* Copyright (c) 2009, 2010, Oracle and/or its affiliates. All rights reserved.
* Use is subject to license terms.
*
- * Copyright (c) 2012, 2016, Intel Corporation.
+ * Copyright (c) 2012, 2017, Intel Corporation.
*/
/*
* This file is part of Lustre, http://www.lustre.org/
#define DEBUG_SUBSYSTEM S_OSD
-#include <lustre_ver.h>
#include <libcfs/libcfs.h>
#include <obd_support.h>
#include <lustre_net.h>
#include <sys/txg.h>
char *osd_obj_tag = "osd_object";
+static int osd_object_sync_delay_us = -1;
static struct dt_object_operations osd_obj_ops;
static struct lu_object_operations osd_lu_obj_ops;
/*
* Retrieve the attributes of a DMU object
*/
-int __osd_object_attr_get(const struct lu_env *env, struct osd_device *o,
- struct osd_object *obj, struct lu_attr *la)
+static int __osd_object_attr_get(const struct lu_env *env, struct osd_device *o,
+ struct osd_object *obj, struct lu_attr *la)
{
struct osa_attr *osa = &osd_oti_get(env)->oti_osa;
sa_bulk_attr_t *bulk = osd_oti_get(env)->oti_attr_bulk;
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];
OBD_SLAB_ALLOC_PTR_GFP(mo, osd_object_kmem, GFP_NOFS);
if (mo != NULL) {
struct lu_object *l;
+ struct lu_object_header *h;
+ struct osd_device *o = osd_dev(d);
l = &mo->oo_dt.do_lu;
- dt_object_init(&mo->oo_dt, NULL, d);
+ if (unlikely(o->od_in_init)) {
+ OBD_ALLOC_PTR(h);
+ if (!h) {
+ OBD_FREE_PTR(mo);
+ return NULL;
+ }
+
+ lu_object_header_init(h);
+ lu_object_init(l, h, d);
+ lu_object_add_top(h, l);
+ mo->oo_header = h;
+ } else {
+ dt_object_init(&mo->oo_dt, NULL, d);
+ mo->oo_header = NULL;
+ }
+
mo->oo_dt.do_ops = &osd_obj_ops;
l->lo_ops = &osd_lu_obj_ops;
INIT_LIST_HEAD(&mo->oo_sa_linkage);
}
}
+static void osd_obj_set_blksize(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);
+ dmu_tx_t *tx;
+ dnode_t *dn = obj->oo_dn;
+ uint32_t blksz;
+ int rc = 0;
+ ENTRY;
+
+ LASSERT(!osd_oti_get(env)->oti_in_trans);
+
+ tx = dmu_tx_create(osd->od_os);
+ if (!tx) {
+ CERROR("%s: fail to create tx to set blksize for "DFID"\n",
+ osd->od_svname, PFID(fid));
+ RETURN_EXIT;
+ }
+
+ dmu_tx_hold_bonus(tx, dn->dn_object);
+ rc = -dmu_tx_assign(tx, TXG_WAIT);
+ if (rc) {
+ dmu_tx_abort(tx);
+ CERROR("%s: fail to assign tx to set blksize for "DFID
+ ": rc = %d\n", osd->od_svname, PFID(fid), rc);
+ RETURN_EXIT;
+ }
+
+ down_write(&obj->oo_guard);
+ if (unlikely((1 << dn->dn_datablkshift) >= PAGE_SIZE))
+ GOTO(out, rc = 1);
+
+ blksz = dn->dn_datablksz;
+ if (!is_power_of_2(blksz))
+ blksz = size_roundup_power2(blksz);
+
+ if (blksz > osd->od_max_blksz)
+ blksz = osd->od_max_blksz;
+ else if (blksz < PAGE_SIZE)
+ blksz = PAGE_SIZE;
+ rc = -dmu_object_set_blocksize(osd->od_os, dn->dn_object, blksz, 0, tx);
+
+ GOTO(out, rc);
+
+out:
+ up_write(&obj->oo_guard);
+ if (rc) {
+ dmu_tx_abort(tx);
+ if (unlikely(obj->oo_dn->dn_maxblkid > 0))
+ rc = 1;
+ if (rc < 0)
+ CERROR("%s: fail to set blksize for "DFID": rc = %d\n",
+ osd->od_svname, PFID(fid), rc);
+ } else {
+ dmu_tx_commit(tx);
+ CDEBUG(D_INODE, "%s: set blksize as %u for "DFID"\n",
+ osd->od_svname, blksz, PFID(fid));
+ }
+}
+
/*
* Concurrency: shouldn't matter.
