static int fld_insert_entry(const struct lu_env *env,
struct lu_server_fld *fld,
const struct lu_seq_range *range)
{
struct thandle *th;
int rc;
ENTRY;
th = dt_trans_create(env, lu2dt_dev(fld->lsf_obj->do_lu.lo_dev));
if (IS_ERR(th))
RETURN(PTR_ERR(th));
rc = fld_declare_index_create(env, fld, range, th);
if (rc != 0) {
if (rc == -EEXIST)
rc = 0;
GOTO(out, rc);
}
rc = dt_trans_start_local(env, lu2dt_dev(fld->lsf_obj->do_lu.lo_dev),
th);
if (rc)
GOTO(out, rc);
rc = fld_index_create(env, fld, range, th);
if (rc == -EEXIST)
rc = 0;
out:
dt_trans_stop(env, lu2dt_dev(fld->lsf_obj->do_lu.lo_dev), th);
RETURN(rc);
}
/**
* Write the special file which contains the list of llog catalogs IDs
*
* This function writes the CATALOG file which contains the array of llog
* catalogs IDs. It is used mostly 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 write
* \param[out] idarray the buffer with the data to write.
* \param[in] fid LLOG_CATALOGS_OID for CATALOG object
*
* \retval 0 on successful write of catalog IDs
* \retval negative value on error
*/
int llog_osd_put_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;
if (count == 0)
RETURN(0);
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))
GOTO(out, rc = -ENOENT);
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);
}
th = dt_trans_create(env, d);
if (IS_ERR(th))
GOTO(out, rc = PTR_ERR(th));
lgi->lgi_buf.lb_len = size;
lgi->lgi_buf.lb_buf = idarray;
rc = dt_declare_record_write(env, o, &lgi->lgi_buf, lgi->lgi_off, th);
if (rc)
GOTO(out, rc);
rc = dt_trans_start_local(env, d, th);
if (rc)
GOTO(out_trans, rc);
rc = dt_record_write(env, o, &lgi->lgi_buf, &lgi->lgi_off, th);
if (rc)
CDEBUG(D_INODE, "can't write CATALOGS at index %d: rc = %d\n",
idx, rc);
out_trans:
dt_trans_stop(env, d, th);
out:
lu_object_put(env, &o->do_lu);
RETURN(rc);
}
/**
* Helper function to open llog or create it if doesn't exist.
* It hides all transaction handling from caller.
*/
int llog_open_create(const struct lu_env *env, struct llog_ctxt *ctxt,
struct llog_handle **res, struct llog_logid *logid,
char *name)
{
struct dt_device *d;
struct thandle *th;
int rc;
rc = llog_open(env, ctxt, res, logid, name, LLOG_OPEN_NEW);
if (rc)
return rc;
if (llog_exist(*res))
return 0;
LASSERT((*res)->lgh_obj != NULL);
d = lu2dt_dev((*res)->lgh_obj->do_lu.lo_dev);
th = dt_trans_create(env, d);
if (IS_ERR(th))
GOTO(out, rc = PTR_ERR(th));
rc = llog_declare_create(env, *res, th);
if (rc == 0) {
rc = dt_trans_start_local(env, d, th);
if (rc == 0)
rc = llog_create(env, *res, th);
}
dt_trans_stop(env, d, th);
out:
if (rc)
llog_close(env, *res);
return rc;
}
/*
* Helper function for write record in llog.
* It hides all transaction handling from caller.
* Valid only with local llog.
*/
int llog_write(const struct lu_env *env, struct llog_handle *loghandle,
struct llog_rec_hdr *rec, int idx)
{
struct dt_device *dt;
struct thandle *th;
int rc;
ENTRY;
LASSERT(loghandle);
LASSERT(loghandle->lgh_ctxt);
LASSERT(loghandle->lgh_obj != NULL);
dt = lu2dt_dev(loghandle->lgh_obj->do_lu.lo_dev);
th = dt_trans_create(env, dt);
if (IS_ERR(th))
RETURN(PTR_ERR(th));
rc = llog_declare_write_rec(env, loghandle, rec, idx, th);
if (rc)
GOTO(out_trans, rc);
th->th_wait_submit = 1;
rc = dt_trans_start_local(env, dt, th);
if (rc)
GOTO(out_trans, rc);
down_write(&loghandle->lgh_lock);
rc = llog_write_rec(env, loghandle, rec, NULL, idx, th);
up_write(&loghandle->lgh_lock);
out_trans:
dt_trans_stop(env, dt, th);
RETURN(rc);
}
/*
* Helper function for write record in llog.
* It hides all transaction handling from caller.
* Valid only with local llog.
