char *name = obddev->obd_type->typ_name;
ldlm_ns_type_t ns_type = LDLM_NS_TYPE_UNKNOWN;
int rc;
- char *cli_name = lustre_cfg_buf(lcfg, 0);
ENTRY;
/* In a more perfect world, we would hang a ptlrpc_client off of
* obd_type and just use the values from there. */
- if (!strcmp(name, LUSTRE_OSC_NAME) ||
- (!(strcmp(name, LUSTRE_OSP_NAME)) &&
- (is_osp_on_mdt(cli_name) &&
- strstr(lustre_cfg_buf(lcfg, 1), "OST") != NULL))) {
- /* OSC or OSP_on_MDT for OSTs */
- rq_portal = OST_REQUEST_PORTAL;
- rp_portal = OSC_REPLY_PORTAL;
- connect_op = OST_CONNECT;
- cli->cl_sp_me = LUSTRE_SP_CLI;
- cli->cl_sp_to = LUSTRE_SP_OST;
- ns_type = LDLM_NS_TYPE_OSC;
+ if (!strcmp(name, LUSTRE_OSC_NAME)) {
+ rq_portal = OST_REQUEST_PORTAL;
+ rp_portal = OSC_REPLY_PORTAL;
+ connect_op = OST_CONNECT;
+ cli->cl_sp_me = LUSTRE_SP_CLI;
+ cli->cl_sp_to = LUSTRE_SP_OST;
+ ns_type = LDLM_NS_TYPE_OSC;
} else if (!strcmp(name, LUSTRE_MDC_NAME) ||
- !strcmp(name, LUSTRE_LWP_NAME) ||
- (!strcmp(name, LUSTRE_OSP_NAME) &&
- (is_osp_on_mdt(cli_name) &&
- strstr(lustre_cfg_buf(lcfg, 1), "OST") == NULL))) {
- /* MDC or OSP_on_MDT for other MDTs */
- rq_portal = MDS_REQUEST_PORTAL;
- rp_portal = MDC_REPLY_PORTAL;
- connect_op = MDS_CONNECT;
- cli->cl_sp_me = LUSTRE_SP_CLI;
- cli->cl_sp_to = LUSTRE_SP_MDT;
- ns_type = LDLM_NS_TYPE_MDC;
+ !strcmp(name, LUSTRE_LWP_NAME)) {
+ rq_portal = MDS_REQUEST_PORTAL;
+ rp_portal = MDC_REPLY_PORTAL;
+ connect_op = MDS_CONNECT;
+ cli->cl_sp_me = LUSTRE_SP_CLI;
+ cli->cl_sp_to = LUSTRE_SP_MDT;
+ ns_type = LDLM_NS_TYPE_MDC;
+ } else if (!strcmp(name, LUSTRE_OSP_NAME)) {
+ if (strstr(lustre_cfg_buf(lcfg, 1), "OST") == NULL) {
+ /* OSP_on_MDT for other MDTs */
+ connect_op = MDS_CONNECT;
+ cli->cl_sp_to = LUSTRE_SP_MDT;
+ ns_type = LDLM_NS_TYPE_MDC;
+ rq_portal = OUT_PORTAL;
+ } else {
+ /* OSP on MDT for OST */
+ connect_op = OST_CONNECT;
+ cli->cl_sp_to = LUSTRE_SP_OST;
+ ns_type = LDLM_NS_TYPE_OSC;
+ rq_portal = OST_REQUEST_PORTAL;
+ }
+ rp_portal = OSC_REPLY_PORTAL;
+ cli->cl_sp_me = LUSTRE_SP_CLI;
} else if (!strcmp(name, LUSTRE_MGC_NAME)) {
rq_portal = MGS_REQUEST_PORTAL;
rp_portal = MGC_REPLY_PORTAL;
cli->cl_avail_grant = 0;
/* FIXME: Should limit this for the sum of all cl_dirty_max. */
cli->cl_dirty_max = OSC_MAX_DIRTY_DEFAULT * 1024 * 1024;
- if (cli->cl_dirty_max >> PAGE_CACHE_SHIFT > num_physpages / 8)
- cli->cl_dirty_max = num_physpages << (PAGE_CACHE_SHIFT - 3);
+ if (cli->cl_dirty_max >> PAGE_CACHE_SHIFT > totalram_pages / 8)
+ cli->cl_dirty_max = totalram_pages << (PAGE_CACHE_SHIFT - 3);
CFS_INIT_LIST_HEAD(&cli->cl_cache_waiters);
CFS_INIT_LIST_HEAD(&cli->cl_loi_ready_list);
CFS_INIT_LIST_HEAD(&cli->cl_loi_hp_ready_list);
cfs_atomic_set(&cli->cl_lru_in_list, 0);
CFS_INIT_LIST_HEAD(&cli->cl_lru_list);
client_obd_list_lock_init(&cli->cl_lru_list_lock);
+ cfs_atomic_set(&cli->cl_unstable_count, 0);
init_waitqueue_head(&cli->cl_destroy_waitq);
cfs_atomic_set(&cli->cl_destroy_in_flight, 0);
cli->cl_max_pages_per_rpc = min_t(int, PTLRPC_MAX_BRW_PAGES,
LNET_MTU >> PAGE_CACHE_SHIFT);
+ /* set cl_chunkbits default value to PAGE_CACHE_SHIFT,
+ * it will be updated at OSC connection time. */
+ cli->cl_chunkbits = PAGE_CACHE_SHIFT;
+
if (!strcmp(name, LUSTRE_MDC_NAME)) {
cli->cl_max_rpcs_in_flight = MDC_MAX_RIF_DEFAULT;
- } else if (num_physpages >> (20 - PAGE_CACHE_SHIFT) <= 128 /* MB */) {
+ } else if (totalram_pages >> (20 - PAGE_CACHE_SHIFT) <= 128 /* MB */) {
