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                 if (IS_ERR(tgt1))
 947                         RETURN(PTR_ERR(tgt1));
 948
 949                 tgt2 = lmv_find_target(lmv, &op_data->op_fid2);
 950                 if (IS_ERR(tgt2))
 951                         RETURN(PTR_ERR(tgt2));
 952
 953                 if ((tgt1->ltd_exp == NULL) || (tgt2->ltd_exp == NULL))
 954                         RETURN(-EINVAL);
 955
 956                 /* only files on same MDT can have their layouts swapped */
 957                 if (tgt1->ltd_idx != tgt2->ltd_idx)
 958                         RETURN(-EPERM);
 959
 960                 rc = obd_iocontrol(cmd, tgt1->ltd_exp, len, karg, uarg);
 961                 break;
 962         }
 963         default:
 964                 for (i = 0; i < count; i++) {
 965                         struct obd_device *mdc_obd;
 966                         int err;
 967
 968                         if (lmv->tgts[i] == NULL ||
 969                             lmv->tgts[i]->ltd_exp == NULL)
 970                                 continue;
 971                         /* ll_umount_begin() sets force flag but for lmv, not
 972                          * mdc. Let's pass it through */
 973                         mdc_obd = class_exp2obd(lmv->tgts[i]->ltd_exp);
 974                         mdc_obd->obd_force = obddev->obd_force;
 975                         err = obd_iocontrol(cmd, lmv->tgts[i]->ltd_exp, len,
 976                                             karg, uarg);
 977                         if (err == -ENODATA && cmd == OBD_IOC_POLL_QUOTACHECK) {
 978                                 RETURN(err);
 979                         } else if (err) {
 980                                 if (lmv->tgts[i]->ltd_active) {
 981                                         CERROR("error: iocontrol MDC %s on MDT"
 982                                                "idx %d cmd %x: err = %d\n",
 983                                                 lmv->tgts[i]->ltd_uuid.uuid,
 984                                                 i, cmd, err);
 985                                         if (!rc)
 986                                                 rc = err;
 987                                 }
 988                         } else
 989                                 set = 1;
 990                 }
 991                 if (!set && !rc)
 992                         rc = -EIO;
 993         }
 994         RETURN(rc);
 995 }
 996
 997 #if 0
 998 static int lmv_all_chars_policy(int count, const char *name,
 999                                 int len)
1000 {
1001         unsigned int c = 0;
1002
1003         while (len > 0)
1004                 c += name[--len];
1005         c = c % count;
1006         return c;
1007 }
1008
1009 static int lmv_nid_policy(struct lmv_obd *lmv)
1010 {
1011         struct obd_import *imp;
1012         __u32              id;
1013
1014         /*
1015          * XXX: To get nid we assume that underlying obd device is mdc.
1016          */
1017         imp = class_exp2cliimp(lmv->tgts[0].ltd_exp);
1018         id = imp->imp_connection->c_self ^ (imp->imp_connection->c_self >> 32);
1019         return id % lmv->desc.ld_tgt_count;
1020 }
1021
1022 static int lmv_choose_mds(struct lmv_obd *lmv, struct md_op_data *op_data,
1023                           placement_policy_t placement)
1024 {
1025         switch (placement) {
1026         case PLACEMENT_CHAR_POLICY:
1027                 return lmv_all_chars_policy(lmv->desc.ld_tgt_count,
1028                                             op_data->op_name,
1029                                             op_data->op_namelen);
1030         case PLACEMENT_NID_POLICY:
1031                 return lmv_nid_policy(lmv);
1032
1033         default:
1034                 break;
1035         }
1036
1037         CERROR("Unsupported placement policy %x\n", placement);
1038         return -EINVAL;
1039 }
1040 #endif
1041
1042 /**
1043  * This is _inode_ placement policy function (not name).
1044  */
1045 static int lmv_placement_policy(struct obd_device *obd,
1046                                 struct md_op_data *op_data,
1047                                 mdsno_t *mds)
1048 {
1049         struct lmv_obd          *lmv = &obd->u.lmv;
1050         ENTRY;
1051
1052         LASSERT(mds != NULL);
1053
1054         if (lmv->desc.ld_tgt_count == 1) {
1055                 *mds = 0;
1056                 RETURN(0);
1057         }
1058
1059         /**
1060          * If stripe_offset is provided during setdirstripe
1061          * (setdirstripe -i xx), xx MDS will be choosen.
1062          */
1063         if (op_data->op_cli_flags & CLI_SET_MEA) {
1064                 struct lmv_user_md *lum;
1065
1066                 lum = (struct lmv_user_md *)op_data->op_data;
1067                 if (lum->lum_type == LMV_STRIPE_TYPE &&
1068                     lum->lum_stripe_offset != -1) {
1069                         if (lum->lum_stripe_offset >= lmv->desc.ld_tgt_count) {
1070                                 CERROR("%s: Stripe_offset %d > MDT count %d:"
1071                                        " rc = %d\n", obd->obd_name,
1072                                        lum->lum_stripe_offset,
1073                                        lmv->desc.ld_tgt_count, -ERANGE);
1074                                 RETURN(-ERANGE);
1075                         }
1076                         *mds = lum->lum_stripe_offset;
1077                         RETURN(0);
1078                 }
1079         }
1080
1081         /* Allocate new fid on target according to operation type and parent
1082          * home mds. */
1083         *mds = op_data->op_mds;
1084         RETURN(0);
1085 }
1086
1087 int __lmv_fid_alloc(struct lmv_obd *lmv, struct lu_fid *fid,
1088                     mdsno_t mds)
1089 {
1090         struct lmv_tgt_desc     *tgt;
1091         int                      rc;
1092         ENTRY;
1093
1094         tgt = lmv_get_target(lmv, mds);
1095         if (IS_ERR(tgt))
1096                 RETURN(PTR_ERR(tgt));
1097
1098         /*
1099          * New seq alloc and FLD setup should be atomic. Otherwise we may find
1100          * on server that seq in new allocated fid is not yet known.
1101          */
1102         mutex_lock(&tgt->ltd_fid_mutex);
1103
1104         if (tgt->ltd_active == 0 || tgt->ltd_exp == NULL)
1105                 GOTO(out, rc = -ENODEV);
1106
1107         /*
1108          * Asking underlaying tgt layer to allocate new fid.
1109          */
1110         rc = obd_fid_alloc(tgt->ltd_exp, fid, NULL);
1111         if (rc > 0) {
1112                 LASSERT(fid_is_sane(fid));
1113                 rc = 0;
1114         }
1115
1116         EXIT;
1117 out:
1118         mutex_unlock(&tgt->ltd_fid_mutex);
1119         return rc;
1120 }
1121
1122 int lmv_fid_alloc(struct obd_export *exp, struct lu_fid *fid,
1123                   struct md_op_data *op_data)
1124 {
1125         struct obd_device     *obd = class_exp2obd(exp);
1126         struct lmv_obd        *lmv = &obd->u.lmv;
1127         mdsno_t                mds = 0;
1128         int                    rc;
1129         ENTRY;
1130
1131         LASSERT(op_data != NULL);
1132         LASSERT(fid != NULL);
1133
1134         rc = lmv_placement_policy(obd, op_data, &mds);
1135         if (rc) {
1136                 CERROR("Can't get target for allocating fid, "
1137                        "rc %d\n", rc);
1138                 RETURN(rc);
1139         }
1140
1141         rc = __lmv_fid_alloc(lmv, fid, mds);
1142         if (rc) {
1143                 CERROR("Can't alloc new fid, rc %d\n", rc);
1144                 RETURN(rc);
1145         }
1146
1147         RETURN(rc);
1148 }
1149
1150 static int lmv_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
1151 {
1152         struct lmv_obd             *lmv = &obd->u.lmv;
1153         struct lprocfs_static_vars  lvars;
1154         struct lmv_desc            *desc;
1155         int                         rc;
1156         ENTRY;
1157
1158         if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
1159                 CERROR("LMV setup requires a descriptor\n");
1160                 RETURN(-EINVAL);
1161         }
1162
1163         desc = (struct lmv_desc *)lustre_cfg_buf(lcfg, 1);
1164         if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
1165                 CERROR("Lmv descriptor size wrong: %d > %d\n",
1166                        (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
1167                 RETURN(-EINVAL);
1168         }
1169
1170         OBD_ALLOC(lmv->tgts, sizeof(*lmv->tgts) * 32);
1171         if (lmv->tgts == NULL)
1172                 RETURN(-ENOMEM);
1173         lmv->tgts_size = 32;
1174
1175         obd_str2uuid(&lmv->desc.ld_uuid, desc->ld_uuid.uuid);
1176         lmv->desc.ld_tgt_count = 0;
1177         lmv->desc.ld_active_tgt_count = 0;
1178         lmv->max_cookiesize = 0;
1179         lmv->max_def_easize = 0;
1180         lmv->max_easize = 0;
1181         lmv->lmv_placement = PLACEMENT_CHAR_POLICY;
1182
1183         spin_lock_init(&lmv->lmv_lock);
1184         mutex_init(&lmv->init_mutex);
1185
1186         lprocfs_lmv_init_vars(&lvars);
1187