*/
int llog_write(const struct lu_env *env, struct llog_handle *loghandle,
struct llog_rec_hdr *rec, struct llog_cookie *reccookie,
int cookiecount, void *buf, int idx)
{
struct dt_device *dt;
struct thandle *th;
int rc;
LASSERT(loghandle);
LASSERT(loghandle->lgh_ctxt);
LASSERT(loghandle->lgh_obj != NULL);
dt = lu2dt_dev(loghandle->lgh_obj->do_lu.lo_dev);
th = dt_trans_create(env, dt);
if (IS_ERR(th))
return PTR_ERR(th);
rc = llog_declare_write_rec(env, loghandle, rec, idx, th);
if (rc)
GOTO(out_trans, rc);
rc = dt_trans_start_local(env, dt, th);
if (rc)
GOTO(out_trans, rc);
down_write(&loghandle->lgh_lock);
rc = llog_write_rec(env, loghandle, rec, reccookie,
cookiecount, buf, idx, th);
up_write(&loghandle->lgh_lock);
out_trans:
dt_trans_stop(env, dt, th);
return rc;
}
static int mdd_convert_linkea(const struct lu_env *env,
struct mdd_device *mdd,
struct mdd_object *o,
const struct lu_name *name)
{
struct thandle *th;
struct lu_fid oldfid;
int rc;
ENTRY;
th = dt_trans_create(env, mdd->mdd_child);
if (IS_ERR(th))
RETURN(PTR_ERR(th));
rc = mdd_declare_links_add(env, o, th, NULL);
if (rc)
GOTO(out, rc);
rc = dt_trans_start_local(env, mdd->mdd_child, th);
if (rc)
GOTO(out, rc);
oldfid.f_seq = FID_SEQ_LOCAL_FILE;
oldfid.f_oid = MDD_ROOT_INDEX_OID;
oldfid.f_ver = 0;
rc = mdd_links_rename(env, o, &oldfid, name, &mdd->mdd_root_fid,
name, th, NULL, 0, 1);
if (rc == -ENOENT || rc == -EEXIST)
rc = 0;
out:
dt_trans_stop(env, mdd->mdd_child, th);
RETURN(rc);
}
static int mdd_convert_lma(const struct lu_env *env, struct mdd_device *mdd,
struct mdd_object *o)
{
struct lustre_mdt_attrs *lma;
struct thandle *th;
struct lu_fid fid;
struct lu_buf buf;
int rc;
ENTRY;
lu_root_fid(&fid);
lma = (struct lustre_mdt_attrs *)&mdd_env_info(env)->mti_xattr_buf;
lustre_lma_init(lma, &fid, 0, 0);
lustre_lma_swab(lma);
buf.lb_buf = lma;
buf.lb_len = sizeof(*lma);
th = dt_trans_create(env, mdd->mdd_child);
if (IS_ERR(th))
RETURN(PTR_ERR(th));
rc = mdo_declare_xattr_set(env, o, &buf, XATTR_NAME_LMA, 0, th);
if (rc)
GOTO(out, rc);
rc = dt_trans_start_local(env, mdd->mdd_child, th);
if (rc)
GOTO(out, rc);
rc = mdo_xattr_set(env, o, &buf, XATTR_NAME_LMA, 0, th, BYPASS_CAPA);
out:
dt_trans_stop(env, mdd->mdd_child, th);
RETURN(rc);
}
/**
* Write fid into last_oid/last_seq file.
**/
int osp_write_last_oid_seq_files(struct lu_env *env, struct osp_device *osp,
struct lu_fid *fid, int sync)
{
struct osp_thread_info *oti = osp_env_info(env);
struct lu_buf *lb_oid = &oti->osi_lb;
struct lu_buf *lb_oseq = &oti->osi_lb2;
loff_t oid_off;
loff_t oseq_off;
struct thandle *th;
int rc;
ENTRY;
/* Note: through f_oid is only 32bits, it will also write
* 64 bits for oid to keep compatiblity with the previous
* version. */
lb_oid->lb_buf = &fid->f_oid;
lb_oid->lb_len = sizeof(obd_id);
oid_off = sizeof(obd_id) * osp->opd_index;
lb_oseq->lb_buf = &fid->f_seq;
lb_oseq->lb_len = sizeof(obd_id);
oseq_off = sizeof(obd_id) * osp->opd_index;
th = dt_trans_create(env, osp->opd_storage);
if (IS_ERR(th))
RETURN(PTR_ERR(th));
th->th_sync |= sync;
rc = dt_declare_record_write(env, osp->opd_last_used_oid_file,
lb_oid->lb_len, oid_off, th);
if (rc != 0)
GOTO(out, rc);
rc = dt_declare_record_write(env, osp->opd_last_used_seq_file,
lb_oseq->lb_len, oseq_off, th);
if (rc != 0)
GOTO(out, rc);
rc = dt_trans_start_local(env, osp->opd_storage, th);
if (rc != 0)
GOTO(out, rc);
rc = dt_record_write(env, osp->opd_last_used_oid_file, lb_oid,
&oid_off, th);
if (rc != 0) {
CERROR("%s: can not write to last seq file: rc = %d\n",
osp->opd_obd->obd_name, rc);
GOTO(out, rc);
}
rc = dt_record_write(env, osp->opd_last_used_seq_file, lb_oseq,
&oseq_off, th);
if (rc) {
CERROR("%s: can not write to last seq file: rc = %d\n",
osp->opd_obd->obd_name, rc);
GOTO(out, rc);
}
out:
dt_trans_stop(env, osp->opd_storage, th);
RETURN(rc);
}
/*
* Update version of an index file
*
* \param env - is the environment passed by the caller
* \param dev - is the backend dt device storing the index file