*/
-int osd_object_init0(const struct lu_env *env, struct osd_object *obj)
+static int osd_object_init0(const struct lu_env *env, struct osd_object *obj)
{
struct osd_device *osd = osd_obj2dev(obj);
const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
int rc = 0;
ENTRY;
- if (obj->oo_dn == NULL)
- RETURN(0);
-
- /* object exist */
+ LASSERT(obj->oo_dn);
rc = osd_object_sa_init(obj, osd);
if (rc)
if (rc)
RETURN(rc);
- if (likely(!fid_is_acct(fid)))
+ if (likely(!fid_is_acct(fid))) {
/* no body operations for accounting objects */
obj->oo_dt.do_body_ops = &osd_body_ops;
+ if (S_ISREG(obj->oo_attr.la_mode) &&
+ obj->oo_dn->dn_maxblkid == 0 &&
+ (1 << obj->oo_dn->dn_datablkshift) < PAGE_SIZE &&
+ (fid_is_idif(fid) || fid_is_norm(fid) ||
+ fid_is_echo(fid)) &&
+ osd->od_is_ost && !osd->od_dt_dev.dd_rdonly)
+ osd_obj_set_blksize(env, osd, obj);
+ }
+
/*
* initialize object before marking it existing
*/
struct lu_buf buf;
int rc;
struct lustre_mdt_attrs *lma;
+ const struct lu_fid *rfid = lu_object_fid(&obj->oo_dt.do_lu);
ENTRY;
CLASSERT(sizeof(info->oti_buf) >= sizeof(*lma));
CWARN("%s: unsupported incompat LMA feature(s) %#x for "
"fid = "DFID"\n", osd_obj2dev(obj)->od_svname,
lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
- PFID(lu_object_fid(&obj->oo_dt.do_lu)));
+ PFID(rfid));
rc = -EOPNOTSUPP;
+ } else if (unlikely(!lu_fid_eq(rfid, &lma->lma_self_fid))) {
+ CERROR("%s: FID-in-LMA "DFID" does not match the "
+ "object self-fid "DFID"\n",
+ osd_obj2dev(obj)->od_svname,
+ PFID(&lma->lma_self_fid), PFID(rfid));
+ rc = -EREMCHG;
+ } else {
+ struct osd_device *osd = osd_obj2dev(obj);
+
+ if (lma->lma_compat & LMAC_STRIPE_INFO &&
+ osd->od_is_ost)
+ obj->oo_pfid_in_lma = 1;
+ if (unlikely(lma->lma_incompat & LMAI_REMOTE_PARENT) &&
+ osd->od_remote_parent_dir != ZFS_NO_OBJECT)
+ lu_object_set_agent_entry(&obj->oo_dt.do_lu);
}
} else if (rc == -ENODATA) {
/* haven't initialize LMA xattr */
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;
}
struct osd_object *obj = osd_obj(l);
struct osd_device *osd = osd_obj2dev(obj);
const struct lu_fid *fid = lu_object_fid(l);
+ struct lustre_scrub *scrub = &osd->od_scrub;
+ struct osd_thread_info *info = osd_oti_get(env);
+ struct luz_direntry *zde = &info->oti_zde;
+ struct osd_idmap_cache *idc;
+ char *name = info->oti_str;
uint64_t oid;
int rc = 0;
+ int rc1;
+ bool remote = false;
ENTRY;
LASSERT(osd_invariant(obj));
if (fid_is_otable_it(&l->lo_header->loh_fid)) {
obj->oo_dt.do_ops = &osd_obj_otable_it_ops;
l->lo_header->loh_attr |= LOHA_EXISTS;
- RETURN(0);
+
+ GOTO(out, rc = 0);
}
- if (conf != NULL && conf->loc_flags & LOC_F_NEW)
+ if (conf && conf->loc_flags & LOC_F_NEW)
GOTO(out, rc = 0);
if (unlikely(fid_is_acct(fid))) {
GOTO(out, rc = 0);
}
- rc = osd_fid_lookup(env, osd, fid, &oid);
- if (rc == 0) {
- LASSERT(obj->oo_dn == NULL);