1188         lprocfs_obd_setup(obd, lvars.obd_vars);
1189 #ifdef LPROCFS
1190         {
1191                 rc = lprocfs_seq_create(obd->obd_proc_entry, "target_obd",
1192                                         0444, &lmv_proc_target_fops, obd);
1193                 if (rc)
1194                         CWARN("%s: error adding LMV target_obd file: rc = %d\n",
1195                                obd->obd_name, rc);
1196        }
1197 #endif
1198         rc = fld_client_init(&lmv->lmv_fld, obd->obd_name,
1199                              LUSTRE_CLI_FLD_HASH_DHT);
1200         if (rc) {
1201                 CERROR("Can't init FLD, err %d\n", rc);
1202                 GOTO(out, rc);
1203         }
1204
1205         RETURN(0);
1206
1207 out:
1208         return rc;
1209 }
1210
1211 static int lmv_cleanup(struct obd_device *obd)
1212 {
1213         struct lmv_obd   *lmv = &obd->u.lmv;
1214         ENTRY;
1215
1216         fld_client_fini(&lmv->lmv_fld);
1217         if (lmv->tgts != NULL) {
1218                 int i;
1219                 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1220                         if (lmv->tgts[i] == NULL)
1221                                 continue;
1222                         lmv_del_target(lmv, i);
1223                 }
1224                 OBD_FREE(lmv->tgts, sizeof(*lmv->tgts) * lmv->tgts_size);
1225                 lmv->tgts_size = 0;
1226         }
1227         RETURN(0);
1228 }
1229
1230 static int lmv_process_config(struct obd_device *obd, obd_count len, void *buf)
1231 {
1232         struct lustre_cfg       *lcfg = buf;
1233         struct obd_uuid         obd_uuid;
1234         int                     gen;
1235         __u32                   index;
1236         int                     rc;
1237         ENTRY;
1238
1239         switch (lcfg->lcfg_command) {
1240         case LCFG_ADD_MDC:
1241                 /* modify_mdc_tgts add 0:lustre-clilmv  1:lustre-MDT0000_UUID
1242                  * 2:0  3:1  4:lustre-MDT0000-mdc_UUID */
1243                 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(obd_uuid.uuid))
1244                         GOTO(out, rc = -EINVAL);
1245
1246                 obd_str2uuid(&obd_uuid,  lustre_cfg_buf(lcfg, 1));
1247
1248                 if (sscanf(lustre_cfg_buf(lcfg, 2), "%d", &index) != 1)
1249                         GOTO(out, rc = -EINVAL);
1250                 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", &gen) != 1)
1251                         GOTO(out, rc = -EINVAL);
1252                 rc = lmv_add_target(obd, &obd_uuid, index, gen);
1253                 GOTO(out, rc);
1254         default:
1255                 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
1256                 GOTO(out, rc = -EINVAL);
1257         }
1258 out:
1259         RETURN(rc);
1260 }
1261
1262 static int lmv_statfs(const struct lu_env *env, struct obd_export *exp,
1263                       struct obd_statfs *osfs, __u64 max_age, __u32 flags)
1264 {
1265         struct obd_device     *obd = class_exp2obd(exp);
1266         struct lmv_obd        *lmv = &obd->u.lmv;
1267         struct obd_statfs     *temp;
1268         int                    rc = 0;
1269         int                    i;
1270         ENTRY;
1271
1272         rc = lmv_check_connect(obd);
1273         if (rc)
1274                 RETURN(rc);
1275
1276         OBD_ALLOC(temp, sizeof(*temp));
1277         if (temp == NULL)
1278                 RETURN(-ENOMEM);
1279
1280         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1281                 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
1282                         continue;
1283
1284                 rc = obd_statfs(env, lmv->tgts[i]->ltd_exp, temp,
1285                                 max_age, flags);
1286                 if (rc) {
1287                         CERROR("can't stat MDS #%d (%s), error %d\n", i,
1288                                lmv->tgts[i]->ltd_exp->exp_obd->obd_name,
1289                                rc);
1290                         GOTO(out_free_temp, rc);
1291                 }
1292
1293                 if (i == 0) {
1294                         *osfs = *temp;
1295                         /* If the statfs is from mount, it will needs
1296                          * retrieve necessary information from MDT0.
1297                          * i.e. mount does not need the merged osfs
1298                          * from all of MDT.
1299                          * And also clients can be mounted as long as
1300                          * MDT0 is in service*/
1301                         if (flags & OBD_STATFS_FOR_MDT0)
1302                                 GOTO(out_free_temp, rc);
1303                 } else {
1304                         osfs->os_bavail += temp->os_bavail;
1305                         osfs->os_blocks += temp->os_blocks;
1306                         osfs->os_ffree += temp->os_ffree;
1307                         osfs->os_files += temp->os_files;
1308                 }
1309         }
1310
1311         EXIT;
1312 out_free_temp:
1313         OBD_FREE(temp, sizeof(*temp));
1314         return rc;
1315 }
1316
1317 static int lmv_getstatus(struct obd_export *exp,
1318                          struct lu_fid *fid,
1319                          struct obd_capa **pc)
1320 {
1321         struct obd_device    *obd = exp->exp_obd;
1322         struct lmv_obd       *lmv = &obd->u.lmv;
1323         int                   rc;
1324         ENTRY;
1325
1326         rc = lmv_check_connect(obd);
1327         if (rc)
1328                 RETURN(rc);
1329
1330         rc = md_getstatus(lmv->tgts[0]->ltd_exp, fid, pc);
1331         RETURN(rc);
1332 }
1333
1334 static int lmv_getxattr(struct obd_export *exp, const struct lu_fid *fid,
1335                         struct obd_capa *oc, obd_valid valid, const char *name,
1336                         const char *input, int input_size, int output_size,
1337                         int flags, struct ptlrpc_request **request)
1338 {
1339         struct obd_device      *obd = exp->exp_obd;
1340         struct lmv_obd         *lmv = &obd->u.lmv;
1341         struct lmv_tgt_desc    *tgt;
1342         int                     rc;
1343         ENTRY;
1344
1345         rc = lmv_check_connect(obd);
1346         if (rc)
1347                 RETURN(rc);
1348
1349         tgt = lmv_find_target(lmv, fid);
1350         if (IS_ERR(tgt))
1351                 RETURN(PTR_ERR(tgt));
1352
1353         rc = md_getxattr(tgt->ltd_exp, fid, oc, valid, name, input,
1354                          input_size, output_size, flags, request);
1355
1356         RETURN(rc);
1357 }
1358
1359 static int lmv_setxattr(struct obd_export *exp, const struct lu_fid *fid,
1360                         struct obd_capa *oc, obd_valid valid, const char *name,
1361                         const char *input, int input_size, int output_size,
1362                         int flags, __u32 suppgid,
1363                         struct ptlrpc_request **request)
1364 {
1365         struct obd_device      *obd = exp->exp_obd;
1366         struct lmv_obd         *lmv = &obd->u.lmv;
1367         struct lmv_tgt_desc    *tgt;
1368         int                     rc;
1369         ENTRY;
1370
1371         rc = lmv_check_connect(obd);
1372         if (rc)
1373                 RETURN(rc);
1374
1375         tgt = lmv_find_target(lmv, fid);
1376         if (IS_ERR(tgt))
1377                 RETURN(PTR_ERR(tgt));
1378
1379         rc = md_setxattr(tgt->ltd_exp, fid, oc, valid, name, input,
1380                          input_size, output_size, flags, suppgid,
1381                          request);
1382
1383         RETURN(rc);
1384 }
1385
1386 static int lmv_getattr(struct obd_export *exp, struct md_op_data *op_data,
1387                        struct ptlrpc_request **request)
1388 {
1389         struct obd_device       *obd = exp->exp_obd;
1390         struct lmv_obd          *lmv = &obd->u.lmv;
1391         struct lmv_tgt_desc     *tgt;
1392         int                      rc;
1393         ENTRY;
1394
1395         rc = lmv_check_connect(obd);
1396         if (rc)
1397                 RETURN(rc);
1398
1399         tgt = lmv_find_target(lmv, &op_data->op_fid1);
1400         if (IS_ERR(tgt))
1401                 RETURN(PTR_ERR(tgt));
1402
1403         if (op_data->op_flags & MF_GET_MDT_IDX) {
1404                 op_data->op_mds = tgt->ltd_idx;
1405                 RETURN(0);
1406         }
1407
1408         rc = md_getattr(tgt->ltd_exp, op_data, request);
1409
1410         RETURN(rc);
1411 }
1412
1413 static int lmv_null_inode(struct obd_export *exp, const struct lu_fid *fid)
1414 {
1415         struct obd_device   *obd = exp->exp_obd;
1416         struct lmv_obd      *lmv = &obd->u.lmv;
1417         int                  i;
1418         int                  rc;
1419         ENTRY;
1420
1421         rc = lmv_check_connect(obd);
1422         if (rc)
1423                 RETURN(rc);
1424
1425         CDEBUG(D_INODE, "CBDATA for "DFID"\n", PFID(fid));
1426
1427         /*
1428          * With DNE every object can have two locks in different namespaces:
1429          * lookup lock in space of MDT storing direntry and update/open lock in
1430          * space of MDT storing inode.
