drivers/staging/lustre/lustre/mdc/mdc_request.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2001, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2011, 2015, Intel Corporation.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  */
  36
  37 #define DEBUG_SUBSYSTEM S_MDC
  38
  39 # include <linux/module.h>
  40 # include <linux/pagemap.h>
  41 # include <linux/miscdevice.h>
  42 # include <linux/init.h>
  43 # include <linux/utsname.h>
  44
  45 #include "../include/lustre_acl.h"
  46 #include "../include/obd_class.h"
  47 #include "../include/lustre_fid.h"
  48 #include "../include/lprocfs_status.h"
  49 #include "../include/lustre_param.h"
  50 #include "../include/lustre_log.h"
  51
  52 #include "mdc_internal.h"
  53
  54 #define REQUEST_MINOR 244
  55
  56 static int mdc_cleanup(struct obd_device *obd);
  57
  58 static inline int mdc_queue_wait(struct ptlrpc_request *req)
  59 {
  60         struct client_obd *cli = &req->rq_import->imp_obd->u.cli;
  61         int rc;
  62
  63         /* mdc_enter_request() ensures that this client has no more
  64          * than cl_max_rpcs_in_flight RPCs simultaneously inf light
  65          * against an MDT. */
  66         rc = mdc_enter_request(cli);
  67         if (rc != 0)
  68                 return rc;
  69
  70         rc = ptlrpc_queue_wait(req);
  71         mdc_exit_request(cli);
  72
  73         return rc;
  74 }
  75
  76 static int mdc_getstatus(struct obd_export *exp, struct lu_fid *rootfid)
  77 {
  78         struct ptlrpc_request *req;
  79         struct mdt_body       *body;
  80         int                 rc;
  81
  82         req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp),
  83                                         &RQF_MDS_GETSTATUS,
  84                                         LUSTRE_MDS_VERSION, MDS_GETSTATUS);
  85         if (req == NULL)
  86                 return -ENOMEM;
  87
  88         mdc_pack_body(req, NULL, 0, 0, -1, 0);
  89         req->rq_send_state = LUSTRE_IMP_FULL;
  90
  91         ptlrpc_request_set_replen(req);
  92
  93         rc = ptlrpc_queue_wait(req);
  94         if (rc)
  95                 goto out;
  96
  97         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
  98         if (body == NULL) {
  99                 rc = -EPROTO;
 100                 goto out;
 101         }
 102
 103         *rootfid = body->fid1;
 104         CDEBUG(D_NET,
 105                "root fid="DFID", last_committed=%llu\n",
 106                PFID(rootfid),
 107                lustre_msg_get_last_committed(req->rq_repmsg));
 108 out:
 109         ptlrpc_req_finished(req);
 110         return rc;
 111 }
 112
 113 /*
 114  * This function now is known to always saying that it will receive 4 buffers
 115  * from server. Even for cases when acl_size and md_size is zero, RPC header
 116  * will contain 4 fields and RPC itself will contain zero size fields. This is
 117  * because mdt_getattr*() _always_ returns 4 fields, but if acl is not needed
 118  * and thus zero, it shrinks it, making zero size. The same story about
 119  * md_size. And this is course of problem when client waits for smaller number
 120  * of fields. This issue will be fixed later when client gets aware of RPC
 121  * layouts.  --umka
 122  */
 123 static int mdc_getattr_common(struct obd_export *exp,
 124                               struct ptlrpc_request *req)
 125 {
 126         struct req_capsule *pill = &req->rq_pill;
 127         struct mdt_body    *body;
 128         void           *eadata;
 129         int              rc;
 130
 131         /* Request message already built. */
 132         rc = ptlrpc_queue_wait(req);
 133         if (rc != 0)
 134                 return rc;
 135
 136         /* sanity check for the reply */
 137         body = req_capsule_server_get(pill, &RMF_MDT_BODY);
 138         if (body == NULL)
 139                 return -EPROTO;
 140
 141         CDEBUG(D_NET, "mode: %o\n", body->mode);
 142
 143         if (body->eadatasize != 0) {
 144                 mdc_update_max_ea_from_body(exp, body);
 145
 146                 eadata = req_capsule_server_sized_get(pill, &RMF_MDT_MD,
 147                                                       body->eadatasize);
 148                 if (eadata == NULL)
 149                         return -EPROTO;
 150         }
 151
 152         if (body->valid & OBD_MD_FLRMTPERM) {
 153                 struct mdt_remote_perm *perm;
 154
 155                 LASSERT(client_is_remote(exp));
 156                 perm = req_capsule_server_swab_get(pill, &RMF_ACL,
 157                                                 lustre_swab_mdt_remote_perm);
 158                 if (perm == NULL)
 159                         return -EPROTO;
 160         }
 161
 162         return 0;
 163 }
 164
 165 static int mdc_getattr(struct obd_export *exp, struct md_op_data *op_data,
 166                 struct ptlrpc_request **request)
 167 {
 168         struct ptlrpc_request *req;
 169         int                 rc;
 170
 171         /* Single MDS without an LMV case */
 172         if (op_data->op_flags & MF_GET_MDT_IDX) {
 173                 op_data->op_mds = 0;
 174                 return 0;
 175         }
 176         *request = NULL;
 177         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_GETATTR);
 178         if (req == NULL)
 179                 return -ENOMEM;
 180
 181         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR);
 182         if (rc) {
 183                 ptlrpc_request_free(req);
 184                 return rc;
 185         }
 186
 187         mdc_pack_body(req, &op_data->op_fid1, op_data->op_valid,
 188                       op_data->op_mode, -1, 0);
 189
 190         req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
 191                              op_data->op_mode);
 192         if (op_data->op_valid & OBD_MD_FLRMTPERM) {
 193                 LASSERT(client_is_remote(exp));
 194                 req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER,
 195                                      sizeof(struct mdt_remote_perm));
 196         }
 197         ptlrpc_request_set_replen(req);
 198
 199         rc = mdc_getattr_common(exp, req);
 200         if (rc)
 201                 ptlrpc_req_finished(req);
 202         else
 203                 *request = req;
 204         return rc;
 205 }
 206
 207 static int mdc_getattr_name(struct obd_export *exp, struct md_op_data *op_data,
 208                      struct ptlrpc_request **request)
 209 {
 210         struct ptlrpc_request *req;
 211         int                 rc;
 212
 213         *request = NULL;
 214         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
 215                                    &RQF_MDS_GETATTR_NAME);
 216         if (req == NULL)
 217                 return -ENOMEM;
 218
 219         req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
 220                              op_data->op_namelen + 1);
 221
 222         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR_NAME);
 223         if (rc) {
 224                 ptlrpc_request_free(req);
 225                 return rc;
 226         }
 227
 228         mdc_pack_body(req, &op_data->op_fid1, op_data->op_valid,
 229                       op_data->op_mode, op_data->op_suppgids[0], 0);
 230
 231         if (op_data->op_name) {
 232                 char *name = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
 233
 234                 LASSERT(strnlen(op_data->op_name, op_data->op_namelen) ==
 235                                 op_data->op_namelen);
 236                 memcpy(name, op_data->op_name, op_data->op_namelen);
 237         }
 238
 239         req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
 240                              op_data->op_mode);
 241         ptlrpc_request_set_replen(req);
 242
 243         rc = mdc_getattr_common(exp, req);
 244         if (rc)
 245                 ptlrpc_req_finished(req);
 246         else
 247                 *request = req;
 248         return rc;
 249 }
 250
 251 static int mdc_is_subdir(struct obd_export *exp,
 252                          const struct lu_fid *pfid,
 253                          const struct lu_fid *cfid,
 254                          struct ptlrpc_request **request)
 255 {
 256         struct ptlrpc_request  *req;
 257         int                  rc;
 258
 259         *request = NULL;
 260         req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp),
 261                                         &RQF_MDS_IS_SUBDIR, LUSTRE_MDS_VERSION,
 262                                         MDS_IS_SUBDIR);
 263         if (req == NULL)
 264                 return -ENOMEM;
 265
 266         mdc_is_subdir_pack(req, pfid, cfid, 0);
 267         ptlrpc_request_set_replen(req);
 268
 269         rc = ptlrpc_queue_wait(req);
 270         if (rc && rc != -EREMOTE)
 271                 ptlrpc_req_finished(req);
 272         else
 273                 *request = req;
 274         return rc;
 275 }
 276
 277 static int mdc_xattr_common(struct obd_export *exp,
 278                             const struct req_format *fmt,
 279                             const struct lu_fid *fid,
 280                             int opcode, u64 valid,
 281                             const char *xattr_name, const char *input,
 282                             int input_size, int output_size, int flags,
 283                             __u32 suppgid, struct ptlrpc_request **request)
 284 {
 285         struct ptlrpc_request *req;
 286         int   xattr_namelen = 0;
 287         char *tmp;
 288         int   rc;
 289
 290         *request = NULL;
 291         req = ptlrpc_request_alloc(class_exp2cliimp(exp), fmt);
 292         if (req == NULL)
 293                 return -ENOMEM;
 294
 295         if (xattr_name) {
 296                 xattr_namelen = strlen(xattr_name) + 1;
 297                 req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
 298                                      xattr_namelen);
 299         }
 300         if (input_size) {
 301                 LASSERT(input);
 302                 req_capsule_set_size(&req->rq_pill, &RMF_EADATA, RCL_CLIENT,
 303                                      input_size);
 304         }
 305
 306         /* Flush local XATTR locks to get rid of a possible cancel RPC */
 307         if (opcode == MDS_REINT && fid_is_sane(fid) &&
 308             exp->exp_connect_data.ocd_ibits_known & MDS_INODELOCK_XATTR) {
 309                 LIST_HEAD(cancels);
 310                 int count;
 311
 312                 /* Without that packing would fail */
 313                 if (input_size == 0)
 314                         req_capsule_set_size(&req->rq_pill, &RMF_EADATA,
 315                                              RCL_CLIENT, 0);
 316
 317                 count = mdc_resource_get_unused(exp, fid,
 318                                                 &cancels, LCK_EX,
 319                                                 MDS_INODELOCK_XATTR);
 320
 321                 rc = mdc_prep_elc_req(exp, req, MDS_REINT, &cancels, count);
 322                 if (rc) {
 323                         ptlrpc_request_free(req);
 324                         return rc;
 325                 }
 326         } else {
 327                 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, opcode);
 328                 if (rc) {
 329                         ptlrpc_request_free(req);
 330                         return rc;
 331                 }
 332         }
 333
 334         if (opcode == MDS_REINT) {
 335                 struct mdt_rec_setxattr *rec;
 336
 337                 CLASSERT(sizeof(struct mdt_rec_setxattr) ==
 338                          sizeof(struct mdt_rec_reint));
 339                 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
 340                 rec->sx_opcode = REINT_SETXATTR;
 341                 rec->sx_fsuid  = from_kuid(&init_user_ns, current_fsuid());
 342                 rec->sx_fsgid  = from_kgid(&init_user_ns, current_fsgid());
 343                 rec->sx_cap    = cfs_curproc_cap_pack();
 344                 rec->sx_suppgid1 = suppgid;
 345                 rec->sx_suppgid2 = -1;
 346                 rec->sx_fid    = *fid;
 347                 rec->sx_valid  = valid | OBD_MD_FLCTIME;
 348                 rec->sx_time   = ktime_get_real_seconds();
 349                 rec->sx_size   = output_size;
 350                 rec->sx_flags  = flags;
 351
 352         } else {
 353                 mdc_pack_body(req, fid, valid, output_size, suppgid, flags);
 354         }
 355
 356         if (xattr_name) {
 357                 tmp = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
 358                 memcpy(tmp, xattr_name, xattr_namelen);
 359         }
 360         if (input_size) {
 361                 tmp = req_capsule_client_get(&req->rq_pill, &RMF_EADATA);
 362                 memcpy(tmp, input, input_size);
 363         }
 364
 365         if (req_capsule_has_field(&req->rq_pill, &RMF_EADATA, RCL_SERVER))
