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         /**
1254          * If stripe_offset is provided during setdirstripe
1255          * (setdirstripe -i xx), xx MDS will be choosen.
1256          */
1257         if (op_data->op_cli_flags & CLI_SET_MEA && op_data->op_data != NULL) {
1258                 struct lmv_user_md *lum;
1259
1260                 lum = op_data->op_data;
1261
1262                 if (le32_to_cpu(lum->lum_stripe_offset) != (__u32)-1) {
1263                         *mds = le32_to_cpu(lum->lum_stripe_offset);
1264                 } else {
1265                         /* -1 means default, which will be in the same MDT with
1266                          * the stripe */
1267                         *mds = op_data->op_mds;
1268                         lum->lum_stripe_offset = cpu_to_le32(op_data->op_mds);
1269                 }
1270         } else {
1271                 /* Allocate new fid on target according to operation type and
1272                  * parent home mds. */
1273                 *mds = op_data->op_mds;
1274         }
1275
1276         RETURN(0);
1277 }
1278
1279 int __lmv_fid_alloc(struct lmv_obd *lmv, struct lu_fid *fid, u32 mds)
1280 {
1281         struct lmv_tgt_desc     *tgt;
1282         int                      rc;
1283         ENTRY;
1284
1285         tgt = lmv_get_target(lmv, mds, NULL);
1286         if (IS_ERR(tgt))
1287                 RETURN(PTR_ERR(tgt));
1288
1289         /*
1290          * New seq alloc and FLD setup should be atomic. Otherwise we may find
1291          * on server that seq in new allocated fid is not yet known.
1292          */
1293         mutex_lock(&tgt->ltd_fid_mutex);
1294
1295         if (tgt->ltd_active == 0 || tgt->ltd_exp == NULL)
1296                 GOTO(out, rc = -ENODEV);
1297
1298         /*
1299          * Asking underlying tgt layer to allocate new fid.
1300          */
1301         rc = obd_fid_alloc(NULL, tgt->ltd_exp, fid, NULL);
1302         if (rc > 0) {
1303                 LASSERT(fid_is_sane(fid));
1304                 rc = 0;
1305         }
1306
1307         EXIT;
1308 out:
1309         mutex_unlock(&tgt->ltd_fid_mutex);
1310         return rc;
1311 }
1312
1313 int lmv_fid_alloc(const struct lu_env *env, struct obd_export *exp,
1314                   struct lu_fid *fid, struct md_op_data *op_data)
1315 {
1316         struct obd_device     *obd = class_exp2obd(exp);
1317         struct lmv_obd        *lmv = &obd->u.lmv;
1318         u32                    mds = 0;
1319         int                    rc;
1320         ENTRY;
1321
1322         LASSERT(op_data != NULL);
1323         LASSERT(fid != NULL);
1324
1325         rc = lmv_placement_policy(obd, op_data, &mds);
1326         if (rc) {
1327                 CERROR("Can't get target for allocating fid, "
1328                        "rc %d\n", rc);
1329                 RETURN(rc);
1330         }
1331
1332         rc = __lmv_fid_alloc(lmv, fid, mds);
1333         if (rc) {
1334                 CERROR("Can't alloc new fid, rc %d\n", rc);
1335                 RETURN(rc);
1336         }
1337
1338         RETURN(rc);
1339 }
1340
1341 static int lmv_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
1342 {
1343         struct lmv_obd  *lmv = &obd->u.lmv;
1344         struct lmv_desc *desc;
1345         int             rc;
1346         ENTRY;
1347
1348         if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
1349                 CERROR("LMV setup requires a descriptor\n");
1350                 RETURN(-EINVAL);
1351         }
1352
1353         desc = (struct lmv_desc *)lustre_cfg_buf(lcfg, 1);
1354         if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
1355                 CERROR("Lmv descriptor size wrong: %d > %d\n",
1356                        (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
1357                 RETURN(-EINVAL);
1358         }
1359
1360         lmv->tgts_size = 32U;
1361         OBD_ALLOC(lmv->tgts, sizeof(*lmv->tgts) * lmv->tgts_size);
1362         if (lmv->tgts == NULL)
1363                 RETURN(-ENOMEM);
1364
1365         obd_str2uuid(&lmv->desc.ld_uuid, desc->ld_uuid.uuid);
1366         lmv->desc.ld_tgt_count = 0;
1367         lmv->desc.ld_active_tgt_count = 0;
1368         lmv->max_cookiesize = 0;
1369         lmv->max_def_easize = 0;
1370         lmv->max_easize = 0;
1371         lmv->lmv_placement = PLACEMENT_CHAR_POLICY;
1372
1373         spin_lock_init(&lmv->lmv_lock);
1374         mutex_init(&lmv->lmv_init_mutex);
1375
1376 #ifdef CONFIG_PROC_FS
1377         obd->obd_vars = lprocfs_lmv_obd_vars;
1378         lprocfs_obd_setup(obd);
1379         lprocfs_alloc_md_stats(obd, 0);
1380         rc = lprocfs_seq_create(obd->obd_proc_entry, "target_obd",
1381                                 0444, &lmv_proc_target_fops, obd);
1382         if (rc)
1383                 CWARN("%s: error adding LMV target_obd file: rc = %d\n",
1384                       obd->obd_name, rc);
1385 #endif
1386         rc = fld_client_init(&lmv->lmv_fld, obd->obd_name,
1387                              LUSTRE_CLI_FLD_HASH_DHT);
1388         if (rc) {
1389                 CERROR("Can't init FLD, err %d\n", rc);
1390                 GOTO(out, rc);
1391         }
1392
1393         RETURN(0);
1394
1395 out:
1396         return rc;
1397 }
1398
1399 static int lmv_cleanup(struct obd_device *obd)
1400 {
1401         struct lmv_obd   *lmv = &obd->u.lmv;
1402         ENTRY;
1403
1404         fld_client_fini(&lmv->lmv_fld);
1405         if (lmv->tgts != NULL) {
1406                 int i;
1407                 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1408                         if (lmv->tgts[i] == NULL)
1409                                 continue;
1410                         lmv_del_target(lmv, i);
1411                 }
1412                 OBD_FREE(lmv->tgts, sizeof(*lmv->tgts) * lmv->tgts_size);
1413                 lmv->tgts_size = 0;
1414         }
1415         RETURN(0);
1416 }
1417
1418 static int lmv_process_config(struct obd_device *obd, size_t len, void *buf)
1419 {
1420         struct lustre_cfg       *lcfg = buf;
1421         struct obd_uuid         obd_uuid;
1422         int                     gen;
1423         __u32                   index;
1424         int                     rc;
1425         ENTRY;
1426
1427         switch (lcfg->lcfg_command) {
1428         case LCFG_ADD_MDC:
1429                 /* modify_mdc_tgts add 0:lustre-clilmv  1:lustre-MDT0000_UUID
1430                  * 2:0  3:1  4:lustre-MDT0000-mdc_UUID */
1431                 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(obd_uuid.uuid))
1432                         GOTO(out, rc = -EINVAL);
1433
1434                 obd_str2uuid(&obd_uuid,  lustre_cfg_buf(lcfg, 1));
1435
1436                 if (sscanf(lustre_cfg_buf(lcfg, 2), "%u", &index) != 1)
1437                         GOTO(out, rc = -EINVAL);
1438                 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", &gen) != 1)
1439                         GOTO(out, rc = -EINVAL);
1440                 rc = lmv_add_target(obd, &obd_uuid, index, gen);
1441                 GOTO(out, rc);
1442         default:
1443                 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
1444                 GOTO(out, rc = -EINVAL);
1445         }
1446 out:
1447         RETURN(rc);
1448 }
1449
1450 static int lmv_statfs(const struct lu_env *env, struct obd_export *exp,
1451                       struct obd_statfs *osfs, __u64 max_age, __u32 flags)
1452 {
1453         struct obd_device       *obd = class_exp2obd(exp);
1454         struct lmv_obd          *lmv = &obd->u.lmv;
1455         struct obd_statfs       *temp;
1456         int                      rc = 0;
1457         __u32                    i;
1458         ENTRY;
1459
1460         rc = lmv_check_connect(obd);
1461         if (rc)
1462                 RETURN(rc);
1463
1464         OBD_ALLOC(temp, sizeof(*temp));
1465         if (temp == NULL)
1466                 RETURN(-ENOMEM);
1467
1468         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1469                 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
1470                         continue;
1471
1472                 rc = obd_statfs(env, lmv->tgts[i]->ltd_exp, temp,
1473                                 max_age, flags);
1474                 if (rc) {
1475                         CERROR("can't stat MDS #%d (%s), error %d\n", i,
1476                                lmv->tgts[i]->ltd_exp->exp_obd->obd_name,
1477                                rc);
1478                         GOTO(out_free_temp, rc);
1479                 }
1480
1481                 if (i == 0) {
1482                         *osfs = *temp;
1483                         /* If the statfs is from mount, it will needs
1484                          * retrieve necessary information from MDT0.
1485                          * i.e. mount does not need the merged osfs
1486                          * from all of MDT.
1487                          * And also clients can be mounted as long as
1488                          * MDT0 is in service*/
1489                         if (flags & OBD_STATFS_FOR_MDT0)
1490                                 GOTO(out_free_temp, rc);
1491                 } else {
1492                         osfs->os_bavail += temp->os_bavail;
1493                         osfs->os_blocks += temp->os_blocks;
1494                         osfs->os_ffree += temp->os_ffree;
1495                         osfs->os_files += temp->os_files;
1496                 }
1497         }
1498
1499         EXIT;
1500 out_free_temp:
1501         OBD_FREE(temp, sizeof(*temp));
1502         return rc;
1503 }
1504
1505 static int lmv_getstatus(struct obd_export *exp,
1506                          struct lu_fid *fid,
1507                          struct obd_capa **pc)
1508 {
1509         struct obd_device    *obd = exp->exp_obd;
1510         struct lmv_obd       *lmv = &obd->u.lmv;
1511         int                   rc;
1512         ENTRY;
1513
1514         rc = lmv_check_connect(obd);
1515         if (rc)
1516                 RETURN(rc);
1517
1518         rc = md_getstatus(lmv->tgts[0]->ltd_exp, fid, pc);
1519         RETURN(rc);
1520 }
1521
1522 static int lmv_getxattr(struct obd_export *exp, const struct lu_fid *fid,
1523                         struct obd_capa *oc, u64 valid, const char *name,
1524                         const char *input, int input_size, int output_size,
1525                         int flags, struct ptlrpc_request **request)
1526 {
1527         struct obd_device      *obd = exp->exp_obd;
1528         struct lmv_obd         *lmv = &obd->u.lmv;
1529         struct lmv_tgt_desc    *tgt;
1530         int                     rc;
1531         ENTRY;
1532
1533         rc = lmv_check_connect(obd);
1534         if (rc)
1535                 RETURN(rc);
1536
1537         tgt = lmv_find_target(lmv, fid);
1538         if (IS_ERR(tgt))
1539                 RETURN(PTR_ERR(tgt));
1540
1541         rc = md_getxattr(tgt->ltd_exp, fid, oc, valid, name, input,
1542                          input_size, output_size, flags, request);
1543
1544         RETURN(rc);
1545 }
1546
1547 static int lmv_setxattr(struct obd_export *exp, const struct lu_fid *fid,
1548                         struct obd_capa *oc, u64 valid, const char *name,
1549                         const char *input, int input_size, int output_size,
1550                         int flags, __u32 suppgid,
1551                         struct ptlrpc_request **request)
1552 {
1553         struct obd_device      *obd = exp->exp_obd;
1554         struct lmv_obd         *lmv = &obd->u.lmv;
1555         struct lmv_tgt_desc    *tgt;
1556         int                     rc;
1557         ENTRY;
1558
1559         rc = lmv_check_connect(obd);
1560         if (rc)
1561                 RETURN(rc);
1562
1563         tgt = lmv_find_target(lmv, fid);
1564         if (IS_ERR(tgt))
1565                 RETURN(PTR_ERR(tgt));
1566
1567         rc = md_setxattr(tgt->ltd_exp, fid, oc, valid, name, input,
1568                          input_size, output_size, flags, suppgid,
1569                          request);
1570
1571         RETURN(rc);
1572 }
1573
1574 static int lmv_getattr(struct obd_export *exp, struct md_op_data *op_data,
1575                        struct ptlrpc_request **request)
1576 {
1577         struct obd_device       *obd = exp->exp_obd;
1578         struct lmv_obd          *lmv = &obd->u.lmv;
1579         struct lmv_tgt_desc     *tgt;
1580         int                      rc;
1581         ENTRY;
1582
1583         rc = lmv_check_connect(obd);
1584         if (rc)
1585                 RETURN(rc);
1586
1587         tgt = lmv_find_target(lmv, &op_data->op_fid1);
1588         if (IS_ERR(tgt))
1589                 RETURN(PTR_ERR(tgt));
1590
1591         if (op_data->op_flags & MF_GET_MDT_IDX) {
1592                 op_data->op_mds = tgt->ltd_idx;
1593                 RETURN(0);
1594         }
1595
1596         rc = md_getattr(tgt->ltd_exp, op_data, request);
1597
1598         RETURN(rc);
1599 }
1600
1601 static int lmv_null_inode(struct obd_export *exp, const struct lu_fid *fid)
1602 {
1603         struct obd_device   *obd = exp->exp_obd;
1604         struct lmv_obd      *lmv = &obd->u.lmv;
1605         __u32                i;
1606         int                  rc;
1607         ENTRY;
1608
1609         rc = lmv_check_connect(obd);
1610         if (rc)
1611                 RETURN(rc);
1612
1613         CDEBUG(D_INODE, "CBDATA for "DFID"\n", PFID(fid));
1614
1615         /*
1616          * With DNE every object can have two locks in different namespaces:
1617          * lookup lock in space of MDT storing direntry and update/open lock in
1618          * space of MDT storing inode.