1431          */
1432         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1433                 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
1434                         continue;
1435                 md_null_inode(lmv->tgts[i]->ltd_exp, fid);
1436         }
1437
1438         RETURN(0);
1439 }
1440
1441 static int lmv_find_cbdata(struct obd_export *exp, const struct lu_fid *fid,
1442                            ldlm_iterator_t it, void *data)
1443 {
1444         struct obd_device   *obd = exp->exp_obd;
1445         struct lmv_obd      *lmv = &obd->u.lmv;
1446         int                  i;
1447         int                  rc;
1448         ENTRY;
1449
1450         rc = lmv_check_connect(obd);
1451         if (rc)
1452                 RETURN(rc);
1453
1454         CDEBUG(D_INODE, "CBDATA for "DFID"\n", PFID(fid));
1455
1456         /*
1457          * With DNE every object can have two locks in different namespaces:
1458          * lookup lock in space of MDT storing direntry and update/open lock in
1459          * space of MDT storing inode.
1460          */
1461         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1462                 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
1463                         continue;
1464                 rc = md_find_cbdata(lmv->tgts[i]->ltd_exp, fid, it, data);
1465                 if (rc)
1466                         RETURN(rc);
1467         }
1468
1469         RETURN(rc);
1470 }
1471
1472
1473 static int lmv_close(struct obd_export *exp, struct md_op_data *op_data,
1474                      struct md_open_data *mod, struct ptlrpc_request **request)
1475 {
1476         struct obd_device     *obd = exp->exp_obd;
1477         struct lmv_obd        *lmv = &obd->u.lmv;
1478         struct lmv_tgt_desc   *tgt;
1479         int                    rc;
1480         ENTRY;
1481
1482         rc = lmv_check_connect(obd);
1483         if (rc)
1484                 RETURN(rc);
1485
1486         tgt = lmv_find_target(lmv, &op_data->op_fid1);
1487         if (IS_ERR(tgt))
1488                 RETURN(PTR_ERR(tgt));
1489
1490         CDEBUG(D_INODE, "CLOSE "DFID"\n", PFID(&op_data->op_fid1));
1491         rc = md_close(tgt->ltd_exp, op_data, mod, request);
1492         RETURN(rc);
1493 }
1494
1495 struct lmv_tgt_desc
1496 *lmv_locate_mds(struct lmv_obd *lmv, struct md_op_data *op_data,
1497                 struct lu_fid *fid)
1498 {
1499         struct lmv_tgt_desc *tgt;
1500
1501         tgt = lmv_find_target(lmv, fid);
1502         if (IS_ERR(tgt))
1503                 return tgt;
1504
1505         op_data->op_mds = tgt->ltd_idx;
1506
1507         return tgt;
1508 }
1509
1510 int lmv_create(struct obd_export *exp, struct md_op_data *op_data,
1511                const void *data, int datalen, int mode, __u32 uid,
1512                __u32 gid, cfs_cap_t cap_effective, __u64 rdev,
1513                struct ptlrpc_request **request)
1514 {
1515         struct obd_device       *obd = exp->exp_obd;
1516         struct lmv_obd          *lmv = &obd->u.lmv;
1517         struct lmv_tgt_desc     *tgt;
1518         int                      rc;
1519         ENTRY;
1520
1521         rc = lmv_check_connect(obd);
1522         if (rc)
1523                 RETURN(rc);
1524
1525         if (!lmv->desc.ld_active_tgt_count)
1526                 RETURN(-EIO);
1527
1528         tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
1529         if (IS_ERR(tgt))
1530                 RETURN(PTR_ERR(tgt));
1531
1532         rc = lmv_fid_alloc(exp, &op_data->op_fid2, op_data);
1533         if (rc)
1534                 RETURN(rc);
1535
1536         CDEBUG(D_INODE, "CREATE '%*s' on "DFID" -> mds #%x\n",
1537                op_data->op_namelen, op_data->op_name, PFID(&op_data->op_fid1),
1538                op_data->op_mds);
1539
1540         op_data->op_flags |= MF_MDC_CANCEL_FID1;
1541         rc = md_create(tgt->ltd_exp, op_data, data, datalen, mode, uid, gid,
1542                        cap_effective, rdev, request);
1543
1544         if (rc == 0) {
1545                 if (*request == NULL)
1546                         RETURN(rc);
1547                 CDEBUG(D_INODE, "Created - "DFID"\n", PFID(&op_data->op_fid2));
1548         }
1549         RETURN(rc);
1550 }
1551
1552 static int lmv_done_writing(struct obd_export *exp,
1553                             struct md_op_data *op_data,
1554                             struct md_open_data *mod)
1555 {
1556         struct obd_device     *obd = exp->exp_obd;
1557         struct lmv_obd        *lmv = &obd->u.lmv;
1558         struct lmv_tgt_desc   *tgt;
1559         int                    rc;
1560         ENTRY;
1561
1562         rc = lmv_check_connect(obd);
1563         if (rc)
1564                 RETURN(rc);
1565
1566         tgt = lmv_find_target(lmv, &op_data->op_fid1);
1567         if (IS_ERR(tgt))
1568                 RETURN(PTR_ERR(tgt));
1569
1570         rc = md_done_writing(tgt->ltd_exp, op_data, mod);
1571         RETURN(rc);
1572 }
1573
1574 static int
1575 lmv_enqueue_remote(struct obd_export *exp, struct ldlm_enqueue_info *einfo,
1576                    struct lookup_intent *it, struct md_op_data *op_data,
1577                    struct lustre_handle *lockh, void *lmm, int lmmsize,
1578                    int extra_lock_flags)
1579 {
1580         struct ptlrpc_request      *req = it->d.lustre.it_data;
1581         struct obd_device          *obd = exp->exp_obd;
1582         struct lmv_obd             *lmv = &obd->u.lmv;
1583         struct lustre_handle        plock;
1584         struct lmv_tgt_desc        *tgt;
1585         struct md_op_data          *rdata;
1586         struct lu_fid               fid1;
1587         struct mdt_body            *body;
1588         int                         rc = 0;
1589         int                         pmode;
1590         ENTRY;
1591
1592         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1593         LASSERT(body != NULL);
1594
1595         if (!(body->valid & OBD_MD_MDS))
1596                 RETURN(0);
1597
1598         CDEBUG(D_INODE, "REMOTE_ENQUEUE '%s' on "DFID" -> "DFID"\n",
1599                LL_IT2STR(it), PFID(&op_data->op_fid1), PFID(&body->fid1));
1600
1601         /*
1602          * We got LOOKUP lock, but we really need attrs.
1603          */
1604         pmode = it->d.lustre.it_lock_mode;
1605         LASSERT(pmode != 0);
1606         memcpy(&plock, lockh, sizeof(plock));
1607         it->d.lustre.it_lock_mode = 0;
1608         it->d.lustre.it_data = NULL;
1609         fid1 = body->fid1;
1610
1611         it->d.lustre.it_disposition &= ~DISP_ENQ_COMPLETE;
1612         ptlrpc_req_finished(req);
1613
1614         tgt = lmv_find_target(lmv, &fid1);
1615         if (IS_ERR(tgt))
1616                 GOTO(out, rc = PTR_ERR(tgt));
1617
1618         OBD_ALLOC_PTR(rdata);
1619         if (rdata == NULL)
1620                 GOTO(out, rc = -ENOMEM);
1621
1622         rdata->op_fid1 = fid1;
1623         rdata->op_bias = MDS_CROSS_REF;
1624
1625         rc = md_enqueue(tgt->ltd_exp, einfo, it, rdata, lockh,
1626                         lmm, lmmsize, NULL, extra_lock_flags);
1627         OBD_FREE_PTR(rdata);
1628         EXIT;
1629 out:
1630         ldlm_lock_decref(&plock, pmode);
1631         return rc;
1632 }
1633
1634 static int
1635 lmv_enqueue(struct obd_export *exp, struct ldlm_enqueue_info *einfo,
1636             struct lookup_intent *it, struct md_op_data *op_data,
1637             struct lustre_handle *lockh, void *lmm, int lmmsize,
1638             struct ptlrpc_request **req, __u64 extra_lock_flags)
1639 {
1640         struct obd_device        *obd = exp->exp_obd;
1641         struct lmv_obd           *lmv = &obd->u.lmv;
1642         struct lmv_tgt_desc      *tgt;
1643         int                       rc;
1644         ENTRY;
1645
1646         rc = lmv_check_connect(obd);
1647         if (rc)
1648                 RETURN(rc);
1649
1650         CDEBUG(D_INODE, "ENQUEUE '%s' on "DFID"\n",
1651                LL_IT2STR(it), PFID(&op_data->op_fid1));
1652
1653         tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
1654         if (IS_ERR(tgt))
1655                 RETURN(PTR_ERR(tgt));
1656
1657         CDEBUG(D_INODE, "ENQUEUE '%s' on "DFID" -> mds #%d\n",
