lustre/lmv/lmv_obd.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  * GPL HEADER START
   5  *
   6  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   7  *
   8  * This program is free software; you can redistribute it and/or modify
   9  * it under the terms of the GNU General Public License version 2 only,
  10  * as published by the Free Software Foundation.
  11  *
  12  * This program is distributed in the hope that it will be useful, but
  13  * WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * General Public License version 2 for more details (a copy is included
  16  * in the LICENSE file that accompanied this code).
  17  *
  18  * You should have received a copy of the GNU General Public License
  19  * version 2 along with this program; If not, see
  20  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  21  *
  22  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  23  * CA 95054 USA or visit www.sun.com if you need additional information or
  24  * have any questions.
  25  *
  26  * GPL HEADER END
  27  */
  28 /*
  29  * Copyright  2008 Sun Microsystems, Inc. All rights reserved
  30  * Use is subject to license terms.
  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 #ifndef EXPORT_SYMTAB
  38 # define EXPORT_SYMTAB
  39 #endif
  40 #define DEBUG_SUBSYSTEM S_LMV
  41 #ifdef __KERNEL__
  42 #include <linux/slab.h>
  43 #include <linux/module.h>
  44 #include <linux/init.h>
  45 #include <linux/slab.h>
  46 #include <linux/pagemap.h>
  47 #include <linux/mm.h>
  48 #include <asm/div64.h>
  49 #include <linux/seq_file.h>
  50 #include <linux/namei.h>
  51 #else
  52 #include <liblustre.h>
  53 #endif
  54
  55 #include <lustre/lustre_idl.h>
  56 #include <lustre_log.h>
  57 #include <obd_support.h>
  58 #include <lustre_lib.h>
  59 #include <lustre_net.h>
  60 #include <obd_class.h>
  61 #include <lprocfs_status.h>
  62 #include <lustre_lite.h>
  63 #include <lustre_fid.h>
  64 #include "lmv_internal.h"
  65
  66 /* object cache. */
  67 cfs_mem_cache_t *lmv_object_cache;
  68 cfs_atomic_t lmv_object_count = CFS_ATOMIC_INIT(0);
  69
  70 static void lmv_activate_target(struct lmv_obd *lmv,
  71                                 struct lmv_tgt_desc *tgt,
  72                                 int activate)
  73 {
  74         if (tgt->ltd_active == activate)
  75                 return;
  76
  77         tgt->ltd_active = activate;
  78         lmv->desc.ld_active_tgt_count += (activate ? 1 : -1);
  79 }
  80
  81 /**
  82  * Error codes:
  83  *
  84  *  -EINVAL  : UUID can't be found in the LMV's target list
  85  *  -ENOTCONN: The UUID is found, but the target connection is bad (!)
  86  *  -EBADF   : The UUID is found, but the OBD of the wrong type (!)
  87  */
  88 static int lmv_set_mdc_active(struct lmv_obd *lmv, struct obd_uuid *uuid,
  89                               int activate)
  90 {
  91         struct lmv_tgt_desc    *tgt;
  92         struct obd_device      *obd;
  93         int                     i;
  94         int                     rc = 0;
  95         ENTRY;
  96
  97         CDEBUG(D_INFO, "Searching in lmv %p for uuid %s (activate=%d)\n",
  98                lmv, uuid->uuid, activate);
  99
 100         cfs_spin_lock(&lmv->lmv_lock);
 101         for (i = 0, tgt = lmv->tgts; i < lmv->desc.ld_tgt_count; i++, tgt++) {
 102                 if (tgt->ltd_exp == NULL)
 103                         continue;
 104
 105                 CDEBUG(D_INFO, "Target idx %d is %s conn "LPX64"\n",
 106                        i, tgt->ltd_uuid.uuid, tgt->ltd_exp->exp_handle.h_cookie);
 107
 108                 if (obd_uuid_equals(uuid, &tgt->ltd_uuid))
 109                         break;
 110         }
 111
 112         if (i == lmv->desc.ld_tgt_count)
 113                 GOTO(out_lmv_lock, rc = -EINVAL);
 114
 115         obd = class_exp2obd(tgt->ltd_exp);
 116         if (obd == NULL)
 117                 GOTO(out_lmv_lock, rc = -ENOTCONN);
 118
 119         CDEBUG(D_INFO, "Found OBD %s=%s device %d (%p) type %s at LMV idx %d\n",
 120                obd->obd_name, obd->obd_uuid.uuid, obd->obd_minor, obd,
 121                obd->obd_type->typ_name, i);
 122         LASSERT(strcmp(obd->obd_type->typ_name, LUSTRE_MDC_NAME) == 0);
 123
 124         if (tgt->ltd_active == activate) {
 125                 CDEBUG(D_INFO, "OBD %p already %sactive!\n", obd,
 126                        activate ? "" : "in");
 127                 GOTO(out_lmv_lock, rc);
 128         }
 129
 130         CDEBUG(D_INFO, "Marking OBD %p %sactive\n", obd,
 131                activate ? "" : "in");
 132         lmv_activate_target(lmv, tgt, activate);
 133         EXIT;
 134
 135  out_lmv_lock:
 136         cfs_spin_unlock(&lmv->lmv_lock);
 137         return rc;
 138 }
 139
 140 static int lmv_set_mdc_data(struct lmv_obd *lmv, struct obd_uuid *uuid,
 141                             struct obd_connect_data *data)
 142 {
 143         struct lmv_tgt_desc    *tgt;
 144         int                     i;
 145         ENTRY;
 146
 147         LASSERT(data != NULL);
 148
 149         cfs_spin_lock(&lmv->lmv_lock);
 150         for (i = 0, tgt = lmv->tgts; i < lmv->desc.ld_tgt_count; i++, tgt++) {
 151                 if (tgt->ltd_exp == NULL)
 152                         continue;
 153
 154                 if (obd_uuid_equals(uuid, &tgt->ltd_uuid)) {
 155                         lmv->datas[tgt->ltd_idx] = *data;
 156                         break;
 157                 }
 158         }
 159         cfs_spin_unlock(&lmv->lmv_lock);
 160         RETURN(0);
 161 }
 162
 163 struct obd_uuid *lmv_get_uuid(struct obd_export *exp) {
 164         struct obd_device *obd = exp->exp_obd;
 165         struct lmv_obd *lmv = &obd->u.lmv;
 166         return obd_get_uuid(lmv->tgts[0].ltd_exp);
 167 }
 168
 169 static int lmv_notify(struct obd_device *obd, struct obd_device *watched,
 170                       enum obd_notify_event ev, void *data)
 171 {
 172         struct obd_connect_data *conn_data;
 173         struct lmv_obd          *lmv = &obd->u.lmv;
 174         struct obd_uuid         *uuid;
 175         int                      rc = 0;
 176         ENTRY;
 177
 178         if (strcmp(watched->obd_type->typ_name, LUSTRE_MDC_NAME)) {
 179                 CERROR("unexpected notification of %s %s!\n",
 180                        watched->obd_type->typ_name,
 181                        watched->obd_name);
 182                 RETURN(-EINVAL);
 183         }
 184
 185         uuid = &watched->u.cli.cl_target_uuid;
 186         if (ev == OBD_NOTIFY_ACTIVE || ev == OBD_NOTIFY_INACTIVE) {
 187                 /*
 188                  * Set MDC as active before notifying the observer, so the
 189                  * observer can use the MDC normally.
 190                  */
 191                 rc = lmv_set_mdc_active(lmv, uuid,
 192                                         ev == OBD_NOTIFY_ACTIVE);
 193                 if (rc) {
 194                         CERROR("%sactivation of %s failed: %d\n",
 195                                ev == OBD_NOTIFY_ACTIVE ? "" : "de",
 196                                uuid->uuid, rc);
 197                         RETURN(rc);
 198                 }
 199         } else if (ev == OBD_NOTIFY_OCD) {
 200                 conn_data = &watched->u.cli.cl_import->imp_connect_data;
 201
 202                 /*
 203                  * Set connect data to desired target, update exp_connect_flags.
 204                  */
 205                 rc = lmv_set_mdc_data(lmv, uuid, conn_data);
 206                 if (rc) {
 207                         CERROR("can't set connect data to target %s, rc %d\n",
 208                                uuid->uuid, rc);
 209                         RETURN(rc);
 210                 }
 211
 212                 /*
 213                  * XXX: Make sure that ocd_connect_flags from all targets are
 214                  * the same. Otherwise one of MDTs runs wrong version or
 215                  * something like this.  --umka
 216                  */
 217                 obd->obd_self_export->exp_connect_flags =
 218                         conn_data->ocd_connect_flags;
 219         }
 220 #if 0
 221         else if (ev == OBD_NOTIFY_DISCON) {
 222                 /*
 223                  * For disconnect event, flush fld cache for failout MDS case.
 224                  */
 225                 fld_client_flush(&lmv->lmv_fld);
 226         }
 227 #endif
 228         /*
 229          * Pass the notification up the chain.
 230          */
 231         if (obd->obd_observer)
 232                 rc = obd_notify(obd->obd_observer, watched, ev, data);
 233
 234         RETURN(rc);
 235 }
 236
 237 /**
 238  * This is fake connect function. Its purpose is to initialize lmv and say
 239  * caller that everything is okay. Real connection will be performed later.
 240  */
 241 static int lmv_connect(const struct lu_env *env,
 242                        struct obd_export **exp, struct obd_device *obd,
 243                        struct obd_uuid *cluuid, struct obd_connect_data *data,
 244                        void *localdata)
 245 {
 246 #ifdef __KERNEL__
 247         struct proc_dir_entry *lmv_proc_dir;
 248 #endif
 249         struct lmv_obd        *lmv = &obd->u.lmv;
 250         struct lustre_handle  conn = { 0 };
 251         int                    rc = 0;
 252         ENTRY;
 253
 254         /*
 255          * We don't want to actually do the underlying connections more than
 256          * once, so keep track.
 257          */
 258         lmv->refcount++;
 259         if (lmv->refcount > 1) {
 260                 *exp = NULL;
 261                 RETURN(0);
 262         }
 263
 264         rc = class_connect(&conn, obd, cluuid);
 265         if (rc) {
 266                 CERROR("class_connection() returned %d\n", rc);
 267                 RETURN(rc);
 268         }
 269
 270         *exp = class_conn2export(&conn);
 271         class_export_get(*exp);
 272
 273         lmv->exp = *exp;
 274         lmv->connected = 0;
 275         lmv->cluuid = *cluuid;
 276
 277         if (data)
 278                 lmv->conn_data = *data;
 279
 280 #ifdef __KERNEL__
 281         lmv_proc_dir = lprocfs_register("target_obds", obd->obd_proc_entry,
 282                                         NULL, NULL);
 283         if (IS_ERR(lmv_proc_dir)) {
 284                 CERROR("could not register /proc/fs/lustre/%s/%s/target_obds.",
 285                        obd->obd_type->typ_name, obd->obd_name);
 286                 lmv_proc_dir = NULL;
 287         }
 288 #endif
 289
 290         /*
 291          * All real clients should perform actual connection right away, because
 292          * it is possible, that LMV will not have opportunity to connect targets
 293          * and MDC stuff will be called directly, for instance while reading
 294          * ../mdc/../kbytesfree procfs file, etc.
 295          */
 296         if (data->ocd_connect_flags & OBD_CONNECT_REAL)
 297                 rc = lmv_check_connect(obd);
 298
 299 #ifdef __KERNEL__
 300         if (rc) {
 301                 if (lmv_proc_dir)
 302                         lprocfs_remove(&lmv_proc_dir);
 303         }
 304 #endif
 305
 306         RETURN(rc);
 307 }
 308
 309 static void lmv_set_timeouts(struct obd_device *obd)
 310 {
 311         struct lmv_tgt_desc   *tgts;
 312         struct lmv_obd        *lmv;
 313         int                    i;
 314
 315         lmv = &obd->u.lmv;
 316         if (lmv->server_timeout == 0)
 317                 return;
 318
 319         if (lmv->connected == 0)
 320                 return;
 321
 322         for (i = 0, tgts = lmv->tgts; i < lmv->desc.ld_tgt_count; i++, tgts++) {
 323                 if (tgts->ltd_exp == NULL)
 324                         continue;
 325
 326                 obd_set_info_async(tgts->ltd_exp, sizeof(KEY_INTERMDS),
 327                                    KEY_INTERMDS, 0, NULL, NULL);
 328         }
 329 }
 330
 331 static int lmv_init_ea_size(struct obd_export *exp, int easize,
 332                             int def_easize, int cookiesize)
 333 {
 334         struct obd_device   *obd = exp->exp_obd;
 335         struct lmv_obd      *lmv = &obd->u.lmv;
 336         int                  i;
 337         int                  rc = 0;
 338         int                  change = 0;
 339         ENTRY;
 340
 341         if (lmv->max_easize < easize) {
 342                 lmv->max_easize = easize;
 343                 change = 1;
 344         }
 345         if (lmv->max_def_easize < def_easize) {
 346                 lmv->max_def_easize = def_easize;
 347                 change = 1;
 348         }
 349         if (lmv->max_cookiesize < cookiesize) {
 350                 lmv->max_cookiesize = cookiesize;
 351                 change = 1;
 352         }
 353         if (change == 0)
 354                 RETURN(0);
 355
 356         if (lmv->connected == 0)
 357                 RETURN(0);
 358
 359         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 360                 if (lmv->tgts[i].ltd_exp == NULL) {
 361                         CWARN("%s: NULL export for %d\n", obd->obd_name, i);
 362                         continue;
 363                 }
 364
 365                 rc = md_init_ea_size(lmv->tgts[i].ltd_exp, easize, def_easize,
 366                                      cookiesize);
 367                 if (rc) {
 368                         CERROR("obd_init_ea_size() failed on MDT target %d, "
 369                                "error %d.\n", i, rc);
 370                         break;
 371                 }
 372         }
 373         RETURN(rc);
 374 }
 375
 376 #define MAX_STRING_SIZE 128
 377
 378 int lmv_connect_mdc(struct obd_device *obd, struct lmv_tgt_desc *tgt)
 379 {
 380 #ifdef __KERNEL__
 381         struct proc_dir_entry   *lmv_proc_dir;
 382 #endif
 383         struct lmv_obd          *lmv = &obd->u.lmv;
 384         struct obd_uuid         *cluuid = &lmv->cluuid;
 385         struct obd_connect_data *mdc_data = NULL;
 386         struct obd_uuid          lmv_mdc_uuid = { "LMV_MDC_UUID" };
 387         struct obd_device       *mdc_obd;
 388         struct obd_export       *mdc_exp;
 389         struct lu_fld_target     target;
 390         int                      rc;
 391         ENTRY;
 392
 393         mdc_obd = class_find_client_obd(&tgt->ltd_uuid, LUSTRE_MDC_NAME,
 394                                         &obd->obd_uuid);
 395         if (!mdc_obd) {
 396                 CERROR("target %s not attached\n", tgt->ltd_uuid.uuid);
 397                 RETURN(-EINVAL);
 398         }
 399
 400         CDEBUG(D_CONFIG, "connect to %s(%s) - %s, %s FOR %s\n",
 401                 mdc_obd->obd_name, mdc_obd->obd_uuid.uuid,
 402                 tgt->ltd_uuid.uuid, obd->obd_uuid.uuid,
 403                 cluuid->uuid);
 404
 405         if (!mdc_obd->obd_set_up) {
 406                 CERROR("target %s is not set up\n", tgt->ltd_uuid.uuid);
 407                 RETURN(-EINVAL);
 408         }
 409
 410         rc = obd_connect(NULL, &mdc_exp, mdc_obd, &lmv_mdc_uuid,
 411                          &lmv->conn_data, NULL);
 412         if (rc) {
 413                 CERROR("target %s connect error %d\n", tgt->ltd_uuid.uuid, rc);
 414                 RETURN(rc);
 415         }
 416
 417         /*
 418          * Init fid sequence client for this mdc and add new fld target.
