lustre/ptlrpc/client.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 #define DEBUG_SUBSYSTEM S_RPC
  26
  27 #include <linux/lustre_ha.h>
  28
  29 void ptlrpc_init_client(struct recovd_obd *recovd,
  30                         void (*recover)(struct ptlrpc_client *recover),
  31                         int req_portal,
  32                         int rep_portal, struct ptlrpc_client *cl)
  33 {
  34         memset(cl, 0, sizeof(*cl));
  35         cl->cli_recovd = recovd;
  36         cl->cli_recover = recover;
  37         if (recovd)
  38                 recovd_cli_manage(recovd, cl);
  39         cl->cli_obd = NULL;
  40         cl->cli_request_portal = req_portal;
  41         cl->cli_reply_portal = rep_portal;
  42         INIT_LIST_HEAD(&cl->cli_sending_head);
  43         INIT_LIST_HEAD(&cl->cli_sent_head);
  44         INIT_LIST_HEAD(&cl->cli_replied_head);
  45         INIT_LIST_HEAD(&cl->cli_replay_head);
  46         INIT_LIST_HEAD(&cl->cli_dying_head);
  47         spin_lock_init(&cl->cli_lock);
  48         sema_init(&cl->cli_rpc_sem, 32);
  49 }
  50
  51 __u8 *ptlrpc_req_to_uuid(struct ptlrpc_request *req)
  52 {
  53         return req->rq_connection->c_remote_uuid;
  54 }
  55
  56 struct ptlrpc_connection *ptlrpc_uuid_to_connection(char *uuid)
  57 {
  58         struct ptlrpc_connection *c;
  59         struct lustre_peer peer;
  60         int err;
  61
  62         err = kportal_uuid_to_peer(uuid, &peer);
  63         if (err != 0) {
  64                 CERROR("cannot find peer %s!\n", uuid);
  65                 return NULL;
  66         }
  67
  68         c = ptlrpc_get_connection(&peer);
  69         if (c)
  70                 c->c_epoch++;
  71
  72         return c;
  73 }
  74
  75 struct ptlrpc_bulk_desc *ptlrpc_prep_bulk(struct ptlrpc_connection *conn)
  76 {
  77         struct ptlrpc_bulk_desc *bulk;
  78
  79         OBD_ALLOC(bulk, sizeof(*bulk));
  80         if (bulk != NULL) {
  81                 bulk->b_connection = ptlrpc_connection_addref(conn);
  82                 init_waitqueue_head(&bulk->b_waitq);
  83         }
  84
  85         return bulk;
  86 }
  87
  88 void ptlrpc_free_bulk(struct ptlrpc_bulk_desc *bulk)
  89 {
  90         ENTRY;
  91         if (bulk == NULL) {
  92                 EXIT;
  93                 return;
  94         }
  95
  96         ptlrpc_put_connection(bulk->b_connection);
  97
  98         OBD_FREE(bulk, sizeof(*bulk));
  99         EXIT;
 100 }
 101
 102 struct ptlrpc_request *ptlrpc_prep_req(struct ptlrpc_client *cl,
 103                                        struct ptlrpc_connection *conn,
 104                                        int opcode, int count, int *lengths,
 105                                        char **bufs)
 106 {
 107         struct ptlrpc_request *request;
 108         int rc;
 109         ENTRY;
 110
 111         OBD_ALLOC(request, sizeof(*request));
 112         if (!request) {
 113                 CERROR("request allocation out of memory\n");
 114                 RETURN(NULL);
 115         }
 116
 117         rc = lustre_pack_msg(count, lengths, bufs,
 118                              &request->rq_reqlen, &request->rq_reqmsg);
 119         if (rc) {
 120                 CERROR("cannot pack request %d\n", rc);
 121                 RETURN(NULL);
 122         }
 123
 124         request->rq_type = PTL_RPC_TYPE_REQUEST;
 125         request->rq_connection = ptlrpc_connection_addref(conn);
 126
 127         request->rq_reqmsg->conn = (__u64)(unsigned long)conn->c_remote_conn;
 128         request->rq_reqmsg->token = conn->c_remote_token;
 129         request->rq_reqmsg->opc = HTON__u32(opcode);