1619          */
1620         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1621                 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
1622                         continue;
1623                 md_null_inode(lmv->tgts[i]->ltd_exp, fid);
1624         }
1625
1626         RETURN(0);
1627 }
1628
1629 static int lmv_find_cbdata(struct obd_export *exp, const struct lu_fid *fid,
1630                            ldlm_iterator_t it, void *data)
1631 {
1632         struct obd_device       *obd = exp->exp_obd;
1633         struct lmv_obd          *lmv = &obd->u.lmv;
1634         int                     i;
1635         int                     tgt;
1636         int                     rc;
1637         ENTRY;
1638
1639         rc = lmv_check_connect(obd);
1640         if (rc)
1641                 RETURN(rc);
1642
1643         CDEBUG(D_INODE, "CBDATA for "DFID"\n", PFID(fid));
1644
1645         /*
1646          * With DNE every object can have two locks in different namespaces:
1647          * lookup lock in space of MDT storing direntry and update/open lock in
1648          * space of MDT storing inode.  Try the MDT that the FID maps to first,
1649          * since this can be easily found, and only try others if that fails.
1650          */
1651         for (i = 0, tgt = lmv_find_target_index(lmv, fid);
1652              i < lmv->desc.ld_tgt_count;
1653              i++, tgt = (tgt + 1) % lmv->desc.ld_tgt_count) {
1654                 if (tgt < 0) {
1655                         CDEBUG(D_HA, "%s: "DFID" is inaccessible: rc = %d\n",
1656                                obd->obd_name, PFID(fid), tgt);
1657                         tgt = 0;
1658                 }
1659
1660                 if (lmv->tgts[tgt] == NULL ||
1661                     lmv->tgts[tgt]->ltd_exp == NULL)
1662                         continue;
1663
1664                 rc = md_find_cbdata(lmv->tgts[tgt]->ltd_exp, fid, it, data);
1665                 if (rc)
1666                         RETURN(rc);
1667         }
1668
1669         RETURN(rc);
1670 }
1671
1672
1673 static int lmv_close(struct obd_export *exp, struct md_op_data *op_data,
1674                      struct md_open_data *mod, struct ptlrpc_request **request)
1675 {
1676         struct obd_device     *obd = exp->exp_obd;
1677         struct lmv_obd        *lmv = &obd->u.lmv;
1678         struct lmv_tgt_desc   *tgt;
1679         int                    rc;
1680         ENTRY;
1681
1682         rc = lmv_check_connect(obd);
1683         if (rc)
1684                 RETURN(rc);
1685
1686         tgt = lmv_find_target(lmv, &op_data->op_fid1);
1687         if (IS_ERR(tgt))
1688                 RETURN(PTR_ERR(tgt));
1689
1690         CDEBUG(D_INODE, "CLOSE "DFID"\n", PFID(&op_data->op_fid1));
1691         rc = md_close(tgt->ltd_exp, op_data, mod, request);
1692         RETURN(rc);
1693 }
1694
1695 /**
1696  * Choosing the MDT by name or FID in @op_data.
1697  * For non-striped directory, it will locate MDT by fid.
1698  * For striped-directory, it will locate MDT by name. And also
1699  * it will reset op_fid1 with the FID of the choosen stripe.
1700  **/
1701 struct lmv_tgt_desc *
1702 lmv_locate_target_for_name(struct lmv_obd *lmv, struct lmv_stripe_md *lsm,
1703                            const char *name, int namelen, struct lu_fid *fid,
1704                            u32 *mds)
1705 {
1706         struct lmv_tgt_desc     *tgt;
1707         const struct lmv_oinfo  *oinfo;
1708
1709         if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_NAME_HASH)) {
1710                 oinfo = &lsm->lsm_md_oinfo[cfs_fail_val];
1711         } else {
1712                 oinfo = lsm_name_to_stripe_info(lsm, name, namelen);
1713                 if (IS_ERR(oinfo))
1714                         RETURN(ERR_CAST(oinfo));
1715         }
1716
1717         *fid = oinfo->lmo_fid;
1718         *mds = oinfo->lmo_mds;
1719         tgt = lmv_get_target(lmv, *mds, NULL);
1720
1721         CDEBUG(D_INFO, "locate on mds %u "DFID"\n", *mds, PFID(fid));
1722         return tgt;
1723 }
1724
1725 /**
1726  * Locate mds by fid or name
1727  *
1728  * For striped directory (lsm != NULL), it will locate the stripe
1729  * by name hash (see lsm_name_to_stripe_info()). Note: if the hash_type
1730  * is unknown, it will return -EBADFD, and lmv_intent_lookup might need
1731  * walk through all of stripes to locate the entry.
1732  *
1733  * For normal direcotry, it will locate MDS by FID directly.
1734  * \param[in] lmv       LMV device
1735  * \param[in] op_data   client MD stack parameters, name, namelen
1736  *                      mds_num etc.
1737  * \param[in] fid       object FID used to locate MDS.
1738  *
1739  * retval               pointer to the lmv_tgt_desc if succeed.
1740  *                      ERR_PTR(errno) if failed.
1741  */
1742 struct lmv_tgt_desc*
1743 lmv_locate_mds(struct lmv_obd *lmv, struct md_op_data *op_data,
1744                struct lu_fid *fid)
1745 {
1746         struct lmv_stripe_md    *lsm = op_data->op_mea1;
1747         struct lmv_tgt_desc     *tgt;
1748
1749         /* During creating VOLATILE file, it should honor the mdt
1750          * index if the file under striped dir is being restored, see
1751          * ct_restore(). */
1752         if (op_data->op_bias & MDS_CREATE_VOLATILE &&
1753             (int)op_data->op_mds != -1 && lsm != NULL) {
1754                 int i;
1755                 tgt = lmv_get_target(lmv, op_data->op_mds, NULL);
1756                 if (IS_ERR(tgt))
1757                         return tgt;
1758
1759                 /* refill the right parent fid */
1760                 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1761                         struct lmv_oinfo *oinfo;
1762
1763                         oinfo = &lsm->lsm_md_oinfo[i];
1764                         if (oinfo->lmo_mds == op_data->op_mds) {
1765                                 *fid = oinfo->lmo_fid;
1766                                 break;
1767                         }
1768                 }
1769
1770                 /* Hmm, can not find the stripe by mdt_index(op_mds) */
1771                 if (i == lsm->lsm_md_stripe_count)
1772                         tgt = ERR_PTR(-EINVAL);
1773
1774                 return tgt;
1775         }
1776
1777         if (lsm == NULL || op_data->op_namelen == 0) {
1778                 tgt = lmv_find_target(lmv, fid);
1779                 if (IS_ERR(tgt))
1780                         return tgt;
1781
1782                 op_data->op_mds = tgt->ltd_idx;
1783                 return tgt;
1784         }
1785
1786         return lmv_locate_target_for_name(lmv, lsm, op_data->op_name,
1787                                           op_data->op_namelen, fid,
1788                                           &op_data->op_mds);
1789 }
1790
1791 int lmv_create(struct obd_export *exp, struct md_op_data *op_data,
1792                 const void *data, size_t datalen, umode_t mode, uid_t uid,
1793                 gid_t gid, cfs_cap_t cap_effective, __u64 rdev,
1794                 struct ptlrpc_request **request)
1795 {
1796         struct obd_device       *obd = exp->exp_obd;
1797         struct lmv_obd          *lmv = &obd->u.lmv;
1798         struct lmv_tgt_desc     *tgt;
1799         int                      rc;
1800         ENTRY;
1801
1802         rc = lmv_check_connect(obd);
1803         if (rc)
1804                 RETURN(rc);
1805
1806         if (!lmv->desc.ld_active_tgt_count)
1807                 RETURN(-EIO);
1808
1809         tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
1810         if (IS_ERR(tgt))
1811                 RETURN(PTR_ERR(tgt));
1812
1813         CDEBUG(D_INODE, "CREATE name '%.*s' on "DFID" -> mds #%x\n",
1814                 (int)op_data->op_namelen, op_data->op_name,
1815                 PFID(&op_data->op_fid1), op_data->op_mds);
1816
1817         rc = lmv_fid_alloc(NULL, exp, &op_data->op_fid2, op_data);
1818         if (rc)
1819                 RETURN(rc);
1820         if (exp_connect_flags(exp) & OBD_CONNECT_DIR_STRIPE) {
1821                 /* Send the create request to the MDT where the object
1822                  * will be located */
1823                 tgt = lmv_find_target(lmv, &op_data->op_fid2);
1824                 if (IS_ERR(tgt))
1825                         RETURN(PTR_ERR(tgt));
1826
1827                 op_data->op_mds = tgt->ltd_idx;
1828         } else {
1829                 CDEBUG(D_CONFIG, "Server doesn't support striped dirs\n");
1830         }
1831
1832         CDEBUG(D_INODE, "CREATE obj "DFID" -> mds #%x\n",
1833                PFID(&op_data->op_fid2), op_data->op_mds);
1834
1835         op_data->op_flags |= MF_MDC_CANCEL_FID1;
1836         rc = md_create(tgt->ltd_exp, op_data, data, datalen, mode, uid, gid,
1837                        cap_effective, rdev, request);
1838         if (rc == 0) {
1839                 if (*request == NULL)
1840                         RETURN(rc);
1841                 CDEBUG(D_INODE, "Created - "DFID"\n", PFID(&op_data->op_fid2));
1842         }
1843         RETURN(rc);
1844 }
1845
1846 static int lmv_done_writing(struct obd_export *exp,
1847                             struct md_op_data *op_data,
1848                             struct md_open_data *mod)
1849 {
1850         struct obd_device     *obd = exp->exp_obd;
1851         struct lmv_obd        *lmv = &obd->u.lmv;
1852         struct lmv_tgt_desc   *tgt;
1853         int                    rc;
1854         ENTRY;
1855
1856         rc = lmv_check_connect(obd);
1857         if (rc)
1858                 RETURN(rc);
1859
1860         tgt = lmv_find_target(lmv, &op_data->op_fid1);
1861         if (IS_ERR(tgt))
1862                 RETURN(PTR_ERR(tgt));
1863
1864         rc = md_done_writing(tgt->ltd_exp, op_data, mod);
1865         RETURN(rc);
1866 }
1867
1868 static int
1869 lmv_enqueue(struct obd_export *exp, struct ldlm_enqueue_info *einfo,
1870             const union ldlm_policy_data *policy,
1871             struct lookup_intent *it, struct md_op_data *op_data,
1872             struct lustre_handle *lockh, __u64 extra_lock_flags)
1873 {
1874         struct obd_device        *obd = exp->exp_obd;
1875         struct lmv_obd           *lmv = &obd->u.lmv;
1876         struct lmv_tgt_desc      *tgt;
1877         int                       rc;
1878         ENTRY;
1879
1880         rc = lmv_check_connect(obd);
1881         if (rc)
1882                 RETURN(rc);
