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.gnu.org/licenses/gpl-2.0.html
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2004, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Use is subject to license terms.
  25  *
  26  * Copyright (c) 2011, 2017, Intel Corporation.
  27  */
  28 /*
  29  * This file is part of Lustre, http://www.lustre.org/
  30  * Lustre is a trademark of Sun Microsystems, Inc.
  31  */
  32
  33 #define DEBUG_SUBSYSTEM S_LMV
  34
  35 #include <linux/file.h>
  36 #include <linux/module.h>
  37 #include <linux/init.h>
  38 #include <linux/user_namespace.h>
  39 #ifdef HAVE_UIDGID_HEADER
  40 # include <linux/uidgid.h>
  41 #endif
  42 #include <linux/slab.h>
  43 #include <linux/pagemap.h>
  44 #include <linux/mm.h>
  45 #include <linux/math64.h>
  46 #include <linux/seq_file.h>
  47 #include <linux/namei.h>
  48
  49 #include <obd_support.h>
  50 #include <lustre_lib.h>
  51 #include <lustre_net.h>
  52 #include <obd_class.h>
  53 #include <lustre_lmv.h>
  54 #include <lprocfs_status.h>
  55 #include <cl_object.h>
  56 #include <lustre_fid.h>
  57 #include <uapi/linux/lustre/lustre_ioctl.h>
  58 #include <lustre_kernelcomm.h>
  59 #include "lmv_internal.h"
  60
  61 static int lmv_check_connect(struct obd_device *obd);
  62
  63 static void lmv_activate_target(struct lmv_obd *lmv,
  64                                 struct lmv_tgt_desc *tgt,
  65                                 int activate)
  66 {
  67         if (tgt->ltd_active == activate)
  68                 return;
  69
  70         tgt->ltd_active = activate;
  71         lmv->desc.ld_active_tgt_count += (activate ? 1 : -1);
  72
  73         tgt->ltd_exp->exp_obd->obd_inactive = !activate;
  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 %#llx\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)
 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
 185         /*
 186          * Pass the notification up the chain.
 187          */
 188         if (obd->obd_observer)
 189                 rc = obd_notify(obd->obd_observer, watched, ev);
 190
 191         RETURN(rc);
 192 }
 193
 194 static int lmv_connect(const struct lu_env *env,
 195                        struct obd_export **pexp, struct obd_device *obd,
 196                        struct obd_uuid *cluuid, struct obd_connect_data *data,
 197                        void *localdata)
 198 {
 199         struct lmv_obd *lmv = &obd->u.lmv;
 200         struct lustre_handle conn = { 0 };
 201         struct obd_export *exp;
 202         int rc;
 203         ENTRY;
 204
 205         rc = class_connect(&conn, obd, cluuid);
 206         if (rc) {
 207                 CERROR("class_connection() returned %d\n", rc);
 208                 RETURN(rc);
 209         }
 210
 211         exp = class_conn2export(&conn);
 212
 213         lmv->connected = 0;
 214         lmv->conn_data = *data;
 215
 216         lmv->lmv_tgts_kobj = kobject_create_and_add("target_obds",
 217                                                     &obd->obd_kset.kobj);
 218         if (!lmv->lmv_tgts_kobj) {
 219                 CERROR("%s: cannot create /sys/fs/lustre/%s/%s/target_obds\n",
 220                        obd->obd_name, obd->obd_type->typ_name, obd->obd_name);
 221         }
 222
 223         rc = lmv_check_connect(obd);
 224         if (rc != 0)
 225                 GOTO(out_sysfs, rc);
 226
 227         *pexp = exp;
 228
 229         RETURN(rc);
 230
 231 out_sysfs:
 232         if (lmv->lmv_tgts_kobj)
 233                 kobject_put(lmv->lmv_tgts_kobj);
 234
 235         class_disconnect(exp);
 236
 237         return rc;
 238 }
 239
 240 static int lmv_init_ea_size(struct obd_export *exp, __u32 easize,
 241                             __u32 def_easize)
 242 {
 243         struct obd_device       *obd = exp->exp_obd;
 244         struct lmv_obd          *lmv = &obd->u.lmv;
 245         __u32                    i;
 246         int                      rc = 0;
 247         int                      change = 0;
 248         ENTRY;
 249
 250         if (lmv->max_easize < easize) {
 251                 lmv->max_easize = easize;
 252                 change = 1;
 253         }
 254         if (lmv->max_def_easize < def_easize) {
 255                 lmv->max_def_easize = def_easize;
 256                 change = 1;
 257         }
 258
 259         if (change == 0)
 260                 RETURN(0);
 261
 262         if (lmv->connected == 0)
 263                 RETURN(0);
 264
 265         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 266                 struct lmv_tgt_desc *tgt = lmv->tgts[i];
 267
 268                 if (tgt == NULL || tgt->ltd_exp == NULL) {
 269                         CWARN("%s: NULL export for %d\n", obd->obd_name, i);
 270                         continue;
 271                 }
 272                 if (!tgt->ltd_active)
 273                         continue;
 274
 275                 rc = md_init_ea_size(tgt->ltd_exp, easize, def_easize);
 276                 if (rc) {
 277                         CERROR("%s: obd_init_ea_size() failed on MDT target %d:"
 278                                " rc = %d\n", obd->obd_name, i, rc);
 279                         break;
 280                 }
 281         }
 282         RETURN(rc);
 283 }
 284
 285 #define MAX_STRING_SIZE 128
 286
 287 int lmv_connect_mdc(struct obd_device *obd, struct lmv_tgt_desc *tgt)
 288 {
 289         struct lmv_obd          *lmv = &obd->u.lmv;
 290         struct obd_device       *mdc_obd;
 291         struct obd_export       *mdc_exp;
 292         struct lu_fld_target     target;
 293         int                      rc;
 294         ENTRY;
 295
 296         mdc_obd = class_find_client_obd(&tgt->ltd_uuid, LUSTRE_MDC_NAME,
 297                                         &obd->obd_uuid);
 298         if (!mdc_obd) {
 299                 CERROR("target %s not attached\n", tgt->ltd_uuid.uuid);
 300                 RETURN(-EINVAL);
 301         }
 302
 303         CDEBUG(D_CONFIG, "connect to %s(%s) - %s, %s\n",
 304                mdc_obd->obd_name, mdc_obd->obd_uuid.uuid,
 305                tgt->ltd_uuid.uuid, obd->obd_uuid.uuid);
 306
 307         if (!mdc_obd->obd_set_up) {
 308                 CERROR("target %s is not set up\n", tgt->ltd_uuid.uuid);
 309                 RETURN(-EINVAL);
 310         }
 311
 312         rc = obd_connect(NULL, &mdc_exp, mdc_obd, &obd->obd_uuid,
 313                          &lmv->conn_data, NULL);
 314         if (rc) {
 315                 CERROR("target %s connect error %d\n", tgt->ltd_uuid.uuid, rc);
 316                 RETURN(rc);
 317         }
 318
 319         /*
 320          * Init fid sequence client for this mdc and add new fld target.
 321          */
 322         rc = obd_fid_init(mdc_obd, mdc_exp, LUSTRE_SEQ_METADATA);
 323         if (rc)
 324                 RETURN(rc);
 325
 326         target.ft_srv = NULL;
 327         target.ft_exp = mdc_exp;
 328         target.ft_idx = tgt->ltd_idx;
 329
 330         fld_client_add_target(&lmv->lmv_fld, &target);
 331
 332         rc = obd_register_observer(mdc_obd, obd);
 333         if (rc) {
 334                 obd_disconnect(mdc_exp);
 335                 CERROR("target %s register_observer error %d\n",
 336                        tgt->ltd_uuid.uuid, rc);
 337                 RETURN(rc);
 338         }
 339
 340         if (obd->obd_observer) {
 341                 /*
 342                  * Tell the observer about the new target.
 343                  */
 344                 rc = obd_notify(obd->obd_observer, mdc_exp->exp_obd,
 345                                 OBD_NOTIFY_ACTIVE);
 346                 if (rc) {
 347                         obd_disconnect(mdc_exp);
 348                         RETURN(rc);
 349                 }
 350         }
 351
 352         tgt->ltd_active = 1;
 353         tgt->ltd_exp = mdc_exp;
 354         lmv->desc.ld_active_tgt_count++;
 355
 356         md_init_ea_size(tgt->ltd_exp, lmv->max_easize, lmv->max_def_easize);
 357
 358         CDEBUG(D_CONFIG, "Connected to %s(%s) successfully (%d)\n",
 359                 mdc_obd->obd_name, mdc_obd->obd_uuid.uuid,
 360                 atomic_read(&obd->obd_refcount));
 361
 362         if (lmv->lmv_tgts_kobj)
 363                 /* Even if we failed to create the link, that's fine */
 364                 rc = sysfs_create_link(lmv->lmv_tgts_kobj,
 365                                        &mdc_obd->obd_kset.kobj,
 366                                        mdc_obd->obd_name);
 367         RETURN(0);
 368 }
 369
 370 static void lmv_del_target(struct lmv_obd *lmv, int index)
 371 {
 372         if (lmv->tgts[index] == NULL)
 373                 return;
 374
 375         OBD_FREE_PTR(lmv->tgts[index]);
 376         lmv->tgts[index] = NULL;
 377         return;
 378 }
 379
 380 static int lmv_add_target(struct obd_device *obd, struct obd_uuid *uuidp,
 381                            __u32 index, int gen)
 382 {
 383         struct obd_device *mdc_obd;
 384         struct lmv_obd      *lmv = &obd->u.lmv;
 385         struct lmv_tgt_desc *tgt;
 386         int                  orig_tgt_count = 0;
 387         int                  rc = 0;
 388         ENTRY;
 389
 390         CDEBUG(D_CONFIG, "Target uuid: %s. index %d\n", uuidp->uuid, index);
 391         mdc_obd = class_find_client_obd(uuidp, LUSTRE_MDC_NAME,
 392                                         &obd->obd_uuid);
 393         if (!mdc_obd) {
 394                 CERROR("%s: Target %s not attached: rc = %d\n",
 395                        obd->obd_name, uuidp->uuid, -EINVAL);
 396                 RETURN(-EINVAL);
 397         }
 398
 399         mutex_lock(&lmv->lmv_init_mutex);
 400         if ((index < lmv->tgts_size) && (lmv->tgts[index] != NULL)) {
 401                 tgt = lmv->tgts[index];
 402                 CERROR("%s: UUID %s already assigned at LMV target index %d:"
 403                        " rc = %d\n", obd->obd_name,
 404                        obd_uuid2str(&tgt->ltd_uuid), index, -EEXIST);
 405                 mutex_unlock(&lmv->lmv_init_mutex);
 406                 RETURN(-EEXIST);
 407         }
 408
 409         if (index >= lmv->tgts_size) {
 410                 /* We need to reallocate the lmv target array. */
 411                 struct lmv_tgt_desc **newtgts, **old = NULL;
 412                 __u32 newsize = 1;
 413                 __u32 oldsize = 0;
 414
 415                 while (newsize < index + 1)
 416                         newsize = newsize << 1;
 417                 OBD_ALLOC(newtgts, sizeof(*newtgts) * newsize);
 418                 if (newtgts == NULL) {
 419                         mutex_unlock(&lmv->lmv_init_mutex);
 420                         RETURN(-ENOMEM);
 421                 }
 422
 423                 if (lmv->tgts_size) {
 424                         memcpy(newtgts, lmv->tgts,
 425                                sizeof(*newtgts) * lmv->tgts_size);
 426                         old = lmv->tgts;
 427                         oldsize = lmv->tgts_size;
 428                 }
 429
 430                 lmv->tgts = newtgts;
 431                 lmv->tgts_size = newsize;
 432                 smp_rmb();
 433                 if (old)
 434                         OBD_FREE(old, sizeof(*old) * oldsize);
 435
 436                 CDEBUG(D_CONFIG, "tgts: %p size: %d\n", lmv->tgts,
 437                        lmv->tgts_size);
 438         }
 439
 440         OBD_ALLOC_PTR(tgt);
 441         if (!tgt) {
 442                 mutex_unlock(&lmv->lmv_init_mutex);
 443                 RETURN(-ENOMEM);
 444         }
 445
 446         mutex_init(&tgt->ltd_fid_mutex);
 447         tgt->ltd_idx = index;
 448         tgt->ltd_uuid = *uuidp;
 449         tgt->ltd_active = 0;
 450         lmv->tgts[index] = tgt;
 451         if (index >= lmv->desc.ld_tgt_count) {
 452                 orig_tgt_count = lmv->desc.ld_tgt_count;
 453                 lmv->desc.ld_tgt_count = index + 1;
 454         }
 455
 456         if (lmv->connected == 0) {
 457                 /* lmv_check_connect() will connect this target. */
 458                 mutex_unlock(&lmv->lmv_init_mutex);
 459                 RETURN(0);
 460         }
 461
 462         /* Otherwise let's connect it ourselves */
 463         mutex_unlock(&lmv->lmv_init_mutex);
 464         rc = lmv_connect_mdc(obd, tgt);
 465         if (rc != 0) {
 466                 spin_lock(&lmv->lmv_lock);
 467                 if (lmv->desc.ld_tgt_count == index + 1)
 468                         lmv->desc.ld_tgt_count = orig_tgt_count;
 469                 memset(tgt, 0, sizeof(*tgt));
 470                 spin_unlock(&lmv->lmv_lock);
 471         } else {
 472                 int easize = sizeof(struct lmv_stripe_md) +
 473                         lmv->desc.ld_tgt_count * sizeof(struct lu_fid);
 474                 lmv_init_ea_size(obd->obd_self_export, easize, 0);
 475         }
 476
 477         RETURN(rc);
 478 }
 479
 480 static int lmv_check_connect(struct obd_device *obd)
 481 {
 482         struct lmv_obd          *lmv = &obd->u.lmv;
 483         struct lmv_tgt_desc     *tgt;
 484         __u32                    i;
 485         int                      rc;
 486         int                      easize;
 487         ENTRY;
 488
 489         if (lmv->connected)
 490                 RETURN(0);
 491
 492         mutex_lock(&lmv->lmv_init_mutex);
 493         if (lmv->connected) {
 494                 mutex_unlock(&lmv->lmv_init_mutex);
 495                 RETURN(0);
 496         }
 497
 498         if (lmv->desc.ld_tgt_count == 0) {
 499                 mutex_unlock(&lmv->lmv_init_mutex);
 500                 CERROR("%s: no targets configured.\n", obd->obd_name);
 501                 RETURN(-EINVAL);
 502         }
 503
 504         LASSERT(lmv->tgts != NULL);
 505
 506         if (lmv->tgts[0] == NULL) {
 507                 mutex_unlock(&lmv->lmv_init_mutex);
 508                 CERROR("%s: no target configured for index 0.\n",
 509                        obd->obd_name);
 510                 RETURN(-EINVAL);
 511         }
 512
 513         CDEBUG(D_CONFIG, "Time to connect %s to %s\n",
 514                obd->obd_uuid.uuid, obd->obd_name);
 515
 516         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 517                 tgt = lmv->tgts[i];
 518                 if (tgt == NULL)
 519                         continue;
 520                 rc = lmv_connect_mdc(obd, tgt);
 521                 if (rc)
 522                         GOTO(out_disc, rc);
 523         }
 524
 525         lmv->connected = 1;
 526         easize = lmv_mds_md_size(lmv->desc.ld_tgt_count, LMV_MAGIC);
 527         lmv_init_ea_size(obd->obd_self_export, easize, 0);
 528         mutex_unlock(&lmv->lmv_init_mutex);
 529         RETURN(0);
 530
 531  out_disc:
 532         while (i-- > 0) {
 533                 int rc2;
 534                 tgt = lmv->tgts[i];
 535                 if (tgt == NULL)
 536                         continue;
 537                 tgt->ltd_active = 0;
 538                 if (tgt->ltd_exp) {
 539                         --lmv->desc.ld_active_tgt_count;
 540                         rc2 = obd_disconnect(tgt->ltd_exp);
 541                         if (rc2) {
 542                                 CERROR("LMV target %s disconnect on "
 543                                        "MDC idx %d: error %d\n",
 544                                        tgt->ltd_uuid.uuid, i, rc2);
 545                         }
 546                 }
 547         }
 548
 549         mutex_unlock(&lmv->lmv_init_mutex);
 550
 551         RETURN(rc);
 552 }
 553
 554 static int lmv_disconnect_mdc(struct obd_device *obd, struct lmv_tgt_desc *tgt)
 555 {
 556         struct lmv_obd         *lmv = &obd->u.lmv;
 557         struct obd_device      *mdc_obd;
 558         int                     rc;
 559         ENTRY;
 560
 561         LASSERT(tgt != NULL);
 562         LASSERT(obd != NULL);
 563
 564         mdc_obd = class_exp2obd(tgt->ltd_exp);
 565
 566         if (mdc_obd) {
 567                 mdc_obd->obd_force = obd->obd_force;
 568                 mdc_obd->obd_fail = obd->obd_fail;
 569                 mdc_obd->obd_no_recov = obd->obd_no_recov;
 570
 571                 if (lmv->lmv_tgts_kobj)
 572                         sysfs_remove_link(lmv->lmv_tgts_kobj,
 573                                           mdc_obd->obd_name);
 574         }
 575
 576         rc = obd_fid_fini(tgt->ltd_exp->exp_obd);
 577         if (rc)
 578                 CERROR("Can't finanize fids factory\n");
 579
 580         CDEBUG(D_INFO, "Disconnected from %s(%s) successfully\n",
 581                tgt->ltd_exp->exp_obd->obd_name,
 582                tgt->ltd_exp->exp_obd->obd_uuid.uuid);
 583
 584         obd_register_observer(tgt->ltd_exp->exp_obd, NULL);
 585         rc = obd_disconnect(tgt->ltd_exp);
 586         if (rc) {
 587                 if (tgt->ltd_active) {
 588                         CERROR("Target %s disconnect error %d\n",
 589                                tgt->ltd_uuid.uuid, rc);
 590                 }
 591         }
 592
 593         lmv_activate_target(lmv, tgt, 0);
 594         tgt->ltd_exp = NULL;
 595         RETURN(0);
 596 }
 597
 598 static int lmv_disconnect(struct obd_export *exp)
 599 {
 600         struct obd_device       *obd = class_exp2obd(exp);
 601         struct lmv_obd          *lmv = &obd->u.lmv;
 602         int                      rc;
 603         __u32                    i;
 604         ENTRY;
 605
 606         if (!lmv->tgts)
 607                 goto out_local;
 608
 609         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 610                 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
 611                         continue;
 612
 613                 lmv_disconnect_mdc(obd, lmv->tgts[i]);
 614         }
 615
 616         if (lmv->lmv_tgts_kobj)
 617                 kobject_put(lmv->lmv_tgts_kobj);
 618
 619 out_local:
 620         /*
 621          * This is the case when no real connection is established by
 622          * lmv_check_connect().
