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 out:
1185         ptlrpc_req_finished(req);
1186         return rc;
1187 }
1188
1189 static int mdc_ioc_hsm_ct_register(struct obd_import *imp, __u32 archives)
1190 {
1191         __u32                   *archive_mask;
1192         struct ptlrpc_request   *req;
1193         int                      rc;
1194
1195         req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_HSM_CT_REGISTER,
1196                                         LUSTRE_MDS_VERSION,
1197                                         MDS_HSM_CT_REGISTER);
1198         if (req == NULL) {
1199                 rc = -ENOMEM;
1200                 goto out;
1201         }
1202
1203         mdc_pack_body(req, NULL, OBD_MD_FLRMTPERM, 0, 0, 0);
1204
1205         /* Copy hsm_progress struct */
1206         archive_mask = req_capsule_client_get(&req->rq_pill,
1207                                               &RMF_MDS_HSM_ARCHIVE);
1208         if (archive_mask == NULL) {
1209                 rc = -EPROTO;
1210                 goto out;
1211         }
1212
1213         *archive_mask = archives;
1214
1215         ptlrpc_request_set_replen(req);
1216
1217         rc = mdc_queue_wait(req);
1218 out:
1219         ptlrpc_req_finished(req);
1220         return rc;
1221 }
1222
1223 static int mdc_ioc_hsm_current_action(struct obd_export *exp,
1224                                       struct md_op_data *op_data)
1225 {
1226         struct hsm_current_action       *hca = op_data->op_data;
1227         struct hsm_current_action       *req_hca;
1228         struct ptlrpc_request           *req;
1229         int                              rc;
1230
1231         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1232                                    &RQF_MDS_HSM_ACTION);
1233         if (req == NULL)
1234                 return -ENOMEM;
1235
1236         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_ACTION);
1237         if (rc) {
1238                 ptlrpc_request_free(req);
1239                 return rc;
1240         }
1241
1242         mdc_pack_body(req, &op_data->op_fid1, OBD_MD_FLRMTPERM, 0,
1243                       op_data->op_suppgids[0], 0);
1244
1245         ptlrpc_request_set_replen(req);
1246
1247         rc = mdc_queue_wait(req);
1248         if (rc)
1249                 goto out;
1250
1251         req_hca = req_capsule_server_get(&req->rq_pill,
1252                                          &RMF_MDS_HSM_CURRENT_ACTION);
1253         if (req_hca == NULL) {
1254                 rc = -EPROTO;
1255                 goto out;
1256         }
1257
1258         *hca = *req_hca;
1259
1260 out:
1261         ptlrpc_req_finished(req);
1262         return rc;
1263 }
1264
1265 static int mdc_ioc_hsm_ct_unregister(struct obd_import *imp)
1266 {
1267         struct ptlrpc_request   *req;
1268         int                      rc;
1269
1270         req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_HSM_CT_UNREGISTER,
1271                                         LUSTRE_MDS_VERSION,
1272                                         MDS_HSM_CT_UNREGISTER);
1273         if (req == NULL) {
1274                 rc = -ENOMEM;
1275                 goto out;
1276         }
1277
1278         mdc_pack_body(req, NULL, OBD_MD_FLRMTPERM, 0, 0, 0);
1279
1280         ptlrpc_request_set_replen(req);
1281
1282         rc = mdc_queue_wait(req);
1283 out:
1284         ptlrpc_req_finished(req);
1285         return rc;
1286 }
1287
1288 static int mdc_ioc_hsm_state_get(struct obd_export *exp,
1289                                  struct md_op_data *op_data)
1290 {
1291         struct hsm_user_state   *hus = op_data->op_data;
1292         struct hsm_user_state   *req_hus;
1293         struct ptlrpc_request   *req;
1294         int                      rc;
1295
1296         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1297                                    &RQF_MDS_HSM_STATE_GET);
1298         if (req == NULL)
1299                 return -ENOMEM;
1300
1301         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_STATE_GET);
1302         if (rc != 0) {
1303                 ptlrpc_request_free(req);
1304                 return rc;
1305         }
1306
1307         mdc_pack_body(req, &op_data->op_fid1, OBD_MD_FLRMTPERM, 0,
1308                       op_data->op_suppgids[0], 0);
1309
1310         ptlrpc_request_set_replen(req);
1311
1312         rc = mdc_queue_wait(req);
1313         if (rc)
1314                 goto out;
1315
1316         req_hus = req_capsule_server_get(&req->rq_pill, &RMF_HSM_USER_STATE);
1317         if (req_hus == NULL) {
1318                 rc = -EPROTO;
1319                 goto out;
1320         }
1321
1322         *hus = *req_hus;
1323
1324 out:
1325         ptlrpc_req_finished(req);
1326         return rc;
1327 }
1328
1329 static int mdc_ioc_hsm_state_set(struct obd_export *exp,
1330                                  struct md_op_data *op_data)
1331 {
1332         struct hsm_state_set    *hss = op_data->op_data;
1333         struct hsm_state_set    *req_hss;
1334         struct ptlrpc_request   *req;
1335         int                      rc;
1336
1337         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1338                                    &RQF_MDS_HSM_STATE_SET);
1339         if (req == NULL)
1340                 return -ENOMEM;
1341
1342         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_STATE_SET);
1343         if (rc) {
1344                 ptlrpc_request_free(req);
1345                 return rc;
1346         }
1347
1348         mdc_pack_body(req, &op_data->op_fid1, OBD_MD_FLRMTPERM, 0,
1349                       op_data->op_suppgids[0], 0);
1350
1351         /* Copy states */
1352         req_hss = req_capsule_client_get(&req->rq_pill, &RMF_HSM_STATE_SET);
1353         if (req_hss == NULL) {
1354                 rc = -EPROTO;
1355                 goto out;
1356         }
1357         *req_hss = *hss;
1358
1359         ptlrpc_request_set_replen(req);
1360
1361         rc = mdc_queue_wait(req);
1362 out:
