/**
* Check whether a slave already own a ldlm lock for the quota identifier \qid.
*
* \param lockh - is the local lock handle from lquota entry.
* \param rlockh - is the remote lock handle of the matched lock, if any.
*
* \retval 0 : on successful look up and \lockh contains the lock handle.
* \retval -ENOENT: no lock found
*/
int qsd_id_lock_match(struct lustre_handle *lockh, struct lustre_handle *rlockh)
{
struct ldlm_lock *lock;
int rc;
ENTRY;
LASSERT(lockh);
if (!lustre_handle_is_used(lockh))
RETURN(-ENOENT);
rc = ldlm_lock_addref_try(lockh, qsd_id_einfo.ei_mode);
if (rc)
RETURN(-ENOENT);
LASSERT(lustre_handle_is_used(lockh));
ldlm_lock_dump_handle(D_QUOTA, lockh);
if (rlockh == NULL)
/* caller not interested in remote handle */
RETURN(0);
/* look up lock associated with local handle and extract remote handle
* to be packed in quota request */
lock = ldlm_handle2lock(lockh);
LASSERT(lock != NULL);
lustre_handle_copy(rlockh, &lock->l_remote_handle);
LDLM_LOCK_PUT(lock);
RETURN(0);
}
int qsd_id_lock_cancel(const struct lu_env *env, struct lquota_entry *lqe)
{
struct qsd_thread_info *qti = qsd_info(env);
int rc;
ENTRY;
lqe_write_lock(lqe);
if (lqe->lqe_pending_write || lqe->lqe_waiting_write ||
lqe->lqe_usage || lqe->lqe_granted) {
lqe_write_unlock(lqe);
RETURN(0);
}
lustre_handle_copy(&qti->qti_lockh, &lqe->lqe_lockh);
if (lustre_handle_is_used(&qti->qti_lockh)) {
memset(&lqe->lqe_lockh, 0, sizeof(lqe->lqe_lockh));
qsd_set_qunit(lqe, 0);
lqe->lqe_edquot = false;
}
lqe_write_unlock(lqe);
rc = qsd_id_lock_match(&qti->qti_lockh, NULL);
if (rc)
RETURN(rc);
ldlm_lock_decref_and_cancel(&qti->qti_lockh, qsd_id_einfo.ei_mode);
RETURN(0);
}
/**
* Invariant that has to be true all of the time.
*/
static int osc_lock_invariant(struct osc_lock *ols)
{
struct ldlm_lock *lock = osc_handle_ptr(&ols->ols_handle);
struct ldlm_lock *olock = ols->ols_lock;
int handle_used = lustre_handle_is_used(&ols->ols_handle);
return
ergo(osc_lock_is_lockless(ols),
ols->ols_locklessable && ols->ols_lock == NULL) ||
(ergo(olock != NULL, handle_used) &&
ergo(olock != NULL,
olock->l_handle.h_cookie == ols->ols_handle.cookie) &&
/*
* Check that ->ols_handle and ->ols_lock are consistent, but
* take into account that they are set at the different time.
*/
ergo(handle_used,
ergo(lock != NULL && olock != NULL, lock == olock) &&
ergo(lock == NULL, olock == NULL)) &&
ergo(ols->ols_state == OLS_CANCELLED,
olock == NULL && !handle_used) &&
/*
* DLM lock is destroyed only after we have seen cancellation
* ast.
*/
ergo(olock != NULL && ols->ols_state < OLS_CANCELLED,
!olock->l_destroyed) &&
ergo(ols->ols_state == OLS_GRANTED,
olock != NULL &&
olock->l_req_mode == olock->l_granted_mode &&
ols->ols_hold));
}
static void mdt_rename_unlock(struct lustre_handle *lh)
{
ENTRY;
LASSERT(lustre_handle_is_used(lh));
ldlm_lock_decref(lh, LCK_EX);
EXIT;
}
/**
* Check whether a qsd instance is all set to send quota request to master.
