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                         /* We no longer want to preserve this for transno-
 731                          * unconditional replay. */
 732                         spin_lock(&cur->rq_lock);
 733                         cur->rq_replay = 0;
 734                         spin_unlock(&cur->rq_lock);
 735                 }
 736         }
 737
 738         if (list_empty(&mod->mod_replay_list)) {
 739                 if (mod->mod_och != NULL)
 740                         mod->mod_och->och_mod = NULL;
 741
 742                 OBD_FREE_PTR(mod);
 743         }
 744 }
 745
 746 int mdc_set_open_replay_data(struct obd_export *exp,
 747                              struct obd_client_handle *och,
 748                              struct ptlrpc_request *open_req)
 749 {
 750         struct md_open_data *mod;
 751         struct mdt_rec_create *rec = lustre_msg_buf(open_req->rq_reqmsg,
 752                                                     DLM_INTENT_REC_OFF,
 753                                                     sizeof(*rec));
 754         struct mdt_body *body = lustre_msg_buf(open_req->rq_repmsg,
 755                                                DLM_REPLY_REC_OFF,
 756                                                sizeof(*body));
 757         struct obd_import *imp = open_req->rq_import;
 758         ENTRY;
 759
 760         LASSERT(rec != NULL);
 761
 762         /* Incoming message in my byte order (it's been swabbed). */
 763         LASSERT_REPSWABBED(open_req, DLM_REPLY_REC_OFF);
 764
 765         /* Outgoing messages always in my byte order. */
 766         LASSERT(body != NULL);
 767
 768         /*Only the import is replayable, we set replay_open data */
 769         if (och && imp->imp_replayable) {
 770                 OBD_ALLOC(mod, sizeof(*mod));
 771                 if (mod == NULL) {
 772                         DEBUG_REQ(D_ERROR, open_req,
 773                                   "Can't allocate md_open_data");
 774                         RETURN(0);
 775                 }
 776                 CFS_INIT_LIST_HEAD(&mod->mod_replay_list);
 777
 778                 spin_lock(&open_req->rq_lock);
 779                 if (!open_req->rq_replay) {
 780                         OBD_FREE(mod, sizeof(*mod));
 781                         spin_unlock(&open_req->rq_lock);
 782                         RETURN(0);
 783                 }
 784
 785                 och->och_mod = mod;
 786                 mod->mod_och = och;
 787                 open_req->rq_cb_data = mod;
 788                 list_add_tail(&open_req->rq_mod_list, &mod->mod_replay_list);
 789                 open_req->rq_commit_cb = mdc_commit_delayed;
 790                 spin_unlock(&open_req->rq_lock);
 791         }
 792
 793         rec->cr_fid2 = body->fid1;
 794         rec->cr_ioepoch = body->ioepoch;
 795         rec->cr_old_handle.cookie = body->handle.cookie;
 796         open_req->rq_replay_cb = mdc_replay_open;
 797         if (!fid_is_sane(&body->fid1)) {
 798                 DEBUG_REQ(D_ERROR, open_req, "Saving replay request with "
 799                           "insane fid");
 800                 LBUG();
 801         }
 802
 803         DEBUG_REQ(D_HA, open_req, "Set up open replay data");
 804         RETURN(0);
 805 }
 806
 807 int mdc_clear_open_replay_data(struct obd_export *exp,
 808                                struct obd_client_handle *och)
 809 {
 810         struct md_open_data *mod = och->och_mod;
 811         ENTRY;
 812
 813         /*
 814          * Don't free the structure now (it happens in mdc_commit_delayed(),
 815          * after the last request is removed from its replay list),
 816          * but make sure that replay doesn't poke at the och, which is about to
 817          * be freed.
 818          */
 819         LASSERT(mod != LP_POISON);
 820         if (mod != NULL)
 821                 mod->mod_och = NULL;
 822
 823         och->och_mod = NULL;
 824         RETURN(0);
 825 }
 826
 827 int mdc_close(struct obd_export *exp, struct md_op_data *op_data,
 828               struct md_open_data *mod, struct ptlrpc_request **request)
 829 {
 830         struct obd_device *obd = class_exp2obd(exp);
 831         int reqsize[4] = { sizeof(struct ptlrpc_body),
 832                            sizeof(struct mdt_epoch),
 833                            sizeof(struct mdt_rec_setattr)};
 834         int repsize[4] = { sizeof(struct ptlrpc_body),
 835                            sizeof(struct mdt_body),
 836                            obd->u.cli.cl_max_mds_easize,
 837                            obd->u.cli.cl_max_mds_cookiesize };
 838         struct ptlrpc_request *req;
 839         int rc;
 840         ENTRY;
 841
 842         reqsize[REQ_REC_OFF + 2] = op_data->op_capa1 ?
