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 DEBUG_SUBSYSTEM S_RPC
  24
  25 #include <linux/obd_support.h>
  26 #include <linux/lustre_lib.h>
  27 #include <linux/lustre_ha.h>
  28
  29 void ptlrpc_init_client(struct recovd_obd *recovd,
  30                         int (*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_delayed_head);
  43         INIT_LIST_HEAD(&cl->cli_sending_head);
  44         INIT_LIST_HEAD(&cl->cli_dying_head);
  45         spin_lock_init(&cl->cli_lock);
  46         sema_init(&cl->cli_rpc_sem, 32);
  47 }
  48
  49 __u8 *ptlrpc_req_to_uuid(struct ptlrpc_request *req)
  50 {
  51         return req->rq_connection->c_remote_uuid;
  52 }
  53
  54 struct ptlrpc_connection *ptlrpc_uuid_to_connection(char *uuid)
  55 {
  56         struct ptlrpc_connection *c;
  57         struct lustre_peer peer;
  58         int err;
  59
  60         err = kportal_uuid_to_peer(uuid, &peer);
  61         if (err != 0) {
  62                 CERROR("cannot find peer %s!\n", uuid);
  63                 return NULL;
  64         }
  65
  66         c = ptlrpc_get_connection(&peer);
  67         if (c) {
  68                 memcpy(c->c_remote_uuid, uuid, sizeof(c->c_remote_uuid));
  69                 c->c_epoch++;
  70         }
  71
  72         return c;
  73 }
  74
  75 void ptlrpc_readdress_connection(struct ptlrpc_connection *conn, char *uuid)
  76 {
  77         struct lustre_peer peer;
  78         int err;
  79
  80         err = kportal_uuid_to_peer(uuid, &peer);
  81         if (err != 0) {
  82                 CERROR("cannot find peer %s!\n", uuid);
  83                 return;
  84         }
  85
  86         memcpy(&conn->c_peer, &peer, sizeof(peer));
  87         return;
  88 }
  89
  90 struct ptlrpc_bulk_desc *ptlrpc_prep_bulk(struct ptlrpc_connection *conn)
  91 {
  92         struct ptlrpc_bulk_desc *desc;
  93
  94         OBD_ALLOC(desc, sizeof(*desc));
  95         if (desc != NULL) {
  96                 desc->b_connection = ptlrpc_connection_addref(conn);
  97                 atomic_set(&desc->b_pages_remaining, 0);
  98                 atomic_set(&desc->b_refcount, 1);
  99                 init_waitqueue_head(&desc->b_waitq);
 100                 INIT_LIST_HEAD(&desc->b_page_list);
 101         }
 102
 103         return desc;
 104 }
 105
 106 struct ptlrpc_bulk_page *ptlrpc_prep_bulk_page(struct ptlrpc_bulk_desc *desc)
 107 {
 108         struct ptlrpc_bulk_page *bulk;
 109
 110         OBD_ALLOC(bulk, sizeof(*bulk));
 111         if (bulk != NULL) {
 112                 bulk->b_desc = desc;
 113                 ptl_set_inv_handle(&bulk->b_md_h);
 114                 ptl_set_inv_handle(&bulk->b_me_h);
 115                 list_add_tail(&bulk->b_link, &desc->b_page_list);
 116                 desc->b_page_count++;
 117                 atomic_inc(&desc->b_pages_remaining);
 118         }
 119         return bulk;
 120 }
 121
 122 void ptlrpc_free_bulk(struct ptlrpc_bulk_desc *desc)
 123 {
 124         struct list_head *tmp, *next;
 125         ENTRY;
 126         if (desc == NULL) {
 127                 EXIT;
 128                 return;
 129         }
 130
 131         list_for_each_safe(tmp, next, &desc->b_page_list) {
 132                 struct ptlrpc_bulk_page *bulk;
 133                 bulk = list_entry(tmp, struct ptlrpc_bulk_page, b_link);
 134                 ptlrpc_free_bulk_page(bulk);
 135         }
 136
 137         ptlrpc_put_connection(desc->b_connection);
 138
 139         OBD_FREE(desc, sizeof(*desc));
