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.gnu.org/licenses/gpl-2.0.html
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2001, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Use is subject to license terms.
  25  *
  26  * Copyright (c) 2011, 2017, Intel Corporation.
  27  */
  28 /*
  29  * This file is part of Lustre, http://www.lustre.org/
  30  * Lustre is a trademark of Sun Microsystems, Inc.
  31  */
  32
  33 #define DEBUG_SUBSYSTEM S_MDC
  34
  35 #include <linux/init.h>
  36 #include <linux/kthread.h>
  37 #include <linux/module.h>
  38 #include <linux/pagemap.h>
  39 #include <linux/user_namespace.h>
  40 #include <linux/utsname.h>
  41 #include <linux/delay.h>
  42 #include <linux/uidgid.h>
  43 #include <linux/device.h>
  44 #include <linux/xarray.h>
  45
  46 #include <lustre_errno.h>
  47
  48 #include <cl_object.h>
  49 #include <llog_swab.h>
  50 #include <lprocfs_status.h>
  51 #include <lustre_acl.h>
  52 #include <lustre_compat.h>
  53 #include <lustre_fid.h>
  54 #include <uapi/linux/lustre/lustre_ioctl.h>
  55 #include <lustre_kernelcomm.h>
  56 #include <lustre_lmv.h>
  57 #include <lustre_log.h>
  58 #include <lustre_swab.h>
  59 #include <obd_class.h>
  60 #include <lustre_osc.h>
  61
  62 #include "mdc_internal.h"
  63
  64 #define REQUEST_MINOR 244
  65
  66 static int mdc_cleanup(struct obd_device *obd);
  67
  68 static inline int mdc_queue_wait(struct ptlrpc_request *req)
  69 {
  70         struct client_obd *cli = &req->rq_import->imp_obd->u.cli;
  71         int rc;
  72
  73         /* obd_get_request_slot() ensures that this client has no more
  74          * than cl_max_rpcs_in_flight RPCs simultaneously inf light
  75          * against an MDT. */
  76         rc = obd_get_request_slot(cli);
  77         if (rc != 0)
  78                 return rc;
  79
  80         rc = ptlrpc_queue_wait(req);
  81         obd_put_request_slot(cli);
  82
  83         return rc;
  84 }
  85
  86 /*
  87  * Send MDS_GET_ROOT RPC to fetch root FID.
  88  *
  89  * If \a fileset is not NULL it should contain a subdirectory off
  90  * the ROOT/ directory to be mounted on the client. Return the FID
  91  * of the subdirectory to the client to mount onto its mountpoint.
  92  *
  93  * \param[in]   imp     MDC import
  94  * \param[in]   fileset fileset name, which could be NULL
  95  * \param[out]  rootfid root FID of this mountpoint
  96  * \param[out]  pc      root capa will be unpacked and saved in this pointer
  97  *
  98  * \retval      0 on success, negative errno on failure
  99  */
 100 static int mdc_get_root(struct obd_export *exp, const char *fileset,
 101                          struct lu_fid *rootfid)
 102 {
 103         struct ptlrpc_request   *req;
 104         struct mdt_body         *body;
 105         int                      rc;
 106
 107         ENTRY;
 108
 109         if (fileset && !(exp_connect_flags(exp) & OBD_CONNECT_SUBTREE))
 110                 RETURN(-ENOTSUPP);
 111
 112         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
 113                                 &RQF_MDS_GET_ROOT);
 114         if (req == NULL)
 115                 RETURN(-ENOMEM);
 116
 117         if (fileset != NULL)
 118                 req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
 119                                      strlen(fileset) + 1);
 120         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GET_ROOT);
 121         if (rc) {
 122                 ptlrpc_request_free(req);
 123                 RETURN(rc);
 124         }
 125         mdc_pack_body(req, NULL, 0, 0, -1, 0);
 126         if (fileset != NULL) {
 127                 char *name = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
 128
 129                 memcpy(name, fileset, strlen(fileset));
 130         }
 131         lustre_msg_add_flags(req->rq_reqmsg, LUSTRE_IMP_FULL);
 132         req->rq_send_state = LUSTRE_IMP_FULL;
 133
 134         ptlrpc_request_set_replen(req);
 135
 136         rc = ptlrpc_queue_wait(req);
 137         if (rc)
 138                 GOTO(out, rc);
 139
 140         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 141         if (body == NULL)
 142                 GOTO(out, rc = -EPROTO);
 143
 144         *rootfid = body->mbo_fid1;
 145         CDEBUG(D_NET, "root fid="DFID", last_committed=%llu\n",
 146                PFID(rootfid), lustre_msg_get_last_committed(req->rq_repmsg));
 147         EXIT;
 148 out:
 149         ptlrpc_req_finished(req);
 150
 151         return rc;
 152 }
 153
 154 /*
 155  * This function now is known to always saying that it will receive 4 buffers
 156  * from server. Even for cases when acl_size and md_size is zero, RPC header
 157  * will contain 4 fields and RPC itself will contain zero size fields. This is
 158  * because mdt_getattr*() _always_ returns 4 fields, but if acl is not needed
 159  * and thus zero, it shrinks it, making zero size. The same story about
 160  * md_size. And this is course of problem when client waits for smaller number
 161  * of fields. This issue will be fixed later when client gets aware of RPC
 162  * layouts.  --umka
 163  */
 164 static int mdc_getattr_common(struct obd_export *exp,
 165                               struct ptlrpc_request *req)
 166 {
 167         struct req_capsule *pill = &req->rq_pill;
 168         struct mdt_body    *body;
 169         void               *eadata;
 170         int                 rc;
 171         ENTRY;
 172
 173         /* Request message already built. */
 174         rc = ptlrpc_queue_wait(req);
 175         if (rc != 0)
 176                 RETURN(rc);
 177
 178         /* sanity check for the reply */
 179         body = req_capsule_server_get(pill, &RMF_MDT_BODY);
 180         if (body == NULL)
 181                 RETURN(-EPROTO);
 182
 183         CDEBUG(D_NET, "mode: %o\n", body->mbo_mode);
 184
 185         mdc_update_max_ea_from_body(exp, body);
 186         if (body->mbo_eadatasize != 0) {
 187                 eadata = req_capsule_server_sized_get(pill, &RMF_MDT_MD,
 188                                                       body->mbo_eadatasize);
 189                 if (eadata == NULL)
 190                         RETURN(-EPROTO);
 191         }
 192
 193         RETURN(0);
 194 }
 195
 196 static void mdc_reset_acl_req(struct ptlrpc_request *req)
 197 {
 198         spin_lock(&req->rq_early_free_lock);
 199         sptlrpc_cli_free_repbuf(req);
 200         req->rq_repbuf = NULL;
 201         req->rq_repbuf_len = 0;
 202         req->rq_repdata = NULL;
 203         req->rq_reqdata_len = 0;
 204         spin_unlock(&req->rq_early_free_lock);
 205 }
 206
 207 static int mdc_getattr(struct obd_export *exp, struct md_op_data *op_data,
 208                        struct ptlrpc_request **request)
 209 {
 210         struct ptlrpc_request *req;
 211         struct obd_import *imp = class_exp2cliimp(exp);
 212         __u32 acl_bufsize = LUSTRE_POSIX_ACL_MAX_SIZE_OLD;
 213         int rc;
 214         ENTRY;
 215
 216         /* Single MDS without an LMV case */
 217         if (op_data->op_flags & MF_GET_MDT_IDX) {
 218                 op_data->op_mds = 0;
 219                 RETURN(0);
 220         }
 221
 222         *request = NULL;
 223         req = ptlrpc_request_alloc(imp, &RQF_MDS_GETATTR);
 224         if (req == NULL)
 225                 RETURN(-ENOMEM);
 226
 227         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR);
 228         if (rc) {
 229                 ptlrpc_request_free(req);
 230                 RETURN(rc);
 231         }
 232
 233 again:
 234         mdc_pack_body(req, &op_data->op_fid1, op_data->op_valid,
 235                       op_data->op_mode, -1, 0);
 236         req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER, acl_bufsize);
 237         req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
 238                              op_data->op_mode);
 239         ptlrpc_request_set_replen(req);
 240
 241         rc = mdc_getattr_common(exp, req);
 242         if (rc) {
 243                 if (rc == -ERANGE) {
 244                         acl_bufsize = min_t(__u32,
 245                                             imp->imp_connect_data.ocd_max_easize,
 246                                             XATTR_SIZE_MAX);
 247                         mdc_reset_acl_req(req);
 248                         goto again;
 249                 }
 250
 251                 ptlrpc_req_finished(req);
 252         } else {
 253                 *request = req;
 254         }
 255
 256         RETURN(rc);
 257 }
 258
 259 static int mdc_getattr_name(struct obd_export *exp, struct md_op_data *op_data,
 260                             struct ptlrpc_request **request)
 261 {
 262         struct ptlrpc_request *req;
 263         struct obd_import *imp = class_exp2cliimp(exp);
 264         __u32 acl_bufsize = LUSTRE_POSIX_ACL_MAX_SIZE_OLD;
 265         int rc;
 266         ENTRY;
 267
 268         *request = NULL;
 269         req = ptlrpc_request_alloc(imp, &RQF_MDS_GETATTR_NAME);
 270         if (req == NULL)
 271                 RETURN(-ENOMEM);
 272
 273         req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
 274                              op_data->op_namelen + 1);
 275
 276         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR_NAME);
 277         if (rc) {
 278                 ptlrpc_request_free(req);
 279                 RETURN(rc);
 280         }
 281
 282         if (op_data->op_name) {
 283                 char *name = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
 284                 LASSERT(strnlen(op_data->op_name, op_data->op_namelen) ==
 285                                 op_data->op_namelen);
 286                 memcpy(name, op_data->op_name, op_data->op_namelen);
 287         }
 288
 289 again:
 290         mdc_pack_body(req, &op_data->op_fid1, op_data->op_valid,
 291                       op_data->op_mode, op_data->op_suppgids[0], 0);
 292         req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
 293                              op_data->op_mode);
 294         req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER, acl_bufsize);
 295         ptlrpc_request_set_replen(req);
 296
 297         rc = mdc_getattr_common(exp, req);
 298         if (rc) {
 299                 if (rc == -ERANGE) {
 300                         acl_bufsize = min_t(__u32,
 301                                             imp->imp_connect_data.ocd_max_easize,
 302                                             XATTR_SIZE_MAX);
 303                         mdc_reset_acl_req(req);
 304                         goto again;
 305                 }
 306
 307                 ptlrpc_req_finished(req);
 308         } else {
 309                 *request = req;
 310         }
 311
 312         RETURN(rc);
 313 }
 314
 315 static int mdc_xattr_common(struct obd_export *exp,const struct req_format *fmt,
 316                             const struct lu_fid *fid, int opcode, u64 valid,
 317                             const char *xattr_name, const char *input,
 318                             int input_size, int output_size, int flags,
 319                             __u32 suppgid, struct ptlrpc_request **request)
 320 {
 321         struct ptlrpc_request *req;
 322         int   xattr_namelen = 0;
 323         char *tmp;
 324         int   rc;
 325         ENTRY;
 326
 327         *request = NULL;
 328         req = ptlrpc_request_alloc(class_exp2cliimp(exp), fmt);
 329         if (req == NULL)
 330                 RETURN(-ENOMEM);
 331
 332         if (xattr_name) {
 333                 xattr_namelen = strlen(xattr_name) + 1;
 334                 req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
 335                                      xattr_namelen);
 336         }
 337         if (input_size)
 338                 LASSERT(input);
 339         req_capsule_set_size(&req->rq_pill, &RMF_EADATA, RCL_CLIENT,
 340                              input_size);
 341
 342         /* get SELinux policy info if any */
 343         rc = sptlrpc_get_sepol(req);
 344         if (rc < 0) {
 345                 ptlrpc_request_free(req);
 346                 RETURN(rc);
 347         }
 348         req_capsule_set_size(&req->rq_pill, &RMF_SELINUX_POL, RCL_CLIENT,
 349                              strlen(req->rq_sepol) ?
 350                              strlen(req->rq_sepol) + 1 : 0);
 351
 352         /* Flush local XATTR locks to get rid of a possible cancel RPC */
 353         if (opcode == MDS_REINT && fid_is_sane(fid) &&
 354             exp->exp_connect_data.ocd_ibits_known & MDS_INODELOCK_XATTR) {
 355                 LIST_HEAD(cancels);
 356                 int count;
 357
 358                 /* Without that packing would fail */
 359                 if (input_size == 0)
 360                         req_capsule_set_size(&req->rq_pill, &RMF_EADATA,
 361                                              RCL_CLIENT, 0);
 362
 363                 count = mdc_resource_get_unused(exp, fid,
 364                                                 &cancels, LCK_EX,
 365                                                 MDS_INODELOCK_XATTR);
 366
 367                 rc = mdc_prep_elc_req(exp, req, MDS_REINT, &cancels, count);
 368                 if (rc) {
 369                         ptlrpc_request_free(req);
 370                         RETURN(rc);
 371                 }
 372         } else {
 373                 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, opcode);
 374                 if (rc) {
 375                         ptlrpc_request_free(req);
 376                         RETURN(rc);
 377                 }
 378         }
 379
 380         if (opcode == MDS_REINT) {
 381                 struct mdt_rec_setxattr *rec;
 382
 383                 BUILD_BUG_ON(sizeof(struct mdt_rec_setxattr) !=
 384                              sizeof(struct mdt_rec_reint));
 385                 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
 386                 rec->sx_opcode = REINT_SETXATTR;
 387                 rec->sx_fsuid  = from_kuid(&init_user_ns, current_fsuid());
 388                 rec->sx_fsgid  = from_kgid(&init_user_ns, current_fsgid());
 389                 rec->sx_cap    = cfs_curproc_cap_pack();
 390                 rec->sx_suppgid1 = suppgid;
 391                 rec->sx_suppgid2 = -1;
 392                 rec->sx_fid    = *fid;
 393                 rec->sx_valid  = valid | OBD_MD_FLCTIME;
 394                 rec->sx_time   = ktime_get_real_seconds();
 395                 rec->sx_size   = output_size;
 396                 rec->sx_flags  = flags;
 397         } else {
 398                 mdc_pack_body(req, fid, valid, output_size, suppgid, flags);
 399         }
 400
 401         if (xattr_name) {
 402                 tmp = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
 403                 memcpy(tmp, xattr_name, xattr_namelen);
 404         }
 405         if (input_size) {
 406                 tmp = req_capsule_client_get(&req->rq_pill, &RMF_EADATA);
 407                 memcpy(tmp, input, input_size);
 408         }
 409
 410         mdc_file_sepol_pack(req);
 411
 412         if (req_capsule_has_field(&req->rq_pill, &RMF_EADATA, RCL_SERVER))
 413                 req_capsule_set_size(&req->rq_pill, &RMF_EADATA,
 414                                      RCL_SERVER, output_size);
 415         ptlrpc_request_set_replen(req);
 416
 417         /* make rpc */
 418         if (opcode == MDS_REINT)
 419                 ptlrpc_get_mod_rpc_slot(req);
 420
 421         rc = ptlrpc_queue_wait(req);
 422
 423         if (opcode == MDS_REINT)
 424                 ptlrpc_put_mod_rpc_slot(req);
 425
 426         if (rc)
 427                 ptlrpc_req_finished(req);
 428         else
 429                 *request = req;
 430         RETURN(rc);
 431 }
 432
 433 static int mdc_setxattr(struct obd_export *exp, const struct lu_fid *fid,
 434                         u64 obd_md_valid, const char *name,
 435                         const void *value, size_t value_size,
 436                         unsigned int xattr_flags, u32 suppgid,
 437                         struct ptlrpc_request **req)
 438 {
 439         LASSERT(obd_md_valid == OBD_MD_FLXATTR ||
 440                 obd_md_valid == OBD_MD_FLXATTRRM);
 441
 442         return mdc_xattr_common(exp, &RQF_MDS_REINT_SETXATTR,
 443                                 fid, MDS_REINT, obd_md_valid, name,
 444                                 value, value_size, 0, xattr_flags, suppgid,
 445                                 req);
 446 }
 447
 448 static int mdc_getxattr(struct obd_export *exp, const struct lu_fid *fid,
 449                         u64 obd_md_valid, const char *name, size_t buf_size,
 450                         struct ptlrpc_request **req)
 451 {
 452         struct mdt_body *body;
 453         int rc;
 454
 455         LASSERT(obd_md_valid == OBD_MD_FLXATTR ||
 456                 obd_md_valid == OBD_MD_FLXATTRLS);
 457
 458         /* The below message is checked in sanity-selinux.sh test_20d */
 459         CDEBUG(D_INFO, "%s: get xattr '%s' for "DFID"\n",
 460                exp->exp_obd->obd_name, name, PFID(fid));
 461         rc = mdc_xattr_common(exp, &RQF_MDS_GETXATTR, fid, MDS_GETXATTR,
 462                               obd_md_valid, name, NULL, 0, buf_size, 0, -1,
 463                               req);
 464         if (rc < 0)
 465                 GOTO(out, rc);
 466
 467         body = req_capsule_server_get(&(*req)->rq_pill, &RMF_MDT_BODY);
 468         if (body == NULL)
 469                 GOTO(out, rc = -EPROTO);
 470
 471         /* only detect the xattr size */
 472         if (buf_size == 0) {
 473                 /* LU-11109: Older MDTs do not distinguish
 474                  * between nonexistent xattrs and zero length
 475                  * values in this case. Newer MDTs will return
 476                  * -ENODATA or set OBD_MD_FLXATTR. */
 477                 GOTO(out, rc = body->mbo_eadatasize);
 478         }
 479
 480         if (body->mbo_eadatasize == 0) {
 481                 /* LU-11109: Newer MDTs set OBD_MD_FLXATTR on
 482                  * success so that we can distinguish between
 483                  * zero length value and nonexistent xattr.