 843                                         sizeof(struct lustre_capa) : 0;
 844         req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
 845                               MDS_CLOSE, 4, reqsize, NULL);
 846         if (req == NULL)
 847                 GOTO(out, rc = -ENOMEM);
 848
 849         /* To avoid a livelock (bug 7034), we need to send CLOSE RPCs to a
 850          * portal whose threads are not taking any DLM locks and are therefore
 851          * always progressing */
 852         /* XXX FIXME bug 249 */
 853         req->rq_request_portal = MDS_READPAGE_PORTAL;
 854
 855         /* Ensure that this close's handle is fixed up during replay. */
 856         if (likely(mod != NULL))
 857                 list_add_tail(&req->rq_mod_list, &mod->mod_replay_list);
 858         else
 859                 CDEBUG(D_HA, "couldn't find open req; expecting close error\n");
 860
 861         mdc_close_pack(req, REQ_REC_OFF, op_data);
 862         ptlrpc_req_set_repsize(req, 4, repsize);
 863         req->rq_commit_cb = mdc_commit_delayed;
 864         req->rq_replay = 1;
 865         LASSERT(req->rq_cb_data == NULL);
 866         req->rq_cb_data = mod;
 867
 868         mdc_get_rpc_lock(obd->u.cli.cl_close_lock, NULL);
 869         rc = ptlrpc_queue_wait(req);
 870         mdc_put_rpc_lock(obd->u.cli.cl_close_lock, NULL);
 871
 872         if (req->rq_repmsg == NULL) {
 873                 CDEBUG(D_HA, "request failed to send: %p, %d\n", req,
 874                        req->rq_status);
 875                 if (rc == 0)
 876                         rc = req->rq_status ? req->rq_status : -EIO;
 877         } else if (rc == 0 || rc == -EAGAIN) {
 878                 rc = lustre_msg_get_status(req->rq_repmsg);
 879                 if (lustre_msg_get_type(req->rq_repmsg) == PTL_RPC_MSG_ERR) {
 880                         DEBUG_REQ(D_ERROR, req, "type == PTL_RPC_MSG_ERR, err "
 881                                   "= %d", rc);
 882                         if (rc > 0)
 883                                 rc = -rc;
 884                 } else if (mod == NULL) {
 885                         if (req->rq_import->imp_replayable)
 886                                 CERROR("Unexpected: can't find md_open_data,"
 887                                        "but close succeeded with replayable imp"
 888                                        "Please tell CFS.\n");
 889                 }
 890                 if (!lustre_swab_repbuf(req, REPLY_REC_OFF,
 891                                         sizeof(struct mdt_body),
 892                                         lustre_swab_mdt_body)) {
 893                         CERROR("Error unpacking mdt_body\n");
 894                         rc = -EPROTO;
 895                 }
 896         }
 897
 898         EXIT;
 899         *request = req;
 900  out:
 901         if (rc != 0 && rc != -EAGAIN && req && req->rq_commit_cb)
 902                 req->rq_commit_cb(req);
 903
 904         return rc;
 905 }
 906
 907 int mdc_done_writing(struct obd_export *exp, struct md_op_data *op_data,
 908                      struct md_open_data *mod)
 909 {
 910         struct obd_device *obd = class_exp2obd(exp);
 911         struct ptlrpc_request *req;
 912         int size[4] = { sizeof(struct ptlrpc_body),
 913                         sizeof(struct mdt_epoch),
 914                         sizeof(struct mdt_rec_setattr)};
 915         int repsize[2] = { sizeof(struct ptlrpc_body),
 916                            sizeof(struct mdt_body)};
 917         int rc;
 918         ENTRY;
 919
 920         if (op_data->op_capa1)
 921                 size[REQ_REC_OFF + 2] = sizeof(struct lustre_capa);
 922         req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
 923                               MDS_DONE_WRITING, 4, size, NULL);
 924         if (req == NULL)
 925                 RETURN(-ENOMEM);
 926
 927         mdc_close_pack(req, REQ_REC_OFF, op_data);
 928
 929         req->rq_replay = 1;
 930         req->rq_cb_data = mod;
 931         req->rq_commit_cb = mdc_commit_delayed;
 932         if (likely(mod != NULL))
 933                 list_add_tail(&req->rq_mod_list, &mod->mod_replay_list);
 934         else
 935                 CDEBUG(D_HA, "couldn't find open req; expecting close error\n");
 936
 937         ptlrpc_req_set_repsize(req, 2, repsize);
 938         mdc_get_rpc_lock(obd->u.cli.cl_close_lock, NULL);
 939         rc = ptlrpc_queue_wait(req);
 940         mdc_put_rpc_lock(obd->u.cli.cl_close_lock, NULL);
 941
 942         /* Close the open replay sequence if an error occured or no SOM
 943          * attribute update is needed. */
 944         if (rc != -EAGAIN)
 945                 ptlrpc_close_replay_seq(req);
 946
 947         if (rc && rc != -EAGAIN && req->rq_commit_cb)
 948                 req->rq_commit_cb(req);
 949
 950         ptlrpc_req_finished(req);
 951         RETURN(rc);
 952 }
 953
 954 #ifdef HAVE_SPLIT_SUPPORT
 955 int mdc_sendpage(struct obd_export *exp, const struct lu_fid *fid,
 956                  const struct page *page, int offset)
 957 {
 958         int rc, size[3] = { sizeof(struct ptlrpc_body), sizeof(struct mdt_body) };
 959         struct obd_import *imp = class_exp2cliimp(exp);
 960         struct ptlrpc_bulk_desc *desc = NULL;
 961         struct ptlrpc_request *req = NULL;
 962         ENTRY;
 963
 964         CDEBUG(D_INODE, "object: "DFID"\n", PFID(fid));
 965
 966         req = ptlrpc_prep_req(imp, LUSTRE_MDS_VERSION, MDS_WRITEPAGE, 3,
 967                               size, NULL);
 968         if (req == NULL)
 969                 GOTO(out, rc = -ENOMEM);
 970
 971         req->rq_request_portal = MDS_READPAGE_PORTAL;
 972
 973         desc = ptlrpc_prep_bulk_imp(req, 1, BULK_GET_SOURCE, MDS_BULK_PORTAL);
 974         if (desc == NULL)
