int mdc_link(struct obd_export *exp, struct md_op_data *op_data,
struct ptlrpc_request **request)
{
struct list_head cancels = LIST_HEAD_INIT(cancels);
struct ptlrpc_request *req;
int count = 0, rc;
ENTRY;
if ((op_data->op_flags & MF_MDC_CANCEL_FID2) &&
(fid_is_sane(&op_data->op_fid2)))
count = mdc_resource_get_unused(exp, &op_data->op_fid2,
&cancels, LCK_EX,
MDS_INODELOCK_UPDATE);
if ((op_data->op_flags & MF_MDC_CANCEL_FID1) &&
(fid_is_sane(&op_data->op_fid1)))
count += mdc_resource_get_unused(exp, &op_data->op_fid1,
&cancels, LCK_EX,
MDS_INODELOCK_UPDATE);
req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_REINT_LINK);
if (req == NULL) {
ldlm_lock_list_put(&cancels, l_bl_ast, count);
RETURN(-ENOMEM);
}
req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
op_data->op_namelen + 1);
/* get SELinux policy info if any */
rc = sptlrpc_get_sepol(req);
if (rc < 0) {
ptlrpc_request_free(req);
RETURN(rc);
}
req_capsule_set_size(&req->rq_pill, &RMF_SELINUX_POL, RCL_CLIENT,
strlen(req->rq_sepol) ?
strlen(req->rq_sepol) + 1 : 0);
rc = mdc_prep_elc_req(exp, req, MDS_REINT, &cancels, count);
if (rc) {
ptlrpc_request_free(req);
RETURN(rc);
}
mdc_link_pack(req, op_data);
ptlrpc_request_set_replen(req);
rc = mdc_reint(req, LUSTRE_IMP_FULL);
*request = req;
if (rc == -ERESTARTSYS)
rc = 0;
RETURN(rc);
}
int mdc_setattr(struct obd_export *exp, struct md_op_data *op_data,
void *ea, size_t ealen, struct ptlrpc_request **request)
{
struct list_head cancels = LIST_HEAD_INIT(cancels);
struct ptlrpc_request *req;
struct mdc_rpc_lock *rpc_lock;
struct obd_device *obd = exp->exp_obd;
int count = 0, rc;
__u64 bits;
ENTRY;
LASSERT(op_data != NULL);
bits = MDS_INODELOCK_UPDATE;
if (op_data->op_attr.ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID))
bits |= MDS_INODELOCK_LOOKUP;
if ((op_data->op_flags & MF_MDC_CANCEL_FID1) &&
(fid_is_sane(&op_data->op_fid1)))
count = mdc_resource_get_unused(exp, &op_data->op_fid1,
&cancels, LCK_EX, bits);
req = ptlrpc_request_alloc(class_exp2cliimp(exp),
&RQF_MDS_REINT_SETATTR);
if (req == NULL) {
ldlm_lock_list_put(&cancels, l_bl_ast, count);
RETURN(-ENOMEM);
}
mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
req_capsule_set_size(&req->rq_pill, &RMF_MDT_EPOCH, RCL_CLIENT, 0);
req_capsule_set_size(&req->rq_pill, &RMF_EADATA, RCL_CLIENT, ealen);
req_capsule_set_size(&req->rq_pill, &RMF_LOGCOOKIES, RCL_CLIENT, 0);
rc = mdc_prep_elc_req(exp, req, MDS_REINT, &cancels, count);
if (rc) {
ptlrpc_request_free(req);
RETURN(rc);
}
rpc_lock = obd->u.cli.cl_rpc_lock;
if (op_data->op_attr.ia_valid & (ATTR_MTIME | ATTR_CTIME))
CDEBUG(D_INODE, "setting mtime "CFS_TIME_T
", ctime "CFS_TIME_T"\n",
LTIME_S(op_data->op_attr.ia_mtime),
LTIME_S(op_data->op_attr.ia_ctime));
mdc_setattr_pack(req, op_data, ea, ealen);
ptlrpc_request_set_replen(req);
rc = mdc_reint(req, rpc_lock, LUSTRE_IMP_FULL);
if (rc == -ERESTARTSYS)
rc = 0;
*request = req;
RETURN(rc);
}
int mdc_unlink(struct obd_export *exp, struct md_op_data *op_data,
struct ptlrpc_request **request)
{
CFS_LIST_HEAD(cancels);
struct obd_device *obd = class_exp2obd(exp);
struct ptlrpc_request *req = *request;
int count = 0, rc;
ENTRY;
LASSERT(req == NULL);
if ((op_data->op_flags & MF_MDC_CANCEL_FID1) &&
(fid_is_sane(&op_data->op_fid1)) &&
!OBD_FAIL_CHECK(OBD_FAIL_LDLM_BL_CALLBACK_NET))
count = mdc_resource_get_unused(exp, &op_data->op_fid1,
&cancels, LCK_EX,
MDS_INODELOCK_UPDATE);
if ((op_data->op_flags & MF_MDC_CANCEL_FID3) &&
(fid_is_sane(&op_data->op_fid3)) &&
!OBD_FAIL_CHECK(OBD_FAIL_LDLM_BL_CALLBACK_NET))
count += mdc_resource_get_unused(exp, &op_data->op_fid3,
&cancels, LCK_EX,
MDS_INODELOCK_FULL);
req = ptlrpc_request_alloc(class_exp2cliimp(exp),
&RQF_MDS_REINT_UNLINK);
if (req == NULL) {
ldlm_lock_list_put(&cancels, l_bl_ast, count);
RETURN(-ENOMEM);
}
mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
op_data->op_namelen + 1);
rc = mdc_prep_elc_req(exp, req, MDS_REINT, &cancels, count);
if (rc) {
ptlrpc_request_free(req);
RETURN(rc);
}
mdc_unlink_pack(req, op_data);
req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
obd->u.cli.cl_max_mds_easize);
req_capsule_set_size(&req->rq_pill, &RMF_LOGCOOKIES, RCL_SERVER,
obd->u.cli.cl_max_mds_cookiesize);
ptlrpc_request_set_replen(req);
*request = req;
rc = mdc_reint(req, obd->u.cli.cl_rpc_lock, LUSTRE_IMP_FULL);
if (rc == -ERESTARTSYS)
rc = 0;
RETURN(rc);
}
/*
* Send non-intent quota request to master.
