lustre/lmv/lmv_obd.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2004, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2011, 2014, Intel Corporation.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  */
  36
  37 #define DEBUG_SUBSYSTEM S_LMV
  38 #include <linux/slab.h>
  39 #include <linux/module.h>
  40 #include <linux/init.h>
  41 #include <linux/slab.h>
  42 #include <linux/pagemap.h>
  43 #include <linux/mm.h>
  44 #include <linux/math64.h>
  45 #include <linux/seq_file.h>
  46 #include <linux/namei.h>
  47
  48 #include <lustre/lustre_idl.h>
  49 #include <obd_support.h>
  50 #include <lustre_lib.h>
  51 #include <lustre_net.h>
  52 #include <obd_class.h>
  53 #include <lustre_lmv.h>
  54 #include <lprocfs_status.h>
  55 #include <cl_object.h>
  56 #include <lustre_fid.h>
  57 #include <lustre_ioctl.h>
  58 #include "lmv_internal.h"
  59
  60 static void lmv_activate_target(struct lmv_obd *lmv,
  61                                 struct lmv_tgt_desc *tgt,
  62                                 int activate)
  63 {
  64         if (tgt->ltd_active == activate)
  65                 return;
  66
  67         tgt->ltd_active = activate;
  68         lmv->desc.ld_active_tgt_count += (activate ? 1 : -1);
  69 }
  70
  71 /**
  72  * Error codes:
  73  *
  74  *  -EINVAL  : UUID can't be found in the LMV's target list
  75  *  -ENOTCONN: The UUID is found, but the target connection is bad (!)
  76  *  -EBADF   : The UUID is found, but the OBD of the wrong type (!)
  77  */
  78 static int lmv_set_mdc_active(struct lmv_obd *lmv,
  79                               const struct obd_uuid *uuid,
  80                               int activate)
  81 {
  82         struct lmv_tgt_desc     *tgt = NULL;
  83         struct obd_device       *obd;
  84         __u32                    i;
  85         int                      rc = 0;
  86         ENTRY;
  87
  88         CDEBUG(D_INFO, "Searching in lmv %p for uuid %s (activate=%d)\n",
  89                         lmv, uuid->uuid, activate);
  90
  91         spin_lock(&lmv->lmv_lock);
  92         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
  93                 tgt = lmv->tgts[i];
  94                 if (tgt == NULL || tgt->ltd_exp == NULL)
  95                         continue;
  96
  97                 CDEBUG(D_INFO, "Target idx %d is %s conn "LPX64"\n", i,
  98                        tgt->ltd_uuid.uuid, tgt->ltd_exp->exp_handle.h_cookie);
  99
 100                 if (obd_uuid_equals(uuid, &tgt->ltd_uuid))
 101                         break;
 102         }
 103
 104         if (i == lmv->desc.ld_tgt_count)
 105                 GOTO(out_lmv_lock, rc = -EINVAL);
 106
 107         obd = class_exp2obd(tgt->ltd_exp);
 108         if (obd == NULL)
 109                 GOTO(out_lmv_lock, rc = -ENOTCONN);
 110
 111         CDEBUG(D_INFO, "Found OBD %s=%s device %d (%p) type %s at LMV idx %d\n",
 112                obd->obd_name, obd->obd_uuid.uuid, obd->obd_minor, obd,
 113                obd->obd_type->typ_name, i);
 114         LASSERT(strcmp(obd->obd_type->typ_name, LUSTRE_MDC_NAME) == 0);
 115
 116         if (tgt->ltd_active == activate) {
 117                 CDEBUG(D_INFO, "OBD %p already %sactive!\n", obd,
 118                        activate ? "" : "in");
 119                 GOTO(out_lmv_lock, rc);
 120         }
 121
 122         CDEBUG(D_INFO, "Marking OBD %p %sactive\n", obd,
 123                activate ? "" : "in");
 124         lmv_activate_target(lmv, tgt, activate);
 125         EXIT;
 126
 127  out_lmv_lock:
 128         spin_unlock(&lmv->lmv_lock);
 129         return rc;
 130 }
 131
 132 struct obd_uuid *lmv_get_uuid(struct obd_export *exp)
 133 {
 134         struct lmv_obd          *lmv = &exp->exp_obd->u.lmv;
 135         struct lmv_tgt_desc     *tgt = lmv->tgts[0];
 136
 137         return (tgt == NULL) ? NULL : obd_get_uuid(tgt->ltd_exp);
 138 }
 139
 140 static int lmv_notify(struct obd_device *obd, struct obd_device *watched,
 141                       enum obd_notify_event ev, void *data)
 142 {
 143         struct obd_connect_data *conn_data;
 144         struct lmv_obd          *lmv = &obd->u.lmv;
 145         struct obd_uuid         *uuid;
 146         int                      rc = 0;
 147         ENTRY;
 148
 149         if (strcmp(watched->obd_type->typ_name, LUSTRE_MDC_NAME)) {
 150                 CERROR("unexpected notification of %s %s!\n",
 151                        watched->obd_type->typ_name,
 152                        watched->obd_name);
 153                 RETURN(-EINVAL);
 154         }
 155
 156         uuid = &watched->u.cli.cl_target_uuid;
 157         if (ev == OBD_NOTIFY_ACTIVE || ev == OBD_NOTIFY_INACTIVE) {
 158                 /*
 159                  * Set MDC as active before notifying the observer, so the
 160                  * observer can use the MDC normally.
 161                  */
 162                 rc = lmv_set_mdc_active(lmv, uuid,
 163                                         ev == OBD_NOTIFY_ACTIVE);
 164                 if (rc) {
 165                         CERROR("%sactivation of %s failed: %d\n",
 166                                ev == OBD_NOTIFY_ACTIVE ? "" : "de",
 167                                uuid->uuid, rc);
 168                         RETURN(rc);
 169                 }
 170         } else if (ev == OBD_NOTIFY_OCD) {
 171                 conn_data = &watched->u.cli.cl_import->imp_connect_data;
 172                 /*
 173                  * XXX: Make sure that ocd_connect_flags from all targets are
 174                  * the same. Otherwise one of MDTs runs wrong version or
 175                  * something like this.  --umka
 176                  */
 177                 obd->obd_self_export->exp_connect_data = *conn_data;
 178         }
 179 #if 0
 180         else if (ev == OBD_NOTIFY_DISCON) {
 181                 /*
 182                  * For disconnect event, flush fld cache for failout MDS case.
 183                  */
 184                 fld_client_flush(&lmv->lmv_fld);
 185         }
 186 #endif
 187         /*
 188          * Pass the notification up the chain.
 189          */
 190         if (obd->obd_observer)
 191                 rc = obd_notify(obd->obd_observer, watched, ev, data);
 192
 193         RETURN(rc);
 194 }
 195
 196 /**
 197  * This is fake connect function. Its purpose is to initialize lmv and say
 198  * caller that everything is okay. Real connection will be performed later.
 199  */
 200 static int lmv_connect(const struct lu_env *env,
 201                        struct obd_export **exp, struct obd_device *obd,
 202                        struct obd_uuid *cluuid, struct obd_connect_data *data,
 203                        void *localdata)
 204 {
 205         struct lmv_obd        *lmv = &obd->u.lmv;
 206         struct lustre_handle  conn = { 0 };
 207         int                    rc = 0;
 208         ENTRY;
 209
 210         /*
 211          * We don't want to actually do the underlying connections more than
 212          * once, so keep track.
 213          */
 214         lmv->refcount++;
 215         if (lmv->refcount > 1) {
 216                 *exp = NULL;
 217                 RETURN(0);
 218         }
 219
 220         rc = class_connect(&conn, obd, cluuid);
 221         if (rc) {
 222                 CERROR("class_connection() returned %d\n", rc);
 223                 RETURN(rc);
 224         }
 225
 226         *exp = class_conn2export(&conn);
 227         class_export_get(*exp);
 228
 229         lmv->exp = *exp;
 230         lmv->connected = 0;
 231         lmv->cluuid = *cluuid;
 232
 233         if (data)
 234                 lmv->conn_data = *data;
 235
 236         if (lmv->targets_proc_entry == NULL) {
 237                 lmv->targets_proc_entry = lprocfs_register("target_obds",
 238                                                            obd->obd_proc_entry,
 239                                                            NULL, NULL);
 240                 if (IS_ERR(lmv->targets_proc_entry)) {
 241                         CERROR("%s: cannot register "
 242                                "/proc/fs/lustre/%s/%s/target_obds\n",
 243                                obd->obd_name, obd->obd_type->typ_name,
 244                                obd->obd_name);
 245                         lmv->targets_proc_entry = NULL;
 246                 }
 247         }
 248
 249         /*
 250          * All real clients should perform actual connection right away, because
 251          * it is possible, that LMV will not have opportunity to connect targets
 252          * and MDC stuff will be called directly, for instance while reading
 253          * ../mdc/../kbytesfree procfs file, etc.
 254          */
 255         if (data != NULL && (data->ocd_connect_flags & OBD_CONNECT_REAL))
 256                 rc = lmv_check_connect(obd);
 257
 258         if (rc && lmv->targets_proc_entry != NULL)
 259                 lprocfs_remove(&lmv->targets_proc_entry);
 260         RETURN(rc);
 261 }
 262
 263 static int lmv_init_ea_size(struct obd_export *exp,
 264                             __u32 easize, __u32 def_easize,
 265                             __u32 cookiesize, __u32 def_cookiesize)
 266 {
 267         struct obd_device       *obd = exp->exp_obd;
 268         struct lmv_obd          *lmv = &obd->u.lmv;
 269         __u32                    i;
 270         int                      rc = 0;
 271         int                      change = 0;
 272         ENTRY;
 273
 274         if (lmv->max_easize < easize) {
 275                 lmv->max_easize = easize;
 276                 change = 1;
 277         }
 278         if (lmv->max_def_easize < def_easize) {
 279                 lmv->max_def_easize = def_easize;
 280                 change = 1;
 281         }
 282         if (lmv->max_cookiesize < cookiesize) {
 283                 lmv->max_cookiesize = cookiesize;
 284                 change = 1;
 285         }
 286         if (lmv->max_def_cookiesize < def_cookiesize) {
 287                 lmv->max_def_cookiesize = def_cookiesize;
 288                 change = 1;
 289         }
 290         if (change == 0)
 291                 RETURN(0);
 292
 293         if (lmv->connected == 0)
 294                 RETURN(0);
 295
 296         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 297                 struct lmv_tgt_desc *tgt = lmv->tgts[i];
 298
 299                 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active) {
 300                         CWARN("%s: NULL export for %d\n", obd->obd_name, i);
 301                         continue;
 302                 }
 303
 304                 rc = md_init_ea_size(tgt->ltd_exp, easize, def_easize,
 305                                      cookiesize, def_cookiesize);
 306                 if (rc) {
 307                         CERROR("%s: obd_init_ea_size() failed on MDT target %d:"
 308                                " rc = %d\n", obd->obd_name, i, rc);
 309                         break;
 310                 }
 311         }
 312         RETURN(rc);
 313 }
 314
 315 #define MAX_STRING_SIZE 128
 316
 317 int lmv_connect_mdc(struct obd_device *obd, struct lmv_tgt_desc *tgt)
 318 {
 319         struct lmv_obd          *lmv = &obd->u.lmv;
 320         struct obd_uuid         *cluuid = &lmv->cluuid;
 321         struct obd_uuid          lmv_mdc_uuid = { "LMV_MDC_UUID" };
 322         struct obd_device       *mdc_obd;
 323         struct obd_export       *mdc_exp;
 324         struct lu_fld_target     target;
 325         int                      rc;
 326         ENTRY;
 327
 328         mdc_obd = class_find_client_obd(&tgt->ltd_uuid, LUSTRE_MDC_NAME,
 329                                         &obd->obd_uuid);
 330         if (!mdc_obd) {
 331                 CERROR("target %s not attached\n", tgt->ltd_uuid.uuid);
 332                 RETURN(-EINVAL);
 333         }
 334
 335         CDEBUG(D_CONFIG, "connect to %s(%s) - %s, %s FOR %s\n",
 336                 mdc_obd->obd_name, mdc_obd->obd_uuid.uuid,
 337                 tgt->ltd_uuid.uuid, obd->obd_uuid.uuid,
 338                 cluuid->uuid);
 339
 340         if (!mdc_obd->obd_set_up) {
 341                 CERROR("target %s is not set up\n", tgt->ltd_uuid.uuid);
 342                 RETURN(-EINVAL);
 343         }
 344
 345         rc = obd_connect(NULL, &mdc_exp, mdc_obd, &lmv_mdc_uuid,
 346                          &lmv->conn_data, NULL);
 347         if (rc) {
 348                 CERROR("target %s connect error %d\n", tgt->ltd_uuid.uuid, rc);
 349                 RETURN(rc);
 350         }
 351
 352         /*
 353          * Init fid sequence client for this mdc and add new fld target.
 354          */
 355         rc = obd_fid_init(mdc_obd, mdc_exp, LUSTRE_SEQ_METADATA);
 356         if (rc)
 357                 RETURN(rc);
 358
 359         target.ft_srv = NULL;
 360         target.ft_exp = mdc_exp;
 361         target.ft_idx = tgt->ltd_idx;
 362
 363         fld_client_add_target(&lmv->lmv_fld, &target);
 364
 365         rc = obd_register_observer(mdc_obd, obd);
 366         if (rc) {
 367                 obd_disconnect(mdc_exp);
 368                 CERROR("target %s register_observer error %d\n",
 369                        tgt->ltd_uuid.uuid, rc);
 370                 RETURN(rc);
 371         }
 372
 373         if (obd->obd_observer) {
 374                 /*
 375                  * Tell the observer about the new target.
