lustre/osc/osc_request.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  * Copyright (C) 2001, 2002 Cluster File Systems, Inc.
   5  *
   6  *  This code is issued under the GNU General Public License.
   7  *  See the file COPYING in this distribution
   8  *
   9  *  Author Peter Braam <braam@clusterfs.com>
  10  *
  11  *  This server is single threaded at present (but can easily be multi
  12  *  threaded). For testing and management it is treated as an
  13  *  obd_device, although it does not export a full OBD method table
  14  *  (the requests are coming in over the wire, so object target
  15  *  modules do not have a full method table.)
  16  *
  17  */
  18
  19 #define EXPORT_SYMTAB
  20 #define DEBUG_SUBSYSTEM S_OSC
  21
  22 #include <linux/module.h>
  23 #include <linux/lustre_dlm.h>
  24 #include <linux/lustre_mds.h> /* for mds_objid */
  25 #include <linux/obd_ost.h>
  26 #include <linux/obd_lov.h>
  27
  28 static void osc_con2cl(struct lustre_handle *conn, struct ptlrpc_client **cl,
  29                        struct ptlrpc_connection **connection,
  30                        struct lustre_handle **rconn)
  31 {
  32         struct obd_export *export = class_conn2export(conn);
  33         struct osc_obd *osc = &export->exp_obd->u.osc;
  34
  35         *cl = osc->osc_client;
  36         *connection = osc->osc_conn;
  37         *rconn = &export->exp_rconnh;
  38 }
  39
  40 static void osc_con2dlmcl(struct lustre_handle *conn, struct ptlrpc_client **cl,
  41                           struct ptlrpc_connection **connection,
  42                           struct lustre_handle **rconn)
  43 {
  44         struct obd_export *export = class_conn2export(conn);
  45         struct osc_obd *osc = &export->exp_obd->u.osc;
  46
  47         *cl = osc->osc_ldlm_client;
  48         *connection = osc->osc_conn;
  49         *rconn = &export->exp_rconnh;
  50 }
  51
  52 static int osc_connect(struct lustre_handle *conn, struct obd_device *obd)
  53 {
  54         struct osc_obd *osc = &obd->u.osc;
  55         //struct obd_import *import;
  56         struct ptlrpc_request *request;
  57         char *tmp = osc->osc_target_uuid;
  58         int rc, size = sizeof(osc->osc_target_uuid);
  59         ENTRY;
  60
  61         /* not used yet
  62         OBD_ALLOC(import, sizeof(*import));
  63         if (!import)
  64                 RETURN(-ENOMEM);
  65          */
  66
  67         MOD_INC_USE_COUNT;
  68         rc = class_connect(conn, obd);
  69         if (rc)
  70                 RETURN(rc);
  71
  72         request = ptlrpc_prep_req(osc->osc_client, osc->osc_conn,
  73                                   OST_CONNECT, 1, &size, &tmp);
  74         if (!request)
  75                 GOTO(out_disco, rc = -ENOMEM);
  76
  77         request->rq_level = LUSTRE_CONN_NEW;
  78         request->rq_replen = lustre_msg_size(0, NULL);
  79         request->rq_reqmsg->addr = -1;
  80         /* Sending our local connection info breaks for local connections
  81         request->rq_reqmsg->addr = conn->addr;
  82         request->rq_reqmsg->cookie = conn->cookie;
  83          */
  84
  85         rc = ptlrpc_queue_wait(request);
  86         rc = ptlrpc_check_status(request, rc);
  87         if (rc) {
  88                 CERROR("%s failed: rc = %d\n", __FUNCTION__, rc);
  89                 GOTO(out, rc);
  90         }
  91
  92         /* XXX eventually maybe more refinement */
  93         osc->osc_conn->c_level = LUSTRE_CONN_FULL;
  94
  95         class_rconn2export(conn, (struct lustre_handle *)request->rq_repmsg);
  96
  97         EXIT;
  98  out:
  99         ptlrpc_free_req(request);
 100  out_disco:
 101         if (rc) {
 102                 class_disconnect(conn);
 103                 MOD_DEC_USE_COUNT;
 104         }
 105         return rc;
 106 }
 107
 108 static int osc_disconnect(struct lustre_handle *conn)
 109 {
 110         struct ptlrpc_request *request;
 111         struct ptlrpc_client *cl;
 112         struct ptlrpc_connection *connection;
 113         struct lustre_handle *rconn;
 114         int rc;
 115         ENTRY;
 116
 117         osc_con2cl(conn, &cl, &connection, &rconn);
 118         request = ptlrpc_prep_req2(cl, connection, rconn,
 119                                    OST_DISCONNECT, 0, NULL, NULL);
 120         if (!request)
 121                 RETURN(-ENOMEM);
 122         request->rq_replen = lustre_msg_size(0, NULL);
 123
 124         rc = ptlrpc_queue_wait(request);
 125         if (rc)
 126                 GOTO(out, rc);
 127         rc = class_disconnect(conn);
 128         if (!rc)
 129                 MOD_DEC_USE_COUNT;
 130
 131  out:
 132         ptlrpc_free_req(request);
 133         return rc;
 134 }
