lustre/lmv/lmv_obd.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.gnu.org/licenses/gpl-2.0.html
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2004, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Use is subject to license terms.
  25  *
  26  * Copyright (c) 2011, 2017, Intel Corporation.
  27  */
  28 /*
  29  * This file is part of Lustre, http://www.lustre.org/
  30  * Lustre is a trademark of Sun Microsystems, Inc.
  31  */
  32
  33 #define DEBUG_SUBSYSTEM S_LMV
  34
  35 #include <linux/file.h>
  36 #include <linux/module.h>
  37 #include <linux/init.h>
  38 #include <linux/user_namespace.h>
  39 #include <linux/uidgid.h>
  40 #include <linux/slab.h>
  41 #include <linux/pagemap.h>
  42 #include <linux/mm.h>
  43 #include <linux/math64.h>
  44 #include <linux/seq_file.h>
  45 #include <linux/namei.h>
  46
  47 #include <obd_support.h>
  48 #include <lustre_lib.h>
  49 #include <lustre_net.h>
  50 #include <obd_class.h>
  51 #include <lustre_lmv.h>
  52 #include <lprocfs_status.h>
  53 #include <cl_object.h>
  54 #include <lustre_fid.h>
  55 #include <uapi/linux/lustre/lustre_ioctl.h>
  56 #include <lustre_kernelcomm.h>
  57 #include "lmv_internal.h"
  58
  59 static int lmv_check_connect(struct obd_device *obd);
  60
  61 void lmv_activate_target(struct lmv_obd *lmv, struct lmv_tgt_desc *tgt,
  62                          int activate)
  63 {
  64         if (tgt->ltd_active == activate)
  65                 return;
  66
  67         tgt->ltd_active = activate;
  68         lmv->lmv_mdt_descs.ltd_lmv_desc.ld_active_tgt_count +=
  69                 (activate ? 1 : -1);
  70
  71         tgt->ltd_exp->exp_obd->obd_inactive = !activate;
  72 }
  73
  74 /**
  75  * Error codes:
  76  *
  77  *  -EINVAL  : UUID can't be found in the LMV's target list
  78  *  -ENOTCONN: The UUID is found, but the target connection is bad (!)
  79  *  -EBADF   : The UUID is found, but the OBD of the wrong type (!)
  80  */
  81 static int lmv_set_mdc_active(struct lmv_obd *lmv,
  82                               const struct obd_uuid *uuid,
  83                               int activate)
  84 {
  85         struct lu_tgt_desc *tgt = NULL;
  86         struct obd_device *obd;
  87         int rc = 0;
  88
  89         ENTRY;
  90
  91         CDEBUG(D_INFO, "Searching in lmv %p for uuid %s (activate=%d)\n",
  92                         lmv, uuid->uuid, activate);
  93
  94         spin_lock(&lmv->lmv_lock);
  95         lmv_foreach_connected_tgt(lmv, tgt) {
  96                 CDEBUG(D_INFO, "Target idx %d is %s conn %#llx\n",
  97                        tgt->ltd_index, tgt->ltd_uuid.uuid,
  98                        tgt->ltd_exp->exp_handle.h_cookie);
  99
 100                 if (obd_uuid_equals(uuid, &tgt->ltd_uuid))
 101                         break;
 102         }
 103
 104         if (!tgt)
 105                 GOTO(out_lmv_lock, rc = -EINVAL);
 106
 107         obd = class_exp2obd(tgt->ltd_exp);
 108         if (obd == NULL)
 109                 GOTO(out_lmv_lock, rc = -ENOTCONN);
 110
 111         CDEBUG(D_INFO, "Found OBD %s=%s device %d (%p) type %s at LMV idx %d\n",
 112                obd->obd_name, obd->obd_uuid.uuid, obd->obd_minor, obd,
 113                obd->obd_type->typ_name, tgt->ltd_index);
 114         LASSERT(strcmp(obd->obd_type->typ_name, LUSTRE_MDC_NAME) == 0);
 115
 116         if (tgt->ltd_active == activate) {
 117                 CDEBUG(D_INFO, "OBD %p already %sactive!\n", obd,
 118                        activate ? "" : "in");
 119                 GOTO(out_lmv_lock, rc);
 120         }
 121
 122         CDEBUG(D_INFO, "Marking OBD %p %sactive\n", obd,
 123                activate ? "" : "in");
 124         lmv_activate_target(lmv, tgt, activate);
 125         EXIT;
 126
 127  out_lmv_lock:
 128         spin_unlock(&lmv->lmv_lock);
 129         return rc;
 130 }
 131
 132 struct obd_uuid *lmv_get_uuid(struct obd_export *exp)
 133 {
 134         struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
 135         struct lmv_tgt_desc *tgt = lmv_tgt(lmv, 0);
 136
 137         return (tgt == NULL) ? NULL : obd_get_uuid(tgt->ltd_exp);
 138 }
 139
 140 static int lmv_notify(struct obd_device *obd, struct obd_device *watched,
 141                       enum obd_notify_event ev)
 142 {
 143         struct obd_connect_data *conn_data;
 144         struct lmv_obd          *lmv = &obd->u.lmv;
 145         struct obd_uuid         *uuid;
 146         int                      rc = 0;
 147         ENTRY;
 148
 149         if (strcmp(watched->obd_type->typ_name, LUSTRE_MDC_NAME)) {
 150                 CERROR("unexpected notification of %s %s!\n",
 151                        watched->obd_type->typ_name,
 152                        watched->obd_name);
 153                 RETURN(-EINVAL);
 154         }
 155
 156         uuid = &watched->u.cli.cl_target_uuid;
 157         if (ev == OBD_NOTIFY_ACTIVE || ev == OBD_NOTIFY_INACTIVE) {
 158                 /*
 159                  * Set MDC as active before notifying the observer, so the
 160                  * observer can use the MDC normally.
 161                  */
 162                 rc = lmv_set_mdc_active(lmv, uuid,
 163                                         ev == OBD_NOTIFY_ACTIVE);
 164                 if (rc) {
 165                         CERROR("%sactivation of %s failed: %d\n",
 166                                ev == OBD_NOTIFY_ACTIVE ? "" : "de",
 167                                uuid->uuid, rc);
 168                         RETURN(rc);
 169                 }
 170         } else if (ev == OBD_NOTIFY_OCD) {
 171                 conn_data = &watched->u.cli.cl_import->imp_connect_data;
 172                 /*
 173                  * XXX: Make sure that ocd_connect_flags from all targets are
 174                  * the same. Otherwise one of MDTs runs wrong version or
 175                  * something like this.  --umka
 176                  */
 177                 obd->obd_self_export->exp_connect_data = *conn_data;
 178         }
 179
 180         /*
 181          * Pass the notification up the chain.
 182          */
 183         if (obd->obd_observer)
 184                 rc = obd_notify(obd->obd_observer, watched, ev);
 185
 186         RETURN(rc);
 187 }
 188
 189 static int lmv_connect(const struct lu_env *env,
 190                        struct obd_export **pexp, struct obd_device *obd,
 191                        struct obd_uuid *cluuid, struct obd_connect_data *data,
 192                        void *localdata)
 193 {
 194         struct lmv_obd *lmv = &obd->u.lmv;
 195         struct lustre_handle conn = { 0 };
 196         struct obd_export *exp;
 197         int rc;
 198         ENTRY;
 199
 200         rc = class_connect(&conn, obd, cluuid);
 201         if (rc) {
 202                 CERROR("class_connection() returned %d\n", rc);
 203                 RETURN(rc);
 204         }
 205
 206         exp = class_conn2export(&conn);
 207
 208         lmv->connected = 0;
 209         lmv->conn_data = *data;
 210         lmv->lmv_cache = localdata;
 211
 212         lmv->lmv_tgts_kobj = kobject_create_and_add("target_obds",
 213                                                     &obd->obd_kset.kobj);
 214         if (!lmv->lmv_tgts_kobj) {
 215                 CERROR("%s: cannot create /sys/fs/lustre/%s/%s/target_obds\n",
 216                        obd->obd_name, obd->obd_type->typ_name, obd->obd_name);
 217         }
 218
 219         rc = lmv_check_connect(obd);
 220         if (rc != 0)
 221                 GOTO(out_sysfs, rc);
 222
 223         *pexp = exp;
 224
 225         RETURN(rc);
 226
 227 out_sysfs:
 228         if (lmv->lmv_tgts_kobj)
 229                 kobject_put(lmv->lmv_tgts_kobj);
 230
 231         class_disconnect(exp);
 232
 233         return rc;
 234 }
 235
 236 static int lmv_init_ea_size(struct obd_export *exp, __u32 easize,
 237                             __u32 def_easize)
 238 {
 239         struct obd_device *obd = exp->exp_obd;
 240         struct lmv_obd *lmv = &obd->u.lmv;
 241         struct lmv_tgt_desc *tgt;
 242         int change = 0;
 243         int rc = 0;
 244
 245         ENTRY;
 246
 247         if (lmv->max_easize < easize) {
 248                 lmv->max_easize = easize;
 249                 change = 1;
 250         }
 251         if (lmv->max_def_easize < def_easize) {
 252                 lmv->max_def_easize = def_easize;
 253                 change = 1;
 254         }
 255
 256         if (change == 0)
 257                 RETURN(0);
 258
 259         if (lmv->connected == 0)
 260                 RETURN(0);
 261
 262         lmv_foreach_connected_tgt(lmv, tgt) {
 263                 if (!tgt->ltd_active)
 264                         continue;
 265
 266                 rc = md_init_ea_size(tgt->ltd_exp, easize, def_easize);
 267                 if (rc) {
 268                         CERROR("%s: obd_init_ea_size() failed on MDT target %d:"
 269                                " rc = %d\n", obd->obd_name, tgt->ltd_index, rc);
 270                         break;
 271                 }
 272         }
 273         RETURN(rc);
 274 }
 275
 276 #define MAX_STRING_SIZE 128
 277
 278 int lmv_connect_mdc(struct obd_device *obd, struct lmv_tgt_desc *tgt)
 279 {
 280         struct lmv_obd          *lmv = &obd->u.lmv;
 281         struct obd_device       *mdc_obd;
 282         struct obd_export       *mdc_exp;
 283         struct lu_fld_target     target;
 284         int                      rc;
 285         ENTRY;
 286
 287         mdc_obd = class_find_client_obd(&tgt->ltd_uuid, LUSTRE_MDC_NAME,
 288                                         &obd->obd_uuid);
 289         if (!mdc_obd) {
 290                 CERROR("target %s not attached\n", tgt->ltd_uuid.uuid);
 291                 RETURN(-EINVAL);
 292         }
 293
 294         CDEBUG(D_CONFIG, "connect to %s(%s) - %s, %s\n",
 295                mdc_obd->obd_name, mdc_obd->obd_uuid.uuid,
 296                tgt->ltd_uuid.uuid, obd->obd_uuid.uuid);
 297
 298         if (!mdc_obd->obd_set_up) {
 299                 CERROR("target %s is not set up\n", tgt->ltd_uuid.uuid);
 300                 RETURN(-EINVAL);
 301         }
 302
 303         rc = obd_connect(NULL, &mdc_exp, mdc_obd, &obd->obd_uuid,
 304                          &lmv->conn_data, lmv->lmv_cache);
 305         if (rc) {
 306                 CERROR("target %s connect error %d\n", tgt->ltd_uuid.uuid, rc);
 307                 RETURN(rc);
 308         }
 309
 310         /*
 311          * Init fid sequence client for this mdc and add new fld target.
 312          */
 313         rc = obd_fid_init(mdc_obd, mdc_exp, LUSTRE_SEQ_METADATA);
 314         if (rc)
 315                 RETURN(rc);
 316
 317         target.ft_srv = NULL;
 318         target.ft_exp = mdc_exp;
 319         target.ft_idx = tgt->ltd_index;
 320
 321         fld_client_add_target(&lmv->lmv_fld, &target);
 322
 323         rc = obd_register_observer(mdc_obd, obd);
 324         if (rc) {
 325                 obd_disconnect(mdc_exp);
 326                 CERROR("target %s register_observer error %d\n",
 327                        tgt->ltd_uuid.uuid, rc);
 328                 RETURN(rc);
 329         }
 330
 331         if (obd->obd_observer) {
 332                 /*
 333                  * Tell the observer about the new target.
 334                  */
 335                 rc = obd_notify(obd->obd_observer, mdc_exp->exp_obd,
 336                                 OBD_NOTIFY_ACTIVE);
 337                 if (rc) {
 338                         obd_disconnect(mdc_exp);
 339                         RETURN(rc);
 340                 }
 341         }
 342
 343         tgt->ltd_active = 1;
 344         tgt->ltd_exp = mdc_exp;
 345         lmv->lmv_mdt_descs.ltd_lmv_desc.ld_active_tgt_count++;
 346
 347         md_init_ea_size(tgt->ltd_exp, lmv->max_easize, lmv->max_def_easize);
 348
 349         rc = lu_qos_add_tgt(&lmv->lmv_qos, tgt);
 350         if (rc) {
 351                 obd_disconnect(mdc_exp);
 352                 RETURN(rc);
 353         }
 354
 355         CDEBUG(D_CONFIG, "Connected to %s(%s) successfully (%d)\n",
 356                 mdc_obd->obd_name, mdc_obd->obd_uuid.uuid,
 357                 atomic_read(&obd->obd_refcount));
 358
 359         lmv_statfs_check_update(obd, tgt);
 360
 361         if (lmv->lmv_tgts_kobj)
 362                 /* Even if we failed to create the link, that's fine */
 363                 rc = sysfs_create_link(lmv->lmv_tgts_kobj,
 364                                        &mdc_obd->obd_kset.kobj,
 365                                        mdc_obd->obd_name);
 366         RETURN(0);
 367 }
 368
 369 static void lmv_del_target(struct lmv_obd *lmv, struct lu_tgt_desc *tgt)
 370 {
 371         LASSERT(tgt);
 372         ltd_del_tgt(&lmv->lmv_mdt_descs, tgt);
 373         OBD_FREE_PTR(tgt);
 374 }
 375
 376 static int lmv_add_target(struct obd_device *obd, struct obd_uuid *uuidp,
 377                            __u32 index, int gen)
 378 {
 379         struct obd_device *mdc_obd;
 380         struct lmv_obd *lmv = &obd->u.lmv;
 381         struct lmv_tgt_desc *tgt;
 382         struct lu_tgt_descs *ltd = &lmv->lmv_mdt_descs;
 383         int rc = 0;
 384
 385         ENTRY;
 386
 387         CDEBUG(D_CONFIG, "Target uuid: %s. index %d\n", uuidp->uuid, index);
 388         mdc_obd = class_find_client_obd(uuidp, LUSTRE_MDC_NAME,
 389                                         &obd->obd_uuid);
 390         if (!mdc_obd) {
 391                 CERROR("%s: Target %s not attached: rc = %d\n",
 392                        obd->obd_name, uuidp->uuid, -EINVAL);
 393                 RETURN(-EINVAL);
 394         }
 395
 396         OBD_ALLOC_PTR(tgt);
 397         if (!tgt)
 398                 RETURN(-ENOMEM);
 399
 400         mutex_init(&tgt->ltd_fid_mutex);
 401         tgt->ltd_index = index;
 402         tgt->ltd_uuid = *uuidp;
 403         tgt->ltd_active = 0;
 404
 405         mutex_lock(&ltd->ltd_mutex);
 406         rc = ltd_add_tgt(ltd, tgt);
 407         mutex_unlock(&ltd->ltd_mutex);
 408
 409         if (rc)
 410                 GOTO(out_tgt, rc);
 411
 412         if (!lmv->connected)
 413                 /* lmv_check_connect() will connect this target. */
 414                 RETURN(0);
 415
 416         rc = lmv_connect_mdc(obd, tgt);
 417         if (!rc) {
 418                 int easize = sizeof(struct lmv_stripe_md) +
 419                         lmv->lmv_mdt_count * sizeof(struct lu_fid);
 420
 421                 lmv_init_ea_size(obd->obd_self_export, easize, 0);
 422         }
 423
 424         RETURN(rc);
 425
 426 out_tgt:
 427         OBD_FREE_PTR(tgt);
 428         return rc;
 429 }
 430
 431 static int lmv_check_connect(struct obd_device *obd)
 432 {
 433         struct lmv_obd *lmv = &obd->u.lmv;
 434         struct lmv_tgt_desc *tgt;
 435         int easize;
 436         int rc;
 437
 438         ENTRY;
 439
 440         if (lmv->connected)
 441                 RETURN(0);
 442
 443         mutex_lock(&lmv->lmv_mdt_descs.ltd_mutex);
 444         if (lmv->connected)
 445                 GOTO(unlock, rc = 0);
 446
 447         if (!lmv->lmv_mdt_count) {
 448                 CERROR("%s: no targets configured: rc = -EINVAL\n",
 449                        obd->obd_name);
 450                 GOTO(unlock, rc = -EINVAL);
 451         }
 452
 453         if (!lmv_mdt0_inited(lmv)) {
 454                 CERROR("%s: no target configured for index 0: rc = -EINVAL.\n",
 455                        obd->obd_name);
 456                 GOTO(unlock, rc = -EINVAL);
 457         }
 458
 459         CDEBUG(D_CONFIG, "Time to connect %s to %s\n",
 460                obd->obd_uuid.uuid, obd->obd_name);
 461
 462         lmv_foreach_tgt(lmv, tgt) {
 463                 rc = lmv_connect_mdc(obd, tgt);
 464                 if (rc)
 465                         GOTO(out_disc, rc);
 466         }
 467
 468         lmv->connected = 1;
 469         easize = lmv_mds_md_size(lmv->lmv_mdt_count, LMV_MAGIC);
 470         lmv_init_ea_size(obd->obd_self_export, easize, 0);
 471         EXIT;
 472 unlock:
 473         mutex_unlock(&lmv->lmv_mdt_descs.ltd_mutex);
 474
 475         return rc;
 476
 477 out_disc:
 478         lmv_foreach_tgt(lmv, tgt) {
 479                 tgt->ltd_active = 0;
 480                 if (!tgt->ltd_exp)
 481                         continue;
 482
 483                 --lmv->lmv_mdt_descs.ltd_lmv_desc.ld_active_tgt_count;
 484                 obd_disconnect(tgt->ltd_exp);
 485         }
 486
 487         goto unlock;
 488 }
 489
 490 static int lmv_disconnect_mdc(struct obd_device *obd, struct lmv_tgt_desc *tgt)
 491 {
 492         struct lmv_obd         *lmv = &obd->u.lmv;
 493         struct obd_device      *mdc_obd;
 494         int                     rc;
 495         ENTRY;
 496
 497         LASSERT(tgt != NULL);
 498         LASSERT(obd != NULL);
 499
 500         mdc_obd = class_exp2obd(tgt->ltd_exp);
 501
 502         if (mdc_obd) {
 503                 mdc_obd->obd_force = obd->obd_force;
 504                 mdc_obd->obd_fail = obd->obd_fail;
 505                 mdc_obd->obd_no_recov = obd->obd_no_recov;
 506
 507                 if (lmv->lmv_tgts_kobj)
 508                         sysfs_remove_link(lmv->lmv_tgts_kobj,
 509                                           mdc_obd->obd_name);
 510         }
 511
 512         rc = obd_fid_fini(tgt->ltd_exp->exp_obd);
 513         if (rc)
 514                 CERROR("Can't finanize fids factory\n");
 515
 516         CDEBUG(D_INFO, "Disconnected from %s(%s) successfully\n",
 517                tgt->ltd_exp->exp_obd->obd_name,
 518                tgt->ltd_exp->exp_obd->obd_uuid.uuid);
 519
 520         obd_register_observer(tgt->ltd_exp->exp_obd, NULL);
 521         rc = obd_disconnect(tgt->ltd_exp);
 522         if (rc) {
 523                 if (tgt->ltd_active) {
 524                         CERROR("Target %s disconnect error %d\n",
 525                                tgt->ltd_uuid.uuid, rc);
 526                 }
 527         }
 528
 529         lmv_activate_target(lmv, tgt, 0);
 530         tgt->ltd_exp = NULL;
 531         RETURN(0);
 532 }
 533
 534 static int lmv_disconnect(struct obd_export *exp)
 535 {
 536         struct obd_device *obd = class_exp2obd(exp);
 537         struct lmv_obd *lmv = &obd->u.lmv;
 538         struct lmv_tgt_desc *tgt;
 539         int rc;
 540
 541         ENTRY;
 542
 543         lmv_foreach_connected_tgt(lmv, tgt)
 544                 lmv_disconnect_mdc(obd, tgt);
 545
 546         if (lmv->lmv_tgts_kobj)
 547                 kobject_put(lmv->lmv_tgts_kobj);
 548
 549         if (!lmv->connected)
 550                 class_export_put(exp);
 551         rc = class_disconnect(exp);
 552         lmv->connected = 0;
 553
 554         RETURN(rc);
 555 }
 556
 557 static int lmv_fid2path(struct obd_export *exp, int len, void *karg,
 558                         void __user *uarg)
 559 {
 560         struct obd_device *obd = class_exp2obd(exp);
 561         struct lmv_obd *lmv = &obd->u.lmv;
 562         struct getinfo_fid2path *gf;
 563         struct lmv_tgt_desc *tgt;
 564         struct getinfo_fid2path *remote_gf = NULL;
 565         struct lu_fid root_fid;
 566         int remote_gf_size = 0;
 567         int rc;
 568
 569         gf = karg;
 570         tgt = lmv_fid2tgt(lmv, &gf->gf_fid);
 571         if (IS_ERR(tgt))
 572                 RETURN(PTR_ERR(tgt));
 573
 574         root_fid = *gf->gf_u.gf_root_fid;
 575         LASSERT(fid_is_sane(&root_fid));
 576
 577 repeat_fid2path:
 578         rc = obd_iocontrol(OBD_IOC_FID2PATH, tgt->ltd_exp, len, gf, uarg);
 579         if (rc != 0 && rc != -EREMOTE)
 580                 GOTO(out_fid2path, rc);
 581
 582         /* If remote_gf != NULL, it means just building the
 583          * path on the remote MDT, copy this path segement to gf */
 584         if (remote_gf != NULL) {
 585                 struct getinfo_fid2path *ori_gf;
 586                 char *ptr;
 587                 int len;
 588
 589                 ori_gf = (struct getinfo_fid2path *)karg;
 590                 if (strlen(ori_gf->gf_u.gf_path) + 1 +
 591                     strlen(gf->gf_u.gf_path) + 1 > ori_gf->gf_pathlen)
 592                         GOTO(out_fid2path, rc = -EOVERFLOW);
 593
 594                 ptr = ori_gf->gf_u.gf_path;
 595
 596                 len = strlen(gf->gf_u.gf_path);
 597                 /* move the current path to the right to release space
 598                  * for closer-to-root part */