* This includes checking whether:
* - the connection to master is set up and usable,
* - the qsd isn't stopping
* - reintegration has been successfully completed and all indexes are
* up-to-date
*
* \param lqe - is the lquota entry for which we would like to send an quota
* request
* \param lockh - is the remote handle of the global lock returned on success
*
* \retval 0 on success, appropriate error on failure
*/
static int qsd_ready(struct lquota_entry *lqe, struct lustre_handle *lockh)
{
struct qsd_qtype_info *qqi = lqe2qqi(lqe);
struct qsd_instance *qsd = qqi->qqi_qsd;
struct obd_import *imp = NULL;
struct ldlm_lock *lock;
ENTRY;
read_lock(&qsd->qsd_lock);
/* is the qsd about to shut down? */
if (qsd->qsd_stopping) {
read_unlock(&qsd->qsd_lock);
LQUOTA_DEBUG(lqe, "dropping quota req since qsd is stopping");
/* Target is about to shut down, client will retry */
RETURN(-EINPROGRESS);
}
/* is the connection to the quota master ready? */
if (qsd->qsd_exp_valid)
imp = class_exp2cliimp(qsd->qsd_exp);
if (imp == NULL || imp->imp_invalid) {
read_unlock(&qsd->qsd_lock);
LQUOTA_DEBUG(lqe, "connection to master not ready");
RETURN(-ENOTCONN);
}
/* In most case, reintegration must have been triggered (when enable
* quota or on OST start), however, in rare race condition (enabling
* quota when starting OSTs), we might miss triggering reintegration
* for some qqi.
*
* If the previous reintegration failed for some reason, we'll
* re-trigger it here as well. */
if (!qqi->qqi_glb_uptodate || !qqi->qqi_slv_uptodate) {
read_unlock(&qsd->qsd_lock);
LQUOTA_DEBUG(lqe, "not up-to-date, dropping request and "
"kicking off reintegration");
qsd_start_reint_thread(qqi);
RETURN(-EINPROGRESS);
}
/* Fill the remote global lock handle, master will check this handle
* to see if the slave is sending request with stale lock */
lustre_handle_copy(lockh, &qqi->qqi_lockh);
read_unlock(&qsd->qsd_lock);
if (!lustre_handle_is_used(lockh))
RETURN(-ENOLCK);
lock = ldlm_handle2lock(lockh);
if (lock == NULL)
RETURN(-ENOLCK);
/* return remote lock handle to be packed in quota request */
lustre_handle_copy(lockh, &lock->l_remote_handle);
LDLM_LOCK_PUT(lock);
RETURN(0);
}
/**
* Invariant that has to be true all of the time.
*/
static int osc_lock_invariant(struct osc_lock *ols)
{
struct ldlm_lock *lock = osc_handle_ptr(&ols->ols_handle);
struct ldlm_lock *olock = ols->ols_lock;
int handle_used = lustre_handle_is_used(&ols->ols_handle);
if (ergo(osc_lock_is_lockless(ols),
ols->ols_locklessable && ols->ols_lock == NULL))
return 1;
/*
* If all the following "ergo"s are true, return 1, otherwise 0
*/
if (!ergo(olock != NULL, handle_used))
return 0;
if (!ergo(olock != NULL,
olock->l_handle.h_cookie == ols->ols_handle.cookie))
return 0;
if (!ergo(handle_used,
ergo(lock != NULL && olock != NULL, lock == olock) &&
ergo(lock == NULL, olock == NULL)))
return 0;
/*
* Check that ->ols_handle and ->ols_lock are consistent, but
* take into account that they are set at the different time.
*/
if (!ergo(ols->ols_state == OLS_CANCELLED,
olock == NULL && !handle_used))
return 0;
/*
* DLM lock is destroyed only after we have seen cancellation
* ast.
*/
if (!ergo(olock != NULL && ols->ols_state < OLS_CANCELLED,
((olock->l_flags & LDLM_FL_DESTROYED) == 0)))
return 0;
if (!ergo(ols->ols_state == OLS_GRANTED,
olock != NULL &&
olock->l_req_mode == olock->l_granted_mode &&
ols->ols_hold))
return 0;
return 1;
}
/*
* 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;
}
/*
* Handle quota request from slave.