1883
1884         CDEBUG(D_INODE, "ENQUEUE '%s' on "DFID"\n",
1885                LL_IT2STR(it), PFID(&op_data->op_fid1));
1886
1887         tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
1888         if (IS_ERR(tgt))
1889                 RETURN(PTR_ERR(tgt));
1890
1891         CDEBUG(D_INODE, "ENQUEUE '%s' on "DFID" -> mds #%u\n",
1892                LL_IT2STR(it), PFID(&op_data->op_fid1), tgt->ltd_idx);
1893
1894         rc = md_enqueue(tgt->ltd_exp, einfo, policy, it, op_data, lockh,
1895                         extra_lock_flags);
1896
1897         RETURN(rc);
1898 }
1899
1900 static int
1901 lmv_getattr_name(struct obd_export *exp,struct md_op_data *op_data,
1902                  struct ptlrpc_request **preq)
1903 {
1904         struct ptlrpc_request   *req = NULL;
1905         struct obd_device       *obd = exp->exp_obd;
1906         struct lmv_obd          *lmv = &obd->u.lmv;
1907         struct lmv_tgt_desc     *tgt;
1908         struct mdt_body         *body;
1909         int                      rc;
1910         ENTRY;
1911
1912         rc = lmv_check_connect(obd);
1913         if (rc)
1914                 RETURN(rc);
1915
1916         tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
1917         if (IS_ERR(tgt))
1918                 RETURN(PTR_ERR(tgt));
1919
1920         CDEBUG(D_INODE, "GETATTR_NAME for %*s on "DFID" -> mds #%d\n",
1921                 (int)op_data->op_namelen, op_data->op_name,
1922                 PFID(&op_data->op_fid1), tgt->ltd_idx);
1923
1924         rc = md_getattr_name(tgt->ltd_exp, op_data, preq);
1925         if (rc != 0)
1926                 RETURN(rc);
1927
1928         body = req_capsule_server_get(&(*preq)->rq_pill, &RMF_MDT_BODY);
1929         LASSERT(body != NULL);
1930
1931         if (body->mbo_valid & OBD_MD_MDS) {
1932                 struct lu_fid rid = body->mbo_fid1;
1933                 CDEBUG(D_INODE, "Request attrs for "DFID"\n",
1934                        PFID(&rid));
1935
1936                 tgt = lmv_find_target(lmv, &rid);
1937                 if (IS_ERR(tgt)) {
1938                         ptlrpc_req_finished(*preq);
1939                         preq = NULL;
1940                         RETURN(PTR_ERR(tgt));
1941                 }
1942
1943                 op_data->op_fid1 = rid;
1944                 op_data->op_valid |= OBD_MD_FLCROSSREF;
1945                 op_data->op_namelen = 0;
1946                 op_data->op_name = NULL;
1947                 rc = md_getattr_name(tgt->ltd_exp, op_data, &req);
1948                 ptlrpc_req_finished(*preq);
1949                 *preq = req;
1950         }
1951
1952         RETURN(rc);
1953 }
1954
1955 #define md_op_data_fid(op_data, fl)                     \
1956         (fl == MF_MDC_CANCEL_FID1 ? &op_data->op_fid1 : \
1957          fl == MF_MDC_CANCEL_FID2 ? &op_data->op_fid2 : \
1958          fl == MF_MDC_CANCEL_FID3 ? &op_data->op_fid3 : \
1959          fl == MF_MDC_CANCEL_FID4 ? &op_data->op_fid4 : \
1960          NULL)
1961
1962 static int lmv_early_cancel(struct obd_export *exp, struct lmv_tgt_desc *tgt,
1963                             struct md_op_data *op_data,
1964                             __u32 op_tgt, ldlm_mode_t mode, int bits, int flag)
1965 {
1966         struct lu_fid          *fid = md_op_data_fid(op_data, flag);
1967         struct obd_device      *obd = exp->exp_obd;
1968         struct lmv_obd         *lmv = &obd->u.lmv;
1969         ldlm_policy_data_t      policy = {{ 0 }};
1970         int                     rc = 0;
1971         ENTRY;
1972
1973         if (!fid_is_sane(fid))
1974                 RETURN(0);
1975
1976         if (tgt == NULL) {
1977                 tgt = lmv_find_target(lmv, fid);
1978                 if (IS_ERR(tgt))
1979                         RETURN(PTR_ERR(tgt));
1980         }
1981
1982         if (tgt->ltd_idx != op_tgt) {
1983                 CDEBUG(D_INODE, "EARLY_CANCEL on "DFID"\n", PFID(fid));
1984                 policy.l_inodebits.bits = bits;
1985                 rc = md_cancel_unused(tgt->ltd_exp, fid, &policy,
1986                                       mode, LCF_ASYNC, NULL);
1987         } else {
1988                 CDEBUG(D_INODE,
1989                        "EARLY_CANCEL skip operation target %d on "DFID"\n",
1990                        op_tgt, PFID(fid));
1991                 op_data->op_flags |= flag;
1992                 rc = 0;
1993         }
1994
1995         RETURN(rc);
1996 }
1997
1998 /*
1999  * llite passes fid of an target inode in op_data->op_fid1 and id of directory in
2000  * op_data->op_fid2
2001  */
2002 static int lmv_link(struct obd_export *exp, struct md_op_data *op_data,
2003                     struct ptlrpc_request **request)
2004 {
2005         struct obd_device       *obd = exp->exp_obd;
2006         struct lmv_obd          *lmv = &obd->u.lmv;
2007         struct lmv_tgt_desc     *tgt;
2008         int                      rc;
2009         ENTRY;
2010
2011         rc = lmv_check_connect(obd);
2012         if (rc)
2013                 RETURN(rc);
2014
2015         LASSERT(op_data->op_namelen != 0);
2016
2017         CDEBUG(D_INODE, "LINK "DFID":%*s to "DFID"\n",
2018                PFID(&op_data->op_fid2), (int)op_data->op_namelen,
2019                op_data->op_name, PFID(&op_data->op_fid1));
2020
2021         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2022         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2023         op_data->op_cap = cfs_curproc_cap_pack();
2024         if (op_data->op_mea2 != NULL) {
2025                 struct lmv_stripe_md    *lsm = op_data->op_mea2;
2026                 const struct lmv_oinfo  *oinfo;
2027
2028                 oinfo = lsm_name_to_stripe_info(lsm, op_data->op_name,
2029                                                 op_data->op_namelen);
2030                 if (IS_ERR(oinfo))
2031                         RETURN(PTR_ERR(oinfo));
2032
2033                 op_data->op_fid2 = oinfo->lmo_fid;
2034         }
2035
2036         tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid2);
2037         if (IS_ERR(tgt))
2038                 RETURN(PTR_ERR(tgt));
2039
2040         /*
2041          * Cancel UPDATE lock on child (fid1).
2042          */
2043         op_data->op_flags |= MF_MDC_CANCEL_FID2;
2044         rc = lmv_early_cancel(exp, NULL, op_data, tgt->ltd_idx, LCK_EX,
2045                               MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID1);
2046         if (rc != 0)
2047                 RETURN(rc);
2048
2049         rc = md_link(tgt->ltd_exp, op_data, request);
2050
2051         RETURN(rc);
2052 }
2053
2054 static int lmv_rename(struct obd_export *exp, struct md_op_data *op_data,
2055                       const char *old, size_t oldlen,
2056                       const char *new, size_t newlen,
2057                       struct ptlrpc_request **request)
2058 {
2059         struct obd_device       *obd = exp->exp_obd;
2060         struct lmv_obd          *lmv = &obd->u.lmv;
2061         struct lmv_tgt_desc     *src_tgt;
2062         int                     rc;
2063         ENTRY;
2064
2065         LASSERT(oldlen != 0);
2066
2067         CDEBUG(D_INODE, "RENAME %.*s in "DFID":%d to %.*s in "DFID":%d\n",
2068                (int)oldlen, old, PFID(&op_data->op_fid1),
2069                op_data->op_mea1 ? op_data->op_mea1->lsm_md_stripe_count : 0,
2070                (int)newlen, new, PFID(&op_data->op_fid2),
2071                op_data->op_mea2 ? op_data->op_mea2->lsm_md_stripe_count : 0);
2072
2073         rc = lmv_check_connect(obd);
2074         if (rc)
2075                 RETURN(rc);
2076
2077         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2078         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2079         op_data->op_cap = cfs_curproc_cap_pack();
2080         if (op_data->op_cli_flags & CLI_MIGRATE) {
2081                 LASSERTF(fid_is_sane(&op_data->op_fid3), "invalid FID "DFID"\n",
2082                          PFID(&op_data->op_fid3));
2083                 rc = lmv_fid_alloc(NULL, exp, &op_data->op_fid2, op_data);
2084                 if (rc)
2085                         RETURN(rc);
2086                 src_tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid3);
2087         } else {
2088                 if (op_data->op_mea1 != NULL) {
2089                         struct lmv_stripe_md    *lsm = op_data->op_mea1;
2090
2091                         src_tgt = lmv_locate_target_for_name(lmv, lsm, old,
2092                                                              oldlen,
2093                                                              &op_data->op_fid1,
2094                                                              &op_data->op_mds);
2095                         if (IS_ERR(src_tgt))
2096                                 RETURN(PTR_ERR(src_tgt));
2097                 } else {
2098                         src_tgt = lmv_find_target(lmv, &op_data->op_fid1);
2099                         if (IS_ERR(src_tgt))
2100                                 RETURN(PTR_ERR(src_tgt));
2101
2102                         op_data->op_mds = src_tgt->ltd_idx;
2103                 }
2104
2105                 if (op_data->op_mea2) {
2106                         struct lmv_stripe_md    *lsm = op_data->op_mea2;
2107                         const struct lmv_oinfo  *oinfo;
2108
2109                         oinfo = lsm_name_to_stripe_info(lsm, new, newlen);
2110                         if (IS_ERR(oinfo))
2111                                 RETURN(PTR_ERR(oinfo));
2112
2113                         op_data->op_fid2 = oinfo->lmo_fid;
2114                 }
2115         }
2116         if (IS_ERR(src_tgt))
2117                 RETURN(PTR_ERR(src_tgt));
2118
2119         /*
2120          * LOOKUP lock on src child (fid3) should also be cancelled for
2121          * src_tgt in mdc_rename.
2122          */
2123         op_data->op_flags |= MF_MDC_CANCEL_FID1 | MF_MDC_CANCEL_FID3;
2124
2125         /*
2126          * Cancel UPDATE locks on tgt parent (fid2), tgt_tgt is its
2127          * own target.
2128          */
2129         rc = lmv_early_cancel(exp, NULL, op_data, src_tgt->ltd_idx,
2130                               LCK_EX, MDS_INODELOCK_UPDATE,
2131                               MF_MDC_CANCEL_FID2);
2132
2133         if (rc != 0)
2134                 RETURN(rc);
2135         /*
2136          * Cancel LOOKUP locks on source child (fid3) for parent tgt_tgt.
