return 0;
}
-static inline int osd_qid_type(struct osd_thandle *oh, int i)
-{
- return (oh->ot_id_type & (1 << i)) ? GRPQUOTA : USRQUOTA;
-}
-
-static inline void osd_qid_set_type(struct osd_thandle *oh, int i, int type)
-{
- oh->ot_id_type |= ((type == GRPQUOTA) ? (1 << i) : 0);
-}
-
-void osd_declare_qid(struct dt_object *dt, struct osd_thandle *oh,
- int type, uid_t id, struct inode *inode)
-{
-#ifdef CONFIG_QUOTA
- int i, allocated = 0;
- struct osd_object *obj;
-
- LASSERT(dt != NULL);
- LASSERT(oh != NULL);
- LASSERTF(oh->ot_id_cnt <= OSD_MAX_UGID_CNT, "count=%u",
- oh->ot_id_cnt);
-
- /* id entry is allocated in the quota file */
- if (inode && inode->i_dquot[type] && inode->i_dquot[type]->dq_off)
- allocated = 1;
-
- for (i = 0; i < oh->ot_id_cnt; i++) {
- if (oh->ot_id_array[i] == id && osd_qid_type(oh, i) == type)
- return;
- }
-
- if (unlikely(i >= OSD_MAX_UGID_CNT)) {
- CERROR("more than %d uid/gids for a transaction?\n", i);
- return;
- }
-
- oh->ot_id_array[i] = id;
- osd_qid_set_type(oh, i, type);
- oh->ot_id_cnt++;
- obj = osd_dt_obj(dt);
- oh->ot_credits += (allocated || id == 0) ?
- 1 : LDISKFS_QUOTA_INIT_BLOCKS(osd_sb(osd_obj2dev(obj)));
-#endif
-}
-
/*
* OSD object methods.
*/
verify = 1;
}
- fid_zero(&oic->oic_fid);
-
/*
* Objects are created as locking anchors or place holders for objects
* yet to be created. No need to osd_oi_lookup() at here because FID
if (IS_ERR(inode)) {
result = PTR_ERR(inode);
if (result == -ENOENT || result == -ESTALE) {
+ fid_zero(&oic->oic_fid);
result = 0;
} else if (result == -EREMCHG) {
th = ERR_PTR(-ENOMEM);
OBD_ALLOC_GFP(oh, sizeof *oh, CFS_ALLOC_IO);
if (oh != NULL) {
+ oh->ot_quota_trans = &oti->oti_quota_trans;
+ memset(oh->ot_quota_trans, 0, sizeof(*oh->ot_quota_trans));
th = &oh->ot_super;
th->th_dev = d;
th->th_result = 0;
lu_device_get(&d->dd_lu_dev);
oh->ot_dev_link = lu_ref_add(&d->dd_lu_dev.ld_reference,
"osd-tx", th);
-
- /*
- * XXX: current rule is that we first start tx,
- * then lock object(s), but we can't use
- * this rule for data (due to locking specifics
- * in ldiskfs). also in long-term we'd like to
- * use usually-used (locks;tx) ordering. so,
- * UGLY thing is that we'll use one ordering for
- * data (ofd) and reverse ordering for metadata
- * (mdd). then at some point we'll fix the latter
- */
- if (dev->od_is_md) {
- LASSERT(oti->oti_r_locks == 0);
- LASSERT(oti->oti_w_locks == 0);
- }
-
oti->oti_txns++;
rc = 0;
} else {
struct osd_thandle *oh;
struct osd_thread_info *oti = osd_oti_get(env);
struct osd_iobuf *iobuf = &oti->oti_iobuf;
-
+ struct qsd_instance *qsd = oti->oti_dev->od_quota_slave;
ENTRY;
oh = container_of0(th, struct osd_thandle, ot_super);
+ if (qsd != NULL)
+ /* inform the quota slave device that the transaction is
+ * stopping */
+ qsd_op_end(env, qsd, oh->ot_quota_trans);
+ oh->ot_quota_trans = NULL;
+
if (oh->ot_handle != NULL) {
handle_t *hdl = oh->ot_handle;
- hdl->h_sync = th->th_sync;
-
/*
* add commit callback
* notice we don't do this in osd_trans_start()
LASSERT(oti->oti_txns == 1);
oti->oti_txns--;
- /*
- * XXX: current rule is that we first start tx,
- * then lock object(s), but we can't use
- * this rule for data (due to locking specifics
- * in ldiskfs). also in long-term we'd like to
- * use usually-used (locks;tx) ordering. so,
- * UGLY thing is that we'll use one ordering for
- * data (ofd) and reverse ordering for metadata
- * (mdd). then at some point we'll fix the latter
- */
- if (osd_dt_dev(th->th_dev)->od_is_md) {
- LASSERT(oti->oti_r_locks == 0);
- LASSERT(oti->oti_w_locks == 0);
- }
rc = dt_txn_hook_stop(env, th);
if (rc != 0)
CERROR("Failure in transaction hook: %d\n", rc);
+
+ /* hook functions might modify th_sync */
+ hdl->h_sync = th->th_sync;
+
oh->ot_handle = NULL;
OSD_CHECK_SLOW_TH(oh, oti->oti_dev,
rc = ldiskfs_journal_stop(hdl));
d = o->oo_dir->od_container.ic_descr;
else
d = NULL;
- return (*p)(env, cookie, LUSTRE_OSD_NAME"-object@%p(i:%p:%lu/%u)[%s]",
+ return (*p)(env, cookie,
+ LUSTRE_OSD_LDISKFS_NAME"-object@%p(i:%p:%lu/%u)[%s]",
o, o->oo_inode,
o->oo_inode ? o->oo_inode->i_ino : 0UL,
o->oo_inode ? o->oo_inode->i_generation : 0,
struct kstatfs *ksfs;
int result = 0;
- if (unlikely(osd->od_mount == NULL))
+ if (unlikely(osd->od_mnt == NULL))
return -EINPROGRESS;
/* osd_lproc.c call this without env, allocate ksfs for that case */
ksfs = &osd_oti_get(env)->oti_ksfs;
}
- cfs_spin_lock(&osd->od_osfs_lock);
- /* cache 1 second */
- if (cfs_time_before_64(osd->od_osfs_age, cfs_time_shift_64(-1))) {
- result = ll_do_statfs(sb, ksfs);
- if (likely(result == 0)) { /* N.B. statfs can't really fail */
- osd->od_osfs_age = cfs_time_current_64();
- statfs_pack(&osd->od_statfs, ksfs);
- }
- }
+ cfs_spin_lock(&osd->od_osfs_lock);
+ /* cache 1 second */
+ if (cfs_time_before_64(osd->od_osfs_age, cfs_time_shift_64(-1))) {
+ result = sb->s_op->statfs(sb->s_root, ksfs);
+ if (likely(result == 0)) { /* N.B. statfs can't really fail */
+ osd->od_osfs_age = cfs_time_current_64();
+ statfs_pack(&osd->od_statfs, ksfs);
+ }
+ }
if (likely(result == 0))
*sfs = osd->od_statfs;
/*
* XXX should be taken from not-yet-existing fs abstraction layer.