 366                 req_capsule_set_size(&req->rq_pill, &RMF_EADATA,
 367                                      RCL_SERVER, output_size);
 368         ptlrpc_request_set_replen(req);
 369
 370         /* make rpc */
 371         if (opcode == MDS_REINT)
 372                 mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
 373
 374         rc = ptlrpc_queue_wait(req);
 375
 376         if (opcode == MDS_REINT)
 377                 mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
 378
 379         if (rc)
 380                 ptlrpc_req_finished(req);
 381         else
 382                 *request = req;
 383         return rc;
 384 }
 385
 386 static int mdc_setxattr(struct obd_export *exp, const struct lu_fid *fid,
 387                         u64 valid, const char *xattr_name,
 388                         const char *input, int input_size, int output_size,
 389                         int flags, __u32 suppgid,
 390                         struct ptlrpc_request **request)
 391 {
 392         return mdc_xattr_common(exp, &RQF_MDS_REINT_SETXATTR,
 393                                 fid, MDS_REINT, valid, xattr_name,
 394                                 input, input_size, output_size, flags,
 395                                 suppgid, request);
 396 }
 397
 398 static int mdc_getxattr(struct obd_export *exp, const struct lu_fid *fid,
 399                         u64 valid, const char *xattr_name,
 400                         const char *input, int input_size, int output_size,
 401                         int flags, struct ptlrpc_request **request)
 402 {
 403         return mdc_xattr_common(exp, &RQF_MDS_GETXATTR,
 404                                 fid, MDS_GETXATTR, valid, xattr_name,
 405                                 input, input_size, output_size, flags,
 406                                 -1, request);
 407 }
 408
 409 #ifdef CONFIG_FS_POSIX_ACL
 410 static int mdc_unpack_acl(struct ptlrpc_request *req, struct lustre_md *md)
 411 {
 412         struct req_capsule     *pill = &req->rq_pill;
 413         struct mdt_body *body = md->body;
 414         struct posix_acl       *acl;
 415         void               *buf;
 416         int                  rc;
 417
 418         if (!body->aclsize)
 419                 return 0;
 420
 421         buf = req_capsule_server_sized_get(pill, &RMF_ACL, body->aclsize);
 422
 423         if (!buf)
 424                 return -EPROTO;
 425
 426         acl = posix_acl_from_xattr(&init_user_ns, buf, body->aclsize);
 427         if (acl == NULL)
 428                 return 0;
 429
 430         if (IS_ERR(acl)) {
 431                 rc = PTR_ERR(acl);
 432                 CERROR("convert xattr to acl: %d\n", rc);
 433                 return rc;
 434         }
 435
 436         rc = posix_acl_valid(acl);
 437         if (rc) {
 438                 CERROR("validate acl: %d\n", rc);
 439                 posix_acl_release(acl);
 440                 return rc;
 441         }
 442
 443         md->posix_acl = acl;
 444         return 0;
 445 }
 446 #else
 447 #define mdc_unpack_acl(req, md) 0
 448 #endif
 449
 450 static int mdc_get_lustre_md(struct obd_export *exp,
 451                              struct ptlrpc_request *req,
 452                              struct obd_export *dt_exp,
 453                              struct obd_export *md_exp,
 454                              struct lustre_md *md)
 455 {
 456         struct req_capsule *pill = &req->rq_pill;
 457         int rc;
 458
 459         LASSERT(md);
 460         memset(md, 0, sizeof(*md));
 461
 462         md->body = req_capsule_server_get(pill, &RMF_MDT_BODY);
 463         LASSERT(md->body != NULL);
 464
 465         if (md->body->valid & OBD_MD_FLEASIZE) {
 466                 int lmmsize;
 467                 struct lov_mds_md *lmm;
 468
 469                 if (!S_ISREG(md->body->mode)) {
 470                         CDEBUG(D_INFO,
 471                                "OBD_MD_FLEASIZE set, should be a regular file, but is not\n");
 472                         rc = -EPROTO;
 473                         goto out;
 474                 }
 475
 476                 if (md->body->eadatasize == 0) {
 477                         CDEBUG(D_INFO,
 478                                "OBD_MD_FLEASIZE set, but eadatasize 0\n");
 479                         rc = -EPROTO;
 480                         goto out;
 481                 }
 482                 lmmsize = md->body->eadatasize;
 483                 lmm = req_capsule_server_sized_get(pill, &RMF_MDT_MD, lmmsize);
 484                 if (!lmm) {
 485                         rc = -EPROTO;
 486                         goto out;
 487                 }
 488
 489                 rc = obd_unpackmd(dt_exp, &md->lsm, lmm, lmmsize);
 490                 if (rc < 0)
 491                         goto out;
 492
 493                 if (rc < sizeof(*md->lsm)) {
 494                         CDEBUG(D_INFO,
 495                                "lsm size too small: rc < sizeof (*md->lsm) (%d < %d)\n",
 496                                rc, (int)sizeof(*md->lsm));
 497                         rc = -EPROTO;
 498                         goto out;
 499                 }
 500
 501         } else if (md->body->valid & OBD_MD_FLDIREA) {
 502                 int lmvsize;
 503                 struct lov_mds_md *lmv;
 504
 505                 if (!S_ISDIR(md->body->mode)) {
 506                         CDEBUG(D_INFO,
 507                                "OBD_MD_FLDIREA set, should be a directory, but is not\n");
 508                         rc = -EPROTO;
 509                         goto out;
 510                 }
 511
 512                 if (md->body->eadatasize == 0) {
 513                         CDEBUG(D_INFO,
 514                                "OBD_MD_FLDIREA is set, but eadatasize 0\n");
 515                         return -EPROTO;
 516                 }
 517                 if (md->body->valid & OBD_MD_MEA) {
 518                         lmvsize = md->body->eadatasize;
 519                         lmv = req_capsule_server_sized_get(pill, &RMF_MDT_MD,
 520                                                            lmvsize);
 521                         if (!lmv) {
 522                                 rc = -EPROTO;
 523                                 goto out;
 524                         }
 525
 526                         rc = obd_unpackmd(md_exp, (void *)&md->mea, lmv,
 527                                           lmvsize);
 528                         if (rc < 0)
 529                                 goto out;
 530
 531                         if (rc < sizeof(*md->mea)) {
 532                                 CDEBUG(D_INFO,
 533                                        "size too small: rc < sizeof(*md->mea) (%d < %d)\n",
 534                                         rc, (int)sizeof(*md->mea));
 535                                 rc = -EPROTO;
 536                                 goto out;
 537                         }
 538                 }
 539         }
 540         rc = 0;
 541
 542         if (md->body->valid & OBD_MD_FLRMTPERM) {
 543                 /* remote permission */
 544                 LASSERT(client_is_remote(exp));
 545                 md->remote_perm = req_capsule_server_swab_get(pill, &RMF_ACL,
 546                                                 lustre_swab_mdt_remote_perm);
 547                 if (!md->remote_perm) {
 548                         rc = -EPROTO;
 549                         goto out;
 550                 }
 551         } else if (md->body->valid & OBD_MD_FLACL) {
 552                 /* for ACL, it's possible that FLACL is set but aclsize is zero.
 553                  * only when aclsize != 0 there's an actual segment for ACL
 554                  * in reply buffer.
 555                  */
 556                 if (md->body->aclsize) {
 557                         rc = mdc_unpack_acl(req, md);
 558                         if (rc)
 559                                 goto out;
 560 #ifdef CONFIG_FS_POSIX_ACL
 561                 } else {
 562                         md->posix_acl = NULL;
 563 #endif
 564                 }
 565         }
 566
 567 out:
 568         if (rc) {
 569 #ifdef CONFIG_FS_POSIX_ACL
 570                 posix_acl_release(md->posix_acl);
 571 #endif
 572                 if (md->lsm)
 573                         obd_free_memmd(dt_exp, &md->lsm);
 574         }
 575         return rc;
 576 }
 577
 578 static int mdc_free_lustre_md(struct obd_export *exp, struct lustre_md *md)
 579 {
 580         return 0;
 581 }
 582
 583 /**
 584  * Handles both OPEN and SETATTR RPCs for OPEN-CLOSE and SETATTR-DONE_WRITING
 585  * RPC chains.
 586  */
 587 void mdc_replay_open(struct ptlrpc_request *req)
 588 {
 589         struct md_open_data *mod = req->rq_cb_data;
 590         struct ptlrpc_request *close_req;
 591         struct obd_client_handle *och;
 592         struct lustre_handle old;
 593         struct mdt_body *body;
 594
 595         if (mod == NULL) {
 596                 DEBUG_REQ(D_ERROR, req,
 597                           "Can't properly replay without open data.");
 598                 return;
 599         }
 600
 601         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 602         LASSERT(body != NULL);
 603
 604         och = mod->mod_och;
 605         if (och != NULL) {
 606                 struct lustre_handle *file_fh;
 607
 608                 LASSERT(och->och_magic == OBD_CLIENT_HANDLE_MAGIC);
 609
 610                 file_fh = &och->och_fh;
 611                 CDEBUG(D_HA, "updating handle from %#llx to %#llx\n",
 612                        file_fh->cookie, body->handle.cookie);
 613                 old = *file_fh;
 614                 *file_fh = body->handle;
 615         }
 616         close_req = mod->mod_close_req;
 617         if (close_req != NULL) {
 618                 __u32 opc = lustre_msg_get_opc(close_req->rq_reqmsg);
 619                 struct mdt_ioepoch *epoch;
 620
 621                 LASSERT(opc == MDS_CLOSE || opc == MDS_DONE_WRITING);
 622                 epoch = req_capsule_client_get(&close_req->rq_pill,
 623                                                &RMF_MDT_EPOCH);
 624                 LASSERT(epoch);
 625
 626                 if (och != NULL)
 627                         LASSERT(!memcmp(&old, &epoch->handle, sizeof(old)));
 628                 DEBUG_REQ(D_HA, close_req, "updating close body with new fh");
 629                 epoch->handle = body->handle;
 630         }
 631 }
 632
 633 void mdc_commit_open(struct ptlrpc_request *req)
 634 {
 635         struct md_open_data *mod = req->rq_cb_data;
 636
 637         if (mod == NULL)
 638                 return;
 639
 640         /**
 641          * No need to touch md_open_data::mod_och, it holds a reference on
 642          * \var mod and will zero references to each other, \var mod will be
 643          * freed after that when md_open_data::mod_och will put the reference.
 644          */
 645
 646         /**
 647          * Do not let open request to disappear as it still may be needed
 648          * for close rpc to happen (it may happen on evict only, otherwise
 649          * ptlrpc_request::rq_replay does not let mdc_commit_open() to be
 650          * called), just mark this rpc as committed to distinguish these 2
 651          * cases, see mdc_close() for details. The open request reference will
 652          * be put along with freeing \var mod.
 653          */
 654         ptlrpc_request_addref(req);
 655         spin_lock(&req->rq_lock);
 656         req->rq_committed = 1;
 657         spin_unlock(&req->rq_lock);
 658         req->rq_cb_data = NULL;
 659         obd_mod_put(mod);
 660 }
 661
 662 int mdc_set_open_replay_data(struct obd_export *exp,
 663                              struct obd_client_handle *och,
 664                              struct lookup_intent *it)
 665 {
 666         struct md_open_data   *mod;
 667         struct mdt_rec_create *rec;
 668         struct mdt_body       *body;
 669         struct ptlrpc_request *open_req = it->d.lustre.it_data;
 670         struct obd_import     *imp = open_req->rq_import;
 671
 672         if (!open_req->rq_replay)
 673                 return 0;
 674
 675         rec = req_capsule_client_get(&open_req->rq_pill, &RMF_REC_REINT);
 676         body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
 677         LASSERT(rec != NULL);
 678         /* Incoming message in my byte order (it's been swabbed). */
 679         /* Outgoing messages always in my byte order. */
 680         LASSERT(body != NULL);
 681
 682         /* Only if the import is replayable, we set replay_open data */
 683         if (och && imp->imp_replayable) {
 684                 mod = obd_mod_alloc();
 685                 if (mod == NULL) {
 686                         DEBUG_REQ(D_ERROR, open_req,
 687                                   "Can't allocate md_open_data");
 688                         return 0;
 689                 }
 690
 691                 /**
 692                  * Take a reference on \var mod, to be freed on mdc_close().