 135
 136 static int osc_getattr(struct lustre_handle *conn, struct obdo *oa)
 137 {
 138         struct ptlrpc_request *request;
 139         struct ptlrpc_client *cl;
 140         struct ptlrpc_connection *connection;
 141         struct lustre_handle *rconn;
 142         struct ost_body *body;
 143         int rc, size = sizeof(*body);
 144         ENTRY;
 145
 146         osc_con2cl(conn, &cl, &connection, &rconn);
 147         request = ptlrpc_prep_req2(cl, connection, rconn,
 148                                    OST_GETATTR, 1, &size, NULL);
 149         if (!request)
 150                 RETURN(-ENOMEM);
 151
 152         body = lustre_msg_buf(request->rq_reqmsg, 0);
 153         memcpy(&body->oa, oa, sizeof(*oa));
 154         body->oa.o_valid = ~0;
 155
 156         request->rq_replen = lustre_msg_size(1, &size);
 157
 158         rc = ptlrpc_queue_wait(request);
 159         rc = ptlrpc_check_status(request, rc);
 160         if (rc) {
 161                 CERROR("%s failed: rc = %d\n", __FUNCTION__, rc);
 162                 GOTO(out, rc);
 163         }
 164
 165         body = lustre_msg_buf(request->rq_repmsg, 0);
 166         CDEBUG(D_INODE, "mode: %o\n", body->oa.o_mode);
 167         if (oa)
 168                 memcpy(oa, &body->oa, sizeof(*oa));
 169
 170         EXIT;
 171  out:
 172         ptlrpc_free_req(request);
 173         return rc;
 174 }
 175
 176 static int osc_open(struct lustre_handle *conn, struct obdo *oa,
 177                     struct lov_stripe_md *md)
 178 {
 179         struct ptlrpc_request *request;
 180         struct ptlrpc_client *cl;
 181         struct ptlrpc_connection *connection;
 182         struct lustre_handle *rconn;
 183         struct ost_body *body;
 184         int rc, size = sizeof(*body);
 185         ENTRY;
 186
 187         osc_con2cl(conn, &cl, &connection, &rconn);
 188         request = ptlrpc_prep_req2(cl, connection, rconn,
 189                                    OST_OPEN, 1, &size, NULL);
 190         if (!request)
 191                 RETURN(-ENOMEM);
 192
 193         body = lustre_msg_buf(request->rq_reqmsg, 0);
 194         memcpy(&body->oa, oa, sizeof(*oa));
 195         body->oa.o_valid = (OBD_MD_FLMODE | OBD_MD_FLID);
 196
 197         request->rq_replen = lustre_msg_size(1, &size);
 198
 199         rc = ptlrpc_queue_wait(request);
 200         rc = ptlrpc_check_status(request, rc);
 201         if (rc)
 202                 GOTO(out, rc);
 203
 204         body = lustre_msg_buf(request->rq_repmsg, 0);
 205         CDEBUG(D_INODE, "mode: %o\n", body->oa.o_mode);
 206         if (oa)
 207                 memcpy(oa, &body->oa, sizeof(*oa));
 208
 209         EXIT;
 210  out:
 211         ptlrpc_free_req(request);
 212         return rc;
 213 }
 214
 215 static int osc_close(struct lustre_handle *conn, struct obdo *oa,
 216                      struct lov_stripe_md *md)
 217 {
 218         struct ptlrpc_request *request;
 219         struct ptlrpc_client *cl;
 220         struct ptlrpc_connection *connection;
 221         struct lustre_handle *rconn;
 222         struct ost_body *body;
 223         int rc, size = sizeof(*body);
 224         ENTRY;
 225
 226         osc_con2cl(conn, &cl, &connection, &rconn);
 227         request = ptlrpc_prep_req2(cl, connection, rconn,
 228                                    OST_CLOSE, 1, &size, NULL);
 229         if (!request)
 230                 RETURN(-ENOMEM);
 231
 232         oa->o_id = md->lmd_object_id;
 233         oa->o_mode = S_IFREG;
 234         oa->o_valid = (OBD_MD_FLMODE | OBD_MD_FLID);
 235         body = lustre_msg_buf(request->rq_reqmsg, 0);
 236         memcpy(&body->oa, oa, sizeof(*oa));
 237
 238         request->rq_replen = lustre_msg_size(1, &size);
 239
 240         rc = ptlrpc_queue_wait(request);
 241         rc = ptlrpc_check_status(request, rc);
 242         if (rc)
 243                 GOTO(out, rc);
 244
 245         body = lustre_msg_buf(request->rq_repmsg, 0);
 246         CDEBUG(D_INODE, "mode: %o\n", body->oa.o_mode);
 247         if (oa)
 248                 memcpy(oa, &body->oa, sizeof(*oa));
 249
 250         EXIT;
 251  out:
 252         ptlrpc_free_req(request);
 253         return rc;
 254 }
 255
 256 static int osc_setattr(struct lustre_handle *conn, struct obdo *oa)
 257 {
 258         struct ptlrpc_request *request;
 259         struct ptlrpc_client *cl;
 260         struct ptlrpc_connection *connection;
 261         struct lustre_handle *rconn;
 262         struct ost_body *body;
 263         int rc, size = sizeof(*body);
 264         ENTRY;
 265
 266         osc_con2cl(conn, &cl, &connection, &rconn);
 267         request = ptlrpc_prep_req2(cl, connection, rconn,
 268                                   OST_SETATTR, 1, &size, NULL);
