int lfsck_bookmark_setup(const struct lu_env *env,
struct lfsck_instance *lfsck)
{
struct dt_object *root;
struct dt_object *obj;
int rc;
ENTRY;
root = dt_locate(env, lfsck->li_bottom, &lfsck->li_local_root_fid);
if (IS_ERR(root))
RETURN(PTR_ERR(root));
dt_try_as_dir(env, root);
obj = local_file_find_or_create(env, lfsck->li_los, root,
lfsck_bookmark_name,
S_IFREG | S_IRUGO | S_IWUSR);
lu_object_put(env, &root->do_lu);
if (IS_ERR(obj))
RETURN(PTR_ERR(obj));
lfsck->li_bookmark_obj = obj;
rc = lfsck_bookmark_load(env, lfsck);
if (rc == -ENODATA)
rc = lfsck_bookmark_init(env, lfsck);
RETURN(rc);
}
/**
* This is helper function to get llog directory object. It is used by named
* llog operations to find/insert/delete llog entry from llog directory.
*
* \param[in] env execution environment
* \param[in] ctxt llog context
*
* \retval dt_object of llog directory
* \retval ERR_PTR of negative value on error
*/
struct dt_object *llog_osd_dir_get(const struct lu_env *env,
struct llog_ctxt *ctxt)
{
struct dt_device *dt;
struct dt_thread_info *dti = dt_info(env);
struct dt_object *dir;
int rc;
dt = ctxt->loc_exp->exp_obd->obd_lvfs_ctxt.dt;
if (ctxt->loc_dir == NULL) {
rc = dt_root_get(env, dt, &dti->dti_fid);
if (rc)
return ERR_PTR(rc);
dir = dt_locate(env, dt, &dti->dti_fid);
if (!IS_ERR(dir) && !dt_try_as_dir(env, dir)) {
lu_object_put(env, &dir->do_lu);
return ERR_PTR(-ENOTDIR);
}
} else {
lu_object_get(&ctxt->loc_dir->do_lu);
dir = ctxt->loc_dir;
}
return dir;
}
/**
* open the PENDING directory for device \a mdd
*
* The PENDING directory persistently tracks files and directories that were
* unlinked from the namespace (nlink == 0) but are still held open by clients.
* Those inodes shouldn't be deleted if the MDS crashes, because the clients
* would not be able to recover and reopen those files. Instead, these inodes
* are linked into the PENDING directory on disk, and only deleted if all
* clients close them, or the MDS finishes client recovery without any client
* reopening them (i.e. former clients didn't join recovery).
* \param d mdd device being started.
*
* \retval 0 success
* \retval -ve index operation error.
*
*/
int orph_index_init(const struct lu_env *env, struct mdd_device *mdd)
{
struct lu_fid fid;
struct dt_object *d;
int rc = 0;
ENTRY;
/* create PENDING dir */
fid_zero(&fid);
rc = mdd_local_file_create(env, mdd, &mdd->mdd_local_root_fid,
orph_index_name, S_IFDIR | S_IRUGO |
S_IWUSR | S_IXUGO, &fid);
if (rc < 0)
RETURN(rc);
d = dt_locate(env, mdd->mdd_child, &fid);
if (IS_ERR(d))
RETURN(PTR_ERR(d));
LASSERT(lu_object_exists(&d->do_lu));
if (!dt_try_as_dir(env, d)) {
CERROR("%s: \"%s\" is not an index: rc = %d\n",
mdd2obd_dev(mdd)->obd_name, orph_index_name, rc);
lu_object_put(env, &d->do_lu);
RETURN(-ENOTDIR);
}
mdd->mdd_orphans = d;
RETURN(0);
}
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;
}
static int __out_tx_index_delete(const struct lu_env *env,
struct dt_object *dt_obj, char *name,
struct thandle_exec_args *ta,
struct update_reply *reply,
int index, char *file, int line)
{
struct tx_arg *arg;
if (dt_try_as_dir(env, dt_obj) == 0) {
ta->ta_err = -ENOTDIR;
return ta->ta_err;
}
LASSERT(ta->ta_handle != NULL);
ta->ta_err = dt_declare_delete(env, dt_obj,
(struct dt_key *)name,
ta->ta_handle);
if (ta->ta_err != 0)
return ta->ta_err;
arg = tx_add_exec(ta, out_tx_index_delete_exec,
out_tx_index_delete_undo, file,
line);
LASSERT(arg);
lu_object_get(&dt_obj->do_lu);
arg->object = dt_obj;
arg->reply = reply;
arg->index = index;
arg->u.insert.key = (struct dt_key *)name;
return 0;
}
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_index_lookup(struct tgt_session_info *tsi)
{
const struct lu_env *env = tsi->tsi_env;
struct tgt_thread_info *tti = tgt_th_info(env);
struct object_update *update = tti->tti_u.update.tti_update;
struct dt_object *obj = tti->tti_u.update.tti_dt_object;
char *name;
int rc;
ENTRY;
if (unlikely(update->ou_result_size < sizeof(tti->tti_fid1)))
return -EPROTO;
if (!lu_object_exists(&obj->do_lu))
RETURN(-ENOENT);
name = object_update_param_get(update, 0, NULL);
if (IS_ERR(name)) {
CERROR("%s: empty name for lookup: rc = %ld\n",
