lustre/lmv/lmv_obd.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2004, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2011, 2012, Intel Corporation.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  */
  36
  37 #define DEBUG_SUBSYSTEM S_LMV
  38 #ifdef __KERNEL__
  39 #include <linux/slab.h>
  40 #include <linux/module.h>
  41 #include <linux/init.h>
  42 #include <linux/slab.h>
  43 #include <linux/pagemap.h>
  44 #include <linux/mm.h>
  45 #include <asm/div64.h>
  46 #include <linux/seq_file.h>
  47 #include <linux/namei.h>
  48 #else
  49 #include <liblustre.h>
  50 #endif
  51
  52 #include <lustre/lustre_idl.h>
  53 #include <obd_support.h>
  54 #include <lustre_lib.h>
  55 #include <lustre_net.h>
  56 #include <obd_class.h>
  57 #include <lprocfs_status.h>
  58 #include <lustre_lite.h>
  59 #include <lustre_fid.h>
  60 #include "lmv_internal.h"
  61
  62 static void lmv_activate_target(struct lmv_obd *lmv,
  63                                 struct lmv_tgt_desc *tgt,
  64                                 int activate)
  65 {
  66         if (tgt->ltd_active == activate)
  67                 return;
  68
  69         tgt->ltd_active = activate;
  70         lmv->desc.ld_active_tgt_count += (activate ? 1 : -1);
  71 }
  72
  73 /**
  74  * Error codes:
  75  *
  76  *  -EINVAL  : UUID can't be found in the LMV's target list
  77  *  -ENOTCONN: The UUID is found, but the target connection is bad (!)
  78  *  -EBADF   : The UUID is found, but the OBD of the wrong type (!)
  79  */
  80 static int lmv_set_mdc_active(struct lmv_obd *lmv, struct obd_uuid *uuid,
  81                               int activate)
  82 {
  83         struct lmv_tgt_desc    *tgt;
  84         struct obd_device      *obd;
  85         int                     i;
  86         int                     rc = 0;
  87         ENTRY;
  88
  89         CDEBUG(D_INFO, "Searching in lmv %p for uuid %s (activate=%d)\n",
  90                lmv, uuid->uuid, activate);
  91
  92         spin_lock(&lmv->lmv_lock);
  93         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
  94                 tgt = lmv->tgts[i];
  95                 if (tgt == NULL || tgt->ltd_exp == NULL)
  96                         continue;
  97
  98                 CDEBUG(D_INFO, "Target idx %d is %s conn "LPX64"\n", i,
  99                        tgt->ltd_uuid.uuid, tgt->ltd_exp->exp_handle.h_cookie);
 100
 101                 if (obd_uuid_equals(uuid, &tgt->ltd_uuid))
 102                         break;
 103         }
 104
 105         if (i == lmv->desc.ld_tgt_count)
 106                 GOTO(out_lmv_lock, rc = -EINVAL);
 107
 108         obd = class_exp2obd(tgt->ltd_exp);
 109         if (obd == NULL)
 110                 GOTO(out_lmv_lock, rc = -ENOTCONN);
 111
 112         CDEBUG(D_INFO, "Found OBD %s=%s device %d (%p) type %s at LMV idx %d\n",
 113                obd->obd_name, obd->obd_uuid.uuid, obd->obd_minor, obd,
 114                obd->obd_type->typ_name, i);
 115         LASSERT(strcmp(obd->obd_type->typ_name, LUSTRE_MDC_NAME) == 0);
 116
 117         if (tgt->ltd_active == activate) {
 118                 CDEBUG(D_INFO, "OBD %p already %sactive!\n", obd,
 119                        activate ? "" : "in");
 120                 GOTO(out_lmv_lock, rc);
 121         }
 122
 123         CDEBUG(D_INFO, "Marking OBD %p %sactive\n", obd,
 124                activate ? "" : "in");
 125         lmv_activate_target(lmv, tgt, activate);
 126         EXIT;
 127
 128  out_lmv_lock:
 129         spin_unlock(&lmv->lmv_lock);
 130         return rc;
 131 }
 132
 133 struct obd_uuid *lmv_get_uuid(struct obd_export *exp)
 134 {
 135         struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
 136
 137         return obd_get_uuid(lmv->tgts[0]->ltd_exp);
 138 }
 139
 140 static int lmv_notify(struct obd_device *obd, struct obd_device *watched,
 141                       enum obd_notify_event ev, void *data)
 142 {
 143         struct obd_connect_data *conn_data;
 144         struct lmv_obd          *lmv = &obd->u.lmv;
 145         struct obd_uuid         *uuid;
 146         int                      rc = 0;
 147         ENTRY;
 148
 149         if (strcmp(watched->obd_type->typ_name, LUSTRE_MDC_NAME)) {
 150                 CERROR("unexpected notification of %s %s!\n",
 151                        watched->obd_type->typ_name,
 152                        watched->obd_name);
 153                 RETURN(-EINVAL);
 154         }
 155
 156         uuid = &watched->u.cli.cl_target_uuid;
 157         if (ev == OBD_NOTIFY_ACTIVE || ev == OBD_NOTIFY_INACTIVE) {
 158                 /*
 159                  * Set MDC as active before notifying the observer, so the
 160                  * observer can use the MDC normally.
 161                  */
 162                 rc = lmv_set_mdc_active(lmv, uuid,
 163                                         ev == OBD_NOTIFY_ACTIVE);
 164                 if (rc) {
 165                         CERROR("%sactivation of %s failed: %d\n",
 166                                ev == OBD_NOTIFY_ACTIVE ? "" : "de",
 167                                uuid->uuid, rc);
 168                         RETURN(rc);
 169                 }
 170         } else if (ev == OBD_NOTIFY_OCD) {
 171                 conn_data = &watched->u.cli.cl_import->imp_connect_data;
 172                 /*
 173                  * XXX: Make sure that ocd_connect_flags from all targets are
 174                  * the same. Otherwise one of MDTs runs wrong version or
 175                  * something like this.  --umka
 176                  */
 177                 obd->obd_self_export->exp_connect_data = *conn_data;
 178         }
 179 #if 0
 180         else if (ev == OBD_NOTIFY_DISCON) {
 181                 /*
 182                  * For disconnect event, flush fld cache for failout MDS case.
 183                  */
 184                 fld_client_flush(&lmv->lmv_fld);
 185         }
 186 #endif
 187         /*
 188          * Pass the notification up the chain.
 189          */
 190         if (obd->obd_observer)
 191                 rc = obd_notify(obd->obd_observer, watched, ev, data);
 192
 193         RETURN(rc);
 194 }
 195
 196 /**
 197  * This is fake connect function. Its purpose is to initialize lmv and say
 198  * caller that everything is okay. Real connection will be performed later.
 199  */
 200 static int lmv_connect(const struct lu_env *env,
 201                        struct obd_export **exp, struct obd_device *obd,
 202                        struct obd_uuid *cluuid, struct obd_connect_data *data,
 203                        void *localdata)
 204 {
 205 #ifdef __KERNEL__
 206         struct proc_dir_entry *lmv_proc_dir;
 207 #endif
 208         struct lmv_obd        *lmv = &obd->u.lmv;
 209         struct lustre_handle  conn = { 0 };
 210         int                    rc = 0;
 211         ENTRY;
 212
 213         /*
 214          * We don't want to actually do the underlying connections more than
 215          * once, so keep track.
 216          */
 217         lmv->refcount++;
 218         if (lmv->refcount > 1) {
 219                 *exp = NULL;
 220                 RETURN(0);
 221         }
 222
 223         rc = class_connect(&conn, obd, cluuid);
 224         if (rc) {
 225                 CERROR("class_connection() returned %d\n", rc);
 226                 RETURN(rc);
 227         }
 228
 229         *exp = class_conn2export(&conn);
 230         class_export_get(*exp);
 231
 232         lmv->exp = *exp;
 233         lmv->connected = 0;
 234         lmv->cluuid = *cluuid;
 235
 236         if (data)
 237                 lmv->conn_data = *data;
 238
 239 #ifdef __KERNEL__
 240         lmv_proc_dir = lprocfs_register("target_obds", obd->obd_proc_entry,
 241                                         NULL, NULL);
 242         if (IS_ERR(lmv_proc_dir)) {
 243                 CERROR("could not register /proc/fs/lustre/%s/%s/target_obds.",
 244                        obd->obd_type->typ_name, obd->obd_name);
 245                 lmv_proc_dir = NULL;
 246         }
 247 #endif
 248
 249         /*
 250          * All real clients should perform actual connection right away, because
 251          * it is possible, that LMV will not have opportunity to connect targets
 252          * and MDC stuff will be called directly, for instance while reading
 253          * ../mdc/../kbytesfree procfs file, etc.
 254          */
 255         if (data->ocd_connect_flags & OBD_CONNECT_REAL)
 256                 rc = lmv_check_connect(obd);
 257
 258 #ifdef __KERNEL__
 259         if (rc) {
 260                 if (lmv_proc_dir)
 261                         lprocfs_remove(&lmv_proc_dir);
 262         }
 263 #endif
 264
 265         RETURN(rc);
 266 }
 267
 268 static void lmv_set_timeouts(struct obd_device *obd)
 269 {
 270         struct lmv_tgt_desc   *tgt;
 271         struct lmv_obd        *lmv;
 272         int                    i;
 273
 274         lmv = &obd->u.lmv;
 275         if (lmv->server_timeout == 0)
 276                 return;
 277
 278         if (lmv->connected == 0)
 279                 return;
 280
 281         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 282                 tgt = lmv->tgts[i];
 283                 if (tgt == NULL || tgt->ltd_exp == NULL || tgt->ltd_active == 0)
 284                         continue;
 285
 286                 obd_set_info_async(NULL, tgt->ltd_exp, sizeof(KEY_INTERMDS),
 287                                    KEY_INTERMDS, 0, NULL, NULL);
 288         }
 289 }
 290
 291 static int lmv_init_ea_size(struct obd_export *exp, int easize,
 292                             int def_easize, int cookiesize)
 293 {
 294         struct obd_device   *obd = exp->exp_obd;
 295         struct lmv_obd      *lmv = &obd->u.lmv;
 296         int                  i;
 297         int                  rc = 0;
 298         int                  change = 0;
 299         ENTRY;
 300
 301         if (lmv->max_easize < easize) {
 302                 lmv->max_easize = easize;
 303                 change = 1;
 304         }
 305         if (lmv->max_def_easize < def_easize) {
 306                 lmv->max_def_easize = def_easize;
 307                 change = 1;
 308         }
 309         if (lmv->max_cookiesize < cookiesize) {
 310                 lmv->max_cookiesize = cookiesize;
 311                 change = 1;
 312         }
 313         if (change == 0)
 314                 RETURN(0);
 315
 316         if (lmv->connected == 0)
 317                 RETURN(0);
 318
 319         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 320                 if (lmv->tgts[i] == NULL ||
 321                     lmv->tgts[i]->ltd_exp == NULL ||
 322                     lmv->tgts[i]->ltd_active == 0) {
 323                         CWARN("%s: NULL export for %d\n", obd->obd_name, i);
 324                         continue;
 325                 }
 326
 327                 rc = md_init_ea_size(lmv->tgts[i]->ltd_exp, easize, def_easize,
 328                                      cookiesize);
 329                 if (rc) {
 330                         CERROR("%s: obd_init_ea_size() failed on MDT target %d:"
 331                                " rc = %d.\n", obd->obd_name, i, rc);
 332                         break;
 333                 }
 334         }
 335         RETURN(rc);
 336 }
 337
 338 #define MAX_STRING_SIZE 128
 339
 340 int lmv_connect_mdc(struct obd_device *obd, struct lmv_tgt_desc *tgt)
 341 {
 342 #ifdef __KERNEL__
 343         struct proc_dir_entry   *lmv_proc_dir;
 344 #endif
 345         struct lmv_obd          *lmv = &obd->u.lmv;
 346         struct obd_uuid         *cluuid = &lmv->cluuid;
 347         struct obd_uuid          lmv_mdc_uuid = { "LMV_MDC_UUID" };
 348         struct obd_device       *mdc_obd;
 349         struct obd_export       *mdc_exp;
 350         struct lu_fld_target     target;
 351         int                      rc;
 352         ENTRY;
 353
 354         mdc_obd = class_find_client_obd(&tgt->ltd_uuid, LUSTRE_MDC_NAME,
 355                                         &obd->obd_uuid);
 356         if (!mdc_obd) {
 357                 CERROR("target %s not attached\n", tgt->ltd_uuid.uuid);
 358                 RETURN(-EINVAL);
 359         }
 360
 361         CDEBUG(D_CONFIG, "connect to %s(%s) - %s, %s FOR %s\n",
 362                 mdc_obd->obd_name, mdc_obd->obd_uuid.uuid,
 363                 tgt->ltd_uuid.uuid, obd->obd_uuid.uuid,
 364                 cluuid->uuid);
 365
 366         if (!mdc_obd->obd_set_up) {
 367                 CERROR("target %s is not set up\n", tgt->ltd_uuid.uuid);
 368                 RETURN(-EINVAL);
 369         }
 370
 371         rc = obd_connect(NULL, &mdc_exp, mdc_obd, &lmv_mdc_uuid,
 372                          &lmv->conn_data, NULL);
 373         if (rc) {
 374                 CERROR("target %s connect error %d\n", tgt->ltd_uuid.uuid, rc);
 375                 RETURN(rc);
 376         }
 377
 378         /*
 379          * Init fid sequence client for this mdc and add new fld target.
