int osc_object_is_contended(struct osc_object *obj)
{
struct osc_device *dev = lu2osc_dev(obj->oo_cl.co_lu.lo_dev);
int osc_contention_time = dev->od_contention_time;
unsigned long cur_time = cfs_time_current();
unsigned long retry_time;
if (OBD_FAIL_CHECK(OBD_FAIL_OSC_OBJECT_CONTENTION))
return 1;
if (!obj->oo_contended)
return 0;
/*
* I like copy-paste. the code is copied from
* ll_file_is_contended.
*/
retry_time = cfs_time_add(obj->oo_contention_time,
cfs_time_seconds(osc_contention_time));
if (cfs_time_after(cur_time, retry_time)) {
osc_object_clear_contended(obj);
return 0;
}
return 1;
}
/**
* Callback handler for receiving incoming glimpse ASTs.
*
* This only can happen on client side. After handling the glimpse AST
* we also consider dropping the lock here if it is unused locally for a
* long time.
*/
static void ldlm_handle_gl_callback(struct ptlrpc_request *req,
struct ldlm_namespace *ns,
struct ldlm_request *dlm_req,
struct ldlm_lock *lock)
{
int rc = -ENOSYS;
LDLM_DEBUG(lock, "client glimpse AST callback handler");
if (lock->l_glimpse_ast != NULL)
rc = lock->l_glimpse_ast(lock, req);
if (req->rq_repmsg != NULL) {
ptlrpc_reply(req);
} else {
req->rq_status = rc;
ptlrpc_error(req);
}
lock_res_and_lock(lock);
if (lock->l_granted_mode == LCK_PW &&
!lock->l_readers && !lock->l_writers &&
cfs_time_after(cfs_time_current(),
cfs_time_add(lock->l_last_used,
cfs_time_seconds(10)))) {
unlock_res_and_lock(lock);
if (ldlm_bl_to_thread_lock(ns, NULL, lock))
ldlm_handle_bl_callback(ns, NULL, lock);
return;
}
unlock_res_and_lock(lock);
LDLM_LOCK_RELEASE(lock);
}
/* Read from wire as much data as possible.
* Returns 0 or 1 on succsess, <0 if error or EOF.
* 0 means partial read, 1 - complete */
int
usocklnd_read_data(usock_conn_t *conn)
{
struct iovec *iov;
int nob;
cfs_time_t t;
LASSERT (conn->uc_rx_nob_wanted != 0);
do {
usock_peer_t *peer = conn->uc_peer;
LASSERT (conn->uc_rx_niov > 0);
nob = libcfs_sock_readv(conn->uc_sock,
conn->uc_rx_iov, conn->uc_rx_niov);
if (nob <= 0) {/* read nothing or error */
if (nob < 0)
conn->uc_errored = 1;
return nob;
}
LASSERT (nob <= conn->uc_rx_nob_wanted);
conn->uc_rx_nob_wanted -= nob;
conn->uc_rx_nob_left -= nob;
t = cfs_time_current();
conn->uc_rx_deadline = cfs_time_add(t, cfs_time_seconds(usock_tuns.ut_timeout));
if(peer != NULL)
peer->up_last_alive = t;
/* "consume" iov */
iov = conn->uc_rx_iov;
do {
LASSERT (conn->uc_rx_niov > 0);
if (nob < iov->iov_len) {
iov->iov_base = (void *)(((unsigned long)(iov->iov_base)) + nob);
iov->iov_len -= nob;
break;
}
nob -= iov->iov_len;
conn->uc_rx_iov = ++iov;
conn->uc_rx_niov--;
} while (nob != 0);
} while (conn->uc_rx_nob_wanted != 0);
return 1; /* read complete */
}
/* Send as much tx data as possible.
* Returns 0 or 1 on succsess, <0 if fatal error.
* 0 means partial send or non-fatal error, 1 - complete.
