static int out_tx_index_insert_exec(struct mdt_thread_info *info,
struct thandle *th, struct tx_arg *arg)
{
struct dt_object *dt_obj = arg->object;
int rc;
LASSERT(dt_obj != NULL && !IS_ERR(dt_obj));
CDEBUG(D_OTHER, "index insert "DFID" name: %s fid "DFID"\n",
PFID(lu_object_fid(&arg->object->do_lu)),
(char *)arg->u.insert.key,
PFID((struct lu_fid *)arg->u.insert.rec));
if (dt_try_as_dir(info->mti_env, dt_obj) == 0)
return -ENOTDIR;
dt_write_lock(info->mti_env, dt_obj, MOR_TGT_CHILD);
rc = dt_insert(info->mti_env, dt_obj, arg->u.insert.rec,
arg->u.insert.key, th, NULL, 0);
dt_write_unlock(info->mti_env, dt_obj);
update_insert_reply(arg->reply, NULL, 0, arg->index, rc);
return rc;
}
static int out_tx_index_insert_undo(struct mdt_thread_info *info,
struct thandle *th, struct tx_arg *arg)
{
struct dt_object *dt_obj = arg->object;
int rc;
LASSERT(dt_obj != NULL && !IS_ERR(dt_obj));
CDEBUG(D_OTHER, "index delete "DFID" name: %s\n",
PFID(lu_object_fid(&arg->object->do_lu)),
(char *)arg->u.insert.key);
if (dt_try_as_dir(info->mti_env, dt_obj) == 0) {
CERROR("%s: "DFID" is not directory: rc = %d\n",
mdt_obd_name(info->mti_mdt),
PFID(lu_object_fid(&dt_obj->do_lu)), -ENOTDIR);
return -ENOTDIR;
}
dt_write_lock(info->mti_env, dt_obj, MOR_TGT_CHILD);
rc = dt_delete(info->mti_env, dt_obj, arg->u.insert.key, th, NULL);
dt_write_unlock(info->mti_env, dt_obj);
update_insert_reply(arg->reply, NULL, 0, arg->index, rc);
return rc;
}
int out_tx_create_exec(struct mdt_thread_info *info, struct thandle *th,
struct tx_arg *arg)
{
struct dt_object *dt_obj = arg->object;
int rc;
LASSERT(dt_obj != NULL && !IS_ERR(dt_obj));
CDEBUG(D_OTHER, "create "DFID": dof %u, mode %o\n",
PFID(lu_object_fid(&arg->object->do_lu)),
arg->u.create.dof.dof_type,
arg->u.create.attr.la_mode & S_IFMT);
dt_write_lock(info->mti_env, dt_obj, MOR_TGT_CHILD);
rc = dt_create(info->mti_env, dt_obj, &arg->u.create.attr,
&arg->u.create.hint, &arg->u.create.dof, th);
dt_write_unlock(info->mti_env, dt_obj);
CDEBUG(D_INFO, "insert create reply mode %o index %d\n",
arg->u.create.attr.la_mode, arg->index);
update_insert_reply(arg->reply, NULL, 0, arg->index, rc);
return rc;
}
static int llog_osd_pad(const struct lu_env *env, struct dt_object *o,
loff_t *off, int len, int index, struct thandle *th)
{
struct llog_thread_info *lgi = llog_info(env);
int rc;
LASSERT(th);
LASSERT(off);
LASSERT(len >= LLOG_MIN_REC_SIZE && (len & 0x7) == 0);
lgi->lgi_tail.lrt_len = lgi->lgi_lrh.lrh_len = len;
lgi->lgi_tail.lrt_index = lgi->lgi_lrh.lrh_index = index;
lgi->lgi_lrh.lrh_type = LLOG_PAD_MAGIC;
lgi->lgi_buf.lb_buf = &lgi->lgi_lrh;
lgi->lgi_buf.lb_len = sizeof(lgi->lgi_lrh);
dt_write_lock(env, o, 0);
rc = dt_record_write(env, o, &lgi->lgi_buf, off, th);
if (rc) {
CERROR("%s: error writing padding record: rc = %d\n",
o->do_lu.lo_dev->ld_obd->obd_name, rc);
GOTO(out, rc);
}
lgi->lgi_buf.lb_buf = &lgi->lgi_tail;
lgi->lgi_buf.lb_len = sizeof(lgi->lgi_tail);
*off += len - sizeof(lgi->lgi_lrh) - sizeof(lgi->lgi_tail);
rc = dt_record_write(env, o, &lgi->lgi_buf, off, th);
if (rc)
CERROR("%s: error writing padding record: rc = %d\n",
o->do_lu.lo_dev->ld_obd->obd_name, rc);
out:
dt_write_unlock(env, o);
return rc;
}
int out_tx_create_exec(const struct lu_env *env, struct thandle *th,
struct tx_arg *arg)
{
struct dt_object *dt_obj = arg->object;
int rc;
CDEBUG(D_OTHER, "%s: create "DFID": dof %u, mode %o\n",
dt_obd_name(th->th_dev),
PFID(lu_object_fid(&arg->object->do_lu)),
arg->u.create.dof.dof_type,
arg->u.create.attr.la_mode & S_IFMT);
dt_write_lock(env, dt_obj, MOR_TGT_CHILD);
rc = dt_create(env, dt_obj, &arg->u.create.attr,
&arg->u.create.hint, &arg->u.create.dof, th);
dt_write_unlock(env, dt_obj);
CDEBUG(D_INFO, "%s: insert create reply %p index %d: rc = %d\n",
dt_obd_name(th->th_dev), arg->reply, arg->index, rc);
update_insert_reply(arg->reply, NULL, 0, arg->index, rc);
return rc;
}
static int out_tx_xattr_set_exec(const struct lu_env *env,
struct thandle *th,
struct tx_arg *arg)
{
struct dt_object *dt_obj = arg->object;
int rc;
CDEBUG(D_INFO, "%s: set xattr buf %p name %s flag %d\n",
dt_obd_name(th->th_dev), arg->u.xattr_set.buf.lb_buf,
arg->u.xattr_set.name, arg->u.xattr_set.flags);
dt_write_lock(env, dt_obj, MOR_TGT_CHILD);
rc = dt_xattr_set(env, dt_obj, &arg->u.xattr_set.buf,
arg->u.xattr_set.name, arg->u.xattr_set.flags,
th, NULL);
dt_write_unlock(env, dt_obj);
/**
* Ignore errors if this is LINK EA
**/
if (unlikely(rc && !strncmp(arg->u.xattr_set.name, XATTR_NAME_LINK,
strlen(XATTR_NAME_LINK))))
rc = 0;
CDEBUG(D_INFO, "%s: insert xattr set reply %p index %d: rc = %d\n",
dt_obd_name(th->th_dev), arg->reply, arg->index, rc);
update_insert_reply(arg->reply, NULL, 0, arg->index, rc);
return rc;
}
static int out_tx_xattr_set_exec(struct mdt_thread_info *info,
struct thandle *th,
struct tx_arg *arg)
{
struct dt_object *dt_obj = arg->object;
int rc;
LASSERT(dt_obj != NULL && !IS_ERR(dt_obj));
CDEBUG(D_OTHER, "attr set "DFID"\n",