 484                  *
 485                  * If OBD_MD_FLXATTR is not set then we keep
 486                  * the old behavior and return -ENODATA for
 487                  * getxattr() when mbo_eadatasize is 0. But
 488                  * -ENODATA only makes sense for getxattr()
 489                  * and not for listxattr(). */
 490                 if (body->mbo_valid & OBD_MD_FLXATTR)
 491                         GOTO(out, rc = 0);
 492                 else if (obd_md_valid == OBD_MD_FLXATTR)
 493                         GOTO(out, rc = -ENODATA);
 494                 else
 495                         GOTO(out, rc = 0);
 496         }
 497
 498         GOTO(out, rc = body->mbo_eadatasize);
 499 out:
 500         if (rc < 0) {
 501                 ptlrpc_req_finished(*req);
 502                 *req = NULL;
 503         }
 504
 505         return rc;
 506 }
 507
 508 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
 509 static int mdc_unpack_acl(struct ptlrpc_request *req, struct lustre_md *md)
 510 {
 511         struct req_capsule     *pill = &req->rq_pill;
 512         struct mdt_body        *body = md->body;
 513         struct posix_acl       *acl;
 514         void                   *buf;
 515         int                     rc;
 516         ENTRY;
 517
 518         if (!body->mbo_aclsize)
 519                 RETURN(0);
 520
 521         buf = req_capsule_server_sized_get(pill, &RMF_ACL, body->mbo_aclsize);
 522
 523         if (!buf)
 524                 RETURN(-EPROTO);
 525
 526         acl = posix_acl_from_xattr(&init_user_ns, buf, body->mbo_aclsize);
 527         if (acl == NULL)
 528                 RETURN(0);
 529         if (IS_ERR(acl)) {
 530                 rc = PTR_ERR(acl);
 531                 CERROR("convert xattr to acl: %d\n", rc);
 532                 RETURN(rc);
 533         }
 534
 535         rc = posix_acl_valid(&init_user_ns, acl);
 536         if (rc) {
 537                 CERROR("validate acl: %d\n", rc);
 538                 posix_acl_release(acl);
 539                 RETURN(rc);
 540         }
 541
 542         md->posix_acl = acl;
 543         RETURN(0);
 544 }
 545 #else
 546 #define mdc_unpack_acl(req, md) 0
 547 #endif
 548
 549 int mdc_get_lustre_md(struct obd_export *exp, struct ptlrpc_request *req,
 550                       struct obd_export *dt_exp, struct obd_export *md_exp,
 551                       struct lustre_md *md)
 552 {
 553         struct req_capsule *pill = &req->rq_pill;
 554         int rc;
 555         ENTRY;
 556
 557         LASSERT(md);
 558         memset(md, 0, sizeof(*md));
 559
 560         md->body = req_capsule_server_get(pill, &RMF_MDT_BODY);
 561         LASSERT(md->body != NULL);
 562
 563         if (md->body->mbo_valid & OBD_MD_FLEASIZE) {
 564                 if (!S_ISREG(md->body->mbo_mode)) {
 565                         CDEBUG(D_INFO, "OBD_MD_FLEASIZE set, should be a "
 566                                "regular file, but is not\n");
 567                         GOTO(out, rc = -EPROTO);
 568                 }
 569
 570                 if (md->body->mbo_eadatasize == 0) {
 571                         CDEBUG(D_INFO, "OBD_MD_FLEASIZE set, "
 572                                "but eadatasize 0\n");
 573                         GOTO(out, rc = -EPROTO);
 574                 }
 575
 576                 md->layout.lb_len = md->body->mbo_eadatasize;
 577                 md->layout.lb_buf = req_capsule_server_sized_get(pill,
 578                                                         &RMF_MDT_MD,
 579                                                         md->layout.lb_len);
 580                 if (md->layout.lb_buf == NULL)
 581                         GOTO(out, rc = -EPROTO);
 582         } else if (md->body->mbo_valid & OBD_MD_FLDIREA) {
 583                 const union lmv_mds_md *lmv;
 584                 size_t lmv_size;
 585
 586                 if (!S_ISDIR(md->body->mbo_mode)) {
 587                         CDEBUG(D_INFO, "OBD_MD_FLDIREA set, should be a "
 588                                "directory, but is not\n");
 589                         GOTO(out, rc = -EPROTO);
 590                 }
 591
 592                 if (md->body->mbo_valid & OBD_MD_MEA) {
 593                         lmv_size = md->body->mbo_eadatasize;
 594                         if (lmv_size == 0) {
 595                                 CDEBUG(D_INFO, "OBD_MD_FLDIREA is set, "
 596                                        "but eadatasize 0\n");
 597                                 RETURN(-EPROTO);
 598                         }
 599
 600                         lmv = req_capsule_server_sized_get(pill, &RMF_MDT_MD,
 601                                                            lmv_size);
 602                         if (lmv == NULL)
 603                                 GOTO(out, rc = -EPROTO);
 604
 605                         rc = md_unpackmd(md_exp, &md->lmv, lmv, lmv_size);
 606                         if (rc < 0)
 607                                 GOTO(out, rc);
 608
 609                         if (rc < (int)sizeof(*md->lmv)) {
 610                                 struct lmv_foreign_md *lfm = md->lfm;
 611
 612                                 /* short (< sizeof(struct lmv_stripe_md))
 613                                  * foreign LMV case
 614                                  */
 615                                 if (lfm->lfm_magic != LMV_MAGIC_FOREIGN) {
 616                                         CDEBUG(D_INFO,
 617                                                "lmv size too small: %d < %d\n",
 618                                                rc, (int)sizeof(*md->lmv));
 619                                         GOTO(out, rc = -EPROTO);
 620                                 }
 621                         }
 622                 }
 623
 624                 /* since 2.12.58 intent_getattr fetches default LMV */
 625                 if (md->body->mbo_valid & OBD_MD_DEFAULT_MEA) {
 626                         lmv_size = sizeof(struct lmv_user_md);
 627                         lmv = req_capsule_server_sized_get(pill,
 628                                                            &RMF_DEFAULT_MDT_MD,
 629                                                            lmv_size);
 630                         if (!lmv)
 631                                 GOTO(out, rc = -EPROTO);
 632
 633                         rc = md_unpackmd(md_exp, &md->default_lmv, lmv,
 634                                          lmv_size);
 635                         if (rc < 0)
 636                                 GOTO(out, rc);
 637
 638                         if (rc < (int)sizeof(*md->default_lmv)) {
 639                                 CDEBUG(D_INFO,
 640                                        "default lmv size too small: %d < %d\n",
 641                                         rc, (int)sizeof(*md->default_lmv));
 642                                 GOTO(out, rc = -EPROTO);
 643                         }
 644                 }
 645         }
 646         rc = 0;
 647
 648         if (md->body->mbo_valid & OBD_MD_FLACL) {
 649                 /* for ACL, it's possible that FLACL is set but aclsize is zero.
 650                  * only when aclsize != 0 there's an actual segment for ACL
 651                  * in reply buffer.
 652                  */
 653                 if (md->body->mbo_aclsize) {
 654                         rc = mdc_unpack_acl(req, md);
 655                         if (rc)
 656                                 GOTO(out, rc);
 657 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
 658                 } else {
 659                         md->posix_acl = NULL;
 660 #endif
 661                 }
 662         }
 663
 664         EXIT;
 665 out:
 666         if (rc) {
 667 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
 668                 posix_acl_release(md->posix_acl);
 669 #endif
 670         }
 671         return rc;
 672 }
 673
 674 int mdc_free_lustre_md(struct obd_export *exp, struct lustre_md *md)
 675 {
 676         ENTRY;
 677         RETURN(0);
 678 }
 679
 680 void mdc_replay_open(struct ptlrpc_request *req)
 681 {
 682         struct md_open_data *mod = req->rq_cb_data;
 683         struct ptlrpc_request *close_req;
 684         struct obd_client_handle *och;
 685         struct lustre_handle old_open_handle = { };
 686         struct mdt_body *body;
 687         ENTRY;
 688
 689         if (mod == NULL) {
 690                 DEBUG_REQ(D_ERROR, req,
 691                           "cannot properly replay without open data");
 692                 EXIT;
 693                 return;
 694         }
 695
 696         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 697         LASSERT(body != NULL);
 698
 699         spin_lock(&req->rq_lock);
 700         och = mod->mod_och;
 701         if (och && och->och_open_handle.cookie)
 702                 req->rq_early_free_repbuf = 1;
 703         else
 704                 req->rq_early_free_repbuf = 0;
 705         spin_unlock(&req->rq_lock);
 706
 707         if (req->rq_early_free_repbuf) {
 708                 struct lustre_handle *file_open_handle;
 709
 710                 LASSERT(och->och_magic == OBD_CLIENT_HANDLE_MAGIC);
 711
 712                 file_open_handle = &och->och_open_handle;
 713                 CDEBUG(D_HA, "updating handle from %#llx to %#llx\n",
 714                        file_open_handle->cookie, body->mbo_open_handle.cookie);
 715                 old_open_handle = *file_open_handle;
 716                 *file_open_handle = body->mbo_open_handle;
 717         }
 718
 719         close_req = mod->mod_close_req;
 720         if (close_req) {
 721                 __u32 opc = lustre_msg_get_opc(close_req->rq_reqmsg);
 722                 struct mdt_ioepoch *epoch;
 723
 724                 LASSERT(opc == MDS_CLOSE);
 725                 epoch = req_capsule_client_get(&close_req->rq_pill,
 726                                                &RMF_MDT_EPOCH);
 727                 LASSERT(epoch);
 728
 729                 if (req->rq_early_free_repbuf)
 730                         LASSERT(old_open_handle.cookie ==
 731                                 epoch->mio_open_handle.cookie);
 732
 733                 DEBUG_REQ(D_HA, close_req, "updating close body with new fh");
 734                 epoch->mio_open_handle = body->mbo_open_handle;
 735         }
 736         EXIT;
 737 }
 738
 739 void mdc_commit_open(struct ptlrpc_request *req)
 740 {
 741         struct md_open_data *mod = req->rq_cb_data;
 742         if (mod == NULL)
 743                 return;
 744
 745         /**
 746          * No need to touch md_open_data::mod_och, it holds a reference on
 747          * \var mod and will zero references to each other, \var mod will be
 748          * freed after that when md_open_data::mod_och will put the reference.
 749          */
 750
 751         /**
 752          * Do not let open request to disappear as it still may be needed
 753          * for close rpc to happen (it may happen on evict only, otherwise
 754          * ptlrpc_request::rq_replay does not let mdc_commit_open() to be
 755          * called), just mark this rpc as committed to distinguish these 2
 756          * cases, see mdc_close() for details. The open request reference will
 757          * be put along with freeing \var mod.
 758          */
 759         ptlrpc_request_addref(req);
 760         spin_lock(&req->rq_lock);
 761         req->rq_committed = 1;
 762         spin_unlock(&req->rq_lock);
 763         req->rq_cb_data = NULL;
 764         obd_mod_put(mod);
 765 }
 766
 767 int mdc_set_open_replay_data(struct obd_export *exp,
 768                              struct obd_client_handle *och,
 769                              struct lookup_intent *it)
 770 {
 771         struct md_open_data     *mod;
 772         struct mdt_rec_create   *rec;
 773         struct mdt_body         *body;
 774         struct ptlrpc_request   *open_req = it->it_request;
 775         struct obd_import       *imp = open_req->rq_import;
 776         ENTRY;
 777
 778         if (!open_req->rq_replay)
 779                 RETURN(0);
 780
 781         rec = req_capsule_client_get(&open_req->rq_pill, &RMF_REC_REINT);
 782         body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
 783         LASSERT(rec != NULL);
 784         /* Incoming message in my byte order (it's been swabbed). */
 785         /* Outgoing messages always in my byte order. */
 786         LASSERT(body != NULL);
 787
 788         /* Only if the import is replayable, we set replay_open data */
 789         if (och && imp->imp_replayable) {
 790                 mod = obd_mod_alloc();
 791                 if (mod == NULL) {
 792                         DEBUG_REQ(D_ERROR, open_req,
 793                                   "cannot allocate md_open_data");
 794                         RETURN(0);
 795                 }
 796
 797                 /**
 798                  * Take a reference on \var mod, to be freed on mdc_close().
