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, 2012, 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 #ifdef __KERNEL__
  39 #include <linux/slab.h>
  40 #include <linux/module.h>
  41 #include <linux/init.h>
  42 #include <linux/slab.h>
  43 #include <linux/pagemap.h>
  44 #include <linux/mm.h>
  45 #include <asm/div64.h>
  46 #include <linux/seq_file.h>
  47 #include <linux/namei.h>
  48 #else
  49 #include <liblustre.h>
  50 #endif
  51
  52 #include <lustre/lustre_idl.h>
  53 #include <obd_support.h>
  54 #include <lustre_lib.h>
  55 #include <lustre_net.h>
  56 #include <obd_class.h>
  57 #include <lprocfs_status.h>
  58 #include <lustre_lite.h>
  59 #include <lustre_fid.h>
  60 #include "lmv_internal.h"
  61
  62 static void lmv_activate_target(struct lmv_obd *lmv,
  63                                 struct lmv_tgt_desc *tgt,
  64                                 int activate)
  65 {
  66         if (tgt->ltd_active == activate)
  67                 return;
  68
  69         tgt->ltd_active = activate;
  70         lmv->desc.ld_active_tgt_count += (activate ? 1 : -1);
  71 }
  72
  73 /**
  74  * Error codes:
  75  *
  76  *  -EINVAL  : UUID can't be found in the LMV's target list
  77  *  -ENOTCONN: The UUID is found, but the target connection is bad (!)
  78  *  -EBADF   : The UUID is found, but the OBD of the wrong type (!)
  79  */
  80 static int lmv_set_mdc_active(struct lmv_obd *lmv, struct obd_uuid *uuid,
  81                               int activate)
  82 {
  83         struct lmv_tgt_desc    *tgt;
  84         struct obd_device      *obd;
  85         int                     i;
  86         int                     rc = 0;
  87         ENTRY;
  88
  89         CDEBUG(D_INFO, "Searching in lmv %p for uuid %s (activate=%d)\n",
  90                lmv, uuid->uuid, activate);
  91
  92         spin_lock(&lmv->lmv_lock);
  93         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
  94                 tgt = lmv->tgts[i];
  95                 if (tgt == NULL || tgt->ltd_exp == NULL)
  96                         continue;
  97
  98                 CDEBUG(D_INFO, "Target idx %d is %s conn "LPX64"\n", i,
  99                        tgt->ltd_uuid.uuid, tgt->ltd_exp->exp_handle.h_cookie);
 100
 101                 if (obd_uuid_equals(uuid, &tgt->ltd_uuid))
 102                         break;
 103         }
 104
 105         if (i == lmv->desc.ld_tgt_count)
 106                 GOTO(out_lmv_lock, rc = -EINVAL);
 107
 108         obd = class_exp2obd(tgt->ltd_exp);
 109         if (obd == NULL)
 110                 GOTO(out_lmv_lock, rc = -ENOTCONN);
 111
 112         CDEBUG(D_INFO, "Found OBD %s=%s device %d (%p) type %s at LMV idx %d\n",
 113                obd->obd_name, obd->obd_uuid.uuid, obd->obd_minor, obd,
 114                obd->obd_type->typ_name, i);
 115         LASSERT(strcmp(obd->obd_type->typ_name, LUSTRE_MDC_NAME) == 0);
 116
 117         if (tgt->ltd_active == activate) {
 118                 CDEBUG(D_INFO, "OBD %p already %sactive!\n", obd,
 119                        activate ? "" : "in");
 120                 GOTO(out_lmv_lock, rc);
 121         }
 122
 123         CDEBUG(D_INFO, "Marking OBD %p %sactive\n", obd,
 124                activate ? "" : "in");
 125         lmv_activate_target(lmv, tgt, activate);
 126         EXIT;
 127
 128  out_lmv_lock:
 129         spin_unlock(&lmv->lmv_lock);
 130         return rc;
 131 }
 132
 133 struct obd_uuid *lmv_get_uuid(struct obd_export *exp)
 134 {
 135         struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
 136
 137         return obd_get_uuid(lmv->tgts[0]->ltd_exp);
 138 }
 139
 140 static int lmv_notify(struct obd_device *obd, struct obd_device *watched,
 141                       enum obd_notify_event ev, void *data)
 142 {
 143         struct obd_connect_data *conn_data;
 144         struct lmv_obd          *lmv = &obd->u.lmv;
 145         struct obd_uuid         *uuid;
 146         int                      rc = 0;
 147         ENTRY;
 148
 149         if (strcmp(watched->obd_type->typ_name, LUSTRE_MDC_NAME)) {
 150                 CERROR("unexpected notification of %s %s!\n",
 151                        watched->obd_type->typ_name,
 152                        watched->obd_name);
 153                 RETURN(-EINVAL);
 154         }
 155
 156         uuid = &watched->u.cli.cl_target_uuid;
 157         if (ev == OBD_NOTIFY_ACTIVE || ev == OBD_NOTIFY_INACTIVE) {
 158                 /*
 159                  * Set MDC as active before notifying the observer, so the
 160                  * observer can use the MDC normally.
 161                  */
 162                 rc = lmv_set_mdc_active(lmv, uuid,
 163                                         ev == OBD_NOTIFY_ACTIVE);
 164                 if (rc) {
 165                         CERROR("%sactivation of %s failed: %d\n",
 166                                ev == OBD_NOTIFY_ACTIVE ? "" : "de",
 167                                uuid->uuid, rc);
 168                         RETURN(rc);
 169                 }
 170         } else if (ev == OBD_NOTIFY_OCD) {
 171                 conn_data = &watched->u.cli.cl_import->imp_connect_data;
 172                 /*
 173                  * XXX: Make sure that ocd_connect_flags from all targets are
 174                  * the same. Otherwise one of MDTs runs wrong version or
 175                  * something like this.  --umka
 176                  */
 177                 obd->obd_self_export->exp_connect_data = *conn_data;
 178         }
 179 #if 0
 180         else if (ev == OBD_NOTIFY_DISCON) {
 181                 /*
 182                  * For disconnect event, flush fld cache for failout MDS case.
 183                  */
 184                 fld_client_flush(&lmv->lmv_fld);
 185         }
 186 #endif
 187         /*
 188          * Pass the notification up the chain.
 189          */
 190         if (obd->obd_observer)
 191                 rc = obd_notify(obd->obd_observer, watched, ev, data);
 192
 193         RETURN(rc);
 194 }
 195
 196 /**
 197  * This is fake connect function. Its purpose is to initialize lmv and say
 198  * caller that everything is okay. Real connection will be performed later.
 199  */
 200 static int lmv_connect(const struct lu_env *env,
 201                        struct obd_export **exp, struct obd_device *obd,
 202                        struct obd_uuid *cluuid, struct obd_connect_data *data,
 203                        void *localdata)
 204 {
 205 #ifdef __KERNEL__
 206         struct proc_dir_entry *lmv_proc_dir;
 207 #endif
 208         struct lmv_obd        *lmv = &obd->u.lmv;
 209         struct lustre_handle  conn = { 0 };
 210         int                    rc = 0;
 211         ENTRY;
 212
 213         /*
 214          * We don't want to actually do the underlying connections more than
 215          * once, so keep track.
 216          */
 217         lmv->refcount++;
 218         if (lmv->refcount > 1) {
 219                 *exp = NULL;
 220                 RETURN(0);
 221         }
 222
 223         rc = class_connect(&conn, obd, cluuid);
 224         if (rc) {
 225                 CERROR("class_connection() returned %d\n", rc);
 226                 RETURN(rc);
 227         }
 228
 229         *exp = class_conn2export(&conn);
 230         class_export_get(*exp);
 231
 232         lmv->exp = *exp;
 233         lmv->connected = 0;
 234         lmv->cluuid = *cluuid;
 235
 236         if (data)
 237                 lmv->conn_data = *data;
 238
 239 #ifdef __KERNEL__
 240         lmv_proc_dir = lprocfs_register("target_obds", obd->obd_proc_entry,
 241                                         NULL, NULL);
 242         if (IS_ERR(lmv_proc_dir)) {
 243                 CERROR("could not register /proc/fs/lustre/%s/%s/target_obds.",
 244                        obd->obd_type->typ_name, obd->obd_name);
 245                 lmv_proc_dir = NULL;
 246         }
 247 #endif
 248
 249         /*
 250          * All real clients should perform actual connection right away, because
 251          * it is possible, that LMV will not have opportunity to connect targets
 252          * and MDC stuff will be called directly, for instance while reading
 253          * ../mdc/../kbytesfree procfs file, etc.
 254          */
 255         if (data->ocd_connect_flags & OBD_CONNECT_REAL)
 256                 rc = lmv_check_connect(obd);
 257
 258 #ifdef __KERNEL__
 259         if (rc) {
 260                 if (lmv_proc_dir)
 261                         lprocfs_remove(&lmv_proc_dir);
 262         }
 263 #endif
 264
 265         RETURN(rc);
 266 }
 267
 268 static void lmv_set_timeouts(struct obd_device *obd)
 269 {
 270         struct lmv_tgt_desc   *tgt;
 271         struct lmv_obd        *lmv;
 272         int                    i;
 273
 274         lmv = &obd->u.lmv;
 275         if (lmv->server_timeout == 0)
 276                 return;
 277
 278         if (lmv->connected == 0)
 279                 return;
 280
 281         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 282                 tgt = lmv->tgts[i];
 283                 if (tgt == NULL || tgt->ltd_exp == NULL || tgt->ltd_active == 0)
 284                         continue;
 285
 286                 obd_set_info_async(NULL, tgt->ltd_exp, sizeof(KEY_INTERMDS),
 287                                    KEY_INTERMDS, 0, NULL, NULL);
 288         }
 289 }
 290
 291 static int lmv_init_ea_size(struct obd_export *exp, int easize,
 292                             int def_easize, int cookiesize)
 293 {
 294         struct obd_device   *obd = exp->exp_obd;
 295         struct lmv_obd      *lmv = &obd->u.lmv;
 296         int                  i;
 297         int                  rc = 0;
 298         int                  change = 0;
 299         ENTRY;
 300
 301         if (lmv->max_easize < easize) {
 302                 lmv->max_easize = easize;
 303                 change = 1;
 304         }
 305         if (lmv->max_def_easize < def_easize) {
 306                 lmv->max_def_easize = def_easize;
 307                 change = 1;
 308         }
 309         if (lmv->max_cookiesize < cookiesize) {
 310                 lmv->max_cookiesize = cookiesize;
 311                 change = 1;
 312         }
 313         if (change == 0)
 314                 RETURN(0);
 315
 316         if (lmv->connected == 0)
 317                 RETURN(0);
 318
 319         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 320                 if (lmv->tgts[i] == NULL ||
 321                     lmv->tgts[i]->ltd_exp == NULL ||
 322                     lmv->tgts[i]->ltd_active == 0) {
 323                         CWARN("%s: NULL export for %d\n", obd->obd_name, i);
 324                         continue;
 325                 }
 326
 327                 rc = md_init_ea_size(lmv->tgts[i]->ltd_exp, easize, def_easize,
 328                                      cookiesize);
 329                 if (rc) {
 330                         CERROR("%s: obd_init_ea_size() failed on MDT target %d:"
 331                                " rc = %d.\n", obd->obd_name, i, rc);
 332                         break;
 333                 }
 334         }
 335         RETURN(rc);
 336 }
 337
 338 #define MAX_STRING_SIZE 128
 339
 340 int lmv_connect_mdc(struct obd_device *obd, struct lmv_tgt_desc *tgt)
 341 {
 342 #ifdef __KERNEL__
 343         struct proc_dir_entry   *lmv_proc_dir;
 344 #endif
 345         struct lmv_obd          *lmv = &obd->u.lmv;
 346         struct obd_uuid         *cluuid = &lmv->cluuid;
 347         struct obd_connect_data *mdc_data = NULL;
 348         struct obd_uuid          lmv_mdc_uuid = { "LMV_MDC_UUID" };
 349         struct obd_device       *mdc_obd;
 350         struct obd_export       *mdc_exp;
 351         struct lu_fld_target     target;
 352         int                      rc;
 353         ENTRY;
 354
 355         mdc_obd = class_find_client_obd(&tgt->ltd_uuid, LUSTRE_MDC_NAME,
 356                                         &obd->obd_uuid);
 357         if (!mdc_obd) {
 358                 CERROR("target %s not attached\n", tgt->ltd_uuid.uuid);
 359                 RETURN(-EINVAL);
 360         }
 361
 362         CDEBUG(D_CONFIG, "connect to %s(%s) - %s, %s FOR %s\n",
 363                 mdc_obd->obd_name, mdc_obd->obd_uuid.uuid,
 364                 tgt->ltd_uuid.uuid, obd->obd_uuid.uuid,
 365                 cluuid->uuid);
 366
 367         if (!mdc_obd->obd_set_up) {
 368                 CERROR("target %s is not set up\n", tgt->ltd_uuid.uuid);
 369                 RETURN(-EINVAL);
 370         }
 371
 372         rc = obd_connect(NULL, &mdc_exp, mdc_obd, &lmv_mdc_uuid,
 373                          &lmv->conn_data, NULL);
 374         if (rc) {
 375                 CERROR("target %s connect error %d\n", tgt->ltd_uuid.uuid, rc);
 376                 RETURN(rc);
 377         }
 378
 379         /*
 380          * Init fid sequence client for this mdc and add new fld target.