 975                 GOTO(out, rc = -ENOMEM);
 976
 977         /* NB req now owns desc and will free it when it gets freed. */
 978         ptlrpc_prep_bulk_page(desc, (struct page *)page, 0, offset);
 979         mdc_readdir_pack(req, REQ_REC_OFF, 0, offset, fid, NULL);
 980
 981         ptlrpc_req_set_repsize(req, 2, size);
 982         rc = ptlrpc_queue_wait(req);
 983         EXIT;
 984 out:
 985         if (req != NULL)
 986                 ptlrpc_req_finished(req);
 987         return rc;
 988 }
 989 EXPORT_SYMBOL(mdc_sendpage);
 990 #endif
 991
 992 int mdc_readpage(struct obd_export *exp, const struct lu_fid *fid,
 993                  struct obd_capa *oc, __u64 offset, struct page *page,
 994                  struct ptlrpc_request **request)
 995 {
 996         int rc, size[3] = { sizeof(struct ptlrpc_body), sizeof(struct mdt_body) };
 997         struct obd_import *imp = class_exp2cliimp(exp);
 998         struct ptlrpc_bulk_desc *desc = NULL;
 999         struct ptlrpc_request *req = NULL;
1000         struct mdt_body *body;
1001         ENTRY;
1002
1003         CDEBUG(D_INODE, "object: "DFID"\n", PFID(fid));
1004
1005         size[REQ_REC_OFF + 1] = oc ? sizeof(struct lustre_capa) : 0;
1006         req = ptlrpc_prep_req(imp, LUSTRE_MDS_VERSION, MDS_READPAGE, 3, size,
1007                               NULL);
1008         if (req == NULL)
1009                 GOTO(out, rc = -ENOMEM);
1010
1011         /* XXX FIXME bug 249 */
1012         req->rq_request_portal = MDS_READPAGE_PORTAL;
1013
1014         desc = ptlrpc_prep_bulk_imp(req, 1, BULK_PUT_SINK, MDS_BULK_PORTAL);
1015         if (desc == NULL)
1016                 GOTO(out, rc = -ENOMEM);
1017
1018         /* NB req now owns desc and will free it when it gets freed */
1019         ptlrpc_prep_bulk_page(desc, page, 0, CFS_PAGE_SIZE);
1020         mdc_readdir_pack(req, REQ_REC_OFF, offset, CFS_PAGE_SIZE, fid, oc);
1021
1022         ptlrpc_req_set_repsize(req, 2, size);
1023         rc = ptlrpc_queue_wait(req);
1024
1025         if (rc == 0) {
1026                 body = lustre_swab_repbuf(req, REPLY_REC_OFF, sizeof(*body),
1027                                           lustre_swab_mdt_body);
1028                 if (body == NULL) {
1029                         CERROR("Can't unpack mdt_body\n");
1030                         GOTO(out, rc = -EPROTO);
1031                 }
1032
1033                 if (req->rq_bulk->bd_nob_transferred != CFS_PAGE_SIZE) {
1034                         CERROR ("Unexpected # bytes transferred: %d"
1035                                 " (%ld expected)\n",
1036                                 req->rq_bulk->bd_nob_transferred,
1037                                 CFS_PAGE_SIZE);
1038                         GOTO(out, rc = -EPROTO);
1039                 }
1040         }
1041
1042         EXIT;
1043  out:
1044         *request = req;
1045         return rc;
1046 }
1047
1048 static int mdc_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1049                          void *karg, void *uarg)
1050 {
1051         struct obd_device *obd = exp->exp_obd;
1052         struct obd_ioctl_data *data = karg;
1053         struct obd_import *imp = obd->u.cli.cl_import;
1054         struct llog_ctxt *ctxt;
1055         int rc;
1056         ENTRY;
1057
1058 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
1059         MOD_INC_USE_COUNT;
1060 #else
1061         if (!try_module_get(THIS_MODULE)) {
1062                 CERROR("Can't get module. Is it alive?");
1063                 return -EINVAL;
1064         }
1065 #endif
1066         switch (cmd) {
1067         case OBD_IOC_CLIENT_RECOVER:
1068                 rc = ptlrpc_recover_import(imp, data->ioc_inlbuf1);
1069                 if (rc < 0)
1070                         GOTO(out, rc);
1071                 GOTO(out, rc = 0);
1072         case IOC_OSC_SET_ACTIVE:
1073                 rc = ptlrpc_set_import_active(imp, data->ioc_offset);
1074                 GOTO(out, rc);
1075         case OBD_IOC_PARSE: {
1076                 ctxt = llog_get_context(exp->exp_obd, LLOG_CONFIG_REPL_CTXT);
1077                 rc = class_config_parse_llog(ctxt, data->ioc_inlbuf1, NULL);
1078                 GOTO(out, rc);
1079         }
1080 #ifdef __KERNEL__
1081         case OBD_IOC_LLOG_INFO:
1082         case OBD_IOC_LLOG_PRINT: {
1083                 ctxt = llog_get_context(obd, LLOG_CONFIG_REPL_CTXT);
1084                 rc = llog_ioctl(ctxt, cmd, data);
1085
1086                 GOTO(out, rc);
1087         }
1088 #endif
1089         case OBD_IOC_POLL_QUOTACHECK:
1090                 rc = lquota_poll_check(quota_interface, exp,
1091                                        (struct if_quotacheck *)karg);
1092                 GOTO(out, rc);
1093         default:
1094                 CERROR("mdc_ioctl(): unrecognised ioctl %#x\n", cmd);
1095                 GOTO(out, rc = -ENOTTY);
1096         }
1097 out:
1098 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
1099         MOD_DEC_USE_COUNT;
1100 #else
1101         module_put(THIS_MODULE);
1102 #endif
1103
1104         return rc;
1105 }
1106
1107 int mdc_set_info_async(struct obd_export *exp, obd_count keylen,
1108                        void *key, obd_count vallen, void *val,
1109                        struct ptlrpc_request_set *set)
1110 {
1111         struct obd_import *imp = class_exp2cliimp(exp);
1112         int rc = -EINVAL;
1113
1114         if (KEY_IS(KEY_INIT_RECOV)) {
1115                 if (vallen != sizeof(int))
1116                         RETURN(-EINVAL);
1117                 spin_lock(&imp->imp_lock);
1118                 imp->imp_initial_recov = *(int *)val;
1119                 spin_unlock(&imp->imp_lock);
1120                 CDEBUG(D_HA, "%s: set imp_initial_recov = %d\n",
1121                        exp->exp_obd->obd_name, imp->imp_initial_recov);