*
* \param env - is the environment passed by the caller
* \param ld - is the lu device associated with the qmt
* \param req - is the quota acquire request
*/
static int qmt_dqacq(const struct lu_env *env, struct lu_device *ld,
struct ptlrpc_request *req)
{
struct qmt_device *qmt = lu2qmt_dev(ld);
struct quota_body *qbody, *repbody;
struct obd_uuid *uuid;
struct ldlm_lock *lock;
struct lquota_entry *lqe;
int pool_id, pool_type, qtype;
int rc;
ENTRY;
qbody = req_capsule_client_get(&req->rq_pill, &RMF_QUOTA_BODY);
if (qbody == NULL)
RETURN(err_serious(-EPROTO));
repbody = req_capsule_server_get(&req->rq_pill, &RMF_QUOTA_BODY);
if (repbody == NULL)
RETURN(err_serious(-EFAULT));
/* verify if global lock is stale */
if (!lustre_handle_is_used(&qbody->qb_glb_lockh))
RETURN(-ENOLCK);
lock = ldlm_handle2lock(&qbody->qb_glb_lockh);
if (lock == NULL)
RETURN(-ENOLCK);
LDLM_LOCK_PUT(lock);
uuid = &req->rq_export->exp_client_uuid;
if (req_is_rel(qbody->qb_flags) + req_is_acq(qbody->qb_flags) +
req_is_preacq(qbody->qb_flags) > 1) {
CERROR("%s: malformed quota request with conflicting flags set "
"(%x) from slave %s\n", qmt->qmt_svname,
qbody->qb_flags, obd_uuid2str(uuid));
RETURN(-EPROTO);
}
if (req_is_acq(qbody->qb_flags) || req_is_preacq(qbody->qb_flags)) {
/* acquire and pre-acquire should use a valid ID lock */
if (!lustre_handle_is_used(&qbody->qb_lockh))
RETURN(-ENOLCK);
lock = ldlm_handle2lock(&qbody->qb_lockh);
if (lock == NULL)
/* no lock associated with this handle */
RETURN(-ENOLCK);
LDLM_DEBUG(lock, "%sacquire request",
req_is_preacq(qbody->qb_flags) ? "pre" : "");
if (!obd_uuid_equals(&lock->l_export->exp_client_uuid, uuid)) {
/* sorry, no way to cheat ... */
LDLM_LOCK_PUT(lock);
RETURN(-ENOLCK);
}
if ((lock->l_flags & LDLM_FL_AST_SENT) != 0) {
struct ptlrpc_service_part *svc;
unsigned int timeout;
svc = req->rq_rqbd->rqbd_svcpt;
timeout = at_est2timeout(at_get(&svc->scp_at_estimate));
timeout = max(timeout, ldlm_timeout);
/* lock is being cancelled, prolong timeout */
ldlm_refresh_waiting_lock(lock, timeout);
}
LDLM_LOCK_PUT(lock);
}
/* extract pool & quota information from global index FID packed in the
* request */
rc = lquota_extract_fid(&qbody->qb_fid, &pool_id, &pool_type, &qtype);
if (rc)
RETURN(-EINVAL);
/* Find the quota entry associated with the quota id */
lqe = qmt_pool_lqe_lookup(env, qmt, pool_id, pool_type, qtype,
&qbody->qb_id);
if (IS_ERR(lqe))
RETURN(PTR_ERR(lqe));
/* process quota request */
rc = qmt_dqacq0(env, lqe, qmt, uuid, qbody->qb_flags, qbody->qb_count,
qbody->qb_usage, repbody);
if (lustre_handle_is_used(&qbody->qb_lockh))
/* return current qunit value only to slaves owning an per-ID
* quota lock. For enqueue, the qunit value will be returned in
