* 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/
- * Lustre is a trademark of Sun Microsystems, Inc.
*
* lustre/osd-zfs/osd_object.c
*
#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;
-extern struct dt_body_operations osd_body_ops;
-static struct dt_object_operations osd_obj_otable_it_ops;
-
-extern struct kmem_cache *osd_object_kmem;
+static const struct dt_object_operations osd_obj_ops;
+static const struct lu_object_operations osd_lu_obj_ops;
+static const struct dt_object_operations osd_obj_otable_it_ops;
static void
osd_object_sa_fini(struct osd_object *obj)
write_lock(&obj->oo_attr_lock);
list_del_init(&obj->oo_sa_linkage);
write_unlock(&obj->oo_attr_lock);
- if (obj->oo_late_xattr) {
+ if (obj->oo_late_xattr && obj->oo_destroyed == 0) {
/*
* 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);
+ if (obj->oo_destroyed == 0) {
+ 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);
/*
* 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;
- int cnt = 0;
+ struct osa_attr *osa = &osd_oti_get(env)->oti_osa;
+ sa_bulk_attr_t *bulk = osd_oti_get(env)->oti_attr_bulk;
+ struct lustre_mdt_attrs *lma;
+ struct lu_buf buf;
+ int cnt = 0;
int rc;
ENTRY;
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;
+ la->la_valid |= LA_ATIME | LA_MTIME | LA_CTIME | LA_BTIME | LA_MODE |
+ LA_TYPE | LA_SIZE | LA_UID | LA_GID | LA_FLAGS |
+ LA_NLINK;
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_CRTIME(o), NULL, osa->btime, 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);
la->la_atime = osa->atime[0];
la->la_mtime = osa->mtime[0];
la->la_ctime = osa->ctime[0];
+ la->la_btime = osa->btime[0];
la->la_mode = osa->mode;
la->la_uid = osa->uid;
la->la_gid = osa->gid;
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;
- }
+ /* Try to get extra flags from LMA */
+ lma = (struct lustre_mdt_attrs *)osd_oti_get(env)->oti_buf;
+ buf.lb_buf = lma;
+ buf.lb_len = sizeof(osd_oti_get(env)->oti_buf);
+ down_read(&obj->oo_guard);
+ rc = osd_xattr_get_lma(env, obj, &buf);
+ if (!rc) {
+ lma->lma_incompat = le32_to_cpu(lma->lma_incompat);
+ obj->oo_lma_flags =
+ lma_to_lustre_flags(lma->lma_incompat);
+ } else if (rc == -ENODATA ||
+ !(S_ISDIR(la->la_mode) &&
+ dt_object_exists(&obj->oo_dt))) {
+ rc = 0;
}
+ up_read(&obj->oo_guard);
if (S_ISCHR(la->la_mode) || S_ISBLK(la->la_mode)) {
rc = -sa_lookup(obj->oo_sa_hdl, SA_ZPL_RDEV(o), &osa->rdev, 8);
dbi = (dmu_buf_impl_t *)db;
DB_DNODE_ENTER(dbi);
*dnp = DB_DNODE(dbi);
+ DB_DNODE_EXIT(dbi);
LASSERT(*dnp != NULL);
return 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);
/*
* 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);
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));
+ BUILD_BUG_ON(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);
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_obj2dev(obj)->od_is_ost)
+ 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 */
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 != -EREMCHG)
+ GOTO(out, rc);
+
+ osd_scrub_refresh_mapping(env, osd, fid, oid, DTO_INDEX_DELETE, true,
+ NULL);
+
+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 (scrub->os_running) {
+ if (!rc) {
+ LASSERT(remote);
+
+ lu_object_set_agent_entry(l);
+ osd_oii_insert(env, osd, fid, oid, false);
+ } else {
+ rc = -EINPROGRESS;
}
