#define DEBUG_SUBSYSTEM S_OSD
-#include <lustre_ver.h>
#include <libcfs/libcfs.h>
#include <obd_support.h>
#include <lustre_net.h>
LASSERT(obj->oo_sa_hdl == NULL);
LASSERT(obj->oo_dn != NULL);
- rc = -sa_handle_get(o->od_os, obj->oo_dn->dn_object, obj,
- SA_HDL_PRIVATE, &obj->oo_sa_hdl);
+ rc = osd_sa_handle_get(obj);
if (rc)
return rc;
/*
* Add object to list of dirty objects in tx handle.
*/
-static void
-osd_object_sa_dirty_add(struct osd_object *obj, struct osd_thandle *oh)
+void osd_object_sa_dirty_add(struct osd_object *obj, struct osd_thandle *oh)
{
if (!list_empty(&obj->oo_sa_linkage))
return;
- down(&oh->ot_sa_lock);
write_lock(&obj->oo_attr_lock);
if (likely(list_empty(&obj->oo_sa_linkage)))
list_add(&obj->oo_sa_linkage, &oh->ot_sa_list);
write_unlock(&obj->oo_attr_lock);
- up(&oh->ot_sa_lock);
}
/*
* Release spill block dbuf hold for all dirty SAs.
*/
-void osd_object_sa_dirty_rele(struct osd_thandle *oh)
+void osd_object_sa_dirty_rele(const struct lu_env *env, struct osd_thandle *oh)
{
struct osd_object *obj;
- down(&oh->ot_sa_lock);
while (!list_empty(&oh->ot_sa_list)) {
obj = list_entry(oh->ot_sa_list.next,
struct osd_object, oo_sa_linkage);
- sa_spill_rele(obj->oo_sa_hdl);
write_lock(&obj->oo_attr_lock);
list_del_init(&obj->oo_sa_linkage);
write_unlock(&obj->oo_attr_lock);
+ if (obj->oo_late_xattr) {
+ /*
+ * take oo_guard to protect oo_sa_xattr buffer
+ * from concurrent update by osd_xattr_set()
+ */
+ LASSERT(oh->ot_assigned != 0);
+ down_write(&obj->oo_guard);
+ if (obj->oo_late_attr_set)
+ __osd_sa_attr_init(env, obj, oh);
+ else if (obj->oo_late_xattr)
+ __osd_sa_xattr_update(env, obj, oh);
+ up_write(&obj->oo_guard);
+ }
+ sa_spill_rele(obj->oo_sa_hdl);
}
- up(&oh->ot_sa_lock);
}
/*
{
struct osa_attr *osa = &osd_oti_get(env)->oti_osa;
sa_bulk_attr_t *bulk = osd_oti_get(env)->oti_attr_bulk;
- sa_handle_t *sa_hdl;
int cnt = 0;
int rc;
ENTRY;
LASSERT(obj->oo_dn != NULL);
- rc = -sa_handle_get(o->od_os, obj->oo_dn->dn_object, NULL,
- SA_HDL_PRIVATE, &sa_hdl);
- if (rc)
- RETURN(rc);
-
la->la_valid |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE | LA_TYPE |
LA_SIZE | LA_UID | LA_GID | LA_FLAGS | LA_NLINK;
SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(o), NULL, &osa->flags, 8);
LASSERT(cnt <= ARRAY_SIZE(osd_oti_get(env)->oti_attr_bulk));
- rc = -sa_bulk_lookup(sa_hdl, bulk, cnt);
+ rc = -sa_bulk_lookup(obj->oo_sa_hdl, bulk, cnt);
if (rc)
GOTO(out_sa, rc);
+#ifdef ZFS_PROJINHERIT
+ if (o->od_projectused_dn && osa->flags & ZFS_PROJID) {
+ rc = -sa_lookup(obj->oo_sa_hdl, SA_ZPL_PROJID(o),
+ &osa->projid, 8);
+ if (rc)
+ GOTO(out_sa, rc);
+
+ la->la_projid = osa->projid;
+ la->la_valid |= LA_PROJID;
+ obj->oo_with_projid = 1;
+ } else {
+ la->la_projid = ZFS_DEFAULT_PROJID;
+ la->la_valid &= ~LA_PROJID;
+ }
+#else
+ la->la_projid = 0;
+ la->la_valid &= ~LA_PROJID;
+#endif
+
la->la_atime = osa->atime[0];
la->la_mtime = osa->mtime[0];
la->la_ctime = osa->ctime[0];
}
if (S_ISCHR(la->la_mode) || S_ISBLK(la->la_mode)) {
- rc = -sa_lookup(sa_hdl, SA_ZPL_RDEV(o), &osa->rdev, 8);
+ rc = -sa_lookup(obj->oo_sa_hdl, SA_ZPL_RDEV(o), &osa->rdev, 8);
if (rc)
GOTO(out_sa, rc);
la->la_rdev = osa->rdev;
la->la_valid |= LA_RDEV;
}
out_sa:
- sa_handle_destroy(sa_hdl);
RETURN(rc);
}
-int __osd_obj2dnode(const struct lu_env *env, objset_t *os,
- uint64_t oid, dnode_t **dnp)
+int __osd_obj2dnode(objset_t *os, uint64_t oid, dnode_t **dnp)
{
- dmu_object_info_t *doi = &osd_oti_get(env)->oti_doi;
dmu_buf_t *db;
dmu_buf_impl_t *dbi;
int rc;
- rc = dmu_bonus_hold(os, oid, osd_obj_tag, &db);
+ rc = -dmu_bonus_hold(os, oid, osd_obj_tag, &db);
if (rc)
return rc;
dbi = (dmu_buf_impl_t *)db;
DB_DNODE_ENTER(dbi);
*dnp = DB_DNODE(dbi);
-
LASSERT(*dnp != NULL);
- dmu_object_info_from_dnode(*dnp, doi);
- if (unlikely (oid != DMU_USERUSED_OBJECT &&
- oid != DMU_GROUPUSED_OBJECT && doi->doi_bonus_type != DMU_OT_SA)) {
- osd_dnode_rele(*dnp);