2137          */
2138         if (fid_is_sane(&op_data->op_fid3)) {
2139                 struct lmv_tgt_desc *tgt;
2140
2141                 tgt = lmv_find_target(lmv, &op_data->op_fid1);
2142                 if (IS_ERR(tgt))
2143                         RETURN(PTR_ERR(tgt));
2144
2145                 /* Cancel LOOKUP lock on its parent */
2146                 rc = lmv_early_cancel(exp, tgt, op_data, src_tgt->ltd_idx,
2147                                       LCK_EX, MDS_INODELOCK_LOOKUP,
2148                                       MF_MDC_CANCEL_FID3);
2149                 if (rc != 0)
2150                         RETURN(rc);
2151
2152                 rc = lmv_early_cancel(exp, NULL, op_data, src_tgt->ltd_idx,
2153                                       LCK_EX, MDS_INODELOCK_FULL,
2154                                       MF_MDC_CANCEL_FID3);
2155                 if (rc != 0)
2156                         RETURN(rc);
2157         }
2158
2159         /*
2160          * Cancel all the locks on tgt child (fid4).
2161          */
2162         if (fid_is_sane(&op_data->op_fid4))
2163                 rc = lmv_early_cancel(exp, NULL, op_data, src_tgt->ltd_idx,
2164                                       LCK_EX, MDS_INODELOCK_FULL,
2165                                       MF_MDC_CANCEL_FID4);
2166
2167         CDEBUG(D_INODE, DFID":m%d to "DFID"\n", PFID(&op_data->op_fid1),
2168                op_data->op_mds, PFID(&op_data->op_fid2));
2169
2170         rc = md_rename(src_tgt->ltd_exp, op_data, old, oldlen, new, newlen,
2171                        request);
2172
2173         RETURN(rc);
2174 }
2175
2176 static int lmv_setattr(struct obd_export *exp, struct md_op_data *op_data,
2177                         void *ea, size_t ealen, void *ea2, size_t ea2len,
2178                         struct ptlrpc_request **request,
2179                         struct md_open_data **mod)
2180 {
2181         struct obd_device       *obd = exp->exp_obd;
2182         struct lmv_obd          *lmv = &obd->u.lmv;
2183         struct lmv_tgt_desc     *tgt;
2184         int                      rc = 0;
2185         ENTRY;
2186
2187         rc = lmv_check_connect(obd);
2188         if (rc)
2189                 RETURN(rc);
2190
2191         CDEBUG(D_INODE, "SETATTR for "DFID", valid 0x%x\n",
2192                PFID(&op_data->op_fid1), op_data->op_attr.ia_valid);
2193
2194         op_data->op_flags |= MF_MDC_CANCEL_FID1;
2195         tgt = lmv_find_target(lmv, &op_data->op_fid1);
2196         if (IS_ERR(tgt))
2197                 RETURN(PTR_ERR(tgt));
2198
2199         rc = md_setattr(tgt->ltd_exp, op_data, ea, ealen, ea2,
2200                         ea2len, request, mod);
2201
2202         RETURN(rc);
2203 }
2204
2205 static int lmv_fsync(struct obd_export *exp, const struct lu_fid *fid,
2206                      struct obd_capa *oc, struct ptlrpc_request **request)
2207 {
2208         struct obd_device       *obd = exp->exp_obd;
2209         struct lmv_obd          *lmv = &obd->u.lmv;
2210         struct lmv_tgt_desc     *tgt;
2211         int                      rc;
2212         ENTRY;
2213
2214         rc = lmv_check_connect(obd);
2215         if (rc != 0)
2216                 RETURN(rc);
2217
2218         tgt = lmv_find_target(lmv, fid);
2219         if (IS_ERR(tgt))
2220                 RETURN(PTR_ERR(tgt));
2221
2222         rc = md_fsync(tgt->ltd_exp, fid, oc, request);
2223         RETURN(rc);
2224 }
2225
2226 /**
2227  * Get current minimum entry from striped directory
2228  *
2229  * This function will search the dir entry, whose hash value is the
2230  * closest(>=) to @hash_offset, from all of sub-stripes, and it is
2231  * only being called for striped directory.
2232  *
2233  * \param[in] exp               export of LMV
2234  * \param[in] op_data           parameters transferred beween client MD stack
2235  *                              stripe_information will be included in this
2236  *                              parameter
2237  * \param[in] cb_op             ldlm callback being used in enqueue in
2238  *                              mdc_read_page
2239  * \param[in] hash_offset       the hash value, which is used to locate
2240  *                              minum(closet) dir entry
2241  * \param[in|out] stripe_offset the caller use this to indicate the stripe
2242  *                              index of last entry, so to avoid hash conflict
2243  *                              between stripes. It will also be used to
2244  *                              return the stripe index of current dir entry.
2245  * \param[in|out] entp          the minum entry and it also is being used
2246  *                              to input the last dir entry to resolve the
2247  *                              hash conflict
2248  *
2249  * \param[out] ppage            the page which holds the minum entry
2250  *
2251  * \retval                      = 0 get the entry successfully
2252  *                              negative errno (< 0) does not get the entry
2253  */
2254 static int lmv_get_min_striped_entry(struct obd_export *exp,
2255                                      struct md_op_data *op_data,
2256                                      struct md_callback *cb_op,
2257                                      __u64 hash_offset, int *stripe_offset,
2258                                      struct lu_dirent **entp,
2259                                      struct page **ppage)
2260 {
2261         struct obd_device       *obd = exp->exp_obd;
2262         struct lmv_obd          *lmv = &obd->u.lmv;
2263         struct lmv_stripe_md    *lsm = op_data->op_mea1;
2264         struct lmv_tgt_desc     *tgt;
2265         int                     stripe_count;
2266         struct lu_dirent        *min_ent = NULL;
2267         struct page             *min_page = NULL;
2268         int                     min_idx = 0;
2269         int                     i;
2270         int                     rc = 0;
2271         ENTRY;
2272
2273         stripe_count = lsm->lsm_md_stripe_count;
2274         for (i = 0; i < stripe_count; i++) {
2275                 struct lu_dirent        *ent = NULL;
2276                 struct page             *page = NULL;
2277                 struct lu_dirpage       *dp;
2278                 __u64                   stripe_hash = hash_offset;
2279
2280                 tgt = lmv_get_target(lmv, lsm->lsm_md_oinfo[i].lmo_mds, NULL);
2281                 if (IS_ERR(tgt))
2282                         GOTO(out, rc = PTR_ERR(tgt));
2283
2284                 /* op_data will be shared by each stripe, so we need
2285                  * reset these value for each stripe */
2286                 op_data->op_fid1 = lsm->lsm_md_oinfo[i].lmo_fid;
2287                 op_data->op_fid2 = lsm->lsm_md_oinfo[i].lmo_fid;
2288                 op_data->op_data = lsm->lsm_md_oinfo[i].lmo_root;
2289 next:
2290                 rc = md_read_page(tgt->ltd_exp, op_data, cb_op, stripe_hash,
2291                                   &page);
2292                 if (rc != 0)
2293                         GOTO(out, rc);
2294
2295                 dp = page_address(page);
2296                 for (ent = lu_dirent_start(dp); ent != NULL;
2297                      ent = lu_dirent_next(ent)) {
2298                         /* Skip dummy entry */
2299                         if (le16_to_cpu(ent->lde_namelen) == 0)
2300                                 continue;
2301
2302                         if (le64_to_cpu(ent->lde_hash) < hash_offset)
2303                                 continue;
2304
2305                         if (le64_to_cpu(ent->lde_hash) == hash_offset &&
2306                             (*entp == ent || i < *stripe_offset))
2307                                 continue;
2308
2309                         /* skip . and .. for other stripes */
2310                         if (i != 0 &&
2311                             (strncmp(ent->lde_name, ".",
2312                                      le16_to_cpu(ent->lde_namelen)) == 0 ||
2313                              strncmp(ent->lde_name, "..",
2314                                      le16_to_cpu(ent->lde_namelen)) == 0))
2315                                 continue;
2316                         break;
2317                 }
2318
2319                 if (ent == NULL) {
2320                         stripe_hash = le64_to_cpu(dp->ldp_hash_end);
2321
2322                         kunmap(page);
2323                         page_cache_release(page);
2324                         page = NULL;
2325
2326                         /* reach the end of current stripe, go to next stripe */
2327                         if (stripe_hash == MDS_DIR_END_OFF)
2328                                 continue;
2329                         else
2330                                 goto next;
2331                 }
2332
2333                 if (min_ent != NULL) {
2334                         if (le64_to_cpu(min_ent->lde_hash) >
2335                             le64_to_cpu(ent->lde_hash)) {
2336                                 min_ent = ent;
2337                                 kunmap(min_page);
2338                                 page_cache_release(min_page);
2339                                 min_idx = i;
2340                                 min_page = page;
2341                         } else {
2342                                 kunmap(page);
2343                                 page_cache_release(page);
2344                                 page = NULL;
2345                         }
2346                 } else {
2347                         min_ent = ent;
2348                         min_page = page;
2349                         min_idx = i;
2350                 }
2351         }
2352
2353 out:
2354         if (*ppage != NULL) {
2355                 kunmap(*ppage);
2356                 page_cache_release(*ppage);
2357         }
2358         *stripe_offset = min_idx;
2359         *entp = min_ent;
2360         *ppage = min_page;
2361         RETURN(rc);
2362 }
2363
2364 /**
2365  * Build dir entry page from a striped directory
2366  *
2367  * This function gets one entry by @offset from a striped directory. It will
2368  * read entries from all of stripes, and choose one closest to the required
2369  * offset(&offset). A few notes
2370  * 1. skip . and .. for non-zero stripes, because there can only have one .
2371  * and .. in a directory.
2372  * 2. op_data will be shared by all of stripes, instead of allocating new
2373  * one, so need to restore before reusing.
2374  * 3. release the entry page if that is not being chosen.
2375  *
2376  * \param[in] exp       obd export refer to LMV
2377  * \param[in] op_data   hold those MD parameters of read_entry
2378  * \param[in] cb_op     ldlm callback being used in enqueue in mdc_read_entry
2379  * \param[out] ldp      the entry being read
2380  * \param[out] ppage    the page holding the entry. Note: because the entry
2381  *                      will be accessed in upper layer, so we need hold the
2382  *                      page until the usages of entry is finished, see
2383  *                      ll_dir_entry_next.