cli->cl_max_rpcs_in_flight = 2;
- } else if (num_physpages >> (20 - PAGE_CACHE_SHIFT) <= 256 /* MB */) {
+ } else if (totalram_pages >> (20 - PAGE_CACHE_SHIFT) <= 256 /* MB */) {
cli->cl_max_rpcs_in_flight = 3;
- } else if (num_physpages >> (20 - PAGE_CACHE_SHIFT) <= 512 /* MB */) {
+ } else if (totalram_pages >> (20 - PAGE_CACHE_SHIFT) <= 512 /* MB */) {
cli->cl_max_rpcs_in_flight = 4;
} else {
if (osc_on_mdt(obddev->obd_name))
int client_obd_cleanup(struct obd_device *obddev)
{
- ENTRY;
+ ENTRY;
- ldlm_namespace_free_post(obddev->obd_namespace);
- obddev->obd_namespace = NULL;
+ ldlm_namespace_free_post(obddev->obd_namespace);
+ obddev->obd_namespace = NULL;
- LASSERT(obddev->u.cli.cl_import == NULL);
+ obd_cleanup_client_import(obddev);
+ LASSERT(obddev->u.cli.cl_import == NULL);
- ldlm_put_ref();
- RETURN(0);
+ ldlm_put_ref();
+ RETURN(0);
}
EXPORT_SYMBOL(client_obd_cleanup);
if (!target) {
deuuidify(str, NULL, &target_start, &target_len);
LCONSOLE_ERROR_MSG(0x137, "%s: not available for connect "
- "from %s (no target)\n", str,
+ "from %s (no target). If you are running "
+ "an HA pair check that the target is "
+ "mounted on the other server.\n", str,
libcfs_nid2str(req->rq_peer.nid));
GOTO(out, rc = -ENODEV);
}
/* Make sure the target isn't cleaned up while we're here. Yes,
* there's still a race between the above check and our incref here.
* Really, class_uuid2obd should take the ref. */
- targref = class_incref(target, __FUNCTION__, cfs_current());
+ targref = class_incref(target, __FUNCTION__, current);
target->obd_conn_inprogress++;
spin_unlock(&target->obd_dev_lock);
libcfs_nid2str(req->rq_peer.nid),
cfs_atomic_read(&export->exp_refcount));
GOTO(out, rc = -EBUSY);
- } else if (req->rq_export != NULL &&
- (cfs_atomic_read(&export->exp_rpc_count) > 1)) {
- /* The current connect RPC has increased exp_rpc_count. */
- LCONSOLE_WARN("%s: Client %s (at %s) refused reconnection, "
- "still busy with %d active RPCs\n",
- target->obd_name, cluuid.uuid,
- libcfs_nid2str(req->rq_peer.nid),
- cfs_atomic_read(&export->exp_rpc_count) - 1);
- spin_lock(&export->exp_lock);
- if (req->rq_export->exp_conn_cnt <
- lustre_msg_get_conn_cnt(req->rq_reqmsg))
- /* try to abort active requests */
- req->rq_export->exp_abort_active_req = 1;
- spin_unlock(&export->exp_lock);
- GOTO(out, rc = -EBUSY);
} else if (lustre_msg_get_conn_cnt(req->rq_reqmsg) == 1) {
if (!strstr(cluuid.uuid, "mdt"))
LCONSOLE_WARN("%s: Rejecting reconnect from the "
export, (long)cfs_time_current_sec(),
export ? (long)export->exp_last_request_time : 0);
- /* If this is the first time a client connects, reset the recovery
+ /* If this is the first time a client connects, reset the recovery
* timer. Discard lightweight connections which might be local. */
if (!lw_client && rc == 0 && target->obd_recovering)
check_and_start_recovery_timer(target, req, export == NULL);
}
LASSERT(lustre_msg_get_conn_cnt(req->rq_reqmsg) > 0);
export->exp_conn_cnt = lustre_msg_get_conn_cnt(req->rq_reqmsg);
- export->exp_abort_active_req = 0;
/* Don't evict liblustre clients for not pinging. */
if (lustre_msg_get_op_flags(req->rq_reqmsg) & MSG_CONNECT_LIBCLIENT) {
target->obd_conn_inprogress--;
spin_unlock(&target->obd_dev_lock);
- class_decref(targref, __func__, cfs_current());
+ class_decref(targref, __func__, current);
}
if (rc)
req->rq_status = rc;
if (req->rq_export->exp_disconnected)
