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);
}
/* This function implies that caller takes care about locking. */
int seq_store_write(struct lu_server_seq *seq,
const struct lu_env *env,
struct thandle *th)
{
struct dt_object *dt_obj = seq->lss_obj;
struct seq_thread_info *info;
struct dt_device *dt_dev;
loff_t pos = 0;
int rc;
ENTRY;
dt_dev = lu2dt_dev(seq->lss_obj->do_lu.lo_dev);
info = lu_context_key_get(&env->le_ctx, &seq_thread_key);
LASSERT(info != NULL);
/* Store ranges in le format. */
range_cpu_to_le(&info->sti_space, &seq->lss_space);
rc = dt_obj->do_body_ops->dbo_write(env, dt_obj,
seq_store_buf(info),
&pos, th, BYPASS_CAPA, 1);
if (rc == sizeof(info->sti_space)) {
CDEBUG(D_INFO, "%s: Space - "DRANGE"\n",
seq->lss_name, PRANGE(&seq->lss_space));
rc = 0;
} else if (rc >= 0) {
rc = -EIO;
}
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);
}
int fld_index_create(struct lu_server_fld *fld,
const struct lu_env *env,
const struct lu_seq_range *range,
struct thandle *th)
{
struct dt_object *dt_obj = fld->lsf_obj;
struct dt_device *dt_dev;
seqno_t start;
int rc;
ENTRY;
start = range->lsr_start;
LASSERT(range_is_sane(range));
dt_dev = lu2dt_dev(fld->lsf_obj->do_lu.lo_dev);
rc = dt_obj->do_index_ops->dio_insert(env, dt_obj,
fld_rec(env, range),
fld_key(env, start),
th, BYPASS_CAPA, 1);
CDEBUG(D_INFO, "%s: insert given range : "DRANGE" rc = %d\n",
fld->lsf_name, PRANGE(range), rc);
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, 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;
}
void fld_trans_stop(struct lu_server_fld *fld,
const struct lu_env *env, struct thandle* th)
{
struct dt_device *dt_dev;
dt_dev = lu2dt_dev(fld->lsf_obj->do_lu.lo_dev);
dt_dev->dd_ops->dt_trans_stop(env, th);
}
struct thandle *fld_trans_create(struct lu_server_fld *fld,
const struct lu_env *env)
{
struct dt_device *dt_dev;
dt_dev = lu2dt_dev(fld->lsf_obj->do_lu.lo_dev);
return dt_dev->dd_ops->dt_trans_create(env, dt_dev);
}
/*
* 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);
}
void seq_store_trans_stop(struct lu_server_seq *seq,
const struct lu_env *env,
struct thandle *th)
{
struct dt_device *dt_dev;
ENTRY;
dt_dev = lu2dt_dev(seq->lss_obj->do_lu.lo_dev);
dt_dev->dd_ops->dt_trans_stop(env, th);
}
static int mdd_connect_to_next(const struct lu_env *env, struct mdd_device *m,
const char *nextdev)
{
struct obd_connect_data *data = NULL;
struct lu_device *lud = mdd2lu_dev(m);
struct obd_device *obd;
int rc;
ENTRY;
LASSERT(m->mdd_child_exp == NULL);
OBD_ALLOC(data, sizeof(*data));
if (data == NULL)
GOTO(out, rc = -ENOMEM);
obd = class_name2obd(nextdev);
if (obd == NULL) {
CERROR("can't locate next device: %s\n", nextdev);
GOTO(out, rc = -ENOTCONN);
}
data->ocd_connect_flags = OBD_CONNECT_VERSION;
data->ocd_version = LUSTRE_VERSION_CODE;
rc = obd_connect(NULL, &m->mdd_child_exp, obd, &obd->obd_uuid, data, NULL);
if (rc) {
CERROR("cannot connect to next dev %s (%d)\n", nextdev, rc);
GOTO(out, rc);
}
lud->ld_site = m->mdd_child_exp->exp_obd->obd_lu_dev->ld_site;
LASSERT(lud->ld_site);
m->mdd_child = lu2dt_dev(m->mdd_child_exp->exp_obd->obd_lu_dev);
m->mdd_bottom = lu2dt_dev(lud->ld_site->ls_bottom_dev);
lu_dev_add_linkage(lud->ld_site, lud);
out:
if (data)
OBD_FREE(data, sizeof(*data));
RETURN(rc);
}
struct thandle* fld_trans_start(struct lu_server_fld *fld,
const struct lu_env *env, int credit)
{
struct fld_thread_info *info;
struct dt_device *dt_dev;
struct txn_param *p;
dt_dev = lu2dt_dev(fld->lsf_obj->do_lu.lo_dev);
info = lu_context_key_get(&env->le_ctx, &fld_thread_key);
p = &info->fti_txn_param;
txn_param_init(p, credit);
return dt_dev->dd_ops->dt_trans_start(env, dt_dev, p);
}
/**
* 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);
}
/**
* Implementation of dt_object_operations::do_object_lock
*
* Enqueue a lock (by ldlm_cli_enqueue()) of remote object on the remote MDT,
* which will lock the object in the global namespace. And because the
* cross-MDT locks are relatively rare compared with normal local MDT operation,
* let's release it right away, instead of putting it into the LRU list.
