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, 2014, 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 #include <linux/kthread.h>
  45 #include <linux/user_namespace.h>
  46 #ifdef HAVE_UIDGID_HEADER
  47 # include <linux/uidgid.h>
  48 #endif
  49
  50 #include <lustre_acl.h>
  51 #include <lustre_ioctl.h>
  52 #include <obd_class.h>
  53 #include <lustre_lmv.h>
  54 #include <lustre_fid.h>
  55 #include <lprocfs_status.h>
  56 #include <lustre_param.h>
  57 #include <lustre_log.h>
  58 #include <lustre_kernelcomm.h>
  59 #include <cl_object.h>
  60
  61 #include "mdc_internal.h"
  62
  63 #define REQUEST_MINOR 244
  64
  65 struct mdc_renew_capa_args {
  66         struct obd_capa        *ra_oc;
  67         renew_capa_cb_t         ra_cb;
  68 };
  69
  70 static int mdc_cleanup(struct obd_device *obd);
  71
  72 static int mdc_unpack_capa(struct obd_export *exp, struct ptlrpc_request *req,
  73                            const struct req_msg_field *field,
  74                            struct obd_capa **oc)
  75 {
  76         struct lustre_capa *capa;
  77         struct obd_capa *c;
  78         ENTRY;
  79
  80         /* swabbed already in mdc_enqueue */
  81         capa = req_capsule_server_get(&req->rq_pill, field);
  82         if (capa == NULL)
  83                 RETURN(-EPROTO);
  84
  85         c = alloc_capa(CAPA_SITE_CLIENT);
  86         if (IS_ERR(c)) {
  87                 CDEBUG(D_INFO, "alloc capa failed!\n");
  88                 RETURN(PTR_ERR(c));
  89         } else {
  90                 c->c_capa = *capa;
  91                 *oc = c;
  92                 RETURN(0);
  93         }
  94 }
  95
  96 static inline int mdc_queue_wait(struct ptlrpc_request *req)
  97 {
  98         struct client_obd *cli = &req->rq_import->imp_obd->u.cli;
  99         int rc;
 100
 101         /* obd_get_request_slot() ensures that this client has no more
 102          * than cl_max_rpcs_in_flight RPCs simultaneously inf light
 103          * against an MDT. */
 104         rc = obd_get_request_slot(cli);
 105         if (rc != 0)
 106                 return rc;
 107
 108         rc = ptlrpc_queue_wait(req);
 109         obd_put_request_slot(cli);
 110
 111         return rc;
 112 }
 113
 114 /* Helper that implements most of mdc_getstatus and signal_completed_replay. */
 115 /* XXX this should become mdc_get_info("key"), sending MDS_GET_INFO RPC */
 116 static int send_getstatus(struct obd_import *imp, struct lu_fid *rootfid,
 117                           struct obd_capa **pc, int level, int msg_flags)
 118 {
 119         struct ptlrpc_request *req;
 120         struct mdt_body       *body;
 121         int                    rc;
 122         ENTRY;
 123
 124         req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_GETSTATUS,
 125                                         LUSTRE_MDS_VERSION, MDS_GETSTATUS);
 126         if (req == NULL)
 127                 RETURN(-ENOMEM);
 128
 129         mdc_pack_body(req, NULL, NULL, 0, 0, -1, 0);
 130         lustre_msg_add_flags(req->rq_reqmsg, msg_flags);
 131         req->rq_send_state = level;
 132
 133         ptlrpc_request_set_replen(req);
 134
 135         rc = ptlrpc_queue_wait(req);
 136         if (rc)
 137                 GOTO(out, rc);
 138
 139         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 140         if (body == NULL)
 141                 GOTO(out, rc = -EPROTO);
 142
 143         if (body->mbo_valid & OBD_MD_FLMDSCAPA) {
 144                 rc = mdc_unpack_capa(NULL, req, &RMF_CAPA1, pc);
 145                 if (rc)
 146                         GOTO(out, rc);
 147         }
 148
 149         *rootfid = body->mbo_fid1;
 150         CDEBUG(D_NET, "root fid="DFID", last_committed="LPU64"\n",
 151                PFID(rootfid), lustre_msg_get_last_committed(req->rq_repmsg));
 152         EXIT;
 153 out:
 154         ptlrpc_req_finished(req);
 155         return rc;
 156 }
 157
 158 /* This should be mdc_get_info("rootfid") */
 159 static int mdc_getstatus(struct obd_export *exp, struct lu_fid *rootfid,
 160                          struct obd_capa **pc)
 161 {
 162         return send_getstatus(class_exp2cliimp(exp), rootfid, pc,
 163                               LUSTRE_IMP_FULL, 0);
 164 }
 165
 166 /*
 167  * This function now is known to always saying that it will receive 4 buffers
 168  * from server. Even for cases when acl_size and md_size is zero, RPC header
 169  * will contain 4 fields and RPC itself will contain zero size fields. This is
 170  * because mdt_getattr*() _always_ returns 4 fields, but if acl is not needed
 171  * and thus zero, it shrinks it, making zero size. The same story about
 172  * md_size. And this is course of problem when client waits for smaller number
 173  * of fields. This issue will be fixed later when client gets aware of RPC
 174  * layouts.  --umka
 175  */
 176 static int mdc_getattr_common(struct obd_export *exp,
 177                               struct ptlrpc_request *req)
 178 {
 179         struct req_capsule *pill = &req->rq_pill;
 180         struct mdt_body    *body;
 181         void               *eadata;
 182         int                 rc;
 183         ENTRY;
 184
 185         /* Request message already built. */
 186         rc = ptlrpc_queue_wait(req);
 187         if (rc != 0)
 188                 RETURN(rc);
 189
 190         /* sanity check for the reply */
 191         body = req_capsule_server_get(pill, &RMF_MDT_BODY);
 192         if (body == NULL)
 193                 RETURN(-EPROTO);
 194
 195         CDEBUG(D_NET, "mode: %o\n", body->mbo_mode);
 196
 197         mdc_update_max_ea_from_body(exp, body);
 198         if (body->mbo_eadatasize != 0) {
 199                 eadata = req_capsule_server_sized_get(pill, &RMF_MDT_MD,
 200                                                       body->mbo_eadatasize);
 201                 if (eadata == NULL)
 202                         RETURN(-EPROTO);
 203         }
 204
 205         if (body->mbo_valid & OBD_MD_FLRMTPERM) {
 206                 struct mdt_remote_perm *perm;
 207
 208                 LASSERT(client_is_remote(exp));
 209                 perm = req_capsule_server_swab_get(pill, &RMF_ACL,
 210                                                 lustre_swab_mdt_remote_perm);
 211                 if (perm == NULL)
 212                         RETURN(-EPROTO);
 213         }
 214
 215         if (body->mbo_valid & OBD_MD_FLMDSCAPA) {
 216                 struct lustre_capa *capa;
 217                 capa = req_capsule_server_get(pill, &RMF_CAPA1);
 218                 if (capa == NULL)
 219                         RETURN(-EPROTO);
 220         }
 221
 222         RETURN(0);
 223 }
 224
 225 static int mdc_getattr(struct obd_export *exp, struct md_op_data *op_data,
 226                        struct ptlrpc_request **request)
 227 {
 228         struct ptlrpc_request *req;
 229         int                    rc;
 230         ENTRY;
 231
 232         /* Single MDS without an LMV case */
 233         if (op_data->op_flags & MF_GET_MDT_IDX) {
 234                 op_data->op_mds = 0;
 235                 RETURN(0);
 236         }
 237         *request = NULL;
 238         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_GETATTR);
 239         if (req == NULL)
 240                 RETURN(-ENOMEM);
 241
 242         mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
 243
 244         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR);
 245         if (rc) {
 246                 ptlrpc_request_free(req);
 247                 RETURN(rc);
 248         }
 249
 250         mdc_pack_body(req, &op_data->op_fid1, op_data->op_capa1,
 251                       op_data->op_valid, op_data->op_mode, -1, 0);
 252
 253         req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
 254                              op_data->op_mode);
 255         if (op_data->op_valid & OBD_MD_FLRMTPERM) {
 256                 LASSERT(client_is_remote(exp));
 257                 req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER,
 258                                      sizeof(struct mdt_remote_perm));
 259         }
 260         ptlrpc_request_set_replen(req);
 261
 262         rc = mdc_getattr_common(exp, req);
 263         if (rc)
 264                 ptlrpc_req_finished(req);
 265         else
 266                 *request = req;
 267         RETURN(rc);
 268 }
 269
 270 static int mdc_getattr_name(struct obd_export *exp, struct md_op_data *op_data,
 271                             struct ptlrpc_request **request)
 272 {
 273         struct ptlrpc_request *req;
 274         int                    rc;
 275         ENTRY;
 276
 277         *request = NULL;
 278         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
 279                                    &RQF_MDS_GETATTR_NAME);
 280         if (req == NULL)
 281                 RETURN(-ENOMEM);
 282
 283         mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
 284         req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
 285                              op_data->op_namelen + 1);
 286
 287         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR_NAME);
 288         if (rc) {
 289                 ptlrpc_request_free(req);
 290                 RETURN(rc);
 291         }
 292
 293         mdc_pack_body(req, &op_data->op_fid1, op_data->op_capa1,
 294                       op_data->op_valid, op_data->op_mode,
 295                       op_data->op_suppgids[0], 0);
 296
 297         if (op_data->op_name) {
 298                 char *name = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
 299                 LASSERT(strnlen(op_data->op_name, op_data->op_namelen) ==
 300                                 op_data->op_namelen);
 301                 memcpy(name, op_data->op_name, op_data->op_namelen);
 302         }
 303
 304         req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
 305                              op_data->op_mode);
 306         ptlrpc_request_set_replen(req);
 307
 308         rc = mdc_getattr_common(exp, req);
 309         if (rc)
 310                 ptlrpc_req_finished(req);
 311         else
 312                 *request = req;
 313         RETURN(rc);
 314 }
 315
 316 static int mdc_xattr_common(struct obd_export *exp,const struct req_format *fmt,
 317                             const struct lu_fid *fid,
 318                             struct obd_capa *oc, int opcode, u64 valid,
 319                             const char *xattr_name, const char *input,
 320                             int input_size, int output_size, int flags,
 321                             __u32 suppgid, struct ptlrpc_request **request)
 322 {
 323         struct ptlrpc_request *req;
 324         int   xattr_namelen = 0;
 325         char *tmp;
 326         int   rc;
 327         ENTRY;
 328
 329         *request = NULL;
 330         req = ptlrpc_request_alloc(class_exp2cliimp(exp), fmt);
 331         if (req == NULL)
 332                 RETURN(-ENOMEM);
 333
 334         mdc_set_capa_size(req, &RMF_CAPA1, oc);
 335         if (xattr_name) {
 336                 xattr_namelen = strlen(xattr_name) + 1;
 337                 req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
 338                                      xattr_namelen);
 339         }
 340         if (input_size) {
 341                 LASSERT(input);
 342                 req_capsule_set_size(&req->rq_pill, &RMF_EADATA, RCL_CLIENT,
 343                                      input_size);
 344         }
 345
 346         /* Flush local XATTR locks to get rid of a possible cancel RPC */
 347         if (opcode == MDS_REINT && fid_is_sane(fid) &&
 348             exp->exp_connect_data.ocd_ibits_known & MDS_INODELOCK_XATTR) {
 349                 struct list_head cancels = LIST_HEAD_INIT(cancels);
 350                 int count;
 351
 352                 /* Without that packing would fail */
 353                 if (input_size == 0)
 354                         req_capsule_set_size(&req->rq_pill, &RMF_EADATA,
 355                                              RCL_CLIENT, 0);
 356
 357                 count = mdc_resource_get_unused(exp, fid,
 358                                                 &cancels, LCK_EX,
 359                                                 MDS_INODELOCK_XATTR);
 360
 361                 rc = mdc_prep_elc_req(exp, req, MDS_REINT, &cancels, count);
 362                 if (rc) {
 363                         ptlrpc_request_free(req);
 364                         RETURN(rc);
 365                 }
 366         } else {
 367                 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, opcode);
 368                 if (rc) {
 369                         ptlrpc_request_free(req);
 370                         RETURN(rc);
 371                 }
 372         }
 373
 374         if (opcode == MDS_REINT) {
 375                 struct mdt_rec_setxattr *rec;
 376
 377                 CLASSERT(sizeof(struct mdt_rec_setxattr) ==
 378                          sizeof(struct mdt_rec_reint));
 379                 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
 380                 rec->sx_opcode = REINT_SETXATTR;
 381                 rec->sx_fsuid  = from_kuid(&init_user_ns, current_fsuid());
 382                 rec->sx_fsgid  = from_kgid(&init_user_ns, current_fsgid());
 383                 rec->sx_cap    = cfs_curproc_cap_pack();
 384                 rec->sx_suppgid1 = suppgid;
 385                 rec->sx_suppgid2 = -1;
 386                 rec->sx_fid    = *fid;
 387                 rec->sx_valid  = valid | OBD_MD_FLCTIME;
 388                 rec->sx_time   = cfs_time_current_sec();
 389                 rec->sx_size   = output_size;
 390                 rec->sx_flags  = flags;
 391
 392                 mdc_pack_capa(req, &RMF_CAPA1, oc);
 393         } else {
 394                 mdc_pack_body(req, fid, oc, valid, output_size, suppgid, flags);
 395         }
 396
 397         if (xattr_name) {
 398                 tmp = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
 399                 memcpy(tmp, xattr_name, xattr_namelen);
 400         }
 401         if (input_size) {
 402                 tmp = req_capsule_client_get(&req->rq_pill, &RMF_EADATA);
 403                 memcpy(tmp, input, input_size);
 404         }
 405
 406         if (req_capsule_has_field(&req->rq_pill, &RMF_EADATA, RCL_SERVER))
 407                 req_capsule_set_size(&req->rq_pill, &RMF_EADATA,
 408                                      RCL_SERVER, output_size);
 409         ptlrpc_request_set_replen(req);
 410
 411         /* make rpc */
 412         if (opcode == MDS_REINT)
 413                 mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
 414
 415         rc = ptlrpc_queue_wait(req);
 416
 417         if (opcode == MDS_REINT)
 418                 mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
 419
 420         if (rc)
 421                 ptlrpc_req_finished(req);
 422         else
 423                 *request = req;
 424         RETURN(rc);
 425 }
 426
 427 static int mdc_setxattr(struct obd_export *exp, const struct lu_fid *fid,
 428                         struct obd_capa *oc, u64 valid,
 429                         const char *xattr_name,
 430                         const char *input, int input_size, int output_size,
