Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1 | /* peer.c: Rx RPC peer management |
| 2 | * |
| 3 | * Copyright (C) 2002 Red Hat, Inc. All Rights Reserved. |
| 4 | * Written by David Howells (dhowells@redhat.com) |
| 5 | * |
| 6 | * This program is free software; you can redistribute it and/or |
| 7 | * modify it under the terms of the GNU General Public License |
| 8 | * as published by the Free Software Foundation; either version |
| 9 | * 2 of the License, or (at your option) any later version. |
| 10 | */ |
| 11 | |
| 12 | #include <linux/sched.h> |
| 13 | #include <linux/slab.h> |
| 14 | #include <linux/module.h> |
| 15 | #include <rxrpc/rxrpc.h> |
| 16 | #include <rxrpc/transport.h> |
| 17 | #include <rxrpc/peer.h> |
| 18 | #include <rxrpc/connection.h> |
| 19 | #include <rxrpc/call.h> |
| 20 | #include <rxrpc/message.h> |
| 21 | #include <linux/udp.h> |
| 22 | #include <linux/ip.h> |
| 23 | #include <net/sock.h> |
| 24 | #include <asm/uaccess.h> |
| 25 | #include <asm/div64.h> |
| 26 | #include "internal.h" |
| 27 | |
| 28 | __RXACCT_DECL(atomic_t rxrpc_peer_count); |
| 29 | LIST_HEAD(rxrpc_peers); |
| 30 | DECLARE_RWSEM(rxrpc_peers_sem); |
| 31 | unsigned long rxrpc_peer_timeout = 12 * 60 * 60; |
| 32 | |
| 33 | static void rxrpc_peer_do_timeout(struct rxrpc_peer *peer); |
| 34 | |
| 35 | static void __rxrpc_peer_timeout(rxrpc_timer_t *timer) |
| 36 | { |
| 37 | struct rxrpc_peer *peer = |
| 38 | list_entry(timer, struct rxrpc_peer, timeout); |
| 39 | |
| 40 | _debug("Rx PEER TIMEOUT [%p{u=%d}]", peer, atomic_read(&peer->usage)); |
| 41 | |
| 42 | rxrpc_peer_do_timeout(peer); |
| 43 | } |
| 44 | |
| 45 | static const struct rxrpc_timer_ops rxrpc_peer_timer_ops = { |
| 46 | .timed_out = __rxrpc_peer_timeout, |
| 47 | }; |
| 48 | |
| 49 | /*****************************************************************************/ |
| 50 | /* |
| 51 | * create a peer record |
| 52 | */ |
| 53 | static int __rxrpc_create_peer(struct rxrpc_transport *trans, __be32 addr, |
| 54 | struct rxrpc_peer **_peer) |
| 55 | { |
| 56 | struct rxrpc_peer *peer; |
| 57 | |
| 58 | _enter("%p,%08x", trans, ntohl(addr)); |
| 59 | |
| 60 | /* allocate and initialise a peer record */ |
| 61 | peer = kmalloc(sizeof(struct rxrpc_peer), GFP_KERNEL); |
| 62 | if (!peer) { |
| 63 | _leave(" = -ENOMEM"); |
| 64 | return -ENOMEM; |
| 65 | } |
| 66 | |
| 67 | memset(peer, 0, sizeof(struct rxrpc_peer)); |
| 68 | atomic_set(&peer->usage, 1); |
| 69 | |
| 70 | INIT_LIST_HEAD(&peer->link); |
| 71 | INIT_LIST_HEAD(&peer->proc_link); |
| 72 | INIT_LIST_HEAD(&peer->conn_idlist); |
| 73 | INIT_LIST_HEAD(&peer->conn_active); |
| 74 | INIT_LIST_HEAD(&peer->conn_graveyard); |
| 75 | spin_lock_init(&peer->conn_gylock); |
| 76 | init_waitqueue_head(&peer->conn_gy_waitq); |
| 77 | rwlock_init(&peer->conn_idlock); |
| 78 | rwlock_init(&peer->conn_lock); |
| 79 | atomic_set(&peer->conn_count, 0); |
| 80 | spin_lock_init(&peer->lock); |
| 81 | rxrpc_timer_init(&peer->timeout, &rxrpc_peer_timer_ops); |
| 82 | |
| 83 | peer->addr.s_addr = addr; |
| 84 | |
| 85 | peer->trans = trans; |
| 86 | peer->ops = trans->peer_ops; |
| 87 | |
| 88 | __RXACCT(atomic_inc(&rxrpc_peer_count)); |
| 89 | *_peer = peer; |
| 90 | _leave(" = 0 (%p)", peer); |
| 91 | |
| 92 | return 0; |
| 93 | } /* end __rxrpc_create_peer() */ |
| 94 | |
| 95 | /*****************************************************************************/ |
| 96 | /* |
