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     *tgt1, *tgt2;
 944
 945                 tgt1 = lmv_find_target(lmv, &op_data->op_fid1);
 946                 tgt2 = lmv_find_target(lmv, &op_data->op_fid2);
 947                 if ((tgt1->ltd_exp == NULL) || (tgt2->ltd_exp == NULL))
 948                         RETURN(-EINVAL);
 949
 950                 /* only files on same MDT can be have their layouts swapped */
 951                 if (tgt1->ltd_idx != tgt2->ltd_idx)
 952                         RETURN(-EPERM);
 953
 954                 rc = obd_iocontrol(cmd, lmv->tgts[0]->ltd_exp, len, karg, uarg);
 955                 break;
 956         }
 957         default:
 958                 for (i = 0; i < count; i++) {
 959                         struct obd_device *mdc_obd;
 960                         int err;
 961
 962                         if (lmv->tgts[i] == NULL ||
 963                             lmv->tgts[i]->ltd_exp == NULL)
 964                                 continue;
 965                         /* ll_umount_begin() sets force flag but for lmv, not
 966                          * mdc. Let's pass it through */
 967                         mdc_obd = class_exp2obd(lmv->tgts[i]->ltd_exp);
 968                         mdc_obd->obd_force = obddev->obd_force;
 969                         err = obd_iocontrol(cmd, lmv->tgts[i]->ltd_exp, len,
 970                                             karg, uarg);
 971                         if (err == -ENODATA && cmd == OBD_IOC_POLL_QUOTACHECK) {
 972                                 RETURN(err);
 973                         } else if (err) {
 974                                 if (lmv->tgts[i]->ltd_active) {
 975                                         CERROR("error: iocontrol MDC %s on MDT"
 976                                                "idx %d cmd %x: err = %d\n",
 977                                                 lmv->tgts[i]->ltd_uuid.uuid,
 978                                                 i, cmd, err);
 979                                         if (!rc)
 980                                                 rc = err;
 981                                 }
 982                         } else
 983                                 set = 1;
 984                 }
 985                 if (!set && !rc)
 986                         rc = -EIO;
 987         }
 988         RETURN(rc);
 989 }
 990
 991 #if 0
 992 static int lmv_all_chars_policy(int count, const char *name,
 993                                 int len)
 994 {
 995         unsigned int c = 0;
 996
 997         while (len > 0)
 998                 c += name[--len];
 999         c = c % count;
1000         return c;
1001 }
1002
1003 static int lmv_nid_policy(struct lmv_obd *lmv)
1004 {
1005         struct obd_import *imp;
1006         __u32              id;
1007
1008         /*
1009          * XXX: To get nid we assume that underlying obd device is mdc.
1010          */
1011         imp = class_exp2cliimp(lmv->tgts[0].ltd_exp);
1012         id = imp->imp_connection->c_self ^ (imp->imp_connection->c_self >> 32);
1013         return id % lmv->desc.ld_tgt_count;
1014 }
1015
1016 static int lmv_choose_mds(struct lmv_obd *lmv, struct md_op_data *op_data,
1017                           placement_policy_t placement)
1018 {
1019         switch (placement) {
1020         case PLACEMENT_CHAR_POLICY:
1021                 return lmv_all_chars_policy(lmv->desc.ld_tgt_count,
1022                                             op_data->op_name,
1023                                             op_data->op_namelen);
1024         case PLACEMENT_NID_POLICY:
1025                 return lmv_nid_policy(lmv);
1026
1027         default:
1028                 break;
1029         }
1030
1031         CERROR("Unsupported placement policy %x\n", placement);
1032         return -EINVAL;
1033 }
1034 #endif
1035
1036 /**
1037  * This is _inode_ placement policy function (not name).
1038  */
1039 static int lmv_placement_policy(struct obd_device *obd,
1040                                 struct md_op_data *op_data,
1041                                 mdsno_t *mds)
1042 {
1043         struct lmv_obd          *lmv = &obd->u.lmv;
1044         ENTRY;
1045
1046         LASSERT(mds != NULL);
1047
1048         if (lmv->desc.ld_tgt_count == 1) {
1049                 *mds = 0;
1050                 RETURN(0);
1051         }
1052
1053         /**
1054          * If stripe_offset is provided during setdirstripe
1055          * (setdirstripe -i xx), xx MDS will be choosen.
1056          */
1057         if (op_data->op_cli_flags & CLI_SET_MEA) {
1058                 struct lmv_user_md *lum;
1059
1060                 lum = (struct lmv_user_md *)op_data->op_data;
1061                 if (lum->lum_type == LMV_STRIPE_TYPE &&
1062                     lum->lum_stripe_offset != -1) {
1063                         if (lum->lum_stripe_offset >= lmv->desc.ld_tgt_count) {
1064                                 CERROR("%s: Stripe_offset %d > MDT count %d:"
1065                                        " rc = %d\n", obd->obd_name,
1066                                        lum->lum_stripe_offset,
1067                                        lmv->desc.ld_tgt_count, -ERANGE);
1068                                 RETURN(-ERANGE);
1069                         }
1070                         *mds = lum->lum_stripe_offset;
1071                         RETURN(0);
1072                 }
1073         }
1074
1075         /* Allocate new fid on target according to operation type and parent
1076          * home mds. */
1077         *mds = op_data->op_mds;
1078         RETURN(0);
1079 }
1080
1081 int __lmv_fid_alloc(struct lmv_obd *lmv, struct lu_fid *fid,
1082                     mdsno_t mds)
1083 {
1084         struct lmv_tgt_desc *tgt;
1085         int                  rc;
1086         ENTRY;
1087
1088         tgt = lmv_get_target(lmv, mds);
1089
1090         /*
1091          * New seq alloc and FLD setup should be atomic. Otherwise we may find
1092          * on server that seq in new allocated fid is not yet known.
1093          */
1094         mutex_lock(&tgt->ltd_fid_mutex);
1095
1096         if (tgt == NULL || tgt->ltd_active == 0 || tgt->ltd_exp == NULL)
1097                 GOTO(out, rc = -ENODEV);
1098
1099         /*
1100          * Asking underlaying tgt layer to allocate new fid.
1101          */
1102         rc = obd_fid_alloc(tgt->ltd_exp, fid, NULL);
1103         if (rc > 0) {
1104                 LASSERT(fid_is_sane(fid));
1105                 rc = 0;
1106         }
1107
1108         EXIT;
1109 out:
1110         mutex_unlock(&tgt->ltd_fid_mutex);
1111         return rc;
1112 }
1113
1114 int lmv_fid_alloc(struct obd_export *exp, struct lu_fid *fid,
1115                   struct md_op_data *op_data)
1116 {
1117         struct obd_device     *obd = class_exp2obd(exp);
1118         struct lmv_obd        *lmv = &obd->u.lmv;
1119         mdsno_t                mds = 0;
1120         int                    rc;
1121         ENTRY;
1122
1123         LASSERT(op_data != NULL);
1124         LASSERT(fid != NULL);
1125
1126         rc = lmv_placement_policy(obd, op_data, &mds);
1127         if (rc) {
1128                 CERROR("Can't get target for allocating fid, "
1129                        "rc %d\n", rc);
1130                 RETURN(rc);
1131         }
1132
1133         rc = __lmv_fid_alloc(lmv, fid, mds);
1134         if (rc) {
1135                 CERROR("Can't alloc new fid, rc %d\n", rc);
1136                 RETURN(rc);
1137         }
1138
1139         RETURN(rc);
1140 }
1141
1142 static int lmv_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
1143 {
1144         struct lmv_obd             *lmv = &obd->u.lmv;
1145         struct lprocfs_static_vars  lvars;
1146         struct lmv_desc            *desc;
1147         int                         rc;
1148         ENTRY;
1149
1150         if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
1151                 CERROR("LMV setup requires a descriptor\n");
1152                 RETURN(-EINVAL);
1153         }
1154
1155         desc = (struct lmv_desc *)lustre_cfg_buf(lcfg, 1);
1156         if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
1157                 CERROR("Lmv descriptor size wrong: %d > %d\n",
1158                        (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
1159                 RETURN(-EINVAL);
1160         }
1161
1162         OBD_ALLOC(lmv->tgts, sizeof(*lmv->tgts) * 32);
1163         if (lmv->tgts == NULL)
1164                 RETURN(-ENOMEM);
1165         lmv->tgts_size = 32;
1166
1167         obd_str2uuid(&lmv->desc.ld_uuid, desc->ld_uuid.uuid);
1168         lmv->desc.ld_tgt_count = 0;
1169         lmv->desc.ld_active_tgt_count = 0;
1170         lmv->max_cookiesize = 0;
1171         lmv->max_def_easize = 0;
1172         lmv->max_easize = 0;
1173         lmv->lmv_placement = PLACEMENT_CHAR_POLICY;
1174
1175         spin_lock_init(&lmv->lmv_lock);
1176         mutex_init(&lmv->init_mutex);
1177
1178         lprocfs_lmv_init_vars(&lvars);
1179
1180         lprocfs_obd_setup(obd, lvars.obd_vars);
1181 #ifdef LPROCFS
1182         {