 799                  * It protects \var mod from being freed on eviction (commit
 800                  * callback is called despite rq_replay flag).
 801                  * Another reference for \var och.
 802                  */
 803                 obd_mod_get(mod);
 804                 obd_mod_get(mod);
 805
 806                 spin_lock(&open_req->rq_lock);
 807                 och->och_mod = mod;
 808                 mod->mod_och = och;
 809                 mod->mod_is_create = it_disposition(it, DISP_OPEN_CREATE) ||
 810                                      it_disposition(it, DISP_OPEN_STRIPE);
 811                 mod->mod_open_req = open_req;
 812                 open_req->rq_cb_data = mod;
 813                 open_req->rq_commit_cb = mdc_commit_open;
 814                 open_req->rq_early_free_repbuf = 1;
 815                 spin_unlock(&open_req->rq_lock);
 816         }
 817
 818         rec->cr_fid2 = body->mbo_fid1;
 819         rec->cr_open_handle_old = body->mbo_open_handle;
 820         open_req->rq_replay_cb = mdc_replay_open;
 821         if (!fid_is_sane(&body->mbo_fid1)) {
 822                 DEBUG_REQ(D_ERROR, open_req,
 823                           "saving replay request with insane FID " DFID,
 824                           PFID(&body->mbo_fid1));
 825                 LBUG();
 826         }
 827
 828         DEBUG_REQ(D_RPCTRACE, open_req, "Set up open replay data");
 829         RETURN(0);
 830 }
 831
 832 static void mdc_free_open(struct md_open_data *mod)
 833 {
 834         int committed = 0;
 835
 836         if (mod->mod_is_create == 0 &&
 837             imp_connect_disp_stripe(mod->mod_open_req->rq_import))
 838                 committed = 1;
 839
 840         /**
 841          * No reason to asssert here if the open request has
 842          * rq_replay == 1. It means that mdc_close failed, and
 843          * close request wasn`t sent. It is not fatal to client.
 844          * The worst thing is eviction if the client gets open lock
 845          **/
 846
 847         DEBUG_REQ(D_RPCTRACE, mod->mod_open_req,
 848                   "free open request, rq_replay=%d",
 849                   mod->mod_open_req->rq_replay);
 850
 851         ptlrpc_request_committed(mod->mod_open_req, committed);
 852         if (mod->mod_close_req)
 853                 ptlrpc_request_committed(mod->mod_close_req, committed);
 854 }
 855
 856 int mdc_clear_open_replay_data(struct obd_export *exp,
 857                                struct obd_client_handle *och)
 858 {
 859         struct md_open_data *mod = och->och_mod;
 860         ENTRY;
 861
 862         /**
 863          * It is possible to not have \var mod in a case of eviction between
 864          * lookup and ll_file_open().
 865          **/
 866         if (mod == NULL)
 867                 RETURN(0);
 868
 869         LASSERT(mod != LP_POISON);
 870         LASSERT(mod->mod_open_req != NULL);
 871
 872         spin_lock(&mod->mod_open_req->rq_lock);
 873         if (mod->mod_och)
 874                 mod->mod_och->och_open_handle.cookie = 0;
 875         mod->mod_open_req->rq_early_free_repbuf = 0;
 876         spin_unlock(&mod->mod_open_req->rq_lock);
 877         mdc_free_open(mod);
 878
 879         mod->mod_och = NULL;
 880         och->och_mod = NULL;
 881         obd_mod_put(mod);
 882
 883         RETURN(0);
 884 }
 885
 886 static int mdc_close(struct obd_export *exp, struct md_op_data *op_data,
 887                      struct md_open_data *mod, struct ptlrpc_request **request)
 888 {
 889         struct obd_device     *obd = class_exp2obd(exp);
 890         struct ptlrpc_request *req;
 891         struct req_format     *req_fmt;
 892         size_t                 u32_count = 0;
 893         int                    rc;
 894         int                    saved_rc = 0;
 895         ENTRY;
 896
 897         CDEBUG(D_INODE, "%s: "DFID" file closed with intent: %x\n",
 898                exp->exp_obd->obd_name, PFID(&op_data->op_fid1),
 899                op_data->op_bias);
 900
 901         if (op_data->op_bias & MDS_CLOSE_INTENT) {
 902                 req_fmt = &RQF_MDS_CLOSE_INTENT;
 903                 if (op_data->op_bias & MDS_HSM_RELEASE) {
 904                         /* allocate a FID for volatile file */
 905                         rc = mdc_fid_alloc(NULL, exp, &op_data->op_fid2,
 906                                            op_data);
 907                         if (rc < 0) {
 908                                 CERROR("%s: "DFID" allocating FID: rc = %d\n",
 909                                        obd->obd_name, PFID(&op_data->op_fid1),
 910                                        rc);
 911                                 /* save the errcode and proceed to close */
 912                                 saved_rc = rc;
 913                         }
 914                 }
 915                 if (op_data->op_bias & MDS_CLOSE_RESYNC_DONE) {
 916                         size_t count = op_data->op_data_size / sizeof(__u32);
 917
 918                         if (count > INLINE_RESYNC_ARRAY_SIZE)
 919                                 u32_count = count;
 920                 }
 921         } else {
 922                 req_fmt = &RQF_MDS_CLOSE;
 923         }
 924
 925         *request = NULL;
 926         if (OBD_FAIL_CHECK(OBD_FAIL_MDC_CLOSE))
 927                 req = NULL;
 928         else
 929                 req = ptlrpc_request_alloc(class_exp2cliimp(exp), req_fmt);
 930
 931         /* Ensure that this close's handle is fixed up during replay. */
 932         if (likely(mod != NULL)) {
 933                 LASSERTF(mod->mod_open_req != NULL &&
 934                          mod->mod_open_req->rq_type != LI_POISON,
 935                          "POISONED open %p!\n", mod->mod_open_req);
 936
 937                 mod->mod_close_req = req;
 938
 939                 DEBUG_REQ(D_RPCTRACE, mod->mod_open_req, "matched open");
 940                 /* We no longer want to preserve this open for replay even
 941                  * though the open was committed. b=3632, b=3633 */
 942                 spin_lock(&mod->mod_open_req->rq_lock);
 943                 mod->mod_open_req->rq_replay = 0;
 944                 spin_unlock(&mod->mod_open_req->rq_lock);
 945         } else {
 946                 CDEBUG(D_HA, "couldn't find open req; expecting close error\n");
 947         }
 948         if (req == NULL) {
 949                 /**
 950                  * TODO: repeat close after errors
 951                  */
 952                 CWARN("%s: close of FID "DFID" failed, file reference will be "
 953                       "dropped when this client unmounts or is evicted\n",
 954                       obd->obd_name, PFID(&op_data->op_fid1));
 955                 GOTO(out, rc = -ENOMEM);
 956         }
 957
 958         if (u32_count > 0)
 959                 req_capsule_set_size(&req->rq_pill, &RMF_U32, RCL_CLIENT,
 960                                      u32_count * sizeof(__u32));
 961
 962         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_CLOSE);
 963         if (rc) {
 964                 ptlrpc_request_free(req);
 965                 req = NULL;
 966                 GOTO(out, rc);
 967         }
 968
 969         /* To avoid a livelock (bug 7034), we need to send CLOSE RPCs to a
 970          * portal whose threads are not taking any DLM locks and are therefore
 971          * always progressing */
 972         req->rq_request_portal = MDS_READPAGE_PORTAL;
 973         ptlrpc_at_set_req_timeout(req);
 974
 975         if (!(exp_connect_flags2(exp) & OBD_CONNECT2_LSOM))
 976                 op_data->op_xvalid &= ~(OP_XVALID_LAZYSIZE |
 977                                         OP_XVALID_LAZYBLOCKS);
 978
 979         mdc_close_pack(req, op_data);
 980
 981         req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
 982                              obd->u.cli.cl_default_mds_easize);
 983
 984         ptlrpc_request_set_replen(req);
 985
 986         ptlrpc_get_mod_rpc_slot(req);
 987         rc = ptlrpc_queue_wait(req);
 988         ptlrpc_put_mod_rpc_slot(req);
 989
 990         if (req->rq_repmsg == NULL) {
 991                 CDEBUG(D_RPCTRACE, "request %p failed to send: rc = %d\n", req,
 992                        req->rq_status);
 993                 if (rc == 0)
 994                         rc = req->rq_status ?: -EIO;
 995         } else if (rc == 0 || rc == -EAGAIN) {
 996                 struct mdt_body *body;
 997
 998                 rc = lustre_msg_get_status(req->rq_repmsg);
 999                 if (lustre_msg_get_type(req->rq_repmsg) == PTL_RPC_MSG_ERR) {
1000                         DEBUG_REQ(D_ERROR, req,
1001                                   "type = PTL_RPC_MSG_ERR: rc = %d", rc);
1002                         if (rc > 0)
1003                                 rc = -rc;
1004                 }
1005                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1006                 if (body == NULL)
1007                         rc = -EPROTO;
1008         } else if (rc == -ESTALE) {
1009                 /**
1010                  * it can be allowed error after 3633 if open was committed and
1011                  * server failed before close was sent. Let's check if mod
1012                  * exists and return no error in that case
1013                  */
1014                 if (mod) {
1015                         DEBUG_REQ(D_HA, req, "Reset ESTALE = %d", rc);
1016                         LASSERT(mod->mod_open_req != NULL);
1017                         if (mod->mod_open_req->rq_committed)
1018                                 rc = 0;
1019                 }
1020         }
1021
1022 out:
1023         if (mod) {
1024                 if (rc != 0)
1025                         mod->mod_close_req = NULL;
1026                 /* Since now, mod is accessed through open_req only,
1027                  * thus close req does not keep a reference on mod anymore. */
1028                 obd_mod_put(mod);
1029         }
1030         *request = req;
1031
1032         RETURN(rc < 0 ? rc : saved_rc);
1033 }
1034
1035 static int mdc_getpage(struct obd_export *exp, const struct lu_fid *fid,
1036                        u64 offset, struct page **pages, int npages,
1037                        struct ptlrpc_request **request)
1038 {
1039         struct ptlrpc_request   *req;
1040         struct ptlrpc_bulk_desc *desc;
1041         int                      i;
1042         int                      resends = 0;
1043         int                      rc;
1044         ENTRY;
1045
1046         *request = NULL;
1047
1048 restart_bulk:
1049         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_READPAGE);
1050         if (req == NULL)
1051                 RETURN(-ENOMEM);
1052
1053         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_READPAGE);
1054         if (rc) {
1055                 ptlrpc_request_free(req);
1056                 RETURN(rc);
1057         }
1058
1059         req->rq_request_portal = MDS_READPAGE_PORTAL;
1060         ptlrpc_at_set_req_timeout(req);
1061
1062         desc = ptlrpc_prep_bulk_imp(req, npages, 1,
1063                                     PTLRPC_BULK_PUT_SINK,
1064                                     MDS_BULK_PORTAL,
1065                                     &ptlrpc_bulk_kiov_pin_ops);
1066         if (desc == NULL) {
1067                 ptlrpc_req_finished(req);
1068                 RETURN(-ENOMEM);
1069         }
1070
1071         /* NB req now owns desc and will free it when it gets freed */
1072         for (i = 0; i < npages; i++)
1073                 desc->bd_frag_ops->add_kiov_frag(desc, pages[i], 0,
1074                                                  PAGE_SIZE);
1075
1076         mdc_readdir_pack(req, offset, PAGE_SIZE * npages, fid);
1077
1078         ptlrpc_request_set_replen(req);
1079         rc = ptlrpc_queue_wait(req);
1080         if (rc) {
1081                 ptlrpc_req_finished(req);
1082                 if (rc != -ETIMEDOUT)
1083                         RETURN(rc);
1084
1085                 resends++;
1086                 if (!client_should_resend(resends, &exp->exp_obd->u.cli)) {
1087                         CERROR("%s: too many resend retries: rc = %d\n",
1088                                exp->exp_obd->obd_name, -EIO);
1089                         RETURN(-EIO);
1090                 }
1091
1092                 /* If a signal interrupts then the timeout returned will
1093                  * not be zero. In that case return -EINTR
1094                  */
1095                 if (msleep_interruptible(resends * 1000))
1096                         RETURN(-EINTR);
1097
1098                 goto restart_bulk;
1099         }
1100
1101         rc = sptlrpc_cli_unwrap_bulk_read(req, req->rq_bulk,
1102                                           req->rq_bulk->bd_nob_transferred);
1103         if (rc < 0) {
1104                 ptlrpc_req_finished(req);
1105                 RETURN(rc);
1106         }
1107
1108         if (req->rq_bulk->bd_nob_transferred & ~LU_PAGE_MASK) {
1109                 CERROR("%s: unexpected bytes transferred: %d (%ld expected)\n",
1110                        exp->exp_obd->obd_name, req->rq_bulk->bd_nob_transferred,
1111                        PAGE_SIZE * npages);
1112                 ptlrpc_req_finished(req);
1113                 RETURN(-EPROTO);
1114         }
1115
1116         *request = req;
1117         RETURN(0);
1118 }
1119
1120 static void mdc_release_page(struct page *page, int remove)
1121 {
1122         if (remove) {
1123                 lock_page(page);
1124                 if (likely(page->mapping != NULL))
1125                         delete_from_page_cache(page);
1126                 unlock_page(page);
1127         }
1128         put_page(page);
1129 }
1130
1131 static struct page *mdc_page_locate(struct address_space *mapping, __u64 *hash,
1132                                     __u64 *start, __u64 *end, int hash64)
1133 {
1134         /*
1135          * Complement of hash is used as an index so that
1136          * radix_tree_gang_lookup() can be used to find a page with starting
1137          * hash _smaller_ than one we are looking for.
1138          */
1139         unsigned long offset = hash_x_index(*hash, hash64);
1140         struct page *page;
1141         unsigned long flags;
1142         int found;
1143
1144         ll_xa_lock_irqsave(&mapping->i_pages, flags);
1145         found = radix_tree_gang_lookup(&mapping->page_tree,
1146                                        (void **)&page, offset, 1);
1147         if (found > 0 && !ll_xa_is_value(page)) {
1148                 struct lu_dirpage *dp;
1149
1150                 get_page(page);
1151                 ll_xa_unlock_irqrestore(&mapping->i_pages, flags);
1152                 /*
1153                  * In contrast to find_lock_page() we are sure that directory
1154                  * page cannot be truncated (while DLM lock is held) and,
1155                  * hence, can avoid restart.
1156                  *
1157                  * In fact, page cannot be locked here at all, because
1158                  * mdc_read_page_remote does synchronous io.
