linux/net/rds/connection.c
<<
>>
Prefs
   1/*
   2 * Copyright (c) 2006 Oracle.  All rights reserved.
   3 *
   4 * This software is available to you under a choice of one of two
   5 * licenses.  You may choose to be licensed under the terms of the GNU
   6 * General Public License (GPL) Version 2, available from the file
   7 * COPYING in the main directory of this source tree, or the
   8 * OpenIB.org BSD license below:
   9 *
  10 *     Redistribution and use in source and binary forms, with or
  11 *     without modification, are permitted provided that the following
  12 *     conditions are met:
  13 *
  14 *      - Redistributions of source code must retain the above
  15 *        copyright notice, this list of conditions and the following
  16 *        disclaimer.
  17 *
  18 *      - Redistributions in binary form must reproduce the above
  19 *        copyright notice, this list of conditions and the following
  20 *        disclaimer in the documentation and/or other materials
  21 *        provided with the distribution.
  22 *
  23 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
  24 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  25 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
  26 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
  27 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
  28 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
  29 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  30 * SOFTWARE.
  31 *
  32 */
  33#include <linux/kernel.h>
  34#include <linux/list.h>
  35#include <net/inet_hashtables.h>
  36
  37#include "rds.h"
  38#include "loop.h"
  39#include "rdma.h"
  40
  41#define RDS_CONNECTION_HASH_BITS 12
  42#define RDS_CONNECTION_HASH_ENTRIES (1 << RDS_CONNECTION_HASH_BITS)
  43#define RDS_CONNECTION_HASH_MASK (RDS_CONNECTION_HASH_ENTRIES - 1)
  44
  45/* converting this to RCU is a chore for another day.. */
  46static DEFINE_SPINLOCK(rds_conn_lock);
  47static unsigned long rds_conn_count;
  48static struct hlist_head rds_conn_hash[RDS_CONNECTION_HASH_ENTRIES];
  49static struct kmem_cache *rds_conn_slab;
  50
  51static struct hlist_head *rds_conn_bucket(__be32 laddr, __be32 faddr)
  52{
  53        /* Pass NULL, don't need struct net for hash */
  54        unsigned long hash = inet_ehashfn(NULL,
  55                                          be32_to_cpu(laddr), 0,
  56                                          be32_to_cpu(faddr), 0);
  57        return &rds_conn_hash[hash & RDS_CONNECTION_HASH_MASK];
  58}
  59
  60#define rds_conn_info_set(var, test, suffix) do {               \
  61        if (test)                                               \
  62                var |= RDS_INFO_CONNECTION_FLAG_##suffix;       \
  63} while (0)
  64
  65static inline int rds_conn_is_sending(struct rds_connection *conn)
  66{
  67        int ret = 0;
  68
  69        if (!mutex_trylock(&conn->c_send_lock))
  70                ret = 1;
  71        else
  72                mutex_unlock(&conn->c_send_lock);
  73
  74        return ret;
  75}
  76
  77static struct rds_connection *rds_conn_lookup(struct hlist_head *head,
  78                                              __be32 laddr, __be32 faddr,
  79                                              struct rds_transport *trans)
  80{
  81        struct rds_connection *conn, *ret = NULL;
  82        struct hlist_node *pos;
  83
  84        hlist_for_each_entry(conn, pos, head, c_hash_node) {
  85                if (conn->c_faddr == faddr && conn->c_laddr == laddr &&
  86                                conn->c_trans == trans) {
  87                        ret = conn;
  88                        break;
  89                }
  90        }
  91        rdsdebug("returning conn %p for %pI4 -> %pI4\n", ret,
  92                 &laddr, &faddr);
  93        return ret;
  94}
  95
  96/*
  97 * This is called by transports as they're bringing down a connection.
  98 * It clears partial message state so that the transport can start sending
  99 * and receiving over this connection again in the future.  It is up to
 100 * the transport to have serialized this call with its send and recv.
 101 */
 102void rds_conn_reset(struct rds_connection *conn)
 103{
 104        rdsdebug("connection %pI4 to %pI4 reset\n",
 105          &conn->c_laddr, &conn->c_faddr);
 106
 107        rds_stats_inc(s_conn_reset);
 108        rds_send_reset(conn);
 109        conn->c_flags = 0;
 110
 111        /* Do not clear next_rx_seq here, else we cannot distinguish
 112         * retransmitted packets from new packets, and will hand all
 113         * of them to the application. That is not consistent with the
 114         * reliability guarantees of RDS. */
 115}
 116
 117/*
 118 * There is only every one 'conn' for a given pair of addresses in the
 119 * system at a time.  They contain messages to be retransmitted and so
 120 * span the lifetime of the actual underlying transport connections.
