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  */
  31
  32 #define DEBUG_SUBSYSTEM S_MDC
  33
  34 #include <linux/init.h>
  35 #include <linux/kthread.h>
  36 #include <linux/module.h>
  37 #include <linux/pagemap.h>
  38 #include <linux/user_namespace.h>
  39 #include <linux/utsname.h>
  40 #include <linux/delay.h>
  41 #include <linux/uidgid.h>
  42 #include <linux/device.h>
  43 #include <linux/xarray.h>
  44
  45 #include <lustre_errno.h>
  46
  47 #include <cl_object.h>
  48 #include <llog_swab.h>
  49 #include <lprocfs_status.h>
  50 #include <lustre_acl.h>
  51 #include <lustre_compat.h>
  52 #include <lustre_fid.h>
  53 #include <uapi/linux/lustre/lustre_ioctl.h>
  54 #include <lustre_kernelcomm.h>
  55 #include <lustre_lmv.h>
  56 #include <lustre_log.h>
  57 #include <lustre_swab.h>
  58 #include <obd_class.h>
  59 #include <lustre_osc.h>
  60
  61 #include "mdc_internal.h"
  62
  63 #define REQUEST_MINOR 244
  64
  65 static int mdc_cleanup(struct obd_device *obd);
  66
  67 static inline int mdc_queue_wait(struct ptlrpc_request *req)
  68 {
  69         struct client_obd *cli = &req->rq_import->imp_obd->u.cli;
  70         int rc;
  71
  72         /* obd_get_request_slot() ensures that this client has no more
  73          * than cl_max_rpcs_in_flight RPCs simultaneously inf light
  74          * against an MDT. */
  75         rc = obd_get_request_slot(cli);
  76         if (rc != 0)
  77                 return rc;
  78
  79         rc = ptlrpc_queue_wait(req);
  80         obd_put_request_slot(cli);
  81
  82         return rc;
  83 }
  84
  85 /*
  86  * Send MDS_GET_ROOT RPC to fetch root FID.
  87  *
  88  * If \a fileset is not NULL it should contain a subdirectory off
  89  * the ROOT/ directory to be mounted on the client. Return the FID
  90  * of the subdirectory to the client to mount onto its mountpoint.
  91  *
  92  * \param[in]   imp     MDC import
  93  * \param[in]   fileset fileset name, which could be NULL
  94  * \param[out]  rootfid root FID of this mountpoint
  95  * \param[out]  pc      root capa will be unpacked and saved in this pointer
  96  *
  97  * \retval      0 on success, negative errno on failure
  98  */
  99 static int mdc_get_root(struct obd_export *exp, const char *fileset,
 100                          struct lu_fid *rootfid)
 101 {
 102         struct ptlrpc_request   *req;
 103         struct mdt_body         *body;
 104         int                      rc;
 105
 106         ENTRY;
 107
 108         if (fileset && !(exp_connect_flags(exp) & OBD_CONNECT_SUBTREE))
 109                 RETURN(-ENOTSUPP);
 110
 111         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
 112                                 &RQF_MDS_GET_ROOT);
 113         if (req == NULL)
 114                 RETURN(-ENOMEM);
 115
 116         if (fileset != NULL)
 117                 req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
 118                                      strlen(fileset) + 1);
 119         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GET_ROOT);
 120         if (rc) {
 121                 ptlrpc_request_free(req);
 122                 RETURN(rc);
 123         }
 124         mdc_pack_body(&req->rq_pill, NULL, 0, 0, -1, 0);
 125         if (fileset != NULL) {
 126                 char *name = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
 127
 128                 memcpy(name, fileset, strlen(fileset));
 129         }
 130         lustre_msg_add_flags(req->rq_reqmsg, LUSTRE_IMP_FULL);
 131         req->rq_send_state = LUSTRE_IMP_FULL;
 132
 133         ptlrpc_request_set_replen(req);
 134
 135         rc = ptlrpc_queue_wait(req);
 136         if (rc)
 137                 GOTO(out, rc);
 138
 139         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 140         if (body == NULL)
 141                 GOTO(out, rc = -EPROTO);
 142
 143         *rootfid = body->mbo_fid1;
 144         CDEBUG(D_NET, "root fid="DFID", last_committed=%llu\n",
 145                PFID(rootfid), lustre_msg_get_last_committed(req->rq_repmsg));
 146         EXIT;
 147 out:
 148         ptlrpc_req_finished(req);
 149
 150         return rc;
 151 }
 152
 153 /*
 154  * This function now is known to always saying that it will receive 4 buffers
 155  * from server. Even for cases when acl_size and md_size is zero, RPC header
 156  * will contain 4 fields and RPC itself will contain zero size fields. This is
 157  * because mdt_getattr*() _always_ returns 4 fields, but if acl is not needed
 158  * and thus zero, it shrinks it, making zero size. The same story about
 159  * md_size. And this is course of problem when client waits for smaller number
 160  * of fields. This issue will be fixed later when client gets aware of RPC
 161  * layouts.  --umka
 162  */
 163 static int mdc_getattr_common(struct obd_export *exp,
 164                               struct ptlrpc_request *req,
 165                               struct md_op_data *op_data)
 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         /* If encryption context was returned by MDT, put it in op_data
 194          * so that caller can set it on inode and save an extra getxattr.
 195          */
 196         if (op_data && op_data->op_valid & OBD_MD_ENCCTX &&
 197             body->mbo_valid & OBD_MD_ENCCTX) {
 198                 op_data->op_file_encctx =
 199                         req_capsule_server_get(pill, &RMF_FILE_ENCCTX);
 200                 op_data->op_file_encctx_size =
 201                         req_capsule_get_size(pill, &RMF_FILE_ENCCTX,
 202                                              RCL_SERVER);
 203         }
 204
 205         RETURN(0);
 206 }
 207
 208 static void mdc_reset_acl_req(struct ptlrpc_request *req)
 209 {
 210         spin_lock(&req->rq_early_free_lock);
 211         sptlrpc_cli_free_repbuf(req);
 212         req->rq_repbuf = NULL;
 213         req->rq_repbuf_len = 0;
 214         req->rq_repdata = NULL;
 215         req->rq_reqdata_len = 0;
 216         spin_unlock(&req->rq_early_free_lock);
 217 }
 218
 219 static int mdc_getattr(struct obd_export *exp, struct md_op_data *op_data,
 220                        struct ptlrpc_request **request)
 221 {
 222         struct ptlrpc_request *req;
 223         struct obd_device *obd = class_exp2obd(exp);
 224         struct obd_import *imp = class_exp2cliimp(exp);
 225         __u32 acl_bufsize = LUSTRE_POSIX_ACL_MAX_SIZE_OLD;
 226         int rc;
 227         ENTRY;
 228
 229         /* Single MDS without an LMV case */
 230         if (op_data->op_flags & MF_GET_MDT_IDX) {
 231                 op_data->op_mds = 0;
 232                 RETURN(0);
 233         }
 234
 235         *request = NULL;
 236         req = ptlrpc_request_alloc(imp, &RQF_MDS_GETATTR);
 237         if (req == NULL)
 238                 RETURN(-ENOMEM);
 239
 240         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR);
 241         if (rc) {
 242                 ptlrpc_request_free(req);
 243                 RETURN(rc);
 244         }
 245
 246         /* LU-15245: avoid deadlock with modifying RPCs on MDS_REQUEST_PORTAL */
 247         req->rq_request_portal = MDS_READPAGE_PORTAL;
 248
 249 again:
 250         mdc_pack_body(&req->rq_pill, &op_data->op_fid1, op_data->op_valid,
 251                       op_data->op_mode, -1, 0);
 252         req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER, acl_bufsize);
 253         req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
 254                              op_data->op_mode);
 255         if (exp_connect_encrypt(exp) && op_data->op_valid & OBD_MD_ENCCTX)
 256                 req_capsule_set_size(&req->rq_pill, &RMF_FILE_ENCCTX,
 257                                      RCL_SERVER,
 258                                      obd->u.cli.cl_max_mds_easize);
 259         else
 260                 req_capsule_set_size(&req->rq_pill, &RMF_FILE_ENCCTX,
 261                                      RCL_SERVER, 0);
 262         ptlrpc_request_set_replen(req);
 263
 264         rc = mdc_getattr_common(exp, req, op_data);
 265         if (rc) {
 266                 if (rc == -ERANGE) {
 267                         acl_bufsize = min_t(__u32,
 268                                             imp->imp_connect_data.ocd_max_easize,
 269                                             XATTR_SIZE_MAX);
 270                         mdc_reset_acl_req(req);
 271                         goto again;
 272                 }
 273
 274                 ptlrpc_req_finished(req);
 275         } else {
 276                 *request = req;
 277         }
 278
 279         RETURN(rc);
 280 }
 281
 282 static int mdc_getattr_name(struct obd_export *exp, struct md_op_data *op_data,
 283                             struct ptlrpc_request **request)
 284 {
 285         struct ptlrpc_request *req;
 286         struct obd_import *imp = class_exp2cliimp(exp);
 287         __u32 acl_bufsize = LUSTRE_POSIX_ACL_MAX_SIZE_OLD;
 288         int rc;
 289         ENTRY;
 290
 291         *request = NULL;
 292         req = ptlrpc_request_alloc(imp, &RQF_MDS_GETATTR_NAME);
 293         if (req == NULL)
 294                 RETURN(-ENOMEM);
 295
 296         req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
 297                              op_data->op_namelen + 1);
 298
 299         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR_NAME);
 300         if (rc) {
 301                 ptlrpc_request_free(req);
 302                 RETURN(rc);
 303         }
 304
 305         if (op_data->op_name) {
 306                 char *name = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
 307                 LASSERT(strnlen(op_data->op_name, op_data->op_namelen) ==
 308                                 op_data->op_namelen);
 309                 memcpy(name, op_data->op_name, op_data->op_namelen);
 310         }
 311
 312 again:
 313         mdc_pack_body(&req->rq_pill, &op_data->op_fid1, op_data->op_valid,
 314                       op_data->op_mode, op_data->op_suppgids[0], 0);
 315         req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
 316                              op_data->op_mode);
 317         req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER, acl_bufsize);
 318         req_capsule_set_size(&req->rq_pill, &RMF_FILE_ENCCTX, RCL_SERVER, 0);
 319         ptlrpc_request_set_replen(req);
 320         if (op_data->op_bias & MDS_FID_OP) {
 321                 struct mdt_body *b = req_capsule_client_get(&req->rq_pill,
 322                                                             &RMF_MDT_BODY);
 323
 324                 if (b) {
 325                         b->mbo_valid |= OBD_MD_NAMEHASH;
 326                         b->mbo_fid2 = op_data->op_fid2;
 327                 }
 328         }
 329
 330         rc = mdc_getattr_common(exp, req, NULL);
 331         if (rc) {
 332                 if (rc == -ERANGE) {
 333                         acl_bufsize = min_t(__u32,
 334                                             imp->imp_connect_data.ocd_max_easize,
 335                                             XATTR_SIZE_MAX);
 336                         mdc_reset_acl_req(req);
 337                         goto again;
 338                 }
 339
 340                 ptlrpc_req_finished(req);
 341         } else {
 342                 *request = req;
 343         }
 344
 345         RETURN(rc);
 346 }
 347
 348 static int mdc_xattr_common(struct obd_export *exp,const struct req_format *fmt,
 349                             const struct lu_fid *fid, int opcode, u64 valid,
 350                             const char *xattr_name, const char *input,
 351                             int input_size, int output_size, int flags,
 352                             __u32 suppgid, struct ptlrpc_request **request)
 353 {
 354         struct ptlrpc_request *req;
 355         int   xattr_namelen = 0;
 356         char *tmp;
 357         int   rc;
 358         ENTRY;
 359
 360         *request = NULL;
 361         req = ptlrpc_request_alloc(class_exp2cliimp(exp), fmt);
 362         if (req == NULL)
 363                 RETURN(-ENOMEM);
 364
 365         if (xattr_name) {
 366                 xattr_namelen = strlen(xattr_name) + 1;
 367                 req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
 368                                      xattr_namelen);
 369         }
 370         if (input_size)
 371                 LASSERT(input);
 372         req_capsule_set_size(&req->rq_pill, &RMF_EADATA, RCL_CLIENT,
 373                              input_size);
 374
 375         /* get SELinux policy info if any */
 376         rc = sptlrpc_get_sepol(req);
 377         if (rc < 0) {
 378                 ptlrpc_request_free(req);
 379                 RETURN(rc);
 380         }
 381         req_capsule_set_size(&req->rq_pill, &RMF_SELINUX_POL, RCL_CLIENT,
 382                              strlen(req->rq_sepol) ?
 383                              strlen(req->rq_sepol) + 1 : 0);
 384
 385         /* Flush local XATTR locks to get rid of a possible cancel RPC */
 386         if (opcode == MDS_REINT && fid_is_sane(fid) &&
 387             exp->exp_connect_data.ocd_ibits_known & MDS_INODELOCK_XATTR) {
 388                 LIST_HEAD(cancels);
 389                 int count;
 390
 391                 /* Without that packing would fail */
 392                 if (input_size == 0)
 393                         req_capsule_set_size(&req->rq_pill, &RMF_EADATA,
 394                                              RCL_CLIENT, 0);
 395
 396                 count = mdc_resource_get_unused(exp, fid,
 397                                                 &cancels, LCK_EX,
 398                                                 MDS_INODELOCK_XATTR);
 399
 400                 rc = mdc_prep_elc_req(exp, req, MDS_REINT, &cancels, count);
 401                 if (rc) {
 402                         ptlrpc_request_free(req);
 403                         RETURN(rc);
 404                 }
 405         } else {
 406                 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, opcode);
 407                 if (rc) {
 408                         ptlrpc_request_free(req);
 409                         RETURN(rc);
 410                 }
 411         }
 412
 413         if (opcode == MDS_REINT) {
 414                 struct mdt_rec_setxattr *rec;
 415
 416                 BUILD_BUG_ON(sizeof(struct mdt_rec_setxattr) !=
 417                              sizeof(struct mdt_rec_reint));
 418                 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
 419                 rec->sx_opcode = REINT_SETXATTR;
 420                 rec->sx_fsuid  = from_kuid(&init_user_ns, current_fsuid());
 421                 rec->sx_fsgid  = from_kgid(&init_user_ns, current_fsgid());
 422                 rec->sx_cap = current_cap().cap[0];
 423                 rec->sx_suppgid1 = suppgid;
 424                 rec->sx_suppgid2 = -1;
 425                 rec->sx_fid    = *fid;
 426                 rec->sx_valid  = valid | OBD_MD_FLCTIME;
 427                 rec->sx_time   = ktime_get_real_seconds();
 428                 rec->sx_size   = output_size;
 429                 rec->sx_flags  = flags;
 430         } else {
 431                 mdc_pack_body(&req->rq_pill, fid, valid, output_size,
 432                               suppgid, flags);
 433                 /* Avoid deadlock with modifying RPCs on MDS_REQUEST_PORTAL.
 434                  * See LU-15245.