* the LVB */
repbody->qb_qunit = lqe->lqe_qunit;
lqe_putref(lqe);
RETURN(rc);
}
/**
* Callback function called when an acquire/release request sent to the master
* is completed
*/
static void qsd_req_completion(const struct lu_env *env,
struct qsd_qtype_info *qqi,
struct quota_body *reqbody,
struct quota_body *repbody,
struct lustre_handle *lockh,
struct lquota_lvb *lvb,
void *arg, int ret)
{
struct lquota_entry *lqe = (struct lquota_entry *)arg;
struct qsd_thread_info *qti;
int rc;
bool adjust = false, cancel = false;
ENTRY;
LASSERT(qqi != NULL && lqe != NULL);
/* environment passed by ptlrpcd is mostly used by CLIO and hasn't the
* DT tags set. */
rc = lu_env_refill_by_tags((struct lu_env *)env, LCT_DT_THREAD, 0);
if (rc) {
LQUOTA_ERROR(lqe, "failed to refill environmnent %d", rc);
lqe_write_lock(lqe);
/* can't afford to adjust quota space with no suitable lu_env */
GOTO(out_noadjust, rc);
}
qti = qsd_info(env);
lqe_write_lock(lqe);
LQUOTA_DEBUG(lqe, "DQACQ returned %d, flags:0x%x", ret,
reqbody->qb_flags);
/* despite -EDQUOT & -EINPROGRESS errors, the master might still
* grant us back quota space to adjust quota overrun */
if (ret != 0 && ret != -EDQUOT && ret != -EINPROGRESS) {
if (ret != -ETIMEDOUT && ret != -ENOTCONN &&
ret != -ESHUTDOWN && ret != -EAGAIN)
/* print errors only if return code is unexpected */
LQUOTA_ERROR(lqe, "DQACQ failed with %d, flags:0x%x",
ret, reqbody->qb_flags);
GOTO(out, ret);
}
/* Set the lqe_lockh */
if (lustre_handle_is_used(lockh) &&
!lustre_handle_equal(lockh, &lqe->lqe_lockh))
lustre_handle_copy(&lqe->lqe_lockh, lockh);
/* If the replied qb_count is zero, it means master didn't process
* the DQACQ since the limit for this ID has been removed, so we
* should not update quota entry & slave index copy neither. */
if (repbody != NULL && repbody->qb_count != 0) {
LQUOTA_DEBUG(lqe, "DQACQ qb_count:"LPU64, repbody->qb_count);
if (req_is_rel(reqbody->qb_flags)) {
if (lqe->lqe_granted < repbody->qb_count) {
LQUOTA_ERROR(lqe, "can't release more space "
"than owned "LPU64"<"LPU64,
lqe->lqe_granted,
repbody->qb_count);
lqe->lqe_granted = 0;
} else {
lqe->lqe_granted -= repbody->qb_count;
}
/* Cancel the per-ID lock initiatively when there
* isn't any usage & grant, which can avoid master
* sending glimpse unnecessarily to this slave on
* quota revoking */
if (!lqe->lqe_pending_write && !lqe->lqe_granted &&
!lqe->lqe_waiting_write && !lqe->lqe_usage)
cancel = true;
} else {
lqe->lqe_granted += repbody->qb_count;
}
qti->qti_rec.lqr_slv_rec.qsr_granted = lqe->lqe_granted;
lqe_write_unlock(lqe);
/* Update the slave index file in the dedicated thread. So far,
* We don't update the version of slave index copy on DQACQ.