 599                 memmove(ptr + len + 1, ptr, strlen(ori_gf->gf_u.gf_path));
 600                 memcpy(ptr, gf->gf_u.gf_path, len);
 601                 ptr[len] = '/';
 602         }
 603
 604         CDEBUG(D_INFO, "%s: get path %s "DFID" rec: %llu ln: %u\n",
 605                tgt->ltd_exp->exp_obd->obd_name,
 606                gf->gf_u.gf_path, PFID(&gf->gf_fid), gf->gf_recno,
 607                gf->gf_linkno);
 608
 609         if (rc == 0)
 610                 GOTO(out_fid2path, rc);
 611
 612         /* sigh, has to go to another MDT to do path building further */
 613         if (remote_gf == NULL) {
 614                 remote_gf_size = sizeof(*remote_gf) + PATH_MAX;
 615                 OBD_ALLOC(remote_gf, remote_gf_size);
 616                 if (remote_gf == NULL)
 617                         GOTO(out_fid2path, rc = -ENOMEM);
 618                 remote_gf->gf_pathlen = PATH_MAX;
 619         }
 620
 621         if (!fid_is_sane(&gf->gf_fid)) {
 622                 CERROR("%s: invalid FID "DFID": rc = %d\n",
 623                        tgt->ltd_exp->exp_obd->obd_name,
 624                        PFID(&gf->gf_fid), -EINVAL);
 625                 GOTO(out_fid2path, rc = -EINVAL);
 626         }
 627
 628         tgt = lmv_fid2tgt(lmv, &gf->gf_fid);
 629         if (IS_ERR(tgt))
 630                 GOTO(out_fid2path, rc = -EINVAL);
 631
 632         remote_gf->gf_fid = gf->gf_fid;
 633         remote_gf->gf_recno = -1;
 634         remote_gf->gf_linkno = -1;
 635         memset(remote_gf->gf_u.gf_path, 0, remote_gf->gf_pathlen);
 636         *remote_gf->gf_u.gf_root_fid = root_fid;
 637         gf = remote_gf;
 638         goto repeat_fid2path;
 639
 640 out_fid2path:
 641         if (remote_gf != NULL)
 642                 OBD_FREE(remote_gf, remote_gf_size);
 643         RETURN(rc);
 644 }
 645
 646 static int lmv_hsm_req_count(struct lmv_obd *lmv,
 647                              const struct hsm_user_request *hur,
 648                              const struct lmv_tgt_desc *tgt_mds)
 649 {
 650         struct lmv_tgt_desc *curr_tgt;
 651         __u32 i;
 652         int nr = 0;
 653
 654         /* count how many requests must be sent to the given target */
 655         for (i = 0; i < hur->hur_request.hr_itemcount; i++) {
 656                 curr_tgt = lmv_fid2tgt(lmv, &hur->hur_user_item[i].hui_fid);
 657                 if (IS_ERR(curr_tgt))
 658                         RETURN(PTR_ERR(curr_tgt));
 659                 if (obd_uuid_equals(&curr_tgt->ltd_uuid, &tgt_mds->ltd_uuid))
 660                         nr++;
 661         }
 662         return nr;
 663 }
 664
 665 static int lmv_hsm_req_build(struct lmv_obd *lmv,
 666                               struct hsm_user_request *hur_in,
 667                               const struct lmv_tgt_desc *tgt_mds,
 668                               struct hsm_user_request *hur_out)
 669 {
 670         __u32 i, nr_out;
 671         struct lmv_tgt_desc *curr_tgt;
 672
 673         /* build the hsm_user_request for the given target */
 674         hur_out->hur_request = hur_in->hur_request;
 675         nr_out = 0;
 676         for (i = 0; i < hur_in->hur_request.hr_itemcount; i++) {
 677                 curr_tgt = lmv_fid2tgt(lmv, &hur_in->hur_user_item[i].hui_fid);
 678                 if (IS_ERR(curr_tgt))
 679                         RETURN(PTR_ERR(curr_tgt));
 680                 if (obd_uuid_equals(&curr_tgt->ltd_uuid, &tgt_mds->ltd_uuid)) {
 681                         hur_out->hur_user_item[nr_out] =
 682                                                 hur_in->hur_user_item[i];
 683                         nr_out++;
 684                 }
 685         }
 686         hur_out->hur_request.hr_itemcount = nr_out;
 687         memcpy(hur_data(hur_out), hur_data(hur_in),
 688                hur_in->hur_request.hr_data_len);
 689
 690         RETURN(0);
 691 }
 692
 693 static int lmv_hsm_ct_unregister(struct obd_device *obd, unsigned int cmd,
 694                                  int len, struct lustre_kernelcomm *lk,
 695                                  void __user *uarg)
 696 {
 697         struct lmv_obd *lmv = &obd->u.lmv;
 698         struct lu_tgt_desc *tgt;
 699         int rc;
 700
 701         ENTRY;
 702
 703         /* unregister request (call from llapi_hsm_copytool_fini) */
 704         lmv_foreach_connected_tgt(lmv, tgt)
 705                 /* best effort: try to clean as much as possible
 706                  * (continue on error) */
 707                 obd_iocontrol(cmd, tgt->ltd_exp, len, lk, uarg);
 708
 709         /* Whatever the result, remove copytool from kuc groups.
 710          * Unreached coordinators will get EPIPE on next requests
 711          * and will unregister automatically.
 712          */
 713         rc = libcfs_kkuc_group_rem(&obd->obd_uuid, lk->lk_uid, lk->lk_group);
 714
 715         RETURN(rc);
 716 }
 717
 718 static int lmv_hsm_ct_register(struct obd_device *obd, unsigned int cmd,
 719                                int len, struct lustre_kernelcomm *lk,
 720                                void __user *uarg)
 721 {
 722         struct lmv_obd *lmv = &obd->u.lmv;
 723         struct file *filp;
 724         bool any_set = false;
 725         struct kkuc_ct_data *kcd;
 726         size_t kcd_size;
 727         struct lu_tgt_desc *tgt;
 728         __u32 i;
 729         int err;
 730         int rc = 0;
 731
 732         ENTRY;
 733
 734         filp = fget(lk->lk_wfd);
 735         if (!filp)
 736                 RETURN(-EBADF);
 737
 738         if (lk->lk_flags & LK_FLG_DATANR)
 739                 kcd_size = offsetof(struct kkuc_ct_data,
 740                                     kcd_archives[lk->lk_data_count]);
 741         else
 742                 kcd_size = sizeof(*kcd);
 743
 744         OBD_ALLOC(kcd, kcd_size);
 745         if (kcd == NULL)
 746                 GOTO(err_fput, rc = -ENOMEM);
 747
 748         kcd->kcd_nr_archives = lk->lk_data_count;
 749         if (lk->lk_flags & LK_FLG_DATANR) {
 750                 kcd->kcd_magic = KKUC_CT_DATA_ARRAY_MAGIC;
 751                 if (lk->lk_data_count > 0)
 752                         memcpy(kcd->kcd_archives, lk->lk_data,
 753                                sizeof(*kcd->kcd_archives) * lk->lk_data_count);
 754         } else {
 755                 kcd->kcd_magic = KKUC_CT_DATA_BITMAP_MAGIC;
 756         }
 757
 758         rc = libcfs_kkuc_group_add(filp, &obd->obd_uuid, lk->lk_uid,
 759                                    lk->lk_group, kcd, kcd_size);
 760         OBD_FREE(kcd, kcd_size);
 761         if (rc)
 762                 GOTO(err_fput, rc);
 763
 764         /* All or nothing: try to register to all MDS.
 765          * In case of failure, unregister from previous MDS,
 766          * except if it because of inactive target. */
 767         lmv_foreach_connected_tgt(lmv, tgt) {
 768                 err = obd_iocontrol(cmd, tgt->ltd_exp, len, lk, uarg);
 769                 if (err) {
 770                         if (tgt->ltd_active) {
 771                                 /* permanent error */
 772                                 CERROR("%s: iocontrol MDC %s on MDT"
 773                                        " idx %d cmd %x: err = %d\n",
 774                                        lmv2obd_dev(lmv)->obd_name,
 775                                        tgt->ltd_uuid.uuid, tgt->ltd_index, cmd,
 776                                        err);
 777                                 rc = err;
 778                                 lk->lk_flags |= LK_FLG_STOP;
 779                                 i = tgt->ltd_index;
 780                                 /* unregister from previous MDS */
 781                                 lmv_foreach_connected_tgt(lmv, tgt) {
 782                                         if (tgt->ltd_index >= i)
 783                                                 break;
 784
 785                                         obd_iocontrol(cmd, tgt->ltd_exp, len,
 786                                                       lk, uarg);
 787                                 }
 788                                 GOTO(err_kkuc_rem, rc);
 789                         }
 790                         /* else: transient error.
 791                          * kuc will register to the missing MDT
 792                          * when it is back */
 793                 } else {
 794                         any_set = true;
 795                 }
 796         }
 797
 798         if (!any_set)
 799                 /* no registration done: return error */
 800                 GOTO(err_kkuc_rem, rc = -ENOTCONN);
 801
 802         RETURN(0);
 803
 804 err_kkuc_rem:
 805         libcfs_kkuc_group_rem(&obd->obd_uuid, lk->lk_uid, lk->lk_group);
 806
 807 err_fput:
 808         fput(filp);
 809         return rc;
 810 }
 811
 812 static int lmv_iocontrol(unsigned int cmd, struct obd_export *exp,
 813                          int len, void *karg, void __user *uarg)
 814 {
 815         struct obd_device *obd = class_exp2obd(exp);
 816         struct lmv_obd *lmv = &obd->u.lmv;
 817         struct lu_tgt_desc *tgt = NULL;
 818         int set = 0;
 819         __u32 count = lmv->lmv_mdt_count;
 820         int rc = 0;
 821
 822         ENTRY;
 823
 824         if (count == 0)
 825                 RETURN(-ENOTTY);
 826
 827         switch (cmd) {
 828         case IOC_OBD_STATFS: {
 829                 struct obd_ioctl_data *data = karg;
 830                 struct obd_device *mdc_obd;
 831                 struct obd_statfs stat_buf = {0};
 832                 __u32 index;
 833
 834                 memcpy(&index, data->ioc_inlbuf2, sizeof(__u32));
 835
 836                 if (index >= lmv->lmv_mdt_descs.ltd_tgts_size)
 837                         RETURN(-ENODEV);
 838
 839                 tgt = lmv_tgt(lmv, index);
 840                 if (!tgt)
 841                         RETURN(-EAGAIN);
 842
 843                 if (!tgt->ltd_active)
 844                         RETURN(-ENODATA);
 845
 846                 mdc_obd = class_exp2obd(tgt->ltd_exp);
 847                 if (!mdc_obd)
 848                         RETURN(-EINVAL);
 849
 850                 /* copy UUID */
 851                 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(mdc_obd),
 852                                  min((int) data->ioc_plen2,
 853                                      (int) sizeof(struct obd_uuid))))
 854                         RETURN(-EFAULT);
 855
 856                 rc = obd_statfs(NULL, tgt->ltd_exp, &stat_buf,
 857                                 ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS,
 858                                 0);
 859                 if (rc)
 860                         RETURN(rc);
 861                 if (copy_to_user(data->ioc_pbuf1, &stat_buf,
 862                                  min((int) data->ioc_plen1,
 863                                      (int) sizeof(stat_buf))))
 864                         RETURN(-EFAULT);
 865                 break;
 866         }
 867         case OBD_IOC_QUOTACTL: {
 868                 struct if_quotactl *qctl = karg;
 869                 struct obd_quotactl *oqctl;
 870
 871                 if (qctl->qc_valid == QC_MDTIDX) {
 872                         tgt = lmv_tgt(lmv, qctl->qc_idx);
 873                 } else if (qctl->qc_valid == QC_UUID) {
 874                         lmv_foreach_tgt(lmv, tgt) {
 875                                 if (!obd_uuid_equals(&tgt->ltd_uuid,
 876                                                      &qctl->obd_uuid))
 877                                         continue;
 878
 879                                 if (!tgt->ltd_exp)
 880                                         RETURN(-EINVAL);
 881
 882                                 break;
 883                         }
 884                 } else {
 885                         RETURN(-EINVAL);
 886                 }
 887
 888                 if (!tgt || !tgt->ltd_exp)
 889                         RETURN(-EINVAL);
 890
 891                 OBD_ALLOC_PTR(oqctl);
 892                 if (!oqctl)
 893                         RETURN(-ENOMEM);
 894
 895                 QCTL_COPY(oqctl, qctl);
 896                 rc = obd_quotactl(tgt->ltd_exp, oqctl);
 897                 if (rc == 0) {
 898                         QCTL_COPY(qctl, oqctl);
 899                         qctl->qc_valid = QC_MDTIDX;
 900                         qctl->obd_uuid = tgt->ltd_uuid;
 901                 }
 902                 OBD_FREE_PTR(oqctl);
 903                 break;
 904         }
 905         case LL_IOC_GET_CONNECT_FLAGS: {
 906                 tgt = lmv_tgt(lmv, 0);
 907                 rc = -ENODATA;
 908                 if (tgt && tgt->ltd_exp)
 909                         rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
 910                 break;
 911         }
 912         case LL_IOC_FID2MDTIDX: {
 913                 struct lu_fid *fid = karg;
 914                 int             mdt_index;
 915
 916                 rc = lmv_fld_lookup(lmv, fid, &mdt_index);
 917                 if (rc != 0)
 918                         RETURN(rc);
 919
 920                 /* Note: this is from llite(see ll_dir_ioctl()), @uarg does not
 921                  * point to user space memory for FID2MDTIDX. */
 922                 *(__u32 *)uarg = mdt_index;
 923                 break;
 924         }
 925         case OBD_IOC_FID2PATH: {
 926                 rc = lmv_fid2path(exp, len, karg, uarg);
 927                 break;
 928         }
 929         case LL_IOC_HSM_STATE_GET:
 930         case LL_IOC_HSM_STATE_SET:
 931         case LL_IOC_HSM_ACTION: {
 932                 struct md_op_data *op_data = karg;
 933
 934                 tgt = lmv_fid2tgt(lmv, &op_data->op_fid1);
 935                 if (IS_ERR(tgt))
 936                         RETURN(PTR_ERR(tgt));
 937
 938                 if (tgt->ltd_exp == NULL)
 939                         RETURN(-EINVAL);
 940
 941                 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
 942                 break;
 943         }
 944         case LL_IOC_HSM_PROGRESS: {
 945                 const struct hsm_progress_kernel *hpk = karg;
 946
 947                 tgt = lmv_fid2tgt(lmv, &hpk->hpk_fid);
 948                 if (IS_ERR(tgt))
 949                         RETURN(PTR_ERR(tgt));
 950                 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
 951                 break;
 952         }
 953         case LL_IOC_HSM_REQUEST: {
 954                 struct hsm_user_request *hur = karg;
 955                 unsigned int reqcount = hur->hur_request.hr_itemcount;
 956
 957                 if (reqcount == 0)
 958                         RETURN(0);
 959
 960                 /* if the request is about a single fid
 961                  * or if there is a single MDS, no need to split
 962                  * the request. */
 963                 if (reqcount == 1 || count == 1) {
 964                         tgt = lmv_fid2tgt(lmv, &hur->hur_user_item[0].hui_fid);
 965                         if (IS_ERR(tgt))
 966                                 RETURN(PTR_ERR(tgt));
 967                         rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
 968                 } else {
 969                         /* split fid list to their respective MDS */
 970                         lmv_foreach_connected_tgt(lmv, tgt) {
 971                                 int nr, rc1;
 972                                 size_t reqlen;
 973                                 struct hsm_user_request *req;
 974
 975                                 nr = lmv_hsm_req_count(lmv, hur, tgt);
 976                                 if (nr < 0)
 977                                         RETURN(nr);
 978                                 if (nr == 0) /* nothing for this MDS */
 979                                         continue;
 980
 981                                 /* build a request with fids for this MDS */
 982                                 reqlen = offsetof(typeof(*hur),
 983                                                   hur_user_item[nr])
 984                                                 + hur->hur_request.hr_data_len;
 985                                 OBD_ALLOC_LARGE(req, reqlen);
 986                                 if (req == NULL)
 987                                         RETURN(-ENOMEM);
 988                                 rc1 = lmv_hsm_req_build(lmv, hur, tgt, req);
 989                                 if (rc1 < 0)
 990                                         GOTO(hsm_req_err, rc1);
 991                                 rc1 = obd_iocontrol(cmd, tgt->ltd_exp, reqlen,
 992                                                     req, uarg);
 993 hsm_req_err:
 994                                 if (rc1 != 0 && rc == 0)
 995                                         rc = rc1;
 996                                 OBD_FREE_LARGE(req, reqlen);
 997                         }
 998                 }
 999                 break;
1000         }
1001         case LL_IOC_LOV_SWAP_LAYOUTS: {
1002                 struct md_op_data *op_data = karg;
1003                 struct lmv_tgt_desc *tgt1, *tgt2;
1004
1005                 tgt1 = lmv_fid2tgt(lmv, &op_data->op_fid1);
1006                 if (IS_ERR(tgt1))
1007                         RETURN(PTR_ERR(tgt1));
1008
1009                 tgt2 = lmv_fid2tgt(lmv, &op_data->op_fid2);
1010                 if (IS_ERR(tgt2))
1011                         RETURN(PTR_ERR(tgt2));
1012
1013                 if ((tgt1->ltd_exp == NULL) || (tgt2->ltd_exp == NULL))
1014                         RETURN(-EINVAL);
1015
1016                 /* only files on same MDT can have their layouts swapped */
1017                 if (tgt1->ltd_index != tgt2->ltd_index)
1018                         RETURN(-EPERM);
1019
1020                 rc = obd_iocontrol(cmd, tgt1->ltd_exp, len, karg, uarg);
1021                 break;
1022         }
1023         case LL_IOC_HSM_CT_START: {
1024                 struct lustre_kernelcomm *lk = karg;
1025                 if (lk->lk_flags & LK_FLG_STOP)
1026                         rc = lmv_hsm_ct_unregister(obd, cmd, len, lk, uarg);
1027                 else
1028                         rc = lmv_hsm_ct_register(obd, cmd, len, lk, uarg);
1029                 break;
1030         }
1031         default:
1032                 lmv_foreach_connected_tgt(lmv, tgt) {
1033                         struct obd_device *mdc_obd;
1034                         int err;
1035
1036                         /* ll_umount_begin() sets force flag but for lmv, not
1037                          * mdc. Let's pass it through */
1038                         mdc_obd = class_exp2obd(tgt->ltd_exp);
1039                         mdc_obd->obd_force = obd->obd_force;
1040                         err = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1041                         if (err) {
1042                                 if (tgt->ltd_active) {
1043                                         CERROR("error: iocontrol MDC %s on MDT"
1044                                                " idx %d cmd %x: err = %d\n",
1045                                                tgt->ltd_uuid.uuid,
1046                                                tgt->ltd_index, cmd, err);
1047                                         if (!rc)
1048                                                 rc = err;
1049                                 }
1050                         } else
1051                                 set = 1;
1052                 }
1053                 if (!set && !rc)
1054                         rc = -EIO;
1055         }
1056         RETURN(rc);
1057 }
1058
1059 int lmv_fid_alloc(const struct lu_env *env, struct obd_export *exp,
1060                   struct lu_fid *fid, struct md_op_data *op_data)
1061 {
1062         struct obd_device *obd = class_exp2obd(exp);
1063         struct lmv_obd *lmv = &obd->u.lmv;
1064         struct lmv_tgt_desc *tgt;
1065         int rc;
1066
1067         ENTRY;
1068
1069         LASSERT(op_data);
1070         LASSERT(fid);
1071
1072         tgt = lmv_tgt(lmv, op_data->op_mds);
1073         if (!tgt)
1074                 RETURN(-ENODEV);
1075
1076         if (!tgt->ltd_active || !tgt->ltd_exp)
1077                 RETURN(-ENODEV);
1078
1079         /*
1080          * New seq alloc and FLD setup should be atomic. Otherwise we may find
1081          * on server that seq in new allocated fid is not yet known.