PFID(lu_object_fid(&arg->object->do_lu)));
dt_write_lock(info->mti_env, dt_obj, MOR_TGT_CHILD);
rc = dt_xattr_set(info->mti_env, dt_obj, &arg->u.xattr_set.buf,
arg->u.xattr_set.name, arg->u.xattr_set.flags,
th, NULL);
dt_write_unlock(info->mti_env, dt_obj);
/**
* Ignore errors if this is LINK EA
**/
if (unlikely(rc && !strncmp(arg->u.xattr_set.name, XATTR_NAME_LINK,
strlen(XATTR_NAME_LINK))))
rc = 0;
update_insert_reply(arg->reply, NULL, 0, arg->index, rc);
CDEBUG(D_INFO, "set xattr buf %p name %s flag %d\n",
arg->u.xattr_set.buf.lb_buf, arg->u.xattr_set.name,
arg->u.xattr_set.flags);
return rc;
}
/*
* Write a global record
*
* \param env - is the environment passed by the caller
* \param obj - is the on-disk global index to be updated
* \param id - index to be updated
* \param rec - record to be written
*/
int lquota_disk_write_glb(const struct lu_env *env, struct dt_object *obj,
__u64 id, struct lquota_glb_rec *rec)
{
struct dt_device *dev = lu2dt_dev(obj->do_lu.lo_dev);
struct thandle *th;
struct dt_key *key = (struct dt_key *)&id;
int rc;
ENTRY;
th = dt_trans_create(env, dev);
if (IS_ERR(th))
RETURN(PTR_ERR(th));
/* the entry with 0 key can always be found in IAM file. */
if (id == 0) {
rc = dt_declare_delete(env, obj, key, th);
if (rc)
GOTO(out, rc);
}
rc = dt_declare_insert(env, obj, (struct dt_rec *)rec, key, th);
if (rc)
GOTO(out, rc);
rc = dt_trans_start_local(env, dev, th);
if (rc)
GOTO(out, rc);
dt_write_lock(env, obj, 0);
if (id == 0) {
struct lquota_glb_rec *tmp;
OBD_ALLOC_PTR(tmp);
if (tmp == NULL)
GOTO(out_lock, rc = -ENOMEM);
rc = dt_lookup(env, obj, (struct dt_rec *)tmp, key,
BYPASS_CAPA);
OBD_FREE_PTR(tmp);
if (rc == 0) {
rc = dt_delete(env, obj, key, th, BYPASS_CAPA);
if (rc)
GOTO(out_lock, rc);
}
rc = 0;
}
rc = dt_insert(env, obj, (struct dt_rec *)rec, key, th, BYPASS_CAPA, 1);
out_lock:
dt_write_unlock(env, obj);
out:
dt_trans_stop(env, dev, th);
RETURN(rc);
}
/**
* Implementation of the llog_operations::lop_create
*
* This function creates the llog according with llog_handle::lgh_obj
* and llog_handle::lgh_name.
*
* \param[in] env execution environment
* \param[in] res llog handle of the current llog
* \param[in] th current transaction handle
*
* \retval 0 on successful create
* \retval negative value on error
*/
static int llog_osd_create(const struct lu_env *env, struct llog_handle *res,
struct thandle *th)
{
struct llog_thread_info *lgi = llog_info(env);
struct dt_insert_rec *rec = &lgi->lgi_dt_rec;
struct local_oid_storage *los;
struct dt_object *o;
int rc = 0;
ENTRY;
LASSERT(env);
o = res->lgh_obj;
LASSERT(o);
/* llog can be already created */
if (dt_object_exists(o))
RETURN(-EEXIST);
los = res->private_data;
LASSERT(los);
dt_write_lock(env, o, 0);
if (!dt_object_exists(o))
rc = llog_osd_create_new_object(env, los, o, th);
else
rc = -EEXIST;
dt_write_unlock(env, o);
if (rc)
RETURN(rc);
if (res->lgh_name) {
struct dt_object *llog_dir;
llog_dir = llog_osd_dir_get(env, res->lgh_ctxt);
if (IS_ERR(llog_dir))
RETURN(PTR_ERR(llog_dir));
logid_to_fid(&res->lgh_id, &lgi->lgi_fid);
rec->rec_fid = &lgi->lgi_fid;
rec->rec_type = S_IFREG;
dt_read_lock(env, llog_dir, 0);
rc = dt_insert(env, llog_dir, (struct dt_rec *)rec,
(struct dt_key *)res->lgh_name,
th, BYPASS_CAPA, 1);
dt_read_unlock(env, llog_dir);
lu_object_put(env, &llog_dir->do_lu);
if (rc)
CERROR("%s: can't create named llog %s: rc = %d\n",
o->do_lu.lo_dev->ld_obd->obd_name,
res->lgh_name, rc);
}
RETURN(rc);
}
static int out_obj_ref_del(const struct lu_env *env,
struct dt_object *dt_obj,
struct thandle *th)
{
int rc;
dt_write_lock(env, dt_obj, MOR_TGT_CHILD);
rc = dt_ref_del(env, dt_obj, th);
dt_write_unlock(env, dt_obj);
return rc;
}
static int out_obj_destroy(const struct lu_env *env, struct dt_object *dt_obj,
struct thandle *th)
{
int rc;
CDEBUG(D_INFO, "%s: destroy "DFID"\n", dt_obd_name(th->th_dev),
PFID(lu_object_fid(&dt_obj->do_lu)));
dt_write_lock(env, dt_obj, MOR_TGT_CHILD);
rc = dt_destroy(env, dt_obj, th);
dt_write_unlock(env, dt_obj);
return rc;
}
static int out_tx_ref_add_undo(struct mdt_thread_info *info, struct thandle *th,
struct tx_arg *arg)
{
struct dt_object *dt_obj = arg->object;
LASSERT(dt_obj != NULL && !IS_ERR(dt_obj));
CDEBUG(D_OTHER, "ref del "DFID"\n",
PFID(lu_object_fid(&arg->object->do_lu)));
dt_write_lock(info->mti_env, dt_obj, MOR_TGT_CHILD);
dt_ref_del(info->mti_env, dt_obj, th);
dt_write_unlock(info->mti_env, dt_obj);
return 0;
}
static int out_tx_ref_add_exec(struct mdt_thread_info *info, struct thandle *th,
struct tx_arg *arg)
{
struct dt_object *dt_obj = arg->object;
LASSERT(dt_obj != NULL && !IS_ERR(dt_obj));
CDEBUG(D_OTHER, "ref add "DFID"\n",
PFID(lu_object_fid(&arg->object->do_lu)));
dt_write_lock(info->mti_env, dt_obj, MOR_TGT_CHILD);
dt_ref_add(info->mti_env, dt_obj, th);
dt_write_unlock(info->mti_env, dt_obj);
update_insert_reply(arg->reply, NULL, 0, arg->index, 0);
return 0;
}
/**
* add an orphan \a obj to the orphan index.