1183                 rc = lprocfs_seq_create(obd->obd_proc_entry, "target_obd",
1184                                         0444, &lmv_proc_target_fops, obd);
1185                 if (rc)
1186                         CWARN("%s: error adding LMV target_obd file: rc = %d\n",
1187                                obd->obd_name, rc);
1188        }
1189 #endif
1190         rc = fld_client_init(&lmv->lmv_fld, obd->obd_name,
1191                              LUSTRE_CLI_FLD_HASH_DHT);
1192         if (rc) {
1193                 CERROR("Can't init FLD, err %d\n", rc);
1194                 GOTO(out, rc);
1195         }
1196
1197         RETURN(0);
1198
1199 out:
1200         return rc;
1201 }
1202
1203 static int lmv_cleanup(struct obd_device *obd)
1204 {
1205         struct lmv_obd   *lmv = &obd->u.lmv;
1206         ENTRY;
1207
1208         fld_client_fini(&lmv->lmv_fld);
1209         if (lmv->tgts != NULL) {
1210                 int i;
1211                 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1212                         if (lmv->tgts[i] == NULL)
1213                                 continue;
1214                         lmv_del_target(lmv, i);
1215                 }
1216                 OBD_FREE(lmv->tgts, sizeof(*lmv->tgts) * lmv->tgts_size);
1217                 lmv->tgts_size = 0;
1218         }
1219         RETURN(0);
1220 }
1221
1222 static int lmv_process_config(struct obd_device *obd, obd_count len, void *buf)
1223 {
1224         struct lustre_cfg       *lcfg = buf;
1225         struct obd_uuid         obd_uuid;
1226         int                     gen;
1227         __u32                   index;
1228         int                     rc;
1229         ENTRY;
1230
1231         switch (lcfg->lcfg_command) {
1232         case LCFG_ADD_MDC:
1233                 /* modify_mdc_tgts add 0:lustre-clilmv  1:lustre-MDT0000_UUID
1234                  * 2:0  3:1  4:lustre-MDT0000-mdc_UUID */
1235                 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(obd_uuid.uuid))
1236                         GOTO(out, rc = -EINVAL);
1237
1238                 obd_str2uuid(&obd_uuid,  lustre_cfg_buf(lcfg, 1));
1239
1240                 if (sscanf(lustre_cfg_buf(lcfg, 2), "%d", &index) != 1)
1241                         GOTO(out, rc = -EINVAL);
1242                 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", &gen) != 1)
1243                         GOTO(out, rc = -EINVAL);
1244                 rc = lmv_add_target(obd, &obd_uuid, index, gen);
1245                 GOTO(out, rc);
1246         default:
1247                 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
1248                 GOTO(out, rc = -EINVAL);
1249         }
1250 out:
1251         RETURN(rc);
1252 }
1253
1254 static int lmv_statfs(const struct lu_env *env, struct obd_export *exp,
1255                       struct obd_statfs *osfs, __u64 max_age, __u32 flags)
1256 {
1257         struct obd_device     *obd = class_exp2obd(exp);
1258         struct lmv_obd        *lmv = &obd->u.lmv;
1259         struct obd_statfs     *temp;
1260         int                    rc = 0;
1261         int                    i;
1262         ENTRY;
1263
1264         rc = lmv_check_connect(obd);
1265         if (rc)
1266                 RETURN(rc);
1267
1268         OBD_ALLOC(temp, sizeof(*temp));
1269         if (temp == NULL)
1270                 RETURN(-ENOMEM);
1271
1272         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1273                 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
1274                         continue;
1275
1276                 rc = obd_statfs(env, lmv->tgts[i]->ltd_exp, temp,
1277                                 max_age, flags);
1278                 if (rc) {
1279                         CERROR("can't stat MDS #%d (%s), error %d\n", i,
1280                                lmv->tgts[i]->ltd_exp->exp_obd->obd_name,
1281                                rc);
1282                         GOTO(out_free_temp, rc);
1283                 }
1284
1285                 if (i == 0) {
1286                         *osfs = *temp;
1287                         /* If the statfs is from mount, it will needs
1288                          * retrieve necessary information from MDT0.
1289                          * i.e. mount does not need the merged osfs
1290                          * from all of MDT.
1291                          * And also clients can be mounted as long as
1292                          * MDT0 is in service*/
1293                         if (flags & OBD_STATFS_FOR_MDT0)
1294                                 GOTO(out_free_temp, rc);
1295                 } else {
1296                         osfs->os_bavail += temp->os_bavail;
1297                         osfs->os_blocks += temp->os_blocks;
1298                         osfs->os_ffree += temp->os_ffree;
1299                         osfs->os_files += temp->os_files;
1300                 }
1301         }
1302
1303         EXIT;
1304 out_free_temp:
1305         OBD_FREE(temp, sizeof(*temp));
1306         return rc;
1307 }
1308
1309 static int lmv_getstatus(struct obd_export *exp,
1310                          struct lu_fid *fid,
1311                          struct obd_capa **pc)
1312 {
1313         struct obd_device    *obd = exp->exp_obd;
1314         struct lmv_obd       *lmv = &obd->u.lmv;
1315         int                   rc;
1316         ENTRY;
1317
1318         rc = lmv_check_connect(obd);
1319         if (rc)
1320                 RETURN(rc);
1321
1322         rc = md_getstatus(lmv->tgts[0]->ltd_exp, fid, pc);
1323         RETURN(rc);
1324 }
1325
1326 static int lmv_getxattr(struct obd_export *exp, const struct lu_fid *fid,
1327                         struct obd_capa *oc, obd_valid valid, const char *name,
1328                         const char *input, int input_size, int output_size,
1329                         int flags, struct ptlrpc_request **request)
1330 {
1331         struct obd_device      *obd = exp->exp_obd;
1332         struct lmv_obd         *lmv = &obd->u.lmv;
1333         struct lmv_tgt_desc    *tgt;
1334         int                     rc;
1335         ENTRY;
1336
1337         rc = lmv_check_connect(obd);
1338         if (rc)
1339                 RETURN(rc);
1340
1341         tgt = lmv_find_target(lmv, fid);
1342         if (IS_ERR(tgt))
1343                 RETURN(PTR_ERR(tgt));
1344
1345         rc = md_getxattr(tgt->ltd_exp, fid, oc, valid, name, input,
1346                          input_size, output_size, flags, request);
1347
1348         RETURN(rc);
1349 }
1350
1351 static int lmv_setxattr(struct obd_export *exp, const struct lu_fid *fid,
1352                         struct obd_capa *oc, obd_valid valid, const char *name,
1353                         const char *input, int input_size, int output_size,
1354                         int flags, __u32 suppgid,
1355                         struct ptlrpc_request **request)
1356 {
1357         struct obd_device      *obd = exp->exp_obd;
1358         struct lmv_obd         *lmv = &obd->u.lmv;
1359         struct lmv_tgt_desc    *tgt;
1360         int                     rc;
1361         ENTRY;
1362
1363         rc = lmv_check_connect(obd);
1364         if (rc)
1365                 RETURN(rc);
1366
1367         tgt = lmv_find_target(lmv, fid);
1368         if (IS_ERR(tgt))
1369                 RETURN(PTR_ERR(tgt));
1370
1371         rc = md_setxattr(tgt->ltd_exp, fid, oc, valid, name, input,
1372                          input_size, output_size, flags, suppgid,
1373                          request);
1374
1375         RETURN(rc);
1376 }
1377
1378 static int lmv_getattr(struct obd_export *exp, struct md_op_data *op_data,
1379                        struct ptlrpc_request **request)
1380 {
1381         struct obd_device       *obd = exp->exp_obd;
1382         struct lmv_obd          *lmv = &obd->u.lmv;
1383         struct lmv_tgt_desc     *tgt;
1384         int                      rc;
1385         ENTRY;
1386
1387         rc = lmv_check_connect(obd);
1388         if (rc)
1389                 RETURN(rc);
1390
1391         tgt = lmv_find_target(lmv, &op_data->op_fid1);
1392         if (IS_ERR(tgt))
1393                 RETURN(PTR_ERR(tgt));
1394
1395         if (op_data->op_flags & MF_GET_MDT_IDX) {
1396                 op_data->op_mds = tgt->ltd_idx;
1397                 RETURN(0);
1398         }
1399
1400         rc = md_getattr(tgt->ltd_exp, op_data, request);
1401
1402         RETURN(rc);
1403 }
1404
1405 static int lmv_null_inode(struct obd_export *exp, const struct lu_fid *fid)
1406 {
1407         struct obd_device   *obd = exp->exp_obd;
1408         struct lmv_obd      *lmv = &obd->u.lmv;
1409         int                  i;
1410         int                  rc;
1411         ENTRY;
1412
1413         rc = lmv_check_connect(obd);
1414         if (rc)
1415                 RETURN(rc);
1416
1417         CDEBUG(D_INODE, "CBDATA for "DFID"\n", PFID(fid));
1418
1419         /*
1420          * With DNE every object can have two locks in different namespaces:
1421          * lookup lock in space of MDT storing direntry and update/open lock in
1422          * space of MDT storing inode.