- *dnp = NULL;
- return -EINVAL;
- }
return 0;
}
RETURN(rc);
}
+/**
+ * Helper function to retrieve DMU object id from fid for accounting object
+ */
+static dnode_t *osd_quota_fid2dmu(const struct osd_device *osd,
+ const struct lu_fid *fid)
+{
+ dnode_t *dn = NULL;
+
+ LASSERT(fid_is_acct(fid));
+
+ switch (fid_oid(fid)) {
+ case ACCT_USER_OID:
+ dn = osd->od_userused_dn;
+ break;
+ case ACCT_GROUP_OID:
+ dn = osd->od_groupused_dn;
+ break;
+#ifdef ZFS_PROJINHERIT
+ case ACCT_PROJECT_OID:
+ dn = osd->od_projectused_dn;
+ break;
+#endif
+ default:
+ break;
+ }
+
+ return dn;
+}
+
/*
* Concurrency: no concurrent access is possible that early in object
* life-cycle.
static int osd_object_init(const struct lu_env *env, struct lu_object *l,
const struct lu_object_conf *conf)
{
- struct osd_object *obj = osd_obj(l);
- struct osd_device *osd = osd_obj2dev(obj);
- uint64_t oid;
- int rc;
+ struct osd_object *obj = osd_obj(l);
+ struct osd_device *osd = osd_obj2dev(obj);
+ const struct lu_fid *fid = lu_object_fid(l);
+ uint64_t oid;
+ int rc = 0;
ENTRY;
LASSERT(osd_invariant(obj));
if (conf != NULL && conf->loc_flags & LOC_F_NEW)
GOTO(out, rc = 0);
- rc = osd_fid_lookup(env, osd, lu_object_fid(l), &oid);
+ if (unlikely(fid_is_acct(fid))) {
+ obj->oo_dn = osd_quota_fid2dmu(osd, fid);
+ if (obj->oo_dn) {
+ obj->oo_dt.do_index_ops = &osd_acct_index_ops;
+ l->lo_header->loh_attr |= LOHA_EXISTS;
+ }
+
+ GOTO(out, rc = 0);
+ }
+
+ rc = osd_fid_lookup(env, osd, fid, &oid);
if (rc == 0) {
LASSERT(obj->oo_dn == NULL);
- rc = __osd_obj2dnode(env, osd->od_os, oid, &obj->oo_dn);
+ rc = __osd_obj2dnode(osd->od_os, oid, &obj->oo_dn);
/* EEXIST will be returned if object is being deleted in ZFS */
if (rc == -EEXIST) {
rc = 0;
LASSERT(obj->oo_destroy == OSD_DESTROY_ASYNC);
/* the object is supposed to be exclusively locked by
- * the caller (osd_object_destroy()), while the transaction
+ * the caller (osd_destroy()), while the transaction
* (oh) is per-thread and not shared */
if (likely(list_empty(&obj->oo_unlinked_linkage))) {
list_add(&obj->oo_unlinked_linkage, &oh->ot_unlinked_list);
up_write(&obj->oo_guard);
}
-static int osd_declare_object_destroy(const struct lu_env *env,
- struct dt_object *dt,
- struct thandle *th)
+static int osd_declare_destroy(const struct lu_env *env, struct dt_object *dt,
+ struct thandle *th)
{
const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
struct osd_object *obj = osd_dt_obj(dt);
struct osd_device *osd = osd_obj2dev(obj);
struct osd_thandle *oh;
+ dnode_t *dn;
int rc;
uint64_t zapid;
ENTRY;
LASSERT(oh->ot_tx != NULL);
/* declare that we'll remove object from fid-dnode mapping */
- zapid = osd_get_name_n_idx(env, osd, fid, NULL, 0);
- dmu_tx_hold_zap(oh->ot_tx, zapid, FALSE, NULL);
+ zapid = osd_get_name_n_idx(env, osd, fid, NULL, 0, &dn);
+ osd_tx_hold_zap(oh->ot_tx, zapid, dn, FALSE, NULL);
osd_declare_xattrs_destroy(env, obj, oh);
- /* declare that we'll remove object from inode accounting ZAPs */
- dmu_tx_hold_zap(oh->ot_tx, osd->od_iusr_oid, FALSE, NULL);
- dmu_tx_hold_zap(oh->ot_tx, osd->od_igrp_oid, FALSE, NULL);
-
/* one less inode */
rc = osd_declare_quota(env, osd, obj->oo_attr.la_uid,
- obj->oo_attr.la_gid, -1, oh, false, NULL, false);
+ obj->oo_attr.la_gid, obj->oo_attr.la_projid,
+ -1, oh, NULL, OSD_QID_INODE);
if (rc)
RETURN(rc);
/* data to be truncated */
rc = osd_declare_quota(env, osd, obj->oo_attr.la_uid,
- obj->oo_attr.la_gid, 0, oh, true, NULL, false);
+ obj->oo_attr.la_gid, obj->oo_attr.la_projid,
+ 0, oh, NULL, OSD_QID_BLK);
if (rc)
RETURN(rc);
dmu_tx_hold_free(oh->ot_tx, obj->oo_dn->dn_object,
0, DMU_OBJECT_END);
else
- dmu_tx_hold_zap(oh->ot_tx, osd->od_unlinkedid, TRUE, NULL);
+ osd_tx_hold_zap(oh->ot_tx, osd->od_unlinked->dn_object,
+ osd->od_unlinked, TRUE, NULL);
/* will help to find FID->ino when this object is being
* added to PENDING/ */
RETURN(0);
}
-static int osd_object_destroy(const struct lu_env *env,
