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) 2001-2003 Cluster File Systems, Inc.
   5  *
   6  *   This file is part of the Lustre file system, http://www.lustre.org
   7  *   Lustre is a trademark of Cluster File Systems, Inc.
   8  *
   9  *   You may have signed or agreed to another license before downloading
  10  *   this software.  If so, you are bound by the terms and conditions
  11  *   of that agreement, and the following does not apply to you.  See the
  12  *   LICENSE file included with this distribution for more information.
  13  *
  14  *   If you did not agree to a different license, then this copy of Lustre
  15  *   is open source software; you can redistribute it and/or modify it
  16  *   under the terms of version 2 of the GNU General Public License as
  17  *   published by the Free Software Foundation.
  18  *
  19  *   In either case, Lustre is distributed in the hope that it will be
  20  *   useful, but WITHOUT ANY WARRANTY; without even the implied warranty
  21  *   of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  22  *   license text for more details.
  23  */
  24
  25 #ifndef EXPORT_SYMTAB
  26 # define EXPORT_SYMTAB
  27 #endif
  28 #define DEBUG_SUBSYSTEM S_MDC
  29
  30 #ifdef __KERNEL__
  31 # include <linux/module.h>
  32 # include <linux/pagemap.h>
  33 # include <linux/miscdevice.h>
  34 # include <linux/init.h>
  35 #else
  36 # include <liblustre.h>
  37 #endif
  38
  39 #include <linux/lustre_acl.h>
  40 #include <obd_class.h>
  41 #include <lustre_dlm.h>
  42 #include <lustre_fid.h>
  43 #include <md_object.h>
  44 #include <lprocfs_status.h>
  45 #include <lustre_param.h>
  46 #include "mdc_internal.h"
  47
  48 static quota_interface_t *quota_interface;
  49
  50 #define REQUEST_MINOR 244
  51
  52 static quota_interface_t *quota_interface;
  53 extern quota_interface_t mdc_quota_interface;
  54
  55 static int mdc_cleanup(struct obd_device *obd);
  56
  57 static struct obd_capa *mdc_unpack_capa(struct ptlrpc_request *req,
  58                                                unsigned int offset)
  59 {
  60         struct lustre_capa *capa;
  61         struct obd_capa *oc;
  62
  63         /* swabbed already in mdc_enqueue */
  64         capa = lustre_msg_buf(req->rq_repmsg, offset, sizeof(*capa));
  65         if (capa == NULL) {
  66                 CERROR("missing capa at offset %d failed!\n", offset);
  67                 return ERR_PTR(-EFAULT);
  68         }
  69
  70         oc = alloc_capa(CAPA_SITE_CLIENT);
  71         if (!oc) {
  72                 CERROR("alloc capa failed!\n");
  73                 return ERR_PTR(-ENOMEM);
  74         }
  75         oc->c_capa = *capa;
  76
  77         return oc;
  78 }
  79
  80 /* Helper that implements most of mdc_getstatus and signal_completed_replay. */
  81 /* XXX this should become mdc_get_info("key"), sending MDS_GET_INFO RPC */
  82 static int send_getstatus(struct obd_import *imp, struct lu_fid *rootfid,
  83                           struct obd_capa **pc, int level, int msg_flags)
  84 {
  85         struct ptlrpc_request *req;
  86         struct mdt_body *body;
  87         int rc, size[3] = { sizeof(struct ptlrpc_body),
  88                             sizeof(*body),
  89                             sizeof(struct lustre_capa) };
  90         ENTRY;
  91
  92         req = ptlrpc_prep_req(imp, LUSTRE_MDS_VERSION, MDS_GETSTATUS, 2, size,
  93                               NULL);
  94         if (!req)
  95                 GOTO(out, rc = -ENOMEM);
  96
  97         req->rq_send_state = level;
  98         ptlrpc_req_set_repsize(req, 3, size);
  99
 100         mdc_pack_req_body(req, REQ_REC_OFF, 0, NULL, NULL, 0, 0);
 101         lustre_msg_add_flags(req->rq_reqmsg, msg_flags);
 102         rc = ptlrpc_queue_wait(req);
 103
 104         if (!rc) {
 105                 body = lustre_swab_repbuf(req, REPLY_REC_OFF, sizeof(*body),
 106                                           lustre_swab_mdt_body);
 107                 if (body == NULL) {
 108                         CERROR ("Can't extract mdt_body\n");
 109                         GOTO (out, rc = -EPROTO);
 110                 }
 111
 112                 *rootfid = body->fid1;
 113
 114                 if (body->valid & OBD_MD_FLMDSCAPA) {
 115                         struct obd_capa *oc;
 116
 117                         oc = mdc_unpack_capa(req, REPLY_REC_OFF + 1);
 118                         if (IS_ERR(oc))
 119                                 GOTO(out, rc = PTR_ERR(oc));
 120                         *pc = oc;
 121                 }
 122
 123                 CDEBUG(D_NET, "root fid="DFID", last_committed="LPU64
 124                        ", last_xid="LPU64"\n",
 125                        PFID(rootfid),
 126                        lustre_msg_get_last_committed(req->rq_repmsg),
 127                        lustre_msg_get_last_xid(req->rq_repmsg));
 128         }
 129
 130         EXIT;
 131  out:
 132         ptlrpc_req_finished(req);
 133         return rc;
 134 }
 135
 136 /* This should be mdc_get_info("rootfid") */
 137 int mdc_getstatus(struct obd_export *exp, struct lu_fid *rootfid,
 138                   struct obd_capa **pc)
 139 {
 140         return send_getstatus(class_exp2cliimp(exp), rootfid, pc,
 141                               LUSTRE_IMP_FULL, 0);
 142 }
 143
 144 /*
 145  * This function now is known to always saying that it will receive 4 buffers
 146  * from server. Even for cases when acl_size and md_size is zero, RPC header
 147  * willcontain 4 fields and RPC itself will contain zero size fields. This is
 148  * because mdt_getattr*() _always_ returns 4 fields, but if acl is not needed
 149  * and thus zero, it shirinks it, making zero size. The same story about
 150  * md_size. And this is course of problem when client waits for smaller number
 151  * of fields. This issue will be fixed later when client gets awar of RPC
 152  * layouts.  --umka
 153  */
 154 static int mdc_getattr_common(struct obd_export *exp, unsigned int ea_size,
 155                               unsigned int acl_size, int mdscapa,
 156                               struct ptlrpc_request *req)
 157 {
 158         struct mdt_body *body;
 159         void *eadata;
 160         int size[5] = { sizeof(struct ptlrpc_body),
 161                         sizeof(*body),
 162                         ea_size,
 163                         acl_size,
 164                         sizeof(struct lustre_capa) };
 165         int offset, rc;
 166         ENTRY;
 167
 168         /* Request message already built. */
 169         if (ea_size)
 170                 CDEBUG(D_INODE, "reserved %u bytes for MD/symlink in packet\n",
 171                        ea_size);
 172         if (acl_size)
 173                 CDEBUG(D_INODE, "reserved %u bytes for ACL\n", acl_size);
 174
 175         ptlrpc_req_set_repsize(req, 5, size);
 176
 177         rc = ptlrpc_queue_wait(req);
 178         if (rc != 0)
 179                 RETURN (rc);
 180
 181         body = lustre_swab_repbuf(req, REPLY_REC_OFF, sizeof(*body),
 182                                   lustre_swab_mdt_body);
 183         if (body == NULL) {
 184                 CERROR ("Can't unpack mdt_body\n");
 185                 RETURN (-EPROTO);
 186         }
 187
 188         CDEBUG(D_NET, "mode: %o\n", body->mode);
 189
 190         offset = REPLY_REC_OFF + 1;
 191         LASSERT_REPSWAB(req, offset);
 192         if (body->eadatasize != 0) {
 193                 /* reply indicates presence of eadata; check it's there... */
 194                 eadata = lustre_msg_buf(req->rq_repmsg, offset++,
 195                                         body->eadatasize);
 196                 if (eadata == NULL) {
 197                         CERROR ("Missing/short eadata\n");
 198                         RETURN (-EPROTO);
 199                 }
 200         }
 201
 202         if (body->valid & OBD_MD_FLMODEASIZE) {
 203                 struct client_obd *cli = &exp->exp_obd->u.cli;
 204
 205                 if (cli->cl_max_mds_easize < body->max_mdsize)
 206                         cli->cl_max_mds_easize = body->max_mdsize;
 207                 if (cli->cl_max_mds_cookiesize < body->max_cookiesize)
 208                         cli->cl_max_mds_cookiesize = body->max_cookiesize;
 209         }
 210
 211         offset += !!body->aclsize;
 212
 213         if (body->valid & OBD_MD_FLMDSCAPA) {
 214                 struct lustre_capa *capa;
 215
 216                 LASSERT(mdscapa);
 217                 capa = lustre_unpack_capa(req->rq_repmsg, offset++);
 218                 if (capa == NULL) {
 219                         CERROR("Missing/short client MDS capability\n");
 220                         RETURN(-EPROTO);
 221                 }
 222         }
 223
 224         RETURN (0);
 225 }
 226
 227 int mdc_getattr(struct obd_export *exp, const struct lu_fid *fid,
 228                 struct obd_capa *oc, obd_valid valid, int ea_size,
 229                 struct ptlrpc_request **request)
 230 {
 231         struct ptlrpc_request *req;
 232         int size[3] = { sizeof(struct ptlrpc_body), sizeof(struct mdt_body) };
 233         int acl_size = 0, rc;
 234         ENTRY;
 235
 236         size[REQ_REC_OFF + 1] = oc ? sizeof(struct lustre_capa) : 0;
 237
 238         /*
 239          * XXX: Do we need to make another request here?  We just did a getattr
 240          * to do the lookup in the first place.