 376                  */
 377                 rc = obd_notify(obd->obd_observer, mdc_exp->exp_obd,
 378                                 OBD_NOTIFY_ACTIVE,
 379                                 (void *)(tgt - lmv->tgts[0]));
 380                 if (rc) {
 381                         obd_disconnect(mdc_exp);
 382                         RETURN(rc);
 383                 }
 384         }
 385
 386         tgt->ltd_active = 1;
 387         tgt->ltd_exp = mdc_exp;
 388         lmv->desc.ld_active_tgt_count++;
 389
 390         md_init_ea_size(tgt->ltd_exp, lmv->max_easize, lmv->max_def_easize,
 391                         lmv->max_cookiesize, lmv->max_def_cookiesize);
 392
 393         CDEBUG(D_CONFIG, "Connected to %s(%s) successfully (%d)\n",
 394                 mdc_obd->obd_name, mdc_obd->obd_uuid.uuid,
 395                 atomic_read(&obd->obd_refcount));
 396
 397         if (lmv->targets_proc_entry != NULL) {
 398                 struct proc_dir_entry *mdc_symlink;
 399
 400                 LASSERT(mdc_obd->obd_type != NULL);
 401                 LASSERT(mdc_obd->obd_type->typ_name != NULL);
 402                 mdc_symlink = lprocfs_add_symlink(mdc_obd->obd_name,
 403                                                   lmv->targets_proc_entry,
 404                                                   "../../../%s/%s",
 405                                                   mdc_obd->obd_type->typ_name,
 406                                                   mdc_obd->obd_name);
 407                 if (mdc_symlink == NULL) {
 408                         CERROR("cannot register LMV target "
 409                                "/proc/fs/lustre/%s/%s/target_obds/%s\n",
 410                                obd->obd_type->typ_name, obd->obd_name,
 411                                mdc_obd->obd_name);
 412                 }
 413         }
 414         RETURN(0);
 415 }
 416
 417 static void lmv_del_target(struct lmv_obd *lmv, int index)
 418 {
 419         if (lmv->tgts[index] == NULL)
 420                 return;
 421
 422         OBD_FREE_PTR(lmv->tgts[index]);
 423         lmv->tgts[index] = NULL;
 424         return;
 425 }
 426
 427 static int lmv_add_target(struct obd_device *obd, struct obd_uuid *uuidp,
 428                            __u32 index, int gen)
 429 {
 430         struct lmv_obd      *lmv = &obd->u.lmv;
 431         struct lmv_tgt_desc *tgt;
 432         int                  orig_tgt_count = 0;
 433         int                  rc = 0;
 434         ENTRY;
 435
 436         CDEBUG(D_CONFIG, "Target uuid: %s. index %d\n", uuidp->uuid, index);
 437
 438         mutex_lock(&lmv->lmv_init_mutex);
 439
 440         if (lmv->desc.ld_tgt_count == 0) {
 441                 struct obd_device *mdc_obd;
 442
 443                 mdc_obd = class_find_client_obd(uuidp, LUSTRE_MDC_NAME,
 444                                                 &obd->obd_uuid);
 445                 if (!mdc_obd) {
 446                         mutex_unlock(&lmv->lmv_init_mutex);
 447                         CERROR("%s: Target %s not attached: rc = %d\n",
 448                                obd->obd_name, uuidp->uuid, -EINVAL);
 449                         RETURN(-EINVAL);
 450                 }
 451         }
 452
 453         if ((index < lmv->tgts_size) && (lmv->tgts[index] != NULL)) {
 454                 tgt = lmv->tgts[index];
 455                 CERROR("%s: UUID %s already assigned at LOV target index %d:"
 456                        " rc = %d\n", obd->obd_name,
 457                        obd_uuid2str(&tgt->ltd_uuid), index, -EEXIST);
 458                 mutex_unlock(&lmv->lmv_init_mutex);
 459                 RETURN(-EEXIST);
 460         }
 461
 462         if (index >= lmv->tgts_size) {
 463                 /* We need to reallocate the lmv target array. */
 464                 struct lmv_tgt_desc **newtgts, **old = NULL;
 465                 __u32 newsize = 1;
 466                 __u32 oldsize = 0;
 467
 468                 while (newsize < index + 1)
 469                         newsize = newsize << 1;
 470                 OBD_ALLOC(newtgts, sizeof(*newtgts) * newsize);
 471                 if (newtgts == NULL) {
 472                         mutex_unlock(&lmv->lmv_init_mutex);
 473                         RETURN(-ENOMEM);
 474                 }
 475
 476                 if (lmv->tgts_size) {
 477                         memcpy(newtgts, lmv->tgts,
 478                                sizeof(*newtgts) * lmv->tgts_size);
 479                         old = lmv->tgts;
 480                         oldsize = lmv->tgts_size;
 481                 }
 482
 483                 lmv->tgts = newtgts;
 484                 lmv->tgts_size = newsize;
 485                 smp_rmb();
 486                 if (old)
 487                         OBD_FREE(old, sizeof(*old) * oldsize);
 488
 489                 CDEBUG(D_CONFIG, "tgts: %p size: %d\n", lmv->tgts,
 490                        lmv->tgts_size);
 491         }
 492
 493         OBD_ALLOC_PTR(tgt);
 494         if (!tgt) {
 495                 mutex_unlock(&lmv->lmv_init_mutex);
 496                 RETURN(-ENOMEM);
 497         }
 498
 499         mutex_init(&tgt->ltd_fid_mutex);
 500         tgt->ltd_idx = index;
 501         tgt->ltd_uuid = *uuidp;
 502         tgt->ltd_active = 0;
 503         lmv->tgts[index] = tgt;
 504         if (index >= lmv->desc.ld_tgt_count) {
 505                 orig_tgt_count = lmv->desc.ld_tgt_count;
 506                 lmv->desc.ld_tgt_count = index + 1;
 507         }
 508
 509         if (lmv->connected) {
 510                 rc = lmv_connect_mdc(obd, tgt);
 511                 if (rc != 0) {
 512                         spin_lock(&lmv->lmv_lock);
 513                         if (lmv->desc.ld_tgt_count == index + 1)
 514                                 lmv->desc.ld_tgt_count = orig_tgt_count;
 515                         memset(tgt, 0, sizeof(*tgt));
 516                         spin_unlock(&lmv->lmv_lock);
 517                 } else {
 518                         int easize = sizeof(struct lmv_stripe_md) +
 519                                 lmv->desc.ld_tgt_count * sizeof(struct lu_fid);
 520                         lmv_init_ea_size(obd->obd_self_export, easize, 0, 0, 0);
 521                 }
 522         }
 523
 524         mutex_unlock(&lmv->lmv_init_mutex);
 525         RETURN(rc);
 526 }
 527
 528 int lmv_check_connect(struct obd_device *obd)
 529 {
 530         struct lmv_obd          *lmv = &obd->u.lmv;
 531         struct lmv_tgt_desc     *tgt;
 532         __u32                    i;
 533         int                      rc;
 534         int                      easize;
 535         ENTRY;
 536
 537         if (lmv->connected)
 538                 RETURN(0);
 539
 540         mutex_lock(&lmv->lmv_init_mutex);
 541         if (lmv->connected) {
 542                 mutex_unlock(&lmv->lmv_init_mutex);
 543                 RETURN(0);
 544         }
 545
 546         if (lmv->desc.ld_tgt_count == 0) {
 547                 mutex_unlock(&lmv->lmv_init_mutex);
 548                 CERROR("%s: no targets configured.\n", obd->obd_name);
 549                 RETURN(-EINVAL);
 550         }
 551
 552         LASSERT(lmv->tgts != NULL);
 553
 554         if (lmv->tgts[0] == NULL) {
 555                 mutex_unlock(&lmv->lmv_init_mutex);
 556                 CERROR("%s: no target configured for index 0.\n",
 557                        obd->obd_name);
 558                 RETURN(-EINVAL);
 559         }
 560
 561         CDEBUG(D_CONFIG, "Time to connect %s to %s\n",
 562                lmv->cluuid.uuid, obd->obd_name);
 563
 564         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 565                 tgt = lmv->tgts[i];
 566                 if (tgt == NULL)
 567                         continue;
 568                 rc = lmv_connect_mdc(obd, tgt);
 569                 if (rc)
 570                         GOTO(out_disc, rc);
 571         }
 572
 573         class_export_put(lmv->exp);
 574         lmv->connected = 1;
 575         easize = lmv_mds_md_size(lmv->desc.ld_tgt_count, LMV_MAGIC);
 576         lmv_init_ea_size(obd->obd_self_export, easize, 0, 0, 0);
 577         mutex_unlock(&lmv->lmv_init_mutex);
 578         RETURN(0);
 579
 580  out_disc:
 581         while (i-- > 0) {
 582                 int rc2;
 583                 tgt = lmv->tgts[i];
 584                 if (tgt == NULL)
 585                         continue;
 586                 tgt->ltd_active = 0;
 587                 if (tgt->ltd_exp) {
 588                         --lmv->desc.ld_active_tgt_count;
 589                         rc2 = obd_disconnect(tgt->ltd_exp);
 590                         if (rc2) {
 591                                 CERROR("LMV target %s disconnect on "
 592                                        "MDC idx %d: error %d\n",
 593                                        tgt->ltd_uuid.uuid, i, rc2);
 594                         }
 595                 }
 596         }
 597         class_disconnect(lmv->exp);
 598         mutex_unlock(&lmv->lmv_init_mutex);
 599         RETURN(rc);
 600 }
 601
 602 static int lmv_disconnect_mdc(struct obd_device *obd, struct lmv_tgt_desc *tgt)
 603 {
 604         struct lmv_obd         *lmv = &obd->u.lmv;
 605         struct obd_device      *mdc_obd;
 606         int                     rc;
 607         ENTRY;
 608
 609         LASSERT(tgt != NULL);
 610         LASSERT(obd != NULL);
 611
 612         mdc_obd = class_exp2obd(tgt->ltd_exp);
 613
 614         if (mdc_obd) {
 615                 mdc_obd->obd_force = obd->obd_force;
 616                 mdc_obd->obd_fail = obd->obd_fail;
 617                 mdc_obd->obd_no_recov = obd->obd_no_recov;
 618         }
 619
 620         if (lmv->targets_proc_entry != NULL)
 621                 lprocfs_remove_proc_entry(mdc_obd->obd_name,
 622                                           lmv->targets_proc_entry);
 623
 624         rc = obd_fid_fini(tgt->ltd_exp->exp_obd);
 625         if (rc)
 626                 CERROR("Can't finanize fids factory\n");
 627
 628         CDEBUG(D_INFO, "Disconnected from %s(%s) successfully\n",
 629                tgt->ltd_exp->exp_obd->obd_name,
 630                tgt->ltd_exp->exp_obd->obd_uuid.uuid);
 631
 632         obd_register_observer(tgt->ltd_exp->exp_obd, NULL);
 633         rc = obd_disconnect(tgt->ltd_exp);
 634         if (rc) {
 635                 if (tgt->ltd_active) {
 636                         CERROR("Target %s disconnect error %d\n",
 637                                tgt->ltd_uuid.uuid, rc);
 638                 }
 639         }
 640
 641         lmv_activate_target(lmv, tgt, 0);
 642         tgt->ltd_exp = NULL;
 643         RETURN(0);
 644 }
 645
 646 static int lmv_disconnect(struct obd_export *exp)
 647 {
 648         struct obd_device       *obd = class_exp2obd(exp);
 649         struct lmv_obd          *lmv = &obd->u.lmv;
 650         int                      rc;
 651         __u32                    i;
 652         ENTRY;
 653
 654         if (!lmv->tgts)
 655                 goto out_local;
 656
 657         /*
 658          * Only disconnect the underlying layers on the final disconnect.
 659          */
 660         lmv->refcount--;
 661         if (lmv->refcount != 0)
 662                 goto out_local;
 663
 664         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 665                 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
 666                         continue;
 667
 668                 lmv_disconnect_mdc(obd, lmv->tgts[i]);
 669         }
 670
 671         if (lmv->targets_proc_entry != NULL)
 672                 lprocfs_remove(&lmv->targets_proc_entry);
 673         else
 674                 CERROR("/proc/fs/lustre/%s/%s/target_obds missing\n",
 675                        obd->obd_type->typ_name, obd->obd_name);
 676
 677 out_local:
 678         /*
 679          * This is the case when no real connection is established by
 680          * lmv_check_connect().
 681          */
 682         if (!lmv->connected)
 683                 class_export_put(exp);
 684         rc = class_disconnect(exp);
 685         if (lmv->refcount == 0)
 686                 lmv->connected = 0;
 687         RETURN(rc);
 688 }
 689
 690 static int lmv_fid2path(struct obd_export *exp, int len, void *karg, void *uarg)
 691 {
 692         struct obd_device       *obddev = class_exp2obd(exp);
 693         struct lmv_obd          *lmv = &obddev->u.lmv;
 694         struct getinfo_fid2path *gf;
 695         struct lmv_tgt_desc     *tgt;
 696         struct getinfo_fid2path *remote_gf = NULL;
 697         int                     remote_gf_size = 0;
 698         int                     rc;
 699
 700         gf = (struct getinfo_fid2path *)karg;
 701         tgt = lmv_find_target(lmv, &gf->gf_fid);
 702         if (IS_ERR(tgt))
 703                 RETURN(PTR_ERR(tgt));
 704
 705 repeat_fid2path:
 706         rc = obd_iocontrol(OBD_IOC_FID2PATH, tgt->ltd_exp, len, gf, uarg);
 707         if (rc != 0 && rc != -EREMOTE)
 708                 GOTO(out_fid2path, rc);
 709
 710         /* If remote_gf != NULL, it means just building the
 711          * path on the remote MDT, copy this path segement to gf */
 712         if (remote_gf != NULL) {
 713                 struct getinfo_fid2path *ori_gf;
 714                 char *ptr;
 715
 716                 ori_gf = (struct getinfo_fid2path *)karg;
 717                 if (strlen(ori_gf->gf_path) +
 718                     strlen(gf->gf_path) > ori_gf->gf_pathlen)
 719                         GOTO(out_fid2path, rc = -EOVERFLOW);
 720
 721                 ptr = ori_gf->gf_path;
 722
 723                 memmove(ptr + strlen(gf->gf_path) + 1, ptr,
 724                         strlen(ori_gf->gf_path));
 725
 726                 strncpy(ptr, gf->gf_path, strlen(gf->gf_path));
 727                 ptr += strlen(gf->gf_path);
 728                 *ptr = '/';
 729         }
 730
 731         CDEBUG(D_INFO, "%s: get path %s "DFID" rec: "LPU64" ln: %u\n",
 732                tgt->ltd_exp->exp_obd->obd_name,
 733                gf->gf_path, PFID(&gf->gf_fid), gf->gf_recno,
 734                gf->gf_linkno);
 735
 736         if (rc == 0)
 737                 GOTO(out_fid2path, rc);
 738
 739         /* sigh, has to go to another MDT to do path building further */
 740         if (remote_gf == NULL) {
 741                 remote_gf_size = sizeof(*remote_gf) + PATH_MAX;
 742                 OBD_ALLOC(remote_gf, remote_gf_size);
 743                 if (remote_gf == NULL)
 744                         GOTO(out_fid2path, rc = -ENOMEM);
 745                 remote_gf->gf_pathlen = PATH_MAX;
 746         }
 747
 748         if (!fid_is_sane(&gf->gf_fid)) {
 749                 CERROR("%s: invalid FID "DFID": rc = %d\n",
 750                        tgt->ltd_exp->exp_obd->obd_name,
 751                        PFID(&gf->gf_fid), -EINVAL);
 752                 GOTO(out_fid2path, rc = -EINVAL);
 753         }
 754
 755         tgt = lmv_find_target(lmv, &gf->gf_fid);
 756         if (IS_ERR(tgt))
 757                 GOTO(out_fid2path, rc = -EINVAL);
 758
 759         remote_gf->gf_fid = gf->gf_fid;
 760         remote_gf->gf_recno = -1;
 761         remote_gf->gf_linkno = -1;
 762         memset(remote_gf->gf_path, 0, remote_gf->gf_pathlen);
 763         gf = remote_gf;
 764         goto repeat_fid2path;
 765
 766 out_fid2path:
 767         if (remote_gf != NULL)
 768                 OBD_FREE(remote_gf, remote_gf_size);
 769         RETURN(rc);
 770 }
 771
 772 static int lmv_hsm_req_count(struct lmv_obd *lmv,
 773                              const struct hsm_user_request *hur,
 774                              const struct lmv_tgt_desc *tgt_mds)
 775 {
 776         __u32                    i;
 777         int                      nr = 0;
 778         struct lmv_tgt_desc     *curr_tgt;
 779
 780         /* count how many requests must be sent to the given target */
 781         for (i = 0; i < hur->hur_request.hr_itemcount; i++) {
 782                 curr_tgt = lmv_find_target(lmv, &hur->hur_user_item[i].hui_fid);
 783                 if (obd_uuid_equals(&curr_tgt->ltd_uuid, &tgt_mds->ltd_uuid))
 784                         nr++;
 785         }
 786         return nr;
 787 }
 788
 789 static void lmv_hsm_req_build(struct lmv_obd *lmv,
 790                               struct hsm_user_request *hur_in,
 791                               const struct lmv_tgt_desc *tgt_mds,
 792                               struct hsm_user_request *hur_out)
 793 {
 794         __u32                    i, nr_out;
 795         struct lmv_tgt_desc     *curr_tgt;
 796
 797         /* build the hsm_user_request for the given target */
 798         hur_out->hur_request = hur_in->hur_request;
 799         nr_out = 0;
 800         for (i = 0; i < hur_in->hur_request.hr_itemcount; i++) {
 801                 curr_tgt = lmv_find_target(lmv,
 802                                            &hur_in->hur_user_item[i].hui_fid);
 803                 if (obd_uuid_equals(&curr_tgt->ltd_uuid, &tgt_mds->ltd_uuid)) {
 804                         hur_out->hur_user_item[nr_out] =
 805                                                 hur_in->hur_user_item[i];
 806                         nr_out++;
 807                 }
 808         }
 809         hur_out->hur_request.hr_itemcount = nr_out;
 810         memcpy(hur_data(hur_out), hur_data(hur_in),
 811                hur_in->hur_request.hr_data_len);
 812 }
 813
 814 static int lmv_hsm_ct_unregister(struct lmv_obd *lmv, unsigned int cmd, int len,
 815                                  struct lustre_kernelcomm *lk, void *uarg)
 816 {
 817         __u32                    i;
 818         int                      rc;
 819         struct kkuc_ct_data     *kcd = NULL;
 820         ENTRY;
 821
 822         /* unregister request (call from llapi_hsm_copytool_fini) */
 823         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 824                 struct lmv_tgt_desc *tgt = lmv->tgts[i];
 825
 826                 if (tgt == NULL || tgt->ltd_exp == NULL)
 827                         continue;
 828                 /* best effort: try to clean as much as possible
 829                  * (continue on error) */
 830                 obd_iocontrol(cmd, tgt->ltd_exp, len, lk, uarg);
 831         }
 832
 833         /* Whatever the result, remove copytool from kuc groups.
 834          * Unreached coordinators will get EPIPE on next requests
 835          * and will unregister automatically.
 836          */
 837         rc = libcfs_kkuc_group_rem(lk->lk_uid, lk->lk_group, (void **)&kcd);
 838         if (kcd != NULL)
 839                 OBD_FREE_PTR(kcd);
 840
 841         RETURN(rc);
 842 }
 843
 844 static int lmv_hsm_ct_register(struct lmv_obd *lmv, unsigned int cmd, int len,
 845                                struct lustre_kernelcomm *lk, void *uarg)
 846 {
 847         struct file             *filp;
 848         __u32                    i, j;
 849         int                      err, rc;
 850         bool                     any_set = false;
 851         struct kkuc_ct_data     *kcd;
 852         ENTRY;
 853
 854         /* All or nothing: try to register to all MDS.
 855          * In case of failure, unregister from previous MDS,
 856          * except if it because of inactive target. */
 857         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 858                 struct lmv_tgt_desc *tgt = lmv->tgts[i];
 859
 860                 if (tgt == NULL || tgt->ltd_exp == NULL)
 861                         continue;
 862                 err = obd_iocontrol(cmd, tgt->ltd_exp, len, lk, uarg);
 863                 if (err) {
 864                         if (tgt->ltd_active) {
 865                                 /* permanent error */
 866                                 CERROR("%s: iocontrol MDC %s on MDT"
 867                                        " idx %d cmd %x: err = %d\n",
 868                                        class_exp2obd(lmv->exp)->obd_name,
 869                                        tgt->ltd_uuid.uuid, i, cmd, err);
 870                                 rc = err;
 871                                 lk->lk_flags |= LK_FLG_STOP;
 872                                 /* unregister from previous MDS */
 873                                 for (j = 0; j < i; j++) {
 874                                         tgt = lmv->tgts[j];
 875                                         if (tgt == NULL || tgt->ltd_exp == NULL)
 876                                                 continue;
 877                                         obd_iocontrol(cmd, tgt->ltd_exp, len,
 878                                                       lk, uarg);
 879                                 }
 880                                 RETURN(rc);
 881                         }
 882                         /* else: transient error.