* \param obj - is the on-disk index that should be updated
* \param ver - is the new version
*/
int lquota_disk_update_ver(const struct lu_env *env, struct dt_device *dev,
struct dt_object *obj, __u64 ver)
{
struct thandle *th;
int rc;
ENTRY;
th = dt_trans_create(env, dev);
if (IS_ERR(th))
RETURN(PTR_ERR(th));
rc = dt_declare_version_set(env, obj, th);
if (rc)
GOTO(out, rc);
rc = dt_trans_start_local(env, dev, th);
if (rc)
GOTO(out, rc);
th->th_sync = 1;
dt_version_set(env, obj, ver, th);
EXIT;
out:
dt_trans_stop(env, dev, th);
return rc;
}
static int lfsck_namespace_delete(const struct lu_env *env,
struct lfsck_component *com,
const struct lu_fid *fid)
{
struct lfsck_instance *lfsck = com->lc_lfsck;
struct lu_fid *key = &lfsck_env_info(env)->lti_fid;
struct thandle *handle;
struct dt_object *obj = com->lc_obj;
int rc;
ENTRY;
handle = dt_trans_create(env, lfsck->li_bottom);
if (IS_ERR(handle))
RETURN(PTR_ERR(handle));
rc = dt_declare_delete(env, obj, (const struct dt_key *)fid, handle);
if (rc != 0)
GOTO(out, rc);
rc = dt_trans_start_local(env, lfsck->li_bottom, handle);
if (rc != 0)
GOTO(out, rc);
fid_cpu_to_be(key, fid);
rc = dt_delete(env, obj, (const struct dt_key *)key, handle,
BYPASS_CAPA);
GOTO(out, rc);
out:
dt_trans_stop(env, lfsck->li_bottom, handle);
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);
}
static int llog_truncate(const struct lu_env *env, struct dt_object *o)
{
struct lu_attr la;
struct thandle *th;
struct dt_device *d;
int rc;
ENTRY;
LASSERT(o);
d = lu2dt_dev(o->do_lu.lo_dev);
LASSERT(d);
rc = dt_attr_get(env, o, &la);
if (rc)
RETURN(rc);
CDEBUG(D_OTHER, "original size "LPU64"\n", la.la_size);
rc = sizeof(struct llog_log_hdr) + sizeof(struct llog_mini_rec);
if (la.la_size < rc) {
CERROR("too small llog: "LPU64"\n", la.la_size);
RETURN(0);
}
/* drop 2 records */
la.la_size = la.la_size - (sizeof(struct llog_mini_rec) * 2);
la.la_valid = LA_SIZE;
th = dt_trans_create(env, d);
if (IS_ERR(th))
RETURN(PTR_ERR(th));
rc = dt_declare_attr_set(env, o, &la, th);
if (rc)
GOTO(stop, rc);
rc = dt_declare_punch(env, o, la.la_size, OBD_OBJECT_EOF, th);
rc = dt_trans_start_local(env, d, th);
if (rc)
GOTO(stop, rc);
rc = dt_punch(env, o, la.la_size, OBD_OBJECT_EOF, th);
if (rc)
GOTO(stop, rc);
rc = dt_attr_set(env, o, &la, th);
if (rc)
GOTO(stop, rc);
stop:
dt_trans_stop(env, d, th);
RETURN(rc);
}
static int lfsck_namespace_store(const struct lu_env *env,
struct lfsck_component *com, bool init)
{
struct dt_object *obj = com->lc_obj;
struct lfsck_instance *lfsck = com->lc_lfsck;
struct thandle *handle;
int len = com->lc_file_size;
int rc;
ENTRY;
lfsck_namespace_cpu_to_le((struct lfsck_namespace *)com->lc_file_disk,
(struct lfsck_namespace *)com->lc_file_ram);
handle = dt_trans_create(env, lfsck->li_bottom);
if (IS_ERR(handle)) {
rc = PTR_ERR(handle);
CERROR("%s: fail to create trans for storing lfsck_namespace: "
"rc = %d\n", lfsck_lfsck2name(lfsck), rc);
RETURN(rc);
}
rc = dt_declare_xattr_set(env, obj,
lfsck_buf_get(env, com->lc_file_disk, len),
XATTR_NAME_LFSCK_NAMESPACE, 0, handle);
if (rc != 0) {
CERROR("%s: fail to declare trans for storing lfsck_namespace: "
"rc = %d\n", lfsck_lfsck2name(lfsck), rc);
GOTO(out, rc);
}
rc = dt_trans_start_local(env, lfsck->li_bottom, handle);
if (rc != 0) {
CERROR("%s: fail to start trans for storing lfsck_namespace: "
"rc = %d\n", lfsck_lfsck2name(lfsck), rc);
GOTO(out, rc);
}
rc = dt_xattr_set(env, obj,
lfsck_buf_get(env, com->lc_file_disk, len),
XATTR_NAME_LFSCK_NAMESPACE,
init ? LU_XATTR_CREATE : LU_XATTR_REPLACE,
handle, BYPASS_CAPA);
if (rc != 0)
CERROR("%s: fail to store lfsck_namespace: len = %d, "
"rc = %d\n", lfsck_lfsck2name(lfsck), len, rc);
GOTO(out, rc);
out:
dt_trans_stop(env, lfsck->li_bottom, handle);
return rc;
}
int lfsck_bookmark_store(const struct lu_env *env, struct lfsck_instance *lfsck)
{
struct thandle *handle;
struct dt_object *obj = lfsck->li_bookmark_obj;
loff_t pos = 0;
int len = sizeof(struct lfsck_bookmark);
int rc;
ENTRY;