- 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_scrub.os_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);
+ CDEBUG_LIMIT(D_LFSCK | D_CONSOLE | D_WARNING,
+ "%s: trigger OI scrub by RPC for "DFID"/%#llx: rc = %d\n",
+ osd_name(osd), PFID(fid), oid, 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);
+ /* obj doesn't contain an lu_object_header, so we don't need call_rcu */
OBD_SLAB_FREE_PTR(obj, osd_object_kmem);
+ if (unlikely(h))
+ lu_object_header_free(h);
}
static int
LASSERT(th != NULL);
LASSERT(dt_object_exists(dt));
- oh = container_of0(th, struct osd_thandle, ot_super);
+ oh = container_of(th, struct osd_thandle, ot_super);
LASSERT(oh->ot_tx != NULL);
+ 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, NULL, 0, &dn);
osd_tx_hold_zap(oh->ot_tx, zapid, dn, FALSE, NULL);
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);
LASSERT(obj->oo_dn != NULL);
- oh = container_of0(th, struct osd_thandle, ot_super);
+ oh = container_of(th, struct osd_thandle, ot_super);
LASSERT(oh != NULL);
LASSERT(oh->ot_tx != NULL);
/* remove obj ref from index dir (it depends) */
zapid = osd_get_name_n_idx(env, osd, fid, buf,
sizeof(info->oti_str), &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
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",
+ CERROR("%s: failed to free %s/%#llx: 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",
+ CERROR("%s: failed to free %s/%#llx: rc = %d\n",
osd->od_svname, buf, oid, rc);
} else { /* asynchronous destroy */
char *key = info->oti_key;
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",
+ CERROR("%s: zap_add_int() failed %s/%#llx: 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);
return rc;
}
-static int osd_attr_get(const struct lu_env *env,
- struct dt_object *dt,
+static int osd_attr_get(const struct lu_env *env, struct dt_object *dt,
struct lu_attr *attr)
{
- struct osd_object *obj = osd_dt_obj(dt);
- uint64_t blocks;
- uint32_t blksize;
- int rc = 0;
+ struct osd_object *obj = osd_dt_obj(dt);
+ struct osd_device *osd = osd_obj2dev(obj);
+ uint64_t blocks;
+ uint32_t blksize;
+ int rc = 0;
down_read(&obj->oo_guard);
read_lock(&obj->oo_attr_lock);
*attr = obj->oo_attr;
- if (obj->oo_lma_flags & LUSTRE_ORPHAN_FL)
+ if (obj->oo_lma_flags & LUSTRE_ORPHAN_FL) {
+ attr->la_valid |= LA_FLAGS;
attr->la_flags |= LUSTRE_ORPHAN_FL;
+ }
+ if (obj->oo_lma_flags & LUSTRE_ENCRYPT_FL) {
+ attr->la_valid |= LA_FLAGS;
+ attr->la_flags |= LUSTRE_ENCRYPT_FL;
+ }
read_unlock(&obj->oo_attr_lock);
+ if (attr->la_valid & LA_FLAGS && attr->la_flags & LUSTRE_ORPHAN_FL)
+ CDEBUG(D_INFO, "%s: set orphan flag on "DFID" (%#llx/%#x)\n",
+ osd_obj2dev(obj)->od_svname,
+ PFID(lu_object_fid(&dt->do_lu)),
+ attr->la_valid, obj->oo_lma_flags);
/* with ZFS_DEBUG zrl_add_debug() called by DB_DNODE_ENTER()
* from within sa_object_size() can block on a mutex, so
sa_object_size(obj->oo_sa_hdl, &blksize, &blocks);
/* we do not control size of indices, so always calculate
* it from number of blocks reported by DMU */
- if (S_ISDIR(attr->la_mode))
+ if (S_ISDIR(attr->la_mode)) {
attr->la_size = 512 * blocks;
+ rc = -zap_count(osd->od_os, obj->oo_dn->dn_object,
+ &attr->la_dirent_count);
+ }
/* Block size may be not set; suggest maximal I/O transfers. */
if (blksize == 0)
blksize = osd_spa_maxblocksize(
return rc;
}