tgt_name(tsi->tsi_tgt), PTR_ERR(name));
RETURN(PTR_ERR(name));
}
dt_read_lock(env, obj, MOR_TGT_CHILD);
if (!dt_try_as_dir(env, obj))
GOTO(out_unlock, rc = -ENOTDIR);
rc = dt_lookup(env, obj, (struct dt_rec *)&tti->tti_fid1,
(struct dt_key *)name);
if (rc < 0)
GOTO(out_unlock, rc);
if (rc == 0)
rc += 1;
out_unlock:
dt_read_unlock(env, obj);
CDEBUG(D_INFO, "lookup "DFID" %s get "DFID" rc %d\n",
PFID(lu_object_fid(&obj->do_lu)), name,
PFID(&tti->tti_fid1), rc);
CDEBUG(D_INFO, "%s: insert lookup reply %p index %d: rc = %d\n",
tgt_name(tsi->tsi_tgt), tti->tti_u.update.tti_update_reply,
0, rc);
object_update_result_insert(tti->tti_u.update.tti_update_reply,
&tti->tti_fid1, sizeof(tti->tti_fid1),
tti->tti_u.update.tti_update_reply_index, rc);
RETURN(rc);
}
static int out_index_lookup(struct tgt_session_info *tsi)
{
const struct lu_env *env = tsi->tsi_env;
struct tgt_thread_info *tti = tgt_th_info(env);
struct update *update = tti->tti_u.update.tti_update;
struct dt_object *obj = tti->tti_u.update.tti_dt_object;
char *name;
int rc;
ENTRY;
if (!lu_object_exists(&obj->do_lu))
RETURN(-ENOENT);
name = (char *)update_param_buf(update, 0, NULL);
if (name == NULL) {
CERROR("%s: empty name for lookup: rc = %d\n",
tgt_name(tsi->tsi_tgt), -EPROTO);
RETURN(err_serious(-EPROTO));
}
dt_read_lock(env, obj, MOR_TGT_CHILD);
if (!dt_try_as_dir(env, obj))
GOTO(out_unlock, rc = -ENOTDIR);
rc = dt_lookup(env, obj, (struct dt_rec *)&tti->tti_fid1,
(struct dt_key *)name, NULL);
if (rc < 0)
GOTO(out_unlock, rc);
if (rc == 0)
rc += 1;
CDEBUG(D_INFO, "lookup "DFID" %s get "DFID" rc %d\n",
PFID(lu_object_fid(&obj->do_lu)), name,
PFID(&tti->tti_fid1), rc);
fid_cpu_to_le(&tti->tti_fid1, &tti->tti_fid1);
out_unlock:
dt_read_unlock(env, obj);
CDEBUG(D_INFO, "%s: insert lookup reply %p index %d: rc = %d\n",
tgt_name(tsi->tsi_tgt), tti->tti_u.update.tti_update_reply,
0, rc);
update_insert_reply(tti->tti_u.update.tti_update_reply,
&tti->tti_fid1, sizeof(tti->tti_fid1), 0, rc);
RETURN(rc);
}
static int orph_index_insert(const struct lu_env *env,
struct mdd_object *obj,
__u32 op,
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);
int rc;
ENTRY;
LASSERT(mdd_write_locked(env, obj) != 0);
LASSERT(!(obj->mod_flags & ORPHAN_OBJ));
mdd_orphan_write_lock(env, mdd);
rc = mdd_orphan_insert_obj(env, mdd, obj, op, th);
if (rc)
GOTO(out, rc);
mdo_ref_add(env, obj, th);
if (!S_ISDIR(mdd_object_type(obj)))
goto out;
mdo_ref_add(env, obj, th);
mdd_orphan_ref_add(env, mdd, th);
/* try best to fixup directory, dont return errors
* from here */
if (!dt_try_as_dir(env, next))
goto out;
next->do_index_ops->dio_delete(env, next,
(const struct dt_key *)dotdot,
th, BYPASS_CAPA);
next->do_index_ops->dio_insert(env, next,
(struct dt_rec *)lf_dor,
(const struct dt_key *)dotdot,
th, BYPASS_CAPA, 1);
out:
if (rc == 0)
obj->mod_flags |= ORPHAN_OBJ;
mdd_orphan_write_unlock(env, mdd);
RETURN(rc);
}
/**
* 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);
}
static int out_index_lookup(struct mdt_thread_info *info)
{
struct update *update = info->mti_u.update.mti_update;
const struct lu_env *env = info->mti_env;
struct dt_object *obj = info->mti_u.update.mti_dt_object;
char *name;
int rc;
ENTRY;
if (!lu_object_exists(&obj->do_lu))
RETURN(-ENOENT);
name = (char *)update_param_buf(update, 0, NULL);
if (name == NULL) {
CERROR("%s: empty name for lookup: rc = %d\n",
mdt_obd_name(info->mti_mdt), -EPROTO);
RETURN(err_serious(-EPROTO));
}
dt_read_lock(env, obj, MOR_TGT_CHILD);
if (!dt_try_as_dir(env, obj))
GOTO(out_unlock, rc = -ENOTDIR);
rc = dt_lookup(env, obj, (struct dt_rec *)&info->mti_tmp_fid1,
(struct dt_key *)name, NULL);
if (rc < 0)
GOTO(out_unlock, rc);
if (rc == 0)
rc += 1;
CDEBUG(D_INFO, "lookup "DFID" %s get "DFID" rc %d\n",
PFID(lu_object_fid(&obj->do_lu)), name,
PFID(&info->mti_tmp_fid1), rc);
fid_cpu_to_le(&info->mti_tmp_fid1, &info->mti_tmp_fid1);
out_unlock:
dt_read_unlock(env, obj);
update_insert_reply(info->mti_u.update.mti_update_reply,
&info->mti_tmp_fid1, sizeof(info->mti_tmp_fid1),
0, rc);
RETURN(rc);
}
int lfsck_bookmark_setup(const struct lu_env *env,
struct lfsck_instance *lfsck)
{
struct dt_object *root;