 130         request->rq_reqmsg->type = HTON__u32(PTL_RPC_MSG_REQUEST);
 131         INIT_LIST_HEAD(&request->rq_list);
 132
 133         /* this will be dec()d once in req_finished, once in free_committed */
 134         atomic_set(&request->rq_refcount, 2);
 135
 136         spin_lock(&conn->c_lock);
 137         request->rq_reqmsg->xid = HTON__u32(++conn->c_xid_out);
 138         spin_unlock(&conn->c_lock);
 139
 140         request->rq_client = cl;
 141
 142         RETURN(request);
 143 }
 144
 145 void ptlrpc_req_finished(struct ptlrpc_request *request)
 146 {
 147         if (request == NULL)
 148                 return;
 149
 150         if (request->rq_repmsg != NULL) {
 151                 OBD_FREE(request->rq_repmsg, request->rq_replen);
 152                 request->rq_repmsg = NULL;
 153         }
 154
 155         if (atomic_dec_and_test(&request->rq_refcount))
 156                 ptlrpc_free_req(request);
 157 }
 158
 159 void ptlrpc_free_req(struct ptlrpc_request *request)
 160 {
 161         if (request == NULL)
 162                 return;
 163
 164         if (request->rq_repmsg != NULL)
 165                 OBD_FREE(request->rq_repmsg, request->rq_replen);
 166         if (request->rq_reqmsg != NULL)
 167                 OBD_FREE(request->rq_reqmsg, request->rq_reqlen);
 168
 169         if (request->rq_client) {
 170                 spin_lock(&request->rq_client->cli_lock);
 171                 list_del(&request->rq_list);
 172                 spin_unlock(&request->rq_client->cli_lock);
 173         }
 174
 175         ptlrpc_put_connection(request->rq_connection);
 176
 177         OBD_FREE(request, sizeof(*request));
 178 }
 179
 180 static int ptlrpc_check_reply(struct ptlrpc_request *req)
 181 {
 182         int rc = 0;
 183
 184         if (req->rq_repmsg != NULL) {
 185                 req->rq_transno = NTOH__u64(req->rq_repmsg->transno);
 186                 req->rq_flags |= PTL_RPC_FL_REPLY;
 187                 GOTO(out, rc = 1);
 188         }
 189
 190         if (req->rq_flags & PTL_RPC_FL_RESEND) {
 191                 CERROR("-- RESEND --\n");
 192                 req->rq_status = -EAGAIN;
 193                 GOTO(out, rc = 1);
 194         }
 195
 196         if (CURRENT_TIME - req->rq_time >= req->rq_timeout) {
 197                 CERROR("-- REQ TIMEOUT --\n");
 198                 /* clear the timeout */
 199                 req->rq_timeout = 0;
 200                 req->rq_flags |= PTL_RPC_FL_TIMEOUT;
 201                 if (req->rq_client && req->rq_client->cli_recovd)
 202                         recovd_cli_fail(req->rq_client);
 203                 GOTO(out, rc = 0);
 204         }
 205
 206         if (req->rq_timeout) {
 207                 schedule_timeout(req->rq_timeout * HZ);
 208         }
 209
 210         if (sigismember(&(current->pending.signal), SIGKILL) ||
 211             sigismember(&(current->pending.signal), SIGTERM) ||
 212             sigismember(&(current->pending.signal), SIGINT)) {
 213                 req->rq_flags |= PTL_RPC_FL_INTR;
 214                 GOTO(out, rc = 1);
 215         }
 216
 217  out:
 218         return rc;
 219 }
 220
 221 int ptlrpc_check_status(struct ptlrpc_request *req, int err)
 222 {
 223         ENTRY;
 224
 225         if (err != 0) {
 226                 CERROR("err is %d\n", err);
 227                 RETURN(err);
 228         }
 229
 230         if (req == NULL) {
 231                 CERROR("req == NULL\n");
 232                 RETURN(-ENOMEM);
 233         }
 234
 235         if (req->rq_repmsg == NULL) {
 236                 CERROR("req->rq_repmsg == NULL\n");
 237                 RETURN(-ENOMEM);
 238         }
 239
 240         if (req->rq_repmsg->type == NTOH__u32(PTL_RPC_MSG_ERR)) {
