lustre/mdc/mdc_request.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  * Copyright (C) 2001-2004 Cluster File Systems, Inc.
   5  *
   6  *   This file is part of Lustre, http://www.sf.net/projects/lustre/
   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 #ifndef EXPORT_SYMTAB
  23 # define EXPORT_SYMTAB
  24 #endif
  25 #define DEBUG_SUBSYSTEM S_MDC
  26
  27 #ifdef __KERNEL__
  28 # include <linux/module.h>
  29 # include <linux/pagemap.h>
  30 # include <linux/miscdevice.h>
  31 # include <linux/init.h>
  32 #else
  33 # include <liblustre.h>
  34 #endif
  35
  36 #include <linux/obd_class.h>
  37 #include <linux/lustre_mds.h>
  38 #include <linux/lustre_dlm.h>
  39 #include <linux/lprocfs_status.h>
  40 #include "mdc_internal.h"
  41
  42 #define REQUEST_MINOR 244
  43
  44 static int mdc_cleanup(struct obd_device *obd, int flags);
  45
  46 extern int mds_queue_req(struct ptlrpc_request *);
  47 /* Helper that implements most of mdc_getstatus and signal_completed_replay. */
  48 /* XXX this should become mdc_get_info("key"), sending MDS_GET_INFO RPC */
  49 static int send_getstatus(struct obd_import *imp, struct ll_fid *rootfid,
  50                           int level, int msg_flags)
  51 {
  52         struct ptlrpc_request *req;
  53         struct mds_body *body;
  54         int rc, size = sizeof(*body);
  55         ENTRY;
  56
  57         req = ptlrpc_prep_req(imp, MDS_GETSTATUS, 1, &size, NULL);
  58         if (!req)
  59                 GOTO(out, rc = -ENOMEM);
  60
  61         body = lustre_msg_buf(req->rq_reqmsg, 0, sizeof (*body));
  62         req->rq_send_state = level;
  63         req->rq_replen = lustre_msg_size(1, &size);
  64
  65         mdc_pack_req_body(req);
  66         req->rq_reqmsg->flags |= msg_flags;
  67         rc = ptlrpc_queue_wait(req);
  68
  69         if (!rc) {
  70                 body = lustre_swab_repbuf (req, 0, sizeof (*body),
  71                                            lustre_swab_mds_body);
  72                 if (body == NULL) {
  73                         CERROR ("Can't extract mds_body\n");
  74                         GOTO (out, rc = -EPROTO);
  75                 }
  76
  77                 memcpy(rootfid, &body->fid1, sizeof(*rootfid));
  78
  79                 CDEBUG(D_NET, "root ino="LPU64", last_committed="LPU64
  80                        ", last_xid="LPU64"\n",
  81                        rootfid->id, req->rq_repmsg->last_committed,
  82                        req->rq_repmsg->last_xid);
  83         }
  84
  85         EXIT;
  86  out:
  87         ptlrpc_req_finished(req);
  88         return rc;
  89 }
  90
  91 /* This should be mdc_get_info("rootfid") */
  92 int mdc_getstatus(struct obd_export *exp, struct ll_fid *rootfid)
  93 {
  94         return send_getstatus(class_exp2cliimp(exp), rootfid, LUSTRE_IMP_FULL,
  95                               0);
  96 }
  97
  98 int mdc_getattr_common(struct obd_export *exp, unsigned int ea_size,
  99                        struct ptlrpc_request *req)
 100 {
 101         struct mds_body *body;
 102         void            *eadata;
 103         int              rc;
 104         int              size[2] = {sizeof(*body), 0};
 105         int              bufcount = 1;
 106         ENTRY;
 107
 108         /* request message already built */
 109
 110         if (ea_size != 0) {
 111                 size[bufcount++] = ea_size;
 112                 CDEBUG(D_INODE, "reserved %u bytes for MD/symlink in packet\n",
 113                        ea_size);
 114         }
 115         req->rq_replen = lustre_msg_size(bufcount, size);
 116
 117         mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
 118         rc = ptlrpc_queue_wait(req);
 119         mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
 120         if (rc != 0)
 121                 RETURN (rc);
 122
 123         body = lustre_swab_repbuf (req, 0, sizeof (*body),
 124                                    lustre_swab_mds_body);
 125         if (body == NULL) {
 126                 CERROR ("Can't unpack mds_body\n");
 127                 RETURN (-EPROTO);
 128         }
 129
 130         CDEBUG(D_NET, "mode: %o\n", body->mode);
 131
 132         LASSERT_REPSWAB (req, 1);
 133         if (body->eadatasize != 0) {
 134                 /* reply indicates presence of eadata; check it's there... */
 135                 eadata = lustre_msg_buf (req->rq_repmsg, 1, body->eadatasize);
 136                 if (eadata == NULL) {
 137                         CERROR ("Missing/short eadata\n");
 138                         RETURN (-EPROTO);
 139                 }
 140         }
 141
 142         RETURN (0);
 143 }
 144
 145 int mdc_getattr(struct obd_export *exp, struct ll_fid *fid,
 146                 unsigned long valid, unsigned int ea_size,
 147                 struct ptlrpc_request **request)
 148 {
 149         struct ptlrpc_request *req;
 150         struct mds_body *body;
 151         int size = sizeof(*body);
 152         int rc;
 153         ENTRY;
 154
 155         /* XXX do we need to make another request here?  We just did a getattr
 156          *     to do the lookup in the first place.