 883                          * kuc will register to the missing MDT
 884                          * when it is back */
 885                 } else {
 886                         any_set = true;
 887                 }
 888         }
 889
 890         if (!any_set)
 891                 /* no registration done: return error */
 892                 RETURN(-ENOTCONN);
 893
 894         /* at least one registration done, with no failure */
 895         filp = fget(lk->lk_wfd);
 896         if (filp == NULL)
 897                 RETURN(-EBADF);
 898
 899         OBD_ALLOC_PTR(kcd);
 900         if (kcd == NULL) {
 901                 fput(filp);
 902                 RETURN(-ENOMEM);
 903         }
 904         kcd->kcd_magic = KKUC_CT_DATA_MAGIC;
 905         kcd->kcd_uuid = lmv->cluuid;
 906         kcd->kcd_archive = lk->lk_data;
 907
 908         rc = libcfs_kkuc_group_add(filp, lk->lk_uid, lk->lk_group, kcd);
 909         if (rc != 0) {
 910                 if (filp != NULL)
 911                         fput(filp);
 912                 OBD_FREE_PTR(kcd);
 913         }
 914
 915         RETURN(rc);
 916 }
 917
 918
 919
 920
 921 static int lmv_iocontrol(unsigned int cmd, struct obd_export *exp,
 922                          int len, void *karg, void *uarg)
 923 {
 924         struct obd_device       *obddev = class_exp2obd(exp);
 925         struct lmv_obd          *lmv = &obddev->u.lmv;
 926         struct lmv_tgt_desc     *tgt = NULL;
 927         __u32                    i = 0;
 928         int                      rc = 0;
 929         int                      set = 0;
 930         __u32                    count = lmv->desc.ld_tgt_count;
 931         ENTRY;
 932
 933         if (count == 0)
 934                 RETURN(-ENOTTY);
 935
 936         switch (cmd) {
 937         case IOC_OBD_STATFS: {
 938                 struct obd_ioctl_data *data = karg;
 939                 struct obd_device *mdc_obd;
 940                 struct obd_statfs stat_buf = {0};
 941                 __u32 index;
 942
 943                 memcpy(&index, data->ioc_inlbuf2, sizeof(__u32));
 944                 if ((index >= count))
 945                         RETURN(-ENODEV);
 946
 947                 tgt = lmv->tgts[index];
 948                 if (tgt == NULL || !tgt->ltd_active)
 949                         RETURN(-ENODATA);
 950
 951                 mdc_obd = class_exp2obd(tgt->ltd_exp);
 952                 if (!mdc_obd)
 953                         RETURN(-EINVAL);
 954
 955                 /* copy UUID */
 956                 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(mdc_obd),
 957                                  min((int) data->ioc_plen2,
 958                                      (int) sizeof(struct obd_uuid))))
 959                         RETURN(-EFAULT);
 960
 961                 rc = obd_statfs(NULL, tgt->ltd_exp, &stat_buf,
 962                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
 963                                 0);
 964                 if (rc)
 965                         RETURN(rc);
 966                 if (copy_to_user(data->ioc_pbuf1, &stat_buf,
 967                                  min((int) data->ioc_plen1,
 968                                      (int) sizeof(stat_buf))))
 969                         RETURN(-EFAULT);
 970                 break;
 971         }
 972         case OBD_IOC_QUOTACTL: {
 973                 struct if_quotactl *qctl = karg;
 974                 struct obd_quotactl *oqctl;
 975
 976                 if (qctl->qc_valid == QC_MDTIDX) {
 977                         if (count <= qctl->qc_idx)
 978                                 RETURN(-EINVAL);
 979
 980                         tgt = lmv->tgts[qctl->qc_idx];
 981                         if (tgt == NULL || tgt->ltd_exp == NULL)
 982                                 RETURN(-EINVAL);
 983                 } else if (qctl->qc_valid == QC_UUID) {
 984                         for (i = 0; i < count; i++) {
 985                                 tgt = lmv->tgts[i];
 986                                 if (tgt == NULL)
 987                                         continue;
 988                                 if (!obd_uuid_equals(&tgt->ltd_uuid,
 989                                                      &qctl->obd_uuid))
 990                                         continue;
 991
 992                                 if (tgt->ltd_exp == NULL)
 993                                         RETURN(-EINVAL);
 994
 995                                 break;
 996                         }
 997                 } else {
 998                         RETURN(-EINVAL);
 999                 }
1000
1001                 if (i >= count)
1002                         RETURN(-EAGAIN);
1003
1004                 LASSERT(tgt != NULL && tgt->ltd_exp != NULL);
1005                 OBD_ALLOC_PTR(oqctl);
1006                 if (!oqctl)
1007                         RETURN(-ENOMEM);
1008
1009                 QCTL_COPY(oqctl, qctl);
1010                 rc = obd_quotactl(tgt->ltd_exp, oqctl);
1011                 if (rc == 0) {
1012                         QCTL_COPY(qctl, oqctl);
1013                         qctl->qc_valid = QC_MDTIDX;
1014                         qctl->obd_uuid = tgt->ltd_uuid;
1015                 }
1016                 OBD_FREE_PTR(oqctl);
1017                 break;
1018         }
1019         case OBD_IOC_CHANGELOG_SEND:
1020         case OBD_IOC_CHANGELOG_CLEAR: {
1021                 struct ioc_changelog *icc = karg;
1022
1023                 if (icc->icc_mdtindex >= count)
1024                         RETURN(-ENODEV);
1025
1026                 tgt = lmv->tgts[icc->icc_mdtindex];
1027                 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active)
1028                         RETURN(-ENODEV);
1029                 rc = obd_iocontrol(cmd, tgt->ltd_exp, sizeof(*icc), icc, NULL);
1030                 break;
1031         }
1032         case LL_IOC_GET_CONNECT_FLAGS: {
1033                 tgt = lmv->tgts[0];
1034                 if (tgt == NULL || tgt->ltd_exp == NULL)
1035                         RETURN(-ENODATA);
1036                 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1037                 break;
1038         }
1039         case LL_IOC_FID2MDTIDX: {
1040                 struct lu_fid *fid = karg;
1041                 int             mdt_index;
1042
1043                 rc = lmv_fld_lookup(lmv, fid, &mdt_index);
1044                 if (rc != 0)
1045                         RETURN(rc);
1046
1047                 /* Note: this is from llite(see ll_dir_ioctl()), @uarg does not
1048                  * point to user space memory for FID2MDTIDX. */
1049                 *(__u32 *)uarg = mdt_index;
1050                 break;
1051         }
1052         case OBD_IOC_FID2PATH: {
1053                 rc = lmv_fid2path(exp, len, karg, uarg);
1054                 break;
1055         }
1056         case LL_IOC_HSM_STATE_GET:
1057         case LL_IOC_HSM_STATE_SET:
1058         case LL_IOC_HSM_ACTION: {
1059                 struct md_op_data       *op_data = karg;
1060
1061                 tgt = lmv_find_target(lmv, &op_data->op_fid1);
1062                 if (IS_ERR(tgt))
1063                         RETURN(PTR_ERR(tgt));
1064
1065                 if (tgt->ltd_exp == NULL)
1066                         RETURN(-EINVAL);
1067
1068                 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1069                 break;
1070         }
1071         case LL_IOC_HSM_PROGRESS: {
1072                 const struct hsm_progress_kernel *hpk = karg;
1073
1074                 tgt = lmv_find_target(lmv, &hpk->hpk_fid);
1075                 if (IS_ERR(tgt))
1076                         RETURN(PTR_ERR(tgt));
1077                 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1078                 break;
1079         }
1080         case LL_IOC_HSM_REQUEST: {
1081                 struct hsm_user_request *hur = karg;
1082                 unsigned int reqcount = hur->hur_request.hr_itemcount;
1083
1084                 if (reqcount == 0)
1085                         RETURN(0);
1086
1087                 /* if the request is about a single fid
1088                  * or if there is a single MDS, no need to split
1089                  * the request. */
1090                 if (reqcount == 1 || count == 1) {
1091                         tgt = lmv_find_target(lmv,
1092                                               &hur->hur_user_item[0].hui_fid);
1093                         if (IS_ERR(tgt))
1094                                 RETURN(PTR_ERR(tgt));
1095                         rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1096                 } else {
1097                         /* split fid list to their respective MDS */
1098                         for (i = 0; i < count; i++) {
1099                                 unsigned int            nr, reqlen;
1100                                 int                     rc1;
1101                                 struct hsm_user_request *req;
1102
1103                                 tgt = lmv->tgts[i];
1104                                 if (tgt == NULL || tgt->ltd_exp == NULL)
1105                                         continue;
1106
1107                                 nr = lmv_hsm_req_count(lmv, hur, tgt);
1108                                 if (nr == 0) /* nothing for this MDS */
1109                                         continue;
1110
1111                                 /* build a request with fids for this MDS */
1112                                 reqlen = offsetof(typeof(*hur),
1113                                                   hur_user_item[nr])
1114                                                 + hur->hur_request.hr_data_len;
1115                                 OBD_ALLOC_LARGE(req, reqlen);
1116                                 if (req == NULL)
1117                                         RETURN(-ENOMEM);
1118
1119                                 lmv_hsm_req_build(lmv, hur, tgt, req);
1120
1121                                 rc1 = obd_iocontrol(cmd, tgt->ltd_exp, reqlen,
1122                                                     req, uarg);
1123                                 if (rc1 != 0 && rc == 0)
1124                                         rc = rc1;
1125                                 OBD_FREE_LARGE(req, reqlen);
1126                         }
1127                 }
1128                 break;
1129         }
1130         case LL_IOC_LOV_SWAP_LAYOUTS: {
1131                 struct md_op_data       *op_data = karg;
1132                 struct lmv_tgt_desc     *tgt1, *tgt2;
1133
1134                 tgt1 = lmv_find_target(lmv, &op_data->op_fid1);
1135                 if (IS_ERR(tgt1))
1136                         RETURN(PTR_ERR(tgt1));
1137
1138                 tgt2 = lmv_find_target(lmv, &op_data->op_fid2);
1139                 if (IS_ERR(tgt2))
1140                         RETURN(PTR_ERR(tgt2));
1141
1142                 if ((tgt1->ltd_exp == NULL) || (tgt2->ltd_exp == NULL))
1143                         RETURN(-EINVAL);
1144
1145                 /* only files on same MDT can have their layouts swapped */
1146                 if (tgt1->ltd_idx != tgt2->ltd_idx)
1147                         RETURN(-EPERM);
1148
1149                 rc = obd_iocontrol(cmd, tgt1->ltd_exp, len, karg, uarg);
1150                 break;
1151         }
1152         case LL_IOC_HSM_CT_START: {
1153                 struct lustre_kernelcomm *lk = karg;
1154                 if (lk->lk_flags & LK_FLG_STOP)
1155                         rc = lmv_hsm_ct_unregister(lmv, cmd, len, lk, uarg);
1156                 else
1157                         rc = lmv_hsm_ct_register(lmv, cmd, len, lk, uarg);
1158                 break;
1159         }
1160         default:
1161                 for (i = 0; i < count; i++) {
1162                         struct obd_device *mdc_obd;
1163                         int err;
1164
1165                         tgt = lmv->tgts[i];
1166                         if (tgt == NULL || tgt->ltd_exp == NULL)
1167                                 continue;
1168                         /* ll_umount_begin() sets force flag but for lmv, not
1169                          * mdc. Let's pass it through */
1170                         mdc_obd = class_exp2obd(tgt->ltd_exp);
1171                         mdc_obd->obd_force = obddev->obd_force;
1172                         err = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1173                         if (err == -ENODATA && cmd == OBD_IOC_POLL_QUOTACHECK) {
1174                                 RETURN(err);
1175                         } else if (err) {
1176                                 if (tgt->ltd_active) {
1177                                         CERROR("error: iocontrol MDC %s on MDT"
1178                                                " idx %d cmd %x: err = %d\n",
1179                                                tgt->ltd_uuid.uuid, i, cmd, err);
1180                                         if (!rc)
1181                                                 rc = err;
1182                                 }
1183                         } else
1184                                 set = 1;
1185                 }
1186                 if (!set && !rc)
1187                         rc = -EIO;
1188         }
1189         RETURN(rc);
1190 }
1191
1192 #if 0
1193 static int lmv_all_chars_policy(int count, const char *name,
1194                                 int len)
1195 {
1196         unsigned int c = 0;
1197
1198         while (len > 0)
1199                 c += name[--len];
1200         c = c % count;
1201         return c;
1202 }
1203
1204 static int lmv_nid_policy(struct lmv_obd *lmv)
1205 {
1206         struct obd_import *imp;
1207         __u32              id;
1208
1209         /*
1210          * XXX: To get nid we assume that underlying obd device is mdc.
1211          */
1212         imp = class_exp2cliimp(lmv->tgts[0].ltd_exp);
1213         id = imp->imp_connection->c_self ^ (imp->imp_connection->c_self >> 32);
1214         return id % lmv->desc.ld_tgt_count;
1215 }
1216
1217 static int lmv_choose_mds(struct lmv_obd *lmv, struct md_op_data *op_data,
1218                           placement_policy_t placement)
1219 {
1220         switch (placement) {
1221         case PLACEMENT_CHAR_POLICY:
1222                 return lmv_all_chars_policy(lmv->desc.ld_tgt_count,
1223                                             op_data->op_name,
1224                                             op_data->op_namelen);
1225         case PLACEMENT_NID_POLICY:
1226                 return lmv_nid_policy(lmv);
1227
1228         default:
1229                 break;
1230         }
1231
1232         CERROR("Unsupported placement policy %x\n", placement);
1233         return -EINVAL;
1234 }
1235 #endif
1236
1237 /**
1238  * This is _inode_ placement policy function (not name).
1239  */
1240 static int lmv_placement_policy(struct obd_device *obd,
1241                                 struct md_op_data *op_data, u32 *mds)
1242 {
1243         struct lmv_obd          *lmv = &obd->u.lmv;
1244         ENTRY;
1245
1246         LASSERT(mds != NULL);
1247
1248         if (lmv->desc.ld_tgt_count == 1) {
1249                 *mds = 0;
1250                 RETURN(0);
1251         }
1252
1253         if (op_data->op_default_stripe_offset != -1) {
1254                 *mds = op_data->op_default_stripe_offset;
1255                 RETURN(0);
1256         }
1257
1258         /**
1259          * If stripe_offset is provided during setdirstripe
1260          * (setdirstripe -i xx), xx MDS will be choosen.
1261          */
1262         if (op_data->op_cli_flags & CLI_SET_MEA && op_data->op_data != NULL) {
1263                 struct lmv_user_md *lum;
1264
1265                 lum = op_data->op_data;
1266
1267                 if (le32_to_cpu(lum->lum_stripe_offset) != (__u32)-1) {
1268                         *mds = le32_to_cpu(lum->lum_stripe_offset);
1269                 } else {
1270                         /* -1 means default, which will be in the same MDT with
1271                          * the stripe */
1272                         *mds = op_data->op_mds;
1273                         lum->lum_stripe_offset = cpu_to_le32(op_data->op_mds);
1274                 }
1275         } else {
1276                 /* Allocate new fid on target according to operation type and
1277                  * parent home mds. */
1278                 *mds = op_data->op_mds;
1279         }
1280
1281         RETURN(0);
1282 }
1283
1284 int __lmv_fid_alloc(struct lmv_obd *lmv, struct lu_fid *fid, u32 mds)
1285 {
1286         struct lmv_tgt_desc     *tgt;
1287         int                      rc;
1288         ENTRY;
1289
1290         tgt = lmv_get_target(lmv, mds, NULL);
1291         if (IS_ERR(tgt))
1292                 RETURN(PTR_ERR(tgt));
1293
1294         /*
1295          * New seq alloc and FLD setup should be atomic. Otherwise we may find
1296          * on server that seq in new allocated fid is not yet known.
1297          */
1298         mutex_lock(&tgt->ltd_fid_mutex);
1299
1300         if (tgt->ltd_active == 0 || tgt->ltd_exp == NULL)
1301                 GOTO(out, rc = -ENODEV);
1302
1303         /*
1304          * Asking underlying tgt layer to allocate new fid.
1305          */
1306         rc = obd_fid_alloc(NULL, tgt->ltd_exp, fid, NULL);
1307         if (rc > 0) {
1308                 LASSERT(fid_is_sane(fid));
1309                 rc = 0;
1310         }
1311
1312         EXIT;
1313 out:
1314         mutex_unlock(&tgt->ltd_fid_mutex);
1315         return rc;
1316 }
1317
1318 int lmv_fid_alloc(const struct lu_env *env, struct obd_export *exp,
1319                   struct lu_fid *fid, struct md_op_data *op_data)
1320 {
1321         struct obd_device     *obd = class_exp2obd(exp);
1322         struct lmv_obd        *lmv = &obd->u.lmv;
1323         u32                    mds = 0;
1324         int                    rc;
1325         ENTRY;
1326
1327         LASSERT(op_data != NULL);
1328         LASSERT(fid != NULL);
1329
1330         rc = lmv_placement_policy(obd, op_data, &mds);
1331         if (rc) {
1332                 CERROR("Can't get target for allocating fid, "
1333                        "rc %d\n", rc);
1334                 RETURN(rc);
1335         }
1336
1337         rc = __lmv_fid_alloc(lmv, fid, mds);
1338         if (rc) {
1339                 CERROR("Can't alloc new fid, rc %d\n", rc);
1340                 RETURN(rc);
1341         }
1342
1343         RETURN(rc);
1344 }
1345
1346 static int lmv_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
1347 {
1348         struct lmv_obd  *lmv = &obd->u.lmv;
1349         struct lmv_desc *desc;
1350         int             rc;
1351         ENTRY;
1352
1353         if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
1354                 CERROR("LMV setup requires a descriptor\n");
1355                 RETURN(-EINVAL);
1356         }
1357
1358         desc = (struct lmv_desc *)lustre_cfg_buf(lcfg, 1);
1359         if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
1360                 CERROR("Lmv descriptor size wrong: %d > %d\n",
1361                        (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
1362                 RETURN(-EINVAL);
1363         }
1364
1365         lmv->tgts_size = 32U;
1366         OBD_ALLOC(lmv->tgts, sizeof(*lmv->tgts) * lmv->tgts_size);
1367         if (lmv->tgts == NULL)
1368                 RETURN(-ENOMEM);
1369
1370         obd_str2uuid(&lmv->desc.ld_uuid, desc->ld_uuid.uuid);
1371         lmv->desc.ld_tgt_count = 0;
1372         lmv->desc.ld_active_tgt_count = 0;
1373         lmv->max_cookiesize = 0;
1374         lmv->max_def_easize = 0;
1375         lmv->max_easize = 0;
1376         lmv->lmv_placement = PLACEMENT_CHAR_POLICY;
1377
1378         spin_lock_init(&lmv->lmv_lock);
1379         mutex_init(&lmv->lmv_init_mutex);
1380
1381 #ifdef CONFIG_PROC_FS
1382         obd->obd_vars = lprocfs_lmv_obd_vars;
1383         lprocfs_obd_setup(obd);
1384         lprocfs_alloc_md_stats(obd, 0);
1385         rc = lprocfs_seq_create(obd->obd_proc_entry, "target_obd",
1386                                 0444, &lmv_proc_target_fops, obd);
1387         if (rc)
1388                 CWARN("%s: error adding LMV target_obd file: rc = %d\n",
1389                       obd->obd_name, rc);
1390 #endif
1391         rc = fld_client_init(&lmv->lmv_fld, obd->obd_name,
1392                              LUSTRE_CLI_FLD_HASH_DHT);
1393         if (rc) {
1394                 CERROR("Can't init FLD, err %d\n", rc);
1395                 GOTO(out, rc);
1396         }
1397
1398         RETURN(0);
1399
1400 out:
1401         return rc;
1402 }
1403
1404 static int lmv_cleanup(struct obd_device *obd)
1405 {
1406         struct lmv_obd   *lmv = &obd->u.lmv;
1407         ENTRY;
1408
1409         fld_client_fini(&lmv->lmv_fld);
1410         if (lmv->tgts != NULL) {
1411                 int i;
1412                 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1413                         if (lmv->tgts[i] == NULL)
1414                                 continue;
1415                         lmv_del_target(lmv, i);
1416                 }
1417                 OBD_FREE(lmv->tgts, sizeof(*lmv->tgts) * lmv->tgts_size);
1418                 lmv->tgts_size = 0;
1419         }
1420         RETURN(0);
1421 }
1422
1423 static int lmv_process_config(struct obd_device *obd, size_t len, void *buf)
1424 {
1425         struct lustre_cfg       *lcfg = buf;
1426         struct obd_uuid         obd_uuid;
1427         int                     gen;
1428         __u32                   index;
1429         int                     rc;
1430         ENTRY;
1431
1432         switch (lcfg->lcfg_command) {
1433         case LCFG_ADD_MDC:
1434                 /* modify_mdc_tgts add 0:lustre-clilmv  1:lustre-MDT0000_UUID
1435                  * 2:0  3:1  4:lustre-MDT0000-mdc_UUID */
1436                 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(obd_uuid.uuid))
1437                         GOTO(out, rc = -EINVAL);
1438
1439                 obd_str2uuid(&obd_uuid,  lustre_cfg_buf(lcfg, 1));
1440
1441                 if (sscanf(lustre_cfg_buf(lcfg, 2), "%u", &index) != 1)
1442                         GOTO(out, rc = -EINVAL);
1443                 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", &gen) != 1)
1444                         GOTO(out, rc = -EINVAL);
1445                 rc = lmv_add_target(obd, &obd_uuid, index, gen);
1446                 GOTO(out, rc);
1447         default:
1448                 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
1449                 GOTO(out, rc = -EINVAL);
1450         }
1451 out:
1452         RETURN(rc);
1453 }
1454
1455 static int lmv_statfs(const struct lu_env *env, struct obd_export *exp,
1456                       struct obd_statfs *osfs, __u64 max_age, __u32 flags)
1457 {
1458         struct obd_device       *obd = class_exp2obd(exp);
1459         struct lmv_obd          *lmv = &obd->u.lmv;
1460         struct obd_statfs       *temp;
1461         int                      rc = 0;
1462         __u32                    i;
1463         ENTRY;
1464
1465         rc = lmv_check_connect(obd);
1466         if (rc)
1467                 RETURN(rc);
1468
1469         OBD_ALLOC(temp, sizeof(*temp));
1470         if (temp == NULL)
1471                 RETURN(-ENOMEM);
1472
1473         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1474                 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
1475                         continue;
1476
1477                 rc = obd_statfs(env, lmv->tgts[i]->ltd_exp, temp,
1478                                 max_age, flags);
1479                 if (rc) {
1480                         CERROR("can't stat MDS #%d (%s), error %d\n", i,
1481                                lmv->tgts[i]->ltd_exp->exp_obd->obd_name,
1482                                rc);
1483                         GOTO(out_free_temp, rc);
1484                 }
1485
1486                 if (i == 0) {
1487                         *osfs = *temp;
1488                         /* If the statfs is from mount, it will needs
1489                          * retrieve necessary information from MDT0.