* Rely on libcfs_sock_writev() for differentiating fatal and
* non-fatal errors. An error should be considered as non-fatal if:
* 1) it still makes sense to continue reading &&
* 2) anyway, poll() will set up POLLHUP|POLLERR flags */
int
usocklnd_send_tx(usock_conn_t *conn, usock_tx_t *tx)
{
struct iovec *iov;
int nob;
cfs_time_t t;
LASSERT (tx->tx_resid != 0);
do {
usock_peer_t *peer = conn->uc_peer;
LASSERT (tx->tx_niov > 0);
nob = libcfs_sock_writev(conn->uc_sock,
tx->tx_iov, tx->tx_niov);
if (nob < 0)
conn->uc_errored = 1;
if (nob <= 0) /* write queue is flow-controlled or error */
return nob;
LASSERT (nob <= tx->tx_resid);
tx->tx_resid -= nob;
t = cfs_time_current();
conn->uc_tx_deadline = cfs_time_add(t, cfs_time_seconds(usock_tuns.ut_timeout));
if(peer != NULL)
peer->up_last_alive = t;
/* "consume" iov */
iov = tx->tx_iov;
do {
LASSERT (tx->tx_niov > 0);
if (nob < iov->iov_len) {
iov->iov_base = (void *)(((unsigned long)(iov->iov_base)) + nob);
iov->iov_len -= nob;
break;
}
nob -= iov->iov_len;
tx->tx_iov = ++iov;
tx->tx_niov--;
} while (nob != 0);
} while (tx->tx_resid != 0);
return 1; /* send complete */
}
int quota_adjust_slave_lqs(struct quota_adjust_qunit *oqaq,
struct lustre_quota_ctxt *qctxt)
{
struct lustre_qunit_size *lqs = NULL;
unsigned long *unit, *tune;
signed long tmp = 0;
cfs_time_t time_limit = 0, *shrink;
int i, rc = 0;
ENTRY;
LASSERT(qctxt);
lqs = quota_search_lqs(LQS_KEY(QAQ_IS_GRP(oqaq), oqaq->qaq_id),
qctxt, QAQ_IS_CREATE_LQS(oqaq) ? 1 : 0);
if (lqs == NULL || IS_ERR(lqs)){
CERROR("fail to find a lqs for %sid %u!\n",
QAQ_IS_GRP(oqaq) ? "g" : "u", oqaq->qaq_id);
RETURN(PTR_ERR(lqs));
}
CDEBUG(D_QUOTA, "before: bunit: %lu, iunit: %lu.\n",
lqs->lqs_bunit_sz, lqs->lqs_iunit_sz);
cfs_spin_lock(&lqs->lqs_lock);
for (i = 0; i < 2; i++) {
if (i == 0 && !QAQ_IS_ADJBLK(oqaq))
continue;
if (i == 1 && !QAQ_IS_ADJINO(oqaq))
continue;
tmp = i ? (lqs->lqs_iunit_sz - oqaq->qaq_iunit_sz) :
(lqs->lqs_bunit_sz - oqaq->qaq_bunit_sz);
shrink = i ? &lqs->lqs_last_ishrink :
&lqs->lqs_last_bshrink;
time_limit = cfs_time_add(i ? lqs->lqs_last_ishrink :
lqs->lqs_last_bshrink,
cfs_time_seconds(qctxt->lqc_switch_seconds));
unit = i ? &lqs->lqs_iunit_sz : &lqs->lqs_bunit_sz;
tune = i ? &lqs->lqs_itune_sz : &lqs->lqs_btune_sz;
/* quota master shrinks */
if (qctxt->lqc_handler && tmp > 0)
*shrink = cfs_time_current();
/* quota master enlarges */
if (qctxt->lqc_handler && tmp < 0) {
/* in case of ping-pong effect, don't enlarge lqs
* in a short time */
if (*shrink &&
cfs_time_before(cfs_time_current(), time_limit))
tmp = 0;
}
/* when setquota, don't enlarge lqs b=18616 */
if (QAQ_IS_CREATE_LQS(oqaq) && tmp < 0)
tmp = 0;
if (tmp != 0) {
*unit = i ? oqaq->qaq_iunit_sz : oqaq->qaq_bunit_sz;
*tune = (*unit) / 2;
}
if (tmp > 0)
rc |= i ? LQS_INO_DECREASE : LQS_BLK_DECREASE;
if (tmp < 0)
rc |= i ? LQS_INO_INCREASE : LQS_BLK_INCREASE;
}
cfs_spin_unlock(&lqs->lqs_lock);
CDEBUG(D_QUOTA, "after: bunit: %lu, iunit: %lu.\n",
lqs->lqs_bunit_sz, lqs->lqs_iunit_sz);
lqs_putref(lqs);
RETURN(rc);
}
/* Send as much tx data as possible.