| 97 | * find a peer record on the specified transport |
| 98 | * - returns (if successful) with peer record usage incremented |
| 99 | * - resurrects it from the graveyard if found there |
| 100 | */ |
| 101 | int rxrpc_peer_lookup(struct rxrpc_transport *trans, __be32 addr, |
| 102 | struct rxrpc_peer **_peer) |
| 103 | { |
| 104 | struct rxrpc_peer *peer, *candidate = NULL; |
| 105 | struct list_head *_p; |
| 106 | int ret; |
| 107 | |
| 108 | _enter("%p{%hu},%08x", trans, trans->port, ntohl(addr)); |
| 109 | |
| 110 | /* [common case] search the transport's active list first */ |
| 111 | read_lock(&trans->peer_lock); |
| 112 | list_for_each(_p, &trans->peer_active) { |
| 113 | peer = list_entry(_p, struct rxrpc_peer, link); |
| 114 | if (peer->addr.s_addr == addr) |
| 115 | goto found_active; |
| 116 | } |
| 117 | read_unlock(&trans->peer_lock); |
| 118 | |
| 119 | /* [uncommon case] not active - create a candidate for a new record */ |
| 120 | ret = __rxrpc_create_peer(trans, addr, &candidate); |
| 121 | if (ret < 0) { |
| 122 | _leave(" = %d", ret); |
| 123 | return ret; |
| 124 | } |
| 125 | |
| 126 | /* search the active list again, just in case it appeared whilst we |
| 127 | * were busy */ |
| 128 | write_lock(&trans->peer_lock); |
| 129 | list_for_each(_p, &trans->peer_active) { |
| 130 | peer = list_entry(_p, struct rxrpc_peer, link); |
| 131 | if (peer->addr.s_addr == addr) |
| 132 | goto found_active_second_chance; |
| 133 | } |
| 134 | |
| 135 | /* search the transport's graveyard list */ |
| 136 | spin_lock(&trans->peer_gylock); |
| 137 | list_for_each(_p, &trans->peer_graveyard) { |
| 138 | peer = list_entry(_p, struct rxrpc_peer, link); |
| 139 | if (peer->addr.s_addr == addr) |
| 140 | goto found_in_graveyard; |
| 141 | } |
| 142 | spin_unlock(&trans->peer_gylock); |
| 143 | |
| 144 | /* we can now add the new candidate to the list |
| 145 | * - tell the application layer that this peer has been added |
| 146 | */ |
| 147 | rxrpc_get_transport(trans); |
| 148 | peer = candidate; |
| 149 | candidate = NULL; |
| 150 | |
| 151 | if (peer->ops && peer->ops->adding) { |
| 152 | ret = peer->ops->adding(peer); |
| 153 | if (ret < 0) { |
| 154 | write_unlock(&trans->peer_lock); |
| 155 | __RXACCT(atomic_dec(&rxrpc_peer_count)); |
| 156 | kfree(peer); |
| 157 | rxrpc_put_transport(trans); |
| 158 | _leave(" = %d", ret); |
| 159 | return ret; |
| 160 | } |
| 161 | } |
| 162 | |
| 163 | atomic_inc(&trans->peer_count); |
| 164 | |
| 165 | make_active: |
| 166 | list_add_tail(&peer->link, &trans->peer_active); |
| 167 | |
| 168 | success_uwfree: |
| 169 | write_unlock(&trans->peer_lock); |
| 170 | |
| 171 | if (candidate) { |
| 172 | __RXACCT(atomic_dec(&rxrpc_peer_count)); |
| 173 | kfree(candidate); |
| 174 | } |
| 175 | |
| 176 | if (list_empty(&peer->proc_link)) { |
| 177 | down_write(&rxrpc_peers_sem); |
| 178 | list_add_tail(&peer->proc_link, &rxrpc_peers); |
| 179 | up_write(&rxrpc_peers_sem); |
| 180 | } |
| 181 | |
| 182 | success: |
| 183 | *_peer = peer; |
| 184 | |
| 185 | _leave(" = 0 (%p{u=%d cc=%d})", |
| 186 | peer, |
| 187 | atomic_read(&peer->usage), |
| 188 | atomic_read(&peer->conn_count)); |
| 189 | return 0; |
| 190 | |
| 191 | /* handle the peer being found in the active list straight off */ |
| 192 | found_active: |
| 193 | rxrpc_get_peer(peer); |
| 194 | read_unlock(&trans->peer_lock); |
| 195 | goto success; |
| 196 | |
