#include "ptlrpc_internal.h"
struct semaphore pinger_sem;
-static struct list_head pinger_imports = CFS_LIST_HEAD_INIT(pinger_imports);
+static CFS_LIST_HEAD(pinger_imports);
int ptlrpc_ping(struct obd_import *imp)
{
struct ptlrpc_request *req;
- int rc = 0;
ENTRY;
- req = ptlrpc_prep_req(imp, LUSTRE_OBD_VERSION, OBD_PING, 1, NULL, NULL);
- if (req) {
- DEBUG_REQ(D_INFO, req, "pinging %s->%s",
- imp->imp_obd->obd_uuid.uuid,
- obd2cli_tgt(imp->imp_obd));
- req->rq_no_resend = req->rq_no_delay = 1;
- ptlrpc_req_set_repsize(req, 1, NULL);
- ptlrpcd_add_req(req);
- } else {
+ req = ptlrpc_request_alloc_pack(imp, &RQF_OBD_PING, LUSTRE_OBD_VERSION,
+ OBD_PING);
+ if (req == NULL) {
CERROR("OOM trying to ping %s->%s\n",
imp->imp_obd->obd_uuid.uuid,
obd2cli_tgt(imp->imp_obd));
- rc = -ENOMEM;
+ RETURN(-ENOMEM);
}
- RETURN(rc);
+ DEBUG_REQ(D_INFO, req, "pinging %s->%s",
+ imp->imp_obd->obd_uuid.uuid, obd2cli_tgt(imp->imp_obd));
+ req->rq_no_resend = req->rq_no_delay = 1;
+ ptlrpc_request_set_replen(req);
+ req->rq_timeout = PING_INTERVAL;
+ ptlrpcd_add_req(req);
+
+ RETURN(0);
}
-static void ptlrpc_update_next_ping(struct obd_import *imp)
+void ptlrpc_update_next_ping(struct obd_import *imp)
{
+#ifdef ENABLE_PINGER
imp->imp_next_ping = cfs_time_shift(
(imp->imp_state == LUSTRE_IMP_DISCON ?
RECONNECT_INTERVAL : PING_INTERVAL));
+#endif /* ENABLE_PINGER */
}
void ptlrpc_ping_import_soon(struct obd_import *imp)
imp->imp_next_ping = cfs_time_current();
}
+static inline int ptlrpc_next_reconnect(struct obd_import *imp)
+{
+ if (imp->imp_server_timeout)
+ return cfs_time_shift(obd_timeout / 2);
+ else
+ return cfs_time_shift(obd_timeout);
+}
+
+static atomic_t suspend_timeouts = ATOMIC_INIT(0);
+static cfs_time_t suspend_wakeup_time = 0;
+
+#ifdef __KERNEL__
+static wait_queue_head_t suspend_timeouts_waitq;
+#endif
+
+cfs_time_t ptlrpc_suspend_wakeup_time(void)
+{
+ return suspend_wakeup_time;
+}
+
+void ptlrpc_deactivate_timeouts(struct obd_import *imp)
+{
+ /*XXX: disabled for now, will be replaced by adaptive timeouts */
+#if 0
+ if (imp->imp_no_timeout)
+ return;
+ imp->imp_no_timeout = 1;
+ atomic_inc(&suspend_timeouts);
+ CDEBUG(D_HA|D_WARNING, "deactivate timeouts %u\n", atomic_read(&suspend_timeouts));
+#endif
+}
+
+void ptlrpc_activate_timeouts(struct obd_import *imp)
+{
+ /*XXX: disabled for now, will be replaced by adaptive timeouts */
+#if 0
+ if (!imp->imp_no_timeout)
+ return;
+ imp->imp_no_timeout = 0;
+ LASSERT(atomic_read(&suspend_timeouts) > 0);
+ if (atomic_dec_and_test(&suspend_timeouts)) {
+ suspend_wakeup_time = cfs_time_current();
+ wake_up(&suspend_timeouts_waitq);
+ }
+ CDEBUG(D_HA|D_WARNING, "activate timeouts %u\n", atomic_read(&suspend_timeouts));
+#endif
+}
+
+int ptlrpc_check_suspend(void)
+{
+ if (atomic_read(&suspend_timeouts))
+ return 1;
+ return 0;
+}
+
+int ptlrpc_check_and_wait_suspend(struct ptlrpc_request *req)
+{
+ struct l_wait_info lwi;
+
+ if (atomic_read(&suspend_timeouts)) {
+ DEBUG_REQ(D_NET, req, "-- suspend %d regular timeout",
+ atomic_read(&suspend_timeouts));