- struct dt_object *dt, struct thandle *th)
+static int osd_destroy(const struct lu_env *env, struct dt_object *dt,
+ struct thandle *th)
{
struct osd_thread_info *info = osd_oti_get(env);
char *buf = info->oti_str;
struct osd_thandle *oh;
int rc;
uint64_t oid, zapid;
+ dnode_t *zdn;
ENTRY;
down_write(&obj->oo_guard);
LASSERT(oh->ot_tx != NULL);
/* remove obj ref from index dir (it depends) */
- zapid = osd_get_name_n_idx(env, osd, fid, buf, sizeof(info->oti_str));
- rc = -zap_remove(osd->od_os, zapid, buf, oh->ot_tx);
+ zapid = osd_get_name_n_idx(env, osd, fid, buf,
+ sizeof(info->oti_str), &zdn);
+ rc = osd_zap_remove(osd, zapid, zdn, buf, oh->ot_tx);
if (rc) {
CERROR("%s: zap_remove(%s) failed: rc = %d\n",
osd->od_svname, buf, rc);
GOTO(out, rc);
}
- /* Remove object from inode accounting. It is not fatal for the destroy
- * operation if something goes wrong while updating accounting, but we
- * still log an error message to notify the administrator */
- rc = -zap_increment_int(osd->od_os, osd->od_iusr_oid,
- obj->oo_attr.la_uid, -1, oh->ot_tx);
- if (rc)
- CERROR("%s: failed to remove "DFID" from accounting ZAP for usr"
- " %d: rc = %d\n", osd->od_svname, PFID(fid),
- obj->oo_attr.la_uid, rc);
- rc = -zap_increment_int(osd->od_os, osd->od_igrp_oid,
- obj->oo_attr.la_gid, -1, oh->ot_tx);
- if (rc)
- CERROR("%s: failed to remove "DFID" from accounting ZAP for grp"
- " %d: rc = %d\n", osd->od_svname, PFID(fid),
- obj->oo_attr.la_gid, rc);
-
oid = obj->oo_dn->dn_object;
if (unlikely(obj->oo_destroy == OSD_DESTROY_NONE)) {
/* this may happen if the destroy wasn't declared
CERROR("%s: failed to free %s %llu: rc = %d\n",
osd->od_svname, buf, oid, rc);
} else { /* asynchronous destroy */
+ char *key = info->oti_key;
+
rc = osd_object_unlinked_add(obj, oh);
if (rc)
GOTO(out, rc);
- rc = -zap_add_int(osd->od_os, osd->od_unlinkedid,
- oid, oh->ot_tx);
+ snprintf(key, sizeof(info->oti_key), "%llx", oid);
+ rc = osd_zap_add(osd, osd->od_unlinked->dn_object,
+ osd->od_unlinked, key, 8, 1, &oid, oh->ot_tx);
if (rc)
CERROR("%s: zap_add_int() failed %s %llu: rc = %d\n",
osd->od_svname, buf, oid, rc);
static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
{
struct osd_object *obj = osd_obj(l);
-
- if (obj->oo_dn != NULL) {
- osd_object_sa_fini(obj);
- if (obj->oo_sa_xattr) {
- nvlist_free(obj->oo_sa_xattr);
- obj->oo_sa_xattr = NULL;
+ const struct lu_fid *fid = lu_object_fid(l);
+
+ if (obj->oo_dn) {
+ if (likely(!fid_is_acct(fid))) {
+ osd_object_sa_fini(obj);
+ if (obj->oo_sa_xattr) {
+ nvlist_free(obj->oo_sa_xattr);
+ obj->oo_sa_xattr = NULL;
+ }
+ osd_dnode_rele(obj->oo_dn);
+ list_del(&obj->oo_sa_linkage);
}
- osd_dnode_rele(obj->oo_dn);
- list_del(&obj->oo_sa_linkage);
obj->oo_dn = NULL;
}
}
return (*p)(env, cookie, LUSTRE_OSD_ZFS_NAME"-object@%p", o);
}
-static void osd_object_read_lock(const struct lu_env *env,
- struct dt_object *dt, unsigned role)
+static void osd_read_lock(const struct lu_env *env, struct dt_object *dt,
+ unsigned role)
{
struct osd_object *obj = osd_dt_obj(dt);
down_read_nested(&obj->oo_sem, role);
}
-static void osd_object_write_lock(const struct lu_env *env,
- struct dt_object *dt, unsigned role)
+static void osd_write_lock(const struct lu_env *env, struct dt_object *dt,
+ unsigned role)
{
struct osd_object *obj = osd_dt_obj(dt);
down_write_nested(&obj->oo_sem, role);
}
-static void osd_object_read_unlock(const struct lu_env *env,
- struct dt_object *dt)
+static void osd_read_unlock(const struct lu_env *env, struct dt_object *dt)
{
struct osd_object *obj = osd_dt_obj(dt);
up_read(&obj->oo_sem);
}
-static void osd_object_write_unlock(const struct lu_env *env,
- struct dt_object *dt)
+static void osd_write_unlock(const struct lu_env *env, struct dt_object *dt)
{
- struct osd_object *obj = osd_dt_obj(dt);
+ struct osd_object *obj = osd_dt_obj(dt);
- LASSERT(osd_invariant(obj));
+ LASSERT(osd_invariant(obj));