1122                 RETURN(0);
1123         }
1124         /* Turn off initial_recov after we try all backup servers once */
1125         if (KEY_IS(KEY_INIT_RECOV_BACKUP)) {
1126                 if (vallen != sizeof(int))
1127                         RETURN(-EINVAL);
1128                 spin_lock(&imp->imp_lock);
1129                 imp->imp_initial_recov_bk = *(int *)val;
1130                 if (imp->imp_initial_recov_bk)
1131                         imp->imp_initial_recov = 1;
1132                 spin_unlock(&imp->imp_lock);
1133                 CDEBUG(D_HA, "%s: set imp_initial_recov_bk = %d\n",
1134                        exp->exp_obd->obd_name, imp->imp_initial_recov_bk);
1135                 RETURN(0);
1136         }
1137         if (KEY_IS(KEY_READ_ONLY)) {
1138                 struct ptlrpc_request *req;
1139                 int size[3] = { sizeof(struct ptlrpc_body), keylen, vallen };
1140                 char *bufs[3] = { NULL, key, val };
1141
1142                 if (vallen != sizeof(int))
1143                         RETURN(-EINVAL);
1144
1145                 spin_lock(&imp->imp_lock);
1146                 if (*((int *)val)) {
1147                         imp->imp_connect_flags_orig |= OBD_CONNECT_RDONLY;
1148                         imp->imp_connect_data.ocd_connect_flags |=
1149                                 OBD_CONNECT_RDONLY;
1150                 } else {
1151                         imp->imp_connect_flags_orig &= ~OBD_CONNECT_RDONLY;
1152                         imp->imp_connect_data.ocd_connect_flags &=
1153                                 ~OBD_CONNECT_RDONLY;
1154                 }
1155                 spin_unlock(&imp->imp_lock);
1156
1157                 req = ptlrpc_prep_req(imp, LUSTRE_MDS_VERSION, MDS_SET_INFO,
1158                                       3, size, bufs);
1159                 if (req == NULL)
1160                         RETURN(-ENOMEM);
1161
1162                 ptlrpc_req_set_repsize(req, 1, NULL);
1163                 if (set) {
1164                         rc = 0;
1165                         ptlrpc_set_add_req(set, req);
1166                         ptlrpc_check_set(set);
1167                 } else {
1168                         rc = ptlrpc_queue_wait(req);
1169                         ptlrpc_req_finished(req);
1170                 }
1171
1172                 RETURN(rc);
1173         }
1174         if (KEY_IS(KEY_FLUSH_CTX)) {
1175                 sptlrpc_import_flush_my_ctx(imp);
1176                 RETURN(0);
1177         }
1178         if (KEY_IS(KEY_MDS_CONN)) {
1179                 struct obd_import *imp = class_exp2cliimp(exp);
1180
1181                 /* mds-mds import */
1182                 spin_lock(&imp->imp_lock);
1183                 imp->imp_server_timeout = 1;
1184                 spin_unlock(&imp->imp_lock);
1185                 imp->imp_client->cli_request_portal = MDS_MDS_PORTAL;
1186                 CDEBUG(D_OTHER|D_WARNING, "%s: timeout / 2\n", exp->exp_obd->obd_name);
1187                 RETURN(0);
1188         }
1189
1190         RETURN(rc);
1191 }
1192
1193 int mdc_get_info(struct obd_export *exp, __u32 keylen, void *key,
1194                  __u32 *vallen, void *val)
1195 {
1196         int rc = -EINVAL;
1197
1198         if (KEY_IS(KEY_MAX_EASIZE)) {
1199                 int mdsize, *max_easize;
1200
1201                 if (*vallen != sizeof(int))
1202                         RETURN(-EINVAL);
1203                 mdsize = *(int*)val;
1204                 if (mdsize > exp->exp_obd->u.cli.cl_max_mds_easize)
1205                         exp->exp_obd->u.cli.cl_max_mds_easize = mdsize;
1206                 max_easize = val;
1207                 *max_easize = exp->exp_obd->u.cli.cl_max_mds_easize;
1208                 RETURN(0);
1209         }
1210         if (KEY_IS(KEY_CONN_DATA)) {
1211                 struct obd_import *imp = class_exp2cliimp(exp);
1212                 struct obd_connect_data *data = val;
1213
1214                 if (*vallen != sizeof(*data))
1215                         RETURN(-EINVAL);
1216
1217                 *data = imp->imp_connect_data;
1218                 RETURN(0);
1219         }
1220
1221         RETURN(rc);
1222 }
1223
1224 static int mdc_statfs(struct obd_device *obd, struct obd_statfs *osfs,
1225                       __u64 max_age)
1226 {
1227         struct ptlrpc_request *req;
1228         struct obd_statfs *msfs;
1229         int rc, size[2] = { sizeof(struct ptlrpc_body), sizeof(*msfs) };
1230         ENTRY;
1231
1232         /* We could possibly pass max_age in the request (as an absolute
1233          * timestamp or a "seconds.usec ago") so the target can avoid doing
1234          * extra calls into the filesystem if that isn't necessary (e.g.
1235          * during mount that would help a bit).  Having relative timestamps
1236          * is not so great if request processing is slow, while absolute
1237          * timestamps are not ideal because they need time synchronization. */
1238         req = ptlrpc_prep_req(obd->u.cli.cl_import, LUSTRE_MDS_VERSION,
1239                               MDS_STATFS, 1, NULL, NULL);
1240         if (!req)
1241                 RETURN(-ENOMEM);
1242
1243         ptlrpc_req_set_repsize(req, 2, size);
1244
1245         rc = ptlrpc_queue_wait(req);
1246
1247         if (rc) {
1248                 /* check connection error first */
1249                 if (obd->u.cli.cl_import->imp_connect_error)
1250                         rc = obd->u.cli.cl_import->imp_connect_error;
1251
1252                 GOTO(out, rc);
1253         }
1254
1255         msfs = lustre_swab_repbuf(req, REPLY_REC_OFF, sizeof(*msfs),
1256                                   lustre_swab_obd_statfs);
1257         if (msfs == NULL) {
1258                 CERROR("Can't unpack obd_statfs\n");
1259                 GOTO(out, rc = -EPROTO);