 380          */
 381         rc = obd_fid_init(mdc_obd, mdc_exp, LUSTRE_SEQ_METADATA);
 382         if (rc)
 383                 RETURN(rc);
 384
 385         target.ft_srv = NULL;
 386         target.ft_exp = mdc_exp;
 387         target.ft_idx = tgt->ltd_idx;
 388
 389         fld_client_add_target(&lmv->lmv_fld, &target);
 390
 391         rc = obd_register_observer(mdc_obd, obd);
 392         if (rc) {
 393                 obd_disconnect(mdc_exp);
 394                 CERROR("target %s register_observer error %d\n",
 395                        tgt->ltd_uuid.uuid, rc);
 396                 RETURN(rc);
 397         }
 398
 399         if (obd->obd_observer) {
 400                 /*
 401                  * Tell the observer about the new target.
 402                  */
 403                 rc = obd_notify(obd->obd_observer, mdc_exp->exp_obd,
 404                                 OBD_NOTIFY_ACTIVE,
 405                                 (void *)(tgt - lmv->tgts[0]));
 406                 if (rc) {
 407                         obd_disconnect(mdc_exp);
 408                         RETURN(rc);
 409                 }
 410         }
 411
 412         tgt->ltd_active = 1;
 413         tgt->ltd_exp = mdc_exp;
 414         lmv->desc.ld_active_tgt_count++;
 415
 416         md_init_ea_size(tgt->ltd_exp, lmv->max_easize,
 417                         lmv->max_def_easize, lmv->max_cookiesize);
 418
 419         CDEBUG(D_CONFIG, "Connected to %s(%s) successfully (%d)\n",
 420                 mdc_obd->obd_name, mdc_obd->obd_uuid.uuid,
 421                 cfs_atomic_read(&obd->obd_refcount));
 422
 423 #ifdef __KERNEL__
 424         lmv_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds");
 425         if (lmv_proc_dir) {
 426                 struct proc_dir_entry *mdc_symlink;
 427
 428                 LASSERT(mdc_obd->obd_type != NULL);
 429                 LASSERT(mdc_obd->obd_type->typ_name != NULL);
 430                 mdc_symlink = lprocfs_add_symlink(mdc_obd->obd_name,
 431                                                   lmv_proc_dir,
 432                                                   "../../../%s/%s",
 433                                                   mdc_obd->obd_type->typ_name,
 434                                                   mdc_obd->obd_name);
 435                 if (mdc_symlink == NULL) {
 436                         CERROR("Could not register LMV target "
 437                                "/proc/fs/lustre/%s/%s/target_obds/%s.",
 438                                obd->obd_type->typ_name, obd->obd_name,
 439                                mdc_obd->obd_name);
 440                         lprocfs_remove(&lmv_proc_dir);
 441                         lmv_proc_dir = NULL;
 442                 }
 443         }
 444 #endif
 445         RETURN(0);
 446 }
 447
 448 static void lmv_del_target(struct lmv_obd *lmv, int index)
 449 {
 450         if (lmv->tgts[index] == NULL)
 451                 return;
 452
 453         OBD_FREE_PTR(lmv->tgts[index]);
 454         lmv->tgts[index] = NULL;
 455         return;
 456 }
 457
 458 static int lmv_add_target(struct obd_device *obd, struct obd_uuid *uuidp,
 459                            __u32 index, int gen)
 460 {
 461         struct lmv_obd      *lmv = &obd->u.lmv;
 462         struct lmv_tgt_desc *tgt;
 463         int                  rc = 0;
 464         ENTRY;
 465
 466         CDEBUG(D_CONFIG, "Target uuid: %s. index %d\n", uuidp->uuid, index);
 467
 468         lmv_init_lock(lmv);
 469
 470         if (lmv->desc.ld_tgt_count == 0) {
 471                 struct obd_device *mdc_obd;
 472
 473                 mdc_obd = class_find_client_obd(uuidp, LUSTRE_MDC_NAME,
 474                                                 &obd->obd_uuid);
 475                 if (!mdc_obd) {
 476                         lmv_init_unlock(lmv);
 477                         CERROR("%s: Target %s not attached: rc = %d\n",
 478                                obd->obd_name, uuidp->uuid, -EINVAL);
 479                         RETURN(-EINVAL);
 480                 }
 481         }
 482
 483         if ((index < lmv->tgts_size) && (lmv->tgts[index] != NULL)) {
 484                 tgt = lmv->tgts[index];
 485                 CERROR("%s: UUID %s already assigned at LOV target index %d:"
 486                        " rc = %d\n", obd->obd_name,
 487                        obd_uuid2str(&tgt->ltd_uuid), index, -EEXIST);
 488                 lmv_init_unlock(lmv);
 489                 RETURN(-EEXIST);
 490         }
 491
 492         if (index >= lmv->tgts_size) {
 493                 /* We need to reallocate the lmv target array. */
 494                 struct lmv_tgt_desc **newtgts, **old = NULL;
 495                 __u32 newsize = 1;
 496                 __u32 oldsize = 0;
 497
 498                 while (newsize < index + 1)
 499                         newsize = newsize << 1;
 500                 OBD_ALLOC(newtgts, sizeof(*newtgts) * newsize);
 501                 if (newtgts == NULL) {
 502                         lmv_init_unlock(lmv);
 503                         RETURN(-ENOMEM);
 504                 }
 505
 506                 if (lmv->tgts_size) {
 507                         memcpy(newtgts, lmv->tgts,
 508                                sizeof(*newtgts) * lmv->tgts_size);
 509                         old = lmv->tgts;
 510                         oldsize = lmv->tgts_size;
 511                 }
 512
 513                 lmv->tgts = newtgts;
 514                 lmv->tgts_size = newsize;
 515                 smp_rmb();
 516                 if (old)
 517                         OBD_FREE(old, sizeof(*old) * oldsize);
 518
 519                 CDEBUG(D_CONFIG, "tgts: %p size: %d\n", lmv->tgts,
 520                        lmv->tgts_size);
 521         }
 522
 523         OBD_ALLOC_PTR(tgt);
 524         if (!tgt) {
 525                 lmv_init_unlock(lmv);
 526                 RETURN(-ENOMEM);
 527         }
 528
 529         mutex_init(&tgt->ltd_fid_mutex);
 530         tgt->ltd_idx = index;
 531         tgt->ltd_uuid = *uuidp;
 532         tgt->ltd_active = 0;
 533         lmv->tgts[index] = tgt;
 534         if (index >= lmv->desc.ld_tgt_count)
 535                 lmv->desc.ld_tgt_count = index + 1;
 536
 537         if (lmv->connected) {
 538                 rc = lmv_connect_mdc(obd, tgt);
 539                 if (rc) {
 540                         spin_lock(&lmv->lmv_lock);
 541                         lmv->desc.ld_tgt_count--;
 542                         memset(tgt, 0, sizeof(*tgt));
 543                         spin_unlock(&lmv->lmv_lock);
 544                 } else {
 545                         int easize = sizeof(struct lmv_stripe_md) +
 546                                      lmv->desc.ld_tgt_count *
 547                                      sizeof(struct lu_fid);
 548                         lmv_init_ea_size(obd->obd_self_export, easize, 0, 0);
 549                 }
 550         }
 551
 552         lmv_init_unlock(lmv);
 553         RETURN(rc);
 554 }
 555
 556 int lmv_check_connect(struct obd_device *obd)
 557 {
 558         struct lmv_obd       *lmv = &obd->u.lmv;
 559         struct lmv_tgt_desc  *tgt;
 560         int                   i;
 561         int                   rc;
 562         int                   easize;
 563         ENTRY;
 564
 565         if (lmv->connected)
 566                 RETURN(0);
 567
 568         lmv_init_lock(lmv);
 569         if (lmv->connected) {
 570                 lmv_init_unlock(lmv);
 571                 RETURN(0);
 572         }
 573
 574         if (lmv->desc.ld_tgt_count == 0) {
 575                 lmv_init_unlock(lmv);
 576                 CERROR("%s: no targets configured.\n", obd->obd_name);
 577                 RETURN(-EINVAL);
 578         }
 579
 580         CDEBUG(D_CONFIG, "Time to connect %s to %s\n",
 581                lmv->cluuid.uuid, obd->obd_name);
 582
 583         LASSERT(lmv->tgts != NULL);
 584
 585         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 586                 tgt = lmv->tgts[i];
 587                 if (tgt == NULL)
 588                         continue;
 589                 rc = lmv_connect_mdc(obd, tgt);
 590                 if (rc)
 591                         GOTO(out_disc, rc);
 592         }
 593
 594         lmv_set_timeouts(obd);
 595         class_export_put(lmv->exp);
 596         lmv->connected = 1;
 597         easize = lmv_get_easize(lmv);
 598         lmv_init_ea_size(obd->obd_self_export, easize, 0, 0);
 599         lmv_init_unlock(lmv);
 600         RETURN(0);
 601
 602  out_disc:
 603         while (i-- > 0) {
 604                 int rc2;
 605                 tgt = lmv->tgts[i];
 606                 if (tgt == NULL)
 607                         continue;
 608                 tgt->ltd_active = 0;
 609                 if (tgt->ltd_exp) {
 610                         --lmv->desc.ld_active_tgt_count;
 611                         rc2 = obd_disconnect(tgt->ltd_exp);
 612                         if (rc2) {
 613                                 CERROR("LMV target %s disconnect on "
 614                                        "MDC idx %d: error %d\n",
 615                                        tgt->ltd_uuid.uuid, i, rc2);
 616                         }
 617                 }
 618         }
 619         class_disconnect(lmv->exp);
 620         lmv_init_unlock(lmv);
 621         RETURN(rc);
 622 }
 623
 624 static int lmv_disconnect_mdc(struct obd_device *obd, struct lmv_tgt_desc *tgt)
 625 {
 626 #ifdef __KERNEL__
 627         struct proc_dir_entry  *lmv_proc_dir;
 628 #endif
 629         struct lmv_obd         *lmv = &obd->u.lmv;
 630         struct obd_device      *mdc_obd;
 631         int                     rc;
 632         ENTRY;
 633
 634         LASSERT(tgt != NULL);
 635         LASSERT(obd != NULL);
 636
 637         mdc_obd = class_exp2obd(tgt->ltd_exp);
 638
 639         if (mdc_obd) {
 640                 mdc_obd->obd_force = obd->obd_force;
 641                 mdc_obd->obd_fail = obd->obd_fail;
 642                 mdc_obd->obd_no_recov = obd->obd_no_recov;
 643         }
 644
 645 #ifdef __KERNEL__
 646         lmv_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds");
 647         if (lmv_proc_dir) {
 648                 struct proc_dir_entry *mdc_symlink;
 649
 650                 mdc_symlink = lprocfs_srch(lmv_proc_dir, mdc_obd->obd_name);
 651                 if (mdc_symlink) {
 652                         lprocfs_remove(&mdc_symlink);
 653                 } else {
 654                         CERROR("/proc/fs/lustre/%s/%s/target_obds/%s missing\n",
 655                                obd->obd_type->typ_name, obd->obd_name,
 656                                mdc_obd->obd_name);
 657                 }
 658         }
 659 #endif
 660         rc = obd_fid_fini(tgt->ltd_exp->exp_obd);
 661         if (rc)
 662                 CERROR("Can't finanize fids factory\n");
 663
 664         CDEBUG(D_INFO, "Disconnected from %s(%s) successfully\n",
 665                tgt->ltd_exp->exp_obd->obd_name,
 666                tgt->ltd_exp->exp_obd->obd_uuid.uuid);
 667
 668         obd_register_observer(tgt->ltd_exp->exp_obd, NULL);
 669         rc = obd_disconnect(tgt->ltd_exp);
 670         if (rc) {
 671                 if (tgt->ltd_active) {
 672                         CERROR("Target %s disconnect error %d\n",
 673                                tgt->ltd_uuid.uuid, rc);
 674                 }
 675         }
 676
 677         lmv_activate_target(lmv, tgt, 0);
 678         tgt->ltd_exp = NULL;
 679         RETURN(0);
 680 }
 681
 682 static int lmv_disconnect(struct obd_export *exp)
 683 {
 684         struct obd_device     *obd = class_exp2obd(exp);
 685 #ifdef __KERNEL__
 686         struct proc_dir_entry *lmv_proc_dir;
 687 #endif
 688         struct lmv_obd        *lmv = &obd->u.lmv;
 689         int                    rc;
 690         int                    i;
 691         ENTRY;
 692
 693         if (!lmv->tgts)
 694                 goto out_local;
 695
 696         /*
 697          * Only disconnect the underlying layers on the final disconnect.
 698          */
 699         lmv->refcount--;
 700         if (lmv->refcount != 0)
 701                 goto out_local;
 702
 703         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 704                 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
 705                         continue;
 706
 707                 lmv_disconnect_mdc(obd, lmv->tgts[i]);
 708         }
 709
 710 #ifdef __KERNEL__
 711         lmv_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds");
 712         if (lmv_proc_dir) {
 713                 lprocfs_remove(&lmv_proc_dir);
 714         } else {
 715                 CERROR("/proc/fs/lustre/%s/%s/target_obds missing\n",
 716                        obd->obd_type->typ_name, obd->obd_name);
 717         }
 718 #endif
 719
 720 out_local:
 721         /*
 722          * This is the case when no real connection is established by
 723          * lmv_check_connect().