 241          */
 242         req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
 243                               MDS_GETATTR, 3, size, NULL);
 244         if (!req)
 245                 GOTO(out, rc = -ENOMEM);
 246
 247         mdc_pack_req_body(req, REQ_REC_OFF, valid, fid, oc, ea_size,
 248                           MDS_BFLAG_EXT_FLAGS/*request "new" flags(bug 9486)*/);
 249
 250         if (valid & OBD_MD_FLRMTPERM)
 251                 acl_size = sizeof(struct mdt_remote_perm);
 252
 253         /* Currently only root inode will call us with FLACL */
 254         else if (valid & OBD_MD_FLACL)
 255                 acl_size = LUSTRE_POSIX_ACL_MAX_SIZE;
 256
 257         rc = mdc_getattr_common(exp, ea_size, acl_size,
 258                                 !!(valid & OBD_MD_FLMDSCAPA), req);
 259         if (rc != 0) {
 260                 ptlrpc_req_finished (req);
 261                 req = NULL;
 262         }
 263  out:
 264         *request = req;
 265         RETURN (rc);
 266 }
 267
 268 int mdc_getattr_name(struct obd_export *exp, const struct lu_fid *fid,
 269                      struct obd_capa *oc, const char *filename, int namelen,
 270                      obd_valid valid, int ea_size,
 271                      struct ptlrpc_request **request)
 272 {
 273         struct ptlrpc_request *req;
 274         struct mdt_body *body;
 275         int size[4] = { sizeof(struct ptlrpc_body), sizeof(*body), 0, namelen};
 276         int rc;
 277         ENTRY;
 278
 279         size[REQ_REC_OFF + 1] = oc ? sizeof(struct lustre_capa) : 0;
 280
 281         req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
 282                               MDS_GETATTR_NAME, 4, size, NULL);
 283         if (!req)
 284                 GOTO(out, rc = -ENOMEM);
 285
 286         mdc_pack_req_body(req, REQ_REC_OFF, valid, fid, oc, ea_size,
 287                           MDS_BFLAG_EXT_FLAGS/*request "new" flags(bug 9486)*/);
 288
 289         if (filename) {
 290                 LASSERT(strnlen(filename, namelen) == namelen - 1);
 291                 memcpy(lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF + 2, namelen),
 292                        filename, namelen);
 293         }
 294
 295         rc = mdc_getattr_common(exp, ea_size, 0, !!(valid & OBD_MD_FLMDSCAPA),
 296                                 req);
 297         if (rc != 0) {
 298                 ptlrpc_req_finished (req);
 299                 req = NULL;
 300         }
 301  out:
 302         *request = req;
 303         RETURN(rc);
 304 }
 305
 306 static int mdc_is_subdir(struct obd_export *exp, const struct lu_fid *pfid,
 307                          const struct lu_fid *cfid, struct ptlrpc_request **request)
 308 {
 309         int size[2] = { sizeof(struct ptlrpc_body),
 310                         sizeof(struct mdt_body) };
 311         struct ptlrpc_request *req;
 312         struct mdt_body *body;
 313         int rc;
 314         ENTRY;
 315
 316         req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
 317                               MDS_IS_SUBDIR, 2, size, NULL);
 318         if (!req)
 319                 GOTO(out, rc = -ENOMEM);
 320
 321         mdc_is_subdir_pack(req, REQ_REC_OFF, pfid, cfid, 0);
 322
 323         ptlrpc_req_set_repsize(req, 2, size);
 324         rc = ptlrpc_queue_wait(req);
 325         if (rc != 0 && rc != -EREMOTE)
 326                 GOTO(out, rc);
 327
 328         body = lustre_swab_repbuf(req, REPLY_REC_OFF, sizeof(*body),
 329                                   lustre_swab_mdt_body);
 330         if (body == NULL) {
 331                 CERROR ("Can't unpack mdt_body\n");
 332                 GOTO(out, rc = -EPROTO);
 333         }
 334         EXIT;
 335  out:
 336         *request = req;
 337         return rc;
 338 }
 339
 340 static
 341 int mdc_xattr_common(struct obd_export *exp, const struct lu_fid *fid,
 342                      struct obd_capa *oc,
 343                      int opcode, obd_valid valid, const char *xattr_name,
 344                      const char *input, int input_size, int output_size,
 345                      int flags, struct ptlrpc_request **request)
 346 {
 347         struct ptlrpc_request *req;
 348         int size[5] = { sizeof(struct ptlrpc_body), sizeof(struct mdt_body) };
 349         int bufcnt = 3, offset = REQ_REC_OFF + 2;
 350         int rc, xattr_namelen = 0, remote_acl = 0;
 351         void *tmp;
 352         ENTRY;
 353
 354         size[REQ_REC_OFF + 1] = oc ? sizeof(struct lustre_capa) : 0;
 355         if (xattr_name) {
 356                 xattr_namelen = strlen(xattr_name) + 1;
 357                 size[bufcnt++] = xattr_namelen;
 358         }
 359         if (input_size) {
 360                 LASSERT(input);
 361                 size[bufcnt++] = input_size;
 362         }
 363
 364         req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
 365                               opcode, bufcnt, size, NULL);
 366         if (!req)
 367                 GOTO(out, rc = -ENOMEM);
 368
 369         /* request data */
 370         mdc_pack_req_body(req, REQ_REC_OFF, valid, fid, oc, output_size, flags);
 371
 372
 373         if (xattr_name) {
 374                 tmp = lustre_msg_buf(req->rq_reqmsg, offset++, xattr_namelen);
 375                 memcpy(tmp, xattr_name, xattr_namelen);
 376                 if (!strcmp(xattr_name, XATTR_NAME_LUSTRE_ACL))
 377                         remote_acl = 1;
 378         }
 379         if (input_size) {
 380                 tmp = lustre_msg_buf(req->rq_reqmsg, offset++, input_size);
 381                 memcpy(tmp, input, input_size);
 382         }
 383
 384         /* reply buffers */
 385         if (opcode == MDS_GETXATTR) {
 386                 size[REPLY_REC_OFF] = sizeof(struct mdt_body);
 387                 bufcnt = 2;
 388         } else {
 389                 bufcnt = 1;
 390         }
 391
 392         /* we do this even output_size is 0, because server is doing that */
 393         size[bufcnt++] = output_size;
 394         ptlrpc_req_set_repsize(req, bufcnt, size);
 395
 396         /* make rpc */
 397         /* NB: set remote acl doesn't need hold rpc lock, because it just
 398          * send command to MDS, and when it's executed on mountpoint on MDS,
 399          * another mdc_xattr_common() will be invoked there. */
 400         if (opcode == MDS_SETXATTR && !remote_acl)
 401                 mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
 402
 403         rc = ptlrpc_queue_wait(req);
 404
 405         if (opcode == MDS_SETXATTR && !remote_acl)
 406                 mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
 407
 408         if (rc != 0)
 409                 GOTO(err_out, rc);
 410
 411         if (opcode == MDS_GETXATTR) {
 412                 struct mdt_body * body = lustre_swab_repbuf(req, REPLY_REC_OFF,
 413                                           sizeof(*body), lustre_swab_mdt_body);
 414                 if (body == NULL) {
 415                         CERROR ("Can't unpack mdt_body\n");
 416                         GOTO(err_out, rc = -EPROTO);
 417                 }
 418         }
 419 out:
 420         *request = req;
 421         RETURN (rc);
 422 err_out:
 423         ptlrpc_req_finished(req);
 424         req = NULL;
 425         goto out;
 426 }
 427
 428 int mdc_setxattr(struct obd_export *exp, const struct lu_fid *fid,
 429                  struct obd_capa *oc, obd_valid valid, const char *xattr_name,
 430                  const char *input, int input_size, int output_size, int flags,
 431                  struct ptlrpc_request **request)
 432 {
 433         return mdc_xattr_common(exp, fid, oc, MDS_SETXATTR, valid, xattr_name,
 434                                 input, input_size, output_size, flags, request);
 435 }
 436
 437 int mdc_getxattr(struct obd_export *exp, const struct lu_fid *fid,
 438                  struct obd_capa *oc, obd_valid valid, const char *xattr_name,
 439                  const char *input, int input_size, int output_size, int flags,
 440                  struct ptlrpc_request **request)
 441 {
 442         return mdc_xattr_common(exp, fid, oc, MDS_GETXATTR, valid, xattr_name,
 443                                 input, input_size, output_size, flags, request);
 444 }
 445
 446 #ifdef CONFIG_FS_POSIX_ACL
 447 static
 448 int mdc_unpack_acl(struct obd_export *exp, struct ptlrpc_request *req,
 449                    struct lustre_md *md, unsigned int offset)
 450 {
 451         struct mdt_body  *body = md->body;
 452         struct posix_acl *acl;
 453         void             *buf;
 454         int               rc;
 455
 456         if (!body->aclsize)
 457                 return 0;
 458
 459         buf = lustre_msg_buf(req->rq_repmsg, offset, body->aclsize);
 460         if (!buf) {
 461                 CERROR("aclsize %u, bufcount %u, bufsize %u\n",
 462                        body->aclsize, lustre_msg_bufcount(req->rq_repmsg),
 463                        (lustre_msg_bufcount(req->rq_repmsg) <= offset) ?