1082          */
1083         mutex_lock(&tgt->ltd_fid_mutex);
1084         rc = obd_fid_alloc(NULL, tgt->ltd_exp, fid, NULL);
1085         mutex_unlock(&tgt->ltd_fid_mutex);
1086         if (rc > 0) {
1087                 LASSERT(fid_is_sane(fid));
1088                 rc = 0;
1089         }
1090
1091         RETURN(rc);
1092 }
1093
1094 static int lmv_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
1095 {
1096         struct lmv_obd *lmv = &obd->u.lmv;
1097         struct lmv_desc *desc;
1098         struct lnet_process_id lnet_id;
1099         int i = 0;
1100         int rc;
1101
1102         ENTRY;
1103
1104         if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
1105                 CERROR("LMV setup requires a descriptor\n");
1106                 RETURN(-EINVAL);
1107         }
1108
1109         desc = (struct lmv_desc *)lustre_cfg_buf(lcfg, 1);
1110         if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
1111                 CERROR("Lmv descriptor size wrong: %d > %d\n",
1112                        (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
1113                 RETURN(-EINVAL);
1114         }
1115
1116         obd_str2uuid(&lmv->lmv_mdt_descs.ltd_lmv_desc.ld_uuid,
1117                      desc->ld_uuid.uuid);
1118         lmv->lmv_mdt_descs.ltd_lmv_desc.ld_tgt_count = 0;
1119         lmv->lmv_mdt_descs.ltd_lmv_desc.ld_active_tgt_count = 0;
1120         lmv->lmv_mdt_descs.ltd_lmv_desc.ld_qos_maxage =
1121                 LMV_DESC_QOS_MAXAGE_DEFAULT;
1122         lmv->max_def_easize = 0;
1123         lmv->max_easize = 0;
1124
1125         spin_lock_init(&lmv->lmv_lock);
1126
1127         /*
1128          * initialize rr_index to lower 32bit of netid, so that client
1129          * can distribute subdirs evenly from the beginning.
1130          */
1131         while (LNetGetId(i++, &lnet_id) != -ENOENT) {
1132                 if (lnet_id.nid != LNET_NID_LO_0) {
1133                         lmv->lmv_qos_rr_index = (u32)lnet_id.nid;
1134                         break;
1135                 }
1136         }
1137
1138         rc = lmv_tunables_init(obd);
1139         if (rc)
1140                 CWARN("%s: error adding LMV sysfs/debugfs files: rc = %d\n",
1141                       obd->obd_name, rc);
1142
1143         rc = fld_client_init(&lmv->lmv_fld, obd->obd_name,
1144                              LUSTRE_CLI_FLD_HASH_DHT);
1145         if (rc)
1146                 CERROR("Can't init FLD, err %d\n", rc);
1147
1148         rc = lu_tgt_descs_init(&lmv->lmv_mdt_descs, true);
1149         if (rc)
1150                 CWARN("%s: error initialize target table: rc = %d\n",
1151                       obd->obd_name, rc);
1152
1153         RETURN(rc);
1154 }
1155
1156 static int lmv_cleanup(struct obd_device *obd)
1157 {
1158         struct lmv_obd *lmv = &obd->u.lmv;
1159         struct lu_tgt_desc *tgt;
1160         struct lu_tgt_desc *tmp;
1161
1162         ENTRY;
1163
1164         fld_client_fini(&lmv->lmv_fld);
1165         lmv_foreach_tgt_safe(lmv, tgt, tmp)
1166                 lmv_del_target(lmv, tgt);
1167         lu_tgt_descs_fini(&lmv->lmv_mdt_descs);
1168
1169         RETURN(0);
1170 }
1171
1172 static int lmv_process_config(struct obd_device *obd, size_t len, void *buf)
1173 {
1174         struct lustre_cfg       *lcfg = buf;
1175         struct obd_uuid         obd_uuid;
1176         int                     gen;
1177         __u32                   index;
1178         int                     rc;
1179         ENTRY;
1180
1181         switch (lcfg->lcfg_command) {
1182         case LCFG_ADD_MDC:
1183                 /* modify_mdc_tgts add 0:lustre-clilmv  1:lustre-MDT0000_UUID
1184                  * 2:0  3:1  4:lustre-MDT0000-mdc_UUID */
1185                 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(obd_uuid.uuid))
1186                         GOTO(out, rc = -EINVAL);
1187
1188                 obd_str2uuid(&obd_uuid,  lustre_cfg_buf(lcfg, 1));
1189
1190                 if (sscanf(lustre_cfg_buf(lcfg, 2), "%u", &index) != 1)
1191                         GOTO(out, rc = -EINVAL);
1192                 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", &gen) != 1)
1193                         GOTO(out, rc = -EINVAL);
1194                 rc = lmv_add_target(obd, &obd_uuid, index, gen);
1195                 GOTO(out, rc);
1196         default:
1197                 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
1198                 GOTO(out, rc = -EINVAL);
1199         }
1200 out:
1201         RETURN(rc);
1202 }
1203
1204 static int lmv_select_statfs_mdt(struct lmv_obd *lmv, __u32 flags)
1205 {
1206         int i;
1207
1208         if (flags & OBD_STATFS_FOR_MDT0)
1209                 return 0;
1210
1211         if (lmv->lmv_statfs_start || lmv->lmv_mdt_count == 1)
1212                 return lmv->lmv_statfs_start;
1213
1214         /* choose initial MDT for this client */
1215         for (i = 0;; i++) {
1216                 struct lnet_process_id lnet_id;
1217                 if (LNetGetId(i, &lnet_id) == -ENOENT)
1218                         break;
1219
1220                 if (lnet_id.nid != LNET_NID_LO_0) {
1221                         /* We dont need a full 64-bit modulus, just enough
1222                          * to distribute the requests across MDTs evenly.
1223                          */
1224                         lmv->lmv_statfs_start = (u32)lnet_id.nid %
1225                                                 lmv->lmv_mdt_count;
1226                         break;
1227                 }
1228         }
1229
1230         return lmv->lmv_statfs_start;
1231 }
1232
1233 static int lmv_statfs(const struct lu_env *env, struct obd_export *exp,
1234                       struct obd_statfs *osfs, time64_t max_age, __u32 flags)
1235 {
1236         struct obd_device *obd = class_exp2obd(exp);
1237         struct lmv_obd *lmv = &obd->u.lmv;
1238         struct obd_statfs *temp;
1239         struct lu_tgt_desc *tgt;
1240         __u32 i;
1241         __u32 idx;
1242         int rc = 0;
1243
1244         ENTRY;
1245
1246         OBD_ALLOC(temp, sizeof(*temp));
1247         if (temp == NULL)
1248                 RETURN(-ENOMEM);
1249
1250         /* distribute statfs among MDTs */
1251         idx = lmv_select_statfs_mdt(lmv, flags);
1252
1253         for (i = 0; i < lmv->lmv_mdt_descs.ltd_tgts_size; i++, idx++) {
1254                 idx = idx % lmv->lmv_mdt_descs.ltd_tgts_size;
1255                 tgt = lmv_tgt(lmv, idx);
1256                 if (!tgt || !tgt->ltd_exp)
1257                         continue;
1258
1259                 rc = obd_statfs(env, tgt->ltd_exp, temp, max_age, flags);
1260                 if (rc) {
1261                         CERROR("%s: can't stat MDS #%d: rc = %d\n",
1262                                tgt->ltd_exp->exp_obd->obd_name, i, rc);
1263                         GOTO(out_free_temp, rc);
1264                 }
1265
1266                 if (temp->os_state & OS_STATE_SUM ||
1267                     flags == OBD_STATFS_FOR_MDT0) {
1268                         /* reset to the last aggregated values
1269                          * and don't sum with non-aggrated data */
1270                         /* If the statfs is from mount, it needs to retrieve
1271                          * necessary information from MDT0. i.e. mount does
1272                          * not need the merged osfs from all of MDT. Also
1273                          * clients can be mounted as long as MDT0 is in
1274                          * service */
1275                         *osfs = *temp;
1276                         break;
1277                 }
1278
1279                 if (i == 0) {
1280                         *osfs = *temp;
1281                 } else {
1282                         osfs->os_bavail += temp->os_bavail;
1283                         osfs->os_blocks += temp->os_blocks;
1284                         osfs->os_ffree += temp->os_ffree;
1285                         osfs->os_files += temp->os_files;
1286                         osfs->os_granted += temp->os_granted;
1287                 }
1288         }
1289
1290         EXIT;
1291 out_free_temp:
1292         OBD_FREE(temp, sizeof(*temp));
1293         return rc;
1294 }
1295
1296 static int lmv_statfs_update(void *cookie, int rc)
1297 {
1298         struct obd_info *oinfo = cookie;
1299         struct obd_device *obd = oinfo->oi_obd;
1300         struct lmv_obd *lmv = &obd->u.lmv;
1301         struct lmv_tgt_desc *tgt = oinfo->oi_tgt;
1302         struct obd_statfs *osfs = oinfo->oi_osfs;
1303
1304         /*
1305          * NB: don't deactivate TGT upon error, because we may not trigger async
1306          * statfs any longer, then there is no chance to activate TGT.
1307          */
1308         if (!rc) {
1309                 spin_lock(&lmv->lmv_lock);
1310                 tgt->ltd_statfs = *osfs;
1311                 tgt->ltd_statfs_age = ktime_get_seconds();
1312                 spin_unlock(&lmv->lmv_lock);
1313                 lmv->lmv_qos.lq_dirty = 1;
1314         }
1315
1316         return rc;
1317 }
1318
1319 /* update tgt statfs async if it's ld_qos_maxage old */
1320 int lmv_statfs_check_update(struct obd_device *obd, struct lmv_tgt_desc *tgt)
1321 {
1322         struct obd_info oinfo = {
1323                 .oi_obd = obd,
1324                 .oi_tgt = tgt,
1325                 .oi_cb_up = lmv_statfs_update,
1326         };
1327         int rc;
1328
1329         if (ktime_get_seconds() - tgt->ltd_statfs_age <
1330             obd->u.lmv.lmv_mdt_descs.ltd_lmv_desc.ld_qos_maxage)
1331                 return 0;
1332
1333         rc = obd_statfs_async(tgt->ltd_exp, &oinfo, 0, NULL);
1334
1335         return rc;
1336 }
1337
1338 static int lmv_get_root(struct obd_export *exp, const char *fileset,
1339                         struct lu_fid *fid)
1340 {
1341         struct obd_device *obd = exp->exp_obd;
1342         struct lmv_obd *lmv = &obd->u.lmv;
1343         struct lu_tgt_desc *tgt = lmv_tgt(lmv, 0);
1344         int rc;
1345
1346         ENTRY;
1347
1348         if (!tgt)
1349                 RETURN(-ENODEV);
1350
1351         rc = md_get_root(tgt->ltd_exp, fileset, fid);
1352         RETURN(rc);
1353 }
1354
1355 static int lmv_getxattr(struct obd_export *exp, const struct lu_fid *fid,
1356                         u64 obd_md_valid, const char *name, size_t buf_size,
1357                         struct ptlrpc_request **req)
1358 {
1359         struct obd_device *obd = exp->exp_obd;
1360         struct lmv_obd *lmv = &obd->u.lmv;
1361         struct lmv_tgt_desc *tgt;
1362         int rc;
1363
1364         ENTRY;
1365
1366         tgt = lmv_fid2tgt(lmv, fid);
1367         if (IS_ERR(tgt))
1368                 RETURN(PTR_ERR(tgt));
1369
1370         rc = md_getxattr(tgt->ltd_exp, fid, obd_md_valid, name, buf_size, req);
1371
1372         RETURN(rc);
1373 }
1374
1375 static int lmv_setxattr(struct obd_export *exp, const struct lu_fid *fid,
1376                         u64 obd_md_valid, const char *name,
1377                         const void *value, size_t value_size,
1378                         unsigned int xattr_flags, u32 suppgid,
1379                         struct ptlrpc_request **req)
1380 {
1381         struct obd_device *obd = exp->exp_obd;
1382         struct lmv_obd *lmv = &obd->u.lmv;
1383         struct lmv_tgt_desc *tgt;
1384         int rc;
1385
1386         ENTRY;
1387
1388         tgt = lmv_fid2tgt(lmv, fid);
1389         if (IS_ERR(tgt))
1390                 RETURN(PTR_ERR(tgt));
1391
1392         rc = md_setxattr(tgt->ltd_exp, fid, obd_md_valid, name,
1393                          value, value_size, xattr_flags, suppgid, req);
1394
1395         RETURN(rc);
1396 }
1397
1398 static int lmv_getattr(struct obd_export *exp, struct md_op_data *op_data,
1399                        struct ptlrpc_request **request)
1400 {
1401         struct obd_device *obd = exp->exp_obd;
1402         struct lmv_obd *lmv = &obd->u.lmv;
1403         struct lmv_tgt_desc *tgt;
1404         int rc;
1405
1406         ENTRY;
1407
1408         tgt = lmv_fid2tgt(lmv, &op_data->op_fid1);
1409         if (IS_ERR(tgt))
1410                 RETURN(PTR_ERR(tgt));
1411
1412         if (op_data->op_flags & MF_GET_MDT_IDX) {
1413                 op_data->op_mds = tgt->ltd_index;
1414                 RETURN(0);
1415         }
1416
1417         rc = md_getattr(tgt->ltd_exp, op_data, request);
1418
1419         RETURN(rc);
1420 }
1421
1422 static int lmv_null_inode(struct obd_export *exp, const struct lu_fid *fid)
1423 {
1424         struct obd_device *obd = exp->exp_obd;
1425         struct lmv_obd *lmv = &obd->u.lmv;
1426         struct lu_tgt_desc *tgt;
1427
1428         ENTRY;
1429
1430         CDEBUG(D_INODE, "CBDATA for "DFID"\n", PFID(fid));
1431
1432         /*
1433          * With DNE every object can have two locks in different namespaces:
1434          * lookup lock in space of MDT storing direntry and update/open lock in
1435          * space of MDT storing inode.