 140         EXIT;
 141 }
 142
 143 void ptlrpc_free_bulk_page(struct ptlrpc_bulk_page *bulk)
 144 {
 145         ENTRY;
 146         if (bulk == NULL) {
 147                 EXIT;
 148                 return;
 149         }
 150
 151         list_del(&bulk->b_link);
 152         bulk->b_desc->b_page_count--;
 153         OBD_FREE(bulk, sizeof(*bulk));
 154         EXIT;
 155 }
 156
 157 struct ptlrpc_request *ptlrpc_prep_req(struct ptlrpc_client *cl,
 158                                        struct ptlrpc_connection *conn,
 159                                        int opcode, int count, int *lengths,
 160                                        char **bufs)
 161 {
 162         struct ptlrpc_request *request;
 163         int rc;
 164         ENTRY;
 165
 166         OBD_ALLOC(request, sizeof(*request));
 167         if (!request) {
 168                 CERROR("request allocation out of memory\n");
 169                 RETURN(NULL);
 170         }
 171
 172         rc = lustre_pack_msg(count, lengths, bufs,
 173                              &request->rq_reqlen, &request->rq_reqmsg);
 174         if (rc) {
 175                 CERROR("cannot pack request %d\n", rc);
 176                 OBD_FREE(request, sizeof(*request));
 177                 RETURN(NULL);
 178         }
 179
 180         request->rq_level = LUSTRE_CONN_FULL;
 181         request->rq_type = PTL_RPC_TYPE_REQUEST;
 182         request->rq_client = cl;
 183         request->rq_connection = ptlrpc_connection_addref(conn);
 184
 185         INIT_LIST_HEAD(&request->rq_list);
 186         INIT_LIST_HEAD(&request->rq_multi);
 187         /* this will be dec()d once in req_finished, once in free_committed */
 188         atomic_set(&request->rq_refcount, 2);
 189
 190         spin_lock(&conn->c_lock);
 191         request->rq_xid = HTON__u32(++conn->c_xid_out);
 192         spin_unlock(&conn->c_lock);
 193
 194         request->rq_reqmsg->magic = PTLRPC_MSG_MAGIC;
 195         request->rq_reqmsg->version = PTLRPC_MSG_VERSION;
 196         request->rq_reqmsg->opc = HTON__u32(opcode);
 197         request->rq_reqmsg->type = HTON__u32(PTL_RPC_MSG_REQUEST);
 198
 199         RETURN(request);
 200 }
 201 struct ptlrpc_request *ptlrpc_prep_req2(struct ptlrpc_client *cl,
 202                                         struct ptlrpc_connection *conn,
 203                                         struct lustre_handle *handle,
 204                                        int opcode, int count, int *lengths,
 205                                        char **bufs)
 206 {
 207         struct ptlrpc_request *req;
 208         req = ptlrpc_prep_req(cl, conn, opcode, count, lengths, bufs);
 209         ptlrpc_hdl2req(req, handle);
 210         return req;
 211 }
 212
 213 void ptlrpc_req_finished(struct ptlrpc_request *request)
 214 {
 215         if (request == NULL)
 216                 return;
 217
 218         if (request->rq_repmsg != NULL) {
 219                 OBD_FREE(request->rq_repmsg, request->rq_replen);
 220                 request->rq_repmsg = NULL;
 221                 request->rq_reply_md.start = NULL;
 222         }
 223
 224         if (atomic_dec_and_test(&request->rq_refcount))
 225                 ptlrpc_free_req(request);
 226 }
 227
 228 void ptlrpc_free_req(struct ptlrpc_request *request)
 229 {
 230         ENTRY;
 231         if (request == NULL) {
 232                 EXIT;
 233                 return;
 234         }
 235
 236         if (request->rq_repmsg != NULL)
