lustre/lmv/lmv_obd.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2004, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2011, 2013, Intel Corporation.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  */
  36
  37 #define DEBUG_SUBSYSTEM S_LMV
  38 #include <linux/slab.h>
  39 #include <linux/module.h>
  40 #include <linux/init.h>
  41 #include <linux/slab.h>
  42 #include <linux/pagemap.h>
  43 #include <linux/mm.h>
  44 #include <linux/math64.h>
  45 #include <linux/seq_file.h>
  46 #include <linux/namei.h>
  47
  48 #include <lustre/lustre_idl.h>
  49 #include <obd_support.h>
  50 #include <lustre_lib.h>
  51 #include <lustre_net.h>
  52 #include <obd_class.h>
  53 #include <lustre_lmv.h>
  54 #include <lprocfs_status.h>
  55 #include <cl_object.h>
  56 #include <lclient.h>
  57 #include <lustre_lite.h>
  58 #include <lustre_fid.h>
  59 #include <lustre_ioctl.h>
  60 #include "lmv_internal.h"
  61
  62 /* This hash is only for testing purpose */
  63 static inline unsigned int
  64 lmv_hash_all_chars(unsigned int count, const char *name, int namelen)
  65 {
  66         unsigned int c = 0;
  67         const unsigned char *p = (const unsigned char *)name;
  68
  69         while (--namelen >= 0)
  70                 c += p[namelen];
  71
  72         c = c % count;
  73
  74         return c;
  75 }
  76
  77 static inline unsigned int
  78 lmv_hash_fnv1a(unsigned int count, const char *name, int namelen)
  79 {
  80         __u64   hash;
  81
  82         hash = lustre_hash_fnv_1a_64(name, namelen);
  83
  84         hash = hash % count;
  85
  86         return hash;
  87 }
  88
  89 int lmv_name_to_stripe_index(__u32 lmv_hash_type, unsigned int stripe_count,
  90                              const char *name, int namelen)
  91 {
  92         int     idx;
  93         __u32   hash_type = lmv_hash_type & LMV_HASH_TYPE_MASK;
  94
  95         LASSERT(namelen > 0);
  96         if (stripe_count <= 1)
  97                 return 0;
  98
  99         /* for migrating object, always start from 0 stripe */
 100         if (lmv_hash_type & LMV_HASH_FLAG_MIGRATION)
 101                 return 0;
 102
 103         switch (hash_type) {
 104         case LMV_HASH_TYPE_ALL_CHARS:
 105                 idx = lmv_hash_all_chars(stripe_count, name, namelen);
 106                 break;
 107         case LMV_HASH_TYPE_FNV_1A_64:
 108                 idx = lmv_hash_fnv1a(stripe_count, name, namelen);
 109                 break;
 110         default:
 111                 idx = -EBADFD;
 112                 break;
 113         }
 114
 115         CDEBUG(D_INFO, "name %.*s hash_type %d idx %d\n", namelen, name,
 116                hash_type, idx);
 117
 118         return idx;
 119 }
 120
 121 static void lmv_activate_target(struct lmv_obd *lmv,
 122                                 struct lmv_tgt_desc *tgt,
 123                                 int activate)
 124 {
 125         if (tgt->ltd_active == activate)
 126                 return;
 127
 128         tgt->ltd_active = activate;
 129         lmv->desc.ld_active_tgt_count += (activate ? 1 : -1);
 130 }
 131
 132 /**
 133  * Error codes:
 134  *
 135  *  -EINVAL  : UUID can't be found in the LMV's target list
 136  *  -ENOTCONN: The UUID is found, but the target connection is bad (!)
 137  *  -EBADF   : The UUID is found, but the OBD of the wrong type (!)
 138  */
 139 static int lmv_set_mdc_active(struct lmv_obd *lmv,
 140                               const struct obd_uuid *uuid,
 141                               int activate)
 142 {
 143         struct lmv_tgt_desc     *tgt = NULL;
 144         struct obd_device       *obd;
 145         __u32                    i;
 146         int                      rc = 0;
 147         ENTRY;
 148
 149         CDEBUG(D_INFO, "Searching in lmv %p for uuid %s (activate=%d)\n",
 150                         lmv, uuid->uuid, activate);
 151
 152         spin_lock(&lmv->lmv_lock);
 153         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 154                 tgt = lmv->tgts[i];
 155                 if (tgt == NULL || tgt->ltd_exp == NULL)
 156                         continue;
 157
 158                 CDEBUG(D_INFO, "Target idx %d is %s conn "LPX64"\n", i,
 159                        tgt->ltd_uuid.uuid, tgt->ltd_exp->exp_handle.h_cookie);
 160
 161                 if (obd_uuid_equals(uuid, &tgt->ltd_uuid))
 162                         break;
 163         }
 164
 165         if (i == lmv->desc.ld_tgt_count)
 166                 GOTO(out_lmv_lock, rc = -EINVAL);
 167
 168         obd = class_exp2obd(tgt->ltd_exp);
 169         if (obd == NULL)
 170                 GOTO(out_lmv_lock, rc = -ENOTCONN);
 171
 172         CDEBUG(D_INFO, "Found OBD %s=%s device %d (%p) type %s at LMV idx %d\n",
 173                obd->obd_name, obd->obd_uuid.uuid, obd->obd_minor, obd,
 174                obd->obd_type->typ_name, i);
 175         LASSERT(strcmp(obd->obd_type->typ_name, LUSTRE_MDC_NAME) == 0);
 176
 177         if (tgt->ltd_active == activate) {
 178                 CDEBUG(D_INFO, "OBD %p already %sactive!\n", obd,
 179                        activate ? "" : "in");
 180                 GOTO(out_lmv_lock, rc);
 181         }
 182
 183         CDEBUG(D_INFO, "Marking OBD %p %sactive\n", obd,
 184                activate ? "" : "in");
 185         lmv_activate_target(lmv, tgt, activate);
 186         EXIT;
 187
 188  out_lmv_lock:
 189         spin_unlock(&lmv->lmv_lock);
 190         return rc;
 191 }
 192
 193 struct obd_uuid *lmv_get_uuid(struct obd_export *exp)
 194 {
 195         struct lmv_obd          *lmv = &exp->exp_obd->u.lmv;
 196         struct lmv_tgt_desc     *tgt = lmv->tgts[0];
 197
 198         return (tgt == NULL) ? NULL : obd_get_uuid(tgt->ltd_exp);
 199 }
 200
 201 static int lmv_notify(struct obd_device *obd, struct obd_device *watched,
 202                       enum obd_notify_event ev, void *data)
 203 {
 204         struct obd_connect_data *conn_data;
 205         struct lmv_obd          *lmv = &obd->u.lmv;
 206         struct obd_uuid         *uuid;
 207         int                      rc = 0;
 208         ENTRY;
 209
 210         if (strcmp(watched->obd_type->typ_name, LUSTRE_MDC_NAME)) {
 211                 CERROR("unexpected notification of %s %s!\n",
 212                        watched->obd_type->typ_name,
 213                        watched->obd_name);
 214                 RETURN(-EINVAL);
 215         }
 216
 217         uuid = &watched->u.cli.cl_target_uuid;
 218         if (ev == OBD_NOTIFY_ACTIVE || ev == OBD_NOTIFY_INACTIVE) {
 219                 /*
 220                  * Set MDC as active before notifying the observer, so the
 221                  * observer can use the MDC normally.
 222                  */
 223                 rc = lmv_set_mdc_active(lmv, uuid,
 224                                         ev == OBD_NOTIFY_ACTIVE);
 225                 if (rc) {
 226                         CERROR("%sactivation of %s failed: %d\n",
 227                                ev == OBD_NOTIFY_ACTIVE ? "" : "de",
 228                                uuid->uuid, rc);
 229                         RETURN(rc);
 230                 }
 231         } else if (ev == OBD_NOTIFY_OCD) {
 232                 conn_data = &watched->u.cli.cl_import->imp_connect_data;
 233                 /*
 234                  * XXX: Make sure that ocd_connect_flags from all targets are
 235                  * the same. Otherwise one of MDTs runs wrong version or
 236                  * something like this.  --umka
 237                  */
 238                 obd->obd_self_export->exp_connect_data = *conn_data;
 239         }
 240 #if 0
 241         else if (ev == OBD_NOTIFY_DISCON) {
 242                 /*
 243                  * For disconnect event, flush fld cache for failout MDS case.
 244                  */
 245                 fld_client_flush(&lmv->lmv_fld);
 246         }
 247 #endif
 248         /*
 249          * Pass the notification up the chain.
 250          */
 251         if (obd->obd_observer)
 252                 rc = obd_notify(obd->obd_observer, watched, ev, data);
 253
 254         RETURN(rc);
 255 }
 256
 257 /**
 258  * This is fake connect function. Its purpose is to initialize lmv and say
 259  * caller that everything is okay. Real connection will be performed later.
 260  */
 261 static int lmv_connect(const struct lu_env *env,
 262                        struct obd_export **exp, struct obd_device *obd,
 263                        struct obd_uuid *cluuid, struct obd_connect_data *data,
 264                        void *localdata)
 265 {
 266         struct lmv_obd        *lmv = &obd->u.lmv;
 267         struct lustre_handle  conn = { 0 };
 268         int                    rc = 0;
 269         ENTRY;
 270
 271         /*
 272          * We don't want to actually do the underlying connections more than
 273          * once, so keep track.
 274          */
 275         lmv->refcount++;
 276         if (lmv->refcount > 1) {
 277                 *exp = NULL;
 278                 RETURN(0);
 279         }
 280
 281         rc = class_connect(&conn, obd, cluuid);
 282         if (rc) {
 283                 CERROR("class_connection() returned %d\n", rc);
 284                 RETURN(rc);
 285         }
 286
 287         *exp = class_conn2export(&conn);
 288         class_export_get(*exp);
 289
 290         lmv->exp = *exp;
 291         lmv->connected = 0;
 292         lmv->cluuid = *cluuid;
 293
 294         if (data)
 295                 lmv->conn_data = *data;
 296
 297         if (lmv->targets_proc_entry == NULL) {
 298                 lmv->targets_proc_entry = lprocfs_seq_register("target_obds",
 299                                                         obd->obd_proc_entry,
 300                                                         NULL, NULL);
 301                 if (IS_ERR(lmv->targets_proc_entry)) {
 302                         CERROR("%s: cannot register "
 303                                "/proc/fs/lustre/%s/%s/target_obds\n",
 304                                obd->obd_name, obd->obd_type->typ_name,
 305                                obd->obd_name);
 306                         lmv->targets_proc_entry = NULL;
 307                 }
 308         }
 309
 310         /*
 311          * All real clients should perform actual connection right away, because
 312          * it is possible, that LMV will not have opportunity to connect targets
 313          * and MDC stuff will be called directly, for instance while reading
 314          * ../mdc/../kbytesfree procfs file, etc.
 315          */
 316         if (data != NULL && (data->ocd_connect_flags & OBD_CONNECT_REAL))
 317                 rc = lmv_check_connect(obd);
 318
 319         if (rc && lmv->targets_proc_entry != NULL)
 320                 lprocfs_remove(&lmv->targets_proc_entry);
 321         RETURN(rc);
 322 }
 323
 324 static void lmv_set_timeouts(struct obd_device *obd)
 325 {
 326         struct lmv_obd          *lmv;
 327         __u32                    i;
 328
 329         lmv = &obd->u.lmv;
 330         if (lmv->server_timeout == 0)
 331                 return;
 332
 333         if (lmv->connected == 0)
 334                 return;
 335
 336         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 337                 struct lmv_tgt_desc *tgt = lmv->tgts[i];
 338
 339                 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active)
 340                         continue;
 341
 342                 obd_set_info_async(NULL, tgt->ltd_exp, sizeof(KEY_INTERMDS),
 343                                    KEY_INTERMDS, 0, NULL, NULL);
 344         }
 345 }
 346
 347 static int lmv_init_ea_size(struct obd_export *exp, int easize,
 348                             int def_easize, int cookiesize, int def_cookiesize)
 349 {
 350         struct obd_device       *obd = exp->exp_obd;
 351         struct lmv_obd          *lmv = &obd->u.lmv;
 352         __u32                    i;
 353         int                      rc = 0;
 354         int                      change = 0;
 355         ENTRY;
 356
 357         if (lmv->max_easize < easize) {
 358                 lmv->max_easize = easize;
 359                 change = 1;
 360         }
 361         if (lmv->max_def_easize < def_easize) {
 362                 lmv->max_def_easize = def_easize;
 363                 change = 1;
 364         }
 365         if (lmv->max_cookiesize < cookiesize) {
 366                 lmv->max_cookiesize = cookiesize;
 367                 change = 1;
 368         }
 369         if (lmv->max_def_cookiesize < def_cookiesize) {
 370                 lmv->max_def_cookiesize = def_cookiesize;
 371                 change = 1;
 372         }
 373         if (change == 0)
 374                 RETURN(0);
 375
 376         if (lmv->connected == 0)
 377                 RETURN(0);
 378
 379         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 380                 struct lmv_tgt_desc *tgt = lmv->tgts[i];
 381
 382                 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active) {
 383                         CWARN("%s: NULL export for %d\n", obd->obd_name, i);
 384                         continue;
 385                 }
 386
 387                 rc = md_init_ea_size(tgt->ltd_exp, easize, def_easize,
 388                                      cookiesize, def_cookiesize);
 389                 if (rc) {
 390                         CERROR("%s: obd_init_ea_size() failed on MDT target %d:"
 391                                " rc = %d\n", obd->obd_name, i, rc);
 392                         break;
 393                 }
 394         }
 395         RETURN(rc);
 396 }
 397
 398 #define MAX_STRING_SIZE 128
 399
 400 int lmv_connect_mdc(struct obd_device *obd, struct lmv_tgt_desc *tgt)
 401 {
 402         struct lmv_obd          *lmv = &obd->u.lmv;
 403         struct obd_uuid         *cluuid = &lmv->cluuid;
 404         struct obd_uuid          lmv_mdc_uuid = { "LMV_MDC_UUID" };
 405         struct obd_device       *mdc_obd;
 406         struct obd_export       *mdc_exp;
 407         struct lu_fld_target     target;
 408         int                      rc;
 409         ENTRY;
 410
 411         mdc_obd = class_find_client_obd(&tgt->ltd_uuid, LUSTRE_MDC_NAME,
 412                                         &obd->obd_uuid);
 413         if (!mdc_obd) {
 414                 CERROR("target %s not attached\n", tgt->ltd_uuid.uuid);
 415                 RETURN(-EINVAL);
 416         }
 417
 418         CDEBUG(D_CONFIG, "connect to %s(%s) - %s, %s FOR %s\n",
 419                 mdc_obd->obd_name, mdc_obd->obd_uuid.uuid,
 420                 tgt->ltd_uuid.uuid, obd->obd_uuid.uuid,
 421                 cluuid->uuid);
 422
 423         if (!mdc_obd->obd_set_up) {
 424                 CERROR("target %s is not set up\n", tgt->ltd_uuid.uuid);
 425                 RETURN(-EINVAL);
 426         }
 427
 428         rc = obd_connect(NULL, &mdc_exp, mdc_obd, &lmv_mdc_uuid,
 429                          &lmv->conn_data, NULL);
 430         if (rc) {
 431                 CERROR("target %s connect error %d\n", tgt->ltd_uuid.uuid, rc);
 432                 RETURN(rc);
 433         }
 434
 435         /*
 436          * Init fid sequence client for this mdc and add new fld target.
 437          */
 438         rc = obd_fid_init(mdc_obd, mdc_exp, LUSTRE_SEQ_METADATA);
 439         if (rc)
 440                 RETURN(rc);
 441
 442         target.ft_srv = NULL;
 443         target.ft_exp = mdc_exp;
 444         target.ft_idx = tgt->ltd_idx;
 445
 446         fld_client_add_target(&lmv->lmv_fld, &target);
 447
 448         rc = obd_register_observer(mdc_obd, obd);
 449         if (rc) {
 450                 obd_disconnect(mdc_exp);
 451                 CERROR("target %s register_observer error %d\n",
 452                        tgt->ltd_uuid.uuid, rc);
 453                 RETURN(rc);
 454         }
 455
 456         if (obd->obd_observer) {
 457                 /*
 458                  * Tell the observer about the new target.
