*
* 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, 2013, Intel Corporation.
- * Use is subject to license terms.
+ *
+ * Copyright (c) 2012, 2017, Intel Corporation.
*/
/*
* This file is part of Lustre, http://www.lustre.org/
* Author: Johann Lombardi <johann@whamcloud.com>
*/
-#ifndef EXPORT_SYMTAB
-# define EXPORT_SYMTAB
-#endif
#define DEBUG_SUBSYSTEM S_OSD
-#include <lustre_ver.h>
#include <libcfs/libcfs.h>
-#include <lustre_fsfilt.h>
#include <obd_support.h>
#include <lustre_net.h>
#include <obd.h>
#include <sys/sa_impl.h>
#include <sys/txg.h>
-static char *osd_obj_tag = "osd_object";
+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;
extern struct dt_body_operations osd_body_ops;
static struct dt_object_operations osd_obj_otable_it_ops;
-extern cfs_mem_cache_t *osd_object_kmem;
+extern struct kmem_cache *osd_object_kmem;
static void
osd_object_sa_fini(struct osd_object *obj)
}
static int
-osd_object_sa_init(struct osd_object *obj, udmu_objset_t *uos)
+osd_object_sa_init(struct osd_object *obj, struct osd_device *o)
{
int rc;
LASSERT(obj->oo_sa_hdl == NULL);
- LASSERT(obj->oo_db != NULL);
+ LASSERT(obj->oo_dn != NULL);
- rc = -sa_handle_get(uos->os, obj->oo_db->db_object, obj,
- SA_HDL_PRIVATE, &obj->oo_sa_hdl);
+ rc = osd_sa_handle_get(obj);
if (rc)
return rc;
/* Cache the xattr object id, valid for the life of the object */
- rc = -sa_lookup(obj->oo_sa_hdl, SA_ZPL_XATTR(uos), &obj->oo_xattr, 8);
+ rc = -sa_lookup(obj->oo_sa_hdl, SA_ZPL_XATTR(o), &obj->oo_xattr, 8);
if (rc == -ENOENT) {
obj->oo_xattr = ZFS_NO_OBJECT;
rc = 0;
/*
* 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 (!cfs_list_empty(&obj->oo_sa_linkage))
+ if (!list_empty(&obj->oo_sa_linkage))
return;
- down(&oh->ot_sa_lock);
write_lock(&obj->oo_attr_lock);
- if (likely(cfs_list_empty(&obj->oo_sa_linkage)))
- cfs_list_add(&obj->oo_sa_linkage, &oh->ot_sa_list);
+ 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 (!cfs_list_empty(&oh->ot_sa_list)) {
- obj = cfs_list_entry(oh->ot_sa_list.next,
- struct osd_object, oo_sa_linkage);
- sa_spill_rele(obj->oo_sa_hdl);
+ while (!list_empty(&oh->ot_sa_list)) {
+ obj = list_entry(oh->ot_sa_list.next,
+ struct osd_object, oo_sa_linkage);
write_lock(&obj->oo_attr_lock);
- cfs_list_del_init(&obj->oo_sa_linkage);
+ 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);
}
/*
/*
* Retrieve the attributes of a DMU object
*/
-int __osd_object_attr_get(const struct lu_env *env, udmu_objset_t *uos,
+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_handle_t *sa_hdl;
- sa_bulk_attr_t *bulk;
+ sa_bulk_attr_t *bulk = osd_oti_get(env)->oti_attr_bulk;
int cnt = 0;
int rc;
ENTRY;
- LASSERT(obj->oo_db != NULL);
-
- rc = -sa_handle_get(uos->os, obj->oo_db->db_object, NULL,
- SA_HDL_PRIVATE, &sa_hdl);
- if (rc)
- RETURN(rc);
-
- OBD_ALLOC(bulk, sizeof(sa_bulk_attr_t) * 9);
- if (bulk == NULL)
- GOTO(out_sa, rc = -ENOMEM);
+ LASSERT(obj->oo_dn != NULL);
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_ATIME(uos), NULL, osa->atime, 16);
- SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(uos), NULL, osa->mtime, 16);
- SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(uos), NULL, osa->ctime, 16);
- SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(uos), NULL, &osa->mode, 8);
- SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_SIZE(uos), NULL, &osa->size, 8);
- SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_LINKS(uos), NULL, &osa->nlink, 8);
- SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_UID(uos), NULL, &osa->uid, 8);
- SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GID(uos), NULL, &osa->gid, 8);
- SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(uos), NULL, &osa->flags, 8);
-
- rc = -sa_bulk_lookup(sa_hdl, bulk, cnt);
+ SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(o), NULL, osa->atime, 16);
+ SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(o), NULL, osa->mtime, 16);
+ SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(o), NULL, osa->ctime, 16);
+ SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(o), NULL, &osa->mode, 8);
+ SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_SIZE(o), NULL, &osa->size, 8);
+ SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_LINKS(o), NULL, &osa->nlink, 8);
+ SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_UID(o), NULL, &osa->uid, 8);
+ SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GID(o), NULL, &osa->gid, 8);
+ 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(obj->oo_sa_hdl, bulk, cnt);
if (rc)
- GOTO(out_bulk, 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_flags = attrs_zfs2fs(osa->flags);
la->la_size = osa->size;
+ /* Try to get extra flag from LMA. Right now, only LMAI_ORPHAN
+ * flags is stored in LMA, and it is only for orphan directory */
+ if (S_ISDIR(la->la_mode) && dt_object_exists(&obj->oo_dt)) {
+ struct osd_thread_info *info = osd_oti_get(env);
+ struct lustre_mdt_attrs *lma;
+ struct lu_buf buf;
+
+ 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) {
+ rc = 0;
+ lma->lma_incompat = le32_to_cpu(lma->lma_incompat);
+ obj->oo_lma_flags =
+ lma_to_lustre_flags(lma->lma_incompat);
+
+ } else if (rc == -ENODATA) {
+ rc = 0;
+ }
+ }
+
if (S_ISCHR(la->la_mode) || S_ISBLK(la->la_mode)) {
- rc = -sa_lookup(sa_hdl, SA_ZPL_RDEV(uos), &osa->rdev, 8);
+ rc = -sa_lookup(obj->oo_sa_hdl, SA_ZPL_RDEV(o), &osa->rdev, 8);
if (rc)
- GOTO(out_bulk, rc);
+ GOTO(out_sa, rc);
la->la_rdev = osa->rdev;
la->la_valid |= LA_RDEV;
}
-out_bulk:
- OBD_FREE(bulk, sizeof(sa_bulk_attr_t) * 9);
out_sa:
- sa_handle_destroy(sa_hdl);
RETURN(rc);
}
-int __osd_obj2dbuf(const struct lu_env *env, objset_t *os,
- uint64_t oid, dmu_buf_t **dbp, void *tag)
+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;
- LASSERT(tag);
-
- rc = -sa_buf_hold(os, oid, tag, dbp);
+ rc = -dmu_bonus_hold(os, oid, osd_obj_tag, &db);
if (rc)
return rc;
- dmu_object_info_from_db(*dbp, doi);
- if (unlikely (oid != DMU_USERUSED_OBJECT &&
- oid != DMU_GROUPUSED_OBJECT && doi->doi_bonus_type != DMU_OT_SA)) {
- sa_buf_rele(*dbp, tag);
- *dbp = NULL;
- return -EINVAL;
- }
-
- LASSERT(*dbp);
- LASSERT((*dbp)->db_object == oid);
- LASSERT((*dbp)->db_offset == -1);
- LASSERT((*dbp)->db_data != NULL);
+ dbi = (dmu_buf_impl_t *)db;
+ DB_DNODE_ENTER(dbi);
+ *dnp = DB_DNODE(dbi);
+ LASSERT(*dnp != NULL);
return 0;
}
{
struct osd_object *mo;
- OBD_SLAB_ALLOC_PTR_GFP(mo, osd_object_kmem, CFS_ALLOC_IO);
