lustre/ost/ost_handler.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  *  Copyright (C) 2001-2003 Cluster File Systems, Inc.
   5  *   Author: Peter J. Braam <braam@clusterfs.com>
   6  *   Author: Phil Schwan <phil@clusterfs.com>
   7  *
   8  *   This file is part of Lustre, http://www.lustre.org.
   9  *
  10  *   Lustre is free software; you can redistribute it and/or
  11  *   modify it under the terms of version 2 of the GNU General Public
  12  *   License as published by the Free Software Foundation.
  13  *
  14  *   Lustre is distributed in the hope that it will be useful,
  15  *   but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  17  *   GNU General Public License for more details.
  18  *
  19  *   You should have received a copy of the GNU General Public License
  20  *   along with Lustre; if not, write to the Free Software
  21  *   Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
  22  *
  23  *  Storage Target Handling functions
  24  *  Lustre Object Server Module (OST)
  25  *
  26  *  This server is single threaded at present (but can easily be multi
  27  *  threaded). For testing and management it is treated as an
  28  *  obd_device, although it does not export a full OBD method table
  29  *  (the requests are coming in over the wire, so object target
  30  *  modules do not have a full method table.)
  31  */
  32
  33 #define EXPORT_SYMTAB
  34 #define DEBUG_SUBSYSTEM S_OST
  35
  36 #include <linux/module.h>
  37 #include <linux/obd_ost.h>
  38 #include <linux/lustre_net.h>
  39 #include <linux/lustre_dlm.h>
  40 #include <linux/lustre_export.h>
  41 #include <linux/init.h>
  42 #include <linux/lprocfs_status.h>
  43 #include <linux/lustre_commit_confd.h>
  44 #include <portals/list.h>
  45
  46 void oti_init(struct obd_trans_info *oti, struct ptlrpc_request *req)
  47 {
  48         if (oti == NULL)
  49                 return;
  50         memset(oti, 0, sizeof *oti);
  51
  52         if (req->rq_repmsg && req->rq_reqmsg != 0)
  53                 oti->oti_transno = req->rq_repmsg->transno;
  54 }
  55
  56 void oti_to_request(struct obd_trans_info *oti, struct ptlrpc_request *req)
  57 {
  58         struct oti_req_ack_lock *ack_lock;
  59         int i;
  60
  61         if (oti == NULL)
  62                 return;
  63
  64         if (req->rq_repmsg)
  65                 req->rq_repmsg->transno = oti->oti_transno;
  66
  67         /* XXX 4 == entries in oti_ack_locks??? */
  68         for (ack_lock = oti->oti_ack_locks, i = 0; i < 4; i++, ack_lock++) {
  69                 if (!ack_lock->mode)
  70                         break;
  71                 memcpy(&req->rq_ack_locks[i].lock, &ack_lock->lock,
  72                        sizeof(req->rq_ack_locks[i].lock));
  73                 req->rq_ack_locks[i].mode = ack_lock->mode;
  74         }
  75 }
  76
  77 static int ost_destroy(struct ptlrpc_request *req, struct obd_trans_info *oti)
  78 {
  79         struct lustre_handle *conn = &req->rq_reqmsg->handle;
  80         struct ost_body *body;
  81         int rc, size = sizeof(*body);
  82         ENTRY;
  83
  84         body = lustre_swab_reqbuf(req, 0, sizeof(*body), lustre_swab_ost_body);
  85         if (body == NULL)
  86                 RETURN(-EFAULT);
  87
  88         rc = lustre_pack_msg(1, &size, NULL, &req->rq_replen, &req->rq_repmsg);
  89         if (rc)
  90                 RETURN(rc);
  91
  92         if (body->oa.o_valid & OBD_MD_FLCOOKIE)
  93                 oti->oti_logcookies = obdo_logcookie(&body->oa);
  94         req->rq_status = obd_destroy(conn, &body->oa, NULL, oti);
  95         RETURN(0);
  96 }
  97
  98 static int ost_getattr(struct ptlrpc_request *req)
  99 {
 100         struct lustre_handle *conn = (struct lustre_handle *)req->rq_reqmsg;
 101         struct ost_body *body, *repbody;
 102         int rc, size = sizeof(*body);
 103         ENTRY;
 104
 105         body = lustre_swab_reqbuf(req, 0, sizeof(*body), lustre_swab_ost_body);
 106         if (body == NULL)
 107                 RETURN(-EFAULT);
 108
 109         rc = lustre_pack_msg(1, &size, NULL, &req->rq_replen, &req->rq_repmsg);
 110         if (rc)
 111                 RETURN(rc);
 112
 113         repbody = lustre_msg_buf (req->rq_repmsg, 0, sizeof(*repbody));
 114         memcpy(&repbody->oa, &body->oa, sizeof(body->oa));
 115         req->rq_status = obd_getattr(conn, &repbody->oa, NULL);
 116         RETURN(0);
 117 }
 118
 119 static int ost_statfs(struct ptlrpc_request *req)
 120 {
 121         struct obd_statfs *osfs;
 122         int rc, size = sizeof(*osfs);
 123         ENTRY;
 124
 125         rc = lustre_pack_msg(1, &size, NULL, &req->rq_replen, &req->rq_repmsg);
 126         if (rc)
 127                 RETURN(rc);
 128
 129         osfs = lustre_msg_buf(req->rq_repmsg, 0, sizeof(*osfs));
 130
 131         req->rq_status = obd_statfs(req->rq_export->exp_obd, osfs, jiffies-HZ);
 132         if (req->rq_status != 0)
 133                 CERROR("ost: statfs failed: rc %d\n", req->rq_status);
 134
 135         RETURN(0);
 136 }
 137
 138 static int ost_syncfs(struct ptlrpc_request *req)
 139 {
 140         struct obd_statfs *osfs;
 141         int rc, size = sizeof(*osfs);
 142         ENTRY;
 143
 144         rc = lustre_pack_msg(0, &size, NULL, &req->rq_replen, &req->rq_repmsg);
 145         if (rc)
 146                 RETURN(rc);
 147
 148         rc = obd_syncfs(req->rq_export);
 149         if (rc) {
 150                 CERROR("ost: syncfs failed: rc %d\n", rc);
 151                 req->rq_status = rc;
 152                 RETURN(rc);
 153         }
 154
 155         RETURN(0);
 156 }
 157
 158 static int ost_open(struct ptlrpc_request *req, struct obd_trans_info *oti)
 159 {
 160         struct lustre_handle *conn = (struct lustre_handle *)req->rq_reqmsg;
 161         struct ost_body *body, *repbody;
 162         int rc, size = sizeof(*repbody);
 163         ENTRY;
 164
 165         body = lustre_swab_reqbuf(req, 0, sizeof(*body), lustre_swab_ost_body);
 166         if (body == NULL)
 167                 RETURN(-EFAULT);
 168
 169         rc = lustre_pack_msg(1, &size, NULL, &req->rq_replen, &req->rq_repmsg);
 170         if (rc)
 171                 RETURN(rc);
 172
 173         repbody = lustre_msg_buf (req->rq_repmsg, 0, sizeof(*repbody));
 174         memcpy(&repbody->oa, &body->oa, sizeof(body->oa));
 175         req->rq_status = obd_open(conn, &repbody->oa, NULL, oti, NULL);
 176         RETURN(0);
 177 }
 178
 179 static int ost_close(struct ptlrpc_request *req, struct obd_trans_info *oti)
 180 {
 181         struct lustre_handle *conn = (struct lustre_handle *)req->rq_reqmsg;
 182         struct ost_body *body, *repbody;