 419          */
 420         rc = obd_fid_init(mdc_exp);
 421         if (rc)
 422                 RETURN(rc);
 423
 424         target.ft_srv = NULL;
 425         target.ft_exp = mdc_exp;
 426         target.ft_idx = tgt->ltd_idx;
 427
 428         fld_client_add_target(&lmv->lmv_fld, &target);
 429
 430         mdc_data = &class_exp2cliimp(mdc_exp)->imp_connect_data;
 431
 432         rc = obd_register_observer(mdc_obd, obd);
 433         if (rc) {
 434                 obd_disconnect(mdc_exp);
 435                 CERROR("target %s register_observer error %d\n",
 436                        tgt->ltd_uuid.uuid, rc);
 437                 RETURN(rc);
 438         }
 439
 440         if (obd->obd_observer) {
 441                 /*
 442                  * Tell the observer about the new target.
 443                  */
 444                 rc = obd_notify(obd->obd_observer, mdc_exp->exp_obd,
 445                                 OBD_NOTIFY_ACTIVE, (void *)(tgt - lmv->tgts));
 446                 if (rc) {
 447                         obd_disconnect(mdc_exp);
 448                         RETURN(rc);
 449                 }
 450         }
 451
 452         tgt->ltd_active = 1;
 453         tgt->ltd_exp = mdc_exp;
 454         lmv->desc.ld_active_tgt_count++;
 455
 456         /*
 457          * Copy connect data, it may be used later.
 458          */
 459         lmv->datas[tgt->ltd_idx] = *mdc_data;
 460
 461         md_init_ea_size(tgt->ltd_exp, lmv->max_easize,
 462                         lmv->max_def_easize, lmv->max_cookiesize);
 463
 464         CDEBUG(D_CONFIG, "Connected to %s(%s) successfully (%d)\n",
 465                 mdc_obd->obd_name, mdc_obd->obd_uuid.uuid,
 466                 cfs_atomic_read(&obd->obd_refcount));
 467
 468 #ifdef __KERNEL__
 469         lmv_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds");
 470         if (lmv_proc_dir) {
 471                 struct proc_dir_entry *mdc_symlink;
 472
 473                 LASSERT(mdc_obd->obd_type != NULL);
 474                 LASSERT(mdc_obd->obd_type->typ_name != NULL);
 475                 mdc_symlink = lprocfs_add_symlink(mdc_obd->obd_name,
 476                                                   lmv_proc_dir,
 477                                                   "../../../%s/%s",
 478                                                   mdc_obd->obd_type->typ_name,
 479                                                   mdc_obd->obd_name);
 480                 if (mdc_symlink == NULL) {
 481                         CERROR("Could not register LMV target "
 482                                "/proc/fs/lustre/%s/%s/target_obds/%s.",
 483                                obd->obd_type->typ_name, obd->obd_name,
 484                                mdc_obd->obd_name);
 485                         lprocfs_remove(&lmv_proc_dir);
 486                         lmv_proc_dir = NULL;
 487                 }
 488         }
 489 #endif
 490         RETURN(0);
 491 }
 492
 493 int lmv_add_target(struct obd_device *obd, struct obd_uuid *tgt_uuid)
 494 {
 495         struct lmv_obd      *lmv = &obd->u.lmv;
 496         struct lmv_tgt_desc *tgt;
 497         int                  rc = 0;
 498         ENTRY;
 499
 500         CDEBUG(D_CONFIG, "Target uuid: %s.\n", tgt_uuid->uuid);
 501
 502         lmv_init_lock(lmv);
 503
 504         if (lmv->desc.ld_active_tgt_count >= LMV_MAX_TGT_COUNT) {
 505                 lmv_init_unlock(lmv);
 506                 CERROR("Can't add %s, LMV module compiled for %d MDCs. "
 507                        "That many MDCs already configured.\n",
 508                        tgt_uuid->uuid, LMV_MAX_TGT_COUNT);
 509                 RETURN(-EINVAL);
 510         }
 511         if (lmv->desc.ld_tgt_count == 0) {
 512                 struct obd_device *mdc_obd;
 513
 514                 mdc_obd = class_find_client_obd(tgt_uuid, LUSTRE_MDC_NAME,
 515                                                 &obd->obd_uuid);
 516                 if (!mdc_obd) {
 517                         lmv_init_unlock(lmv);
 518                         CERROR("Target %s not attached\n", tgt_uuid->uuid);
 519                         RETURN(-EINVAL);
 520                 }
 521
 522                 rc = obd_llog_init(obd, &obd->obd_olg, mdc_obd, NULL);
 523                 if (rc) {
 524                         lmv_init_unlock(lmv);
 525                         CERROR("lmv failed to setup llogging subsystems\n");
 526                 }
 527         }
 528         cfs_spin_lock(&lmv->lmv_lock);
 529         tgt = lmv->tgts + lmv->desc.ld_tgt_count++;
 530         tgt->ltd_uuid = *tgt_uuid;
 531         cfs_spin_unlock(&lmv->lmv_lock);
 532
 533         if (lmv->connected) {
 534                 rc = lmv_connect_mdc(obd, tgt);
 535                 if (rc) {
 536                         cfs_spin_lock(&lmv->lmv_lock);
 537                         lmv->desc.ld_tgt_count--;
 538                         memset(tgt, 0, sizeof(*tgt));
 539                         cfs_spin_unlock(&lmv->lmv_lock);
 540                 } else {
 541                         int easize = sizeof(struct lmv_stripe_md) +
 542                                      lmv->desc.ld_tgt_count *
 543                                      sizeof(struct lu_fid);
 544                         lmv_init_ea_size(obd->obd_self_export, easize, 0, 0);
 545                 }
 546         }
 547
 548         lmv_init_unlock(lmv);
 549         RETURN(rc);
 550 }
 551
 552 int lmv_check_connect(struct obd_device *obd)
 553 {
 554         struct lmv_obd       *lmv = &obd->u.lmv;
 555         struct lmv_tgt_desc  *tgt;
 556         int                   i;
 557         int                   rc;
 558         int                   easize;
 559         ENTRY;
 560
 561         if (lmv->connected)
 562                 RETURN(0);
 563
 564         lmv_init_lock(lmv);
 565         if (lmv->connected) {
 566                 lmv_init_unlock(lmv);
 567                 RETURN(0);
 568         }
 569
 570         if (lmv->desc.ld_tgt_count == 0) {
 571                 CERROR("%s: no targets configured.\n", obd->obd_name);
 572                 RETURN(-EINVAL);
 573         }
 574
 575         CDEBUG(D_CONFIG, "Time to connect %s to %s\n",
 576                lmv->cluuid.uuid, obd->obd_name);
 577
 578         LASSERT(lmv->tgts != NULL);
 579
 580         for (i = 0, tgt = lmv->tgts; i < lmv->desc.ld_tgt_count; i++, tgt++) {
 581                 rc = lmv_connect_mdc(obd, tgt);
 582                 if (rc)
 583                         GOTO(out_disc, rc);
 584         }
 585
 586         lmv_set_timeouts(obd);
 587         class_export_put(lmv->exp);
 588         lmv->connected = 1;
 589         easize = lmv_get_easize(lmv);
 590         lmv_init_ea_size(obd->obd_self_export, easize, 0, 0);
 591         lmv_init_unlock(lmv);
 592         RETURN(0);
 593
 594  out_disc:
 595         while (i-- > 0) {
 596                 int rc2;
 597                 --tgt;
 598                 tgt->ltd_active = 0;
 599                 if (tgt->ltd_exp) {
 600                         --lmv->desc.ld_active_tgt_count;
 601                         rc2 = obd_disconnect(tgt->ltd_exp);
 602                         if (rc2) {
 603                                 CERROR("LMV target %s disconnect on "
 604                                        "MDC idx %d: error %d\n",
 605                                        tgt->ltd_uuid.uuid, i, rc2);
 606                         }
 607                 }
 608         }
 609         class_disconnect(lmv->exp);
 610         lmv_init_unlock(lmv);
 611         RETURN(rc);
 612 }
 613
 614 static int lmv_disconnect_mdc(struct obd_device *obd, struct lmv_tgt_desc *tgt)
 615 {
 616 #ifdef __KERNEL__
 617         struct proc_dir_entry  *lmv_proc_dir;
 618 #endif
 619         struct lmv_obd         *lmv = &obd->u.lmv;
 620         struct obd_device      *mdc_obd;
 621         int                     rc;
 622         ENTRY;
 623
 624         LASSERT(tgt != NULL);
 625         LASSERT(obd != NULL);
 626
 627         mdc_obd = class_exp2obd(tgt->ltd_exp);
 628
 629         if (mdc_obd)
 630                 mdc_obd->obd_no_recov = obd->obd_no_recov;
 631
 632 #ifdef __KERNEL__
 633         lmv_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds");
 634         if (lmv_proc_dir) {
 635                 struct proc_dir_entry *mdc_symlink;
 636
 637                 mdc_symlink = lprocfs_srch(lmv_proc_dir, mdc_obd->obd_name);
 638                 if (mdc_symlink) {
 639                         lprocfs_remove(&mdc_symlink);
 640                 } else {
 641                         CERROR("/proc/fs/lustre/%s/%s/target_obds/%s missing\n",
 642                                obd->obd_type->typ_name, obd->obd_name,
 643                                mdc_obd->obd_name);
 644                 }
 645         }
 646 #endif
 647         rc = obd_fid_fini(tgt->ltd_exp);
 648         if (rc)
 649                 CERROR("Can't finanize fids factory\n");
 650
 651         CDEBUG(D_INFO, "Disconnected from %s(%s) successfully\n",
 652                tgt->ltd_exp->exp_obd->obd_name,
 653                tgt->ltd_exp->exp_obd->obd_uuid.uuid);
 654
 655         obd_register_observer(tgt->ltd_exp->exp_obd, NULL);
 656         rc = obd_disconnect(tgt->ltd_exp);
 657         if (rc) {
 658                 if (tgt->ltd_active) {
 659                         CERROR("Target %s disconnect error %d\n",
 660                                tgt->ltd_uuid.uuid, rc);
 661                 }
 662         }
 663
 664         lmv_activate_target(lmv, tgt, 0);
 665         tgt->ltd_exp = NULL;
 666         RETURN(0);
 667 }
 668
 669 static int lmv_disconnect(struct obd_export *exp)
 670 {
 671         struct obd_device     *obd = class_exp2obd(exp);
 672 #ifdef __KERNEL__
 673         struct proc_dir_entry *lmv_proc_dir;
 674 #endif
 675         struct lmv_obd        *lmv = &obd->u.lmv;
 676         int                    rc;
 677         int                    i;
 678         ENTRY;
 679
 680         if (!lmv->tgts)
 681                 goto out_local;
 682
 683         /*
 684          * Only disconnect the underlying layers on the final disconnect.
 685          */
 686         lmv->refcount--;
 687         if (lmv->refcount != 0)
 688                 goto out_local;
 689
 690         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 691                 if (lmv->tgts[i].ltd_exp == NULL)
 692                         continue;
 693                 lmv_disconnect_mdc(obd, &lmv->tgts[i]);
 694         }
 695
 696 #ifdef __KERNEL__
 697         lmv_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds");
 698         if (lmv_proc_dir) {
 699                 lprocfs_remove(&lmv_proc_dir);
 700         } else {
 701                 CERROR("/proc/fs/lustre/%s/%s/target_obds missing\n",
 702                        obd->obd_type->typ_name, obd->obd_name);
 703         }
 704 #endif
 705
 706 out_local:
 707         /*
 708          * This is the case when no real connection is established by
 709          * lmv_check_connect().
 710          */
 711         if (!lmv->connected)
 712                 class_export_put(exp);
 713         rc = class_disconnect(exp);
 714         if (lmv->refcount == 0)
 715                 lmv->connected = 0;
 716         RETURN(rc);
 717 }
 718
 719 static int lmv_iocontrol(unsigned int cmd, struct obd_export *exp,
 720                          int len, void *karg, void *uarg)
 721 {
 722         struct obd_device    *obddev = class_exp2obd(exp);
 723         struct lmv_obd       *lmv = &obddev->u.lmv;
 724         int                   i = 0;
 725         int                   rc = 0;
 726         int                   set = 0;
 727         int                   count = lmv->desc.ld_tgt_count;
 728         ENTRY;
 729
 730         if (count == 0)
 731                 RETURN(-ENOTTY);
 732
 733         switch (cmd) {
 734         case IOC_OBD_STATFS: {
 735                 struct obd_ioctl_data *data = karg;
 736                 struct obd_device *mdc_obd;
 737                 struct obd_statfs stat_buf = {0};
 738                 __u32 index;
 739
 740                 memcpy(&index, data->ioc_inlbuf2, sizeof(__u32));
 741                 if ((index >= count))
 742                         RETURN(-ENODEV);
 743
 744                 if (!lmv->tgts[index].ltd_active)
 745                         RETURN(-ENODATA);
 746
 747                 mdc_obd = class_exp2obd(lmv->tgts[index].ltd_exp);
 748                 if (!mdc_obd)
 749                         RETURN(-EINVAL);
 750
 751                 /* copy UUID */
 752                 if (cfs_copy_to_user(data->ioc_pbuf2, obd2cli_tgt(mdc_obd),
 753                                      min((int) data->ioc_plen2,
 754                                          (int) sizeof(struct obd_uuid))))
 755                         RETURN(-EFAULT);
 756
 757                 rc = obd_statfs(mdc_obd, &stat_buf,
 758                                 cfs_time_current_64() - CFS_HZ, 0);
 759                 if (rc)
 760                         RETURN(rc);
 761                 if (cfs_copy_to_user(data->ioc_pbuf1, &stat_buf,
 762                                      min((int) data->ioc_plen1,
 763                                          (int) sizeof(stat_buf))))
 764                         RETURN(-EFAULT);
 765                 break;
 766         }
 767         case OBD_IOC_QUOTACTL: {
 768                 struct if_quotactl *qctl = karg;
 769                 struct lmv_tgt_desc *tgt = NULL;
 770                 struct obd_quotactl *oqctl;
 771
 772                 if (qctl->qc_valid == QC_MDTIDX) {
 773                         if (qctl->qc_idx < 0 || count <= qctl->qc_idx)
 774                                 RETURN(-EINVAL);
 775
 776                         tgt = &lmv->tgts[qctl->qc_idx];
 777                         if (!tgt->ltd_exp)
 778                                 RETURN(-EINVAL);
 779                 } else if (qctl->qc_valid == QC_UUID) {
 780                         for (i = 0; i < count; i++) {
 781                                 tgt = &lmv->tgts[i];
 782                                 if (!obd_uuid_equals(&tgt->ltd_uuid,
 783                                                      &qctl->obd_uuid))
 784                                         continue;
 785
 786                                 if (tgt->ltd_exp == NULL)
 787                                         RETURN(-EINVAL);
 788
 789                                 break;
 790                         }
 791                 } else {
 792                         RETURN(-EINVAL);
 793                 }
 794
 795                 if (i >= count)
 796                         RETURN(-EAGAIN);
 797
 798                 LASSERT(tgt && tgt->ltd_exp);
 799                 OBD_ALLOC_PTR(oqctl);
 800                 if (!oqctl)
 801                         RETURN(-ENOMEM);
 802
 803                 QCTL_COPY(oqctl, qctl);
 804                 rc = obd_quotactl(tgt->ltd_exp, oqctl);
 805                 if (rc == 0) {
 806                         QCTL_COPY(qctl, oqctl);
 807                         qctl->qc_valid = QC_MDTIDX;
 808                         qctl->obd_uuid = tgt->ltd_uuid;
 809                 }
 810                 OBD_FREE_PTR(oqctl);
 811                 break;
 812         }
 813         case OBD_IOC_CHANGELOG_SEND:
 814         case OBD_IOC_CHANGELOG_CLEAR: {
 815                 struct ioc_changelog *icc = karg;
 816
 817                 if (icc->icc_mdtindex >= count)
 818                         RETURN(-ENODEV);
 819
 820                 rc = obd_iocontrol(cmd, lmv->tgts[icc->icc_mdtindex].ltd_exp,
 821                                    sizeof(*icc), icc, NULL);
 822                 break;
 823         }
 824         case LL_IOC_GET_CONNECT_FLAGS: {
 825                 rc = obd_iocontrol(cmd, lmv->tgts[0].ltd_exp, len, karg, uarg);
 826                 break;
 827         }
 828
 829         default : {
 830                 for (i = 0; i < count; i++) {
 831                         int err;
 832
 833                         if (lmv->tgts[i].ltd_exp == NULL)
 834                                 continue;
 835
 836                         err = obd_iocontrol(cmd, lmv->tgts[i].ltd_exp, len,
 837                                             karg, uarg);
 838                         if (err == -ENODATA && cmd == OBD_IOC_POLL_QUOTACHECK) {
 839                                 RETURN(err);
 840                         } else if (err) {
 841                                 if (lmv->tgts[i].ltd_active) {
 842                                         CERROR("error: iocontrol MDC %s on MDT"
 843                                                "idx %d cmd %x: err = %d\n",
 844                                                 lmv->tgts[i].ltd_uuid.uuid,
 845                                                 i, cmd, err);
 846                                         if (!rc)
 847                                                 rc = err;
 848                                 }
 849                         } else
 850                                 set = 1;
 851                 }
 852                 if (!set && !rc)
 853                         rc = -EIO;
 854         }
 855         }
 856         RETURN(rc);
 857 }
 858
 859 static int lmv_all_chars_policy(int count, const char *name,
 860                                 int len)
 861 {
 862         unsigned int c = 0;
 863
 864         while (len > 0)
 865                 c += name[--len];
 866         c = c % count;
 867         return c;
 868 }
 869
 870 static int lmv_nid_policy(struct lmv_obd *lmv)
 871 {
 872         struct obd_import *imp;
 873         __u32              id;
 874
 875         /*
 876          * XXX: To get nid we assume that underlying obd device is mdc.
