lustre/mdc/mdc_request.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  * GPL HEADER START
   5  *
   6  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   7  *
   8  * This program is free software; you can redistribute it and/or modify
   9  * it under the terms of the GNU General Public License version 2 only,
  10  * as published by the Free Software Foundation.
  11  *
  12  * This program is distributed in the hope that it will be useful, but
  13  * WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * General Public License version 2 for more details (a copy is included
  16  * in the LICENSE file that accompanied this code).
  17  *
  18  * You should have received a copy of the GNU General Public License
  19  * version 2 along with this program; If not, see
  20  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  21  *
  22  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  23  * CA 95054 USA or visit www.sun.com if you need additional information or
  24  * have any questions.
  25  *
  26  * GPL HEADER END
  27  */
  28 /*
  29  * Copyright  2008 Sun Microsystems, Inc. All rights reserved
  30  * Use is subject to license terms.
  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 #ifndef EXPORT_SYMTAB
  38 # define EXPORT_SYMTAB
  39 #endif
  40 #define DEBUG_SUBSYSTEM S_MDC
  41
  42 #ifdef __KERNEL__
  43 # include <linux/module.h>
  44 # include <linux/pagemap.h>
  45 # include <linux/miscdevice.h>
  46 # include <linux/init.h>
  47 #else
  48 # include <liblustre.h>
  49 #endif
  50
  51 #include <lustre_acl.h>
  52 #include <obd_class.h>
  53 #include <lustre_dlm.h>
  54 #include <lustre_fid.h>
  55 #include <md_object.h>
  56 #include <lprocfs_status.h>
  57 #include <lustre_param.h>
  58 #include "mdc_internal.h"
  59 #include <lustre/lustre_idl.h>
  60
  61 #define REQUEST_MINOR 244
  62
  63 static quota_interface_t *quota_interface;
  64 extern quota_interface_t mdc_quota_interface;
  65
  66 static int mdc_cleanup(struct obd_device *obd);
  67
  68 int mdc_unpack_capa(struct obd_export *exp, struct ptlrpc_request *req,
  69                     const struct req_msg_field *field, struct obd_capa **oc)
  70 {
  71         struct lustre_capa *capa;
  72         struct obd_capa *c;
  73         ENTRY;
  74
  75         /* swabbed already in mdc_enqueue */
  76         capa = req_capsule_server_get(&req->rq_pill, field);
  77         if (capa == NULL)
  78                 RETURN(-EPROTO);
  79
  80         c = alloc_capa(CAPA_SITE_CLIENT);
  81         if (IS_ERR(c)) {
  82                 CDEBUG(D_INFO, "alloc capa failed!\n");
  83                 RETURN(PTR_ERR(c));
  84         } else {
  85                 c->c_capa = *capa;
  86                 *oc = c;
  87                 RETURN(0);
  88         }
  89 }
  90
  91 /* Helper that implements most of mdc_getstatus and signal_completed_replay. */
  92 /* XXX this should become mdc_get_info("key"), sending MDS_GET_INFO RPC */
  93 static int send_getstatus(struct obd_import *imp, struct lu_fid *rootfid,
  94                           struct obd_capa **pc, int level, int msg_flags)
  95 {
  96         struct ptlrpc_request *req;
  97         struct mdt_body       *body;
  98         int                    rc;
  99         ENTRY;
 100
 101         req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_GETSTATUS,
 102                                         LUSTRE_MDS_VERSION, MDS_GETSTATUS);
 103         if (req == NULL)
 104                 RETURN(-ENOMEM);
 105
 106         mdc_pack_body(req, NULL, NULL, 0, 0, -1, 0);
 107         lustre_msg_add_flags(req->rq_reqmsg, msg_flags);
 108         req->rq_send_state = level;
 109
 110         ptlrpc_request_set_replen(req);
 111
 112         rc = ptlrpc_queue_wait(req);
 113         if (rc)
 114                 GOTO(out, rc);
 115
 116         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 117         if (body == NULL)
 118                 GOTO(out, rc = -EPROTO);
 119
 120         if (body->valid & OBD_MD_FLMDSCAPA) {
 121                 rc = mdc_unpack_capa(NULL, req, &RMF_CAPA1, pc);
 122                 if (rc)
 123                         GOTO(out, rc);
 124         }
 125
 126         *rootfid = body->fid1;
 127         CDEBUG(D_NET,
 128                "root fid="DFID", last_committed="LPU64", last_xid="LPU64"\n",
 129                PFID(rootfid),
 130                lustre_msg_get_last_committed(req->rq_repmsg),
 131                lustre_msg_get_last_xid(req->rq_repmsg));
 132         EXIT;
 133 out:
 134         ptlrpc_req_finished(req);
 135         return rc;
 136 }
 137
 138 /* This should be mdc_get_info("rootfid") */
 139 int mdc_getstatus(struct obd_export *exp, struct lu_fid *rootfid,
 140                   struct obd_capa **pc)
 141 {
 142         return send_getstatus(class_exp2cliimp(exp), rootfid, pc,
 143                               LUSTRE_IMP_FULL, 0);
 144 }
 145
 146 /*
 147  * This function now is known to always saying that it will receive 4 buffers
 148  * from server. Even for cases when acl_size and md_size is zero, RPC header
 149  * will contain 4 fields and RPC itself will contain zero size fields. This is
 150  * because mdt_getattr*() _always_ returns 4 fields, but if acl is not needed
 151  * and thus zero, it shrinks it, making zero size. The same story about
 152  * md_size. And this is course of problem when client waits for smaller number
 153  * of fields. This issue will be fixed later when client gets aware of RPC
 154  * layouts.  --umka
 155  */
 156 static int mdc_getattr_common(struct obd_export *exp,
 157                               struct ptlrpc_request *req)
 158 {
 159         struct req_capsule *pill = &req->rq_pill;
 160         struct mdt_body    *body;
 161         void               *eadata;
 162         int                 rc;
 163         ENTRY;
 164
 165         /* Request message already built. */
 166         rc = ptlrpc_queue_wait(req);
 167         if (rc != 0)
 168                 RETURN(rc);
 169
 170         /* sanity check for the reply */
 171         body = req_capsule_server_get(pill, &RMF_MDT_BODY);
 172         if (body == NULL)
 173                 RETURN(-EPROTO);
 174
 175         CDEBUG(D_NET, "mode: %o\n", body->mode);
 176
 177         if (body->eadatasize != 0) {
 178                 eadata = req_capsule_server_sized_get(pill, &RMF_MDT_MD,
 179                                                       body->eadatasize);
 180                 if (eadata == NULL)
 181                         RETURN(-EPROTO);
 182         }
 183
 184         if (body->valid & OBD_MD_FLRMTPERM) {
 185                 struct mdt_remote_perm *perm;
 186
 187                 LASSERT(client_is_remote(exp));
 188                 perm = req_capsule_server_swab_get(pill, &RMF_ACL,
 189                                                 lustre_swab_mdt_remote_perm);
 190                 if (perm == NULL)
 191                         RETURN(-EPROTO);
 192         }
 193
 194         if (body->valid & OBD_MD_FLMDSCAPA) {
 195                 struct lustre_capa *capa;
 196                 capa = req_capsule_server_get(pill, &RMF_CAPA1);
 197                 if (capa == NULL)
 198                         RETURN(-EPROTO);
 199         }
 200
 201         RETURN(0);
 202 }
 203
 204 int mdc_getattr(struct obd_export *exp, const struct lu_fid *fid,
 205                 struct obd_capa *oc, obd_valid valid, int ea_size,
 206                 struct ptlrpc_request **request)
 207 {
 208         struct ptlrpc_request *req;
 209         int                    rc;
 210         ENTRY;
 211
 212         *request = NULL;
 213         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_GETATTR);
 214         if (req == NULL)
 215                 RETURN(-ENOMEM);
 216
 217         mdc_set_capa_size(req, &RMF_CAPA1, oc);
 218
 219         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR);
 220         if (rc) {
 221                 ptlrpc_request_free(req);
 222                 RETURN(rc);
 223         }
 224
 225         /* MDS_BFLAG_EXT_FLAGS: request "new" flags(bug 9486) */
 226         mdc_pack_body(req, fid, oc, valid, ea_size, -1, MDS_BFLAG_EXT_FLAGS);
 227
 228         req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER, ea_size);
 229         if (valid & OBD_MD_FLRMTPERM) {
 230                 LASSERT(client_is_remote(exp));
 231                 req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER,
 232                                      sizeof(struct mdt_remote_perm));
 233         }
 234         ptlrpc_request_set_replen(req);
 235
 236         rc = mdc_getattr_common(exp, req);
 237         if (rc)
 238                 ptlrpc_req_finished(req);
 239         else
 240                 *request = req;
 241         RETURN(rc);
 242 }
 243
 244 int mdc_getattr_name(struct obd_export *exp, const struct lu_fid *fid,
 245                      struct obd_capa *oc, const char *filename, int namelen,
 246                      obd_valid valid, int ea_size, __u32 suppgid,
 247                      struct ptlrpc_request **request)
 248 {
 249         struct ptlrpc_request *req;
 250         int                    rc;
 251         ENTRY;
 252
 253         *request = NULL;
 254         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
 255                                    &RQF_MDS_GETATTR_NAME);
 256         if (req == NULL)
 257                 RETURN(-ENOMEM);
 258
 259         mdc_set_capa_size(req, &RMF_CAPA1, oc);
 260         req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT, namelen);
 261
 262         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR_NAME);
 263         if (rc) {
 264                 ptlrpc_request_free(req);
 265                 RETURN(rc);
 266         }
 267
 268         /* MDS_BFLAG_EXT_FLAGS: request "new" flags(bug 9486) */
 269         mdc_pack_body(req, fid, oc, valid, ea_size, suppgid,
 270                       MDS_BFLAG_EXT_FLAGS);
 271
 272         if (filename) {
 273                 char *name = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
 274                 LASSERT(strnlen(filename, namelen) == namelen - 1);
 275                 memcpy(name, filename, namelen);
 276         }
 277
 278         req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER, ea_size);
 279         ptlrpc_request_set_replen(req);
 280
 281         rc = mdc_getattr_common(exp, req);
 282         if (rc)
 283                 ptlrpc_req_finished(req);
 284         else
 285                 *request = req;
 286         RETURN(rc);
 287 }
 288
 289 static int mdc_is_subdir(struct obd_export *exp,
 290                          const struct lu_fid *pfid,
 291                          const struct lu_fid *cfid,
 292                          struct ptlrpc_request **request)
 293 {
 294         struct ptlrpc_request  *req;
 295         int                     rc;
 296
 297         ENTRY;
 298
 299         *request = NULL;
 300         req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp),