 464                                 -1 : lustre_msg_buflen(req->rq_repmsg, offset));
 465                 return -EPROTO;
 466         }
 467
 468         acl = posix_acl_from_xattr(buf, body->aclsize);
 469         if (IS_ERR(acl)) {
 470                 rc = PTR_ERR(acl);
 471                 CERROR("convert xattr to acl: %d\n", rc);
 472                 return rc;
 473         }
 474
 475         rc = posix_acl_valid(acl);
 476         if (rc) {
 477                 CERROR("validate acl: %d\n", rc);
 478                 posix_acl_release(acl);
 479                 return rc;
 480         }
 481
 482         md->posix_acl = acl;
 483         return 0;
 484 }
 485 #else
 486 #define mdc_unpack_acl(exp, req, md, offset) 0
 487 #endif
 488
 489 int mdc_get_lustre_md(struct obd_export *exp, struct ptlrpc_request *req,
 490                       int offset, struct obd_export *dt_exp,
 491                       struct obd_export *md_exp,
 492                       struct lustre_md *md)
 493 {
 494         int rc;
 495         ENTRY;
 496
 497         LASSERT(md);
 498         memset(md, 0, sizeof(*md));
 499
 500         md->body = lustre_msg_buf(req->rq_repmsg, offset, sizeof (*md->body));
 501         LASSERT (md->body != NULL);
 502         LASSERT_REPSWABBED(req, offset);
 503         offset++;
 504
 505         if (md->body->valid & OBD_MD_FLEASIZE) {
 506                 int lmmsize;
 507                 struct lov_mds_md *lmm;
 508
 509                 if (!S_ISREG(md->body->mode)) {
 510                         CERROR("OBD_MD_FLEASIZE set, should be a regular file, "
 511                                "but is not\n");
 512                         GOTO(out, rc = -EPROTO);
 513                 }
 514
 515                 if (md->body->eadatasize == 0) {
 516                         CERROR("OBD_MD_FLEASIZE set, but eadatasize 0\n");
 517                         GOTO(out, rc = -EPROTO);
 518                 }
 519                 lmmsize = md->body->eadatasize;
 520                 lmm = lustre_msg_buf(req->rq_repmsg, offset, lmmsize);
 521                 if (!lmm) {
 522                         CERROR ("incorrect message: lmm == 0\n");
 523                         GOTO(out, rc = -EPROTO);
 524                 }
 525                 LASSERT_REPSWABBED(req, offset);
 526
 527                 rc = obd_unpackmd(dt_exp, &md->lsm, lmm, lmmsize);
 528                 if (rc < 0)
 529                         GOTO(out, rc);
 530
 531                 if (rc < sizeof(*md->lsm)) {
 532                         CERROR ("lsm size too small:  rc < sizeof (*md->lsm) "
 533                                 "(%d < %d)\n", rc, sizeof(*md->lsm));
 534                         GOTO(out, rc = -EPROTO);
 535                 }
 536
 537                 offset++;
 538         } else if (md->body->valid & OBD_MD_FLDIREA) {
 539                 int lmvsize;
 540                 struct lov_mds_md *lmv;
 541
 542                 if(!S_ISDIR(md->body->mode)) {
 543                         CERROR("OBD_MD_FLDIREA set, should be a directory, but "
 544                                "is not\n");
 545                         GOTO(out, rc = -EPROTO);
 546                 }
 547
 548                 if (md->body->eadatasize == 0) {
 549                         CERROR("OBD_MD_FLDIREA is set, but eadatasize 0\n");
 550                         RETURN(-EPROTO);
 551                 }
 552                 if (md->body->valid & OBD_MD_MEA) {
 553                         lmvsize = md->body->eadatasize;
 554                         lmv = lustre_msg_buf(req->rq_repmsg, offset, lmvsize);
 555                         if (!lmv) {
 556                                 CERROR ("incorrect message: lmv == 0\n");
 557                                 GOTO(out, rc = -EPROTO);
 558                         }
 559
 560                         LASSERT_REPSWABBED(req, offset);
 561
 562                         rc = obd_unpackmd(md_exp, (void *)&md->mea, lmv,
 563                                           lmvsize);
 564                         if (rc < 0)
 565                                 GOTO(out, rc);
 566
 567                         if (rc < sizeof(*md->mea)) {
 568                                 CERROR ("size too small:  rc < sizeof(*md->mea) "
 569                                         "(%d < %d)\n", rc, sizeof(*md->mea));
 570                                 GOTO(out, rc = -EPROTO);
 571                         }
 572                 }
 573                 offset++;
 574         }
 575         rc = 0;
 576
 577         /* remote permission */
 578         if (md->body->valid & OBD_MD_FLRMTPERM) {
 579                 md->remote_perm = lustre_msg_buf(req->rq_repmsg, offset++,
 580                                                 sizeof(struct mdt_remote_perm));
 581                 if (!md->remote_perm) {
 582                         CERROR ("incorrect message: remote_perm == 0\n");
 583                         GOTO(out, rc = -EPROTO);
 584                 }
 585         }
 586
 587         /* for ACL, it's possible that FLACL is set but aclsize is zero.  only
 588          * when aclsize != 0 there's an actual segment for ACL in reply
 589          * buffer. */
 590         else if (md->body->valid & OBD_MD_FLACL) {
 591                 if (md->body->aclsize) {
 592                         rc = mdc_unpack_acl(dt_exp, req, md, offset++);
 593                         if (rc)
 594                                 GOTO(out, rc);
 595 #ifdef CONFIG_FS_POSIX_ACL
 596                 } else {
 597                         md->posix_acl = NULL;
 598 #endif
 599                 }
 600         }
 601
 602         if (md->body->valid & OBD_MD_FLMDSCAPA) {
 603                 struct obd_capa *oc = mdc_unpack_capa(req, offset++);
 604
 605                 if (IS_ERR(oc))
 606                         GOTO(out, rc = PTR_ERR(oc));
 607                 md->mds_capa = oc;
 608         }
 609
 610         if (md->body->valid & OBD_MD_FLOSSCAPA) {
 611                 struct obd_capa *oc = mdc_unpack_capa(req, offset++);
 612
 613                 if (IS_ERR(oc))
 614                         GOTO(out, rc = PTR_ERR(oc));
 615                 md->oss_capa = oc;
 616         }
 617
 618         EXIT;
 619 out:
 620         if (rc) {
 621                 if (md->oss_capa)
 622                         free_capa(md->oss_capa);
 623                 if (md->mds_capa)
 624                         free_capa(md->mds_capa);
 625 #ifdef CONFIG_FS_POSIX_ACL
 626                 posix_acl_release(md->posix_acl);
 627 #endif
 628                 if (md->lsm)
 629                         obd_free_memmd(dt_exp, &md->lsm);
 630         }
 631         return rc;
 632 }
 633
 634 int mdc_free_lustre_md(struct obd_export *exp, struct lustre_md *md)
 635 {
 636         ENTRY;
 637         RETURN(0);
 638 }
 639
 640 static void mdc_replay_open(struct ptlrpc_request *req)
 641 {
 642         struct md_open_data *mod = req->rq_cb_data;
 643         struct ptlrpc_request *cur, *tmp;
 644         struct obd_client_handle *och;
 645         struct lustre_handle old;
 646         struct mdt_body *body;
 647         ENTRY;
 648
 649         if (mod == NULL) {
 650                 DEBUG_REQ(D_ERROR, req,
 651                           "Can't properly replay without open data.");
 652                 EXIT;
 653                 return;
 654         }
 655
 656         body = lustre_swab_repbuf(req, DLM_REPLY_REC_OFF, sizeof(*body),
 657                                   lustre_swab_mdt_body);
 658         LASSERT(body != NULL);
 659
 660         och = mod->mod_och;
 661         if (och != NULL) {
 662                 struct lustre_handle *file_fh;
 663
 664                 LASSERT(och->och_magic == OBD_CLIENT_HANDLE_MAGIC);
 665
 666                 file_fh = &och->och_fh;
 667                 CDEBUG(D_HA, "updating handle from "LPX64" to "LPX64"\n",
 668                        file_fh->cookie, body->handle.cookie);
 669                 old = *file_fh;
 670                 *file_fh = body->handle;
 671         }
 672
 673         list_for_each_entry_safe(cur, tmp, &mod->mod_replay_list, rq_mod_list) {
 674                 int opc = lustre_msg_get_opc(cur->rq_reqmsg);
 675                 struct mdt_epoch *epoch = NULL;
 676
 677                 if (opc == MDS_CLOSE || opc == MDS_DONE_WRITING) {
 678                         epoch = lustre_msg_buf(cur->rq_reqmsg,
 679                                                REQ_REC_OFF, sizeof(*epoch));
 680                         LASSERT(epoch);
 681                         DEBUG_REQ(D_HA, cur, "updating %s body with new fh",
 682                                   opc == MDS_CLOSE ? "CLOSE" : "DONE_WRITING");
 683                 } else if (opc == MDS_REINT) {
 684                         struct mdt_rec_setattr *rec;
 685
 686                         /* Check this is REINT_SETATTR. */
 687                         rec = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF,
 688                                              sizeof (*rec));
 689                         LASSERT(rec && rec->sa_opcode == REINT_SETATTR);
 690
 691                         epoch = lustre_msg_buf(cur->rq_reqmsg,
 692                                                REQ_REC_OFF + 2, sizeof(*epoch));
 693                         LASSERT(epoch);
 694                         DEBUG_REQ(D_HA, cur, "updating REINT_SETATTR body "
 695                                   "with new fh");
 696                 }
 697                 if (epoch) {
 698                         if (och != NULL)
 699                                 LASSERT(!memcmp(&old, &epoch->handle,
 700                                                 sizeof(old)));