 269         if (!request)
 270                 RETURN(-ENOMEM);
 271
 272         body = lustre_msg_buf(request->rq_reqmsg, 0);
 273         memcpy(&body->oa, oa, sizeof(*oa));
 274
 275         request->rq_replen = lustre_msg_size(1, &size);
 276
 277         rc = ptlrpc_queue_wait(request);
 278         rc = ptlrpc_check_status(request, rc);
 279         GOTO(out, rc);
 280
 281  out:
 282         ptlrpc_free_req(request);
 283         return rc;
 284 }
 285
 286 static int osc_create(struct lustre_handle *conn, struct obdo *oa,
 287                       struct lov_stripe_md **ea)
 288 {
 289         struct ptlrpc_request *request;
 290         struct ptlrpc_client *cl;
 291         struct ptlrpc_connection *connection;
 292         struct lustre_handle *rconn;
 293         struct ost_body *body;
 294         int rc, size = sizeof(*body);
 295         ENTRY;
 296
 297         if (!oa) {
 298                 CERROR("oa NULL\n");
 299                 RETURN(-EINVAL);
 300         }
 301
 302         if (!ea) {
 303                 LBUG();
 304         }
 305
 306         if (!*ea) {
 307                 OBD_ALLOC(*ea, oa->o_easize);
 308                 if (!*ea)
 309                         RETURN(-ENOMEM);
 310                 (*ea)->lmd_size = oa->o_easize;
 311         }
 312
 313         osc_con2cl(conn, &cl, &connection, &rconn);
 314         request = ptlrpc_prep_req2(cl, connection, rconn,
 315                                   OST_CREATE, 1, &size, NULL);
 316         if (!request)
 317                 RETURN(-ENOMEM);
 318
 319         body = lustre_msg_buf(request->rq_reqmsg, 0);
 320         memcpy(&body->oa, oa, sizeof(*oa));
 321
 322         request->rq_replen = lustre_msg_size(1, &size);
 323
 324         rc = ptlrpc_queue_wait(request);
 325         rc = ptlrpc_check_status(request, rc);
 326         if (rc)
 327                 GOTO(out, rc);
 328
 329         body = lustre_msg_buf(request->rq_repmsg, 0);
 330         memcpy(oa, &body->oa, sizeof(*oa));
 331
 332         (*ea)->lmd_object_id = oa->o_id;
 333         (*ea)->lmd_stripe_count = 1;
 334         EXIT;
 335  out:
 336         ptlrpc_free_req(request);
 337         return rc;
 338 }
 339
 340 static int osc_punch(struct lustre_handle *conn, struct obdo *oa,
 341                      struct lov_stripe_md *md, obd_size count,
 342                      obd_off offset)
 343 {
 344         struct ptlrpc_request *request;
 345         struct ptlrpc_client *cl;
 346         struct ptlrpc_connection *connection;
 347         struct lustre_handle *rconn;
 348         struct ost_body *body;
 349         int rc, size = sizeof(*body);
 350         ENTRY;
 351
 352         if (!oa) {
 353                 CERROR("oa NULL\n");
 354                 RETURN(-EINVAL);
 355         }
 356         osc_con2cl(conn, &cl, &connection, &rconn);
 357         request = ptlrpc_prep_req2(cl, connection, rconn,
 358                                    OST_PUNCH, 1, &size, NULL);
 359         if (!request)
 360                 RETURN(-ENOMEM);
 361
 362         body = lustre_msg_buf(request->rq_reqmsg, 0);
 363         memcpy(&body->oa, oa, sizeof(*oa));
 364         body->oa.o_blocks = count;
 365         body->oa.o_valid |= OBD_MD_FLBLOCKS;
 366
 367         request->rq_replen = lustre_msg_size(1, &size);
 368
 369         rc = ptlrpc_queue_wait(request);
 370         rc = ptlrpc_check_status(request, rc);
 371         if (rc)
 372                 GOTO(out, rc);
 373
 374         body = lustre_msg_buf(request->rq_repmsg, 0);
 375         memcpy(oa, &body->oa, sizeof(*oa));
 376
 377         EXIT;
 378  out:
 379         ptlrpc_free_req(request);
 380         return rc;
 381 }
 382
 383 static int osc_destroy(struct lustre_handle *conn, struct obdo *oa,
 384                        struct lov_stripe_md *ea)
 385 {
 386         struct ptlrpc_request *request;
 387         struct ptlrpc_client *cl;
 388         struct ptlrpc_connection *connection;
 389         struct lustre_handle *rconn;
 390         struct ost_body *body;
 391         int rc, size = sizeof(*body);
 392         ENTRY;
 393
 394         if (!oa) {
 395                 CERROR("oa NULL\n");
 396                 RETURN(-EINVAL);
 397         }
 398         osc_con2cl(conn, &cl, &connection, &rconn);
 399         request = ptlrpc_prep_req2(cl, connection, rconn,
 400                                    OST_DESTROY, 1, &size, NULL);
 401         if (!request)
 402                 RETURN(-ENOMEM);
 403
 404         body = lustre_msg_buf(request->rq_reqmsg, 0);
 405         memcpy(&body->oa, oa, sizeof(*oa));
 406         body->oa.o_valid = ~0;
 407
 408         request->rq_replen = lustre_msg_size(1, &size);
 409