*
* \param[in] env execution environment
* \param[in] dt object to be locked
* \param[out] lh lock handle
* \param[in] einfo enqueue information
* \param[in] policy lock policy
*
* \retval ELDLM_OK if locking the object succeeds.
* \retval negative errno if locking fails.
*/
static int osp_md_object_lock(const struct lu_env *env,
struct dt_object *dt,
struct lustre_handle *lh,
struct ldlm_enqueue_info *einfo,
union ldlm_policy_data *policy)
{
struct ldlm_res_id *res_id;
struct dt_device *dt_dev = lu2dt_dev(dt->do_lu.lo_dev);
struct osp_device *osp = dt2osp_dev(dt_dev);
struct ptlrpc_request *req;
int rc = 0;
__u64 flags = 0;
enum ldlm_mode mode;
res_id = einfo->ei_res_id;
LASSERT(res_id != NULL);
mode = ldlm_lock_match(osp->opd_obd->obd_namespace,
LDLM_FL_BLOCK_GRANTED, res_id,
einfo->ei_type, policy,
einfo->ei_mode, lh, 0);
if (mode > 0)
return ELDLM_OK;
if (einfo->ei_nonblock)
flags |= LDLM_FL_BLOCK_NOWAIT;
req = ldlm_enqueue_pack(osp->opd_exp, 0);
if (IS_ERR(req))
RETURN(PTR_ERR(req));
rc = ldlm_cli_enqueue(osp->opd_exp, &req, einfo, res_id,
(const union ldlm_policy_data *)policy,
&flags, NULL, 0, LVB_T_NONE, lh, 0);
ptlrpc_req_finished(req);
if (rc == ELDLM_OK) {
struct ldlm_lock *lock;
lock = __ldlm_handle2lock(lh, 0);
ldlm_set_cbpending(lock);
LDLM_LOCK_PUT(lock);
}
return rc == ELDLM_OK ? 0 : -EIO;
}
/*
* Connect a quota master to the backend OSD device.
*
* \param env - is the environment passed by the caller
* \param qmt - is the quota master target to be connected
* \param cfg - is the configuration log record from which we need to extract
* the service name of the backend OSD device to connect to.
*
* \retval - 0 on success, appropriate error on failure
*/
static int qmt_connect_to_osd(const struct lu_env *env, struct qmt_device *qmt,
struct lustre_cfg *cfg)
{
struct obd_connect_data *data = NULL;
struct obd_device *obd;
struct lu_device *ld = qmt2lu_dev(qmt);
int rc;
ENTRY;
LASSERT(qmt->qmt_child_exp == NULL);
OBD_ALLOC_PTR(data);
if (data == NULL)
GOTO(out, rc = -ENOMEM);
/* look-up OBD device associated with the backend OSD device.
* The MDT is kind enough to pass the OBD name in QMT configuration */
obd = class_name2obd(lustre_cfg_string(cfg, 3));
if (obd == NULL) {
CERROR("%s: can't locate backend osd device: %s\n",
qmt->qmt_svname, lustre_cfg_string(cfg, 3));
GOTO(out, rc = -ENOTCONN);
}
data->ocd_connect_flags = OBD_CONNECT_VERSION;
data->ocd_version = LUSTRE_VERSION_CODE;
/* connect to OSD device */
rc = obd_connect(NULL, &qmt->qmt_child_exp, obd, &obd->obd_uuid, data,
NULL);
if (rc) {
CERROR("%s: cannot connect to osd dev %s (%d)\n",
qmt->qmt_svname, obd->obd_name, rc);
GOTO(out, rc);
}
/* initialize site (although it isn't used anywhere) and lu_device
* pointer to next device */
qmt->qmt_child = lu2dt_dev(qmt->qmt_child_exp->exp_obd->obd_lu_dev);
ld->ld_site = qmt->qmt_child_exp->exp_obd->obd_lu_dev->ld_site;
EXIT;
out:
if (data)
OBD_FREE_PTR(data);
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);
}
/**
* Implementation of dt_object_operations::do_object_lock
*
* Enqueue a lock (by ldlm_cli_enqueue()) of remote object on the remote MDT,
* which will lock the object in the global namespace. And because the
* cross-MDT locks are relatively rare compared with normal local MDT operation,
* let's release it right away, instead of putting it into the LRU list.