1423          */
1424         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1425                 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
1426                         continue;
1427                 md_null_inode(lmv->tgts[i]->ltd_exp, fid);
1428         }
1429
1430         RETURN(0);
1431 }
1432
1433 static int lmv_find_cbdata(struct obd_export *exp, const struct lu_fid *fid,
1434                            ldlm_iterator_t it, void *data)
1435 {
1436         struct obd_device   *obd = exp->exp_obd;
1437         struct lmv_obd      *lmv = &obd->u.lmv;
1438         int                  i;
1439         int                  rc;
1440         ENTRY;
1441
1442         rc = lmv_check_connect(obd);
1443         if (rc)
1444                 RETURN(rc);
1445
1446         CDEBUG(D_INODE, "CBDATA for "DFID"\n", PFID(fid));
1447
1448         /*
1449          * With DNE every object can have two locks in different namespaces:
1450          * lookup lock in space of MDT storing direntry and update/open lock in
1451          * space of MDT storing inode.
1452          */
1453         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1454                 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
1455                         continue;
1456                 rc = md_find_cbdata(lmv->tgts[i]->ltd_exp, fid, it, data);
1457                 if (rc)
1458                         RETURN(rc);
1459         }
1460
1461         RETURN(rc);
1462 }
1463
1464
1465 static int lmv_close(struct obd_export *exp, struct md_op_data *op_data,
1466                      struct md_open_data *mod, struct ptlrpc_request **request)
1467 {
1468         struct obd_device     *obd = exp->exp_obd;
1469         struct lmv_obd        *lmv = &obd->u.lmv;
1470         struct lmv_tgt_desc   *tgt;
1471         int                    rc;
1472         ENTRY;
1473
1474         rc = lmv_check_connect(obd);
1475         if (rc)
1476                 RETURN(rc);
1477
1478         tgt = lmv_find_target(lmv, &op_data->op_fid1);
1479         if (IS_ERR(tgt))
1480                 RETURN(PTR_ERR(tgt));
1481
1482         CDEBUG(D_INODE, "CLOSE "DFID"\n", PFID(&op_data->op_fid1));
1483         rc = md_close(tgt->ltd_exp, op_data, mod, request);
1484         RETURN(rc);
1485 }
1486
1487 struct lmv_tgt_desc
1488 *lmv_locate_mds(struct lmv_obd *lmv, struct md_op_data *op_data,
1489                 struct lu_fid *fid)
1490 {
1491         struct lmv_tgt_desc *tgt;
1492
1493         tgt = lmv_find_target(lmv, fid);
1494         op_data->op_mds = tgt->ltd_idx;
1495
1496         return tgt;
1497 }
1498
1499 int lmv_create(struct obd_export *exp, struct md_op_data *op_data,
1500                const void *data, int datalen, int mode, __u32 uid,
1501                __u32 gid, cfs_cap_t cap_effective, __u64 rdev,
1502                struct ptlrpc_request **request)
1503 {
1504         struct obd_device       *obd = exp->exp_obd;
1505         struct lmv_obd          *lmv = &obd->u.lmv;
1506         struct lmv_tgt_desc     *tgt;
1507         int                      rc;
1508         ENTRY;
1509
1510         rc = lmv_check_connect(obd);
1511         if (rc)
1512                 RETURN(rc);
1513
1514         if (!lmv->desc.ld_active_tgt_count)
1515                 RETURN(-EIO);
1516
1517         tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
1518         if (IS_ERR(tgt))
1519                 RETURN(PTR_ERR(tgt));
1520
1521         rc = lmv_fid_alloc(exp, &op_data->op_fid2, op_data);
1522         if (rc)
1523                 RETURN(rc);
1524
1525         CDEBUG(D_INODE, "CREATE '%*s' on "DFID" -> mds #%x\n",
1526                op_data->op_namelen, op_data->op_name, PFID(&op_data->op_fid1),
1527                op_data->op_mds);
1528
1529         op_data->op_flags |= MF_MDC_CANCEL_FID1;
1530         rc = md_create(tgt->ltd_exp, op_data, data, datalen, mode, uid, gid,
1531                        cap_effective, rdev, request);
1532
1533         if (rc == 0) {
1534                 if (*request == NULL)
1535                         RETURN(rc);
1536                 CDEBUG(D_INODE, "Created - "DFID"\n", PFID(&op_data->op_fid2));
1537         }
1538         RETURN(rc);
1539 }
1540
1541 static int lmv_done_writing(struct obd_export *exp,
1542                             struct md_op_data *op_data,
1543                             struct md_open_data *mod)
1544 {
1545         struct obd_device     *obd = exp->exp_obd;
1546         struct lmv_obd        *lmv = &obd->u.lmv;
1547         struct lmv_tgt_desc   *tgt;
1548         int                    rc;
1549         ENTRY;
1550
1551         rc = lmv_check_connect(obd);
1552         if (rc)
1553                 RETURN(rc);
1554
1555         tgt = lmv_find_target(lmv, &op_data->op_fid1);
1556         if (IS_ERR(tgt))
1557                 RETURN(PTR_ERR(tgt));
1558
1559         rc = md_done_writing(tgt->ltd_exp, op_data, mod);
1560         RETURN(rc);
1561 }
1562
1563 static int
1564 lmv_enqueue_remote(struct obd_export *exp, struct ldlm_enqueue_info *einfo,
1565                    struct lookup_intent *it, struct md_op_data *op_data,
1566                    struct lustre_handle *lockh, void *lmm, int lmmsize,
1567                    int extra_lock_flags)
1568 {
1569         struct ptlrpc_request      *req = it->d.lustre.it_data;
1570         struct obd_device          *obd = exp->exp_obd;
1571         struct lmv_obd             *lmv = &obd->u.lmv;
1572         struct lustre_handle        plock;
1573         struct lmv_tgt_desc        *tgt;
1574         struct md_op_data          *rdata;
1575         struct lu_fid               fid1;
1576         struct mdt_body            *body;
1577         int                         rc = 0;
1578         int                         pmode;
1579         ENTRY;
1580
1581         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1582         LASSERT(body != NULL);
1583
1584         if (!(body->valid & OBD_MD_MDS))
1585                 RETURN(0);
1586
1587         CDEBUG(D_INODE, "REMOTE_ENQUEUE '%s' on "DFID" -> "DFID"\n",
1588                LL_IT2STR(it), PFID(&op_data->op_fid1), PFID(&body->fid1));
1589
1590         /*
1591          * We got LOOKUP lock, but we really need attrs.
1592          */
1593         pmode = it->d.lustre.it_lock_mode;
1594         LASSERT(pmode != 0);
1595         memcpy(&plock, lockh, sizeof(plock));
1596         it->d.lustre.it_lock_mode = 0;
1597         it->d.lustre.it_data = NULL;
1598         fid1 = body->fid1;
1599
1600         it->d.lustre.it_disposition &= ~DISP_ENQ_COMPLETE;
1601         ptlrpc_req_finished(req);
1602
1603         tgt = lmv_find_target(lmv, &fid1);
1604         if (IS_ERR(tgt))
1605                 GOTO(out, rc = PTR_ERR(tgt));
1606
1607         OBD_ALLOC_PTR(rdata);
1608         if (rdata == NULL)
1609                 GOTO(out, rc = -ENOMEM);
1610
1611         rdata->op_fid1 = fid1;
1612         rdata->op_bias = MDS_CROSS_REF;
1613
1614         rc = md_enqueue(tgt->ltd_exp, einfo, it, rdata, lockh,
1615                         lmm, lmmsize, NULL, extra_lock_flags);
1616         OBD_FREE_PTR(rdata);
1617         EXIT;
1618 out:
1619         ldlm_lock_decref(&plock, pmode);
1620         return rc;
1621 }
1622
1623 static int
1624 lmv_enqueue(struct obd_export *exp, struct ldlm_enqueue_info *einfo,
1625             struct lookup_intent *it, struct md_op_data *op_data,
1626             struct lustre_handle *lockh, void *lmm, int lmmsize,
1627             struct ptlrpc_request **req, __u64 extra_lock_flags)
1628 {
1629         struct obd_device        *obd = exp->exp_obd;
1630         struct lmv_obd           *lmv = &obd->u.lmv;
1631         struct lmv_tgt_desc      *tgt;
1632         int                       rc;
1633         ENTRY;
1634
1635         rc = lmv_check_connect(obd);
1636         if (rc)
1637                 RETURN(rc);
1638
1639         CDEBUG(D_INODE, "ENQUEUE '%s' on "DFID"\n",
1640                LL_IT2STR(it), PFID(&op_data->op_fid1));
1641
1642         tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
1643         if (IS_ERR(tgt))
1644                 RETURN(PTR_ERR(tgt));
1645
1646         CDEBUG(D_INODE, "ENQUEUE '%s' on "DFID" -> mds #%d\n",