* \param obj file or directory.
* \param th transaction for index insert.
*
* \pre obj nlink == 0 && obj->mod_count != 0
*
* \retval 0 success
* \retval -ve index operation error.
*/
int mdd_orphan_insert(const struct lu_env *env, struct mdd_object *obj,
struct thandle *th)
{
struct mdd_device *mdd = mdo2mdd(&obj->mod_obj);
struct dt_object *dor = mdd->mdd_orphans;
const struct lu_fid *lf_dor = lu_object_fid(&dor->do_lu);
struct dt_object *next = mdd_object_child(obj);
struct dt_insert_rec *rec = &mdd_env_info(env)->mti_dt_rec;
int rc;
ENTRY;
LASSERT(mdd_write_locked(env, obj) != 0);
LASSERT(!(obj->mod_flags & ORPHAN_OBJ));
dt_write_lock(env, mdd->mdd_orphans, MOR_TGT_ORPHAN);
rc = mdd_orphan_insert_obj(env, mdd, obj, th);
if (rc)
GOTO(out, rc);
mdo_ref_add(env, obj, th);
if (!S_ISDIR(mdd_object_type(obj)))
GOTO(out, rc = 0);
mdo_ref_add(env, obj, th);
dt_ref_add(env, mdd->mdd_orphans, th);
/* try best to fixup directory, do not return errors from here */
if (!dt_try_as_dir(env, next))
GOTO(out, rc = 0);
dt_delete(env, next, (const struct dt_key *)dotdot, th);
rec->rec_fid = lf_dor;
rec->rec_type = S_IFDIR;
dt_insert(env, next, (const struct dt_rec *)rec,
(const struct dt_key *)dotdot, th);
out:
if (rc == 0)
obj->mod_flags |= ORPHAN_OBJ;
dt_write_unlock(env, mdd->mdd_orphans);
RETURN(rc);
}
/**
* delete an orphan \a obj from orphan index.
* \param obj file or directory.
* \param th transaction for index deletion and object destruction.
*
* \pre obj->mod_count == 0 && ORPHAN_OBJ is set for obj.
*
* \retval 0 success
* \retval -ve index operation error.
*/
int mdd_orphan_delete(const struct lu_env *env, struct mdd_object *obj,
struct thandle *th)
{
struct mdd_device *mdd = mdo2mdd(&obj->mod_obj);
struct dt_object *dor = mdd->mdd_orphans;
struct dt_key *key;
int rc = 0;
ENTRY;
LASSERT(mdd_write_locked(env, obj) != 0);
LASSERT(obj->mod_flags & ORPHAN_OBJ);
LASSERT(obj->mod_count == 0);
LASSERT(dor);
key = mdd_orphan_key_fill(env, mdo2fid(obj));
dt_write_lock(env, mdd->mdd_orphans, MOR_TGT_ORPHAN);
if (OBD_FAIL_CHECK(OBD_FAIL_MDS_ORPHAN_DELETE))
goto ref_del;
rc = dt_delete(env, mdd->mdd_orphans, key, th);
if (rc == -ENOENT) {
key = mdd_orphan_key_fill_20(env, mdo2fid(obj));
rc = dt_delete(env, mdd->mdd_orphans, key, th);
}
ref_del:
if (!rc) {
/* lov objects will be destroyed by caller */
mdo_ref_del(env, obj, th);
if (S_ISDIR(mdd_object_type(obj))) {
mdo_ref_del(env, obj, th);
dt_ref_del(env, mdd->mdd_orphans, th);
}
obj->mod_flags &= ~ORPHAN_OBJ;
} else {
CERROR("%s: could not delete orphan object "DFID": rc = %d\n",
mdd2obd_dev(mdd)->obd_name, PFID(mdo2fid(obj)), rc);
}
dt_write_unlock(env, mdd->mdd_orphans);
RETURN(rc);
}
/**
* All of the xxx_undo will be used once execution failed,
* But because all of the required resource has been reserved in
* declare phase, i.e. if declare succeed, it should make sure
* the following executing phase succeed in anyway, so these undo
* should be useless for most of the time in Phase I
**/
int out_tx_create_undo(struct mdt_thread_info *info, struct thandle *th,
struct tx_arg *arg)
{
struct dt_object *dt_obj = arg->object;
int rc;
LASSERT(dt_obj != NULL && !IS_ERR(dt_obj));
dt_write_lock(info->mti_env, dt_obj, MOR_TGT_CHILD);
rc = dt_destroy(info->mti_env, dt_obj, th);
dt_write_unlock(info->mti_env, dt_obj);
/* we don't like double failures */
if (rc != 0)
CERROR("%s: undo failure, we are doomed!: rc = %d\n",
mdt_obd_name(info->mti_mdt), rc);
return rc;
}
static int out_tx_write_exec(const struct lu_env *env, struct thandle *th,
struct tx_arg *arg)
{
struct dt_object *dt_obj = arg->object;
int rc;
dt_write_lock(env, dt_obj, MOR_TGT_CHILD);
rc = dt_record_write(env, dt_obj, &arg->u.write.buf,
&arg->u.write.pos, th);
dt_write_unlock(env, dt_obj);
if (rc == 0)
rc = arg->u.write.buf.lb_len;
object_update_result_insert(arg->reply, NULL, 0, arg->index, rc);
return rc > 0 ? 0 : rc;
}
static int out_obj_index_delete(const struct lu_env *env,
struct dt_object *dt_obj,
const struct dt_key *key,
struct thandle *th)
{
int rc;
CDEBUG(D_INFO, "%s: index delete "DFID" name: %s\n",
dt_obd_name(th->th_dev), PFID(lu_object_fid(&dt_obj->do_lu)),
(char *)key);
if (dt_try_as_dir(env, dt_obj) == 0)
return -ENOTDIR;
dt_write_lock(env, dt_obj, MOR_TGT_CHILD);
rc = dt_delete(env, dt_obj, key, th, NULL);
dt_write_unlock(env, dt_obj);
return rc;
}
static int out_tx_attr_set_exec(struct mdt_thread_info *info,
struct thandle *th, struct tx_arg *arg)
{
struct dt_object *dt_obj = arg->object;
int rc;
LASSERT(dt_obj != NULL && !IS_ERR(dt_obj));
CDEBUG(D_OTHER, "attr set "DFID"\n",
PFID(lu_object_fid(&arg->object->do_lu)));
dt_write_lock(info->mti_env, dt_obj, MOR_TGT_CHILD);
rc = dt_attr_set(info->mti_env, dt_obj, &arg->u.attr_set.attr,
th, NULL);
dt_write_unlock(info->mti_env, dt_obj);
update_insert_reply(arg->reply, NULL, 0, arg->index, rc);
return rc;
}
static int out_tx_attr_set_exec(const struct lu_env *env, struct thandle *th,
struct tx_arg *arg)
{
struct dt_object *dt_obj = arg->object;
int rc;
CDEBUG(D_OTHER, "%s: attr set "DFID"\n", dt_obd_name(th->th_dev),
PFID(lu_object_fid(&dt_obj->do_lu)));
dt_write_lock(env, dt_obj, MOR_TGT_CHILD);
rc = dt_attr_set(env, dt_obj, &arg->u.attr_set.attr, th, NULL);
dt_write_unlock(env, dt_obj);
CDEBUG(D_INFO, "%s: insert attr_set reply %p index %d: rc = %d\n",
dt_obd_name(th->th_dev), arg->reply, arg->index, rc);
update_insert_reply(arg->reply, NULL, 0, arg->index, rc);
return rc;
}
static int out_obj_index_insert(const struct lu_env *env,
struct dt_object *dt_obj,
const struct dt_rec *rec,
const struct dt_key *key,
struct thandle *th)
{
int rc;
CDEBUG(D_INFO, "%s: index insert "DFID" name: %s fid "DFID"\n",
dt_obd_name(th->th_dev), PFID(lu_object_fid(&dt_obj->do_lu)),
(char *)key, PFID((struct lu_fid *)rec));
if (dt_try_as_dir(env, dt_obj) == 0)
return -ENOTDIR;
dt_write_lock(env, dt_obj, MOR_TGT_CHILD);
rc = dt_insert(env, dt_obj, rec, key, th, NULL, 0);
dt_write_unlock(env, dt_obj);
return rc;
}
/*
* Update a record in a quota index file.
*
* \param env - is the environment passed by the caller
* \param th - is the transaction to use for disk writes
* \param obj - is the on-disk index to be updated.
* \param id - is the key to be updated
* \param rec - is the input record containing the new quota settings.
* \param flags - can be LQUOTA_BUMP_VER or LQUOTA_SET_VER.
* \param ver - is the new version of the index if LQUOTA_SET_VER is set or is
* used to return the new version of the index when
* LQUOTA_BUMP_VER is set.