struct dt_object *obj;
int rc;
ENTRY;
root = dt_locate(env, lfsck->li_bottom, &lfsck->li_local_root_fid);
if (IS_ERR(root))
RETURN(PTR_ERR(root));
if (unlikely(!dt_try_as_dir(env, root))) {
lfsck_object_put(env, root);
RETURN(-ENOTDIR);
}
obj = local_file_find_or_create(env, lfsck->li_los, root,
LFSCK_BOOKMARK,
S_IFREG | S_IRUGO | S_IWUSR);
lfsck_object_put(env, root);
if (IS_ERR(obj))
RETURN(PTR_ERR(obj));
lfsck->li_bookmark_obj = obj;
rc = lfsck_bookmark_load(env, lfsck);
if (rc == 0) {
struct lfsck_bookmark *mb = &lfsck->li_bookmark_ram;
/* It is upgraded from old release, set it as
* LFSCK_ASYNC_WIN_DEFAULT to avoid memory pressure. */
if (unlikely(mb->lb_async_windows == 0)) {
mb->lb_async_windows = LFSCK_ASYNC_WIN_DEFAULT;
mutex_lock(&lfsck->li_mutex);
rc = lfsck_bookmark_store(env, lfsck);
mutex_unlock(&lfsck->li_mutex);
}
} else if (rc == -ENODATA) {
rc = lfsck_bookmark_init(env, lfsck);
}
RETURN(rc);
}
/*
* To enable DNE functionality we need FID of /ROOT directory
* (which is / as seen by the clients) to belong to MDT0 and
* not to FID_SEQ_LOCAL_FILE or some other local sequence,
* which can be used by any node, so can't be part of FLDB.
*
* Pre-production code was using FID_SEQ_LOCAL_FILE for /ROOT
* making few existing setups incompatibile with DNE. This
* applies to ZFS-based setups only as ldiskfs-based setups
* are still using IGIF to identify /ROOT.
*
* The intention of this code is to fix on-disk state to use
* FID_SEQ_ROOT for /ROOT:
* - "." and ".." references in /ROOT itself and it`s subdirectories
* - LinkEA in all the objects listed in /ROOT
*
* Given only ZFS is affected where "." and ".." are not stored, we need to:
* - delete "." and ".." from /ROOT and its subdirectories
* - rename references in LinkEA in all the objects listed in /ROOT
*
* This code is subject for removal in 2.5
*/
static int mdd_convert_remove_dots(const struct lu_env *env,
struct mdd_device *mdd,
struct mdd_object *o)
{
struct thandle *th;
const struct dt_key *dot = (const struct dt_key *)".";
const struct dt_key *dotdot = (const struct dt_key *)"..";
int rc;
if (dt_try_as_dir(env, mdd_object_child(o)) == 0)
RETURN(-ENOTDIR);
/* remove "."/".." and do not insert them back - not stored in ZFS */
th = dt_trans_create(env, mdd->mdd_child);
if (IS_ERR(th))
RETURN(PTR_ERR(th));
rc = dt_declare_delete(env, mdd_object_child(o), dot, th);
if (rc)
GOTO(out, rc);
rc = dt_declare_delete(env, mdd_object_child(o), dotdot, th);
if (rc)
GOTO(out, rc);
rc = dt_trans_start_local(env, mdd->mdd_child, th);
if (rc)
GOTO(out, rc);
/* ignore non-existing "."/".." - we stored them on disk for
* pre-production systems, but this is not how regular ZFS works */
rc = dt_delete(env, mdd_object_child(o), dot, th, BYPASS_CAPA);
if (rc == -ENOENT)
rc = 0;
if (rc)
GOTO(out, rc);
rc = dt_delete(env, mdd_object_child(o), dotdot, th, BYPASS_CAPA);
if (rc == -ENOENT)
rc = 0;
if (rc)
GOTO(out, rc);
out:
dt_trans_stop(env, mdd->mdd_child, th);
RETURN(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_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;
}
static int __out_tx_index_insert(struct mdt_thread_info *info,
struct dt_object *dt_obj,
char *name, struct lu_fid *fid,
struct thandle_exec_args *th,
struct update_reply *reply,
int index, char *file, int line)
{
struct tx_arg *arg;
LASSERT(th->ta_handle != NULL);
if (lu_object_exists(&dt_obj->do_lu)) {
if (dt_try_as_dir(info->mti_env, dt_obj) == 0) {
th->ta_err = -ENOTDIR;
return th->ta_err;
}
th->ta_err = dt_declare_insert(info->mti_env, dt_obj,
(struct dt_rec *)fid,
(struct dt_key *)name,
th->ta_handle);
}
if (th->ta_err != 0)
return th->ta_err;
arg = tx_add_exec(th, out_tx_index_insert_exec,
out_tx_index_insert_undo, file,
line);
LASSERT(arg);
lu_object_get(&dt_obj->do_lu);
arg->object = dt_obj;
arg->reply = reply;
arg->index = index;
arg->u.insert.rec = (struct dt_rec *)fid;
arg->u.insert.key = (struct dt_key *)name;
return 0;
}
/**
* open the PENDING directory for device \a mdd
*
* The PENDING directory persistently tracks files and directories that were
* unlinked from the namespace (nlink == 0) but are still held open by clients.