1647                LL_IT2STR(it), PFID(&op_data->op_fid1), tgt->ltd_idx);
1648
1649         rc = md_enqueue(tgt->ltd_exp, einfo, it, op_data, lockh,
1650                         lmm, lmmsize, req, extra_lock_flags);
1651
1652         if (rc == 0 && it && it->it_op == IT_OPEN) {
1653                 rc = lmv_enqueue_remote(exp, einfo, it, op_data, lockh,
1654                                         lmm, lmmsize, extra_lock_flags);
1655         }
1656         RETURN(rc);
1657 }
1658
1659 static int
1660 lmv_getattr_name(struct obd_export *exp,struct md_op_data *op_data,
1661                  struct ptlrpc_request **request)
1662 {
1663         struct ptlrpc_request   *req = NULL;
1664         struct obd_device       *obd = exp->exp_obd;
1665         struct lmv_obd          *lmv = &obd->u.lmv;
1666         struct lmv_tgt_desc     *tgt;
1667         struct mdt_body         *body;
1668         int                      rc;
1669         ENTRY;
1670
1671         rc = lmv_check_connect(obd);
1672         if (rc)
1673                 RETURN(rc);
1674
1675         tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
1676         if (IS_ERR(tgt))
1677                 RETURN(PTR_ERR(tgt));
1678
1679         CDEBUG(D_INODE, "GETATTR_NAME for %*s on "DFID" -> mds #%d\n",
1680                op_data->op_namelen, op_data->op_name, PFID(&op_data->op_fid1),
1681                tgt->ltd_idx);
1682
1683         rc = md_getattr_name(tgt->ltd_exp, op_data, request);
1684         if (rc != 0)
1685                 RETURN(rc);
1686
1687         body = req_capsule_server_get(&(*request)->rq_pill,
1688                                       &RMF_MDT_BODY);
1689         LASSERT(body != NULL);
1690
1691         if (body->valid & OBD_MD_MDS) {
1692                 struct lu_fid rid = body->fid1;
1693                 CDEBUG(D_INODE, "Request attrs for "DFID"\n",
1694                        PFID(&rid));
1695
1696                 tgt = lmv_find_target(lmv, &rid);
1697                 if (IS_ERR(tgt)) {
1698                         ptlrpc_req_finished(*request);
1699                         RETURN(PTR_ERR(tgt));
1700                 }
1701
1702                 op_data->op_fid1 = rid;
1703                 op_data->op_valid |= OBD_MD_FLCROSSREF;
1704                 op_data->op_namelen = 0;
1705                 op_data->op_name = NULL;
1706                 rc = md_getattr_name(tgt->ltd_exp, op_data, &req);
1707                 ptlrpc_req_finished(*request);
1708                 *request = req;
1709         }
1710
1711         RETURN(rc);
1712 }
1713
1714 #define md_op_data_fid(op_data, fl)                     \
1715         (fl == MF_MDC_CANCEL_FID1 ? &op_data->op_fid1 : \
1716          fl == MF_MDC_CANCEL_FID2 ? &op_data->op_fid2 : \
1717          fl == MF_MDC_CANCEL_FID3 ? &op_data->op_fid3 : \
1718          fl == MF_MDC_CANCEL_FID4 ? &op_data->op_fid4 : \
1719          NULL)
1720
1721 static int lmv_early_cancel(struct obd_export *exp, struct md_op_data *op_data,
1722                             int op_tgt, ldlm_mode_t mode, int bits, int flag)
1723 {
1724         struct lu_fid          *fid = md_op_data_fid(op_data, flag);
1725         struct obd_device      *obd = exp->exp_obd;
1726         struct lmv_obd         *lmv = &obd->u.lmv;
1727         struct lmv_tgt_desc    *tgt;
1728         ldlm_policy_data_t      policy = {{0}};
1729         int                     rc = 0;
1730         ENTRY;
1731
1732         if (!fid_is_sane(fid))
1733                 RETURN(0);
1734
1735         tgt = lmv_find_target(lmv, fid);
1736         if (IS_ERR(tgt))
1737                 RETURN(PTR_ERR(tgt));
1738
1739         if (tgt->ltd_idx != op_tgt) {
1740                 CDEBUG(D_INODE, "EARLY_CANCEL on "DFID"\n", PFID(fid));
1741                 policy.l_inodebits.bits = bits;
1742                 rc = md_cancel_unused(tgt->ltd_exp, fid, &policy,
1743                                       mode, LCF_ASYNC, NULL);
1744         } else {
1745                 CDEBUG(D_INODE,
1746                        "EARLY_CANCEL skip operation target %d on "DFID"\n",
1747                        op_tgt, PFID(fid));
1748                 op_data->op_flags |= flag;
1749                 rc = 0;
1750         }
1751
1752         RETURN(rc);
1753 }
1754
1755 /*
1756  * llite passes fid of an target inode in op_data->op_fid1 and id of directory in
1757  * op_data->op_fid2
1758  */
1759 static int lmv_link(struct obd_export *exp, struct md_op_data *op_data,
1760                     struct ptlrpc_request **request)
1761 {
1762         struct obd_device       *obd = exp->exp_obd;
1763         struct lmv_obd          *lmv = &obd->u.lmv;
1764         struct lmv_tgt_desc     *tgt;
1765         int                      rc;
1766         ENTRY;
1767
1768         rc = lmv_check_connect(obd);
1769         if (rc)
1770                 RETURN(rc);
1771
1772         LASSERT(op_data->op_namelen != 0);
1773
1774         CDEBUG(D_INODE, "LINK "DFID":%*s to "DFID"\n",
1775                PFID(&op_data->op_fid2), op_data->op_namelen,
1776                op_data->op_name, PFID(&op_data->op_fid1));
1777
1778         op_data->op_fsuid = cfs_curproc_fsuid();
1779         op_data->op_fsgid = cfs_curproc_fsgid();
1780         op_data->op_cap = cfs_curproc_cap_pack();
1781         tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid2);
1782         if (IS_ERR(tgt))
1783                 RETURN(PTR_ERR(tgt));
1784
1785         /*
1786          * Cancel UPDATE lock on child (fid1).
1787          */
1788         op_data->op_flags |= MF_MDC_CANCEL_FID2;
1789         rc = lmv_early_cancel(exp, op_data, tgt->ltd_idx, LCK_EX,
1790                               MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID1);
1791         if (rc != 0)
1792                 RETURN(rc);
1793
1794         rc = md_link(tgt->ltd_exp, op_data, request);
1795
1796         RETURN(rc);
1797 }
1798
1799 static int lmv_rename(struct obd_export *exp, struct md_op_data *op_data,
1800                       const char *old, int oldlen, const char *new, int newlen,
1801                       struct ptlrpc_request **request)
1802 {
1803         struct obd_device       *obd = exp->exp_obd;
1804         struct lmv_obd          *lmv = &obd->u.lmv;
1805         struct lmv_tgt_desc     *src_tgt;
1806         struct lmv_tgt_desc     *tgt_tgt;
1807         int                     rc;
1808         ENTRY;
1809
1810         LASSERT(oldlen != 0);
1811
1812         CDEBUG(D_INODE, "RENAME %*s in "DFID" to %*s in "DFID"\n",
1813                oldlen, old, PFID(&op_data->op_fid1),
1814                newlen, new, PFID(&op_data->op_fid2));
1815
1816         rc = lmv_check_connect(obd);
1817         if (rc)
1818                 RETURN(rc);
1819
1820         op_data->op_fsuid = cfs_curproc_fsuid();
1821         op_data->op_fsgid = cfs_curproc_fsgid();
1822         op_data->op_cap = cfs_curproc_cap_pack();
1823         src_tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
1824         if (IS_ERR(src_tgt))
1825                 RETURN(PTR_ERR(src_tgt));
1826
1827         tgt_tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid2);
1828         if (IS_ERR(tgt_tgt))
1829                 RETURN(PTR_ERR(tgt_tgt));
1830         /*
1831          * LOOKUP lock on src child (fid3) should also be cancelled for
1832          * src_tgt in mdc_rename.
1833          */
1834         op_data->op_flags |= MF_MDC_CANCEL_FID1 | MF_MDC_CANCEL_FID3;
1835
1836         /*
1837          * Cancel UPDATE locks on tgt parent (fid2), tgt_tgt is its
1838          * own target.
1839          */
1840         rc = lmv_early_cancel(exp, op_data, src_tgt->ltd_idx,
1841                               LCK_EX, MDS_INODELOCK_UPDATE,
1842                               MF_MDC_CANCEL_FID2);
1843
1844         /*
1845          * Cancel LOOKUP locks on tgt child (fid4) for parent tgt_tgt.
1846          */
1847         if (rc == 0) {
1848                 rc = lmv_early_cancel(exp, op_data, src_tgt->ltd_idx,
1849                                       LCK_EX, MDS_INODELOCK_LOOKUP,
1850                                       MF_MDC_CANCEL_FID4);
1851         }
1852
1853         /*
1854          * Cancel all the locks on tgt child (fid4).
