kernel/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, 2012, 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 int mdc_get_lustre_md(struct obd_export *exp, struct ptlrpc_request *req,
 451                       struct obd_export *dt_exp, struct obd_export *md_exp,
 452                       struct lustre_md *md)
 453 {
 454         struct req_capsule *pill = &req->rq_pill;
 455         int rc;
 456
 457         LASSERT(md);
 458         memset(md, 0, sizeof(*md));
 459
 460         md->body = req_capsule_server_get(pill, &RMF_MDT_BODY);
 461         LASSERT(md->body != NULL);
 462
 463         if (md->body->valid & OBD_MD_FLEASIZE) {
 464                 int lmmsize;
 465                 struct lov_mds_md *lmm;
 466
 467                 if (!S_ISREG(md->body->mode)) {
 468                         CDEBUG(D_INFO,
 469                                "OBD_MD_FLEASIZE set, should be a regular file, but is not\n");
 470                         rc = -EPROTO;
 471                         goto out;
 472                 }
 473
 474                 if (md->body->eadatasize == 0) {
 475                         CDEBUG(D_INFO,
 476                                "OBD_MD_FLEASIZE set, but eadatasize 0\n");
 477                         rc = -EPROTO;
 478                         goto out;
 479                 }
 480                 lmmsize = md->body->eadatasize;
 481                 lmm = req_capsule_server_sized_get(pill, &RMF_MDT_MD, lmmsize);
 482                 if (!lmm) {
 483                         rc = -EPROTO;
 484                         goto out;
 485                 }
 486
 487                 rc = obd_unpackmd(dt_exp, &md->lsm, lmm, lmmsize);
 488                 if (rc < 0)
 489                         goto out;
 490
 491                 if (rc < sizeof(*md->lsm)) {
 492                         CDEBUG(D_INFO,
 493                                "lsm size too small: rc < sizeof (*md->lsm) (%d < %d)\n",
 494                                rc, (int)sizeof(*md->lsm));
 495                         rc = -EPROTO;
 496                         goto out;
 497                 }
 498
 499         } else if (md->body->valid & OBD_MD_FLDIREA) {
 500                 int lmvsize;
 501                 struct lov_mds_md *lmv;
 502
 503                 if (!S_ISDIR(md->body->mode)) {
 504                         CDEBUG(D_INFO,
 505                                "OBD_MD_FLDIREA set, should be a directory, but is not\n");
 506                         rc = -EPROTO;
 507                         goto out;
 508                 }
 509
 510                 if (md->body->eadatasize == 0) {
 511                         CDEBUG(D_INFO,
 512                                "OBD_MD_FLDIREA is set, but eadatasize 0\n");
 513                         return -EPROTO;
 514                 }
 515                 if (md->body->valid & OBD_MD_MEA) {
 516                         lmvsize = md->body->eadatasize;
 517                         lmv = req_capsule_server_sized_get(pill, &RMF_MDT_MD,
 518                                                            lmvsize);
 519                         if (!lmv) {
 520                                 rc = -EPROTO;
 521                                 goto out;
 522                         }
 523
 524                         rc = obd_unpackmd(md_exp, (void *)&md->mea, lmv,
 525                                           lmvsize);
 526                         if (rc < 0)
 527                                 goto out;
 528
 529                         if (rc < sizeof(*md->mea)) {
 530                                 CDEBUG(D_INFO,
 531                                        "size too small: rc < sizeof(*md->mea) (%d < %d)\n",
 532                                         rc, (int)sizeof(*md->mea));
 533                                 rc = -EPROTO;
 534                                 goto out;
 535                         }
 536                 }
 537         }
 538         rc = 0;
 539
 540         if (md->body->valid & OBD_MD_FLRMTPERM) {
 541                 /* remote permission */
 542                 LASSERT(client_is_remote(exp));
 543                 md->remote_perm = req_capsule_server_swab_get(pill, &RMF_ACL,
 544                                                 lustre_swab_mdt_remote_perm);
 545                 if (!md->remote_perm) {
 546                         rc = -EPROTO;
 547                         goto out;
 548                 }
 549         } else if (md->body->valid & OBD_MD_FLACL) {
 550                 /* for ACL, it's possible that FLACL is set but aclsize is zero.
 551                  * only when aclsize != 0 there's an actual segment for ACL
 552                  * in reply buffer.
 553                  */
 554                 if (md->body->aclsize) {
 555                         rc = mdc_unpack_acl(req, md);
 556                         if (rc)
 557                                 goto out;
 558 #ifdef CONFIG_FS_POSIX_ACL
 559                 } else {
 560                         md->posix_acl = NULL;
 561 #endif
 562                 }
 563         }
 564
 565 out:
 566         if (rc) {
 567 #ifdef CONFIG_FS_POSIX_ACL
 568                 posix_acl_release(md->posix_acl);
 569 #endif
 570                 if (md->lsm)
 571                         obd_free_memmd(dt_exp, &md->lsm);
 572         }
 573         return rc;
 574 }
 575
 576 int mdc_free_lustre_md(struct obd_export *exp, struct lustre_md *md)
 577 {
 578         return 0;
 579 }
 580
 581 /**
 582  * Handles both OPEN and SETATTR RPCs for OPEN-CLOSE and SETATTR-DONE_WRITING
 583  * RPC chains.
 584  */
 585 void mdc_replay_open(struct ptlrpc_request *req)
 586 {
 587         struct md_open_data *mod = req->rq_cb_data;
 588         struct ptlrpc_request *close_req;
 589         struct obd_client_handle *och;
 590         struct lustre_handle old;
 591         struct mdt_body *body;
 592
 593         if (mod == NULL) {
 594                 DEBUG_REQ(D_ERROR, req,
 595                           "Can't properly replay without open data.");
 596                 return;
 597         }
 598
 599         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 600         LASSERT(body != NULL);
 601
 602         och = mod->mod_och;
 603         if (och != NULL) {
 604                 struct lustre_handle *file_fh;
 605
 606                 LASSERT(och->och_magic == OBD_CLIENT_HANDLE_MAGIC);
 607
 608                 file_fh = &och->och_fh;
 609                 CDEBUG(D_HA, "updating handle from %#llx to %#llx\n",
 610                        file_fh->cookie, body->handle.cookie);
 611                 old = *file_fh;
 612                 *file_fh = body->handle;
 613         }
 614         close_req = mod->mod_close_req;
 615         if (close_req != NULL) {
 616                 __u32 opc = lustre_msg_get_opc(close_req->rq_reqmsg);
 617                 struct mdt_ioepoch *epoch;
 618
 619                 LASSERT(opc == MDS_CLOSE || opc == MDS_DONE_WRITING);
 620                 epoch = req_capsule_client_get(&close_req->rq_pill,
 621                                                &RMF_MDT_EPOCH);
 622                 LASSERT(epoch);
 623
 624                 if (och != NULL)
 625                         LASSERT(!memcmp(&old, &epoch->handle, sizeof(old)));
 626                 DEBUG_REQ(D_HA, close_req, "updating close body with new fh");
 627                 epoch->handle = body->handle;
 628         }
 629 }
 630
 631 void mdc_commit_open(struct ptlrpc_request *req)
 632 {
 633         struct md_open_data *mod = req->rq_cb_data;
 634
 635         if (mod == NULL)
 636                 return;
 637
 638         /**
 639          * No need to touch md_open_data::mod_och, it holds a reference on
 640          * \var mod and will zero references to each other, \var mod will be
 641          * freed after that when md_open_data::mod_och will put the reference.
 642          */
 643
 644         /**
 645          * Do not let open request to disappear as it still may be needed
 646          * for close rpc to happen (it may happen on evict only, otherwise
 647          * ptlrpc_request::rq_replay does not let mdc_commit_open() to be
 648          * called), just mark this rpc as committed to distinguish these 2
 649          * cases, see mdc_close() for details. The open request reference will
 650          * be put along with freeing \var mod.
 651          */
 652         ptlrpc_request_addref(req);
 653         spin_lock(&req->rq_lock);
 654         req->rq_committed = 1;
 655         spin_unlock(&req->rq_lock);
 656         req->rq_cb_data = NULL;
 657         obd_mod_put(mod);
 658 }
 659
 660 int mdc_set_open_replay_data(struct obd_export *exp,
 661                              struct obd_client_handle *och,
 662                              struct lookup_intent *it)
 663 {
 664         struct md_open_data   *mod;
 665         struct mdt_rec_create *rec;
 666         struct mdt_body       *body;
 667         struct ptlrpc_request *open_req = it->d.lustre.it_data;
 668         struct obd_import     *imp = open_req->rq_import;
 669
 670         if (!open_req->rq_replay)
 671                 return 0;
 672
 673         rec = req_capsule_client_get(&open_req->rq_pill, &RMF_REC_REINT);
 674         body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
 675         LASSERT(rec != NULL);
 676         /* Incoming message in my byte order (it's been swabbed). */
 677         /* Outgoing messages always in my byte order. */
 678         LASSERT(body != NULL);
 679
 680         /* Only if the import is replayable, we set replay_open data */
 681         if (och && imp->imp_replayable) {
 682                 mod = obd_mod_alloc();
 683                 if (mod == NULL) {
 684                         DEBUG_REQ(D_ERROR, open_req,
 685                                   "Can't allocate md_open_data");
 686                         return 0;
 687                 }
 688
 689                 /**
 690                  * Take a reference on \var mod, to be freed on mdc_close().
 691                  * It protects \var mod from being freed on eviction (commit
 692                  * callback is called despite rq_replay flag).
 693                  * Another reference for \var och.