 381          */
 382         rc = obd_fid_init(mdc_obd, mdc_exp, LUSTRE_SEQ_METADATA);
 383         if (rc)
 384                 RETURN(rc);
 385
 386         target.ft_srv = NULL;
 387         target.ft_exp = mdc_exp;
 388         target.ft_idx = tgt->ltd_idx;
 389
 390         fld_client_add_target(&lmv->lmv_fld, &target);
 391
 392         mdc_data = &class_exp2cliimp(mdc_exp)->imp_connect_data;
 393
 394         rc = obd_register_observer(mdc_obd, obd);
 395         if (rc) {
 396                 obd_disconnect(mdc_exp);
 397                 CERROR("target %s register_observer error %d\n",
 398                        tgt->ltd_uuid.uuid, rc);
 399                 RETURN(rc);
 400         }
 401
 402         if (obd->obd_observer) {
 403                 /*
 404                  * Tell the observer about the new target.
 405                  */
 406                 rc = obd_notify(obd->obd_observer, mdc_exp->exp_obd,
 407                                 OBD_NOTIFY_ACTIVE,
 408                                 (void *)(tgt - lmv->tgts[0]));
 409                 if (rc) {
 410                         obd_disconnect(mdc_exp);
 411                         RETURN(rc);
 412                 }
 413         }
 414
 415         tgt->ltd_active = 1;
 416         tgt->ltd_exp = mdc_exp;
 417         lmv->desc.ld_active_tgt_count++;
 418
 419         md_init_ea_size(tgt->ltd_exp, lmv->max_easize,
 420                         lmv->max_def_easize, lmv->max_cookiesize);
 421
 422         CDEBUG(D_CONFIG, "Connected to %s(%s) successfully (%d)\n",
 423                 mdc_obd->obd_name, mdc_obd->obd_uuid.uuid,
 424                 cfs_atomic_read(&obd->obd_refcount));
 425
 426 #ifdef __KERNEL__
 427         lmv_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds");
 428         if (lmv_proc_dir) {
 429                 struct proc_dir_entry *mdc_symlink;
 430
 431                 LASSERT(mdc_obd->obd_type != NULL);
 432                 LASSERT(mdc_obd->obd_type->typ_name != NULL);
 433                 mdc_symlink = lprocfs_add_symlink(mdc_obd->obd_name,
 434                                                   lmv_proc_dir,
 435                                                   "../../../%s/%s",
 436                                                   mdc_obd->obd_type->typ_name,
 437                                                   mdc_obd->obd_name);
 438                 if (mdc_symlink == NULL) {
 439                         CERROR("Could not register LMV target "
 440                                "/proc/fs/lustre/%s/%s/target_obds/%s.",
 441                                obd->obd_type->typ_name, obd->obd_name,
 442                                mdc_obd->obd_name);
 443                         lprocfs_remove(&lmv_proc_dir);
 444                         lmv_proc_dir = NULL;
 445                 }
 446         }
 447 #endif
 448         RETURN(0);
 449 }
 450
 451 static void lmv_del_target(struct lmv_obd *lmv, int index)
 452 {
 453         if (lmv->tgts[index] == NULL)
 454                 return;
 455
 456         OBD_FREE_PTR(lmv->tgts[index]);
 457         lmv->tgts[index] = NULL;
 458         return;
 459 }
 460
 461 static int lmv_add_target(struct obd_device *obd, struct obd_uuid *uuidp,
 462                            __u32 index, int gen)
 463 {
 464         struct lmv_obd      *lmv = &obd->u.lmv;
 465         struct lmv_tgt_desc *tgt;
 466         int                  rc = 0;
 467         ENTRY;
 468
 469         CDEBUG(D_CONFIG, "Target uuid: %s. index %d\n", uuidp->uuid, index);
 470
 471         lmv_init_lock(lmv);
 472
 473         if (lmv->desc.ld_tgt_count == 0) {
 474                 struct obd_device *mdc_obd;
 475
 476                 mdc_obd = class_find_client_obd(uuidp, LUSTRE_MDC_NAME,
 477                                                 &obd->obd_uuid);
 478                 if (!mdc_obd) {
 479                         lmv_init_unlock(lmv);
 480                         CERROR("%s: Target %s not attached: rc = %d\n",
 481                                obd->obd_name, uuidp->uuid, -EINVAL);
 482                         RETURN(-EINVAL);
 483                 }
 484         }
 485
 486         if ((index < lmv->tgts_size) && (lmv->tgts[index] != NULL)) {
 487                 tgt = lmv->tgts[index];
 488                 CERROR("%s: UUID %s already assigned at LOV target index %d:"
 489                        " rc = %d\n", obd->obd_name,
 490                        obd_uuid2str(&tgt->ltd_uuid), index, -EEXIST);
 491                 lmv_init_unlock(lmv);
 492                 RETURN(-EEXIST);
 493         }
 494
 495         if (index >= lmv->tgts_size) {
 496                 /* We need to reallocate the lmv target array. */
 497                 struct lmv_tgt_desc **newtgts, **old = NULL;
 498                 __u32 newsize = 1;
 499                 __u32 oldsize = 0;
 500
 501                 while (newsize < index + 1)
 502                         newsize = newsize << 1;
 503                 OBD_ALLOC(newtgts, sizeof(*newtgts) * newsize);
 504                 if (newtgts == NULL) {
 505                         lmv_init_unlock(lmv);
 506                         RETURN(-ENOMEM);
 507                 }
 508
 509                 if (lmv->tgts_size) {
 510                         memcpy(newtgts, lmv->tgts,
 511                                sizeof(*newtgts) * lmv->tgts_size);
 512                         old = lmv->tgts;
 513                         oldsize = lmv->tgts_size;
 514                 }
 515
 516                 lmv->tgts = newtgts;
 517                 lmv->tgts_size = newsize;
 518                 smp_rmb();
 519                 if (old)
 520                         OBD_FREE(old, sizeof(*old) * oldsize);
 521
 522                 CDEBUG(D_CONFIG, "tgts: %p size: %d\n", lmv->tgts,
 523                        lmv->tgts_size);
 524         }
 525
 526         OBD_ALLOC_PTR(tgt);
 527         if (!tgt) {
 528                 lmv_init_unlock(lmv);
 529                 RETURN(-ENOMEM);
 530         }
 531
 532         mutex_init(&tgt->ltd_fid_mutex);
 533         tgt->ltd_idx = index;
 534         tgt->ltd_uuid = *uuidp;
 535         tgt->ltd_active = 0;
 536         lmv->tgts[index] = tgt;
 537         if (index >= lmv->desc.ld_tgt_count)
 538                 lmv->desc.ld_tgt_count = index + 1;
 539
 540         if (lmv->connected) {
 541                 rc = lmv_connect_mdc(obd, tgt);
 542                 if (rc) {
 543                         spin_lock(&lmv->lmv_lock);
 544                         lmv->desc.ld_tgt_count--;
 545                         memset(tgt, 0, sizeof(*tgt));
 546                         spin_unlock(&lmv->lmv_lock);
 547                 } else {
 548                         int easize = sizeof(struct lmv_stripe_md) +
 549                                      lmv->desc.ld_tgt_count *
 550                                      sizeof(struct lu_fid);
 551                         lmv_init_ea_size(obd->obd_self_export, easize, 0, 0);
 552                 }
 553         }
 554
 555         lmv_init_unlock(lmv);
 556         RETURN(rc);
 557 }
 558
 559 int lmv_check_connect(struct obd_device *obd)
 560 {
 561         struct lmv_obd       *lmv = &obd->u.lmv;
 562         struct lmv_tgt_desc  *tgt;
 563         int                   i;
 564         int                   rc;
 565         int                   easize;
 566         ENTRY;
 567
 568         if (lmv->connected)
 569                 RETURN(0);
 570
 571         lmv_init_lock(lmv);
 572         if (lmv->connected) {
 573                 lmv_init_unlock(lmv);
 574                 RETURN(0);
 575         }
 576
 577         if (lmv->desc.ld_tgt_count == 0) {
 578                 lmv_init_unlock(lmv);
 579                 CERROR("%s: no targets configured.\n", obd->obd_name);
 580                 RETURN(-EINVAL);
 581         }
 582
 583         CDEBUG(D_CONFIG, "Time to connect %s to %s\n",
 584                lmv->cluuid.uuid, obd->obd_name);
 585
 586         LASSERT(lmv->tgts != NULL);
 587
 588         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 589                 tgt = lmv->tgts[i];
 590                 if (tgt == NULL)
 591                         continue;
 592                 rc = lmv_connect_mdc(obd, tgt);
 593                 if (rc)
 594                         GOTO(out_disc, rc);
 595         }
 596
 597         lmv_set_timeouts(obd);
 598         class_export_put(lmv->exp);
 599         lmv->connected = 1;
 600         easize = lmv_get_easize(lmv);
 601         lmv_init_ea_size(obd->obd_self_export, easize, 0, 0);
 602         lmv_init_unlock(lmv);
 603         RETURN(0);
 604
 605  out_disc:
 606         while (i-- > 0) {
 607                 int rc2;
 608                 tgt = lmv->tgts[i];
 609                 if (tgt == NULL)
 610                         continue;
 611                 tgt->ltd_active = 0;
 612                 if (tgt->ltd_exp) {
 613                         --lmv->desc.ld_active_tgt_count;
 614                         rc2 = obd_disconnect(tgt->ltd_exp);
 615                         if (rc2) {
 616                                 CERROR("LMV target %s disconnect on "
 617                                        "MDC idx %d: error %d\n",
 618                                        tgt->ltd_uuid.uuid, i, rc2);
 619                         }
 620                 }
 621         }
 622         class_disconnect(lmv->exp);
 623         lmv_init_unlock(lmv);
 624         RETURN(rc);
 625 }
 626
 627 static int lmv_disconnect_mdc(struct obd_device *obd, struct lmv_tgt_desc *tgt)
 628 {
 629 #ifdef __KERNEL__
 630         struct proc_dir_entry  *lmv_proc_dir;
 631 #endif
 632         struct lmv_obd         *lmv = &obd->u.lmv;
 633         struct obd_device      *mdc_obd;
 634         int                     rc;
 635         ENTRY;
 636
 637         LASSERT(tgt != NULL);
 638         LASSERT(obd != NULL);
 639
 640         mdc_obd = class_exp2obd(tgt->ltd_exp);
 641
 642         if (mdc_obd) {
 643                 mdc_obd->obd_force = obd->obd_force;
 644                 mdc_obd->obd_fail = obd->obd_fail;
 645                 mdc_obd->obd_no_recov = obd->obd_no_recov;
 646         }
 647
 648 #ifdef __KERNEL__
 649         lmv_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds");
 650         if (lmv_proc_dir) {
 651                 struct proc_dir_entry *mdc_symlink;
 652
 653                 mdc_symlink = lprocfs_srch(lmv_proc_dir, mdc_obd->obd_name);
 654                 if (mdc_symlink) {
 655                         lprocfs_remove(&mdc_symlink);
 656                 } else {
 657                         CERROR("/proc/fs/lustre/%s/%s/target_obds/%s missing\n",
 658                                obd->obd_type->typ_name, obd->obd_name,
 659                                mdc_obd->obd_name);
 660                 }
 661         }
 662 #endif
 663         rc = obd_fid_fini(tgt->ltd_exp->exp_obd);
 664         if (rc)
 665                 CERROR("Can't finanize fids factory\n");
 666
 667         CDEBUG(D_INFO, "Disconnected from %s(%s) successfully\n",
 668                tgt->ltd_exp->exp_obd->obd_name,
 669                tgt->ltd_exp->exp_obd->obd_uuid.uuid);
 670
 671         obd_register_observer(tgt->ltd_exp->exp_obd, NULL);
 672         rc = obd_disconnect(tgt->ltd_exp);
 673         if (rc) {
 674                 if (tgt->ltd_active) {
 675                         CERROR("Target %s disconnect error %d\n",
 676                                tgt->ltd_uuid.uuid, rc);
 677                 }
 678         }
 679
 680         lmv_activate_target(lmv, tgt, 0);
 681         tgt->ltd_exp = NULL;
 682         RETURN(0);
 683 }
 684
 685 static int lmv_disconnect(struct obd_export *exp)
 686 {
 687         struct obd_device     *obd = class_exp2obd(exp);
 688 #ifdef __KERNEL__
 689         struct proc_dir_entry *lmv_proc_dir;
 690 #endif
 691         struct lmv_obd        *lmv = &obd->u.lmv;
 692         int                    rc;
 693         int                    i;
 694         ENTRY;
 695
 696         if (!lmv->tgts)
 697                 goto out_local;
 698
 699         /*
 700          * Only disconnect the underlying layers on the final disconnect.