 157          */
 158         req = ptlrpc_prep_req(class_exp2cliimp(exp), MDS_GETATTR, 1, &size,
 159                               NULL);
 160         if (!req)
 161                 GOTO(out, rc = -ENOMEM);
 162
 163         body = lustre_msg_buf(req->rq_reqmsg, 0, sizeof (*body));
 164         memcpy(&body->fid1, fid, sizeof(*fid));
 165         body->valid = valid;
 166         body->eadatasize = ea_size;
 167         mdc_pack_req_body(req);
 168
 169         rc = mdc_getattr_common(exp, ea_size, req);
 170         if (rc != 0) {
 171                 ptlrpc_req_finished (req);
 172                 req = NULL;
 173         }
 174  out:
 175         *request = req;
 176         RETURN (rc);
 177 }
 178
 179 int mdc_getattr_name(struct obd_export *exp, struct ll_fid *fid,
 180                      char *filename, int namelen, unsigned long valid,
 181                      unsigned int ea_size, struct ptlrpc_request **request)
 182 {
 183         struct ptlrpc_request *req;
 184         struct mds_body *body;
 185         int rc, size[2] = {sizeof(*body), namelen};
 186         ENTRY;
 187
 188         req = ptlrpc_prep_req(class_exp2cliimp(exp), MDS_GETATTR_NAME, 2,
 189                               size, NULL);
 190         if (!req)
 191                 GOTO(out, rc = -ENOMEM);
 192
 193         body = lustre_msg_buf(req->rq_reqmsg, 0, sizeof (*body));
 194         memcpy(&body->fid1, fid, sizeof(*fid));
 195         body->valid = valid;
 196         body->eadatasize = ea_size;
 197         mdc_pack_req_body(req);
 198
 199         if (filename != NULL) {
 200                 LASSERT (strnlen (filename, namelen) == namelen - 1);
 201                 memcpy(lustre_msg_buf(req->rq_reqmsg, 1, namelen),
 202                        filename, namelen);
 203         } else {
 204                 LASSERT(namelen == 1);
 205         }
 206
 207         rc = mdc_getattr_common(exp, ea_size, req);
 208         if (rc != 0) {
 209                 ptlrpc_req_finished (req);
 210                 req = NULL;
 211         }
 212  out:
 213         *request = req;
 214         RETURN(rc);
 215 }
 216
 217 /* This should be called with both the request and the reply still packed. */
 218 int mdc_store_inode_generation(struct obd_export *exp,
 219                                 struct ptlrpc_request *req, int reqoff,
 220                                 int repoff)
 221 {
 222         struct mds_rec_create *rec =
 223                 lustre_msg_buf(req->rq_reqmsg, reqoff, sizeof(*rec));
 224         struct mds_body *body =
 225                 lustre_msg_buf(req->rq_repmsg, repoff, sizeof(*body));
 226
 227         LASSERT (rec != NULL);
 228         LASSERT (body != NULL);
 229
 230         memcpy(&rec->cr_replayfid, &body->fid1, sizeof rec->cr_replayfid);
 231         DEBUG_REQ(D_HA, req, "storing generation %u for ino "LPU64,
 232                   rec->cr_replayfid.generation, rec->cr_replayfid.id);
 233         return 0;
 234 }
 235
 236 int mdc_req2lustre_md(struct obd_export *exp_mdc, struct ptlrpc_request *req,
 237                       unsigned int offset, struct obd_export *exp_osc,
 238                       struct lustre_md *md)
 239 {
 240         int rc = 0;
 241         ENTRY;
 242
 243         LASSERT(md);
 244         memset(md, 0, sizeof(*md));
 245
 246         md->body = lustre_msg_buf(req->rq_repmsg, offset, sizeof (*md->body));
 247         LASSERT (md->body != NULL);
 248         LASSERT_REPSWABBED (req, offset);
 249
 250         if (!(md->body->valid & OBD_MD_FLEASIZE) &&
 251             !(md->body->valid & OBD_MD_FLDIREA))
 252                 RETURN(0);
 253
 254         /* ea is presented in reply, parse it */
 255         if (S_ISREG(md->body->mode)) {
 256                 int lmmsize;
 257                 struct lov_mds_md *lmm;
 258
 259                 if (md->body->eadatasize == 0) {
 260                         CERROR ("OBD_MD_FLEASIZE set, but eadatasize 0\n");
 261                         RETURN(-EPROTO);
 262                 }
 263                 lmmsize = md->body->eadatasize;
 264                 lmm = lustre_msg_buf(req->rq_repmsg, offset + 1, lmmsize);
 265                 LASSERT (lmm != NULL);
 266                 LASSERT_REPSWABBED (req, offset + 1);
 267
 268                 rc = obd_unpackmd(exp_osc, &md->lsm, lmm, lmmsize);
 269                 if (rc >= 0) {
 270                         LASSERT (rc >= sizeof (*md->lsm));
 271                         rc = 0;
 272                 }
 273         } else if (S_ISDIR(md->body->mode)) {
 274                 struct mea *mea;
 275                 int mdsize;
 276                 LASSERT(exp_mdc != NULL);
 277                 /* dir can be non-splitted */
 278                 if (md->body->eadatasize == 0)
 279                         RETURN(0);
 280
 281                 mdsize = md->body->eadatasize;
 282                 mea = lustre_msg_buf(req->rq_repmsg, offset + 1, mdsize);
 283                 LASSERT(mea != NULL);
 284
 285                 rc = obd_unpackmd(exp_mdc, (void *) &md->mea,
 286                                   (void *) mea, mdsize);
 287                 if (rc >= 0) {
 288                         LASSERT (rc >= sizeof (*md->mea));
 289                         rc = 0;
 290                 }
 291         } else {
 292                 LASSERT(0);
 293         }
 294         RETURN(rc);
 295 }
 296
 297 static void mdc_commit_open(struct ptlrpc_request *req)
 298 {
 299         struct mdc_open_data *mod = req->rq_cb_data;
 300         if (mod == NULL)
 301                 return;
 302
 303         if (mod->mod_close_req != NULL)
 304                 mod->mod_close_req->rq_cb_data = NULL;
 305
 306         if (mod->mod_och != NULL)
 307                 mod->mod_och->och_mod = NULL;
 308
 309         OBD_FREE(mod, sizeof(*mod));
 310         req->rq_cb_data = NULL;
 311 }
 312
 313 static void mdc_replay_open(struct ptlrpc_request *req)
 314 {
 315         struct mdc_open_data *mod = req->rq_cb_data;
 316         struct obd_client_handle *och;
 317         struct ptlrpc_request *close_req;
 318         struct lustre_handle old;
 319         struct mds_body *body;
 320         ENTRY;
 321
 322         body = lustre_swab_repbuf(req, 1, sizeof(*body), lustre_swab_mds_body);
 323         LASSERT (body != NULL);
 324
 325         if (mod == NULL) {
 326                 DEBUG_REQ(D_ERROR, req,
 327                           "can't properly replay without open data");
 328                 EXIT;
