lustre/lmv/lmv_obd.c

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