- 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;
- GOTO(out, rc);
+ idc = osd_idc_find(env, osd, fid);
+ if (idc && !idc->oic_remote && idc->oic_dnode != ZFS_NO_OBJECT) {
+ oid = idc->oic_dnode;
+ goto zget;
+ }
+
+ rc = -ENOENT;
+ if (!list_empty(&osd->od_scrub.os_inconsistent_items))
+ rc = osd_oii_lookup(osd, fid, &oid);
+
+ if (rc)
+ rc = osd_fid_lookup(env, osd, fid, &oid);
+
+ if (rc == -ENOENT) {
+ if (likely(!(fid_is_norm(fid) || fid_is_igif(fid)) ||
+ fid_is_on_ost(env, osd, fid) ||
+ !zfs_test_bit(osd_oi_fid2idx(osd, fid),
+ scrub->os_file.sf_oi_bitmap)))
+ GOTO(out, rc = 0);
+
+ rc = -EREMCHG;
+ goto trigger;
+ }
+
+ if (rc)
+ GOTO(out, rc);
+
+zget:
+ LASSERT(obj->oo_dn == NULL);
+
+ rc = __osd_obj2dnode(osd->od_os, oid, &obj->oo_dn);
+ /* EEXIST will be returned if object is being deleted in ZFS */
+ if (rc == -EEXIST)
+ GOTO(out, rc = 0);
+
+ if (rc) {
+ CERROR("%s: lookup "DFID"/%#llx failed: rc = %d\n",
+ osd->od_svname, PFID(lu_object_fid(l)), oid, rc);
+ GOTO(out, rc);
+ }
+
+ rc = osd_object_init0(env, obj);
+ if (rc)
+ GOTO(out, rc);
+
+ if (unlikely(obj->oo_header))
+ GOTO(out, rc = 0);
+
+ rc = osd_check_lma(env, obj);
+ if ((!rc && !remote) || (rc != -EREMCHG))
+ GOTO(out, rc);
+
+trigger:
+ /* We still have chance to get the valid dnode: for the object that is
+ * referenced by remote name entry, the object on the local MDT will be
+ * linked under the dir /REMOTE_PARENT_DIR with its FID string as name.
+ *
+ * During the OI scrub, if we cannot find the OI mapping, we may still
+ * have change to map the FID to local OID via lookup the dir
+ * /REMOTE_PARENT_DIR. */
+ if (!remote && !fid_is_on_ost(env, osd, fid)) {
+ osd_fid2str(name, fid, sizeof(info->oti_str));
+ rc = osd_zap_lookup(osd, osd->od_remote_parent_dir,
+ NULL, name, 8, 3, (void *)zde);
+ if (!rc) {
+ oid = zde->lzd_reg.zde_dnode;
+ osd_dnode_rele(obj->oo_dn);
+ obj->oo_dn = NULL;
+ remote = true;
+ goto zget;
}
- if (rc != 0) {
- CERROR("%s: lookup "DFID"/%#llx failed: rc = %d\n",
- osd->od_svname, PFID(lu_object_fid(l)), oid, rc);
- GOTO(out, rc);
+ }
+
+ /* The case someone triggered the OI scrub already. */
+ if (thread_is_running(&scrub->os_thread)) {
+ if (!rc) {
+ LASSERT(remote);
+
+ lu_object_set_agent_entry(l);
+ osd_oii_insert(env, osd, fid, oid, false);
+ } else {
+ rc = -EINPROGRESS;
}
- LASSERT(obj->oo_dn);
- rc = osd_object_init0(env, obj);
- if (rc != 0)
- GOTO(out, rc);
- rc = osd_check_lma(env, obj);
- if (rc != 0)
- GOTO(out, rc);
- } else if (rc == -ENOENT) {
- rc = 0;
+ GOTO(out, rc);
}
- LASSERT(osd_invariant(obj));
+
+ /* The case NOT allow to trigger OI scrub automatically. */
+ if (osd->od_auto_scrub_interval == AS_NEVER)
+ GOTO(out, rc);
+
+ /* It is me to trigger the OI scrub. */
+ rc1 = osd_scrub_start(env, osd, SS_CLEAR_DRYRUN |
+ SS_CLEAR_FAILOUT | SS_AUTO_FULL);
+ LCONSOLE_WARN("%s: trigger OI scrub by RPC for the "DFID": rc = %d\n",