 121 *
 122 * For now they are not garbage collected once they're created.  They
 123 * are torn down as the module is removed, if ever.
 124 */
 125static struct rds_connection *__rds_conn_create(__be32 laddr, __be32 faddr,
 126                                       struct rds_transport *trans, gfp_t gfp,
 127                                       int is_outgoing)
 128{
 129        struct rds_connection *conn, *tmp, *parent = NULL;
 130        struct hlist_head *head = rds_conn_bucket(laddr, faddr);
 131        unsigned long flags;
 132        int ret;
 133
 134        spin_lock_irqsave(&rds_conn_lock, flags);
 135        conn = rds_conn_lookup(head, laddr, faddr, trans);
 136        if (conn
 137         && conn->c_loopback
 138         && conn->c_trans != &rds_loop_transport
 139         && !is_outgoing) {
 140                /* This is a looped back IB connection, and we're
 141                 * called by the code handling the incoming connect.
 142                 * We need a second connection object into which we
 143                 * can stick the other QP. */
 144                parent = conn;
 145                conn = parent->c_passive;
 146        }
 147        spin_unlock_irqrestore(&rds_conn_lock, flags);
 148        if (conn)
 149                goto out;
 150
 151        conn = kmem_cache_alloc(rds_conn_slab, gfp);
 152        if (conn == NULL) {
 153                conn = ERR_PTR(-ENOMEM);
 154                goto out;
 155        }
 156
 157        memset(conn, 0, sizeof(*conn));
 158
 159        INIT_HLIST_NODE(&conn->c_hash_node);
 160        conn->c_version = RDS_PROTOCOL_3_0;
 161        conn->c_laddr = laddr;
 162        conn->c_faddr = faddr;
 163        spin_lock_init(&conn->c_lock);
 164        conn->c_next_tx_seq = 1;
 165
 166        mutex_init(&conn->c_send_lock);
 167        INIT_LIST_HEAD(&conn->c_send_queue);
 168        INIT_LIST_HEAD(&conn->c_retrans);
 169
 170        ret = rds_cong_get_maps(conn);
 171        if (ret) {
 172                kmem_cache_free(rds_conn_slab, conn);
 173                conn = ERR_PTR(ret);
 174                goto out;
 175        }
 176
 177        /*
 178         * This is where a connection becomes loopback.  If *any* RDS sockets
 179         * can bind to the destination address then we'd rather the messages
 180         * flow through loopback rather than either transport.
 181         */
 182        if (rds_trans_get_preferred(faddr)) {
 183                conn->c_loopback = 1;
 184                if (is_outgoing && trans->t_prefer_loopback) {
 185                        /* "outgoing" connection - and the transport
 186                         * says it wants the connection handled by the
 187                         * loopback transport. This is what TCP does.
 188                         */
 189                        trans = &rds_loop_transport;
 190                }
 191        }
 192
 193        conn->c_trans = trans;
 194
 195        ret = trans->conn_alloc(conn, gfp);
 196        if (ret) {
 197                kmem_cache_free(rds_conn_slab, conn);
 198                conn = ERR_PTR(ret);
 199                goto out;
 200        }
 201
 202        atomic_set(&conn->c_state, RDS_CONN_DOWN);
 203        conn->c_reconnect_jiffies = 0;
 204        INIT_DELAYED_WORK(&conn->c_send_w, rds_send_worker);
 205        INIT_DELAYED_WORK(&conn->c_recv_w, rds_recv_worker);
 206        INIT_DELAYED_WORK(&conn->c_conn_w, rds_connect_worker);
 207        INIT_WORK(&conn->c_down_w, rds_shutdown_worker);
 208        mutex_init(&conn->c_cm_lock);
 209        conn->c_flags = 0;
 210
 211        rdsdebug("allocated conn %p for %pI4 -> %pI4 over %s %s\n",
 212          conn, &laddr, &faddr,
 213          trans->t_name ? trans->t_name : "[unknown]",
 214          is_outgoing ? "(outgoing)" : "");
 215
 216        spin_lock_irqsave(&rds_conn_lock, flags);
 217        if (parent == NULL) {
 218                tmp = rds_conn_lookup(head, laddr, faddr, trans);
 219                if (tmp == NULL)
 220                        hlist_add_head(&conn->c_hash_node, head);
 221        } else {
 222                tmp = parent->c_passive;
 223                if (!tmp)
 224                        parent->c_passive = conn;
 225        }
 226
 227        if (tmp) {
 228                trans->conn_free(conn->c_transport_data);
 229                kmem_cache_free(rds_conn_slab, conn);
 230                conn = tmp;
 231        } else {
 232                rds_cong_add_conn(conn);
 233                rds_conn_count++;
 234        }
 235
 236        spin_unlock_irqrestore(&rds_conn_lock, flags);
 237
 238out:
 239        return conn;
 240}
 241