1855          */
1856         if (rc == 0)
1857                 rc = lmv_early_cancel(exp, op_data, src_tgt->ltd_idx,
1858                                       LCK_EX, MDS_INODELOCK_FULL,
1859                                       MF_MDC_CANCEL_FID4);
1860
1861         if (rc == 0)
1862                 rc = md_rename(src_tgt->ltd_exp, op_data, old, oldlen,
1863                                new, newlen, request);
1864         RETURN(rc);
1865 }
1866
1867 static int lmv_setattr(struct obd_export *exp, struct md_op_data *op_data,
1868                        void *ea, int ealen, void *ea2, int ea2len,
1869                        struct ptlrpc_request **request,
1870                        struct md_open_data **mod)
1871 {
1872         struct obd_device       *obd = exp->exp_obd;
1873         struct lmv_obd          *lmv = &obd->u.lmv;
1874         struct lmv_tgt_desc     *tgt;
1875         int                      rc = 0;
1876         ENTRY;
1877
1878         rc = lmv_check_connect(obd);
1879         if (rc)
1880                 RETURN(rc);
1881
1882         CDEBUG(D_INODE, "SETATTR for "DFID", valid 0x%x\n",
1883                PFID(&op_data->op_fid1), op_data->op_attr.ia_valid);
1884
1885         op_data->op_flags |= MF_MDC_CANCEL_FID1;
1886         tgt = lmv_find_target(lmv, &op_data->op_fid1);
1887         if (IS_ERR(tgt))
1888                 RETURN(PTR_ERR(tgt));
1889
1890         rc = md_setattr(tgt->ltd_exp, op_data, ea, ealen, ea2,
1891                         ea2len, request, mod);
1892
1893         RETURN(rc);
1894 }
1895
1896 static int lmv_sync(struct obd_export *exp, const struct lu_fid *fid,
1897                     struct obd_capa *oc, struct ptlrpc_request **request)
1898 {
1899         struct obd_device         *obd = exp->exp_obd;
1900         struct lmv_obd            *lmv = &obd->u.lmv;
1901         struct lmv_tgt_desc       *tgt;
1902         int                        rc;
1903         ENTRY;
1904
1905         rc = lmv_check_connect(obd);
1906         if (rc)
1907                 RETURN(rc);
1908
1909         tgt = lmv_find_target(lmv, fid);
1910         if (IS_ERR(tgt))
1911                 RETURN(PTR_ERR(tgt));
1912
1913         rc = md_sync(tgt->ltd_exp, fid, oc, request);
1914         RETURN(rc);
1915 }
1916
1917 static int lmv_readpage(struct obd_export *exp, struct md_op_data *op_data,
1918                         struct page **pages, struct ptlrpc_request **request)
1919 {
1920         struct obd_device       *obd = exp->exp_obd;
1921         struct lmv_obd          *lmv = &obd->u.lmv;
1922         __u64                    offset = op_data->op_offset;
1923         int                      rc;
1924         int                      i;
1925         /* number of pages read, in CFS_PAGE_SIZE */
1926         int                      nrdpgs;
1927         /* number of pages transferred in LU_PAGE_SIZE */
1928         int                      nlupgs;
1929         struct lmv_tgt_desc     *tgt;
1930         struct lu_dirpage       *dp;
1931         struct lu_dirent        *ent;
1932         ENTRY;
1933
1934         rc = lmv_check_connect(obd);
1935         if (rc)
1936                 RETURN(rc);
1937
1938         CDEBUG(D_INODE, "READPAGE at "LPX64" from "DFID"\n",
1939                offset, PFID(&op_data->op_fid1));
1940
1941         /*
1942          * This case handle directory lookup in clustered metadata case (i.e.
1943          * split directory is located on multiple md servers.)
1944          * each server keeps directory entries for certain range of hashes.
1945          * E.g. we have N server and suppose hash range is 0 to MAX_HASH.
1946          * first server will keep records with hashes [ 0 ... MAX_HASH /N  - 1],
1947          * second one with hashes [MAX_HASH / N ... 2 * MAX_HASH / N] and
1948          * so on....
1949          *      readdir can simply start reading entries from 0 - N server in
1950          * order but that will not scale well as all client will request dir in
1951          * to server in same order.
1952          * Following algorithm does optimization:
1953          * Instead of doing readdir in 1, 2, ...., N order, client with a
1954          * rank R does readdir in R, R + 1, ..., N, 1, ... R - 1 order.
1955          * (every client has rank R)
1956          *      But ll_readdir() expect offset range [0 to MAX_HASH/N) but
1957          * since client ask dir from MDS{R} client has pages with offsets
1958          * [R*MAX_HASH/N ... (R + 1)*MAX_HASH/N] there for we do hash_adj
1959          * on hash  values that we get.
1960          * Since these codes might be still useful for sharded directory, so
1961          * Keeping this code for further reference
1962         if (0) {
1963                 LASSERT(nr > 0);
1964                 seg_size = MAX_HASH_SIZE;
1965                 do_div(seg_size, nr);
1966                 los      = obj->lo_stripes;
1967                 tgt      = lmv_get_target(lmv, los[0].ls_mds);
1968                 rank     = lmv_node_rank(tgt->ltd_exp, fid) % nr;
1969                 tgt_tmp  = offset;
1970                 do_div(tgt_tmp, seg_size);
1971                 tgt0_idx = do_div(tgt_tmp,  nr);
1972                 tgt_idx  = (tgt0_idx + rank) % nr;
1973
1974                 if (tgt_idx < tgt0_idx)
1975                          * Wrap around.
1976                          *
1977                          * Last segment has unusual length due to division
1978                          * rounding.
1979                         hash_adj = MAX_HASH_SIZE - seg_size * nr;
1980                 else
1981                         hash_adj = 0;
1982
1983                 hash_adj += rank * seg_size;
1984
1985                 CDEBUG(D_INODE, "Readpage hash adjustment: %x "LPX64" "
1986                        LPX64"/%x -> "LPX64"/%x\n", rank, hash_adj,
1987                        offset, tgt0_idx, offset + hash_adj, tgt_idx);
1988
1989                 offset = (offset + hash_adj) & MAX_HASH_SIZE;
1990                 rid = lsm->mea_oinfo[tgt_idx].lmo_fid;
1991                 tgt = lmv_get_target(lmv, lsm->mea_oinfo[tgt_idx].lmo_mds);
1992
1993                 CDEBUG(D_INODE, "Forward to "DFID" with offset %lu i %d\n",
1994                        PFID(&rid), (unsigned long)offset, tgt_idx);
1995         }
1996         */
1997         tgt = lmv_find_target(lmv, &op_data->op_fid1);
1998         if (IS_ERR(tgt))
1999                 RETURN(PTR_ERR(tgt));
2000
2001         rc = md_readpage(tgt->ltd_exp, op_data, pages, request);
2002         if (rc != 0)
2003                 RETURN(rc);
2004
2005         nrdpgs = ((*request)->rq_bulk->bd_nob_transferred + CFS_PAGE_SIZE - 1)
2006                  >> CFS_PAGE_SHIFT;
2007         nlupgs = (*request)->rq_bulk->bd_nob_transferred >> LU_PAGE_SHIFT;
2008         LASSERT(!((*request)->rq_bulk->bd_nob_transferred & ~LU_PAGE_MASK));
2009         LASSERT(nrdpgs > 0 && nrdpgs <= op_data->op_npages);
2010
2011         CDEBUG(D_INODE, "read %d(%d)/%d pages\n", nrdpgs, nlupgs,
2012                op_data->op_npages);
2013
2014         for (i = 0; i < nrdpgs; i++) {
2015 #if CFS_PAGE_SIZE > LU_PAGE_SIZE
2016                 struct lu_dirpage *first;
2017                 __u64 hash_end = 0;
2018                 __u32 flags = 0;
2019 #endif
2020                 struct lu_dirent *tmp = NULL;
2021
2022                 dp = cfs_kmap(pages[i]);
2023                 ent = lu_dirent_start(dp);
2024 #if CFS_PAGE_SIZE > LU_PAGE_SIZE
2025                 first = dp;
2026                 hash_end = dp->ldp_hash_end;
2027 repeat:
2028 #endif
2029                 nlupgs--;
2030
2031                 for (tmp = ent; ent != NULL;
2032                      tmp = ent, ent = lu_dirent_next(ent));
2033 #if CFS_PAGE_SIZE > LU_PAGE_SIZE
2034                 dp = (struct lu_dirpage *)((char *)dp + LU_PAGE_SIZE);
2035                 if (((unsigned long)dp & ~CFS_PAGE_MASK) && nlupgs > 0) {
2036                         ent = lu_dirent_start(dp);
2037
2038                         if (tmp) {
2039                                 /* enlarge the end entry lde_reclen from 0 to
2040                                  * first entry of next lu_dirpage, in this way
2041                                  * several lu_dirpages can be stored into one
2042                                  * client page on client. */
2043                                 tmp = ((void *)tmp) +
2044                                       le16_to_cpu(tmp->lde_reclen);
2045                                 tmp->lde_reclen =
2046                                         cpu_to_le16((char *)(dp->ldp_entries) -
2047                                                     (char *)tmp);
2048                                 goto repeat;
2049                         }
2050                 }
2051                 first->ldp_hash_end = hash_end;
2052                 first->ldp_flags &= ~cpu_to_le32(LDF_COLLIDE);
2053                 first->ldp_flags |= flags & cpu_to_le32(LDF_COLLIDE);
2054 #else
2055                 SET_BUT_UNUSED(tmp);
2056 #endif
2057                 cfs_kunmap(pages[i]);
2058         }
2059         RETURN(rc);
2060 }
2061
2062 static int lmv_unlink(struct obd_export *exp, struct md_op_data *op_data,
2063                       struct ptlrpc_request **request)
2064 {
2065         struct obd_device       *obd = exp->exp_obd;
2066         struct lmv_obd          *lmv = &obd->u.lmv;
2067         struct lmv_tgt_desc     *tgt = NULL;
2068         struct mdt_body         *body;
2069         int                     rc;
2070         ENTRY;
2071
2072         rc = lmv_check_connect(obd);
2073         if (rc)
2074                 RETURN(rc);
2075 retry:
2076         /* Send unlink requests to the MDT where the child is located */
2077         if (likely(!fid_is_zero(&op_data->op_fid2)))
2078                 tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid2);
2079         else
2080                 tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
2081         if (IS_ERR(tgt))
2082                 RETURN(PTR_ERR(tgt));
2083
2084         op_data->op_fsuid = cfs_curproc_fsuid();
2085         op_data->op_fsgid = cfs_curproc_fsgid();
2086         op_data->op_cap = cfs_curproc_cap_pack();
2087
2088         /*
2089          * If child's fid is given, cancel unused locks for it if it is from
2090          * another export than parent.