*/
+ param->ddp_mnt = osd_dt_dev(dev)->od_mnt;
param->ddp_max_name_len = LDISKFS_NAME_LEN;
param->ddp_max_nlink = LDISKFS_LINK_MAX;
param->ddp_block_shift = sb->s_blocksize_bits;
+ param->ddp_mount_type = LDD_MT_LDISKFS;
param->ddp_mntopts = 0;
if (test_opt(sb, XATTR_USER))
param->ddp_mntopts |= MNTOPT_USERXATTR;
*/
static int osd_sync(const struct lu_env *env, struct dt_device *d)
{
- CDEBUG(D_HA, "syncing OSD %s\n", LUSTRE_OSD_NAME);
+ CDEBUG(D_HA, "syncing OSD %s\n", LUSTRE_OSD_LDISKFS_NAME);
return ldiskfs_force_commit(osd_sb(osd_dt_dev(d)));
}
struct super_block *s = osd_sb(osd_dt_dev(d));
ENTRY;
- CDEBUG(D_HA, "async commit OSD %s\n", LUSTRE_OSD_NAME);
+ CDEBUG(D_HA, "async commit OSD %s\n", LUSTRE_OSD_LDISKFS_NAME);
RETURN(s->s_op->sync_fs(s, 0));
}
int rc;
ENTRY;
- CERROR("*** setting device %s read-only ***\n", LUSTRE_OSD_NAME);
+ CERROR("*** setting %s read-only ***\n", osd_dt_dev(d)->od_svname);
rc = __lvfs_set_rdonly(sb->s_bdev, LDISKFS_SB(sb)->journal_bdev);
RETURN(rc);
attr->la_flags = LDISKFS_I(inode)->i_flags;
attr->la_nlink = inode->i_nlink;
attr->la_rdev = inode->i_rdev;
- attr->la_blksize = ll_inode_blksize(inode);
- attr->la_blkbits = inode->i_blkbits;
+ attr->la_blksize = 1 << inode->i_blkbits;
+ attr->la_blkbits = inode->i_blkbits;
}
static int osd_attr_get(const struct lu_env *env,
const struct lu_attr *attr,
struct thandle *handle)
{
- struct osd_thandle *oh;
- struct osd_object *obj;
+ struct osd_thandle *oh;
+ struct osd_object *obj;
+ struct osd_thread_info *info = osd_oti_get(env);
+ struct lquota_id_info *qi = &info->oti_qi;
+ long long bspace;
+ int rc = 0;
+ bool allocated;
+ ENTRY;
- LASSERT(dt != NULL);
- LASSERT(handle != NULL);
+ LASSERT(dt != NULL);
+ LASSERT(handle != NULL);
- obj = osd_dt_obj(dt);
- LASSERT(osd_invariant(obj));
+ obj = osd_dt_obj(dt);
+ LASSERT(osd_invariant(obj));
- oh = container_of0(handle, struct osd_thandle, ot_super);
- LASSERT(oh->ot_handle == NULL);
+ oh = container_of0(handle, struct osd_thandle, ot_super);
+ LASSERT(oh->ot_handle == NULL);
- OSD_DECLARE_OP(oh, attr_set);
- oh->ot_credits += osd_dto_credits_noquota[DTO_ATTR_SET_BASE];
+ OSD_DECLARE_OP(oh, attr_set);
+ oh->ot_credits += osd_dto_credits_noquota[DTO_ATTR_SET_BASE];
- if (attr && attr->la_valid & LA_UID) {
- if (obj->oo_inode)
- osd_declare_qid(dt, oh, USRQUOTA, obj->oo_inode->i_uid,
- obj->oo_inode);
- osd_declare_qid(dt, oh, USRQUOTA, attr->la_uid, NULL);
- }
- if (attr && attr->la_valid & LA_GID) {
- if (obj->oo_inode)
- osd_declare_qid(dt, oh, GRPQUOTA, obj->oo_inode->i_gid,
- obj->oo_inode);
- osd_declare_qid(dt, oh, GRPQUOTA, attr->la_gid, NULL);
- }
+ if (attr == NULL || obj->oo_inode == NULL)
+ RETURN(rc);
- return 0;
+ bspace = obj->oo_inode->i_blocks;
+ bspace <<= obj->oo_inode->i_sb->s_blocksize_bits;
+ bspace = toqb(bspace);
+
+ /* Changing ownership is always preformed by super user, it should not
+ * fail with EDQUOT.
