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  */
  31
  32 #define DEBUG_SUBSYSTEM S_LMV
  33
  34 #include <linux/file.h>
  35 #include <linux/module.h>
  36 #include <linux/init.h>
  37 #include <linux/user_namespace.h>
  38 #include <linux/uidgid.h>
  39 #include <linux/slab.h>
  40 #include <linux/pagemap.h>
  41 #include <linux/mm.h>
  42 #include <linux/math64.h>
  43 #include <linux/seq_file.h>
  44 #include <linux/namei.h>
  45
  46 #include <obd_support.h>
  47 #include <lustre_lib.h>
  48 #include <lustre_net.h>
  49 #include <obd_class.h>
  50 #include <lustre_lmv.h>
  51 #include <lprocfs_status.h>
  52 #include <cl_object.h>
  53 #include <lustre_fid.h>
  54 #include <uapi/linux/lustre/lustre_ioctl.h>
  55 #include <lustre_ioctl_old.h>
  56 #include <lustre_kernelcomm.h>
  57 #include "lmv_internal.h"
  58
  59 static int lmv_check_connect(struct obd_device *obd);
  60 static inline bool lmv_op_default_rr_mkdir(const struct md_op_data *op_data);
  61
  62 void lmv_activate_target(struct lmv_obd *lmv, struct lmv_tgt_desc *tgt,
  63                          int activate)
  64 {
  65         if (tgt->ltd_active == activate)
  66                 return;
  67
  68         tgt->ltd_active = activate;
  69         lmv->lmv_mdt_descs.ltd_lmv_desc.ld_active_tgt_count +=
  70                 (activate ? 1 : -1);
  71
  72         tgt->ltd_exp->exp_obd->obd_inactive = !activate;
  73 }
  74
  75 /**
  76  * Error codes:
  77  *
  78  *  -EINVAL  : UUID can't be found in the LMV's target list
  79  *  -ENOTCONN: The UUID is found, but the target connection is bad (!)
  80  *  -EBADF   : The UUID is found, but the OBD of the wrong type (!)
  81  */
  82 static int lmv_set_mdc_active(struct lmv_obd *lmv,
  83                               const struct obd_uuid *uuid,
  84                               int activate)
  85 {
  86         struct lu_tgt_desc *tgt = NULL;
  87         struct obd_device *obd;
  88         int rc = 0;
  89
  90         ENTRY;
  91
  92         CDEBUG(D_INFO, "Searching in lmv %p for uuid %s (activate=%d)\n",
  93                         lmv, uuid->uuid, activate);
  94
  95         spin_lock(&lmv->lmv_lock);
  96         lmv_foreach_connected_tgt(lmv, tgt) {
  97                 CDEBUG(D_INFO, "Target idx %d is %s conn %#llx\n",
  98                        tgt->ltd_index, tgt->ltd_uuid.uuid,
  99                        tgt->ltd_exp->exp_handle.h_cookie);
 100
 101                 if (obd_uuid_equals(uuid, &tgt->ltd_uuid))
 102                         break;
 103         }
 104
 105         if (!tgt)
 106                 GOTO(out_lmv_lock, rc = -EINVAL);
 107
 108         obd = class_exp2obd(tgt->ltd_exp);
 109         if (obd == NULL)
 110                 GOTO(out_lmv_lock, rc = -ENOTCONN);
 111
 112         CDEBUG(D_INFO, "Found OBD %s=%s device %d (%p) type %s at LMV idx %d\n",
 113                obd->obd_name, obd->obd_uuid.uuid, obd->obd_minor, obd,
 114                obd->obd_type->typ_name, tgt->ltd_index);
 115         LASSERT(strcmp(obd->obd_type->typ_name, LUSTRE_MDC_NAME) == 0);
 116
 117         if (tgt->ltd_active == activate) {
 118                 CDEBUG(D_INFO, "OBD %p already %sactive!\n", obd,
 119                        activate ? "" : "in");
 120                 GOTO(out_lmv_lock, rc);
 121         }
 122
 123         CDEBUG(D_INFO, "Marking OBD %p %sactive\n", obd,
 124                activate ? "" : "in");
 125         lmv_activate_target(lmv, tgt, activate);
 126         EXIT;
 127
 128  out_lmv_lock:
 129         spin_unlock(&lmv->lmv_lock);
 130         return rc;
 131 }
 132
 133 static struct obd_uuid *lmv_get_uuid(struct obd_export *exp)
 134 {
 135         struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
 136         struct lmv_tgt_desc *tgt = lmv_tgt(lmv, 0);
 137
 138         return tgt ? obd_get_uuid(tgt->ltd_exp) : NULL;
 139 }
 140
 141 static int lmv_notify(struct obd_device *obd, struct obd_device *watched,
 142                       enum obd_notify_event ev)
 143 {
 144         struct obd_connect_data *conn_data;
 145         struct lmv_obd          *lmv = &obd->u.lmv;
 146         struct obd_uuid         *uuid;
 147         int                      rc = 0;
 148         ENTRY;
 149
 150         if (strcmp(watched->obd_type->typ_name, LUSTRE_MDC_NAME)) {
 151                 CERROR("unexpected notification of %s %s!\n",
 152                        watched->obd_type->typ_name,
 153                        watched->obd_name);
 154                 RETURN(-EINVAL);
 155         }
 156
 157         uuid = &watched->u.cli.cl_target_uuid;
 158         if (ev == OBD_NOTIFY_ACTIVE || ev == OBD_NOTIFY_INACTIVE) {
 159                 /*
 160                  * Set MDC as active before notifying the observer, so the
 161                  * observer can use the MDC normally.
 162                  */
 163                 rc = lmv_set_mdc_active(lmv, uuid,
 164                                         ev == OBD_NOTIFY_ACTIVE);
 165                 if (rc) {
 166                         CERROR("%sactivation of %s failed: %d\n",
 167                                ev == OBD_NOTIFY_ACTIVE ? "" : "de",
 168                                uuid->uuid, rc);
 169                         RETURN(rc);
 170                 }
 171         } else if (ev == OBD_NOTIFY_OCD) {
 172                 conn_data = &watched->u.cli.cl_import->imp_connect_data;
 173                 /*
 174                  * XXX: Make sure that ocd_connect_flags from all targets are
 175                  * the same. Otherwise one of MDTs runs wrong version or
 176                  * something like this.  --umka
 177                  */
 178                 obd->obd_self_export->exp_connect_data = *conn_data;
 179         }
 180
 181         /*
 182          * Pass the notification up the chain.
 183          */
 184         if (obd->obd_observer)
 185                 rc = obd_notify(obd->obd_observer, watched, ev);
 186
 187         RETURN(rc);
 188 }
 189
 190 static int lmv_connect(const struct lu_env *env,
 191                        struct obd_export **pexp, struct obd_device *obd,
 192                        struct obd_uuid *cluuid, struct obd_connect_data *data,
 193                        void *localdata)
 194 {
 195         struct lmv_obd *lmv = &obd->u.lmv;
 196         struct lustre_handle conn = { 0 };
 197         struct obd_export *exp;
 198         int rc;
 199         ENTRY;
 200
 201         rc = class_connect(&conn, obd, cluuid);
 202         if (rc) {
 203                 CERROR("class_connection() returned %d\n", rc);
 204                 RETURN(rc);
 205         }
 206
 207         exp = class_conn2export(&conn);
 208
 209         lmv->connected = 0;
 210         lmv->conn_data = *data;
 211         lmv->lmv_cache = localdata;
 212
 213         lmv->lmv_tgts_kobj = kobject_create_and_add("target_obds",
 214                                                     &obd->obd_kset.kobj);
 215         if (!lmv->lmv_tgts_kobj) {
 216                 CERROR("%s: cannot create /sys/fs/lustre/%s/%s/target_obds\n",
 217                        obd->obd_name, obd->obd_type->typ_name, obd->obd_name);
 218         }
 219
 220         rc = lmv_check_connect(obd);
 221         if (rc != 0)
 222                 GOTO(out_sysfs, rc);
 223
 224         *pexp = exp;
 225
 226         RETURN(rc);
 227
 228 out_sysfs:
 229         if (lmv->lmv_tgts_kobj)
 230                 kobject_put(lmv->lmv_tgts_kobj);
 231
 232         class_disconnect(exp);
 233
 234         return rc;
 235 }
 236
 237 static int lmv_init_ea_size(struct obd_export *exp, __u32 easize,
 238                             __u32 def_easize)
 239 {
 240         struct obd_device *obd = exp->exp_obd;
 241         struct lmv_obd *lmv = &obd->u.lmv;
 242         struct lmv_tgt_desc *tgt;
 243         int change = 0;
 244         int rc = 0;
 245
 246         ENTRY;
 247
 248         if (lmv->max_easize < easize) {
 249                 lmv->max_easize = easize;
 250                 change = 1;
 251         }
 252         if (lmv->max_def_easize < def_easize) {
 253                 lmv->max_def_easize = def_easize;
 254                 change = 1;
 255         }
 256
 257         if (change == 0)
 258                 RETURN(0);
 259
 260         if (lmv->connected == 0)
 261                 RETURN(0);
 262
 263         lmv_foreach_connected_tgt(lmv, tgt) {
 264                 if (!tgt->ltd_active)
 265                         continue;
 266
 267                 rc = md_init_ea_size(tgt->ltd_exp, easize, def_easize);
 268                 if (rc) {
 269                         CERROR("%s: obd_init_ea_size() failed on MDT target %d:"
 270                                " rc = %d\n", obd->obd_name, tgt->ltd_index, rc);
 271                         break;
 272                 }
 273         }
 274         RETURN(rc);
 275 }
 276
 277 #define MAX_STRING_SIZE 128
 278
 279 static int lmv_connect_mdc(struct obd_device *obd, struct lmv_tgt_desc *tgt)
 280 {
 281         struct lmv_obd *lmv = &obd->u.lmv;
 282         struct obd_device *mdc_obd;
 283         struct obd_export *mdc_exp;
 284         struct lu_fld_target target;
 285         int  rc;
 286         ENTRY;
 287
 288         mdc_obd = class_find_client_obd(&tgt->ltd_uuid, LUSTRE_MDC_NAME,
 289                                         &obd->obd_uuid);
 290         if (!mdc_obd) {
 291                 CERROR("target %s not attached\n", tgt->ltd_uuid.uuid);
 292                 RETURN(-EINVAL);
 293         }
 294
 295         CDEBUG(D_CONFIG, "connect to %s(%s) - %s, %s\n",
 296                mdc_obd->obd_name, mdc_obd->obd_uuid.uuid,
 297                tgt->ltd_uuid.uuid, obd->obd_uuid.uuid);
 298
 299         if (!mdc_obd->obd_set_up) {
 300                 CERROR("target %s is not set up\n", tgt->ltd_uuid.uuid);
 301                 RETURN(-EINVAL);
 302         }
 303
 304         rc = obd_connect(NULL, &mdc_exp, mdc_obd, &obd->obd_uuid,
 305                          &lmv->conn_data, lmv->lmv_cache);
 306         if (rc) {
 307                 CERROR("target %s connect error %d\n", tgt->ltd_uuid.uuid, rc);
 308                 RETURN(rc);
 309         }
 310
 311         /*
 312          * Init fid sequence client for this mdc and add new fld target.
 313          */
 314         rc = obd_fid_init(mdc_obd, mdc_exp, LUSTRE_SEQ_METADATA);
 315         if (rc)
 316                 RETURN(rc);
 317
 318         target.ft_srv = NULL;
 319         target.ft_exp = mdc_exp;
 320         target.ft_idx = tgt->ltd_index;
 321
 322         fld_client_add_target(&lmv->lmv_fld, &target);
 323
 324         rc = obd_register_observer(mdc_obd, obd);
 325         if (rc) {
 326                 obd_disconnect(mdc_exp);
 327                 CERROR("target %s register_observer error %d\n",
 328                        tgt->ltd_uuid.uuid, rc);
 329                 RETURN(rc);
 330         }
 331
 332         if (obd->obd_observer) {
 333                 /*
 334                  * Tell the observer about the new target.
 335                  */
 336                 rc = obd_notify(obd->obd_observer, mdc_exp->exp_obd,
 337                                 OBD_NOTIFY_ACTIVE);
 338                 if (rc) {
 339                         obd_disconnect(mdc_exp);
 340                         RETURN(rc);
 341                 }
 342         }
 343
 344         tgt->ltd_active = 1;
 345         tgt->ltd_exp = mdc_exp;
 346         lmv->lmv_mdt_descs.ltd_lmv_desc.ld_active_tgt_count++;
 347
 348         md_init_ea_size(tgt->ltd_exp, lmv->max_easize, lmv->max_def_easize);
 349
 350         rc = lu_qos_add_tgt(&lmv->lmv_qos, tgt);
 351         if (rc) {
 352                 obd_disconnect(mdc_exp);
 353                 RETURN(rc);
 354         }
 355
 356         CDEBUG(D_CONFIG, "Connected to %s(%s) successfully (%d)\n",
 357                mdc_obd->obd_name, mdc_obd->obd_uuid.uuid,
 358                atomic_read(&obd->obd_refcount));
 359
 360         lmv_statfs_check_update(obd, tgt);
 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, struct lu_tgt_desc *tgt)
 371 {
 372         LASSERT(tgt);
 373         ltd_del_tgt(&lmv->lmv_mdt_descs, tgt);
 374         OBD_FREE_PTR(tgt);
 375 }
 376
 377 static int lmv_add_target(struct obd_device *obd, struct obd_uuid *uuidp,
 378                            __u32 index, int gen)
 379 {
 380         struct obd_device *mdc_obd;
 381         struct lmv_obd *lmv = &obd->u.lmv;
 382         struct lmv_tgt_desc *tgt;
 383         struct lu_tgt_descs *ltd = &lmv->lmv_mdt_descs;
 384         int rc = 0;
 385
 386         ENTRY;
 387
 388         CDEBUG(D_CONFIG, "Target uuid: %s. index %d\n", uuidp->uuid, index);
 389         mdc_obd = class_find_client_obd(uuidp, LUSTRE_MDC_NAME,
 390                                         &obd->obd_uuid);
 391         if (!mdc_obd) {
 392                 CERROR("%s: Target %s not attached: rc = %d\n",
 393                        obd->obd_name, uuidp->uuid, -EINVAL);
 394                 RETURN(-EINVAL);
 395         }
 396
 397         OBD_ALLOC_PTR(tgt);
 398         if (!tgt)
 399                 RETURN(-ENOMEM);
 400
 401         mutex_init(&tgt->ltd_fid_mutex);
 402         tgt->ltd_index = index;
 403         tgt->ltd_uuid = *uuidp;
 404         tgt->ltd_active = 0;
 405
 406         mutex_lock(&ltd->ltd_mutex);
 407         rc = ltd_add_tgt(ltd, tgt);
 408         mutex_unlock(&ltd->ltd_mutex);
 409
 410         if (rc)
 411                 GOTO(out_tgt, rc);
 412
 413         if (!lmv->connected)
 414                 /* lmv_check_connect() will connect this target. */
 415                 RETURN(0);
 416
 417         rc = lmv_connect_mdc(obd, tgt);
 418         if (!rc) {
 419                 int easize = sizeof(struct lmv_stripe_md) +
 420                         lmv->lmv_mdt_count * sizeof(struct lu_fid);
 421
 422                 lmv_init_ea_size(obd->obd_self_export, easize, 0);
 423         }
 424
 425         RETURN(rc);
 426
 427 out_tgt:
 428         OBD_FREE_PTR(tgt);
 429         return rc;
 430 }
 431
 432 static int lmv_check_connect(struct obd_device *obd)
 433 {
 434         struct lmv_obd *lmv = &obd->u.lmv;
 435         struct lmv_tgt_desc *tgt;
 436         int easize;
 437         int rc;
 438
 439         ENTRY;
 440
 441         if (lmv->connected)
 442                 RETURN(0);
 443
 444         mutex_lock(&lmv->lmv_mdt_descs.ltd_mutex);
 445         if (lmv->connected)
 446                 GOTO(unlock, rc = 0);
 447
 448         if (!lmv->lmv_mdt_count) {
 449                 CERROR("%s: no targets configured: rc = -EINVAL\n",
 450                        obd->obd_name);
 451                 GOTO(unlock, rc = -EINVAL);
 452         }
 453
 454         if (!lmv_mdt0_inited(lmv)) {
 455                 CERROR("%s: no target configured for index 0: rc = -EINVAL.\n",
 456                        obd->obd_name);
 457                 GOTO(unlock, rc = -EINVAL);
 458         }
 459
 460         CDEBUG(D_CONFIG, "Time to connect %s to %s\n",
 461                obd->obd_uuid.uuid, obd->obd_name);
 462
 463         lmv_foreach_tgt(lmv, tgt) {
 464                 rc = lmv_connect_mdc(obd, tgt);
 465                 if (rc)
 466                         GOTO(out_disc, rc);
 467         }
 468
 469         lmv->connected = 1;
 470         easize = lmv_mds_md_size(lmv->lmv_mdt_count, LMV_MAGIC);
 471         lmv_init_ea_size(obd->obd_self_export, easize, 0);
 472         EXIT;
 473 unlock:
 474         mutex_unlock(&lmv->lmv_mdt_descs.ltd_mutex);
 475
 476         return rc;
 477
 478 out_disc:
 479         lmv_foreach_tgt(lmv, tgt) {
 480                 tgt->ltd_active = 0;
 481                 if (!tgt->ltd_exp)
 482                         continue;
 483
 484                 --lmv->lmv_mdt_descs.ltd_lmv_desc.ld_active_tgt_count;
 485                 obd_disconnect(tgt->ltd_exp);
 486         }
 487
 488         goto unlock;
 489 }
 490
 491 static int lmv_disconnect_mdc(struct obd_device *obd, struct lmv_tgt_desc *tgt)
 492 {
 493         struct lmv_obd *lmv = &obd->u.lmv;
 494         struct obd_device *mdc_obd;
 495         int rc;
 496         ENTRY;
 497
 498         LASSERT(tgt != NULL);
 499         LASSERT(obd != NULL);
 500
 501         mdc_obd = class_exp2obd(tgt->ltd_exp);
 502
 503         if (mdc_obd) {
 504                 mdc_obd->obd_force = obd->obd_force;
 505                 mdc_obd->obd_fail = obd->obd_fail;
 506                 mdc_obd->obd_no_recov = obd->obd_no_recov;
 507
 508                 if (lmv->lmv_tgts_kobj)
 509                         sysfs_remove_link(lmv->lmv_tgts_kobj,
 510                                           mdc_obd->obd_name);
 511         }
 512
 513         rc = lu_qos_del_tgt(&lmv->lmv_qos, tgt);
 514         if (rc)
 515                 CERROR("%s: Can't del target from QoS table: rc = %d\n",
 516                        tgt->ltd_exp->exp_obd->obd_name, rc);
 517
 518         rc = fld_client_del_target(&lmv->lmv_fld, tgt->ltd_index);
 519         if (rc)
 520                 CERROR("%s: Can't del fld targets: rc = %d\n",
 521                        tgt->ltd_exp->exp_obd->obd_name, rc);
 522
 523         rc = obd_fid_fini(tgt->ltd_exp->exp_obd);
 524         if (rc)
 525                 CERROR("%s: Can't finalize fids factory: rc = %d\n",
 526                        tgt->ltd_exp->exp_obd->obd_name, rc);
 527
 528         CDEBUG(D_INFO, "Disconnected from %s(%s) successfully\n",
 529                tgt->ltd_exp->exp_obd->obd_name,
 530                tgt->ltd_exp->exp_obd->obd_uuid.uuid);
 531
 532         lmv_activate_target(lmv, tgt, 0);
 533         obd_register_observer(tgt->ltd_exp->exp_obd, NULL);
 534         rc = obd_disconnect(tgt->ltd_exp);
 535         if (rc) {
 536                 CERROR("%s: Target %s disconnect error: rc = %d\n",
 537                        tgt->ltd_exp->exp_obd->obd_name,
 538                        tgt->ltd_uuid.uuid, rc);
 539         }
 540         tgt->ltd_exp = NULL;
 541         RETURN(0);
 542 }
 543
 544 static int lmv_disconnect(struct obd_export *exp)
 545 {
 546         struct obd_device *obd = class_exp2obd(exp);
 547         struct lmv_obd *lmv = &obd->u.lmv;
 548         struct lmv_tgt_desc *tgt;
 549         int rc;
 550
 551         ENTRY;
 552
 553         lmv_foreach_connected_tgt(lmv, tgt)
 554                 lmv_disconnect_mdc(obd, tgt);
 555
 556         if (lmv->lmv_tgts_kobj)
 557                 kobject_put(lmv->lmv_tgts_kobj);
 558
 559         lmv->connected = 0;
 560         rc = class_disconnect(exp);
 561
 562         RETURN(rc);
 563 }
 564
 565 static int lmv_fid2path(struct obd_export *exp, int len, void *karg,
 566                         void __user *uarg)
 567 {
 568         struct obd_device *obd = class_exp2obd(exp);
 569         struct lmv_obd *lmv = &obd->u.lmv;
 570         struct getinfo_fid2path *gf;
 571         struct lmv_tgt_desc *tgt;
 572         struct getinfo_fid2path *remote_gf = NULL;
 573         struct lu_fid root_fid;
 574         int remote_gf_size = 0;
 575         int currentisenc = 0;
 576         int globalisenc = 0;
 577         int rc;
 578
 579         gf = karg;
 580         tgt = lmv_fid2tgt(lmv, &gf->gf_fid);
 581         if (IS_ERR(tgt))
 582                 RETURN(PTR_ERR(tgt));
 583
 584         root_fid = *gf->gf_u.gf_root_fid;
 585         LASSERT(fid_is_sane(&root_fid));
 586
 587 repeat_fid2path:
 588         rc = obd_iocontrol(OBD_IOC_FID2PATH, tgt->ltd_exp, len, gf, uarg);
 589         if (rc != 0 && rc != -EREMOTE)
 590                 GOTO(out_fid2path, rc);
 591
 592         if (gf->gf_u.gf_path[0] == '/') {
 593                 /* by convention, server side (mdt_path_current()) puts
 594                  * a leading '/' to tell client that we are dealing with
 595                  * an encrypted file
 596                  */
 597                 currentisenc = 1;
 598                 globalisenc = 1;
 599         } else {
 600                 currentisenc = 0;
 601         }
 602
 603         /* If remote_gf != NULL, it means just building the
 604          * path on the remote MDT, copy this path segment to gf.