+ lwi = LWI_INTR(NULL, NULL);
+ l_wait_event(suspend_timeouts_waitq,
+ atomic_read(&suspend_timeouts) == 0, &lwi);
+ DEBUG_REQ(D_NET, req, "-- recharge regular timeout");
+ return 1;
+ }
+ return 0;
+}
+
#ifdef __KERNEL__
+
+static void ptlrpc_pinger_process_import(struct obd_import *imp,
+ unsigned long this_ping)
+{
+ int force, level;
+
+ spin_lock(&imp->imp_lock);
+ level = imp->imp_state;
+ force = imp->imp_force_verify;
+ if (force)
+ imp->imp_force_verify = 0;
+ spin_unlock(&imp->imp_lock);
+
+ CDEBUG(level == LUSTRE_IMP_FULL ? D_INFO : D_HA,
+ "level %s/%u force %u deactive %u pingable %u\n",
+ ptlrpc_import_state_name(level), level,
+ force, imp->imp_deactive, imp->imp_pingable);
+
+ if (cfs_time_aftereq(imp->imp_next_ping - 5 * CFS_TICK,
+ this_ping) && force == 0)
+ return;
+
+ if (level == LUSTRE_IMP_DISCON && !imp->imp_deactive) {
+ /* wait at least a timeout before trying recovery again */
+ imp->imp_next_ping = ptlrpc_next_reconnect(imp);
+ ptlrpc_initiate_recovery(imp);
+ } else if (level != LUSTRE_IMP_FULL ||
+ imp->imp_obd->obd_no_recov ||
+ imp->imp_deactive) {
+ CDEBUG(D_HA, "not pinging %s (in recovery "
+ " or recovery disabled: %s)\n",
+ obd2cli_tgt(imp->imp_obd),
+ ptlrpc_import_state_name(level));
+ } else if (imp->imp_pingable || force) {
+ ptlrpc_ping(imp);
+ }
+}
+
static int ptlrpc_pinger_main(void *arg)
{
struct ptlrpc_svc_data *data = (struct ptlrpc_svc_data *)arg;
struct obd_import *imp =
list_entry(iter, struct obd_import,
imp_pinger_chain);
- int force, level;
-
- spin_lock(&imp->imp_lock);
- level = imp->imp_state;
- force = imp->imp_force_verify;
- imp->imp_force_verify = 0;
- spin_unlock(&imp->imp_lock);
-
- CDEBUG(level == LUSTRE_IMP_FULL ? D_INFO : D_HA,
- "level %s/%u force %u deactive %u pingable %u\n",
- ptlrpc_import_state_name(level), level,
- force, imp->imp_deactive, imp->imp_pingable);
-
- if (force ||
- /* if the next ping is within, say, 5 jiffies from
- now, go ahead and ping. See note below. */
- cfs_time_aftereq(this_ping,
- imp->imp_next_ping - 5 * CFS_TICK)) {
- if (level == LUSTRE_IMP_DISCON &&
- !imp->imp_deactive) {
- /* wait at least a timeout before
- trying recovery again. */
- imp->imp_next_ping = cfs_time_shift(obd_timeout);
- ptlrpc_initiate_recovery(imp);
- } else if (level != LUSTRE_IMP_FULL ||
- imp->imp_obd->obd_no_recov ||
- imp->imp_deactive) {
- CDEBUG(D_HA, "not pinging %s "
- "(in recovery: %s or recovery "
- "disabled: %u/%u)\n",
- obd2cli_tgt(imp->imp_obd),
- ptlrpc_import_state_name(level),
- imp->imp_deactive,
- imp->imp_obd->obd_no_recov);
- } else if (imp->imp_pingable || force) {
- ptlrpc_ping(imp);
- }
- } else {
- if (!imp->imp_pingable)
- continue;
- CDEBUG(D_INFO,
- "don't need to ping %s ("CFS_TIME_T
- " > "CFS_TIME_T")\n",
- obd2cli_tgt(imp->imp_obd),
- imp->imp_next_ping, this_ping);
- }
+ ptlrpc_pinger_process_import(imp, this_ping);
/* obd_timeout might have changed */
- if (cfs_time_after(imp->imp_next_ping,
+ if (imp->imp_pingable && imp->imp_next_ping &&
+ cfs_time_after(imp->imp_next_ping,
cfs_time_add(this_ping,