 693                  * It protects \var mod from being freed on eviction (commit
 694                  * callback is called despite rq_replay flag).
 695                  * Another reference for \var och.
 696                  */
 697                 obd_mod_get(mod);
 698                 obd_mod_get(mod);
 699
 700                 spin_lock(&open_req->rq_lock);
 701                 och->och_mod = mod;
 702                 mod->mod_och = och;
 703                 mod->mod_is_create = it_disposition(it, DISP_OPEN_CREATE) ||
 704                                      it_disposition(it, DISP_OPEN_STRIPE);
 705                 mod->mod_open_req = open_req;
 706                 open_req->rq_cb_data = mod;
 707                 open_req->rq_commit_cb = mdc_commit_open;
 708                 spin_unlock(&open_req->rq_lock);
 709         }
 710
 711         rec->cr_fid2 = body->fid1;
 712         rec->cr_ioepoch = body->ioepoch;
 713         rec->cr_old_handle.cookie = body->handle.cookie;
 714         open_req->rq_replay_cb = mdc_replay_open;
 715         if (!fid_is_sane(&body->fid1)) {
 716                 DEBUG_REQ(D_ERROR, open_req,
 717                           "Saving replay request with insane fid");
 718                 LBUG();
 719         }
 720
 721         DEBUG_REQ(D_RPCTRACE, open_req, "Set up open replay data");
 722         return 0;
 723 }
 724
 725 static void mdc_free_open(struct md_open_data *mod)
 726 {
 727         int committed = 0;
 728
 729         if (mod->mod_is_create == 0 &&
 730             imp_connect_disp_stripe(mod->mod_open_req->rq_import))
 731                 committed = 1;
 732
 733         LASSERT(mod->mod_open_req->rq_replay == 0);
 734
 735         DEBUG_REQ(D_RPCTRACE, mod->mod_open_req, "free open request\n");
 736
 737         ptlrpc_request_committed(mod->mod_open_req, committed);
 738         if (mod->mod_close_req)
 739                 ptlrpc_request_committed(mod->mod_close_req, committed);
 740 }
 741
 742 static int mdc_clear_open_replay_data(struct obd_export *exp,
 743                                       struct obd_client_handle *och)
 744 {
 745         struct md_open_data *mod = och->och_mod;
 746
 747         /**
 748          * It is possible to not have \var mod in a case of eviction between
 749          * lookup and ll_file_open().
 750          **/
 751         if (mod == NULL)
 752                 return 0;
 753
 754         LASSERT(mod != LP_POISON);
 755         LASSERT(mod->mod_open_req != NULL);
 756         mdc_free_open(mod);
 757
 758         mod->mod_och = NULL;
 759         och->och_mod = NULL;
 760         obd_mod_put(mod);
 761
 762         return 0;
 763 }
 764
 765 /* Prepares the request for the replay by the given reply */
 766 static void mdc_close_handle_reply(struct ptlrpc_request *req,
 767                                    struct md_op_data *op_data, int rc) {
 768         struct mdt_body  *repbody;
 769         struct mdt_ioepoch *epoch;
 770
 771         if (req && rc == -EAGAIN) {
 772                 repbody = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 773                 epoch = req_capsule_client_get(&req->rq_pill, &RMF_MDT_EPOCH);
 774
 775                 epoch->flags |= MF_SOM_AU;
 776                 if (repbody->valid & OBD_MD_FLGETATTRLOCK)
 777                         op_data->op_flags |= MF_GETATTR_LOCK;
 778         }
 779 }
 780
 781 static int mdc_close(struct obd_export *exp, struct md_op_data *op_data,
 782                      struct md_open_data *mod, struct ptlrpc_request **request)
 783 {
 784         struct obd_device     *obd = class_exp2obd(exp);
 785         struct ptlrpc_request *req;
 786         struct req_format     *req_fmt;
 787         int                    rc;
 788         int                    saved_rc = 0;
 789
 790         req_fmt = &RQF_MDS_CLOSE;
 791         if (op_data->op_bias & MDS_HSM_RELEASE) {
 792                 req_fmt = &RQF_MDS_RELEASE_CLOSE;
 793
 794                 /* allocate a FID for volatile file */
 795                 rc = mdc_fid_alloc(exp, &op_data->op_fid2, op_data);
 796                 if (rc < 0) {
 797                         CERROR("%s: "DFID" failed to allocate FID: %d\n",
 798                                obd->obd_name, PFID(&op_data->op_fid1), rc);
 799                         /* save the errcode and proceed to close */
 800                         saved_rc = rc;
 801                 }
 802         }
 803
 804         *request = NULL;
 805         req = ptlrpc_request_alloc(class_exp2cliimp(exp), req_fmt);
 806         if (req == NULL)
 807                 return -ENOMEM;
 808
 809         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_CLOSE);
 810         if (rc) {
 811                 ptlrpc_request_free(req);
 812                 return rc;
 813         }
 814
 815         /* To avoid a livelock (bug 7034), we need to send CLOSE RPCs to a
 816          * portal whose threads are not taking any DLM locks and are therefore
 817          * always progressing */
 818         req->rq_request_portal = MDS_READPAGE_PORTAL;
 819         ptlrpc_at_set_req_timeout(req);
 820
 821         /* Ensure that this close's handle is fixed up during replay. */
 822         if (likely(mod != NULL)) {
 823                 LASSERTF(mod->mod_open_req != NULL &&
 824                          mod->mod_open_req->rq_type != LI_POISON,
 825                          "POISONED open %p!\n", mod->mod_open_req);
 826
 827                 mod->mod_close_req = req;
 828
 829                 DEBUG_REQ(D_HA, mod->mod_open_req, "matched open");
 830                 /* We no longer want to preserve this open for replay even
 831                  * though the open was committed. b=3632, b=3633 */
 832                 spin_lock(&mod->mod_open_req->rq_lock);
 833                 mod->mod_open_req->rq_replay = 0;
 834                 spin_unlock(&mod->mod_open_req->rq_lock);
 835         } else {
 836                  CDEBUG(D_HA,
 837                         "couldn't find open req; expecting close error\n");
 838         }
 839
 840         mdc_close_pack(req, op_data);
 841
 842         req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
 843                              obd->u.cli.cl_default_mds_easize);
 844         req_capsule_set_size(&req->rq_pill, &RMF_LOGCOOKIES, RCL_SERVER,
 845                              obd->u.cli.cl_default_mds_cookiesize);
 846
 847         ptlrpc_request_set_replen(req);
 848
 849         mdc_get_rpc_lock(obd->u.cli.cl_close_lock, NULL);
 850         rc = ptlrpc_queue_wait(req);
 851         mdc_put_rpc_lock(obd->u.cli.cl_close_lock, NULL);
 852
 853         if (req->rq_repmsg == NULL) {
 854                 CDEBUG(D_RPCTRACE, "request failed to send: %p, %d\n", req,
 855                        req->rq_status);
 856                 if (rc == 0)
 857                         rc = req->rq_status ?: -EIO;
 858         } else if (rc == 0 || rc == -EAGAIN) {
 859                 struct mdt_body *body;
 860
 861                 rc = lustre_msg_get_status(req->rq_repmsg);
 862                 if (lustre_msg_get_type(req->rq_repmsg) == PTL_RPC_MSG_ERR) {
 863                         DEBUG_REQ(D_ERROR, req,
 864                                   "type == PTL_RPC_MSG_ERR, err = %d", rc);
 865                         if (rc > 0)
 866                                 rc = -rc;
 867                 }
 868                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 869                 if (body == NULL)
 870                         rc = -EPROTO;
 871         } else if (rc == -ESTALE) {
 872                 /**
 873                  * it can be allowed error after 3633 if open was committed and
 874                  * server failed before close was sent. Let's check if mod
 875                  * exists and return no error in that case
 876                  */
 877                 if (mod) {
 878                         DEBUG_REQ(D_HA, req, "Reset ESTALE = %d", rc);
 879                         LASSERT(mod->mod_open_req != NULL);
 880                         if (mod->mod_open_req->rq_committed)
 881                                 rc = 0;
 882                 }
 883         }
 884
 885         if (mod) {
 886                 if (rc != 0)
 887                         mod->mod_close_req = NULL;
 888                 /* Since now, mod is accessed through open_req only,
 889                  * thus close req does not keep a reference on mod anymore. */
 890                 obd_mod_put(mod);
 891         }
 892         *request = req;
 893         mdc_close_handle_reply(req, op_data, rc);
 894         return rc < 0 ? rc : saved_rc;
 895 }
 896
 897 static int mdc_done_writing(struct obd_export *exp, struct md_op_data *op_data,
 898                             struct md_open_data *mod)
 899 {
 900         struct obd_device     *obd = class_exp2obd(exp);
 901         struct ptlrpc_request *req;
 902         int                 rc;
 903
 904         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
 905                                    &RQF_MDS_DONE_WRITING);
 906         if (req == NULL)
 907                 return -ENOMEM;
 908
 909         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_DONE_WRITING);
 910         if (rc) {
 911                 ptlrpc_request_free(req);
 912                 return rc;
 913         }
 914
 915         if (mod != NULL) {
 916                 LASSERTF(mod->mod_open_req != NULL &&
 917                          mod->mod_open_req->rq_type != LI_POISON,
 918                          "POISONED setattr %p!\n", mod->mod_open_req);
 919
 920                 mod->mod_close_req = req;
 921                 DEBUG_REQ(D_HA, mod->mod_open_req, "matched setattr");
 922                 /* We no longer want to preserve this setattr for replay even
 923                  * though the open was committed. b=3632, b=3633 */
 924                 spin_lock(&mod->mod_open_req->rq_lock);
 925                 mod->mod_open_req->rq_replay = 0;
 926                 spin_unlock(&mod->mod_open_req->rq_lock);
 927         }
 928
 929         mdc_close_pack(req, op_data);
 930         ptlrpc_request_set_replen(req);
 931
 932         mdc_get_rpc_lock(obd->u.cli.cl_close_lock, NULL);
 933         rc = ptlrpc_queue_wait(req);
 934         mdc_put_rpc_lock(obd->u.cli.cl_close_lock, NULL);
 935
 936         if (rc == -ESTALE) {
 937                 /**
 938                  * it can be allowed error after 3633 if open or setattr were
 939                  * committed and server failed before close was sent.