1260         }
1261
1262         memcpy(osfs, msfs, sizeof(*msfs));
1263         EXIT;
1264 out:
1265         ptlrpc_req_finished(req);
1266
1267         return rc;
1268 }
1269
1270 static int mdc_pin(struct obd_export *exp, const struct lu_fid *fid,
1271                    struct obd_capa *oc,
1272                    struct obd_client_handle *handle, int flag)
1273 {
1274         struct ptlrpc_request *req;
1275         struct mdt_body *body;
1276         int rc, size[3] = { sizeof(struct ptlrpc_body), sizeof(*body) };
1277         ENTRY;
1278
1279         size[REQ_REC_OFF + 1] = oc ? sizeof(struct lustre_capa) : 0;
1280         req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
1281                               MDS_PIN, 3, size, NULL);
1282         if (req == NULL)
1283                 RETURN(-ENOMEM);
1284
1285         body = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF, sizeof (*body));
1286         body->fid1 = *fid;
1287         body->flags = flag;
1288         mdc_pack_capa(req, REQ_REC_OFF + 1, oc);
1289
1290         ptlrpc_req_set_repsize(req, 2, size);
1291
1292         mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
1293         rc = ptlrpc_queue_wait(req);
1294         mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
1295         if (rc) {
1296                 CERROR("pin failed: %d\n", rc);
1297                 ptlrpc_req_finished(req);
1298                 RETURN(rc);
1299         }
1300
1301         body = lustre_swab_repbuf(req, REPLY_REC_OFF, sizeof(*body),
1302                                   lustre_swab_mdt_body);
1303         if (body == NULL) {
1304                 ptlrpc_req_finished(req);
1305                 RETURN(rc);
1306         }
1307
1308         memcpy(&handle->och_fh, &body->handle, sizeof(body->handle));
1309         handle->och_magic = OBD_CLIENT_HANDLE_MAGIC;
1310
1311         OBD_ALLOC(handle->och_mod, sizeof(*handle->och_mod));
1312         if (handle->och_mod == NULL) {
1313                 DEBUG_REQ(D_ERROR, req, "can't allocate md_open_data");
1314                 RETURN(-ENOMEM);
1315         }
1316
1317         /* will be dropped by unpin */
1318         CFS_INIT_LIST_HEAD(&handle->och_mod->mod_replay_list);
1319         list_add_tail(&req->rq_mod_list, &handle->och_mod->mod_replay_list);
1320
1321         RETURN(rc);
1322 }
1323
1324 static int mdc_unpin(struct obd_export *exp,
1325                      struct obd_client_handle *handle, int flag)
1326 {
1327         struct ptlrpc_request *req;
1328         struct mdt_body *body;
1329         int rc, size[2] = { sizeof(struct ptlrpc_body), sizeof(*body) };
1330         ENTRY;
1331
1332         if (handle->och_magic != OBD_CLIENT_HANDLE_MAGIC)
1333                 RETURN(0);
1334
1335         req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
1336                               MDS_CLOSE, 2, size, NULL);
1337         if (req == NULL)
1338                 RETURN(-ENOMEM);
1339
1340         body = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF, sizeof(*body));
1341         memcpy(&body->handle, &handle->och_fh, sizeof(body->handle));
1342         body->flags = flag;
1343
1344         ptlrpc_req_set_repsize(req, 1, NULL);
1345         mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
1346         rc = ptlrpc_queue_wait(req);
1347         mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
1348
1349         if (rc != 0)
1350                 CERROR("unpin failed: %d\n", rc);
1351
1352         ptlrpc_req_finished(req);
1353
1354         LASSERT(!list_empty(&handle->och_mod->mod_replay_list));
1355         req = list_entry(handle->och_mod->mod_replay_list.next,
1356                          typeof(*req), rq_mod_list);
1357         list_del_init(&req->rq_mod_list);
1358         ptlrpc_req_finished(req);
1359         LASSERT(list_empty(&handle->och_mod->mod_replay_list));
1360
1361         OBD_FREE(handle->och_mod, sizeof(*handle->och_mod));
1362         RETURN(rc);
1363 }
1364
1365 int mdc_sync(struct obd_export *exp, const struct lu_fid *fid,
1366              struct obd_capa *oc, struct ptlrpc_request **request)
1367 {
1368         struct ptlrpc_request *req;
1369         int size[3] = { sizeof(struct ptlrpc_body), sizeof(struct mdt_body) };
1370         int rc;
1371         ENTRY;
1372
1373         size[REQ_REC_OFF + 1] = oc ? sizeof(struct lustre_capa) : 0;
1374         req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
1375                               MDS_SYNC, 3, size, NULL);
1376         if (!req)
1377                 RETURN(rc = -ENOMEM);
1378
1379         mdc_pack_req_body(req, REQ_REC_OFF, 0, fid, oc, 0, 0);
1380
1381         ptlrpc_req_set_repsize(req, 2, size);
1382
1383         rc = ptlrpc_queue_wait(req);
1384         if (rc || request == NULL)
1385                 ptlrpc_req_finished(req);
1386         else
1387                 *request = req;
1388
1389         RETURN(rc);
1390 }
1391
1392 static int mdc_import_event(struct obd_device *obd, struct obd_import *imp,
1393                             enum obd_import_event event)
1394 {
1395         int rc = 0;
1396
1397         LASSERT(imp->imp_obd == obd);
1398
1399         switch (event) {
1400         case IMP_EVENT_DISCON: {
1401 #if 0
1402                 /* XXX Pass event up to OBDs stack. used only for FLD now */
1403                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_DISCON, NULL);
1404 #endif
1405                 break;
1406         }
1407         case IMP_EVENT_INACTIVE: {
1408                 struct client_obd *cli = &obd->u.cli;
1409                 /*
1410                  * Flush current sequence to make client obtain new one
1411                  * from server in case of disconnect/reconnect.
1412                  * If range is already empty then no need to flush it.