+ 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;
- CFS_INIT_LIST_HEAD(&mo->oo_sa_linkage);
+ INIT_LIST_HEAD(&mo->oo_sa_linkage);
+ INIT_LIST_HEAD(&mo->oo_unlinked_linkage);
init_rwsem(&mo->oo_sem);
- sema_init(&mo->oo_guard, 1);
+ init_rwsem(&mo->oo_guard);
rwlock_init(&mo->oo_attr_lock);
+ mo->oo_destroy = OSD_DESTROY_NONE;
return l;
} else {
return NULL;
}
}
+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)
{
struct osd_device *osd = osd_obj2dev(obj);
- udmu_objset_t *uos = &osd->od_objset;
- const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
+ const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
int rc = 0;
ENTRY;
- if (obj->oo_db == NULL)
- RETURN(0);
+ LASSERT(obj->oo_dn);
- /* object exist */
-
- rc = osd_object_sa_init(obj, uos);
+ rc = osd_object_sa_init(obj, osd);
if (rc)
RETURN(rc);
/* cache attrs in object */
- rc = __osd_object_attr_get(env, &osd->od_objset,
- obj, &obj->oo_attr);
+ rc = __osd_object_attr_get(env, osd, obj, &obj->oo_attr);
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
*/
obj->oo_dt.do_lu.lo_header->loh_attr |= obj->oo_attr.la_mode & S_IFMT;
- cfs_mb();
+ smp_mb();
obj->oo_dt.do_lu.lo_header->loh_attr |= LOHA_EXISTS;
RETURN(0);
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));
buf.lb_buf = lma;
buf.lb_len = sizeof(info->oti_buf);
- rc = osd_xattr_get(env, &obj->oo_dt, &buf, XATTR_NAME_LMA, BYPASS_CAPA);
+ rc = osd_xattr_get(env, &obj->oo_dt, &buf, XATTR_NAME_LMA);
if (rc > 0) {
rc = 0;
lustre_lma_swab(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 */
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);
+ 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);
}
- rc = osd_fid_lookup(env, osd, lu_object_fid(l), &oid);
- if (rc == 0) {
- LASSERT(obj->oo_db == NULL);
- rc = __osd_obj2dbuf(env, osd->od_objset.os, oid,
- &obj->oo_db, osd_obj_tag);
- if (rc != 0) {
- CERROR("%s: lookup "DFID"/"LPX64" failed: rc = %d\n",
- osd->od_svname, PFID(lu_object_fid(l)), oid, rc);
- GOTO(out, rc);
+ if (conf && conf->loc_flags & LOC_F_NEW)
+ GOTO(out, rc = 0);
+
+ 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;
}
- LASSERT(obj->oo_db);
- 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 = 0);
}
- LASSERT(osd_invariant(obj));
+
+ 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;
+ }
+ }
+
+ /* 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;
+ }
+
+ GOTO(out, rc);
+ }
+
+ /* 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 void __osd_declare_object_destroy(const struct lu_env *env,
- struct osd_object *obj,
- struct osd_thandle *oh)
+static int
+osd_object_unlinked_add(struct osd_object *obj, struct osd_thandle *oh)
{
- struct osd_device *osd = osd_obj2dev(obj);
- udmu_objset_t *uos = &osd->od_objset;
- dmu_buf_t *db = obj->oo_db;
- zap_attribute_t *za = &osd_oti_get(env)->oti_za;
- uint64_t oid = db->db_object, xid;
- dmu_tx_t *tx = oh->ot_tx;
- zap_cursor_t *zc;
- int rc = 0;
-
- dmu_tx_hold_free(tx, oid, 0, DMU_OBJECT_END);
+ int rc = -EBUSY;
- /* zap holding xattrs */
- if (obj->oo_xattr != ZFS_NO_OBJECT) {
- oid = obj->oo_xattr;
+ LASSERT(obj->oo_destroy == OSD_DESTROY_ASYNC);
- dmu_tx_hold_free(tx, oid, 0, DMU_OBJECT_END);
-
- rc = -udmu_zap_cursor_init(&zc, uos, oid, 0);
- if (rc)
- goto out;
+ /* the object is supposed to be exclusively locked by
+ * 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);
+ rc = 0;
+ }
- while ((rc = -zap_cursor_retrieve(zc, za)) == 0) {
- BUG_ON(za->za_integer_length != sizeof(uint64_t));
- BUG_ON(za->za_num_integers != 1);
+ return rc;
+}
- rc = -zap_lookup(uos->os, obj->oo_xattr, za->za_name,
- sizeof(uint64_t), 1, &xid);
- if (rc) {
- CERROR("%s: xattr lookup failed: rc = %d\n",
- osd->od_svname, rc);
- goto out_err;
- }
- dmu_tx_hold_free(tx, xid, 0, DMU_OBJECT_END);
+/* Default to max data size covered by a level-1 indirect block */
+static unsigned long osd_sync_destroy_max_size =
+ 1UL << (DN_MAX_INDBLKSHIFT - SPA_BLKPTRSHIFT + SPA_MAXBLOCKSHIFT);
+module_param(osd_sync_destroy_max_size, ulong, 0444);
+MODULE_PARM_DESC(osd_sync_destroy_max_size, "Maximum object size to use synchronous destroy.");
- zap_cursor_advance(zc);
- }
- if (rc == -ENOENT)
- rc = 0;
-out_err:
- udmu_zap_cursor_fini(zc);
+static inline void
+osd_object_set_destroy_type(struct osd_object *obj)
+{
+ /*
+ * Lock-less OST_WRITE can race with OST_DESTROY, so set destroy type
+ * only once and use it consistently thereafter.
+ */
+ down_write(&obj->oo_guard);
+ if (obj->oo_destroy == OSD_DESTROY_NONE) {
+ if (obj->oo_attr.la_size <= osd_sync_destroy_max_size)
+ obj->oo_destroy = OSD_DESTROY_SYNC;
+ else /* Larger objects are destroyed asynchronously */
+ obj->oo_destroy = OSD_DESTROY_ASYNC;
}
-out:
- if (rc && tx->tx_err == 0)
- tx->tx_err = -rc;
+ 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)
{
- char *buf = osd_oti_get(env)->oti_str;
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;
+ uint64_t zapid;
ENTRY;
LASSERT(th != NULL);
oh = container_of0(th, struct osd_thandle, ot_super);
LASSERT(oh->ot_tx != NULL);
- /* declare that we'll destroy the object */
- __osd_declare_object_destroy(env, obj, oh);
+ dmu_tx_mark_netfree(oh->ot_tx);
/* declare that we'll remove object from fid-dnode mapping */
- zapid = osd_get_name_n_idx(env, osd, fid, buf);
- dmu_tx_hold_bonus(oh->ot_tx, zapid);
- dmu_tx_hold_zap(oh->ot_tx, zapid, 0, buf);
+ zapid = osd_get_name_n_idx(env, osd, fid, NULL, 0, &dn);
+ osd_tx_hold_zap(oh->ot_tx, zapid, dn, FALSE, NULL);
- /* declare that we'll remove object from inode accounting ZAPs */
- dmu_tx_hold_bonus(oh->ot_tx, osd->od_iusr_oid);
- dmu_tx_hold_zap(oh->ot_tx, osd->od_iusr_oid, 0, buf);
- dmu_tx_hold_bonus(oh->ot_tx, osd->od_igrp_oid);
- dmu_tx_hold_zap(oh->ot_tx, osd->od_igrp_oid, 0, buf);
+ osd_declare_xattrs_destroy(env, obj, oh);
/* 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);
- RETURN(rc);
-}
-
-int __osd_object_free(udmu_objset_t *uos, uint64_t oid, dmu_tx_t *tx)
-{
- LASSERT(uos->objects != 0);
- spin_lock(&uos->lock);
- uos->objects--;
- spin_unlock(&uos->lock);
-
- return -dmu_object_free(uos->os, oid, tx);
-}
+ obj->oo_attr.la_gid, obj->oo_attr.la_projid,
+ 0, oh, NULL, OSD_QID_BLK);
+ if (rc)
+ RETURN(rc);
-/*
- * Delete a DMU object
- *
- * The transaction passed to this routine must have
- * dmu_tx_hold_free(tx, oid, 0, DMU_OBJECT_END) called
- * and then assigned to a transaction group.