 301                                         &RQF_MDS_IS_SUBDIR, LUSTRE_MDS_VERSION,
 302                                         MDS_IS_SUBDIR);
 303         if (req == NULL)
 304                 RETURN(-ENOMEM);
 305
 306         mdc_is_subdir_pack(req, pfid, cfid, 0);
 307         ptlrpc_request_set_replen(req);
 308
 309         rc = ptlrpc_queue_wait(req);
 310         if (rc && rc != -EREMOTE)
 311                 ptlrpc_req_finished(req);
 312         else
 313                 *request = req;
 314         RETURN(rc);
 315 }
 316
 317 static int mdc_xattr_common(struct obd_export *exp,const struct req_format *fmt,
 318                             const struct lu_fid *fid,
 319                             struct obd_capa *oc, int opcode, obd_valid valid,
 320                             const char *xattr_name, const char *input,
 321                             int input_size, int output_size, int flags,
 322                             __u32 suppgid, struct ptlrpc_request **request)
 323 {
 324         struct ptlrpc_request *req;
 325         int   xattr_namelen = 0;
 326         char *tmp;
 327         int   rc;
 328         ENTRY;
 329
 330         *request = NULL;
 331         req = ptlrpc_request_alloc(class_exp2cliimp(exp), fmt);
 332         if (req == NULL)
 333                 RETURN(-ENOMEM);
 334
 335         mdc_set_capa_size(req, &RMF_CAPA1, oc);
 336         if (xattr_name) {
 337                 xattr_namelen = strlen(xattr_name) + 1;
 338                 req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
 339                                      xattr_namelen);
 340         }
 341         if (input_size) {
 342                 LASSERT(input);
 343                 req_capsule_set_size(&req->rq_pill, &RMF_EADATA, RCL_CLIENT,
 344                                      input_size);
 345         }
 346
 347         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, opcode);
 348         if (rc) {
 349                 ptlrpc_request_free(req);
 350                 RETURN(rc);
 351         }
 352
 353         if (opcode == MDS_REINT) {
 354                 struct mdt_rec_setxattr *rec;
 355
 356                 CLASSERT(sizeof(struct mdt_rec_setxattr) ==
 357                          sizeof(struct mdt_rec_reint));
 358                 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
 359                 rec->sx_opcode = REINT_SETXATTR;
 360                 /* TODO:
 361                  *  cfs_curproc_fs{u,g}id() should replace
 362                  *  current->fs{u,g}id for portability.
 363                  */
 364                 rec->sx_fsuid  = current->fsuid;
 365                 rec->sx_fsgid  = current->fsgid;
 366                 rec->sx_cap    = cfs_curproc_cap_pack();
 367                 rec->sx_suppgid1 = suppgid;
 368                 rec->sx_suppgid2 = -1;
 369                 rec->sx_fid    = *fid;
 370                 rec->sx_valid  = valid | OBD_MD_FLCTIME;
 371                 rec->sx_time   = cfs_time_current_sec();
 372                 rec->sx_size   = output_size;
 373                 rec->sx_flags  = flags;
 374
 375                 mdc_pack_capa(req, &RMF_CAPA1, oc);
 376         } else {
 377                 mdc_pack_body(req, fid, oc, valid, output_size, suppgid, flags);
 378         }
 379
 380         if (xattr_name) {
 381                 tmp = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
 382                 memcpy(tmp, xattr_name, xattr_namelen);
 383         }
 384         if (input_size) {
 385                 tmp = req_capsule_client_get(&req->rq_pill, &RMF_EADATA);
 386                 memcpy(tmp, input, input_size);
 387         }
 388
 389         if (req_capsule_has_field(&req->rq_pill, &RMF_EADATA, RCL_SERVER))
 390                 req_capsule_set_size(&req->rq_pill, &RMF_EADATA,
 391                                      RCL_SERVER, output_size);
 392         ptlrpc_request_set_replen(req);
 393
 394         /* make rpc */
 395         if (opcode == MDS_REINT)
 396                 mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
 397
 398         rc = ptlrpc_queue_wait(req);
 399
 400         if (opcode == MDS_REINT)
 401                 mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
 402
 403         if (rc)
 404                 ptlrpc_req_finished(req);
 405         else
 406                 *request = req;
 407         RETURN(rc);
 408 }
 409
 410 int mdc_setxattr(struct obd_export *exp, const struct lu_fid *fid,
 411                  struct obd_capa *oc, obd_valid valid, const char *xattr_name,
 412                  const char *input, int input_size, int output_size,
 413                  int flags, __u32 suppgid, struct ptlrpc_request **request)
 414 {
 415         return mdc_xattr_common(exp, &RQF_MDS_REINT_SETXATTR,
 416                                 fid, oc, MDS_REINT, valid, xattr_name,
 417                                 input, input_size, output_size, flags,
 418                                 suppgid, request);
 419 }
 420
 421 int mdc_getxattr(struct obd_export *exp, const struct lu_fid *fid,
 422                  struct obd_capa *oc, obd_valid valid, const char *xattr_name,
 423                  const char *input, int input_size, int output_size,
 424                  int flags, struct ptlrpc_request **request)
 425 {
 426         return mdc_xattr_common(exp, &RQF_MDS_GETXATTR,
 427                                 fid, oc, MDS_GETXATTR, valid, xattr_name,
 428                                 input, input_size, output_size, flags,
 429                                 -1, request);
 430 }
 431
 432 #ifdef CONFIG_FS_POSIX_ACL
 433 static int mdc_unpack_acl(struct ptlrpc_request *req, struct lustre_md *md)
 434 {
 435         struct req_capsule     *pill = &req->rq_pill;
 436         struct mdt_body        *body = md->body;
 437         struct posix_acl       *acl;
 438         void                   *buf;
 439         int                     rc;
 440         ENTRY;
 441
 442         if (!body->aclsize)
 443                 RETURN(0);
 444
 445         buf = req_capsule_server_sized_get(pill, &RMF_ACL, body->aclsize);
 446
 447         if (!buf)
 448                 RETURN(-EPROTO);
 449
 450         acl = posix_acl_from_xattr(buf, body->aclsize);
 451         if (IS_ERR(acl)) {
 452                 rc = PTR_ERR(acl);
 453                 CERROR("convert xattr to acl: %d\n", rc);
 454                 RETURN(rc);
 455         }
 456
 457         rc = posix_acl_valid(acl);
 458         if (rc) {
 459                 CERROR("validate acl: %d\n", rc);
 460                 posix_acl_release(acl);
 461                 RETURN(rc);
 462         }
 463
 464         md->posix_acl = acl;
 465         RETURN(0);
 466 }
 467 #else
 468 #define mdc_unpack_acl(req, md) 0
 469 #endif
 470
 471 int mdc_get_lustre_md(struct obd_export *exp, struct ptlrpc_request *req,
 472                       struct obd_export *dt_exp, struct obd_export *md_exp,
 473                       struct lustre_md *md)
 474 {
 475         struct req_capsule *pill = &req->rq_pill;
 476         int rc;
 477         ENTRY;
 478
 479         LASSERT(md);
 480         memset(md, 0, sizeof(*md));
 481
 482         md->body = req_capsule_server_get(pill, &RMF_MDT_BODY);
 483         LASSERT(md->body != NULL);
 484
 485         if (md->body->valid & OBD_MD_FLEASIZE) {
 486                 int lmmsize;
 487                 struct lov_mds_md *lmm;
 488
 489                 if (!S_ISREG(md->body->mode)) {
 490                         CDEBUG(D_INFO, "OBD_MD_FLEASIZE set, should be a "
 491                                "regular file, but is not\n");
 492                         GOTO(out, rc = -EPROTO);
 493                 }
 494
 495                 if (md->body->eadatasize == 0) {
 496                         CDEBUG(D_INFO, "OBD_MD_FLEASIZE set, "
 497                                "but eadatasize 0\n");
 498                         GOTO(out, rc = -EPROTO);
 499                 }
 500                 lmmsize = md->body->eadatasize;
 501                 lmm = req_capsule_server_sized_get(pill, &RMF_MDT_MD, lmmsize);
 502                 if (!lmm)
 503                         GOTO(out, rc = -EPROTO);
 504
 505                 rc = obd_unpackmd(dt_exp, &md->lsm, lmm, lmmsize);
 506                 if (rc < 0)
 507                         GOTO(out, rc);
 508
 509                 if (rc < sizeof(*md->lsm)) {
 510                         CDEBUG(D_INFO, "lsm size too small: "
 511                                "rc < sizeof (*md->lsm) (%d < %d)\n",
 512                                rc, (int)sizeof(*md->lsm));
 513                         GOTO(out, rc = -EPROTO);
 514                 }
 515
 516         } else if (md->body->valid & OBD_MD_FLDIREA) {
 517                 int lmvsize;
 518                 struct lov_mds_md *lmv;
 519
 520                 if(!S_ISDIR(md->body->mode)) {
 521                         CDEBUG(D_INFO, "OBD_MD_FLDIREA set, should be a "
 522                                "directory, but is not\n");
 523                         GOTO(out, rc = -EPROTO);
 524                 }
 525
 526                 if (md->body->eadatasize == 0) {
 527                         CDEBUG(D_INFO, "OBD_MD_FLDIREA is set, "
 528                                "but eadatasize 0\n");
 529                         RETURN(-EPROTO);
 530                 }
 531                 if (md->body->valid & OBD_MD_MEA) {
 532                         lmvsize = md->body->eadatasize;
 533                         lmv = req_capsule_server_sized_get(pill, &RMF_MDT_MD,
 534                                                            lmvsize);
 535                         if (!lmv)
 536                                 GOTO(out, rc = -EPROTO);
 537
 538                         rc = obd_unpackmd(md_exp, (void *)&md->mea, lmv,
 539                                           lmvsize);
 540                         if (rc < 0)
 541                                 GOTO(out, rc);
 542
 543                         if (rc < sizeof(*md->mea)) {
 544                                 CDEBUG(D_INFO, "size too small:  "
 545                                        "rc < sizeof(*md->mea) (%d < %d)\n",
 546                                         rc, (int)sizeof(*md->mea));
 547                                 GOTO(out, rc = -EPROTO);
 548                         }
 549                 }
 550         }
 551         rc = 0;
 552
 553         if (md->body->valid & OBD_MD_FLRMTPERM) {
 554                 /* remote permission */
 555                 LASSERT(client_is_remote(exp));
 556                 md->remote_perm = req_capsule_server_swab_get(pill, &RMF_ACL,
 557                                                 lustre_swab_mdt_remote_perm);
 558                 if (!md->remote_perm)
 559                         GOTO(out, rc = -EPROTO);
 560         }
 561         else if (md->body->valid & OBD_MD_FLACL) {
 562                 /* for ACL, it's possible that FLACL is set but aclsize is zero.
