*
* You should have received a copy of the GNU General Public License
* version 2 along with this program; If not, see
- * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
- *
- * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
- * CA 95054 USA or visit www.sun.com if you need additional information or
- * have any questions.
+ * http://www.gnu.org/licenses/gpl-2.0.html
*
* GPL HEADER END
*/
* Copyright (c) 2009, 2010, Oracle and/or its affiliates. All rights reserved.
* Use is subject to license terms.
*
- * Copyright (c) 2012, 2015, Intel Corporation.
+ * Copyright (c) 2012, 2016, Intel Corporation.
*/
/*
* This file is part of Lustre, http://www.lustre.org/
};
static const struct named_oid oids[] = {
- { LAST_RECV_OID, LAST_RCVD },
- { OFD_LAST_GROUP_OID, "LAST_GROUP" },
- { LLOG_CATALOGS_OID, "CATALOGS" },
- { MGS_CONFIGS_OID, NULL /*MOUNT_CONFIGS_DIR*/ },
- { FID_SEQ_SRV_OID, "seq_srv" },
- { FID_SEQ_CTL_OID, "seq_ctl" },
- { FLD_INDEX_OID, "fld" },
- { MDD_LOV_OBJ_OID, LOV_OBJID },
- { OFD_HEALTH_CHECK_OID, HEALTH_CHECK },
- { ACCT_USER_OID, "acct_usr_inode" },
- { ACCT_GROUP_OID, "acct_grp_inode" },
- { REPLY_DATA_OID, REPLY_DATA },
- { 0, NULL }
+ { .oid = LAST_RECV_OID, .name = LAST_RCVD },
+ { .oid = OFD_LAST_GROUP_OID, .name = "LAST_GROUP" },
+ { .oid = LLOG_CATALOGS_OID, .name = "CATALOGS" },
+ { .oid = MGS_CONFIGS_OID, /*MOUNT_CONFIGS_DIR*/ },
+ { .oid = FID_SEQ_SRV_OID, .name = "seq_srv" },
+ { .oid = FID_SEQ_CTL_OID, .name = "seq_ctl" },
+ { .oid = FLD_INDEX_OID, .name = "fld" },
+ { .oid = MDD_LOV_OBJ_OID, .name = LOV_OBJID },
+ { .oid = OFD_HEALTH_CHECK_OID, .name = HEALTH_CHECK },
+ { .oid = ACCT_USER_OID, .name = "acct_usr_inode" },
+ { .oid = ACCT_GROUP_OID, .name = "acct_grp_inode" },
+ { .oid = ACCT_PROJECT_OID, .name = "acct_prj_inode" },
+ { .oid = REPLY_DATA_OID, .name = REPLY_DATA },
+ { .oid = 0 }
};
static char *oid2name(const unsigned long oid)
{
struct zpl_direntry *zde = &osd_oti_get(env)->oti_zde.lzd_reg;
struct lu_attr *la = &osd_oti_get(env)->oti_la;
- dmu_buf_t *db;
+ sa_handle_t *sa_hdl = NULL;
dmu_tx_t *tx;
+ uint64_t oid;
int rc;
/* verify it doesn't already exist */
if (rc == 0)
return -EEXIST;
+ if (o->od_dt_dev.dd_rdonly)
+ return -EROFS;
+
/* create fid-to-dnode index */
tx = dmu_tx_create(o->od_os);
if (tx == NULL)
dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, 1, NULL);
dmu_tx_hold_bonus(tx, parent);
dmu_tx_hold_zap(tx, parent, TRUE, name);
- LASSERT(tx->tx_objset->os_sa);
dmu_tx_hold_sa_create(tx, ZFS_SA_BASE_ATTR_SIZE);
rc = -dmu_tx_assign(tx, TXG_WAIT);
return rc;
}
+ oid = osd_zap_create_flags(o->od_os, 0, ZAP_FLAG_HASH64,
+ DMU_OT_DIRECTORY_CONTENTS,
+ 14, /* == ZFS fzap_default_block_shift */
+ DN_MAX_INDBLKSHIFT,
+ 0, tx);
+
+ rc = -sa_handle_get(o->od_os, oid, NULL, SA_HDL_PRIVATE, &sa_hdl);
+ if (rc)
+ goto commit;
la->la_valid = LA_MODE | LA_UID | LA_GID;
la->la_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
la->la_uid = la->la_gid = 0;
- __osd_zap_create(env, o, &db, tx, la, parent, 0);
+ rc = __osd_attr_init(env, o, sa_hdl, tx, la, parent);
+ sa_handle_destroy(sa_hdl);
+ if (rc)
+ goto commit;
- zde->zde_dnode = db->db_object;
+ zde->zde_dnode = oid;
zde->zde_pad = 0;
zde->zde_type = IFTODT(S_IFDIR);
rc = -zap_add(o->od_os, parent, name, 8, 1, (void *)zde, tx);
+commit:
+ if (rc)
+ dmu_object_free(o->od_os, oid, tx);
dmu_tx_commit(tx);
- *child = db->db_object;
- sa_buf_rele(db, osd_obj_tag);
+ if (rc == 0)
+ *child = oid;
return rc;
}
rc = osd_fld_lookup(env, osd, fid_seq(fid), range);
if (rc != 0) {