lfsck_bookmark_cpu_to_le(&lfsck->li_bookmark_disk,
&lfsck->li_bookmark_ram);
handle = dt_trans_create(env, lfsck->li_bottom);
if (IS_ERR(handle)) {
rc = PTR_ERR(handle);
CERROR("%s: fail to create trans for storing lfsck_bookmark: "
"rc = %d\n", lfsck_lfsck2name(lfsck), rc);
RETURN(rc);
}
rc = dt_declare_record_write(env, obj,
lfsck_buf_get(env,
&lfsck->li_bookmark_disk, len),
0, handle);
if (rc != 0) {
CERROR("%s: fail to declare trans for storing lfsck_bookmark: "
"rc = %d\n", lfsck_lfsck2name(lfsck), rc);
GOTO(out, rc);
}
rc = dt_trans_start_local(env, lfsck->li_bottom, handle);
if (rc != 0) {
CERROR("%s: fail to start trans for storing lfsck_bookmark: "
"rc = %d\n", lfsck_lfsck2name(lfsck), rc);
GOTO(out, rc);
}
rc = dt_record_write(env, obj,
lfsck_buf_get(env, &lfsck->li_bookmark_disk, len),
&pos, handle);
if (rc != 0)
CERROR("%s: fail to store lfsck_bookmark: expected = %d, "
"rc = %d\n", lfsck_lfsck2name(lfsck), len, rc);
GOTO(out, rc);
out:
dt_trans_stop(env, lfsck->li_bottom, handle);
return rc;
}
/**
* Helper function to open llog or create it if doesn't exist.
* It hides all transaction handling from caller.
*/
int llog_open_create(const struct lu_env *env, struct llog_ctxt *ctxt,
struct llog_handle **res, struct llog_logid *logid,
char *name)
{
struct dt_device *d;
struct thandle *th;
int rc;
ENTRY;
rc = llog_open(env, ctxt, res, logid, name, LLOG_OPEN_NEW);
if (rc)
RETURN(rc);
if (llog_exist(*res))
RETURN(0);
LASSERT((*res)->lgh_obj != NULL);
d = lu2dt_dev((*res)->lgh_obj->do_lu.lo_dev);
th = dt_trans_create(env, d);
if (IS_ERR(th))
GOTO(out, rc = PTR_ERR(th));
/* Create update llog object synchronously, which
* happens during inialization process see
* lod_sub_prep_llog(), to make sure the update
* llog object is created before corss-MDT writing
* updates into the llog object */
if (ctxt->loc_flags & LLOG_CTXT_FLAG_NORMAL_FID)
th->th_sync = 1;
th->th_wait_submit = 1;
rc = llog_declare_create(env, *res, th);
if (rc == 0) {
rc = dt_trans_start_local(env, d, th);
if (rc == 0)
rc = llog_create(env, *res, th);
}
dt_trans_stop(env, d, th);
out:
if (rc)
llog_close(env, *res);
RETURN(rc);
}
int llog_destroy(const struct lu_env *env, struct llog_handle *handle)
{
struct llog_operations *lop;
struct dt_device *dt;
struct thandle *th;
int rc;
ENTRY;
rc = llog_handle2ops(handle, &lop);
if (rc < 0)
RETURN(rc);
if (lop->lop_destroy == NULL)
RETURN(-EOPNOTSUPP);
if (handle->lgh_obj == NULL) {
/* if lgh_obj == NULL, then it is from client side destroy */
rc = lop->lop_destroy(env, handle, NULL);
RETURN(rc);
}
if (!dt_object_exists(handle->lgh_obj))
RETURN(0);
dt = lu2dt_dev(handle->lgh_obj->do_lu.lo_dev);
th = dt_trans_create(env, dt);
if (IS_ERR(th))
RETURN(PTR_ERR(th));
rc = llog_declare_destroy(env, handle, th);
if (rc != 0)
GOTO(out_trans, rc);
rc = dt_trans_start_local(env, dt, th);
if (rc < 0)
GOTO(out_trans, rc);
rc = lop->lop_destroy(env, handle, th);
out_trans:
dt_trans_stop(env, dt, th);
RETURN(rc);
}
static int write_capa_keys(const struct lu_env *env,
struct mdt_device *mdt,
struct lustre_capa_key *keys)
{
struct mdt_thread_info *mti;
struct lustre_capa_key *tmp;
struct thandle *th;
loff_t off = 0;
int i, rc;
mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
th = dt_trans_create(env, mdt->mdt_bottom);
if (IS_ERR(th))
RETURN(PTR_ERR(th));
rc = dt_declare_record_write(env, mdt->mdt_ck_obj,
mdt_buf_const(env, NULL,
sizeof(*tmp) * 3), 0, th);
if (rc)
goto stop;
rc = dt_trans_start_local(env, mdt->mdt_bottom, th);
if (rc)
goto stop;
tmp = &mti->mti_capa_key;
for (i = 0; i < 2; i++) {
lck_cpu_to_le(tmp, &keys[i]);
rc = dt_record_write(env, mdt->mdt_ck_obj,
mdt_buf_const(env, tmp, sizeof(*tmp)),
&off, th);
if (rc)
break;
}
stop:
dt_trans_stop(env, mdt->mdt_bottom, th);
CDEBUG(D_INFO, "write capability keys rc = %d:\n", rc);
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);
}
int lfsck_bookmark_store(const struct lu_env *env, struct lfsck_instance *lfsck)
{
struct thandle *handle;