1490                          * i.e. mount does not need the merged osfs
1491                          * from all of MDT.
1492                          * And also clients can be mounted as long as
1493                          * MDT0 is in service*/
1494                         if (flags & OBD_STATFS_FOR_MDT0)
1495                                 GOTO(out_free_temp, rc);
1496                 } else {
1497                         osfs->os_bavail += temp->os_bavail;
1498                         osfs->os_blocks += temp->os_blocks;
1499                         osfs->os_ffree += temp->os_ffree;
1500                         osfs->os_files += temp->os_files;
1501                 }
1502         }
1503
1504         EXIT;
1505 out_free_temp:
1506         OBD_FREE(temp, sizeof(*temp));
1507         return rc;
1508 }
1509
1510 static int lmv_getstatus(struct obd_export *exp,
1511                          struct lu_fid *fid,
1512                          struct obd_capa **pc)
1513 {
1514         struct obd_device    *obd = exp->exp_obd;
1515         struct lmv_obd       *lmv = &obd->u.lmv;
1516         int                   rc;
1517         ENTRY;
1518
1519         rc = lmv_check_connect(obd);
1520         if (rc)
1521                 RETURN(rc);
1522
1523         rc = md_getstatus(lmv->tgts[0]->ltd_exp, fid, pc);
1524         RETURN(rc);
1525 }
1526
1527 static int lmv_getxattr(struct obd_export *exp, const struct lu_fid *fid,
1528                         struct obd_capa *oc, u64 valid, const char *name,
1529                         const char *input, int input_size, int output_size,
1530                         int flags, struct ptlrpc_request **request)
1531 {
1532         struct obd_device      *obd = exp->exp_obd;
1533         struct lmv_obd         *lmv = &obd->u.lmv;
1534         struct lmv_tgt_desc    *tgt;
1535         int                     rc;
1536         ENTRY;
1537
1538         rc = lmv_check_connect(obd);
1539         if (rc)
1540                 RETURN(rc);
1541
1542         tgt = lmv_find_target(lmv, fid);
1543         if (IS_ERR(tgt))
1544                 RETURN(PTR_ERR(tgt));
1545
1546         rc = md_getxattr(tgt->ltd_exp, fid, oc, valid, name, input,
1547                          input_size, output_size, flags, request);
1548
1549         RETURN(rc);
1550 }
1551
1552 static int lmv_setxattr(struct obd_export *exp, const struct lu_fid *fid,
1553                         struct obd_capa *oc, u64 valid, const char *name,
1554                         const char *input, int input_size, int output_size,
1555                         int flags, __u32 suppgid,
1556                         struct ptlrpc_request **request)
1557 {
1558         struct obd_device      *obd = exp->exp_obd;
1559         struct lmv_obd         *lmv = &obd->u.lmv;
1560         struct lmv_tgt_desc    *tgt;
1561         int                     rc;
1562         ENTRY;
1563
1564         rc = lmv_check_connect(obd);
1565         if (rc)
1566                 RETURN(rc);
1567
1568         tgt = lmv_find_target(lmv, fid);
1569         if (IS_ERR(tgt))
1570                 RETURN(PTR_ERR(tgt));
1571
1572         rc = md_setxattr(tgt->ltd_exp, fid, oc, valid, name, input,
1573                          input_size, output_size, flags, suppgid,
1574                          request);
1575
1576         RETURN(rc);
1577 }
1578
1579 static int lmv_getattr(struct obd_export *exp, struct md_op_data *op_data,
1580                        struct ptlrpc_request **request)
1581 {
1582         struct obd_device       *obd = exp->exp_obd;
1583         struct lmv_obd          *lmv = &obd->u.lmv;
1584         struct lmv_tgt_desc     *tgt;
1585         int                      rc;
1586         ENTRY;
1587
1588         rc = lmv_check_connect(obd);
1589         if (rc)
1590                 RETURN(rc);
1591
1592         tgt = lmv_find_target(lmv, &op_data->op_fid1);
1593         if (IS_ERR(tgt))
1594                 RETURN(PTR_ERR(tgt));
1595
1596         if (op_data->op_flags & MF_GET_MDT_IDX) {
1597                 op_data->op_mds = tgt->ltd_idx;
1598                 RETURN(0);
1599         }
1600
1601         rc = md_getattr(tgt->ltd_exp, op_data, request);
1602
1603         RETURN(rc);
1604 }
1605
1606 static int lmv_null_inode(struct obd_export *exp, const struct lu_fid *fid)
1607 {
1608         struct obd_device   *obd = exp->exp_obd;
1609         struct lmv_obd      *lmv = &obd->u.lmv;
1610         __u32                i;
1611         int                  rc;
1612         ENTRY;
1613
1614         rc = lmv_check_connect(obd);
1615         if (rc)
1616                 RETURN(rc);
1617
1618         CDEBUG(D_INODE, "CBDATA for "DFID"\n", PFID(fid));
1619
1620         /*
1621          * With DNE every object can have two locks in different namespaces:
1622          * lookup lock in space of MDT storing direntry and update/open lock in
1623          * space of MDT storing inode.
1624          */
1625         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1626                 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
1627                         continue;
1628                 md_null_inode(lmv->tgts[i]->ltd_exp, fid);
1629         }
1630
1631         RETURN(0);
1632 }
1633
1634 static int lmv_find_cbdata(struct obd_export *exp, const struct lu_fid *fid,
1635                            ldlm_iterator_t it, void *data)
1636 {
1637         struct obd_device       *obd = exp->exp_obd;
1638         struct lmv_obd          *lmv = &obd->u.lmv;
1639         int                     i;
1640         int                     tgt;
1641         int                     rc;
1642         ENTRY;
1643
1644         rc = lmv_check_connect(obd);
1645         if (rc)
1646                 RETURN(rc);
1647
1648         CDEBUG(D_INODE, "CBDATA for "DFID"\n", PFID(fid));
1649
1650         /*
1651          * With DNE every object can have two locks in different namespaces:
1652          * lookup lock in space of MDT storing direntry and update/open lock in
1653          * space of MDT storing inode.  Try the MDT that the FID maps to first,
1654          * since this can be easily found, and only try others if that fails.
1655          */
1656         for (i = 0, tgt = lmv_find_target_index(lmv, fid);
1657              i < lmv->desc.ld_tgt_count;
1658              i++, tgt = (tgt + 1) % lmv->desc.ld_tgt_count) {
1659                 if (tgt < 0) {
1660                         CDEBUG(D_HA, "%s: "DFID" is inaccessible: rc = %d\n",
1661                                obd->obd_name, PFID(fid), tgt);
1662                         tgt = 0;
1663                 }
1664
1665                 if (lmv->tgts[tgt] == NULL ||
1666                     lmv->tgts[tgt]->ltd_exp == NULL)
1667                         continue;
1668
1669                 rc = md_find_cbdata(lmv->tgts[tgt]->ltd_exp, fid, it, data);
1670                 if (rc)
1671                         RETURN(rc);
1672         }
1673
1674         RETURN(rc);
1675 }
1676
1677
1678 static int lmv_close(struct obd_export *exp, struct md_op_data *op_data,
1679                      struct md_open_data *mod, struct ptlrpc_request **request)
1680 {
1681         struct obd_device     *obd = exp->exp_obd;
1682         struct lmv_obd        *lmv = &obd->u.lmv;
1683         struct lmv_tgt_desc   *tgt;
1684         int                    rc;
1685         ENTRY;
1686
1687         rc = lmv_check_connect(obd);
1688         if (rc)
1689                 RETURN(rc);
1690
1691         tgt = lmv_find_target(lmv, &op_data->op_fid1);
1692         if (IS_ERR(tgt))
1693                 RETURN(PTR_ERR(tgt));
1694
1695         CDEBUG(D_INODE, "CLOSE "DFID"\n", PFID(&op_data->op_fid1));
1696         rc = md_close(tgt->ltd_exp, op_data, mod, request);
1697         RETURN(rc);
1698 }
1699
1700 /**
1701  * Choosing the MDT by name or FID in @op_data.
1702  * For non-striped directory, it will locate MDT by fid.
1703  * For striped-directory, it will locate MDT by name. And also
1704  * it will reset op_fid1 with the FID of the choosen stripe.
1705  **/
1706 struct lmv_tgt_desc *
1707 lmv_locate_target_for_name(struct lmv_obd *lmv, struct lmv_stripe_md *lsm,
1708                            const char *name, int namelen, struct lu_fid *fid,
1709                            u32 *mds)
1710 {
1711         struct lmv_tgt_desc     *tgt;
1712         const struct lmv_oinfo  *oinfo;
1713
1714         if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_NAME_HASH)) {
1715                 if (cfs_fail_val >= lsm->lsm_md_stripe_count)
1716                         RETURN(ERR_PTR(-EBADF));
1717                 oinfo = &lsm->lsm_md_oinfo[cfs_fail_val];
1718         } else {
1719                 oinfo = lsm_name_to_stripe_info(lsm, name, namelen);
1720                 if (IS_ERR(oinfo))
1721                         RETURN(ERR_CAST(oinfo));
1722         }
1723
1724         if (fid != NULL)
1725                 *fid = oinfo->lmo_fid;
1726         if (mds != NULL)
1727                 *mds = oinfo->lmo_mds;
1728
1729         tgt = lmv_get_target(lmv, oinfo->lmo_mds, NULL);
1730
1731         CDEBUG(D_INFO, "locate on mds %u "DFID"\n", oinfo->lmo_mds,
1732                PFID(&oinfo->lmo_fid));
1733         return tgt;
1734 }
1735
1736 /**
1737  * Locate mds by fid or name
1738  *
1739  * For striped directory (lsm != NULL), it will locate the stripe
1740  * by name hash (see lsm_name_to_stripe_info()). Note: if the hash_type
1741  * is unknown, it will return -EBADFD, and lmv_intent_lookup might need
1742  * walk through all of stripes to locate the entry.
1743  *
1744  * For normal direcotry, it will locate MDS by FID directly.
1745  * \param[in] lmv       LMV device
1746  * \param[in] op_data   client MD stack parameters, name, namelen
1747  *                      mds_num etc.
1748  * \param[in] fid       object FID used to locate MDS.
1749  *
1750  * retval               pointer to the lmv_tgt_desc if succeed.
1751  *                      ERR_PTR(errno) if failed.