- *
- * This will release db and set it to NULL to prevent further dbuf releases.
- */
-static int __osd_object_destroy(const struct lu_env *env,
- struct osd_object *obj,
- dmu_tx_t *tx, void *tag)
-{
- struct osd_device *osd = osd_obj2dev(obj);
- udmu_objset_t *uos = &osd->od_objset;
- uint64_t xid;
- zap_attribute_t *za = &osd_oti_get(env)->oti_za;
- zap_cursor_t *zc;
- int rc;
+ osd_object_set_destroy_type(obj);
+ if (obj->oo_destroy == OSD_DESTROY_SYNC)
+ dmu_tx_hold_free(oh->ot_tx, obj->oo_dn->dn_object,
+ 0, DMU_OBJECT_END);
+ else
+ osd_tx_hold_zap(oh->ot_tx, osd->od_unlinked->dn_object,
+ osd->od_unlinked, TRUE, NULL);
- /* Assert that the transaction has been assigned to a
- transaction group. */
- LASSERT(tx->tx_txg != 0);
+ /* 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);
- /* zap holding xattrs */
- if (obj->oo_xattr != ZFS_NO_OBJECT) {
- rc = -udmu_zap_cursor_init(&zc, uos, obj->oo_xattr, 0);
- if (rc)
- return rc;
- while ((rc = -zap_cursor_retrieve(zc, za)) == 0) {
- BUG_ON(za->za_integer_length != sizeof(uint64_t));
- BUG_ON(za->za_num_integers != 1);
-
- rc = -zap_lookup(uos->os, obj->oo_xattr, za->za_name,
- sizeof(uint64_t), 1, &xid);
- if (rc) {
- CERROR("%s: lookup xattr %s failed: rc = %d\n",
- osd->od_svname, za->za_name, rc);
- continue;
- }
- rc = __osd_object_free(uos, xid, tx);
- if (rc)
- CERROR("%s: fetch xattr %s failed: rc = %d\n",
- osd->od_svname, za->za_name, rc);
- zap_cursor_advance(zc);
- }
- udmu_zap_cursor_fini(zc);
+ /* will help to find FID->ino when this object is being
+ * added to PENDING/ */
+ osd_idc_find_and_init(env, osd, obj);
- rc = __osd_object_free(uos, obj->oo_xattr, tx);
- if (rc)
- CERROR("%s: freeing xattr failed: rc = %d\n",
- osd->od_svname, rc);
- }
-
- return __osd_object_free(uos, obj->oo_db->db_object, tx);
+ 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)
{
- char *buf = osd_oti_get(env)->oti_str;
+ struct osd_thread_info *info = osd_oti_get(env);
+ char *buf = info->oti_str;
struct osd_object *obj = osd_dt_obj(dt);
struct osd_device *osd = osd_obj2dev(obj);
const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
struct osd_thandle *oh;
int rc;
- uint64_t zapid;
+ uint64_t oid, zapid;
+ dnode_t *zdn;
ENTRY;
- LASSERT(obj->oo_db != NULL);
- LASSERT(dt_object_exists(dt));
- LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
+ down_write(&obj->oo_guard);
+
+ if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
+ GOTO(out, rc = -ENOENT);
+
+ LASSERT(obj->oo_dn != NULL);
oh = container_of0(th, struct osd_thandle, ot_super);
LASSERT(oh != NULL);
LASSERT(oh->ot_tx != NULL);
- zapid = osd_get_name_n_idx(env, osd, fid, buf);
-
/* remove obj ref from index dir (it depends) */
- rc = -zap_remove(osd->od_objset.os, zapid, buf, oh->ot_tx);
+ zapid = osd_get_name_n_idx(env, osd, fid, buf,
+ sizeof(info->oti_str), &zdn);
+ rc = osd_xattrs_destroy(env, obj, oh);
if (rc) {
- CERROR("%s: zap_remove() failed: rc = %d\n",
- osd->od_svname, rc);
+ CERROR("%s: cannot destroy xattrs for %s: 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_objset.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_objset.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);
+ 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);
+ }
- /* kill object */
- rc = __osd_object_destroy(env, obj, oh->ot_tx, osd_obj_tag);
- if (rc) {
- CERROR("%s: __osd_object_destroy() failed: rc = %d\n",
- osd->od_svname, 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
+ * e.g. when the object is created and then destroyed
+ * in the same transaction - we don't need additional
+ * space for destroy specifically */
+ LASSERT(obj->oo_attr.la_size <= osd_sync_destroy_max_size);
+ rc = -dmu_object_free(osd->od_os, oid, oh->ot_tx);
+ if (rc)
+ CERROR("%s: failed to free %s %llu: rc = %d\n",
+ osd->od_svname, buf, oid, rc);
+ } else if (obj->oo_destroy == OSD_DESTROY_SYNC) {
+ rc = -dmu_object_free(osd->od_os, oid, oh->ot_tx);
+ if (rc)
+ 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);
+
+ 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);
}
+ /* 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);
-
+ if (rc == 0)
+ obj->oo_destroyed = 1;
+ up_write(&obj->oo_guard);
RETURN (0);
}
static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
{
struct osd_object *obj = osd_obj(l);
-
- if (obj->oo_db != 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);
}
- sa_buf_rele(obj->oo_db, osd_obj_tag);
- cfs_list_del(&obj->oo_sa_linkage);
- obj->oo_db = NULL;
+ 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);
LASSERT(osd_invariant(obj));
- down_read(&obj->oo_sem);
+ 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);
LASSERT(osd_invariant(obj));
- down_write(&obj->oo_sem);
+ 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;
static int osd_attr_get(const struct lu_env *env,
struct dt_object *dt,
- struct lu_attr *attr,
- struct lustre_capa *capa)
+ struct lu_attr *attr)
{
struct osd_object *obj = osd_dt_obj(dt);
uint64_t blocks;
uint32_t blksize;
+ int rc = 0;
+
+ down_read(&obj->oo_guard);
+
+ 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(dt_object_exists(dt));
LASSERT(osd_invariant(obj));
- LASSERT(obj->oo_db);
+ LASSERT(obj->oo_dn);
read_lock(&obj->oo_attr_lock);
*attr = obj->oo_attr;
+ if (obj->oo_lma_flags & LUSTRE_ORPHAN_FL)
+ attr->la_flags |= LUSTRE_ORPHAN_FL;
read_unlock(&obj->oo_attr_lock);
/* with ZFS_DEBUG zrl_add_debug() called by DB_DNODE_ENTER()
attr->la_size = 512 * blocks;
/* Block size may be not set; suggest maximal I/O transfers. */
if (blksize == 0)
- blksize = 1ULL << SPA_MAXBLOCKSHIFT;
+ blksize = osd_spa_maxblocksize(
+ dmu_objset_spa(osd_obj2dev(obj)->od_os));
attr->la_blksize = blksize;
attr->la_blocks = blocks;
attr->la_valid |= LA_BLOCKS | LA_BLKSIZE;
- return 0;
+out:
+ up_read(&obj->oo_guard);
+ return rc;
}
/* Simple wrapper on top of qsd API which implement quota transfer for osd
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;
if (unlikely(qsd == NULL))
return 0;