 242struct rds_connection *rds_conn_create(__be32 laddr, __be32 faddr,
 243                                       struct rds_transport *trans, gfp_t gfp)
 244{
 245        return __rds_conn_create(laddr, faddr, trans, gfp, 0);
 246}
 247
 248struct rds_connection *rds_conn_create_outgoing(__be32 laddr, __be32 faddr,
 249                                       struct rds_transport *trans, gfp_t gfp)
 250{
 251        return __rds_conn_create(laddr, faddr, trans, gfp, 1);
 252}
 253
 254void rds_conn_destroy(struct rds_connection *conn)
 255{
 256        struct rds_message *rm, *rtmp;
 257
 258        rdsdebug("freeing conn %p for %pI4 -> "
 259                 "%pI4\n", conn, &conn->c_laddr,
 260                 &conn->c_faddr);
 261
 262        hlist_del_init(&conn->c_hash_node);
 263
 264        /* wait for the rds thread to shut it down */
 265        atomic_set(&conn->c_state, RDS_CONN_ERROR);
 266        cancel_delayed_work(&conn->c_conn_w);
 267        queue_work(rds_wq, &conn->c_down_w);
 268        flush_workqueue(rds_wq);
 269
 270        /* tear down queued messages */
 271        list_for_each_entry_safe(rm, rtmp,
 272                                 &conn->c_send_queue,
 273                                 m_conn_item) {
 274                list_del_init(&rm->m_conn_item);
 275                BUG_ON(!list_empty(&rm->m_sock_item));
 276                rds_message_put(rm);
 277        }
 278        if (conn->c_xmit_rm)
 279                rds_message_put(conn->c_xmit_rm);
 280
 281        conn->c_trans->conn_free(conn->c_transport_data);
 282
 283        /*
 284         * The congestion maps aren't freed up here.  They're
 285         * freed by rds_cong_exit() after all the connections
 286         * have been freed.
 287         */
 288        rds_cong_remove_conn(conn);
 289
 290        BUG_ON(!list_empty(&conn->c_retrans));
 291        kmem_cache_free(rds_conn_slab, conn);
 292
 293        rds_conn_count--;
 294}
 295
 296static void rds_conn_message_info(struct socket *sock, unsigned int len,
 297                                  struct rds_info_iterator *iter,
 298                                  struct rds_info_lengths *lens,
 299                                  int want_send)
 300{
 301        struct hlist_head *head;
 302        struct hlist_node *pos;
 303        struct list_head *list;
 304        struct rds_connection *conn;
 305        struct rds_message *rm;
 306        unsigned long flags;
 307        unsigned int total = 0;
 308        size_t i;
 309
 310        len /= sizeof(struct rds_info_message);
 311
 312        spin_lock_irqsave(&rds_conn_lock, flags);
 313
 314        for (i = 0, head = rds_conn_hash; i < ARRAY_SIZE(rds_conn_hash);
 315             i++, head++) {
 316                hlist_for_each_entry(conn, pos, head, c_hash_node) {
 317                        if (want_send)
 318                                list = &conn->c_send_queue;
 319                        else
 320                                list = &conn->c_retrans;
 321
 322                        spin_lock(&conn->c_lock);
 323
 324                        /* XXX too lazy to maintain counts.. */
 325                        list_for_each_entry(rm, list, m_conn_item) {
 326                                total++;
 327                                if (total <= len)
 328                                        rds_inc_info_copy(&rm->m_inc, iter,
 329                                                          conn->c_laddr,
 330                                                          conn->c_faddr, 0);
 331                        }
 332
 333                        spin_unlock(&conn->c_lock);
 334                }
 335        }
 336
 337        spin_unlock_irqrestore(&rds_conn_lock, flags);
 338
 339        lens->nr = total;
 340        lens->each = sizeof(struct rds_info_message);
 341}
 342
 343static void rds_conn_message_info_send(struct socket *sock, unsigned int len,
 344                                       struct rds_info_iterator *iter,
 345                                       struct rds_info_lengths *lens)
 346{
 347        rds_conn_message_info(sock, len, iter, lens, 1);
 348}
 349
 350static void rds_conn_message_info_retrans(struct socket *sock,
 351                                          unsigned int len,
 352                                          struct rds_info_iterator *iter,
 353                                          struct rds_info_lengths *lens)
 354{
 355        rds_conn_message_info(sock, len, iter, lens, 0);
 356}
 357
 358void rds_for_each_conn_info(struct socket *sock, unsigned int len,
 359                          struct rds_info_iterator *iter,