* Those inodes shouldn't be deleted if the MDS crashes, because the clients
* would not be able to recover and reopen those files. Instead, these inodes
* are linked into the PENDING directory on disk, and only deleted if all
* clients close them, or the MDS finishes client recovery without any client
* reopening them (i.e. former clients didn't join recovery).
* \param d mdd device being started.
*
* \retval 0 success
* \retval -ve index operation error.
*
*/
int orph_index_init(const struct lu_env *env, struct mdd_device *mdd)
{
struct lu_fid fid;
struct dt_object *d;
int rc = 0;
ENTRY;
d = dt_store_open(env, mdd->mdd_child, "", orph_index_name, &fid);
if (!IS_ERR(d)) {
mdd->mdd_orphans = d;
if (!dt_try_as_dir(env, d)) {
rc = -ENOTDIR;
CERROR("\"%s\" is not an index! : rc = %d\n",
orph_index_name, rc);
}
} else {
CERROR("cannot find \"%s\" obj %d\n",
orph_index_name, (int)PTR_ERR(d));
rc = PTR_ERR(d);
}
RETURN(rc);
}
int mdd_compat_fixes(const struct lu_env *env, struct mdd_device *mdd)
{
struct mdd_thread_info *info = mdd_env_info(env);
struct mdd_object *root;
struct dt_object *o;
struct lustre_mdt_attrs *lma;
struct lu_buf buf;
int rc;
ENTRY;
/* IGIF FIDS are valid for old 1.8 and 2.[123] ROOT and are kept.
* Normal FIDs used by Xyratex 1.8->2.1 upgrade tool are also kept. */
if (fid_is_igif(&mdd->mdd_root_fid) || fid_is_norm(&mdd->mdd_root_fid))
RETURN(0);
/*
* FID is supposed to be FID_SEQ_ROOT for:
* - new ldiskfs fs
* - new ZFS fs
* - old ZFS fs, by now processed with osd_convert_root_to_new_seq()
*/
if (fid_seq(&mdd->mdd_root_fid) != FID_SEQ_ROOT) {
CERROR("%s: wrong FID "DFID" is used for /ROOT\n",
mdd2obd_dev(mdd)->obd_name,
PFID(&mdd->mdd_root_fid));
RETURN(-EINVAL);
}
root = mdd_object_find(env, mdd, &mdd->mdd_root_fid);
if (IS_ERR(root))
RETURN(PTR_ERR(root));
o = mdd_object_child(root);
CDEBUG(D_OTHER, "/ROOT = "DFID"\n", PFID(&mdd->mdd_root_fid));
if (dt_try_as_dir(env, o) == 0) {
CERROR("%s: not a directory\n", mdd2obd_dev(mdd)->obd_name);
GOTO(out, rc = -ENOTDIR);
}
lma = (struct lustre_mdt_attrs *)&info->mti_xattr_buf;
CLASSERT(sizeof(info->mti_xattr_buf) >= LMA_OLD_SIZE);
buf.lb_len = LMA_OLD_SIZE;
buf.lb_buf = lma;
rc = mdo_xattr_get(env, root, &buf, XATTR_NAME_LMA, BYPASS_CAPA);
if (rc < 0 && rc != -ENODATA) {
CERROR("%s: can't fetch LMA: rc = %d\n",
mdd2obd_dev(mdd)->obd_name, rc);
GOTO(out, rc);
}
lustre_lma_swab(lma);
if (lu_fid_eq(&lma->lma_self_fid, &mdd->mdd_root_fid)) {
/* /ROOT has been converted already
* or was correct from the beginning */
CDEBUG(D_OTHER, "%s: converted already\n",
mdd2obd_dev(mdd)->obd_name);
GOTO(out, rc = 0);
}
/* this is supposed to happen only on pre-production ZFS backend */
if (strcmp(mdd->mdd_bottom->dd_lu_dev.ld_type->ldt_name,
LUSTRE_OSD_ZFS_NAME) != 0) {
CERROR("%s: "DFID" is used on ldiskfs?!\n",
mdd2obd_dev(mdd)->obd_name, PFID(&mdd->mdd_root_fid));
GOTO(out, rc = -ENOTSUPP);
}
LCONSOLE_INFO("%s: FID of /ROOT has been changed. "
"Please remount the clients.\n",
mdd2obd_dev(mdd)->obd_name);
/* Fill FLDB first */
rc = mdd_fill_fldb(env, mdd);
if (rc)
GOTO(out, rc);
/* remove ./.. from /ROOT */
rc = mdd_convert_remove_dots(env, mdd, root);
if (rc)
GOTO(out, rc);
/* go over the directory, fix all the objects */
rc = mdd_fix_children(env, mdd, o);
if (rc)
GOTO(out, rc);
/* Update LMA on /ROOT. Done for simplicity in MDD, not in osd-zfs.