 605          */
 606         if (remote_gf != NULL) {
 607                 struct getinfo_fid2path *ori_gf;
 608                 int oldisenc = 0;
 609                 char *ptr;
 610                 int len;
 611
 612                 ori_gf = (struct getinfo_fid2path *)karg;
 613                 if (strlen(ori_gf->gf_u.gf_path) + 1 +
 614                     strlen(gf->gf_u.gf_path) + 1 > ori_gf->gf_pathlen)
 615                         GOTO(out_fid2path, rc = -EOVERFLOW);
 616
 617                 ptr = ori_gf->gf_u.gf_path;
 618                 oldisenc = ptr[0] == '/';
 619
 620                 len = strlen(gf->gf_u.gf_path);
 621                 if (len) {
 622                         /* move the current path to the right to release space
 623                          * for closer-to-root part
 624                          */
 625                         memmove(ptr + len - currentisenc + 1 + globalisenc,
 626                                 ptr + oldisenc,
 627                                 strlen(ori_gf->gf_u.gf_path) - oldisenc + 1);
 628                         if (globalisenc)
 629                                 *(ptr++) = '/';
 630                         memcpy(ptr, gf->gf_u.gf_path + currentisenc,
 631                                len - currentisenc);
 632                         ptr[len - currentisenc] = '/';
 633                 }
 634         }
 635
 636         CDEBUG(D_INFO, "%s: get path %s "DFID" rec: %llu ln: %u\n",
 637                tgt->ltd_exp->exp_obd->obd_name,
 638                gf->gf_u.gf_path, PFID(&gf->gf_fid), gf->gf_recno,
 639                gf->gf_linkno);
 640
 641         if (rc == 0)
 642                 GOTO(out_fid2path, rc);
 643
 644         /* sigh, has to go to another MDT to do path building further */
 645         if (remote_gf == NULL) {
 646                 remote_gf_size = sizeof(*remote_gf) + len - sizeof(*gf);
 647                 OBD_ALLOC(remote_gf, remote_gf_size);
 648                 if (remote_gf == NULL)
 649                         GOTO(out_fid2path, rc = -ENOMEM);
 650                 remote_gf->gf_pathlen = len - sizeof(*gf);
 651         }
 652
 653         if (!fid_is_sane(&gf->gf_fid)) {
 654                 CERROR("%s: invalid FID "DFID": rc = %d\n",
 655                        tgt->ltd_exp->exp_obd->obd_name,
 656                        PFID(&gf->gf_fid), -EINVAL);
 657                 GOTO(out_fid2path, rc = -EINVAL);
 658         }
 659
 660         tgt = lmv_fid2tgt(lmv, &gf->gf_fid);
 661         if (IS_ERR(tgt))
 662                 GOTO(out_fid2path, rc = -EINVAL);
 663
 664         remote_gf->gf_fid = gf->gf_fid;
 665         remote_gf->gf_recno = -1;
 666         remote_gf->gf_linkno = -1;
 667         memset(remote_gf->gf_u.gf_path, 0, remote_gf->gf_pathlen);
 668         *remote_gf->gf_u.gf_root_fid = root_fid;
 669         gf = remote_gf;
 670         goto repeat_fid2path;
 671
 672 out_fid2path:
 673         if (remote_gf != NULL)
 674                 OBD_FREE(remote_gf, remote_gf_size);
 675         RETURN(rc);
 676 }
 677
 678 static int lmv_hsm_req_count(struct lmv_obd *lmv,
 679                              const struct hsm_user_request *hur,
 680                              const struct lmv_tgt_desc *tgt_mds)
 681 {
 682         struct lmv_tgt_desc *curr_tgt;
 683         __u32 i;
 684         int nr = 0;
 685
 686         /* count how many requests must be sent to the given target */
 687         for (i = 0; i < hur->hur_request.hr_itemcount; i++) {
 688                 curr_tgt = lmv_fid2tgt(lmv, &hur->hur_user_item[i].hui_fid);
 689                 if (IS_ERR(curr_tgt))
 690                         RETURN(PTR_ERR(curr_tgt));
 691                 if (obd_uuid_equals(&curr_tgt->ltd_uuid, &tgt_mds->ltd_uuid))
 692                         nr++;
 693         }
 694         return nr;
 695 }
 696
 697 static int lmv_hsm_req_build(struct lmv_obd *lmv,
 698                               struct hsm_user_request *hur_in,
 699                               const struct lmv_tgt_desc *tgt_mds,
 700                               struct hsm_user_request *hur_out)
 701 {
 702         __u32 i, nr_out;
 703         struct lmv_tgt_desc *curr_tgt;
 704
 705         /* build the hsm_user_request for the given target */
 706         hur_out->hur_request = hur_in->hur_request;
 707         nr_out = 0;
 708         for (i = 0; i < hur_in->hur_request.hr_itemcount; i++) {
 709                 curr_tgt = lmv_fid2tgt(lmv, &hur_in->hur_user_item[i].hui_fid);
 710                 if (IS_ERR(curr_tgt))
 711                         RETURN(PTR_ERR(curr_tgt));
 712                 if (obd_uuid_equals(&curr_tgt->ltd_uuid, &tgt_mds->ltd_uuid)) {
 713                         hur_out->hur_user_item[nr_out] =
 714                                                 hur_in->hur_user_item[i];
 715                         nr_out++;
 716                 }
 717         }
 718         hur_out->hur_request.hr_itemcount = nr_out;
 719         memcpy(hur_data(hur_out), hur_data(hur_in),
 720                hur_in->hur_request.hr_data_len);
 721
 722         RETURN(0);
 723 }
 724
 725 static int lmv_hsm_ct_unregister(struct obd_device *obd, unsigned int cmd,
 726                                  int len, struct lustre_kernelcomm *lk,
 727                                  void __user *uarg)
 728 {
 729         struct lmv_obd *lmv = &obd->u.lmv;
 730         struct lu_tgt_desc *tgt;
 731         int rc;
 732
 733         ENTRY;
 734
 735         /* unregister request (call from llapi_hsm_copytool_fini) */
 736         lmv_foreach_connected_tgt(lmv, tgt)
 737                 /* best effort: try to clean as much as possible
 738                  * (continue on error) */
 739                 obd_iocontrol(cmd, tgt->ltd_exp, len, lk, uarg);
 740
 741         /* Whatever the result, remove copytool from kuc groups.
 742          * Unreached coordinators will get EPIPE on next requests
 743          * and will unregister automatically.
 744          */
 745         rc = libcfs_kkuc_group_rem(&obd->obd_uuid, lk->lk_uid, lk->lk_group);
 746
 747         RETURN(rc);
 748 }
 749
 750 static int lmv_hsm_ct_register(struct obd_device *obd, unsigned int cmd,
 751                                int len, struct lustre_kernelcomm *lk,
 752                                void __user *uarg)
 753 {
 754         struct lmv_obd *lmv = &obd->u.lmv;
 755         struct file *filp;
 756         bool any_set = false;
 757         struct kkuc_ct_data *kcd;
 758         size_t kcd_size;
 759         struct lu_tgt_desc *tgt;
 760         __u32 i;
 761         int err;
 762         int rc = 0;
 763
 764         ENTRY;
 765
 766         filp = fget(lk->lk_wfd);
 767         if (!filp)
 768                 RETURN(-EBADF);
 769
 770         if (lk->lk_flags & LK_FLG_DATANR)
 771                 kcd_size = offsetof(struct kkuc_ct_data,
 772                                     kcd_archives[lk->lk_data_count]);
 773         else
 774                 kcd_size = sizeof(*kcd);
 775
 776         OBD_ALLOC(kcd, kcd_size);
 777         if (kcd == NULL)
 778                 GOTO(err_fput, rc = -ENOMEM);
 779
 780         kcd->kcd_nr_archives = lk->lk_data_count;
 781         if (lk->lk_flags & LK_FLG_DATANR) {
 782                 kcd->kcd_magic = KKUC_CT_DATA_ARRAY_MAGIC;
 783                 if (lk->lk_data_count > 0)
 784                         memcpy(kcd->kcd_archives, lk->lk_data,
 785                                sizeof(*kcd->kcd_archives) * lk->lk_data_count);
 786         } else {
 787                 kcd->kcd_magic = KKUC_CT_DATA_BITMAP_MAGIC;
 788         }
 789
 790         rc = libcfs_kkuc_group_add(filp, &obd->obd_uuid, lk->lk_uid,
 791                                    lk->lk_group, kcd, kcd_size);
 792         OBD_FREE(kcd, kcd_size);
 793         if (rc)
 794                 GOTO(err_fput, rc);
 795
 796         /* All or nothing: try to register to all MDS.
 797          * In case of failure, unregister from previous MDS,
 798          * except if it because of inactive target. */
 799         lmv_foreach_connected_tgt(lmv, tgt) {
 800                 err = obd_iocontrol(cmd, tgt->ltd_exp, len, lk, uarg);
 801                 if (err) {
 802                         if (tgt->ltd_active) {
 803                                 /* permanent error */
 804                                 CERROR("%s: iocontrol MDC %s on MDT"
 805                                        " idx %d cmd %x: err = %d\n",
 806                                        lmv2obd_dev(lmv)->obd_name,
 807                                        tgt->ltd_uuid.uuid, tgt->ltd_index, cmd,
 808                                        err);
 809                                 rc = err;
 810                                 lk->lk_flags |= LK_FLG_STOP;
 811                                 i = tgt->ltd_index;
 812                                 /* unregister from previous MDS */
 813                                 lmv_foreach_connected_tgt(lmv, tgt) {
 814                                         if (tgt->ltd_index >= i)
 815                                                 break;
 816
 817                                         obd_iocontrol(cmd, tgt->ltd_exp, len,
 818                                                       lk, uarg);
 819                                 }
 820                                 GOTO(err_kkuc_rem, rc);
 821                         }
 822                         /* else: transient error.
 823                          * kuc will register to the missing MDT
 824                          * when it is back */
 825                 } else {
 826                         any_set = true;
 827                 }
 828         }
 829
 830         if (!any_set)
 831                 /* no registration done: return error */
 832                 GOTO(err_kkuc_rem, rc = -ENOTCONN);
 833
 834         RETURN(0);
 835
 836 err_kkuc_rem:
 837         libcfs_kkuc_group_rem(&obd->obd_uuid, lk->lk_uid, lk->lk_group);
 838
 839 err_fput:
 840         fput(filp);
 841         return rc;
 842 }
 843
 844 static int lmv_iocontrol(unsigned int cmd, struct obd_export *exp,
 845                          int len, void *karg, void __user *uarg)
 846 {
 847         struct obd_device *obd = class_exp2obd(exp);
 848         struct lmv_obd *lmv = &obd->u.lmv;
 849         struct lu_tgt_desc *tgt = NULL;
 850         int set = 0;
 851         __u32 count = lmv->lmv_mdt_count;
 852         int rc = 0;
 853
 854         ENTRY;
 855         CDEBUG(D_IOCTL, "%s: cmd=%x len=%u karg=%pK uarg=%pK\n",
 856                exp->exp_obd->obd_name, cmd, len, karg, uarg);
 857         if (count == 0)
 858                 RETURN(-ENOTTY);
 859
 860         /* exit early for unknown ioctl types */
 861         if (unlikely(_IOC_TYPE(cmd) != 'f' && !IOC_OSC_SET_ACTIVE_ALLOW(cmd)))
 862                 RETURN(OBD_IOC_ERROR(obd->obd_name, cmd, "unknown", -ENOTTY));
 863
 864         /* handle commands that don't use @karg first */
 865         switch (cmd) {
 866         case LL_IOC_GET_CONNECT_FLAGS:
 867                 tgt = lmv_tgt(lmv, 0);
 868                 rc = -ENODATA;
 869                 if (tgt && tgt->ltd_exp)
 870                         rc = obd_iocontrol(cmd, tgt->ltd_exp, len, NULL, uarg);
 871                 RETURN(rc);
 872         }
 873
 874         if (unlikely(karg == NULL))
 875                 RETURN(OBD_IOC_ERROR(obd->obd_name, cmd, "karg=NULL", -EINVAL));
 876
 877         switch (cmd) {
 878         case IOC_OBD_STATFS: {
 879                 struct obd_ioctl_data *data = karg;
 880                 struct obd_device *mdc_obd;
 881                 struct obd_statfs stat_buf = {0};
 882                 __u32 index;
 883
 884                 memcpy(&index, data->ioc_inlbuf2, sizeof(__u32));
 885
 886                 if (index >= lmv->lmv_mdt_descs.ltd_tgts_size)
 887                         RETURN(-ENODEV);
 888
 889                 tgt = lmv_tgt(lmv, index);
 890                 if (!tgt)
 891                         RETURN(-EAGAIN);
 892
 893                 if (!tgt->ltd_active)
 894                         RETURN(-ENODATA);
 895
 896                 mdc_obd = class_exp2obd(tgt->ltd_exp);
 897                 if (!mdc_obd)
 898                         RETURN(-EINVAL);
 899
 900                 /* copy UUID */
 901                 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(mdc_obd),
 902                                  min((int) data->ioc_plen2,
 903                                      (int) sizeof(struct obd_uuid))))
 904                         RETURN(-EFAULT);
 905
 906                 rc = obd_statfs(NULL, tgt->ltd_exp, &stat_buf,
 907                                 ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS,
 908                                 0);
 909                 if (rc)
 910                         RETURN(rc);
 911                 if (copy_to_user(data->ioc_pbuf1, &stat_buf,
 912                                  min((int) data->ioc_plen1,
 913                                      (int) sizeof(stat_buf))))
 914                         RETURN(-EFAULT);
 915                 break;
 916         }
 917         case OBD_IOC_QUOTACTL: {
 918                 struct if_quotactl *qctl = karg;
 919                 struct obd_quotactl *oqctl;
 920                 struct obd_import *imp;
 921
 922                 if (qctl->qc_valid == QC_MDTIDX) {
 923                         tgt = lmv_tgt(lmv, qctl->qc_idx);
 924                 } else if (qctl->qc_valid == QC_UUID) {
 925                         lmv_foreach_tgt(lmv, tgt) {
 926                                 if (!obd_uuid_equals(&tgt->ltd_uuid,
 927                                                      &qctl->obd_uuid))
 928                                         continue;
 929
 930                                 if (!tgt->ltd_exp)
 931                                         RETURN(-EINVAL);
 932
 933                                 break;
 934                         }
 935                 } else {
 936                         RETURN(-EINVAL);
 937                 }
 938
 939                 if (!tgt)
 940                         RETURN(-ENODEV);
 941
 942                 if (!tgt->ltd_exp)
 943                         RETURN(-EINVAL);
 944
 945                 imp = class_exp2cliimp(tgt->ltd_exp);
 946                 if (!tgt->ltd_active && imp->imp_state != LUSTRE_IMP_IDLE) {
 947                         qctl->qc_valid = QC_MDTIDX;
 948                         qctl->obd_uuid = tgt->ltd_uuid;
 949                         RETURN(-ENODATA);
 950                 }
 951
 952                 OBD_ALLOC_PTR(oqctl);
 953                 if (!oqctl)
 954                         RETURN(-ENOMEM);
 955
 956                 QCTL_COPY(oqctl, qctl);
 957                 rc = obd_quotactl(tgt->ltd_exp, oqctl);
 958                 if (rc == 0) {
 959                         QCTL_COPY(qctl, oqctl);
 960                         qctl->qc_valid = QC_MDTIDX;
 961                         qctl->obd_uuid = tgt->ltd_uuid;
 962                 }
 963                 OBD_FREE_PTR(oqctl);
 964                 break;
 965         }
 966         case LL_IOC_FID2MDTIDX: {
 967                 struct lu_fid *fid = karg;
 968                 int             mdt_index;
 969
 970                 rc = lmv_fld_lookup(lmv, fid, &mdt_index);
 971                 if (rc != 0)
 972                         RETURN(rc);
 973
 974                 /* Note: this is from llite(see ll_dir_ioctl()), @uarg does not
 975                  * point to user space memory for FID2MDTIDX. */
 976                 *(__u32 *)uarg = mdt_index;
 977                 break;
 978         }
 979         case OBD_IOC_FID2PATH: {
 980                 rc = lmv_fid2path(exp, len, karg, uarg);
 981                 break;
 982         }
 983         case LL_IOC_HSM_STATE_GET:
 984         case LL_IOC_HSM_STATE_SET:
 985         case LL_IOC_HSM_ACTION: {
 986                 struct md_op_data *op_data = karg;
 987
 988                 tgt = lmv_fid2tgt(lmv, &op_data->op_fid1);
 989                 if (IS_ERR(tgt))
 990                         RETURN(PTR_ERR(tgt));
 991
 992                 if (tgt->ltd_exp == NULL)
 993                         RETURN(-EINVAL);
 994
 995                 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
 996                 break;
 997         }
 998         case LL_IOC_HSM_PROGRESS: {
 999                 const struct hsm_progress_kernel *hpk = karg;
1000
1001                 tgt = lmv_fid2tgt(lmv, &hpk->hpk_fid);
1002                 if (IS_ERR(tgt))
1003                         RETURN(PTR_ERR(tgt));
1004                 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1005                 break;
1006         }
1007         case LL_IOC_HSM_REQUEST: {
1008                 struct hsm_user_request *hur = karg;
1009                 unsigned int reqcount = hur->hur_request.hr_itemcount;
1010
1011                 if (reqcount == 0)
1012                         RETURN(0);
1013
1014                 /* if the request is about a single fid
1015                  * or if there is a single MDS, no need to split
1016                  * the request. */
1017                 if (reqcount == 1 || count == 1) {
1018                         tgt = lmv_fid2tgt(lmv, &hur->hur_user_item[0].hui_fid);
1019                         if (IS_ERR(tgt))
1020                                 RETURN(PTR_ERR(tgt));
1021                         rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1022                 } else {
1023                         /* split fid list to their respective MDS */
1024                         lmv_foreach_connected_tgt(lmv, tgt) {
1025                                 int nr, rc1;
1026                                 size_t reqlen;
1027                                 struct hsm_user_request *req;
1028
1029                                 nr = lmv_hsm_req_count(lmv, hur, tgt);
1030                                 if (nr < 0)
1031                                         RETURN(nr);
1032                                 if (nr == 0) /* nothing for this MDS */
1033                                         continue;
1034
1035                                 /* build a request with fids for this MDS */
1036                                 reqlen = offsetof(typeof(*hur),
1037                                                   hur_user_item[nr])
1038                                                 + hur->hur_request.hr_data_len;
1039                                 OBD_ALLOC_LARGE(req, reqlen);
1040                                 if (req == NULL)
1041                                         RETURN(-ENOMEM);
1042                                 rc1 = lmv_hsm_req_build(lmv, hur, tgt, req);
1043                                 if (rc1 < 0)
1044                                         GOTO(hsm_req_err, rc1);
1045                                 rc1 = obd_iocontrol(cmd, tgt->ltd_exp, reqlen,
1046                                                     req, uarg);
1047 hsm_req_err:
1048                                 if (rc1 != 0 && rc == 0)
1049                                         rc = rc1;
1050                                 OBD_FREE_LARGE(req, reqlen);
1051                         }
1052                 }
1053                 break;
1054         }
1055         case LL_IOC_LOV_SWAP_LAYOUTS: {
1056                 struct md_op_data *op_data = karg;
1057                 struct lmv_tgt_desc *tgt1, *tgt2;
1058
1059                 tgt1 = lmv_fid2tgt(lmv, &op_data->op_fid1);
1060                 if (IS_ERR(tgt1))
1061                         RETURN(PTR_ERR(tgt1));
1062
1063                 tgt2 = lmv_fid2tgt(lmv, &op_data->op_fid2);
1064                 if (IS_ERR(tgt2))
1065                         RETURN(PTR_ERR(tgt2));
1066
1067                 if ((tgt1->ltd_exp == NULL) || (tgt2->ltd_exp == NULL))
1068                         RETURN(-EINVAL);
1069
1070                 /* only files on same MDT can have their layouts swapped */
1071                 if (tgt1->ltd_index != tgt2->ltd_index)
1072                         RETURN(-EPERM);
1073
1074                 rc = obd_iocontrol(cmd, tgt1->ltd_exp, len, karg, uarg);
1075                 break;
1076         }
1077         case LL_IOC_HSM_CT_START: {
1078                 struct lustre_kernelcomm *lk = karg;
1079
1080                 if (lk->lk_flags & LK_FLG_STOP)
1081                         rc = lmv_hsm_ct_unregister(obd, cmd, len, lk, uarg);
1082                 else
1083                         rc = lmv_hsm_ct_register(obd, cmd, len, lk, uarg);
1084                 break;
1085         }
1086         default:
1087                 lmv_foreach_connected_tgt(lmv, tgt) {
1088                         struct obd_device *mdc_obd;
1089                         int err;
1090
1091                         /* ll_umount_begin() sets force flag but for lmv, not
1092                          * mdc. Let's pass it through */
1093                         mdc_obd = class_exp2obd(tgt->ltd_exp);
1094                         mdc_obd->obd_force = obd->obd_force;
1095                         err = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1096                         if (err) {
1097                                 if (tgt->ltd_active) {
1098                                         OBD_IOC_ERROR(obd->obd_name, cmd,
1099                                                       tgt->ltd_uuid.uuid, err);
1100                                         if (!rc)
1101                                                 rc = err;
1102                                         if (unlikely(err == -ENOTTY))
1103                                                 break;
1104                                 }
1105                         } else {
1106                                 set = 1;
1107                         }
1108                 }
1109                 if (!set && !rc)
1110                         rc = -EIO;
1111                 break;
1112         }
1113         RETURN(rc);
1114 }
1115
1116 int lmv_fid_alloc(const struct lu_env *env, struct obd_export *exp,
1117                   struct lu_fid *fid, struct md_op_data *op_data)
1118 {
1119         struct obd_device *obd = class_exp2obd(exp);
1120         struct lmv_obd *lmv = &obd->u.lmv;
1121         struct lmv_tgt_desc *tgt;
1122         int rc;
1123
1124         ENTRY;
1125
1126         LASSERT(op_data);
1127         LASSERT(fid);
1128
1129         tgt = lmv_tgt(lmv, op_data->op_mds);
1130         if (!tgt)
1131                 RETURN(-ENODEV);
1132
1133         if (!tgt->ltd_active || !tgt->ltd_exp)
1134                 RETURN(-ENODEV);
1135
1136         /*
1137          * New seq alloc and FLD setup should be atomic. Otherwise we may find
1138          * on server that seq in new allocated fid is not yet known.