 329                 return;
 330         }
 331
 332         och = mod->mod_och;
 333         if (och != NULL) {
 334                 struct lustre_handle *file_fh;
 335                 LASSERT(och->och_magic == OBD_CLIENT_HANDLE_MAGIC);
 336                 file_fh = &och->och_fh;
 337                 CDEBUG(D_HA, "updating handle from "LPX64" to "LPX64"\n",
 338                        file_fh->cookie, body->handle.cookie);
 339                 memcpy(&old, file_fh, sizeof(old));
 340                 memcpy(file_fh, &body->handle, sizeof(*file_fh));
 341         }
 342
 343         close_req = mod->mod_close_req;
 344         if (close_req != NULL) {
 345                 struct mds_body *close_body;
 346                 LASSERT(close_req->rq_reqmsg->opc == MDS_CLOSE);
 347                 close_body = lustre_msg_buf(close_req->rq_reqmsg, 0,
 348                                             sizeof(*close_body));
 349                 if (och != NULL)
 350                         LASSERT(!memcmp(&old, &close_body->handle, sizeof old));
 351                 DEBUG_REQ(D_HA, close_req, "updating close body with new fh");
 352                 memcpy(&close_body->handle, &body->handle,
 353                        sizeof(close_body->handle));
 354         }
 355
 356         EXIT;
 357 }
 358
 359 int  mdc_set_open_replay_data(struct obd_export *exp,
 360                               struct obd_client_handle *och,
 361                               struct ptlrpc_request *open_req)
 362 {
 363         struct mdc_open_data *mod;
 364         struct mds_rec_create *rec =
 365                 lustre_msg_buf(open_req->rq_reqmsg, 2, sizeof(*rec));
 366         struct mds_body *body =
 367                 lustre_msg_buf(open_req->rq_repmsg, 1, sizeof(*body));
 368
 369         LASSERT(rec != NULL);
 370         /* outgoing messages always in my byte order */
 371         LASSERT(body != NULL);
 372         /* incoming message in my byte order (it's been swabbed) */
 373         LASSERT_REPSWABBED(open_req, 1);
 374
 375         OBD_ALLOC(mod, sizeof(*mod));
 376         if (mod == NULL) {
 377                 DEBUG_REQ(D_ERROR, open_req, "can't allocate mdc_open_data");
 378                 return 0;
 379         }
 380
 381         och->och_mod = mod;
 382         mod->mod_och = och;
 383         mod->mod_open_req = open_req;
 384
 385         memcpy(&rec->cr_replayfid, &body->fid1, sizeof rec->cr_replayfid);
 386         open_req->rq_replay_cb = mdc_replay_open;
 387         open_req->rq_commit_cb = mdc_commit_open;
 388         open_req->rq_cb_data = mod;
 389         DEBUG_REQ(D_HA, open_req, "set up replay data");
 390         return 0;
 391 }
 392
 393 int mdc_clear_open_replay_data(struct obd_export *exp,
 394                                struct obd_client_handle *och)
 395 {
 396         struct mdc_open_data *mod = och->och_mod;
 397
 398         /* Don't free the structure now (it happens in mdc_commit_open, after
 399          * we're sure we won't need to fix up the close request in the future),
 400          * but make sure that replay doesn't poke at the och, which is about to
 401          * be freed. */
 402         LASSERT(mod != LP_POISON);
 403         if (mod != NULL)
 404                 mod->mod_och = NULL;
 405         och->och_mod = NULL;
 406         return 0;
 407 }
 408
 409 static void mdc_commit_close(struct ptlrpc_request *req)
 410 {
 411         struct mdc_open_data *mod = req->rq_cb_data;
 412         struct ptlrpc_request *open_req;
 413         struct obd_import *imp = req->rq_import;
 414
 415         DEBUG_REQ(D_HA, req, "close req committed");
 416         if (mod == NULL)
 417                 return;
 418
 419         mod->mod_close_req = NULL;
 420         req->rq_cb_data = NULL;
 421         req->rq_commit_cb = NULL;
 422
 423         open_req = mod->mod_open_req;
 424         LASSERT(open_req != NULL);
 425         LASSERT(open_req != LP_POISON);
 426         LASSERT(open_req->rq_type != LI_POISON);
 427
 428         DEBUG_REQ(D_HA, open_req, "open req balanced");
 429         if (open_req->rq_transno == 0) {
 430                 DEBUG_REQ(D_ERROR, open_req, "BUG 3892  open");
 431                 DEBUG_REQ(D_ERROR, req, "BUG 3892 close");
 432                 LASSERTF(open_req->rq_transno != 0, "BUG 3892");
 433         }
 434         LASSERT(open_req->rq_import == imp);
 435
 436         /* We no longer want to preserve this for transno-unconditional
 437          * replay. */
 438         spin_lock(&open_req->rq_lock);
 439         open_req->rq_replay = 0;
 440         spin_unlock(&open_req->rq_lock);
 441 }
 442
 443 static int mdc_close_interpret(struct ptlrpc_request *req, void *data, int rc)
 444 {
 445         union ptlrpc_async_args *aa = data;
 446         struct mdc_rpc_lock *rpc_lock;
 447         struct obd_device *obd = aa->pointer_arg[1];
 448         unsigned long flags;
 449
 450         spin_lock_irqsave(&req->rq_lock, flags);
 451         rpc_lock = aa->pointer_arg[0];
 452         aa->pointer_arg[0] = NULL;
 453         spin_unlock_irqrestore (&req->rq_lock, flags);
 454
 455         if (rpc_lock == NULL) {
 456                 CERROR("called with NULL rpc_lock\n");
 457         } else {
 458                 mdc_put_rpc_lock(rpc_lock, NULL);
 459                 LASSERTF(rpc_lock == obd->u.cli.cl_rpc_lock, "%p != %p\n",
 460                          rpc_lock, obd->u.cli.cl_rpc_lock);
 461         }
 462         wake_up(&req->rq_reply_waitq);
 463         RETURN(rc);
 464 }
 465
 466 /* We can't use ptlrpc_check_reply, because we don't want to wake up for
 467  * anything but a reply or an error. */
 468 static int mdc_close_check_reply(struct ptlrpc_request *req)
 469 {
 470         int rc = 0;
 471         unsigned long flags;
 472
 473         spin_lock_irqsave(&req->rq_lock, flags);
 474         if (req->rq_async_args.pointer_arg[0] == NULL)
 475                 rc = 1;
 476         spin_unlock_irqrestore (&req->rq_lock, flags);
 477         return rc;
 478 }
 479
 480 static int go_back_to_sleep(void *unused)
 481 {
 482         return 0;
 483 }
 484
 485 int mdc_close(struct obd_export *exp, struct obdo *oa,
 486               struct obd_client_handle *och, struct ptlrpc_request **request)
 487 {
 488         struct obd_device *obd = class_exp2obd(exp);
 489         int reqsize = sizeof(struct mds_body);
 490         int rc, repsize[3] = {sizeof(struct mds_body),
 491                               obd->u.cli.cl_max_mds_easize,
