/**
* 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_xattr_get(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 lu_buf *lbuf = &tti->tti_buf;
struct object_update_reply *reply = tti->tti_u.update.tti_update_reply;
struct dt_object *obj = tti->tti_u.update.tti_dt_object;
char *name;
struct object_update_result *update_result;
int idx = tti->tti_u.update.tti_update_reply_index;
int rc;
ENTRY;
if (!lu_object_exists(&obj->do_lu)) {
set_bit(LU_OBJECT_HEARD_BANSHEE,
&obj->do_lu.lo_header->loh_flags);
RETURN(-ENOENT);
}
name = object_update_param_get(update, 0, NULL);
if (IS_ERR(name)) {
CERROR("%s: empty name for xattr get: rc = %ld\n",
tgt_name(tsi->tsi_tgt), PTR_ERR(name));
RETURN(PTR_ERR(name));
}
update_result = object_update_result_get(reply, 0, NULL);
if (update_result == NULL) {
CERROR("%s: empty name for xattr get: rc = %d\n",
tgt_name(tsi->tsi_tgt), -EPROTO);
RETURN(-EPROTO);
}
lbuf->lb_len = (int)tti->tti_u.update.tti_update->ou_result_size;
lbuf->lb_buf = update_result->our_data;
if (lbuf->lb_len == 0)
lbuf->lb_buf = 0;
dt_read_lock(env, obj, MOR_TGT_CHILD);
rc = dt_xattr_get(env, obj, lbuf, name);
dt_read_unlock(env, obj);
if (rc < 0)
lbuf->lb_len = 0;
CDEBUG(D_INFO, "%s: "DFID" get xattr %s len %d\n",
tgt_name(tsi->tsi_tgt), PFID(lu_object_fid(&obj->do_lu)),
name, (int)lbuf->lb_len);
GOTO(out, rc);
out:
object_update_result_insert(reply, lbuf->lb_buf, lbuf->lb_len, idx, rc);
RETURN(0);
}
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 out_xattr_get(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 lu_buf *lbuf = &tti->tti_buf;
struct update_reply *reply = tti->tti_u.update.tti_update_reply;
struct dt_object *obj = tti->tti_u.update.tti_dt_object;
char *name;
void *ptr;
int rc;
ENTRY;
name = (char *)update_param_buf(update, 0, NULL);
if (name == NULL) {
CERROR("%s: empty name for xattr get: rc = %d\n",
tgt_name(tsi->tsi_tgt), -EPROTO);
RETURN(err_serious(-EPROTO));
}
ptr = update_get_buf_internal(reply, 0, NULL);
LASSERT(ptr != NULL);
/* The first 4 bytes(int) are used to store the result */
lbuf->lb_buf = (char *)ptr + sizeof(int);
lbuf->lb_len = UPDATE_BUFFER_SIZE - sizeof(struct update_reply);
dt_read_lock(env, obj, MOR_TGT_CHILD);
rc = dt_xattr_get(env, obj, lbuf, name, NULL);
dt_read_unlock(env, obj);
if (rc < 0) {
lbuf->lb_len = 0;
GOTO(out, rc);
}
if (rc == 0) {
lbuf->lb_len = 0;
GOTO(out, rc = -ENOENT);
}
lbuf->lb_len = rc;
rc = 0;
CDEBUG(D_INFO, "%s: "DFID" get xattr %s len %d\n",
tgt_name(tsi->tsi_tgt), PFID(lu_object_fid(&obj->do_lu)),
name, (int)lbuf->lb_len);
out:
*(int *)ptr = rc;
reply->ur_lens[0] = lbuf->lb_len + sizeof(int);
RETURN(rc);
}
/*
* Retrieve quota settings from disk for a particular identifier.
*
* \param env - is the environment passed by the caller
* \param obj - is the on-disk index where quota settings are stored.
* \param id - is the key to be updated
* \param rec - is the output record where to store quota settings.