1159                  */
1160                 wait_on_page_locked(page);
1161                 if (PageUptodate(page)) {
1162                         dp = kmap(page);
1163                         if (BITS_PER_LONG == 32 && hash64) {
1164                                 *start = le64_to_cpu(dp->ldp_hash_start) >> 32;
1165                                 *end   = le64_to_cpu(dp->ldp_hash_end) >> 32;
1166                                 *hash  = *hash >> 32;
1167                         } else {
1168                                 *start = le64_to_cpu(dp->ldp_hash_start);
1169                                 *end   = le64_to_cpu(dp->ldp_hash_end);
1170                         }
1171                         if (unlikely(*start == 1 && *hash == 0))
1172                                 *hash = *start;
1173                         else
1174                                 LASSERTF(*start <= *hash, "start = %#llx"
1175                                          ",end = %#llx,hash = %#llx\n",
1176                                          *start, *end, *hash);
1177                         CDEBUG(D_VFSTRACE, "offset %lx [%#llx %#llx],"
1178                               " hash %#llx\n", offset, *start, *end, *hash);
1179                         if (*hash > *end) {
1180                                 kunmap(page);
1181                                 mdc_release_page(page, 0);
1182                                 page = NULL;
1183                         } else if (*end != *start && *hash == *end) {
1184                                 /*
1185                                  * upon hash collision, remove this page,
1186                                  * otherwise put page reference, and
1187                                  * mdc_read_page_remote() will issue RPC to
1188                                  * fetch the page we want.
1189                                  */
1190                                 kunmap(page);
1191                                 mdc_release_page(page,
1192                                     le32_to_cpu(dp->ldp_flags) & LDF_COLLIDE);
1193                                 page = NULL;
1194                         }
1195                 } else {
1196                         put_page(page);
1197                         page = ERR_PTR(-EIO);
1198                 }
1199         } else {
1200                 ll_xa_unlock_irqrestore(&mapping->i_pages, flags);
1201                 page = NULL;
1202         }
1203         return page;
1204 }
1205
1206 /*
1207  * Adjust a set of pages, each page containing an array of lu_dirpages,
1208  * so that each page can be used as a single logical lu_dirpage.
1209  *
1210  * A lu_dirpage is laid out as follows, where s = ldp_hash_start,
1211  * e = ldp_hash_end, f = ldp_flags, p = padding, and each "ent" is a
1212  * struct lu_dirent.  It has size up to LU_PAGE_SIZE. The ldp_hash_end
1213  * value is used as a cookie to request the next lu_dirpage in a
1214  * directory listing that spans multiple pages (two in this example):
1215  *   ________
1216  *  |        |
1217  * .|--------v-------   -----.
1218  * |s|e|f|p|ent|ent| ... |ent|
1219  * '--|--------------   -----'   Each PAGE contains a single
1220  *    '------.                   lu_dirpage.
1221  * .---------v-------   -----.
1222  * |s|e|f|p|ent| 0 | ... | 0 |
1223  * '-----------------   -----'
1224  *
1225  * However, on hosts where the native VM page size (PAGE_SIZE) is
1226  * larger than LU_PAGE_SIZE, a single host page may contain multiple
1227  * lu_dirpages. After reading the lu_dirpages from the MDS, the
1228  * ldp_hash_end of the first lu_dirpage refers to the one immediately
1229  * after it in the same PAGE (arrows simplified for brevity, but
1230  * in general e0==s1, e1==s2, etc.):
1231  *
1232  * .--------------------   -----.
1233  * |s0|e0|f0|p|ent|ent| ... |ent|
1234  * |---v----------------   -----|
1235  * |s1|e1|f1|p|ent|ent| ... |ent|
1236  * |---v----------------   -----|  Here, each PAGE contains
1237  *             ...                 multiple lu_dirpages.
1238  * |---v----------------   -----|
1239  * |s'|e'|f'|p|ent|ent| ... |ent|
1240  * '---|----------------   -----'
1241  *     v
1242  * .----------------------------.
1243  * |        next PAGE           |
1244  *
1245  * This structure is transformed into a single logical lu_dirpage as follows:
1246  *
1247  * - Replace e0 with e' so the request for the next lu_dirpage gets the page
1248  *   labeled 'next PAGE'.
1249  *
1250  * - Copy the LDF_COLLIDE flag from f' to f0 to correctly reflect whether
1251  *   a hash collision with the next page exists.
1252  *
1253  * - Adjust the lde_reclen of the ending entry of each lu_dirpage to span
1254  *   to the first entry of the next lu_dirpage.
1255  */
1256 #if PAGE_SIZE > LU_PAGE_SIZE
1257 static void mdc_adjust_dirpages(struct page **pages, int cfs_pgs, int lu_pgs)
1258 {
1259         int i;
1260
1261         for (i = 0; i < cfs_pgs; i++) {
1262                 struct lu_dirpage *dp = kmap(pages[i]);
1263                 struct lu_dirpage *first = dp;
1264                 struct lu_dirent *end_dirent = NULL;
1265                 struct lu_dirent *ent;
1266                 __u64 hash_end = dp->ldp_hash_end;
1267                 __u32 flags = dp->ldp_flags;
1268
1269                 while (--lu_pgs > 0) {
1270                         ent = lu_dirent_start(dp);
1271                         for (end_dirent = ent; ent != NULL;
1272                              end_dirent = ent, ent = lu_dirent_next(ent));
1273
1274                         /* Advance dp to next lu_dirpage. */
1275                         dp = (struct lu_dirpage *)((char *)dp + LU_PAGE_SIZE);
1276
1277                         /* Check if we've reached the end of the PAGE. */
1278                         if (!((unsigned long)dp & ~PAGE_MASK))
1279                                 break;
1280
1281                         /* Save the hash and flags of this lu_dirpage. */
1282                         hash_end = dp->ldp_hash_end;
1283                         flags = dp->ldp_flags;
1284
1285                         /* Check if lu_dirpage contains no entries. */
1286                         if (end_dirent == NULL)
1287                                 break;
1288
1289                         /* Enlarge the end entry lde_reclen from 0 to
1290                          * first entry of next lu_dirpage. */
1291                         LASSERT(le16_to_cpu(end_dirent->lde_reclen) == 0);
1292                         end_dirent->lde_reclen =
1293                                 cpu_to_le16((char *)(dp->ldp_entries) -
1294                                             (char *)end_dirent);
1295                 }
1296
1297                 first->ldp_hash_end = hash_end;
1298                 first->ldp_flags &= ~cpu_to_le32(LDF_COLLIDE);
1299                 first->ldp_flags |= flags & cpu_to_le32(LDF_COLLIDE);
1300
1301                 kunmap(pages[i]);
1302         }
1303         LASSERTF(lu_pgs == 0, "left = %d\n", lu_pgs);
1304 }
1305 #else
1306 #define mdc_adjust_dirpages(pages, cfs_pgs, lu_pgs) do {} while (0)
1307 #endif  /* PAGE_SIZE > LU_PAGE_SIZE */
1308
1309 /* parameters for readdir page */
1310 struct readpage_param {
1311         struct md_op_data       *rp_mod;
1312         __u64                   rp_off;
1313         int                     rp_hash64;
1314         struct obd_export       *rp_exp;
1315         struct md_callback      *rp_cb;
1316 };
1317
1318 /**
1319  * Read pages from server.
1320  *
1321  * Page in MDS_READPAGE RPC is packed in LU_PAGE_SIZE, and each page contains
1322  * a header lu_dirpage which describes the start/end hash, and whether this
1323  * page is empty (contains no dir entry) or hash collide with next page.
1324  * After client receives reply, several pages will be integrated into dir page
1325  * in PAGE_SIZE (if PAGE_SIZE greater than LU_PAGE_SIZE), and the
1326  * lu_dirpage for this integrated page will be adjusted.
1327  **/
1328 static int mdc_read_page_remote(void *data, struct page *page0)
1329 {
1330         struct readpage_param *rp = data;
1331         struct page **page_pool;
1332         struct page *page;
1333         struct lu_dirpage *dp;
1334         struct md_op_data *op_data = rp->rp_mod;
1335         struct ptlrpc_request *req;
1336         int max_pages;
1337         struct inode *inode;
1338         struct lu_fid *fid;
1339         int rd_pgs = 0; /* number of pages actually read */
1340         int npages;
1341         int i;
1342         int rc;
1343         ENTRY;
1344
1345         max_pages = rp->rp_exp->exp_obd->u.cli.cl_max_pages_per_rpc;
1346         inode = op_data->op_data;
1347         fid = &op_data->op_fid1;
1348         LASSERT(inode != NULL);
1349
1350         OBD_ALLOC_PTR_ARRAY(page_pool, max_pages);
1351         if (page_pool != NULL) {
1352                 page_pool[0] = page0;
1353         } else {
1354                 page_pool = &page0;
1355                 max_pages = 1;
1356         }
1357
1358         for (npages = 1; npages < max_pages; npages++) {
1359                 page = page_cache_alloc(inode->i_mapping);
1360                 if (page == NULL)
1361                         break;
1362                 page_pool[npages] = page;
1363         }
1364
1365         rc = mdc_getpage(rp->rp_exp, fid, rp->rp_off, page_pool, npages, &req);
1366         if (rc < 0) {
1367                 /* page0 is special, which was added into page cache early */
1368                 delete_from_page_cache(page0);
1369         } else {
1370                 int lu_pgs;
1371
1372                 rd_pgs = (req->rq_bulk->bd_nob_transferred + PAGE_SIZE - 1) >>
1373                         PAGE_SHIFT;
1374                 lu_pgs = req->rq_bulk->bd_nob_transferred >> LU_PAGE_SHIFT;
1375                 LASSERT(!(req->rq_bulk->bd_nob_transferred & ~LU_PAGE_MASK));
1376
1377                 CDEBUG(D_INODE, "read %d(%d) pages\n", rd_pgs, lu_pgs);
1378
1379                 mdc_adjust_dirpages(page_pool, rd_pgs, lu_pgs);
1380
1381                 SetPageUptodate(page0);
1382         }
1383         unlock_page(page0);
1384
1385         ptlrpc_req_finished(req);
1386         CDEBUG(D_CACHE, "read %d/%d pages\n", rd_pgs, npages);
1387         for (i = 1; i < npages; i++) {
1388                 unsigned long   offset;
1389                 __u64           hash;
1390                 int ret;
1391
1392                 page = page_pool[i];
1393
1394                 if (rc < 0 || i >= rd_pgs) {
1395                         put_page(page);
1396                         continue;
1397                 }
1398
1399                 SetPageUptodate(page);
1400
1401                 dp = kmap(page);
1402                 hash = le64_to_cpu(dp->ldp_hash_start);
1403                 kunmap(page);
1404
1405                 offset = hash_x_index(hash, rp->rp_hash64);
1406
1407                 prefetchw(&page->flags);
1408                 ret = add_to_page_cache_lru(page, inode->i_mapping, offset,
1409                                             GFP_KERNEL);
1410                 if (ret == 0)
1411                         unlock_page(page);
1412                 else
1413                         CDEBUG(D_VFSTRACE, "page %lu add to page cache failed:"
1414                                " rc = %d\n", offset, ret);
1415                 put_page(page);
1416         }
1417
1418         if (page_pool != &page0)
1419                 OBD_FREE_PTR_ARRAY(page_pool, max_pages);
1420
1421         RETURN(rc);
1422 }
1423
1424 /**
1425  * Read dir page from cache first, if it can not find it, read it from
1426  * server and add into the cache.
1427  *
1428  * \param[in] exp       MDC export
1429  * \param[in] op_data   client MD stack parameters, transfering parameters
1430  *                      between different layers on client MD stack.
1431  * \param[in] cb_op     callback required for ldlm lock enqueue during
1432  *                      read page
1433  * \param[in] hash_offset the hash offset of the page to be read
1434  * \param[in] ppage     the page to be read
1435  *
1436  * retval               = 0 get the page successfully
1437  *                      errno(<0) get the page failed
1438  */
1439 static int mdc_read_page(struct obd_export *exp, struct md_op_data *op_data,
1440                          struct md_callback *cb_op, __u64 hash_offset,
1441                          struct page **ppage)
1442 {
1443         struct lookup_intent    it = { .it_op = IT_READDIR };
1444         struct page             *page;
1445         struct inode            *dir = op_data->op_data;
1446         struct address_space    *mapping;
1447         struct lu_dirpage       *dp;
1448         __u64                   start = 0;
1449         __u64                   end = 0;
1450         struct lustre_handle    lockh;
1451         struct ptlrpc_request   *enq_req = NULL;
1452         struct readpage_param   rp_param;
1453         int rc;
1454
1455         ENTRY;
1456
1457         *ppage = NULL;
1458
1459         LASSERT(dir != NULL);
1460         mapping = dir->i_mapping;
1461
1462         rc = mdc_intent_lock(exp, op_data, &it, &enq_req,
1463                              cb_op->md_blocking_ast, 0);
1464         if (enq_req != NULL)
1465                 ptlrpc_req_finished(enq_req);
1466
1467         if (rc < 0) {
1468                 CERROR("%s: "DFID" lock enqueue fails: rc = %d\n",
1469                        exp->exp_obd->obd_name, PFID(&op_data->op_fid1), rc);
1470                 RETURN(rc);
1471         }
1472
1473         rc = 0;
1474         lockh.cookie = it.it_lock_handle;
1475         mdc_set_lock_data(exp, &lockh, dir, NULL);
1476
1477         rp_param.rp_off = hash_offset;
1478         rp_param.rp_hash64 = op_data->op_cli_flags & CLI_HASH64;
1479         page = mdc_page_locate(mapping, &rp_param.rp_off, &start, &end,
1480                                rp_param.rp_hash64);
1481         if (IS_ERR(page)) {
1482                 CERROR("%s: dir page locate: "DFID" at %llu: rc %ld\n",
1483                        exp->exp_obd->obd_name, PFID(&op_data->op_fid1),
1484                        rp_param.rp_off, PTR_ERR(page));
1485                 GOTO(out_unlock, rc = PTR_ERR(page));
1486         } else if (page != NULL) {
1487                 /*
1488                  * XXX nikita: not entirely correct handling of a corner case:
1489                  * suppose hash chain of entries with hash value HASH crosses
1490                  * border between pages P0 and P1. First both P0 and P1 are
1491                  * cached, seekdir() is called for some entry from the P0 part
1492                  * of the chain. Later P0 goes out of cache. telldir(HASH)
1493                  * happens and finds P1, as it starts with matching hash
1494                  * value. Remaining entries from P0 part of the chain are
1495                  * skipped. (Is that really a bug?)
1496                  *
1497                  * Possible solutions: 0. don't cache P1 is such case, handle
1498                  * it as an "overflow" page. 1. invalidate all pages at
1499                  * once. 2. use HASH|1 as an index for P1.