 431                         int flags, __u32 suppgid,
 432                         struct ptlrpc_request **request)
 433 {
 434         return mdc_xattr_common(exp, &RQF_MDS_REINT_SETXATTR,
 435                                 fid, oc, MDS_REINT, valid, xattr_name,
 436                                 input, input_size, output_size, flags,
 437                                 suppgid, request);
 438 }
 439
 440 static int mdc_getxattr(struct obd_export *exp, const struct lu_fid *fid,
 441                         struct obd_capa *oc, u64 valid,
 442                         const char *xattr_name,
 443                         const char *input, int input_size, int output_size,
 444                         int flags, struct ptlrpc_request **request)
 445 {
 446         return mdc_xattr_common(exp, &RQF_MDS_GETXATTR,
 447                                 fid, oc, MDS_GETXATTR, valid, xattr_name,
 448                                 input, input_size, output_size, flags,
 449                                 -1, request);
 450 }
 451
 452 #ifdef CONFIG_FS_POSIX_ACL
 453 static int mdc_unpack_acl(struct ptlrpc_request *req, struct lustre_md *md)
 454 {
 455         struct req_capsule     *pill = &req->rq_pill;
 456         struct mdt_body        *body = md->body;
 457         struct posix_acl       *acl;
 458         void                   *buf;
 459         int                     rc;
 460         ENTRY;
 461
 462         if (!body->mbo_aclsize)
 463                 RETURN(0);
 464
 465         buf = req_capsule_server_sized_get(pill, &RMF_ACL, body->mbo_aclsize);
 466
 467         if (!buf)
 468                 RETURN(-EPROTO);
 469
 470         acl = posix_acl_from_xattr(&init_user_ns, buf, body->mbo_aclsize);
 471         if (acl == NULL)
 472                 RETURN(0);
 473         if (IS_ERR(acl)) {
 474                 rc = PTR_ERR(acl);
 475                 CERROR("convert xattr to acl: %d\n", rc);
 476                 RETURN(rc);
 477         }
 478
 479         rc = posix_acl_valid(acl);
 480         if (rc) {
 481                 CERROR("validate acl: %d\n", rc);
 482                 posix_acl_release(acl);
 483                 RETURN(rc);
 484         }
 485
 486         md->posix_acl = acl;
 487         RETURN(0);
 488 }
 489 #else
 490 #define mdc_unpack_acl(req, md) 0
 491 #endif
 492
 493 int mdc_get_lustre_md(struct obd_export *exp, struct ptlrpc_request *req,
 494                       struct obd_export *dt_exp, struct obd_export *md_exp,
 495                       struct lustre_md *md)
 496 {
 497         struct req_capsule *pill = &req->rq_pill;
 498         int rc;
 499         ENTRY;
 500
 501         LASSERT(md);
 502         memset(md, 0, sizeof(*md));
 503
 504         md->body = req_capsule_server_get(pill, &RMF_MDT_BODY);
 505         LASSERT(md->body != NULL);
 506
 507         if (md->body->mbo_valid & OBD_MD_FLEASIZE) {
 508                 int lmmsize;
 509                 struct lov_mds_md *lmm;
 510
 511                 if (!S_ISREG(md->body->mbo_mode)) {
 512                         CDEBUG(D_INFO, "OBD_MD_FLEASIZE set, should be a "
 513                                "regular file, but is not\n");
 514                         GOTO(out, rc = -EPROTO);
 515                 }
 516
 517                 if (md->body->mbo_eadatasize == 0) {
 518                         CDEBUG(D_INFO, "OBD_MD_FLEASIZE set, "
 519                                "but eadatasize 0\n");
 520                         GOTO(out, rc = -EPROTO);
 521                 }
 522
 523                 lmmsize = md->body->mbo_eadatasize;
 524                 lmm = req_capsule_server_sized_get(pill, &RMF_MDT_MD, lmmsize);
 525                 if (!lmm)
 526                         GOTO(out, rc = -EPROTO);
 527
 528                 rc = obd_unpackmd(dt_exp, &md->lsm, lmm, lmmsize);
 529                 if (rc < 0)
 530                         GOTO(out, rc);
 531
 532                 if (rc < (typeof(rc))sizeof(*md->lsm)) {
 533                         CDEBUG(D_INFO, "lsm size too small: "
 534                                "rc < sizeof (*md->lsm) (%d < %d)\n",
 535                                rc, (int)sizeof(*md->lsm));
 536                         GOTO(out, rc = -EPROTO);
 537                 }
 538
 539         } else if (md->body->mbo_valid & OBD_MD_FLDIREA) {
 540                 int lmvsize;
 541                 struct lov_mds_md *lmv;
 542
 543                 if (!S_ISDIR(md->body->mbo_mode)) {
 544                         CDEBUG(D_INFO, "OBD_MD_FLDIREA set, should be a "
 545                                "directory, but is not\n");
 546                         GOTO(out, rc = -EPROTO);
 547                 }
 548
 549                 if (md->body->mbo_eadatasize == 0) {
 550                         CDEBUG(D_INFO, "OBD_MD_FLDIREA is set, "
 551                                "but eadatasize 0\n");
 552                         RETURN(-EPROTO);
 553                 }
 554
 555                 if (md->body->mbo_valid & OBD_MD_MEA) {
 556                         lmvsize = md->body->mbo_eadatasize;
 557                         lmv = req_capsule_server_sized_get(pill, &RMF_MDT_MD,
 558                                                            lmvsize);
 559                         if (!lmv)
 560                                 GOTO(out, rc = -EPROTO);
 561
 562                         rc = obd_unpackmd(md_exp, (void *)&md->lmv, lmv,
 563                                           lmvsize);
 564                         if (rc < 0)
 565                                 GOTO(out, rc);
 566
 567                         if (rc < (typeof(rc))sizeof(*md->lmv)) {
 568                                 CDEBUG(D_INFO, "size too small:  "
 569                                        "rc < sizeof(*md->lmv) (%d < %d)\n",
 570                                         rc, (int)sizeof(*md->lmv));
 571                                 GOTO(out, rc = -EPROTO);
 572                         }
 573                 }
 574         }
 575         rc = 0;
 576
 577         if (md->body->mbo_valid & OBD_MD_FLRMTPERM) {
 578                 /* remote permission */
 579                 LASSERT(client_is_remote(exp));
 580                 md->remote_perm = req_capsule_server_swab_get(pill, &RMF_ACL,
 581                                                 lustre_swab_mdt_remote_perm);
 582                 if (!md->remote_perm)
 583                         GOTO(out, rc = -EPROTO);
 584         } else if (md->body->mbo_valid & OBD_MD_FLACL) {
 585                 /* for ACL, it's possible that FLACL is set but aclsize is zero.
 586                  * only when aclsize != 0 there's an actual segment for ACL
 587                  * in reply buffer.
 588                  */
 589                 if (md->body->mbo_aclsize) {
 590                         rc = mdc_unpack_acl(req, md);
 591                         if (rc)
 592                                 GOTO(out, rc);
 593 #ifdef CONFIG_FS_POSIX_ACL
 594                 } else {
 595                         md->posix_acl = NULL;
 596 #endif
 597                 }
 598         }
 599         if (md->body->mbo_valid & OBD_MD_FLMDSCAPA) {
 600                 struct obd_capa *oc = NULL;
 601
 602                 rc = mdc_unpack_capa(NULL, req, &RMF_CAPA1, &oc);
 603                 if (rc)
 604                         GOTO(out, rc);
 605                 md->mds_capa = oc;
 606         }
 607
 608         if (md->body->mbo_valid & OBD_MD_FLOSSCAPA) {
 609                 struct obd_capa *oc = NULL;
 610
 611                 rc = mdc_unpack_capa(NULL, req, &RMF_CAPA2, &oc);
 612                 if (rc)
 613                         GOTO(out, rc);
 614                 md->oss_capa = oc;
 615         }
 616
 617         EXIT;
 618 out:
 619         if (rc) {
 620                 if (md->oss_capa) {
 621                         capa_put(md->oss_capa);
 622                         md->oss_capa = NULL;
 623                 }
 624                 if (md->mds_capa) {
 625                         capa_put(md->mds_capa);
 626                         md->mds_capa = NULL;
 627                 }
 628 #ifdef CONFIG_FS_POSIX_ACL
 629                 posix_acl_release(md->posix_acl);
 630 #endif
 631                 if (md->lsm)
 632                         obd_free_memmd(dt_exp, &md->lsm);
 633         }
 634         return rc;
 635 }
 636
 637 int mdc_free_lustre_md(struct obd_export *exp, struct lustre_md *md)
 638 {
 639         ENTRY;
 640         RETURN(0);
 641 }
 642
 643 void mdc_replay_open(struct ptlrpc_request *req)
 644 {
 645         struct md_open_data *mod = req->rq_cb_data;
 646         struct ptlrpc_request *close_req;
 647         struct obd_client_handle *och;
 648         struct lustre_handle old;
 649         struct mdt_body *body;
 650         ENTRY;
 651
 652         if (mod == NULL) {
 653                 DEBUG_REQ(D_ERROR, req,
 654                           "Can't properly replay without open data.");
 655                 EXIT;
 656                 return;
 657         }
 658
 659         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 660         LASSERT(body != NULL);
 661
 662         och = mod->mod_och;
 663         if (och != NULL) {
 664                 struct lustre_handle *file_fh;
 665
 666                 LASSERT(och->och_magic == OBD_CLIENT_HANDLE_MAGIC);
 667
 668                 file_fh = &och->och_fh;
 669                 CDEBUG(D_HA, "updating handle from "LPX64" to "LPX64"\n",
 670                        file_fh->cookie, body->mbo_handle.cookie);
 671                 old = *file_fh;
 672                 *file_fh = body->mbo_handle;
 673         }
 674         close_req = mod->mod_close_req;
 675         if (close_req != NULL) {
 676                 __u32 opc = lustre_msg_get_opc(close_req->rq_reqmsg);
 677                 struct mdt_ioepoch *epoch;
 678
 679                 LASSERT(opc == MDS_CLOSE);
 680                 epoch = req_capsule_client_get(&close_req->rq_pill,
 681                                                &RMF_MDT_EPOCH);
 682                 LASSERT(epoch);
 683
 684                 if (och != NULL)
 685                         LASSERT(!memcmp(&old, &epoch->mio_handle, sizeof(old)));
 686
 687                 DEBUG_REQ(D_HA, close_req, "updating close body with new fh");
 688                 epoch->mio_handle = body->mbo_handle;
 689         }
 690         EXIT;
 691 }
 692
 693 void mdc_commit_open(struct ptlrpc_request *req)
 694 {
 695         struct md_open_data *mod = req->rq_cb_data;
 696         if (mod == NULL)
 697                 return;
 698
 699         /**
 700          * No need to touch md_open_data::mod_och, it holds a reference on
 701          * \var mod and will zero references to each other, \var mod will be
 702          * freed after that when md_open_data::mod_och will put the reference.
 703          */
 704
 705         /**
 706          * Do not let open request to disappear as it still may be needed
 707          * for close rpc to happen (it may happen on evict only, otherwise
 708          * ptlrpc_request::rq_replay does not let mdc_commit_open() to be
 709          * called), just mark this rpc as committed to distinguish these 2
 710          * cases, see mdc_close() for details. The open request reference will
 711          * be put along with freeing \var mod.
 712          */
 713         ptlrpc_request_addref(req);
 714         spin_lock(&req->rq_lock);
 715         req->rq_committed = 1;
 716         spin_unlock(&req->rq_lock);
 717         req->rq_cb_data = NULL;
 718         obd_mod_put(mod);
 719 }
 720
 721 int mdc_set_open_replay_data(struct obd_export *exp,
 722                              struct obd_client_handle *och,
 723                              struct lookup_intent *it)
 724 {
 725         struct md_open_data     *mod;
 726         struct mdt_rec_create   *rec;
 727         struct mdt_body         *body;
 728         struct ptlrpc_request   *open_req = it->d.lustre.it_data;
 729         struct obd_import       *imp = open_req->rq_import;
 730         ENTRY;
 731
 732         if (!open_req->rq_replay)
 733                 RETURN(0);
 734
 735         rec = req_capsule_client_get(&open_req->rq_pill, &RMF_REC_REINT);
 736         body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
 737         LASSERT(rec != NULL);
 738         /* Incoming message in my byte order (it's been swabbed). */
 739         /* Outgoing messages always in my byte order. */
 740         LASSERT(body != NULL);
 741
 742         /* Only if the import is replayable, we set replay_open data */
 743         if (och && imp->imp_replayable) {
 744                 mod = obd_mod_alloc();
 745                 if (mod == NULL) {
 746                         DEBUG_REQ(D_ERROR, open_req,
 747                                   "Can't allocate md_open_data");
 748                         RETURN(0);
 749                 }
 750
 751                 /**
 752                  * Take a reference on \var mod, to be freed on mdc_close().
 753                  * It protects \var mod from being freed on eviction (commit
 754                  * callback is called despite rq_replay flag).
 755                  * Another reference for \var och.
 756                  */
 757                 obd_mod_get(mod);
 758                 obd_mod_get(mod);
 759
 760                 spin_lock(&open_req->rq_lock);
 761                 och->och_mod = mod;
 762                 mod->mod_och = och;
 763                 mod->mod_is_create = it_disposition(it, DISP_OPEN_CREATE) ||
 764                                      it_disposition(it, DISP_OPEN_STRIPE);
 765                 mod->mod_open_req = open_req;
 766                 open_req->rq_cb_data = mod;
 767                 open_req->rq_commit_cb = mdc_commit_open;
 768                 spin_unlock(&open_req->rq_lock);
 769         }
 770
 771         rec->cr_fid2 = body->mbo_fid1;
 772         rec->cr_ioepoch = body->mbo_ioepoch;
 773         rec->cr_old_handle.cookie = body->mbo_handle.cookie;
 774         open_req->rq_replay_cb = mdc_replay_open;
 775         if (!fid_is_sane(&body->mbo_fid1)) {
 776                 DEBUG_REQ(D_ERROR, open_req, "Saving replay request with "
 777                           "insane fid");
 778                 LBUG();
 779         }
 780
 781         DEBUG_REQ(D_RPCTRACE, open_req, "Set up open replay data");
 782         RETURN(0);
 783 }
 784
 785 static void mdc_free_open(struct md_open_data *mod)
 786 {
 787         int committed = 0;
 788
 789         if (mod->mod_is_create == 0 &&
 790             imp_connect_disp_stripe(mod->mod_open_req->rq_import))
 791                 committed = 1;
 792
 793         LASSERT(mod->mod_open_req->rq_replay == 0);
 794
 795         DEBUG_REQ(D_RPCTRACE, mod->mod_open_req, "free open request\n");
 796
 797         ptlrpc_request_committed(mod->mod_open_req, committed);
 798         if (mod->mod_close_req)
 799                 ptlrpc_request_committed(mod->mod_close_req, committed);
 800 }
 801
 802 int mdc_clear_open_replay_data(struct obd_export *exp,
 803                                struct obd_client_handle *och)
 804 {
 805         struct md_open_data *mod = och->och_mod;
 806         ENTRY;
 807
 808         /**
 809          * It is possible to not have \var mod in a case of eviction between
 810          * lookup and ll_file_open().