 701          */
 702         lmv->refcount--;
 703         if (lmv->refcount != 0)
 704                 goto out_local;
 705
 706         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 707                 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
 708                         continue;
 709
 710                 lmv_disconnect_mdc(obd, lmv->tgts[i]);
 711         }
 712
 713 #ifdef __KERNEL__
 714         lmv_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds");
 715         if (lmv_proc_dir) {
 716                 lprocfs_remove(&lmv_proc_dir);
 717         } else {
 718                 CERROR("/proc/fs/lustre/%s/%s/target_obds missing\n",
 719                        obd->obd_type->typ_name, obd->obd_name);
 720         }
 721 #endif
 722
 723 out_local:
 724         /*
 725          * This is the case when no real connection is established by
 726          * lmv_check_connect().
 727          */
 728         if (!lmv->connected)
 729                 class_export_put(exp);
 730         rc = class_disconnect(exp);
 731         if (lmv->refcount == 0)
 732                 lmv->connected = 0;
 733         RETURN(rc);
 734 }
 735
 736 static int lmv_iocontrol(unsigned int cmd, struct obd_export *exp,
 737                          int len, void *karg, void *uarg)
 738 {
 739         struct obd_device    *obddev = class_exp2obd(exp);
 740         struct lmv_obd       *lmv = &obddev->u.lmv;
 741         int                   i = 0;
 742         int                   rc = 0;
 743         int                   set = 0;
 744         int                   count = lmv->desc.ld_tgt_count;
 745         ENTRY;
 746
 747         if (count == 0)
 748                 RETURN(-ENOTTY);
 749
 750         switch (cmd) {
 751         case IOC_OBD_STATFS: {
 752                 struct obd_ioctl_data *data = karg;
 753                 struct obd_device *mdc_obd;
 754                 struct obd_statfs stat_buf = {0};
 755                 __u32 index;
 756
 757                 memcpy(&index, data->ioc_inlbuf2, sizeof(__u32));
 758                 if ((index >= count))
 759                         RETURN(-ENODEV);
 760
 761                 if (lmv->tgts[index] == NULL ||
 762                     lmv->tgts[index]->ltd_active == 0)
 763                         RETURN(-ENODATA);
 764
 765                 mdc_obd = class_exp2obd(lmv->tgts[index]->ltd_exp);
 766                 if (!mdc_obd)
 767                         RETURN(-EINVAL);
 768
 769                 /* copy UUID */
 770                 if (cfs_copy_to_user(data->ioc_pbuf2, obd2cli_tgt(mdc_obd),
 771                                      min((int) data->ioc_plen2,
 772                                          (int) sizeof(struct obd_uuid))))
 773                         RETURN(-EFAULT);
 774
 775                 rc = obd_statfs(NULL, lmv->tgts[index]->ltd_exp, &stat_buf,
 776                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
 777                                 0);
 778                 if (rc)
 779                         RETURN(rc);
 780                 if (cfs_copy_to_user(data->ioc_pbuf1, &stat_buf,
 781                                      min((int) data->ioc_plen1,
 782                                          (int) sizeof(stat_buf))))
 783                         RETURN(-EFAULT);
 784                 break;
 785         }
 786         case OBD_IOC_QUOTACTL: {
 787                 struct if_quotactl *qctl = karg;
 788                 struct lmv_tgt_desc *tgt = NULL;
 789                 struct obd_quotactl *oqctl;
 790
 791                 if (qctl->qc_valid == QC_MDTIDX) {
 792                         if (qctl->qc_idx < 0 || count <= qctl->qc_idx)
 793                                 RETURN(-EINVAL);
 794
 795                         tgt = lmv->tgts[qctl->qc_idx];
 796                         if (tgt == NULL || tgt->ltd_exp == NULL)
 797                                 RETURN(-EINVAL);
 798                 } else if (qctl->qc_valid == QC_UUID) {
 799                         for (i = 0; i < count; i++) {
 800                                 tgt = lmv->tgts[i];
 801                                 if (tgt == NULL)
 802                                         continue;
 803                                 if (!obd_uuid_equals(&tgt->ltd_uuid,
 804                                                      &qctl->obd_uuid))
 805                                         continue;
 806
 807                                 if (tgt->ltd_exp == NULL)
 808                                         RETURN(-EINVAL);
 809
 810                                 break;
 811                         }
 812                 } else {
 813                         RETURN(-EINVAL);
 814                 }
 815
 816                 if (i >= count)
 817                         RETURN(-EAGAIN);
 818
 819                 LASSERT(tgt && tgt->ltd_exp);
 820                 OBD_ALLOC_PTR(oqctl);
 821                 if (!oqctl)
 822                         RETURN(-ENOMEM);
 823
 824                 QCTL_COPY(oqctl, qctl);
 825                 rc = obd_quotactl(tgt->ltd_exp, oqctl);
 826                 if (rc == 0) {
 827                         QCTL_COPY(qctl, oqctl);
 828                         qctl->qc_valid = QC_MDTIDX;
 829                         qctl->obd_uuid = tgt->ltd_uuid;
 830                 }
 831                 OBD_FREE_PTR(oqctl);
 832                 break;
 833         }
 834         case OBD_IOC_CHANGELOG_SEND:
 835         case OBD_IOC_CHANGELOG_CLEAR: {
 836                 struct ioc_changelog *icc = karg;
 837
 838                 if (icc->icc_mdtindex >= count)
 839                         RETURN(-ENODEV);
 840
 841                 if (lmv->tgts[icc->icc_mdtindex] == NULL ||
 842                     lmv->tgts[icc->icc_mdtindex]->ltd_exp == NULL ||
 843                     lmv->tgts[icc->icc_mdtindex]->ltd_active == 0)
 844                         RETURN(-ENODEV);
 845                 rc = obd_iocontrol(cmd, lmv->tgts[icc->icc_mdtindex]->ltd_exp,
 846                                    sizeof(*icc), icc, NULL);
 847                 break;
 848         }
 849         case LL_IOC_GET_CONNECT_FLAGS: {
 850                 if (lmv->tgts[0] == NULL)
 851                         RETURN(-ENODATA);
 852                 rc = obd_iocontrol(cmd, lmv->tgts[0]->ltd_exp, len, karg, uarg);
 853                 break;
 854         }
 855         case OBD_IOC_FID2PATH: {
 856                 struct getinfo_fid2path *gf;
 857                 struct lmv_tgt_desc     *tgt;
 858
 859                 gf = (struct getinfo_fid2path *)karg;
 860                 tgt = lmv_find_target(lmv, &gf->gf_fid);
 861                 if (IS_ERR(tgt))
 862                         RETURN(PTR_ERR(tgt));
 863                 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
 864                 break;
 865         }
 866         case LL_IOC_HSM_STATE_GET:
 867         case LL_IOC_HSM_STATE_SET:
 868         case LL_IOC_HSM_ACTION:
 869         case LL_IOC_LOV_SWAP_LAYOUTS: {
 870                 struct md_op_data       *op_data = karg;
 871                 struct lmv_tgt_desc     *tgt;
 872
 873                 tgt = lmv_find_target(lmv, &op_data->op_fid1);
 874                 if (!tgt->ltd_exp)
 875                         RETURN(-EINVAL);
 876
 877                 rc = obd_iocontrol(cmd, lmv->tgts[0]->ltd_exp, len, karg, uarg);
 878                 break;
 879         }
 880         default:
 881                 for (i = 0; i < count; i++) {
 882                         struct obd_device *mdc_obd;
 883                         int err;
 884
 885                         if (lmv->tgts[i] == NULL ||
 886                             lmv->tgts[i]->ltd_exp == NULL)
 887                                 continue;
 888                         /* ll_umount_begin() sets force flag but for lmv, not
 889                          * mdc. Let's pass it through */
 890                         mdc_obd = class_exp2obd(lmv->tgts[i]->ltd_exp);
 891                         mdc_obd->obd_force = obddev->obd_force;
 892                         err = obd_iocontrol(cmd, lmv->tgts[i]->ltd_exp, len,
 893                                             karg, uarg);
 894                         if (err == -ENODATA && cmd == OBD_IOC_POLL_QUOTACHECK) {
 895                                 RETURN(err);
 896                         } else if (err) {
 897                                 if (lmv->tgts[i]->ltd_active) {
 898                                         CERROR("error: iocontrol MDC %s on MDT"
 899                                                "idx %d cmd %x: err = %d\n",
 900                                                 lmv->tgts[i]->ltd_uuid.uuid,
 901                                                 i, cmd, err);
 902                                         if (!rc)
 903                                                 rc = err;
 904                                 }
 905                         } else
 906                                 set = 1;
 907                 }
 908                 if (!set && !rc)
 909                         rc = -EIO;
 910         }
 911         RETURN(rc);
 912 }
 913
 914 #if 0
 915 static int lmv_all_chars_policy(int count, const char *name,
 916                                 int len)
 917 {
 918         unsigned int c = 0;
 919
 920         while (len > 0)
 921                 c += name[--len];
 922         c = c % count;
 923         return c;
 924 }
 925
 926 static int lmv_nid_policy(struct lmv_obd *lmv)
 927 {
 928         struct obd_import *imp;
 929         __u32              id;
 930
 931         /*
 932          * XXX: To get nid we assume that underlying obd device is mdc.
 933          */
 934         imp = class_exp2cliimp(lmv->tgts[0].ltd_exp);
 935         id = imp->imp_connection->c_self ^ (imp->imp_connection->c_self >> 32);
 936         return id % lmv->desc.ld_tgt_count;
 937 }
 938
 939 static int lmv_choose_mds(struct lmv_obd *lmv, struct md_op_data *op_data,
 940                           placement_policy_t placement)
 941 {
 942         switch (placement) {
 943         case PLACEMENT_CHAR_POLICY:
 944                 return lmv_all_chars_policy(lmv->desc.ld_tgt_count,
 945                                             op_data->op_name,
 946                                             op_data->op_namelen);
 947         case PLACEMENT_NID_POLICY:
 948                 return lmv_nid_policy(lmv);
 949
 950         default:
 951                 break;
 952         }
 953
 954         CERROR("Unsupported placement policy %x\n", placement);
 955         return -EINVAL;
 956 }
 957 #endif
 958
 959 /**
 960  * This is _inode_ placement policy function (not name).
 961  */
 962 static int lmv_placement_policy(struct obd_device *obd,
 963                                 struct md_op_data *op_data,
 964                                 mdsno_t *mds)
 965 {
 966         struct lmv_obd          *lmv = &obd->u.lmv;
 967         ENTRY;
 968
 969         LASSERT(mds != NULL);
 970
 971         if (lmv->desc.ld_tgt_count == 1) {
 972                 *mds = 0;
 973                 RETURN(0);
 974         }
 975
 976         /**
 977          * If stripe_offset is provided during setdirstripe
 978          * (setdirstripe -i xx), xx MDS will be choosen.
 979          */
 980         if (op_data->op_cli_flags & CLI_SET_MEA) {
 981                 struct lmv_user_md *lum;
 982
 983                 lum = (struct lmv_user_md *)op_data->op_data;
 984                 if (lum->lum_type == LMV_STRIPE_TYPE &&
 985                     lum->lum_stripe_offset != -1) {
 986                         if (lum->lum_stripe_offset >= lmv->desc.ld_tgt_count) {
 987                                 CERROR("%s: Stripe_offset %d > MDT count %d:"
 988                                        " rc = %d\n", obd->obd_name,
 989                                        lum->lum_stripe_offset,
 990                                        lmv->desc.ld_tgt_count, -ERANGE);
 991                                 RETURN(-ERANGE);
 992                         }
 993                         *mds = lum->lum_stripe_offset;
 994                         RETURN(0);
 995                 }
 996         }
 997
 998         /* Allocate new fid on target according to operation type and parent
 999          * home mds. */
1000         *mds = op_data->op_mds;
1001         RETURN(0);
1002 }
1003
1004 int __lmv_fid_alloc(struct lmv_obd *lmv, struct lu_fid *fid,
1005                     mdsno_t mds)
1006 {
1007         struct lmv_tgt_desc *tgt;
1008         int                  rc;
1009         ENTRY;
1010
1011         tgt = lmv_get_target(lmv, mds);
1012
1013         /*
1014          * New seq alloc and FLD setup should be atomic. Otherwise we may find
1015          * on server that seq in new allocated fid is not yet known.
1016          */
1017         mutex_lock(&tgt->ltd_fid_mutex);
1018
1019         if (tgt == NULL || tgt->ltd_active == 0 || tgt->ltd_exp == NULL)
1020                 GOTO(out, rc = -ENODEV);
1021
1022         /*
1023          * Asking underlaying tgt layer to allocate new fid.