 459                  */
 460                 rc = obd_notify(obd->obd_observer, mdc_exp->exp_obd,
 461                                 OBD_NOTIFY_ACTIVE,
 462                                 (void *)(tgt - lmv->tgts[0]));
 463                 if (rc) {
 464                         obd_disconnect(mdc_exp);
 465                         RETURN(rc);
 466                 }
 467         }
 468
 469         tgt->ltd_active = 1;
 470         tgt->ltd_exp = mdc_exp;
 471         lmv->desc.ld_active_tgt_count++;
 472
 473         md_init_ea_size(tgt->ltd_exp, lmv->max_easize, lmv->max_def_easize,
 474                         lmv->max_cookiesize, lmv->max_def_cookiesize);
 475
 476         CDEBUG(D_CONFIG, "Connected to %s(%s) successfully (%d)\n",
 477                 mdc_obd->obd_name, mdc_obd->obd_uuid.uuid,
 478                 atomic_read(&obd->obd_refcount));
 479
 480         if (lmv->targets_proc_entry != NULL) {
 481                 struct proc_dir_entry *mdc_symlink;
 482
 483                 LASSERT(mdc_obd->obd_type != NULL);
 484                 LASSERT(mdc_obd->obd_type->typ_name != NULL);
 485                 mdc_symlink = lprocfs_add_symlink(mdc_obd->obd_name,
 486                                                   lmv->targets_proc_entry,
 487                                                   "../../../%s/%s",
 488                                                   mdc_obd->obd_type->typ_name,
 489                                                   mdc_obd->obd_name);
 490                 if (mdc_symlink == NULL) {
 491                         CERROR("cannot register LMV target "
 492                                "/proc/fs/lustre/%s/%s/target_obds/%s\n",
 493                                obd->obd_type->typ_name, obd->obd_name,
 494                                mdc_obd->obd_name);
 495                 }
 496         }
 497         RETURN(0);
 498 }
 499
 500 static void lmv_del_target(struct lmv_obd *lmv, int index)
 501 {
 502         if (lmv->tgts[index] == NULL)
 503                 return;
 504
 505         OBD_FREE_PTR(lmv->tgts[index]);
 506         lmv->tgts[index] = NULL;
 507         return;
 508 }
 509
 510 static int lmv_add_target(struct obd_device *obd, struct obd_uuid *uuidp,
 511                            __u32 index, int gen)
 512 {
 513         struct lmv_obd      *lmv = &obd->u.lmv;
 514         struct lmv_tgt_desc *tgt;
 515         int                  orig_tgt_count = 0;
 516         int                  rc = 0;
 517         ENTRY;
 518
 519         CDEBUG(D_CONFIG, "Target uuid: %s. index %d\n", uuidp->uuid, index);
 520
 521         lmv_init_lock(lmv);
 522
 523         if (lmv->desc.ld_tgt_count == 0) {
 524                 struct obd_device *mdc_obd;
 525
 526                 mdc_obd = class_find_client_obd(uuidp, LUSTRE_MDC_NAME,
 527                                                 &obd->obd_uuid);
 528                 if (!mdc_obd) {
 529                         lmv_init_unlock(lmv);
 530                         CERROR("%s: Target %s not attached: rc = %d\n",
 531                                obd->obd_name, uuidp->uuid, -EINVAL);
 532                         RETURN(-EINVAL);
 533                 }
 534         }
 535
 536         if ((index < lmv->tgts_size) && (lmv->tgts[index] != NULL)) {
 537                 tgt = lmv->tgts[index];
 538                 CERROR("%s: UUID %s already assigned at LOV target index %d:"
 539                        " rc = %d\n", obd->obd_name,
 540                        obd_uuid2str(&tgt->ltd_uuid), index, -EEXIST);
 541                 lmv_init_unlock(lmv);
 542                 RETURN(-EEXIST);
 543         }
 544
 545         if (index >= lmv->tgts_size) {
 546                 /* We need to reallocate the lmv target array. */
 547                 struct lmv_tgt_desc **newtgts, **old = NULL;
 548                 __u32 newsize = 1;
 549                 __u32 oldsize = 0;
 550
 551                 while (newsize < index + 1)
 552                         newsize = newsize << 1;
 553                 OBD_ALLOC(newtgts, sizeof(*newtgts) * newsize);
 554                 if (newtgts == NULL) {
 555                         lmv_init_unlock(lmv);
 556                         RETURN(-ENOMEM);
 557                 }
 558
 559                 if (lmv->tgts_size) {
 560                         memcpy(newtgts, lmv->tgts,
 561                                sizeof(*newtgts) * lmv->tgts_size);
 562                         old = lmv->tgts;
 563                         oldsize = lmv->tgts_size;
 564                 }
 565
 566                 lmv->tgts = newtgts;
 567                 lmv->tgts_size = newsize;
 568                 smp_rmb();
 569                 if (old)
 570                         OBD_FREE(old, sizeof(*old) * oldsize);
 571
 572                 CDEBUG(D_CONFIG, "tgts: %p size: %d\n", lmv->tgts,
 573                        lmv->tgts_size);
 574         }
 575
 576         OBD_ALLOC_PTR(tgt);
 577         if (!tgt) {
 578                 lmv_init_unlock(lmv);
 579                 RETURN(-ENOMEM);
 580         }
 581
 582         mutex_init(&tgt->ltd_fid_mutex);
 583         tgt->ltd_idx = index;
 584         tgt->ltd_uuid = *uuidp;
 585         tgt->ltd_active = 0;
 586         lmv->tgts[index] = tgt;
 587         if (index >= lmv->desc.ld_tgt_count) {
 588                 orig_tgt_count = lmv->desc.ld_tgt_count;
 589                 lmv->desc.ld_tgt_count = index + 1;
 590         }
 591
 592         if (lmv->connected) {
 593                 rc = lmv_connect_mdc(obd, tgt);
 594                 if (rc != 0) {
 595                         spin_lock(&lmv->lmv_lock);
 596                         if (lmv->desc.ld_tgt_count == index + 1)
 597                                 lmv->desc.ld_tgt_count = orig_tgt_count;
 598                         memset(tgt, 0, sizeof(*tgt));
 599                         spin_unlock(&lmv->lmv_lock);
 600                 } else {
 601                         int easize = sizeof(struct lmv_stripe_md) +
 602                                 lmv->desc.ld_tgt_count * sizeof(struct lu_fid);
 603                         lmv_init_ea_size(obd->obd_self_export, easize, 0, 0, 0);
 604                 }
 605         }
 606
 607         lmv_init_unlock(lmv);
 608         RETURN(rc);
 609 }
 610
 611 int lmv_check_connect(struct obd_device *obd)
 612 {
 613         struct lmv_obd          *lmv = &obd->u.lmv;
 614         struct lmv_tgt_desc     *tgt;
 615         __u32                    i;
 616         int                      rc;
 617         int                      easize;
 618         ENTRY;
 619
 620         if (lmv->connected)
 621                 RETURN(0);
 622
 623         lmv_init_lock(lmv);
 624         if (lmv->connected) {
 625                 lmv_init_unlock(lmv);
 626                 RETURN(0);
 627         }
 628
 629         if (lmv->desc.ld_tgt_count == 0) {
 630                 lmv_init_unlock(lmv);
 631                 CERROR("%s: no targets configured.\n", obd->obd_name);
 632                 RETURN(-EINVAL);
 633         }
 634
 635         LASSERT(lmv->tgts != NULL);
 636
 637         if (lmv->tgts[0] == NULL) {
 638                 lmv_init_unlock(lmv);
 639                 CERROR("%s: no target configured for index 0.\n",
 640                        obd->obd_name);
 641                 RETURN(-EINVAL);
 642         }
 643
 644         CDEBUG(D_CONFIG, "Time to connect %s to %s\n",
 645                lmv->cluuid.uuid, obd->obd_name);
 646
 647         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 648                 tgt = lmv->tgts[i];
 649                 if (tgt == NULL)
 650                         continue;
 651                 rc = lmv_connect_mdc(obd, tgt);
 652                 if (rc)
 653                         GOTO(out_disc, rc);
 654         }
 655
 656         lmv_set_timeouts(obd);
 657         class_export_put(lmv->exp);
 658         lmv->connected = 1;
 659         easize = lmv_mds_md_size(lmv->desc.ld_tgt_count, LMV_MAGIC);
 660         lmv_init_ea_size(obd->obd_self_export, easize, 0, 0, 0);
 661         lmv_init_unlock(lmv);
 662         RETURN(0);
 663
 664  out_disc:
 665         while (i-- > 0) {
 666                 int rc2;
 667                 tgt = lmv->tgts[i];
 668                 if (tgt == NULL)
 669                         continue;
 670                 tgt->ltd_active = 0;
 671                 if (tgt->ltd_exp) {
 672                         --lmv->desc.ld_active_tgt_count;
 673                         rc2 = obd_disconnect(tgt->ltd_exp);
 674                         if (rc2) {
 675                                 CERROR("LMV target %s disconnect on "
 676                                        "MDC idx %d: error %d\n",
 677                                        tgt->ltd_uuid.uuid, i, rc2);
 678                         }
 679                 }
 680         }
 681         class_disconnect(lmv->exp);
 682         lmv_init_unlock(lmv);
 683         RETURN(rc);
 684 }
 685
 686 static int lmv_disconnect_mdc(struct obd_device *obd, struct lmv_tgt_desc *tgt)
 687 {
 688         struct lmv_obd         *lmv = &obd->u.lmv;
 689         struct obd_device      *mdc_obd;
 690         int                     rc;
 691         ENTRY;
 692
 693         LASSERT(tgt != NULL);
 694         LASSERT(obd != NULL);
 695
 696         mdc_obd = class_exp2obd(tgt->ltd_exp);
 697
 698         if (mdc_obd) {
 699                 mdc_obd->obd_force = obd->obd_force;
 700                 mdc_obd->obd_fail = obd->obd_fail;
 701                 mdc_obd->obd_no_recov = obd->obd_no_recov;
 702         }
 703
 704         if (lmv->targets_proc_entry != NULL)
 705                 lprocfs_remove_proc_entry(mdc_obd->obd_name,
 706                                           lmv->targets_proc_entry);
 707
 708         rc = obd_fid_fini(tgt->ltd_exp->exp_obd);
 709         if (rc)
 710                 CERROR("Can't finanize fids factory\n");
 711
 712         CDEBUG(D_INFO, "Disconnected from %s(%s) successfully\n",
 713                tgt->ltd_exp->exp_obd->obd_name,
 714                tgt->ltd_exp->exp_obd->obd_uuid.uuid);
 715
 716         obd_register_observer(tgt->ltd_exp->exp_obd, NULL);
 717         rc = obd_disconnect(tgt->ltd_exp);
 718         if (rc) {
 719                 if (tgt->ltd_active) {
 720                         CERROR("Target %s disconnect error %d\n",
 721                                tgt->ltd_uuid.uuid, rc);
 722                 }
 723         }
 724
 725         lmv_activate_target(lmv, tgt, 0);
 726         tgt->ltd_exp = NULL;
 727         RETURN(0);
 728 }
 729
 730 static int lmv_disconnect(struct obd_export *exp)
 731 {
 732         struct obd_device       *obd = class_exp2obd(exp);
 733         struct lmv_obd          *lmv = &obd->u.lmv;
 734         int                      rc;
 735         __u32                    i;
 736         ENTRY;
 737
 738         if (!lmv->tgts)
 739                 goto out_local;
 740
 741         /*
 742          * Only disconnect the underlying layers on the final disconnect.
 743          */
 744         lmv->refcount--;
 745         if (lmv->refcount != 0)
 746                 goto out_local;
 747
 748         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 749                 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
 750                         continue;
 751
 752                 lmv_disconnect_mdc(obd, lmv->tgts[i]);
 753         }
 754
 755         if (lmv->targets_proc_entry != NULL)
 756                 lprocfs_remove(&lmv->targets_proc_entry);
 757         else
 758                 CERROR("/proc/fs/lustre/%s/%s/target_obds missing\n",
 759                        obd->obd_type->typ_name, obd->obd_name);
 760
 761 out_local:
 762         /*
 763          * This is the case when no real connection is established by
 764          * lmv_check_connect().
 765          */
 766         if (!lmv->connected)
 767                 class_export_put(exp);
 768         rc = class_disconnect(exp);
 769         if (lmv->refcount == 0)
 770                 lmv->connected = 0;
 771         RETURN(rc);
 772 }
 773
 774 static int lmv_fid2path(struct obd_export *exp, int len, void *karg, void *uarg)
 775 {
 776         struct obd_device       *obddev = class_exp2obd(exp);
 777         struct lmv_obd          *lmv = &obddev->u.lmv;
 778         struct getinfo_fid2path *gf;
 779         struct lmv_tgt_desc     *tgt;
 780         struct getinfo_fid2path *remote_gf = NULL;
 781         int                     remote_gf_size = 0;
 782         int                     rc;
 783
 784         gf = (struct getinfo_fid2path *)karg;
 785         tgt = lmv_find_target(lmv, &gf->gf_fid);
 786         if (IS_ERR(tgt))
 787                 RETURN(PTR_ERR(tgt));
 788
 789 repeat_fid2path:
 790         rc = obd_iocontrol(OBD_IOC_FID2PATH, tgt->ltd_exp, len, gf, uarg);
 791         if (rc != 0 && rc != -EREMOTE)
 792                 GOTO(out_fid2path, rc);
 793
 794         /* If remote_gf != NULL, it means just building the
 795          * path on the remote MDT, copy this path segement to gf */
 796         if (remote_gf != NULL) {
 797                 struct getinfo_fid2path *ori_gf;
 798                 char *ptr;
 799
 800                 ori_gf = (struct getinfo_fid2path *)karg;
 801                 if (strlen(ori_gf->gf_path) +
 802                     strlen(gf->gf_path) > ori_gf->gf_pathlen)
 803                         GOTO(out_fid2path, rc = -EOVERFLOW);
 804
 805                 ptr = ori_gf->gf_path;
 806
 807                 memmove(ptr + strlen(gf->gf_path) + 1, ptr,
 808                         strlen(ori_gf->gf_path));
 809
 810                 strncpy(ptr, gf->gf_path, strlen(gf->gf_path));
 811                 ptr += strlen(gf->gf_path);
 812                 *ptr = '/';
 813         }
 814
 815         CDEBUG(D_INFO, "%s: get path %s "DFID" rec: "LPU64" ln: %u\n",
 816                tgt->ltd_exp->exp_obd->obd_name,
 817                gf->gf_path, PFID(&gf->gf_fid), gf->gf_recno,
 818                gf->gf_linkno);
 819
 820         if (rc == 0)
 821                 GOTO(out_fid2path, rc);
 822
 823         /* sigh, has to go to another MDT to do path building further */
 824         if (remote_gf == NULL) {
 825                 remote_gf_size = sizeof(*remote_gf) + PATH_MAX;
 826                 OBD_ALLOC(remote_gf, remote_gf_size);
 827                 if (remote_gf == NULL)
 828                         GOTO(out_fid2path, rc = -ENOMEM);
 829                 remote_gf->gf_pathlen = PATH_MAX;
 830         }
 831
 832         if (!fid_is_sane(&gf->gf_fid)) {
 833                 CERROR("%s: invalid FID "DFID": rc = %d\n",
 834                        tgt->ltd_exp->exp_obd->obd_name,
 835                        PFID(&gf->gf_fid), -EINVAL);
 836                 GOTO(out_fid2path, rc = -EINVAL);
 837         }
 838
 839         tgt = lmv_find_target(lmv, &gf->gf_fid);
 840         if (IS_ERR(tgt))
 841                 GOTO(out_fid2path, rc = -EINVAL);
 842
 843         remote_gf->gf_fid = gf->gf_fid;
 844         remote_gf->gf_recno = -1;
 845         remote_gf->gf_linkno = -1;
 846         memset(remote_gf->gf_path, 0, remote_gf->gf_pathlen);
 847         gf = remote_gf;
 848         goto repeat_fid2path;
 849
 850 out_fid2path:
 851         if (remote_gf != NULL)
 852                 OBD_FREE(remote_gf, remote_gf_size);
 853         RETURN(rc);
 854 }
 855
 856 static int lmv_hsm_req_count(struct lmv_obd *lmv,
 857                              const struct hsm_user_request *hur,
 858                              const struct lmv_tgt_desc *tgt_mds)
 859 {
 860         __u32                    i;
 861         int                      nr = 0;
 862         struct lmv_tgt_desc     *curr_tgt;
 863
 864         /* count how many requests must be sent to the given target */
 865         for (i = 0; i < hur->hur_request.hr_itemcount; i++) {
 866                 curr_tgt = lmv_find_target(lmv, &hur->hur_user_item[i].hui_fid);
 867                 if (obd_uuid_equals(&curr_tgt->ltd_uuid, &tgt_mds->ltd_uuid))
 868                         nr++;
 869         }
 870         return nr;
 871 }
 872
 873 static void lmv_hsm_req_build(struct lmv_obd *lmv,
 874                               struct hsm_user_request *hur_in,
 875                               const struct lmv_tgt_desc *tgt_mds,
 876                               struct hsm_user_request *hur_out)
 877 {
 878         __u32                    i, nr_out;
 879         struct lmv_tgt_desc     *curr_tgt;
 880
 881         /* build the hsm_user_request for the given target */
 882         hur_out->hur_request = hur_in->hur_request;
 883         nr_out = 0;
 884         for (i = 0; i < hur_in->hur_request.hr_itemcount; i++) {
 885                 curr_tgt = lmv_find_target(lmv,
 886                                            &hur_in->hur_user_item[i].hui_fid);
 887                 if (obd_uuid_equals(&curr_tgt->ltd_uuid, &tgt_mds->ltd_uuid)) {
 888                         hur_out->hur_user_item[nr_out] =
 889                                                 hur_in->hur_user_item[i];
 890                         nr_out++;
 891                 }
 892         }
 893         hur_out->hur_request.hr_itemcount = nr_out;
 894         memcpy(hur_data(hur_out), hur_data(hur_in),
 895                hur_in->hur_request.hr_data_len);
 896 }
 897
 898 static int lmv_hsm_ct_unregister(struct lmv_obd *lmv, unsigned int cmd, int len,
 899                                  struct lustre_kernelcomm *lk, void *uarg)
 900 {
 901         __u32                    i;
 902         int                      rc;
 903         struct kkuc_ct_data     *kcd = NULL;
 904         ENTRY;
 905
 906         /* unregister request (call from llapi_hsm_copytool_fini) */
 907         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 908                 struct lmv_tgt_desc *tgt = lmv->tgts[i];
 909
 910                 if (tgt == NULL || tgt->ltd_exp == NULL)
 911                         continue;
 912                 /* best effort: try to clean as much as possible
 913                  * (continue on error) */
 914                 obd_iocontrol(cmd, tgt->ltd_exp, len, lk, uarg);
 915         }
 916
 917         /* Whatever the result, remove copytool from kuc groups.
 918          * Unreached coordinators will get EPIPE on next requests
 919          * and will unregister automatically.
 920          */
 921         rc = libcfs_kkuc_group_rem(lk->lk_uid, lk->lk_group, (void **)&kcd);
 922         if (kcd != NULL)
 923                 OBD_FREE_PTR(kcd);
 924
 925         RETURN(rc);
 926 }
 927
 928 static int lmv_hsm_ct_register(struct lmv_obd *lmv, unsigned int cmd, int len,
 929                                struct lustre_kernelcomm *lk, void *uarg)
 930 {
 931         struct file             *filp;
 932         __u32                    i, j;
 933         int                      err, rc;
 934         bool                     any_set = false;
 935         struct kkuc_ct_data     *kcd;
 936         ENTRY;
 937
 938         /* All or nothing: try to register to all MDS.