* No locking is necessary since nobody can change
* lqe->lqe_granted while lqe->lqe_pending_req > 0 */
qsd_upd_schedule(qqi, lqe, &lqe->lqe_id, &qti->qti_rec, 0,
false);
lqe_write_lock(lqe);
}
/* extract information from lvb */
if (ret == 0 && lvb != 0) {
if (lvb->lvb_id_qunit != 0)
qsd_set_qunit(lqe, lvb->lvb_id_qunit);
qsd_set_edquot(lqe, !!(lvb->lvb_flags & LQUOTA_FL_EDQUOT));
} else if (repbody != NULL && repbody->qb_qunit != 0) {
qsd_set_qunit(lqe, repbody->qb_qunit);
}
/* turn off pre-acquire if it failed with -EDQUOT. This is done to avoid
* flooding the master with acquire request. Pre-acquire will be turned
* on again as soon as qunit is modified */
if (req_is_preacq(reqbody->qb_flags) && ret == -EDQUOT)
lqe->lqe_nopreacq = true;
out:
adjust = qsd_adjust_needed(lqe);
if (reqbody && req_is_acq(reqbody->qb_flags) && ret != -EDQUOT) {
lqe->lqe_acq_rc = ret;
lqe->lqe_acq_time = cfs_time_current_64();
}
out_noadjust:
qsd_request_exit(lqe);
lqe_write_unlock(lqe);
/* release reference on per-ID lock */
if (lustre_handle_is_used(lockh))
ldlm_lock_decref(lockh, qsd_id_einfo.ei_mode);
if (cancel) {
qsd_adjust_schedule(lqe, false, true);
} else if (adjust) {
if (!ret || ret == -EDQUOT)
qsd_adjust_schedule(lqe, false, false);
else
qsd_adjust_schedule(lqe, true, false);
}
lqe_putref(lqe);
if (lvb)
OBD_FREE_PTR(lvb);
EXIT;
}
/**
* Check whether any quota space adjustment (pre-acquire/release/report) is
* needed for a given quota ID. If a non-null \a qbody is passed, then the
* \a qbody structure (qb_count/flags/usage) is filled with appropriate data
* to be packed in the quota request.
*
* \param lqe - is the lquota entry for which we would like to adjust quota
* space.
* \param qbody - is the quota body to fill, if not NULL.
*
* \retval true - space adjustment is required and \a qbody is filled, if not
* NULL
* \retval false - no space adjustment required
*/
static bool qsd_calc_adjust(struct lquota_entry *lqe, struct quota_body *qbody)
{
__u64 usage, granted;
ENTRY;
usage = lqe->lqe_usage;
usage += lqe->lqe_pending_write + lqe->lqe_waiting_write;
granted = lqe->lqe_granted;
if (qbody != NULL)
qbody->qb_flags = 0;
if (!lqe->lqe_enforced) {
/* quota not enforced any more for this ID */
if (granted != 0) {
/* release all quota space unconditionally */
LQUOTA_DEBUG(lqe, "not enforced, releasing all space");
if (qbody != NULL) {
qbody->qb_count = granted;
qbody->qb_flags = QUOTA_DQACQ_FL_REL;
}
RETURN(true);
}
RETURN(false);
}
if (!lustre_handle_is_used(&lqe->lqe_lockh)) {
/* No valid per-ID lock
* When reporting quota (during reintegration or on setquota
* glimpse), we should release granted space if usage is 0.