*
* \retval - 0 on success, appropriate error on failure
*/
int lquota_disk_write(const struct lu_env *env, struct thandle *th,
struct dt_object *obj, union lquota_id *id,
struct dt_rec *rec, __u32 flags, __u64 *ver)
{
struct lquota_thread_info *qti = lquota_info(env);
struct dt_key *key = (struct dt_key *)&id->qid_uid;
int rc;
ENTRY;
LASSERT(dt_object_exists(obj));
LASSERT(obj->do_index_ops != NULL);
/* lock index */
dt_write_lock(env, obj, 0);
/* check whether there is already an existing record for this ID */
rc = dt_lookup(env, obj, (struct dt_rec *)&qti->qti_rec, key,
BYPASS_CAPA);
if (rc == 0) {
/* delete existing record in order to replace it */
rc = dt_delete(env, obj, key, th, BYPASS_CAPA);
if (rc)
GOTO(out, rc);
} else if (rc == -ENOENT) {
/* probably first insert */
rc = 0;
} else {
GOTO(out, rc);
}
if (rec != NULL) {
/* insert record with updated quota settings */
rc = dt_insert(env, obj, rec, key, th, BYPASS_CAPA, 1);
if (rc) {
/* try to insert the old one */
rc = dt_insert(env, obj, (struct dt_rec *)&qti->qti_rec,
key, th, BYPASS_CAPA, 1);
LASSERTF(rc == 0, "failed to insert record in quota "
"index "DFID,
PFID(lu_object_fid(&obj->do_lu)));
GOTO(out, rc);
}
}
if (flags != 0) {
LASSERT(ver);
if (flags & LQUOTA_BUMP_VER) {
/* caller wants to bump the version, let's first read
* it */
*ver = dt_version_get(env, obj);
(*ver)++;
} else {
LASSERT(flags & LQUOTA_SET_VER);
}
dt_version_set(env, obj, *ver, th);
}
EXIT;
out:
dt_write_unlock(env, obj);
return rc;
}
static int llog_osd_destroy(const struct lu_env *env,
struct llog_handle *loghandle)
{
struct llog_thread_info *lgi = llog_info(env);
struct llog_ctxt *ctxt;
struct dt_object *o, *llog_dir = NULL;
struct dt_device *d;
struct thandle *th;
char *name = NULL;
int rc;
ENTRY;
ctxt = loghandle->lgh_ctxt;
LASSERT(ctxt);
o = loghandle->lgh_obj;
LASSERT(o);
d = lu2dt_dev(o->do_lu.lo_dev);
LASSERT(d);
LASSERT(d == ctxt->loc_exp->exp_obd->obd_lvfs_ctxt.dt);
th = dt_trans_create(env, d);
if (IS_ERR(th))
RETURN(PTR_ERR(th));
if (loghandle->lgh_name) {
llog_dir = llog_osd_dir_get(env, ctxt);
if (IS_ERR(llog_dir))
GOTO(out_trans, rc = PTR_ERR(llog_dir));
dt_declare_ref_del(env, o, th);
name = loghandle->lgh_name;
rc = dt_declare_delete(env, llog_dir,
(struct dt_key *)name, th);
if (rc)
GOTO(out_trans, rc);
}
dt_declare_ref_del(env, o, th);
rc = dt_declare_destroy(env, o, th);
if (rc)
GOTO(out_trans, rc);
rc = dt_trans_start_local(env, d, th);
if (rc)
GOTO(out_trans, rc);
dt_write_lock(env, o, 0);
if (dt_object_exists(o)) {
if (name) {
dt_ref_del(env, o, th);
dt_read_lock(env, llog_dir, 0);
rc = dt_delete(env, llog_dir,
(struct dt_key *) name,
th, BYPASS_CAPA);
dt_read_unlock(env, llog_dir);
if (rc) {
CERROR("%s: can't remove llog %s: rc = %d\n",
o->do_lu.lo_dev->ld_obd->obd_name,
name, rc);
GOTO(out_unlock, rc);
}
}
/*
* XXX: compatibility bits
* on old filesystems llogs are referenced by the name
* on the new ones they are referenced by OI and by
* the name
*/
rc = dt_attr_get(env, o, &lgi->lgi_attr, NULL);
if (rc)
GOTO(out_unlock, rc);
LASSERT(lgi->lgi_attr.la_nlink < 2);
if (lgi->lgi_attr.la_nlink == 1)
dt_ref_del(env, o, th);
rc = dt_destroy(env, o, th);