 939          * In case of failure, unregister from previous MDS,
 940          * except if it because of inactive target. */
 941         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 942                 struct lmv_tgt_desc *tgt = lmv->tgts[i];
 943
 944                 if (tgt == NULL || tgt->ltd_exp == NULL)
 945                         continue;
 946                 err = obd_iocontrol(cmd, tgt->ltd_exp, len, lk, uarg);
 947                 if (err) {
 948                         if (tgt->ltd_active) {
 949                                 /* permanent error */
 950                                 CERROR("%s: iocontrol MDC %s on MDT"
 951                                        " idx %d cmd %x: err = %d\n",
 952                                        class_exp2obd(lmv->exp)->obd_name,
 953                                        tgt->ltd_uuid.uuid, i, cmd, err);
 954                                 rc = err;
 955                                 lk->lk_flags |= LK_FLG_STOP;
 956                                 /* unregister from previous MDS */
 957                                 for (j = 0; j < i; j++) {
 958                                         tgt = lmv->tgts[j];
 959                                         if (tgt == NULL || tgt->ltd_exp == NULL)
 960                                                 continue;
 961                                         obd_iocontrol(cmd, tgt->ltd_exp, len,
 962                                                       lk, uarg);
 963                                 }
 964                                 RETURN(rc);
 965                         }
 966                         /* else: transient error.
 967                          * kuc will register to the missing MDT
 968                          * when it is back */
 969                 } else {
 970                         any_set = true;
 971                 }
 972         }
 973
 974         if (!any_set)
 975                 /* no registration done: return error */
 976                 RETURN(-ENOTCONN);
 977
 978         /* at least one registration done, with no failure */
 979         filp = fget(lk->lk_wfd);
 980         if (filp == NULL)
 981                 RETURN(-EBADF);
 982
 983         OBD_ALLOC_PTR(kcd);
 984         if (kcd == NULL) {
 985                 fput(filp);
 986                 RETURN(-ENOMEM);
 987         }
 988         kcd->kcd_magic = KKUC_CT_DATA_MAGIC;
 989         kcd->kcd_uuid = lmv->cluuid;
 990         kcd->kcd_archive = lk->lk_data;
 991
 992         rc = libcfs_kkuc_group_add(filp, lk->lk_uid, lk->lk_group, kcd);
 993         if (rc != 0) {
 994                 if (filp != NULL)
 995                         fput(filp);
 996                 OBD_FREE_PTR(kcd);
 997         }
 998
 999         RETURN(rc);
1000 }
1001
1002
1003
1004
1005 static int lmv_iocontrol(unsigned int cmd, struct obd_export *exp,
1006                          int len, void *karg, void *uarg)
1007 {
1008         struct obd_device       *obddev = class_exp2obd(exp);
1009         struct lmv_obd          *lmv = &obddev->u.lmv;
1010         struct lmv_tgt_desc     *tgt = NULL;
1011         __u32                    i = 0;
1012         int                      rc = 0;
1013         int                      set = 0;
1014         __u32                    count = lmv->desc.ld_tgt_count;
1015         ENTRY;
1016
1017         if (count == 0)
1018                 RETURN(-ENOTTY);
1019
1020         switch (cmd) {
1021         case IOC_OBD_STATFS: {
1022                 struct obd_ioctl_data *data = karg;
1023                 struct obd_device *mdc_obd;
1024                 struct obd_statfs stat_buf = {0};
1025                 __u32 index;
1026
1027                 memcpy(&index, data->ioc_inlbuf2, sizeof(__u32));
1028                 if ((index >= count))
1029                         RETURN(-ENODEV);
1030
1031                 tgt = lmv->tgts[index];
1032                 if (tgt == NULL || !tgt->ltd_active)
1033                         RETURN(-ENODATA);
1034
1035                 mdc_obd = class_exp2obd(tgt->ltd_exp);
1036                 if (!mdc_obd)
1037                         RETURN(-EINVAL);
1038
1039                 /* copy UUID */
1040                 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(mdc_obd),
1041                                  min((int) data->ioc_plen2,
1042                                      (int) sizeof(struct obd_uuid))))
1043                         RETURN(-EFAULT);
1044
1045                 rc = obd_statfs(NULL, tgt->ltd_exp, &stat_buf,
1046                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1047                                 0);
1048                 if (rc)
1049                         RETURN(rc);
1050                 if (copy_to_user(data->ioc_pbuf1, &stat_buf,
1051                                  min((int) data->ioc_plen1,
1052                                      (int) sizeof(stat_buf))))
1053                         RETURN(-EFAULT);
1054                 break;
1055         }
1056         case OBD_IOC_QUOTACTL: {
1057                 struct if_quotactl *qctl = karg;
1058                 struct obd_quotactl *oqctl;
1059
1060                 if (qctl->qc_valid == QC_MDTIDX) {
1061                         if (count <= qctl->qc_idx)
1062                                 RETURN(-EINVAL);
1063
1064                         tgt = lmv->tgts[qctl->qc_idx];
1065                         if (tgt == NULL || tgt->ltd_exp == NULL)
1066                                 RETURN(-EINVAL);
1067                 } else if (qctl->qc_valid == QC_UUID) {
1068                         for (i = 0; i < count; i++) {
1069                                 tgt = lmv->tgts[i];
1070                                 if (tgt == NULL)
1071                                         continue;
1072                                 if (!obd_uuid_equals(&tgt->ltd_uuid,
1073                                                      &qctl->obd_uuid))
1074                                         continue;
1075
1076                                 if (tgt->ltd_exp == NULL)
1077                                         RETURN(-EINVAL);
1078
1079                                 break;
1080                         }
1081                 } else {
1082                         RETURN(-EINVAL);
1083                 }
1084
1085                 if (i >= count)
1086                         RETURN(-EAGAIN);
1087
1088                 LASSERT(tgt != NULL && tgt->ltd_exp != NULL);
1089                 OBD_ALLOC_PTR(oqctl);
1090                 if (!oqctl)
1091                         RETURN(-ENOMEM);
1092
1093                 QCTL_COPY(oqctl, qctl);
1094                 rc = obd_quotactl(tgt->ltd_exp, oqctl);
1095                 if (rc == 0) {
1096                         QCTL_COPY(qctl, oqctl);
1097                         qctl->qc_valid = QC_MDTIDX;
1098                         qctl->obd_uuid = tgt->ltd_uuid;
1099                 }
1100                 OBD_FREE_PTR(oqctl);
1101                 break;
1102         }
1103         case OBD_IOC_CHANGELOG_SEND:
1104         case OBD_IOC_CHANGELOG_CLEAR: {
1105                 struct ioc_changelog *icc = karg;
1106
1107                 if (icc->icc_mdtindex >= count)
1108                         RETURN(-ENODEV);
1109
1110                 tgt = lmv->tgts[icc->icc_mdtindex];
1111                 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active)
1112                         RETURN(-ENODEV);
1113                 rc = obd_iocontrol(cmd, tgt->ltd_exp, sizeof(*icc), icc, NULL);
1114                 break;
1115         }
1116         case LL_IOC_GET_CONNECT_FLAGS: {
1117                 tgt = lmv->tgts[0];
1118                 if (tgt == NULL || tgt->ltd_exp == NULL)
1119                         RETURN(-ENODATA);
1120                 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1121                 break;
1122         }
1123         case LL_IOC_FID2MDTIDX: {
1124                 struct lu_fid *fid = karg;
1125                 int             mdt_index;
1126
1127                 rc = lmv_fld_lookup(lmv, fid, &mdt_index);
1128                 if (rc != 0)
1129                         RETURN(rc);
1130
1131                 /* Note: this is from llite(see ll_dir_ioctl()), @uarg does not
1132                  * point to user space memory for FID2MDTIDX. */
1133                 *(__u32 *)uarg = mdt_index;
1134                 break;
1135         }
1136         case OBD_IOC_FID2PATH: {
1137                 rc = lmv_fid2path(exp, len, karg, uarg);
1138                 break;
1139         }
1140         case LL_IOC_HSM_STATE_GET:
1141         case LL_IOC_HSM_STATE_SET:
1142         case LL_IOC_HSM_ACTION: {
1143                 struct md_op_data       *op_data = karg;
1144
1145                 tgt = lmv_find_target(lmv, &op_data->op_fid1);
1146                 if (IS_ERR(tgt))
1147                         RETURN(PTR_ERR(tgt));
1148
1149                 if (tgt->ltd_exp == NULL)
1150                         RETURN(-EINVAL);
1151
1152                 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1153                 break;
1154         }
1155         case LL_IOC_HSM_PROGRESS: {
1156                 const struct hsm_progress_kernel *hpk = karg;
1157
1158                 tgt = lmv_find_target(lmv, &hpk->hpk_fid);
1159                 if (IS_ERR(tgt))
1160                         RETURN(PTR_ERR(tgt));
1161                 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1162                 break;
1163         }
1164         case LL_IOC_HSM_REQUEST: {
1165                 struct hsm_user_request *hur = karg;
1166                 unsigned int reqcount = hur->hur_request.hr_itemcount;
1167
1168                 if (reqcount == 0)
1169                         RETURN(0);
1170
1171                 /* if the request is about a single fid
1172                  * or if there is a single MDS, no need to split
1173                  * the request. */
1174                 if (reqcount == 1 || count == 1) {
1175                         tgt = lmv_find_target(lmv,
1176                                               &hur->hur_user_item[0].hui_fid);
1177                         if (IS_ERR(tgt))
1178                                 RETURN(PTR_ERR(tgt));
1179                         rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1180                 } else {
1181                         /* split fid list to their respective MDS */
1182                         for (i = 0; i < count; i++) {
1183                                 unsigned int            nr, reqlen;
1184                                 int                     rc1;
1185                                 struct hsm_user_request *req;
1186
1187                                 tgt = lmv->tgts[i];
1188                                 if (tgt == NULL || tgt->ltd_exp == NULL)
1189                                         continue;
1190
1191                                 nr = lmv_hsm_req_count(lmv, hur, tgt);
1192                                 if (nr == 0) /* nothing for this MDS */
1193                                         continue;
1194
1195                                 /* build a request with fids for this MDS */
1196                                 reqlen = offsetof(typeof(*hur),
1197                                                   hur_user_item[nr])
1198                                                 + hur->hur_request.hr_data_len;
1199                                 OBD_ALLOC_LARGE(req, reqlen);
1200                                 if (req == NULL)
1201                                         RETURN(-ENOMEM);
1202
1203                                 lmv_hsm_req_build(lmv, hur, tgt, req);
1204
1205                                 rc1 = obd_iocontrol(cmd, tgt->ltd_exp, reqlen,
1206                                                     req, uarg);
1207                                 if (rc1 != 0 && rc == 0)
1208                                         rc = rc1;
1209                                 OBD_FREE_LARGE(req, reqlen);
1210                         }
1211                 }
1212                 break;
1213         }
1214         case LL_IOC_LOV_SWAP_LAYOUTS: {
1215                 struct md_op_data       *op_data = karg;
1216                 struct lmv_tgt_desc     *tgt1, *tgt2;
1217
1218                 tgt1 = lmv_find_target(lmv, &op_data->op_fid1);
1219                 if (IS_ERR(tgt1))
1220                         RETURN(PTR_ERR(tgt1));
1221
1222                 tgt2 = lmv_find_target(lmv, &op_data->op_fid2);
1223                 if (IS_ERR(tgt2))
1224                         RETURN(PTR_ERR(tgt2));
1225
1226                 if ((tgt1->ltd_exp == NULL) || (tgt2->ltd_exp == NULL))
1227                         RETURN(-EINVAL);
1228
1229                 /* only files on same MDT can have their layouts swapped */
1230                 if (tgt1->ltd_idx != tgt2->ltd_idx)
1231                         RETURN(-EPERM);
1232
1233                 rc = obd_iocontrol(cmd, tgt1->ltd_exp, len, karg, uarg);
1234                 break;
1235         }
1236         case LL_IOC_HSM_CT_START: {
1237                 struct lustre_kernelcomm *lk = karg;
1238                 if (lk->lk_flags & LK_FLG_STOP)
1239                         rc = lmv_hsm_ct_unregister(lmv, cmd, len, lk, uarg);
1240                 else
1241                         rc = lmv_hsm_ct_register(lmv, cmd, len, lk, uarg);
1242                 break;
1243         }
1244         default:
1245                 for (i = 0; i < count; i++) {
1246                         struct obd_device *mdc_obd;
1247                         int err;
1248
1249                         tgt = lmv->tgts[i];
1250                         if (tgt == NULL || tgt->ltd_exp == NULL)
1251                                 continue;
1252                         /* ll_umount_begin() sets force flag but for lmv, not
1253                          * mdc. Let's pass it through */
1254                         mdc_obd = class_exp2obd(tgt->ltd_exp);
1255                         mdc_obd->obd_force = obddev->obd_force;
1256                         err = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1257                         if (err == -ENODATA && cmd == OBD_IOC_POLL_QUOTACHECK) {
1258                                 RETURN(err);
1259                         } else if (err) {
1260                                 if (tgt->ltd_active) {
1261                                         CERROR("error: iocontrol MDC %s on MDT"
1262                                                " idx %d cmd %x: err = %d\n",
1263                                                tgt->ltd_uuid.uuid, i, cmd, err);
1264                                         if (!rc)
1265                                                 rc = err;
1266                                 }
1267                         } else
1268                                 set = 1;
1269                 }
1270                 if (!set && !rc)
1271                         rc = -EIO;
1272         }
1273         RETURN(rc);
1274 }
1275
1276 #if 0
1277 static int lmv_all_chars_policy(int count, const char *name,
1278                                 int len)
1279 {
1280         unsigned int c = 0;
1281
1282         while (len > 0)
1283                 c += name[--len];
1284         c = c % count;
1285         return c;
1286 }
1287
1288 static int lmv_nid_policy(struct lmv_obd *lmv)
1289 {
1290         struct obd_import *imp;
1291         __u32              id;
1292
1293         /*
1294          * XXX: To get nid we assume that underlying obd device is mdc.
1295          */
1296         imp = class_exp2cliimp(lmv->tgts[0].ltd_exp);
1297         id = imp->imp_connection->c_self ^ (imp->imp_connection->c_self >> 32);
1298         return id % lmv->desc.ld_tgt_count;
1299 }
1300
1301 static int lmv_choose_mds(struct lmv_obd *lmv, struct md_op_data *op_data,
1302                           placement_policy_t placement)
1303 {
1304         switch (placement) {
1305         case PLACEMENT_CHAR_POLICY:
1306                 return lmv_all_chars_policy(lmv->desc.ld_tgt_count,
1307                                             op_data->op_name,
1308                                             op_data->op_namelen);
1309         case PLACEMENT_NID_POLICY:
1310                 return lmv_nid_policy(lmv);
1311
1312         default:
1313                 break;
1314         }
1315
1316         CERROR("Unsupported placement policy %x\n", placement);
1317         return -EINVAL;
1318 }
1319 #endif
1320
1321 /**
1322  * This is _inode_ placement policy function (not name).
1323  */
1324 static int lmv_placement_policy(struct obd_device *obd,
1325                                 struct md_op_data *op_data, u32 *mds)
1326 {
1327         struct lmv_obd          *lmv = &obd->u.lmv;
1328         ENTRY;
1329
1330         LASSERT(mds != NULL);
1331
1332         if (lmv->desc.ld_tgt_count == 1) {
1333                 *mds = 0;
1334                 RETURN(0);
1335         }
1336
1337         /**
1338          * If stripe_offset is provided during setdirstripe
1339          * (setdirstripe -i xx), xx MDS will be choosen.
1340          */
1341         if (op_data->op_cli_flags & CLI_SET_MEA && op_data->op_data != NULL) {
1342                 struct lmv_user_md *lum;
1343
1344                 lum = op_data->op_data;
1345
1346                 if (le32_to_cpu(lum->lum_stripe_offset) != (__u32)-1) {
1347                         *mds = le32_to_cpu(lum->lum_stripe_offset);
1348                 } else {
1349                         /* -1 means default, which will be in the same MDT with
1350                          * the stripe */
1351                         *mds = op_data->op_mds;
1352                         lum->lum_stripe_offset = cpu_to_le32(op_data->op_mds);
1353                 }
1354         } else {
1355                 /* Allocate new fid on target according to operation type and
1356                  * parent home mds. */
1357                 *mds = op_data->op_mds;
1358         }
1359
1360         RETURN(0);
1361 }
1362
1363 int __lmv_fid_alloc(struct lmv_obd *lmv, struct lu_fid *fid, u32 mds)
1364 {
1365         struct lmv_tgt_desc     *tgt;
1366         int                      rc;
1367         ENTRY;
1368
1369         tgt = lmv_get_target(lmv, mds, NULL);
1370         if (IS_ERR(tgt))
1371                 RETURN(PTR_ERR(tgt));
1372
1373         /*
1374          * New seq alloc and FLD setup should be atomic. Otherwise we may find
1375          * on server that seq in new allocated fid is not yet known.
1376          */
1377         mutex_lock(&tgt->ltd_fid_mutex);
1378
1379         if (tgt->ltd_active == 0 || tgt->ltd_exp == NULL)
1380                 GOTO(out, rc = -ENODEV);
1381
1382         /*
1383          * Asking underlying tgt layer to allocate new fid.