1658                LL_IT2STR(it), PFID(&op_data->op_fid1), tgt->ltd_idx);
1659
1660         rc = md_enqueue(tgt->ltd_exp, einfo, it, op_data, lockh,
1661                         lmm, lmmsize, req, extra_lock_flags);
1662
1663         if (rc == 0 && it && it->it_op == IT_OPEN) {
1664                 rc = lmv_enqueue_remote(exp, einfo, it, op_data, lockh,
1665                                         lmm, lmmsize, extra_lock_flags);
1666         }
1667         RETURN(rc);
1668 }
1669
1670 static int
1671 lmv_getattr_name(struct obd_export *exp,struct md_op_data *op_data,
1672                  struct ptlrpc_request **request)
1673 {
1674         struct ptlrpc_request   *req = NULL;
1675         struct obd_device       *obd = exp->exp_obd;
1676         struct lmv_obd          *lmv = &obd->u.lmv;
1677         struct lmv_tgt_desc     *tgt;
1678         struct mdt_body         *body;
1679         int                      rc;
1680         ENTRY;
1681
1682         rc = lmv_check_connect(obd);
1683         if (rc)
1684                 RETURN(rc);
1685
1686         tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
1687         if (IS_ERR(tgt))
1688                 RETURN(PTR_ERR(tgt));
1689
1690         CDEBUG(D_INODE, "GETATTR_NAME for %*s on "DFID" -> mds #%d\n",
1691                op_data->op_namelen, op_data->op_name, PFID(&op_data->op_fid1),
1692                tgt->ltd_idx);
1693
1694         rc = md_getattr_name(tgt->ltd_exp, op_data, request);
1695         if (rc != 0)
1696                 RETURN(rc);
1697
1698         body = req_capsule_server_get(&(*request)->rq_pill,
1699                                       &RMF_MDT_BODY);
1700         LASSERT(body != NULL);
1701
1702         if (body->valid & OBD_MD_MDS) {
1703                 struct lu_fid rid = body->fid1;
1704                 CDEBUG(D_INODE, "Request attrs for "DFID"\n",
1705                        PFID(&rid));
1706
1707                 tgt = lmv_find_target(lmv, &rid);
1708                 if (IS_ERR(tgt)) {
1709                         ptlrpc_req_finished(*request);
1710                         RETURN(PTR_ERR(tgt));
1711                 }
1712
1713                 op_data->op_fid1 = rid;
1714                 op_data->op_valid |= OBD_MD_FLCROSSREF;
1715                 op_data->op_namelen = 0;
1716                 op_data->op_name = NULL;
1717                 rc = md_getattr_name(tgt->ltd_exp, op_data, &req);
1718                 ptlrpc_req_finished(*request);
1719                 *request = req;
1720         }
1721
1722         RETURN(rc);
1723 }
1724
1725 #define md_op_data_fid(op_data, fl)                     \
1726         (fl == MF_MDC_CANCEL_FID1 ? &op_data->op_fid1 : \
1727          fl == MF_MDC_CANCEL_FID2 ? &op_data->op_fid2 : \
1728          fl == MF_MDC_CANCEL_FID3 ? &op_data->op_fid3 : \
1729          fl == MF_MDC_CANCEL_FID4 ? &op_data->op_fid4 : \
1730          NULL)
1731
1732 static int lmv_early_cancel(struct obd_export *exp, struct md_op_data *op_data,
1733                             int op_tgt, ldlm_mode_t mode, int bits, int flag)
1734 {
1735         struct lu_fid          *fid = md_op_data_fid(op_data, flag);
1736         struct obd_device      *obd = exp->exp_obd;
1737         struct lmv_obd         *lmv = &obd->u.lmv;
1738         struct lmv_tgt_desc    *tgt;
1739         ldlm_policy_data_t      policy = {{0}};
1740         int                     rc = 0;
1741         ENTRY;
1742
1743         if (!fid_is_sane(fid))
1744                 RETURN(0);
1745
1746         tgt = lmv_find_target(lmv, fid);
1747         if (IS_ERR(tgt))
1748                 RETURN(PTR_ERR(tgt));
1749
1750         if (tgt->ltd_idx != op_tgt) {
1751                 CDEBUG(D_INODE, "EARLY_CANCEL on "DFID"\n", PFID(fid));
1752                 policy.l_inodebits.bits = bits;
1753                 rc = md_cancel_unused(tgt->ltd_exp, fid, &policy,
1754                                       mode, LCF_ASYNC, NULL);
1755         } else {
1756                 CDEBUG(D_INODE,
1757                        "EARLY_CANCEL skip operation target %d on "DFID"\n",
1758                        op_tgt, PFID(fid));
1759                 op_data->op_flags |= flag;
1760                 rc = 0;
1761         }
1762
1763         RETURN(rc);
1764 }
1765
1766 /*
1767  * llite passes fid of an target inode in op_data->op_fid1 and id of directory in
1768  * op_data->op_fid2
1769  */
1770 static int lmv_link(struct obd_export *exp, struct md_op_data *op_data,
1771                     struct ptlrpc_request **request)
1772 {
1773         struct obd_device       *obd = exp->exp_obd;
1774         struct lmv_obd          *lmv = &obd->u.lmv;
1775         struct lmv_tgt_desc     *tgt;
1776         int                      rc;
1777         ENTRY;
1778
1779         rc = lmv_check_connect(obd);
1780         if (rc)
1781                 RETURN(rc);
1782
1783         LASSERT(op_data->op_namelen != 0);
1784
1785         CDEBUG(D_INODE, "LINK "DFID":%*s to "DFID"\n",
1786                PFID(&op_data->op_fid2), op_data->op_namelen,
1787                op_data->op_name, PFID(&op_data->op_fid1));
1788
1789         op_data->op_fsuid = cfs_curproc_fsuid();
1790         op_data->op_fsgid = cfs_curproc_fsgid();
1791         op_data->op_cap = cfs_curproc_cap_pack();
1792         tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid2);
1793         if (IS_ERR(tgt))
1794                 RETURN(PTR_ERR(tgt));
1795
1796         /*
1797          * Cancel UPDATE lock on child (fid1).
1798          */
1799         op_data->op_flags |= MF_MDC_CANCEL_FID2;
1800         rc = lmv_early_cancel(exp, op_data, tgt->ltd_idx, LCK_EX,
1801                               MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID1);
1802         if (rc != 0)
1803                 RETURN(rc);
1804
1805         rc = md_link(tgt->ltd_exp, op_data, request);
1806
1807         RETURN(rc);
1808 }
1809
1810 static int lmv_rename(struct obd_export *exp, struct md_op_data *op_data,
1811                       const char *old, int oldlen, const char *new, int newlen,
1812                       struct ptlrpc_request **request)
1813 {
1814         struct obd_device       *obd = exp->exp_obd;
1815         struct lmv_obd          *lmv = &obd->u.lmv;
1816         struct lmv_tgt_desc     *src_tgt;
1817         struct lmv_tgt_desc     *tgt_tgt;
1818         int                     rc;
1819         ENTRY;
1820
1821         LASSERT(oldlen != 0);
1822
1823         CDEBUG(D_INODE, "RENAME %*s in "DFID" to %*s in "DFID"\n",
1824                oldlen, old, PFID(&op_data->op_fid1),
1825                newlen, new, PFID(&op_data->op_fid2));
1826
1827         rc = lmv_check_connect(obd);
1828         if (rc)
1829                 RETURN(rc);
1830
1831         op_data->op_fsuid = cfs_curproc_fsuid();
1832         op_data->op_fsgid = cfs_curproc_fsgid();
1833         op_data->op_cap = cfs_curproc_cap_pack();
1834         src_tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
1835         if (IS_ERR(src_tgt))
1836                 RETURN(PTR_ERR(src_tgt));
1837
1838         tgt_tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid2);
1839         if (IS_ERR(tgt_tgt))
1840                 RETURN(PTR_ERR(tgt_tgt));
1841         /*
1842          * LOOKUP lock on src child (fid3) should also be cancelled for
1843          * src_tgt in mdc_rename.
1844          */
1845         op_data->op_flags |= MF_MDC_CANCEL_FID1 | MF_MDC_CANCEL_FID3;
1846
1847         /*
1848          * Cancel UPDATE locks on tgt parent (fid2), tgt_tgt is its
1849          * own target.
1850          */
1851         rc = lmv_early_cancel(exp, op_data, src_tgt->ltd_idx,
1852                               LCK_EX, MDS_INODELOCK_UPDATE,
1853                               MF_MDC_CANCEL_FID2);
1854
1855         /*
1856          * Cancel LOOKUP locks on tgt child (fid4) for parent tgt_tgt.
1857          */
1858         if (rc == 0) {
1859                 rc = lmv_early_cancel(exp, op_data, src_tgt->ltd_idx,
1860                                       LCK_EX, MDS_INODELOCK_LOOKUP,
1861                                       MF_MDC_CANCEL_FID4);
1862         }
1863
1864         /*
1865          * Cancel all the locks on tgt child (fid4).