 940                  * Let's check if mod exists and return no error in that case
 941                  */
 942                 if (mod) {
 943                         LASSERT(mod->mod_open_req != NULL);
 944                         if (mod->mod_open_req->rq_committed)
 945                                 rc = 0;
 946                 }
 947         }
 948
 949         if (mod) {
 950                 if (rc != 0)
 951                         mod->mod_close_req = NULL;
 952                 LASSERT(mod->mod_open_req != NULL);
 953                 mdc_free_open(mod);
 954
 955                 /* Since now, mod is accessed through setattr req only,
 956                  * thus DW req does not keep a reference on mod anymore. */
 957                 obd_mod_put(mod);
 958         }
 959
 960         mdc_close_handle_reply(req, op_data, rc);
 961         ptlrpc_req_finished(req);
 962         return rc;
 963 }
 964
 965 static int mdc_readpage(struct obd_export *exp, struct md_op_data *op_data,
 966                         struct page **pages, struct ptlrpc_request **request)
 967 {
 968         struct ptlrpc_request   *req;
 969         struct ptlrpc_bulk_desc *desc;
 970         int                   i;
 971         wait_queue_head_t             waitq;
 972         int                   resends = 0;
 973         struct l_wait_info       lwi;
 974         int                   rc;
 975
 976         *request = NULL;
 977         init_waitqueue_head(&waitq);
 978
 979 restart_bulk:
 980         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_READPAGE);
 981         if (req == NULL)
 982                 return -ENOMEM;
 983
 984         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_READPAGE);
 985         if (rc) {
 986                 ptlrpc_request_free(req);
 987                 return rc;
 988         }
 989
 990         req->rq_request_portal = MDS_READPAGE_PORTAL;
 991         ptlrpc_at_set_req_timeout(req);
 992
 993         desc = ptlrpc_prep_bulk_imp(req, op_data->op_npages, 1, BULK_PUT_SINK,
 994                                     MDS_BULK_PORTAL);
 995         if (desc == NULL) {
 996                 ptlrpc_request_free(req);
 997                 return -ENOMEM;
 998         }
 999
1000         /* NB req now owns desc and will free it when it gets freed */
1001         for (i = 0; i < op_data->op_npages; i++)
1002                 ptlrpc_prep_bulk_page_pin(desc, pages[i], 0, PAGE_CACHE_SIZE);
1003
1004         mdc_readdir_pack(req, op_data->op_offset,
1005                          PAGE_CACHE_SIZE * op_data->op_npages,
1006                          &op_data->op_fid1);
1007
1008         ptlrpc_request_set_replen(req);
1009         rc = ptlrpc_queue_wait(req);
1010         if (rc) {
1011                 ptlrpc_req_finished(req);
1012                 if (rc != -ETIMEDOUT)
1013                         return rc;
1014
1015                 resends++;
1016                 if (!client_should_resend(resends, &exp->exp_obd->u.cli)) {
1017                         CERROR("too many resend retries, returning error\n");
1018                         return -EIO;
1019                 }
1020                 lwi = LWI_TIMEOUT_INTR(cfs_time_seconds(resends),
1021                                        NULL, NULL, NULL);
1022                 l_wait_event(waitq, 0, &lwi);
1023
1024                 goto restart_bulk;
1025         }
1026
1027         rc = sptlrpc_cli_unwrap_bulk_read(req, req->rq_bulk,
1028                                           req->rq_bulk->bd_nob_transferred);
1029         if (rc < 0) {
1030                 ptlrpc_req_finished(req);
1031                 return rc;
1032         }
1033
1034         if (req->rq_bulk->bd_nob_transferred & ~LU_PAGE_MASK) {
1035                 CERROR("Unexpected # bytes transferred: %d (%ld expected)\n",
1036                         req->rq_bulk->bd_nob_transferred,
1037                         PAGE_CACHE_SIZE * op_data->op_npages);
1038                 ptlrpc_req_finished(req);
1039                 return -EPROTO;
1040         }
1041
1042         *request = req;
1043         return 0;
1044 }
1045
1046 static int mdc_statfs(const struct lu_env *env,
1047                       struct obd_export *exp, struct obd_statfs *osfs,
1048                       __u64 max_age, __u32 flags)
1049 {
1050         struct obd_device     *obd = class_exp2obd(exp);
1051         struct ptlrpc_request *req;
1052         struct obd_statfs     *msfs;
1053         struct obd_import     *imp = NULL;
1054         int                 rc;
1055
1056         /*
1057          * Since the request might also come from lprocfs, so we need
1058          * sync this with client_disconnect_export Bug15684
1059          */
1060         down_read(&obd->u.cli.cl_sem);
1061         if (obd->u.cli.cl_import)
1062                 imp = class_import_get(obd->u.cli.cl_import);
1063         up_read(&obd->u.cli.cl_sem);
1064         if (!imp)
1065                 return -ENODEV;
1066
1067         req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_STATFS,
1068                                         LUSTRE_MDS_VERSION, MDS_STATFS);
1069         if (req == NULL) {
1070                 rc = -ENOMEM;
1071                 goto output;
1072         }
1073
1074         ptlrpc_request_set_replen(req);
1075
1076         if (flags & OBD_STATFS_NODELAY) {
1077                 /* procfs requests not want stay in wait for avoid deadlock */
1078                 req->rq_no_resend = 1;
1079                 req->rq_no_delay = 1;
1080         }
1081
1082         rc = ptlrpc_queue_wait(req);
1083         if (rc) {
1084                 /* check connection error first */
1085                 if (imp->imp_connect_error)
1086                         rc = imp->imp_connect_error;
1087                 goto out;
1088         }
1089
1090         msfs = req_capsule_server_get(&req->rq_pill, &RMF_OBD_STATFS);
1091         if (msfs == NULL) {
1092                 rc = -EPROTO;
1093                 goto out;
1094         }
1095
1096         *osfs = *msfs;
1097 out:
1098         ptlrpc_req_finished(req);
1099 output:
1100         class_import_put(imp);
1101         return rc;
1102 }
1103
1104 static int mdc_ioc_fid2path(struct obd_export *exp, struct getinfo_fid2path *gf)
1105 {
1106         __u32 keylen, vallen;
1107         void *key;
1108         int rc;
1109
1110         if (gf->gf_pathlen > PATH_MAX)
1111                 return -ENAMETOOLONG;
1112         if (gf->gf_pathlen < 2)
1113                 return -EOVERFLOW;
1114
1115         /* Key is KEY_FID2PATH + getinfo_fid2path description */
1116         keylen = cfs_size_round(sizeof(KEY_FID2PATH)) + sizeof(*gf);
1117         key = kzalloc(keylen, GFP_NOFS);
1118         if (!key)
1119                 return -ENOMEM;
1120         memcpy(key, KEY_FID2PATH, sizeof(KEY_FID2PATH));
1121         memcpy(key + cfs_size_round(sizeof(KEY_FID2PATH)), gf, sizeof(*gf));
1122
1123         CDEBUG(D_IOCTL, "path get "DFID" from %llu #%d\n",
1124                PFID(&gf->gf_fid), gf->gf_recno, gf->gf_linkno);
1125
1126         if (!fid_is_sane(&gf->gf_fid)) {
1127                 rc = -EINVAL;
1128                 goto out;
1129         }
1130
1131         /* Val is struct getinfo_fid2path result plus path */
1132         vallen = sizeof(*gf) + gf->gf_pathlen;
1133
1134         rc = obd_get_info(NULL, exp, keylen, key, &vallen, gf, NULL);
1135         if (rc != 0 && rc != -EREMOTE)
1136                 goto out;
1137
1138         if (vallen <= sizeof(*gf)) {
1139                 rc = -EPROTO;
1140                 goto out;
1141         } else if (vallen > sizeof(*gf) + gf->gf_pathlen) {
1142                 rc = -EOVERFLOW;
1143                 goto out;
1144         }
1145
1146         CDEBUG(D_IOCTL, "path get "DFID" from %llu #%d\n%s\n",
1147                PFID(&gf->gf_fid), gf->gf_recno, gf->gf_linkno, gf->gf_path);
1148
1149 out:
1150         kfree(key);
1151         return rc;
1152 }
1153
1154 static int mdc_ioc_hsm_progress(struct obd_export *exp,
1155                                 struct hsm_progress_kernel *hpk)
1156 {
1157         struct obd_import               *imp = class_exp2cliimp(exp);
1158         struct hsm_progress_kernel      *req_hpk;
1159         struct ptlrpc_request           *req;
1160         int                              rc;
1161
1162         req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_HSM_PROGRESS,
1163                                         LUSTRE_MDS_VERSION, MDS_HSM_PROGRESS);
1164         if (req == NULL) {
1165                 rc = -ENOMEM;
1166                 goto out;
1167         }
1168
1169         mdc_pack_body(req, NULL, OBD_MD_FLRMTPERM, 0, 0, 0);
1170
1171         /* Copy hsm_progress struct */
1172         req_hpk = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_PROGRESS);
1173         if (req_hpk == NULL) {
1174                 rc = -EPROTO;
1175                 goto out;
1176         }
1177
1178         *req_hpk = *hpk;
1179         req_hpk->hpk_errval = lustre_errno_hton(hpk->hpk_errval);
1180
1181         ptlrpc_request_set_replen(req);
1182
1183         rc = mdc_queue_wait(req);
1184         goto out;
1185 out:
1186         ptlrpc_req_finished(req);
1187         return rc;
1188 }
1189
1190 static int mdc_ioc_hsm_ct_register(struct obd_import *imp, __u32 archives)
1191 {
1192         __u32                   *archive_mask;
1193         struct ptlrpc_request   *req;
1194         int                      rc;
1195
1196         req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_HSM_CT_REGISTER,
1197                                         LUSTRE_MDS_VERSION,
1198                                         MDS_HSM_CT_REGISTER);
1199         if (req == NULL) {
1200                 rc = -ENOMEM;
1201                 goto out;
1202         }
1203
1204         mdc_pack_body(req, NULL, OBD_MD_FLRMTPERM, 0, 0, 0);
1205
1206         /* Copy hsm_progress struct */
1207         archive_mask = req_capsule_client_get(&req->rq_pill,
1208                                               &RMF_MDS_HSM_ARCHIVE);
1209         if (archive_mask == NULL) {
1210                 rc = -EPROTO;
1211                 goto out;
1212         }
1213
1214         *archive_mask = archives;
1215
1216         ptlrpc_request_set_replen(req);
1217
1218         rc = mdc_queue_wait(req);
1219         goto out;
1220 out:
1221         ptlrpc_req_finished(req);
1222         return rc;
1223 }
1224
1225 static int mdc_ioc_hsm_current_action(struct obd_export *exp,
1226                                       struct md_op_data *op_data)
1227 {
1228         struct hsm_current_action       *hca = op_data->op_data;
1229         struct hsm_current_action       *req_hca;
1230         struct ptlrpc_request           *req;
1231         int                              rc;
1232
1233         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1234                                    &RQF_MDS_HSM_ACTION);
1235         if (req == NULL)
1236                 return -ENOMEM;
1237
1238         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_ACTION);
1239         if (rc) {
1240                 ptlrpc_request_free(req);
1241                 return rc;
1242         }
1243
1244         mdc_pack_body(req, &op_data->op_fid1, OBD_MD_FLRMTPERM, 0,
1245                       op_data->op_suppgids[0], 0);
1246
1247         ptlrpc_request_set_replen(req);
1248
1249         rc = mdc_queue_wait(req);
1250         if (rc)
1251                 goto out;
1252
1253         req_hca = req_capsule_server_get(&req->rq_pill,
1254                                          &RMF_MDS_HSM_CURRENT_ACTION);
1255         if (req_hca == NULL) {
1256                 rc = -EPROTO;
1257                 goto out;
1258         }
1259
1260         *hca = *req_hca;
1261
1262 out:
1263         ptlrpc_req_finished(req);
1264         return rc;
1265 }
1266
1267 static int mdc_ioc_hsm_ct_unregister(struct obd_import *imp)
1268 {
1269         struct ptlrpc_request   *req;
1270         int                      rc;
1271
1272         req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_HSM_CT_UNREGISTER,
1273                                         LUSTRE_MDS_VERSION,
1274                                         MDS_HSM_CT_UNREGISTER);
1275         if (req == NULL) {
1276                 rc = -ENOMEM;
1277                 goto out;
1278         }
1279
1280         mdc_pack_body(req, NULL, OBD_MD_FLRMTPERM, 0, 0, 0);
1281
1282         ptlrpc_request_set_replen(req);
1283
1284         rc = mdc_queue_wait(req);
1285         goto out;
1286 out:
1287         ptlrpc_req_finished(req);
1288         return rc;
1289 }
1290
1291 static int mdc_ioc_hsm_state_get(struct obd_export *exp,
1292                                  struct md_op_data *op_data)
1293 {
1294         struct hsm_user_state   *hus = op_data->op_data;
1295         struct hsm_user_state   *req_hus;
1296         struct ptlrpc_request   *req;
1297         int                      rc;
1298
1299         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1300                                    &RQF_MDS_HSM_STATE_GET);
1301         if (req == NULL)
1302                 return -ENOMEM;
1303
1304         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_STATE_GET);
1305         if (rc != 0) {
1306                 ptlrpc_request_free(req);
1307                 return rc;
1308         }
1309
1310         mdc_pack_body(req, &op_data->op_fid1, OBD_MD_FLRMTPERM, 0,
1311                       op_data->op_suppgids[0], 0);
1312
1313         ptlrpc_request_set_replen(req);
1314
1315         rc = mdc_queue_wait(req);
1316         if (rc)
1317                 goto out;
1318
1319         req_hus = req_capsule_server_get(&req->rq_pill, &RMF_HSM_USER_STATE);