 811          **/
 812         if (mod == NULL)
 813                 RETURN(0);
 814
 815         LASSERT(mod != LP_POISON);
 816         LASSERT(mod->mod_open_req != NULL);
 817         mdc_free_open(mod);
 818
 819         mod->mod_och = NULL;
 820         och->och_mod = NULL;
 821         obd_mod_put(mod);
 822
 823         RETURN(0);
 824 }
 825
 826 static int mdc_close(struct obd_export *exp, struct md_op_data *op_data,
 827                      struct md_open_data *mod, struct ptlrpc_request **request)
 828 {
 829         struct obd_device     *obd = class_exp2obd(exp);
 830         struct ptlrpc_request *req;
 831         struct req_format     *req_fmt;
 832         int                    rc;
 833         int                    saved_rc = 0;
 834         ENTRY;
 835
 836         req_fmt = &RQF_MDS_CLOSE;
 837         if (op_data->op_bias & MDS_HSM_RELEASE) {
 838                 req_fmt = &RQF_MDS_RELEASE_CLOSE;
 839
 840                 /* allocate a FID for volatile file */
 841                 rc = mdc_fid_alloc(NULL, exp, &op_data->op_fid2, op_data);
 842                 if (rc < 0) {
 843                         CERROR("%s: "DFID" failed to allocate FID: %d\n",
 844                                obd->obd_name, PFID(&op_data->op_fid1), rc);
 845                         /* save the errcode and proceed to close */
 846                         saved_rc = rc;
 847                 }
 848         }
 849
 850         *request = NULL;
 851         req = ptlrpc_request_alloc(class_exp2cliimp(exp), req_fmt);
 852         if (req == NULL)
 853                 RETURN(-ENOMEM);
 854
 855         mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
 856
 857         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_CLOSE);
 858         if (rc) {
 859                 ptlrpc_request_free(req);
 860                 RETURN(rc);
 861         }
 862
 863         /* To avoid a livelock (bug 7034), we need to send CLOSE RPCs to a
 864          * portal whose threads are not taking any DLM locks and are therefore
 865          * always progressing */
 866         req->rq_request_portal = MDS_READPAGE_PORTAL;
 867         ptlrpc_at_set_req_timeout(req);
 868
 869         /* Ensure that this close's handle is fixed up during replay. */
 870         if (likely(mod != NULL)) {
 871                 LASSERTF(mod->mod_open_req != NULL &&
 872                          mod->mod_open_req->rq_type != LI_POISON,
 873                          "POISONED open %p!\n", mod->mod_open_req);
 874
 875                 mod->mod_close_req = req;
 876
 877                 DEBUG_REQ(D_HA, mod->mod_open_req, "matched open");
 878                 /* We no longer want to preserve this open for replay even
 879                  * though the open was committed. b=3632, b=3633 */
 880                 spin_lock(&mod->mod_open_req->rq_lock);
 881                 mod->mod_open_req->rq_replay = 0;
 882                 spin_unlock(&mod->mod_open_req->rq_lock);
 883         } else {
 884                  CDEBUG(D_HA, "couldn't find open req; expecting close error\n");
 885         }
 886
 887         mdc_close_pack(req, op_data);
 888
 889         req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
 890                              obd->u.cli.cl_default_mds_easize);
 891
 892         ptlrpc_request_set_replen(req);
 893
 894         mdc_get_rpc_lock(obd->u.cli.cl_close_lock, NULL);
 895         rc = ptlrpc_queue_wait(req);
 896         mdc_put_rpc_lock(obd->u.cli.cl_close_lock, NULL);
 897
 898         if (req->rq_repmsg == NULL) {
 899                 CDEBUG(D_RPCTRACE, "request failed to send: %p, %d\n", req,
 900                        req->rq_status);
 901                 if (rc == 0)
 902                         rc = req->rq_status ?: -EIO;
 903         } else if (rc == 0 || rc == -EAGAIN) {
 904                 struct mdt_body *body;
 905
 906                 rc = lustre_msg_get_status(req->rq_repmsg);
 907                 if (lustre_msg_get_type(req->rq_repmsg) == PTL_RPC_MSG_ERR) {
 908                         DEBUG_REQ(D_ERROR, req, "type == PTL_RPC_MSG_ERR, err "
 909                                   "= %d", rc);
 910                         if (rc > 0)
 911                                 rc = -rc;
 912                 }
 913                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 914                 if (body == NULL)
 915                         rc = -EPROTO;
 916         } else if (rc == -ESTALE) {
 917                 /**
 918                  * it can be allowed error after 3633 if open was committed and
 919                  * server failed before close was sent. Let's check if mod
 920                  * exists and return no error in that case
 921                  */
 922                 if (mod) {
 923                         DEBUG_REQ(D_HA, req, "Reset ESTALE = %d", rc);
 924                         LASSERT(mod->mod_open_req != NULL);
 925                         if (mod->mod_open_req->rq_committed)
 926                                 rc = 0;
 927                 }
 928         }
 929
 930         if (mod) {
 931                 if (rc != 0)
 932                         mod->mod_close_req = NULL;
 933                 /* Since now, mod is accessed through open_req only,
 934                  * thus close req does not keep a reference on mod anymore. */
 935                 obd_mod_put(mod);
 936         }
 937         *request = req;
 938
 939         RETURN(rc < 0 ? rc : saved_rc);
 940 }
 941
 942 static int mdc_getpage(struct obd_export *exp, const struct lu_fid *fid,
 943                        __u64 offset, struct obd_capa *oc,
 944                        struct page **pages, int npages,
 945                        struct ptlrpc_request **request)
 946 {
 947         struct ptlrpc_request   *req;
 948         struct ptlrpc_bulk_desc *desc;
 949         int                      i;
 950         wait_queue_head_t        waitq;
 951         int                      resends = 0;
 952         struct l_wait_info       lwi;
 953         int                      rc;
 954         ENTRY;
 955
 956         *request = NULL;
 957         init_waitqueue_head(&waitq);
 958
 959 restart_bulk:
 960         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_READPAGE);
 961         if (req == NULL)
 962                 RETURN(-ENOMEM);
 963
 964         mdc_set_capa_size(req, &RMF_CAPA1, oc);
 965
 966         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_READPAGE);
 967         if (rc) {
 968                 ptlrpc_request_free(req);
 969                 RETURN(rc);
 970         }
 971
 972         req->rq_request_portal = MDS_READPAGE_PORTAL;
 973         ptlrpc_at_set_req_timeout(req);
 974
 975         desc = ptlrpc_prep_bulk_imp(req, npages, 1, BULK_PUT_SINK,
 976                                     MDS_BULK_PORTAL);
 977         if (desc == NULL) {
 978                 ptlrpc_request_free(req);
 979                 RETURN(-ENOMEM);
 980         }
 981
 982         /* NB req now owns desc and will free it when it gets freed */
 983         for (i = 0; i < npages; i++)
 984                 ptlrpc_prep_bulk_page_pin(desc, pages[i], 0, PAGE_CACHE_SIZE);
 985
 986         mdc_readdir_pack(req, offset, PAGE_CACHE_SIZE * npages, fid, oc);
 987
 988         ptlrpc_request_set_replen(req);
 989         rc = ptlrpc_queue_wait(req);
 990         if (rc) {
 991                 ptlrpc_req_finished(req);
 992                 if (rc != -ETIMEDOUT)
 993                         RETURN(rc);
 994
 995                 resends++;
 996                 if (!client_should_resend(resends, &exp->exp_obd->u.cli)) {
 997                         CERROR("%s: too many resend retries: rc = %d\n",
 998                                exp->exp_obd->obd_name, -EIO);
 999                         RETURN(-EIO);
1000                 }
1001                 lwi = LWI_TIMEOUT_INTR(cfs_time_seconds(resends), NULL, NULL,
1002                                        NULL);
1003                 l_wait_event(waitq, 0, &lwi);
1004
1005                 goto restart_bulk;
1006         }
1007
1008         rc = sptlrpc_cli_unwrap_bulk_read(req, req->rq_bulk,
1009                                           req->rq_bulk->bd_nob_transferred);
1010         if (rc < 0) {
1011                 ptlrpc_req_finished(req);
1012                 RETURN(rc);
1013         }
1014
1015         if (req->rq_bulk->bd_nob_transferred & ~LU_PAGE_MASK) {
1016                 CERROR("%s: unexpected bytes transferred: %d (%ld expected)\n",
1017                        exp->exp_obd->obd_name, req->rq_bulk->bd_nob_transferred,
1018                        PAGE_CACHE_SIZE * npages);
1019                 ptlrpc_req_finished(req);
1020                 RETURN(-EPROTO);
1021         }
1022
1023         *request = req;
1024         RETURN(0);
1025 }
1026
1027 static void mdc_release_page(struct page *page, int remove)
1028 {
1029         if (remove) {
1030                 lock_page(page);
1031                 if (likely(page->mapping != NULL))
1032                         truncate_complete_page(page->mapping, page);
1033                 unlock_page(page);
1034         }
1035         page_cache_release(page);
1036 }
1037
1038 static struct page *mdc_page_locate(struct address_space *mapping, __u64 *hash,
1039                                     __u64 *start, __u64 *end, int hash64)
1040 {
1041         /*
1042          * Complement of hash is used as an index so that
1043          * radix_tree_gang_lookup() can be used to find a page with starting
1044          * hash _smaller_ than one we are looking for.
1045          */
1046         unsigned long offset = hash_x_index(*hash, hash64);
1047         struct page *page;
1048         int found;
1049
1050         spin_lock_irq(&mapping->tree_lock);
1051         found = radix_tree_gang_lookup(&mapping->page_tree,
1052                                        (void **)&page, offset, 1);
1053         if (found > 0 && !radix_tree_exceptional_entry(page)) {
1054                 struct lu_dirpage *dp;
1055
1056                 page_cache_get(page);
1057                 spin_unlock_irq(&mapping->tree_lock);
1058                 /*
1059                  * In contrast to find_lock_page() we are sure that directory
1060                  * page cannot be truncated (while DLM lock is held) and,
1061                  * hence, can avoid restart.
1062                  *
1063                  * In fact, page cannot be locked here at all, because
1064                  * mdc_read_page_remote does synchronous io.
1065                  */
1066                 wait_on_page_locked(page);
1067                 if (PageUptodate(page)) {
1068                         dp = kmap(page);
1069                         if (BITS_PER_LONG == 32 && hash64) {
1070                                 *start = le64_to_cpu(dp->ldp_hash_start) >> 32;
1071                                 *end   = le64_to_cpu(dp->ldp_hash_end) >> 32;
1072                                 *hash  = *hash >> 32;
1073                         } else {
1074                                 *start = le64_to_cpu(dp->ldp_hash_start);
1075                                 *end   = le64_to_cpu(dp->ldp_hash_end);
1076                         }
1077                         if (unlikely(*start == 1 && *hash == 0))
1078                                 *hash = *start;
1079                         else
1080                                 LASSERTF(*start <= *hash, "start = "LPX64
1081                                          ",end = "LPX64",hash = "LPX64"\n",
1082                                          *start, *end, *hash);
1083                         CDEBUG(D_VFSTRACE, "offset %lx ["LPX64" "LPX64"],"
1084                               " hash "LPX64"\n", offset, *start, *end, *hash);
1085                         if (*hash > *end) {
1086                                 kunmap(page);
1087                                 mdc_release_page(page, 0);
1088                                 page = NULL;
1089                         } else if (*end != *start && *hash == *end) {
1090                                 /*
1091                                  * upon hash collision, remove this page,
1092                                  * otherwise put page reference, and
1093                                  * mdc_read_page_remote() will issue RPC to
1094                                  * fetch the page we want.
1095                                  */
1096                                 kunmap(page);
1097                                 mdc_release_page(page,
1098                                     le32_to_cpu(dp->ldp_flags) & LDF_COLLIDE);
1099                                 page = NULL;
1100                         }
1101                 } else {
1102                         page_cache_release(page);
1103                         page = ERR_PTR(-EIO);
1104                 }
1105         } else {
1106                 spin_unlock_irq(&mapping->tree_lock);
1107                 page = NULL;
1108         }
1109         return page;
1110 }
1111
1112 /*
1113  * Adjust a set of pages, each page containing an array of lu_dirpages,
1114  * so that each page can be used as a single logical lu_dirpage.
1115  *
1116  * A lu_dirpage is laid out as follows, where s = ldp_hash_start,
1117  * e = ldp_hash_end, f = ldp_flags, p = padding, and each "ent" is a
1118  * struct lu_dirent.  It has size up to LU_PAGE_SIZE. The ldp_hash_end
1119  * value is used as a cookie to request the next lu_dirpage in a
1120  * directory listing that spans multiple pages (two in this example):
1121  *   ________
1122  *  |        |
1123  * .|--------v-------   -----.
1124  * |s|e|f|p|ent|ent| ... |ent|
1125  * '--|--------------   -----'   Each PAGE contains a single
1126  *    '------.                   lu_dirpage.
1127  * .---------v-------   -----.
1128  * |s|e|f|p|ent| 0 | ... | 0 |
1129  * '-----------------   -----'
1130  *
1131  * However, on hosts where the native VM page size (PAGE_CACHE_SIZE) is
1132  * larger than LU_PAGE_SIZE, a single host page may contain multiple
1133  * lu_dirpages. After reading the lu_dirpages from the MDS, the
1134  * ldp_hash_end of the first lu_dirpage refers to the one immediately
1135  * after it in the same PAGE (arrows simplified for brevity, but
1136  * in general e0==s1, e1==s2, etc.):
1137  *
1138  * .--------------------   -----.
1139  * |s0|e0|f0|p|ent|ent| ... |ent|
1140  * |---v----------------   -----|
1141  * |s1|e1|f1|p|ent|ent| ... |ent|
1142  * |---v----------------   -----|  Here, each PAGE contains
1143  *             ...                 multiple lu_dirpages.
1144  * |---v----------------   -----|
1145  * |s'|e'|f'|p|ent|ent| ... |ent|
1146  * '---|----------------   -----'
1147  *     v
1148  * .----------------------------.
1149  * |        next PAGE           |
1150  *
1151  * This structure is transformed into a single logical lu_dirpage as follows:
1152  *
1153  * - Replace e0 with e' so the request for the next lu_dirpage gets the page
1154  *   labeled 'next PAGE'.
1155  *
1156  * - Copy the LDF_COLLIDE flag from f' to f0 to correctly reflect whether
1157  *   a hash collision with the next page exists.
1158  *
1159  * - Adjust the lde_reclen of the ending entry of each lu_dirpage to span
1160  *   to the first entry of the next lu_dirpage.