 360                          struct rds_info_lengths *lens,
 361                          int (*visitor)(struct rds_connection *, void *),
 362                          size_t item_len)
 363{
 364        uint64_t buffer[(item_len + 7) / 8];
 365        struct hlist_head *head;
 366        struct hlist_node *pos;
 367        struct hlist_node *tmp;
 368        struct rds_connection *conn;
 369        unsigned long flags;
 370        size_t i;
 371
 372        spin_lock_irqsave(&rds_conn_lock, flags);
 373
 374        lens->nr = 0;
 375        lens->each = item_len;
 376
 377        for (i = 0, head = rds_conn_hash; i < ARRAY_SIZE(rds_conn_hash);
 378             i++, head++) {
 379                hlist_for_each_entry_safe(conn, pos, tmp, head, c_hash_node) {
 380
 381                        /* XXX no c_lock usage.. */
 382                        if (!visitor(conn, buffer))
 383                                continue;
 384
 385                        /* We copy as much as we can fit in the buffer,
 386                         * but we count all items so that the caller
 387                         * can resize the buffer. */
 388                        if (len >= item_len) {
 389                                rds_info_copy(iter, buffer, item_len);
 390                                len -= item_len;
 391                        }
 392                        lens->nr++;
 393                }
 394        }
 395
 396        spin_unlock_irqrestore(&rds_conn_lock, flags);
 397}
 398
 399static int rds_conn_info_visitor(struct rds_connection *conn,
 400                                  void *buffer)
 401{
 402        struct rds_info_connection *cinfo = buffer;
 403
 404        cinfo->next_tx_seq = conn->c_next_tx_seq;
 405        cinfo->next_rx_seq = conn->c_next_rx_seq;
 406        cinfo->laddr = conn->c_laddr;
 407        cinfo->faddr = conn->c_faddr;
 408        strncpy(cinfo->transport, conn->c_trans->t_name,
 409                sizeof(cinfo->transport));
 410        cinfo->flags = 0;
 411
 412        rds_conn_info_set(cinfo->flags,
 413                          rds_conn_is_sending(conn), SENDING);
 414        /* XXX Future: return the state rather than these funky bits */
 415        rds_conn_info_set(cinfo->flags,
 416                          atomic_read(&conn->c_state) == RDS_CONN_CONNECTING,
 417                          CONNECTING);
 418        rds_conn_info_set(cinfo->flags,
 419                          atomic_read(&conn->c_state) == RDS_CONN_UP,
 420                          CONNECTED);
 421        return 1;
 422}
 423
 424static void rds_conn_info(struct socket *sock, unsigned int len,
 425                          struct rds_info_iterator *iter,
 426                          struct rds_info_lengths *lens)
 427{
 428        rds_for_each_conn_info(sock, len, iter, lens,
 429                                rds_conn_info_visitor,
 430                                sizeof(struct rds_info_connection));
 431}
 432
 433int __init rds_conn_init(void)
 434{
 435        rds_conn_slab = kmem_cache_create("rds_connection",
 436                                          sizeof(struct rds_connection),
 437                                          0, 0, NULL);
 438        if (rds_conn_slab == NULL)
 439                return -ENOMEM;
 440
 441        rds_info_register_func(RDS_INFO_CONNECTIONS, rds_conn_info);
 442        rds_info_register_func(RDS_INFO_SEND_MESSAGES,
 443                               rds_conn_message_info_send);
 444        rds_info_register_func(RDS_INFO_RETRANS_MESSAGES,
 445                               rds_conn_message_info_retrans);
 446
 447        return 0;
 448}
 449
 450void rds_conn_exit(void)
 451{
 452        rds_loop_exit();
 453
 454        WARN_ON(!hlist_empty(rds_conn_hash));
 455
 456        kmem_cache_destroy(rds_conn_slab);
 457
 458        rds_info_deregister_func(RDS_INFO_CONNECTIONS, rds_conn_info);
 459        rds_info_deregister_func(RDS_INFO_SEND_MESSAGES,
 460                                 rds_conn_message_info_send);
 461        rds_info_deregister_func(RDS_INFO_RETRANS_MESSAGES,
 462                                 rds_conn_message_info_retrans);
 463}
 464
 465/*
 466 * Force a disconnect
 467 */
 468void rds_conn_drop(struct rds_connection *conn)
 469{
 470        atomic_set(&conn->c_state, RDS_CONN_ERROR);
 471        queue_work(rds_wq, &conn->c_down_w);
 472}
 473
 474/*
 475 * An error occurred on the connection
 476 */
 477void
 478__rds_conn_error(struct rds_connection *conn, const char *fmt, ...)
 479{
 480        va_list ap;
 481
 482        va_start(ap, fmt);
 483        vprintk(fmt, ap);
 484        va_end(ap);
 485
 486        rds_conn_drop(conn);
 487}
 488