lustre/ptlrpc/service.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  *  Copyright (C) 2002 Cluster File Systems, Inc.
   5  *
   6  *   This file is part of Lustre, http://www.lustre.org.
   7  *
   8  *   Lustre is free software; you can redistribute it and/or
   9  *   modify it under the terms of version 2 of the GNU General Public
  10  *   License as published by the Free Software Foundation.
  11  *
  12  *   Lustre is distributed in the hope that it will be useful,
  13  *   but WITHOUT ANY WARRANTY; without even the implied warranty of
  14  *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  15  *   GNU General Public License for more details.
  16  *
  17  *   You should have received a copy of the GNU General Public License
  18  *   along with Lustre; if not, write to the Free Software
  19  *   Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
  20  *
  21  */
  22
  23 #define EXPORT_SYMTAB
  24
  25 #include <linux/config.h>
  26 #include <linux/module.h>
  27 #include <linux/kernel.h>
  28
  29 #define DEBUG_SUBSYSTEM S_RPC
  30
  31 #include <linux/obd_support.h>
  32 #include <linux/obd_class.h>
  33 #include <linux/lustre_net.h>
  34
  35 extern int request_in_callback(ptl_event_t *ev, void *data);
  36 extern int ptl_handled_rpc(struct ptlrpc_service *service, void *start);
  37
  38 static int ptlrpc_check_event(struct ptlrpc_service *svc)
  39 {
  40         int rc = 0;
  41
  42         spin_lock(&svc->srv_lock);
  43         if (sigismember(&(current->pending.signal), SIGKILL) ||
  44             sigismember(&(current->pending.signal), SIGTERM) ||
  45             sigismember(&(current->pending.signal), SIGINT)) {
  46                 svc->srv_flags |= SVC_KILLED;
  47                 GOTO(out, rc = 1);
  48         }
  49
  50         if (svc->srv_flags & SVC_STOPPING)
  51                 GOTO(out, rc = 1);
  52
  53         if (svc->srv_flags & SVC_EVENT)
  54                 LBUG();
  55
  56         if (ptl_is_valid_handle(&svc->srv_eq_h)) {
  57                 int err;
  58                 err = PtlEQGet(svc->srv_eq_h, &svc->srv_ev);
  59
  60                 if (err == PTL_OK) {
  61                         svc->srv_flags |= SVC_EVENT;
  62                         GOTO(out, rc = 1);
  63                 }
  64
  65                 if (err != PTL_EQ_EMPTY) {
  66                         CDEBUG(D_NET, "BUG: PtlEQGet returned %d\n", rc);
  67                         LBUG();
  68                 }
  69
  70                 GOTO(out, rc = 0);
  71         }
  72
  73         if (!list_empty(&svc->srv_reqs)) {
  74                 svc->srv_flags |= SVC_LIST;
  75                 GOTO(out, rc = 1);
  76         }
  77
  78         EXIT;
  79  out:
  80         spin_unlock(&svc->srv_lock);
  81         return rc;
  82 }
  83
  84 struct ptlrpc_service *
  85 ptlrpc_init_svc(__u32 bufsize, int req_portal, int rep_portal, char *uuid,
  86                 svc_handler_t handler)
  87 {
  88         int err;
  89         int rc, i;
  90         struct ptlrpc_service *service;
  91
  92         OBD_ALLOC(service, sizeof(*service));
  93         if (!service) {
  94                 LBUG();
  95                 RETURN(NULL);
  96         }
  97
  98         spin_lock_init(&service->srv_lock);
  99         INIT_LIST_HEAD(&service->srv_reqs);
 100         init_waitqueue_head(&service->srv_ctl_waitq);
 101         init_waitqueue_head(&service->srv_waitq);
 102
 103         service->srv_thread = NULL;
 104         service->srv_flags = 0;
 105
 106         service->srv_buf_size = bufsize;
 107         service->srv_rep_portal = rep_portal;
 108         service->srv_req_portal = req_portal;
 109         service->srv_handler = handler;
 110
 111         err = kportal_uuid_to_peer(uuid, &service->srv_self);
 112         if (err) {
 113                 CERROR("cannot get peer for uuid %s", uuid);
 114                 GOTO(err_free, NULL);
 115         }
 116
 117         service->srv_ring_length = RPC_RING_LENGTH;