up_write(&obj->oo_sem);
}
-static int osd_object_write_locked(const struct lu_env *env,
- struct dt_object *dt)
+static int osd_write_locked(const struct lu_env *env, struct dt_object *dt)
{
struct osd_object *obj = osd_dt_obj(dt);
int rc = 1;
if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
GOTO(out, rc = -ENOENT);
+ if (unlikely(fid_is_acct(lu_object_fid(&dt->do_lu))))
+ GOTO(out, rc = 0);
+
LASSERT(osd_invariant(obj));
LASSERT(obj->oo_dn);
* anything else */
}
- if (attr && (attr->la_valid & (LA_UID | LA_GID))) {
+ if (attr && (attr->la_valid & (LA_UID | LA_GID | LA_PROJID))) {
sa_object_size(obj->oo_sa_hdl, &blksize, &bspace);
bspace = toqb(bspace * blksize);
}
if (attr && attr->la_valid & LA_UID) {
- /* account for user inode tracking ZAP update */
- dmu_tx_hold_zap(oh->ot_tx, osd->od_iusr_oid, FALSE, NULL);
-
/* quota enforcement for user */
if (attr->la_uid != obj->oo_attr.la_uid) {
rc = qsd_transfer(env, osd->od_quota_slave,
}
}
if (attr && attr->la_valid & LA_GID) {
- /* account for user inode tracking ZAP update */
- dmu_tx_hold_zap(oh->ot_tx, osd->od_igrp_oid, FALSE, NULL);
-
/* quota enforcement for group */
if (attr->la_gid != obj->oo_attr.la_gid) {
rc = qsd_transfer(env, osd->od_quota_slave,
GOTO(out, rc);
}
}
-
+#ifdef ZFS_PROJINHERIT
+ if (attr && attr->la_valid & LA_PROJID) {
+ if (!osd->od_projectused_dn)
+ GOTO(out, rc = -EOPNOTSUPP);
+
+ /* Usually, if project quota is upgradable for the device,
+ * then the upgrade will be done before or when mount the
+ * device. So when we come here, this project should have
+ * project ID attribute already (that is zero by default).
+ * Otherwise, there was something wrong during the former
+ * upgrade, let's return failure to report that.
+ *
+ * Please note that, different from other attributes, you
+ * can NOT simply set the project ID attribute under such
+ * case, because adding (NOT change) project ID attribute
+ * needs to change the object's attribute layout to match
+ * zfs backend quota accounting requirement. */
+ if (unlikely(!obj->oo_with_projid))
+ GOTO(out, rc = -ENXIO);
+
+ /* quota enforcement for project */
+ if (attr->la_projid != obj->oo_attr.la_projid) {
+ rc = qsd_transfer(env, osd->od_quota_slave,
+ &oh->ot_quota_trans, PRJQUOTA,
+ obj->oo_attr.la_projid,
+ attr->la_projid, bspace,
+ &info->oti_qi);
+ if (rc)
+ GOTO(out, rc);
+ }
+ }
+#endif
out:
up_read(&obj->oo_guard);
RETURN(rc);
if (rc < 0) {
CWARN("%s: failed to set LMA flags: rc = %d\n",
osd->od_svname, rc);
- RETURN(rc);
+ GOTO(out, rc);
}
}
}
- /* do both accounting updates outside oo_attr_lock below */
- if ((valid & LA_UID) && (la->la_uid != obj->oo_attr.la_uid)) {
- /* Update user accounting. Failure isn't fatal, but we still
- * log an error message */
- rc = -zap_increment_int(osd->od_os, osd->od_iusr_oid,
- la->la_uid, 1, oh->ot_tx);
- if (rc)
- CERROR("%s: failed to update accounting ZAP for user "
- "%d (%d)\n", osd->od_svname, la->la_uid, rc);
- rc = -zap_increment_int(osd->od_os, osd->od_iusr_oid,
- obj->oo_attr.la_uid, -1, oh->ot_tx);
- if (rc)
- CERROR("%s: failed to update accounting ZAP for user "
- "%d (%d)\n", osd->od_svname,
- obj->oo_attr.la_uid, rc);
- }
- if ((valid & LA_GID) && (la->la_gid != obj->oo_attr.la_gid)) {
- /* Update group accounting. Failure isn't fatal, but we still
- * log an error message */
- rc = -zap_increment_int(osd->od_os, osd->od_igrp_oid,
- la->la_gid, 1, oh->ot_tx);
- if (rc)
- CERROR("%s: failed to update accounting ZAP for user "
- "%d (%d)\n", osd->od_svname, la->la_gid, rc);
- rc = -zap_increment_int(osd->od_os, osd->od_igrp_oid,
- obj->oo_attr.la_gid, -1, oh->ot_tx);
- if (rc)
- CERROR("%s: failed to update accounting ZAP for user "
- "%d (%d)\n", osd->od_svname,
- obj->oo_attr.la_gid, rc);
- }
-
write_lock(&obj->oo_attr_lock);
cnt = 0;
+
+ if (valid & LA_PROJID) {
+#ifdef ZFS_PROJINHERIT
+ /* osd_declare_attr_set() must be called firstly.