 435                  */
 436                 req->rq_request_portal = MDS_READPAGE_PORTAL;
 437         }
 438
 439         if (xattr_name) {
 440                 tmp = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
 441                 memcpy(tmp, xattr_name, xattr_namelen);
 442         }
 443         if (input_size) {
 444                 tmp = req_capsule_client_get(&req->rq_pill, &RMF_EADATA);
 445                 memcpy(tmp, input, input_size);
 446         }
 447
 448         mdc_file_sepol_pack(&req->rq_pill);
 449
 450         if (req_capsule_has_field(&req->rq_pill, &RMF_EADATA, RCL_SERVER))
 451                 req_capsule_set_size(&req->rq_pill, &RMF_EADATA,
 452                                      RCL_SERVER, output_size);
 453         ptlrpc_request_set_replen(req);
 454
 455         /* make rpc */
 456         if (opcode == MDS_REINT)
 457                 ptlrpc_get_mod_rpc_slot(req);
 458
 459         rc = ptlrpc_queue_wait(req);
 460
 461         if (opcode == MDS_REINT)
 462                 ptlrpc_put_mod_rpc_slot(req);
 463
 464         if (rc)
 465                 ptlrpc_req_finished(req);
 466         else
 467                 *request = req;
 468         RETURN(rc);
 469 }
 470
 471 static int mdc_setxattr(struct obd_export *exp, const struct lu_fid *fid,
 472                         u64 obd_md_valid, const char *name,
 473                         const void *value, size_t value_size,
 474                         unsigned int xattr_flags, u32 suppgid,
 475                         struct ptlrpc_request **req)
 476 {
 477         LASSERT(obd_md_valid == OBD_MD_FLXATTR ||
 478                 obd_md_valid == OBD_MD_FLXATTRRM);
 479
 480         return mdc_xattr_common(exp, &RQF_MDS_REINT_SETXATTR,
 481                                 fid, MDS_REINT, obd_md_valid, name,
 482                                 value, value_size, 0, xattr_flags, suppgid,
 483                                 req);
 484 }
 485
 486 static int mdc_getxattr(struct obd_export *exp, const struct lu_fid *fid,
 487                         u64 obd_md_valid, const char *name, size_t buf_size,
 488                         struct ptlrpc_request **req)
 489 {
 490         struct mdt_body *body;
 491         int rc;
 492
 493         LASSERT(obd_md_valid == OBD_MD_FLXATTR ||
 494                 obd_md_valid == OBD_MD_FLXATTRLS);
 495
 496         /* Message below is checked in sanity-selinux test_20d
 497          * and sanity-sec test_49
 498          */
 499         CDEBUG(D_INFO, "%s: get xattr '%s' for "DFID"\n",
 500                exp->exp_obd->obd_name, name, PFID(fid));
 501         rc = mdc_xattr_common(exp, &RQF_MDS_GETXATTR, fid, MDS_GETXATTR,
 502                               obd_md_valid, name, NULL, 0, buf_size, 0, -1,
 503                               req);
 504         if (rc < 0)
 505                 GOTO(out, rc);
 506
 507         body = req_capsule_server_get(&(*req)->rq_pill, &RMF_MDT_BODY);
 508         if (body == NULL)
 509                 GOTO(out, rc = -EPROTO);
 510
 511         /* only detect the xattr size */
 512         if (buf_size == 0) {
 513                 /* LU-11109: Older MDTs do not distinguish
 514                  * between nonexistent xattrs and zero length
 515                  * values in this case. Newer MDTs will return
 516                  * -ENODATA or set OBD_MD_FLXATTR. */
 517                 GOTO(out, rc = body->mbo_eadatasize);
 518         }
 519
 520         if (body->mbo_eadatasize == 0) {
 521                 /* LU-11109: Newer MDTs set OBD_MD_FLXATTR on
 522                  * success so that we can distinguish between
 523                  * zero length value and nonexistent xattr.
 524                  *
 525                  * If OBD_MD_FLXATTR is not set then we keep
 526                  * the old behavior and return -ENODATA for
 527                  * getxattr() when mbo_eadatasize is 0. But
 528                  * -ENODATA only makes sense for getxattr()
 529                  * and not for listxattr(). */
 530                 if (body->mbo_valid & OBD_MD_FLXATTR)
 531                         GOTO(out, rc = 0);
 532                 else if (obd_md_valid == OBD_MD_FLXATTR)
 533                         GOTO(out, rc = -ENODATA);
 534                 else
 535                         GOTO(out, rc = 0);
 536         }
 537
 538         GOTO(out, rc = body->mbo_eadatasize);
 539 out:
 540         if (rc < 0) {
 541                 ptlrpc_req_finished(*req);
 542                 *req = NULL;
 543         }
 544
 545         return rc;
 546 }
 547
 548 static int mdc_get_lustre_md(struct obd_export *exp, struct req_capsule *pill,
 549                              struct obd_export *dt_exp,
 550                              struct obd_export *md_exp,
 551                              struct lustre_md *md)
 552 {
 553         int rc;
 554         ENTRY;
 555
 556         LASSERT(md);
 557         memset(md, 0, sizeof(*md));
 558
 559         md->body = req_capsule_server_get(pill, &RMF_MDT_BODY);
 560         LASSERT(md->body != NULL);
 561
 562         if (md->body->mbo_valid & OBD_MD_FLEASIZE) {
 563                 if (!S_ISREG(md->body->mbo_mode)) {
 564                         CDEBUG(D_INFO, "OBD_MD_FLEASIZE set, should be a "
 565                                "regular file, but is not\n");
 566                         GOTO(out, rc = -EPROTO);
 567                 }
 568
 569                 if (md->body->mbo_eadatasize == 0) {
 570                         CDEBUG(D_INFO, "OBD_MD_FLEASIZE set, "
 571                                "but eadatasize 0\n");
 572                         GOTO(out, rc = -EPROTO);
 573                 }
 574
 575                 md->layout.lb_len = md->body->mbo_eadatasize;
 576                 md->layout.lb_buf = req_capsule_server_sized_get(pill,
 577                                                         &RMF_MDT_MD,
 578                                                         md->layout.lb_len);
 579                 if (md->layout.lb_buf == NULL)
 580                         GOTO(out, rc = -EPROTO);
 581         } else if (md->body->mbo_valid & OBD_MD_FLDIREA) {
 582                 const union lmv_mds_md *lmv;
 583                 size_t lmv_size;
 584
 585                 if (!S_ISDIR(md->body->mbo_mode)) {
 586                         CDEBUG(D_INFO, "OBD_MD_FLDIREA set, should be a "
 587                                "directory, but is not\n");
 588                         GOTO(out, rc = -EPROTO);
 589                 }
 590
 591                 if (md_exp->exp_obd->obd_type->typ_lu == &mdc_device_type) {
 592                         CERROR("%s: no LMV, upgrading from old version?\n",
 593                                md_exp->exp_obd->obd_name);
 594
 595                         GOTO(out_acl, rc = 0);
 596                 }
 597
 598                 if (md->body->mbo_valid & OBD_MD_MEA) {
 599                         lmv_size = md->body->mbo_eadatasize;
 600                         if (lmv_size == 0) {
 601                                 CDEBUG(D_INFO, "OBD_MD_FLDIREA is set, "
 602                                        "but eadatasize 0\n");
 603                                 RETURN(-EPROTO);
 604                         }
 605
 606                         lmv = req_capsule_server_sized_get(pill, &RMF_MDT_MD,
 607                                                            lmv_size);
 608                         if (lmv == NULL)
 609                                 GOTO(out, rc = -EPROTO);
 610
 611                         rc = md_unpackmd(md_exp, &md->lmv, lmv, lmv_size);
 612                         if (rc < 0)
 613                                 GOTO(out, rc);
 614
 615                         if (rc < (int)sizeof(*md->lmv)) {
 616                                 struct lmv_foreign_md *lfm = md->lfm;
 617
 618                                 /* short (< sizeof(struct lmv_stripe_md))
 619                                  * foreign LMV case
 620                                  */
 621                                 if (lfm->lfm_magic != LMV_MAGIC_FOREIGN) {
 622                                         CDEBUG(D_INFO,
 623                                                "lmv size too small: %d < %d\n",
 624                                                rc, (int)sizeof(*md->lmv));
 625                                         GOTO(out, rc = -EPROTO);
 626                                 }
 627                         }
 628                 }
 629
 630                 /* since 2.12.58 intent_getattr fetches default LMV */
 631                 if (md->body->mbo_valid & OBD_MD_DEFAULT_MEA) {
 632                         lmv_size = sizeof(struct lmv_user_md);
 633                         lmv = req_capsule_server_sized_get(pill,
 634                                                            &RMF_DEFAULT_MDT_MD,
 635                                                            lmv_size);
 636                         if (!lmv)
 637                                 GOTO(out, rc = -EPROTO);
 638
 639                         rc = md_unpackmd(md_exp, &md->default_lmv, lmv,
 640                                          lmv_size);
 641                         if (rc < 0)
 642                                 GOTO(out, rc);
 643
 644                         if (rc < (int)sizeof(*md->default_lmv)) {
 645                                 CDEBUG(D_INFO,
 646                                        "default lmv size too small: %d < %d\n",
 647                                         rc, (int)sizeof(*md->default_lmv));
 648                                 GOTO(out, rc = -EPROTO);
 649                         }
 650                 }
 651         }
 652         rc = 0;
 653
 654 out_acl:
 655         if (md->body->mbo_valid & OBD_MD_FLACL) {
 656                 /* for ACL, it's possible that FLACL is set but aclsize is zero.
 657                  * only when aclsize != 0 there's an actual segment for ACL
 658                  * in reply buffer.
 659                  */
 660                 rc = mdc_unpack_acl(pill, md);
 661                 if (rc)
 662                         GOTO(out, rc);
 663         }
 664
 665         EXIT;
 666 out:
 667         if (rc)
 668                 lmd_clear_acl(md);
 669
 670         return rc;
 671 }
 672
 673 static int mdc_free_lustre_md(struct obd_export *exp, struct lustre_md *md)
 674 {
 675         ENTRY;
 676         RETURN(0);
 677 }
 678
 679 void mdc_replay_open(struct ptlrpc_request *req)
 680 {
 681         struct md_open_data *mod = req->rq_cb_data;
 682         struct ptlrpc_request *close_req;
 683         struct obd_client_handle *och;
 684         struct lustre_handle old_open_handle = { };
 685         struct mdt_body *body;
 686         struct ldlm_reply *rep;
 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         rep = req_capsule_server_get(&req->rq_pill, &RMF_DLM_REP);
 700         if (rep != NULL && rep->lock_policy_res2 != 0)
 701                 DEBUG_REQ(D_ERROR, req, "Open request replay failed with %ld ",
 702                         (long int)rep->lock_policy_res2);
 703
 704         spin_lock(&req->rq_lock);
 705         och = mod->mod_och;
 706         if (och && och->och_open_handle.cookie)
 707                 req->rq_early_free_repbuf = 1;
 708         else
 709                 req->rq_early_free_repbuf = 0;
 710         spin_unlock(&req->rq_lock);
 711
 712         if (req->rq_early_free_repbuf) {
 713                 struct lustre_handle *file_open_handle;
 714
 715                 LASSERT(och->och_magic == OBD_CLIENT_HANDLE_MAGIC);
 716
 717                 file_open_handle = &och->och_open_handle;
 718                 CDEBUG(D_HA, "updating handle from %#llx to %#llx\n",
 719                        file_open_handle->cookie, body->mbo_open_handle.cookie);
 720                 old_open_handle = *file_open_handle;
 721                 *file_open_handle = body->mbo_open_handle;
 722         }
 723
 724         close_req = mod->mod_close_req;
 725         if (close_req) {
 726                 __u32 opc = lustre_msg_get_opc(close_req->rq_reqmsg);
 727                 struct mdt_ioepoch *epoch;
 728
 729                 LASSERT(opc == MDS_CLOSE);
 730                 epoch = req_capsule_client_get(&close_req->rq_pill,
 731                                                &RMF_MDT_EPOCH);
 732                 LASSERT(epoch);
 733
 734                 if (req->rq_early_free_repbuf)
 735                         LASSERT(old_open_handle.cookie ==
 736                                 epoch->mio_open_handle.cookie);
 737
 738                 DEBUG_REQ(D_HA, close_req, "updating close body with new fh");
 739                 epoch->mio_open_handle = body->mbo_open_handle;
 740         }
 741         EXIT;
 742 }
 743
 744 void mdc_commit_open(struct ptlrpc_request *req)
 745 {
 746         struct md_open_data *mod = req->rq_cb_data;
 747         if (mod == NULL)
 748                 return;
 749
 750         /**
 751          * No need to touch md_open_data::mod_och, it holds a reference on
 752          * \var mod and will zero references to each other, \var mod will be
 753          * freed after that when md_open_data::mod_och will put the reference.
 754          */
 755
 756         /**
 757          * Do not let open request to disappear as it still may be needed
 758          * for close rpc to happen (it may happen on evict only, otherwise
 759          * ptlrpc_request::rq_replay does not let mdc_commit_open() to be
 760          * called), just mark this rpc as committed to distinguish these 2
 761          * cases, see mdc_close() for details. The open request reference will
 762          * be put along with freeing \var mod.
 763          */
 764         ptlrpc_request_addref(req);
 765         spin_lock(&req->rq_lock);
 766         req->rq_committed = 1;
 767         spin_unlock(&req->rq_lock);
 768         req->rq_cb_data = NULL;
 769         obd_mod_put(mod);
 770 }
 771
 772 int mdc_set_open_replay_data(struct obd_export *exp,
 773                              struct obd_client_handle *och,
 774                              struct lookup_intent *it)
 775 {
 776         struct md_open_data     *mod;
 777         struct mdt_rec_create   *rec;
 778         struct mdt_body         *body;
 779         struct ptlrpc_request   *open_req = it->it_request;
 780         struct obd_import       *imp = open_req->rq_import;
 781         ENTRY;
 782
 783         if (!open_req->rq_replay)
 784                 RETURN(0);
 785
 786         rec = req_capsule_client_get(&open_req->rq_pill, &RMF_REC_REINT);
 787         body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
 788         LASSERT(rec != NULL);
 789         /* Incoming message in my byte order (it's been swabbed). */
 790         /* Outgoing messages always in my byte order. */
 791         LASSERT(body != NULL);
 792
 793         /* Only if the import is replayable, we set replay_open data */
 794         if (och && imp->imp_replayable) {
 795                 mod = obd_mod_alloc();
 796                 if (mod == NULL) {
 797                         DEBUG_REQ(D_ERROR, open_req,
 798                                   "cannot allocate md_open_data");
 799                         RETURN(0);
 800                 }
 801
 802                 /**
 803                  * Take a reference on \var mod, to be freed on mdc_close().
 804                  * It protects \var mod from being freed on eviction (commit
 805                  * callback is called despite rq_replay flag).
 806                  * Another reference for \var och.