 623          */
 624         if (!lmv->connected)
 625                 class_export_put(exp);
 626         rc = class_disconnect(exp);
 627         lmv->connected = 0;
 628
 629         RETURN(rc);
 630 }
 631
 632 static int lmv_fid2path(struct obd_export *exp, int len, void *karg,
 633                         void __user *uarg)
 634 {
 635         struct obd_device       *obddev = class_exp2obd(exp);
 636         struct lmv_obd          *lmv = &obddev->u.lmv;
 637         struct getinfo_fid2path *gf;
 638         struct lmv_tgt_desc     *tgt;
 639         struct getinfo_fid2path *remote_gf = NULL;
 640         struct lu_fid           root_fid;
 641         int                     remote_gf_size = 0;
 642         int                     rc;
 643
 644         gf = karg;
 645         tgt = lmv_find_target(lmv, &gf->gf_fid);
 646         if (IS_ERR(tgt))
 647                 RETURN(PTR_ERR(tgt));
 648
 649         root_fid = *gf->gf_u.gf_root_fid;
 650         LASSERT(fid_is_sane(&root_fid));
 651
 652 repeat_fid2path:
 653         rc = obd_iocontrol(OBD_IOC_FID2PATH, tgt->ltd_exp, len, gf, uarg);
 654         if (rc != 0 && rc != -EREMOTE)
 655                 GOTO(out_fid2path, rc);
 656
 657         /* If remote_gf != NULL, it means just building the
 658          * path on the remote MDT, copy this path segement to gf */
 659         if (remote_gf != NULL) {
 660                 struct getinfo_fid2path *ori_gf;
 661                 char *ptr;
 662                 int len;
 663
 664                 ori_gf = (struct getinfo_fid2path *)karg;
 665                 if (strlen(ori_gf->gf_u.gf_path) + 1 +
 666                     strlen(gf->gf_u.gf_path) + 1 > ori_gf->gf_pathlen)
 667                         GOTO(out_fid2path, rc = -EOVERFLOW);
 668
 669                 ptr = ori_gf->gf_u.gf_path;
 670
 671                 len = strlen(gf->gf_u.gf_path);
 672                 /* move the current path to the right to release space
 673                  * for closer-to-root part */
 674                 memmove(ptr + len + 1, ptr, strlen(ori_gf->gf_u.gf_path));
 675                 memcpy(ptr, gf->gf_u.gf_path, len);
 676                 ptr[len] = '/';
 677         }
 678
 679         CDEBUG(D_INFO, "%s: get path %s "DFID" rec: %llu ln: %u\n",
 680                tgt->ltd_exp->exp_obd->obd_name,
 681                gf->gf_u.gf_path, PFID(&gf->gf_fid), gf->gf_recno,
 682                gf->gf_linkno);
 683
 684         if (rc == 0)
 685                 GOTO(out_fid2path, rc);
 686
 687         /* sigh, has to go to another MDT to do path building further */
 688         if (remote_gf == NULL) {
 689                 remote_gf_size = sizeof(*remote_gf) + PATH_MAX;
 690                 OBD_ALLOC(remote_gf, remote_gf_size);
 691                 if (remote_gf == NULL)
 692                         GOTO(out_fid2path, rc = -ENOMEM);
 693                 remote_gf->gf_pathlen = PATH_MAX;
 694         }
 695
 696         if (!fid_is_sane(&gf->gf_fid)) {
 697                 CERROR("%s: invalid FID "DFID": rc = %d\n",
 698                        tgt->ltd_exp->exp_obd->obd_name,
 699                        PFID(&gf->gf_fid), -EINVAL);
 700                 GOTO(out_fid2path, rc = -EINVAL);
 701         }
 702
 703         tgt = lmv_find_target(lmv, &gf->gf_fid);
 704         if (IS_ERR(tgt))
 705                 GOTO(out_fid2path, rc = -EINVAL);
 706
 707         remote_gf->gf_fid = gf->gf_fid;
 708         remote_gf->gf_recno = -1;
 709         remote_gf->gf_linkno = -1;
 710         memset(remote_gf->gf_u.gf_path, 0, remote_gf->gf_pathlen);
 711         *remote_gf->gf_u.gf_root_fid = root_fid;
 712         gf = remote_gf;
 713         goto repeat_fid2path;
 714
 715 out_fid2path:
 716         if (remote_gf != NULL)
 717                 OBD_FREE(remote_gf, remote_gf_size);
 718         RETURN(rc);
 719 }
 720
 721 static int lmv_hsm_req_count(struct lmv_obd *lmv,
 722                              const struct hsm_user_request *hur,
 723                              const struct lmv_tgt_desc *tgt_mds)
 724 {
 725         __u32                    i;
 726         int                      nr = 0;
 727         struct lmv_tgt_desc     *curr_tgt;
 728
 729         /* count how many requests must be sent to the given target */
 730         for (i = 0; i < hur->hur_request.hr_itemcount; i++) {
 731                 curr_tgt = lmv_find_target(lmv, &hur->hur_user_item[i].hui_fid);
 732                 if (IS_ERR(curr_tgt))
 733                         RETURN(PTR_ERR(curr_tgt));
 734                 if (obd_uuid_equals(&curr_tgt->ltd_uuid, &tgt_mds->ltd_uuid))
 735                         nr++;
 736         }
 737         return nr;
 738 }
 739
 740 static int lmv_hsm_req_build(struct lmv_obd *lmv,
 741                               struct hsm_user_request *hur_in,
 742                               const struct lmv_tgt_desc *tgt_mds,
 743                               struct hsm_user_request *hur_out)
 744 {
 745         __u32                    i, nr_out;
 746         struct lmv_tgt_desc     *curr_tgt;
 747
 748         /* build the hsm_user_request for the given target */
 749         hur_out->hur_request = hur_in->hur_request;
 750         nr_out = 0;
 751         for (i = 0; i < hur_in->hur_request.hr_itemcount; i++) {
 752                 curr_tgt = lmv_find_target(lmv,
 753                                            &hur_in->hur_user_item[i].hui_fid);
 754                 if (IS_ERR(curr_tgt))
 755                         RETURN(PTR_ERR(curr_tgt));
 756                 if (obd_uuid_equals(&curr_tgt->ltd_uuid, &tgt_mds->ltd_uuid)) {
 757                         hur_out->hur_user_item[nr_out] =
 758                                                 hur_in->hur_user_item[i];
 759                         nr_out++;
 760                 }
 761         }
 762         hur_out->hur_request.hr_itemcount = nr_out;
 763         memcpy(hur_data(hur_out), hur_data(hur_in),
 764                hur_in->hur_request.hr_data_len);
 765
 766         RETURN(0);
 767 }
 768
 769 static int lmv_hsm_ct_unregister(struct obd_device *obd, unsigned int cmd,
 770                                  int len, struct lustre_kernelcomm *lk,
 771                                  void __user *uarg)
 772 {
 773         struct lmv_obd *lmv = &obd->u.lmv;
 774         __u32   i;
 775         int     rc;
 776         ENTRY;
 777
 778         /* unregister request (call from llapi_hsm_copytool_fini) */
 779         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 780                 struct lmv_tgt_desc *tgt = lmv->tgts[i];
 781
 782                 if (tgt == NULL || tgt->ltd_exp == NULL)
 783                         continue;
 784                 /* best effort: try to clean as much as possible
 785                  * (continue on error) */
 786                 obd_iocontrol(cmd, tgt->ltd_exp, len, lk, uarg);
 787         }
 788
 789         /* Whatever the result, remove copytool from kuc groups.
 790          * Unreached coordinators will get EPIPE on next requests
 791          * and will unregister automatically.
 792          */
 793         rc = libcfs_kkuc_group_rem(&obd->obd_uuid, lk->lk_uid, lk->lk_group);
 794
 795         RETURN(rc);
 796 }
 797
 798 static int lmv_hsm_ct_register(struct obd_device *obd, unsigned int cmd,
 799                                int len, struct lustre_kernelcomm *lk,
 800                                void __user *uarg)
 801 {
 802         struct lmv_obd *lmv = &obd->u.lmv;
 803         struct file *filp;
 804         __u32 i, j;
 805         int err;
 806         bool any_set = false;
 807         struct kkuc_ct_data *kcd;
 808         size_t kcd_size;
 809         int rc = 0;
 810         ENTRY;
 811
 812         filp = fget(lk->lk_wfd);
 813         if (!filp)
 814                 RETURN(-EBADF);
 815
 816         if (lk->lk_flags & LK_FLG_DATANR)
 817                 kcd_size = offsetof(struct kkuc_ct_data,
 818                                     kcd_archives[lk->lk_data_count]);
 819         else
 820                 kcd_size = sizeof(*kcd);
 821
 822         OBD_ALLOC(kcd, kcd_size);
 823         if (kcd == NULL)
 824                 GOTO(err_fput, rc = -ENOMEM);
 825
 826         kcd->kcd_nr_archives = lk->lk_data_count;
 827         if (lk->lk_flags & LK_FLG_DATANR) {
 828                 kcd->kcd_magic = KKUC_CT_DATA_ARRAY_MAGIC;
 829                 if (lk->lk_data_count > 0)
 830                         memcpy(kcd->kcd_archives, lk->lk_data,
 831                                sizeof(*kcd->kcd_archives) * lk->lk_data_count);
 832         } else {
 833                 kcd->kcd_magic = KKUC_CT_DATA_BITMAP_MAGIC;
 834         }
 835
 836         rc = libcfs_kkuc_group_add(filp, &obd->obd_uuid, lk->lk_uid,
 837                                    lk->lk_group, kcd, kcd_size);
 838         OBD_FREE(kcd, kcd_size);
 839         if (rc)
 840                 GOTO(err_fput, rc);
 841
 842         /* All or nothing: try to register to all MDS.
 843          * In case of failure, unregister from previous MDS,
 844          * except if it because of inactive target. */
 845         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 846                 struct lmv_tgt_desc *tgt = lmv->tgts[i];
 847
 848                 if (tgt == NULL || tgt->ltd_exp == NULL)
 849                         continue;
 850
 851                 err = obd_iocontrol(cmd, tgt->ltd_exp, len, lk, uarg);
 852                 if (err) {
 853                         if (tgt->ltd_active) {
 854                                 /* permanent error */
 855                                 CERROR("%s: iocontrol MDC %s on MDT"
 856                                        " idx %d cmd %x: err = %d\n",
 857                                        lmv2obd_dev(lmv)->obd_name,
 858                                        tgt->ltd_uuid.uuid, i, cmd, err);
 859                                 rc = err;
 860                                 lk->lk_flags |= LK_FLG_STOP;
 861                                 /* unregister from previous MDS */
 862                                 for (j = 0; j < i; j++) {
 863                                         tgt = lmv->tgts[j];
 864                                         if (tgt == NULL || tgt->ltd_exp == NULL)
 865                                                 continue;
 866                                         obd_iocontrol(cmd, tgt->ltd_exp, len,
 867                                                       lk, uarg);
 868                                 }
 869                                 GOTO(err_kkuc_rem, rc);
 870                         }
 871                         /* else: transient error.
 872                          * kuc will register to the missing MDT
 873                          * when it is back */
 874                 } else {
 875                         any_set = true;
 876                 }
 877         }
 878
 879         if (!any_set)
 880                 /* no registration done: return error */
 881                 GOTO(err_kkuc_rem, rc = -ENOTCONN);
 882
 883         RETURN(0);
 884
 885 err_kkuc_rem:
 886         libcfs_kkuc_group_rem(&obd->obd_uuid, lk->lk_uid, lk->lk_group);
 887
 888 err_fput:
 889         fput(filp);
 890         return rc;
 891 }
 892
 893
 894
 895
 896 static int lmv_iocontrol(unsigned int cmd, struct obd_export *exp,
 897                          int len, void *karg, void __user *uarg)
 898 {
 899         struct obd_device       *obddev = class_exp2obd(exp);
 900         struct lmv_obd          *lmv = &obddev->u.lmv;
 901         struct lmv_tgt_desc     *tgt = NULL;
 902         __u32                    i = 0;
 903         int                      rc = 0;
 904         int                      set = 0;
 905         __u32                    count = lmv->desc.ld_tgt_count;
 906         ENTRY;
 907
 908         if (count == 0)
 909                 RETURN(-ENOTTY);
 910
 911         switch (cmd) {
 912         case IOC_OBD_STATFS: {
 913                 struct obd_ioctl_data *data = karg;
 914                 struct obd_device *mdc_obd;
 915                 struct obd_statfs stat_buf = {0};
 916                 __u32 index;
 917
 918                 memcpy(&index, data->ioc_inlbuf2, sizeof(__u32));
 919                 if ((index >= count))
 920                         RETURN(-ENODEV);
 921
 922                 tgt = lmv->tgts[index];
 923                 if (tgt == NULL || !tgt->ltd_active)
 924                         RETURN(-ENODATA);
 925
 926                 mdc_obd = class_exp2obd(tgt->ltd_exp);
 927                 if (!mdc_obd)
 928                         RETURN(-EINVAL);
 929
 930                 /* copy UUID */
 931                 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(mdc_obd),
 932                                  min((int) data->ioc_plen2,
 933                                      (int) sizeof(struct obd_uuid))))
 934                         RETURN(-EFAULT);
 935
 936                 rc = obd_statfs(NULL, tgt->ltd_exp, &stat_buf,
 937                                 ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS,
 938                                 0);
 939                 if (rc)
 940                         RETURN(rc);
 941                 if (copy_to_user(data->ioc_pbuf1, &stat_buf,
 942                                  min((int) data->ioc_plen1,
 943                                      (int) sizeof(stat_buf))))
 944                         RETURN(-EFAULT);
 945                 break;
 946         }
 947         case OBD_IOC_QUOTACTL: {
 948                 struct if_quotactl *qctl = karg;
 949                 struct obd_quotactl *oqctl;
 950
 951                 if (qctl->qc_valid == QC_MDTIDX) {
 952                         if (count <= qctl->qc_idx)
 953                                 RETURN(-EINVAL);
 954
 955                         tgt = lmv->tgts[qctl->qc_idx];
 956                         if (tgt == NULL || tgt->ltd_exp == NULL)
 957                                 RETURN(-EINVAL);
 958                 } else if (qctl->qc_valid == QC_UUID) {
 959                         for (i = 0; i < count; i++) {
 960                                 tgt = lmv->tgts[i];
 961                                 if (tgt == NULL)
 962                                         continue;
 963                                 if (!obd_uuid_equals(&tgt->ltd_uuid,
 964                                                      &qctl->obd_uuid))
 965                                         continue;
 966
 967                                 if (tgt->ltd_exp == NULL)
 968                                         RETURN(-EINVAL);
 969
 970                                 break;
 971                         }
 972                 } else {
 973                         RETURN(-EINVAL);
 974                 }
 975
 976                 if (i >= count)
 977                         RETURN(-EAGAIN);
 978
 979                 LASSERT(tgt != NULL && tgt->ltd_exp != NULL);
 980                 OBD_ALLOC_PTR(oqctl);
 981                 if (!oqctl)
 982                         RETURN(-ENOMEM);
 983
 984                 QCTL_COPY(oqctl, qctl);
 985                 rc = obd_quotactl(tgt->ltd_exp, oqctl);
 986                 if (rc == 0) {
 987                         QCTL_COPY(qctl, oqctl);
 988                         qctl->qc_valid = QC_MDTIDX;
 989                         qctl->obd_uuid = tgt->ltd_uuid;
 990                 }
 991                 OBD_FREE_PTR(oqctl);
 992                 break;
 993         }
 994         case LL_IOC_GET_CONNECT_FLAGS: {
 995                 tgt = lmv->tgts[0];
 996                 if (tgt == NULL || tgt->ltd_exp == NULL)
 997                         RETURN(-ENODATA);
 998                 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
 999                 break;
1000         }
1001         case LL_IOC_FID2MDTIDX: {
1002                 struct lu_fid *fid = karg;
1003                 int             mdt_index;
1004
1005                 rc = lmv_fld_lookup(lmv, fid, &mdt_index);
1006                 if (rc != 0)
1007                         RETURN(rc);
1008
1009                 /* Note: this is from llite(see ll_dir_ioctl()), @uarg does not
1010                  * point to user space memory for FID2MDTIDX. */
1011                 *(__u32 *)uarg = mdt_index;
1012                 break;
1013         }
1014         case OBD_IOC_FID2PATH: {
1015                 rc = lmv_fid2path(exp, len, karg, uarg);
1016                 break;
1017         }
1018         case LL_IOC_HSM_STATE_GET:
1019         case LL_IOC_HSM_STATE_SET:
1020         case LL_IOC_HSM_ACTION: {
1021                 struct md_op_data       *op_data = karg;
1022
1023                 tgt = lmv_find_target(lmv, &op_data->op_fid1);
1024                 if (IS_ERR(tgt))
1025                         RETURN(PTR_ERR(tgt));
1026
1027                 if (tgt->ltd_exp == NULL)
1028                         RETURN(-EINVAL);
1029
1030                 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1031                 break;
1032         }
1033         case LL_IOC_HSM_PROGRESS: {
1034                 const struct hsm_progress_kernel *hpk = karg;
1035
1036                 tgt = lmv_find_target(lmv, &hpk->hpk_fid);
1037                 if (IS_ERR(tgt))
1038                         RETURN(PTR_ERR(tgt));
1039                 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1040                 break;
1041         }
1042         case LL_IOC_HSM_REQUEST: {
1043                 struct hsm_user_request *hur = karg;
1044                 unsigned int reqcount = hur->hur_request.hr_itemcount;
1045
1046                 if (reqcount == 0)
1047                         RETURN(0);
1048
1049                 /* if the request is about a single fid
1050                  * or if there is a single MDS, no need to split
1051                  * the request. */
1052                 if (reqcount == 1 || count == 1) {
1053                         tgt = lmv_find_target(lmv,
1054                                               &hur->hur_user_item[0].hui_fid);
1055                         if (IS_ERR(tgt))
1056                                 RETURN(PTR_ERR(tgt));
1057                         rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1058                 } else {
1059                         /* split fid list to their respective MDS */
1060                         for (i = 0; i < count; i++) {
1061                                 int nr, rc1;
1062                                 size_t reqlen;
1063                                 struct hsm_user_request *req;
1064
1065                                 tgt = lmv->tgts[i];
1066                                 if (tgt == NULL || tgt->ltd_exp == NULL)