 563                  * only when aclsize != 0 there's an actual segment for ACL
 564                  * in reply buffer.
 565                  */
 566                 if (md->body->aclsize) {
 567                         rc = mdc_unpack_acl(req, md);
 568                         if (rc)
 569                                 GOTO(out, rc);
 570 #ifdef CONFIG_FS_POSIX_ACL
 571                 } else {
 572                         md->posix_acl = NULL;
 573 #endif
 574                 }
 575         }
 576         if (md->body->valid & OBD_MD_FLMDSCAPA) {
 577                 struct obd_capa *oc = NULL;
 578
 579                 rc = mdc_unpack_capa(NULL, req, &RMF_CAPA1, &oc);
 580                 if (rc)
 581                         GOTO(out, rc);
 582                 md->mds_capa = oc;
 583         }
 584
 585         if (md->body->valid & OBD_MD_FLOSSCAPA) {
 586                 struct obd_capa *oc = NULL;
 587
 588                 rc = mdc_unpack_capa(NULL, req, &RMF_CAPA2, &oc);
 589                 if (rc)
 590                         GOTO(out, rc);
 591                 md->oss_capa = oc;
 592         }
 593
 594         EXIT;
 595 out:
 596         if (rc) {
 597                 if (md->oss_capa) {
 598                         capa_put(md->oss_capa);
 599                         md->oss_capa = NULL;
 600                 }
 601                 if (md->mds_capa) {
 602                         capa_put(md->mds_capa);
 603                         md->mds_capa = NULL;
 604                 }
 605 #ifdef CONFIG_FS_POSIX_ACL
 606                 posix_acl_release(md->posix_acl);
 607 #endif
 608                 if (md->lsm)
 609                         obd_free_memmd(dt_exp, &md->lsm);
 610         }
 611         return rc;
 612 }
 613
 614 int mdc_free_lustre_md(struct obd_export *exp, struct lustre_md *md)
 615 {
 616         ENTRY;
 617         RETURN(0);
 618 }
 619
 620 /**
 621  * Handles both OPEN and SETATTR RPCs for OPEN-CLOSE and SETATTR-DONE_WRITING
 622  * RPC chains.
 623  */
 624 void mdc_replay_open(struct ptlrpc_request *req)
 625 {
 626         struct md_open_data *mod = req->rq_cb_data;
 627         struct ptlrpc_request *close_req;
 628         struct obd_client_handle *och;
 629         struct lustre_handle old;
 630         struct mdt_body *body;
 631         ENTRY;
 632
 633         if (mod == NULL) {
 634                 DEBUG_REQ(D_ERROR, req,
 635                           "Can't properly replay without open data.");
 636                 EXIT;
 637                 return;
 638         }
 639
 640         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 641         LASSERT(body != NULL);
 642
 643         och = mod->mod_och;
 644         if (och != NULL) {
 645                 struct lustre_handle *file_fh;
 646
 647                 LASSERT(och->och_magic == OBD_CLIENT_HANDLE_MAGIC);
 648
 649                 file_fh = &och->och_fh;
 650                 CDEBUG(D_HA, "updating handle from "LPX64" to "LPX64"\n",
 651                        file_fh->cookie, body->handle.cookie);
 652                 old = *file_fh;
 653                 *file_fh = body->handle;
 654         }
 655         close_req = mod->mod_close_req;
 656         if (close_req != NULL) {
 657                 __u32 opc = lustre_msg_get_opc(close_req->rq_reqmsg);
 658                 struct mdt_epoch *epoch;
 659
 660                 LASSERT(opc == MDS_CLOSE || opc == MDS_DONE_WRITING);
 661                 epoch = req_capsule_client_get(&close_req->rq_pill,
 662                                                &RMF_MDT_EPOCH);
 663                 LASSERT(epoch);
 664
 665                 if (och != NULL)
 666                         LASSERT(!memcmp(&old, &epoch->handle, sizeof(old)));
 667                 DEBUG_REQ(D_HA, close_req, "updating close body with new fh");
 668                 epoch->handle = body->handle;
 669         }
 670         EXIT;
 671 }
 672
 673 void mdc_commit_open(struct ptlrpc_request *req)
 674 {
 675         struct md_open_data *mod = req->rq_cb_data;
 676         if (mod == NULL)
 677                 return;
 678
 679         if (mod->mod_och != NULL)
 680                 mod->mod_och->och_mod = NULL;
 681
 682         OBD_FREE(mod, sizeof(*mod));
 683         req->rq_cb_data = NULL;
 684 }
 685
 686 int mdc_set_open_replay_data(struct obd_export *exp,
 687                              struct obd_client_handle *och,
 688                              struct ptlrpc_request *open_req)
 689 {
 690         struct md_open_data   *mod;
 691         struct mdt_rec_create *rec;
 692         struct mdt_body       *body;
 693         struct obd_import     *imp = open_req->rq_import;
 694         ENTRY;
 695
 696         if (!open_req->rq_replay)
 697                 RETURN(0);
 698
 699         rec = req_capsule_client_get(&open_req->rq_pill, &RMF_REC_REINT);
 700         body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
 701         LASSERT(rec != NULL);
 702         /* Incoming message in my byte order (it's been swabbed). */
 703         /* Outgoing messages always in my byte order. */
 704         LASSERT(body != NULL);
 705
 706         /* Only if the import is replayable, we set replay_open data */
 707         if (och && imp->imp_replayable) {
 708                 OBD_ALLOC_PTR(mod);
 709                 if (mod == NULL) {
 710                         DEBUG_REQ(D_ERROR, open_req,
 711                                   "Can't allocate md_open_data");
 712                         RETURN(0);
 713                 }
 714
 715                 spin_lock(&open_req->rq_lock);
 716                 och->och_mod = mod;
 717                 mod->mod_och = och;
 718                 mod->mod_open_req = open_req;
 719                 open_req->rq_cb_data = mod;
 720                 open_req->rq_commit_cb = mdc_commit_open;
 721                 spin_unlock(&open_req->rq_lock);
 722         }
 723
 724         rec->cr_fid2 = body->fid1;
 725         rec->cr_ioepoch = body->ioepoch;
 726         rec->cr_old_handle.cookie = body->handle.cookie;
 727         open_req->rq_replay_cb = mdc_replay_open;
 728         if (!fid_is_sane(&body->fid1)) {
 729                 DEBUG_REQ(D_ERROR, open_req, "Saving replay request with "
 730                           "insane fid");
 731                 LBUG();
 732         }
 733
 734         DEBUG_REQ(D_RPCTRACE, open_req, "Set up open replay data");
 735         RETURN(0);
 736 }
 737
 738 int mdc_clear_open_replay_data(struct obd_export *exp,
 739                                struct obd_client_handle *och)
 740 {
 741         struct md_open_data *mod = och->och_mod;
 742         ENTRY;
 743
 744         /*
 745          * Don't free the structure now (it happens in mdc_commit_open(), after
 746          * we're sure we won't need to fix up the close request in the future),
 747          * but make sure that replay doesn't poke at the och, which is about to
 748          * be freed.