 724          */
 725         if (!lmv->connected)
 726                 class_export_put(exp);
 727         rc = class_disconnect(exp);
 728         if (lmv->refcount == 0)
 729                 lmv->connected = 0;
 730         RETURN(rc);
 731 }
 732
 733 static int lmv_fid2path(struct obd_export *exp, int len, void *karg, void *uarg)
 734 {
 735         struct obd_device       *obddev = class_exp2obd(exp);
 736         struct lmv_obd          *lmv = &obddev->u.lmv;
 737         struct getinfo_fid2path *gf;
 738         struct lmv_tgt_desc     *tgt;
 739         struct getinfo_fid2path *remote_gf = NULL;
 740         int                     remote_gf_size = 0;
 741         int                     rc;
 742
 743         gf = (struct getinfo_fid2path *)karg;
 744         tgt = lmv_find_target(lmv, &gf->gf_fid);
 745         if (IS_ERR(tgt))
 746                 RETURN(PTR_ERR(tgt));
 747
 748 repeat_fid2path:
 749         rc = obd_iocontrol(OBD_IOC_FID2PATH, tgt->ltd_exp, len, gf, uarg);
 750         if (rc != 0 && rc != -EREMOTE)
 751                 GOTO(out_fid2path, rc);
 752
 753         /* If remote_gf != NULL, it means just building the
 754          * path on the remote MDT, copy this path segement to gf */
 755         if (remote_gf != NULL) {
 756                 struct getinfo_fid2path *ori_gf;
 757                 char *ptr;
 758
 759                 ori_gf = (struct getinfo_fid2path *)karg;
 760                 if (strlen(ori_gf->gf_path) +
 761                     strlen(gf->gf_path) > ori_gf->gf_pathlen)
 762                         GOTO(out_fid2path, rc = -EOVERFLOW);
 763
 764                 ptr = ori_gf->gf_path;
 765
 766                 memmove(ptr + strlen(gf->gf_path) + 1, ptr,
 767                         strlen(ori_gf->gf_path));
 768
 769                 strncpy(ptr, gf->gf_path, strlen(gf->gf_path));
 770                 ptr += strlen(gf->gf_path);
 771                 *ptr = '/';
 772         }
 773
 774         CDEBUG(D_INFO, "%s: get path %s "DFID" rec: "LPU64" ln: %u\n",
 775                tgt->ltd_exp->exp_obd->obd_name,
 776                gf->gf_path, PFID(&gf->gf_fid), gf->gf_recno,
 777                gf->gf_linkno);
 778
 779         if (rc == 0)
 780                 GOTO(out_fid2path, rc);
 781
 782         /* sigh, has to go to another MDT to do path building further */
 783         if (remote_gf == NULL) {
 784                 remote_gf_size = sizeof(*remote_gf) + PATH_MAX;
 785                 OBD_ALLOC(remote_gf, remote_gf_size);
 786                 if (remote_gf == NULL)
 787                         GOTO(out_fid2path, rc = -ENOMEM);
 788                 remote_gf->gf_pathlen = PATH_MAX;
 789         }
 790
 791         if (!fid_is_sane(&gf->gf_fid)) {
 792                 CERROR("%s: invalid FID "DFID": rc = %d\n",
 793                        tgt->ltd_exp->exp_obd->obd_name,
 794                        PFID(&gf->gf_fid), -EINVAL);
 795                 GOTO(out_fid2path, rc = -EINVAL);
 796         }
 797
 798         tgt = lmv_find_target(lmv, &gf->gf_fid);
 799         if (IS_ERR(tgt))
 800                 GOTO(out_fid2path, rc = -EINVAL);
 801
 802         remote_gf->gf_fid = gf->gf_fid;
 803         remote_gf->gf_recno = -1;
 804         remote_gf->gf_linkno = -1;
 805         memset(remote_gf->gf_path, 0, remote_gf->gf_pathlen);
 806         gf = remote_gf;
 807         goto repeat_fid2path;
 808
 809 out_fid2path:
 810         if (remote_gf != NULL)
 811                 OBD_FREE(remote_gf, remote_gf_size);
 812         RETURN(rc);
 813 }
 814
 815 static int lmv_iocontrol(unsigned int cmd, struct obd_export *exp,
 816                          int len, void *karg, void *uarg)
 817 {
 818         struct obd_device    *obddev = class_exp2obd(exp);
 819         struct lmv_obd       *lmv = &obddev->u.lmv;
 820         int                   i = 0;
 821         int                   rc = 0;
 822         int                   set = 0;
 823         int                   count = lmv->desc.ld_tgt_count;
 824         ENTRY;
 825
 826         if (count == 0)
 827                 RETURN(-ENOTTY);
 828
 829         switch (cmd) {
 830         case IOC_OBD_STATFS: {
 831                 struct obd_ioctl_data *data = karg;
 832                 struct obd_device *mdc_obd;
 833                 struct obd_statfs stat_buf = {0};
 834                 __u32 index;
 835
 836                 memcpy(&index, data->ioc_inlbuf2, sizeof(__u32));
 837                 if ((index >= count))
 838                         RETURN(-ENODEV);
 839
 840                 if (lmv->tgts[index] == NULL ||
 841                     lmv->tgts[index]->ltd_active == 0)
 842                         RETURN(-ENODATA);
 843
 844                 mdc_obd = class_exp2obd(lmv->tgts[index]->ltd_exp);
 845                 if (!mdc_obd)
 846                         RETURN(-EINVAL);
 847
 848                 /* copy UUID */
 849                 if (cfs_copy_to_user(data->ioc_pbuf2, obd2cli_tgt(mdc_obd),
 850                                      min((int) data->ioc_plen2,
 851                                          (int) sizeof(struct obd_uuid))))
 852                         RETURN(-EFAULT);
 853
 854                 rc = obd_statfs(NULL, lmv->tgts[index]->ltd_exp, &stat_buf,
 855                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
 856                                 0);
 857                 if (rc)
 858                         RETURN(rc);
 859                 if (cfs_copy_to_user(data->ioc_pbuf1, &stat_buf,
 860                                      min((int) data->ioc_plen1,
 861                                          (int) sizeof(stat_buf))))
 862                         RETURN(-EFAULT);
 863                 break;
 864         }
 865         case OBD_IOC_QUOTACTL: {
 866                 struct if_quotactl *qctl = karg;
 867                 struct lmv_tgt_desc *tgt = NULL;
 868                 struct obd_quotactl *oqctl;
 869
 870                 if (qctl->qc_valid == QC_MDTIDX) {
 871                         if (qctl->qc_idx < 0 || count <= qctl->qc_idx)
 872                                 RETURN(-EINVAL);
 873
 874                         tgt = lmv->tgts[qctl->qc_idx];
 875                         if (tgt == NULL || tgt->ltd_exp == NULL)
 876                                 RETURN(-EINVAL);
 877                 } else if (qctl->qc_valid == QC_UUID) {
 878                         for (i = 0; i < count; i++) {
 879                                 tgt = lmv->tgts[i];
 880                                 if (tgt == NULL)
 881                                         continue;
 882                                 if (!obd_uuid_equals(&tgt->ltd_uuid,
 883                                                      &qctl->obd_uuid))
 884                                         continue;
 885
 886                                 if (tgt->ltd_exp == NULL)
 887                                         RETURN(-EINVAL);
 888
 889                                 break;
 890                         }
 891                 } else {
 892                         RETURN(-EINVAL);
 893                 }
 894
 895                 if (i >= count)
 896                         RETURN(-EAGAIN);
 897
 898                 LASSERT(tgt && tgt->ltd_exp);
 899                 OBD_ALLOC_PTR(oqctl);
 900                 if (!oqctl)
 901                         RETURN(-ENOMEM);
 902
 903                 QCTL_COPY(oqctl, qctl);
 904                 rc = obd_quotactl(tgt->ltd_exp, oqctl);
 905                 if (rc == 0) {
 906                         QCTL_COPY(qctl, oqctl);
 907                         qctl->qc_valid = QC_MDTIDX;
 908                         qctl->obd_uuid = tgt->ltd_uuid;
 909                 }
 910                 OBD_FREE_PTR(oqctl);
 911                 break;
 912         }
 913         case OBD_IOC_CHANGELOG_SEND:
 914         case OBD_IOC_CHANGELOG_CLEAR: {
 915                 struct ioc_changelog *icc = karg;
 916
 917                 if (icc->icc_mdtindex >= count)
 918                         RETURN(-ENODEV);
 919
 920                 if (lmv->tgts[icc->icc_mdtindex] == NULL ||
 921                     lmv->tgts[icc->icc_mdtindex]->ltd_exp == NULL ||
 922                     lmv->tgts[icc->icc_mdtindex]->ltd_active == 0)
 923                         RETURN(-ENODEV);
 924                 rc = obd_iocontrol(cmd, lmv->tgts[icc->icc_mdtindex]->ltd_exp,
 925                                    sizeof(*icc), icc, NULL);
 926                 break;
 927         }
 928         case LL_IOC_GET_CONNECT_FLAGS: {
 929                 if (lmv->tgts[0] == NULL)
 930                         RETURN(-ENODATA);
 931                 rc = obd_iocontrol(cmd, lmv->tgts[0]->ltd_exp, len, karg, uarg);
 932                 break;
 933         }
 934         case OBD_IOC_FID2PATH: {
 935                 rc = lmv_fid2path(exp, len, karg, uarg);
 936                 break;
 937         }
 938         case LL_IOC_HSM_STATE_GET:
 939         case LL_IOC_HSM_STATE_SET:
 940         case LL_IOC_HSM_ACTION:
 941         case LL_IOC_LOV_SWAP_LAYOUTS: {
 942                 struct md_op_data       *op_data = karg;
 943                 struct lmv_tgt_desc     *tgt;
 944
 945                 tgt = lmv_find_target(lmv, &op_data->op_fid1);
 946                 if (!tgt->ltd_exp)
 947                         RETURN(-EINVAL);
 948
 949                 rc = obd_iocontrol(cmd, lmv->tgts[0]->ltd_exp, len, karg, uarg);
 950                 break;
 951         }
 952         default:
 953                 for (i = 0; i < count; i++) {
 954                         struct obd_device *mdc_obd;
 955                         int err;
 956
 957                         if (lmv->tgts[i] == NULL ||
 958                             lmv->tgts[i]->ltd_exp == NULL)
 959                                 continue;
 960                         /* ll_umount_begin() sets force flag but for lmv, not
 961                          * mdc. Let's pass it through */
 962                         mdc_obd = class_exp2obd(lmv->tgts[i]->ltd_exp);
 963                         mdc_obd->obd_force = obddev->obd_force;
 964                         err = obd_iocontrol(cmd, lmv->tgts[i]->ltd_exp, len,
 965                                             karg, uarg);
 966                         if (err == -ENODATA && cmd == OBD_IOC_POLL_QUOTACHECK) {
 967                                 RETURN(err);
 968                         } else if (err) {
 969                                 if (lmv->tgts[i]->ltd_active) {
 970                                         CERROR("error: iocontrol MDC %s on MDT"
 971                                                "idx %d cmd %x: err = %d\n",
 972                                                 lmv->tgts[i]->ltd_uuid.uuid,
 973                                                 i, cmd, err);
 974                                         if (!rc)
 975                                                 rc = err;
 976                                 }
 977                         } else
 978                                 set = 1;
 979                 }
 980                 if (!set && !rc)
 981                         rc = -EIO;
 982         }
 983         RETURN(rc);
 984 }
 985
 986 #if 0
 987 static int lmv_all_chars_policy(int count, const char *name,
 988                                 int len)
 989 {
 990         unsigned int c = 0;
 991
 992         while (len > 0)
 993                 c += name[--len];
 994         c = c % count;
 995         return c;
 996 }
 997
 998 static int lmv_nid_policy(struct lmv_obd *lmv)
 999 {
1000         struct obd_import *imp;
1001         __u32              id;
1002
1003         /*
1004          * XXX: To get nid we assume that underlying obd device is mdc.
1005          */
1006         imp = class_exp2cliimp(lmv->tgts[0].ltd_exp);
1007         id = imp->imp_connection->c_self ^ (imp->imp_connection->c_self >> 32);
1008         return id % lmv->desc.ld_tgt_count;
1009 }
1010
1011 static int lmv_choose_mds(struct lmv_obd *lmv, struct md_op_data *op_data,
1012                           placement_policy_t placement)
1013 {
1014         switch (placement) {
1015         case PLACEMENT_CHAR_POLICY:
1016                 return lmv_all_chars_policy(lmv->desc.ld_tgt_count,
1017                                             op_data->op_name,
1018                                             op_data->op_namelen);
1019         case PLACEMENT_NID_POLICY:
1020                 return lmv_nid_policy(lmv);
1021
1022         default:
1023                 break;
1024         }
1025
1026         CERROR("Unsupported placement policy %x\n", placement);
1027         return -EINVAL;
1028 }
1029 #endif
1030
1031 /**
1032  * This is _inode_ placement policy function (not name).
1033  */
1034 static int lmv_placement_policy(struct obd_device *obd,
1035                                 struct md_op_data *op_data,
1036                                 mdsno_t *mds)
1037 {
1038         struct lmv_obd          *lmv = &obd->u.lmv;
1039         ENTRY;
1040
1041         LASSERT(mds != NULL);
1042
1043         if (lmv->desc.ld_tgt_count == 1) {
1044                 *mds = 0;
1045                 RETURN(0);
1046         }
1047
1048         /**
1049          * If stripe_offset is provided during setdirstripe
1050          * (setdirstripe -i xx), xx MDS will be choosen.