1139          */
1140         mutex_lock(&tgt->ltd_fid_mutex);
1141         rc = obd_fid_alloc(NULL, tgt->ltd_exp, fid, NULL);
1142         mutex_unlock(&tgt->ltd_fid_mutex);
1143         if (rc > 0) {
1144                 LASSERT(fid_is_sane(fid));
1145                 rc = 0;
1146         }
1147
1148         RETURN(rc);
1149 }
1150
1151 static int lmv_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
1152 {
1153         struct lmv_obd *lmv = &obd->u.lmv;
1154         struct lmv_desc *desc;
1155         struct lnet_processid lnet_id;
1156         int i = 0;
1157         int rc;
1158
1159         ENTRY;
1160
1161         if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
1162                 CERROR("LMV setup requires a descriptor\n");
1163                 RETURN(-EINVAL);
1164         }
1165
1166         desc = (struct lmv_desc *)lustre_cfg_buf(lcfg, 1);
1167         if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
1168                 CERROR("Lmv descriptor size wrong: %d > %d\n",
1169                        (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
1170                 RETURN(-EINVAL);
1171         }
1172
1173         obd_str2uuid(&lmv->lmv_mdt_descs.ltd_lmv_desc.ld_uuid,
1174                      desc->ld_uuid.uuid);
1175         lmv->lmv_mdt_descs.ltd_lmv_desc.ld_tgt_count = 0;
1176         lmv->lmv_mdt_descs.ltd_lmv_desc.ld_active_tgt_count = 0;
1177         lmv->lmv_mdt_descs.ltd_lmv_desc.ld_qos_maxage =
1178                 LMV_DESC_QOS_MAXAGE_DEFAULT;
1179         lmv->max_def_easize = 0;
1180         lmv->max_easize = 0;
1181
1182         spin_lock_init(&lmv->lmv_lock);
1183
1184         /*
1185          * initialize rr_index to lower 32bit of netid, so that client
1186          * can distribute subdirs evenly from the beginning.
1187          */
1188         while (LNetGetId(i++, &lnet_id) != -ENOENT) {
1189                 if (!nid_is_lo0(&lnet_id.nid)) {
1190                         lmv->lmv_qos_rr_index = ntohl(lnet_id.nid.nid_addr[0]);
1191                         break;
1192                 }
1193         }
1194
1195         rc = lmv_tunables_init(obd);
1196         if (rc)
1197                 CWARN("%s: error adding LMV sysfs/debugfs files: rc = %d\n",
1198                       obd->obd_name, rc);
1199
1200         rc = fld_client_init(&lmv->lmv_fld, obd->obd_name,
1201                              LUSTRE_CLI_FLD_HASH_DHT);
1202         if (rc)
1203                 CERROR("Can't init FLD, err %d\n", rc);
1204
1205         rc = lu_tgt_descs_init(&lmv->lmv_mdt_descs, true);
1206         if (rc)
1207                 CWARN("%s: error initialize target table: rc = %d\n",
1208                       obd->obd_name, rc);
1209
1210         RETURN(rc);
1211 }
1212
1213 static int lmv_cleanup(struct obd_device *obd)
1214 {
1215         struct lmv_obd *lmv = &obd->u.lmv;
1216         struct lu_tgt_desc *tgt;
1217         struct lu_tgt_desc *tmp;
1218
1219         ENTRY;
1220
1221         fld_client_fini(&lmv->lmv_fld);
1222         fld_client_debugfs_fini(&lmv->lmv_fld);
1223
1224         lprocfs_obd_cleanup(obd);
1225         lprocfs_free_md_stats(obd);
1226
1227         lmv_foreach_tgt_safe(lmv, tgt, tmp)
1228                 lmv_del_target(lmv, tgt);
1229         lu_tgt_descs_fini(&lmv->lmv_mdt_descs);
1230
1231         RETURN(0);
1232 }
1233
1234 static int lmv_process_config(struct obd_device *obd, size_t len, void *buf)
1235 {
1236         struct lustre_cfg       *lcfg = buf;
1237         struct obd_uuid         obd_uuid;
1238         int                     gen;
1239         __u32                   index;
1240         int                     rc;
1241         ENTRY;
1242
1243         switch (lcfg->lcfg_command) {
1244         case LCFG_ADD_MDC:
1245                 /* modify_mdc_tgts add 0:lustre-clilmv  1:lustre-MDT0000_UUID
1246                  * 2:0  3:1  4:lustre-MDT0000-mdc_UUID */
1247                 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(obd_uuid.uuid))
1248                         GOTO(out, rc = -EINVAL);
1249
1250                 obd_str2uuid(&obd_uuid,  lustre_cfg_buf(lcfg, 1));
1251
1252                 if (sscanf(lustre_cfg_buf(lcfg, 2), "%u", &index) != 1)
1253                         GOTO(out, rc = -EINVAL);
1254                 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", &gen) != 1)
1255                         GOTO(out, rc = -EINVAL);
1256                 rc = lmv_add_target(obd, &obd_uuid, index, gen);
1257                 GOTO(out, rc);
1258         default:
1259                 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
1260                 GOTO(out, rc = -EINVAL);
1261         }
1262 out:
1263         RETURN(rc);
1264 }
1265
1266 static int lmv_select_statfs_mdt(struct lmv_obd *lmv, __u32 flags)
1267 {
1268         int i;
1269
1270         if (flags & OBD_STATFS_FOR_MDT0)
1271                 return 0;
1272
1273         if (lmv->lmv_statfs_start || lmv->lmv_mdt_count == 1)
1274                 return lmv->lmv_statfs_start;
1275
1276         /* choose initial MDT for this client */
1277         for (i = 0;; i++) {
1278                 struct lnet_processid lnet_id;
1279                 if (LNetGetId(i, &lnet_id) == -ENOENT)
1280                         break;
1281
1282                 if (!nid_is_lo0(&lnet_id.nid)) {
1283                         /* We dont need a full 64-bit modulus, just enough
1284                          * to distribute the requests across MDTs evenly.
1285                          */
1286                         lmv->lmv_statfs_start = nidhash(&lnet_id.nid) %
1287                                                 lmv->lmv_mdt_count;
1288                         break;
1289                 }
1290         }
1291
1292         return lmv->lmv_statfs_start;
1293 }
1294
1295 static int lmv_statfs(const struct lu_env *env, struct obd_export *exp,
1296                       struct obd_statfs *osfs, time64_t max_age, __u32 flags)
1297 {
1298         struct obd_device *obd = class_exp2obd(exp);
1299         struct lmv_obd *lmv = &obd->u.lmv;
1300         struct obd_statfs *temp;
1301         struct lu_tgt_desc *tgt;
1302         __u32 i;
1303         __u32 idx;
1304         int rc = 0;
1305         int err = 0;
1306
1307         ENTRY;
1308
1309         OBD_ALLOC(temp, sizeof(*temp));
1310         if (temp == NULL)
1311                 RETURN(-ENOMEM);
1312
1313         /* distribute statfs among MDTs */
1314         idx = lmv_select_statfs_mdt(lmv, flags);
1315
1316         for (i = 0; i < lmv->lmv_mdt_descs.ltd_tgts_size; i++, idx++) {
1317                 idx = idx % lmv->lmv_mdt_descs.ltd_tgts_size;
1318                 tgt = lmv_tgt(lmv, idx);
1319                 if (!tgt || !tgt->ltd_exp)
1320                         continue;
1321
1322                 rc = obd_statfs(env, tgt->ltd_exp, temp, max_age,
1323                                 flags | OBD_STATFS_NESTED);
1324                 if (rc) {
1325                         CERROR("%s: can't stat MDS #%d: rc = %d\n",
1326                                tgt->ltd_exp->exp_obd->obd_name, i, rc);
1327                         err = rc;
1328                         /* Try another MDT */
1329                         if (flags & OBD_STATFS_SUM)
1330                                 continue;
1331                         GOTO(out_free_temp, rc);
1332                 }
1333
1334                 if (temp->os_state & OS_STATFS_SUM ||
1335                     flags == OBD_STATFS_FOR_MDT0) {
1336                         /* reset to the last aggregated values
1337                          * and don't sum with non-aggrated data */
1338                         /* If the statfs is from mount, it needs to retrieve
1339                          * necessary information from MDT0. i.e. mount does
1340                          * not need the merged osfs from all of MDT. Also
1341                          * clients can be mounted as long as MDT0 is in
1342                          * service */
1343                         *osfs = *temp;
1344                         GOTO(out_free_temp, rc);
1345                 }
1346
1347                 if (i == 0) {
1348                         *osfs = *temp;
1349                 } else {
1350                         osfs->os_bavail += temp->os_bavail;
1351                         osfs->os_blocks += temp->os_blocks;
1352                         osfs->os_ffree += temp->os_ffree;
1353                         osfs->os_files += temp->os_files;
1354                         osfs->os_granted += temp->os_granted;
1355                 }
1356         }
1357         /* There is no stats from some MDTs, data incomplete */
1358         if (err)
1359                 rc = err;
1360 out_free_temp:
1361         OBD_FREE(temp, sizeof(*temp));
1362         RETURN(rc);
1363 }
1364
1365 static int lmv_statfs_update(void *cookie, int rc)
1366 {
1367         struct obd_info *oinfo = cookie;
1368         struct obd_device *obd = oinfo->oi_obd;
1369         struct lmv_obd *lmv = &obd->u.lmv;
1370         struct lmv_tgt_desc *tgt = oinfo->oi_tgt;
1371         struct obd_statfs *osfs = oinfo->oi_osfs;
1372
1373         /*
1374          * NB: don't deactivate TGT upon error, because we may not trigger async
1375          * statfs any longer, then there is no chance to activate TGT.
1376          */
1377         if (!rc) {
1378                 spin_lock(&lmv->lmv_lock);
1379                 tgt->ltd_statfs = *osfs;
1380                 tgt->ltd_statfs_age = ktime_get_seconds();
1381                 spin_unlock(&lmv->lmv_lock);
1382                 set_bit(LQ_DIRTY, &lmv->lmv_qos.lq_flags);
1383         }
1384
1385         return rc;
1386 }
1387
1388 /* update tgt statfs async if it's ld_qos_maxage old */
1389 int lmv_statfs_check_update(struct obd_device *obd, struct lmv_tgt_desc *tgt)
1390 {
1391         struct obd_info oinfo = {
1392                 .oi_obd = obd,
1393                 .oi_tgt = tgt,
1394                 .oi_cb_up = lmv_statfs_update,
1395         };
1396         int rc;
1397
1398         if (ktime_get_seconds() - tgt->ltd_statfs_age <
1399             obd->u.lmv.lmv_mdt_descs.ltd_lmv_desc.ld_qos_maxage)
1400                 return 0;
1401
1402         rc = obd_statfs_async(tgt->ltd_exp, &oinfo, 0, NULL);
1403
1404         return rc;
1405 }
1406
1407 static int lmv_get_root(struct obd_export *exp, const char *fileset,
1408                         struct lu_fid *fid)
1409 {
1410         struct obd_device *obd = exp->exp_obd;
1411         struct lmv_obd *lmv = &obd->u.lmv;
1412         struct lu_tgt_desc *tgt = lmv_tgt(lmv, 0);
1413         int rc;
1414
1415         ENTRY;
1416
1417         if (!tgt)
1418                 RETURN(-ENODEV);
1419
1420         rc = md_get_root(tgt->ltd_exp, fileset, fid);
1421         RETURN(rc);
1422 }
1423
1424 static int lmv_getxattr(struct obd_export *exp, const struct lu_fid *fid,
1425                         u64 obd_md_valid, const char *name, size_t buf_size,
1426                         struct ptlrpc_request **req)
1427 {
1428         struct obd_device *obd = exp->exp_obd;
1429         struct lmv_obd *lmv = &obd->u.lmv;
1430         struct lmv_tgt_desc *tgt;
1431         int rc;
1432
1433         ENTRY;
1434
1435         tgt = lmv_fid2tgt(lmv, fid);
1436         if (IS_ERR(tgt))
1437                 RETURN(PTR_ERR(tgt));
1438
1439         rc = md_getxattr(tgt->ltd_exp, fid, obd_md_valid, name, buf_size, req);
1440
1441         RETURN(rc);
1442 }
1443
1444 static int lmv_setxattr(struct obd_export *exp, const struct lu_fid *fid,
1445                         u64 obd_md_valid, const char *name,
1446                         const void *value, size_t value_size,
1447                         unsigned int xattr_flags, u32 suppgid,
1448                         struct ptlrpc_request **req)
1449 {
1450         struct obd_device *obd = exp->exp_obd;
1451         struct lmv_obd *lmv = &obd->u.lmv;
1452         struct lmv_tgt_desc *tgt;
1453         int rc;
1454
1455         ENTRY;
1456
1457         tgt = lmv_fid2tgt(lmv, fid);
1458         if (IS_ERR(tgt))
1459                 RETURN(PTR_ERR(tgt));
1460
1461         rc = md_setxattr(tgt->ltd_exp, fid, obd_md_valid, name,
1462                          value, value_size, xattr_flags, suppgid, req);
1463
1464         RETURN(rc);
1465 }
1466
1467 static int lmv_getattr(struct obd_export *exp, struct md_op_data *op_data,
1468                        struct ptlrpc_request **request)
1469 {
1470         struct obd_device *obd = exp->exp_obd;
1471         struct lmv_obd *lmv = &obd->u.lmv;
1472         struct lmv_tgt_desc *tgt;
1473         int rc;
1474
1475         ENTRY;
1476
1477         tgt = lmv_fid2tgt(lmv, &op_data->op_fid1);
1478         if (IS_ERR(tgt))
1479                 RETURN(PTR_ERR(tgt));
1480
1481         if (op_data->op_flags & MF_GET_MDT_IDX) {
1482                 op_data->op_mds = tgt->ltd_index;
1483                 RETURN(0);
1484         }
1485
1486         rc = md_getattr(tgt->ltd_exp, op_data, request);
1487
1488         RETURN(rc);
1489 }
1490
1491 static int lmv_null_inode(struct obd_export *exp, const struct lu_fid *fid)
1492 {
1493         struct obd_device *obd = exp->exp_obd;
1494         struct lmv_obd *lmv = &obd->u.lmv;
1495         struct lu_tgt_desc *tgt;
1496
1497         ENTRY;
1498
1499         CDEBUG(D_INODE, "CBDATA for "DFID"\n", PFID(fid));
1500
1501         /*
1502          * With DNE every object can have two locks in different namespaces:
1503          * lookup lock in space of MDT storing direntry and update/open lock in
1504          * space of MDT storing inode.
1505          */
1506         lmv_foreach_connected_tgt(lmv, tgt)
1507                 md_null_inode(tgt->ltd_exp, fid);
1508
1509         RETURN(0);
1510 }
1511
1512 static int lmv_close(struct obd_export *exp, struct md_op_data *op_data,
1513                      struct md_open_data *mod, struct ptlrpc_request **request)
1514 {
1515         struct obd_device *obd = exp->exp_obd;
1516         struct lmv_obd *lmv = &obd->u.lmv;
1517         struct lmv_tgt_desc *tgt;
1518         int rc;
1519
1520         ENTRY;
1521
1522         tgt = lmv_fid2tgt(lmv, &op_data->op_fid1);
1523         if (IS_ERR(tgt))
1524                 RETURN(PTR_ERR(tgt));
1525
1526         CDEBUG(D_INODE, "CLOSE "DFID"\n", PFID(&op_data->op_fid1));
1527         rc = md_close(tgt->ltd_exp, op_data, mod, request);
1528         RETURN(rc);
1529 }
1530
1531 static struct lu_tgt_desc *lmv_locate_tgt_qos(struct lmv_obd *lmv,
1532                                               struct md_op_data *op_data)
1533 {
1534         struct lu_tgt_desc *tgt, *cur = NULL;
1535         __u64 total_avail = 0;
1536         __u64 total_weight = 0;
1537         __u64 cur_weight = 0;
1538         int total_usable = 0;
1539         __u64 rand;
1540         int rc;
1541
1542         ENTRY;
1543
1544         if (!ltd_qos_is_usable(&lmv->lmv_mdt_descs))
1545                 RETURN(ERR_PTR(-EAGAIN));
1546
1547         down_write(&lmv->lmv_qos.lq_rw_sem);
1548
1549         if (!ltd_qos_is_usable(&lmv->lmv_mdt_descs))
1550                 GOTO(unlock, tgt = ERR_PTR(-EAGAIN));
1551
1552         rc = ltd_qos_penalties_calc(&lmv->lmv_mdt_descs);
1553         if (rc)
1554                 GOTO(unlock, tgt = ERR_PTR(rc));
1555
1556         lmv_foreach_tgt(lmv, tgt) {
1557                 if (!tgt->ltd_exp || !tgt->ltd_active) {
1558                         tgt->ltd_qos.ltq_usable = 0;
1559                         continue;
1560                 }
1561
1562                 tgt->ltd_qos.ltq_usable = 1;
1563                 lu_tgt_qos_weight_calc(tgt, true);
1564                 if (tgt->ltd_index == op_data->op_mds)
1565                         cur = tgt;
1566                 total_avail += tgt->ltd_qos.ltq_avail;
1567                 total_weight += tgt->ltd_qos.ltq_weight;
1568                 total_usable++;
1569         }
1570
1571         /* If current MDT has above-average space and dir is not aleady using
1572          * round-robin to spread across more MDTs, stay on the parent MDT
1573          * to avoid creating needless remote MDT directories.  Remote dirs
1574          * close to the root balance space more effectively than bottom dirs,
1575          * so prefer to create remote dirs at top level of directory tree.
1576          * "16 / (dir_depth + 10)" is the factor to make it less likely
1577          * for top-level directories to stay local unless they have more than
1578          * average free space, while deep dirs prefer local until more full.
