198211489SZach Brown /* -*- mode: c; c-basic-offset: 8; -*- 298211489SZach Brown * vim: noexpandtab sw=8 ts=8 sts=0: 398211489SZach Brown * 498211489SZach Brown * Copyright (C) 2005 Oracle. All rights reserved. 598211489SZach Brown * 698211489SZach Brown * This program is free software; you can redistribute it and/or 798211489SZach Brown * modify it under the terms of the GNU General Public 898211489SZach Brown * License as published by the Free Software Foundation; either 998211489SZach Brown * version 2 of the License, or (at your option) any later version. 1098211489SZach Brown * 1198211489SZach Brown * This program is distributed in the hope that it will be useful, 1298211489SZach Brown * but WITHOUT ANY WARRANTY; without even the implied warranty of 1398211489SZach Brown * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU 1498211489SZach Brown * General Public License for more details. 1598211489SZach Brown * 1698211489SZach Brown * You should have received a copy of the GNU General Public 1798211489SZach Brown * License along with this program; if not, write to the 1898211489SZach Brown * Free Software Foundation, Inc., 59 Temple Place - Suite 330, 1998211489SZach Brown * Boston, MA 021110-1307, USA. 2098211489SZach Brown */ 2198211489SZach Brown 2298211489SZach Brown #ifndef O2CLUSTER_TCP_INTERNAL_H 2398211489SZach Brown #define O2CLUSTER_TCP_INTERNAL_H 2498211489SZach Brown 2598211489SZach Brown #define O2NET_MSG_MAGIC ((u16)0xfa55) 2698211489SZach Brown #define O2NET_MSG_STATUS_MAGIC ((u16)0xfa56) 2798211489SZach Brown #define O2NET_MSG_KEEP_REQ_MAGIC ((u16)0xfa57) 2898211489SZach Brown #define O2NET_MSG_KEEP_RESP_MAGIC ((u16)0xfa58) 2998211489SZach Brown 3098211489SZach Brown /* we're delaying our quorum decision so that heartbeat will have timed 3198211489SZach Brown * out truly dead nodes by the time we come around to making decisions 3298211489SZach Brown * on their number */ 3398211489SZach Brown #define O2NET_QUORUM_DELAY_MS ((o2hb_dead_threshold + 2) * O2HB_REGION_TIMEOUT_MS) 3498211489SZach Brown 3598211489SZach Brown /* 3698211489SZach Brown * This version number represents quite a lot, unfortunately. It not 3798211489SZach Brown * only represents the raw network message protocol on the wire but also 3898211489SZach Brown * locking semantics of the file system using the protocol. It should 3998211489SZach Brown * be somewhere else, I'm sure, but right now it isn't. 4098211489SZach Brown * 41*828ae6afSAndrew Beekhof * New in version 5: 42*828ae6afSAndrew Beekhof * - Network timeout checking protocol 43*828ae6afSAndrew Beekhof * 4424c19ef4SMark Fasheh * New in version 4: 4524c19ef4SMark Fasheh * - Remove i_generation from lock names for better stat performance. 4624c19ef4SMark Fasheh * 47379dfe9dSMark Fasheh * New in version 3: 48379dfe9dSMark Fasheh * - Replace dentry votes with a cluster lock 49379dfe9dSMark Fasheh * 5098211489SZach Brown * New in version 2: 5198211489SZach Brown * - full 64 bit i_size in the metadata lock lvbs 5298211489SZach Brown * - introduction of "rw" lock and pushing meta/data locking down 5398211489SZach Brown */ 54*828ae6afSAndrew Beekhof #define O2NET_PROTOCOL_VERSION 5ULL 5598211489SZach Brown struct o2net_handshake { 5698211489SZach Brown __be64 protocol_version; 5798211489SZach Brown __be64 connector_id; 58*828ae6afSAndrew Beekhof __be32 o2hb_heartbeat_timeout_ms; 59*828ae6afSAndrew Beekhof __be32 o2net_idle_timeout_ms; 60*828ae6afSAndrew Beekhof __be32 o2net_keepalive_delay_ms; 61*828ae6afSAndrew Beekhof __be32 o2net_reconnect_delay_ms; 6298211489SZach Brown }; 6398211489SZach Brown 6498211489SZach Brown struct o2net_node { 6598211489SZach Brown /* this is never called from int/bh */ 6698211489SZach Brown spinlock_t nn_lock; 6798211489SZach Brown 6898211489SZach Brown /* set the moment an sc is allocated and a connect is started */ 6998211489SZach Brown struct o2net_sock_container *nn_sc; 7098211489SZach Brown /* _valid is only set after the handshake passes and tx can happen */ 7198211489SZach Brown unsigned nn_sc_valid:1; 7298211489SZach Brown /* if this is set tx just returns it */ 7398211489SZach Brown int nn_persistent_error; 7498211489SZach Brown 7598211489SZach Brown /* threads waiting for an sc to arrive wait on the wq for generation 7698211489SZach Brown * to increase. it is increased when a connecting socket succeeds 7798211489SZach Brown * or fails or when an accepted socket is attached. */ 7898211489SZach Brown wait_queue_head_t nn_sc_wq; 7998211489SZach Brown 8098211489SZach Brown struct idr nn_status_idr; 8198211489SZach Brown struct list_head nn_status_list; 8298211489SZach Brown 8398211489SZach Brown /* connects are attempted from when heartbeat comes up until either hb 8498211489SZach Brown * goes down, the node is unconfigured, no connect attempts succeed 8598211489SZach Brown * before O2NET_CONN_IDLE_DELAY, or a connect succeeds. connect_work 8698211489SZach Brown * is queued from set_nn_state both from hb up and from itself if a 8798211489SZach Brown * connect attempt fails and so can be