1051          */
1052         if (op_data->op_cli_flags & CLI_SET_MEA) {
1053                 struct lmv_user_md *lum;
1054
1055                 lum = (struct lmv_user_md *)op_data->op_data;
1056                 if (lum->lum_type == LMV_STRIPE_TYPE &&
1057                     lum->lum_stripe_offset != -1) {
1058                         if (lum->lum_stripe_offset >= lmv->desc.ld_tgt_count) {
1059                                 CERROR("%s: Stripe_offset %d > MDT count %d:"
1060                                        " rc = %d\n", obd->obd_name,
1061                                        lum->lum_stripe_offset,
1062                                        lmv->desc.ld_tgt_count, -ERANGE);
1063                                 RETURN(-ERANGE);
1064                         }
1065                         *mds = lum->lum_stripe_offset;
1066                         RETURN(0);
1067                 }
1068         }
1069
1070         /* Allocate new fid on target according to operation type and parent
1071          * home mds. */
1072         *mds = op_data->op_mds;
1073         RETURN(0);
1074 }
1075
1076 int __lmv_fid_alloc(struct lmv_obd *lmv, struct lu_fid *fid,
1077                     mdsno_t mds)
1078 {
1079         struct lmv_tgt_desc *tgt;
1080         int                  rc;
1081         ENTRY;
1082
1083         tgt = lmv_get_target(lmv, mds);
1084
1085         /*
1086          * New seq alloc and FLD setup should be atomic. Otherwise we may find
1087          * on server that seq in new allocated fid is not yet known.
1088          */
1089         mutex_lock(&tgt->ltd_fid_mutex);
1090
1091         if (tgt == NULL || tgt->ltd_active == 0 || tgt->ltd_exp == NULL)
1092                 GOTO(out, rc = -ENODEV);
1093
1094         /*
1095          * Asking underlaying tgt layer to allocate new fid.
1096          */
1097         rc = obd_fid_alloc(tgt->ltd_exp, fid, NULL);
1098         if (rc > 0) {
1099                 LASSERT(fid_is_sane(fid));
1100                 rc = 0;
1101         }
1102
1103         EXIT;
1104 out:
1105         mutex_unlock(&tgt->ltd_fid_mutex);
1106         return rc;
1107 }
1108
1109 int lmv_fid_alloc(struct obd_export *exp, struct lu_fid *fid,
1110                   struct md_op_data *op_data)
1111 {
1112         struct obd_device     *obd = class_exp2obd(exp);
1113         struct lmv_obd        *lmv = &obd->u.lmv;
1114         mdsno_t                mds = 0;
1115         int                    rc;
1116         ENTRY;
1117
1118         LASSERT(op_data != NULL);
1119         LASSERT(fid != NULL);
1120
1121         rc = lmv_placement_policy(obd, op_data, &mds);
1122         if (rc) {
1123                 CERROR("Can't get target for allocating fid, "
1124                        "rc %d\n", rc);
1125                 RETURN(rc);
1126         }
1127
1128         rc = __lmv_fid_alloc(lmv, fid, mds);
1129         if (rc) {
1130                 CERROR("Can't alloc new fid, rc %d\n", rc);
1131                 RETURN(rc);
1132         }
1133
1134         RETURN(rc);
1135 }
1136
1137 static int lmv_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
1138 {
1139         struct lmv_obd             *lmv = &obd->u.lmv;
1140         struct lprocfs_static_vars  lvars;
1141         struct lmv_desc            *desc;
1142         int                         rc;
1143         ENTRY;
1144
1145         if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
1146                 CERROR("LMV setup requires a descriptor\n");
1147                 RETURN(-EINVAL);
1148         }
1149
1150         desc = (struct lmv_desc *)lustre_cfg_buf(lcfg, 1);
1151         if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
1152                 CERROR("Lmv descriptor size wrong: %d > %d\n",
1153                        (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
1154                 RETURN(-EINVAL);
1155         }
1156
1157         OBD_ALLOC(lmv->tgts, sizeof(*lmv->tgts) * 32);
1158         if (lmv->tgts == NULL)
1159                 RETURN(-ENOMEM);
1160         lmv->tgts_size = 32;
1161
1162         obd_str2uuid(&lmv->desc.ld_uuid, desc->ld_uuid.uuid);
1163         lmv->desc.ld_tgt_count = 0;
1164         lmv->desc.ld_active_tgt_count = 0;
1165         lmv->max_cookiesize = 0;
1166         lmv->max_def_easize = 0;
1167         lmv->max_easize = 0;
1168         lmv->lmv_placement = PLACEMENT_CHAR_POLICY;
1169
1170         spin_lock_init(&lmv->lmv_lock);
1171         mutex_init(&lmv->init_mutex);
1172
1173         lprocfs_lmv_init_vars(&lvars);
1174
1175         lprocfs_obd_setup(obd, lvars.obd_vars);
1176 #ifdef LPROCFS
1177         {
1178                 rc = lprocfs_seq_create(obd->obd_proc_entry, "target_obd",
1179                                         0444, &lmv_proc_target_fops, obd);
1180                 if (rc)
1181                         CWARN("%s: error adding LMV target_obd file: rc = %d\n",
1182                                obd->obd_name, rc);
1183        }
1184 #endif
1185         rc = fld_client_init(&lmv->lmv_fld, obd->obd_name,
1186                              LUSTRE_CLI_FLD_HASH_DHT);
1187         if (rc) {
1188                 CERROR("Can't init FLD, err %d\n", rc);
1189                 GOTO(out, rc);
1190         }
1191
1192         RETURN(0);
1193
1194 out:
1195         return rc;
1196 }
1197
1198 static int lmv_cleanup(struct obd_device *obd)
1199 {
1200         struct lmv_obd   *lmv = &obd->u.lmv;
1201         ENTRY;
1202
1203         fld_client_fini(&lmv->lmv_fld);
1204         if (lmv->tgts != NULL) {
1205                 int i;
1206                 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1207                         if (lmv->tgts[i] == NULL)
1208                                 continue;
1209                         lmv_del_target(lmv, i);
1210                 }
1211                 OBD_FREE(lmv->tgts, sizeof(*lmv->tgts) * lmv->tgts_size);
1212                 lmv->tgts_size = 0;
1213         }
1214         RETURN(0);
1215 }
1216
1217 static int lmv_process_config(struct obd_device *obd, obd_count len, void *buf)
1218 {
1219         struct lustre_cfg       *lcfg = buf;
1220         struct obd_uuid         obd_uuid;
1221         int                     gen;
1222         __u32                   index;
1223         int                     rc;
1224         ENTRY;
1225
1226         switch (lcfg->lcfg_command) {
1227         case LCFG_ADD_MDC:
1228                 /* modify_mdc_tgts add 0:lustre-clilmv  1:lustre-MDT0000_UUID
1229                  * 2:0  3:1  4:lustre-MDT0000-mdc_UUID */
1230                 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(obd_uuid.uuid))
1231                         GOTO(out, rc = -EINVAL);
1232
1233                 obd_str2uuid(&obd_uuid,  lustre_cfg_buf(lcfg, 1));
1234
1235                 if (sscanf(lustre_cfg_buf(lcfg, 2), "%d", &index) != 1)
1236                         GOTO(out, rc = -EINVAL);
1237                 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", &gen) != 1)
1238                         GOTO(out, rc = -EINVAL);
1239                 rc = lmv_add_target(obd, &obd_uuid, index, gen);
1240                 GOTO(out, rc);
1241         default:
1242                 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
1243                 GOTO(out, rc = -EINVAL);
1244         }
1245 out:
1246         RETURN(rc);
1247 }
1248
1249 static int lmv_statfs(const struct lu_env *env, struct obd_export *exp,
1250                       struct obd_statfs *osfs, __u64 max_age, __u32 flags)
1251 {
1252         struct obd_device     *obd = class_exp2obd(exp);
1253         struct lmv_obd        *lmv = &obd->u.lmv;
1254         struct obd_statfs     *temp;
1255         int                    rc = 0;
1256         int                    i;
1257         ENTRY;
1258
1259         rc = lmv_check_connect(obd);
1260         if (rc)
1261                 RETURN(rc);
1262
1263         OBD_ALLOC(temp, sizeof(*temp));
1264         if (temp == NULL)
1265                 RETURN(-ENOMEM);
1266
1267         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1268                 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
1269                         continue;
1270
1271                 rc = obd_statfs(env, lmv->tgts[i]->ltd_exp, temp,
1272                                 max_age, flags);
1273                 if (rc) {
1274                         CERROR("can't stat MDS #%d (%s), error %d\n", i,
1275                                lmv->tgts[i]->ltd_exp->exp_obd->obd_name,
1276                                rc);
1277                         GOTO(out_free_temp, rc);
1278                 }
1279
1280                 if (i == 0) {
1281                         *osfs = *temp;
1282                         /* If the statfs is from mount, it will needs
1283                          * retrieve necessary information from MDT0.
1284                          * i.e. mount does not need the merged osfs
1285                          * from all of MDT.
1286                          * And also clients can be mounted as long as
1287                          * MDT0 is in service*/
1288                         if (flags & OBD_STATFS_FOR_MDT0)
1289                                 GOTO(out_free_temp, rc);
1290                 } else {
1291                         osfs->os_bavail += temp->os_bavail;
1292                         osfs->os_blocks += temp->os_blocks;
1293                         osfs->os_ffree += temp->os_ffree;
1294                         osfs->os_files += temp->os_files;
1295                 }
1296         }
1297
1298         EXIT;
1299 out_free_temp:
1300         OBD_FREE(temp, sizeof(*temp));
1301         return rc;
1302 }
1303
1304 static int lmv_getstatus(struct obd_export *exp,
1305                          struct lu_fid *fid,
1306                          struct obd_capa **pc)
1307 {
1308         struct obd_device    *obd = exp->exp_obd;
1309         struct lmv_obd       *lmv = &obd->u.lmv;
1310         int                   rc;
1311         ENTRY;
1312
1313         rc = lmv_check_connect(obd);
1314         if (rc)
1315                 RETURN(rc);
1316
1317         rc = md_getstatus(lmv->tgts[0]->ltd_exp, fid, pc);
1318         RETURN(rc);
1319 }
1320
1321 static int lmv_getxattr(struct obd_export *exp, const struct lu_fid *fid,
1322                         struct obd_capa *oc, obd_valid valid, const char *name,
1323                         const char *input, int input_size, int output_size,
1324                         int flags, struct ptlrpc_request **request)
1325 {
1326         struct obd_device      *obd = exp->exp_obd;
1327         struct lmv_obd         *lmv = &obd->u.lmv;
1328         struct lmv_tgt_desc    *tgt;
1329         int                     rc;
1330         ENTRY;
1331
1332         rc = lmv_check_connect(obd);
1333         if (rc)
1334                 RETURN(rc);
1335
1336         tgt = lmv_find_target(lmv, fid);
1337         if (IS_ERR(tgt))
1338                 RETURN(PTR_ERR(tgt));
1339
1340         rc = md_getxattr(tgt->ltd_exp, fid, oc, valid, name, input,
1341                          input_size, output_size, flags, request);
1342
1343         RETURN(rc);
1344 }
1345
1346 static int lmv_setxattr(struct obd_export *exp, const struct lu_fid *fid,
1347                         struct obd_capa *oc, obd_valid valid, const char *name,
1348                         const char *input, int input_size, int output_size,
1349                         int flags, __u32 suppgid,
1350                         struct ptlrpc_request **request)
1351 {
1352         struct obd_device      *obd = exp->exp_obd;
1353         struct lmv_obd         *lmv = &obd->u.lmv;
1354         struct lmv_tgt_desc    *tgt;
1355         int                     rc;
1356         ENTRY;
1357
1358         rc = lmv_check_connect(obd);
1359         if (rc)
1360                 RETURN(rc);
1361
1362         tgt = lmv_find_target(lmv, fid);
1363         if (IS_ERR(tgt))
1364                 RETURN(PTR_ERR(tgt));
1365
1366         rc = md_setxattr(tgt->ltd_exp, fid, oc, valid, name, input,
1367                          input_size, output_size, flags, suppgid,
1368                          request);
1369
1370         RETURN(rc);
1371 }
1372
1373 static int lmv_getattr(struct obd_export *exp, struct md_op_data *op_data,
1374                        struct ptlrpc_request **request)
1375 {
1376         struct obd_device       *obd = exp->exp_obd;
1377         struct lmv_obd          *lmv = &obd->u.lmv;
1378         struct lmv_tgt_desc     *tgt;
1379         int                      rc;
1380         ENTRY;
1381
1382         rc = lmv_check_connect(obd);
1383         if (rc)
1384                 RETURN(rc);
1385
1386         tgt = lmv_find_target(lmv, &op_data->op_fid1);
1387         if (IS_ERR(tgt))
1388                 RETURN(PTR_ERR(tgt));
1389
1390         if (op_data->op_flags & MF_GET_MDT_IDX) {
1391                 op_data->op_mds = tgt->ltd_idx;
1392                 RETURN(0);
1393         }
1394
1395         rc = md_getattr(tgt->ltd_exp, op_data, request);
1396
1397         RETURN(rc);
1398 }
1399
1400 static int lmv_null_inode(struct obd_export *exp, const struct lu_fid *fid)
1401 {
1402         struct obd_device   *obd = exp->exp_obd;
1403         struct lmv_obd      *lmv = &obd->u.lmv;
1404         int                  i;
1405         int                  rc;
1406         ENTRY;
1407
1408         rc = lmv_check_connect(obd);
1409         if (rc)
1410                 RETURN(rc);
1411
1412         CDEBUG(D_INODE, "CBDATA for "DFID"\n", PFID(fid));
1413
1414         /*
1415          * With DNE every object can have two locks in different namespaces:
1416          * lookup lock in space of MDT storing direntry and update/open lock in
1417          * space of MDT storing inode.
