DOSTID"\n",
- ofd_obd(ofd)->obd_name, POSTID(&oa->o_oi));
+ ofd_obd(ofd)->obd_name, seq, oid);
+
if (!skip_orphan) {
rc = ofd_seq_last_oid_write(env, ofd, oseq);
} else {
/* don't reuse orphan object, return last used objid */
- ostid_set_id(&oa->o_oi, last);
+ ostid_set_id(oi, last);
rc = 0;
}
+
+ GOTO(out_put, rc);
+
out_put:
ofd_seq_put(env, oseq);
- RETURN(rc);
+ return rc;
}
int ofd_create(const struct lu_env *env, struct obd_export *exp,
}
}
+ /* This can happen if a new OST is formatted and installed
+ * in place of an old one at the same index. Instead of
+ * precreating potentially millions of deleted old objects
+ * (possibly filling the OST), only precreate the last batch.
+ * LFSCK will eventually clean up any orphans. LU-14 */
+ if (diff > 5 * OST_MAX_PRECREATE) {
+ diff = OST_MAX_PRECREATE / 2;
+ LCONSOLE_WARN("%s: precreate FID "DOSTID" is over %u "
+ "larger than the LAST_ID "DOSTID", only "
+ "precreating the last %u objects.\n",
+ ofd_name(ofd), POSTID(&oa->o_oi),
+ 5 * OST_MAX_PRECREATE,
+ POSTID(&oseq->os_oi), diff);
+ ofd_seq_last_oid_set(oseq, ostid_id(&oa->o_oi) - diff);
+ }
+
while (diff > 0) {
next_id = ofd_seq_last_oid(oseq) + 1;
count = ofd_precreate_batch(ofd, diff);
info = ofd_info_init(env, exp);
- rc = ostid_to_fid(&info->fti_fid, &oinfo->oi_oa->o_oi, 0);
- if (rc != 0)
- GOTO(out, rc);
+ info->fti_fid = oinfo->oi_oa->o_oi.oi_fid;
rc = ofd_auth_capa(exp, &info->fti_fid, ostid_seq(&oinfo->oi_oa->o_oi),
oinfo_capa(oinfo), CAPA_OPC_META_READ);
if (rc)
}
info = ofd_info_init(env, exp);
- rc = ostid_to_fid(&info->fti_fid, &oinfo->oi_oa->o_oi, 0);
- if (rc != 0)
- GOTO(out, rc);
-
+ info->fti_fid = oinfo->oi_oa->o_oi.oi_fid;
rc = ofd_auth_capa(exp, &info->fti_fid, ostid_seq(&oinfo->oi_oa->o_oi),
oinfo_capa(oinfo), CAPA_OPC_OSS_TRUNC);
if (rc)
ostid_set_seq(&ostid, *(__u64 *)data->ioc_inlbuf4);
ostid_set_id(&ostid, *(__u64 *)data->ioc_inlbuf3);
- rc = ostid_to_fid(&fid, &ostid, 0);
+ rc = ostid_to_fid(&fid, &ostid,
+ ofd->ofd_lut.lut_lsd.lsd_osd_index);
if (rc != 0)
GOTO(out, rc);
} else {