 183         int rc, size = sizeof(*repbody);
 184         ENTRY;
 185
 186         body = lustre_swab_reqbuf(req, 0, sizeof(*body), lustre_swab_ost_body);
 187         if (body == NULL)
 188                 RETURN(-EFAULT);
 189
 190         rc = lustre_pack_msg(1, &size, NULL, &req->rq_replen, &req->rq_repmsg);
 191         if (rc)
 192                 RETURN(rc);
 193
 194         repbody = lustre_msg_buf(req->rq_repmsg, 0, sizeof(*repbody));
 195         memcpy(&repbody->oa, &body->oa, sizeof(body->oa));
 196         req->rq_status = obd_close(conn, &repbody->oa, NULL, oti);
 197         RETURN(0);
 198 }
 199
 200 static int ost_create(struct ptlrpc_request *req, struct obd_trans_info *oti)
 201 {
 202         struct lustre_handle *conn = (struct lustre_handle *)req->rq_reqmsg;
 203         struct ost_body *body, *repbody;
 204         int rc, size = sizeof(*repbody);
 205         ENTRY;
 206
 207         body = lustre_swab_reqbuf(req, 0, sizeof(*body), lustre_swab_ost_body);
 208         if (body == NULL)
 209                 RETURN(-EFAULT);
 210
 211         rc = lustre_pack_msg(1, &size, NULL, &req->rq_replen, &req->rq_repmsg);
 212         if (rc)
 213                 RETURN(rc);
 214
 215         repbody = lustre_msg_buf (req->rq_repmsg, 0, sizeof(*repbody));
 216         memcpy(&repbody->oa, &body->oa, sizeof(body->oa));
 217         oti->oti_logcookies = obdo_logcookie(&repbody->oa);
 218         req->rq_status = obd_create(conn, &repbody->oa, NULL, oti);
 219         //obd_log_cancel(conn, NULL, 1, oti->oti_logcookies, 0);
 220         RETURN(0);
 221 }
 222
 223 static int ost_punch(struct ptlrpc_request *req, struct obd_trans_info *oti)
 224 {
 225         struct lustre_handle *conn = (struct lustre_handle *)req->rq_reqmsg;
 226         struct ost_body *body, *repbody;
 227         int rc, size = sizeof(*repbody);
 228         ENTRY;
 229
 230         body = lustre_swab_reqbuf(req, 0, sizeof(*body), lustre_swab_ost_body);
 231         if (body == NULL)
 232                 RETURN(-EFAULT);
 233
 234         if ((body->oa.o_valid & (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS)) !=
 235             (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS))
 236                 RETURN(-EINVAL);
 237
 238         rc = lustre_pack_msg(1, &size, NULL, &req->rq_replen, &req->rq_repmsg);
 239         if (rc)
 240                 RETURN(rc);
 241
 242         repbody = lustre_msg_buf(req->rq_repmsg, 0, sizeof(*repbody));
 243         memcpy(&repbody->oa, &body->oa, sizeof(body->oa));
 244         req->rq_status = obd_punch(conn, &repbody->oa, NULL, repbody->oa.o_size,
 245                                    repbody->oa.o_blocks, oti);
 246         RETURN(0);
 247 }
 248
 249 static int ost_setattr(struct ptlrpc_request *req, struct obd_trans_info *oti)
 250 {
 251         struct lustre_handle *conn = &req->rq_reqmsg->handle;
 252         struct ost_body *body, *repbody;
 253         int rc, size = sizeof(*repbody);
 254         ENTRY;
 255
 256         body = lustre_swab_reqbuf(req, 0, sizeof(*body), lustre_swab_ost_body);
 257         if (body == NULL)
 258                 RETURN(-EFAULT);
 259
 260         rc = lustre_pack_msg(1, &size, NULL, &req->rq_replen, &req->rq_repmsg);
 261         if (rc)
 262                 RETURN(rc);
 263
 264         repbody = lustre_msg_buf(req->rq_repmsg, 0, sizeof(*repbody));
 265         memcpy(&repbody->oa, &body->oa, sizeof(body->oa));
 266
 267         req->rq_status = obd_setattr(conn, &repbody->oa, NULL, oti);
 268         RETURN(0);
 269 }
 270
 271 static int ost_bulk_timeout(void *data)
 272 {
 273         ENTRY;
 274         /* We don't fail the connection here, because having the export
 275          * killed makes the (vital) call to commitrw very sad.
 276          */
 277         RETURN(1);
 278 }
 279
 280 static int get_per_page_niobufs(struct obd_ioobj *ioo, int nioo,
 281                                 struct niobuf_remote *rnb, int nrnb,
 282                                 struct niobuf_remote **pp_rnbp)
 283 {
 284         /* Copy a remote niobuf, splitting it into page-sized chunks
 285          * and setting ioo[i].ioo_bufcnt accordingly */
 286         struct niobuf_remote *pp_rnb;
 287         int   i;
 288         int   j;
 289         int   page;
 290         int   rnbidx = 0;
 291         int   npages = 0;
 292
 293         /* first count and check the number of pages required */
 294         for (i = 0; i < nioo; i++)
 295                 for (j = 0; j < ioo->ioo_bufcnt; j++, rnbidx++) {
 296                         obd_off offset = rnb[rnbidx].offset;
 297                         obd_off p0 = offset >> PAGE_SHIFT;
 298                         obd_off pn = (offset + rnb[rnbidx].len - 1)>>PAGE_SHIFT;
 299
 300                         LASSERT(rnbidx < nrnb);
 301
 302                         npages += (pn + 1 - p0);
 303
 304                         if (rnb[rnbidx].len == 0) {
 305                                 CERROR("zero len BRW: obj %d objid "LPX64
 306                                        " buf %u\n", i, ioo[i].ioo_id, j);
 307                                 return -EINVAL;
 308                         }
 309                         if (j > 0 &&
 310                             rnb[rnbidx].offset <= rnb[rnbidx-1].offset) {
 311                                 CERROR("unordered BRW: obj %d objid "LPX64
 312                                        " buf %u offset "LPX64" <= "LPX64"\n",
 313                                        i, ioo[i].ioo_id, j, rnb[rnbidx].offset,
 314                                        rnb[rnbidx].offset);
 315                                 return -EINVAL;
 316                         }
 317                 }
 318
 319         LASSERT(rnbidx == nrnb);
 320
 321         if (npages == nrnb) {       /* all niobufs are for single pages */
 322                 *pp_rnbp = rnb;
 323                 return npages;
 324         }
 325
 326         OBD_ALLOC(pp_rnb, sizeof(*pp_rnb) * npages);
 327         if (pp_rnb == NULL)
 328                 return -ENOMEM;
 329
 330         /* now do the actual split */
 331         page = rnbidx = 0;
 332         for (i = 0; i < nioo; i++) {
 333                 int  obj_pages = 0;
 334
 335                 for (j = 0; j < ioo[i].ioo_bufcnt; j++, rnbidx++) {
 336                         obd_off off = rnb[rnbidx].offset;
 337                         int     nob = rnb[rnbidx].len;
 338
 339                         LASSERT(rnbidx < nrnb);
 340                         do {
 341                                 obd_off  poff = off & (PAGE_SIZE - 1);
 342                                 int      pnob = (poff + nob > PAGE_SIZE) ?