1436          */
1437         lmv_foreach_connected_tgt(lmv, tgt)
1438                 md_null_inode(tgt->ltd_exp, fid);
1439
1440         RETURN(0);
1441 }
1442
1443 static int lmv_close(struct obd_export *exp, struct md_op_data *op_data,
1444                      struct md_open_data *mod, struct ptlrpc_request **request)
1445 {
1446         struct obd_device *obd = exp->exp_obd;
1447         struct lmv_obd *lmv = &obd->u.lmv;
1448         struct lmv_tgt_desc *tgt;
1449         int rc;
1450
1451         ENTRY;
1452
1453         tgt = lmv_fid2tgt(lmv, &op_data->op_fid1);
1454         if (IS_ERR(tgt))
1455                 RETURN(PTR_ERR(tgt));
1456
1457         CDEBUG(D_INODE, "CLOSE "DFID"\n", PFID(&op_data->op_fid1));
1458         rc = md_close(tgt->ltd_exp, op_data, mod, request);
1459         RETURN(rc);
1460 }
1461
1462 static struct lu_tgt_desc *lmv_locate_tgt_qos(struct lmv_obd *lmv, __u32 *mdt)
1463 {
1464         struct lu_tgt_desc *tgt;
1465         __u64 total_weight = 0;
1466         __u64 cur_weight = 0;
1467         __u64 rand;
1468         int rc;
1469
1470         ENTRY;
1471
1472         if (!ltd_qos_is_usable(&lmv->lmv_mdt_descs))
1473                 RETURN(ERR_PTR(-EAGAIN));
1474
1475         down_write(&lmv->lmv_qos.lq_rw_sem);
1476
1477         if (!ltd_qos_is_usable(&lmv->lmv_mdt_descs))
1478                 GOTO(unlock, tgt = ERR_PTR(-EAGAIN));
1479
1480         rc = ltd_qos_penalties_calc(&lmv->lmv_mdt_descs);
1481         if (rc)
1482                 GOTO(unlock, tgt = ERR_PTR(rc));
1483
1484         lmv_foreach_tgt(lmv, tgt) {
1485                 tgt->ltd_qos.ltq_usable = 0;
1486                 if (!tgt->ltd_exp || !tgt->ltd_active)
1487                         continue;
1488
1489                 tgt->ltd_qos.ltq_usable = 1;
1490                 lu_tgt_qos_weight_calc(tgt);
1491                 total_weight += tgt->ltd_qos.ltq_weight;
1492         }
1493
1494         rand = lu_prandom_u64_max(total_weight);
1495
1496         lmv_foreach_connected_tgt(lmv, tgt) {
1497                 if (!tgt->ltd_qos.ltq_usable)
1498                         continue;
1499
1500                 cur_weight += tgt->ltd_qos.ltq_weight;
1501                 if (cur_weight < rand)
1502                         continue;
1503
1504                 *mdt = tgt->ltd_index;
1505                 ltd_qos_update(&lmv->lmv_mdt_descs, tgt, &total_weight);
1506                 GOTO(unlock, rc = 0);
1507         }
1508
1509         /* no proper target found */
1510         GOTO(unlock, tgt = ERR_PTR(-EAGAIN));
1511 unlock:
1512         up_write(&lmv->lmv_qos.lq_rw_sem);
1513
1514         return tgt;
1515 }
1516
1517 static struct lu_tgt_desc *lmv_locate_tgt_rr(struct lmv_obd *lmv, __u32 *mdt)
1518 {
1519         struct lu_tgt_desc *tgt;
1520         int i;
1521         int index;
1522
1523         ENTRY;
1524
1525         spin_lock(&lmv->lmv_qos.lq_rr.lqr_alloc);
1526         for (i = 0; i < lmv->lmv_mdt_descs.ltd_tgts_size; i++) {
1527                 index = (i + lmv->lmv_qos_rr_index) %
1528                         lmv->lmv_mdt_descs.ltd_tgts_size;
1529                 tgt = lmv_tgt(lmv, index);
1530                 if (!tgt || !tgt->ltd_exp || !tgt->ltd_active)
1531                         continue;
1532
1533                 *mdt = tgt->ltd_index;
1534                 lmv->lmv_qos_rr_index = (*mdt + 1) %
1535                                         lmv->lmv_mdt_descs.ltd_tgts_size;
1536                 spin_unlock(&lmv->lmv_qos.lq_rr.lqr_alloc);
1537
1538                 RETURN(tgt);
1539         }
1540         spin_unlock(&lmv->lmv_qos.lq_rr.lqr_alloc);
1541
1542         RETURN(ERR_PTR(-ENODEV));
1543 }
1544
1545 static struct lmv_tgt_desc *
1546 lmv_locate_tgt_by_name(struct lmv_obd *lmv, struct lmv_stripe_md *lsm,
1547                        const char *name, int namelen, struct lu_fid *fid,
1548                        __u32 *mds, bool new_layout)
1549 {
1550         struct lmv_tgt_desc *tgt;
1551         const struct lmv_oinfo *oinfo;
1552
1553         if (!lmv_dir_striped(lsm) || !namelen) {
1554                 tgt = lmv_fid2tgt(lmv, fid);
1555                 if (IS_ERR(tgt))
1556                         return tgt;
1557
1558                 *mds = tgt->ltd_index;
1559                 return tgt;
1560         }
1561
1562         if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_NAME_HASH)) {
1563                 if (cfs_fail_val >= lsm->lsm_md_stripe_count)
1564                         return ERR_PTR(-EBADF);
1565                 oinfo = &lsm->lsm_md_oinfo[cfs_fail_val];
1566         } else {
1567                 oinfo = lsm_name_to_stripe_info(lsm, name, namelen, new_layout);
1568                 if (IS_ERR(oinfo))
1569                         return ERR_CAST(oinfo);
1570         }
1571
1572         *fid = oinfo->lmo_fid;
1573         *mds = oinfo->lmo_mds;
1574         tgt = lmv_tgt(lmv, oinfo->lmo_mds);
1575
1576         CDEBUG(D_INODE, "locate MDT %u parent "DFID"\n", *mds, PFID(fid));
1577
1578         return tgt ? tgt : ERR_PTR(-ENODEV);
1579 }
1580
1581 /**
1582  * Locate MDT of op_data->op_fid1
1583  *
1584  * For striped directory, it will locate the stripe by name hash, if hash_type
1585  * is unknown, it will return the stripe specified by 'op_data->op_stripe_index'
1586  * which is set outside, and if dir is migrating, 'op_data->op_new_layout'
1587  * indicates whether old or new layout is used to locate.
1588  *
1589  * For plain direcotry, it just locate the MDT of op_data->op_fid1.
1590  *
1591  * \param[in] lmv       LMV device
1592  * \param[in] op_data   client MD stack parameters, name, namelen
1593  *                      mds_num etc.
1594  *
1595  * retval               pointer to the lmv_tgt_desc if succeed.
1596  *                      ERR_PTR(errno) if failed.
1597  */
1598 struct lmv_tgt_desc *
1599 lmv_locate_tgt(struct lmv_obd *lmv, struct md_op_data *op_data)
1600 {
1601         struct lmv_stripe_md *lsm = op_data->op_mea1;
1602         struct lmv_oinfo *oinfo;
1603         struct lmv_tgt_desc *tgt;
1604
1605         if (lmv_dir_foreign(lsm))
1606                 return ERR_PTR(-ENODATA);
1607
1608         /* During creating VOLATILE file, it should honor the mdt
1609          * index if the file under striped dir is being restored, see
1610          * ct_restore(). */
1611         if (op_data->op_bias & MDS_CREATE_VOLATILE &&
1612             op_data->op_mds != LMV_OFFSET_DEFAULT) {
1613                 tgt = lmv_tgt(lmv, op_data->op_mds);
1614                 if (!tgt)
1615                         return ERR_PTR(-ENODEV);
1616
1617                 if (lmv_dir_striped(lsm)) {
1618                         int i;
1619
1620                         /* refill the right parent fid */
1621                         for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1622                                 oinfo = &lsm->lsm_md_oinfo[i];
1623                                 if (oinfo->lmo_mds == op_data->op_mds) {
1624                                         op_data->op_fid1 = oinfo->lmo_fid;
1625                                         break;
1626                                 }
1627                         }
1628
1629                         if (i == lsm->lsm_md_stripe_count)
1630                                 op_data->op_fid1 = lsm->lsm_md_oinfo[0].lmo_fid;
1631                 }
1632         } else if (lmv_dir_bad_hash(lsm)) {
1633                 LASSERT(op_data->op_stripe_index < lsm->lsm_md_stripe_count);
1634                 oinfo = &lsm->lsm_md_oinfo[op_data->op_stripe_index];
1635
1636                 op_data->op_fid1 = oinfo->lmo_fid;
1637                 op_data->op_mds = oinfo->lmo_mds;
1638                 tgt = lmv_tgt(lmv, oinfo->lmo_mds);
1639                 if (!tgt)
1640                         return ERR_PTR(-ENODEV);
1641         } else {
1642                 tgt = lmv_locate_tgt_by_name(lmv, op_data->op_mea1,
1643                                 op_data->op_name, op_data->op_namelen,
1644                                 &op_data->op_fid1, &op_data->op_mds,
1645                                 op_data->op_new_layout);
1646         }
1647
1648         return tgt;
1649 }
1650
1651 /* Locate MDT of op_data->op_fid2 for link/rename */
1652 static struct lmv_tgt_desc *
1653 lmv_locate_tgt2(struct lmv_obd *lmv, struct md_op_data *op_data)
1654 {
1655         struct lmv_tgt_desc *tgt;
1656         int rc;
1657
1658         LASSERT(op_data->op_name);
1659         if (lmv_dir_layout_changing(op_data->op_mea2)) {
1660                 struct lu_fid fid1 = op_data->op_fid1;
1661                 struct lmv_stripe_md *lsm1 = op_data->op_mea1;
1662                 struct ptlrpc_request *request = NULL;
1663
1664                 /*
1665                  * avoid creating new file under old layout of migrating
1666                  * directory, check it here.
1667                  */
1668                 tgt = lmv_locate_tgt_by_name(lmv, op_data->op_mea2,
1669                                 op_data->op_name, op_data->op_namelen,
1670                                 &op_data->op_fid2, &op_data->op_mds, false);
1671                 if (IS_ERR(tgt))
1672                         RETURN(tgt);
1673
1674                 op_data->op_fid1 = op_data->op_fid2;
1675                 op_data->op_mea1 = op_data->op_mea2;
1676                 rc = md_getattr_name(tgt->ltd_exp, op_data, &request);
1677                 op_data->op_fid1 = fid1;
1678                 op_data->op_mea1 = lsm1;
1679                 if (!rc) {
1680                         ptlrpc_req_finished(request);
1681                         RETURN(ERR_PTR(-EEXIST));
1682                 }
1683
1684                 if (rc != -ENOENT)
1685                         RETURN(ERR_PTR(rc));
1686         }
1687
1688         return lmv_locate_tgt_by_name(lmv, op_data->op_mea2, op_data->op_name,
1689                                 op_data->op_namelen, &op_data->op_fid2,
1690                                 &op_data->op_mds, true);
1691 }
1692
1693 int lmv_old_layout_lookup(struct lmv_obd *lmv, struct md_op_data *op_data)
1694 {
1695         struct lu_tgt_desc *tgt;
1696         struct ptlrpc_request *request;
1697         int rc;
1698
1699         LASSERT(lmv_dir_layout_changing(op_data->op_mea1));
1700         LASSERT(!op_data->op_new_layout);
1701
1702         tgt = lmv_locate_tgt(lmv, op_data);
1703         if (IS_ERR(tgt))
1704                 return PTR_ERR(tgt);
1705
1706         rc = md_getattr_name(tgt->ltd_exp, op_data, &request);
1707         if (!rc) {
1708                 ptlrpc_req_finished(request);
1709                 return -EEXIST;
1710         }
1711
1712         return rc;
1713 }
1714
1715 /* mkdir by QoS in two cases:
1716  * 1. 'lfs mkdir -i -1'
1717  * 2. parent default LMV master_mdt_index is -1
1718  *
1719  * NB, mkdir by QoS only if parent is not striped, this is to avoid remote
1720  * directories under striped directory.
1721  */
1722 static inline bool lmv_op_qos_mkdir(const struct md_op_data *op_data)
1723 {
1724         const struct lmv_stripe_md *lsm = op_data->op_default_mea1;
1725         const struct lmv_user_md *lum = op_data->op_data;
1726
1727         if (op_data->op_code != LUSTRE_OPC_MKDIR)
1728                 return false;
1729
1730         if (lmv_dir_striped(op_data->op_mea1))
1731                 return false;
1732
1733         if (op_data->op_cli_flags & CLI_SET_MEA && lum &&
1734             (le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC ||
1735              le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC_SPECIFIC) &&
1736             le32_to_cpu(lum->lum_stripe_offset) == LMV_OFFSET_DEFAULT)
1737                 return true;
1738
1739         if (lsm && lsm->lsm_md_master_mdt_index == LMV_OFFSET_DEFAULT)
1740                 return true;
1741
1742         return false;
1743 }
1744
1745 /* 'lfs mkdir -i <specific_MDT>' */
1746 static inline bool lmv_op_user_specific_mkdir(const struct md_op_data *op_data)
1747 {
1748         const struct lmv_user_md *lum = op_data->op_data;
1749
1750         return op_data->op_code == LUSTRE_OPC_MKDIR &&
1751                op_data->op_cli_flags & CLI_SET_MEA && lum &&
1752                (le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC ||
1753                 le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC_SPECIFIC) &&
1754                le32_to_cpu(lum->lum_stripe_offset) != LMV_OFFSET_DEFAULT;
1755 }
1756
1757 /* parent default LMV master_mdt_index is not -1. */
1758 static inline bool
1759 lmv_op_default_specific_mkdir(const struct md_op_data *op_data)
1760 {
1761         return op_data->op_code == LUSTRE_OPC_MKDIR &&
1762                op_data->op_default_mea1 &&
1763                op_data->op_default_mea1->lsm_md_master_mdt_index !=
1764                         LMV_OFFSET_DEFAULT;
1765 }
1766 int lmv_create(struct obd_export *exp, struct md_op_data *op_data,
1767                 const void *data, size_t datalen, umode_t mode, uid_t uid,
1768                 gid_t gid, cfs_cap_t cap_effective, __u64 rdev,
1769                 struct ptlrpc_request **request)
1770 {
1771         struct obd_device *obd = exp->exp_obd;
1772         struct lmv_obd *lmv = &obd->u.lmv;
1773         struct lmv_tgt_desc *tgt;
1774         struct mdt_body *repbody;
1775         int rc;
1776
1777         ENTRY;
1778
1779         if (!lmv->lmv_mdt_descs.ltd_lmv_desc.ld_active_tgt_count)
1780                 RETURN(-EIO);
1781
1782         if (lmv_dir_bad_hash(op_data->op_mea1))
1783                 RETURN(-EBADF);
1784
1785         if (lmv_dir_layout_changing(op_data->op_mea1)) {
1786                 /*
1787                  * if parent is migrating, create() needs to lookup existing
1788                  * name in both old and new layout, check old layout on client.
1789                  */
1790                 rc = lmv_old_layout_lookup(lmv, op_data);
1791                 if (rc != -ENOENT)
1792                         RETURN(rc);
1793
1794                 op_data->op_new_layout = true;
1795         }
1796
1797         tgt = lmv_locate_tgt(lmv, op_data);
1798         if (IS_ERR(tgt))
1799                 RETURN(PTR_ERR(tgt));
1800
1801         if (lmv_op_user_specific_mkdir(op_data)) {
1802                 struct lmv_user_md *lum = op_data->op_data;
1803
1804                 op_data->op_mds = le32_to_cpu(lum->lum_stripe_offset);
1805                 tgt = lmv_tgt(lmv, op_data->op_mds);
1806                 if (!tgt)
1807                         RETURN(-ENODEV);
1808         } else if (lmv_op_default_specific_mkdir(op_data)) {
1809                 op_data->op_mds =
1810                         op_data->op_default_mea1->lsm_md_master_mdt_index;
1811                 tgt = lmv_tgt(lmv, op_data->op_mds);
1812                 if (!tgt)
1813                         RETURN(-ENODEV);
1814         } else if (lmv_op_qos_mkdir(op_data)) {
1815                 tgt = lmv_locate_tgt_qos(lmv, &op_data->op_mds);
1816                 if (tgt == ERR_PTR(-EAGAIN))
1817                         tgt = lmv_locate_tgt_rr(lmv, &op_data->op_mds);
1818                 if (IS_ERR(tgt))
1819                         RETURN(PTR_ERR(tgt));
1820                 /*
1821                  * only update statfs after QoS mkdir, this means the cached
1822                  * statfs may be stale, and current mkdir may not follow QoS
1823                  * accurately, but it's not serious, and avoids periodic statfs
1824                  * when client doesn't mkdir by QoS.