*
* \retval - 0 on success, appropriate error on failure
*/
int lquota_disk_read(const struct lu_env *env, struct dt_object *obj,
union lquota_id *id, struct dt_rec *rec)
{
int rc;
ENTRY;
LASSERT(dt_object_exists(obj));
LASSERT(obj->do_index_ops != NULL);
/* lookup on-disk record from index file */
dt_read_lock(env, obj, 0);
rc = dt_lookup(env, obj, rec, (struct dt_key *)&id->qid_uid);
dt_read_unlock(env, obj);
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 object_update *update = tti->tti_u.update.tti_update;
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 (unlikely(update->ou_result_size < sizeof(*obdo)))
return -EPROTO;
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);
if (rc)
GOTO(out_unlock, rc);
obdo->o_valid = 0;
obdo_from_la(obdo, la, la->la_valid);
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);
}
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);
}
/**
* Implementation of the llog_operations::lop_open
*
* This function opens the llog by its logid or by name, it may open also
* non existent llog and assing then new id to it.
* The llog_open/llog_close pair works similar to lu_object_find/put,
* the object may not exist prior open. The result of open is just dt_object
* in the llog header.
*
* \param[in] env execution environment
* \param[in] handle llog handle of the current llog
* \param[in] logid logid of llog to open (nameless llog)
* \param[in] name name of llog to open (named llog)
* \param[in] open_param
* LLOG_OPEN_NEW - new llog, may not exist
* LLOG_OPEN_EXIST - old llog, must exist
*
* \retval 0 on successful open, llog_handle::lgh_obj
* contains the dt_object of the llog.
* \retval negative value on error
*/
static int llog_osd_open(const struct lu_env *env, struct llog_handle *handle,
struct llog_logid *logid, char *name,
enum llog_open_param open_param)
{
struct llog_thread_info *lgi = llog_info(env);
struct llog_ctxt *ctxt = handle->lgh_ctxt;
struct dt_object *o;
struct dt_device *dt;
struct ls_device *ls;
struct local_oid_storage *los;
int rc = 0;
ENTRY;
LASSERT(env);
LASSERT(ctxt);
LASSERT(ctxt->loc_exp);
LASSERT(ctxt->loc_exp->exp_obd);
dt = ctxt->loc_exp->exp_obd->obd_lvfs_ctxt.dt;
LASSERT(dt);
ls = ls_device_get(dt);
if (IS_ERR(ls))
RETURN(PTR_ERR(ls));
mutex_lock(&ls->ls_los_mutex);
los = dt_los_find(ls, name != NULL ? FID_SEQ_LLOG_NAME : FID_SEQ_LLOG);
mutex_unlock(&ls->ls_los_mutex);
LASSERT(los);
ls_device_put(env, ls);
LASSERT(handle);
if (logid != NULL) {
logid_to_fid(logid, &lgi->lgi_fid);
} else if (name) {
struct dt_object *llog_dir;
llog_dir = llog_osd_dir_get(env, ctxt);
if (IS_ERR(llog_dir))
GOTO(out, rc = PTR_ERR(llog_dir));
dt_read_lock(env, llog_dir, 0);
rc = dt_lookup_dir(env, llog_dir, name, &lgi->lgi_fid);
dt_read_unlock(env, llog_dir);
lu_object_put(env, &llog_dir->do_lu);
if (rc == -ENOENT && open_param == LLOG_OPEN_NEW) {
/* generate fid for new llog */
rc = local_object_fid_generate(env, los,
&lgi->lgi_fid);
}
if (rc < 0)
GOTO(out, rc);
OBD_ALLOC(handle->lgh_name, strlen(name) + 1);
if (handle->lgh_name)
strcpy(handle->lgh_name, name);
else
GOTO(out, rc = -ENOMEM);
} else {