1500                  */
1501                 GOTO(hash_collision, page);
1502         }
1503
1504         rp_param.rp_exp = exp;
1505         rp_param.rp_mod = op_data;
1506         page = read_cache_page(mapping,
1507                                hash_x_index(rp_param.rp_off,
1508                                             rp_param.rp_hash64),
1509                                mdc_read_page_remote, &rp_param);
1510         if (IS_ERR(page)) {
1511                 CDEBUG(D_INFO, "%s: read cache page: "DFID" at %llu: %ld\n",
1512                        exp->exp_obd->obd_name, PFID(&op_data->op_fid1),
1513                        rp_param.rp_off, PTR_ERR(page));
1514                 GOTO(out_unlock, rc = PTR_ERR(page));
1515         }
1516
1517         wait_on_page_locked(page);
1518         (void)kmap(page);
1519         if (!PageUptodate(page)) {
1520                 CERROR("%s: page not updated: "DFID" at %llu: rc %d\n",
1521                        exp->exp_obd->obd_name, PFID(&op_data->op_fid1),
1522                        rp_param.rp_off, -5);
1523                 goto fail;
1524         }
1525         if (!PageChecked(page))
1526                 SetPageChecked(page);
1527         if (PageError(page)) {
1528                 CERROR("%s: page error: "DFID" at %llu: rc %d\n",
1529                        exp->exp_obd->obd_name, PFID(&op_data->op_fid1),
1530                        rp_param.rp_off, -5);
1531                 goto fail;
1532         }
1533
1534 hash_collision:
1535         dp = page_address(page);
1536         if (BITS_PER_LONG == 32 && rp_param.rp_hash64) {
1537                 start = le64_to_cpu(dp->ldp_hash_start) >> 32;
1538                 end   = le64_to_cpu(dp->ldp_hash_end) >> 32;
1539                 rp_param.rp_off = hash_offset >> 32;
1540         } else {
1541                 start = le64_to_cpu(dp->ldp_hash_start);
1542                 end   = le64_to_cpu(dp->ldp_hash_end);
1543                 rp_param.rp_off = hash_offset;
1544         }
1545         if (end == start) {
1546                 LASSERT(start == rp_param.rp_off);
1547                 CWARN("Page-wide hash collision: %#lx\n", (unsigned long)end);
1548 #if BITS_PER_LONG == 32
1549                 CWARN("Real page-wide hash collision at [%llu %llu] with "
1550                       "hash %llu\n", le64_to_cpu(dp->ldp_hash_start),
1551                       le64_to_cpu(dp->ldp_hash_end), hash_offset);
1552 #endif
1553
1554                 /*
1555                  * Fetch whole overflow chain...
1556                  *
1557                  * XXX not yet.
1558                  */
1559                 goto fail;
1560         }
1561         *ppage = page;
1562 out_unlock:
1563         ldlm_lock_decref(&lockh, it.it_lock_mode);
1564         return rc;
1565 fail:
1566         kunmap(page);
1567         mdc_release_page(page, 1);
1568         rc = -EIO;
1569         goto out_unlock;
1570 }
1571
1572 static int mdc_statfs_interpret(const struct lu_env *env,
1573                                 struct ptlrpc_request *req, void *args, int rc)
1574 {
1575         struct obd_info *oinfo = args;
1576         struct obd_statfs *osfs;
1577
1578         if (!rc) {
1579                 osfs = req_capsule_server_get(&req->rq_pill, &RMF_OBD_STATFS);
1580                 if (!osfs)
1581                         return -EPROTO;
1582
1583                 oinfo->oi_osfs = osfs;
1584
1585                 CDEBUG(D_CACHE, "blocks=%llu free=%llu avail=%llu "
1586                        "objects=%llu free=%llu state=%x\n",
1587                         osfs->os_blocks, osfs->os_bfree, osfs->os_bavail,
1588                         osfs->os_files, osfs->os_ffree, osfs->os_state);
1589         }
1590
1591         oinfo->oi_cb_up(oinfo, rc);
1592
1593         return rc;
1594 }
1595
1596 static int mdc_statfs_async(struct obd_export *exp,
1597                             struct obd_info *oinfo, time64_t max_age,
1598                             struct ptlrpc_request_set *unused)
1599 {
1600         struct ptlrpc_request *req;
1601         struct obd_info *aa;
1602
1603         req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp), &RQF_MDS_STATFS,
1604                                         LUSTRE_MDS_VERSION, MDS_STATFS);
1605         if (req == NULL)
1606                 return -ENOMEM;
1607
1608         ptlrpc_request_set_replen(req);
1609         req->rq_interpret_reply = mdc_statfs_interpret;
1610
1611         aa = ptlrpc_req_async_args(aa, req);
1612         *aa = *oinfo;
1613
1614         ptlrpcd_add_req(req);
1615
1616         return 0;
1617 }
1618
1619 static int mdc_statfs(const struct lu_env *env,
1620                       struct obd_export *exp, struct obd_statfs *osfs,
1621                       time64_t max_age, __u32 flags)
1622 {
1623         struct obd_device *obd = class_exp2obd(exp);
1624         struct req_format *fmt;
1625         struct ptlrpc_request *req;
1626         struct obd_statfs *msfs;
1627         struct obd_import *imp = NULL;
1628         int rc;
1629         ENTRY;
1630
1631         /*
1632          * Since the request might also come from lprocfs, so we need
1633          * sync this with client_disconnect_export Bug15684
1634          */
1635         down_read(&obd->u.cli.cl_sem);
1636         if (obd->u.cli.cl_import)
1637                 imp = class_import_get(obd->u.cli.cl_import);
1638         up_read(&obd->u.cli.cl_sem);
1639         if (!imp)
1640                 RETURN(-ENODEV);
1641
1642         fmt = &RQF_MDS_STATFS;
1643         if ((exp_connect_flags2(exp) & OBD_CONNECT2_SUM_STATFS) &&
1644             (flags & OBD_STATFS_SUM))
1645                 fmt = &RQF_MDS_STATFS_NEW;
1646         req = ptlrpc_request_alloc_pack(imp, fmt, LUSTRE_MDS_VERSION,
1647                                         MDS_STATFS);
1648         if (req == NULL)
1649                 GOTO(output, rc = -ENOMEM);
1650
1651         if ((flags & OBD_STATFS_SUM) &&
1652             (exp_connect_flags2(exp) & OBD_CONNECT2_SUM_STATFS)) {
1653                 /* request aggregated states */
1654                 struct mdt_body *body;
1655
1656                 body = req_capsule_client_get(&req->rq_pill, &RMF_MDT_BODY);
1657                 if (body == NULL)
1658                         GOTO(out, rc = -EPROTO);
1659                 body->mbo_valid = OBD_MD_FLAGSTATFS;
1660         }
1661
1662         ptlrpc_request_set_replen(req);
1663
1664         if (flags & OBD_STATFS_NODELAY) {
1665                 /* procfs requests not want stay in wait for avoid deadlock */
1666                 req->rq_no_resend = 1;
1667                 req->rq_no_delay = 1;
1668         }
1669
1670         rc = ptlrpc_queue_wait(req);
1671         if (rc) {
1672                 /* check connection error first */
1673                 if (imp->imp_connect_error)
1674                         rc = imp->imp_connect_error;
1675                 GOTO(out, rc);
1676         }
1677
1678         msfs = req_capsule_server_get(&req->rq_pill, &RMF_OBD_STATFS);
1679         if (msfs == NULL)
1680                 GOTO(out, rc = -EPROTO);
1681
1682         *osfs = *msfs;
1683         EXIT;
1684 out:
1685         ptlrpc_req_finished(req);
1686 output:
1687         class_import_put(imp);
1688         return rc;
1689 }
1690
1691 static int mdc_ioc_fid2path(struct obd_export *exp, struct getinfo_fid2path *gf)
1692 {
1693         __u32 keylen, vallen;
1694         void *key;
1695         int rc;
1696
1697         if (gf->gf_pathlen > PATH_MAX)
1698                 RETURN(-ENAMETOOLONG);
1699         if (gf->gf_pathlen < 2)
1700                 RETURN(-EOVERFLOW);
1701
1702         /* Key is KEY_FID2PATH + getinfo_fid2path description */
1703         keylen = cfs_size_round(sizeof(KEY_FID2PATH) + sizeof(*gf) +
1704                                 sizeof(struct lu_fid));
1705         OBD_ALLOC(key, keylen);
1706         if (key == NULL)
1707                 RETURN(-ENOMEM);
1708         memcpy(key, KEY_FID2PATH, sizeof(KEY_FID2PATH));
1709         memcpy(key + cfs_size_round(sizeof(KEY_FID2PATH)), gf, sizeof(*gf));
1710         memcpy(key + cfs_size_round(sizeof(KEY_FID2PATH)) + sizeof(*gf),
1711                gf->gf_u.gf_root_fid, sizeof(struct lu_fid));
1712         CDEBUG(D_IOCTL, "path get "DFID" from %llu #%d\n",
1713                PFID(&gf->gf_fid), gf->gf_recno, gf->gf_linkno);
1714
1715         if (!fid_is_sane(&gf->gf_fid))
1716                 GOTO(out, rc = -EINVAL);
1717
1718         /* Val is struct getinfo_fid2path result plus path */
1719         vallen = sizeof(*gf) + gf->gf_pathlen;
1720
1721         rc = obd_get_info(NULL, exp, keylen, key, &vallen, gf);
1722         if (rc != 0 && rc != -EREMOTE)
1723                 GOTO(out, rc);
1724
1725         if (vallen <= sizeof(*gf))
1726                 GOTO(out, rc = -EPROTO);
1727         if (vallen > sizeof(*gf) + gf->gf_pathlen)
1728                 GOTO(out, rc = -EOVERFLOW);
1729
1730         CDEBUG(D_IOCTL, "path got "DFID" from %llu #%d: %s\n",
1731                PFID(&gf->gf_fid), gf->gf_recno, gf->gf_linkno,
1732                gf->gf_pathlen < 512 ? gf->gf_u.gf_path :
1733                /* only log the last 512 characters of the path */
1734                gf->gf_u.gf_path + gf->gf_pathlen - 512);
1735
1736 out:
1737         OBD_FREE(key, keylen);
1738         return rc;
1739 }
1740
1741 static int mdc_ioc_hsm_progress(struct obd_export *exp,
1742                                 struct hsm_progress_kernel *hpk)
1743 {
1744         struct obd_import               *imp = class_exp2cliimp(exp);
1745         struct hsm_progress_kernel      *req_hpk;
1746         struct ptlrpc_request           *req;
1747         int                              rc;
1748         ENTRY;
1749
1750         req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_HSM_PROGRESS,
1751                                         LUSTRE_MDS_VERSION, MDS_HSM_PROGRESS);
1752         if (req == NULL)
1753                 GOTO(out, rc = -ENOMEM);
1754
1755         mdc_pack_body(req, NULL, 0, 0, -1, 0);
1756
1757         /* Copy hsm_progress struct */
1758         req_hpk = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_PROGRESS);
1759         if (req_hpk == NULL)
1760                 GOTO(out, rc = -EPROTO);
1761
1762         *req_hpk = *hpk;
1763         req_hpk->hpk_errval = lustre_errno_hton(hpk->hpk_errval);
1764
1765         ptlrpc_request_set_replen(req);
1766
1767         ptlrpc_get_mod_rpc_slot(req);
1768         rc = ptlrpc_queue_wait(req);
1769         ptlrpc_put_mod_rpc_slot(req);
1770
1771         GOTO(out, rc);
1772 out:
1773         ptlrpc_req_finished(req);
1774         return rc;
1775 }
1776 /**
1777  * Send hsm_ct_register to MDS
1778  *
1779  * \param[in]   imp             import
1780  * \param[in]   archive_count   if in bitmap format, it is the bitmap,
1781  *                              else it is the count of archive_ids
1782  * \param[in]   archives        if in bitmap format, it is NULL,
1783  *                              else it is archive_id lists
1784  */
1785 static int mdc_ioc_hsm_ct_register(struct obd_import *imp, __u32 archive_count,
1786                                    __u32 *archives)
1787 {
1788         struct ptlrpc_request *req;
1789         __u32 *archive_array;
1790         size_t archives_size;
1791         int rc;
1792         ENTRY;
1793
1794         req = ptlrpc_request_alloc(imp, &RQF_MDS_HSM_CT_REGISTER);
1795         if (req == NULL)
1796                 RETURN(-ENOMEM);
1797
1798         if (archives != NULL)
1799                 archives_size = sizeof(*archive_array) * archive_count;
1800         else
1801                 archives_size = sizeof(archive_count);
1802
1803         req_capsule_set_size(&req->rq_pill, &RMF_MDS_HSM_ARCHIVE,
1804                              RCL_CLIENT, archives_size);
1805
1806         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_CT_REGISTER);
1807         if (rc) {
1808                 ptlrpc_request_free(req);
1809                 RETURN(-ENOMEM);
1810         }
1811
1812         mdc_pack_body(req, NULL, 0, 0, -1, 0);
1813
1814         archive_array = req_capsule_client_get(&req->rq_pill,
1815                                                &RMF_MDS_HSM_ARCHIVE);
1816         if (archive_array == NULL)
1817                 GOTO(out, rc = -EPROTO);
1818
1819         if (archives != NULL)
1820                 memcpy(archive_array, archives, archives_size);
1821         else
1822                 *archive_array = archive_count;
1823
1824         ptlrpc_request_set_replen(req);
1825         req->rq_no_resend = 1;
1826
1827         rc = mdc_queue_wait(req);
1828         GOTO(out, rc);
1829 out:
1830         ptlrpc_req_finished(req);
1831         return rc;
1832 }
1833
1834 static int mdc_ioc_hsm_current_action(struct obd_export *exp,
1835                                       struct md_op_data *op_data)
1836 {
1837         struct hsm_current_action       *hca = op_data->op_data;
1838         struct hsm_current_action       *req_hca;
1839         struct ptlrpc_request           *req;
1840         int                              rc;
1841         ENTRY;
1842
1843         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1844                                    &RQF_MDS_HSM_ACTION);
1845         if (req == NULL)
1846                 RETURN(-ENOMEM);