+ *
+ * We still need to call the osd_declare_qid() to calculate the journal
+ * credits for updating quota accounting files and to trigger quota
+ * space adjustment once the operation is completed.*/
+ if ((attr->la_valid & LA_UID) != 0 &&
+ attr->la_uid != obj->oo_inode->i_uid) {
+ qi->lqi_type = USRQUOTA;
+
+ /* inode accounting */
+ qi->lqi_is_blk = false;
+
+ /* one more inode for the new owner ... */
+ qi->lqi_id.qid_uid = attr->la_uid;
+ qi->lqi_space = 1;
+ allocated = (attr->la_uid == 0) ? true : false;
+ rc = osd_declare_qid(env, oh, qi, allocated, NULL);
+ if (rc == -EDQUOT || rc == -EINPROGRESS)
+ rc = 0;
+ if (rc)
+ RETURN(rc);
+
+ /* and one less inode for the current uid */
+ qi->lqi_id.qid_uid = obj->oo_inode->i_uid;
+ qi->lqi_space = -1;
+ rc = osd_declare_qid(env, oh, qi, true, NULL);
+ if (rc == -EDQUOT || 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 = attr->la_uid;
+ qi->lqi_space = bspace;
+ allocated = (attr->la_uid == 0) ? true : false;
+ rc = osd_declare_qid(env, oh, qi, allocated, 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 = obj->oo_inode->i_uid;
+ qi->lqi_space = -bspace;
+ rc = osd_declare_qid(env, oh, qi, true, NULL);
+ if (rc == -EDQUOT || rc == -EINPROGRESS)
+ rc = 0;
+ if (rc)
+ RETURN(rc);
+ }
+
+ if (attr->la_valid & LA_GID &&
+ attr->la_gid != obj->oo_inode->i_gid) {
+ qi->lqi_type = GRPQUOTA;
+
+ /* inode accounting */
+ qi->lqi_is_blk = false;
+
+ /* one more inode for the new group owner ... */
+ qi->lqi_id.qid_gid = attr->la_gid;
+ qi->lqi_space = 1;
+ allocated = (attr->la_gid == 0) ? true : false;
+ rc = osd_declare_qid(env, oh, qi, allocated, NULL);
+ if (rc == -EDQUOT || rc == -EINPROGRESS)
+ rc = 0;
+ if (rc)
+ RETURN(rc);
+
+ /* and one less inode for the current gid */
+ qi->lqi_id.qid_gid = obj->oo_inode->i_gid;
+ qi->lqi_space = -1;
+ rc = osd_declare_qid(env, oh, qi, true, NULL);
+ if (rc == -EDQUOT || 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_gid = attr->la_gid;
+ qi->lqi_space = bspace;
+ allocated = (attr->la_gid == 0) ? true : false;
+ rc = osd_declare_qid(env, oh, qi, allocated, NULL);
+ if (rc == -EDQUOT || rc == -EINPROGRESS)
+ rc = 0;
+ if (rc)
+ RETURN(rc);
+
+ /* and finally less blocks for the current owner */
+ qi->lqi_id.qid_gid = obj->oo_inode->i_gid;
+ qi->lqi_space = -bspace;
+ rc = osd_declare_qid(env, oh, qi, true, NULL);
+ if (rc == -EDQUOT || rc == -EINPROGRESS)
+ rc = 0;
+ if (rc)
+ RETURN(rc);
+ }
+
+ RETURN(rc);
}
static int osd_inode_setattr(const struct lu_env *env,
static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
- struct dt_object *parent, cfs_umode_t child_mode)
+ struct dt_object *parent, struct dt_object *child,
+ cfs_umode_t child_mode)
{
LASSERT(ah);
struct dt_object_format *dof,
struct thandle *handle)
{
- struct osd_thandle *oh;
+ struct osd_thandle *oh;
+ int rc;
+ ENTRY;
LASSERT(handle != NULL);
if (fid_is_norm(lu_object_fid(&dt->do_lu))) {
OSD_DECLARE_OP(oh, insert);
oh->ot_credits += osd_dto_credits_noquota[DTO_INDEX_INSERT];
+ /* Reuse idle OI block may cause additional one OI block
+ * to be changed. */
+ oh->ot_credits += 1;
}
/* If this is directory, then we expect . and .. to be inserted as
* well. The one directory block always needs to be created for the
oh->ot_credits += osd_dto_credits_noquota[DTO_WRITE_BASE];
}
- if (attr) {
- osd_declare_qid(dt, oh, USRQUOTA, attr->la_uid, NULL);
- osd_declare_qid(dt, oh, GRPQUOTA, attr->la_gid, NULL);
- }
- return 0;
+ if (!attr)
+ RETURN(0);
+
+ rc = osd_declare_inode_qid(env, attr->la_uid, attr->la_gid, 1, oh,
+ false, false, NULL, false);
+ RETURN(rc);
}
static int osd_object_create(const struct lu_env *env, struct dt_object *dt,
struct osd_object *obj = osd_dt_obj(dt);
struct inode *inode = obj->oo_inode;
struct osd_thandle *oh;
-
+ int rc;
ENTRY;
oh = container_of0(th, struct osd_thandle, ot_super);
OSD_DECLARE_OP(oh, destroy);
OSD_DECLARE_OP(oh, delete);
oh->ot_credits += osd_dto_credits_noquota[DTO_OBJECT_DELETE];
- oh->ot_credits += osd_dto_credits_noquota[DTO_INDEX_DELETE];
-
- osd_declare_qid(dt, oh, USRQUOTA, inode->i_uid, inode);