1413                  */
1414                 if (cli->cl_seq != NULL &&
1415                     !range_is_exhausted(&cli->cl_seq->lcs_space)) {
1416                         seq_client_flush(cli->cl_seq);
1417                 }
1418
1419                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_INACTIVE, NULL);
1420                 break;
1421         }
1422         case IMP_EVENT_INVALIDATE: {
1423                 struct ldlm_namespace *ns = obd->obd_namespace;
1424
1425                 ldlm_namespace_cleanup(ns, LDLM_FL_LOCAL_ONLY);
1426
1427                 break;
1428         }
1429         case IMP_EVENT_ACTIVE: {
1430                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_ACTIVE, NULL);
1431                 break;
1432         }
1433         case IMP_EVENT_OCD:
1434                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_OCD, NULL);
1435                 break;
1436
1437         default:
1438                 CERROR("Unknown import event %x\n", event);
1439                 LBUG();
1440         }
1441         RETURN(rc);
1442 }
1443
1444 static int mdc_fid_init(struct obd_export *exp)
1445 {
1446         struct client_obd *cli = &exp->exp_obd->u.cli;
1447         char *prefix;
1448         int rc;
1449         ENTRY;
1450
1451         OBD_ALLOC_PTR(cli->cl_seq);
1452         if (cli->cl_seq == NULL)
1453                 RETURN(-ENOMEM);
1454
1455         OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
1456         if (prefix == NULL)
1457                 GOTO(out_free_seq, rc = -ENOMEM);
1458
1459         snprintf(prefix, MAX_OBD_NAME + 5, "srv-%s",
1460                  exp->exp_obd->obd_name);
1461
1462         /* Init client side sequence-manager */
1463         rc = seq_client_init(cli->cl_seq, exp,
1464                              LUSTRE_SEQ_METADATA,
1465                              prefix, NULL);
1466         OBD_FREE(prefix, MAX_OBD_NAME + 5);
1467         if (rc)
1468                 GOTO(out_free_seq, rc);
1469
1470         RETURN(rc);
1471 out_free_seq:
1472         OBD_FREE_PTR(cli->cl_seq);
1473         cli->cl_seq = NULL;
1474         return rc;
1475 }
1476
1477 static int mdc_fid_fini(struct obd_export *exp)
1478 {
1479         struct client_obd *cli = &exp->exp_obd->u.cli;
1480         ENTRY;
1481
1482         if (cli->cl_seq != NULL) {
1483                 seq_client_fini(cli->cl_seq);
1484                 OBD_FREE_PTR(cli->cl_seq);
1485                 cli->cl_seq = NULL;
1486         }
1487
1488         RETURN(0);
1489 }
1490
1491 int mdc_fid_alloc(struct obd_export *exp, struct lu_fid *fid,
1492                   struct md_op_data *op_data)
1493 {
1494         struct client_obd *cli = &exp->exp_obd->u.cli;
1495         struct lu_client_seq *seq = cli->cl_seq;
1496         ENTRY;
1497         RETURN(seq_client_alloc_fid(seq, fid));
1498 }
1499
1500 /* XXX This method is used only to clear current fid seq
1501  * once fld/mds insert failed */
1502 static int mdc_fid_delete(struct obd_export *exp, const struct lu_fid *fid)
1503 {
1504         struct client_obd *cli = &exp->exp_obd->u.cli;
1505
1506         seq_client_flush(cli->cl_seq);
1507         return 0;
1508 }
1509
1510 static int mdc_setup(struct obd_device *obd, struct lustre_cfg *cfg)
1511 {
1512         struct client_obd *cli = &obd->u.cli;
1513         struct lprocfs_static_vars lvars;
1514         int rc;
1515         ENTRY;
1516
1517         OBD_ALLOC(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
1518         if (!cli->cl_rpc_lock)
1519                 RETURN(-ENOMEM);
1520         mdc_init_rpc_lock(cli->cl_rpc_lock);
1521
1522         ptlrpcd_addref();
1523
1524         OBD_ALLOC(cli->cl_setattr_lock, sizeof (*cli->cl_setattr_lock));
1525         if (!cli->cl_setattr_lock)
1526                 GOTO(err_rpc_lock, rc = -ENOMEM);
1527         mdc_init_rpc_lock(cli->cl_setattr_lock);
1528
1529         OBD_ALLOC(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
1530         if (!cli->cl_close_lock)
1531                 GOTO(err_setattr_lock, rc = -ENOMEM);
1532         mdc_init_rpc_lock(cli->cl_close_lock);
1533
1534         rc = client_obd_setup(obd, cfg);
1535         if (rc)
1536                 GOTO(err_close_lock, rc);
1537         lprocfs_init_vars(mdc, &lvars);
1538         lprocfs_obd_setup(obd, lvars.obd_vars);
1539         ptlrpc_lprocfs_register_obd(obd);
1540
1541         rc = obd_llog_init(obd, NULL, obd, 0, NULL, NULL);
1542         if (rc) {
1543                 mdc_cleanup(obd);
1544                 CERROR("failed to setup llogging subsystems\n");
1545         }
1546
1547         RETURN(rc);
1548
1549 err_close_lock:
1550         OBD_FREE(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
1551 err_setattr_lock:
1552         OBD_FREE(cli->cl_setattr_lock, sizeof (*cli->cl_setattr_lock));
1553 err_rpc_lock:
1554         OBD_FREE(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
1555         ptlrpcd_decref();
1556         RETURN(rc);
1557 }
1558
1559 /* Initialize the default and maximum LOV EA and cookie sizes.  This allows
1560  * us to make MDS RPCs with large enough reply buffers to hold the
1561  * maximum-sized (= maximum striped) EA and cookie without having to
1562  * calculate this (via a call into the LOV + OSCs) each time we make an RPC. */
1563 int mdc_init_ea_size(struct obd_export *exp, int easize,
1564                      int def_easize, int cookiesize)
1565 {
1566         struct obd_device *obd = exp->exp_obd;
1567         struct client_obd *cli = &obd->u.cli;
1568         ENTRY;
1569
1570         if (cli->cl_max_mds_easize < easize)
1571                 cli->cl_max_mds_easize = easize;
1572
1573         if (cli->cl_default_mds_easize < def_easize)
1574                 cli->cl_default_mds_easize = def_easize;
1575
1576         if (cli->cl_max_mds_cookiesize < cookiesize)
1577                 cli->cl_max_mds_cookiesize = cookiesize;
1578
1579         RETURN(0);
1580 }