 807                  */
 808                 obd_mod_get(mod);
 809                 obd_mod_get(mod);
 810
 811                 spin_lock(&open_req->rq_lock);
 812                 och->och_mod = mod;
 813                 mod->mod_och = och;
 814                 mod->mod_is_create = it_disposition(it, DISP_OPEN_CREATE) ||
 815                                      it_disposition(it, DISP_OPEN_STRIPE);
 816                 mod->mod_open_req = open_req;
 817                 open_req->rq_cb_data = mod;
 818                 open_req->rq_commit_cb = mdc_commit_open;
 819                 open_req->rq_early_free_repbuf = 1;
 820                 spin_unlock(&open_req->rq_lock);
 821         }
 822
 823         rec->cr_fid2 = body->mbo_fid1;
 824         rec->cr_open_handle_old = body->mbo_open_handle;
 825         open_req->rq_replay_cb = mdc_replay_open;
 826         if (!fid_is_sane(&body->mbo_fid1)) {
 827                 DEBUG_REQ(D_ERROR, open_req,
 828                           "saving replay request with insane FID " DFID,
 829                           PFID(&body->mbo_fid1));
 830                 LBUG();
 831         }
 832
 833         DEBUG_REQ(D_RPCTRACE, open_req, "Set up open replay data");
 834         RETURN(0);
 835 }
 836
 837 static void mdc_free_open(struct md_open_data *mod)
 838 {
 839         int committed = 0;
 840
 841         if (mod->mod_is_create == 0 &&
 842             imp_connect_disp_stripe(mod->mod_open_req->rq_import))
 843                 committed = 1;
 844
 845         /**
 846          * No reason to asssert here if the open request has
 847          * rq_replay == 1. It means that mdc_close failed, and
 848          * close request wasn`t sent. It is not fatal to client.
 849          * The worst thing is eviction if the client gets open lock
 850          **/
 851
 852         DEBUG_REQ(D_RPCTRACE, mod->mod_open_req,
 853                   "free open request, rq_replay=%d",
 854                   mod->mod_open_req->rq_replay);
 855
 856         ptlrpc_request_committed(mod->mod_open_req, committed);
 857         if (mod->mod_close_req)
 858                 ptlrpc_request_committed(mod->mod_close_req, committed);
 859 }
 860
 861 static int mdc_clear_open_replay_data(struct obd_export *exp,
 862                                       struct obd_client_handle *och)
 863 {
 864         struct md_open_data *mod = och->och_mod;
 865         ENTRY;
 866
 867         /**
 868          * It is possible to not have \var mod in a case of eviction between
 869          * lookup and ll_file_open().
 870          **/
 871         if (mod == NULL)
 872                 RETURN(0);
 873
 874         LASSERT(mod != LP_POISON);
 875         LASSERT(mod->mod_open_req != NULL);
 876
 877         spin_lock(&mod->mod_open_req->rq_lock);
 878         if (mod->mod_och)
 879                 mod->mod_och->och_open_handle.cookie = 0;
 880         mod->mod_open_req->rq_early_free_repbuf = 0;
 881         spin_unlock(&mod->mod_open_req->rq_lock);
 882         mdc_free_open(mod);
 883
 884         mod->mod_och = NULL;
 885         och->och_mod = NULL;
 886         obd_mod_put(mod);
 887
 888         RETURN(0);
 889 }
 890
 891 static int mdc_close(struct obd_export *exp, struct md_op_data *op_data,
 892                      struct md_open_data *mod, struct ptlrpc_request **request)
 893 {
 894         struct obd_device     *obd = class_exp2obd(exp);
 895         struct ptlrpc_request *req;
 896         struct req_format     *req_fmt;
 897         size_t                 u32_count = 0;
 898         int                    rc;
 899         int                    saved_rc = 0;
 900         ENTRY;
 901
 902         CDEBUG(D_INODE, "%s: "DFID" file closed with intent: %x\n",
 903                exp->exp_obd->obd_name, PFID(&op_data->op_fid1),
 904                op_data->op_bias);
 905
 906         if (op_data->op_bias & MDS_CLOSE_INTENT) {
 907                 req_fmt = &RQF_MDS_CLOSE_INTENT;
 908                 if (op_data->op_bias & MDS_HSM_RELEASE) {
 909                         /* allocate a FID for volatile file */
 910                         rc = mdc_fid_alloc(NULL, exp, &op_data->op_fid2,
 911                                            op_data);
 912                         if (rc < 0) {
 913                                 CERROR("%s: "DFID" allocating FID: rc = %d\n",
 914                                        obd->obd_name, PFID(&op_data->op_fid1),
 915                                        rc);
 916                                 /* save the errcode and proceed to close */
 917                                 saved_rc = rc;
 918                         }
 919                 }
 920                 if (op_data->op_bias & MDS_CLOSE_RESYNC_DONE) {
 921                         size_t count = op_data->op_data_size / sizeof(__u32);
 922
 923                         if (count > INLINE_RESYNC_ARRAY_SIZE)
 924                                 u32_count = count;
 925                 }
 926         } else {
 927                 req_fmt = &RQF_MDS_CLOSE;
 928         }
 929
 930         *request = NULL;
 931         if (OBD_FAIL_CHECK(OBD_FAIL_MDC_CLOSE))
 932                 req = NULL;
 933         else
 934                 req = ptlrpc_request_alloc(class_exp2cliimp(exp), req_fmt);
 935
 936         /* Ensure that this close's handle is fixed up during replay. */
 937         if (likely(mod != NULL)) {
 938                 LASSERTF(mod->mod_open_req != NULL &&
 939                          mod->mod_open_req->rq_type != LI_POISON,
 940                          "POISONED open %p!\n", mod->mod_open_req);
 941
 942                 mod->mod_close_req = req;
 943
 944                 DEBUG_REQ(D_RPCTRACE, mod->mod_open_req, "matched open");
 945                 /* We no longer want to preserve this open for replay even
 946                  * though the open was committed. b=3632, b=3633 */
 947                 spin_lock(&mod->mod_open_req->rq_lock);
 948                 mod->mod_open_req->rq_replay = 0;
 949                 spin_unlock(&mod->mod_open_req->rq_lock);
 950         } else {
 951                 CDEBUG(D_HA, "couldn't find open req; expecting close error\n");
 952         }
 953         if (req == NULL) {
 954                 /**
 955                  * TODO: repeat close after errors
 956                  */
 957                 CWARN("%s: close of FID "DFID" failed, file reference will be "
 958                       "dropped when this client unmounts or is evicted\n",
 959                       obd->obd_name, PFID(&op_data->op_fid1));
 960                 GOTO(out, rc = -ENOMEM);
 961         }
 962
 963         if (u32_count > 0)
 964                 req_capsule_set_size(&req->rq_pill, &RMF_U32, RCL_CLIENT,
 965                                      u32_count * sizeof(__u32));
 966
 967         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_CLOSE);
 968         if (rc) {
 969                 ptlrpc_request_free(req);
 970                 req = NULL;
 971                 GOTO(out, rc);
 972         }
 973
 974         /* To avoid a livelock (bug 7034), we need to send CLOSE RPCs to a
 975          * portal whose threads are not taking any DLM locks and are therefore
 976          * always progressing */
 977         req->rq_request_portal = MDS_READPAGE_PORTAL;
 978         ptlrpc_at_set_req_timeout(req);
 979
 980         if (!obd->u.cli.cl_lsom_update ||
 981             !(exp_connect_flags2(exp) & OBD_CONNECT2_LSOM))
 982                 op_data->op_xvalid &= ~(OP_XVALID_LAZYSIZE |
 983                                         OP_XVALID_LAZYBLOCKS);
 984
 985         mdc_close_pack(&req->rq_pill, op_data);
 986
 987         req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
 988                              obd->u.cli.cl_default_mds_easize);
 989
 990         ptlrpc_request_set_replen(req);
 991
 992         ptlrpc_get_mod_rpc_slot(req);
 993         rc = ptlrpc_queue_wait(req);
 994         ptlrpc_put_mod_rpc_slot(req);
 995
 996         if (req->rq_repmsg == NULL) {
 997                 CDEBUG(D_RPCTRACE, "request %p failed to send: rc = %d\n", req,
 998                        req->rq_status);
 999                 if (rc == 0)
1000                         rc = req->rq_status ?: -EIO;
1001         } else if (rc == 0 || rc == -EAGAIN) {
1002                 struct mdt_body *body;
1003
1004                 rc = lustre_msg_get_status(req->rq_repmsg);
1005                 if (lustre_msg_get_type(req->rq_repmsg) == PTL_RPC_MSG_ERR) {
1006                         DEBUG_REQ(D_ERROR, req,
1007                                   "type = PTL_RPC_MSG_ERR: rc = %d", rc);
1008                         if (rc > 0)
1009                                 rc = -rc;
1010                 }
1011                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1012                 if (body == NULL)
1013                         rc = -EPROTO;
1014         } else if (rc == -ESTALE) {
1015                 /**
1016                  * it can be allowed error after 3633 if open was committed and
1017                  * server failed before close was sent. Let's check if mod
1018                  * exists and return no error in that case
1019                  */
1020                 if (mod) {
1021                         DEBUG_REQ(D_HA, req, "Reset ESTALE = %d", rc);
1022                         LASSERT(mod->mod_open_req != NULL);
1023                         if (mod->mod_open_req->rq_committed)
1024                                 rc = 0;
1025                 }
1026         }
1027
1028 out:
1029         if (mod) {
1030                 if (rc != 0)
1031                         mod->mod_close_req = NULL;
1032                 /* Since now, mod is accessed through open_req only,
1033                  * thus close req does not keep a reference on mod anymore. */
1034                 obd_mod_put(mod);
1035         }
1036         *request = req;
1037
1038         RETURN(rc < 0 ? rc : saved_rc);
1039 }
1040
1041 static int mdc_getpage(struct obd_export *exp, const struct lu_fid *fid,
1042                        u64 offset, struct page **pages, int npages,
1043                        struct ptlrpc_request **request)
1044 {
1045         struct ptlrpc_request   *req;
1046         struct ptlrpc_bulk_desc *desc;
1047         int                      i;
1048         int                      resends = 0;
1049         int                      rc;
1050         ENTRY;
1051
1052         *request = NULL;
1053
1054 restart_bulk:
1055         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_READPAGE);
1056         if (req == NULL)
1057                 RETURN(-ENOMEM);
1058
1059         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_READPAGE);
1060         if (rc) {
1061                 ptlrpc_request_free(req);
1062                 RETURN(rc);
1063         }
1064
1065         req->rq_request_portal = MDS_READPAGE_PORTAL;
1066         ptlrpc_at_set_req_timeout(req);
1067
1068         desc = ptlrpc_prep_bulk_imp(req, npages, 1,
1069                                     PTLRPC_BULK_PUT_SINK,
1070                                     MDS_BULK_PORTAL,
1071                                     &ptlrpc_bulk_kiov_pin_ops);
1072         if (desc == NULL) {
1073                 ptlrpc_req_finished(req);
1074                 RETURN(-ENOMEM);
1075         }
1076
1077         /* NB req now owns desc and will free it when it gets freed */
1078         for (i = 0; i < npages; i++)
1079                 desc->bd_frag_ops->add_kiov_frag(desc, pages[i], 0,
1080                                                  PAGE_SIZE);
1081
1082         mdc_readdir_pack(&req->rq_pill, offset, PAGE_SIZE * npages, fid);
1083
1084         ptlrpc_request_set_replen(req);
1085         rc = ptlrpc_queue_wait(req);
1086         if (rc) {
1087                 ptlrpc_req_finished(req);
1088                 if (rc != -ETIMEDOUT)
1089                         RETURN(rc);
1090
1091                 resends++;
1092                 if (!client_should_resend(resends, &exp->exp_obd->u.cli)) {
1093                         CERROR("%s: too many resend retries: rc = %d\n",
1094                                exp->exp_obd->obd_name, -EIO);
1095                         RETURN(-EIO);
1096                 }
1097
1098                 /* If a signal interrupts then the timeout returned will
1099                  * not be zero. In that case return -EINTR
1100                  */
1101                 if (msleep_interruptible(resends * 1000))
1102                         RETURN(-EINTR);
1103
1104                 goto restart_bulk;
1105         }
1106
1107         rc = sptlrpc_cli_unwrap_bulk_read(req, req->rq_bulk,
1108                                           req->rq_bulk->bd_nob_transferred);
1109         if (rc < 0) {
1110                 ptlrpc_req_finished(req);
1111                 RETURN(rc);
1112         }
1113
1114         if (req->rq_bulk->bd_nob_transferred & ~LU_PAGE_MASK) {
1115                 CERROR("%s: unexpected bytes transferred: %d (%ld expected)\n",
1116                        exp->exp_obd->obd_name, req->rq_bulk->bd_nob_transferred,
1117                        PAGE_SIZE * npages);
1118                 ptlrpc_req_finished(req);
1119                 RETURN(-EPROTO);
1120         }
1121
1122         *request = req;
1123         RETURN(0);
1124 }
1125
1126 static void mdc_release_page(struct page *page, int remove)
1127 {
1128         if (remove) {
1129                 lock_page(page);
1130                 if (likely(page->mapping != NULL))
1131                         delete_from_page_cache(page);
1132                 unlock_page(page);
1133         }
1134         put_page(page);
1135 }
1136
1137 static struct page *mdc_page_locate(struct address_space *mapping, __u64 *hash,
1138                                     __u64 *start, __u64 *end, int hash64)
1139 {
1140         /*
1141          * Complement of hash is used as an index so that
1142          * radix_tree_gang_lookup() can be used to find a page with starting
1143          * hash _smaller_ than one we are looking for.
1144          */
1145         unsigned long offset = hash_x_index(*hash, hash64);
1146         struct page *page;
1147         unsigned long flags;
1148         int found;
1149
1150         ll_xa_lock_irqsave(&mapping->i_pages, flags);
1151         found = radix_tree_gang_lookup(&mapping->page_tree,
1152                                        (void **)&page, offset, 1);
1153         if (found > 0 && !ll_xa_is_value(page)) {
1154                 struct lu_dirpage *dp;
1155
1156                 get_page(page);
1157                 ll_xa_unlock_irqrestore(&mapping->i_pages, flags);
1158                 /*
1159                  * In contrast to find_lock_page() we are sure that directory
1160                  * page cannot be truncated (while DLM lock is held) and,
1161                  * hence, can avoid restart.
1162                  *
1163                  * In fact, page cannot be locked here at all, because
1164                  * mdc_read_page_remote does synchronous io.
1165                  */
1166                 wait_on_page_locked(page);
1167                 if (PageUptodate(page)) {
1168                         dp = kmap(page);
1169                         if (BITS_PER_LONG == 32 && hash64) {
1170                                 *start = le64_to_cpu(dp->ldp_hash_start) >> 32;
1171                                 *end   = le64_to_cpu(dp->ldp_hash_end) >> 32;
1172                                 *hash  = *hash >> 32;
1173                         } else {
1174                                 *start = le64_to_cpu(dp->ldp_hash_start);
1175                                 *end   = le64_to_cpu(dp->ldp_hash_end);
1176                         }
1177                         if (unlikely(*start == 1 && *hash == 0))
1178                                 *hash = *start;
1179                         else
1180                                 LASSERTF(*start <= *hash, "start = %#llx"
1181                                          ",end = %#llx,hash = %#llx\n",
1182                                          *start, *end, *hash);
1183                         CDEBUG(D_VFSTRACE, "offset %lx [%#llx %#llx],"
1184                               " hash %#llx\n", offset, *start, *end, *hash);
1185                         if (*hash > *end) {
1186                                 kunmap(page);
1187                                 mdc_release_page(page, 0);
1188                                 page = NULL;
1189                         } else if (*end != *start && *hash == *end) {
1190                                 /*
1191                                  * upon hash collision, remove this page,
1192                                  * otherwise put page reference, and
1193                                  * mdc_read_page_remote() will issue RPC to
1194                                  * fetch the page we want.