- osd_declare_qid(dt, oh, GRPQUOTA, inode->i_gid, inode);
+ /* XXX: So far, only normal fid needs to be inserted into the OI,
+ * so only normal fid needs to be removed from the OI also. */
+ if (fid_is_norm(lu_object_fid(&dt->do_lu))) {
+ oh->ot_credits += osd_dto_credits_noquota[DTO_INDEX_DELETE];
+ /* Recycle idle OI leaf may cause additional three OI blocks
+ * to be changed. */
+ oh->ot_credits += 3;
+ }
- RETURN(0);
+ /* one less inode */
+ rc = osd_declare_inode_qid(env, inode->i_uid, inode->i_gid, -1, oh,
+ false, true, NULL, false);
+ if (rc)
+ RETURN(rc);
+ /* data to be truncated */
+ rc = osd_declare_inode_qid(env, inode->i_uid, inode->i_gid, 0, oh, true,
+ true, NULL, false);
+ RETURN(rc);
}
static int osd_object_destroy(const struct lu_env *env,
LASSERT(dt_object_exists(dt));
LASSERT(inode->i_op != NULL && inode->i_op->getxattr != NULL);
- LASSERT(osd_read_locked(env, obj) || osd_write_locked(env, obj));
if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
return -EACCES;
const struct lu_buf *buf, const char *name,
int fl, struct thandle *handle)
{
- struct osd_thandle *oh;
+ struct osd_thandle *oh;
- LASSERT(handle != NULL);
+ LASSERT(handle != NULL);
- if (strcmp(name, XATTR_NAME_VERSION) == 0) {
- /* no credits for version */
- return 0;
- }
+ oh = container_of0(handle, struct osd_thandle, ot_super);
+ LASSERT(oh->ot_handle == NULL);
- oh = container_of0(handle, struct osd_thandle, ot_super);
- LASSERT(oh->ot_handle == NULL);
-
- OSD_DECLARE_OP(oh, xattr_set);
- oh->ot_credits += osd_dto_credits_noquota[DTO_XATTR_SET];
+ OSD_DECLARE_OP(oh, xattr_set);
+ if (strcmp(name, XATTR_NAME_VERSION) == 0)
+ oh->ot_credits += osd_dto_credits_noquota[DTO_ATTR_SET_BASE];
+ else
+ oh->ot_credits += osd_dto_credits_noquota[DTO_XATTR_SET];
- return 0;
+ return 0;
}
/*
return result;
result = iam_container_setup(bag);
- if (result != 0)
- goto out;
-
- if (osd_obj2dev(obj)->od_iop_mode) {
- u32 ptr = bag->ic_descr->id_ops->id_root_ptr(bag);
-
- bag->ic_root_bh = ldiskfs_bread(NULL, obj->oo_inode,
- ptr, 0, &result);
- }
-
- out:
if (result == 0)
obj->oo_dt.do_index_ops = &osd_index_iam_ops;
else
}
LINVRNT(osd_invariant(obj));
+ if (is_quota_glb_feat(feat))
+ result = osd_quota_migration(env, dt, feat);
+
return result;
}
struct thandle *handle,
struct lustre_capa *capa)
{
- struct osd_object *obj = osd_dt_obj(dt);
- struct osd_thandle *oh;
- struct iam_path_descr *ipd;
- struct iam_container *bag = &obj->oo_dir->od_container;
- int rc;
+ struct osd_thread_info *oti = osd_oti_get(env);
+ struct osd_object *obj = osd_dt_obj(dt);
+ struct osd_thandle *oh;
+ struct iam_path_descr *ipd;
+ struct iam_container *bag = &obj->oo_dir->od_container;
+ int rc;
ENTRY;
LASSERT(oh->ot_handle != NULL);
LASSERT(oh->ot_handle->h_transaction != NULL);
+ if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
+ /* swab quota uid/gid provided by caller */
+ oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
+ key = (const struct dt_key *)&oti->oti_quota_id;
+ }
+
rc = iam_delete(oh->ot_handle, bag, (const struct iam_key *)key, ipd);
osd_ipd_put(env, bag, ipd);
LINVRNT(osd_invariant(obj));
struct thandle *handle)
{
struct osd_thandle *oh;
+ struct inode *inode;
+ int rc;
+ ENTRY;
LASSERT(dt_object_exists(dt));
LASSERT(handle != NULL);
OSD_DECLARE_OP(oh, delete);
oh->ot_credits += osd_dto_credits_noquota[DTO_INDEX_DELETE];
- LASSERT(osd_dt_obj(dt)->oo_inode);
- osd_declare_qid(dt, oh, USRQUOTA, osd_dt_obj(dt)->oo_inode->i_uid,
- osd_dt_obj(dt)->oo_inode);
- osd_declare_qid(dt, oh, GRPQUOTA, osd_dt_obj(dt)->oo_inode->i_gid,
- osd_dt_obj(dt)->oo_inode);
+ inode = osd_dt_obj(dt)->oo_inode;
+ LASSERT(inode);
- return 0;
+ rc = osd_declare_inode_qid(env, inode->i_uid, inode->i_gid, 0, oh,
+ true, true, NULL, false);
+ RETURN(rc);
}
static inline int osd_get_fid_from_dentry(struct ldiskfs_dir_entry_2 *de,
/* got ipd now we can start iterator. */
iam_it_init(it, bag, 0, ipd);
+ if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
+ /* swab quota uid/gid provided by caller */
+ oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
+ key = (const struct dt_key *)&oti->oti_quota_id;
+ }
+