- LASSERT(qtype >= 0 && qtype < MAXQUOTAS);
+ LASSERT(qtype >= 0 && qtype < LL_MAXQUOTAS);
qi->lqi_type = qtype;
/* inode accounting */
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;
struct thandle *handle)
{
struct osd_thread_info *info = osd_oti_get(env);
- char *buf = osd_oti_get(env)->oti_str;
struct osd_object *obj = osd_dt_obj(dt);
struct osd_device *osd = osd_obj2dev(obj);
+ dmu_tx_hold_t *txh;
struct osd_thandle *oh;
uint64_t bspace;
uint32_t blksize;
- int rc;
+ int rc = 0;
+ bool found;
ENTRY;
- if (!dt_object_exists(dt)) {
- /* XXX: sanity check that object creation is declared */
- RETURN(0);
- }
LASSERT(handle != NULL);
LASSERT(osd_invariant(obj));
oh = container_of0(handle, struct osd_thandle, ot_super);
+ down_read(&obj->oo_guard);
+ if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
+ GOTO(out, rc = 0);
+
LASSERT(obj->oo_sa_hdl != NULL);
- dmu_tx_hold_sa(oh->ot_tx, obj->oo_sa_hdl, 0);
+ LASSERT(oh->ot_tx != NULL);
+ /* regular attributes are part of the bonus buffer */
+ /* let's check whether this object is already part of
+ * transaction.. */
+ found = false;
+ for (txh = list_head(&oh->ot_tx->tx_holds); txh;
+ txh = list_next(&oh->ot_tx->tx_holds, txh)) {
+ if (txh->txh_dnode == NULL)
+ continue;
+ if (txh->txh_dnode->dn_object != obj->oo_dn->dn_object)
+ continue;
+ /* this object is part of the transaction already
+ * we don't need to declare bonus again */
+ found = true;
+ break;
+ }
+ if (!found)
+ dmu_tx_hold_bonus(oh->ot_tx, obj->oo_dn->dn_object);
+ if (oh->ot_tx->tx_err != 0)
+ GOTO(out, rc = -oh->ot_tx->tx_err);
+
+ if (attr && attr->la_valid & LA_FLAGS) {
+ /* LMA is usually a part of bonus, no need to declare
+ * anything else */
+ }
- sa_object_size(obj->oo_sa_hdl, &blksize, &bspace);
- bspace = toqb(bspace * blksize);
+ if (attr && (attr->la_valid & (LA_UID | LA_GID | LA_PROJID))) {
+ sa_object_size(obj->oo_sa_hdl, &blksize, &bspace);
+ bspace = toqb(bspace * 512);
- if (attr && attr->la_valid & LA_UID) {
- /* account for user inode tracking ZAP update */
- dmu_tx_hold_bonus(oh->ot_tx, osd->od_iusr_oid);
- dmu_tx_hold_zap(oh->ot_tx, osd->od_iusr_oid, TRUE, buf);
+ 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) {
/* quota enforcement for user */
if (attr->la_uid != obj->oo_attr.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)
- RETURN(rc);
+ GOTO(out, rc);
}
}
if (attr && attr->la_valid & LA_GID) {
- /* account for user inode tracking ZAP update */
- dmu_tx_hold_bonus(oh->ot_tx, osd->od_igrp_oid);
- dmu_tx_hold_zap(oh->ot_tx, osd->od_igrp_oid, TRUE, buf);
-
/* quota enforcement for group */
if (attr->la_gid != obj->oo_attr.la_gid) {
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)
- RETURN(rc);
+ GOTO(out, rc);
}
}
-
- RETURN(0);
+#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);
}
/*
* to a transaction group.
*/
static int osd_attr_set(const struct lu_env *env, struct dt_object *dt,
- const struct lu_attr *la, struct thandle *handle,
- struct lustre_capa *capa)
+ const struct lu_attr *la, struct thandle *handle)
{
+ struct osd_thread_info *info = osd_oti_get(env);
+ sa_bulk_attr_t *bulk = osd_oti_get(env)->oti_attr_bulk;
struct osd_object *obj = osd_dt_obj(dt);
struct osd_device *osd = osd_obj2dev(obj);
- udmu_objset_t *uos = &osd->od_objset;
struct osd_thandle *oh;
- struct osa_attr *osa = &osd_oti_get(env)->oti_osa;
- sa_bulk_attr_t *bulk;
+ struct osa_attr *osa = &info->oti_osa;
+ __u64 valid = la->la_valid;
int cnt;
int rc = 0;
ENTRY;
+
+ down_read(&obj->oo_guard);
+ if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
+ GOTO(out, rc = -ENOENT);
+
LASSERT(handle != NULL);
- LASSERT(dt_object_exists(dt));
LASSERT(osd_invariant(obj));
LASSERT(obj->oo_sa_hdl);
transaction group. */
LASSERT(oh->ot_tx->tx_txg != 0);
- if (la->la_valid == 0)
- RETURN(0);
-
- OBD_ALLOC(bulk, sizeof(sa_bulk_attr_t) * 10);
- if (bulk == NULL)
- RETURN(-ENOMEM);
+ 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);
+ }
+ up_read(&obj->oo_guard);
- /* do both accounting updates outside oo_attr_lock below */
- if ((la->la_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_objset.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_objset.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);
+ RETURN(rc > 0 ? 0 : rc);
}
- if ((la->la_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_objset.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_objset.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);
+
+ /* 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_CTIME && la->la_ctime == obj->oo_attr.la_ctime)
+ valid &= ~LA_CTIME;
+
+ if (valid & LA_MTIME && la->la_mtime == obj->oo_attr.la_mtime)
+ valid &= ~LA_MTIME;
+
+ if (valid & LA_ATIME && la->la_atime == obj->oo_attr.la_atime)
+ valid &= ~LA_ATIME;
+
+ if (valid == 0)
+ GOTO(out, rc = 0);
+
+ if (valid & LA_FLAGS) {
+ struct lustre_mdt_attrs *lma;
+ struct lu_buf buf;
+
+ 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) {
+ lma->lma_incompat =
+ le32_to_cpu(lma->lma_incompat);
+ lma->lma_incompat |=
+ lustre_to_lma_flags(la->la_flags);
+ lma->lma_incompat =
+ cpu_to_le32(lma->lma_incompat);
+ buf.lb_buf = lma;
+ buf.lb_len = sizeof(*lma);
+ rc = osd_xattr_set_internal(env, obj, &buf,
+ XATTR_NAME_LMA,
+ LU_XATTR_REPLACE,
+ oh);
+ }
+ if (rc < 0) {
+ CWARN("%s: failed to set LMA flags: rc = %d\n",
+ osd->od_svname, rc);
+ GOTO(out, rc);
+ }
+ }
}
write_lock(&obj->oo_attr_lock);
cnt = 0;
- if (la->la_valid & LA_ATIME) {
+
+ 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(uos), NULL,
+ SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(osd), NULL,
osa->atime, 16);
}
- if (la->la_valid & LA_MTIME) {
+ if (valid & LA_MTIME) {
osa->mtime[0] = obj->oo_attr.la_mtime = la->la_mtime;
- SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(uos), NULL,
+ SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(osd), NULL,
osa->mtime, 16);
}
- if (la->la_valid & LA_CTIME) {
+ if (valid & LA_CTIME) {
osa->ctime[0] = obj->oo_attr.la_ctime = la->la_ctime;
- SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(uos), NULL,
+ SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(osd), NULL,
osa->ctime, 16);
}
- if (la->la_valid & LA_MODE) {
+ if (valid & LA_MODE) {
/* mode is stored along with type, so read it first */
obj->oo_attr.la_mode = (obj->oo_attr.la_mode & S_IFMT) |
(la->la_mode & ~S_IFMT);
osa->mode = obj->oo_attr.la_mode;
- SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(uos), NULL,
+ SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(osd), NULL,
&osa->mode, 8);
}
- if (la->la_valid & LA_SIZE) {
+ if (valid & LA_SIZE) {
osa->size = obj->oo_attr.la_size = la->la_size;
- SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_SIZE(uos), NULL,
+ SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_SIZE(osd), NULL,
&osa->size, 8);
}
- if (la->la_valid & LA_NLINK) {
+ if (valid & LA_NLINK) {
osa->nlink = obj->oo_attr.la_nlink = la->la_nlink;
- SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_LINKS(uos), NULL,
+ SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_LINKS(osd), NULL,
&osa->nlink, 8);
}
- if (la->la_valid & LA_RDEV) {
+ if (valid & LA_RDEV) {
osa->rdev = obj->oo_attr.la_rdev = la->la_rdev;
- SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_RDEV(uos), NULL,
+ SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_RDEV(osd), NULL,
&osa->rdev, 8);
}
- if (la->la_valid & LA_FLAGS) {
+ if (valid & LA_FLAGS) {
osa->flags = attrs_fs2zfs(la->la_flags);
/* many flags are not supported by zfs, so ensure a good cached
* copy */
obj->oo_attr.la_flags = attrs_zfs2fs(osa->flags);
- SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(uos), NULL,
+#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);
}
- if (la->la_valid & LA_UID) {
+ if (valid & LA_UID) {
osa->uid = obj->oo_attr.la_uid = la->la_uid;
- SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_UID(uos), NULL,
+ SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_UID(osd), NULL,
&osa->uid, 8);
}
- if (la->la_valid & LA_GID) {
+ if (valid & LA_GID) {
osa->gid = obj->oo_attr.la_gid = la->la_gid;
- SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GID(uos), NULL,
+ SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GID(osd), NULL,
&osa->gid, 8);
}
- obj->oo_attr.la_valid |= la->la_valid;
+ obj->oo_attr.la_valid |= valid;
write_unlock(&obj->oo_attr_lock);
+ LASSERT(cnt <= ARRAY_SIZE(osd_oti_get(env)->oti_attr_bulk));
rc = osd_object_sa_bulk_update(obj, bulk, cnt, oh);
- OBD_FREE(bulk, sizeof(sa_bulk_attr_t) * 10);
+out:
+ up_read(&obj->oo_guard);
RETURN(rc);
}
static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
struct dt_object *parent, struct dt_object *child,
- cfs_umode_t child_mode)
+ umode_t child_mode)
{
LASSERT(ah);
- memset(ah, 0, sizeof(*ah));
ah->dah_parent = parent;
ah->dah_mode = child_mode;
+
+ if (parent != NULL && !dt_object_remote(parent)) {
+ /* will help to find FID->ino at dt_insert("..") */
+ struct osd_object *pobj = osd_dt_obj(parent);
+
+ osd_idc_find_and_init(env, osd_obj2dev(pobj), pobj);
+ }
}
-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)
{
- char *buf = osd_oti_get(env)->oti_str;
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;
- int rc;
+ dnode_t *dn;
+ int rc, dnode_size;
ENTRY;
LASSERT(dof);
oh = container_of0(handle, struct osd_thandle, ot_super);
LASSERT(oh->ot_tx != NULL);
+ /* this is the minimum set of EAs on every Lustre object */
+ 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);
+
switch (dof->dof_type) {
case DFT_DIR:
+ dt->do_index_ops = &osd_dir_ops;
case DFT_INDEX:
/* for zap create */
- dmu_tx_hold_zap(oh->ot_tx, DMU_NEW_OBJECT, 1, NULL);
+ dmu_tx_hold_zap(oh->ot_tx, DMU_NEW_OBJECT, FALSE, NULL);
+ dmu_tx_hold_sa_create(oh->ot_tx, dnode_size);
break;
case DFT_REGULAR:
case DFT_SYM:
case DFT_NODE:
/* first, we'll create new object */
- dmu_tx_hold_bonus(oh->ot_tx, DMU_NEW_OBJECT);
+ dmu_tx_hold_sa_create(oh->ot_tx, dnode_size);
break;
default:
}
/* and we'll add it to some mapping */
- zapid = osd_get_name_n_idx(env, osd, fid, buf);
- dmu_tx_hold_bonus(oh->ot_tx, zapid);
- dmu_tx_hold_zap(oh->ot_tx, zapid, TRUE, buf);
-
- /* we will also update inode accounting ZAPs */
- dmu_tx_hold_bonus(oh->ot_tx, osd->od_iusr_oid);
- dmu_tx_hold_zap(oh->ot_tx, osd->od_iusr_oid, TRUE, buf);
- dmu_tx_hold_bonus(oh->ot_tx, osd->od_igrp_oid);
- dmu_tx_hold_zap(oh->ot_tx, osd->od_igrp_oid, TRUE, buf);
+ zapid = osd_get_name_n_idx(env, osd, fid, NULL, 0, &dn);
+ osd_tx_hold_zap(oh->ot_tx, zapid, dn, TRUE, NULL);
- dmu_tx_hold_sa_create(oh->ot_tx, ZFS_SA_BASE_ATTR_SIZE);
+ /* will help to find FID->ino mapping at dt_insert() */
+ osd_idc_find_and_init(env, osd, obj);
- __osd_xattr_declare_set(env, obj, sizeof(struct lustre_mdt_attrs),
- XATTR_NAME_LMA, oh);
+ rc = osd_declare_quota(env, osd, attr->la_uid, attr->la_gid,
+ attr->la_projid, 1, oh, NULL, OSD_QID_INODE);
- rc = osd_declare_quota(env, osd, attr->la_uid, attr->la_gid, 1, oh,
- false, NULL, false);
RETURN(rc);
}
-int __osd_attr_init(const struct lu_env *env, udmu_objset_t *uos, uint64_t oid,
- dmu_tx_t *tx, struct lu_attr *la, uint64_t parent)
+int __osd_attr_init(const struct lu_env *env, struct osd_device *osd,
+ 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;
- sa_handle_t *sa_hdl;
+ sa_bulk_attr_t *bulk = osd_oti_get(env)->oti_attr_bulk;
struct osa_attr *osa = &osd_oti_get(env)->oti_osa;
uint64_t gen;
uint64_t crtime[2];
timestruc_t now;
int cnt;
int rc;
+ char *dxattr = NULL;
+ size_t sa_size;
- gethrestime(&now);
- gen = dmu_tx_get_txg(tx);
+ LASSERT(sa_hdl);
+
+ gen = dmu_tx_get_txg(tx);
+ gethrestime(&now);
ZFS_TIME_ENCODE(&now, crtime);
osa->atime[0] = la->la_atime;
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;
-
- /* Now add in all of the "SA" attributes */
- rc = -sa_handle_get(uos->os, oid, NULL, SA_HDL_PRIVATE, &sa_hdl);
- if (rc)
- return rc;
-
- OBD_ALLOC(bulk, sizeof(sa_bulk_attr_t) * 13);
- if (bulk == NULL) {
- rc = -ENOMEM;
- goto out;
+#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.