* Correct LMA will imply the whole directory has been coverted
* successfully, otherwise it will be retried on next mount. */
rc = mdd_convert_lma(env, mdd, root);
out:
mdd_object_put(env, root);
RETURN(rc);
}
static int out_attr_get(struct tgt_session_info *tsi)
{
const struct lu_env *env = tsi->tsi_env;
struct tgt_thread_info *tti = tgt_th_info(env);
struct obdo *obdo = &tti->tti_u.update.tti_obdo;
struct lu_attr *la = &tti->tti_attr;
struct dt_object *obj = tti->tti_u.update.tti_dt_object;
int rc;
ENTRY;
if (!lu_object_exists(&obj->do_lu))
RETURN(-ENOENT);
dt_read_lock(env, obj, MOR_TGT_CHILD);
rc = dt_attr_get(env, obj, la, NULL);
if (rc)
GOTO(out_unlock, rc);
/*
* If it is a directory, we will also check whether the
* directory is empty.
* la_flags = 0 : Empty.
* = 1 : Not empty.
*/
la->la_flags = 0;
if (S_ISDIR(la->la_mode)) {
struct dt_it *it;
const struct dt_it_ops *iops;
if (!dt_try_as_dir(env, obj))
GOTO(out_unlock, rc = -ENOTDIR);
iops = &obj->do_index_ops->dio_it;
it = iops->init(env, obj, LUDA_64BITHASH, BYPASS_CAPA);
if (!IS_ERR(it)) {
int result;
result = iops->get(env, it, (const void *)"");
if (result > 0) {
int i;
for (result = 0, i = 0; result == 0 && i < 3;
++i)
result = iops->next(env, it);
if (result == 0)
la->la_flags = 1;
} else if (result == 0)
/*
* Huh? Index contains no zero key?
*/
rc = -EIO;
iops->put(env, it);
iops->fini(env, it);
}
}
obdo->o_valid = 0;
obdo_from_la(obdo, la, la->la_valid);
obdo_cpu_to_le(obdo, obdo);
lustre_set_wire_obdo(NULL, obdo, obdo);
out_unlock:
dt_read_unlock(env, obj);
CDEBUG(D_INFO, "%s: insert attr get reply %p index %d: rc = %d\n",
tgt_name(tsi->tsi_tgt), tti->tti_u.update.tti_update_reply,
0, rc);
update_insert_reply(tti->tti_u.update.tti_update_reply, obdo,
sizeof(*obdo), 0, rc);
RETURN(rc);
}
/*
* Set up quota directory (either "quota_master" or "quota_slave") for a QMT or
* QSD instance. This function is also used to create per-pool directory on
* the quota master.
* The directory is created with a local sequence if it does not exist already.
* This function is called at ->ldo_prepare time when the full device stack is
* configured.
*
* \param env - is the environment passed by the caller
* \param dev - is the dt_device where to create the quota directory
* \param parent - is the parent directory. If not specified, the directory
* will be created under the root directory
* \param name - is the name of quota directory to be created
*
* \retval - pointer to quota root dt_object on success, appropriate error
* on failure
*/
struct dt_object *lquota_disk_dir_find_create(const struct lu_env *env,
struct dt_device *dev,
struct dt_object *parent,
const char *name)
{
struct lquota_thread_info *qti = lquota_info(env);
struct dt_object *qt_dir = NULL;
struct local_oid_storage *los = NULL;
int rc;
ENTRY;
/* Set up local storage to create the quota directory.