 410         rc = ptlrpc_queue_wait(request);
 411         rc = ptlrpc_check_status(request, rc);
 412         if (rc)
 413                 GOTO(out, rc);
 414
 415         body = lustre_msg_buf(request->rq_repmsg, 0);
 416         memcpy(oa, &body->oa, sizeof(*oa));
 417
 418         EXIT;
 419  out:
 420         ptlrpc_free_req(request);
 421         return rc;
 422 }
 423
 424 struct osc_brw_cb_data {
 425         brw_callback_t callback;
 426         void *cb_data;
 427         void *obd_data;
 428         size_t obd_size;
 429 };
 430
 431 /* Our bulk-unmapping bottom half. */
 432 static void unmap_and_decref_bulk_desc(void *data)
 433 {
 434         struct ptlrpc_bulk_desc *desc = data;
 435         struct list_head *tmp;
 436         ENTRY;
 437
 438         /* This feels wrong to me. */
 439         list_for_each(tmp, &desc->b_page_list) {
 440                 struct ptlrpc_bulk_page *bulk;
 441                 bulk = list_entry(tmp, struct ptlrpc_bulk_page, b_link);
 442
 443                 kunmap(bulk->b_page);
 444         }
 445
 446         ptlrpc_bulk_decref(desc);
 447         EXIT;
 448 }
 449
 450 static void brw_finish(struct ptlrpc_bulk_desc *desc, void *data)
 451 {
 452         struct osc_brw_cb_data *cb_data = data;
 453         ENTRY;
 454
 455         if (desc->b_flags & PTL_RPC_FL_INTR)
 456                 CERROR("got signal\n");
 457
 458         if (cb_data->callback)
 459                 cb_data->callback(cb_data->cb_data);
 460
 461         OBD_FREE(cb_data->obd_data, cb_data->obd_size);
 462         OBD_FREE(cb_data, sizeof(*cb_data));
 463
 464         /* We can't kunmap the desc from interrupt context, so we do it from
 465          * the bottom half above. */
 466         INIT_TQUEUE(&desc->b_queue, 0, 0);
 467         PREPARE_TQUEUE(&desc->b_queue, unmap_and_decref_bulk_desc, desc);
 468         schedule_task(&desc->b_queue);
 469
 470         EXIT;
 471 }
 472
 473 static int osc_brw_read(struct lustre_handle *conn, struct lov_stripe_md *md,
 474                         obd_count page_count, struct page **page_array,
 475                         obd_size *count, obd_off *offset, obd_flag *flags,
 476                         brw_callback_t callback, void *data)
 477 {
 478         struct ptlrpc_client *cl;
 479         struct ptlrpc_connection *connection;
 480         struct lustre_handle *rconn;
 481         struct ptlrpc_request *request = NULL;
 482         struct ptlrpc_bulk_desc *desc = NULL;
 483         struct ost_body *body;
 484         struct osc_brw_cb_data *cb_data = NULL;
 485         int rc, size[3] = {sizeof(*body)};
 486         void *iooptr, *nioptr;
 487         int mapped = 0;
 488         ENTRY;
 489
 490         size[1] = sizeof(struct obd_ioobj);
 491         size[2] = page_count * sizeof(struct niobuf_remote);
 492
 493         osc_con2cl(conn, &cl, &connection, &rconn);
 494         request = ptlrpc_prep_req2(cl, connection, rconn,
 495                                    OST_READ, 3, size, NULL);
 496         if (!request)
 497                 RETURN(-ENOMEM);
 498
 499         body = lustre_msg_buf(request->rq_reqmsg, 0);
 500
 501         desc = ptlrpc_prep_bulk(connection);
 502         if (!desc)
 503                 GOTO(out_req, rc = -ENOMEM);
 504         desc->b_portal = OST_BULK_PORTAL;
 505         desc->b_cb = brw_finish;
 506         OBD_ALLOC(cb_data, sizeof(*cb_data));
 507         if (!cb_data)
 508                 GOTO(out_desc, rc = -ENOMEM);
 509
 510         cb_data->callback = callback;
 511         cb_data->cb_data = data;
 512         desc->b_cb_data = cb_data;
 513
 514         iooptr = lustre_msg_buf(request->rq_reqmsg, 1);
 515         nioptr = lustre_msg_buf(request->rq_reqmsg, 2);
 516         ost_pack_ioo(&iooptr, md, page_count);
 517         /* end almost identical to brw_write case */
 518
 519         for (mapped = 0; mapped < page_count; mapped++) {
 520                 struct ptlrpc_bulk_page *bulk = ptlrpc_prep_bulk_page(desc);
 521                 if (bulk == NULL)
 522                         GOTO(out_unmap, rc = -ENOMEM);
 523
 524                 spin_lock(&connection->c_lock);
 525                 bulk->b_xid = ++connection->c_xid_out;
 526                 spin_unlock(&connection->c_lock);
 527
 528                 bulk->b_buf = kmap(page_array[mapped]);
 529                 bulk->b_page = page_array[mapped];
 530                 bulk->b_buflen = PAGE_SIZE;
 531                 ost_pack_niobuf(&nioptr, offset[mapped], count[mapped],
 532                                 flags[mapped], bulk->b_xid);
 533         }
 534
 535         /*
 536          * Register the bulk first, because the reply could arrive out of order,
 537          * and we want to be ready for the bulk data.