1363         ptlrpc_req_finished(req);
1364         return rc;
1365 }
1366
1367 static int mdc_ioc_hsm_request(struct obd_export *exp,
1368                                struct hsm_user_request *hur)
1369 {
1370         struct obd_import       *imp = class_exp2cliimp(exp);
1371         struct ptlrpc_request   *req;
1372         struct hsm_request      *req_hr;
1373         struct hsm_user_item    *req_hui;
1374         char                    *req_opaque;
1375         int                      rc;
1376
1377         req = ptlrpc_request_alloc(imp, &RQF_MDS_HSM_REQUEST);
1378         if (req == NULL) {
1379                 rc = -ENOMEM;
1380                 goto out;
1381         }
1382
1383         req_capsule_set_size(&req->rq_pill, &RMF_MDS_HSM_USER_ITEM, RCL_CLIENT,
1384                              hur->hur_request.hr_itemcount
1385                              * sizeof(struct hsm_user_item));
1386         req_capsule_set_size(&req->rq_pill, &RMF_GENERIC_DATA, RCL_CLIENT,
1387                              hur->hur_request.hr_data_len);
1388
1389         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_REQUEST);
1390         if (rc) {
1391                 ptlrpc_request_free(req);
1392                 return rc;
1393         }
1394
1395         mdc_pack_body(req, NULL, OBD_MD_FLRMTPERM, 0, 0, 0);
1396
1397         /* Copy hsm_request struct */
1398         req_hr = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_REQUEST);
1399         if (req_hr == NULL) {
1400                 rc = -EPROTO;
1401                 goto out;
1402         }
1403         *req_hr = hur->hur_request;
1404
1405         /* Copy hsm_user_item structs */
1406         req_hui = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_USER_ITEM);
1407         if (req_hui == NULL) {
1408                 rc = -EPROTO;
1409                 goto out;
1410         }
1411         memcpy(req_hui, hur->hur_user_item,
1412                hur->hur_request.hr_itemcount * sizeof(struct hsm_user_item));
1413
1414         /* Copy opaque field */
1415         req_opaque = req_capsule_client_get(&req->rq_pill, &RMF_GENERIC_DATA);
1416         if (req_opaque == NULL) {
1417                 rc = -EPROTO;
1418                 goto out;
1419         }
1420         memcpy(req_opaque, hur_data(hur), hur->hur_request.hr_data_len);
1421
1422         ptlrpc_request_set_replen(req);
1423
1424         rc = mdc_queue_wait(req);
1425 out:
1426         ptlrpc_req_finished(req);
1427         return rc;
1428 }
1429
1430 static struct kuc_hdr *changelog_kuc_hdr(char *buf, int len, int flags)
1431 {
1432         struct kuc_hdr *lh = (struct kuc_hdr *)buf;
1433
1434         LASSERT(len <= KUC_CHANGELOG_MSG_MAXSIZE);
1435
1436         lh->kuc_magic = KUC_MAGIC;
1437         lh->kuc_transport = KUC_TRANSPORT_CHANGELOG;
1438         lh->kuc_flags = flags;
1439         lh->kuc_msgtype = CL_RECORD;
1440         lh->kuc_msglen = len;
1441         return lh;
1442 }
1443
1444 #define D_CHANGELOG 0
1445
1446 struct changelog_show {
1447         __u64           cs_startrec;
1448         __u32           cs_flags;
1449         struct file     *cs_fp;
1450         char            *cs_buf;
1451         struct obd_device *cs_obd;
1452 };
1453
1454 static int changelog_kkuc_cb(const struct lu_env *env, struct llog_handle *llh,
1455                              struct llog_rec_hdr *hdr, void *data)
1456 {
1457         struct changelog_show *cs = data;
1458         struct llog_changelog_rec *rec = (struct llog_changelog_rec *)hdr;
1459         struct kuc_hdr *lh;
1460         int len, rc;
1461
1462         if (rec->cr_hdr.lrh_type != CHANGELOG_REC) {
1463                 rc = -EINVAL;
1464                 CERROR("%s: not a changelog rec %x/%d: rc = %d\n",
1465                        cs->cs_obd->obd_name, rec->cr_hdr.lrh_type,
1466                        rec->cr.cr_type, rc);
1467                 return rc;
1468         }
1469
1470         if (rec->cr.cr_index < cs->cs_startrec) {
1471                 /* Skip entries earlier than what we are interested in */
1472                 CDEBUG(D_CHANGELOG, "rec=%llu start=%llu\n",
1473                        rec->cr.cr_index, cs->cs_startrec);
1474                 return 0;
1475         }
1476
1477         CDEBUG(D_CHANGELOG, "%llu %02d%-5s %llu 0x%x t="DFID" p="DFID
1478                 " %.*s\n", rec->cr.cr_index, rec->cr.cr_type,
1479                 changelog_type2str(rec->cr.cr_type), rec->cr.cr_time,
1480                 rec->cr.cr_flags & CLF_FLAGMASK,
1481                 PFID(&rec->cr.cr_tfid), PFID(&rec->cr.cr_pfid),
1482                 rec->cr.cr_namelen, changelog_rec_name(&rec->cr));
1483
1484         len = sizeof(*lh) + changelog_rec_size(&rec->cr) + rec->cr.cr_namelen;
1485
1486         /* Set up the message */
1487         lh = changelog_kuc_hdr(cs->cs_buf, len, cs->cs_flags);
1488         memcpy(lh + 1, &rec->cr, len - sizeof(*lh));
1489
1490         rc = libcfs_kkuc_msg_put(cs->cs_fp, lh);
1491         CDEBUG(D_CHANGELOG, "kucmsg fp %p len %d rc %d\n", cs->cs_fp, len, rc);
1492
1493         return rc;
1494 }
1495
1496 static int mdc_changelog_send_thread(void *csdata)
1497 {
1498         struct changelog_show *cs = csdata;
1499         struct llog_ctxt *ctxt = NULL;
1500         struct llog_handle *llh = NULL;
1501         struct kuc_hdr *kuch;
1502         int rc;
1503
1504         CDEBUG(D_CHANGELOG, "changelog to fp=%p start %llu\n",
1505                cs->cs_fp, cs->cs_startrec);
1506
1507         cs->cs_buf = kzalloc(KUC_CHANGELOG_MSG_MAXSIZE, GFP_NOFS);
1508         if (!cs->cs_buf) {
1509                 rc = -ENOMEM;
1510                 goto out;
1511         }
1512
1513         /* Set up the remote catalog handle */
1514         ctxt = llog_get_context(cs->cs_obd, LLOG_CHANGELOG_REPL_CTXT);
1515         if (ctxt == NULL) {
1516                 rc = -ENOENT;
1517                 goto out;
1518         }
1519         rc = llog_open(NULL, ctxt, &llh, NULL, CHANGELOG_CATALOG,
1520                        LLOG_OPEN_EXISTS);
1521         if (rc) {
1522                 CERROR("%s: fail to open changelog catalog: rc = %d\n",