1825                  */
1826                 lmv_statfs_check_update(obd, tgt);
1827         }
1828
1829 retry:
1830         rc = lmv_fid_alloc(NULL, exp, &op_data->op_fid2, op_data);
1831         if (rc)
1832                 RETURN(rc);
1833
1834         CDEBUG(D_INODE, "CREATE name '%.*s' "DFID" on "DFID" -> mds #%x\n",
1835                 (int)op_data->op_namelen, op_data->op_name,
1836                 PFID(&op_data->op_fid2), PFID(&op_data->op_fid1),
1837                 op_data->op_mds);
1838
1839         op_data->op_flags |= MF_MDC_CANCEL_FID1;
1840         rc = md_create(tgt->ltd_exp, op_data, data, datalen, mode, uid, gid,
1841                        cap_effective, rdev, request);
1842         if (rc == 0) {
1843                 if (*request == NULL)
1844                         RETURN(rc);
1845                 CDEBUG(D_INODE, "Created - "DFID"\n", PFID(&op_data->op_fid2));
1846         }
1847
1848         /* dir restripe needs to send to MDT where dir is located */
1849         if (rc != -EREMOTE ||
1850             !(exp_connect_flags2(exp) & OBD_CONNECT2_CRUSH))
1851                 RETURN(rc);
1852
1853         repbody = req_capsule_server_get(&(*request)->rq_pill, &RMF_MDT_BODY);
1854         if (repbody == NULL)
1855                 RETURN(-EPROTO);
1856
1857         /* Not cross-ref case, just get out of here. */
1858         if (likely(!(repbody->mbo_valid & OBD_MD_MDS)))
1859                 RETURN(rc);
1860
1861         op_data->op_fid2 = repbody->mbo_fid1;
1862         ptlrpc_req_finished(*request);
1863         *request = NULL;
1864
1865         tgt = lmv_fid2tgt(lmv, &op_data->op_fid2);
1866         if (IS_ERR(tgt))
1867                 RETURN(PTR_ERR(tgt));
1868
1869         op_data->op_mds = tgt->ltd_index;
1870         goto retry;
1871 }
1872
1873 static int
1874 lmv_enqueue(struct obd_export *exp, struct ldlm_enqueue_info *einfo,
1875             const union ldlm_policy_data *policy, struct md_op_data *op_data,
1876             struct lustre_handle *lockh, __u64 extra_lock_flags)
1877 {
1878         struct obd_device *obd = exp->exp_obd;
1879         struct lmv_obd *lmv = &obd->u.lmv;
1880         struct lmv_tgt_desc *tgt;
1881         int rc;
1882
1883         ENTRY;
1884
1885         CDEBUG(D_INODE, "ENQUEUE on "DFID"\n", PFID(&op_data->op_fid1));
1886
1887         tgt = lmv_fid2tgt(lmv, &op_data->op_fid1);
1888         if (IS_ERR(tgt))
1889                 RETURN(PTR_ERR(tgt));
1890
1891         CDEBUG(D_INODE, "ENQUEUE on "DFID" -> mds #%u\n",
1892                PFID(&op_data->op_fid1), tgt->ltd_index);
1893
1894         rc = md_enqueue(tgt->ltd_exp, einfo, policy, op_data, lockh,
1895                         extra_lock_flags);
1896
1897         RETURN(rc);
1898 }
1899
1900 int
1901 lmv_getattr_name(struct obd_export *exp,struct md_op_data *op_data,
1902                  struct ptlrpc_request **preq)
1903 {
1904         struct obd_device *obd = exp->exp_obd;
1905         struct lmv_obd *lmv = &obd->u.lmv;
1906         struct lmv_tgt_desc *tgt;
1907         struct mdt_body *body;
1908         int rc;
1909
1910         ENTRY;
1911
1912 retry:
1913         tgt = lmv_locate_tgt(lmv, op_data);
1914         if (IS_ERR(tgt))
1915                 RETURN(PTR_ERR(tgt));
1916
1917         CDEBUG(D_INODE, "GETATTR_NAME for %*s on "DFID" -> mds #%d\n",
1918                 (int)op_data->op_namelen, op_data->op_name,
1919                 PFID(&op_data->op_fid1), tgt->ltd_index);
1920
1921         rc = md_getattr_name(tgt->ltd_exp, op_data, preq);
1922         if (rc == -ENOENT && lmv_dir_retry_check_update(op_data)) {
1923                 ptlrpc_req_finished(*preq);
1924                 *preq = NULL;
1925                 goto retry;
1926         }
1927
1928         if (rc)
1929                 RETURN(rc);
1930
1931         body = req_capsule_server_get(&(*preq)->rq_pill, &RMF_MDT_BODY);
1932         LASSERT(body != NULL);
1933
1934         if (body->mbo_valid & OBD_MD_MDS) {
1935                 op_data->op_fid1 = body->mbo_fid1;
1936                 op_data->op_valid |= OBD_MD_FLCROSSREF;
1937                 op_data->op_namelen = 0;
1938                 op_data->op_name = NULL;
1939
1940                 ptlrpc_req_finished(*preq);
1941                 *preq = NULL;
1942
1943                 goto retry;
1944         }
1945
1946         RETURN(rc);
1947 }
1948
1949 #define md_op_data_fid(op_data, fl)                     \
1950         (fl == MF_MDC_CANCEL_FID1 ? &op_data->op_fid1 : \
1951          fl == MF_MDC_CANCEL_FID2 ? &op_data->op_fid2 : \
1952          fl == MF_MDC_CANCEL_FID3 ? &op_data->op_fid3 : \
1953          fl == MF_MDC_CANCEL_FID4 ? &op_data->op_fid4 : \
1954          NULL)
1955
1956 static int lmv_early_cancel(struct obd_export *exp, struct lmv_tgt_desc *tgt,
1957                             struct md_op_data *op_data, __u32 op_tgt,
1958                             enum ldlm_mode mode, int bits, int flag)
1959 {
1960         struct lu_fid *fid = md_op_data_fid(op_data, flag);
1961         struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
1962         union ldlm_policy_data policy = { { 0 } };
1963         int rc = 0;
1964         ENTRY;
1965
1966         if (!fid_is_sane(fid))
1967                 RETURN(0);
1968
1969         if (tgt == NULL) {
1970                 tgt = lmv_fid2tgt(lmv, fid);
1971                 if (IS_ERR(tgt))
1972                         RETURN(PTR_ERR(tgt));
1973         }
1974
1975         if (tgt->ltd_index != op_tgt) {
1976                 CDEBUG(D_INODE, "EARLY_CANCEL on "DFID"\n", PFID(fid));
1977                 policy.l_inodebits.bits = bits;
1978                 rc = md_cancel_unused(tgt->ltd_exp, fid, &policy,
1979                                       mode, LCF_ASYNC, NULL);
1980         } else {
1981                 CDEBUG(D_INODE,
1982                        "EARLY_CANCEL skip operation target %d on "DFID"\n",
1983                        op_tgt, PFID(fid));
1984                 op_data->op_flags |= flag;
1985                 rc = 0;
1986         }
1987
1988         RETURN(rc);
1989 }
1990
1991 /*
1992  * llite passes fid of an target inode in op_data->op_fid1 and id of directory in
1993  * op_data->op_fid2
1994  */
1995 static int lmv_link(struct obd_export *exp, struct md_op_data *op_data,
1996                     struct ptlrpc_request **request)
1997 {
1998         struct obd_device       *obd = exp->exp_obd;
1999         struct lmv_obd          *lmv = &obd->u.lmv;
2000         struct lmv_tgt_desc     *tgt;
2001         int                      rc;
2002         ENTRY;
2003
2004         LASSERT(op_data->op_namelen != 0);
2005
2006         CDEBUG(D_INODE, "LINK "DFID":%*s to "DFID"\n",
2007                PFID(&op_data->op_fid2), (int)op_data->op_namelen,
2008                op_data->op_name, PFID(&op_data->op_fid1));
2009
2010         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2011         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2012         op_data->op_cap = cfs_curproc_cap_pack();
2013
2014         tgt = lmv_locate_tgt2(lmv, op_data);
2015         if (IS_ERR(tgt))
2016                 RETURN(PTR_ERR(tgt));
2017
2018         /*
2019          * Cancel UPDATE lock on child (fid1).
2020          */
2021         op_data->op_flags |= MF_MDC_CANCEL_FID2;
2022         rc = lmv_early_cancel(exp, NULL, op_data, tgt->ltd_index, LCK_EX,
2023                               MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID1);
2024         if (rc != 0)
2025                 RETURN(rc);
2026
2027         rc = md_link(tgt->ltd_exp, op_data, request);
2028
2029         RETURN(rc);
2030 }
2031
2032 static int lmv_migrate(struct obd_export *exp, struct md_op_data *op_data,
2033                         const char *name, size_t namelen,
2034                         struct ptlrpc_request **request)
2035 {
2036         struct obd_device *obd = exp->exp_obd;
2037         struct lmv_obd *lmv = &obd->u.lmv;
2038         struct lmv_stripe_md *lsm = op_data->op_mea1;
2039         struct lmv_tgt_desc *parent_tgt;
2040         struct lmv_tgt_desc *sp_tgt;
2041         struct lmv_tgt_desc *tp_tgt = NULL;
2042         struct lmv_tgt_desc *child_tgt;
2043         struct lmv_tgt_desc *tgt;
2044         struct lu_fid target_fid = { 0 };
2045         int rc;
2046
2047         ENTRY;
2048
2049         LASSERT(op_data->op_cli_flags & CLI_MIGRATE);
2050
2051         CDEBUG(D_INODE, "MIGRATE "DFID"/%.*s\n",
2052                PFID(&op_data->op_fid1), (int)namelen, name);
2053
2054         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2055         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2056         op_data->op_cap = cfs_curproc_cap_pack();
2057
2058         parent_tgt = lmv_fid2tgt(lmv, &op_data->op_fid1);
2059         if (IS_ERR(parent_tgt))
2060                 RETURN(PTR_ERR(parent_tgt));
2061
2062         if (lmv_dir_striped(lsm)) {
2063                 const struct lmv_oinfo *oinfo;
2064
2065                 oinfo = lsm_name_to_stripe_info(lsm, name, namelen, false);
2066                 if (IS_ERR(oinfo))
2067                         RETURN(PTR_ERR(oinfo));
2068
2069                 /* save source stripe FID in fid4 temporarily for ELC */
2070                 op_data->op_fid4 = oinfo->lmo_fid;
2071                 sp_tgt = lmv_tgt(lmv, oinfo->lmo_mds);
2072                 if (!sp_tgt)
2073                         RETURN(-ENODEV);
2074
2075                 /*
2076                  * if parent is being migrated too, fill op_fid2 with target
2077                  * stripe fid, otherwise the target stripe is not created yet.
2078                  */
2079                 if (lmv_dir_layout_changing(lsm)) {
2080                         oinfo = lsm_name_to_stripe_info(lsm, name, namelen,
2081                                                         true);
2082                         if (IS_ERR(oinfo))
2083                                 RETURN(PTR_ERR(oinfo));
2084
2085                         op_data->op_fid2 = oinfo->lmo_fid;
2086                         tp_tgt = lmv_tgt(lmv, oinfo->lmo_mds);
2087                         if (!tp_tgt)
2088                                 RETURN(-ENODEV);
2089                 }
2090         } else {
2091                 sp_tgt = parent_tgt;
2092         }
2093
2094         child_tgt = lmv_fid2tgt(lmv, &op_data->op_fid3);
2095         if (IS_ERR(child_tgt))
2096                 RETURN(PTR_ERR(child_tgt));
2097
2098         /* for directory, migrate to MDT specified by lum_stripe_offset;
2099          * otherwise migrate to the target stripe of parent, but parent
2100          * directory may have finished migration (normally current file too),
2101          * allocate FID on MDT lum_stripe_offset, and server will check
2102          * whether file was migrated already.
2103          */
2104         if (S_ISDIR(op_data->op_mode) || !tp_tgt) {
2105                 struct lmv_user_md *lum = op_data->op_data;
2106
2107                 op_data->op_mds = le32_to_cpu(lum->lum_stripe_offset);
2108         } else  {
2109                 op_data->op_mds = tp_tgt->ltd_index;
2110         }
2111         rc = lmv_fid_alloc(NULL, exp, &target_fid, op_data);
2112         if (rc)
2113                 RETURN(rc);
2114
2115         /*
2116          * for directory, send migrate request to the MDT where the object will
2117          * be migrated to, because we can't create a striped directory remotely.
2118          *
2119          * otherwise, send to the MDT where source is located because regular
2120          * file may open lease.
2121          *
2122          * NB. if MDT doesn't support DIR_MIGRATE, send to source MDT too for
2123          * backward compatibility.
2124          */
2125         if (S_ISDIR(op_data->op_mode) &&
2126             (exp_connect_flags2(exp) & OBD_CONNECT2_DIR_MIGRATE)) {
2127                 tgt = lmv_fid2tgt(lmv, &target_fid);
2128                 if (IS_ERR(tgt))
2129                         RETURN(PTR_ERR(tgt));
2130         } else {
2131                 tgt = child_tgt;
2132         }
2133
2134         /* cancel UPDATE lock of parent master object */
2135         rc = lmv_early_cancel(exp, parent_tgt, op_data, tgt->ltd_index, LCK_EX,
2136                               MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID1);
2137         if (rc)
2138                 RETURN(rc);
2139
2140         /* cancel UPDATE lock of source parent */
2141         if (sp_tgt != parent_tgt) {
2142                 /*
2143                  * migrate RPC packs master object FID, because we can only pack
2144                  * two FIDs in reint RPC, but MDS needs to know both source
2145                  * parent and target parent, and it will obtain them from master
2146                  * FID and LMV, the other FID in RPC is kept for target.
2147                  *
2148                  * since this FID is not passed to MDC, cancel it anyway.
2149                  */
2150                 rc = lmv_early_cancel(exp, sp_tgt, op_data, -1, LCK_EX,
2151                                       MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID4);
2152                 if (rc)
2153                         RETURN(rc);
2154
2155                 op_data->op_flags &= ~MF_MDC_CANCEL_FID4;
2156         }
2157         op_data->op_fid4 = target_fid;
2158
2159         /* cancel UPDATE locks of target parent */
2160         rc = lmv_early_cancel(exp, tp_tgt, op_data, tgt->ltd_index, LCK_EX,
2161                               MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID2);
2162         if (rc)
2163                 RETURN(rc);
2164
2165         /* cancel LOOKUP lock of source if source is remote object */
2166         if (child_tgt != sp_tgt) {
2167                 rc = lmv_early_cancel(exp, sp_tgt, op_data, tgt->ltd_index,
2168                                       LCK_EX, MDS_INODELOCK_LOOKUP,
2169                                       MF_MDC_CANCEL_FID3);
2170                 if (rc)
2171                         RETURN(rc);
2172         }
2173
2174         /* cancel ELC locks of source */
2175         rc = lmv_early_cancel(exp, child_tgt, op_data, tgt->ltd_index, LCK_EX,
2176                               MDS_INODELOCK_ELC, MF_MDC_CANCEL_FID3);
2177         if (rc)
2178                 RETURN(rc);
2179
2180         rc = md_rename(tgt->ltd_exp, op_data, name, namelen, NULL, 0, request);
2181
2182         RETURN(rc);
2183 }
2184
2185 static int lmv_rename(struct obd_export *exp, struct md_op_data *op_data,
2186                       const char *old, size_t oldlen,
2187                       const char *new, size_t newlen,
2188                       struct ptlrpc_request **request)
2189 {
2190         struct obd_device *obd = exp->exp_obd;
2191         struct lmv_obd *lmv = &obd->u.lmv;
2192         struct lmv_tgt_desc *sp_tgt;
2193         struct lmv_tgt_desc *tp_tgt = NULL;
2194         struct lmv_tgt_desc *src_tgt = NULL;
2195         struct lmv_tgt_desc *tgt;
2196         struct mdt_body *body;
2197         int rc;
2198
2199         ENTRY;
2200
2201         LASSERT(oldlen != 0);
2202
2203         if (op_data->op_cli_flags & CLI_MIGRATE) {
2204                 rc = lmv_migrate(exp, op_data, old, oldlen, request);
2205                 RETURN(rc);
2206         }
2207
2208         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2209         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2210         op_data->op_cap = cfs_curproc_cap_pack();
2211
2212         op_data->op_name = new;
2213         op_data->op_namelen = newlen;
2214
2215         tp_tgt = lmv_locate_tgt2(lmv, op_data);
2216         if (IS_ERR(tp_tgt))
2217                 RETURN(PTR_ERR(tp_tgt));
2218
2219         /* Since the target child might be destroyed, and it might become
2220          * orphan, and we can only check orphan on the local MDT right now, so
2221          * we send rename request to the MDT where target child is located. If
2222          * target child does not exist, then it will send the request to the
2223          * target parent */
2224         if (fid_is_sane(&op_data->op_fid4)) {
2225                 tgt = lmv_fid2tgt(lmv, &op_data->op_fid4);
2226                 if (IS_ERR(tgt))
2227                         RETURN(PTR_ERR(tgt));
2228         } else {
2229                 tgt = tp_tgt;
2230         }
2231
2232         op_data->op_flags |= MF_MDC_CANCEL_FID4;
2233
2234         /* cancel UPDATE locks of target parent */
2235         rc = lmv_early_cancel(exp, tp_tgt, op_data, tgt->ltd_index, LCK_EX,
2236                               MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID2);
2237         if (rc != 0)
2238                 RETURN(rc);
2239
2240         if (fid_is_sane(&op_data->op_fid4)) {
2241                 /* cancel LOOKUP lock of target on target parent */
2242                 if (tgt != tp_tgt) {
2243                         rc = lmv_early_cancel(exp, tp_tgt, op_data,
2244                                               tgt->ltd_index, LCK_EX,
2245                                               MDS_INODELOCK_LOOKUP,
2246                                               MF_MDC_CANCEL_FID4);
2247                         if (rc != 0)
2248                                 RETURN(rc);
2249                 }
2250         }
2251
2252         if (fid_is_sane(&op_data->op_fid3)) {
2253                 src_tgt = lmv_fid2tgt(lmv, &op_data->op_fid3);
2254                 if (IS_ERR(src_tgt))
2255                         RETURN(PTR_ERR(src_tgt));
2256
2257                 /* cancel ELC locks of source */
2258                 rc = lmv_early_cancel(exp, src_tgt, op_data, tgt->ltd_index,
2259                                       LCK_EX, MDS_INODELOCK_ELC,
2260                                       MF_MDC_CANCEL_FID3);
2261                 if (rc != 0)
2262                         RETURN(rc);
2263         }
2264
2265         op_data->op_name = old;
2266         op_data->op_namelen = oldlen;
2267 retry:
2268         sp_tgt = lmv_locate_tgt(lmv, op_data);
2269         if (IS_ERR(sp_tgt))
2270                 RETURN(PTR_ERR(sp_tgt));
2271
2272         /* cancel UPDATE locks of source parent */
2273         rc = lmv_early_cancel(exp, sp_tgt, op_data, tgt->ltd_index, LCK_EX,
2274                               MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID1);
2275         if (rc != 0)
2276                 RETURN(rc);
2277
2278         if (fid_is_sane(&op_data->op_fid3)) {
2279                 /* cancel LOOKUP lock of source on source parent */
2280                 if (src_tgt != sp_tgt) {
2281                         rc = lmv_early_cancel(exp, sp_tgt, op_data,
2282                                               tgt->ltd_index, LCK_EX,
2283                                               MDS_INODELOCK_LOOKUP,
2284                                               MF_MDC_CANCEL_FID3);
2285                         if (rc != 0)
2286                                 RETURN(rc);
2287                 }
2288         }
2289
2290 rename:
2291         CDEBUG(D_INODE, "RENAME "DFID"/%.*s to "DFID"/%.*s\n",
2292                 PFID(&op_data->op_fid1), (int)oldlen, old,
2293                 PFID(&op_data->op_fid2), (int)newlen, new);