*/
cnt = 0;
- SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(uos), NULL, &osa->mode, 8);
- SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_SIZE(uos), NULL, &osa->size, 8);
- SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GEN(uos), NULL, &gen, 8);
- SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_UID(uos), NULL, &osa->uid, 8);
- SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GID(uos), NULL, &osa->gid, 8);
- SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_PARENT(uos), NULL, &parent, 8);
- SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(uos), NULL, &osa->flags, 8);
- SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(uos), NULL, osa->atime, 16);
- SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(uos), NULL, osa->mtime, 16);
- SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(uos), NULL, osa->ctime, 16);
- SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CRTIME(uos), NULL, crtime, 16);
- SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_LINKS(uos), NULL, &osa->nlink, 8);
- SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_RDEV(uos), NULL, &osa->rdev, 8);
+ SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(osd), NULL, &osa->mode, 8);
+ SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_SIZE(osd), NULL, &osa->size, 8);
+ SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GEN(osd), NULL, &gen, 8);
+ SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_UID(osd), NULL, &osa->uid, 8);
+ SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GID(osd), NULL, &osa->gid, 8);
+ SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_PARENT(osd), NULL, &parent, 8);
+ SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(osd), NULL, &osa->flags, 8);
+ SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(osd), NULL, osa->atime, 16);
+ SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(osd), NULL, osa->mtime, 16);
+ 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;
+}
+
+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;
+
+ /* take dnode_t from tx to save on dnode#->dnode_t lookup */
+ for (txh = list_tail(&tx->tx_holds); txh;
+ txh = list_prev(&tx->tx_holds, txh)) {
+ dnode_t *dn = txh->txh_dnode;
+ dmu_buf_impl_t *db;
+
+ if (dn == NULL)
+ continue;
+ if (dn->dn_object != oid)
+ continue;
+ db = dn->dn_bonus;
+ if (db == NULL) {
+ rw_enter(&dn->dn_struct_rwlock, RW_WRITER);
+ if (dn->dn_bonus == NULL)
+ dbuf_create_bonus(dn);
+ rw_exit(&dn->dn_struct_rwlock);
+ }
+ db = dn->dn_bonus;
+ LASSERT(db);
+ LASSERT(dn->dn_handle);
+ DB_DNODE_ENTER(db);
+ if (refcount_add(&db->db_holds, osd_obj_tag) == 1) {
+ refcount_add(&dn->dn_holds, tag);
+ 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(tx->tx_objset, oid, dnp);
- OBD_FREE(bulk, sizeof(sa_bulk_attr_t) * 13);
-out:
- sa_handle_destroy(sa_hdl);
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, udmu_objset_t *uos,
- dmu_buf_t **dbp, dmu_tx_t *tx, struct lu_attr *la,
- uint64_t parent, void *tag)
+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)
{
+ dmu_object_type_t type = DMU_OT_PLAIN_FILE_CONTENTS;
uint64_t oid;
- int rc;
-
- LASSERT(tag);
- spin_lock(&uos->lock);
- uos->objects++;
- spin_unlock(&uos->lock);
-
- /* Assert that the transaction has been assigned to a
- transaction group. */
- LASSERT(tx->tx_txg != 0);
-
- /* Create a new DMU object. */
- oid = dmu_object_alloc(uos->os, DMU_OT_PLAIN_FILE_CONTENTS, 0,
- DMU_OT_SA, DN_MAX_BONUSLEN, tx);
- rc = -sa_buf_hold(uos->os, oid, tag, dbp);
- if (rc)
- return rc;
+ int size;
+
+ /* Use DMU_OTN_UINT8_METADATA for local objects so their data blocks
+ * would get an additional ditto copy */
+ if (unlikely(S_ISREG(la->la_mode) &&
+ fid_seq_is_local_file(fid_seq(fid))))
+ type = DMU_OTN_UINT8_METADATA;
+
+ /* Create a new DMU object using the default dnode size. */
+ 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;
la->la_nlink = 1;
- return __osd_attr_init(env, uos, oid, tx, la, parent);
+ return osd_find_new_dnode(env, tx, oid, dnp);
}
/*
* will also require a FAT ZAP. If there is a new type of micro ZAP created
* then we might need to re-evaluate the use of this flag and instead do
* a conversion from the different internal ZAP hash formats being used. */
-int __osd_zap_create(const struct lu_env *env, udmu_objset_t *uos,
- dmu_buf_t **zap_dbp, dmu_tx_t *tx,
- struct lu_attr *la, uint64_t parent,
- void *tag, zap_flags_t flags)
+int __osd_zap_create(const struct lu_env *env, struct osd_device *osd,
+ dnode_t **dnp, dmu_tx_t *tx, struct lu_attr *la,
+ unsigned dnsize, zap_flags_t flags)
{
uint64_t oid;
- int rc;
-
- LASSERT(tag);
-
- spin_lock(&uos->lock);
- uos->objects++;
- spin_unlock(&uos->lock);
/* Assert that the transaction has been assigned to a
transaction group. */
LASSERT(tx->tx_txg != 0);
+ *dnp = NULL;
- oid = zap_create_flags(uos->os, 0, flags | ZAP_FLAG_HASH64,
- DMU_OT_DIRECTORY_CONTENTS, 12, 12,
- DMU_OT_SA, DN_MAX_BONUSLEN, tx);
+ oid = osd_zap_create_flags(osd->od_os, 0, flags | ZAP_FLAG_HASH64,
+ DMU_OT_DIRECTORY_CONTENTS,
+ 14, /* == ZFS fzap_default_blockshift */
+ DN_MAX_INDBLKSHIFT, /* indirect blockshift */
+ dnsize, tx);
- rc = -sa_buf_hold(uos->os, oid, tag, zap_dbp);
- if (rc)
- return rc;
-
- LASSERT(la->la_valid & LA_MODE);
la->la_size = 2;
la->la_nlink = 1;
- return __osd_attr_init(env, uos, oid, tx, la, parent);
+ return osd_find_new_dnode(env, tx, oid, dnp);
}
-static dmu_buf_t *osd_mkidx(const struct lu_env *env, struct osd_device *osd,
- struct lu_attr *la, uint64_t parent,
- struct osd_thandle *oh)
+static dnode_t *osd_mkidx(const struct lu_env *env, struct osd_object *obj,
+ struct lu_attr *la, struct osd_thandle *oh)
{
- dmu_buf_t *db;
- int rc;
+ struct osd_device *osd = osd_obj2dev(obj);
+ dnode_t *dn;
+ int rc;
/* Index file should be created as regular file in order not to confuse
* ZPL which could interpret them as directory.