1579          *    depth=0 -> 160%, depth=3 -> 123%, depth=6 -> 100%,
1580          *    depth=9 -> 84%, depth=12 -> 73%, depth=15 -> 64%
1581          */
1582         if (!lmv_op_default_rr_mkdir(op_data)) {
1583                 rand = total_avail * 16 /
1584                         (total_usable * (op_data->op_dir_depth + 10));
1585                 if (cur && cur->ltd_qos.ltq_avail >= rand) {
1586                         tgt = cur;
1587                         GOTO(unlock, tgt);
1588                 }
1589         }
1590
1591         rand = lu_prandom_u64_max(total_weight);
1592
1593         lmv_foreach_connected_tgt(lmv, tgt) {
1594                 if (!tgt->ltd_qos.ltq_usable)
1595                         continue;
1596
1597                 cur_weight += tgt->ltd_qos.ltq_weight;
1598                 if (cur_weight < rand)
1599                         continue;
1600
1601                 ltd_qos_update(&lmv->lmv_mdt_descs, tgt, &total_weight);
1602                 GOTO(unlock, tgt);
1603         }
1604
1605         /* no proper target found */
1606         GOTO(unlock, tgt = ERR_PTR(-EAGAIN));
1607 unlock:
1608         up_write(&lmv->lmv_qos.lq_rw_sem);
1609
1610         return tgt;
1611 }
1612
1613 static struct lu_tgt_desc *lmv_locate_tgt_rr(struct lmv_obd *lmv)
1614 {
1615         struct lu_tgt_desc *tgt;
1616         int i;
1617         int index;
1618
1619         ENTRY;
1620
1621         spin_lock(&lmv->lmv_lock);
1622         for (i = 0; i < lmv->lmv_mdt_descs.ltd_tgts_size; i++) {
1623                 index = (i + lmv->lmv_qos_rr_index) %
1624                         lmv->lmv_mdt_descs.ltd_tgts_size;
1625                 tgt = lmv_tgt(lmv, index);
1626                 if (!tgt || !tgt->ltd_exp || !tgt->ltd_active)
1627                         continue;
1628
1629                 lmv->lmv_qos_rr_index = (tgt->ltd_index + 1) %
1630                                         lmv->lmv_mdt_descs.ltd_tgts_size;
1631                 spin_unlock(&lmv->lmv_lock);
1632
1633                 RETURN(tgt);
1634         }
1635         spin_unlock(&lmv->lmv_lock);
1636
1637         RETURN(ERR_PTR(-ENODEV));
1638 }
1639
1640 /* locate MDT which is less full (avoid the most full MDT) */
1641 static struct lu_tgt_desc *lmv_locate_tgt_lf(struct lmv_obd *lmv)
1642 {
1643         struct lu_tgt_desc *min = NULL;
1644         struct lu_tgt_desc *tgt;
1645         __u64 avail = 0;
1646         __u64 rand;
1647
1648         ENTRY;
1649
1650         if (!ltd_qos_is_usable(&lmv->lmv_mdt_descs))
1651                 RETURN(ERR_PTR(-EAGAIN));
1652
1653         down_write(&lmv->lmv_qos.lq_rw_sem);
1654
1655         if (!ltd_qos_is_usable(&lmv->lmv_mdt_descs))
1656                 GOTO(unlock, tgt = ERR_PTR(-EAGAIN));
1657
1658         lmv_foreach_tgt(lmv, tgt) {
1659                 if (!tgt->ltd_exp || !tgt->ltd_active) {
1660                         tgt->ltd_qos.ltq_usable = 0;
1661                         continue;
1662                 }
1663
1664                 tgt->ltd_qos.ltq_usable = 1;
1665                 lu_tgt_qos_weight_calc(tgt, true);
1666                 avail += tgt->ltd_qos.ltq_avail;
1667                 if (!min || min->ltd_qos.ltq_avail > tgt->ltd_qos.ltq_avail)
1668                         min = tgt;
1669         }
1670
1671         /* avoid the most full MDT */
1672         if (min)
1673                 avail -= min->ltd_qos.ltq_avail;
1674
1675         rand = lu_prandom_u64_max(avail);
1676         avail = 0;
1677         lmv_foreach_connected_tgt(lmv, tgt) {
1678                 if (!tgt->ltd_qos.ltq_usable)
1679                         continue;
1680
1681                 if (tgt == min)
1682                         continue;
1683
1684                 avail += tgt->ltd_qos.ltq_avail;
1685                 if (avail < rand)
1686                         continue;
1687
1688                 GOTO(unlock, tgt);
1689         }
1690
1691         /* no proper target found */
1692         GOTO(unlock, tgt = ERR_PTR(-EAGAIN));
1693 unlock:
1694         up_write(&lmv->lmv_qos.lq_rw_sem);
1695
1696         RETURN(tgt);
1697 }
1698
1699 /* locate MDT by file name, for striped directory, the file name hash decides
1700  * which stripe its dirent is stored.
1701  */
1702 static struct lmv_tgt_desc *
1703 lmv_locate_tgt_by_name(struct lmv_obd *lmv, struct lmv_stripe_md *lsm,
1704                        const char *name, int namelen, struct lu_fid *fid,
1705                        __u32 *mds, bool new_layout)
1706 {
1707         struct lmv_tgt_desc *tgt;
1708         const struct lmv_oinfo *oinfo;
1709
1710         if (!lmv_dir_striped(lsm) || !namelen) {
1711                 tgt = lmv_fid2tgt(lmv, fid);
1712                 if (IS_ERR(tgt))
1713                         return tgt;
1714
1715                 *mds = tgt->ltd_index;
1716                 return tgt;
1717         }
1718
1719         if (CFS_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_NAME_HASH)) {
1720                 if (cfs_fail_val >= lsm->lsm_md_stripe_count)
1721                         return ERR_PTR(-EBADF);
1722                 oinfo = &lsm->lsm_md_oinfo[cfs_fail_val];
1723         } else {
1724                 oinfo = lsm_name_to_stripe_info(lsm, name, namelen, new_layout);
1725                 if (IS_ERR(oinfo))
1726                         return ERR_CAST(oinfo);
1727         }
1728
1729         /* check stripe FID is sane */
1730         if (!fid_is_sane(&oinfo->lmo_fid))
1731                 return ERR_PTR(-ENODEV);
1732
1733         *fid = oinfo->lmo_fid;
1734         *mds = oinfo->lmo_mds;
1735         tgt = lmv_tgt(lmv, oinfo->lmo_mds);
1736
1737         CDEBUG(D_INODE, "locate MDT %u parent "DFID"\n", *mds, PFID(fid));
1738
1739         return tgt ? tgt : ERR_PTR(-ENODEV);
1740 }
1741
1742 /**
1743  * Locate MDT of op_data->op_fid1
1744  *
1745  * For striped directory, it will locate the stripe by name hash, if hash_type
1746  * is unknown, it will return the stripe specified by 'op_data->op_stripe_index'
1747  * which is set outside, and if dir is migrating, 'op_data->op_new_layout'
1748  * indicates whether old or new layout is used to locate.
1749  *
1750  * For plain direcotry, it just locate the MDT of op_data->op_fid1.
1751  *
1752  * \param[in] lmv               LMV device
1753  * \param[in/out] op_data       client MD stack parameters, name, namelen etc,
1754  *                              op_mds and op_fid1 will be updated if op_mea1
1755  *                              indicates fid1 represents a striped directory.
1756  *
1757  * retval               pointer to the lmv_tgt_desc if succeed.
1758  *                      ERR_PTR(errno) if failed.
1759  */
1760 struct lmv_tgt_desc *
1761 lmv_locate_tgt(struct lmv_obd *lmv, struct md_op_data *op_data)
1762 {
1763         struct lmv_stripe_md *lsm = op_data->op_mea1;
1764         struct lmv_oinfo *oinfo;
1765         struct lmv_tgt_desc *tgt;
1766
1767         if (lmv_dir_foreign(lsm))
1768                 return ERR_PTR(-ENODATA);
1769
1770         /* During creating VOLATILE file, it should honor the mdt
1771          * index if the file under striped dir is being restored, see
1772          * ct_restore(). */
1773         if (op_data->op_bias & MDS_CREATE_VOLATILE &&
1774             op_data->op_mds != LMV_OFFSET_DEFAULT) {
1775                 tgt = lmv_tgt(lmv, op_data->op_mds);
1776                 if (!tgt)
1777                         return ERR_PTR(-ENODEV);
1778
1779                 if (lmv_dir_striped(lsm)) {
1780                         int i;
1781
1782                         /* refill the right parent fid */
1783                         for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1784                                 oinfo = &lsm->lsm_md_oinfo[i];
1785                                 if (oinfo->lmo_mds == op_data->op_mds) {
1786                                         op_data->op_fid1 = oinfo->lmo_fid;
1787                                         break;
1788                                 }
1789                         }
1790
1791                         if (i == lsm->lsm_md_stripe_count)
1792                                 op_data->op_fid1 = lsm->lsm_md_oinfo[0].lmo_fid;
1793                 }
1794         } else if (lmv_dir_bad_hash(lsm)) {
1795                 LASSERT(op_data->op_stripe_index < lsm->lsm_md_stripe_count);
1796                 oinfo = &lsm->lsm_md_oinfo[op_data->op_stripe_index];
1797
1798                 op_data->op_fid1 = oinfo->lmo_fid;
1799                 op_data->op_mds = oinfo->lmo_mds;
1800                 tgt = lmv_tgt(lmv, oinfo->lmo_mds);
1801                 if (!tgt)
1802                         return ERR_PTR(-ENODEV);
1803         } else {
1804                 tgt = lmv_locate_tgt_by_name(lmv, op_data->op_mea1,
1805                                 op_data->op_name, op_data->op_namelen,
1806                                 &op_data->op_fid1, &op_data->op_mds,
1807                                 op_data->op_new_layout);
1808         }
1809
1810         return tgt;
1811 }
1812
1813 /* Locate MDT of op_data->op_fid2 for link/rename */
1814 static struct lmv_tgt_desc *
1815 lmv_locate_tgt2(struct lmv_obd *lmv, struct md_op_data *op_data)
1816 {
1817         struct lmv_tgt_desc *tgt;
1818         int rc;
1819
1820         LASSERT(op_data->op_name);
1821         if (lmv_dir_layout_changing(op_data->op_mea2)) {
1822                 struct lu_fid fid1 = op_data->op_fid1;
1823                 struct lmv_stripe_md *lsm1 = op_data->op_mea1;
1824                 struct ptlrpc_request *request = NULL;
1825
1826                 /*
1827                  * avoid creating new file under old layout of migrating
1828                  * directory, check it here.
1829                  */
1830                 tgt = lmv_locate_tgt_by_name(lmv, op_data->op_mea2,
1831                                 op_data->op_name, op_data->op_namelen,
1832                                 &op_data->op_fid2, &op_data->op_mds, false);
1833                 if (IS_ERR(tgt))
1834                         RETURN(tgt);
1835
1836                 op_data->op_fid1 = op_data->op_fid2;
1837                 op_data->op_mea1 = op_data->op_mea2;
1838                 rc = md_getattr_name(tgt->ltd_exp, op_data, &request);
1839                 op_data->op_fid1 = fid1;
1840                 op_data->op_mea1 = lsm1;
1841                 if (!rc) {
1842                         ptlrpc_req_finished(request);
1843                         RETURN(ERR_PTR(-EEXIST));
1844                 }
1845
1846                 if (rc != -ENOENT)
1847                         RETURN(ERR_PTR(rc));
1848         }
1849
1850         return lmv_locate_tgt_by_name(lmv, op_data->op_mea2, op_data->op_name,
1851                                 op_data->op_namelen, &op_data->op_fid2,
1852                                 &op_data->op_mds, true);
1853 }
1854
1855 int lmv_old_layout_lookup(struct lmv_obd *lmv, struct md_op_data *op_data)
1856 {
1857         struct lu_tgt_desc *tgt;
1858         struct ptlrpc_request *request;
1859         int rc;
1860
1861         LASSERT(lmv_dir_layout_changing(op_data->op_mea1));
1862         LASSERT(!op_data->op_new_layout);
1863
1864         tgt = lmv_locate_tgt(lmv, op_data);
1865         if (IS_ERR(tgt))
1866                 return PTR_ERR(tgt);
1867
1868         rc = md_getattr_name(tgt->ltd_exp, op_data, &request);
1869         if (!rc) {
1870                 ptlrpc_req_finished(request);
1871                 return -EEXIST;
1872         }
1873
1874         return rc;
1875 }
1876
1877 /* mkdir by QoS upon 'lfs mkdir -i -1'.
1878  *
1879  * NB, mkdir by QoS only if parent is not striped, this is to avoid remote
1880  * directories under striped directory.
1881  */
1882 static inline bool lmv_op_user_qos_mkdir(const struct md_op_data *op_data)
1883 {
1884         const struct lmv_user_md *lum = op_data->op_data;
1885
1886         if (op_data->op_code != LUSTRE_OPC_MKDIR)
1887                 return false;
1888
1889         if (lmv_dir_striped(op_data->op_mea1))
1890                 return false;
1891
1892         return (op_data->op_cli_flags & CLI_SET_MEA) && lum &&
1893                le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC &&
1894                le32_to_cpu(lum->lum_stripe_offset) == LMV_OFFSET_DEFAULT;
1895 }
1896
1897 /* mkdir by QoS if either ROOT or parent default LMV is space balanced. */
1898 static inline bool lmv_op_default_qos_mkdir(const struct md_op_data *op_data)
1899 {
1900         const struct lmv_stripe_md *lsm = op_data->op_default_mea1;
1901
1902         if (op_data->op_code != LUSTRE_OPC_MKDIR)
1903                 return false;
1904
1905         if (lmv_dir_striped(op_data->op_mea1))
1906                 return false;
1907
1908         return (op_data->op_flags & MF_QOS_MKDIR) ||
1909                (lsm && lsm->lsm_md_master_mdt_index == LMV_OFFSET_DEFAULT);
1910 }
1911
1912 /* if parent default LMV is space balanced, and
1913  * 1. max_inherit_rr is set
1914  * 2. or parent is ROOT
1915  * mkdir roundrobin. Or if parent doesn't have default LMV, while ROOT default
1916  * LMV requests roundrobin mkdir, do the same.
1917  * NB, this needs to check server is balanced, which is done by caller.
1918  */
1919 static inline bool lmv_op_default_rr_mkdir(const struct md_op_data *op_data)
1920 {
1921         const struct lmv_stripe_md *lsm = op_data->op_default_mea1;
1922
1923         return (op_data->op_flags & MF_RR_MKDIR) ||
1924                (lsm && lsm->lsm_md_max_inherit_rr != LMV_INHERIT_RR_NONE) ||
1925                fid_is_root(&op_data->op_fid1);
1926 }
1927
1928 /* 'lfs mkdir -i <specific_MDT>' */
1929 static inline bool lmv_op_user_specific_mkdir(const struct md_op_data *op_data)
1930 {
1931         const struct lmv_user_md *lum = op_data->op_data;
1932
1933         return op_data->op_code == LUSTRE_OPC_MKDIR &&
1934                op_data->op_cli_flags & CLI_SET_MEA && lum &&
1935                (le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC ||
1936                 le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC_SPECIFIC) &&
1937                le32_to_cpu(lum->lum_stripe_offset) != LMV_OFFSET_DEFAULT;
1938 }
1939
1940 /* parent default LMV master_mdt_index is not -1. */
1941 static inline bool
1942 lmv_op_default_specific_mkdir(const struct md_op_data *op_data)
1943 {
1944         return op_data->op_code == LUSTRE_OPC_MKDIR &&
1945                op_data->op_default_mea1 &&
1946                op_data->op_default_mea1->lsm_md_master_mdt_index !=
1947                         LMV_OFFSET_DEFAULT;
1948 }
1949
1950 /* locate MDT by space usage */
1951 static struct lu_tgt_desc *lmv_locate_tgt_by_space(struct lmv_obd *lmv,
1952                                                    struct md_op_data *op_data,
1953                                                    struct lmv_tgt_desc *tgt)
1954 {
1955         struct lmv_tgt_desc *tmp = tgt;
1956
1957         tgt = lmv_locate_tgt_qos(lmv, op_data);
1958         if (tgt == ERR_PTR(-EAGAIN)) {
1959                 if (ltd_qos_is_balanced(&lmv->lmv_mdt_descs) &&
1960                     !lmv_op_default_rr_mkdir(op_data) &&
1961                     !lmv_op_user_qos_mkdir(op_data))
1962                         /* if not necessary, don't create remote directory. */
1963                         tgt = tmp;
1964                 else
1965                         tgt = lmv_locate_tgt_rr(lmv);
1966         }
1967
1968         /*
1969          * only update statfs after QoS mkdir, this means the cached statfs may
1970          * be stale, and current mkdir may not follow QoS accurately, but it's
1971          * not serious, and avoids periodic statfs when client doesn't mkdir by
1972          * QoS.
1973          */
1974         if (!IS_ERR(tgt)) {
1975                 op_data->op_mds = tgt->ltd_index;
1976                 lmv_statfs_check_update(lmv2obd_dev(lmv), tgt);
1977         }
1978
1979         return tgt;
1980 }
1981
1982 int lmv_create(struct obd_export *exp, struct md_op_data *op_data,
1983                 const void *data, size_t datalen, umode_t mode, uid_t uid,
1984                 gid_t gid, kernel_cap_t cap_effective, __u64 rdev,
1985                 struct ptlrpc_request **request)
1986 {
1987         struct obd_device *obd = exp->exp_obd;
1988         struct lmv_obd *lmv = &obd->u.lmv;
1989         struct lmv_tgt_desc *tgt;
1990         struct mdt_body *repbody;
1991         int rc;
1992
1993         ENTRY;
1994
1995         if (!lmv->lmv_mdt_descs.ltd_lmv_desc.ld_active_tgt_count)
1996                 RETURN(-EIO);
1997
1998         if (lmv_dir_bad_hash(op_data->op_mea1))
1999                 RETURN(-EBADF);
2000
2001         if (lmv_dir_layout_changing(op_data->op_mea1)) {
2002                 /*
2003                  * if parent is migrating, create() needs to lookup existing
2004                  * name in both old and new layout, check old layout on client.
2005                  */
2006                 rc = lmv_old_layout_lookup(lmv, op_data);
2007                 if (rc != -ENOENT)
2008                         RETURN(rc);
2009
2010                 op_data->op_new_layout = true;
2011         }
2012
2013         tgt = lmv_locate_tgt(lmv, op_data);
2014         if (IS_ERR(tgt))
2015                 RETURN(PTR_ERR(tgt));
2016
2017         /* the order to apply policy in mkdir:
2018          * 1. is "lfs mkdir -i N"? mkdir on MDT N.
2019          * 2. is "lfs mkdir -i -1"? mkdir by space usage.
2020          * 3. is starting MDT specified in default LMV? mkdir on MDT N.
2021          * 4. is default LMV space balanced? mkdir by space usage.
2022          */
2023         if (lmv_op_user_specific_mkdir(op_data)) {
2024                 struct lmv_user_md *lum = op_data->op_data;
2025
2026                 op_data->op_mds = le32_to_cpu(lum->lum_stripe_offset);
2027                 tgt = lmv_tgt(lmv, op_data->op_mds);
2028                 if (!tgt)
2029                         RETURN(-ENODEV);
2030         } else if (lmv_op_user_qos_mkdir(op_data)) {
2031                 tgt = lmv_locate_tgt_by_space(lmv, op_data, tgt);
2032                 if (IS_ERR(tgt))
2033                         RETURN(PTR_ERR(tgt));
2034         } else if (lmv_op_default_specific_mkdir(op_data)) {
2035                 op_data->op_mds =
2036                         op_data->op_default_mea1->lsm_md_master_mdt_index;
2037                 tgt = lmv_tgt(lmv, op_data->op_mds);
2038                 if (!tgt)
2039                         RETURN(-ENODEV);
2040         } else if (lmv_op_default_qos_mkdir(op_data)) {
2041                 tgt = lmv_locate_tgt_by_space(lmv, op_data, tgt);
2042                 if (IS_ERR(tgt))
2043                         RETURN(PTR_ERR(tgt));
2044         }
2045
2046 retry:
2047         rc = lmv_fid_alloc(NULL, exp, &op_data->op_fid2, op_data);
2048         if (rc)
2049                 RETURN(rc);
2050
2051         CDEBUG(D_INODE, "CREATE name '%.*s' "DFID" on "DFID" -> mds #%x\n",
2052                 (int)op_data->op_namelen, op_data->op_name,
2053                 PFID(&op_data->op_fid2), PFID(&op_data->op_fid1),
2054                 op_data->op_mds);
2055
2056         op_data->op_flags |= MF_MDC_CANCEL_FID1;
2057         rc = md_create(tgt->ltd_exp, op_data, data, datalen, mode, uid, gid,
2058                        cap_effective, rdev, request);
2059         if (rc == 0) {
2060                 if (*request == NULL)
2061                         RETURN(rc);
2062                 CDEBUG(D_INODE, "Created - "DFID"\n", PFID(&op_data->op_fid2));
2063         }
2064
2065         /* dir restripe needs to send to MDT where dir is located */
2066         if (rc != -EREMOTE ||
2067             !(exp_connect_flags2(exp) & OBD_CONNECT2_CRUSH))
2068                 RETURN(rc);
2069
2070         repbody = req_capsule_server_get(&(*request)->rq_pill, &RMF_MDT_BODY);
2071         if (repbody == NULL)
2072                 RETURN(-EPROTO);
2073
2074         /* Not cross-ref case, just get out of here. */
2075         if (likely(!(repbody->mbo_valid & OBD_MD_MDS)))
2076                 RETURN(rc);
2077
2078         op_data->op_fid2 = repbody->mbo_fid1;
2079         ptlrpc_req_finished(*request);
2080         *request = NULL;
2081
2082         tgt = lmv_fid2tgt(lmv, &op_data->op_fid2);
2083         if (IS_ERR(tgt))
2084                 RETURN(PTR_ERR(tgt));
2085
2086         op_data->op_mds = tgt->ltd_index;
2087         goto retry;
2088 }
2089
2090 static int
2091 lmv_enqueue(struct obd_export *exp, struct ldlm_enqueue_info *einfo,
2092             const union ldlm_policy_data *policy, struct md_op_data *op_data,
2093             struct lustre_handle *lockh, __u64 extra_lock_flags)
2094 {
2095         struct obd_device *obd = exp->exp_obd;
2096         struct lmv_obd *lmv = &obd->u.lmv;
2097         struct lmv_tgt_desc *tgt;
2098         int rc;
2099
2100         ENTRY;
2101
2102         CDEBUG(D_INODE, "ENQUEUE on "DFID"\n", PFID(&op_data->op_fid1));
2103
2104         tgt = lmv_fid2tgt(lmv, &op_data->op_fid1);
2105         if (IS_ERR(tgt))
2106                 RETURN(PTR_ERR(tgt));
2107
2108         CDEBUG(D_INODE, "ENQUEUE on "DFID" -> mds #%u\n",
2109                PFID(&op_data->op_fid1), tgt->ltd_index);
2110
2111         rc = md_enqueue(tgt->ltd_exp, einfo, policy, op_data, lockh,
2112                         extra_lock_flags);
2113
2114         RETURN(rc);
2115 }
2116
2117 int
2118 lmv_getattr_name(struct obd_export *exp,struct md_op_data *op_data,
2119                  struct ptlrpc_request **preq)
2120 {
2121         struct obd_device *obd = exp->exp_obd;
2122         struct lmv_obd *lmv = &obd->u.lmv;
2123         struct lmv_tgt_desc *tgt;
2124         struct mdt_body *body;
2125         int rc;
2126
2127         ENTRY;
2128
2129 retry:
2130         if (op_data->op_namelen == 2 &&
2131             op_data->op_name[0] == '.' && op_data->op_name[1] == '.')