+ * If osd::od_projectused_dn is not set, then we
+ * can not arrive at here. */
+ LASSERT(osd->od_projectused_dn);
+ LASSERT(obj->oo_with_projid);
+
+ osa->projid = obj->oo_attr.la_projid = la->la_projid;
+ SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_PROJID(osd), NULL,
+ &osa->projid, 8);
+#else
+ valid &= ~LA_PROJID;
+#endif
+ }
+
if (valid & LA_ATIME) {
osa->atime[0] = obj->oo_attr.la_atime = la->la_atime;
SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(osd), NULL,
/* many flags are not supported by zfs, so ensure a good cached
* copy */
obj->oo_attr.la_flags = attrs_zfs2fs(osa->flags);
+#ifdef ZFS_PROJINHERIT
+ if (obj->oo_with_projid)
+ osa->flags |= ZFS_PROJID;
+#endif
SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(osd), NULL,
&osa->flags, 8);
}
}
}
-static int osd_declare_object_create(const struct lu_env *env,
- struct dt_object *dt,
- struct lu_attr *attr,
- struct dt_allocation_hint *hint,
- struct dt_object_format *dof,
- struct thandle *handle)
+static int osd_declare_create(const struct lu_env *env, struct dt_object *dt,
+ struct lu_attr *attr,
+ struct dt_allocation_hint *hint,
+ struct dt_object_format *dof,
+ struct thandle *handle)
{
const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
struct osd_object *obj = osd_dt_obj(dt);
struct osd_device *osd = osd_obj2dev(obj);
struct osd_thandle *oh;
uint64_t zapid;
+ dnode_t *dn;
int rc, dnode_size;
ENTRY;
}
/* and we'll add it to some mapping */
- zapid = osd_get_name_n_idx(env, osd, fid, NULL, 0);
- dmu_tx_hold_zap(oh->ot_tx, zapid, TRUE, NULL);
-
- /* we will also update inode accounting ZAPs */
- dmu_tx_hold_zap(oh->ot_tx, osd->od_iusr_oid, FALSE, NULL);
- dmu_tx_hold_zap(oh->ot_tx, osd->od_igrp_oid, FALSE, NULL);
+ zapid = osd_get_name_n_idx(env, osd, fid, NULL, 0, &dn);
+ osd_tx_hold_zap(oh->ot_tx, zapid, dn, TRUE, NULL);
/* will help to find FID->ino mapping at dt_insert() */
osd_idc_find_and_init(env, osd, obj);
- rc = osd_declare_quota(env, osd, attr->la_uid, attr->la_gid, 1, oh,
- false, NULL, false);
+ rc = osd_declare_quota(env, osd, attr->la_uid, attr->la_gid,
+ attr->la_projid, 1, oh, NULL, OSD_QID_INODE);
RETURN(rc);
}
int __osd_attr_init(const struct lu_env *env, struct osd_device *osd,
- sa_handle_t *sa_hdl, dmu_tx_t *tx,
- struct lu_attr *la, uint64_t parent)
+ struct osd_object *obj, sa_handle_t *sa_hdl, dmu_tx_t *tx,
+ struct lu_attr *la, uint64_t parent,
+ nvlist_t *xattr)
{
sa_bulk_attr_t *bulk = osd_oti_get(env)->oti_attr_bulk;
struct osa_attr *osa = &osd_oti_get(env)->oti_osa;
timestruc_t now;
int cnt;
int rc;
+ char *dxattr = NULL;
+ size_t sa_size;
+
LASSERT(sa_hdl);
osa->gid = la->la_gid;
osa->rdev = la->la_rdev;
osa->nlink = la->la_nlink;
- osa->flags = attrs_fs2zfs(la->la_flags);
+ if (la->la_valid & LA_FLAGS)
+ osa->flags = attrs_fs2zfs(la->la_flags);
+ else
+ osa->flags = 0;
osa->size = la->la_size;
+#ifdef ZFS_PROJINHERIT
+ if (osd->od_projectused_dn) {
+ if (la->la_valid & LA_PROJID)
+ osa->projid = la->la_projid;
+ else
+ osa->projid = ZFS_DEFAULT_PROJID;
+ osa->flags |= ZFS_PROJID;
+ if (obj)
+ obj->oo_with_projid = 1;
+ } else {
+ osa->flags &= ~ZFS_PROJID;
+ }
+#endif
/*
* we need to create all SA below upon object create.