1581
1582 static int mdc_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
1583 {
1584         int rc = 0;
1585         ENTRY;
1586
1587         switch (stage) {
1588         case OBD_CLEANUP_EARLY:
1589         case OBD_CLEANUP_EXPORTS:
1590                 /* If we set up but never connected, the
1591                    client import will not have been cleaned. */
1592                 if (obd->u.cli.cl_import) {
1593                         struct obd_import *imp;
1594                         imp = obd->u.cli.cl_import;
1595                         CERROR("client import never connected\n");
1596                         ptlrpc_invalidate_import(imp);
1597                         ptlrpc_free_rq_pool(imp->imp_rq_pool);
1598                         class_destroy_import(imp);
1599                         obd->u.cli.cl_import = NULL;
1600                 }
1601                 break;
1602         case OBD_CLEANUP_SELF_EXP:
1603                 rc = obd_llog_finish(obd, 0);
1604                 if (rc != 0)
1605                         CERROR("failed to cleanup llogging subsystems\n");
1606         case OBD_CLEANUP_OBD:
1607                 break;
1608         }
1609         RETURN(rc);
1610 }
1611
1612 static int mdc_cleanup(struct obd_device *obd)
1613 {
1614         struct client_obd *cli = &obd->u.cli;
1615
1616         OBD_FREE(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
1617         OBD_FREE(cli->cl_setattr_lock, sizeof (*cli->cl_setattr_lock));
1618         OBD_FREE(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
1619
1620         ptlrpc_lprocfs_unregister_obd(obd);
1621         lprocfs_obd_cleanup(obd);
1622         ptlrpcd_decref();
1623
1624         return client_obd_cleanup(obd);
1625 }
1626
1627
1628 static int mdc_llog_init(struct obd_device *obd, struct obd_llogs *llogs,
1629                          struct obd_device *tgt,
1630                          int count, struct llog_catid *logid,
1631                          struct obd_uuid *uuid)
1632 {
1633         struct llog_ctxt *ctxt;
1634         int rc;
1635         ENTRY;
1636
1637         rc = llog_setup(obd, llogs, LLOG_CONFIG_REPL_CTXT, tgt, 0, NULL,
1638                         &llog_client_ops);
1639         if (rc == 0) {
1640                 ctxt = llog_get_context(obd, LLOG_CONFIG_REPL_CTXT);
1641                 ctxt->loc_imp = obd->u.cli.cl_import;
1642         }
1643
1644         rc = llog_setup(obd, llogs, LLOG_LOVEA_REPL_CTXT, tgt, 0, NULL,
1645                        &llog_client_ops);
1646         if (rc == 0) {
1647                 ctxt = llog_get_context(obd, LLOG_LOVEA_REPL_CTXT);
1648                 ctxt->loc_imp = obd->u.cli.cl_import;
1649         }
1650
1651         RETURN(rc);
1652 }
1653
1654 static int mdc_llog_finish(struct obd_device *obd, int count)
1655 {
1656         int rc;
1657         ENTRY;
1658
1659         rc = llog_cleanup(llog_get_context(obd, LLOG_LOVEA_REPL_CTXT));
1660         if (rc) {
1661                 CERROR("can not cleanup LLOG_CONFIG_REPL_CTXT rc %d\n", rc);
1662         }
1663         rc = llog_cleanup(llog_get_context(obd, LLOG_CONFIG_REPL_CTXT));
1664         RETURN(rc);
1665 }
1666
1667 static int mdc_process_config(struct obd_device *obd, obd_count len, void *buf)
1668 {
1669         struct lustre_cfg *lcfg = buf;
1670         struct lprocfs_static_vars lvars;
1671         int rc = 0;
1672
1673         lprocfs_init_vars(mdc, &lvars);
1674
1675         rc = class_process_proc_param(PARAM_MDC, lvars.obd_vars, lcfg, obd);
1676         return(rc);
1677 }
1678
1679 /* get remote permission for current user on fid */
1680 int mdc_get_remote_perm(struct obd_export *exp, const struct lu_fid *fid,
1681                         struct obd_capa *oc, struct ptlrpc_request **request)
1682 {
1683         struct ptlrpc_request *req;
1684         struct mdt_body *body;
1685         struct mdt_remote_perm *perm;
1686         int size[5] = { sizeof(struct ptlrpc_body), sizeof(*body) };
1687         int rc;
1688         ENTRY;
1689
1690         size[REQ_REC_OFF + 1] = oc ? sizeof(struct lustre_capa) : 0;
1691
1692         *request = NULL;
1693         req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
1694                               MDS_GETATTR, 3, size, NULL);
1695         if (!req)
1696                 RETURN(-ENOMEM);
1697
1698         mdc_pack_req_body(req, REQ_REC_OFF, OBD_MD_FLRMTPERM, fid, oc, 0, 0);
1699
1700         size[REPLY_REC_OFF + 1] = sizeof(*perm);
1701         ptlrpc_req_set_repsize(req, 5, size);
1702         rc = ptlrpc_queue_wait(req);
1703         if (rc) {
1704                 ptlrpc_req_finished(req);
1705                 RETURN(rc);
1706         }
1707
1708         body = lustre_swab_repbuf(req, REPLY_REC_OFF, sizeof(*body),
1709                                   lustre_swab_mdt_body);
1710         LASSERT(body);
1711         LASSERT(body->valid & OBD_MD_FLRMTPERM);
1712
1713         perm = lustre_swab_repbuf(req, REPLY_REC_OFF + 1, sizeof(*perm),
1714                                   lustre_swab_mdt_remote_perm);
1715         LASSERT(perm);
1716
1717         *request = req;
1718         RETURN(0);
1719 }
1720
1721 static int mdc_interpret_renew_capa(struct ptlrpc_request *req, void *unused,
1722                                     int status)
1723 {
1724         struct obd_capa *oc = req->rq_async_args.pointer_arg[0];
1725         renew_capa_cb_t cb = req->rq_async_args.pointer_arg[1];
1726         struct mdt_body *body = NULL;
1727         struct lustre_capa *capa;
1728         ENTRY;
1729
1730         if (status)
1731                 GOTO(out, capa = ERR_PTR(status));
1732
1733         body = lustre_swab_repbuf(req, REPLY_REC_OFF, sizeof(*body),
1734                                   lustre_swab_mdt_body);
1735         if (body == NULL)
1736                 GOTO(out, capa = ERR_PTR(-EFAULT));
1737
1738         if ((body->valid & OBD_MD_FLOSSCAPA) == 0)
1739                 GOTO(out, capa = ERR_PTR(-ENOENT));