1161  */
1162 #if PAGE_CACHE_SIZE > LU_PAGE_SIZE
1163 static void mdc_adjust_dirpages(struct page **pages, int cfs_pgs, int lu_pgs)
1164 {
1165         int i;
1166
1167         for (i = 0; i < cfs_pgs; i++) {
1168                 struct lu_dirpage       *dp = kmap(pages[i]);
1169                 struct lu_dirpage       *first = dp;
1170                 struct lu_dirent        *end_dirent = NULL;
1171                 struct lu_dirent        *ent;
1172                 __u64           hash_end = le64_to_cpu(dp->ldp_hash_end);
1173                 __u32           flags = le32_to_cpu(dp->ldp_flags);
1174
1175                 while (--lu_pgs > 0) {
1176                         ent = lu_dirent_start(dp);
1177                         for (end_dirent = ent; ent != NULL;
1178                              end_dirent = ent, ent = lu_dirent_next(ent));
1179
1180                         /* Advance dp to next lu_dirpage. */
1181                         dp = (struct lu_dirpage *)((char *)dp + LU_PAGE_SIZE);
1182
1183                         /* Check if we've reached the end of the PAGE. */
1184                         if (!((unsigned long)dp & ~PAGE_MASK))
1185                                 break;
1186
1187                         /* Save the hash and flags of this lu_dirpage. */
1188                         hash_end = le64_to_cpu(dp->ldp_hash_end);
1189                         flags = le32_to_cpu(dp->ldp_flags);
1190
1191                         /* Check if lu_dirpage contains no entries. */
1192                         if (end_dirent == NULL)
1193                                 break;
1194
1195                         /* Enlarge the end entry lde_reclen from 0 to
1196                          * first entry of next lu_dirpage. */
1197                         LASSERT(le16_to_cpu(end_dirent->lde_reclen) == 0);
1198                         end_dirent->lde_reclen =
1199                                 cpu_to_le16((char *)(dp->ldp_entries) -
1200                                             (char *)end_dirent);
1201                 }
1202
1203                 first->ldp_hash_end = hash_end;
1204                 first->ldp_flags &= ~cpu_to_le32(LDF_COLLIDE);
1205                 first->ldp_flags |= flags & cpu_to_le32(LDF_COLLIDE);
1206
1207                 kunmap(pages[i]);
1208         }
1209         LASSERTF(lu_pgs == 0, "left = %d\n", lu_pgs);
1210 }
1211 #else
1212 #define mdc_adjust_dirpages(pages, cfs_pgs, lu_pgs) do {} while (0)
1213 #endif  /* PAGE_CACHE_SIZE > LU_PAGE_SIZE */
1214
1215 /* parameters for readdir page */
1216 struct readpage_param {
1217         struct md_op_data       *rp_mod;
1218         __u64                   rp_off;
1219         int                     rp_hash64;
1220         struct obd_export       *rp_exp;
1221         struct md_callback      *rp_cb;
1222 };
1223
1224 #ifndef HAVE_DELETE_FROM_PAGE_CACHE
1225 static inline void delete_from_page_cache(struct page *page)
1226 {
1227         remove_from_page_cache(page);
1228         page_cache_release(page);
1229 }
1230 #endif
1231
1232 /**
1233  * Read pages from server.
1234  *
1235  * Page in MDS_READPAGE RPC is packed in LU_PAGE_SIZE, and each page contains
1236  * a header lu_dirpage which describes the start/end hash, and whether this
1237  * page is empty (contains no dir entry) or hash collide with next page.
1238  * After client receives reply, several pages will be integrated into dir page
1239  * in PAGE_SIZE (if PAGE_SIZE greater than LU_PAGE_SIZE), and the
1240  * lu_dirpage for this integrated page will be adjusted.
1241  **/
1242 static int mdc_read_page_remote(void *data, struct page *page0)
1243 {
1244         struct readpage_param   *rp = data;
1245         struct page             **page_pool;
1246         struct page             *page;
1247         struct lu_dirpage       *dp;
1248         int                     rd_pgs = 0; /* number of pages read actually */
1249         int                     npages;
1250         struct md_op_data       *op_data = rp->rp_mod;
1251         struct ptlrpc_request   *req;
1252         int                     max_pages = op_data->op_max_pages;
1253         struct inode            *inode;
1254         struct lu_fid           *fid;
1255         int                     i;
1256         int                     rc;
1257         ENTRY;
1258
1259         LASSERT(max_pages > 0 && max_pages <= PTLRPC_MAX_BRW_PAGES);
1260         inode = op_data->op_data;
1261         fid = &op_data->op_fid1;
1262         LASSERT(inode != NULL);
1263
1264         OBD_ALLOC(page_pool, sizeof(page_pool[0]) * max_pages);
1265         if (page_pool != NULL) {
1266                 page_pool[0] = page0;
1267         } else {
1268                 page_pool = &page0;
1269                 max_pages = 1;
1270         }
1271
1272         for (npages = 1; npages < max_pages; npages++) {
1273                 page = page_cache_alloc_cold(inode->i_mapping);
1274                 if (page == NULL)
1275                         break;
1276                 page_pool[npages] = page;
1277         }
1278
1279         rc = mdc_getpage(rp->rp_exp, fid, rp->rp_off, op_data->op_capa1,
1280                          page_pool, npages, &req);
1281         if (rc < 0) {
1282                 /* page0 is special, which was added into page cache early */
1283                 delete_from_page_cache(page0);
1284         } else {
1285                 int lu_pgs;
1286
1287                 rd_pgs = (req->rq_bulk->bd_nob_transferred +
1288                             PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1289                 lu_pgs = req->rq_bulk->bd_nob_transferred >>
1290                                                         LU_PAGE_SHIFT;
1291                 LASSERT(!(req->rq_bulk->bd_nob_transferred & ~LU_PAGE_MASK));
1292
1293                 CDEBUG(D_INODE, "read %d(%d) pages\n", rd_pgs, lu_pgs);
1294
1295                 mdc_adjust_dirpages(page_pool, rd_pgs, lu_pgs);
1296
1297                 SetPageUptodate(page0);
1298         }
1299         unlock_page(page0);
1300
1301         ptlrpc_req_finished(req);
1302         CDEBUG(D_CACHE, "read %d/%d pages\n", rd_pgs, npages);
1303         for (i = 1; i < npages; i++) {
1304                 unsigned long   offset;
1305                 __u64           hash;
1306                 int ret;
1307
1308                 page = page_pool[i];
1309
1310                 if (rc < 0 || i >= rd_pgs) {
1311                         page_cache_release(page);
1312                         continue;
1313                 }
1314
1315                 SetPageUptodate(page);
1316
1317                 dp = kmap(page);
1318                 hash = le64_to_cpu(dp->ldp_hash_start);
1319                 kunmap(page);
1320
1321                 offset = hash_x_index(hash, rp->rp_hash64);
1322
1323                 prefetchw(&page->flags);
1324                 ret = add_to_page_cache_lru(page, inode->i_mapping, offset,
1325                                             GFP_KERNEL);
1326                 if (ret == 0)
1327                         unlock_page(page);
1328                 else
1329                         CDEBUG(D_VFSTRACE, "page %lu add to page cache failed:"
1330                                " rc = %d\n", offset, ret);
1331                 page_cache_release(page);
1332         }
1333
1334         if (page_pool != &page0)
1335                 OBD_FREE(page_pool, sizeof(page_pool[0]) * max_pages);
1336
1337         RETURN(rc);
1338 }
1339
1340 /**
1341  * Read dir page from cache first, if it can not find it, read it from
1342  * server and add into the cache.
1343  *
1344  * \param[in] exp       MDC export
1345  * \param[in] op_data   client MD stack parameters, transfering parameters
1346  *                      between different layers on client MD stack.
1347  * \param[in] cb_op     callback required for ldlm lock enqueue during
1348  *                      read page
1349  * \param[in] hash_offset the hash offset of the page to be read
1350  * \param[in] ppage     the page to be read
1351  *
1352  * retval               = 0 get the page successfully
1353  *                      errno(<0) get the page failed
1354  */
1355 static int mdc_read_page(struct obd_export *exp, struct md_op_data *op_data,
1356                          struct md_callback *cb_op, __u64 hash_offset,
1357                          struct page **ppage)
1358 {
1359         struct lookup_intent    it = { .it_op = IT_READDIR };
1360         struct page             *page;
1361         struct inode            *dir = op_data->op_data;
1362         struct address_space    *mapping;
1363         struct lu_dirpage       *dp;
1364         __u64                   start = 0;
1365         __u64                   end = 0;
1366         struct lustre_handle    lockh;
1367         struct ptlrpc_request   *enq_req = NULL;
1368         struct readpage_param   rp_param;
1369         int rc;
1370
1371         ENTRY;
1372
1373         *ppage = NULL;
1374
1375         LASSERT(dir != NULL);
1376         mapping = dir->i_mapping;
1377
1378         rc = mdc_intent_lock(exp, op_data, &it, &enq_req,
1379                              cb_op->md_blocking_ast, 0);
1380         if (enq_req != NULL)
1381                 ptlrpc_req_finished(enq_req);
1382
1383         if (rc < 0) {
1384                 CERROR("%s: "DFID" lock enqueue fails: rc = %d\n",
1385                        exp->exp_obd->obd_name, PFID(&op_data->op_fid1), rc);
1386                 RETURN(rc);
1387         }
1388
1389         rc = 0;
1390         mdc_set_lock_data(exp, &it.d.lustre.it_lock_handle, dir, NULL);
1391
1392         rp_param.rp_off = hash_offset;
1393         rp_param.rp_hash64 = op_data->op_cli_flags & CLI_HASH64;
1394         page = mdc_page_locate(mapping, &rp_param.rp_off, &start, &end,
1395                                rp_param.rp_hash64);
1396         if (IS_ERR(page)) {
1397                 CERROR("%s: dir page locate: "DFID" at "LPU64": rc %ld\n",
1398                        exp->exp_obd->obd_name, PFID(&op_data->op_fid1),
1399                        rp_param.rp_off, PTR_ERR(page));
1400                 GOTO(out_unlock, rc = PTR_ERR(page));
1401         } else if (page != NULL) {
1402                 /*
1403                  * XXX nikita: not entirely correct handling of a corner case:
1404                  * suppose hash chain of entries with hash value HASH crosses
1405                  * border between pages P0 and P1. First both P0 and P1 are
1406                  * cached, seekdir() is called for some entry from the P0 part
1407                  * of the chain. Later P0 goes out of cache. telldir(HASH)
1408                  * happens and finds P1, as it starts with matching hash
1409                  * value. Remaining entries from P0 part of the chain are
1410                  * skipped. (Is that really a bug?)
1411                  *
1412                  * Possible solutions: 0. don't cache P1 is such case, handle
1413                  * it as an "overflow" page. 1. invalidate all pages at
1414                  * once. 2. use HASH|1 as an index for P1.
1415                  */
1416                 GOTO(hash_collision, page);
1417         }
1418
1419         rp_param.rp_exp = exp;
1420         rp_param.rp_mod = op_data;
1421         page = read_cache_page(mapping,
1422                                hash_x_index(rp_param.rp_off,
1423                                             rp_param.rp_hash64),
1424                                mdc_read_page_remote, &rp_param);
1425         if (IS_ERR(page)) {
1426                 CDEBUG(D_INFO, "%s: read cache page: "DFID" at "LPU64": %ld\n",
1427                        exp->exp_obd->obd_name, PFID(&op_data->op_fid1),
1428                        rp_param.rp_off, PTR_ERR(page));
1429                 GOTO(out_unlock, rc = PTR_ERR(page));
1430         }
1431
1432         wait_on_page_locked(page);
1433         (void)kmap(page);
1434         if (!PageUptodate(page)) {
1435                 CERROR("%s: page not updated: "DFID" at "LPU64": rc %d\n",
1436                        exp->exp_obd->obd_name, PFID(&op_data->op_fid1),
1437                        rp_param.rp_off, -5);
1438                 goto fail;
1439         }
1440         if (!PageChecked(page))
1441                 SetPageChecked(page);
1442         if (PageError(page)) {
1443                 CERROR("%s: page error: "DFID" at "LPU64": rc %d\n",
1444                        exp->exp_obd->obd_name, PFID(&op_data->op_fid1),
1445                        rp_param.rp_off, -5);
1446                 goto fail;
1447         }
1448
1449 hash_collision:
1450         dp = page_address(page);
1451         if (BITS_PER_LONG == 32 && rp_param.rp_hash64) {
1452                 start = le64_to_cpu(dp->ldp_hash_start) >> 32;
1453                 end   = le64_to_cpu(dp->ldp_hash_end) >> 32;
1454                 rp_param.rp_off = hash_offset >> 32;
1455         } else {
1456                 start = le64_to_cpu(dp->ldp_hash_start);
1457                 end   = le64_to_cpu(dp->ldp_hash_end);
1458                 rp_param.rp_off = hash_offset;
1459         }
1460         if (end == start) {
1461                 LASSERT(start == rp_param.rp_off);
1462                 CWARN("Page-wide hash collision: %#lx\n", (unsigned long)end);
1463 #if BITS_PER_LONG == 32
1464                 CWARN("Real page-wide hash collision at ["LPU64" "LPU64"] with "
1465                       "hash "LPU64"\n", le64_to_cpu(dp->ldp_hash_start),
1466                       le64_to_cpu(dp->ldp_hash_end), hash_offset);
1467 #endif
1468
1469                 /*
1470                  * Fetch whole overflow chain...
1471                  *
1472                  * XXX not yet.