1384          */
1385         rc = obd_fid_alloc(NULL, tgt->ltd_exp, fid, NULL);
1386         if (rc > 0) {
1387                 LASSERT(fid_is_sane(fid));
1388                 rc = 0;
1389         }
1390
1391         EXIT;
1392 out:
1393         mutex_unlock(&tgt->ltd_fid_mutex);
1394         return rc;
1395 }
1396
1397 int lmv_fid_alloc(const struct lu_env *env, struct obd_export *exp,
1398                   struct lu_fid *fid, struct md_op_data *op_data)
1399 {
1400         struct obd_device     *obd = class_exp2obd(exp);
1401         struct lmv_obd        *lmv = &obd->u.lmv;
1402         u32                    mds = 0;
1403         int                    rc;
1404         ENTRY;
1405
1406         LASSERT(op_data != NULL);
1407         LASSERT(fid != NULL);
1408
1409         rc = lmv_placement_policy(obd, op_data, &mds);
1410         if (rc) {
1411                 CERROR("Can't get target for allocating fid, "
1412                        "rc %d\n", rc);
1413                 RETURN(rc);
1414         }
1415
1416         rc = __lmv_fid_alloc(lmv, fid, mds);
1417         if (rc) {
1418                 CERROR("Can't alloc new fid, rc %d\n", rc);
1419                 RETURN(rc);
1420         }
1421
1422         RETURN(rc);
1423 }
1424
1425 static int lmv_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
1426 {
1427         struct lmv_obd  *lmv = &obd->u.lmv;
1428         struct lmv_desc *desc;
1429         int             rc;
1430         ENTRY;
1431
1432         if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
1433                 CERROR("LMV setup requires a descriptor\n");
1434                 RETURN(-EINVAL);
1435         }
1436
1437         desc = (struct lmv_desc *)lustre_cfg_buf(lcfg, 1);
1438         if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
1439                 CERROR("Lmv descriptor size wrong: %d > %d\n",
1440                        (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
1441                 RETURN(-EINVAL);
1442         }
1443
1444         OBD_ALLOC(lmv->tgts, sizeof(*lmv->tgts) * 32);
1445         if (lmv->tgts == NULL)
1446                 RETURN(-ENOMEM);
1447         lmv->tgts_size = 32;
1448
1449         obd_str2uuid(&lmv->desc.ld_uuid, desc->ld_uuid.uuid);
1450         lmv->desc.ld_tgt_count = 0;
1451         lmv->desc.ld_active_tgt_count = 0;
1452         lmv->max_cookiesize = 0;
1453         lmv->max_def_easize = 0;
1454         lmv->max_easize = 0;
1455         lmv->lmv_placement = PLACEMENT_CHAR_POLICY;
1456
1457         spin_lock_init(&lmv->lmv_lock);
1458         mutex_init(&lmv->init_mutex);
1459
1460 #ifdef LPROCFS
1461         obd->obd_vars = lprocfs_lmv_obd_vars;
1462         lprocfs_obd_setup(obd);
1463         lprocfs_alloc_md_stats(obd, 0);
1464         rc = lprocfs_seq_create(obd->obd_proc_entry, "target_obd",
1465                                 0444, &lmv_proc_target_fops, obd);
1466         if (rc)
1467                 CWARN("%s: error adding LMV target_obd file: rc = %d\n",
1468                       obd->obd_name, rc);
1469 #endif
1470         rc = fld_client_init(&lmv->lmv_fld, obd->obd_name,
1471                              LUSTRE_CLI_FLD_HASH_DHT);
1472         if (rc) {
1473                 CERROR("Can't init FLD, err %d\n", rc);
1474                 GOTO(out, rc);
1475         }
1476
1477         RETURN(0);
1478
1479 out:
1480         return rc;
1481 }
1482
1483 static int lmv_cleanup(struct obd_device *obd)
1484 {
1485         struct lmv_obd   *lmv = &obd->u.lmv;
1486         ENTRY;
1487
1488         fld_client_fini(&lmv->lmv_fld);
1489         if (lmv->tgts != NULL) {
1490                 int i;
1491                 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1492                         if (lmv->tgts[i] == NULL)
1493                                 continue;
1494                         lmv_del_target(lmv, i);
1495                 }
1496                 OBD_FREE(lmv->tgts, sizeof(*lmv->tgts) * lmv->tgts_size);
1497                 lmv->tgts_size = 0;
1498         }
1499         RETURN(0);
1500 }
1501
1502 static int lmv_process_config(struct obd_device *obd, obd_count len, void *buf)
1503 {
1504         struct lustre_cfg       *lcfg = buf;
1505         struct obd_uuid         obd_uuid;
1506         int                     gen;
1507         __u32                   index;
1508         int                     rc;
1509         ENTRY;
1510
1511         switch (lcfg->lcfg_command) {
1512         case LCFG_ADD_MDC:
1513                 /* modify_mdc_tgts add 0:lustre-clilmv  1:lustre-MDT0000_UUID
1514                  * 2:0  3:1  4:lustre-MDT0000-mdc_UUID */
1515                 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(obd_uuid.uuid))
1516                         GOTO(out, rc = -EINVAL);
1517
1518                 obd_str2uuid(&obd_uuid,  lustre_cfg_buf(lcfg, 1));
1519
1520                 if (sscanf(lustre_cfg_buf(lcfg, 2), "%u", &index) != 1)
1521                         GOTO(out, rc = -EINVAL);
1522                 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", &gen) != 1)
1523                         GOTO(out, rc = -EINVAL);
1524                 rc = lmv_add_target(obd, &obd_uuid, index, gen);
1525                 GOTO(out, rc);
1526         default:
1527                 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
1528                 GOTO(out, rc = -EINVAL);
1529         }
1530 out:
1531         RETURN(rc);
1532 }
1533
1534 static int lmv_statfs(const struct lu_env *env, struct obd_export *exp,
1535                       struct obd_statfs *osfs, __u64 max_age, __u32 flags)
1536 {
1537         struct obd_device       *obd = class_exp2obd(exp);
1538         struct lmv_obd          *lmv = &obd->u.lmv;
1539         struct obd_statfs       *temp;
1540         int                      rc = 0;
1541         __u32                    i;
1542         ENTRY;
1543
1544         rc = lmv_check_connect(obd);
1545         if (rc)
1546                 RETURN(rc);
1547
1548         OBD_ALLOC(temp, sizeof(*temp));
1549         if (temp == NULL)
1550                 RETURN(-ENOMEM);
1551
1552         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1553                 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
1554                         continue;
1555
1556                 rc = obd_statfs(env, lmv->tgts[i]->ltd_exp, temp,
1557                                 max_age, flags);
1558                 if (rc) {
1559                         CERROR("can't stat MDS #%d (%s), error %d\n", i,
1560                                lmv->tgts[i]->ltd_exp->exp_obd->obd_name,
1561                                rc);
1562                         GOTO(out_free_temp, rc);
1563                 }
1564
1565                 if (i == 0) {
1566                         *osfs = *temp;
1567                         /* If the statfs is from mount, it will needs
1568                          * retrieve necessary information from MDT0.
1569                          * i.e. mount does not need the merged osfs
1570                          * from all of MDT.
1571                          * And also clients can be mounted as long as
1572                          * MDT0 is in service*/
1573                         if (flags & OBD_STATFS_FOR_MDT0)
1574                                 GOTO(out_free_temp, rc);
1575                 } else {
1576                         osfs->os_bavail += temp->os_bavail;
1577                         osfs->os_blocks += temp->os_blocks;
1578                         osfs->os_ffree += temp->os_ffree;
1579                         osfs->os_files += temp->os_files;
1580                 }
1581         }
1582
1583         EXIT;
1584 out_free_temp:
1585         OBD_FREE(temp, sizeof(*temp));
1586         return rc;
1587 }
1588
1589 static int lmv_getstatus(struct obd_export *exp,
1590                          struct lu_fid *fid,
1591                          struct obd_capa **pc)
1592 {
1593         struct obd_device    *obd = exp->exp_obd;
1594         struct lmv_obd       *lmv = &obd->u.lmv;
1595         int                   rc;
1596         ENTRY;
1597
1598         rc = lmv_check_connect(obd);
1599         if (rc)
1600                 RETURN(rc);
1601
1602         rc = md_getstatus(lmv->tgts[0]->ltd_exp, fid, pc);
1603         RETURN(rc);
1604 }
1605
1606 static int lmv_getxattr(struct obd_export *exp, const struct lu_fid *fid,
1607                         struct obd_capa *oc, obd_valid valid, const char *name,
1608                         const char *input, int input_size, int output_size,
1609                         int flags, struct ptlrpc_request **request)
1610 {
1611         struct obd_device      *obd = exp->exp_obd;
1612         struct lmv_obd         *lmv = &obd->u.lmv;
1613         struct lmv_tgt_desc    *tgt;
1614         int                     rc;
1615         ENTRY;
1616
1617         rc = lmv_check_connect(obd);
1618         if (rc)
1619                 RETURN(rc);
1620
1621         tgt = lmv_find_target(lmv, fid);
1622         if (IS_ERR(tgt))
1623                 RETURN(PTR_ERR(tgt));
1624
1625         rc = md_getxattr(tgt->ltd_exp, fid, oc, valid, name, input,
1626                          input_size, output_size, flags, request);
1627
1628         RETURN(rc);
1629 }
1630
1631 static int lmv_setxattr(struct obd_export *exp, const struct lu_fid *fid,
1632                         struct obd_capa *oc, obd_valid valid, const char *name,
1633                         const char *input, int input_size, int output_size,
1634                         int flags, __u32 suppgid,
1635                         struct ptlrpc_request **request)
1636 {
1637         struct obd_device      *obd = exp->exp_obd;
1638         struct lmv_obd         *lmv = &obd->u.lmv;
1639         struct lmv_tgt_desc    *tgt;
1640         int                     rc;
1641         ENTRY;
1642
1643         rc = lmv_check_connect(obd);
1644         if (rc)
1645                 RETURN(rc);
1646
1647         tgt = lmv_find_target(lmv, fid);
1648         if (IS_ERR(tgt))
1649                 RETURN(PTR_ERR(tgt));
1650
1651         rc = md_setxattr(tgt->ltd_exp, fid, oc, valid, name, input,
1652                          input_size, output_size, flags, suppgid,
1653                          request);
1654
1655         RETURN(rc);
1656 }
1657
1658 static int lmv_getattr(struct obd_export *exp, struct md_op_data *op_data,
1659                        struct ptlrpc_request **request)
1660 {
1661         struct obd_device       *obd = exp->exp_obd;
1662         struct lmv_obd          *lmv = &obd->u.lmv;
1663         struct lmv_tgt_desc     *tgt;
1664         int                      rc;
1665         ENTRY;
1666
1667         rc = lmv_check_connect(obd);
1668         if (rc)
1669                 RETURN(rc);
1670
1671         tgt = lmv_find_target(lmv, &op_data->op_fid1);
1672         if (IS_ERR(tgt))
1673                 RETURN(PTR_ERR(tgt));
1674
1675         if (op_data->op_flags & MF_GET_MDT_IDX) {
1676                 op_data->op_mds = tgt->ltd_idx;
1677                 RETURN(0);
1678         }
1679
1680         rc = md_getattr(tgt->ltd_exp, op_data, request);
1681
1682         RETURN(rc);
1683 }
1684
1685 static int lmv_null_inode(struct obd_export *exp, const struct lu_fid *fid)
1686 {
1687         struct obd_device   *obd = exp->exp_obd;
1688         struct lmv_obd      *lmv = &obd->u.lmv;
1689         __u32                i;
1690         int                  rc;
1691         ENTRY;
1692
1693         rc = lmv_check_connect(obd);
1694         if (rc)
1695                 RETURN(rc);
1696
1697         CDEBUG(D_INODE, "CBDATA for "DFID"\n", PFID(fid));
1698
1699         /*
1700          * With DNE every object can have two locks in different namespaces:
1701          * lookup lock in space of MDT storing direntry and update/open lock in
1702          * space of MDT storing inode.
1703          */
1704         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1705                 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
1706                         continue;
1707                 md_null_inode(lmv->tgts[i]->ltd_exp, fid);
1708         }
1709
1710         RETURN(0);
1711 }
1712
1713 static int lmv_find_cbdata(struct obd_export *exp, const struct lu_fid *fid,
1714                            ldlm_iterator_t it, void *data)
1715 {
1716         struct obd_device       *obd = exp->exp_obd;
1717         struct lmv_obd          *lmv = &obd->u.lmv;
1718         int                     i;
1719         int                     tgt;
1720         int                     rc;
1721         ENTRY;
1722
1723         rc = lmv_check_connect(obd);
1724         if (rc)
1725                 RETURN(rc);
1726
1727         CDEBUG(D_INODE, "CBDATA for "DFID"\n", PFID(fid));
1728
1729         /*
1730          * With DNE every object can have two locks in different namespaces:
1731          * lookup lock in space of MDT storing direntry and update/open lock in
1732          * space of MDT storing inode.  Try the MDT that the FID maps to first,
1733          * since this can be easily found, and only try others if that fails.
1734          */
1735         for (i = 0, tgt = lmv_find_target_index(lmv, fid);
1736              i < lmv->desc.ld_tgt_count;
1737              i++, tgt = (tgt + 1) % lmv->desc.ld_tgt_count) {
1738                 if (tgt < 0) {
1739                         CDEBUG(D_HA, "%s: "DFID" is inaccessible: rc = %d\n",
1740                                obd->obd_name, PFID(fid), tgt);
1741                         tgt = 0;
1742                 }
1743
1744                 if (lmv->tgts[tgt] == NULL ||
1745                     lmv->tgts[tgt]->ltd_exp == NULL)
1746                         continue;
1747
1748                 rc = md_find_cbdata(lmv->tgts[tgt]->ltd_exp, fid, it, data);
1749                 if (rc)
1750                         RETURN(rc);
1751         }
1752
1753         RETURN(rc);
1754 }
1755
1756
1757 static int lmv_close(struct obd_export *exp, struct md_op_data *op_data,
1758                      struct md_open_data *mod, struct ptlrpc_request **request)
1759 {
1760         struct obd_device     *obd = exp->exp_obd;
1761         struct lmv_obd        *lmv = &obd->u.lmv;
1762         struct lmv_tgt_desc   *tgt;
1763         int                    rc;
1764         ENTRY;
1765
1766         rc = lmv_check_connect(obd);
1767         if (rc)
1768                 RETURN(rc);
1769
1770         tgt = lmv_find_target(lmv, &op_data->op_fid1);
1771         if (IS_ERR(tgt))
1772                 RETURN(PTR_ERR(tgt));
1773
1774         CDEBUG(D_INODE, "CLOSE "DFID"\n", PFID(&op_data->op_fid1));
1775         rc = md_close(tgt->ltd_exp, op_data, mod, request);
1776         RETURN(rc);
1777 }
1778
1779 /**
1780  * Choosing the MDT by name or FID in @op_data.
1781  * For non-striped directory, it will locate MDT by fid.
1782  * For striped-directory, it will locate MDT by name. And also
1783  * it will reset op_fid1 with the FID of the choosen stripe.
1784  **/
1785 struct lmv_tgt_desc *
1786 lmv_locate_target_for_name(struct lmv_obd *lmv, struct lmv_stripe_md *lsm,
1787                            const char *name, int namelen, struct lu_fid *fid,
1788                            u32 *mds)
1789 {
1790         struct lmv_tgt_desc     *tgt;
1791         const struct lmv_oinfo  *oinfo;
1792
1793         oinfo = lsm_name_to_stripe_info(lsm, name, namelen);
1794         if (IS_ERR(oinfo))
1795                 RETURN(ERR_CAST(oinfo));
1796         *fid = oinfo->lmo_fid;
1797         *mds = oinfo->lmo_mds;
1798         tgt = lmv_get_target(lmv, *mds, NULL);
1799
1800         CDEBUG(D_INFO, "locate on mds %u "DFID"\n", *mds, PFID(fid));
1801         return tgt;
1802 }
1803
1804 /**
1805  * Locate mds by fid or name
1806  *
1807  * For striped directory (lsm != NULL), it will locate the stripe
1808  * by name hash (see lsm_name_to_stripe_info()). Note: if the hash_type
1809  * is unknown, it will return -EBADFD, and lmv_intent_lookup might need
1810  * walk through all of stripes to locate the entry.
1811  *
1812  * For normal direcotry, it will locate MDS by FID directly.
1813  * \param[in] lmv       LMV device
1814  * \param[in] op_data   client MD stack parameters, name, namelen
1815  *                      mds_num etc.
1816  * \param[in] fid       object FID used to locate MDS.
1817  *
1818  * retval               pointer to the lmv_tgt_desc if succeed.
1819  *                      ERR_PTR(errno) if failed.