1067                                         continue;
1068
1069                                 nr = lmv_hsm_req_count(lmv, hur, tgt);
1070                                 if (nr < 0)
1071                                         RETURN(nr);
1072                                 if (nr == 0) /* nothing for this MDS */
1073                                         continue;
1074
1075                                 /* build a request with fids for this MDS */
1076                                 reqlen = offsetof(typeof(*hur),
1077                                                   hur_user_item[nr])
1078                                                 + hur->hur_request.hr_data_len;
1079                                 OBD_ALLOC_LARGE(req, reqlen);
1080                                 if (req == NULL)
1081                                         RETURN(-ENOMEM);
1082                                 rc1 = lmv_hsm_req_build(lmv, hur, tgt, req);
1083                                 if (rc1 < 0)
1084                                         GOTO(hsm_req_err, rc1);
1085                                 rc1 = obd_iocontrol(cmd, tgt->ltd_exp, reqlen,
1086                                                     req, uarg);
1087 hsm_req_err:
1088                                 if (rc1 != 0 && rc == 0)
1089                                         rc = rc1;
1090                                 OBD_FREE_LARGE(req, reqlen);
1091                         }
1092                 }
1093                 break;
1094         }
1095         case LL_IOC_LOV_SWAP_LAYOUTS: {
1096                 struct md_op_data       *op_data = karg;
1097                 struct lmv_tgt_desc     *tgt1, *tgt2;
1098
1099                 tgt1 = lmv_find_target(lmv, &op_data->op_fid1);
1100                 if (IS_ERR(tgt1))
1101                         RETURN(PTR_ERR(tgt1));
1102
1103                 tgt2 = lmv_find_target(lmv, &op_data->op_fid2);
1104                 if (IS_ERR(tgt2))
1105                         RETURN(PTR_ERR(tgt2));
1106
1107                 if ((tgt1->ltd_exp == NULL) || (tgt2->ltd_exp == NULL))
1108                         RETURN(-EINVAL);
1109
1110                 /* only files on same MDT can have their layouts swapped */
1111                 if (tgt1->ltd_idx != tgt2->ltd_idx)
1112                         RETURN(-EPERM);
1113
1114                 rc = obd_iocontrol(cmd, tgt1->ltd_exp, len, karg, uarg);
1115                 break;
1116         }
1117         case LL_IOC_HSM_CT_START: {
1118                 struct lustre_kernelcomm *lk = karg;
1119                 if (lk->lk_flags & LK_FLG_STOP)
1120                         rc = lmv_hsm_ct_unregister(obddev, cmd, len, lk, uarg);
1121                 else
1122                         rc = lmv_hsm_ct_register(obddev, cmd, len, lk, uarg);
1123                 break;
1124         }
1125         default:
1126                 for (i = 0; i < count; i++) {
1127                         struct obd_device *mdc_obd;
1128                         int err;
1129
1130                         tgt = lmv->tgts[i];
1131                         if (tgt == NULL || tgt->ltd_exp == NULL)
1132                                 continue;
1133                         /* ll_umount_begin() sets force flag but for lmv, not
1134                          * mdc. Let's pass it through */
1135                         mdc_obd = class_exp2obd(tgt->ltd_exp);
1136                         mdc_obd->obd_force = obddev->obd_force;
1137                         err = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1138                         if (err) {
1139                                 if (tgt->ltd_active) {
1140                                         CERROR("error: iocontrol MDC %s on MDT"
1141                                                " idx %d cmd %x: err = %d\n",
1142                                                tgt->ltd_uuid.uuid, i, cmd, err);
1143                                         if (!rc)
1144                                                 rc = err;
1145                                 }
1146                         } else
1147                                 set = 1;
1148                 }
1149                 if (!set && !rc)
1150                         rc = -EIO;
1151         }
1152         RETURN(rc);
1153 }
1154
1155 /**
1156  * This is _inode_ placement policy function (not name).
1157  */
1158 static int lmv_placement_policy(struct obd_device *obd,
1159                                 struct md_op_data *op_data, u32 *mds)
1160 {
1161         struct lmv_obd     *lmv = &obd->u.lmv;
1162         struct lmv_user_md *lum;
1163
1164         ENTRY;
1165
1166         LASSERT(mds != NULL);
1167
1168         if (lmv->desc.ld_tgt_count == 1) {
1169                 *mds = 0;
1170                 RETURN(0);
1171         }
1172
1173         lum = op_data->op_data;
1174         /* Choose MDS by
1175          * 1. See if the stripe offset is specified by lum.
1176          * 2. Then check if there is default stripe offset.
1177          * 3. Finally choose MDS by name hash if the parent
1178          *    is striped directory. (see lmv_locate_tgt()). */
1179         if (op_data->op_cli_flags & CLI_SET_MEA && lum != NULL &&
1180             le32_to_cpu(lum->lum_stripe_offset) != (__u32)-1) {
1181                 *mds = le32_to_cpu(lum->lum_stripe_offset);
1182         } else if (op_data->op_default_stripe_offset != (__u32)-1) {
1183                 *mds = op_data->op_default_stripe_offset;
1184                 op_data->op_mds = *mds;
1185                 /* Correct the stripe offset in lum */
1186                 if (lum != NULL)
1187                         lum->lum_stripe_offset = cpu_to_le32(*mds);
1188         } else {
1189                 *mds = op_data->op_mds;
1190         }
1191
1192         RETURN(0);
1193 }
1194
1195 int __lmv_fid_alloc(struct lmv_obd *lmv, struct lu_fid *fid, u32 mds)
1196 {
1197         struct lmv_tgt_desc     *tgt;
1198         int                      rc;
1199         ENTRY;
1200
1201         tgt = lmv_get_target(lmv, mds, NULL);
1202         if (IS_ERR(tgt))
1203                 RETURN(PTR_ERR(tgt));
1204
1205         /*
1206          * New seq alloc and FLD setup should be atomic. Otherwise we may find
1207          * on server that seq in new allocated fid is not yet known.
1208          */
1209         mutex_lock(&tgt->ltd_fid_mutex);
1210
1211         if (tgt->ltd_active == 0 || tgt->ltd_exp == NULL)
1212                 GOTO(out, rc = -ENODEV);
1213
1214         /*
1215          * Asking underlying tgt layer to allocate new fid.
1216          */
1217         rc = obd_fid_alloc(NULL, tgt->ltd_exp, fid, NULL);
1218         if (rc > 0) {
1219                 LASSERT(fid_is_sane(fid));
1220                 rc = 0;
1221         }
1222
1223         EXIT;
1224 out:
1225         mutex_unlock(&tgt->ltd_fid_mutex);
1226         return rc;
1227 }
1228
1229 int lmv_fid_alloc(const struct lu_env *env, struct obd_export *exp,
1230                   struct lu_fid *fid, struct md_op_data *op_data)
1231 {
1232         struct obd_device     *obd = class_exp2obd(exp);
1233         struct lmv_obd        *lmv = &obd->u.lmv;
1234         u32                    mds = 0;
1235         int                    rc;
1236         ENTRY;
1237
1238         LASSERT(op_data != NULL);
1239         LASSERT(fid != NULL);
1240
1241         rc = lmv_placement_policy(obd, op_data, &mds);
1242         if (rc) {
1243                 CERROR("Can't get target for allocating fid, "
1244                        "rc %d\n", rc);
1245                 RETURN(rc);
1246         }
1247
1248         rc = __lmv_fid_alloc(lmv, fid, mds);
1249         if (rc) {
1250                 CERROR("Can't alloc new fid, rc %d\n", rc);
1251                 RETURN(rc);
1252         }
1253
1254         RETURN(rc);
1255 }
1256
1257 static int lmv_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
1258 {
1259         struct lmv_obd  *lmv = &obd->u.lmv;
1260         struct lmv_desc *desc;
1261         int             rc;
1262         ENTRY;
1263
1264         if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
1265                 CERROR("LMV setup requires a descriptor\n");
1266                 RETURN(-EINVAL);
1267         }
1268
1269         desc = (struct lmv_desc *)lustre_cfg_buf(lcfg, 1);
1270         if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
1271                 CERROR("Lmv descriptor size wrong: %d > %d\n",
1272                        (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
1273                 RETURN(-EINVAL);
1274         }
1275
1276         lmv->tgts_size = 32U;
1277         OBD_ALLOC(lmv->tgts, sizeof(*lmv->tgts) * lmv->tgts_size);
1278         if (lmv->tgts == NULL)
1279                 RETURN(-ENOMEM);
1280
1281         obd_str2uuid(&lmv->desc.ld_uuid, desc->ld_uuid.uuid);
1282         lmv->desc.ld_tgt_count = 0;
1283         lmv->desc.ld_active_tgt_count = 0;
1284         lmv->max_def_easize = 0;
1285         lmv->max_easize = 0;
1286
1287         spin_lock_init(&lmv->lmv_lock);
1288         mutex_init(&lmv->lmv_init_mutex);
1289
1290         rc = lmv_tunables_init(obd);
1291         if (rc)
1292                 CWARN("%s: error adding LMV sysfs/debugfs files: rc = %d\n",
1293                       obd->obd_name, rc);
1294
1295         rc = fld_client_init(&lmv->lmv_fld, obd->obd_name,
1296                              LUSTRE_CLI_FLD_HASH_DHT);
1297         if (rc) {
1298                 CERROR("Can't init FLD, err %d\n", rc);
1299                 GOTO(out, rc);
1300         }
1301
1302         RETURN(0);
1303
1304 out:
1305         return rc;
1306 }
1307
1308 static int lmv_cleanup(struct obd_device *obd)
1309 {
1310         struct lmv_obd   *lmv = &obd->u.lmv;
1311         ENTRY;
1312
1313         fld_client_fini(&lmv->lmv_fld);
1314         if (lmv->tgts != NULL) {
1315                 int i;
1316                 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1317                         if (lmv->tgts[i] == NULL)
1318                                 continue;
1319                         lmv_del_target(lmv, i);
1320                 }
1321                 OBD_FREE(lmv->tgts, sizeof(*lmv->tgts) * lmv->tgts_size);
1322                 lmv->tgts_size = 0;
1323         }
1324         RETURN(0);
1325 }
1326
1327 static int lmv_process_config(struct obd_device *obd, size_t len, void *buf)
1328 {
1329         struct lustre_cfg       *lcfg = buf;
1330         struct obd_uuid         obd_uuid;
1331         int                     gen;
1332         __u32                   index;
1333         int                     rc;
1334         ENTRY;
1335
1336         switch (lcfg->lcfg_command) {
1337         case LCFG_ADD_MDC:
1338                 /* modify_mdc_tgts add 0:lustre-clilmv  1:lustre-MDT0000_UUID
1339                  * 2:0  3:1  4:lustre-MDT0000-mdc_UUID */
1340                 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(obd_uuid.uuid))
1341                         GOTO(out, rc = -EINVAL);
1342
1343                 obd_str2uuid(&obd_uuid,  lustre_cfg_buf(lcfg, 1));
1344
1345                 if (sscanf(lustre_cfg_buf(lcfg, 2), "%u", &index) != 1)
1346                         GOTO(out, rc = -EINVAL);
1347                 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", &gen) != 1)
1348                         GOTO(out, rc = -EINVAL);
1349                 rc = lmv_add_target(obd, &obd_uuid, index, gen);
1350                 GOTO(out, rc);
1351         default:
1352                 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
1353                 GOTO(out, rc = -EINVAL);
1354         }
1355 out:
1356         RETURN(rc);
1357 }
1358
1359 static int lmv_select_statfs_mdt(struct lmv_obd *lmv, __u32 flags)
1360 {
1361         int i;
1362
1363         if (flags & OBD_STATFS_FOR_MDT0)
1364                 return 0;
1365
1366         if (lmv->lmv_statfs_start || lmv->desc.ld_tgt_count == 1)
1367                 return lmv->lmv_statfs_start;
1368
1369         /* choose initial MDT for this client */
1370         for (i = 0;; i++) {
1371                 struct lnet_process_id lnet_id;
1372                 if (LNetGetId(i, &lnet_id) == -ENOENT)
1373                         break;
1374
1375                 if (lnet_id.nid != LNET_NID_LO_0) {
1376                         /* We dont need a full 64-bit modulus, just enough
1377                          * to distribute the requests across MDTs evenly.
1378                          */
1379                         lmv->lmv_statfs_start =
1380                                 (u32)lnet_id.nid % lmv->desc.ld_tgt_count;
1381                         break;
1382                 }
1383         }
1384
1385         return lmv->lmv_statfs_start;
1386 }
1387
1388 static int lmv_statfs(const struct lu_env *env, struct obd_export *exp,
1389                       struct obd_statfs *osfs, time64_t max_age, __u32 flags)
1390 {
1391         struct obd_device       *obd = class_exp2obd(exp);
1392         struct lmv_obd          *lmv = &obd->u.lmv;
1393         struct obd_statfs       *temp;
1394         int                      rc = 0;
1395         __u32                    i, idx;
1396         ENTRY;
1397
1398         OBD_ALLOC(temp, sizeof(*temp));
1399         if (temp == NULL)
1400                 RETURN(-ENOMEM);
1401
1402         /* distribute statfs among MDTs */
1403         idx = lmv_select_statfs_mdt(lmv, flags);
1404
1405         for (i = 0; i < lmv->desc.ld_tgt_count; i++, idx++) {
1406                 idx = idx % lmv->desc.ld_tgt_count;
1407                 if (lmv->tgts[idx] == NULL || lmv->tgts[idx]->ltd_exp == NULL)
1408                         continue;
1409
1410                 rc = obd_statfs(env, lmv->tgts[idx]->ltd_exp, temp,
1411                                 max_age, flags);
1412                 if (rc) {
1413                         CERROR("%s: can't stat MDS #%d: rc = %d\n",
1414                                lmv->tgts[idx]->ltd_exp->exp_obd->obd_name, i,
1415                                rc);
1416                         GOTO(out_free_temp, rc);
1417                 }
1418
1419                 if (temp->os_state & OS_STATE_SUM ||
1420                     flags == OBD_STATFS_FOR_MDT0) {
1421                         /* reset to the last aggregated values
1422                          * and don't sum with non-aggrated data */
1423                         /* If the statfs is from mount, it needs to retrieve
1424                          * necessary information from MDT0. i.e. mount does
1425                          * not need the merged osfs from all of MDT. Also
1426                          * clients can be mounted as long as MDT0 is in
1427                          * service */
1428                         *osfs = *temp;
1429                         break;
1430                 }
1431
1432                 if (i == 0) {
1433                         *osfs = *temp;
1434                 } else {
1435                         osfs->os_bavail += temp->os_bavail;
1436                         osfs->os_blocks += temp->os_blocks;
1437                         osfs->os_ffree += temp->os_ffree;
1438                         osfs->os_files += temp->os_files;
1439                         osfs->os_granted += temp->os_granted;
1440                 }
1441         }
1442
1443         EXIT;
1444 out_free_temp:
1445         OBD_FREE(temp, sizeof(*temp));
1446         return rc;
1447 }
1448
1449 static int lmv_get_root(struct obd_export *exp, const char *fileset,
1450                         struct lu_fid *fid)
1451 {
1452         struct obd_device    *obd = exp->exp_obd;
1453         struct lmv_obd       *lmv = &obd->u.lmv;
1454         int                   rc;
1455         ENTRY;
1456
1457         rc = md_get_root(lmv->tgts[0]->ltd_exp, fileset, fid);
1458         RETURN(rc);
1459 }
1460
1461 static int lmv_getxattr(struct obd_export *exp, const struct lu_fid *fid,
1462                         u64 obd_md_valid, const char *name, size_t buf_size,
1463                         struct ptlrpc_request **req)
1464 {
1465         struct obd_device      *obd = exp->exp_obd;
1466         struct lmv_obd         *lmv = &obd->u.lmv;
1467         struct lmv_tgt_desc    *tgt;
1468         int                     rc;
1469         ENTRY;
1470
1471         tgt = lmv_find_target(lmv, fid);
1472         if (IS_ERR(tgt))
1473                 RETURN(PTR_ERR(tgt));
1474
1475         rc = md_getxattr(tgt->ltd_exp, fid, obd_md_valid, name, buf_size, req);
1476
1477         RETURN(rc);
1478 }
1479
1480 static int lmv_setxattr(struct obd_export *exp, const struct lu_fid *fid,
1481                         u64 obd_md_valid, const char *name,
1482                         const void *value, size_t value_size,
1483                         unsigned int xattr_flags, u32 suppgid,
1484                         struct ptlrpc_request **req)
1485 {
1486         struct obd_device      *obd = exp->exp_obd;
1487         struct lmv_obd         *lmv = &obd->u.lmv;
1488         struct lmv_tgt_desc    *tgt;
1489         int                     rc;
1490         ENTRY;
1491
1492         tgt = lmv_find_target(lmv, fid);
1493         if (IS_ERR(tgt))
1494                 RETURN(PTR_ERR(tgt));
1495
1496         rc = md_setxattr(tgt->ltd_exp, fid, obd_md_valid, name,
1497                          value, value_size, xattr_flags, suppgid, req);
1498
1499         RETURN(rc);
1500 }
1501
1502 static int lmv_getattr(struct obd_export *exp, struct md_op_data *op_data,
1503                        struct ptlrpc_request **request)
1504 {
1505         struct obd_device       *obd = exp->exp_obd;
1506         struct lmv_obd          *lmv = &obd->u.lmv;
1507         struct lmv_tgt_desc     *tgt;
1508         int                      rc;
1509         ENTRY;
1510
1511         tgt = lmv_find_target(lmv, &op_data->op_fid1);
1512         if (IS_ERR(tgt))
1513                 RETURN(PTR_ERR(tgt));
1514
1515         if (op_data->op_flags & MF_GET_MDT_IDX) {
1516                 op_data->op_mds = tgt->ltd_idx;
1517                 RETURN(0);
1518         }
1519
1520         rc = md_getattr(tgt->ltd_exp, op_data, request);
1521
1522         RETURN(rc);
1523 }
1524
1525 static int lmv_null_inode(struct obd_export *exp, const struct lu_fid *fid)
1526 {
1527         struct obd_device   *obd = exp->exp_obd;
1528         struct lmv_obd      *lmv = &obd->u.lmv;
1529         __u32                i;
1530         ENTRY;
1531
1532         CDEBUG(D_INODE, "CBDATA for "DFID"\n", PFID(fid));
1533
1534         /*
1535          * With DNE every object can have two locks in different namespaces:
1536          * lookup lock in space of MDT storing direntry and update/open lock in
1537          * space of MDT storing inode.