*
* \param env - the environment passed by the caller
* \param exp - is the export to use to send the acquire RPC
* \param qbody - quota body to be packed in request
* \param sync - synchronous or asynchronous
* \param completion - completion callback
* \param qqi - is the qsd_qtype_info structure to pass to the completion
* function
* \param lqe - is the qid entry to be processed
*
* \retval 0 - success
* \retval -ve - appropriate errors
*/
int qsd_send_dqacq(const struct lu_env *env, struct obd_export *exp,
struct quota_body *qbody, bool sync,
qsd_req_completion_t completion, struct qsd_qtype_info *qqi,
struct lustre_handle *lockh, struct lquota_entry *lqe)
{
struct ptlrpc_request *req;
struct quota_body *req_qbody;
struct qsd_async_args *aa;
int rc;
ENTRY;
LASSERT(exp);
req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_QUOTA_DQACQ);
if (req == NULL)
GOTO(out, rc = -ENOMEM);
req->rq_no_resend = req->rq_no_delay = 1;
req->rq_no_retry_einprogress = 1;
rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, QUOTA_DQACQ);
if (rc) {
ptlrpc_request_free(req);
GOTO(out, rc);
}
req->rq_request_portal = MDS_READPAGE_PORTAL;
req_qbody = req_capsule_client_get(&req->rq_pill, &RMF_QUOTA_BODY);
*req_qbody = *qbody;
ptlrpc_request_set_replen(req);
CLASSERT(sizeof(*aa) <= sizeof(req->rq_async_args));
aa = ptlrpc_req_async_args(req);
aa->aa_exp = exp;
aa->aa_qqi = qqi;
aa->aa_arg = (void *)lqe;
aa->aa_completion = completion;
lustre_handle_copy(&aa->aa_lockh, lockh);
if (sync) {
rc = ptlrpc_queue_wait(req);
rc = qsd_dqacq_interpret(env, req, aa, rc);
ptlrpc_req_finished(req);
} else {
req->rq_interpret_reply = qsd_dqacq_interpret;
ptlrpcd_add_req(req);
}
RETURN(rc);
out:
completion(env, qqi, qbody, NULL, lockh, NULL, lqe, rc);
return rc;
}
static int osc_io_data_version_start(const struct lu_env *env,
const struct cl_io_slice *slice)
{
struct cl_data_version_io *dv = &slice->cis_io->u.ci_data_version;
struct osc_io *oio = cl2osc_io(env, slice);
struct obdo *oa = &oio->oi_oa;
struct osc_async_cbargs *cbargs = &oio->oi_cbarg;
struct osc_object *obj = cl2osc(slice->cis_obj);
struct lov_oinfo *loi = obj->oo_oinfo;
struct obd_export *exp = osc_export(obj);
struct ptlrpc_request *req;
struct ost_body *body;
struct osc_data_version_args *dva;
int rc;
ENTRY;
memset(oa, 0, sizeof(*oa));
oa->o_oi = loi->loi_oi;
oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
if (dv->dv_flags & (LL_DV_RD_FLUSH | LL_DV_WR_FLUSH)) {
oa->o_valid |= OBD_MD_FLFLAGS;
oa->o_flags |= OBD_FL_SRVLOCK;
if (dv->dv_flags & LL_DV_WR_FLUSH)
oa->o_flags |= OBD_FL_FLUSH;
}
init_completion(&cbargs->opc_sync);
req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_OST_GETATTR);
if (req == NULL)
RETURN(-ENOMEM);
rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_GETATTR);
if (rc < 0) {
ptlrpc_request_free(req);
RETURN(rc);
}
body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
lustre_set_wire_obdo(&req->rq_import->imp_connect_data, &body->oa, oa);
ptlrpc_request_set_replen(req);
req->rq_interpret_reply = osc_data_version_interpret;
CLASSERT(sizeof(*dva) <= sizeof(req->rq_async_args));
dva = ptlrpc_req_async_args(req);
dva->dva_oio = oio;
ptlrpcd_add_req(req);
RETURN(0);
}
int mdc_setattr(struct obd_export *exp, struct md_op_data *op_data,
void *ea, size_t ealen, struct ptlrpc_request **request)
{
LIST_HEAD(cancels);
struct ptlrpc_request *req;
int count = 0, rc;
__u64 bits;
bits = MDS_INODELOCK_UPDATE;
if (op_data->op_attr.ia_valid & (ATTR_MODE | ATTR_UID | ATTR_GID))
bits |= MDS_INODELOCK_LOOKUP;
if ((op_data->op_flags & MF_MDC_CANCEL_FID1) &&
(fid_is_sane(&op_data->op_fid1)))
count = mdc_resource_get_unused(exp, &op_data->op_fid1,
&cancels, LCK_EX, bits);
req = ptlrpc_request_alloc(class_exp2cliimp(exp),
&RQF_MDS_REINT_SETATTR);
if (!req) {
ldlm_lock_list_put(&cancels, l_bl_ast, count);
return -ENOMEM;
}
req_capsule_set_size(&req->rq_pill, &RMF_MDT_EPOCH, RCL_CLIENT, 0);
req_capsule_set_size(&req->rq_pill, &RMF_EADATA, RCL_CLIENT, ealen);
req_capsule_set_size(&req->rq_pill, &RMF_LOGCOOKIES, RCL_CLIENT, 0);
rc = mdc_prep_elc_req(exp, req, MDS_REINT, &cancels, count);
if (rc) {
ptlrpc_request_free(req);
return rc;
}
if (op_data->op_attr.ia_valid & (ATTR_MTIME | ATTR_CTIME))
CDEBUG(D_INODE, "setting mtime %ld, ctime %ld\n",
LTIME_S(op_data->op_attr.ia_mtime),
LTIME_S(op_data->op_attr.ia_ctime));
mdc_setattr_pack(req, op_data, ea, ealen);
ptlrpc_request_set_replen(req);
rc = mdc_reint(req, LUSTRE_IMP_FULL);
if (rc == -ERESTARTSYS)
rc = 0;
*request = req;
return rc;
}
int mdc_unlink(struct obd_export *exp, struct md_op_data *op_data,
struct ptlrpc_request **request)
{
LIST_HEAD(cancels);
struct obd_device *obd = class_exp2obd(exp);
struct ptlrpc_request *req = *request;
int count = 0, rc;
LASSERT(!req);
if ((op_data->op_flags & MF_MDC_CANCEL_FID1) &&
(fid_is_sane(&op_data->op_fid1)))
count = mdc_resource_get_unused(exp, &op_data->op_fid1,
&cancels, LCK_EX,
MDS_INODELOCK_UPDATE);
if ((op_data->op_flags & MF_MDC_CANCEL_FID3) &&
(fid_is_sane(&op_data->op_fid3)))
count += mdc_resource_get_unused(exp, &op_data->op_fid3,
&cancels, LCK_EX,
MDS_INODELOCK_FULL);
req = ptlrpc_request_alloc(class_exp2cliimp(exp),
&RQF_MDS_REINT_UNLINK);
if (!req) {
ldlm_lock_list_put(&cancels, l_bl_ast, count);
return -ENOMEM;
}
req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
op_data->op_namelen + 1);
rc = mdc_prep_elc_req(exp, req, MDS_REINT, &cancels, count);
if (rc) {
ptlrpc_request_free(req);
return rc;
}
mdc_unlink_pack(req, op_data);
req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
obd->u.cli.cl_default_mds_easize);
ptlrpc_request_set_replen(req);
*request = req;
rc = mdc_reint(req, LUSTRE_IMP_FULL);
if (rc == -ERESTARTSYS)
rc = 0;
return rc;
}
int mdc_link(struct obd_export *exp, struct md_op_data *op_data,
struct ptlrpc_request **request)
{
CFS_LIST_HEAD(cancels);
struct obd_device *obd = exp->exp_obd;
struct ptlrpc_request *req;
int count = 0, rc;
ENTRY;
if ((op_data->op_flags & MF_MDC_CANCEL_FID2) &&
(fid_is_sane(&op_data->op_fid2)))
count = mdc_resource_get_unused(exp, &op_data->op_fid2,
&cancels, LCK_EX,
MDS_INODELOCK_UPDATE);
if ((op_data->op_flags & MF_MDC_CANCEL_FID1) &&
(fid_is_sane(&op_data->op_fid1)))
count += mdc_resource_get_unused(exp, &op_data->op_fid1,
&cancels, LCK_EX,
MDS_INODELOCK_UPDATE);
req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_REINT_LINK);
if (req == NULL) {
ldlm_lock_list_put(&cancels, l_bl_ast, count);
RETURN(-ENOMEM);
}
mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
mdc_set_capa_size(req, &RMF_CAPA2, op_data->op_capa2);
req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
op_data->op_namelen + 1);
rc = mdc_prep_elc_req(exp, req, &cancels, count);
if (rc) {
ptlrpc_request_free(req);
RETURN(rc);
}
mdc_link_pack(req, op_data);
ptlrpc_request_set_replen(req);
rc = mdc_reint(req, obd->u.cli.cl_rpc_lock, LUSTRE_IMP_FULL);
*request = req;
if (rc == -ERESTARTSYS)
rc = 0;
RETURN(rc);
}
static int osp_statfs_update(struct osp_device *d)
{
struct ptlrpc_request *req;
struct obd_import *imp;
union ptlrpc_async_args *aa;
int rc;
ENTRY;
CDEBUG(D_CACHE, "going to update statfs\n");
imp = d->opd_obd->u.cli.cl_import;
LASSERT(imp);
req = ptlrpc_request_alloc(imp, &RQF_OST_STATFS);
if (req == NULL)
RETURN(-ENOMEM);
rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_STATFS);
if (rc) {
ptlrpc_request_free(req);
RETURN(rc);
}
ptlrpc_request_set_replen(req);
req->rq_request_portal = OST_CREATE_PORTAL;
ptlrpc_at_set_req_timeout(req);
req->rq_interpret_reply = (ptlrpc_interpterer_t)osp_statfs_interpret;
aa = ptlrpc_req_async_args(req);
aa->pointer_arg[0] = d;
/*
* no updates till reply
*/
cfs_timer_disarm(&d->opd_statfs_timer);
d->opd_statfs_fresh_till = cfs_time_shift(obd_timeout * 1000);
d->opd_statfs_update_in_progress = 1;
ptlrpcd_add_req(req, PDL_POLICY_ROUND, -1);
RETURN(0);
}
/**
* asks OST to clean precreate orphans
* and gets next id for new objects
*/
static int osp_precreate_cleanup_orphans(struct lu_env *env,
struct osp_device *d)
{
struct osp_thread_info *osi = osp_env_info(env);
struct lu_fid *last_fid = &osi->osi_fid;
struct ptlrpc_request *req = NULL;
struct obd_import *imp;
struct ost_body *body;
struct l_wait_info lwi = { 0 };
int update_status = 0;
int rc;
int diff;
ENTRY;
/*
* wait for local recovery to finish, so we can cleanup orphans
* orphans are all objects since "last used" (assigned), but
* there might be objects reserved and in some cases they won't
* be used. we can't cleanup them till we're sure they won't be
* used. also can't we allow new reservations because they may
* end up getting orphans being cleaned up below. so we block
* new reservations and wait till all reserved objects either
* user or released.