*
* \param[in] env execution environment
* \param[in] dt object to be locked
* \param[out] lh lock handle
* \param[in] einfo enqueue information
* \param[in] policy lock policy
*
* \retval ELDLM_OK if locking the object succeeds.
* \retval negative errno if locking fails.
*/
static int osp_md_object_lock(const struct lu_env *env,
struct dt_object *dt,
struct lustre_handle *lh,
struct ldlm_enqueue_info *einfo,
union ldlm_policy_data *policy)
{
struct ldlm_res_id *res_id;
struct dt_device *dt_dev = lu2dt_dev(dt->do_lu.lo_dev);
struct osp_device *osp = dt2osp_dev(dt_dev);
struct lu_device *top_device;
struct ptlrpc_request *req;
int rc = 0;
__u64 flags = LDLM_FL_NO_LRU;
res_id = einfo->ei_res_id;
LASSERT(res_id != NULL);
if (einfo->ei_nonblock)
flags |= LDLM_FL_BLOCK_NOWAIT;
if (einfo->ei_mode & (LCK_EX | LCK_PW))
flags |= LDLM_FL_COS_INCOMPAT;
req = ldlm_enqueue_pack(osp->opd_exp, 0);
if (IS_ERR(req))
RETURN(PTR_ERR(req));
/* During recovery, it needs to let OSP send enqueue
* without checking recoverying status, in case the
* other target is being recovered at the same time,
* and if we wait here for the import to be recovered,
* it might cause deadlock */
top_device = dt_dev->dd_lu_dev.ld_site->ls_top_dev;
if (top_device->ld_obd->obd_recovering)
req->rq_allow_replay = 1;
rc = ldlm_cli_enqueue(osp->opd_exp, &req, einfo, res_id,
(const union ldlm_policy_data *)policy,
&flags, NULL, 0, LVB_T_NONE, lh, 0);
ptlrpc_req_finished(req);
return rc == ELDLM_OK ? 0 : -EIO;
}
struct thandle *seq_store_trans_start(struct lu_server_seq *seq,
const struct lu_env *env, int credit,
int sync)
{
struct seq_thread_info *info;
struct dt_device *dt_dev;
struct thandle *th;
ENTRY;
dt_dev = lu2dt_dev(seq->lss_obj->do_lu.lo_dev);
info = lu_context_key_get(&env->le_ctx, &seq_thread_key);
LASSERT(info != NULL);
txn_param_init(&info->sti_txn, credit);
if (sync)
txn_param_sync(&info->sti_txn);
th = dt_dev->dd_ops->dt_trans_start(env, dt_dev, &info->sti_txn);
return th;
}
int fld_index_delete(struct lu_server_fld *fld,
const struct lu_env *env,
struct lu_seq_range *range,
struct thandle *th)
{
struct dt_object *dt_obj = fld->lsf_obj;
struct dt_device *dt_dev;
seqno_t seq = range->lsr_start;
int rc;
ENTRY;
dt_dev = lu2dt_dev(fld->lsf_obj->do_lu.lo_dev);
rc = dt_obj->do_index_ops->dio_delete(env, dt_obj,
fld_key(env, seq), th,
BYPASS_CAPA);
CDEBUG(D_INFO, "%s: delete given range : "DRANGE" rc = %d\n",
fld->lsf_name, PRANGE(range), rc);
RETURN(rc);
}
/**
* Implementation of the llog_operations::lop_prev_block
*
* This function finds the llog block to return which contains
* record with required index but in reverse order - from end of llog
* to the beginning.
* It is main part of reverse llog processing.
*
* \param[in] env execution environment
* \param[in] loghandle llog handle of the current llog
* \param[in] prev_idx target index to find
* \param[in] buf pointer to data buffer to fill
* \param[in] len required len to read, it is LLOG_CHUNK_SIZE usually.