if (rc)
GOTO(out_unlock, rc);
}
out_unlock:
dt_write_unlock(env, o);
out_trans:
dt_trans_stop(env, d, th);
if (llog_dir != NULL)
lu_object_put(env, &llog_dir->do_lu);
RETURN(rc);
}
static int llog_osd_write_blob(const struct lu_env *env, struct dt_object *o,
struct llog_rec_hdr *rec, void *buf,
loff_t *off, struct thandle *th)
{
struct llog_thread_info *lgi = llog_info(env);
int buflen = rec->lrh_len;
int rc;
ENTRY;
LASSERT(env);
LASSERT(o);
if (buflen == 0)
CWARN("0-length record\n");
CDEBUG(D_OTHER, "write blob with type %x, buf %p/%u at off %llu\n",
rec->lrh_type, buf, buflen, *off);
if (!buf) {
lgi->lgi_buf.lb_len = buflen;
lgi->lgi_buf.lb_buf = rec;
rc = dt_record_write(env, o, &lgi->lgi_buf, off, th);
if (rc)
CERROR("%s: error writing log record: rc = %d\n",
o->do_lu.lo_dev->ld_obd->obd_name, rc);
RETURN(rc);
}
/* the buf case */
/* protect the following 3 writes from concurrent read */
dt_write_lock(env, o, 0);
rec->lrh_len = sizeof(*rec) + buflen + sizeof(lgi->lgi_tail);
lgi->lgi_buf.lb_len = sizeof(*rec);
lgi->lgi_buf.lb_buf = rec;
rc = dt_record_write(env, o, &lgi->lgi_buf, off, th);
if (rc) {
CERROR("%s: error writing log hdr: rc = %d\n",
o->do_lu.lo_dev->ld_obd->obd_name, rc);
GOTO(out, rc);
}
lgi->lgi_buf.lb_len = buflen;
lgi->lgi_buf.lb_buf = buf;
rc = dt_record_write(env, o, &lgi->lgi_buf, off, th);
if (rc) {
CERROR("%s: error writing log buffer: rc = %d\n",
o->do_lu.lo_dev->ld_obd->obd_name, rc);
GOTO(out, rc);
}
lgi->lgi_tail.lrt_len = rec->lrh_len;
lgi->lgi_tail.lrt_index = rec->lrh_index;
lgi->lgi_buf.lb_len = sizeof(lgi->lgi_tail);
lgi->lgi_buf.lb_buf = &lgi->lgi_tail;
rc = dt_record_write(env, o, &lgi->lgi_buf, off, th);
if (rc)
CERROR("%s: error writing log tail: rc = %d\n",
o->do_lu.lo_dev->ld_obd->obd_name, rc);
out:
dt_write_unlock(env, o);
RETURN(rc);
}
static inline void mdd_orphan_write_lock(const struct lu_env *env,
struct mdd_device *mdd)
{
struct dt_object *dor = mdd->mdd_orphans;
dt_write_lock(env, dor, MOR_TGT_ORPHAN);
}
int ofd_precreate_objects(const struct lu_env *env, struct ofd_device *ofd,
obd_id id, struct ofd_seq *oseq, int nr, int sync)
{
struct ofd_thread_info *info = ofd_info(env);
struct ofd_object *fo = NULL;
struct dt_object *next;
struct thandle *th;
struct ofd_object **batch;
struct lu_fid *fid = &info->fti_fid;
obd_id tmp;
int rc;
int i;
int objects = 0;
int nr_saved = nr;
ENTRY;
/* Don't create objects beyond the valid range for this SEQ */
if (unlikely(fid_seq_is_mdt0(ostid_seq(&oseq->os_oi)) &&
(id + nr) >= IDIF_MAX_OID)) {
CERROR("%s:"DOSTID" hit the IDIF_MAX_OID (1<<48)!\n",
ofd_name(ofd), id, ostid_seq(&oseq->os_oi));
RETURN(rc = -ENOSPC);
} else if (unlikely(!fid_seq_is_mdt0(ostid_seq(&oseq->os_oi)) &&
(id + nr) >= OBIF_MAX_OID)) {
CERROR("%s:"DOSTID" hit the OBIF_MAX_OID (1<<32)!\n",
ofd_name(ofd), id, ostid_seq(&oseq->os_oi));
RETURN(rc = -ENOSPC);
}
OBD_ALLOC(batch, nr_saved * sizeof(struct ofd_object *));
if (batch == NULL)
RETURN(-ENOMEM);
info->fti_attr.la_valid = LA_TYPE | LA_MODE;
/*
* We mark object SUID+SGID to flag it for accepting UID+GID from
* client on first write. Currently the permission bits on the OST are
* never used, so this is OK.