*/
spin_lock(&d->opd_pre_lock);
d->opd_pre_recovering = 1;
spin_unlock(&d->opd_pre_lock);
/*
* The locking above makes sure the opd_pre_reserved check below will
* catch all osp_precreate_reserve() calls who find
* "!opd_pre_recovering".
*/
l_wait_event(d->opd_pre_waitq,
(!d->opd_pre_reserved && d->opd_recovery_completed) ||
!osp_precreate_running(d) || d->opd_got_disconnected,
&lwi);
if (!osp_precreate_running(d) || d->opd_got_disconnected)
GOTO(out, rc = -EAGAIN);
CDEBUG(D_HA, "%s: going to cleanup orphans since "DFID"\n",
d->opd_obd->obd_name, PFID(&d->opd_last_used_fid));
*last_fid = d->opd_last_used_fid;
/* The OSP should already get the valid seq now */
LASSERT(!fid_is_zero(last_fid));
if (fid_oid(&d->opd_last_used_fid) < 2) {
/* lastfid looks strange... ask OST */
rc = osp_get_lastfid_from_ost(env, d);
if (rc)
GOTO(out, rc);
}
imp = d->opd_obd->u.cli.cl_import;
LASSERT(imp);
req = ptlrpc_request_alloc(imp, &RQF_OST_CREATE);
if (req == NULL)
GOTO(out, rc = -ENOMEM);
rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_CREATE);
if (rc) {
ptlrpc_request_free(req);
req = NULL;
GOTO(out, rc);
}
body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
if (body == NULL)
GOTO(out, rc = -EPROTO);
body->oa.o_flags = OBD_FL_DELORPHAN;
body->oa.o_valid = OBD_MD_FLFLAGS | OBD_MD_FLGROUP;
fid_to_ostid(&d->opd_last_used_fid, &body->oa.o_oi);
ptlrpc_request_set_replen(req);
/* Don't resend the delorphan req */
req->rq_no_resend = req->rq_no_delay = 1;
rc = ptlrpc_queue_wait(req);
if (rc) {
update_status = 1;
GOTO(out, rc);
}
body = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
if (body == NULL)
GOTO(out, rc = -EPROTO);
/*
* OST provides us with id new pool starts from in body->oa.o_id
*/
ostid_to_fid(last_fid, &body->oa.o_oi, d->opd_index);
spin_lock(&d->opd_pre_lock);
diff = lu_fid_diff(&d->opd_last_used_fid, last_fid);
if (diff > 0) {
d->opd_pre_grow_count = OST_MIN_PRECREATE + diff;
d->opd_pre_last_created_fid = d->opd_last_used_fid;
} else {
d->opd_pre_grow_count = OST_MIN_PRECREATE;
d->opd_pre_last_created_fid = *last_fid;
}
/*
* This empties the pre-creation pool and effectively blocks any new
* reservations.
*/
LASSERT(fid_oid(&d->opd_pre_last_created_fid) <=
LUSTRE_DATA_SEQ_MAX_WIDTH);
d->opd_pre_used_fid = d->opd_pre_last_created_fid;
d->opd_pre_grow_slow = 0;
spin_unlock(&d->opd_pre_lock);
CDEBUG(D_HA, "%s: Got last_id "DFID" from OST, last_created "DFID
"last_used is "DFID"\n", d->opd_obd->obd_name, PFID(last_fid),
PFID(&d->opd_pre_last_created_fid), PFID(&d->opd_last_used_fid));
out:
if (req)
ptlrpc_req_finished(req);
spin_lock(&d->opd_pre_lock);
d->opd_pre_recovering = 0;
spin_unlock(&d->opd_pre_lock);
/*
* If rc is zero, the pre-creation window should have been emptied.
* Since waking up the herd would be useless without pre-created
* objects, we defer the signal to osp_precreate_send() in that case.