 118         service->srv_id.nid = PTL_ID_ANY;
 119         service->srv_id.pid = PTL_ID_ANY;
 120
 121         rc = PtlEQAlloc(service->srv_self.peer_ni, 128,
 122                         request_in_callback,
 123                         service, &(service->srv_eq_h));
 124
 125         if (rc != PTL_OK) {
 126                 CERROR("PtlEQAlloc failed: %d\n", rc);
 127                 LBUG();
 128                 GOTO(err_free, NULL);
 129         }
 130
 131         for (i = 0; i < service->srv_ring_length; i++) {
 132                 OBD_ALLOC(service->srv_buf[i], service->srv_buf_size);
 133                 if (service->srv_buf[i] == NULL) {
 134                         CERROR("no memory\n");
 135                         LBUG();
 136                         GOTO(err_ring, NULL);
 137                 }
 138                 service->srv_ref_count[i] = 0;
 139                 ptlrpc_link_svc_me(service, i);
 140         }
 141
 142         CDEBUG(D_NET, "Starting service listening on portal %d\n",
 143                service->srv_req_portal);
 144
 145         return service;
 146 err_ring:
 147         service->srv_ring_length = i;
 148         rpc_unregister_service(service); // XXX verify this is right
 149         PtlEQFree(service->srv_eq_h);
 150 err_free:
 151         OBD_FREE(service, sizeof(*service));
 152         return NULL;
 153 }
 154
 155 static int ptlrpc_main(void *arg)
 156 {
 157         int rc;
 158         struct ptlrpc_svc_data *data = (struct ptlrpc_svc_data *)arg;
 159         struct obd_device *obddev = data->dev;
 160         struct ptlrpc_service *svc = data->svc;
 161
 162         ENTRY;
 163
 164         lock_kernel();
 165         daemonize();
 166         spin_lock_irq(&current->sigmask_lock);
 167         sigfillset(&current->blocked);
 168         recalc_sigpending(current);
 169         spin_unlock_irq(&current->sigmask_lock);
 170
 171         sprintf(current->comm, data->name);
 172
 173         /* Record that the  thread is running */
 174         svc->srv_thread = current;
 175         svc->srv_flags = SVC_RUNNING;
 176         wake_up(&svc->srv_ctl_waitq);
 177
 178         /* XXX maintain a list of all managed devices: insert here */
 179
 180         /* And now, loop forever on requests */
 181         while (1) {
 182                 wait_event(svc->srv_waitq, ptlrpc_check_event(svc));
 183
 184                 spin_lock(&svc->srv_lock);
 185                 if (svc->srv_flags & SVC_SIGNAL) {
 186                         spin_unlock(&svc->srv_lock);
 187                         EXIT;
 188                         break;
 189                 }
 190
 191                 if (svc->srv_flags & SVC_STOPPING) {
 192                         spin_unlock(&svc->srv_lock);
 193                         EXIT;
 194                         break;
 195                 }
 196
 197                 if (svc->srv_flags & SVC_EVENT) {
 198                         struct ptlrpc_request request;
 199                         void *start;
 200                         svc->srv_flags = SVC_RUNNING;
 201
 202                         /* FIXME: If we move to an event-driven model,
 203                          * we should put the request on the stack of
 204                          * mds_handle instead. */
 205                         start = svc->srv_ev.mem_desc.start;
 206                         memset(&request, 0, sizeof(request));
 207                         request.rq_obd = obddev;
 208                         request.rq_reqbuf = (svc->srv_ev.mem_desc.start +
 209                                              svc->srv_ev.offset);
 210                         request.rq_reqlen = svc->srv_ev.mem_desc.length;
 211                         request.rq_xid = svc->srv_ev.match_bits;
 212                         CDEBUG(D_NET, "got req %d\n", request.rq_xid);
 213
 214                         request.rq_peer.peer_nid = svc->srv_ev.initiator.nid;
 215                         /* FIXME: this NI should be the incoming NI.