 877          */
 878         imp = class_exp2cliimp(lmv->tgts[0].ltd_exp);
 879         id = imp->imp_connection->c_self ^ (imp->imp_connection->c_self >> 32);
 880         return id % lmv->desc.ld_tgt_count;
 881 }
 882
 883 static int lmv_choose_mds(struct lmv_obd *lmv, struct md_op_data *op_data,
 884                           placement_policy_t placement)
 885 {
 886         switch (placement) {
 887         case PLACEMENT_CHAR_POLICY:
 888                 return lmv_all_chars_policy(lmv->desc.ld_tgt_count,
 889                                             op_data->op_name,
 890                                             op_data->op_namelen);
 891         case PLACEMENT_NID_POLICY:
 892                 return lmv_nid_policy(lmv);
 893
 894         default:
 895                 break;
 896         }
 897
 898         CERROR("Unsupported placement policy %x\n", placement);
 899         return -EINVAL;
 900 }
 901
 902 /**
 903  * This is _inode_ placement policy function (not name).
 904  */
 905 static int lmv_placement_policy(struct obd_device *obd,
 906                                 struct md_op_data *op_data,
 907                                 mdsno_t *mds)
 908 {
 909         struct lmv_obd          *lmv = &obd->u.lmv;
 910         struct lmv_object       *obj;
 911         int                      rc;
 912         ENTRY;
 913
 914         LASSERT(mds != NULL);
 915
 916         if (lmv->desc.ld_tgt_count == 1) {
 917                 *mds = 0;
 918                 RETURN(0);
 919         }
 920
 921         /*
 922          * Allocate new fid on target according to operation type and parent
 923          * home mds.
 924          */
 925         obj = lmv_object_find(obd, &op_data->op_fid1);
 926         if (obj != NULL || op_data->op_name == NULL ||
 927             op_data->op_opc != LUSTRE_OPC_MKDIR) {
 928                 /*
 929                  * Allocate fid for non-dir or for null name or for case parent
 930                  * dir is split.
 931                  */
 932                 if (obj) {
 933                         lmv_object_put(obj);
 934
 935                         /*
 936                          * If we have this flag turned on, and we see that
 937                          * parent dir is split, this means, that caller did not
 938                          * notice split yet. This is race and we would like to
 939                          * let caller know that.
 940                          */
 941                         if (op_data->op_bias & MDS_CHECK_SPLIT)
 942                                 RETURN(-ERESTART);
 943                 }
 944
 945                 /*
 946                  * Allocate new fid on same mds where parent fid is located and
 947                  * where operation will be sent. In case of split dir, ->op_fid1
 948                  * and ->op_mds here will contain fid and mds of slave directory
 949                  * object (assigned by caller).
 950                  */
 951                 *mds = op_data->op_mds;
 952                 rc = 0;
 953         } else {
 954                 /*
 955                  * Parent directory is not split and we want to create a
 956                  * directory in it. Let's calculate where to place it according
 957                  * to operation data @op_data.
 958                  */
 959                 *mds = lmv_choose_mds(lmv, op_data, lmv->lmv_placement);
 960                 rc = 0;
 961         }
 962
 963         if (rc) {
 964                 CERROR("Can't choose MDS, err = %d\n", rc);
 965         } else {
 966                 LASSERT(*mds < lmv->desc.ld_tgt_count);
 967         }
 968
 969         RETURN(rc);
 970 }
 971
 972 int __lmv_fid_alloc(struct lmv_obd *lmv, struct lu_fid *fid,
 973                     mdsno_t mds)
 974 {
 975         struct lmv_tgt_desc *tgt;
 976         int                  rc;
 977         ENTRY;
 978
 979         tgt = lmv_get_target(lmv, mds);
 980
 981         /*
 982          * New seq alloc and FLD setup should be atomic. Otherwise we may find
 983          * on server that seq in new allocated fid is not yet known.
 984          */
 985         cfs_down(&tgt->ltd_fid_sem);
 986
 987         if (!tgt->ltd_active)
 988                 GOTO(out, rc = -ENODEV);
 989
 990         /*
 991          * Asking underlaying tgt layer to allocate new fid.
 992          */
 993         rc = obd_fid_alloc(tgt->ltd_exp, fid, NULL);
 994         if (rc > 0) {
 995                 LASSERT(fid_is_sane(fid));
 996                 rc = 0;
 997         }
 998
 999         EXIT;
1000 out:
1001         cfs_up(&tgt->ltd_fid_sem);
1002         return rc;
1003 }
1004
1005 int lmv_fid_alloc(struct obd_export *exp, struct lu_fid *fid,
1006                   struct md_op_data *op_data)
1007 {
1008         struct obd_device     *obd = class_exp2obd(exp);
1009         struct lmv_obd        *lmv = &obd->u.lmv;
1010         mdsno_t                mds;
1011         int                    rc;
1012         ENTRY;
1013
1014         LASSERT(op_data != NULL);
1015         LASSERT(fid != NULL);
1016
1017         rc = lmv_placement_policy(obd, op_data, &mds);
1018         if (rc) {
1019                 CERROR("Can't get target for allocating fid, "
1020                        "rc %d\n", rc);
1021                 RETURN(rc);
1022         }
1023
1024         rc = __lmv_fid_alloc(lmv, fid, mds);
1025         if (rc) {
1026                 CERROR("Can't alloc new fid, rc %d\n", rc);
1027                 RETURN(rc);
1028         }
1029
1030         RETURN(rc);
1031 }
1032
1033 static int lmv_fid_delete(struct obd_export *exp, const struct lu_fid *fid)
1034 {
1035         ENTRY;
1036         LASSERT(exp != NULL && fid != NULL);
1037         if (lmv_object_delete(exp, fid)) {
1038                 CDEBUG(D_INODE, "Object "DFID" is destroyed.\n",
1039                        PFID(fid));
1040         }
1041         RETURN(0);
1042 }
1043
1044 static int lmv_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
1045 {
1046         struct lmv_obd             *lmv = &obd->u.lmv;
1047         struct lprocfs_static_vars  lvars;
1048         struct lmv_desc            *desc;
1049         int                         rc;
1050         int                         i = 0;
1051         ENTRY;
1052
1053         if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
1054                 CERROR("LMV setup requires a descriptor\n");
1055                 RETURN(-EINVAL);
1056         }
1057
1058         desc = (struct lmv_desc *)lustre_cfg_buf(lcfg, 1);
1059         if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
1060                 CERROR("Lmv descriptor size wrong: %d > %d\n",
1061                        (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
1062                 RETURN(-EINVAL);
1063         }
1064
1065         lmv->tgts_size = LMV_MAX_TGT_COUNT * sizeof(struct lmv_tgt_desc);
1066
1067         OBD_ALLOC(lmv->tgts, lmv->tgts_size);
1068         if (lmv->tgts == NULL)
1069                 RETURN(-ENOMEM);
1070
1071         for (i = 0; i < LMV_MAX_TGT_COUNT; i++) {
1072                 cfs_sema_init(&lmv->tgts[i].ltd_fid_sem, 1);
1073                 lmv->tgts[i].ltd_idx = i;
1074         }
1075
1076         lmv->datas_size = LMV_MAX_TGT_COUNT * sizeof(struct obd_connect_data);
1077
1078         OBD_ALLOC(lmv->datas, lmv->datas_size);
1079         if (lmv->datas == NULL)
1080                 GOTO(out_free_tgts, rc = -ENOMEM);
1081
1082         obd_str2uuid(&lmv->desc.ld_uuid, desc->ld_uuid.uuid);
1083         lmv->desc.ld_tgt_count = 0;
1084         lmv->desc.ld_active_tgt_count = 0;
1085         lmv->max_cookiesize = 0;
1086         lmv->max_def_easize = 0;
1087         lmv->max_easize = 0;
1088         lmv->lmv_placement = PLACEMENT_CHAR_POLICY;
1089
1090         cfs_spin_lock_init(&lmv->lmv_lock);
1091         cfs_sema_init(&lmv->init_sem, 1);
1092
1093         rc = lmv_object_setup(obd);
1094         if (rc) {
1095                 CERROR("Can't setup LMV object manager, error %d.\n", rc);
1096                 GOTO(out_free_datas, rc);
1097         }
1098
1099         lprocfs_lmv_init_vars(&lvars);
1100         lprocfs_obd_setup(obd, lvars.obd_vars);
1101 #ifdef LPROCFS
1102         {
1103                 rc = lprocfs_seq_create(obd->obd_proc_entry, "target_obd_status",
1104                                         0444, &lmv_proc_target_fops, obd);
1105                 if (rc)
1106                         CWARN("Error adding target_obd_stats file (%d)\n", rc);
1107        }
1108 #endif
1109         rc = fld_client_init(&lmv->lmv_fld, obd->obd_name,
1110                              LUSTRE_CLI_FLD_HASH_DHT);
1111         if (rc) {
1112                 CERROR("Can't init FLD, err %d\n", rc);
1113                 GOTO(out_free_datas, rc);
1114         }
1115
1116         RETURN(0);
1117
1118 out_free_datas:
1119         OBD_FREE(lmv->datas, lmv->datas_size);
1120         lmv->datas = NULL;
1121 out_free_tgts:
1122         OBD_FREE(lmv->tgts, lmv->tgts_size);
1123         lmv->tgts = NULL;
1124         return rc;
1125 }
1126
1127 static int lmv_cleanup(struct obd_device *obd)
1128 {
1129         struct lmv_obd   *lmv = &obd->u.lmv;
1130         ENTRY;
1131
1132         fld_client_fini(&lmv->lmv_fld);
1133         lprocfs_obd_cleanup(obd);
1134         lmv_object_cleanup(obd);
1135         OBD_FREE(lmv->datas, lmv->datas_size);
1136         OBD_FREE(lmv->tgts, lmv->tgts_size);
1137
1138         RETURN(0);
1139 }
1140
1141 static int lmv_process_config(struct obd_device *obd, obd_count len, void *buf)
1142 {
1143         struct lustre_cfg     *lcfg = buf;
1144         struct obd_uuid        tgt_uuid;
1145         int                    rc;
1146         ENTRY;
1147
1148         switch(lcfg->lcfg_command) {
1149         case LCFG_ADD_MDC:
1150                 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(tgt_uuid.uuid))
1151                         GOTO(out, rc = -EINVAL);
1152
1153                 obd_str2uuid(&tgt_uuid, lustre_cfg_string(lcfg, 1));
1154                 rc = lmv_add_target(obd, &tgt_uuid);
1155                 GOTO(out, rc);
1156         default: {
1157                 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
1158                 GOTO(out, rc = -EINVAL);
1159         }
1160         }
1161 out:
1162         RETURN(rc);
1163 }
1164
1165 static int lmv_statfs(struct obd_device *obd, struct obd_statfs *osfs,
1166                       __u64 max_age, __u32 flags)
1167 {
1168         struct lmv_obd        *lmv = &obd->u.lmv;
1169         struct obd_statfs     *temp;
1170         int                    rc = 0;
1171         int                    i;
1172         ENTRY;
1173
1174         rc = lmv_check_connect(obd);
1175         if (rc)
1176                 RETURN(rc);
1177
1178         OBD_ALLOC(temp, sizeof(*temp));
1179         if (temp == NULL)
1180                 RETURN(-ENOMEM);
1181
1182         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1183                 if (lmv->tgts[i].ltd_exp == NULL)
1184                         continue;
1185
1186                 rc = obd_statfs(lmv->tgts[i].ltd_exp->exp_obd, temp,
1187                                 max_age, flags);
1188                 if (rc) {
1189                         CERROR("can't stat MDS #%d (%s), error %d\n", i,
1190                                lmv->tgts[i].ltd_exp->exp_obd->obd_name,
1191                                rc);
1192                         GOTO(out_free_temp, rc);
1193                 }
1194                 if (i == 0) {
1195                         *osfs = *temp;
1196                 } else {
1197                         osfs->os_bavail += temp->os_bavail;
1198                         osfs->os_blocks += temp->os_blocks;
1199                         osfs->os_ffree += temp->os_ffree;
1200                         osfs->os_files += temp->os_files;
1201                 }
1202         }
1203
1204         EXIT;
1205 out_free_temp:
1206         OBD_FREE(temp, sizeof(*temp));
1207         return rc;
1208 }
1209
1210 static int lmv_getstatus(struct obd_export *exp,
1211                          struct lu_fid *fid,
1212                          struct obd_capa **pc)
1213 {
1214         struct obd_device    *obd = exp->exp_obd;
1215         struct lmv_obd       *lmv = &obd->u.lmv;
1216         int                   rc;
1217         ENTRY;
1218
1219         rc = lmv_check_connect(obd);
1220         if (rc)
1221                 RETURN(rc);
1222
1223         rc = md_getstatus(lmv->tgts[0].ltd_exp, fid, pc);
1224         RETURN(rc);
1225 }
1226
1227 static int lmv_getxattr(struct obd_export *exp, const struct lu_fid *fid,
1228                         struct obd_capa *oc, obd_valid valid, const char *name,
1229                         const char *input, int input_size, int output_size,
1230                         int flags, struct ptlrpc_request **request)
1231 {
1232         struct obd_device      *obd = exp->exp_obd;
1233         struct lmv_obd         *lmv = &obd->u.lmv;
1234         struct lmv_tgt_desc    *tgt;
1235         int                     rc;
1236         ENTRY;
1237
1238         rc = lmv_check_connect(obd);
1239         if (rc)
1240                 RETURN(rc);
1241
1242         tgt = lmv_find_target(lmv, fid);
1243         if (IS_ERR(tgt))
1244                 RETURN(PTR_ERR(tgt));
1245
1246         rc = md_getxattr(tgt->ltd_exp, fid, oc, valid, name, input,
1247                          input_size, output_size, flags, request);
1248
1249         RETURN(rc);
1250 }
1251
1252 static int lmv_setxattr(struct obd_export *exp, const struct lu_fid *fid,
1253                         struct obd_capa *oc, obd_valid valid, const char *name,
1254                         const char *input, int input_size, int output_size,
1255                         int flags, __u32 suppgid,
1256                         struct ptlrpc_request **request)
1257 {
1258         struct obd_device      *obd = exp->exp_obd;
1259         struct lmv_obd         *lmv = &obd->u.lmv;
1260         struct lmv_tgt_desc    *tgt;
1261         int                     rc;
1262         ENTRY;
1263
1264         rc = lmv_check_connect(obd);
1265         if (rc)
1266                 RETURN(rc);
1267
1268         tgt = lmv_find_target(lmv, fid);
1269         if (IS_ERR(tgt))
1270                 RETURN(PTR_ERR(tgt));
1271
1272         rc = md_setxattr(tgt->ltd_exp, fid, oc, valid, name, input,
1273                          input_size, output_size, flags, suppgid,
1274                          request);
1275
1276         RETURN(rc);
1277 }
1278
1279 static int lmv_getattr(struct obd_export *exp, struct md_op_data *op_data,
1280                        struct ptlrpc_request **request)
1281 {
1282         struct obd_device       *obd = exp->exp_obd;
1283         struct lmv_obd          *lmv = &obd->u.lmv;
1284         struct lmv_tgt_desc     *tgt;
1285         struct lmv_object       *obj;
1286         int                      rc;
1287         int                      i;
1288         ENTRY;
1289
1290         rc = lmv_check_connect(obd);
1291         if (rc)
1292                 RETURN(rc);
1293
1294         tgt = lmv_find_target(lmv, &op_data->op_fid1);
1295         if (IS_ERR(tgt))
1296                 RETURN(PTR_ERR(tgt));
1297
1298         if (op_data->op_valid & OBD_MD_MDTIDX) {
1299                 op_data->op_mds = tgt->ltd_idx;
1300                 RETURN(0);
1301         }
1302
1303         rc = md_getattr(tgt->ltd_exp, op_data, request);
1304         if (rc)
1305                 RETURN(rc);
1306
1307         obj = lmv_object_find_lock(obd, &op_data->op_fid1);
1308
1309         CDEBUG(D_INODE, "GETATTR for "DFID" %s\n", PFID(&op_data->op_fid1),
1310                obj ? "(split)" : "");
1311
1312         /*
1313          * If object is split, then we loop over all the slaves and gather size
1314          * attribute. In ideal world we would have to gather also mds field from
1315          * all slaves, as object is spread over the cluster and this is
1316          * definitely interesting information and it is not good to loss it,
1317          * but...