1538          */
1539         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1540                 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
1541                         continue;
1542                 md_null_inode(lmv->tgts[i]->ltd_exp, fid);
1543         }
1544
1545         RETURN(0);
1546 }
1547
1548 static int lmv_close(struct obd_export *exp, struct md_op_data *op_data,
1549                      struct md_open_data *mod, struct ptlrpc_request **request)
1550 {
1551         struct obd_device     *obd = exp->exp_obd;
1552         struct lmv_obd        *lmv = &obd->u.lmv;
1553         struct lmv_tgt_desc   *tgt;
1554         int                    rc;
1555         ENTRY;
1556
1557         tgt = lmv_find_target(lmv, &op_data->op_fid1);
1558         if (IS_ERR(tgt))
1559                 RETURN(PTR_ERR(tgt));
1560
1561         CDEBUG(D_INODE, "CLOSE "DFID"\n", PFID(&op_data->op_fid1));
1562         rc = md_close(tgt->ltd_exp, op_data, mod, request);
1563         RETURN(rc);
1564 }
1565
1566 struct lmv_tgt_desc*
1567 __lmv_locate_tgt(struct lmv_obd *lmv, struct lmv_stripe_md *lsm,
1568                  const char *name, int namelen, struct lu_fid *fid, u32 *mds,
1569                  bool post_migrate)
1570 {
1571         struct lmv_tgt_desc *tgt;
1572         const struct lmv_oinfo *oinfo;
1573
1574         if (lsm == NULL || namelen == 0) {
1575                 tgt = lmv_find_target(lmv, fid);
1576                 if (IS_ERR(tgt))
1577                         return tgt;
1578
1579                 LASSERT(mds);
1580                 *mds = tgt->ltd_idx;
1581                 return tgt;
1582         }
1583
1584         if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_NAME_HASH)) {
1585                 if (cfs_fail_val >= lsm->lsm_md_stripe_count)
1586                         return ERR_PTR(-EBADF);
1587                 oinfo = &lsm->lsm_md_oinfo[cfs_fail_val];
1588         } else {
1589                 oinfo = lsm_name_to_stripe_info(lsm, name, namelen,
1590                                                 post_migrate);
1591                 if (IS_ERR(oinfo))
1592                         return ERR_CAST(oinfo);
1593         }
1594
1595         if (fid != NULL)
1596                 *fid = oinfo->lmo_fid;
1597         if (mds != NULL)
1598                 *mds = oinfo->lmo_mds;
1599         /* check stripe FID is sane */
1600         if (!fid_is_sane(&oinfo->lmo_fid))
1601                 return ERR_PTR(-ENODEV);
1602
1603         tgt = lmv_get_target(lmv, oinfo->lmo_mds, NULL);
1604
1605         CDEBUG(D_INFO, "locate on mds %u "DFID"\n", oinfo->lmo_mds,
1606                PFID(&oinfo->lmo_fid));
1607
1608         return tgt;
1609 }
1610
1611
1612 /**
1613  * Locate mdt by fid or name
1614  *
1615  * For striped directory, it will locate the stripe by name hash, if hash_type
1616  * is unknown, it will return the stripe specified by 'op_data->op_stripe_index'
1617  * which is set outside, and if dir is migrating, 'op_data->op_post_migrate'
1618  * indicates whether old or new layout is used to locate.
1619  *
1620  * For normal direcotry, it will locate MDS by FID directly.
1621  *
1622  * \param[in] lmv               LMV device
1623  * \param[in/out] op_data       client MD stack parameters, name, namelen etc,
1624  *                              op_mds and op_fid1 will be updated if op_mea1
1625  *                              indicates fid1 represents a striped directory.
1626  * \param[out] fid              object FID used to locate MDS.
1627  *
1628  * retval               pointer to the lmv_tgt_desc if succeed.
1629  *                      ERR_PTR(errno) if failed.
1630  */
1631 struct lmv_tgt_desc*
1632 lmv_locate_tgt(struct lmv_obd *lmv, struct md_op_data *op_data,
1633                struct lu_fid *fid)
1634 {
1635         struct lmv_stripe_md *lsm = op_data->op_mea1;
1636         struct lmv_oinfo *oinfo;
1637         struct lmv_tgt_desc *tgt;
1638
1639         /* During creating VOLATILE file, it should honor the mdt
1640          * index if the file under striped dir is being restored, see
1641          * ct_restore(). */
1642         if (op_data->op_bias & MDS_CREATE_VOLATILE &&
1643             (int)op_data->op_mds != -1) {
1644                 tgt = lmv_get_target(lmv, op_data->op_mds, NULL);
1645                 if (IS_ERR(tgt))
1646                         return tgt;
1647
1648                 if (lsm) {
1649                         int i;
1650
1651                         /* refill the right parent fid */
1652                         for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1653                                 oinfo = &lsm->lsm_md_oinfo[i];
1654                                 if (oinfo->lmo_mds == op_data->op_mds) {
1655                                         *fid = oinfo->lmo_fid;
1656                                         break;
1657                                 }
1658                         }
1659
1660                         if (i == lsm->lsm_md_stripe_count)
1661                                 *fid = lsm->lsm_md_oinfo[0].lmo_fid;
1662                 }
1663         } else if (lmv_is_dir_bad_hash(lsm)) {
1664                 LASSERT(op_data->op_stripe_index < lsm->lsm_md_stripe_count);
1665                 oinfo = &lsm->lsm_md_oinfo[op_data->op_stripe_index];
1666
1667                 *fid = oinfo->lmo_fid;
1668                 op_data->op_mds = oinfo->lmo_mds;
1669                 tgt = lmv_get_target(lmv, oinfo->lmo_mds, NULL);
1670         } else {
1671                 tgt = __lmv_locate_tgt(lmv, lsm, op_data->op_name,
1672                                        op_data->op_namelen, fid,
1673                                        &op_data->op_mds,
1674                                        op_data->op_post_migrate);
1675         }
1676
1677         return tgt;
1678 }
1679
1680 int lmv_create(struct obd_export *exp, struct md_op_data *op_data,
1681                 const void *data, size_t datalen, umode_t mode, uid_t uid,
1682                 gid_t gid, cfs_cap_t cap_effective, __u64 rdev,
1683                 struct ptlrpc_request **request)
1684 {
1685         struct obd_device       *obd = exp->exp_obd;
1686         struct lmv_obd          *lmv = &obd->u.lmv;
1687         struct lmv_tgt_desc     *tgt;
1688         int                      rc;
1689         ENTRY;
1690
1691         if (!lmv->desc.ld_active_tgt_count)
1692                 RETURN(-EIO);
1693
1694         if (lmv_is_dir_bad_hash(op_data->op_mea1))
1695                 RETURN(-EBADF);
1696
1697         if (lmv_is_dir_migrating(op_data->op_mea1)) {
1698                 /*
1699                  * if parent is migrating, create() needs to lookup existing
1700                  * name, to avoid creating new file under old layout of
1701                  * migrating directory, check old layout here.
1702                  */
1703                 tgt = lmv_locate_tgt(lmv, op_data, &op_data->op_fid1);
1704                 if (IS_ERR(tgt))
1705                         RETURN(PTR_ERR(tgt));
1706
1707                 rc = md_getattr_name(tgt->ltd_exp, op_data, request);
1708                 if (!rc) {
1709                         ptlrpc_req_finished(*request);
1710                         *request = NULL;
1711                         RETURN(-EEXIST);
1712                 }
1713
1714                 if (rc != -ENOENT)
1715                         RETURN(rc);
1716
1717                 op_data->op_post_migrate = true;
1718         }
1719
1720         tgt = lmv_locate_tgt(lmv, op_data, &op_data->op_fid1);
1721         if (IS_ERR(tgt))
1722                 RETURN(PTR_ERR(tgt));
1723
1724         CDEBUG(D_INODE, "CREATE name '%.*s' on "DFID" -> mds #%x\n",
1725                 (int)op_data->op_namelen, op_data->op_name,
1726                 PFID(&op_data->op_fid1), op_data->op_mds);
1727
1728         rc = lmv_fid_alloc(NULL, exp, &op_data->op_fid2, op_data);
1729         if (rc)
1730                 RETURN(rc);
1731
1732         if (exp_connect_flags(exp) & OBD_CONNECT_DIR_STRIPE) {
1733                 /* Send the create request to the MDT where the object
1734                  * will be located */
1735                 tgt = lmv_find_target(lmv, &op_data->op_fid2);
1736                 if (IS_ERR(tgt))
1737                         RETURN(PTR_ERR(tgt));
1738
1739                 op_data->op_mds = tgt->ltd_idx;
1740         } else {
1741                 CDEBUG(D_CONFIG, "Server doesn't support striped dirs\n");
1742         }
1743
1744         CDEBUG(D_INODE, "CREATE obj "DFID" -> mds #%x\n",
1745                PFID(&op_data->op_fid2), op_data->op_mds);
1746
1747         op_data->op_flags |= MF_MDC_CANCEL_FID1;
1748         rc = md_create(tgt->ltd_exp, op_data, data, datalen, mode, uid, gid,
1749                        cap_effective, rdev, request);
1750         if (rc == 0) {
1751                 if (*request == NULL)
1752                         RETURN(rc);
1753                 CDEBUG(D_INODE, "Created - "DFID"\n", PFID(&op_data->op_fid2));
1754         }
1755         RETURN(rc);
1756 }
1757
1758 static int
1759 lmv_enqueue(struct obd_export *exp, struct ldlm_enqueue_info *einfo,
1760             const union ldlm_policy_data *policy, struct md_op_data *op_data,
1761             struct lustre_handle *lockh, __u64 extra_lock_flags)
1762 {
1763         struct obd_device        *obd = exp->exp_obd;
1764         struct lmv_obd           *lmv = &obd->u.lmv;
1765         struct lmv_tgt_desc      *tgt;
1766         int                       rc;
1767         ENTRY;
1768
1769         CDEBUG(D_INODE, "ENQUEUE on "DFID"\n", PFID(&op_data->op_fid1));
1770
1771         tgt = lmv_find_target(lmv, &op_data->op_fid1);
1772         if (IS_ERR(tgt))
1773                 RETURN(PTR_ERR(tgt));
1774
1775         CDEBUG(D_INODE, "ENQUEUE on "DFID" -> mds #%u\n",
1776                PFID(&op_data->op_fid1), tgt->ltd_idx);
1777
1778         rc = md_enqueue(tgt->ltd_exp, einfo, policy, op_data, lockh,
1779                         extra_lock_flags);
1780
1781         RETURN(rc);
1782 }
1783
1784 int
1785 lmv_getattr_name(struct obd_export *exp,struct md_op_data *op_data,
1786                  struct ptlrpc_request **preq)
1787 {
1788         struct obd_device *obd = exp->exp_obd;
1789         struct lmv_obd *lmv = &obd->u.lmv;
1790         struct lmv_tgt_desc *tgt;
1791         struct mdt_body *body;
1792         int rc;
1793
1794         ENTRY;
1795
1796 retry:
1797         tgt = lmv_locate_tgt(lmv, op_data, &op_data->op_fid1);
1798         if (IS_ERR(tgt))
1799                 RETURN(PTR_ERR(tgt));
1800
1801         CDEBUG(D_INODE, "GETATTR_NAME for %*s on "DFID" -> mds #%d\n",
1802                 (int)op_data->op_namelen, op_data->op_name,
1803                 PFID(&op_data->op_fid1), tgt->ltd_idx);
1804
1805         rc = md_getattr_name(tgt->ltd_exp, op_data, preq);
1806         if (rc == -ENOENT && lmv_dir_retry_check_update(op_data)) {
1807                 ptlrpc_req_finished(*preq);
1808                 *preq = NULL;
1809                 goto retry;
1810         }
1811
1812         if (rc)
1813                 RETURN(rc);
1814
1815         body = req_capsule_server_get(&(*preq)->rq_pill, &RMF_MDT_BODY);
1816         LASSERT(body != NULL);
1817
1818         if (body->mbo_valid & OBD_MD_MDS) {
1819                 op_data->op_fid1 = body->mbo_fid1;
1820                 op_data->op_valid |= OBD_MD_FLCROSSREF;
1821                 op_data->op_namelen = 0;
1822                 op_data->op_name = NULL;
1823
1824                 ptlrpc_req_finished(*preq);
1825                 *preq = NULL;
1826
1827                 goto retry;
1828         }
1829
1830         RETURN(rc);
1831 }
1832
1833 #define md_op_data_fid(op_data, fl)                     \
1834         (fl == MF_MDC_CANCEL_FID1 ? &op_data->op_fid1 : \
1835          fl == MF_MDC_CANCEL_FID2 ? &op_data->op_fid2 : \
1836          fl == MF_MDC_CANCEL_FID3 ? &op_data->op_fid3 : \
1837          fl == MF_MDC_CANCEL_FID4 ? &op_data->op_fid4 : \
1838          NULL)
1839
1840 static int lmv_early_cancel(struct obd_export *exp, struct lmv_tgt_desc *tgt,
1841                             struct md_op_data *op_data, __u32 op_tgt,
1842                             enum ldlm_mode mode, int bits, int flag)
1843 {
1844         struct lu_fid *fid = md_op_data_fid(op_data, flag);
1845         struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
1846         union ldlm_policy_data policy = { { 0 } };
1847         int rc = 0;
1848         ENTRY;
1849
1850         if (!fid_is_sane(fid))
1851                 RETURN(0);
1852
1853         if (tgt == NULL) {
1854                 tgt = lmv_find_target(lmv, fid);
1855                 if (IS_ERR(tgt))
1856                         RETURN(PTR_ERR(tgt));
1857         }
1858
1859         if (tgt->ltd_idx != op_tgt) {
1860                 CDEBUG(D_INODE, "EARLY_CANCEL on "DFID"\n", PFID(fid));
1861                 policy.l_inodebits.bits = bits;
1862                 rc = md_cancel_unused(tgt->ltd_exp, fid, &policy,
1863                                       mode, LCF_ASYNC, NULL);
1864         } else {
1865                 CDEBUG(D_INODE,
1866                        "EARLY_CANCEL skip operation target %d on "DFID"\n",
1867                        op_tgt, PFID(fid));
1868                 op_data->op_flags |= flag;
1869                 rc = 0;
1870         }
1871
1872         RETURN(rc);
1873 }
1874
1875 /*
1876  * llite passes fid of an target inode in op_data->op_fid1 and id of directory in
1877  * op_data->op_fid2
1878  */
1879 static int lmv_link(struct obd_export *exp, struct md_op_data *op_data,
1880                     struct ptlrpc_request **request)
1881 {
1882         struct obd_device       *obd = exp->exp_obd;
1883         struct lmv_obd          *lmv = &obd->u.lmv;
1884         struct lmv_tgt_desc     *tgt;
1885         int                      rc;
1886         ENTRY;
1887
1888         LASSERT(op_data->op_namelen != 0);
1889
1890         CDEBUG(D_INODE, "LINK "DFID":%*s to "DFID"\n",
1891                PFID(&op_data->op_fid2), (int)op_data->op_namelen,
1892                op_data->op_name, PFID(&op_data->op_fid1));
1893
1894         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
1895         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
1896         op_data->op_cap = cfs_curproc_cap_pack();
1897
1898         if (lmv_is_dir_migrating(op_data->op_mea2)) {
1899                 struct lu_fid fid1 = op_data->op_fid1;
1900                 struct lmv_stripe_md *lsm1 = op_data->op_mea1;
1901
1902                 /*
1903                  * avoid creating new file under old layout of migrating
1904                  * directory, check it here.