1523                        cs->cs_obd->obd_name, rc);
1524                 goto out;
1525         }
1526         rc = llog_init_handle(NULL, llh, LLOG_F_IS_CAT, NULL);
1527         if (rc) {
1528                 CERROR("llog_init_handle failed %d\n", rc);
1529                 goto out;
1530         }
1531
1532         rc = llog_cat_process(NULL, llh, changelog_kkuc_cb, cs, 0, 0);
1533
1534         /* Send EOF no matter what our result */
1535         kuch = changelog_kuc_hdr(cs->cs_buf, sizeof(*kuch), cs->cs_flags);
1536         if (kuch) {
1537                 kuch->kuc_msgtype = CL_EOF;
1538                 libcfs_kkuc_msg_put(cs->cs_fp, kuch);
1539         }
1540
1541 out:
1542         fput(cs->cs_fp);
1543         if (llh)
1544                 llog_cat_close(NULL, llh);
1545         if (ctxt)
1546                 llog_ctxt_put(ctxt);
1547         kfree(cs->cs_buf);
1548         kfree(cs);
1549         return rc;
1550 }
1551
1552 static int mdc_ioc_changelog_send(struct obd_device *obd,
1553                                   struct ioc_changelog *icc)
1554 {
1555         struct changelog_show *cs;
1556         int rc;
1557
1558         /* Freed in mdc_changelog_send_thread */
1559         cs = kzalloc(sizeof(*cs), GFP_NOFS);
1560         if (!cs)
1561                 return -ENOMEM;
1562
1563         cs->cs_obd = obd;
1564         cs->cs_startrec = icc->icc_recno;
1565         /* matching fput in mdc_changelog_send_thread */
1566         cs->cs_fp = fget(icc->icc_id);
1567         cs->cs_flags = icc->icc_flags;
1568
1569         /*
1570          * New thread because we should return to user app before
1571          * writing into our pipe
1572          */
1573         rc = PTR_ERR(kthread_run(mdc_changelog_send_thread, cs,
1574                                  "mdc_clg_send_thread"));
1575         if (!IS_ERR_VALUE(rc)) {
1576                 CDEBUG(D_CHANGELOG, "start changelog thread\n");
1577                 return 0;
1578         }
1579
1580         CERROR("Failed to start changelog thread: %d\n", rc);
1581         kfree(cs);
1582         return rc;
1583 }
1584
1585 static int mdc_ioc_hsm_ct_start(struct obd_export *exp,
1586                                 struct lustre_kernelcomm *lk);
1587
1588 static int mdc_quotacheck(struct obd_device *unused, struct obd_export *exp,
1589                           struct obd_quotactl *oqctl)
1590 {
1591         struct client_obd       *cli = &exp->exp_obd->u.cli;
1592         struct ptlrpc_request   *req;
1593         struct obd_quotactl     *body;
1594         int                   rc;
1595
1596         req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp),
1597                                         &RQF_MDS_QUOTACHECK, LUSTRE_MDS_VERSION,
1598                                         MDS_QUOTACHECK);
1599         if (req == NULL)
1600                 return -ENOMEM;
1601
1602         body = req_capsule_client_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
1603         *body = *oqctl;
1604
1605         ptlrpc_request_set_replen(req);
1606
1607         /* the next poll will find -ENODATA, that means quotacheck is
1608          * going on */
1609         cli->cl_qchk_stat = -ENODATA;
1610         rc = ptlrpc_queue_wait(req);
1611         if (rc)
1612                 cli->cl_qchk_stat = rc;
1613         ptlrpc_req_finished(req);
1614         return rc;
1615 }
1616
1617 static int mdc_quota_poll_check(struct obd_export *exp,
1618                                 struct if_quotacheck *qchk)
1619 {
1620         struct client_obd *cli = &exp->exp_obd->u.cli;
1621         int rc;
1622
1623         qchk->obd_uuid = cli->cl_target_uuid;
1624         memcpy(qchk->obd_type, LUSTRE_MDS_NAME, strlen(LUSTRE_MDS_NAME));
1625
1626         rc = cli->cl_qchk_stat;
1627         /* the client is not the previous one */
1628         if (rc == CL_NOT_QUOTACHECKED)
1629                 rc = -EINTR;
1630         return rc;
1631 }
1632
1633 static int mdc_quotactl(struct obd_device *unused, struct obd_export *exp,
1634                         struct obd_quotactl *oqctl)
1635 {
1636         struct ptlrpc_request   *req;
1637         struct obd_quotactl     *oqc;
1638         int                   rc;
1639
1640         req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp),
1641                                         &RQF_MDS_QUOTACTL, LUSTRE_MDS_VERSION,
1642                                         MDS_QUOTACTL);
1643         if (req == NULL)
1644                 return -ENOMEM;
1645
1646         oqc = req_capsule_client_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
1647         *oqc = *oqctl;
1648
1649         ptlrpc_request_set_replen(req);
1650         ptlrpc_at_set_req_timeout(req);
1651         req->rq_no_resend = 1;
1652
1653         rc = ptlrpc_queue_wait(req);
1654         if (rc)
1655                 CERROR("ptlrpc_queue_wait failed, rc: %d\n", rc);
1656
1657         if (req->rq_repmsg) {
1658                 oqc = req_capsule_server_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
1659                 if (oqc) {
1660                         *oqctl = *oqc;
1661                 } else if (!rc) {
1662                         CERROR("Can't unpack obd_quotactl\n");
1663                         rc = -EPROTO;
1664                 }
1665         } else if (!rc) {
1666                 CERROR("Can't unpack obd_quotactl\n");
1667                 rc = -EPROTO;
1668         }
1669         ptlrpc_req_finished(req);
1670
1671         return rc;
1672 }
1673
1674 static int mdc_ioc_swap_layouts(struct obd_export *exp,
1675                                 struct md_op_data *op_data)
1676 {
1677         LIST_HEAD(cancels);
1678         struct ptlrpc_request   *req;
1679         int                      rc, count;
1680         struct mdc_swap_layouts *msl, *payload;
1681
1682         msl = op_data->op_data;
1683
1684         /* When the MDT will get the MDS_SWAP_LAYOUTS RPC the
1685          * first thing it will do is to cancel the 2 layout
1686          * locks hold by this client.