 237                 OBD_FREE(request->rq_repmsg, request->rq_replen);
 238         if (request->rq_reqmsg != NULL)
 239                 OBD_FREE(request->rq_reqmsg, request->rq_reqlen);
 240
 241         if (request->rq_client) {
 242                 spin_lock(&request->rq_client->cli_lock);
 243                 list_del_init(&request->rq_list);
 244                 spin_unlock(&request->rq_client->cli_lock);
 245         }
 246
 247         ptlrpc_put_connection(request->rq_connection);
 248         list_del(&request->rq_multi);
 249         OBD_FREE(request, sizeof(*request));
 250         EXIT;
 251 }
 252
 253 static int ptlrpc_check_reply(struct ptlrpc_request *req)
 254 {
 255         int rc = 0;
 256
 257         if (req->rq_repmsg != NULL) {
 258                 req->rq_transno = NTOH__u64(req->rq_repmsg->transno);
 259                 req->rq_flags |= PTL_RPC_FL_REPLIED;
 260                 GOTO(out, rc = 1);
 261         }
 262
 263         if (req->rq_flags & PTL_RPC_FL_RESEND) {
 264                 CERROR("-- RESEND --\n");
 265                 GOTO(out, rc = 1);
 266         }
 267
 268         if (req->rq_flags & PTL_RPC_FL_RECOVERY) {
 269                 CERROR("-- RESTART --\n");
 270                 GOTO(out, rc = 1);
 271         }
 272
 273
 274         if (CURRENT_TIME - req->rq_time >= req->rq_timeout) {
 275                 CERROR("-- REQ TIMEOUT ON CONNID %d XID %Ld --\n",
 276                        req->rq_connid, (unsigned long long)req->rq_xid);
 277                 /* clear the timeout */
 278                 req->rq_timeout = 0;
 279                 req->rq_connection->c_level = LUSTRE_CONN_RECOVD;
 280                 req->rq_flags |= PTL_RPC_FL_TIMEOUT;
 281                 if (req->rq_client && req->rq_client->cli_recovd)
 282                         recovd_cli_fail(req->rq_client);
 283                 if (req->rq_level < LUSTRE_CONN_FULL)
 284                         rc = 1;
 285                 else
 286                         rc = 0;
 287                 GOTO(out, rc);
 288         }
 289
 290         if (req->rq_timeout) {
 291                 schedule_timeout(req->rq_timeout * HZ);
 292         }
 293
 294         if (l_killable_pending(current)) {
 295                 req->rq_flags |= PTL_RPC_FL_INTR;
 296                 GOTO(out, rc = 1);
 297         }
 298
 299  out:
 300         CDEBUG(D_NET, "req = %p, rc = %d\n", req, rc);
 301         return rc;
 302 }
 303
 304 int ptlrpc_check_status(struct ptlrpc_request *req, int err)
 305 {
 306         ENTRY;
 307
 308         if (err != 0) {
 309                 CERROR("err is %d\n", err);
 310                 RETURN(err);
 311         }
 312
 313         if (req == NULL) {
 314                 CERROR("req == NULL\n");
 315                 RETURN(-ENOMEM);
 316         }
 317
 318         if (req->rq_repmsg == NULL) {
 319                 CERROR("req->rq_repmsg == NULL\n");
 320                 RETURN(-ENOMEM);
 321         }
 322
 323         if (req->rq_repmsg->type == NTOH__u32(PTL_RPC_MSG_ERR)) {
 324                 CERROR("req->rq_repmsg->type == PTL_RPC_MSG_ERR\n");
 325                 RETURN(-EINVAL);
 326         }
 327
 328         if (req->rq_repmsg->status != 0) {
 329                 if (req->rq_repmsg->status < 0)
 330                         CERROR("req->rq_repmsg->status is %d\n",
 331                                req->rq_repmsg->status);
 332                 else
 333                         CDEBUG(D_INFO, "req->rq_repmsg->status is %d\n",