* We use the sequence reserved for local named objects */
lu_local_name_obj_fid(&qti->qti_fid, 1);
rc = local_oid_storage_init(env, dev, &qti->qti_fid, &los);
if (rc)
RETURN(ERR_PTR(rc));
if (parent == NULL) {
/* Fetch dt object associated with root directory */
rc = dt_root_get(env, dev, &qti->qti_fid);
if (rc)
GOTO(out, rc);
parent = dt_locate_at(env, dev, &qti->qti_fid,
dev->dd_lu_dev.ld_site->ls_top_dev, NULL);
if (IS_ERR(parent))
GOTO(out, rc = PTR_ERR(parent));
} else {
lu_object_get(&parent->do_lu);
}
/* create quota directory to be used for all quota index files */
qt_dir = local_file_find_or_create(env, los, parent, name, S_IFDIR |
S_IRUGO | S_IWUSR | S_IXUGO);
if (IS_ERR(qt_dir))
GOTO(out, rc = PTR_ERR(qt_dir));
/* local_oid_storage_fini() will finalize the local storage device,
* we have to open the object in another device stack */
qti->qti_fid = qt_dir->do_lu.lo_header->loh_fid;
lu_object_put_nocache(env, &qt_dir->do_lu);
qt_dir = dt_locate(env, dev, &qti->qti_fid);
if (IS_ERR(qt_dir))
GOTO(out, rc = PTR_ERR(qt_dir));
if (!dt_try_as_dir(env, qt_dir))
GOTO(out, rc = -ENOTDIR);
EXIT;
out:
if (parent != NULL && !IS_ERR(parent))
lu_object_put(env, &parent->do_lu);
if (los != NULL)
local_oid_storage_fini(env, los);
if (rc) {
if (qt_dir != NULL && !IS_ERR(qt_dir))
lu_object_put(env, &qt_dir->do_lu);
qt_dir = ERR_PTR(rc);
}
return qt_dir;
}
/**
* \retval +ve repaired
* \retval 0 no need to repair
* \retval -ve error cases
*/
static int lfsck_namespace_double_scan_one(const struct lu_env *env,
struct lfsck_component *com,
struct dt_object *child, __u8 flags)
{
struct lfsck_thread_info *info = lfsck_env_info(env);
struct lu_attr *la = &info->lti_la;
struct lu_name *cname = &info->lti_name;
struct lu_fid *pfid = &info->lti_fid;
struct lu_fid *cfid = &info->lti_fid2;
struct lfsck_instance *lfsck = com->lc_lfsck;
struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
struct lfsck_namespace *ns = com->lc_file_ram;
struct linkea_data ldata = { 0 };
struct thandle *handle = NULL;
bool locked = false;
bool update = false;
int rc;
ENTRY;
if (com->lc_journal) {
again:
LASSERT(!locked);
update = false;
com->lc_journal = 1;
handle = dt_trans_create(env, lfsck->li_next);
if (IS_ERR(handle))
RETURN(rc = PTR_ERR(handle));
rc = dt_declare_xattr_set(env, child,
lfsck_buf_get_const(env, NULL, DEFAULT_LINKEA_SIZE),
XATTR_NAME_LINK, 0, handle);
if (rc != 0)
GOTO(stop, rc);
rc = dt_trans_start(env, lfsck->li_next, handle);
if (rc != 0)
GOTO(stop, rc);
dt_write_lock(env, child, MOR_TGT_CHILD);
locked = true;
}
if (unlikely(lfsck_is_dead_obj(child)))
GOTO(stop, rc = 0);
rc = dt_attr_get(env, child, la, BYPASS_CAPA);
if (rc == 0)
rc = lfsck_links_read(env, child, &ldata);
if (rc != 0) {
if ((bk->lb_param & LPF_DRYRUN) &&
(rc == -EINVAL || rc == -ENODATA))
rc = 1;
GOTO(stop, rc);
}
linkea_first_entry(&ldata);
while (ldata.ld_lee != NULL) {
struct dt_object *parent = NULL;
rc = lfsck_linkea_entry_unpack(lfsck, &ldata, cname, pfid);
if (rc > 0)
update = true;
if (!fid_is_sane(pfid))
goto shrink;
parent = lfsck_object_find(env, lfsck, pfid);
if (parent == NULL)
goto shrink;
else if (IS_ERR(parent))
GOTO(stop, rc = PTR_ERR(parent));
if (!dt_object_exists(parent))
goto shrink;
/* XXX: Currently, skip remote object, the consistency for
* remote object will be processed in LFSCK phase III. */
if (dt_object_remote(parent)) {
lfsck_object_put(env, parent);
linkea_next_entry(&ldata);
continue;
}
if (unlikely(!dt_try_as_dir(env, parent)))
goto shrink;
/* To guarantee the 'name' is terminated with '0'. */
memcpy(info->lti_key, cname->ln_name, cname->ln_namelen);
info->lti_key[cname->ln_namelen] = 0;
cname->ln_name = info->lti_key;
rc = dt_lookup(env, parent, (struct dt_rec *)cfid,
(const struct dt_key *)cname->ln_name,
BYPASS_CAPA);
if (rc != 0 && rc != -ENOENT) {
lfsck_object_put(env, parent);
GOTO(stop, rc);
}
if (rc == 0) {
if (lu_fid_eq(cfid, lfsck_dto2fid(child))) {
lfsck_object_put(env, parent);
linkea_next_entry(&ldata);
continue;
}
goto shrink;
}
/* If there is no name entry in the parent dir and the object