struct dt_object *obj = lfsck->li_bookmark_obj;
struct dt_device *dev = lfsck_obj2dev(obj);
loff_t pos = 0;
int len = sizeof(struct lfsck_bookmark);
int rc;
ENTRY;
lfsck_bookmark_cpu_to_le(&lfsck->li_bookmark_disk,
&lfsck->li_bookmark_ram);
handle = dt_trans_create(env, dev);
if (IS_ERR(handle))
GOTO(log, rc = PTR_ERR(handle));
rc = dt_declare_record_write(env, obj,
lfsck_buf_get(env,
&lfsck->li_bookmark_disk, len),
0, handle);
if (rc != 0)
GOTO(out, rc);
rc = dt_trans_start_local(env, dev, handle);
if (rc != 0)
GOTO(out, rc);
rc = dt_record_write(env, obj,
lfsck_buf_get(env, &lfsck->li_bookmark_disk, len),
&pos, handle);
GOTO(out, rc);
out:
dt_trans_stop(env, dev, handle);
log:
if (rc != 0)
CDEBUG(D_LFSCK, "%s: fail to store lfsck_bookmark: rc = %d\n",
lfsck_lfsck2name(lfsck), rc);
return rc;
}
int ofd_object_destroy(const struct lu_env *env, struct ofd_object *fo,
int orphan)
{
struct ofd_device *ofd = ofd_obj2dev(fo);
struct thandle *th;
int rc = 0;
ENTRY;
ofd_write_lock(env, fo);
if (!ofd_object_exists(fo))
GOTO(unlock, rc = -ENOENT);
th = ofd_trans_create(env, ofd);
if (IS_ERR(th))
GOTO(unlock, rc = PTR_ERR(th));
dt_declare_ref_del(env, ofd_object_child(fo), th);
dt_declare_destroy(env, ofd_object_child(fo), th);
if (orphan)
rc = dt_trans_start_local(env, ofd->ofd_osd, th);
else
rc = ofd_trans_start(env, ofd, NULL, th);
if (rc)
GOTO(stop, rc);
ofd_fmd_drop(ofd_info(env)->fti_exp, &fo->ofo_header.loh_fid);
dt_ref_del(env, ofd_object_child(fo), th);
dt_destroy(env, ofd_object_child(fo), th);
stop:
ofd_trans_stop(env, ofd, th, rc);
unlock:
ofd_write_unlock(env, fo);
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);
}
/*
* Update slave or global index copy.
*
* \param env - the environment passed by the caller
* \param qqi - is the qsd_type_info structure managing the index to be
* update
* \param qid - is the identifier for which we need to update the quota
* settings
* \param global - is set to true when updating the global index copy and to
* false for the slave index copy.
* \param ver - is the new version of the index. If equal to 0, the version
* of the index isn't changed
* \param rec - is the updated record to insert in the index file
*/
int qsd_update_index(const struct lu_env *env, struct qsd_qtype_info *qqi,
union lquota_id *qid, bool global, __u64 ver, void *rec)
{
struct thandle *th = NULL;
struct dt_object *obj;
__u64 *new_verp = NULL;
int flags = 0;
int rc;
ENTRY;
obj = global ? qqi->qqi_glb_obj : qqi->qqi_slv_obj;
/* allocate transaction */
th = dt_trans_create(env, qqi->qqi_qsd->qsd_dev);
if (IS_ERR(th))
RETURN(PTR_ERR(th));
/* reserve enough credits to update record in index file */
rc = lquota_disk_declare_write(env, th, obj, qid);
if (rc)
GOTO(out, rc);
/* start local transaction */
rc = dt_trans_start_local(env, qqi->qqi_qsd->qsd_dev, th);
if (rc)
GOTO(out, rc);
if (global) {
/* Update record in global index copy */
struct lquota_glb_rec *glb_rec = (struct lquota_glb_rec *)rec;
CDEBUG(D_QUOTA, "%s: updating global index hardlimit: "LPU64", "
"softlimit: "LPU64" for id "LPU64"\n",
qqi->qqi_qsd->qsd_svname, glb_rec->qbr_hardlimit,
glb_rec->qbr_softlimit, qid->qid_uid);
} else {
/* Update record in slave index copy */
struct lquota_slv_rec *slv_rec = (struct lquota_slv_rec *)rec;
CDEBUG(D_QUOTA, "%s: update granted to "LPU64" for id "LPU64
"\n", qqi->qqi_qsd->qsd_svname, slv_rec->qsr_granted,
qid->qid_uid);
}
if (ver != 0) {
new_verp = &ver;
flags = LQUOTA_SET_VER;
}
/* write new record to index file */
rc = lquota_disk_write(env, th, obj, qid, (struct dt_rec *)rec, flags,
new_verp);
EXIT;
out:
dt_trans_stop(env, qqi->qqi_qsd->qsd_dev, th);
if (rc)
CERROR("%s: failed to update %s index copy for id "LPU64", rc:"
"%d\n", qqi->qqi_qsd->qsd_svname,
global ? "global" : "slave", qid->qid_uid, rc);
else if (flags == LQUOTA_SET_VER)
qsd_bump_version(qqi, ver, global);
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);
}
/**
* Set attributes of object during write bulk IO processing.