LASSERTF(open_param & LLOG_OPEN_NEW, "%#x\n", open_param);
/* generate fid for new llog */
rc = local_object_fid_generate(env, los, &lgi->lgi_fid);
if (rc < 0)
GOTO(out, rc);
}
o = ls_locate(env, ls, &lgi->lgi_fid, NULL);
if (IS_ERR(o))
GOTO(out_name, rc = PTR_ERR(o));
/* No new llog is expected but doesn't exist */
if (open_param != LLOG_OPEN_NEW && !dt_object_exists(o))
GOTO(out_put, rc = -ENOENT);
fid_to_logid(&lgi->lgi_fid, &handle->lgh_id);
handle->lgh_obj = o;
handle->private_data = los;
LASSERT(handle->lgh_ctxt);
RETURN(rc);
out_put:
lu_object_put(env, &o->do_lu);
out_name:
if (handle->lgh_name != NULL)
OBD_FREE(handle->lgh_name, strlen(name) + 1);
out:
dt_los_put(los);
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);
}
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_prev_block(const struct lu_env *env,
struct llog_handle *loghandle,
int prev_idx, void *buf, int len)
{
struct llog_thread_info *lgi = llog_info(env);
struct dt_object *o;
struct dt_device *dt;
loff_t cur_offset;
int rc;
ENTRY;
if (len == 0 || len & (LLOG_CHUNK_SIZE - 1))
RETURN(-EINVAL);
CDEBUG(D_OTHER, "looking for log index %u\n", prev_idx);
LASSERT(loghandle);
LASSERT(loghandle->lgh_ctxt);
o = loghandle->lgh_obj;
LASSERT(o);
LASSERT(dt_object_exists(o));
dt = lu2dt_dev(o->do_lu.lo_dev);
LASSERT(dt);
cur_offset = LLOG_CHUNK_SIZE;
llog_skip_over(&cur_offset, 0, prev_idx);
rc = dt_attr_get(env, o, &lgi->lgi_attr, BYPASS_CAPA);
if (rc)
GOTO(out, rc);
while (cur_offset < lgi->lgi_attr.la_size) {
struct llog_rec_hdr *rec, *last_rec;
struct llog_rec_tail *tail;
lgi->lgi_buf.lb_len = len;
lgi->lgi_buf.lb_buf = buf;
/* It is OK to have locking around dt_read() only, see
* comment in llog_osd_next_block for details
*/
dt_read_lock(env, o, 0);
rc = dt_read(env, o, &lgi->lgi_buf, &cur_offset);
dt_read_unlock(env, o);
if (rc < 0) {
CERROR("%s: can't read llog block from log "DFID
" offset "LPU64": rc = %d\n",
o->do_lu.lo_dev->ld_obd->obd_name,
PFID(lu_object_fid(&o->do_lu)), cur_offset, rc);
GOTO(out, rc);
}
if (rc == 0) /* end of file, nothing to do */
GOTO(out, rc);
if (rc < sizeof(*tail)) {
CERROR("%s: invalid llog block at log id "LPU64"/%u "
"offset "LPU64"\n",
o->do_lu.lo_dev->ld_obd->obd_name,
loghandle->lgh_id.lgl_oid,
loghandle->lgh_id.lgl_ogen, cur_offset);
GOTO(out, rc = -EINVAL);
}
rec = buf;
if (LLOG_REC_HDR_NEEDS_SWABBING(rec))
lustre_swab_llog_rec(rec);
tail = (struct llog_rec_tail *)((char *)buf + rc -
sizeof(struct llog_rec_tail));
/* get the last record in block */
last_rec = (struct llog_rec_hdr *)((char *)buf + rc -
le32_to_cpu(tail->lrt_len));
if (LLOG_REC_HDR_NEEDS_SWABBING(last_rec))
lustre_swab_llog_rec(last_rec);
LASSERT(last_rec->lrh_index == tail->lrt_index);
/* this shouldn't happen */
if (tail->lrt_index == 0) {
CERROR("%s: invalid llog tail at log id "LPU64"/%u "
"offset "LPU64"\n",
o->do_lu.lo_dev->ld_obd->obd_name,
loghandle->lgh_id.lgl_oid,
loghandle->lgh_id.lgl_ogen, cur_offset);
GOTO(out, rc = -EINVAL);
}
if (tail->lrt_index < prev_idx)
continue;
/* sanity check that the start of the new buffer is no farther
* than the record that we wanted. This shouldn't happen. */