2384  *
2385  * retval               =0 if get entry successfully
2386  *                      <0 cannot get entry
2387  */
2388 static int lmv_read_striped_page(struct obd_export *exp,
2389                                  struct md_op_data *op_data,
2390                                  struct md_callback *cb_op,
2391                                  __u64 offset, struct page **ppage)
2392 {
2393         struct obd_device       *obd = exp->exp_obd;
2394         struct lu_fid           master_fid = op_data->op_fid1;
2395         struct inode            *master_inode = op_data->op_data;
2396         __u64                   hash_offset = offset;
2397         struct lu_dirpage       *dp;
2398         struct page             *min_ent_page = NULL;
2399         struct page             *ent_page = NULL;
2400         struct lu_dirent        *ent;
2401         void                    *area;
2402         int                     ent_idx = 0;
2403         struct lu_dirent        *min_ent = NULL;
2404         struct lu_dirent        *last_ent;
2405         size_t                  left_bytes;
2406         int                     rc;
2407         ENTRY;
2408
2409         rc = lmv_check_connect(obd);
2410         if (rc)
2411                 RETURN(rc);
2412
2413         /* Allocate a page and read entries from all of stripes and fill
2414          * the page by hash order */
2415         ent_page = alloc_page(GFP_KERNEL);
2416         if (ent_page == NULL)
2417                 RETURN(-ENOMEM);
2418
2419         /* Initialize the entry page */
2420         dp = kmap(ent_page);
2421         memset(dp, 0, sizeof(*dp));
2422         dp->ldp_hash_start = cpu_to_le64(offset);
2423         dp->ldp_flags |= LDF_COLLIDE;
2424
2425         area = dp + 1;
2426         left_bytes = PAGE_CACHE_SIZE - sizeof(*dp);
2427         ent = area;
2428         last_ent = ent;
2429         do {
2430                 __u16   ent_size;
2431
2432                 /* Find the minum entry from all sub-stripes */
2433                 rc = lmv_get_min_striped_entry(exp, op_data, cb_op, hash_offset,
2434                                                &ent_idx, &min_ent,
2435                                                &min_ent_page);
2436                 if (rc != 0)
2437                         GOTO(out, rc);
2438
2439                 /* If it can not get minum entry, it means it already reaches
2440                  * the end of this directory */
2441                 if (min_ent == NULL) {
2442                         last_ent->lde_reclen = 0;
2443                         hash_offset = MDS_DIR_END_OFF;
2444                         GOTO(out, rc);
2445                 }
2446
2447                 ent_size = le16_to_cpu(min_ent->lde_reclen);
2448
2449                 /* the last entry lde_reclen is 0, but it might not
2450                  * the end of this entry of this temporay entry */
2451                 if (ent_size == 0)
2452                         ent_size = lu_dirent_calc_size(
2453                                         le16_to_cpu(min_ent->lde_namelen),
2454                                         le32_to_cpu(min_ent->lde_attrs));
2455                 if (ent_size > left_bytes) {
2456                         last_ent->lde_reclen = cpu_to_le16(0);
2457                         hash_offset = le64_to_cpu(min_ent->lde_hash);
2458                         GOTO(out, rc);
2459                 }
2460
2461                 memcpy(ent, min_ent, ent_size);
2462
2463                 /* Replace . with master FID and Replace .. with the parent FID
2464                  * of master object */
2465                 if (strncmp(ent->lde_name, ".",
2466                             le16_to_cpu(ent->lde_namelen)) == 0 &&
2467                     le16_to_cpu(ent->lde_namelen) == 1)
2468                         fid_cpu_to_le(&ent->lde_fid, &master_fid);
2469                 else if (strncmp(ent->lde_name, "..",
2470                                    le16_to_cpu(ent->lde_namelen)) == 0 &&
2471                            le16_to_cpu(ent->lde_namelen) == 2)
2472                         fid_cpu_to_le(&ent->lde_fid, &op_data->op_fid3);
2473
2474                 left_bytes -= ent_size;
2475                 ent->lde_reclen = cpu_to_le16(ent_size);
2476                 last_ent = ent;
2477                 ent = (void *)ent + ent_size;
2478                 hash_offset = le64_to_cpu(min_ent->lde_hash);
2479                 if (hash_offset == MDS_DIR_END_OFF) {
2480                         last_ent->lde_reclen = 0;
2481                         break;
2482                 }
2483         } while (1);
2484 out:
2485         if (min_ent_page != NULL) {
2486                 kunmap(min_ent_page);
2487                 page_cache_release(min_ent_page);
2488         }
2489
2490         if (unlikely(rc != 0)) {
2491                 __free_page(ent_page);
2492                 ent_page = NULL;
2493         } else {
2494                 if (ent == area)
2495                         dp->ldp_flags |= LDF_EMPTY;
2496                 dp->ldp_flags = cpu_to_le32(dp->ldp_flags);
2497                 dp->ldp_hash_end = cpu_to_le64(hash_offset);
2498         }
2499
2500         /* We do not want to allocate md_op_data during each
2501          * dir entry reading, so op_data will be shared by every stripe,
2502          * then we need to restore it back to original value before
2503          * return to the upper layer */
2504         op_data->op_fid1 = master_fid;
2505         op_data->op_fid2 = master_fid;
2506         op_data->op_data = master_inode;
2507
2508         *ppage = ent_page;
2509
2510         RETURN(rc);
2511 }
2512
2513 int lmv_read_page(struct obd_export *exp, struct md_op_data *op_data,
2514                   struct md_callback *cb_op, __u64 offset,
2515                   struct page **ppage)
2516 {
2517         struct obd_device       *obd = exp->exp_obd;
2518         struct lmv_obd          *lmv = &obd->u.lmv;
2519         struct lmv_stripe_md    *lsm = op_data->op_mea1;
2520         struct lmv_tgt_desc     *tgt;
2521         int                     rc;
2522         ENTRY;
2523
2524         rc = lmv_check_connect(obd);
2525         if (rc != 0)
2526                 RETURN(rc);
2527
2528         if (unlikely(lsm != NULL)) {
2529                 rc = lmv_read_striped_page(exp, op_data, cb_op, offset, ppage);
2530                 RETURN(rc);
2531         }
2532
2533         tgt = lmv_find_target(lmv, &op_data->op_fid1);
2534         if (IS_ERR(tgt))
2535                 RETURN(PTR_ERR(tgt));
2536
2537         rc = md_read_page(tgt->ltd_exp, op_data, cb_op, offset, ppage);
2538
2539         RETURN(rc);
2540 }
2541
2542 /**
2543  * Unlink a file/directory
2544  *
2545  * Unlink a file or directory under the parent dir. The unlink request
2546  * usually will be sent to the MDT where the child is located, but if
2547  * the client does not have the child FID then request will be sent to the
2548  * MDT where the parent is located.
2549  *
2550  * If the parent is a striped directory then it also needs to locate which
2551  * stripe the name of the child is located, and replace the parent FID
2552  * (@op->op_fid1) with the stripe FID. Note: if the stripe is unknown,
2553  * it will walk through all of sub-stripes until the child is being
2554  * unlinked finally.
2555  *
2556  * \param[in] exp       export refer to LMV
2557  * \param[in] op_data   different parameters transferred beween client
2558  *                      MD stacks, name, namelen, FIDs etc.
2559  *                      op_fid1 is the parent FID, op_fid2 is the child
2560  *                      FID.
2561  * \param[out] request  point to the request of unlink.
2562  *
2563  * retval               0 if succeed
2564  *                      negative errno if failed.
2565  */
2566 static int lmv_unlink(struct obd_export *exp, struct md_op_data *op_data,
2567                       struct ptlrpc_request **request)
2568 {
2569         struct obd_device       *obd = exp->exp_obd;
2570         struct lmv_obd          *lmv = &obd->u.lmv;
2571         struct lmv_tgt_desc     *tgt = NULL;
2572         struct lmv_tgt_desc     *parent_tgt = NULL;
2573         struct mdt_body         *body;
2574         int                     rc;
2575         int                     stripe_index = 0;
2576         struct lmv_stripe_md    *lsm = op_data->op_mea1;
2577         ENTRY;
2578
2579         rc = lmv_check_connect(obd);
2580         if (rc)
2581                 RETURN(rc);
2582 retry_unlink:
2583         /* For striped dir, we need to locate the parent as well */
2584         if (lsm != NULL) {
2585                 struct lmv_tgt_desc *tmp;
2586
2587                 LASSERT(op_data->op_name != NULL &&
2588                         op_data->op_namelen != 0);
2589
2590                 tmp = lmv_locate_target_for_name(lmv, lsm,
2591                                                  op_data->op_name,
2592                                                  op_data->op_namelen,
2593                                                  &op_data->op_fid1,
2594                                                  &op_data->op_mds);
2595
2596                 /* return -EBADFD means unknown hash type, might
2597                  * need try all sub-stripe here */
2598                 if (IS_ERR(tmp) && PTR_ERR(tmp) != -EBADFD)
2599                         RETURN(PTR_ERR(tmp));
2600
2601                 /* Note: both migrating dir and unknown hash dir need to
2602                  * try all of sub-stripes, so we need start search the
2603                  * name from stripe 0, but migrating dir is already handled
2604                  * inside lmv_locate_target_for_name(), so we only check
2605                  * unknown hash type directory here */
2606                 if (!lmv_is_known_hash_type(lsm->lsm_md_hash_type)) {
2607                         struct lmv_oinfo *oinfo;
2608
2609                         oinfo = &lsm->lsm_md_oinfo[stripe_index];
2610
2611                         op_data->op_fid1 = oinfo->lmo_fid;
2612                         op_data->op_mds = oinfo->lmo_mds;
2613                 }
2614         }
2615
2616 try_next_stripe:
2617         /* Send unlink requests to the MDT where the child is located */
2618         if (likely(!fid_is_zero(&op_data->op_fid2)))
2619                 tgt = lmv_find_target(lmv, &op_data->op_fid2);
2620         else if (lsm != NULL)
2621                 tgt = lmv_get_target(lmv, op_data->op_mds, NULL);
2622         else
2623                 tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
2624
2625         if (IS_ERR(tgt))
2626                 RETURN(PTR_ERR(tgt));
2627
2628         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2629         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2630         op_data->op_cap = cfs_curproc_cap_pack();
2631
2632         /*
2633          * If child's fid is given, cancel unused locks for it if it is from
2634          * another export than parent.
2635          *
2636          * LOOKUP lock for child (fid3) should also be cancelled on parent
2637          * tgt_tgt in mdc_unlink().
2638          */
2639         op_data->op_flags |= MF_MDC_CANCEL_FID1 | MF_MDC_CANCEL_FID3;
2640
2641         /*
2642          * Cancel FULL locks on child (fid3).
2643          */
2644         parent_tgt = lmv_find_target(lmv, &op_data->op_fid1);
2645         if (IS_ERR(parent_tgt))
2646                 RETURN(PTR_ERR(parent_tgt));
2647
2648         if (parent_tgt != tgt) {
2649                 rc = lmv_early_cancel(exp, parent_tgt, op_data, tgt->ltd_idx,
2650                                       LCK_EX, MDS_INODELOCK_LOOKUP,
2651                                       MF_MDC_CANCEL_FID3);
2652         }
2653
2654         rc = lmv_early_cancel(exp, NULL, op_data, tgt->ltd_idx, LCK_EX,
2655                               MDS_INODELOCK_FULL, MF_MDC_CANCEL_FID3);
2656         if (rc != 0)
2657                 RETURN(rc);
2658
2659         CDEBUG(D_INODE, "unlink with fid="DFID"/"DFID" -> mds #%u\n",
2660                PFID(&op_data->op_fid1), PFID(&op_data->op_fid2), tgt->ltd_idx);
2661
2662         rc = md_unlink(tgt->ltd_exp, op_data, request);
2663         if (rc != 0 && rc != -EREMOTE && rc != -ENOENT)
2664                 RETURN(rc);
2665
2666         /* Try next stripe if it is needed. */
2667         if (rc == -ENOENT && lsm != NULL && lmv_need_try_all_stripes(lsm)) {
2668                 struct lmv_oinfo *oinfo;
2669
2670                 stripe_index++;
2671                 if (stripe_index >= lsm->lsm_md_stripe_count)
2672                         RETURN(rc);
2673
2674                 oinfo = &lsm->lsm_md_oinfo[stripe_index];
2675
2676                 op_data->op_fid1 = oinfo->lmo_fid;
2677                 op_data->op_mds = oinfo->lmo_mds;
2678
2679                 ptlrpc_req_finished(*request);
2680                 *request = NULL;
2681
2682                 goto try_next_stripe;
2683         }
2684
2685         body = req_capsule_server_get(&(*request)->rq_pill, &RMF_MDT_BODY);
2686         if (body == NULL)
2687                 RETURN(-EPROTO);
2688
2689         /* Not cross-ref case, just get out of here. */
2690         if (likely(!(body->mbo_valid & OBD_MD_MDS)))
2691                 RETURN(rc);
2692
2693         CDEBUG(D_INODE, "%s: try unlink to another MDT for "DFID"\n",
2694                exp->exp_obd->obd_name, PFID(&body->mbo_fid1));
2695
2696         /* This is a remote object, try remote MDT, Note: it may
2697          * try more than 1 time here, Considering following case
2698          * /mnt/lustre is root on MDT0, remote1 is on MDT1
2699          * 1. Initially A does not know where remote1 is, it send
2700          *    unlink RPC to MDT0, MDT0 return -EREMOTE, it will
2701          *    resend unlink RPC to MDT1 (retry 1st time).
2702          *
2703          * 2. During the unlink RPC in flight,
2704          *    client B mv /mnt/lustre/remote1 /mnt/lustre/remote2
2705          *    and create new remote1, but on MDT0
2706          *
2707          * 3. MDT1 get unlink RPC(from A), then do remote lock on
2708          *    /mnt/lustre, then lookup get fid of remote1, and find
2709          *    it is remote dir again, and replay -EREMOTE again.