*
* Change object attributes and write parent FID into extended
* attributes when needed.
*
* \param[in] env execution environment
* \param[in] ofd OFD device
* \param[in] ofd_obj OFD object
* \param[in] la object attributes
* \param[in] ff parent FID
*
* \retval 0 on successful attributes update
* \retval negative value on error
*/
static int
ofd_write_attr_set(const struct lu_env *env, struct ofd_device *ofd,
struct ofd_object *ofd_obj, struct lu_attr *la,
struct filter_fid *ff)
{
struct ofd_thread_info *info = ofd_info(env);
__u64 valid = la->la_valid;
int rc;
struct thandle *th;
struct dt_object *dt_obj;
int ff_needed = 0;
ENTRY;
LASSERT(la);
dt_obj = ofd_object_child(ofd_obj);
LASSERT(dt_obj != NULL);
la->la_valid &= LA_UID | LA_GID;
rc = ofd_attr_handle_ugid(env, ofd_obj, la, 0 /* !is_setattr */);
if (rc != 0)
GOTO(out, rc);
if (ff != NULL) {
rc = ofd_object_ff_load(env, ofd_obj);
if (rc == -ENODATA)
ff_needed = 1;
else if (rc < 0)
GOTO(out, rc);
}
if (!la->la_valid && !ff_needed)
/* no attributes to set */
GOTO(out, rc = 0);
th = ofd_trans_create(env, ofd);
if (IS_ERR(th))
GOTO(out, rc = PTR_ERR(th));
if (la->la_valid) {
rc = dt_declare_attr_set(env, dt_obj, la, th);
if (rc)
GOTO(out_tx, rc);
}
if (ff_needed) {
if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_UNMATCHED_PAIR1))
ff->ff_parent.f_oid = cpu_to_le32(1UL << 31);
if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_UNMATCHED_PAIR2))
ff->ff_parent.f_oid =
cpu_to_le32(le32_to_cpu(ff->ff_parent.f_oid) - 1);
info->fti_buf.lb_buf = ff;
info->fti_buf.lb_len = sizeof(*ff);
rc = dt_declare_xattr_set(env, dt_obj, &info->fti_buf,
XATTR_NAME_FID, 0, th);
if (rc)
GOTO(out_tx, rc);
}
/* We don't need a transno for this operation which will be re-executed
* anyway when the OST_WRITE (with a transno assigned) is replayed */
rc = dt_trans_start_local(env, ofd->ofd_osd , th);
if (rc)
GOTO(out_tx, rc);
/* set uid/gid */
if (la->la_valid) {
rc = dt_attr_set(env, dt_obj, la, th);
if (rc)
GOTO(out_tx, rc);
}
/* set filter fid EA */
if (ff_needed) {
if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_NOPFID))
GOTO(out_tx, rc);
rc = dt_xattr_set(env, dt_obj, &info->fti_buf, XATTR_NAME_FID,
0, th);
if (rc == 0) {
ofd_obj->ofo_pfid.f_seq = le64_to_cpu(ff->ff_parent.f_seq);
ofd_obj->ofo_pfid.f_oid = le32_to_cpu(ff->ff_parent.f_oid);
/* Currently, the filter_fid::ff_parent::f_ver is not
* the real parent MDT-object's FID::f_ver, instead it
* is the OST-object index in its parent MDT-object's
* layout EA. */
ofd_obj->ofo_pfid.f_stripe_idx =
le32_to_cpu(ff->ff_parent.f_stripe_idx);
}
}
GOTO(out_tx, rc);
out_tx:
dt_trans_stop(env, ofd->ofd_osd, th);
out:
la->la_valid = valid;
return rc;
}
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);
}
/* returns negative on error; 0 if success; 1 if success & log destroyed */
int llog_cancel_rec(const struct lu_env *env, struct llog_handle *loghandle,
int index)
{
struct llog_thread_info *lgi = llog_info(env);
struct dt_device *dt;
struct llog_log_hdr *llh = loghandle->lgh_hdr;
struct thandle *th;
int rc;
int rc1;
bool subtract_count = false;
ENTRY;
CDEBUG(D_RPCTRACE, "Canceling %d in log "DOSTID"\n", index,
POSTID(&loghandle->lgh_id.lgl_oi));
if (index == 0) {
CERROR("Can't cancel index 0 which is header\n");
RETURN(-EINVAL);
}
LASSERT(loghandle != NULL);
LASSERT(loghandle->lgh_ctxt != NULL);