if (rec->lrh_index > prev_idx) {
CERROR("%s: missed desired record? %u > %u\n",
o->do_lu.lo_dev->ld_obd->obd_name,
rec->lrh_index, prev_idx);
GOTO(out, rc = -ENOENT);
}
GOTO(out, rc = 0);
}
GOTO(out, rc = -EIO);
out:
return rc;
}
/* sets:
* - cur_offset to the furthest point read in the log file
* - cur_idx to the log index preceeding cur_offset
* returns -EIO/-EINVAL on error
*/
static int llog_osd_next_block(const struct lu_env *env,
struct llog_handle *loghandle, int *cur_idx,
int next_idx, __u64 *cur_offset, void *buf,
int len)
{
struct llog_thread_info *lgi = llog_info(env);
struct dt_object *o;
struct dt_device *dt;
int rc;
ENTRY;
LASSERT(env);
LASSERT(lgi);
if (len == 0 || len & (LLOG_CHUNK_SIZE - 1))
RETURN(-EINVAL);
CDEBUG(D_OTHER, "looking for log index %u (cur idx %u off "LPU64")\n",
next_idx, *cur_idx, *cur_offset);
LASSERT(loghandle);
LASSERT(loghandle->lgh_ctxt);
o = loghandle->lgh_obj;
LASSERT(o);
LASSERT(dt_object_exists(o));
dt = lu2dt_dev(o->do_lu.lo_dev);
LASSERT(dt);
rc = dt_attr_get(env, o, &lgi->lgi_attr, BYPASS_CAPA);
if (rc)
GOTO(out, rc);
while (*cur_offset < lgi->lgi_attr.la_size) {
struct llog_rec_hdr *rec, *last_rec;
struct llog_rec_tail *tail;
llog_skip_over(cur_offset, *cur_idx, next_idx);
/* read up to next LLOG_CHUNK_SIZE block */
lgi->lgi_buf.lb_len = LLOG_CHUNK_SIZE -
(*cur_offset & (LLOG_CHUNK_SIZE - 1));
lgi->lgi_buf.lb_buf = buf;
/* Note: read lock is not needed around la_size get above at
* the time of dt_attr_get(). There are only two cases that
* matter. Either la_size == cur_offset, in which case the
* entire read is skipped, or la_size > cur_offset and the loop
* is entered and this thread is blocked at dt_read_lock()
* until the write is completed. When the write completes, then
* the dt_read() will be done with the full length, and will
* get the full data.
*/
dt_read_lock(env, o, 0);
rc = dt_read(env, o, &lgi->lgi_buf, cur_offset);
dt_read_unlock(env, o);
if (rc < 0) {
CERROR("%s: can't read llog block from log "DFID
" offset "LPU64": rc = %d\n",
o->do_lu.lo_dev->ld_obd->obd_name,
PFID(lu_object_fid(&o->do_lu)), *cur_offset,
rc);
GOTO(out, rc);
}
if (rc < len) {
/* signal the end of the valid buffer to
* llog_process */
memset(buf + rc, 0, len - rc);
}
if (rc == 0) /* end of file, nothing to do */
GOTO(out, rc);
if (rc < sizeof(*tail)) {
CERROR("%s: invalid llog block at log id "LPU64"/%u "
"offset "LPU64"\n",
o->do_lu.lo_dev->ld_obd->obd_name,
loghandle->lgh_id.lgl_oid,
loghandle->lgh_id.lgl_ogen, *cur_offset);
GOTO(out, rc = -EINVAL);
}
rec = buf;
if (LLOG_REC_HDR_NEEDS_SWABBING(rec))
lustre_swab_llog_rec(rec);
tail = (struct llog_rec_tail *)((char *)buf + rc -
sizeof(struct llog_rec_tail));
/* get the last record in block */
last_rec = (struct llog_rec_hdr *)((char *)buf + rc -
le32_to_cpu(tail->lrt_len));
if (LLOG_REC_HDR_NEEDS_SWABBING(last_rec))
lustre_swab_llog_rec(last_rec);
LASSERT(last_rec->lrh_index == tail->lrt_index);
*cur_idx = tail->lrt_index;
/* this shouldn't happen */
if (tail->lrt_index == 0) {
CERROR("%s: invalid llog tail at log id "LPU64"/%u "
"offset "LPU64"\n",
