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