2091          *
2092          * LOOKUP lock for child (fid3) should also be cancelled on parent
2093          * tgt_tgt in mdc_unlink().
2094          */
2095         op_data->op_flags |= MF_MDC_CANCEL_FID1 | MF_MDC_CANCEL_FID3;
2096
2097         /*
2098          * Cancel FULL locks on child (fid3).
2099          */
2100         rc = lmv_early_cancel(exp, op_data, tgt->ltd_idx, LCK_EX,
2101                               MDS_INODELOCK_FULL, MF_MDC_CANCEL_FID3);
2102
2103         if (rc != 0)
2104                 RETURN(rc);
2105
2106         CDEBUG(D_INODE, "unlink with fid="DFID"/"DFID" -> mds #%d\n",
2107                PFID(&op_data->op_fid1), PFID(&op_data->op_fid2), tgt->ltd_idx);
2108
2109         rc = md_unlink(tgt->ltd_exp, op_data, request);
2110         if (rc != 0 && rc != -EREMOTE)
2111                 RETURN(rc);
2112
2113         body = req_capsule_server_get(&(*request)->rq_pill, &RMF_MDT_BODY);
2114         if (body == NULL)
2115                 RETURN(-EPROTO);
2116
2117         /* Not cross-ref case, just get out of here. */
2118         if (likely(!(body->valid & OBD_MD_MDS)))
2119                 RETURN(0);
2120
2121         CDEBUG(D_INODE, "%s: try unlink to another MDT for "DFID"\n",
2122                exp->exp_obd->obd_name, PFID(&body->fid1));
2123
2124         /* This is a remote object, try remote MDT, Note: it may
2125          * try more than 1 time here, Considering following case
2126          * /mnt/lustre is root on MDT0, remote1 is on MDT1
2127          * 1. Initially A does not know where remote1 is, it send
2128          *    unlink RPC to MDT0, MDT0 return -EREMOTE, it will
2129          *    resend unlink RPC to MDT1 (retry 1st time).
2130          *
2131          * 2. During the unlink RPC in flight,
2132          *    client B mv /mnt/lustre/remote1 /mnt/lustre/remote2
2133          *    and create new remote1, but on MDT0
2134          *
2135          * 3. MDT1 get unlink RPC(from A), then do remote lock on
2136          *    /mnt/lustre, then lookup get fid of remote1, and find
2137          *    it is remote dir again, and replay -EREMOTE again.
2138          *
2139          * 4. Then A will resend unlink RPC to MDT0. (retry 2nd times).
2140          *
2141          * In theory, it might try unlimited time here, but it should
2142          * be very rare case.  */
2143         op_data->op_fid2 = body->fid1;
2144         ptlrpc_req_finished(*request);
2145         *request = NULL;
2146
2147         goto retry;
2148 }
2149
2150 static int lmv_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
2151 {
2152         struct lmv_obd *lmv = &obd->u.lmv;
2153         int rc = 0;
2154
2155         switch (stage) {
2156         case OBD_CLEANUP_EARLY:
2157                 /* XXX: here should be calling obd_precleanup() down to
2158                  * stack. */
2159                 break;
2160         case OBD_CLEANUP_EXPORTS:
2161                 fld_client_proc_fini(&lmv->lmv_fld);
2162                 lprocfs_obd_cleanup(obd);
2163                 break;
2164         default:
2165                 break;
2166         }
2167         RETURN(rc);
2168 }
2169
2170 static int lmv_get_info(const struct lu_env *env, struct obd_export *exp,
2171                         __u32 keylen, void *key, __u32 *vallen, void *val,
2172                         struct lov_stripe_md *lsm)
2173 {
2174         struct obd_device       *obd;
2175         struct lmv_obd          *lmv;
2176         int                      rc = 0;
2177         ENTRY;
2178
2179         obd = class_exp2obd(exp);
2180         if (obd == NULL) {
2181                 CDEBUG(D_IOCTL, "Invalid client cookie "LPX64"\n",
2182                        exp->exp_handle.h_cookie);
2183                 RETURN(-EINVAL);
2184         }
2185
2186         lmv = &obd->u.lmv;
2187         if (keylen >= strlen("remote_flag") && !strcmp(key, "remote_flag")) {
2188                 struct lmv_tgt_desc *tgt;
2189                 int i;
2190
2191                 rc = lmv_check_connect(obd);
2192                 if (rc)
2193                         RETURN(rc);
2194
2195                 LASSERT(*vallen == sizeof(__u32));
2196                 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2197                         tgt = lmv->tgts[i];
2198                         /*
2199                          * All tgts should be connected when this gets called.
2200                          */
2201                         if (tgt == NULL || tgt->ltd_exp == NULL)
2202                                 continue;
2203
2204                         if (!obd_get_info(env, tgt->ltd_exp, keylen, key,
2205                                           vallen, val, NULL))
2206                                 RETURN(0);
2207                 }
2208                 RETURN(-EINVAL);
2209         } else if (KEY_IS(KEY_MAX_EASIZE) || KEY_IS(KEY_CONN_DATA)) {
2210                 rc = lmv_check_connect(obd);
2211                 if (rc)
2212                         RETURN(rc);
2213
2214                 /*
2215                  * Forwarding this request to first MDS, it should know LOV
2216                  * desc.
2217                  */
2218                 rc = obd_get_info(env, lmv->tgts[0]->ltd_exp, keylen, key,
2219                                   vallen, val, NULL);
2220                 if (!rc && KEY_IS(KEY_CONN_DATA))
2221                         exp->exp_connect_data = *(struct obd_connect_data *)val;
2222                 RETURN(rc);
2223         } else if (KEY_IS(KEY_TGT_COUNT)) {
2224                 *((int *)val) = lmv->desc.ld_tgt_count;
2225                 RETURN(0);
2226         }
2227
2228         CDEBUG(D_IOCTL, "Invalid key\n");
2229         RETURN(-EINVAL);
2230 }
2231
2232 int lmv_set_info_async(const struct lu_env *env, struct obd_export *exp,
2233                        obd_count keylen, void *key, obd_count vallen,
2234                        void *val, struct ptlrpc_request_set *set)
2235 {
2236         struct lmv_tgt_desc    *tgt;
2237         struct obd_device      *obd;
2238         struct lmv_obd         *lmv;
2239         int rc = 0;
2240         ENTRY;
2241
2242         obd = class_exp2obd(exp);
2243         if (obd == NULL) {
2244                 CDEBUG(D_IOCTL, "Invalid client cookie "LPX64"\n",
2245                        exp->exp_handle.h_cookie);
2246                 RETURN(-EINVAL);
2247         }
2248         lmv = &obd->u.lmv;
2249
2250         if (KEY_IS(KEY_READ_ONLY) || KEY_IS(KEY_FLUSH_CTX)) {
2251                 int i, err = 0;
2252
2253                 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2254                         tgt = lmv->tgts[i];
2255
2256                         if (tgt == NULL || tgt->ltd_exp == NULL)
2257                                 continue;
2258
2259                         err = obd_set_info_async(env, tgt->ltd_exp,
2260                                                  keylen, key, vallen, val, set);
2261                         if (err && rc == 0)
2262                                 rc = err;
2263                 }
2264
2265                 RETURN(rc);
2266         }
2267
2268         RETURN(-EINVAL);
2269 }
2270
2271 int lmv_packmd(struct obd_export *exp, struct lov_mds_md **lmmp,
2272                struct lov_stripe_md *lsm)
2273 {
2274         struct obd_device         *obd = class_exp2obd(exp);
2275         struct lmv_obd            *lmv = &obd->u.lmv;
2276         struct lmv_stripe_md      *meap;
2277         struct lmv_stripe_md      *lsmp;
2278         int                        mea_size;
2279         int                        i;
2280         ENTRY;
2281
2282         mea_size = lmv_get_easize(lmv);
2283         if (!lmmp)
2284                 RETURN(mea_size);
2285
2286         if (*lmmp && !lsm) {
2287                 OBD_FREE_LARGE(*lmmp, mea_size);
2288                 *lmmp = NULL;
2289                 RETURN(0);
2290         }
2291
2292         if (*lmmp == NULL) {
2293                 OBD_ALLOC_LARGE(*lmmp, mea_size);
2294                 if (*lmmp == NULL)
2295                         RETURN(-ENOMEM);
2296         }
2297
2298         if (!lsm)
2299                 RETURN(mea_size);
2300
2301         lsmp = (struct lmv_stripe_md *)lsm;
2302         meap = (struct lmv_stripe_md *)*lmmp;
2303
2304         if (lsmp->mea_magic != MEA_MAGIC_LAST_CHAR &&
2305             lsmp->mea_magic != MEA_MAGIC_ALL_CHARS)
2306                 RETURN(-EINVAL);
2307
2308         meap->mea_magic = cpu_to_le32(lsmp->mea_magic);
2309         meap->mea_count = cpu_to_le32(lsmp->mea_count);
2310         meap->mea_master = cpu_to_le32(lsmp->mea_master);
2311
2312         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2313                 meap->mea_ids[i] = lsmp->mea_ids[i];
2314                 fid_cpu_to_le(&meap->mea_ids[i], &lsmp->mea_ids[i]);
2315         }
2316
2317         RETURN(mea_size);
2318 }
2319
2320 int lmv_unpackmd(struct obd_export *exp, struct lov_stripe_md **lsmp,
2321                  struct lov_mds_md *lmm, int lmm_size)
2322 {
2323         struct obd_device          *obd = class_exp2obd(exp);
2324         struct lmv_stripe_md      **tmea = (struct lmv_stripe_md **)lsmp;
2325         struct lmv_stripe_md       *mea = (struct lmv_stripe_md *)lmm;
2326         struct lmv_obd             *lmv = &obd->u.lmv;
2327         int                         mea_size;
2328         int                         i;
2329         __u32                       magic;
2330         ENTRY;
2331
2332         mea_size = lmv_get_easize(lmv);
2333         if (lsmp == NULL)
2334                 return mea_size;
2335
2336         if (*lsmp != NULL && lmm == NULL) {
2337                 OBD_FREE_LARGE(*tmea, mea_size);
2338                 *lsmp = NULL;
2339                 RETURN(0);
2340         }
2341
2342         LASSERT(mea_size == lmm_size);
2343
2344         OBD_ALLOC_LARGE(*tmea, mea_size);
2345         if (*tmea == NULL)
2346                 RETURN(-ENOMEM);
2347
2348         if (!lmm)
2349                 RETURN(mea_size);
2350
2351         if (mea->mea_magic == MEA_MAGIC_LAST_CHAR ||
2352             mea->mea_magic == MEA_MAGIC_ALL_CHARS ||
2353             mea->mea_magic == MEA_MAGIC_HASH_SEGMENT)
2354         {
2355                 magic = le32_to_cpu(mea->mea_magic);
2356         } else {
2357                 /*
2358                  * Old mea is not handled here.
