int cfs_timer_is_armed(cfs_timer_t *l)
{
if (cfs_time_before(cfs_time_current(), l->expires))
return 1;
else
return 0;
}
static inline void update_capa_timer(struct obd_capa *ocapa, cfs_time_t expiry)
{
if (cfs_time_before(expiry, ll_capa_timer.expires) ||
!timer_pending(&ll_capa_timer)) {
mod_timer(&ll_capa_timer, expiry);
DEBUG_CAPA(D_SEC, &ocapa->c_capa,
"ll_capa_timer update: %lu/%lu by", expiry, jiffies);
}
}
/*
* statfs
*/
static inline int osp_statfs_need_update(struct osp_device *d)
{
return !cfs_time_before(cfs_time_current(),
d->opd_statfs_fresh_till);
}
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);
}
static int quota_chk_acq_common(struct obd_device *obd, struct obd_export *exp,
const unsigned int id[], int pending[],
int count, quota_acquire acquire,
struct obd_trans_info *oti, int isblk,
struct inode *inode, int frags)
{
struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
struct timeval work_start;
struct timeval work_end;
long timediff;
struct l_wait_info lwi = { 0 };
int rc = 0, cycle = 0, count_err = 1;
ENTRY;
if (!quota_is_set(obd, id, isblk ? QB_SET : QI_SET))
RETURN(0);
if (isblk && (exp->exp_failed || exp->exp_abort_active_req))
/* If the client has been evicted or if it
* timed out and tried to reconnect already,
* abort the request immediately */
RETURN(-ENOTCONN);
CDEBUG(D_QUOTA, "check quota for %s\n", obd->obd_name);
pending[USRQUOTA] = pending[GRPQUOTA] = 0;
/* Unfortunately, if quota master is too busy to handle the
* pre-dqacq in time and quota hash on ost is used up, we
* have to wait for the completion of in flight dqacq/dqrel,
* in order to get enough quota for write b=12588 */
cfs_gettimeofday(&work_start);
while ((rc = quota_check_common(obd, id, pending, count, cycle, isblk,
inode, frags)) &
QUOTA_RET_ACQUOTA) {
cfs_spin_lock(&qctxt->lqc_lock);
if (!qctxt->lqc_import && oti) {
cfs_spin_unlock(&qctxt->lqc_lock);
LASSERT(oti && oti->oti_thread &&
oti->oti_thread->t_watchdog);
lc_watchdog_disable(oti->oti_thread->t_watchdog);
CDEBUG(D_QUOTA, "sleep for quota master\n");
l_wait_event(qctxt->lqc_wait_for_qmaster, check_qm(qctxt),
&lwi);
CDEBUG(D_QUOTA, "wake up when quota master is back\n");
lc_watchdog_touch(oti->oti_thread->t_watchdog,
CFS_GET_TIMEOUT(oti->oti_thread->t_svc));
} else {
cfs_spin_unlock(&qctxt->lqc_lock);
}
cycle++;
if (isblk)
OBD_FAIL_TIMEOUT(OBD_FAIL_OST_HOLD_WRITE_RPC, 90);
/* after acquire(), we should run quota_check_common again
* so that we confirm there are enough quota to finish write */
rc = acquire(obd, id, oti, isblk);
/* please reference to dqacq_completion for the below */
/* a new request is finished, try again */
if (rc == QUOTA_REQ_RETURNED) {
CDEBUG(D_QUOTA, "finish a quota req, try again\n");
continue;
}
/* it is out of quota already */
if (rc == -EDQUOT) {
CDEBUG(D_QUOTA, "out of quota, return -EDQUOT\n");
break;
}
/* Related quota has been disabled by master, but enabled by
* slave, do not try again. */
if (unlikely(rc == -ESRCH)) {
CERROR("mismatched quota configuration, stop try.\n");
break;
}
if (isblk && (exp->exp_failed || exp->exp_abort_active_req))
/* The client has been evicted or tried to