*/
if (rc != 0) {
if (update_status) {
CERROR("%s: cannot cleanup orphans: rc = %d\n",
d->opd_obd->obd_name, rc);
/* we can't proceed from here, OST seem to
* be in a bad shape, better to wait for
* a new instance of the server and repeat
* from the beginning. notify possible waiters
* this OSP isn't quite functional yet */
osp_pre_update_status(d, rc);
} else {
wake_up(&d->opd_pre_user_waitq);
}
}
RETURN(rc);
}
static int osp_get_lastfid_from_ost(const struct lu_env *env,
struct osp_device *d)
{
struct ptlrpc_request *req = NULL;
struct obd_import *imp;
struct lu_fid *last_fid;
char *tmp;
int rc;
ENTRY;
imp = d->opd_obd->u.cli.cl_import;
LASSERT(imp);
req = ptlrpc_request_alloc(imp, &RQF_OST_GET_INFO_LAST_FID);
if (req == NULL)
RETURN(-ENOMEM);
req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_KEY, RCL_CLIENT,
sizeof(KEY_LAST_FID));
rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_GET_INFO);
if (rc) {
ptlrpc_request_free(req);
RETURN(rc);
}
tmp = req_capsule_client_get(&req->rq_pill, &RMF_GETINFO_KEY);
memcpy(tmp, KEY_LAST_FID, sizeof(KEY_LAST_FID));
req->rq_no_delay = req->rq_no_resend = 1;
last_fid = req_capsule_client_get(&req->rq_pill, &RMF_FID);
fid_cpu_to_le(last_fid, &d->opd_last_used_fid);
ptlrpc_request_set_replen(req);
rc = ptlrpc_queue_wait(req);
if (rc) {
/* bad-bad OST.. let sysadm sort this out */
if (rc == -ENOTSUPP) {
CERROR("%s: server does not support FID: rc = %d\n",
d->opd_obd->obd_name, -ENOTSUPP);
}
ptlrpc_set_import_active(imp, 0);
GOTO(out, rc);
}
last_fid = req_capsule_server_get(&req->rq_pill, &RMF_FID);
if (last_fid == NULL) {
CERROR("%s: Got last_fid failed.\n", d->opd_obd->obd_name);
GOTO(out, rc = -EPROTO);
}
if (!fid_is_sane(last_fid)) {
CERROR("%s: Got insane last_fid "DFID"\n",
d->opd_obd->obd_name, PFID(last_fid));
GOTO(out, rc = -EPROTO);
}
/* Only update the last used fid, if the OST has objects for
* this sequence, i.e. fid_oid > 0 */
if (fid_oid(last_fid) > 0)
d->opd_last_used_fid = *last_fid;
CDEBUG(D_HA, "%s: Got last_fid "DFID"\n", d->opd_obd->obd_name,
PFID(last_fid));
out:
ptlrpc_req_finished(req);
RETURN(rc);
}
static int osp_precreate_send(const struct lu_env *env, struct osp_device *d)
{
struct osp_thread_info *oti = osp_env_info(env);
struct ptlrpc_request *req;
struct obd_import *imp;
struct ost_body *body;
int rc, grow, diff;
struct lu_fid *fid = &oti->osi_fid;
ENTRY;
/* don't precreate new objects till OST healthy and has free space */
if (unlikely(d->opd_pre_status)) {
CDEBUG(D_INFO, "%s: don't send new precreate: rc = %d\n",
d->opd_obd->obd_name, d->opd_pre_status);
RETURN(0);
}
/*
* if not connection/initialization is compeleted, ignore
*/
imp = d->opd_obd->u.cli.cl_import;
LASSERT(imp);
req = ptlrpc_request_alloc(imp, &RQF_OST_CREATE);
if (req == NULL)
RETURN(-ENOMEM);
req->rq_request_portal = OST_CREATE_PORTAL;
/* we should not resend create request - anyway we will have delorphan
* and kill these objects */
req->rq_no_delay = req->rq_no_resend = 1;
rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_CREATE);
if (rc) {
ptlrpc_request_free(req);
RETURN(rc);
}
spin_lock(&d->opd_pre_lock);
if (d->opd_pre_grow_count > d->opd_pre_max_grow_count / 2)
d->opd_pre_grow_count = d->opd_pre_max_grow_count / 2;
grow = d->opd_pre_grow_count;
spin_unlock(&d->opd_pre_lock);
body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
LASSERT(body);
*fid = d->opd_pre_last_created_fid;
rc = osp_precreate_fids(env, d, fid, &grow);
if (rc == 1) {
/* Current seq has been used up*/
if (!osp_is_fid_client(d)) {
osp_pre_update_status(d, -ENOSPC);
rc = -ENOSPC;
}
wake_up(&d->opd_pre_waitq);
GOTO(out_req, rc);
}
if (!osp_is_fid_client(d)) {
/* Non-FID client will always send seq 0 because of
* compatiblity */
LASSERTF(fid_is_idif(fid), "Invalid fid "DFID"\n", PFID(fid));
fid->f_seq = 0;
}
fid_to_ostid(fid, &body->oa.o_oi);
body->oa.o_valid = OBD_MD_FLGROUP;
ptlrpc_request_set_replen(req);
rc = ptlrpc_queue_wait(req);
if (rc) {
CERROR("%s: can't precreate: rc = %d\n", d->opd_obd->obd_name,
rc);
GOTO(out_req, rc);
}
LASSERT(req->rq_transno == 0);
body = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
if (body == NULL)
GOTO(out_req, rc = -EPROTO);
ostid_to_fid(fid, &body->oa.o_oi, d->opd_index);
LASSERTF(lu_fid_diff(fid, &d->opd_pre_used_fid) > 0,
"reply fid "DFID" pre used fid "DFID"\n", PFID(fid),
PFID(&d->opd_pre_used_fid));
diff = lu_fid_diff(fid, &d->opd_pre_last_created_fid);
spin_lock(&d->opd_pre_lock);
if (diff < grow) {
/* the OST has not managed to create all the
* objects we asked for */
d->opd_pre_grow_count = max(diff, OST_MIN_PRECREATE);
d->opd_pre_grow_slow = 1;
} else {
/* the OST is able to keep up with the work,
* we could consider increasing grow_count
* next time if needed */
d->opd_pre_grow_slow = 0;
}
d->opd_pre_last_created_fid = *fid;
spin_unlock(&d->opd_pre_lock);
CDEBUG(D_HA, "%s: current precreated pool: "DFID"-"DFID"\n",
d->opd_obd->obd_name, PFID(&d->opd_pre_used_fid),
PFID(&d->opd_pre_last_created_fid));
out_req:
/* now we can wakeup all users awaiting for objects */
osp_pre_update_status(d, rc);
wake_up(&d->opd_pre_user_waitq);
ptlrpc_req_finished(req);
RETURN(rc);
}
int osp_object_truncate(const struct lu_env *env, struct dt_object *dt,
__u64 size)
{
struct osp_device *d = lu2osp_dev(dt->do_lu.lo_dev);
struct ptlrpc_request *req = NULL;
struct obd_import *imp;
struct ost_body *body;
struct obdo *oa = NULL;
int rc;
ENTRY;
imp = d->opd_obd->u.cli.cl_import;
LASSERT(imp);
req = ptlrpc_request_alloc(imp, &RQF_OST_PUNCH);
if (req == NULL)
RETURN(-ENOMEM);
/* XXX: capa support? */
/* osc_set_capa_size(req, &RMF_CAPA1, capa); */
rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_PUNCH);
if (rc) {
ptlrpc_request_free(req);
RETURN(rc);
}
/*
* XXX: decide how do we do here with resend
* if we don't resend, then client may see wrong file size
* if we do resend, then MDS thread can get stuck for quite long
*/
req->rq_no_resend = req->rq_no_delay = 1;
req->rq_request_portal = OST_IO_PORTAL; /* bug 7198 */
ptlrpc_at_set_req_timeout(req);
OBD_ALLOC_PTR(oa);
if (oa == NULL)
GOTO(out, rc = -ENOMEM);
rc = fid_to_ostid(lu_object_fid(&dt->do_lu), &oa->o_oi);
LASSERT(rc == 0);
oa->o_size = size;
oa->o_blocks = OBD_OBJECT_EOF;
oa->o_valid = OBD_MD_FLSIZE | OBD_MD_FLBLOCKS |
OBD_MD_FLID | OBD_MD_FLGROUP;
body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
LASSERT(body);
lustre_set_wire_obdo(&req->rq_import->imp_connect_data, &body->oa, oa);
/* XXX: capa support? */
/* osc_pack_capa(req, body, capa); */
ptlrpc_request_set_replen(req);
rc = ptlrpc_queue_wait(req);
if (rc)
CERROR("can't punch object: %d\n", rc);
out:
ptlrpc_req_finished(req);
if (oa)
OBD_FREE_PTR(oa);
RETURN(rc);
}
int gss_do_ctx_fini_rpc(struct gss_cli_ctx *gctx)
{
struct ptlrpc_cli_ctx *ctx = &gctx->gc_base;
struct obd_import *imp = ctx->cc_sec->ps_import;
struct ptlrpc_request *req;
struct ptlrpc_user_desc *pud;
int rc;
ENTRY;
LASSERT(atomic_read(&ctx->cc_refcount) > 0);
if (cli_ctx_is_error(ctx) || !cli_ctx_is_uptodate(ctx)) {
CDEBUG(D_SEC, "ctx %p(%u->%s) not uptodate, "
"don't send destroy rpc\n", ctx,
ctx->cc_vcred.vc_uid, sec2target_str(ctx->cc_sec));
RETURN(0);
}
might_sleep();
CWARN("%s ctx %p idx "LPX64" (%u->%s)\n",
sec_is_reverse(ctx->cc_sec) ?