+ /* During upgrade, OST FLDB might not be loaded because
+ * OST FLDB is not created until 2.6, so if some DNE
+ * filesystem upgrade from 2.5 to 2.7/2.8, they will
+ * not be able to find the sequence from local FLDB
+ * cache see fld_index_init(). */
+ if (rc == -ENOENT && osd->od_is_ost)
+ RETURN(1);
+
if (rc != -ENOENT)
CERROR("%s: "DFID" lookup failed: rc = %d\n",
osd_name(osd), PFID(fid), rc);
oi.oi_zapid = osd->od_O_id;
sprintf(seq_name, (fid_seq_is_rsvd(seq) ||
- fid_seq_is_mdt0(seq)) ? LPU64 : LPX64i,
+ fid_seq_is_mdt0(seq)) ? "%llu" : "%llx",
fid_seq_is_idif(seq) ? 0 : seq);
rc = osd_oi_find_or_create(env, osd, oi.oi_zapid, seq_name, &odb);
*/
static uint64_t
osd_get_idx_for_ost_obj(const struct lu_env *env, struct osd_device *osd,
- const struct lu_fid *fid, char *buf)
+ const struct lu_fid *fid, char *buf, int bufsize)
{
struct osd_seq *osd_seq;
unsigned long b;
b = id % OSD_OST_MAP_SIZE;
LASSERT(osd_seq->os_compat_dirs[b]);
- sprintf(buf, LPU64, id);
+ if (buf)
+ snprintf(buf, bufsize, "%llu", id);
return osd_seq->os_compat_dirs[b];
}
LASSERT(osd->od_oi_table != NULL);
oi = osd->od_oi_table[fid_seq(fid) & (osd->od_oi_count - 1)];
- osd_fid2str(buf, fid);
+ if (buf)
+ osd_fid2str(buf, fid);
return oi->oi_zapid;
}
uint64_t osd_get_name_n_idx(const struct lu_env *env, struct osd_device *osd,
- const struct lu_fid *fid, char *buf)
+ const struct lu_fid *fid, char *buf, int bufsize)
{
uint64_t zapid;
LASSERT(fid);
- LASSERT(buf);
if (fid_is_on_ost(env, osd, fid) == 1 || fid_seq(fid) == FID_SEQ_ECHO) {
- zapid = osd_get_idx_for_ost_obj(env, osd, fid, buf);
+ zapid = osd_get_idx_for_ost_obj(env, osd, fid, buf, bufsize);
} else if (unlikely(fid_seq(fid) == FID_SEQ_LOCAL_FILE)) {
/* special objects with fixed known fids get their name */
char *name = oid2name(fid_oid(fid));
if (name) {
zapid = osd->od_root;
- strcpy(buf, name);
+ if (buf)
+ strncpy(buf, name, bufsize);
if (fid_is_acct(fid))
zapid = MASTER_NODE_OBJ;
} else {
fid_oid(fid) == OSD_FS_ROOT_OID;
}
+static inline int osd_oid(struct osd_device *dev, __u32 local_oid,
+ uint64_t *oid)
+{
+ switch (local_oid) {
+ case ACCT_USER_OID:
+ *oid = dev->od_iusr_oid;
+ return 0;
+ case ACCT_GROUP_OID:
+ *oid = dev->od_igrp_oid;
+ return 0;
+ case ACCT_PROJECT_OID:
+ /* TODO: real oid */
+ CERROR("unsupported quota oid: %#x\n", local_oid);
+ return -ENOTSUPP;
+ }
+
+ LASSERTF(0, "invalid oid: %u for quota type", local_oid);
+ return -ENOTSUPP;
+}
+
int osd_fid_lookup(const struct lu_env *env, struct osd_device *dev,
const struct lu_fid *fid, uint64_t *oid)
{
RETURN(-ENOENT);
if (unlikely(fid_is_acct(fid))) {
- if (fid_oid(fid) == ACCT_USER_OID)
- *oid = dev->od_iusr_oid;
- else
- *oid = dev->od_igrp_oid;
+ rc = osd_oid(dev, fid_oid(fid), oid);
+ if (rc)
+ RETURN(rc);
} else if (unlikely(fid_is_fs_root(fid))) {
*oid = dev->od_root;
} else {
- zapid = osd_get_name_n_idx(env, dev, fid, buf);
-
+ zapid = osd_get_name_n_idx(env, dev, fid, buf,
+ sizeof(info->oti_buf));
rc = -zap_lookup(dev->od_os, zapid, buf,
8, 1, &info->oti_zde);
if (rc)
oi = o->od_oi_table[key];
if (oi) {
- if (oi->oi_db)
- sa_buf_rele(oi->oi_db, osd_obj_tag);
+ if (oi->oi_dn)
+ osd_dnode_rele(oi->oi_dn);
OBD_FREE_PTR(oi);
o->od_oi_table[key] = NULL;
}
}
o->od_oi_table[key] = oi;
- __osd_obj2dbuf(env, o->od_os, oi->oi_zapid, &oi->oi_db);
+ __osd_obj2dnode(env, o->od_os, oi->oi_zapid, &oi->oi_dn);
return 0;
}
return 0;
}
+
+/*
+ * the following set of functions are used to maintain per-thread
+ * cache of FID->ino mapping. this mechanism is used to avoid
+ * expensive LU/OI lookups.