1687          * So the client must cancel its layout locks on the 2 fids
1688          * with the request RPC to avoid extra RPC round trips
1689          */
1690         count = mdc_resource_get_unused(exp, &op_data->op_fid1, &cancels,
1691                                         LCK_CR, MDS_INODELOCK_LAYOUT);
1692         count += mdc_resource_get_unused(exp, &op_data->op_fid2, &cancels,
1693                                          LCK_CR, MDS_INODELOCK_LAYOUT);
1694
1695         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1696                                    &RQF_MDS_SWAP_LAYOUTS);
1697         if (req == NULL) {
1698                 ldlm_lock_list_put(&cancels, l_bl_ast, count);
1699                 return -ENOMEM;
1700         }
1701
1702         rc = mdc_prep_elc_req(exp, req, MDS_SWAP_LAYOUTS, &cancels, count);
1703         if (rc) {
1704                 ptlrpc_request_free(req);
1705                 return rc;
1706         }
1707
1708         mdc_swap_layouts_pack(req, op_data);
1709
1710         payload = req_capsule_client_get(&req->rq_pill, &RMF_SWAP_LAYOUTS);
1711         LASSERT(payload);
1712
1713         *payload = *msl;
1714
1715         ptlrpc_request_set_replen(req);
1716
1717         rc = ptlrpc_queue_wait(req);
1718
1719         ptlrpc_req_finished(req);
1720         return rc;
1721 }
1722
1723 static int mdc_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1724                          void *karg, void *uarg)
1725 {
1726         struct obd_device *obd = exp->exp_obd;
1727         struct obd_ioctl_data *data = karg;
1728         struct obd_import *imp = obd->u.cli.cl_import;
1729         int rc;
1730
1731         if (!try_module_get(THIS_MODULE)) {
1732                 CERROR("Can't get module. Is it alive?");
1733                 return -EINVAL;
1734         }
1735         switch (cmd) {
1736         case OBD_IOC_CHANGELOG_SEND:
1737                 rc = mdc_ioc_changelog_send(obd, karg);
1738                 goto out;
1739         case OBD_IOC_CHANGELOG_CLEAR: {
1740                 struct ioc_changelog *icc = karg;
1741                 struct changelog_setinfo cs = {
1742                         .cs_recno = icc->icc_recno,
1743                         .cs_id = icc->icc_id
1744                 };
1745
1746                 rc = obd_set_info_async(NULL, exp, strlen(KEY_CHANGELOG_CLEAR),
1747                                         KEY_CHANGELOG_CLEAR, sizeof(cs), &cs,
1748                                         NULL);
1749                 goto out;
1750         }
1751         case OBD_IOC_FID2PATH:
1752                 rc = mdc_ioc_fid2path(exp, karg);
1753                 goto out;
1754         case LL_IOC_HSM_CT_START:
1755                 rc = mdc_ioc_hsm_ct_start(exp, karg);
1756                 /* ignore if it was already registered on this MDS. */
1757                 if (rc == -EEXIST)
1758                         rc = 0;
1759                 goto out;
1760         case LL_IOC_HSM_PROGRESS:
1761                 rc = mdc_ioc_hsm_progress(exp, karg);
1762                 goto out;
1763         case LL_IOC_HSM_STATE_GET:
1764                 rc = mdc_ioc_hsm_state_get(exp, karg);
1765                 goto out;
1766         case LL_IOC_HSM_STATE_SET:
1767                 rc = mdc_ioc_hsm_state_set(exp, karg);
1768                 goto out;
1769         case LL_IOC_HSM_ACTION:
1770                 rc = mdc_ioc_hsm_current_action(exp, karg);
1771                 goto out;
1772         case LL_IOC_HSM_REQUEST:
1773                 rc = mdc_ioc_hsm_request(exp, karg);
1774                 goto out;
1775         case OBD_IOC_CLIENT_RECOVER:
1776                 rc = ptlrpc_recover_import(imp, data->ioc_inlbuf1, 0);
1777                 if (rc < 0)
1778                         goto out;
1779                 rc = 0;
1780                 goto out;
1781         case IOC_OSC_SET_ACTIVE:
1782                 rc = ptlrpc_set_import_active(imp, data->ioc_offset);
1783                 goto out;
1784         case OBD_IOC_POLL_QUOTACHECK:
1785                 rc = mdc_quota_poll_check(exp, (struct if_quotacheck *)karg);
1786                 goto out;
1787         case OBD_IOC_PING_TARGET:
1788                 rc = ptlrpc_obd_ping(obd);
1789                 goto out;
1790         /*
1791          * Normally IOC_OBD_STATFS, OBD_IOC_QUOTACTL iocontrol are handled by
1792          * LMV instead of MDC. But when the cluster is upgraded from 1.8,
1793          * there'd be no LMV layer thus we might be called here. Eventually
1794          * this code should be removed.
1795          * bz20731, LU-592.
1796          */
1797         case IOC_OBD_STATFS: {
1798                 struct obd_statfs stat_buf = {0};
1799
1800                 if (*((__u32 *) data->ioc_inlbuf2) != 0) {
1801                         rc = -ENODEV;
1802                         goto out;
1803                 }
1804
1805                 /* copy UUID */
1806                 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(obd),
1807                                  min_t(size_t, data->ioc_plen2,
1808                                                sizeof(struct obd_uuid)))) {
1809                         rc = -EFAULT;
1810                         goto out;
1811                 }
1812
1813                 rc = mdc_statfs(NULL, obd->obd_self_export, &stat_buf,
1814                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1815                                 0);
1816                 if (rc != 0)
1817                         goto out;
1818
1819                 if (copy_to_user(data->ioc_pbuf1, &stat_buf,
1820                                  min_t(size_t, data->ioc_plen1,
1821                                                sizeof(stat_buf)))) {
1822                         rc = -EFAULT;
1823                         goto out;
1824                 }
1825
1826                 rc = 0;
1827                 goto out;
1828         }
1829         case OBD_IOC_QUOTACTL: {
1830                 struct if_quotactl *qctl = karg;
1831                 struct obd_quotactl *oqctl;
1832
1833                 oqctl = kzalloc(sizeof(*oqctl), GFP_NOFS);
1834                 if (!oqctl) {
1835                         rc = -ENOMEM;
1836                         goto out;
1837                 }
1838
1839                 QCTL_COPY(oqctl, qctl);
1840                 rc = obd_quotactl(exp, oqctl);
1841                 if (rc == 0) {
1842                         QCTL_COPY(qctl, oqctl);
1843                         qctl->qc_valid = QC_MDTIDX;