1418          */
1419         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1420                 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
1421                         continue;
1422                 md_null_inode(lmv->tgts[i]->ltd_exp, fid);
1423         }
1424
1425         RETURN(0);
1426 }
1427
1428 static int lmv_find_cbdata(struct obd_export *exp, const struct lu_fid *fid,
1429                            ldlm_iterator_t it, void *data)
1430 {
1431         struct obd_device   *obd = exp->exp_obd;
1432         struct lmv_obd      *lmv = &obd->u.lmv;
1433         int                  i;
1434         int                  rc;
1435         ENTRY;
1436
1437         rc = lmv_check_connect(obd);
1438         if (rc)
1439                 RETURN(rc);
1440
1441         CDEBUG(D_INODE, "CBDATA for "DFID"\n", PFID(fid));
1442
1443         /*
1444          * With DNE every object can have two locks in different namespaces:
1445          * lookup lock in space of MDT storing direntry and update/open lock in
1446          * space of MDT storing inode.
1447          */
1448         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1449                 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
1450                         continue;
1451                 rc = md_find_cbdata(lmv->tgts[i]->ltd_exp, fid, it, data);
1452                 if (rc)
1453                         RETURN(rc);
1454         }
1455
1456         RETURN(rc);
1457 }
1458
1459
1460 static int lmv_close(struct obd_export *exp, struct md_op_data *op_data,
1461                      struct md_open_data *mod, struct ptlrpc_request **request)
1462 {
1463         struct obd_device     *obd = exp->exp_obd;
1464         struct lmv_obd        *lmv = &obd->u.lmv;
1465         struct lmv_tgt_desc   *tgt;
1466         int                    rc;
1467         ENTRY;
1468
1469         rc = lmv_check_connect(obd);
1470         if (rc)
1471                 RETURN(rc);
1472
1473         tgt = lmv_find_target(lmv, &op_data->op_fid1);
1474         if (IS_ERR(tgt))
1475                 RETURN(PTR_ERR(tgt));
1476
1477         CDEBUG(D_INODE, "CLOSE "DFID"\n", PFID(&op_data->op_fid1));
1478         rc = md_close(tgt->ltd_exp, op_data, mod, request);
1479         RETURN(rc);
1480 }
1481
1482 struct lmv_tgt_desc
1483 *lmv_locate_mds(struct lmv_obd *lmv, struct md_op_data *op_data,
1484                 struct lu_fid *fid)
1485 {
1486         struct lmv_tgt_desc *tgt;
1487
1488         tgt = lmv_find_target(lmv, fid);
1489         op_data->op_mds = tgt->ltd_idx;
1490
1491         return tgt;
1492 }
1493
1494 int lmv_create(struct obd_export *exp, struct md_op_data *op_data,
1495                const void *data, int datalen, int mode, __u32 uid,
1496                __u32 gid, cfs_cap_t cap_effective, __u64 rdev,
1497                struct ptlrpc_request **request)
1498 {
1499         struct obd_device       *obd = exp->exp_obd;
1500         struct lmv_obd          *lmv = &obd->u.lmv;
1501         struct lmv_tgt_desc     *tgt;
1502         int                      rc;
1503         ENTRY;
1504
1505         rc = lmv_check_connect(obd);
1506         if (rc)
1507                 RETURN(rc);
1508
1509         if (!lmv->desc.ld_active_tgt_count)
1510                 RETURN(-EIO);
1511
1512         tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
1513         if (IS_ERR(tgt))
1514                 RETURN(PTR_ERR(tgt));
1515
1516         rc = lmv_fid_alloc(exp, &op_data->op_fid2, op_data);
1517         if (rc)
1518                 RETURN(rc);
1519
1520         CDEBUG(D_INODE, "CREATE '%*s' on "DFID" -> mds #%x\n",
1521                op_data->op_namelen, op_data->op_name, PFID(&op_data->op_fid1),
1522                op_data->op_mds);
1523
1524         op_data->op_flags |= MF_MDC_CANCEL_FID1;
1525         rc = md_create(tgt->ltd_exp, op_data, data, datalen, mode, uid, gid,
1526                        cap_effective, rdev, request);
1527
1528         if (rc == 0) {
1529                 if (*request == NULL)
1530                         RETURN(rc);
1531                 CDEBUG(D_INODE, "Created - "DFID"\n", PFID(&op_data->op_fid2));
1532         }
1533         RETURN(rc);
1534 }
1535
1536 static int lmv_done_writing(struct obd_export *exp,
1537                             struct md_op_data *op_data,
1538                             struct md_open_data *mod)
1539 {
1540         struct obd_device     *obd = exp->exp_obd;
1541         struct lmv_obd        *lmv = &obd->u.lmv;
1542         struct lmv_tgt_desc   *tgt;
1543         int                    rc;
1544         ENTRY;
1545
1546         rc = lmv_check_connect(obd);
1547         if (rc)
1548                 RETURN(rc);
1549
1550         tgt = lmv_find_target(lmv, &op_data->op_fid1);
1551         if (IS_ERR(tgt))
1552                 RETURN(PTR_ERR(tgt));
1553
1554         rc = md_done_writing(tgt->ltd_exp, op_data, mod);
1555         RETURN(rc);
1556 }
1557
1558 static int
1559 lmv_enqueue_remote(struct obd_export *exp, struct ldlm_enqueue_info *einfo,
1560                    struct lookup_intent *it, struct md_op_data *op_data,
1561                    struct lustre_handle *lockh, void *lmm, int lmmsize,
1562                    int extra_lock_flags)
1563 {
1564         struct ptlrpc_request      *req = it->d.lustre.it_data;
1565         struct obd_device          *obd = exp->exp_obd;
1566         struct lmv_obd             *lmv = &obd->u.lmv;
1567         struct lustre_handle        plock;
1568         struct lmv_tgt_desc        *tgt;
1569         struct md_op_data          *rdata;
1570         struct lu_fid               fid1;
1571         struct mdt_body            *body;
1572         int                         rc = 0;
1573         int                         pmode;
1574         ENTRY;
1575
1576         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1577         LASSERT(body != NULL);
1578
1579         if (!(body->valid & OBD_MD_MDS))
1580                 RETURN(0);
1581
1582         CDEBUG(D_INODE, "REMOTE_ENQUEUE '%s' on "DFID" -> "DFID"\n",
1583                LL_IT2STR(it), PFID(&op_data->op_fid1), PFID(&body->fid1));
1584
1585         /*
1586          * We got LOOKUP lock, but we really need attrs.
1587          */
1588         pmode = it->d.lustre.it_lock_mode;
1589         LASSERT(pmode != 0);
1590         memcpy(&plock, lockh, sizeof(plock));
1591         it->d.lustre.it_lock_mode = 0;
1592         it->d.lustre.it_data = NULL;
1593         fid1 = body->fid1;
1594
1595         it->d.lustre.it_disposition &= ~DISP_ENQ_COMPLETE;
1596         ptlrpc_req_finished(req);
1597
1598         tgt = lmv_find_target(lmv, &fid1);
1599         if (IS_ERR(tgt))
1600                 GOTO(out, rc = PTR_ERR(tgt));
1601
1602         OBD_ALLOC_PTR(rdata);
1603         if (rdata == NULL)
1604                 GOTO(out, rc = -ENOMEM);
1605
1606         rdata->op_fid1 = fid1;
1607         rdata->op_bias = MDS_CROSS_REF;
1608
1609         rc = md_enqueue(tgt->ltd_exp, einfo, it, rdata, lockh,
1610                         lmm, lmmsize, NULL, extra_lock_flags);
1611         OBD_FREE_PTR(rdata);
1612         EXIT;
1613 out:
1614         ldlm_lock_decref(&plock, pmode);
1615         return rc;
1616 }
1617
1618 static int
1619 lmv_enqueue(struct obd_export *exp, struct ldlm_enqueue_info *einfo,
1620             struct lookup_intent *it, struct md_op_data *op_data,
1621             struct lustre_handle *lockh, void *lmm, int lmmsize,
1622             struct ptlrpc_request **req, __u64 extra_lock_flags)
1623 {
1624         struct obd_device        *obd = exp->exp_obd;
1625         struct lmv_obd           *lmv = &obd->u.lmv;
1626         struct lmv_tgt_desc      *tgt;
1627         int                       rc;
1628         ENTRY;
1629
1630         rc = lmv_check_connect(obd);
1631         if (rc)
1632                 RETURN(rc);
1633
1634         CDEBUG(D_INODE, "ENQUEUE '%s' on "DFID"\n",
1635                LL_IT2STR(it), PFID(&op_data->op_fid1));
1636
1637         tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
1638         if (IS_ERR(tgt))
1639                 RETURN(PTR_ERR(tgt));
1640
1641         CDEBUG(D_INODE, "ENQUEUE '%s' on "DFID" -> mds #%d\n",
1642                LL_IT2STR(it), PFID(&op_data->op_fid1), tgt->ltd_idx);
1643
1644         rc = md_enqueue(tgt->ltd_exp, einfo, it, op_data, lockh,
1645                         lmm, lmmsize, req, extra_lock_flags);
1646
1647         if (rc == 0 && it && it->it_op == IT_OPEN) {
1648                 rc = lmv_enqueue_remote(exp, einfo, it, op_data, lockh,
1649                                         lmm, lmmsize, extra_lock_flags);
1650         }
1651         RETURN(rc);
1652 }
1653
1654 static int
1655 lmv_getattr_name(struct obd_export *exp,struct md_op_data *op_data,
1656                  struct ptlrpc_request **request)
1657 {
1658         struct ptlrpc_request   *req = NULL;
1659         struct obd_device       *obd = exp->exp_obd;
1660         struct lmv_obd          *lmv = &obd->u.lmv;
1661         struct lmv_tgt_desc     *tgt;
1662         struct mdt_body         *body;
1663         int                      rc;
1664         ENTRY;
1665
1666         rc = lmv_check_connect(obd);
1667         if (rc)
1668                 RETURN(rc);
1669
1670         tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
1671         if (IS_ERR(tgt))
1672                 RETURN(PTR_ERR(tgt));
1673
1674         CDEBUG(D_INODE, "GETATTR_NAME for %*s on "DFID" -> mds #%d\n",
1675                op_data->op_namelen, op_data->op_name, PFID(&op_data->op_fid1),
1676                tgt->ltd_idx);
1677
1678         rc = md_getattr_name(tgt->ltd_exp, op_data, request);
1679         if (rc != 0)
1680                 RETURN(rc);
1681
1682         body = req_capsule_server_get(&(*request)->rq_pill,
1683                                       &RMF_MDT_BODY);
1684         LASSERT(body != NULL);
1685
1686         if (body->valid & OBD_MD_MDS) {
1687                 struct lu_fid rid = body->fid1;
1688                 CDEBUG(D_INODE, "Request attrs for "DFID"\n",
1689                        PFID(&rid));
1690
1691                 tgt = lmv_find_target(lmv, &rid);
1692                 if (IS_ERR(tgt)) {
1693                         ptlrpc_req_finished(*request);
1694                         RETURN(PTR_ERR(tgt));
1695                 }
1696
1697                 op_data->op_fid1 = rid;
1698                 op_data->op_valid |= OBD_MD_FLCROSSREF;
1699                 op_data->op_namelen = 0;
1700                 op_data->op_name = NULL;
1701                 rc = md_getattr_name(tgt->ltd_exp, op_data, &req);
1702                 ptlrpc_req_finished(*request);
1703                 *request = req;
1704         }
1705
1706         RETURN(rc);
1707 }
1708
1709 #define md_op_data_fid(op_data, fl)                     \
1710         (fl == MF_MDC_CANCEL_FID1 ? &op_data->op_fid1 : \
1711          fl == MF_MDC_CANCEL_FID2 ? &op_data->op_fid2 : \
1712          fl == MF_MDC_CANCEL_FID3 ? &op_data->op_fid3 : \
1713          fl == MF_MDC_CANCEL_FID4 ? &op_data->op_fid4 : \
1714          NULL)
1715
1716 static int lmv_early_cancel(struct obd_export *exp, struct md_op_data *op_data,
1717                             int op_tgt, ldlm_mode_t mode, int bits, int flag)
1718 {
1719         struct lu_fid          *fid = md_op_data_fid(op_data, flag);
1720         struct obd_device      *obd = exp->exp_obd;
1721         struct lmv_obd         *lmv = &obd->u.lmv;
1722         struct lmv_tgt_desc    *tgt;
1723         ldlm_policy_data_t      policy = {{0}};
1724         int                     rc = 0;
1725         ENTRY;
1726
1727         if (!fid_is_sane(fid))
1728                 RETURN(0);
1729
1730         tgt = lmv_find_target(lmv, fid);
1731         if (IS_ERR(tgt))
1732                 RETURN(PTR_ERR(tgt));
1733
1734         if (tgt->ltd_idx != op_tgt) {
1735                 CDEBUG(D_INODE, "EARLY_CANCEL on "DFID"\n", PFID(fid));
1736                 policy.l_inodebits.bits = bits;
1737                 rc = md_cancel_unused(tgt->ltd_exp, fid, &policy,
1738                                       mode, LCF_ASYNC, NULL);
1739         } else {
1740                 CDEBUG(D_INODE,
1741                        "EARLY_CANCEL skip operation target %d on "DFID"\n",
1742                        op_tgt, PFID(fid));
1743                 op_data->op_flags |= flag;
1744                 rc = 0;
1745         }
1746
1747         RETURN(rc);
1748 }
1749
1750 /*
1751  * llite passes fid of an target inode in op_data->op_fid1 and id of directory in
1752  * op_data->op_fid2
1753  */
1754 static int lmv_link(struct obd_export *exp, struct md_op_data *op_data,
1755                     struct ptlrpc_request **request)
1756 {
1757         struct obd_device       *obd = exp->exp_obd;
1758         struct lmv_obd          *lmv = &obd->u.lmv;
1759         struct lmv_tgt_desc     *tgt;
1760         int                      rc;
1761         ENTRY;
1762
1763         rc = lmv_check_connect(obd);
1764         if (rc)
1765                 RETURN(rc);
1766
1767         LASSERT(op_data->op_namelen != 0);
1768
1769         CDEBUG(D_INODE, "LINK "DFID":%*s to "DFID"\n",
1770                PFID(&op_data->op_fid2), op_data->op_namelen,
1771                op_data->op_name, PFID(&op_data->op_fid1));
1772
1773         op_data->op_fsuid = cfs_curproc_fsuid();
1774         op_data->op_fsgid = cfs_curproc_fsgid();
1775         op_data->op_cap = cfs_curproc_cap_pack();
1776         tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid2);
1777         if (IS_ERR(tgt))
1778                 RETURN(PTR_ERR(tgt));
1779
1780         /*
1781          * Cancel UPDATE lock on child (fid1).