 538          *
 539          * One reference is released when brw_finish is complete, the
 540          * other here when we finish waiting on it if we don't have a callback.
 541          *
 542          * We don't reference the bulk descriptor again here if there is a
 543          * callback, so we don't need an additional refcount on it.
 544          *
 545          * On error, we never do the brw_finish, so we handle all decrefs.
 546          */
 547         if (!callback)
 548                 ptlrpc_bulk_addref(desc);
 549         rc = ptlrpc_register_bulk(desc);
 550         if (rc)
 551                 GOTO(out_desc2, rc);
 552
 553         request->rq_replen = lustre_msg_size(1, size);
 554         rc = ptlrpc_queue_wait(request);
 555         rc = ptlrpc_check_status(request, rc);
 556
 557         /* XXX: Mike, this is the only place I'm not sure of.  If we had
 558          *      an error here, will we always call brw_finish?  If yes, then
 559          *      out_desc_2 will do too much and we should jump to out_desc.
 560          *      If maybe, then we are screwed, and we need to set things up
 561          *      so that bulk_sink_callback is called for each bulk page,
 562          *      even on error so brw_finish is always called.  It would need
 563          *      to be passed an error code as a parameter to know what to do.
 564          *
 565          *      That would also help with the partial completion case, so
 566          *      we could say in brw_finish "these pages are done, don't
 567          *      restart them" and osc_brw callers can know this.
 568          */
 569         if (rc)
 570                 GOTO(out_desc2, rc);
 571
 572         /* Callbacks cause asynchronous handling. */
 573         if (callback)
 574                 GOTO(out_req, rc = 0);
 575
 576         /* If there's no callback function, sleep here until complete. */
 577         l_wait_event_killable(desc->b_waitq, ptlrpc_check_bulk_received(desc));
 578         if (desc->b_flags & PTL_RPC_FL_INTR)
 579                 GOTO(out_desc, rc = -EINTR);
 580
 581         EXIT;
 582 out_desc:
 583         ptlrpc_bulk_decref(desc);
 584 out_req:
 585         ptlrpc_req_finished(request);
 586         RETURN(rc);
 587
 588         /* Clean up on error. */
 589 out_desc2:
 590         if (!callback)
 591                 ptlrpc_bulk_decref(desc);
 592 out_unmap:
 593         while (mapped-- > 0)
 594                 kunmap(page_array[mapped]);
 595         OBD_FREE(cb_data, sizeof(*cb_data));
 596         goto out_desc;
 597 }
 598
 599 static int osc_brw_write(struct lustre_handle *conn,
 600                          struct lov_stripe_md *md, obd_count page_count,
 601                          struct page **pagearray, obd_size *count,
 602                          obd_off *offset, obd_flag *flags,
 603                          brw_callback_t callback, void *data)
 604 {
 605         struct ptlrpc_client *cl;
 606         struct ptlrpc_connection *connection;
 607         struct lustre_handle *rconn;
 608         struct ptlrpc_request *request = NULL;
 609         struct ptlrpc_bulk_desc *desc = NULL;
 610         struct ost_body *body;
 611         struct niobuf_local *local = NULL;
 612         struct niobuf_remote *remote;
 613         struct osc_brw_cb_data *cb_data = NULL;
 614         int rc, j, size[3] = {sizeof(*body)};
 615         void *iooptr, *nioptr;
 616         int mapped = 0;
 617         ENTRY;