GOTO(reqcopy_put, rc = 0);
- rc = lu_context_init(&req->rq_recov_session, LCT_SESSION);
+ rc = lu_context_init(&req->rq_recov_session, LCT_SERVER_SESSION);
if (rc) {
CERROR("Failure to initialize session: %d\n", rc);
GOTO(reqcopy_put, rc);
thread->t_env = env;
thread->t_id = -1; /* force filter_iobuf_get/put to use local buffers */
env->le_ctx.lc_thread = thread;
- thread->t_data = NULL;
- thread->t_watchdog = NULL;
+ tgt_io_thread_init(thread); /* init thread_big_cache for IO requests */
+ thread->t_watchdog = NULL;
CDEBUG(D_HA, "%s: started recovery thread pid %d\n", obd->obd_name,
current_pid());
trd->trd_processing_task = 0;
complete(&trd->trd_finishing);
- OBD_FREE_PTR(thread);
- OBD_FREE_PTR(env);
- RETURN(rc);
+ tgt_io_thread_done(thread);
+ OBD_FREE_PTR(thread);
+ OBD_FREE_PTR(env);
+ RETURN(rc);
}
static int target_start_recovery_thread(struct lu_target *lut,
int target_bulk_io(struct obd_export *exp, struct ptlrpc_bulk_desc *desc,
struct l_wait_info *lwi)
{
- struct ptlrpc_request *req = desc->bd_req;
- int rc = 0;
- ENTRY;
+ struct ptlrpc_request *req = desc->bd_req;
+ time_t start = cfs_time_current_sec();
+ int rc = 0;
+
+ ENTRY;
/* If there is eviction in progress, wait for it to finish. */
- if (unlikely(cfs_atomic_read(&exp->exp_obd->obd_evict_inprogress))) {
- *lwi = LWI_INTR(NULL, NULL);
- rc = l_wait_event(exp->exp_obd->obd_evict_inprogress_waitq,
- !cfs_atomic_read(&exp->exp_obd->
- obd_evict_inprogress),
- lwi);
- }
+ if (unlikely(cfs_atomic_read(&exp->exp_obd->obd_evict_inprogress))) {
+ *lwi = LWI_INTR(NULL, NULL);
+ rc = l_wait_event(exp->exp_obd->obd_evict_inprogress_waitq,
+ !cfs_atomic_read(&exp->exp_obd->
+ obd_evict_inprogress),
+ lwi);
+ }
- /* Check if client was evicted or tried to reconnect already. */
- if (exp->exp_failed || exp->exp_abort_active_req) {
- rc = -ENOTCONN;
- } else {
- if (desc->bd_type == BULK_PUT_SINK)
- rc = sptlrpc_svc_wrap_bulk(req, desc);
- if (rc == 0)
- rc = ptlrpc_start_bulk_transfer(desc);
- }
-
- if (rc == 0 && OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE)) {
- ptlrpc_abort_bulk(desc);
- } else if (rc == 0) {
- time_t start = cfs_time_current_sec();
- do {
- long timeoutl = req->rq_deadline - cfs_time_current_sec();
- cfs_duration_t timeout = timeoutl <= 0 ?
- CFS_TICK : cfs_time_seconds(timeoutl);
- *lwi = LWI_TIMEOUT_INTERVAL(timeout,
- cfs_time_seconds(1),
- target_bulk_timeout,
- desc);
- rc = l_wait_event(desc->bd_waitq,
- !ptlrpc_server_bulk_active(desc) ||
- exp->exp_failed ||
- exp->exp_abort_active_req,
- lwi);
- LASSERT(rc == 0 || rc == -ETIMEDOUT);
- /* Wait again if we changed deadline. */
- } while ((rc == -ETIMEDOUT) &&
- (req->rq_deadline > cfs_time_current_sec()));
-
- if (rc == -ETIMEDOUT) {
- DEBUG_REQ(D_ERROR, req,
- "timeout on bulk %s after %ld%+lds",
- bulk2type(desc),
- req->rq_deadline - start,
- cfs_time_current_sec() -
- req->rq_deadline);
- ptlrpc_abort_bulk(desc);
- } else if (exp->exp_failed) {
- DEBUG_REQ(D_ERROR, req, "Eviction on bulk %s",
- bulk2type(desc));
- rc = -ENOTCONN;
- ptlrpc_abort_bulk(desc);
- } else if (exp->exp_abort_active_req) {
- DEBUG_REQ(D_ERROR, req, "Reconnect on bulk %s",
- bulk2type(desc));
- /* We don't reply anyway. */
- rc = -ETIMEDOUT;
- ptlrpc_abort_bulk(desc);
- } else if (desc->bd_failure ||
- desc->bd_nob_transferred != desc->bd_nob) {
- DEBUG_REQ(D_ERROR, req, "%s bulk %s %d(%d)",
- desc->bd_failure ?