2710          *
2711          * 4. Then A will resend unlink RPC to MDT0. (retry 2nd times).
2712          *
2713          * In theory, it might try unlimited time here, but it should
2714          * be very rare case.  */
2715         op_data->op_fid2 = body->mbo_fid1;
2716         ptlrpc_req_finished(*request);
2717         *request = NULL;
2718
2719         goto retry_unlink;
2720 }
2721
2722 static int lmv_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
2723 {
2724         struct lmv_obd *lmv = &obd->u.lmv;
2725         int rc = 0;
2726
2727         switch (stage) {
2728         case OBD_CLEANUP_EARLY:
2729                 /* XXX: here should be calling obd_precleanup() down to
2730                  * stack. */
2731                 break;
2732         case OBD_CLEANUP_EXPORTS:
2733                 fld_client_proc_fini(&lmv->lmv_fld);
2734                 lprocfs_obd_cleanup(obd);
2735                 lprocfs_free_md_stats(obd);
2736                 break;
2737         default:
2738                 break;
2739         }
2740         RETURN(rc);
2741 }
2742
2743 static int lmv_get_info(const struct lu_env *env, struct obd_export *exp,
2744                         __u32 keylen, void *key, __u32 *vallen, void *val,
2745                         struct lov_stripe_md *lsm)
2746 {
2747         struct obd_device       *obd;
2748         struct lmv_obd          *lmv;
2749         int                      rc = 0;
2750         ENTRY;
2751
2752         obd = class_exp2obd(exp);
2753         if (obd == NULL) {
2754                 CDEBUG(D_IOCTL, "Invalid client cookie "LPX64"\n",
2755                        exp->exp_handle.h_cookie);
2756                 RETURN(-EINVAL);
2757         }
2758
2759         lmv = &obd->u.lmv;
2760         if (keylen >= strlen("remote_flag") && !strcmp(key, "remote_flag")) {
2761                 int i;
2762
2763                 rc = lmv_check_connect(obd);
2764                 if (rc)
2765                         RETURN(rc);
2766
2767                 LASSERT(*vallen == sizeof(__u32));
2768                 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2769                         struct lmv_tgt_desc *tgt = lmv->tgts[i];
2770                         /*
2771                          * All tgts should be connected when this gets called.
2772                          */
2773                         if (tgt == NULL || tgt->ltd_exp == NULL)
2774                                 continue;
2775
2776                         if (!obd_get_info(env, tgt->ltd_exp, keylen, key,
2777                                           vallen, val, NULL))
2778                                 RETURN(0);
2779                 }
2780                 RETURN(-EINVAL);
2781         } else if (KEY_IS(KEY_MAX_EASIZE) ||
2782                    KEY_IS(KEY_DEFAULT_EASIZE) ||
2783                    KEY_IS(KEY_MAX_COOKIESIZE) ||
2784                    KEY_IS(KEY_DEFAULT_COOKIESIZE) ||
2785                    KEY_IS(KEY_CONN_DATA)) {
2786                 rc = lmv_check_connect(obd);
2787                 if (rc)
2788                         RETURN(rc);
2789
2790                 /*
2791                  * Forwarding this request to first MDS, it should know LOV
2792                  * desc.
2793                  */
2794                 rc = obd_get_info(env, lmv->tgts[0]->ltd_exp, keylen, key,
2795                                   vallen, val, NULL);
2796                 if (!rc && KEY_IS(KEY_CONN_DATA))
2797                         exp->exp_connect_data = *(struct obd_connect_data *)val;
2798                 RETURN(rc);
2799         } else if (KEY_IS(KEY_TGT_COUNT)) {
2800                 *((int *)val) = lmv->desc.ld_tgt_count;
2801                 RETURN(0);
2802         }
2803
2804         CDEBUG(D_IOCTL, "Invalid key\n");
2805         RETURN(-EINVAL);
2806 }
2807
2808 int lmv_set_info_async(const struct lu_env *env, struct obd_export *exp,
2809                        obd_count keylen, void *key, obd_count vallen,
2810                        void *val, struct ptlrpc_request_set *set)
2811 {
2812         struct lmv_tgt_desc    *tgt = NULL;
2813         struct obd_device      *obd;
2814         struct lmv_obd         *lmv;
2815         int rc = 0;
2816         ENTRY;
2817
2818         obd = class_exp2obd(exp);
2819         if (obd == NULL) {
2820                 CDEBUG(D_IOCTL, "Invalid client cookie "LPX64"\n",
2821                        exp->exp_handle.h_cookie);
2822                 RETURN(-EINVAL);
2823         }
2824         lmv = &obd->u.lmv;
2825
2826         if (KEY_IS(KEY_READ_ONLY) || KEY_IS(KEY_FLUSH_CTX)) {
2827                 int i, err = 0;
2828
2829                 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2830                         tgt = lmv->tgts[i];
2831
2832                         if (tgt == NULL || tgt->ltd_exp == NULL)
2833                                 continue;
2834
2835                         err = obd_set_info_async(env, tgt->ltd_exp,
2836                                                  keylen, key, vallen, val, set);
2837                         if (err && rc == 0)
2838                                 rc = err;
2839                 }
2840
2841                 RETURN(rc);
2842         }
2843
2844         RETURN(-EINVAL);
2845 }
2846
2847 static int lmv_pack_md_v1(const struct lmv_stripe_md *lsm,
2848                           struct lmv_mds_md_v1 *lmm1)
2849 {
2850         int     cplen;
2851         int     i;
2852
2853         lmm1->lmv_magic = cpu_to_le32(lsm->lsm_md_magic);
2854         lmm1->lmv_stripe_count = cpu_to_le32(lsm->lsm_md_stripe_count);
2855         lmm1->lmv_master_mdt_index = cpu_to_le32(lsm->lsm_md_master_mdt_index);
2856         lmm1->lmv_hash_type = cpu_to_le32(lsm->lsm_md_hash_type);
2857         cplen = strlcpy(lmm1->lmv_pool_name, lsm->lsm_md_pool_name,
2858                         sizeof(lmm1->lmv_pool_name));
2859         if (cplen >= sizeof(lmm1->lmv_pool_name))
2860                 return -E2BIG;
2861
2862         for (i = 0; i < lsm->lsm_md_stripe_count; i++)
2863                 fid_cpu_to_le(&lmm1->lmv_stripe_fids[i],
2864                               &lsm->lsm_md_oinfo[i].lmo_fid);
2865         return 0;
2866 }
2867
2868 int lmv_pack_md(union lmv_mds_md **lmmp, const struct lmv_stripe_md *lsm,
2869                 int stripe_count)
2870 {
2871         int     lmm_size = 0;
2872         bool    allocated = false;
2873         int     rc = 0;
2874         ENTRY;
2875
2876         LASSERT(lmmp != NULL);
2877         /* Free lmm */
2878         if (*lmmp != NULL && lsm == NULL) {
2879                 int stripe_count;
2880
2881                 stripe_count = lmv_mds_md_stripe_count_get(*lmmp);
2882                 lmm_size = lmv_mds_md_size(stripe_count,
2883                                            le32_to_cpu((*lmmp)->lmv_magic));
2884                 if (lmm_size == 0)
2885                         RETURN(-EINVAL);
2886                 OBD_FREE(*lmmp, lmm_size);
2887                 *lmmp = NULL;
2888                 RETURN(0);
2889         }
2890
2891         /* Alloc lmm */
2892         if (*lmmp == NULL && lsm == NULL) {
2893                 lmm_size = lmv_mds_md_size(stripe_count, LMV_MAGIC);
2894                 LASSERT(lmm_size > 0);
2895                 OBD_ALLOC(*lmmp, lmm_size);
2896                 if (*lmmp == NULL)
2897                         RETURN(-ENOMEM);
2898                 lmv_mds_md_stripe_count_set(*lmmp, stripe_count);
2899                 (*lmmp)->lmv_magic = cpu_to_le32(LMV_MAGIC);
2900                 RETURN(lmm_size);
2901         }
2902
2903         /* pack lmm */
2904         LASSERT(lsm != NULL);
2905         lmm_size = lmv_mds_md_size(lsm->lsm_md_stripe_count, lsm->lsm_md_magic);
2906         if (*lmmp == NULL) {
2907                 OBD_ALLOC(*lmmp, lmm_size);
2908                 if (*lmmp == NULL)
2909                         RETURN(-ENOMEM);
2910                 allocated = true;
2911         }
2912
2913         switch (lsm->lsm_md_magic) {
2914         case LMV_MAGIC_V1:
2915                 rc = lmv_pack_md_v1(lsm, &(*lmmp)->lmv_md_v1);
2916                 break;
2917         default:
2918                 rc = -EINVAL;
2919                 break;
2920         }
2921
2922         if (rc != 0 && allocated) {
2923                 OBD_FREE(*lmmp, lmm_size);
2924                 *lmmp = NULL;
2925         }
2926
2927         RETURN(lmm_size);
2928 }
2929 EXPORT_SYMBOL(lmv_pack_md);
2930
2931 static int lmv_unpack_md_v1(struct obd_export *exp, struct lmv_stripe_md *lsm,
2932                             const struct lmv_mds_md_v1 *lmm1)
2933 {
2934         struct lmv_obd  *lmv = &exp->exp_obd->u.lmv;
2935         int             stripe_count;
2936         int             cplen;
2937         int             i;
2938         int             rc = 0;
2939         ENTRY;
2940
2941         lsm->lsm_md_magic = le32_to_cpu(lmm1->lmv_magic);
2942         lsm->lsm_md_stripe_count = le32_to_cpu(lmm1->lmv_stripe_count);
2943         lsm->lsm_md_master_mdt_index = le32_to_cpu(lmm1->lmv_master_mdt_index);
2944         if (OBD_FAIL_CHECK(OBD_FAIL_UNKNOWN_LMV_STRIPE))
2945                 lsm->lsm_md_hash_type = LMV_HASH_TYPE_UNKNOWN;
2946         else
2947                 lsm->lsm_md_hash_type = le32_to_cpu(lmm1->lmv_hash_type);
2948         lsm->lsm_md_layout_version = le32_to_cpu(lmm1->lmv_layout_version);
2949         cplen = strlcpy(lsm->lsm_md_pool_name, lmm1->lmv_pool_name,
2950                         sizeof(lsm->lsm_md_pool_name));
2951
2952         if (cplen >= sizeof(lsm->lsm_md_pool_name))
2953                 RETURN(-E2BIG);
2954
2955         CDEBUG(D_INFO, "unpack lsm count %d, master %d hash_type %d"
2956                "layout_version %d\n", lsm->lsm_md_stripe_count,
2957                lsm->lsm_md_master_mdt_index, lsm->lsm_md_hash_type,
2958                lsm->lsm_md_layout_version);
2959
2960         stripe_count = le32_to_cpu(lmm1->lmv_stripe_count);
2961         for (i = 0; i < stripe_count; i++) {
2962                 fid_le_to_cpu(&lsm->lsm_md_oinfo[i].lmo_fid,
2963                               &lmm1->lmv_stripe_fids[i]);
2964                 rc = lmv_fld_lookup(lmv, &lsm->lsm_md_oinfo[i].lmo_fid,
2965                                     &lsm->lsm_md_oinfo[i].lmo_mds);
2966                 if (rc != 0)
2967                         RETURN(rc);
2968                 CDEBUG(D_INFO, "unpack fid #%d "DFID"\n", i,
2969                        PFID(&lsm->lsm_md_oinfo[i].lmo_fid));
2970         }
2971
2972         RETURN(rc);
2973 }
2974
2975 int lmv_unpack_md(struct obd_export *exp, struct lmv_stripe_md **lsmp,
2976                   const union lmv_mds_md *lmm, int stripe_count)
2977 {
2978         struct lmv_stripe_md     *lsm;
2979         int                      lsm_size;
2980         int                      rc;
2981         bool                     allocated = false;
2982         ENTRY;
2983
2984         LASSERT(lsmp != NULL);
2985
2986         lsm = *lsmp;
2987         /* Free memmd */
2988         if (lsm != NULL && lmm == NULL) {
2989                 int i;
2990                 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
2991                         /* For migrating inode, the master stripe and master
2992                          * object will be the same, so do not need iput, see
2993                          * ll_update_lsm_md */
2994                         if (!(lsm->lsm_md_hash_type & LMV_HASH_FLAG_MIGRATION &&
2995                               i == 0) && lsm->lsm_md_oinfo[i].lmo_root != NULL)
2996                                 iput(lsm->lsm_md_oinfo[i].lmo_root);
2997                 }
2998                 lsm_size = lmv_stripe_md_size(lsm->lsm_md_stripe_count);
2999                 OBD_FREE(lsm, lsm_size);
3000                 *lsmp = NULL;
3001                 RETURN(0);
3002         }
3003
3004         /* Alloc memmd */
3005         if (lsm == NULL && lmm == NULL) {
3006                 lsm_size = lmv_stripe_md_size(stripe_count);
3007                 OBD_ALLOC(lsm, lsm_size);
3008                 if (lsm == NULL)
3009                         RETURN(-ENOMEM);
3010                 lsm->lsm_md_stripe_count = stripe_count;
3011                 *lsmp = lsm;
3012                 RETURN(0);
3013         }
3014
3015         if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_STRIPE)
3016                 RETURN(-EPERM);
3017
3018         /* Unpack memmd */
3019         if (le32_to_cpu(lmm->lmv_magic) != LMV_MAGIC_V1 &&
3020             le32_to_cpu(lmm->lmv_magic) != LMV_USER_MAGIC) {
3021                 CERROR("%s: invalid lmv magic %x: rc = %d\n",
3022                        exp->exp_obd->obd_name, le32_to_cpu(lmm->lmv_magic),
3023                        -EIO);
3024                 RETURN(-EIO);
3025         }
3026
3027         if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_V1)
3028                 lsm_size = lmv_stripe_md_size(lmv_mds_md_stripe_count_get(lmm));
3029         else
3030                 /**
3031                  * Unpack default dirstripe(lmv_user_md) to lmv_stripe_md,
3032                  * stripecount should be 0 then.