1195                                  */
1196                                 kunmap(page);
1197                                 mdc_release_page(page,
1198                                     le32_to_cpu(dp->ldp_flags) & LDF_COLLIDE);
1199                                 page = NULL;
1200                         }
1201                 } else {
1202                         put_page(page);
1203                         page = ERR_PTR(-EIO);
1204                 }
1205         } else {
1206                 ll_xa_unlock_irqrestore(&mapping->i_pages, flags);
1207                 page = NULL;
1208         }
1209         return page;
1210 }
1211
1212 /*
1213  * Adjust a set of pages, each page containing an array of lu_dirpages,
1214  * so that each page can be used as a single logical lu_dirpage.
1215  *
1216  * A lu_dirpage is laid out as follows, where s = ldp_hash_start,
1217  * e = ldp_hash_end, f = ldp_flags, p = padding, and each "ent" is a
1218  * struct lu_dirent.  It has size up to LU_PAGE_SIZE. The ldp_hash_end
1219  * value is used as a cookie to request the next lu_dirpage in a
1220  * directory listing that spans multiple pages (two in this example):
1221  *   ________
1222  *  |        |
1223  * .|--------v-------   -----.
1224  * |s|e|f|p|ent|ent| ... |ent|
1225  * '--|--------------   -----'   Each PAGE contains a single
1226  *    '------.                   lu_dirpage.
1227  * .---------v-------   -----.
1228  * |s|e|f|p|ent| 0 | ... | 0 |
1229  * '-----------------   -----'
1230  *
1231  * However, on hosts where the native VM page size (PAGE_SIZE) is
1232  * larger than LU_PAGE_SIZE, a single host page may contain multiple
1233  * lu_dirpages. After reading the lu_dirpages from the MDS, the
1234  * ldp_hash_end of the first lu_dirpage refers to the one immediately
1235  * after it in the same PAGE (arrows simplified for brevity, but
1236  * in general e0==s1, e1==s2, etc.):
1237  *
1238  * .--------------------   -----.
1239  * |s0|e0|f0|p|ent|ent| ... |ent|
1240  * |---v----------------   -----|
1241  * |s1|e1|f1|p|ent|ent| ... |ent|
1242  * |---v----------------   -----|  Here, each PAGE contains
1243  *             ...                 multiple lu_dirpages.
1244  * |---v----------------   -----|
1245  * |s'|e'|f'|p|ent|ent| ... |ent|
1246  * '---|----------------   -----'
1247  *     v
1248  * .----------------------------.
1249  * |        next PAGE           |
1250  *
1251  * This structure is transformed into a single logical lu_dirpage as follows:
1252  *
1253  * - Replace e0 with e' so the request for the next lu_dirpage gets the page
1254  *   labeled 'next PAGE'.
1255  *
1256  * - Copy the LDF_COLLIDE flag from f' to f0 to correctly reflect whether
1257  *   a hash collision with the next page exists.
1258  *
1259  * - Adjust the lde_reclen of the ending entry of each lu_dirpage to span
1260  *   to the first entry of the next lu_dirpage.
1261  */
1262 #if PAGE_SIZE > LU_PAGE_SIZE
1263 static void mdc_adjust_dirpages(struct page **pages, int cfs_pgs, int lu_pgs)
1264 {
1265         int i;
1266
1267         for (i = 0; i < cfs_pgs; i++) {
1268                 struct lu_dirpage *dp = kmap(pages[i]);
1269                 struct lu_dirpage *first = dp;
1270                 struct lu_dirent *end_dirent = NULL;
1271                 struct lu_dirent *ent;
1272                 __u64 hash_end = dp->ldp_hash_end;
1273                 __u32 flags = dp->ldp_flags;
1274
1275                 while (--lu_pgs > 0) {
1276                         ent = lu_dirent_start(dp);
1277                         for (end_dirent = ent; ent != NULL;
1278                              end_dirent = ent, ent = lu_dirent_next(ent));
1279
1280                         /* Advance dp to next lu_dirpage. */
1281                         dp = (struct lu_dirpage *)((char *)dp + LU_PAGE_SIZE);
1282
1283                         /* Check if we've reached the end of the PAGE. */
1284                         if (!((unsigned long)dp & ~PAGE_MASK))
1285                                 break;
1286
1287                         /* Save the hash and flags of this lu_dirpage. */
1288                         hash_end = dp->ldp_hash_end;
1289                         flags = dp->ldp_flags;
1290
1291                         /* Check if lu_dirpage contains no entries. */
1292                         if (end_dirent == NULL)
1293                                 break;
1294
1295                         /* Enlarge the end entry lde_reclen from 0 to
1296                          * first entry of next lu_dirpage. */
1297                         LASSERT(le16_to_cpu(end_dirent->lde_reclen) == 0);
1298                         end_dirent->lde_reclen =
1299                                 cpu_to_le16((char *)(dp->ldp_entries) -
1300                                             (char *)end_dirent);
1301                 }
1302
1303                 first->ldp_hash_end = hash_end;
1304                 first->ldp_flags &= ~cpu_to_le32(LDF_COLLIDE);
1305                 first->ldp_flags |= flags & cpu_to_le32(LDF_COLLIDE);
1306
1307                 kunmap(pages[i]);
1308         }
1309         LASSERTF(lu_pgs == 0, "left = %d\n", lu_pgs);
1310 }
1311 #else
1312 #define mdc_adjust_dirpages(pages, cfs_pgs, lu_pgs) do {} while (0)
1313 #endif  /* PAGE_SIZE > LU_PAGE_SIZE */
1314
1315 /* parameters for readdir page */
1316 struct readpage_param {
1317         struct md_op_data       *rp_mod;
1318         __u64                   rp_off;
1319         int                     rp_hash64;
1320         struct obd_export       *rp_exp;
1321 };
1322
1323 /**
1324  * Read pages from server.
1325  *
1326  * Page in MDS_READPAGE RPC is packed in LU_PAGE_SIZE, and each page contains
1327  * a header lu_dirpage which describes the start/end hash, and whether this
1328  * page is empty (contains no dir entry) or hash collide with next page.
1329  * After client receives reply, several pages will be integrated into dir page
1330  * in PAGE_SIZE (if PAGE_SIZE greater than LU_PAGE_SIZE), and the
1331  * lu_dirpage for this integrated page will be adjusted.
1332  **/
1333 static int mdc_read_page_remote(void *data, struct page *page0)
1334 {
1335         struct readpage_param *rp = data;
1336         struct page **page_pool;
1337         struct page *page;
1338         struct lu_dirpage *dp;
1339         struct md_op_data *op_data = rp->rp_mod;
1340         struct ptlrpc_request *req;
1341         int max_pages;
1342         struct inode *inode;
1343         struct lu_fid *fid;
1344         int rd_pgs = 0; /* number of pages actually read */
1345         int npages;
1346         int i;
1347         int rc;
1348         ENTRY;
1349
1350         max_pages = rp->rp_exp->exp_obd->u.cli.cl_max_pages_per_rpc;
1351         inode = op_data->op_data;
1352         fid = &op_data->op_fid1;
1353         LASSERT(inode != NULL);
1354
1355         OBD_ALLOC_PTR_ARRAY_LARGE(page_pool, max_pages);
1356         if (page_pool != NULL) {
1357                 page_pool[0] = page0;
1358         } else {
1359                 page_pool = &page0;
1360                 max_pages = 1;
1361         }
1362
1363         for (npages = 1; npages < max_pages; npages++) {
1364                 page = page_cache_alloc(inode->i_mapping);
1365                 if (page == NULL)
1366                         break;
1367                 page_pool[npages] = page;
1368         }
1369
1370         rc = mdc_getpage(rp->rp_exp, fid, rp->rp_off, page_pool, npages, &req);
1371         if (rc < 0) {
1372                 /* page0 is special, which was added into page cache early */
1373                 delete_from_page_cache(page0);
1374         } else {
1375                 int lu_pgs;
1376
1377                 rd_pgs = (req->rq_bulk->bd_nob_transferred + PAGE_SIZE - 1) >>
1378                         PAGE_SHIFT;
1379                 lu_pgs = req->rq_bulk->bd_nob_transferred >> LU_PAGE_SHIFT;
1380                 LASSERT(!(req->rq_bulk->bd_nob_transferred & ~LU_PAGE_MASK));
1381
1382                 CDEBUG(D_INODE, "read %d(%d) pages\n", rd_pgs, lu_pgs);
1383
1384                 mdc_adjust_dirpages(page_pool, rd_pgs, lu_pgs);
1385
1386                 SetPageUptodate(page0);
1387         }
1388         unlock_page(page0);
1389
1390         ptlrpc_req_finished(req);
1391         CDEBUG(D_CACHE, "read %d/%d pages\n", rd_pgs, npages);
1392         for (i = 1; i < npages; i++) {
1393                 unsigned long   offset;
1394                 __u64           hash;
1395                 int ret;
1396
1397                 page = page_pool[i];
1398
1399                 if (rc < 0 || i >= rd_pgs) {
1400                         put_page(page);
1401                         continue;
1402                 }
1403
1404                 SetPageUptodate(page);
1405
1406                 dp = kmap(page);
1407                 hash = le64_to_cpu(dp->ldp_hash_start);
1408                 kunmap(page);
1409
1410                 offset = hash_x_index(hash, rp->rp_hash64);
1411
1412                 prefetchw(&page->flags);
1413                 ret = add_to_page_cache_lru(page, inode->i_mapping, offset,
1414                                             GFP_KERNEL);
1415                 if (ret == 0)
1416                         unlock_page(page);
1417                 else
1418                         CDEBUG(D_VFSTRACE, "page %lu add to page cache failed:"
1419                                " rc = %d\n", offset, ret);
1420                 put_page(page);
1421         }
1422
1423         if (page_pool != &page0)
1424                 OBD_FREE_PTR_ARRAY_LARGE(page_pool, max_pages);
1425
1426         RETURN(rc);
1427 }
1428
1429 /**
1430  * Read dir page from cache first, if it can not find it, read it from
1431  * server and add into the cache.
1432  *
1433  * \param[in] exp       MDC export
1434  * \param[in] op_data   client MD stack parameters, transfering parameters
1435  *                      between different layers on client MD stack.
1436  * \param[in] mrinfo    callback required for ldlm lock enqueue during
1437  *                      read page
1438  * \param[in] hash_offset the hash offset of the page to be read
1439  * \param[in] ppage     the page to be read
1440  *
1441  * retval               = 0 get the page successfully
1442  *                      errno(<0) get the page failed
1443  */
1444 static int mdc_read_page(struct obd_export *exp, struct md_op_data *op_data,
1445                          struct md_readdir_info *mrinfo, __u64 hash_offset,
1446                          struct page **ppage)
1447 {
1448         struct lookup_intent    it = { .it_op = IT_READDIR };
1449         struct page             *page;
1450         struct inode            *dir = op_data->op_data;
1451         struct address_space    *mapping;
1452         struct lu_dirpage       *dp;
1453         __u64                   start = 0;
1454         __u64                   end = 0;
1455         struct lustre_handle    lockh;
1456         struct ptlrpc_request   *enq_req = NULL;
1457         struct readpage_param   rp_param;
1458         int rc;
1459
1460         ENTRY;
1461
1462         *ppage = NULL;
1463
1464         LASSERT(dir != NULL);
1465         mapping = dir->i_mapping;
1466
1467         rc = mdc_intent_lock(exp, op_data, &it, &enq_req,
1468                              mrinfo->mr_blocking_ast, 0);
1469         if (enq_req != NULL)
1470                 ptlrpc_req_finished(enq_req);
1471
1472         if (rc < 0) {
1473                 CERROR("%s: "DFID" lock enqueue fails: rc = %d\n",
1474                        exp->exp_obd->obd_name, PFID(&op_data->op_fid1), rc);
1475                 RETURN(rc);
1476         }
1477
1478         rc = 0;
1479         lockh.cookie = it.it_lock_handle;
1480         mdc_set_lock_data(exp, &lockh, dir, NULL);
1481
1482         rp_param.rp_off = hash_offset;
1483         rp_param.rp_hash64 = op_data->op_cli_flags & CLI_HASH64;
1484         page = mdc_page_locate(mapping, &rp_param.rp_off, &start, &end,
1485                                rp_param.rp_hash64);
1486         if (IS_ERR(page)) {
1487                 CERROR("%s: dir page locate: "DFID" at %llu: rc %ld\n",
1488                        exp->exp_obd->obd_name, PFID(&op_data->op_fid1),
1489                        rp_param.rp_off, PTR_ERR(page));
1490                 GOTO(out_unlock, rc = PTR_ERR(page));
1491         } else if (page != NULL) {
1492                 /*
1493                  * XXX nikita: not entirely correct handling of a corner case:
1494                  * suppose hash chain of entries with hash value HASH crosses
1495                  * border between pages P0 and P1. First both P0 and P1 are
1496                  * cached, seekdir() is called for some entry from the P0 part
1497                  * of the chain. Later P0 goes out of cache. telldir(HASH)
1498                  * happens and finds P1, as it starts with matching hash
1499                  * value. Remaining entries from P0 part of the chain are
1500                  * skipped. (Is that really a bug?)
1501                  *
1502                  * Possible solutions: 0. don't cache P1 is such case, handle
1503                  * it as an "overflow" page. 1. invalidate all pages at
1504                  * once. 2. use HASH|1 as an index for P1.