1782          */
1783         op_data->op_flags |= MF_MDC_CANCEL_FID2;
1784         rc = lmv_early_cancel(exp, op_data, tgt->ltd_idx, LCK_EX,
1785                               MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID1);
1786         if (rc != 0)
1787                 RETURN(rc);
1788
1789         rc = md_link(tgt->ltd_exp, op_data, request);
1790
1791         RETURN(rc);
1792 }
1793
1794 static int lmv_rename(struct obd_export *exp, struct md_op_data *op_data,
1795                       const char *old, int oldlen, const char *new, int newlen,
1796                       struct ptlrpc_request **request)
1797 {
1798         struct obd_device       *obd = exp->exp_obd;
1799         struct lmv_obd          *lmv = &obd->u.lmv;
1800         struct lmv_tgt_desc     *src_tgt;
1801         struct lmv_tgt_desc     *tgt_tgt;
1802         int                     rc;
1803         ENTRY;
1804
1805         LASSERT(oldlen != 0);
1806
1807         CDEBUG(D_INODE, "RENAME %*s in "DFID" to %*s in "DFID"\n",
1808                oldlen, old, PFID(&op_data->op_fid1),
1809                newlen, new, PFID(&op_data->op_fid2));
1810
1811         rc = lmv_check_connect(obd);
1812         if (rc)
1813                 RETURN(rc);
1814
1815         op_data->op_fsuid = cfs_curproc_fsuid();
1816         op_data->op_fsgid = cfs_curproc_fsgid();
1817         op_data->op_cap = cfs_curproc_cap_pack();
1818         src_tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
1819         if (IS_ERR(src_tgt))
1820                 RETURN(PTR_ERR(src_tgt));
1821
1822         tgt_tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid2);
1823         if (IS_ERR(tgt_tgt))
1824                 RETURN(PTR_ERR(tgt_tgt));
1825         /*
1826          * LOOKUP lock on src child (fid3) should also be cancelled for
1827          * src_tgt in mdc_rename.
1828          */
1829         op_data->op_flags |= MF_MDC_CANCEL_FID1 | MF_MDC_CANCEL_FID3;
1830
1831         /*
1832          * Cancel UPDATE locks on tgt parent (fid2), tgt_tgt is its
1833          * own target.
1834          */
1835         rc = lmv_early_cancel(exp, op_data, src_tgt->ltd_idx,
1836                               LCK_EX, MDS_INODELOCK_UPDATE,
1837                               MF_MDC_CANCEL_FID2);
1838
1839         /*
1840          * Cancel LOOKUP locks on tgt child (fid4) for parent tgt_tgt.
1841          */
1842         if (rc == 0) {
1843                 rc = lmv_early_cancel(exp, op_data, src_tgt->ltd_idx,
1844                                       LCK_EX, MDS_INODELOCK_LOOKUP,
1845                                       MF_MDC_CANCEL_FID4);
1846         }
1847
1848         /*
1849          * Cancel all the locks on tgt child (fid4).
1850          */
1851         if (rc == 0)
1852                 rc = lmv_early_cancel(exp, op_data, src_tgt->ltd_idx,
1853                                       LCK_EX, MDS_INODELOCK_FULL,
1854                                       MF_MDC_CANCEL_FID4);
1855
1856         if (rc == 0)
1857                 rc = md_rename(src_tgt->ltd_exp, op_data, old, oldlen,
1858                                new, newlen, request);
1859         RETURN(rc);
1860 }
1861
1862 static int lmv_setattr(struct obd_export *exp, struct md_op_data *op_data,
1863                        void *ea, int ealen, void *ea2, int ea2len,
1864                        struct ptlrpc_request **request,
1865                        struct md_open_data **mod)
1866 {
1867         struct obd_device       *obd = exp->exp_obd;
1868         struct lmv_obd          *lmv = &obd->u.lmv;
1869         struct lmv_tgt_desc     *tgt;
1870         int                      rc = 0;
1871         ENTRY;
1872
1873         rc = lmv_check_connect(obd);
1874         if (rc)
1875                 RETURN(rc);
1876
1877         CDEBUG(D_INODE, "SETATTR for "DFID", valid 0x%x\n",
1878                PFID(&op_data->op_fid1), op_data->op_attr.ia_valid);
1879
1880         op_data->op_flags |= MF_MDC_CANCEL_FID1;
1881         tgt = lmv_find_target(lmv, &op_data->op_fid1);
1882         if (IS_ERR(tgt))
1883                 RETURN(PTR_ERR(tgt));
1884
1885         rc = md_setattr(tgt->ltd_exp, op_data, ea, ealen, ea2,
1886                         ea2len, request, mod);
1887
1888         RETURN(rc);
1889 }
1890
1891 static int lmv_sync(struct obd_export *exp, const struct lu_fid *fid,
1892                     struct obd_capa *oc, struct ptlrpc_request **request)
1893 {
1894         struct obd_device         *obd = exp->exp_obd;
1895         struct lmv_obd            *lmv = &obd->u.lmv;
1896         struct lmv_tgt_desc       *tgt;
1897         int                        rc;
1898         ENTRY;
1899
1900         rc = lmv_check_connect(obd);
1901         if (rc)
1902                 RETURN(rc);
1903
1904         tgt = lmv_find_target(lmv, fid);
1905         if (IS_ERR(tgt))
1906                 RETURN(PTR_ERR(tgt));
1907
1908         rc = md_sync(tgt->ltd_exp, fid, oc, request);
1909         RETURN(rc);
1910 }
1911
1912 static int lmv_readpage(struct obd_export *exp, struct md_op_data *op_data,
1913                         struct page **pages, struct ptlrpc_request **request)
1914 {
1915         struct obd_device       *obd = exp->exp_obd;
1916         struct lmv_obd          *lmv = &obd->u.lmv;
1917         __u64                    offset = op_data->op_offset;
1918         int                      rc;
1919         int                      i;
1920         /* number of pages read, in CFS_PAGE_SIZE */
1921         int                      nrdpgs;
1922         /* number of pages transferred in LU_PAGE_SIZE */
1923         int                      nlupgs;
1924         struct lmv_tgt_desc     *tgt;
1925         struct lu_dirpage       *dp;
1926         struct lu_dirent        *ent;
1927         ENTRY;
1928
1929         rc = lmv_check_connect(obd);
1930         if (rc)
1931                 RETURN(rc);
1932
1933         CDEBUG(D_INODE, "READPAGE at "LPX64" from "DFID"\n",
1934                offset, PFID(&op_data->op_fid1));
1935
1936         /*
1937          * This case handle directory lookup in clustered metadata case (i.e.
1938          * split directory is located on multiple md servers.)
1939          * each server keeps directory entries for certain range of hashes.
1940          * E.g. we have N server and suppose hash range is 0 to MAX_HASH.
1941          * first server will keep records with hashes [ 0 ... MAX_HASH /N  - 1],
1942          * second one with hashes [MAX_HASH / N ... 2 * MAX_HASH / N] and
1943          * so on....
1944          *      readdir can simply start reading entries from 0 - N server in
1945          * order but that will not scale well as all client will request dir in
1946          * to server in same order.
1947          * Following algorithm does optimization:
1948          * Instead of doing readdir in 1, 2, ...., N order, client with a
1949          * rank R does readdir in R, R + 1, ..., N, 1, ... R - 1 order.
1950          * (every client has rank R)
1951          *      But ll_readdir() expect offset range [0 to MAX_HASH/N) but
1952          * since client ask dir from MDS{R} client has pages with offsets
1953          * [R*MAX_HASH/N ... (R + 1)*MAX_HASH/N] there for we do hash_adj
1954          * on hash  values that we get.
1955          * Since these codes might be still useful for sharded directory, so
1956          * Keeping this code for further reference
1957         if (0) {
1958                 LASSERT(nr > 0);
1959                 seg_size = MAX_HASH_SIZE;
1960                 do_div(seg_size, nr);
1961                 los      = obj->lo_stripes;
1962                 tgt      = lmv_get_target(lmv, los[0].ls_mds);
1963                 rank     = lmv_node_rank(tgt->ltd_exp, fid) % nr;
1964                 tgt_tmp  = offset;
1965                 do_div(tgt_tmp, seg_size);
1966                 tgt0_idx = do_div(tgt_tmp,  nr);
1967                 tgt_idx  = (tgt0_idx + rank) % nr;
1968
1969                 if (tgt_idx < tgt0_idx)
1970                          * Wrap around.
1971                          *
1972                          * Last segment has unusual length due to division
1973                          * rounding.
1974                         hash_adj = MAX_HASH_SIZE - seg_size * nr;
1975                 else
1976                         hash_adj = 0;
1977
1978                 hash_adj += rank * seg_size;
1979
1980                 CDEBUG(D_INODE, "Readpage hash adjustment: %x "LPX64" "
1981                        LPX64"/%x -> "LPX64"/%x\n", rank, hash_adj,
1982                        offset, tgt0_idx, offset + hash_adj, tgt_idx);
1983
1984                 offset = (offset + hash_adj) & MAX_HASH_SIZE;
1985                 rid = lsm->mea_oinfo[tgt_idx].lmo_fid;
1986                 tgt = lmv_get_target(lmv, lsm->mea_oinfo[tgt_idx].lmo_mds);
1987
1988                 CDEBUG(D_INODE, "Forward to "DFID" with offset %lu i %d\n",
1989                        PFID(&rid), (unsigned long)offset, tgt_idx);
1990         }
1991         */
1992         tgt = lmv_find_target(lmv, &op_data->op_fid1);
1993         if (IS_ERR(tgt))
1994                 RETURN(PTR_ERR(tgt));
1995
1996         rc = md_readpage(tgt->ltd_exp, op_data, pages, request);
1997         if (rc != 0)
1998                 RETURN(rc);
1999
2000         nrdpgs = ((*request)->rq_bulk->bd_nob_transferred + CFS_PAGE_SIZE - 1)
2001                  >> CFS_PAGE_SHIFT;
2002         nlupgs = (*request)->rq_bulk->bd_nob_transferred >> LU_PAGE_SHIFT;
2003         LASSERT(!((*request)->rq_bulk->bd_nob_transferred & ~LU_PAGE_MASK));
2004         LASSERT(nrdpgs > 0 && nrdpgs <= op_data->op_npages);
2005
2006         CDEBUG(D_INODE, "read %d(%d)/%d pages\n", nrdpgs, nlupgs,
2007                op_data->op_npages);
2008
2009         for (i = 0; i < nrdpgs; i++) {
2010 #if CFS_PAGE_SIZE > LU_PAGE_SIZE
2011                 struct lu_dirpage *first;
2012                 __u64 hash_end = 0;
2013                 __u32 flags = 0;
2014 #endif
2015                 struct lu_dirent *tmp = NULL;
2016
2017                 dp = cfs_kmap(pages[i]);
2018                 ent = lu_dirent_start(dp);
2019 #if CFS_PAGE_SIZE > LU_PAGE_SIZE
2020                 first = dp;
2021                 hash_end = dp->ldp_hash_end;
2022 repeat:
2023 #endif
2024                 nlupgs--;
2025
2026                 for (tmp = ent; ent != NULL;
2027                      tmp = ent, ent = lu_dirent_next(ent));
2028 #if CFS_PAGE_SIZE > LU_PAGE_SIZE
2029                 dp = (struct lu_dirpage *)((char *)dp + LU_PAGE_SIZE);
2030                 if (((unsigned long)dp & ~CFS_PAGE_MASK) && nlupgs > 0) {
2031                         ent = lu_dirent_start(dp);
2032
2033                         if (tmp) {
2034                                 /* enlarge the end entry lde_reclen from 0 to
2035                                  * first entry of next lu_dirpage, in this way
2036                                  * several lu_dirpages can be stored into one
2037                                  * client page on client. */
2038                                 tmp = ((void *)tmp) +
2039                                       le16_to_cpu(tmp->lde_reclen);
2040                                 tmp->lde_reclen =
2041                                         cpu_to_le16((char *)(dp->ldp_entries) -
2042                                                     (char *)tmp);
2043                                 goto repeat;
2044                         }
2045                 }
2046                 first->ldp_hash_end = hash_end;
2047                 first->ldp_flags &= ~cpu_to_le32(LDF_COLLIDE);
2048                 first->ldp_flags |= flags & cpu_to_le32(LDF_COLLIDE);
2049 #else
2050                 SET_BUT_UNUSED(tmp);
2051 #endif
2052                 cfs_kunmap(pages[i]);
2053         }
2054         RETURN(rc);
2055 }
2056
2057 static int lmv_unlink(struct obd_export *exp, struct md_op_data *op_data,
2058                       struct ptlrpc_request **request)
2059 {
2060         struct obd_device       *obd = exp->exp_obd;
2061         struct lmv_obd          *lmv = &obd->u.lmv;
2062         struct lmv_tgt_desc     *tgt = NULL;
2063         struct mdt_body         *body;
2064         int                     rc;
2065         ENTRY;
2066
2067         rc = lmv_check_connect(obd);
2068         if (rc)
2069                 RETURN(rc);
2070 retry:
2071         /* Send unlink requests to the MDT where the child is located */
2072         if (likely(!fid_is_zero(&op_data->op_fid2)))
2073                 tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid2);
2074         else
2075                 tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
2076         if (IS_ERR(tgt))
2077                 RETURN(PTR_ERR(tgt));
2078
2079         op_data->op_fsuid = cfs_curproc_fsuid();
2080         op_data->op_fsgid = cfs_curproc_fsgid();
2081         op_data->op_cap = cfs_curproc_cap_pack();
2082
2083         /*
2084          * If child's fid is given, cancel unused locks for it if it is from
2085          * another export than parent.