*
* XXX The attribute order matters since the accounting callback relies
* on static offsets (i.e. SA_*_OFFSET, see zfs_space_delta_cb()) to
- * look up the UID/GID attributes. Moreover, the callback does not seem
- * to support the spill block.
+ * look up the UID/GID/PROJID attributes. Moreover, the callback does
+ * not seem to support the spill block.
* We define attributes in the same order as SA_*_OFFSET in order to
* work around the problem. See ORI-610.
*/
SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(osd), NULL, osa->ctime, 16);
SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CRTIME(osd), NULL, crtime, 16);
SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_LINKS(osd), NULL, &osa->nlink, 8);
+#ifdef ZFS_PROJINHERIT
+ if (osd->od_projectused_dn)
+ SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_PROJID(osd), NULL,
+ &osa->projid, 8);
+#endif
SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_RDEV(osd), NULL, &osa->rdev, 8);
LASSERT(cnt <= ARRAY_SIZE(osd_oti_get(env)->oti_attr_bulk));
+ if (xattr) {
+ rc = -nvlist_size(xattr, &sa_size, NV_ENCODE_XDR);
+ LASSERT(rc == 0);
+
+ dxattr = osd_zio_buf_alloc(sa_size);
+ LASSERT(dxattr);
+
+ rc = -nvlist_pack(xattr, &dxattr, &sa_size,
+ NV_ENCODE_XDR, KM_SLEEP);
+ LASSERT(rc == 0);
+
+ SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_DXATTR(osd),
+ NULL, dxattr, sa_size);
+ }
+
rc = -sa_replace_all_by_template(sa_hdl, bulk, cnt, tx);
+ if (dxattr)
+ osd_zio_buf_free(dxattr, sa_size);
return rc;
}
atomic_inc_32(&dn->dn_dbufs_count);
}
*dnp = dn;
+ dbuf_read(db, NULL, DB_RF_MUST_SUCCEED | DB_RF_NOPREFETCH);
break;
}
if (unlikely(*dnp == NULL))
- rc = __osd_obj2dnode(env, tx->tx_objset, oid, dnp);
+ rc = __osd_obj2dnode(tx->tx_objset, oid, dnp);
return rc;
}
+#ifdef HAVE_DMU_OBJECT_ALLOC_DNSIZE
+static int osd_find_dnsize(struct osd_object *obj)
+{
+ struct osd_device *osd = osd_obj2dev(obj);
+ int dnsize;
+
+ if (osd->od_dnsize == ZFS_DNSIZE_AUTO) {
+ dnsize = DNODE_MIN_SIZE;
+ do {
+ if (DN_BONUS_SIZE(dnsize) >= obj->oo_ea_in_bonus + 32)
+ break;
+ dnsize <<= 1;
+ } while (dnsize < DNODE_MAX_SIZE);
+ if (dnsize > DNODE_MAX_SIZE)
+ dnsize = DNODE_MAX_SIZE;
+ } else if (osd->od_dnsize == ZFS_DNSIZE_1K) {
+ dnsize = 1024;
+ } else if (osd->od_dnsize == ZFS_DNSIZE_2K) {
+ dnsize = 2048;
+ } else if (osd->od_dnsize == ZFS_DNSIZE_4K) {
+ dnsize = 4096;
+ } else if (osd->od_dnsize == ZFS_DNSIZE_8K) {
+ dnsize = 8192;
+ } else if (osd->od_dnsize == ZFS_DNSIZE_16K) {
+ dnsize = 16384;
+ } else {
+ dnsize = DNODE_MIN_SIZE;
+ }
+ return dnsize;
+}
+#else
+static int inline osd_find_dnsize(struct osd_object *obj)
+{
+ return DN_MAX_BONUSLEN;
+}
+#endif
+
/*
* The transaction passed to this routine must have
* dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT) called and then assigned
type = DMU_OTN_UINT8_METADATA;
/* Create a new DMU object using the default dnode size. */
- oid = osd_dmu_object_alloc(osd->od_os, type, 0, 0, tx);
+ oid = osd_dmu_object_alloc(osd->od_os, type, 0,
+ osd_find_dnsize(obj), tx);
LASSERT(la->la_valid & LA_MODE);
la->la_size = 0;
* a conversion from the different internal ZAP hash formats being used. */
int __osd_zap_create(const struct lu_env *env, struct osd_device *osd,
dnode_t **dnp, dmu_tx_t *tx, struct lu_attr *la,
- zap_flags_t flags)
+ unsigned dnsize, zap_flags_t flags)
{
uint64_t oid;
DMU_OT_DIRECTORY_CONTENTS,
14, /* == ZFS fzap_default_blockshift */
DN_MAX_INDBLKSHIFT, /* indirect blockshift */
- 0, tx);
+ dnsize, tx);
la->la_size = 2;
la->la_nlink = 1;
* binary keys */
LASSERT(S_ISREG(la->la_mode));
rc = __osd_zap_create(env, osd_obj2dev(obj), &dn, oh->ot_tx, la,
- ZAP_FLAG_UINT64_KEY);
+ osd_find_dnsize(obj), ZAP_FLAG_UINT64_KEY);
if (rc)
return ERR_PTR(rc);
return dn;
int rc;
LASSERT(S_ISDIR(la->la_mode));
- rc = __osd_zap_create(env, osd_obj2dev(obj), &dn, oh->ot_tx, la, 0);
+ rc = __osd_zap_create(env, osd_obj2dev(obj), &dn, oh->ot_tx, la,
+ osd_find_dnsize(obj), 0);
if (rc)
return ERR_PTR(rc);
return dn;
/*
* Concurrency: @dt is write locked.