"server finishing reverse" : "client finishing forward",
ctx, gss_handle_to_u64(&gctx->gc_handle),
ctx->cc_vcred.vc_uid, sec2target_str(ctx->cc_sec));
gctx->gc_proc = PTLRPC_GSS_PROC_DESTROY;
req = ptlrpc_request_alloc(imp, &RQF_SEC_CTX);
if (req == NULL) {
CWARN("ctx %p(%u): fail to prepare rpc, destroy locally\n",
ctx, ctx->cc_vcred.vc_uid);
GOTO(out, rc = -ENOMEM);
}
rc = ptlrpc_request_bufs_pack(req, LUSTRE_OBD_VERSION, SEC_CTX_FINI,
NULL, ctx);
if (rc) {
ptlrpc_request_free(req);
GOTO(out_ref, rc);
}
/* fix the user desc */
if (req->rq_pack_udesc) {
/* we rely the fact that this request is in AUTH mode,
* and user_desc at offset 2. */
pud = lustre_msg_buf(req->rq_reqbuf, 2, sizeof(*pud));
LASSERT(pud);
pud->pud_uid = pud->pud_fsuid = ctx->cc_vcred.vc_uid;
pud->pud_gid = pud->pud_fsgid = ctx->cc_vcred.vc_gid;
pud->pud_cap = 0;
pud->pud_ngroups = 0;
}
req->rq_phase = RQ_PHASE_RPC;
rc = ptl_send_rpc(req, 1);
if (rc)
CWARN("ctx %p(%u->%s): rpc error %d, destroy locally\n", ctx,
ctx->cc_vcred.vc_uid, sec2target_str(ctx->cc_sec), rc);
out_ref:
ptlrpc_req_finished(req);
out:
RETURN(rc);
}
static int osc_object_fiemap(const struct lu_env *env, struct cl_object *obj,
struct ll_fiemap_info_key *fmkey,
struct fiemap *fiemap, size_t *buflen)
{
struct obd_export *exp = osc_export(cl2osc(obj));
struct ldlm_res_id resid;
union ldlm_policy_data policy;
struct lustre_handle lockh;
enum ldlm_mode mode = LCK_MINMODE;
struct ptlrpc_request *req;
struct fiemap *reply;
char *tmp;
int rc;
ENTRY;
fmkey->lfik_oa.o_oi = cl2osc(obj)->oo_oinfo->loi_oi;
if (!(fmkey->lfik_fiemap.fm_flags & FIEMAP_FLAG_SYNC))
goto skip_locking;
policy.l_extent.start = fmkey->lfik_fiemap.fm_start & PAGE_MASK;
if (OBD_OBJECT_EOF - fmkey->lfik_fiemap.fm_length <=
fmkey->lfik_fiemap.fm_start + PAGE_SIZE - 1)
policy.l_extent.end = OBD_OBJECT_EOF;
else
policy.l_extent.end = (fmkey->lfik_fiemap.fm_start +
fmkey->lfik_fiemap.fm_length +
PAGE_SIZE - 1) & PAGE_MASK;
ostid_build_res_name(&fmkey->lfik_oa.o_oi, &resid);
mode = ldlm_lock_match(exp->exp_obd->obd_namespace,
LDLM_FL_BLOCK_GRANTED | LDLM_FL_LVB_READY,
&resid, LDLM_EXTENT, &policy,
LCK_PR | LCK_PW, &lockh, 0);
if (mode) { /* lock is cached on client */
if (mode != LCK_PR) {
ldlm_lock_addref(&lockh, LCK_PR);
ldlm_lock_decref(&lockh, LCK_PW);
}
} else { /* no cached lock, needs acquire lock on server side */
fmkey->lfik_oa.o_valid |= OBD_MD_FLFLAGS;
fmkey->lfik_oa.o_flags |= OBD_FL_SRVLOCK;
}
skip_locking:
req = ptlrpc_request_alloc(class_exp2cliimp(exp),
&RQF_OST_GET_INFO_FIEMAP);
if (req == NULL)
GOTO(drop_lock, rc = -ENOMEM);
req_capsule_set_size(&req->rq_pill, &RMF_FIEMAP_KEY, RCL_CLIENT,
sizeof(*fmkey));
req_capsule_set_size(&req->rq_pill, &RMF_FIEMAP_VAL, RCL_CLIENT,
*buflen);
req_capsule_set_size(&req->rq_pill, &RMF_FIEMAP_VAL, RCL_SERVER,
*buflen);
rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_GET_INFO);
if (rc != 0) {
ptlrpc_request_free(req);
GOTO(drop_lock, rc);
}
tmp = req_capsule_client_get(&req->rq_pill, &RMF_FIEMAP_KEY);
memcpy(tmp, fmkey, sizeof(*fmkey));
tmp = req_capsule_client_get(&req->rq_pill, &RMF_FIEMAP_VAL);
memcpy(tmp, fiemap, *buflen);
ptlrpc_request_set_replen(req);
rc = ptlrpc_queue_wait(req);
if (rc != 0)
GOTO(fini_req, rc);
reply = req_capsule_server_get(&req->rq_pill, &RMF_FIEMAP_VAL);
if (reply == NULL)
GOTO(fini_req, rc = -EPROTO);
memcpy(fiemap, reply, *buflen);
fini_req:
ptlrpc_req_finished(req);
drop_lock:
if (mode)
ldlm_lock_decref(&lockh, LCK_PR);
RETURN(rc);
}
/*
* Get intent per-ID lock or global-index lock from master.