 343                                                 PAGE_SIZE - poff : nob;
 344
 345                                 LASSERT(page < npages);
 346                                 pp_rnb[page].len = pnob;
 347                                 pp_rnb[page].offset = off;
 348                                 pp_rnb[page].flags = rnb->flags;
 349
 350                                 CDEBUG(D_PAGE, "   obj %d id "LPX64
 351                                        "page %d(%d) "LPX64" for %d\n",
 352                                        i, ioo[i].ioo_id, obj_pages, page,
 353                                        pp_rnb[page].offset, pp_rnb[page].len);
 354                                 page++;
 355                                 obj_pages++;
 356
 357                                 off += pnob;
 358                                 nob -= pnob;
 359                         } while (nob > 0);
 360                         LASSERT(nob == 0);
 361                 }
 362                 ioo[i].ioo_bufcnt = obj_pages;
 363         }
 364         LASSERT(page == npages);
 365
 366         *pp_rnbp = pp_rnb;
 367         return npages;
 368 }
 369
 370 static void free_per_page_niobufs (int npages, struct niobuf_remote *pp_rnb,
 371                                    struct niobuf_remote *rnb)
 372 {
 373         if (pp_rnb == rnb)                      /* didn't allocate above */
 374                 return;
 375
 376         OBD_FREE(pp_rnb, sizeof(*pp_rnb) * npages);
 377 }
 378
 379 #if CHECKSUM_BULK
 380 __u64 ost_checksum_bulk (struct ptlrpc_bulk_desc *desc)
 381 {
 382         __u64             cksum = 0;
 383         struct ptlrpc_bulk_page *bp;
 384
 385         list_for_each_entry(bp, &desc->bd_page_list, bp_link) {
 386                 ost_checksum(&cksum, kmap(bp->bp_page) + bp->bp_pageoffset,
 387                              bp->bp_buflen);
 388                 kunmap(bp->bp_page);
 389         }
 390 }
 391 #endif
 392
 393 static int ost_brw_read(struct ptlrpc_request *req)
 394 {
 395         struct ptlrpc_bulk_desc *desc;
 396         struct niobuf_remote    *remote_nb;
 397         struct niobuf_remote    *pp_rnb;
 398         struct niobuf_local     *local_nb;
 399         struct obd_ioobj        *ioo;
 400         struct ost_body         *body, *repbody;
 401         struct l_wait_info       lwi;
 402         struct obd_trans_info    oti = { 0 };
 403         int                      size[1] = { sizeof(*body) };
 404         int                      comms_error = 0;
 405         int                      niocount;
 406         int                      npages;
 407         int                      nob = 0;
 408         int                      rc;
 409         int                      i;
 410         ENTRY;
 411
 412         if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_READ_BULK))
 413                 GOTO(out, rc = -EIO);
 414
 415         body = lustre_swab_reqbuf(req, 0, sizeof(*body), lustre_swab_ost_body);
 416         if (body == NULL) {
 417                 CERROR("Missing/short ost_body\n");
 418                 GOTO(out, rc = -EFAULT);
 419         }
 420
 421         /* BUG 974: when we send back cache grants, don't clear this flag */
 422         body->oa.o_valid &= ~OBD_MD_FLRDEV;
 423
 424         ioo = lustre_swab_reqbuf(req, 1, sizeof(*ioo), lustre_swab_obd_ioobj);
 425         if (ioo == NULL) {
 426                 CERROR("Missing/short ioobj\n");
 427                 GOTO(out, rc = -EFAULT);
 428         }
 429
 430         niocount = ioo->ioo_bufcnt;
 431         remote_nb = lustre_swab_reqbuf(req, 2, niocount * sizeof(*remote_nb),
 432                                        lustre_swab_niobuf_remote);
 433         if (remote_nb == NULL) {
 434                 CERROR("Missing/short niobuf\n");
 435                 GOTO(out, rc = -EFAULT);
 436         }
 437         if (lustre_msg_swabbed(req->rq_reqmsg)) { /* swab remaining niobufs */
 438                 for (i = 1; i < niocount; i++)
 439                         lustre_swab_niobuf_remote (&remote_nb[i]);
 440         }
 441
 442         size[0] = sizeof(*body);
 443         rc = lustre_pack_msg(1, size, NULL, &req->rq_replen, &req->rq_repmsg);
 444         if (rc)
 445                 GOTO(out, rc);
 446
 447         /* FIXME all niobuf splitting should be done in obdfilter if needed */
 448         /* CAVEAT EMPTOR this sets ioo->ioo_bufcnt to # pages */
 449         npages = get_per_page_niobufs(ioo, 1, remote_nb, niocount, &pp_rnb);
 450         if (npages < 0)
 451                 GOTO(out, rc = npages);
 452
 453         OBD_ALLOC(local_nb, sizeof(*local_nb) * npages);
 454         if (local_nb == NULL)
 455                 GOTO(out_pp_rnb, rc = -ENOMEM);
 456
 457         desc = ptlrpc_prep_bulk_exp(req, BULK_PUT_SOURCE, OST_BULK_PORTAL);
 458         if (desc == NULL)
 459                 GOTO(out_local, rc = -ENOMEM);
 460
 461         rc = obd_preprw(OBD_BRW_READ, req->rq_export, &body->oa, 1,
 462                         ioo, npages, pp_rnb, local_nb, &oti);
 463         if (rc != 0)
 464                 GOTO(out_bulk, rc);
 465
 466         nob = 0;
 467         for (i = 0; i < npages; i++) {
 468                 int page_rc = local_nb[i].rc;
 469
 470                 if (page_rc < 0) {              /* error */
 471                         rc = page_rc;
 472                         break;
 473                 }
 474
 475                 LASSERT(page_rc <= pp_rnb[i].len);
 476                 nob += page_rc;
 477                 if (page_rc != 0) {             /* some data! */
 478                         LASSERT (local_nb[i].page != NULL);
 479                         rc = ptlrpc_prep_bulk_page(desc, local_nb[i].page,
 480                                                    pp_rnb[i].offset& ~PAGE_MASK,
 481                                                    page_rc);
 482                         if (rc != 0)
 483                                 break;
 484                 }
 485
 486                 if (page_rc != pp_rnb[i].len) { /* short read */
 487                         /* All subsequent pages should be 0 */