1318          */
1319         if (obj) {
1320                 struct mdt_body *body;
1321
1322                 if (*request == NULL) {
1323                         lmv_object_put(obj);
1324                         RETURN(rc);
1325                 }
1326
1327                 body = req_capsule_server_get(&(*request)->rq_pill,
1328                                               &RMF_MDT_BODY);
1329                 LASSERT(body != NULL);
1330
1331                 for (i = 0; i < obj->lo_objcount; i++) {
1332                         if (lmv->tgts[i].ltd_exp == NULL) {
1333                                 CWARN("%s: NULL export for %d\n",
1334                                       obd->obd_name, i);
1335                                 continue;
1336                         }
1337
1338                         /*
1339                          * Skip master object.
1340                          */
1341                         if (lu_fid_eq(&obj->lo_fid, &obj->lo_stripes[i].ls_fid))
1342                                 continue;
1343
1344                         body->size += obj->lo_stripes[i].ls_size;
1345                 }
1346
1347                 lmv_object_put_unlock(obj);
1348         }
1349
1350         RETURN(rc);
1351 }
1352
1353 static int lmv_change_cbdata(struct obd_export *exp, const struct lu_fid *fid,
1354                              ldlm_iterator_t it, void *data)
1355 {
1356         struct obd_device   *obd = exp->exp_obd;
1357         struct lmv_obd      *lmv = &obd->u.lmv;
1358         int                  i;
1359         int                  rc;
1360         ENTRY;
1361
1362         rc = lmv_check_connect(obd);
1363         if (rc)
1364                 RETURN(rc);
1365
1366         CDEBUG(D_INODE, "CBDATA for "DFID"\n", PFID(fid));
1367
1368         /*
1369          * With CMD every object can have two locks in different namespaces:
1370          * lookup lock in space of mds storing direntry and update/open lock in
1371          * space of mds storing inode.
1372          */
1373         for (i = 0; i < lmv->desc.ld_tgt_count; i++)
1374                 md_change_cbdata(lmv->tgts[i].ltd_exp, fid, it, data);
1375
1376         RETURN(0);
1377 }
1378
1379 static int lmv_find_cbdata(struct obd_export *exp, const struct lu_fid *fid,
1380                            ldlm_iterator_t it, void *data)
1381 {
1382         struct obd_device   *obd = exp->exp_obd;
1383         struct lmv_obd      *lmv = &obd->u.lmv;
1384         int                  i;
1385         int                  rc;
1386         ENTRY;
1387
1388         rc = lmv_check_connect(obd);
1389         if (rc)
1390                 RETURN(rc);
1391
1392         CDEBUG(D_INODE, "CBDATA for "DFID"\n", PFID(fid));
1393
1394         /*
1395          * With CMD every object can have two locks in different namespaces:
1396          * lookup lock in space of mds storing direntry and update/open lock in
1397          * space of mds storing inode.
1398          */
1399         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1400                 rc = md_find_cbdata(lmv->tgts[i].ltd_exp, fid, it, data);
1401                 if (rc)
1402                         RETURN(rc);
1403         }
1404
1405         RETURN(rc);
1406 }
1407
1408
1409 static int lmv_close(struct obd_export *exp, struct md_op_data *op_data,
1410                      struct md_open_data *mod, struct ptlrpc_request **request)
1411 {
1412         struct obd_device     *obd = exp->exp_obd;
1413         struct lmv_obd        *lmv = &obd->u.lmv;
1414         struct lmv_tgt_desc   *tgt;
1415         int                    rc;
1416         ENTRY;
1417
1418         rc = lmv_check_connect(obd);
1419         if (rc)
1420                 RETURN(rc);
1421
1422         tgt = lmv_find_target(lmv, &op_data->op_fid1);
1423         if (IS_ERR(tgt))
1424                 RETURN(PTR_ERR(tgt));
1425
1426         CDEBUG(D_INODE, "CLOSE "DFID"\n", PFID(&op_data->op_fid1));
1427         rc = md_close(tgt->ltd_exp, op_data, mod, request);
1428         RETURN(rc);
1429 }
1430
1431 /**
1432  * Called in the case MDS returns -ERESTART on create on open, what means that
1433  * directory is split and its LMV presentation object has to be updated.
1434  */
1435 int lmv_handle_split(struct obd_export *exp, const struct lu_fid *fid)
1436 {
1437         struct obd_device       *obd = exp->exp_obd;
1438         struct lmv_obd          *lmv = &obd->u.lmv;
1439         struct ptlrpc_request   *req = NULL;
1440         struct lmv_tgt_desc     *tgt;
1441         struct lmv_object       *obj;
1442         struct lustre_md         md;
1443         struct md_op_data       *op_data;
1444         int                      mealen;
1445         int                      rc;
1446         __u64                    valid;
1447         ENTRY;
1448
1449         md.mea = NULL;
1450         mealen = lmv_get_easize(lmv);
1451
1452         valid = OBD_MD_FLEASIZE | OBD_MD_FLDIREA | OBD_MD_MEA;
1453
1454         tgt = lmv_find_target(lmv, fid);
1455         if (IS_ERR(tgt))
1456                 RETURN(PTR_ERR(tgt));
1457
1458         /*
1459          * Time to update mea of parent fid.
1460          */
1461
1462         OBD_ALLOC_PTR(op_data);
1463         if (op_data == NULL)
1464                 RETURN(-ENOMEM);
1465
1466         op_data->op_fid1 = *fid;
1467         op_data->op_mode = mealen;
1468         op_data->op_valid = valid;
1469
1470         rc = md_getattr(tgt->ltd_exp, op_data, &req);
1471         OBD_FREE_PTR(op_data);
1472         if (rc) {
1473                 CERROR("md_getattr() failed, error %d\n", rc);
1474                 GOTO(cleanup, rc);
1475         }
1476
1477         rc = md_get_lustre_md(tgt->ltd_exp, req, NULL, exp, &md);
1478         if (rc) {
1479                 CERROR("md_get_lustre_md() failed, error %d\n", rc);
1480                 GOTO(cleanup, rc);
1481         }
1482
1483         if (md.mea == NULL)
1484                 GOTO(cleanup, rc = -ENODATA);
1485
1486         obj = lmv_object_create(exp, fid, md.mea);
1487         if (IS_ERR(obj))
1488                 rc = PTR_ERR(obj);
1489         else
1490                 lmv_object_put(obj);
1491
1492         obd_free_memmd(exp, (void *)&md.mea);
1493         EXIT;
1494 cleanup:
1495         if (req)
1496                 ptlrpc_req_finished(req);
1497         return rc;
1498 }
1499
1500 int lmv_create(struct obd_export *exp, struct md_op_data *op_data,
1501                const void *data, int datalen, int mode, __u32 uid,
1502                __u32 gid, cfs_cap_t cap_effective, __u64 rdev,
1503                struct ptlrpc_request **request)
1504 {
1505         struct obd_device       *obd = exp->exp_obd;
1506         struct lmv_obd          *lmv = &obd->u.lmv;
1507         struct lmv_tgt_desc     *tgt;
1508         struct lmv_object       *obj;
1509         int                      rc;
1510         int                      loop = 0;
1511         int                      sidx;
1512         ENTRY;
1513
1514         rc = lmv_check_connect(obd);
1515         if (rc)
1516                 RETURN(rc);
1517
1518         if (!lmv->desc.ld_active_tgt_count)
1519                 RETURN(-EIO);
1520 repeat:
1521         ++loop;
1522         LASSERT(loop <= 2);
1523
1524         obj = lmv_object_find(obd, &op_data->op_fid1);
1525         if (obj) {
1526                 sidx = raw_name2idx(obj->lo_hashtype, obj->lo_objcount,
1527                                        op_data->op_name, op_data->op_namelen);
1528                 op_data->op_fid1 = obj->lo_stripes[sidx].ls_fid;
1529                 op_data->op_bias &= ~MDS_CHECK_SPLIT;
1530                 op_data->op_mds = obj->lo_stripes[sidx].ls_mds;
1531                 tgt = lmv_get_target(lmv, op_data->op_mds);
1532                 lmv_object_put(obj);
1533         } else {
1534                 tgt = lmv_find_target(lmv, &op_data->op_fid1);
1535                 op_data->op_bias |= MDS_CHECK_SPLIT;
1536                 op_data->op_mds = tgt->ltd_idx;
1537         }
1538
1539         if (IS_ERR(tgt))
1540                 RETURN(PTR_ERR(tgt));
1541
1542         rc = lmv_fid_alloc(exp, &op_data->op_fid2, op_data);
1543         if (rc == -ERESTART)
1544                 goto repeat;
1545         else if (rc)
1546                 RETURN(rc);
1547
1548         CDEBUG(D_INODE, "CREATE '%*s' on "DFID" -> mds #%x\n",
1549                op_data->op_namelen, op_data->op_name, PFID(&op_data->op_fid1),
1550                op_data->op_mds);
1551
1552         op_data->op_flags |= MF_MDC_CANCEL_FID1;
1553         rc = md_create(tgt->ltd_exp, op_data, data, datalen, mode, uid, gid,
1554                        cap_effective, rdev, request);
1555         if (rc == 0) {
1556                 if (*request == NULL)
1557                         RETURN(rc);
1558                 CDEBUG(D_INODE, "Created - "DFID"\n", PFID(&op_data->op_fid2));
1559         } else if (rc == -ERESTART) {
1560                 LASSERT(*request != NULL);
1561                 DEBUG_REQ(D_WARNING|D_RPCTRACE, *request,
1562                           "Got -ERESTART during create!\n");
1563                 ptlrpc_req_finished(*request);
1564                 *request = NULL;
1565
1566                 /*
1567                  * Directory got split. Time to update local object and repeat
1568                  * the request with proper MDS.
1569                  */
1570                 rc = lmv_handle_split(exp, &op_data->op_fid1);
1571                 if (rc == 0) {
1572                         rc = lmv_allocate_slaves(obd, &op_data->op_fid1,
1573                                                  op_data, &op_data->op_fid2);
1574                         if (rc)
1575                                 RETURN(rc);
1576                         goto repeat;
1577                 }
1578         }
1579         RETURN(rc);
1580 }
1581
1582 static int lmv_done_writing(struct obd_export *exp,
1583                             struct md_op_data *op_data,
1584                             struct md_open_data *mod)
1585 {
1586         struct obd_device     *obd = exp->exp_obd;
1587         struct lmv_obd        *lmv = &obd->u.lmv;
1588         struct lmv_tgt_desc   *tgt;
1589         int                    rc;
1590         ENTRY;
1591
1592         rc = lmv_check_connect(obd);
1593         if (rc)
1594                 RETURN(rc);
1595
1596         tgt = lmv_find_target(lmv, &op_data->op_fid1);
1597         if (IS_ERR(tgt))
1598                 RETURN(PTR_ERR(tgt));
1599
1600         rc = md_done_writing(tgt->ltd_exp, op_data, mod);
1601         RETURN(rc);
1602 }
1603
1604 static int
1605 lmv_enqueue_slaves(struct obd_export *exp, struct ldlm_enqueue_info *einfo,
1606                    struct lookup_intent *it, struct md_op_data *op_data,
1607                    struct lustre_handle *lockh, void *lmm, int lmmsize)
1608 {
1609         struct obd_device     *obd = exp->exp_obd;
1610         struct lmv_obd        *lmv = &obd->u.lmv;
1611         struct lmv_stripe_md  *mea = op_data->op_mea1;
1612         struct md_op_data     *op_data2;
1613         struct lmv_tgt_desc   *tgt;
1614         int                    i;
1615         int                    rc = 0;
1616         ENTRY;
1617
1618         OBD_ALLOC_PTR(op_data2);
1619         if (op_data2 == NULL)
1620                 RETURN(-ENOMEM);
1621
1622         LASSERT(mea != NULL);
1623         for (i = 0; i < mea->mea_count; i++) {
1624                 memset(op_data2, 0, sizeof(*op_data2));
1625                 op_data2->op_fid1 = mea->mea_ids[i];
1626                 op_data2->op_bias = 0;
1627
1628                 tgt = lmv_find_target(lmv, &op_data2->op_fid1);
1629                 if (IS_ERR(tgt))
1630                         GOTO(cleanup, rc = PTR_ERR(tgt));
1631
1632                 if (tgt->ltd_exp == NULL)
1633                         continue;
1634
1635                 rc = md_enqueue(tgt->ltd_exp, einfo, it, op_data2,
1636                                 lockh + i, lmm, lmmsize, NULL, 0);
1637
1638                 CDEBUG(D_INODE, "Take lock on slave "DFID" -> %d/%d\n",
1639                        PFID(&mea->mea_ids[i]), rc, it->d.lustre.it_status);
1640
1641                 if (rc)
1642                         GOTO(cleanup, rc);
1643
1644                 if (it->d.lustre.it_data) {
1645                         struct ptlrpc_request *req;
1646                         req = (struct ptlrpc_request *)it->d.lustre.it_data;
1647                         ptlrpc_req_finished(req);
1648                 }
1649
1650                 if (it->d.lustre.it_status)
1651                         GOTO(cleanup, rc = it->d.lustre.it_status);
1652         }
1653
1654         EXIT;
1655 cleanup:
1656         OBD_FREE_PTR(op_data2);
1657
1658         if (rc != 0) {
1659                 /*
1660                  * Drop all taken locks.