 701                         epoch->handle = body->handle;
 702                 }
 703         }
 704         EXIT;
 705 }
 706
 707 void mdc_commit_delayed(struct ptlrpc_request *req)
 708 {
 709         struct md_open_data *mod = req->rq_cb_data;
 710         struct ptlrpc_request *cur, *tmp;
 711
 712         DEBUG_REQ(D_HA, req, "req committed");
 713
 714         if (mod == NULL)
 715                 return;
 716
 717         req->rq_cb_data = NULL;
 718         req->rq_commit_cb = NULL;
 719         list_del_init(&req->rq_mod_list);
 720         if (req->rq_sequence) {
 721                 list_for_each_entry_safe(cur, tmp, &mod->mod_replay_list,
 722                                          rq_mod_list)
 723                 {
 724                         LASSERT(cur != LP_POISON);
 725                         LASSERT(cur->rq_type != LI_POISON);
 726                         DEBUG_REQ(D_HA, cur, "req balanced");
 727                         LASSERT(cur->rq_transno != 0);
 728                         LASSERT(cur->rq_import == req->rq_import);
 729
 730                         list_del_init(&cur->rq_mod_list);
 731                         /* We no longer want to preserve this for transno-
 732                          * unconditional replay. */
 733                         spin_lock(&cur->rq_lock);
 734                         cur->rq_replay = 0;
 735                         spin_unlock(&cur->rq_lock);
 736                 }
 737         }
 738
 739         if (list_empty(&mod->mod_replay_list)) {
 740                 if (mod->mod_och != NULL)
 741                         mod->mod_och->och_mod = NULL;
 742
 743                 OBD_FREE_PTR(mod);
 744         }
 745 }
 746
 747 int mdc_set_open_replay_data(struct obd_export *exp,
 748                              struct obd_client_handle *och,
 749                              struct ptlrpc_request *open_req)
 750 {
 751         struct md_open_data *mod;
 752         struct mdt_rec_create *rec = lustre_msg_buf(open_req->rq_reqmsg,
 753                                                     DLM_INTENT_REC_OFF,
 754                                                     sizeof(*rec));
 755         struct mdt_body *body = lustre_msg_buf(open_req->rq_repmsg,
 756                                                DLM_REPLY_REC_OFF,
 757                                                sizeof(*body));
 758         struct obd_import *imp = open_req->rq_import;
 759         ENTRY;
 760
 761         LASSERT(rec != NULL);
 762
 763         /* Incoming message in my byte order (it's been swabbed). */
 764         LASSERT_REPSWABBED(open_req, DLM_REPLY_REC_OFF);
 765
 766         /* Outgoing messages always in my byte order. */
 767         LASSERT(body != NULL);
 768
 769         /*Only the import is replayable, we set replay_open data */
 770         if (och && imp->imp_replayable) {
 771                 OBD_ALLOC(mod, sizeof(*mod));
 772                 if (mod == NULL) {
 773                         DEBUG_REQ(D_ERROR, open_req,
 774                                   "Can't allocate md_open_data");
 775                         RETURN(0);
 776                 }
 777                 CFS_INIT_LIST_HEAD(&mod->mod_replay_list);
 778
 779                 spin_lock(&open_req->rq_lock);
 780                 if (!open_req->rq_replay) {
 781                         OBD_FREE(mod, sizeof(*mod));
 782                         spin_unlock(&open_req->rq_lock);
 783                         RETURN(0);
 784                 }
 785
 786                 och->och_mod = mod;
 787                 mod->mod_och = och;
 788                 open_req->rq_cb_data = mod;
 789                 list_add_tail(&open_req->rq_mod_list, &mod->mod_replay_list);
 790                 open_req->rq_commit_cb = mdc_commit_delayed;
 791                 spin_unlock(&open_req->rq_lock);
 792         }
 793
 794         rec->cr_fid2 = body->fid1;
 795         rec->cr_ioepoch = body->ioepoch;
 796         rec->cr_old_handle.cookie = body->handle.cookie;
 797         open_req->rq_replay_cb = mdc_replay_open;
 798         if (!fid_is_sane(&body->fid1)) {
 799                 DEBUG_REQ(D_ERROR, open_req, "Saving replay request with "
 800                           "insane fid");
 801                 LBUG();
 802         }
 803
 804         DEBUG_REQ(D_HA, open_req, "Set up open replay data");
 805         RETURN(0);
 806 }
 807
 808 int mdc_clear_open_replay_data(struct obd_export *exp,
 809                                struct obd_client_handle *och)
 810 {
 811         struct md_open_data *mod = och->och_mod;
 812         ENTRY;
 813
 814         /*
 815          * Don't free the structure now (it happens in mdc_commit_delayed(),
 816          * after the last request is removed from its replay list),
 817          * but make sure that replay doesn't poke at the och, which is about to
 818          * be freed.
 819          */
 820         LASSERT(mod != LP_POISON);
 821         if (mod != NULL)
 822                 mod->mod_och = NULL;
 823
 824         och->och_mod = NULL;
 825         RETURN(0);
 826 }
 827
 828 int mdc_close(struct obd_export *exp, struct md_op_data *op_data,
 829               struct md_open_data *mod, struct ptlrpc_request **request)
 830 {
 831         struct obd_device *obd = class_exp2obd(exp);
 832         int reqsize[4] = { sizeof(struct ptlrpc_body),
 833                            sizeof(struct mdt_epoch),
 834                            sizeof(struct mdt_rec_setattr)};
 835         int repsize[4] = { sizeof(struct ptlrpc_body),
 836                            sizeof(struct mdt_body),
 837                            obd->u.cli.cl_max_mds_easize,
 838                            obd->u.cli.cl_max_mds_cookiesize };
 839         struct ptlrpc_request *req;
 840         int rc;
 841         ENTRY;
 842
 843         reqsize[REQ_REC_OFF + 2] = op_data->op_capa1 ?
 844                                         sizeof(struct lustre_capa) : 0;
 845         req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
 846                               MDS_CLOSE, 4, reqsize, NULL);
 847         if (req == NULL)
 848                 GOTO(out, rc = -ENOMEM);
 849
 850         /* To avoid a livelock (bug 7034), we need to send CLOSE RPCs to a
 851          * portal whose threads are not taking any DLM locks and are therefore
 852          * always progressing */
 853         /* XXX FIXME bug 249 */
 854         req->rq_request_portal = MDS_READPAGE_PORTAL;
 855
 856         /* Ensure that this close's handle is fixed up during replay. */
 857         if (likely(mod != NULL))
 858                 list_add_tail(&req->rq_mod_list, &mod->mod_replay_list);
 859         else
 860                 CDEBUG(D_HA, "couldn't find open req; expecting close error\n");
 861
 862         mdc_close_pack(req, REQ_REC_OFF, op_data);
 863         ptlrpc_req_set_repsize(req, 4, repsize);
 864         req->rq_commit_cb = mdc_commit_delayed;
 865         req->rq_replay = 1;
 866         LASSERT(req->rq_cb_data == NULL);
 867         req->rq_cb_data = mod;
 868
 869         mdc_get_rpc_lock(obd->u.cli.cl_close_lock, NULL);
 870         rc = ptlrpc_queue_wait(req);
 871         mdc_put_rpc_lock(obd->u.cli.cl_close_lock, NULL);
 872
 873         if (req->rq_repmsg == NULL) {
 874                 CDEBUG(D_HA, "request failed to send: %p, %d\n", req,
 875                        req->rq_status);
 876                 if (rc == 0)
 877                         rc = req->rq_status ? req->rq_status : -EIO;
 878         } else if (rc == 0 || rc == -EAGAIN) {
 879                 rc = lustre_msg_get_status(req->rq_repmsg);
 880                 if (lustre_msg_get_type(req->rq_repmsg) == PTL_RPC_MSG_ERR) {
 881                         DEBUG_REQ(D_ERROR, req, "type == PTL_RPC_MSG_ERR, err "
 882                                   "= %d", rc);
 883                         if (rc > 0)
 884                                 rc = -rc;
 885                 } else if (mod == NULL) {
 886                         if (req->rq_import->imp_replayable)
 887                                 CERROR("Unexpected: can't find md_open_data,"
 888                                        "but close succeeded with replayable imp"
 889                                        "Please tell CFS.\n");
 890                 }
 891                 if (!lustre_swab_repbuf(req, REPLY_REC_OFF,
 892                                         sizeof(struct mdt_body),
 893                                         lustre_swab_mdt_body)) {
 894                         CERROR("Error unpacking mdt_body\n");
 895                         rc = -EPROTO;
 896                 }
 897         }
 898
 899         EXIT;
 900         *request = req;
 901  out:
 902         if (rc != 0 && rc != -EAGAIN && req && req->rq_commit_cb)
 903                 req->rq_commit_cb(req);
 904
 905         return rc;
 906 }
 907
 908 int mdc_done_writing(struct obd_export *exp, struct md_op_data *op_data,
 909                      struct md_open_data *mod)
 910 {
 911         struct obd_device *obd = class_exp2obd(exp);
 912         struct ptlrpc_request *req;
 913         int size[4] = { sizeof(struct ptlrpc_body),
 914                         sizeof(struct mdt_epoch),
 915                         sizeof(struct mdt_rec_setattr)};
 916         int repsize[2] = { sizeof(struct ptlrpc_body),
 917                            sizeof(struct mdt_body)};
 918         int rc;
 919         ENTRY;
 920
 921         if (op_data->op_capa1)
 922                 size[REQ_REC_OFF + 2] = sizeof(struct lustre_capa);
 923         req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
 924                               MDS_DONE_WRITING, 4, size, NULL);