*/
-static int osd_object_create(const struct lu_env *env, struct dt_object *dt,
- struct lu_attr *attr,
- struct dt_allocation_hint *hint,
- struct dt_object_format *dof,
- struct thandle *th)
+static int osd_create(const struct lu_env *env, struct dt_object *dt,
+ struct lu_attr *attr, struct dt_allocation_hint *hint,
+ struct dt_object_format *dof, struct thandle *th)
{
struct osd_thread_info *info = osd_oti_get(env);
struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
struct osd_device *osd = osd_obj2dev(obj);
char *buf = info->oti_str;
struct osd_thandle *oh;
- dnode_t *dn = NULL;
+ dnode_t *dn = NULL, *zdn = NULL;
uint64_t zapid, parent = 0;
int rc;
ENTRY;
+ LASSERT(!fid_is_acct(fid));
+
/* concurrent create declarations should not see
* the object inconsistent (db, attr, etc).
* in regular cases acquisition should be cheap */
LASSERT(th != NULL);
oh = container_of0(th, struct osd_thandle, ot_super);
- /*
- * XXX missing: Quote handling.
- */
-
LASSERT(obj->oo_dn == NULL);
/* to follow ZFS on-disk format we need
obj->oo_attr = *attr;
obj->oo_attr.la_valid |= LA_SIZE | LA_NLINK | LA_TYPE;
+#ifdef ZFS_PROJINHERIT
+ if (osd->od_projectused_dn) {
+ if (!(obj->oo_attr.la_valid & LA_PROJID))
+ obj->oo_attr.la_projid = ZFS_DEFAULT_PROJID;
+ obj->oo_with_projid = 1;
+ }
+#endif
+
dn = osd_create_type_f(dof->dof_type)(env, obj, &obj->oo_attr, oh);
if (IS_ERR(dn)) {
rc = PTR_ERR(dn);
zde->zde_dnode = dn->dn_object;
zde->zde_type = IFTODT(attr->la_mode & S_IFMT);
- zapid = osd_get_name_n_idx(env, osd, fid, buf, sizeof(info->oti_str));
-
- rc = -zap_add(osd->od_os, zapid, buf, 8, 1, zde, oh->ot_tx);
+ zapid = osd_get_name_n_idx(env, osd, fid, buf,
+ sizeof(info->oti_str), &zdn);
+ rc = osd_zap_add(osd, zapid, zdn, buf, 8, 1, zde, oh->ot_tx);
if (rc)
GOTO(out, rc);
-
- /* Now add in all of the "SA" attributes */
- rc = -sa_handle_get(osd->od_os, dn->dn_object, NULL,
- SA_HDL_PRIVATE, &obj->oo_sa_hdl);
- if (rc)
- GOTO(out, rc);
-
- /* configure new osd object */
obj->oo_dn = dn;
- parent = parent != 0 ? parent : zapid;
- rc = __osd_attr_init(env, osd, obj->oo_sa_hdl, oh->ot_tx,
- &obj->oo_attr, parent);
+ /* Now add in all of the "SA" attributes */
+ rc = osd_sa_handle_get(obj);
if (rc)
GOTO(out, rc);
- /* XXX: oo_lma_flags */
- obj->oo_dt.do_lu.lo_header->loh_attr |= obj->oo_attr.la_mode & S_IFMT;
- smp_mb();
- obj->oo_dt.do_lu.lo_header->loh_attr |= LOHA_EXISTS;
- if (likely(!fid_is_acct(lu_object_fid(&obj->oo_dt.do_lu))))
- /* no body operations for accounting objects */
- obj->oo_dt.do_body_ops = &osd_body_ops;
-
rc = -nvlist_alloc(&obj->oo_sa_xattr, NV_UNIQUE_NAME, KM_SLEEP);
if (rc)
GOTO(out, rc);
/* initialize LMA */
- lustre_lma_init(lma, lu_object_fid(&obj->oo_dt.do_lu), 0, 0);
+ lustre_lma_init(lma, fid, 0, 0);
lustre_lma_swab(lma);
rc = -nvlist_add_byte_array(obj->oo_sa_xattr, XATTR_NAME_LMA,
(uchar_t *)lma, sizeof(*lma));
if (rc)
GOTO(out, rc);
- rc = __osd_sa_xattr_update(env, obj, oh);
+
+ /* configure new osd object */
+ obj->oo_parent = parent != 0 ? parent : zapid;
+ obj->oo_late_attr_set = 1;
+ rc = __osd_sa_xattr_schedule_update(env, obj, oh);
if (rc)
GOTO(out, rc);
- osd_idc_find_and_init(env, osd, obj);
- /* Add new object to inode accounting.