 618
 619         size[1] = sizeof(struct obd_ioobj);
 620         size[2] = page_count * sizeof(*remote);
 621
 622         osc_con2cl(conn, &cl, &connection, &rconn);
 623         request = ptlrpc_prep_req2(cl, connection, rconn,
 624                                    OST_WRITE, 3, size, NULL);
 625         if (!request)
 626                 RETURN(-ENOMEM);
 627
 628         body = lustre_msg_buf(request->rq_reqmsg, 0);
 629
 630         desc = ptlrpc_prep_bulk(connection);
 631         if (!desc)
 632                 GOTO(out_req, rc = -ENOMEM);
 633         desc->b_portal = OSC_BULK_PORTAL;
 634         desc->b_cb = brw_finish;
 635         OBD_ALLOC(cb_data, sizeof(*cb_data));
 636         if (!cb_data)
 637                 GOTO(out_desc, rc = -ENOMEM);
 638
 639         cb_data->callback = callback;
 640         cb_data->cb_data = data;
 641         desc->b_cb_data = cb_data;
 642
 643         iooptr = lustre_msg_buf(request->rq_reqmsg, 1);
 644         nioptr = lustre_msg_buf(request->rq_reqmsg, 2);
 645         ost_pack_ioo(&iooptr, md, page_count);
 646         /* end almost identical to brw_read case */
 647
 648         OBD_ALLOC(local, page_count * sizeof(*local));
 649         if (!local)
 650                 GOTO(out_cb, rc = -ENOMEM);
 651
 652         cb_data->obd_data = local;
 653         cb_data->obd_size = page_count * sizeof(*local);
 654
 655         for (mapped = 0; mapped < page_count; mapped++) {
 656                 local[mapped].addr = kmap(pagearray[mapped]);
 657                 local[mapped].offset = offset[mapped];
 658                 local[mapped].len = count[mapped];
 659                 ost_pack_niobuf(&nioptr, offset[mapped], count[mapped],
 660                                 flags[mapped], 0);
 661         }
 662
 663         size[1] = page_count * sizeof(*remote);
 664         request->rq_replen = lustre_msg_size(2, size);
 665         rc = ptlrpc_queue_wait(request);
 666         rc = ptlrpc_check_status(request, rc);
 667         if (rc)
 668                 GOTO(out_unmap, rc);
 669
 670         nioptr = lustre_msg_buf(request->rq_repmsg, 1);
 671         if (!nioptr)
 672                 GOTO(out_unmap, rc = -EINVAL);
 673
 674         if (request->rq_repmsg->buflens[1] != size[1]) {
 675                 CERROR("buffer length wrong (%d vs. %d)\n",
 676                        request->rq_repmsg->buflens[1], size[1]);
 677                 GOTO(out_unmap, rc = -EINVAL);
 678         }
 679
 680         for (j = 0; j < page_count; j++) {
 681                 struct ptlrpc_bulk_page *bulk;
 682
 683                 ost_unpack_niobuf(&nioptr, &remote);
 684
 685                 bulk = ptlrpc_prep_bulk_page(desc);
 686                 if (!bulk)
 687                         GOTO(out_unmap, rc = -ENOMEM);
 688
 689                 bulk->b_buf = (void *)(unsigned long)local[j].addr;
 690                 bulk->b_buflen = local[j].len;
 691                 bulk->b_xid = remote->xid;
 692                 bulk->b_page = pagearray[j];
 693         }
 694
 695         if (desc->b_page_count != page_count)
 696                 LBUG();
 697
 698         /*
 699          * One reference is released when brw_finish is complete, the
 700          * other here when we finish waiting on it if we don't have a callback.
 701          *
 702          * We don't reference the bulk descriptor again here if there is a
 703          * callback, so we don't need an additional refcount on it.