1752  */
1753 struct lmv_tgt_desc*
1754 lmv_locate_mds(struct lmv_obd *lmv, struct md_op_data *op_data,
1755                struct lu_fid *fid)
1756 {
1757         struct lmv_stripe_md    *lsm = op_data->op_mea1;
1758         struct lmv_tgt_desc     *tgt;
1759
1760         /* During creating VOLATILE file, it should honor the mdt
1761          * index if the file under striped dir is being restored, see
1762          * ct_restore(). */
1763         if (op_data->op_bias & MDS_CREATE_VOLATILE &&
1764             (int)op_data->op_mds != -1 && lsm != NULL) {
1765                 int i;
1766                 tgt = lmv_get_target(lmv, op_data->op_mds, NULL);
1767                 if (IS_ERR(tgt))
1768                         return tgt;
1769
1770                 /* refill the right parent fid */
1771                 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1772                         struct lmv_oinfo *oinfo;
1773
1774                         oinfo = &lsm->lsm_md_oinfo[i];
1775                         if (oinfo->lmo_mds == op_data->op_mds) {
1776                                 *fid = oinfo->lmo_fid;
1777                                 break;
1778                         }
1779                 }
1780
1781                 /* Hmm, can not find the stripe by mdt_index(op_mds) */
1782                 if (i == lsm->lsm_md_stripe_count)
1783                         tgt = ERR_PTR(-EINVAL);
1784
1785                 return tgt;
1786         }
1787
1788         if (lsm == NULL || op_data->op_namelen == 0) {
1789                 tgt = lmv_find_target(lmv, fid);
1790                 if (IS_ERR(tgt))
1791                         return tgt;
1792
1793                 op_data->op_mds = tgt->ltd_idx;
1794                 return tgt;
1795         }
1796
1797         return lmv_locate_target_for_name(lmv, lsm, op_data->op_name,
1798                                           op_data->op_namelen, fid,
1799                                           &op_data->op_mds);
1800 }
1801
1802 int lmv_create(struct obd_export *exp, struct md_op_data *op_data,
1803                 const void *data, size_t datalen, umode_t mode, uid_t uid,
1804                 gid_t gid, cfs_cap_t cap_effective, __u64 rdev,
1805                 struct ptlrpc_request **request)
1806 {
1807         struct obd_device       *obd = exp->exp_obd;
1808         struct lmv_obd          *lmv = &obd->u.lmv;
1809         struct lmv_tgt_desc     *tgt;
1810         int                      rc;
1811         ENTRY;
1812
1813         rc = lmv_check_connect(obd);
1814         if (rc)
1815                 RETURN(rc);
1816
1817         if (!lmv->desc.ld_active_tgt_count)
1818                 RETURN(-EIO);
1819
1820         tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
1821         if (IS_ERR(tgt))
1822                 RETURN(PTR_ERR(tgt));
1823
1824         CDEBUG(D_INODE, "CREATE name '%.*s' on "DFID" -> mds #%x\n",
1825                 (int)op_data->op_namelen, op_data->op_name,
1826                 PFID(&op_data->op_fid1), op_data->op_mds);
1827
1828         rc = lmv_fid_alloc(NULL, exp, &op_data->op_fid2, op_data);
1829         if (rc)
1830                 RETURN(rc);
1831         if (exp_connect_flags(exp) & OBD_CONNECT_DIR_STRIPE) {
1832                 /* Send the create request to the MDT where the object
1833                  * will be located */
1834                 tgt = lmv_find_target(lmv, &op_data->op_fid2);
1835                 if (IS_ERR(tgt))
1836                         RETURN(PTR_ERR(tgt));
1837
1838                 op_data->op_mds = tgt->ltd_idx;
1839         } else {
1840                 CDEBUG(D_CONFIG, "Server doesn't support striped dirs\n");
1841         }
1842
1843         CDEBUG(D_INODE, "CREATE obj "DFID" -> mds #%x\n",
1844                PFID(&op_data->op_fid2), op_data->op_mds);
1845
1846         op_data->op_flags |= MF_MDC_CANCEL_FID1;
1847         rc = md_create(tgt->ltd_exp, op_data, data, datalen, mode, uid, gid,
1848                        cap_effective, rdev, request);
1849         if (rc == 0) {
1850                 if (*request == NULL)
1851                         RETURN(rc);
1852                 CDEBUG(D_INODE, "Created - "DFID"\n", PFID(&op_data->op_fid2));
1853         }
1854         RETURN(rc);
1855 }
1856
1857 static int lmv_done_writing(struct obd_export *exp,
1858                             struct md_op_data *op_data,
1859                             struct md_open_data *mod)
1860 {
1861         struct obd_device     *obd = exp->exp_obd;
1862         struct lmv_obd        *lmv = &obd->u.lmv;
1863         struct lmv_tgt_desc   *tgt;
1864         int                    rc;
1865         ENTRY;
1866
1867         rc = lmv_check_connect(obd);
1868         if (rc)
1869                 RETURN(rc);
1870
1871         tgt = lmv_find_target(lmv, &op_data->op_fid1);
1872         if (IS_ERR(tgt))
1873                 RETURN(PTR_ERR(tgt));
1874
1875         rc = md_done_writing(tgt->ltd_exp, op_data, mod);
1876         RETURN(rc);
1877 }
1878
1879 static int
1880 lmv_enqueue(struct obd_export *exp, struct ldlm_enqueue_info *einfo,
1881             const union ldlm_policy_data *policy,
1882             struct lookup_intent *it, struct md_op_data *op_data,
1883             struct lustre_handle *lockh, __u64 extra_lock_flags)
1884 {
1885         struct obd_device        *obd = exp->exp_obd;
1886         struct lmv_obd           *lmv = &obd->u.lmv;
1887         struct lmv_tgt_desc      *tgt;
1888         int                       rc;
1889         ENTRY;
1890
1891         rc = lmv_check_connect(obd);
1892         if (rc)
1893                 RETURN(rc);
1894
1895         CDEBUG(D_INODE, "ENQUEUE '%s' on "DFID"\n",
1896                LL_IT2STR(it), PFID(&op_data->op_fid1));
1897
1898         tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
1899         if (IS_ERR(tgt))
1900                 RETURN(PTR_ERR(tgt));
1901
1902         CDEBUG(D_INODE, "ENQUEUE '%s' on "DFID" -> mds #%u\n",
1903                LL_IT2STR(it), PFID(&op_data->op_fid1), tgt->ltd_idx);
1904
1905         rc = md_enqueue(tgt->ltd_exp, einfo, policy, it, op_data, lockh,
1906                         extra_lock_flags);
1907
1908         RETURN(rc);
1909 }
1910
1911 static int
1912 lmv_getattr_name(struct obd_export *exp,struct md_op_data *op_data,
1913                  struct ptlrpc_request **preq)
1914 {
1915         struct ptlrpc_request   *req = NULL;
1916         struct obd_device       *obd = exp->exp_obd;
1917         struct lmv_obd          *lmv = &obd->u.lmv;
1918         struct lmv_tgt_desc     *tgt;
1919         struct mdt_body         *body;
1920         int                      rc;
1921         ENTRY;
1922
1923         rc = lmv_check_connect(obd);
1924         if (rc)
1925                 RETURN(rc);
1926
1927         tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
1928         if (IS_ERR(tgt))
1929                 RETURN(PTR_ERR(tgt));
1930
1931         CDEBUG(D_INODE, "GETATTR_NAME for %*s on "DFID" -> mds #%d\n",
1932                 (int)op_data->op_namelen, op_data->op_name,
1933                 PFID(&op_data->op_fid1), tgt->ltd_idx);
1934
1935         rc = md_getattr_name(tgt->ltd_exp, op_data, preq);
1936         if (rc != 0)
1937                 RETURN(rc);
1938
1939         body = req_capsule_server_get(&(*preq)->rq_pill, &RMF_MDT_BODY);
1940         LASSERT(body != NULL);
1941
1942         if (body->mbo_valid & OBD_MD_MDS) {
1943                 struct lu_fid rid = body->mbo_fid1;
1944                 CDEBUG(D_INODE, "Request attrs for "DFID"\n",
1945                        PFID(&rid));
1946
1947                 tgt = lmv_find_target(lmv, &rid);
1948                 if (IS_ERR(tgt)) {
1949                         ptlrpc_req_finished(*preq);
1950                         preq = NULL;
1951                         RETURN(PTR_ERR(tgt));
1952                 }
1953
1954                 op_data->op_fid1 = rid;
1955                 op_data->op_valid |= OBD_MD_FLCROSSREF;
1956                 op_data->op_namelen = 0;
1957                 op_data->op_name = NULL;
1958                 rc = md_getattr_name(tgt->ltd_exp, op_data, &req);
1959                 ptlrpc_req_finished(*preq);
1960                 *preq = req;
1961         }
1962
1963         RETURN(rc);
1964 }
1965
1966 #define md_op_data_fid(op_data, fl)                     \
1967         (fl == MF_MDC_CANCEL_FID1 ? &op_data->op_fid1 : \
1968          fl == MF_MDC_CANCEL_FID2 ? &op_data->op_fid2 : \
1969          fl == MF_MDC_CANCEL_FID3 ? &op_data->op_fid3 : \
1970          fl == MF_MDC_CANCEL_FID4 ? &op_data->op_fid4 : \
1971          NULL)
1972
1973 static int lmv_early_cancel(struct obd_export *exp, struct lmv_tgt_desc *tgt,
1974                             struct md_op_data *op_data,
1975                             __u32 op_tgt, ldlm_mode_t mode, int bits, int flag)
1976 {
1977         struct lu_fid          *fid = md_op_data_fid(op_data, flag);
1978         struct obd_device      *obd = exp->exp_obd;
1979         struct lmv_obd         *lmv = &obd->u.lmv;
1980         ldlm_policy_data_t      policy = {{ 0 }};
1981         int                     rc = 0;
1982         ENTRY;
1983
1984         if (!fid_is_sane(fid))
1985                 RETURN(0);
1986
1987         if (tgt == NULL) {
1988                 tgt = lmv_find_target(lmv, fid);
1989                 if (IS_ERR(tgt))
1990                         RETURN(PTR_ERR(tgt));
1991         }
1992
1993         if (tgt->ltd_idx != op_tgt) {
1994                 CDEBUG(D_INODE, "EARLY_CANCEL on "DFID"\n", PFID(fid));
1995                 policy.l_inodebits.bits = bits;
1996                 rc = md_cancel_unused(tgt->ltd_exp, fid, &policy,
1997                                       mode, LCF_ASYNC, NULL);
1998         } else {
1999                 CDEBUG(D_INODE,
2000                        "EARLY_CANCEL skip operation target %d on "DFID"\n",
2001                        op_tgt, PFID(fid));
2002                 op_data->op_flags |= flag;
2003                 rc = 0;
2004         }
2005
2006         RETURN(rc);
2007 }
2008
2009 /*
2010  * llite passes fid of an target inode in op_data->op_fid1 and id of directory in
2011  * op_data->op_fid2
2012  */
2013 static int lmv_link(struct obd_export *exp, struct md_op_data *op_data,
2014                     struct ptlrpc_request **request)
2015 {
2016         struct obd_device       *obd = exp->exp_obd;
2017         struct lmv_obd          *lmv = &obd->u.lmv;
2018         struct lmv_tgt_desc     *tgt;
2019         int                      rc;
2020         ENTRY;
2021
2022         rc = lmv_check_connect(obd);
2023         if (rc)
2024                 RETURN(rc);
2025
2026         LASSERT(op_data->op_namelen != 0);
2027
2028         CDEBUG(D_INODE, "LINK "DFID":%*s to "DFID"\n",
2029                PFID(&op_data->op_fid2), (int)op_data->op_namelen,
2030                op_data->op_name, PFID(&op_data->op_fid1));
2031
2032         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2033         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2034         op_data->op_cap = cfs_curproc_cap_pack();
2035         if (op_data->op_mea2 != NULL) {
2036                 struct lmv_stripe_md    *lsm = op_data->op_mea2;
2037                 const struct lmv_oinfo  *oinfo;
2038
2039                 oinfo = lsm_name_to_stripe_info(lsm, op_data->op_name,
2040                                                 op_data->op_namelen);
2041                 if (IS_ERR(oinfo))
2042                         RETURN(PTR_ERR(oinfo));
2043
2044                 op_data->op_fid2 = oinfo->lmo_fid;
2045         }
2046
2047         tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid2);
2048         if (IS_ERR(tgt))
2049                 RETURN(PTR_ERR(tgt));
2050
2051         /*
2052          * Cancel UPDATE lock on child (fid1).
2053          */
2054         op_data->op_flags |= MF_MDC_CANCEL_FID2;
2055         rc = lmv_early_cancel(exp, NULL, op_data, tgt->ltd_idx, LCK_EX,
2056                               MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID1);
2057         if (rc != 0)
2058                 RETURN(rc);
2059
2060         rc = md_link(tgt->ltd_exp, op_data, request);
2061
2062         RETURN(rc);
2063 }
2064
2065 static int lmv_rename(struct obd_export *exp, struct md_op_data *op_data,
2066                       const char *old, size_t oldlen,
2067                       const char *new, size_t newlen,
2068                       struct ptlrpc_request **request)
2069 {
2070         struct obd_device       *obd = exp->exp_obd;
2071         struct lmv_obd          *lmv = &obd->u.lmv;
2072         struct lmv_tgt_desc     *src_tgt;
2073         int                     rc;
2074         ENTRY;
2075
2076         LASSERT(oldlen != 0);
2077
2078         CDEBUG(D_INODE, "RENAME %.*s in "DFID":%d to %.*s in "DFID":%d\n",
2079                (int)oldlen, old, PFID(&op_data->op_fid1),
2080                op_data->op_mea1 ? op_data->op_mea1->lsm_md_stripe_count : 0,
2081                (int)newlen, new, PFID(&op_data->op_fid2),
2082                op_data->op_mea2 ? op_data->op_mea2->lsm_md_stripe_count : 0);
2083
2084         rc = lmv_check_connect(obd);
2085         if (rc)
2086                 RETURN(rc);
2087
2088         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2089         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2090         op_data->op_cap = cfs_curproc_cap_pack();
2091         if (op_data->op_cli_flags & CLI_MIGRATE) {
2092                 LASSERTF(fid_is_sane(&op_data->op_fid3), "invalid FID "DFID"\n",
2093                          PFID(&op_data->op_fid3));
2094
2095                 if (op_data->op_mea1 != NULL) {
2096                         struct lmv_stripe_md    *lsm = op_data->op_mea1;
2097                         struct lmv_tgt_desc     *tmp;
2098
2099                         /* Fix the parent fid for striped dir */
2100                         tmp = lmv_locate_target_for_name(lmv, lsm, old,
2101                                                          oldlen,
2102                                                          &op_data->op_fid1,
2103                                                          NULL);
2104                         if (IS_ERR(tmp))
2105                                 RETURN(PTR_ERR(tmp));
2106                 }
2107
2108                 rc = lmv_fid_alloc(NULL, exp, &op_data->op_fid2, op_data);
2109                 if (rc != 0)
2110                         RETURN(rc);
2111
2112                 src_tgt = lmv_find_target(lmv, &op_data->op_fid3);
2113         } else {
2114                 if (op_data->op_mea1 != NULL) {
2115                         struct lmv_stripe_md    *lsm = op_data->op_mea1;
2116
2117                         src_tgt = lmv_locate_target_for_name(lmv, lsm, old,
2118                                                              oldlen,
2119                                                              &op_data->op_fid1,
2120                                                              &op_data->op_mds);
2121                         if (IS_ERR(src_tgt))
2122                                 RETURN(PTR_ERR(src_tgt));
2123                 } else {
2124                         src_tgt = lmv_find_target(lmv, &op_data->op_fid1);
2125                         if (IS_ERR(src_tgt))
2126                                 RETURN(PTR_ERR(src_tgt));
2127
2128                         op_data->op_mds = src_tgt->ltd_idx;
2129                 }
2130
2131                 if (op_data->op_mea2) {
2132                         struct lmv_stripe_md    *lsm = op_data->op_mea2;
2133                         const struct lmv_oinfo  *oinfo;
2134
2135                         oinfo = lsm_name_to_stripe_info(lsm, new, newlen);
2136                         if (IS_ERR(oinfo))
2137                                 RETURN(PTR_ERR(oinfo));
2138
2139                         op_data->op_fid2 = oinfo->lmo_fid;
2140                 }
2141         }
2142         if (IS_ERR(src_tgt))
2143                 RETURN(PTR_ERR(src_tgt));
2144
2145         /*
2146          * LOOKUP lock on src child (fid3) should also be cancelled for
2147          * src_tgt in mdc_rename.
2148          */
2149         op_data->op_flags |= MF_MDC_CANCEL_FID1 | MF_MDC_CANCEL_FID3;
2150
2151         /*
2152          * Cancel UPDATE locks on tgt parent (fid2), tgt_tgt is its
2153          * own target.
2154          */
2155         rc = lmv_early_cancel(exp, NULL, op_data, src_tgt->ltd_idx,
2156                               LCK_EX, MDS_INODELOCK_UPDATE,
2157                               MF_MDC_CANCEL_FID2);
2158
2159         if (rc != 0)
2160                 RETURN(rc);
2161         /*
2162          * Cancel LOOKUP locks on source child (fid3) for parent tgt_tgt.
2163          */
2164         if (fid_is_sane(&op_data->op_fid3)) {
2165                 struct lmv_tgt_desc *tgt;
2166
2167                 tgt = lmv_find_target(lmv, &op_data->op_fid1);
2168                 if (IS_ERR(tgt))
2169                         RETURN(PTR_ERR(tgt));
2170
2171                 /* Cancel LOOKUP lock on its parent */
2172                 rc = lmv_early_cancel(exp, tgt, op_data, src_tgt->ltd_idx,
2173                                       LCK_EX, MDS_INODELOCK_LOOKUP,
2174                                       MF_MDC_CANCEL_FID3);
2175                 if (rc != 0)
2176                         RETURN(rc);
2177
2178                 rc = lmv_early_cancel(exp, NULL, op_data, src_tgt->ltd_idx,
2179                                       LCK_EX, MDS_INODELOCK_FULL,
2180                                       MF_MDC_CANCEL_FID3);
2181                 if (rc != 0)
2182                         RETURN(rc);
2183         }
2184
2185         /*
2186          * Cancel all the locks on tgt child (fid4).
2187          */
2188         if (fid_is_sane(&op_data->op_fid4))
2189                 rc = lmv_early_cancel(exp, NULL, op_data, src_tgt->ltd_idx,
2190                                       LCK_EX, MDS_INODELOCK_FULL,
2191                                       MF_MDC_CANCEL_FID4);
2192
2193         CDEBUG(D_INODE, DFID":m%d to "DFID"\n", PFID(&op_data->op_fid1),
2194                op_data->op_mds, PFID(&op_data->op_fid2));
2195
2196         rc = md_rename(src_tgt->ltd_exp, op_data, old, oldlen, new, newlen,
2197                        request);
2198
2199         RETURN(rc);
2200 }
2201
2202 static int lmv_setattr(struct obd_export *exp, struct md_op_data *op_data,
2203                         void *ea, size_t ealen, void *ea2, size_t ea2len,
2204                         struct ptlrpc_request **request,
2205                         struct md_open_data **mod)
2206 {
2207         struct obd_device       *obd = exp->exp_obd;
2208         struct lmv_obd          *lmv = &obd->u.lmv;
2209         struct lmv_tgt_desc     *tgt;
2210         int                      rc = 0;
2211         ENTRY;
2212
2213         rc = lmv_check_connect(obd);
2214         if (rc)
2215                 RETURN(rc);
2216
2217         CDEBUG(D_INODE, "SETATTR for "DFID", valid 0x%x\n",
2218                PFID(&op_data->op_fid1), op_data->op_attr.ia_valid);
2219
2220         op_data->op_flags |= MF_MDC_CANCEL_FID1;
2221         tgt = lmv_find_target(lmv, &op_data->op_fid1);
2222         if (IS_ERR(tgt))
2223                 RETURN(PTR_ERR(tgt));
2224
2225         rc = md_setattr(tgt->ltd_exp, op_data, ea, ealen, ea2,
2226                         ea2len, request, mod);
2227
2228         RETURN(rc);
2229 }
2230
2231 static int lmv_fsync(struct obd_export *exp, const struct lu_fid *fid,
2232                      struct obd_capa *oc, struct ptlrpc_request **request)
2233 {
2234         struct obd_device       *obd = exp->exp_obd;
2235         struct lmv_obd          *lmv = &obd->u.lmv;
2236         struct lmv_tgt_desc     *tgt;
2237         int                      rc;
2238         ENTRY;
2239
2240         rc = lmv_check_connect(obd);
2241         if (rc != 0)
2242                 RETURN(rc);
2243
2244         tgt = lmv_find_target(lmv, fid);
2245         if (IS_ERR(tgt))
2246                 RETURN(PTR_ERR(tgt));
2247
2248         rc = md_fsync(tgt->ltd_exp, fid, oc, request);
2249         RETURN(rc);
2250 }
2251
2252 /**
2253  * Get current minimum entry from striped directory
2254  *
2255  * This function will search the dir entry, whose hash value is the
2256  * closest(>=) to @hash_offset, from all of sub-stripes, and it is
2257  * only being called for striped directory.
2258  *
2259  * \param[in] exp               export of LMV
2260  * \param[in] op_data           parameters transferred beween client MD stack
2261  *                              stripe_information will be included in this
2262  *                              parameter
2263  * \param[in] cb_op             ldlm callback being used in enqueue in
2264  *                              mdc_read_page
2265  * \param[in] hash_offset       the hash value, which is used to locate
2266  *                              minum(closet) dir entry
2267  * \param[in|out] stripe_offset the caller use this to indicate the stripe
2268  *                              index of last entry, so to avoid hash conflict
2269  *                              between stripes. It will also be used to
2270  *                              return the stripe index of current dir entry.