1024          */
1025         rc = obd_fid_alloc(tgt->ltd_exp, fid, NULL);
1026         if (rc > 0) {
1027                 LASSERT(fid_is_sane(fid));
1028                 rc = 0;
1029         }
1030
1031         EXIT;
1032 out:
1033         mutex_unlock(&tgt->ltd_fid_mutex);
1034         return rc;
1035 }
1036
1037 int lmv_fid_alloc(struct obd_export *exp, struct lu_fid *fid,
1038                   struct md_op_data *op_data)
1039 {
1040         struct obd_device     *obd = class_exp2obd(exp);
1041         struct lmv_obd        *lmv = &obd->u.lmv;
1042         mdsno_t                mds = 0;
1043         int                    rc;
1044         ENTRY;
1045
1046         LASSERT(op_data != NULL);
1047         LASSERT(fid != NULL);
1048
1049         rc = lmv_placement_policy(obd, op_data, &mds);
1050         if (rc) {
1051                 CERROR("Can't get target for allocating fid, "
1052                        "rc %d\n", rc);
1053                 RETURN(rc);
1054         }
1055
1056         rc = __lmv_fid_alloc(lmv, fid, mds);
1057         if (rc) {
1058                 CERROR("Can't alloc new fid, rc %d\n", rc);
1059                 RETURN(rc);
1060         }
1061
1062         RETURN(rc);
1063 }
1064
1065 static int lmv_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
1066 {
1067         struct lmv_obd             *lmv = &obd->u.lmv;
1068         struct lprocfs_static_vars  lvars;
1069         struct lmv_desc            *desc;
1070         int                         rc;
1071         ENTRY;
1072
1073         if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
1074                 CERROR("LMV setup requires a descriptor\n");
1075                 RETURN(-EINVAL);
1076         }
1077
1078         desc = (struct lmv_desc *)lustre_cfg_buf(lcfg, 1);
1079         if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
1080                 CERROR("Lmv descriptor size wrong: %d > %d\n",
1081                        (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
1082                 RETURN(-EINVAL);
1083         }
1084
1085         OBD_ALLOC(lmv->tgts, sizeof(*lmv->tgts) * 32);
1086         if (lmv->tgts == NULL)
1087                 RETURN(-ENOMEM);
1088         lmv->tgts_size = 32;
1089
1090         obd_str2uuid(&lmv->desc.ld_uuid, desc->ld_uuid.uuid);
1091         lmv->desc.ld_tgt_count = 0;
1092         lmv->desc.ld_active_tgt_count = 0;
1093         lmv->max_cookiesize = 0;
1094         lmv->max_def_easize = 0;
1095         lmv->max_easize = 0;
1096         lmv->lmv_placement = PLACEMENT_CHAR_POLICY;
1097
1098         spin_lock_init(&lmv->lmv_lock);
1099         mutex_init(&lmv->init_mutex);
1100
1101         lprocfs_lmv_init_vars(&lvars);
1102
1103         lprocfs_obd_setup(obd, lvars.obd_vars);
1104 #ifdef LPROCFS
1105         {
1106                 rc = lprocfs_seq_create(obd->obd_proc_entry, "target_obd",
1107                                         0444, &lmv_proc_target_fops, obd);
1108                 if (rc)
1109                         CWARN("%s: error adding LMV target_obd file: rc = %d\n",
1110                                obd->obd_name, rc);
1111        }
1112 #endif
1113         rc = fld_client_init(&lmv->lmv_fld, obd->obd_name,
1114                              LUSTRE_CLI_FLD_HASH_DHT);
1115         if (rc) {
1116                 CERROR("Can't init FLD, err %d\n", rc);
1117                 GOTO(out, rc);
1118         }
1119
1120         RETURN(0);
1121
1122 out:
1123         return rc;
1124 }
1125
1126 static int lmv_cleanup(struct obd_device *obd)
1127 {
1128         struct lmv_obd   *lmv = &obd->u.lmv;
1129         ENTRY;
1130
1131         fld_client_fini(&lmv->lmv_fld);
1132         if (lmv->tgts != NULL) {
1133                 int i;
1134                 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1135                         if (lmv->tgts[i] == NULL)
1136                                 continue;
1137                         lmv_del_target(lmv, i);
1138                 }
1139                 OBD_FREE(lmv->tgts, sizeof(*lmv->tgts) * lmv->tgts_size);
1140                 lmv->tgts_size = 0;
1141         }
1142         RETURN(0);
1143 }
1144
1145 static int lmv_process_config(struct obd_device *obd, obd_count len, void *buf)
1146 {
1147         struct lustre_cfg       *lcfg = buf;
1148         struct obd_uuid         obd_uuid;
1149         int                     gen;
1150         __u32                   index;
1151         int                     rc;
1152         ENTRY;
1153
1154         switch (lcfg->lcfg_command) {
1155         case LCFG_ADD_MDC:
1156                 /* modify_mdc_tgts add 0:lustre-clilmv  1:lustre-MDT0000_UUID
1157                  * 2:0  3:1  4:lustre-MDT0000-mdc_UUID */
1158                 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(obd_uuid.uuid))
1159                         GOTO(out, rc = -EINVAL);
1160
1161                 obd_str2uuid(&obd_uuid,  lustre_cfg_buf(lcfg, 1));
1162
1163                 if (sscanf(lustre_cfg_buf(lcfg, 2), "%d", &index) != 1)
1164                         GOTO(out, rc = -EINVAL);
1165                 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", &gen) != 1)
1166                         GOTO(out, rc = -EINVAL);
1167                 rc = lmv_add_target(obd, &obd_uuid, index, gen);
1168                 GOTO(out, rc);
1169         default:
1170                 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
1171                 GOTO(out, rc = -EINVAL);
1172         }
1173 out:
1174         RETURN(rc);
1175 }
1176
1177 static int lmv_statfs(const struct lu_env *env, struct obd_export *exp,
1178                       struct obd_statfs *osfs, __u64 max_age, __u32 flags)
1179 {
1180         struct obd_device     *obd = class_exp2obd(exp);
1181         struct lmv_obd        *lmv = &obd->u.lmv;
1182         struct obd_statfs     *temp;
1183         int                    rc = 0;
1184         int                    i;
1185         ENTRY;
1186
1187         rc = lmv_check_connect(obd);
1188         if (rc)
1189                 RETURN(rc);
1190
1191         OBD_ALLOC(temp, sizeof(*temp));
1192         if (temp == NULL)
1193                 RETURN(-ENOMEM);
1194
1195         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1196                 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
1197                         continue;
1198
1199                 rc = obd_statfs(env, lmv->tgts[i]->ltd_exp, temp,
1200                                 max_age, flags);
1201                 if (rc) {
1202                         CERROR("can't stat MDS #%d (%s), error %d\n", i,
1203                                lmv->tgts[i]->ltd_exp->exp_obd->obd_name,
1204                                rc);
1205                         GOTO(out_free_temp, rc);
1206                 }
1207
1208                 if (i == 0) {
1209                         *osfs = *temp;
1210                         /* If the statfs is from mount, it will needs
1211                          * retrieve necessary information from MDT0.
1212                          * i.e. mount does not need the merged osfs
1213                          * from all of MDT.
1214                          * And also clients can be mounted as long as
1215                          * MDT0 is in service*/
1216                         if (flags & OBD_STATFS_FOR_MDT0)
1217                                 GOTO(out_free_temp, rc);
1218                 } else {
1219                         osfs->os_bavail += temp->os_bavail;
1220                         osfs->os_blocks += temp->os_blocks;
1221                         osfs->os_ffree += temp->os_ffree;
1222                         osfs->os_files += temp->os_files;
1223                 }
1224         }
1225
1226         EXIT;
1227 out_free_temp:
1228         OBD_FREE(temp, sizeof(*temp));
1229         return rc;
1230 }
1231
1232 static int lmv_getstatus(struct obd_export *exp,
1233                          struct lu_fid *fid,
1234                          struct obd_capa **pc)
1235 {
1236         struct obd_device    *obd = exp->exp_obd;
1237         struct lmv_obd       *lmv = &obd->u.lmv;
1238         int                   rc;
1239         ENTRY;
1240
1241         rc = lmv_check_connect(obd);
1242         if (rc)
1243                 RETURN(rc);
1244
1245         rc = md_getstatus(lmv->tgts[0]->ltd_exp, fid, pc);
1246         RETURN(rc);
1247 }
1248
1249 static int lmv_getxattr(struct obd_export *exp, const struct lu_fid *fid,
1250                         struct obd_capa *oc, obd_valid valid, const char *name,
1251                         const char *input, int input_size, int output_size,
1252                         int flags, struct ptlrpc_request **request)
1253 {
1254         struct obd_device      *obd = exp->exp_obd;
1255         struct lmv_obd         *lmv = &obd->u.lmv;
1256         struct lmv_tgt_desc    *tgt;
1257         int                     rc;
1258         ENTRY;
1259
1260         rc = lmv_check_connect(obd);
1261         if (rc)
1262                 RETURN(rc);
1263
1264         tgt = lmv_find_target(lmv, fid);
1265         if (IS_ERR(tgt))
1266                 RETURN(PTR_ERR(tgt));
1267
1268         rc = md_getxattr(tgt->ltd_exp, fid, oc, valid, name, input,
1269                          input_size, output_size, flags, request);
1270
1271         RETURN(rc);
1272 }
1273
1274 static int lmv_setxattr(struct obd_export *exp, const struct lu_fid *fid,
1275                         struct obd_capa *oc, obd_valid valid, const char *name,
1276                         const char *input, int input_size, int output_size,
1277                         int flags, __u32 suppgid,
1278                         struct ptlrpc_request **request)
1279 {
1280         struct obd_device      *obd = exp->exp_obd;
1281         struct lmv_obd         *lmv = &obd->u.lmv;
1282         struct lmv_tgt_desc    *tgt;
1283         int                     rc;
1284         ENTRY;
1285
1286         rc = lmv_check_connect(obd);
1287         if (rc)
1288                 RETURN(rc);
1289
1290         tgt = lmv_find_target(lmv, fid);
1291         if (IS_ERR(tgt))
1292                 RETURN(PTR_ERR(tgt));
1293
1294         rc = md_setxattr(tgt->ltd_exp, fid, oc, valid, name, input,
1295                          input_size, output_size, flags, suppgid,
1296                          request);
1297
1298         RETURN(rc);
1299 }
1300
1301 static int lmv_getattr(struct obd_export *exp, struct md_op_data *op_data,
1302                        struct ptlrpc_request **request)
1303 {
1304         struct obd_device       *obd = exp->exp_obd;
1305         struct lmv_obd          *lmv = &obd->u.lmv;
1306         struct lmv_tgt_desc     *tgt;
1307         int                      rc;
1308         ENTRY;
1309
1310         rc = lmv_check_connect(obd);
1311         if (rc)
1312                 RETURN(rc);
1313
1314         tgt = lmv_find_target(lmv, &op_data->op_fid1);
1315         if (IS_ERR(tgt))
1316                 RETURN(PTR_ERR(tgt));
1317
1318         if (op_data->op_flags & MF_GET_MDT_IDX) {
1319                 op_data->op_mds = tgt->ltd_idx;
1320                 RETURN(0);
1321         }
1322
1323         rc = md_getattr(tgt->ltd_exp, op_data, request);
1324
1325         RETURN(rc);
1326 }
1327
1328 static int lmv_null_inode(struct obd_export *exp, const struct lu_fid *fid)
1329 {
1330         struct obd_device   *obd = exp->exp_obd;
1331         struct lmv_obd      *lmv = &obd->u.lmv;
1332         int                  i;
1333         int                  rc;
1334         ENTRY;
1335
1336         rc = lmv_check_connect(obd);
1337         if (rc)
1338                 RETURN(rc);
1339
1340         CDEBUG(D_INODE, "CBDATA for "DFID"\n", PFID(fid));
1341
1342         /*
1343          * With DNE every object can have two locks in different namespaces:
1344          * lookup lock in space of MDT storing direntry and update/open lock in
1345          * space of MDT storing inode.
1346          */
1347         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1348                 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
1349                         continue;
1350                 md_null_inode(lmv->tgts[i]->ltd_exp, fid);
1351         }
1352
1353         RETURN(0);
1354 }
1355
1356 static int lmv_find_cbdata(struct obd_export *exp, const struct lu_fid *fid,
1357                            ldlm_iterator_t it, void *data)
1358 {
1359         struct obd_device   *obd = exp->exp_obd;
1360         struct lmv_obd      *lmv = &obd->u.lmv;
1361         int                  i;
1362         int                  rc;
1363         ENTRY;
1364
1365         rc = lmv_check_connect(obd);
1366         if (rc)
1367                 RETURN(rc);
1368
1369         CDEBUG(D_INODE, "CBDATA for "DFID"\n", PFID(fid));
1370
1371         /*
1372          * With DNE every object can have two locks in different namespaces:
1373          * lookup lock in space of MDT storing direntry and update/open lock in
1374          * space of MDT storing inode.