1847
1848         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_ACTION);
1849         if (rc) {
1850                 ptlrpc_request_free(req);
1851                 RETURN(rc);
1852         }
1853
1854         mdc_pack_body(req, &op_data->op_fid1, 0, 0,
1855                       op_data->op_suppgids[0], 0);
1856
1857         ptlrpc_request_set_replen(req);
1858
1859         rc = mdc_queue_wait(req);
1860         if (rc)
1861                 GOTO(out, rc);
1862
1863         req_hca = req_capsule_server_get(&req->rq_pill,
1864                                          &RMF_MDS_HSM_CURRENT_ACTION);
1865         if (req_hca == NULL)
1866                 GOTO(out, rc = -EPROTO);
1867
1868         *hca = *req_hca;
1869
1870         EXIT;
1871 out:
1872         ptlrpc_req_finished(req);
1873         return rc;
1874 }
1875
1876 static int mdc_ioc_hsm_ct_unregister(struct obd_import *imp)
1877 {
1878         struct ptlrpc_request   *req;
1879         int                      rc;
1880         ENTRY;
1881
1882         req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_HSM_CT_UNREGISTER,
1883                                         LUSTRE_MDS_VERSION,
1884                                         MDS_HSM_CT_UNREGISTER);
1885         if (req == NULL)
1886                 GOTO(out, rc = -ENOMEM);
1887
1888         mdc_pack_body(req, NULL, 0, 0, -1, 0);
1889
1890         ptlrpc_request_set_replen(req);
1891
1892         rc = mdc_queue_wait(req);
1893         GOTO(out, rc);
1894 out:
1895         ptlrpc_req_finished(req);
1896         return rc;
1897 }
1898
1899 static int mdc_ioc_hsm_state_get(struct obd_export *exp,
1900                                  struct md_op_data *op_data)
1901 {
1902         struct hsm_user_state   *hus = op_data->op_data;
1903         struct hsm_user_state   *req_hus;
1904         struct ptlrpc_request   *req;
1905         int                      rc;
1906         ENTRY;
1907
1908         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1909                                    &RQF_MDS_HSM_STATE_GET);
1910         if (req == NULL)
1911                 RETURN(-ENOMEM);
1912
1913         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_STATE_GET);
1914         if (rc != 0) {
1915                 ptlrpc_request_free(req);
1916                 RETURN(rc);
1917         }
1918
1919         mdc_pack_body(req, &op_data->op_fid1, 0, 0,
1920                       op_data->op_suppgids[0], 0);
1921
1922         ptlrpc_request_set_replen(req);
1923
1924         rc = mdc_queue_wait(req);
1925         if (rc)
1926                 GOTO(out, rc);
1927
1928         req_hus = req_capsule_server_get(&req->rq_pill, &RMF_HSM_USER_STATE);
1929         if (req_hus == NULL)
1930                 GOTO(out, rc = -EPROTO);
1931
1932         *hus = *req_hus;
1933
1934         EXIT;
1935 out:
1936         ptlrpc_req_finished(req);
1937         return rc;
1938 }
1939
1940 static int mdc_ioc_hsm_state_set(struct obd_export *exp,
1941                                  struct md_op_data *op_data)
1942 {
1943         struct hsm_state_set    *hss = op_data->op_data;
1944         struct hsm_state_set    *req_hss;
1945         struct ptlrpc_request   *req;
1946         int                      rc;
1947         ENTRY;
1948
1949         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1950                                    &RQF_MDS_HSM_STATE_SET);
1951         if (req == NULL)
1952                 RETURN(-ENOMEM);
1953
1954         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_STATE_SET);
1955         if (rc) {
1956                 ptlrpc_request_free(req);
1957                 RETURN(rc);
1958         }
1959
1960         mdc_pack_body(req, &op_data->op_fid1, 0, 0,
1961                       op_data->op_suppgids[0], 0);
1962
1963         /* Copy states */
1964         req_hss = req_capsule_client_get(&req->rq_pill, &RMF_HSM_STATE_SET);
1965         if (req_hss == NULL)
1966                 GOTO(out, rc = -EPROTO);
1967         *req_hss = *hss;
1968
1969         ptlrpc_request_set_replen(req);
1970
1971         ptlrpc_get_mod_rpc_slot(req);
1972         rc = ptlrpc_queue_wait(req);
1973         ptlrpc_put_mod_rpc_slot(req);
1974
1975         GOTO(out, rc);
1976 out:
1977         ptlrpc_req_finished(req);
1978         return rc;
1979 }
1980
1981 static int mdc_ioc_hsm_request(struct obd_export *exp,
1982                                struct hsm_user_request *hur)
1983 {
1984         struct obd_import       *imp = class_exp2cliimp(exp);
1985         struct ptlrpc_request   *req;
1986         struct hsm_request      *req_hr;
1987         struct hsm_user_item    *req_hui;
1988         char                    *req_opaque;
1989         int                      rc;
1990         ENTRY;
1991
1992         req = ptlrpc_request_alloc(imp, &RQF_MDS_HSM_REQUEST);
1993         if (req == NULL)
1994                 GOTO(out, rc = -ENOMEM);
1995
1996         req_capsule_set_size(&req->rq_pill, &RMF_MDS_HSM_USER_ITEM, RCL_CLIENT,
1997                              hur->hur_request.hr_itemcount
1998                              * sizeof(struct hsm_user_item));
1999         req_capsule_set_size(&req->rq_pill, &RMF_GENERIC_DATA, RCL_CLIENT,
2000                              hur->hur_request.hr_data_len);
2001
2002         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_REQUEST);
2003         if (rc) {
2004                 ptlrpc_request_free(req);
2005                 RETURN(rc);
2006         }
2007
2008         mdc_pack_body(req, NULL, 0, 0, -1, 0);
2009
2010         /* Copy hsm_request struct */
2011         req_hr = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_REQUEST);
2012         if (req_hr == NULL)
2013                 GOTO(out, rc = -EPROTO);
2014         *req_hr = hur->hur_request;
2015
2016         /* Copy hsm_user_item structs */
2017         req_hui = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_USER_ITEM);
2018         if (req_hui == NULL)
2019                 GOTO(out, rc = -EPROTO);
2020         memcpy(req_hui, hur->hur_user_item,
2021                hur->hur_request.hr_itemcount * sizeof(struct hsm_user_item));
2022
2023         /* Copy opaque field */
2024         req_opaque = req_capsule_client_get(&req->rq_pill, &RMF_GENERIC_DATA);
2025         if (req_opaque == NULL)
2026                 GOTO(out, rc = -EPROTO);
2027         memcpy(req_opaque, hur_data(hur), hur->hur_request.hr_data_len);
2028
2029         ptlrpc_request_set_replen(req);
2030
2031         ptlrpc_get_mod_rpc_slot(req);
2032         rc = ptlrpc_queue_wait(req);
2033         ptlrpc_put_mod_rpc_slot(req);
2034
2035         GOTO(out, rc);
2036
2037 out:
2038         ptlrpc_req_finished(req);
2039         return rc;
2040 }
2041
2042 static int mdc_ioc_hsm_ct_start(struct obd_export *exp,
2043                                 struct lustre_kernelcomm *lk);
2044
2045 static int mdc_quotactl(struct obd_device *unused, struct obd_export *exp,
2046                         struct obd_quotactl *oqctl)
2047 {
2048         struct ptlrpc_request *req;
2049         struct obd_quotactl *oqc;
2050         int rc;
2051         ENTRY;
2052
2053         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_QUOTACTL);
2054         if (req == NULL)
2055                 RETURN(-ENOMEM);
2056
2057
2058         if (LUSTRE_Q_CMD_IS_POOL(oqctl->qc_cmd))
2059                 req_capsule_set_size(&req->rq_pill,
2060                                      &RMF_OBD_QUOTACTL,
2061                                      RCL_CLIENT,
2062                                      sizeof(*oqc) + LOV_MAXPOOLNAME + 1);
2063
2064         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION,
2065                                  MDS_QUOTACTL);
2066         if (rc) {
2067                 ptlrpc_request_free(req);
2068                 RETURN(rc);
2069         }
2070
2071         oqc = req_capsule_client_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
2072         QCTL_COPY(oqc, oqctl);
2073
2074         ptlrpc_request_set_replen(req);
2075         ptlrpc_at_set_req_timeout(req);
2076
2077         rc = ptlrpc_queue_wait(req);
2078         if (rc) {
2079                 CERROR("%s: ptlrpc_queue_wait failed: rc = %d\n",
2080                        exp->exp_obd->obd_name, rc);
2081                 GOTO(out, rc);
2082         }
2083
2084         if (req->rq_repmsg &&
2085             (oqc = req_capsule_server_get(&req->rq_pill, &RMF_OBD_QUOTACTL))) {
2086                 QCTL_COPY(oqctl, oqc);
2087         } else if (!rc) {
2088                 rc = -EPROTO;
2089                 CERROR("%s: cannot unpack obd_quotactl: rc = %d\n",
2090                         exp->exp_obd->obd_name, rc);
2091         }
2092 out:
2093         ptlrpc_req_finished(req);
2094
2095         RETURN(rc);
2096 }
2097
2098 static int mdc_ioc_swap_layouts(struct obd_export *exp,
2099                                 struct md_op_data *op_data)
2100 {
2101         LIST_HEAD(cancels);
2102         struct ptlrpc_request   *req;
2103         int                      rc, count;
2104         struct mdc_swap_layouts *msl, *payload;
2105         ENTRY;
2106
2107         msl = op_data->op_data;
2108
2109         /* When the MDT will get the MDS_SWAP_LAYOUTS RPC the
2110          * first thing it will do is to cancel the 2 layout
2111          * locks held by this client.
2112          * So the client must cancel its layout locks on the 2 fids
2113          * with the request RPC to avoid extra RPC round trips.
2114          */
2115         count = mdc_resource_get_unused(exp, &op_data->op_fid1, &cancels,
2116                                         LCK_EX, MDS_INODELOCK_LAYOUT |
2117                                         MDS_INODELOCK_XATTR);
2118         count += mdc_resource_get_unused(exp, &op_data->op_fid2, &cancels,
2119                                          LCK_EX, MDS_INODELOCK_LAYOUT |
2120                                          MDS_INODELOCK_XATTR);
2121
2122         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
2123                                    &RQF_MDS_SWAP_LAYOUTS);
2124         if (req == NULL) {
2125                 ldlm_lock_list_put(&cancels, l_bl_ast, count);
2126                 RETURN(-ENOMEM);
2127         }
2128
2129         rc = mdc_prep_elc_req(exp, req, MDS_SWAP_LAYOUTS, &cancels, count);
2130         if (rc) {
2131                 ptlrpc_request_free(req);
2132                 RETURN(rc);
2133         }
2134
2135         mdc_swap_layouts_pack(req, op_data);
2136
2137         payload = req_capsule_client_get(&req->rq_pill, &RMF_SWAP_LAYOUTS);
2138         LASSERT(payload);
2139
2140         *payload = *msl;
2141
2142         ptlrpc_request_set_replen(req);
2143
2144         rc = ptlrpc_queue_wait(req);
2145         if (rc)
2146                 GOTO(out, rc);
2147         EXIT;
2148
2149 out:
2150         ptlrpc_req_finished(req);
2151         return rc;
2152 }
2153
2154 static int mdc_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
2155                          void *karg, void __user *uarg)
2156 {
2157         struct obd_device *obd = exp->exp_obd;
2158         struct obd_ioctl_data *data = karg;
2159         struct obd_import *imp = obd->u.cli.cl_import;
2160         int rc;
2161         ENTRY;
2162
2163         if (!try_module_get(THIS_MODULE)) {
2164                 CERROR("%s: cannot get module '%s'\n", obd->obd_name,
2165                        module_name(THIS_MODULE));
2166                 return -EINVAL;
2167         }
2168         switch (cmd) {
2169         case OBD_IOC_FID2PATH:
2170                 rc = mdc_ioc_fid2path(exp, karg);
2171                 GOTO(out, rc);
2172         case LL_IOC_HSM_CT_START:
2173                 rc = mdc_ioc_hsm_ct_start(exp, karg);
2174                 /* ignore if it was already registered on this MDS. */
2175                 if (rc == -EEXIST)
2176                         rc = 0;
2177                 GOTO(out, rc);
2178         case LL_IOC_HSM_PROGRESS:
2179                 rc = mdc_ioc_hsm_progress(exp, karg);
2180                 GOTO(out, rc);
2181         case LL_IOC_HSM_STATE_GET:
2182                 rc = mdc_ioc_hsm_state_get(exp, karg);
2183                 GOTO(out, rc);
2184         case LL_IOC_HSM_STATE_SET:
2185                 rc = mdc_ioc_hsm_state_set(exp, karg);
2186                 GOTO(out, rc);
2187         case LL_IOC_HSM_ACTION:
2188                 rc = mdc_ioc_hsm_current_action(exp, karg);
2189                 GOTO(out, rc);
2190         case LL_IOC_HSM_REQUEST:
2191                 rc = mdc_ioc_hsm_request(exp, karg);
2192                 GOTO(out, rc);
2193         case OBD_IOC_CLIENT_RECOVER:
2194                 rc = ptlrpc_recover_import(imp, data->ioc_inlbuf1, 0);
2195                 if (rc < 0)
2196                         GOTO(out, rc);
2197                 GOTO(out, rc = 0);
2198         case IOC_OSC_SET_ACTIVE:
2199                 rc = ptlrpc_set_import_active(imp, data->ioc_offset);
2200                 GOTO(out, rc);
2201         /*
2202          * Normally IOC_OBD_STATFS, OBD_IOC_QUOTACTL iocontrol are handled by
2203          * LMV instead of MDC. But when the cluster is upgraded from 1.8,
2204          * there'd be no LMV layer thus we might be called here. Eventually
2205          * this code should be removed.
2206          * bz20731, LU-592.