1320         if (req_hus == NULL) {
1321                 rc = -EPROTO;
1322                 goto out;
1323         }
1324
1325         *hus = *req_hus;
1326
1327 out:
1328         ptlrpc_req_finished(req);
1329         return rc;
1330 }
1331
1332 static int mdc_ioc_hsm_state_set(struct obd_export *exp,
1333                                  struct md_op_data *op_data)
1334 {
1335         struct hsm_state_set    *hss = op_data->op_data;
1336         struct hsm_state_set    *req_hss;
1337         struct ptlrpc_request   *req;
1338         int                      rc;
1339
1340         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1341                                    &RQF_MDS_HSM_STATE_SET);
1342         if (req == NULL)
1343                 return -ENOMEM;
1344
1345         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_STATE_SET);
1346         if (rc) {
1347                 ptlrpc_request_free(req);
1348                 return rc;
1349         }
1350
1351         mdc_pack_body(req, &op_data->op_fid1, OBD_MD_FLRMTPERM, 0,
1352                       op_data->op_suppgids[0], 0);
1353
1354         /* Copy states */
1355         req_hss = req_capsule_client_get(&req->rq_pill, &RMF_HSM_STATE_SET);
1356         if (req_hss == NULL) {
1357                 rc = -EPROTO;
1358                 goto out;
1359         }
1360         *req_hss = *hss;
1361
1362         ptlrpc_request_set_replen(req);
1363
1364         rc = mdc_queue_wait(req);
1365         goto out;
1366
1367 out:
1368         ptlrpc_req_finished(req);
1369         return rc;
1370 }
1371
1372 static int mdc_ioc_hsm_request(struct obd_export *exp,
1373                                struct hsm_user_request *hur)
1374 {
1375         struct obd_import       *imp = class_exp2cliimp(exp);
1376         struct ptlrpc_request   *req;
1377         struct hsm_request      *req_hr;
1378         struct hsm_user_item    *req_hui;
1379         char                    *req_opaque;
1380         int                      rc;
1381
1382         req = ptlrpc_request_alloc(imp, &RQF_MDS_HSM_REQUEST);
1383         if (req == NULL) {
1384                 rc = -ENOMEM;
1385                 goto out;
1386         }
1387
1388         req_capsule_set_size(&req->rq_pill, &RMF_MDS_HSM_USER_ITEM, RCL_CLIENT,
1389                              hur->hur_request.hr_itemcount
1390                              * sizeof(struct hsm_user_item));
1391         req_capsule_set_size(&req->rq_pill, &RMF_GENERIC_DATA, RCL_CLIENT,
1392                              hur->hur_request.hr_data_len);
1393
1394         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_REQUEST);
1395         if (rc) {
1396                 ptlrpc_request_free(req);
1397                 return rc;
1398         }
1399
1400         mdc_pack_body(req, NULL, OBD_MD_FLRMTPERM, 0, 0, 0);
1401
1402         /* Copy hsm_request struct */
1403         req_hr = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_REQUEST);
1404         if (req_hr == NULL) {
1405                 rc = -EPROTO;
1406                 goto out;
1407         }
1408         *req_hr = hur->hur_request;
1409
1410         /* Copy hsm_user_item structs */
1411         req_hui = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_USER_ITEM);
1412         if (req_hui == NULL) {
1413                 rc = -EPROTO;
1414                 goto out;
1415         }
1416         memcpy(req_hui, hur->hur_user_item,
1417                hur->hur_request.hr_itemcount * sizeof(struct hsm_user_item));
1418
1419         /* Copy opaque field */
1420         req_opaque = req_capsule_client_get(&req->rq_pill, &RMF_GENERIC_DATA);
1421         if (req_opaque == NULL) {
1422                 rc = -EPROTO;
1423                 goto out;
1424         }
1425         memcpy(req_opaque, hur_data(hur), hur->hur_request.hr_data_len);
1426
1427         ptlrpc_request_set_replen(req);
1428
1429         rc = mdc_queue_wait(req);
1430         goto out;
1431
1432 out:
1433         ptlrpc_req_finished(req);
1434         return rc;
1435 }
1436
1437 static struct kuc_hdr *changelog_kuc_hdr(char *buf, int len, int flags)
1438 {
1439         struct kuc_hdr *lh = (struct kuc_hdr *)buf;
1440
1441         LASSERT(len <= KUC_CHANGELOG_MSG_MAXSIZE);
1442
1443         lh->kuc_magic = KUC_MAGIC;
1444         lh->kuc_transport = KUC_TRANSPORT_CHANGELOG;
1445         lh->kuc_flags = flags;
1446         lh->kuc_msgtype = CL_RECORD;
1447         lh->kuc_msglen = len;
1448         return lh;
1449 }
1450
1451 #define D_CHANGELOG 0
1452
1453 struct changelog_show {
1454         __u64           cs_startrec;
1455         __u32           cs_flags;
1456         struct file     *cs_fp;
1457         char            *cs_buf;
1458         struct obd_device *cs_obd;
1459 };
1460
1461 static int changelog_kkuc_cb(const struct lu_env *env, struct llog_handle *llh,
1462                              struct llog_rec_hdr *hdr, void *data)
1463 {
1464         struct changelog_show *cs = data;
1465         struct llog_changelog_rec *rec = (struct llog_changelog_rec *)hdr;
1466         struct kuc_hdr *lh;
1467         int len, rc;
1468
1469         if (rec->cr_hdr.lrh_type != CHANGELOG_REC) {
1470                 rc = -EINVAL;
1471                 CERROR("%s: not a changelog rec %x/%d: rc = %d\n",
1472                        cs->cs_obd->obd_name, rec->cr_hdr.lrh_type,
1473                        rec->cr.cr_type, rc);
1474                 return rc;
1475         }
1476
1477         if (rec->cr.cr_index < cs->cs_startrec) {
1478                 /* Skip entries earlier than what we are interested in */
1479                 CDEBUG(D_CHANGELOG, "rec=%llu start=%llu\n",
1480                        rec->cr.cr_index, cs->cs_startrec);
1481                 return 0;
1482         }
1483
1484         CDEBUG(D_CHANGELOG, "%llu %02d%-5s %llu 0x%x t="DFID" p="DFID
1485                 " %.*s\n", rec->cr.cr_index, rec->cr.cr_type,
1486                 changelog_type2str(rec->cr.cr_type), rec->cr.cr_time,
1487                 rec->cr.cr_flags & CLF_FLAGMASK,
1488                 PFID(&rec->cr.cr_tfid), PFID(&rec->cr.cr_pfid),
1489                 rec->cr.cr_namelen, changelog_rec_name(&rec->cr));
1490
1491         len = sizeof(*lh) + changelog_rec_size(&rec->cr) + rec->cr.cr_namelen;
1492
1493         /* Set up the message */
1494         lh = changelog_kuc_hdr(cs->cs_buf, len, cs->cs_flags);
1495         memcpy(lh + 1, &rec->cr, len - sizeof(*lh));
1496
1497         rc = libcfs_kkuc_msg_put(cs->cs_fp, lh);
1498         CDEBUG(D_CHANGELOG, "kucmsg fp %p len %d rc %d\n", cs->cs_fp, len, rc);
1499
1500         return rc;
1501 }
1502
1503 static int mdc_changelog_send_thread(void *csdata)
1504 {
1505         struct changelog_show *cs = csdata;
1506         struct llog_ctxt *ctxt = NULL;
1507         struct llog_handle *llh = NULL;
1508         struct kuc_hdr *kuch;
1509         int rc;
1510
1511         CDEBUG(D_CHANGELOG, "changelog to fp=%p start %llu\n",
1512                cs->cs_fp, cs->cs_startrec);
1513
1514         cs->cs_buf = kzalloc(KUC_CHANGELOG_MSG_MAXSIZE, GFP_NOFS);
1515         if (!cs->cs_buf) {
1516                 rc = -ENOMEM;
1517                 goto out;
1518         }
1519
1520         /* Set up the remote catalog handle */
1521         ctxt = llog_get_context(cs->cs_obd, LLOG_CHANGELOG_REPL_CTXT);
1522         if (ctxt == NULL) {
1523                 rc = -ENOENT;
1524                 goto out;
1525         }
1526         rc = llog_open(NULL, ctxt, &llh, NULL, CHANGELOG_CATALOG,
1527                        LLOG_OPEN_EXISTS);
1528         if (rc) {
1529                 CERROR("%s: fail to open changelog catalog: rc = %d\n",
1530                        cs->cs_obd->obd_name, rc);
1531                 goto out;
1532         }
1533         rc = llog_init_handle(NULL, llh, LLOG_F_IS_CAT, NULL);
1534         if (rc) {
1535                 CERROR("llog_init_handle failed %d\n", rc);
1536                 goto out;
1537         }
1538
1539         rc = llog_cat_process(NULL, llh, changelog_kkuc_cb, cs, 0, 0);
1540
1541         /* Send EOF no matter what our result */
1542         kuch = changelog_kuc_hdr(cs->cs_buf, sizeof(*kuch), cs->cs_flags);
1543         if (kuch) {
1544                 kuch->kuc_msgtype = CL_EOF;
1545                 libcfs_kkuc_msg_put(cs->cs_fp, kuch);
1546         }
1547
1548 out:
1549         fput(cs->cs_fp);
1550         if (llh)
1551                 llog_cat_close(NULL, llh);
1552         if (ctxt)
1553                 llog_ctxt_put(ctxt);
1554         kfree(cs->cs_buf);
1555         kfree(cs);
1556         return rc;
1557 }
1558
1559 static int mdc_ioc_changelog_send(struct obd_device *obd,
1560                                   struct ioc_changelog *icc)
1561 {
1562         struct changelog_show *cs;
1563         int rc;
1564
1565         /* Freed in mdc_changelog_send_thread */
1566         cs = kzalloc(sizeof(*cs), GFP_NOFS);
1567         if (!cs)
1568                 return -ENOMEM;
1569
1570         cs->cs_obd = obd;
1571         cs->cs_startrec = icc->icc_recno;
1572         /* matching fput in mdc_changelog_send_thread */
1573         cs->cs_fp = fget(icc->icc_id);
1574         cs->cs_flags = icc->icc_flags;
1575
1576         /*
1577          * New thread because we should return to user app before
1578          * writing into our pipe
1579          */
1580         rc = PTR_ERR(kthread_run(mdc_changelog_send_thread, cs,
1581                                  "mdc_clg_send_thread"));
1582         if (!IS_ERR_VALUE(rc)) {
1583                 CDEBUG(D_CHANGELOG, "start changelog thread\n");
1584                 return 0;
1585         }
1586
1587         CERROR("Failed to start changelog thread: %d\n", rc);
1588         kfree(cs);
1589         return rc;
1590 }
1591
1592 static int mdc_ioc_hsm_ct_start(struct obd_export *exp,
1593                                 struct lustre_kernelcomm *lk);
1594
1595 static int mdc_quotacheck(struct obd_device *unused, struct obd_export *exp,
1596                           struct obd_quotactl *oqctl)
1597 {
1598         struct client_obd       *cli = &exp->exp_obd->u.cli;
1599         struct ptlrpc_request   *req;
1600         struct obd_quotactl     *body;
1601         int                   rc;
1602
1603         req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp),
1604                                         &RQF_MDS_QUOTACHECK, LUSTRE_MDS_VERSION,
1605                                         MDS_QUOTACHECK);
1606         if (req == NULL)
1607                 return -ENOMEM;
1608
1609         body = req_capsule_client_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
1610         *body = *oqctl;
1611
1612         ptlrpc_request_set_replen(req);
1613
1614         /* the next poll will find -ENODATA, that means quotacheck is
1615          * going on */
1616         cli->cl_qchk_stat = -ENODATA;
1617         rc = ptlrpc_queue_wait(req);
1618         if (rc)
1619                 cli->cl_qchk_stat = rc;
1620         ptlrpc_req_finished(req);
1621         return rc;
1622 }
1623
1624 static int mdc_quota_poll_check(struct obd_export *exp,
1625                                 struct if_quotacheck *qchk)
1626 {
1627         struct client_obd *cli = &exp->exp_obd->u.cli;
1628         int rc;
1629
1630         qchk->obd_uuid = cli->cl_target_uuid;
1631         memcpy(qchk->obd_type, LUSTRE_MDS_NAME, strlen(LUSTRE_MDS_NAME));
1632
1633         rc = cli->cl_qchk_stat;
1634         /* the client is not the previous one */
1635         if (rc == CL_NOT_QUOTACHECKED)
1636                 rc = -EINTR;
1637         return rc;
1638 }
1639
1640 static int mdc_quotactl(struct obd_device *unused, struct obd_export *exp,
1641                         struct obd_quotactl *oqctl)
1642 {
1643         struct ptlrpc_request   *req;
1644         struct obd_quotactl     *oqc;
1645         int                   rc;
1646
1647         req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp),
1648                                         &RQF_MDS_QUOTACTL, LUSTRE_MDS_VERSION,
1649                                         MDS_QUOTACTL);
1650         if (req == NULL)
1651                 return -ENOMEM;
1652
1653         oqc = req_capsule_client_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
1654         *oqc = *oqctl;
1655
1656         ptlrpc_request_set_replen(req);
1657         ptlrpc_at_set_req_timeout(req);
1658         req->rq_no_resend = 1;
1659
1660         rc = ptlrpc_queue_wait(req);
1661         if (rc)
1662                 CERROR("ptlrpc_queue_wait failed, rc: %d\n", rc);
1663
1664         if (req->rq_repmsg) {
1665                 oqc = req_capsule_server_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
1666                 if (oqc) {
1667                         *oqctl = *oqc;
1668                 } else if (!rc) {
1669                         CERROR("Can't unpack obd_quotactl\n");
1670                         rc = -EPROTO;
1671                 }
1672         } else if (!rc) {
1673                 CERROR("Can't unpack obd_quotactl\n");
1674                 rc = -EPROTO;
1675         }
1676         ptlrpc_req_finished(req);
1677
1678         return rc;
1679 }
1680
1681 static int mdc_ioc_swap_layouts(struct obd_export *exp,
1682                                 struct md_op_data *op_data)
1683 {
1684         LIST_HEAD(cancels);
1685         struct ptlrpc_request   *req;
1686         int                      rc, count;
1687         struct mdc_swap_layouts *msl, *payload;
1688
1689         msl = op_data->op_data;
1690
1691         /* When the MDT will get the MDS_SWAP_LAYOUTS RPC the
1692          * first thing it will do is to cancel the 2 layout
1693          * locks hold by this client.