 488                         while(++i < npages)
 489                                 LASSERT(local_nb[i].rc == 0);
 490                         break;
 491                 }
 492         }
 493
 494         if (rc == 0) {
 495                 rc = ptlrpc_bulk_put(desc);
 496                 if (rc == 0) {
 497                         lwi = LWI_TIMEOUT(obd_timeout * HZ, ost_bulk_timeout,
 498                                           desc);
 499                         rc = l_wait_event(desc->bd_waitq,
 500                                           ptlrpc_bulk_complete(desc), &lwi);
 501                         if (rc) {
 502                                 LASSERT(rc == -ETIMEDOUT);
 503                                 CERROR ("timeout waiting for bulk PUT\n");
 504                                 ptlrpc_abort_bulk(desc);
 505                         }
 506                 } else {
 507                         CERROR("ptlrpc_bulk_put failed RC: %d\n", rc);
 508                 }
 509                 comms_error = rc != 0;
 510         }
 511
 512         /* Must commit after prep above in all cases */
 513         rc = obd_commitrw(OBD_BRW_READ, req->rq_export, &body->oa, 1,
 514                           ioo, npages, local_nb, &oti);
 515
 516         repbody = lustre_msg_buf(req->rq_repmsg, 0, sizeof(*repbody));
 517         memcpy(&repbody->oa, &body->oa, sizeof(repbody->oa));
 518
 519 #if CHECKSUM_BULK
 520         if (rc == 0) {
 521                 repbody->oa.o_rdev = ost_checksum_bulk(desc);
 522                 repbody->oa.o_valid |= OBD_MD_FLCKSUM;
 523         }
 524 #endif
 525
 526  out_bulk:
 527         ptlrpc_free_bulk(desc);
 528  out_local:
 529         OBD_FREE(local_nb, sizeof(*local_nb) * npages);
 530  out_pp_rnb:
 531         free_per_page_niobufs(npages, pp_rnb, remote_nb);
 532  out:
 533         LASSERT(rc <= 0);
 534         if (rc == 0) {
 535                 req->rq_status = nob;
 536                 ptlrpc_reply(req);
 537         } else if (!comms_error) {
 538                 /* only reply if comms OK */
 539                 req->rq_status = rc;
 540                 ptlrpc_error(req);
 541         } else {
 542                 if (req->rq_repmsg != NULL) {
 543                         /* reply out callback would free */
 544                         OBD_FREE(req->rq_repmsg, req->rq_replen);
 545                 }
 546                 CERROR("bulk IO comms error: evicting %s@%s nid "LPU64"\n",
 547                        req->rq_export->exp_client_uuid.uuid,
 548                        req->rq_connection->c_remote_uuid.uuid,
 549                        req->rq_connection->c_peer.peer_nid);
 550                 ptlrpc_fail_export(req->rq_export);
 551         }
 552
 553         RETURN(rc);
 554 }
 555
 556 static int ost_brw_write(struct ptlrpc_request *req, struct obd_trans_info *oti)
 557 {
 558         struct ptlrpc_bulk_desc *desc;
 559         struct niobuf_remote    *remote_nb;
 560         struct niobuf_remote    *pp_rnb;
 561         struct niobuf_local     *local_nb;
 562         struct obd_ioobj        *ioo;
 563         struct ost_body         *body, *repbody;
 564         struct l_wait_info       lwi;
 565         __u32                   *rcs;
 566         int                      size[2] = { sizeof(*body) };
 567         int                      objcount, niocount, npages;
 568         int                      comms_error = 0;
 569         int                      rc, rc2, swab, i, j;
 570         ENTRY;
 571
 572         if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_WRITE_BULK))
 573                 GOTO(out, rc = -EIO);
 574
 575         /* pause before transaction has been started */
 576         OBD_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_BULK | OBD_FAIL_ONCE,
 577                          obd_timeout +1);
 578
 579         swab = lustre_msg_swabbed(req->rq_reqmsg);
 580         body = lustre_swab_reqbuf(req, 0, sizeof(*body), lustre_swab_ost_body);
 581         if (body == NULL) {
 582                 CERROR("Missing/short ost_body\n");
 583                 GOTO(out, rc = -EFAULT);
 584         }
 585
 586         /* BUG 974: when we send back cache grants, don't clear this flag */
 587         body->oa.o_valid &= ~OBD_MD_FLRDEV;
 588
 589         LASSERT_REQSWAB(req, 1);
 590         objcount = req->rq_reqmsg->buflens[1] / sizeof(*ioo);
 591         if (objcount == 0) {
 592                 CERROR("Missing/short ioobj\n");
 593                 GOTO(out, rc = -EFAULT);
 594         }
 595         ioo = lustre_msg_buf (req->rq_reqmsg, 1, objcount * sizeof(*ioo));
 596         LASSERT (ioo != NULL);
 597         for (niocount = i = 0; i < objcount; i++) {
 598                 if (swab)
 599                         lustre_swab_obd_ioobj (&ioo[i]);
 600                 if (ioo[i].ioo_bufcnt == 0) {
 601                         CERROR("ioo[%d] has zero bufcnt\n", i);
 602                         GOTO(out, rc = -EFAULT);
 603                 }
 604                 niocount += ioo[i].ioo_bufcnt;
 605         }
 606
 607         remote_nb = lustre_swab_reqbuf(req, 2, niocount * sizeof(*remote_nb),
 608                                        lustre_swab_niobuf_remote);
 609         if (remote_nb == NULL) {
 610                 CERROR("Missing/short niobuf\n");
 611                 GOTO(out, rc = -EFAULT);
 612         }
 613         if (swab) {                             /* swab the remaining niobufs */
 614                 for (i = 1; i < niocount; i++)
 615                         lustre_swab_niobuf_remote (&remote_nb[i]);
 616         }
 617
 618         size[1] = niocount * sizeof(*rcs);
 619         rc = lustre_pack_msg(2, size, NULL, &req->rq_replen,
 620                              &req->rq_repmsg);
 621         if (rc != 0)
 622                 GOTO(out, rc);
 623         rcs = lustre_msg_buf(req->rq_repmsg, 1, niocount * sizeof(*rcs));
 624
 625         /* FIXME all niobuf splitting should be done in obdfilter if needed */
 626         /* CAVEAT EMPTOR this sets ioo->ioo_bufcnt to # pages */
 627         npages = get_per_page_niobufs(ioo, objcount,remote_nb,niocount,&pp_rnb);