rc = iam_it_get(it, (struct iam_key *)key);
if (rc >= 0) {
if (S_ISDIR(obj->oo_inode->i_mode))
iam_rec = (struct iam_rec *) rec;
iam_reccpy(&it->ii_path.ip_leaf, (struct iam_rec *)iam_rec);
+
if (S_ISDIR(obj->oo_inode->i_mode))
osd_fid_unpack((struct lu_fid *) rec,
(struct osd_fid_pack *)iam_rec);
+ else if (fid_is_quota(lu_object_fid(&dt->do_lu)))
+ osd_quota_unpack(obj, rec);
}
+
iam_it_put(it);
iam_it_fini(it);
osd_ipd_put(env, bag, ipd);
cfs_cap_t save = cfs_curproc_cap_pack();
#endif
struct osd_thread_info *oti = osd_oti_get(env);
- struct iam_rec *iam_rec = (struct iam_rec *)oti->oti_ldp;
+ struct iam_rec *iam_rec;
int rc;
ENTRY;
else
cfs_cap_lower(CFS_CAP_SYS_RESOURCE);
#endif
- if (S_ISDIR(obj->oo_inode->i_mode))
- osd_fid_pack((struct osd_fid_pack *)iam_rec, rec, &oti->oti_fid);
- else
- iam_rec = (struct iam_rec *) rec;
+ if (S_ISDIR(obj->oo_inode->i_mode)) {
+ iam_rec = (struct iam_rec *)oti->oti_ldp;
+ osd_fid_pack((struct osd_fid_pack *)iam_rec, rec, &oti->oti_fid);
+ } else if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
+ /* pack quota uid/gid */
+ oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
+ key = (const struct dt_key *)&oti->oti_quota_id;
+ /* pack quota record */
+ rec = osd_quota_pack(obj, rec, &oti->oti_quota_rec);
+ iam_rec = (struct iam_rec *)rec;
+ } else {
+ iam_rec = (struct iam_rec *)rec;
+ }
+
rc = iam_insert(oh->ot_handle, bag, (const struct iam_key *)key,
iam_rec, ipd);
#ifdef HAVE_QUOTA_SUPPORT
rc = osd_ea_fid_get(env, obj, ino, fid, &oic->oic_lid);
else
osd_id_gen(&oic->oic_lid, ino, OSD_OII_NOGEN);
-
- if (rc != 0 || !fid_is_norm(fid))
+ if (rc != 0 || !fid_is_norm(fid)) {
+ fid_zero(&oic->oic_fid);
GOTO(out, rc);
+ }
oic->oic_fid = *fid;
if ((scrub->os_pos_current <= ino) &&
struct lu_object *luch;
struct lu_object *lo;
- luch = lu_object_find(env, ludev, fid, NULL);
+ /*
+ * at this point topdev might not exist yet
+ * (i.e. MGS is preparing profiles). so we can
+ * not rely on topdev and instead lookup with
+ * our device passed as topdev. this can't work
+ * if the object isn't cached yet (as osd doesn't
+ * allocate lu_header). IOW, the object must be
+ * in the cache, otherwise lu_object_alloc() crashes
+ * -bzzz
+ */
+ luch = lu_object_find_at(env, ludev, fid, NULL);
if (!IS_ERR(luch)) {
if (lu_object_exists(luch)) {
lo = lu_object_locate(luch->lo_header, ludev->ld_type);
struct thandle *handle)
{
struct osd_thandle *oh;
+ struct inode *inode;
+ int rc;
+ ENTRY;
LASSERT(dt_object_exists(dt));
LASSERT(handle != NULL);
OSD_DECLARE_OP(oh, insert);
oh->ot_credits += osd_dto_credits_noquota[DTO_INDEX_INSERT];
- LASSERT(osd_dt_obj(dt)->oo_inode);
- osd_declare_qid(dt, oh, USRQUOTA, osd_dt_obj(dt)->oo_inode->i_uid,
- osd_dt_obj(dt)->oo_inode);
- osd_declare_qid(dt, oh, GRPQUOTA, osd_dt_obj(dt)->oo_inode->i_gid,
- osd_dt_obj(dt)->oo_inode);
+ inode = osd_dt_obj(dt)->oo_inode;
+ LASSERT(inode);
- return 0;
+ /* We ignore block quota on meta pool (MDTs), so needn't
+ * calculate how many blocks will be consumed by this index
+ * insert */
+ rc = osd_declare_inode_qid(env, inode->i_uid, inode->i_gid, 0, oh,
+ true, true, NULL, false);
+ RETURN(rc);
}
/**
static int osd_it_iam_get(const struct lu_env *env,
struct dt_it *di, const struct dt_key *key)
{
- struct osd_it_iam *it = (struct osd_it_iam *)di;
+ struct osd_thread_info *oti = osd_oti_get(env);
+ struct osd_it_iam *it = (struct osd_it_iam *)di;
+
+ if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
+ /* swab quota uid/gid */
+ oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
+ key = (struct dt_key *)&oti->oti_quota_id;
+ }
return iam_it_get(&it->oi_it, (const struct iam_key *)key);
}
static struct dt_key *osd_it_iam_key(const struct lu_env *env,
const struct dt_it *di)
{
- struct osd_it_iam *it = (struct osd_it_iam *)di;
+ struct osd_thread_info *oti = osd_oti_get(env);
+ struct osd_it_iam *it = (struct osd_it_iam *)di;
+ struct osd_object *obj = it->oi_obj;
+ struct dt_key *key;
- return (struct dt_key *)iam_it_key_get(&it->oi_it);
+ key = (struct dt_key *)iam_it_key_get(&it->oi_it);
+
+ if (!IS_ERR(key) && fid_is_quota(lu_object_fid(&obj->oo_dt.do_lu))) {