 334                                req->rq_repmsg->status);
 335                 /* XXX: translate this error from net to host */
 336                 RETURN(req->rq_repmsg->status);
 337         }
 338
 339         RETURN(0);
 340 }
 341
 342 static void ptlrpc_cleanup_request_buf(struct ptlrpc_request *request)
 343 {
 344         OBD_FREE(request->rq_reqmsg, request->rq_reqlen);
 345         request->rq_reqmsg = NULL;
 346         request->rq_reqlen = 0;
 347 }
 348
 349 /* Abort this request and cleanup any resources associated with it. */
 350 static int ptlrpc_abort(struct ptlrpc_request *request)
 351 {
 352         /* First remove the ME for the reply; in theory, this means
 353          * that we can tear down the buffer safely. */
 354         PtlMEUnlink(request->rq_reply_me_h);
 355         OBD_FREE(request->rq_reply_md.start, request->rq_replen);
 356         request->rq_repmsg = NULL;
 357         request->rq_replen = 0;
 358         return 0;
 359 }
 360
 361 /* caller must lock cli */
 362 void ptlrpc_free_committed(struct ptlrpc_client *cli)
 363 {
 364         struct list_head *tmp, *saved;
 365         struct ptlrpc_request *req;
 366
 367         list_for_each_safe(tmp, saved, &cli->cli_sending_head) {
 368                 req = list_entry(tmp, struct ptlrpc_request, rq_list);
 369
 370                 if ( (req->rq_flags & PTL_RPC_FL_REPLAY) ) {
 371                         CDEBUG(D_INFO, "Retaining request %Ld for replay\n",
 372                                req->rq_xid);
 373                         continue;
 374                 }
 375
 376                 /* not yet committed */
 377                 if (req->rq_transno > cli->cli_last_committed)
 378                         break;
 379
 380                 CDEBUG(D_INFO, "Marking request %Ld as committed ("
 381                        "transno=%Lu, last_committed=%Lu\n",
 382                        req->rq_xid, req->rq_transno,
 383                        cli->cli_last_committed);
 384                 if (atomic_dec_and_test(&req->rq_refcount)) {
 385                         /* we do this to prevent free_req deadlock */
 386                         list_del_init(&req->rq_list);
 387                         req->rq_client = NULL;
 388                         ptlrpc_free_req(req);
 389                 } else {
 390                         list_del_init(&req->rq_list);
 391                         list_add(&req->rq_list, &cli->cli_dying_head);
 392                 }
 393         }
 394
 395         EXIT;
 396         return;
 397 }
 398
 399 void ptlrpc_cleanup_client(struct ptlrpc_client *cli)
 400 {
 401         struct list_head *tmp, *saved;
 402         struct ptlrpc_request *req;
 403         ENTRY;
 404
 405         spin_lock(&cli->cli_lock);
 406         list_for_each_safe(tmp, saved, &cli->cli_sending_head) {
 407                 req = list_entry(tmp, struct ptlrpc_request, rq_list);
 408                 CDEBUG(D_INFO, "Cleaning req %p from sending list.\n", req);
 409                 list_del_init(&req->rq_list);
 410                 req->rq_client = NULL;
 411                 ptlrpc_free_req(req);
 412         }
 413         list_for_each_safe(tmp, saved, &cli->cli_dying_head) {
 414                 req = list_entry(tmp, struct ptlrpc_request, rq_list);
 415                 CERROR("Request %p is on the dying list at cleanup!\n", req);
 416                 list_del_init(&req->rq_list);
 417                 req->rq_client = NULL;
 418                 ptlrpc_free_req(req);
 419         }