 492                               obd->u.cli.cl_max_mds_cookiesize};
 493         struct ptlrpc_request *req;
 494         struct mdc_open_data *mod;
 495         struct l_wait_info lwi;
 496         ENTRY;
 497
 498         req = ptlrpc_prep_req(class_exp2cliimp(exp), MDS_CLOSE, 1, &reqsize,
 499                               NULL);
 500         if (req == NULL)
 501                 GOTO(out, rc = -ENOMEM);
 502
 503         /* Ensure that this close's handle is fixed up during replay. */
 504         LASSERT(och != NULL);
 505         mod = och->och_mod;
 506         if (likely(mod != NULL)) {
 507                 mod->mod_close_req = req;
 508                 LASSERT(mod->mod_open_req->rq_type != LI_POISON);
 509                 DEBUG_REQ(D_HA, mod->mod_open_req, "matched open req %p",
 510                           mod->mod_open_req);
 511         } else {
 512                 CDEBUG(D_HA, "couldn't find open req; expecting close error\n");
 513         }
 514
 515         mdc_close_pack(req, 0, oa, oa->o_valid, och);
 516
 517         req->rq_replen = lustre_msg_size(3, repsize);
 518         req->rq_commit_cb = mdc_commit_close;
 519         LASSERT(req->rq_cb_data == NULL);
 520         req->rq_cb_data = mod;
 521
 522         /* We hand a ref to the rpcd here, so we need another one of our own. */
 523         ptlrpc_request_addref(req);
 524
 525         mdc_get_rpc_lock(obd->u.cli.cl_rpc_lock, NULL);
 526         req->rq_interpret_reply = mdc_close_interpret;
 527         req->rq_async_args.pointer_arg[0] = obd->u.cli.cl_rpc_lock;
 528         req->rq_async_args.pointer_arg[1] = obd;
 529         ptlrpcd_add_req(req);
 530         lwi = LWI_TIMEOUT_INTR(MAX(req->rq_timeout * HZ, 1), go_back_to_sleep,
 531                                NULL, NULL);
 532         rc = l_wait_event(req->rq_reply_waitq, mdc_close_check_reply(req),
 533                           &lwi);
 534         if (req->rq_repmsg == NULL) {
 535                 CDEBUG(D_HA, "request failed to send: %p, %d\n", req,
 536                        req->rq_status);
 537                 if (rc == 0)
 538                         rc = req->rq_status ? req->rq_status : -EIO;
 539         } else if (rc == 0) {
 540                 rc = req->rq_repmsg->status;
 541                 if (req->rq_repmsg->type == PTL_RPC_MSG_ERR) {
 542                         DEBUG_REQ(D_ERROR, req, "type == PTL_RPC_MSG_ERR, err "
 543                                   "= %d", rc);
 544                         if (rc > 0)
 545                                 rc = -rc;
 546                 } else if (mod == NULL) {
 547                         CERROR("Unexpected: can't find mdc_open_data, but the "
 548                                "close succeeded.  Please tell CFS.\n");
 549                 }
 550                 if (!lustre_swab_repbuf(req, 0, sizeof(struct mds_body),
 551                                         lustre_swab_mds_body)) {
 552                         CERROR("Error unpacking mds_body\n");
 553                         rc = -EPROTO;
 554                 }
 555         }
 556         if (req->rq_async_args.pointer_arg[0] != NULL) {
 557                 CERROR("returned without dropping rpc_lock: rc %d\n", rc);
 558                 mdc_close_interpret(req, &req->rq_async_args, rc);
 559                 portals_debug_dumplog();
 560         }
 561
 562         EXIT;
 563  out:
 564         *request = req;
 565         return rc;
 566 }
 567
 568 int mdc_done_writing(struct obd_export *exp, struct obdo *obdo)
 569 {
 570         struct ptlrpc_request *req;
 571         struct mds_body *body;
 572         int rc, size = sizeof(*body);
 573         ENTRY;
 574
 575         req = ptlrpc_prep_req(class_exp2cliimp(exp), MDS_DONE_WRITING, 1,
 576                               &size, NULL);
 577         if (req == NULL)
 578                 RETURN(-ENOMEM);
 579
 580         body = lustre_msg_buf(req->rq_reqmsg, 0, sizeof(*body));
 581         mdc_pack_fid(&body->fid1, obdo->o_id, 0, obdo->o_mode);
 582         body->size = obdo->o_size;
 583         body->blocks = obdo->o_blocks;
 584         body->flags = obdo->o_flags;
 585         body->valid = obdo->o_valid;
 586 //        memcpy(&body->handle, &och->och_fh, sizeof(body->handle));
 587
 588         req->rq_replen = lustre_msg_size(1, &size);
 589
 590         rc = ptlrpc_queue_wait(req);
 591         ptlrpc_req_finished(req);
 592         RETURN(rc);
 593 }
 594
 595 int mdc_readpage(struct obd_export *exp, struct ll_fid *mdc_fid,
 596                  __u64 offset, struct page *page,
 597                  struct ptlrpc_request **request)
 598 {
 599         struct obd_import *imp = class_exp2cliimp(exp);
 600         struct ptlrpc_request *req = NULL;
 601         struct ptlrpc_bulk_desc *desc = NULL;
 602         struct mds_body *body;
 603         int rc, size = sizeof(*body);
 604         ENTRY;
 605
 606         CDEBUG(D_INODE, "inode: %ld\n", (long)mdc_fid->id);
 607
 608         req = ptlrpc_prep_req(imp, MDS_READPAGE, 1, &size, NULL);
 609         if (req == NULL)
 610                 GOTO(out, rc = -ENOMEM);
 611         /* XXX FIXME bug 249 */
 612         req->rq_request_portal = MDS_READPAGE_PORTAL;
 613
 614         desc = ptlrpc_prep_bulk_imp(req, 1, BULK_PUT_SINK, MDS_BULK_PORTAL);
 615         if (desc == NULL)
 616                 GOTO(out, rc = -ENOMEM);
 617         /* NB req now owns desc and will free it when it gets freed */
 618
 619         ptlrpc_prep_bulk_page(desc, page, 0, PAGE_CACHE_SIZE);
 620
 621         mdc_readdir_pack(req, offset, PAGE_CACHE_SIZE, mdc_fid);
 622
 623         req->rq_replen = lustre_msg_size(1, &size);
 624         rc = ptlrpc_queue_wait(req);
 625
 626         if (rc == 0) {
 627                 body = lustre_swab_repbuf(req, 0, sizeof (*body),
 628                                           lustre_swab_mds_body);
 629                 if (body == NULL) {
 630                         CERROR("Can't unpack mds_body\n");
 631                         GOTO(out, rc = -EPROTO);
 632                 }
 633
 634                 if (req->rq_bulk->bd_nob_transferred != PAGE_CACHE_SIZE) {
 635                         CERROR ("Unexpected # bytes transferred: %d"
 636                                 " (%ld expected)\n",
 637                                 req->rq_bulk->bd_nob_transferred,
 638                                 PAGE_CACHE_SIZE);
 639                         GOTO (out, rc = -EPROTO);
 640                 }
 641         }
 642