2132                 tgt = lmv_fid2tgt(lmv, &op_data->op_fid1);
2133         else
2134                 tgt = lmv_locate_tgt(lmv, op_data);
2135         if (IS_ERR(tgt))
2136                 RETURN(PTR_ERR(tgt));
2137
2138         CDEBUG(D_INODE, "GETATTR_NAME for %*s on "DFID" -> mds #%d\n",
2139                 (int)op_data->op_namelen, op_data->op_name,
2140                 PFID(&op_data->op_fid1), tgt->ltd_index);
2141
2142         rc = md_getattr_name(tgt->ltd_exp, op_data, preq);
2143         if (rc == -ENOENT && lmv_dir_retry_check_update(op_data)) {
2144                 ptlrpc_req_finished(*preq);
2145                 *preq = NULL;
2146                 goto retry;
2147         }
2148
2149         if (rc)
2150                 RETURN(rc);
2151
2152         body = req_capsule_server_get(&(*preq)->rq_pill, &RMF_MDT_BODY);
2153         LASSERT(body != NULL);
2154
2155         if (body->mbo_valid & OBD_MD_MDS) {
2156                 op_data->op_fid1 = body->mbo_fid1;
2157                 op_data->op_valid |= OBD_MD_FLCROSSREF;
2158                 op_data->op_namelen = 0;
2159                 op_data->op_name = NULL;
2160
2161                 ptlrpc_req_finished(*preq);
2162                 *preq = NULL;
2163
2164                 goto retry;
2165         }
2166
2167         RETURN(rc);
2168 }
2169
2170 #define md_op_data_fid(op_data, fl)                     \
2171         (fl == MF_MDC_CANCEL_FID1 ? &op_data->op_fid1 : \
2172          fl == MF_MDC_CANCEL_FID2 ? &op_data->op_fid2 : \
2173          fl == MF_MDC_CANCEL_FID3 ? &op_data->op_fid3 : \
2174          fl == MF_MDC_CANCEL_FID4 ? &op_data->op_fid4 : \
2175          NULL)
2176
2177 static int lmv_early_cancel(struct obd_export *exp, struct lmv_tgt_desc *tgt,
2178                             struct md_op_data *op_data, __u32 op_tgt,
2179                             enum ldlm_mode mode, int bits, int flag)
2180 {
2181         struct lu_fid *fid = md_op_data_fid(op_data, flag);
2182         struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
2183         union ldlm_policy_data policy = { { 0 } };
2184         int rc = 0;
2185         ENTRY;
2186
2187         if (!fid_is_sane(fid))
2188                 RETURN(0);
2189
2190         if (tgt == NULL) {
2191                 tgt = lmv_fid2tgt(lmv, fid);
2192                 if (IS_ERR(tgt))
2193                         RETURN(PTR_ERR(tgt));
2194         }
2195
2196         if (tgt->ltd_index != op_tgt) {
2197                 CDEBUG(D_INODE, "EARLY_CANCEL on "DFID"\n", PFID(fid));
2198                 policy.l_inodebits.bits = bits;
2199                 rc = md_cancel_unused(tgt->ltd_exp, fid, &policy,
2200                                       mode, LCF_ASYNC, NULL);
2201         } else {
2202                 CDEBUG(D_INODE,
2203                        "EARLY_CANCEL skip operation target %d on "DFID"\n",
2204                        op_tgt, PFID(fid));
2205                 op_data->op_flags |= flag;
2206                 rc = 0;
2207         }
2208
2209         RETURN(rc);
2210 }
2211
2212 /*
2213  * llite passes fid of an target inode in op_data->op_fid1 and id of directory in
2214  * op_data->op_fid2
2215  */
2216 static int lmv_link(struct obd_export *exp, struct md_op_data *op_data,
2217                     struct ptlrpc_request **request)
2218 {
2219         struct obd_device       *obd = exp->exp_obd;
2220         struct lmv_obd          *lmv = &obd->u.lmv;
2221         struct lmv_tgt_desc     *tgt;
2222         int                      rc;
2223         ENTRY;
2224
2225         LASSERT(op_data->op_namelen != 0);
2226
2227         CDEBUG(D_INODE, "LINK "DFID":%*s to "DFID"\n",
2228                PFID(&op_data->op_fid2), (int)op_data->op_namelen,
2229                op_data->op_name, PFID(&op_data->op_fid1));
2230
2231         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2232         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2233         op_data->op_cap = current_cap();
2234
2235         tgt = lmv_locate_tgt2(lmv, op_data);
2236         if (IS_ERR(tgt))
2237                 RETURN(PTR_ERR(tgt));
2238
2239         /*
2240          * Cancel UPDATE lock on child (fid1).
2241          */
2242         op_data->op_flags |= MF_MDC_CANCEL_FID2;
2243         rc = lmv_early_cancel(exp, NULL, op_data, tgt->ltd_index, LCK_EX,
2244                               MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID1);
2245         if (rc != 0)
2246                 RETURN(rc);
2247
2248         rc = md_link(tgt->ltd_exp, op_data, request);
2249
2250         RETURN(rc);
2251 }
2252
2253 /* migrate the top directory */
2254 static inline bool lmv_op_topdir_migrate(const struct md_op_data *op_data)
2255 {
2256         if (!S_ISDIR(op_data->op_mode))
2257                 return false;
2258
2259         if (lmv_dir_layout_changing(op_data->op_mea1))
2260                 return false;
2261
2262         return true;
2263 }
2264
2265 /* migrate top dir to specific MDTs */
2266 static inline bool lmv_topdir_specific_migrate(const struct md_op_data *op_data)
2267 {
2268         const struct lmv_user_md *lum = op_data->op_data;
2269
2270         if (!lmv_op_topdir_migrate(op_data))
2271                 return false;
2272
2273         return le32_to_cpu(lum->lum_stripe_offset) != LMV_OFFSET_DEFAULT;
2274 }
2275
2276 /* migrate top dir in QoS mode if user issued "lfs migrate -m -1..." */
2277 static inline bool lmv_topdir_qos_migrate(const struct md_op_data *op_data)
2278 {
2279         const struct lmv_user_md *lum = op_data->op_data;
2280
2281         if (!lmv_op_topdir_migrate(op_data))
2282                 return false;
2283
2284         return le32_to_cpu(lum->lum_stripe_offset) == LMV_OFFSET_DEFAULT;
2285 }
2286
2287 static inline bool lmv_subdir_specific_migrate(const struct md_op_data *op_data)
2288 {
2289         const struct lmv_user_md *lum = op_data->op_data;
2290
2291         if (!S_ISDIR(op_data->op_mode))
2292                 return false;
2293
2294         if (!lmv_dir_layout_changing(op_data->op_mea1))
2295                 return false;
2296
2297         return le32_to_cpu(lum->lum_stripe_offset) != LMV_OFFSET_DEFAULT;
2298 }
2299
2300 static int lmv_migrate(struct obd_export *exp, struct md_op_data *op_data,
2301                         const char *name, size_t namelen,
2302                         struct ptlrpc_request **request)
2303 {
2304         struct obd_device *obd = exp->exp_obd;
2305         struct lmv_obd *lmv = &obd->u.lmv;
2306         struct lmv_stripe_md *lsm = op_data->op_mea1;
2307         struct lmv_tgt_desc *parent_tgt;
2308         struct lmv_tgt_desc *sp_tgt;
2309         struct lmv_tgt_desc *tp_tgt = NULL;
2310         struct lmv_tgt_desc *child_tgt;
2311         struct lmv_tgt_desc *tgt;
2312         struct lu_fid target_fid = { 0 };
2313         int rc;
2314
2315         ENTRY;
2316
2317         LASSERT(op_data->op_cli_flags & CLI_MIGRATE);
2318
2319         CDEBUG(D_INODE, "MIGRATE "DFID"/%.*s\n",
2320                PFID(&op_data->op_fid1), (int)namelen, name);
2321
2322         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2323         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2324         op_data->op_cap = current_cap();
2325
2326         parent_tgt = lmv_fid2tgt(lmv, &op_data->op_fid1);
2327         if (IS_ERR(parent_tgt))
2328                 RETURN(PTR_ERR(parent_tgt));
2329
2330         if (lmv_dir_striped(lsm)) {
2331                 const struct lmv_oinfo *oinfo;
2332
2333                 oinfo = lsm_name_to_stripe_info(lsm, name, namelen, false);
2334                 if (IS_ERR(oinfo))
2335                         RETURN(PTR_ERR(oinfo));
2336
2337                 /* save source stripe FID in fid4 temporarily for ELC */
2338                 op_data->op_fid4 = oinfo->lmo_fid;
2339                 sp_tgt = lmv_tgt(lmv, oinfo->lmo_mds);
2340                 if (!sp_tgt)
2341                         RETURN(-ENODEV);
2342
2343                 /*
2344                  * if parent is being migrated too, fill op_fid2 with target
2345                  * stripe fid, otherwise the target stripe is not created yet.
2346                  */
2347                 if (lmv_dir_layout_changing(lsm)) {
2348                         oinfo = lsm_name_to_stripe_info(lsm, name, namelen,
2349                                                         true);
2350                         if (IS_ERR(oinfo))
2351                                 RETURN(PTR_ERR(oinfo));
2352
2353                         op_data->op_fid2 = oinfo->lmo_fid;
2354                         tp_tgt = lmv_tgt(lmv, oinfo->lmo_mds);
2355                         if (!tp_tgt)
2356                                 RETURN(-ENODEV);
2357
2358                         /* parent unchanged and update namespace only */
2359                         if (lu_fid_eq(&op_data->op_fid4, &op_data->op_fid2) &&
2360                             op_data->op_bias & MDS_MIGRATE_NSONLY)
2361                                 RETURN(-EALREADY);
2362                 }
2363         } else {
2364                 sp_tgt = parent_tgt;
2365         }
2366
2367         child_tgt = lmv_fid2tgt(lmv, &op_data->op_fid3);
2368         if (IS_ERR(child_tgt))
2369                 RETURN(PTR_ERR(child_tgt));
2370
2371         if (lmv_topdir_specific_migrate(op_data)) {
2372                 struct lmv_user_md *lum = op_data->op_data;
2373
2374                 op_data->op_mds = le32_to_cpu(lum->lum_stripe_offset);
2375         } else if (lmv_topdir_qos_migrate(op_data)) {
2376                 tgt = lmv_locate_tgt_lf(lmv);
2377                 if (tgt == ERR_PTR(-EAGAIN))
2378                         tgt = lmv_locate_tgt_rr(lmv);
2379                 if (IS_ERR(tgt))
2380                         RETURN(PTR_ERR(tgt));
2381
2382                 op_data->op_mds = tgt->ltd_index;
2383         } else if (lmv_subdir_specific_migrate(op_data)) {
2384                 struct lmv_user_md *lum = op_data->op_data;
2385                 __u32 i;
2386
2387                 LASSERT(tp_tgt);
2388                 if (le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC_SPECIFIC) {
2389                         /* adjust MDTs in lum, since subdir is located on where
2390                          * its parent stripe is, not the first specified MDT.
2391                          */
2392                         for (i = 0; i < le32_to_cpu(lum->lum_stripe_count);
2393                              i++) {
2394                                 if (le32_to_cpu(lum->lum_objects[i].lum_mds) ==
2395                                     tp_tgt->ltd_index)
2396                                         break;
2397                         }
2398
2399                         if (i == le32_to_cpu(lum->lum_stripe_count))
2400                                 RETURN(-ENODEV);
2401
2402                         lum->lum_objects[i].lum_mds =
2403                                 lum->lum_objects[0].lum_mds;
2404                         lum->lum_objects[0].lum_mds =
2405                                 cpu_to_le32(tp_tgt->ltd_index);
2406                 }
2407                 /* NB, the above adjusts subdir migration for command like
2408                  * "lfs migrate -m 0,1,2 ...", but for migration like
2409                  * "lfs migrate -m 0 -c 2 ...", the top dir is migrated to MDT0
2410                  * and MDT1, however its subdir may be migrated to MDT1 and MDT2
2411                  */
2412
2413                 lum->lum_stripe_offset = cpu_to_le32(tp_tgt->ltd_index);
2414                 op_data->op_mds = tp_tgt->ltd_index;
2415         } else if (tp_tgt) {
2416                 op_data->op_mds = tp_tgt->ltd_index;
2417         } else {
2418                 op_data->op_mds = sp_tgt->ltd_index;
2419         }
2420
2421         rc = lmv_fid_alloc(NULL, exp, &target_fid, op_data);
2422         if (rc)
2423                 RETURN(rc);
2424
2425         /*
2426          * for directory, send migrate request to the MDT where the object will
2427          * be migrated to, because we can't create a striped directory remotely.
2428          *
2429          * otherwise, send to the MDT where source is located because regular
2430          * file may open lease.
2431          *
2432          * NB. if MDT doesn't support DIR_MIGRATE, send to source MDT too for
2433          * backward compatibility.
2434          */
2435         if (S_ISDIR(op_data->op_mode) &&
2436             (exp_connect_flags2(exp) & OBD_CONNECT2_DIR_MIGRATE)) {
2437                 tgt = lmv_fid2tgt(lmv, &target_fid);
2438                 if (IS_ERR(tgt))
2439                         RETURN(PTR_ERR(tgt));
2440         } else {
2441                 tgt = child_tgt;
2442         }
2443
2444         /* cancel UPDATE lock of parent master object */
2445         rc = lmv_early_cancel(exp, parent_tgt, op_data, tgt->ltd_index, LCK_EX,
2446                               MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID1);
2447         if (rc)
2448                 RETURN(rc);
2449
2450         /* cancel UPDATE lock of source parent */
2451         if (sp_tgt != parent_tgt) {
2452                 /*
2453                  * migrate RPC packs master object FID, because we can only pack
2454                  * two FIDs in reint RPC, but MDS needs to know both source
2455                  * parent and target parent, and it will obtain them from master
2456                  * FID and LMV, the other FID in RPC is kept for target.
2457                  *
2458                  * since this FID is not passed to MDC, cancel it anyway.
2459                  */
2460                 rc = lmv_early_cancel(exp, sp_tgt, op_data, -1, LCK_EX,
2461                                       MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID4);
2462                 if (rc)
2463                         RETURN(rc);
2464
2465                 op_data->op_flags &= ~MF_MDC_CANCEL_FID4;
2466         }
2467         op_data->op_fid4 = target_fid;
2468
2469         /* cancel UPDATE locks of target parent */
2470         rc = lmv_early_cancel(exp, tp_tgt, op_data, tgt->ltd_index, LCK_EX,
2471                               MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID2);
2472         if (rc)
2473                 RETURN(rc);
2474
2475         /* cancel LOOKUP lock of source if source is remote object */
2476         if (child_tgt != sp_tgt) {
2477                 rc = lmv_early_cancel(exp, sp_tgt, op_data, tgt->ltd_index,
2478                                       LCK_EX, MDS_INODELOCK_LOOKUP,
2479                                       MF_MDC_CANCEL_FID3);
2480                 if (rc)
2481                         RETURN(rc);
2482         }
2483
2484         /* cancel ELC locks of source */
2485         rc = lmv_early_cancel(exp, child_tgt, op_data, tgt->ltd_index, LCK_EX,
2486                               MDS_INODELOCK_ELC, MF_MDC_CANCEL_FID3);
2487         if (rc)
2488                 RETURN(rc);
2489
2490         rc = md_rename(tgt->ltd_exp, op_data, name, namelen, NULL, 0, request);
2491
2492         RETURN(rc);
2493 }
2494
2495 static int lmv_rename(struct obd_export *exp, struct md_op_data *op_data,
2496                       const char *old, size_t oldlen,
2497                       const char *new, size_t newlen,
2498                       struct ptlrpc_request **request)
2499 {
2500         struct obd_device *obd = exp->exp_obd;
2501         struct lmv_obd *lmv = &obd->u.lmv;
2502         struct lmv_tgt_desc *sp_tgt;
2503         struct lmv_tgt_desc *tp_tgt = NULL;
2504         struct lmv_tgt_desc *src_tgt = NULL;
2505         struct lmv_tgt_desc *tgt;
2506         struct mdt_body *body;
2507         int rc;
2508
2509         ENTRY;
2510
2511         LASSERT(oldlen != 0);
2512
2513         if (op_data->op_cli_flags & CLI_MIGRATE) {
2514                 rc = lmv_migrate(exp, op_data, old, oldlen, request);
2515                 RETURN(rc);
2516         }
2517
2518         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2519         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2520         op_data->op_cap = current_cap();
2521
2522         op_data->op_name = new;
2523         op_data->op_namelen = newlen;
2524
2525         tp_tgt = lmv_locate_tgt2(lmv, op_data);
2526         if (IS_ERR(tp_tgt))
2527                 RETURN(PTR_ERR(tp_tgt));
2528
2529         /* Since the target child might be destroyed, and it might become
2530          * orphan, and we can only check orphan on the local MDT right now, so
2531          * we send rename request to the MDT where target child is located. If
2532          * target child does not exist, then it will send the request to the
2533          * target parent */
2534         if (fid_is_sane(&op_data->op_fid4)) {
2535                 tgt = lmv_fid2tgt(lmv, &op_data->op_fid4);
2536                 if (IS_ERR(tgt))
2537                         RETURN(PTR_ERR(tgt));
2538         } else {
2539                 tgt = tp_tgt;
2540         }
2541
2542         op_data->op_flags |= MF_MDC_CANCEL_FID4;
2543
2544         /* cancel UPDATE locks of target parent */
2545         rc = lmv_early_cancel(exp, tp_tgt, op_data, tgt->ltd_index, LCK_EX,
2546                               MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID2);
2547         if (rc != 0)
2548                 RETURN(rc);
2549
2550         if (fid_is_sane(&op_data->op_fid4)) {
2551                 /* cancel LOOKUP lock of target on target parent */
2552                 if (tgt != tp_tgt) {
2553                         rc = lmv_early_cancel(exp, tp_tgt, op_data,
2554                                               tgt->ltd_index, LCK_EX,
2555                                               MDS_INODELOCK_LOOKUP,
2556                                               MF_MDC_CANCEL_FID4);
2557                         if (rc != 0)
2558                                 RETURN(rc);
2559                 }
2560         }
2561
2562         if (fid_is_sane(&op_data->op_fid3)) {
2563                 src_tgt = lmv_fid2tgt(lmv, &op_data->op_fid3);
2564                 if (IS_ERR(src_tgt))
2565                         RETURN(PTR_ERR(src_tgt));
2566
2567                 /* cancel ELC locks of source */
2568                 rc = lmv_early_cancel(exp, src_tgt, op_data, tgt->ltd_index,
2569                                       LCK_EX, MDS_INODELOCK_ELC,
2570                                       MF_MDC_CANCEL_FID3);
2571                 if (rc != 0)
2572                         RETURN(rc);
2573         }
2574
2575         op_data->op_name = old;
2576         op_data->op_namelen = oldlen;
2577 retry:
2578         sp_tgt = lmv_locate_tgt(lmv, op_data);
2579         if (IS_ERR(sp_tgt))
2580                 RETURN(PTR_ERR(sp_tgt));
2581
2582         /* cancel UPDATE locks of source parent */
2583         rc = lmv_early_cancel(exp, sp_tgt, op_data, tgt->ltd_index, LCK_EX,
2584                               MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID1);
2585         if (rc != 0)
2586                 RETURN(rc);
2587
2588         if (fid_is_sane(&op_data->op_fid3)) {
2589                 /* cancel LOOKUP lock of source on source parent */
2590                 if (src_tgt != sp_tgt) {
2591                         rc = lmv_early_cancel(exp, sp_tgt, op_data,
2592                                               tgt->ltd_index, LCK_EX,
2593                                               MDS_INODELOCK_LOOKUP,
2594                                               MF_MDC_CANCEL_FID3);
2595                         if (rc != 0)
2596                                 RETURN(rc);
2597                 }
2598         }
2599
2600 rename:
2601         CDEBUG(D_INODE, "RENAME "DFID"/%.*s to "DFID"/%.*s\n",
2602                 PFID(&op_data->op_fid1), (int)oldlen, old,
2603                 PFID(&op_data->op_fid2), (int)newlen, new);
2604
2605         rc = md_rename(tgt->ltd_exp, op_data, old, oldlen, new, newlen,
2606                         request);
2607         if (rc == -ENOENT && lmv_dir_retry_check_update(op_data)) {
2608                 ptlrpc_req_finished(*request);
2609                 *request = NULL;
2610                 goto retry;
2611         }
2612
2613         if (rc && rc != -EXDEV)
2614                 RETURN(rc);
2615
2616         body = req_capsule_server_get(&(*request)->rq_pill, &RMF_MDT_BODY);
2617         if (body == NULL)
2618                 RETURN(-EPROTO);
2619
2620         /* Not cross-ref case, just get out of here. */
2621         if (likely(!(body->mbo_valid & OBD_MD_MDS)))
2622                 RETURN(rc);
2623
2624         op_data->op_fid4 = body->mbo_fid1;
2625
2626         ptlrpc_req_finished(*request);
2627         *request = NULL;
2628
2629         tgt = lmv_fid2tgt(lmv, &op_data->op_fid4);
2630         if (IS_ERR(tgt))
2631                 RETURN(PTR_ERR(tgt));
2632
2633         if (fid_is_sane(&op_data->op_fid4)) {
2634                 /* cancel LOOKUP lock of target on target parent */
2635                 if (tgt != tp_tgt) {
2636                         rc = lmv_early_cancel(exp, tp_tgt, op_data,
2637                                               tgt->ltd_index, LCK_EX,
2638                                               MDS_INODELOCK_LOOKUP,
2639                                               MF_MDC_CANCEL_FID4);
2640                         if (rc != 0)
2641                                 RETURN(rc);