1905                  */
1906                 tgt = __lmv_locate_tgt(lmv, op_data->op_mea2, op_data->op_name,
1907                                        op_data->op_namelen, &op_data->op_fid2,
1908                                        &op_data->op_mds, false);
1909                 tgt = lmv_locate_tgt(lmv, op_data, &op_data->op_fid1);
1910                 if (IS_ERR(tgt))
1911                         RETURN(PTR_ERR(tgt));
1912
1913                 op_data->op_fid1 = op_data->op_fid2;
1914                 op_data->op_mea1 = op_data->op_mea2;
1915                 rc = md_getattr_name(tgt->ltd_exp, op_data, request);
1916                 op_data->op_fid1 = fid1;
1917                 op_data->op_mea1 = lsm1;
1918                 if (!rc) {
1919                         ptlrpc_req_finished(*request);
1920                         *request = NULL;
1921                         RETURN(-EEXIST);
1922                 }
1923
1924                 if (rc != -ENOENT)
1925                         RETURN(rc);
1926         }
1927
1928         tgt = __lmv_locate_tgt(lmv, op_data->op_mea2, op_data->op_name,
1929                                op_data->op_namelen, &op_data->op_fid2,
1930                                &op_data->op_mds, true);
1931         if (IS_ERR(tgt))
1932                 RETURN(PTR_ERR(tgt));
1933
1934         /*
1935          * Cancel UPDATE lock on child (fid1).
1936          */
1937         op_data->op_flags |= MF_MDC_CANCEL_FID2;
1938         rc = lmv_early_cancel(exp, NULL, op_data, tgt->ltd_idx, LCK_EX,
1939                               MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID1);
1940         if (rc != 0)
1941                 RETURN(rc);
1942
1943         rc = md_link(tgt->ltd_exp, op_data, request);
1944
1945         RETURN(rc);
1946 }
1947
1948 static int lmv_migrate(struct obd_export *exp, struct md_op_data *op_data,
1949                         const char *name, size_t namelen,
1950                         struct ptlrpc_request **request)
1951 {
1952         struct obd_device *obd = exp->exp_obd;
1953         struct lmv_obd *lmv = &obd->u.lmv;
1954         struct lmv_stripe_md *lsm = op_data->op_mea1;
1955         struct lmv_tgt_desc *parent_tgt;
1956         struct lmv_tgt_desc *sp_tgt;
1957         struct lmv_tgt_desc *tp_tgt = NULL;
1958         struct lmv_tgt_desc *child_tgt;
1959         struct lmv_tgt_desc *tgt;
1960         struct lu_fid target_fid;
1961         int rc;
1962
1963         ENTRY;
1964
1965         LASSERT(op_data->op_cli_flags & CLI_MIGRATE);
1966
1967         CDEBUG(D_INODE, "MIGRATE "DFID"/%.*s\n",
1968                PFID(&op_data->op_fid1), (int)namelen, name);
1969
1970         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
1971         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
1972         op_data->op_cap = cfs_curproc_cap_pack();
1973
1974         parent_tgt = lmv_find_target(lmv, &op_data->op_fid1);
1975         if (IS_ERR(parent_tgt))
1976                 RETURN(PTR_ERR(parent_tgt));
1977
1978         if (lsm) {
1979                 __u32 hash_type = lsm->lsm_md_hash_type;
1980                 __u32 stripe_count = lsm->lsm_md_stripe_count;
1981
1982                 /*
1983                  * old stripes are appended after new stripes for migrating
1984                  * directory.
1985                  */
1986                 if (lsm->lsm_md_hash_type & LMV_HASH_FLAG_MIGRATION) {
1987                         hash_type = lsm->lsm_md_migrate_hash;
1988                         stripe_count -= lsm->lsm_md_migrate_offset;
1989                 }
1990
1991                 rc = lmv_name_to_stripe_index(hash_type, stripe_count, name,
1992                                               namelen);
1993                 if (rc < 0)
1994                         RETURN(rc);
1995
1996                 if (lsm->lsm_md_hash_type & LMV_HASH_FLAG_MIGRATION)
1997                         rc += lsm->lsm_md_migrate_offset;
1998
1999                 /* save it in fid4 temporarily for early cancel */
2000                 op_data->op_fid4 = lsm->lsm_md_oinfo[rc].lmo_fid;
2001                 sp_tgt = lmv_get_target(lmv, lsm->lsm_md_oinfo[rc].lmo_mds,
2002                                         NULL);
2003                 if (IS_ERR(sp_tgt))
2004                         RETURN(PTR_ERR(sp_tgt));
2005
2006                 /*
2007                  * if parent is being migrated too, fill op_fid2 with target
2008                  * stripe fid, otherwise the target stripe is not created yet.
2009                  */
2010                 if (lsm->lsm_md_hash_type & LMV_HASH_FLAG_MIGRATION) {
2011                         hash_type = lsm->lsm_md_hash_type &
2012                                     ~LMV_HASH_FLAG_MIGRATION;
2013                         stripe_count = lsm->lsm_md_migrate_offset;
2014
2015                         rc = lmv_name_to_stripe_index(hash_type, stripe_count,
2016                                                       name, namelen);
2017                         if (rc < 0)
2018                                 RETURN(rc);
2019
2020                         op_data->op_fid2 = lsm->lsm_md_oinfo[rc].lmo_fid;
2021                         tp_tgt = lmv_get_target(lmv,
2022                                                 lsm->lsm_md_oinfo[rc].lmo_mds,
2023                                                 NULL);
2024                         if (IS_ERR(tp_tgt))
2025                                 RETURN(PTR_ERR(tp_tgt));
2026                 }
2027         } else {
2028                 sp_tgt = parent_tgt;
2029         }
2030
2031         child_tgt = lmv_find_target(lmv, &op_data->op_fid3);
2032         if (IS_ERR(child_tgt))
2033                 RETURN(PTR_ERR(child_tgt));
2034
2035         if (!S_ISDIR(op_data->op_mode) && tp_tgt)
2036                 rc = __lmv_fid_alloc(lmv, &target_fid, tp_tgt->ltd_idx);
2037         else
2038                 rc = lmv_fid_alloc(NULL, exp, &target_fid, op_data);
2039         if (rc)
2040                 RETURN(rc);
2041
2042         /*
2043          * for directory, send migrate request to the MDT where the object will
2044          * be migrated to, because we can't create a striped directory remotely.
2045          *
2046          * otherwise, send to the MDT where source is located because regular
2047          * file may open lease.
2048          *
2049          * NB. if MDT doesn't support DIR_MIGRATE, send to source MDT too for
2050          * backward compatibility.
2051          */
2052         if (S_ISDIR(op_data->op_mode) &&
2053             (exp_connect_flags2(exp) & OBD_CONNECT2_DIR_MIGRATE)) {
2054                 tgt = lmv_find_target(lmv, &target_fid);
2055                 if (IS_ERR(tgt))
2056                         RETURN(PTR_ERR(tgt));
2057         } else {
2058                 tgt = child_tgt;
2059         }
2060
2061         /* cancel UPDATE lock of parent master object */
2062         rc = lmv_early_cancel(exp, parent_tgt, op_data, tgt->ltd_idx, LCK_EX,
2063                               MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID1);
2064         if (rc)
2065                 RETURN(rc);
2066
2067         /* cancel UPDATE lock of source parent */
2068         if (sp_tgt != parent_tgt) {
2069                 /*
2070                  * migrate RPC packs master object FID, because we can only pack
2071                  * two FIDs in reint RPC, but MDS needs to know both source
2072                  * parent and target parent, and it will obtain them from master
2073                  * FID and LMV, the other FID in RPC is kept for target.
2074                  *
2075                  * since this FID is not passed to MDC, cancel it anyway.
2076                  */
2077                 rc = lmv_early_cancel(exp, sp_tgt, op_data, -1, LCK_EX,
2078                                       MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID4);
2079                 if (rc)
2080                         RETURN(rc);
2081
2082                 op_data->op_flags &= ~MF_MDC_CANCEL_FID4;
2083         }
2084         op_data->op_fid4 = target_fid;
2085
2086         /* cancel UPDATE locks of target parent */
2087         rc = lmv_early_cancel(exp, tp_tgt, op_data, tgt->ltd_idx, LCK_EX,
2088                               MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID2);
2089         if (rc)
2090                 RETURN(rc);
2091
2092         /* cancel LOOKUP lock of source if source is remote object */
2093         if (child_tgt != sp_tgt) {
2094                 rc = lmv_early_cancel(exp, sp_tgt, op_data, tgt->ltd_idx,
2095                                       LCK_EX, MDS_INODELOCK_LOOKUP,
2096                                       MF_MDC_CANCEL_FID3);
2097                 if (rc)
2098                         RETURN(rc);
2099         }
2100
2101         /* cancel ELC locks of source */
2102         rc = lmv_early_cancel(exp, child_tgt, op_data, tgt->ltd_idx, LCK_EX,
2103                               MDS_INODELOCK_ELC, MF_MDC_CANCEL_FID3);
2104         if (rc)
2105                 RETURN(rc);
2106
2107         rc = md_rename(tgt->ltd_exp, op_data, name, namelen, NULL, 0, request);
2108
2109         RETURN(rc);
2110 }
2111
2112 static int lmv_rename(struct obd_export *exp, struct md_op_data *op_data,
2113                       const char *old, size_t oldlen,
2114                       const char *new, size_t newlen,
2115                       struct ptlrpc_request **request)
2116 {
2117         struct obd_device *obd = exp->exp_obd;
2118         struct lmv_obd *lmv = &obd->u.lmv;
2119         struct lmv_tgt_desc *sp_tgt;
2120         struct lmv_tgt_desc *tp_tgt = NULL;
2121         struct lmv_tgt_desc *src_tgt = NULL;
2122         struct lmv_tgt_desc *tgt;
2123         struct mdt_body *body;
2124         int rc;
2125
2126         ENTRY;
2127
2128         LASSERT(oldlen != 0);
2129
2130         if (op_data->op_cli_flags & CLI_MIGRATE) {
2131                 rc = lmv_migrate(exp, op_data, old, oldlen, request);
2132                 RETURN(rc);
2133         }
2134
2135         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2136         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2137         op_data->op_cap = cfs_curproc_cap_pack();
2138
2139         if (lmv_is_dir_migrating(op_data->op_mea2)) {
2140                 struct lu_fid fid1 = op_data->op_fid1;
2141                 struct lmv_stripe_md *lsm1 = op_data->op_mea1;
2142
2143                 /*
2144                  * we avoid creating new file under old layout of migrating
2145                  * directory, if there is an existing file with new name under
2146                  * old layout, we can't unlink file in old layout and rename to
2147                  * new layout in one transaction, so return -EBUSY here.`
2148                  */
2149                 tgt = __lmv_locate_tgt(lmv, op_data->op_mea2, new, newlen,
2150                                        &op_data->op_fid2, &op_data->op_mds,
2151                                        false);
2152                 if (IS_ERR(tgt))
2153                         RETURN(PTR_ERR(tgt));
2154
2155                 op_data->op_fid1 = op_data->op_fid2;
2156                 op_data->op_mea1 = op_data->op_mea2;
2157                 op_data->op_name = new;
2158                 op_data->op_namelen = newlen;
2159                 rc = md_getattr_name(tgt->ltd_exp, op_data, request);
2160                 op_data->op_fid1 = fid1;
2161                 op_data->op_mea1 = lsm1;
2162                 op_data->op_name = NULL;
2163                 op_data->op_namelen = 0;
2164                 if (!rc) {
2165                         ptlrpc_req_finished(*request);
2166                         *request = NULL;
2167                         RETURN(-EBUSY);
2168                 }
2169
2170                 if (rc != -ENOENT)
2171                         RETURN(rc);
2172         }
2173
2174         /* rename to new layout for migrating directory */
2175         tp_tgt = __lmv_locate_tgt(lmv, op_data->op_mea2, new, newlen,
2176                                   &op_data->op_fid2, &op_data->op_mds, true);
2177         if (IS_ERR(tp_tgt))
2178                 RETURN(PTR_ERR(tp_tgt));
2179
2180         /* Since the target child might be destroyed, and it might become
2181          * orphan, and we can only check orphan on the local MDT right now, so
2182          * we send rename request to the MDT where target child is located. If
2183          * target child does not exist, then it will send the request to the
2184          * target parent */
2185         if (fid_is_sane(&op_data->op_fid4)) {
2186                 tgt = lmv_find_target(lmv, &op_data->op_fid4);
2187                 if (IS_ERR(tgt))
2188                         RETURN(PTR_ERR(tgt));
2189         } else {
2190                 tgt = tp_tgt;
2191         }
2192
2193         op_data->op_flags |= MF_MDC_CANCEL_FID4;
2194
2195         /* cancel UPDATE locks of target parent */
2196         rc = lmv_early_cancel(exp, tp_tgt, op_data, tgt->ltd_idx, LCK_EX,
2197                               MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID2);
2198         if (rc != 0)
2199                 RETURN(rc);
2200
2201         if (fid_is_sane(&op_data->op_fid4)) {
2202                 /* cancel LOOKUP lock of target on target parent */
2203                 if (tgt != tp_tgt) {
2204                         rc = lmv_early_cancel(exp, tp_tgt, op_data,
2205                                               tgt->ltd_idx, LCK_EX,
2206                                               MDS_INODELOCK_LOOKUP,
2207                                               MF_MDC_CANCEL_FID4);
2208                         if (rc != 0)
2209                                 RETURN(rc);
2210                 }
2211         }
2212
2213         if (fid_is_sane(&op_data->op_fid3)) {
2214                 src_tgt = lmv_find_target(lmv, &op_data->op_fid3);
2215                 if (IS_ERR(src_tgt))
2216                         RETURN(PTR_ERR(src_tgt));
2217
2218                 /* cancel ELC locks of source */
2219                 rc = lmv_early_cancel(exp, src_tgt, op_data, tgt->ltd_idx,
2220                                       LCK_EX, MDS_INODELOCK_ELC,
2221                                       MF_MDC_CANCEL_FID3);
2222                 if (rc != 0)
2223                         RETURN(rc);
2224         }
2225
2226 retry:
2227         sp_tgt = __lmv_locate_tgt(lmv, op_data->op_mea1, old, oldlen,
2228                                   &op_data->op_fid1, &op_data->op_mds,
2229                                   op_data->op_post_migrate);
2230         if (IS_ERR(sp_tgt))
2231                 RETURN(PTR_ERR(sp_tgt));
2232
2233         /* cancel UPDATE locks of source parent */
2234         rc = lmv_early_cancel(exp, sp_tgt, op_data, tgt->ltd_idx, LCK_EX,
2235                               MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID1);
2236         if (rc != 0)
2237                 RETURN(rc);
2238
2239         if (fid_is_sane(&op_data->op_fid3)) {
2240                 /* cancel LOOKUP lock of source on source parent */
2241                 if (src_tgt != sp_tgt) {
2242                         rc = lmv_early_cancel(exp, sp_tgt, op_data,
2243                                               tgt->ltd_idx, LCK_EX,
2244                                               MDS_INODELOCK_LOOKUP,
2245                                               MF_MDC_CANCEL_FID3);
2246                         if (rc != 0)
2247                                 RETURN(rc);
2248                 }
2249         }
2250
2251 rename:
2252         CDEBUG(D_INODE, "RENAME "DFID"/%.*s to "DFID"/%.*s\n",
2253                 PFID(&op_data->op_fid1), (int)oldlen, old,
2254                 PFID(&op_data->op_fid2), (int)newlen, new);
2255
2256         rc = md_rename(tgt->ltd_exp, op_data, old, oldlen, new, newlen,
2257                         request);
2258         if (rc == -ENOENT && lmv_dir_retry_check_update(op_data)) {
2259                 ptlrpc_req_finished(*request);
2260                 *request = NULL;
2261                 goto retry;
2262         }
2263
2264         if (rc && rc != -EXDEV)
2265                 RETURN(rc);
2266
2267         body = req_capsule_server_get(&(*request)->rq_pill, &RMF_MDT_BODY);
2268         if (body == NULL)
2269                 RETURN(-EPROTO);
2270
2271         /* Not cross-ref case, just get out of here. */
2272         if (likely(!(body->mbo_valid & OBD_MD_MDS)))
2273                 RETURN(rc);
2274
2275         op_data->op_fid4 = body->mbo_fid1;
2276
2277         ptlrpc_req_finished(*request);
2278         *request = NULL;
2279
2280         tgt = lmv_find_target(lmv, &op_data->op_fid4);
2281         if (IS_ERR(tgt))
2282                 RETURN(PTR_ERR(tgt));
2283
2284         if (fid_is_sane(&op_data->op_fid4)) {
2285                 /* cancel LOOKUP lock of target on target parent */
2286                 if (tgt != tp_tgt) {
2287                         rc = lmv_early_cancel(exp, tp_tgt, op_data,
2288                                               tgt->ltd_idx, LCK_EX,
2289                                               MDS_INODELOCK_LOOKUP,
2290                                               MF_MDC_CANCEL_FID4);
2291                         if (rc != 0)
2292                                 RETURN(rc);
2293                 }
2294         }
2295