1473                  */
1474                 goto fail;
1475         }
1476         *ppage = page;
1477 out_unlock:
1478         lockh.cookie = it.d.lustre.it_lock_handle;
1479         ldlm_lock_decref(&lockh, it.d.lustre.it_lock_mode);
1480         it.d.lustre.it_lock_handle = 0;
1481         return rc;
1482 fail:
1483         kunmap(page);
1484         mdc_release_page(page, 1);
1485         rc = -EIO;
1486         goto out_unlock;
1487 }
1488
1489
1490 static int mdc_statfs(const struct lu_env *env,
1491                       struct obd_export *exp, struct obd_statfs *osfs,
1492                       __u64 max_age, __u32 flags)
1493 {
1494         struct obd_device     *obd = class_exp2obd(exp);
1495         struct ptlrpc_request *req;
1496         struct obd_statfs     *msfs;
1497         struct obd_import     *imp = NULL;
1498         int                    rc;
1499         ENTRY;
1500
1501         /*
1502          * Since the request might also come from lprocfs, so we need
1503          * sync this with client_disconnect_export Bug15684
1504          */
1505         down_read(&obd->u.cli.cl_sem);
1506         if (obd->u.cli.cl_import)
1507                 imp = class_import_get(obd->u.cli.cl_import);
1508         up_read(&obd->u.cli.cl_sem);
1509         if (!imp)
1510                 RETURN(-ENODEV);
1511
1512         req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_STATFS,
1513                                         LUSTRE_MDS_VERSION, MDS_STATFS);
1514         if (req == NULL)
1515                 GOTO(output, rc = -ENOMEM);
1516
1517         ptlrpc_request_set_replen(req);
1518
1519         if (flags & OBD_STATFS_NODELAY) {
1520                 /* procfs requests not want stay in wait for avoid deadlock */
1521                 req->rq_no_resend = 1;
1522                 req->rq_no_delay = 1;
1523         }
1524
1525         rc = ptlrpc_queue_wait(req);
1526         if (rc) {
1527                 /* check connection error first */
1528                 if (imp->imp_connect_error)
1529                         rc = imp->imp_connect_error;
1530                 GOTO(out, rc);
1531         }
1532
1533         msfs = req_capsule_server_get(&req->rq_pill, &RMF_OBD_STATFS);
1534         if (msfs == NULL)
1535                 GOTO(out, rc = -EPROTO);
1536
1537         *osfs = *msfs;
1538         EXIT;
1539 out:
1540         ptlrpc_req_finished(req);
1541 output:
1542         class_import_put(imp);
1543         return rc;
1544 }
1545
1546 static int mdc_ioc_fid2path(struct obd_export *exp, struct getinfo_fid2path *gf)
1547 {
1548         __u32 keylen, vallen;
1549         void *key;
1550         int rc;
1551
1552         if (gf->gf_pathlen > PATH_MAX)
1553                 RETURN(-ENAMETOOLONG);
1554         if (gf->gf_pathlen < 2)
1555                 RETURN(-EOVERFLOW);
1556
1557         /* Key is KEY_FID2PATH + getinfo_fid2path description */
1558         keylen = cfs_size_round(sizeof(KEY_FID2PATH)) + sizeof(*gf);
1559         OBD_ALLOC(key, keylen);
1560         if (key == NULL)
1561                 RETURN(-ENOMEM);
1562         memcpy(key, KEY_FID2PATH, sizeof(KEY_FID2PATH));
1563         memcpy(key + cfs_size_round(sizeof(KEY_FID2PATH)), gf, sizeof(*gf));
1564
1565         CDEBUG(D_IOCTL, "path get "DFID" from "LPU64" #%d\n",
1566                PFID(&gf->gf_fid), gf->gf_recno, gf->gf_linkno);
1567
1568         if (!fid_is_sane(&gf->gf_fid))
1569                 GOTO(out, rc = -EINVAL);
1570
1571         /* Val is struct getinfo_fid2path result plus path */
1572         vallen = sizeof(*gf) + gf->gf_pathlen;
1573
1574         rc = obd_get_info(NULL, exp, keylen, key, &vallen, gf);
1575         if (rc != 0 && rc != -EREMOTE)
1576                 GOTO(out, rc);
1577
1578         if (vallen <= sizeof(*gf))
1579                 GOTO(out, rc = -EPROTO);
1580         else if (vallen > sizeof(*gf) + gf->gf_pathlen)
1581                 GOTO(out, rc = -EOVERFLOW);
1582
1583         CDEBUG(D_IOCTL, "path get "DFID" from "LPU64" #%d\n%s\n",
1584                PFID(&gf->gf_fid), gf->gf_recno, gf->gf_linkno, gf->gf_path);
1585
1586 out:
1587         OBD_FREE(key, keylen);
1588         return rc;
1589 }
1590
1591 static int mdc_ioc_hsm_progress(struct obd_export *exp,
1592                                 struct hsm_progress_kernel *hpk)
1593 {
1594         struct obd_import               *imp = class_exp2cliimp(exp);
1595         struct hsm_progress_kernel      *req_hpk;
1596         struct ptlrpc_request           *req;
1597         int                              rc;
1598         ENTRY;
1599
1600         req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_HSM_PROGRESS,
1601                                         LUSTRE_MDS_VERSION, MDS_HSM_PROGRESS);
1602         if (req == NULL)
1603                 GOTO(out, rc = -ENOMEM);
1604
1605         mdc_pack_body(req, NULL, NULL, OBD_MD_FLRMTPERM, 0, -1, 0);
1606
1607         /* Copy hsm_progress struct */
1608         req_hpk = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_PROGRESS);
1609         if (req_hpk == NULL)
1610                 GOTO(out, rc = -EPROTO);
1611
1612         *req_hpk = *hpk;
1613         req_hpk->hpk_errval = lustre_errno_hton(hpk->hpk_errval);
1614
1615         ptlrpc_request_set_replen(req);
1616
1617         mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
1618         rc = ptlrpc_queue_wait(req);
1619         mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
1620
1621         GOTO(out, rc);
1622 out:
1623         ptlrpc_req_finished(req);
1624         return rc;
1625 }
1626
1627 static int mdc_ioc_hsm_ct_register(struct obd_import *imp, __u32 archives)
1628 {
1629         __u32                   *archive_mask;
1630         struct ptlrpc_request   *req;
1631         int                      rc;
1632         ENTRY;
1633
1634         req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_HSM_CT_REGISTER,
1635                                         LUSTRE_MDS_VERSION,
1636                                         MDS_HSM_CT_REGISTER);
1637         if (req == NULL)
1638                 GOTO(out, rc = -ENOMEM);
1639
1640         mdc_pack_body(req, NULL, NULL, OBD_MD_FLRMTPERM, 0, -1, 0);
1641
1642         /* Copy hsm_progress struct */
1643         archive_mask = req_capsule_client_get(&req->rq_pill,
1644                                               &RMF_MDS_HSM_ARCHIVE);
1645         if (archive_mask == NULL)
1646                 GOTO(out, rc = -EPROTO);
1647
1648         *archive_mask = archives;
1649
1650         ptlrpc_request_set_replen(req);
1651
1652         rc = mdc_queue_wait(req);
1653         GOTO(out, rc);
1654 out:
1655         ptlrpc_req_finished(req);
1656         return rc;
1657 }
1658
1659 static int mdc_ioc_hsm_current_action(struct obd_export *exp,
1660                                       struct md_op_data *op_data)
1661 {
1662         struct hsm_current_action       *hca = op_data->op_data;
1663         struct hsm_current_action       *req_hca;
1664         struct ptlrpc_request           *req;
1665         int                              rc;
1666         ENTRY;
1667
1668         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1669                                    &RQF_MDS_HSM_ACTION);
1670         if (req == NULL)
1671                 RETURN(-ENOMEM);
1672
1673         mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
1674
1675         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_ACTION);
1676         if (rc) {
1677                 ptlrpc_request_free(req);
1678                 RETURN(rc);
1679         }
1680
1681         mdc_pack_body(req, &op_data->op_fid1, op_data->op_capa1,
1682                       OBD_MD_FLRMTPERM, 0, op_data->op_suppgids[0], 0);
1683
1684         ptlrpc_request_set_replen(req);
1685
1686         rc = mdc_queue_wait(req);
1687         if (rc)
1688                 GOTO(out, rc);
1689
1690         req_hca = req_capsule_server_get(&req->rq_pill,
1691                                          &RMF_MDS_HSM_CURRENT_ACTION);
1692         if (req_hca == NULL)
1693                 GOTO(out, rc = -EPROTO);
1694
1695         *hca = *req_hca;
1696
1697         EXIT;
1698 out:
1699         ptlrpc_req_finished(req);
1700         return rc;
1701 }
1702
1703 static int mdc_ioc_hsm_ct_unregister(struct obd_import *imp)
1704 {
1705         struct ptlrpc_request   *req;
1706         int                      rc;
1707         ENTRY;
1708
1709         req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_HSM_CT_UNREGISTER,
1710                                         LUSTRE_MDS_VERSION,
1711                                         MDS_HSM_CT_UNREGISTER);
1712         if (req == NULL)
1713                 GOTO(out, rc = -ENOMEM);
1714
1715         mdc_pack_body(req, NULL, NULL, OBD_MD_FLRMTPERM, 0, -1, 0);
1716
1717         ptlrpc_request_set_replen(req);
1718
1719         rc = mdc_queue_wait(req);
1720         GOTO(out, rc);
1721 out:
1722         ptlrpc_req_finished(req);
1723         return rc;
1724 }
1725
1726 static int mdc_ioc_hsm_state_get(struct obd_export *exp,
1727                                  struct md_op_data *op_data)
1728 {
1729         struct hsm_user_state   *hus = op_data->op_data;
1730         struct hsm_user_state   *req_hus;
1731         struct ptlrpc_request   *req;
1732         int                      rc;
1733         ENTRY;
1734
1735         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1736                                    &RQF_MDS_HSM_STATE_GET);
1737         if (req == NULL)
1738                 RETURN(-ENOMEM);
1739
1740         mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
1741
1742         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_STATE_GET);
1743         if (rc != 0) {
1744                 ptlrpc_request_free(req);
1745                 RETURN(rc);
1746         }
1747
1748         mdc_pack_body(req, &op_data->op_fid1, op_data->op_capa1,
1749                       OBD_MD_FLRMTPERM, 0, op_data->op_suppgids[0], 0);
1750
1751         ptlrpc_request_set_replen(req);
1752
1753         rc = mdc_queue_wait(req);
1754         if (rc)
1755                 GOTO(out, rc);
1756
1757         req_hus = req_capsule_server_get(&req->rq_pill, &RMF_HSM_USER_STATE);
1758         if (req_hus == NULL)
1759                 GOTO(out, rc = -EPROTO);
1760
1761         *hus = *req_hus;
1762
1763         EXIT;
1764 out:
1765         ptlrpc_req_finished(req);
1766         return rc;
1767 }
1768
1769 static int mdc_ioc_hsm_state_set(struct obd_export *exp,
1770                                  struct md_op_data *op_data)
1771 {
1772         struct hsm_state_set    *hss = op_data->op_data;
1773         struct hsm_state_set    *req_hss;
1774         struct ptlrpc_request   *req;
1775         int                      rc;
1776         ENTRY;
1777
1778         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1779                                    &RQF_MDS_HSM_STATE_SET);
1780         if (req == NULL)
1781                 RETURN(-ENOMEM);
1782
1783         mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
1784
1785         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_STATE_SET);
1786         if (rc) {
1787                 ptlrpc_request_free(req);
1788                 RETURN(rc);
1789         }
1790
1791         mdc_pack_body(req, &op_data->op_fid1, op_data->op_capa1,
1792                       OBD_MD_FLRMTPERM, 0, op_data->op_suppgids[0], 0);
1793
1794         /* Copy states */
1795         req_hss = req_capsule_client_get(&req->rq_pill, &RMF_HSM_STATE_SET);
1796         if (req_hss == NULL)
1797                 GOTO(out, rc = -EPROTO);
1798         *req_hss = *hss;
1799
1800         ptlrpc_request_set_replen(req);
1801
1802         mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
1803         rc = ptlrpc_queue_wait(req);
1804         mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
1805
1806         GOTO(out, rc);
1807 out:
1808         ptlrpc_req_finished(req);
1809         return rc;
1810 }
1811
1812 static int mdc_ioc_hsm_request(struct obd_export *exp,
1813                                struct hsm_user_request *hur)
1814 {
1815         struct obd_import       *imp = class_exp2cliimp(exp);
1816         struct ptlrpc_request   *req;
1817         struct hsm_request      *req_hr;
1818         struct hsm_user_item    *req_hui;
1819         char                    *req_opaque;
1820         int                      rc;
1821         ENTRY;
1822
1823         req = ptlrpc_request_alloc(imp, &RQF_MDS_HSM_REQUEST);
1824         if (req == NULL)
1825                 GOTO(out, rc = -ENOMEM);
1826
1827         req_capsule_set_size(&req->rq_pill, &RMF_MDS_HSM_USER_ITEM, RCL_CLIENT,
1828                              hur->hur_request.hr_itemcount
1829                              * sizeof(struct hsm_user_item));
1830         req_capsule_set_size(&req->rq_pill, &RMF_GENERIC_DATA, RCL_CLIENT,
1831                              hur->hur_request.hr_data_len);
1832
1833         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_REQUEST);
1834         if (rc) {
1835                 ptlrpc_request_free(req);
1836                 RETURN(rc);
1837         }
1838
1839         mdc_pack_body(req, NULL, NULL, OBD_MD_FLRMTPERM, 0, -1, 0);
1840
1841         /* Copy hsm_request struct */
1842         req_hr = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_REQUEST);
1843         if (req_hr == NULL)
1844                 GOTO(out, rc = -EPROTO);
1845         *req_hr = hur->hur_request;
1846
1847         /* Copy hsm_user_item structs */
1848         req_hui = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_USER_ITEM);
1849         if (req_hui == NULL)
1850                 GOTO(out, rc = -EPROTO);
1851         memcpy(req_hui, hur->hur_user_item,
1852                hur->hur_request.hr_itemcount * sizeof(struct hsm_user_item));
1853
1854         /* Copy opaque field */