 628         if (npages < 0)
 629                 GOTO(out, rc = npages);
 630
 631         OBD_ALLOC(local_nb, sizeof(*local_nb) * npages);
 632         if (local_nb == NULL)
 633                 GOTO(out_pp_rnb, rc = -ENOMEM);
 634
 635         desc = ptlrpc_prep_bulk_exp(req, BULK_GET_SINK, OST_BULK_PORTAL);
 636         if (desc == NULL)
 637                 GOTO(out_local, rc = -ENOMEM);
 638
 639         rc = obd_preprw(OBD_BRW_WRITE, req->rq_export, &body->oa, objcount,
 640                         ioo, npages, pp_rnb, local_nb, oti);
 641         if (rc != 0)
 642                 GOTO(out_bulk, rc);
 643
 644         /* NB Having prepped, we must commit... */
 645
 646         for (i = 0; i < npages; i++) {
 647                 rc = ptlrpc_prep_bulk_page(desc, local_nb[i].page,
 648                                            pp_rnb[i].offset & (PAGE_SIZE - 1),
 649                                            pp_rnb[i].len);
 650                 if (rc != 0)
 651                         break;
 652         }
 653
 654         if (rc == 0) {
 655                 rc = ptlrpc_bulk_get(desc);
 656                 if (rc == 0) {
 657                         lwi = LWI_TIMEOUT(obd_timeout * HZ, ost_bulk_timeout,
 658                                           desc);
 659                         rc = l_wait_event(desc->bd_waitq,
 660                                           ptlrpc_bulk_complete(desc), &lwi);
 661                         if (rc) {
 662                                 LASSERT(rc == -ETIMEDOUT);
 663                                 CERROR("timeout waiting for bulk GET\n");
 664                                 ptlrpc_abort_bulk(desc);
 665                         }
 666                 } else {
 667                         CERROR("ptlrpc_bulk_get failed RC: %d\n", rc);
 668                 }
 669                 comms_error = rc != 0;
 670         }
 671
 672         repbody = lustre_msg_buf(req->rq_repmsg, 0, sizeof(*repbody));
 673         memcpy(&repbody->oa, &body->oa, sizeof(repbody->oa));
 674
 675 #if CHECKSUM_BULK
 676         if (rc == 0 && (body->oa.o_valid & OBD_MD_FLCKSUM) != 0) {
 677                 static int cksum_counter;
 678                 __u64 client_cksum = body->oa.o_rdev;
 679                 __u64 cksum = ost_checksum_bulk(desc);
 680
 681                 if (client_cksum != cksum) {
 682                         CERROR("Bad checksum: client "LPX64", server "LPX64
 683                                ", client NID "LPX64"\n", client_cksum, cksum,
 684                                req->rq_connection->c_peer.peer_nid);
 685                         cksum_counter = 1;
 686                         repbody->oa.o_rdev = cksum;
 687                 } else {
 688                         cksum_counter++;
 689                         if ((cksum_counter & (-cksum_counter)) == cksum_counter)
 690                                 CERROR("Checksum %d from "LPX64": "LPX64" OK\n",
 691                                         cksum_counter,
 692                                         req->rq_connection->c_peer.peer_nid,
 693                                         cksum);
 694                 }
 695         }
 696 #endif
 697         /* Must commit after prep above in all cases */
 698         rc2 = obd_commitrw(OBD_BRW_WRITE, req->rq_export, &repbody->oa,
 699                            objcount, ioo, npages, local_nb, oti);
 700
 701         if (rc == 0) {
 702                 /* set per-requested niobuf return codes */
 703                 for (i = j = 0; i < niocount; i++) {
 704                         int nob = remote_nb[i].len;
 705
 706                         rcs[i] = 0;
 707                         do {
 708                                 LASSERT(j < npages);
 709                                 if (local_nb[j].rc < 0)
 710                                         rcs[i] = local_nb[j].rc;
 711                                 nob -= pp_rnb[j].len;
 712                                 j++;
 713                         } while (nob > 0);
 714                         LASSERT(nob == 0);
 715                 }
 716                 LASSERT(j == npages);
 717         }
 718         if (rc == 0)
 719                 rc = rc2;
 720
 721  out_bulk:
 722         ptlrpc_free_bulk(desc);
 723  out_local:
 724         OBD_FREE(local_nb, sizeof(*local_nb) * npages);
 725  out_pp_rnb:
 726         free_per_page_niobufs(npages, pp_rnb, remote_nb);
 727  out:
 728         if (rc == 0) {
 729                 oti_to_request(oti, req);
 730                 rc = ptlrpc_reply(req);
 731         } else if (!comms_error) {
 732                 /* Only reply if there was no comms problem with bulk */
 733                 req->rq_status = rc;
 734                 ptlrpc_error(req);
 735         } else {
 736                 if (req->rq_repmsg != NULL) {
 737                         /* reply out callback would free */
 738                         OBD_FREE (req->rq_repmsg, req->rq_replen);
 739                 }
 740                 CERROR("bulk IO comms error: evicting %s@%s nid "LPU64"\n",
 741                        req->rq_export->exp_client_uuid.uuid,
 742                        req->rq_connection->c_remote_uuid.uuid,
 743                        req->rq_connection->c_peer.peer_nid);
 744                 ptlrpc_fail_export(req->rq_export);
 745         }
 746         RETURN(rc);
 747 }
 748
 749 static int ost_san_brw(struct ptlrpc_request *req, int cmd)
 750 {
 751         struct niobuf_remote *remote_nb, *res_nb;
 752         struct obd_ioobj *ioo;
 753         struct ost_body *body, *repbody;
 754         int rc, i, j, objcount, niocount, size[2] = {sizeof(*body)};
 755         int n;
 756         int swab;
 757         ENTRY;
 758
 759         /* XXX not set to use latest protocol */
 760
 761         swab = lustre_msg_swabbed(req->rq_reqmsg);
 762         body = lustre_swab_reqbuf(req, 0, sizeof(*body), lustre_swab_ost_body);
 763         if (body == NULL) {
 764                 CERROR("Missing/short ost_body\n");
 765                 GOTO(out, rc = -EFAULT);
 766         }
 767