+ */
+struct osd_idmap_cache *osd_idc_find(const struct lu_env *env,
+ struct osd_device *osd,
+ const struct lu_fid *fid)
+{
+ struct osd_thread_info *oti = osd_oti_get(env);
+ struct osd_idmap_cache *idc = oti->oti_ins_cache;
+ int i;
+
+ for (i = 0; i < oti->oti_ins_cache_used; i++) {
+ if (!lu_fid_eq(&idc[i].oic_fid, fid))
+ continue;
+ if (idc[i].oic_dev != osd)
+ continue;
+
+ return idc + i;
+ }
+
+ return NULL;
+}
+
+struct osd_idmap_cache *osd_idc_add(const struct lu_env *env,
+ struct osd_device *osd,
+ const struct lu_fid *fid)
+{
+ struct osd_thread_info *oti = osd_oti_get(env);
+ struct osd_idmap_cache *idc;
+ int i;
+
+ if (unlikely(oti->oti_ins_cache_used >= oti->oti_ins_cache_size)) {
+ i = oti->oti_ins_cache_size * 2;
+ LASSERT(i < 1000);
+ if (i == 0)
+ i = OSD_INS_CACHE_SIZE;
+ OBD_ALLOC(idc, sizeof(*idc) * i);
+ if (idc == NULL)
+ return ERR_PTR(-ENOMEM);
+ if (oti->oti_ins_cache != NULL) {
+ memcpy(idc, oti->oti_ins_cache,
+ oti->oti_ins_cache_used * sizeof(*idc));
+ OBD_FREE(oti->oti_ins_cache,
+ oti->oti_ins_cache_used * sizeof(*idc));
+ }
+ oti->oti_ins_cache = idc;
+ oti->oti_ins_cache_size = i;
+ }
+
+ idc = &oti->oti_ins_cache[oti->oti_ins_cache_used++];
+ idc->oic_fid = *fid;
+ idc->oic_dev = osd;
+ idc->oic_dnode = 0;
+ idc->oic_remote = 0;
+
+ return idc;
+}
+
+/**
+ * Lookup mapping for the given fid in the cache
+ *
+ * Initialize a new one if not found. the initialization checks whether
+ * the object is local or remote. for the local objects, OI is used to
+ * learn dnode#. the function is used when the caller has no information
+ * about the object, e.g. at dt_insert().
+ */
+struct osd_idmap_cache *osd_idc_find_or_init(const struct lu_env *env,
+ struct osd_device *osd,
+ const struct lu_fid *fid)
+{
+ struct osd_idmap_cache *idc;
+ int rc;
+
+ idc = osd_idc_find(env, osd, fid);
+ if (idc != NULL)
+ return idc;
+
+ /* new mapping is needed */
+ idc = osd_idc_add(env, osd, fid);
+ if (IS_ERR(idc))
+ return idc;
+
+ /* initialize it */
+ rc = osd_remote_fid(env, osd, fid);
+ if (unlikely(rc < 0))
+ return ERR_PTR(rc);
+
+ if (rc == 0) {
+ /* the object is local, lookup in OI */
+ uint64_t dnode;
+
+ rc = osd_fid_lookup(env, osd, fid, &dnode);
+ if (unlikely(rc < 0)) {
+ CERROR("%s: can't lookup: rc = %d\n",
+ osd->od_svname, rc);
+ return ERR_PTR(rc);
+ }
+ LASSERT(dnode < (1ULL << DN_MAX_OBJECT_SHIFT));
+ idc->oic_dnode = dnode;
+ } else {
+ /* the object is remote */
+ idc->oic_remote = 1;
+ }
+
+ return idc;
+}
+
+/*
+ * lookup mapping for given FID and fill it from the given object.
+ * the object is local by definition.
+ */
+int osd_idc_find_and_init(const struct lu_env *env, struct osd_device *osd,
+ struct osd_object *obj)
+{
+ const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
+ struct osd_idmap_cache *idc;
+
+ idc = osd_idc_find(env, osd, fid);
+ if (idc != NULL) {
+ if (obj->oo_dn == NULL)
+ return 0;
+ idc->oic_dnode = obj->oo_dn->dn_object;
+ return 0;
+ }
+
+ /* new mapping is needed */
+ idc = osd_idc_add(env, osd, fid);
+ if (IS_ERR(idc))
+ return PTR_ERR(idc);
+
+ if (obj->oo_dn)
+ idc->oic_dnode = obj->oo_dn->dn_object;
+
+ return 0;
+}
git://git.whamcloud.com / fs / lustre-release.git / blobdiff