1866          */
1867         if (rc == 0)
1868                 rc = lmv_early_cancel(exp, op_data, src_tgt->ltd_idx,
1869                                       LCK_EX, MDS_INODELOCK_FULL,
1870                                       MF_MDC_CANCEL_FID4);
1871
1872         if (rc == 0)
1873                 rc = md_rename(src_tgt->ltd_exp, op_data, old, oldlen,
1874                                new, newlen, request);
1875         RETURN(rc);
1876 }
1877
1878 static int lmv_setattr(struct obd_export *exp, struct md_op_data *op_data,
1879                        void *ea, int ealen, void *ea2, int ea2len,
1880                        struct ptlrpc_request **request,
1881                        struct md_open_data **mod)
1882 {
1883         struct obd_device       *obd = exp->exp_obd;
1884         struct lmv_obd          *lmv = &obd->u.lmv;
1885         struct lmv_tgt_desc     *tgt;
1886         int                      rc = 0;
1887         ENTRY;
1888
1889         rc = lmv_check_connect(obd);
1890         if (rc)
1891                 RETURN(rc);
1892
1893         CDEBUG(D_INODE, "SETATTR for "DFID", valid 0x%x\n",
1894                PFID(&op_data->op_fid1), op_data->op_attr.ia_valid);
1895
1896         op_data->op_flags |= MF_MDC_CANCEL_FID1;
1897         tgt = lmv_find_target(lmv, &op_data->op_fid1);
1898         if (IS_ERR(tgt))
1899                 RETURN(PTR_ERR(tgt));
1900
1901         rc = md_setattr(tgt->ltd_exp, op_data, ea, ealen, ea2,
1902                         ea2len, request, mod);
1903
1904         RETURN(rc);
1905 }
1906
1907 static int lmv_sync(struct obd_export *exp, const struct lu_fid *fid,
1908                     struct obd_capa *oc, struct ptlrpc_request **request)
1909 {
1910         struct obd_device         *obd = exp->exp_obd;
1911         struct lmv_obd            *lmv = &obd->u.lmv;
1912         struct lmv_tgt_desc       *tgt;
1913         int                        rc;
1914         ENTRY;
1915
1916         rc = lmv_check_connect(obd);
1917         if (rc)
1918                 RETURN(rc);
1919
1920         tgt = lmv_find_target(lmv, fid);
1921         if (IS_ERR(tgt))
1922                 RETURN(PTR_ERR(tgt));
1923
1924         rc = md_sync(tgt->ltd_exp, fid, oc, request);
1925         RETURN(rc);
1926 }
1927
1928 static int lmv_readpage(struct obd_export *exp, struct md_op_data *op_data,
1929                         struct page **pages, struct ptlrpc_request **request)
1930 {
1931         struct obd_device       *obd = exp->exp_obd;
1932         struct lmv_obd          *lmv = &obd->u.lmv;
1933         __u64                    offset = op_data->op_offset;
1934         int                      rc;
1935         int                      i;
1936         /* number of pages read, in CFS_PAGE_SIZE */
1937         int                      nrdpgs;
1938         /* number of pages transferred in LU_PAGE_SIZE */
1939         int                      nlupgs;
1940         struct lmv_tgt_desc     *tgt;
1941         struct lu_dirpage       *dp;
1942         struct lu_dirent        *ent;
1943         ENTRY;
1944
1945         rc = lmv_check_connect(obd);
1946         if (rc)
1947                 RETURN(rc);
1948
1949         CDEBUG(D_INODE, "READPAGE at "LPX64" from "DFID"\n",
1950                offset, PFID(&op_data->op_fid1));
1951
1952         /*
1953          * This case handle directory lookup in clustered metadata case (i.e.
1954          * split directory is located on multiple md servers.)
1955          * each server keeps directory entries for certain range of hashes.
1956          * E.g. we have N server and suppose hash range is 0 to MAX_HASH.
1957          * first server will keep records with hashes [ 0 ... MAX_HASH /N  - 1],
1958          * second one with hashes [MAX_HASH / N ... 2 * MAX_HASH / N] and
1959          * so on....
1960          *      readdir can simply start reading entries from 0 - N server in
1961          * order but that will not scale well as all client will request dir in
1962          * to server in same order.
1963          * Following algorithm does optimization:
1964          * Instead of doing readdir in 1, 2, ...., N order, client with a
1965          * rank R does readdir in R, R + 1, ..., N, 1, ... R - 1 order.
1966          * (every client has rank R)
1967          *      But ll_readdir() expect offset range [0 to MAX_HASH/N) but
1968          * since client ask dir from MDS{R} client has pages with offsets
1969          * [R*MAX_HASH/N ... (R + 1)*MAX_HASH/N] there for we do hash_adj
1970          * on hash  values that we get.
1971          * Since these codes might be still useful for sharded directory, so
1972          * Keeping this code for further reference
1973         if (0) {
1974                 LASSERT(nr > 0);
1975                 seg_size = MAX_HASH_SIZE;
1976                 do_div(seg_size, nr);
1977                 los      = obj->lo_stripes;
1978                 tgt      = lmv_get_target(lmv, los[0].ls_mds);
1979                 rank     = lmv_node_rank(tgt->ltd_exp, fid) % nr;
1980                 tgt_tmp  = offset;
1981                 do_div(tgt_tmp, seg_size);
1982                 tgt0_idx = do_div(tgt_tmp,  nr);
1983                 tgt_idx  = (tgt0_idx + rank) % nr;
1984
1985                 if (tgt_idx < tgt0_idx)
1986                          * Wrap around.
1987                          *
1988                          * Last segment has unusual length due to division
1989                          * rounding.
1990                         hash_adj = MAX_HASH_SIZE - seg_size * nr;
1991                 else
1992                         hash_adj = 0;
1993
1994                 hash_adj += rank * seg_size;
1995
1996                 CDEBUG(D_INODE, "Readpage hash adjustment: %x "LPX64" "
1997                        LPX64"/%x -> "LPX64"/%x\n", rank, hash_adj,
1998                        offset, tgt0_idx, offset + hash_adj, tgt_idx);
1999
2000                 offset = (offset + hash_adj) & MAX_HASH_SIZE;
2001                 rid = lsm->mea_oinfo[tgt_idx].lmo_fid;
2002                 tgt = lmv_get_target(lmv, lsm->mea_oinfo[tgt_idx].lmo_mds);
2003
2004                 CDEBUG(D_INODE, "Forward to "DFID" with offset %lu i %d\n",
2005                        PFID(&rid), (unsigned long)offset, tgt_idx);
2006         }
2007         */
2008         tgt = lmv_find_target(lmv, &op_data->op_fid1);
2009         if (IS_ERR(tgt))
2010                 RETURN(PTR_ERR(tgt));
2011
2012         rc = md_readpage(tgt->ltd_exp, op_data, pages, request);
2013         if (rc != 0)
2014                 RETURN(rc);
2015
2016         nrdpgs = ((*request)->rq_bulk->bd_nob_transferred + CFS_PAGE_SIZE - 1)
2017                  >> CFS_PAGE_SHIFT;
2018         nlupgs = (*request)->rq_bulk->bd_nob_transferred >> LU_PAGE_SHIFT;
2019         LASSERT(!((*request)->rq_bulk->bd_nob_transferred & ~LU_PAGE_MASK));
2020         LASSERT(nrdpgs > 0 && nrdpgs <= op_data->op_npages);
2021
2022         CDEBUG(D_INODE, "read %d(%d)/%d pages\n", nrdpgs, nlupgs,
2023                op_data->op_npages);
2024
2025         for (i = 0; i < nrdpgs; i++) {
2026 #if CFS_PAGE_SIZE > LU_PAGE_SIZE
2027                 struct lu_dirpage *first;
2028                 __u64 hash_end = 0;
2029                 __u32 flags = 0;
2030 #endif
2031                 struct lu_dirent *tmp = NULL;
2032
2033                 dp = cfs_kmap(pages[i]);
2034                 ent = lu_dirent_start(dp);
2035 #if CFS_PAGE_SIZE > LU_PAGE_SIZE
2036                 first = dp;
2037                 hash_end = dp->ldp_hash_end;
2038 repeat:
2039 #endif
2040                 nlupgs--;
2041
2042                 for (tmp = ent; ent != NULL;
2043                      tmp = ent, ent = lu_dirent_next(ent));
2044 #if CFS_PAGE_SIZE > LU_PAGE_SIZE
2045                 dp = (struct lu_dirpage *)((char *)dp + LU_PAGE_SIZE);
2046                 if (((unsigned long)dp & ~CFS_PAGE_MASK) && nlupgs > 0) {
2047                         ent = lu_dirent_start(dp);
2048
2049                         if (tmp) {
2050                                 /* enlarge the end entry lde_reclen from 0 to
2051                                  * first entry of next lu_dirpage, in this way
2052                                  * several lu_dirpages can be stored into one
2053                                  * client page on client. */
2054                                 tmp = ((void *)tmp) +
2055                                       le16_to_cpu(tmp->lde_reclen);
2056                                 tmp->lde_reclen =
2057                                         cpu_to_le16((char *)(dp->ldp_entries) -
2058                                                     (char *)tmp);
2059                                 goto repeat;
2060                         }
2061                 }
2062                 first->ldp_hash_end = hash_end;
2063                 first->ldp_flags &= ~cpu_to_le32(LDF_COLLIDE);
2064                 first->ldp_flags |= flags & cpu_to_le32(LDF_COLLIDE);
2065 #else
2066                 SET_BUT_UNUSED(tmp);
2067 #endif
2068                 cfs_kunmap(pages[i]);
2069         }
2070         RETURN(rc);
2071 }
2072
2073 static int lmv_unlink(struct obd_export *exp, struct md_op_data *op_data,
2074                       struct ptlrpc_request **request)
2075 {
2076         struct obd_device       *obd = exp->exp_obd;
2077         struct lmv_obd          *lmv = &obd->u.lmv;
2078         struct lmv_tgt_desc     *tgt = NULL;
2079         struct mdt_body         *body;
2080         int                     rc;
2081         ENTRY;
2082
2083         rc = lmv_check_connect(obd);
2084         if (rc)
2085                 RETURN(rc);
2086 retry:
2087         /* Send unlink requests to the MDT where the child is located */
2088         if (likely(!fid_is_zero(&op_data->op_fid2)))
2089                 tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid2);
2090         else
2091                 tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
2092         if (IS_ERR(tgt))
2093                 RETURN(PTR_ERR(tgt));
2094
2095         op_data->op_fsuid = cfs_curproc_fsuid();
2096         op_data->op_fsgid = cfs_curproc_fsgid();
2097         op_data->op_cap = cfs_curproc_cap_pack();
2098
2099         /*
2100          * If child's fid is given, cancel unused locks for it if it is from
2101          * another export than parent.