 694                  */
 695                 obd_mod_get(mod);
 696                 obd_mod_get(mod);
 697
 698                 spin_lock(&open_req->rq_lock);
 699                 och->och_mod = mod;
 700                 mod->mod_och = och;
 701                 mod->mod_is_create = it_disposition(it, DISP_OPEN_CREATE) ||
 702                                      it_disposition(it, DISP_OPEN_STRIPE);
 703                 mod->mod_open_req = open_req;
 704                 open_req->rq_cb_data = mod;
 705                 open_req->rq_commit_cb = mdc_commit_open;
 706                 spin_unlock(&open_req->rq_lock);
 707         }
 708
 709         rec->cr_fid2 = body->fid1;
 710         rec->cr_ioepoch = body->ioepoch;
 711         rec->cr_old_handle.cookie = body->handle.cookie;
 712         open_req->rq_replay_cb = mdc_replay_open;
 713         if (!fid_is_sane(&body->fid1)) {
 714                 DEBUG_REQ(D_ERROR, open_req,
 715                           "Saving replay request with insane fid");
 716                 LBUG();
 717         }
 718
 719         DEBUG_REQ(D_RPCTRACE, open_req, "Set up open replay data");
 720         return 0;
 721 }
 722
 723 static void mdc_free_open(struct md_open_data *mod)
 724 {
 725         int committed = 0;
 726
 727         if (mod->mod_is_create == 0 &&
 728             imp_connect_disp_stripe(mod->mod_open_req->rq_import))
 729                 committed = 1;
 730
 731         LASSERT(mod->mod_open_req->rq_replay == 0);
 732
 733         DEBUG_REQ(D_RPCTRACE, mod->mod_open_req, "free open request\n");
 734
 735         ptlrpc_request_committed(mod->mod_open_req, committed);
 736         if (mod->mod_close_req)
 737                 ptlrpc_request_committed(mod->mod_close_req, committed);
 738 }
 739
 740 int mdc_clear_open_replay_data(struct obd_export *exp,
 741                                struct obd_client_handle *och)
 742 {
 743         struct md_open_data *mod = och->och_mod;
 744
 745         /**
 746          * It is possible to not have \var mod in a case of eviction between
 747          * lookup and ll_file_open().
 748          **/
 749         if (mod == NULL)
 750                 return 0;
 751
 752         LASSERT(mod != LP_POISON);
 753         LASSERT(mod->mod_open_req != NULL);
 754         mdc_free_open(mod);
 755
 756         mod->mod_och = NULL;
 757         och->och_mod = NULL;
 758         obd_mod_put(mod);
 759
 760         return 0;
 761 }
 762
 763 /* Prepares the request for the replay by the given reply */
 764 static void mdc_close_handle_reply(struct ptlrpc_request *req,
 765                                    struct md_op_data *op_data, int rc) {
 766         struct mdt_body  *repbody;
 767         struct mdt_ioepoch *epoch;
 768
 769         if (req && rc == -EAGAIN) {
 770                 repbody = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 771                 epoch = req_capsule_client_get(&req->rq_pill, &RMF_MDT_EPOCH);
 772
 773                 epoch->flags |= MF_SOM_AU;
 774                 if (repbody->valid & OBD_MD_FLGETATTRLOCK)
 775                         op_data->op_flags |= MF_GETATTR_LOCK;
 776         }
 777 }
 778
 779 static int mdc_close(struct obd_export *exp, struct md_op_data *op_data,
 780                      struct md_open_data *mod, struct ptlrpc_request **request)
 781 {
 782         struct obd_device     *obd = class_exp2obd(exp);
 783         struct ptlrpc_request *req;
 784         struct req_format     *req_fmt;
 785         int                    rc;
 786         int                    saved_rc = 0;
 787
 788         req_fmt = &RQF_MDS_CLOSE;
 789         if (op_data->op_bias & MDS_HSM_RELEASE) {
 790                 req_fmt = &RQF_MDS_RELEASE_CLOSE;
 791
 792                 /* allocate a FID for volatile file */
 793                 rc = mdc_fid_alloc(exp, &op_data->op_fid2, op_data);
 794                 if (rc < 0) {
 795                         CERROR("%s: "DFID" failed to allocate FID: %d\n",
 796                                obd->obd_name, PFID(&op_data->op_fid1), rc);
 797                         /* save the errcode and proceed to close */
 798                         saved_rc = rc;
 799                 }
 800         }
 801
 802         *request = NULL;
 803         req = ptlrpc_request_alloc(class_exp2cliimp(exp), req_fmt);
 804         if (req == NULL)
 805                 return -ENOMEM;
 806
 807         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_CLOSE);
 808         if (rc) {
 809                 ptlrpc_request_free(req);
 810                 return rc;
 811         }
 812
 813         /* To avoid a livelock (bug 7034), we need to send CLOSE RPCs to a
 814          * portal whose threads are not taking any DLM locks and are therefore
 815          * always progressing */
 816         req->rq_request_portal = MDS_READPAGE_PORTAL;
 817         ptlrpc_at_set_req_timeout(req);
 818
 819         /* Ensure that this close's handle is fixed up during replay. */
 820         if (likely(mod != NULL)) {
 821                 LASSERTF(mod->mod_open_req != NULL &&
 822                          mod->mod_open_req->rq_type != LI_POISON,
 823                          "POISONED open %p!\n", mod->mod_open_req);
 824
 825                 mod->mod_close_req = req;
 826
 827                 DEBUG_REQ(D_HA, mod->mod_open_req, "matched open");
 828                 /* We no longer want to preserve this open for replay even
 829                  * though the open was committed. b=3632, b=3633 */
 830                 spin_lock(&mod->mod_open_req->rq_lock);
 831                 mod->mod_open_req->rq_replay = 0;
 832                 spin_unlock(&mod->mod_open_req->rq_lock);
 833         } else {
 834                  CDEBUG(D_HA,
 835                         "couldn't find open req; expecting close error\n");
 836         }
 837
 838         mdc_close_pack(req, op_data);
 839
 840         req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
 841                              obd->u.cli.cl_default_mds_easize);
 842         req_capsule_set_size(&req->rq_pill, &RMF_LOGCOOKIES, RCL_SERVER,
 843                              obd->u.cli.cl_default_mds_cookiesize);
 844
 845         ptlrpc_request_set_replen(req);
 846
 847         mdc_get_rpc_lock(obd->u.cli.cl_close_lock, NULL);
 848         rc = ptlrpc_queue_wait(req);
 849         mdc_put_rpc_lock(obd->u.cli.cl_close_lock, NULL);
 850
 851         if (req->rq_repmsg == NULL) {
 852                 CDEBUG(D_RPCTRACE, "request failed to send: %p, %d\n", req,
 853                        req->rq_status);
 854                 if (rc == 0)
 855                         rc = req->rq_status ?: -EIO;
 856         } else if (rc == 0 || rc == -EAGAIN) {
 857                 struct mdt_body *body;
 858
 859                 rc = lustre_msg_get_status(req->rq_repmsg);
 860                 if (lustre_msg_get_type(req->rq_repmsg) == PTL_RPC_MSG_ERR) {
 861                         DEBUG_REQ(D_ERROR, req,
 862                                   "type == PTL_RPC_MSG_ERR, err = %d", rc);
 863                         if (rc > 0)
 864                                 rc = -rc;
 865                 }
 866                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 867                 if (body == NULL)
 868                         rc = -EPROTO;
 869         } else if (rc == -ESTALE) {
 870                 /**
 871                  * it can be allowed error after 3633 if open was committed and
 872                  * server failed before close was sent. Let's check if mod
 873                  * exists and return no error in that case
 874                  */
 875                 if (mod) {
 876                         DEBUG_REQ(D_HA, req, "Reset ESTALE = %d", rc);
 877                         LASSERT(mod->mod_open_req != NULL);
 878                         if (mod->mod_open_req->rq_committed)
 879                                 rc = 0;
 880                 }
 881         }
 882
 883         if (mod) {
 884                 if (rc != 0)
 885                         mod->mod_close_req = NULL;
 886                 /* Since now, mod is accessed through open_req only,
 887                  * thus close req does not keep a reference on mod anymore. */
 888                 obd_mod_put(mod);
 889         }
 890         *request = req;
 891         mdc_close_handle_reply(req, op_data, rc);
 892         return rc < 0 ? rc : saved_rc;
 893 }
 894
 895 static int mdc_done_writing(struct obd_export *exp, struct md_op_data *op_data,
 896                             struct md_open_data *mod)
 897 {
 898         struct obd_device     *obd = class_exp2obd(exp);
 899         struct ptlrpc_request *req;
 900         int                 rc;
 901
 902         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
 903                                    &RQF_MDS_DONE_WRITING);
 904         if (req == NULL)
 905                 return -ENOMEM;
 906
 907         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_DONE_WRITING);
 908         if (rc) {
 909                 ptlrpc_request_free(req);
 910                 return rc;
 911         }
 912
 913         if (mod != NULL) {
 914                 LASSERTF(mod->mod_open_req != NULL &&
 915                          mod->mod_open_req->rq_type != LI_POISON,
 916                          "POISONED setattr %p!\n", mod->mod_open_req);
 917
 918                 mod->mod_close_req = req;
 919                 DEBUG_REQ(D_HA, mod->mod_open_req, "matched setattr");
 920                 /* We no longer want to preserve this setattr for replay even
 921                  * though the open was committed. b=3632, b=3633 */
 922                 spin_lock(&mod->mod_open_req->rq_lock);
 923                 mod->mod_open_req->rq_replay = 0;
 924                 spin_unlock(&mod->mod_open_req->rq_lock);
 925         }
 926
 927         mdc_close_pack(req, op_data);
 928         ptlrpc_request_set_replen(req);
 929
 930         mdc_get_rpc_lock(obd->u.cli.cl_close_lock, NULL);
 931         rc = ptlrpc_queue_wait(req);
 932         mdc_put_rpc_lock(obd->u.cli.cl_close_lock, NULL);
 933
 934         if (rc == -ESTALE) {
 935                 /**
 936                  * it can be allowed error after 3633 if open or setattr were
 937                  * committed and server failed before close was sent.