2296         goto rename;
2297 }
2298
2299 static int lmv_setattr(struct obd_export *exp, struct md_op_data *op_data,
2300                        void *ea, size_t ealen, struct ptlrpc_request **request)
2301 {
2302         struct obd_device       *obd = exp->exp_obd;
2303         struct lmv_obd          *lmv = &obd->u.lmv;
2304         struct lmv_tgt_desc     *tgt;
2305         int                      rc = 0;
2306         ENTRY;
2307
2308         CDEBUG(D_INODE, "SETATTR for "DFID", valid 0x%x/0x%x\n",
2309                PFID(&op_data->op_fid1), op_data->op_attr.ia_valid,
2310                op_data->op_xvalid);
2311
2312         op_data->op_flags |= MF_MDC_CANCEL_FID1;
2313         tgt = lmv_find_target(lmv, &op_data->op_fid1);
2314         if (IS_ERR(tgt))
2315                 RETURN(PTR_ERR(tgt));
2316
2317         rc = md_setattr(tgt->ltd_exp, op_data, ea, ealen, request);
2318
2319         RETURN(rc);
2320 }
2321
2322 static int lmv_fsync(struct obd_export *exp, const struct lu_fid *fid,
2323                      struct ptlrpc_request **request)
2324 {
2325         struct obd_device       *obd = exp->exp_obd;
2326         struct lmv_obd          *lmv = &obd->u.lmv;
2327         struct lmv_tgt_desc     *tgt;
2328         int                      rc;
2329         ENTRY;
2330
2331         tgt = lmv_find_target(lmv, fid);
2332         if (IS_ERR(tgt))
2333                 RETURN(PTR_ERR(tgt));
2334
2335         rc = md_fsync(tgt->ltd_exp, fid, request);
2336         RETURN(rc);
2337 }
2338
2339 struct stripe_dirent {
2340         struct page             *sd_page;
2341         struct lu_dirpage       *sd_dp;
2342         struct lu_dirent        *sd_ent;
2343         bool                     sd_eof;
2344 };
2345
2346 struct lmv_dir_ctxt {
2347         struct lmv_obd          *ldc_lmv;
2348         struct md_op_data       *ldc_op_data;
2349         struct md_callback      *ldc_cb_op;
2350         __u64                    ldc_hash;
2351         int                      ldc_count;
2352         struct stripe_dirent     ldc_stripes[0];
2353 };
2354
2355 static inline void stripe_dirent_unload(struct stripe_dirent *stripe)
2356 {
2357         if (stripe->sd_page) {
2358                 kunmap(stripe->sd_page);
2359                 put_page(stripe->sd_page);
2360                 stripe->sd_page = NULL;
2361                 stripe->sd_ent = NULL;
2362         }
2363 }
2364
2365 static inline void put_lmv_dir_ctxt(struct lmv_dir_ctxt *ctxt)
2366 {
2367         int i;
2368
2369         for (i = 0; i < ctxt->ldc_count; i++)
2370                 stripe_dirent_unload(&ctxt->ldc_stripes[i]);
2371 }
2372
2373 /* if @ent is dummy, or . .., get next */
2374 static struct lu_dirent *stripe_dirent_get(struct lmv_dir_ctxt *ctxt,
2375                                            struct lu_dirent *ent,
2376                                            int stripe_index)
2377 {
2378         for (; ent; ent = lu_dirent_next(ent)) {
2379                 /* Skip dummy entry */
2380                 if (le16_to_cpu(ent->lde_namelen) == 0)
2381                         continue;
2382
2383                 /* skip . and .. for other stripes */
2384                 if (stripe_index &&
2385                     (strncmp(ent->lde_name, ".",
2386                              le16_to_cpu(ent->lde_namelen)) == 0 ||
2387                      strncmp(ent->lde_name, "..",
2388                              le16_to_cpu(ent->lde_namelen)) == 0))
2389                         continue;
2390
2391                 if (le64_to_cpu(ent->lde_hash) >= ctxt->ldc_hash)
2392                         break;
2393         }
2394
2395         return ent;
2396 }
2397
2398 static struct lu_dirent *stripe_dirent_load(struct lmv_dir_ctxt *ctxt,
2399                                             struct stripe_dirent *stripe,
2400                                             int stripe_index)
2401 {
2402         struct md_op_data *op_data = ctxt->ldc_op_data;
2403         struct lmv_oinfo *oinfo;
2404         struct lu_fid fid = op_data->op_fid1;
2405         struct inode *inode = op_data->op_data;
2406         struct lmv_tgt_desc *tgt;
2407         struct lu_dirent *ent = stripe->sd_ent;
2408         __u64 hash = ctxt->ldc_hash;
2409         int rc = 0;
2410
2411         ENTRY;
2412
2413         LASSERT(stripe == &ctxt->ldc_stripes[stripe_index]);
2414         LASSERT(!ent);
2415
2416         do {
2417                 if (stripe->sd_page) {
2418                         __u64 end = le64_to_cpu(stripe->sd_dp->ldp_hash_end);
2419
2420                         /* @hash should be the last dirent hash */
2421                         LASSERTF(hash <= end,
2422                                  "ctxt@%p stripe@%p hash %llx end %llx\n",
2423                                  ctxt, stripe, hash, end);
2424                         /* unload last page */
2425                         stripe_dirent_unload(stripe);
2426                         /* eof */
2427                         if (end == MDS_DIR_END_OFF) {
2428                                 stripe->sd_eof = true;
2429                                 break;
2430                         }
2431                         hash = end;
2432                 }
2433
2434                 oinfo = &op_data->op_mea1->lsm_md_oinfo[stripe_index];
2435                 if (!oinfo->lmo_root) {
2436                         rc = -ENOENT;
2437                         break;
2438                 }
2439
2440                 tgt = lmv_get_target(ctxt->ldc_lmv, oinfo->lmo_mds, NULL);
2441                 if (IS_ERR(tgt)) {
2442                         rc = PTR_ERR(tgt);
2443                         break;
2444                 }
2445
2446                 /* op_data is shared by stripes, reset after use */
2447                 op_data->op_fid1 = oinfo->lmo_fid;
2448                 op_data->op_fid2 = oinfo->lmo_fid;
2449                 op_data->op_data = oinfo->lmo_root;
2450
2451                 rc = md_read_page(tgt->ltd_exp, op_data, ctxt->ldc_cb_op, hash,
2452                                   &stripe->sd_page);
2453
2454                 op_data->op_fid1 = fid;
2455                 op_data->op_fid2 = fid;
2456                 op_data->op_data = inode;
2457
2458                 if (rc)
2459                         break;
2460
2461                 stripe->sd_dp = page_address(stripe->sd_page);
2462                 ent = stripe_dirent_get(ctxt, lu_dirent_start(stripe->sd_dp),
2463                                         stripe_index);
2464                 /* in case a page filled with ., .. and dummy, read next */
2465         } while (!ent);
2466
2467         stripe->sd_ent = ent;
2468         if (rc) {
2469                 LASSERT(!ent);
2470                 /* treat error as eof, so dir can be partially accessed */
2471                 stripe->sd_eof = true;
2472                 LCONSOLE_WARN("dir "DFID" stripe %d readdir failed: %d, "
2473                               "directory is partially accessed!\n",
2474                               PFID(&ctxt->ldc_op_data->op_fid1), stripe_index,
2475                               rc);
2476         }
2477
2478         RETURN(ent);
2479 }
2480
2481 static int lmv_file_resync(struct obd_export *exp, struct md_op_data *data)
2482 {
2483         struct obd_device       *obd = exp->exp_obd;
2484         struct lmv_obd          *lmv = &obd->u.lmv;
2485         struct lmv_tgt_desc     *tgt;
2486         int                      rc;
2487         ENTRY;
2488
2489         rc = lmv_check_connect(obd);
2490         if (rc != 0)
2491                 RETURN(rc);
2492
2493         tgt = lmv_find_target(lmv, &data->op_fid1);
2494         if (IS_ERR(tgt))
2495                 RETURN(PTR_ERR(tgt));
2496
2497         data->op_flags |= MF_MDC_CANCEL_FID1;
2498         rc = md_file_resync(tgt->ltd_exp, data);
2499         RETURN(rc);
2500 }
2501
2502 /**
2503  * Get dirent with the closest hash for striped directory
2504  *
2505  * This function will search the dir entry, whose hash value is the
2506  * closest(>=) to hash from all of sub-stripes, and it is only being called
2507  * for striped directory.
2508  *
2509  * \param[in] ctxt              dir read context
2510  *
2511  * \retval                      dirent get the entry successfully
2512  *                              NULL does not get the entry, normally it means
2513  *                              it reaches the end of the directory, while read
2514  *                              stripe dirent error is ignored to allow partial
2515  *                              access.
2516  */
2517 static struct lu_dirent *lmv_dirent_next(struct lmv_dir_ctxt *ctxt)
2518 {
2519         struct stripe_dirent *stripe;
2520         struct lu_dirent *ent = NULL;
2521         int i;
2522         int min = -1;
2523
2524         /* TODO: optimize with k-way merge sort */
2525         for (i = 0; i < ctxt->ldc_count; i++) {
2526                 stripe = &ctxt->ldc_stripes[i];
2527                 if (stripe->sd_eof)
2528                         continue;
2529
2530                 if (!stripe->sd_ent) {
2531                         stripe_dirent_load(ctxt, stripe, i);
2532                         if (!stripe->sd_ent) {
2533                                 LASSERT(stripe->sd_eof);
2534                                 continue;
2535                         }
2536                 }
2537
2538                 if (min == -1 ||
2539                     le64_to_cpu(ctxt->ldc_stripes[min].sd_ent->lde_hash) >
2540                     le64_to_cpu(stripe->sd_ent->lde_hash)) {
2541                         min = i;
2542                         if (le64_to_cpu(stripe->sd_ent->lde_hash) ==
2543                             ctxt->ldc_hash)
2544                                 break;
2545                 }
2546         }
2547
2548         if (min != -1) {
2549                 stripe = &ctxt->ldc_stripes[min];
2550                 ent = stripe->sd_ent;
2551                 /* pop found dirent */
2552                 stripe->sd_ent = stripe_dirent_get(ctxt, lu_dirent_next(ent),
2553                                                    min);
2554         }
2555
2556         return ent;
2557 }
2558
2559 /**
2560  * Build dir entry page for striped directory
2561  *
2562  * This function gets one entry by @offset from a striped directory. It will
2563  * read entries from all of stripes, and choose one closest to the required
2564  * offset(&offset). A few notes
2565  * 1. skip . and .. for non-zero stripes, because there can only have one .
2566  * and .. in a directory.
2567  * 2. op_data will be shared by all of stripes, instead of allocating new
2568  * one, so need to restore before reusing.
2569  *
2570  * \param[in] exp       obd export refer to LMV
2571  * \param[in] op_data   hold those MD parameters of read_entry
2572  * \param[in] cb_op     ldlm callback being used in enqueue in mdc_read_entry
2573  * \param[in] offset    starting hash offset
2574  * \param[out] ppage    the page holding the entry. Note: because the entry
2575  *                      will be accessed in upper layer, so we need hold the
2576  *                      page until the usages of entry is finished, see
2577  *                      ll_dir_entry_next.
2578  *
2579  * retval               =0 if get entry successfully
2580  *                      <0 cannot get entry
2581  */
2582 static int lmv_striped_read_page(struct obd_export *exp,
2583                                  struct md_op_data *op_data,
2584                                  struct md_callback *cb_op,
2585                                  __u64 offset, struct page **ppage)
2586 {
2587         struct page *page = NULL;
2588         struct lu_dirpage *dp;
2589         void *start;
2590         struct lu_dirent *ent;
2591         struct lu_dirent *last_ent;
2592         int stripe_count;
2593         struct lmv_dir_ctxt *ctxt;
2594         struct lu_dirent *next = NULL;
2595         __u16 ent_size;
2596         size_t left_bytes;
2597         int rc = 0;
2598         ENTRY;
2599
2600         /* Allocate a page and read entries from all of stripes and fill
2601          * the page by hash order */
2602         page = alloc_page(GFP_KERNEL);
2603         if (!page)
2604                 RETURN(-ENOMEM);
2605
2606         /* Initialize the entry page */
2607         dp = kmap(page);
2608         memset(dp, 0, sizeof(*dp));
2609         dp->ldp_hash_start = cpu_to_le64(offset);
2610
2611         start = dp + 1;
2612         left_bytes = PAGE_SIZE - sizeof(*dp);
2613         ent = start;
2614         last_ent = ent;
2615
2616         /* initalize dir read context */
2617         stripe_count = op_data->op_mea1->lsm_md_stripe_count;
2618         OBD_ALLOC(ctxt, offsetof(typeof(*ctxt), ldc_stripes[stripe_count]));
2619         if (!ctxt)
2620                 GOTO(free_page, rc = -ENOMEM);
2621         ctxt->ldc_lmv = &exp->exp_obd->u.lmv;
2622         ctxt->ldc_op_data = op_data;
2623         ctxt->ldc_cb_op = cb_op;
2624         ctxt->ldc_hash = offset;
2625         ctxt->ldc_count = stripe_count;
2626
2627         while (1) {
2628                 next = lmv_dirent_next(ctxt);
2629
2630                 /* end of directory */
2631                 if (!next) {
2632                         ctxt->ldc_hash = MDS_DIR_END_OFF;
2633                         break;
2634                 }
2635                 ctxt->ldc_hash = le64_to_cpu(next->lde_hash);
2636
2637                 ent_size = le16_to_cpu(next->lde_reclen);
2638
2639                 /* the last entry lde_reclen is 0, but it might not be the last
2640                  * one of this temporay dir page */
2641                 if (!ent_size)
2642                         ent_size = lu_dirent_calc_size(
2643                                         le16_to_cpu(next->lde_namelen),
2644                                         le32_to_cpu(next->lde_attrs));
2645                 /* page full */
2646                 if (ent_size > left_bytes)
2647                         break;
2648
2649                 memcpy(ent, next, ent_size);
2650
2651                 /* Replace . with master FID and Replace .. with the parent FID
2652                  * of master object */
2653                 if (strncmp(ent->lde_name, ".",
2654                             le16_to_cpu(ent->lde_namelen)) == 0 &&
2655                     le16_to_cpu(ent->lde_namelen) == 1)
2656                         fid_cpu_to_le(&ent->lde_fid, &op_data->op_fid1);
2657                 else if (strncmp(ent->lde_name, "..",
2658                                    le16_to_cpu(ent->lde_namelen)) == 0 &&
2659                            le16_to_cpu(ent->lde_namelen) == 2)
2660                         fid_cpu_to_le(&ent->lde_fid, &op_data->op_fid3);
2661
2662                 CDEBUG(D_INODE, "entry %.*s hash %#llx\n",
2663                        le16_to_cpu(ent->lde_namelen), ent->lde_name,
2664                        le64_to_cpu(ent->lde_hash));
2665
2666                 left_bytes -= ent_size;
2667                 ent->lde_reclen = cpu_to_le16(ent_size);
2668                 last_ent = ent;
2669                 ent = (void *)ent + ent_size;
2670         };
2671
2672         last_ent->lde_reclen = 0;
2673
2674         if (ent == start)
2675                 dp->ldp_flags |= LDF_EMPTY;
2676         else if (ctxt->ldc_hash == le64_to_cpu(last_ent->lde_hash))
2677                 dp->ldp_flags |= LDF_COLLIDE;
2678         dp->ldp_flags = cpu_to_le32(dp->ldp_flags);
2679         dp->ldp_hash_end = cpu_to_le64(ctxt->ldc_hash);
2680
2681         put_lmv_dir_ctxt(ctxt);
2682         OBD_FREE(ctxt, offsetof(typeof(*ctxt), ldc_stripes[stripe_count]));
2683
2684         *ppage = page;
2685
2686         RETURN(0);
2687
2688 free_page:
2689         kunmap(page);
2690         __free_page(page);
2691
2692         return rc;
2693 }
2694
2695 int lmv_read_page(struct obd_export *exp, struct md_op_data *op_data,
2696                   struct md_callback *cb_op, __u64 offset,
2697                   struct page **ppage)
2698 {
2699         struct obd_device       *obd = exp->exp_obd;
2700         struct lmv_obd          *lmv = &obd->u.lmv;
2701         struct lmv_stripe_md    *lsm = op_data->op_mea1;
2702         struct lmv_tgt_desc     *tgt;
2703         int                     rc;
2704         ENTRY;
2705
2706         if (unlikely(lsm != NULL)) {
2707                 rc = lmv_striped_read_page(exp, op_data, cb_op, offset, ppage);
2708                 RETURN(rc);
2709         }
2710
2711         tgt = lmv_find_target(lmv, &op_data->op_fid1);
2712         if (IS_ERR(tgt))
2713                 RETURN(PTR_ERR(tgt));
2714
2715         rc = md_read_page(tgt->ltd_exp, op_data, cb_op, offset, ppage);
2716
2717         RETURN(rc);
2718 }
2719
2720 /**
2721  * Unlink a file/directory
2722  *
2723  * Unlink a file or directory under the parent dir. The unlink request
2724  * usually will be sent to the MDT where the child is located, but if
2725  * the client does not have the child FID then request will be sent to the
2726  * MDT where the parent is located.
2727  *
2728  * If the parent is a striped directory then it also needs to locate which
2729  * stripe the name of the child is located, and replace the parent FID
2730  * (@op->op_fid1) with the stripe FID. Note: if the stripe is unknown,
2731  * it will walk through all of sub-stripes until the child is being
2732  * unlinked finally.