2102          *
2103          * LOOKUP lock for child (fid3) should also be cancelled on parent
2104          * tgt_tgt in mdc_unlink().
2105          */
2106         op_data->op_flags |= MF_MDC_CANCEL_FID1 | MF_MDC_CANCEL_FID3;
2107
2108         /*
2109          * Cancel FULL locks on child (fid3).
2110          */
2111         rc = lmv_early_cancel(exp, op_data, tgt->ltd_idx, LCK_EX,
2112                               MDS_INODELOCK_FULL, MF_MDC_CANCEL_FID3);
2113
2114         if (rc != 0)
2115                 RETURN(rc);
2116
2117         CDEBUG(D_INODE, "unlink with fid="DFID"/"DFID" -> mds #%d\n",
2118                PFID(&op_data->op_fid1), PFID(&op_data->op_fid2), tgt->ltd_idx);
2119
2120         rc = md_unlink(tgt->ltd_exp, op_data, request);
2121         if (rc != 0 && rc != -EREMOTE)
2122                 RETURN(rc);
2123
2124         body = req_capsule_server_get(&(*request)->rq_pill, &RMF_MDT_BODY);
2125         if (body == NULL)
2126                 RETURN(-EPROTO);
2127
2128         /* Not cross-ref case, just get out of here. */
2129         if (likely(!(body->valid & OBD_MD_MDS)))
2130                 RETURN(0);
2131
2132         CDEBUG(D_INODE, "%s: try unlink to another MDT for "DFID"\n",
2133                exp->exp_obd->obd_name, PFID(&body->fid1));
2134
2135         /* This is a remote object, try remote MDT, Note: it may
2136          * try more than 1 time here, Considering following case
2137          * /mnt/lustre is root on MDT0, remote1 is on MDT1
2138          * 1. Initially A does not know where remote1 is, it send
2139          *    unlink RPC to MDT0, MDT0 return -EREMOTE, it will
2140          *    resend unlink RPC to MDT1 (retry 1st time).
2141          *
2142          * 2. During the unlink RPC in flight,
2143          *    client B mv /mnt/lustre/remote1 /mnt/lustre/remote2
2144          *    and create new remote1, but on MDT0
2145          *
2146          * 3. MDT1 get unlink RPC(from A), then do remote lock on
2147          *    /mnt/lustre, then lookup get fid of remote1, and find
2148          *    it is remote dir again, and replay -EREMOTE again.
2149          *
2150          * 4. Then A will resend unlink RPC to MDT0. (retry 2nd times).
2151          *
2152          * In theory, it might try unlimited time here, but it should
2153          * be very rare case.  */
2154         op_data->op_fid2 = body->fid1;
2155         ptlrpc_req_finished(*request);
2156         *request = NULL;
2157
2158         goto retry;
2159 }
2160
2161 static int lmv_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
2162 {
2163         struct lmv_obd *lmv = &obd->u.lmv;
2164         int rc = 0;
2165
2166         switch (stage) {
2167         case OBD_CLEANUP_EARLY:
2168                 /* XXX: here should be calling obd_precleanup() down to
2169                  * stack. */
2170                 break;
2171         case OBD_CLEANUP_EXPORTS:
2172                 fld_client_proc_fini(&lmv->lmv_fld);
2173                 lprocfs_obd_cleanup(obd);
2174                 break;
2175         default:
2176                 break;
2177         }
2178         RETURN(rc);
2179 }
2180
2181 static int lmv_get_info(const struct lu_env *env, struct obd_export *exp,
2182                         __u32 keylen, void *key, __u32 *vallen, void *val,
2183                         struct lov_stripe_md *lsm)
2184 {
2185         struct obd_device       *obd;
2186         struct lmv_obd          *lmv;
2187         int                      rc = 0;
2188         ENTRY;
2189
2190         obd = class_exp2obd(exp);
2191         if (obd == NULL) {
2192                 CDEBUG(D_IOCTL, "Invalid client cookie "LPX64"\n",
2193                        exp->exp_handle.h_cookie);
2194                 RETURN(-EINVAL);
2195         }
2196
2197         lmv = &obd->u.lmv;
2198         if (keylen >= strlen("remote_flag") && !strcmp(key, "remote_flag")) {
2199                 struct lmv_tgt_desc *tgt;
2200                 int i;
2201
2202                 rc = lmv_check_connect(obd);
2203                 if (rc)
2204                         RETURN(rc);
2205
2206                 LASSERT(*vallen == sizeof(__u32));
2207                 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2208                         tgt = lmv->tgts[i];
2209                         /*
2210                          * All tgts should be connected when this gets called.
2211                          */
2212                         if (tgt == NULL || tgt->ltd_exp == NULL)
2213                                 continue;
2214
2215                         if (!obd_get_info(env, tgt->ltd_exp, keylen, key,
2216                                           vallen, val, NULL))
2217                                 RETURN(0);
2218                 }
2219                 RETURN(-EINVAL);
2220         } else if (KEY_IS(KEY_MAX_EASIZE) || KEY_IS(KEY_CONN_DATA)) {
2221                 rc = lmv_check_connect(obd);
2222                 if (rc)
2223                         RETURN(rc);
2224
2225                 /*
2226                  * Forwarding this request to first MDS, it should know LOV
2227                  * desc.
2228                  */
2229                 rc = obd_get_info(env, lmv->tgts[0]->ltd_exp, keylen, key,
2230                                   vallen, val, NULL);
2231                 if (!rc && KEY_IS(KEY_CONN_DATA))
2232                         exp->exp_connect_data = *(struct obd_connect_data *)val;
2233                 RETURN(rc);
2234         } else if (KEY_IS(KEY_TGT_COUNT)) {
2235                 *((int *)val) = lmv->desc.ld_tgt_count;
2236                 RETURN(0);
2237         }
2238
2239         CDEBUG(D_IOCTL, "Invalid key\n");
2240         RETURN(-EINVAL);
2241 }
2242
2243 int lmv_set_info_async(const struct lu_env *env, struct obd_export *exp,
2244                        obd_count keylen, void *key, obd_count vallen,
2245                        void *val, struct ptlrpc_request_set *set)
2246 {
2247         struct lmv_tgt_desc    *tgt;
2248         struct obd_device      *obd;
2249         struct lmv_obd         *lmv;
2250         int rc = 0;
2251         ENTRY;
2252
2253         obd = class_exp2obd(exp);
2254         if (obd == NULL) {
2255                 CDEBUG(D_IOCTL, "Invalid client cookie "LPX64"\n",
2256                        exp->exp_handle.h_cookie);
2257                 RETURN(-EINVAL);
2258         }
2259         lmv = &obd->u.lmv;
2260
2261         if (KEY_IS(KEY_READ_ONLY) || KEY_IS(KEY_FLUSH_CTX)) {
2262                 int i, err = 0;
2263
2264                 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2265                         tgt = lmv->tgts[i];
2266
2267                         if (tgt == NULL || tgt->ltd_exp == NULL)
2268                                 continue;
2269
2270                         err = obd_set_info_async(env, tgt->ltd_exp,
2271                                                  keylen, key, vallen, val, set);
2272                         if (err && rc == 0)
2273                                 rc = err;
2274                 }
2275
2276                 RETURN(rc);
2277         }
2278
2279         RETURN(-EINVAL);
2280 }
2281
2282 int lmv_packmd(struct obd_export *exp, struct lov_mds_md **lmmp,
2283                struct lov_stripe_md *lsm)
2284 {
2285         struct obd_device         *obd = class_exp2obd(exp);
2286         struct lmv_obd            *lmv = &obd->u.lmv;
2287         struct lmv_stripe_md      *meap;
2288         struct lmv_stripe_md      *lsmp;
2289         int                        mea_size;
2290         int                        i;
2291         ENTRY;
2292
2293         mea_size = lmv_get_easize(lmv);
2294         if (!lmmp)
2295                 RETURN(mea_size);
2296
2297         if (*lmmp && !lsm) {
2298                 OBD_FREE_LARGE(*lmmp, mea_size);
2299                 *lmmp = NULL;
2300                 RETURN(0);
2301         }
2302
2303         if (*lmmp == NULL) {
2304                 OBD_ALLOC_LARGE(*lmmp, mea_size);
2305                 if (*lmmp == NULL)
2306                         RETURN(-ENOMEM);
2307         }
2308
2309         if (!lsm)
2310                 RETURN(mea_size);
2311
2312         lsmp = (struct lmv_stripe_md *)lsm;
2313         meap = (struct lmv_stripe_md *)*lmmp;
2314
2315         if (lsmp->mea_magic != MEA_MAGIC_LAST_CHAR &&
2316             lsmp->mea_magic != MEA_MAGIC_ALL_CHARS)
2317                 RETURN(-EINVAL);
2318
2319         meap->mea_magic = cpu_to_le32(lsmp->mea_magic);
2320         meap->mea_count = cpu_to_le32(lsmp->mea_count);
2321         meap->mea_master = cpu_to_le32(lsmp->mea_master);
2322
2323         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2324                 meap->mea_ids[i] = lsmp->mea_ids[i];
2325                 fid_cpu_to_le(&meap->mea_ids[i], &lsmp->mea_ids[i]);
2326         }
2327
2328         RETURN(mea_size);
2329 }
2330
2331 int lmv_unpackmd(struct obd_export *exp, struct lov_stripe_md **lsmp,
2332                  struct lov_mds_md *lmm, int lmm_size)
2333 {
2334         struct obd_device          *obd = class_exp2obd(exp);
2335         struct lmv_stripe_md      **tmea = (struct lmv_stripe_md **)lsmp;
2336         struct lmv_stripe_md       *mea = (struct lmv_stripe_md *)lmm;
2337         struct lmv_obd             *lmv = &obd->u.lmv;
2338         int                         mea_size;
2339         int                         i;
2340         __u32                       magic;
2341         ENTRY;
2342
2343         mea_size = lmv_get_easize(lmv);
2344         if (lsmp == NULL)
2345                 return mea_size;
2346
2347         if (*lsmp != NULL && lmm == NULL) {
2348                 OBD_FREE_LARGE(*tmea, mea_size);
2349                 *lsmp = NULL;
2350                 RETURN(0);
2351         }
2352
2353         LASSERT(mea_size == lmm_size);
2354
2355         OBD_ALLOC_LARGE(*tmea, mea_size);
2356         if (*tmea == NULL)
2357                 RETURN(-ENOMEM);
2358
2359         if (!lmm)
2360                 RETURN(mea_size);
2361
2362         if (mea->mea_magic == MEA_MAGIC_LAST_CHAR ||
2363             mea->mea_magic == MEA_MAGIC_ALL_CHARS ||
2364             mea->mea_magic == MEA_MAGIC_HASH_SEGMENT)
2365         {
2366                 magic = le32_to_cpu(mea->mea_magic);
2367         } else {
2368                 /*
2369                  * Old mea is not handled here.