 749          */
 750         LASSERT(mod != LP_POISON);
 751         if (mod != NULL)
 752                 mod->mod_och = NULL;
 753
 754         och->och_mod = NULL;
 755         RETURN(0);
 756 }
 757
 758 int mdc_close(struct obd_export *exp, struct md_op_data *op_data,
 759               struct md_open_data *mod, struct ptlrpc_request **request)
 760 {
 761         struct obd_device     *obd = class_exp2obd(exp);
 762         struct ptlrpc_request *req;
 763         int                    rc;
 764         ENTRY;
 765
 766         *request = NULL;
 767         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_CLOSE);
 768         if (req == NULL)
 769                 RETURN(-ENOMEM);
 770
 771         mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
 772
 773         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_CLOSE);
 774         if (rc) {
 775                 ptlrpc_request_free(req);
 776                 RETURN(rc);
 777         }
 778
 779         /* To avoid a livelock (bug 7034), we need to send CLOSE RPCs to a
 780          * portal whose threads are not taking any DLM locks and are therefore
 781          * always progressing */
 782         req->rq_request_portal = MDS_READPAGE_PORTAL;
 783         ptlrpc_at_set_req_timeout(req);
 784
 785         /* Ensure that this close's handle is fixed up during replay. */
 786         if (likely(mod != NULL)) {
 787                 LASSERTF(mod->mod_open_req->rq_type != LI_POISON,
 788                          "POISONED open %p!\n", mod->mod_open_req);
 789
 790                 mod->mod_close_req = req;
 791                 DEBUG_REQ(D_HA, mod->mod_open_req, "matched open");
 792                 /* We no longer want to preserve this open for replay even
 793                  * though the open was committed. b=3632, b=3633 */
 794                 spin_lock(&mod->mod_open_req->rq_lock);
 795                 mod->mod_open_req->rq_replay = 0;
 796                 spin_unlock(&mod->mod_open_req->rq_lock);
 797         } else {
 798                  CDEBUG(D_HA, "couldn't find open req; expecting close error\n");
 799         }
 800
 801         mdc_close_pack(req, op_data);
 802
 803         req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
 804                              obd->u.cli.cl_max_mds_easize);
 805         req_capsule_set_size(&req->rq_pill, &RMF_LOGCOOKIES, RCL_SERVER,
 806                              obd->u.cli.cl_max_mds_cookiesize);
 807
 808         ptlrpc_request_set_replen(req);
 809
 810         mdc_get_rpc_lock(obd->u.cli.cl_close_lock, NULL);
 811         rc = ptlrpc_queue_wait(req);
 812         mdc_put_rpc_lock(obd->u.cli.cl_close_lock, NULL);
 813
 814         if (req->rq_repmsg == NULL) {
 815                 CDEBUG(D_RPCTRACE, "request failed to send: %p, %d\n", req,
 816                        req->rq_status);
 817                 if (rc == 0)
 818                         rc = req->rq_status ?: -EIO;
 819         } else if (rc == 0 || rc == -EAGAIN) {
 820                 struct mdt_body *body;
 821
 822                 rc = lustre_msg_get_status(req->rq_repmsg);
 823                 if (lustre_msg_get_type(req->rq_repmsg) == PTL_RPC_MSG_ERR) {
 824                         DEBUG_REQ(D_ERROR, req, "type == PTL_RPC_MSG_ERR, err "
 825                                   "= %d", rc);
 826                         if (rc > 0)
 827                                 rc = -rc;
 828                 }
 829                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 830                 if (body == NULL)
 831                         rc = -EPROTO;
 832         }
 833
 834         if (rc != 0 && mod)
 835                  mod->mod_close_req = NULL;
 836
 837         *request = req;
 838         RETURN(rc);
 839 }
 840
 841 int mdc_done_writing(struct obd_export *exp, struct md_op_data *op_data,
 842                      struct md_open_data *mod)
 843 {
 844         struct obd_device     *obd = class_exp2obd(exp);
 845         struct ptlrpc_request *req;
 846         int                    rc;
 847         ENTRY;
 848
 849         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
 850                                    &RQF_MDS_DONE_WRITING);
 851         if (req == NULL)
 852                 RETURN(-ENOMEM);
 853
 854         mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
 855         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_DONE_WRITING);
 856         if (rc) {
 857                 ptlrpc_request_free(req);
 858                 RETURN(rc);
 859         }
 860
 861         if (mod != NULL) {
 862                 LASSERTF(mod->mod_open_req->rq_type != LI_POISON,
 863                          "POISONED setattr %p!\n", mod->mod_open_req);
 864
 865                 mod->mod_close_req = req;
 866                 DEBUG_REQ(D_HA, mod->mod_open_req, "matched setattr");
 867                 /* We no longer want to preserve this open for replay even
 868                  * though the open was committed. b=3632, b=3633 */
 869                 spin_lock(&mod->mod_open_req->rq_lock);
 870                 mod->mod_open_req->rq_replay = 0;
 871                 spin_unlock(&mod->mod_open_req->rq_lock);
 872         }
 873
 874         mdc_close_pack(req, op_data);
 875         ptlrpc_request_set_replen(req);
 876
 877         mdc_get_rpc_lock(obd->u.cli.cl_close_lock, NULL);
 878         rc = ptlrpc_queue_wait(req);
 879         mdc_put_rpc_lock(obd->u.cli.cl_close_lock, NULL);
 880
 881         ptlrpc_req_finished(req);
 882         RETURN(rc);
 883 }
 884
 885 #ifdef HAVE_SPLIT_SUPPORT
 886 int mdc_sendpage(struct obd_export *exp, const struct lu_fid *fid,
 887                  const struct page *page, int offset)
 888 {
 889         struct ptlrpc_request   *req;
 890         struct ptlrpc_bulk_desc *desc;
 891         int                      rc;
 892         ENTRY;
 893
 894         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_WRITEPAGE);
 895         if (req == NULL)
 896                 RETURN(-ENOMEM);
 897
 898         /* FIXME: capa doesn't support split yet */
 899         mdc_set_capa_size(req, &RMF_CAPA1, NULL);
 900
 901         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_WRITEPAGE);
 902         if (rc) {
 903                 ptlrpc_request_free(req);
 904                 RETURN(rc);
 905         }
 906
 907         req->rq_request_portal = MDS_READPAGE_PORTAL;
 908         ptlrpc_at_set_req_timeout(req);
 909
 910         desc = ptlrpc_prep_bulk_imp(req, 1, BULK_GET_SOURCE, MDS_BULK_PORTAL);
 911         if (desc == NULL)
 912                 GOTO(out, rc = -ENOMEM);
 913
 914         /* NB req now owns desc and will free it when it gets freed. */
 915         ptlrpc_prep_bulk_page(desc, (struct page *)page, 0, offset);
 916         mdc_readdir_pack(req, 0, offset, fid, NULL);
 917
 918         ptlrpc_request_set_replen(req);
 919         rc = ptlrpc_queue_wait(req);
 920         if (rc)
 921                 GOTO(out, rc);
 922
 923         rc = sptlrpc_cli_unwrap_bulk_write(req, req->rq_bulk);
 924 out:
 925         ptlrpc_req_finished(req);
 926         return rc;
 927 }
 928 EXPORT_SYMBOL(mdc_sendpage);
 929 #endif
 930
 931 int mdc_readpage(struct obd_export *exp, const struct lu_fid *fid,
 932                  struct obd_capa *oc, __u64 offset, struct page *page,
 933                  struct ptlrpc_request **request)
 934 {
 935         struct ptlrpc_request   *req;
 936         struct ptlrpc_bulk_desc *desc;
 937         int                      rc;
 938         ENTRY;
 939
 940         *request = NULL;
 941         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_READPAGE);
 942         if (req == NULL)
 943                 RETURN(-ENOMEM);
 944
 945         mdc_set_capa_size(req, &RMF_CAPA1, oc);
 946
 947         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_READPAGE);
 948         if (rc) {
 949                 ptlrpc_request_free(req);
 950                 RETURN(rc);
 951         }
 952
 953         req->rq_request_portal = MDS_READPAGE_PORTAL;
 954         ptlrpc_at_set_req_timeout(req);
 955
 956         desc = ptlrpc_prep_bulk_imp(req, 1, BULK_PUT_SINK, MDS_BULK_PORTAL);
 957         if (desc == NULL) {
 958                 ptlrpc_request_free(req);
 959                 RETURN(-ENOMEM);
 960         }
 961
 962         /* NB req now owns desc and will free it when it gets freed */
 963         ptlrpc_prep_bulk_page(desc, page, 0, CFS_PAGE_SIZE);
 964         mdc_readdir_pack(req, offset, CFS_PAGE_SIZE, fid, oc);
 965
 966         ptlrpc_request_set_replen(req);
 967         rc = ptlrpc_queue_wait(req);
 968         if (rc) {
 969                 ptlrpc_req_finished(req);
 970                 RETURN(rc);
 971         }
 972
 973         rc = sptlrpc_cli_unwrap_bulk_read(req, req->rq_bulk,
 974                                           req->rq_bulk->bd_nob_transferred);
 975         if (rc < 0) {
 976                 ptlrpc_req_finished(req);
 977                 RETURN(rc);
 978         }
 979
 980         if (req->rq_bulk->bd_nob_transferred != CFS_PAGE_SIZE) {
 981                 CERROR("Unexpected # bytes transferred: %d (%ld expected)\n",
 982                         req->rq_bulk->bd_nob_transferred, CFS_PAGE_SIZE);
 983                 ptlrpc_req_finished(req);
 984                 RETURN(-EPROTO);
 985         }
 986
 987         *request = req;
 988         RETURN(0);
 989 }
 990
 991 static int mdc_ioc_fid2path(struct obd_export *exp, struct getinfo_fid2path *gf)
 992 {
 993         __u32 keylen, vallen;
 994         void *key;
 995         int rc;
 996
 997         if (gf->gf_pathlen > PATH_MAX)
 998                 RETURN(-ENAMETOOLONG);
 999         if (gf->gf_pathlen < 2)
1000                 RETURN(-EOVERFLOW);
1001
1002         /* Key is KEY_FID2PATH + getinfo_fid2path description */
1003         keylen = size_round(sizeof(KEY_FID2PATH)) + sizeof(*gf);