1844                         qctl->obd_uuid = obd->u.cli.cl_target_uuid;
1845                 }
1846
1847                 kfree(oqctl);
1848                 goto out;
1849         }
1850         case LL_IOC_GET_CONNECT_FLAGS:
1851                 if (copy_to_user(uarg, exp_connect_flags_ptr(exp),
1852                                  sizeof(*exp_connect_flags_ptr(exp)))) {
1853                         rc = -EFAULT;
1854                         goto out;
1855                 }
1856
1857                 rc = 0;
1858                 goto out;
1859         case LL_IOC_LOV_SWAP_LAYOUTS:
1860                 rc = mdc_ioc_swap_layouts(exp, karg);
1861                 goto out;
1862         default:
1863                 CERROR("unrecognised ioctl: cmd = %#x\n", cmd);
1864                 rc = -ENOTTY;
1865                 goto out;
1866         }
1867 out:
1868         module_put(THIS_MODULE);
1869
1870         return rc;
1871 }
1872
1873 static int mdc_get_info_rpc(struct obd_export *exp,
1874                             u32 keylen, void *key,
1875                             int vallen, void *val)
1876 {
1877         struct obd_import      *imp = class_exp2cliimp(exp);
1878         struct ptlrpc_request  *req;
1879         char               *tmp;
1880         int                  rc = -EINVAL;
1881
1882         req = ptlrpc_request_alloc(imp, &RQF_MDS_GET_INFO);
1883         if (req == NULL)
1884                 return -ENOMEM;
1885
1886         req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_KEY,
1887                              RCL_CLIENT, keylen);
1888         req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_VALLEN,
1889                              RCL_CLIENT, sizeof(__u32));
1890
1891         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GET_INFO);
1892         if (rc) {
1893                 ptlrpc_request_free(req);
1894                 return rc;
1895         }
1896
1897         tmp = req_capsule_client_get(&req->rq_pill, &RMF_GETINFO_KEY);
1898         memcpy(tmp, key, keylen);
1899         tmp = req_capsule_client_get(&req->rq_pill, &RMF_GETINFO_VALLEN);
1900         memcpy(tmp, &vallen, sizeof(__u32));
1901
1902         req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_VAL,
1903                              RCL_SERVER, vallen);
1904         ptlrpc_request_set_replen(req);
1905
1906         rc = ptlrpc_queue_wait(req);
1907         /* -EREMOTE means the get_info result is partial, and it needs to
1908          * continue on another MDT, see fid2path part in lmv_iocontrol */
1909         if (rc == 0 || rc == -EREMOTE) {
1910                 tmp = req_capsule_server_get(&req->rq_pill, &RMF_GETINFO_VAL);
1911                 memcpy(val, tmp, vallen);
1912                 if (ptlrpc_rep_need_swab(req)) {
1913                         if (KEY_IS(KEY_FID2PATH))
1914                                 lustre_swab_fid2path(val);
1915                 }
1916         }
1917         ptlrpc_req_finished(req);
1918
1919         return rc;
1920 }
1921
1922 static void lustre_swab_hai(struct hsm_action_item *h)
1923 {
1924         __swab32s(&h->hai_len);
1925         __swab32s(&h->hai_action);
1926         lustre_swab_lu_fid(&h->hai_fid);
1927         lustre_swab_lu_fid(&h->hai_dfid);
1928         __swab64s(&h->hai_cookie);
1929         __swab64s(&h->hai_extent.offset);
1930         __swab64s(&h->hai_extent.length);
1931         __swab64s(&h->hai_gid);
1932 }
1933
1934 static void lustre_swab_hal(struct hsm_action_list *h)
1935 {
1936         struct hsm_action_item  *hai;
1937         int                      i;
1938
1939         __swab32s(&h->hal_version);
1940         __swab32s(&h->hal_count);
1941         __swab32s(&h->hal_archive_id);
1942         __swab64s(&h->hal_flags);
1943         hai = hai_zero(h);
1944         for (i = 0; i < h->hal_count; i++, hai = hai_next(hai))
1945                 lustre_swab_hai(hai);
1946 }
1947
1948 static void lustre_swab_kuch(struct kuc_hdr *l)
1949 {
1950         __swab16s(&l->kuc_magic);
1951         /* __u8 l->kuc_transport */
1952         __swab16s(&l->kuc_msgtype);
1953         __swab16s(&l->kuc_msglen);
1954 }
1955
1956 static int mdc_ioc_hsm_ct_start(struct obd_export *exp,
1957                                 struct lustre_kernelcomm *lk)
1958 {
1959         struct obd_import  *imp = class_exp2cliimp(exp);
1960         __u32               archive = lk->lk_data;
1961         int                 rc = 0;
1962
1963         if (lk->lk_group != KUC_GRP_HSM) {
1964                 CERROR("Bad copytool group %d\n", lk->lk_group);
1965                 return -EINVAL;
1966         }
1967
1968         CDEBUG(D_HSM, "CT start r%d w%d u%d g%d f%#x\n", lk->lk_rfd, lk->lk_wfd,
1969                lk->lk_uid, lk->lk_group, lk->lk_flags);
1970
1971         if (lk->lk_flags & LK_FLG_STOP) {
1972                 /* Unregister with the coordinator */
1973                 rc = mdc_ioc_hsm_ct_unregister(imp);
1974         } else {
1975                 rc = mdc_ioc_hsm_ct_register(imp, archive);
1976         }
1977
1978         return rc;
1979 }
1980
1981 /**
1982  * Send a message to any listening copytools
1983  * @param val KUC message (kuc_hdr + hsm_action_list)
1984  * @param len total length of message
1985  */
1986 static int mdc_hsm_copytool_send(int len, void *val)
1987 {
1988         struct kuc_hdr          *lh = (struct kuc_hdr *)val;
1989         struct hsm_action_list  *hal = (struct hsm_action_list *)(lh + 1);
1990
1991         if (len < sizeof(*lh) + sizeof(*hal)) {
1992                 CERROR("Short HSM message %d < %d\n", len,
1993                        (int) (sizeof(*lh) + sizeof(*hal)));
1994                 return -EPROTO;
1995         }
1996         if (lh->kuc_magic == __swab16(KUC_MAGIC)) {
1997                 lustre_swab_kuch(lh);
1998                 lustre_swab_hal(hal);
1999         } else if (lh->kuc_magic != KUC_MAGIC) {
2000                 CERROR("Bad magic %x!=%x\n", lh->kuc_magic, KUC_MAGIC);
2001                 return -EPROTO;
2002         }
2003
2004         CDEBUG(D_HSM,
2005                "Received message mg=%x t=%d m=%d l=%d actions=%d on %s\n",
2006                lh->kuc_magic, lh->kuc_transport, lh->kuc_msgtype,
2007                lh->kuc_msglen, hal->hal_count, hal->hal_fsname);
2008
2009         /* Broadcast to HSM listeners */
2010         return libcfs_kkuc_group_put(KUC_GRP_HSM, lh);
2011 }
2012
2013 /**
2014  * callback function passed to kuc for re-registering each HSM copytool
2015  * running on MDC, after MDT shutdown/recovery.