2370                  */
2371                 CERROR("Old not supportable EA is found\n");
2372                 LBUG();
2373         }
2374
2375         (*tmea)->mea_magic = magic;
2376         (*tmea)->mea_count = le32_to_cpu(mea->mea_count);
2377         (*tmea)->mea_master = le32_to_cpu(mea->mea_master);
2378
2379         for (i = 0; i < (*tmea)->mea_count; i++) {
2380                 (*tmea)->mea_ids[i] = mea->mea_ids[i];
2381                 fid_le_to_cpu(&(*tmea)->mea_ids[i], &(*tmea)->mea_ids[i]);
2382         }
2383         RETURN(mea_size);
2384 }
2385
2386 static int lmv_cancel_unused(struct obd_export *exp, const struct lu_fid *fid,
2387                              ldlm_policy_data_t *policy, ldlm_mode_t mode,
2388                              ldlm_cancel_flags_t flags, void *opaque)
2389 {
2390         struct obd_device       *obd = exp->exp_obd;
2391         struct lmv_obd          *lmv = &obd->u.lmv;
2392         int                      rc = 0;
2393         int                      err;
2394         int                      i;
2395         ENTRY;
2396
2397         LASSERT(fid != NULL);
2398
2399         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2400                 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL ||
2401                     lmv->tgts[i]->ltd_active == 0)
2402                         continue;
2403
2404                 err = md_cancel_unused(lmv->tgts[i]->ltd_exp, fid,
2405                                        policy, mode, flags, opaque);
2406                 if (!rc)
2407                         rc = err;
2408         }
2409         RETURN(rc);
2410 }
2411
2412 int lmv_set_lock_data(struct obd_export *exp, __u64 *lockh, void *data,
2413                       __u64 *bits)
2414 {
2415         struct lmv_obd          *lmv = &exp->exp_obd->u.lmv;
2416         int                      rc;
2417         ENTRY;
2418
2419         rc =  md_set_lock_data(lmv->tgts[0]->ltd_exp, lockh, data, bits);
2420         RETURN(rc);
2421 }
2422
2423 ldlm_mode_t lmv_lock_match(struct obd_export *exp, __u64 flags,
2424                            const struct lu_fid *fid, ldlm_type_t type,
2425                            ldlm_policy_data_t *policy, ldlm_mode_t mode,
2426                            struct lustre_handle *lockh)
2427 {
2428         struct obd_device       *obd = exp->exp_obd;
2429         struct lmv_obd          *lmv = &obd->u.lmv;
2430         ldlm_mode_t              rc;
2431         int                      i;
2432         ENTRY;
2433
2434         CDEBUG(D_INODE, "Lock match for "DFID"\n", PFID(fid));
2435
2436         /*
2437          * With CMD every object can have two locks in different namespaces:
2438          * lookup lock in space of mds storing direntry and update/open lock in
2439          * space of mds storing inode. Thus we check all targets, not only that
2440          * one fid was created in.
2441          */
2442         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2443                 if (lmv->tgts[i] == NULL ||
2444                     lmv->tgts[i]->ltd_exp == NULL ||
2445                     lmv->tgts[i]->ltd_active == 0)
2446                         continue;
2447
2448                 rc = md_lock_match(lmv->tgts[i]->ltd_exp, flags, fid,
2449                                    type, policy, mode, lockh);
2450                 if (rc)
2451                         RETURN(rc);
2452         }
2453
2454         RETURN(0);
2455 }
2456
2457 int lmv_get_lustre_md(struct obd_export *exp, struct ptlrpc_request *req,
2458                       struct obd_export *dt_exp, struct obd_export *md_exp,
2459                       struct lustre_md *md)
2460 {
2461         struct lmv_obd          *lmv = &exp->exp_obd->u.lmv;
2462
2463         return md_get_lustre_md(lmv->tgts[0]->ltd_exp, req, dt_exp, md_exp, md);
2464 }
2465
2466 int lmv_free_lustre_md(struct obd_export *exp, struct lustre_md *md)
2467 {
2468         struct obd_device       *obd = exp->exp_obd;
2469         struct lmv_obd          *lmv = &obd->u.lmv;
2470         ENTRY;
2471
2472         if (md->mea)
2473                 obd_free_memmd(exp, (void *)&md->mea);
2474         RETURN(md_free_lustre_md(lmv->tgts[0]->ltd_exp, md));
2475 }
2476
2477 int lmv_set_open_replay_data(struct obd_export *exp,
2478                              struct obd_client_handle *och,
2479                              struct ptlrpc_request *open_req)
2480 {
2481         struct obd_device       *obd = exp->exp_obd;
2482         struct lmv_obd          *lmv = &obd->u.lmv;
2483         struct lmv_tgt_desc     *tgt;
2484         ENTRY;
2485
2486         tgt = lmv_find_target(lmv, &och->och_fid);
2487         if (IS_ERR(tgt))
2488                 RETURN(PTR_ERR(tgt));
2489
2490         RETURN(md_set_open_replay_data(tgt->ltd_exp, och, open_req));
2491 }
2492
2493 int lmv_clear_open_replay_data(struct obd_export *exp,
2494                                struct obd_client_handle *och)
2495 {
2496         struct obd_device       *obd = exp->exp_obd;
2497         struct lmv_obd          *lmv = &obd->u.lmv;
2498         struct lmv_tgt_desc     *tgt;
2499         ENTRY;
2500
2501         tgt = lmv_find_target(lmv, &och->och_fid);
2502         if (IS_ERR(tgt))
2503                 RETURN(PTR_ERR(tgt));
2504
2505         RETURN(md_clear_open_replay_data(tgt->ltd_exp, och));
2506 }
2507
2508 static int lmv_get_remote_perm(struct obd_export *exp,
2509                                const struct lu_fid *fid,
2510                                struct obd_capa *oc, __u32 suppgid,
2511                                struct ptlrpc_request **request)
2512 {
2513         struct obd_device       *obd = exp->exp_obd;
2514         struct lmv_obd          *lmv = &obd->u.lmv;
2515         struct lmv_tgt_desc     *tgt;
2516         int                      rc;
2517         ENTRY;
2518
2519         rc = lmv_check_connect(obd);
2520         if (rc)
2521                 RETURN(rc);
2522
2523         tgt = lmv_find_target(lmv, fid);
2524         if (IS_ERR(tgt))
2525                 RETURN(PTR_ERR(tgt));
2526
2527         rc = md_get_remote_perm(tgt->ltd_exp, fid, oc, suppgid, request);
2528         RETURN(rc);
2529 }
2530
2531 static int lmv_renew_capa(struct obd_export *exp, struct obd_capa *oc,
2532                           renew_capa_cb_t cb)
2533 {
2534         struct obd_device       *obd = exp->exp_obd;
2535         struct lmv_obd          *lmv = &obd->u.lmv;
2536         struct lmv_tgt_desc     *tgt;
2537         int                      rc;
2538         ENTRY;
2539
2540         rc = lmv_check_connect(obd);
2541         if (rc)
2542                 RETURN(rc);
2543
2544         tgt = lmv_find_target(lmv, &oc->c_capa.lc_fid);
2545         if (IS_ERR(tgt))
2546                 RETURN(PTR_ERR(tgt));
2547
2548         rc = md_renew_capa(tgt->ltd_exp, oc, cb);
2549         RETURN(rc);
2550 }
2551
2552 int lmv_unpack_capa(struct obd_export *exp, struct ptlrpc_request *req,
2553                     const struct req_msg_field *field, struct obd_capa **oc)
2554 {
2555         struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
2556
2557         return md_unpack_capa(lmv->tgts[0]->ltd_exp, req, field, oc);
2558 }
2559
2560 int lmv_intent_getattr_async(struct obd_export *exp,
2561                              struct md_enqueue_info *minfo,
2562                              struct ldlm_enqueue_info *einfo)
2563 {
2564         struct md_op_data       *op_data = &minfo->mi_data;
2565         struct obd_device       *obd = exp->exp_obd;
2566         struct lmv_obd          *lmv = &obd->u.lmv;