+ /* swab quota uid/gid */
+ oti->oti_quota_id = le64_to_cpu(*((__u64 *)key));
+ key = (struct dt_key *)&oti->oti_quota_id;
+ }
+
+ return key;
}
/**
/* IAM does not store object type in IAM index (dir) */
osd_it_pack_dirent(lde, fid, hash, name, namelen,
0, LUDA_FID);
+ } else if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
+ iam_reccpy(&it->oi_it.ii_path.ip_leaf,
+ (struct iam_rec *)dtrec);
+ osd_quota_unpack(it->oi_obj, dtrec);
} else {
iam_reccpy(&it->oi_it.ii_path.ip_leaf,
(struct iam_rec *)dtrec);
it->oie_file.f_pos = 0;
it->oie_file.f_dentry = obj_dentry;
if (attr & LUDA_64BITHASH)
- it->oie_file.f_flags = O_64BITHASH;
+ it->oie_file.f_mode |= FMODE_64BITHASH;
else
- it->oie_file.f_flags = O_32BITHASH;
+ it->oie_file.f_mode |= FMODE_32BITHASH;
it->oie_file.f_mapping = obj->oo_inode->i_mapping;
it->oie_file.f_op = obj->oo_inode->i_fop;
it->oie_file.private_data = NULL;
it->oie_it_dirent++;
RETURN(0);
} else {
- if (it->oie_file.f_pos == LDISKFS_HTREE_EOF)
+ if (it->oie_file.f_pos == ldiskfs_get_htree_eof(&it->oie_file))
rc = +1;
else
rc = osd_ldiskfs_it_fill(env, di);
it->oie_dirent->oied_name,
it->oie_dirent->oied_namelen,
it->oie_dirent->oied_type, attr);
-
- if (!fid_is_norm(fid))
+ if (!fid_is_norm(fid)) {
+ fid_zero(&oic->oic_fid);
RETURN(0);
+ }
oic->oic_fid = *fid;
if ((scrub->os_pos_current <= ino) &&
static int osd_mount(const struct lu_env *env,
struct osd_device *o, struct lustre_cfg *cfg)
{
- struct lustre_mount_info *lmi;
- const char *dev = lustre_cfg_string(cfg, 0);
- struct lustre_disk_data *ldd;
- struct lustre_sb_info *lsi;
- int rc = 0;
-
+ const char *name = lustre_cfg_string(cfg, 0);
+ const char *dev = lustre_cfg_string(cfg, 1);
+ const char *opts;
+ unsigned long page, s_flags, lmd_flags = 0;
+ struct page *__page;
+ struct file_system_type *type;
+ char *options = NULL;
+ char *str;
+ int rc = 0;
ENTRY;
+ if (o->od_mnt != NULL)
+ RETURN(0);
+
o->od_fsops = fsfilt_get_ops(mt_str(LDD_MT_LDISKFS));
if (o->od_fsops == NULL) {
CERROR("Can't find fsfilt_ldiskfs\n");
RETURN(-ENOTSUPP);
}
- if (o->od_mount != NULL) {
- CERROR("Already mounted (%s)\n", dev);
- RETURN(-EEXIST);
- }
+ OBD_PAGE_ALLOC(__page, CFS_ALLOC_STD);
+ if (__page == NULL)
+ GOTO(out, rc = -ENOMEM);
+
+ str = lustre_cfg_string(cfg, 2);
+ s_flags = simple_strtoul(str, NULL, 0);
+ str = strstr(str, ":");
+ if (str)
+ lmd_flags = simple_strtoul(str + 1, NULL, 0);
+ opts = lustre_cfg_string(cfg, 3);
+ page = (unsigned long)cfs_page_address(__page);
+ options = (char *)page;
+ *options = '\0';
+ if (opts == NULL)
+ strcat(options, "user_xattr,acl");
+ else
+ strcat(options, opts);
- /* get mount */
- lmi = server_get_mount(dev);
- if (lmi == NULL) {
- CERROR("Cannot get mount info for %s!\n", dev);
- RETURN(-EFAULT);
- }
+ /* Glom up mount options */
+ if (*options != '\0')
+ strcat(options, ",");
+ strlcat(options, "no_mbcache", CFS_PAGE_SIZE);
- LASSERT(lmi != NULL);
- /* save lustre_mount_info in dt_device */
- o->od_mount = lmi;
- o->od_mnt = lmi->lmi_mnt;
+ type = get_fs_type("ldiskfs");
+ if (!type) {
+ CERROR("%s: cannot find ldiskfs module\n", name);
+ GOTO(out, rc = -ENODEV);
+ }
+
+ o->od_mnt = vfs_kern_mount(type, s_flags, dev, options);
+ cfs_module_put(type->owner);
- lsi = s2lsi(lmi->lmi_sb);
- ldd = lsi->lsi_ldd;
+ if (IS_ERR(o->od_mnt)) {
+ rc = PTR_ERR(o->od_mnt);
+ CERROR("%s: can't mount %s: %d\n", name, dev, rc);
+ o->od_mnt = NULL;
+ GOTO(out, rc);
+ }
- if (ldd->ldd_flags & LDD_F_IAM_DIR) {
+ if (lvfs_check_rdonly(o->od_mnt->mnt_sb->s_bdev)) {
+ CERROR("%s: underlying device %s is marked as read-only. "
+ "Setup failed\n", name, dev);
+ mntput(o->od_mnt);
+ o->od_mnt = NULL;
+ GOTO(out, rc = -EROFS);
+ }
+
+ if (!LDISKFS_HAS_COMPAT_FEATURE(o->od_mnt->mnt_sb,
+ LDISKFS_FEATURE_COMPAT_HAS_JOURNAL)) {
+ CERROR("%s: device %s is mounted w/o journal\n", name, dev);
+ mntput(o->od_mnt);
+ o->od_mnt = NULL;
+ GOTO(out, rc = -EINVAL);
+ }
+
+ if (lmd_flags & LMD_FLG_IAM) {
o->od_iop_mode = 0;
- LCONSOLE_WARN("%s: OSD: IAM mode enabled\n", dev);
+ LCONSOLE_WARN("%s: OSD: IAM mode enabled\n", name);
} else