1375          */
1376         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1377                 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
1378                         continue;
1379                 rc = md_find_cbdata(lmv->tgts[i]->ltd_exp, fid, it, data);
1380                 if (rc)
1381                         RETURN(rc);
1382         }
1383
1384         RETURN(rc);
1385 }
1386
1387
1388 static int lmv_close(struct obd_export *exp, struct md_op_data *op_data,
1389                      struct md_open_data *mod, struct ptlrpc_request **request)
1390 {
1391         struct obd_device     *obd = exp->exp_obd;
1392         struct lmv_obd        *lmv = &obd->u.lmv;
1393         struct lmv_tgt_desc   *tgt;
1394         int                    rc;
1395         ENTRY;
1396
1397         rc = lmv_check_connect(obd);
1398         if (rc)
1399                 RETURN(rc);
1400
1401         tgt = lmv_find_target(lmv, &op_data->op_fid1);
1402         if (IS_ERR(tgt))
1403                 RETURN(PTR_ERR(tgt));
1404
1405         CDEBUG(D_INODE, "CLOSE "DFID"\n", PFID(&op_data->op_fid1));
1406         rc = md_close(tgt->ltd_exp, op_data, mod, request);
1407         RETURN(rc);
1408 }
1409
1410 struct lmv_tgt_desc
1411 *lmv_locate_mds(struct lmv_obd *lmv, struct md_op_data *op_data,
1412                 struct lu_fid *fid)
1413 {
1414         struct lmv_tgt_desc *tgt;
1415
1416         tgt = lmv_find_target(lmv, fid);
1417         op_data->op_mds = tgt->ltd_idx;
1418
1419         return tgt;
1420 }
1421
1422 int lmv_create(struct obd_export *exp, struct md_op_data *op_data,
1423                const void *data, int datalen, int mode, __u32 uid,
1424                __u32 gid, cfs_cap_t cap_effective, __u64 rdev,
1425                struct ptlrpc_request **request)
1426 {
1427         struct obd_device       *obd = exp->exp_obd;
1428         struct lmv_obd          *lmv = &obd->u.lmv;
1429         struct lmv_tgt_desc     *tgt;
1430         int                      rc;
1431         ENTRY;
1432
1433         rc = lmv_check_connect(obd);
1434         if (rc)
1435                 RETURN(rc);
1436
1437         if (!lmv->desc.ld_active_tgt_count)
1438                 RETURN(-EIO);
1439
1440         tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
1441         if (IS_ERR(tgt))
1442                 RETURN(PTR_ERR(tgt));
1443
1444         rc = lmv_fid_alloc(exp, &op_data->op_fid2, op_data);
1445         if (rc)
1446                 RETURN(rc);
1447
1448         CDEBUG(D_INODE, "CREATE '%*s' on "DFID" -> mds #%x\n",
1449                op_data->op_namelen, op_data->op_name, PFID(&op_data->op_fid1),
1450                op_data->op_mds);
1451
1452         op_data->op_flags |= MF_MDC_CANCEL_FID1;
1453         rc = md_create(tgt->ltd_exp, op_data, data, datalen, mode, uid, gid,
1454                        cap_effective, rdev, request);
1455
1456         if (rc == 0) {
1457                 if (*request == NULL)
1458                         RETURN(rc);
1459                 CDEBUG(D_INODE, "Created - "DFID"\n", PFID(&op_data->op_fid2));
1460         }
1461         RETURN(rc);
1462 }
1463
1464 static int lmv_done_writing(struct obd_export *exp,
1465                             struct md_op_data *op_data,
1466                             struct md_open_data *mod)
1467 {
1468         struct obd_device     *obd = exp->exp_obd;
1469         struct lmv_obd        *lmv = &obd->u.lmv;
1470         struct lmv_tgt_desc   *tgt;
1471         int                    rc;
1472         ENTRY;
1473
1474         rc = lmv_check_connect(obd);
1475         if (rc)
1476                 RETURN(rc);
1477
1478         tgt = lmv_find_target(lmv, &op_data->op_fid1);
1479         if (IS_ERR(tgt))
1480                 RETURN(PTR_ERR(tgt));
1481
1482         rc = md_done_writing(tgt->ltd_exp, op_data, mod);
1483         RETURN(rc);
1484 }
1485
1486 static int
1487 lmv_enqueue_remote(struct obd_export *exp, struct ldlm_enqueue_info *einfo,
1488                    struct lookup_intent *it, struct md_op_data *op_data,
1489                    struct lustre_handle *lockh, void *lmm, int lmmsize,
1490                    int extra_lock_flags)
1491 {
1492         struct ptlrpc_request      *req = it->d.lustre.it_data;
1493         struct obd_device          *obd = exp->exp_obd;
1494         struct lmv_obd             *lmv = &obd->u.lmv;
1495         struct lustre_handle        plock;
1496         struct lmv_tgt_desc        *tgt;
1497         struct md_op_data          *rdata;
1498         struct lu_fid               fid1;
1499         struct mdt_body            *body;
1500         int                         rc = 0;
1501         int                         pmode;
1502         ENTRY;
1503
1504         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1505         LASSERT(body != NULL);
1506
1507         if (!(body->valid & OBD_MD_MDS))
1508                 RETURN(0);
1509
1510         CDEBUG(D_INODE, "REMOTE_ENQUEUE '%s' on "DFID" -> "DFID"\n",
1511                LL_IT2STR(it), PFID(&op_data->op_fid1), PFID(&body->fid1));
1512
1513         /*
1514          * We got LOOKUP lock, but we really need attrs.
1515          */
1516         pmode = it->d.lustre.it_lock_mode;
1517         LASSERT(pmode != 0);
1518         memcpy(&plock, lockh, sizeof(plock));
1519         it->d.lustre.it_lock_mode = 0;
1520         it->d.lustre.it_data = NULL;
1521         fid1 = body->fid1;
1522
1523         it->d.lustre.it_disposition &= ~DISP_ENQ_COMPLETE;
1524         ptlrpc_req_finished(req);
1525
1526         tgt = lmv_find_target(lmv, &fid1);
1527         if (IS_ERR(tgt))
1528                 GOTO(out, rc = PTR_ERR(tgt));
1529
1530         OBD_ALLOC_PTR(rdata);
1531         if (rdata == NULL)
1532                 GOTO(out, rc = -ENOMEM);
1533
1534         rdata->op_fid1 = fid1;
1535         rdata->op_bias = MDS_CROSS_REF;
1536
1537         rc = md_enqueue(tgt->ltd_exp, einfo, it, rdata, lockh,
1538                         lmm, lmmsize, NULL, extra_lock_flags);
1539         OBD_FREE_PTR(rdata);
1540         EXIT;
1541 out:
1542         ldlm_lock_decref(&plock, pmode);
1543         return rc;
1544 }
1545
1546 static int
1547 lmv_enqueue(struct obd_export *exp, struct ldlm_enqueue_info *einfo,
1548             struct lookup_intent *it, struct md_op_data *op_data,
1549             struct lustre_handle *lockh, void *lmm, int lmmsize,
1550             struct ptlrpc_request **req, __u64 extra_lock_flags)
1551 {
1552         struct obd_device        *obd = exp->exp_obd;
1553         struct lmv_obd           *lmv = &obd->u.lmv;
1554         struct lmv_tgt_desc      *tgt;
1555         int                       rc;
1556         ENTRY;
1557
1558         rc = lmv_check_connect(obd);
1559         if (rc)
1560                 RETURN(rc);
1561
1562         CDEBUG(D_INODE, "ENQUEUE '%s' on "DFID"\n",
1563                LL_IT2STR(it), PFID(&op_data->op_fid1));
1564
1565         tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
1566         if (IS_ERR(tgt))
1567                 RETURN(PTR_ERR(tgt));
1568
1569         CDEBUG(D_INODE, "ENQUEUE '%s' on "DFID" -> mds #%d\n",
1570                LL_IT2STR(it), PFID(&op_data->op_fid1), tgt->ltd_idx);
1571
1572         rc = md_enqueue(tgt->ltd_exp, einfo, it, op_data, lockh,
1573                         lmm, lmmsize, req, extra_lock_flags);
1574
1575         if (rc == 0 && it && it->it_op == IT_OPEN) {
1576                 rc = lmv_enqueue_remote(exp, einfo, it, op_data, lockh,
1577                                         lmm, lmmsize, extra_lock_flags);
1578         }
1579         RETURN(rc);
1580 }
1581
1582 static int
1583 lmv_getattr_name(struct obd_export *exp,struct md_op_data *op_data,
1584                  struct ptlrpc_request **request)
1585 {
1586         struct ptlrpc_request   *req = NULL;
1587         struct obd_device       *obd = exp->exp_obd;
1588         struct lmv_obd          *lmv = &obd->u.lmv;
1589         struct lmv_tgt_desc     *tgt;
1590         struct mdt_body         *body;
1591         int                      rc;
1592         ENTRY;
1593
1594         rc = lmv_check_connect(obd);
1595         if (rc)
1596                 RETURN(rc);
1597
1598         tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
1599         if (IS_ERR(tgt))
1600                 RETURN(PTR_ERR(tgt));
1601
1602         CDEBUG(D_INODE, "GETATTR_NAME for %*s on "DFID" -> mds #%d\n",
1603                op_data->op_namelen, op_data->op_name, PFID(&op_data->op_fid1),
1604                tgt->ltd_idx);
1605
1606         rc = md_getattr_name(tgt->ltd_exp, op_data, request);
1607         if (rc != 0)
1608                 RETURN(rc);
1609
1610         body = req_capsule_server_get(&(*request)->rq_pill,
1611                                       &RMF_MDT_BODY);
1612         LASSERT(body != NULL);
1613
1614         if (body->valid & OBD_MD_MDS) {
1615                 struct lu_fid rid = body->fid1;
1616                 CDEBUG(D_INODE, "Request attrs for "DFID"\n",
1617                        PFID(&rid));
1618
1619                 tgt = lmv_find_target(lmv, &rid);
1620                 if (IS_ERR(tgt)) {
1621                         ptlrpc_req_finished(*request);
1622                         RETURN(PTR_ERR(tgt));
1623                 }
1624
1625                 op_data->op_fid1 = rid;
1626                 op_data->op_valid |= OBD_MD_FLCROSSREF;
1627                 op_data->op_namelen = 0;
1628                 op_data->op_name = NULL;
1629                 rc = md_getattr_name(tgt->ltd_exp, op_data, &req);
1630                 ptlrpc_req_finished(*request);
1631                 *request = req;
1632         }
1633
1634         RETURN(rc);
1635 }
1636
1637 #define md_op_data_fid(op_data, fl)                     \
1638         (fl == MF_MDC_CANCEL_FID1 ? &op_data->op_fid1 : \
1639          fl == MF_MDC_CANCEL_FID2 ? &op_data->op_fid2 : \
1640          fl == MF_MDC_CANCEL_FID3 ? &op_data->op_fid3 : \
1641          fl == MF_MDC_CANCEL_FID4 ? &op_data->op_fid4 : \
1642          NULL)
1643
1644 static int lmv_early_cancel(struct obd_export *exp, struct md_op_data *op_data,
1645                             int op_tgt, ldlm_mode_t mode, int bits, int flag)
1646 {
1647         struct lu_fid          *fid = md_op_data_fid(op_data, flag);
1648         struct obd_device      *obd = exp->exp_obd;
1649         struct lmv_obd         *lmv = &obd->u.lmv;
1650         struct lmv_tgt_desc    *tgt;
1651         ldlm_policy_data_t      policy = {{0}};
1652         int                     rc = 0;
1653         ENTRY;
1654
1655         if (!fid_is_sane(fid))
1656                 RETURN(0);
1657
1658         tgt = lmv_find_target(lmv, fid);
1659         if (IS_ERR(tgt))
1660                 RETURN(PTR_ERR(tgt));
1661
1662         if (tgt->ltd_idx != op_tgt) {
1663                 CDEBUG(D_INODE, "EARLY_CANCEL on "DFID"\n", PFID(fid));
1664                 policy.l_inodebits.bits = bits;
1665                 rc = md_cancel_unused(tgt->ltd_exp, fid, &policy,
1666                                       mode, LCF_ASYNC, NULL);
1667         } else {
1668                 CDEBUG(D_INODE,
1669                        "EARLY_CANCEL skip operation target %d on "DFID"\n",
1670                        op_tgt, PFID(fid));
1671                 op_data->op_flags |= flag;
1672                 rc = 0;
1673         }
1674
1675         RETURN(rc);
1676 }
1677
1678 /*
1679  * llite passes fid of an target inode in op_data->op_fid1 and id of directory in
1680  * op_data->op_fid2
1681  */
1682 static int lmv_link(struct obd_export *exp, struct md_op_data *op_data,
1683                     struct ptlrpc_request **request)
1684 {
1685         struct obd_device       *obd = exp->exp_obd;
1686         struct lmv_obd          *lmv = &obd->u.lmv;
1687         struct lmv_tgt_desc     *tgt;
1688         int                      rc;
1689         ENTRY;
1690
1691         rc = lmv_check_connect(obd);
1692         if (rc)
1693                 RETURN(rc);
1694
1695         LASSERT(op_data->op_namelen != 0);
1696
1697         CDEBUG(D_INODE, "LINK "DFID":%*s to "DFID"\n",
1698                PFID(&op_data->op_fid2), op_data->op_namelen,
1699                op_data->op_name, PFID(&op_data->op_fid1));
1700
1701         op_data->op_fsuid = cfs_curproc_fsuid();
1702         op_data->op_fsgid = cfs_curproc_fsgid();
1703         op_data->op_cap = cfs_curproc_cap_pack();
1704         tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid2);
1705         if (IS_ERR(tgt))
1706                 RETURN(PTR_ERR(tgt));
1707
1708         /*
1709          * Cancel UPDATE lock on child (fid1).