3033                  */
3034                 lsm_size = lmv_stripe_md_size(0);
3035
3036         lsm_size = lmv_stripe_md_size(lmv_mds_md_stripe_count_get(lmm));
3037         if (lsm == NULL) {
3038                 OBD_ALLOC(lsm, lsm_size);
3039                 if (lsm == NULL)
3040                         RETURN(-ENOMEM);
3041                 allocated = true;
3042                 *lsmp = lsm;
3043         }
3044
3045         switch (le32_to_cpu(lmm->lmv_magic)) {
3046         case LMV_MAGIC_V1:
3047                 rc = lmv_unpack_md_v1(exp, lsm, &lmm->lmv_md_v1);
3048                 break;
3049         default:
3050                 CERROR("%s: unrecognized magic %x\n", exp->exp_obd->obd_name,
3051                        le32_to_cpu(lmm->lmv_magic));
3052                 rc = -EINVAL;
3053                 break;
3054         }
3055
3056         if (rc != 0 && allocated) {
3057                 OBD_FREE(lsm, lsm_size);
3058                 *lsmp = NULL;
3059                 lsm_size = rc;
3060         }
3061         RETURN(lsm_size);
3062 }
3063
3064 int lmv_alloc_memmd(struct lmv_stripe_md **lsmp, int stripes)
3065 {
3066         return lmv_unpack_md(NULL, lsmp, NULL, stripes);
3067 }
3068 EXPORT_SYMBOL(lmv_alloc_memmd);
3069
3070 void lmv_free_memmd(struct lmv_stripe_md *lsm)
3071 {
3072         lmv_unpack_md(NULL, &lsm, NULL, 0);
3073 }
3074 EXPORT_SYMBOL(lmv_free_memmd);
3075
3076 int lmv_unpackmd(struct obd_export *exp, struct lov_stripe_md **lsmp,
3077                  struct lov_mds_md *lmm, int disk_len)
3078 {
3079         return lmv_unpack_md(exp, (struct lmv_stripe_md **)lsmp,
3080                              (union lmv_mds_md *)lmm, disk_len);
3081 }
3082
3083 int lmv_packmd(struct obd_export *exp, struct lov_mds_md **lmmp,
3084                struct lov_stripe_md *lsm)
3085 {
3086         struct obd_device               *obd = exp->exp_obd;
3087         struct lmv_obd                  *lmv_obd = &obd->u.lmv;
3088         const struct lmv_stripe_md      *lmv = (struct lmv_stripe_md *)lsm;
3089         int                             stripe_count;
3090
3091         if (lmmp == NULL) {
3092                 if (lsm != NULL)
3093                         stripe_count = lmv->lsm_md_stripe_count;
3094                 else
3095                         stripe_count = lmv_obd->desc.ld_tgt_count;
3096
3097                 return lmv_mds_md_size(stripe_count, LMV_MAGIC_V1);
3098         }
3099
3100         return lmv_pack_md((union lmv_mds_md **)lmmp, lmv, 0);
3101 }
3102
3103 static int lmv_cancel_unused(struct obd_export *exp, const struct lu_fid *fid,
3104                              ldlm_policy_data_t *policy, ldlm_mode_t mode,
3105                              ldlm_cancel_flags_t flags, void *opaque)
3106 {
3107         struct obd_device       *obd = exp->exp_obd;
3108         struct lmv_obd          *lmv = &obd->u.lmv;
3109         int                      rc = 0;
3110         int                      err;
3111         __u32                    i;
3112         ENTRY;
3113
3114         LASSERT(fid != NULL);
3115
3116         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
3117                 struct lmv_tgt_desc *tgt = lmv->tgts[i];
3118
3119                 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active)
3120                         continue;
3121
3122                 err = md_cancel_unused(tgt->ltd_exp, fid, policy, mode, flags,
3123                                        opaque);
3124                 if (!rc)
3125                         rc = err;
3126         }
3127         RETURN(rc);
3128 }
3129
3130 int lmv_set_lock_data(struct obd_export *exp, __u64 *lockh, void *data,
3131                       __u64 *bits)
3132 {
3133         struct lmv_obd          *lmv = &exp->exp_obd->u.lmv;
3134         struct lmv_tgt_desc     *tgt = lmv->tgts[0];
3135         int                      rc;
3136         ENTRY;
3137
3138         if (tgt == NULL || tgt->ltd_exp == NULL)
3139                 RETURN(-EINVAL);
3140         rc =  md_set_lock_data(tgt->ltd_exp, lockh, data, bits);
3141         RETURN(rc);
3142 }
3143
3144 ldlm_mode_t lmv_lock_match(struct obd_export *exp, __u64 flags,
3145                            const struct lu_fid *fid, ldlm_type_t type,
3146                            ldlm_policy_data_t *policy, ldlm_mode_t mode,
3147                            struct lustre_handle *lockh)
3148 {
3149         struct obd_device       *obd = exp->exp_obd;
3150         struct lmv_obd          *lmv = &obd->u.lmv;
3151         ldlm_mode_t             rc;
3152         int                     tgt;
3153         int                     i;
3154         ENTRY;
3155
3156         CDEBUG(D_INODE, "Lock match for "DFID"\n", PFID(fid));
3157
3158         /*
3159          * With DNE every object can have two locks in different namespaces:
3160          * lookup lock in space of MDT storing direntry and update/open lock in
3161          * space of MDT storing inode.  Try the MDT that the FID maps to first,
3162          * since this can be easily found, and only try others if that fails.
3163          */
3164         for (i = 0, tgt = lmv_find_target_index(lmv, fid);
3165              i < lmv->desc.ld_tgt_count;
3166              i++, tgt = (tgt + 1) % lmv->desc.ld_tgt_count) {
3167                 if (tgt < 0) {
3168                         CDEBUG(D_HA, "%s: "DFID" is inaccessible: rc = %d\n",
3169                                obd->obd_name, PFID(fid), tgt);
3170                         tgt = 0;
3171                 }
3172
3173                 if (lmv->tgts[tgt] == NULL ||
3174                     lmv->tgts[tgt]->ltd_exp == NULL ||
3175                     lmv->tgts[tgt]->ltd_active == 0)
3176                         continue;
3177
3178                 rc = md_lock_match(lmv->tgts[tgt]->ltd_exp, flags, fid,
3179                                    type, policy, mode, lockh);
3180                 if (rc)
3181                         RETURN(rc);
3182         }
3183
3184         RETURN(0);
3185 }
3186
3187 int lmv_get_lustre_md(struct obd_export *exp, struct ptlrpc_request *req,
3188                       struct obd_export *dt_exp, struct obd_export *md_exp,
3189                       struct lustre_md *md)
3190 {
3191         struct lmv_obd          *lmv = &exp->exp_obd->u.lmv;
3192         struct lmv_tgt_desc     *tgt = lmv->tgts[0];
3193
3194         if (tgt == NULL || tgt->ltd_exp == NULL)
3195                 RETURN(-EINVAL);
3196
3197         return md_get_lustre_md(lmv->tgts[0]->ltd_exp, req, dt_exp, md_exp, md);
3198 }
3199
3200 int lmv_free_lustre_md(struct obd_export *exp, struct lustre_md *md)
3201 {
3202         struct obd_device       *obd = exp->exp_obd;
3203         struct lmv_obd          *lmv = &obd->u.lmv;
3204         struct lmv_tgt_desc     *tgt = lmv->tgts[0];
3205         ENTRY;
3206
3207         if (md->lmv != NULL) {
3208                 lmv_free_memmd(md->lmv);
3209                 md->lmv = NULL;
3210         }
3211         if (tgt == NULL || tgt->ltd_exp == NULL)
3212                 RETURN(-EINVAL);
3213         RETURN(md_free_lustre_md(lmv->tgts[0]->ltd_exp, md));
3214 }
3215
3216 int lmv_set_open_replay_data(struct obd_export *exp,
3217                              struct obd_client_handle *och,
3218                              struct lookup_intent *it)
3219 {
3220         struct obd_device       *obd = exp->exp_obd;
3221         struct lmv_obd          *lmv = &obd->u.lmv;
3222         struct lmv_tgt_desc     *tgt;
3223         ENTRY;
3224
3225         tgt = lmv_find_target(lmv, &och->och_fid);
3226         if (IS_ERR(tgt))
3227                 RETURN(PTR_ERR(tgt));
3228
3229         RETURN(md_set_open_replay_data(tgt->ltd_exp, och, it));
3230 }
3231
3232 int lmv_clear_open_replay_data(struct obd_export *exp,
3233                                struct obd_client_handle *och)
3234 {
3235         struct obd_device       *obd = exp->exp_obd;
3236         struct lmv_obd          *lmv = &obd->u.lmv;
3237         struct lmv_tgt_desc     *tgt;
3238         ENTRY;
3239
3240         tgt = lmv_find_target(lmv, &och->och_fid);
3241         if (IS_ERR(tgt))
3242                 RETURN(PTR_ERR(tgt));
3243
3244         RETURN(md_clear_open_replay_data(tgt->ltd_exp, och));
3245 }
3246
3247 static int lmv_get_remote_perm(struct obd_export *exp,
3248                                const struct lu_fid *fid,
3249                                struct obd_capa *oc, __u32 suppgid,
3250                                struct ptlrpc_request **request)
3251 {
3252         struct obd_device       *obd = exp->exp_obd;
3253         struct lmv_obd          *lmv = &obd->u.lmv;
3254         struct lmv_tgt_desc     *tgt;
3255         int                      rc;
3256         ENTRY;
3257
3258         rc = lmv_check_connect(obd);
3259         if (rc)
3260                 RETURN(rc);
3261
3262         tgt = lmv_find_target(lmv, fid);
3263         if (IS_ERR(tgt))
3264                 RETURN(PTR_ERR(tgt));
3265
3266         rc = md_get_remote_perm(tgt->ltd_exp, fid, oc, suppgid, request);
3267         RETURN(rc);
3268 }
3269
3270 static int lmv_renew_capa(struct obd_export *exp, struct obd_capa *oc,
3271                           renew_capa_cb_t cb)
3272 {
3273         struct obd_device       *obd = exp->exp_obd;
3274         struct lmv_obd          *lmv = &obd->u.lmv;
3275         struct lmv_tgt_desc     *tgt;
3276         int                      rc;
3277         ENTRY;
3278
3279         rc = lmv_check_connect(obd);
3280         if (rc)
3281                 RETURN(rc);
3282
3283         tgt = lmv_find_target(lmv, &oc->c_capa.lc_fid);
3284         if (IS_ERR(tgt))
3285                 RETURN(PTR_ERR(tgt));
3286
3287         rc = md_renew_capa(tgt->ltd_exp, oc, cb);
3288         RETURN(rc);
3289 }
3290
3291 int lmv_unpack_capa(struct obd_export *exp, struct ptlrpc_request *req,
3292                     const struct req_msg_field *field, struct obd_capa **oc)
3293 {
3294         struct lmv_obd          *lmv = &exp->exp_obd->u.lmv;
3295         struct lmv_tgt_desc     *tgt = lmv->tgts[0];
3296
3297         if (tgt == NULL || tgt->ltd_exp == NULL)
3298                 RETURN(-EINVAL);
3299         return md_unpack_capa(tgt->ltd_exp, req, field, oc);
3300 }
3301
3302 int lmv_intent_getattr_async(struct obd_export *exp,