2733  *
2734  * \param[in] exp       export refer to LMV
2735  * \param[in] op_data   different parameters transferred beween client
2736  *                      MD stacks, name, namelen, FIDs etc.
2737  *                      op_fid1 is the parent FID, op_fid2 is the child
2738  *                      FID.
2739  * \param[out] request  point to the request of unlink.
2740  *
2741  * retval               0 if succeed
2742  *                      negative errno if failed.
2743  */
2744 static int lmv_unlink(struct obd_export *exp, struct md_op_data *op_data,
2745                       struct ptlrpc_request **request)
2746 {
2747         struct obd_device *obd = exp->exp_obd;
2748         struct lmv_obd *lmv = &obd->u.lmv;
2749         struct lmv_tgt_desc *tgt;
2750         struct lmv_tgt_desc *parent_tgt;
2751         struct mdt_body *body;
2752         int rc;
2753
2754         ENTRY;
2755
2756         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2757         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2758         op_data->op_cap = cfs_curproc_cap_pack();
2759
2760 retry:
2761         parent_tgt = lmv_locate_tgt(lmv, op_data, &op_data->op_fid1);
2762         if (IS_ERR(parent_tgt))
2763                 RETURN(PTR_ERR(parent_tgt));
2764
2765         if (likely(!fid_is_zero(&op_data->op_fid2))) {
2766                 tgt = lmv_find_target(lmv, &op_data->op_fid2);
2767                 if (IS_ERR(tgt))
2768                         RETURN(PTR_ERR(tgt));
2769         } else {
2770                 tgt = parent_tgt;
2771         }
2772
2773         /*
2774          * If child's fid is given, cancel unused locks for it if it is from
2775          * another export than parent.
2776          *
2777          * LOOKUP lock for child (fid3) should also be cancelled on parent
2778          * tgt_tgt in mdc_unlink().
2779          */
2780         op_data->op_flags |= MF_MDC_CANCEL_FID1 | MF_MDC_CANCEL_FID3;
2781
2782         if (parent_tgt != tgt)
2783                 rc = lmv_early_cancel(exp, parent_tgt, op_data, tgt->ltd_idx,
2784                                       LCK_EX, MDS_INODELOCK_LOOKUP,
2785                                       MF_MDC_CANCEL_FID3);
2786
2787         rc = lmv_early_cancel(exp, NULL, op_data, tgt->ltd_idx, LCK_EX,
2788                               MDS_INODELOCK_ELC, MF_MDC_CANCEL_FID3);
2789         if (rc)
2790                 RETURN(rc);
2791
2792         CDEBUG(D_INODE, "unlink with fid="DFID"/"DFID" -> mds #%u\n",
2793                PFID(&op_data->op_fid1), PFID(&op_data->op_fid2), tgt->ltd_idx);
2794
2795         rc = md_unlink(tgt->ltd_exp, op_data, request);
2796         if (rc == -ENOENT && lmv_dir_retry_check_update(op_data)) {
2797                 ptlrpc_req_finished(*request);
2798                 *request = NULL;
2799                 goto retry;
2800         }
2801
2802         if (rc != -EREMOTE)
2803                 RETURN(rc);
2804
2805         body = req_capsule_server_get(&(*request)->rq_pill, &RMF_MDT_BODY);
2806         if (body == NULL)
2807                 RETURN(-EPROTO);
2808
2809         /* Not cross-ref case, just get out of here. */
2810         if (likely(!(body->mbo_valid & OBD_MD_MDS)))
2811                 RETURN(rc);
2812
2813         /* This is a remote object, try remote MDT. */
2814         op_data->op_fid2 = body->mbo_fid1;
2815         ptlrpc_req_finished(*request);
2816         *request = NULL;
2817
2818         tgt = lmv_find_target(lmv, &op_data->op_fid2);
2819         if (IS_ERR(tgt))
2820                 RETURN(PTR_ERR(tgt));
2821
2822         goto retry;
2823 }
2824
2825 static int lmv_precleanup(struct obd_device *obd)
2826 {
2827         ENTRY;
2828         libcfs_kkuc_group_rem(&obd->obd_uuid, 0, KUC_GRP_HSM);
2829         fld_client_debugfs_fini(&obd->u.lmv.lmv_fld);
2830         lprocfs_obd_cleanup(obd);
2831         lprocfs_free_md_stats(obd);
2832         RETURN(0);
2833 }
2834
2835 /**
2836  * Get by key a value associated with a LMV device.
2837  *
2838  * Dispatch request to lower-layer devices as needed.
2839  *
2840  * \param[in] env               execution environment for this thread
2841  * \param[in] exp               export for the LMV device
2842  * \param[in] keylen            length of key identifier
2843  * \param[in] key               identifier of key to get value for
2844  * \param[in] vallen            size of \a val
2845  * \param[out] val              pointer to storage location for value
2846  * \param[in] lsm               optional striping metadata of object
2847  *
2848  * \retval 0            on success
2849  * \retval negative     negated errno on failure
2850  */
2851 static int lmv_get_info(const struct lu_env *env, struct obd_export *exp,
2852                         __u32 keylen, void *key, __u32 *vallen, void *val)
2853 {
2854         struct obd_device       *obd;
2855         struct lmv_obd          *lmv;
2856         int                      rc = 0;
2857         ENTRY;
2858
2859         obd = class_exp2obd(exp);
2860         if (obd == NULL) {
2861                 CDEBUG(D_IOCTL, "Invalid client cookie %#llx\n",
2862                        exp->exp_handle.h_cookie);
2863                 RETURN(-EINVAL);
2864         }
2865
2866         lmv = &obd->u.lmv;
2867         if (keylen >= strlen("remote_flag") && !strcmp(key, "remote_flag")) {
2868                 int i;
2869
2870                 LASSERT(*vallen == sizeof(__u32));
2871                 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2872                         struct lmv_tgt_desc *tgt = lmv->tgts[i];
2873                         /*
2874                          * All tgts should be connected when this gets called.
2875                          */
2876                         if (tgt == NULL || tgt->ltd_exp == NULL)
2877                                 continue;
2878
2879                         if (!obd_get_info(env, tgt->ltd_exp, keylen, key,
2880                                           vallen, val))
2881                                 RETURN(0);
2882                 }
2883                 RETURN(-EINVAL);
2884         } else if (KEY_IS(KEY_MAX_EASIZE) ||
2885                    KEY_IS(KEY_DEFAULT_EASIZE) ||
2886                    KEY_IS(KEY_CONN_DATA)) {
2887                 /*
2888                  * Forwarding this request to first MDS, it should know LOV
2889                  * desc.
2890                  */
2891                 rc = obd_get_info(env, lmv->tgts[0]->ltd_exp, keylen, key,
2892                                   vallen, val);
2893                 if (!rc && KEY_IS(KEY_CONN_DATA))
2894                         exp->exp_connect_data = *(struct obd_connect_data *)val;
2895                 RETURN(rc);
2896         } else if (KEY_IS(KEY_TGT_COUNT)) {
2897                 *((int *)val) = lmv->desc.ld_tgt_count;
2898                 RETURN(0);
2899         }
2900
2901         CDEBUG(D_IOCTL, "Invalid key\n");
2902         RETURN(-EINVAL);
2903 }
2904
2905 static int lmv_rmfid(struct obd_export *exp, struct fid_array *fa,
2906                      int *__rcs, struct ptlrpc_request_set *_set)
2907 {
2908         struct obd_device *obddev = class_exp2obd(exp);
2909         struct ptlrpc_request_set *set = _set;
2910         struct lmv_obd *lmv = &obddev->u.lmv;
2911         int tgt_count = lmv->desc.ld_tgt_count;
2912         struct fid_array *fat, **fas = NULL;
2913         int i, rc, **rcs = NULL;
2914
2915         if (!set) {
2916                 set = ptlrpc_prep_set();
2917                 if (!set)
2918                         RETURN(-ENOMEM);
2919         }
2920
2921         /* split FIDs by targets */
2922         OBD_ALLOC(fas, sizeof(fas) * tgt_count);
2923         if (fas == NULL)
2924                 GOTO(out, rc = -ENOMEM);
2925         OBD_ALLOC(rcs, sizeof(int *) * tgt_count);
2926         if (rcs == NULL)
2927                 GOTO(out_fas, rc = -ENOMEM);
2928
2929         for (i = 0; i < fa->fa_nr; i++) {
2930                 unsigned int idx;
2931
2932                 rc = lmv_fld_lookup(lmv, &fa->fa_fids[i], &idx);
2933                 if (rc) {
2934                         CDEBUG(D_OTHER, "can't lookup "DFID": rc = %d\n",
2935                                PFID(&fa->fa_fids[i]), rc);
2936                         continue;
2937                 }
2938                 LASSERT(idx < tgt_count);
2939                 if (!fas[idx])
2940                         OBD_ALLOC(fas[idx], offsetof(struct fid_array,
2941                                   fa_fids[fa->fa_nr]));
2942                 if (!fas[idx])
2943                         GOTO(out, rc = -ENOMEM);
2944                 if (!rcs[idx])
2945                         OBD_ALLOC(rcs[idx], sizeof(int) * fa->fa_nr);
2946                 if (!rcs[idx])
2947                         GOTO(out, rc = -ENOMEM);
2948
2949                 fat = fas[idx];
2950                 fat->fa_fids[fat->fa_nr++] = fa->fa_fids[i];
2951         }
2952
2953         for (i = 0; i < tgt_count; i++) {
2954                 fat = fas[i];
2955                 if (!fat || fat->fa_nr == 0)
2956                         continue;
2957                 rc = md_rmfid(lmv->tgts[i]->ltd_exp, fat, rcs[i], set);
2958         }
2959
2960         rc = ptlrpc_set_wait(NULL, set);
2961         if (rc == 0) {
2962                 int j = 0;
2963                 for (i = 0; i < tgt_count; i++) {
2964                         fat = fas[i];
2965                         if (!fat || fat->fa_nr == 0)
2966                                 continue;
2967                         /* copy FIDs back */
2968                         memcpy(fa->fa_fids + j, fat->fa_fids,
2969                                fat->fa_nr * sizeof(struct lu_fid));
2970                         /* copy rcs back */
2971                         memcpy(__rcs + j, rcs[i], fat->fa_nr * sizeof(**rcs));
2972                         j += fat->fa_nr;
2973                 }
2974         }
2975         if (set != _set)
2976                 ptlrpc_set_destroy(set);
2977
2978 out:
2979         for (i = 0; i < tgt_count; i++) {
2980                 if (fas && fas[i])
2981                         OBD_FREE(fas[i], offsetof(struct fid_array,
2982                                                 fa_fids[fa->fa_nr]));
2983                 if (rcs && rcs[i])
2984                         OBD_FREE(rcs[i], sizeof(int) * fa->fa_nr);
2985         }
2986         if (rcs)
2987                 OBD_FREE(rcs, sizeof(int *) * tgt_count);
2988 out_fas:
2989         if (fas)
2990                 OBD_FREE(fas, sizeof(fas) * tgt_count);
2991
2992         RETURN(rc);
2993 }
2994
2995 /**
2996  * Asynchronously set by key a value associated with a LMV device.
2997  *
2998  * Dispatch request to lower-layer devices as needed.
2999  *
3000  * \param[in] env       execution environment for this thread
3001  * \param[in] exp       export for the LMV device
3002  * \param[in] keylen    length of key identifier
3003  * \param[in] key       identifier of key to store value for
3004  * \param[in] vallen    size of value to store
3005  * \param[in] val       pointer to data to be stored
3006  * \param[in] set       optional list of related ptlrpc requests
3007  *
3008  * \retval 0            on success
3009  * \retval negative     negated errno on failure
3010  */
3011 int lmv_set_info_async(const struct lu_env *env, struct obd_export *exp,
3012                         __u32 keylen, void *key, __u32 vallen, void *val,
3013                         struct ptlrpc_request_set *set)
3014 {
3015         struct lmv_tgt_desc     *tgt = NULL;
3016         struct obd_device       *obd;
3017         struct lmv_obd          *lmv;
3018         int rc = 0;
3019         ENTRY;
3020
3021         obd = class_exp2obd(exp);
3022         if (obd == NULL) {
3023                 CDEBUG(D_IOCTL, "Invalid client cookie %#llx\n",
3024                        exp->exp_handle.h_cookie);
3025                 RETURN(-EINVAL);
3026         }
3027         lmv = &obd->u.lmv;
3028
3029         if (KEY_IS(KEY_READ_ONLY) || KEY_IS(KEY_FLUSH_CTX) ||
3030             KEY_IS(KEY_DEFAULT_EASIZE)) {
3031                 int i, err = 0;
3032
3033                 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
3034                         tgt = lmv->tgts[i];
3035
3036                         if (tgt == NULL || tgt->ltd_exp == NULL)
3037                                 continue;
3038
3039                         err = obd_set_info_async(env, tgt->ltd_exp,
3040                                                  keylen, key, vallen, val, set);
3041                         if (err && rc == 0)
3042                                 rc = err;
3043                 }
3044
3045                 RETURN(rc);
3046         }
3047
3048         RETURN(-EINVAL);
3049 }
3050
3051 static int lmv_unpack_md_v1(struct obd_export *exp, struct lmv_stripe_md *lsm,
3052                             const struct lmv_mds_md_v1 *lmm1)
3053 {
3054         struct lmv_obd  *lmv = &exp->exp_obd->u.lmv;
3055         int             stripe_count;
3056         int             cplen;
3057         int             i;
3058         int             rc = 0;
3059         ENTRY;
3060
3061         lsm->lsm_md_magic = le32_to_cpu(lmm1->lmv_magic);
3062         lsm->lsm_md_stripe_count = le32_to_cpu(lmm1->lmv_stripe_count);
3063         lsm->lsm_md_master_mdt_index = le32_to_cpu(lmm1->lmv_master_mdt_index);
3064         if (OBD_FAIL_CHECK(OBD_FAIL_UNKNOWN_LMV_STRIPE))
3065                 lsm->lsm_md_hash_type = LMV_HASH_TYPE_UNKNOWN;
3066         else
3067                 lsm->lsm_md_hash_type = le32_to_cpu(lmm1->lmv_hash_type);
3068         lsm->lsm_md_layout_version = le32_to_cpu(lmm1->lmv_layout_version);
3069         lsm->lsm_md_migrate_offset = le32_to_cpu(lmm1->lmv_migrate_offset);
3070         lsm->lsm_md_migrate_hash = le32_to_cpu(lmm1->lmv_migrate_hash);
3071         cplen = strlcpy(lsm->lsm_md_pool_name, lmm1->lmv_pool_name,
3072                         sizeof(lsm->lsm_md_pool_name));
3073
3074         if (cplen >= sizeof(lsm->lsm_md_pool_name))
3075                 RETURN(-E2BIG);
3076
3077         CDEBUG(D_INFO, "unpack lsm count %d, master %d hash_type %#x "
3078                "layout_version %d\n", lsm->lsm_md_stripe_count,
3079                lsm->lsm_md_master_mdt_index, lsm->lsm_md_hash_type,
3080                lsm->lsm_md_layout_version);
3081
3082         stripe_count = le32_to_cpu(lmm1->lmv_stripe_count);
3083         for (i = 0; i < stripe_count; i++) {
3084                 fid_le_to_cpu(&lsm->lsm_md_oinfo[i].lmo_fid,
3085                               &lmm1->lmv_stripe_fids[i]);
3086                 /*
3087                  * set default value -1, so lmv_locate_tgt() knows this stripe
3088                  * target is not initialized.
3089                  */
3090                 lsm->lsm_md_oinfo[i].lmo_mds = (u32)-1;
3091                 if (!fid_is_sane(&lsm->lsm_md_oinfo[i].lmo_fid))
3092                         continue;
3093
3094                 rc = lmv_fld_lookup(lmv, &lsm->lsm_md_oinfo[i].lmo_fid,
3095                                     &lsm->lsm_md_oinfo[i].lmo_mds);
3096                 if (rc == -ENOENT)
3097                         continue;
3098
3099                 if (rc)
3100                         RETURN(rc);
3101
3102                 CDEBUG(D_INFO, "unpack fid #%d "DFID"\n", i,
3103                        PFID(&lsm->lsm_md_oinfo[i].lmo_fid));
3104         }
3105
3106         RETURN(rc);
3107 }
3108
3109 static int lmv_unpackmd(struct obd_export *exp, struct lmv_stripe_md **lsmp,
3110                         const union lmv_mds_md *lmm, size_t lmm_size)
3111 {
3112         struct lmv_stripe_md     *lsm;
3113         int                      lsm_size;
3114         int                      rc;
3115         bool                     allocated = false;
3116         ENTRY;
3117
3118         LASSERT(lsmp != NULL);
3119
3120         lsm = *lsmp;
3121         /* Free memmd */
3122         if (lsm != NULL && lmm == NULL) {
3123                 int i;
3124
3125                 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
3126                         if (lsm->lsm_md_oinfo[i].lmo_root)
3127                                 iput(lsm->lsm_md_oinfo[i].lmo_root);
3128                 }
3129                 lsm_size = lmv_stripe_md_size(lsm->lsm_md_stripe_count);
3130                 OBD_FREE(lsm, lsm_size);
3131                 *lsmp = NULL;
3132                 RETURN(0);
3133         }
3134
3135         if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_STRIPE)
3136                 RETURN(-EPERM);
3137
3138         /* Unpack memmd */
3139         if (le32_to_cpu(lmm->lmv_magic) != LMV_MAGIC_V1 &&
3140             le32_to_cpu(lmm->lmv_magic) != LMV_USER_MAGIC) {
3141                 CERROR("%s: invalid lmv magic %x: rc = %d\n",
3142                        exp->exp_obd->obd_name, le32_to_cpu(lmm->lmv_magic),
3143                        -EIO);
3144                 RETURN(-EIO);
3145         }
3146
3147         if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_V1)
3148                 lsm_size = lmv_stripe_md_size(lmv_mds_md_stripe_count_get(lmm));
3149         else
3150                 /**
3151                  * Unpack default dirstripe(lmv_user_md) to lmv_stripe_md,
3152                  * stripecount should be 0 then.