1661                  */
1662                 while (--i >= 0) {
1663                         if (lockh[i].cookie)
1664                                 ldlm_lock_decref(lockh + i, einfo->ei_mode);
1665                         lockh[i].cookie = 0;
1666                 }
1667         }
1668         return rc;
1669 }
1670
1671 static int
1672 lmv_enqueue_remote(struct obd_export *exp, struct ldlm_enqueue_info *einfo,
1673                    struct lookup_intent *it, struct md_op_data *op_data,
1674                    struct lustre_handle *lockh, void *lmm, int lmmsize,
1675                    int extra_lock_flags)
1676 {
1677         struct ptlrpc_request      *req = it->d.lustre.it_data;
1678         struct obd_device          *obd = exp->exp_obd;
1679         struct lmv_obd             *lmv = &obd->u.lmv;
1680         struct lustre_handle        plock;
1681         struct lmv_tgt_desc        *tgt;
1682         struct md_op_data          *rdata;
1683         struct lu_fid               fid1;
1684         struct mdt_body            *body;
1685         int                         rc = 0;
1686         int                         pmode;
1687         ENTRY;
1688
1689         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1690         LASSERT(body != NULL);
1691
1692         if (!(body->valid & OBD_MD_MDS))
1693                 RETURN(0);
1694
1695         CDEBUG(D_INODE, "REMOTE_ENQUEUE '%s' on "DFID" -> "DFID"\n",
1696                LL_IT2STR(it), PFID(&op_data->op_fid1), PFID(&body->fid1));
1697
1698         /*
1699          * We got LOOKUP lock, but we really need attrs.
1700          */
1701         pmode = it->d.lustre.it_lock_mode;
1702         LASSERT(pmode != 0);
1703         memcpy(&plock, lockh, sizeof(plock));
1704         it->d.lustre.it_lock_mode = 0;
1705         it->d.lustre.it_data = NULL;
1706         fid1 = body->fid1;
1707
1708         it->d.lustre.it_disposition &= ~DISP_ENQ_COMPLETE;
1709         ptlrpc_req_finished(req);
1710
1711         tgt = lmv_find_target(lmv, &fid1);
1712         if (IS_ERR(tgt))
1713                 GOTO(out, rc = PTR_ERR(tgt));
1714
1715         OBD_ALLOC_PTR(rdata);
1716         if (rdata == NULL)
1717                 GOTO(out, rc = -ENOMEM);
1718
1719         rdata->op_fid1 = fid1;
1720         rdata->op_bias = MDS_CROSS_REF;
1721
1722         rc = md_enqueue(tgt->ltd_exp, einfo, it, rdata, lockh,
1723                         lmm, lmmsize, NULL, extra_lock_flags);
1724         OBD_FREE_PTR(rdata);
1725         EXIT;
1726 out:
1727         ldlm_lock_decref(&plock, pmode);
1728         return rc;
1729 }
1730
1731 static int
1732 lmv_enqueue(struct obd_export *exp, struct ldlm_enqueue_info *einfo,
1733             struct lookup_intent *it, struct md_op_data *op_data,
1734             struct lustre_handle *lockh, void *lmm, int lmmsize,
1735             struct ptlrpc_request **req, int extra_lock_flags)
1736 {
1737         struct obd_device        *obd = exp->exp_obd;
1738         struct lmv_obd           *lmv = &obd->u.lmv;
1739         struct lmv_tgt_desc      *tgt;
1740         struct lmv_object        *obj;
1741         int                       sidx;
1742         int                       rc;
1743         ENTRY;
1744
1745         rc = lmv_check_connect(obd);
1746         if (rc)
1747                 RETURN(rc);
1748
1749         CDEBUG(D_INODE, "ENQUEUE '%s' on "DFID"\n",
1750                LL_IT2STR(it), PFID(&op_data->op_fid1));
1751
1752         if (op_data->op_mea1 && it && it->it_op == IT_UNLINK) {
1753                 rc = lmv_enqueue_slaves(exp, einfo, it, op_data,
1754                                         lockh, lmm, lmmsize);
1755                 RETURN(rc);
1756         }
1757
1758         obj = lmv_object_find(obd, &op_data->op_fid1);
1759         if (obj && op_data->op_namelen) {
1760                 sidx = raw_name2idx(obj->lo_hashtype,
1761                                        obj->lo_objcount,
1762                                        (char *)op_data->op_name,
1763                                        op_data->op_namelen);
1764                 op_data->op_fid1 = obj->lo_stripes[sidx].ls_fid;
1765                 tgt = lmv_get_target(lmv, obj->lo_stripes[sidx].ls_mds);
1766         } else {
1767                 tgt = lmv_find_target(lmv, &op_data->op_fid1);
1768         }
1769         if (obj)
1770                 lmv_object_put(obj);
1771
1772         if (IS_ERR(tgt))
1773                 RETURN(PTR_ERR(tgt));
1774
1775         CDEBUG(D_INODE, "ENQUEUE '%s' on "DFID" -> mds #%d\n",
1776                LL_IT2STR(it), PFID(&op_data->op_fid1), tgt->ltd_idx);
1777
1778         rc = md_enqueue(tgt->ltd_exp, einfo, it, op_data, lockh,
1779                         lmm, lmmsize, req, extra_lock_flags);
1780
1781         if (rc == 0 && it && it->it_op == IT_OPEN) {
1782                 rc = lmv_enqueue_remote(exp, einfo, it, op_data, lockh,
1783                                         lmm, lmmsize, extra_lock_flags);
1784         }
1785         RETURN(rc);
1786 }
1787
1788 static int
1789 lmv_getattr_name(struct obd_export *exp,struct md_op_data *op_data,
1790                  struct ptlrpc_request **request)
1791 {
1792         struct ptlrpc_request   *req = NULL;
1793         struct obd_device       *obd = exp->exp_obd;
1794         struct lmv_obd          *lmv = &obd->u.lmv;
1795         struct lu_fid            rid = op_data->op_fid1;
1796         struct lmv_tgt_desc     *tgt;
1797         struct mdt_body         *body;
1798         struct lmv_object       *obj;
1799         obd_valid                valid = op_data->op_valid;
1800         int                      rc;
1801         int                      loop = 0;
1802         int                      sidx;
1803         ENTRY;
1804
1805         rc = lmv_check_connect(obd);
1806         if (rc)
1807                 RETURN(rc);
1808
1809 repeat:
1810         ++loop;
1811         LASSERT(loop <= 2);
1812         obj = lmv_object_find(obd, &rid);
1813         if (obj) {
1814                 sidx = raw_name2idx(obj->lo_hashtype, obj->lo_objcount,
1815                                     op_data->op_name, op_data->op_namelen);
1816                 rid = obj->lo_stripes[sidx].ls_fid;
1817                 tgt = lmv_get_target(lmv, obj->lo_stripes[sidx].ls_mds);
1818                 op_data->op_mds = obj->lo_stripes[sidx].ls_mds;
1819                 valid &= ~OBD_MD_FLCKSPLIT;
1820                 lmv_object_put(obj);
1821         } else {
1822                 tgt = lmv_find_target(lmv, &rid);
1823                 valid |= OBD_MD_FLCKSPLIT;
1824                 op_data->op_mds = tgt->ltd_idx;
1825         }
1826         if (IS_ERR(tgt))
1827                 RETURN(PTR_ERR(tgt));
1828
1829         CDEBUG(D_INODE, "GETATTR_NAME for %*s on "DFID" - "DFID" -> mds #%d\n",
1830                op_data->op_namelen, op_data->op_name, PFID(&op_data->op_fid1),
1831                PFID(&rid), tgt->ltd_idx);
1832
1833         op_data->op_valid = valid;
1834         op_data->op_fid1 = rid;
1835         rc = md_getattr_name(tgt->ltd_exp, op_data, request);
1836         if (rc == 0) {
1837                 body = req_capsule_server_get(&(*request)->rq_pill,
1838                                               &RMF_MDT_BODY);
1839                 LASSERT(body != NULL);
1840
1841                 if (body->valid & OBD_MD_MDS) {
1842                         rid = body->fid1;
1843                         CDEBUG(D_INODE, "Request attrs for "DFID"\n",
1844                                PFID(&rid));
1845
1846                         tgt = lmv_find_target(lmv, &rid);
1847                         if (IS_ERR(tgt)) {
1848                                 ptlrpc_req_finished(*request);
1849                                 RETURN(PTR_ERR(tgt));
1850                         }
1851
1852                         op_data->op_fid1 = rid;
1853                         op_data->op_valid |= OBD_MD_FLCROSSREF;
1854                         op_data->op_namelen = 0;
1855                         op_data->op_name = NULL;
1856                         rc = md_getattr_name(tgt->ltd_exp, op_data, &req);
1857                         ptlrpc_req_finished(*request);
1858                         *request = req;
1859                 }
1860         } else if (rc == -ERESTART) {
1861                 LASSERT(*request != NULL);
1862                 DEBUG_REQ(D_WARNING|D_RPCTRACE, *request,
1863                           "Got -ERESTART during getattr!\n");
1864                 ptlrpc_req_finished(*request);
1865                 *request = NULL;
1866
1867                 /*
1868                  * Directory got split. Time to update local object and repeat
1869                  * the request with proper MDS.
1870                  */
1871                 rc = lmv_handle_split(exp, &rid);
1872                 if (rc == 0)
1873                         goto repeat;
1874         }
1875         RETURN(rc);
1876 }
1877
1878 #define md_op_data_fid(op_data, fl)                     \
1879         (fl == MF_MDC_CANCEL_FID1 ? &op_data->op_fid1 : \
1880          fl == MF_MDC_CANCEL_FID2 ? &op_data->op_fid2 : \
1881          fl == MF_MDC_CANCEL_FID3 ? &op_data->op_fid3 : \
1882          fl == MF_MDC_CANCEL_FID4 ? &op_data->op_fid4 : \
1883          NULL)
1884
1885 static int lmv_early_cancel_slaves(struct obd_export *exp,
1886                                    struct md_op_data *op_data, int op_tgt,
1887                                    ldlm_mode_t mode, int bits, int flag)
1888 {
1889         struct obd_device      *obd = exp->exp_obd;
1890         struct lmv_obd         *lmv = &obd->u.lmv;
1891         ldlm_policy_data_t      policy = {{0}};
1892         struct lu_fid          *op_fid;
1893         struct lu_fid          *st_fid;
1894         struct lmv_tgt_desc    *tgt;
1895         struct lmv_object      *obj;
1896         int                     rc = 0;
1897         int                     i;
1898         ENTRY;
1899
1900         op_fid = md_op_data_fid(op_data, flag);
1901         if (!fid_is_sane(op_fid))
1902                 RETURN(0);
1903
1904         obj = lmv_object_find(obd, op_fid);
1905         if (obj == NULL)
1906                 RETURN(-EALREADY);
1907
1908         policy.l_inodebits.bits = bits;
1909         for (i = 0; i < obj->lo_objcount; i++) {
1910                 tgt = lmv_get_target(lmv, obj->lo_stripes[i].ls_mds);
1911                 st_fid = &obj->lo_stripes[i].ls_fid;
1912                 if (op_tgt != tgt->ltd_idx) {
1913                         CDEBUG(D_INODE, "EARLY_CANCEL slave "DFID" -> mds #%d\n",
1914                                PFID(st_fid), tgt->ltd_idx);
1915                         rc = md_cancel_unused(tgt->ltd_exp, st_fid, &policy,
1916                                               mode, LDLM_FL_ASYNC, NULL);
1917                         if (rc)
1918                                 GOTO(out_put_obj, rc);
1919                 } else {
1920                         CDEBUG(D_INODE,
1921                                "EARLY_CANCEL skip operation target %d on "DFID"\n",
1922                                op_tgt, PFID(st_fid));
1923                         /*
1924                          * Do not cancel locks for operation target, they will
1925                          * be handled later in underlaying layer when calling
1926                          * function we run on behalf of.
1927                          */
1928                         *op_fid = *st_fid;
1929                         op_data->op_flags |= flag;
1930                 }
1931         }
1932         EXIT;
1933 out_put_obj:
1934         lmv_object_put(obj);
1935         return rc;
1936 }
1937
1938 static int lmv_early_cancel(struct obd_export *exp, struct md_op_data *op_data,
1939                             int op_tgt, ldlm_mode_t mode, int bits, int flag)
1940 {
1941         struct lu_fid          *fid = md_op_data_fid(op_data, flag);
1942         struct obd_device      *obd = exp->exp_obd;
1943         struct lmv_obd         *lmv = &obd->u.lmv;
1944         struct lmv_tgt_desc    *tgt;
1945         ldlm_policy_data_t      policy = {{0}};
1946         struct lmv_object      *obj;
1947         int                     rc = 0;
1948         ENTRY;
1949
1950         if (!fid_is_sane(fid))
1951                 RETURN(0);
1952
1953         obj = lmv_object_find(obd, fid);
1954         if (obj) {
1955                 rc = lmv_early_cancel_slaves(exp, op_data, op_tgt, mode,
1956                                              bits, flag);
1957                 lmv_object_put(obj);
1958         } else {
1959                 tgt = lmv_find_target(lmv, fid);
1960                 if (IS_ERR(tgt))
1961                         RETURN(PTR_ERR(tgt));
1962
1963                 if (tgt->ltd_idx != op_tgt) {
1964                         CDEBUG(D_INODE, "EARLY_CANCEL on "DFID"\n", PFID(fid));
1965                         policy.l_inodebits.bits = bits;
1966                         rc = md_cancel_unused(tgt->ltd_exp, fid, &policy,
1967                                               mode, LDLM_FL_ASYNC, NULL);
1968                 } else {
1969                         CDEBUG(D_INODE,
1970                                "EARLY_CANCEL skip operation target %d on "DFID"\n",
1971                                op_tgt, PFID(fid));
1972                         op_data->op_flags |= flag;
1973                         rc = 0;
1974                 }
1975
1976         }
1977         RETURN(rc);
1978 }
1979
1980 /*
1981  * llite passes fid of an target inode in op_data->op_fid1 and id of directory in
1982  * op_data->op_fid2
1983  */
1984 static int lmv_link(struct obd_export *exp, struct md_op_data *op_data,
1985                     struct ptlrpc_request **request)
1986 {
1987         struct obd_device       *obd = exp->exp_obd;
1988         struct lmv_obd          *lmv = &obd->u.lmv;
1989         struct lmv_tgt_desc     *tgt;
1990         struct lmv_object       *obj;
1991         int                      rc;
1992         int                      loop = 0;
1993         mdsno_t                  mds;
1994         int                      sidx;
1995         ENTRY;
1996
1997         rc = lmv_check_connect(obd);
1998         if (rc)
1999                 RETURN(rc);
2000
2001 repeat:
2002         ++loop;
2003         LASSERT(loop <= 2);
2004         LASSERT(op_data->op_namelen != 0);
2005
2006         CDEBUG(D_INODE, "LINK "DFID":%*s to "DFID"\n",
2007                PFID(&op_data->op_fid2), op_data->op_namelen,
2008                op_data->op_name, PFID(&op_data->op_fid1));
2009
2010         obj = lmv_object_find(obd, &op_data->op_fid2);
2011         if (obj) {
2012                 sidx = raw_name2idx(obj->lo_hashtype,
2013                                     obj->lo_objcount,
2014                                     op_data->op_name,
2015                                     op_data->op_namelen);
2016                 op_data->op_fid2 = obj->lo_stripes[sidx].ls_fid;
2017                 mds = obj->lo_stripes[sidx].ls_mds;
2018                 lmv_object_put(obj);
2019         } else {
2020                 rc = lmv_fld_lookup(lmv, &op_data->op_fid2, &mds);
2021                 if (rc)
2022                         RETURN(rc);
2023         }
2024
2025         CDEBUG(D_INODE, "Forward to mds #%x ("DFID")\n",
2026                mds, PFID(&op_data->op_fid1));
2027
2028         op_data->op_fsuid = cfs_curproc_fsuid();
2029         op_data->op_fsgid = cfs_curproc_fsgid();
2030         op_data->op_cap = cfs_curproc_cap_pack();
2031         tgt = lmv_get_target(lmv, mds);
2032
2033         /*
2034          * Cancel UPDATE lock on child (fid1).