2271  * \param[in|out] entp          the minum entry and it also is being used
2272  *                              to input the last dir entry to resolve the
2273  *                              hash conflict
2274  *
2275  * \param[out] ppage            the page which holds the minum entry
2276  *
2277  * \retval                      = 0 get the entry successfully
2278  *                              negative errno (< 0) does not get the entry
2279  */
2280 static int lmv_get_min_striped_entry(struct obd_export *exp,
2281                                      struct md_op_data *op_data,
2282                                      struct md_callback *cb_op,
2283                                      __u64 hash_offset, int *stripe_offset,
2284                                      struct lu_dirent **entp,
2285                                      struct page **ppage)
2286 {
2287         struct obd_device       *obd = exp->exp_obd;
2288         struct lmv_obd          *lmv = &obd->u.lmv;
2289         struct lmv_stripe_md    *lsm = op_data->op_mea1;
2290         struct lmv_tgt_desc     *tgt;
2291         int                     stripe_count;
2292         struct lu_dirent        *min_ent = NULL;
2293         struct page             *min_page = NULL;
2294         int                     min_idx = 0;
2295         int                     i;
2296         int                     rc = 0;
2297         ENTRY;
2298
2299         stripe_count = lsm->lsm_md_stripe_count;
2300         for (i = 0; i < stripe_count; i++) {
2301                 struct lu_dirent        *ent = NULL;
2302                 struct page             *page = NULL;
2303                 struct lu_dirpage       *dp;
2304                 __u64                   stripe_hash = hash_offset;
2305
2306                 tgt = lmv_get_target(lmv, lsm->lsm_md_oinfo[i].lmo_mds, NULL);
2307                 if (IS_ERR(tgt))
2308                         GOTO(out, rc = PTR_ERR(tgt));
2309
2310                 /* op_data will be shared by each stripe, so we need
2311                  * reset these value for each stripe */
2312                 op_data->op_fid1 = lsm->lsm_md_oinfo[i].lmo_fid;
2313                 op_data->op_fid2 = lsm->lsm_md_oinfo[i].lmo_fid;
2314                 op_data->op_data = lsm->lsm_md_oinfo[i].lmo_root;
2315 next:
2316                 rc = md_read_page(tgt->ltd_exp, op_data, cb_op, stripe_hash,
2317                                   &page);
2318                 if (rc != 0)
2319                         GOTO(out, rc);
2320
2321                 dp = page_address(page);
2322                 for (ent = lu_dirent_start(dp); ent != NULL;
2323                      ent = lu_dirent_next(ent)) {
2324                         /* Skip dummy entry */
2325                         if (le16_to_cpu(ent->lde_namelen) == 0)
2326                                 continue;
2327
2328                         if (le64_to_cpu(ent->lde_hash) < hash_offset)
2329                                 continue;
2330
2331                         if (le64_to_cpu(ent->lde_hash) == hash_offset &&
2332                             (*entp == ent || i < *stripe_offset))
2333                                 continue;
2334
2335                         /* skip . and .. for other stripes */
2336                         if (i != 0 &&
2337                             (strncmp(ent->lde_name, ".",
2338                                      le16_to_cpu(ent->lde_namelen)) == 0 ||
2339                              strncmp(ent->lde_name, "..",
2340                                      le16_to_cpu(ent->lde_namelen)) == 0))
2341                                 continue;
2342                         break;
2343                 }
2344
2345                 if (ent == NULL) {
2346                         stripe_hash = le64_to_cpu(dp->ldp_hash_end);
2347
2348                         kunmap(page);
2349                         page_cache_release(page);
2350                         page = NULL;
2351
2352                         /* reach the end of current stripe, go to next stripe */
2353                         if (stripe_hash == MDS_DIR_END_OFF)
2354                                 continue;
2355                         else
2356                                 goto next;
2357                 }
2358
2359                 if (min_ent != NULL) {
2360                         if (le64_to_cpu(min_ent->lde_hash) >
2361                             le64_to_cpu(ent->lde_hash)) {
2362                                 min_ent = ent;
2363                                 kunmap(min_page);
2364                                 page_cache_release(min_page);
2365                                 min_idx = i;
2366                                 min_page = page;
2367                         } else {
2368                                 kunmap(page);
2369                                 page_cache_release(page);
2370                                 page = NULL;
2371                         }
2372                 } else {
2373                         min_ent = ent;
2374                         min_page = page;
2375                         min_idx = i;
2376                 }
2377         }
2378
2379 out:
2380         if (*ppage != NULL) {
2381                 kunmap(*ppage);
2382                 page_cache_release(*ppage);
2383         }
2384         *stripe_offset = min_idx;
2385         *entp = min_ent;
2386         *ppage = min_page;
2387         RETURN(rc);
2388 }
2389
2390 /**
2391  * Build dir entry page from a striped directory
2392  *
2393  * This function gets one entry by @offset from a striped directory. It will
2394  * read entries from all of stripes, and choose one closest to the required
2395  * offset(&offset). A few notes
2396  * 1. skip . and .. for non-zero stripes, because there can only have one .
2397  * and .. in a directory.
2398  * 2. op_data will be shared by all of stripes, instead of allocating new
2399  * one, so need to restore before reusing.
2400  * 3. release the entry page if that is not being chosen.
2401  *
2402  * \param[in] exp       obd export refer to LMV
2403  * \param[in] op_data   hold those MD parameters of read_entry
2404  * \param[in] cb_op     ldlm callback being used in enqueue in mdc_read_entry
2405  * \param[out] ldp      the entry being read
2406  * \param[out] ppage    the page holding the entry. Note: because the entry
2407  *                      will be accessed in upper layer, so we need hold the
2408  *                      page until the usages of entry is finished, see
2409  *                      ll_dir_entry_next.
2410  *
2411  * retval               =0 if get entry successfully
2412  *                      <0 cannot get entry
2413  */
2414 static int lmv_read_striped_page(struct obd_export *exp,
2415                                  struct md_op_data *op_data,
2416                                  struct md_callback *cb_op,
2417                                  __u64 offset, struct page **ppage)
2418 {
2419         struct obd_device       *obd = exp->exp_obd;
2420         struct lu_fid           master_fid = op_data->op_fid1;
2421         struct inode            *master_inode = op_data->op_data;
2422         __u64                   hash_offset = offset;
2423         struct lu_dirpage       *dp;
2424         struct page             *min_ent_page = NULL;
2425         struct page             *ent_page = NULL;
2426         struct lu_dirent        *ent;
2427         void                    *area;
2428         int                     ent_idx = 0;
2429         struct lu_dirent        *min_ent = NULL;
2430         struct lu_dirent        *last_ent;
2431         size_t                  left_bytes;
2432         int                     rc;
2433         ENTRY;
2434
2435         rc = lmv_check_connect(obd);
2436         if (rc)
2437                 RETURN(rc);
2438
2439         /* Allocate a page and read entries from all of stripes and fill
2440          * the page by hash order */
2441         ent_page = alloc_page(GFP_KERNEL);
2442         if (ent_page == NULL)
2443                 RETURN(-ENOMEM);
2444
2445         /* Initialize the entry page */
2446         dp = kmap(ent_page);
2447         memset(dp, 0, sizeof(*dp));
2448         dp->ldp_hash_start = cpu_to_le64(offset);
2449         dp->ldp_flags |= LDF_COLLIDE;
2450
2451         area = dp + 1;
2452         left_bytes = PAGE_CACHE_SIZE - sizeof(*dp);
2453         ent = area;
2454         last_ent = ent;
2455         do {
2456                 __u16   ent_size;
2457
2458                 /* Find the minum entry from all sub-stripes */
2459                 rc = lmv_get_min_striped_entry(exp, op_data, cb_op, hash_offset,
2460                                                &ent_idx, &min_ent,
2461                                                &min_ent_page);
2462                 if (rc != 0)
2463                         GOTO(out, rc);
2464
2465                 /* If it can not get minum entry, it means it already reaches
2466                  * the end of this directory */
2467                 if (min_ent == NULL) {
2468                         last_ent->lde_reclen = 0;
2469                         hash_offset = MDS_DIR_END_OFF;
2470                         GOTO(out, rc);
2471                 }
2472
2473                 ent_size = le16_to_cpu(min_ent->lde_reclen);
2474
2475                 /* the last entry lde_reclen is 0, but it might not
2476                  * the end of this entry of this temporay entry */
2477                 if (ent_size == 0)
2478                         ent_size = lu_dirent_calc_size(
2479                                         le16_to_cpu(min_ent->lde_namelen),
2480                                         le32_to_cpu(min_ent->lde_attrs));
2481                 if (ent_size > left_bytes) {
2482                         last_ent->lde_reclen = cpu_to_le16(0);
2483                         hash_offset = le64_to_cpu(min_ent->lde_hash);
2484                         GOTO(out, rc);
2485                 }
2486
2487                 memcpy(ent, min_ent, ent_size);
2488
2489                 /* Replace . with master FID and Replace .. with the parent FID
2490                  * of master object */
2491                 if (strncmp(ent->lde_name, ".",
2492                             le16_to_cpu(ent->lde_namelen)) == 0 &&
2493                     le16_to_cpu(ent->lde_namelen) == 1)
2494                         fid_cpu_to_le(&ent->lde_fid, &master_fid);
2495                 else if (strncmp(ent->lde_name, "..",
2496                                    le16_to_cpu(ent->lde_namelen)) == 0 &&
2497                            le16_to_cpu(ent->lde_namelen) == 2)
2498                         fid_cpu_to_le(&ent->lde_fid, &op_data->op_fid3);
2499
2500                 left_bytes -= ent_size;
2501                 ent->lde_reclen = cpu_to_le16(ent_size);
2502                 last_ent = ent;
2503                 ent = (void *)ent + ent_size;
2504                 hash_offset = le64_to_cpu(min_ent->lde_hash);
2505                 if (hash_offset == MDS_DIR_END_OFF) {
2506                         last_ent->lde_reclen = 0;
2507                         break;
2508                 }
2509         } while (1);
2510 out:
2511         if (min_ent_page != NULL) {
2512                 kunmap(min_ent_page);
2513                 page_cache_release(min_ent_page);
2514         }
2515
2516         if (unlikely(rc != 0)) {
2517                 __free_page(ent_page);
2518                 ent_page = NULL;
2519         } else {
2520                 if (ent == area)
2521                         dp->ldp_flags |= LDF_EMPTY;
2522                 dp->ldp_flags = cpu_to_le32(dp->ldp_flags);
2523                 dp->ldp_hash_end = cpu_to_le64(hash_offset);
2524         }
2525
2526         /* We do not want to allocate md_op_data during each
2527          * dir entry reading, so op_data will be shared by every stripe,
2528          * then we need to restore it back to original value before
2529          * return to the upper layer */
2530         op_data->op_fid1 = master_fid;
2531         op_data->op_fid2 = master_fid;
2532         op_data->op_data = master_inode;
2533
2534         *ppage = ent_page;
2535
2536         RETURN(rc);
2537 }
2538
2539 int lmv_read_page(struct obd_export *exp, struct md_op_data *op_data,
2540                   struct md_callback *cb_op, __u64 offset,
2541                   struct page **ppage)
2542 {
2543         struct obd_device       *obd = exp->exp_obd;
2544         struct lmv_obd          *lmv = &obd->u.lmv;
2545         struct lmv_stripe_md    *lsm = op_data->op_mea1;
2546         struct lmv_tgt_desc     *tgt;
2547         int                     rc;
2548         ENTRY;
2549
2550         rc = lmv_check_connect(obd);
2551         if (rc != 0)
2552                 RETURN(rc);
2553
2554         if (unlikely(lsm != NULL)) {
2555                 rc = lmv_read_striped_page(exp, op_data, cb_op, offset, ppage);
2556                 RETURN(rc);
2557         }
2558
2559         tgt = lmv_find_target(lmv, &op_data->op_fid1);
2560         if (IS_ERR(tgt))
2561                 RETURN(PTR_ERR(tgt));
2562
2563         rc = md_read_page(tgt->ltd_exp, op_data, cb_op, offset, ppage);
2564
2565         RETURN(rc);
2566 }
2567
2568 /**
2569  * Unlink a file/directory
2570  *
2571  * Unlink a file or directory under the parent dir. The unlink request
2572  * usually will be sent to the MDT where the child is located, but if
2573  * the client does not have the child FID then request will be sent to the
2574  * MDT where the parent is located.
2575  *
2576  * If the parent is a striped directory then it also needs to locate which
2577  * stripe the name of the child is located, and replace the parent FID
2578  * (@op->op_fid1) with the stripe FID. Note: if the stripe is unknown,
2579  * it will walk through all of sub-stripes until the child is being
2580  * unlinked finally.
2581  *
2582  * \param[in] exp       export refer to LMV
2583  * \param[in] op_data   different parameters transferred beween client
2584  *                      MD stacks, name, namelen, FIDs etc.
2585  *                      op_fid1 is the parent FID, op_fid2 is the child
2586  *                      FID.
2587  * \param[out] request  point to the request of unlink.
2588  *
2589  * retval               0 if succeed
2590  *                      negative errno if failed.
2591  */
2592 static int lmv_unlink(struct obd_export *exp, struct md_op_data *op_data,
2593                       struct ptlrpc_request **request)
2594 {
2595         struct obd_device       *obd = exp->exp_obd;
2596         struct lmv_obd          *lmv = &obd->u.lmv;
2597         struct lmv_tgt_desc     *tgt = NULL;
2598         struct lmv_tgt_desc     *parent_tgt = NULL;
2599         struct mdt_body         *body;
2600         int                     rc;
2601         int                     stripe_index = 0;
2602         struct lmv_stripe_md    *lsm = op_data->op_mea1;
2603         ENTRY;
2604
2605         rc = lmv_check_connect(obd);
2606         if (rc)
2607                 RETURN(rc);
2608 retry_unlink:
2609         /* For striped dir, we need to locate the parent as well */
2610         if (lsm != NULL) {
2611                 struct lmv_tgt_desc *tmp;
2612
2613                 LASSERT(op_data->op_name != NULL &&
2614                         op_data->op_namelen != 0);
2615
2616                 tmp = lmv_locate_target_for_name(lmv, lsm,
2617                                                  op_data->op_name,
2618                                                  op_data->op_namelen,
2619                                                  &op_data->op_fid1,
2620                                                  &op_data->op_mds);
2621
2622                 /* return -EBADFD means unknown hash type, might
2623                  * need try all sub-stripe here */
2624                 if (IS_ERR(tmp) && PTR_ERR(tmp) != -EBADFD)
2625                         RETURN(PTR_ERR(tmp));
2626
2627                 /* Note: both migrating dir and unknown hash dir need to
2628                  * try all of sub-stripes, so we need start search the
2629                  * name from stripe 0, but migrating dir is already handled
2630                  * inside lmv_locate_target_for_name(), so we only check
2631                  * unknown hash type directory here */
2632                 if (!lmv_is_known_hash_type(lsm->lsm_md_hash_type)) {
2633                         struct lmv_oinfo *oinfo;
2634
2635                         oinfo = &lsm->lsm_md_oinfo[stripe_index];
2636
2637                         op_data->op_fid1 = oinfo->lmo_fid;
2638                         op_data->op_mds = oinfo->lmo_mds;
2639                 }
2640         }
2641
2642 try_next_stripe:
2643         /* Send unlink requests to the MDT where the child is located */
2644         if (likely(!fid_is_zero(&op_data->op_fid2)))
2645                 tgt = lmv_find_target(lmv, &op_data->op_fid2);
2646         else if (lsm != NULL)
2647                 tgt = lmv_get_target(lmv, op_data->op_mds, NULL);
2648         else
2649                 tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
2650
2651         if (IS_ERR(tgt))
2652                 RETURN(PTR_ERR(tgt));
2653
2654         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2655         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2656         op_data->op_cap = cfs_curproc_cap_pack();
2657
2658         /*
2659          * If child's fid is given, cancel unused locks for it if it is from
2660          * another export than parent.
2661          *
2662          * LOOKUP lock for child (fid3) should also be cancelled on parent
2663          * tgt_tgt in mdc_unlink().
2664          */
2665         op_data->op_flags |= MF_MDC_CANCEL_FID1 | MF_MDC_CANCEL_FID3;
2666
2667         /*
2668          * Cancel FULL locks on child (fid3).
2669          */
2670         parent_tgt = lmv_find_target(lmv, &op_data->op_fid1);
2671         if (IS_ERR(parent_tgt))
2672                 RETURN(PTR_ERR(parent_tgt));
2673
2674         if (parent_tgt != tgt) {
2675                 rc = lmv_early_cancel(exp, parent_tgt, op_data, tgt->ltd_idx,
2676                                       LCK_EX, MDS_INODELOCK_LOOKUP,
2677                                       MF_MDC_CANCEL_FID3);
2678         }
2679
2680         rc = lmv_early_cancel(exp, NULL, op_data, tgt->ltd_idx, LCK_EX,
2681                               MDS_INODELOCK_FULL, MF_MDC_CANCEL_FID3);
2682         if (rc != 0)
2683                 RETURN(rc);
2684
2685         CDEBUG(D_INODE, "unlink with fid="DFID"/"DFID" -> mds #%u\n",
2686                PFID(&op_data->op_fid1), PFID(&op_data->op_fid2), tgt->ltd_idx);
2687
2688         rc = md_unlink(tgt->ltd_exp, op_data, request);
2689         if (rc != 0 && rc != -EREMOTE && rc != -ENOENT)
2690                 RETURN(rc);
2691
2692         /* Try next stripe if it is needed. */
2693         if (rc == -ENOENT && lsm != NULL && lmv_need_try_all_stripes(lsm)) {
2694                 struct lmv_oinfo *oinfo;
2695
2696                 stripe_index++;
2697                 if (stripe_index >= lsm->lsm_md_stripe_count)
2698                         RETURN(rc);
2699
2700                 oinfo = &lsm->lsm_md_oinfo[stripe_index];
2701
2702                 op_data->op_fid1 = oinfo->lmo_fid;
2703                 op_data->op_mds = oinfo->lmo_mds;
2704
2705                 ptlrpc_req_finished(*request);
2706                 *request = NULL;
2707
2708                 goto try_next_stripe;
2709         }
2710
2711         body = req_capsule_server_get(&(*request)->rq_pill, &RMF_MDT_BODY);
2712         if (body == NULL)
2713                 RETURN(-EPROTO);
2714
2715         /* Not cross-ref case, just get out of here. */
2716         if (likely(!(body->mbo_valid & OBD_MD_MDS)))
2717                 RETURN(rc);
2718
2719         CDEBUG(D_INODE, "%s: try unlink to another MDT for "DFID"\n",
2720                exp->exp_obd->obd_name, PFID(&body->mbo_fid1));
2721
2722         /* This is a remote object, try remote MDT, Note: it may
2723          * try more than 1 time here, Considering following case
2724          * /mnt/lustre is root on MDT0, remote1 is on MDT1
2725          * 1. Initially A does not know where remote1 is, it send
2726          *    unlink RPC to MDT0, MDT0 return -EREMOTE, it will
2727          *    resend unlink RPC to MDT1 (retry 1st time).
2728          *
2729          * 2. During the unlink RPC in flight,
2730          *    client B mv /mnt/lustre/remote1 /mnt/lustre/remote2
2731          *    and create new remote1, but on MDT0
2732          *
2733          * 3. MDT1 get unlink RPC(from A), then do remote lock on
2734          *    /mnt/lustre, then lookup get fid of remote1, and find
2735          *    it is remote dir again, and replay -EREMOTE again.
2736          *
2737          * 4. Then A will resend unlink RPC to MDT0. (retry 2nd times).