1820  */
1821 struct lmv_tgt_desc*
1822 lmv_locate_mds(struct lmv_obd *lmv, struct md_op_data *op_data,
1823                struct lu_fid *fid)
1824 {
1825         struct lmv_stripe_md    *lsm = op_data->op_mea1;
1826         struct lmv_tgt_desc     *tgt;
1827
1828         /* During creating VOLATILE file, it should honor the mdt
1829          * index if the file under striped dir is being restored, see
1830          * ct_restore(). */
1831         if (op_data->op_bias & MDS_CREATE_VOLATILE &&
1832             (int)op_data->op_mds != -1 && lsm != NULL) {
1833                 int i;
1834                 tgt = lmv_get_target(lmv, op_data->op_mds, NULL);
1835                 if (IS_ERR(tgt))
1836                         return tgt;
1837
1838                 /* refill the right parent fid */
1839                 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1840                         struct lmv_oinfo *oinfo;
1841
1842                         oinfo = &lsm->lsm_md_oinfo[i];
1843                         if (oinfo->lmo_mds == op_data->op_mds) {
1844                                 *fid = oinfo->lmo_fid;
1845                                 break;
1846                         }
1847                 }
1848
1849                 /* Hmm, can not find the stripe by mdt_index(op_mds) */
1850                 if (i == lsm->lsm_md_stripe_count)
1851                         tgt = ERR_PTR(-EINVAL);
1852
1853                 return tgt;
1854         }
1855
1856         if (lsm == NULL || op_data->op_namelen == 0) {
1857                 tgt = lmv_find_target(lmv, fid);
1858                 if (IS_ERR(tgt))
1859                         return tgt;
1860
1861                 op_data->op_mds = tgt->ltd_idx;
1862                 return tgt;
1863         }
1864
1865         return lmv_locate_target_for_name(lmv, lsm, op_data->op_name,
1866                                           op_data->op_namelen, fid,
1867                                           &op_data->op_mds);
1868 }
1869
1870 int lmv_create(struct obd_export *exp, struct md_op_data *op_data,
1871                const void *data, int datalen, int mode, __u32 uid,
1872                __u32 gid, cfs_cap_t cap_effective, __u64 rdev,
1873                struct ptlrpc_request **request)
1874 {
1875         struct obd_device       *obd = exp->exp_obd;
1876         struct lmv_obd          *lmv = &obd->u.lmv;
1877         struct lmv_tgt_desc     *tgt;
1878         int                      rc;
1879         ENTRY;
1880
1881         rc = lmv_check_connect(obd);
1882         if (rc)
1883                 RETURN(rc);
1884
1885         if (!lmv->desc.ld_active_tgt_count)
1886                 RETURN(-EIO);
1887
1888         tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
1889         if (IS_ERR(tgt))
1890                 RETURN(PTR_ERR(tgt));
1891
1892         CDEBUG(D_INODE, "CREATE name '%.*s' on "DFID" -> mds #%x\n",
1893                op_data->op_namelen, op_data->op_name, PFID(&op_data->op_fid1),
1894                op_data->op_mds);
1895
1896         rc = lmv_fid_alloc(NULL, exp, &op_data->op_fid2, op_data);
1897         if (rc)
1898                 RETURN(rc);
1899
1900         /* Send the create request to the MDT where the object
1901          * will be located */
1902         tgt = lmv_find_target(lmv, &op_data->op_fid2);
1903         if (IS_ERR(tgt))
1904                 RETURN(PTR_ERR(tgt));
1905
1906         op_data->op_mds = tgt->ltd_idx;
1907
1908         CDEBUG(D_INODE, "CREATE obj "DFID" -> mds #%x\n",
1909                PFID(&op_data->op_fid2), op_data->op_mds);
1910
1911         op_data->op_flags |= MF_MDC_CANCEL_FID1;
1912         rc = md_create(tgt->ltd_exp, op_data, data, datalen, mode, uid, gid,
1913                        cap_effective, rdev, request);
1914         if (rc == 0) {
1915                 if (*request == NULL)
1916                         RETURN(rc);
1917                 CDEBUG(D_INODE, "Created - "DFID"\n", PFID(&op_data->op_fid2));
1918         }
1919         RETURN(rc);
1920 }
1921
1922 static int lmv_done_writing(struct obd_export *exp,
1923                             struct md_op_data *op_data,
1924                             struct md_open_data *mod)
1925 {
1926         struct obd_device     *obd = exp->exp_obd;
1927         struct lmv_obd        *lmv = &obd->u.lmv;
1928         struct lmv_tgt_desc   *tgt;
1929         int                    rc;
1930         ENTRY;
1931
1932         rc = lmv_check_connect(obd);
1933         if (rc)
1934                 RETURN(rc);
1935
1936         tgt = lmv_find_target(lmv, &op_data->op_fid1);
1937         if (IS_ERR(tgt))
1938                 RETURN(PTR_ERR(tgt));
1939
1940         rc = md_done_writing(tgt->ltd_exp, op_data, mod);
1941         RETURN(rc);
1942 }
1943
1944 static int
1945 lmv_enqueue(struct obd_export *exp, struct ldlm_enqueue_info *einfo,
1946             const union ldlm_policy_data *policy,
1947             struct lookup_intent *it, struct md_op_data *op_data,
1948             struct lustre_handle *lockh, __u64 extra_lock_flags)
1949 {
1950         struct obd_device        *obd = exp->exp_obd;
1951         struct lmv_obd           *lmv = &obd->u.lmv;
1952         struct lmv_tgt_desc      *tgt;
1953         int                       rc;
1954         ENTRY;
1955
1956         rc = lmv_check_connect(obd);
1957         if (rc)
1958                 RETURN(rc);
1959
1960         CDEBUG(D_INODE, "ENQUEUE '%s' on "DFID"\n",
1961                LL_IT2STR(it), PFID(&op_data->op_fid1));
1962
1963         tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
1964         if (IS_ERR(tgt))
1965                 RETURN(PTR_ERR(tgt));
1966
1967         CDEBUG(D_INODE, "ENQUEUE '%s' on "DFID" -> mds #%d\n",
1968                LL_IT2STR(it), PFID(&op_data->op_fid1), tgt->ltd_idx);
1969
1970         rc = md_enqueue(tgt->ltd_exp, einfo, policy, it, op_data, lockh,
1971                         extra_lock_flags);
1972
1973         RETURN(rc);
1974 }
1975
1976 static int
1977 lmv_getattr_name(struct obd_export *exp,struct md_op_data *op_data,
1978                  struct ptlrpc_request **preq)
1979 {
1980         struct ptlrpc_request   *req = NULL;
1981         struct obd_device       *obd = exp->exp_obd;
1982         struct lmv_obd          *lmv = &obd->u.lmv;
1983         struct lmv_tgt_desc     *tgt;
1984         struct mdt_body         *body;
1985         int                      rc;
1986         ENTRY;
1987
1988         rc = lmv_check_connect(obd);
1989         if (rc)
1990                 RETURN(rc);
1991
1992         tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
1993         if (IS_ERR(tgt))
1994                 RETURN(PTR_ERR(tgt));
1995
1996         CDEBUG(D_INODE, "GETATTR_NAME for %*s on "DFID" -> mds #%d\n",
1997                op_data->op_namelen, op_data->op_name, PFID(&op_data->op_fid1),
1998                tgt->ltd_idx);
1999
2000         rc = md_getattr_name(tgt->ltd_exp, op_data, preq);
2001         if (rc != 0)
2002                 RETURN(rc);
2003
2004         body = req_capsule_server_get(&(*preq)->rq_pill, &RMF_MDT_BODY);
2005         LASSERT(body != NULL);
2006
2007         if (body->mbo_valid & OBD_MD_MDS) {
2008                 struct lu_fid rid = body->mbo_fid1;
2009                 CDEBUG(D_INODE, "Request attrs for "DFID"\n",
2010                        PFID(&rid));
2011
2012                 tgt = lmv_find_target(lmv, &rid);
2013                 if (IS_ERR(tgt)) {
2014                         ptlrpc_req_finished(*preq);
2015                         preq = NULL;
2016                         RETURN(PTR_ERR(tgt));
2017                 }
2018
2019                 op_data->op_fid1 = rid;
2020                 op_data->op_valid |= OBD_MD_FLCROSSREF;
2021                 op_data->op_namelen = 0;
2022                 op_data->op_name = NULL;
2023                 rc = md_getattr_name(tgt->ltd_exp, op_data, &req);
2024                 ptlrpc_req_finished(*preq);
2025                 *preq = req;
2026         }
2027
2028         RETURN(rc);
2029 }
2030
2031 #define md_op_data_fid(op_data, fl)                     \
2032         (fl == MF_MDC_CANCEL_FID1 ? &op_data->op_fid1 : \
2033          fl == MF_MDC_CANCEL_FID2 ? &op_data->op_fid2 : \
2034          fl == MF_MDC_CANCEL_FID3 ? &op_data->op_fid3 : \
2035          fl == MF_MDC_CANCEL_FID4 ? &op_data->op_fid4 : \
2036          NULL)
2037
2038 static int lmv_early_cancel(struct obd_export *exp, struct lmv_tgt_desc *tgt,
2039                             struct md_op_data *op_data,
2040                             int op_tgt, ldlm_mode_t mode, int bits, int flag)
2041 {
2042         struct lu_fid          *fid = md_op_data_fid(op_data, flag);
2043         struct obd_device      *obd = exp->exp_obd;
2044         struct lmv_obd         *lmv = &obd->u.lmv;
2045         ldlm_policy_data_t      policy = {{ 0 }};
2046         int                     rc = 0;
2047         ENTRY;
2048
2049         if (!fid_is_sane(fid))
2050                 RETURN(0);
2051
2052         if (tgt == NULL) {
2053                 tgt = lmv_find_target(lmv, fid);
2054                 if (IS_ERR(tgt))
2055                         RETURN(PTR_ERR(tgt));
2056         }
2057
2058         if (tgt->ltd_idx != op_tgt) {
2059                 CDEBUG(D_INODE, "EARLY_CANCEL on "DFID"\n", PFID(fid));
2060                 policy.l_inodebits.bits = bits;
2061                 rc = md_cancel_unused(tgt->ltd_exp, fid, &policy,
2062                                       mode, LCF_ASYNC, NULL);
2063         } else {
2064                 CDEBUG(D_INODE,
2065                        "EARLY_CANCEL skip operation target %d on "DFID"\n",
2066                        op_tgt, PFID(fid));
2067                 op_data->op_flags |= flag;
2068                 rc = 0;
2069         }
2070
2071         RETURN(rc);
2072 }
2073
2074 /*
2075  * llite passes fid of an target inode in op_data->op_fid1 and id of directory in
2076  * op_data->op_fid2
2077  */
2078 static int lmv_link(struct obd_export *exp, struct md_op_data *op_data,
2079                     struct ptlrpc_request **request)
2080 {
2081         struct obd_device       *obd = exp->exp_obd;
2082         struct lmv_obd          *lmv = &obd->u.lmv;
2083         struct lmv_tgt_desc     *tgt;
2084         int                      rc;
2085         ENTRY;
2086
2087         rc = lmv_check_connect(obd);
2088         if (rc)
2089                 RETURN(rc);
2090
2091         LASSERT(op_data->op_namelen != 0);
2092
2093         CDEBUG(D_INODE, "LINK "DFID":%*s to "DFID"\n",
2094                PFID(&op_data->op_fid2), op_data->op_namelen,
2095                op_data->op_name, PFID(&op_data->op_fid1));
2096
2097         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2098         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2099         op_data->op_cap = cfs_curproc_cap_pack();
2100         if (op_data->op_mea2 != NULL) {
2101                 struct lmv_stripe_md    *lsm = op_data->op_mea2;
2102                 const struct lmv_oinfo  *oinfo;
2103
2104                 oinfo = lsm_name_to_stripe_info(lsm, op_data->op_name,
2105                                                 op_data->op_namelen);
2106                 if (IS_ERR(oinfo))
2107                         RETURN(PTR_ERR(oinfo));
2108
2109                 op_data->op_fid2 = oinfo->lmo_fid;
2110         }
2111
2112         tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid2);
2113         if (IS_ERR(tgt))
2114                 RETURN(PTR_ERR(tgt));
2115
2116         /*
2117          * Cancel UPDATE lock on child (fid1).
2118          */
2119         op_data->op_flags |= MF_MDC_CANCEL_FID2;
2120         rc = lmv_early_cancel(exp, NULL, op_data, tgt->ltd_idx, LCK_EX,
2121                               MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID1);
2122         if (rc != 0)
2123                 RETURN(rc);
2124
2125         rc = md_link(tgt->ltd_exp, op_data, request);
2126
2127         RETURN(rc);
2128 }
2129
2130 static int lmv_rename(struct obd_export *exp, struct md_op_data *op_data,
2131                       const char *old, int oldlen, const char *new, int newlen,
2132                       struct ptlrpc_request **request)
2133 {
2134         struct obd_device       *obd = exp->exp_obd;
2135         struct lmv_obd          *lmv = &obd->u.lmv;
2136         struct lmv_tgt_desc     *src_tgt;
2137         int                     rc;
2138         ENTRY;
2139
2140         LASSERT(oldlen != 0);
2141
2142         CDEBUG(D_INODE, "RENAME %.*s in "DFID":%d to %.*s in "DFID":%d\n",
2143                oldlen, old, PFID(&op_data->op_fid1),
2144                op_data->op_mea1 ? op_data->op_mea1->lsm_md_stripe_count : 0,
2145                newlen, new, PFID(&op_data->op_fid2),
2146                op_data->op_mea2 ? op_data->op_mea2->lsm_md_stripe_count : 0);
2147
2148         rc = lmv_check_connect(obd);
2149         if (rc)
2150                 RETURN(rc);
2151
2152         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2153         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2154         op_data->op_cap = cfs_curproc_cap_pack();
2155         if (op_data->op_cli_flags & CLI_MIGRATE) {
2156                 LASSERTF(fid_is_sane(&op_data->op_fid3), "invalid FID "DFID"\n",
2157                          PFID(&op_data->op_fid3));
2158                 rc = lmv_fid_alloc(NULL, exp, &op_data->op_fid2, op_data);
2159                 if (rc)
2160                         RETURN(rc);
2161                 src_tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid3);
2162         } else {
2163                 if (op_data->op_mea1 != NULL) {
2164                         struct lmv_stripe_md    *lsm = op_data->op_mea1;
2165
2166                         src_tgt = lmv_locate_target_for_name(lmv, lsm, old,
2167                                                              oldlen,
2168                                                              &op_data->op_fid1,
2169                                                              &op_data->op_mds);
2170                         if (IS_ERR(src_tgt))
2171                                 RETURN(PTR_ERR(src_tgt));
2172                 } else {
2173                         src_tgt = lmv_find_target(lmv, &op_data->op_fid1);
2174                         if (IS_ERR(src_tgt))
2175                                 RETURN(PTR_ERR(src_tgt));
2176
2177                         op_data->op_mds = src_tgt->ltd_idx;
2178                 }
2179
2180                 if (op_data->op_mea2) {
2181                         struct lmv_stripe_md    *lsm = op_data->op_mea2;
2182                         const struct lmv_oinfo  *oinfo;
2183
2184                         oinfo = lsm_name_to_stripe_info(lsm, new, newlen);
2185                         if (IS_ERR(oinfo))
2186                                 RETURN(PTR_ERR(oinfo));
2187
2188                         op_data->op_fid2 = oinfo->lmo_fid;
2189                 }
2190         }
2191         if (IS_ERR(src_tgt))
2192                 RETURN(PTR_ERR(src_tgt));
2193
2194         /*
2195          * LOOKUP lock on src child (fid3) should also be cancelled for
2196          * src_tgt in mdc_rename.
2197          */
2198         op_data->op_flags |= MF_MDC_CANCEL_FID1 | MF_MDC_CANCEL_FID3;
2199
2200         /*
2201          * Cancel UPDATE locks on tgt parent (fid2), tgt_tgt is its
2202          * own target.
2203          */
2204         rc = lmv_early_cancel(exp, NULL, op_data, src_tgt->ltd_idx,
2205                               LCK_EX, MDS_INODELOCK_UPDATE,
2206                               MF_MDC_CANCEL_FID2);
2207
2208         if (rc != 0)
2209                 RETURN(rc);
2210         /*
2211          * Cancel LOOKUP locks on source child (fid3) for parent tgt_tgt.
2212          */
2213         if (fid_is_sane(&op_data->op_fid3)) {
2214                 struct lmv_tgt_desc *tgt;
2215
2216                 tgt = lmv_find_target(lmv, &op_data->op_fid1);
2217                 if (IS_ERR(tgt))
2218                         RETURN(PTR_ERR(tgt));
2219
2220                 /* Cancel LOOKUP lock on its parent */
2221                 rc = lmv_early_cancel(exp, tgt, op_data, src_tgt->ltd_idx,
2222                                       LCK_EX, MDS_INODELOCK_LOOKUP,
2223                                       MF_MDC_CANCEL_FID3);
2224                 if (rc != 0)
2225                         RETURN(rc);
2226
2227                 rc = lmv_early_cancel(exp, NULL, op_data, src_tgt->ltd_idx,
2228                                       LCK_EX, MDS_INODELOCK_FULL,
2229                                       MF_MDC_CANCEL_FID3);
2230                 if (rc != 0)
2231                         RETURN(rc);
2232         }
2233
2234         /*
2235          * Cancel all the locks on tgt child (fid4).
2236          */
2237         if (fid_is_sane(&op_data->op_fid4))
2238                 rc = lmv_early_cancel(exp, NULL, op_data, src_tgt->ltd_idx,
2239                                       LCK_EX, MDS_INODELOCK_FULL,
2240                                       MF_MDC_CANCEL_FID4);
2241
2242         CDEBUG(D_INODE, DFID":m%d to "DFID"\n", PFID(&op_data->op_fid1),
2243                op_data->op_mds, PFID(&op_data->op_fid2));
2244
2245         rc = md_rename(src_tgt->ltd_exp, op_data, old, oldlen, new, newlen,
2246                        request);
2247
2248         RETURN(rc);
2249 }
2250
2251 static int lmv_setattr(struct obd_export *exp, struct md_op_data *op_data,
2252                        void *ea, int ealen, void *ea2, int ea2len,
2253                        struct ptlrpc_request **request,
2254                        struct md_open_data **mod)
2255 {
2256         struct obd_device       *obd = exp->exp_obd;
2257         struct lmv_obd          *lmv = &obd->u.lmv;
2258         struct lmv_tgt_desc     *tgt;
2259         int                      rc = 0;
2260         ENTRY;
2261
2262         rc = lmv_check_connect(obd);
2263         if (rc)
2264                 RETURN(rc);
2265
2266         CDEBUG(D_INODE, "SETATTR for "DFID", valid 0x%x\n",
2267                PFID(&op_data->op_fid1), op_data->op_attr.ia_valid);
2268
2269         op_data->op_flags |= MF_MDC_CANCEL_FID1;
2270         tgt = lmv_find_target(lmv, &op_data->op_fid1);
2271         if (IS_ERR(tgt))
2272                 RETURN(PTR_ERR(tgt));
2273
2274         rc = md_setattr(tgt->ltd_exp, op_data, ea, ealen, ea2,
2275                         ea2len, request, mod);
2276
2277         RETURN(rc);
2278 }
2279
2280 static int lmv_fsync(struct obd_export *exp, const struct lu_fid *fid,
2281                      struct obd_capa *oc, struct ptlrpc_request **request)
2282 {
2283         struct obd_device       *obd = exp->exp_obd;
2284         struct lmv_obd          *lmv = &obd->u.lmv;
2285         struct lmv_tgt_desc     *tgt;
2286         int                      rc;
2287         ENTRY;
2288
2289         rc = lmv_check_connect(obd);
2290         if (rc != 0)
2291                 RETURN(rc);
2292
2293         tgt = lmv_find_target(lmv, fid);
2294         if (IS_ERR(tgt))
2295                 RETURN(PTR_ERR(tgt));
2296
2297         rc = md_fsync(tgt->ltd_exp, fid, oc, request);
2298         RETURN(rc);
2299 }
2300
2301 /**
2302  * Get current minimum entry from striped directory
2303  *
2304  * This function will search the dir entry, whose hash value is the
2305  * closest(>=) to @hash_offset, from all of sub-stripes, and it is
2306  * only being called for striped directory.
2307  *
2308  * \param[in] exp               export of LMV
2309  * \param[in] op_data           parameters transferred beween client MD stack
2310  *                              stripe_information will be included in this
2311  *                              parameter
2312  * \param[in] cb_op             ldlm callback being used in enqueue in
2313  *                              mdc_read_page
2314  * \param[in] hash_offset       the hash value, which is used to locate
2315  *                              minum(closet) dir entry
2316  * \param[in|out] stripe_offset the caller use this to indicate the stripe
2317  *                              index of last entry, so to avoid hash conflict
2318  *                              between stripes. It will also be used to
2319  *                              return the stripe index of current dir entry.