 216                          * We don't know how to find that from here. */
 217                         request.rq_peer.peer_ni = svc->srv_self.peer_ni;
 218                         svc->srv_flags &= ~SVC_EVENT;
 219
 220                         spin_unlock(&svc->srv_lock);
 221                         rc = svc->srv_handler(obddev, svc, &request);
 222                         ptl_handled_rpc(svc, start);
 223                         continue;
 224                 }
 225
 226                 if (svc->srv_flags & SVC_LIST) {
 227                         struct ptlrpc_request *request;
 228                         svc->srv_flags = SVC_RUNNING;
 229
 230                         request = list_entry(svc->srv_reqs.next,
 231                                              struct ptlrpc_request,
 232                                              rq_list);
 233                         list_del(&request->rq_list);
 234                         spin_unlock(&svc->srv_lock);
 235                         rc = svc->srv_handler(obddev, svc, request);
 236                         continue;
 237                 }
 238                 CERROR("unknown break in service");
 239                 spin_unlock(&svc->srv_lock);
 240                 EXIT;
 241                 break;
 242         }
 243
 244         svc->srv_thread = NULL;
 245         svc->srv_flags = SVC_STOPPED;
 246         wake_up(&svc->srv_ctl_waitq);
 247         CDEBUG(D_NET, "svc exiting process %d\n", current->pid);
 248         return 0;
 249 }
 250
 251 void ptlrpc_stop_thread(struct ptlrpc_service *svc)
 252 {
 253         svc->srv_flags = SVC_STOPPING;
 254
 255         wake_up(&svc->srv_waitq);
 256         wait_event_interruptible(svc->srv_ctl_waitq,
 257                                  (svc->srv_flags & SVC_STOPPED));
 258 }
 259
 260 int ptlrpc_start_thread(struct obd_device *dev, struct ptlrpc_service *svc,
 261                                 char *name)
 262 {
 263         struct ptlrpc_svc_data d;
 264         int rc;
 265         ENTRY;
 266
 267         d.dev = dev;
 268         d.svc = svc;
 269         d.name = name;
 270
 271         init_waitqueue_head(&svc->srv_waitq);
 272
 273         init_waitqueue_head(&svc->srv_ctl_waitq);
 274         rc = kernel_thread(ptlrpc_main, (void *) &d,
 275                            CLONE_VM | CLONE_FS | CLONE_FILES);
 276         if (rc < 0) {
 277                 CERROR("cannot start thread\n");
 278                 RETURN(-EINVAL);
 279         }
 280         wait_event(svc->srv_ctl_waitq, svc->srv_flags & SVC_RUNNING);
 281
 282         RETURN(0);
 283 }
 284
 285 int rpc_unregister_service(struct ptlrpc_service *service)
 286 {
 287         int rc, i;
 288
 289         for (i = 0; i < service->srv_ring_length; i++) {
 290                 if (ptl_is_valid_handle(&(service->srv_me_h[i]))) {
 291                         rc = PtlMEUnlink(service->srv_me_h[i]);
 292                         if (rc)
 293                                 CERROR("PtlMEUnlink failed: %d\n", rc);
 294                         ptl_set_inv_handle(&(service->srv_me_h[i]));
 295                 }
 296
 297                 if (service->srv_buf[i] != NULL)
 298                         OBD_FREE(service->srv_buf[i], service->srv_buf_size);
 299                 service->srv_buf[i] = NULL;
 300         }
 301
 302         rc = PtlEQFree(service->srv_eq_h);
 303         if (rc)
 304                 CERROR("PtlEQFree failed: %d\n", rc);
 305
 306         return 0;
 307 }