o->od_iop_mode = 1;
+ if (lmd_flags & LMD_FLG_NOSCRUB)
+ o->od_scrub.os_no_scrub = 1;
- if (ldd->ldd_flags & LDD_F_SV_TYPE_OST) {
- rc = osd_compat_init(o);
- if (rc)
- CERROR("%s: can't initialize compats: %d\n", dev, rc);
- }
+out:
+ if (__page)
+ OBD_PAGE_FREE(__page);
+ if (rc)
+ fsfilt_put_ops(o->od_fsops);
RETURN(rc);
}
int rc;
ENTRY;
+ rc = osd_shutdown(env, osd_dev(d));
+
osd_compat_fini(osd_dev(d));
shrink_dcache_sb(osd_sb(osd_dev(d)));
RETURN (ERR_PTR(rc));
}
- if (osd_dev(d)->od_mount)
- server_put_mount(osd_dev(d)->od_mount->lmi_name,
- osd_dev(d)->od_mount->lmi_mnt);
- osd_dev(d)->od_mount = NULL;
+ if (osd_dev(d)->od_mnt) {
+ mntput(osd_dev(d)->od_mnt);
+ osd_dev(d)->od_mnt = NULL;
+ }
RETURN(NULL);
}
+static int osd_device_init0(const struct lu_env *env,
+ struct osd_device *o,
+ struct lustre_cfg *cfg)
+{
+ struct lu_device *l = osd2lu_dev(o);
+ struct osd_thread_info *info;
+ int rc;
+
+ /* if the module was re-loaded, env can loose its keys */
+ rc = lu_env_refill((struct lu_env *) env);
+ if (rc)
+ GOTO(out, rc);
+ info = osd_oti_get(env);
+ LASSERT(info);
+
+ l->ld_ops = &osd_lu_ops;
+ o->od_dt_dev.dd_ops = &osd_dt_ops;
+
+ cfs_spin_lock_init(&o->od_osfs_lock);
+ cfs_mutex_init(&o->od_otable_mutex);
+ o->od_osfs_age = cfs_time_shift_64(-1000);
+
+ o->od_capa_hash = init_capa_hash();
+ if (o->od_capa_hash == NULL)
+ GOTO(out, rc = -ENOMEM);
+
+ o->od_read_cache = 1;
+ o->od_writethrough_cache = 1;
+
+ rc = osd_mount(env, o, cfg);
+ if (rc)
+ GOTO(out_capa, rc);
+
+ /* setup scrub, including OI files initialization */
+ rc = osd_scrub_setup(env, o);
+ if (rc < 0)
+ GOTO(out_mnt, rc);
+
+ strncpy(o->od_svname, lustre_cfg_string(cfg, 4),
+ sizeof(o->od_svname) - 1);
+
+ if (strstr(o->od_svname, "-OST")) {
+ rc = osd_compat_init(o);
+ if (rc != 0)
+ GOTO(out_mnt, rc);
+ }
+
+ rc = lu_site_init(&o->od_site, l);
+ if (rc)
+ GOTO(out_compat, rc);
+
+ rc = lu_site_init_finish(&o->od_site);
+ if (rc)
+ GOTO(out_compat, rc);
+
+ rc = osd_procfs_init(o, o->od_svname);
+ if (rc != 0) {
+ CERROR("%s: can't initialize procfs: rc = %d\n",
+ o->od_svname, rc);
+ GOTO(out_compat, rc);
+ }
+
+ LASSERT(l->ld_site->ls_linkage.next && l->ld_site->ls_linkage.prev);
+
+ RETURN(0);
+out_compat:
+ osd_compat_fini(o);
+out_mnt:
+ osd_oi_fini(info, o);
+ osd_shutdown(env, o);
+ mntput(o->od_mnt);
+ o->od_mnt = NULL;
+out_capa:
+ cleanup_capa_hash(o->od_capa_hash);
+out:
+ RETURN(rc);
+}
+
static struct lu_device *osd_device_alloc(const struct lu_env *env,
struct lu_device_type *t,
struct lustre_cfg *cfg)
{
- struct lu_device *l;
- struct osd_device *o;
-
- OBD_ALLOC_PTR(o);
- if (o != NULL) {
- int result;
-
- result = dt_device_init(&o->od_dt_dev, t);
- if (result == 0) {
- l = osd2lu_dev(o);
- l->ld_ops = &osd_lu_ops;
- o->od_dt_dev.dd_ops = &osd_dt_ops;
- cfs_spin_lock_init(&o->od_osfs_lock);
- cfs_mutex_init(&o->od_otable_mutex);
- o->od_osfs_age = cfs_time_shift_64(-1000);
- o->od_capa_hash = init_capa_hash();
- if (o->od_capa_hash == NULL) {
- dt_device_fini(&o->od_dt_dev);
- l = ERR_PTR(-ENOMEM);
- }
- } else
- l = ERR_PTR(result);
+ struct osd_device *o;
+ int rc;
- if (IS_ERR(l))
- OBD_FREE_PTR(o);
- } else
- l = ERR_PTR(-ENOMEM);
- return l;
+ OBD_ALLOC_PTR(o);
+ if (o == NULL)
+ return ERR_PTR(-ENOMEM);
+
+ rc = dt_device_init(&o->od_dt_dev, t);
+ if (rc == 0) {
+ rc = osd_device_init0(env, o, cfg);
+ if (rc)
+ dt_device_fini(&o->od_dt_dev);
+ }
+
+ if (unlikely(rc != 0))
+ OBD_FREE_PTR(o);
+
+ return rc == 0 ? osd2lu_dev(o) : ERR_PTR(rc);
}
static struct lu_device *osd_device_free(const struct lu_env *env,
ENTRY;
cleanup_capa_hash(o->od_capa_hash);
+ /* XXX: make osd top device in order to release reference */
+ d->ld_site->ls_top_dev = d;
+ lu_site_purge(env, d->ld_site, -1);
+ if (!cfs_hash_is_empty(d->ld_site->ls_obj_hash)) {
+ LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_ERROR, NULL);
+ lu_site_print(env, d->ld_site, &msgdata, lu_cdebug_printer);
+ }
+ lu_site_fini(&o->od_site);
dt_device_fini(&o->od_dt_dev);
OBD_FREE_PTR(o);
RETURN(NULL);
break;
case LCFG_CLEANUP:
lu_dev_del_linkage(d->ld_site, d);
- err = osd_shutdown(env, o);