cfs_time_seconds(PING_INTERVAL))))
ptlrpc_update_next_ping(imp);
}
mutex_up(&pinger_sem);
+ /* update memory usage info */
+ obd_update_maxusage();
/* Wait until the next ping time, or until we're stopped. */
time_to_next_ping = cfs_time_sub(cfs_time_add(this_ping,
- cfs_time_seconds(PING_INTERVAL)),
+ cfs_time_seconds(PING_INTERVAL)),
cfs_time_current());
/* The ping sent by ptlrpc_send_rpc may get sent out
if (pinger_thread != NULL)
RETURN(-EALREADY);
- OBD_ALLOC(pinger_thread, sizeof(*pinger_thread));
+ OBD_ALLOC_PTR(pinger_thread);
if (pinger_thread == NULL)
RETURN(-ENOMEM);
cfs_waitq_init(&pinger_thread->t_ctl_waitq);
-
+ cfs_waitq_init(&suspend_timeouts_waitq);
+
d.name = "ll_ping";
d.thread = pinger_thread;
l_wait_event(pinger_thread->t_ctl_waitq,
(pinger_thread->t_flags & SVC_STOPPED), &lwi);
- OBD_FREE(pinger_thread, sizeof(*pinger_thread));
+ OBD_FREE_PTR(pinger_thread);
pinger_thread = NULL;
RETURN(rc);
}
mutex_down(&pinger_sem);
CDEBUG(D_HA, "adding pingable import %s->%s\n",
imp->imp_obd->obd_uuid.uuid, obd2cli_tgt(imp->imp_obd));
+ /* if we add to pinger we want recovery on this import */
+ imp->imp_obd->obd_no_recov = 0;
ptlrpc_update_next_ping(imp);
/* XXX sort, blah blah */
list_add_tail(&imp->imp_pinger_chain, &pinger_imports);
list_del_init(&imp->imp_pinger_chain);
CDEBUG(D_HA, "removing pingable import %s->%s\n",
imp->imp_obd->obd_uuid.uuid, obd2cli_tgt(imp->imp_obd));
+ /* if we remove from pinger we don't want recovery on this import */
+ imp->imp_obd->obd_no_recov = 1;
class_import_put(imp);
mutex_up(&pinger_sem);
RETURN(0);
int ping_evictor_wake(struct obd_export *exp)
{
spin_lock(&pet_lock);
- if (pet_exp) {
+ if (pet_exp || (pet_state != PET_READY)) {
/* eventually the new obd will call here again. */
spin_unlock(&pet_lock);
return 1;
obd = pet_exp->exp_obd;
spin_unlock(&pet_lock);
- expire_time = CURRENT_SECONDS - (3 * obd_timeout / 2);
+ expire_time = cfs_time_current_sec() - (3 * obd_timeout / 2);
CDEBUG(D_HA, "evicting all exports of obd %s older than %ld\n",
obd->obd_name, expire_time);
if (expire_time > exp->exp_last_request_time) {
class_export_get(exp);
spin_unlock(&obd->obd_dev_lock);
- LCONSOLE_WARN("%s: haven't heard from client %s"
+ LCONSOLE_WARN("%s: haven't heard from client %s"
" (at %s) in %ld seconds. I think"
" it's dead, and I am evicting"
- " it.\n", obd->obd_name,
+ " it. exp %p, cur %ld expire %ld"
+ " last %ld\n",
+ obd->obd_name,
obd_uuid2str(&exp->exp_client_uuid),
obd_export_nid2str(exp),
- (long)(CURRENT_SECONDS -
- exp->exp_last_request_time));
+ (long)(cfs_time_current_sec() -
+ exp->exp_last_request_time),
+ exp, (long)cfs_time_current_sec(),
+ (long)expire_time,
+ (long)exp->exp_last_request_time);
CDEBUG(D_HA, "Last request was at %ld\n",
exp->exp_last_request_time);
class_fail_export(exp);
continue;
}
- req = ptlrpc_prep_req(imp, LUSTRE_OBD_VERSION, OBD_PING,
- 1, NULL, NULL);
- if (!req) {
- CERROR("out of memory\n");
+ req = ptlrpc_request_alloc_pack(imp, &RQF_OBD_PING,
+ LUSTRE_OBD_VERSION,
+ OBD_PING);
+ if (req == NULL) {