* Otherwise, if the usage is less than granted, we need to
* acquire the per-ID lock to make sure the unused grant can be
* reclaimed by per-ID lock glimpse. */
if (usage == 0) {
/* no on-disk usage and no outstanding activity, release
* space */
if (granted != 0) {
LQUOTA_DEBUG(lqe, "no usage, releasing all "
"space");
if (qbody != NULL) {
qbody->qb_count = granted;
qbody->qb_flags = QUOTA_DQACQ_FL_REL;
}
RETURN(true);
}
LQUOTA_DEBUG(lqe, "no usage + no granted, nothing to "
"do");
RETURN(false);
}
if (lqe->lqe_usage < lqe->lqe_granted) {
/* holding quota space w/o any lock, enqueue per-ID lock
* again */
LQUOTA_DEBUG(lqe, "(re)acquiring per-ID lock");
if (qbody != NULL) {
qbody->qb_count = 0;
qbody->qb_flags = QUOTA_DQACQ_FL_ACQ;
}
RETURN(true);
}
if (lqe->lqe_usage > lqe->lqe_granted) {
/* quota overrun, report usage */
LQUOTA_DEBUG(lqe, "overrun, reporting usage");
if (qbody != NULL) {
qbody->qb_usage = lqe->lqe_usage;
qbody->qb_flags = QUOTA_DQACQ_FL_REPORT;
}
RETURN(true);
}
LQUOTA_DEBUG(lqe, "granted matches usage, nothing to do");
RETURN(false);
}
/* valid per-ID lock
* Apply good old quota qunit adjustment logic which has been around
* since lustre 1.4:
* 1. release spare quota space? */
if (granted > usage + lqe->lqe_qunit) {
/* pre-release quota space */
if (qbody == NULL)
RETURN(true);
qbody->qb_count = granted - usage;
/* if usage == 0, release all granted space */
if (usage) {
/* try to keep one qunit of quota space */
qbody->qb_count -= lqe->lqe_qunit;
/* but don't release less than qtune to avoid releasing
* space too often */
if (qbody->qb_count < lqe->lqe_qtune)
qbody->qb_count = lqe->lqe_qtune;
}
qbody->qb_flags = QUOTA_DQACQ_FL_REL;
RETURN(true);
}
/* 2. Any quota overrun? */
if (lqe->lqe_usage > lqe->lqe_granted) {
/* we overconsumed quota space, we report usage in request so
* that master can adjust it unconditionally */
if (qbody == NULL)
RETURN(true);
qbody->qb_usage = lqe->lqe_usage;
granted = lqe->lqe_usage;
qbody->qb_flags = QUOTA_DQACQ_FL_REPORT;
}
/* 3. Time to pre-acquire? */
if (!lqe->lqe_edquot && !lqe->lqe_nopreacq && usage > 0 &&
lqe->lqe_qunit != 0 && granted < usage + lqe->lqe_qtune) {
/* To pre-acquire quota space, we report how much spare quota
* space the slave currently owns, then the master will grant us
* back how much we can pretend given the current state of
* affairs */
if (qbody == NULL)
RETURN(true);
if (granted <= usage)
qbody->qb_count = 0;
else
qbody->qb_count = granted - usage;
qbody->qb_flags |= QUOTA_DQACQ_FL_PREACQ;
RETURN(true);
}
if (qbody != NULL)
RETURN(qbody->qb_flags != 0);
else
RETURN(false);
}
/*
* VBR: rename versions in reply: 0 - src parent; 1 - tgt parent;
* 2 - src child; 3 - tgt child.
* Update on disk version of src child.
*/
static int mdt_reint_rename(struct mdt_thread_info *info,
struct mdt_lock_handle *lhc)
{
struct mdt_reint_record *rr = &info->mti_rr;
struct md_attr *ma = &info->mti_attr;
struct ptlrpc_request *req = mdt_info_req(info);
struct mdt_object *msrcdir;
struct mdt_object *mtgtdir;
struct mdt_object *mold;
struct mdt_object *mnew = NULL;