 938                  * Let's check if mod exists and return no error in that case
 939                  */
 940                 if (mod) {
 941                         LASSERT(mod->mod_open_req != NULL);
 942                         if (mod->mod_open_req->rq_committed)
 943                                 rc = 0;
 944                 }
 945         }
 946
 947         if (mod) {
 948                 if (rc != 0)
 949                         mod->mod_close_req = NULL;
 950                 LASSERT(mod->mod_open_req != NULL);
 951                 mdc_free_open(mod);
 952
 953                 /* Since now, mod is accessed through setattr req only,
 954                  * thus DW req does not keep a reference on mod anymore. */
 955                 obd_mod_put(mod);
 956         }
 957
 958         mdc_close_handle_reply(req, op_data, rc);
 959         ptlrpc_req_finished(req);
 960         return rc;
 961 }
 962
 963 static int mdc_readpage(struct obd_export *exp, struct md_op_data *op_data,
 964                         struct page **pages, struct ptlrpc_request **request)
 965 {
 966         struct ptlrpc_request   *req;
 967         struct ptlrpc_bulk_desc *desc;
 968         int                   i;
 969         wait_queue_head_t             waitq;
 970         int                   resends = 0;
 971         struct l_wait_info       lwi;
 972         int                   rc;
 973
 974         *request = NULL;
 975         init_waitqueue_head(&waitq);
 976
 977 restart_bulk:
 978         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_READPAGE);
 979         if (req == NULL)
 980                 return -ENOMEM;
 981
 982         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_READPAGE);
 983         if (rc) {
 984                 ptlrpc_request_free(req);
 985                 return rc;
 986         }
 987
 988         req->rq_request_portal = MDS_READPAGE_PORTAL;
 989         ptlrpc_at_set_req_timeout(req);
 990
 991         desc = ptlrpc_prep_bulk_imp(req, op_data->op_npages, 1, BULK_PUT_SINK,
 992                                     MDS_BULK_PORTAL);
 993         if (desc == NULL) {
 994                 ptlrpc_request_free(req);
 995                 return -ENOMEM;
 996         }
 997
 998         /* NB req now owns desc and will free it when it gets freed */
 999         for (i = 0; i < op_data->op_npages; i++)
1000                 ptlrpc_prep_bulk_page_pin(desc, pages[i], 0, PAGE_CACHE_SIZE);
1001
1002         mdc_readdir_pack(req, op_data->op_offset,
1003                          PAGE_CACHE_SIZE * op_data->op_npages,
1004                          &op_data->op_fid1);
1005
1006         ptlrpc_request_set_replen(req);
1007         rc = ptlrpc_queue_wait(req);
1008         if (rc) {
1009                 ptlrpc_req_finished(req);
1010                 if (rc != -ETIMEDOUT)
1011                         return rc;
1012
1013                 resends++;
1014                 if (!client_should_resend(resends, &exp->exp_obd->u.cli)) {
1015                         CERROR("too many resend retries, returning error\n");
1016                         return -EIO;
1017                 }
1018                 lwi = LWI_TIMEOUT_INTR(cfs_time_seconds(resends),
1019                                        NULL, NULL, NULL);
1020                 l_wait_event(waitq, 0, &lwi);
1021
1022                 goto restart_bulk;
1023         }
1024
1025         rc = sptlrpc_cli_unwrap_bulk_read(req, req->rq_bulk,
1026                                           req->rq_bulk->bd_nob_transferred);
1027         if (rc < 0) {
1028                 ptlrpc_req_finished(req);
1029                 return rc;
1030         }
1031
1032         if (req->rq_bulk->bd_nob_transferred & ~LU_PAGE_MASK) {
1033                 CERROR("Unexpected # bytes transferred: %d (%ld expected)\n",
1034                         req->rq_bulk->bd_nob_transferred,
1035                         PAGE_CACHE_SIZE * op_data->op_npages);
1036                 ptlrpc_req_finished(req);
1037                 return -EPROTO;
1038         }
1039
1040         *request = req;
1041         return 0;
1042 }
1043
1044 static int mdc_statfs(const struct lu_env *env,
1045                       struct obd_export *exp, struct obd_statfs *osfs,
1046                       __u64 max_age, __u32 flags)
1047 {
1048         struct obd_device     *obd = class_exp2obd(exp);
1049         struct ptlrpc_request *req;
1050         struct obd_statfs     *msfs;
1051         struct obd_import     *imp = NULL;
1052         int                 rc;
1053
1054         /*
1055          * Since the request might also come from lprocfs, so we need
1056          * sync this with client_disconnect_export Bug15684
1057          */
1058         down_read(&obd->u.cli.cl_sem);
1059         if (obd->u.cli.cl_import)
1060                 imp = class_import_get(obd->u.cli.cl_import);
1061         up_read(&obd->u.cli.cl_sem);
1062         if (!imp)
1063                 return -ENODEV;
1064
1065         req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_STATFS,
1066                                         LUSTRE_MDS_VERSION, MDS_STATFS);
1067         if (req == NULL) {
1068                 rc = -ENOMEM;
1069                 goto output;
1070         }
1071
1072         ptlrpc_request_set_replen(req);
1073
1074         if (flags & OBD_STATFS_NODELAY) {
1075                 /* procfs requests not want stay in wait for avoid deadlock */
1076                 req->rq_no_resend = 1;
1077                 req->rq_no_delay = 1;
1078         }
1079
1080         rc = ptlrpc_queue_wait(req);
1081         if (rc) {
1082                 /* check connection error first */
1083                 if (imp->imp_connect_error)
1084                         rc = imp->imp_connect_error;
1085                 goto out;
1086         }
1087
1088         msfs = req_capsule_server_get(&req->rq_pill, &RMF_OBD_STATFS);
1089         if (msfs == NULL) {
1090                 rc = -EPROTO;
1091                 goto out;
1092         }
1093
1094         *osfs = *msfs;
1095 out:
1096         ptlrpc_req_finished(req);
1097 output:
1098         class_import_put(imp);
1099         return rc;
1100 }
1101
1102 static int mdc_ioc_fid2path(struct obd_export *exp, struct getinfo_fid2path *gf)
1103 {
1104         __u32 keylen, vallen;
1105         void *key;
1106         int rc;
1107
1108         if (gf->gf_pathlen > PATH_MAX)
1109                 return -ENAMETOOLONG;
1110         if (gf->gf_pathlen < 2)
1111                 return -EOVERFLOW;
1112
1113         /* Key is KEY_FID2PATH + getinfo_fid2path description */
1114         keylen = cfs_size_round(sizeof(KEY_FID2PATH)) + sizeof(*gf);
1115         key = kzalloc(keylen, GFP_NOFS);
1116         if (!key)
1117                 return -ENOMEM;
1118         memcpy(key, KEY_FID2PATH, sizeof(KEY_FID2PATH));
1119         memcpy(key + cfs_size_round(sizeof(KEY_FID2PATH)), gf, sizeof(*gf));
1120
1121         CDEBUG(D_IOCTL, "path get "DFID" from %llu #%d\n",
1122                PFID(&gf->gf_fid), gf->gf_recno, gf->gf_linkno);
1123
1124         if (!fid_is_sane(&gf->gf_fid)) {
1125                 rc = -EINVAL;
1126                 goto out;
1127         }
1128
1129         /* Val is struct getinfo_fid2path result plus path */
1130         vallen = sizeof(*gf) + gf->gf_pathlen;
1131
1132         rc = obd_get_info(NULL, exp, keylen, key, &vallen, gf, NULL);
1133         if (rc != 0 && rc != -EREMOTE)
1134                 goto out;
1135
1136         if (vallen <= sizeof(*gf)) {
1137                 rc = -EPROTO;
1138                 goto out;
1139         } else if (vallen > sizeof(*gf) + gf->gf_pathlen) {
1140                 rc = -EOVERFLOW;
1141                 goto out;
1142         }
1143
1144         CDEBUG(D_IOCTL, "path get "DFID" from %llu #%d\n%s\n",
1145                PFID(&gf->gf_fid), gf->gf_recno, gf->gf_linkno, gf->gf_path);
1146
1147 out:
1148         kfree(key);
1149         return rc;
1150 }
1151
1152 static int mdc_ioc_hsm_progress(struct obd_export *exp,
1153                                 struct hsm_progress_kernel *hpk)
1154 {
1155         struct obd_import               *imp = class_exp2cliimp(exp);
1156         struct hsm_progress_kernel      *req_hpk;
1157         struct ptlrpc_request           *req;
1158         int                              rc;
1159
1160         req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_HSM_PROGRESS,
1161                                         LUSTRE_MDS_VERSION, MDS_HSM_PROGRESS);
1162         if (req == NULL) {
1163                 rc = -ENOMEM;
1164                 goto out;
1165         }
1166
1167         mdc_pack_body(req, NULL, OBD_MD_FLRMTPERM, 0, 0, 0);
1168
1169         /* Copy hsm_progress struct */
1170         req_hpk = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_PROGRESS);
1171         if (req_hpk == NULL) {
1172                 rc = -EPROTO;
1173                 goto out;
1174         }
1175
1176         *req_hpk = *hpk;
1177         req_hpk->hpk_errval = lustre_errno_hton(hpk->hpk_errval);
1178
1179         ptlrpc_request_set_replen(req);
1180
1181         rc = mdc_queue_wait(req);
1182         goto out;
1183 out:
1184         ptlrpc_req_finished(req);
1185         return rc;
1186 }
1187
1188 static int mdc_ioc_hsm_ct_register(struct obd_import *imp, __u32 archives)
1189 {
1190         __u32                   *archive_mask;
1191         struct ptlrpc_request   *req;
1192         int                      rc;
1193
1194         req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_HSM_CT_REGISTER,
1195                                         LUSTRE_MDS_VERSION,
1196                                         MDS_HSM_CT_REGISTER);
1197         if (req == NULL) {
1198                 rc = -ENOMEM;
1199                 goto out;
1200         }
1201
1202         mdc_pack_body(req, NULL, OBD_MD_FLRMTPERM, 0, 0, 0);
1203
1204         /* Copy hsm_progress struct */
1205         archive_mask = req_capsule_client_get(&req->rq_pill,