o->do_lu.lo_dev->ld_obd->obd_name,
loghandle->lgh_id.lgl_oid,
loghandle->lgh_id.lgl_ogen, *cur_offset);
GOTO(out, rc = -EINVAL);
}
if (tail->lrt_index < next_idx)
continue;
/* sanity check that the start of the new buffer is no farther
* than the record that we wanted. This shouldn't happen. */
if (rec->lrh_index > next_idx) {
CERROR("%s: missed desired record? %u > %u\n",
o->do_lu.lo_dev->ld_obd->obd_name,
rec->lrh_index, next_idx);
GOTO(out, rc = -ENOENT);
}
GOTO(out, rc = 0);
}
GOTO(out, rc = -EIO);
out:
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);
}
static int out_read(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;
struct object_update_reply *reply = tti->tti_u.update.tti_update_reply;
int index = tti->tti_u.update.tti_update_reply_index;
struct lu_rdbuf *rdbuf;
struct object_update_result *update_result;
struct out_read_reply *orr;
void *tmp;
size_t size;
size_t total_size = 0;
__u64 pos;
unsigned int i;
unsigned int nbufs;
int rc = 0;
ENTRY;
update_result = object_update_result_get(reply, index, NULL);
LASSERT(update_result != NULL);
update_result->our_datalen = sizeof(*orr);
if (!lu_object_exists(&obj->do_lu))
GOTO(out, rc = -ENOENT);
tmp = object_update_param_get(update, 0, NULL);
if (IS_ERR(tmp)) {
CERROR("%s: empty size for read: rc = %ld\n",
tgt_name(tsi->tsi_tgt), PTR_ERR(tmp));
GOTO(out, rc = PTR_ERR(tmp));
}
size = le64_to_cpu(*(size_t *)(tmp));
tmp = object_update_param_get(update, 1, NULL);
if (IS_ERR(tmp)) {
CERROR("%s: empty pos for read: rc = %ld\n",
tgt_name(tsi->tsi_tgt), PTR_ERR(tmp));
GOTO(out, rc = PTR_ERR(tmp));
}
pos = le64_to_cpu(*(__u64 *)(tmp));
/* Put the offset into the begining of the buffer in reply */
orr = (struct out_read_reply *)update_result->our_data;
nbufs = (size + OUT_BULK_BUFFER_SIZE - 1) / OUT_BULK_BUFFER_SIZE;
OBD_ALLOC(rdbuf, sizeof(struct lu_rdbuf) +
nbufs * sizeof(rdbuf->rb_bufs[0]));
if (rdbuf == NULL)
GOTO(out, rc = -ENOMEM);
rdbuf->rb_nbufs = 0;
total_size = 0;
for (i = 0; i < nbufs; i++) {
__u32 read_size;
read_size = size > OUT_BULK_BUFFER_SIZE ?
OUT_BULK_BUFFER_SIZE : size;
OBD_ALLOC(rdbuf->rb_bufs[i].lb_buf, read_size);
if (rdbuf->rb_bufs[i].lb_buf == NULL)
GOTO(out_free, rc = -ENOMEM);
rdbuf->rb_bufs[i].lb_len = read_size;
dt_read_lock(env, obj, MOR_TGT_CHILD);
rc = dt_read(env, obj, &rdbuf->rb_bufs[i], &pos);
dt_read_unlock(env, obj);
total_size += rc < 0 ? 0 : rc;
if (rc <= 0)
break;
rdbuf->rb_nbufs++;
size -= read_size;
}
/* send pages to client */
rc = tgt_send_buffer(tsi, rdbuf);
if (rc < 0)
GOTO(out_free, rc);
orr->orr_size = total_size;
orr->orr_offset = pos;
orr_cpu_to_le(orr, orr);
update_result->our_datalen += orr->orr_size;
out_free:
for (i = 0; i < nbufs; i++) {
if (rdbuf->rb_bufs[i].lb_buf != NULL) {
OBD_FREE(rdbuf->rb_bufs[i].lb_buf,
rdbuf->rb_bufs[i].lb_len);
}
}
OBD_FREE(rdbuf, sizeof(struct lu_rdbuf) +
nbufs * sizeof(rdbuf->rb_bufs[0]));
out:
/* Insert read buffer */
update_result->our_rc = ptlrpc_status_hton(rc);
reply->ourp_lens[index] = cfs_size_round(update_result->our_datalen +
sizeof(*update_result));
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);
}