 768         ioo = lustre_swab_reqbuf(req, 1, sizeof(*ioo), lustre_swab_obd_ioobj);
 769         if (ioo == NULL) {
 770                 CERROR("Missing/short ioobj\n");
 771                 GOTO(out, rc = -EFAULT);
 772         }
 773         objcount = req->rq_reqmsg->buflens[1] / sizeof(*ioo);
 774         niocount = ioo[0].ioo_bufcnt;
 775         for (i = 1; i < objcount; i++) {
 776                 if (swab)
 777                         lustre_swab_obd_ioobj (&ioo[i]);
 778                 niocount += ioo[i].ioo_bufcnt;
 779         }
 780
 781         remote_nb = lustre_swab_reqbuf(req, 2, niocount * sizeof(*remote_nb),
 782                                        lustre_swab_niobuf_remote);
 783         if (remote_nb == NULL) {
 784                 CERROR("Missing/short niobuf\n");
 785                 GOTO(out, rc = -EFAULT);
 786         }
 787         if (swab) {                             /* swab the remaining niobufs */
 788                 for (i = 1; i < niocount; i++)
 789                         lustre_swab_niobuf_remote (&remote_nb[i]);
 790         }
 791
 792         for (i = n = 0; i < objcount; i++) {
 793                 for (j = 0; j < ioo[i].ioo_bufcnt; j++, n++) {
 794                         if (remote_nb[n].len == 0) {
 795                                 CERROR("zero len BRW: objid "LPX64" buf %u\n",
 796                                        ioo[i].ioo_id, j);
 797                                 GOTO(out, rc = -EINVAL);
 798                         }
 799                         if (j && remote_nb[n].offset <= remote_nb[n-1].offset) {
 800                                 CERROR("unordered BRW: objid "LPX64
 801                                        " buf %u offset "LPX64" <= "LPX64"\n",
 802                                        ioo[i].ioo_id, j, remote_nb[n].offset,
 803                                        remote_nb[n-1].offset);
 804                                 GOTO(out, rc = -EINVAL);
 805                         }
 806                 }
 807         }
 808
 809         size[1] = niocount * sizeof(*remote_nb);
 810         rc = lustre_pack_msg(2, size, NULL, &req->rq_replen, &req->rq_repmsg);
 811         if (rc)
 812                 GOTO(out, rc);
 813
 814         req->rq_status = obd_san_preprw(cmd, req->rq_export, &body->oa,
 815                                         objcount, ioo, niocount, remote_nb);
 816
 817         if (req->rq_status)
 818                 GOTO(out, rc = 0);
 819
 820         repbody = lustre_msg_buf(req->rq_repmsg, 0, sizeof(*repbody));
 821         memcpy(&repbody->oa, &body->oa, sizeof(body->oa));
 822
 823         res_nb = lustre_msg_buf(req->rq_repmsg, 1, size[1]);
 824         memcpy(res_nb, remote_nb, size[1]);
 825         rc = 0;
 826 out:
 827         if (rc) {
 828                 OBD_FREE(req->rq_repmsg, req->rq_replen);
 829                 req->rq_repmsg = NULL;
 830                 req->rq_status = rc;
 831                 ptlrpc_error(req);
 832         } else
 833                 ptlrpc_reply(req);
 834
 835         return rc;
 836 }
 837
 838 static int ost_log_cancel(struct ptlrpc_request *req)
 839 {
 840         struct lustre_handle *conn;
 841         struct llog_cookie *logcookies;
 842         int num_cookies, rc = 0;
 843         ENTRY;
 844
 845         logcookies = lustre_msg_buf(req->rq_reqmsg, 0, sizeof(*logcookies));
 846         if (logcookies == NULL) {
 847                 DEBUG_REQ(D_HA, req, "no cookies sent");
 848                 RETURN(-EFAULT);
 849         }
 850         num_cookies = req->rq_reqmsg->buflens[0] / sizeof(*logcookies);
 851
 852         /* workaround until we don't need to send replies */
 853         rc = lustre_pack_msg(0, NULL, NULL, &req->rq_replen, &req->rq_repmsg);
 854         if (rc)
 855                 RETURN(rc);
 856         req->rq_repmsg->status = 0;
 857         /* end workaround */
 858
 859         conn = (struct lustre_handle *)&req->rq_reqmsg->handle;
 860         rc = obd_log_cancel(conn, NULL, num_cookies, logcookies, 0);
 861
 862         RETURN(rc);
 863 }
 864
 865 static int ost_set_info(struct ptlrpc_request *req)
 866 {
 867         struct lustre_handle *conn;
 868         char *key;
 869         int keylen, rc = 0;
 870         ENTRY;
 871
 872         key = lustre_msg_buf(req->rq_reqmsg, 0, 1);
 873         if (key == NULL) {
 874                 DEBUG_REQ(D_HA, req, "no set_info key");
 875                 RETURN(-EFAULT);
 876         }
 877         keylen = req->rq_reqmsg->buflens[0];
 878
 879         rc = lustre_pack_msg(0, NULL, NULL, &req->rq_replen, &req->rq_repmsg);
 880         if (rc)
 881                 RETURN(rc);
 882
 883         conn = (struct lustre_handle *)&req->rq_reqmsg->handle;
 884         rc = obd_set_info(conn, keylen, key, 0, NULL);
 885         req->rq_repmsg->status = 0;
 886         RETURN(rc);
 887 }
 888
 889 static int filter_recovery_request(struct ptlrpc_request *req,
 890                                    struct obd_device *obd, int *process)
 891 {
 892         switch (req->rq_reqmsg->opc) {
 893         case OST_CONNECT: /* This will never get here, but for completeness. */
 894         case OST_DISCONNECT:
 895                *process = 1;
 896                RETURN(0);
 897
 898         case OBD_PING:
 899         case OST_CLOSE:
 900         case OST_CREATE:
 901         case OST_DESTROY:
 902         case OST_OPEN:
 903         case OST_PUNCH:
 904         case OST_SETATTR:
 905         case OST_SYNCFS:
 906         case OST_WRITE:
 907         case OBD_LOG_CANCEL:
 908         case LDLM_ENQUEUE:
 909                 *process = target_queue_recovery_request(req, obd);
 910                 RETURN(0);
 911
 912         default:
 913                 DEBUG_REQ(D_ERROR, req, "not permitted during recovery");
 914                 *process = 0;
 915                 /* XXX what should we set rq_status to here? */
 916                 req->rq_status = -EAGAIN;
 917                 RETURN(ptlrpc_error(req));
 918         }
 919 }
 920
 921
 922
 923 static int ost_handle(struct ptlrpc_request *req)
 924 {
 925         struct obd_trans_info trans_info = { 0, };
