lustre/lmv/lmv_obd.c

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