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