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