 241                 CERROR("req->rq_repmsg->type == PTL_RPC_MSG_ERR\n");
 242                 RETURN(-EINVAL);
 243         }
 244
 245         if (req->rq_repmsg->status != 0) {
 246                 CERROR("req->rq_repmsg->status is %d\n",
 247                        req->rq_repmsg->status);
 248                 /* XXX: translate this error from net to host */
 249                 RETURN(req->rq_repmsg->status);
 250         }
 251
 252         RETURN(0);
 253 }
 254
 255 static void ptlrpc_cleanup_request_buf(struct ptlrpc_request *request)
 256 {
 257         OBD_FREE(request->rq_reqmsg, request->rq_reqlen);
 258         request->rq_reqmsg = NULL;
 259         request->rq_reqlen = 0;
 260 }
 261
 262 /* Abort this request and cleanup any resources associated with it. */
 263 static int ptlrpc_abort(struct ptlrpc_request *request)
 264 {
 265         /* First remove the ME for the reply; in theory, this means
 266          * that we can tear down the buffer safely. */
 267         PtlMEUnlink(request->rq_reply_me_h);
 268         OBD_FREE(request->rq_reply_md.start, request->rq_replen);
 269         request->rq_repmsg = NULL;
 270         request->rq_replen = 0;
 271         return 0;
 272 }
 273
 274 /* caller must lock cli */
 275 void ptlrpc_free_committed(struct ptlrpc_client *cli)
 276 {
 277         struct list_head *tmp, *saved;
 278         struct ptlrpc_request *req;
 279
 280         list_for_each_safe(tmp, saved, &cli->cli_replied_head) {
 281                 req = list_entry(tmp, struct ptlrpc_request, rq_list);
 282
 283                 /* not yet committed */
 284                 if (req->rq_transno > cli->cli_last_committed)
 285                         break;
 286
 287                 /* retain for replay if flagged */
 288                 list_del(&req->rq_list);
 289                 if (req->rq_flags & PTL_RPC_FL_RETAIN) {
 290                         list_add(&req->rq_list, &cli->cli_replay_head);
 291                 } else {
 292                         CDEBUG(D_INFO, "Marking request %p as committed ("
 293                                "transno=%Lu, last_committed=%Lu\n", req,
 294                                req->rq_transno, cli->cli_last_committed);
 295                         if (atomic_dec_and_test(&req->rq_refcount))
 296                                 ptlrpc_free_req(req);
 297                         else
 298                                 list_add(&req->rq_list, &cli->cli_dying_head);
 299                 }
 300         }
 301
 302         EXIT;
 303         return;
 304 }
 305
 306 void ptlrpc_cleanup_client(struct ptlrpc_client *cli)
 307 {
 308         struct list_head *tmp, *saved;
 309         struct ptlrpc_request *req;
 310         ENTRY;
 311
 312         spin_lock(&cli->cli_lock);
 313         list_for_each_safe(tmp, saved, &cli->cli_replied_head) {
 314                 req = list_entry(tmp, struct ptlrpc_request, rq_list);
 315                 /* We do this to prevent ptlrpc_free_req from taking cli_lock */
 316                 CDEBUG(D_INFO, "Cleaning req %p from replied list.\n", req);
 317                 list_del(&req->rq_list);
 318                 req->rq_client = NULL;
 319                 ptlrpc_free_req(req);
 320         }
 321         list_for_each_safe(tmp, saved, &cli->cli_sent_head) {
 322                 req = list_entry(tmp, struct ptlrpc_request, rq_list);
 323                 CDEBUG(D_INFO, "Cleaning req %p from sent list.\n", req);
 324                 list_del(&req->rq_list);
 325                 req->rq_client = NULL;
 326                 ptlrpc_free_req(req);
 327         }
 328         list_for_each_safe(tmp, saved, &cli->cli_replay_head) {