- "network error on" : "truncated",
- bulk2type(desc),
- desc->bd_nob_transferred,
- desc->bd_nob);
- /* XXX Should this be a different errno? */
- rc = -ETIMEDOUT;
- } else if (desc->bd_type == BULK_GET_SINK) {
- rc = sptlrpc_svc_unwrap_bulk(req, desc);
- }
- } else {
- DEBUG_REQ(D_ERROR, req, "bulk %s failed: rc %d",
- bulk2type(desc), rc);
- }
+ /* Check if client was evicted or reconnected already. */
+ if (exp->exp_failed ||
+ exp->exp_conn_cnt > lustre_msg_get_conn_cnt(req->rq_reqmsg)) {
+ rc = -ENOTCONN;
+ } else {
+ if (desc->bd_type == BULK_PUT_SINK)
+ rc = sptlrpc_svc_wrap_bulk(req, desc);
+ if (rc == 0)
+ rc = ptlrpc_start_bulk_transfer(desc);
+ }
- RETURN(rc);
+ if (rc < 0) {
+ DEBUG_REQ(D_ERROR, req, "bulk %s failed: rc %d",
+ bulk2type(desc), rc);
+ RETURN(rc);
+ }
+
+ if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE)) {
+ ptlrpc_abort_bulk(desc);
+ RETURN(0);
+ }
+
+ do {
+ long timeoutl = req->rq_deadline - cfs_time_current_sec();
+ cfs_duration_t timeout = timeoutl <= 0 ?
+ CFS_TICK : cfs_time_seconds(timeoutl);
+
+ *lwi = LWI_TIMEOUT_INTERVAL(timeout, cfs_time_seconds(1),
+ target_bulk_timeout, desc);
+ rc = l_wait_event(desc->bd_waitq,
+ !ptlrpc_server_bulk_active(desc) ||
+ exp->exp_failed ||
+ exp->exp_conn_cnt >
+ lustre_msg_get_conn_cnt(req->rq_reqmsg),
+ lwi);
+ LASSERT(rc == 0 || rc == -ETIMEDOUT);
+ /* Wait again if we changed deadline. */
+ } while ((rc == -ETIMEDOUT) &&
+ (req->rq_deadline > cfs_time_current_sec()));
+
+ if (rc == -ETIMEDOUT) {
+ DEBUG_REQ(D_ERROR, req, "timeout on bulk %s after %ld%+lds",
+ bulk2type(desc), req->rq_deadline - start,
+ cfs_time_current_sec() - req->rq_deadline);
+ ptlrpc_abort_bulk(desc);
+ } else if (exp->exp_failed) {
+ DEBUG_REQ(D_ERROR, req, "Eviction on bulk %s",
+ bulk2type(desc));
+ rc = -ENOTCONN;
+ ptlrpc_abort_bulk(desc);
+ } else if (exp->exp_conn_cnt >
+ lustre_msg_get_conn_cnt(req->rq_reqmsg)) {
+ DEBUG_REQ(D_ERROR, req, "Reconnect on bulk %s",
+ bulk2type(desc));
+ /* We don't reply anyway. */
+ rc = -ETIMEDOUT;
+ ptlrpc_abort_bulk(desc);
+ } else if (desc->bd_failure ||
+ desc->bd_nob_transferred != desc->bd_nob) {
+ DEBUG_REQ(D_ERROR, req, "%s bulk %s %d(%d)",
+ desc->bd_failure ? "network error on" : "truncated",
+ bulk2type(desc), desc->bd_nob_transferred,
+ desc->bd_nob);
+ /* XXX Should this be a different errno? */
+ rc = -ETIMEDOUT;
+ } else if (desc->bd_type == BULK_GET_SINK) {
+ rc = sptlrpc_svc_unwrap_bulk(req, desc);
+ }
+
+ RETURN(rc);
}
EXPORT_SYMBOL(target_bulk_io);
git://git.whamcloud.com / fs / lustre-release.git / blobdiff