1855         req_opaque = req_capsule_client_get(&req->rq_pill, &RMF_GENERIC_DATA);
1856         if (req_opaque == NULL)
1857                 GOTO(out, rc = -EPROTO);
1858         memcpy(req_opaque, hur_data(hur), hur->hur_request.hr_data_len);
1859
1860         ptlrpc_request_set_replen(req);
1861
1862         mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
1863         rc = ptlrpc_queue_wait(req);
1864         mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
1865
1866         GOTO(out, rc);
1867
1868 out:
1869         ptlrpc_req_finished(req);
1870         return rc;
1871 }
1872
1873 static struct kuc_hdr *changelog_kuc_hdr(char *buf, size_t len, __u32 flags)
1874 {
1875         struct kuc_hdr *lh = (struct kuc_hdr *)buf;
1876
1877         LASSERT(len <= KUC_CHANGELOG_MSG_MAXSIZE);
1878
1879         lh->kuc_magic = KUC_MAGIC;
1880         lh->kuc_transport = KUC_TRANSPORT_CHANGELOG;
1881         lh->kuc_flags = flags;
1882         lh->kuc_msgtype = CL_RECORD;
1883         lh->kuc_msglen = len;
1884         return lh;
1885 }
1886
1887 struct changelog_show {
1888         __u64                            cs_startrec;
1889         enum changelog_send_flag         cs_flags;
1890         struct file                     *cs_fp;
1891         char                            *cs_buf;
1892         struct obd_device               *cs_obd;
1893 };
1894
1895 static inline char *cs_obd_name(struct changelog_show *cs)
1896 {
1897         return cs->cs_obd->obd_name;
1898 }
1899
1900 static int changelog_kkuc_cb(const struct lu_env *env, struct llog_handle *llh,
1901                              struct llog_rec_hdr *hdr, void *data)
1902 {
1903         struct changelog_show           *cs = data;
1904         struct llog_changelog_rec       *rec = (struct llog_changelog_rec *)hdr;
1905         struct kuc_hdr                  *lh;
1906         size_t                           len;
1907         int                              rc;
1908         ENTRY;
1909
1910         if (rec->cr_hdr.lrh_type != CHANGELOG_REC) {
1911                 rc = -EINVAL;
1912                 CERROR("%s: not a changelog rec %x/%d: rc = %d\n",
1913                        cs_obd_name(cs), rec->cr_hdr.lrh_type,
1914                        rec->cr.cr_type, rc);
1915                 RETURN(rc);
1916         }
1917
1918         if (rec->cr.cr_index < cs->cs_startrec) {
1919                 /* Skip entries earlier than what we are interested in */
1920                 CDEBUG(D_HSM, "rec="LPU64" start="LPU64"\n",
1921                        rec->cr.cr_index, cs->cs_startrec);
1922                 RETURN(0);
1923         }
1924
1925         CDEBUG(D_HSM, LPU64" %02d%-5s "LPU64" 0x%x t="DFID" p="DFID" %.*s\n",
1926                rec->cr.cr_index, rec->cr.cr_type,
1927                changelog_type2str(rec->cr.cr_type), rec->cr.cr_time,
1928                rec->cr.cr_flags & CLF_FLAGMASK,
1929                PFID(&rec->cr.cr_tfid), PFID(&rec->cr.cr_pfid),
1930                rec->cr.cr_namelen, changelog_rec_name(&rec->cr));
1931
1932         len = sizeof(*lh) + changelog_rec_size(&rec->cr) + rec->cr.cr_namelen;
1933
1934         /* Set up the message */
1935         lh = changelog_kuc_hdr(cs->cs_buf, len, cs->cs_flags);
1936         memcpy(lh + 1, &rec->cr, len - sizeof(*lh));
1937
1938         rc = libcfs_kkuc_msg_put(cs->cs_fp, lh);
1939         CDEBUG(D_HSM, "kucmsg fp %p len %zu rc %d\n", cs->cs_fp, len, rc);
1940
1941         RETURN(rc);
1942 }
1943
1944 static int mdc_changelog_send_thread(void *csdata)
1945 {
1946         struct changelog_show   *cs = csdata;
1947         struct llog_ctxt        *ctxt = NULL;
1948         struct llog_handle      *llh = NULL;
1949         struct kuc_hdr          *kuch;
1950         enum llog_flag           flags = LLOG_F_IS_CAT;
1951         int                      rc;
1952
1953         CDEBUG(D_HSM, "changelog to fp=%p start "LPU64"\n",
1954                cs->cs_fp, cs->cs_startrec);
1955
1956         OBD_ALLOC(cs->cs_buf, KUC_CHANGELOG_MSG_MAXSIZE);
1957         if (cs->cs_buf == NULL)
1958                 GOTO(out, rc = -ENOMEM);
1959
1960         /* Set up the remote catalog handle */
1961         ctxt = llog_get_context(cs->cs_obd, LLOG_CHANGELOG_REPL_CTXT);
1962         if (ctxt == NULL)
1963                 GOTO(out, rc = -ENOENT);
1964         rc = llog_open(NULL, ctxt, &llh, NULL, CHANGELOG_CATALOG,
1965                        LLOG_OPEN_EXISTS);
1966         if (rc) {
1967                 CERROR("%s: fail to open changelog catalog: rc = %d\n",
1968                        cs_obd_name(cs), rc);
1969                 GOTO(out, rc);
1970         }
1971
1972         if (cs->cs_flags & CHANGELOG_FLAG_JOBID)
1973                 flags |= LLOG_F_EXT_JOBID;
1974
1975         rc = llog_init_handle(NULL, llh, flags, NULL);
1976         if (rc) {
1977                 CERROR("llog_init_handle failed %d\n", rc);
1978                 GOTO(out, rc);
1979         }
1980
1981         rc = llog_cat_process(NULL, llh, changelog_kkuc_cb, cs, 0, 0);
1982
1983         /* Send EOF no matter what our result */
1984         kuch = changelog_kuc_hdr(cs->cs_buf, sizeof(*kuch), cs->cs_flags);
1985         kuch->kuc_msgtype = CL_EOF;
1986         libcfs_kkuc_msg_put(cs->cs_fp, kuch);
1987
1988 out:
1989         fput(cs->cs_fp);
1990         if (llh)
1991                 llog_cat_close(NULL, llh);
1992         if (ctxt)
1993                 llog_ctxt_put(ctxt);
1994         if (cs->cs_buf)
1995                 OBD_FREE(cs->cs_buf, KUC_CHANGELOG_MSG_MAXSIZE);
1996         OBD_FREE_PTR(cs);
1997         return rc;
1998 }
1999
2000 static int mdc_ioc_changelog_send(struct obd_device *obd,
2001                                   struct ioc_changelog *icc)
2002 {
2003         struct changelog_show *cs;
2004         struct task_struct *task;
2005         int rc;
2006
2007         /* Freed in mdc_changelog_send_thread */
2008         OBD_ALLOC_PTR(cs);
2009         if (!cs)
2010                 return -ENOMEM;
2011
2012         cs->cs_obd = obd;
2013         cs->cs_startrec = icc->icc_recno;
2014         /* matching fput in mdc_changelog_send_thread */
2015         cs->cs_fp = fget(icc->icc_id);
2016         cs->cs_flags = icc->icc_flags;
2017
2018         /*
2019          * New thread because we should return to user app before
2020          * writing into our pipe
2021          */
2022         task = kthread_run(mdc_changelog_send_thread, cs,
2023                            "mdc_clg_send_thread");
2024         if (IS_ERR(task)) {
2025                 rc = PTR_ERR(task);
2026                 CERROR("%s: cannot start changelog thread: rc = %d\n",
2027                        cs_obd_name(cs), rc);
2028                 OBD_FREE_PTR(cs);
2029         } else {
2030                 rc = 0;
2031                 CDEBUG(D_HSM, "%s: started changelog thread\n",
2032                        cs_obd_name(cs));
2033         }
2034
2035         return rc;
2036 }
2037
2038 static int mdc_ioc_hsm_ct_start(struct obd_export *exp,
2039                                 struct lustre_kernelcomm *lk);
2040
2041 static int mdc_quotactl(struct obd_device *unused, struct obd_export *exp,
2042                         struct obd_quotactl *oqctl)
2043 {
2044         struct ptlrpc_request   *req;
2045         struct obd_quotactl     *oqc;
2046         int                      rc;
2047         ENTRY;
2048
2049         req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp),
2050                                         &RQF_MDS_QUOTACTL, LUSTRE_MDS_VERSION,
2051                                         MDS_QUOTACTL);
2052         if (req == NULL)
2053                 RETURN(-ENOMEM);
2054
2055         oqc = req_capsule_client_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
2056         *oqc = *oqctl;
2057
2058         ptlrpc_request_set_replen(req);
2059         ptlrpc_at_set_req_timeout(req);
2060         req->rq_no_resend = 1;
2061
2062         rc = ptlrpc_queue_wait(req);
2063         if (rc)
2064                 CERROR("ptlrpc_queue_wait failed, rc: %d\n", rc);
2065
2066         if (req->rq_repmsg &&
2067             (oqc = req_capsule_server_get(&req->rq_pill, &RMF_OBD_QUOTACTL))) {
2068                 *oqctl = *oqc;
2069         } else if (!rc) {
2070                 CERROR ("Can't unpack obd_quotactl\n");
2071                 rc = -EPROTO;
2072         }
2073         ptlrpc_req_finished(req);
2074
2075         RETURN(rc);
2076 }
2077
2078 static int mdc_ioc_swap_layouts(struct obd_export *exp,
2079                                 struct md_op_data *op_data)
2080 {
2081         struct list_head cancels = LIST_HEAD_INIT(cancels);
2082         struct ptlrpc_request   *req;
2083         int                      rc, count;
2084         struct mdc_swap_layouts *msl, *payload;
2085         ENTRY;
2086
2087         msl = op_data->op_data;
2088
2089         /* When the MDT will get the MDS_SWAP_LAYOUTS RPC the
2090          * first thing it will do is to cancel the 2 layout
2091          * locks hold by this client.
2092          * So the client must cancel its layout locks on the 2 fids
2093          * with the request RPC to avoid extra RPC round trips
2094          */
2095         count = mdc_resource_get_unused(exp, &op_data->op_fid1, &cancels,
2096                                         LCK_EX, MDS_INODELOCK_LAYOUT |
2097                                         MDS_INODELOCK_XATTR);
2098         count += mdc_resource_get_unused(exp, &op_data->op_fid2, &cancels,
2099                                          LCK_EX, MDS_INODELOCK_LAYOUT |
2100                                          MDS_INODELOCK_XATTR);
2101
2102         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
2103                                    &RQF_MDS_SWAP_LAYOUTS);
2104         if (req == NULL) {
2105                 ldlm_lock_list_put(&cancels, l_bl_ast, count);
2106                 RETURN(-ENOMEM);
2107         }
2108
2109         mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
2110         mdc_set_capa_size(req, &RMF_CAPA2, op_data->op_capa2);
2111
2112         rc = mdc_prep_elc_req(exp, req, MDS_SWAP_LAYOUTS, &cancels, count);
2113         if (rc) {
2114                 ptlrpc_request_free(req);
2115                 RETURN(rc);
2116         }
2117
2118         mdc_swap_layouts_pack(req, op_data);
2119
2120         payload = req_capsule_client_get(&req->rq_pill, &RMF_SWAP_LAYOUTS);
2121         LASSERT(payload);
2122
2123         *payload = *msl;
2124
2125         ptlrpc_request_set_replen(req);
2126
2127         rc = ptlrpc_queue_wait(req);
2128         if (rc)
2129                 GOTO(out, rc);
2130         EXIT;
2131
2132 out:
2133         ptlrpc_req_finished(req);
2134         return rc;
2135 }
2136
2137 static int mdc_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
2138                          void *karg, void __user *uarg)
2139 {
2140         struct obd_device *obd = exp->exp_obd;
2141         struct obd_ioctl_data *data = karg;
2142         struct obd_import *imp = obd->u.cli.cl_import;
2143         int rc;
2144         ENTRY;
2145
2146         if (!try_module_get(THIS_MODULE)) {
2147                 CERROR("%s: cannot get module '%s'\n", obd->obd_name,
2148                        module_name(THIS_MODULE));
2149                 return -EINVAL;
2150         }
2151         switch (cmd) {
2152         case OBD_IOC_CHANGELOG_SEND:
2153                 rc = mdc_ioc_changelog_send(obd, karg);
2154                 GOTO(out, rc);
2155         case OBD_IOC_CHANGELOG_CLEAR: {
2156                 struct ioc_changelog *icc = karg;
2157                 struct changelog_setinfo cs =
2158                         {.cs_recno = icc->icc_recno, .cs_id = icc->icc_id};
2159                 rc = obd_set_info_async(NULL, exp, strlen(KEY_CHANGELOG_CLEAR),
2160                                         KEY_CHANGELOG_CLEAR, sizeof(cs), &cs,
2161                                         NULL);
2162                 GOTO(out, rc);
2163         }
2164         case OBD_IOC_FID2PATH:
2165                 rc = mdc_ioc_fid2path(exp, karg);
2166                 GOTO(out, rc);
2167         case LL_IOC_HSM_CT_START:
2168                 rc = mdc_ioc_hsm_ct_start(exp, karg);
2169                 /* ignore if it was already registered on this MDS. */
2170                 if (rc == -EEXIST)
2171                         rc = 0;
2172                 GOTO(out, rc);
2173         case LL_IOC_HSM_PROGRESS:
2174                 rc = mdc_ioc_hsm_progress(exp, karg);
2175                 GOTO(out, rc);
2176         case LL_IOC_HSM_STATE_GET:
2177                 rc = mdc_ioc_hsm_state_get(exp, karg);
2178                 GOTO(out, rc);
2179         case LL_IOC_HSM_STATE_SET:
2180                 rc = mdc_ioc_hsm_state_set(exp, karg);
2181                 GOTO(out, rc);
2182         case LL_IOC_HSM_ACTION:
2183                 rc = mdc_ioc_hsm_current_action(exp, karg);
2184                 GOTO(out, rc);
2185         case LL_IOC_HSM_REQUEST:
2186                 rc = mdc_ioc_hsm_request(exp, karg);
2187                 GOTO(out, rc);
2188         case OBD_IOC_CLIENT_RECOVER:
2189                 rc = ptlrpc_recover_import(imp, data->ioc_inlbuf1, 0);
2190                 if (rc < 0)
2191                         GOTO(out, rc);
2192                 GOTO(out, rc = 0);
2193         case IOC_OSC_SET_ACTIVE:
2194                 rc = ptlrpc_set_import_active(imp, data->ioc_offset);
2195                 GOTO(out, rc);
2196         case OBD_IOC_PING_TARGET:
2197                 rc = ptlrpc_obd_ping(obd);
2198                 GOTO(out, rc);
2199         /*
2200          * Normally IOC_OBD_STATFS, OBD_IOC_QUOTACTL iocontrol are handled by
2201          * LMV instead of MDC. But when the cluster is upgraded from 1.8,
2202          * there'd be no LMV layer thus we might be called here. Eventually
2203          * this code should be removed.
2204          * bz20731, LU-592.