+ osd_name(osd), PFID(fid), rc1);
+ if (!rc) {
+ LASSERT(remote);
+
+ lu_object_set_agent_entry(l);
+ if (!rc1)
+ osd_oii_insert(env, osd, fid, oid, false);
+ } else {
+ if (!rc1)
+ rc = -EINPROGRESS;
+ else
+ rc = -EREMCHG;
+ }
+
+ GOTO(out, rc);
+
out:
RETURN(rc);
}
static void osd_object_free(const struct lu_env *env, struct lu_object *l)
{
struct osd_object *obj = osd_obj(l);
+ struct lu_object_header *h = obj->oo_header;
LASSERT(osd_invariant(obj));
dt_object_fini(&obj->oo_dt);
OBD_SLAB_FREE_PTR(obj, osd_object_kmem);
+ if (unlikely(h)) {
+ lu_object_header_fini(h);
+ OBD_FREE_PTR(h);
+ }
}
static int
/* 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);
osd_tx_hold_zap(oh->ot_tx, osd->od_unlinked->dn_object,
osd->od_unlinked, TRUE, NULL);
+ /* remove agent entry (if have) from remote parent */
+ if (lu_object_has_agent_entry(&obj->oo_dt.do_lu))
+ osd_tx_hold_zap(oh->ot_tx, osd->od_remote_parent_dir,
+ NULL, FALSE, NULL);
+
/* will help to find FID->ino when this object is being
* added to PENDING/ */
osd_idc_find_and_init(env, osd, obj);
/* remove obj ref from index dir (it depends) */
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);
- }
-
rc = osd_xattrs_destroy(env, obj, oh);
if (rc) {
CERROR("%s: cannot destroy xattrs for %s: rc = %d\n",
GOTO(out, rc);
}
+ if (lu_object_has_agent_entry(&obj->oo_dt.do_lu)) {
+ rc = osd_delete_from_remote_parent(env, osd, obj, oh, true);
+ if (rc)
+ GOTO(out, rc);
+ }
+
oid = obj->oo_dn->dn_object;
if (unlikely(obj->oo_destroy == OSD_DESTROY_NONE)) {
/* this may happen if the destroy wasn't declared
osd->od_svname, buf, oid, rc);
}
+ /* Remove the OI mapping after the destroy to handle the race with
+ * OI scrub that may insert missed OI mapping during the interval. */
+ rc = osd_zap_remove(osd, zapid, zdn, buf, oh->ot_tx);
+ if (unlikely(rc == -ENOENT))
+ rc = 0;
+ if (rc)
+ CERROR("%s: zap_remove(%s) failed: rc = %d\n",
+ osd->od_svname, buf, rc);
+
+ GOTO(out, rc);
+
out:
/* not needed in the cache anymore */
set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
struct qsd_instance *qsd,
struct lquota_trans *trans, int qtype,
__u64 orig_id, __u64 new_id, __u64 bspace,
- struct lquota_id_info *qi)
+ struct lquota_id_info *qi, bool ignore_edquot)
{
int rc;
qi->lqi_id.qid_uid = new_id;
qi->lqi_space = 1;
rc = qsd_op_begin(env, qsd, trans, qi, NULL);
- if (rc == -EDQUOT || rc == -EINPROGRESS)
+ if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
rc = 0;
if (rc)
return rc;
qi->lqi_id.qid_uid = new_id;
qi->lqi_space = bspace;
rc = qsd_op_begin(env, qsd, trans, qi, NULL);
- if (rc == -EDQUOT || rc == -EINPROGRESS)
+ if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
rc = 0;
if (rc)
return rc;
* 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);
+ bspace = toqb(bspace * 512);
+
+ CDEBUG(D_QUOTA, "%s: enforce quota on UID %u, GID %u,"
