net/rds/threads.c

   1 /*
   2  * Copyright (c) 2006, 2018 Oracle and/or its affiliates. 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/random.h>
  35 #include <linux/export.h>
  36
  37 #include "rds.h"
  38
  39 /*
  40  * All of connection management is simplified by serializing it through
  41  * work queues that execute in a connection managing thread.
  42  *
  43  * TCP wants to send acks through sendpage() in response to data_ready(),
  44  * but it needs a process context to do so.
  45  *
  46  * The receive paths need to allocate but can't drop packets (!) so we have
  47  * a thread around to block allocating if the receive fast path sees an
  48  * allocation failure.
  49  */
  50
  51 /* Grand Unified Theory of connection life cycle:
  52  * At any point in time, the connection can be in one of these states:
  53  * DOWN, CONNECTING, UP, DISCONNECTING, ERROR
  54  *
  55  * The following transitions are possible:
  56  *  ANY           -> ERROR
  57  *  UP            -> DISCONNECTING
  58  *  ERROR         -> DISCONNECTING
  59  *  DISCONNECTING -> DOWN
  60  *  DOWN          -> CONNECTING
  61  *  CONNECTING    -> UP
  62  *
  63  * Transition to state DISCONNECTING/DOWN:
  64  *  -   Inside the shutdown worker; synchronizes with xmit path
  65  *      through RDS_IN_XMIT, and with connection management callbacks
  66  *      via c_cm_lock.
  67  *
  68  *      For receive callbacks, we rely on the underlying transport
  69  *      (TCP, IB/RDMA) to provide the necessary synchronisation.
  70  */
  71 struct workqueue_struct *rds_wq;
  72 EXPORT_SYMBOL_GPL(rds_wq);
  73
  74 void rds_connect_path_complete(struct rds_conn_path *cp, int curr)
  75 {
  76         if (!rds_conn_path_transition(cp, curr, RDS_CONN_UP)) {
  77                 printk(KERN_WARNING "%s: Cannot transition to state UP, "
  78                                 "current state is %d\n",
  79                                 __func__,
  80                                 atomic_read(&cp->cp_state));
  81                 rds_conn_path_drop(cp, false);
  82                 return;
  83         }
  84
  85         rdsdebug("conn %p for %pI6c to %pI6c complete\n",
  86                  cp->cp_conn, &cp->cp_conn->c_laddr, &cp->cp_conn->c_faddr);
  87
  88         cp->cp_reconnect_jiffies = 0;
  89         set_bit(0, &cp->cp_conn->c_map_queued);
  90         rcu_read_lock();
  91         if (!rds_destroy_pending(cp->cp_conn)) {
  92                 queue_delayed_work(rds_wq, &cp->cp_send_w, 0);
  93                 queue_delayed_work(rds_wq, &cp->cp_recv_w, 0);
  94         }
  95         rcu_read_unlock();
  96 }
  97 EXPORT_SYMBOL_GPL(rds_connect_path_complete);
  98
  99 void rds_connect_complete(struct rds_connection *conn)
 100 {
 101         rds_connect_path_complete(&conn->c_path[0], RDS_CONN_CONNECTING);
 102 }
 103 EXPORT_SYMBOL_GPL(rds_connect_complete);
 104
 105 /*
 106  * This random exponential backoff is relied on to eventually resolve racing
 107  * connects.
 108  *
 109  * If connect attempts race then both parties drop both connections and come
 110  * here to wait for a random amount of time before trying again.  Eventually
 111  * the backoff range will be so much greater than the time it takes to
 112  * establish a connection that one of the pair will establish the connection
 113  * before the other's random delay fires.
 114  *
 115  * Connection attempts that arrive while a connection is already established
 116  * are also considered to be racing connects.  This lets a connection from
 117  * a rebooted machine replace an existing stale connection before the transport
 118  * notices that the connection has failed.