*
* \param env - the environment passed by the caller
* \param exp - is the export to use to send the intent RPC
* \param qbody - quota body to be packed in request
* \param sync - synchronous or asynchronous (pre-acquire)
* \param it_op - IT_QUOTA_DQACQ or IT_QUOTA_CONN
* \param completion - completion callback
* \param qqi - is the qsd_qtype_info structure to pass to the completion
* function
* \param lvb - is the lvb associated with the lock and returned by the
* server
* \param arg - is an opaq argument passed to the completion callback
*
* \retval 0 - success
* \retval -ve - appropriate errors
*/
int qsd_intent_lock(const struct lu_env *env, struct obd_export *exp,
struct quota_body *qbody, bool sync, int it_op,
qsd_req_completion_t completion, struct qsd_qtype_info *qqi,
struct lquota_lvb *lvb, void *arg)
{
struct qsd_thread_info *qti = qsd_info(env);
struct ptlrpc_request *req;
struct qsd_async_args *aa = NULL;
struct ldlm_intent *lit;
struct quota_body *req_qbody;
__u64 flags = LDLM_FL_HAS_INTENT;
int rc;
ENTRY;
LASSERT(exp != NULL);
LASSERT(!lustre_handle_is_used(&qbody->qb_lockh));
memset(&qti->qti_lockh, 0, sizeof(qti->qti_lockh));
req = ptlrpc_request_alloc(class_exp2cliimp(exp),
&RQF_LDLM_INTENT_QUOTA);
if (req == NULL)
GOTO(out, rc = -ENOMEM);
req->rq_no_retry_einprogress = 1;
rc = ldlm_prep_enqueue_req(exp, req, NULL, 0);
if (rc) {
ptlrpc_request_free(req);
GOTO(out, rc);
}
lit = req_capsule_client_get(&req->rq_pill, &RMF_LDLM_INTENT);
lit->opc = (__u64)it_op;
req_qbody = req_capsule_client_get(&req->rq_pill, &RMF_QUOTA_BODY);
*req_qbody = *qbody;
req_capsule_set_size(&req->rq_pill, &RMF_DLM_LVB, RCL_SERVER,
sizeof(*lvb));
ptlrpc_request_set_replen(req);
switch(it_op) {
case IT_QUOTA_CONN:
/* build resource name associated with global index */
fid_build_reg_res_name(&qbody->qb_fid, &qti->qti_resid);
/* copy einfo template and fill ei_cbdata with qqi pointer */
memcpy(&qti->qti_einfo, &qsd_glb_einfo, sizeof(qti->qti_einfo));
qti->qti_einfo.ei_cbdata = qqi;
/* don't cancel global lock on memory pressure */
flags |= LDLM_FL_NO_LRU;
break;
case IT_QUOTA_DQACQ:
/* build resource name associated for per-ID quota lock */
fid_build_quota_res_name(&qbody->qb_fid, &qbody->qb_id,
&qti->qti_resid);
/* copy einfo template and fill ei_cbdata with lqe pointer */
memcpy(&qti->qti_einfo, &qsd_id_einfo, sizeof(qti->qti_einfo));
qti->qti_einfo.ei_cbdata = arg;
break;
default:
LASSERTF(0, "invalid it_op %d", it_op);
}
/* build lock enqueue request */
rc = ldlm_cli_enqueue(exp, &req, &qti->qti_einfo, &qti->qti_resid, NULL,
&flags, (void *)lvb, sizeof(*lvb), LVB_T_LQUOTA,
&qti->qti_lockh, 1);
if (rc < 0) {
ptlrpc_req_finished(req);
GOTO(out, rc);
}
/* grab reference on backend structure for the new lock */
switch(it_op) {
case IT_QUOTA_CONN:
/* grab reference on qqi for new lock */
#ifdef USE_LU_REF
{
struct ldlm_lock *lock;
lock = ldlm_handle2lock(&qti->qti_lockh);
if (lock == NULL) {
ptlrpc_req_finished(req);
GOTO(out, rc = -ENOLCK);
}
lu_ref_add(&qqi->qqi_reference, "glb_lock", lock);
LDLM_LOCK_PUT(lock);
}
#endif
qqi_getref(qqi);
break;
case IT_QUOTA_DQACQ:
/* grab reference on lqe for new lock */
lqe_getref((struct lquota_entry *)arg);
/* all acquire/release request are sent with no_resend and
* no_delay flag */
req->rq_no_resend = req->rq_no_delay = 1;
break;
default:
break;
}
CLASSERT(sizeof(*aa) <= sizeof(req->rq_async_args));
aa = ptlrpc_req_async_args(req);
aa->aa_exp = exp;
aa->aa_qqi = qqi;
aa->aa_arg = arg;
aa->aa_lvb = lvb;
aa->aa_completion = completion;
lustre_handle_copy(&aa->aa_lockh, &qti->qti_lockh);
if (sync) {
/* send lock enqueue request and wait for completion */
rc = ptlrpc_queue_wait(req);
rc = qsd_intent_interpret(env, req, aa, rc);
ptlrpc_req_finished(req);
} else {
/* queue lock request and return */
req->rq_interpret_reply = qsd_intent_interpret;
ptlrpcd_add_req(req, PDL_POLICY_LOCAL, -1);
}
RETURN(rc);
out:
completion(env, qqi, qbody, NULL, &qti->qti_lockh, lvb, arg, rc);
return rc;
}
/*
* Fetch a global or slave index from the QMT.
*
* \param env - the environment passed by the caller
* \param exp - is the export to use to issue the OBD_IDX_READ RPC
* \param ii - is the index information to be packed in the request
* on success, the index information returned by the server
* is copied there.
* \param npages - is the number of pages in the pages array
* \param pages - is an array of @npages pages
*
* \retval 0 - success
* \retval -ve - appropriate errors
*/
int qsd_fetch_index(const struct lu_env *env, struct obd_export *exp,
struct idx_info *ii, unsigned int npages,
struct page **pages, bool *need_swab)
{
struct ptlrpc_request *req;
struct idx_info *req_ii;
struct ptlrpc_bulk_desc *desc;
int rc, i;
ENTRY;
LASSERT(exp);
req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_OBD_IDX_READ);
if (req == NULL)
RETURN(-ENOMEM);
rc = ptlrpc_request_pack(req, LUSTRE_OBD_VERSION, OBD_IDX_READ);
if (rc) {
ptlrpc_request_free(req);
RETURN(rc);
}
req->rq_request_portal = MDS_READPAGE_PORTAL;
ptlrpc_at_set_req_timeout(req);
/* allocate bulk descriptor */
desc = ptlrpc_prep_bulk_imp(req, npages, 1, BULK_PUT_SINK,
MDS_BULK_PORTAL);
if (desc == NULL) {
ptlrpc_request_free(req);
RETURN(-ENOMEM);
}
/* req now owns desc and will free it when it gets freed */
for (i = 0; i < npages; i++)
ptlrpc_prep_bulk_page_pin(desc, pages[i], 0, PAGE_CACHE_SIZE);
/* pack index information in request */
req_ii = req_capsule_client_get(&req->rq_pill, &RMF_IDX_INFO);
*req_ii = *ii;
ptlrpc_request_set_replen(req);
/* send request to master and wait for RPC to complete */
rc = ptlrpc_queue_wait(req);
if (rc)
GOTO(out, rc);
rc = sptlrpc_cli_unwrap_bulk_read(req, req->rq_bulk,
req->rq_bulk->bd_nob_transferred);
if (rc < 0)
GOTO(out, rc);
else
/* sptlrpc_cli_unwrap_bulk_read() returns the number of bytes
* transferred*/
rc = 0;
req_ii = req_capsule_server_get(&req->rq_pill, &RMF_IDX_INFO);
*ii = *req_ii;
*need_swab = ptlrpc_rep_need_swab(req);
EXIT;
out:
ptlrpc_req_finished(req);
return rc;
}
/* This is a callback from the llog_* functions.
* Assumes caller has already pushed us into the kernel context.
*/
static int llog_client_open(const struct lu_env *env,
struct llog_handle *lgh, struct llog_logid *logid,
char *name, enum llog_open_param open_param)
{
struct obd_import *imp;
struct llogd_body *body;
struct llog_ctxt *ctxt = lgh->lgh_ctxt;
struct ptlrpc_request *req = NULL;
int rc;
LLOG_CLIENT_ENTRY(ctxt, imp);
/* client cannot create llog */
LASSERTF(open_param != LLOG_OPEN_NEW, "%#x\n", open_param);
LASSERT(lgh);
req = ptlrpc_request_alloc(imp, &RQF_LLOG_ORIGIN_HANDLE_CREATE);
if (!req) {
rc = -ENOMEM;
goto out;
}
if (name)
req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
strlen(name) + 1);
rc = ptlrpc_request_pack(req, LUSTRE_LOG_VERSION,
LLOG_ORIGIN_HANDLE_CREATE);
if (rc) {
ptlrpc_request_free(req);
req = NULL;
goto out;
}
ptlrpc_request_set_replen(req);
body = req_capsule_client_get(&req->rq_pill, &RMF_LLOGD_BODY);
if (logid)
body->lgd_logid = *logid;
body->lgd_ctxt_idx = ctxt->loc_idx - 1;
if (name) {
char *tmp;
tmp = req_capsule_client_sized_get(&req->rq_pill, &RMF_NAME,
strlen(name) + 1);
LASSERT(tmp);
strcpy(tmp, name);
}
rc = ptlrpc_queue_wait(req);
if (rc)
goto out;
body = req_capsule_server_get(&req->rq_pill, &RMF_LLOGD_BODY);
if (!body) {
rc = -EFAULT;
goto out;
}
lgh->lgh_id = body->lgd_logid;
lgh->lgh_ctxt = ctxt;
out:
LLOG_CLIENT_EXIT(ctxt, imp);
ptlrpc_req_finished(req);
return rc;
}
int mdc_create(struct obd_export *exp, struct md_op_data *op_data,
const void *data, int datalen, int mode, __u32 uid, __u32 gid,
cfs_cap_t cap_effective, __u64 rdev,
struct ptlrpc_request **request)
{
struct ptlrpc_request *req;
int level, rc;
int count, resends = 0;
struct obd_import *import = exp->exp_obd->u.cli.cl_import;
int generation = import->imp_generation;
CFS_LIST_HEAD(cancels);
ENTRY;
/* For case if upper layer did not alloc fid, do it now. */
if (!fid_is_sane(&op_data->op_fid2)) {
/*
* mdc_fid_alloc() may return errno 1 in case of switch to new
* sequence, handle this.