2016  * @param data archive id served by the copytool
2017  * @param cb_arg callback argument (obd_import)
2018  */
2019 static int mdc_hsm_ct_reregister(__u32 data, void *cb_arg)
2020 {
2021         struct obd_import       *imp = (struct obd_import *)cb_arg;
2022         __u32                    archive = data;
2023         int                      rc;
2024
2025         CDEBUG(D_HA, "recover copytool registration to MDT (archive=%#x)\n",
2026                archive);
2027         rc = mdc_ioc_hsm_ct_register(imp, archive);
2028
2029         /* ignore error if the copytool is already registered */
2030         return ((rc != 0) && (rc != -EEXIST)) ? rc : 0;
2031 }
2032
2033 static int mdc_set_info_async(const struct lu_env *env,
2034                               struct obd_export *exp,
2035                               u32 keylen, void *key,
2036                               u32 vallen, void *val,
2037                               struct ptlrpc_request_set *set)
2038 {
2039         struct obd_import       *imp = class_exp2cliimp(exp);
2040         int                      rc;
2041
2042         if (KEY_IS(KEY_READ_ONLY)) {
2043                 if (vallen != sizeof(int))
2044                         return -EINVAL;
2045
2046                 spin_lock(&imp->imp_lock);
2047                 if (*((int *)val)) {
2048                         imp->imp_connect_flags_orig |= OBD_CONNECT_RDONLY;
2049                         imp->imp_connect_data.ocd_connect_flags |=
2050                                                         OBD_CONNECT_RDONLY;
2051                 } else {
2052                         imp->imp_connect_flags_orig &= ~OBD_CONNECT_RDONLY;
2053                         imp->imp_connect_data.ocd_connect_flags &=
2054                                                         ~OBD_CONNECT_RDONLY;
2055                 }
2056                 spin_unlock(&imp->imp_lock);
2057
2058                 rc = do_set_info_async(imp, MDS_SET_INFO, LUSTRE_MDS_VERSION,
2059                                        keylen, key, vallen, val, set);
2060                 return rc;
2061         }
2062         if (KEY_IS(KEY_SPTLRPC_CONF)) {
2063                 sptlrpc_conf_client_adapt(exp->exp_obd);
2064                 return 0;
2065         }
2066         if (KEY_IS(KEY_FLUSH_CTX)) {
2067                 sptlrpc_import_flush_my_ctx(imp);
2068                 return 0;
2069         }
2070         if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
2071                 rc = do_set_info_async(imp, MDS_SET_INFO, LUSTRE_MDS_VERSION,
2072                                        keylen, key, vallen, val, set);
2073                 return rc;
2074         }
2075         if (KEY_IS(KEY_HSM_COPYTOOL_SEND)) {
2076                 rc = mdc_hsm_copytool_send(vallen, val);
2077                 return rc;
2078         }
2079
2080         CERROR("Unknown key %s\n", (char *)key);
2081         return -EINVAL;
2082 }
2083
2084 static int mdc_get_info(const struct lu_env *env, struct obd_export *exp,
2085                         __u32 keylen, void *key, __u32 *vallen, void *val,
2086                         struct lov_stripe_md *lsm)
2087 {
2088         int rc = -EINVAL;
2089
2090         if (KEY_IS(KEY_MAX_EASIZE)) {
2091                 int mdsize, *max_easize;
2092
2093                 if (*vallen != sizeof(int))
2094                         return -EINVAL;
2095                 mdsize = *(int *)val;
2096                 if (mdsize > exp->exp_obd->u.cli.cl_max_mds_easize)
2097                         exp->exp_obd->u.cli.cl_max_mds_easize = mdsize;
2098                 max_easize = val;
2099                 *max_easize = exp->exp_obd->u.cli.cl_max_mds_easize;
2100                 return 0;
2101         } else if (KEY_IS(KEY_DEFAULT_EASIZE)) {
2102                 int *default_easize;
2103
2104                 if (*vallen != sizeof(int))
2105                         return -EINVAL;
2106                 default_easize = val;
2107                 *default_easize = exp->exp_obd->u.cli.cl_default_mds_easize;
2108                 return 0;
2109         } else if (KEY_IS(KEY_CONN_DATA)) {
2110                 struct obd_import *imp = class_exp2cliimp(exp);
2111                 struct obd_connect_data *data = val;
2112
2113                 if (*vallen != sizeof(*data))
2114                         return -EINVAL;
2115
2116                 *data = imp->imp_connect_data;
2117                 return 0;
2118         } else if (KEY_IS(KEY_TGT_COUNT)) {
2119                 *((int *)val) = 1;
2120                 return 0;
2121         }
2122
2123         rc = mdc_get_info_rpc(exp, keylen, key, *vallen, val);
2124
2125         return rc;
2126 }
2127
2128 static int mdc_sync(struct obd_export *exp, const struct lu_fid *fid,
2129                     struct ptlrpc_request **request)
2130 {
2131         struct ptlrpc_request *req;
2132         int                 rc;
2133
2134         *request = NULL;
2135         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_SYNC);
2136         if (req == NULL)
2137                 return -ENOMEM;
2138
2139         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_SYNC);
2140         if (rc) {
2141                 ptlrpc_request_free(req);
2142                 return rc;
2143         }
2144
2145         mdc_pack_body(req, fid, 0, 0, -1, 0);
2146
2147         ptlrpc_request_set_replen(req);
2148
2149         rc = ptlrpc_queue_wait(req);
2150         if (rc)
2151                 ptlrpc_req_finished(req);
2152         else
2153                 *request = req;
2154         return rc;
2155 }
2156
2157 static int mdc_import_event(struct obd_device *obd, struct obd_import *imp,
2158                             enum obd_import_event event)
2159 {
2160         int rc = 0;
2161
2162         LASSERT(imp->imp_obd == obd);
2163
2164         switch (event) {
2165         case IMP_EVENT_DISCON: {
2166 #if 0
2167                 /* XXX Pass event up to OBDs stack. used only for FLD now */
2168                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_DISCON, NULL);
2169 #endif
2170                 break;
2171         }
2172         case IMP_EVENT_INACTIVE: {
2173                 struct client_obd *cli = &obd->u.cli;
2174                 /*
2175                  * Flush current sequence to make client obtain new one
2176                  * from server in case of disconnect/reconnect.