LASSERT(loghandle->lgh_obj != NULL);
dt = lu2dt_dev(loghandle->lgh_obj->do_lu.lo_dev);
th = dt_trans_create(env, dt);
if (IS_ERR(th))
RETURN(PTR_ERR(th));
rc = llog_declare_write_rec(env, loghandle, &llh->llh_hdr, index, th);
if (rc < 0)
GOTO(out_trans, rc);
if ((llh->llh_flags & LLOG_F_ZAP_WHEN_EMPTY))
rc = llog_declare_destroy(env, loghandle, th);
th->th_wait_submit = 1;
rc = dt_trans_start_local(env, dt, th);
if (rc < 0)
GOTO(out_trans, rc);
down_write(&loghandle->lgh_lock);
/* clear bitmap */
mutex_lock(&loghandle->lgh_hdr_mutex);
if (!ext2_clear_bit(index, LLOG_HDR_BITMAP(llh))) {
CDEBUG(D_RPCTRACE, "Catalog index %u already clear?\n", index);
GOTO(out_unlock, rc);
}
loghandle->lgh_hdr->llh_count--;
subtract_count = true;
/* Pass this index to llog_osd_write_rec(), which will use the index
* to only update the necesary bitmap. */
lgi->lgi_cookie.lgc_index = index;
/* update header */
rc = llog_write_rec(env, loghandle, &llh->llh_hdr, &lgi->lgi_cookie,
LLOG_HEADER_IDX, th);
if (rc != 0)
GOTO(out_unlock, rc);
if ((llh->llh_flags & LLOG_F_ZAP_WHEN_EMPTY) &&
(llh->llh_count == 1) &&
((loghandle->lgh_last_idx == LLOG_HDR_BITMAP_SIZE(llh) - 1) ||
(loghandle->u.phd.phd_cat_handle != NULL &&
loghandle->u.phd.phd_cat_handle->u.chd.chd_current_log !=
loghandle))) {
/* never try to destroy it again */
llh->llh_flags &= ~LLOG_F_ZAP_WHEN_EMPTY;
rc = llog_trans_destroy(env, loghandle, th);
if (rc < 0) {
/* Sigh, can not destroy the final plain llog, but
* the bitmap has been clearly, so the record can not
* be accessed anymore, let's return 0 for now, and
* the orphan will be handled by LFSCK. */
CERROR("%s: can't destroy empty llog #"DOSTID
"#%08x: rc = %d\n",
loghandle->lgh_ctxt->loc_obd->obd_name,
POSTID(&loghandle->lgh_id.lgl_oi),
loghandle->lgh_id.lgl_ogen, rc);
GOTO(out_unlock, rc);
}
rc = LLOG_DEL_PLAIN;
}
out_unlock:
mutex_unlock(&loghandle->lgh_hdr_mutex);
up_write(&loghandle->lgh_lock);
out_trans:
rc1 = dt_trans_stop(env, dt, th);
if (rc == 0)
rc = rc1;
if (rc < 0 && subtract_count) {
mutex_lock(&loghandle->lgh_hdr_mutex);
loghandle->lgh_hdr->llh_count++;
ext2_set_bit(index, LLOG_HDR_BITMAP(llh));
mutex_unlock(&loghandle->lgh_hdr_mutex);
}
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);
}
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;
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;
/* prepare objects */
ostid_set_seq(&info->fti_ostid, ostid_seq(&oseq->os_oi));
for (i = 0; i < nr; i++) {
ostid_set_id(&info->fti_ostid, id + i);
rc = ostid_to_fid(&info->fti_fid, &info->fti_ostid, 0);
if (rc) {
if (i == 0)
GOTO(out, rc);
nr = i;
break;
}
fo = ofd_object_find(env, ofd, &info->fti_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, sizeof(tmp),
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(&info->fti_fid));
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(&info->fti_fid), nr);
for (i = 0; i < nr; i++) {
fo = batch[i];
LASSERT(fo);
if (likely(!ofd_object_exists(fo))) {
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;
if (objects > 0) {
tmp = cpu_to_le64(ofd_seq_last_oid(oseq));
rc = dt_record_write(env, oseq->os_lastid_obj,
&info->fti_buf, &info->fti_off, th);
}
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);
}
/**
* Set attributes of object during write bulk IO processing.
*
* Change object attributes and write parent FID into extended
* attributes when needed.