-/* Simple wrapper on top of qsd API which implement quota transfer for osd
- * setattr needs. As a reminder, only the root user can change ownership of
- * a file, that's why EDQUOT & EINPROGRESS errors are discarded */
-static inline int qsd_transfer(const struct lu_env *env,
- struct qsd_instance *qsd,
- struct lquota_trans *trans, int qtype,
- __u64 orig_id, __u64 new_id, __u64 bspace,
- struct lquota_id_info *qi)
-{
- int rc;
-
- if (unlikely(qsd == NULL))
- return 0;
-
- LASSERT(qtype >= 0 && qtype < LL_MAXQUOTAS);
- qi->lqi_type = qtype;
-
- /* inode accounting */
- qi->lqi_is_blk = false;
-
- /* one more inode for the new owner ... */
- 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)
- rc = 0;
- if (rc)
- return rc;
-
- /* and one less inode for the current id */
- qi->lqi_id.qid_uid = orig_id;;
- qi->lqi_space = -1;
- /* can't get EDQUOT when reducing usage */
- rc = qsd_op_begin(env, qsd, trans, qi, NULL);
- if (rc == -EINPROGRESS)
- rc = 0;
- if (rc)
- return rc;
-
- /* block accounting */
- qi->lqi_is_blk = true;
-
- /* more blocks for the new owner ... */
- 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)
- rc = 0;
- if (rc)
- return rc;
-
- /* and finally less blocks for the current owner */
- qi->lqi_id.qid_uid = orig_id;
- qi->lqi_space = -bspace;
- rc = qsd_op_begin(env, qsd, trans, qi, NULL);
- /* can't get EDQUOT when reducing usage */
- if (rc == -EINPROGRESS)
- rc = 0;
- return rc;
-}
-
static int osd_declare_attr_set(const struct lu_env *env,
struct dt_object *dt,
const struct lu_attr *attr,
LASSERT(handle != NULL);
LASSERT(osd_invariant(obj));
- oh = container_of0(handle, struct osd_thandle, ot_super);
+ oh = container_of(handle, struct osd_thandle, ot_super);
down_read(&obj->oo_guard);
if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
- GOTO(out, rc = 0);
+ GOTO(out_sem, rc = 0);
LASSERT(obj->oo_sa_hdl != NULL);
LASSERT(oh->ot_tx != NULL);
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);
+ GOTO(out_sem, rc = -oh->ot_tx->tx_err);
if (attr && attr->la_valid & LA_FLAGS) {
/* LMA is usually a part of bonus, no need to declare
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);
}
+ /* to preserve locking order - qsd_transfer() may need to flush
+ * currently running transaction when we're out of quota.
+ */
+ up_read(&obj->oo_guard);
- 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);
- if (rc)
- GOTO(out, rc);
- }
+ /* quota enforcement for user */
+ if (attr && attr->la_valid & LA_UID &&
+ attr->la_uid != obj->oo_attr.la_uid) {
+ rc = qsd_transfer(env, osd_def_qsd(osd),
+ &oh->ot_quota_trans, USRQUOTA,
+ obj->oo_attr.la_uid, attr->la_uid,
+ bspace, &info->oti_qi);
+ if (rc)
+ GOTO(out, rc);
}
- if (attr && attr->la_valid & LA_GID) {
- /* 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);
- if (rc)
- GOTO(out, rc);
- }
+
+ /* quota enforcement for group */
+ if (attr && attr->la_valid & LA_GID &&
+ attr->la_gid != obj->oo_attr.la_gid) {
+ rc = qsd_transfer(env, osd_def_qsd(osd),
+ &oh->ot_quota_trans, GRPQUOTA,
+ obj->oo_attr.la_gid, attr->la_gid,
+ bspace, &info->oti_qi);
+ if (rc)
+ GOTO(out, rc);
}
#ifdef ZFS_PROJINHERIT
- if (attr && attr->la_valid & LA_PROJID) {
+ /* quota enforcement for project */
+ if (attr && attr->la_valid & LA_PROJID &&
+ attr->la_projid != obj->oo_attr.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.