2294
2295         rc = md_rename(tgt->ltd_exp, op_data, old, oldlen, new, newlen,
2296                         request);
2297         if (rc == -ENOENT && lmv_dir_retry_check_update(op_data)) {
2298                 ptlrpc_req_finished(*request);
2299                 *request = NULL;
2300                 goto retry;
2301         }
2302
2303         if (rc && rc != -EXDEV)
2304                 RETURN(rc);
2305
2306         body = req_capsule_server_get(&(*request)->rq_pill, &RMF_MDT_BODY);
2307         if (body == NULL)
2308                 RETURN(-EPROTO);
2309
2310         /* Not cross-ref case, just get out of here. */
2311         if (likely(!(body->mbo_valid & OBD_MD_MDS)))
2312                 RETURN(rc);
2313
2314         op_data->op_fid4 = body->mbo_fid1;
2315
2316         ptlrpc_req_finished(*request);
2317         *request = NULL;
2318
2319         tgt = lmv_fid2tgt(lmv, &op_data->op_fid4);
2320         if (IS_ERR(tgt))
2321                 RETURN(PTR_ERR(tgt));
2322
2323         if (fid_is_sane(&op_data->op_fid4)) {
2324                 /* cancel LOOKUP lock of target on target parent */
2325                 if (tgt != tp_tgt) {
2326                         rc = lmv_early_cancel(exp, tp_tgt, op_data,
2327                                               tgt->ltd_index, LCK_EX,
2328                                               MDS_INODELOCK_LOOKUP,
2329                                               MF_MDC_CANCEL_FID4);
2330                         if (rc != 0)
2331                                 RETURN(rc);
2332                 }
2333         }
2334
2335         goto rename;
2336 }
2337
2338 static int lmv_setattr(struct obd_export *exp, struct md_op_data *op_data,
2339                        void *ea, size_t ealen, struct ptlrpc_request **request)
2340 {
2341         struct obd_device *obd = exp->exp_obd;
2342         struct lmv_obd *lmv = &obd->u.lmv;
2343         struct lmv_tgt_desc *tgt;
2344         int rc = 0;
2345
2346         ENTRY;
2347
2348         CDEBUG(D_INODE, "SETATTR for "DFID", valid 0x%x/0x%x\n",
2349                PFID(&op_data->op_fid1), op_data->op_attr.ia_valid,
2350                op_data->op_xvalid);
2351
2352         op_data->op_flags |= MF_MDC_CANCEL_FID1;
2353         tgt = lmv_fid2tgt(lmv, &op_data->op_fid1);
2354         if (IS_ERR(tgt))
2355                 RETURN(PTR_ERR(tgt));
2356
2357         rc = md_setattr(tgt->ltd_exp, op_data, ea, ealen, request);
2358
2359         RETURN(rc);
2360 }
2361
2362 static int lmv_fsync(struct obd_export *exp, const struct lu_fid *fid,
2363                      struct ptlrpc_request **request)
2364 {
2365         struct obd_device *obd = exp->exp_obd;
2366         struct lmv_obd *lmv = &obd->u.lmv;
2367         struct lmv_tgt_desc *tgt;
2368         int rc;
2369
2370         ENTRY;
2371
2372         tgt = lmv_fid2tgt(lmv, fid);
2373         if (IS_ERR(tgt))
2374                 RETURN(PTR_ERR(tgt));
2375
2376         rc = md_fsync(tgt->ltd_exp, fid, request);
2377         RETURN(rc);
2378 }
2379
2380 struct stripe_dirent {
2381         struct page             *sd_page;
2382         struct lu_dirpage       *sd_dp;
2383         struct lu_dirent        *sd_ent;
2384         bool                     sd_eof;
2385 };
2386
2387 struct lmv_dir_ctxt {
2388         struct lmv_obd          *ldc_lmv;
2389         struct md_op_data       *ldc_op_data;
2390         struct md_callback      *ldc_cb_op;
2391         __u64                    ldc_hash;
2392         int                      ldc_count;
2393         struct stripe_dirent     ldc_stripes[0];
2394 };
2395
2396 static inline void stripe_dirent_unload(struct stripe_dirent *stripe)
2397 {
2398         if (stripe->sd_page) {
2399                 kunmap(stripe->sd_page);
2400                 put_page(stripe->sd_page);
2401                 stripe->sd_page = NULL;
2402                 stripe->sd_ent = NULL;
2403         }
2404 }
2405
2406 static inline void put_lmv_dir_ctxt(struct lmv_dir_ctxt *ctxt)
2407 {
2408         int i;
2409
2410         for (i = 0; i < ctxt->ldc_count; i++)
2411                 stripe_dirent_unload(&ctxt->ldc_stripes[i]);
2412 }
2413
2414 /* if @ent is dummy, or . .., get next */
2415 static struct lu_dirent *stripe_dirent_get(struct lmv_dir_ctxt *ctxt,
2416                                            struct lu_dirent *ent,
2417                                            int stripe_index)
2418 {
2419         for (; ent; ent = lu_dirent_next(ent)) {
2420                 /* Skip dummy entry */
2421                 if (le16_to_cpu(ent->lde_namelen) == 0)
2422                         continue;
2423
2424                 /* skip . and .. for other stripes */
2425                 if (stripe_index &&
2426                     (strncmp(ent->lde_name, ".",
2427                              le16_to_cpu(ent->lde_namelen)) == 0 ||
2428                      strncmp(ent->lde_name, "..",
2429                              le16_to_cpu(ent->lde_namelen)) == 0))
2430                         continue;
2431
2432                 if (le64_to_cpu(ent->lde_hash) >= ctxt->ldc_hash)
2433                         break;
2434         }
2435
2436         return ent;
2437 }
2438
2439 static struct lu_dirent *stripe_dirent_load(struct lmv_dir_ctxt *ctxt,
2440                                             struct stripe_dirent *stripe,
2441                                             int stripe_index)
2442 {
2443         struct md_op_data *op_data = ctxt->ldc_op_data;
2444         struct lmv_oinfo *oinfo;
2445         struct lu_fid fid = op_data->op_fid1;
2446         struct inode *inode = op_data->op_data;
2447         struct lmv_tgt_desc *tgt;
2448         struct lu_dirent *ent = stripe->sd_ent;
2449         __u64 hash = ctxt->ldc_hash;
2450         int rc = 0;
2451
2452         ENTRY;
2453
2454         LASSERT(stripe == &ctxt->ldc_stripes[stripe_index]);
2455         LASSERT(!ent);
2456
2457         do {
2458                 if (stripe->sd_page) {
2459                         __u64 end = le64_to_cpu(stripe->sd_dp->ldp_hash_end);
2460
2461                         /* @hash should be the last dirent hash */
2462                         LASSERTF(hash <= end,
2463                                  "ctxt@%p stripe@%p hash %llx end %llx\n",
2464                                  ctxt, stripe, hash, end);
2465                         /* unload last page */
2466                         stripe_dirent_unload(stripe);
2467                         /* eof */
2468                         if (end == MDS_DIR_END_OFF) {
2469                                 stripe->sd_eof = true;
2470                                 break;
2471                         }
2472                         hash = end;
2473                 }
2474
2475                 oinfo = &op_data->op_mea1->lsm_md_oinfo[stripe_index];
2476                 if (!oinfo->lmo_root) {
2477                         rc = -ENOENT;
2478                         break;
2479                 }
2480
2481                 tgt = lmv_tgt(ctxt->ldc_lmv, oinfo->lmo_mds);
2482                 if (!tgt) {
2483                         rc = -ENODEV;
2484                         break;
2485                 }
2486
2487                 /* op_data is shared by stripes, reset after use */
2488                 op_data->op_fid1 = oinfo->lmo_fid;
2489                 op_data->op_fid2 = oinfo->lmo_fid;
2490                 op_data->op_data = oinfo->lmo_root;
2491
2492                 rc = md_read_page(tgt->ltd_exp, op_data, ctxt->ldc_cb_op, hash,
2493                                   &stripe->sd_page);
2494
2495                 op_data->op_fid1 = fid;
2496                 op_data->op_fid2 = fid;
2497                 op_data->op_data = inode;
2498
2499                 if (rc)
2500                         break;
2501
2502                 stripe->sd_dp = page_address(stripe->sd_page);
2503                 ent = stripe_dirent_get(ctxt, lu_dirent_start(stripe->sd_dp),
2504                                         stripe_index);
2505                 /* in case a page filled with ., .. and dummy, read next */
2506         } while (!ent);
2507
2508         stripe->sd_ent = ent;
2509         if (rc) {
2510                 LASSERT(!ent);
2511                 /* treat error as eof, so dir can be partially accessed */
2512                 stripe->sd_eof = true;
2513                 LCONSOLE_WARN("dir "DFID" stripe %d readdir failed: %d, "
2514                               "directory is partially accessed!\n",
2515                               PFID(&ctxt->ldc_op_data->op_fid1), stripe_index,
2516                               rc);
2517         }
2518
2519         RETURN(ent);
2520 }
2521
2522 static int lmv_file_resync(struct obd_export *exp, struct md_op_data *data)
2523 {
2524         struct obd_device *obd = exp->exp_obd;
2525         struct lmv_obd *lmv = &obd->u.lmv;
2526         struct lmv_tgt_desc *tgt;
2527         int rc;
2528
2529         ENTRY;
2530
2531         rc = lmv_check_connect(obd);
2532         if (rc != 0)
2533                 RETURN(rc);
2534
2535         tgt = lmv_fid2tgt(lmv, &data->op_fid1);
2536         if (IS_ERR(tgt))
2537                 RETURN(PTR_ERR(tgt));
2538
2539         data->op_flags |= MF_MDC_CANCEL_FID1;
2540         rc = md_file_resync(tgt->ltd_exp, data);
2541         RETURN(rc);
2542 }
2543
2544 /**
2545  * Get dirent with the closest hash for striped directory
2546  *
2547  * This function will search the dir entry, whose hash value is the
2548  * closest(>=) to hash from all of sub-stripes, and it is only being called
2549  * for striped directory.
2550  *
2551  * \param[in] ctxt              dir read context
2552  *
2553  * \retval                      dirent get the entry successfully
2554  *                              NULL does not get the entry, normally it means
2555  *                              it reaches the end of the directory, while read
2556  *                              stripe dirent error is ignored to allow partial
2557  *                              access.
2558  */
2559 static struct lu_dirent *lmv_dirent_next(struct lmv_dir_ctxt *ctxt)
2560 {
2561         struct stripe_dirent *stripe;
2562         struct lu_dirent *ent = NULL;
2563         int i;
2564         int min = -1;
2565
2566         /* TODO: optimize with k-way merge sort */
2567         for (i = 0; i < ctxt->ldc_count; i++) {
2568                 stripe = &ctxt->ldc_stripes[i];
2569                 if (stripe->sd_eof)
2570                         continue;
2571
2572                 if (!stripe->sd_ent) {
2573                         stripe_dirent_load(ctxt, stripe, i);
2574                         if (!stripe->sd_ent) {
2575                                 LASSERT(stripe->sd_eof);
2576                                 continue;
2577                         }
2578                 }
2579
2580                 if (min == -1 ||
2581                     le64_to_cpu(ctxt->ldc_stripes[min].sd_ent->lde_hash) >
2582                     le64_to_cpu(stripe->sd_ent->lde_hash)) {
2583                         min = i;
2584                         if (le64_to_cpu(stripe->sd_ent->lde_hash) ==
2585                             ctxt->ldc_hash)
2586                                 break;
2587                 }
2588         }
2589
2590         if (min != -1) {
2591                 stripe = &ctxt->ldc_stripes[min];
2592                 ent = stripe->sd_ent;
2593                 /* pop found dirent */
2594                 stripe->sd_ent = stripe_dirent_get(ctxt, lu_dirent_next(ent),
2595                                                    min);
2596         }
2597
2598         return ent;
2599 }
2600
2601 /**
2602  * Build dir entry page for striped directory
2603  *
2604  * This function gets one entry by @offset from a striped directory. It will
2605  * read entries from all of stripes, and choose one closest to the required
2606  * offset(&offset). A few notes
2607  * 1. skip . and .. for non-zero stripes, because there can only have one .
2608  * and .. in a directory.
2609  * 2. op_data will be shared by all of stripes, instead of allocating new
2610  * one, so need to restore before reusing.
2611  *
2612  * \param[in] exp       obd export refer to LMV
2613  * \param[in] op_data   hold those MD parameters of read_entry
2614  * \param[in] cb_op     ldlm callback being used in enqueue in mdc_read_entry
2615  * \param[in] offset    starting hash offset
2616  * \param[out] ppage    the page holding the entry. Note: because the entry
2617  *                      will be accessed in upper layer, so we need hold the
2618  *                      page until the usages of entry is finished, see
2619  *                      ll_dir_entry_next.
2620  *
2621  * retval               =0 if get entry successfully
2622  *                      <0 cannot get entry
2623  */
2624 static int lmv_striped_read_page(struct obd_export *exp,
2625                                  struct md_op_data *op_data,
2626                                  struct md_callback *cb_op,
2627                                  __u64 offset, struct page **ppage)
2628 {
2629         struct page *page = NULL;
2630         struct lu_dirpage *dp;
2631         void *start;
2632         struct lu_dirent *ent;
2633         struct lu_dirent *last_ent;
2634         int stripe_count;
2635         struct lmv_dir_ctxt *ctxt;
2636         struct lu_dirent *next = NULL;
2637         __u16 ent_size;
2638         size_t left_bytes;
2639         int rc = 0;
2640         ENTRY;
2641
2642         /* Allocate a page and read entries from all of stripes and fill
2643          * the page by hash order */
2644         page = alloc_page(GFP_KERNEL);
2645         if (!page)
2646                 RETURN(-ENOMEM);
2647
2648         /* Initialize the entry page */
2649         dp = kmap(page);
2650         memset(dp, 0, sizeof(*dp));
2651         dp->ldp_hash_start = cpu_to_le64(offset);
2652
2653         start = dp + 1;
2654         left_bytes = PAGE_SIZE - sizeof(*dp);
2655         ent = start;
2656         last_ent = ent;
2657
2658         /* initalize dir read context */
2659         stripe_count = op_data->op_mea1->lsm_md_stripe_count;
2660         OBD_ALLOC(ctxt, offsetof(typeof(*ctxt), ldc_stripes[stripe_count]));
2661         if (!ctxt)
2662                 GOTO(free_page, rc = -ENOMEM);
2663         ctxt->ldc_lmv = &exp->exp_obd->u.lmv;
2664         ctxt->ldc_op_data = op_data;
2665         ctxt->ldc_cb_op = cb_op;
2666         ctxt->ldc_hash = offset;
2667         ctxt->ldc_count = stripe_count;
2668
2669         while (1) {
2670                 next = lmv_dirent_next(ctxt);
2671
2672                 /* end of directory */
2673                 if (!next) {
2674                         ctxt->ldc_hash = MDS_DIR_END_OFF;
2675                         break;
2676                 }
2677                 ctxt->ldc_hash = le64_to_cpu(next->lde_hash);
2678
2679                 ent_size = le16_to_cpu(next->lde_reclen);
2680
2681                 /* the last entry lde_reclen is 0, but it might not be the last
2682                  * one of this temporay dir page */
2683                 if (!ent_size)
2684                         ent_size = lu_dirent_calc_size(
2685                                         le16_to_cpu(next->lde_namelen),
2686                                         le32_to_cpu(next->lde_attrs));
2687                 /* page full */
2688                 if (ent_size > left_bytes)
2689                         break;
2690
2691                 memcpy(ent, next, ent_size);
2692
2693                 /* Replace . with master FID and Replace .. with the parent FID
2694                  * of master object */
2695                 if (strncmp(ent->lde_name, ".",
2696                             le16_to_cpu(ent->lde_namelen)) == 0 &&
2697                     le16_to_cpu(ent->lde_namelen) == 1)
2698                         fid_cpu_to_le(&ent->lde_fid, &op_data->op_fid1);
2699                 else if (strncmp(ent->lde_name, "..",
2700                                    le16_to_cpu(ent->lde_namelen)) == 0 &&
2701                            le16_to_cpu(ent->lde_namelen) == 2)
2702                         fid_cpu_to_le(&ent->lde_fid, &op_data->op_fid3);
2703
2704                 CDEBUG(D_INODE, "entry %.*s hash %#llx\n",
2705                        le16_to_cpu(ent->lde_namelen), ent->lde_name,
2706                        le64_to_cpu(ent->lde_hash));
2707
2708                 left_bytes -= ent_size;
2709                 ent->lde_reclen = cpu_to_le16(ent_size);
2710                 last_ent = ent;
2711                 ent = (void *)ent + ent_size;
2712         };
2713
2714         last_ent->lde_reclen = 0;
2715
2716         if (ent == start)
2717                 dp->ldp_flags |= LDF_EMPTY;
2718         else if (ctxt->ldc_hash == le64_to_cpu(last_ent->lde_hash))
2719                 dp->ldp_flags |= LDF_COLLIDE;
2720         dp->ldp_flags = cpu_to_le32(dp->ldp_flags);
2721         dp->ldp_hash_end = cpu_to_le64(ctxt->ldc_hash);
2722
2723         put_lmv_dir_ctxt(ctxt);
2724         OBD_FREE(ctxt, offsetof(typeof(*ctxt), ldc_stripes[stripe_count]));
2725
2726         *ppage = page;
2727
2728         RETURN(0);
2729
2730 free_page:
2731         kunmap(page);
2732         __free_page(page);
2733
2734         return rc;
2735 }
2736
2737 int lmv_read_page(struct obd_export *exp, struct md_op_data *op_data,
2738                   struct md_callback *cb_op, __u64 offset,
2739                   struct page **ppage)
2740 {
2741         struct obd_device *obd = exp->exp_obd;
2742         struct lmv_obd *lmv = &obd->u.lmv;
2743         struct lmv_tgt_desc *tgt;
2744         int rc;
2745
2746         ENTRY;
2747
2748         if (unlikely(lmv_dir_foreign(op_data->op_mea1)))
2749                 RETURN(-ENODATA);
2750
2751         if (unlikely(lmv_dir_striped(op_data->op_mea1))) {
2752                 rc = lmv_striped_read_page(exp, op_data, cb_op, offset, ppage);
2753                 RETURN(rc);
2754         }
2755
2756         tgt = lmv_fid2tgt(lmv, &op_data->op_fid1);
2757         if (IS_ERR(tgt))
2758                 RETURN(PTR_ERR(tgt));
2759
2760         rc = md_read_page(tgt->ltd_exp, op_data, cb_op, offset, ppage);
2761
2762         RETURN(rc);
2763 }
2764
2765 /**
2766  * Unlink a file/directory
2767  *
2768  * Unlink a file or directory under the parent dir. The unlink request
2769  * usually will be sent to the MDT where the child is located, but if
2770  * the client does not have the child FID then request will be sent to the
2771  * MDT where the parent is located.
2772  *
2773  * If the parent is a striped directory then it also needs to locate which
2774  * stripe the name of the child is located, and replace the parent FID
2775  * (@op->op_fid1) with the stripe FID. Note: if the stripe is unknown,
2776  * it will walk through all of sub-stripes until the child is being
2777  * unlinked finally.
2778  *
2779  * \param[in] exp       export refer to LMV
2780  * \param[in] op_data   different parameters transferred beween client
2781  *                      MD stacks, name, namelen, FIDs etc.
2782  *                      op_fid1 is the parent FID, op_fid2 is the child
2783  *                      FID.