+ "the quota space is %lld (%u)\n", osd->od_svname,
+ attr->la_uid, attr->la_gid, bspace, blksize);
}
if (attr && attr->la_valid & LA_UID) {
rc = qsd_transfer(env, osd->od_quota_slave,
&oh->ot_quota_trans, USRQUOTA,
obj->oo_attr.la_uid, attr->la_uid,
- bspace, &info->oti_qi);
+ bspace, &info->oti_qi, true);
if (rc)
GOTO(out, rc);
}
rc = qsd_transfer(env, osd->od_quota_slave,
&oh->ot_quota_trans, GRPQUOTA,
obj->oo_attr.la_gid, attr->la_gid,
- bspace, &info->oti_qi);
+ bspace, &info->oti_qi,
+ !(attr->la_flags &
+ LUSTRE_SET_SYNC_FL));
if (rc)
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, true);
+ if (rc)
+ GOTO(out, rc);
+ }
+ }
+#endif
out:
up_read(&obj->oo_guard);
RETURN(rc);
transaction group. */
LASSERT(oh->ot_tx->tx_txg != 0);
+ if (OBD_FAIL_CHECK(OBD_FAIL_OSD_FID_MAPPING) && !osd->od_is_ost) {
+ struct zpl_direntry *zde = &info->oti_zde.lzd_reg;
+ char *buf = info->oti_str;
+ dnode_t *zdn = NULL;
+ uint64_t zapid;
+
+ zapid = osd_get_name_n_idx(env, osd, lu_object_fid(&dt->do_lu),
+ buf, sizeof(info->oti_str), &zdn);
+ rc = osd_zap_lookup(osd, zapid, zdn, buf, 8,
+ sizeof(*zde) / 8, zde);
+ if (!rc) {
+ zde->zde_dnode -= 1;
+ rc = -zap_update(osd->od_os, zapid, buf, 8,
+ sizeof(*zde) / 8, zde, oh->ot_tx);
+ }
+ if (rc > 0)
+ rc = 0;
+ GOTO(out, rc);
+ }
+
/* Only allow set size for regular file */
if (!S_ISREG(dt->do_lu.lo_header->loh_attr))
valid &= ~(LA_SIZE | LA_BLOCKS);
if (valid & LA_FLAGS) {
struct lustre_mdt_attrs *lma;
struct lu_buf buf;
+ int size = 0;
if (la->la_flags & LUSTRE_LMA_FL_MASKS) {
+ LASSERT(!obj->oo_pfid_in_lma);
CLASSERT(sizeof(info->oti_buf) >= sizeof(*lma));
lma = (struct lustre_mdt_attrs *)&info->oti_buf;
buf.lb_buf = lma;
buf.lb_len = sizeof(info->oti_buf);
- rc = osd_xattr_get(env, &obj->oo_dt, &buf,
- XATTR_NAME_LMA);
- if (rc > 0) {
+
+ /* Please do NOT call osd_xattr_get() directly, that
+ * will cause recursive down_read() on oo_guard. */
+ rc = osd_xattr_get_internal(env, obj, &buf,
+ XATTR_NAME_LMA, &size);
+ if (!rc && unlikely(size < sizeof(*lma))) {
+ rc = -EINVAL;
+ } else if (!rc) {
lma->lma_incompat =
le32_to_cpu(lma->lma_incompat);
lma->lma_incompat |=
if (rc < 0) {
CWARN("%s: failed to set LMA flags: rc = %d\n",
osd->od_svname, rc);
- RETURN(rc);
+ GOTO(out, 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);
}
LASSERT(oh->ot_tx != NULL);
/* this is the minimum set of EAs on every Lustre object */
- obj->oo_ea_in_bonus = ZFS_SA_BASE_ATTR_SIZE +
- sizeof(__u64) + /* VBR VERSION */
- sizeof(struct lustre_mdt_attrs); /* LMA */
+ obj->oo_ea_in_bonus = OSD_BASE_EA_IN_BONUS;
/* reserve 32 bytes for extra stuff like ACLs */
dnode_size = size_roundup_power2(obj->oo_ea_in_bonus + 32);
/* 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 osd_object *obj, sa_handle_t *sa_hdl, dmu_tx_t *tx,
struct lu_attr *la, uint64_t parent,
nvlist_t *xattr)
{