2086          *
2087          * LOOKUP lock for child (fid3) should also be cancelled on parent
2088          * tgt_tgt in mdc_unlink().
2089          */
2090         op_data->op_flags |= MF_MDC_CANCEL_FID1 | MF_MDC_CANCEL_FID3;
2091
2092         /*
2093          * Cancel FULL locks on child (fid3).
2094          */
2095         rc = lmv_early_cancel(exp, op_data, tgt->ltd_idx, LCK_EX,
2096                               MDS_INODELOCK_FULL, MF_MDC_CANCEL_FID3);
2097
2098         if (rc != 0)
2099                 RETURN(rc);
2100
2101         CDEBUG(D_INODE, "unlink with fid="DFID"/"DFID" -> mds #%d\n",
2102                PFID(&op_data->op_fid1), PFID(&op_data->op_fid2), tgt->ltd_idx);
2103
2104         rc = md_unlink(tgt->ltd_exp, op_data, request);
2105         if (rc != 0 && rc != -EREMOTE)
2106                 RETURN(rc);
2107
2108         body = req_capsule_server_get(&(*request)->rq_pill, &RMF_MDT_BODY);
2109         if (body == NULL)
2110                 RETURN(-EPROTO);
2111
2112         /* Not cross-ref case, just get out of here. */
2113         if (likely(!(body->valid & OBD_MD_MDS)))
2114                 RETURN(0);
2115
2116         CDEBUG(D_INODE, "%s: try unlink to another MDT for "DFID"\n",
2117                exp->exp_obd->obd_name, PFID(&body->fid1));
2118
2119         /* This is a remote object, try remote MDT, Note: it may
2120          * try more than 1 time here, Considering following case
2121          * /mnt/lustre is root on MDT0, remote1 is on MDT1
2122          * 1. Initially A does not know where remote1 is, it send
2123          *    unlink RPC to MDT0, MDT0 return -EREMOTE, it will
2124          *    resend unlink RPC to MDT1 (retry 1st time).
2125          *
2126          * 2. During the unlink RPC in flight,
2127          *    client B mv /mnt/lustre/remote1 /mnt/lustre/remote2
2128          *    and create new remote1, but on MDT0
2129          *
2130          * 3. MDT1 get unlink RPC(from A), then do remote lock on
2131          *    /mnt/lustre, then lookup get fid of remote1, and find
2132          *    it is remote dir again, and replay -EREMOTE again.
2133          *
2134          * 4. Then A will resend unlink RPC to MDT0. (retry 2nd times).
2135          *
2136          * In theory, it might try unlimited time here, but it should
2137          * be very rare case.  */
2138         op_data->op_fid2 = body->fid1;
2139         ptlrpc_req_finished(*request);
2140         *request = NULL;
2141
2142         goto retry;
2143 }
2144
2145 static int lmv_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
2146 {
2147         struct lmv_obd *lmv = &obd->u.lmv;
2148         int rc = 0;
2149
2150         switch (stage) {
2151         case OBD_CLEANUP_EARLY:
2152                 /* XXX: here should be calling obd_precleanup() down to
2153                  * stack. */
2154                 break;
2155         case OBD_CLEANUP_EXPORTS:
2156                 fld_client_proc_fini(&lmv->lmv_fld);
2157                 lprocfs_obd_cleanup(obd);
2158                 break;
2159         default:
2160                 break;
2161         }
2162         RETURN(rc);
2163 }
2164
2165 static int lmv_get_info(const struct lu_env *env, struct obd_export *exp,
2166                         __u32 keylen, void *key, __u32 *vallen, void *val,
2167                         struct lov_stripe_md *lsm)
2168 {
2169         struct obd_device       *obd;
2170         struct lmv_obd          *lmv;
2171         int                      rc = 0;
2172         ENTRY;
2173
2174         obd = class_exp2obd(exp);
2175         if (obd == NULL) {
2176                 CDEBUG(D_IOCTL, "Invalid client cookie "LPX64"\n",
2177                        exp->exp_handle.h_cookie);
2178                 RETURN(-EINVAL);
2179         }
2180
2181         lmv = &obd->u.lmv;
2182         if (keylen >= strlen("remote_flag") && !strcmp(key, "remote_flag")) {
2183                 struct lmv_tgt_desc *tgt;
2184                 int i;
2185
2186                 rc = lmv_check_connect(obd);
2187                 if (rc)
2188                         RETURN(rc);
2189
2190                 LASSERT(*vallen == sizeof(__u32));
2191                 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2192                         tgt = lmv->tgts[i];
2193                         /*
2194                          * All tgts should be connected when this gets called.
2195                          */
2196                         if (tgt == NULL || tgt->ltd_exp == NULL)
2197                                 continue;
2198
2199                         if (!obd_get_info(env, tgt->ltd_exp, keylen, key,
2200                                           vallen, val, NULL))
2201                                 RETURN(0);
2202                 }
2203                 RETURN(-EINVAL);
2204         } else if (KEY_IS(KEY_MAX_EASIZE) || KEY_IS(KEY_CONN_DATA)) {
2205                 rc = lmv_check_connect(obd);
2206                 if (rc)
2207                         RETURN(rc);
2208
2209                 /*
2210                  * Forwarding this request to first MDS, it should know LOV
2211                  * desc.
2212                  */
2213                 rc = obd_get_info(env, lmv->tgts[0]->ltd_exp, keylen, key,
2214                                   vallen, val, NULL);
2215                 if (!rc && KEY_IS(KEY_CONN_DATA))
2216                         exp->exp_connect_data = *(struct obd_connect_data *)val;
2217                 RETURN(rc);
2218         } else if (KEY_IS(KEY_TGT_COUNT)) {
2219                 *((int *)val) = lmv->desc.ld_tgt_count;
2220                 RETURN(0);
2221         }
2222
2223         CDEBUG(D_IOCTL, "Invalid key\n");
2224         RETURN(-EINVAL);
2225 }
2226
2227 int lmv_set_info_async(const struct lu_env *env, struct obd_export *exp,
2228                        obd_count keylen, void *key, obd_count vallen,
2229                        void *val, struct ptlrpc_request_set *set)
2230 {
2231         struct lmv_tgt_desc    *tgt;
2232         struct obd_device      *obd;
2233         struct lmv_obd         *lmv;
2234         int rc = 0;
2235         ENTRY;
2236
2237         obd = class_exp2obd(exp);
2238         if (obd == NULL) {
2239                 CDEBUG(D_IOCTL, "Invalid client cookie "LPX64"\n",
2240                        exp->exp_handle.h_cookie);
2241                 RETURN(-EINVAL);
2242         }
2243         lmv = &obd->u.lmv;
2244
2245         if (KEY_IS(KEY_READ_ONLY) || KEY_IS(KEY_FLUSH_CTX)) {
2246                 int i, err = 0;
2247
2248                 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2249                         tgt = lmv->tgts[i];
2250
2251                         if (tgt == NULL || tgt->ltd_exp == NULL)
2252                                 continue;
2253
2254                         err = obd_set_info_async(env, tgt->ltd_exp,
2255                                                  keylen, key, vallen, val, set);
2256                         if (err && rc == 0)
2257                                 rc = err;
2258                 }
2259
2260                 RETURN(rc);
2261         }
2262
2263         RETURN(-EINVAL);
2264 }
2265
2266 int lmv_packmd(struct obd_export *exp, struct lov_mds_md **lmmp,
2267                struct lov_stripe_md *lsm)
2268 {
2269         struct obd_device         *obd = class_exp2obd(exp);
2270         struct lmv_obd            *lmv = &obd->u.lmv;
2271         struct lmv_stripe_md      *meap;
2272         struct lmv_stripe_md      *lsmp;
2273         int                        mea_size;
2274         int                        i;
2275         ENTRY;
2276
2277         mea_size = lmv_get_easize(lmv);
2278         if (!lmmp)
2279                 RETURN(mea_size);
2280
2281         if (*lmmp && !lsm) {
2282                 OBD_FREE_LARGE(*lmmp, mea_size);
2283                 *lmmp = NULL;
2284                 RETURN(0);
2285         }
2286
2287         if (*lmmp == NULL) {
2288                 OBD_ALLOC_LARGE(*lmmp, mea_size);
2289                 if (*lmmp == NULL)
2290                         RETURN(-ENOMEM);
2291         }
2292
2293         if (!lsm)
2294                 RETURN(mea_size);
2295
2296         lsmp = (struct lmv_stripe_md *)lsm;
2297         meap = (struct lmv_stripe_md *)*lmmp;
2298
2299         if (lsmp->mea_magic != MEA_MAGIC_LAST_CHAR &&
2300             lsmp->mea_magic != MEA_MAGIC_ALL_CHARS)
2301                 RETURN(-EINVAL);
2302
2303         meap->mea_magic = cpu_to_le32(lsmp->mea_magic);
2304         meap->mea_count = cpu_to_le32(lsmp->mea_count);
2305         meap->mea_master = cpu_to_le32(lsmp->mea_master);
2306
2307         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2308                 meap->mea_ids[i] = meap->mea_ids[i];
2309                 fid_cpu_to_le(&meap->mea_ids[i], &meap->mea_ids[i]);
2310         }
2311
2312         RETURN(mea_size);
2313 }
2314
2315 int lmv_unpackmd(struct obd_export *exp, struct lov_stripe_md **lsmp,
2316                  struct lov_mds_md *lmm, int lmm_size)
2317 {
2318         struct obd_device          *obd = class_exp2obd(exp);
2319         struct lmv_stripe_md      **tmea = (struct lmv_stripe_md **)lsmp;
2320         struct lmv_stripe_md       *mea = (struct lmv_stripe_md *)lmm;
2321         struct lmv_obd             *lmv = &obd->u.lmv;
2322         int                         mea_size;
2323         int                         i;
2324         __u32                       magic;
2325         ENTRY;
2326
2327         mea_size = lmv_get_easize(lmv);
2328         if (lsmp == NULL)
2329                 return mea_size;
2330
2331         if (*lsmp != NULL && lmm == NULL) {
2332                 OBD_FREE_LARGE(*tmea, mea_size);
2333                 *lsmp = NULL;
2334                 RETURN(0);
2335         }
2336
2337         LASSERT(mea_size == lmm_size);
2338
2339         OBD_ALLOC_LARGE(*tmea, mea_size);
2340         if (*tmea == NULL)
2341                 RETURN(-ENOMEM);
2342
2343         if (!lmm)
2344                 RETURN(mea_size);
2345
2346         if (mea->mea_magic == MEA_MAGIC_LAST_CHAR ||
2347             mea->mea_magic == MEA_MAGIC_ALL_CHARS ||
2348             mea->mea_magic == MEA_MAGIC_HASH_SEGMENT)
2349         {
2350                 magic = le32_to_cpu(mea->mea_magic);
2351         } else {
2352                 /*
2353                  * Old mea is not handled here.
2354                  */
2355                 CERROR("Old not supportable EA is found\n");
2356                 LBUG();
2357         }
2358
2359         (*tmea)->mea_magic = magic;
2360         (*tmea)->mea_count = le32_to_cpu(mea->mea_count);
2361         (*tmea)->mea_master = le32_to_cpu(mea->mea_master);
2362
2363         for (i = 0; i < (*tmea)->mea_count; i++) {
2364                 (*tmea)->mea_ids[i] = mea->mea_ids[i];
2365                 fid_le_to_cpu(&(*tmea)->mea_ids[i], &(*tmea)->mea_ids[i]);
2366         }
2367         RETURN(mea_size);
2368 }
2369
2370 static int lmv_cancel_unused(struct obd_export *exp, const struct lu_fid *fid,
2371                              ldlm_policy_data_t *policy, ldlm_mode_t mode,
2372                              ldlm_cancel_flags_t flags, void *opaque)
2373 {
2374         struct obd_device       *obd = exp->exp_obd;
2375         struct lmv_obd          *lmv = &obd->u.lmv;
2376         int                      rc = 0;
2377         int                      err;
2378         int                      i;
2379         ENTRY;
2380
2381         LASSERT(fid != NULL);
2382
2383         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2384                 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL ||
2385                     lmv->tgts[i]->ltd_active == 0)
2386                         continue;
2387
2388                 err = md_cancel_unused(lmv->tgts[i]->ltd_exp, fid,
2389                                        policy, mode, flags, opaque);
2390                 if (!rc)
2391                         rc = err;
2392         }
2393         RETURN(rc);
2394 }
2395
2396 int lmv_set_lock_data(struct obd_export *exp, __u64 *lockh, void *data,
2397                       __u64 *bits)
2398 {
2399         struct lmv_obd          *lmv = &exp->exp_obd->u.lmv;
2400         int                      rc;
2401         ENTRY;
2402
2403         rc =  md_set_lock_data(lmv->tgts[0]->ltd_exp, lockh, data, bits);
2404         RETURN(rc);
2405 }
2406
2407 ldlm_mode_t lmv_lock_match(struct obd_export *exp, __u64 flags,
2408                            const struct lu_fid *fid, ldlm_type_t type,
2409                            ldlm_policy_data_t *policy, ldlm_mode_t mode,
2410                            struct lustre_handle *lockh)
2411 {
2412         struct obd_device       *obd = exp->exp_obd;
2413         struct lmv_obd          *lmv = &obd->u.lmv;
2414         ldlm_mode_t              rc;
2415         int                      i;
2416         ENTRY;
2417
2418         CDEBUG(D_INODE, "Lock match for "DFID"\n", PFID(fid));
2419
2420         /*
2421          * With CMD every object can have two locks in different namespaces:
2422          * lookup lock in space of mds storing direntry and update/open lock in
2423          * space of mds storing inode. Thus we check all targets, not only that
2424          * one fid was created in.