3303                              struct md_enqueue_info *minfo,
3304                              struct ldlm_enqueue_info *einfo)
3305 {
3306         struct md_op_data       *op_data = &minfo->mi_data;
3307         struct obd_device       *obd = exp->exp_obd;
3308         struct lmv_obd          *lmv = &obd->u.lmv;
3309         struct lmv_tgt_desc     *tgt = NULL;
3310         int                      rc;
3311         ENTRY;
3312
3313         rc = lmv_check_connect(obd);
3314         if (rc)
3315                 RETURN(rc);
3316
3317         tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
3318         if (IS_ERR(tgt))
3319                 RETURN(PTR_ERR(tgt));
3320
3321         rc = md_intent_getattr_async(tgt->ltd_exp, minfo, einfo);
3322         RETURN(rc);
3323 }
3324
3325 int lmv_revalidate_lock(struct obd_export *exp, struct lookup_intent *it,
3326                         struct lu_fid *fid, __u64 *bits)
3327 {
3328         struct obd_device       *obd = exp->exp_obd;
3329         struct lmv_obd          *lmv = &obd->u.lmv;
3330         struct lmv_tgt_desc     *tgt;
3331         int                      rc;
3332         ENTRY;
3333
3334         rc = lmv_check_connect(obd);
3335         if (rc)
3336                 RETURN(rc);
3337
3338         tgt = lmv_find_target(lmv, fid);
3339         if (IS_ERR(tgt))
3340                 RETURN(PTR_ERR(tgt));
3341
3342         rc = md_revalidate_lock(tgt->ltd_exp, it, fid, bits);
3343         RETURN(rc);
3344 }
3345
3346 int lmv_get_fid_from_lsm(struct obd_export *exp,
3347                          const struct lmv_stripe_md *lsm,
3348                          const char *name, int namelen, struct lu_fid *fid)
3349 {
3350         const struct lmv_oinfo *oinfo;
3351
3352         LASSERT(lsm != NULL);
3353         oinfo = lsm_name_to_stripe_info(lsm, name, namelen);
3354         if (IS_ERR(oinfo))
3355                 return PTR_ERR(oinfo);
3356
3357         *fid = oinfo->lmo_fid;
3358
3359         RETURN(0);
3360 }
3361
3362 /**
3363  * For lmv, only need to send request to master MDT, and the master MDT will
3364  * process with other slave MDTs. The only exception is Q_GETOQUOTA for which
3365  * we directly fetch data from the slave MDTs.
3366  */
3367 int lmv_quotactl(struct obd_device *unused, struct obd_export *exp,
3368                  struct obd_quotactl *oqctl)
3369 {
3370         struct obd_device   *obd = class_exp2obd(exp);
3371         struct lmv_obd      *lmv = &obd->u.lmv;
3372         struct lmv_tgt_desc *tgt = lmv->tgts[0];
3373         int                  rc = 0;
3374         __u32                i;
3375         __u64                curspace, curinodes;
3376         ENTRY;
3377
3378         if (tgt == NULL ||
3379             tgt->ltd_exp == NULL ||
3380             !tgt->ltd_active ||
3381             lmv->desc.ld_tgt_count == 0) {
3382                 CERROR("master lmv inactive\n");
3383                 RETURN(-EIO);
3384         }
3385
3386         if (oqctl->qc_cmd != Q_GETOQUOTA) {
3387                 rc = obd_quotactl(tgt->ltd_exp, oqctl);
3388                 RETURN(rc);
3389         }
3390
3391         curspace = curinodes = 0;
3392         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
3393                 int err;
3394                 tgt = lmv->tgts[i];
3395
3396                 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active)
3397                         continue;
3398
3399                 err = obd_quotactl(tgt->ltd_exp, oqctl);
3400                 if (err) {
3401                         CERROR("getquota on mdt %d failed. %d\n", i, err);
3402                         if (!rc)
3403                                 rc = err;
3404                 } else {
3405                         curspace += oqctl->qc_dqblk.dqb_curspace;
3406                         curinodes += oqctl->qc_dqblk.dqb_curinodes;
3407                 }
3408         }
3409         oqctl->qc_dqblk.dqb_curspace = curspace;
3410         oqctl->qc_dqblk.dqb_curinodes = curinodes;
3411
3412         RETURN(rc);
3413 }
3414
3415 int lmv_quotacheck(struct obd_device *unused, struct obd_export *exp,
3416                    struct obd_quotactl *oqctl)
3417 {
3418         struct obd_device       *obd = class_exp2obd(exp);
3419         struct lmv_obd          *lmv = &obd->u.lmv;
3420         struct lmv_tgt_desc     *tgt;
3421         __u32                    i;
3422         int                      rc = 0;
3423         ENTRY;
3424
3425         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
3426                 int err;
3427                 tgt = lmv->tgts[i];
3428                 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active) {
3429                         CERROR("lmv idx %d inactive\n", i);
3430                         RETURN(-EIO);
3431                 }
3432
3433                 err = obd_quotacheck(tgt->ltd_exp, oqctl);
3434                 if (err && !rc)
3435                         rc = err;
3436         }
3437
3438         RETURN(rc);
3439 }
3440
3441 static int lmv_merge_attr(struct obd_export *exp,
3442                           const struct lmv_stripe_md *lsm,
3443                           struct cl_attr *attr,
3444                           ldlm_blocking_callback cb_blocking)
3445 {
3446         int rc;
3447         int i;
3448
3449         rc = lmv_revalidate_slaves(exp, lsm, cb_blocking, 0);
3450         if (rc < 0)
3451                 return rc;
3452
3453         for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
3454                 struct inode *inode = lsm->lsm_md_oinfo[i].lmo_root;
3455
3456                 CDEBUG(D_INFO, ""DFID" size %llu, blocks %llu nlink %u,"
3457                        " atime %lu ctime %lu, mtime %lu.\n",
3458                        PFID(&lsm->lsm_md_oinfo[i].lmo_fid),
3459                        i_size_read(inode), (unsigned long long)inode->i_blocks,
3460                        inode->i_nlink, LTIME_S(inode->i_atime),
3461                        LTIME_S(inode->i_ctime), LTIME_S(inode->i_mtime));
3462
3463                 /* for slave stripe, it needs to subtract nlink for . and .. */
3464                 if (i != 0)
3465                         attr->cat_nlink += inode->i_nlink - 2;
3466                 else
3467                         attr->cat_nlink = inode->i_nlink;
3468
3469                 attr->cat_size += i_size_read(inode);
3470                 attr->cat_blocks += inode->i_blocks;
3471
3472                 if (attr->cat_atime < LTIME_S(inode->i_atime))
3473                         attr->cat_atime = LTIME_S(inode->i_atime);
3474
3475                 if (attr->cat_ctime < LTIME_S(inode->i_ctime))
3476                         attr->cat_ctime = LTIME_S(inode->i_ctime);
3477
3478                 if (attr->cat_mtime < LTIME_S(inode->i_mtime))
3479                         attr->cat_mtime = LTIME_S(inode->i_mtime);
3480         }
3481         return 0;
3482 }
3483
3484 struct obd_ops lmv_obd_ops = {
3485         .o_owner                = THIS_MODULE,
3486         .o_setup                = lmv_setup,
3487         .o_cleanup              = lmv_cleanup,
3488         .o_precleanup           = lmv_precleanup,
3489         .o_process_config       = lmv_process_config,
3490         .o_connect              = lmv_connect,
3491         .o_disconnect           = lmv_disconnect,
3492         .o_statfs               = lmv_statfs,
3493         .o_get_info             = lmv_get_info,
3494         .o_set_info_async       = lmv_set_info_async,
3495         .o_packmd               = lmv_packmd,
3496         .o_unpackmd             = lmv_unpackmd,
3497         .o_notify               = lmv_notify,
3498         .o_get_uuid             = lmv_get_uuid,
3499         .o_iocontrol            = lmv_iocontrol,
3500         .o_quotacheck           = lmv_quotacheck,
3501         .o_quotactl             = lmv_quotactl
3502 };
3503
3504 struct md_ops lmv_md_ops = {
3505         .m_getstatus            = lmv_getstatus,
3506         .m_null_inode           = lmv_null_inode,
3507         .m_find_cbdata          = lmv_find_cbdata,
3508         .m_close                = lmv_close,
3509         .m_create               = lmv_create,
3510         .m_done_writing         = lmv_done_writing,
3511         .m_enqueue              = lmv_enqueue,
3512         .m_getattr              = lmv_getattr,
3513         .m_getxattr             = lmv_getxattr,
3514         .m_getattr_name         = lmv_getattr_name,
3515         .m_intent_lock          = lmv_intent_lock,
3516         .m_link                 = lmv_link,
3517         .m_rename               = lmv_rename,
3518         .m_setattr              = lmv_setattr,
3519         .m_setxattr             = lmv_setxattr,
3520         .m_fsync                = lmv_fsync,
3521         .m_read_page            = lmv_read_page,
3522         .m_unlink               = lmv_unlink,
3523         .m_init_ea_size         = lmv_init_ea_size,
3524         .m_cancel_unused        = lmv_cancel_unused,
3525         .m_set_lock_data        = lmv_set_lock_data,
3526         .m_lock_match           = lmv_lock_match,
3527         .m_get_lustre_md        = lmv_get_lustre_md,
3528         .m_free_lustre_md       = lmv_free_lustre_md,
3529         .m_merge_attr           = lmv_merge_attr,
3530         .m_set_open_replay_data = lmv_set_open_replay_data,
3531         .m_clear_open_replay_data = lmv_clear_open_replay_data,
3532         .m_renew_capa           = lmv_renew_capa,
3533         .m_unpack_capa          = lmv_unpack_capa,
3534         .m_get_remote_perm      = lmv_get_remote_perm,
3535         .m_intent_getattr_async = lmv_intent_getattr_async,
3536         .m_revalidate_lock      = lmv_revalidate_lock,
3537         .m_get_fid_from_lsm     = lmv_get_fid_from_lsm,
3538 };
3539
3540 int __init lmv_init(void)
3541 {
3542         return class_register_type(&lmv_obd_ops, &lmv_md_ops, true, NULL,
3543                                    LUSTRE_LMV_NAME, NULL);
3544 }
3545
3546 static void lmv_exit(void)
3547 {
3548         class_unregister_type(LUSTRE_LMV_NAME);
3549 }
3550
3551 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
3552 MODULE_DESCRIPTION("Lustre Logical Metadata Volume OBD driver");
3553 MODULE_LICENSE("GPL");
3554
3555 module_init(lmv_init);
3556 module_exit(lmv_exit);