*/
rc = mdc_fid_alloc(exp, &op_data->op_fid2, op_data);
if (rc < 0) {
CERROR("Can't alloc new fid, rc %d\n", rc);
RETURN(rc);
}
}
rebuild:
count = 0;
if ((op_data->op_flags & MF_MDC_CANCEL_FID1) &&
(fid_is_sane(&op_data->op_fid1)))
count = mdc_resource_get_unused(exp, &op_data->op_fid1,
&cancels, LCK_EX,
MDS_INODELOCK_UPDATE);
req = ptlrpc_request_alloc(class_exp2cliimp(exp),
&RQF_MDS_REINT_CREATE_RMT_ACL);
if (req == NULL) {
ldlm_lock_list_put(&cancels, l_bl_ast, count);
RETURN(-ENOMEM);
}
mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
op_data->op_namelen + 1);
req_capsule_set_size(&req->rq_pill, &RMF_EADATA, RCL_CLIENT,
data && datalen ? datalen : 0);
rc = mdc_prep_elc_req(exp, req, MDS_REINT, &cancels, count);
if (rc) {
ptlrpc_request_free(req);
RETURN(rc);
}
/*
* mdc_create_pack() fills msg->bufs[1] with name and msg->bufs[2] with
* tgt, for symlinks or lov MD data.
*/
mdc_create_pack(req, op_data, data, datalen, mode, uid,
gid, cap_effective, rdev);
ptlrpc_request_set_replen(req);
/* ask ptlrpc not to resend on EINPROGRESS since we have our own retry
* logic here */
req->rq_no_retry_einprogress = 1;
if (resends) {
req->rq_generation_set = 1;
req->rq_import_generation = generation;
req->rq_sent = cfs_time_current_sec() + resends;
}
level = LUSTRE_IMP_FULL;
resend:
rc = mdc_reint(req, exp->exp_obd->u.cli.cl_rpc_lock, level);
/* Resend if we were told to. */
if (rc == -ERESTARTSYS) {
level = LUSTRE_IMP_RECOVER;
goto resend;
} else if (rc == -EINPROGRESS) {
/* Retry create infinitely until succeed or get other
* error code. */
ptlrpc_req_finished(req);
resends++;
CDEBUG(D_HA, "%s: resend:%d create on "DFID"/"DFID"\n",
exp->exp_obd->obd_name, resends,
PFID(&op_data->op_fid1), PFID(&op_data->op_fid2));
if (generation == import->imp_generation) {
goto rebuild;
} else {
CDEBUG(D_HA, "resend cross eviction\n");
RETURN(-EIO);
}
} else if (rc == 0) {
struct mdt_body *body;
struct lustre_capa *capa;
body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
LASSERT(body);
if (body->valid & OBD_MD_FLMDSCAPA) {
capa = req_capsule_server_get(&req->rq_pill,
&RMF_CAPA1);
if (capa == NULL)
rc = -EPROTO;
}
}
*request = req;
RETURN(rc);
}
int mdc_create(struct obd_export *exp, struct md_op_data *op_data,
const void *data, int datalen, int mode, __u32 uid, __u32 gid,
cfs_cap_t cap_effective, __u64 rdev,
struct ptlrpc_request **request)
{
struct ptlrpc_request *req;
int level, rc;
int count = 0;
CFS_LIST_HEAD(cancels);
ENTRY;
/* For case if upper layer did not alloc fid, do it now. */
if (!fid_is_sane(&op_data->op_fid2)) {
/*
* mdc_fid_alloc() may return errno 1 in case of switch to new
* sequence, handle this.
*/
rc = mdc_fid_alloc(exp, &op_data->op_fid2, op_data);
if (rc < 0) {
CERROR("Can't alloc new fid, rc %d\n", rc);
RETURN(rc);
}
}
if ((op_data->op_flags & MF_MDC_CANCEL_FID1) &&
(fid_is_sane(&op_data->op_fid1)))
count = mdc_resource_get_unused(exp, &op_data->op_fid1,
&cancels, LCK_EX,
MDS_INODELOCK_UPDATE);
req = ptlrpc_request_alloc(class_exp2cliimp(exp),
&RQF_MDS_REINT_CREATE_RMT_ACL);
if (req == NULL) {
ldlm_lock_list_put(&cancels, l_bl_ast, count);
RETURN(-ENOMEM);
}
mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
op_data->op_namelen + 1);
req_capsule_set_size(&req->rq_pill, &RMF_EADATA, RCL_CLIENT,
data && datalen ? datalen : 0);
rc = mdc_prep_elc_req(exp, req, &cancels, count);
if (rc) {
ptlrpc_request_free(req);
RETURN(rc);
}
/*
* mdc_create_pack() fills msg->bufs[1] with name and msg->bufs[2] with
* tgt, for symlinks or lov MD data.
*/
mdc_create_pack(req, op_data, data, datalen, mode, uid,
gid, cap_effective, rdev);
ptlrpc_request_set_replen(req);
level = LUSTRE_IMP_FULL;
resend:
rc = mdc_reint(req, exp->exp_obd->u.cli.cl_rpc_lock, level);
/* Resend if we were told to. */
if (rc == -ERESTARTSYS) {
level = LUSTRE_IMP_RECOVER;
goto resend;
} else if (rc == 0) {
struct mdt_body *body;
struct lustre_capa *capa;
body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
LASSERT(body);
if (body->valid & OBD_MD_FLMDSCAPA) {
capa = req_capsule_server_get(&req->rq_pill,
&RMF_CAPA1);
if (capa == NULL)
rc = -EPROTO;
}
}
*request = req;
RETURN(rc);
}
/* If mdc_setattr is called with an 'iattr', then it is a normal RPC that
* should take the normal semaphore and go to the normal portal.