| 197 | /* handle resurrecting a peer from the graveyard */ |
| 198 | found_in_graveyard: |
| 199 | rxrpc_get_peer(peer); |
| 200 | rxrpc_get_transport(peer->trans); |
| 201 | rxrpc_krxtimod_del_timer(&peer->timeout); |
| 202 | list_del_init(&peer->link); |
| 203 | spin_unlock(&trans->peer_gylock); |
| 204 | goto make_active; |
| 205 | |
| 206 | /* handle finding the peer on the second time through the active |
| 207 | * list */ |
| 208 | found_active_second_chance: |
| 209 | rxrpc_get_peer(peer); |
| 210 | goto success_uwfree; |
| 211 | |
| 212 | } /* end rxrpc_peer_lookup() */ |
| 213 | |
| 214 | /*****************************************************************************/ |
| 215 | /* |
| 216 | * finish with a peer record |
| 217 | * - it gets sent to the graveyard from where it can be resurrected or timed |
| 218 | * out |
| 219 | */ |
| 220 | void rxrpc_put_peer(struct rxrpc_peer *peer) |
| 221 | { |
| 222 | struct rxrpc_transport *trans = peer->trans; |
| 223 | |
| 224 | _enter("%p{cc=%d a=%08x}", |
| 225 | peer, |
| 226 | atomic_read(&peer->conn_count), |
| 227 | ntohl(peer->addr.s_addr)); |
| 228 | |
| 229 | /* sanity check */ |
| 230 | if (atomic_read(&peer->usage) <= 0) |
| 231 | BUG(); |
| 232 | |
| 233 | write_lock(&trans->peer_lock); |
| 234 | spin_lock(&trans->peer_gylock); |
| 235 | if (likely(!atomic_dec_and_test(&peer->usage))) { |
| 236 | spin_unlock(&trans->peer_gylock); |
| 237 | write_unlock(&trans->peer_lock); |
| 238 | _leave(""); |
| 239 | return; |
| 240 | } |
| 241 | |
| 242 | /* move to graveyard queue */ |
| 243 | list_del(&peer->link); |
| 244 | write_unlock(&trans->peer_lock); |
| 245 | |
| 246 | list_add_tail(&peer->link, &trans->peer_graveyard); |
| 247 | |
| 248 | BUG_ON(!list_empty(&peer->conn_active)); |
| 249 | |
| 250 | rxrpc_krxtimod_add_timer(&peer->timeout, rxrpc_peer_timeout * HZ); |
| 251 | |
| 252 | spin_unlock(&trans->peer_gylock); |
| 253 | |
| 254 | rxrpc_put_transport(trans); |
| 255 | |
| 256 | _leave(" [killed]"); |
| 257 | } /* end rxrpc_put_peer() */ |
| 258 | |
| 259 | /*****************************************************************************/ |
| 260 | /* |
| 261 | * handle a peer timing out in the graveyard |
| 262 | * - called from krxtimod |
| 263 | */ |
| 264 | static void rxrpc_peer_do_timeout(struct rxrpc_peer *peer) |
| 265 | { |
| 266 | struct rxrpc_transport *trans = peer->trans; |
| 267 | |
| 268 | _enter("%p{u=%d cc=%d a=%08x}", |
| 269 | peer, |
| 270 | atomic_read(&peer->usage), |
| 271 | atomic_read(&peer->conn_count), |
| 272 | ntohl(peer->addr.s_addr)); |
| 273 | |
| 274 | BUG_ON(atomic_read(&peer->usage) < 0); |
| 275 | |
| 276 | /* remove from graveyard if still dead */ |
| 277 | spin_lock(&trans->peer_gylock); |
| 278 | if (atomic_read(&peer->usage) == 0) |
| 279 | list_del_init(&peer->link); |
| 280 | else |
| 281 | peer = NULL; |
| 282 | spin_unlock(&trans->peer_gylock); |
| 283 | |
| 284 | if (!peer) { |
| 285 | _leave(""); |
| 286 | return; /* resurrected */ |
| 287 | } |
| 288 | |
| 289 | /* clear all connections on this peer */ |
| 290 | rxrpc_conn_clearall(peer); |
| 291 | |
| 292 | BUG_ON(!list_empty(&peer->conn_active)); |
| 293 | BUG_ON(!list_empty(&peer->conn_graveyard)); |
| 294 | |
| 295 | /* inform the application layer */ |
| 296 | if (peer->ops && peer->ops->discarding) |
| 297 | peer->ops->discarding(peer); |
| 298 | |
| 299 | if (!list_empty(&peer->proc_link)) { |
| 300 | down_write(&rxrpc_peers_sem); |