struct mdt_lock_handle *lh_srcdirp;
struct mdt_lock_handle *lh_tgtdirp;
struct mdt_lock_handle *lh_oldp;
struct mdt_lock_handle *lh_newp;
struct lu_fid *old_fid = &info->mti_tmp_fid1;
struct lu_fid *new_fid = &info->mti_tmp_fid2;
struct lustre_handle rename_lh = { 0 };
struct lu_name slname = { 0 };
struct lu_name *lname;
int rc;
ENTRY;
if (info->mti_dlm_req)
ldlm_request_cancel(req, info->mti_dlm_req, 0);
DEBUG_REQ(D_INODE, req, "rename "DFID"/%s to "DFID"/%s",
PFID(rr->rr_fid1), rr->rr_name,
PFID(rr->rr_fid2), rr->rr_tgt);
rc = mdt_rename_lock(info, &rename_lh);
if (rc) {
CERROR("Can't lock FS for rename, rc %d\n", rc);
RETURN(rc);
}
lh_newp = &info->mti_lh[MDT_LH_NEW];
/* step 1: lock the source dir. */
lh_srcdirp = &info->mti_lh[MDT_LH_PARENT];
mdt_lock_pdo_init(lh_srcdirp, LCK_PW, rr->rr_name,
rr->rr_namelen);
msrcdir = mdt_object_find_lock(info, rr->rr_fid1, lh_srcdirp,
MDS_INODELOCK_UPDATE);
if (IS_ERR(msrcdir))
GOTO(out_rename_lock, rc = PTR_ERR(msrcdir));
if (mdt_object_obf(msrcdir))
GOTO(out_unlock_source, rc = -EPERM);
rc = mdt_version_get_check_save(info, msrcdir, 0);
if (rc)
GOTO(out_unlock_source, rc);
/* step 2: find & lock the target dir. */
lh_tgtdirp = &info->mti_lh[MDT_LH_CHILD];
mdt_lock_pdo_init(lh_tgtdirp, LCK_PW, rr->rr_tgt,
rr->rr_tgtlen);
if (lu_fid_eq(rr->rr_fid1, rr->rr_fid2)) {
mdt_object_get(info->mti_env, msrcdir);
mtgtdir = msrcdir;
if (lh_tgtdirp->mlh_pdo_hash != lh_srcdirp->mlh_pdo_hash) {
rc = mdt_pdir_hash_lock(info, lh_tgtdirp, mtgtdir,
MDS_INODELOCK_UPDATE);
if (rc)
GOTO(out_unlock_source, rc);
OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_PDO_LOCK2, 10);
}
} else {
mtgtdir = mdt_object_find(info->mti_env, info->mti_mdt,
rr->rr_fid2);
if (IS_ERR(mtgtdir))
GOTO(out_unlock_source, rc = PTR_ERR(mtgtdir));
if (mdt_object_obf(mtgtdir))
GOTO(out_put_target, rc = -EPERM);
/* check early, the real version will be saved after locking */
rc = mdt_version_get_check(info, mtgtdir, 1);
if (rc)
GOTO(out_put_target, rc);
if (unlikely(mdt_object_remote(mtgtdir))) {
CDEBUG(D_INFO, "Source dir "DFID" target dir "DFID
"on different MDTs\n", PFID(rr->rr_fid1),
PFID(rr->rr_fid2));
GOTO(out_put_target, rc = -EXDEV);
} else {
if (likely(mdt_object_exists(mtgtdir))) {
/* we lock the target dir if it is local */
rc = mdt_object_lock(info, mtgtdir, lh_tgtdirp,
MDS_INODELOCK_UPDATE,
MDT_LOCAL_LOCK);
if (rc != 0)
GOTO(out_put_target, rc);
/* get and save correct version after locking */
mdt_version_get_save(info, mtgtdir, 1);
} else {
GOTO(out_put_target, rc = -ESTALE);
}
}
}
/* step 3: find & lock the old object. */
lname = mdt_name(info->mti_env, (char *)rr->rr_name, rr->rr_namelen);
mdt_name_copy(&slname, lname);
fid_zero(old_fid);
rc = mdt_lookup_version_check(info, msrcdir, &slname, old_fid, 2);
if (rc != 0)
GOTO(out_unlock_target, rc);
if (lu_fid_eq(old_fid, rr->rr_fid1) || lu_fid_eq(old_fid, rr->rr_fid2))
GOTO(out_unlock_target, rc = -EINVAL);
mold = mdt_object_find(info->mti_env, info->mti_mdt, old_fid);
if (IS_ERR(mold))
GOTO(out_unlock_target, rc = PTR_ERR(mold));