1206                                               &RMF_MDS_HSM_ARCHIVE);
1207         if (archive_mask == NULL) {
1208                 rc = -EPROTO;
1209                 goto out;
1210         }
1211
1212         *archive_mask = archives;
1213
1214         ptlrpc_request_set_replen(req);
1215
1216         rc = mdc_queue_wait(req);
1217         goto out;
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         goto out;
1284 out:
1285         ptlrpc_req_finished(req);
1286         return rc;
1287 }
1288
1289 static int mdc_ioc_hsm_state_get(struct obd_export *exp,
1290                                  struct md_op_data *op_data)
1291 {
1292         struct hsm_user_state   *hus = op_data->op_data;
1293         struct hsm_user_state   *req_hus;
1294         struct ptlrpc_request   *req;
1295         int                      rc;
1296
1297         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1298                                    &RQF_MDS_HSM_STATE_GET);
1299         if (req == NULL)
1300                 return -ENOMEM;
1301
1302         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_STATE_GET);
1303         if (rc != 0) {
1304                 ptlrpc_request_free(req);
1305                 return rc;
1306         }
1307
1308         mdc_pack_body(req, &op_data->op_fid1, OBD_MD_FLRMTPERM, 0,
1309                       op_data->op_suppgids[0], 0);
1310
1311         ptlrpc_request_set_replen(req);
1312
1313         rc = mdc_queue_wait(req);
1314         if (rc)
1315                 goto out;
1316
1317         req_hus = req_capsule_server_get(&req->rq_pill, &RMF_HSM_USER_STATE);
1318         if (req_hus == NULL) {
1319                 rc = -EPROTO;
1320                 goto out;
1321         }
1322
1323         *hus = *req_hus;
1324
1325 out:
1326         ptlrpc_req_finished(req);
1327         return rc;
1328 }
1329
1330 static int mdc_ioc_hsm_state_set(struct obd_export *exp,
1331                                  struct md_op_data *op_data)
1332 {
1333         struct hsm_state_set    *hss = op_data->op_data;
1334         struct hsm_state_set    *req_hss;
1335         struct ptlrpc_request   *req;
1336         int                      rc;
1337
1338         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1339                                    &RQF_MDS_HSM_STATE_SET);
1340         if (req == NULL)
1341                 return -ENOMEM;
1342
1343         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_STATE_SET);
1344         if (rc) {
1345                 ptlrpc_request_free(req);
1346                 return rc;
1347         }
1348
1349         mdc_pack_body(req, &op_data->op_fid1, OBD_MD_FLRMTPERM, 0,
1350                       op_data->op_suppgids[0], 0);
1351
1352         /* Copy states */
1353         req_hss = req_capsule_client_get(&req->rq_pill, &RMF_HSM_STATE_SET);
1354         if (req_hss == NULL) {
1355                 rc = -EPROTO;
1356                 goto out;
1357         }
1358         *req_hss = *hss;
1359
1360         ptlrpc_request_set_replen(req);
1361
1362         rc = mdc_queue_wait(req);
1363         goto out;
1364
1365 out:
1366         ptlrpc_req_finished(req);
1367         return rc;
1368 }
1369
1370 static int mdc_ioc_hsm_request(struct obd_export *exp,
1371                                struct hsm_user_request *hur)
1372 {
1373         struct obd_import       *imp = class_exp2cliimp(exp);
1374         struct ptlrpc_request   *req;
1375         struct hsm_request      *req_hr;
1376         struct hsm_user_item    *req_hui;
1377         char                    *req_opaque;
1378         int                      rc;
1379
1380         req = ptlrpc_request_alloc(imp, &RQF_MDS_HSM_REQUEST);
1381         if (req == NULL) {
1382                 rc = -ENOMEM;
1383                 goto out;
1384         }
1385
1386         req_capsule_set_size(&req->rq_pill, &RMF_MDS_HSM_USER_ITEM, RCL_CLIENT,
1387                              hur->hur_request.hr_itemcount
1388                              * sizeof(struct hsm_user_item));
1389         req_capsule_set_size(&req->rq_pill, &RMF_GENERIC_DATA, RCL_CLIENT,
1390                              hur->hur_request.hr_data_len);
1391
1392         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_REQUEST);
1393         if (rc) {
1394                 ptlrpc_request_free(req);
1395                 return rc;
1396         }
1397
1398         mdc_pack_body(req, NULL, OBD_MD_FLRMTPERM, 0, 0, 0);
1399
1400         /* Copy hsm_request struct */
1401         req_hr = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_REQUEST);
1402         if (req_hr == NULL) {
1403                 rc = -EPROTO;
1404                 goto out;
1405         }
1406         *req_hr = hur->hur_request;
1407
1408         /* Copy hsm_user_item structs */
1409         req_hui = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_USER_ITEM);
1410         if (req_hui == NULL) {
1411                 rc = -EPROTO;
1412                 goto out;
1413         }
1414         memcpy(req_hui, hur->hur_user_item,
1415                hur->hur_request.hr_itemcount * sizeof(struct hsm_user_item));
1416
1417         /* Copy opaque field */
1418         req_opaque = req_capsule_client_get(&req->rq_pill, &RMF_GENERIC_DATA);
1419         if (req_opaque == NULL) {
1420                 rc = -EPROTO;
1421                 goto out;
1422         }
1423         memcpy(req_opaque, hur_data(hur), hur->hur_request.hr_data_len);
1424
1425         ptlrpc_request_set_replen(req);
1426
1427         rc = mdc_queue_wait(req);
1428         goto out;
1429
1430 out:
1431         ptlrpc_req_finished(req);
1432         return rc;
1433 }
1434
1435 static struct kuc_hdr *changelog_kuc_hdr(char *buf, int len, int flags)
1436 {
1437         struct kuc_hdr *lh = (struct kuc_hdr *)buf;
1438
1439         LASSERT(len <= KUC_CHANGELOG_MSG_MAXSIZE);
1440
1441         lh->kuc_magic = KUC_MAGIC;
1442         lh->kuc_transport = KUC_TRANSPORT_CHANGELOG;
1443         lh->kuc_flags = flags;
1444         lh->kuc_msgtype = CL_RECORD;
1445         lh->kuc_msglen = len;
1446         return lh;
1447 }
1448
1449 #define D_CHANGELOG 0
1450
1451 struct changelog_show {
1452         __u64           cs_startrec;
1453         __u32           cs_flags;
1454         struct file     *cs_fp;
1455         char            *cs_buf;
1456         struct obd_device *cs_obd;
1457 };
1458
1459 static int changelog_kkuc_cb(const struct lu_env *env, struct llog_handle *llh,
1460                              struct llog_rec_hdr *hdr, void *data)
1461 {
1462         struct changelog_show *cs = data;
1463         struct llog_changelog_rec *rec = (struct llog_changelog_rec *)hdr;
1464         struct kuc_hdr *lh;
1465         int len, rc;
1466
1467         if (rec->cr_hdr.lrh_type != CHANGELOG_REC) {
1468                 rc = -EINVAL;
1469                 CERROR("%s: not a changelog rec %x/%d: rc = %d\n",
1470                        cs->cs_obd->obd_name, rec->cr_hdr.lrh_type,
1471                        rec->cr.cr_type, rc);
1472                 return rc;
1473         }
1474
1475         if (rec->cr.cr_index < cs->cs_startrec) {
1476                 /* Skip entries earlier than what we are interested in */
1477                 CDEBUG(D_CHANGELOG, "rec=%llu start=%llu\n",
1478                        rec->cr.cr_index, cs->cs_startrec);
1479                 return 0;
1480         }
1481
1482         CDEBUG(D_CHANGELOG, "%llu %02d%-5s %llu 0x%x t="DFID" p="DFID
1483                 " %.*s\n", rec->cr.cr_index, rec->cr.cr_type,
1484                 changelog_type2str(rec->cr.cr_type), rec->cr.cr_time,
1485                 rec->cr.cr_flags & CLF_FLAGMASK,
1486                 PFID(&rec->cr.cr_tfid), PFID(&rec->cr.cr_pfid),
1487                 rec->cr.cr_namelen, changelog_rec_name(&rec->cr));
1488
1489         len = sizeof(*lh) + changelog_rec_size(&rec->cr) + rec->cr.cr_namelen;
1490
1491         /* Set up the message */
1492         lh = changelog_kuc_hdr(cs->cs_buf, len, cs->cs_flags);
1493         memcpy(lh + 1, &rec->cr, len - sizeof(*lh));
1494
1495         rc = libcfs_kkuc_msg_put(cs->cs_fp, lh);
1496         CDEBUG(D_CHANGELOG, "kucmsg fp %p len %d rc %d\n", cs->cs_fp, len, rc);
1497
1498         return rc;
1499 }
1500
1501 static int mdc_changelog_send_thread(void *csdata)
1502 {
1503         struct changelog_show *cs = csdata;
1504         struct llog_ctxt *ctxt = NULL;
1505         struct llog_handle *llh = NULL;
1506         struct kuc_hdr *kuch;
1507         int rc;
1508
1509         CDEBUG(D_CHANGELOG, "changelog to fp=%p start %llu\n",
1510                cs->cs_fp, cs->cs_startrec);
1511
1512         cs->cs_buf = kzalloc(KUC_CHANGELOG_MSG_MAXSIZE, GFP_NOFS);
1513         if (!cs->cs_buf) {
1514                 rc = -ENOMEM;
1515                 goto out;
1516         }
1517
1518         /* Set up the remote catalog handle */
1519         ctxt = llog_get_context(cs->cs_obd, LLOG_CHANGELOG_REPL_CTXT);
1520         if (ctxt == NULL) {
1521                 rc = -ENOENT;
1522                 goto out;
1523         }
1524         rc = llog_open(NULL, ctxt, &llh, NULL, CHANGELOG_CATALOG,
1525                        LLOG_OPEN_EXISTS);
1526         if (rc) {
1527                 CERROR("%s: fail to open changelog catalog: rc = %d\n",
1528                        cs->cs_obd->obd_name, rc);
1529                 goto out;
1530         }
1531         rc = llog_init_handle(NULL, llh, LLOG_F_IS_CAT, NULL);
1532         if (rc) {
1533                 CERROR("llog_init_handle failed %d\n", rc);
1534                 goto out;
1535         }
1536
1537         rc = llog_cat_process(NULL, llh, changelog_kkuc_cb, cs, 0, 0);
1538
1539         /* Send EOF no matter what our result */
1540         kuch = changelog_kuc_hdr(cs->cs_buf, sizeof(*kuch), cs->cs_flags);
1541         if (kuch) {
1542                 kuch->kuc_msgtype = CL_EOF;
1543                 libcfs_kkuc_msg_put(cs->cs_fp, kuch);