2642                 }
2643         }
2644
2645         goto rename;
2646 }
2647
2648 static int lmv_setattr(struct obd_export *exp, struct md_op_data *op_data,
2649                        void *ea, size_t ealen, struct ptlrpc_request **request)
2650 {
2651         struct obd_device *obd = exp->exp_obd;
2652         struct lmv_obd *lmv = &obd->u.lmv;
2653         struct lmv_tgt_desc *tgt;
2654         int rc = 0;
2655
2656         ENTRY;
2657
2658         CDEBUG(D_INODE, "SETATTR for "DFID", valid 0x%x/0x%x\n",
2659                PFID(&op_data->op_fid1), op_data->op_attr.ia_valid,
2660                op_data->op_xvalid);
2661
2662         op_data->op_flags |= MF_MDC_CANCEL_FID1;
2663         tgt = lmv_fid2tgt(lmv, &op_data->op_fid1);
2664         if (IS_ERR(tgt))
2665                 RETURN(PTR_ERR(tgt));
2666
2667         rc = md_setattr(tgt->ltd_exp, op_data, ea, ealen, request);
2668
2669         RETURN(rc);
2670 }
2671
2672 static int lmv_fsync(struct obd_export *exp, const struct lu_fid *fid,
2673                      struct ptlrpc_request **request)
2674 {
2675         struct obd_device *obd = exp->exp_obd;
2676         struct lmv_obd *lmv = &obd->u.lmv;
2677         struct lmv_tgt_desc *tgt;
2678         int rc;
2679
2680         ENTRY;
2681
2682         tgt = lmv_fid2tgt(lmv, fid);
2683         if (IS_ERR(tgt))
2684                 RETURN(PTR_ERR(tgt));
2685
2686         rc = md_fsync(tgt->ltd_exp, fid, request);
2687         RETURN(rc);
2688 }
2689
2690 struct stripe_dirent {
2691         struct page             *sd_page;
2692         struct lu_dirpage       *sd_dp;
2693         struct lu_dirent        *sd_ent;
2694         bool                     sd_eof;
2695 };
2696
2697 struct lmv_dir_ctxt {
2698         struct lmv_obd          *ldc_lmv;
2699         struct md_op_data       *ldc_op_data;
2700         struct md_readdir_info  *ldc_mrinfo;
2701         __u64                    ldc_hash;
2702         int                      ldc_count;
2703         struct stripe_dirent     ldc_stripes[0];
2704 };
2705
2706 static inline void stripe_dirent_unload(struct stripe_dirent *stripe)
2707 {
2708         if (stripe->sd_page) {
2709                 kunmap(stripe->sd_page);
2710                 put_page(stripe->sd_page);
2711                 stripe->sd_page = NULL;
2712                 stripe->sd_ent = NULL;
2713         }
2714 }
2715
2716 static inline void put_lmv_dir_ctxt(struct lmv_dir_ctxt *ctxt)
2717 {
2718         int i;
2719
2720         for (i = 0; i < ctxt->ldc_count; i++)
2721                 stripe_dirent_unload(&ctxt->ldc_stripes[i]);
2722 }
2723
2724 /* if @ent is dummy, or . .., get next */
2725 static struct lu_dirent *stripe_dirent_get(struct lmv_dir_ctxt *ctxt,
2726                                            struct lu_dirent *ent,
2727                                            int stripe_index)
2728 {
2729         for (; ent; ent = lu_dirent_next(ent)) {
2730                 /* Skip dummy entry */
2731                 if (le16_to_cpu(ent->lde_namelen) == 0)
2732                         continue;
2733
2734                 /* skip . and .. for other stripes */
2735                 if (stripe_index &&
2736                     (strncmp(ent->lde_name, ".",
2737                              le16_to_cpu(ent->lde_namelen)) == 0 ||
2738                      strncmp(ent->lde_name, "..",
2739                              le16_to_cpu(ent->lde_namelen)) == 0))
2740                         continue;
2741
2742                 if (le64_to_cpu(ent->lde_hash) >= ctxt->ldc_hash)
2743                         break;
2744         }
2745
2746         return ent;
2747 }
2748
2749 static struct lu_dirent *stripe_dirent_load(struct lmv_dir_ctxt *ctxt,
2750                                             struct stripe_dirent *stripe,
2751                                             int stripe_index)
2752 {
2753         struct md_op_data *op_data = ctxt->ldc_op_data;
2754         struct lmv_oinfo *oinfo;
2755         struct lu_fid fid = op_data->op_fid1;
2756         struct inode *inode = op_data->op_data;
2757         struct lmv_tgt_desc *tgt;
2758         struct lu_dirent *ent = stripe->sd_ent;
2759         __u64 hash = ctxt->ldc_hash;
2760         int rc = 0;
2761
2762         ENTRY;
2763
2764         LASSERT(stripe == &ctxt->ldc_stripes[stripe_index]);
2765         LASSERT(!ent);
2766
2767         do {
2768                 if (stripe->sd_page) {
2769                         __u64 end = le64_to_cpu(stripe->sd_dp->ldp_hash_end);
2770
2771                         /* @hash should be the last dirent hash */
2772                         LASSERTF(hash <= end,
2773                                  "ctxt@%p stripe@%p hash %llx end %llx\n",
2774                                  ctxt, stripe, hash, end);
2775                         /* unload last page */
2776                         stripe_dirent_unload(stripe);
2777                         /* eof */
2778                         if (end == MDS_DIR_END_OFF) {
2779                                 stripe->sd_eof = true;
2780                                 break;
2781                         }
2782                         hash = end;
2783                 }
2784
2785                 oinfo = &op_data->op_mea1->lsm_md_oinfo[stripe_index];
2786                 if (!oinfo->lmo_root) {
2787                         rc = -ENOENT;
2788                         break;
2789                 }
2790
2791                 tgt = lmv_tgt(ctxt->ldc_lmv, oinfo->lmo_mds);
2792                 if (!tgt) {
2793                         rc = -ENODEV;
2794                         break;
2795                 }
2796
2797                 /* op_data is shared by stripes, reset after use */
2798                 op_data->op_fid1 = oinfo->lmo_fid;
2799                 op_data->op_fid2 = oinfo->lmo_fid;
2800                 op_data->op_data = oinfo->lmo_root;
2801
2802                 rc = md_read_page(tgt->ltd_exp, op_data, ctxt->ldc_mrinfo, hash,
2803                                   &stripe->sd_page);
2804
2805                 op_data->op_fid1 = fid;
2806                 op_data->op_fid2 = fid;
2807                 op_data->op_data = inode;
2808
2809                 if (rc)
2810                         break;
2811
2812                 stripe->sd_dp = page_address(stripe->sd_page);
2813                 ent = stripe_dirent_get(ctxt, lu_dirent_start(stripe->sd_dp),
2814                                         stripe_index);
2815                 /* in case a page filled with ., .. and dummy, read next */
2816         } while (!ent);
2817
2818         stripe->sd_ent = ent;
2819         if (rc) {
2820                 LASSERT(!ent);
2821                 /* treat error as eof, so dir can be partially accessed */
2822                 stripe->sd_eof = true;
2823                 ctxt->ldc_mrinfo->mr_partial_readdir_rc = rc;
2824                 LCONSOLE_WARN("dir "DFID" stripe %d readdir failed: %d, "
2825                               "directory is partially accessed!\n",
2826                               PFID(&ctxt->ldc_op_data->op_fid1), stripe_index,
2827                               rc);
2828         }
2829
2830         RETURN(ent);
2831 }
2832
2833 static int lmv_file_resync(struct obd_export *exp, struct md_op_data *data)
2834 {
2835         struct obd_device *obd = exp->exp_obd;
2836         struct lmv_obd *lmv = &obd->u.lmv;
2837         struct lmv_tgt_desc *tgt;
2838         int rc;
2839
2840         ENTRY;
2841
2842         rc = lmv_check_connect(obd);
2843         if (rc != 0)
2844                 RETURN(rc);
2845
2846         tgt = lmv_fid2tgt(lmv, &data->op_fid1);
2847         if (IS_ERR(tgt))
2848                 RETURN(PTR_ERR(tgt));
2849
2850         data->op_flags |= MF_MDC_CANCEL_FID1;
2851         rc = md_file_resync(tgt->ltd_exp, data);
2852         RETURN(rc);
2853 }
2854
2855 /**
2856  * Get dirent with the closest hash for striped directory
2857  *
2858  * This function will search the dir entry, whose hash value is the
2859  * closest(>=) to hash from all of sub-stripes, and it is only being called
2860  * for striped directory.
2861  *
2862  * \param[in] ctxt              dir read context
2863  *
2864  * \retval                      dirent get the entry successfully
2865  *                              NULL does not get the entry, normally it means
2866  *                              it reaches the end of the directory, while read
2867  *                              stripe dirent error is ignored to allow partial
2868  *                              access.
2869  */
2870 static struct lu_dirent *lmv_dirent_next(struct lmv_dir_ctxt *ctxt)
2871 {
2872         struct stripe_dirent *stripe;
2873         struct lu_dirent *ent = NULL;
2874         int i;
2875         int min = -1;
2876
2877         /* TODO: optimize with k-way merge sort */
2878         for (i = 0; i < ctxt->ldc_count; i++) {
2879                 stripe = &ctxt->ldc_stripes[i];
2880                 if (stripe->sd_eof)
2881                         continue;
2882
2883                 if (!stripe->sd_ent) {
2884                         stripe_dirent_load(ctxt, stripe, i);
2885                         if (!stripe->sd_ent) {
2886                                 LASSERT(stripe->sd_eof);
2887                                 continue;
2888                         }
2889                 }
2890
2891                 if (min == -1 ||
2892                     le64_to_cpu(ctxt->ldc_stripes[min].sd_ent->lde_hash) >
2893                     le64_to_cpu(stripe->sd_ent->lde_hash)) {
2894                         min = i;
2895                         if (le64_to_cpu(stripe->sd_ent->lde_hash) ==
2896                             ctxt->ldc_hash)
2897                                 break;
2898                 }
2899         }
2900
2901         if (min != -1) {
2902                 stripe = &ctxt->ldc_stripes[min];
2903                 ent = stripe->sd_ent;
2904                 /* pop found dirent */
2905                 stripe->sd_ent = stripe_dirent_get(ctxt, lu_dirent_next(ent),
2906                                                    min);
2907         }
2908
2909         return ent;
2910 }
2911
2912 /**
2913  * Build dir entry page for striped directory
2914  *
2915  * This function gets one entry by @offset from a striped directory. It will
2916  * read entries from all of stripes, and choose one closest to the required
2917  * offset(&offset). A few notes
2918  * 1. skip . and .. for non-zero stripes, because there can only have one .
2919  * and .. in a directory.
2920  * 2. op_data will be shared by all of stripes, instead of allocating new
2921  * one, so need to restore before reusing.
2922  *
2923  * \param[in] exp       obd export refer to LMV
2924  * \param[in] op_data   hold those MD parameters of read_entry
2925  * \param[in] mrinfo    ldlm callback being used in enqueue in mdc_read_entry,
2926  *                      and partial readdir result will be stored in it.
2927  * \param[in] offset    starting hash offset
2928  * \param[out] ppage    the page holding the entry. Note: because the entry
2929  *                      will be accessed in upper layer, so we need hold the
2930  *                      page until the usages of entry is finished, see
2931  *                      ll_dir_entry_next.
2932  *
2933  * retval               =0 if get entry successfully
2934  *                      <0 cannot get entry
2935  */
2936 static int lmv_striped_read_page(struct obd_export *exp,
2937                                  struct md_op_data *op_data,
2938                                  struct md_readdir_info *mrinfo, __u64 offset,
2939                                  struct page **ppage)
2940 {
2941         struct page *page = NULL;
2942         struct lu_dirpage *dp;
2943         void *start;
2944         struct lu_dirent *ent;
2945         struct lu_dirent *last_ent;
2946         int stripe_count;
2947         struct lmv_dir_ctxt *ctxt;
2948         struct lu_dirent *next = NULL;
2949         __u16 ent_size;
2950         size_t left_bytes;
2951         int rc = 0;
2952         ENTRY;
2953
2954         /* Allocate a page and read entries from all of stripes and fill
2955          * the page by hash order */
2956         page = alloc_page(GFP_KERNEL);
2957         if (!page)
2958                 RETURN(-ENOMEM);
2959
2960         /* Initialize the entry page */
2961         dp = kmap(page);
2962         memset(dp, 0, sizeof(*dp));
2963         dp->ldp_hash_start = cpu_to_le64(offset);
2964
2965         start = dp + 1;
2966         left_bytes = PAGE_SIZE - sizeof(*dp);
2967         ent = start;
2968         last_ent = ent;
2969
2970         /* initalize dir read context */
2971         stripe_count = op_data->op_mea1->lsm_md_stripe_count;
2972         OBD_ALLOC(ctxt, offsetof(typeof(*ctxt), ldc_stripes[stripe_count]));
2973         if (!ctxt)
2974                 GOTO(free_page, rc = -ENOMEM);
2975         ctxt->ldc_lmv = &exp->exp_obd->u.lmv;
2976         ctxt->ldc_op_data = op_data;
2977         ctxt->ldc_mrinfo = mrinfo;
2978         ctxt->ldc_hash = offset;
2979         ctxt->ldc_count = stripe_count;
2980
2981         while (1) {
2982                 next = lmv_dirent_next(ctxt);
2983
2984                 /* end of directory */
2985                 if (!next) {
2986                         ctxt->ldc_hash = MDS_DIR_END_OFF;
2987                         break;
2988                 }
2989                 ctxt->ldc_hash = le64_to_cpu(next->lde_hash);
2990
2991                 ent_size = le16_to_cpu(next->lde_reclen);
2992
2993                 /* the last entry lde_reclen is 0, but it might not be the last
2994                  * one of this temporay dir page */
2995                 if (!ent_size)
2996                         ent_size = lu_dirent_calc_size(
2997                                         le16_to_cpu(next->lde_namelen),
2998                                         le32_to_cpu(next->lde_attrs));
2999                 /* page full */
3000                 if (ent_size > left_bytes)
3001                         break;
3002
3003                 memcpy(ent, next, ent_size);
3004
3005                 /* Replace . with master FID and Replace .. with the parent FID
3006                  * of master object */
3007                 if (strncmp(ent->lde_name, ".",
3008                             le16_to_cpu(ent->lde_namelen)) == 0 &&
3009                     le16_to_cpu(ent->lde_namelen) == 1)
3010                         fid_cpu_to_le(&ent->lde_fid, &op_data->op_fid1);
3011                 else if (strncmp(ent->lde_name, "..",
3012                                    le16_to_cpu(ent->lde_namelen)) == 0 &&
3013                            le16_to_cpu(ent->lde_namelen) == 2)
3014                         fid_cpu_to_le(&ent->lde_fid, &op_data->op_fid3);
3015
3016                 CDEBUG(D_INODE, "entry %.*s hash %#llx\n",
3017                        le16_to_cpu(ent->lde_namelen), ent->lde_name,
3018                        le64_to_cpu(ent->lde_hash));
3019
3020                 left_bytes -= ent_size;
3021                 ent->lde_reclen = cpu_to_le16(ent_size);
3022                 last_ent = ent;
3023                 ent = (void *)ent + ent_size;
3024         };
3025
3026         last_ent->lde_reclen = 0;
3027
3028         if (ent == start)
3029                 dp->ldp_flags |= LDF_EMPTY;
3030         else if (ctxt->ldc_hash == le64_to_cpu(last_ent->lde_hash))
3031                 dp->ldp_flags |= LDF_COLLIDE;
3032         dp->ldp_flags = cpu_to_le32(dp->ldp_flags);
3033         dp->ldp_hash_end = cpu_to_le64(ctxt->ldc_hash);
3034
3035         put_lmv_dir_ctxt(ctxt);
3036         OBD_FREE(ctxt, offsetof(typeof(*ctxt), ldc_stripes[stripe_count]));
3037
3038         *ppage = page;
3039
3040         RETURN(0);
3041
3042 free_page:
3043         kunmap(page);
3044         __free_page(page);
3045
3046         return rc;
3047 }
3048
3049 static int lmv_read_page(struct obd_export *exp, struct md_op_data *op_data,
3050                          struct md_readdir_info *mrinfo, __u64 offset,
3051                          struct page **ppage)
3052 {
3053         struct obd_device *obd = exp->exp_obd;
3054         struct lmv_obd *lmv = &obd->u.lmv;
3055         struct lmv_tgt_desc *tgt;
3056         int rc;
3057
3058         ENTRY;
3059
3060         if (unlikely(lmv_dir_foreign(op_data->op_mea1)))
3061                 RETURN(-ENODATA);
3062
3063         if (unlikely(lmv_dir_striped(op_data->op_mea1))) {
3064                 rc = lmv_striped_read_page(exp, op_data, mrinfo, offset, ppage);
3065                 RETURN(rc);
3066         }
3067
3068         tgt = lmv_fid2tgt(lmv, &op_data->op_fid1);
3069         if (IS_ERR(tgt))
3070                 RETURN(PTR_ERR(tgt));
3071
3072         rc = md_read_page(tgt->ltd_exp, op_data, mrinfo, offset, ppage);
3073
3074         RETURN(rc);
3075 }
3076
3077 /**
3078  * Unlink a file/directory
3079  *
3080  * Unlink a file or directory under the parent dir. The unlink request
3081  * usually will be sent to the MDT where the child is located, but if
3082  * the client does not have the child FID then request will be sent to the
3083  * MDT where the parent is located.
3084  *
3085  * If the parent is a striped directory then it also needs to locate which
3086  * stripe the name of the child is located, and replace the parent FID
3087  * (@op->op_fid1) with the stripe FID. Note: if the stripe is unknown,
3088  * it will walk through all of sub-stripes until the child is being
3089  * unlinked finally.
3090  *
3091  * \param[in] exp       export refer to LMV
3092  * \param[in] op_data   different parameters transferred beween client
3093  *                      MD stacks, name, namelen, FIDs etc.
3094  *                      op_fid1 is the parent FID, op_fid2 is the child
3095  *                      FID.
3096  * \param[out] request  point to the request of unlink.
3097  *
3098  * retval               0 if succeed
3099  *                      negative errno if failed.
3100  */
3101 static int lmv_unlink(struct obd_export *exp, struct md_op_data *op_data,
3102                       struct ptlrpc_request **request)
3103 {
3104         struct obd_device *obd = exp->exp_obd;
3105         struct lmv_obd *lmv = &obd->u.lmv;
3106         struct lmv_tgt_desc *tgt;
3107         struct lmv_tgt_desc *parent_tgt;
3108         struct mdt_body *body;
3109         int rc;
3110
3111         ENTRY;
3112
3113         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
3114         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
3115         op_data->op_cap = current_cap();
3116
3117 retry:
3118         parent_tgt = lmv_locate_tgt(lmv, op_data);
3119         if (IS_ERR(parent_tgt))
3120                 RETURN(PTR_ERR(parent_tgt));
3121
3122         if (likely(!fid_is_zero(&op_data->op_fid2))) {
3123                 tgt = lmv_fid2tgt(lmv, &op_data->op_fid2);
3124                 if (IS_ERR(tgt))
3125                         RETURN(PTR_ERR(tgt));
3126         } else {
3127                 tgt = parent_tgt;
3128         }
3129
3130         /*
3131          * If child's fid is given, cancel unused locks for it if it is from
3132          * another export than parent.
3133          *
3134          * LOOKUP lock for child (fid3) should also be cancelled on parent
3135          * tgt_tgt in mdc_unlink().
3136          */
3137         op_data->op_flags |= MF_MDC_CANCEL_FID1 | MF_MDC_CANCEL_FID3;
3138
3139         if (parent_tgt != tgt)
3140                 rc = lmv_early_cancel(exp, parent_tgt, op_data, tgt->ltd_index,
3141                                       LCK_EX, MDS_INODELOCK_LOOKUP,
3142                                       MF_MDC_CANCEL_FID3);
3143
3144         rc = lmv_early_cancel(exp, NULL, op_data, tgt->ltd_index, LCK_EX,
3145                               MDS_INODELOCK_ELC, MF_MDC_CANCEL_FID3);
3146         if (rc)
3147                 RETURN(rc);
3148
3149         CDEBUG(D_INODE, "unlink with fid="DFID"/"DFID" -> mds #%u\n",
3150                PFID(&op_data->op_fid1), PFID(&op_data->op_fid2),
3151                tgt->ltd_index);
3152
3153         rc = md_unlink(tgt->ltd_exp, op_data, request);
3154         if (rc == -ENOENT && lmv_dir_retry_check_update(op_data)) {
3155                 ptlrpc_req_finished(*request);
3156                 *request = NULL;
3157                 goto retry;
3158         }
3159
3160         if (rc != -EREMOTE)
3161                 RETURN(rc);
3162
3163         body = req_capsule_server_get(&(*request)->rq_pill, &RMF_MDT_BODY);
3164         if (body == NULL)
3165                 RETURN(-EPROTO);
3166
3167         /* Not cross-ref case, just get out of here. */
3168         if (likely(!(body->mbo_valid & OBD_MD_MDS)))
3169                 RETURN(rc);
3170
3171         /* This is a remote object, try remote MDT. */
3172         op_data->op_fid2 = body->mbo_fid1;
3173         ptlrpc_req_finished(*request);
3174         *request = NULL;
3175
3176         tgt = lmv_fid2tgt(lmv, &op_data->op_fid2);
3177         if (IS_ERR(tgt))
3178                 RETURN(PTR_ERR(tgt));
3179
3180         goto retry;
3181 }
3182
3183 static int lmv_precleanup(struct obd_device *obd)
3184 {
3185         ENTRY;
3186         libcfs_kkuc_group_rem(&obd->obd_uuid, 0, KUC_GRP_HSM);
3187         RETURN(0);
3188 }
3189
3190 /**
3191  * Get by key a value associated with a LMV device.