 643         EXIT;
 644  out:
 645         *request = req;
 646         return rc;
 647 }
 648
 649 static int mdc_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
 650                          void *karg, void *uarg)
 651 {
 652         struct obd_device *obd = exp->exp_obd;
 653         struct obd_ioctl_data *data = karg;
 654         struct obd_import *imp = obd->u.cli.cl_import;
 655         struct llog_ctxt *ctxt;
 656         int rc;
 657         ENTRY;
 658
 659 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
 660         MOD_INC_USE_COUNT;
 661 #else
 662         if (!try_module_get(THIS_MODULE)) {
 663                 CERROR("Can't get module. Is it alive?");
 664                 return -EINVAL;
 665         }
 666 #endif
 667         switch (cmd) {
 668         case OBD_IOC_CLIENT_RECOVER:
 669                 rc = ptlrpc_recover_import(imp, data->ioc_inlbuf1);
 670                 if (rc < 0)
 671                         GOTO(out, rc);
 672                 GOTO(out, rc = 0);
 673         case IOC_OSC_SET_ACTIVE:
 674                 rc = ptlrpc_set_import_active(imp, data->ioc_offset);
 675                 GOTO(out, rc);
 676         case OBD_IOC_PARSE: {
 677                 ctxt = llog_get_context(&exp->exp_obd->obd_llogs,
 678                                         LLOG_CONFIG_REPL_CTXT);
 679                 rc = class_config_process_llog(ctxt, data->ioc_inlbuf1, NULL);
 680                 GOTO(out, rc);
 681         }
 682 #ifdef __KERNEL__
 683         case OBD_IOC_LLOG_INFO:
 684         case OBD_IOC_LLOG_PRINT: {
 685                 ctxt = llog_get_context(&obd->obd_llogs, LLOG_CONFIG_REPL_CTXT);
 686                 rc = llog_ioctl(ctxt, cmd, data);
 687
 688                 GOTO(out, rc);
 689         }
 690 #endif
 691         default:
 692                 CERROR("mdc_ioctl(): unrecognised ioctl %#x\n", cmd);
 693                 GOTO(out, rc = -ENOTTY);
 694         }
 695 out:
 696 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
 697         MOD_DEC_USE_COUNT;
 698 #else
 699         module_put(THIS_MODULE);
 700 #endif
 701
 702         return rc;
 703 }
 704
 705 int mdc_set_info(struct obd_export *exp, obd_count keylen,
 706                  void *key, obd_count vallen, void *val)
 707 {
 708         int rc = -EINVAL;
 709
 710         if (keylen == strlen("initial_recov") &&
 711             memcmp(key, "initial_recov", strlen("initial_recov")) == 0) {
 712                 struct obd_import *imp = exp->exp_obd->u.cli.cl_import;
 713                 if (vallen != sizeof(int))
 714                         RETURN(-EINVAL);
 715                 imp->imp_initial_recov = *(int *)val;
 716                 CDEBUG(D_HA, "%s: set imp_no_init_recov = %d\n",
 717                        exp->exp_obd->obd_name,
 718                        imp->imp_initial_recov);
 719                 RETURN(0);
 720         } else if (keylen >= strlen("client") && strcmp(key, "client") == 0) {
 721                 struct ptlrpc_request *req;
 722                 char *bufs[1] = {key};
 723                 int rc;
 724                 req = ptlrpc_prep_req(class_exp2cliimp(exp), OST_SET_INFO, 1,
 725                                       &keylen, bufs);
 726                 if (req == NULL)
 727                         RETURN(-ENOMEM);
 728
 729                 req->rq_replen = lustre_msg_size(0, NULL);
 730                 rc = ptlrpc_queue_wait(req);
 731                 ptlrpc_req_finished(req);
 732                 RETURN(rc);
 733         } else if (keylen >= strlen("mds_num") && strcmp(key, "mds_num") == 0) {
 734                 struct ptlrpc_request *req;
 735                 int rc, size[2] = {keylen, vallen};
 736                 char *bufs[2] = {key, val};
 737
 738                 req = ptlrpc_prep_req(class_exp2cliimp(exp), OST_SET_INFO, 2,
 739                                       size, bufs);
 740                 if (req == NULL)
 741                         RETURN(-ENOMEM);
 742
 743                 req->rq_replen = lustre_msg_size(0, NULL);
 744                 rc = ptlrpc_queue_wait(req);
 745                 ptlrpc_req_finished(req);
 746                 RETURN(rc);
 747         } else if (keylen >= strlen("inter_mds") && strcmp(key, "inter_mds") == 0) {
 748                 struct obd_import *imp = class_exp2cliimp(exp);
 749                 imp->imp_server_timeout = 1;
 750                 CDEBUG(D_OTHER, "%s: timeout / 2\n", exp->exp_obd->obd_name);
 751                 RETURN(0);
 752         }
 753         RETURN(rc);
 754 }
 755
 756 static int mdc_statfs(struct obd_device *obd, struct obd_statfs *osfs,
 757                       unsigned long max_age)
 758 {
 759         struct ptlrpc_request *req;
 760         struct obd_statfs *msfs;
 761         int rc, size = sizeof(*msfs);
 762         ENTRY;
 763
 764         /* We could possibly pass max_age in the request (as an absolute
 765          * timestamp or a "seconds.usec ago") so the target can avoid doing
 766          * extra calls into the filesystem if that isn't necessary (e.g.
 767          * during mount that would help a bit).  Having relative timestamps
 768          * is not so great if request processing is slow, while absolute
 769          * timestamps are not ideal because they need time synchronization. */
 770         req = ptlrpc_prep_req(obd->u.cli.cl_import, MDS_STATFS, 0, NULL, NULL);
 771         if (!req)
 772                 RETURN(-ENOMEM);
 773
 774         req->rq_replen = lustre_msg_size(1, &size);
 775
 776         mdc_get_rpc_lock(obd->u.cli.cl_rpc_lock, NULL);
 777         rc = ptlrpc_queue_wait(req);
 778         mdc_put_rpc_lock(obd->u.cli.cl_rpc_lock, NULL);
 779
 780         if (rc)
 781                 GOTO(out, rc);
 782
 783         msfs = lustre_swab_repbuf(req, 0, sizeof(*msfs),lustre_swab_obd_statfs);
 784         if (msfs == NULL) {
 785                 CERROR("Can't unpack obd_statfs\n");
 786                 GOTO(out, rc = -EPROTO);
 787         }
 788
 789         memcpy(osfs, msfs, sizeof (*msfs));
 790         EXIT;
 791 out:
 792         ptlrpc_req_finished(req);
 793
 794         return rc;
 795 }
 796
 797 static int mdc_pin(struct obd_export *exp, obd_id ino, __u32 gen, int type,
 798                    struct obd_client_handle *handle, int flag)
 799 {
 800         struct ptlrpc_request *req;
 801         struct mds_body *body;
 802         int rc, size = sizeof(*body);
 803         ENTRY;
 804
 805         req = ptlrpc_prep_req(class_exp2cliimp(exp), MDS_PIN, 1, &size, NULL);