*/
info->fti_attr.la_mode = S_IFREG | S_ISUID | S_ISGID | 0666;
info->fti_dof.dof_type = dt_mode_to_dft(S_IFREG);
/* Initialize a/c/m time so any client timestamp will always
* be newer and update the inode. ctime = 0 is also handled
* specially in osd_inode_setattr(). See LU-221, LU-1042 */
info->fti_attr.la_valid |= LA_ATIME | LA_MTIME | LA_CTIME;
info->fti_attr.la_atime = 0;
info->fti_attr.la_mtime = 0;
info->fti_attr.la_ctime = 0;
LASSERT(id != 0);
/* prepare objects */
*fid = *lu_object_fid(&oseq->os_lastid_obj->do_lu);
for (i = 0; i < nr; i++) {
rc = fid_set_id(fid, id + i);
if (rc != 0) {
if (i == 0)
GOTO(out, rc);
nr = i;
break;
}
fo = ofd_object_find(env, ofd, fid);
if (IS_ERR(fo)) {
if (i == 0)
GOTO(out, rc = PTR_ERR(fo));
nr = i;
break;
}
ofd_write_lock(env, fo);
batch[i] = fo;
}
info->fti_buf.lb_buf = &tmp;
info->fti_buf.lb_len = sizeof(tmp);
info->fti_off = 0;
th = ofd_trans_create(env, ofd);
if (IS_ERR(th))
GOTO(out, rc = PTR_ERR(th));
th->th_sync |= sync;
rc = dt_declare_record_write(env, oseq->os_lastid_obj, &info->fti_buf,
info->fti_off, th);
if (rc)
GOTO(trans_stop, rc);
for (i = 0; i < nr; i++) {
fo = batch[i];
LASSERT(fo);
if (unlikely(ofd_object_exists(fo))) {
/* object may exist being re-created by write replay */
CDEBUG(D_INODE, "object "LPX64"/"LPX64" exists: "
DFID"\n", ostid_seq(&oseq->os_oi), id,
PFID(lu_object_fid(&fo->ofo_obj.do_lu)));
continue;
}
next = ofd_object_child(fo);
LASSERT(next != NULL);
rc = dt_declare_create(env, next, &info->fti_attr, NULL,
&info->fti_dof, th);
if (rc) {
nr = i;
break;
}
}
rc = dt_trans_start_local(env, ofd->ofd_osd, th);
if (rc)
GOTO(trans_stop, rc);
CDEBUG(D_OTHER, "%s: create new object "DFID" nr %d\n",
ofd_name(ofd), PFID(fid), nr);
LASSERT(nr > 0);
/* When the LFSCK scanning the whole device to verify the LAST_ID file
* consistency, it will load the last_id into RAM firstly, and compare
* the last_id with each OST-object's ID. If the later one is larger,
* then it will regard the LAST_ID file crashed. But during the LFSCK
* scanning, the OFD may continue to create new OST-objects. Those new
* created OST-objects will have larger IDs than the LFSCK known ones.
* So from the LFSCK view, it needs to re-load the last_id from disk
* file, and if the latest last_id is still smaller than the object's
* ID, then the LAST_ID file is real crashed.
*
* To make above mechanism to work, before OFD pre-create OST-objects,
* it needs to update the LAST_ID file firstly, otherwise, the LFSCK
* may cannot get latest last_id although new OST-object created. */
if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_SKIP_LASTID)) {
tmp = cpu_to_le64(id + nr - 1);
dt_write_lock(env, oseq->os_lastid_obj, 0);
rc = dt_record_write(env, oseq->os_lastid_obj,
&info->fti_buf, &info->fti_off, th);
dt_write_unlock(env, oseq->os_lastid_obj);
if (rc != 0)
GOTO(trans_stop, rc);
}
for (i = 0; i < nr; i++) {
fo = batch[i];
LASSERT(fo);
/* Only the new created objects need to be recorded. */
if (ofd->ofd_osd->dd_record_fid_accessed) {
lfsck_pack_rfa(&ofd_info(env)->fti_lr,
lu_object_fid(&fo->ofo_obj.do_lu));
lfsck_in_notify(env, ofd->ofd_osd,
&ofd_info(env)->fti_lr);
}
if (likely(!ofd_object_exists(fo) &&
!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DANGLING))) {
next = ofd_object_child(fo);
LASSERT(next != NULL);
rc = dt_create(env, next, &info->fti_attr, NULL,
&info->fti_dof, th);
if (rc)
break;
LASSERT(ofd_object_exists(fo));
}
ofd_seq_last_oid_set(oseq, id + i);
}
objects = i;
/* NOT all the wanted objects have been created,
* set the LAST_ID as the real created. */
if (unlikely(objects < nr)) {
int rc1;
info->fti_off = 0;
tmp = cpu_to_le64(ofd_seq_last_oid(oseq));
dt_write_lock(env, oseq->os_lastid_obj, 0);
rc1 = dt_record_write(env, oseq->os_lastid_obj,
&info->fti_buf, &info->fti_off, th);
dt_write_unlock(env, oseq->os_lastid_obj);
if (rc1 != 0)
CERROR("%s: fail to reset the LAST_ID for seq ("LPX64
") from "LPU64" to "LPU64"\n", ofd_name(ofd),
ostid_seq(&oseq->os_oi), id + nr - 1,
ofd_seq_last_oid(oseq));
}
trans_stop:
ofd_trans_stop(env, ofd, th, rc);
out:
for (i = 0; i < nr_saved; i++) {
fo = batch[i];
if (fo) {
ofd_write_unlock(env, fo);