 925         if (req == NULL)
 926                 RETURN(-ENOMEM);
 927
 928         mdc_close_pack(req, REQ_REC_OFF, op_data);
 929
 930         req->rq_replay = 1;
 931         req->rq_cb_data = mod;
 932         req->rq_commit_cb = mdc_commit_delayed;
 933         if (likely(mod != NULL))
 934                 list_add_tail(&req->rq_mod_list, &mod->mod_replay_list);
 935         else
 936                 CDEBUG(D_HA, "couldn't find open req; expecting close error\n");
 937
 938         ptlrpc_req_set_repsize(req, 2, repsize);
 939         mdc_get_rpc_lock(obd->u.cli.cl_close_lock, NULL);
 940         rc = ptlrpc_queue_wait(req);
 941         mdc_put_rpc_lock(obd->u.cli.cl_close_lock, NULL);
 942
 943         /* Close the open replay sequence if an error occured or no SOM
 944          * attribute update is needed. */
 945         if (rc != -EAGAIN)
 946                 ptlrpc_close_replay_seq(req);
 947
 948         if (rc && rc != -EAGAIN && req->rq_commit_cb)
 949                 req->rq_commit_cb(req);
 950
 951         ptlrpc_req_finished(req);
 952         RETURN(rc);
 953 }
 954
 955 #ifdef HAVE_SPLIT_SUPPORT
 956 int mdc_sendpage(struct obd_export *exp, const struct lu_fid *fid,
 957                  const struct page *page, int offset)
 958 {
 959         int rc, size[3] = { sizeof(struct ptlrpc_body), sizeof(struct mdt_body) };
 960         struct obd_import *imp = class_exp2cliimp(exp);
 961         struct ptlrpc_bulk_desc *desc = NULL;
 962         struct ptlrpc_request *req = NULL;
 963         ENTRY;
 964
 965         CDEBUG(D_INODE, "object: "DFID"\n", PFID(fid));
 966
 967         req = ptlrpc_prep_req(imp, LUSTRE_MDS_VERSION, MDS_WRITEPAGE, 3,
 968                               size, NULL);
 969         if (req == NULL)
 970                 GOTO(out, rc = -ENOMEM);
 971
 972         req->rq_request_portal = MDS_READPAGE_PORTAL;
 973
 974         desc = ptlrpc_prep_bulk_imp(req, 1, BULK_GET_SOURCE, MDS_BULK_PORTAL);
 975         if (desc == NULL)
 976                 GOTO(out, rc = -ENOMEM);
 977
 978         /* NB req now owns desc and will free it when it gets freed. */
 979         ptlrpc_prep_bulk_page(desc, (struct page *)page, 0, offset);
 980         mdc_readdir_pack(req, REQ_REC_OFF, 0, offset, fid, NULL);
 981
 982         ptlrpc_req_set_repsize(req, 2, size);
 983         rc = ptlrpc_queue_wait(req);
 984         EXIT;
 985 out:
 986         if (req != NULL)
 987                 ptlrpc_req_finished(req);
 988         return rc;
 989 }
 990 EXPORT_SYMBOL(mdc_sendpage);
 991 #endif
 992
 993 int mdc_readpage(struct obd_export *exp, const struct lu_fid *fid,
 994                  struct obd_capa *oc, __u64 offset, struct page *page,
 995                  struct ptlrpc_request **request)
 996 {
 997         int rc, size[3] = { sizeof(struct ptlrpc_body), sizeof(struct mdt_body) };
 998         struct obd_import *imp = class_exp2cliimp(exp);
 999         struct ptlrpc_bulk_desc *desc = NULL;
1000         struct ptlrpc_request *req = NULL;
1001         struct mdt_body *body;
1002         ENTRY;
1003
1004         CDEBUG(D_INODE, "object: "DFID"\n", PFID(fid));
1005
1006         size[REQ_REC_OFF + 1] = oc ? sizeof(struct lustre_capa) : 0;
1007         req = ptlrpc_prep_req(imp, LUSTRE_MDS_VERSION, MDS_READPAGE, 3, size,
1008                               NULL);
1009         if (req == NULL)
1010                 GOTO(out, rc = -ENOMEM);
1011
1012         /* XXX FIXME bug 249 */
1013         req->rq_request_portal = MDS_READPAGE_PORTAL;
1014
1015         desc = ptlrpc_prep_bulk_imp(req, 1, BULK_PUT_SINK, MDS_BULK_PORTAL);
1016         if (desc == NULL)
1017                 GOTO(out, rc = -ENOMEM);
1018
1019         /* NB req now owns desc and will free it when it gets freed */
1020         ptlrpc_prep_bulk_page(desc, page, 0, CFS_PAGE_SIZE);
1021         mdc_readdir_pack(req, REQ_REC_OFF, offset, CFS_PAGE_SIZE, fid, oc);
1022
1023         ptlrpc_req_set_repsize(req, 2, size);
1024         rc = ptlrpc_queue_wait(req);
1025
1026         if (rc == 0) {
1027                 body = lustre_swab_repbuf(req, REPLY_REC_OFF, sizeof(*body),
1028                                           lustre_swab_mdt_body);
1029                 if (body == NULL) {
1030                         CERROR("Can't unpack mdt_body\n");
1031                         GOTO(out, rc = -EPROTO);
1032                 }
1033
1034                 if (req->rq_bulk->bd_nob_transferred != CFS_PAGE_SIZE) {
1035                         CERROR ("Unexpected # bytes transferred: %d"
1036                                 " (%ld expected)\n",
1037                                 req->rq_bulk->bd_nob_transferred,
1038                                 CFS_PAGE_SIZE);
1039                         GOTO(out, rc = -EPROTO);
1040                 }
1041         }
1042
1043         EXIT;
1044  out:
1045         *request = req;
1046         return rc;
1047 }
1048
1049 static int mdc_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1050                          void *karg, void *uarg)
1051 {
1052         struct obd_device *obd = exp->exp_obd;
1053         struct obd_ioctl_data *data = karg;
1054         struct obd_import *imp = obd->u.cli.cl_import;
1055         struct llog_ctxt *ctxt;
1056         int rc;
1057         ENTRY;
1058
1059 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
1060         MOD_INC_USE_COUNT;
1061 #else
1062         if (!try_module_get(THIS_MODULE)) {
1063                 CERROR("Can't get module. Is it alive?");
1064                 return -EINVAL;
1065         }
1066 #endif
1067         switch (cmd) {
1068         case OBD_IOC_CLIENT_RECOVER:
1069                 rc = ptlrpc_recover_import(imp, data->ioc_inlbuf1);
1070                 if (rc < 0)
1071                         GOTO(out, rc);
1072                 GOTO(out, rc = 0);
1073         case IOC_OSC_SET_ACTIVE:
1074                 rc = ptlrpc_set_import_active(imp, data->ioc_offset);
1075                 GOTO(out, rc);
1076         case OBD_IOC_PARSE: {
1077                 ctxt = llog_get_context(exp->exp_obd, LLOG_CONFIG_REPL_CTXT);
1078                 rc = class_config_parse_llog(ctxt, data->ioc_inlbuf1, NULL);
1079                 GOTO(out, rc);
1080         }
1081 #ifdef __KERNEL__
1082         case OBD_IOC_LLOG_INFO:
1083         case OBD_IOC_LLOG_PRINT: {
1084                 ctxt = llog_get_context(obd, LLOG_CONFIG_REPL_CTXT);
1085                 rc = llog_ioctl(ctxt, cmd, data);
1086
1087                 GOTO(out, rc);
1088         }
1089 #endif
1090         case OBD_IOC_POLL_QUOTACHECK:
1091                 rc = lquota_poll_check(quota_interface, exp,
1092                                        (struct if_quotacheck *)karg);
1093                 GOTO(out, rc);
1094         default:
1095                 CERROR("mdc_ioctl(): unrecognised ioctl %#x\n", cmd);
1096                 GOTO(out, rc = -ENOTTY);
1097         }
1098 out:
1099 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
1100         MOD_DEC_USE_COUNT;
1101 #else
1102         module_put(THIS_MODULE);
1103 #endif
1104
1105         return rc;
1106 }
1107
1108 int mdc_set_info_async(struct obd_export *exp, obd_count keylen,
1109                        void *key, obd_count vallen, void *val,
1110                        struct ptlrpc_request_set *set)
1111 {
1112         struct obd_import *imp = class_exp2cliimp(exp);
1113         int rc = -EINVAL;
1114
1115         if (KEY_IS(KEY_INIT_RECOV)) {
1116                 if (vallen != sizeof(int))
1117                         RETURN(-EINVAL);
1118                 spin_lock(&imp->imp_lock);
1119                 imp->imp_initial_recov = *(int *)val;
1120                 spin_unlock(&imp->imp_lock);
1121                 CDEBUG(D_HA, "%s: set imp_initial_recov = %d\n",
1122                        exp->exp_obd->obd_name, imp->imp_initial_recov);
1123                 RETURN(0);
1124         }
1125         /* Turn off initial_recov after we try all backup servers once */
1126         if (KEY_IS(KEY_INIT_RECOV_BACKUP)) {
1127                 if (vallen != sizeof(int))
1128                         RETURN(-EINVAL);
1129                 spin_lock(&imp->imp_lock);
1130                 imp->imp_initial_recov_bk = *(int *)val;
1131                 if (imp->imp_initial_recov_bk)
1132                         imp->imp_initial_recov = 1;
1133                 spin_unlock(&imp->imp_lock);
1134                 CDEBUG(D_HA, "%s: set imp_initial_recov_bk = %d\n",
1135                        exp->exp_obd->obd_name, imp->imp_initial_recov_bk);
1136                 RETURN(0);
1137         }
1138         if (KEY_IS(KEY_READ_ONLY)) {
1139                 struct ptlrpc_request *req;
1140                 int size[3] = { sizeof(struct ptlrpc_body), keylen, vallen };
1141                 char *bufs[3] = { NULL, key, val };
1142
1143                 if (vallen != sizeof(int))
1144                         RETURN(-EINVAL);
1145
1146                 spin_lock(&imp->imp_lock);
1147                 if (*((int *)val)) {
1148                         imp->imp_connect_flags_orig |= OBD_CONNECT_RDONLY;
1149                         imp->imp_connect_data.ocd_connect_flags |=
1150                                 OBD_CONNECT_RDONLY;
1151                 } else {
1152                         imp->imp_connect_flags_orig &= ~OBD_CONNECT_RDONLY;
1153                         imp->imp_connect_data.ocd_connect_flags &=
1154                                 ~OBD_CONNECT_RDONLY;
1155                 }
1156                 spin_unlock(&imp->imp_lock);
1157
1158                 req = ptlrpc_prep_req(imp, LUSTRE_MDS_VERSION, MDS_SET_INFO,
1159                                       3, size, bufs);
1160                 if (req == NULL)