 119  *
 120  * We should *always* start with a random backoff; otherwise a broken connection
 121  * will always take several iterations to be re-established.
 122  */
 123 void rds_queue_reconnect(struct rds_conn_path *cp)
 124 {
 125         unsigned long rand;
 126         struct rds_connection *conn = cp->cp_conn;
 127
 128         rdsdebug("conn %p for %pI6c to %pI6c reconnect jiffies %lu\n",
 129                  conn, &conn->c_laddr, &conn->c_faddr,
 130                  cp->cp_reconnect_jiffies);
 131
 132         /* let peer with smaller addr initiate reconnect, to avoid duels */
 133         if (conn->c_trans->t_type == RDS_TRANS_TCP &&
 134             rds_addr_cmp(&conn->c_laddr, &conn->c_faddr) >= 0)
 135                 return;
 136
 137         set_bit(RDS_RECONNECT_PENDING, &cp->cp_flags);
 138         if (cp->cp_reconnect_jiffies == 0) {
 139                 cp->cp_reconnect_jiffies = rds_sysctl_reconnect_min_jiffies;
 140                 rcu_read_lock();
 141                 if (!rds_destroy_pending(cp->cp_conn))
 142                         queue_delayed_work(rds_wq, &cp->cp_conn_w, 0);
 143                 rcu_read_unlock();
 144                 return;
 145         }
 146
 147         get_random_bytes(&rand, sizeof(rand));
 148         rdsdebug("%lu delay %lu ceil conn %p for %pI6c -> %pI6c\n",
 149                  rand % cp->cp_reconnect_jiffies, cp->cp_reconnect_jiffies,
 150                  conn, &conn->c_laddr, &conn->c_faddr);
 151         rcu_read_lock();
 152         if (!rds_destroy_pending(cp->cp_conn))
 153                 queue_delayed_work(rds_wq, &cp->cp_conn_w,
 154                                    rand % cp->cp_reconnect_jiffies);
 155         rcu_read_unlock();
 156
 157         cp->cp_reconnect_jiffies = min(cp->cp_reconnect_jiffies * 2,
 158                                         rds_sysctl_reconnect_max_jiffies);
 159 }
 160
 161 void rds_connect_worker(struct work_struct *work)
 162 {
 163         struct rds_conn_path *cp = container_of(work,
 164                                                 struct rds_conn_path,
 165                                                 cp_conn_w.work);
 166         struct rds_connection *conn = cp->cp_conn;
 167         int ret;
 168
 169         if (cp->cp_index > 0 &&
 170             rds_addr_cmp(&cp->cp_conn->c_laddr, &cp->cp_conn->c_faddr) >= 0)
 171                 return;
 172         clear_bit(RDS_RECONNECT_PENDING, &cp->cp_flags);
 173         ret = rds_conn_path_transition(cp, RDS_CONN_DOWN, RDS_CONN_CONNECTING);
 174         if (ret) {
 175                 ret = conn->c_trans->conn_path_connect(cp);
 176                 rdsdebug("conn %p for %pI6c to %pI6c dispatched, ret %d\n",
 177                          conn, &conn->c_laddr, &conn->c_faddr, ret);
 178
 179                 if (ret) {
 180                         if (rds_conn_path_transition(cp,
 181                                                      RDS_CONN_CONNECTING,
 182                                                      RDS_CONN_DOWN))
 183                                 rds_queue_reconnect(cp);
 184                         else
 185                                 rds_conn_path_error(cp, "connect failed\n");
 186                 }
 187         }
 188 }
 189
 190 void rds_send_worker(struct work_struct *work)
 191 {
 192         struct rds_conn_path *cp = container_of(work,
 193                                                 struct rds_conn_path,
 194                                                 cp_send_w.work);
 195         int ret;
 196
 197         if (rds_conn_path_state(cp) == RDS_CONN_UP) {
 198                 clear_bit(RDS_LL_SEND_FULL, &cp->cp_flags);
 199                 ret = rds_send_xmit(cp);
 200                 cond_resched();
 201                 rdsdebug("conn %p ret %d\n", cp->cp_conn, ret);