*
* If it is called with iattr->ia_valid & ATTR_FROM_OPEN, then it is a
* magic open-path setattr that should take the setattr semaphore and
* go to the setattr portal. */
int mdc_setattr(struct obd_export *exp, struct md_op_data *op_data,
void *ea, int ealen, void *ea2, int ea2len,
struct ptlrpc_request **request, struct md_open_data **mod)
{
CFS_LIST_HEAD(cancels);
struct ptlrpc_request *req;
struct mdc_rpc_lock *rpc_lock;
struct obd_device *obd = exp->exp_obd;
int count = 0, rc;
__u64 bits;
ENTRY;
LASSERT(op_data != NULL);
bits = MDS_INODELOCK_UPDATE;
if (op_data->op_attr.ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID))
bits |= MDS_INODELOCK_LOOKUP;
if ((op_data->op_flags & MF_MDC_CANCEL_FID1) &&
(fid_is_sane(&op_data->op_fid1)))
count = mdc_resource_get_unused(exp, &op_data->op_fid1,
&cancels, LCK_EX, bits);
req = ptlrpc_request_alloc(class_exp2cliimp(exp),
&RQF_MDS_REINT_SETATTR);
if (req == NULL) {
ldlm_lock_list_put(&cancels, l_bl_ast, count);
RETURN(-ENOMEM);
}
mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
if ((op_data->op_flags & (MF_SOM_CHANGE | MF_EPOCH_OPEN)) == 0)
req_capsule_set_size(&req->rq_pill, &RMF_MDT_EPOCH, RCL_CLIENT,
0);
req_capsule_set_size(&req->rq_pill, &RMF_EADATA, RCL_CLIENT, ealen);
req_capsule_set_size(&req->rq_pill, &RMF_LOGCOOKIES, RCL_CLIENT,
ea2len);
rc = mdc_prep_elc_req(exp, req, &cancels, count);
if (rc) {
ptlrpc_request_free(req);
RETURN(rc);
}
if (op_data->op_attr.ia_valid & ATTR_FROM_OPEN) {
req->rq_request_portal = MDS_SETATTR_PORTAL;
ptlrpc_at_set_req_timeout(req);
rpc_lock = obd->u.cli.cl_setattr_lock;
} else {
rpc_lock = obd->u.cli.cl_rpc_lock;
}
if (op_data->op_attr.ia_valid & (ATTR_MTIME | ATTR_CTIME))
CDEBUG(D_INODE, "setting mtime "CFS_TIME_T
", ctime "CFS_TIME_T"\n",
LTIME_S(op_data->op_attr.ia_mtime),
LTIME_S(op_data->op_attr.ia_ctime));
mdc_setattr_pack(req, op_data, ea, ealen, ea2, ea2len);
ptlrpc_request_set_replen(req);
if (mod && (op_data->op_flags & MF_EPOCH_OPEN) &&
req->rq_import->imp_replayable)
{
LASSERT(*mod == NULL);
*mod = obd_mod_alloc();
if (*mod == NULL) {
DEBUG_REQ(D_ERROR, req, "Can't allocate "
"md_open_data");
} else {
req->rq_replay = 1;
req->rq_cb_data = *mod;
(*mod)->mod_open_req = req;
req->rq_commit_cb = mdc_commit_open;
/**
* Take an extra reference on \var mod, it protects \var
* mod from being freed on eviction (commit callback is
* called despite rq_replay flag).
* Will be put on mdc_done_writing().
*/
obd_mod_get(*mod);
}
}
rc = mdc_reint(req, rpc_lock, LUSTRE_IMP_FULL);
/* Save the obtained info in the original RPC for the replay case. */
if (rc == 0 && (op_data->op_flags & MF_EPOCH_OPEN)) {
struct mdt_ioepoch *epoch;
struct mdt_body *body;
epoch = req_capsule_client_get(&req->rq_pill, &RMF_MDT_EPOCH);
body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
LASSERT(epoch != NULL);
LASSERT(body != NULL);
epoch->handle = body->handle;
epoch->ioepoch = body->ioepoch;
req->rq_replay_cb = mdc_replay_open;
/** bug 3633, open may be committed and estale answer is not error */
} else if (rc == -ESTALE && (op_data->op_flags & MF_SOM_CHANGE)) {
rc = 0;
} else if (rc == -ERESTARTSYS) {
rc = 0;
}
*request = req;
if (rc && req->rq_commit_cb) {
/* Put an extra reference on \var mod on error case. */
obd_mod_put(*mod);
req->rq_commit_cb(req);
}
RETURN(rc);
}
int mdc_create(struct obd_export *exp, struct md_op_data *op_data,
const void *data, size_t datalen,
umode_t mode, uid_t uid, gid_t gid,
cfs_cap_t cap_effective, __u64 rdev,
struct ptlrpc_request **request)
{
struct ptlrpc_request *req;
int level, rc;
int count, resends = 0;
struct obd_import *import = exp->exp_obd->u.cli.cl_import;
int generation = import->imp_generation;
struct list_head cancels = LIST_HEAD_INIT(cancels);
ENTRY;
/* For case if upper layer did not alloc fid, do it now. */
if (!fid_is_sane(&op_data->op_fid2)) {
/*
* mdc_fid_alloc() may return errno 1 in case of switch to new
* sequence, handle this.
*/
rc = mdc_fid_alloc(NULL, exp, &op_data->op_fid2, op_data);
if (rc < 0)
RETURN(rc);
}
rebuild:
count = 0;
if ((op_data->op_flags & MF_MDC_CANCEL_FID1) &&
(fid_is_sane(&op_data->op_fid1)))
count = mdc_resource_get_unused(exp, &op_data->op_fid1,
&cancels, LCK_EX,
MDS_INODELOCK_UPDATE);
req = ptlrpc_request_alloc(class_exp2cliimp(exp),
&RQF_MDS_REINT_CREATE_ACL);
if (req == NULL) {
ldlm_lock_list_put(&cancels, l_bl_ast, count);
RETURN(-ENOMEM);
}
req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
op_data->op_namelen + 1);
req_capsule_set_size(&req->rq_pill, &RMF_EADATA, RCL_CLIENT,
data && datalen ? datalen : 0);
req_capsule_set_size(&req->rq_pill, &RMF_FILE_SECCTX_NAME,
RCL_CLIENT, op_data->op_file_secctx_name != NULL ?
strlen(op_data->op_file_secctx_name) + 1 : 0);
req_capsule_set_size(&req->rq_pill, &RMF_FILE_SECCTX, RCL_CLIENT,
op_data->op_file_secctx_size);
/* get SELinux policy info if any */
rc = sptlrpc_get_sepol(req);
if (rc < 0) {
ptlrpc_request_free(req);
RETURN(rc);
}
req_capsule_set_size(&req->rq_pill, &RMF_SELINUX_POL, RCL_CLIENT,
strlen(req->rq_sepol) ?
strlen(req->rq_sepol) + 1 : 0);
rc = mdc_prep_elc_req(exp, req, MDS_REINT, &cancels, count);
if (rc) {
ptlrpc_request_free(req);
RETURN(rc);
}
/*
* mdc_create_pack() fills msg->bufs[1] with name and msg->bufs[2] with
* tgt, for symlinks or lov MD data.
*/
mdc_create_pack(req, op_data, data, datalen, mode, uid,
gid, cap_effective, rdev);
ptlrpc_request_set_replen(req);
/* ask ptlrpc not to resend on EINPROGRESS since we have our own retry
* logic here */
req->rq_no_retry_einprogress = 1;
if (resends) {
req->rq_generation_set = 1;
req->rq_import_generation = generation;
req->rq_sent = ktime_get_real_seconds() + resends;
}
level = LUSTRE_IMP_FULL;
resend:
rc = mdc_reint(req, level);
/* Resend if we were told to. */
if (rc == -ERESTARTSYS) {
level = LUSTRE_IMP_RECOVER;
goto resend;
} else if (rc == -EINPROGRESS) {
/* Retry create infinitely until succeed or get other
* error code or interrupted. */
ptlrpc_req_finished(req);
if (generation == import->imp_generation) {
if (signal_pending(current))
RETURN(-EINTR);
resends++;
CDEBUG(D_HA, "%s: resend:%d create on "DFID"/"DFID"\n",
exp->exp_obd->obd_name, resends,
PFID(&op_data->op_fid1),
PFID(&op_data->op_fid2));
goto rebuild;
} else {
CDEBUG(D_HA, "resend cross eviction\n");
RETURN(-EIO);
}
}
*request = req;
RETURN(rc);
}