 806         if (req == NULL)
 807                 RETURN(-ENOMEM);
 808
 809         body = lustre_msg_buf(req->rq_reqmsg, 0, sizeof (*body));
 810         mdc_pack_fid(&body->fid1, ino, gen, type);
 811         body->flags = flag;
 812
 813         req->rq_replen = lustre_msg_size(1, &size);
 814
 815         mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
 816         rc = ptlrpc_queue_wait(req);
 817         mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
 818         if (rc) {
 819                 CERROR("pin failed: %d\n", rc);
 820                 ptlrpc_req_finished(req);
 821                 RETURN(rc);
 822         }
 823
 824         body = lustre_swab_repbuf(req, 0, sizeof(*body), lustre_swab_mds_body);
 825         if (body == NULL) {
 826                 ptlrpc_req_finished(req);
 827                 RETURN(rc);
 828         }
 829
 830         memcpy(&handle->och_fh, &body->handle, sizeof(body->handle));
 831         handle->och_magic = OBD_CLIENT_HANDLE_MAGIC;
 832
 833         OBD_ALLOC(handle->och_mod, sizeof(*handle->och_mod));
 834         if (handle->och_mod == NULL) {
 835                 DEBUG_REQ(D_ERROR, req, "can't allocate mdc_open_data");
 836                 RETURN(-ENOMEM);
 837         }
 838         handle->och_mod->mod_open_req = req; /* will be dropped by unpin */
 839
 840         RETURN(rc);
 841 }
 842
 843 static int mdc_unpin(struct obd_export *exp,
 844                      struct obd_client_handle *handle, int flag)
 845 {
 846         struct ptlrpc_request *req;
 847         struct mds_body *body;
 848         int rc, size = sizeof(*body);
 849         ENTRY;
 850
 851         if (handle->och_magic != OBD_CLIENT_HANDLE_MAGIC)
 852                 RETURN(0);
 853
 854         req = ptlrpc_prep_req(class_exp2cliimp(exp), MDS_CLOSE, 1, &size, NULL);
 855         if (req == NULL)
 856                 RETURN(-ENOMEM);
 857
 858         body = lustre_msg_buf(req->rq_reqmsg, 0, sizeof(*body));
 859         memcpy(&body->handle, &handle->och_fh, sizeof(body->handle));
 860         body->flags = flag;
 861
 862         req->rq_replen = lustre_msg_size(0, NULL);
 863         mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
 864         rc = ptlrpc_queue_wait(req);
 865         mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
 866
 867         if (rc != 0)
 868                 CERROR("unpin failed: %d\n", rc);
 869
 870         ptlrpc_req_finished(req);
 871         ptlrpc_req_finished(handle->och_mod->mod_open_req);
 872         OBD_FREE(handle->och_mod, sizeof(*handle->och_mod));
 873         RETURN(rc);
 874 }
 875
 876 int mdc_sync(struct obd_export *exp, struct ll_fid *fid,
 877              struct ptlrpc_request **request)
 878 {
 879         struct ptlrpc_request *req;
 880         struct mds_body *body;
 881         int size = sizeof(*body);
 882         int rc;
 883         ENTRY;
 884
 885         req = ptlrpc_prep_req(class_exp2cliimp(exp), MDS_SYNC, 1,&size,NULL);
 886         if (!req)
 887                 RETURN(rc = -ENOMEM);
 888
 889         if (fid) {
 890                 body = lustre_msg_buf(req->rq_reqmsg, 0, sizeof (*body));
 891                 memcpy(&body->fid1, fid, sizeof(*fid));
 892                 mdc_pack_req_body(req);
 893         }
 894
 895         req->rq_replen = lustre_msg_size(1, &size);
 896
 897         rc = ptlrpc_queue_wait(req);
 898         if (rc || request == NULL)
 899                 ptlrpc_req_finished(req);
 900         else
 901                 *request = req;
 902
 903         RETURN(rc);
 904 }
 905
 906 static int mdc_import_event(struct obd_device *obd, struct obd_import *imp,
 907                             enum obd_import_event event)
 908 {
 909         int rc = 0;
 910
 911         LASSERT(imp->imp_obd == obd);
 912
 913         switch (event) {
 914         case IMP_EVENT_DISCON: {
 915                 break;
 916         }
 917         case IMP_EVENT_INACTIVE: {
 918                 if (obd->obd_observer)
 919                         rc = obd_notify(obd->obd_observer, obd, 0);
 920                 break;
 921         }
 922         case IMP_EVENT_INVALIDATE: {
 923                 struct ldlm_namespace *ns = obd->obd_namespace;
 924
 925                 ldlm_namespace_cleanup(ns, LDLM_FL_LOCAL_ONLY);
 926
 927                 break;
 928         }
 929         case IMP_EVENT_ACTIVE: {
 930                 if (obd->obd_observer)
 931                         rc = obd_notify(obd->obd_observer, obd, 1);
 932                 break;
 933         }
 934         default:
 935                 CERROR("Unknown import event %d\n", event);
 936                 LBUG();
 937         }
 938         RETURN(rc);
 939 }
 940
 941 static int mdc_attach(struct obd_device *dev, obd_count len, void *data)
 942 {
 943         struct lprocfs_static_vars lvars;
 944
 945         lprocfs_init_vars(mdc, &lvars);
 946         return lprocfs_obd_attach(dev, lvars.obd_vars);
 947 }
 948
 949 static int mdc_detach(struct obd_device *dev)
 950 {
 951         return lprocfs_obd_detach(dev);
 952 }
 953
 954 static int mdc_setup(struct obd_device *obd, obd_count len, void *buf)
 955 {
 956         struct client_obd *cli = &obd->u.cli;
 957         int rc;
 958         ENTRY;
 959
 960         OBD_ALLOC(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
 961         if (!cli->cl_rpc_lock)
 962                 RETURN(-ENOMEM);
 963         mdc_init_rpc_lock(cli->cl_rpc_lock);
 964
 965         ptlrpcd_addref();
 966
 967         OBD_ALLOC(cli->cl_setattr_lock, sizeof (*cli->cl_setattr_lock));
 968         if (!cli->cl_setattr_lock)
 969                 GOTO(err_rpc_lock, rc = -ENOMEM);
 970         mdc_init_rpc_lock(cli->cl_setattr_lock);
 971
 972         rc = client_obd_setup(obd, len, buf);
 973         if (rc)
 974                 GOTO(err_setattr_lock, rc);
 975
 976         rc = obd_llog_init(obd, &obd->obd_llogs, obd, 0, NULL);
 977         if (rc) {
 978                 mdc_cleanup(obd, 0);
 979                 CERROR("failed to setup llogging subsystems\n");
 980         }
 981
 982         RETURN(rc);
 983
 984 err_setattr_lock:
 985         OBD_FREE(cli->cl_setattr_lock, sizeof (*cli->cl_setattr_lock));
 986 err_rpc_lock:
 987         OBD_FREE(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
 988         ptlrpcd_decref();
 989         RETURN(rc);
 990 }
 991
 992 static int mdc_init_ea_size(struct obd_export *exp, int easize, int cookiesize)
 993 {