2177                  */
2178                 if (cli->cl_seq != NULL)
2179                         seq_client_flush(cli->cl_seq);
2180
2181                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_INACTIVE, NULL);
2182                 break;
2183         }
2184         case IMP_EVENT_INVALIDATE: {
2185                 struct ldlm_namespace *ns = obd->obd_namespace;
2186
2187                 ldlm_namespace_cleanup(ns, LDLM_FL_LOCAL_ONLY);
2188
2189                 break;
2190         }
2191         case IMP_EVENT_ACTIVE:
2192                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_ACTIVE, NULL);
2193                 /* redo the kuc registration after reconnecting */
2194                 if (rc == 0)
2195                         /* re-register HSM agents */
2196                         rc = libcfs_kkuc_group_foreach(KUC_GRP_HSM,
2197                                                        mdc_hsm_ct_reregister,
2198                                                        (void *)imp);
2199                 break;
2200         case IMP_EVENT_OCD:
2201                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_OCD, NULL);
2202                 break;
2203         case IMP_EVENT_DEACTIVATE:
2204         case IMP_EVENT_ACTIVATE:
2205                 break;
2206         default:
2207                 CERROR("Unknown import event %x\n", event);
2208                 LBUG();
2209         }
2210         return rc;
2211 }
2212
2213 int mdc_fid_alloc(struct obd_export *exp, struct lu_fid *fid,
2214                   struct md_op_data *op_data)
2215 {
2216         struct client_obd *cli = &exp->exp_obd->u.cli;
2217         struct lu_client_seq *seq = cli->cl_seq;
2218
2219         return seq_client_alloc_fid(NULL, seq, fid);
2220 }
2221
2222 static struct obd_uuid *mdc_get_uuid(struct obd_export *exp)
2223 {
2224         struct client_obd *cli = &exp->exp_obd->u.cli;
2225
2226         return &cli->cl_target_uuid;
2227 }
2228
2229 /**
2230  * Determine whether the lock can be canceled before replaying it during
2231  * recovery, non zero value will be return if the lock can be canceled,
2232  * or zero returned for not
2233  */
2234 static int mdc_cancel_for_recovery(struct ldlm_lock *lock)
2235 {
2236         if (lock->l_resource->lr_type != LDLM_IBITS)
2237                 return 0;
2238
2239         /* FIXME: if we ever get into a situation where there are too many
2240          * opened files with open locks on a single node, then we really
2241          * should replay these open locks to reget it */
2242         if (lock->l_policy_data.l_inodebits.bits & MDS_INODELOCK_OPEN)
2243                 return 0;
2244
2245         return 1;
2246 }
2247
2248 static int mdc_resource_inode_free(struct ldlm_resource *res)
2249 {
2250         if (res->lr_lvb_inode)
2251                 res->lr_lvb_inode = NULL;
2252
2253         return 0;
2254 }
2255
2256 static struct ldlm_valblock_ops inode_lvbo = {
2257         .lvbo_free = mdc_resource_inode_free,
2258 };
2259
2260 static int mdc_llog_init(struct obd_device *obd)
2261 {
2262         struct obd_llog_group   *olg = &obd->obd_olg;
2263         struct llog_ctxt        *ctxt;
2264         int                      rc;
2265
2266         rc = llog_setup(NULL, obd, olg, LLOG_CHANGELOG_REPL_CTXT, obd,
2267                         &llog_client_ops);
2268         if (rc)
2269                 return rc;
2270
2271         ctxt = llog_group_get_ctxt(olg, LLOG_CHANGELOG_REPL_CTXT);
2272         llog_initiator_connect(ctxt);
2273         llog_ctxt_put(ctxt);
2274
2275         return 0;
2276 }
2277
2278 static void mdc_llog_finish(struct obd_device *obd)
2279 {
2280         struct llog_ctxt *ctxt;
2281
2282         ctxt = llog_get_context(obd, LLOG_CHANGELOG_REPL_CTXT);
2283         if (ctxt)
2284                 llog_cleanup(NULL, ctxt);
2285 }
2286
2287 static int mdc_setup(struct obd_device *obd, struct lustre_cfg *cfg)
2288 {
2289         struct client_obd *cli = &obd->u.cli;
2290         struct lprocfs_static_vars lvars = { NULL };
2291         int rc;
2292
2293         cli->cl_rpc_lock = kzalloc(sizeof(*cli->cl_rpc_lock), GFP_NOFS);
2294         if (!cli->cl_rpc_lock)
2295                 return -ENOMEM;
2296         mdc_init_rpc_lock(cli->cl_rpc_lock);
2297
2298         ptlrpcd_addref();
2299
2300         cli->cl_close_lock = kzalloc(sizeof(*cli->cl_close_lock), GFP_NOFS);
2301         if (!cli->cl_close_lock) {
2302                 rc = -ENOMEM;
2303                 goto err_rpc_lock;
2304         }
2305         mdc_init_rpc_lock(cli->cl_close_lock);
2306
2307         rc = client_obd_setup(obd, cfg);
2308         if (rc)
2309                 goto err_close_lock;
2310         lprocfs_mdc_init_vars(&lvars);
2311         lprocfs_obd_setup(obd, lvars.obd_vars, lvars.sysfs_vars);
2312         sptlrpc_lprocfs_cliobd_attach(obd);
2313         ptlrpc_lprocfs_register_obd(obd);
2314
2315         ns_register_cancel(obd->obd_namespace, mdc_cancel_for_recovery);
2316
2317         obd->obd_namespace->ns_lvbo = &inode_lvbo;
2318
2319         rc = mdc_llog_init(obd);
2320         if (rc) {
2321                 mdc_cleanup(obd);
2322                 CERROR("failed to setup llogging subsystems\n");
2323         }
2324
2325         return rc;
2326
2327 err_close_lock:
2328         kfree(cli->cl_close_lock);
2329 err_rpc_lock:
2330         kfree(cli->cl_rpc_lock);
2331         ptlrpcd_decref();
2332         return rc;
2333 }
2334
2335 /* Initialize the default and maximum LOV EA and cookie sizes.  This allows
2336  * us to make MDS RPCs with large enough reply buffers to hold a default
2337  * sized EA and cookie without having to calculate this (via a call into the
2338  * LOV + OSCs) each time we make an RPC.  The maximum size is also tracked
2339  * but not used to avoid wastefully vmalloc()'ing large reply buffers when
2340  * a large number of stripes is possible.  If a larger reply buffer is
2341  * required it will be reallocated in the ptlrpc layer due to overflow.