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));
return rc;
}
-static int osd_find_new_dnode(const struct lu_env *env, dmu_tx_t *tx,
- uint64_t oid, dnode_t **dnp)
+int osd_find_new_dnode(const struct lu_env *env, dmu_tx_t *tx,
+ uint64_t oid, dnode_t **dnp)
{
dmu_tx_hold_t *txh;
int rc = 0;
return rc;
}
+#ifdef HAVE_DMU_OBJECT_ALLOC_DNSIZE
+int osd_find_dnsize(struct osd_device *osd, int ea_in_bonus)
+{
+ int dnsize;
+
+ if (osd->od_dnsize == ZFS_DNSIZE_AUTO) {
+ dnsize = DNODE_MIN_SIZE;
+ do {
+ if (DN_BONUS_SIZE(dnsize) >= 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;
+}
+#endif
+
/*
* The transaction passed to this routine must have
* dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT) called and then assigned
* to a transaction group.
*/
-int __osd_object_create(const struct lu_env *env, struct osd_object *obj,
+int __osd_object_create(const struct lu_env *env, struct osd_device *osd,
+ struct osd_object *obj, const struct lu_fid *fid,
dnode_t **dnp, dmu_tx_t *tx, struct lu_attr *la)
{
- struct osd_device *osd = osd_obj2dev(obj);
- const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
- dmu_object_type_t type = DMU_OT_PLAIN_FILE_CONTENTS;
+ dmu_object_type_t type = DMU_OT_PLAIN_FILE_CONTENTS;
uint64_t oid;
+ int size;
/* Use DMU_OTN_UINT8_METADATA for local objects so their data blocks
* would get an additional ditto copy */
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);
+ if (obj)
+ size = obj->oo_ea_in_bonus;
+ else
+ size = OSD_BASE_EA_IN_BONUS;
+ oid = osd_dmu_object_alloc(osd->od_os, type, 0,
+ osd_find_dnsize(osd, size), 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;
static dnode_t *osd_mkidx(const struct lu_env *env, struct osd_object *obj,
struct lu_attr *la, struct osd_thandle *oh)
{
+ struct osd_device *osd = osd_obj2dev(obj);
dnode_t *dn;
int rc;
* We set ZAP_FLAG_UINT64_KEY to let ZFS know than we are going to use
* 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);
+ rc = __osd_zap_create(env, osd, &dn, oh->ot_tx, la,
+ osd_find_dnsize(osd, obj->oo_ea_in_bonus), ZAP_FLAG_UINT64_KEY);
if (rc)
return ERR_PTR(rc);
return dn;
static dnode_t *osd_mkdir(const struct lu_env *env, struct osd_object *obj,
struct lu_attr *la, struct osd_thandle *oh)
{
+ struct osd_device *osd = osd_obj2dev(obj);
dnode_t *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, &dn, oh->ot_tx, la,
+ osd_find_dnsize(osd, obj->oo_ea_in_bonus), 0);
if (rc)
return ERR_PTR(rc);
return dn;
int rc;
LASSERT(S_ISREG(la->la_mode));
- rc = __osd_object_create(env, obj, &dn, oh->ot_tx, la);
+ rc = __osd_object_create(env, osd, obj, fid, &dn, oh->ot_tx, la);
if (rc)
return ERR_PTR(rc);
- if ((fid_is_idif(fid) || fid_is_norm(fid) || fid_is_echo(fid)) &&
- osd->od_is_ost) {
+ if ((fid_is_idif(fid) || fid_is_norm(fid) || fid_is_echo(fid))) {
/* The minimum block size must be at least page size otherwise