2359                  */
2360                 CERROR("Old not supportable EA is found\n");
2361                 LBUG();
2362         }
2363
2364         (*tmea)->mea_magic = magic;
2365         (*tmea)->mea_count = le32_to_cpu(mea->mea_count);
2366         (*tmea)->mea_master = le32_to_cpu(mea->mea_master);
2367
2368         for (i = 0; i < (*tmea)->mea_count; i++) {
2369                 (*tmea)->mea_ids[i] = mea->mea_ids[i];
2370                 fid_le_to_cpu(&(*tmea)->mea_ids[i], &(*tmea)->mea_ids[i]);
2371         }
2372         RETURN(mea_size);
2373 }
2374
2375 static int lmv_cancel_unused(struct obd_export *exp, const struct lu_fid *fid,
2376                              ldlm_policy_data_t *policy, ldlm_mode_t mode,
2377                              ldlm_cancel_flags_t flags, void *opaque)
2378 {
2379         struct obd_device       *obd = exp->exp_obd;
2380         struct lmv_obd          *lmv = &obd->u.lmv;
2381         int                      rc = 0;
2382         int                      err;
2383         int                      i;
2384         ENTRY;
2385
2386         LASSERT(fid != NULL);
2387
2388         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2389                 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL ||
2390                     lmv->tgts[i]->ltd_active == 0)
2391                         continue;
2392
2393                 err = md_cancel_unused(lmv->tgts[i]->ltd_exp, fid,
2394                                        policy, mode, flags, opaque);
2395                 if (!rc)
2396                         rc = err;
2397         }
2398         RETURN(rc);
2399 }
2400
2401 int lmv_set_lock_data(struct obd_export *exp, __u64 *lockh, void *data,
2402                       __u64 *bits)
2403 {
2404         struct lmv_obd          *lmv = &exp->exp_obd->u.lmv;
2405         int                      rc;
2406         ENTRY;
2407
2408         rc =  md_set_lock_data(lmv->tgts[0]->ltd_exp, lockh, data, bits);
2409         RETURN(rc);
2410 }
2411
2412 ldlm_mode_t lmv_lock_match(struct obd_export *exp, __u64 flags,
2413                            const struct lu_fid *fid, ldlm_type_t type,
2414                            ldlm_policy_data_t *policy, ldlm_mode_t mode,
2415                            struct lustre_handle *lockh)
2416 {
2417         struct obd_device       *obd = exp->exp_obd;
2418         struct lmv_obd          *lmv = &obd->u.lmv;
2419         ldlm_mode_t              rc;
2420         int                      i;
2421         ENTRY;
2422
2423         CDEBUG(D_INODE, "Lock match for "DFID"\n", PFID(fid));
2424
2425         /*
2426          * With CMD every object can have two locks in different namespaces:
2427          * lookup lock in space of mds storing direntry and update/open lock in
2428          * space of mds storing inode. Thus we check all targets, not only that
2429          * one fid was created in.
2430          */
2431         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2432                 if (lmv->tgts[i] == NULL ||
2433                     lmv->tgts[i]->ltd_exp == NULL ||
2434                     lmv->tgts[i]->ltd_active == 0)
2435                         continue;
2436
2437                 rc = md_lock_match(lmv->tgts[i]->ltd_exp, flags, fid,
2438                                    type, policy, mode, lockh);
2439                 if (rc)
2440                         RETURN(rc);
2441         }
2442
2443         RETURN(0);
2444 }
2445
2446 int lmv_get_lustre_md(struct obd_export *exp, struct ptlrpc_request *req,
2447                       struct obd_export *dt_exp, struct obd_export *md_exp,
2448                       struct lustre_md *md)
2449 {
2450         struct lmv_obd          *lmv = &exp->exp_obd->u.lmv;
2451
2452         return md_get_lustre_md(lmv->tgts[0]->ltd_exp, req, dt_exp, md_exp, md);
2453 }
2454
2455 int lmv_free_lustre_md(struct obd_export *exp, struct lustre_md *md)
2456 {
2457         struct obd_device       *obd = exp->exp_obd;
2458         struct lmv_obd          *lmv = &obd->u.lmv;
2459         ENTRY;
2460
2461         if (md->mea)
2462                 obd_free_memmd(exp, (void *)&md->mea);
2463         RETURN(md_free_lustre_md(lmv->tgts[0]->ltd_exp, md));
2464 }
2465
2466 int lmv_set_open_replay_data(struct obd_export *exp,
2467                              struct obd_client_handle *och,
2468                              struct ptlrpc_request *open_req)
2469 {
2470         struct obd_device       *obd = exp->exp_obd;
2471         struct lmv_obd          *lmv = &obd->u.lmv;
2472         struct lmv_tgt_desc     *tgt;
2473         ENTRY;
2474
2475         tgt = lmv_find_target(lmv, &och->och_fid);
2476         if (IS_ERR(tgt))
2477                 RETURN(PTR_ERR(tgt));
2478
2479         RETURN(md_set_open_replay_data(tgt->ltd_exp, och, open_req));
2480 }
2481
2482 int lmv_clear_open_replay_data(struct obd_export *exp,
2483                                struct obd_client_handle *och)
2484 {
2485         struct obd_device       *obd = exp->exp_obd;
2486         struct lmv_obd          *lmv = &obd->u.lmv;
2487         struct lmv_tgt_desc     *tgt;
2488         ENTRY;
2489
2490         tgt = lmv_find_target(lmv, &och->och_fid);
2491         if (IS_ERR(tgt))
2492                 RETURN(PTR_ERR(tgt));
2493
2494         RETURN(md_clear_open_replay_data(tgt->ltd_exp, och));
2495 }
2496
2497 static int lmv_get_remote_perm(struct obd_export *exp,
2498                                const struct lu_fid *fid,
2499                                struct obd_capa *oc, __u32 suppgid,
2500                                struct ptlrpc_request **request)
2501 {
2502         struct obd_device       *obd = exp->exp_obd;
2503         struct lmv_obd          *lmv = &obd->u.lmv;
2504         struct lmv_tgt_desc     *tgt;
2505         int                      rc;
2506         ENTRY;
2507
2508         rc = lmv_check_connect(obd);
2509         if (rc)
2510                 RETURN(rc);
2511
2512         tgt = lmv_find_target(lmv, fid);
2513         if (IS_ERR(tgt))
2514                 RETURN(PTR_ERR(tgt));
2515
2516         rc = md_get_remote_perm(tgt->ltd_exp, fid, oc, suppgid, request);
2517         RETURN(rc);
2518 }
2519
2520 static int lmv_renew_capa(struct obd_export *exp, struct obd_capa *oc,
2521                           renew_capa_cb_t cb)
2522 {
2523         struct obd_device       *obd = exp->exp_obd;
2524         struct lmv_obd          *lmv = &obd->u.lmv;
2525         struct lmv_tgt_desc     *tgt;
2526         int                      rc;
2527         ENTRY;
2528
2529         rc = lmv_check_connect(obd);
2530         if (rc)
2531                 RETURN(rc);
2532
2533         tgt = lmv_find_target(lmv, &oc->c_capa.lc_fid);
2534         if (IS_ERR(tgt))
2535                 RETURN(PTR_ERR(tgt));
2536
2537         rc = md_renew_capa(tgt->ltd_exp, oc, cb);
2538         RETURN(rc);
2539 }
2540
2541 int lmv_unpack_capa(struct obd_export *exp, struct ptlrpc_request *req,
2542                     const struct req_msg_field *field, struct obd_capa **oc)
2543 {
2544         struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
2545
2546         return md_unpack_capa(lmv->tgts[0]->ltd_exp, req, field, oc);
2547 }
2548
2549 int lmv_intent_getattr_async(struct obd_export *exp,
2550                              struct md_enqueue_info *minfo,
2551                              struct ldlm_enqueue_info *einfo)
2552 {
2553         struct md_op_data       *op_data = &minfo->mi_data;
2554         struct obd_device       *obd = exp->exp_obd;
2555         struct lmv_obd          *lmv = &obd->u.lmv;