* to reconnect already, abort the request */
RETURN(-ENOTCONN);
/* -EBUSY and others, wait a second and try again */
if (rc < 0) {
cfs_waitq_t waitq;
struct l_wait_info lwi;
if (oti && oti->oti_thread && oti->oti_thread->t_watchdog)
lc_watchdog_touch(oti->oti_thread->t_watchdog,
CFS_GET_TIMEOUT(oti->oti_thread->t_svc));
CDEBUG(D_QUOTA, "rc: %d, count_err: %d\n", rc,
count_err++);
cfs_waitq_init(&waitq);
lwi = LWI_TIMEOUT(cfs_time_seconds(min(cycle, 10)), NULL,
NULL);
l_wait_event(waitq, 0, &lwi);
}
if (rc < 0 || cycle % 10 == 0) {
cfs_spin_lock(&last_print_lock);
if (last_print == 0 ||
cfs_time_before((last_print + cfs_time_seconds(30)),
cfs_time_current())) {
last_print = cfs_time_current();
cfs_spin_unlock(&last_print_lock);
CWARN("still haven't managed to acquire quota "
"space from the quota master after %d "
"retries (err=%d, rc=%d)\n",
cycle, count_err - 1, rc);
} else {
cfs_spin_unlock(&last_print_lock);
}
}
CDEBUG(D_QUOTA, "recheck quota with rc: %d, cycle: %d\n", rc,
cycle);
}
cfs_gettimeofday(&work_end);
timediff = cfs_timeval_sub(&work_end, &work_start, NULL);
lprocfs_counter_add(qctxt->lqc_stats,
isblk ? LQUOTA_WAIT_FOR_CHK_BLK :
LQUOTA_WAIT_FOR_CHK_INO,
timediff);
if (rc > 0)
rc = 0;
RETURN(rc);
}
static int mdt_ck_thread_main(void *args)
{
struct mdt_device *mdt = args;
struct ptlrpc_thread *thread = &mdt->mdt_ck_thread;
struct lustre_capa_key *bkey = &mdt->mdt_capa_keys[0],
*rkey = &mdt->mdt_capa_keys[1];
struct lustre_capa_key *tmp;
struct lu_env env;
struct mdt_thread_info *info;
struct md_device *next;
struct l_wait_info lwi = { 0 };
mdsno_t mdsnum;
int rc;
ENTRY;
unshare_fs_struct();
cfs_block_allsigs();
thread_set_flags(thread, SVC_RUNNING);
wake_up(&thread->t_ctl_waitq);
rc = lu_env_init(&env, LCT_MD_THREAD|LCT_REMEMBER|LCT_NOREF);
if (rc)
RETURN(rc);
thread->t_env = &env;
env.le_ctx.lc_thread = thread;
env.le_ctx.lc_cookie = 0x1;
info = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
LASSERT(info != NULL);
tmp = &info->mti_capa_key;
mdsnum = mdt_seq_site(mdt)->ss_node_id;
while (1) {
l_wait_event(thread->t_ctl_waitq,
thread_is_stopping(thread) ||
thread_is_event(thread),
&lwi);
if (thread_is_stopping(thread))
break;
thread_clear_flags(thread, SVC_EVENT);
if (cfs_time_before(cfs_time_current(), mdt->mdt_ck_expiry))
break;
*tmp = *rkey;
make_capa_key(tmp, mdsnum, rkey->lk_keyid);
next = mdt->mdt_child;
rc = next->md_ops->mdo_update_capa_key(&env, next, tmp);
if (!rc) {
spin_lock(&capa_lock);
*bkey = *rkey;
*rkey = *tmp;
spin_unlock(&capa_lock);
rc = write_capa_keys(&env, mdt, mdt->mdt_capa_keys);
if (rc) {
spin_lock(&capa_lock);
*rkey = *bkey;
memset(bkey, 0, sizeof(*bkey));
spin_unlock(&capa_lock);
} else {
set_capa_key_expiry(mdt);
DEBUG_CAPA_KEY(D_SEC, rkey, "new");
}
}
if (rc) {
DEBUG_CAPA_KEY(D_ERROR, rkey, "update failed for");
/* next retry is in 300 sec */
mdt->mdt_ck_expiry = jiffies + 300 * HZ;
}
cfs_timer_arm(&mdt->mdt_ck_timer, mdt->mdt_ck_expiry);
CDEBUG(D_SEC, "mdt_ck_timer %lu\n", mdt->mdt_ck_expiry);
}
lu_env_fini(&env);
thread_set_flags(thread, SVC_STOPPED);
wake_up(&thread->t_ctl_waitq);
RETURN(0);
}