ofd_object_put(env, fo);
}
}
OBD_FREE(batch, nr_saved * sizeof(struct ofd_object *));
CDEBUG((objects == 0 && rc == 0) ? D_ERROR : D_OTHER,
"created %d/%d objects: %d\n", objects, nr_saved, rc);
LASSERT(ergo(objects == 0, rc < 0));
RETURN(objects > 0 ? objects : rc);
}
static int mdd_orphan_destroy(const struct lu_env *env, struct mdd_object *obj,
struct dt_key *key)
{
struct thandle *th = NULL;
struct mdd_device *mdd = mdo2mdd(&obj->mod_obj);
bool orphan_exists = true;
int rc = 0;
ENTRY;
th = mdd_trans_create(env, mdd);
if (IS_ERR(th)) {
rc = PTR_ERR(th);
if (rc != -EINPROGRESS)
CERROR("%s: cannot get orphan thandle: rc = %d\n",
mdd2obd_dev(mdd)->obd_name, rc);
RETURN(rc);
}
mdd_write_lock(env, obj, MOR_TGT_CHILD);
rc = mdd_orphan_declare_delete(env, obj, th);
if (rc == -ENOENT)
orphan_exists = false;
else if (rc)
GOTO(unlock, rc);
if (orphan_exists) {
rc = mdo_declare_destroy(env, obj, th);
if (rc)
GOTO(unlock, rc);
}
rc = mdd_trans_start(env, mdd, th);
if (rc)
GOTO(unlock, rc);
if (likely(obj->mod_count == 0)) {
dt_write_lock(env, mdd->mdd_orphans, MOR_TGT_ORPHAN);
rc = dt_delete(env, mdd->mdd_orphans, key, th);
if (rc) {
CERROR("%s: could not delete orphan "DFID": rc = %d\n",
mdd2obd_dev(mdd)->obd_name, PFID(mdo2fid(obj)),
rc);
} else if (orphan_exists) {
mdo_ref_del(env, obj, th);
if (S_ISDIR(mdd_object_type(obj))) {
mdo_ref_del(env, obj, th);
dt_ref_del(env, mdd->mdd_orphans, th);
}
rc = mdo_destroy(env, obj, th);
} else {
CWARN("%s: orphan %s "DFID" doesn't exist\n",
mdd2obd_dev(mdd)->obd_name, (char *)key,
PFID(mdo2fid(obj)));
}
dt_write_unlock(env, mdd->mdd_orphans);
}
unlock:
mdd_write_unlock(env, obj);
rc = mdd_trans_stop(env, mdd, 0, th);
RETURN(rc);
}
/**
* Implementation of the llog_operations::lop_destroy
*
* This function destroys the llog and deletes also entry in the
* llog directory in case of named llog. Llog should be opened prior that.
* Destroy method is not part of external transaction and does everything
* inside.
*
* \param[in] env execution environment
* \param[in] loghandle llog handle of the current llog
*
* \retval 0 on successful destroy
* \retval negative value on error
*/
static int llog_osd_destroy(const struct lu_env *env,
struct llog_handle *loghandle)
{
struct llog_ctxt *ctxt;
struct dt_object *o, *llog_dir = NULL;
struct dt_device *d;
struct thandle *th;
char *name = NULL;
int rc;
ENTRY;
ctxt = loghandle->lgh_ctxt;
LASSERT(ctxt);
o = loghandle->lgh_obj;
LASSERT(o);
d = lu2dt_dev(o->do_lu.lo_dev);
LASSERT(d);
LASSERT(d == ctxt->loc_exp->exp_obd->obd_lvfs_ctxt.dt);
th = dt_trans_create(env, d);
if (IS_ERR(th))
RETURN(PTR_ERR(th));
if (loghandle->lgh_name) {
llog_dir = llog_osd_dir_get(env, ctxt);
if (IS_ERR(llog_dir))
GOTO(out_trans, rc = PTR_ERR(llog_dir));
name = loghandle->lgh_name;
rc = dt_declare_delete(env, llog_dir,
(struct dt_key *)name, th);
if (rc)
GOTO(out_trans, rc);
}
dt_declare_ref_del(env, o, th);
rc = dt_declare_destroy(env, o, th);
if (rc)
GOTO(out_trans, rc);
rc = dt_trans_start_local(env, d, th);
if (rc)
GOTO(out_trans, rc);
dt_write_lock(env, o, 0);
if (dt_object_exists(o)) {
if (name) {
dt_read_lock(env, llog_dir, 0);
rc = dt_delete(env, llog_dir,
(struct dt_key *) name,
th, BYPASS_CAPA);
dt_read_unlock(env, llog_dir);
if (rc) {
CERROR("%s: can't remove llog %s: rc = %d\n",
o->do_lu.lo_dev->ld_obd->obd_name,
name, rc);
GOTO(out_unlock, rc);
}
}
dt_ref_del(env, o, th);
rc = dt_destroy(env, o, th);
if (rc)
GOTO(out_unlock, rc);
}
out_unlock:
dt_write_unlock(env, o);
out_trans:
dt_trans_stop(env, d, th);
if (llog_dir != NULL)
lu_object_put(env, &llog_dir->do_lu);
RETURN(rc);
}
/**
* Read the special file which contains the list of llog catalogs IDs
*
* This function reads the CATALOGS file which contains the array of llog
* catalogs IDs. The main purpose of this file is to store OSP llogs indexed
* by OST/MDT number.
*
* \param[in] env execution environment
* \param[in] d corresponding storage device
* \param[in] idx position to start from, usually OST/MDT index
* \param[in] count how many catalog IDs to read
* \param[out] idarray the buffer for the data. If it is NULL then
* function returns just number of catalog IDs
* in the file.