1544         }
1545
1546 out:
1547         fput(cs->cs_fp);
1548         if (llh)
1549                 llog_cat_close(NULL, llh);
1550         if (ctxt)
1551                 llog_ctxt_put(ctxt);
1552         kfree(cs->cs_buf);
1553         kfree(cs);
1554         return rc;
1555 }
1556
1557 static int mdc_ioc_changelog_send(struct obd_device *obd,
1558                                   struct ioc_changelog *icc)
1559 {
1560         struct changelog_show *cs;
1561         int rc;
1562
1563         /* Freed in mdc_changelog_send_thread */
1564         cs = kzalloc(sizeof(*cs), GFP_NOFS);
1565         if (!cs)
1566                 return -ENOMEM;
1567
1568         cs->cs_obd = obd;
1569         cs->cs_startrec = icc->icc_recno;
1570         /* matching fput in mdc_changelog_send_thread */
1571         cs->cs_fp = fget(icc->icc_id);
1572         cs->cs_flags = icc->icc_flags;
1573
1574         /*
1575          * New thread because we should return to user app before
1576          * writing into our pipe
1577          */
1578         rc = PTR_ERR(kthread_run(mdc_changelog_send_thread, cs,
1579                                  "mdc_clg_send_thread"));
1580         if (!IS_ERR_VALUE(rc)) {
1581                 CDEBUG(D_CHANGELOG, "start changelog thread\n");
1582                 return 0;
1583         }
1584
1585         CERROR("Failed to start changelog thread: %d\n", rc);
1586         kfree(cs);
1587         return rc;
1588 }
1589
1590 static int mdc_ioc_hsm_ct_start(struct obd_export *exp,
1591                                 struct lustre_kernelcomm *lk);
1592
1593 static int mdc_quotacheck(struct obd_device *unused, struct obd_export *exp,
1594                           struct obd_quotactl *oqctl)
1595 {
1596         struct client_obd       *cli = &exp->exp_obd->u.cli;
1597         struct ptlrpc_request   *req;
1598         struct obd_quotactl     *body;
1599         int                   rc;
1600
1601         req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp),
1602                                         &RQF_MDS_QUOTACHECK, LUSTRE_MDS_VERSION,
1603                                         MDS_QUOTACHECK);
1604         if (req == NULL)
1605                 return -ENOMEM;
1606
1607         body = req_capsule_client_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
1608         *body = *oqctl;
1609
1610         ptlrpc_request_set_replen(req);
1611
1612         /* the next poll will find -ENODATA, that means quotacheck is
1613          * going on */
1614         cli->cl_qchk_stat = -ENODATA;
1615         rc = ptlrpc_queue_wait(req);
1616         if (rc)
1617                 cli->cl_qchk_stat = rc;
1618         ptlrpc_req_finished(req);
1619         return rc;
1620 }
1621
1622 static int mdc_quota_poll_check(struct obd_export *exp,
1623                                 struct if_quotacheck *qchk)
1624 {
1625         struct client_obd *cli = &exp->exp_obd->u.cli;
1626         int rc;
1627
1628         qchk->obd_uuid = cli->cl_target_uuid;
1629         memcpy(qchk->obd_type, LUSTRE_MDS_NAME, strlen(LUSTRE_MDS_NAME));
1630
1631         rc = cli->cl_qchk_stat;
1632         /* the client is not the previous one */
1633         if (rc == CL_NOT_QUOTACHECKED)
1634                 rc = -EINTR;
1635         return rc;
1636 }
1637
1638 static int mdc_quotactl(struct obd_device *unused, struct obd_export *exp,
1639                         struct obd_quotactl *oqctl)
1640 {
1641         struct ptlrpc_request   *req;
1642         struct obd_quotactl     *oqc;
1643         int                   rc;
1644
1645         req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp),
1646                                         &RQF_MDS_QUOTACTL, LUSTRE_MDS_VERSION,
1647                                         MDS_QUOTACTL);
1648         if (req == NULL)
1649                 return -ENOMEM;
1650
1651         oqc = req_capsule_client_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
1652         *oqc = *oqctl;
1653
1654         ptlrpc_request_set_replen(req);
1655         ptlrpc_at_set_req_timeout(req);
1656         req->rq_no_resend = 1;
1657
1658         rc = ptlrpc_queue_wait(req);
1659         if (rc)
1660                 CERROR("ptlrpc_queue_wait failed, rc: %d\n", rc);
1661
1662         if (req->rq_repmsg) {
1663                 oqc = req_capsule_server_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
1664                 if (oqc) {
1665                         *oqctl = *oqc;
1666                 } else if (!rc) {
1667                         CERROR("Can't unpack obd_quotactl\n");
1668                         rc = -EPROTO;
1669                 }
1670         } else if (!rc) {
1671                 CERROR("Can't unpack obd_quotactl\n");
1672                 rc = -EPROTO;
1673         }
1674         ptlrpc_req_finished(req);
1675
1676         return rc;
1677 }
1678
1679 static int mdc_ioc_swap_layouts(struct obd_export *exp,
1680                                 struct md_op_data *op_data)
1681 {
1682         LIST_HEAD(cancels);
1683         struct ptlrpc_request   *req;
1684         int                      rc, count;
1685         struct mdc_swap_layouts *msl, *payload;
1686
1687         msl = op_data->op_data;
1688
1689         /* When the MDT will get the MDS_SWAP_LAYOUTS RPC the
1690          * first thing it will do is to cancel the 2 layout
1691          * locks hold by this client.
1692          * So the client must cancel its layout locks on the 2 fids
1693          * with the request RPC to avoid extra RPC round trips
1694          */
1695         count = mdc_resource_get_unused(exp, &op_data->op_fid1, &cancels,
1696                                         LCK_CR, MDS_INODELOCK_LAYOUT);
1697         count += mdc_resource_get_unused(exp, &op_data->op_fid2, &cancels,
1698                                          LCK_CR, MDS_INODELOCK_LAYOUT);
1699
1700         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1701                                    &RQF_MDS_SWAP_LAYOUTS);
1702         if (req == NULL) {
1703                 ldlm_lock_list_put(&cancels, l_bl_ast, count);
1704                 return -ENOMEM;
1705         }
1706
1707         rc = mdc_prep_elc_req(exp, req, MDS_SWAP_LAYOUTS, &cancels, count);
1708         if (rc) {
1709                 ptlrpc_request_free(req);
1710                 return rc;
1711         }
1712
1713         mdc_swap_layouts_pack(req, op_data);
1714
1715         payload = req_capsule_client_get(&req->rq_pill, &RMF_SWAP_LAYOUTS);
1716         LASSERT(payload);
1717
1718         *payload = *msl;
1719
1720         ptlrpc_request_set_replen(req);
1721
1722         rc = ptlrpc_queue_wait(req);
1723
1724         ptlrpc_req_finished(req);
1725         return rc;
1726 }
1727
1728 static int mdc_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1729                          void *karg, void *uarg)
1730 {
1731         struct obd_device *obd = exp->exp_obd;
1732         struct obd_ioctl_data *data = karg;
1733         struct obd_import *imp = obd->u.cli.cl_import;
1734         int rc;
1735
1736         if (!try_module_get(THIS_MODULE)) {
1737                 CERROR("Can't get module. Is it alive?");
1738                 return -EINVAL;
1739         }
1740         switch (cmd) {
1741         case OBD_IOC_CHANGELOG_SEND:
1742                 rc = mdc_ioc_changelog_send(obd, karg);
1743                 goto out;
1744         case OBD_IOC_CHANGELOG_CLEAR: {
1745                 struct ioc_changelog *icc = karg;
1746                 struct changelog_setinfo cs = {
1747                         .cs_recno = icc->icc_recno,
1748                         .cs_id = icc->icc_id
1749                 };
1750
1751                 rc = obd_set_info_async(NULL, exp, strlen(KEY_CHANGELOG_CLEAR),
1752                                         KEY_CHANGELOG_CLEAR, sizeof(cs), &cs,
1753                                         NULL);
1754                 goto out;
1755         }
1756         case OBD_IOC_FID2PATH:
1757                 rc = mdc_ioc_fid2path(exp, karg);
1758                 goto out;
1759         case LL_IOC_HSM_CT_START:
1760                 rc = mdc_ioc_hsm_ct_start(exp, karg);
1761                 /* ignore if it was already registered on this MDS. */
1762                 if (rc == -EEXIST)
1763                         rc = 0;
1764                 goto out;
1765         case LL_IOC_HSM_PROGRESS:
1766                 rc = mdc_ioc_hsm_progress(exp, karg);
1767                 goto out;
1768         case LL_IOC_HSM_STATE_GET:
1769                 rc = mdc_ioc_hsm_state_get(exp, karg);
1770                 goto out;
1771         case LL_IOC_HSM_STATE_SET:
1772                 rc = mdc_ioc_hsm_state_set(exp, karg);
1773                 goto out;
1774         case LL_IOC_HSM_ACTION:
1775                 rc = mdc_ioc_hsm_current_action(exp, karg);
1776                 goto out;
1777         case LL_IOC_HSM_REQUEST:
1778                 rc = mdc_ioc_hsm_request(exp, karg);
1779                 goto out;
1780         case OBD_IOC_CLIENT_RECOVER:
1781                 rc = ptlrpc_recover_import(imp, data->ioc_inlbuf1, 0);
1782                 if (rc < 0)
1783                         goto out;
1784                 rc = 0;
1785                 goto out;
1786         case IOC_OSC_SET_ACTIVE:
1787                 rc = ptlrpc_set_import_active(imp, data->ioc_offset);
1788                 goto out;
1789         case OBD_IOC_POLL_QUOTACHECK:
1790                 rc = mdc_quota_poll_check(exp, (struct if_quotacheck *)karg);
1791                 goto out;
1792         case OBD_IOC_PING_TARGET:
1793                 rc = ptlrpc_obd_ping(obd);
1794                 goto out;
1795         /*
1796          * Normally IOC_OBD_STATFS, OBD_IOC_QUOTACTL iocontrol are handled by
1797          * LMV instead of MDC. But when the cluster is upgraded from 1.8,
1798          * there'd be no LMV layer thus we might be called here. Eventually
1799          * this code should be removed.
