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