1505                  */
1506                 GOTO(hash_collision, page);
1507         }
1508
1509         rp_param.rp_exp = exp;
1510         rp_param.rp_mod = op_data;
1511         page = read_cache_page(mapping,
1512                                hash_x_index(rp_param.rp_off,
1513                                             rp_param.rp_hash64),
1514                                mdc_read_page_remote, &rp_param);
1515         if (IS_ERR(page)) {
1516                 CDEBUG(D_INFO, "%s: read cache page: "DFID" at %llu: %ld\n",
1517                        exp->exp_obd->obd_name, PFID(&op_data->op_fid1),
1518                        rp_param.rp_off, PTR_ERR(page));
1519                 GOTO(out_unlock, rc = PTR_ERR(page));
1520         }
1521
1522         wait_on_page_locked(page);
1523         (void)kmap(page);
1524         if (!PageUptodate(page)) {
1525                 CERROR("%s: page not updated: "DFID" at %llu: rc %d\n",
1526                        exp->exp_obd->obd_name, PFID(&op_data->op_fid1),
1527                        rp_param.rp_off, -5);
1528                 goto fail;
1529         }
1530         if (!PageChecked(page))
1531                 SetPageChecked(page);
1532         if (PageError(page)) {
1533                 CERROR("%s: page error: "DFID" at %llu: rc %d\n",
1534                        exp->exp_obd->obd_name, PFID(&op_data->op_fid1),
1535                        rp_param.rp_off, -5);
1536                 goto fail;
1537         }
1538
1539 hash_collision:
1540         dp = page_address(page);
1541         if (BITS_PER_LONG == 32 && rp_param.rp_hash64) {
1542                 start = le64_to_cpu(dp->ldp_hash_start) >> 32;
1543                 end   = le64_to_cpu(dp->ldp_hash_end) >> 32;
1544                 rp_param.rp_off = hash_offset >> 32;
1545         } else {
1546                 start = le64_to_cpu(dp->ldp_hash_start);
1547                 end   = le64_to_cpu(dp->ldp_hash_end);
1548                 rp_param.rp_off = hash_offset;
1549         }
1550         if (end == start) {
1551                 LASSERT(start == rp_param.rp_off);
1552                 CWARN("Page-wide hash collision: %#lx\n", (unsigned long)end);
1553 #if BITS_PER_LONG == 32
1554                 CWARN("Real page-wide hash collision at [%llu %llu] with "
1555                       "hash %llu\n", le64_to_cpu(dp->ldp_hash_start),
1556                       le64_to_cpu(dp->ldp_hash_end), hash_offset);
1557 #endif
1558
1559                 /*
1560                  * Fetch whole overflow chain...
1561                  *
1562                  * XXX not yet.
1563                  */
1564                 goto fail;
1565         }
1566         *ppage = page;
1567 out_unlock:
1568         ldlm_lock_decref(&lockh, it.it_lock_mode);
1569         return rc;
1570 fail:
1571         kunmap(page);
1572         mdc_release_page(page, 1);
1573         rc = -EIO;
1574         goto out_unlock;
1575 }
1576
1577 static int mdc_statfs_interpret(const struct lu_env *env,
1578                                 struct ptlrpc_request *req, void *args, int rc)
1579 {
1580         struct obd_info *oinfo = args;
1581         struct obd_statfs *osfs;
1582
1583         if (!rc) {
1584                 osfs = req_capsule_server_get(&req->rq_pill, &RMF_OBD_STATFS);
1585                 if (!osfs)
1586                         return -EPROTO;
1587
1588                 oinfo->oi_osfs = osfs;
1589
1590                 CDEBUG(D_CACHE, "blocks=%llu free=%llu avail=%llu "
1591                        "objects=%llu free=%llu state=%x\n",
1592                         osfs->os_blocks, osfs->os_bfree, osfs->os_bavail,
1593                         osfs->os_files, osfs->os_ffree, osfs->os_state);
1594         }
1595
1596         oinfo->oi_cb_up(oinfo, rc);
1597
1598         return rc;
1599 }
1600
1601 static int mdc_statfs_async(struct obd_export *exp,
1602                             struct obd_info *oinfo, time64_t max_age,
1603                             struct ptlrpc_request_set *unused)
1604 {
1605         struct ptlrpc_request *req;
1606         struct obd_info *aa;
1607
1608         req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp), &RQF_MDS_STATFS,
1609                                         LUSTRE_MDS_VERSION, MDS_STATFS);
1610         if (req == NULL)
1611                 return -ENOMEM;
1612
1613         ptlrpc_request_set_replen(req);
1614         req->rq_interpret_reply = mdc_statfs_interpret;
1615
1616         aa = ptlrpc_req_async_args(aa, req);
1617         *aa = *oinfo;
1618
1619         ptlrpcd_add_req(req);
1620
1621         return 0;
1622 }
1623
1624 static int mdc_statfs(const struct lu_env *env,
1625                       struct obd_export *exp, struct obd_statfs *osfs,
1626                       time64_t max_age, __u32 flags)
1627 {
1628         struct obd_device *obd = class_exp2obd(exp);
1629         struct req_format *fmt;
1630         struct ptlrpc_request *req;
1631         struct obd_statfs *msfs;
1632         struct obd_import *imp, *imp0;
1633         int rc;
1634         ENTRY;
1635
1636         /*
1637          * Since the request might also come from lprocfs, so we need
1638          * sync this with client_disconnect_export Bug15684
1639          */
1640         with_imp_locked(obd, imp0, rc)
1641                 imp = class_import_get(imp0);
1642         if (rc)
1643                 RETURN(rc);
1644
1645         fmt = &RQF_MDS_STATFS;
1646         if ((exp_connect_flags2(exp) & OBD_CONNECT2_SUM_STATFS) &&
1647             (flags & OBD_STATFS_SUM))
1648                 fmt = &RQF_MDS_STATFS_NEW;
1649         req = ptlrpc_request_alloc_pack(imp, fmt, LUSTRE_MDS_VERSION,
1650                                         MDS_STATFS);
1651         if (req == NULL)
1652                 GOTO(output, rc = -ENOMEM);
1653         req->rq_allow_intr = 1;
1654
1655         if ((flags & OBD_STATFS_SUM) &&
1656             (exp_connect_flags2(exp) & OBD_CONNECT2_SUM_STATFS)) {
1657                 /* request aggregated states */
1658                 struct mdt_body *body;
1659
1660                 body = req_capsule_client_get(&req->rq_pill, &RMF_MDT_BODY);
1661                 if (body == NULL)
1662                         GOTO(out, rc = -EPROTO);
1663                 body->mbo_valid = OBD_MD_FLAGSTATFS;
1664         }
1665
1666         ptlrpc_request_set_replen(req);
1667
1668         if (flags & OBD_STATFS_NODELAY) {
1669                 /* procfs requests not want stay in wait for avoid deadlock */
1670                 req->rq_no_resend = 1;
1671                 req->rq_no_delay = 1;
1672         }
1673
1674         rc = ptlrpc_queue_wait(req);
1675         if (rc) {
1676                 /* check connection error first */
1677                 if (imp->imp_connect_error)
1678                         rc = imp->imp_connect_error;
1679                 GOTO(out, rc);
1680         }
1681
1682         msfs = req_capsule_server_get(&req->rq_pill, &RMF_OBD_STATFS);
1683         if (msfs == NULL)
1684                 GOTO(out, rc = -EPROTO);
1685
1686         *osfs = *msfs;
1687         EXIT;
1688 out:
1689         ptlrpc_req_finished(req);
1690 output:
1691         class_import_put(imp);
1692         return rc;
1693 }
1694
1695 static int mdc_ioc_fid2path(struct obd_export *exp, struct getinfo_fid2path *gf)
1696 {
1697         __u32 keylen, vallen;
1698         void *key;
1699         int rc;
1700
1701         if (gf->gf_pathlen > PATH_MAX)
1702                 RETURN(-ENAMETOOLONG);
1703         if (gf->gf_pathlen < 2)
1704                 RETURN(-EOVERFLOW);
1705
1706         /* Key is KEY_FID2PATH + getinfo_fid2path description */
1707         keylen = cfs_size_round(sizeof(KEY_FID2PATH) + sizeof(*gf) +
1708                                 sizeof(struct lu_fid));
1709         OBD_ALLOC(key, keylen);
1710         if (key == NULL)
1711                 RETURN(-ENOMEM);
1712         memcpy(key, KEY_FID2PATH, sizeof(KEY_FID2PATH));
1713         memcpy(key + cfs_size_round(sizeof(KEY_FID2PATH)), gf, sizeof(*gf));
1714         memcpy(key + cfs_size_round(sizeof(KEY_FID2PATH)) + sizeof(*gf),
1715                gf->gf_u.gf_root_fid, sizeof(struct lu_fid));
1716         CDEBUG(D_IOCTL, "path get "DFID" from %llu #%d\n",
1717                PFID(&gf->gf_fid), gf->gf_recno, gf->gf_linkno);
1718
1719         if (!fid_is_sane(&gf->gf_fid))
1720                 GOTO(out, rc = -EINVAL);
1721
1722         /* Val is struct getinfo_fid2path result plus path */
1723         vallen = sizeof(*gf) + gf->gf_pathlen;
1724
1725         rc = obd_get_info(NULL, exp, keylen, key, &vallen, gf);
1726         if (rc != 0 && rc != -EREMOTE)
1727                 GOTO(out, rc);
1728
1729         if (vallen <= sizeof(*gf))
1730                 GOTO(out, rc = -EPROTO);
1731         if (vallen > sizeof(*gf) + gf->gf_pathlen)
1732                 GOTO(out, rc = -EOVERFLOW);
1733
1734         CDEBUG(D_IOCTL, "path got "DFID" from %llu #%d: %s\n",
1735                PFID(&gf->gf_fid), gf->gf_recno, gf->gf_linkno,
1736                gf->gf_pathlen < 512 ? gf->gf_u.gf_path :
1737                /* only log the last 512 characters of the path */
1738                gf->gf_u.gf_path + gf->gf_pathlen - 512);
1739
1740 out:
1741         OBD_FREE(key, keylen);
1742         return rc;
1743 }
1744
1745 static int mdc_ioc_hsm_progress(struct obd_export *exp,
1746                                 struct hsm_progress_kernel *hpk)
1747 {
1748         struct obd_import               *imp = class_exp2cliimp(exp);
1749         struct hsm_progress_kernel      *req_hpk;
1750         struct ptlrpc_request           *req;
1751         int                              rc;
1752         ENTRY;
1753
1754         req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_HSM_PROGRESS,
1755                                         LUSTRE_MDS_VERSION, MDS_HSM_PROGRESS);
1756         if (req == NULL)
1757                 GOTO(out, rc = -ENOMEM);
1758
1759         mdc_pack_body(&req->rq_pill, NULL, 0, 0, -1, 0);
1760
1761         /* Copy hsm_progress struct */
1762         req_hpk = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_PROGRESS);
1763         if (req_hpk == NULL)
1764                 GOTO(out, rc = -EPROTO);
1765
1766         *req_hpk = *hpk;
1767         req_hpk->hpk_errval = lustre_errno_hton(hpk->hpk_errval);
1768
1769         ptlrpc_request_set_replen(req);
1770
1771         ptlrpc_get_mod_rpc_slot(req);
1772         rc = ptlrpc_queue_wait(req);
1773         ptlrpc_put_mod_rpc_slot(req);
1774
1775         GOTO(out, rc);
1776 out:
1777         ptlrpc_req_finished(req);
1778         return rc;
1779 }
1780 /**
1781  * Send hsm_ct_register to MDS
1782  *
1783  * \param[in]   imp             import
1784  * \param[in]   archive_count   if in bitmap format, it is the bitmap,
1785  *                              else it is the count of archive_ids
1786  * \param[in]   archives        if in bitmap format, it is NULL,
1787  *                              else it is archive_id lists
1788  */
1789 static int mdc_ioc_hsm_ct_register(struct obd_import *imp, __u32 archive_count,
1790                                    __u32 *archives)
1791 {
1792         struct ptlrpc_request *req;
1793         __u32 *archive_array;
1794         size_t archives_size;
1795         int rc;
1796         ENTRY;
1797
1798         req = ptlrpc_request_alloc(imp, &RQF_MDS_HSM_CT_REGISTER);
1799         if (req == NULL)
1800                 RETURN(-ENOMEM);
1801
1802         if (archives != NULL)
1803                 archives_size = sizeof(*archive_array) * archive_count;
1804         else
1805                 archives_size = sizeof(archive_count);
1806
1807         req_capsule_set_size(&req->rq_pill, &RMF_MDS_HSM_ARCHIVE,
1808                              RCL_CLIENT, archives_size);
1809
1810         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_CT_REGISTER);
1811         if (rc) {
1812                 ptlrpc_request_free(req);
1813                 RETURN(-ENOMEM);
1814         }
1815
1816         mdc_pack_body(&req->rq_pill, NULL, 0, 0, -1, 0);
1817
1818         archive_array = req_capsule_client_get(&req->rq_pill,
1819                                                &RMF_MDS_HSM_ARCHIVE);
1820         if (archive_array == NULL)
1821                 GOTO(out, rc = -EPROTO);
1822
1823         if (archives != NULL)
1824                 memcpy(archive_array, archives, archives_size);
1825         else
1826                 *archive_array = archive_count;
1827
1828         ptlrpc_request_set_replen(req);
1829         req->rq_no_resend = 1;
1830
1831         rc = mdc_queue_wait(req);
1832         GOTO(out, rc);
1833 out:
1834         ptlrpc_req_finished(req);
1835         return rc;
1836 }
1837
1838 static int mdc_ioc_hsm_current_action(struct obd_export *exp,
1839                                       struct md_op_data *op_data)
1840 {
1841         struct hsm_current_action       *hca = op_data->op_data;
1842         struct hsm_current_action       *req_hca;
1843         struct ptlrpc_request           *req;
1844         int                              rc;
1845         ENTRY;
1846
1847         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1848                                    &RQF_MDS_HSM_ACTION);
1849         if (req == NULL)
1850                 RETURN(-ENOMEM);
1851
1852         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_ACTION);
1853         if (rc) {
1854                 ptlrpc_request_free(req);
1855                 RETURN(rc);
1856         }
1857
1858         mdc_pack_body(&req->rq_pill, &op_data->op_fid1, 0, 0,
1859                       op_data->op_suppgids[0], 0);
1860
1861         ptlrpc_request_set_replen(req);
1862
1863         rc = mdc_queue_wait(req);
1864         if (rc)
1865                 GOTO(out, rc);
1866
1867         req_hca = req_capsule_server_get(&req->rq_pill,
1868                                          &RMF_MDS_HSM_CURRENT_ACTION);
1869         if (req_hca == NULL)
1870                 GOTO(out, rc = -EPROTO);
1871
1872         *hca = *req_hca;
1873
1874         EXIT;
1875 out:
1876         ptlrpc_req_finished(req);
1877         return rc;
1878 }
1879
1880 static int mdc_ioc_hsm_ct_unregister(struct obd_import *imp)
1881 {
1882         struct ptlrpc_request   *req;
1883         int                      rc;
1884         ENTRY;
1885
1886         req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_HSM_CT_UNREGISTER,
1887                                         LUSTRE_MDS_VERSION,
1888                                         MDS_HSM_CT_UNREGISTER);
1889         if (req == NULL)
1890                 GOTO(out, rc = -ENOMEM);
1891
1892         mdc_pack_body(&req->rq_pill, NULL, 0, 0, -1, 0);
1893
1894         ptlrpc_request_set_replen(req);
1895
1896         rc = mdc_queue_wait(req);
1897         GOTO(out, rc);
1898 out:
1899         ptlrpc_req_finished(req);
1900         return rc;
1901 }
1902
1903 static int mdc_ioc_hsm_state_get(struct obd_export *exp,
1904                                  struct md_op_data *op_data)
1905 {
1906         struct hsm_user_state   *hus = op_data->op_data;
1907         struct hsm_user_state   *req_hus;
1908         struct ptlrpc_request   *req;
1909         int                      rc;
1910         ENTRY;
1911
1912         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1913                                    &RQF_MDS_HSM_STATE_GET);
1914         if (req == NULL)
1915                 RETURN(-ENOMEM);
1916