| 301 | list_del(&peer->proc_link); |
| 302 | up_write(&rxrpc_peers_sem); |
| 303 | } |
| 304 | |
| 305 | __RXACCT(atomic_dec(&rxrpc_peer_count)); |
| 306 | kfree(peer); |
| 307 | |
| 308 | /* if the graveyard is now empty, wake up anyone waiting for that */ |
| 309 | if (atomic_dec_and_test(&trans->peer_count)) |
| 310 | wake_up(&trans->peer_gy_waitq); |
| 311 | |
| 312 | _leave(" [destroyed]"); |
| 313 | } /* end rxrpc_peer_do_timeout() */ |
| 314 | |
| 315 | /*****************************************************************************/ |
| 316 | /* |
| 317 | * clear all peer records from a transport endpoint |
| 318 | */ |
| 319 | void rxrpc_peer_clearall(struct rxrpc_transport *trans) |
| 320 | { |
| 321 | DECLARE_WAITQUEUE(myself,current); |
| 322 | |
| 323 | struct rxrpc_peer *peer; |
| 324 | int err; |
| 325 | |
| 326 | _enter("%p",trans); |
| 327 | |
| 328 | /* there shouldn't be any active peers remaining */ |
| 329 | BUG_ON(!list_empty(&trans->peer_active)); |
| 330 | |
| 331 | /* manually timeout all peers in the graveyard */ |
| 332 | spin_lock(&trans->peer_gylock); |
| 333 | while (!list_empty(&trans->peer_graveyard)) { |
| 334 | peer = list_entry(trans->peer_graveyard.next, |
| 335 | struct rxrpc_peer, link); |
| 336 | _debug("Clearing peer %p\n", peer); |
| 337 | err = rxrpc_krxtimod_del_timer(&peer->timeout); |
| 338 | spin_unlock(&trans->peer_gylock); |
| 339 | |
| 340 | if (err == 0) |
| 341 | rxrpc_peer_do_timeout(peer); |
| 342 | |
| 343 | spin_lock(&trans->peer_gylock); |
| 344 | } |
| 345 | spin_unlock(&trans->peer_gylock); |
| 346 | |
| 347 | /* wait for the the peer graveyard to be completely cleared */ |
| 348 | set_current_state(TASK_UNINTERRUPTIBLE); |
| 349 | add_wait_queue(&trans->peer_gy_waitq, &myself); |
| 350 | |
| 351 | while (atomic_read(&trans->peer_count) != 0) { |
| 352 | schedule(); |
| 353 | set_current_state(TASK_UNINTERRUPTIBLE); |
| 354 | } |
| 355 | |
| 356 | remove_wait_queue(&trans->peer_gy_waitq, &myself); |
| 357 | set_current_state(TASK_RUNNING); |
| 358 | |
| 359 | _leave(""); |
| 360 | } /* end rxrpc_peer_clearall() */ |
| 361 | |
| 362 | /*****************************************************************************/ |
| 363 | /* |
| 364 | * calculate and cache the Round-Trip-Time for a message and its response |
| 365 | */ |
| 366 | void rxrpc_peer_calculate_rtt(struct rxrpc_peer *peer, |
| 367 | struct rxrpc_message *msg, |
| 368 | struct rxrpc_message *resp) |
| 369 | { |
| 370 | unsigned long long rtt; |
| 371 | int loop; |
| 372 | |
| 373 | _enter("%p,%p,%p", peer, msg, resp); |
| 374 | |
| 375 | /* calculate the latest RTT */ |
| 376 | rtt = resp->stamp.tv_sec - msg->stamp.tv_sec; |
| 377 | rtt *= 1000000UL; |
| 378 | rtt += resp->stamp.tv_usec - msg->stamp.tv_usec; |
| 379 | |
| 380 | /* add to cache */ |
| 381 | peer->rtt_cache[peer->rtt_point] = rtt; |
| 382 | peer->rtt_point++; |
| 383 | peer->rtt_point %= RXRPC_RTT_CACHE_SIZE; |
| 384 | |
| 385 | if (peer->rtt_usage < RXRPC_RTT_CACHE_SIZE) |
| 386 | peer->rtt_usage++; |
| 387 | |
| 388 | /* recalculate RTT */ |
| 389 | rtt = 0; |
| 390 | for (loop = peer->rtt_usage - 1; loop >= 0; loop--) |
| 391 | rtt += peer->rtt_cache[loop]; |
| 392 | |
| 393 | do_div(rtt, peer->rtt_usage); |
| 394 | peer->rtt = rtt; |
| 395 | |
| 396 | _leave(" RTT=%lu.%lums", |
| 397 | (long) (peer->rtt / 1000), (long) (peer->rtt % 1000)); |
| 398 | |
| 399 | } /* end rxrpc_peer_calculate_rtt() */ |