2035          */
2036         op_data->op_flags |= MF_MDC_CANCEL_FID2;
2037         rc = lmv_early_cancel(exp, op_data, tgt->ltd_idx, LCK_EX,
2038                               MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID1);
2039         if (rc == 0)
2040                 rc = md_link(tgt->ltd_exp, op_data, request);
2041         if (rc == -ERESTART) {
2042                 LASSERT(*request != NULL);
2043                 DEBUG_REQ(D_WARNING|D_RPCTRACE, *request,
2044                           "Got -ERESTART during link!\n");
2045                 ptlrpc_req_finished(*request);
2046                 *request = NULL;
2047
2048                 /*
2049                  * Directory got split. Time to update local object and repeat
2050                  * the request with proper MDS.
2051                  */
2052                 rc = lmv_handle_split(exp, &op_data->op_fid2);
2053                 if (rc == 0)
2054                         goto repeat;
2055         }
2056
2057         RETURN(rc);
2058 }
2059
2060 static int lmv_rename(struct obd_export *exp, struct md_op_data *op_data,
2061                       const char *old, int oldlen, const char *new, int newlen,
2062                       struct ptlrpc_request **request)
2063 {
2064         struct obd_device       *obd = exp->exp_obd;
2065         struct lmv_obd          *lmv = &obd->u.lmv;
2066         struct lmv_tgt_desc     *src_tgt;
2067         struct lmv_tgt_desc     *tgt_tgt;
2068         int                      rc;
2069         int                      sidx;
2070         int                      loop = 0;
2071         struct lmv_object       *obj;
2072         mdsno_t                  mds1;
2073         mdsno_t                  mds2;
2074         ENTRY;
2075
2076         LASSERT(oldlen != 0);
2077
2078         CDEBUG(D_INODE, "RENAME %*s in "DFID" to %*s in "DFID"\n",
2079                oldlen, old, PFID(&op_data->op_fid1),
2080                newlen, new, PFID(&op_data->op_fid2));
2081
2082         rc = lmv_check_connect(obd);
2083         if (rc)
2084                 RETURN(rc);
2085
2086 repeat:
2087         ++loop;
2088         LASSERT(loop <= 2);
2089         obj = lmv_object_find(obd, &op_data->op_fid1);
2090         if (obj) {
2091                 sidx = raw_name2idx(obj->lo_hashtype, obj->lo_objcount,
2092                                     (char *)old, oldlen);
2093                 op_data->op_fid1 = obj->lo_stripes[sidx].ls_fid;
2094                 mds1 = obj->lo_stripes[sidx].ls_mds;
2095                 CDEBUG(D_INODE, "Parent obj "DFID"\n", PFID(&op_data->op_fid1));
2096                 lmv_object_put(obj);
2097         } else {
2098                 rc = lmv_fld_lookup(lmv, &op_data->op_fid1, &mds1);
2099                 if (rc)
2100                         RETURN(rc);
2101         }
2102
2103         obj = lmv_object_find(obd, &op_data->op_fid2);
2104         if (obj) {
2105                 /*
2106                  * Directory is already split, so we have to forward request to
2107                  * the right MDS.
2108                  */
2109                 sidx = raw_name2idx(obj->lo_hashtype, obj->lo_objcount,
2110                                        (char *)new, newlen);
2111
2112                 mds2 = obj->lo_stripes[sidx].ls_mds;
2113                 op_data->op_fid2 = obj->lo_stripes[sidx].ls_fid;
2114                 CDEBUG(D_INODE, "Parent obj "DFID"\n", PFID(&op_data->op_fid2));
2115                 lmv_object_put(obj);
2116         } else {
2117                 rc = lmv_fld_lookup(lmv, &op_data->op_fid2, &mds2);
2118                 if (rc)
2119                         RETURN(rc);
2120         }
2121
2122         op_data->op_fsuid = cfs_curproc_fsuid();
2123         op_data->op_fsgid = cfs_curproc_fsgid();
2124         op_data->op_cap = cfs_curproc_cap_pack();
2125
2126         src_tgt = lmv_get_target(lmv, mds1);
2127         tgt_tgt = lmv_get_target(lmv, mds2);
2128
2129         /*
2130          * LOOKUP lock on src child (fid3) should also be cancelled for
2131          * src_tgt in mdc_rename.
2132          */
2133         op_data->op_flags |= MF_MDC_CANCEL_FID1 | MF_MDC_CANCEL_FID3;
2134
2135         /*
2136          * Cancel UPDATE locks on tgt parent (fid2), tgt_tgt is its
2137          * own target.
2138          */
2139         rc = lmv_early_cancel(exp, op_data, src_tgt->ltd_idx,
2140                               LCK_EX, MDS_INODELOCK_UPDATE,
2141                               MF_MDC_CANCEL_FID2);
2142
2143         /*
2144          * Cancel LOOKUP locks on tgt child (fid4) for parent tgt_tgt.
2145          */
2146         if (rc == 0) {
2147                 rc = lmv_early_cancel(exp, op_data, src_tgt->ltd_idx,
2148                                       LCK_EX, MDS_INODELOCK_LOOKUP,
2149                                       MF_MDC_CANCEL_FID4);
2150         }
2151
2152         /*
2153          * Cancel all the locks on tgt child (fid4).
2154          */
2155         if (rc == 0)
2156                 rc = lmv_early_cancel(exp, op_data, src_tgt->ltd_idx,
2157                                       LCK_EX, MDS_INODELOCK_FULL,
2158                                       MF_MDC_CANCEL_FID4);
2159
2160         if (rc == 0)
2161                 rc = md_rename(src_tgt->ltd_exp, op_data, old, oldlen,
2162                                new, newlen, request);
2163
2164         if (rc == -ERESTART) {
2165                 LASSERT(*request != NULL);
2166                 DEBUG_REQ(D_WARNING|D_RPCTRACE, *request,
2167                           "Got -ERESTART during rename!\n");
2168                 ptlrpc_req_finished(*request);
2169                 *request = NULL;
2170
2171                 /*
2172                  * Directory got split. Time to update local object and repeat
2173                  * the request with proper MDS.
2174                  */
2175                 rc = lmv_handle_split(exp, &op_data->op_fid1);
2176                 if (rc == 0)
2177                         goto repeat;
2178         }
2179         RETURN(rc);
2180 }
2181
2182 static int lmv_setattr(struct obd_export *exp, struct md_op_data *op_data,
2183                        void *ea, int ealen, void *ea2, int ea2len,
2184                        struct ptlrpc_request **request,
2185                        struct md_open_data **mod)
2186 {
2187         struct obd_device       *obd = exp->exp_obd;
2188         struct lmv_obd          *lmv = &obd->u.lmv;
2189         struct ptlrpc_request   *req;
2190         struct lmv_tgt_desc     *tgt;
2191         struct lmv_object       *obj;
2192         int                      rc = 0;
2193         int                      i;
2194         ENTRY;
2195
2196         rc = lmv_check_connect(obd);
2197         if (rc)
2198                 RETURN(rc);
2199
2200         obj = lmv_object_find(obd, &op_data->op_fid1);
2201
2202         CDEBUG(D_INODE, "SETATTR for "DFID", valid 0x%x%s\n",
2203                PFID(&op_data->op_fid1), op_data->op_attr.ia_valid,
2204                obj ? ", split" : "");
2205
2206         op_data->op_flags |= MF_MDC_CANCEL_FID1;
2207         if (obj) {
2208                 for (i = 0; i < obj->lo_objcount; i++) {
2209                         op_data->op_fid1 = obj->lo_stripes[i].ls_fid;
2210
2211                         tgt = lmv_get_target(lmv, obj->lo_stripes[i].ls_mds);
2212                         if (IS_ERR(tgt)) {
2213                                 rc = PTR_ERR(tgt);
2214                                 break;
2215                         }
2216
2217                         rc = md_setattr(tgt->ltd_exp, op_data, ea, ealen,
2218                                         ea2, ea2len, &req, mod);
2219
2220                         if (lu_fid_eq(&obj->lo_fid, &obj->lo_stripes[i].ls_fid)) {
2221                                 /*
2222                                  * This is master object and this request should
2223                                  * be returned back to llite.
2224                                  */
2225                                 *request = req;
2226                         } else {
2227                                 ptlrpc_req_finished(req);
2228                         }
2229
2230                         if (rc)
2231                                 break;
2232                 }
2233                 lmv_object_put(obj);
2234         } else {
2235                 tgt = lmv_find_target(lmv, &op_data->op_fid1);
2236                 if (IS_ERR(tgt))
2237                         RETURN(PTR_ERR(tgt));
2238
2239                 rc = md_setattr(tgt->ltd_exp, op_data, ea, ealen, ea2,
2240                                 ea2len, request, mod);
2241         }
2242         RETURN(rc);
2243 }
2244
2245 static int lmv_sync(struct obd_export *exp, const struct lu_fid *fid,
2246                     struct obd_capa *oc, struct ptlrpc_request **request)
2247 {
2248         struct obd_device         *obd = exp->exp_obd;
2249         struct lmv_obd            *lmv = &obd->u.lmv;
2250         struct lmv_tgt_desc       *tgt;
2251         int                        rc;
2252         ENTRY;
2253
2254         rc = lmv_check_connect(obd);
2255         if (rc)
2256                 RETURN(rc);
2257
2258         tgt = lmv_find_target(lmv, fid);
2259         if (IS_ERR(tgt))
2260                 RETURN(PTR_ERR(tgt));
2261
2262         rc = md_sync(tgt->ltd_exp, fid, oc, request);
2263         RETURN(rc);
2264 }
2265
2266 /**
2267  * Main purpose of LMV blocking ast is to remove split directory LMV
2268  * presentation object (struct lmv_object) attached to the lock being revoked.
2269  */
2270 int lmv_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2271                      void *data, int flag)
2272 {
2273         struct lustre_handle    lockh;
2274         struct lmv_object      *obj;
2275         int                     rc;
2276         ENTRY;
2277
2278         switch (flag) {
2279         case LDLM_CB_BLOCKING:
2280                 ldlm_lock2handle(lock, &lockh);
2281                 rc = ldlm_cli_cancel(&lockh);
2282                 if (rc < 0) {
2283                         CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
2284                         RETURN(rc);
2285                 }
2286                 break;
2287         case LDLM_CB_CANCELING:
2288                 /*
2289                  * Time to drop cached attrs for split directory object
2290                  */
2291                 obj = lock->l_ast_data;
2292                 if (obj) {
2293                         CDEBUG(D_INODE, "Cancel %s on "LPU64"/"LPU64
2294                                ", master "DFID"\n",
2295                                lock->l_resource->lr_name.name[3] == 1 ?
2296                                "LOOKUP" : "UPDATE",
2297                                lock->l_resource->lr_name.name[0],
2298                                lock->l_resource->lr_name.name[1],
2299                                PFID(&obj->lo_fid));
2300                         lmv_object_put(obj);
2301                 }
2302                 break;
2303         default:
2304                 LBUG();
2305         }
2306         RETURN(0);
2307 }
2308
2309 static void lmv_hash_adjust(__u64 *hash, __u64 hash_adj)
2310 {
2311         __u64         val;
2312
2313         val = le64_to_cpu(*hash);
2314         if (val < hash_adj)
2315                 val += MAX_HASH_SIZE;
2316         if (val != DIR_END_OFF)
2317                 *hash = cpu_to_le64(val - hash_adj);
2318 }
2319
2320 static __u32 lmv_node_rank(struct obd_export *exp, const struct lu_fid *fid)
2321 {
2322         __u64              id;
2323         struct obd_import *imp;
2324
2325         /*
2326          * XXX: to get nid we assume that underlying obd device is mdc.
2327          */
2328         imp  = class_exp2cliimp(exp);
2329         id   = imp->imp_connection->c_self + fid_flatten(fid);
2330
2331         CDEBUG(D_INODE, "Readpage node rank: "LPX64" "DFID" "LPX64" "LPX64"\n",
2332                imp->imp_connection->c_self, PFID(fid), id, id ^ (id >> 32));
2333
2334         return id ^ (id >> 32);
2335 }
2336
2337 static int lmv_readpage(struct obd_export *exp, const struct lu_fid *fid,
2338                         struct obd_capa *oc, __u64 offset64, struct page *page,
2339                         struct ptlrpc_request **request)
2340 {
2341         struct obd_device       *obd = exp->exp_obd;
2342         struct lmv_obd          *lmv = &obd->u.lmv;
2343         struct lu_fid            rid = *fid;
2344         struct lmv_object       *obj;
2345         __u64                    offset;
2346         __u64                    hash_adj = 0;
2347         __u32                    rank = 0;
2348         __u64                    seg_size = 0;
2349         __u64                    tgt_tmp = 0;
2350         int                      tgt_idx = 0;
2351         int                      tgt0_idx = 0;
2352         int                      rc;
2353         int                      nr = 0;
2354         struct lmv_stripe       *los;
2355         struct lmv_tgt_desc     *tgt;
2356         struct lu_dirpage       *dp;
2357         struct lu_dirent        *ent;
2358         ENTRY;
2359
2360         offset = offset64;
2361
2362         rc = lmv_check_connect(obd);
2363         if (rc)
2364                 RETURN(rc);
2365
2366         CDEBUG(D_INODE, "READPAGE at "LPX64" from "DFID"\n", offset, PFID(&rid));
2367
2368         /*
2369          * This case handle directory lookup in clustered metadata case (i.e.
2370          * split directory is located on multiple md servers.)
2371          * each server keeps directory entries for certain range of hashes.
2372          * E.g. we have N server and suppose hash range is 0 to MAX_HASH.
2373          * first server will keep records with hashes [ 0 ... MAX_HASH / N  - 1],
2374          * second one with hashes [MAX_HASH / N ... 2 * MAX_HASH / N] and
2375          * so on....
2376          *      readdir can simply start reading entries from 0 - N server in
2377          * order but that will not scale well as all client will request dir in
2378          * to server in same order.