1694          * So the client must cancel its layout locks on the 2 fids
1695          * with the request RPC to avoid extra RPC round trips
1696          */
1697         count = mdc_resource_get_unused(exp, &op_data->op_fid1, &cancels,
1698                                         LCK_CR, MDS_INODELOCK_LAYOUT);
1699         count += mdc_resource_get_unused(exp, &op_data->op_fid2, &cancels,
1700                                          LCK_CR, MDS_INODELOCK_LAYOUT);
1701
1702         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1703                                    &RQF_MDS_SWAP_LAYOUTS);
1704         if (req == NULL) {
1705                 ldlm_lock_list_put(&cancels, l_bl_ast, count);
1706                 return -ENOMEM;
1707         }
1708
1709         rc = mdc_prep_elc_req(exp, req, MDS_SWAP_LAYOUTS, &cancels, count);
1710         if (rc) {
1711                 ptlrpc_request_free(req);
1712                 return rc;
1713         }
1714
1715         mdc_swap_layouts_pack(req, op_data);
1716
1717         payload = req_capsule_client_get(&req->rq_pill, &RMF_SWAP_LAYOUTS);
1718         LASSERT(payload);
1719
1720         *payload = *msl;
1721
1722         ptlrpc_request_set_replen(req);
1723
1724         rc = ptlrpc_queue_wait(req);
1725
1726         ptlrpc_req_finished(req);
1727         return rc;
1728 }
1729
1730 static int mdc_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1731                          void *karg, void *uarg)
1732 {
1733         struct obd_device *obd = exp->exp_obd;
1734         struct obd_ioctl_data *data = karg;
1735         struct obd_import *imp = obd->u.cli.cl_import;
1736         int rc;
1737
1738         if (!try_module_get(THIS_MODULE)) {
1739                 CERROR("Can't get module. Is it alive?");
1740                 return -EINVAL;
1741         }
1742         switch (cmd) {
1743         case OBD_IOC_CHANGELOG_SEND:
1744                 rc = mdc_ioc_changelog_send(obd, karg);
1745                 goto out;
1746         case OBD_IOC_CHANGELOG_CLEAR: {
1747                 struct ioc_changelog *icc = karg;
1748                 struct changelog_setinfo cs = {
1749                         .cs_recno = icc->icc_recno,
1750                         .cs_id = icc->icc_id
1751                 };
1752
1753                 rc = obd_set_info_async(NULL, exp, strlen(KEY_CHANGELOG_CLEAR),
1754                                         KEY_CHANGELOG_CLEAR, sizeof(cs), &cs,
1755                                         NULL);
1756                 goto out;
1757         }
1758         case OBD_IOC_FID2PATH:
1759                 rc = mdc_ioc_fid2path(exp, karg);
1760                 goto out;
1761         case LL_IOC_HSM_CT_START:
1762                 rc = mdc_ioc_hsm_ct_start(exp, karg);
1763                 /* ignore if it was already registered on this MDS. */
1764                 if (rc == -EEXIST)
1765                         rc = 0;
1766                 goto out;
1767         case LL_IOC_HSM_PROGRESS:
1768                 rc = mdc_ioc_hsm_progress(exp, karg);
1769                 goto out;
1770         case LL_IOC_HSM_STATE_GET:
1771                 rc = mdc_ioc_hsm_state_get(exp, karg);
1772                 goto out;
1773         case LL_IOC_HSM_STATE_SET:
1774                 rc = mdc_ioc_hsm_state_set(exp, karg);
1775                 goto out;
1776         case LL_IOC_HSM_ACTION:
1777                 rc = mdc_ioc_hsm_current_action(exp, karg);
1778                 goto out;
1779         case LL_IOC_HSM_REQUEST:
1780                 rc = mdc_ioc_hsm_request(exp, karg);
1781                 goto out;
1782         case OBD_IOC_CLIENT_RECOVER:
1783                 rc = ptlrpc_recover_import(imp, data->ioc_inlbuf1, 0);
1784                 if (rc < 0)
1785                         goto out;
1786                 rc = 0;
1787                 goto out;
1788         case IOC_OSC_SET_ACTIVE:
1789                 rc = ptlrpc_set_import_active(imp, data->ioc_offset);
1790                 goto out;
1791         case OBD_IOC_POLL_QUOTACHECK:
1792                 rc = mdc_quota_poll_check(exp, (struct if_quotacheck *)karg);
1793                 goto out;
1794         case OBD_IOC_PING_TARGET:
1795                 rc = ptlrpc_obd_ping(obd);
1796                 goto out;
1797         /*
1798          * Normally IOC_OBD_STATFS, OBD_IOC_QUOTACTL iocontrol are handled by
1799          * LMV instead of MDC. But when the cluster is upgraded from 1.8,
1800          * there'd be no LMV layer thus we might be called here. Eventually
1801          * this code should be removed.
1802          * bz20731, LU-592.
1803          */
1804         case IOC_OBD_STATFS: {
1805                 struct obd_statfs stat_buf = {0};
1806
1807                 if (*((__u32 *) data->ioc_inlbuf2) != 0) {
1808                         rc = -ENODEV;
1809                         goto out;
1810                 }
1811
1812                 /* copy UUID */
1813                 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(obd),
1814                                  min_t(size_t, data->ioc_plen2,
1815                                                sizeof(struct obd_uuid)))) {
1816                         rc = -EFAULT;
1817                         goto out;
1818                 }
1819
1820                 rc = mdc_statfs(NULL, obd->obd_self_export, &stat_buf,
1821                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1822                                 0);
1823                 if (rc != 0)
1824                         goto out;
1825
1826                 if (copy_to_user(data->ioc_pbuf1, &stat_buf,
1827                                  min_t(size_t, data->ioc_plen1,
1828                                                sizeof(stat_buf)))) {
1829                         rc = -EFAULT;
1830                         goto out;
1831                 }
1832
1833                 rc = 0;
1834                 goto out;
1835         }
1836         case OBD_IOC_QUOTACTL: {
1837                 struct if_quotactl *qctl = karg;
1838                 struct obd_quotactl *oqctl;
1839
1840                 oqctl = kzalloc(sizeof(*oqctl), GFP_NOFS);
1841                 if (!oqctl) {
1842                         rc = -ENOMEM;
1843                         goto out;
1844                 }
1845
1846                 QCTL_COPY(oqctl, qctl);
1847                 rc = obd_quotactl(exp, oqctl);
1848                 if (rc == 0) {
1849                         QCTL_COPY(qctl, oqctl);
1850                         qctl->qc_valid = QC_MDTIDX;
1851                         qctl->obd_uuid = obd->u.cli.cl_target_uuid;
1852                 }
1853
1854                 kfree(oqctl);
1855                 goto out;
1856         }
1857         case LL_IOC_GET_CONNECT_FLAGS:
1858                 if (copy_to_user(uarg, exp_connect_flags_ptr(exp),
1859                                  sizeof(*exp_connect_flags_ptr(exp)))) {
1860                         rc = -EFAULT;
1861                         goto out;
1862                 }
1863
1864                 rc = 0;
1865                 goto out;
1866         case LL_IOC_LOV_SWAP_LAYOUTS:
1867                 rc = mdc_ioc_swap_layouts(exp, karg);
1868                 goto out;
1869         default:
1870                 CERROR("unrecognised ioctl: cmd = %#x\n", cmd);
1871                 rc = -ENOTTY;
1872                 goto out;
1873         }
1874 out:
1875         module_put(THIS_MODULE);
1876
1877         return rc;
1878 }
1879
1880 static int mdc_get_info_rpc(struct obd_export *exp,
1881                             u32 keylen, void *key,
1882                             int vallen, void *val)
1883 {
1884         struct obd_import      *imp = class_exp2cliimp(exp);
1885         struct ptlrpc_request  *req;
1886         char               *tmp;
1887         int                  rc = -EINVAL;
1888
1889         req = ptlrpc_request_alloc(imp, &RQF_MDS_GET_INFO);
1890         if (req == NULL)
1891                 return -ENOMEM;
1892
1893         req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_KEY,
1894                              RCL_CLIENT, keylen);
1895         req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_VALLEN,
1896                              RCL_CLIENT, sizeof(__u32));
1897
1898         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GET_INFO);
1899         if (rc) {
1900                 ptlrpc_request_free(req);
1901                 return rc;
1902         }
1903
1904         tmp = req_capsule_client_get(&req->rq_pill, &RMF_GETINFO_KEY);
1905         memcpy(tmp, key, keylen);
1906         tmp = req_capsule_client_get(&req->rq_pill, &RMF_GETINFO_VALLEN);
1907         memcpy(tmp, &vallen, sizeof(__u32));
1908
1909         req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_VAL,
1910                              RCL_SERVER, vallen);
1911         ptlrpc_request_set_replen(req);
1912
1913         rc = ptlrpc_queue_wait(req);
1914         /* -EREMOTE means the get_info result is partial, and it needs to
1915          * continue on another MDT, see fid2path part in lmv_iocontrol */
1916         if (rc == 0 || rc == -EREMOTE) {
1917                 tmp = req_capsule_server_get(&req->rq_pill, &RMF_GETINFO_VAL);
1918                 memcpy(val, tmp, vallen);
1919                 if (ptlrpc_rep_need_swab(req)) {
1920                         if (KEY_IS(KEY_FID2PATH))
1921                                 lustre_swab_fid2path(val);
1922                 }
1923         }
1924         ptlrpc_req_finished(req);
1925
1926         return rc;
1927 }
1928
1929 static void lustre_swab_hai(struct hsm_action_item *h)
1930 {
1931         __swab32s(&h->hai_len);
1932         __swab32s(&h->hai_action);
1933         lustre_swab_lu_fid(&h->hai_fid);