* link count is less than the linkea entries count, then the
* linkea entry should be removed. */
if (ldata.ld_leh->leh_reccount > la->la_nlink)
goto shrink;
/* XXX: For the case of there is a linkea entry, but without
* name entry pointing to the object and its hard links
* count is not less than the object name entries count,
* then seems we should add the 'missed' name entry back
* to namespace, but before LFSCK phase III finished, we
* do not know whether the object has some inconsistency
* on other MDTs. So now, do NOT add the name entry back
* to the namespace, but keep the linkEA entry. LU-2914 */
lfsck_object_put(env, parent);
linkea_next_entry(&ldata);
continue;
shrink:
if (parent != NULL)
lfsck_object_put(env, parent);
if (bk->lb_param & LPF_DRYRUN)
RETURN(1);
CDEBUG(D_LFSCK, "Remove linkEA: "DFID"[%.*s], "DFID"\n",
PFID(lfsck_dto2fid(child)), cname->ln_namelen, cname->ln_name,
PFID(pfid));
linkea_del_buf(&ldata, cname);
update = true;
}
if (update) {
if (!com->lc_journal) {
com->lc_journal = 1;
goto again;
}
rc = lfsck_links_write(env, child, &ldata, handle);
}
GOTO(stop, rc);
stop:
if (locked) {
/* XXX: For the case linkea entries count does not match the object hard
* links count, we cannot update the later one simply. Before LFSCK
* phase III finished, we cannot know whether there are some remote
* name entries to be repaired or not. LU-2914 */
if (rc == 0 && !lfsck_is_dead_obj(child) &&
ldata.ld_leh != NULL &&
ldata.ld_leh->leh_reccount != la->la_nlink)
CWARN("%s: the object "DFID" linkEA entry count %u "
"may not match its hardlink count %u\n",
lfsck_lfsck2name(lfsck), PFID(cfid),
ldata.ld_leh->leh_reccount, la->la_nlink);
dt_write_unlock(env, child);
}
if (handle != NULL)
dt_trans_stop(env, lfsck->li_next, handle);
if (rc == 0 && update) {
ns->ln_objs_nlink_repaired++;
rc = 1;
}
return rc;
}
static int lfsck_namespace_reset(const struct lu_env *env,
struct lfsck_component *com, bool init)
{
struct lfsck_instance *lfsck = com->lc_lfsck;
struct lfsck_namespace *ns = com->lc_file_ram;
struct dt_object *root;
struct dt_object *dto;
int rc;
ENTRY;
root = dt_locate(env, lfsck->li_bottom, &lfsck->li_local_root_fid);
if (IS_ERR(root))
RETURN(PTR_ERR(root));
if (unlikely(!dt_try_as_dir(env, root))) {
lu_object_put(env, &root->do_lu);
RETURN(-ENOTDIR);
}
down_write(&com->lc_sem);
if (init) {
memset(ns, 0, sizeof(*ns));
} else {
__u32 count = ns->ln_success_count;
__u64 last_time = ns->ln_time_last_complete;
memset(ns, 0, sizeof(*ns));
ns->ln_success_count = count;
ns->ln_time_last_complete = last_time;
}
ns->ln_magic = LFSCK_NAMESPACE_MAGIC;
ns->ln_status = LS_INIT;
rc = local_object_unlink(env, lfsck->li_bottom, root,
lfsck_namespace_name);
if (rc != 0)
GOTO(out, rc);
lfsck_object_put(env, com->lc_obj);
com->lc_obj = NULL;
dto = local_index_find_or_create(env, lfsck->li_los, root,
lfsck_namespace_name,
S_IFREG | S_IRUGO | S_IWUSR,
&dt_lfsck_features);
if (IS_ERR(dto))
GOTO(out, rc = PTR_ERR(dto));
com->lc_obj = dto;
rc = dto->do_ops->do_index_try(env, dto, &dt_lfsck_features);
if (rc != 0)
GOTO(out, rc);
rc = lfsck_namespace_store(env, com, true);
GOTO(out, rc);
out:
up_write(&com->lc_sem);
lu_object_put(env, &root->do_lu);
return rc;
}
static int out_attr_get(struct tgt_session_info *tsi)
{
const struct lu_env *env = tsi->tsi_env;
struct tgt_thread_info *tti = tgt_th_info(env);
struct obdo *obdo = &tti->tti_u.update.tti_obdo;
struct lu_attr *la = &tti->tti_attr;
struct dt_object *obj = tti->tti_u.update.tti_dt_object;
int idx = tti->tti_u.update.tti_update_reply_index;
int rc;
ENTRY;
if (!lu_object_exists(&obj->do_lu)) {
/* Usually, this will be called when the master MDT try
* to init a remote object(see osp_object_init), so if
* the object does not exist on slave, we need set BANSHEE flag,
* so the object can be removed from the cache immediately */
set_bit(LU_OBJECT_HEARD_BANSHEE,
&obj->do_lu.lo_header->loh_flags);
RETURN(-ENOENT);
}
dt_read_lock(env, obj, MOR_TGT_CHILD);
rc = dt_attr_get(env, obj, la, NULL);
if (rc)
GOTO(out_unlock, rc);
/*
* If it is a directory, we will also check whether the
* directory is empty.