3153                  */
3154                 lsm_size = lmv_stripe_md_size(0);
3155
3156         lsm_size = lmv_stripe_md_size(lmv_mds_md_stripe_count_get(lmm));
3157         if (lsm == NULL) {
3158                 OBD_ALLOC(lsm, lsm_size);
3159                 if (lsm == NULL)
3160                         RETURN(-ENOMEM);
3161                 allocated = true;
3162                 *lsmp = lsm;
3163         }
3164
3165         switch (le32_to_cpu(lmm->lmv_magic)) {
3166         case LMV_MAGIC_V1:
3167                 rc = lmv_unpack_md_v1(exp, lsm, &lmm->lmv_md_v1);
3168                 break;
3169         default:
3170                 CERROR("%s: unrecognized magic %x\n", exp->exp_obd->obd_name,
3171                        le32_to_cpu(lmm->lmv_magic));
3172                 rc = -EINVAL;
3173                 break;
3174         }
3175
3176         if (rc != 0 && allocated) {
3177                 OBD_FREE(lsm, lsm_size);
3178                 *lsmp = NULL;
3179                 lsm_size = rc;
3180         }
3181         RETURN(lsm_size);
3182 }
3183
3184 void lmv_free_memmd(struct lmv_stripe_md *lsm)
3185 {
3186         lmv_unpackmd(NULL, &lsm, NULL, 0);
3187 }
3188 EXPORT_SYMBOL(lmv_free_memmd);
3189
3190 static int lmv_cancel_unused(struct obd_export *exp, const struct lu_fid *fid,
3191                              union ldlm_policy_data *policy,
3192                              enum ldlm_mode mode, enum ldlm_cancel_flags flags,
3193                              void *opaque)
3194 {
3195         struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
3196         int rc = 0;
3197         __u32 i;
3198         ENTRY;
3199
3200         LASSERT(fid != NULL);
3201
3202         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
3203                 struct lmv_tgt_desc *tgt = lmv->tgts[i];
3204                 int err;
3205
3206                 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active)
3207                         continue;
3208
3209                 err = md_cancel_unused(tgt->ltd_exp, fid, policy, mode, flags,
3210                                        opaque);
3211                 if (!rc)
3212                         rc = err;
3213         }
3214         RETURN(rc);
3215 }
3216
3217 static int lmv_set_lock_data(struct obd_export *exp,
3218                              const struct lustre_handle *lockh,
3219                              void *data, __u64 *bits)
3220 {
3221         struct lmv_obd          *lmv = &exp->exp_obd->u.lmv;
3222         struct lmv_tgt_desc     *tgt = lmv->tgts[0];
3223         int                      rc;
3224         ENTRY;
3225
3226         if (tgt == NULL || tgt->ltd_exp == NULL)
3227                 RETURN(-EINVAL);
3228         rc =  md_set_lock_data(tgt->ltd_exp, lockh, data, bits);
3229         RETURN(rc);
3230 }
3231
3232 enum ldlm_mode lmv_lock_match(struct obd_export *exp, __u64 flags,
3233                               const struct lu_fid *fid, enum ldlm_type type,
3234                               union ldlm_policy_data *policy,
3235                               enum ldlm_mode mode, struct lustre_handle *lockh)
3236 {
3237         struct obd_device       *obd = exp->exp_obd;
3238         struct lmv_obd          *lmv = &obd->u.lmv;
3239         enum ldlm_mode          rc;
3240         int                     tgt;
3241         int                     i;
3242         ENTRY;
3243
3244         CDEBUG(D_INODE, "Lock match for "DFID"\n", PFID(fid));
3245
3246         /*
3247          * With DNE every object can have two locks in different namespaces:
3248          * lookup lock in space of MDT storing direntry and update/open lock in
3249          * space of MDT storing inode.  Try the MDT that the FID maps to first,
3250          * since this can be easily found, and only try others if that fails.
3251          */
3252         for (i = 0, tgt = lmv_find_target_index(lmv, fid);
3253              i < lmv->desc.ld_tgt_count;
3254              i++, tgt = (tgt + 1) % lmv->desc.ld_tgt_count) {
3255                 if (tgt < 0) {
3256                         CDEBUG(D_HA, "%s: "DFID" is inaccessible: rc = %d\n",
3257                                obd->obd_name, PFID(fid), tgt);
3258                         tgt = 0;
3259                 }
3260
3261                 if (lmv->tgts[tgt] == NULL ||
3262                     lmv->tgts[tgt]->ltd_exp == NULL ||
3263                     lmv->tgts[tgt]->ltd_active == 0)
3264                         continue;
3265
3266                 rc = md_lock_match(lmv->tgts[tgt]->ltd_exp, flags, fid,
3267                                    type, policy, mode, lockh);
3268                 if (rc)
3269                         RETURN(rc);
3270         }
3271
3272         RETURN(0);
3273 }
3274
3275 int lmv_get_lustre_md(struct obd_export *exp, struct ptlrpc_request *req,
3276                       struct obd_export *dt_exp, struct obd_export *md_exp,
3277                       struct lustre_md *md)
3278 {
3279         struct lmv_obd          *lmv = &exp->exp_obd->u.lmv;
3280         struct lmv_tgt_desc     *tgt = lmv->tgts[0];
3281
3282         if (tgt == NULL || tgt->ltd_exp == NULL)
3283                 RETURN(-EINVAL);
3284
3285         return md_get_lustre_md(lmv->tgts[0]->ltd_exp, req, dt_exp, md_exp, md);
3286 }
3287
3288 int lmv_free_lustre_md(struct obd_export *exp, struct lustre_md *md)
3289 {
3290         struct obd_device       *obd = exp->exp_obd;
3291         struct lmv_obd          *lmv = &obd->u.lmv;
3292         struct lmv_tgt_desc     *tgt = lmv->tgts[0];
3293         ENTRY;
3294
3295         if (md->lmv != NULL) {
3296                 lmv_free_memmd(md->lmv);
3297                 md->lmv = NULL;
3298         }
3299         if (tgt == NULL || tgt->ltd_exp == NULL)
3300                 RETURN(-EINVAL);
3301         RETURN(md_free_lustre_md(lmv->tgts[0]->ltd_exp, md));
3302 }
3303
3304 int lmv_set_open_replay_data(struct obd_export *exp,
3305                              struct obd_client_handle *och,
3306                              struct lookup_intent *it)
3307 {
3308         struct obd_device       *obd = exp->exp_obd;
3309         struct lmv_obd          *lmv = &obd->u.lmv;
3310         struct lmv_tgt_desc     *tgt;
3311         ENTRY;
3312
3313         tgt = lmv_find_target(lmv, &och->och_fid);
3314         if (IS_ERR(tgt))
3315                 RETURN(PTR_ERR(tgt));
3316
3317         RETURN(md_set_open_replay_data(tgt->ltd_exp, och, it));
3318 }
3319
3320 int lmv_clear_open_replay_data(struct obd_export *exp,
3321                                struct obd_client_handle *och)
3322 {
3323         struct obd_device       *obd = exp->exp_obd;
3324         struct lmv_obd          *lmv = &obd->u.lmv;
3325         struct lmv_tgt_desc     *tgt;
3326         ENTRY;
3327
3328         tgt = lmv_find_target(lmv, &och->och_fid);
3329         if (IS_ERR(tgt))
3330                 RETURN(PTR_ERR(tgt));
3331
3332         RETURN(md_clear_open_replay_data(tgt->ltd_exp, och));
3333 }
3334
3335 int lmv_intent_getattr_async(struct obd_export *exp,
3336                              struct md_enqueue_info *minfo)
3337 {
3338         struct md_op_data *op_data = &minfo->mi_data;
3339         struct obd_device *obd = exp->exp_obd;
3340         struct lmv_obd *lmv = &obd->u.lmv;
3341         struct lmv_tgt_desc *ptgt;
3342         struct lmv_tgt_desc *ctgt;
3343         int rc;
3344
3345         ENTRY;
3346
3347         if (!fid_is_sane(&op_data->op_fid2))
3348                 RETURN(-EINVAL);
3349
3350         ptgt = lmv_locate_tgt(lmv, op_data, &op_data->op_fid1);
3351         if (IS_ERR(ptgt))
3352                 RETURN(PTR_ERR(ptgt));
3353
3354         ctgt = lmv_find_target(lmv, &op_data->op_fid2);
3355         if (IS_ERR(ctgt))
3356                 RETURN(PTR_ERR(ctgt));
3357
3358         /* remote object needs two RPCs to lookup and getattr, considering the
3359          * complexity, don't support statahead for now.
3360          */
3361         if (ptgt != ctgt)
3362                 RETURN(-EREMOTE);
3363
3364         rc = md_intent_getattr_async(ptgt->ltd_exp, minfo);
3365
3366         RETURN(rc);
3367 }
3368
3369 int lmv_revalidate_lock(struct obd_export *exp, struct lookup_intent *it,
3370                         struct lu_fid *fid, __u64 *bits)
3371 {
3372         struct obd_device       *obd = exp->exp_obd;
3373         struct lmv_obd          *lmv = &obd->u.lmv;
3374         struct lmv_tgt_desc     *tgt;
3375         int                      rc;
3376         ENTRY;
3377
3378         tgt = lmv_find_target(lmv, fid);
3379         if (IS_ERR(tgt))
3380                 RETURN(PTR_ERR(tgt));
3381
3382         rc = md_revalidate_lock(tgt->ltd_exp, it, fid, bits);
3383         RETURN(rc);
3384 }
3385
3386 int lmv_get_fid_from_lsm(struct obd_export *exp,
3387                          const struct lmv_stripe_md *lsm,
3388                          const char *name, int namelen, struct lu_fid *fid)
3389 {
3390         const struct lmv_oinfo *oinfo;
3391
3392         LASSERT(lsm != NULL);
3393         oinfo = lsm_name_to_stripe_info(lsm, name, namelen, false);
3394         if (IS_ERR(oinfo))
3395                 return PTR_ERR(oinfo);
3396
3397         *fid = oinfo->lmo_fid;
3398
3399         RETURN(0);
3400 }
3401
3402 /**
3403  * For lmv, only need to send request to master MDT, and the master MDT will
3404  * process with other slave MDTs. The only exception is Q_GETOQUOTA for which
3405  * we directly fetch data from the slave MDTs.
3406  */
3407 int lmv_quotactl(struct obd_device *unused, struct obd_export *exp,
3408                  struct obd_quotactl *oqctl)
3409 {
3410         struct obd_device   *obd = class_exp2obd(exp);
3411         struct lmv_obd      *lmv = &obd->u.lmv;
3412         struct lmv_tgt_desc *tgt = lmv->tgts[0];
3413         int                  rc = 0;
3414         __u32                i;
3415         __u64                curspace, curinodes;
3416         ENTRY;
3417
3418         if (tgt == NULL ||
3419             tgt->ltd_exp == NULL ||
3420             !tgt->ltd_active ||
3421             lmv->desc.ld_tgt_count == 0) {
3422                 CERROR("master lmv inactive\n");
3423                 RETURN(-EIO);
3424         }
3425
3426         if (oqctl->qc_cmd != Q_GETOQUOTA) {
3427                 rc = obd_quotactl(tgt->ltd_exp, oqctl);
3428                 RETURN(rc);
3429         }
3430
3431         curspace = curinodes = 0;
3432         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
3433                 int err;
3434                 tgt = lmv->tgts[i];
3435
3436                 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active)
3437                         continue;
3438
3439                 err = obd_quotactl(tgt->ltd_exp, oqctl);
3440                 if (err) {
3441                         CERROR("getquota on mdt %d failed. %d\n", i, err);
3442                         if (!rc)
3443                                 rc = err;
3444                 } else {
3445                         curspace += oqctl->qc_dqblk.dqb_curspace;
3446                         curinodes += oqctl->qc_dqblk.dqb_curinodes;
3447                 }
3448         }
3449         oqctl->qc_dqblk.dqb_curspace = curspace;
3450         oqctl->qc_dqblk.dqb_curinodes = curinodes;
3451
3452         RETURN(rc);
3453 }
3454
3455 static int lmv_merge_attr(struct obd_export *exp,
3456                           const struct lu_fid *fid,
3457                           const struct lmv_stripe_md *lsm,
3458                           struct cl_attr *attr,
3459                           ldlm_blocking_callback cb_blocking)
3460 {
3461         int rc;
3462         int i;
3463
3464         rc = lmv_revalidate_slaves(exp, fid, lsm, cb_blocking, 0);
3465         if (rc < 0)
3466                 return rc;
3467
3468         for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
3469                 struct inode *inode = lsm->lsm_md_oinfo[i].lmo_root;
3470
3471                 if (!inode)
3472                         continue;
3473
3474                 CDEBUG(D_INFO,
3475                        "" DFID " size %llu, blocks %llu nlink %u, atime %lld ctime %lld, mtime %lld.\n",
3476                        PFID(&lsm->lsm_md_oinfo[i].lmo_fid),
3477                        i_size_read(inode), (unsigned long long)inode->i_blocks,
3478                        inode->i_nlink, (s64)inode->i_atime.tv_sec,
3479                        (s64)inode->i_ctime.tv_sec, (s64)inode->i_mtime.tv_sec);
3480
3481                 /* for slave stripe, it needs to subtract nlink for . and .. */
3482                 if (i != 0)
3483                         attr->cat_nlink += inode->i_nlink - 2;
3484                 else
3485                         attr->cat_nlink = inode->i_nlink;
3486
3487                 attr->cat_size += i_size_read(inode);
3488                 attr->cat_blocks += inode->i_blocks;
3489
3490                 if (attr->cat_atime < inode->i_atime.tv_sec)
3491                         attr->cat_atime = inode->i_atime.tv_sec;
3492
3493                 if (attr->cat_ctime < inode->i_ctime.tv_sec)
3494                         attr->cat_ctime = inode->i_ctime.tv_sec;
3495
3496                 if (attr->cat_mtime < inode->i_mtime.tv_sec)
3497                         attr->cat_mtime = inode->i_mtime.tv_sec;
3498         }
3499         return 0;
3500 }
3501
3502 struct obd_ops lmv_obd_ops = {
3503         .o_owner                = THIS_MODULE,
3504         .o_setup                = lmv_setup,
3505         .o_cleanup              = lmv_cleanup,
3506         .o_precleanup           = lmv_precleanup,
3507         .o_process_config       = lmv_process_config,
3508         .o_connect              = lmv_connect,
3509         .o_disconnect           = lmv_disconnect,
3510         .o_statfs               = lmv_statfs,
3511         .o_get_info             = lmv_get_info,
3512         .o_set_info_async       = lmv_set_info_async,
3513         .o_notify               = lmv_notify,
3514         .o_get_uuid             = lmv_get_uuid,
3515         .o_iocontrol            = lmv_iocontrol,
3516         .o_quotactl             = lmv_quotactl
3517 };
3518
3519 struct md_ops lmv_md_ops = {
3520         .m_get_root             = lmv_get_root,
3521         .m_null_inode           = lmv_null_inode,
3522         .m_close                = lmv_close,
3523         .m_create               = lmv_create,
3524         .m_enqueue              = lmv_enqueue,
3525         .m_getattr              = lmv_getattr,
3526         .m_getxattr             = lmv_getxattr,
3527         .m_getattr_name         = lmv_getattr_name,
3528         .m_intent_lock          = lmv_intent_lock,
3529         .m_link                 = lmv_link,
3530         .m_rename               = lmv_rename,
3531         .m_setattr              = lmv_setattr,
3532         .m_setxattr             = lmv_setxattr,
3533         .m_fsync                = lmv_fsync,
3534         .m_file_resync          = lmv_file_resync,
3535         .m_read_page            = lmv_read_page,
3536         .m_unlink               = lmv_unlink,
3537         .m_init_ea_size         = lmv_init_ea_size,
3538         .m_cancel_unused        = lmv_cancel_unused,
3539         .m_set_lock_data        = lmv_set_lock_data,
3540         .m_lock_match           = lmv_lock_match,
3541         .m_get_lustre_md        = lmv_get_lustre_md,
3542         .m_free_lustre_md       = lmv_free_lustre_md,
3543         .m_merge_attr           = lmv_merge_attr,
3544         .m_set_open_replay_data = lmv_set_open_replay_data,
3545         .m_clear_open_replay_data = lmv_clear_open_replay_data,
3546         .m_intent_getattr_async = lmv_intent_getattr_async,
3547         .m_revalidate_lock      = lmv_revalidate_lock,
3548         .m_get_fid_from_lsm     = lmv_get_fid_from_lsm,
3549         .m_unpackmd             = lmv_unpackmd,
3550         .m_rmfid                = lmv_rmfid,
3551 };
3552
3553 static int __init lmv_init(void)
3554 {
3555         return class_register_type(&lmv_obd_ops, &lmv_md_ops, true, NULL,
3556                                    LUSTRE_LMV_NAME, NULL);
3557 }
3558
3559 static void __exit lmv_exit(void)
3560 {
3561         class_unregister_type(LUSTRE_LMV_NAME);
3562 }
3563
3564 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
3565 MODULE_DESCRIPTION("Lustre Logical Metadata Volume");
3566 MODULE_VERSION(LUSTRE_VERSION_STRING);
3567 MODULE_LICENSE("GPL");
3568
3569 module_init(lmv_init);
3570 module_exit(lmv_exit);