1917         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_STATE_GET);
1918         if (rc != 0) {
1919                 ptlrpc_request_free(req);
1920                 RETURN(rc);
1921         }
1922
1923         mdc_pack_body(&req->rq_pill, &op_data->op_fid1, 0, 0,
1924                       op_data->op_suppgids[0], 0);
1925
1926         ptlrpc_request_set_replen(req);
1927
1928         rc = mdc_queue_wait(req);
1929         if (rc)
1930                 GOTO(out, rc);
1931
1932         req_hus = req_capsule_server_get(&req->rq_pill, &RMF_HSM_USER_STATE);
1933         if (req_hus == NULL)
1934                 GOTO(out, rc = -EPROTO);
1935
1936         *hus = *req_hus;
1937
1938         EXIT;
1939 out:
1940         ptlrpc_req_finished(req);
1941         return rc;
1942 }
1943
1944 static int mdc_ioc_hsm_state_set(struct obd_export *exp,
1945                                  struct md_op_data *op_data)
1946 {
1947         struct hsm_state_set    *hss = op_data->op_data;
1948         struct hsm_state_set    *req_hss;
1949         struct ptlrpc_request   *req;
1950         int                      rc;
1951         ENTRY;
1952
1953         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1954                                    &RQF_MDS_HSM_STATE_SET);
1955         if (req == NULL)
1956                 RETURN(-ENOMEM);
1957
1958         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_STATE_SET);
1959         if (rc) {
1960                 ptlrpc_request_free(req);
1961                 RETURN(rc);
1962         }
1963
1964         mdc_pack_body(&req->rq_pill, &op_data->op_fid1, 0, 0,
1965                       op_data->op_suppgids[0], 0);
1966
1967         /* Copy states */
1968         req_hss = req_capsule_client_get(&req->rq_pill, &RMF_HSM_STATE_SET);
1969         if (req_hss == NULL)
1970                 GOTO(out, rc = -EPROTO);
1971         *req_hss = *hss;
1972
1973         ptlrpc_request_set_replen(req);
1974
1975         ptlrpc_get_mod_rpc_slot(req);
1976         rc = ptlrpc_queue_wait(req);
1977         ptlrpc_put_mod_rpc_slot(req);
1978
1979         GOTO(out, rc);
1980 out:
1981         ptlrpc_req_finished(req);
1982         return rc;
1983 }
1984
1985 static int mdc_ioc_hsm_request(struct obd_export *exp,
1986                                struct hsm_user_request *hur)
1987 {
1988         struct obd_import       *imp = class_exp2cliimp(exp);
1989         struct ptlrpc_request   *req;
1990         struct hsm_request      *req_hr;
1991         struct hsm_user_item    *req_hui;
1992         char                    *req_opaque;
1993         int                      rc;
1994         ENTRY;
1995
1996         req = ptlrpc_request_alloc(imp, &RQF_MDS_HSM_REQUEST);
1997         if (req == NULL)
1998                 GOTO(out, rc = -ENOMEM);
1999
2000         req_capsule_set_size(&req->rq_pill, &RMF_MDS_HSM_USER_ITEM, RCL_CLIENT,
2001                              hur->hur_request.hr_itemcount
2002                              * sizeof(struct hsm_user_item));
2003         req_capsule_set_size(&req->rq_pill, &RMF_GENERIC_DATA, RCL_CLIENT,
2004                              hur->hur_request.hr_data_len);
2005
2006         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_REQUEST);
2007         if (rc) {
2008                 ptlrpc_request_free(req);
2009                 RETURN(rc);
2010         }
2011
2012         mdc_pack_body(&req->rq_pill, NULL, 0, 0, -1, 0);
2013
2014         /* Copy hsm_request struct */
2015         req_hr = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_REQUEST);
2016         if (req_hr == NULL)
2017                 GOTO(out, rc = -EPROTO);
2018         *req_hr = hur->hur_request;
2019
2020         /* Copy hsm_user_item structs */
2021         req_hui = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_USER_ITEM);
2022         if (req_hui == NULL)
2023                 GOTO(out, rc = -EPROTO);
2024         memcpy(req_hui, hur->hur_user_item,
2025                hur->hur_request.hr_itemcount * sizeof(struct hsm_user_item));
2026
2027         /* Copy opaque field */
2028         req_opaque = req_capsule_client_get(&req->rq_pill, &RMF_GENERIC_DATA);
2029         if (req_opaque == NULL)
2030                 GOTO(out, rc = -EPROTO);
2031         memcpy(req_opaque, hur_data(hur), hur->hur_request.hr_data_len);
2032
2033         ptlrpc_request_set_replen(req);
2034
2035         ptlrpc_get_mod_rpc_slot(req);
2036         rc = ptlrpc_queue_wait(req);
2037         ptlrpc_put_mod_rpc_slot(req);
2038
2039         GOTO(out, rc);
2040
2041 out:
2042         ptlrpc_req_finished(req);
2043         return rc;
2044 }
2045
2046 static int mdc_ioc_hsm_ct_start(struct obd_export *exp,
2047                                 struct lustre_kernelcomm *lk);
2048
2049 static int mdc_quotactl(struct obd_device *unused, struct obd_export *exp,
2050                         struct obd_quotactl *oqctl)
2051 {
2052         struct ptlrpc_request *req;
2053         struct obd_quotactl *oqc;
2054         int rc;
2055         ENTRY;
2056
2057         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_QUOTACTL);
2058         if (req == NULL)
2059                 RETURN(-ENOMEM);
2060
2061
2062         if (LUSTRE_Q_CMD_IS_POOL(oqctl->qc_cmd))
2063                 req_capsule_set_size(&req->rq_pill,
2064                                      &RMF_OBD_QUOTACTL,
2065                                      RCL_CLIENT,
2066                                      sizeof(*oqc) + LOV_MAXPOOLNAME + 1);
2067
2068         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION,
2069                                  MDS_QUOTACTL);
2070         if (rc) {
2071                 ptlrpc_request_free(req);
2072                 RETURN(rc);
2073         }
2074
2075         oqc = req_capsule_client_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
2076         QCTL_COPY(oqc, oqctl);
2077
2078         ptlrpc_request_set_replen(req);
2079         ptlrpc_at_set_req_timeout(req);
2080
2081         rc = ptlrpc_queue_wait(req);
2082         if (rc) {
2083                 CERROR("%s: ptlrpc_queue_wait failed: rc = %d\n",
2084                        exp->exp_obd->obd_name, rc);
2085                 GOTO(out, rc);
2086         }
2087
2088         if (req->rq_repmsg &&
2089             (oqc = req_capsule_server_get(&req->rq_pill, &RMF_OBD_QUOTACTL))) {
2090                 QCTL_COPY(oqctl, oqc);
2091         } else if (!rc) {
2092                 rc = -EPROTO;
2093                 CERROR("%s: cannot unpack obd_quotactl: rc = %d\n",
2094                         exp->exp_obd->obd_name, rc);
2095         }
2096 out:
2097         ptlrpc_req_finished(req);
2098
2099         RETURN(rc);
2100 }
2101
2102 static int mdc_ioc_swap_layouts(struct obd_export *exp,
2103                                 struct md_op_data *op_data)
2104 {
2105         LIST_HEAD(cancels);
2106         struct ptlrpc_request   *req;
2107         int                      rc, count;
2108         struct mdc_swap_layouts *msl, *payload;
2109         ENTRY;
2110
2111         msl = op_data->op_data;
2112
2113         /* When the MDT will get the MDS_SWAP_LAYOUTS RPC the
2114          * first thing it will do is to cancel the 2 layout
2115          * locks held by this client.
2116          * So the client must cancel its layout locks on the 2 fids
2117          * with the request RPC to avoid extra RPC round trips.
2118          */
2119         count = mdc_resource_get_unused(exp, &op_data->op_fid1, &cancels,
2120                                         LCK_EX, MDS_INODELOCK_LAYOUT |
2121                                         MDS_INODELOCK_XATTR);
2122         count += mdc_resource_get_unused(exp, &op_data->op_fid2, &cancels,
2123                                          LCK_EX, MDS_INODELOCK_LAYOUT |
2124                                          MDS_INODELOCK_XATTR);
2125
2126         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
2127                                    &RQF_MDS_SWAP_LAYOUTS);
2128         if (req == NULL) {
2129                 ldlm_lock_list_put(&cancels, l_bl_ast, count);
2130                 RETURN(-ENOMEM);
2131         }
2132
2133         rc = mdc_prep_elc_req(exp, req, MDS_SWAP_LAYOUTS, &cancels, count);
2134         if (rc) {
2135                 ptlrpc_request_free(req);
2136                 RETURN(rc);
2137         }
2138
2139         mdc_swap_layouts_pack(&req->rq_pill, op_data);
2140
2141         payload = req_capsule_client_get(&req->rq_pill, &RMF_SWAP_LAYOUTS);
2142         LASSERT(payload);
2143
2144         *payload = *msl;
2145
2146         ptlrpc_request_set_replen(req);
2147
2148         rc = ptlrpc_queue_wait(req);
2149         if (rc)
2150                 GOTO(out, rc);
2151         EXIT;
2152
2153 out:
2154         ptlrpc_req_finished(req);
2155         return rc;
2156 }
2157
2158 static int mdc_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
2159                          void *karg, void __user *uarg)
2160 {
2161         struct obd_device *obd = exp->exp_obd;
2162         struct obd_ioctl_data *data = karg;
2163         struct obd_import *imp = obd->u.cli.cl_import;
2164         int rc;
2165         ENTRY;
2166
2167         if (!try_module_get(THIS_MODULE)) {
2168                 CERROR("%s: cannot get module '%s'\n", obd->obd_name,
2169                        module_name(THIS_MODULE));
2170                 return -EINVAL;
2171         }
2172         switch (cmd) {
2173         case OBD_IOC_FID2PATH:
2174                 rc = mdc_ioc_fid2path(exp, karg);
2175                 GOTO(out, rc);
2176         case LL_IOC_HSM_CT_START:
2177                 rc = mdc_ioc_hsm_ct_start(exp, karg);
2178                 /* ignore if it was already registered on this MDS. */
2179                 if (rc == -EEXIST)
2180                         rc = 0;
2181                 GOTO(out, rc);
2182         case LL_IOC_HSM_PROGRESS:
2183                 rc = mdc_ioc_hsm_progress(exp, karg);
2184                 GOTO(out, rc);
2185         case LL_IOC_HSM_STATE_GET:
2186                 rc = mdc_ioc_hsm_state_get(exp, karg);
2187                 GOTO(out, rc);
2188         case LL_IOC_HSM_STATE_SET:
2189                 rc = mdc_ioc_hsm_state_set(exp, karg);
2190                 GOTO(out, rc);
2191         case LL_IOC_HSM_ACTION:
2192                 rc = mdc_ioc_hsm_current_action(exp, karg);
2193                 GOTO(out, rc);
2194         case LL_IOC_HSM_REQUEST:
2195                 rc = mdc_ioc_hsm_request(exp, karg);
2196                 GOTO(out, rc);
2197         case OBD_IOC_CLIENT_RECOVER:
2198                 rc = ptlrpc_recover_import(imp, data->ioc_inlbuf1, 0);
2199                 if (rc < 0)
2200                         GOTO(out, rc);
2201                 GOTO(out, rc = 0);
2202         case IOC_OSC_SET_ACTIVE:
2203                 rc = ptlrpc_set_import_active(imp, data->ioc_offset);
2204                 GOTO(out, rc);
2205         /*
2206          * Normally IOC_OBD_STATFS, OBD_IOC_QUOTACTL iocontrol are handled by
2207          * LMV instead of MDC. But when the cluster is upgraded from 1.8,
2208          * there'd be no LMV layer thus we might be called here. Eventually
2209          * this code should be removed.
2210          * bz20731, LU-592.
2211          */
2212         case IOC_OBD_STATFS: {
2213                 struct obd_statfs stat_buf = {0};
2214
2215                 if (*((__u32 *) data->ioc_inlbuf2) != 0)
2216                         GOTO(out, rc = -ENODEV);
2217
2218                 /* copy UUID */
2219                 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(obd),
2220                                  min((int)data->ioc_plen2,
2221                                      (int)sizeof(struct obd_uuid))))
2222                         GOTO(out, rc = -EFAULT);
2223
2224                 rc = mdc_statfs(NULL, obd->obd_self_export, &stat_buf,
2225                                 ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS,
2226                                 0);
2227                 if (rc != 0)
2228                         GOTO(out, rc);
2229
2230                 if (copy_to_user(data->ioc_pbuf1, &stat_buf,
2231                                      min((int) data->ioc_plen1,
2232                                          (int) sizeof(stat_buf))))
2233                         GOTO(out, rc = -EFAULT);
2234
2235                 GOTO(out, rc = 0);
2236         }
2237         case OBD_IOC_QUOTACTL: {
2238                 struct if_quotactl *qctl = karg;
2239                 struct obd_quotactl *oqctl;
2240
2241                 OBD_ALLOC_PTR(oqctl);
2242                 if (oqctl == NULL)
2243                         GOTO(out, rc = -ENOMEM);
2244
2245                 QCTL_COPY(oqctl, qctl);
2246                 rc = obd_quotactl(exp, oqctl);
2247                 if (rc == 0) {
2248                         QCTL_COPY(qctl, oqctl);
2249                         qctl->qc_valid = QC_MDTIDX;
2250                         qctl->obd_uuid = obd->u.cli.cl_target_uuid;
2251                 }
2252
2253                 OBD_FREE_PTR(oqctl);
2254                 GOTO(out, rc);
2255         }
2256         case LL_IOC_GET_CONNECT_FLAGS:
2257                 if (copy_to_user(uarg, exp_connect_flags_ptr(exp),
2258                                  sizeof(*exp_connect_flags_ptr(exp))))
2259                         GOTO(out, rc = -EFAULT);
2260
2261                 GOTO(out, rc = 0);
2262         case LL_IOC_LOV_SWAP_LAYOUTS:
2263                 rc = mdc_ioc_swap_layouts(exp, karg);
2264                 GOTO(out, rc);
2265         default:
2266                 CERROR("unrecognised ioctl: cmd = %#x\n", cmd);
2267                 GOTO(out, rc = -ENOTTY);
2268         }
2269 out:
2270         module_put(THIS_MODULE);
2271
2272         return rc;
2273 }
2274
2275 static int mdc_get_info_rpc(struct obd_export *exp,
2276                             u32 keylen, void *key,
2277                             u32 vallen, void *val)
2278 {
2279         struct obd_import      *imp = class_exp2cliimp(exp);
2280         struct ptlrpc_request  *req;
2281         char                   *tmp;
2282         int                     rc = -EINVAL;
2283         ENTRY;
2284
2285         req = ptlrpc_request_alloc(imp, &RQF_MDS_GET_INFO);
2286         if (req == NULL)
2287                 RETURN(-ENOMEM);
2288
2289         req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_KEY,
2290                              RCL_CLIENT, keylen);
2291         req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_VALLEN,
2292                              RCL_CLIENT, sizeof(vallen));
2293
2294         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GET_INFO);
2295         if (rc) {
2296                 ptlrpc_request_free(req);
2297                 RETURN(rc);
2298         }
2299
2300         tmp = req_capsule_client_get(&req->rq_pill, &RMF_GETINFO_KEY);
2301         memcpy(tmp, key, keylen);
2302         tmp = req_capsule_client_get(&req->rq_pill, &RMF_GETINFO_VALLEN);
2303         memcpy(tmp, &vallen, sizeof(vallen));
2304
2305         req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_VAL,
2306                              RCL_SERVER, vallen);
2307         ptlrpc_request_set_replen(req);
2308
2309         /* if server failed to resolve FID, and OI scrub not able to fix it, it
2310          * will return -EINPROGRESS, ptlrpc_queue_wait() will keep retrying,
2311          * set request interruptible to avoid deadlock.