* Returns 0 or 1 on succsess, <0 if fatal error.
* 0 means partial send or non-fatal error, 1 - complete.
* Rely on libcfs_sock_writev() for differentiating fatal and
* non-fatal errors. An error should be considered as non-fatal if:
* 1) it still makes sense to continue reading &&
* 2) anyway, poll() will set up POLLHUP|POLLERR flags */
int
usocklnd_send_tx(usock_conn_t *conn, usock_tx_t *tx)
{
struct iovec *iov;
int nob;
struct lnet_xport *lx = conn->uc_lx;
struct pfl_opstat *opst;
cfs_time_t t;
LASSERT (tx->tx_resid != 0);
do {
usock_peer_t *peer = conn->uc_peer;
LASSERT (tx->tx_niov > 0);
nob = lx_writev(lx, tx->tx_iov, tx->tx_niov);
if (nob < 0)
conn->uc_errored = 1;
if (nob <= 0) /* write queue is flow-controlled or error */
return nob;
if (peer && peer->up_ni)
opst = peer->up_ni->ni_iostats.wr;
else if (conn->uc_ni)
opst = conn->uc_ni->ni_iostats.wr;
else
opst = usock_pasv_iostats.wr;
pfl_opstat_add(opst, nob);
if (peer)
pfl_opstat_add(peer->up_iostats.wr, nob);
pfl_opstat_add(usock_aggr_iostats.wr, nob);
LASSERT (nob <= tx->tx_resid);
tx->tx_resid -= nob;
t = cfs_time_current();
conn->uc_tx_deadline = cfs_time_add(t, cfs_time_seconds(usock_tuns.ut_timeout));
if(peer != NULL)
peer->up_last_alive = t;
/* "consume" iov */
iov = tx->tx_iov;
do {
LASSERT (tx->tx_niov > 0);
if ((size_t)nob < iov->iov_len) {
iov->iov_base = (void *)(((unsigned long)(iov->iov_base)) + nob);
iov->iov_len -= nob;
break;
}
nob -= iov->iov_len;
tx->tx_iov = ++iov;
tx->tx_niov--;
} while (nob != 0);
} while (tx->tx_resid != 0);
return 1; /* send complete */
}
/* Read from wire as much data as possible.
* Returns 0 or 1 on succsess, <0 if error or EOF.
* 0 means partial read, 1 - complete */
int
usocklnd_read_data(usock_conn_t *conn)
{
struct pfl_opstat *opst;
struct iovec *iov;
int nob;
cfs_time_t t;
LASSERT (conn->uc_rx_nob_wanted != 0);
do {
usock_peer_t *peer = conn->uc_peer;
LASSERT (conn->uc_rx_niov > 0);
nob = lx_readv(conn->uc_lx, conn->uc_rx_iov, conn->uc_rx_niov);
if (nob <= 0) {/* read nothing or error */
if (nob < 0)
conn->uc_errored = 1;
return nob;
}
if (peer && peer->up_ni)
opst = peer->up_ni->ni_iostats.rd;
else if (conn->uc_ni)
opst = conn->uc_ni->ni_iostats.rd;
else
opst = usock_pasv_iostats.rd;
pfl_opstat_add(opst, nob);
if (peer)
pfl_opstat_add(peer->up_iostats.rd, nob);
pfl_opstat_add(usock_aggr_iostats.rd, nob);
LASSERT (nob <= conn->uc_rx_nob_wanted);
conn->uc_rx_nob_wanted -= nob;
conn->uc_rx_nob_left -= nob;
t = cfs_time_current();
conn->uc_rx_deadline = cfs_time_add(t, cfs_time_seconds(usock_tuns.ut_timeout));
if(peer != NULL)
peer->up_last_alive = t;
/* "consume" iov */
iov = conn->uc_rx_iov;
do {
LASSERT (conn->uc_rx_niov > 0);
if ((size_t)nob < iov->iov_len) {
iov->iov_base = (void *)(((unsigned long)(iov->iov_base)) + nob);
iov->iov_len -= nob;
break;
}
nob -= iov->iov_len;
conn->uc_rx_iov = ++iov;
conn->uc_rx_niov--;
} while (nob != 0);
} while (conn->uc_rx_nob_wanted != 0);
return 1; /* read complete */
}