1004         OBD_ALLOC(key, keylen);
1005         if (key == NULL)
1006                 RETURN(-ENOMEM);
1007         memcpy(key, KEY_FID2PATH, sizeof(KEY_FID2PATH));
1008         memcpy(key + size_round(sizeof(KEY_FID2PATH)), gf, sizeof(*gf));
1009
1010         CDEBUG(D_IOCTL, "path get "DFID" from "LPU64" #%d\n",
1011                PFID(&gf->gf_fid), gf->gf_recno, gf->gf_linkno);
1012
1013         if (!fid_is_sane(&gf->gf_fid))
1014                 GOTO(out, rc = -EINVAL);
1015
1016         /* Val is struct getinfo_fid2path result plus path */
1017         vallen = sizeof(*gf) + gf->gf_pathlen;
1018
1019         rc = obd_get_info(exp, keylen, key, &vallen, gf, NULL);
1020         if (rc)
1021                 GOTO(out, rc);
1022
1023         if (vallen <= sizeof(*gf))
1024                 GOTO(out, rc = -EPROTO);
1025         else if (vallen > sizeof(*gf) + gf->gf_pathlen)
1026                 GOTO(out, rc = -EOVERFLOW);
1027
1028         CDEBUG(D_IOCTL, "path get "DFID" from "LPU64" #%d\n%s\n",
1029                PFID(&gf->gf_fid), gf->gf_recno, gf->gf_linkno, gf->gf_path);
1030
1031 out:
1032         OBD_FREE(key, keylen);
1033         return rc;
1034 }
1035
1036 static int mdc_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1037                          void *karg, void *uarg)
1038 {
1039         struct obd_device *obd = exp->exp_obd;
1040         struct obd_ioctl_data *data = karg;
1041         struct obd_import *imp = obd->u.cli.cl_import;
1042         struct llog_ctxt *ctxt;
1043         int rc;
1044         ENTRY;
1045
1046         if (!try_module_get(THIS_MODULE)) {
1047                 CERROR("Can't get module. Is it alive?");
1048                 return -EINVAL;
1049         }
1050         switch (cmd) {
1051         case OBD_IOC_CHANGELOG_CLEAR: {
1052                 struct ioc_changelog_clear *icc = karg;
1053                 struct changelog_setinfo cs =
1054                         {icc->icc_recno, icc->icc_id};
1055                 rc = obd_set_info_async(exp, strlen(KEY_CHANGELOG_CLEAR),
1056                                         KEY_CHANGELOG_CLEAR, sizeof(cs), &cs,
1057                                         NULL);
1058                 GOTO(out, rc);
1059         }
1060         case OBD_IOC_FID2PATH: {
1061                 rc = mdc_ioc_fid2path(exp, karg);
1062                 GOTO(out, rc);
1063         }
1064         case OBD_IOC_CLIENT_RECOVER:
1065                 rc = ptlrpc_recover_import(imp, data->ioc_inlbuf1);
1066                 if (rc < 0)
1067                         GOTO(out, rc);
1068                 GOTO(out, rc = 0);
1069         case IOC_OSC_SET_ACTIVE:
1070                 rc = ptlrpc_set_import_active(imp, data->ioc_offset);
1071                 GOTO(out, rc);
1072         case OBD_IOC_PARSE: {
1073                 ctxt = llog_get_context(exp->exp_obd, LLOG_CONFIG_REPL_CTXT);
1074                 rc = class_config_parse_llog(ctxt, data->ioc_inlbuf1, NULL);
1075                 llog_ctxt_put(ctxt);
1076                 GOTO(out, rc);
1077         }
1078 #ifdef __KERNEL__
1079         case OBD_IOC_LLOG_INFO:
1080         case OBD_IOC_LLOG_PRINT: {
1081                 ctxt = llog_get_context(obd, LLOG_CONFIG_REPL_CTXT);
1082                 rc = llog_ioctl(ctxt, cmd, data);
1083                 llog_ctxt_put(ctxt);
1084                 GOTO(out, rc);
1085         }
1086 #endif
1087         case OBD_IOC_POLL_QUOTACHECK:
1088                 rc = lquota_poll_check(quota_interface, exp,
1089                                        (struct if_quotacheck *)karg);
1090                 GOTO(out, rc);
1091         case OBD_IOC_PING_TARGET:
1092                 rc = ptlrpc_obd_ping(obd);
1093                 GOTO(out, rc);
1094         default:
1095                 CERROR("mdc_ioctl(): unrecognised ioctl %#x\n", cmd);
1096                 GOTO(out, rc = -ENOTTY);
1097         }
1098 out:
1099         module_put(THIS_MODULE);
1100
1101         return rc;
1102 }
1103
1104 int mdc_get_info_rpc(struct obd_export *exp,
1105                      obd_count keylen, void *key,
1106                      int vallen, void *val)
1107 {
1108         struct obd_import      *imp = class_exp2cliimp(exp);
1109         struct ptlrpc_request  *req;
1110         char                   *tmp;
1111         int                     rc = -EINVAL;
1112         ENTRY;
1113
1114         req = ptlrpc_request_alloc(imp, &RQF_MDS_GET_INFO);
1115         if (req == NULL)
1116                 RETURN(-ENOMEM);
1117
1118         req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_KEY,
1119                              RCL_CLIENT, keylen);
1120         req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_VALLEN,
1121                              RCL_CLIENT, sizeof(__u32));
1122
1123         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GET_INFO);
1124         if (rc) {
1125                 ptlrpc_request_free(req);
1126                 RETURN(rc);
1127         }
1128
1129         tmp = req_capsule_client_get(&req->rq_pill, &RMF_GETINFO_KEY);
1130         memcpy(tmp, key, keylen);
1131         tmp = req_capsule_client_get(&req->rq_pill, &RMF_GETINFO_VALLEN);
1132         memcpy(tmp, &vallen, sizeof(__u32));
1133
1134         req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_VAL,
1135                              RCL_SERVER, vallen);
1136         ptlrpc_request_set_replen(req);
1137
1138         rc = ptlrpc_queue_wait(req);
1139         if (rc == 0) {
1140                 tmp = req_capsule_server_get(&req->rq_pill, &RMF_GETINFO_VAL);
1141                 memcpy(val, tmp, vallen);
1142                 if (lustre_msg_swabbed(req->rq_repmsg)) {
1143                         if (KEY_IS(KEY_FID2PATH)) {
1144                                 lustre_swab_fid2path(val);
1145                         }
1146                 }
1147         }
1148         ptlrpc_req_finished(req);
1149
1150         RETURN(rc);
1151 }
1152
1153 static void lustre_swab_hai(struct hsm_action_item *h)
1154 {
1155         __swab32s(&h->hai_len);
1156         __swab32s(&h->hai_action);
1157         lustre_swab_lu_fid(&h->hai_fid);
1158         __swab64s(&h->hai_cookie);
1159         __swab64s(&h->hai_extent_start);
1160         __swab64s(&h->hai_extent_end);
1161         __swab64s(&h->hai_gid);
1162 }
1163
1164 static void lustre_swab_hal(struct hsm_action_list *h)
1165 {
1166         struct hsm_action_item *hai;
1167         int i;
1168
1169         __swab32s(&h->hal_version);
1170         __swab32s(&h->hal_count);
1171         __swab32s(&h->hal_archive_num);
1172         hai = hai_zero(h);
1173         for (i = 0; i < h->hal_count; i++) {
1174                 lustre_swab_hai(hai);
1175                 hai = hai_next(hai);
1176         }
1177 }
1178
1179 /**
1180  * Send a message to any listening copytools, nonblocking
1181  * @param val LNL message (lnl_hdr + hsm_action_list)
1182  * @param len total length of message
1183  */
1184 static int mdc_hsm_copytool_send(int len, void *val)
1185 {
1186         struct lnl_hdr *lh = (struct lnl_hdr *)val;
1187         struct hsm_action_list *hal = (struct hsm_action_list *)(lh + 1);
1188         int rc;
1189         ENTRY;
1190
1191         if (len < sizeof(*lh) + sizeof(*hal)) {
1192                 CERROR("Short HSM message %d < %d\n", len,
1193                       (int) (sizeof(*lh) + sizeof(*hal)));
1194                 RETURN(-EPROTO);
1195         }
1196         if (lh->lnl_magic == __swab16(LNL_MAGIC)) {
1197                 lustre_swab_lnlh(lh);
1198                 lustre_swab_hal(hal);
1199         } else if (lh->lnl_magic != LNL_MAGIC) {
1200                 CERROR("Bad magic %x!=%x\n", lh->lnl_magic, LNL_MAGIC);
1201                 RETURN(-EPROTO);
1202         }
1203
1204         CDEBUG(D_IOCTL, " Received message mg=%x t=%d m=%d l=%d actions=%d\n",
1205                lh->lnl_magic, lh->lnl_transport, lh->lnl_msgtype,
1206                lh->lnl_msglen, hal->hal_count);
1207
1208         /* Broadcast to HSM listeners */
1209         rc = libcfs_klnl_msg_put(0, LNL_GRP_HSM, lh);
1210
1211         RETURN(rc);
1212 }
1213
1214 int mdc_set_info_async(struct obd_export *exp,
1215                        obd_count keylen, void *key,
1216                        obd_count vallen, void *val,
1217                        struct ptlrpc_request_set *set)
1218 {
1219         struct obd_import *imp = class_exp2cliimp(exp);
1220         int                rc = -EINVAL;
1221         ENTRY;
1222
1223         if (KEY_IS(KEY_INIT_RECOV)) {
1224                 if (vallen != sizeof(int))
1225                         RETURN(-EINVAL);
1226                 spin_lock(&imp->imp_lock);
1227                 imp->imp_initial_recov = *(int *)val;
1228                 spin_unlock(&imp->imp_lock);
1229                 CDEBUG(D_HA, "%s: set imp_initial_recov = %d\n",
1230                        exp->exp_obd->obd_name, imp->imp_initial_recov);
1231                 RETURN(0);
1232         }
1233         /* Turn off initial_recov after we try all backup servers once */
1234         if (KEY_IS(KEY_INIT_RECOV_BACKUP)) {
1235                 if (vallen != sizeof(int))
1236                         RETURN(-EINVAL);
1237                 spin_lock(&imp->imp_lock);
1238                 imp->imp_initial_recov_bk = *(int *)val;
1239                 if (imp->imp_initial_recov_bk)
1240                         imp->imp_initial_recov = 1;
1241                 spin_unlock(&imp->imp_lock);
1242                 CDEBUG(D_HA, "%s: set imp_initial_recov_bk = %d\n",
1243                        exp->exp_obd->obd_name, imp->imp_initial_recov_bk);
1244                 RETURN(0);
1245         }
1246         if (KEY_IS(KEY_READ_ONLY)) {
1247                 if (vallen != sizeof(int))
1248                         RETURN(-EINVAL);
1249
1250                 spin_lock(&imp->imp_lock);
1251                 if (*((int *)val)) {