2379          * Following algorithm does optimization:
2380          * Instead of doing readdir in 1, 2, ...., N order, client with a
2381          * rank R does readdir in R, R + 1, ..., N, 1, ... R - 1 order.
2382          * (every client has rank R)
2383          *      But ll_readdir() expect offset range [0 to MAX_HASH/N) but
2384          * since client ask dir from MDS{R} client has pages with offsets
2385          * [R*MAX_HASH/N ... (R + 1)*MAX_HASH/N] there for we do hash_adj
2386          * on hash  values that we get.
2387          */
2388         obj = lmv_object_find_lock(obd, fid);
2389         if (obj) {
2390                 nr       = obj->lo_objcount;
2391                 LASSERT(nr > 0);
2392                 seg_size = MAX_HASH_SIZE;
2393                 do_div(seg_size, nr);
2394                 los      = obj->lo_stripes;
2395                 tgt      = lmv_get_target(lmv, los[0].ls_mds);
2396                 rank     = lmv_node_rank(tgt->ltd_exp, fid) % nr;
2397                 tgt_tmp  = offset;
2398                 do_div(tgt_tmp, seg_size);
2399                 tgt0_idx = do_div(tgt_tmp,  nr);
2400                 tgt_idx  = (tgt0_idx + rank) % nr;
2401
2402                 if (tgt_idx < tgt0_idx)
2403                         /*
2404                          * Wrap around.
2405                          *
2406                          * Last segment has unusual length due to division
2407                          * rounding.
2408                          */
2409                         hash_adj = MAX_HASH_SIZE - seg_size * nr;
2410                 else
2411                         hash_adj = 0;
2412
2413                 hash_adj += rank * seg_size;
2414
2415                 CDEBUG(D_INODE, "Readpage hash adjustment: %x "LPX64" "
2416                        LPX64"/%x -> "LPX64"/%x\n", rank, hash_adj,
2417                        offset, tgt0_idx, offset + hash_adj, tgt_idx);
2418
2419                 offset = (offset + hash_adj) & MAX_HASH_SIZE;
2420                 rid = obj->lo_stripes[tgt_idx].ls_fid;
2421                 tgt = lmv_get_target(lmv, los[tgt_idx].ls_mds);
2422
2423                 CDEBUG(D_INODE, "Forward to "DFID" with offset %lu i %d\n",
2424                        PFID(&rid), (unsigned long)offset, tgt_idx);
2425         } else
2426                 tgt = lmv_find_target(lmv, &rid);
2427
2428         if (IS_ERR(tgt))
2429                 GOTO(cleanup, rc = PTR_ERR(tgt));
2430
2431         rc = md_readpage(tgt->ltd_exp, &rid, oc, offset, page, request);
2432         if (rc)
2433                 GOTO(cleanup, rc);
2434         if (obj) {
2435                 dp = cfs_kmap(page);
2436
2437                 lmv_hash_adjust(&dp->ldp_hash_start, hash_adj);
2438                 lmv_hash_adjust(&dp->ldp_hash_end,   hash_adj);
2439                 LASSERT(le64_to_cpu(dp->ldp_hash_start) <= offset64);
2440
2441                 for (ent = lu_dirent_start(dp); ent != NULL;
2442                      ent = lu_dirent_next(ent))
2443                         lmv_hash_adjust(&ent->lde_hash, hash_adj);
2444
2445                 if (tgt0_idx != nr - 1) {
2446                         __u64 end;
2447
2448                         end = le64_to_cpu(dp->ldp_hash_end);
2449                         if (end == DIR_END_OFF) {
2450                                 dp->ldp_hash_end = cpu_to_le32(seg_size *
2451                                                                (tgt0_idx + 1));
2452                                 CDEBUG(D_INODE,
2453                                        ""DFID" reset end "LPX64" tgt %d\n",
2454                                        PFID(&rid),
2455                                        (__u64)le64_to_cpu(dp->ldp_hash_end), tgt_idx);
2456                         }
2457                 }
2458                 cfs_kunmap(page);
2459         }
2460         EXIT;
2461 cleanup:
2462         if (obj)
2463                 lmv_object_put_unlock(obj);
2464         return rc;
2465 }
2466
2467 static int lmv_unlink(struct obd_export *exp, struct md_op_data *op_data,
2468                       struct ptlrpc_request **request)
2469 {
2470         struct obd_device       *obd = exp->exp_obd;
2471         struct lmv_obd          *lmv = &obd->u.lmv;
2472         struct lmv_tgt_desc     *tgt = NULL;
2473         struct lmv_object       *obj;
2474         int                      rc;
2475         int                      sidx;
2476         int                      loop = 0;
2477         ENTRY;
2478
2479         rc = lmv_check_connect(obd);
2480         if (rc)
2481                 RETURN(rc);
2482
2483 repeat:
2484         ++loop;
2485         LASSERT(loop <= 2);
2486         LASSERT(op_data->op_namelen != 0);
2487
2488         obj = lmv_object_find(obd, &op_data->op_fid1);
2489         if (obj) {
2490                 sidx = raw_name2idx(obj->lo_hashtype,
2491                                     obj->lo_objcount,
2492                                     op_data->op_name,
2493                                     op_data->op_namelen);
2494                 op_data->op_bias &= ~MDS_CHECK_SPLIT;
2495                 op_data->op_fid1 = obj->lo_stripes[sidx].ls_fid;
2496                 tgt = lmv_get_target(lmv,
2497                                      obj->lo_stripes[sidx].ls_mds);
2498                 lmv_object_put(obj);
2499                 CDEBUG(D_INODE, "UNLINK '%*s' in "DFID" -> %u\n",
2500                        op_data->op_namelen, op_data->op_name,
2501                        PFID(&op_data->op_fid1), sidx);
2502         }
2503
2504         if (tgt == NULL) {
2505                 tgt = lmv_find_target(lmv, &op_data->op_fid1);
2506                 if (IS_ERR(tgt))
2507                         RETURN(PTR_ERR(tgt));
2508                 op_data->op_bias |= MDS_CHECK_SPLIT;
2509         }
2510
2511         op_data->op_fsuid = cfs_curproc_fsuid();
2512         op_data->op_fsgid = cfs_curproc_fsgid();
2513         op_data->op_cap = cfs_curproc_cap_pack();
2514
2515         /*
2516          * If child's fid is given, cancel unused locks for it if it is from
2517          * another export than parent.
2518          *
2519          * LOOKUP lock for child (fid3) should also be cancelled on parent
2520          * tgt_tgt in mdc_unlink().
2521          */
2522         op_data->op_flags |= MF_MDC_CANCEL_FID1 | MF_MDC_CANCEL_FID3;
2523
2524         /*
2525          * Cancel FULL locks on child (fid3).
2526          */
2527         rc = lmv_early_cancel(exp, op_data, tgt->ltd_idx, LCK_EX,
2528                               MDS_INODELOCK_FULL, MF_MDC_CANCEL_FID3);
2529
2530         if (rc == 0)
2531                 rc = md_unlink(tgt->ltd_exp, op_data, request);
2532
2533         if (rc == -ERESTART) {
2534                 LASSERT(*request != NULL);
2535                 DEBUG_REQ(D_WARNING|D_RPCTRACE, *request,
2536                           "Got -ERESTART during unlink!\n");
2537                 ptlrpc_req_finished(*request);
2538                 *request = NULL;
2539
2540                 /*
2541                  * Directory got split. Time to update local object and repeat
2542                  * the request with proper MDS.
2543                  */
2544                 rc = lmv_handle_split(exp, &op_data->op_fid1);
2545                 if (rc == 0)
2546                         goto repeat;
2547         }
2548         RETURN(rc);
2549 }
2550
2551 static int lmv_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
2552 {
2553         int        rc = 0;
2554
2555         switch (stage) {
2556         case OBD_CLEANUP_EARLY:
2557                 /* XXX: here should be calling obd_precleanup() down to
2558                  * stack. */
2559                 break;
2560         case OBD_CLEANUP_EXPORTS:
2561                 rc = obd_llog_finish(obd, 0);
2562                 if (rc != 0)
2563                         CERROR("failed to cleanup llogging subsystems\n");
2564                 break;
2565         default:
2566                 break;
2567         }
2568         RETURN(rc);
2569 }
2570
2571 static int lmv_get_info(struct obd_export *exp, __u32 keylen,
2572                         void *key, __u32 *vallen, void *val,
2573                         struct lov_stripe_md *lsm)
2574 {
2575         struct obd_device       *obd;
2576         struct lmv_obd          *lmv;
2577         int                      rc = 0;
2578         ENTRY;
2579
2580         obd = class_exp2obd(exp);
2581         if (obd == NULL) {
2582                 CDEBUG(D_IOCTL, "Invalid client cookie "LPX64"\n",
2583                        exp->exp_handle.h_cookie);
2584                 RETURN(-EINVAL);
2585         }
2586
2587         lmv = &obd->u.lmv;
2588         if (keylen >= strlen("remote_flag") && !strcmp(key, "remote_flag")) {
2589                 struct lmv_tgt_desc *tgts;
2590                 int i;
2591
2592                 rc = lmv_check_connect(obd);
2593                 if (rc)
2594                         RETURN(rc);
2595
2596                 LASSERT(*vallen == sizeof(__u32));
2597                 for (i = 0, tgts = lmv->tgts; i < lmv->desc.ld_tgt_count;
2598                      i++, tgts++) {
2599
2600                         /*
2601                          * All tgts should be connected when this gets called.
2602                          */
2603                         if (!tgts || !tgts->ltd_exp) {
2604                                 CERROR("target not setup?\n");
2605                                 continue;
2606                         }
2607
2608                         if (!obd_get_info(tgts->ltd_exp, keylen, key,
2609                                           vallen, val, NULL))
2610                                 RETURN(0);
2611                 }
2612                 RETURN(-EINVAL);
2613         } else if (KEY_IS(KEY_MAX_EASIZE) || KEY_IS(KEY_CONN_DATA)) {
2614                 rc = lmv_check_connect(obd);
2615                 if (rc)
2616                         RETURN(rc);
2617
2618                 /*
2619                  * Forwarding this request to first MDS, it should know LOV
2620                  * desc.
2621                  */
2622                 rc = obd_get_info(lmv->tgts[0].ltd_exp, keylen, key,
2623                                   vallen, val, NULL);
2624                 if (!rc && KEY_IS(KEY_CONN_DATA)) {
2625                         exp->exp_connect_flags =
2626                         ((struct obd_connect_data *)val)->ocd_connect_flags;
2627                 }
2628                 RETURN(rc);
2629         }
2630
2631         CDEBUG(D_IOCTL, "Invalid key\n");
2632         RETURN(-EINVAL);
2633 }
2634
2635 int lmv_set_info_async(struct obd_export *exp, obd_count keylen,
2636                        void *key, obd_count vallen, void *val,
2637                        struct ptlrpc_request_set *set)
2638 {
2639         struct lmv_tgt_desc    *tgt;
2640         struct obd_device      *obd;
2641         struct lmv_obd         *lmv;
2642         int rc = 0;
2643         ENTRY;
2644
2645         obd = class_exp2obd(exp);
2646         if (obd == NULL) {
2647                 CDEBUG(D_IOCTL, "Invalid client cookie "LPX64"\n",
2648                        exp->exp_handle.h_cookie);
2649                 RETURN(-EINVAL);
2650         }
2651         lmv = &obd->u.lmv;
2652
2653         if (KEY_IS(KEY_READ_ONLY) || KEY_IS(KEY_FLUSH_CTX) ||
2654             KEY_IS(KEY_INIT_RECOV_BACKUP)) {
2655                 int i, err = 0;
2656
2657                 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2658                         tgt = &lmv->tgts[i];
2659
2660                         if (!tgt->ltd_exp)
2661                                 continue;
2662
2663                         err = obd_set_info_async(tgt->ltd_exp,
2664                                                  keylen, key, vallen, val, set);
2665                         if (err && rc == 0)
2666                                 rc = err;
2667                 }
2668
2669                 RETURN(rc);
2670         }
2671
2672         RETURN(-EINVAL);
2673 }
2674
2675 int lmv_packmd(struct obd_export *exp, struct lov_mds_md **lmmp,
2676                struct lov_stripe_md *lsm)
2677 {
2678         struct obd_device         *obd = class_exp2obd(exp);
2679         struct lmv_obd            *lmv = &obd->u.lmv;
2680         struct lmv_stripe_md      *meap;
2681         struct lmv_stripe_md      *lsmp;
2682         int                        mea_size;
2683         int                        i;
2684         ENTRY;
2685
2686         mea_size = lmv_get_easize(lmv);
2687         if (!lmmp)
2688                 RETURN(mea_size);
2689
2690         if (*lmmp && !lsm) {
2691                 OBD_FREE(*lmmp, mea_size);
2692                 *lmmp = NULL;
2693                 RETURN(0);
2694         }
2695
2696         if (*lmmp == NULL) {
2697                 OBD_ALLOC(*lmmp, mea_size);
2698                 if (*lmmp == NULL)
2699                         RETURN(-ENOMEM);
2700         }
2701
2702         if (!lsm)
2703                 RETURN(mea_size);
2704
2705         lsmp = (struct lmv_stripe_md *)lsm;
2706         meap = (struct lmv_stripe_md *)*lmmp;
2707
2708         if (lsmp->mea_magic != MEA_MAGIC_LAST_CHAR &&
2709             lsmp->mea_magic != MEA_MAGIC_ALL_CHARS)
2710                 RETURN(-EINVAL);
2711
2712         meap->mea_magic = cpu_to_le32(lsmp->mea_magic);
2713         meap->mea_count = cpu_to_le32(lsmp->mea_count);
2714         meap->mea_master = cpu_to_le32(lsmp->mea_master);
2715
2716         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2717                 meap->mea_ids[i] = meap->mea_ids[i];
2718                 fid_cpu_to_le(&meap->mea_ids[i], &meap->mea_ids[i]);
2719         }
2720
2721         RETURN(mea_size);
2722 }
2723
2724 int lmv_unpackmd(struct obd_export *exp, struct lov_stripe_md **lsmp,
2725                  struct lov_mds_md *lmm, int lmm_size)
2726 {
2727         struct obd_device          *obd = class_exp2obd(exp);
2728         struct lmv_stripe_md      **tmea = (struct lmv_stripe_md **)lsmp;
2729         struct lmv_stripe_md       *mea = (struct lmv_stripe_md *)lmm;
2730         struct lmv_obd             *lmv = &obd->u.lmv;
2731         int                         mea_size;
2732         int                         i;
2733         __u32                       magic;
2734         ENTRY;
2735
2736         mea_size = lmv_get_easize(lmv);
2737         if (lsmp == NULL)
2738                 return mea_size;
2739
2740         if (*lsmp != NULL && lmm == NULL) {
2741                 OBD_FREE(*tmea, mea_size);
2742                 *lsmp = NULL;
2743                 RETURN(0);
2744         }
2745
2746         LASSERT(mea_size == lmm_size);
2747
2748         OBD_ALLOC(*tmea, mea_size);
2749         if (*tmea == NULL)
2750                 RETURN(-ENOMEM);
2751
2752         if (!lmm)
2753                 RETURN(mea_size);
2754
2755         if (mea->mea_magic == MEA_MAGIC_LAST_CHAR ||
2756             mea->mea_magic == MEA_MAGIC_ALL_CHARS ||
2757             mea->mea_magic == MEA_MAGIC_HASH_SEGMENT)
2758         {
2759                 magic = le32_to_cpu(mea->mea_magic);
2760         } else {
2761                 /*
2762                  * Old mea is not handled here.