 329                 req = list_entry(tmp, struct ptlrpc_request, rq_list);
 330                 CERROR("Request %p is on the replay list at cleanup!\n", req);
 331                 list_del(&req->rq_list);
 332                 req->rq_client = NULL;
 333                 ptlrpc_free_req(req);
 334         }
 335         list_for_each_safe(tmp, saved, &cli->cli_sending_head) {
 336                 req = list_entry(tmp, struct ptlrpc_request, rq_list);
 337                 CDEBUG(D_INFO, "Cleaning req %p from sending list.\n", req);
 338                 list_del(&req->rq_list);
 339                 req->rq_client = NULL;
 340                 ptlrpc_free_req(req);
 341         }
 342         list_for_each_safe(tmp, saved, &cli->cli_dying_head) {
 343                 req = list_entry(tmp, struct ptlrpc_request, rq_list);
 344                 CERROR("Request %p is on the dying list at cleanup!\n", req);
 345                 list_del(&req->rq_list);
 346                 req->rq_client = NULL;
 347                 ptlrpc_free_req(req);
 348         }
 349         spin_unlock(&cli->cli_lock);
 350         EXIT;
 351         return;
 352 }
 353
 354 int ptlrpc_queue_wait(struct ptlrpc_request *req)
 355 {
 356         int rc = 0;
 357         ENTRY;
 358
 359         init_waitqueue_head(&req->rq_wait_for_rep);
 360  resend:
 361         req->rq_time = CURRENT_TIME;
 362         req->rq_timeout = 3;
 363         rc = ptl_send_rpc(req);
 364         if (rc) {
 365                 CERROR("error %d, opcode %d\n", rc, req->rq_reqmsg->opc);
 366                 ptlrpc_cleanup_request_buf(req);
 367                 up(&req->rq_client->cli_rpc_sem);
 368                 RETURN(-rc);
 369         }
 370
 371         CDEBUG(D_OTHER, "-- sleeping\n");
 372         wait_event_interruptible(req->rq_wait_for_rep, ptlrpc_check_reply(req));
 373         CDEBUG(D_OTHER, "-- done\n");
 374
 375         if (req->rq_flags & PTL_RPC_FL_RESEND) {
 376                 req->rq_flags &= ~PTL_RPC_FL_RESEND;
 377                 goto resend;
 378         }
 379
 380         //ptlrpc_cleanup_request_buf(req);
 381         up(&req->rq_client->cli_rpc_sem);
 382         if (req->rq_flags & PTL_RPC_FL_INTR) {
 383                 /* Clean up the dangling reply buffers */
 384                 ptlrpc_abort(req);
 385                 GOTO(out, rc = -EINTR);
 386         }
 387
 388         if (! (req->rq_flags & PTL_RPC_FL_REPLY)) {
 389                 CERROR("Unknown reason for wakeup\n");
 390                 /* XXX Phil - I end up here when I kill obdctl */
 391                 ptlrpc_abort(req);
 392                 GOTO(out, rc = -EINTR);
 393         }
 394
 395         rc = lustre_unpack_msg(req->rq_repmsg, req->rq_replen);
 396         if (rc) {
 397                 CERROR("unpack_rep failed: %d\n", rc);
 398                 GOTO(out, rc);
 399         }
 400         CDEBUG(D_NET, "got rep %d\n", req->rq_repmsg->xid);
 401         if (req->rq_repmsg->status == 0)
 402                 CDEBUG(D_NET, "--> buf %p len %d status %d\n", req->rq_repmsg,
 403                        req->rq_replen, req->rq_repmsg->status);
 404
 405         spin_lock(&req->rq_client->cli_lock);
 406         /* add to the tail of the replied head */
 407         list_del(&req->rq_list);
 408         list_add(&req->rq_list, req->rq_client->cli_replied_head.prev);
 409
 410         req->rq_client->cli_last_rcvd = req->rq_repmsg->last_rcvd;
 411         req->rq_client->cli_last_committed = req->rq_repmsg->last_committed;
 412         ptlrpc_free_committed(req->rq_client);
 413         spin_unlock(&req->rq_client->cli_lock);
 414
 415         EXIT;
 416  out:
 417         return rc;
 418 }