* 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->od_objset, &db, oh->ot_tx, la,
- parent, osd_obj_tag, 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 db;
+ return dn;
}
-static dmu_buf_t *osd_mkdir(const struct lu_env *env, struct osd_device *osd,
- struct lu_attr *la, uint64_t parent,
- struct osd_thandle *oh)
+static dnode_t *osd_mkdir(const struct lu_env *env, struct osd_object *obj,
+ struct lu_attr *la, struct osd_thandle *oh)
{
- dmu_buf_t *db;
- int rc;
+ struct osd_device *osd = osd_obj2dev(obj);
+ dnode_t *dn;
+ int rc;
LASSERT(S_ISDIR(la->la_mode));
- rc = __osd_zap_create(env, &osd->od_objset, &db, oh->ot_tx, la,
- parent, osd_obj_tag, 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 db;
+ return dn;
}
-static dmu_buf_t* osd_mkreg(const struct lu_env *env, struct osd_device *osd,
- struct lu_attr *la, uint64_t parent,
- struct osd_thandle *oh)
+static dnode_t *osd_mkreg(const struct lu_env *env, struct osd_object *obj,
+ struct lu_attr *la, struct osd_thandle *oh)
{
- dmu_buf_t *db;
- int rc;
+ const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
+ struct osd_device *osd = osd_obj2dev(obj);
+ dnode_t *dn;
+ int rc;
LASSERT(S_ISREG(la->la_mode));
- rc = __osd_object_create(env, &osd->od_objset, &db, oh->ot_tx, la,
- parent, osd_obj_tag);
+ rc = __osd_object_create(env, osd, obj, fid, &dn, oh->ot_tx, la);
if (rc)
return ERR_PTR(rc);
- /*
- * XXX: a hack, OST to use bigger blocksize. we need
- * a method in OSD API to control this from OFD/MDD
- */
- if (!lu_device_is_md(osd2lu_dev(osd))) {
- rc = -dmu_object_set_blocksize(osd->od_objset.os,
- db->db_object,
- 128 << 10, 0, oh->ot_tx);
+ 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
+ * RDMA due to subpage transfer size */
+ rc = -dmu_object_set_blocksize(osd->od_os, dn->dn_object,
+ PAGE_SIZE, 0, oh->ot_tx);
if (unlikely(rc)) {
CERROR("%s: can't change blocksize: %d\n",
osd->od_svname, rc);
}
}
- return db;
+ return dn;
}
-static dmu_buf_t *osd_mksym(const struct lu_env *env, struct osd_device *osd,
- struct lu_attr *la, uint64_t parent,
- struct osd_thandle *oh)
+static dnode_t *osd_mksym(const struct lu_env *env, struct osd_object *obj,
+ struct lu_attr *la, struct osd_thandle *oh)
{
- dmu_buf_t *db;
- int rc;
+ dnode_t *dn;
+ int rc;
LASSERT(S_ISLNK(la->la_mode));
- rc = __osd_object_create(env, &osd->od_objset, &db, oh->ot_tx, la,
- parent, osd_obj_tag);
+ 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 db;
+ return dn;
}
-static dmu_buf_t *osd_mknod(const struct lu_env *env, struct osd_device *osd,
- struct lu_attr *la, uint64_t parent,
- struct osd_thandle *oh)
+static dnode_t *osd_mknod(const struct lu_env *env, struct osd_object *obj,
+ struct lu_attr *la, struct osd_thandle *oh)
{
- dmu_buf_t *db;
- int rc;
+ dnode_t *dn;
+ int rc;
- la->la_valid = LA_MODE;
if (S_ISCHR(la->la_mode) || S_ISBLK(la->la_mode))
la->la_valid |= LA_RDEV;
- rc = __osd_object_create(env, &osd->od_objset, &db, oh->ot_tx, la,
- parent, osd_obj_tag);
+ 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 db;
+ return dn;
}
-typedef dmu_buf_t *(*osd_obj_type_f)(const struct lu_env *env,
- struct osd_device *osd,
- struct lu_attr *la,
- uint64_t parent,
- struct osd_thandle *oh);
+typedef dnode_t *(*osd_obj_type_f)(const struct lu_env *env,
+ struct osd_object *obj,
+ struct lu_attr *la,
+ struct osd_thandle *oh);
static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
{
}
/*
- * Primitives for directory (i.e. ZAP) handling
+ * Concurrency: @dt is write locked.
*/
-static inline int osd_init_lma(const struct lu_env *env, struct osd_object *obj,
- const struct lu_fid *fid, struct osd_thandle *oh)
+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 lu_buf buf;
- int rc;
-
- lustre_lma_init(lma, fid, 0);
- lustre_lma_swab(lma);
- buf.lb_buf = lma;
- buf.lb_len = sizeof(*lma);
-
- rc = osd_xattr_set_internal(env, obj, &buf, XATTR_NAME_LMA,
- LU_XATTR_CREATE, oh, BYPASS_CAPA);
-
- return rc;
-}
-
-/*
- * 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)
-{
- struct zpl_direntry *zde = &osd_oti_get(env)->oti_zde.lzd_reg;
+ struct zpl_direntry *zde = &info->oti_zde.lzd_reg;
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);
- char *buf = osd_oti_get(env)->oti_str;
+ char *buf = info->oti_str;
struct osd_thandle *oh;
- dmu_buf_t *db;
- uint64_t zapid;
+ dnode_t *dn = NULL, *zdn = NULL;
+ uint64_t zapid, parent = 0;
int rc;
+ __u32 compat = 0;
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 */
- down(&obj->oo_guard);
+ down_write(&obj->oo_guard);
+
+ if (unlikely(dt_object_exists(dt)))
+ GOTO(out, rc = -EEXIST);
LASSERT(osd_invariant(obj));
- LASSERT(!dt_object_exists(dt));
LASSERT(dof != NULL);
LASSERT(th != NULL);
oh = container_of0(th, struct osd_thandle, ot_super);
- /*
- * XXX missing: Quote handling.
- */
-
- LASSERT(obj->oo_db == NULL);
+ LASSERT(obj->oo_dn == NULL);
/* to follow ZFS on-disk format we need
* to initialize parent dnode properly */
- zapid = 0;
- if (hint && hint->dah_parent)
- zapid = osd_dt_obj(hint->dah_parent)->oo_db->db_object;
+ if (hint != NULL && hint->dah_parent != NULL &&
+ !dt_object_remote(hint->dah_parent))
+ parent = osd_dt_obj(hint->dah_parent)->oo_dn->dn_object;
+
+ /* we may fix some attributes, better do not change the source */
+ 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
- db = osd_create_type_f(dof->dof_type)(env, osd, attr, zapid, oh);
- if (IS_ERR(db))
- GOTO(out, rc = PTR_ERR(th));
+ dn = osd_create_type_f(dof->dof_type)(env, obj, &obj->oo_attr, oh);
+ if (IS_ERR(dn)) {
+ rc = PTR_ERR(dn);
+ dn = NULL;
+ GOTO(out, rc);
+ }
zde->zde_pad = 0;
- zde->zde_dnode = db->db_object;
+ 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);
+ 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 = -zap_add(osd->od_objset.os, zapid, buf, 8, 1, zde, oh->ot_tx);
+ rc = osd_zap_add(osd, zapid, zdn, buf, 8, 1, zde, oh->ot_tx);
if (rc)
GOTO(out, rc);
- /* Add new object to inode accounting.