2763                  */
2764                 CERROR("Old not supportable EA is found\n");
2765                 LBUG();
2766         }
2767
2768         (*tmea)->mea_magic = magic;
2769         (*tmea)->mea_count = le32_to_cpu(mea->mea_count);
2770         (*tmea)->mea_master = le32_to_cpu(mea->mea_master);
2771
2772         for (i = 0; i < (*tmea)->mea_count; i++) {
2773                 (*tmea)->mea_ids[i] = mea->mea_ids[i];
2774                 fid_le_to_cpu(&(*tmea)->mea_ids[i], &(*tmea)->mea_ids[i]);
2775         }
2776         RETURN(mea_size);
2777 }
2778
2779 static int lmv_cancel_unused(struct obd_export *exp, const struct lu_fid *fid,
2780                              ldlm_policy_data_t *policy, ldlm_mode_t mode,
2781                              int flags, void *opaque)
2782 {
2783         struct obd_device       *obd = exp->exp_obd;
2784         struct lmv_obd          *lmv = &obd->u.lmv;
2785         int                      rc = 0;
2786         int                      err;
2787         int                      i;
2788         ENTRY;
2789
2790         LASSERT(fid != NULL);
2791
2792         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2793                 if (!lmv->tgts[i].ltd_exp || !lmv->tgts[i].ltd_active)
2794                         continue;
2795
2796                 err = md_cancel_unused(lmv->tgts[i].ltd_exp, fid,
2797                                        policy, mode, flags, opaque);
2798                 if (!rc)
2799                         rc = err;
2800         }
2801         RETURN(rc);
2802 }
2803
2804 int lmv_set_lock_data(struct obd_export *exp, __u64 *lockh, void *data,
2805                       __u32 *bits)
2806 {
2807         struct obd_device       *obd = exp->exp_obd;
2808         struct lmv_obd          *lmv = &obd->u.lmv;
2809         int                      rc;
2810         ENTRY;
2811
2812         rc =  md_set_lock_data(lmv->tgts[0].ltd_exp, lockh, data, bits);
2813         RETURN(rc);
2814 }
2815
2816 ldlm_mode_t lmv_lock_match(struct obd_export *exp, int flags,
2817                            const struct lu_fid *fid, ldlm_type_t type,
2818                            ldlm_policy_data_t *policy, ldlm_mode_t mode,
2819                            struct lustre_handle *lockh)
2820 {
2821         struct obd_device       *obd = exp->exp_obd;
2822         struct lmv_obd          *lmv = &obd->u.lmv;
2823         ldlm_mode_t              rc;
2824         int                      i;
2825         ENTRY;
2826
2827         CDEBUG(D_INODE, "Lock match for "DFID"\n", PFID(fid));
2828
2829         /*
2830          * With CMD every object can have two locks in different namespaces:
2831          * lookup lock in space of mds storing direntry and update/open lock in
2832          * space of mds storing inode. Thus we check all targets, not only that
2833          * one fid was created in.
2834          */
2835         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2836                 rc = md_lock_match(lmv->tgts[i].ltd_exp, flags, fid,
2837                                    type, policy, mode, lockh);
2838                 if (rc)
2839                         RETURN(rc);
2840         }
2841
2842         RETURN(0);
2843 }
2844
2845 int lmv_get_lustre_md(struct obd_export *exp, struct ptlrpc_request *req,
2846                       struct obd_export *dt_exp, struct obd_export *md_exp,
2847                       struct lustre_md *md)
2848 {
2849         struct obd_device       *obd = exp->exp_obd;
2850         struct lmv_obd          *lmv = &obd->u.lmv;
2851         int                      rc;
2852         ENTRY;
2853         rc = md_get_lustre_md(lmv->tgts[0].ltd_exp, req, dt_exp, md_exp, md);
2854         RETURN(rc);
2855 }
2856
2857 int lmv_free_lustre_md(struct obd_export *exp, struct lustre_md *md)
2858 {
2859         struct obd_device       *obd = exp->exp_obd;
2860         struct lmv_obd          *lmv = &obd->u.lmv;
2861         ENTRY;
2862
2863         if (md->mea)
2864                 obd_free_memmd(exp, (void *)&md->mea);
2865         RETURN(md_free_lustre_md(lmv->tgts[0].ltd_exp, md));
2866 }
2867
2868 int lmv_set_open_replay_data(struct obd_export *exp,
2869                              struct obd_client_handle *och,
2870                              struct ptlrpc_request *open_req)
2871 {
2872         struct obd_device       *obd = exp->exp_obd;
2873         struct lmv_obd          *lmv = &obd->u.lmv;
2874         struct lmv_tgt_desc     *tgt;
2875         ENTRY;
2876
2877         tgt = lmv_find_target(lmv, &och->och_fid);
2878         if (IS_ERR(tgt))
2879                 RETURN(PTR_ERR(tgt));
2880
2881         RETURN(md_set_open_replay_data(tgt->ltd_exp, och, open_req));
2882 }
2883
2884 int lmv_clear_open_replay_data(struct obd_export *exp,
2885                                struct obd_client_handle *och)
2886 {
2887         struct obd_device       *obd = exp->exp_obd;
2888         struct lmv_obd          *lmv = &obd->u.lmv;
2889         struct lmv_tgt_desc     *tgt;
2890         ENTRY;
2891
2892         tgt = lmv_find_target(lmv, &och->och_fid);
2893         if (IS_ERR(tgt))
2894                 RETURN(PTR_ERR(tgt));
2895
2896         RETURN(md_clear_open_replay_data(tgt->ltd_exp, och));
2897 }
2898
2899 static int lmv_get_remote_perm(struct obd_export *exp,
2900                                const struct lu_fid *fid,
2901                                struct obd_capa *oc, __u32 suppgid,
2902                                struct ptlrpc_request **request)
2903 {
2904         struct obd_device       *obd = exp->exp_obd;
2905         struct lmv_obd          *lmv = &obd->u.lmv;
2906         struct lmv_tgt_desc     *tgt;
2907         int                      rc;
2908         ENTRY;
2909
2910         rc = lmv_check_connect(obd);
2911         if (rc)
2912                 RETURN(rc);
2913
2914         tgt = lmv_find_target(lmv, fid);
2915         if (IS_ERR(tgt))
2916                 RETURN(PTR_ERR(tgt));
2917
2918         rc = md_get_remote_perm(tgt->ltd_exp, fid, oc, suppgid, request);
2919         RETURN(rc);
2920 }
2921
2922 static int lmv_renew_capa(struct obd_export *exp, struct obd_capa *oc,
2923                           renew_capa_cb_t cb)
2924 {
2925         struct obd_device       *obd = exp->exp_obd;
2926         struct lmv_obd          *lmv = &obd->u.lmv;
2927         struct lmv_tgt_desc     *tgt;
2928         int                      rc;
2929         ENTRY;
2930
2931         rc = lmv_check_connect(obd);
2932         if (rc)
2933                 RETURN(rc);
2934
2935         tgt = lmv_find_target(lmv, &oc->c_capa.lc_fid);
2936         if (IS_ERR(tgt))
2937                 RETURN(PTR_ERR(tgt));
2938
2939         rc = md_renew_capa(tgt->ltd_exp, oc, cb);
2940         RETURN(rc);
2941 }
2942
2943 int lmv_unpack_capa(struct obd_export *exp, struct ptlrpc_request *req,
2944                     const struct req_msg_field *field, struct obd_capa **oc)
2945 {
2946         struct obd_device *obd = exp->exp_obd;
2947         struct lmv_obd *lmv = &obd->u.lmv;
2948         int rc;
2949
2950         ENTRY;
2951         rc = md_unpack_capa(lmv->tgts[0].ltd_exp, req, field, oc);
2952         RETURN(rc);
2953 }
2954
2955 int lmv_intent_getattr_async(struct obd_export *exp,
2956                              struct md_enqueue_info *minfo,
2957                              struct ldlm_enqueue_info *einfo)
2958 {
2959         struct md_op_data       *op_data = &minfo->mi_data;
2960         struct obd_device       *obd = exp->exp_obd;
2961         struct lmv_obd          *lmv = &obd->u.lmv;
2962         struct lmv_object       *obj;
2963         struct lmv_tgt_desc     *tgt;
2964         int                      rc;
2965         int                      sidx;
2966         ENTRY;
2967
2968         rc = lmv_check_connect(obd);
2969         if (rc)
2970                 RETURN(rc);
2971
2972         if (!fid_is_sane(&op_data->op_fid2)) {
2973                 obj = lmv_object_find(obd, &op_data->op_fid1);
2974                 if (obj && op_data->op_namelen) {
2975                         sidx = raw_name2idx(obj->lo_hashtype,
2976                                             obj->lo_objcount,
2977                                             (char *)op_data->op_name,
2978                                             op_data->op_namelen);
2979                         op_data->op_fid1 = obj->lo_stripes[sidx].ls_fid;
2980                         tgt = lmv_get_target(lmv,
2981                                              obj->lo_stripes[sidx].ls_mds);
2982                         CDEBUG(D_INODE,
2983                                "Choose slave dir ("DFID") -> mds #%d\n",
2984                                PFID(&op_data->op_fid1), tgt->ltd_idx);
2985                 } else {
2986                         tgt = lmv_find_target(lmv, &op_data->op_fid1);
2987                 }
2988                 if (obj)
2989                         lmv_object_put(obj);
2990         } else {
2991                 op_data->op_fid1 = op_data->op_fid2;
2992                 tgt = lmv_find_target(lmv, &op_data->op_fid2);
2993                 op_data->op_bias = MDS_CROSS_REF;
2994                 /*
2995                  * Unfortunately, we have to lie to MDC/MDS to retrieve
2996                  * attributes llite needs.
2997                 */
2998                 if (minfo->mi_it.it_op & IT_LOOKUP)
2999                         minfo->mi_it.it_op = IT_GETATTR;
3000         }
3001
3002         if (IS_ERR(tgt))
3003                 RETURN(PTR_ERR(tgt));
3004
3005         rc = md_intent_getattr_async(tgt->ltd_exp, minfo, einfo);
3006         RETURN(rc);
3007 }
3008
3009 int lmv_revalidate_lock(struct obd_export *exp, struct lookup_intent *it,
3010                         struct lu_fid *fid, __u32 *bits)
3011 {
3012         struct obd_device       *obd = exp->exp_obd;
3013         struct lmv_obd          *lmv = &obd->u.lmv;
3014         struct lmv_tgt_desc     *tgt;
3015         int                      rc;
3016         ENTRY;
3017
3018         rc = lmv_check_connect(obd);
3019         if (rc)
3020                 RETURN(rc);
3021
3022         tgt = lmv_find_target(lmv, fid);
3023         if (IS_ERR(tgt))
3024                 RETURN(PTR_ERR(tgt));
3025
3026         rc = md_revalidate_lock(tgt->ltd_exp, it, fid, bits);
3027         RETURN(rc);
3028 }
3029
3030
3031 struct obd_ops lmv_obd_ops = {
3032         .o_owner                = THIS_MODULE,
3033         .o_setup                = lmv_setup,
3034         .o_cleanup              = lmv_cleanup,
3035         .o_precleanup           = lmv_precleanup,
3036         .o_process_config       = lmv_process_config,
3037         .o_connect              = lmv_connect,
3038         .o_disconnect           = lmv_disconnect,
3039         .o_statfs               = lmv_statfs,
3040         .o_get_info             = lmv_get_info,
3041         .o_set_info_async       = lmv_set_info_async,
3042         .o_packmd               = lmv_packmd,
3043         .o_unpackmd             = lmv_unpackmd,
3044         .o_notify               = lmv_notify,
3045         .o_get_uuid             = lmv_get_uuid,
3046         .o_iocontrol            = lmv_iocontrol,
3047         .o_fid_delete           = lmv_fid_delete
3048 };
3049
3050 struct md_ops lmv_md_ops = {
3051         .m_getstatus            = lmv_getstatus,
3052         .m_change_cbdata        = lmv_change_cbdata,
3053         .m_find_cbdata          = lmv_find_cbdata,
3054         .m_close                = lmv_close,
3055         .m_create               = lmv_create,
3056         .m_done_writing         = lmv_done_writing,
3057         .m_enqueue              = lmv_enqueue,
3058         .m_getattr              = lmv_getattr,
3059         .m_getxattr             = lmv_getxattr,
3060         .m_getattr_name         = lmv_getattr_name,
3061         .m_intent_lock          = lmv_intent_lock,
3062         .m_link                 = lmv_link,
3063         .m_rename               = lmv_rename,
3064         .m_setattr              = lmv_setattr,
3065         .m_setxattr             = lmv_setxattr,
3066         .m_sync                 = lmv_sync,
3067         .m_readpage             = lmv_readpage,
3068         .m_unlink               = lmv_unlink,
3069         .m_init_ea_size         = lmv_init_ea_size,
3070         .m_cancel_unused        = lmv_cancel_unused,
3071         .m_set_lock_data        = lmv_set_lock_data,
3072         .m_lock_match           = lmv_lock_match,
3073         .m_get_lustre_md        = lmv_get_lustre_md,
3074         .m_free_lustre_md       = lmv_free_lustre_md,
3075         .m_set_open_replay_data = lmv_set_open_replay_data,
3076         .m_clear_open_replay_data = lmv_clear_open_replay_data,
3077         .m_renew_capa           = lmv_renew_capa,
3078         .m_unpack_capa          = lmv_unpack_capa,
3079         .m_get_remote_perm      = lmv_get_remote_perm,
3080         .m_intent_getattr_async = lmv_intent_getattr_async,
3081         .m_revalidate_lock      = lmv_revalidate_lock
3082 };
3083
3084 static quota_interface_t *quota_interface;
3085 extern quota_interface_t lmv_quota_interface;
3086
3087 int __init lmv_init(void)
3088 {
3089         struct lprocfs_static_vars lvars;
3090         int                        rc;
3091
3092         lmv_object_cache = cfs_mem_cache_create("lmv_objects",
3093                                                 sizeof(struct lmv_object),
3094                                                 0, 0);
3095         if (!lmv_object_cache) {
3096                 CERROR("Error allocating lmv objects cache\n");
3097                 return -ENOMEM;
3098         }
3099
3100         lprocfs_lmv_init_vars(&lvars);
3101
3102         cfs_request_module("lquota");
3103         quota_interface = PORTAL_SYMBOL_GET(lmv_quota_interface);
3104         init_obd_quota_ops(quota_interface, &lmv_obd_ops);
3105
3106         rc = class_register_type(&lmv_obd_ops, &lmv_md_ops,
3107                                  lvars.module_vars, LUSTRE_LMV_NAME, NULL);
3108         if (rc) {
3109                 if (quota_interface)
3110                         PORTAL_SYMBOL_PUT(lmv_quota_interface);
3111                 cfs_mem_cache_destroy(lmv_object_cache);
3112         }
3113
3114         return rc;
3115 }
3116
3117 #ifdef __KERNEL__
3118 static void lmv_exit(void)
3119 {
3120         if (quota_interface)
3121                 PORTAL_SYMBOL_PUT(lmv_quota_interface);
3122
3123         class_unregister_type(LUSTRE_LMV_NAME);
3124
3125         LASSERTF(cfs_atomic_read(&lmv_object_count) == 0,
3126                  "Can't free lmv objects cache, %d object(s) busy\n",
3127                  cfs_atomic_read(&lmv_object_count));
3128         cfs_mem_cache_destroy(lmv_object_cache);
3129 }
3130
3131 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
3132 MODULE_DESCRIPTION("Lustre Logical Metadata Volume OBD driver");
3133 MODULE_LICENSE("GPL");
3134
3135 module_init(lmv_init);
3136 module_exit(lmv_exit);
3137 #endif