if (mdt_object_remote(mold)) {
mdt_object_put(info->mti_env, mold);
CDEBUG(D_INFO, "Source child "DFID" is on another MDT\n",
PFID(old_fid));
GOTO(out_unlock_target, rc = -EXDEV);
}
if (mdt_object_obf(mold)) {
mdt_object_put(info->mti_env, mold);
GOTO(out_unlock_target, rc = -EPERM);
}
lh_oldp = &info->mti_lh[MDT_LH_OLD];
mdt_lock_reg_init(lh_oldp, LCK_EX);
rc = mdt_object_lock(info, mold, lh_oldp, MDS_INODELOCK_LOOKUP,
MDT_CROSS_LOCK);
if (rc != 0) {
mdt_object_put(info->mti_env, mold);
GOTO(out_unlock_target, rc);
}
info->mti_mos = mold;
/* save version after locking */
mdt_version_get_save(info, mold, 2);
mdt_set_capainfo(info, 2, old_fid, BYPASS_CAPA);
/* step 4: find & lock the new object. */
/* new target object may not exist now */
lname = mdt_name(info->mti_env, (char *)rr->rr_tgt, rr->rr_tgtlen);
/* lookup with version checking */
fid_zero(new_fid);
rc = mdt_lookup_version_check(info, mtgtdir, lname, new_fid, 3);
if (rc == 0) {
/* the new_fid should have been filled at this moment */
if (lu_fid_eq(old_fid, new_fid))
GOTO(out_unlock_old, rc);
if (lu_fid_eq(new_fid, rr->rr_fid1) ||
lu_fid_eq(new_fid, rr->rr_fid2))
GOTO(out_unlock_old, rc = -EINVAL);
mdt_lock_reg_init(lh_newp, LCK_EX);
mnew = mdt_object_find(info->mti_env, info->mti_mdt, new_fid);
if (IS_ERR(mnew))
GOTO(out_unlock_old, rc = PTR_ERR(mnew));
if (mdt_object_obf(mnew)) {
mdt_object_put(info->mti_env, mnew);
GOTO(out_unlock_old, rc = -EPERM);
}
if (mdt_object_remote(mnew)) {
mdt_object_put(info->mti_env, mnew);
CDEBUG(D_INFO, "src child "DFID" is on another MDT\n",
PFID(new_fid));
GOTO(out_unlock_old, rc = -EXDEV);
}
rc = mdt_object_lock(info, mnew, lh_newp,
MDS_INODELOCK_FULL, MDT_CROSS_LOCK);
if (rc != 0) {
mdt_object_put(info->mti_env, mnew);
GOTO(out_unlock_old, rc);
}
/* get and save version after locking */
mdt_version_get_save(info, mnew, 3);
mdt_set_capainfo(info, 3, new_fid, BYPASS_CAPA);
} else if (rc != -EREMOTE && rc != -ENOENT) {
GOTO(out_unlock_old, rc);
} else {
mdt_enoent_version_save(info, 3);
}
/* step 5: rename it */
mdt_reint_init_ma(info, ma);
mdt_fail_write(info->mti_env, info->mti_mdt->mdt_bottom,
OBD_FAIL_MDS_REINT_RENAME_WRITE);
/* Check if @dst is subdir of @src. */
rc = mdt_rename_sanity(info, old_fid);
if (rc)
GOTO(out_unlock_new, rc);
rc = mdo_rename(info->mti_env, mdt_object_child(msrcdir),
mdt_object_child(mtgtdir), old_fid, &slname,
(mnew ? mdt_object_child(mnew) : NULL),
lname, ma);
/* handle last link of tgt object */
if (rc == 0) {
mdt_counter_incr(req, LPROC_MDT_RENAME);
if (mnew)
mdt_handle_last_unlink(info, mnew, ma);
mdt_rename_counter_tally(info, info->mti_mdt, req,
msrcdir, mtgtdir);
}
EXIT;
out_unlock_new:
if (mnew)
mdt_object_unlock_put(info, mnew, lh_newp, rc);
out_unlock_old:
mdt_object_unlock_put(info, mold, lh_oldp, rc);
out_unlock_target:
mdt_object_unlock(info, mtgtdir, lh_tgtdirp, rc);
out_put_target:
mdt_object_put(info->mti_env, mtgtdir);
out_unlock_source:
mdt_object_unlock_put(info, msrcdir, lh_srcdirp, rc);
out_rename_lock:
if (lustre_handle_is_used(&rename_lh))
mdt_rename_unlock(&rename_lh);
return rc;
}