 202                 switch (ret) {
 203                 case -EAGAIN:
 204                         rds_stats_inc(s_send_immediate_retry);
 205                         queue_delayed_work(rds_wq, &cp->cp_send_w, 0);
 206                         break;
 207                 case -ENOMEM:
 208                         rds_stats_inc(s_send_delayed_retry);
 209                         queue_delayed_work(rds_wq, &cp->cp_send_w, 2);
 210                 default:
 211                         break;
 212                 }
 213         }
 214 }
 215
 216 void rds_recv_worker(struct work_struct *work)
 217 {
 218         struct rds_conn_path *cp = container_of(work,
 219                                                 struct rds_conn_path,
 220                                                 cp_recv_w.work);
 221         int ret;
 222
 223         if (rds_conn_path_state(cp) == RDS_CONN_UP) {
 224                 ret = cp->cp_conn->c_trans->recv_path(cp);
 225                 rdsdebug("conn %p ret %d\n", cp->cp_conn, ret);
 226                 switch (ret) {
 227                 case -EAGAIN:
 228                         rds_stats_inc(s_recv_immediate_retry);
 229                         queue_delayed_work(rds_wq, &cp->cp_recv_w, 0);
 230                         break;
 231                 case -ENOMEM:
 232                         rds_stats_inc(s_recv_delayed_retry);
 233                         queue_delayed_work(rds_wq, &cp->cp_recv_w, 2);
 234                 default:
 235                         break;
 236                 }
 237         }
 238 }
 239
 240 void rds_shutdown_worker(struct work_struct *work)
 241 {
 242         struct rds_conn_path *cp = container_of(work,
 243                                                 struct rds_conn_path,
 244                                                 cp_down_w);
 245
 246         rds_conn_shutdown(cp);
 247 }
 248
 249 void rds_threads_exit(void)
 250 {
 251         destroy_workqueue(rds_wq);
 252 }
 253
 254 int rds_threads_init(void)
 255 {
 256         rds_wq = create_singlethread_workqueue("krdsd");
 257         if (!rds_wq)
 258                 return -ENOMEM;
 259
 260         return 0;
 261 }
 262
 263 /* Compare two IPv6 addresses.  Return 0 if the two addresses are equal.
 264  * Return 1 if the first is greater.  Return -1 if the second is greater.
 265  */
 266 int rds_addr_cmp(const struct in6_addr *addr1,
 267                  const struct in6_addr *addr2)
 268 {
 269 #if defined(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) && BITS_PER_LONG == 64
 270         const __be64 *a1, *a2;
 271         u64 x, y;
 272
 273         a1 = (__be64 *)addr1;
 274         a2 = (__be64 *)addr2;
 275
 276         if (*a1 != *a2) {
 277                 if (be64_to_cpu(*a1) < be64_to_cpu(*a2))
 278                         return -1;
 279                 else
 280                         return 1;
 281         } else {
 282                 x = be64_to_cpu(*++a1);
 283                 y = be64_to_cpu(*++a2);
 284                 if (x < y)
 285                         return -1;
 286                 else if (x > y)
 287                         return 1;
 288                 else
 289                         return 0;
 290         }
 291 #else
 292         u32 a, b;
 293         int i;
 294
 295         for (i = 0; i < 4; i++) {
 296                 if (addr1->s6_addr32[i] != addr2->s6_addr32[i]) {
 297                         a = ntohl(addr1->s6_addr32[i]);
 298                         b = ntohl(addr2->s6_addr32[i]);
 299                         if (a < b)
 300                                 return -1;
 301                         else if (a > b)
 302                                 return 1;
 303                 }
 304         }
 305         return 0;
 306 #endif
 307 }
 308 EXPORT_SYMBOL_GPL(rds_addr_cmp);