1800          * bz20731, LU-592.
1801          */
1802         case IOC_OBD_STATFS: {
1803                 struct obd_statfs stat_buf = {0};
1804
1805                 if (*((__u32 *) data->ioc_inlbuf2) != 0) {
1806                         rc = -ENODEV;
1807                         goto out;
1808                 }
1809
1810                 /* copy UUID */
1811                 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(obd),
1812                                  min_t(size_t, data->ioc_plen2,
1813                                                sizeof(struct obd_uuid)))) {
1814                         rc = -EFAULT;
1815                         goto out;
1816                 }
1817
1818                 rc = mdc_statfs(NULL, obd->obd_self_export, &stat_buf,
1819                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1820                                 0);
1821                 if (rc != 0)
1822                         goto out;
1823
1824                 if (copy_to_user(data->ioc_pbuf1, &stat_buf,
1825                                  min_t(size_t, data->ioc_plen1,
1826                                                sizeof(stat_buf)))) {
1827                         rc = -EFAULT;
1828                         goto out;
1829                 }
1830
1831                 rc = 0;
1832                 goto out;
1833         }
1834         case OBD_IOC_QUOTACTL: {
1835                 struct if_quotactl *qctl = karg;
1836                 struct obd_quotactl *oqctl;
1837
1838                 oqctl = kzalloc(sizeof(*oqctl), GFP_NOFS);
1839                 if (!oqctl) {
1840                         rc = -ENOMEM;
1841                         goto out;
1842                 }
1843
1844                 QCTL_COPY(oqctl, qctl);
1845                 rc = obd_quotactl(exp, oqctl);
1846                 if (rc == 0) {
1847                         QCTL_COPY(qctl, oqctl);
1848                         qctl->qc_valid = QC_MDTIDX;
1849                         qctl->obd_uuid = obd->u.cli.cl_target_uuid;
1850                 }
1851
1852                 kfree(oqctl);
1853                 goto out;
1854         }
1855         case LL_IOC_GET_CONNECT_FLAGS:
1856                 if (copy_to_user(uarg, exp_connect_flags_ptr(exp),
1857                                  sizeof(*exp_connect_flags_ptr(exp)))) {
1858                         rc = -EFAULT;
1859                         goto out;
1860                 }
1861
1862                 rc = 0;
1863                 goto out;
1864         case LL_IOC_LOV_SWAP_LAYOUTS:
1865                 rc = mdc_ioc_swap_layouts(exp, karg);
1866                 goto out;
1867         default:
1868                 CERROR("unrecognised ioctl: cmd = %#x\n", cmd);
1869                 rc = -ENOTTY;
1870                 goto out;
1871         }
1872 out:
1873         module_put(THIS_MODULE);
1874
1875         return rc;
1876 }
1877
1878 static int mdc_get_info_rpc(struct obd_export *exp,
1879                             u32 keylen, void *key,
1880                             int vallen, void *val)
1881 {
1882         struct obd_import      *imp = class_exp2cliimp(exp);
1883         struct ptlrpc_request  *req;
1884         char               *tmp;
1885         int                  rc = -EINVAL;
1886
1887         req = ptlrpc_request_alloc(imp, &RQF_MDS_GET_INFO);
1888         if (req == NULL)
1889                 return -ENOMEM;
1890
1891         req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_KEY,
1892                              RCL_CLIENT, keylen);
1893         req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_VALLEN,
1894                              RCL_CLIENT, sizeof(__u32));
1895
1896         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GET_INFO);
1897         if (rc) {
1898                 ptlrpc_request_free(req);
1899                 return rc;
1900         }
1901
1902         tmp = req_capsule_client_get(&req->rq_pill, &RMF_GETINFO_KEY);
1903         memcpy(tmp, key, keylen);
1904         tmp = req_capsule_client_get(&req->rq_pill, &RMF_GETINFO_VALLEN);
1905         memcpy(tmp, &vallen, sizeof(__u32));
1906
1907         req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_VAL,
1908                              RCL_SERVER, vallen);
1909         ptlrpc_request_set_replen(req);
1910
1911         rc = ptlrpc_queue_wait(req);
1912         /* -EREMOTE means the get_info result is partial, and it needs to
1913          * continue on another MDT, see fid2path part in lmv_iocontrol */
1914         if (rc == 0 || rc == -EREMOTE) {
1915                 tmp = req_capsule_server_get(&req->rq_pill, &RMF_GETINFO_VAL);
1916                 memcpy(val, tmp, vallen);
1917                 if (ptlrpc_rep_need_swab(req)) {
1918                         if (KEY_IS(KEY_FID2PATH))
1919                                 lustre_swab_fid2path(val);
1920                 }
1921         }
1922         ptlrpc_req_finished(req);
1923
1924         return rc;
1925 }
1926
1927 static void lustre_swab_hai(struct hsm_action_item *h)
1928 {
1929         __swab32s(&h->hai_len);
1930         __swab32s(&h->hai_action);
1931         lustre_swab_lu_fid(&h->hai_fid);
1932         lustre_swab_lu_fid(&h->hai_dfid);
1933         __swab64s(&h->hai_cookie);
1934         __swab64s(&h->hai_extent.offset);
1935         __swab64s(&h->hai_extent.length);
1936         __swab64s(&h->hai_gid);
1937 }
1938
1939 static void lustre_swab_hal(struct hsm_action_list *h)
1940 {
1941         struct hsm_action_item  *hai;
1942         int                      i;
1943
1944         __swab32s(&h->hal_version);
1945         __swab32s(&h->hal_count);
1946         __swab32s(&h->hal_archive_id);
1947         __swab64s(&h->hal_flags);
1948         hai = hai_zero(h);
1949         for (i = 0; i < h->hal_count; i++, hai = hai_next(hai))
1950                 lustre_swab_hai(hai);
1951 }
1952
1953 static void lustre_swab_kuch(struct kuc_hdr *l)
1954 {
1955         __swab16s(&l->kuc_magic);
1956         /* __u8 l->kuc_transport */
1957         __swab16s(&l->kuc_msgtype);
1958         __swab16s(&l->kuc_msglen);
1959 }
1960
1961 static int mdc_ioc_hsm_ct_start(struct obd_export *exp,
1962                                 struct lustre_kernelcomm *lk)
1963 {
1964         struct obd_import  *imp = class_exp2cliimp(exp);
1965         __u32               archive = lk->lk_data;
1966         int                 rc = 0;
1967
1968         if (lk->lk_group != KUC_GRP_HSM) {
1969                 CERROR("Bad copytool group %d\n", lk->lk_group);
1970                 return -EINVAL;
1971         }
1972
1973         CDEBUG(D_HSM, "CT start r%d w%d u%d g%d f%#x\n", lk->lk_rfd, lk->lk_wfd,
1974                lk->lk_uid, lk->lk_group, lk->lk_flags);
1975
1976         if (lk->lk_flags & LK_FLG_STOP) {
1977                 /* Unregister with the coordinator */
1978                 rc = mdc_ioc_hsm_ct_unregister(imp);
1979         } else {
1980                 rc = mdc_ioc_hsm_ct_register(imp, archive);
1981         }
1982
1983         return rc;
1984 }
1985
1986 /**
1987  * Send a message to any listening copytools
1988  * @param val KUC message (kuc_hdr + hsm_action_list)
1989  * @param len total length of message
1990  */
1991 static int mdc_hsm_copytool_send(int len, void *val)
1992 {
1993         struct kuc_hdr          *lh = (struct kuc_hdr *)val;
1994         struct hsm_action_list  *hal = (struct hsm_action_list *)(lh + 1);
1995
1996         if (len < sizeof(*lh) + sizeof(*hal)) {
1997                 CERROR("Short HSM message %d < %d\n", len,
1998                        (int) (sizeof(*lh) + sizeof(*hal)));
1999                 return -EPROTO;
2000         }
2001         if (lh->kuc_magic == __swab16(KUC_MAGIC)) {
2002                 lustre_swab_kuch(lh);
2003                 lustre_swab_hal(hal);
2004         } else if (lh->kuc_magic != KUC_MAGIC) {
2005                 CERROR("Bad magic %x!=%x\n", lh->kuc_magic, KUC_MAGIC);
2006                 return -EPROTO;
2007         }
2008
2009         CDEBUG(D_HSM,
2010                "Received message mg=%x t=%d m=%d l=%d actions=%d on %s\n",
2011                lh->kuc_magic, lh->kuc_transport, lh->kuc_msgtype,
2012                lh->kuc_msglen, hal->hal_count, hal->hal_fsname);
2013
2014         /* Broadcast to HSM listeners */
2015         return libcfs_kkuc_group_put(KUC_GRP_HSM, lh);
2016 }
2017
2018 /**
2019  * callback function passed to kuc for re-registering each HSM copytool
2020  * running on MDC, after MDT shutdown/recovery.