 926         struct obd_trans_info *oti = &trans_info;
 927         int should_process, fail = OBD_FAIL_OST_ALL_REPLY_NET, rc = 0;
 928         ENTRY;
 929
 930         /* XXX identical to MDS */
 931         if (req->rq_reqmsg->opc != OST_CONNECT) {
 932                 struct obd_device *obd;
 933                 int abort_recovery, recovering;
 934
 935                 if (req->rq_export == NULL) {
 936                         CDEBUG(D_HA, "operation %d on unconnected OST\n",
 937                                req->rq_reqmsg->opc);
 938                         req->rq_status = -ENOTCONN;
 939                         GOTO(out, rc = -ENOTCONN);
 940                 }
 941
 942                 obd = req->rq_export->exp_obd;
 943
 944                 /* Check for aborted recovery. */
 945                 spin_lock_bh(&obd->obd_processing_task_lock);
 946                 abort_recovery = obd->obd_abort_recovery;
 947                 recovering = obd->obd_recovering;
 948                 spin_unlock_bh(&obd->obd_processing_task_lock);
 949                 if (abort_recovery) {
 950                         target_abort_recovery(obd);
 951                 } else if (recovering) {
 952                         rc = filter_recovery_request(req, obd, &should_process);
 953                         if (rc || !should_process)
 954                                 RETURN(rc);
 955                 }
 956         }
 957
 958         if (strcmp(req->rq_obd->obd_type->typ_name, "ost") != 0)
 959                 GOTO(out, rc = -EINVAL);
 960
 961         oti_init(oti, req);
 962
 963         switch (req->rq_reqmsg->opc) {
 964         case OST_CONNECT:
 965                 CDEBUG(D_INODE, "connect\n");
 966                 OBD_FAIL_RETURN(OBD_FAIL_OST_CONNECT_NET, 0);
 967                 rc = target_handle_connect(req, ost_handle);
 968                 break;
 969         case OST_DISCONNECT:
 970                 CDEBUG(D_INODE, "disconnect\n");
 971                 OBD_FAIL_RETURN(OBD_FAIL_OST_DISCONNECT_NET, 0);
 972                 rc = target_handle_disconnect(req);
 973                 break;
 974         case OST_CREATE:
 975                 CDEBUG(D_INODE, "create\n");
 976                 OBD_FAIL_RETURN(OBD_FAIL_OST_CREATE_NET, 0);
 977                 rc = ost_create(req, oti);
 978                 break;
 979         case OST_DESTROY:
 980                 CDEBUG(D_INODE, "destroy\n");
 981                 OBD_FAIL_RETURN(OBD_FAIL_OST_DESTROY_NET, 0);
 982                 rc = ost_destroy(req, oti);
 983                 break;
 984         case OST_GETATTR:
 985                 CDEBUG(D_INODE, "getattr\n");
 986                 OBD_FAIL_RETURN(OBD_FAIL_OST_GETATTR_NET, 0);
 987                 rc = ost_getattr(req);
 988                 break;
 989         case OST_SETATTR:
 990                 CDEBUG(D_INODE, "setattr\n");
 991                 OBD_FAIL_RETURN(OBD_FAIL_OST_SETATTR_NET, 0);
 992                 rc = ost_setattr(req, oti);
 993                 break;
 994         case OST_OPEN:
 995                 CDEBUG(D_INODE, "open\n");
 996                 OBD_FAIL_RETURN(OBD_FAIL_OST_OPEN_NET, 0);
 997                 rc = ost_open(req, oti);
 998                 break;
 999         case OST_CLOSE:
1000                 CDEBUG(D_INODE, "close\n");
1001                 OBD_FAIL_RETURN(OBD_FAIL_OST_CLOSE_NET, 0);
1002                 rc = ost_close(req, oti);
1003                 break;
1004         case OST_WRITE:
1005                 CDEBUG(D_INODE, "write\n");
1006                 OBD_FAIL_RETURN(OBD_FAIL_OST_BRW_NET, 0);
1007                 rc = ost_brw_write(req, oti);
1008                 /* ost_brw sends its own replies */
1009                 RETURN(rc);
1010         case OST_READ:
1011                 CDEBUG(D_INODE, "read\n");
1012                 OBD_FAIL_RETURN(OBD_FAIL_OST_BRW_NET, 0);
1013                 rc = ost_brw_read(req);
1014                 /* ost_brw sends its own replies */
1015                 RETURN(rc);
1016         case OST_SAN_READ:
1017                 CDEBUG(D_INODE, "san read\n");
1018                 OBD_FAIL_RETURN(OBD_FAIL_OST_BRW_NET, 0);
1019                 rc = ost_san_brw(req, OBD_BRW_READ);
1020                 /* ost_san_brw sends its own replies */
1021                 RETURN(rc);
1022         case OST_SAN_WRITE:
1023                 CDEBUG(D_INODE, "san write\n");
1024                 OBD_FAIL_RETURN(OBD_FAIL_OST_BRW_NET, 0);
1025                 rc = ost_san_brw(req, OBD_BRW_WRITE);
1026                 /* ost_san_brw sends its own replies */
1027                 RETURN(rc);
1028         case OST_PUNCH:
1029                 CDEBUG(D_INODE, "punch\n");
1030                 OBD_FAIL_RETURN(OBD_FAIL_OST_PUNCH_NET, 0);
1031                 rc = ost_punch(req, oti);
1032                 break;
1033         case OST_STATFS:
1034                 CDEBUG(D_INODE, "statfs\n");
1035                 OBD_FAIL_RETURN(OBD_FAIL_OST_STATFS_NET, 0);
1036                 rc = ost_statfs(req);
1037                 break;
1038         case OST_SYNCFS:
1039                 CDEBUG(D_INODE, "sync\n");
1040                 OBD_FAIL_RETURN(OBD_FAIL_OST_SYNCFS_NET, 0);
1041                 rc = ost_syncfs(req);
1042                 break;
1043         case OST_SET_INFO:
1044                 DEBUG_REQ(D_INODE, req, "set_info");
1045                 rc = ost_set_info(req);
1046         case OBD_PING:
1047                 DEBUG_REQ(D_INODE, req, "ping");
1048                 rc = target_handle_ping(req);
1049                 break;
1050         case OBD_LOG_CANCEL:
1051                 CDEBUG(D_INODE, "log cancel\n");
1052                 OBD_FAIL_RETURN(OBD_FAIL_OBD_LOG_CANCEL_NET, 0);
1053                 rc = ost_log_cancel(req);
1054                 break;
1055         case LDLM_ENQUEUE:
1056                 CDEBUG(D_INODE, "enqueue\n");
1057                 OBD_FAIL_RETURN(OBD_FAIL_LDLM_ENQUEUE, 0);
1058                 rc = ldlm_handle_enqueue(req, ldlm_server_completion_ast,
1059                                          ldlm_server_blocking_ast);
1060                 fail = OBD_FAIL_OST_LDLM_REPLY_NET;