- break;
+ err = osd_shutdown(env, o);
+ break;
default:
err = -ENOSYS;
}
RETURN(0);
}
+/*
+ * we use exports to track all osd users
+ */
+static int osd_obd_connect(const struct lu_env *env, struct obd_export **exp,
+ struct obd_device *obd, struct obd_uuid *cluuid,
+ struct obd_connect_data *data, void *localdata)
+{
+ struct osd_device *osd = osd_dev(obd->obd_lu_dev);
+ struct lustre_handle conn;
+ int rc;
+ ENTRY;
+
+ CDEBUG(D_CONFIG, "connect #%d\n", osd->od_connects);
+
+ rc = class_connect(&conn, obd, cluuid);
+ if (rc)
+ RETURN(rc);
+
+ *exp = class_conn2export(&conn);
+
+ cfs_spin_lock(&osd->od_osfs_lock);
+ osd->od_connects++;
+ cfs_spin_unlock(&osd->od_osfs_lock);
+
+ RETURN(0);
+}
+
+/*
+ * once last export (we don't count self-export) disappeared
+ * osd can be released
+ */
+static int osd_obd_disconnect(struct obd_export *exp)
+{
+ struct obd_device *obd = exp->exp_obd;
+ struct osd_device *osd = osd_dev(obd->obd_lu_dev);
+ int rc, release = 0;
+ ENTRY;
+
+ /* Only disconnect the underlying layers on the final disconnect. */
+ cfs_spin_lock(&osd->od_osfs_lock);
+ osd->od_connects--;
+ if (osd->od_connects == 0)
+ release = 1;
+ cfs_spin_unlock(&osd->od_osfs_lock);
+
+ rc = class_disconnect(exp); /* bz 9811 */
+
+ if (rc == 0 && release)
+ class_manual_cleanup(obd);
+ RETURN(rc);
+}
+
static int osd_prepare(const struct lu_env *env, struct lu_device *pdev,
struct lu_device *dev)
{
struct osd_device *osd = osd_dev(dev);
- int result;
+ int result = 0;
ENTRY;
- /* 1. setup scrub, including OI files initialization */
- result = osd_scrub_setup(env, osd);
- if (result < 0)
- RETURN(result);
-
- /* 2. setup quota slave instance */
- osd->od_quota_slave = qsd_init(env, osd->od_svname, &osd->od_dt_dev,
- osd->od_proc_entry);
- if (IS_ERR(osd->od_quota_slave)) {
- result = PTR_ERR(osd->od_quota_slave);
- osd->od_quota_slave = NULL;
- RETURN(result);
- }
-
-#if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2,3,50,0)
+#if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 3, 55, 0)
/* Unfortunately, the current MDD implementation relies on some specific
* code to be executed in the OSD layer. Since OFD now also uses the OSD
* module, we need a way to skip the metadata-specific code when running
#warning "all is_md checks must be removed from osd-ldiskfs"
#endif
- if (!osd->od_is_md)
- RETURN(0);
+ if (osd->od_is_md) {
+ /* 1. setup local objects */
+ result = llo_local_objects_setup(env, lu2md_dev(pdev),
+ lu2dt_dev(dev));
+ if (result)
+ RETURN(result);
+ }
- /* 3. setup local objects */
- result = llo_local_objects_setup(env, lu2md_dev(pdev), lu2dt_dev(dev));
- RETURN(result);
+ /* 2. setup quota slave instance */
+ osd->od_quota_slave = qsd_init(env, osd->od_svname, &osd->od_dt_dev,
+ osd->od_proc_entry);
+ if (IS_ERR(osd->od_quota_slave)) {
+ result = PTR_ERR(osd->od_quota_slave);
+ osd->od_quota_slave = NULL;
+ }
+
+ RETURN(result);
}
static const struct lu_object_operations osd_lu_obj_ops = {
.ldto_device_fini = osd_device_fini
};
-static struct lu_device_type osd_device_type = {
+struct lu_device_type osd_device_type = {
.ldt_tags = LU_DEVICE_DT,
- .ldt_name = LUSTRE_OSD_NAME,
+ .ldt_name = LUSTRE_OSD_LDISKFS_NAME,
.ldt_ops = &osd_device_type_ops,
.ldt_ctx_tags = LCT_LOCAL,
};
* lprocfs legacy support.
*/
static struct obd_ops osd_obd_device_ops = {
- .o_owner = THIS_MODULE
+ .o_owner = THIS_MODULE,
+ .o_connect = osd_obd_connect,
+ .o_disconnect = osd_obd_disconnect
};
static int __init osd_mod_init(void)
osd_oi_mod_init();
lprocfs_osd_init_vars(&lvars);
return class_register_type(&osd_obd_device_ops, NULL, lvars.module_vars,
- LUSTRE_OSD_NAME, &osd_device_type);
+ LUSTRE_OSD_LDISKFS_NAME, &osd_device_type);
}
static void __exit osd_mod_exit(void)
{
- class_unregister_type(LUSTRE_OSD_NAME);
+ class_unregister_type(LUSTRE_OSD_LDISKFS_NAME);
}
MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
-MODULE_DESCRIPTION("Lustre Object Storage Device ("LUSTRE_OSD_NAME")");
+MODULE_DESCRIPTION("Lustre Object Storage Device ("LUSTRE_OSD_LDISKFS_NAME")");
MODULE_LICENSE("GPL");
cfs_module(osd, "0.1.0", osd_mod_init, osd_mod_exit);