1934         lustre_swab_lu_fid(&h->hai_dfid);
1935         __swab64s(&h->hai_cookie);
1936         __swab64s(&h->hai_extent.offset);
1937         __swab64s(&h->hai_extent.length);
1938         __swab64s(&h->hai_gid);
1939 }
1940
1941 static void lustre_swab_hal(struct hsm_action_list *h)
1942 {
1943         struct hsm_action_item  *hai;
1944         int                      i;
1945
1946         __swab32s(&h->hal_version);
1947         __swab32s(&h->hal_count);
1948         __swab32s(&h->hal_archive_id);
1949         __swab64s(&h->hal_flags);
1950         hai = hai_zero(h);
1951         for (i = 0; i < h->hal_count; i++, hai = hai_next(hai))
1952                 lustre_swab_hai(hai);
1953 }
1954
1955 static void lustre_swab_kuch(struct kuc_hdr *l)
1956 {
1957         __swab16s(&l->kuc_magic);
1958         /* __u8 l->kuc_transport */
1959         __swab16s(&l->kuc_msgtype);
1960         __swab16s(&l->kuc_msglen);
1961 }
1962
1963 static int mdc_ioc_hsm_ct_start(struct obd_export *exp,
1964                                 struct lustre_kernelcomm *lk)
1965 {
1966         struct obd_import  *imp = class_exp2cliimp(exp);
1967         __u32               archive = lk->lk_data;
1968         int                 rc = 0;
1969
1970         if (lk->lk_group != KUC_GRP_HSM) {
1971                 CERROR("Bad copytool group %d\n", lk->lk_group);
1972                 return -EINVAL;
1973         }
1974
1975         CDEBUG(D_HSM, "CT start r%d w%d u%d g%d f%#x\n", lk->lk_rfd, lk->lk_wfd,
1976                lk->lk_uid, lk->lk_group, lk->lk_flags);
1977
1978         if (lk->lk_flags & LK_FLG_STOP) {
1979                 /* Unregister with the coordinator */
1980                 rc = mdc_ioc_hsm_ct_unregister(imp);
1981         } else {
1982                 rc = mdc_ioc_hsm_ct_register(imp, archive);
1983         }
1984
1985         return rc;
1986 }
1987
1988 /**
1989  * Send a message to any listening copytools
1990  * @param val KUC message (kuc_hdr + hsm_action_list)
1991  * @param len total length of message
1992  */
1993 static int mdc_hsm_copytool_send(int len, void *val)
1994 {
1995         struct kuc_hdr          *lh = (struct kuc_hdr *)val;
1996         struct hsm_action_list  *hal = (struct hsm_action_list *)(lh + 1);
1997
1998         if (len < sizeof(*lh) + sizeof(*hal)) {
1999                 CERROR("Short HSM message %d < %d\n", len,
2000                        (int) (sizeof(*lh) + sizeof(*hal)));
2001                 return -EPROTO;
2002         }
2003         if (lh->kuc_magic == __swab16(KUC_MAGIC)) {
2004                 lustre_swab_kuch(lh);
2005                 lustre_swab_hal(hal);
2006         } else if (lh->kuc_magic != KUC_MAGIC) {
2007                 CERROR("Bad magic %x!=%x\n", lh->kuc_magic, KUC_MAGIC);
2008                 return -EPROTO;
2009         }
2010
2011         CDEBUG(D_HSM,
2012                "Received message mg=%x t=%d m=%d l=%d actions=%d on %s\n",
2013                lh->kuc_magic, lh->kuc_transport, lh->kuc_msgtype,
2014                lh->kuc_msglen, hal->hal_count, hal->hal_fsname);
2015
2016         /* Broadcast to HSM listeners */
2017         return libcfs_kkuc_group_put(KUC_GRP_HSM, lh);
2018 }
2019
2020 /**
2021  * callback function passed to kuc for re-registering each HSM copytool
2022  * running on MDC, after MDT shutdown/recovery.
2023  * @param data archive id served by the copytool
2024  * @param cb_arg callback argument (obd_import)
2025  */
2026 static int mdc_hsm_ct_reregister(__u32 data, void *cb_arg)
2027 {
2028         struct obd_import       *imp = (struct obd_import *)cb_arg;
2029         __u32                    archive = data;
2030         int                      rc;
2031
2032         CDEBUG(D_HA, "recover copytool registration to MDT (archive=%#x)\n",
2033                archive);
2034         rc = mdc_ioc_hsm_ct_register(imp, archive);
2035
2036         /* ignore error if the copytool is already registered */
2037         return ((rc != 0) && (rc != -EEXIST)) ? rc : 0;
2038 }
2039
2040 static int mdc_set_info_async(const struct lu_env *env,
2041                               struct obd_export *exp,
2042                               u32 keylen, void *key,
2043                               u32 vallen, void *val,
2044                               struct ptlrpc_request_set *set)
2045 {
2046         struct obd_import       *imp = class_exp2cliimp(exp);
2047         int                      rc;
2048
2049         if (KEY_IS(KEY_READ_ONLY)) {
2050                 if (vallen != sizeof(int))
2051                         return -EINVAL;
2052
2053                 spin_lock(&imp->imp_lock);
2054                 if (*((int *)val)) {
2055                         imp->imp_connect_flags_orig |= OBD_CONNECT_RDONLY;
2056                         imp->imp_connect_data.ocd_connect_flags |=
2057                                                         OBD_CONNECT_RDONLY;
2058                 } else {
2059                         imp->imp_connect_flags_orig &= ~OBD_CONNECT_RDONLY;
2060                         imp->imp_connect_data.ocd_connect_flags &=
2061                                                         ~OBD_CONNECT_RDONLY;
2062                 }
2063                 spin_unlock(&imp->imp_lock);
2064
2065                 rc = do_set_info_async(imp, MDS_SET_INFO, LUSTRE_MDS_VERSION,
2066                                        keylen, key, vallen, val, set);
2067                 return rc;
2068         }
2069         if (KEY_IS(KEY_SPTLRPC_CONF)) {
2070                 sptlrpc_conf_client_adapt(exp->exp_obd);
2071                 return 0;
2072         }
2073         if (KEY_IS(KEY_FLUSH_CTX)) {
2074                 sptlrpc_import_flush_my_ctx(imp);
2075                 return 0;
2076         }
2077         if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
2078                 rc = do_set_info_async(imp, MDS_SET_INFO, LUSTRE_MDS_VERSION,
2079                                        keylen, key, vallen, val, set);
2080                 return rc;
2081         }
2082         if (KEY_IS(KEY_HSM_COPYTOOL_SEND)) {
2083                 rc = mdc_hsm_copytool_send(vallen, val);
2084                 return rc;
2085         }
2086
2087         CERROR("Unknown key %s\n", (char *)key);
2088         return -EINVAL;
2089 }
2090
2091 static int mdc_get_info(const struct lu_env *env, struct obd_export *exp,
2092                         __u32 keylen, void *key, __u32 *vallen, void *val,
2093                         struct lov_stripe_md *lsm)
2094 {
2095         int rc = -EINVAL;
2096
2097         if (KEY_IS(KEY_MAX_EASIZE)) {
2098                 int mdsize, *max_easize;
2099
2100                 if (*vallen != sizeof(int))
2101                         return -EINVAL;
2102                 mdsize = *(int *)val;
2103                 if (mdsize > exp->exp_obd->u.cli.cl_max_mds_easize)
2104                         exp->exp_obd->u.cli.cl_max_mds_easize = mdsize;
2105                 max_easize = val;
2106                 *max_easize = exp->exp_obd->u.cli.cl_max_mds_easize;
2107                 return 0;
2108         } else if (KEY_IS(KEY_DEFAULT_EASIZE)) {
2109                 int *default_easize;
2110
2111                 if (*vallen != sizeof(int))
2112                         return -EINVAL;
2113                 default_easize = val;
2114                 *default_easize = exp->exp_obd->u.cli.cl_default_mds_easize;
2115                 return 0;
2116         } else if (KEY_IS(KEY_CONN_DATA)) {
2117                 struct obd_import *imp = class_exp2cliimp(exp);
2118                 struct obd_connect_data *data = val;
2119
2120                 if (*vallen != sizeof(*data))
2121                         return -EINVAL;
2122
2123                 *data = imp->imp_connect_data;
2124                 return 0;
2125         } else if (KEY_IS(KEY_TGT_COUNT)) {
2126                 *((int *)val) = 1;
2127                 return 0;
2128         }
2129
2130         rc = mdc_get_info_rpc(exp, keylen, key, *vallen, val);
2131
2132         return rc;
2133 }
2134
2135 static int mdc_sync(struct obd_export *exp, const struct lu_fid *fid,
2136                     struct ptlrpc_request **request)
2137 {
2138         struct ptlrpc_request *req;
2139         int                 rc;
2140
2141         *request = NULL;
2142         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_SYNC);
2143         if (req == NULL)
2144                 return -ENOMEM;
2145
2146         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_SYNC);
2147         if (rc) {
2148                 ptlrpc_request_free(req);
2149                 return rc;
2150         }
2151
2152         mdc_pack_body(req, fid, 0, 0, -1, 0);
2153
2154         ptlrpc_request_set_replen(req);
2155
2156         rc = ptlrpc_queue_wait(req);
2157         if (rc)
2158                 ptlrpc_req_finished(req);
2159         else
2160                 *request = req;
2161         return rc;
2162 }
2163
2164 static int mdc_import_event(struct obd_device *obd, struct obd_import *imp,
2165                             enum obd_import_event event)
2166 {
2167         int rc = 0;
2168
2169         LASSERT(imp->imp_obd == obd);
2170
2171         switch (event) {
2172         case IMP_EVENT_DISCON: {
2173 #if 0
2174                 /* XXX Pass event up to OBDs stack. used only for FLD now */
2175                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_DISCON, NULL);
2176 #endif
2177                 break;
2178         }
2179         case IMP_EVENT_INACTIVE: {
2180                 struct client_obd *cli = &obd->u.cli;
2181                 /*
2182                  * Flush current sequence to make client obtain new one
2183                  * from server in case of disconnect/reconnect.