1161                         RETURN(-ENOMEM);
1162
1163                 ptlrpc_req_set_repsize(req, 1, NULL);
1164                 if (set) {
1165                         rc = 0;
1166                         ptlrpc_set_add_req(set, req);
1167                         ptlrpc_check_set(set);
1168                 } else {
1169                         rc = ptlrpc_queue_wait(req);
1170                         ptlrpc_req_finished(req);
1171                 }
1172
1173                 RETURN(rc);
1174         }
1175         if (KEY_IS(KEY_FLUSH_CTX)) {
1176                 sptlrpc_import_flush_my_ctx(imp);
1177                 RETURN(0);
1178         }
1179         if (KEY_IS(KEY_MDS_CONN)) {
1180                 struct obd_import *imp = class_exp2cliimp(exp);
1181
1182                 /* mds-mds import */
1183                 spin_lock(&imp->imp_lock);
1184                 imp->imp_server_timeout = 1;
1185                 spin_unlock(&imp->imp_lock);
1186                 imp->imp_client->cli_request_portal = MDS_MDS_PORTAL;
1187                 CDEBUG(D_OTHER|D_WARNING, "%s: timeout / 2\n", exp->exp_obd->obd_name);
1188                 RETURN(0);
1189         }
1190
1191         RETURN(rc);
1192 }
1193
1194 int mdc_get_info(struct obd_export *exp, __u32 keylen, void *key,
1195                  __u32 *vallen, void *val)
1196 {
1197         int rc = -EINVAL;
1198
1199         if (KEY_IS(KEY_MAX_EASIZE)) {
1200                 int mdsize, *max_easize;
1201
1202                 if (*vallen != sizeof(int))
1203                         RETURN(-EINVAL);
1204                 mdsize = *(int*)val;
1205                 if (mdsize > exp->exp_obd->u.cli.cl_max_mds_easize)
1206                         exp->exp_obd->u.cli.cl_max_mds_easize = mdsize;
1207                 max_easize = val;
1208                 *max_easize = exp->exp_obd->u.cli.cl_max_mds_easize;
1209                 RETURN(0);
1210         }
1211         if (KEY_IS(KEY_CONN_DATA)) {
1212                 struct obd_import *imp = class_exp2cliimp(exp);
1213                 struct obd_connect_data *data = val;
1214
1215                 if (*vallen != sizeof(*data))
1216                         RETURN(-EINVAL);
1217
1218                 *data = imp->imp_connect_data;
1219                 RETURN(0);
1220         }
1221
1222         RETURN(rc);
1223 }
1224
1225 static int mdc_statfs(struct obd_device *obd, struct obd_statfs *osfs,
1226                       __u64 max_age)
1227 {
1228         struct ptlrpc_request *req;
1229         struct obd_statfs *msfs;
1230         int rc, size[2] = { sizeof(struct ptlrpc_body), sizeof(*msfs) };
1231         ENTRY;
1232
1233         /* We could possibly pass max_age in the request (as an absolute
1234          * timestamp or a "seconds.usec ago") so the target can avoid doing
1235          * extra calls into the filesystem if that isn't necessary (e.g.
1236          * during mount that would help a bit).  Having relative timestamps
1237          * is not so great if request processing is slow, while absolute
1238          * timestamps are not ideal because they need time synchronization. */
1239         req = ptlrpc_prep_req(obd->u.cli.cl_import, LUSTRE_MDS_VERSION,
1240                               MDS_STATFS, 1, NULL, NULL);
1241         if (!req)
1242                 RETURN(-ENOMEM);
1243
1244         ptlrpc_req_set_repsize(req, 2, size);
1245
1246         rc = ptlrpc_queue_wait(req);
1247
1248         if (rc) {
1249                 /* check connection error first */
1250                 if (obd->u.cli.cl_import->imp_connect_error)
1251                         rc = obd->u.cli.cl_import->imp_connect_error;
1252
1253                 GOTO(out, rc);
1254         }
1255
1256         msfs = lustre_swab_repbuf(req, REPLY_REC_OFF, sizeof(*msfs),
1257                                   lustre_swab_obd_statfs);
1258         if (msfs == NULL) {
1259                 CERROR("Can't unpack obd_statfs\n");
1260                 GOTO(out, rc = -EPROTO);
1261         }
1262
1263         memcpy(osfs, msfs, sizeof(*msfs));
1264         EXIT;
1265 out:
1266         ptlrpc_req_finished(req);
1267
1268         return rc;
1269 }
1270
1271 static int mdc_pin(struct obd_export *exp, const struct lu_fid *fid,
1272                    struct obd_capa *oc,
1273                    struct obd_client_handle *handle, int flag)
1274 {
1275         struct ptlrpc_request *req;
1276         struct mdt_body *body;
1277         int rc, size[3] = { sizeof(struct ptlrpc_body), sizeof(*body) };
1278         ENTRY;
1279
1280         size[REQ_REC_OFF + 1] = oc ? sizeof(struct lustre_capa) : 0;
1281         req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
1282                               MDS_PIN, 3, size, NULL);
1283         if (req == NULL)
1284                 RETURN(-ENOMEM);
1285
1286         body = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF, sizeof (*body));
1287         body->fid1 = *fid;
1288         body->flags = flag;
1289         mdc_pack_capa(req, REQ_REC_OFF + 1, oc);
1290
1291         ptlrpc_req_set_repsize(req, 2, size);
1292
1293         mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
1294         rc = ptlrpc_queue_wait(req);
1295         mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
1296         if (rc) {
1297                 CERROR("pin failed: %d\n", rc);
1298                 ptlrpc_req_finished(req);
1299                 RETURN(rc);
1300         }
1301
1302         body = lustre_swab_repbuf(req, REPLY_REC_OFF, sizeof(*body),
1303                                   lustre_swab_mdt_body);
1304         if (body == NULL) {
1305                 ptlrpc_req_finished(req);
1306                 RETURN(rc);
1307         }
1308
1309         memcpy(&handle->och_fh, &body->handle, sizeof(body->handle));
1310         handle->och_magic = OBD_CLIENT_HANDLE_MAGIC;
1311
1312         OBD_ALLOC(handle->och_mod, sizeof(*handle->och_mod));
1313         if (handle->och_mod == NULL) {
1314                 DEBUG_REQ(D_ERROR, req, "can't allocate md_open_data");
1315                 RETURN(-ENOMEM);
1316         }
1317
1318         /* will be dropped by unpin */
1319         CFS_INIT_LIST_HEAD(&handle->och_mod->mod_replay_list);
1320         list_add_tail(&req->rq_mod_list, &handle->och_mod->mod_replay_list);
1321
1322         RETURN(rc);
1323 }
1324
1325 static int mdc_unpin(struct obd_export *exp,
1326                      struct obd_client_handle *handle, int flag)
1327 {
1328         struct ptlrpc_request *req;
1329         struct mdt_body *body;
1330         int rc, size[2] = { sizeof(struct ptlrpc_body), sizeof(*body) };
1331         ENTRY;
1332
1333         if (handle->och_magic != OBD_CLIENT_HANDLE_MAGIC)
1334                 RETURN(0);
1335
1336         req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
1337                               MDS_CLOSE, 2, size, NULL);
1338         if (req == NULL)
1339                 RETURN(-ENOMEM);
1340
1341         body = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF, sizeof(*body));
1342         memcpy(&body->handle, &handle->och_fh, sizeof(body->handle));
1343         body->flags = flag;
1344
1345         ptlrpc_req_set_repsize(req, 1, NULL);
1346         mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
1347         rc = ptlrpc_queue_wait(req);
1348         mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
1349
1350         if (rc != 0)
1351                 CERROR("unpin failed: %d\n", rc);
1352
1353         ptlrpc_req_finished(req);
1354
1355         LASSERT(!list_empty(&handle->och_mod->mod_replay_list));
1356         req = list_entry(handle->och_mod->mod_replay_list.next,
1357                          typeof(*req), rq_mod_list);
1358         list_del_init(&req->rq_mod_list);
1359         ptlrpc_req_finished(req);
1360         LASSERT(list_empty(&handle->och_mod->mod_replay_list));
1361
1362         OBD_FREE(handle->och_mod, sizeof(*handle->och_mod));
1363         RETURN(rc);
1364 }
1365
1366 int mdc_sync(struct obd_export *exp, const struct lu_fid *fid,
1367              struct obd_capa *oc, struct ptlrpc_request **request)
1368 {
1369         struct ptlrpc_request *req;
1370         int size[3] = { sizeof(struct ptlrpc_body), sizeof(struct mdt_body) };
1371         int rc;
1372         ENTRY;
1373
1374         size[REQ_REC_OFF + 1] = oc ? sizeof(struct lustre_capa) : 0;
1375         req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
1376                               MDS_SYNC, 3, size, NULL);
1377         if (!req)
1378                 RETURN(rc = -ENOMEM);
1379
1380         mdc_pack_req_body(req, REQ_REC_OFF, 0, fid, oc, 0, 0);
1381
1382         ptlrpc_req_set_repsize(req, 2, size);
1383
1384         rc = ptlrpc_queue_wait(req);
1385         if (rc || request == NULL)
1386                 ptlrpc_req_finished(req);
1387         else
1388                 *request = req;
1389
1390         RETURN(rc);
1391 }
1392
1393 static int mdc_import_event(struct obd_device *obd, struct obd_import *imp,
1394                             enum obd_import_event event)
1395 {
1396         int rc = 0;
1397
1398         LASSERT(imp->imp_obd == obd);
1399
1400         switch (event) {
1401         case IMP_EVENT_DISCON: {
1402 #if 0
1403                 /* XXX Pass event up to OBDs stack. used only for FLD now */
1404                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_DISCON, NULL);
1405 #endif
1406                 break;
1407         }
1408         case IMP_EVENT_INACTIVE: {
1409                 struct client_obd *cli = &obd->u.cli;
1410                 /*
1411                  * Flush current sequence to make client obtain new one
1412                  * from server in case of disconnect/reconnect.
