1 /* -*- mode: c; c-basic-offset: 8; -*- 2 * vim: noexpandtab sw=8 ts=8 sts=0: 3 * 4 * Copyright (C) 2005 Oracle. All rights reserved. 5 * 6 * This program is free software; you can redistribute it and/or 7 * modify it under the terms of the GNU General Public 8 * License as published by the Free Software Foundation; either 9 * version 2 of the License, or (at your option) any later version. 10 * 11 * This program is distributed in the hope that it will be useful, 12 * but WITHOUT ANY WARRANTY; without even the implied warranty of 13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU 14 * General Public License for more details. 15 * 16 * You should have received a copy of the GNU General Public 17 * License along with this program; if not, write to the 18 * Free Software Foundation, Inc., 59 Temple Place - Suite 330, 19 * Boston, MA 021110-1307, USA. 20 */ 21 22 #ifndef O2CLUSTER_TCP_INTERNAL_H 23 #define O2CLUSTER_TCP_INTERNAL_H 24 25 #define O2NET_MSG_MAGIC ((u16)0xfa55) 26 #define O2NET_MSG_STATUS_MAGIC ((u16)0xfa56) 27 #define O2NET_MSG_KEEP_REQ_MAGIC ((u16)0xfa57) 28 #define O2NET_MSG_KEEP_RESP_MAGIC ((u16)0xfa58) 29 30 /* we're delaying our quorum decision so that heartbeat will have timed 31 * out truly dead nodes by the time we come around to making decisions 32 * on their number */ 33 #define O2NET_QUORUM_DELAY_MS ((o2hb_dead_threshold + 2) * O2HB_REGION_TIMEOUT_MS) 34 35 /* 36 * This version number represents quite a lot, unfortunately. It not 37 * only represents the raw network message protocol on the wire but also 38 * locking semantics of the file system using the protocol. It should 39 * be somewhere else, I'm sure, but right now it isn't. 40 * 41 * With version 11, we separate out the filesystem locking portion. The 42 * filesystem now has a major.minor version it negotiates. Version 11 43 * introduces this negotiation to the o2dlm protocol, and as such the 44 * version here in tcp_internal.h should not need to be bumped for 45 * filesystem locking changes. 46 * 47 * New in version 12 48 * - Negotiate hb timeout when storage is down. 49 * 50 * New in version 11 51 * - Negotiation of filesystem locking in the dlm join. 52 * 53 * New in version 10: 54 * - Meta/data locks combined 55 * 56 * New in version 9: 57 * - All votes removed 58 * 59 * New in version 8: 60 * - Replace delete inode votes with a cluster lock 61 * 62 * New in version 7: 63 * - DLM join domain includes the live nodemap 64 * 65 * New in version 6: 66 * - DLM lockres remote refcount fixes. 67 * 68 * New in version 5: 69 * - Network timeout checking protocol 70 * 71 * New in version 4: 72 * - Remove i_generation from lock names for better stat performance. 73 * 74 * New in version 3: 75 * - Replace dentry votes with a cluster lock 76 * 77 * New in version 2: 78 * - full 64 bit i_size in the metadata lock lvbs 79 * - introduction of "rw" lock and pushing meta/data locking down 80 */ 81 #define O2NET_PROTOCOL_VERSION 12ULL 82 struct o2net_handshake { 83 __be64 protocol_version; 84 __be64 connector_id; 85 __be32 o2hb_heartbeat_timeout_ms; 86 __be32 o2net_idle_timeout_ms; 87 __be32 o2net_keepalive_delay_ms; 88 __be32 o2net_reconnect_delay_ms; 89 }; 90 91 struct o2net_node { 92 /* this is never called from int/bh */ 93 spinlock_t nn_lock; 94 95 /* set the moment an sc is allocated and a connect is started */ 96 struct o2net_sock_container *nn_sc; 97 /* _valid is only set after the handshake passes and tx can happen */ 98 unsigned nn_sc_valid:1; 99 /* if this is set tx just returns it */ 100 int nn_persistent_error; 101 /* It is only set to 1 after the idle time out. */ 102 atomic_t nn_timeout; 103 104 /* threads waiting for an sc to arrive wait on the wq for generation 105 * to increase. it is increased when a connecting socket succeeds 106 * or fails or when an accepted socket is attached. */ 107 wait_queue_head_t nn_sc_wq; 108 109 struct idr nn_status_idr; 110 struct list_head nn_status_list; 111 112 /* connects are attempted from when heartbeat comes up until either hb 113 * goes down, the node is unconfigured, or a connect succeeds. 