*
* \retval 0 on successful buffer read
* \retval negative value on error
*/
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;
rc = dt_read(env, o, &lgi->lgi_buf, &cur_offset);
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 "DOSTID"/%u "
"offset "LPU64"\n",
o->do_lu.lo_dev->ld_obd->obd_name,
POSTID(&loghandle->lgh_id.lgl_oi),
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 "DOSTID"/%u "
"offset "LPU64"\n",
o->do_lu.lo_dev->ld_obd->obd_name,
POSTID(&loghandle->lgh_id.lgl_oi),
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;
}
/**
* 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_test_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
{
struct obd_device *tgt;
struct llog_ctxt *ctxt;
struct dt_object *o;
struct lu_env env;
struct lu_context test_session;
int rc;
if (lcfg->lcfg_bufcount < 2) {
CERROR("requires a TARGET OBD name\n");
return -EINVAL;
}
if (lcfg->lcfg_buflens[1] < 1) {
CERROR("requires a TARGET OBD name\n");
return -EINVAL;
}
/* disk obd */
tgt = class_name2obd(lustre_cfg_string(lcfg, 1));
if (!tgt || !tgt->obd_attached || !tgt->obd_set_up) {
CERROR("target device not attached or not set up (%s)\n",
lustre_cfg_string(lcfg, 1));
return -EINVAL;
}
rc = lu_env_init(&env, LCT_LOCAL | LCT_MG_THREAD);
if (rc)
return rc;
rc = lu_context_init(&test_session, LCT_SESSION);
if (rc)
GOTO(cleanup_env, rc);
test_session.lc_thread = (struct ptlrpc_thread *)current;
lu_context_enter(&test_session);
env.le_ses = &test_session;
CWARN("Setup llog-test device over %s device\n",
lustre_cfg_string(lcfg, 1));
OBD_SET_CTXT_MAGIC(&obd->obd_lvfs_ctxt);
obd->obd_lvfs_ctxt.dt = lu2dt_dev(tgt->obd_lu_dev);
rc = llog_setup(&env, tgt, &tgt->obd_olg, LLOG_TEST_ORIG_CTXT, tgt,
&llog_osd_ops);
if (rc)
GOTO(cleanup_session, rc);
/* use MGS llog dir for tests */
ctxt = llog_get_context(tgt, LLOG_CONFIG_ORIG_CTXT);
LASSERT(ctxt);
o = ctxt->loc_dir;
llog_ctxt_put(ctxt);
ctxt = llog_get_context(tgt, LLOG_TEST_ORIG_CTXT);
LASSERT(ctxt);
ctxt->loc_dir = o;
llog_ctxt_put(ctxt);
llog_test_rand = cfs_rand();
rc = llog_run_tests(&env, tgt);
if (rc)
llog_test_cleanup(obd);
cleanup_session:
lu_context_exit(&test_session);
lu_context_fini(&test_session);
cleanup_env:
lu_env_fini(&env);
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);
}
/* 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;
}
/**
* Implementation of the llog_operations::lop_next_block
*
* This function finds the the next llog block to return which contains
* record with required index. It is main part of llog processing.
*
* \param[in] env execution environment
* \param[in] loghandle llog handle of the current llog
* \param[in,out] cur_idx index preceeding cur_offset
* \param[in] next_idx target index to find
* \param[in,out] cur_offset furtherst point read in the file
* \param[in] buf pointer to data buffer to fill
* \param[in] len required len to read, it is
* LLOG_CHUNK_SIZE usually.
*
* \retval 0 on successful buffer read
* \retval negative value 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;
rc = dt_read(env, o, &lgi->lgi_buf, cur_offset);
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 "DOSTID"/%u "
"offset "LPU64"\n",
o->do_lu.lo_dev->ld_obd->obd_name,
POSTID(&loghandle->lgh_id.lgl_oi),
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 -
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 "DOSTID"/%u "
"offset "LPU64"\n",
o->do_lu.lo_dev->ld_obd->obd_name,
POSTID(&loghandle->lgh_id.lgl_oi),
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);
}
/* Trim unsupported extensions for compat w/ older clients */
if (!(loghandle->lgh_hdr->llh_flags & LLOG_F_EXT_JOBID))
changelog_block_trim_ext(rec, last_rec,
CLF_VERSION | CLF_RENAME);
GOTO(out, rc = 0);
}
GOTO(out, rc = -EIO);
out:
return 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);
}