 704          */
 705         if (!callback)
 706                 ptlrpc_bulk_addref(desc);
 707         rc = ptlrpc_send_bulk(desc);
 708
 709         /* XXX: Mike, same question as in osc_brw_read. */
 710         if (rc)
 711                 GOTO(out_desc2, rc);
 712
 713         /* Callbacks cause asynchronous handling. */
 714         if (callback)
 715                 GOTO(out_req, rc = 0);
 716
 717         /* If there's no callback function, sleep here until complete. */
 718         l_wait_event_killable(desc->b_waitq, ptlrpc_check_bulk_sent(desc));
 719         if (desc->b_flags & PTL_RPC_FL_INTR)
 720                 GOTO(out_desc, rc = -EINTR);
 721
 722         EXIT;
 723 out_desc:
 724         ptlrpc_bulk_decref(desc);
 725 out_req:
 726         ptlrpc_req_finished(request);
 727         return rc;
 728
 729         /* Clean up on error. */
 730 out_desc2:
 731         if (!callback)
 732                 ptlrpc_bulk_decref(desc);
 733 out_unmap:
 734         while (mapped-- > 0)
 735                 kunmap(pagearray[mapped]);
 736
 737         OBD_FREE(local, page_count * sizeof(*local));
 738 out_cb:
 739         OBD_FREE(cb_data, sizeof(*cb_data));
 740         goto out_desc;
 741 }
 742
 743 static int osc_brw(int cmd, struct lustre_handle *conn,
 744                    struct lov_stripe_md *md, obd_count page_count,
 745                    struct page **page_array, obd_size *count, obd_off *offset,
 746                    obd_flag *flags, brw_callback_t callback, void *data)
 747 {
 748         if (cmd & OBD_BRW_WRITE)
 749                 return osc_brw_write(conn, md, page_count, page_array, count,
 750                                      offset, flags, callback, data);
 751         else
 752                 return osc_brw_read(conn, md, page_count, page_array, count,
 753                                     offset, flags, callback, data);
 754 }
 755
 756 static int osc_enqueue(struct lustre_handle *conn,
 757                        struct lustre_handle *parent_lock, __u64 *res_id,
 758                        __u32 type, void *extentp, int extent_len, __u32 mode,
 759                        int *flags, void *callback, void *data, int datalen,
 760                        struct lustre_handle *lockh)
 761 {
 762         struct obd_device *obddev = class_conn2obd(conn);
 763         struct ptlrpc_connection *connection;
 764         struct ptlrpc_client *cl;
 765         struct lustre_handle *rconn;
 766         struct ldlm_extent *extent = extentp;
 767         int rc;
 768         __u32 mode2;
 769
 770         /* Filesystem locks are given a bit of special treatment: first we
 771          * fixup the lock to start and end on page boundaries. */
 772         extent->start &= PAGE_MASK;
 773         extent->end = (extent->end + PAGE_SIZE - 1) & PAGE_MASK;
 774
 775         /* Next, search for already existing extent locks that will cover us */
 776         osc_con2dlmcl(conn, &cl, &connection, &rconn);
 777         rc = ldlm_lock_match(obddev->obd_namespace, res_id, type, extent,
 778                              sizeof(extent), mode, lockh);
 779         if (rc == 1) {
 780                 /* We already have a lock, and it's referenced */
 781                 return 0;
 782         }
 783
 784         /* Next, search for locks that we can upgrade (if we're trying to write)
 785          * or are more than we need (if we're trying to read).  Because the VFS
 786          * and page cache already protect us locally, lots of readers/writers
 787          * can share a single PW lock. */
 788         if (mode == LCK_PW)
 789                 mode2 = LCK_PR;
 790         else
 791                 mode2 = LCK_PW;
 792
 793         rc = ldlm_lock_match(obddev->obd_namespace, res_id, type, extent,
 794                              sizeof(extent), mode2, lockh);
 795         if (rc == 1) {
 796                 int flags;
 797                 /* FIXME: This is not incredibly elegant, but it might
 798                  * be more elegant than adding another parameter to
 799                  * lock_match.  I want a second opinion. */
 800                 ldlm_lock_addref(lockh, mode);
 801                 ldlm_lock_decref(lockh, mode2);
 802
 803                 if (mode == LCK_PR)
 804                         return 0;
 805
 806                 rc = ldlm_cli_convert(cl, lockh, rconn, mode, &flags);
 807                 if (rc)
 808                         LBUG();
 809
 810                 return rc;
 811         }
 812
 813         rc = ldlm_cli_enqueue(cl, connection, rconn, NULL,obddev->obd_namespace,
 814                               parent_lock, res_id, type, extent, sizeof(extent),
 815                               mode, flags, callback, data, datalen, lockh);
 816         return rc;
 817 }
 818
 819 static int osc_cancel(struct lustre_handle *oconn, __u32 mode,
 820                       struct lustre_handle *lockh)
 821 {
 822         ENTRY;
 823
 824         ldlm_lock_decref(lockh, mode);
 825
 826         RETURN(0);
 827 }
 828
 829 static int osc_setup(struct obd_device *obddev, obd_count len, void *buf)
 830 {
 831         struct obd_ioctl_data* data = buf;
 832         struct osc_obd *osc = &obddev->u.osc;