2320  * \param[in|out] entp          the minum entry and it also is being used
2321  *                              to input the last dir entry to resolve the
2322  *                              hash conflict
2323  *
2324  * \param[out] ppage            the page which holds the minum entry
2325  *
2326  * \retval                      = 0 get the entry successfully
2327  *                              negative errno (< 0) does not get the entry
2328  */
2329 static int lmv_get_min_striped_entry(struct obd_export *exp,
2330                                      struct md_op_data *op_data,
2331                                      struct md_callback *cb_op,
2332                                      __u64 hash_offset, int *stripe_offset,
2333                                      struct lu_dirent **entp,
2334                                      struct page **ppage)
2335 {
2336         struct obd_device       *obd = exp->exp_obd;
2337         struct lmv_obd          *lmv = &obd->u.lmv;
2338         struct lmv_stripe_md    *lsm = op_data->op_mea1;
2339         struct lmv_tgt_desc     *tgt;
2340         int                     stripe_count;
2341         struct lu_dirent        *min_ent = NULL;
2342         struct page             *min_page = NULL;
2343         int                     min_idx = 0;
2344         int                     i;
2345         int                     rc = 0;
2346         ENTRY;
2347
2348         stripe_count = lsm->lsm_md_stripe_count;
2349         for (i = 0; i < stripe_count; i++) {
2350                 struct lu_dirent        *ent = NULL;
2351                 struct page             *page = NULL;
2352                 struct lu_dirpage       *dp;
2353                 __u64                   stripe_hash = hash_offset;
2354
2355                 tgt = lmv_get_target(lmv, lsm->lsm_md_oinfo[i].lmo_mds, NULL);
2356                 if (IS_ERR(tgt))
2357                         GOTO(out, rc = PTR_ERR(tgt));
2358
2359                 /* op_data will be shared by each stripe, so we need
2360                  * reset these value for each stripe */
2361                 op_data->op_stripe_offset = i;
2362                 op_data->op_fid1 = lsm->lsm_md_oinfo[i].lmo_fid;
2363                 op_data->op_fid2 = lsm->lsm_md_oinfo[i].lmo_fid;
2364                 op_data->op_data = lsm->lsm_md_oinfo[i].lmo_root;
2365 next:
2366                 rc = md_read_page(tgt->ltd_exp, op_data, cb_op, stripe_hash,
2367                                   &page);
2368                 if (rc != 0)
2369                         GOTO(out, rc);
2370
2371                 dp = page_address(page);
2372                 for (ent = lu_dirent_start(dp); ent != NULL;
2373                      ent = lu_dirent_next(ent)) {
2374                         /* Skip dummy entry */
2375                         if (le16_to_cpu(ent->lde_namelen) == 0)
2376                                 continue;
2377
2378                         if (le64_to_cpu(ent->lde_hash) < hash_offset)
2379                                 continue;
2380
2381                         if (le64_to_cpu(ent->lde_hash) == hash_offset &&
2382                             (*entp == ent || i < *stripe_offset))
2383                                 continue;
2384
2385                         /* skip . and .. for other stripes */
2386                         if (i != 0 &&
2387                             (strncmp(ent->lde_name, ".",
2388                                      le16_to_cpu(ent->lde_namelen)) == 0 ||
2389                              strncmp(ent->lde_name, "..",
2390                                      le16_to_cpu(ent->lde_namelen)) == 0))
2391                                 continue;
2392                         break;
2393                 }
2394
2395                 if (ent == NULL) {
2396                         stripe_hash = le64_to_cpu(dp->ldp_hash_end);
2397
2398                         kunmap(page);
2399                         page_cache_release(page);
2400                         page = NULL;
2401
2402                         /* reach the end of current stripe, go to next stripe */
2403                         if (stripe_hash == MDS_DIR_END_OFF)
2404                                 continue;
2405                         else
2406                                 goto next;
2407                 }
2408
2409                 if (min_ent != NULL) {
2410                         if (le64_to_cpu(min_ent->lde_hash) >
2411                             le64_to_cpu(ent->lde_hash)) {
2412                                 min_ent = ent;
2413                                 kunmap(min_page);
2414                                 page_cache_release(min_page);
2415                                 min_idx = i;
2416                                 min_page = page;
2417                         } else {
2418                                 kunmap(page);
2419                                 page_cache_release(page);
2420                                 page = NULL;
2421                         }
2422                 } else {
2423                         min_ent = ent;
2424                         min_page = page;
2425                         min_idx = i;
2426                 }
2427         }
2428
2429 out:
2430         if (*ppage != NULL) {
2431                 kunmap(*ppage);
2432                 page_cache_release(*ppage);
2433         }
2434         *stripe_offset = min_idx;
2435         *entp = min_ent;
2436         *ppage = min_page;
2437         RETURN(rc);
2438 }
2439
2440 /**
2441  * Build dir entry page from a striped directory
2442  *
2443  * This function gets one entry by @offset from a striped directory. It will
2444  * read entries from all of stripes, and choose one closest to the required
2445  * offset(&offset). A few notes
2446  * 1. skip . and .. for non-zero stripes, because there can only have one .
2447  * and .. in a directory.
2448  * 2. op_data will be shared by all of stripes, instead of allocating new
2449  * one, so need to restore before reusing.
2450  * 3. release the entry page if that is not being chosen.
2451  *
2452  * \param[in] exp       obd export refer to LMV
2453  * \param[in] op_data   hold those MD parameters of read_entry
2454  * \param[in] cb_op     ldlm callback being used in enqueue in mdc_read_entry
2455  * \param[out] ldp      the entry being read
2456  * \param[out] ppage    the page holding the entry. Note: because the entry
2457  *                      will be accessed in upper layer, so we need hold the
2458  *                      page until the usages of entry is finished, see
2459  *                      ll_dir_entry_next.
2460  *
2461  * retval               =0 if get entry successfully
2462  *                      <0 cannot get entry
2463  */
2464 static int lmv_read_striped_page(struct obd_export *exp,
2465                                  struct md_op_data *op_data,
2466                                  struct md_callback *cb_op,
2467                                  __u64 offset, struct page **ppage)
2468 {
2469         struct obd_device       *obd = exp->exp_obd;
2470         struct lu_fid           master_fid = op_data->op_fid1;
2471         struct inode            *master_inode = op_data->op_data;
2472         __u64                   hash_offset = offset;
2473         struct lu_dirpage       *dp;
2474         struct page             *min_ent_page = NULL;
2475         struct page             *ent_page = NULL;
2476         struct lu_dirent        *ent;
2477         void                    *area;
2478         int                     ent_idx = 0;
2479         struct lu_dirent        *min_ent = NULL;
2480         struct lu_dirent        *last_ent;
2481         int                     left_bytes;
2482         int                     rc;
2483         ENTRY;
2484
2485         rc = lmv_check_connect(obd);
2486         if (rc)
2487                 RETURN(rc);
2488
2489         /* Allocate a page and read entries from all of stripes and fill
2490          * the page by hash order */
2491         ent_page = alloc_page(GFP_KERNEL);
2492         if (ent_page == NULL)
2493                 RETURN(-ENOMEM);
2494
2495         /* Initialize the entry page */
2496         dp = kmap(ent_page);
2497         memset(dp, 0, sizeof(*dp));
2498         dp->ldp_hash_start = cpu_to_le64(offset);
2499         dp->ldp_flags |= LDF_COLLIDE;
2500
2501         area = dp + 1;
2502         left_bytes = PAGE_CACHE_SIZE - sizeof(*dp);
2503         ent = area;
2504         last_ent = ent;
2505         do {
2506                 __u16   ent_size;
2507
2508                 /* Find the minum entry from all sub-stripes */
2509                 rc = lmv_get_min_striped_entry(exp, op_data, cb_op, hash_offset,
2510                                                &ent_idx, &min_ent,
2511                                                &min_ent_page);
2512                 if (rc != 0)
2513                         GOTO(out, rc);
2514
2515                 /* If it can not get minum entry, it means it already reaches
2516                  * the end of this directory */
2517                 if (min_ent == NULL) {
2518                         last_ent->lde_reclen = 0;
2519                         hash_offset = MDS_DIR_END_OFF;
2520                         GOTO(out, rc);
2521                 }
2522
2523                 ent_size = le16_to_cpu(min_ent->lde_reclen);
2524
2525                 /* the last entry lde_reclen is 0, but it might not
2526                  * the end of this entry of this temporay entry */
2527                 if (ent_size == 0)
2528                         ent_size = lu_dirent_calc_size(
2529                                         le16_to_cpu(min_ent->lde_namelen),
2530                                         le32_to_cpu(min_ent->lde_attrs));
2531                 if (ent_size > left_bytes) {
2532                         last_ent->lde_reclen = cpu_to_le16(0);
2533                         hash_offset = le64_to_cpu(min_ent->lde_hash);
2534                         GOTO(out, rc);
2535                 }
2536
2537                 memcpy(ent, min_ent, ent_size);
2538
2539                 /* Replace . with master FID and Replace .. with the parent FID
2540                  * of master object */
2541                 if (strncmp(ent->lde_name, ".",
2542                             le16_to_cpu(ent->lde_namelen)) == 0 &&
2543                     le16_to_cpu(ent->lde_namelen) == 1)
2544                         fid_cpu_to_le(&ent->lde_fid, &master_fid);
2545                 else if (strncmp(ent->lde_name, "..",
2546                                    le16_to_cpu(ent->lde_namelen)) == 0 &&
2547                            le16_to_cpu(ent->lde_namelen) == 2)
2548                         fid_cpu_to_le(&ent->lde_fid, &op_data->op_fid3);
2549
2550                 left_bytes -= ent_size;
2551                 ent->lde_reclen = cpu_to_le16(ent_size);
2552                 last_ent = ent;
2553                 ent = (void *)ent + ent_size;
2554                 hash_offset = le64_to_cpu(min_ent->lde_hash);
2555                 if (hash_offset == MDS_DIR_END_OFF) {
2556                         last_ent->lde_reclen = 0;
2557                         break;
2558                 }
2559         } while (1);
2560 out:
2561         if (min_ent_page != NULL) {
2562                 kunmap(min_ent_page);
2563                 page_cache_release(min_ent_page);
2564         }
2565
2566         if (unlikely(rc != 0)) {
2567                 __free_page(ent_page);
2568                 ent_page = NULL;
2569         } else {
2570                 if (ent == area)
2571                         dp->ldp_flags |= LDF_EMPTY;
2572                 dp->ldp_flags = cpu_to_le32(dp->ldp_flags);
2573                 dp->ldp_hash_end = cpu_to_le64(hash_offset);
2574         }
2575
2576         /* We do not want to allocate md_op_data during each
2577          * dir entry reading, so op_data will be shared by every stripe,
2578          * then we need to restore it back to original value before
2579          * return to the upper layer */
2580         op_data->op_fid1 = master_fid;
2581         op_data->op_fid2 = master_fid;
2582         op_data->op_data = master_inode;
2583
2584         *ppage = ent_page;
2585
2586         RETURN(rc);
2587 }
2588
2589 int lmv_read_page(struct obd_export *exp, struct md_op_data *op_data,
2590                   struct md_callback *cb_op, __u64 offset,
2591                   struct page **ppage)
2592 {
2593         struct obd_device       *obd = exp->exp_obd;
2594         struct lmv_obd          *lmv = &obd->u.lmv;
2595         struct lmv_stripe_md    *lsm = op_data->op_mea1;
2596         struct lmv_tgt_desc     *tgt;
2597         int                     rc;
2598         ENTRY;
2599
2600         rc = lmv_check_connect(obd);
2601         if (rc != 0)
2602                 RETURN(rc);
2603
2604         if (unlikely(lsm != NULL)) {
2605                 rc = lmv_read_striped_page(exp, op_data, cb_op, offset, ppage);
2606                 RETURN(rc);
2607         }
2608
2609         tgt = lmv_find_target(lmv, &op_data->op_fid1);
2610         if (IS_ERR(tgt))
2611                 RETURN(PTR_ERR(tgt));
2612
2613         rc = md_read_page(tgt->ltd_exp, op_data, cb_op, offset, ppage);
2614
2615         RETURN(rc);
2616 }
2617
2618 /**
2619  * Unlink a file/directory
2620  *
2621  * Unlink a file or directory under the parent dir. The unlink request
2622  * usually will be sent to the MDT where the child is located, but if
2623  * the client does not have the child FID then request will be sent to the
2624  * MDT where the parent is located.
2625  *
2626  * If the parent is a striped directory then it also needs to locate which
2627  * stripe the name of the child is located, and replace the parent FID
2628  * (@op->op_fid1) with the stripe FID. Note: if the stripe is unknown,
2629  * it will walk through all of sub-stripes until the child is being
2630  * unlinked finally.
2631  *
2632  * \param[in] exp       export refer to LMV
2633  * \param[in] op_data   different parameters transferred beween client
2634  *                      MD stacks, name, namelen, FIDs etc.
2635  *                      op_fid1 is the parent FID, op_fid2 is the child
2636  *                      FID.
2637  * \param[out] request  point to the request of unlink.
2638  *
2639  * retval               0 if succeed
2640  *                      negative errno if failed.
2641  */
2642 static int lmv_unlink(struct obd_export *exp, struct md_op_data *op_data,
2643                       struct ptlrpc_request **request)
2644 {
2645         struct obd_device       *obd = exp->exp_obd;
2646         struct lmv_obd          *lmv = &obd->u.lmv;
2647         struct lmv_tgt_desc     *tgt = NULL;
2648         struct lmv_tgt_desc     *parent_tgt = NULL;
2649         struct mdt_body         *body;
2650         int                     rc;
2651         int                     stripe_index = 0;
2652         struct lmv_stripe_md    *lsm = op_data->op_mea1;
2653         ENTRY;
2654
2655         rc = lmv_check_connect(obd);
2656         if (rc)
2657                 RETURN(rc);
2658 retry_unlink:
2659         /* For striped dir, we need to locate the parent as well */
2660         if (lsm != NULL) {
2661                 struct lmv_tgt_desc *tmp;
2662
2663                 LASSERT(op_data->op_name != NULL &&
2664                         op_data->op_namelen != 0);
2665
2666                 tmp = lmv_locate_target_for_name(lmv, lsm,
2667                                                  op_data->op_name,
2668                                                  op_data->op_namelen,
2669                                                  &op_data->op_fid1,
2670                                                  &op_data->op_mds);
2671
2672                 /* return -EBADFD means unknown hash type, might
2673                  * need try all sub-stripe here */
2674                 if (IS_ERR(tmp) && PTR_ERR(tmp) != -EBADFD)
2675                         RETURN(PTR_ERR(tmp));
2676
2677                 /* Note: both migrating dir and unknown hash dir need to
2678                  * try all of sub-stripes, so we need start search the
2679                  * name from stripe 0, but migrating dir is already handled
2680                  * inside lmv_locate_target_for_name(), so we only check
2681                  * unknown hash type directory here */
2682                 if (!lmv_is_known_hash_type(lsm)) {
2683                         struct lmv_oinfo *oinfo;
2684
2685                         oinfo = &lsm->lsm_md_oinfo[stripe_index];
2686
2687                         op_data->op_fid1 = oinfo->lmo_fid;
2688                         op_data->op_mds = oinfo->lmo_mds;
2689                 }
2690         }
2691
2692 try_next_stripe:
2693         /* Send unlink requests to the MDT where the child is located */
2694         if (likely(!fid_is_zero(&op_data->op_fid2)))
2695                 tgt = lmv_find_target(lmv, &op_data->op_fid2);
2696         else if (lsm != NULL)
2697                 tgt = lmv_get_target(lmv, op_data->op_mds, NULL);
2698         else
2699                 tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
2700
2701         if (IS_ERR(tgt))
2702                 RETURN(PTR_ERR(tgt));
2703
2704         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2705         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2706         op_data->op_cap = cfs_curproc_cap_pack();
2707
2708         /*
2709          * If child's fid is given, cancel unused locks for it if it is from
2710          * another export than parent.
2711          *
2712          * LOOKUP lock for child (fid3) should also be cancelled on parent
2713          * tgt_tgt in mdc_unlink().
2714          */
2715         op_data->op_flags |= MF_MDC_CANCEL_FID1 | MF_MDC_CANCEL_FID3;
2716
2717         /*
2718          * Cancel FULL locks on child (fid3).
2719          */
2720         parent_tgt = lmv_find_target(lmv, &op_data->op_fid1);
2721         if (IS_ERR(parent_tgt))
2722                 RETURN(PTR_ERR(parent_tgt));
2723
2724         if (parent_tgt != tgt) {
2725                 rc = lmv_early_cancel(exp, parent_tgt, op_data, tgt->ltd_idx,
2726                                       LCK_EX, MDS_INODELOCK_LOOKUP,
2727                                       MF_MDC_CANCEL_FID3);
2728         }
2729
2730         rc = lmv_early_cancel(exp, NULL, op_data, tgt->ltd_idx, LCK_EX,
2731                               MDS_INODELOCK_FULL, MF_MDC_CANCEL_FID3);
2732         if (rc != 0)
2733                 RETURN(rc);
2734
2735         CDEBUG(D_INODE, "unlink with fid="DFID"/"DFID" -> mds #%d\n",
2736                PFID(&op_data->op_fid1), PFID(&op_data->op_fid2), tgt->ltd_idx);
2737
2738         rc = md_unlink(tgt->ltd_exp, op_data, request);
2739         if (rc != 0 && rc != -EREMOTE && rc != -ENOENT)
2740                 RETURN(rc);
2741
2742         /* Try next stripe if it is needed. */
2743         if (rc == -ENOENT && lsm != NULL && lmv_need_try_all_stripes(lsm)) {
2744                 struct lmv_oinfo *oinfo;
2745
2746                 stripe_index++;
2747                 if (stripe_index >= lsm->lsm_md_stripe_count)
2748                         RETURN(rc);
2749
2750                 oinfo = &lsm->lsm_md_oinfo[stripe_index];
2751
2752                 op_data->op_fid1 = oinfo->lmo_fid;
2753                 op_data->op_mds = oinfo->lmo_mds;
2754
2755                 ptlrpc_req_finished(*request);
2756                 *request = NULL;
2757
2758                 goto try_next_stripe;
2759         }
2760
2761         body = req_capsule_server_get(&(*request)->rq_pill, &RMF_MDT_BODY);
2762         if (body == NULL)
2763                 RETURN(-EPROTO);
2764
2765         /* Not cross-ref case, just get out of here. */
2766         if (likely(!(body->mbo_valid & OBD_MD_MDS)))
2767                 RETURN(rc);
2768
2769         CDEBUG(D_INODE, "%s: try unlink to another MDT for "DFID"\n",
2770                exp->exp_obd->obd_name, PFID(&body->mbo_fid1));
2771
2772         /* This is a remote object, try remote MDT, Note: it may
2773          * try more than 1 time here, Considering following case
2774          * /mnt/lustre is root on MDT0, remote1 is on MDT1
2775          * 1. Initially A does not know where remote1 is, it send
2776          *    unlink RPC to MDT0, MDT0 return -EREMOTE, it will
2777          *    resend unlink RPC to MDT1 (retry 1st time).