1252                         imp->imp_connect_flags_orig |= OBD_CONNECT_RDONLY;
1253                         imp->imp_connect_data.ocd_connect_flags |= OBD_CONNECT_RDONLY;
1254                 } else {
1255                         imp->imp_connect_flags_orig &= ~OBD_CONNECT_RDONLY;
1256                         imp->imp_connect_data.ocd_connect_flags &= ~OBD_CONNECT_RDONLY;
1257                 }
1258                 spin_unlock(&imp->imp_lock);
1259
1260                 rc = target_set_info_rpc(imp, MDS_SET_INFO,
1261                                          keylen, key, vallen, val, set);
1262                 RETURN(rc);
1263         }
1264         if (KEY_IS(KEY_SPTLRPC_CONF)) {
1265                 sptlrpc_conf_client_adapt(exp->exp_obd);
1266                 RETURN(0);
1267         }
1268         if (KEY_IS(KEY_FLUSH_CTX)) {
1269                 sptlrpc_import_flush_my_ctx(imp);
1270                 RETURN(0);
1271         }
1272         if (KEY_IS(KEY_MDS_CONN)) {
1273                 /* mds-mds import */
1274                 spin_lock(&imp->imp_lock);
1275                 imp->imp_server_timeout = 1;
1276                 spin_unlock(&imp->imp_lock);
1277                 imp->imp_client->cli_request_portal = MDS_MDS_PORTAL;
1278                 CDEBUG(D_OTHER, "%s: timeout / 2\n", exp->exp_obd->obd_name);
1279                 RETURN(0);
1280         }
1281         if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
1282                 rc = target_set_info_rpc(imp, MDS_SET_INFO,
1283                                          keylen, key, vallen, val, set);
1284                 RETURN(rc);
1285         }
1286         if (KEY_IS(KEY_HSM_COPYTOOL_SEND)) {
1287                 rc = mdc_hsm_copytool_send(vallen, val);
1288                 RETURN(rc);
1289         }
1290
1291         RETURN(rc);
1292 }
1293
1294 int mdc_get_info(struct obd_export *exp, __u32 keylen, void *key,
1295                  __u32 *vallen, void *val, struct lov_stripe_md *lsm)
1296 {
1297         int rc = -EINVAL;
1298
1299         if (KEY_IS(KEY_MAX_EASIZE)) {
1300                 int mdsize, *max_easize;
1301
1302                 if (*vallen != sizeof(int))
1303                         RETURN(-EINVAL);
1304                 mdsize = *(int*)val;
1305                 if (mdsize > exp->exp_obd->u.cli.cl_max_mds_easize)
1306                         exp->exp_obd->u.cli.cl_max_mds_easize = mdsize;
1307                 max_easize = val;
1308                 *max_easize = exp->exp_obd->u.cli.cl_max_mds_easize;
1309                 RETURN(0);
1310         }
1311         if (KEY_IS(KEY_CONN_DATA)) {
1312                 struct obd_import *imp = class_exp2cliimp(exp);
1313                 struct obd_connect_data *data = val;
1314
1315                 if (*vallen != sizeof(*data))
1316                         RETURN(-EINVAL);
1317
1318                 *data = imp->imp_connect_data;
1319                 RETURN(0);
1320         }
1321
1322         rc = mdc_get_info_rpc(exp, keylen, key, *vallen, val);
1323
1324         RETURN(rc);
1325 }
1326
1327 static int mdc_statfs(struct obd_device *obd, struct obd_statfs *osfs,
1328                       __u64 max_age, __u32 flags)
1329 {
1330         struct ptlrpc_request *req;
1331         struct obd_statfs     *msfs;
1332         struct obd_import     *imp = NULL;
1333         int                    rc;
1334         ENTRY;
1335
1336
1337         /*Since the request might also come from lprocfs, so we need
1338          *sync this with client_disconnect_export Bug15684*/
1339         down_read(&obd->u.cli.cl_sem);
1340         if (obd->u.cli.cl_import)
1341                 imp = class_import_get(obd->u.cli.cl_import);
1342         up_read(&obd->u.cli.cl_sem);
1343         if (!imp)
1344                 RETURN(-ENODEV);
1345
1346         req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_STATFS,
1347                                         LUSTRE_MDS_VERSION, MDS_STATFS);
1348         if (req == NULL)
1349                 GOTO(output, rc = -ENOMEM);
1350
1351         ptlrpc_request_set_replen(req);
1352
1353         if (flags & OBD_STATFS_NODELAY) {
1354                 /* procfs requests not want stay in wait for avoid deadlock */
1355                 req->rq_no_resend = 1;
1356                 req->rq_no_delay = 1;
1357         }
1358
1359         rc = ptlrpc_queue_wait(req);
1360         if (rc) {
1361                 /* check connection error first */
1362                 if (imp->imp_connect_error)
1363                         rc = imp->imp_connect_error;
1364                 GOTO(out, rc);
1365         }
1366
1367         msfs = req_capsule_server_get(&req->rq_pill, &RMF_OBD_STATFS);
1368         if (msfs == NULL)
1369                 GOTO(out, rc = -EPROTO);
1370
1371         *osfs = *msfs;
1372         EXIT;
1373 out:
1374         ptlrpc_req_finished(req);
1375 output:
1376         class_import_put(imp);
1377         return rc;
1378 }
1379
1380 static int mdc_pin(struct obd_export *exp, const struct lu_fid *fid,
1381                    struct obd_capa *oc, struct obd_client_handle *handle,
1382                    int flags)
1383 {
1384         struct ptlrpc_request *req;
1385         struct mdt_body       *body;
1386         int                    rc;
1387         ENTRY;
1388
1389         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_PIN);
1390         if (req == NULL)
1391                 RETURN(-ENOMEM);
1392
1393         mdc_set_capa_size(req, &RMF_CAPA1, oc);
1394
1395         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_PIN);
1396         if (rc) {
1397                 ptlrpc_request_free(req);
1398                 RETURN(rc);
1399         }
1400
1401         mdc_pack_body(req, fid, oc, 0, 0, -1, flags);
1402
1403         ptlrpc_request_set_replen(req);
1404
1405         mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
1406         rc = ptlrpc_queue_wait(req);
1407         mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
1408         if (rc) {
1409                 CERROR("Pin failed: %d\n", rc);
1410                 GOTO(err_out, rc);
1411         }
1412
1413         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1414         if (body == NULL)
1415                 GOTO(err_out, rc = -EPROTO);
1416
1417         handle->och_fh = body->handle;
1418         handle->och_magic = OBD_CLIENT_HANDLE_MAGIC;
1419
1420         OBD_ALLOC_PTR(handle->och_mod);
1421         if (handle->och_mod == NULL) {
1422                 DEBUG_REQ(D_ERROR, req, "can't allocate md_open_data");
1423                 GOTO(err_out, rc = -ENOMEM);
1424         }
1425         handle->och_mod->mod_open_req = req; /* will be dropped by unpin */
1426
1427         RETURN(0);
1428
1429 err_out:
1430         ptlrpc_req_finished(req);
1431         RETURN(rc);
1432 }
1433
1434 static int mdc_unpin(struct obd_export *exp, struct obd_client_handle *handle,
1435                      int flag)
1436 {
1437         struct ptlrpc_request *req;
1438         struct mdt_body       *body;
1439         int                    rc;
1440         ENTRY;
1441
1442         req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp), &RQF_MDS_UNPIN,
1443                                         LUSTRE_MDS_VERSION, MDS_UNPIN);
1444         if (req == NULL)
1445                 RETURN(-ENOMEM);
1446
1447         body = req_capsule_client_get(&req->rq_pill, &RMF_MDT_BODY);
1448         body->handle = handle->och_fh;
1449         body->flags = flag;
1450
1451         ptlrpc_request_set_replen(req);
1452
1453         mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
1454         rc = ptlrpc_queue_wait(req);
1455         mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
1456
1457         if (rc != 0)
1458                 CERROR("Unpin failed: %d\n", rc);
1459
1460         ptlrpc_req_finished(req);
1461         ptlrpc_req_finished(handle->och_mod->mod_open_req);
1462
1463         OBD_FREE(handle->och_mod, sizeof(*handle->och_mod));
1464         RETURN(rc);
1465 }
1466
1467 int mdc_sync(struct obd_export *exp, const struct lu_fid *fid,
1468              struct obd_capa *oc, struct ptlrpc_request **request)
1469 {
1470         struct ptlrpc_request *req;
1471         int                    rc;
1472         ENTRY;
1473
1474         *request = NULL;
1475         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_SYNC);
1476         if (req == NULL)
1477                 RETURN(-ENOMEM);
1478
1479         mdc_set_capa_size(req, &RMF_CAPA1, oc);
1480
1481         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_SYNC);
1482         if (rc) {
1483                 ptlrpc_request_free(req);
1484                 RETURN(rc);
1485         }
1486
1487         mdc_pack_body(req, fid, oc, 0, 0, -1, 0);
1488
1489         ptlrpc_request_set_replen(req);
1490
1491         rc = ptlrpc_queue_wait(req);
1492         if (rc)
1493                 ptlrpc_req_finished(req);
1494         else
1495                 *request = req;
1496         RETURN(rc);
1497 }
1498
1499 static int mdc_import_event(struct obd_device *obd, struct obd_import *imp,
1500                             enum obd_import_event event)
1501 {
1502         int rc = 0;
1503
1504         LASSERT(imp->imp_obd == obd);
1505
1506         switch (event) {
1507         case IMP_EVENT_DISCON: {
1508 #if 0
1509                 /* XXX Pass event up to OBDs stack. used only for FLD now */
1510                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_DISCON, NULL);
1511 #endif
1512                 break;
1513         }
1514         case IMP_EVENT_INACTIVE: {
1515                 struct client_obd *cli = &obd->u.cli;
1516                 /*
1517                  * Flush current sequence to make client obtain new one
1518                  * from server in case of disconnect/reconnect.
1519                  * If range is already empty then no need to flush it.