 833         char server_uuid[37];
 834         int rc;
 835         ENTRY;
 836
 837         if (data->ioc_inllen1 < 1) {
 838                 CERROR("osc setup requires a TARGET UUID\n");
 839                 RETURN(-EINVAL);
 840         }
 841
 842         if (data->ioc_inllen1 > 37) {
 843                 CERROR("osc TARGET UUID must be less than 38 characters\n");
 844                 RETURN(-EINVAL);
 845         }
 846
 847         if (data->ioc_inllen2 < 1) {
 848                 CERROR("osc setup requires a SERVER UUID\n");
 849                 RETURN(-EINVAL);
 850         }
 851
 852         if (data->ioc_inllen2 > 37) {
 853                 CERROR("osc SERVER UUID must be less than 38 characters\n");
 854                 RETURN(-EINVAL);
 855         }
 856
 857         memcpy(osc->osc_target_uuid, data->ioc_inlbuf1, data->ioc_inllen1);
 858         memcpy(server_uuid, data->ioc_inlbuf2, MIN(data->ioc_inllen2,
 859                                                    sizeof(server_uuid)));
 860
 861         osc->osc_conn = ptlrpc_uuid_to_connection(server_uuid);
 862         if (!osc->osc_conn)
 863                 RETURN(-ENOENT);
 864
 865         obddev->obd_namespace =
 866                 ldlm_namespace_new("osc", LDLM_NAMESPACE_CLIENT);
 867         if (obddev->obd_namespace == NULL)
 868                 GOTO(out_conn, rc = -ENOMEM);
 869
 870         OBD_ALLOC(osc->osc_client, sizeof(*osc->osc_client));
 871         if (osc->osc_client == NULL)
 872                 GOTO(out_ns, rc = -ENOMEM);
 873
 874         OBD_ALLOC(osc->osc_ldlm_client, sizeof(*osc->osc_ldlm_client));
 875         if (osc->osc_ldlm_client == NULL)
 876                 GOTO(out_client, rc = -ENOMEM);
 877
 878         ptlrpc_init_client(NULL, NULL, OST_REQUEST_PORTAL, OSC_REPLY_PORTAL,
 879                            osc->osc_client);
 880         ptlrpc_init_client(NULL, NULL, LDLM_REQUEST_PORTAL, LDLM_REPLY_PORTAL,
 881                            osc->osc_ldlm_client);
 882         osc->osc_client->cli_name = "osc";
 883         osc->osc_ldlm_client->cli_name = "ldlm";
 884
 885         MOD_INC_USE_COUNT;
 886         RETURN(0);
 887
 888  out_client:
 889         OBD_FREE(osc->osc_client, sizeof(*osc->osc_client));
 890  out_ns:
 891         ldlm_namespace_free(obddev->obd_namespace);
 892  out_conn:
 893         ptlrpc_put_connection(osc->osc_conn);
 894         return rc;
 895 }
 896
 897 static int osc_cleanup(struct obd_device * obddev)
 898 {
 899         struct osc_obd *osc = &obddev->u.osc;
 900
 901         ldlm_namespace_free(obddev->obd_namespace);
 902
 903         ptlrpc_cleanup_client(osc->osc_client);
 904         OBD_FREE(osc->osc_client, sizeof(*osc->osc_client));
 905         ptlrpc_cleanup_client(osc->osc_ldlm_client);
 906         OBD_FREE(osc->osc_ldlm_client, sizeof(*osc->osc_ldlm_client));
 907         ptlrpc_put_connection(osc->osc_conn);
 908
 909         MOD_DEC_USE_COUNT;
 910         return 0;
 911 }
 912
 913 static int osc_statfs(struct lustre_handle *conn, struct statfs *sfs)
 914 {
 915         struct ptlrpc_request *request;
 916         struct ptlrpc_client *cl;
 917         struct ptlrpc_connection *connection;
 918         struct lustre_handle *rconn;
 919         struct obd_statfs *osfs;
 920         int rc, size = sizeof(*osfs);
 921         ENTRY;
 922
 923         osc_con2cl(conn, &cl, &connection, &rconn);
 924         request = ptlrpc_prep_req2(cl, connection, rconn,
 925                                    OST_STATFS, 0, NULL, NULL);
 926         if (!request)
 927                 RETURN(-ENOMEM);
 928
 929         request->rq_replen = lustre_msg_size(1, &size);
 930
 931         rc = ptlrpc_queue_wait(request);
 932         rc = ptlrpc_check_status(request, rc);
 933         if (rc) {
 934                 CERROR("%s failed: rc = %d\n", __FUNCTION__, rc);
 935                 GOTO(out, rc);
 936         }
 937
 938         osfs = lustre_msg_buf(request->rq_repmsg, 0);
 939         obd_statfs_unpack(osfs, sfs);
 940
 941         EXIT;
 942  out:
 943         ptlrpc_free_req(request);
 944         return rc;
 945 }
 946
 947 struct obd_ops osc_obd_ops = {
 948         o_setup:        osc_setup,
 949         o_cleanup:      osc_cleanup,
 950         o_statfs:       osc_statfs,
 951         o_create:       osc_create,
 952         o_destroy:      osc_destroy,
 953         o_getattr:      osc_getattr,
 954         o_setattr:      osc_setattr,
 955         o_open:         osc_open,
 956         o_close:        osc_close,
 957         o_connect:      osc_connect,
 958         o_disconnect:   osc_disconnect,
 959         o_brw:          osc_brw,
 960         o_punch:        osc_punch,
 961         o_enqueue:      osc_enqueue,
 962         o_cancel:       osc_cancel
 963 };
 964
 965 static int __init osc_init(void)
 966 {
 967         return class_register_type(&osc_obd_ops, LUSTRE_OSC_NAME);
 968 }
 969
 970 static void __exit osc_exit(void)
 971 {
 972         class_unregister_type(LUSTRE_OSC_NAME);
 973 }
 974
 975 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
 976 MODULE_DESCRIPTION("Lustre Object Storage Client (OSC) v1.0");
 977 MODULE_LICENSE("GPL");
 978
 979 module_init(osc_init);
 980 module_exit(osc_exit);