 420         spin_unlock(&cli->cli_lock);
 421
 422         EXIT;
 423         return;
 424 }
 425
 426 void ptlrpc_continue_req(struct ptlrpc_request *req)
 427 {
 428         ENTRY;
 429         CDEBUG(D_INODE, "continue delayed request %Ld opc %d\n",
 430                req->rq_xid, req->rq_reqmsg->opc);
 431         wake_up(&req->rq_wait_for_rep);
 432         EXIT;
 433 }
 434
 435 void ptlrpc_resend_req(struct ptlrpc_request *req)
 436 {
 437         ENTRY;
 438         CDEBUG(D_INODE, "resend request %Ld, opc %d\n",
 439                req->rq_xid, req->rq_reqmsg->opc);
 440         req->rq_status = -EAGAIN;
 441         req->rq_level = LUSTRE_CONN_RECOVD;
 442         req->rq_flags |= PTL_RPC_FL_RESEND;
 443         req->rq_flags &= ~PTL_RPC_FL_TIMEOUT;
 444         wake_up(&req->rq_wait_for_rep);
 445         EXIT;
 446 }
 447
 448 void ptlrpc_restart_req(struct ptlrpc_request *req)
 449 {
 450         ENTRY;
 451         CDEBUG(D_INODE, "restart completed request %Ld, opc %d\n",
 452                req->rq_xid, req->rq_reqmsg->opc);
 453         req->rq_status = -ERESTARTSYS;
 454         req->rq_flags |= PTL_RPC_FL_RECOVERY;
 455         req->rq_flags &= ~PTL_RPC_FL_TIMEOUT;
 456         wake_up(&req->rq_wait_for_rep);
 457         EXIT;
 458 }
 459
 460 int ptlrpc_queue_wait(struct ptlrpc_request *req)
 461 {
 462         int rc = 0;
 463         struct ptlrpc_client *cli = req->rq_client;
 464         ENTRY;
 465
 466         init_waitqueue_head(&req->rq_wait_for_rep);
 467         CDEBUG(D_NET, "subsys: %s req %Ld opc %d level %d, conn level %d\n",
 468                cli->cli_name, req->rq_xid, req->rq_reqmsg->opc, req->rq_level,
 469                req->rq_connection->c_level);
 470
 471         /* XXX probably both an import and connection level are needed */
 472         if (req->rq_level > req->rq_connection->c_level) {
 473                 CERROR("process %d waiting for recovery (%d > %d)\n",
 474                        current->pid, req->rq_level, req->rq_connection->c_level);
 475                 spin_lock(&cli->cli_lock);
 476                 list_del_init(&req->rq_list);
 477                 list_add(&req->rq_list, cli->cli_delayed_head.prev);
 478                 spin_unlock(&cli->cli_lock);
 479                 l_wait_event_killable
 480                         (req->rq_wait_for_rep,
 481                          req->rq_level <= req->rq_connection->c_level);
 482                 spin_lock(&cli->cli_lock);
 483                 list_del_init(&req->rq_list);
 484                 spin_unlock(&cli->cli_lock);
 485                 CERROR("process %d resumed\n", current->pid);
 486         }
 487  resend:
 488         req->rq_time = CURRENT_TIME;
 489         req->rq_timeout = 100;
 490         rc = ptl_send_rpc(req);
 491         if (rc) {
 492                 CERROR("error %d, opcode %d\n", rc, req->rq_reqmsg->opc);
 493                 if ( rc > 0 )
 494                         rc = -rc;
 495                 ptlrpc_cleanup_request_buf(req);
 496                 up(&cli->cli_rpc_sem);
 497                 RETURN(-rc);
 498         }
 499
 500         spin_lock(&cli->cli_lock);
 501         list_del_init(&req->rq_list);
 502         list_add_tail(&req->rq_list, &cli->cli_sending_head);
 503         spin_unlock(&cli->cli_lock);
 504
 505         CDEBUG(D_OTHER, "-- sleeping\n");
 506         l_wait_event_killable(req->rq_wait_for_rep, ptlrpc_check_reply(req));
 507         CDEBUG(D_OTHER, "-- done\n");
 508
 509         if (req->rq_flags & PTL_RPC_FL_RESEND) {