2784  * \param[out] request  point to the request of unlink.
2785  *
2786  * retval               0 if succeed
2787  *                      negative errno if failed.
2788  */
2789 static int lmv_unlink(struct obd_export *exp, struct md_op_data *op_data,
2790                       struct ptlrpc_request **request)
2791 {
2792         struct obd_device *obd = exp->exp_obd;
2793         struct lmv_obd *lmv = &obd->u.lmv;
2794         struct lmv_tgt_desc *tgt;
2795         struct lmv_tgt_desc *parent_tgt;
2796         struct mdt_body *body;
2797         int rc;
2798
2799         ENTRY;
2800
2801         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2802         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2803         op_data->op_cap = cfs_curproc_cap_pack();
2804
2805 retry:
2806         parent_tgt = lmv_locate_tgt(lmv, op_data);
2807         if (IS_ERR(parent_tgt))
2808                 RETURN(PTR_ERR(parent_tgt));
2809
2810         if (likely(!fid_is_zero(&op_data->op_fid2))) {
2811                 tgt = lmv_fid2tgt(lmv, &op_data->op_fid2);
2812                 if (IS_ERR(tgt))
2813                         RETURN(PTR_ERR(tgt));
2814         } else {
2815                 tgt = parent_tgt;
2816         }
2817
2818         /*
2819          * If child's fid is given, cancel unused locks for it if it is from
2820          * another export than parent.
2821          *
2822          * LOOKUP lock for child (fid3) should also be cancelled on parent
2823          * tgt_tgt in mdc_unlink().
2824          */
2825         op_data->op_flags |= MF_MDC_CANCEL_FID1 | MF_MDC_CANCEL_FID3;
2826
2827         if (parent_tgt != tgt)
2828                 rc = lmv_early_cancel(exp, parent_tgt, op_data, tgt->ltd_index,
2829                                       LCK_EX, MDS_INODELOCK_LOOKUP,
2830                                       MF_MDC_CANCEL_FID3);
2831
2832         rc = lmv_early_cancel(exp, NULL, op_data, tgt->ltd_index, LCK_EX,
2833                               MDS_INODELOCK_ELC, MF_MDC_CANCEL_FID3);
2834         if (rc)
2835                 RETURN(rc);
2836
2837         CDEBUG(D_INODE, "unlink with fid="DFID"/"DFID" -> mds #%u\n",
2838                PFID(&op_data->op_fid1), PFID(&op_data->op_fid2),
2839                tgt->ltd_index);
2840
2841         rc = md_unlink(tgt->ltd_exp, op_data, request);
2842         if (rc == -ENOENT && lmv_dir_retry_check_update(op_data)) {
2843                 ptlrpc_req_finished(*request);
2844                 *request = NULL;
2845                 goto retry;
2846         }
2847
2848         if (rc != -EREMOTE)
2849                 RETURN(rc);
2850
2851         body = req_capsule_server_get(&(*request)->rq_pill, &RMF_MDT_BODY);
2852         if (body == NULL)
2853                 RETURN(-EPROTO);
2854
2855         /* Not cross-ref case, just get out of here. */
2856         if (likely(!(body->mbo_valid & OBD_MD_MDS)))
2857                 RETURN(rc);
2858
2859         /* This is a remote object, try remote MDT. */
2860         op_data->op_fid2 = body->mbo_fid1;
2861         ptlrpc_req_finished(*request);
2862         *request = NULL;
2863
2864         tgt = lmv_fid2tgt(lmv, &op_data->op_fid2);
2865         if (IS_ERR(tgt))
2866                 RETURN(PTR_ERR(tgt));
2867
2868         goto retry;
2869 }
2870
2871 static int lmv_precleanup(struct obd_device *obd)
2872 {
2873         ENTRY;
2874         libcfs_kkuc_group_rem(&obd->obd_uuid, 0, KUC_GRP_HSM);
2875         fld_client_debugfs_fini(&obd->u.lmv.lmv_fld);
2876         lprocfs_obd_cleanup(obd);
2877         lprocfs_free_md_stats(obd);
2878         RETURN(0);
2879 }
2880
2881 /**
2882  * Get by key a value associated with a LMV device.
2883  *
2884  * Dispatch request to lower-layer devices as needed.
2885  *
2886  * \param[in] env               execution environment for this thread
2887  * \param[in] exp               export for the LMV device
2888  * \param[in] keylen            length of key identifier
2889  * \param[in] key               identifier of key to get value for
2890  * \param[in] vallen            size of \a val
2891  * \param[out] val              pointer to storage location for value
2892  * \param[in] lsm               optional striping metadata of object
2893  *
2894  * \retval 0            on success
2895  * \retval negative     negated errno on failure
2896  */
2897 static int lmv_get_info(const struct lu_env *env, struct obd_export *exp,
2898                         __u32 keylen, void *key, __u32 *vallen, void *val)
2899 {
2900         struct obd_device *obd;
2901         struct lmv_obd *lmv;
2902         struct lu_tgt_desc *tgt;
2903         int rc = 0;
2904
2905         ENTRY;
2906
2907         obd = class_exp2obd(exp);
2908         if (obd == NULL) {
2909                 CDEBUG(D_IOCTL, "Invalid client cookie %#llx\n",
2910                        exp->exp_handle.h_cookie);
2911                 RETURN(-EINVAL);
2912         }
2913
2914         lmv = &obd->u.lmv;
2915         if (keylen >= strlen("remote_flag") && !strcmp(key, "remote_flag")) {
2916                 LASSERT(*vallen == sizeof(__u32));
2917                 lmv_foreach_connected_tgt(lmv, tgt) {
2918                         if (!obd_get_info(env, tgt->ltd_exp, keylen, key,
2919                                           vallen, val))
2920                                 RETURN(0);
2921                 }
2922                 RETURN(-EINVAL);
2923         } else if (KEY_IS(KEY_MAX_EASIZE) ||
2924                    KEY_IS(KEY_DEFAULT_EASIZE) ||
2925                    KEY_IS(KEY_CONN_DATA)) {
2926                 /*
2927                  * Forwarding this request to first MDS, it should know LOV
2928                  * desc.
2929                  */
2930                 tgt = lmv_tgt(lmv, 0);
2931                 if (!tgt)
2932                         RETURN(-ENODEV);
2933
2934                 rc = obd_get_info(env, tgt->ltd_exp, keylen, key, vallen, val);
2935                 if (!rc && KEY_IS(KEY_CONN_DATA))
2936                         exp->exp_connect_data = *(struct obd_connect_data *)val;
2937                 RETURN(rc);
2938         } else if (KEY_IS(KEY_TGT_COUNT)) {
2939                 *((int *)val) = lmv->lmv_mdt_descs.ltd_lmv_desc.ld_tgt_count;
2940                 RETURN(0);
2941         }
2942
2943         CDEBUG(D_IOCTL, "Invalid key\n");
2944         RETURN(-EINVAL);
2945 }
2946
2947 static int lmv_rmfid(struct obd_export *exp, struct fid_array *fa,
2948                      int *__rcs, struct ptlrpc_request_set *_set)
2949 {
2950         struct obd_device *obd = class_exp2obd(exp);
2951         struct ptlrpc_request_set *set = _set;
2952         struct lmv_obd *lmv = &obd->u.lmv;
2953         int tgt_count = lmv->lmv_mdt_count;
2954         struct lu_tgt_desc *tgt;
2955         struct fid_array *fat, **fas = NULL;
2956         int i, rc, **rcs = NULL;
2957
2958         if (!set) {
2959                 set = ptlrpc_prep_set();
2960                 if (!set)
2961                         RETURN(-ENOMEM);
2962         }
2963
2964         /* split FIDs by targets */
2965         OBD_ALLOC_PTR_ARRAY(fas, tgt_count);
2966         if (fas == NULL)
2967                 GOTO(out, rc = -ENOMEM);
2968         OBD_ALLOC_PTR_ARRAY(rcs, tgt_count);
2969         if (rcs == NULL)
2970                 GOTO(out_fas, rc = -ENOMEM);
2971
2972         for (i = 0; i < fa->fa_nr; i++) {
2973                 unsigned int idx;
2974
2975                 rc = lmv_fld_lookup(lmv, &fa->fa_fids[i], &idx);
2976                 if (rc) {
2977                         CDEBUG(D_OTHER, "can't lookup "DFID": rc = %d\n",
2978                                PFID(&fa->fa_fids[i]), rc);
2979                         continue;
2980                 }
2981                 LASSERT(idx < tgt_count);
2982                 if (!fas[idx])
2983                         OBD_ALLOC(fas[idx], offsetof(struct fid_array,
2984                                   fa_fids[fa->fa_nr]));
2985                 if (!fas[idx])
2986                         GOTO(out, rc = -ENOMEM);
2987                 if (!rcs[idx])
2988                         OBD_ALLOC_PTR_ARRAY(rcs[idx], fa->fa_nr);
2989                 if (!rcs[idx])
2990                         GOTO(out, rc = -ENOMEM);
2991
2992                 fat = fas[idx];
2993                 fat->fa_fids[fat->fa_nr++] = fa->fa_fids[i];
2994         }
2995
2996         lmv_foreach_connected_tgt(lmv, tgt) {
2997                 fat = fas[tgt->ltd_index];
2998                 if (!fat || fat->fa_nr == 0)
2999                         continue;
3000                 rc = md_rmfid(tgt->ltd_exp, fat, rcs[tgt->ltd_index], set);
3001         }
3002
3003         rc = ptlrpc_set_wait(NULL, set);
3004         if (rc == 0) {
3005                 int j = 0;
3006                 for (i = 0; i < tgt_count; i++) {
3007                         fat = fas[i];
3008                         if (!fat || fat->fa_nr == 0)
3009                                 continue;
3010                         /* copy FIDs back */
3011                         memcpy(fa->fa_fids + j, fat->fa_fids,
3012                                fat->fa_nr * sizeof(struct lu_fid));
3013                         /* copy rcs back */
3014                         memcpy(__rcs + j, rcs[i], fat->fa_nr * sizeof(**rcs));
3015                         j += fat->fa_nr;
3016                 }
3017         }
3018         if (set != _set)
3019                 ptlrpc_set_destroy(set);
3020
3021 out:
3022         for (i = 0; i < tgt_count; i++) {
3023                 if (fas && fas[i])
3024                         OBD_FREE(fas[i], offsetof(struct fid_array,
3025                                                 fa_fids[fa->fa_nr]));
3026                 if (rcs && rcs[i])
3027                         OBD_FREE_PTR_ARRAY(rcs[i], fa->fa_nr);
3028         }
3029         if (rcs)
3030                 OBD_FREE_PTR_ARRAY(rcs, tgt_count);
3031 out_fas:
3032         if (fas)
3033                 OBD_FREE_PTR_ARRAY(fas, tgt_count);
3034
3035         RETURN(rc);
3036 }
3037
3038 /**
3039  * Asynchronously set by key a value associated with a LMV device.
3040  *
3041  * Dispatch request to lower-layer devices as needed.
3042  *
3043  * \param[in] env       execution environment for this thread
3044  * \param[in] exp       export for the LMV device
3045  * \param[in] keylen    length of key identifier
3046  * \param[in] key       identifier of key to store value for
3047  * \param[in] vallen    size of value to store
3048  * \param[in] val       pointer to data to be stored
3049  * \param[in] set       optional list of related ptlrpc requests
3050  *
3051  * \retval 0            on success
3052  * \retval negative     negated errno on failure
3053  */
3054 int lmv_set_info_async(const struct lu_env *env, struct obd_export *exp,
3055                         __u32 keylen, void *key, __u32 vallen, void *val,
3056                         struct ptlrpc_request_set *set)
3057 {
3058         struct lmv_tgt_desc     *tgt = NULL;
3059         struct obd_device       *obd;
3060         struct lmv_obd          *lmv;
3061         int rc = 0;
3062         ENTRY;
3063
3064         obd = class_exp2obd(exp);
3065         if (obd == NULL) {
3066                 CDEBUG(D_IOCTL, "Invalid client cookie %#llx\n",
3067                        exp->exp_handle.h_cookie);
3068                 RETURN(-EINVAL);
3069         }
3070         lmv = &obd->u.lmv;
3071
3072         if (KEY_IS(KEY_READ_ONLY) || KEY_IS(KEY_FLUSH_CTX) ||
3073             KEY_IS(KEY_DEFAULT_EASIZE)) {
3074                 int err = 0;
3075
3076                 lmv_foreach_connected_tgt(lmv, tgt) {
3077                         err = obd_set_info_async(env, tgt->ltd_exp,
3078                                                  keylen, key, vallen, val, set);
3079                         if (err && rc == 0)
3080                                 rc = err;
3081                 }
3082
3083                 RETURN(rc);
3084         }
3085
3086         RETURN(-EINVAL);
3087 }
3088
3089 static int lmv_unpack_md_v1(struct obd_export *exp, struct lmv_stripe_md *lsm,
3090                             const struct lmv_mds_md_v1 *lmm1)
3091 {
3092         struct lmv_obd  *lmv = &exp->exp_obd->u.lmv;
3093         int             stripe_count;
3094         int             cplen;
3095         int             i;
3096         int             rc = 0;
3097         ENTRY;
3098
3099         lsm->lsm_md_magic = le32_to_cpu(lmm1->lmv_magic);
3100         lsm->lsm_md_stripe_count = le32_to_cpu(lmm1->lmv_stripe_count);
3101         lsm->lsm_md_master_mdt_index = le32_to_cpu(lmm1->lmv_master_mdt_index);
3102         if (OBD_FAIL_CHECK(OBD_FAIL_UNKNOWN_LMV_STRIPE))
3103                 lsm->lsm_md_hash_type = LMV_HASH_TYPE_UNKNOWN;
3104         else
3105                 lsm->lsm_md_hash_type = le32_to_cpu(lmm1->lmv_hash_type);
3106         lsm->lsm_md_layout_version = le32_to_cpu(lmm1->lmv_layout_version);
3107         lsm->lsm_md_migrate_offset = le32_to_cpu(lmm1->lmv_migrate_offset);
3108         lsm->lsm_md_migrate_hash = le32_to_cpu(lmm1->lmv_migrate_hash);
3109         cplen = strlcpy(lsm->lsm_md_pool_name, lmm1->lmv_pool_name,
3110                         sizeof(lsm->lsm_md_pool_name));
3111
3112         if (cplen >= sizeof(lsm->lsm_md_pool_name))
3113                 RETURN(-E2BIG);
3114
3115         CDEBUG(D_INFO, "unpack lsm count %d/%d, master %d hash_type %#x/%#x "
3116                "layout_version %d\n", lsm->lsm_md_stripe_count,
3117                lsm->lsm_md_migrate_offset, lsm->lsm_md_master_mdt_index,
3118                lsm->lsm_md_hash_type, lsm->lsm_md_migrate_hash,
3119                lsm->lsm_md_layout_version);
3120
3121         stripe_count = le32_to_cpu(lmm1->lmv_stripe_count);
3122         for (i = 0; i < stripe_count; i++) {
3123                 fid_le_to_cpu(&lsm->lsm_md_oinfo[i].lmo_fid,
3124                               &lmm1->lmv_stripe_fids[i]);
3125                 /*
3126                  * set default value -1, so lmv_locate_tgt() knows this stripe
3127                  * target is not initialized.
3128                  */
3129                 lsm->lsm_md_oinfo[i].lmo_mds = LMV_OFFSET_DEFAULT;
3130                 if (!fid_is_sane(&lsm->lsm_md_oinfo[i].lmo_fid))
3131                         continue;
3132
3133                 rc = lmv_fld_lookup(lmv, &lsm->lsm_md_oinfo[i].lmo_fid,
3134                                     &lsm->lsm_md_oinfo[i].lmo_mds);
3135                 if (rc == -ENOENT)
3136                         continue;
3137
3138                 if (rc)
3139                         RETURN(rc);
3140
3141                 CDEBUG(D_INFO, "unpack fid #%d "DFID"\n", i,
3142                        PFID(&lsm->lsm_md_oinfo[i].lmo_fid));
3143         }
3144
3145         RETURN(rc);
3146 }
3147
3148 static inline int lmv_unpack_user_md(struct obd_export *exp,
3149                                      struct lmv_stripe_md *lsm,
3150                                      const struct lmv_user_md *lmu)
3151 {
3152         lsm->lsm_md_magic = le32_to_cpu(lmu->lum_magic);
3153         lsm->lsm_md_stripe_count = le32_to_cpu(lmu->lum_stripe_count);
3154         lsm->lsm_md_master_mdt_index = le32_to_cpu(lmu->lum_stripe_offset);
3155         lsm->lsm_md_hash_type = le32_to_cpu(lmu->lum_hash_type);
3156         lsm->lsm_md_pool_name[LOV_MAXPOOLNAME] = 0;
3157
3158         return 0;
3159 }
3160
3161 static int lmv_unpackmd(struct obd_export *exp, struct lmv_stripe_md **lsmp,
3162                         const union lmv_mds_md *lmm, size_t lmm_size)
3163 {
3164         struct lmv_stripe_md     *lsm;
3165         int                      lsm_size;
3166         int                      rc;
3167         bool                     allocated = false;
3168         ENTRY;
3169
3170         LASSERT(lsmp != NULL);
3171
3172         lsm = *lsmp;
3173         /* Free memmd */
3174         if (lsm != NULL && lmm == NULL) {
3175                 int i;
3176                 struct lmv_foreign_md *lfm = (struct lmv_foreign_md *)lsm;
3177
3178                 if (lfm->lfm_magic == LMV_MAGIC_FOREIGN) {
3179                         size_t lfm_size;
3180
3181                         lfm_size = lfm->lfm_length + offsetof(typeof(*lfm),
3182                                                               lfm_value[0]);
3183                         OBD_FREE_LARGE(lfm, lfm_size);
3184                         RETURN(0);
3185                 }
3186
3187                 if (lmv_dir_striped(lsm)) {
3188                         for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
3189                                 if (lsm->lsm_md_oinfo[i].lmo_root)
3190                                         iput(lsm->lsm_md_oinfo[i].lmo_root);
3191                         }
3192                         lsm_size = lmv_stripe_md_size(lsm->lsm_md_stripe_count);
3193                 } else {
3194                         lsm_size = lmv_stripe_md_size(0);
3195                 }
3196                 OBD_FREE(lsm, lsm_size);
3197                 *lsmp = NULL;
3198                 RETURN(0);
3199         }
3200
3201         /* foreign lmv case */
3202         if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_FOREIGN) {
3203                 struct lmv_foreign_md *lfm = (struct lmv_foreign_md *)lsm;
3204
3205                 if (lfm == NULL) {
3206                         OBD_ALLOC_LARGE(lfm, lmm_size);
3207                         if (lfm == NULL)
3208                                 RETURN(-ENOMEM);
3209                         *lsmp = (struct lmv_stripe_md *)lfm;
3210                 }
3211                 lfm->lfm_magic = le32_to_cpu(lmm->lmv_foreign_md.lfm_magic);
3212                 lfm->lfm_length = le32_to_cpu(lmm->lmv_foreign_md.lfm_length);
3213                 lfm->lfm_type = le32_to_cpu(lmm->lmv_foreign_md.lfm_type);
3214                 lfm->lfm_flags = le32_to_cpu(lmm->lmv_foreign_md.lfm_flags);
3215                 memcpy(&lfm->lfm_value, &lmm->lmv_foreign_md.lfm_value,
3216                        lfm->lfm_length);
3217                 RETURN(lmm_size);
3218         }
3219
3220         if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_STRIPE)
3221                 RETURN(-EPERM);
3222
3223         /* Unpack memmd */
3224         if (le32_to_cpu(lmm->lmv_magic) != LMV_MAGIC_V1 &&
3225             le32_to_cpu(lmm->lmv_magic) != LMV_USER_MAGIC) {
3226                 CERROR("%s: invalid lmv magic %x: rc = %d\n",
3227                        exp->exp_obd->obd_name, le32_to_cpu(lmm->lmv_magic),
3228                        -EIO);
3229                 RETURN(-EIO);
3230         }
3231
3232         if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_V1)
3233                 lsm_size = lmv_stripe_md_size(lmv_mds_md_stripe_count_get(lmm));
3234         else
3235                 /**
3236                  * Unpack default dirstripe(lmv_user_md) to lmv_stripe_md,
3237                  * stripecount should be 0 then.