1061                 break;
1062         case LDLM_CONVERT:
1063                 CDEBUG(D_INODE, "convert\n");
1064                 OBD_FAIL_RETURN(OBD_FAIL_LDLM_CONVERT, 0);
1065                 rc = ldlm_handle_convert(req);
1066                 break;
1067         case LDLM_CANCEL:
1068                 CDEBUG(D_INODE, "cancel\n");
1069                 OBD_FAIL_RETURN(OBD_FAIL_LDLM_CANCEL, 0);
1070                 rc = ldlm_handle_cancel(req);
1071                 break;
1072         case LDLM_BL_CALLBACK:
1073         case LDLM_CP_CALLBACK:
1074                 CDEBUG(D_INODE, "callback\n");
1075                 CERROR("callbacks should not happen on OST\n");
1076                 /* fall through */
1077         default:
1078                 CERROR("Unexpected opcode %d\n", req->rq_reqmsg->opc);
1079                 req->rq_status = -ENOTSUPP;
1080                 rc = ptlrpc_error(req);
1081                 RETURN(rc);
1082         }
1083
1084         EXIT;
1085         /* If we're DISCONNECTing, the export_data is already freed */
1086         if (!rc && req->rq_reqmsg->opc != OST_DISCONNECT) {
1087                 struct obd_device *obd  = req->rq_export->exp_obd;
1088                 if (!obd->obd_no_transno) {
1089                         req->rq_repmsg->last_committed =
1090                                 obd->obd_last_committed;
1091                 } else {
1092                         DEBUG_REQ(D_IOCTL, req,
1093                                   "not sending last_committed update");
1094                 }
1095                 CDEBUG(D_INFO, "last_committed "LPU64", xid "LPX64"\n",
1096                        obd->obd_last_committed, req->rq_xid);
1097         }
1098
1099 out:
1100         if (lustre_msg_get_flags(req->rq_reqmsg) & MSG_LAST_REPLAY) {
1101                 struct obd_device *obd = req->rq_export->exp_obd;
1102
1103                 if (obd && obd->obd_recovering) {
1104                         DEBUG_REQ(D_HA, req, "LAST_REPLAY, queuing reply");
1105                         return target_queue_final_reply(req, rc);
1106                 }
1107                 /* Lost a race with recovery; let the error path DTRT. */
1108                 rc = req->rq_status = -ENOTCONN;
1109         }
1110
1111         if (!rc)
1112                 oti_to_request(oti, req);
1113
1114         target_send_reply(req, rc, fail);
1115         return 0;
1116 }
1117
1118 static int ost_setup(struct obd_device *obddev, obd_count len, void *buf)
1119 {
1120         struct ost_obd *ost = &obddev->u.ost;
1121         int err, i;
1122         ENTRY;
1123
1124 #ifdef ENABLE_ORPHANS
1125         err = llog_start_commit_thread();
1126         if (err < 0)
1127                 RETURN(err);
1128 #endif
1129
1130         ost->ost_service = ptlrpc_init_svc(OST_NEVENTS, OST_NBUFS,
1131                                            OST_BUFSIZE, OST_MAXREQSIZE,
1132                                            OST_REQUEST_PORTAL, OSC_REPLY_PORTAL,
1133                                            ost_handle, "ost", obddev);
1134         if (!ost->ost_service) {
1135                 CERROR("failed to start service\n");
1136                 RETURN(-ENOMEM);
1137         }
1138
1139         for (i = 0; i < OST_NUM_THREADS; i++) {
1140                 char name[32];
1141                 sprintf(name, "ll_ost_%02d", i);
1142                 err = ptlrpc_start_thread(obddev, ost->ost_service, name);
1143                 if (err) {
1144                         CERROR("error starting thread #%d: rc %d\n", i, err);
1145                         RETURN(-EINVAL);
1146                 }
1147         }
1148
1149         RETURN(0);
1150 }
1151
1152 static int ost_cleanup(struct obd_device *obddev, int flags)
1153 {
1154         struct ost_obd *ost = &obddev->u.ost;
1155         int err = 0;
1156         ENTRY;
1157
1158         if (obddev->obd_recovering)
1159                 target_cancel_recovery_timer(obddev);
1160
1161         ptlrpc_stop_all_threads(ost->ost_service);
1162         ptlrpc_unregister_service(ost->ost_service);
1163
1164         RETURN(err);
1165 }
1166
1167 int ost_attach(struct obd_device *dev, obd_count len, void *data)
1168 {
1169         struct lprocfs_static_vars lvars;
1170
1171         lprocfs_init_vars(ost,&lvars);
1172         return lprocfs_obd_attach(dev, lvars.obd_vars);
1173 }
1174
1175 int ost_detach(struct obd_device *dev)
1176 {
1177         return lprocfs_obd_detach(dev);
1178 }
1179
1180 /* I don't think this function is ever used, since nothing
1181  * connects directly to this module.
1182  */
1183 static int ost_connect(struct lustre_handle *conn,
1184                        struct obd_device *obd, struct obd_uuid *cluuid)
1185 {
1186         struct obd_export *exp;
1187         int rc;
1188         ENTRY;
1189
1190         if (!conn || !obd || !cluuid)
1191                 RETURN(-EINVAL);
1192
1193         rc = class_connect(conn, obd, cluuid);
1194         if (rc)
1195                 RETURN(rc);
1196         exp = class_conn2export(conn);
1197         LASSERT(exp);
1198         class_export_put(exp);
1199
1200         RETURN(0);
1201 }
1202
1203 /* use obd ops to offer management infrastructure */
1204 static struct obd_ops ost_obd_ops = {
1205         o_owner:        THIS_MODULE,
1206         o_attach:       ost_attach,
1207         o_detach:       ost_detach,
1208         o_setup:        ost_setup,
1209         o_cleanup:      ost_cleanup,
1210         o_connect:      ost_connect,
1211 };
1212
1213 static int __init ost_init(void)
1214 {
1215         struct lprocfs_static_vars lvars;
1216         ENTRY;
1217
1218         lprocfs_init_vars(ost,&lvars);
1219         RETURN(class_register_type(&ost_obd_ops, lvars.module_vars,
1220                                    LUSTRE_OST_NAME));
1221 }
1222
1223 static void /*__exit*/ ost_exit(void)
1224 {
1225         class_unregister_type(LUSTRE_OST_NAME);
1226 }
1227
1228 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
1229 MODULE_DESCRIPTION("Lustre Object Storage Target (OST) v0.01");
1230 MODULE_LICENSE("GPL");
1231
1232 module_init(ost_init);
1233 module_exit(ost_exit);