+ /* 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. */
+ * Please note that, different from other attributes,
+ * you can NOT simply set the project ID attribute under
+ * such case, because adding (NOT change) project ID
+ * attribute needs to change the object's attribute
+ * layout to match zfs backend quota accounting
+ * requirement.
+ */
if (unlikely(!obj->oo_with_projid))
GOTO(out, rc = -ENXIO);
- /* quota enforcement for project */
- if (attr->la_projid != obj->oo_attr.la_projid) {
- rc = qsd_transfer(env, osd->od_quota_slave,
- &oh->ot_quota_trans, PRJQUOTA,
- obj->oo_attr.la_projid,
- attr->la_projid, bspace,
- &info->oti_qi);
- if (rc)
- GOTO(out, rc);
- }
+ rc = qsd_transfer(env, osd_def_qsd(osd),
+ &oh->ot_quota_trans, PRJQUOTA,
+ obj->oo_attr.la_projid,
+ attr->la_projid, bspace,
+ &info->oti_qi);
+ if (rc)
+ GOTO(out, rc);
}
#endif
out:
+ RETURN(rc);
+out_sem:
up_read(&obj->oo_guard);
RETURN(rc);
}
LASSERT(osd_invariant(obj));
LASSERT(obj->oo_sa_hdl);
- oh = container_of0(handle, struct osd_thandle, ot_super);
+ oh = container_of(handle, struct osd_thandle, ot_super);
/* Assert that the transaction has been assigned to a
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));
+ BUILD_BUG_ON(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 |=
CWARN("%s: failed to set LMA flags: rc = %d\n",
osd->od_svname, rc);
GOTO(out, rc);
+ } else {
+ obj->oo_lma_flags =
+ la->la_flags & LUSTRE_LMA_FL_MASKS;
}
}
}
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);
+ if (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;
+ 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
#endif
+ valid &= ~LA_PROJID;
}
if (valid & LA_ATIME) {
* copy */
obj->oo_attr.la_flags = attrs_zfs2fs(osa->flags);
#ifdef ZFS_PROJINHERIT
- if (obj->oo_with_projid)
+ if (obj->oo_with_projid && osd->od_projectused_dn)
osa->flags |= ZFS_PROJID;
#endif
SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(osd), NULL,
}
LASSERT(handle != NULL);
- oh = container_of0(handle, struct osd_thandle, ot_super);
+ oh = container_of(handle, struct osd_thandle, ot_super);
LASSERT(oh->ot_tx != NULL);
/* this is the minimum set of EAs on every Lustre object */
switch (dof->dof_type) {
case DFT_DIR:
dt->do_index_ops = &osd_dir_ops;
+ /* fallthrough */
case DFT_INDEX:
/* for zap create */
dmu_tx_hold_zap(oh->ot_tx, DMU_NEW_OBJECT, FALSE, NULL);
struct lu_attr *la, uint64_t parent,
nvlist_t *xattr)
{
- sa_bulk_attr_t *bulk = osd_oti_get(env)->oti_attr_bulk;
- struct osa_attr *osa = &osd_oti_get(env)->oti_osa;
- uint64_t gen;
- uint64_t crtime[2];
- timestruc_t now;
- int cnt;
- int rc;
+ 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;
+ inode_timespec_t now;
+ int cnt;
+ int rc;
char *dxattr = NULL;
size_t sa_size;
gen = dmu_tx_get_txg(tx);
gethrestime(&now);
- ZFS_TIME_ENCODE(&now, crtime);
+ ZFS_TIME_ENCODE(&now, osa->btime);
osa->atime[0] = la->la_atime;
osa->ctime[0] = la->la_ctime;
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_CRTIME(osd), NULL, osa->btime, 16);
SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_LINKS(osd), NULL, &osa->nlink, 8);
#ifdef ZFS_PROJINHERIT
if (osd->od_projectused_dn)
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;
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);