3192  *
3193  * Dispatch request to lower-layer devices as needed.
3194  *
3195  * \param[in] env               execution environment for this thread
3196  * \param[in] exp               export for the LMV device
3197  * \param[in] keylen            length of key identifier
3198  * \param[in] key               identifier of key to get value for
3199  * \param[in] vallen            size of \a val
3200  * \param[out] val              pointer to storage location for value
3201  * \param[in] lsm               optional striping metadata of object
3202  *
3203  * \retval 0            on success
3204  * \retval negative     negated errno on failure
3205  */
3206 static int lmv_get_info(const struct lu_env *env, struct obd_export *exp,
3207                         __u32 keylen, void *key, __u32 *vallen, void *val)
3208 {
3209         struct obd_device *obd;
3210         struct lmv_obd *lmv;
3211         struct lu_tgt_desc *tgt;
3212         int rc = 0;
3213
3214         ENTRY;
3215
3216         obd = class_exp2obd(exp);
3217         if (obd == NULL) {
3218                 CDEBUG(D_IOCTL, "Invalid client cookie %#llx\n",
3219                        exp->exp_handle.h_cookie);
3220                 RETURN(-EINVAL);
3221         }
3222
3223         lmv = &obd->u.lmv;
3224         if (keylen >= strlen("remote_flag") && !strcmp(key, "remote_flag")) {
3225                 LASSERT(*vallen == sizeof(__u32));
3226                 lmv_foreach_connected_tgt(lmv, tgt) {
3227                         if (!obd_get_info(env, tgt->ltd_exp, keylen, key,
3228                                           vallen, val))
3229                                 RETURN(0);
3230                 }
3231                 RETURN(-EINVAL);
3232         } else if (KEY_IS(KEY_MAX_EASIZE) ||
3233                    KEY_IS(KEY_DEFAULT_EASIZE) ||
3234                    KEY_IS(KEY_CONN_DATA)) {
3235                 /*
3236                  * Forwarding this request to first MDS, it should know LOV
3237                  * desc.
3238                  */
3239                 tgt = lmv_tgt(lmv, 0);
3240                 if (!tgt)
3241                         RETURN(-ENODEV);
3242
3243                 rc = obd_get_info(env, tgt->ltd_exp, keylen, key, vallen, val);
3244                 if (!rc && KEY_IS(KEY_CONN_DATA))
3245                         exp->exp_connect_data = *(struct obd_connect_data *)val;
3246                 RETURN(rc);
3247         } else if (KEY_IS(KEY_TGT_COUNT)) {
3248                 *((int *)val) = lmv->lmv_mdt_descs.ltd_tgts_size;
3249                 RETURN(0);
3250         }
3251
3252         CDEBUG(D_IOCTL, "Invalid key\n");
3253         RETURN(-EINVAL);
3254 }
3255
3256 static int lmv_rmfid(struct obd_export *exp, struct fid_array *fa,
3257                      int *__rcs, struct ptlrpc_request_set *_set)
3258 {
3259         struct obd_device *obd = class_exp2obd(exp);
3260         struct ptlrpc_request_set *set = _set;
3261         struct lmv_obd *lmv = &obd->u.lmv;
3262         int tgt_count = lmv->lmv_mdt_count;
3263         struct lu_tgt_desc *tgt;
3264         struct fid_array *fat, **fas = NULL;
3265         int i, rc, **rcs = NULL;
3266
3267         if (!set) {
3268                 set = ptlrpc_prep_set();
3269                 if (!set)
3270                         RETURN(-ENOMEM);
3271         }
3272
3273         /* split FIDs by targets */
3274         OBD_ALLOC_PTR_ARRAY(fas, tgt_count);
3275         if (fas == NULL)
3276                 GOTO(out, rc = -ENOMEM);
3277         OBD_ALLOC_PTR_ARRAY(rcs, tgt_count);
3278         if (rcs == NULL)
3279                 GOTO(out_fas, rc = -ENOMEM);
3280
3281         for (i = 0; i < fa->fa_nr; i++) {
3282                 unsigned int idx;
3283
3284                 rc = lmv_fld_lookup(lmv, &fa->fa_fids[i], &idx);
3285                 if (rc) {
3286                         CDEBUG(D_OTHER, "can't lookup "DFID": rc = %d\n",
3287                                PFID(&fa->fa_fids[i]), rc);
3288                         continue;
3289                 }
3290                 LASSERT(idx < tgt_count);
3291                 if (!fas[idx])
3292                         OBD_ALLOC(fas[idx], offsetof(struct fid_array,
3293                                   fa_fids[fa->fa_nr]));
3294                 if (!fas[idx])
3295                         GOTO(out, rc = -ENOMEM);
3296                 if (!rcs[idx])
3297                         OBD_ALLOC_PTR_ARRAY(rcs[idx], fa->fa_nr);
3298                 if (!rcs[idx])
3299                         GOTO(out, rc = -ENOMEM);
3300
3301                 fat = fas[idx];
3302                 fat->fa_fids[fat->fa_nr++] = fa->fa_fids[i];
3303         }
3304
3305         lmv_foreach_connected_tgt(lmv, tgt) {
3306                 fat = fas[tgt->ltd_index];
3307                 if (!fat || fat->fa_nr == 0)
3308                         continue;
3309                 rc = md_rmfid(tgt->ltd_exp, fat, rcs[tgt->ltd_index], set);
3310         }
3311
3312         rc = ptlrpc_set_wait(NULL, set);
3313         if (rc == 0) {
3314                 int j = 0;
3315                 for (i = 0; i < tgt_count; i++) {
3316                         fat = fas[i];
3317                         if (!fat || fat->fa_nr == 0)
3318                                 continue;
3319                         /* copy FIDs back */
3320                         memcpy(fa->fa_fids + j, fat->fa_fids,
3321                                fat->fa_nr * sizeof(struct lu_fid));
3322                         /* copy rcs back */
3323                         memcpy(__rcs + j, rcs[i], fat->fa_nr * sizeof(**rcs));
3324                         j += fat->fa_nr;
3325                 }
3326         }
3327         if (set != _set)
3328                 ptlrpc_set_destroy(set);
3329
3330 out:
3331         for (i = 0; i < tgt_count; i++) {
3332                 if (fas && fas[i])
3333                         OBD_FREE(fas[i], offsetof(struct fid_array,
3334                                                 fa_fids[fa->fa_nr]));
3335                 if (rcs && rcs[i])
3336                         OBD_FREE_PTR_ARRAY(rcs[i], fa->fa_nr);
3337         }
3338         if (rcs)
3339                 OBD_FREE_PTR_ARRAY(rcs, tgt_count);
3340 out_fas:
3341         if (fas)
3342                 OBD_FREE_PTR_ARRAY(fas, tgt_count);
3343
3344         RETURN(rc);
3345 }
3346
3347 /**
3348  * Asynchronously set by key a value associated with a LMV device.
3349  *
3350  * Dispatch request to lower-layer devices as needed.
3351  *
3352  * \param[in] env       execution environment for this thread
3353  * \param[in] exp       export for the LMV device
3354  * \param[in] keylen    length of key identifier
3355  * \param[in] key       identifier of key to store value for
3356  * \param[in] vallen    size of value to store
3357  * \param[in] val       pointer to data to be stored
3358  * \param[in] set       optional list of related ptlrpc requests
3359  *
3360  * \retval 0            on success
3361  * \retval negative     negated errno on failure
3362  */
3363 static int lmv_set_info_async(const struct lu_env *env, struct obd_export *exp,
3364                               __u32 keylen, void *key, __u32 vallen, void *val,
3365                               struct ptlrpc_request_set *set)
3366 {
3367         struct lmv_tgt_desc *tgt;
3368         struct obd_device *obd;
3369         struct lmv_obd *lmv;
3370         int rc = 0;
3371         ENTRY;
3372
3373         obd = class_exp2obd(exp);
3374         if (obd == NULL) {
3375                 CDEBUG(D_IOCTL, "Invalid client cookie %#llx\n",
3376                        exp->exp_handle.h_cookie);
3377                 RETURN(-EINVAL);
3378         }
3379         lmv = &obd->u.lmv;
3380
3381         if (KEY_IS(KEY_READ_ONLY) || KEY_IS(KEY_FLUSH_CTX) ||
3382             KEY_IS(KEY_DEFAULT_EASIZE)) {
3383                 int err = 0;
3384
3385                 lmv_foreach_connected_tgt(lmv, tgt) {
3386                         err = obd_set_info_async(env, tgt->ltd_exp,
3387                                                  keylen, key, vallen, val, set);
3388                         if (err && rc == 0)
3389                                 rc = err;
3390                 }
3391
3392                 RETURN(rc);
3393         }
3394
3395         RETURN(-EINVAL);
3396 }
3397
3398 static int lmv_unpack_md_v1(struct obd_export *exp, struct lmv_stripe_md *lsm,
3399                             const struct lmv_mds_md_v1 *lmm1)
3400 {
3401         struct lmv_obd  *lmv = &exp->exp_obd->u.lmv;
3402         int             stripe_count;
3403         int             cplen;
3404         int             i;
3405         int             rc = 0;
3406         ENTRY;
3407
3408         lsm->lsm_md_magic = le32_to_cpu(lmm1->lmv_magic);
3409         lsm->lsm_md_stripe_count = le32_to_cpu(lmm1->lmv_stripe_count);
3410         lsm->lsm_md_master_mdt_index = le32_to_cpu(lmm1->lmv_master_mdt_index);
3411         if (CFS_FAIL_CHECK(OBD_FAIL_LMV_UNKNOWN_STRIPE))
3412                 lsm->lsm_md_hash_type = cfs_fail_val ?: LMV_HASH_TYPE_UNKNOWN;
3413         else
3414                 lsm->lsm_md_hash_type = le32_to_cpu(lmm1->lmv_hash_type);
3415         lsm->lsm_md_layout_version = le32_to_cpu(lmm1->lmv_layout_version);
3416         lsm->lsm_md_migrate_offset = le32_to_cpu(lmm1->lmv_migrate_offset);
3417         lsm->lsm_md_migrate_hash = le32_to_cpu(lmm1->lmv_migrate_hash);
3418         cplen = strlcpy(lsm->lsm_md_pool_name, lmm1->lmv_pool_name,
3419                         sizeof(lsm->lsm_md_pool_name));
3420
3421         if (cplen >= sizeof(lsm->lsm_md_pool_name))
3422                 RETURN(-E2BIG);
3423
3424         CDEBUG(D_INFO, "unpack lsm count %d/%d, master %d hash_type %#x/%#x "
3425                "layout_version %d\n", lsm->lsm_md_stripe_count,
3426                lsm->lsm_md_migrate_offset, lsm->lsm_md_master_mdt_index,
3427                lsm->lsm_md_hash_type, lsm->lsm_md_migrate_hash,
3428                lsm->lsm_md_layout_version);
3429
3430         stripe_count = le32_to_cpu(lmm1->lmv_stripe_count);
3431         for (i = 0; i < stripe_count; i++) {
3432                 fid_le_to_cpu(&lsm->lsm_md_oinfo[i].lmo_fid,
3433                               &lmm1->lmv_stripe_fids[i]);
3434                 /*
3435                  * set default value -1, so lmv_locate_tgt() knows this stripe
3436                  * target is not initialized.
3437                  */
3438                 lsm->lsm_md_oinfo[i].lmo_mds = LMV_OFFSET_DEFAULT;
3439                 if (!fid_is_sane(&lsm->lsm_md_oinfo[i].lmo_fid))
3440                         continue;
3441
3442                 rc = lmv_fld_lookup(lmv, &lsm->lsm_md_oinfo[i].lmo_fid,
3443                                     &lsm->lsm_md_oinfo[i].lmo_mds);
3444                 if (rc == -ENOENT)
3445                         continue;
3446
3447                 if (rc)
3448                         RETURN(rc);
3449
3450                 CDEBUG(D_INFO, "unpack fid #%d "DFID"\n", i,
3451                        PFID(&lsm->lsm_md_oinfo[i].lmo_fid));
3452         }
3453
3454         RETURN(rc);
3455 }
3456
3457 static inline int lmv_unpack_user_md(struct obd_export *exp,
3458                                      struct lmv_stripe_md *lsm,
3459                                      const struct lmv_user_md *lmu)
3460 {
3461         lsm->lsm_md_magic = le32_to_cpu(lmu->lum_magic);
3462         lsm->lsm_md_stripe_count = le32_to_cpu(lmu->lum_stripe_count);
3463         lsm->lsm_md_master_mdt_index = le32_to_cpu(lmu->lum_stripe_offset);
3464         lsm->lsm_md_hash_type = le32_to_cpu(lmu->lum_hash_type);
3465         lsm->lsm_md_max_inherit = lmu->lum_max_inherit;
3466         lsm->lsm_md_max_inherit_rr = lmu->lum_max_inherit_rr;
3467         lsm->lsm_md_pool_name[LOV_MAXPOOLNAME] = 0;
3468
3469         return 0;
3470 }
3471
3472 static int lmv_unpackmd(struct obd_export *exp, struct lmv_stripe_md **lsmp,
3473                         const union lmv_mds_md *lmm, size_t lmm_size)
3474 {
3475         struct lmv_stripe_md     *lsm;
3476         int                      lsm_size;
3477         int                      rc;
3478         bool                     allocated = false;
3479         ENTRY;
3480
3481         LASSERT(lsmp != NULL);
3482
3483         lsm = *lsmp;
3484         /* Free memmd */
3485         if (lsm != NULL && lmm == NULL) {
3486                 int i;
3487                 struct lmv_foreign_md *lfm = (struct lmv_foreign_md *)lsm;
3488
3489                 if (lfm->lfm_magic == LMV_MAGIC_FOREIGN) {
3490                         size_t lfm_size;
3491
3492                         lfm_size = lfm->lfm_length + offsetof(typeof(*lfm),
3493                                                               lfm_value[0]);
3494                         OBD_FREE_LARGE(lfm, lfm_size);
3495                         RETURN(0);
3496                 }
3497
3498                 if (lmv_dir_striped(lsm)) {
3499                         for (i = 0; i < lsm->lsm_md_stripe_count; i++)
3500                                 iput(lsm->lsm_md_oinfo[i].lmo_root);
3501                         lsm_size = lmv_stripe_md_size(lsm->lsm_md_stripe_count);
3502                 } else {
3503                         lsm_size = lmv_stripe_md_size(0);
3504                 }
3505                 OBD_FREE(lsm, lsm_size);
3506                 *lsmp = NULL;
3507                 RETURN(0);
3508         }
3509
3510         /* foreign lmv case */
3511         if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_FOREIGN) {
3512                 struct lmv_foreign_md *lfm = (struct lmv_foreign_md *)lsm;
3513
3514                 if (lfm == NULL) {
3515                         OBD_ALLOC_LARGE(lfm, lmm_size);
3516                         if (lfm == NULL)
3517                                 RETURN(-ENOMEM);
3518                         *lsmp = (struct lmv_stripe_md *)lfm;
3519                 }
3520                 lfm->lfm_magic = le32_to_cpu(lmm->lmv_foreign_md.lfm_magic);
3521                 lfm->lfm_length = le32_to_cpu(lmm->lmv_foreign_md.lfm_length);
3522                 lfm->lfm_type = le32_to_cpu(lmm->lmv_foreign_md.lfm_type);
3523                 lfm->lfm_flags = le32_to_cpu(lmm->lmv_foreign_md.lfm_flags);
3524                 memcpy(&lfm->lfm_value, &lmm->lmv_foreign_md.lfm_value,
3525                        lfm->lfm_length);
3526                 RETURN(lmm_size);
3527         }
3528
3529         if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_STRIPE)
3530                 RETURN(-EPERM);
3531
3532         /* Unpack memmd */
3533         if (le32_to_cpu(lmm->lmv_magic) != LMV_MAGIC_V1 &&
3534             le32_to_cpu(lmm->lmv_magic) != LMV_USER_MAGIC) {
3535                 CERROR("%s: invalid lmv magic %x: rc = %d\n",
3536                        exp->exp_obd->obd_name, le32_to_cpu(lmm->lmv_magic),
3537                        -EIO);
3538                 RETURN(-EIO);
3539         }
3540
3541         if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_V1)
3542                 lsm_size = lmv_stripe_md_size(lmv_mds_md_stripe_count_get(lmm));
3543         else
3544                 /**
3545                  * Unpack default dirstripe(lmv_user_md) to lmv_stripe_md,
3546                  * stripecount should be 0 then.
3547                  */
3548                 lsm_size = lmv_stripe_md_size(0);
3549
3550         if (lsm == NULL) {
3551                 OBD_ALLOC(lsm, lsm_size);
3552                 if (lsm == NULL)
3553                         RETURN(-ENOMEM);
3554                 allocated = true;
3555                 *lsmp = lsm;
3556         }
3557
3558         switch (le32_to_cpu(lmm->lmv_magic)) {
3559         case LMV_MAGIC_V1:
3560                 rc = lmv_unpack_md_v1(exp, lsm, &lmm->lmv_md_v1);
3561                 break;
3562         case LMV_USER_MAGIC:
3563                 rc = lmv_unpack_user_md(exp, lsm, &lmm->lmv_user_md);
3564                 break;
3565         default:
3566                 CERROR("%s: unrecognized magic %x\n", exp->exp_obd->obd_name,
3567                        le32_to_cpu(lmm->lmv_magic));
3568                 rc = -EINVAL;
3569                 break;
3570         }
3571
3572         if (rc != 0 && allocated) {
3573                 OBD_FREE(lsm, lsm_size);
3574                 *lsmp = NULL;
3575                 lsm_size = rc;
3576         }
3577         RETURN(lsm_size);
3578 }
3579
3580 void lmv_free_memmd(struct lmv_stripe_md *lsm)
3581 {
3582         lmv_unpackmd(NULL, &lsm, NULL, 0);
3583 }
3584 EXPORT_SYMBOL(lmv_free_memmd);
3585
3586 static int lmv_cancel_unused(struct obd_export *exp, const struct lu_fid *fid,
3587                              union ldlm_policy_data *policy,
3588                              enum ldlm_mode mode, enum ldlm_cancel_flags flags,
3589                              void *opaque)
3590 {
3591         struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
3592         struct lu_tgt_desc *tgt;
3593         int err;
3594         int rc = 0;
3595
3596         ENTRY;
3597
3598         LASSERT(fid != NULL);
3599
3600         lmv_foreach_connected_tgt(lmv, tgt) {
3601                 if (!tgt->ltd_active)
3602                         continue;
3603
3604                 err = md_cancel_unused(tgt->ltd_exp, fid, policy, mode, flags,
3605                                        opaque);
3606                 if (!rc)
3607                         rc = err;
3608         }
3609         RETURN(rc);
3610 }
3611
3612 static int lmv_set_lock_data(struct obd_export *exp,
3613                              const struct lustre_handle *lockh,
3614                              void *data, __u64 *bits)
3615 {
3616         struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
3617         struct lmv_tgt_desc *tgt = lmv_tgt(lmv, 0);
3618         int rc;
3619
3620         ENTRY;
3621
3622         if (tgt == NULL || tgt->ltd_exp == NULL)
3623                 RETURN(-EINVAL);
3624         rc =  md_set_lock_data(tgt->ltd_exp, lockh, data, bits);
3625         RETURN(rc);
3626 }
3627
3628 static enum ldlm_mode
3629 lmv_lock_match(struct obd_export *exp, __u64 flags,
3630                const struct lu_fid *fid, enum ldlm_type type,
3631                union ldlm_policy_data *policy,
3632                enum ldlm_mode mode, struct lustre_handle *lockh)
3633 {
3634         struct obd_device *obd = exp->exp_obd;
3635         struct lmv_obd *lmv = &obd->u.lmv;
3636         struct lu_tgt_desc *tgt;
3637         __u64 bits = policy->l_inodebits.bits;
3638         enum ldlm_mode rc = LCK_MINMODE;
3639         int index;
3640         int i;
3641
3642         /* only one bit is set */
3643         LASSERT(bits && !(bits & (bits - 1)));
3644         /* With DNE every object can have two locks in different namespaces:
3645          * lookup lock in space of MDT storing direntry and update/open lock in
3646          * space of MDT storing inode.  Try the MDT that the FID maps to first,
3647          * since this can be easily found, and only try others if that fails.