 510                 req->rq_flags &= ~PTL_RPC_FL_RESEND;
 511                 goto resend;
 512         }
 513
 514         up(&cli->cli_rpc_sem);
 515         if (req->rq_flags & PTL_RPC_FL_TIMEOUT)
 516                 GOTO(out, rc = -ETIMEDOUT);
 517
 518         if (req->rq_flags & PTL_RPC_FL_INTR) {
 519                 /* Clean up the dangling reply buffers */
 520                 ptlrpc_abort(req);
 521                 GOTO(out, rc = -EINTR);
 522         }
 523
 524         if (!(req->rq_flags & PTL_RPC_FL_REPLIED))
 525                 GOTO(out, rc = req->rq_status);
 526
 527         rc = lustre_unpack_msg(req->rq_repmsg, req->rq_replen);
 528         if (rc) {
 529                 CERROR("unpack_rep failed: %d\n", rc);
 530                 GOTO(out, rc);
 531         }
 532         CDEBUG(D_NET, "got rep %Ld\n", req->rq_xid);
 533         if (req->rq_repmsg->status == 0)
 534                 CDEBUG(D_NET, "--> buf %p len %d status %d\n", req->rq_repmsg,
 535                        req->rq_replen, req->rq_repmsg->status);
 536
 537         spin_lock(&cli->cli_lock);
 538         cli->cli_last_rcvd = req->rq_repmsg->last_rcvd;
 539         cli->cli_last_committed = req->rq_repmsg->last_committed;
 540         ptlrpc_free_committed(cli);
 541         spin_unlock(&cli->cli_lock);
 542
 543         EXIT;
 544  out:
 545         return rc;
 546 }
 547
 548 int ptlrpc_replay_req(struct ptlrpc_request *req)
 549 {
 550         int rc = 0;
 551         struct ptlrpc_client *cli = req->rq_client;
 552         ENTRY;
 553
 554         init_waitqueue_head(&req->rq_wait_for_rep);
 555         CDEBUG(D_NET, "req %Ld opc %d level %d, conn level %d\n",
 556                req->rq_xid, req->rq_reqmsg->opc, req->rq_level,
 557                req->rq_connection->c_level);
 558
 559         req->rq_time = CURRENT_TIME;
 560         req->rq_timeout = 100;
 561         rc = ptl_send_rpc(req);
 562         if (rc) {
 563                 CERROR("error %d, opcode %d\n", rc, req->rq_reqmsg->opc);
 564                 ptlrpc_cleanup_request_buf(req);
 565                 up(&cli->cli_rpc_sem);
 566                 RETURN(-rc);
 567         }
 568
 569         CDEBUG(D_OTHER, "-- sleeping\n");
 570         l_wait_event_killable(req->rq_wait_for_rep, ptlrpc_check_reply(req));
 571         CDEBUG(D_OTHER, "-- done\n");
 572
 573         up(&cli->cli_rpc_sem);
 574
 575         if (!(req->rq_flags & PTL_RPC_FL_REPLIED)) {
 576                 CERROR("Unknown reason for wakeup\n");
 577                 /* XXX Phil - I end up here when I kill obdctl */
 578                 ptlrpc_abort(req);
 579                 GOTO(out, rc = -EINTR);
 580         }
 581
 582         rc = lustre_unpack_msg(req->rq_repmsg, req->rq_replen);
 583         if (rc) {
 584                 CERROR("unpack_rep failed: %d\n", rc);
 585                 GOTO(out, rc);
 586         }
 587
 588         CDEBUG(D_NET, "got rep %Ld\n", req->rq_xid);
 589         if (req->rq_repmsg->status == 0)
 590                 CDEBUG(D_NET, "--> buf %p len %d status %d\n", req->rq_repmsg,
 591                        req->rq_replen, req->rq_repmsg->status);
 592         else {
 593                 CERROR("recovery failed: ");
 594                 CERROR("req %Ld opc %d level %d, conn level %d\n",
 595                        req->rq_xid, req->rq_reqmsg->opc, req->rq_level,
 596                        req->rq_connection->c_level);
 597                 LBUG();
 598         }
 599
 600  out:
 601         RETURN(rc);
 602 }