2205          */
2206         case IOC_OBD_STATFS: {
2207                 struct obd_statfs stat_buf = {0};
2208
2209                 if (*((__u32 *) data->ioc_inlbuf2) != 0)
2210                         GOTO(out, rc = -ENODEV);
2211
2212                 /* copy UUID */
2213                 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(obd),
2214                                  min((int)data->ioc_plen2,
2215                                      (int)sizeof(struct obd_uuid))))
2216                         GOTO(out, rc = -EFAULT);
2217
2218                 rc = mdc_statfs(NULL, obd->obd_self_export, &stat_buf,
2219                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
2220                                 0);
2221                 if (rc != 0)
2222                         GOTO(out, rc);
2223
2224                 if (copy_to_user(data->ioc_pbuf1, &stat_buf,
2225                                      min((int) data->ioc_plen1,
2226                                          (int) sizeof(stat_buf))))
2227                         GOTO(out, rc = -EFAULT);
2228
2229                 GOTO(out, rc = 0);
2230         }
2231         case OBD_IOC_QUOTACTL: {
2232                 struct if_quotactl *qctl = karg;
2233                 struct obd_quotactl *oqctl;
2234
2235                 OBD_ALLOC_PTR(oqctl);
2236                 if (oqctl == NULL)
2237                         GOTO(out, rc = -ENOMEM);
2238
2239                 QCTL_COPY(oqctl, qctl);
2240                 rc = obd_quotactl(exp, oqctl);
2241                 if (rc == 0) {
2242                         QCTL_COPY(qctl, oqctl);
2243                         qctl->qc_valid = QC_MDTIDX;
2244                         qctl->obd_uuid = obd->u.cli.cl_target_uuid;
2245                 }
2246
2247                 OBD_FREE_PTR(oqctl);
2248                 GOTO(out, rc);
2249         }
2250         case LL_IOC_GET_CONNECT_FLAGS:
2251                 if (copy_to_user(uarg, exp_connect_flags_ptr(exp),
2252                                  sizeof(*exp_connect_flags_ptr(exp))))
2253                         GOTO(out, rc = -EFAULT);
2254
2255                 GOTO(out, rc = 0);
2256         case LL_IOC_LOV_SWAP_LAYOUTS:
2257                 rc = mdc_ioc_swap_layouts(exp, karg);
2258                 GOTO(out, rc);
2259         default:
2260                 CERROR("unrecognised ioctl: cmd = %#x\n", cmd);
2261                 GOTO(out, rc = -ENOTTY);
2262         }
2263 out:
2264         module_put(THIS_MODULE);
2265
2266         return rc;
2267 }
2268
2269 static int mdc_get_info_rpc(struct obd_export *exp,
2270                             u32 keylen, void *key,
2271                             u32 vallen, void *val)
2272 {
2273         struct obd_import      *imp = class_exp2cliimp(exp);
2274         struct ptlrpc_request  *req;
2275         char                   *tmp;
2276         int                     rc = -EINVAL;
2277         ENTRY;
2278
2279         req = ptlrpc_request_alloc(imp, &RQF_MDS_GET_INFO);
2280         if (req == NULL)
2281                 RETURN(-ENOMEM);
2282
2283         req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_KEY,
2284                              RCL_CLIENT, keylen);
2285         req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_VALLEN,
2286                              RCL_CLIENT, sizeof(vallen));
2287
2288         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GET_INFO);
2289         if (rc) {
2290                 ptlrpc_request_free(req);
2291                 RETURN(rc);
2292         }
2293
2294         tmp = req_capsule_client_get(&req->rq_pill, &RMF_GETINFO_KEY);
2295         memcpy(tmp, key, keylen);
2296         tmp = req_capsule_client_get(&req->rq_pill, &RMF_GETINFO_VALLEN);
2297         memcpy(tmp, &vallen, sizeof(vallen));
2298
2299         req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_VAL,
2300                              RCL_SERVER, vallen);
2301         ptlrpc_request_set_replen(req);
2302
2303         rc = ptlrpc_queue_wait(req);
2304         /* -EREMOTE means the get_info result is partial, and it needs to
2305          * continue on another MDT, see fid2path part in lmv_iocontrol */
2306         if (rc == 0 || rc == -EREMOTE) {
2307                 tmp = req_capsule_server_get(&req->rq_pill, &RMF_GETINFO_VAL);
2308                 memcpy(val, tmp, vallen);
2309                 if (ptlrpc_rep_need_swab(req)) {
2310                         if (KEY_IS(KEY_FID2PATH))
2311                                 lustre_swab_fid2path(val);
2312                 }
2313         }
2314         ptlrpc_req_finished(req);
2315
2316         RETURN(rc);
2317 }
2318
2319 static void lustre_swab_hai(struct hsm_action_item *h)
2320 {
2321         __swab32s(&h->hai_len);
2322         __swab32s(&h->hai_action);
2323         lustre_swab_lu_fid(&h->hai_fid);
2324         lustre_swab_lu_fid(&h->hai_dfid);
2325         __swab64s(&h->hai_cookie);
2326         __swab64s(&h->hai_extent.offset);
2327         __swab64s(&h->hai_extent.length);
2328         __swab64s(&h->hai_gid);
2329 }
2330
2331 static void lustre_swab_hal(struct hsm_action_list *h)
2332 {
2333         struct hsm_action_item  *hai;
2334         __u32                    i;
2335
2336         __swab32s(&h->hal_version);
2337         __swab32s(&h->hal_count);
2338         __swab32s(&h->hal_archive_id);
2339         __swab64s(&h->hal_flags);
2340         hai = hai_first(h);
2341         for (i = 0; i < h->hal_count; i++, hai = hai_next(hai))
2342                 lustre_swab_hai(hai);
2343 }
2344
2345 static void lustre_swab_kuch(struct kuc_hdr *l)
2346 {
2347         __swab16s(&l->kuc_magic);
2348         /* __u8 l->kuc_transport */
2349         __swab16s(&l->kuc_msgtype);
2350         __swab16s(&l->kuc_msglen);
2351 }
2352
2353 static int mdc_ioc_hsm_ct_start(struct obd_export *exp,
2354                                 struct lustre_kernelcomm *lk)
2355 {
2356         struct obd_import  *imp = class_exp2cliimp(exp);
2357         __u32               archive = lk->lk_data;
2358         int                 rc = 0;
2359
2360         if (lk->lk_group != KUC_GRP_HSM) {
2361                 CERROR("Bad copytool group %d\n", lk->lk_group);
2362                 return -EINVAL;
2363         }
2364
2365         CDEBUG(D_HSM, "CT start r%d w%d u%d g%d f%#x\n", lk->lk_rfd, lk->lk_wfd,
2366                lk->lk_uid, lk->lk_group, lk->lk_flags);
2367
2368         if (lk->lk_flags & LK_FLG_STOP) {
2369                 /* Unregister with the coordinator */
2370                 rc = mdc_ioc_hsm_ct_unregister(imp);
2371         } else {
2372                 rc = mdc_ioc_hsm_ct_register(imp, archive);
2373         }
2374
2375         return rc;
2376 }
2377
2378 /**
2379  * Send a message to any listening copytools
2380  * @param val KUC message (kuc_hdr + hsm_action_list)
2381  * @param len total length of message
2382  */
2383 static int mdc_hsm_copytool_send(size_t len, void *val)
2384 {
2385         struct kuc_hdr          *lh = (struct kuc_hdr *)val;
2386         struct hsm_action_list  *hal = (struct hsm_action_list *)(lh + 1);
2387         int                      rc;
2388         ENTRY;
2389
2390         if (len < sizeof(*lh) + sizeof(*hal)) {
2391                 CERROR("Short HSM message %zu < %zu\n", len,
2392                        sizeof(*lh) + sizeof(*hal));
2393                 RETURN(-EPROTO);
2394         }
2395         if (lh->kuc_magic == __swab16(KUC_MAGIC)) {
2396                 lustre_swab_kuch(lh);
2397                 lustre_swab_hal(hal);
2398         } else if (lh->kuc_magic != KUC_MAGIC) {
2399                 CERROR("Bad magic %x!=%x\n", lh->kuc_magic, KUC_MAGIC);
2400                 RETURN(-EPROTO);
2401         }
2402
2403         CDEBUG(D_HSM, " Received message mg=%x t=%d m=%d l=%d actions=%d "
2404                "on %s\n",
2405                lh->kuc_magic, lh->kuc_transport, lh->kuc_msgtype,
2406                lh->kuc_msglen, hal->hal_count, hal->hal_fsname);
2407
2408         /* Broadcast to HSM listeners */
2409         rc = libcfs_kkuc_group_put(KUC_GRP_HSM, lh);
2410
2411         RETURN(rc);
2412 }
2413
2414 /**
2415  * callback function passed to kuc for re-registering each HSM copytool
2416  * running on MDC, after MDT shutdown/recovery.
2417  * @param data copytool registration data
2418  * @param cb_arg callback argument (obd_import)
2419  */
2420 static int mdc_hsm_ct_reregister(void *data, void *cb_arg)
2421 {
2422         struct kkuc_ct_data     *kcd = data;
2423         struct obd_import       *imp = (struct obd_import *)cb_arg;
2424         int                      rc;
2425
2426         if (kcd == NULL || kcd->kcd_magic != KKUC_CT_DATA_MAGIC)
2427                 return -EPROTO;
2428
2429         if (!obd_uuid_equals(&kcd->kcd_uuid, &imp->imp_obd->obd_uuid))
2430                 return 0;
2431
2432         CDEBUG(D_HA, "%s: recover copytool registration to MDT (archive=%#x)\n",
2433                imp->imp_obd->obd_name, kcd->kcd_archive);
2434         rc = mdc_ioc_hsm_ct_register(imp, kcd->kcd_archive);
2435
2436         /* ignore error if the copytool is already registered */
2437         return (rc == -EEXIST) ? 0 : rc;
2438 }
2439
2440 /**
2441  * Re-establish all kuc contexts with MDT
2442  * after MDT shutdown/recovery.
2443  */
2444 static int mdc_kuc_reregister(struct obd_import *imp)
2445 {
2446         /* re-register HSM agents */
2447         return libcfs_kkuc_group_foreach(KUC_GRP_HSM, mdc_hsm_ct_reregister,
2448                                          (void *)imp);
2449 }
2450
2451 static int mdc_set_info_async(const struct lu_env *env,
2452                               struct obd_export *exp,
2453                               u32 keylen, void *key,
2454                               u32 vallen, void *val,
2455                               struct ptlrpc_request_set *set)
2456 {
2457         struct obd_import       *imp = class_exp2cliimp(exp);
2458         int                      rc;
2459         ENTRY;
2460
2461         if (KEY_IS(KEY_READ_ONLY)) {
2462                 if (vallen != sizeof(int))
2463                         RETURN(-EINVAL);
2464
2465                 spin_lock(&imp->imp_lock);
2466                 if (*((int *)val)) {
2467                         imp->imp_connect_flags_orig |= OBD_CONNECT_RDONLY;
2468                         imp->imp_connect_data.ocd_connect_flags |=
2469                                                         OBD_CONNECT_RDONLY;
2470                 } else {
2471                         imp->imp_connect_flags_orig &= ~OBD_CONNECT_RDONLY;
2472                         imp->imp_connect_data.ocd_connect_flags &=
2473                                                         ~OBD_CONNECT_RDONLY;
2474                 }
2475                 spin_unlock(&imp->imp_lock);
2476
2477                 rc = do_set_info_async(imp, MDS_SET_INFO, LUSTRE_MDS_VERSION,
2478                                        keylen, key, vallen, val, set);
2479                 RETURN(rc);
2480         }
2481         if (KEY_IS(KEY_SPTLRPC_CONF)) {
2482                 sptlrpc_conf_client_adapt(exp->exp_obd);
2483                 RETURN(0);
2484         }
2485         if (KEY_IS(KEY_FLUSH_CTX)) {
2486                 sptlrpc_import_flush_my_ctx(imp);
2487                 RETURN(0);
2488         }
2489         if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
2490                 rc = do_set_info_async(imp, MDS_SET_INFO, LUSTRE_MDS_VERSION,
2491                                        keylen, key, vallen, val, set);
2492                 RETURN(rc);
2493         }
2494         if (KEY_IS(KEY_HSM_COPYTOOL_SEND)) {
2495                 rc = mdc_hsm_copytool_send(vallen, val);
2496                 RETURN(rc);
2497         }
2498
2499         if (KEY_IS(KEY_DEFAULT_EASIZE)) {
2500                 __u32 *default_easize = val;
2501
2502                 exp->exp_obd->u.cli.cl_default_mds_easize = *default_easize;
2503                 RETURN(0);
2504         }
2505
2506         CERROR("Unknown key %s\n", (char *)key);
2507         RETURN(-EINVAL);
2508 }
2509
2510 static int mdc_get_info(const struct lu_env *env, struct obd_export *exp,
2511                         __u32 keylen, void *key, __u32 *vallen, void *val)
2512 {
2513         int rc = -EINVAL;
2514
2515         if (KEY_IS(KEY_MAX_EASIZE)) {
2516                 __u32 mdsize, *max_easize;
2517
2518                 if (*vallen != sizeof(int))
2519                         RETURN(-EINVAL);
2520                 mdsize = *(__u32 *)val;
2521                 if (mdsize > exp->exp_obd->u.cli.cl_max_mds_easize)
2522                         exp->exp_obd->u.cli.cl_max_mds_easize = mdsize;
2523                 max_easize = val;
2524                 *max_easize = exp->exp_obd->u.cli.cl_max_mds_easize;
2525                 RETURN(0);
2526         } else if (KEY_IS(KEY_DEFAULT_EASIZE)) {
2527                 __u32 *default_easize;
2528
2529                 if (*vallen != sizeof(int))
2530                         RETURN(-EINVAL);
2531                 default_easize = val;
2532                 *default_easize = exp->exp_obd->u.cli.cl_default_mds_easize;
2533                 RETURN(0);
2534         } else if (KEY_IS(KEY_CONN_DATA)) {
2535                 struct obd_import *imp = class_exp2cliimp(exp);
2536                 struct obd_connect_data *data = val;
2537
2538                 if (*vallen != sizeof(*data))
2539                         RETURN(-EINVAL);
2540
2541                 *data = imp->imp_connect_data;
2542                 RETURN(0);
2543         } else if (KEY_IS(KEY_TGT_COUNT)) {
2544                 *((__u32 *)val) = 1;
2545                 RETURN(0);
2546         }
2547
2548         rc = mdc_get_info_rpc(exp, keylen, key, *vallen, val);
2549
2550         RETURN(rc);
2551 }
2552
2553 static int mdc_fsync(struct obd_export *exp, const struct lu_fid *fid,
2554                      struct obd_capa *oc, struct ptlrpc_request **request)
2555 {
2556         struct ptlrpc_request *req;
2557         int                    rc;
2558         ENTRY;
2559
2560         *request = NULL;
2561         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_SYNC);
2562         if (req == NULL)
2563                 RETURN(-ENOMEM);
2564
2565         mdc_set_capa_size(req, &RMF_CAPA1, oc);
2566
2567         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_SYNC);
2568         if (rc) {
2569                 ptlrpc_request_free(req);
2570                 RETURN(rc);
2571         }
2572
2573         mdc_pack_body(req, fid, oc, 0, 0, -1, 0);
2574
2575         ptlrpc_request_set_replen(req);
2576
2577         rc = ptlrpc_queue_wait(req);
2578         if (rc)
2579                 ptlrpc_req_finished(req);
2580         else
2581                 *request = req;
2582         RETURN(rc);
2583 }
2584
2585 static int mdc_import_event(struct obd_device *obd, struct obd_import *imp,
2586                             enum obd_import_event event)
2587 {
2588         int rc = 0;
2589
2590         LASSERT(imp->imp_obd == obd);
2591
2592         switch (event) {
2593         case IMP_EVENT_DISCON: {
2594 #if 0
2595                 /* XXX Pass event up to OBDs stack. used only for FLD now */
2596                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_DISCON, NULL);
2597 #endif
2598                 break;
2599         }
2600         case IMP_EVENT_INACTIVE: {
2601                 struct client_obd *cli = &obd->u.cli;
2602                 /*
2603                  * Flush current sequence to make client obtain new one
2604                  * from server in case of disconnect/reconnect.