- * Errors are not considered as fatal */
- rc = -zap_increment_int(osd->od_objset.os, osd->od_iusr_oid,
- (attr->la_valid & LA_UID) ? attr->la_uid : 0, 1,
- oh->ot_tx);
+skip_add:
+ obj->oo_dn = dn;
+ /* Now add in all of the "SA" attributes */
+ rc = osd_sa_handle_get(obj);
+ if (rc)
+ GOTO(out, rc);
+
+ rc = -nvlist_alloc(&obj->oo_sa_xattr, NV_UNIQUE_NAME, KM_SLEEP);
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_objset.os, osd->od_igrp_oid,
- (attr->la_valid & LA_GID) ? attr->la_gid : 0, 1,
- oh->ot_tx);
+ GOTO(out, rc);
+
+ /* initialize LMA */
+ 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));
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);
+ GOTO(out, rc);
/* configure new osd object */
- obj->oo_db = db;
- rc = osd_object_init0(env, obj);
- LASSERT(ergo(rc == 0, dt_object_exists(dt)));
- LASSERT(osd_invariant(obj));
+ 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);
- rc = osd_init_lma(env, obj, fid, oh);
- if (rc) {
- CERROR("%s: can not set LMA on "DFID": rc = %d\n",
- osd->od_svname, PFID(fid), rc);
- /* ignore errors during LMA initialization */
- rc = 0;
- }
+ /* 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:
- up(&obj->oo_guard);
+ 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)
{
+ 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);
}
/*
* 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;
struct osd_device *osd = osd_obj2dev(obj);
- udmu_objset_t *uos = &osd->od_objset;
uint64_t nlink;
int rc;
ENTRY;
+ down_read(&obj->oo_guard);
+ if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
+ GOTO(out, rc = -ENOENT);
+
LASSERT(osd_invariant(obj));
- LASSERT(dt_object_exists(dt));
LASSERT(obj->oo_sa_hdl != NULL);
oh = container_of0(handle, struct osd_thandle, ot_super);
nlink = ++obj->oo_attr.la_nlink;
write_unlock(&obj->oo_attr_lock);
- rc = osd_object_sa_update(obj, SA_ZPL_LINKS(uos), &nlink, 8, oh);
- return rc;
+ rc = osd_object_sa_update(obj, SA_ZPL_LINKS(osd), &nlink, 8, oh);
+
+out:
+ up_read(&obj->oo_guard);
+ 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)
{
+ 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);
}
/*
* 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;
struct osd_device *osd = osd_obj2dev(obj);
- udmu_objset_t *uos = &osd->od_objset;
uint64_t nlink;
int rc;
ENTRY;
+ down_read(&obj->oo_guard);
+
+ if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
+ GOTO(out, rc = -ENOENT);
+
LASSERT(osd_invariant(obj));
- LASSERT(dt_object_exists(dt));
LASSERT(obj->oo_sa_hdl != NULL);
oh = container_of0(handle, struct osd_thandle, ot_super);
nlink = --obj->oo_attr.la_nlink;
write_unlock(&obj->oo_attr_lock);
- rc = osd_object_sa_update(obj, SA_ZPL_LINKS(uos), &nlink, 8, oh);
- return rc;
+ rc = osd_object_sa_update(obj, SA_ZPL_LINKS(osd), &nlink, 8, oh);
+
+out:
+ up_read(&obj->oo_guard);
+ RETURN(rc);
}
-static int capa_is_sane(const struct lu_env *env, struct osd_device *dev,
- struct lustre_capa *capa, struct lustre_capa_key *keys)
+static int osd_object_sync(const struct lu_env *env, struct dt_object *dt,
+ __u64 start, __u64 end)
{
- struct osd_thread_info *oti = osd_oti_get(env);
- struct obd_capa *oc;
- int i, rc = 0;
+ struct osd_device *osd = osd_obj2dev(osd_dt_obj(dt));
ENTRY;
- oc = capa_lookup(dev->od_capa_hash, capa, 0);
- if (oc) {
- if (capa_is_expired(oc)) {
- DEBUG_CAPA(D_ERROR, capa, "expired");
- rc = -ESTALE;
- }
- capa_put(oc);
- RETURN(rc);
- }
-
- spin_lock(&capa_lock);
- for (i = 0; i < 2; i++) {
- if (keys[i].lk_keyid == capa->lc_keyid) {
- oti->oti_capa_key = keys[i];
- break;
- }
- }
- spin_unlock(&capa_lock);
-
- if (i == 2) {
- DEBUG_CAPA(D_ERROR, capa, "no matched capa key");
- RETURN(-ESTALE);
- }
-
- rc = capa_hmac(oti->oti_capa.lc_hmac, capa, oti->oti_capa_key.lk_key);
- if (rc)
- RETURN(rc);
- if (memcmp(oti->oti_capa.lc_hmac, capa->lc_hmac, sizeof(capa->lc_hmac)))
- {
- DEBUG_CAPA(D_ERROR, capa, "HMAC mismatch");
- RETURN(-EACCES);
+ /* XXX: no other option than syncing the whole filesystem until we
+ * 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) {
+ if (osd_object_sync_delay_us < 0)
+ txg_wait_synced(dmu_objset_pool(osd->od_os), 0ULL);
+ else
+ udelay(osd_object_sync_delay_us);
}
- oc = capa_add(dev->od_capa_hash, capa);
- capa_put(oc);
-
RETURN(0);
}
-static int osd_object_auth(const struct lu_env *env, struct dt_object *dt,
- struct lustre_capa *capa, __u64 opc)
+static int osd_invalidate(const struct lu_env *env, struct dt_object *dt)
{
- const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
- struct osd_device *dev = osd_dev(dt->do_lu.lo_dev);
- int rc;
-
- if (!dev->od_fl_capa)
- return 0;
-
- if (capa == BYPASS_CAPA)
- return 0;
-
- if (!capa) {
- CERROR("no capability is provided for fid "DFID"\n", PFID(fid));
- return -EACCES;
- }
-
- if (!lu_fid_eq(fid, &capa->lc_fid)) {
- DEBUG_CAPA(D_ERROR, capa, "fid "DFID" mismatch with",PFID(fid));
- return -EACCES;
- }
-
- if (!capa_opc_supported(capa, opc)) {
- DEBUG_CAPA(D_ERROR, capa, "opc "LPX64" not supported by", opc);
- return -EACCES;
- }
-
- if ((rc = capa_is_sane(env, dev, capa, dev->od_capa_keys))) {
- DEBUG_CAPA(D_ERROR, capa, "insane (rc %d)", rc);
- return -EACCES;
- }
-
return 0;
}
-static struct obd_capa *osd_capa_get(const struct lu_env *env,
- struct dt_object *dt,
- struct lustre_capa *old,
- __u64 opc)
-{
- struct osd_thread_info *info = osd_oti_get(env);
- const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
- struct osd_object *obj = osd_dt_obj(dt);
- struct osd_device *dev = osd_obj2dev(obj);
- struct lustre_capa_key *key = &info->oti_capa_key;
- struct lustre_capa *capa = &info->oti_capa;
- struct obd_capa *oc;
- int rc;
- ENTRY;
-
- if (!dev->od_fl_capa)
- RETURN(ERR_PTR(-ENOENT));
-
- LASSERT(dt_object_exists(dt));
- LASSERT(osd_invariant(obj));
-
- /* renewal sanity check */
- if (old && osd_object_auth(env, dt, old, opc))
- RETURN(ERR_PTR(-EACCES));
-
- capa->lc_fid = *fid;
- capa->lc_opc = opc;
- capa->lc_uid = 0;
- capa->lc_flags = dev->od_capa_alg << 24;
- capa->lc_timeout = dev->od_capa_timeout;
- capa->lc_expiry = 0;
-
- oc = capa_lookup(dev->od_capa_hash, capa, 1);
- if (oc) {
- LASSERT(!capa_is_expired(oc));
- RETURN(oc);
- }
-
- spin_lock(&capa_lock);
- *key = dev->od_capa_keys[1];
- spin_unlock(&capa_lock);
-
- capa->lc_keyid = key->lk_keyid;
- capa->lc_expiry = cfs_time_current_sec() + dev->od_capa_timeout;
-
- rc = capa_hmac(capa->lc_hmac, capa, key->lk_key);
- if (rc) {
- DEBUG_CAPA(D_ERROR, capa, "HMAC failed: %d for", rc);
- LBUG();
- RETURN(ERR_PTR(rc));
- }
-
- oc = capa_add(dev->od_capa_hash, capa);
- RETURN(oc);
-}
-
-static int osd_object_sync(const struct lu_env *env, struct dt_object *dt)
-{
- struct osd_device *osd = osd_obj2dev(osd_dt_obj(dt));
- ENTRY;
-
- /* XXX: no other option than syncing the whole filesystem until we
- * support ZIL */
- txg_wait_synced(dmu_objset_pool(osd->od_objset.os), 0ULL);
-
- RETURN(0);
-}
-
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,
.do_declare_xattr_del = osd_declare_xattr_del,
.do_xattr_del = osd_xattr_del,
.do_xattr_list = osd_xattr_list,
- .do_capa_get = osd_capa_get,
.do_object_sync = osd_object_sync,
+ .do_invalidate = osd_invalidate,
};
static struct lu_object_operations osd_lu_obj_ops = {
static int osd_otable_it_attr_get(const struct lu_env *env,
struct dt_object *dt,
- struct lu_attr *attr,
- struct lustre_capa *capa)
+ struct lu_attr *attr)
{
attr->la_valid = 0;
return 0;
}
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,
};
+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");