2342  */
2343 static int mdc_init_ea_size(struct obd_export *exp, int easize,
2344                             int def_easize, int cookiesize, int def_cookiesize)
2345 {
2346         struct obd_device *obd = exp->exp_obd;
2347         struct client_obd *cli = &obd->u.cli;
2348
2349         if (cli->cl_max_mds_easize < easize)
2350                 cli->cl_max_mds_easize = easize;
2351
2352         if (cli->cl_default_mds_easize < def_easize)
2353                 cli->cl_default_mds_easize = def_easize;
2354
2355         if (cli->cl_max_mds_cookiesize < cookiesize)
2356                 cli->cl_max_mds_cookiesize = cookiesize;
2357
2358         if (cli->cl_default_mds_cookiesize < def_cookiesize)
2359                 cli->cl_default_mds_cookiesize = def_cookiesize;
2360
2361         return 0;
2362 }
2363
2364 static int mdc_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
2365 {
2366         switch (stage) {
2367         case OBD_CLEANUP_EARLY:
2368                 break;
2369         case OBD_CLEANUP_EXPORTS:
2370                 /* Failsafe, ok if racy */
2371                 if (obd->obd_type->typ_refcnt <= 1)
2372                         libcfs_kkuc_group_rem(0, KUC_GRP_HSM);
2373
2374                 obd_cleanup_client_import(obd);
2375                 ptlrpc_lprocfs_unregister_obd(obd);
2376                 lprocfs_obd_cleanup(obd);
2377
2378                 mdc_llog_finish(obd);
2379                 break;
2380         }
2381         return 0;
2382 }
2383
2384 static int mdc_cleanup(struct obd_device *obd)
2385 {
2386         struct client_obd *cli = &obd->u.cli;
2387
2388         kfree(cli->cl_rpc_lock);
2389         kfree(cli->cl_close_lock);
2390
2391         ptlrpcd_decref();
2392
2393         return client_obd_cleanup(obd);
2394 }
2395
2396 static int mdc_process_config(struct obd_device *obd, u32 len, void *buf)
2397 {
2398         struct lustre_cfg *lcfg = buf;
2399         struct lprocfs_static_vars lvars = { NULL };
2400         int rc = 0;
2401
2402         lprocfs_mdc_init_vars(&lvars);
2403         switch (lcfg->lcfg_command) {
2404         default:
2405                 rc = class_process_proc_param(PARAM_MDC, lvars.obd_vars,
2406                                               lcfg, obd);
2407                 if (rc > 0)
2408                         rc = 0;
2409                 break;
2410         }
2411         return rc;
2412 }
2413
2414 /* get remote permission for current user on fid */
2415 static int mdc_get_remote_perm(struct obd_export *exp, const struct lu_fid *fid,
2416                                __u32 suppgid, struct ptlrpc_request **request)
2417 {
2418         struct ptlrpc_request  *req;
2419         int                 rc;
2420
2421         LASSERT(client_is_remote(exp));
2422
2423         *request = NULL;
2424         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_GETATTR);
2425         if (req == NULL)
2426                 return -ENOMEM;
2427
2428         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR);
2429         if (rc) {
2430                 ptlrpc_request_free(req);
2431                 return rc;
2432         }
2433
2434         mdc_pack_body(req, fid, OBD_MD_FLRMTPERM, 0, suppgid, 0);
2435
2436         req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER,
2437                              sizeof(struct mdt_remote_perm));
2438
2439         ptlrpc_request_set_replen(req);
2440
2441         rc = ptlrpc_queue_wait(req);
2442         if (rc)
2443                 ptlrpc_req_finished(req);
2444         else
2445                 *request = req;
2446         return rc;
2447 }
2448
2449 static struct obd_ops mdc_obd_ops = {
2450         .owner          = THIS_MODULE,
2451         .setup          = mdc_setup,
2452         .precleanup     = mdc_precleanup,
2453         .cleanup        = mdc_cleanup,
2454         .add_conn       = client_import_add_conn,
2455         .del_conn       = client_import_del_conn,
2456         .connect        = client_connect_import,
2457         .disconnect     = client_disconnect_export,
2458         .iocontrol      = mdc_iocontrol,
2459         .set_info_async = mdc_set_info_async,
2460         .statfs         = mdc_statfs,
2461         .fid_init       = client_fid_init,
2462         .fid_fini       = client_fid_fini,
2463         .fid_alloc      = mdc_fid_alloc,
2464         .import_event   = mdc_import_event,
2465         .get_info       = mdc_get_info,
2466         .process_config = mdc_process_config,
2467         .get_uuid       = mdc_get_uuid,
2468         .quotactl       = mdc_quotactl,
2469         .quotacheck     = mdc_quotacheck
2470 };
2471
2472 static struct md_ops mdc_md_ops = {
2473         .getstatus              = mdc_getstatus,
2474         .null_inode             = mdc_null_inode,
2475         .find_cbdata            = mdc_find_cbdata,
2476         .close                  = mdc_close,
2477         .create                 = mdc_create,
2478         .done_writing           = mdc_done_writing,
2479         .enqueue                = mdc_enqueue,
2480         .getattr                = mdc_getattr,
2481         .getattr_name           = mdc_getattr_name,
2482         .intent_lock            = mdc_intent_lock,
2483         .link                   = mdc_link,
2484         .is_subdir              = mdc_is_subdir,
2485         .rename                 = mdc_rename,
2486         .setattr                = mdc_setattr,
2487         .setxattr               = mdc_setxattr,
2488         .getxattr               = mdc_getxattr,
2489         .sync                   = mdc_sync,
2490         .readpage               = mdc_readpage,
2491         .unlink                 = mdc_unlink,
2492         .cancel_unused          = mdc_cancel_unused,
2493         .init_ea_size           = mdc_init_ea_size,
2494         .set_lock_data          = mdc_set_lock_data,
2495         .lock_match             = mdc_lock_match,
2496         .get_lustre_md          = mdc_get_lustre_md,
2497         .free_lustre_md         = mdc_free_lustre_md,
2498         .set_open_replay_data   = mdc_set_open_replay_data,
2499         .clear_open_replay_data = mdc_clear_open_replay_data,
2500         .get_remote_perm        = mdc_get_remote_perm,
2501         .intent_getattr_async   = mdc_intent_getattr_async,
2502         .revalidate_lock        = mdc_revalidate_lock
2503 };
2504
2505 static int __init mdc_init(void)
2506 {
2507         struct lprocfs_static_vars lvars = { NULL };
2508
2509         lprocfs_mdc_init_vars(&lvars);
2510
2511         return class_register_type(&mdc_obd_ops, &mdc_md_ops,
2512                                  LUSTRE_MDC_NAME, NULL);
2513 }
2514
2515 static void /*__exit*/ mdc_exit(void)
2516 {
2517         class_unregister_type(LUSTRE_MDC_NAME);
2518 }
2519
2520 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
2521 MODULE_DESCRIPTION("Lustre Metadata Client");
2522 MODULE_LICENSE("GPL");
2523
2524 module_init(mdc_init);
2525 module_exit(mdc_exit);