 994         struct obd_device *obd = exp->exp_obd;
 995         struct client_obd *cli = &obd->u.cli;
 996         ENTRY;
 997
 998         if (cli->cl_max_mds_easize < easize)
 999                 cli->cl_max_mds_easize = easize;
1000         if (cli->cl_max_mds_cookiesize < cookiesize)
1001                 cli->cl_max_mds_cookiesize = cookiesize;
1002         RETURN(0);
1003 }
1004
1005 static int mdc_precleanup(struct obd_device *obd, int flags)
1006 {
1007         int rc = 0;
1008
1009         rc = obd_llog_finish(obd, &obd->obd_llogs, 0);
1010         if (rc != 0)
1011                 CERROR("failed to cleanup llogging subsystems\n");
1012
1013         RETURN(rc);
1014 }
1015
1016 static int mdc_cleanup(struct obd_device *obd, int flags)
1017 {
1018         struct client_obd *cli = &obd->u.cli;
1019
1020         OBD_FREE(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
1021         OBD_FREE(cli->cl_setattr_lock, sizeof (*cli->cl_setattr_lock));
1022
1023         ptlrpcd_decref();
1024
1025         return client_obd_cleanup(obd, flags);
1026 }
1027
1028
1029 static int mdc_llog_init(struct obd_device *obd, struct obd_llogs *llogs,
1030                          struct obd_device *tgt, int count,
1031                          struct llog_catid *logid)
1032 {
1033         struct llog_ctxt *ctxt;
1034         int rc;
1035         ENTRY;
1036
1037         rc = obd_llog_setup(obd, llogs, LLOG_CONFIG_REPL_CTXT, tgt, 0, NULL,
1038                             &llog_client_ops);
1039         if (rc == 0) {
1040                 ctxt = llog_get_context(llogs, LLOG_CONFIG_REPL_CTXT);
1041                 ctxt->loc_imp = obd->u.cli.cl_import;
1042         }
1043
1044         RETURN(rc);
1045 }
1046
1047 static int mdc_llog_finish(struct obd_device *obd,
1048                            struct obd_llogs *llogs, int count)
1049 {
1050         int rc;
1051         ENTRY;
1052
1053         rc = obd_llog_cleanup(llog_get_context(llogs, LLOG_CONFIG_REPL_CTXT));
1054         RETURN(rc);
1055 }
1056 static struct obd_device *mdc_get_real_obd(struct obd_export *exp,
1057                                            char *name, int len)
1058 {
1059        ENTRY;
1060        RETURN(exp->exp_obd);
1061 }
1062
1063 static int mdc_get_info(struct obd_export *exp, obd_count keylen,
1064                         void *key, __u32 *vallen, void *val)
1065 {
1066         ENTRY;
1067         if (!vallen || !val)
1068                 RETURN(-EFAULT);
1069
1070         if (keylen >= strlen("mdsize") && strcmp(key, "mdsize") == 0) {
1071                 struct ptlrpc_request *req;
1072                 __u32 *reply;
1073                 char *bufs[1] = {key};
1074                 int rc;
1075                 req = ptlrpc_prep_req(class_exp2cliimp(exp), OST_GET_INFO, 1,
1076                                       &keylen, bufs);
1077                 if (req == NULL)
1078                         RETURN(-ENOMEM);
1079
1080                 req->rq_replen = lustre_msg_size(1, vallen);
1081                 rc = ptlrpc_queue_wait(req);
1082                 if (rc)
1083                         GOTO(out, rc);
1084
1085                 /* FIXME: right swap routine here! */
1086                 reply = lustre_swab_repbuf(req, 0, sizeof(*reply), NULL);
1087                 if (reply == NULL) {
1088                         CERROR("Can't unpack mdsize\n");
1089                         GOTO(out, rc = -EPROTO);
1090                 }
1091                 *((__u32 *)val) = *reply;
1092         out:
1093                 ptlrpc_req_finished(req);
1094                 RETURN(rc);
1095         }
1096         RETURN(-EINVAL);
1097 }
1098
1099 int mdc_obj_create(struct obd_export *exp, struct obdo *oa,
1100                     struct lov_stripe_md **ea, struct obd_trans_info *oti)
1101 {
1102         struct ptlrpc_request *request;
1103         struct ost_body *body;
1104         int rc, size = sizeof(*body);
1105         ENTRY;
1106
1107         LASSERT(oa);
1108
1109         request = ptlrpc_prep_req(class_exp2cliimp(exp), OST_CREATE,
1110                                   1, &size, NULL);
1111         if (!request)
1112                 GOTO(out_req, rc = -ENOMEM);
1113
1114         body = lustre_msg_buf(request->rq_reqmsg, 0, sizeof (*body));
1115         memcpy(&body->oa, oa, sizeof(body->oa));
1116
1117         request->rq_replen = lustre_msg_size(1, &size);
1118         rc = ptlrpc_queue_wait(request);
1119         if (rc)
1120                 GOTO(out_req, rc);
1121
1122         body = lustre_swab_repbuf(request, 0, sizeof(*body),
1123                                   lustre_swab_ost_body);
1124         if (body == NULL) {
1125                 CERROR ("can't unpack ost_body\n");
1126                 GOTO (out_req, rc = -EPROTO);
1127         }
1128
1129         memcpy(oa, &body->oa, sizeof(*oa));
1130
1131         /* store ino/generation for recovery */
1132         body = lustre_msg_buf(request->rq_reqmsg, 0, sizeof (*body));
1133         body->oa.o_id = oa->o_id;
1134         body->oa.o_generation = oa->o_generation;
1135
1136         CDEBUG(D_HA, "transno: "LPD64"\n", request->rq_repmsg->transno);
1137         EXIT;
1138 out_req:
1139         ptlrpc_req_finished(request);
1140         return rc;
1141 }
1142
1143 int mdc_brw(int rw, struct obd_export *exp, struct obdo *oa,
1144                 struct lov_stripe_md *ea, obd_count oa_bufs,
1145                 struct brw_page *pgarr, struct obd_trans_info *oti)
1146 {
1147         struct ptlrpc_bulk_desc *desc;
1148         struct niobuf_remote *niobuf;
1149         struct ptlrpc_request *req;
1150         struct obd_ioobj *ioobj;
1151         struct ost_body *body;
1152         int err, opc, i;
1153         int size[3];
1154
1155         opc = ((rw & OBD_BRW_WRITE) != 0) ? OST_WRITE : OST_READ;
1156
1157         size[0] = sizeof(*body);
1158         size[1] = sizeof(*ioobj);
1159         size[2] = oa_bufs * sizeof(*niobuf);
1160
1161         req = ptlrpc_prep_req(class_exp2cliimp(exp), opc, 3, size, NULL);
1162         LASSERT(req != NULL);
1163
1164         if (opc == OST_WRITE)
1165                 desc = ptlrpc_prep_bulk_imp(req, oa_bufs, BULK_GET_SOURCE,
1166                                             OST_BULK_PORTAL);
1167         else
1168                 desc = ptlrpc_prep_bulk_imp(req, oa_bufs, BULK_PUT_SINK,
1169                                             OST_BULK_PORTAL);
1170         LASSERT(desc != NULL);
1171
1172         body = lustre_msg_buf(req->rq_reqmsg, 0, sizeof(*body));