1520                  */
1521                 if (cli->cl_seq != NULL &&
1522                     !range_is_exhausted(&cli->cl_seq->lcs_space)) {
1523                         seq_client_flush(cli->cl_seq);
1524                 }
1525
1526                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_INACTIVE, NULL);
1527                 break;
1528         }
1529         case IMP_EVENT_INVALIDATE: {
1530                 struct ldlm_namespace *ns = obd->obd_namespace;
1531
1532                 ldlm_namespace_cleanup(ns, LDLM_FL_LOCAL_ONLY);
1533
1534                 break;
1535         }
1536         case IMP_EVENT_ACTIVE: {
1537                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_ACTIVE, NULL);
1538                 break;
1539         }
1540         case IMP_EVENT_OCD:
1541                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_OCD, NULL);
1542                 break;
1543
1544         default:
1545                 CERROR("Unknown import event %x\n", event);
1546                 LBUG();
1547         }
1548         RETURN(rc);
1549 }
1550
1551 static int mdc_fid_init(struct obd_export *exp)
1552 {
1553         struct client_obd *cli = &exp->exp_obd->u.cli;
1554         char *prefix;
1555         int rc;
1556         ENTRY;
1557
1558         OBD_ALLOC_PTR(cli->cl_seq);
1559         if (cli->cl_seq == NULL)
1560                 RETURN(-ENOMEM);
1561
1562         OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
1563         if (prefix == NULL)
1564                 GOTO(out_free_seq, rc = -ENOMEM);
1565
1566         snprintf(prefix, MAX_OBD_NAME + 5, "srv-%s",
1567                  exp->exp_obd->obd_name);
1568
1569         /* Init client side sequence-manager */
1570         rc = seq_client_init(cli->cl_seq, exp,
1571                              LUSTRE_SEQ_METADATA,
1572                              prefix, NULL);
1573         OBD_FREE(prefix, MAX_OBD_NAME + 5);
1574         if (rc)
1575                 GOTO(out_free_seq, rc);
1576
1577         RETURN(rc);
1578 out_free_seq:
1579         OBD_FREE_PTR(cli->cl_seq);
1580         cli->cl_seq = NULL;
1581         return rc;
1582 }
1583
1584 static int mdc_fid_fini(struct obd_export *exp)
1585 {
1586         struct client_obd *cli = &exp->exp_obd->u.cli;
1587         ENTRY;
1588
1589         if (cli->cl_seq != NULL) {
1590                 seq_client_fini(cli->cl_seq);
1591                 OBD_FREE_PTR(cli->cl_seq);
1592                 cli->cl_seq = NULL;
1593         }
1594
1595         RETURN(0);
1596 }
1597
1598 int mdc_fid_alloc(struct obd_export *exp, struct lu_fid *fid,
1599                   struct md_op_data *op_data)
1600 {
1601         struct client_obd *cli = &exp->exp_obd->u.cli;
1602         struct lu_client_seq *seq = cli->cl_seq;
1603         ENTRY;
1604         RETURN(seq_client_alloc_fid(seq, fid));
1605 }
1606
1607 /* XXX This method is used only to clear current fid seq
1608  * once fld/mds insert failed */
1609 static int mdc_fid_delete(struct obd_export *exp, const struct lu_fid *fid)
1610 {
1611         struct client_obd *cli = &exp->exp_obd->u.cli;
1612
1613         seq_client_flush(cli->cl_seq);
1614         return 0;
1615 }
1616
1617 struct obd_uuid *mdc_get_uuid(struct obd_export *exp) {
1618         struct client_obd *cli = &exp->exp_obd->u.cli;
1619         return &cli->cl_target_uuid;
1620 }
1621
1622 static int mdc_setup(struct obd_device *obd, struct lustre_cfg *cfg)
1623 {
1624         struct client_obd *cli = &obd->u.cli;
1625         struct lprocfs_static_vars lvars = { 0 };
1626         int rc;
1627         ENTRY;
1628
1629         OBD_ALLOC(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
1630         if (!cli->cl_rpc_lock)
1631                 RETURN(-ENOMEM);
1632         mdc_init_rpc_lock(cli->cl_rpc_lock);
1633
1634         ptlrpcd_addref();
1635
1636         OBD_ALLOC(cli->cl_setattr_lock, sizeof (*cli->cl_setattr_lock));
1637         if (!cli->cl_setattr_lock)
1638                 GOTO(err_rpc_lock, rc = -ENOMEM);
1639         mdc_init_rpc_lock(cli->cl_setattr_lock);
1640
1641         OBD_ALLOC(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
1642         if (!cli->cl_close_lock)
1643                 GOTO(err_setattr_lock, rc = -ENOMEM);
1644         mdc_init_rpc_lock(cli->cl_close_lock);
1645
1646         rc = client_obd_setup(obd, cfg);
1647         if (rc)
1648                 GOTO(err_close_lock, rc);
1649         lprocfs_mdc_init_vars(&lvars);
1650         lprocfs_obd_setup(obd, lvars.obd_vars);
1651         sptlrpc_lprocfs_cliobd_attach(obd);
1652         ptlrpc_lprocfs_register_obd(obd);
1653
1654         rc = obd_llog_init(obd, &obd->obd_olg, obd, NULL);
1655         if (rc) {
1656                 mdc_cleanup(obd);
1657                 CERROR("failed to setup llogging subsystems\n");
1658         }
1659
1660         /* ignore errors */
1661         libcfs_klnl_start(LNL_TRANSPORT_HSM);
1662
1663         RETURN(rc);
1664
1665 err_close_lock:
1666         OBD_FREE(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
1667 err_setattr_lock:
1668         OBD_FREE(cli->cl_setattr_lock, sizeof (*cli->cl_setattr_lock));
1669 err_rpc_lock:
1670         OBD_FREE(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
1671         ptlrpcd_decref();
1672         RETURN(rc);
1673 }
1674
1675 /* Initialize the default and maximum LOV EA and cookie sizes.  This allows
1676  * us to make MDS RPCs with large enough reply buffers to hold the
1677  * maximum-sized (= maximum striped) EA and cookie without having to
1678  * calculate this (via a call into the LOV + OSCs) each time we make an RPC. */
1679 static int mdc_init_ea_size(struct obd_export *exp, int easize,
1680                      int def_easize, int cookiesize)
1681 {
1682         struct obd_device *obd = exp->exp_obd;
1683         struct client_obd *cli = &obd->u.cli;
1684         ENTRY;
1685
1686         if (cli->cl_max_mds_easize < easize)
1687                 cli->cl_max_mds_easize = easize;
1688
1689         if (cli->cl_default_mds_easize < def_easize)
1690                 cli->cl_default_mds_easize = def_easize;
1691
1692         if (cli->cl_max_mds_cookiesize < cookiesize)
1693                 cli->cl_max_mds_cookiesize = cookiesize;
1694
1695         RETURN(0);
1696 }
1697
1698 static int mdc_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
1699 {
1700         int rc = 0;
1701         ENTRY;
1702
1703         switch (stage) {
1704         case OBD_CLEANUP_EARLY:
1705         case OBD_CLEANUP_EXPORTS:
1706                 /* If we set up but never connected, the
1707                    client import will not have been cleaned. */
1708                 if (obd->u.cli.cl_import) {
1709                         struct obd_import *imp;
1710                         down_write(&obd->u.cli.cl_sem);
1711                         imp = obd->u.cli.cl_import;
1712                         CERROR("client import never connected\n");
1713                         ptlrpc_invalidate_import(imp);
1714                         class_destroy_import(imp);
1715                         up_write(&obd->u.cli.cl_sem);
1716                         obd->u.cli.cl_import = NULL;
1717                 }
1718                 rc = obd_llog_finish(obd, 0);
1719                 if (rc != 0)
1720                         CERROR("failed to cleanup llogging subsystems\n");
1721                 break;
1722         }
1723         RETURN(rc);
1724 }
1725
1726 static int mdc_cleanup(struct obd_device *obd)
1727 {
1728         struct client_obd *cli = &obd->u.cli;
1729
1730         libcfs_klnl_stop(LNL_TRANSPORT_HSM, LNL_GRP_HSM);
1731
1732         OBD_FREE(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
1733         OBD_FREE(cli->cl_setattr_lock, sizeof (*cli->cl_setattr_lock));
1734         OBD_FREE(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
1735
1736         ptlrpc_lprocfs_unregister_obd(obd);
1737         lprocfs_obd_cleanup(obd);
1738         ptlrpcd_decref();
1739
1740         return client_obd_cleanup(obd);
1741 }
1742
1743
1744 static int mdc_llog_init(struct obd_device *obd, struct obd_llog_group *olg,
1745                          struct obd_device *tgt, int *index)
1746 {
1747         struct llog_ctxt *ctxt;
1748         int rc;
1749         ENTRY;
1750
1751         LASSERT(olg == &obd->obd_olg);
1752
1753         rc = llog_setup(obd, olg, LLOG_LOVEA_REPL_CTXT, tgt, 0, NULL,
1754                         &llog_client_ops);
1755         if (rc)
1756                 RETURN(rc);
1757
1758         ctxt = llog_get_context(obd, LLOG_LOVEA_REPL_CTXT);
1759         llog_initiator_connect(ctxt);
1760         llog_ctxt_put(ctxt);
1761
1762         rc = llog_setup(obd, olg, LLOG_CHANGELOG_REPL_CTXT, tgt, 0, NULL,
1763                         &llog_client_ops);
1764         if (rc == 0) {
1765                 ctxt = llog_group_get_ctxt(olg, LLOG_CHANGELOG_REPL_CTXT);
1766                 llog_initiator_connect(ctxt);
1767                 llog_ctxt_put(ctxt);
1768         }
1769
1770         RETURN(rc);
1771 }
1772
1773 static int mdc_llog_finish(struct obd_device *obd, int count)
1774 {
1775         struct llog_ctxt *ctxt;
1776         int rc = 0;
1777         ENTRY;
1778