1710          */
1711         op_data->op_flags |= MF_MDC_CANCEL_FID2;
1712         rc = lmv_early_cancel(exp, op_data, tgt->ltd_idx, LCK_EX,
1713                               MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID1);
1714         if (rc != 0)
1715                 RETURN(rc);
1716
1717         rc = md_link(tgt->ltd_exp, op_data, request);
1718
1719         RETURN(rc);
1720 }
1721
1722 static int lmv_rename(struct obd_export *exp, struct md_op_data *op_data,
1723                       const char *old, int oldlen, const char *new, int newlen,
1724                       struct ptlrpc_request **request)
1725 {
1726         struct obd_device       *obd = exp->exp_obd;
1727         struct lmv_obd          *lmv = &obd->u.lmv;
1728         struct lmv_tgt_desc     *src_tgt;
1729         struct lmv_tgt_desc     *tgt_tgt;
1730         int                     rc;
1731         ENTRY;
1732
1733         LASSERT(oldlen != 0);
1734
1735         CDEBUG(D_INODE, "RENAME %*s in "DFID" to %*s in "DFID"\n",
1736                oldlen, old, PFID(&op_data->op_fid1),
1737                newlen, new, PFID(&op_data->op_fid2));
1738
1739         rc = lmv_check_connect(obd);
1740         if (rc)
1741                 RETURN(rc);
1742
1743         op_data->op_fsuid = cfs_curproc_fsuid();
1744         op_data->op_fsgid = cfs_curproc_fsgid();
1745         op_data->op_cap = cfs_curproc_cap_pack();
1746         src_tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
1747         if (IS_ERR(src_tgt))
1748                 RETURN(PTR_ERR(src_tgt));
1749
1750         tgt_tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid2);
1751         if (IS_ERR(tgt_tgt))
1752                 RETURN(PTR_ERR(tgt_tgt));
1753         /*
1754          * LOOKUP lock on src child (fid3) should also be cancelled for
1755          * src_tgt in mdc_rename.
1756          */
1757         op_data->op_flags |= MF_MDC_CANCEL_FID1 | MF_MDC_CANCEL_FID3;
1758
1759         /*
1760          * Cancel UPDATE locks on tgt parent (fid2), tgt_tgt is its
1761          * own target.
1762          */
1763         rc = lmv_early_cancel(exp, op_data, src_tgt->ltd_idx,
1764                               LCK_EX, MDS_INODELOCK_UPDATE,
1765                               MF_MDC_CANCEL_FID2);
1766
1767         /*
1768          * Cancel LOOKUP locks on tgt child (fid4) for parent tgt_tgt.
1769          */
1770         if (rc == 0) {
1771                 rc = lmv_early_cancel(exp, op_data, src_tgt->ltd_idx,
1772                                       LCK_EX, MDS_INODELOCK_LOOKUP,
1773                                       MF_MDC_CANCEL_FID4);
1774         }
1775
1776         /*
1777          * Cancel all the locks on tgt child (fid4).
1778          */
1779         if (rc == 0)
1780                 rc = lmv_early_cancel(exp, op_data, src_tgt->ltd_idx,
1781                                       LCK_EX, MDS_INODELOCK_FULL,
1782                                       MF_MDC_CANCEL_FID4);
1783
1784         if (rc == 0)
1785                 rc = md_rename(src_tgt->ltd_exp, op_data, old, oldlen,
1786                                new, newlen, request);
1787         RETURN(rc);
1788 }
1789
1790 static int lmv_setattr(struct obd_export *exp, struct md_op_data *op_data,
1791                        void *ea, int ealen, void *ea2, int ea2len,
1792                        struct ptlrpc_request **request,
1793                        struct md_open_data **mod)
1794 {
1795         struct obd_device       *obd = exp->exp_obd;
1796         struct lmv_obd          *lmv = &obd->u.lmv;
1797         struct lmv_tgt_desc     *tgt;
1798         int                      rc = 0;
1799         ENTRY;
1800
1801         rc = lmv_check_connect(obd);
1802         if (rc)
1803                 RETURN(rc);
1804
1805         CDEBUG(D_INODE, "SETATTR for "DFID", valid 0x%x\n",
1806                PFID(&op_data->op_fid1), op_data->op_attr.ia_valid);
1807
1808         op_data->op_flags |= MF_MDC_CANCEL_FID1;
1809         tgt = lmv_find_target(lmv, &op_data->op_fid1);
1810         if (IS_ERR(tgt))
1811                 RETURN(PTR_ERR(tgt));
1812
1813         rc = md_setattr(tgt->ltd_exp, op_data, ea, ealen, ea2,
1814                         ea2len, request, mod);
1815
1816         RETURN(rc);
1817 }
1818
1819 static int lmv_sync(struct obd_export *exp, const struct lu_fid *fid,
1820                     struct obd_capa *oc, struct ptlrpc_request **request)
1821 {
1822         struct obd_device         *obd = exp->exp_obd;
1823         struct lmv_obd            *lmv = &obd->u.lmv;
1824         struct lmv_tgt_desc       *tgt;
1825         int                        rc;
1826         ENTRY;
1827
1828         rc = lmv_check_connect(obd);
1829         if (rc)
1830                 RETURN(rc);
1831
1832         tgt = lmv_find_target(lmv, fid);
1833         if (IS_ERR(tgt))
1834                 RETURN(PTR_ERR(tgt));
1835
1836         rc = md_sync(tgt->ltd_exp, fid, oc, request);
1837         RETURN(rc);
1838 }
1839
1840 static int lmv_readpage(struct obd_export *exp, struct md_op_data *op_data,
1841                         struct page **pages, struct ptlrpc_request **request)
1842 {
1843         struct obd_device       *obd = exp->exp_obd;
1844         struct lmv_obd          *lmv = &obd->u.lmv;
1845         __u64                    offset = op_data->op_offset;
1846         int                      rc;
1847         int                      i;
1848         /* number of pages read, in CFS_PAGE_SIZE */
1849         int                      nrdpgs;
1850         /* number of pages transferred in LU_PAGE_SIZE */
1851         int                      nlupgs;
1852         struct lmv_tgt_desc     *tgt;
1853         struct lu_dirpage       *dp;
1854         struct lu_dirent        *ent;
1855         ENTRY;
1856
1857         rc = lmv_check_connect(obd);
1858         if (rc)
1859                 RETURN(rc);
1860
1861         CDEBUG(D_INODE, "READPAGE at "LPX64" from "DFID"\n",
1862                offset, PFID(&op_data->op_fid1));
1863
1864         /*
1865          * This case handle directory lookup in clustered metadata case (i.e.
1866          * split directory is located on multiple md servers.)
1867          * each server keeps directory entries for certain range of hashes.
1868          * E.g. we have N server and suppose hash range is 0 to MAX_HASH.
1869          * first server will keep records with hashes [ 0 ... MAX_HASH /N  - 1],
1870          * second one with hashes [MAX_HASH / N ... 2 * MAX_HASH / N] and
1871          * so on....
1872          *      readdir can simply start reading entries from 0 - N server in
1873          * order but that will not scale well as all client will request dir in
1874          * to server in same order.
1875          * Following algorithm does optimization:
1876          * Instead of doing readdir in 1, 2, ...., N order, client with a
1877          * rank R does readdir in R, R + 1, ..., N, 1, ... R - 1 order.
1878          * (every client has rank R)
1879          *      But ll_readdir() expect offset range [0 to MAX_HASH/N) but
1880          * since client ask dir from MDS{R} client has pages with offsets
1881          * [R*MAX_HASH/N ... (R + 1)*MAX_HASH/N] there for we do hash_adj
1882          * on hash  values that we get.
1883          * Since these codes might be still useful for sharded directory, so
1884          * Keeping this code for further reference
1885         if (0) {
1886                 LASSERT(nr > 0);
1887                 seg_size = MAX_HASH_SIZE;
1888                 do_div(seg_size, nr);
1889                 los      = obj->lo_stripes;
1890                 tgt      = lmv_get_target(lmv, los[0].ls_mds);
1891                 rank     = lmv_node_rank(tgt->ltd_exp, fid) % nr;
1892                 tgt_tmp  = offset;
1893                 do_div(tgt_tmp, seg_size);
1894                 tgt0_idx = do_div(tgt_tmp,  nr);
1895                 tgt_idx  = (tgt0_idx + rank) % nr;
1896
1897                 if (tgt_idx < tgt0_idx)
1898                          * Wrap around.
1899                          *
1900                          * Last segment has unusual length due to division
1901                          * rounding.
1902                         hash_adj = MAX_HASH_SIZE - seg_size * nr;
1903                 else
1904                         hash_adj = 0;
1905
1906                 hash_adj += rank * seg_size;
1907
1908                 CDEBUG(D_INODE, "Readpage hash adjustment: %x "LPX64" "
1909                        LPX64"/%x -> "LPX64"/%x\n", rank, hash_adj,
1910                        offset, tgt0_idx, offset + hash_adj, tgt_idx);
1911
1912                 offset = (offset + hash_adj) & MAX_HASH_SIZE;
1913                 rid = lsm->mea_oinfo[tgt_idx].lmo_fid;
1914                 tgt = lmv_get_target(lmv, lsm->mea_oinfo[tgt_idx].lmo_mds);
1915
1916                 CDEBUG(D_INODE, "Forward to "DFID" with offset %lu i %d\n",
1917                        PFID(&rid), (unsigned long)offset, tgt_idx);
1918         }
1919         */
1920         tgt = lmv_find_target(lmv, &op_data->op_fid1);
1921         if (IS_ERR(tgt))
1922                 RETURN(PTR_ERR(tgt));
1923
1924         rc = md_readpage(tgt->ltd_exp, op_data, pages, request);
1925         if (rc != 0)
1926                 RETURN(rc);
1927
1928         nrdpgs = ((*request)->rq_bulk->bd_nob_transferred + CFS_PAGE_SIZE - 1)
1929                  >> CFS_PAGE_SHIFT;
1930         nlupgs = (*request)->rq_bulk->bd_nob_transferred >> LU_PAGE_SHIFT;
1931         LASSERT(!((*request)->rq_bulk->bd_nob_transferred & ~LU_PAGE_MASK));
1932         LASSERT(nrdpgs > 0 && nrdpgs <= op_data->op_npages);
1933
1934         CDEBUG(D_INODE, "read %d(%d)/%d pages\n", nrdpgs, nlupgs,
1935                op_data->op_npages);
1936
1937         for (i = 0; i < nrdpgs; i++) {
1938 #if CFS_PAGE_SIZE > LU_PAGE_SIZE
1939                 struct lu_dirpage *first;
1940                 __u64 hash_end = 0;
1941                 __u32 flags = 0;
1942 #endif
1943                 struct lu_dirent *tmp = NULL;
1944
1945                 dp = cfs_kmap(pages[i]);
1946                 ent = lu_dirent_start(dp);
1947 #if CFS_PAGE_SIZE > LU_PAGE_SIZE
1948                 first = dp;
1949                 hash_end = dp->ldp_hash_end;
1950 repeat:
1951 #endif
1952                 nlupgs--;
1953
1954                 for (tmp = ent; ent != NULL;
1955                      tmp = ent, ent = lu_dirent_next(ent));
1956 #if CFS_PAGE_SIZE > LU_PAGE_SIZE
1957                 dp = (struct lu_dirpage *)((char *)dp + LU_PAGE_SIZE);
1958                 if (((unsigned long)dp & ~CFS_PAGE_MASK) && nlupgs > 0) {
1959                         ent = lu_dirent_start(dp);
1960
1961                         if (tmp) {
1962                                 /* enlarge the end entry lde_reclen from 0 to
1963                                  * first entry of next lu_dirpage, in this way
1964                                  * several lu_dirpages can be stored into one
1965                                  * client page on client. */
1966                                 tmp = ((void *)tmp) +
1967                                       le16_to_cpu(tmp->lde_reclen);
1968                                 tmp->lde_reclen =
1969                                         cpu_to_le16((char *)(dp->ldp_entries) -
1970                                                     (char *)tmp);
1971                                 goto repeat;
1972                         }
1973                 }
1974                 first->ldp_hash_end = hash_end;
1975                 first->ldp_flags &= ~cpu_to_le32(LDF_COLLIDE);
1976                 first->ldp_flags |= flags & cpu_to_le32(LDF_COLLIDE);
1977 #else
1978                 SET_BUT_UNUSED(tmp);
1979 #endif
1980                 cfs_kunmap(pages[i]);
1981         }
1982         RETURN(rc);
1983 }
1984
1985 static int lmv_unlink(struct obd_export *exp, struct md_op_data *op_data,
1986                       struct ptlrpc_request **request)
1987 {
1988         struct obd_device       *obd = exp->exp_obd;
1989         struct lmv_obd          *lmv = &obd->u.lmv;
1990         struct lmv_tgt_desc     *tgt = NULL;
1991         struct mdt_body         *body;
1992         int                     rc;
1993         ENTRY;
1994
1995         rc = lmv_check_connect(obd);
1996         if (rc)
1997                 RETURN(rc);
1998 retry:
1999         /* Send unlink requests to the MDT where the child is located */
2000         if (likely(!fid_is_zero(&op_data->op_fid2)))
2001                 tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid2);
2002         else
2003                 tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
2004         if (IS_ERR(tgt))
2005                 RETURN(PTR_ERR(tgt));
2006
2007         op_data->op_fsuid = cfs_curproc_fsuid();
2008         op_data->op_fsgid = cfs_curproc_fsgid();
2009         op_data->op_cap = cfs_curproc_cap_pack();
2010
2011         /*
2012          * If child's fid is given, cancel unused locks for it if it is from
2013          * another export than parent.
2014          *
2015          * LOOKUP lock for child (fid3) should also be cancelled on parent
2016          * tgt_tgt in mdc_unlink().