* it will break the assumption in tgt_thread_big_cache where
* the array size is PTLRPC_MAX_BRW_PAGES. It will also affect
int rc;
LASSERT(S_ISLNK(la->la_mode));
- rc = __osd_object_create(env, obj, &dn, oh->ot_tx, la);
+ rc = __osd_object_create(env, osd_obj2dev(obj), obj,
+ lu_object_fid(&obj->oo_dt.do_lu),
+ &dn, oh->ot_tx, la);
if (rc)
return ERR_PTR(rc);
return dn;
if (S_ISCHR(la->la_mode) || S_ISBLK(la->la_mode))
la->la_valid |= LA_RDEV;
- rc = __osd_object_create(env, obj, &dn, oh->ot_tx, la);
+ rc = __osd_object_create(env, osd_obj2dev(obj), obj,
+ lu_object_fid(&obj->oo_dt.do_lu),
+ &dn, oh->ot_tx, la);
if (rc)
return ERR_PTR(rc);
return dn;
dnode_t *dn = NULL, *zdn = NULL;
uint64_t zapid, parent = 0;
int rc;
+ __u32 compat = 0;
ENTRY;
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);
zapid = osd_get_name_n_idx(env, osd, fid, buf,
sizeof(info->oti_str), &zdn);
+ if (CFS_FAIL_CHECK(OBD_FAIL_OSD_NO_OI_ENTRY) ||
+ (osd->od_is_ost && OBD_FAIL_CHECK(OBD_FAIL_OSD_COMPAT_NO_ENTRY)))
+ goto skip_add;
+
+ if (osd->od_is_ost && OBD_FAIL_CHECK(OBD_FAIL_OSD_COMPAT_INVALID_ENTRY))
+ zde->zde_dnode++;
+
rc = osd_zap_add(osd, zapid, zdn, buf, 8, 1, zde, oh->ot_tx);
if (rc)
GOTO(out, rc);
+
+skip_add:
obj->oo_dn = dn;
/* Now add in all of the "SA" attributes */
rc = osd_sa_handle_get(obj);
GOTO(out, rc);
/* initialize LMA */
- lustre_lma_init(lma, fid, 0, 0);
+ if (fid_is_idif(fid) || (fid_is_norm(fid) && osd->od_is_ost))
+ compat |= LMAC_FID_ON_OST;
+ lustre_lma_init(lma, fid, compat, 0);
lustre_lma_swab(lma);
rc = -nvlist_add_byte_array(obj->oo_sa_xattr, XATTR_NAME_LMA,
(uchar_t *)lma, sizeof(*lma));
static int osd_declare_ref_add(const struct lu_env *env, struct dt_object *dt,
struct thandle *th)
{
+ osd_idc_find_and_init(env, osd_dev(dt->do_lu.lo_dev), osd_dt_obj(dt));
return osd_declare_attr_set(env, dt, NULL, th);
}
static int osd_declare_ref_del(const struct lu_env *env, struct dt_object *dt,
struct thandle *handle)
{
+ osd_idc_find_and_init(env, osd_dev(dt->do_lu.lo_dev), osd_dt_obj(dt));
return osd_declare_attr_set(env, dt, NULL, handle);
}
* support ZIL. If the object tracked the txg that it was last
* modified in, it could pass that txg here instead of "0". Maybe
* the changes are already committed, so no wait is needed at all? */
- if (!osd->od_dt_dev.dd_rdonly)
- txg_wait_synced(dmu_objset_pool(osd->od_os), 0ULL);
+ if (!osd->od_dt_dev.dd_rdonly) {
+ if (osd_object_sync_delay_us < 0)
+ txg_wait_synced(dmu_objset_pool(osd->od_os), 0ULL);
+ else
+ udelay(osd_object_sync_delay_us);
+ }
RETURN(0);
}
.do_attr_get = osd_otable_it_attr_get,
.do_index_try = osd_index_try,
};
+
+module_param(osd_object_sync_delay_us, int, 0644);
+MODULE_PARM_DESC(osd_object_sync_delay_us,
+ "If zero or larger delay N usec instead of doing object sync");
git://git.whamcloud.com / fs / lustre-release.git / blobdiff