2778          *
2779          * 2. During the unlink RPC in flight,
2780          *    client B mv /mnt/lustre/remote1 /mnt/lustre/remote2
2781          *    and create new remote1, but on MDT0
2782          *
2783          * 3. MDT1 get unlink RPC(from A), then do remote lock on
2784          *    /mnt/lustre, then lookup get fid of remote1, and find
2785          *    it is remote dir again, and replay -EREMOTE again.
2786          *
2787          * 4. Then A will resend unlink RPC to MDT0. (retry 2nd times).
2788          *
2789          * In theory, it might try unlimited time here, but it should
2790          * be very rare case.  */
2791         op_data->op_fid2 = body->mbo_fid1;
2792         ptlrpc_req_finished(*request);
2793         *request = NULL;
2794
2795         goto retry_unlink;
2796 }
2797
2798 static int lmv_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
2799 {
2800         struct lmv_obd *lmv = &obd->u.lmv;
2801         int rc = 0;
2802
2803         switch (stage) {
2804         case OBD_CLEANUP_EARLY:
2805                 /* XXX: here should be calling obd_precleanup() down to
2806                  * stack. */
2807                 break;
2808         case OBD_CLEANUP_EXPORTS:
2809                 fld_client_proc_fini(&lmv->lmv_fld);
2810                 lprocfs_obd_cleanup(obd);
2811                 lprocfs_free_md_stats(obd);
2812                 break;
2813         default:
2814                 break;
2815         }
2816         RETURN(rc);
2817 }
2818
2819 static int lmv_get_info(const struct lu_env *env, struct obd_export *exp,
2820                         __u32 keylen, void *key, __u32 *vallen, void *val,
2821                         struct lov_stripe_md *lsm)
2822 {
2823         struct obd_device       *obd;
2824         struct lmv_obd          *lmv;
2825         int                      rc = 0;
2826         ENTRY;
2827
2828         obd = class_exp2obd(exp);
2829         if (obd == NULL) {
2830                 CDEBUG(D_IOCTL, "Invalid client cookie "LPX64"\n",
2831                        exp->exp_handle.h_cookie);
2832                 RETURN(-EINVAL);
2833         }
2834
2835         lmv = &obd->u.lmv;
2836         if (keylen >= strlen("remote_flag") && !strcmp(key, "remote_flag")) {
2837                 int i;
2838
2839                 rc = lmv_check_connect(obd);
2840                 if (rc)
2841                         RETURN(rc);
2842
2843                 LASSERT(*vallen == sizeof(__u32));
2844                 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2845                         struct lmv_tgt_desc *tgt = lmv->tgts[i];
2846                         /*
2847                          * All tgts should be connected when this gets called.
2848                          */
2849                         if (tgt == NULL || tgt->ltd_exp == NULL)
2850                                 continue;
2851
2852                         if (!obd_get_info(env, tgt->ltd_exp, keylen, key,
2853                                           vallen, val, NULL))
2854                                 RETURN(0);
2855                 }
2856                 RETURN(-EINVAL);
2857         } else if (KEY_IS(KEY_MAX_EASIZE) ||
2858                    KEY_IS(KEY_DEFAULT_EASIZE) ||
2859                    KEY_IS(KEY_MAX_COOKIESIZE) ||
2860                    KEY_IS(KEY_DEFAULT_COOKIESIZE) ||
2861                    KEY_IS(KEY_CONN_DATA)) {
2862                 rc = lmv_check_connect(obd);
2863                 if (rc)
2864                         RETURN(rc);
2865
2866                 /*
2867                  * Forwarding this request to first MDS, it should know LOV
2868                  * desc.
2869                  */
2870                 rc = obd_get_info(env, lmv->tgts[0]->ltd_exp, keylen, key,
2871                                   vallen, val, NULL);
2872                 if (!rc && KEY_IS(KEY_CONN_DATA))
2873                         exp->exp_connect_data = *(struct obd_connect_data *)val;
2874                 RETURN(rc);
2875         } else if (KEY_IS(KEY_TGT_COUNT)) {
2876                 *((int *)val) = lmv->desc.ld_tgt_count;
2877                 RETURN(0);
2878         }
2879
2880         CDEBUG(D_IOCTL, "Invalid key\n");
2881         RETURN(-EINVAL);
2882 }
2883
2884 int lmv_set_info_async(const struct lu_env *env, struct obd_export *exp,
2885                        obd_count keylen, void *key, obd_count vallen,
2886                        void *val, struct ptlrpc_request_set *set)
2887 {
2888         struct lmv_tgt_desc    *tgt = NULL;
2889         struct obd_device      *obd;
2890         struct lmv_obd         *lmv;
2891         int rc = 0;
2892         ENTRY;
2893
2894         obd = class_exp2obd(exp);
2895         if (obd == NULL) {
2896                 CDEBUG(D_IOCTL, "Invalid client cookie "LPX64"\n",
2897                        exp->exp_handle.h_cookie);
2898                 RETURN(-EINVAL);
2899         }
2900         lmv = &obd->u.lmv;
2901
2902         if (KEY_IS(KEY_READ_ONLY) || KEY_IS(KEY_FLUSH_CTX)) {
2903                 int i, err = 0;
2904
2905                 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2906                         tgt = lmv->tgts[i];
2907
2908                         if (tgt == NULL || tgt->ltd_exp == NULL)
2909                                 continue;
2910
2911                         err = obd_set_info_async(env, tgt->ltd_exp,
2912                                                  keylen, key, vallen, val, set);
2913                         if (err && rc == 0)
2914                                 rc = err;
2915                 }
2916
2917                 RETURN(rc);
2918         }
2919
2920         RETURN(-EINVAL);
2921 }
2922
2923 static int lmv_pack_md_v1(const struct lmv_stripe_md *lsm,
2924                           struct lmv_mds_md_v1 *lmm1)
2925 {
2926         int     cplen;
2927         int     i;
2928
2929         lmm1->lmv_magic = cpu_to_le32(lsm->lsm_md_magic);
2930         lmm1->lmv_stripe_count = cpu_to_le32(lsm->lsm_md_stripe_count);
2931         lmm1->lmv_master_mdt_index = cpu_to_le32(lsm->lsm_md_master_mdt_index);
2932         lmm1->lmv_hash_type = cpu_to_le32(lsm->lsm_md_hash_type);
2933         cplen = strlcpy(lmm1->lmv_pool_name, lsm->lsm_md_pool_name,
2934                         sizeof(lmm1->lmv_pool_name));
2935         if (cplen >= sizeof(lmm1->lmv_pool_name))
2936                 return -E2BIG;
2937
2938         for (i = 0; i < lsm->lsm_md_stripe_count; i++)
2939                 fid_cpu_to_le(&lmm1->lmv_stripe_fids[i],
2940                               &lsm->lsm_md_oinfo[i].lmo_fid);
2941         return 0;
2942 }
2943
2944 int lmv_pack_md(union lmv_mds_md **lmmp, const struct lmv_stripe_md *lsm,
2945                 int stripe_count)
2946 {
2947         int     lmm_size = 0;
2948         bool    allocated = false;
2949         int     rc = 0;
2950         ENTRY;
2951
2952         LASSERT(lmmp != NULL);
2953         /* Free lmm */
2954         if (*lmmp != NULL && lsm == NULL) {
2955                 int stripe_count;
2956
2957                 stripe_count = lmv_mds_md_stripe_count_get(*lmmp);
2958                 lmm_size = lmv_mds_md_size(stripe_count,
2959                                            le32_to_cpu((*lmmp)->lmv_magic));
2960                 if (lmm_size == 0)
2961                         RETURN(-EINVAL);
2962                 OBD_FREE(*lmmp, lmm_size);
2963                 *lmmp = NULL;
2964                 RETURN(0);
2965         }
2966
2967         /* Alloc lmm */
2968         if (*lmmp == NULL && lsm == NULL) {
2969                 lmm_size = lmv_mds_md_size(stripe_count, LMV_MAGIC);
2970                 LASSERT(lmm_size > 0);
2971                 OBD_ALLOC(*lmmp, lmm_size);
2972                 if (*lmmp == NULL)
2973                         RETURN(-ENOMEM);
2974                 lmv_mds_md_stripe_count_set(*lmmp, stripe_count);
2975                 (*lmmp)->lmv_magic = cpu_to_le32(LMV_MAGIC);
2976                 RETURN(lmm_size);
2977         }
2978
2979         /* pack lmm */
2980         LASSERT(lsm != NULL);
2981         lmm_size = lmv_mds_md_size(lsm->lsm_md_stripe_count, lsm->lsm_md_magic);
2982         if (*lmmp == NULL) {
2983                 OBD_ALLOC(*lmmp, lmm_size);
2984                 if (*lmmp == NULL)
2985                         RETURN(-ENOMEM);
2986                 allocated = true;
2987         }
2988
2989         switch (lsm->lsm_md_magic) {
2990         case LMV_MAGIC_V1:
2991                 rc = lmv_pack_md_v1(lsm, &(*lmmp)->lmv_md_v1);
2992                 break;
2993         default:
2994                 rc = -EINVAL;
2995                 break;
2996         }
2997
2998         if (rc != 0 && allocated) {
2999                 OBD_FREE(*lmmp, lmm_size);
3000                 *lmmp = NULL;
3001         }
3002
3003         RETURN(lmm_size);
3004 }
3005 EXPORT_SYMBOL(lmv_pack_md);
3006
3007 static int lmv_unpack_md_v1(struct obd_export *exp, struct lmv_stripe_md *lsm,
3008                             const struct lmv_mds_md_v1 *lmm1)
3009 {
3010         struct lmv_obd  *lmv = &exp->exp_obd->u.lmv;
3011         int             stripe_count;
3012         int             cplen;
3013         int             i;
3014         int             rc = 0;
3015         ENTRY;
3016
3017         lsm->lsm_md_magic = le32_to_cpu(lmm1->lmv_magic);
3018         lsm->lsm_md_stripe_count = le32_to_cpu(lmm1->lmv_stripe_count);
3019         lsm->lsm_md_master_mdt_index = le32_to_cpu(lmm1->lmv_master_mdt_index);
3020         if (OBD_FAIL_CHECK(OBD_FAIL_UNKNOWN_LMV_STRIPE))
3021                 lsm->lsm_md_hash_type = LMV_HASH_TYPE_UNKNOWN;
3022         else
3023                 lsm->lsm_md_hash_type = le32_to_cpu(lmm1->lmv_hash_type);
3024         lsm->lsm_md_layout_version = le32_to_cpu(lmm1->lmv_layout_version);
3025         cplen = strlcpy(lsm->lsm_md_pool_name, lmm1->lmv_pool_name,
3026                         sizeof(lsm->lsm_md_pool_name));
3027
3028         if (cplen >= sizeof(lsm->lsm_md_pool_name))
3029                 RETURN(-E2BIG);
3030
3031         CDEBUG(D_INFO, "unpack lsm count %d, master %d hash_type %d"
3032                "layout_version %d\n", lsm->lsm_md_stripe_count,
3033                lsm->lsm_md_master_mdt_index, lsm->lsm_md_hash_type,
3034                lsm->lsm_md_layout_version);
3035
3036         stripe_count = le32_to_cpu(lmm1->lmv_stripe_count);
3037         for (i = 0; i < stripe_count; i++) {
3038                 fid_le_to_cpu(&lsm->lsm_md_oinfo[i].lmo_fid,
3039                               &lmm1->lmv_stripe_fids[i]);
3040                 rc = lmv_fld_lookup(lmv, &lsm->lsm_md_oinfo[i].lmo_fid,
3041                                     &lsm->lsm_md_oinfo[i].lmo_mds);
3042                 if (rc != 0)
3043                         RETURN(rc);
3044                 CDEBUG(D_INFO, "unpack fid #%d "DFID"\n", i,
3045                        PFID(&lsm->lsm_md_oinfo[i].lmo_fid));
3046         }
3047
3048         RETURN(rc);
3049 }
3050
3051 int lmv_unpack_md(struct obd_export *exp, struct lmv_stripe_md **lsmp,
3052                   const union lmv_mds_md *lmm, int stripe_count)
3053 {
3054         struct lmv_stripe_md     *lsm;
3055         int                      lsm_size;
3056         int                      rc;
3057         bool                     allocated = false;
3058         ENTRY;
3059
3060         LASSERT(lsmp != NULL);
3061
3062         lsm = *lsmp;
3063         /* Free memmd */
3064         if (lsm != NULL && lmm == NULL) {
3065                 int i;
3066                 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
3067                         /* For migrating inode, the master stripe and master
3068                          * object will be the same, so do not need iput, see
3069                          * ll_update_lsm_md */
3070                         if (!(lsm->lsm_md_hash_type & LMV_HASH_FLAG_MIGRATION &&
3071                               i == 0) && lsm->lsm_md_oinfo[i].lmo_root != NULL)
3072                                 iput(lsm->lsm_md_oinfo[i].lmo_root);
3073                 }
3074                 lsm_size = lmv_stripe_md_size(lsm->lsm_md_stripe_count);
3075                 OBD_FREE(lsm, lsm_size);
3076                 *lsmp = NULL;
3077                 RETURN(0);
3078         }
3079
3080         /* Alloc memmd */
3081         if (lsm == NULL && lmm == NULL) {
3082                 lsm_size = lmv_stripe_md_size(stripe_count);
3083                 OBD_ALLOC(lsm, lsm_size);
3084                 if (lsm == NULL)
3085                         RETURN(-ENOMEM);
3086                 lsm->lsm_md_stripe_count = stripe_count;
3087                 *lsmp = lsm;
3088                 RETURN(0);
3089         }
3090
3091         if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_STRIPE)
3092                 RETURN(-EPERM);
3093
3094         /* Unpack memmd */
3095         if (le32_to_cpu(lmm->lmv_magic) != LMV_MAGIC_V1 &&
3096             le32_to_cpu(lmm->lmv_magic) != LMV_USER_MAGIC) {
3097                 CERROR("%s: invalid lmv magic %x: rc = %d\n",
3098                        exp->exp_obd->obd_name, le32_to_cpu(lmm->lmv_magic),
3099                        -EIO);
3100                 RETURN(-EIO);
3101         }
3102
3103         if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_V1)
3104                 lsm_size = lmv_stripe_md_size(lmv_mds_md_stripe_count_get(lmm));
3105         else
3106                 /**
3107                  * Unpack default dirstripe(lmv_user_md) to lmv_stripe_md,
3108                  * stripecount should be 0 then.
3109                  */
3110                 lsm_size = lmv_stripe_md_size(0);
3111
3112         lsm_size = lmv_stripe_md_size(lmv_mds_md_stripe_count_get(lmm));
3113         if (lsm == NULL) {
3114                 OBD_ALLOC(lsm, lsm_size);
3115                 if (lsm == NULL)
3116                         RETURN(-ENOMEM);
3117                 allocated = true;
3118                 *lsmp = lsm;
3119         }
3120
3121         switch (le32_to_cpu(lmm->lmv_magic)) {
3122         case LMV_MAGIC_V1:
3123                 rc = lmv_unpack_md_v1(exp, lsm, &lmm->lmv_md_v1);
3124                 break;
3125         default:
3126                 CERROR("%s: unrecognized magic %x\n", exp->exp_obd->obd_name,
3127                        le32_to_cpu(lmm->lmv_magic));
3128                 rc = -EINVAL;
3129                 break;
3130         }
3131
3132         if (rc != 0 && allocated) {
3133                 OBD_FREE(lsm, lsm_size);
3134                 *lsmp = NULL;
3135                 lsm_size = rc;
3136         }
3137         RETURN(lsm_size);
3138 }
3139
3140 int lmv_alloc_memmd(struct lmv_stripe_md **lsmp, int stripes)
3141 {
3142         return lmv_unpack_md(NULL, lsmp, NULL, stripes);
3143 }
3144 EXPORT_SYMBOL(lmv_alloc_memmd);
3145
3146 void lmv_free_memmd(struct lmv_stripe_md *lsm)
3147 {
3148         lmv_unpack_md(NULL, &lsm, NULL, 0);
3149 }
3150 EXPORT_SYMBOL(lmv_free_memmd);
3151
3152 int lmv_unpackmd(struct obd_export *exp, struct lov_stripe_md **lsmp,
3153                  struct lov_mds_md *lmm, int disk_len)
3154 {
3155         return lmv_unpack_md(exp, (struct lmv_stripe_md **)lsmp,
3156                              (union lmv_mds_md *)lmm, disk_len);
3157 }
3158
3159 int lmv_packmd(struct obd_export *exp, struct lov_mds_md **lmmp,
3160                struct lov_stripe_md *lsm)
3161 {
3162         struct obd_device               *obd = exp->exp_obd;
3163         struct lmv_obd                  *lmv_obd = &obd->u.lmv;
3164         const struct lmv_stripe_md      *lmv = (struct lmv_stripe_md *)lsm;
3165         int                             stripe_count;
3166
3167         if (lmmp == NULL) {
3168                 if (lsm != NULL)
3169                         stripe_count = lmv->lsm_md_stripe_count;
3170                 else
3171                         stripe_count = lmv_obd->desc.ld_tgt_count;
3172
3173                 return lmv_mds_md_size(stripe_count, LMV_MAGIC_V1);
3174         }
3175
3176         return lmv_pack_md((union lmv_mds_md **)lmmp, lmv, 0);
3177 }
3178
3179 static int lmv_cancel_unused(struct obd_export *exp, const struct lu_fid *fid,
3180                              ldlm_policy_data_t *policy, ldlm_mode_t mode,
3181                              ldlm_cancel_flags_t flags, void *opaque)
3182 {
3183         struct obd_device       *obd = exp->exp_obd;
3184         struct lmv_obd          *lmv = &obd->u.lmv;
3185         int                      rc = 0;
3186         int                      err;
3187         __u32                    i;
3188         ENTRY;
3189
3190         LASSERT(fid != NULL);
3191
3192         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
3193                 struct lmv_tgt_desc *tgt = lmv->tgts[i];
3194
3195                 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active)
3196                         continue;
3197
3198                 err = md_cancel_unused(tgt->ltd_exp, fid, policy, mode, flags,
3199                                        opaque);
3200                 if (!rc)
3201                         rc = err;
3202         }
3203         RETURN(rc);
3204 }
3205
3206 int lmv_set_lock_data(struct obd_export *exp, __u64 *lockh, void *data,
3207                       __u64 *bits)
3208 {
3209         struct lmv_obd          *lmv = &exp->exp_obd->u.lmv;
3210         struct lmv_tgt_desc     *tgt = lmv->tgts[0];
3211         int                      rc;
3212         ENTRY;
3213
3214         if (tgt == NULL || tgt->ltd_exp == NULL)
3215                 RETURN(-EINVAL);
3216         rc =  md_set_lock_data(tgt->ltd_exp, lockh, data, bits);
3217         RETURN(rc);
3218 }
3219
3220 ldlm_mode_t lmv_lock_match(struct obd_export *exp, __u64 flags,
3221                            const struct lu_fid *fid, ldlm_type_t type,
3222                            ldlm_policy_data_t *policy, ldlm_mode_t mode,
3223                            struct lustre_handle *lockh)
3224 {
3225         struct obd_device       *obd = exp->exp_obd;
3226         struct lmv_obd          *lmv = &obd->u.lmv;
3227         ldlm_mode_t             rc;
3228         int                     tgt;
3229         int                     i;
3230         ENTRY;
3231
3232         CDEBUG(D_INODE, "Lock match for "DFID"\n", PFID(fid));
3233
3234         /*
3235          * With DNE every object can have two locks in different namespaces:
3236          * lookup lock in space of MDT storing direntry and update/open lock in
3237          * space of MDT storing inode.  Try the MDT that the FID maps to first,
3238          * since this can be easily found, and only try others if that fails.