2605                  */
2606                 if (cli->cl_seq != NULL)
2607                         seq_client_flush(cli->cl_seq);
2608
2609                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_INACTIVE, NULL);
2610                 break;
2611         }
2612         case IMP_EVENT_INVALIDATE: {
2613                 struct ldlm_namespace *ns = obd->obd_namespace;
2614
2615                 ldlm_namespace_cleanup(ns, LDLM_FL_LOCAL_ONLY);
2616
2617                 break;
2618         }
2619         case IMP_EVENT_ACTIVE:
2620                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_ACTIVE, NULL);
2621                 /* redo the kuc registration after reconnecting */
2622                 if (rc == 0)
2623                         rc = mdc_kuc_reregister(imp);
2624                 break;
2625         case IMP_EVENT_OCD:
2626                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_OCD, NULL);
2627                 break;
2628         case IMP_EVENT_DEACTIVATE:
2629         case IMP_EVENT_ACTIVATE:
2630                 break;
2631         default:
2632                 CERROR("Unknown import event %x\n", event);
2633                 LBUG();
2634         }
2635         RETURN(rc);
2636 }
2637
2638 int mdc_fid_alloc(const struct lu_env *env, struct obd_export *exp,
2639                   struct lu_fid *fid, struct md_op_data *op_data)
2640 {
2641         struct client_obd *cli = &exp->exp_obd->u.cli;
2642         struct lu_client_seq *seq = cli->cl_seq;
2643         ENTRY;
2644         RETURN(seq_client_alloc_fid(env, seq, fid));
2645 }
2646
2647 static struct obd_uuid *mdc_get_uuid(struct obd_export *exp)
2648 {
2649         struct client_obd *cli = &exp->exp_obd->u.cli;
2650         return &cli->cl_target_uuid;
2651 }
2652
2653 /**
2654  * Determine whether the lock can be canceled before replaying it during
2655  * recovery, non zero value will be return if the lock can be canceled,
2656  * or zero returned for not
2657  */
2658 static int mdc_cancel_weight(struct ldlm_lock *lock)
2659 {
2660         if (lock->l_resource->lr_type != LDLM_IBITS)
2661                 RETURN(0);
2662
2663         /* FIXME: if we ever get into a situation where there are too many
2664          * opened files with open locks on a single node, then we really
2665          * should replay these open locks to reget it */
2666         if (lock->l_policy_data.l_inodebits.bits & MDS_INODELOCK_OPEN)
2667                 RETURN(0);
2668
2669         RETURN(1);
2670 }
2671
2672 static int mdc_resource_inode_free(struct ldlm_resource *res)
2673 {
2674         if (res->lr_lvb_inode)
2675                 res->lr_lvb_inode = NULL;
2676
2677         return 0;
2678 }
2679
2680 static struct ldlm_valblock_ops inode_lvbo = {
2681         .lvbo_free = mdc_resource_inode_free
2682 };
2683
2684 static int mdc_llog_init(struct obd_device *obd)
2685 {
2686         struct obd_llog_group   *olg = &obd->obd_olg;
2687         struct llog_ctxt        *ctxt;
2688         int                      rc;
2689
2690         ENTRY;
2691
2692         rc = llog_setup(NULL, obd, olg, LLOG_CHANGELOG_REPL_CTXT, obd,
2693                         &llog_client_ops);
2694         if (rc < 0)
2695                 RETURN(rc);
2696
2697         ctxt = llog_group_get_ctxt(olg, LLOG_CHANGELOG_REPL_CTXT);
2698         llog_initiator_connect(ctxt);
2699         llog_ctxt_put(ctxt);
2700
2701         RETURN(0);
2702 }
2703
2704 static void mdc_llog_finish(struct obd_device *obd)
2705 {
2706         struct llog_ctxt *ctxt;
2707
2708         ENTRY;
2709
2710         ctxt = llog_get_context(obd, LLOG_CHANGELOG_REPL_CTXT);
2711         if (ctxt != NULL)
2712                 llog_cleanup(NULL, ctxt);
2713
2714         EXIT;
2715 }
2716
2717 static int mdc_setup(struct obd_device *obd, struct lustre_cfg *cfg)
2718 {
2719         struct client_obd               *cli = &obd->u.cli;
2720         int                             rc;
2721         ENTRY;
2722
2723         OBD_ALLOC(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
2724         if (!cli->cl_rpc_lock)
2725                 RETURN(-ENOMEM);
2726         mdc_init_rpc_lock(cli->cl_rpc_lock);
2727
2728         rc = ptlrpcd_addref();
2729         if (rc < 0)
2730                 GOTO(err_rpc_lock, rc);
2731
2732         OBD_ALLOC(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
2733         if (!cli->cl_close_lock)
2734                 GOTO(err_ptlrpcd_decref, rc = -ENOMEM);
2735         mdc_init_rpc_lock(cli->cl_close_lock);
2736
2737         rc = client_obd_setup(obd, cfg);
2738         if (rc)
2739                 GOTO(err_close_lock, rc);
2740 #ifdef CONFIG_PROC_FS
2741         obd->obd_vars = lprocfs_mdc_obd_vars;
2742         lprocfs_obd_setup(obd);
2743         lprocfs_alloc_md_stats(obd, 0);
2744 #endif
2745         sptlrpc_lprocfs_cliobd_attach(obd);
2746         ptlrpc_lprocfs_register_obd(obd);
2747
2748         ns_register_cancel(obd->obd_namespace, mdc_cancel_weight);
2749
2750         obd->obd_namespace->ns_lvbo = &inode_lvbo;
2751
2752         rc = mdc_llog_init(obd);
2753         if (rc) {
2754                 mdc_cleanup(obd);
2755                 CERROR("failed to setup llogging subsystems\n");
2756         }
2757
2758         RETURN(rc);
2759
2760 err_close_lock:
2761         OBD_FREE(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
2762 err_ptlrpcd_decref:
2763         ptlrpcd_decref();
2764 err_rpc_lock:
2765         OBD_FREE(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
2766         RETURN(rc);
2767 }
2768
2769 /* Initialize the default and maximum LOV EA sizes.  This allows
2770  * us to make MDS RPCs with large enough reply buffers to hold a default
2771  * sized EA without having to calculate this (via a call into the
2772  * LOV + OSCs) each time we make an RPC.  The maximum size is also tracked
2773  * but not used to avoid wastefully vmalloc()'ing large reply buffers when
2774  * a large number of stripes is possible.  If a larger reply buffer is
2775  * required it will be reallocated in the ptlrpc layer due to overflow.
2776  */
2777 static int mdc_init_ea_size(struct obd_export *exp, __u32 easize,
2778                             __u32 def_easize)
2779 {
2780         struct obd_device *obd = exp->exp_obd;
2781         struct client_obd *cli = &obd->u.cli;
2782         ENTRY;
2783
2784         if (cli->cl_max_mds_easize < easize)
2785                 cli->cl_max_mds_easize = easize;
2786
2787         if (cli->cl_default_mds_easize < def_easize)
2788                 cli->cl_default_mds_easize = def_easize;
2789
2790         RETURN(0);
2791 }
2792
2793 static int mdc_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
2794 {
2795         int rc = 0;
2796         ENTRY;
2797
2798         switch (stage) {
2799         case OBD_CLEANUP_EARLY:
2800                 break;
2801         case OBD_CLEANUP_EXPORTS:
2802                 /* Failsafe, ok if racy */
2803                 if (obd->obd_type->typ_refcnt <= 1)
2804                         libcfs_kkuc_group_rem(0, KUC_GRP_HSM, NULL);
2805
2806                 obd_cleanup_client_import(obd);
2807                 ptlrpc_lprocfs_unregister_obd(obd);
2808                 lprocfs_obd_cleanup(obd);
2809                 lprocfs_free_md_stats(obd);
2810                 mdc_llog_finish(obd);
2811                 break;
2812         }
2813         RETURN(rc);
2814 }
2815
2816 static int mdc_cleanup(struct obd_device *obd)
2817 {
2818         struct client_obd *cli = &obd->u.cli;
2819
2820         OBD_FREE(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
2821         OBD_FREE(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
2822
2823         ptlrpcd_decref();
2824
2825         return client_obd_cleanup(obd);
2826 }
2827
2828 static int mdc_process_config(struct obd_device *obd, size_t len, void *buf)
2829 {
2830         struct lustre_cfg *lcfg = buf;
2831         int rc = class_process_proc_param(PARAM_MDC, obd->obd_vars, lcfg, obd);
2832         return (rc > 0 ? 0: rc);
2833 }
2834
2835
2836 /* get remote permission for current user on fid */
2837 static int mdc_get_remote_perm(struct obd_export *exp, const struct lu_fid *fid,
2838                                struct obd_capa *oc, __u32 suppgid,
2839                                struct ptlrpc_request **request)
2840 {
2841         struct ptlrpc_request  *req;
2842         int                    rc;
2843         ENTRY;
2844
2845         LASSERT(client_is_remote(exp));
2846
2847         *request = NULL;
2848         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_GETATTR);
2849         if (req == NULL)
2850                 RETURN(-ENOMEM);
2851
2852         mdc_set_capa_size(req, &RMF_CAPA1, oc);
2853
2854         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR);
2855         if (rc) {
2856                 ptlrpc_request_free(req);
2857                 RETURN(rc);
2858         }
2859
2860         mdc_pack_body(req, fid, oc, OBD_MD_FLRMTPERM, 0, suppgid, 0);
2861
2862         req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER,
2863                              sizeof(struct mdt_remote_perm));
2864
2865         ptlrpc_request_set_replen(req);
2866
2867         rc = ptlrpc_queue_wait(req);
2868         if (rc)
2869                 ptlrpc_req_finished(req);
2870         else
2871                 *request = req;
2872         RETURN(rc);
2873 }
2874
2875 static int mdc_interpret_renew_capa(const struct lu_env *env,
2876                                     struct ptlrpc_request *req, void *args,
2877                                     int status)
2878 {
2879         struct mdc_renew_capa_args *ra = args;
2880         struct mdt_body *body = NULL;
2881         struct lustre_capa *capa;
2882         ENTRY;
2883
2884         if (status)
2885                 GOTO(out, capa = ERR_PTR(status));
2886
2887         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2888         if (body == NULL)
2889                 GOTO(out, capa = ERR_PTR(-EFAULT));
2890
2891         if ((body->mbo_valid & OBD_MD_FLOSSCAPA) == 0)
2892                 GOTO(out, capa = ERR_PTR(-ENOENT));
2893
2894         capa = req_capsule_server_get(&req->rq_pill, &RMF_CAPA2);
2895         if (!capa)
2896                 GOTO(out, capa = ERR_PTR(-EFAULT));
2897         EXIT;
2898 out:
2899         ra->ra_cb(ra->ra_oc, capa);
2900         return 0;
2901 }
2902
2903 static int mdc_renew_capa(struct obd_export *exp, struct obd_capa *oc,
2904                           renew_capa_cb_t cb)
2905 {
2906         struct ptlrpc_request *req;
2907         struct mdc_renew_capa_args *ra;
2908         ENTRY;
2909
2910         req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp), &RQF_MDS_GETATTR,
2911                                         LUSTRE_MDS_VERSION, MDS_GETATTR);
2912         if (req == NULL)
2913                 RETURN(-ENOMEM);
2914
2915         /* NB, OBD_MD_FLOSSCAPA is set here, but it doesn't necessarily mean the
2916          * capa to renew is oss capa.
2917          */
2918         mdc_pack_body(req, &oc->c_capa.lc_fid, oc, OBD_MD_FLOSSCAPA, 0, -1, 0);
2919         ptlrpc_request_set_replen(req);
2920
2921         CLASSERT(sizeof(*ra) <= sizeof(req->rq_async_args));
2922         ra = ptlrpc_req_async_args(req);
2923         ra->ra_oc = oc;
2924         ra->ra_cb = cb;
2925         req->rq_interpret_reply = mdc_interpret_renew_capa;
2926         ptlrpcd_add_req(req, PDL_POLICY_LOCAL, -1);
2927         RETURN(0);
2928 }
2929
2930 static struct obd_ops mdc_obd_ops = {
2931         .o_owner            = THIS_MODULE,
2932         .o_setup            = mdc_setup,
2933         .o_precleanup       = mdc_precleanup,
2934         .o_cleanup          = mdc_cleanup,
2935         .o_add_conn         = client_import_add_conn,
2936         .o_del_conn         = client_import_del_conn,
2937         .o_connect          = client_connect_import,
2938         .o_disconnect       = client_disconnect_export,
2939         .o_iocontrol        = mdc_iocontrol,
2940         .o_set_info_async   = mdc_set_info_async,
2941         .o_statfs           = mdc_statfs,
2942         .o_fid_init         = client_fid_init,
2943         .o_fid_fini         = client_fid_fini,
2944         .o_fid_alloc        = mdc_fid_alloc,
2945         .o_import_event     = mdc_import_event,
2946         .o_get_info         = mdc_get_info,
2947         .o_process_config   = mdc_process_config,
2948         .o_get_uuid         = mdc_get_uuid,
2949         .o_quotactl         = mdc_quotactl,
2950 };
2951
2952 static struct md_ops mdc_md_ops = {
2953         .m_getstatus        = mdc_getstatus,
2954         .m_null_inode       = mdc_null_inode,
2955         .m_find_cbdata      = mdc_find_cbdata,
2956         .m_close            = mdc_close,
2957         .m_create           = mdc_create,
2958         .m_enqueue          = mdc_enqueue,
2959         .m_getattr          = mdc_getattr,
2960         .m_getattr_name     = mdc_getattr_name,
2961         .m_intent_lock      = mdc_intent_lock,
2962         .m_link             = mdc_link,
2963         .m_rename           = mdc_rename,
2964         .m_setattr          = mdc_setattr,
2965         .m_setxattr         = mdc_setxattr,
2966         .m_getxattr         = mdc_getxattr,
2967         .m_fsync                = mdc_fsync,
2968         .m_read_page            = mdc_read_page,
2969         .m_unlink           = mdc_unlink,
2970         .m_cancel_unused    = mdc_cancel_unused,
2971         .m_init_ea_size     = mdc_init_ea_size,
2972         .m_set_lock_data    = mdc_set_lock_data,
2973         .m_lock_match       = mdc_lock_match,
2974         .m_get_lustre_md    = mdc_get_lustre_md,
2975         .m_free_lustre_md   = mdc_free_lustre_md,
2976         .m_set_open_replay_data = mdc_set_open_replay_data,
2977         .m_clear_open_replay_data = mdc_clear_open_replay_data,
2978         .m_renew_capa       = mdc_renew_capa,
2979         .m_unpack_capa      = mdc_unpack_capa,
2980         .m_get_remote_perm  = mdc_get_remote_perm,
2981         .m_intent_getattr_async = mdc_intent_getattr_async,
2982         .m_revalidate_lock      = mdc_revalidate_lock
2983 };
2984
2985 static int __init mdc_init(void)
2986 {
2987         return class_register_type(&mdc_obd_ops, &mdc_md_ops, true, NULL,
2988                                    LUSTRE_MDC_NAME, NULL);
2989 }
2990
2991 static void /*__exit*/ mdc_exit(void)
2992 {
2993         class_unregister_type(LUSTRE_MDC_NAME);
2994 }
2995
2996 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
2997 MODULE_DESCRIPTION("Lustre Metadata Client");
2998 MODULE_LICENSE("GPL");
2999
3000 module_init(mdc_init);
3001 module_exit(mdc_exit);