2738          *
2739          * In theory, it might try unlimited time here, but it should
2740          * be very rare case.  */
2741         op_data->op_fid2 = body->mbo_fid1;
2742         ptlrpc_req_finished(*request);
2743         *request = NULL;
2744
2745         goto retry_unlink;
2746 }
2747
2748 static int lmv_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
2749 {
2750         struct lmv_obd *lmv = &obd->u.lmv;
2751         int rc = 0;
2752
2753         switch (stage) {
2754         case OBD_CLEANUP_EARLY:
2755                 /* XXX: here should be calling obd_precleanup() down to
2756                  * stack. */
2757                 break;
2758         case OBD_CLEANUP_EXPORTS:
2759                 fld_client_proc_fini(&lmv->lmv_fld);
2760                 lprocfs_obd_cleanup(obd);
2761                 lprocfs_free_md_stats(obd);
2762                 break;
2763         default:
2764                 break;
2765         }
2766         RETURN(rc);
2767 }
2768
2769 /**
2770  * Get by key a value associated with a LMV device.
2771  *
2772  * Dispatch request to lower-layer devices as needed.
2773  *
2774  * \param[in] env               execution environment for this thread
2775  * \param[in] exp               export for the LMV device
2776  * \param[in] keylen            length of key identifier
2777  * \param[in] key               identifier of key to get value for
2778  * \param[in] vallen            size of \a val
2779  * \param[out] val              pointer to storage location for value
2780  * \param[in] lsm               optional striping metadata of object
2781  *
2782  * \retval 0            on success
2783  * \retval negative     negated errno on failure
2784  */
2785 static int lmv_get_info(const struct lu_env *env, struct obd_export *exp,
2786                         __u32 keylen, void *key, __u32 *vallen, void *val,
2787                         struct lov_stripe_md *lsm)
2788 {
2789         struct obd_device       *obd;
2790         struct lmv_obd          *lmv;
2791         int                      rc = 0;
2792         ENTRY;
2793
2794         obd = class_exp2obd(exp);
2795         if (obd == NULL) {
2796                 CDEBUG(D_IOCTL, "Invalid client cookie "LPX64"\n",
2797                        exp->exp_handle.h_cookie);
2798                 RETURN(-EINVAL);
2799         }
2800
2801         lmv = &obd->u.lmv;
2802         if (keylen >= strlen("remote_flag") && !strcmp(key, "remote_flag")) {
2803                 int i;
2804
2805                 rc = lmv_check_connect(obd);
2806                 if (rc)
2807                         RETURN(rc);
2808
2809                 LASSERT(*vallen == sizeof(__u32));
2810                 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2811                         struct lmv_tgt_desc *tgt = lmv->tgts[i];
2812                         /*
2813                          * All tgts should be connected when this gets called.
2814                          */
2815                         if (tgt == NULL || tgt->ltd_exp == NULL)
2816                                 continue;
2817
2818                         if (!obd_get_info(env, tgt->ltd_exp, keylen, key,
2819                                           vallen, val, NULL))
2820                                 RETURN(0);
2821                 }
2822                 RETURN(-EINVAL);
2823         } else if (KEY_IS(KEY_MAX_EASIZE) ||
2824                    KEY_IS(KEY_DEFAULT_EASIZE) ||
2825                    KEY_IS(KEY_MAX_COOKIESIZE) ||
2826                    KEY_IS(KEY_DEFAULT_COOKIESIZE) ||
2827                    KEY_IS(KEY_CONN_DATA)) {
2828                 rc = lmv_check_connect(obd);
2829                 if (rc)
2830                         RETURN(rc);
2831
2832                 /*
2833                  * Forwarding this request to first MDS, it should know LOV
2834                  * desc.
2835                  */
2836                 rc = obd_get_info(env, lmv->tgts[0]->ltd_exp, keylen, key,
2837                                   vallen, val, NULL);
2838                 if (!rc && KEY_IS(KEY_CONN_DATA))
2839                         exp->exp_connect_data = *(struct obd_connect_data *)val;
2840                 RETURN(rc);
2841         } else if (KEY_IS(KEY_TGT_COUNT)) {
2842                 *((int *)val) = lmv->desc.ld_tgt_count;
2843                 RETURN(0);
2844         }
2845
2846         CDEBUG(D_IOCTL, "Invalid key\n");
2847         RETURN(-EINVAL);
2848 }
2849
2850 /**
2851  * Asynchronously set by key a value associated with a LMV device.
2852  *
2853  * Dispatch request to lower-layer devices as needed.
2854  *
2855  * \param[in] env       execution environment for this thread
2856  * \param[in] exp       export for the LMV device
2857  * \param[in] keylen    length of key identifier
2858  * \param[in] key       identifier of key to store value for
2859  * \param[in] vallen    size of value to store
2860  * \param[in] val       pointer to data to be stored
2861  * \param[in] set       optional list of related ptlrpc requests
2862  *
2863  * \retval 0            on success
2864  * \retval negative     negated errno on failure
2865  */
2866 int lmv_set_info_async(const struct lu_env *env, struct obd_export *exp,
2867                        obd_count keylen, void *key, obd_count vallen,
2868                        void *val, struct ptlrpc_request_set *set)
2869 {
2870         struct lmv_tgt_desc     *tgt = NULL;
2871         struct obd_device       *obd;
2872         struct lmv_obd          *lmv;
2873         int rc = 0;
2874         ENTRY;
2875
2876         obd = class_exp2obd(exp);
2877         if (obd == NULL) {
2878                 CDEBUG(D_IOCTL, "Invalid client cookie "LPX64"\n",
2879                        exp->exp_handle.h_cookie);
2880                 RETURN(-EINVAL);
2881         }
2882         lmv = &obd->u.lmv;
2883
2884         if (KEY_IS(KEY_READ_ONLY) || KEY_IS(KEY_FLUSH_CTX) ||
2885             KEY_IS(KEY_DEFAULT_EASIZE)) {
2886                 int i, err = 0;
2887
2888                 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2889                         tgt = lmv->tgts[i];
2890
2891                         if (tgt == NULL || tgt->ltd_exp == NULL)
2892                                 continue;
2893
2894                         err = obd_set_info_async(env, tgt->ltd_exp,
2895                                                  keylen, key, vallen, val, set);
2896                         if (err && rc == 0)
2897                                 rc = err;
2898                 }
2899
2900                 RETURN(rc);
2901         }
2902
2903         RETURN(-EINVAL);
2904 }
2905
2906 static int lmv_pack_md_v1(const struct lmv_stripe_md *lsm,
2907                           struct lmv_mds_md_v1 *lmm1)
2908 {
2909         int     cplen;
2910         int     i;
2911
2912         lmm1->lmv_magic = cpu_to_le32(lsm->lsm_md_magic);
2913         lmm1->lmv_stripe_count = cpu_to_le32(lsm->lsm_md_stripe_count);
2914         lmm1->lmv_master_mdt_index = cpu_to_le32(lsm->lsm_md_master_mdt_index);
2915         lmm1->lmv_hash_type = cpu_to_le32(lsm->lsm_md_hash_type);
2916         cplen = strlcpy(lmm1->lmv_pool_name, lsm->lsm_md_pool_name,
2917                         sizeof(lmm1->lmv_pool_name));
2918         if (cplen >= sizeof(lmm1->lmv_pool_name))
2919                 return -E2BIG;
2920
2921         for (i = 0; i < lsm->lsm_md_stripe_count; i++)
2922                 fid_cpu_to_le(&lmm1->lmv_stripe_fids[i],
2923                               &lsm->lsm_md_oinfo[i].lmo_fid);
2924         return 0;
2925 }
2926
2927 int lmv_pack_md(union lmv_mds_md **lmmp, const struct lmv_stripe_md *lsm,
2928                 int stripe_count)
2929 {
2930         int     lmm_size = 0;
2931         bool    allocated = false;
2932         int     rc = 0;
2933         ENTRY;
2934
2935         LASSERT(lmmp != NULL);
2936         /* Free lmm */
2937         if (*lmmp != NULL && lsm == NULL) {
2938                 int stripe_count;
2939
2940                 stripe_count = lmv_mds_md_stripe_count_get(*lmmp);
2941                 lmm_size = lmv_mds_md_size(stripe_count,
2942                                            le32_to_cpu((*lmmp)->lmv_magic));
2943                 if (lmm_size == 0)
2944                         RETURN(-EINVAL);
2945                 OBD_FREE(*lmmp, lmm_size);
2946                 *lmmp = NULL;
2947                 RETURN(0);
2948         }
2949
2950         /* Alloc lmm */
2951         if (*lmmp == NULL && lsm == NULL) {
2952                 lmm_size = lmv_mds_md_size(stripe_count, LMV_MAGIC);
2953                 LASSERT(lmm_size > 0);
2954                 OBD_ALLOC(*lmmp, lmm_size);
2955                 if (*lmmp == NULL)
2956                         RETURN(-ENOMEM);
2957                 lmv_mds_md_stripe_count_set(*lmmp, stripe_count);
2958                 (*lmmp)->lmv_magic = cpu_to_le32(LMV_MAGIC);
2959                 RETURN(lmm_size);
2960         }
2961
2962         /* pack lmm */
2963         LASSERT(lsm != NULL);
2964         lmm_size = lmv_mds_md_size(lsm->lsm_md_stripe_count, lsm->lsm_md_magic);
2965         if (*lmmp == NULL) {
2966                 OBD_ALLOC(*lmmp, lmm_size);
2967                 if (*lmmp == NULL)
2968                         RETURN(-ENOMEM);
2969                 allocated = true;
2970         }
2971
2972         switch (lsm->lsm_md_magic) {
2973         case LMV_MAGIC_V1:
2974                 rc = lmv_pack_md_v1(lsm, &(*lmmp)->lmv_md_v1);
2975                 break;
2976         default:
2977                 rc = -EINVAL;
2978                 break;
2979         }
2980
2981         if (rc != 0 && allocated) {
2982                 OBD_FREE(*lmmp, lmm_size);
2983                 *lmmp = NULL;
2984         }
2985
2986         RETURN(lmm_size);
2987 }
2988
2989 static int lmv_unpack_md_v1(struct obd_export *exp, struct lmv_stripe_md *lsm,
2990                             const struct lmv_mds_md_v1 *lmm1)
2991 {
2992         struct lmv_obd  *lmv = &exp->exp_obd->u.lmv;
2993         int             stripe_count;
2994         int             cplen;
2995         int             i;
2996         int             rc = 0;
2997         ENTRY;
2998
2999         lsm->lsm_md_magic = le32_to_cpu(lmm1->lmv_magic);
3000         lsm->lsm_md_stripe_count = le32_to_cpu(lmm1->lmv_stripe_count);
3001         lsm->lsm_md_master_mdt_index = le32_to_cpu(lmm1->lmv_master_mdt_index);
3002         if (OBD_FAIL_CHECK(OBD_FAIL_UNKNOWN_LMV_STRIPE))
3003                 lsm->lsm_md_hash_type = LMV_HASH_TYPE_UNKNOWN;
3004         else
3005                 lsm->lsm_md_hash_type = le32_to_cpu(lmm1->lmv_hash_type);
3006         lsm->lsm_md_layout_version = le32_to_cpu(lmm1->lmv_layout_version);
3007         cplen = strlcpy(lsm->lsm_md_pool_name, lmm1->lmv_pool_name,
3008                         sizeof(lsm->lsm_md_pool_name));
3009
3010         if (cplen >= sizeof(lsm->lsm_md_pool_name))
3011                 RETURN(-E2BIG);
3012
3013         CDEBUG(D_INFO, "unpack lsm count %d, master %d hash_type %d"
3014                "layout_version %d\n", lsm->lsm_md_stripe_count,
3015                lsm->lsm_md_master_mdt_index, lsm->lsm_md_hash_type,
3016                lsm->lsm_md_layout_version);
3017
3018         stripe_count = le32_to_cpu(lmm1->lmv_stripe_count);
3019         for (i = 0; i < stripe_count; i++) {
3020                 fid_le_to_cpu(&lsm->lsm_md_oinfo[i].lmo_fid,
3021                               &lmm1->lmv_stripe_fids[i]);
3022                 rc = lmv_fld_lookup(lmv, &lsm->lsm_md_oinfo[i].lmo_fid,
3023                                     &lsm->lsm_md_oinfo[i].lmo_mds);
3024                 if (rc != 0)
3025                         RETURN(rc);
3026                 CDEBUG(D_INFO, "unpack fid #%d "DFID"\n", i,
3027                        PFID(&lsm->lsm_md_oinfo[i].lmo_fid));
3028         }
3029
3030         RETURN(rc);
3031 }
3032
3033 int lmv_unpack_md(struct obd_export *exp, struct lmv_stripe_md **lsmp,
3034                   const union lmv_mds_md *lmm, int stripe_count)
3035 {
3036         struct lmv_stripe_md     *lsm;
3037         int                      lsm_size;
3038         int                      rc;
3039         bool                     allocated = false;
3040         ENTRY;
3041
3042         LASSERT(lsmp != NULL);
3043
3044         lsm = *lsmp;
3045         /* Free memmd */
3046         if (lsm != NULL && lmm == NULL) {
3047                 int i;
3048                 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
3049                         /* For migrating inode, the master stripe and master
3050                          * object will be the same, so do not need iput, see
3051                          * ll_update_lsm_md */
3052                         if (!(lsm->lsm_md_hash_type & LMV_HASH_FLAG_MIGRATION &&
3053                               i == 0) && lsm->lsm_md_oinfo[i].lmo_root != NULL)
3054                                 iput(lsm->lsm_md_oinfo[i].lmo_root);
3055                 }
3056                 lsm_size = lmv_stripe_md_size(lsm->lsm_md_stripe_count);
3057                 OBD_FREE(lsm, lsm_size);
3058                 *lsmp = NULL;
3059                 RETURN(0);
3060         }
3061
3062         /* Alloc memmd */
3063         if (lsm == NULL && lmm == NULL) {
3064                 lsm_size = lmv_stripe_md_size(stripe_count);
3065                 OBD_ALLOC(lsm, lsm_size);
3066                 if (lsm == NULL)
3067                         RETURN(-ENOMEM);
3068                 lsm->lsm_md_stripe_count = stripe_count;
3069                 *lsmp = lsm;
3070                 RETURN(0);
3071         }
3072
3073         if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_STRIPE)
3074                 RETURN(-EPERM);
3075
3076         /* Unpack memmd */
3077         if (le32_to_cpu(lmm->lmv_magic) != LMV_MAGIC_V1 &&
3078             le32_to_cpu(lmm->lmv_magic) != LMV_USER_MAGIC) {
3079                 CERROR("%s: invalid lmv magic %x: rc = %d\n",
3080                        exp->exp_obd->obd_name, le32_to_cpu(lmm->lmv_magic),
3081                        -EIO);
3082                 RETURN(-EIO);
3083         }
3084
3085         if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_V1)
3086                 lsm_size = lmv_stripe_md_size(lmv_mds_md_stripe_count_get(lmm));
3087         else
3088                 /**
3089                  * Unpack default dirstripe(lmv_user_md) to lmv_stripe_md,
3090                  * stripecount should be 0 then.
3091                  */
3092                 lsm_size = lmv_stripe_md_size(0);
3093
3094         lsm_size = lmv_stripe_md_size(lmv_mds_md_stripe_count_get(lmm));
3095         if (lsm == NULL) {
3096                 OBD_ALLOC(lsm, lsm_size);
3097                 if (lsm == NULL)
3098                         RETURN(-ENOMEM);
3099                 allocated = true;
3100                 *lsmp = lsm;
3101         }
3102
3103         switch (le32_to_cpu(lmm->lmv_magic)) {
3104         case LMV_MAGIC_V1:
3105                 rc = lmv_unpack_md_v1(exp, lsm, &lmm->lmv_md_v1);
3106                 break;
3107         default:
3108                 CERROR("%s: unrecognized magic %x\n", exp->exp_obd->obd_name,
3109                        le32_to_cpu(lmm->lmv_magic));
3110                 rc = -EINVAL;
3111                 break;
3112         }
3113
3114         if (rc != 0 && allocated) {
3115                 OBD_FREE(lsm, lsm_size);
3116                 *lsmp = NULL;
3117                 lsm_size = rc;
3118         }
3119         RETURN(lsm_size);
3120 }
3121
3122 int lmv_alloc_memmd(struct lmv_stripe_md **lsmp, int stripes)
3123 {
3124         return lmv_unpack_md(NULL, lsmp, NULL, stripes);
3125 }
3126
3127 void lmv_free_memmd(struct lmv_stripe_md *lsm)
3128 {
3129         lmv_unpack_md(NULL, &lsm, NULL, 0);
3130 }
3131 EXPORT_SYMBOL(lmv_free_memmd);
3132
3133 int lmv_unpackmd(struct obd_export *exp, struct lov_stripe_md **lsmp,
3134                  struct lov_mds_md *lmm, int disk_len)
3135 {
3136         return lmv_unpack_md(exp, (struct lmv_stripe_md **)lsmp,
3137                              (union lmv_mds_md *)lmm, disk_len);
3138 }
3139
3140 int lmv_packmd(struct obd_export *exp, struct lov_mds_md **lmmp,
3141                struct lov_stripe_md *lsm)
3142 {
3143         struct obd_device               *obd = exp->exp_obd;
3144         struct lmv_obd                  *lmv_obd = &obd->u.lmv;
3145         const struct lmv_stripe_md      *lmv = (struct lmv_stripe_md *)lsm;
3146         int                             stripe_count;
3147
3148         if (lmmp == NULL) {
3149                 if (lsm != NULL)
3150                         stripe_count = lmv->lsm_md_stripe_count;
3151                 else
3152                         stripe_count = lmv_obd->desc.ld_tgt_count;
3153
3154                 return lmv_mds_md_size(stripe_count, LMV_MAGIC_V1);
3155         }
3156
3157         return lmv_pack_md((union lmv_mds_md **)lmmp, lmv, 0);
3158 }
3159
3160 static int lmv_cancel_unused(struct obd_export *exp, const struct lu_fid *fid,
3161                              ldlm_policy_data_t *policy, ldlm_mode_t mode,
3162                              ldlm_cancel_flags_t flags, void *opaque)
3163 {
3164         struct obd_device       *obd = exp->exp_obd;
3165         struct lmv_obd          *lmv = &obd->u.lmv;
3166         int                      rc = 0;
3167         int                      err;
3168         __u32                    i;
3169         ENTRY;
3170
3171         LASSERT(fid != NULL);
3172
3173         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
3174                 struct lmv_tgt_desc *tgt = lmv->tgts[i];
3175
3176                 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active)
3177                         continue;
3178
3179                 err = md_cancel_unused(tgt->ltd_exp, fid, policy, mode, flags,
3180                                        opaque);
3181                 if (!rc)
3182                         rc = err;
3183         }
3184         RETURN(rc);
3185 }
3186
3187 int lmv_set_lock_data(struct obd_export *exp, __u64 *lockh, void *data,
3188                       __u64 *bits)
3189 {
3190         struct lmv_obd          *lmv = &exp->exp_obd->u.lmv;
3191         struct lmv_tgt_desc     *tgt = lmv->tgts[0];
3192         int                      rc;
3193         ENTRY;
3194
3195         if (tgt == NULL || tgt->ltd_exp == NULL)
3196                 RETURN(-EINVAL);
3197         rc =  md_set_lock_data(tgt->ltd_exp, lockh, data, bits);
3198         RETURN(rc);
3199 }
3200
3201 ldlm_mode_t lmv_lock_match(struct obd_export *exp, __u64 flags,
3202                            const struct lu_fid *fid, ldlm_type_t type,
3203                            ldlm_policy_data_t *policy, ldlm_mode_t mode,
3204                            struct lustre_handle *lockh)
3205 {
3206         struct obd_device       *obd = exp->exp_obd;
3207         struct lmv_obd          *lmv = &obd->u.lmv;
3208         ldlm_mode_t             rc;
3209         int                     tgt;
3210         int                     i;
3211         ENTRY;
3212
3213         CDEBUG(D_INODE, "Lock match for "DFID"\n", PFID(fid));
3214
3215         /*
3216          * With DNE every object can have two locks in different namespaces:
3217          * lookup lock in space of MDT storing direntry and update/open lock in
3218          * space of MDT storing inode.  Try the MDT that the FID maps to first,
3219          * since this can be easily found, and only try others if that fails.