3238                  */
3239                 lsm_size = lmv_stripe_md_size(0);
3240
3241         if (lsm == NULL) {
3242                 OBD_ALLOC(lsm, lsm_size);
3243                 if (lsm == NULL)
3244                         RETURN(-ENOMEM);
3245                 allocated = true;
3246                 *lsmp = lsm;
3247         }
3248
3249         switch (le32_to_cpu(lmm->lmv_magic)) {
3250         case LMV_MAGIC_V1:
3251                 rc = lmv_unpack_md_v1(exp, lsm, &lmm->lmv_md_v1);
3252                 break;
3253         case LMV_USER_MAGIC:
3254                 rc = lmv_unpack_user_md(exp, lsm, &lmm->lmv_user_md);
3255                 break;
3256         default:
3257                 CERROR("%s: unrecognized magic %x\n", exp->exp_obd->obd_name,
3258                        le32_to_cpu(lmm->lmv_magic));
3259                 rc = -EINVAL;
3260                 break;
3261         }
3262
3263         if (rc != 0 && allocated) {
3264                 OBD_FREE(lsm, lsm_size);
3265                 *lsmp = NULL;
3266                 lsm_size = rc;
3267         }
3268         RETURN(lsm_size);
3269 }
3270
3271 void lmv_free_memmd(struct lmv_stripe_md *lsm)
3272 {
3273         lmv_unpackmd(NULL, &lsm, NULL, 0);
3274 }
3275 EXPORT_SYMBOL(lmv_free_memmd);
3276
3277 static int lmv_cancel_unused(struct obd_export *exp, const struct lu_fid *fid,
3278                              union ldlm_policy_data *policy,
3279                              enum ldlm_mode mode, enum ldlm_cancel_flags flags,
3280                              void *opaque)
3281 {
3282         struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
3283         struct lu_tgt_desc *tgt;
3284         int err;
3285         int rc = 0;
3286
3287         ENTRY;
3288
3289         LASSERT(fid != NULL);
3290
3291         lmv_foreach_connected_tgt(lmv, tgt) {
3292                 if (!tgt->ltd_active)
3293                         continue;
3294
3295                 err = md_cancel_unused(tgt->ltd_exp, fid, policy, mode, flags,
3296                                        opaque);
3297                 if (!rc)
3298                         rc = err;
3299         }
3300         RETURN(rc);
3301 }
3302
3303 static int lmv_set_lock_data(struct obd_export *exp,
3304                              const struct lustre_handle *lockh,
3305                              void *data, __u64 *bits)
3306 {
3307         struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
3308         struct lmv_tgt_desc *tgt = lmv_tgt(lmv, 0);
3309         int rc;
3310
3311         ENTRY;
3312
3313         if (tgt == NULL || tgt->ltd_exp == NULL)
3314                 RETURN(-EINVAL);
3315         rc =  md_set_lock_data(tgt->ltd_exp, lockh, data, bits);
3316         RETURN(rc);
3317 }
3318
3319 enum ldlm_mode lmv_lock_match(struct obd_export *exp, __u64 flags,
3320                               const struct lu_fid *fid, enum ldlm_type type,
3321                               union ldlm_policy_data *policy,
3322                               enum ldlm_mode mode, struct lustre_handle *lockh)
3323 {
3324         struct obd_device *obd = exp->exp_obd;
3325         struct lmv_obd *lmv = &obd->u.lmv;
3326         enum ldlm_mode rc;
3327         struct lu_tgt_desc *tgt;
3328         int i;
3329         int index;
3330
3331         ENTRY;
3332
3333         CDEBUG(D_INODE, "Lock match for "DFID"\n", PFID(fid));
3334
3335         /*
3336          * With DNE every object can have two locks in different namespaces:
3337          * lookup lock in space of MDT storing direntry and update/open lock in
3338          * space of MDT storing inode.  Try the MDT that the FID maps to first,
3339          * since this can be easily found, and only try others if that fails.
3340          */
3341         for (i = 0, index = lmv_fid2tgt_index(lmv, fid);
3342              i < lmv->lmv_mdt_descs.ltd_tgts_size;
3343              i++, index = (index + 1) % lmv->lmv_mdt_descs.ltd_tgts_size) {
3344                 if (index < 0) {
3345                         CDEBUG(D_HA, "%s: "DFID" is inaccessible: rc = %d\n",
3346                                obd->obd_name, PFID(fid), index);
3347                         index = 0;
3348                 }
3349
3350                 tgt = lmv_tgt(lmv, index);
3351                 if (!tgt || !tgt->ltd_exp || !tgt->ltd_active)
3352                         continue;
3353
3354                 rc = md_lock_match(tgt->ltd_exp, flags, fid, type, policy, mode,
3355                                    lockh);
3356                 if (rc)
3357                         RETURN(rc);
3358         }
3359
3360         RETURN(0);
3361 }
3362
3363 int lmv_get_lustre_md(struct obd_export *exp, struct ptlrpc_request *req,
3364                       struct obd_export *dt_exp, struct obd_export *md_exp,
3365                       struct lustre_md *md)
3366 {
3367         struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
3368         struct lmv_tgt_desc *tgt = lmv_tgt(lmv, 0);
3369
3370         if (!tgt || !tgt->ltd_exp)
3371                 return -EINVAL;
3372
3373         return md_get_lustre_md(tgt->ltd_exp, req, dt_exp, md_exp, md);
3374 }
3375
3376 int lmv_free_lustre_md(struct obd_export *exp, struct lustre_md *md)
3377 {
3378         struct obd_device *obd = exp->exp_obd;
3379         struct lmv_obd *lmv = &obd->u.lmv;
3380         struct lmv_tgt_desc *tgt = lmv_tgt(lmv, 0);
3381
3382         ENTRY;
3383
3384         if (md->default_lmv) {
3385                 lmv_free_memmd(md->default_lmv);
3386                 md->default_lmv = NULL;
3387         }
3388         if (md->lmv != NULL) {
3389                 lmv_free_memmd(md->lmv);
3390                 md->lmv = NULL;
3391         }
3392         if (!tgt || !tgt->ltd_exp)
3393                 RETURN(-EINVAL);
3394         RETURN(md_free_lustre_md(tgt->ltd_exp, md));
3395 }
3396
3397 int lmv_set_open_replay_data(struct obd_export *exp,
3398                              struct obd_client_handle *och,
3399                              struct lookup_intent *it)
3400 {
3401         struct obd_device *obd = exp->exp_obd;
3402         struct lmv_obd *lmv = &obd->u.lmv;
3403         struct lmv_tgt_desc *tgt;
3404
3405         ENTRY;
3406
3407         tgt = lmv_fid2tgt(lmv, &och->och_fid);
3408         if (IS_ERR(tgt))
3409                 RETURN(PTR_ERR(tgt));
3410
3411         RETURN(md_set_open_replay_data(tgt->ltd_exp, och, it));
3412 }
3413
3414 int lmv_clear_open_replay_data(struct obd_export *exp,
3415                                struct obd_client_handle *och)
3416 {
3417         struct obd_device *obd = exp->exp_obd;
3418         struct lmv_obd *lmv = &obd->u.lmv;
3419         struct lmv_tgt_desc *tgt;
3420
3421         ENTRY;
3422
3423         tgt = lmv_fid2tgt(lmv, &och->och_fid);
3424         if (IS_ERR(tgt))
3425                 RETURN(PTR_ERR(tgt));
3426
3427         RETURN(md_clear_open_replay_data(tgt->ltd_exp, och));
3428 }
3429
3430 int lmv_intent_getattr_async(struct obd_export *exp,
3431                              struct md_enqueue_info *minfo)
3432 {
3433         struct md_op_data *op_data = &minfo->mi_data;
3434         struct obd_device *obd = exp->exp_obd;
3435         struct lmv_obd *lmv = &obd->u.lmv;
3436         struct lmv_tgt_desc *ptgt;
3437         struct lmv_tgt_desc *ctgt;
3438         int rc;
3439
3440         ENTRY;
3441
3442         if (!fid_is_sane(&op_data->op_fid2))
3443                 RETURN(-EINVAL);
3444
3445         ptgt = lmv_locate_tgt(lmv, op_data);
3446         if (IS_ERR(ptgt))
3447                 RETURN(PTR_ERR(ptgt));
3448
3449         ctgt = lmv_fid2tgt(lmv, &op_data->op_fid2);
3450         if (IS_ERR(ctgt))
3451                 RETURN(PTR_ERR(ctgt));
3452
3453         /*
3454          * remote object needs two RPCs to lookup and getattr, considering the
3455          * complexity don't support statahead for now.
3456          */
3457         if (ctgt != ptgt)
3458                 RETURN(-EREMOTE);
3459
3460         rc = md_intent_getattr_async(ptgt->ltd_exp, minfo);
3461
3462         RETURN(rc);
3463 }
3464
3465 int lmv_revalidate_lock(struct obd_export *exp, struct lookup_intent *it,
3466                         struct lu_fid *fid, __u64 *bits)
3467 {
3468         struct obd_device *obd = exp->exp_obd;
3469         struct lmv_obd *lmv = &obd->u.lmv;
3470         struct lmv_tgt_desc *tgt;
3471         int rc;
3472
3473         ENTRY;
3474
3475         tgt = lmv_fid2tgt(lmv, fid);
3476         if (IS_ERR(tgt))
3477                 RETURN(PTR_ERR(tgt));
3478
3479         rc = md_revalidate_lock(tgt->ltd_exp, it, fid, bits);
3480         RETURN(rc);
3481 }
3482
3483 int lmv_get_fid_from_lsm(struct obd_export *exp,
3484                          const struct lmv_stripe_md *lsm,
3485                          const char *name, int namelen, struct lu_fid *fid)
3486 {
3487         const struct lmv_oinfo *oinfo;
3488
3489         LASSERT(lmv_dir_striped(lsm));
3490
3491         oinfo = lsm_name_to_stripe_info(lsm, name, namelen, false);
3492         if (IS_ERR(oinfo))
3493                 return PTR_ERR(oinfo);
3494
3495         *fid = oinfo->lmo_fid;
3496
3497         RETURN(0);
3498 }
3499
3500 /**
3501  * For lmv, only need to send request to master MDT, and the master MDT will
3502  * process with other slave MDTs. The only exception is Q_GETOQUOTA for which
3503  * we directly fetch data from the slave MDTs.
3504  */
3505 int lmv_quotactl(struct obd_device *unused, struct obd_export *exp,
3506                  struct obd_quotactl *oqctl)
3507 {
3508         struct obd_device *obd = class_exp2obd(exp);
3509         struct lmv_obd *lmv = &obd->u.lmv;
3510         struct lmv_tgt_desc *tgt = lmv_tgt(lmv, 0);
3511         __u64 curspace, curinodes;
3512         int rc = 0;
3513
3514         ENTRY;
3515
3516         if (!tgt || !tgt->ltd_exp || !tgt->ltd_active) {
3517                 CERROR("master lmv inactive\n");
3518                 RETURN(-EIO);
3519         }
3520
3521         if (oqctl->qc_cmd != Q_GETOQUOTA) {
3522                 rc = obd_quotactl(tgt->ltd_exp, oqctl);
3523                 RETURN(rc);
3524         }
3525
3526         curspace = curinodes = 0;
3527         lmv_foreach_connected_tgt(lmv, tgt) {
3528                 int err;
3529
3530                 if (!tgt->ltd_active)
3531                         continue;
3532
3533                 err = obd_quotactl(tgt->ltd_exp, oqctl);
3534                 if (err) {
3535                         CERROR("getquota on mdt %d failed. %d\n",
3536                                tgt->ltd_index, err);
3537                         if (!rc)
3538                                 rc = err;
3539                 } else {
3540                         curspace += oqctl->qc_dqblk.dqb_curspace;
3541                         curinodes += oqctl->qc_dqblk.dqb_curinodes;
3542                 }
3543         }
3544         oqctl->qc_dqblk.dqb_curspace = curspace;
3545         oqctl->qc_dqblk.dqb_curinodes = curinodes;
3546
3547         RETURN(rc);
3548 }
3549
3550 static int lmv_merge_attr(struct obd_export *exp,
3551                           const struct lmv_stripe_md *lsm,
3552                           struct cl_attr *attr,
3553                           ldlm_blocking_callback cb_blocking)
3554 {
3555         int rc;
3556         int i;
3557
3558         if (!lmv_dir_striped(lsm))
3559                 return 0;
3560
3561         rc = lmv_revalidate_slaves(exp, lsm, cb_blocking, 0);
3562         if (rc < 0)
3563                 return rc;
3564
3565         for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
3566                 struct inode *inode = lsm->lsm_md_oinfo[i].lmo_root;
3567
3568                 if (!inode)
3569                         continue;
3570
3571                 CDEBUG(D_INFO,
3572                        "" DFID " size %llu, blocks %llu nlink %u, atime %lld ctime %lld, mtime %lld.\n",
3573                        PFID(&lsm->lsm_md_oinfo[i].lmo_fid),
3574                        i_size_read(inode), (unsigned long long)inode->i_blocks,
3575                        inode->i_nlink, (s64)inode->i_atime.tv_sec,
3576                        (s64)inode->i_ctime.tv_sec, (s64)inode->i_mtime.tv_sec);
3577
3578                 /* for slave stripe, it needs to subtract nlink for . and .. */
3579                 if (i != 0)
3580                         attr->cat_nlink += inode->i_nlink - 2;
3581                 else
3582                         attr->cat_nlink = inode->i_nlink;
3583
3584                 attr->cat_size += i_size_read(inode);
3585                 attr->cat_blocks += inode->i_blocks;
3586
3587                 if (attr->cat_atime < inode->i_atime.tv_sec)
3588                         attr->cat_atime = inode->i_atime.tv_sec;
3589
3590                 if (attr->cat_ctime < inode->i_ctime.tv_sec)
3591                         attr->cat_ctime = inode->i_ctime.tv_sec;
3592
3593                 if (attr->cat_mtime < inode->i_mtime.tv_sec)
3594                         attr->cat_mtime = inode->i_mtime.tv_sec;
3595         }
3596         return 0;
3597 }
3598
3599 static const struct obd_ops lmv_obd_ops = {
3600         .o_owner                = THIS_MODULE,
3601         .o_setup                = lmv_setup,
3602         .o_cleanup              = lmv_cleanup,
3603         .o_precleanup           = lmv_precleanup,
3604         .o_process_config       = lmv_process_config,
3605         .o_connect              = lmv_connect,
3606         .o_disconnect           = lmv_disconnect,
3607         .o_statfs               = lmv_statfs,
3608         .o_get_info             = lmv_get_info,
3609         .o_set_info_async       = lmv_set_info_async,
3610         .o_notify               = lmv_notify,
3611         .o_get_uuid             = lmv_get_uuid,
3612         .o_fid_alloc            = lmv_fid_alloc,
3613         .o_iocontrol            = lmv_iocontrol,
3614         .o_quotactl             = lmv_quotactl
3615 };
3616
3617 static const struct md_ops lmv_md_ops = {
3618         .m_get_root             = lmv_get_root,
3619         .m_null_inode           = lmv_null_inode,
3620         .m_close                = lmv_close,
3621         .m_create               = lmv_create,
3622         .m_enqueue              = lmv_enqueue,
3623         .m_getattr              = lmv_getattr,
3624         .m_getxattr             = lmv_getxattr,
3625         .m_getattr_name         = lmv_getattr_name,
3626         .m_intent_lock          = lmv_intent_lock,
3627         .m_link                 = lmv_link,
3628         .m_rename               = lmv_rename,
3629         .m_setattr              = lmv_setattr,
3630         .m_setxattr             = lmv_setxattr,
3631         .m_fsync                = lmv_fsync,
3632         .m_file_resync          = lmv_file_resync,
3633         .m_read_page            = lmv_read_page,
3634         .m_unlink               = lmv_unlink,
3635         .m_init_ea_size         = lmv_init_ea_size,
3636         .m_cancel_unused        = lmv_cancel_unused,
3637         .m_set_lock_data        = lmv_set_lock_data,
3638         .m_lock_match           = lmv_lock_match,
3639         .m_get_lustre_md        = lmv_get_lustre_md,
3640         .m_free_lustre_md       = lmv_free_lustre_md,
3641         .m_merge_attr           = lmv_merge_attr,
3642         .m_set_open_replay_data = lmv_set_open_replay_data,
3643         .m_clear_open_replay_data = lmv_clear_open_replay_data,
3644         .m_intent_getattr_async = lmv_intent_getattr_async,
3645         .m_revalidate_lock      = lmv_revalidate_lock,
3646         .m_get_fid_from_lsm     = lmv_get_fid_from_lsm,
3647         .m_unpackmd             = lmv_unpackmd,
3648         .m_rmfid                = lmv_rmfid,
3649 };
3650
3651 static int __init lmv_init(void)
3652 {
3653         return class_register_type(&lmv_obd_ops, &lmv_md_ops, true, NULL,
3654                                    LUSTRE_LMV_NAME, NULL);
3655 }
3656
3657 static void __exit lmv_exit(void)
3658 {
3659         class_unregister_type(LUSTRE_LMV_NAME);
3660 }
3661
3662 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
3663 MODULE_DESCRIPTION("Lustre Logical Metadata Volume");
3664 MODULE_VERSION(LUSTRE_VERSION_STRING);
3665 MODULE_LICENSE("GPL");
3666
3667 module_init(lmv_init);
3668 module_exit(lmv_exit);