self-arming. shutdown is 8898211489SZach Brown * careful to first mark the nn such that no connects will be attempted 8998211489SZach Brown * before canceling delayed connect work and flushing the queue. */ 90c4028958SDavid Howells struct delayed_work nn_connect_work; 9198211489SZach Brown unsigned long nn_last_connect_attempt; 9298211489SZach Brown 9398211489SZach Brown /* this is queued as nodes come up and is canceled when a connection is 9498211489SZach Brown * established. this expiring gives up on the node and errors out 9598211489SZach Brown * transmits */ 96c4028958SDavid Howells struct delayed_work nn_connect_expired; 9798211489SZach Brown 9898211489SZach Brown /* after we give up on a socket we wait a while before deciding 9998211489SZach Brown * that it is still heartbeating and that we should do some 10098211489SZach Brown * quorum work */ 101c4028958SDavid Howells struct delayed_work nn_still_up; 10298211489SZach Brown }; 10398211489SZach Brown 10498211489SZach Brown struct o2net_sock_container { 10598211489SZach Brown struct kref sc_kref; 10698211489SZach Brown /* the next two are vaild for the life time of the sc */ 10798211489SZach Brown struct socket *sc_sock; 10898211489SZach Brown struct o2nm_node *sc_node; 10998211489SZach Brown 11098211489SZach Brown /* all of these sc work structs hold refs on the sc while they are 11198211489SZach Brown * queued. they should not be able to ref a freed sc. the teardown 11298211489SZach Brown * race is with o2net_wq destruction in o2net_stop_listening() */ 11398211489SZach Brown 11498211489SZach Brown /* rx and connect work are generated from socket callbacks. sc 11598211489SZach Brown * shutdown removes the callbacks and then flushes the work queue */ 11698211489SZach Brown struct work_struct sc_rx_work; 11798211489SZach Brown struct work_struct sc_connect_work; 11898211489SZach Brown /* shutdown work is triggered in two ways. the simple way is 11998211489SZach Brown * for a code path calls ensure_shutdown which gets a lock, removes 12098211489SZach Brown * the sc from the nn, and queues the work. in this case the 12198211489SZach Brown * work is single-shot. the work is also queued from a sock 12298211489SZach Brown * callback, though, and in this case the work will find the sc 12398211489SZach Brown * still on the nn and will call ensure_shutdown itself.. this 12498211489SZach Brown * ends up triggering the shutdown work again, though nothing 12598211489SZach Brown * will be done in that second iteration. so work queue teardown 12698211489SZach Brown * has to be careful to remove the sc from the nn before waiting 12798211489SZach Brown * on the work queue so that the shutdown work doesn't remove the 12898211489SZach Brown * sc and rearm itself. 12998211489SZach Brown */ 13098211489SZach Brown struct work_struct sc_shutdown_work; 13198211489SZach Brown 13298211489SZach Brown struct timer_list sc_idle_timeout; 133c4028958SDavid Howells struct delayed_work sc_keepalive_work; 13498211489SZach Brown 13598211489SZach Brown unsigned sc_handshake_ok:1; 13698211489SZach Brown 13798211489SZach Brown struct page *sc_page; 13898211489SZach Brown size_t sc_page_off; 13998211489SZach Brown 14098211489SZach Brown /* original handlers for the sockets */ 14198211489SZach Brown void (*sc_state_change)(struct sock *sk); 14298211489SZach Brown void (*sc_data_ready)(struct sock *sk, int bytes); 14398211489SZach Brown 14498211489SZach Brown struct timeval sc_tv_timer; 14598211489SZach Brown struct timeval sc_tv_data_ready; 14698211489SZach Brown struct timeval sc_tv_advance_start; 14798211489SZach Brown struct timeval sc_tv_advance_stop; 14898211489SZach Brown struct timeval sc_tv_func_start; 14998211489SZach Brown struct timeval sc_tv_func_stop; 15098211489SZach Brown u32 sc_msg_key; 15198211489SZach Brown u16 sc_msg_type; 15298211489SZach Brown }; 15398211489SZach Brown 15498211489SZach Brown struct o2net_msg_handler { 15598211489SZach Brown struct rb_node nh_node; 15698211489SZach Brown u32 nh_max_len; 15798211489SZach Brown u32 nh_msg_type; 15898211489SZach Brown u32 nh_key; 15998211489SZach Brown o2net_msg_handler_func *nh_func; 16098211489SZach Brown o2net_msg_handler_func *nh_func_data; 16198211489SZach Brown struct kref nh_kref; 16298211489SZach Brown struct list_head nh_unregister_item; 16398211489SZach Brown }; 16498211489SZach Brown 16598211489SZach Brown enum o2net_system_error { 16698211489SZach Brown O2NET_ERR_NONE = 0, 16798211489SZach Brown O2NET_ERR_NO_HNDLR, 16898211489SZach Brown O2NET_ERR_OVERFLOW, 16998211489SZach Brown O2NET_ERR_DIED, 17098211489SZach Brown O2NET_ERR_MAX 17198211489SZach Brown }; 17298211489SZach Brown 17398211489SZach Brown struct o2net_status_wait { 17498211489SZach Brown enum o2net_system_error ns_sys_status; 17598211489SZach Brown s32 ns_status; 17698211489SZach Brown int ns_id; 17798211489SZach Brown wait_queue_head_t ns_wq; 17898211489SZach Brown struct list_head ns_node_item; 17998211489SZach Brown }; 18098211489SZach Brown 18198211489SZach Brown #endif /* O2CLUSTER_TCP_INTERNAL_H */ 182