114 * connect_work is queued from set_nn_state both from hb up and from 115 * itself if a connect attempt fails and so can be self-arming. 116 * shutdown is careful to first mark the nn such that no connects will 117 * be attempted before canceling delayed connect work and flushing the 118 * queue. */ 119 struct delayed_work nn_connect_work; 120 unsigned long nn_last_connect_attempt; 121 122 /* this is queued as nodes come up and is canceled when a connection is 123 * established. this expiring gives up on the node and errors out 124 * transmits */ 125 struct delayed_work nn_connect_expired; 126 127 /* after we give up on a socket we wait a while before deciding 128 * that it is still heartbeating and that we should do some 129 * quorum work */ 130 struct delayed_work nn_still_up; 131 }; 132 133 struct o2net_sock_container { 134 struct kref sc_kref; 135 /* the next two are valid for the life time of the sc */ 136 struct socket *sc_sock; 137 struct o2nm_node *sc_node; 138 139 /* all of these sc work structs hold refs on the sc while they are 140 * queued. they should not be able to ref a freed sc. the teardown 141 * race is with o2net_wq destruction in o2net_stop_listening() */ 142 143 /* rx and connect work are generated from socket callbacks. sc 144 * shutdown removes the callbacks and then flushes the work queue */ 145 struct work_struct sc_rx_work; 146 struct work_struct sc_connect_work; 147 /* shutdown work is triggered in two ways. the simple way is 148 * for a code path calls ensure_shutdown which gets a lock, removes 149 * the sc from the nn, and queues the work. in this case the 150 * work is single-shot. the work is also queued from a sock 151 * callback, though, and in this case the work will find the sc 152 * still on the nn and will call ensure_shutdown itself.. this 153 * ends up triggering the shutdown work again, though nothing 154 * will be done in that second iteration. so work queue teardown 155 * has to be careful to remove the sc from the nn before waiting 156 * on the work queue so that the shutdown work doesn't remove the 157 * sc and rearm itself. 158 */ 159 struct work_struct sc_shutdown_work; 160 161 struct timer_list sc_idle_timeout; 162 struct delayed_work sc_keepalive_work; 163 164 unsigned sc_handshake_ok:1; 165 166 struct page *sc_page; 167 size_t sc_page_off; 168 169 /* original handlers for the sockets */ 170 void (*sc_state_change)(struct sock *sk); 171 void (*sc_data_ready)(struct sock *sk); 172 173 u32 sc_msg_key; 174 u16 sc_msg_type; 175 176 #ifdef CONFIG_DEBUG_FS 177 struct list_head sc_net_debug_item; 178 ktime_t sc_tv_timer; 179 ktime_t sc_tv_data_ready; 180 ktime_t sc_tv_advance_start; 181 ktime_t sc_tv_advance_stop; 182 ktime_t sc_tv_func_start; 183 ktime_t sc_tv_func_stop; 184 #endif 185 #ifdef CONFIG_OCFS2_FS_STATS 186 ktime_t sc_tv_acquiry_total; 187 ktime_t sc_tv_send_total; 188 ktime_t sc_tv_status_total; 189 u32 sc_send_count; 190 u32 sc_recv_count; 191 ktime_t sc_tv_process_total; 192 #endif 193 struct mutex sc_send_lock; 194 }; 195 196 struct o2net_msg_handler { 197 struct rb_node nh_node; 198 u32 nh_max_len; 199 u32 nh_msg_type; 200 u32 nh_key; 201 o2net_msg_handler_func *nh_func; 202 o2net_msg_handler_func *nh_func_data; 203 o2net_post_msg_handler_func 204 *nh_post_func; 205 struct kref nh_kref; 206 struct list_head nh_unregister_item; 207 }; 208 209 enum o2net_system_error { 210 O2NET_ERR_NONE = 0, 211 O2NET_ERR_NO_HNDLR, 212 O2NET_ERR_OVERFLOW, 213 O2NET_ERR_DIED, 214 O2NET_ERR_MAX 215 }; 216 217 struct o2net_status_wait { 218 enum o2net_system_error ns_sys_status; 219 s32 ns_status; 220 int ns_id; 221 wait_queue_head_t ns_wq; 222 struct list_head ns_node_item; 223 }; 224 225 #ifdef CONFIG_DEBUG_FS 226 /* just for state dumps */ 227 struct o2net_send_tracking { 228 struct list_head st_net_debug_item; 229 struct task_struct *st_task; 230 struct o2net_sock_container *st_sc; 231 u32 st_id; 232 u32 st_msg_type; 233 u32 st_msg_key; 234 u8 st_node; 235 ktime_t st_sock_time; 236 ktime_t st_send_time; 237 ktime_t st_status_time; 238 }; 239 #else 240 struct o2net_send_tracking { 241 u32 dummy; 242 }; 243 #endif /* CONFIG_DEBUG_FS */ 244 245 #endif /* O2CLUSTER_TCP_INTERNAL_H */ 246