3239          */
3240         for (i = 0, tgt = lmv_find_target_index(lmv, fid);
3241              i < lmv->desc.ld_tgt_count;
3242              i++, tgt = (tgt + 1) % lmv->desc.ld_tgt_count) {
3243                 if (tgt < 0) {
3244                         CDEBUG(D_HA, "%s: "DFID" is inaccessible: rc = %d\n",
3245                                obd->obd_name, PFID(fid), tgt);
3246                         tgt = 0;
3247                 }
3248
3249                 if (lmv->tgts[tgt] == NULL ||
3250                     lmv->tgts[tgt]->ltd_exp == NULL ||
3251                     lmv->tgts[tgt]->ltd_active == 0)
3252                         continue;
3253
3254                 rc = md_lock_match(lmv->tgts[tgt]->ltd_exp, flags, fid,
3255                                    type, policy, mode, lockh);
3256                 if (rc)
3257                         RETURN(rc);
3258         }
3259
3260         RETURN(0);
3261 }
3262
3263 int lmv_get_lustre_md(struct obd_export *exp, struct ptlrpc_request *req,
3264                       struct obd_export *dt_exp, struct obd_export *md_exp,
3265                       struct lustre_md *md)
3266 {
3267         struct lmv_obd          *lmv = &exp->exp_obd->u.lmv;
3268         struct lmv_tgt_desc     *tgt = lmv->tgts[0];
3269
3270         if (tgt == NULL || tgt->ltd_exp == NULL)
3271                 RETURN(-EINVAL);
3272
3273         return md_get_lustre_md(lmv->tgts[0]->ltd_exp, req, dt_exp, md_exp, md);
3274 }
3275
3276 int lmv_free_lustre_md(struct obd_export *exp, struct lustre_md *md)
3277 {
3278         struct obd_device       *obd = exp->exp_obd;
3279         struct lmv_obd          *lmv = &obd->u.lmv;
3280         struct lmv_tgt_desc     *tgt = lmv->tgts[0];
3281         ENTRY;
3282
3283         if (md->lmv != NULL) {
3284                 lmv_free_memmd(md->lmv);
3285                 md->lmv = NULL;
3286         }
3287         if (tgt == NULL || tgt->ltd_exp == NULL)
3288                 RETURN(-EINVAL);
3289         RETURN(md_free_lustre_md(lmv->tgts[0]->ltd_exp, md));
3290 }
3291
3292 int lmv_set_open_replay_data(struct obd_export *exp,
3293                              struct obd_client_handle *och,
3294                              struct lookup_intent *it)
3295 {
3296         struct obd_device       *obd = exp->exp_obd;
3297         struct lmv_obd          *lmv = &obd->u.lmv;
3298         struct lmv_tgt_desc     *tgt;
3299         ENTRY;
3300
3301         tgt = lmv_find_target(lmv, &och->och_fid);
3302         if (IS_ERR(tgt))
3303                 RETURN(PTR_ERR(tgt));
3304
3305         RETURN(md_set_open_replay_data(tgt->ltd_exp, och, it));
3306 }
3307
3308 int lmv_clear_open_replay_data(struct obd_export *exp,
3309                                struct obd_client_handle *och)
3310 {
3311         struct obd_device       *obd = exp->exp_obd;
3312         struct lmv_obd          *lmv = &obd->u.lmv;
3313         struct lmv_tgt_desc     *tgt;
3314         ENTRY;
3315
3316         tgt = lmv_find_target(lmv, &och->och_fid);
3317         if (IS_ERR(tgt))
3318                 RETURN(PTR_ERR(tgt));
3319
3320         RETURN(md_clear_open_replay_data(tgt->ltd_exp, och));
3321 }
3322
3323 static int lmv_get_remote_perm(struct obd_export *exp,
3324                                const struct lu_fid *fid,
3325                                struct obd_capa *oc, __u32 suppgid,
3326                                struct ptlrpc_request **request)
3327 {
3328         struct obd_device       *obd = exp->exp_obd;
3329         struct lmv_obd          *lmv = &obd->u.lmv;
3330         struct lmv_tgt_desc     *tgt;
3331         int                      rc;
3332         ENTRY;
3333
3334         rc = lmv_check_connect(obd);
3335         if (rc)
3336                 RETURN(rc);
3337
3338         tgt = lmv_find_target(lmv, fid);
3339         if (IS_ERR(tgt))
3340                 RETURN(PTR_ERR(tgt));
3341
3342         rc = md_get_remote_perm(tgt->ltd_exp, fid, oc, suppgid, request);
3343         RETURN(rc);
3344 }
3345
3346 static int lmv_renew_capa(struct obd_export *exp, struct obd_capa *oc,
3347                           renew_capa_cb_t cb)
3348 {
3349         struct obd_device       *obd = exp->exp_obd;
3350         struct lmv_obd          *lmv = &obd->u.lmv;
3351         struct lmv_tgt_desc     *tgt;
3352         int                      rc;
3353         ENTRY;
3354
3355         rc = lmv_check_connect(obd);
3356         if (rc)
3357                 RETURN(rc);
3358
3359         tgt = lmv_find_target(lmv, &oc->c_capa.lc_fid);
3360         if (IS_ERR(tgt))
3361                 RETURN(PTR_ERR(tgt));
3362
3363         rc = md_renew_capa(tgt->ltd_exp, oc, cb);
3364         RETURN(rc);
3365 }
3366
3367 int lmv_unpack_capa(struct obd_export *exp, struct ptlrpc_request *req,
3368                     const struct req_msg_field *field, struct obd_capa **oc)
3369 {
3370         struct lmv_obd          *lmv = &exp->exp_obd->u.lmv;
3371         struct lmv_tgt_desc     *tgt = lmv->tgts[0];
3372
3373         if (tgt == NULL || tgt->ltd_exp == NULL)
3374                 RETURN(-EINVAL);
3375         return md_unpack_capa(tgt->ltd_exp, req, field, oc);
3376 }
3377
3378 int lmv_intent_getattr_async(struct obd_export *exp,
3379                              struct md_enqueue_info *minfo,
3380                              struct ldlm_enqueue_info *einfo)
3381 {
3382         struct md_op_data       *op_data = &minfo->mi_data;
3383         struct obd_device       *obd = exp->exp_obd;
3384         struct lmv_obd          *lmv = &obd->u.lmv;
3385         struct lmv_tgt_desc     *tgt = NULL;
3386         int                      rc;
3387         ENTRY;
3388
3389         rc = lmv_check_connect(obd);
3390         if (rc)
3391                 RETURN(rc);
3392
3393         tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
3394         if (IS_ERR(tgt))
3395                 RETURN(PTR_ERR(tgt));
3396
3397         rc = md_intent_getattr_async(tgt->ltd_exp, minfo, einfo);
3398         RETURN(rc);
3399 }
3400
3401 int lmv_revalidate_lock(struct obd_export *exp, struct lookup_intent *it,
3402                         struct lu_fid *fid, __u64 *bits)
3403 {
3404         struct obd_device       *obd = exp->exp_obd;
3405         struct lmv_obd          *lmv = &obd->u.lmv;
3406         struct lmv_tgt_desc     *tgt;
3407         int                      rc;
3408         ENTRY;
3409
3410         rc = lmv_check_connect(obd);
3411         if (rc)
3412                 RETURN(rc);
3413
3414         tgt = lmv_find_target(lmv, fid);
3415         if (IS_ERR(tgt))
3416                 RETURN(PTR_ERR(tgt));
3417
3418         rc = md_revalidate_lock(tgt->ltd_exp, it, fid, bits);
3419         RETURN(rc);
3420 }
3421
3422 int lmv_get_fid_from_lsm(struct obd_export *exp,
3423                          const struct lmv_stripe_md *lsm,
3424                          const char *name, int namelen, struct lu_fid *fid)
3425 {
3426         const struct lmv_oinfo *oinfo;
3427
3428         LASSERT(lsm != NULL);
3429         oinfo = lsm_name_to_stripe_info(lsm, name, namelen);
3430         if (IS_ERR(oinfo))
3431                 return PTR_ERR(oinfo);
3432
3433         *fid = oinfo->lmo_fid;
3434
3435         RETURN(0);
3436 }
3437
3438 /**
3439  * For lmv, only need to send request to master MDT, and the master MDT will
3440  * process with other slave MDTs. The only exception is Q_GETOQUOTA for which
3441  * we directly fetch data from the slave MDTs.
3442  */
3443 int lmv_quotactl(struct obd_device *unused, struct obd_export *exp,
3444                  struct obd_quotactl *oqctl)
3445 {
3446         struct obd_device   *obd = class_exp2obd(exp);
3447         struct lmv_obd      *lmv = &obd->u.lmv;
3448         struct lmv_tgt_desc *tgt = lmv->tgts[0];
3449         int                  rc = 0;
3450         __u32                i;
3451         __u64                curspace, curinodes;
3452         ENTRY;
3453
3454         if (tgt == NULL ||
3455             tgt->ltd_exp == NULL ||
3456             !tgt->ltd_active ||
3457             lmv->desc.ld_tgt_count == 0) {
3458                 CERROR("master lmv inactive\n");
3459                 RETURN(-EIO);
3460         }
3461
3462         if (oqctl->qc_cmd != Q_GETOQUOTA) {
3463                 rc = obd_quotactl(tgt->ltd_exp, oqctl);
3464                 RETURN(rc);
3465         }
3466
3467         curspace = curinodes = 0;
3468         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
3469                 int err;
3470                 tgt = lmv->tgts[i];
3471
3472                 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active)
3473                         continue;
3474
3475                 err = obd_quotactl(tgt->ltd_exp, oqctl);
3476                 if (err) {
3477                         CERROR("getquota on mdt %d failed. %d\n", i, err);
3478                         if (!rc)
3479                                 rc = err;
3480                 } else {
3481                         curspace += oqctl->qc_dqblk.dqb_curspace;
3482                         curinodes += oqctl->qc_dqblk.dqb_curinodes;
3483                 }
3484         }
3485         oqctl->qc_dqblk.dqb_curspace = curspace;
3486         oqctl->qc_dqblk.dqb_curinodes = curinodes;
3487
3488         RETURN(rc);
3489 }
3490
3491 int lmv_quotacheck(struct obd_device *unused, struct obd_export *exp,
3492                    struct obd_quotactl *oqctl)
3493 {
3494         struct obd_device       *obd = class_exp2obd(exp);
3495         struct lmv_obd          *lmv = &obd->u.lmv;
3496         struct lmv_tgt_desc     *tgt;
3497         __u32                    i;
3498         int                      rc = 0;
3499         ENTRY;
3500
3501         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
3502                 int err;
3503                 tgt = lmv->tgts[i];
3504                 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active) {
3505                         CERROR("lmv idx %d inactive\n", i);
3506                         RETURN(-EIO);
3507                 }
3508
3509                 err = obd_quotacheck(tgt->ltd_exp, oqctl);
3510                 if (err && !rc)
3511                         rc = err;
3512         }
3513
3514         RETURN(rc);
3515 }
3516
3517 int lmv_update_lsm_md(struct obd_export *exp, struct lmv_stripe_md *lsm,
3518                       struct mdt_body *body, ldlm_blocking_callback cb_blocking)
3519 {
3520         return lmv_revalidate_slaves(exp, body, lsm, cb_blocking, 0);
3521 }
3522
3523 int lmv_merge_attr(struct obd_export *exp, const struct lmv_stripe_md *lsm,
3524                    struct cl_attr *attr)
3525 {
3526         int i;
3527
3528         for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
3529                 struct inode *inode = lsm->lsm_md_oinfo[i].lmo_root;
3530
3531                 CDEBUG(D_INFO, ""DFID" size %llu, nlink %u, atime %lu ctime"
3532                        "%lu, mtime %lu.\n", PFID(&lsm->lsm_md_oinfo[i].lmo_fid),
3533                        i_size_read(inode), inode->i_nlink,
3534                        LTIME_S(inode->i_atime), LTIME_S(inode->i_ctime),
3535                        LTIME_S(inode->i_mtime));
3536
3537                 /* for slave stripe, it needs to subtract nlink for . and .. */
3538                 if (i != 0)
3539                         attr->cat_nlink += inode->i_nlink - 2;
3540                 else
3541                         attr->cat_nlink = inode->i_nlink;
3542
3543                 attr->cat_size += i_size_read(inode);
3544
3545                 if (attr->cat_atime < LTIME_S(inode->i_atime))
3546                         attr->cat_atime = LTIME_S(inode->i_atime);
3547
3548                 if (attr->cat_ctime < LTIME_S(inode->i_ctime))
3549                         attr->cat_ctime = LTIME_S(inode->i_ctime);
3550
3551                 if (attr->cat_mtime < LTIME_S(inode->i_mtime))
3552                         attr->cat_mtime = LTIME_S(inode->i_mtime);
3553         }
3554         return 0;
3555 }
3556
3557 struct obd_ops lmv_obd_ops = {
3558         .o_owner                = THIS_MODULE,
3559         .o_setup                = lmv_setup,
3560         .o_cleanup              = lmv_cleanup,
3561         .o_precleanup           = lmv_precleanup,
3562         .o_process_config       = lmv_process_config,
3563         .o_connect              = lmv_connect,
3564         .o_disconnect           = lmv_disconnect,
3565         .o_statfs               = lmv_statfs,
3566         .o_get_info             = lmv_get_info,
3567         .o_set_info_async       = lmv_set_info_async,
3568         .o_packmd               = lmv_packmd,
3569         .o_unpackmd             = lmv_unpackmd,
3570         .o_notify               = lmv_notify,
3571         .o_get_uuid             = lmv_get_uuid,
3572         .o_iocontrol            = lmv_iocontrol,
3573         .o_quotacheck           = lmv_quotacheck,
3574         .o_quotactl             = lmv_quotactl
3575 };
3576
3577 struct md_ops lmv_md_ops = {
3578         .m_getstatus            = lmv_getstatus,
3579         .m_null_inode           = lmv_null_inode,
3580         .m_find_cbdata          = lmv_find_cbdata,
3581         .m_close                = lmv_close,
3582         .m_create               = lmv_create,
3583         .m_done_writing         = lmv_done_writing,
3584         .m_enqueue              = lmv_enqueue,
3585         .m_getattr              = lmv_getattr,
3586         .m_getxattr             = lmv_getxattr,
3587         .m_getattr_name         = lmv_getattr_name,
3588         .m_intent_lock          = lmv_intent_lock,
3589         .m_link                 = lmv_link,
3590         .m_rename               = lmv_rename,
3591         .m_setattr              = lmv_setattr,
3592         .m_setxattr             = lmv_setxattr,
3593         .m_fsync                = lmv_fsync,
3594         .m_read_page            = lmv_read_page,
3595         .m_unlink               = lmv_unlink,
3596         .m_init_ea_size         = lmv_init_ea_size,
3597         .m_cancel_unused        = lmv_cancel_unused,
3598         .m_set_lock_data        = lmv_set_lock_data,
3599         .m_lock_match           = lmv_lock_match,
3600         .m_get_lustre_md        = lmv_get_lustre_md,
3601         .m_free_lustre_md       = lmv_free_lustre_md,
3602         .m_update_lsm_md        = lmv_update_lsm_md,
3603         .m_merge_attr           = lmv_merge_attr,
3604         .m_set_open_replay_data = lmv_set_open_replay_data,
3605         .m_clear_open_replay_data = lmv_clear_open_replay_data,
3606         .m_renew_capa           = lmv_renew_capa,
3607         .m_unpack_capa          = lmv_unpack_capa,
3608         .m_get_remote_perm      = lmv_get_remote_perm,
3609         .m_intent_getattr_async = lmv_intent_getattr_async,
3610         .m_revalidate_lock      = lmv_revalidate_lock,
3611         .m_get_fid_from_lsm     = lmv_get_fid_from_lsm,
3612 };
3613
3614 int __init lmv_init(void)
3615 {
3616         return class_register_type(&lmv_obd_ops, &lmv_md_ops, true, NULL,
3617                                    LUSTRE_LMV_NAME, NULL);
3618 }
3619
3620 static void lmv_exit(void)
3621 {
3622         class_unregister_type(LUSTRE_LMV_NAME);
3623 }
3624
3625 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
3626 MODULE_DESCRIPTION("Lustre Logical Metadata Volume OBD driver");
3627 MODULE_LICENSE("GPL");
3628
3629 module_init(lmv_init);
3630 module_exit(lmv_exit);