1740
1741         capa = lustre_unpack_capa(req->rq_repmsg, REPLY_REC_OFF + 1);
1742         if (!capa)
1743                 GOTO(out, capa = ERR_PTR(-EFAULT));
1744         EXIT;
1745 out:
1746         cb(oc, capa);
1747         return 0;
1748 }
1749
1750 static int mdc_renew_capa(struct obd_export *exp, struct obd_capa *oc,
1751                           renew_capa_cb_t cb)
1752 {
1753         struct ptlrpc_request *req;
1754         int size[5] = { sizeof(struct ptlrpc_body),
1755                         sizeof(struct mdt_body),
1756                         sizeof(struct lustre_capa) };
1757         ENTRY;
1758
1759         req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
1760                               MDS_GETATTR, 3, size, NULL);
1761         if (!req)
1762                 RETURN(-ENOMEM);
1763
1764         mdc_pack_req_body(req, REQ_REC_OFF, OBD_MD_FLOSSCAPA,
1765                           &oc->c_capa.lc_fid, oc, 0, 0);
1766
1767         ptlrpc_req_set_repsize(req, 5, size);
1768         req->rq_async_args.pointer_arg[0] = oc;
1769         req->rq_async_args.pointer_arg[1] = cb;
1770         req->rq_interpret_reply = mdc_interpret_renew_capa;
1771         ptlrpcd_add_req(req);
1772
1773         RETURN(0);
1774 }
1775
1776 static int mdc_connect(const struct lu_env *env,
1777                        struct lustre_handle *dlm_handle,
1778                        struct obd_device *obd, struct obd_uuid *cluuid,
1779                        struct obd_connect_data *data) {
1780         struct obd_import *imp = obd->u.cli.cl_import;
1781
1782         /* mds-mds import features */
1783         if (data && (data->ocd_connect_flags & OBD_CONNECT_MDS_MDS)) {
1784                 spin_lock(&imp->imp_lock);
1785                 imp->imp_server_timeout = 1;
1786                 spin_unlock(&imp->imp_lock);
1787                 imp->imp_client->cli_request_portal = MDS_MDS_PORTAL;
1788                 CDEBUG(D_OTHER, "%s: Set 'mds' portal and timeout\n",
1789                        obd->obd_name);
1790         }
1791
1792         return client_connect_import(env, dlm_handle, obd, cluuid, data);
1793 }
1794
1795 struct obd_ops mdc_obd_ops = {
1796         .o_owner            = THIS_MODULE,
1797         .o_setup            = mdc_setup,
1798         .o_precleanup       = mdc_precleanup,
1799         .o_cleanup          = mdc_cleanup,
1800         .o_add_conn         = client_import_add_conn,
1801         .o_del_conn         = client_import_del_conn,
1802         .o_connect          = mdc_connect,
1803         .o_disconnect       = client_disconnect_export,
1804         .o_iocontrol        = mdc_iocontrol,
1805         .o_set_info_async   = mdc_set_info_async,
1806         .o_statfs           = mdc_statfs,
1807         .o_pin              = mdc_pin,
1808         .o_unpin            = mdc_unpin,
1809         .o_fid_init         = mdc_fid_init,
1810         .o_fid_fini         = mdc_fid_fini,
1811         .o_fid_alloc        = mdc_fid_alloc,
1812         .o_fid_delete       = mdc_fid_delete,
1813         .o_import_event     = mdc_import_event,
1814         .o_llog_init        = mdc_llog_init,
1815         .o_llog_finish      = mdc_llog_finish,
1816         .o_get_info         = mdc_get_info,
1817         .o_process_config  = mdc_process_config,
1818 };
1819
1820 struct md_ops mdc_md_ops = {
1821         .m_getstatus        = mdc_getstatus,
1822         .m_change_cbdata    = mdc_change_cbdata,
1823         .m_close            = mdc_close,
1824         .m_create           = mdc_create,
1825         .m_done_writing     = mdc_done_writing,
1826         .m_enqueue          = mdc_enqueue,
1827         .m_getattr          = mdc_getattr,
1828         .m_getattr_name     = mdc_getattr_name,
1829         .m_intent_lock      = mdc_intent_lock,
1830         .m_link             = mdc_link,
1831         .m_is_subdir        = mdc_is_subdir,
1832         .m_rename           = mdc_rename,
1833         .m_setattr          = mdc_setattr,
1834         .m_setxattr         = mdc_setxattr,
1835         .m_getxattr         = mdc_getxattr,
1836         .m_sync             = mdc_sync,
1837         .m_readpage         = mdc_readpage,
1838         .m_unlink           = mdc_unlink,
1839         .m_cancel_unused    = mdc_cancel_unused,
1840         .m_init_ea_size     = mdc_init_ea_size,
1841         .m_set_lock_data    = mdc_set_lock_data,
1842         .m_lock_match       = mdc_lock_match,
1843         .m_get_lustre_md    = mdc_get_lustre_md,
1844         .m_free_lustre_md   = mdc_free_lustre_md,
1845         .m_set_open_replay_data = mdc_set_open_replay_data,
1846         .m_clear_open_replay_data = mdc_clear_open_replay_data,
1847         .m_get_remote_perm  = mdc_get_remote_perm,
1848         .m_renew_capa       = mdc_renew_capa
1849 };
1850
1851 extern quota_interface_t mdc_quota_interface;
1852
1853 int __init mdc_init(void)
1854 {
1855         int rc;
1856         struct lprocfs_static_vars lvars;
1857         lprocfs_init_vars(mdc, &lvars);
1858
1859         request_module("lquota");
1860         quota_interface = PORTAL_SYMBOL_GET(mdc_quota_interface);
1861         init_obd_quota_ops(quota_interface, &mdc_obd_ops);
1862
1863         rc = class_register_type(&mdc_obd_ops, &mdc_md_ops, lvars.module_vars,
1864                                  LUSTRE_MDC_NAME, NULL);
1865         if (rc && quota_interface)
1866                 PORTAL_SYMBOL_PUT(mdc_quota_interface);
1867
1868         RETURN(rc);
1869 }
1870
1871 #ifdef __KERNEL__
1872 static void /*__exit*/ mdc_exit(void)
1873 {
1874         if (quota_interface)
1875                 PORTAL_SYMBOL_PUT(mdc_quota_interface);
1876
1877         class_unregister_type(LUSTRE_MDC_NAME);
1878 }
1879
1880 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
1881 MODULE_DESCRIPTION("Lustre Metadata Client");
1882 MODULE_LICENSE("GPL");
1883
1884 module_init(mdc_init);
1885 module_exit(mdc_exit);
1886 #endif