3220          */
3221         for (i = 0, tgt = lmv_find_target_index(lmv, fid);
3222              i < lmv->desc.ld_tgt_count;
3223              i++, tgt = (tgt + 1) % lmv->desc.ld_tgt_count) {
3224                 if (tgt < 0) {
3225                         CDEBUG(D_HA, "%s: "DFID" is inaccessible: rc = %d\n",
3226                                obd->obd_name, PFID(fid), tgt);
3227                         tgt = 0;
3228                 }
3229
3230                 if (lmv->tgts[tgt] == NULL ||
3231                     lmv->tgts[tgt]->ltd_exp == NULL ||
3232                     lmv->tgts[tgt]->ltd_active == 0)
3233                         continue;
3234
3235                 rc = md_lock_match(lmv->tgts[tgt]->ltd_exp, flags, fid,
3236                                    type, policy, mode, lockh);
3237                 if (rc)
3238                         RETURN(rc);
3239         }
3240
3241         RETURN(0);
3242 }
3243
3244 int lmv_get_lustre_md(struct obd_export *exp, struct ptlrpc_request *req,
3245                       struct obd_export *dt_exp, struct obd_export *md_exp,
3246                       struct lustre_md *md)
3247 {
3248         struct lmv_obd          *lmv = &exp->exp_obd->u.lmv;
3249         struct lmv_tgt_desc     *tgt = lmv->tgts[0];
3250
3251         if (tgt == NULL || tgt->ltd_exp == NULL)
3252                 RETURN(-EINVAL);
3253
3254         return md_get_lustre_md(lmv->tgts[0]->ltd_exp, req, dt_exp, md_exp, md);
3255 }
3256
3257 int lmv_free_lustre_md(struct obd_export *exp, struct lustre_md *md)
3258 {
3259         struct obd_device       *obd = exp->exp_obd;
3260         struct lmv_obd          *lmv = &obd->u.lmv;
3261         struct lmv_tgt_desc     *tgt = lmv->tgts[0];
3262         ENTRY;
3263
3264         if (md->lmv != NULL) {
3265                 lmv_free_memmd(md->lmv);
3266                 md->lmv = NULL;
3267         }
3268         if (tgt == NULL || tgt->ltd_exp == NULL)
3269                 RETURN(-EINVAL);
3270         RETURN(md_free_lustre_md(lmv->tgts[0]->ltd_exp, md));
3271 }
3272
3273 int lmv_set_open_replay_data(struct obd_export *exp,
3274                              struct obd_client_handle *och,
3275                              struct lookup_intent *it)
3276 {
3277         struct obd_device       *obd = exp->exp_obd;
3278         struct lmv_obd          *lmv = &obd->u.lmv;
3279         struct lmv_tgt_desc     *tgt;
3280         ENTRY;
3281
3282         tgt = lmv_find_target(lmv, &och->och_fid);
3283         if (IS_ERR(tgt))
3284                 RETURN(PTR_ERR(tgt));
3285
3286         RETURN(md_set_open_replay_data(tgt->ltd_exp, och, it));
3287 }
3288
3289 int lmv_clear_open_replay_data(struct obd_export *exp,
3290                                struct obd_client_handle *och)
3291 {
3292         struct obd_device       *obd = exp->exp_obd;
3293         struct lmv_obd          *lmv = &obd->u.lmv;
3294         struct lmv_tgt_desc     *tgt;
3295         ENTRY;
3296
3297         tgt = lmv_find_target(lmv, &och->och_fid);
3298         if (IS_ERR(tgt))
3299                 RETURN(PTR_ERR(tgt));
3300
3301         RETURN(md_clear_open_replay_data(tgt->ltd_exp, och));
3302 }
3303
3304 static int lmv_get_remote_perm(struct obd_export *exp,
3305                                const struct lu_fid *fid,
3306                                struct obd_capa *oc, __u32 suppgid,
3307                                struct ptlrpc_request **request)
3308 {
3309         struct obd_device       *obd = exp->exp_obd;
3310         struct lmv_obd          *lmv = &obd->u.lmv;
3311         struct lmv_tgt_desc     *tgt;
3312         int                      rc;
3313         ENTRY;
3314
3315         rc = lmv_check_connect(obd);
3316         if (rc)
3317                 RETURN(rc);
3318
3319         tgt = lmv_find_target(lmv, fid);
3320         if (IS_ERR(tgt))
3321                 RETURN(PTR_ERR(tgt));
3322
3323         rc = md_get_remote_perm(tgt->ltd_exp, fid, oc, suppgid, request);
3324         RETURN(rc);
3325 }
3326
3327 static int lmv_renew_capa(struct obd_export *exp, struct obd_capa *oc,
3328                           renew_capa_cb_t cb)
3329 {
3330         struct obd_device       *obd = exp->exp_obd;
3331         struct lmv_obd          *lmv = &obd->u.lmv;
3332         struct lmv_tgt_desc     *tgt;
3333         int                      rc;
3334         ENTRY;
3335
3336         rc = lmv_check_connect(obd);
3337         if (rc)
3338                 RETURN(rc);
3339
3340         tgt = lmv_find_target(lmv, &oc->c_capa.lc_fid);
3341         if (IS_ERR(tgt))
3342                 RETURN(PTR_ERR(tgt));
3343
3344         rc = md_renew_capa(tgt->ltd_exp, oc, cb);
3345         RETURN(rc);
3346 }
3347
3348 int lmv_unpack_capa(struct obd_export *exp, struct ptlrpc_request *req,
3349                     const struct req_msg_field *field, struct obd_capa **oc)
3350 {
3351         struct lmv_obd          *lmv = &exp->exp_obd->u.lmv;
3352         struct lmv_tgt_desc     *tgt = lmv->tgts[0];
3353
3354         if (tgt == NULL || tgt->ltd_exp == NULL)
3355                 RETURN(-EINVAL);
3356         return md_unpack_capa(tgt->ltd_exp, req, field, oc);
3357 }
3358
3359 int lmv_intent_getattr_async(struct obd_export *exp,
3360                              struct md_enqueue_info *minfo,
3361                              struct ldlm_enqueue_info *einfo)
3362 {
3363         struct md_op_data       *op_data = &minfo->mi_data;
3364         struct obd_device       *obd = exp->exp_obd;
3365         struct lmv_obd          *lmv = &obd->u.lmv;
3366         struct lmv_tgt_desc     *tgt = NULL;
3367         int                      rc;
3368         ENTRY;
3369
3370         rc = lmv_check_connect(obd);
3371         if (rc)
3372                 RETURN(rc);
3373
3374         tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
3375         if (IS_ERR(tgt))
3376                 RETURN(PTR_ERR(tgt));
3377
3378         rc = md_intent_getattr_async(tgt->ltd_exp, minfo, einfo);
3379         RETURN(rc);
3380 }
3381
3382 int lmv_revalidate_lock(struct obd_export *exp, struct lookup_intent *it,
3383                         struct lu_fid *fid, __u64 *bits)
3384 {
3385         struct obd_device       *obd = exp->exp_obd;
3386         struct lmv_obd          *lmv = &obd->u.lmv;
3387         struct lmv_tgt_desc     *tgt;
3388         int                      rc;
3389         ENTRY;
3390
3391         rc = lmv_check_connect(obd);
3392         if (rc)
3393                 RETURN(rc);
3394
3395         tgt = lmv_find_target(lmv, fid);
3396         if (IS_ERR(tgt))
3397                 RETURN(PTR_ERR(tgt));
3398
3399         rc = md_revalidate_lock(tgt->ltd_exp, it, fid, bits);
3400         RETURN(rc);
3401 }
3402
3403 int lmv_get_fid_from_lsm(struct obd_export *exp,
3404                          const struct lmv_stripe_md *lsm,
3405                          const char *name, int namelen, struct lu_fid *fid)
3406 {
3407         const struct lmv_oinfo *oinfo;
3408
3409         LASSERT(lsm != NULL);
3410         oinfo = lsm_name_to_stripe_info(lsm, name, namelen);
3411         if (IS_ERR(oinfo))
3412                 return PTR_ERR(oinfo);
3413
3414         *fid = oinfo->lmo_fid;
3415
3416         RETURN(0);
3417 }
3418
3419 /**
3420  * For lmv, only need to send request to master MDT, and the master MDT will
3421  * process with other slave MDTs. The only exception is Q_GETOQUOTA for which
3422  * we directly fetch data from the slave MDTs.
3423  */
3424 int lmv_quotactl(struct obd_device *unused, struct obd_export *exp,
3425                  struct obd_quotactl *oqctl)
3426 {
3427         struct obd_device   *obd = class_exp2obd(exp);
3428         struct lmv_obd      *lmv = &obd->u.lmv;
3429         struct lmv_tgt_desc *tgt = lmv->tgts[0];
3430         int                  rc = 0;
3431         __u32                i;
3432         __u64                curspace, curinodes;
3433         ENTRY;
3434
3435         if (tgt == NULL ||
3436             tgt->ltd_exp == NULL ||
3437             !tgt->ltd_active ||
3438             lmv->desc.ld_tgt_count == 0) {
3439                 CERROR("master lmv inactive\n");
3440                 RETURN(-EIO);
3441         }
3442
3443         if (oqctl->qc_cmd != Q_GETOQUOTA) {
3444                 rc = obd_quotactl(tgt->ltd_exp, oqctl);
3445                 RETURN(rc);
3446         }
3447
3448         curspace = curinodes = 0;
3449         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
3450                 int err;
3451                 tgt = lmv->tgts[i];
3452
3453                 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active)
3454                         continue;
3455
3456                 err = obd_quotactl(tgt->ltd_exp, oqctl);
3457                 if (err) {
3458                         CERROR("getquota on mdt %d failed. %d\n", i, err);
3459                         if (!rc)
3460                                 rc = err;
3461                 } else {
3462                         curspace += oqctl->qc_dqblk.dqb_curspace;
3463                         curinodes += oqctl->qc_dqblk.dqb_curinodes;
3464                 }
3465         }
3466         oqctl->qc_dqblk.dqb_curspace = curspace;
3467         oqctl->qc_dqblk.dqb_curinodes = curinodes;
3468
3469         RETURN(rc);
3470 }
3471
3472 int lmv_quotacheck(struct obd_device *unused, struct obd_export *exp,
3473                    struct obd_quotactl *oqctl)
3474 {
3475         struct obd_device       *obd = class_exp2obd(exp);
3476         struct lmv_obd          *lmv = &obd->u.lmv;
3477         struct lmv_tgt_desc     *tgt;
3478         __u32                    i;
3479         int                      rc = 0;
3480         ENTRY;
3481
3482         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
3483                 int err;
3484                 tgt = lmv->tgts[i];
3485                 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active) {
3486                         CERROR("lmv idx %d inactive\n", i);
3487                         RETURN(-EIO);
3488                 }
3489
3490                 err = obd_quotacheck(tgt->ltd_exp, oqctl);
3491                 if (err && !rc)
3492                         rc = err;
3493         }
3494
3495         RETURN(rc);
3496 }
3497
3498 static int lmv_merge_attr(struct obd_export *exp,
3499                           const struct lmv_stripe_md *lsm,
3500                           struct cl_attr *attr,
3501                           ldlm_blocking_callback cb_blocking)
3502 {
3503         int rc;
3504         int i;
3505
3506         rc = lmv_revalidate_slaves(exp, lsm, cb_blocking, 0);
3507         if (rc < 0)
3508                 return rc;
3509
3510         for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
3511                 struct inode *inode = lsm->lsm_md_oinfo[i].lmo_root;
3512
3513                 CDEBUG(D_INFO, ""DFID" size %llu, blocks %llu nlink %u,"
3514                        " atime %lu ctime %lu, mtime %lu.\n",
3515                        PFID(&lsm->lsm_md_oinfo[i].lmo_fid),
3516                        i_size_read(inode), (unsigned long long)inode->i_blocks,
3517                        inode->i_nlink, LTIME_S(inode->i_atime),
3518                        LTIME_S(inode->i_ctime), LTIME_S(inode->i_mtime));
3519
3520                 /* for slave stripe, it needs to subtract nlink for . and .. */
3521                 if (i != 0)
3522                         attr->cat_nlink += inode->i_nlink - 2;
3523                 else
3524                         attr->cat_nlink = inode->i_nlink;
3525
3526                 attr->cat_size += i_size_read(inode);
3527                 attr->cat_blocks += inode->i_blocks;
3528
3529                 if (attr->cat_atime < LTIME_S(inode->i_atime))
3530                         attr->cat_atime = LTIME_S(inode->i_atime);
3531
3532                 if (attr->cat_ctime < LTIME_S(inode->i_ctime))
3533                         attr->cat_ctime = LTIME_S(inode->i_ctime);
3534
3535                 if (attr->cat_mtime < LTIME_S(inode->i_mtime))
3536                         attr->cat_mtime = LTIME_S(inode->i_mtime);
3537         }
3538         return 0;
3539 }
3540
3541 struct obd_ops lmv_obd_ops = {
3542         .o_owner                = THIS_MODULE,
3543         .o_setup                = lmv_setup,
3544         .o_cleanup              = lmv_cleanup,
3545         .o_precleanup           = lmv_precleanup,
3546         .o_process_config       = lmv_process_config,
3547         .o_connect              = lmv_connect,
3548         .o_disconnect           = lmv_disconnect,
3549         .o_statfs               = lmv_statfs,
3550         .o_get_info             = lmv_get_info,
3551         .o_set_info_async       = lmv_set_info_async,
3552         .o_packmd               = lmv_packmd,
3553         .o_unpackmd             = lmv_unpackmd,
3554         .o_notify               = lmv_notify,
3555         .o_get_uuid             = lmv_get_uuid,
3556         .o_iocontrol            = lmv_iocontrol,
3557         .o_quotacheck           = lmv_quotacheck,
3558         .o_quotactl             = lmv_quotactl
3559 };
3560
3561 struct md_ops lmv_md_ops = {
3562         .m_getstatus            = lmv_getstatus,
3563         .m_null_inode           = lmv_null_inode,
3564         .m_find_cbdata          = lmv_find_cbdata,
3565         .m_close                = lmv_close,
3566         .m_create               = lmv_create,
3567         .m_done_writing         = lmv_done_writing,
3568         .m_enqueue              = lmv_enqueue,
3569         .m_getattr              = lmv_getattr,
3570         .m_getxattr             = lmv_getxattr,
3571         .m_getattr_name         = lmv_getattr_name,
3572         .m_intent_lock          = lmv_intent_lock,
3573         .m_link                 = lmv_link,
3574         .m_rename               = lmv_rename,
3575         .m_setattr              = lmv_setattr,
3576         .m_setxattr             = lmv_setxattr,
3577         .m_fsync                = lmv_fsync,
3578         .m_read_page            = lmv_read_page,
3579         .m_unlink               = lmv_unlink,
3580         .m_init_ea_size         = lmv_init_ea_size,
3581         .m_cancel_unused        = lmv_cancel_unused,
3582         .m_set_lock_data        = lmv_set_lock_data,
3583         .m_lock_match           = lmv_lock_match,
3584         .m_get_lustre_md        = lmv_get_lustre_md,
3585         .m_free_lustre_md       = lmv_free_lustre_md,
3586         .m_merge_attr           = lmv_merge_attr,
3587         .m_set_open_replay_data = lmv_set_open_replay_data,
3588         .m_clear_open_replay_data = lmv_clear_open_replay_data,
3589         .m_renew_capa           = lmv_renew_capa,
3590         .m_unpack_capa          = lmv_unpack_capa,
3591         .m_get_remote_perm      = lmv_get_remote_perm,
3592         .m_intent_getattr_async = lmv_intent_getattr_async,
3593         .m_revalidate_lock      = lmv_revalidate_lock,
3594         .m_get_fid_from_lsm     = lmv_get_fid_from_lsm,
3595 };
3596
3597 int __init lmv_init(void)
3598 {
3599         return class_register_type(&lmv_obd_ops, &lmv_md_ops, true, NULL,
3600                                    LUSTRE_LMV_NAME, NULL);
3601 }
3602
3603 static void lmv_exit(void)
3604 {
3605         class_unregister_type(LUSTRE_LMV_NAME);
3606 }
3607
3608 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
3609 MODULE_DESCRIPTION("Lustre Logical Metadata Volume OBD driver");
3610 MODULE_LICENSE("GPL");
3611
3612 module_init(lmv_init);
3613 module_exit(lmv_exit);