2184                  */
2185                 if (cli->cl_seq != NULL)
2186                         seq_client_flush(cli->cl_seq);
2187
2188                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_INACTIVE, NULL);
2189                 break;
2190         }
2191         case IMP_EVENT_INVALIDATE: {
2192                 struct ldlm_namespace *ns = obd->obd_namespace;
2193
2194                 ldlm_namespace_cleanup(ns, LDLM_FL_LOCAL_ONLY);
2195
2196                 break;
2197         }
2198         case IMP_EVENT_ACTIVE:
2199                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_ACTIVE, NULL);
2200                 /* redo the kuc registration after reconnecting */
2201                 if (rc == 0)
2202                         /* re-register HSM agents */
2203                         rc = libcfs_kkuc_group_foreach(KUC_GRP_HSM,
2204                                                        mdc_hsm_ct_reregister,
2205                                                        (void *)imp);
2206                 break;
2207         case IMP_EVENT_OCD:
2208                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_OCD, NULL);
2209                 break;
2210         case IMP_EVENT_DEACTIVATE:
2211         case IMP_EVENT_ACTIVATE:
2212                 break;
2213         default:
2214                 CERROR("Unknown import event %x\n", event);
2215                 LBUG();
2216         }
2217         return rc;
2218 }
2219
2220 int mdc_fid_alloc(struct obd_export *exp, struct lu_fid *fid,
2221                   struct md_op_data *op_data)
2222 {
2223         struct client_obd *cli = &exp->exp_obd->u.cli;
2224         struct lu_client_seq *seq = cli->cl_seq;
2225
2226         return seq_client_alloc_fid(NULL, seq, fid);
2227 }
2228
2229 static struct obd_uuid *mdc_get_uuid(struct obd_export *exp)
2230 {
2231         struct client_obd *cli = &exp->exp_obd->u.cli;
2232
2233         return &cli->cl_target_uuid;
2234 }
2235
2236 /**
2237  * Determine whether the lock can be canceled before replaying it during
2238  * recovery, non zero value will be return if the lock can be canceled,
2239  * or zero returned for not
2240  */
2241 static int mdc_cancel_for_recovery(struct ldlm_lock *lock)
2242 {
2243         if (lock->l_resource->lr_type != LDLM_IBITS)
2244                 return 0;
2245
2246         /* FIXME: if we ever get into a situation where there are too many
2247          * opened files with open locks on a single node, then we really
2248          * should replay these open locks to reget it */
2249         if (lock->l_policy_data.l_inodebits.bits & MDS_INODELOCK_OPEN)
2250                 return 0;
2251
2252         return 1;
2253 }
2254
2255 static int mdc_resource_inode_free(struct ldlm_resource *res)
2256 {
2257         if (res->lr_lvb_inode)
2258                 res->lr_lvb_inode = NULL;
2259
2260         return 0;
2261 }
2262
2263 static struct ldlm_valblock_ops inode_lvbo = {
2264         .lvbo_free = mdc_resource_inode_free,
2265 };
2266
2267 static int mdc_llog_init(struct obd_device *obd)
2268 {
2269         struct obd_llog_group   *olg = &obd->obd_olg;
2270         struct llog_ctxt        *ctxt;
2271         int                      rc;
2272
2273         rc = llog_setup(NULL, obd, olg, LLOG_CHANGELOG_REPL_CTXT, obd,
2274                         &llog_client_ops);
2275         if (rc)
2276                 return rc;
2277
2278         ctxt = llog_group_get_ctxt(olg, LLOG_CHANGELOG_REPL_CTXT);
2279         llog_initiator_connect(ctxt);
2280         llog_ctxt_put(ctxt);
2281
2282         return 0;
2283 }
2284
2285 static void mdc_llog_finish(struct obd_device *obd)
2286 {
2287         struct llog_ctxt *ctxt;
2288
2289         ctxt = llog_get_context(obd, LLOG_CHANGELOG_REPL_CTXT);
2290         if (ctxt)
2291                 llog_cleanup(NULL, ctxt);
2292 }
2293
2294 static int mdc_setup(struct obd_device *obd, struct lustre_cfg *cfg)
2295 {
2296         struct client_obd *cli = &obd->u.cli;
2297         struct lprocfs_static_vars lvars = { NULL };
2298         int rc;
2299
2300         cli->cl_rpc_lock = kzalloc(sizeof(*cli->cl_rpc_lock), GFP_NOFS);
2301         if (!cli->cl_rpc_lock)
2302                 return -ENOMEM;
2303         mdc_init_rpc_lock(cli->cl_rpc_lock);
2304
2305         ptlrpcd_addref();
2306
2307         cli->cl_close_lock = kzalloc(sizeof(*cli->cl_close_lock), GFP_NOFS);
2308         if (!cli->cl_close_lock) {
2309                 rc = -ENOMEM;
2310                 goto err_rpc_lock;
2311         }
2312         mdc_init_rpc_lock(cli->cl_close_lock);
2313
2314         rc = client_obd_setup(obd, cfg);
2315         if (rc)
2316                 goto err_close_lock;
2317         lprocfs_mdc_init_vars(&lvars);
2318         lprocfs_obd_setup(obd, lvars.obd_vars, lvars.sysfs_vars);
2319         sptlrpc_lprocfs_cliobd_attach(obd);
2320         ptlrpc_lprocfs_register_obd(obd);
2321
2322         ns_register_cancel(obd->obd_namespace, mdc_cancel_for_recovery);
2323
2324         obd->obd_namespace->ns_lvbo = &inode_lvbo;
2325
2326         rc = mdc_llog_init(obd);
2327         if (rc) {
2328                 mdc_cleanup(obd);
2329                 CERROR("failed to setup llogging subsystems\n");
2330         }
2331
2332         return rc;
2333
2334 err_close_lock:
2335         kfree(cli->cl_close_lock);
2336 err_rpc_lock:
2337         kfree(cli->cl_rpc_lock);
2338         ptlrpcd_decref();
2339         return rc;
2340 }
2341
2342 /* Initialize the default and maximum LOV EA and cookie sizes.  This allows
2343  * us to make MDS RPCs with large enough reply buffers to hold a default
2344  * sized EA and cookie without having to calculate this (via a call into the
2345  * LOV + OSCs) each time we make an RPC.  The maximum size is also tracked
2346  * but not used to avoid wastefully vmalloc()'ing large reply buffers when
2347  * a large number of stripes is possible.  If a larger reply buffer is
2348  * required it will be reallocated in the ptlrpc layer due to overflow.
2349  */
2350 static int mdc_init_ea_size(struct obd_export *exp, int easize,
2351                             int def_easize, int cookiesize, int def_cookiesize)
2352 {
2353         struct obd_device *obd = exp->exp_obd;
2354         struct client_obd *cli = &obd->u.cli;
2355
2356         if (cli->cl_max_mds_easize < easize)
2357                 cli->cl_max_mds_easize = easize;
2358
2359         if (cli->cl_default_mds_easize < def_easize)
2360                 cli->cl_default_mds_easize = def_easize;
2361
2362         if (cli->cl_max_mds_cookiesize < cookiesize)
2363                 cli->cl_max_mds_cookiesize = cookiesize;
2364
2365         if (cli->cl_default_mds_cookiesize < def_cookiesize)
2366                 cli->cl_default_mds_cookiesize = def_cookiesize;
2367
2368         return 0;
2369 }
2370
2371 static int mdc_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
2372 {
2373         switch (stage) {
2374         case OBD_CLEANUP_EARLY:
2375                 break;
2376         case OBD_CLEANUP_EXPORTS:
2377                 /* Failsafe, ok if racy */
2378                 if (obd->obd_type->typ_refcnt <= 1)
2379                         libcfs_kkuc_group_rem(0, KUC_GRP_HSM);
2380
2381                 obd_cleanup_client_import(obd);
2382                 ptlrpc_lprocfs_unregister_obd(obd);
2383                 lprocfs_obd_cleanup(obd);
2384
2385                 mdc_llog_finish(obd);
2386                 break;
2387         }
2388         return 0;
2389 }
2390
2391 static int mdc_cleanup(struct obd_device *obd)
2392 {
2393         struct client_obd *cli = &obd->u.cli;
2394
2395         kfree(cli->cl_rpc_lock);
2396         kfree(cli->cl_close_lock);
2397
2398         ptlrpcd_decref();
2399
2400         return client_obd_cleanup(obd);
2401 }
2402
2403 static int mdc_process_config(struct obd_device *obd, u32 len, void *buf)
2404 {
2405         struct lustre_cfg *lcfg = buf;
2406         struct lprocfs_static_vars lvars = { NULL };
2407         int rc = 0;
2408
2409         lprocfs_mdc_init_vars(&lvars);
2410         switch (lcfg->lcfg_command) {
2411         default:
2412                 rc = class_process_proc_param(PARAM_MDC, lvars.obd_vars,
2413                                               lcfg, obd);
2414                 if (rc > 0)
2415                         rc = 0;
2416                 break;
2417         }
2418         return rc;
2419 }
2420
2421 /* get remote permission for current user on fid */
2422 static int mdc_get_remote_perm(struct obd_export *exp, const struct lu_fid *fid,
2423                                __u32 suppgid, struct ptlrpc_request **request)
2424 {
2425         struct ptlrpc_request  *req;
2426         int                 rc;
2427
2428         LASSERT(client_is_remote(exp));
2429
2430         *request = NULL;
2431         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_GETATTR);
2432         if (req == NULL)
2433                 return -ENOMEM;
2434
2435         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR);
2436         if (rc) {
2437                 ptlrpc_request_free(req);
2438                 return rc;
2439         }
2440
2441         mdc_pack_body(req, fid, OBD_MD_FLRMTPERM, 0, suppgid, 0);
2442
2443         req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER,
2444                              sizeof(struct mdt_remote_perm));
2445
2446         ptlrpc_request_set_replen(req);
2447
2448         rc = ptlrpc_queue_wait(req);
2449         if (rc)
2450                 ptlrpc_req_finished(req);
2451         else
2452                 *request = req;
2453         return rc;
2454 }
2455
2456 static struct obd_ops mdc_obd_ops = {
2457         .owner          = THIS_MODULE,
2458         .setup          = mdc_setup,
2459         .precleanup     = mdc_precleanup,
2460         .cleanup        = mdc_cleanup,
2461         .add_conn       = client_import_add_conn,
2462         .del_conn       = client_import_del_conn,
2463         .connect        = client_connect_import,
2464         .disconnect     = client_disconnect_export,
2465         .iocontrol      = mdc_iocontrol,
2466         .set_info_async = mdc_set_info_async,
2467         .statfs         = mdc_statfs,
2468         .fid_init       = client_fid_init,
2469         .fid_fini       = client_fid_fini,
2470         .fid_alloc      = mdc_fid_alloc,
2471         .import_event   = mdc_import_event,
2472         .get_info       = mdc_get_info,
2473         .process_config = mdc_process_config,
2474         .get_uuid       = mdc_get_uuid,
2475         .quotactl       = mdc_quotactl,
2476         .quotacheck     = mdc_quotacheck
2477 };
2478
2479 static struct md_ops mdc_md_ops = {
2480         .getstatus              = mdc_getstatus,
2481         .null_inode             = mdc_null_inode,
2482         .find_cbdata            = mdc_find_cbdata,
2483         .close                  = mdc_close,
2484         .create                 = mdc_create,
2485         .done_writing           = mdc_done_writing,
2486         .enqueue                = mdc_enqueue,
2487         .getattr                = mdc_getattr,
2488         .getattr_name           = mdc_getattr_name,
2489         .intent_lock            = mdc_intent_lock,
2490         .link                   = mdc_link,
2491         .is_subdir              = mdc_is_subdir,
2492         .rename                 = mdc_rename,
2493         .setattr                = mdc_setattr,
2494         .setxattr               = mdc_setxattr,
2495         .getxattr               = mdc_getxattr,
2496         .sync                   = mdc_sync,
2497         .readpage               = mdc_readpage,
2498         .unlink                 = mdc_unlink,
2499         .cancel_unused          = mdc_cancel_unused,
2500         .init_ea_size           = mdc_init_ea_size,
2501         .set_lock_data          = mdc_set_lock_data,
2502         .lock_match             = mdc_lock_match,
2503         .get_lustre_md          = mdc_get_lustre_md,
2504         .free_lustre_md         = mdc_free_lustre_md,
2505         .set_open_replay_data   = mdc_set_open_replay_data,
2506         .clear_open_replay_data = mdc_clear_open_replay_data,
2507         .get_remote_perm        = mdc_get_remote_perm,
2508         .intent_getattr_async   = mdc_intent_getattr_async,
2509         .revalidate_lock        = mdc_revalidate_lock
2510 };
2511
2512 static int __init mdc_init(void)
2513 {
2514         struct lprocfs_static_vars lvars = { NULL };
2515
2516         lprocfs_mdc_init_vars(&lvars);
2517
2518         return class_register_type(&mdc_obd_ops, &mdc_md_ops,
2519                                  LUSTRE_MDC_NAME, NULL);
2520 }
2521
2522 static void /*__exit*/ mdc_exit(void)
2523 {
2524         class_unregister_type(LUSTRE_MDC_NAME);
2525 }
2526
2527 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
2528 MODULE_DESCRIPTION("Lustre Metadata Client");
2529 MODULE_LICENSE("GPL");
2530
2531 module_init(mdc_init);
2532 module_exit(mdc_exit);