2312          */
2313         if (KEY_IS(KEY_FID2PATH))
2314                 req->rq_allow_intr = 1;
2315
2316         rc = ptlrpc_queue_wait(req);
2317         /* -EREMOTE means the get_info result is partial, and it needs to
2318          * continue on another MDT, see fid2path part in lmv_iocontrol */
2319         if (rc == 0 || rc == -EREMOTE) {
2320                 tmp = req_capsule_server_get(&req->rq_pill, &RMF_GETINFO_VAL);
2321                 memcpy(val, tmp, vallen);
2322                 if (req_capsule_rep_need_swab(&req->rq_pill)) {
2323                         if (KEY_IS(KEY_FID2PATH))
2324                                 lustre_swab_fid2path(val);
2325                 }
2326         }
2327         ptlrpc_req_finished(req);
2328
2329         RETURN(rc);
2330 }
2331
2332 static void lustre_swab_hai(struct hsm_action_item *h)
2333 {
2334         __swab32s(&h->hai_len);
2335         __swab32s(&h->hai_action);
2336         lustre_swab_lu_fid(&h->hai_fid);
2337         lustre_swab_lu_fid(&h->hai_dfid);
2338         __swab64s(&h->hai_cookie);
2339         __swab64s(&h->hai_extent.offset);
2340         __swab64s(&h->hai_extent.length);
2341         __swab64s(&h->hai_gid);
2342 }
2343
2344 static void lustre_swab_hal(struct hsm_action_list *h)
2345 {
2346         struct hsm_action_item  *hai;
2347         __u32                    i;
2348
2349         __swab32s(&h->hal_version);
2350         __swab32s(&h->hal_count);
2351         __swab32s(&h->hal_archive_id);
2352         __swab64s(&h->hal_flags);
2353         hai = hai_first(h);
2354         for (i = 0; i < h->hal_count; i++, hai = hai_next(hai))
2355                 lustre_swab_hai(hai);
2356 }
2357
2358 static void lustre_swab_kuch(struct kuc_hdr *l)
2359 {
2360         __swab16s(&l->kuc_magic);
2361         /* __u8 l->kuc_transport */
2362         __swab16s(&l->kuc_msgtype);
2363         __swab16s(&l->kuc_msglen);
2364 }
2365
2366 static int mdc_ioc_hsm_ct_start(struct obd_export *exp,
2367                                 struct lustre_kernelcomm *lk)
2368 {
2369         struct obd_import *imp = class_exp2cliimp(exp);
2370         int rc = 0;
2371
2372         if (lk->lk_group != KUC_GRP_HSM) {
2373                 CERROR("Bad copytool group %d\n", lk->lk_group);
2374                 return -EINVAL;
2375         }
2376
2377         CDEBUG(D_HSM, "CT start r%d w%d u%d g%d f%#x\n", lk->lk_rfd, lk->lk_wfd,
2378                lk->lk_uid, lk->lk_group, lk->lk_flags);
2379
2380         if (lk->lk_flags & LK_FLG_STOP) {
2381                 /* Unregister with the coordinator */
2382                 rc = mdc_ioc_hsm_ct_unregister(imp);
2383         } else {
2384                 __u32 *archives = NULL;
2385
2386                 if ((lk->lk_flags & LK_FLG_DATANR) && lk->lk_data_count > 0)
2387                         archives = lk->lk_data;
2388
2389                 rc = mdc_ioc_hsm_ct_register(imp, lk->lk_data_count, archives);
2390         }
2391
2392         return rc;
2393 }
2394
2395 /**
2396  * Send a message to any listening copytools
2397  * @param val KUC message (kuc_hdr + hsm_action_list)
2398  * @param len total length of message
2399  */
2400 static int mdc_hsm_copytool_send(const struct obd_uuid *uuid,
2401                                  size_t len, void *val)
2402 {
2403         struct kuc_hdr          *lh = (struct kuc_hdr *)val;
2404         struct hsm_action_list  *hal = (struct hsm_action_list *)(lh + 1);
2405         int                      rc;
2406         ENTRY;
2407
2408         if (len < sizeof(*lh) + sizeof(*hal)) {
2409                 CERROR("Short HSM message %zu < %zu\n", len,
2410                        sizeof(*lh) + sizeof(*hal));
2411                 RETURN(-EPROTO);
2412         }
2413         if (lh->kuc_magic == __swab16(KUC_MAGIC)) {
2414                 lustre_swab_kuch(lh);
2415                 lustre_swab_hal(hal);
2416         } else if (lh->kuc_magic != KUC_MAGIC) {
2417                 CERROR("Bad magic %x!=%x\n", lh->kuc_magic, KUC_MAGIC);
2418                 RETURN(-EPROTO);
2419         }
2420
2421         CDEBUG(D_HSM, " Received message mg=%x t=%d m=%d l=%d actions=%d "
2422                "on %s\n",
2423                lh->kuc_magic, lh->kuc_transport, lh->kuc_msgtype,
2424                lh->kuc_msglen, hal->hal_count, hal->hal_fsname);
2425
2426         /* Broadcast to HSM listeners */
2427         rc = libcfs_kkuc_group_put(uuid, KUC_GRP_HSM, lh);
2428
2429         RETURN(rc);
2430 }
2431
2432 /**
2433  * callback function passed to kuc for re-registering each HSM copytool
2434  * running on MDC, after MDT shutdown/recovery.
2435  * @param data copytool registration data
2436  * @param cb_arg callback argument (obd_import)
2437  */
2438 static int mdc_hsm_ct_reregister(void *data, void *cb_arg)
2439 {
2440         struct obd_import *imp = (struct obd_import *)cb_arg;
2441         struct kkuc_ct_data *kcd = data;
2442         __u32 *archives = NULL;
2443         int rc;
2444
2445         if (kcd == NULL ||
2446             (kcd->kcd_magic != KKUC_CT_DATA_ARRAY_MAGIC &&
2447              kcd->kcd_magic != KKUC_CT_DATA_BITMAP_MAGIC))
2448                 return -EPROTO;
2449
2450         if (kcd->kcd_magic == KKUC_CT_DATA_BITMAP_MAGIC) {
2451                 CDEBUG(D_HA, "%s: recover copytool registration to MDT "
2452                        "(archive=%#x)\n", imp->imp_obd->obd_name,
2453                        kcd->kcd_nr_archives);
2454         } else {
2455                 CDEBUG(D_HA, "%s: recover copytool registration to MDT "
2456                        "(archive nr = %u)\n",
2457                        imp->imp_obd->obd_name, kcd->kcd_nr_archives);
2458                 if (kcd->kcd_nr_archives != 0)
2459                         archives = kcd->kcd_archives;
2460         }
2461
2462         rc = mdc_ioc_hsm_ct_register(imp, kcd->kcd_nr_archives, archives);
2463         /* ignore error if the copytool is already registered */
2464         return (rc == -EEXIST) ? 0 : rc;
2465 }
2466
2467 /**
2468  * Re-establish all kuc contexts with MDT
2469  * after MDT shutdown/recovery.
2470  */
2471 static int mdc_kuc_reregister(struct obd_import *imp)
2472 {
2473         /* re-register HSM agents */
2474         return libcfs_kkuc_group_foreach(&imp->imp_obd->obd_uuid, KUC_GRP_HSM,
2475                                          mdc_hsm_ct_reregister, imp);
2476 }
2477
2478 static int mdc_set_info_async(const struct lu_env *env,
2479                               struct obd_export *exp,
2480                               u32 keylen, void *key,
2481                               u32 vallen, void *val,
2482                               struct ptlrpc_request_set *set)
2483 {
2484         struct obd_import       *imp = class_exp2cliimp(exp);
2485         int                      rc;
2486         ENTRY;
2487
2488         if (KEY_IS(KEY_READ_ONLY)) {
2489                 if (vallen != sizeof(int))
2490                         RETURN(-EINVAL);
2491
2492                 spin_lock(&imp->imp_lock);
2493                 if (*((int *)val)) {
2494                         imp->imp_connect_flags_orig |= OBD_CONNECT_RDONLY;
2495                         imp->imp_connect_data.ocd_connect_flags |=
2496                                                         OBD_CONNECT_RDONLY;
2497                 } else {
2498                         imp->imp_connect_flags_orig &= ~OBD_CONNECT_RDONLY;
2499                         imp->imp_connect_data.ocd_connect_flags &=
2500                                                         ~OBD_CONNECT_RDONLY;
2501                 }
2502                 spin_unlock(&imp->imp_lock);
2503
2504                 rc = do_set_info_async(imp, MDS_SET_INFO, LUSTRE_MDS_VERSION,
2505                                        keylen, key, vallen, val, set);
2506                 RETURN(rc);
2507         }
2508         if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
2509                 rc = do_set_info_async(imp, MDS_SET_INFO, LUSTRE_MDS_VERSION,
2510                                        keylen, key, vallen, val, set);
2511                 RETURN(rc);
2512         }
2513         if (KEY_IS(KEY_HSM_COPYTOOL_SEND)) {
2514                 rc = mdc_hsm_copytool_send(&imp->imp_obd->obd_uuid, vallen,
2515                                            val);
2516                 RETURN(rc);
2517         }
2518
2519         if (KEY_IS(KEY_DEFAULT_EASIZE)) {
2520                 __u32 *default_easize = val;
2521
2522                 exp->exp_obd->u.cli.cl_default_mds_easize = *default_easize;
2523                 RETURN(0);
2524         }
2525
2526         rc = osc_set_info_async(env, exp, keylen, key, vallen, val, set);
2527         RETURN(rc);
2528 }
2529
2530 static int mdc_get_info(const struct lu_env *env, struct obd_export *exp,
2531                         __u32 keylen, void *key, __u32 *vallen, void *val)
2532 {
2533         int rc = -EINVAL;
2534
2535         if (KEY_IS(KEY_MAX_EASIZE)) {
2536                 __u32 mdsize, *max_easize;
2537
2538                 if (*vallen != sizeof(int))
2539                         RETURN(-EINVAL);
2540                 mdsize = *(__u32 *)val;
2541                 if (mdsize > exp->exp_obd->u.cli.cl_max_mds_easize)
2542                         exp->exp_obd->u.cli.cl_max_mds_easize = mdsize;
2543                 max_easize = val;
2544                 *max_easize = exp->exp_obd->u.cli.cl_max_mds_easize;
2545                 RETURN(0);
2546         } else if (KEY_IS(KEY_DEFAULT_EASIZE)) {
2547                 __u32 *default_easize;
2548
2549                 if (*vallen != sizeof(int))
2550                         RETURN(-EINVAL);
2551                 default_easize = val;
2552                 *default_easize = exp->exp_obd->u.cli.cl_default_mds_easize;
2553                 RETURN(0);
2554         } else if (KEY_IS(KEY_CONN_DATA)) {
2555                 struct obd_import *imp = class_exp2cliimp(exp);
2556                 struct obd_connect_data *data = val;
2557
2558                 if (*vallen != sizeof(*data))
2559                         RETURN(-EINVAL);
2560
2561                 *data = imp->imp_connect_data;
2562                 RETURN(0);
2563         } else if (KEY_IS(KEY_TGT_COUNT)) {
2564                 *((__u32 *)val) = 1;
2565                 RETURN(0);
2566         }
2567
2568         rc = mdc_get_info_rpc(exp, keylen, key, *vallen, val);
2569
2570         RETURN(rc);
2571 }
2572
2573 static int mdc_fsync(struct obd_export *exp, const struct lu_fid *fid,
2574                      struct ptlrpc_request **request)
2575 {
2576         struct ptlrpc_request *req;
2577         int                    rc;
2578         ENTRY;
2579
2580         *request = NULL;
2581         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_SYNC);
2582         if (req == NULL)
2583                 RETURN(-ENOMEM);
2584
2585         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_SYNC);
2586         if (rc) {
2587                 ptlrpc_request_free(req);
2588                 RETURN(rc);
2589         }
2590
2591         mdc_pack_body(&req->rq_pill, fid, 0, 0, -1, 0);
2592
2593         ptlrpc_request_set_replen(req);
2594
2595         rc = ptlrpc_queue_wait(req);
2596         if (rc)
2597                 ptlrpc_req_finished(req);
2598         else
2599                 *request = req;
2600         RETURN(rc);
2601 }
2602
2603 struct mdc_rmfid_args {
2604         int *mra_rcs;
2605         int mra_nr;
2606 };
2607
2608 int mdc_rmfid_interpret(const struct lu_env *env, struct ptlrpc_request *req,
2609                           void *args, int rc)
2610 {
2611         struct mdc_rmfid_args *aa;
2612         int *rcs, size;
2613         ENTRY;
2614
2615         if (!rc) {
2616                 aa = ptlrpc_req_async_args(aa, req);
2617
2618                 size = req_capsule_get_size(&req->rq_pill, &RMF_RCS,
2619                                             RCL_SERVER);
2620                 LASSERT(size == sizeof(int) * aa->mra_nr);
2621                 rcs = req_capsule_server_get(&req->rq_pill, &RMF_RCS);
2622                 LASSERT(rcs);
2623                 LASSERT(aa->mra_rcs);
2624                 LASSERT(aa->mra_nr);
2625                 memcpy(aa->mra_rcs, rcs, size);
2626         }
2627
2628         RETURN(rc);
2629 }
2630
2631 static int mdc_rmfid(struct obd_export *exp, struct fid_array *fa,
2632                      int *rcs, struct ptlrpc_request_set *set)
2633 {
2634         struct ptlrpc_request *req;
2635         struct mdc_rmfid_args *aa;
2636         struct mdt_body *b;
2637         struct lu_fid *tmp;
2638         int rc, flen;
2639         ENTRY;
2640
2641         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_RMFID);
2642         if (req == NULL)
2643                 RETURN(-ENOMEM);
2644
2645         flen = fa->fa_nr * sizeof(struct lu_fid);
2646         req_capsule_set_size(&req->rq_pill, &RMF_FID_ARRAY,
2647                              RCL_CLIENT, flen);
2648         req_capsule_set_size(&req->rq_pill, &RMF_FID_ARRAY,
2649                              RCL_SERVER, flen);
2650         req_capsule_set_size(&req->rq_pill, &RMF_RCS,
2651                              RCL_SERVER, fa->fa_nr * sizeof(__u32));
2652         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_RMFID);
2653         if (rc) {
2654                 ptlrpc_request_free(req);
2655                 RETURN(rc);
2656         }
2657         tmp = req_capsule_client_get(&req->rq_pill, &RMF_FID_ARRAY);
2658         memcpy(tmp, fa->fa_fids, flen);
2659
2660         mdc_pack_body(&req->rq_pill, NULL, 0, 0, -1, 0);
2661         b = req_capsule_client_get(&req->rq_pill, &RMF_MDT_BODY);
2662         b->mbo_ctime = ktime_get_real_seconds();
2663
2664         ptlrpc_request_set_replen(req);
2665
2666         LASSERT(rcs);
2667         aa = ptlrpc_req_async_args(aa, req);
2668         aa->mra_rcs = rcs;
2669         aa->mra_nr = fa->fa_nr;
2670         req->rq_interpret_reply = mdc_rmfid_interpret;
2671
2672         ptlrpc_set_add_req(set, req);
2673         ptlrpc_check_set(NULL, set);
2674
2675         RETURN(rc);
2676 }
2677
2678 static int mdc_import_event(struct obd_device *obd, struct obd_import *imp,
2679                             enum obd_import_event event)
2680 {
2681         struct client_obd *cli = &obd->u.cli;
2682         int rc = 0;
2683
2684         LASSERT(imp->imp_obd == obd);
2685
2686         switch (event) {
2687         case IMP_EVENT_DISCON:
2688                 spin_lock(&cli->cl_loi_list_lock);
2689                 cli->cl_avail_grant = 0;
2690                 cli->cl_lost_grant = 0;
2691                 spin_unlock(&cli->cl_loi_list_lock);
2692                 break;
2693         case IMP_EVENT_INACTIVE:
2694                 /*
2695                  * Flush current sequence to make client obtain new one
2696                  * from server in case of disconnect/reconnect.