- * Errors are not considered as fatal */
- rc = -zap_increment_int(osd->od_os, osd->od_iusr_oid,
- (attr->la_valid & LA_UID) ? attr->la_uid : 0, 1,
- oh->ot_tx);
- if (rc)
- CERROR("%s: failed to add "DFID" to accounting ZAP for usr %d "
- "(%d)\n", osd->od_svname, PFID(fid), attr->la_uid, rc);
- rc = -zap_increment_int(osd->od_os, osd->od_igrp_oid,
- (attr->la_valid & LA_GID) ? attr->la_gid : 0, 1,
- oh->ot_tx);
- if (rc)
- CERROR("%s: failed to add "DFID" to accounting ZAP for grp %d "
- "(%d)\n", osd->od_svname, PFID(fid), attr->la_gid, rc);
+ /* XXX: oo_lma_flags */
+ obj->oo_dt.do_lu.lo_header->loh_attr |= obj->oo_attr.la_mode & S_IFMT;
+ if (likely(!fid_is_acct(lu_object_fid(&obj->oo_dt.do_lu))))
+ /* no body operations for accounting objects */
+ obj->oo_dt.do_body_ops = &osd_body_ops;
+
+ osd_idc_find_and_init(env, osd, obj);
out:
if (unlikely(rc && dn)) {
dmu_object_free(osd->od_os, dn->dn_object, oh->ot_tx);
osd_dnode_rele(dn);
obj->oo_dn = NULL;
+ } else if (!rc) {
+ obj->oo_dt.do_lu.lo_header->loh_attr |= LOHA_EXISTS;
}
up_write(&obj->oo_guard);
RETURN(rc);
}
-static int osd_declare_object_ref_add(const struct lu_env *env,
- struct dt_object *dt,
- struct thandle *th)
+static int osd_declare_ref_add(const struct lu_env *env, struct dt_object *dt,
+ struct thandle *th)
{
return osd_declare_attr_set(env, dt, NULL, th);
}
/*
* Concurrency: @dt is write locked.
*/
-static int osd_object_ref_add(const struct lu_env *env,
- struct dt_object *dt,
- struct thandle *handle)
+static int osd_ref_add(const struct lu_env *env, struct dt_object *dt,
+ struct thandle *handle)
{
struct osd_object *obj = osd_dt_obj(dt);
struct osd_thandle *oh;
RETURN(rc);
}
-static int osd_declare_object_ref_del(const struct lu_env *env,
- struct dt_object *dt,
- struct thandle *handle)
+static int osd_declare_ref_del(const struct lu_env *env, struct dt_object *dt,
+ struct thandle *handle)
{
return osd_declare_attr_set(env, dt, NULL, handle);
}
/*
* Concurrency: @dt is write locked.
*/
-static int osd_object_ref_del(const struct lu_env *env,
- struct dt_object *dt,
- struct thandle *handle)
+static int osd_ref_del(const struct lu_env *env, struct dt_object *dt,
+ struct thandle *handle)
{
struct osd_object *obj = osd_dt_obj(dt);
struct osd_thandle *oh;
}
static struct dt_object_operations osd_obj_ops = {
- .do_read_lock = osd_object_read_lock,
- .do_write_lock = osd_object_write_lock,
- .do_read_unlock = osd_object_read_unlock,
- .do_write_unlock = osd_object_write_unlock,
- .do_write_locked = osd_object_write_locked,
+ .do_read_lock = osd_read_lock,
+ .do_write_lock = osd_write_lock,
+ .do_read_unlock = osd_read_unlock,
+ .do_write_unlock = osd_write_unlock,
+ .do_write_locked = osd_write_locked,
.do_attr_get = osd_attr_get,
.do_declare_attr_set = osd_declare_attr_set,
.do_attr_set = osd_attr_set,
.do_ah_init = osd_ah_init,
- .do_declare_create = osd_declare_object_create,
- .do_create = osd_object_create,
- .do_declare_destroy = osd_declare_object_destroy,
- .do_destroy = osd_object_destroy,
+ .do_declare_create = osd_declare_create,
+ .do_create = osd_create,
+ .do_declare_destroy = osd_declare_destroy,
+ .do_destroy = osd_destroy,
.do_index_try = osd_index_try,
- .do_declare_ref_add = osd_declare_object_ref_add,
- .do_ref_add = osd_object_ref_add,
- .do_declare_ref_del = osd_declare_object_ref_del,
- .do_ref_del = osd_object_ref_del,
+ .do_declare_ref_add = osd_declare_ref_add,
+ .do_ref_add = osd_ref_add,
+ .do_declare_ref_del = osd_declare_ref_del,
+ .do_ref_del = osd_ref_del,
.do_xattr_get = osd_xattr_get,
.do_declare_xattr_set = osd_declare_xattr_set,
.do_xattr_set = osd_xattr_set,
}
static struct dt_object_operations osd_obj_otable_it_ops = {
- .do_attr_get = osd_otable_it_attr_get,
- .do_index_try = osd_index_try,
+ .do_attr_get = osd_otable_it_attr_get,
+ .do_index_try = osd_index_try,
};