*
* \param[in] env execution environment
* \param[in] ofd OFD device
* \param[in] ofd_obj OFD object
* \param[in] la object attributes
* \param[in] oa obdo
*
* \retval 0 on successful attributes update
* \retval negative value on error
*/
static int
ofd_write_attr_set(const struct lu_env *env, struct ofd_device *ofd,
struct ofd_object *ofd_obj, struct lu_attr *la,
struct obdo *oa)
{
struct ofd_thread_info *info = ofd_info(env);
struct filter_fid *ff = &info->fti_mds_fid;
__u64 valid = la->la_valid;
struct thandle *th;
struct dt_object *dt_obj;
int fl = 0;
int rc;
ENTRY;
LASSERT(la);
dt_obj = ofd_object_child(ofd_obj);
LASSERT(dt_obj != NULL);
la->la_valid &= LA_UID | LA_GID | LA_PROJID;
rc = ofd_attr_handle_id(env, ofd_obj, la, 0 /* !is_setattr */);
if (rc != 0)
GOTO(out, rc);
fl = ofd_object_ff_update(env, ofd_obj, oa, ff);
if (fl < 0)
GOTO(out, rc = fl);
if (!la->la_valid && !fl)
/* no attributes to set */
GOTO(out, rc = 0);
th = ofd_trans_create(env, ofd);
if (IS_ERR(th))
GOTO(out, rc = PTR_ERR(th));
if (la->la_valid) {
rc = dt_declare_attr_set(env, dt_obj, la, th);
if (rc)
GOTO(out_tx, rc);
}
if (fl) {
if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_UNMATCHED_PAIR1))
ff->ff_parent.f_oid = cpu_to_le32(1UL << 31);
else if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_UNMATCHED_PAIR2))
le32_add_cpu(&ff->ff_parent.f_oid, -1);
rc = dt_declare_xattr_set(env, dt_obj, &info->fti_buf,
XATTR_NAME_FID, 0, th);
if (rc)
GOTO(out_tx, rc);
}
/* We don't need a transno for this operation which will be re-executed
* anyway when the OST_WRITE (with a transno assigned) is replayed */
rc = dt_trans_start_local(env, ofd->ofd_osd , th);
if (rc)
GOTO(out_tx, rc);
/* set uid/gid/projid */
if (la->la_valid) {
rc = dt_attr_set(env, dt_obj, la, th);
if (rc)
GOTO(out_tx, rc);
}
/* set filter fid EA.
* FIXME: it holds read lock of ofd object to modify the XATTR_NAME_FID
* while the write lock should be held. However, it should work because
* write RPCs only modify ff_{parent,layout} and those information will
* be the same from all the write RPCs. The reason that fl is not used
* in dt_xattr_set() is to allow this race. */
if (fl) {
if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_NOPFID))
GOTO(out_tx, rc);
info->fti_buf.lb_buf = ff;
info->fti_buf.lb_len = sizeof(*ff);
rc = dt_xattr_set(env, dt_obj, &info->fti_buf, XATTR_NAME_FID,
0, th);
if (rc == 0)
filter_fid_le_to_cpu(&ofd_obj->ofo_ff, ff, sizeof(*ff));
}
GOTO(out_tx, rc);
out_tx:
dt_trans_stop(env, ofd->ofd_osd, th);
out:
la->la_valid = valid;
return rc;
}
static int lfsck_namespace_update(const struct lu_env *env,
struct lfsck_component *com,
const struct lu_fid *fid,
__u8 flags, bool force)
{
struct lfsck_instance *lfsck = com->lc_lfsck;
struct lu_fid *key = &lfsck_env_info(env)->lti_fid;
struct thandle *handle;
struct dt_object *obj = com->lc_obj;
int rc;
bool exist = false;
__u8 tf;
ENTRY;
rc = lfsck_namespace_lookup(env, com, fid, &tf);
if (rc != 0 && rc != -ENOENT)
RETURN(rc);
if (rc == 0) {
if (!force || flags == tf)
RETURN(0);
exist = true;
handle = dt_trans_create(env, lfsck->li_bottom);
if (IS_ERR(handle))
RETURN(PTR_ERR(handle));
rc = dt_declare_delete(env, obj, (const struct dt_key *)fid,
handle);
if (rc != 0)
GOTO(out, rc);
} else {
handle = dt_trans_create(env, lfsck->li_bottom);
if (IS_ERR(handle))
RETURN(PTR_ERR(handle));
}
rc = dt_declare_insert(env, obj, (const struct dt_rec *)&flags,
(const struct dt_key *)fid, handle);
if (rc != 0)
GOTO(out, rc);
rc = dt_trans_start_local(env, lfsck->li_bottom, handle);
if (rc != 0)
GOTO(out, rc);
fid_cpu_to_be(key, fid);
if (exist) {
rc = dt_delete(env, obj, (const struct dt_key *)key, handle,
BYPASS_CAPA);
if (rc != 0) {
CERROR("%s: fail to insert "DFID": rc = %d\n",
lfsck_lfsck2name(com->lc_lfsck), PFID(fid), rc);
GOTO(out, rc);
}
}
rc = dt_insert(env, obj, (const struct dt_rec *)&flags,
(const struct dt_key *)key, handle, BYPASS_CAPA, 1);
GOTO(out, rc);
out:
dt_trans_stop(env, lfsck->li_bottom, handle);
return rc;
}