2207          */
2208         case IOC_OBD_STATFS: {
2209                 struct obd_statfs stat_buf = {0};
2210
2211                 if (*((__u32 *) data->ioc_inlbuf2) != 0)
2212                         GOTO(out, rc = -ENODEV);
2213
2214                 /* copy UUID */
2215                 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(obd),
2216                                  min((int)data->ioc_plen2,
2217                                      (int)sizeof(struct obd_uuid))))
2218                         GOTO(out, rc = -EFAULT);
2219
2220                 rc = mdc_statfs(NULL, obd->obd_self_export, &stat_buf,
2221                                 ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS,
2222                                 0);
2223                 if (rc != 0)
2224                         GOTO(out, rc);
2225
2226                 if (copy_to_user(data->ioc_pbuf1, &stat_buf,
2227                                      min((int) data->ioc_plen1,
2228                                          (int) sizeof(stat_buf))))
2229                         GOTO(out, rc = -EFAULT);
2230
2231                 GOTO(out, rc = 0);
2232         }
2233         case OBD_IOC_QUOTACTL: {
2234                 struct if_quotactl *qctl = karg;
2235                 struct obd_quotactl *oqctl;
2236
2237                 OBD_ALLOC_PTR(oqctl);
2238                 if (oqctl == NULL)
2239                         GOTO(out, rc = -ENOMEM);
2240
2241                 QCTL_COPY(oqctl, qctl);
2242                 rc = obd_quotactl(exp, oqctl);
2243                 if (rc == 0) {
2244                         QCTL_COPY(qctl, oqctl);
2245                         qctl->qc_valid = QC_MDTIDX;
2246                         qctl->obd_uuid = obd->u.cli.cl_target_uuid;
2247                 }
2248
2249                 OBD_FREE_PTR(oqctl);
2250                 GOTO(out, rc);
2251         }
2252         case LL_IOC_GET_CONNECT_FLAGS:
2253                 if (copy_to_user(uarg, exp_connect_flags_ptr(exp),
2254                                  sizeof(*exp_connect_flags_ptr(exp))))
2255                         GOTO(out, rc = -EFAULT);
2256
2257                 GOTO(out, rc = 0);
2258         case LL_IOC_LOV_SWAP_LAYOUTS:
2259                 rc = mdc_ioc_swap_layouts(exp, karg);
2260                 GOTO(out, rc);
2261         default:
2262                 CERROR("unrecognised ioctl: cmd = %#x\n", cmd);
2263                 GOTO(out, rc = -ENOTTY);
2264         }
2265 out:
2266         module_put(THIS_MODULE);
2267
2268         return rc;
2269 }
2270
2271 static int mdc_get_info_rpc(struct obd_export *exp,
2272                             u32 keylen, void *key,
2273                             u32 vallen, void *val)
2274 {
2275         struct obd_import      *imp = class_exp2cliimp(exp);
2276         struct ptlrpc_request  *req;
2277         char                   *tmp;
2278         int                     rc = -EINVAL;
2279         ENTRY;
2280
2281         req = ptlrpc_request_alloc(imp, &RQF_MDS_GET_INFO);
2282         if (req == NULL)
2283                 RETURN(-ENOMEM);
2284
2285         req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_KEY,
2286                              RCL_CLIENT, keylen);
2287         req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_VALLEN,
2288                              RCL_CLIENT, sizeof(vallen));
2289
2290         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GET_INFO);
2291         if (rc) {
2292                 ptlrpc_request_free(req);
2293                 RETURN(rc);
2294         }
2295
2296         tmp = req_capsule_client_get(&req->rq_pill, &RMF_GETINFO_KEY);
2297         memcpy(tmp, key, keylen);
2298         tmp = req_capsule_client_get(&req->rq_pill, &RMF_GETINFO_VALLEN);
2299         memcpy(tmp, &vallen, sizeof(vallen));
2300
2301         req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_VAL,
2302                              RCL_SERVER, vallen);
2303         ptlrpc_request_set_replen(req);
2304
2305         rc = ptlrpc_queue_wait(req);
2306         /* -EREMOTE means the get_info result is partial, and it needs to
2307          * continue on another MDT, see fid2path part in lmv_iocontrol */
2308         if (rc == 0 || rc == -EREMOTE) {
2309                 tmp = req_capsule_server_get(&req->rq_pill, &RMF_GETINFO_VAL);
2310                 memcpy(val, tmp, vallen);
2311                 if (ptlrpc_rep_need_swab(req)) {
2312                         if (KEY_IS(KEY_FID2PATH))
2313                                 lustre_swab_fid2path(val);
2314                 }
2315         }
2316         ptlrpc_req_finished(req);
2317
2318         RETURN(rc);
2319 }
2320
2321 static void lustre_swab_hai(struct hsm_action_item *h)
2322 {
2323         __swab32s(&h->hai_len);
2324         __swab32s(&h->hai_action);
2325         lustre_swab_lu_fid(&h->hai_fid);
2326         lustre_swab_lu_fid(&h->hai_dfid);
2327         __swab64s(&h->hai_cookie);
2328         __swab64s(&h->hai_extent.offset);
2329         __swab64s(&h->hai_extent.length);
2330         __swab64s(&h->hai_gid);
2331 }
2332
2333 static void lustre_swab_hal(struct hsm_action_list *h)
2334 {
2335         struct hsm_action_item  *hai;
2336         __u32                    i;
2337
2338         __swab32s(&h->hal_version);
2339         __swab32s(&h->hal_count);
2340         __swab32s(&h->hal_archive_id);
2341         __swab64s(&h->hal_flags);
2342         hai = hai_first(h);
2343         for (i = 0; i < h->hal_count; i++, hai = hai_next(hai))
2344                 lustre_swab_hai(hai);
2345 }
2346
2347 static void lustre_swab_kuch(struct kuc_hdr *l)
2348 {
2349         __swab16s(&l->kuc_magic);
2350         /* __u8 l->kuc_transport */
2351         __swab16s(&l->kuc_msgtype);
2352         __swab16s(&l->kuc_msglen);
2353 }
2354
2355 static int mdc_ioc_hsm_ct_start(struct obd_export *exp,
2356                                 struct lustre_kernelcomm *lk)
2357 {
2358         struct obd_import *imp = class_exp2cliimp(exp);
2359         int rc = 0;
2360
2361         if (lk->lk_group != KUC_GRP_HSM) {
2362                 CERROR("Bad copytool group %d\n", lk->lk_group);
2363                 return -EINVAL;
2364         }
2365
2366         CDEBUG(D_HSM, "CT start r%d w%d u%d g%d f%#x\n", lk->lk_rfd, lk->lk_wfd,
2367                lk->lk_uid, lk->lk_group, lk->lk_flags);
2368
2369         if (lk->lk_flags & LK_FLG_STOP) {
2370                 /* Unregister with the coordinator */
2371                 rc = mdc_ioc_hsm_ct_unregister(imp);
2372         } else {
2373                 __u32 *archives = NULL;
2374
2375                 if ((lk->lk_flags & LK_FLG_DATANR) && lk->lk_data_count > 0)
2376                         archives = lk->lk_data;
2377
2378                 rc = mdc_ioc_hsm_ct_register(imp, lk->lk_data_count, archives);
2379         }
2380
2381         return rc;
2382 }
2383
2384 /**
2385  * Send a message to any listening copytools
2386  * @param val KUC message (kuc_hdr + hsm_action_list)
2387  * @param len total length of message
2388  */
2389 static int mdc_hsm_copytool_send(const struct obd_uuid *uuid,
2390                                  size_t len, void *val)
2391 {
2392         struct kuc_hdr          *lh = (struct kuc_hdr *)val;
2393         struct hsm_action_list  *hal = (struct hsm_action_list *)(lh + 1);
2394         int                      rc;
2395         ENTRY;
2396
2397         if (len < sizeof(*lh) + sizeof(*hal)) {
2398                 CERROR("Short HSM message %zu < %zu\n", len,
2399                        sizeof(*lh) + sizeof(*hal));
2400                 RETURN(-EPROTO);
2401         }
2402         if (lh->kuc_magic == __swab16(KUC_MAGIC)) {
2403                 lustre_swab_kuch(lh);
2404                 lustre_swab_hal(hal);
2405         } else if (lh->kuc_magic != KUC_MAGIC) {
2406                 CERROR("Bad magic %x!=%x\n", lh->kuc_magic, KUC_MAGIC);
2407                 RETURN(-EPROTO);
2408         }
2409
2410         CDEBUG(D_HSM, " Received message mg=%x t=%d m=%d l=%d actions=%d "
2411                "on %s\n",
2412                lh->kuc_magic, lh->kuc_transport, lh->kuc_msgtype,
2413                lh->kuc_msglen, hal->hal_count, hal->hal_fsname);
2414
2415         /* Broadcast to HSM listeners */
2416         rc = libcfs_kkuc_group_put(uuid, KUC_GRP_HSM, lh);
2417
2418         RETURN(rc);
2419 }
2420
2421 /**
2422  * callback function passed to kuc for re-registering each HSM copytool
2423  * running on MDC, after MDT shutdown/recovery.
2424  * @param data copytool registration data
2425  * @param cb_arg callback argument (obd_import)
2426  */
2427 static int mdc_hsm_ct_reregister(void *data, void *cb_arg)
2428 {
2429         struct obd_import *imp = (struct obd_import *)cb_arg;
2430         struct kkuc_ct_data *kcd = data;
2431         __u32 *archives = NULL;
2432         int rc;
2433
2434         if (kcd == NULL ||
2435             (kcd->kcd_magic != KKUC_CT_DATA_ARRAY_MAGIC &&
2436              kcd->kcd_magic != KKUC_CT_DATA_BITMAP_MAGIC))
2437                 return -EPROTO;
2438
2439         if (kcd->kcd_magic == KKUC_CT_DATA_BITMAP_MAGIC) {
2440                 CDEBUG(D_HA, "%s: recover copytool registration to MDT "
2441                        "(archive=%#x)\n", imp->imp_obd->obd_name,
2442                        kcd->kcd_nr_archives);
2443         } else {
2444                 CDEBUG(D_HA, "%s: recover copytool registration to MDT "
2445                        "(archive nr = %u)\n",
2446                        imp->imp_obd->obd_name, kcd->kcd_nr_archives);
2447                 if (kcd->kcd_nr_archives != 0)
2448                         archives = kcd->kcd_archives;
2449         }
2450
2451         rc = mdc_ioc_hsm_ct_register(imp, kcd->kcd_nr_archives, archives);
2452         /* ignore error if the copytool is already registered */
2453         return (rc == -EEXIST) ? 0 : rc;
2454 }
2455
2456 /**
2457  * Re-establish all kuc contexts with MDT
2458  * after MDT shutdown/recovery.
2459  */
2460 static int mdc_kuc_reregister(struct obd_import *imp)
2461 {
2462         /* re-register HSM agents */
2463         return libcfs_kkuc_group_foreach(&imp->imp_obd->obd_uuid, KUC_GRP_HSM,
2464                                          mdc_hsm_ct_reregister, imp);
2465 }
2466
2467 static int mdc_set_info_async(const struct lu_env *env,
2468                               struct obd_export *exp,
2469                               u32 keylen, void *key,
2470                               u32 vallen, void *val,
2471                               struct ptlrpc_request_set *set)
2472 {
2473         struct obd_import       *imp = class_exp2cliimp(exp);
2474         int                      rc;
2475         ENTRY;
2476
2477         if (KEY_IS(KEY_READ_ONLY)) {
2478                 if (vallen != sizeof(int))
2479                         RETURN(-EINVAL);
2480
2481                 spin_lock(&imp->imp_lock);
2482                 if (*((int *)val)) {
2483                         imp->imp_connect_flags_orig |= OBD_CONNECT_RDONLY;
2484                         imp->imp_connect_data.ocd_connect_flags |=
2485                                                         OBD_CONNECT_RDONLY;
2486                 } else {
2487                         imp->imp_connect_flags_orig &= ~OBD_CONNECT_RDONLY;
2488                         imp->imp_connect_data.ocd_connect_flags &=
2489                                                         ~OBD_CONNECT_RDONLY;
2490                 }
2491                 spin_unlock(&imp->imp_lock);
2492
2493                 rc = do_set_info_async(imp, MDS_SET_INFO, LUSTRE_MDS_VERSION,
2494                                        keylen, key, vallen, val, set);
2495                 RETURN(rc);
2496         }
2497         if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
2498                 rc = do_set_info_async(imp, MDS_SET_INFO, LUSTRE_MDS_VERSION,
2499                                        keylen, key, vallen, val, set);
2500                 RETURN(rc);
2501         }
2502         if (KEY_IS(KEY_HSM_COPYTOOL_SEND)) {
2503                 rc = mdc_hsm_copytool_send(&imp->imp_obd->obd_uuid, vallen,
2504                                            val);
2505                 RETURN(rc);
2506         }
2507
2508         if (KEY_IS(KEY_DEFAULT_EASIZE)) {
2509                 __u32 *default_easize = val;
2510
2511                 exp->exp_obd->u.cli.cl_default_mds_easize = *default_easize;
2512                 RETURN(0);
2513         }
2514
2515         rc = osc_set_info_async(env, exp, keylen, key, vallen, val, set);
2516         RETURN(rc);
2517 }
2518
2519 static int mdc_get_info(const struct lu_env *env, struct obd_export *exp,
2520                         __u32 keylen, void *key, __u32 *vallen, void *val)
2521 {
2522         int rc = -EINVAL;
2523
2524         if (KEY_IS(KEY_MAX_EASIZE)) {
2525                 __u32 mdsize, *max_easize;
2526
2527                 if (*vallen != sizeof(int))
2528                         RETURN(-EINVAL);
2529                 mdsize = *(__u32 *)val;
2530                 if (mdsize > exp->exp_obd->u.cli.cl_max_mds_easize)
2531                         exp->exp_obd->u.cli.cl_max_mds_easize = mdsize;
2532                 max_easize = val;
2533                 *max_easize = exp->exp_obd->u.cli.cl_max_mds_easize;
2534                 RETURN(0);
2535         } else if (KEY_IS(KEY_DEFAULT_EASIZE)) {
2536                 __u32 *default_easize;
2537
2538                 if (*vallen != sizeof(int))
2539                         RETURN(-EINVAL);
2540                 default_easize = val;
2541                 *default_easize = exp->exp_obd->u.cli.cl_default_mds_easize;
2542                 RETURN(0);
2543         } else if (KEY_IS(KEY_CONN_DATA)) {
2544                 struct obd_import *imp = class_exp2cliimp(exp);
2545                 struct obd_connect_data *data = val;
2546
2547                 if (*vallen != sizeof(*data))
2548                         RETURN(-EINVAL);
2549
2550                 *data = imp->imp_connect_data;
2551                 RETURN(0);
2552         } else if (KEY_IS(KEY_TGT_COUNT)) {
2553                 *((__u32 *)val) = 1;
2554                 RETURN(0);
2555         }
2556
2557         rc = mdc_get_info_rpc(exp, keylen, key, *vallen, val);
2558
2559         RETURN(rc);
2560 }
2561
2562 static int mdc_fsync(struct obd_export *exp, const struct lu_fid *fid,
2563                      struct ptlrpc_request **request)
2564 {
2565         struct ptlrpc_request *req;
2566         int                    rc;
2567         ENTRY;
2568
2569         *request = NULL;
2570         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_SYNC);
2571         if (req == NULL)
2572                 RETURN(-ENOMEM);
2573
2574         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_SYNC);
2575         if (rc) {
2576                 ptlrpc_request_free(req);
2577                 RETURN(rc);
2578         }
2579
2580         mdc_pack_body(req, fid, 0, 0, -1, 0);
2581
2582         ptlrpc_request_set_replen(req);
2583
2584         rc = ptlrpc_queue_wait(req);
2585         if (rc)
2586                 ptlrpc_req_finished(req);
2587         else
2588                 *request = req;
2589         RETURN(rc);
2590 }
2591
2592 struct mdc_rmfid_args {
2593         int *mra_rcs;
2594         int mra_nr;
2595 };
2596
2597 int mdc_rmfid_interpret(const struct lu_env *env, struct ptlrpc_request *req,
2598                           void *args, int rc)
2599 {
2600         struct mdc_rmfid_args *aa;
2601         int *rcs, size;
2602         ENTRY;
2603
2604         if (!rc) {
2605                 aa = ptlrpc_req_async_args(aa, req);
2606
2607                 size = req_capsule_get_size(&req->rq_pill, &RMF_RCS,
2608                                             RCL_SERVER);
2609                 LASSERT(size == sizeof(int) * aa->mra_nr);
2610                 rcs = req_capsule_server_get(&req->rq_pill, &RMF_RCS);
2611                 LASSERT(rcs);
2612                 LASSERT(aa->mra_rcs);
2613                 LASSERT(aa->mra_nr);
2614                 memcpy(aa->mra_rcs, rcs, size);
2615         }
2616
2617         RETURN(rc);
2618 }
2619
2620 static int mdc_rmfid(struct obd_export *exp, struct fid_array *fa,
2621                      int *rcs, struct ptlrpc_request_set *set)
2622 {
2623         struct ptlrpc_request *req;
2624         struct mdc_rmfid_args *aa;
2625         struct mdt_body *b;
2626         struct lu_fid *tmp;
2627         int rc, flen;
2628         ENTRY;
2629
2630         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_RMFID);
2631         if (req == NULL)
2632                 RETURN(-ENOMEM);
2633
2634         flen = fa->fa_nr * sizeof(struct lu_fid);
2635         req_capsule_set_size(&req->rq_pill, &RMF_FID_ARRAY,
2636                              RCL_CLIENT, flen);
2637         req_capsule_set_size(&req->rq_pill, &RMF_FID_ARRAY,
2638                              RCL_SERVER, flen);
2639         req_capsule_set_size(&req->rq_pill, &RMF_RCS,
2640                              RCL_SERVER, fa->fa_nr * sizeof(__u32));
2641         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_RMFID);
2642         if (rc) {
2643                 ptlrpc_request_free(req);
2644                 RETURN(rc);
2645         }
2646         tmp = req_capsule_client_get(&req->rq_pill, &RMF_FID_ARRAY);
2647         memcpy(tmp, fa->fa_fids, flen);
2648
2649         mdc_pack_body(req, NULL, 0, 0, -1, 0);
2650         b = req_capsule_client_get(&req->rq_pill, &RMF_MDT_BODY);
2651         b->mbo_ctime = ktime_get_real_seconds();
2652
2653         ptlrpc_request_set_replen(req);
2654
2655         LASSERT(rcs);
2656         aa = ptlrpc_req_async_args(aa, req);
2657         aa->mra_rcs = rcs;
2658         aa->mra_nr = fa->fa_nr;
2659         req->rq_interpret_reply = mdc_rmfid_interpret;
2660
2661         ptlrpc_set_add_req(set, req);
2662         ptlrpc_check_set(NULL, set);
2663
2664         RETURN(rc);
2665 }
2666
2667 static int mdc_import_event(struct obd_device *obd, struct obd_import *imp,
2668                             enum obd_import_event event)
2669 {
2670         struct client_obd *cli = &obd->u.cli;
2671         int rc = 0;
2672
2673         LASSERT(imp->imp_obd == obd);
2674
2675         switch (event) {
2676         case IMP_EVENT_DISCON:
2677                 spin_lock(&cli->cl_loi_list_lock);
2678                 cli->cl_avail_grant = 0;
2679                 cli->cl_lost_grant = 0;
2680                 spin_unlock(&cli->cl_loi_list_lock);
2681                 break;
2682         case IMP_EVENT_INACTIVE:
2683                 /*
2684                  * Flush current sequence to make client obtain new one
2685                  * from server in case of disconnect/reconnect.