2021  * @param data archive id served by the copytool
2022  * @param cb_arg callback argument (obd_import)
2023  */
2024 static int mdc_hsm_ct_reregister(__u32 data, void *cb_arg)
2025 {
2026         struct obd_import       *imp = (struct obd_import *)cb_arg;
2027         __u32                    archive = data;
2028         int                      rc;
2029
2030         CDEBUG(D_HA, "recover copytool registration to MDT (archive=%#x)\n",
2031                archive);
2032         rc = mdc_ioc_hsm_ct_register(imp, archive);
2033
2034         /* ignore error if the copytool is already registered */
2035         return ((rc != 0) && (rc != -EEXIST)) ? rc : 0;
2036 }
2037
2038 /**
2039  * Re-establish all kuc contexts with MDT
2040  * after MDT shutdown/recovery.
2041  */
2042 static int mdc_kuc_reregister(struct obd_import *imp)
2043 {
2044         /* re-register HSM agents */
2045         return libcfs_kkuc_group_foreach(KUC_GRP_HSM, mdc_hsm_ct_reregister,
2046                                          (void *)imp);
2047 }
2048
2049 static int mdc_set_info_async(const struct lu_env *env,
2050                               struct obd_export *exp,
2051                               u32 keylen, void *key,
2052                               u32 vallen, void *val,
2053                               struct ptlrpc_request_set *set)
2054 {
2055         struct obd_import       *imp = class_exp2cliimp(exp);
2056         int                      rc;
2057
2058         if (KEY_IS(KEY_READ_ONLY)) {
2059                 if (vallen != sizeof(int))
2060                         return -EINVAL;
2061
2062                 spin_lock(&imp->imp_lock);
2063                 if (*((int *)val)) {
2064                         imp->imp_connect_flags_orig |= OBD_CONNECT_RDONLY;
2065                         imp->imp_connect_data.ocd_connect_flags |=
2066                                                         OBD_CONNECT_RDONLY;
2067                 } else {
2068                         imp->imp_connect_flags_orig &= ~OBD_CONNECT_RDONLY;
2069                         imp->imp_connect_data.ocd_connect_flags &=
2070                                                         ~OBD_CONNECT_RDONLY;
2071                 }
2072                 spin_unlock(&imp->imp_lock);
2073
2074                 rc = do_set_info_async(imp, MDS_SET_INFO, LUSTRE_MDS_VERSION,
2075                                        keylen, key, vallen, val, set);
2076                 return rc;
2077         }
2078         if (KEY_IS(KEY_SPTLRPC_CONF)) {
2079                 sptlrpc_conf_client_adapt(exp->exp_obd);
2080                 return 0;
2081         }
2082         if (KEY_IS(KEY_FLUSH_CTX)) {
2083                 sptlrpc_import_flush_my_ctx(imp);
2084                 return 0;
2085         }
2086         if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
2087                 rc = do_set_info_async(imp, MDS_SET_INFO, LUSTRE_MDS_VERSION,
2088                                        keylen, key, vallen, val, set);
2089                 return rc;
2090         }
2091         if (KEY_IS(KEY_HSM_COPYTOOL_SEND)) {
2092                 rc = mdc_hsm_copytool_send(vallen, val);
2093                 return rc;
2094         }
2095
2096         CERROR("Unknown key %s\n", (char *)key);
2097         return -EINVAL;
2098 }
2099
2100 static int mdc_get_info(const struct lu_env *env, struct obd_export *exp,
2101                         __u32 keylen, void *key, __u32 *vallen, void *val,
2102                         struct lov_stripe_md *lsm)
2103 {
2104         int rc = -EINVAL;
2105
2106         if (KEY_IS(KEY_MAX_EASIZE)) {
2107                 int mdsize, *max_easize;
2108
2109                 if (*vallen != sizeof(int))
2110                         return -EINVAL;
2111                 mdsize = *(int *)val;
2112                 if (mdsize > exp->exp_obd->u.cli.cl_max_mds_easize)
2113                         exp->exp_obd->u.cli.cl_max_mds_easize = mdsize;
2114                 max_easize = val;
2115                 *max_easize = exp->exp_obd->u.cli.cl_max_mds_easize;
2116                 return 0;
2117         } else if (KEY_IS(KEY_DEFAULT_EASIZE)) {
2118                 int *default_easize;
2119
2120                 if (*vallen != sizeof(int))
2121                         return -EINVAL;
2122                 default_easize = val;
2123                 *default_easize = exp->exp_obd->u.cli.cl_default_mds_easize;
2124                 return 0;
2125         } else if (KEY_IS(KEY_CONN_DATA)) {
2126                 struct obd_import *imp = class_exp2cliimp(exp);
2127                 struct obd_connect_data *data = val;
2128
2129                 if (*vallen != sizeof(*data))
2130                         return -EINVAL;
2131
2132                 *data = imp->imp_connect_data;
2133                 return 0;
2134         } else if (KEY_IS(KEY_TGT_COUNT)) {
2135                 *((int *)val) = 1;
2136                 return 0;
2137         }
2138
2139         rc = mdc_get_info_rpc(exp, keylen, key, *vallen, val);
2140
2141         return rc;
2142 }
2143
2144 static int mdc_sync(struct obd_export *exp, const struct lu_fid *fid,
2145                     struct ptlrpc_request **request)
2146 {
2147         struct ptlrpc_request *req;
2148         int                 rc;
2149
2150         *request = NULL;
2151         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_SYNC);
2152         if (req == NULL)
2153                 return -ENOMEM;
2154
2155         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_SYNC);
2156         if (rc) {
2157                 ptlrpc_request_free(req);
2158                 return rc;
2159         }
2160
2161         mdc_pack_body(req, fid, 0, 0, -1, 0);
2162
2163         ptlrpc_request_set_replen(req);
2164
2165         rc = ptlrpc_queue_wait(req);
2166         if (rc)
2167                 ptlrpc_req_finished(req);
2168         else
2169                 *request = req;
2170         return rc;
2171 }
2172
2173 static int mdc_import_event(struct obd_device *obd, struct obd_import *imp,
2174                             enum obd_import_event event)
2175 {
2176         int rc = 0;
2177
2178         LASSERT(imp->imp_obd == obd);
2179
2180         switch (event) {
2181         case IMP_EVENT_DISCON: {
2182 #if 0
2183                 /* XXX Pass event up to OBDs stack. used only for FLD now */
2184                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_DISCON, NULL);
2185 #endif
2186                 break;
2187         }
2188         case IMP_EVENT_INACTIVE: {
2189                 struct client_obd *cli = &obd->u.cli;
2190                 /*
2191                  * Flush current sequence to make client obtain new one
2192                  * from server in case of disconnect/reconnect.
2193                  */
2194                 if (cli->cl_seq != NULL)
2195                         seq_client_flush(cli->cl_seq);
2196
2197                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_INACTIVE, NULL);
2198                 break;
2199         }
2200         case IMP_EVENT_INVALIDATE: {
2201                 struct ldlm_namespace *ns = obd->obd_namespace;
2202
2203                 ldlm_namespace_cleanup(ns, LDLM_FL_LOCAL_ONLY);
2204
2205                 break;
2206         }
2207         case IMP_EVENT_ACTIVE:
2208                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_ACTIVE, NULL);
2209                 /* redo the kuc registration after reconnecting */
2210                 if (rc == 0)
2211                         rc = mdc_kuc_reregister(imp);
2212                 break;
2213         case IMP_EVENT_OCD:
2214                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_OCD, NULL);
2215                 break;
2216         case IMP_EVENT_DEACTIVATE:
2217         case IMP_EVENT_ACTIVATE:
2218                 break;
2219         default:
2220                 CERROR("Unknown import event %x\n", event);
2221                 LBUG();
2222         }
2223         return rc;
2224 }
2225
2226 int mdc_fid_alloc(struct obd_export *exp, struct lu_fid *fid,
2227                   struct md_op_data *op_data)
2228 {
2229         struct client_obd *cli = &exp->exp_obd->u.cli;
2230         struct lu_client_seq *seq = cli->cl_seq;
2231
2232         return seq_client_alloc_fid(NULL, seq, fid);
2233 }
2234
2235 static struct obd_uuid *mdc_get_uuid(struct obd_export *exp)
2236 {
2237         struct client_obd *cli = &exp->exp_obd->u.cli;
2238
2239         return &cli->cl_target_uuid;
2240 }
2241
2242 /**
2243  * Determine whether the lock can be canceled before replaying it during
2244  * recovery, non zero value will be return if the lock can be canceled,
2245  * or zero returned for not
2246  */
2247 static int mdc_cancel_for_recovery(struct ldlm_lock *lock)
2248 {
2249         if (lock->l_resource->lr_type != LDLM_IBITS)
2250                 return 0;
2251
2252         /* FIXME: if we ever get into a situation where there are too many
2253          * opened files with open locks on a single node, then we really
2254          * should replay these open locks to reget it */
2255         if (lock->l_policy_data.l_inodebits.bits & MDS_INODELOCK_OPEN)
2256                 return 0;
2257
2258         return 1;
2259 }
2260
2261 static int mdc_resource_inode_free(struct ldlm_resource *res)
2262 {
2263         if (res->lr_lvb_inode)
2264                 res->lr_lvb_inode = NULL;
2265
2266         return 0;
2267 }
2268
2269 static struct ldlm_valblock_ops inode_lvbo = {
2270         .lvbo_free = mdc_resource_inode_free,
2271 };
2272
2273 static int mdc_llog_init(struct obd_device *obd)
2274 {
2275         struct obd_llog_group   *olg = &obd->obd_olg;
2276         struct llog_ctxt        *ctxt;
2277         int                      rc;
2278
2279         rc = llog_setup(NULL, obd, olg, LLOG_CHANGELOG_REPL_CTXT, obd,
2280                         &llog_client_ops);
2281         if (rc)
2282                 return rc;
2283
2284         ctxt = llog_group_get_ctxt(olg, LLOG_CHANGELOG_REPL_CTXT);
2285         llog_initiator_connect(ctxt);
2286         llog_ctxt_put(ctxt);
2287
2288         return 0;
2289 }
2290
2291 static void mdc_llog_finish(struct obd_device *obd)
2292 {
2293         struct llog_ctxt *ctxt;
2294
2295         ctxt = llog_get_context(obd, LLOG_CHANGELOG_REPL_CTXT);
2296         if (ctxt)
2297                 llog_cleanup(NULL, ctxt);
2298 }
2299
2300 static int mdc_setup(struct obd_device *obd, struct lustre_cfg *cfg)
2301 {
2302         struct client_obd *cli = &obd->u.cli;
2303         struct lprocfs_static_vars lvars = { NULL };
2304         int rc;
2305
2306         cli->cl_rpc_lock = kzalloc(sizeof(*cli->cl_rpc_lock), GFP_NOFS);
2307         if (!cli->cl_rpc_lock)
2308                 return -ENOMEM;
2309         mdc_init_rpc_lock(cli->cl_rpc_lock);
2310
2311         ptlrpcd_addref();
2312
2313         cli->cl_close_lock = kzalloc(sizeof(*cli->cl_close_lock), GFP_NOFS);
2314         if (!cli->cl_close_lock) {
2315                 rc = -ENOMEM;
2316                 goto err_rpc_lock;
2317         }
2318         mdc_init_rpc_lock(cli->cl_close_lock);
2319
2320         rc = client_obd_setup(obd, cfg);
2321         if (rc)
2322                 goto err_close_lock;
2323         lprocfs_mdc_init_vars(&lvars);
2324         lprocfs_obd_setup(obd, lvars.obd_vars, lvars.sysfs_vars);
2325         sptlrpc_lprocfs_cliobd_attach(obd);
2326         ptlrpc_lprocfs_register_obd(obd);
2327
2328         ns_register_cancel(obd->obd_namespace, mdc_cancel_for_recovery);
2329
2330         obd->obd_namespace->ns_lvbo = &inode_lvbo;
2331
2332         rc = mdc_llog_init(obd);
2333         if (rc) {
2334                 mdc_cleanup(obd);
2335                 CERROR("failed to setup llogging subsystems\n");
2336         }
2337
2338         return rc;
2339
2340 err_close_lock:
2341         kfree(cli->cl_close_lock);
2342 err_rpc_lock:
2343         kfree(cli->cl_rpc_lock);
2344         ptlrpcd_decref();
2345         return rc;
2346 }
2347
2348 /* Initialize the default and maximum LOV EA and cookie sizes.  This allows
2349  * us to make MDS RPCs with large enough reply buffers to hold a default
2350  * sized EA and cookie without having to calculate this (via a call into the
2351  * LOV + OSCs) each time we make an RPC.  The maximum size is also tracked
2352  * but not used to avoid wastefully vmalloc()'ing large reply buffers when
2353  * a large number of stripes is possible.  If a larger reply buffer is
2354  * required it will be reallocated in the ptlrpc layer due to overflow.