2017          */
2018         op_data->op_flags |= MF_MDC_CANCEL_FID1 | MF_MDC_CANCEL_FID3;
2019
2020         /*
2021          * Cancel FULL locks on child (fid3).
2022          */
2023         rc = lmv_early_cancel(exp, op_data, tgt->ltd_idx, LCK_EX,
2024                               MDS_INODELOCK_FULL, MF_MDC_CANCEL_FID3);
2025
2026         if (rc != 0)
2027                 RETURN(rc);
2028
2029         CDEBUG(D_INODE, "unlink with fid="DFID"/"DFID" -> mds #%d\n",
2030                PFID(&op_data->op_fid1), PFID(&op_data->op_fid2), tgt->ltd_idx);
2031
2032         rc = md_unlink(tgt->ltd_exp, op_data, request);
2033         if (rc != 0 && rc != -EREMOTE)
2034                 RETURN(rc);
2035
2036         body = req_capsule_server_get(&(*request)->rq_pill, &RMF_MDT_BODY);
2037         if (body == NULL)
2038                 RETURN(-EPROTO);
2039
2040         /* Not cross-ref case, just get out of here. */
2041         if (likely(!(body->valid & OBD_MD_MDS)))
2042                 RETURN(0);
2043
2044         CDEBUG(D_INODE, "%s: try unlink to another MDT for "DFID"\n",
2045                exp->exp_obd->obd_name, PFID(&body->fid1));
2046
2047         /* This is a remote object, try remote MDT, Note: it may
2048          * try more than 1 time here, Considering following case
2049          * /mnt/lustre is root on MDT0, remote1 is on MDT1
2050          * 1. Initially A does not know where remote1 is, it send
2051          *    unlink RPC to MDT0, MDT0 return -EREMOTE, it will
2052          *    resend unlink RPC to MDT1 (retry 1st time).
2053          *
2054          * 2. During the unlink RPC in flight,
2055          *    client B mv /mnt/lustre/remote1 /mnt/lustre/remote2
2056          *    and create new remote1, but on MDT0
2057          *
2058          * 3. MDT1 get unlink RPC(from A), then do remote lock on
2059          *    /mnt/lustre, then lookup get fid of remote1, and find
2060          *    it is remote dir again, and replay -EREMOTE again.
2061          *
2062          * 4. Then A will resend unlink RPC to MDT0. (retry 2nd times).
2063          *
2064          * In theory, it might try unlimited time here, but it should
2065          * be very rare case.  */
2066         op_data->op_fid2 = body->fid1;
2067         ptlrpc_req_finished(*request);
2068         *request = NULL;
2069
2070         goto retry;
2071 }
2072
2073 static int lmv_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
2074 {
2075         struct lmv_obd *lmv = &obd->u.lmv;
2076         int rc = 0;
2077
2078         switch (stage) {
2079         case OBD_CLEANUP_EARLY:
2080                 /* XXX: here should be calling obd_precleanup() down to
2081                  * stack. */
2082                 break;
2083         case OBD_CLEANUP_EXPORTS:
2084                 fld_client_proc_fini(&lmv->lmv_fld);
2085                 lprocfs_obd_cleanup(obd);
2086                 break;
2087         default:
2088                 break;
2089         }
2090         RETURN(rc);
2091 }
2092
2093 static int lmv_get_info(const struct lu_env *env, struct obd_export *exp,
2094                         __u32 keylen, void *key, __u32 *vallen, void *val,
2095                         struct lov_stripe_md *lsm)
2096 {
2097         struct obd_device       *obd;
2098         struct lmv_obd          *lmv;
2099         int                      rc = 0;
2100         ENTRY;
2101
2102         obd = class_exp2obd(exp);
2103         if (obd == NULL) {
2104                 CDEBUG(D_IOCTL, "Invalid client cookie "LPX64"\n",
2105                        exp->exp_handle.h_cookie);
2106                 RETURN(-EINVAL);
2107         }
2108
2109         lmv = &obd->u.lmv;
2110         if (keylen >= strlen("remote_flag") && !strcmp(key, "remote_flag")) {
2111                 struct lmv_tgt_desc *tgt;
2112                 int i;
2113
2114                 rc = lmv_check_connect(obd);
2115                 if (rc)
2116                         RETURN(rc);
2117
2118                 LASSERT(*vallen == sizeof(__u32));
2119                 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2120                         tgt = lmv->tgts[i];
2121                         /*
2122                          * All tgts should be connected when this gets called.
2123                          */
2124                         if (tgt == NULL || tgt->ltd_exp == NULL)
2125                                 continue;
2126
2127                         if (!obd_get_info(env, tgt->ltd_exp, keylen, key,
2128                                           vallen, val, NULL))
2129                                 RETURN(0);
2130                 }
2131                 RETURN(-EINVAL);
2132         } else if (KEY_IS(KEY_MAX_EASIZE) || KEY_IS(KEY_CONN_DATA)) {
2133                 rc = lmv_check_connect(obd);
2134                 if (rc)
2135                         RETURN(rc);
2136
2137                 /*
2138                  * Forwarding this request to first MDS, it should know LOV
2139                  * desc.
2140                  */
2141                 rc = obd_get_info(env, lmv->tgts[0]->ltd_exp, keylen, key,
2142                                   vallen, val, NULL);
2143                 if (!rc && KEY_IS(KEY_CONN_DATA))
2144                         exp->exp_connect_data = *(struct obd_connect_data *)val;
2145                 RETURN(rc);
2146         } else if (KEY_IS(KEY_TGT_COUNT)) {
2147                 *((int *)val) = lmv->desc.ld_tgt_count;
2148                 RETURN(0);
2149         }
2150
2151         CDEBUG(D_IOCTL, "Invalid key\n");
2152         RETURN(-EINVAL);
2153 }
2154
2155 int lmv_set_info_async(const struct lu_env *env, struct obd_export *exp,
2156                        obd_count keylen, void *key, obd_count vallen,
2157                        void *val, struct ptlrpc_request_set *set)
2158 {
2159         struct lmv_tgt_desc    *tgt;
2160         struct obd_device      *obd;
2161         struct lmv_obd         *lmv;
2162         int rc = 0;
2163         ENTRY;
2164
2165         obd = class_exp2obd(exp);
2166         if (obd == NULL) {
2167                 CDEBUG(D_IOCTL, "Invalid client cookie "LPX64"\n",
2168                        exp->exp_handle.h_cookie);
2169                 RETURN(-EINVAL);
2170         }
2171         lmv = &obd->u.lmv;
2172
2173         if (KEY_IS(KEY_READ_ONLY) || KEY_IS(KEY_FLUSH_CTX)) {
2174                 int i, err = 0;
2175
2176                 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2177                         tgt = lmv->tgts[i];
2178
2179                         if (tgt == NULL || tgt->ltd_exp == NULL)
2180                                 continue;
2181
2182                         err = obd_set_info_async(env, tgt->ltd_exp,
2183                                                  keylen, key, vallen, val, set);
2184                         if (err && rc == 0)
2185                                 rc = err;
2186                 }
2187
2188                 RETURN(rc);
2189         }
2190
2191         RETURN(-EINVAL);
2192 }
2193
2194 int lmv_packmd(struct obd_export *exp, struct lov_mds_md **lmmp,
2195                struct lov_stripe_md *lsm)
2196 {
2197         struct obd_device         *obd = class_exp2obd(exp);
2198         struct lmv_obd            *lmv = &obd->u.lmv;
2199         struct lmv_stripe_md      *meap;
2200         struct lmv_stripe_md      *lsmp;
2201         int                        mea_size;
2202         int                        i;
2203         ENTRY;
2204
2205         mea_size = lmv_get_easize(lmv);
2206         if (!lmmp)
2207                 RETURN(mea_size);
2208
2209         if (*lmmp && !lsm) {
2210                 OBD_FREE_LARGE(*lmmp, mea_size);
2211                 *lmmp = NULL;
2212                 RETURN(0);
2213         }
2214
2215         if (*lmmp == NULL) {
2216                 OBD_ALLOC_LARGE(*lmmp, mea_size);
2217                 if (*lmmp == NULL)
2218                         RETURN(-ENOMEM);
2219         }
2220
2221         if (!lsm)
2222                 RETURN(mea_size);
2223
2224         lsmp = (struct lmv_stripe_md *)lsm;
2225         meap = (struct lmv_stripe_md *)*lmmp;
2226
2227         if (lsmp->mea_magic != MEA_MAGIC_LAST_CHAR &&
2228             lsmp->mea_magic != MEA_MAGIC_ALL_CHARS)
2229                 RETURN(-EINVAL);
2230
2231         meap->mea_magic = cpu_to_le32(lsmp->mea_magic);
2232         meap->mea_count = cpu_to_le32(lsmp->mea_count);
2233         meap->mea_master = cpu_to_le32(lsmp->mea_master);
2234
2235         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2236                 meap->mea_ids[i] = meap->mea_ids[i];
2237                 fid_cpu_to_le(&meap->mea_ids[i], &meap->mea_ids[i]);
2238         }
2239
2240         RETURN(mea_size);
2241 }
2242
2243 int lmv_unpackmd(struct obd_export *exp, struct lov_stripe_md **lsmp,
2244                  struct lov_mds_md *lmm, int lmm_size)
2245 {
2246         struct obd_device          *obd = class_exp2obd(exp);
2247         struct lmv_stripe_md      **tmea = (struct lmv_stripe_md **)lsmp;
2248         struct lmv_stripe_md       *mea = (struct lmv_stripe_md *)lmm;
2249         struct lmv_obd             *lmv = &obd->u.lmv;
2250         int                         mea_size;
2251         int                         i;
2252         __u32                       magic;
2253         ENTRY;
2254
2255         mea_size = lmv_get_easize(lmv);
2256         if (lsmp == NULL)
2257                 return mea_size;
2258
2259         if (*lsmp != NULL && lmm == NULL) {
2260                 OBD_FREE_LARGE(*tmea, mea_size);
2261                 *lsmp = NULL;
2262                 RETURN(0);
2263         }
2264
2265         LASSERT(mea_size == lmm_size);
2266
2267         OBD_ALLOC_LARGE(*tmea, mea_size);
2268         if (*tmea == NULL)
2269                 RETURN(-ENOMEM);
2270
2271         if (!lmm)
2272                 RETURN(mea_size);
2273
2274         if (mea->mea_magic == MEA_MAGIC_LAST_CHAR ||
2275             mea->mea_magic == MEA_MAGIC_ALL_CHARS ||
2276             mea->mea_magic == MEA_MAGIC_HASH_SEGMENT)
2277         {
2278                 magic = le32_to_cpu(mea->mea_magic);
2279         } else {
2280                 /*
2281                  * Old mea is not handled here.
2282                  */
2283                 CERROR("Old not supportable EA is found\n");
2284                 LBUG();
2285         }
2286
2287         (*tmea)->mea_magic = magic;
2288         (*tmea)->mea_count = le32_to_cpu(mea->mea_count);
2289         (*tmea)->mea_master = le32_to_cpu(mea->mea_master);
2290
2291         for (i = 0; i < (*tmea)->mea_count; i++) {
2292                 (*tmea)->mea_ids[i] = mea->mea_ids[i];
2293                 fid_le_to_cpu(&(*tmea)->mea_ids[i], &(*tmea)->mea_ids[i]);
2294         }
2295         RETURN(mea_size);
2296 }
2297
2298 static int lmv_cancel_unused(struct obd_export *exp, const struct lu_fid *fid,
2299                              ldlm_policy_data_t *policy, ldlm_mode_t mode,
2300                              ldlm_cancel_flags_t flags, void *opaque)
2301 {
2302         struct obd_device       *obd = exp->exp_obd;
2303         struct lmv_obd          *lmv = &obd->u.lmv;
2304         int                      rc = 0;
2305         int                      err;
2306         int                      i;
2307         ENTRY;
2308
2309         LASSERT(fid != NULL);
2310
2311         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2312                 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL ||
2313                     lmv->tgts[i]->ltd_active == 0)
2314                         continue;
2315
2316                 err = md_cancel_unused(lmv->tgts[i]->ltd_exp, fid,
2317                                        policy, mode, flags, opaque);
2318                 if (!rc)
2319                         rc = err;
2320         }
2321         RETURN(rc);
2322 }
2323
2324 int lmv_set_lock_data(struct obd_export *exp, __u64 *lockh, void *data,
2325                       __u64 *bits)
2326 {
2327         struct lmv_obd          *lmv = &exp->exp_obd->u.lmv;
2328         int                      rc;
2329         ENTRY;
2330
2331         rc =  md_set_lock_data(lmv->tgts[0]->ltd_exp, lockh, data, bits);
2332         RETURN(rc);
2333 }
2334
2335 ldlm_mode_t lmv_lock_match(struct obd_export *exp, __u64 flags,
2336                            const struct lu_fid *fid, ldlm_type_t type,
2337                            ldlm_policy_data_t *policy, ldlm_mode_t mode,
2338                            struct lustre_handle *lockh)
2339 {
2340         struct obd_device       *obd = exp->exp_obd;
2341         struct lmv_obd          *lmv = &obd->u.lmv;
2342         ldlm_mode_t              rc;
2343         int                      i;
2344         ENTRY;
2345
2346         CDEBUG(D_INODE, "Lock match for "DFID"\n", PFID(fid));
2347
2348         /*
2349          * With CMD every object can have two locks in different namespaces:
2350          * lookup lock in space of mds storing direntry and update/open lock in
2351          * space of mds storing inode. Thus we check all targets, not only that
2352          * one fid was created in.