2425          */
2426         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2427                 if (lmv->tgts[i] == NULL ||
2428                     lmv->tgts[i]->ltd_exp == NULL ||
2429                     lmv->tgts[i]->ltd_active == 0)
2430                         continue;
2431
2432                 rc = md_lock_match(lmv->tgts[i]->ltd_exp, flags, fid,
2433                                    type, policy, mode, lockh);
2434                 if (rc)
2435                         RETURN(rc);
2436         }
2437
2438         RETURN(0);
2439 }
2440
2441 int lmv_get_lustre_md(struct obd_export *exp, struct ptlrpc_request *req,
2442                       struct obd_export *dt_exp, struct obd_export *md_exp,
2443                       struct lustre_md *md)
2444 {
2445         struct lmv_obd          *lmv = &exp->exp_obd->u.lmv;
2446
2447         return md_get_lustre_md(lmv->tgts[0]->ltd_exp, req, dt_exp, md_exp, md);
2448 }
2449
2450 int lmv_free_lustre_md(struct obd_export *exp, struct lustre_md *md)
2451 {
2452         struct obd_device       *obd = exp->exp_obd;
2453         struct lmv_obd          *lmv = &obd->u.lmv;
2454         ENTRY;
2455
2456         if (md->mea)
2457                 obd_free_memmd(exp, (void *)&md->mea);
2458         RETURN(md_free_lustre_md(lmv->tgts[0]->ltd_exp, md));
2459 }
2460
2461 int lmv_set_open_replay_data(struct obd_export *exp,
2462                              struct obd_client_handle *och,
2463                              struct ptlrpc_request *open_req)
2464 {
2465         struct obd_device       *obd = exp->exp_obd;
2466         struct lmv_obd          *lmv = &obd->u.lmv;
2467         struct lmv_tgt_desc     *tgt;
2468         ENTRY;
2469
2470         tgt = lmv_find_target(lmv, &och->och_fid);
2471         if (IS_ERR(tgt))
2472                 RETURN(PTR_ERR(tgt));
2473
2474         RETURN(md_set_open_replay_data(tgt->ltd_exp, och, open_req));
2475 }
2476
2477 int lmv_clear_open_replay_data(struct obd_export *exp,
2478                                struct obd_client_handle *och)
2479 {
2480         struct obd_device       *obd = exp->exp_obd;
2481         struct lmv_obd          *lmv = &obd->u.lmv;
2482         struct lmv_tgt_desc     *tgt;
2483         ENTRY;
2484
2485         tgt = lmv_find_target(lmv, &och->och_fid);
2486         if (IS_ERR(tgt))
2487                 RETURN(PTR_ERR(tgt));
2488
2489         RETURN(md_clear_open_replay_data(tgt->ltd_exp, och));
2490 }
2491
2492 static int lmv_get_remote_perm(struct obd_export *exp,
2493                                const struct lu_fid *fid,
2494                                struct obd_capa *oc, __u32 suppgid,
2495                                struct ptlrpc_request **request)
2496 {
2497         struct obd_device       *obd = exp->exp_obd;
2498         struct lmv_obd          *lmv = &obd->u.lmv;
2499         struct lmv_tgt_desc     *tgt;
2500         int                      rc;
2501         ENTRY;
2502
2503         rc = lmv_check_connect(obd);
2504         if (rc)
2505                 RETURN(rc);
2506
2507         tgt = lmv_find_target(lmv, fid);
2508         if (IS_ERR(tgt))
2509                 RETURN(PTR_ERR(tgt));
2510
2511         rc = md_get_remote_perm(tgt->ltd_exp, fid, oc, suppgid, request);
2512         RETURN(rc);
2513 }
2514
2515 static int lmv_renew_capa(struct obd_export *exp, struct obd_capa *oc,
2516                           renew_capa_cb_t cb)
2517 {
2518         struct obd_device       *obd = exp->exp_obd;
2519         struct lmv_obd          *lmv = &obd->u.lmv;
2520         struct lmv_tgt_desc     *tgt;
2521         int                      rc;
2522         ENTRY;
2523
2524         rc = lmv_check_connect(obd);
2525         if (rc)
2526                 RETURN(rc);
2527
2528         tgt = lmv_find_target(lmv, &oc->c_capa.lc_fid);
2529         if (IS_ERR(tgt))
2530                 RETURN(PTR_ERR(tgt));
2531
2532         rc = md_renew_capa(tgt->ltd_exp, oc, cb);
2533         RETURN(rc);
2534 }
2535
2536 int lmv_unpack_capa(struct obd_export *exp, struct ptlrpc_request *req,
2537                     const struct req_msg_field *field, struct obd_capa **oc)
2538 {
2539         struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
2540
2541         return md_unpack_capa(lmv->tgts[0]->ltd_exp, req, field, oc);
2542 }
2543
2544 int lmv_intent_getattr_async(struct obd_export *exp,
2545                              struct md_enqueue_info *minfo,
2546                              struct ldlm_enqueue_info *einfo)
2547 {
2548         struct md_op_data       *op_data = &minfo->mi_data;
2549         struct obd_device       *obd = exp->exp_obd;
2550         struct lmv_obd          *lmv = &obd->u.lmv;
2551         struct lmv_tgt_desc     *tgt = NULL;
2552         int                      rc;
2553         ENTRY;
2554
2555         rc = lmv_check_connect(obd);
2556         if (rc)
2557                 RETURN(rc);
2558
2559         tgt = lmv_find_target(lmv, &op_data->op_fid1);
2560         if (IS_ERR(tgt))
2561                 RETURN(PTR_ERR(tgt));
2562
2563         rc = md_intent_getattr_async(tgt->ltd_exp, minfo, einfo);
2564         RETURN(rc);
2565 }
2566
2567 int lmv_revalidate_lock(struct obd_export *exp, struct lookup_intent *it,
2568                         struct lu_fid *fid, __u64 *bits)
2569 {
2570         struct obd_device       *obd = exp->exp_obd;
2571         struct lmv_obd          *lmv = &obd->u.lmv;
2572         struct lmv_tgt_desc     *tgt;
2573         int                      rc;
2574         ENTRY;
2575
2576         rc = lmv_check_connect(obd);
2577         if (rc)
2578                 RETURN(rc);
2579
2580         tgt = lmv_find_target(lmv, fid);
2581         if (IS_ERR(tgt))
2582                 RETURN(PTR_ERR(tgt));
2583
2584         rc = md_revalidate_lock(tgt->ltd_exp, it, fid, bits);
2585         RETURN(rc);
2586 }
2587
2588 /**
2589  * For lmv, only need to send request to master MDT, and the master MDT will
2590  * process with other slave MDTs. The only exception is Q_GETOQUOTA for which
2591  * we directly fetch data from the slave MDTs.
2592  */
2593 int lmv_quotactl(struct obd_device *unused, struct obd_export *exp,
2594                  struct obd_quotactl *oqctl)
2595 {
2596         struct obd_device   *obd = class_exp2obd(exp);
2597         struct lmv_obd      *lmv = &obd->u.lmv;
2598         struct lmv_tgt_desc *tgt = lmv->tgts[0];
2599         int                  rc = 0, i;
2600         __u64                curspace, curinodes;
2601         ENTRY;
2602
2603         if (!lmv->desc.ld_tgt_count || !tgt->ltd_active) {
2604                 CERROR("master lmv inactive\n");
2605                 RETURN(-EIO);
2606         }
2607
2608         if (oqctl->qc_cmd != Q_GETOQUOTA) {
2609                 rc = obd_quotactl(tgt->ltd_exp, oqctl);
2610                 RETURN(rc);
2611         }
2612
2613         curspace = curinodes = 0;
2614         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2615                 int err;
2616                 tgt = lmv->tgts[i];
2617
2618                 if (tgt == NULL || tgt->ltd_exp == NULL || tgt->ltd_active == 0)
2619                         continue;
2620                 if (!tgt->ltd_active) {
2621                         CDEBUG(D_HA, "mdt %d is inactive.\n", i);
2622                         continue;
2623                 }
2624
2625                 err = obd_quotactl(tgt->ltd_exp, oqctl);
2626                 if (err) {
2627                         CERROR("getquota on mdt %d failed. %d\n", i, err);
2628                         if (!rc)
2629                                 rc = err;
2630                 } else {
2631                         curspace += oqctl->qc_dqblk.dqb_curspace;
2632                         curinodes += oqctl->qc_dqblk.dqb_curinodes;
2633                 }
2634         }
2635         oqctl->qc_dqblk.dqb_curspace = curspace;
2636         oqctl->qc_dqblk.dqb_curinodes = curinodes;
2637
2638         RETURN(rc);
2639 }
2640
2641 int lmv_quotacheck(struct obd_device *unused, struct obd_export *exp,
2642                    struct obd_quotactl *oqctl)
2643 {
2644         struct obd_device   *obd = class_exp2obd(exp);
2645         struct lmv_obd      *lmv = &obd->u.lmv;
2646         struct lmv_tgt_desc *tgt;
2647         int                  i, rc = 0;
2648         ENTRY;
2649
2650         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2651                 int err;
2652                 tgt = lmv->tgts[i];
2653                 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active) {
2654                         CERROR("lmv idx %d inactive\n", i);
2655                         RETURN(-EIO);
2656                 }
2657
2658                 err = obd_quotacheck(tgt->ltd_exp, oqctl);
2659                 if (err && !rc)
2660                         rc = err;
2661         }
2662
2663         RETURN(rc);
2664 }
2665
2666 struct obd_ops lmv_obd_ops = {
2667         .o_owner                = THIS_MODULE,
2668         .o_setup                = lmv_setup,
2669         .o_cleanup              = lmv_cleanup,
2670         .o_precleanup           = lmv_precleanup,
2671         .o_process_config       = lmv_process_config,
2672         .o_connect              = lmv_connect,
2673         .o_disconnect           = lmv_disconnect,
2674         .o_statfs               = lmv_statfs,
2675         .o_get_info             = lmv_get_info,
2676         .o_set_info_async       = lmv_set_info_async,
2677         .o_packmd               = lmv_packmd,
2678         .o_unpackmd             = lmv_unpackmd,
2679         .o_notify               = lmv_notify,
2680         .o_get_uuid             = lmv_get_uuid,
2681         .o_iocontrol            = lmv_iocontrol,
2682         .o_quotacheck           = lmv_quotacheck,
2683         .o_quotactl             = lmv_quotactl
2684 };
2685
2686 struct md_ops lmv_md_ops = {
2687         .m_getstatus            = lmv_getstatus,
2688         .m_null_inode           = lmv_null_inode,
2689         .m_find_cbdata          = lmv_find_cbdata,
2690         .m_close                = lmv_close,
2691         .m_create               = lmv_create,
2692         .m_done_writing         = lmv_done_writing,
2693         .m_enqueue              = lmv_enqueue,
2694         .m_getattr              = lmv_getattr,
2695         .m_getxattr             = lmv_getxattr,
2696         .m_getattr_name         = lmv_getattr_name,
2697         .m_intent_lock          = lmv_intent_lock,
2698         .m_link                 = lmv_link,
2699         .m_rename               = lmv_rename,
2700         .m_setattr              = lmv_setattr,
2701         .m_setxattr             = lmv_setxattr,
2702         .m_sync                 = lmv_sync,
2703         .m_readpage             = lmv_readpage,
2704         .m_unlink               = lmv_unlink,
2705         .m_init_ea_size         = lmv_init_ea_size,
2706         .m_cancel_unused        = lmv_cancel_unused,
2707         .m_set_lock_data        = lmv_set_lock_data,
2708         .m_lock_match           = lmv_lock_match,
2709         .m_get_lustre_md        = lmv_get_lustre_md,
2710         .m_free_lustre_md       = lmv_free_lustre_md,
2711         .m_set_open_replay_data = lmv_set_open_replay_data,
2712         .m_clear_open_replay_data = lmv_clear_open_replay_data,
2713         .m_renew_capa           = lmv_renew_capa,
2714         .m_unpack_capa          = lmv_unpack_capa,
2715         .m_get_remote_perm      = lmv_get_remote_perm,
2716         .m_intent_getattr_async = lmv_intent_getattr_async,
2717         .m_revalidate_lock      = lmv_revalidate_lock
2718 };
2719
2720 int __init lmv_init(void)
2721 {
2722         struct lprocfs_static_vars lvars;
2723         int                        rc;
2724
2725         lprocfs_lmv_init_vars(&lvars);
2726
2727         rc = class_register_type(&lmv_obd_ops, &lmv_md_ops,
2728                                  lvars.module_vars, LUSTRE_LMV_NAME, NULL);
2729         return rc;
2730 }
2731
2732 #ifdef __KERNEL__
2733 static void lmv_exit(void)
2734 {
2735         class_unregister_type(LUSTRE_LMV_NAME);
2736 }
2737
2738 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
2739 MODULE_DESCRIPTION("Lustre Logical Metadata Volume OBD driver");
2740 MODULE_LICENSE("GPL");
2741
2742 module_init(lmv_init);
2743 module_exit(lmv_exit);
2744 #endif