+ if (zfs_refcount_add(&db->db_holds, osd_obj_tag) == 1) {
+ zfs_refcount_add(&dn->dn_holds, osd_obj_tag);
atomic_inc_32(&dn->dn_dbufs_count);
}
*dnp = dn;
+ DB_DNODE_EXIT(db);
dbuf_read(db, NULL, DB_RF_MUST_SUCCEED | DB_RF_NOPREFETCH);
break;
}
osd->od_svname, rc);
return ERR_PTR(rc);
}
+ } else if ((fid_is_llog(fid))) {
+ rc = -dmu_object_set_blocksize(osd->od_os, dn->dn_object,
+ LLOG_MIN_CHUNK_SIZE, 0, oh->ot_tx);
+ if (unlikely(rc)) {
+ CERROR("%s: can't change blocksize: %d\n",
+ osd->od_svname, rc);
+ return ERR_PTR(rc);
+ }
}
return dn;
dnode_t *dn = NULL, *zdn = NULL;
uint64_t zapid, parent = 0;
int rc;
+ __u32 compat = 0;
ENTRY;
LASSERT(dof != NULL);
LASSERT(th != NULL);
- oh = container_of0(th, struct osd_thandle, ot_super);
+ oh = container_of(th, struct osd_thandle, ot_super);
LASSERT(obj->oo_dn == NULL);
/* we may fix some attributes, better do not change the source */
obj->oo_attr = *attr;
+ obj->oo_attr.la_size = 0;
+ obj->oo_attr.la_nlink = 0;
obj->oo_attr.la_valid |= LA_SIZE | LA_NLINK | LA_TYPE;
#ifdef ZFS_PROJINHERIT
zde->zde_pad = 0;
zde->zde_dnode = dn->dn_object;
- zde->zde_type = IFTODT(attr->la_mode & S_IFMT);
+ zde->zde_type = S_DT(attr->la_mode & S_IFMT);
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);
}
LASSERT(osd_invariant(obj));
LASSERT(obj->oo_sa_hdl != NULL);
- oh = container_of0(handle, struct osd_thandle, ot_super);
+ oh = container_of(handle, struct osd_thandle, ot_super);
write_lock(&obj->oo_attr_lock);
nlink = ++obj->oo_attr.la_nlink;
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);
}
LASSERT(osd_invariant(obj));
LASSERT(obj->oo_sa_hdl != NULL);
- oh = container_of0(handle, struct osd_thandle, ot_super);
+ oh = container_of(handle, struct osd_thandle, ot_super);
LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
write_lock(&obj->oo_attr_lock);
__u64 start, __u64 end)
{
struct osd_device *osd = osd_obj2dev(osd_dt_obj(dt));
+ uint64_t txg = 0;
ENTRY;
- /* 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)
- txg_wait_synced(dmu_objset_pool(osd->od_os), 0ULL);
+ if (osd->od_dt_dev.dd_rdonly)
+ RETURN(0);
+
+ txg = osd_db_dirty_txg(osd_dt_obj(dt)->oo_dn->dn_dbuf);
+ if (txg) {
+ /* the object is dirty or being synced */
+ if (osd_object_sync_delay_us < 0)
+ txg_wait_synced(dmu_objset_pool(osd->od_os), txg);
+ else
+ udelay(osd_object_sync_delay_us);
+ }
RETURN(0);
}
return 0;
}
-static struct dt_object_operations osd_obj_ops = {
+static bool osd_check_stale(struct dt_object *dt)
+{
+ return false;
+}
+
+static const struct dt_object_operations osd_obj_ops = {
.do_read_lock = osd_read_lock,
.do_write_lock = osd_write_lock,
.do_read_unlock = osd_read_unlock,
.do_xattr_list = osd_xattr_list,
.do_object_sync = osd_object_sync,
.do_invalidate = osd_invalidate,
+ .do_check_stale = osd_check_stale,
};
-static struct lu_object_operations osd_lu_obj_ops = {
+static const struct lu_object_operations osd_lu_obj_ops = {
.loo_object_init = osd_object_init,
.loo_object_delete = osd_object_delete,
.loo_object_release = osd_object_release,
return 0;
}
-static struct dt_object_operations osd_obj_otable_it_ops = {
+static const struct dt_object_operations osd_obj_otable_it_ops = {
.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