* \param[in] fid LLOG_CATALOGS_OID for CATALOG object
*
* \retval 0 on successful read of catalog IDs
* \retval negative value on error
* \retval positive value which is number of records in
* the file if \a idarray is NULL
*/
int llog_osd_get_cat_list(const struct lu_env *env, struct dt_device *d,
int idx, int count, struct llog_catid *idarray,
const struct lu_fid *fid)
{
struct llog_thread_info *lgi = llog_info(env);
struct dt_object *o = NULL;
struct thandle *th;
int rc, size;
ENTRY;
LASSERT(d);
size = sizeof(*idarray) * count;
lgi->lgi_off = idx * sizeof(*idarray);
lgi->lgi_fid = *fid;
o = dt_locate(env, d, &lgi->lgi_fid);
if (IS_ERR(o))
RETURN(PTR_ERR(o));
if (!dt_object_exists(o)) {
th = dt_trans_create(env, d);
if (IS_ERR(th))
GOTO(out, rc = PTR_ERR(th));
lgi->lgi_attr.la_valid = LA_MODE;
lgi->lgi_attr.la_mode = S_IFREG | S_IRUGO | S_IWUSR;
lgi->lgi_dof.dof_type = dt_mode_to_dft(S_IFREG);
rc = dt_declare_create(env, o, &lgi->lgi_attr, NULL,
&lgi->lgi_dof, th);
if (rc)
GOTO(out_trans, rc);
rc = dt_trans_start_local(env, d, th);
if (rc)
GOTO(out_trans, rc);
dt_write_lock(env, o, 0);
if (!dt_object_exists(o))
rc = dt_create(env, o, &lgi->lgi_attr, NULL,
&lgi->lgi_dof, th);
dt_write_unlock(env, o);
out_trans:
dt_trans_stop(env, d, th);
if (rc)
GOTO(out, rc);
}
rc = dt_attr_get(env, o, &lgi->lgi_attr, BYPASS_CAPA);
if (rc)
GOTO(out, rc);
if (!S_ISREG(lgi->lgi_attr.la_mode)) {
CERROR("%s: CATALOGS is not a regular file!: mode = %o\n",
o->do_lu.lo_dev->ld_obd->obd_name,
lgi->lgi_attr.la_mode);
GOTO(out, rc = -ENOENT);
}
CDEBUG(D_CONFIG, "cat list: disk size=%d, read=%d\n",
(int)lgi->lgi_attr.la_size, size);
/* return just number of llogs */
if (idarray == NULL) {
rc = lgi->lgi_attr.la_size / sizeof(*idarray);
GOTO(out, rc);
}
/* read for new ost index or for empty file */
memset(idarray, 0, size);
if (lgi->lgi_attr.la_size <= lgi->lgi_off)
GOTO(out, rc = 0);
if (lgi->lgi_attr.la_size < lgi->lgi_off + size)
size = lgi->lgi_attr.la_size - lgi->lgi_off;
lgi->lgi_buf.lb_buf = idarray;
lgi->lgi_buf.lb_len = size;
rc = dt_record_read(env, o, &lgi->lgi_buf, &lgi->lgi_off);
if (rc) {
CERROR("%s: error reading CATALOGS: rc = %d\n",
o->do_lu.lo_dev->ld_obd->obd_name, rc);
GOTO(out, rc);
}
EXIT;
out:
lu_object_put(env, &o->do_lu);
RETURN(rc);
}
static int llog_osd_write_blob(const struct lu_env *env, struct dt_object *o,
struct llog_rec_hdr *rec, void *buf,
loff_t *off, struct thandle *th)
{
struct llog_thread_info *lgi = llog_info(env);
int buflen = rec->lrh_len;
int rc;
LASSERT(env);
LASSERT(o);
if (buflen == 0)
CWARN("0-length record\n");
CDEBUG(D_OTHER, "write blob with type %x, buf %p/%u at off %llu\n",
rec->lrh_type, buf, buflen, *off);
lgi->lgi_attr.la_valid = LA_SIZE;
lgi->lgi_attr.la_size = *off;
if (!buf) {
lgi->lgi_buf.lb_len = buflen;
lgi->lgi_buf.lb_buf = rec;
rc = dt_record_write(env, o, &lgi->lgi_buf, off, th);
if (rc)
CERROR("%s: error writing log record: rc = %d\n",
o->do_lu.lo_dev->ld_obd->obd_name, rc);
GOTO(out, rc);
}
/* the buf case */
/* protect the following 3 writes from concurrent read */
dt_write_lock(env, o, 0);
rec->lrh_len = sizeof(*rec) + buflen + sizeof(lgi->lgi_tail);
lgi->lgi_buf.lb_len = sizeof(*rec);
lgi->lgi_buf.lb_buf = rec;
rc = dt_record_write(env, o, &lgi->lgi_buf, off, th);
if (rc) {
CERROR("%s: error writing log hdr: rc = %d\n",
o->do_lu.lo_dev->ld_obd->obd_name, rc);
GOTO(out_unlock, rc);
}
lgi->lgi_buf.lb_len = buflen;
lgi->lgi_buf.lb_buf = buf;
rc = dt_record_write(env, o, &lgi->lgi_buf, off, th);
if (rc) {
CERROR("%s: error writing log buffer: rc = %d\n",
o->do_lu.lo_dev->ld_obd->obd_name, rc);
GOTO(out_unlock, rc);
}
lgi->lgi_tail.lrt_len = rec->lrh_len;
lgi->lgi_tail.lrt_index = rec->lrh_index;
lgi->lgi_buf.lb_len = sizeof(lgi->lgi_tail);
lgi->lgi_buf.lb_buf = &lgi->lgi_tail;
rc = dt_record_write(env, o, &lgi->lgi_buf, off, th);
if (rc)
CERROR("%s: error writing log tail: rc = %d\n",
o->do_lu.lo_dev->ld_obd->obd_name, rc);
out_unlock:
dt_write_unlock(env, o);
out:
/* cleanup the content written above */
if (rc) {
dt_punch(env, o, lgi->lgi_attr.la_size, OBD_OBJECT_EOF, th,
BYPASS_CAPA);
dt_attr_set(env, o, &lgi->lgi_attr, th, BYPASS_CAPA);
}
return rc;
}