1173         ioobj = lustre_msg_buf(req->rq_reqmsg, 1, sizeof(*ioobj));
1174         niobuf = lustre_msg_buf(req->rq_reqmsg, 2, oa_bufs * sizeof(*niobuf));
1175
1176         memcpy(&body->oa, oa, sizeof(*oa));
1177         obdo_to_ioobj(oa, ioobj);
1178         ioobj->ioo_bufcnt = oa_bufs;
1179
1180         for (i = 0; i < oa_bufs; i++, niobuf++) {
1181                 struct brw_page *pg = &pgarr[i];
1182
1183                 LASSERT(pg->count > 0);
1184                 LASSERT((pg->off & ~PAGE_MASK) + pg->count <= PAGE_SIZE);
1185
1186                 ptlrpc_prep_bulk_page(desc, pg->pg, pg->off & ~PAGE_MASK,
1187                                       pg->count);
1188
1189                 niobuf->offset = pg->off;
1190                 niobuf->len = pg->count;
1191                 niobuf->flags = pg->flag;
1192         }
1193
1194         /* size[0] still sizeof (*body) */
1195         if (opc == OST_WRITE) {
1196                 /* 1 RC per niobuf */
1197                 size[1] = sizeof(__u32) * oa_bufs;
1198                 req->rq_replen = lustre_msg_size(2, size);
1199         } else {
1200                 /* 1 RC for the whole I/O */
1201                 req->rq_replen = lustre_msg_size(1, size);
1202         }
1203         err = ptlrpc_queue_wait(req);
1204         LASSERT(err == 0);
1205
1206         ptlrpc_req_finished(req);
1207         return 0;
1208 }
1209
1210 static int mdc_valid_attrs(struct obd_export *exp, struct ll_fid *fid)
1211 {
1212         struct ldlm_res_id res_id = { .name = {0} };
1213         struct obd_device *obd = exp->exp_obd;
1214         struct lustre_handle lockh;
1215         ldlm_policy_data_t policy;
1216         int flags;
1217         ENTRY;
1218
1219         res_id.name[0] = fid->id;
1220         res_id.name[1] = fid->generation;
1221         policy.l_inodebits.bits = MDS_INODELOCK_UPDATE;
1222
1223         CDEBUG(D_INFO, "trying to match res "LPU64"\n", res_id.name[0]);
1224
1225         /* FIXME use LDLM_FL_TEST_LOCK instead */
1226         flags = LDLM_FL_BLOCK_GRANTED | LDLM_FL_CBPENDING;
1227         if (ldlm_lock_match(obd->obd_namespace, flags, &res_id,
1228                             LDLM_IBITS, &policy, LCK_PR, &lockh)) {
1229                 ldlm_lock_decref(&lockh, LCK_PR);
1230                 RETURN(1);
1231         }
1232
1233         if (ldlm_lock_match(obd->obd_namespace, flags, &res_id,
1234                             LDLM_IBITS, &policy, LCK_PW, &lockh)) {
1235                 ldlm_lock_decref(&lockh, LCK_PW);
1236                 RETURN(1);
1237         }
1238         RETURN(0);
1239 }
1240 static int mdc_change_cbdata_name(struct obd_export *exp, struct ll_fid *pfid,
1241                                   char *name, int len, struct ll_fid *cfid,
1242                                   ldlm_iterator_t it, void *data)
1243 {
1244         int rc;
1245         rc = mdc_change_cbdata(exp, cfid, it, data);
1246         RETURN(rc);
1247 }
1248
1249 struct obd_ops mdc_obd_ops = {
1250         .o_owner         = THIS_MODULE,
1251         .o_attach        = mdc_attach,
1252         .o_detach        = mdc_detach,
1253         .o_setup         = mdc_setup,
1254         .o_precleanup    = mdc_precleanup,
1255         .o_cleanup       = mdc_cleanup,
1256         .o_connect       = client_connect_import,
1257         .o_disconnect    = client_disconnect_export,
1258         .o_iocontrol     = mdc_iocontrol,
1259         .o_set_info      = mdc_set_info,
1260         .o_statfs        = mdc_statfs,
1261         .o_pin           = mdc_pin,
1262         .o_unpin         = mdc_unpin,
1263         .o_import_event  = mdc_import_event,
1264         .o_llog_init     = mdc_llog_init,
1265         .o_llog_finish   = mdc_llog_finish,
1266         .o_create        = mdc_obj_create,
1267         .o_set_info      = mdc_set_info,
1268         .o_get_info      = mdc_get_info,
1269         .o_brw           = mdc_brw,
1270         .o_init_ea_size  = mdc_init_ea_size,
1271 };
1272
1273 struct md_ops mdc_md_ops = {
1274         .m_getstatus     = mdc_getstatus,
1275         .m_getattr       = mdc_getattr,
1276         .m_change_cbdata = mdc_change_cbdata,
1277         .m_close         = mdc_close,
1278         .m_create        = mdc_create,
1279         .m_done_writing  = mdc_done_writing,
1280         .m_enqueue       = mdc_enqueue,
1281         .m_getattr_name  = mdc_getattr_name,
1282         .m_intent_lock   = mdc_intent_lock,
1283         .m_link          = mdc_link,
1284         .m_rename        = mdc_rename,
1285         .m_setattr       = mdc_setattr,
1286         .m_sync          = mdc_sync,
1287         .m_readpage      = mdc_readpage,
1288         .m_unlink        = mdc_unlink,
1289         .m_valid_attrs   = mdc_valid_attrs,
1290         .m_req2lustre_md = mdc_req2lustre_md,
1291         .m_set_open_replay_data   = mdc_set_open_replay_data,
1292         .m_clear_open_replay_data = mdc_clear_open_replay_data,
1293         .m_store_inode_generation = mdc_store_inode_generation,
1294         .m_set_lock_data = mdc_set_lock_data,
1295         .m_get_real_obd  = mdc_get_real_obd,
1296         .m_change_cbdata_name = mdc_change_cbdata_name,
1297         .m_change_cbdata = mdc_change_cbdata,
1298 };
1299
1300 int __init mdc_init(void)
1301 {
1302         struct lprocfs_static_vars lvars;
1303         lprocfs_init_vars(mdc, &lvars);
1304         return class_register_type(&mdc_obd_ops, &mdc_md_ops, lvars.module_vars,
1305                                    LUSTRE_MDC_NAME);
1306 }
1307
1308 #ifdef __KERNEL__
1309 static void /*__exit*/ mdc_exit(void)
1310 {
1311         class_unregister_type(LUSTRE_MDC_NAME);
1312 }
1313
1314 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
1315 MODULE_DESCRIPTION("Lustre Metadata Client");
1316 MODULE_LICENSE("GPL");
1317
1318 EXPORT_SYMBOL(mdc_req2lustre_md);
1319 EXPORT_SYMBOL(mdc_change_cbdata);
1320 EXPORT_SYMBOL(mdc_getstatus);
1321 EXPORT_SYMBOL(mdc_getattr);
1322 EXPORT_SYMBOL(mdc_getattr_name);
1323 EXPORT_SYMBOL(mdc_create);
1324 EXPORT_SYMBOL(mdc_unlink);
1325 EXPORT_SYMBOL(mdc_rename);
1326 EXPORT_SYMBOL(mdc_link);
1327 EXPORT_SYMBOL(mdc_readpage);
1328 EXPORT_SYMBOL(mdc_setattr);
1329 EXPORT_SYMBOL(mdc_close);
1330 EXPORT_SYMBOL(mdc_done_writing);
1331 EXPORT_SYMBOL(mdc_sync);
1332 EXPORT_SYMBOL(mdc_set_open_replay_data);
1333 EXPORT_SYMBOL(mdc_clear_open_replay_data);
1334 EXPORT_SYMBOL(mdc_store_inode_generation);
1335
1336 module_init(mdc_init);
1337 module_exit(mdc_exit);
1338 #endif