1413                  * If range is already empty then no need to flush it.
1414                  */
1415                 if (cli->cl_seq != NULL &&
1416                     !range_is_exhausted(&cli->cl_seq->lcs_space)) {
1417                         seq_client_flush(cli->cl_seq);
1418                 }
1419
1420                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_INACTIVE, NULL);
1421                 break;
1422         }
1423         case IMP_EVENT_INVALIDATE: {
1424                 struct ldlm_namespace *ns = obd->obd_namespace;
1425
1426                 ldlm_namespace_cleanup(ns, LDLM_FL_LOCAL_ONLY);
1427
1428                 break;
1429         }
1430         case IMP_EVENT_ACTIVE: {
1431                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_ACTIVE, NULL);
1432                 break;
1433         }
1434         case IMP_EVENT_OCD:
1435                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_OCD, NULL);
1436                 break;
1437
1438         default:
1439                 CERROR("Unknown import event %x\n", event);
1440                 LBUG();
1441         }
1442         RETURN(rc);
1443 }
1444
1445 static int mdc_fid_init(struct obd_export *exp)
1446 {
1447         struct client_obd *cli = &exp->exp_obd->u.cli;
1448         char *prefix;
1449         int rc;
1450         ENTRY;
1451
1452         OBD_ALLOC_PTR(cli->cl_seq);
1453         if (cli->cl_seq == NULL)
1454                 RETURN(-ENOMEM);
1455
1456         OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
1457         if (prefix == NULL)
1458                 GOTO(out_free_seq, rc = -ENOMEM);
1459
1460         snprintf(prefix, MAX_OBD_NAME + 5, "srv-%s",
1461                  exp->exp_obd->obd_name);
1462
1463         /* Init client side sequence-manager */
1464         rc = seq_client_init(cli->cl_seq, exp,
1465                              LUSTRE_SEQ_METADATA,
1466                              prefix, NULL);
1467         OBD_FREE(prefix, MAX_OBD_NAME + 5);
1468         if (rc)
1469                 GOTO(out_free_seq, rc);
1470
1471         RETURN(rc);
1472 out_free_seq:
1473         OBD_FREE_PTR(cli->cl_seq);
1474         cli->cl_seq = NULL;
1475         return rc;
1476 }
1477
1478 static int mdc_fid_fini(struct obd_export *exp)
1479 {
1480         struct client_obd *cli = &exp->exp_obd->u.cli;
1481         ENTRY;
1482
1483         if (cli->cl_seq != NULL) {
1484                 seq_client_fini(cli->cl_seq);
1485                 OBD_FREE_PTR(cli->cl_seq);
1486                 cli->cl_seq = NULL;
1487         }
1488
1489         RETURN(0);
1490 }
1491
1492 int mdc_fid_alloc(struct obd_export *exp, struct lu_fid *fid,
1493                   struct md_op_data *op_data)
1494 {
1495         struct client_obd *cli = &exp->exp_obd->u.cli;
1496         struct lu_client_seq *seq = cli->cl_seq;
1497         ENTRY;
1498         RETURN(seq_client_alloc_fid(seq, fid));
1499 }
1500
1501 /* XXX This method is used only to clear current fid seq
1502  * once fld/mds insert failed */
1503 static int mdc_fid_delete(struct obd_export *exp, const struct lu_fid *fid)
1504 {
1505         struct client_obd *cli = &exp->exp_obd->u.cli;
1506
1507         seq_client_flush(cli->cl_seq);
1508         return 0;
1509 }
1510
1511 static int mdc_setup(struct obd_device *obd, struct lustre_cfg *cfg)
1512 {
1513         struct client_obd *cli = &obd->u.cli;
1514         struct lprocfs_static_vars lvars;
1515         int rc;
1516         ENTRY;
1517
1518         OBD_ALLOC(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
1519         if (!cli->cl_rpc_lock)
1520                 RETURN(-ENOMEM);
1521         mdc_init_rpc_lock(cli->cl_rpc_lock);
1522
1523         ptlrpcd_addref();
1524
1525         OBD_ALLOC(cli->cl_setattr_lock, sizeof (*cli->cl_setattr_lock));
1526         if (!cli->cl_setattr_lock)
1527                 GOTO(err_rpc_lock, rc = -ENOMEM);
1528         mdc_init_rpc_lock(cli->cl_setattr_lock);
1529
1530         OBD_ALLOC(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
1531         if (!cli->cl_close_lock)
1532                 GOTO(err_setattr_lock, rc = -ENOMEM);
1533         mdc_init_rpc_lock(cli->cl_close_lock);
1534
1535         rc = client_obd_setup(obd, cfg);
1536         if (rc)
1537                 GOTO(err_close_lock, rc);
1538         lprocfs_init_vars(mdc, &lvars);
1539         lprocfs_obd_setup(obd, lvars.obd_vars);
1540         ptlrpc_lprocfs_register_obd(obd);
1541
1542         rc = obd_llog_init(obd, NULL, obd, 0, NULL, NULL);
1543         if (rc) {
1544                 mdc_cleanup(obd);
1545                 CERROR("failed to setup llogging subsystems\n");
1546         }
1547
1548         RETURN(rc);
1549
1550 err_close_lock:
1551         OBD_FREE(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
1552 err_setattr_lock:
1553         OBD_FREE(cli->cl_setattr_lock, sizeof (*cli->cl_setattr_lock));
1554 err_rpc_lock:
1555         OBD_FREE(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
1556         ptlrpcd_decref();
1557         RETURN(rc);
1558 }
1559
1560 /* Initialize the default and maximum LOV EA and cookie sizes.  This allows
1561  * us to make MDS RPCs with large enough reply buffers to hold the
1562  * maximum-sized (= maximum striped) EA and cookie without having to
1563  * calculate this (via a call into the LOV + OSCs) each time we make an RPC. */
1564 int mdc_init_ea_size(struct obd_export *exp, int easize,
1565                      int def_easize, int cookiesize)
1566 {
1567         struct obd_device *obd = exp->exp_obd;
1568         struct client_obd *cli = &obd->u.cli;
1569         ENTRY;
1570
1571         if (cli->cl_max_mds_easize < easize)
1572                 cli->cl_max_mds_easize = easize;
1573
1574         if (cli->cl_default_mds_easize < def_easize)
1575                 cli->cl_default_mds_easize = def_easize;
1576
1577         if (cli->cl_max_mds_cookiesize < cookiesize)
1578                 cli->cl_max_mds_cookiesize = cookiesize;
1579
1580         RETURN(0);
1581 }
1582
1583 static int mdc_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
1584 {
1585         int rc = 0;
1586         ENTRY;
1587
1588         switch (stage) {
1589         case OBD_CLEANUP_EARLY:
1590         case OBD_CLEANUP_EXPORTS:
1591                 /* If we set up but never connected, the
1592                    client import will not have been cleaned. */
1593                 if (obd->u.cli.cl_import) {
1594                         struct obd_import *imp;
1595                         imp = obd->u.cli.cl_import;
1596                         CERROR("client import never connected\n");
1597                         ptlrpc_invalidate_import(imp);
1598                         ptlrpc_free_rq_pool(imp->imp_rq_pool);
1599                         class_destroy_import(imp);
1600                         obd->u.cli.cl_import = NULL;
1601                 }
1602                 break;
1603         case OBD_CLEANUP_SELF_EXP:
1604                 rc = obd_llog_finish(obd, 0);
1605                 if (rc != 0)
1606                         CERROR("failed to cleanup llogging subsystems\n");
1607         case OBD_CLEANUP_OBD:
1608                 break;
1609         }
1610         RETURN(rc);
1611 }
1612
1613 static int mdc_cleanup(struct obd_device *obd)
1614 {
1615         struct client_obd *cli = &obd->u.cli;
1616
1617         OBD_FREE(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
1618         OBD_FREE(cli->cl_setattr_lock, sizeof (*cli->cl_setattr_lock));
1619         OBD_FREE(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
1620
1621         ptlrpc_lprocfs_unregister_obd(obd);
1622         lprocfs_obd_cleanup(obd);
1623         ptlrpcd_decref();
1624
1625         return client_obd_cleanup(obd);
1626 }
1627
1628
1629 static int mdc_llog_init(struct obd_device *obd, struct obd_llogs *llogs,
1630                          struct obd_device *tgt,
1631                          int count, struct llog_catid *logid,
1632                          struct obd_uuid *uuid)
1633 {
1634         struct llog_ctxt *ctxt;
1635         int rc;
1636         ENTRY;
1637
1638         rc = llog_setup(obd, llogs, LLOG_CONFIG_REPL_CTXT, tgt, 0, NULL,
1639                         &llog_client_ops);
1640         if (rc == 0) {
1641                 ctxt = llog_get_context(obd, LLOG_CONFIG_REPL_CTXT);
1642                 ctxt->loc_imp = obd->u.cli.cl_import;
1643         }
1644
1645         rc = llog_setup(obd, llogs, LLOG_LOVEA_REPL_CTXT, tgt, 0, NULL,
1646                        &llog_client_ops);
1647         if (rc == 0) {
1648                 ctxt = llog_get_context(obd, LLOG_LOVEA_REPL_CTXT);
1649                 ctxt->loc_imp = obd->u.cli.cl_import;
1650         }
1651
1652         RETURN(rc);
1653 }
1654
1655 static int mdc_llog_finish(struct obd_device *obd, int count)
1656 {
1657         int rc;
1658         ENTRY;
1659
1660         rc = llog_cleanup(llog_get_context(obd, LLOG_LOVEA_REPL_CTXT));