* la_flags = 0 : Empty.
* = 1 : Not empty.
*/
la->la_flags = 0;
if (S_ISDIR(la->la_mode)) {
struct dt_it *it;
const struct dt_it_ops *iops;
if (!dt_try_as_dir(env, obj))
GOTO(out_unlock, rc = -ENOTDIR);
iops = &obj->do_index_ops->dio_it;
it = iops->init(env, obj, LUDA_64BITHASH, BYPASS_CAPA);
if (!IS_ERR(it)) {
int result;
result = iops->get(env, it, (const void *)"");
if (result > 0) {
int i;
for (result = 0, i = 0; result == 0 && i < 3;
++i)
result = iops->next(env, it);
if (result == 0)
la->la_flags = 1;
} else if (result == 0)
/*
* Huh? Index contains no zero key?
*/
rc = -EIO;
iops->put(env, it);
iops->fini(env, it);
}
}
obdo->o_valid = 0;
obdo_from_la(obdo, la, la->la_valid);
obdo_cpu_to_le(obdo, obdo);
lustre_set_wire_obdo(NULL, obdo, obdo);
out_unlock:
dt_read_unlock(env, obj);
CDEBUG(D_INFO, "%s: insert attr get reply %p index %d: rc = %d\n",
tgt_name(tsi->tsi_tgt), tti->tti_u.update.tti_update_reply,
0, rc);
object_update_result_insert(tti->tti_u.update.tti_update_reply, obdo,
sizeof(*obdo), idx, rc);
RETURN(rc);
}
int lfsck_namespace_setup(const struct lu_env *env,
struct lfsck_instance *lfsck)
{
struct lfsck_component *com;
struct lfsck_namespace *ns;
struct dt_object *root = NULL;
struct dt_object *obj;
int rc;
ENTRY;
LASSERT(lfsck->li_master);
OBD_ALLOC_PTR(com);
if (com == NULL)
RETURN(-ENOMEM);
CFS_INIT_LIST_HEAD(&com->lc_link);
CFS_INIT_LIST_HEAD(&com->lc_link_dir);
init_rwsem(&com->lc_sem);
atomic_set(&com->lc_ref, 1);
com->lc_lfsck = lfsck;
com->lc_type = LT_NAMESPACE;
com->lc_ops = &lfsck_namespace_ops;
com->lc_file_size = sizeof(struct lfsck_namespace);
OBD_ALLOC(com->lc_file_ram, com->lc_file_size);
if (com->lc_file_ram == NULL)
GOTO(out, rc = -ENOMEM);
OBD_ALLOC(com->lc_file_disk, com->lc_file_size);
if (com->lc_file_disk == NULL)
GOTO(out, rc = -ENOMEM);
root = dt_locate(env, lfsck->li_bottom, &lfsck->li_local_root_fid);
if (IS_ERR(root))
GOTO(out, rc = PTR_ERR(root));
if (unlikely(!dt_try_as_dir(env, root)))
GOTO(out, rc = -ENOTDIR);
obj = local_index_find_or_create(env, lfsck->li_los, root,
lfsck_namespace_name,
S_IFREG | S_IRUGO | S_IWUSR,
&dt_lfsck_features);
if (IS_ERR(obj))
GOTO(out, rc = PTR_ERR(obj));
com->lc_obj = obj;
rc = obj->do_ops->do_index_try(env, obj, &dt_lfsck_features);
if (rc != 0)
GOTO(out, rc);
rc = lfsck_namespace_load(env, com);
if (rc > 0)
rc = lfsck_namespace_reset(env, com, true);
else if (rc == -ENODATA)
rc = lfsck_namespace_init(env, com);
if (rc != 0)
GOTO(out, rc);
ns = com->lc_file_ram;
switch (ns->ln_status) {
case LS_INIT:
case LS_COMPLETED:
case LS_FAILED:
case LS_STOPPED:
spin_lock(&lfsck->li_lock);
cfs_list_add_tail(&com->lc_link, &lfsck->li_list_idle);
spin_unlock(&lfsck->li_lock);
break;
default:
CERROR("%s: unknown lfsck_namespace status: rc = %u\n",
lfsck_lfsck2name(lfsck), ns->ln_status);
/* fall through */
case LS_SCANNING_PHASE1:
case LS_SCANNING_PHASE2:
/* No need to store the status to disk right now.
* If the system crashed before the status stored,
* it will be loaded back when next time. */
ns->ln_status = LS_CRASHED;
/* fall through */
case LS_PAUSED:
case LS_CRASHED:
spin_lock(&lfsck->li_lock);
cfs_list_add_tail(&com->lc_link, &lfsck->li_list_scan);
cfs_list_add_tail(&com->lc_link_dir, &lfsck->li_list_dir);
spin_unlock(&lfsck->li_lock);
break;
}
GOTO(out, rc = 0);
out:
if (root != NULL && !IS_ERR(root))
lu_object_put(env, &root->do_lu);
if (rc != 0)
lfsck_component_cleanup(env, com);
return rc;
}