3648          */
3649         if (bits == MDS_INODELOCK_LOOKUP) {
3650                 for (i = 0, index = lmv_fid2tgt_index(lmv, fid);
3651                      i < lmv->lmv_mdt_descs.ltd_tgts_size; i++,
3652                      index = (index + 1) % lmv->lmv_mdt_descs.ltd_tgts_size) {
3653                         if (index < 0) {
3654                                 CDEBUG(D_HA,
3655                                        "%s: "DFID" is inaccessible: rc = %d\n",
3656                                        obd->obd_name, PFID(fid), index);
3657                                 index = 0;
3658                         }
3659                         tgt = lmv_tgt(lmv, index);
3660                         if (!tgt || !tgt->ltd_exp || !tgt->ltd_active)
3661                                 continue;
3662                         rc = md_lock_match(tgt->ltd_exp, flags, fid, type,
3663                                            policy, mode, lockh);
3664                         if (rc)
3665                                 break;
3666                 }
3667         } else {
3668                 tgt = lmv_fid2tgt(lmv, fid);
3669                 if (!IS_ERR(tgt) && tgt->ltd_exp && tgt->ltd_active)
3670                         rc = md_lock_match(tgt->ltd_exp, flags, fid, type,
3671                                            policy, mode, lockh);
3672         }
3673
3674         CDEBUG(D_INODE, "Lock match for "DFID": %d\n", PFID(fid), rc);
3675
3676         return rc;
3677 }
3678
3679 static int
3680 lmv_get_lustre_md(struct obd_export *exp, struct req_capsule *pill,
3681                   struct obd_export *dt_exp, struct obd_export *md_exp,
3682                   struct lustre_md *md)
3683 {
3684         struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
3685         struct lmv_tgt_desc *tgt = lmv_tgt(lmv, 0);
3686
3687         if (!tgt || !tgt->ltd_exp)
3688                 return -EINVAL;
3689
3690         return md_get_lustre_md(tgt->ltd_exp, pill, dt_exp, md_exp, md);
3691 }
3692
3693 static int lmv_free_lustre_md(struct obd_export *exp, struct lustre_md *md)
3694 {
3695         struct obd_device *obd = exp->exp_obd;
3696         struct lmv_obd *lmv = &obd->u.lmv;
3697         struct lmv_tgt_desc *tgt = lmv_tgt(lmv, 0);
3698
3699         ENTRY;
3700
3701         if (md->default_lmv) {
3702                 lmv_free_memmd(md->default_lmv);
3703                 md->default_lmv = NULL;
3704         }
3705         if (md->lmv != NULL) {
3706                 lmv_free_memmd(md->lmv);
3707                 md->lmv = NULL;
3708         }
3709         if (!tgt || !tgt->ltd_exp)
3710                 RETURN(-EINVAL);
3711         RETURN(md_free_lustre_md(tgt->ltd_exp, md));
3712 }
3713
3714 static int lmv_set_open_replay_data(struct obd_export *exp,
3715                                     struct obd_client_handle *och,
3716                                     struct lookup_intent *it)
3717 {
3718         struct obd_device *obd = exp->exp_obd;
3719         struct lmv_obd *lmv = &obd->u.lmv;
3720         struct lmv_tgt_desc *tgt;
3721
3722         ENTRY;
3723
3724         tgt = lmv_fid2tgt(lmv, &och->och_fid);
3725         if (IS_ERR(tgt))
3726                 RETURN(PTR_ERR(tgt));
3727
3728         RETURN(md_set_open_replay_data(tgt->ltd_exp, och, it));
3729 }
3730
3731 static int lmv_clear_open_replay_data(struct obd_export *exp,
3732                                       struct obd_client_handle *och)
3733 {
3734         struct obd_device *obd = exp->exp_obd;
3735         struct lmv_obd *lmv = &obd->u.lmv;
3736         struct lmv_tgt_desc *tgt;
3737
3738         ENTRY;
3739
3740         tgt = lmv_fid2tgt(lmv, &och->och_fid);
3741         if (IS_ERR(tgt))
3742                 RETURN(PTR_ERR(tgt));
3743
3744         RETURN(md_clear_open_replay_data(tgt->ltd_exp, och));
3745 }
3746
3747 static int lmv_intent_getattr_async(struct obd_export *exp,
3748                                     struct md_op_item *item)
3749 {
3750         struct md_op_data *op_data = &item->mop_data;
3751         struct obd_device *obd = exp->exp_obd;
3752         struct lmv_obd *lmv = &obd->u.lmv;
3753         struct lmv_tgt_desc *ptgt;
3754         struct lmv_tgt_desc *ctgt;
3755         int rc;
3756
3757         ENTRY;
3758
3759         if (!fid_is_sane(&op_data->op_fid2))
3760                 RETURN(-EINVAL);
3761
3762         ptgt = lmv_locate_tgt(lmv, op_data);
3763         if (IS_ERR(ptgt))
3764                 RETURN(PTR_ERR(ptgt));
3765
3766         ctgt = lmv_fid2tgt(lmv, &op_data->op_fid2);
3767         if (IS_ERR(ctgt))
3768                 RETURN(PTR_ERR(ctgt));
3769
3770         /*
3771          * remote object needs two RPCs to lookup and getattr, considering the
3772          * complexity don't support statahead for now.
3773          */
3774         if (ctgt != ptgt)
3775                 RETURN(-EREMOTE);
3776
3777         rc = md_intent_getattr_async(ptgt->ltd_exp, item);
3778
3779         RETURN(rc);
3780 }
3781
3782 static int lmv_revalidate_lock(struct obd_export *exp, struct lookup_intent *it,
3783                                struct lu_fid *fid, __u64 *bits)
3784 {
3785         struct obd_device *obd = exp->exp_obd;
3786         struct lmv_obd *lmv = &obd->u.lmv;
3787         struct lmv_tgt_desc *tgt;
3788         int rc;
3789
3790         ENTRY;
3791
3792         tgt = lmv_fid2tgt(lmv, fid);
3793         if (IS_ERR(tgt))
3794                 RETURN(PTR_ERR(tgt));
3795
3796         rc = md_revalidate_lock(tgt->ltd_exp, it, fid, bits);
3797         RETURN(rc);
3798 }
3799
3800 static int lmv_get_fid_from_lsm(struct obd_export *exp,
3801                                 const struct lmv_stripe_md *lsm,
3802                                 const char *name, int namelen,
3803                                 struct lu_fid *fid)
3804 {
3805         const struct lmv_oinfo *oinfo;
3806
3807         if (!lmv_dir_striped(lsm))
3808                 RETURN(-ESTALE);
3809
3810         oinfo = lsm_name_to_stripe_info(lsm, name, namelen, false);
3811         if (IS_ERR(oinfo))
3812                 return PTR_ERR(oinfo);
3813
3814         *fid = oinfo->lmo_fid;
3815
3816         RETURN(0);
3817 }
3818
3819 /**
3820  * For lmv, only need to send request to master MDT, and the master MDT will
3821  * process with other slave MDTs. The only exception is Q_GETOQUOTA for which
3822  * we directly fetch data from the slave MDTs.
3823  */
3824 static int lmv_quotactl(struct obd_device *unused, struct obd_export *exp,
3825                         struct obd_quotactl *oqctl)
3826 {
3827         struct obd_device *obd = class_exp2obd(exp);
3828         struct lmv_obd *lmv = &obd->u.lmv;
3829         struct lmv_tgt_desc *tgt = lmv_tgt(lmv, 0);
3830         __u64 curspace, curinodes;
3831         int rc = 0;
3832
3833         ENTRY;
3834
3835         if (!tgt || !tgt->ltd_exp || !tgt->ltd_active) {
3836                 CERROR("master lmv inactive\n");
3837                 RETURN(-EIO);
3838         }
3839
3840         if (oqctl->qc_cmd != Q_GETOQUOTA) {
3841                 rc = obd_quotactl(tgt->ltd_exp, oqctl);
3842                 RETURN(rc);
3843         }
3844
3845         curspace = curinodes = 0;
3846         lmv_foreach_connected_tgt(lmv, tgt) {
3847                 int err;
3848
3849                 if (!tgt->ltd_active)
3850                         continue;
3851
3852                 err = obd_quotactl(tgt->ltd_exp, oqctl);
3853                 if (err) {
3854                         CERROR("getquota on mdt %d failed. %d\n",
3855                                tgt->ltd_index, err);
3856                         if (!rc)
3857                                 rc = err;
3858                 } else {
3859                         curspace += oqctl->qc_dqblk.dqb_curspace;
3860                         curinodes += oqctl->qc_dqblk.dqb_curinodes;
3861                 }
3862         }
3863         oqctl->qc_dqblk.dqb_curspace = curspace;
3864         oqctl->qc_dqblk.dqb_curinodes = curinodes;
3865
3866         RETURN(rc);
3867 }
3868
3869 static int lmv_merge_attr(struct obd_export *exp,
3870                           const struct lmv_stripe_md *lsm,
3871                           struct cl_attr *attr,
3872                           ldlm_blocking_callback cb_blocking)
3873 {
3874         int rc;
3875         int i;
3876
3877         if (!lmv_dir_striped(lsm))
3878                 return 0;
3879
3880         rc = lmv_revalidate_slaves(exp, lsm, cb_blocking, 0);
3881         if (rc < 0)
3882                 return rc;
3883
3884         for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
3885                 struct inode *inode = lsm->lsm_md_oinfo[i].lmo_root;
3886
3887                 if (!inode)
3888                         continue;
3889
3890                 CDEBUG(D_INFO,
3891                        "" DFID " size %llu, blocks %llu nlink %u, atime %lld ctime %lld, mtime %lld.\n",
3892                        PFID(&lsm->lsm_md_oinfo[i].lmo_fid),
3893                        i_size_read(inode), (unsigned long long)inode->i_blocks,
3894                        inode->i_nlink, (s64)inode->i_atime.tv_sec,
3895                        (s64)inode->i_ctime.tv_sec, (s64)inode->i_mtime.tv_sec);
3896
3897                 /* for slave stripe, it needs to subtract nlink for . and .. */
3898                 if (i != 0)
3899                         attr->cat_nlink += inode->i_nlink - 2;
3900                 else
3901                         attr->cat_nlink = inode->i_nlink;
3902
3903                 attr->cat_size += i_size_read(inode);
3904                 attr->cat_blocks += inode->i_blocks;
3905
3906                 if (attr->cat_atime < inode->i_atime.tv_sec)
3907                         attr->cat_atime = inode->i_atime.tv_sec;
3908
3909                 if (attr->cat_ctime < inode->i_ctime.tv_sec)
3910                         attr->cat_ctime = inode->i_ctime.tv_sec;
3911
3912                 if (attr->cat_mtime < inode->i_mtime.tv_sec)
3913                         attr->cat_mtime = inode->i_mtime.tv_sec;
3914         }
3915         return 0;
3916 }
3917
3918 static struct lu_batch *lmv_batch_create(struct obd_export *exp,
3919                                          enum lu_batch_flags flags,
3920                                          __u32 max_count)
3921 {
3922         struct lu_batch *bh;
3923         struct lmv_batch *lbh;
3924
3925         ENTRY;
3926         OBD_ALLOC_PTR(lbh);
3927         if (!lbh)
3928                 RETURN(ERR_PTR(-ENOMEM));
3929
3930         bh = &lbh->lbh_super;
3931         bh->lbt_flags = flags;
3932         bh->lbt_max_count = max_count;
3933
3934         if (flags & BATCH_FL_RQSET) {
3935                 bh->lbt_rqset = ptlrpc_prep_set();
3936                 if (bh->lbt_rqset == NULL) {
3937                         OBD_FREE_PTR(lbh);
3938                         RETURN(ERR_PTR(-ENOMEM));
3939                 }
3940         }
3941
3942         INIT_LIST_HEAD(&lbh->lbh_sub_batch_list);
3943         RETURN(bh);
3944 }
3945
3946 static int lmv_batch_stop(struct obd_export *exp, struct lu_batch *bh)
3947 {
3948         struct lmv_batch *lbh;
3949         struct lmvsub_batch *sub;
3950         struct lmvsub_batch *tmp;
3951         int rc = 0;
3952
3953         ENTRY;
3954
3955         lbh = container_of(bh, struct lmv_batch, lbh_super);
3956         list_for_each_entry_safe(sub, tmp, &lbh->lbh_sub_batch_list,
3957                                  sbh_sub_item) {
3958                 list_del(&sub->sbh_sub_item);
3959                 rc = md_batch_stop(sub->sbh_tgt->ltd_exp, sub->sbh_sub);
3960                 if (rc < 0) {
3961                         CERROR("%s: stop batch processing failed: rc = %d\n",
3962                                exp->exp_obd->obd_name, rc);
3963                         if (bh->lbt_result == 0)
3964                                 bh->lbt_result = rc;
3965                 }
3966                 OBD_FREE_PTR(sub);
3967         }
3968
3969         if (bh->lbt_flags & BATCH_FL_RQSET) {
3970                 rc = ptlrpc_set_wait(NULL, bh->lbt_rqset);
3971                 ptlrpc_set_destroy(bh->lbt_rqset);
3972         }
3973
3974         OBD_FREE_PTR(lbh);
3975         RETURN(rc);
3976 }
3977
3978 static int lmv_batch_flush(struct obd_export *exp, struct lu_batch *bh,
3979                            bool wait)
3980 {
3981         struct lmv_batch *lbh;
3982         struct lmvsub_batch *sub;
3983         int rc = 0;
3984         int rc1;
3985
3986         ENTRY;
3987
3988         lbh = container_of(bh, struct lmv_batch, lbh_super);
3989         list_for_each_entry(sub, &lbh->lbh_sub_batch_list, sbh_sub_item) {
3990                 rc1 = md_batch_flush(sub->sbh_tgt->ltd_exp, sub->sbh_sub, wait);
3991                 if (rc1 < 0) {
3992                         CERROR("%s: stop batch processing failed: rc = %d\n",
3993                                exp->exp_obd->obd_name, rc);
3994                         if (bh->lbt_result == 0)
3995                                 bh->lbt_result = rc;
3996
3997                         if (rc == 0)
3998                                 rc = rc1;
3999                 }
4000         }
4001
4002         if (wait && bh->lbt_flags & BATCH_FL_RQSET) {
4003                 rc1 = ptlrpc_set_wait(NULL, bh->lbt_rqset);
4004                 if (rc == 0)
4005                         rc = rc1;
4006         }
4007
4008         RETURN(rc);
4009 }
4010
4011 static inline struct lmv_tgt_desc *
4012 lmv_batch_locate_tgt(struct lmv_obd *lmv, struct md_op_item *item)
4013 {
4014         struct md_op_data *op_data = &item->mop_data;
4015         struct lmv_tgt_desc *tgt;
4016
4017         switch (item->mop_opc) {
4018         case MD_OP_GETATTR: {
4019                 struct lmv_tgt_desc *ptgt;
4020
4021                 if (!fid_is_sane(&op_data->op_fid2))
4022                         RETURN(ERR_PTR(-EINVAL));
4023
4024                 ptgt = lmv_locate_tgt(lmv, op_data);
4025                 if (IS_ERR(ptgt))
4026                         RETURN(ptgt);
4027
4028                 tgt = lmv_fid2tgt(lmv, &op_data->op_fid2);
4029                 if (IS_ERR(tgt))
4030                         RETURN(tgt);
4031
4032                 /*
4033                  * Remote object needs two RPCs to lookup and getattr,
4034                  * considering the complexity don't support statahead for now.
4035                  */
4036                 if (tgt != ptgt)
4037                         RETURN(ERR_PTR(-EREMOTE));
4038
4039                 break;
4040         }
4041         default:
4042                 tgt = ERR_PTR(-ENOTSUPP);
4043         }
4044
4045         return tgt;
4046 }
4047
4048 struct lu_batch *lmv_batch_lookup_sub(struct lmv_batch *lbh,
4049                                       struct lmv_tgt_desc *tgt)
4050 {
4051         struct lmvsub_batch *sub;
4052
4053         list_for_each_entry(sub, &lbh->lbh_sub_batch_list, sbh_sub_item) {
4054                 if (sub->sbh_tgt == tgt)
4055                         return sub->sbh_sub;
4056         }
4057
4058         return NULL;
4059 }
4060
4061 struct lu_batch *lmv_batch_get_sub(struct lmv_batch *lbh,
4062                                    struct lmv_tgt_desc *tgt)
4063 {
4064         struct lmvsub_batch *sbh;
4065         struct lu_batch *child_bh;
4066         struct lu_batch *bh;
4067
4068         ENTRY;
4069
4070         child_bh = lmv_batch_lookup_sub(lbh, tgt);
4071         if (child_bh != NULL)
4072                 RETURN(child_bh);
4073
4074         OBD_ALLOC_PTR(sbh);
4075         if (sbh == NULL)
4076                 RETURN(ERR_PTR(-ENOMEM));
4077
4078         INIT_LIST_HEAD(&sbh->sbh_sub_item);
4079         sbh->sbh_tgt = tgt;
4080
4081         bh = &lbh->lbh_super;
4082         child_bh = md_batch_create(tgt->ltd_exp, bh->lbt_flags,
4083                                    bh->lbt_max_count);
4084         if (IS_ERR(child_bh)) {
4085                 OBD_FREE_PTR(sbh);
4086                 RETURN(child_bh);
4087         }
4088
4089         child_bh->lbt_rqset = bh->lbt_rqset;
4090         sbh->sbh_sub = child_bh;
4091         list_add(&sbh->sbh_sub_item, &lbh->lbh_sub_batch_list);
4092         RETURN(child_bh);
4093 }
4094
4095 static int lmv_batch_add(struct obd_export *exp, struct lu_batch *bh,
4096                          struct md_op_item *item)
4097 {
4098         struct obd_device *obd = exp->exp_obd;
4099         struct lmv_obd *lmv = &obd->u.lmv;
4100         struct lmv_tgt_desc *tgt;
4101         struct lmv_batch *lbh;
4102         struct lu_batch *child_bh;
4103         int rc;
4104
4105         ENTRY;
4106
4107         tgt = lmv_batch_locate_tgt(lmv, item);
4108         if (IS_ERR(tgt))
4109                 RETURN(PTR_ERR(tgt));
4110
4111         lbh = container_of(bh, struct lmv_batch, lbh_super);
4112         child_bh = lmv_batch_get_sub(lbh, tgt);
4113         if (IS_ERR(child_bh))
4114                 RETURN(PTR_ERR(child_bh));
4115
4116         rc = md_batch_add(tgt->ltd_exp, child_bh, item);
4117         RETURN(rc);
4118 }
4119
4120 static const struct obd_ops lmv_obd_ops = {
4121         .o_owner                = THIS_MODULE,
4122         .o_setup                = lmv_setup,
4123         .o_cleanup              = lmv_cleanup,
4124         .o_precleanup           = lmv_precleanup,
4125         .o_process_config       = lmv_process_config,
4126         .o_connect              = lmv_connect,
4127         .o_disconnect           = lmv_disconnect,
4128         .o_statfs               = lmv_statfs,
4129         .o_get_info             = lmv_get_info,
4130         .o_set_info_async       = lmv_set_info_async,
4131         .o_notify               = lmv_notify,
4132         .o_get_uuid             = lmv_get_uuid,
4133         .o_fid_alloc            = lmv_fid_alloc,
4134         .o_iocontrol            = lmv_iocontrol,
4135         .o_quotactl             = lmv_quotactl
4136 };
4137
4138 static const struct md_ops lmv_md_ops = {
4139         .m_get_root             = lmv_get_root,
4140         .m_null_inode           = lmv_null_inode,
4141         .m_close                = lmv_close,
4142         .m_create               = lmv_create,
4143         .m_enqueue              = lmv_enqueue,
4144         .m_getattr              = lmv_getattr,
4145         .m_getxattr             = lmv_getxattr,
4146         .m_getattr_name         = lmv_getattr_name,
4147         .m_intent_lock          = lmv_intent_lock,
4148         .m_link                 = lmv_link,
4149         .m_rename               = lmv_rename,
4150         .m_setattr              = lmv_setattr,
4151         .m_setxattr             = lmv_setxattr,
4152         .m_fsync                = lmv_fsync,
4153         .m_file_resync          = lmv_file_resync,
4154         .m_read_page            = lmv_read_page,
4155         .m_unlink               = lmv_unlink,
4156         .m_init_ea_size         = lmv_init_ea_size,
4157         .m_cancel_unused        = lmv_cancel_unused,
4158         .m_set_lock_data        = lmv_set_lock_data,
4159         .m_lock_match           = lmv_lock_match,
4160         .m_get_lustre_md        = lmv_get_lustre_md,
4161         .m_free_lustre_md       = lmv_free_lustre_md,
4162         .m_merge_attr           = lmv_merge_attr,
4163         .m_set_open_replay_data = lmv_set_open_replay_data,
4164         .m_clear_open_replay_data = lmv_clear_open_replay_data,
4165         .m_intent_getattr_async = lmv_intent_getattr_async,
4166         .m_revalidate_lock      = lmv_revalidate_lock,
4167         .m_get_fid_from_lsm     = lmv_get_fid_from_lsm,
4168         .m_unpackmd             = lmv_unpackmd,
4169         .m_rmfid                = lmv_rmfid,
4170         .m_batch_create         = lmv_batch_create,
4171         .m_batch_add            = lmv_batch_add,
4172         .m_batch_stop           = lmv_batch_stop,
4173         .m_batch_flush          = lmv_batch_flush,
4174 };
4175
4176 static int __init lmv_init(void)
4177 {
4178         return class_register_type(&lmv_obd_ops, &lmv_md_ops, true,
4179                                    LUSTRE_LMV_NAME, NULL);
4180 }
4181
4182 static void __exit lmv_exit(void)
4183 {
4184         class_unregister_type(LUSTRE_LMV_NAME);
4185 }
4186
4187 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
4188 MODULE_DESCRIPTION("Lustre Logical Metadata Volume");
4189 MODULE_VERSION(LUSTRE_VERSION_STRING);
4190 MODULE_LICENSE("GPL");
4191
4192 module_init(lmv_init);
4193 module_exit(lmv_exit);