2353          */
2354         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2355                 if (lmv->tgts[i] == NULL ||
2356                     lmv->tgts[i]->ltd_exp == NULL ||
2357                     lmv->tgts[i]->ltd_active == 0)
2358                         continue;
2359
2360                 rc = md_lock_match(lmv->tgts[i]->ltd_exp, flags, fid,
2361                                    type, policy, mode, lockh);
2362                 if (rc)
2363                         RETURN(rc);
2364         }
2365
2366         RETURN(0);
2367 }
2368
2369 int lmv_get_lustre_md(struct obd_export *exp, struct ptlrpc_request *req,
2370                       struct obd_export *dt_exp, struct obd_export *md_exp,
2371                       struct lustre_md *md)
2372 {
2373         struct lmv_obd          *lmv = &exp->exp_obd->u.lmv;
2374
2375         return md_get_lustre_md(lmv->tgts[0]->ltd_exp, req, dt_exp, md_exp, md);
2376 }
2377
2378 int lmv_free_lustre_md(struct obd_export *exp, struct lustre_md *md)
2379 {
2380         struct obd_device       *obd = exp->exp_obd;
2381         struct lmv_obd          *lmv = &obd->u.lmv;
2382         ENTRY;
2383
2384         if (md->mea)
2385                 obd_free_memmd(exp, (void *)&md->mea);
2386         RETURN(md_free_lustre_md(lmv->tgts[0]->ltd_exp, md));
2387 }
2388
2389 int lmv_set_open_replay_data(struct obd_export *exp,
2390                              struct obd_client_handle *och,
2391                              struct ptlrpc_request *open_req)
2392 {
2393         struct obd_device       *obd = exp->exp_obd;
2394         struct lmv_obd          *lmv = &obd->u.lmv;
2395         struct lmv_tgt_desc     *tgt;
2396         ENTRY;
2397
2398         tgt = lmv_find_target(lmv, &och->och_fid);
2399         if (IS_ERR(tgt))
2400                 RETURN(PTR_ERR(tgt));
2401
2402         RETURN(md_set_open_replay_data(tgt->ltd_exp, och, open_req));
2403 }
2404
2405 int lmv_clear_open_replay_data(struct obd_export *exp,
2406                                struct obd_client_handle *och)
2407 {
2408         struct obd_device       *obd = exp->exp_obd;
2409         struct lmv_obd          *lmv = &obd->u.lmv;
2410         struct lmv_tgt_desc     *tgt;
2411         ENTRY;
2412
2413         tgt = lmv_find_target(lmv, &och->och_fid);
2414         if (IS_ERR(tgt))
2415                 RETURN(PTR_ERR(tgt));
2416
2417         RETURN(md_clear_open_replay_data(tgt->ltd_exp, och));
2418 }
2419
2420 static int lmv_get_remote_perm(struct obd_export *exp,
2421                                const struct lu_fid *fid,
2422                                struct obd_capa *oc, __u32 suppgid,
2423                                struct ptlrpc_request **request)
2424 {
2425         struct obd_device       *obd = exp->exp_obd;
2426         struct lmv_obd          *lmv = &obd->u.lmv;
2427         struct lmv_tgt_desc     *tgt;
2428         int                      rc;
2429         ENTRY;
2430
2431         rc = lmv_check_connect(obd);
2432         if (rc)
2433                 RETURN(rc);
2434
2435         tgt = lmv_find_target(lmv, fid);
2436         if (IS_ERR(tgt))
2437                 RETURN(PTR_ERR(tgt));
2438
2439         rc = md_get_remote_perm(tgt->ltd_exp, fid, oc, suppgid, request);
2440         RETURN(rc);
2441 }
2442
2443 static int lmv_renew_capa(struct obd_export *exp, struct obd_capa *oc,
2444                           renew_capa_cb_t cb)
2445 {
2446         struct obd_device       *obd = exp->exp_obd;
2447         struct lmv_obd          *lmv = &obd->u.lmv;
2448         struct lmv_tgt_desc     *tgt;
2449         int                      rc;
2450         ENTRY;
2451
2452         rc = lmv_check_connect(obd);
2453         if (rc)
2454                 RETURN(rc);
2455
2456         tgt = lmv_find_target(lmv, &oc->c_capa.lc_fid);
2457         if (IS_ERR(tgt))
2458                 RETURN(PTR_ERR(tgt));
2459
2460         rc = md_renew_capa(tgt->ltd_exp, oc, cb);
2461         RETURN(rc);
2462 }
2463
2464 int lmv_unpack_capa(struct obd_export *exp, struct ptlrpc_request *req,
2465                     const struct req_msg_field *field, struct obd_capa **oc)
2466 {
2467         struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
2468
2469         return md_unpack_capa(lmv->tgts[0]->ltd_exp, req, field, oc);
2470 }
2471
2472 int lmv_intent_getattr_async(struct obd_export *exp,
2473                              struct md_enqueue_info *minfo,
2474                              struct ldlm_enqueue_info *einfo)
2475 {
2476         struct md_op_data       *op_data = &minfo->mi_data;
2477         struct obd_device       *obd = exp->exp_obd;
2478         struct lmv_obd          *lmv = &obd->u.lmv;
2479         struct lmv_tgt_desc     *tgt = NULL;
2480         int                      rc;
2481         ENTRY;
2482
2483         rc = lmv_check_connect(obd);
2484         if (rc)
2485                 RETURN(rc);
2486
2487         tgt = lmv_find_target(lmv, &op_data->op_fid1);
2488         if (IS_ERR(tgt))
2489                 RETURN(PTR_ERR(tgt));
2490
2491         rc = md_intent_getattr_async(tgt->ltd_exp, minfo, einfo);
2492         RETURN(rc);
2493 }
2494
2495 int lmv_revalidate_lock(struct obd_export *exp, struct lookup_intent *it,
2496                         struct lu_fid *fid, __u64 *bits)
2497 {
2498         struct obd_device       *obd = exp->exp_obd;
2499         struct lmv_obd          *lmv = &obd->u.lmv;
2500         struct lmv_tgt_desc     *tgt;
2501         int                      rc;
2502         ENTRY;
2503
2504         rc = lmv_check_connect(obd);
2505         if (rc)
2506                 RETURN(rc);
2507
2508         tgt = lmv_find_target(lmv, fid);
2509         if (IS_ERR(tgt))
2510                 RETURN(PTR_ERR(tgt));
2511
2512         rc = md_revalidate_lock(tgt->ltd_exp, it, fid, bits);
2513         RETURN(rc);
2514 }
2515
2516 /**
2517  * For lmv, only need to send request to master MDT, and the master MDT will
2518  * process with other slave MDTs. The only exception is Q_GETOQUOTA for which
2519  * we directly fetch data from the slave MDTs.
2520  */
2521 int lmv_quotactl(struct obd_device *unused, struct obd_export *exp,
2522                  struct obd_quotactl *oqctl)
2523 {
2524         struct obd_device   *obd = class_exp2obd(exp);
2525         struct lmv_obd      *lmv = &obd->u.lmv;
2526         struct lmv_tgt_desc *tgt = lmv->tgts[0];
2527         int                  rc = 0, i;
2528         __u64                curspace, curinodes;
2529         ENTRY;
2530
2531         if (!lmv->desc.ld_tgt_count || !tgt->ltd_active) {
2532                 CERROR("master lmv inactive\n");
2533                 RETURN(-EIO);
2534         }
2535
2536         if (oqctl->qc_cmd != Q_GETOQUOTA) {
2537                 rc = obd_quotactl(tgt->ltd_exp, oqctl);
2538                 RETURN(rc);
2539         }
2540
2541         curspace = curinodes = 0;
2542         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2543                 int err;
2544                 tgt = lmv->tgts[i];
2545
2546                 if (tgt == NULL || tgt->ltd_exp == NULL || tgt->ltd_active == 0)
2547                         continue;
2548                 if (!tgt->ltd_active) {
2549                         CDEBUG(D_HA, "mdt %d is inactive.\n", i);
2550                         continue;
2551                 }
2552
2553                 err = obd_quotactl(tgt->ltd_exp, oqctl);
2554                 if (err) {
2555                         CERROR("getquota on mdt %d failed. %d\n", i, err);
2556                         if (!rc)
2557                                 rc = err;
2558                 } else {
2559                         curspace += oqctl->qc_dqblk.dqb_curspace;
2560                         curinodes += oqctl->qc_dqblk.dqb_curinodes;
2561                 }
2562         }
2563         oqctl->qc_dqblk.dqb_curspace = curspace;
2564         oqctl->qc_dqblk.dqb_curinodes = curinodes;
2565
2566         RETURN(rc);
2567 }
2568
2569 int lmv_quotacheck(struct obd_device *unused, struct obd_export *exp,
2570                    struct obd_quotactl *oqctl)
2571 {
2572         struct obd_device   *obd = class_exp2obd(exp);
2573         struct lmv_obd      *lmv = &obd->u.lmv;
2574         struct lmv_tgt_desc *tgt;
2575         int                  i, rc = 0;
2576         ENTRY;
2577
2578         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2579                 int err;
2580                 tgt = lmv->tgts[i];
2581                 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active) {
2582                         CERROR("lmv idx %d inactive\n", i);
2583                         RETURN(-EIO);
2584                 }
2585
2586                 err = obd_quotacheck(tgt->ltd_exp, oqctl);
2587                 if (err && !rc)
2588                         rc = err;
2589         }
2590
2591         RETURN(rc);
2592 }
2593
2594 struct obd_ops lmv_obd_ops = {
2595         .o_owner                = THIS_MODULE,
2596         .o_setup                = lmv_setup,
2597         .o_cleanup              = lmv_cleanup,
2598         .o_precleanup           = lmv_precleanup,
2599         .o_process_config       = lmv_process_config,
2600         .o_connect              = lmv_connect,
2601         .o_disconnect           = lmv_disconnect,
2602         .o_statfs               = lmv_statfs,
2603         .o_get_info             = lmv_get_info,
2604         .o_set_info_async       = lmv_set_info_async,
2605         .o_packmd               = lmv_packmd,
2606         .o_unpackmd             = lmv_unpackmd,
2607         .o_notify               = lmv_notify,
2608         .o_get_uuid             = lmv_get_uuid,
2609         .o_iocontrol            = lmv_iocontrol,
2610         .o_quotacheck           = lmv_quotacheck,
2611         .o_quotactl             = lmv_quotactl
2612 };
2613
2614 struct md_ops lmv_md_ops = {
2615         .m_getstatus            = lmv_getstatus,
2616         .m_null_inode           = lmv_null_inode,
2617         .m_find_cbdata          = lmv_find_cbdata,
2618         .m_close                = lmv_close,
2619         .m_create               = lmv_create,
2620         .m_done_writing         = lmv_done_writing,
2621         .m_enqueue              = lmv_enqueue,
2622         .m_getattr              = lmv_getattr,
2623         .m_getxattr             = lmv_getxattr,
2624         .m_getattr_name         = lmv_getattr_name,
2625         .m_intent_lock          = lmv_intent_lock,
2626         .m_link                 = lmv_link,
2627         .m_rename               = lmv_rename,
2628         .m_setattr              = lmv_setattr,
2629         .m_setxattr             = lmv_setxattr,
2630         .m_sync                 = lmv_sync,
2631         .m_readpage             = lmv_readpage,
2632         .m_unlink               = lmv_unlink,
2633         .m_init_ea_size         = lmv_init_ea_size,
2634         .m_cancel_unused        = lmv_cancel_unused,
2635         .m_set_lock_data        = lmv_set_lock_data,
2636         .m_lock_match           = lmv_lock_match,
2637         .m_get_lustre_md        = lmv_get_lustre_md,
2638         .m_free_lustre_md       = lmv_free_lustre_md,
2639         .m_set_open_replay_data = lmv_set_open_replay_data,
2640         .m_clear_open_replay_data = lmv_clear_open_replay_data,
2641         .m_renew_capa           = lmv_renew_capa,
2642         .m_unpack_capa          = lmv_unpack_capa,
2643         .m_get_remote_perm      = lmv_get_remote_perm,
2644         .m_intent_getattr_async = lmv_intent_getattr_async,
2645         .m_revalidate_lock      = lmv_revalidate_lock
2646 };
2647
2648 int __init lmv_init(void)
2649 {
2650         struct lprocfs_static_vars lvars;
2651         int                        rc;
2652
2653         lprocfs_lmv_init_vars(&lvars);
2654
2655         rc = class_register_type(&lmv_obd_ops, &lmv_md_ops,
2656                                  lvars.module_vars, LUSTRE_LMV_NAME, NULL);
2657         return rc;
2658 }
2659
2660 #ifdef __KERNEL__
2661 static void lmv_exit(void)
2662 {
2663         class_unregister_type(LUSTRE_LMV_NAME);
2664 }
2665
2666 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
2667 MODULE_DESCRIPTION("Lustre Logical Metadata Volume OBD driver");
2668 MODULE_LICENSE("GPL");
2669
2670 module_init(lmv_init);
2671 module_exit(lmv_exit);
2672 #endif