2686                  */
2687                 down_read(&cli->cl_seq_rwsem);
2688                 if (cli->cl_seq)
2689                         seq_client_flush(cli->cl_seq);
2690                 up_read(&cli->cl_seq_rwsem);
2691
2692                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_INACTIVE);
2693                 break;
2694         case IMP_EVENT_INVALIDATE: {
2695                 struct ldlm_namespace *ns = obd->obd_namespace;
2696                 struct lu_env *env;
2697                 __u16 refcheck;
2698
2699                 ldlm_namespace_cleanup(ns, LDLM_FL_LOCAL_ONLY);
2700
2701                 env = cl_env_get(&refcheck);
2702                 if (!IS_ERR(env)) {
2703                         /* Reset grants. All pages go to failing rpcs due to
2704                          * the invalid import.
2705                          */
2706                         osc_io_unplug(env, cli, NULL);
2707
2708                         cfs_hash_for_each_nolock(ns->ns_rs_hash,
2709                                                  osc_ldlm_resource_invalidate,
2710                                                  env, 0);
2711                         cl_env_put(env, &refcheck);
2712                         ldlm_namespace_cleanup(ns, LDLM_FL_LOCAL_ONLY);
2713                 } else {
2714                         rc = PTR_ERR(env);
2715                 }
2716                 break;
2717         }
2718         case IMP_EVENT_ACTIVE:
2719                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_ACTIVE);
2720                 /* redo the kuc registration after reconnecting */
2721                 if (rc == 0)
2722                         rc = mdc_kuc_reregister(imp);
2723                 break;
2724         case IMP_EVENT_OCD: {
2725                 struct obd_connect_data *ocd = &imp->imp_connect_data;
2726
2727                 if (OCD_HAS_FLAG(ocd, GRANT))
2728                         osc_init_grant(cli, ocd);
2729
2730                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_OCD);
2731                 break;
2732         }
2733         case IMP_EVENT_DEACTIVATE:
2734         case IMP_EVENT_ACTIVATE:
2735                 break;
2736         default:
2737                 CERROR("Unknown import event %x\n", event);
2738                 LBUG();
2739         }
2740         RETURN(rc);
2741 }
2742
2743 int mdc_fid_alloc(const struct lu_env *env, struct obd_export *exp,
2744                   struct lu_fid *fid, struct md_op_data *op_data)
2745 {
2746         struct client_obd *cli = &exp->exp_obd->u.cli;
2747         int rc = -EIO;
2748
2749         ENTRY;
2750
2751         down_read(&cli->cl_seq_rwsem);
2752         if (cli->cl_seq)
2753                 rc = seq_client_alloc_fid(env, cli->cl_seq, fid);
2754         up_read(&cli->cl_seq_rwsem);
2755
2756         RETURN(rc);
2757 }
2758
2759 static struct obd_uuid *mdc_get_uuid(struct obd_export *exp)
2760 {
2761         struct client_obd *cli = &exp->exp_obd->u.cli;
2762         return &cli->cl_target_uuid;
2763 }
2764
2765 /**
2766  * Determine whether the lock can be canceled before replaying it during
2767  * recovery, non zero value will be return if the lock can be canceled,
2768  * or zero returned for not
2769  */
2770 static int mdc_cancel_weight(struct ldlm_lock *lock)
2771 {
2772         if (lock->l_resource->lr_type != LDLM_IBITS)
2773                 RETURN(0);
2774
2775         /* FIXME: if we ever get into a situation where there are too many
2776          * opened files with open locks on a single node, then we really
2777          * should replay these open locks to reget it */
2778         if (lock->l_policy_data.l_inodebits.bits & MDS_INODELOCK_OPEN)
2779                 RETURN(0);
2780
2781         /* Special case for DoM locks, cancel only unused and granted locks */
2782         if (ldlm_has_dom(lock) &&
2783             (lock->l_granted_mode != lock->l_req_mode ||
2784              osc_ldlm_weigh_ast(lock) != 0))
2785                 RETURN(0);
2786
2787         RETURN(1);
2788 }
2789
2790 static int mdc_resource_inode_free(struct ldlm_resource *res)
2791 {
2792         if (res->lr_lvb_inode)
2793                 res->lr_lvb_inode = NULL;
2794
2795         return 0;
2796 }
2797
2798 static struct ldlm_valblock_ops inode_lvbo = {
2799         .lvbo_free = mdc_resource_inode_free
2800 };
2801
2802 static int mdc_llog_init(struct obd_device *obd)
2803 {
2804         struct obd_llog_group   *olg = &obd->obd_olg;
2805         struct llog_ctxt        *ctxt;
2806         int                      rc;
2807
2808         ENTRY;
2809
2810         rc = llog_setup(NULL, obd, olg, LLOG_CHANGELOG_REPL_CTXT, obd,
2811                         &llog_client_ops);
2812         if (rc < 0)
2813                 RETURN(rc);
2814
2815         ctxt = llog_group_get_ctxt(olg, LLOG_CHANGELOG_REPL_CTXT);
2816         llog_initiator_connect(ctxt);
2817         llog_ctxt_put(ctxt);
2818
2819         RETURN(0);
2820 }
2821
2822 static void mdc_llog_finish(struct obd_device *obd)
2823 {
2824         struct llog_ctxt *ctxt;
2825
2826         ENTRY;
2827
2828         ctxt = llog_get_context(obd, LLOG_CHANGELOG_REPL_CTXT);
2829         if (ctxt != NULL)
2830                 llog_cleanup(NULL, ctxt);
2831
2832         EXIT;
2833 }
2834
2835 int mdc_setup(struct obd_device *obd, struct lustre_cfg *cfg)
2836 {
2837         int rc;
2838
2839         ENTRY;
2840
2841         rc = osc_setup_common(obd, cfg);
2842         if (rc < 0)
2843                 RETURN(rc);
2844
2845         rc = mdc_tunables_init(obd);
2846         if (rc)
2847                 GOTO(err_osc_cleanup, rc);
2848
2849         obd->u.cli.cl_dom_min_inline_repsize = MDC_DOM_DEF_INLINE_REPSIZE;
2850
2851         ns_register_cancel(obd->obd_namespace, mdc_cancel_weight);
2852
2853         obd->obd_namespace->ns_lvbo = &inode_lvbo;
2854
2855         rc = mdc_llog_init(obd);
2856         if (rc) {
2857                 CERROR("%s: failed to setup llogging subsystems: rc = %d\n",
2858                        obd->obd_name, rc);
2859                 GOTO(err_llog_cleanup, rc);
2860         }
2861
2862         rc = mdc_changelog_cdev_init(obd);
2863         if (rc) {
2864                 CERROR("%s: failed to setup changelog char device: rc = %d\n",
2865                        obd->obd_name, rc);
2866                 GOTO(err_changelog_cleanup, rc);
2867         }
2868
2869         RETURN(rc);
2870
2871 err_changelog_cleanup:
2872         mdc_llog_finish(obd);
2873 err_llog_cleanup:
2874         lprocfs_free_md_stats(obd);
2875         ptlrpc_lprocfs_unregister_obd(obd);
2876 err_osc_cleanup:
2877         osc_cleanup_common(obd);
2878         return rc;
2879 }
2880
2881 /* Initialize the default and maximum LOV EA sizes.  This allows
2882  * us to make MDS RPCs with large enough reply buffers to hold a default
2883  * sized EA without having to calculate this (via a call into the
2884  * LOV + OSCs) each time we make an RPC.  The maximum size is also tracked
2885  * but not used to avoid wastefully vmalloc()'ing large reply buffers when
2886  * a large number of stripes is possible.  If a larger reply buffer is
2887  * required it will be reallocated in the ptlrpc layer due to overflow.
2888  */
2889 static int mdc_init_ea_size(struct obd_export *exp, __u32 easize,
2890                             __u32 def_easize)
2891 {
2892         struct obd_device *obd = exp->exp_obd;
2893         struct client_obd *cli = &obd->u.cli;
2894         ENTRY;
2895
2896         if (cli->cl_max_mds_easize < easize)
2897                 cli->cl_max_mds_easize = easize;
2898
2899         if (cli->cl_default_mds_easize < def_easize)
2900                 cli->cl_default_mds_easize = def_easize;
2901
2902         RETURN(0);
2903 }
2904
2905 static int mdc_precleanup(struct obd_device *obd)
2906 {
2907         ENTRY;
2908
2909         osc_precleanup_common(obd);
2910         mdc_changelog_cdev_finish(obd);
2911
2912         obd_cleanup_client_import(obd);
2913         ptlrpc_lprocfs_unregister_obd(obd);
2914         lprocfs_free_md_stats(obd);
2915         mdc_llog_finish(obd);
2916         RETURN(0);
2917 }
2918
2919 static int mdc_cleanup(struct obd_device *obd)
2920 {
2921         return osc_cleanup_common(obd);
2922 }
2923
2924 static const struct obd_ops mdc_obd_ops = {
2925         .o_owner            = THIS_MODULE,
2926         .o_setup            = mdc_setup,
2927         .o_precleanup       = mdc_precleanup,
2928         .o_cleanup          = mdc_cleanup,
2929         .o_add_conn         = client_import_add_conn,
2930         .o_del_conn         = client_import_del_conn,
2931         .o_connect          = client_connect_import,
2932         .o_reconnect        = osc_reconnect,
2933         .o_disconnect       = osc_disconnect,
2934         .o_iocontrol        = mdc_iocontrol,
2935         .o_set_info_async   = mdc_set_info_async,
2936         .o_statfs           = mdc_statfs,
2937         .o_statfs_async     = mdc_statfs_async,
2938         .o_fid_init         = client_fid_init,
2939         .o_fid_fini         = client_fid_fini,
2940         .o_fid_alloc        = mdc_fid_alloc,
2941         .o_import_event     = mdc_import_event,
2942         .o_get_info         = mdc_get_info,
2943         .o_get_uuid         = mdc_get_uuid,
2944         .o_quotactl         = mdc_quotactl,
2945 };
2946
2947 static const struct md_ops mdc_md_ops = {
2948         .m_get_root         = mdc_get_root,
2949         .m_null_inode       = mdc_null_inode,
2950         .m_close            = mdc_close,
2951         .m_create           = mdc_create,
2952         .m_enqueue          = mdc_enqueue,
2953         .m_getattr          = mdc_getattr,
2954         .m_getattr_name     = mdc_getattr_name,
2955         .m_intent_lock      = mdc_intent_lock,
2956         .m_link             = mdc_link,
2957         .m_rename           = mdc_rename,
2958         .m_setattr          = mdc_setattr,
2959         .m_setxattr         = mdc_setxattr,
2960         .m_getxattr         = mdc_getxattr,
2961         .m_fsync                = mdc_fsync,
2962         .m_file_resync          = mdc_file_resync,
2963         .m_read_page            = mdc_read_page,
2964         .m_unlink           = mdc_unlink,
2965         .m_cancel_unused    = mdc_cancel_unused,
2966         .m_init_ea_size     = mdc_init_ea_size,
2967         .m_set_lock_data    = mdc_set_lock_data,
2968         .m_lock_match       = mdc_lock_match,
2969         .m_get_lustre_md    = mdc_get_lustre_md,
2970         .m_free_lustre_md   = mdc_free_lustre_md,
2971         .m_set_open_replay_data = mdc_set_open_replay_data,
2972         .m_clear_open_replay_data = mdc_clear_open_replay_data,
2973         .m_intent_getattr_async = mdc_intent_getattr_async,
2974         .m_revalidate_lock      = mdc_revalidate_lock,
2975         .m_rmfid                = mdc_rmfid,
2976 };
2977
2978 dev_t mdc_changelog_dev;
2979 struct class *mdc_changelog_class;
2980 static int __init mdc_init(void)
2981 {
2982         int rc = 0;
2983         rc = alloc_chrdev_region(&mdc_changelog_dev, 0,
2984                                  MDC_CHANGELOG_DEV_COUNT,
2985                                  MDC_CHANGELOG_DEV_NAME);
2986         if (rc)
2987                 return rc;
2988
2989         mdc_changelog_class = class_create(THIS_MODULE, MDC_CHANGELOG_DEV_NAME);
2990         if (IS_ERR(mdc_changelog_class)) {
2991                 rc = PTR_ERR(mdc_changelog_class);
2992                 goto out_dev;
2993         }
2994
2995         rc = class_register_type(&mdc_obd_ops, &mdc_md_ops, true, NULL,
2996                                  LUSTRE_MDC_NAME, &mdc_device_type);
2997         if (rc)
2998                 goto out_dev;
2999
3000         return 0;
3001
3002 out_dev:
3003         unregister_chrdev_region(mdc_changelog_dev, MDC_CHANGELOG_DEV_COUNT);
3004         return rc;
3005 }
3006
3007 static void __exit mdc_exit(void)
3008 {
3009         class_destroy(mdc_changelog_class);
3010         unregister_chrdev_region(mdc_changelog_dev, MDC_CHANGELOG_DEV_COUNT);
3011         class_unregister_type(LUSTRE_MDC_NAME);
3012 }
3013
3014 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
3015 MODULE_DESCRIPTION("Lustre Metadata Client");
3016 MODULE_VERSION(LUSTRE_VERSION_STRING);
3017 MODULE_LICENSE("GPL");
3018
3019 module_init(mdc_init);
3020 module_exit(mdc_exit);