2697                  */
2698                 down_read(&cli->cl_seq_rwsem);
2699                 if (cli->cl_seq)
2700                         seq_client_flush(cli->cl_seq);
2701                 up_read(&cli->cl_seq_rwsem);
2702
2703                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_INACTIVE);
2704                 break;
2705         case IMP_EVENT_INVALIDATE: {
2706                 struct ldlm_namespace *ns = obd->obd_namespace;
2707                 struct lu_env *env;
2708                 __u16 refcheck;
2709
2710                 ldlm_namespace_cleanup(ns, LDLM_FL_LOCAL_ONLY);
2711
2712                 env = cl_env_get(&refcheck);
2713                 if (!IS_ERR(env)) {
2714                         /* Reset grants. All pages go to failing rpcs due to
2715                          * the invalid import.
2716                          */
2717                         osc_io_unplug(env, cli, NULL);
2718
2719                         cfs_hash_for_each_nolock(ns->ns_rs_hash,
2720                                                  osc_ldlm_resource_invalidate,
2721                                                  env, 0);
2722                         cl_env_put(env, &refcheck);
2723                         ldlm_namespace_cleanup(ns, LDLM_FL_LOCAL_ONLY);
2724                 } else {
2725                         rc = PTR_ERR(env);
2726                 }
2727                 break;
2728         }
2729         case IMP_EVENT_ACTIVE:
2730                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_ACTIVE);
2731                 /* redo the kuc registration after reconnecting */
2732                 if (rc == 0)
2733                         rc = mdc_kuc_reregister(imp);
2734                 break;
2735         case IMP_EVENT_OCD: {
2736                 struct obd_connect_data *ocd = &imp->imp_connect_data;
2737
2738                 if (OCD_HAS_FLAG(ocd, GRANT))
2739                         osc_init_grant(cli, ocd);
2740
2741                 md_init_ea_size(obd->obd_self_export, ocd->ocd_max_easize, 0);
2742                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_OCD);
2743                 break;
2744         }
2745         case IMP_EVENT_DEACTIVATE:
2746         case IMP_EVENT_ACTIVATE:
2747                 break;
2748         default:
2749                 CERROR("Unknown import event %x\n", event);
2750                 LBUG();
2751         }
2752         RETURN(rc);
2753 }
2754
2755 int mdc_fid_alloc(const struct lu_env *env, struct obd_export *exp,
2756                   struct lu_fid *fid, struct md_op_data *op_data)
2757 {
2758         struct client_obd *cli = &exp->exp_obd->u.cli;
2759         int rc = -EIO;
2760
2761         ENTRY;
2762
2763         down_read(&cli->cl_seq_rwsem);
2764         if (cli->cl_seq)
2765                 rc = seq_client_alloc_fid(env, cli->cl_seq, fid);
2766         up_read(&cli->cl_seq_rwsem);
2767
2768         RETURN(rc);
2769 }
2770
2771 static struct obd_uuid *mdc_get_uuid(struct obd_export *exp)
2772 {
2773         struct client_obd *cli = &exp->exp_obd->u.cli;
2774         return &cli->cl_target_uuid;
2775 }
2776
2777 /**
2778  * Determine whether the lock can be canceled before replaying it during
2779  * recovery, non zero value will be return if the lock can be canceled,
2780  * or zero returned for not
2781  */
2782 static int mdc_cancel_weight(struct ldlm_lock *lock)
2783 {
2784         if (lock->l_resource->lr_type != LDLM_IBITS)
2785                 RETURN(0);
2786
2787         /* FIXME: if we ever get into a situation where there are too many
2788          * opened files with open locks on a single node, then we really
2789          * should replay these open locks to reget it */
2790         if (lock->l_policy_data.l_inodebits.bits & MDS_INODELOCK_OPEN)
2791                 RETURN(0);
2792
2793         /* Special case for DoM locks, cancel only unused and granted locks */
2794         if (ldlm_has_dom(lock) &&
2795             (lock->l_granted_mode != lock->l_req_mode ||
2796              osc_ldlm_weigh_ast(lock) != 0))
2797                 RETURN(0);
2798
2799         RETURN(1);
2800 }
2801
2802 static int mdc_resource_inode_free(struct ldlm_resource *res)
2803 {
2804         if (res->lr_lvb_inode)
2805                 res->lr_lvb_inode = NULL;
2806
2807         return 0;
2808 }
2809
2810 static struct ldlm_valblock_ops inode_lvbo = {
2811         .lvbo_free = mdc_resource_inode_free
2812 };
2813
2814 static int mdc_llog_init(struct obd_device *obd)
2815 {
2816         struct obd_llog_group   *olg = &obd->obd_olg;
2817         struct llog_ctxt        *ctxt;
2818         int                      rc;
2819
2820         ENTRY;
2821
2822         rc = llog_setup(NULL, obd, olg, LLOG_CHANGELOG_REPL_CTXT, obd,
2823                         &llog_client_ops);
2824         if (rc < 0)
2825                 RETURN(rc);
2826
2827         ctxt = llog_group_get_ctxt(olg, LLOG_CHANGELOG_REPL_CTXT);
2828         llog_initiator_connect(ctxt);
2829         llog_ctxt_put(ctxt);
2830
2831         RETURN(0);
2832 }
2833
2834 static void mdc_llog_finish(struct obd_device *obd)
2835 {
2836         struct llog_ctxt *ctxt;
2837
2838         ENTRY;
2839
2840         ctxt = llog_get_context(obd, LLOG_CHANGELOG_REPL_CTXT);
2841         if (ctxt != NULL)
2842                 llog_cleanup(NULL, ctxt);
2843
2844         EXIT;
2845 }
2846
2847 int mdc_setup(struct obd_device *obd, struct lustre_cfg *cfg)
2848 {
2849         int rc;
2850
2851         ENTRY;
2852
2853         rc = osc_setup_common(obd, cfg);
2854         if (rc < 0)
2855                 RETURN(rc);
2856
2857         rc = mdc_tunables_init(obd);
2858         if (rc)
2859                 GOTO(err_osc_cleanup, rc);
2860
2861         obd->u.cli.cl_dom_min_inline_repsize = MDC_DOM_DEF_INLINE_REPSIZE;
2862         obd->u.cli.cl_lsom_update = true;
2863
2864         ns_register_cancel(obd->obd_namespace, mdc_cancel_weight);
2865
2866         obd->obd_namespace->ns_lvbo = &inode_lvbo;
2867
2868         rc = mdc_llog_init(obd);
2869         if (rc) {
2870                 CERROR("%s: failed to setup llogging subsystems: rc = %d\n",
2871                        obd->obd_name, rc);
2872                 GOTO(err_llog_cleanup, rc);
2873         }
2874
2875         rc = mdc_changelog_cdev_init(obd);
2876         if (rc) {
2877                 CERROR("%s: failed to setup changelog char device: rc = %d\n",
2878                        obd->obd_name, rc);
2879                 GOTO(err_changelog_cleanup, rc);
2880         }
2881
2882         RETURN(rc);
2883
2884 err_changelog_cleanup:
2885         mdc_llog_finish(obd);
2886 err_llog_cleanup:
2887         lprocfs_free_md_stats(obd);
2888         ptlrpc_lprocfs_unregister_obd(obd);
2889 err_osc_cleanup:
2890         osc_cleanup_common(obd);
2891         return rc;
2892 }
2893
2894 /* Initialize the default and maximum LOV EA sizes.  This allows
2895  * us to make MDS RPCs with large enough reply buffers to hold a default
2896  * sized EA without having to calculate this (via a call into the
2897  * LOV + OSCs) each time we make an RPC.  The maximum size is also tracked
2898  * but not used to avoid wastefully vmalloc()'ing large reply buffers when
2899  * a large number of stripes is possible.  If a larger reply buffer is
2900  * required it will be reallocated in the ptlrpc layer due to overflow.
2901  */
2902 static int mdc_init_ea_size(struct obd_export *exp, __u32 easize,
2903                             __u32 def_easize)
2904 {
2905         struct obd_device *obd = exp->exp_obd;
2906         struct client_obd *cli = &obd->u.cli;
2907         ENTRY;
2908
2909         if (cli->cl_max_mds_easize < easize)
2910                 cli->cl_max_mds_easize = easize;
2911
2912         if (cli->cl_default_mds_easize < def_easize)
2913                 cli->cl_default_mds_easize = def_easize;
2914
2915         RETURN(0);
2916 }
2917
2918 static int mdc_precleanup(struct obd_device *obd)
2919 {
2920         ENTRY;
2921
2922         osc_precleanup_common(obd);
2923         mdc_changelog_cdev_finish(obd);
2924
2925         obd_cleanup_client_import(obd);
2926         ptlrpc_lprocfs_unregister_obd(obd);
2927         lprocfs_free_md_stats(obd);
2928         mdc_llog_finish(obd);
2929         RETURN(0);
2930 }
2931
2932 static int mdc_cleanup(struct obd_device *obd)
2933 {
2934         return osc_cleanup_common(obd);
2935 }
2936
2937 static const struct obd_ops mdc_obd_ops = {
2938         .o_owner            = THIS_MODULE,
2939         .o_setup            = mdc_setup,
2940         .o_precleanup       = mdc_precleanup,
2941         .o_cleanup          = mdc_cleanup,
2942         .o_add_conn         = client_import_add_conn,
2943         .o_del_conn         = client_import_del_conn,
2944         .o_connect          = client_connect_import,
2945         .o_reconnect        = osc_reconnect,
2946         .o_disconnect       = osc_disconnect,
2947         .o_iocontrol        = mdc_iocontrol,
2948         .o_set_info_async   = mdc_set_info_async,
2949         .o_statfs           = mdc_statfs,
2950         .o_statfs_async     = mdc_statfs_async,
2951         .o_fid_init         = client_fid_init,
2952         .o_fid_fini         = client_fid_fini,
2953         .o_fid_alloc        = mdc_fid_alloc,
2954         .o_import_event     = mdc_import_event,
2955         .o_get_info         = mdc_get_info,
2956         .o_get_uuid         = mdc_get_uuid,
2957         .o_quotactl         = mdc_quotactl,
2958 };
2959
2960 static const struct md_ops mdc_md_ops = {
2961         .m_get_root         = mdc_get_root,
2962         .m_null_inode       = mdc_null_inode,
2963         .m_close            = mdc_close,
2964         .m_create           = mdc_create,
2965         .m_enqueue          = mdc_enqueue,
2966         .m_getattr          = mdc_getattr,
2967         .m_getattr_name     = mdc_getattr_name,
2968         .m_intent_lock      = mdc_intent_lock,
2969         .m_link             = mdc_link,
2970         .m_rename           = mdc_rename,
2971         .m_setattr          = mdc_setattr,
2972         .m_setxattr         = mdc_setxattr,
2973         .m_getxattr         = mdc_getxattr,
2974         .m_fsync                = mdc_fsync,
2975         .m_file_resync          = mdc_file_resync,
2976         .m_read_page            = mdc_read_page,
2977         .m_unlink           = mdc_unlink,
2978         .m_cancel_unused    = mdc_cancel_unused,
2979         .m_init_ea_size     = mdc_init_ea_size,
2980         .m_set_lock_data    = mdc_set_lock_data,
2981         .m_lock_match       = mdc_lock_match,
2982         .m_get_lustre_md    = mdc_get_lustre_md,
2983         .m_free_lustre_md   = mdc_free_lustre_md,
2984         .m_set_open_replay_data = mdc_set_open_replay_data,
2985         .m_clear_open_replay_data = mdc_clear_open_replay_data,
2986         .m_intent_getattr_async = mdc_intent_getattr_async,
2987         .m_revalidate_lock      = mdc_revalidate_lock,
2988         .m_rmfid                = mdc_rmfid,
2989 };
2990
2991 dev_t mdc_changelog_dev;
2992 struct class *mdc_changelog_class;
2993 static int __init mdc_init(void)
2994 {
2995         int rc = 0;
2996         rc = alloc_chrdev_region(&mdc_changelog_dev, 0,
2997                                  MDC_CHANGELOG_DEV_COUNT,
2998                                  MDC_CHANGELOG_DEV_NAME);
2999         if (rc)
3000                 return rc;
3001
3002         mdc_changelog_class = class_create(THIS_MODULE, MDC_CHANGELOG_DEV_NAME);
3003         if (IS_ERR(mdc_changelog_class)) {
3004                 rc = PTR_ERR(mdc_changelog_class);
3005                 goto out_dev;
3006         }
3007
3008         rc = class_register_type(&mdc_obd_ops, &mdc_md_ops, true,
3009                                  LUSTRE_MDC_NAME, &mdc_device_type);
3010         if (rc)
3011                 goto out_class;
3012
3013         return 0;
3014
3015 out_class:
3016         class_destroy(mdc_changelog_class);
3017 out_dev:
3018         unregister_chrdev_region(mdc_changelog_dev, MDC_CHANGELOG_DEV_COUNT);
3019         return rc;
3020 }
3021
3022 static void __exit mdc_exit(void)
3023 {
3024         class_unregister_type(LUSTRE_MDC_NAME);
3025         class_destroy(mdc_changelog_class);
3026         unregister_chrdev_region(mdc_changelog_dev, MDC_CHANGELOG_DEV_COUNT);
3027         idr_destroy(&mdc_changelog_minor_idr);
3028 }
3029
3030 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
3031 MODULE_DESCRIPTION("Lustre Metadata Client");
3032 MODULE_VERSION(LUSTRE_VERSION_STRING);
3033 MODULE_LICENSE("GPL");
3034
3035 module_init(mdc_init);
3036 module_exit(mdc_exit);