+ CERROR("OOM trying to ping %s->%s\n",
+ imp->imp_obd->obd_uuid.uuid,
+ obd2cli_tgt(imp->imp_obd));
break;
}
+
req->rq_no_resend = 1;
- ptlrpc_req_set_repsize(req, 1, NULL);
+ ptlrpc_request_set_replen(req);
req->rq_send_state = LUSTRE_IMP_FULL;
req->rq_phase = RQ_PHASE_RPC;
req->rq_import_generation = generation;
ptlrpc_set_add_req(set, req);
} else {
- CDEBUG(D_HA, "don't need to ping %s ("CFS_TIME_T" > "
- CFS_TIME_T")\n", obd2cli_tgt(imp->imp_obd),
+ CDEBUG(D_INFO, "don't need to ping %s ("CFS_TIME_T
+ " > "CFS_TIME_T")\n", obd2cli_tgt(imp->imp_obd),
imp->imp_next_ping, pd->pd_this_ping);
}
}
/* Might be empty, that's OK. */
if (set->set_remaining == 0)
- CDEBUG(D_HA, "nothing to ping\n");
+ CDEBUG(D_RPCTRACE, "nothing to ping\n");
list_for_each(iter, &set->set_requests) {
struct ptlrpc_request *req =
list_entry(iter, struct ptlrpc_request,
rq_set_chain);
- DEBUG_REQ(D_HA, req, "pinging %s->%s",
+ DEBUG_REQ(D_RPCTRACE, req, "pinging %s->%s",
req->rq_import->imp_obd->obd_uuid.uuid,
obd2cli_tgt(req->rq_import->imp_obd));
(void)ptl_send_rpc(req, 0);
rc = ptlrpc_check_set(set);
/* not finished, and we are not expired, simply return */
- if (!rc && cfs_time_before(curtime,
- cfs_time_add(pd->pd_this_ping,
- cfs_time_seconds(PING_INTERVAL)))) {
- CDEBUG(D_HA, "not finished, but also not expired\n");
+ if (!rc && cfs_time_before(curtime, cfs_time_add(pd->pd_this_ping,
+ cfs_time_seconds(PING_INTERVAL)))) {
+ CDEBUG(D_RPCTRACE, "not finished, but also not expired\n");
pd->pd_recursion--;
return 0;
}
req = list_entry(iter, struct ptlrpc_request,
rq_set_chain);
- if (req->rq_replied)
+ if (req->rq_phase == RQ_PHASE_COMPLETE)
continue;
req->rq_phase = RQ_PHASE_COMPLETE;
continue;
}
- CDEBUG(D_HA, "pinger initiate expire_one_request\n");
+ CDEBUG(D_RPCTRACE, "pinger initiate expire_one_request\n");
ptlrpc_expire_one_request(req);
}
mutex_up(&pinger_sem);
cfs_time_seconds(PING_INTERVAL));
pd->pd_this_ping = 0; /* XXX for debug */
- CDEBUG(D_HA, "finished a round ping\n");
+ CDEBUG(D_INFO, "finished a round ping\n");
pd->pd_recursion--;
return 0;
}
{
#ifdef ENABLE_PINGER
memset(&pinger_args, 0, sizeof(pinger_args));
- pinger_callback = liblustre_register_wait_callback(&pinger_check_rpcs,
+ pinger_callback = liblustre_register_wait_callback("pinger_check_rpcs",
+ &pinger_check_rpcs,
&pinger_args);
#endif
return 0;
void ptlrpc_pinger_wake_up()
{
#ifdef ENABLE_PINGER
- ENTRY;
/* XXX force pinger to run, if needed */
struct obd_import *imp;
+ ENTRY;
list_for_each_entry(imp, &pinger_imports, imp_pinger_chain) {
- CDEBUG(D_HA, "Checking that we need to do anything about import"
- " %s->%s\n", imp->imp_obd->obd_uuid.uuid,
- obd2cli_tgt(imp->imp_obd));
+ CDEBUG(D_RPCTRACE, "checking import %s->%s\n",
+ imp->imp_obd->obd_uuid.uuid, obd2cli_tgt(imp->imp_obd));
#ifdef ENABLE_LIBLUSTRE_RECOVERY
if (imp->imp_state == LUSTRE_IMP_DISCON && !imp->imp_deactive)
#else
obd2cli_tgt(imp->imp_obd), imp->imp_state,
imp->imp_deactive);
}
-#endif
EXIT;
+#endif
}
#endif /* !__KERNEL__ */
git://git.whamcloud.com / fs / lustre-release.git / blobdiff