1779         ctxt = llog_get_context(obd, LLOG_LOVEA_REPL_CTXT);
1780         if (ctxt)
1781                 rc = llog_cleanup(ctxt);
1782
1783         ctxt = llog_get_context(obd, LLOG_CHANGELOG_REPL_CTXT);
1784         if (ctxt)
1785                 rc = llog_cleanup(ctxt);
1786
1787         RETURN(rc);
1788 }
1789
1790 static int mdc_process_config(struct obd_device *obd, obd_count len, void *buf)
1791 {
1792         struct lustre_cfg *lcfg = buf;
1793         struct lprocfs_static_vars lvars = { 0 };
1794         int rc = 0;
1795
1796         lprocfs_mdc_init_vars(&lvars);
1797         switch (lcfg->lcfg_command) {
1798         default:
1799                 rc = class_process_proc_param(PARAM_MDC, lvars.obd_vars,
1800                                               lcfg, obd);
1801                 if (rc > 0)
1802                         rc = 0;
1803                 break;
1804         }
1805         return(rc);
1806 }
1807
1808
1809 /* get remote permission for current user on fid */
1810 int mdc_get_remote_perm(struct obd_export *exp, const struct lu_fid *fid,
1811                         struct obd_capa *oc, __u32 suppgid,
1812                         struct ptlrpc_request **request)
1813 {
1814         struct ptlrpc_request  *req;
1815         int                    rc;
1816         ENTRY;
1817
1818         LASSERT(client_is_remote(exp));
1819
1820         *request = NULL;
1821         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_GETATTR);
1822         if (req == NULL)
1823                 RETURN(-ENOMEM);
1824
1825         mdc_set_capa_size(req, &RMF_CAPA1, oc);
1826
1827         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR);
1828         if (rc) {
1829                 ptlrpc_request_free(req);
1830                 RETURN(rc);
1831         }
1832
1833         mdc_pack_body(req, fid, oc, OBD_MD_FLRMTPERM, 0, suppgid, 0);
1834
1835         req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER,
1836                              sizeof(struct mdt_remote_perm));
1837
1838         ptlrpc_request_set_replen(req);
1839
1840         rc = ptlrpc_queue_wait(req);
1841         if (rc)
1842                 ptlrpc_req_finished(req);
1843         else
1844                 *request = req;
1845         RETURN(rc);
1846 }
1847
1848 static int mdc_interpret_renew_capa(const struct lu_env *env,
1849                                     struct ptlrpc_request *req, void *unused,
1850                                     int status)
1851 {
1852         struct obd_capa *oc = req->rq_async_args.pointer_arg[0];
1853         renew_capa_cb_t cb = req->rq_async_args.pointer_arg[1];
1854         struct mdt_body *body = NULL;
1855         struct lustre_capa *capa;
1856         ENTRY;
1857
1858         if (status)
1859                 GOTO(out, capa = ERR_PTR(status));
1860
1861         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1862         if (body == NULL)
1863                 GOTO(out, capa = ERR_PTR(-EFAULT));
1864
1865         if ((body->valid & OBD_MD_FLOSSCAPA) == 0)
1866                 GOTO(out, capa = ERR_PTR(-ENOENT));
1867
1868         capa = req_capsule_server_get(&req->rq_pill, &RMF_CAPA2);
1869         if (!capa)
1870                 GOTO(out, capa = ERR_PTR(-EFAULT));
1871         EXIT;
1872 out:
1873         cb(oc, capa);
1874         return 0;
1875 }
1876
1877 static int mdc_renew_capa(struct obd_export *exp, struct obd_capa *oc,
1878                           renew_capa_cb_t cb)
1879 {
1880         struct ptlrpc_request *req;
1881         ENTRY;
1882
1883         req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp), &RQF_MDS_GETATTR,
1884                                         LUSTRE_MDS_VERSION, MDS_GETATTR);
1885         if (req == NULL)
1886                 RETURN(-ENOMEM);
1887
1888         /* NB, OBD_MD_FLOSSCAPA is set here, but it doesn't necessarily mean the
1889          * capa to renew is oss capa.
1890          */
1891         mdc_pack_body(req, &oc->c_capa.lc_fid, oc, OBD_MD_FLOSSCAPA, 0, -1, 0);
1892         ptlrpc_request_set_replen(req);
1893
1894         req->rq_async_args.pointer_arg[0] = oc;
1895         req->rq_async_args.pointer_arg[1] = cb;
1896         req->rq_interpret_reply = mdc_interpret_renew_capa;
1897         ptlrpcd_add_req(req, PSCOPE_OTHER);
1898         RETURN(0);
1899 }
1900
1901 static int mdc_connect(const struct lu_env *env,
1902                        struct obd_export **exp,
1903                        struct obd_device *obd, struct obd_uuid *cluuid,
1904                        struct obd_connect_data *data,
1905                        void *localdata)
1906 {
1907         struct obd_import *imp = obd->u.cli.cl_import;
1908
1909         /* mds-mds import features */
1910         if (data && (data->ocd_connect_flags & OBD_CONNECT_MDS_MDS)) {
1911                 spin_lock(&imp->imp_lock);
1912                 imp->imp_server_timeout = 1;
1913                 spin_unlock(&imp->imp_lock);
1914                 imp->imp_client->cli_request_portal = MDS_MDS_PORTAL;
1915                 CDEBUG(D_OTHER, "%s: Set 'mds' portal and timeout\n",
1916                        obd->obd_name);
1917         }
1918
1919         return client_connect_import(env, exp, obd, cluuid, data, NULL);
1920 }
1921
1922 struct obd_ops mdc_obd_ops = {
1923         .o_owner            = THIS_MODULE,
1924         .o_setup            = mdc_setup,
1925         .o_precleanup       = mdc_precleanup,
1926         .o_cleanup          = mdc_cleanup,
1927         .o_add_conn         = client_import_add_conn,
1928         .o_del_conn         = client_import_del_conn,
1929         .o_connect          = mdc_connect,
1930         .o_disconnect       = client_disconnect_export,
1931         .o_iocontrol        = mdc_iocontrol,
1932         .o_set_info_async   = mdc_set_info_async,
1933         .o_statfs           = mdc_statfs,
1934         .o_pin              = mdc_pin,
1935         .o_unpin            = mdc_unpin,
1936         .o_fid_init         = mdc_fid_init,
1937         .o_fid_fini         = mdc_fid_fini,
1938         .o_fid_alloc        = mdc_fid_alloc,
1939         .o_fid_delete       = mdc_fid_delete,
1940         .o_import_event     = mdc_import_event,
1941         .o_llog_init        = mdc_llog_init,
1942         .o_llog_finish      = mdc_llog_finish,
1943         .o_get_info         = mdc_get_info,
1944         .o_process_config   = mdc_process_config,
1945         .o_get_uuid         = mdc_get_uuid,
1946 };
1947
1948 struct md_ops mdc_md_ops = {
1949         .m_getstatus        = mdc_getstatus,
1950         .m_change_cbdata    = mdc_change_cbdata,
1951         .m_close            = mdc_close,
1952         .m_create           = mdc_create,
1953         .m_done_writing     = mdc_done_writing,
1954         .m_enqueue          = mdc_enqueue,
1955         .m_getattr          = mdc_getattr,
1956         .m_getattr_name     = mdc_getattr_name,
1957         .m_intent_lock      = mdc_intent_lock,
1958         .m_link             = mdc_link,
1959         .m_is_subdir        = mdc_is_subdir,
1960         .m_rename           = mdc_rename,
1961         .m_setattr          = mdc_setattr,
1962         .m_setxattr         = mdc_setxattr,
1963         .m_getxattr         = mdc_getxattr,
1964         .m_sync             = mdc_sync,
1965         .m_readpage         = mdc_readpage,
1966         .m_unlink           = mdc_unlink,
1967         .m_cancel_unused    = mdc_cancel_unused,
1968         .m_init_ea_size     = mdc_init_ea_size,
1969         .m_set_lock_data    = mdc_set_lock_data,
1970         .m_lock_match       = mdc_lock_match,
1971         .m_get_lustre_md    = mdc_get_lustre_md,
1972         .m_free_lustre_md   = mdc_free_lustre_md,
1973         .m_set_open_replay_data = mdc_set_open_replay_data,
1974         .m_clear_open_replay_data = mdc_clear_open_replay_data,
1975         .m_renew_capa       = mdc_renew_capa,
1976         .m_unpack_capa      = mdc_unpack_capa,
1977         .m_get_remote_perm  = mdc_get_remote_perm,
1978         .m_intent_getattr_async = mdc_intent_getattr_async,
1979         .m_revalidate_lock      = mdc_revalidate_lock
1980 };
1981
1982 int __init mdc_init(void)
1983 {
1984         int rc;
1985         struct lprocfs_static_vars lvars = { 0 };
1986         lprocfs_mdc_init_vars(&lvars);
1987
1988         request_module("lquota");
1989         quota_interface = PORTAL_SYMBOL_GET(mdc_quota_interface);
1990         init_obd_quota_ops(quota_interface, &mdc_obd_ops);
1991
1992         rc = class_register_type(&mdc_obd_ops, &mdc_md_ops, lvars.module_vars,
1993                                  LUSTRE_MDC_NAME, NULL);
1994         if (rc && quota_interface)
1995                 PORTAL_SYMBOL_PUT(mdc_quota_interface);
1996
1997         RETURN(rc);
1998 }
1999
2000 #ifdef __KERNEL__
2001 static void /*__exit*/ mdc_exit(void)
2002 {
2003         if (quota_interface)
2004                 PORTAL_SYMBOL_PUT(mdc_quota_interface);
2005
2006         class_unregister_type(LUSTRE_MDC_NAME);
2007 }
2008
2009 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
2010 MODULE_DESCRIPTION("Lustre Metadata Client");
2011 MODULE_LICENSE("GPL");
2012
2013 module_init(mdc_init);
2014 module_exit(mdc_exit);
2015 #endif