2355  */
2356 static int mdc_init_ea_size(struct obd_export *exp, int easize,
2357                             int def_easize, int cookiesize, int def_cookiesize)
2358 {
2359         struct obd_device *obd = exp->exp_obd;
2360         struct client_obd *cli = &obd->u.cli;
2361
2362         if (cli->cl_max_mds_easize < easize)
2363                 cli->cl_max_mds_easize = easize;
2364
2365         if (cli->cl_default_mds_easize < def_easize)
2366                 cli->cl_default_mds_easize = def_easize;
2367
2368         if (cli->cl_max_mds_cookiesize < cookiesize)
2369                 cli->cl_max_mds_cookiesize = cookiesize;
2370
2371         if (cli->cl_default_mds_cookiesize < def_cookiesize)
2372                 cli->cl_default_mds_cookiesize = def_cookiesize;
2373
2374         return 0;
2375 }
2376
2377 static int mdc_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
2378 {
2379         switch (stage) {
2380         case OBD_CLEANUP_EARLY:
2381                 break;
2382         case OBD_CLEANUP_EXPORTS:
2383                 /* Failsafe, ok if racy */
2384                 if (obd->obd_type->typ_refcnt <= 1)
2385                         libcfs_kkuc_group_rem(0, KUC_GRP_HSM);
2386
2387                 obd_cleanup_client_import(obd);
2388                 ptlrpc_lprocfs_unregister_obd(obd);
2389                 lprocfs_obd_cleanup(obd);
2390
2391                 mdc_llog_finish(obd);
2392                 break;
2393         }
2394         return 0;
2395 }
2396
2397 static int mdc_cleanup(struct obd_device *obd)
2398 {
2399         struct client_obd *cli = &obd->u.cli;
2400
2401         kfree(cli->cl_rpc_lock);
2402         kfree(cli->cl_close_lock);
2403
2404         ptlrpcd_decref();
2405
2406         return client_obd_cleanup(obd);
2407 }
2408
2409 static int mdc_process_config(struct obd_device *obd, u32 len, void *buf)
2410 {
2411         struct lustre_cfg *lcfg = buf;
2412         struct lprocfs_static_vars lvars = { NULL };
2413         int rc = 0;
2414
2415         lprocfs_mdc_init_vars(&lvars);
2416         switch (lcfg->lcfg_command) {
2417         default:
2418                 rc = class_process_proc_param(PARAM_MDC, lvars.obd_vars,
2419                                               lcfg, obd);
2420                 if (rc > 0)
2421                         rc = 0;
2422                 break;
2423         }
2424         return rc;
2425 }
2426
2427 /* get remote permission for current user on fid */
2428 static int mdc_get_remote_perm(struct obd_export *exp, const struct lu_fid *fid,
2429                                __u32 suppgid, struct ptlrpc_request **request)
2430 {
2431         struct ptlrpc_request  *req;
2432         int                 rc;
2433
2434         LASSERT(client_is_remote(exp));
2435
2436         *request = NULL;
2437         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_GETATTR);
2438         if (req == NULL)
2439                 return -ENOMEM;
2440
2441         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR);
2442         if (rc) {
2443                 ptlrpc_request_free(req);
2444                 return rc;
2445         }
2446
2447         mdc_pack_body(req, fid, OBD_MD_FLRMTPERM, 0, suppgid, 0);
2448
2449         req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER,
2450                              sizeof(struct mdt_remote_perm));
2451
2452         ptlrpc_request_set_replen(req);
2453
2454         rc = ptlrpc_queue_wait(req);
2455         if (rc)
2456                 ptlrpc_req_finished(req);
2457         else
2458                 *request = req;
2459         return rc;
2460 }
2461
2462 static struct obd_ops mdc_obd_ops = {
2463         .o_owner            = THIS_MODULE,
2464         .o_setup            = mdc_setup,
2465         .o_precleanup       = mdc_precleanup,
2466         .o_cleanup        = mdc_cleanup,
2467         .o_add_conn      = client_import_add_conn,
2468         .o_del_conn      = client_import_del_conn,
2469         .o_connect          = client_connect_import,
2470         .o_disconnect       = client_disconnect_export,
2471         .o_iocontrol    = mdc_iocontrol,
2472         .o_set_info_async   = mdc_set_info_async,
2473         .o_statfs          = mdc_statfs,
2474         .o_fid_init         = client_fid_init,
2475         .o_fid_fini         = client_fid_fini,
2476         .o_fid_alloc    = mdc_fid_alloc,
2477         .o_import_event     = mdc_import_event,
2478         .o_get_info      = mdc_get_info,
2479         .o_process_config   = mdc_process_config,
2480         .o_get_uuid      = mdc_get_uuid,
2481         .o_quotactl      = mdc_quotactl,
2482         .o_quotacheck       = mdc_quotacheck
2483 };
2484
2485 static struct md_ops mdc_md_ops = {
2486         .m_getstatus    = mdc_getstatus,
2487         .m_null_inode       = mdc_null_inode,
2488         .m_find_cbdata      = mdc_find_cbdata,
2489         .m_close            = mdc_close,
2490         .m_create          = mdc_create,
2491         .m_done_writing     = mdc_done_writing,
2492         .m_enqueue        = mdc_enqueue,
2493         .m_getattr        = mdc_getattr,
2494         .m_getattr_name     = mdc_getattr_name,
2495         .m_intent_lock      = mdc_intent_lock,
2496         .m_link      = mdc_link,
2497         .m_is_subdir    = mdc_is_subdir,
2498         .m_rename          = mdc_rename,
2499         .m_setattr        = mdc_setattr,
2500         .m_setxattr      = mdc_setxattr,
2501         .m_getxattr      = mdc_getxattr,
2502         .m_sync      = mdc_sync,
2503         .m_readpage      = mdc_readpage,
2504         .m_unlink          = mdc_unlink,
2505         .m_cancel_unused    = mdc_cancel_unused,
2506         .m_init_ea_size     = mdc_init_ea_size,
2507         .m_set_lock_data    = mdc_set_lock_data,
2508         .m_lock_match       = mdc_lock_match,
2509         .m_get_lustre_md    = mdc_get_lustre_md,
2510         .m_free_lustre_md   = mdc_free_lustre_md,
2511         .m_set_open_replay_data = mdc_set_open_replay_data,
2512         .m_clear_open_replay_data = mdc_clear_open_replay_data,
2513         .m_get_remote_perm  = mdc_get_remote_perm,
2514         .m_intent_getattr_async = mdc_intent_getattr_async,
2515         .m_revalidate_lock      = mdc_revalidate_lock
2516 };
2517
2518 static int __init mdc_init(void)
2519 {
2520         struct lprocfs_static_vars lvars = { NULL };
2521
2522         lprocfs_mdc_init_vars(&lvars);
2523
2524         return class_register_type(&mdc_obd_ops, &mdc_md_ops,
2525                                  LUSTRE_MDC_NAME, NULL);
2526 }
2527
2528 static void /*__exit*/ mdc_exit(void)
2529 {
2530         class_unregister_type(LUSTRE_MDC_NAME);
2531 }
2532
2533 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
2534 MODULE_DESCRIPTION("Lustre Metadata Client");
2535 MODULE_LICENSE("GPL");
2536
2537 module_init(mdc_init);
2538 module_exit(mdc_exit);