2567         struct lmv_tgt_desc     *tgt = NULL;
2568         int                      rc;
2569         ENTRY;
2570
2571         rc = lmv_check_connect(obd);
2572         if (rc)
2573                 RETURN(rc);
2574
2575         tgt = lmv_find_target(lmv, &op_data->op_fid1);
2576         if (IS_ERR(tgt))
2577                 RETURN(PTR_ERR(tgt));
2578
2579         rc = md_intent_getattr_async(tgt->ltd_exp, minfo, einfo);
2580         RETURN(rc);
2581 }
2582
2583 int lmv_revalidate_lock(struct obd_export *exp, struct lookup_intent *it,
2584                         struct lu_fid *fid, __u64 *bits)
2585 {
2586         struct obd_device       *obd = exp->exp_obd;
2587         struct lmv_obd          *lmv = &obd->u.lmv;
2588         struct lmv_tgt_desc     *tgt;
2589         int                      rc;
2590         ENTRY;
2591
2592         rc = lmv_check_connect(obd);
2593         if (rc)
2594                 RETURN(rc);
2595
2596         tgt = lmv_find_target(lmv, fid);
2597         if (IS_ERR(tgt))
2598                 RETURN(PTR_ERR(tgt));
2599
2600         rc = md_revalidate_lock(tgt->ltd_exp, it, fid, bits);
2601         RETURN(rc);
2602 }
2603
2604 /**
2605  * For lmv, only need to send request to master MDT, and the master MDT will
2606  * process with other slave MDTs. The only exception is Q_GETOQUOTA for which
2607  * we directly fetch data from the slave MDTs.
2608  */
2609 int lmv_quotactl(struct obd_device *unused, struct obd_export *exp,
2610                  struct obd_quotactl *oqctl)
2611 {
2612         struct obd_device   *obd = class_exp2obd(exp);
2613         struct lmv_obd      *lmv = &obd->u.lmv;
2614         struct lmv_tgt_desc *tgt = lmv->tgts[0];
2615         int                  rc = 0, i;
2616         __u64                curspace, curinodes;
2617         ENTRY;
2618
2619         if (!lmv->desc.ld_tgt_count || !tgt->ltd_active) {
2620                 CERROR("master lmv inactive\n");
2621                 RETURN(-EIO);
2622         }
2623
2624         if (oqctl->qc_cmd != Q_GETOQUOTA) {
2625                 rc = obd_quotactl(tgt->ltd_exp, oqctl);
2626                 RETURN(rc);
2627         }
2628
2629         curspace = curinodes = 0;
2630         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2631                 int err;
2632                 tgt = lmv->tgts[i];
2633
2634                 if (tgt == NULL || tgt->ltd_exp == NULL || tgt->ltd_active == 0)
2635                         continue;
2636                 if (!tgt->ltd_active) {
2637                         CDEBUG(D_HA, "mdt %d is inactive.\n", i);
2638                         continue;
2639                 }
2640
2641                 err = obd_quotactl(tgt->ltd_exp, oqctl);
2642                 if (err) {
2643                         CERROR("getquota on mdt %d failed. %d\n", i, err);
2644                         if (!rc)
2645                                 rc = err;
2646                 } else {
2647                         curspace += oqctl->qc_dqblk.dqb_curspace;
2648                         curinodes += oqctl->qc_dqblk.dqb_curinodes;
2649                 }
2650         }
2651         oqctl->qc_dqblk.dqb_curspace = curspace;
2652         oqctl->qc_dqblk.dqb_curinodes = curinodes;
2653
2654         RETURN(rc);
2655 }
2656
2657 int lmv_quotacheck(struct obd_device *unused, struct obd_export *exp,
2658                    struct obd_quotactl *oqctl)
2659 {
2660         struct obd_device   *obd = class_exp2obd(exp);
2661         struct lmv_obd      *lmv = &obd->u.lmv;
2662         struct lmv_tgt_desc *tgt;
2663         int                  i, rc = 0;
2664         ENTRY;
2665
2666         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2667                 int err;
2668                 tgt = lmv->tgts[i];
2669                 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active) {
2670                         CERROR("lmv idx %d inactive\n", i);
2671                         RETURN(-EIO);
2672                 }
2673
2674                 err = obd_quotacheck(tgt->ltd_exp, oqctl);
2675                 if (err && !rc)
2676                         rc = err;
2677         }
2678
2679         RETURN(rc);
2680 }
2681
2682 struct obd_ops lmv_obd_ops = {
2683         .o_owner                = THIS_MODULE,
2684         .o_setup                = lmv_setup,
2685         .o_cleanup              = lmv_cleanup,
2686         .o_precleanup           = lmv_precleanup,
2687         .o_process_config       = lmv_process_config,
2688         .o_connect              = lmv_connect,
2689         .o_disconnect           = lmv_disconnect,
2690         .o_statfs               = lmv_statfs,
2691         .o_get_info             = lmv_get_info,
2692         .o_set_info_async       = lmv_set_info_async,
2693         .o_packmd               = lmv_packmd,
2694         .o_unpackmd             = lmv_unpackmd,
2695         .o_notify               = lmv_notify,
2696         .o_get_uuid             = lmv_get_uuid,
2697         .o_iocontrol            = lmv_iocontrol,
2698         .o_quotacheck           = lmv_quotacheck,
2699         .o_quotactl             = lmv_quotactl
2700 };
2701
2702 struct md_ops lmv_md_ops = {
2703         .m_getstatus            = lmv_getstatus,
2704         .m_null_inode           = lmv_null_inode,
2705         .m_find_cbdata          = lmv_find_cbdata,
2706         .m_close                = lmv_close,
2707         .m_create               = lmv_create,
2708         .m_done_writing         = lmv_done_writing,
2709         .m_enqueue              = lmv_enqueue,
2710         .m_getattr              = lmv_getattr,
2711         .m_getxattr             = lmv_getxattr,
2712         .m_getattr_name         = lmv_getattr_name,
2713         .m_intent_lock          = lmv_intent_lock,
2714         .m_link                 = lmv_link,
2715         .m_rename               = lmv_rename,
2716         .m_setattr              = lmv_setattr,
2717         .m_setxattr             = lmv_setxattr,
2718         .m_sync                 = lmv_sync,
2719         .m_readpage             = lmv_readpage,
2720         .m_unlink               = lmv_unlink,
2721         .m_init_ea_size         = lmv_init_ea_size,
2722         .m_cancel_unused        = lmv_cancel_unused,
2723         .m_set_lock_data        = lmv_set_lock_data,
2724         .m_lock_match           = lmv_lock_match,
2725         .m_get_lustre_md        = lmv_get_lustre_md,
2726         .m_free_lustre_md       = lmv_free_lustre_md,
2727         .m_set_open_replay_data = lmv_set_open_replay_data,
2728         .m_clear_open_replay_data = lmv_clear_open_replay_data,
2729         .m_renew_capa           = lmv_renew_capa,
2730         .m_unpack_capa          = lmv_unpack_capa,
2731         .m_get_remote_perm      = lmv_get_remote_perm,
2732         .m_intent_getattr_async = lmv_intent_getattr_async,
2733         .m_revalidate_lock      = lmv_revalidate_lock
2734 };
2735
2736 int __init lmv_init(void)
2737 {
2738         struct lprocfs_static_vars lvars;
2739         int                        rc;
2740
2741         lprocfs_lmv_init_vars(&lvars);
2742
2743         rc = class_register_type(&lmv_obd_ops, &lmv_md_ops,
2744                                  lvars.module_vars, LUSTRE_LMV_NAME, NULL);
2745         return rc;
2746 }
2747
2748 #ifdef __KERNEL__
2749 static void lmv_exit(void)
2750 {
2751         class_unregister_type(LUSTRE_LMV_NAME);
2752 }
2753
2754 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
2755 MODULE_DESCRIPTION("Lustre Logical Metadata Volume OBD driver");
2756 MODULE_LICENSE("GPL");
2757
2758 module_init(lmv_init);
2759 module_exit(lmv_exit);
2760 #endif