2556         struct lmv_tgt_desc     *tgt = NULL;
2557         int                      rc;
2558         ENTRY;
2559
2560         rc = lmv_check_connect(obd);
2561         if (rc)
2562                 RETURN(rc);
2563
2564         tgt = lmv_find_target(lmv, &op_data->op_fid1);
2565         if (IS_ERR(tgt))
2566                 RETURN(PTR_ERR(tgt));
2567
2568         rc = md_intent_getattr_async(tgt->ltd_exp, minfo, einfo);
2569         RETURN(rc);
2570 }
2571
2572 int lmv_revalidate_lock(struct obd_export *exp, struct lookup_intent *it,
2573                         struct lu_fid *fid, __u64 *bits)
2574 {
2575         struct obd_device       *obd = exp->exp_obd;
2576         struct lmv_obd          *lmv = &obd->u.lmv;
2577         struct lmv_tgt_desc     *tgt;
2578         int                      rc;
2579         ENTRY;
2580
2581         rc = lmv_check_connect(obd);
2582         if (rc)
2583                 RETURN(rc);
2584
2585         tgt = lmv_find_target(lmv, fid);
2586         if (IS_ERR(tgt))
2587                 RETURN(PTR_ERR(tgt));
2588
2589         rc = md_revalidate_lock(tgt->ltd_exp, it, fid, bits);
2590         RETURN(rc);
2591 }
2592
2593 /**
2594  * For lmv, only need to send request to master MDT, and the master MDT will
2595  * process with other slave MDTs. The only exception is Q_GETOQUOTA for which
2596  * we directly fetch data from the slave MDTs.
2597  */
2598 int lmv_quotactl(struct obd_device *unused, struct obd_export *exp,
2599                  struct obd_quotactl *oqctl)
2600 {
2601         struct obd_device   *obd = class_exp2obd(exp);
2602         struct lmv_obd      *lmv = &obd->u.lmv;
2603         struct lmv_tgt_desc *tgt = lmv->tgts[0];
2604         int                  rc = 0, i;
2605         __u64                curspace, curinodes;
2606         ENTRY;
2607
2608         if (!lmv->desc.ld_tgt_count || !tgt->ltd_active) {
2609                 CERROR("master lmv inactive\n");
2610                 RETURN(-EIO);
2611         }
2612
2613         if (oqctl->qc_cmd != Q_GETOQUOTA) {
2614                 rc = obd_quotactl(tgt->ltd_exp, oqctl);
2615                 RETURN(rc);
2616         }
2617
2618         curspace = curinodes = 0;
2619         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2620                 int err;
2621                 tgt = lmv->tgts[i];
2622
2623                 if (tgt == NULL || tgt->ltd_exp == NULL || tgt->ltd_active == 0)
2624                         continue;
2625                 if (!tgt->ltd_active) {
2626                         CDEBUG(D_HA, "mdt %d is inactive.\n", i);
2627                         continue;
2628                 }
2629
2630                 err = obd_quotactl(tgt->ltd_exp, oqctl);
2631                 if (err) {
2632                         CERROR("getquota on mdt %d failed. %d\n", i, err);
2633                         if (!rc)
2634                                 rc = err;
2635                 } else {
2636                         curspace += oqctl->qc_dqblk.dqb_curspace;
2637                         curinodes += oqctl->qc_dqblk.dqb_curinodes;
2638                 }
2639         }
2640         oqctl->qc_dqblk.dqb_curspace = curspace;
2641         oqctl->qc_dqblk.dqb_curinodes = curinodes;
2642
2643         RETURN(rc);
2644 }
2645
2646 int lmv_quotacheck(struct obd_device *unused, struct obd_export *exp,
2647                    struct obd_quotactl *oqctl)
2648 {
2649         struct obd_device   *obd = class_exp2obd(exp);
2650         struct lmv_obd      *lmv = &obd->u.lmv;
2651         struct lmv_tgt_desc *tgt;
2652         int                  i, rc = 0;
2653         ENTRY;
2654
2655         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2656                 int err;
2657                 tgt = lmv->tgts[i];
2658                 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active) {
2659                         CERROR("lmv idx %d inactive\n", i);
2660                         RETURN(-EIO);
2661                 }
2662
2663                 err = obd_quotacheck(tgt->ltd_exp, oqctl);
2664                 if (err && !rc)
2665                         rc = err;
2666         }
2667
2668         RETURN(rc);
2669 }
2670
2671 struct obd_ops lmv_obd_ops = {
2672         .o_owner                = THIS_MODULE,
2673         .o_setup                = lmv_setup,
2674         .o_cleanup              = lmv_cleanup,
2675         .o_precleanup           = lmv_precleanup,
2676         .o_process_config       = lmv_process_config,
2677         .o_connect              = lmv_connect,
2678         .o_disconnect           = lmv_disconnect,
2679         .o_statfs               = lmv_statfs,
2680         .o_get_info             = lmv_get_info,
2681         .o_set_info_async       = lmv_set_info_async,
2682         .o_packmd               = lmv_packmd,
2683         .o_unpackmd             = lmv_unpackmd,
2684         .o_notify               = lmv_notify,
2685         .o_get_uuid             = lmv_get_uuid,
2686         .o_iocontrol            = lmv_iocontrol,
2687         .o_quotacheck           = lmv_quotacheck,
2688         .o_quotactl             = lmv_quotactl
2689 };
2690
2691 struct md_ops lmv_md_ops = {
2692         .m_getstatus            = lmv_getstatus,
2693         .m_null_inode           = lmv_null_inode,
2694         .m_find_cbdata          = lmv_find_cbdata,
2695         .m_close                = lmv_close,
2696         .m_create               = lmv_create,
2697         .m_done_writing         = lmv_done_writing,
2698         .m_enqueue              = lmv_enqueue,
2699         .m_getattr              = lmv_getattr,
2700         .m_getxattr             = lmv_getxattr,
2701         .m_getattr_name         = lmv_getattr_name,
2702         .m_intent_lock          = lmv_intent_lock,
2703         .m_link                 = lmv_link,
2704         .m_rename               = lmv_rename,
2705         .m_setattr              = lmv_setattr,
2706         .m_setxattr             = lmv_setxattr,
2707         .m_sync                 = lmv_sync,
2708         .m_readpage             = lmv_readpage,
2709         .m_unlink               = lmv_unlink,
2710         .m_init_ea_size         = lmv_init_ea_size,
2711         .m_cancel_unused        = lmv_cancel_unused,
2712         .m_set_lock_data        = lmv_set_lock_data,
2713         .m_lock_match           = lmv_lock_match,
2714         .m_get_lustre_md        = lmv_get_lustre_md,
2715         .m_free_lustre_md       = lmv_free_lustre_md,
2716         .m_set_open_replay_data = lmv_set_open_replay_data,
2717         .m_clear_open_replay_data = lmv_clear_open_replay_data,
2718         .m_renew_capa           = lmv_renew_capa,
2719         .m_unpack_capa          = lmv_unpack_capa,
2720         .m_get_remote_perm      = lmv_get_remote_perm,
2721         .m_intent_getattr_async = lmv_intent_getattr_async,
2722         .m_revalidate_lock      = lmv_revalidate_lock
2723 };
2724
2725 int __init lmv_init(void)
2726 {
2727         struct lprocfs_static_vars lvars;
2728         int                        rc;
2729
2730         lprocfs_lmv_init_vars(&lvars);
2731
2732         rc = class_register_type(&lmv_obd_ops, &lmv_md_ops,
2733                                  lvars.module_vars, LUSTRE_LMV_NAME, NULL);
2734         return rc;
2735 }
2736
2737 #ifdef __KERNEL__
2738 static void lmv_exit(void)
2739 {
2740         class_unregister_type(LUSTRE_LMV_NAME);
2741 }
2742
2743 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
2744 MODULE_DESCRIPTION("Lustre Logical Metadata Volume OBD driver");
2745 MODULE_LICENSE("GPL");
2746
2747 module_init(lmv_init);
2748 module_exit(lmv_exit);
2749 #endif