1661         if (rc) {
1662                 CERROR("can not cleanup LLOG_CONFIG_REPL_CTXT rc %d\n", rc);
1663         }
1664         rc = llog_cleanup(llog_get_context(obd, LLOG_CONFIG_REPL_CTXT));
1665         RETURN(rc);
1666 }
1667
1668 static int mdc_process_config(struct obd_device *obd, obd_count len, void *buf)
1669 {
1670         struct lustre_cfg *lcfg = buf;
1671         struct lprocfs_static_vars lvars;
1672         int rc = 0;
1673
1674         lprocfs_init_vars(mdc, &lvars);
1675
1676         rc = class_process_proc_param(PARAM_MDC, lvars.obd_vars, lcfg, obd);
1677         return(rc);
1678 }
1679
1680 /* get remote permission for current user on fid */
1681 int mdc_get_remote_perm(struct obd_export *exp, const struct lu_fid *fid,
1682                         struct obd_capa *oc, struct ptlrpc_request **request)
1683 {
1684         struct ptlrpc_request *req;
1685         struct mdt_body *body;
1686         struct mdt_remote_perm *perm;
1687         int size[5] = { sizeof(struct ptlrpc_body), sizeof(*body) };
1688         int rc;
1689         ENTRY;
1690
1691         size[REQ_REC_OFF + 1] = oc ? sizeof(struct lustre_capa) : 0;
1692
1693         *request = NULL;
1694         req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
1695                               MDS_GETATTR, 3, size, NULL);
1696         if (!req)
1697                 RETURN(-ENOMEM);
1698
1699         mdc_pack_req_body(req, REQ_REC_OFF, OBD_MD_FLRMTPERM, fid, oc, 0, 0);
1700
1701         size[REPLY_REC_OFF + 1] = sizeof(*perm);
1702         ptlrpc_req_set_repsize(req, 5, size);
1703         rc = ptlrpc_queue_wait(req);
1704         if (rc) {
1705                 ptlrpc_req_finished(req);
1706                 RETURN(rc);
1707         }
1708
1709         body = lustre_swab_repbuf(req, REPLY_REC_OFF, sizeof(*body),
1710                                   lustre_swab_mdt_body);
1711         LASSERT(body);
1712         LASSERT(body->valid & OBD_MD_FLRMTPERM);
1713
1714         perm = lustre_swab_repbuf(req, REPLY_REC_OFF + 1, sizeof(*perm),
1715                                   lustre_swab_mdt_remote_perm);
1716         LASSERT(perm);
1717
1718         *request = req;
1719         RETURN(0);
1720 }
1721
1722 static int mdc_interpret_renew_capa(struct ptlrpc_request *req, void *unused,
1723                                     int status)
1724 {
1725         struct obd_capa *oc = req->rq_async_args.pointer_arg[0];
1726         renew_capa_cb_t cb = req->rq_async_args.pointer_arg[1];
1727         struct mdt_body *body = NULL;
1728         struct lustre_capa *capa;
1729         ENTRY;
1730
1731         if (status)
1732                 GOTO(out, capa = ERR_PTR(status));
1733
1734         body = lustre_swab_repbuf(req, REPLY_REC_OFF, sizeof(*body),
1735                                   lustre_swab_mdt_body);
1736         if (body == NULL)
1737                 GOTO(out, capa = ERR_PTR(-EFAULT));
1738
1739         if ((body->valid & OBD_MD_FLOSSCAPA) == 0)
1740                 GOTO(out, capa = ERR_PTR(-ENOENT));
1741
1742         capa = lustre_unpack_capa(req->rq_repmsg, REPLY_REC_OFF + 1);
1743         if (!capa)
1744                 GOTO(out, capa = ERR_PTR(-EFAULT));
1745         EXIT;
1746 out:
1747         cb(oc, capa);
1748         return 0;
1749 }
1750
1751 static int mdc_renew_capa(struct obd_export *exp, struct obd_capa *oc,
1752                           renew_capa_cb_t cb)
1753 {
1754         struct ptlrpc_request *req;
1755         int size[5] = { sizeof(struct ptlrpc_body),
1756                         sizeof(struct mdt_body),
1757                         sizeof(struct lustre_capa) };
1758         ENTRY;
1759
1760         req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
1761                               MDS_GETATTR, 3, size, NULL);
1762         if (!req)
1763                 RETURN(-ENOMEM);
1764
1765         mdc_pack_req_body(req, REQ_REC_OFF, OBD_MD_FLOSSCAPA,
1766                           &oc->c_capa.lc_fid, oc, 0, 0);
1767
1768         ptlrpc_req_set_repsize(req, 5, size);
1769         req->rq_async_args.pointer_arg[0] = oc;
1770         req->rq_async_args.pointer_arg[1] = cb;
1771         req->rq_interpret_reply = mdc_interpret_renew_capa;
1772         ptlrpcd_add_req(req);
1773
1774         RETURN(0);
1775 }
1776
1777 static int mdc_connect(const struct lu_env *env,
1778                        struct lustre_handle *dlm_handle,
1779                        struct obd_device *obd, struct obd_uuid *cluuid,
1780                        struct obd_connect_data *data) {
1781         struct obd_import *imp = obd->u.cli.cl_import;
1782
1783         /* mds-mds import features */
1784         if (data && (data->ocd_connect_flags & OBD_CONNECT_MDS_MDS)) {
1785                 spin_lock(&imp->imp_lock);
1786                 imp->imp_server_timeout = 1;
1787                 spin_unlock(&imp->imp_lock);
1788                 imp->imp_client->cli_request_portal = MDS_MDS_PORTAL;
1789                 CDEBUG(D_OTHER, "%s: Set 'mds' portal and timeout\n",
1790                        obd->obd_name);
1791         }
1792
1793         return client_connect_import(env, dlm_handle, obd, cluuid, data);
1794 }
1795
1796 struct obd_ops mdc_obd_ops = {
1797         .o_owner            = THIS_MODULE,
1798         .o_setup            = mdc_setup,
1799         .o_precleanup       = mdc_precleanup,
1800         .o_cleanup          = mdc_cleanup,
1801         .o_add_conn         = client_import_add_conn,
1802         .o_del_conn         = client_import_del_conn,
1803         .o_connect          = mdc_connect,
1804         .o_disconnect       = client_disconnect_export,
1805         .o_iocontrol        = mdc_iocontrol,
1806         .o_set_info_async   = mdc_set_info_async,
1807         .o_statfs           = mdc_statfs,
1808         .o_pin              = mdc_pin,
1809         .o_unpin            = mdc_unpin,
1810         .o_fid_init         = mdc_fid_init,
1811         .o_fid_fini         = mdc_fid_fini,
1812         .o_fid_alloc        = mdc_fid_alloc,
1813         .o_fid_delete       = mdc_fid_delete,
1814         .o_import_event     = mdc_import_event,
1815         .o_llog_init        = mdc_llog_init,
1816         .o_llog_finish      = mdc_llog_finish,
1817         .o_get_info         = mdc_get_info,
1818         .o_process_config  = mdc_process_config,
1819 };
1820
1821 struct md_ops mdc_md_ops = {
1822         .m_getstatus        = mdc_getstatus,
1823         .m_change_cbdata    = mdc_change_cbdata,
1824         .m_close            = mdc_close,
1825         .m_create           = mdc_create,
1826         .m_done_writing     = mdc_done_writing,
1827         .m_enqueue          = mdc_enqueue,
1828         .m_getattr          = mdc_getattr,
1829         .m_getattr_name     = mdc_getattr_name,
1830         .m_intent_lock      = mdc_intent_lock,
1831         .m_link             = mdc_link,
1832         .m_is_subdir        = mdc_is_subdir,
1833         .m_rename           = mdc_rename,
1834         .m_setattr          = mdc_setattr,
1835         .m_setxattr         = mdc_setxattr,
1836         .m_getxattr         = mdc_getxattr,
1837         .m_sync             = mdc_sync,
1838         .m_readpage         = mdc_readpage,
1839         .m_unlink           = mdc_unlink,
1840         .m_cancel_unused    = mdc_cancel_unused,
1841         .m_init_ea_size     = mdc_init_ea_size,
1842         .m_set_lock_data    = mdc_set_lock_data,
1843         .m_lock_match       = mdc_lock_match,
1844         .m_get_lustre_md    = mdc_get_lustre_md,
1845         .m_free_lustre_md   = mdc_free_lustre_md,
1846         .m_set_open_replay_data = mdc_set_open_replay_data,
1847         .m_clear_open_replay_data = mdc_clear_open_replay_data,
1848         .m_get_remote_perm  = mdc_get_remote_perm,
1849         .m_renew_capa       = mdc_renew_capa
1850 };
1851
1852 extern quota_interface_t mdc_quota_interface;
1853
1854 int __init mdc_init(void)
1855 {
1856         int rc;
1857         struct lprocfs_static_vars lvars;
1858         lprocfs_init_vars(mdc, &lvars);
1859
1860         request_module("lquota");
1861         quota_interface = PORTAL_SYMBOL_GET(mdc_quota_interface);
1862         init_obd_quota_ops(quota_interface, &mdc_obd_ops);
1863
1864         rc = class_register_type(&mdc_obd_ops, &mdc_md_ops, lvars.module_vars,
1865                                  LUSTRE_MDC_NAME, NULL);
1866         if (rc && quota_interface)
1867                 PORTAL_SYMBOL_PUT(mdc_quota_interface);
1868
1869         RETURN(rc);
1870 }
1871
1872 #ifdef __KERNEL__
1873 static void /*__exit*/ mdc_exit(void)
1874 {
1875         if (quota_interface)
1876                 PORTAL_SYMBOL_PUT(mdc_quota_interface);
1877
1878         class_unregister_type(LUSTRE_MDC_NAME);
1879 }
1880
1881 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
1882 MODULE_DESCRIPTION("Lustre Metadata Client");
1883 MODULE_LICENSE("GPL");
1884
1885 module_init(mdc_init);
1886 module_exit(mdc_exit);
1887 #endif